THE ENVIRONMENT AND EVOLUTION OF GALAXIES

ASTROPHYSICS AND
SPACE SCIENCE LIBRARY

A SERIES OF BOOKS ON THE RECENT DEVELOPMENTS

OF SPACE SCIENCE AND OF GENERAL GEOPHYSICS AND ASTROPHYSICS
PUBLISHED IN CONNECTION WITH THE JOURNAL
SPACE SCIENCE REVIEWS

Editorial Board

R. L. F. BOYD, University College, London, England

W. B. BURTON, Sterrewacht, Leiden, The Netherlands
C. DE JAGER, University of Utrecht, The Netherlands
J. KLECZEK, Czechoslovak Academy of Sciences, Ondfejov, Czechoslovakia
Z. KOPAL t, University of Manchester, England
R. LUST, Max-Planck-Institutfor Meteorologie, Hamburg, Germany
L. I. SEDOV, Academy of Sciences, Moscow, Russia
Z. SvESTKA, Laboratory for Space Research, Utrecht, The Netherlands

VOLUME 188
PROCEEDINGS
 THE ENVIRONMENT AND
EVOLUTION OF GALAXIES
PROCEEDINGS OF THE THIRD TETONS SUMMER SCHOOL
HELD IN GRAND TETON NATIONAL PARK, WYOMING, V.S.A., JULY 1992

Edited by

J. MICHAEL SHULL
University of C%rado at Bou/der,
Joint Institute for Laboratory Astrophysics,
Boulder, Colorado, U.S.A.
and

HARLEY A. THRONSON, JR.

University of Wyoming,
Wyoming Infrared Observatory,
Laramie, Wyoming, U.S.A.

SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

Library of Congress Cataloging-in-Publication Data
The Environment and evolution of galaxles I edlted by J. Mlehael Shull
and Harley A. Thronson, Jr.
p. em. -- (Astrophys 1es and spaee se i enee 11 brary ; v. 188)
Ineludes blbllographleal referenees and index.
ISBN 978-0-7923-2542-0 ISBN 978-94-011-1882-8 (eBook)
DOI 10.1007/978-94-011-1882-8
1. Galaxies--Evolution--Congresses. 1. Shull, J. Mlehael.
II. Thronson, Harley A. III. Serles.
QB857.5.E96E58 1993
523.1' 12--de20 93-21159

ISBN 978-0-7923-2542-0

Printed an acid-free paper

AlI Rights Reserved

© 1993 Springer Science+Business Media Dordrecht
Originally published by Kluwer Academic Publishers in 1993
No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical,
including photocopying, recording or by any information storage and
retrieval system, without written permission from the copyright owner.
 T ABLE OF CONTENTS

PREFACE ix
ACKNOWLEDGEMENTS xi

GASlROPHYSICAL COSMOLOGY - INTRODUCTION AND OVERVIEW 3

J. Richard Bond

RECENT RESULTS FROM COBE 27

C.L. Bennett, N.W. Boggess, M.G. Hauser, J.C. Mather, G.P' Smoot, and
E.L. Wright

THE FORMATION OF GALAXIES 59

Marc Davis

EXPERIMENTAL GALAXY FORMATION 69

August E. Evrard

POPULATION SYNTHESIS AND SPEClRAL EVOLUTION OF GALAXIES 91

Gustavo Bruzual A.

SEARCHING FOR HIDDEN GALAXIES 105

J.I. Davies

PANEL DISCUSSION: THE NATURE OF HIGH-REDSHIFT OBJECTS 127

J. Michael Shull

GALAXY EVOLUTION AS A FUNCTION OF ENVIRONMENT 131

Richard Ellis

DISTANT GALAXIES: EVOLUTIONARY CLUES 143

Simon Lilly

OLD, RED, AND DEAD RADIO GALAXIES AT LARGE REDSHIFf 151

Hyron Spinrad

PRIMEVAL GALAXIES, THE IGM, AND THE QSO-PROTOGALAXY

CONNECTION 155
Timothy M. Heckman
vi

THE EXTRAGALACTIC X-RAY BACKGROUND (1 - 20 keV) 159

Richard E. Griffiths

THE EVOLUTION OF THE INTERGALACTIC MEDIUM AND L YMAN-

ALPHA CLOUDS: COMPARISON OF THEORIES AND OBSERVATIONS 191
Stanislaw Bajtlik

PROPERTIES OF THE LYMAN a. FOREST AT HIGH AND LOW REDSHIFTS 213

Ray J. Weymann

QSO ABSORPTION LINES: IMPLICATIONS FOR GALAXY FORMATION

AND EVOLUTION 237
Kenneth M. Lanzetta

THE PROPERTIES OF ABSORPTION-LINE SELECTED HIGH-REDSHIFT

GALAXIES 263
Charles C. Steidel

MOLECULAR GAS AND THE YIELD OF YOUNG STARS IN GALAXIES 295

Judith S. Young

THE PHYSICAL INTERPRETATION OF MORPHOLOGY 305

Rosemary F.G. Wyse

MERGERS AND THE ORIGIN OF EARLY-TYPE GALAXIES 327

Lars Hemquist

INTERGALACTIC AND EXTENDED ATOMIC GAS 345

Jacqueline van Gorkum

ENERGY INPUT FROM AGNs 369

Mitchell C. Begelman

COOLING FLOWS IN CLUSTERS OF GALAXIES 383

Richard Mushotzky

EVOLUTION OF CLUSTERS AND THE INTRACLUSTER MEDIUM 409

Megan Donahue

THE QUASAR LUMINOSITY FUNCTION 433

BJ. Boyle

GALACTIC SUPERWINDS 455

Timothy M. Heckman, Matthew D. Lehnert, and Lee Armus
 vii

THE FIRST STARS 499

Fred C. Adams

THE BIRTH OF STARS IN GALAXIES 533

Robert C. Kennicutt, Jr.

THE FIRST QSs AND THEIR INFLUENCE ON THE IN1ERGALACTIC

MEDIUM 559
Jill Bechtold

INDEX 579
 PREFACE
J. MICHAEL SHULL & HARLEY THRONSON
University of Colorado and University of Wyoming

1. Introduction

The 1992 Tetons Conference on The Envi'r onment and Evolution of Galaxies
was the third in our series of meetings dealing with the interstellar medium
(ISM) in all its forms. The 1986 conference on Interstellar Processes was
followed by The Interstellar Medium of Galaxies in 1989. As a continuation
of our outward expansion into the universe, it was perhaps natural that our
third meeting should consider galaxy formation and the relationships among
galaxies, structure, and the intergalactic medium (IGM). Tetons III is also
the first formal joint Wyoming - Colorado conference, although the first two
involved substantial unofficial collaboration.
As we discllssed this concept with our organizing committee, the topics to
be covered by the meeting were expanded to include large-scale structure,
the new COBE and ROSAT data, galaxy population synthesis, "cooling
flows"in clusters, and AGN /starburst energy-input to galaxies. However, in
choosing our speakers, we attempted to highlight several themes more than
others - for example, the IGM and high-redshift galaxies. The latter topic
was chosen as the subject ofthe Panel Discussion (another Tetons tradition).
Perhaps owing to the breadth of the topics, and undoubtedly due to the
splendid natural setting, Tetons III was the largest of our meetings, with 305
registered participants. We are pleased that the success of our conferences
continues to grow, but we suspect that we have reached a natural size limit.
Tetons III was also intended to be a return to the "Summer School" format
of Tetons I, with an emphasis on pedagogical talks as well as current research
results. We therefore made special efforts to solicit speakers who would
translate these talks into valuable written contributions. We also endeavored
to invite a substantial number of younger speakers, in order to stimulate new
ways of interpreting the vast literature in the topics of this meeting. Rob
Kennicutt, speaking on the last day, made a perceptive remark that he was
one of the oldest speakers at the conference (at which point Ray VVeymann
begged to differ). For the benefit of those who count such things, we note
that 5 of the 23 invited speakers were women.
Organizing and running a conference of this magnitude requires the efforts
of many people, together with substantial funding - "the mother's milk
of astrophysics" (with apologies to Paul Tsongas). The administrations
and departments of the Universities of Colorado and Wyoming provided
generous assistance for the conference advertising, organization, and sec-
retarial support. In particular, we thank Dr. Risa Palm (Dean, Graduate

ix
x

School, CU), Dr. Oliver Walter (Dean, Arts & Sciences, U Wyo.), and our
respective department chairs and center directors (Glen Rebka and Lee
Schick at Wyoming; Ellen Zweibel and Ted Snow at Colorado) for their
support. We are also extremely grateful to the National Science Foundation
(NSF) and the National Aeronautics and Space Administration (NASA) for
their financial support, which allowed us to provide partial travel support
for young astronomers and to subsidize, once again, the publication of an
inexpensive soft-cover version of the procedings. In the spirit of a summer
school, we hope that this book will be used by a large number of students
and younger astronomers in their classes and research - for at least as long
as the time until Tetons IV.
During the final editing process, we have benefitted greatly from the
efforts of a group of CU grad students and postdocs (Mark Fardal, Mark
Giroux, Phil Maloney, and Jon Slavin) who assisted and advised the editors .
Special thanks is due to Janet Ferguson-Shaw at CU, who rescued valuable
files after a disk crash. We thank Gert Kiers and Michiel Kolman of Kluwer
Academic Publishers for their assistance in guiding this book to its final
form. Finally, we thank President and Mrs. Reagan for appearing at lunch
during the conference's last day, demonstrating their appreciation for the
astronomical and astrological significance of our meeting. We hope that this
book will be used by a large number of students and younger astronomers
in theor classes and research - at least until Tetons IV (July 1995).
We close with some comments on Wyoming, courtesy ofMitch Begelman,
who originally presented them at his invited talk, July 8, 1992.

TOP 10 REASONS WYOMING IS AN ENLIGHTENED STATE

10. Its Observatory is only 1 hour from campus. '

9. It's the only state with 2 "Little A mer'ica" truck stops.
8. There are solar-powered rest areas on freeways.
7. The first real-estate developers were British aristrocrats.
6. They invented the first "Watergate-type" scandal ('Teappt Dome', 1927).
5. First county public library system.
4. First voting rights for women {1869}.
3. First woman governor of a state (1925).
2. Bottles of (good) wine for invited speake'rs {1992}.
1. National Public Radio can be heard in the wilderness (KUWR/FM 91.9)
 Scientific Organizing Committee
Mitchell Begelman Colin Norman
J ames Binney Paul Shapiro
Alan Heavens J. Michael Shull
David Hollenbach John Stocke
Richard Kron Hyron Spinrad
Shri Kulkarni Harley Thronson
Art Wolfe

Local Organizing Committee

David Barnaby, J. Michael Shull, Harley Thronson
Kim Bella, Theresa Calovini, Darcy Collins, Jim Dove,
Evelyn Haskell, Steve Martin, Marce Mitchum, Steve Penton, and Kara Peterson

Editorial Assistance
Gert Kiers and Michiel Kolman
Kluwer Academic Publishers

Financial and Logistical Support

Department of Physics f3 Astronomy
College of Arts f3 Sciences
Office of the Vice-President for Research
University of Wyoming

Department of Astrophysics) Planetary) f3 Atmospheric Sciences

The Graduate School
The Center for Astrophysics f3 Space Astronomy
University of Colorado

The National Science Foundation

The National Aeronautics and Space Administration

Photography by Harley Thronson

Group photographs by Eric Lufkin and Cristina Thronson

xi
 GASTROPHYSICAL COSMOLOGY

Introduction and Overview

J. RICHARD BOND
CIAR Cosmology Program, Canadian Institute for Theoretical Astrophysics

Abstract. Over the past decade, cosmology has largely focussed on gravitational forces,
or, if gas was included, by linear perturbation calculations or by highly simplified and
symmetric nonlinear models, with some notable pioneering exceptions. We hope that a
Universe dominated by elementary particle relics will be largely understandable in such
simple terms, and, with the COBE observation of CMB anisotropy, there is hope that our
wishes will be realized. We would then know the redshifts when dark matter condensations
of various masses formed. However to understand the formation and evolution of galaxies,
quasars, Lya clouds, the IGM, clusters, and all of the other directly observable beasts that
inhabit the cosmic zoo we must include in full gory detail the effects of the gas as it shock
heats, photoionizes, radiatively cools, suffers fragmentation, forms stars, is transported
outward in galactic winds, and inward in galactic and larger mass mergers; and then
there are magnetic fields. Once complex gas processes are included, calculations become
frustratingly difficult to make definitive, but there is reason for theoretical optimism,
especially with the rapidly growing data-bases to guide us. Conferences like Tetons '92
that bring together primarily ISM types and primarily cosmological types to assess where
we stand in the development of a fully dissipative gas-dynamical theory of cosmc structure
formation are essential for progress. In this overview, I will try to address the scales over
which we think gastrophysical processes will be most relevant and the impact they may
play in our overall cosmological understanding.
Key words: microwave background - cosmology - galaxy formation

1. Spatial Wavebands for Cosmic Density Fluctuations

Given the gravitational instability of the Universe and the high degree of
isotropy of the cosmic background radiation, viewing the development of
cosmic structure through the evolution and coupling of perturbation eigen-
modes is a very instructive approach. On scales much smaller than any global
curvature scale, these eigenmodes are simply plane waves characterized by
spatial wavenumbers k (with the average time-dependent expansion factor
a(t) of the Universe factored out). We may roughly divide the problem of
cosmic structure formation into various wavebands, which we discuss in turn
in the following sections. (We normalize a to be unity now so that comov-
ing wavelengths, 21rk- 1 , are expressed in current cosmic length units. Since
these are estimated from recession velocities, the unit is the h- 1 Mpc, where
h is the Hubble parameter in units of 100 km 8- 1 Mpc- 1 •
• ULSS: The realm of ultra-Iarge-scale-structure corresponds to spatial co-
moving (inverse) wavenumbers k- 1 in excess of a few times the Hubble ra-
dius, cH01 = 3000 h- 1 Mpc. Of course, we only access our 'Hubble patch',
which we think is more likely to be just a tiny bit of the Universe rather
than the bulk of it, although we cannot know what lies beyond nor how

3
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution o/Galaxies, 3-26.
© 1993 Kluwer Academic Publishers.
4

geometrically complex it is. Mean properties of our bit such as curvature

are what we usually misname 'global parameters'.
• VLSS: Over the realm of very-large-scale-structure, from the horizon scale
(k- 1 "" 2cHOl for Einstein - deSitter models) down to say k- 1 "" 100 h- 1
Mpc, density and velocity fluctuations are so small that no influence on
observed cosmic structures will be likely, but gravitational potential per-
turbations are large enough to be observed through their influence on the
cosmic background radiation (CMB) - COBE and other large angle experi-
ments probe VLSS directly. In most models, the fluctuation spectrum shape
is unmodified over its initial shape (the transfer function is approximately
unity)" which further simplifies the interpretation of the VLSS CMB probes
(§2).
• LSS: Over the realm of large scale structure, from"" 100 h- 1 Mpcdown
to about 5h- 1 Mpc, the transfer function does change, and depends upon
the type and density of dark matter, on the values of A, n, H o, etc .. How-
ever, we believe that the evolution of the waves in this band is sufficiently
linear that first-order perturbation calculations of LSS probes may be valid.
Intermediate-angle CMB anisotropy experiments in conjunction with the
VLSS probes will be particularly invaluable for pinning down cosmological
parameters. Building a consistent picture with LSS flow and clustering data
will be a major area over the next few years.
• MSS, SSS, VSSS, USSS: Below LSS lie wavebands for which gas physics
will have been extremely important, if not dominant, in determining the
nature of the objects we see and how they are clustered. Fluctuations are
nonlinear in these regimes. In a hierarchical model, nonlinearity at differ-
ent scales will occur at sufficiently different epochs that we divide the gas-
trophysical realms into medium, small, very small and ultra small, bands
responsible for the construction of, respectively: clusters and groups (""
10 14 - 15 M 0 )i bright galaxies ("" 10 11 - 12 M 0 )i dwarf galaxies and Lyman al-
pha clouds ("" 109 - 10 M 0 )i and the first gas clouds to collapse ("" 106 - 7 M 0 ),
which make the first stars. Of course, significant gas dynamical processing
may obscure the hierarchical relationship between object and primordial
fluctuation waveband.

2. The Ultra-Large Scale Parameters of our Hubble Patch

All that we know of ultra-large structure are single values, averages over
our local Hubble patch that probably tell us little about the entire global
structure of the Universe, in spite of the folklore that n for our patch defines
the global fate. Thus even if we measure an average density jj < PCf"it now,
our ancestors, if they exist in such- a distant future, may learn that jj > Pcrit
over far larger scales than HOI. Our current Hubble patch would then be
 5

a local void expanding into denser ridges, and the whole may eventually
collapse; and beyond this, an ultra-ultra large patch within which our island
ofreheated stuff resides will still be accelerating (i.e., inflating) - according
to the theory of stochastic inflation. The evolution of the substructure in our
Hubble patch and the propagation of photons in it very much depends upon
these mean parameters.
T-y: The relic photon temperature is now well determined, from COBE's FI-
RAS and Gush, Halpern, and Wishnow's COBRA experiments. The FIRAS
team has recently announced T-y = 2.726 ± 0.005 (1 (F error, Mather et al.
1993).
N v : The number of light relic neutrinos is 3, determined within a few percent,
using the CERN LEP and SLAC SLC data on Zo boson decay. With T-y and
a little weak interaction theory, we get Tv, and thus the number density of
light relic neutrinos; the direct detection of this sea seems very unlikely.
fiB: The baryon abundance parameter appears to be well constrained by Big
Bang nucleosynthesis, especially with advances in neutron lifetime measure-
ments and determination of the light neutrino number. The upper bound,
based primarily upon the requirement that 4He is not overproduced, is
fiB ;S 0.064 (2h)-2 (e.g., Olive et al. 1990). If the combination of He 3 +D
is not mostly destroyed during stellar evolution, there is also a lower bound,
fiB .<.0.038 (2h)-2. Steigman and collaborators use Lithium observations to
argue that the error on fiB ~ 0.06 (2h)-2 is under 10%! If nucleosynthesis
was inhomogeneous, fiB could be as high as 0.3; however, the quark-hadron
phase transition would have to have been first order in this theory, and re-
cent lattice gauge theory simulations suggest it wasn't (with the caveat that
this relies upon an ability to scale from very small computational lattices to
predict the continuum result).
Remnant Dark Matter: The allowed range for fiB should be compared with
the rv 0.007 observed in luminous baryons, suggesting at least some baryonic
dark matter exists . The form it takes, whether in Jupiters and brown dwarfs,
in very massive black holes, or in warm gas, is a decidedly gastrophysical
problem.
Relic Dark Matter: Dynamical estimates of the amount of dark matter from
combining the observed mass-to-light ratios in galaxies and clusters with the
observed luminosity density typically give rv 0.1 - 0.3 of the closure density.
Comparison of IRAS galaxy redshift surveys with surveys in which radial
distance is also estimated gives a value nearer unity. Biasing of the galaxy
distribution relative to the dark matter distribution would raise these esti-
mates. Thus, there is no evidence against fi in clustering dark matter being
nearly unity. The most popular candidate for the clustering dark matter
is cold dark matter - massive slowly moving elementary particle relics of
6

the early universe. Most people's best bet for CDM remains the lightest
supersymmetric particle, even though the LEP results from CERN on the
Zo-boson lifetime give surprisingly strong constraints on the properties of
generic massive CDM candidates with interactions that are weak or weaker
than weak. The axion remains as viable as ever.
In the past few years hot dark matter in the form of a light neutrino (in the
ev range and presumably a v r ) got a powerful boost from the proposed MSW
neutrino oscillation solution to the solar neutrino problem, together with the
seesaw mechanism for neutrino mass generation. An astrophysical solution is
difficult to reconcile with the combination of Gallex, Sage, Homestake and
Kamiokande data. (However, the Gallex result is only about 20' from the
standard solar model.) Although v-dominated adiabatic models are strongly
ruled out by CMB limits, v's are the best dark matter for cosmic string
models, and CDM models with a moderate fraction (511.1 '" 0.1-0.3) of light
neutrinos give one of the nicest explanations of large-scale clustering data
in the post-COBE era (§4).
to: Renzini (1992) gives (13-15)±3 Gyr for globular cluster ages, with the
biggest uncertainty (20-25%) coming from the distance modulus, and with
smaller uncertainties from metallicity, helium diffusion, etc. Nuclear cos-
mochronology is more uncertain, with ages of the oldest heavy elements
anywhere from 10 to 20 Gyr being possible. The age of the Milky Way disk
from white dwarf cooling and open cluster ages is about 8-10 Gyr. To these
one must add estimate,s for formation times. A 15 Gyr age for a universe
with 51 ~ 1 in cold matter and baryons - as in the 'standard CDM model'
- would require a Hubble constant below 45; Ho = 65 requires a 10 Gyr
total age.
Ho: All of the 'astronomically-calibrated' methods to determine distance
give, according to most practitioners (e.g., Tonry 1992), values of Ho around
80. What is rather compelling is not each method by itself, but the consis-
tency of the result from Tully-Fisher distances (84 ± 10), planetary nebulae
(80 ± 10) and galaxy surface brightness fluctuations (80 ± 10). Methods
based on 'physics' tend to give lower values. If Type Ia supernovae are stan-
dard candles, they give rv 50 ± 10 (e.g., Sandage et al.1992 calibrated SNla
with a Cepheid in a distant galaxy using the Hubble Telescope and got
46). Kirshner et al. (1992) quote 60 ± 10 for the Baade-Wesselink (mov-
ing photosphere) method for Type II supernovae, but their distances are
consistent with those obtained for individual objects using the 'astronomy'
methods. Combining Sunyaev-Zeldovich and X-ray observations of clusters
(with all of the uncertainties that that entails) gives the theoretically happy
Ho = 40 ± 9 km s-1 Mpc- 1 for Abe11665 (Birkinshaw et al.1991), but SZ/X
estimates of Ho have always been on the low side. Gravitational lensing time
delays have tended to give low Ho but the modelling uncertainties are great.
 7

[Hoto](A, 0): Even with non-zero A or negative curvature, Ho > 70 and a 15

Gyr age lead to a devastating conclusion for cosmology. To emphasize this,
we show what happens to ages when we take the density in nonrelativistic
matter (On!' = OB+Ocdm) to be 0.2 and take OA to be 0.8. Ho = 100 gives 11
Gyr, Ho = 80 gives 13.5 Gyr and Ho = 70 gives 15 Gyr. For these H o , open
universes with 0 = Onr = 0.2 give 8.5 Gyr, 11 Gyr, and 12 Gyr, respectively.
Although it is possible to reconcile the high Ho with lower Onr, (0.1 with
OA = 0.9 gives 16 Gyr for Ho = 80), the amount of clustered dark matter
appears to be in excess of 0.2, and A-energy does not cluster. Intermediate
values for Hoto can be obtained if there is a form of matter present whose
equation of state results in a slower density falloff, intermediate between the
rv a- 2 falloff of curvature energy density and the rv a O falloff of cosmological

constant energy density, with a the expansion factor. Any decline slower
than rv a- 3 requires that the pressure of such matter be negative, but this
is possible if potential energy dominates over kinetic, and is realized, for
example, with scalar fields, and, indeed, is a prerequisite for inflation.
Cosmic Coincidences: As Dicke and Peebles have emphasized, to have non-
zero A or non-zero curvature becoming dynamically important just at the
present epoch would involve a remarkable coincidence, but we have learned
to live with other apparent coincidences, e.g., the nearness of the epochs of
recombination and of the transition from radiation to matter domination;
and the similarity of the dark matter and baryon densities. Anthropic argu-
ments that restrict the values of 0 and A so humans can exist (e.g., requiring
a matter-dominated epoch to ensure sufficient perturbation growth) make
most physicists cringe. Still, one might argue that a molecular-life selection
function has been applied to at least one of the set of all possible accelerated
patches that underwent reheating.
0: The argument that if our Hubble patch is part of a much larger region of
space that inflated its mean curvature would be nearly unity is quite com-
pelling. Curvature is a combination of a squared gradient and the Laplacian
acting on a large scale gravitational potential, a unified picture within which
to view curvature fluctuations and the curvature mean. Thus, I consider the
tininess of fluctuations in the curvature on scales around HOI, as shown by
COBE, to strongly suggest that the mean curvature will also be tiny. Other-
wise we need a spatial cosmic coincidence, a great increase in amplitude for
o
wavelengths beyond H lll - 01- 1 / 2 (something that e.g., double inflation
could, in principle, give). The traditional mathematical picture of a smooth
expanding ball( oon) or saddle of constant mean curvature with tiny ripples
of unrelated origin added as an afterthought was the historical approach,
but it seems contrived to me. In any case 0 doesn't do nearly as well as A
in solving the Hoto problem.
A: Before embracing a A-dominated cosmology, with density parameter
8

OA == A/(87rGPcrit) in excess of 1/3, we should recall that the Steady State

theory was a deSitter space theory with non-zero A. But if we are now in a
(new) period of accelerated expansion, there are no prospects for a reheat-
ing episode like the one that must have ended the last period of inflation, if
indeed there was one: no ongoing matter creation a la Hoyle et al., just a
'cold death' of the Universe. Non-zero A does help explain the LSS (§4).
A non-zero mean scalar field with vacuum (potential) energy density
(V(~)) = (3 x 1O- 12 Gev)4flAh2 can be a superposition of very long ULSS
waves - unlike curvature energy it involves no gradient. The physics case
against A is well known: whatever particle scale we refer it to, this density
is tiny; e.g.,in Planck energy density units (m~ ,. . ., 1094 gcm- 3 ), we have
(V(~));S 1O-122m~flAh2. The 'vacuum' energy density is also expected to
renormalize its value during the cooling of the universe (before the first 3
minutes). Thus, it is not only hard to understand why the effective A is
so small now, but even if it is zero we really have no idea why it should
be so - even with Euclidean wormhole physics. Since inflation relies on
a large effective A generated by the inflaton field to drive the accelerating
expansion (e.g., (V(~)) '" (10 16 Gev)4 in chaotic inflation when LSS waves
were generated), we should not be too surprised if inflation and its Zeldovich
density spectrum are modified when the A problem is finally solved.
Non-zero A or vacuum energy can also be viewed as just another form of
dark matter; for that matter, so can curvature energy, fl cUl' v == 1 - fl.

3. Theories for the Initial Fluctuations Within our Hubble Patch

Quantum Noise: The main paradigm after a decade remains the inflation
one, with quantum zero point oscillations of scalar fields providing the source
of adiabatic (curvature) fluctuations. In the standard stochastic inflation
picture, most of the original volume that got into acceleration would have
passed into deceleration, accompanied by particle production, and eventually
a candidate domain for the bit of the Universe that we see, although by far
the largest current physical volume in the universe would still be undergoing
acceleration. A )..~4 potential invoked in chaotic inflation would need ).. ,. . .,
10- 13 in order that the quantum noise satisfies CMB observations. Such a
tiny ).. is unnatural since, even if one began with such a small coupling, it
would generally become far larger through radiative corrections. There have
been many attempts to construct models with natural, stable, small effective
couplings; e.g., Freese, Frieman, & Olinto (1990) propose a 'natural inflation'
in which the inflaton is the phase of a complex field with a radiatively-
protected effective).. of order (mGUT / mp )4, where mGUT ,....., 1016 Ge V is the
GUT energy scale and mp '" 1019 GeV is the Planck mass.
Power-Law Breaking of Scale Invariance: Exact scale invariance (i.e., n3 = 1,
 9

where the initial rms density perturbations are d(T~( k) / dIn k <X k3+ n .) can-
not occur in inflation. Most models give approximate power laws with n" be-
low unity, but not by much. This gives a little more power in LSS and VLSS
bands. To get more complex spectra (mountains, valleys, plateaus) a spa-
tial cosmic coincidence must be built into the potential surface of the scalar
fields that drive inflation, or special (very unlikely) initial conditions must
be invoked for our patch (e.g., Salopek et al.1989 and references therein).
This is distasteful, but all our VLSS, LSS and MSS observations probe the
structure of only a tiny section of the scalar field potential surface.
Isocurvature Baryon Perturbations: What is difficult to arrange in infla-
tion models is less power than scale invariant spectra give. This and the
requirement that the dark matter would be purely baryonic (in spite of
the !lB ;S 0.064 (2h)-2 nucleosynthesis limit) are the main reasons for not
taking those special isocurvature baryon models that are able (maybe) to
reproduce the cosmic structure we observe very seriously. These models have
initial fluctuations in the baryon number, but no net fluctuation in the curva-
ture, so are orthogonal to the usual adiabatic modes (that e.g., the standard
CDM model presupposes). Nearly scale invariant isocurvature spectra are
very strongly ruled out by CMB constraints. (At least one particle-physics
motivated model exists in which scale invariance on small scales goes over
to the observationally required white noise spectrum on large scales, but a
cosmic coincidence is required for the transformation scale.)
Topological Defects: In spite of inflation's ability to smooth our Hubble patch
and generate fluctuations through quantum oscillations, there is a healthy
skepticism that it is the only path to cosmic smoothness and that the inflaton
potential will have just the right coupling to give the 'observed' perturba-
tions. Among searches for alternative generation mechanisms, topological
defects in field configurations formed during cosmological phase transitions
have been the most promising. In spite of field smoothing through gradi-
ent interactions and radiative losses once there is causal contact, these field
defects do last long enough to generate matter density perturbations be-
fore disappearing. Examples are the 1D cosmic strings and the 3D 'global'
monopoles and textures; 2D domain walls do not lead to viable models.
Defect models generate approximately scale invariant initial fluctuations,
but these are decidedly non-Gaussian and so have unique features for galaxy
formation: in particular, some objects can form shortly after recombination
and result in early reionization of the Universe, which could lower small
angle CMB anisotropies below observability.
Constraints on Ezplosion-induced Structure: Explosion-driven or radiation-
pressure-driven structure formation, amplifying small seed fluctuations, would
release more 'y-distortion' energy through Compton cooling of electrons than
the current strong FIRAS limits allow - at most 0.01 % of the total energy
10

in the CMB at the 95% confidence limit (Mather et al. 1993). Nonetheless,
local explosions and radiation forces still have lots of room to locally amplify
(or deamplify) structure over the UMSS, VSSS, SSS and even MSSS bands.

4. COSInic Fluctuation Spectra and their VLSS and LSS Probes

In Figure l(a), we show the wavebands probed by various large-scale struc-

ture observations (large scale streaming velocities LSSV, the angular corre-
lation of galaxies Wgg(O), the power spectrum and redshift-space correlation
function of galaxies as probed by the QDOT redshift survey, the correlation
function of clusters of galaxies ~cc). The best indicator for large-scale power
is the angular correlation function of galaxies.
The range covered by these LSS probes should be contrasted with the
range covered by microwave anisotropy experiments, each of which can be
well characterized by filters which act upon a 'power spectrum for t1T /T
fluctuations' (see Bond & Efstathiou 1987 for a precise definition). Filter
functions are shown for the COBE (7 0 beam) DMR experiment, the MIT
(3.8 0 beam) balloon experiment of Page et al. (1990), the UCSB 1991 South
Pole (1.5 0 beam) experiment of Gaier et ai. (1992) and Schuster et ai. (1993),
the UCSB 1989 South Pole (0.5 0 beam) experiment of Meinhold & Lu-
bin (1991) and the Caltech OVRO (1.8' beam) experiment of Readhead et
ai. (1989). The balloon-borne UCB/UCSB MAX and Goddard/MIT experi-
ments have filters which cover about the same range as the SP89 experiment
and there is a new Caltech experiment planned to cover the region between
SP89 and OVRO.
Thus CMB anisotropy experiments cover the entire VLSS and LSS bands.
Below rv 5 h- 1 Mpc, primary anisotropies of the CMB (those one calculates
from linear perturbation theory and which are easiest to interpret) are ba-
sically erased if hydrogen recombination is standard (SR line in Fig. 1), so
photons decouple from baryons at z rv 1000 and freely stream to us; if there
is an early injection of energy which ionizes the medium, photon decoupling
would not have occurred until a lower (fiB-dependent) redshift and would
erase t1T /T power on scales typically below the NR (no recombination) line
shown.
The light long-dashed filter curves at smaller scales show the regions
of the spectrum probed by the VLA, by the SCUBA array on the sub-
mm telescope JCMT, by the IRAM mm-dish, and by the OVRO mm-array.
Although their beams are too small to see primary CMB anisotropies, they
will provide invaluable probes of secondary anisotropies (those generated
by nonlinear effects, including redshifted dust emission from galaxies and
Thomson scattering from nonlinear structures in the pregalactic medium.)
Observed power spectra (actually their square roots) are shown as hatched
regions for density fluctuations inferred from COBE and for galaxy fluctu-
 11

at ions inferred from the APM and ROE Wgg data. The long wavelength
hatched curve is the DMR-normalized scale invariant spectrum (assuming
an nnr = 1 model, and including a 30% -e rror budget). The heavy curve
extending the hatched Wgg power into smaller distances is the power cor-
responding to the well known ~gg( r) = (r I rOgg) -"I 3D correlation function
form, where the CfAl redshift survey values have been taken, rOgg = 5.4 h- 1
Mpcand"Y = 1.8. Power spectra derived from the CfA2 (Vogel et al.1992),
QDOT (Kaiser et al. 1991), and mAS 1.2Jy (Fisher et al. 1992) redshift sur-
veys are compatible with the range inferred from Wgg when account is taken
of redshift space distortions and a possible clustering offset between IRAS
and optically identified galaxies (see Davis, these proceedings). Remarkably,
cluster-cluster correlations also seem to be compatible with this spectrum
(Dalton et al. 1992; Nichol et al. 1992), as do galaxy-cluster cross correlations
(Efstathiou 1992).
Power spectra for gravitational potential fluctuations are related to those
for the density through the Poisson-Newton equation V2~ = 47rGa 2 bp; thus
they are flat for scale invariant spectra on large scales (dO"i(k) = k-4dO"~(k)
Q( kn.-1dln k) . Power spectra for large-scale streaming velocities are related

to those in density though the continuity equation (dO"~(k) = k-2dO"~(k)

Q( kn.-1dlnk) .
The power spectra for primary t:J..T IT fluctuations are more complex than
these, because they include effects associated with geometrical ripples in the
past light cone (Sachs-Wolfe effect), with the flow of electrons at photon de-
coupling, the degree of photon compression at decoupling, and the damping
associated with the width of decoupling mentioned above.
In Figure (la), we also show the band of waves whose local construCtive
interference leads in time to collapsed virialized dark matter condensations of
the type shown. Of course, such a characterization presupposes that there ~re
waves of significant amplitudes to form the structures. This will generally be
true for hierarchical models, in which the rms power in density fluctuations in
each waveband is monotonically increasing with decreasing scale. However,
damping processes or tilted initial spectra may require some of the shorter
distance structure to arise from fragmentation and other non-gravitational
effects.
In Figure (lb) , linear density-density power spectra normalized to the
COBE DMR data are compared with the galaxy data for the models in
Table 1. To translate into galaxy-galaxy power spectra to match the second
hatched region, it is usual to multiply by a single constant biasing factor for
the galaxies in question. There are so many orders of magnitude in the Figure
that the LSS ' extra power' problem is not that evident, a strong indicator
that nearly scale-invariant spectra may be on the right track. The insert
Figure (lc) gives a closeup of the galaxy clustering regime, with the (linear)
12

1000

(a) 10 16
cIs gps
~,.......... LSSV
,:,t.

.s tco(r)
'11'.,(1':1)
"0

"
•
b
II
1
"0
.
~,.......
,:,t.
C

"0

"
.. t
b
"0
-..

(b)
____ . COM
_____ HC
_ _ VAC/C
t!. ___ C40
.......
,:,t.
1
.s
"0

"
•
b
II.

.......
"0

0.008.boOl 0.001 0.01 0.1 1

k hMpc- 1

Fig. 1. (a) Cosmic waveband probes; (b) p power spectra; (c) galaxy power spectra.
 13

biasing factor bg = 0";1 now included. This shows that the shape of the
CDM spectrum does not agree with the LSS power. Scale dependent biasing
which suppresses power on small scales could be one way out (Carlberg &
Couchman 1991). However the other models in Table 1 can explain the LSS
power, although, like CDM, each has at least one "Achilles heel" that may
be fatal.
Characteristic Waveband Amplitudes: To discuss differences that various
models of structure formation give over these realm.s, we characterize the
amplitude in the associated wavebands by relative rms fluctuations in the
mass in spheres of radius R (in h- 1 Mpc), assuming linear growth of per-
turbations: O"R == (~M( < R)j lW"),.m$ . The average mass initially within the
sphere is related to R by M ~ 10 12 .4(Rjh- 1 Mpc)3M0.
It has become standard to characterize the amplitude of primordial fluc-
tuations by 0"8. The radius R = 8 h- 1 Mpcroughly divides LSS from MSS.
The observed rms fluctuations in the galaxy distribution are unity on this
scale, but this includes nonlinear and galaxy biasing effects. If the dynamics
is sufficiently linear on this scale, then to agree with the unity observation,
the mass fluctuations would have to be amplified by a factor bg = 0";;1. Thus
0";1 has come to be known as the biasing factor. However, nonlinear effects
could lead to smaller amplifying factors, differences between galaxy types
could imply different biasing factors and complexities associated with the
formation and merging of bright galaxies and other gastrophysical processes
could make the biasing factors functions of environment and of scale. O"R
scales with the normalization 0"8.
Typical values of COBE-normalized O"R'S for the different bands are
shown in Table 1 for some popular post-COBE models of structure forma-
tion. The notation for the models is that of Bardeen et al. 1987 (BBE). All
models but one have an initially scale invariant (n$ = 1) adiabatic spectrum.
CDM denotes the standard model with Ho = 50, !lB = 0.05, !lcdm = 1 - !lB.
A high-0"8 CDM model has problems explaining the relatively quiescent pair
velocities of galaxies on small scales, and overproduces rich clusters (§5).
The redshift znl(R) at which an rms perturbation of scale R reaches
unity, assuming a linear extrapolation of the density evolution, is simply
1 + znl(R) = O"R for CDM Universes, since O"R ex: the scale factor a. This
provides a first (somewhat low) estimate for when dark halos of mass M
form in abundance (a few percent of the mass will be in such halos at
1 + Z '" 1.40"R).
Other models shown in the Table are: VACjC, a A i= 0 model with
Ho = 80 km s-l Mpc- 1 , !lA = 0 .75 and !lcdm = 0.25. For Universes with
A i= 0, O"R evolution slows down once !lA dominates, so 1 + znl(R) will
be higher than O"R; we list these in brackets after the O"R values when the
difference is significant.
HC is a mixed hot and cold dark matter model, with !l" = 0.3 in light
14

TABLE I
Table 1: Characteristic Density Fluctuation Levels
Realm Probes [dM/Ml,.m. CDM VAC/C HC C40

ULSS "Global" 0 0 0 0
VLSS COBE 0"300 0.0050"8 0.012 0.006 0.004
LSS W gg , Pgg(k) 0"25 0.30"8 0.37 0.27 0.19
ea, ecg, Vb .. ,,,
MSS cIs 0"8 0.94 1 0.7 0.6
gps 0"4 1.80"8 1.6 1 1.0
SSS gals, QSOs 0"0.5 60"8 4.3(6} 1.8 3.1
VSSS dG, Ly Q: 0"0.1 110"8 6.8(9} 5.2
USSS 1st stars 0"0.01 210"8 10(14) 9

FLAW Wgg Vb .. ,,, ZgJ? H o, vb",'"

massive neutrinos (mv = 7 ev) and flcd'Tn 0.7 in CDM (van Dalen &
Schaeffer 1992). In BBE, it was argued that HC models with significant
1/ content would not form galaxies early enough, based upon an fiv = 0.4
(mv = 10 eV), flcdm = 0.5, fiB = 0.1 HC model, but it was also shown that
it would give LSS correlations and flows within the current observational
range. The HC model in the Table is not quite as bad, but 0"0.5 may still be
too small; however, one can roughly match the LSS data with fiv as low as
rv 0.1 (mv ~ 3 ev). Thus HC shows much promise. N-body studies described

by Davis and Evrard, these proceedings, support these conclusions.

C40 is a CDM model, but with Ho = 40. It also has a slight spectral index
change, ns = 0.95 as suggested by chaotic inflation, and has fiB = 0.1, the
nucleosynthesis upper limit for Ho = 40. All of these effects conspire to give
just about enough LSS.
For cosmic string, monopole and texture models, the COBE-inspired
value for 0"8 would have to be below 0.3 or so (Bennett, Bouchet, & Steb-
bins 1992; Bennett & Rhie 1992; Pen, Spergel, & Turok 1993). It will not be
easy for defect proponents to concoct a viable model for structure formation
and LSS flows with such an amplitude, although more detailed simulations
are needed to decide. Remarkably, the cosmic texture model in an Ho = 50
CDM universe seems to give a power spectrum with the same shape as that
inferred from LSS clustering observations (Park, Spergel, & Turok 1991).
DMR Likelihood: See Bennett et al., these proceedings, for a detailed discus-
sion of COBE's DMR anisotropy experiment. The strength of the detection
is indicated by the likelihood curve shown in Figure 2 for the standard CDM
model. (This is based on a Bayesian treatment of the '90 A+B X 53 A+B'
correlation function data given by Smoot et al. (1992) and assuming a Gaus-
sian distribution of errors and using a uniform weight all sky approximation
to treat the theoretical variance in C(O).) The relation between 0"8 shown
 15

here and the value of the quadrupole used to normalize the angular power
spectrum is Qrms,pS = 14.90"sJLK, which gives a maximum likelihood value
of 14JLK. This compares with the value derived using the quoted value of
the rms anisotropies on a 10° scale, Qrms,PS = 14.30"sJLK. The observed
quadrupole on the sky is 13.4 ± 5JLK. With the Bayesian analysis used here,
there are about 15% errors (at the one 0" level) on the normalization O"s, while
the errors are about 20% if one just uses the 10° rms anisotropies. Seljak &
Bertschinger (1992) also use a likelihood approach on the correlation data
to derive a value similar to the one I get for Qrms,PS. The DMR team derive
15.3JLK for this '90 A+B X 53 A+B', 9% larger than that given here, with
bigger error bars on their effective O"s, and actually adopt the larger value of
17 JLK for a scale invariant spectrum based upon the analysis in the Wright
et al. (1992) paper. The difference between their result and that shown here
might be explained in part by the influence of Galactic cuts, by their use
of uncorrelated chi-squared rather than Bayesian statistics and correlation
function differences from DMR map to DMR map.
While these effects may cause o"s to be up to 20% higher than the values
shown in the Table, there a number of physical effects which will lower the
value. Effects associated with the gentle breaking of scale invariance which
bring the spectral index of the fluctuations below ns = 1, the influence of
gravity waves on large angle anisotropies, and Ho and fiB modifications
all cause o"s to decrease. For a fairly conservative inflation model, (chaotic
inflation with a A<jJ4 potential), the spectral index in the VSS to LSS regime is
0.95, lowering o"s by 10%; gravity waves further lower it by 10% (Starobinsky
1985; Abbott & Wise 1985, and a plethora of post-COBE papers); using
fiB = 0.05 indicated by primordial nucleosynthesis gives a further lowering
to o"s = 0.75. Models with ns even smaller are certainly possible in inflation
and lead to even more drastic modifications. Lowering Ho to 40 yields a
further 30% lowering. Thus decreases of at least 25% are quite plausible over
the naive COBE-normalized O"s. On the other hand, there are many inflation
models which can give ns quite close to unity and there are some with
negligible gravity wave corrections, so we must currently live with combined
uncertainties of ab " ut 30% in O"s, for 'standard inflationary models'. This is
not that much better than the constraints we had before DMR's observation,
just based upon the ability of the models to form the structure we observe;
but of course now the inflationary models have increased weight. However,
the observational error bars on DMR can go down a factor of 2 when all 4
years of data are analyzed.
Whatever the final resolution of the specific amplitude, the strength of the
detection shown by the likelihood function of Figure 2 will survive. We note
that large-scale streaming velocities of galaxies indicate a high amplitude for
the spectrum (Bertschinger et al. 1990; Efstathiou, Bond, & White 1992).
MIT Likelihood: Also shown in Figure 2 are likelihood curves for the smaller
16

CMS expls vs 08=0.05 n.= 1 h=0.5 CDM

(Q_-ps = 15J.LK 0'8)
3

"0
o
o
:::2
-Q)
- -
9pl(4chs)
.... I
~
:.:3
" ,
, ,II V

9pl(c'N4?
1 "" ,
'" )

Fig. 2. DMR, MIT, SP91 likelihood functions for a standard CDM model.

angle MIT 3.8 0 and the UCSB 1.50 SP91 experiments. A positive cross-
correlation of the MIT map with the DMR maps has recently been reported
by Ganga et al. (1992), indicating that both experiments are seeing the same
pattern of bumps on the sky. I have been analyzing the MIT data using
the Bayesian approach on the complete map (rather than on correlation
functions) and have indeed found a significant detection at the same level
as my DMR correlation function analysis gives, with 27% one <F errors.
Although these curves are plotted against <Fs for the specific CDM model
shown, the same agreement will exist for any scale-invariant model, except
the value of <Fs will differ, as in Table 1. (A caveat here is that NR models
will give slightly different answers for the two experiments, but the errors
are large enough to accommodate them.)
SP91 Likelihood: Smaller angle experiments such as the SP91 experiment
are sensitive to the details of the cosmological model, in particular the value
of fiB, since they probe gas flows at the time of recombination as well as the
 17

gravitational potential fluctuations that the larger angle experiments probe.

It is unclear at the present time how to interpret the anisotropies that are
now observed in this and other intermediate-angle experiments, since the pri-
mordial signal may be contaminated by Milky Way or extragalactic sources,
especially by synchrotron radiation at lower frequencies and Galactic dust
emission at higher frequencies. The data which has been analyzed for SP91
so far consists of one 9-point scan of the sky and one 13-point scan of a
different region, each in 4 channels centred around 30 GHz. If contamina-
tion exists it is more likely to be from synchrotron sources than from dust
at these frequencies . The SP91 likelihood functions of Figure 2 are calcu-
lated assuming that the only signal is a cosmic primary !::J.T IT signal from a
scale invariant CDM model with h = 0.5. When all 4 channels are analyzed
simulataneously, both scans give a maximum to the likelihood, but if chan-
nel 4 of the 9-point scan is analyzed alone, there is no maximum - i.e ., no
indication of a signal, but within large statistical errors. The all-channel
likelihood functions get substantially broader than those shown if we add a
simplified synchrotron radiation signal of Gaussian-distributed white noise
with an angular 'coherence angle' optimally-sized to take away as much of
the signal as possible from the primary !::J.T IT. Although a maximum to the
likelihood remains in the 13-point data, it disappears in the 9-point data,
but again with wide errors. At this point I think that all we can say is that
the likelihood contours for these two scans are not clearly inconsistent with
the CDM model or in the other models of Table 1 (Bond & Efstathiou 1993).
The next year or so will be very exciting as more high precision anisotropy
data probing a variety of resolution scales come in to show us whether we
are converging upon a specific cosmological model.

5. Are Clusters Clean Cosmic Probes? Not Clean, Just Crucial

Clusters aren't simple: Theorists have long dreamed that clusters were simple
deep potential wells, nicely virialized, relatively isolated and spherical, with
a uniform temperature Tx, dark matter velocity dispersion VDM, and galaxy
velocity dispersion Vg all simply reflecting the binding energy per mass of
the beasts. If so, then determining the abundances of clusters above Tx from
X-ray satellite observations, above Vg from galaxy redshifts, or above VVM
from gravitational lensing would fix not only the amplitude of the density
fluctuations on MSS, but also the shapes. Of course, it isn't so.
Clusters and eTs : Nonetheless, cluster abundances do provide an especially
sensitive measure of eTs, the conventional fluctuation normalizer. This is easy
to see using a simple argument from Bardeen et al. (1986) and Efstathiou,
Bond, & White (1992) : an unperturbed sphere ofradius 8 h- 1 Mpccontains
1.2 X 1015nnrh5~ M 0 , and if the region when it virializes is compacted into
18

an average overdensity of 178 relative to the background (as in spherical

top hat collapses), then its final radius is nearly an Abell radius, 1.5 h- 1
Mpc. If '30" peaks' on this scale are virializing now, then 0"8 ~ 0.6 follows if
we set 30"8 ~ 1.69, where 1.69 is the average linear overdensity needed for
a spherical shell to have collapsed to a point. If they virialized at z = 0.2,
then 0"8 ~ 0.7.
BBE give more precise estimates of the number of O"'s required to get the
Abell richness 1 abundance for clusters virializing now using the spherical
approximation for the models in Table 1, among others. We got 0"8 ~ 0.7-
0.8. Frenk et al. got 0"8 '" 0.3-0.5 based on projection corrections of cluster
vg's, Bond & Myers (1990) got 0"8 '" 0.6-0.9 based on the Edge et al. (1990)
sample of X-ray clusters, and White et al. (1992) got 0"8 ~ 0.5-0.6. Carlberg
& Couchman (1992) raised the spectre of 0"8 being in excess of unity, with
the explanation of the paucity of very high velocity dispersion clusters a
result of velocity biasing in which Vg/VDM is substantially below unity.
It is outrageous that the data are still sufficiently murky that one can
still argue about a factor of two in 0"8, since quite small variations in 0"8
dramatically change the predicted number of high Tx and Vg clusters as a
function of redshift. Yet the situation is promising. To have at least one Tx =
14 keV cluster at redshift 0.2 - and one was found by the Ginga satellite
(Arnaud et al.1992) - strongly constrains how low 0"8 can be (provided
complex gastrophysics does not modify the fully virialized assumption). By
contrast, the richest cluster in Abell's catalogue, Abell 665 at z = 0.18, has
a more modest Tx = 8.2 keV. (Its well-observed SZ decrement of tl.T /T =
-1.46 X 10- 4 (Birkinshaw et al.1991) agrees with the X-ray density and
temperature estimates.
The cluster sy~tem could also give the shape of O"R around R = 8 h- 1
Mpc. Indeed, naive modelling of the current X-ray observations supports
a shallower power spectrum than CDM gives over the spatial waveband
responsible for cluster production, not unlike the one the Wgg data suggests,
but this is even more dependent upon uncertain gas dynamical issues.
Fortunately we are now in the midst of a world-wide assault on clus-
ters, with the development of catalogues beyond Abell, using for example,
the APM and ROE Southern Sky surveys (Dalton et al.1992; Nichol et
al. 1992), with new X-ray satellites (ROSAT of course and BBXRT and the
recently launched ASTRO-D which is in many ways ideal for cluster Tx de-
termination), and multi-object spectrographs that can be used to get much
more accurate Vg determinations. Sunyaev-Zeldovich experiments are also
maturing, both in the radio, using the Owens Valley dishes (with COMA
now seen in SZ), the 5-km array in Cambridge, and in the mm and sub-mm,
aboard balloons. Accompanying this is a tremendous burst of theoretical
work modelling with gas dynamics the realistic formation and evolution of
clusters. Although theorists have a long way to go and much parameter
 19

space to cover, we can look forward to well-calibrated cluster models to

wake theorists from their dreams: in a hierarchy, clusters will often have
had a recent major merger, or be in the throes of one, equilibrium may well
not prevail, and the core physics that X-rays probe is complicated by gas
inhomogeneities, cooling flows, metals, magnetic fields, radio galaxies, etc.
Donahue and Mushotzky describe the state of this art in these proceedings.
Clusters and Large Scale Power: Clusters have been one of the main indi-
cators of large scale power since the Bahcall and Soneira and Klypin and
Kopylov 1983 estimations of the correlation length of Abell clusters. BBE
showed how dismally the CDM model fared with this data, but also that HC
and VAC/C models fared much better (but still not enough for the {ee data
of that time). Now the observations of {ee appear to be compatible with the
observations of W gg . Figure 3 shows that the ratio of {ee to egg on large scales
as predicted by the hierarchical peaks model of cluster formation described
by Bond and Myers in these proceedings for a wgg-inspired power spectrum
agrees with the data.
A shape compatibility between {ee and egg, if it holds up, is very signifi-
cant. Nonuniform biasing, bg(r, t), of galaxies relative to the mass distribu-
tion (due, e.g., to gas dynamical, merging or radiative effects) could always
be invoked to explain the excess power in Pgg(k) relative to that in the CDM
model. However, the large biasing factor, be, of clusters is expected to be rel-
atively uniform, since the thresholding criterion - "have you collapsed by
now?" - is a local timing question. Thus biasing factors for galaxies may
actually be relatively uniform too. Further, hierarchical peak theory predicts
[{eg/{ggp/2 = (b e/b g )1/2, which Efstathiou (1992) shows is a good approxi-
mation to his data. On the other hand, we must be wary of overinterpreting
these new {ee's, since the samples are small and the error bars are large.

6. The Gastrophysical Reahns

Gastrophysics is important below about 8 h- 1 Mpc, a band which forms

clusters and also large voids. In Table 1, the redshifts Znl signal when per-
vasive structure forms on those mass scales. For viable post-COBE models,
there is a very large range: in some, galaxy halos would virialize at high
redshift, and in others quite late. And even with specification of the halo
formation epoch, dissipative phenomena greatly impede the translation into
a full theory of galaxy formation.
The Hierarchy with Gastrophysical Processes: If our interests are just to
nail down the parameters characterizing the primordial fluctuations, it is
best to use clean probes, those on large scales where we do not expect
gas dynamics to have had a large effect. This includes CMB anisotropies,
and, we hope, large scale streaming velocities. If biasing on large scales is a
20

10
9
B
7
roa=5.4: cis at z=O

-
~
6
~ BS fractal
1:1 5
<44r
"-t: 4 ED
.......
<44r -I,
II
~
3

-
III
.t:J
"-
c: 2
"""
"
.......
.t:J
~

(°8=0.7.1)
\ \

APM 1 ,,~
\ '\l
r=0.2 e.g. VAC/C \
____ _ . r=0.5 sld COM "
"
10- 0.001
n{>M)

Fig. 3. Cluster-cluster correlation amplitude as a function of cluster abundance for a

power spectrum that reproduces the APM Wgg and {cc data (solid curves, denoted
by r = 0.2 as per Efstathiou, Bond, & White 1992), and also for the standard
CDM model (short-dashed, r = 0.5). The upper curves (at small nez) for each of
the 2 cases have 0"8 = 1, the lower have 0"8 = 1.4; biasing of clusters is so strong
that their clustering is relatively insensitive to the amount of dynamics which 0"8
measures. Al and A3 denote Abell richness 1 and 3 clusters (Bahcall & West 1992,
Postman et al.1992), APM o and APM l refer to clusters found in the APM survey
(Dalton et al.1992) with an algorithm less subject to projection contamination
than Abell's, and ED refers to the Edinburgh-Durham cluster sample found in the
ROE Southern Sky catalogue (Nichol et al.1992). CDM does not have enough LSS
power to explain the shape of {ee . To put {ee relative to egg, a clustering length of
5.4 h- 1 Mpcwas adopted, as in Fig.1. The long-dashed curve is the Bahcall scaling
law for the correlation strength with richness. Although it was motivated by a
phenomenological fractal description, this figure shows it can be fully understood
within Gaussian inflation fluctuation theory.
 21

linear phenomenon, then the correlation functions of galaxies and clusters

for l' ~ 10 h- 1 Mpcshould be linear probes of the mass density fluctuations.
However, such probes might be affected by gas dynamics (e.g., environmental
influences on the Tully-Fisher relation between infrared flux and 21 cm line
widths of spiral galaxies could give false impressions of LSS flows).
Merging: The essence of the hierarchy is merging and this may be the defin-
ing principle for the formation ofthe most interesting objects in the universe,
the great energy releasers like starbursts and AGNs and radio galaxies. That
close encounters will be important for them is suggested by their enhanced
clustering over that of normal galaxies. And even the large scale galaxy dis-
tribution may depend upon the details of how galaxies merge in groups,
since larger groups are more clustered than smaller groups and than gen-
eral galaxies. In purely collisionless computations, galactic mass dark matter
halos lose their identity to larger mass halos as the hierarchy develops. A
challenge to gastrophysicists is to show the extent to which dissipation ar-
rests this overmerging problem for galaxies themselves. Hernquist (these
proceedings) describes the current state of the art on gaseous mergers.
Feedback: Energy output from collapsed objects could enhance 01' damp
structure development. Hierarchical models like the CDM model rely on
feedback to suppress catastrophic star formation at z '" 10 - 20, when gas
clouds of about 106 - 10 7 M0 collapse. Such small clouds are quite frag-
ile, easily unbound by the energy they produce or by energy incident from
outside. Later, at least by z '" 5, we rely on feedback to highly ionize the
IGM via UV photons or shock-heating to explain the Gunn-Peterson effect.
How close the UV output from QSOs and other AGNs alone comes to ex-
plaining this is described by Bechtold and Boyle, these proceedings. A hot
medium can significantly smooth the gas (over a scale k- 1 '" 0.1 h- 1 Mpc,
i.e.,,,, 10 10 M 0 , if the temperature is 2 X 10 4 Kat z = 5), but not the dark
matter, which clusters into deeper and deeper potential wells with larger
and larger virial velocity dispersions until finally gas can begin to fall back
in. Changing UV backgrounds could accentuate this late infall. Effects like
these may have something to do with the mystery of faint blue galaxies, as
described in these proceedings.
Cooling and Fragmentation: The highly naive spherical 'top hat' picture of
galaxy formation embodied in the 'baryon density v.s. virial temperature'
diagrams a la Rees and Ostriker continues to be incredibly influential in
shaping our thoughts - e.g., the folklore that supermassive galaxies don't
form because they can't cool fast enough, although why there is an upper
limit to the luminosities of cD galaxies is still mysterious (e.g., Tremaine
1990). Combined with feedback assumptions and a calculational procedure
for the abundance of objects as a function of Tv and redshift (Press-Schechter
or peaks mass functions), this simple model leads to useful zeroth order pic-
22

tures for how changes in the fluctuation spectrum, in feedback and in star
formation assumptions may affect galaxies at high redshift, as Davis de-
scribes in these proceedings. But there are many many parameters one can
fool with in this game. Refinements include a 2-phase hot/cold (10 6 K/I04K)
protogalactic medium, a natural byproduct of fragmentation through cooling
instab,lities, but hard to include explicitly in full galaxy formation calcula-
tions because of the small size of the cool clouds.
FIRAS Constraints: Why have we not directly detected the waste heat of
gastrophysical feedback? FIRAS now limits the excess energy from 500J-Lm
to 5000J-Lm (the CMB peak is at 1400 J-Lm) to be ~ 3 X 10- 4 of the total in
the CMB, as long as it deviates from a blackbody and does not mimic the
Galactic spectrum fit to the FIRAS data (Mather et al.1993). Even if some
extragalactic component mimicked Galactic dust emission within the errors,
there will still be a strong limit on high-redshift dust emission models.
For example, if the nuclear energy burnt in a population of very massive
stars (VMOs) of density fl. shining at redshift z. were to all find its way
into this band, the FIRAS constraint would be fl.h 2 ~ 10- 5 (1 + z.)/5,
with even more powerful constraints possible from accreting dust-shrouded
black holes, and nearly as powerful ones from '" 20 M0 stars. The heavy
elements generated by dust-shrouded 20 M0 stars radiating into this band
cannot exceed flmetal.,h2 ~ 10- 5 . 4(1 + z.)/5, corresponding to '" 10- 4.3 or
so of metals in bright galaxies. To avoid the conclusion that luminous dust-
shrouded starbursts at high redshift are not that abundant, one could invoke
hot grains, e.g.,50K dust delivers its peak at about (350 J-Lm)(l + zgal)/5, so
FIRAS might be sensitive only to the long-wavelength tail.
Bursting in a dust-free environment is another way out. Since COBE's
DIRBE is plagued by difficult foreground subtractions before it can get to the
residual cosmological signal, near-infrared limits on background light are still
relatively weak; indeed as much (unclustered) light energy can reside there
as in all of the CMB. To have most of the non-luminous dark baryons locked
away in black hole remnants would require such a dust-free environment. Of
course all of these radiant energy limits will be irrelevant if they are locked
up in Jupiters or brown dwarfs.
Polluting Winds: The internal energy generation of big galaxies, especially
of radio and active galaxies and starbursts, can have potentially devastating
effects upon themselves and their environment. While the 1061 ergs released
in these relatively rare seed structures could not trigger a chain-reaction
explosion scenario which generates all of the observed large scale structure
as in the Cowie-Ostriker-Ikeuchi story, because of FIRAS constraints, they
can stimulate star formation over moderate Mpc scales, and high-redshift
radio galaxy observations in fact show that this happens.
The galactic superwinds we know of spew forth kinetic energy, photons
 23

above the Lyman edge, dust and metals in gas form, magnetic fields, and so
on. The galaxy M82 is considered to be the Rosetta Stone for superwindsj
Kronberg and collaborators give evidence for magnetic field transport and
there is also evidence for entrained ISM propagating into the IGM, with
accompanying heavy element pollution. Arp 220 is another of our Rosetta
Stones, the canonical starburst, with a far infrared flux of '" 1045 L 0 , a
similar kinetic energy release in superwinds, yielding a total energy release
of '" 1060 ergs with '" 109 M0 consumed if the starburst has a duration
of about 10 7 - 108 yr. Arp 220 is undoubtedly a merger, probably of two
spirals, a phenomena which should be ubiquitous in hierarchical models -
especially at higher redshift when the merger rate is expected to have been
higher. After all, that is how giant galaxies would have been assembled in
the hierarchy. Heckman et al., in these procedings, describe the powerful
effects superwinds may have had.
Lya Clouds and dGs: Tiny and moderate mass dwarfs, and the IGM gen-
erally, are undoubtedly profoundly affected by feedback, even if the energy
is coming from outside rather than from inside. I have long felt that this is
the key to understanding the dramatic evolution of the LyO' cloud system
- as dwarf galaxies blowing apart. In part this view is coloured by the hope
that LyO' clouds represent a primordial population rather than a secondary
population of no clear significance for constrainting models. However, if the
amplitude 0"0.1 is too high, as in the COBE-normalized CDM model, then
dwarf galaxy scale objects would have virialized at high redshift and then
merged into more massive entities, leaving LyO' clouds to form in other ways,
maybe as remnants of fragmenting shells from galactic winds. An even bigger
puzzle in all models are the faint blue galaxies, a population which dramat-
ically fades from view below z '" 0.3, a redshift of no obvious significance in
the hierarchy.
Elliptical Ages and the Earliest QSOs: But we should contrast the dG puz-
zles that suggest lower values for 0"0.1 with the apparent venerable age of
elliptical galaxies, globular clusters, and the existence of quasars at z > 4,
which all point to significant early activity and thus high 0"0.5' Bruzual, in
these proceedings, shows that with advances in synthetic spectra modelling,
e.g., including the very luminous asymptotic giant branch stars, one still
finds support for elliptical formation at redshifts between 10 and 20. This is
anathema to all of the Table 1 models, even for the low bias factor models
that the COBE result would indicate for the CDM model.
Hydro: Detailed 3D gas-dynamical codes for simulating the formation of
cosmic structures are under active development as Evrard describes in these
proceedings. Unlike pure N-body studies, however, where the large-distance
structure is largely unaffected by the nonlinear evolution on small scales,
gas processes are expected to couple small scales to larger ones. What this
24

means computationally is that, given the finite resolutions that we are ca-
pable of following with 3D codes, we must model by hand what may be
termed sub-grid physics to feed into an upward cascade of influence. Thus,
star formation, supernova explosions, galactic winds, H II regions, ionized
IGM, cooling instabilities and flows, fragmentation, dynamo generation of
magnetic fields in collapsing objects - all of the problems that make ISM
studies so challenging - face us in cosmology now that we have entered the
realms of gastrophysics.
Thus, hydro models in cosmology can at best be illustrative rather than
definitive. Resolution is fundamental for gas dynamics. Although the usual
approach of N-body studies using large boxes with fixed Eulerian grids can
be good in some contexts to determine gross gas behaviour (e.g., Cen &
Ostriker 1992), fully adaptive Lagrangian 3D hydro codes (e.g., smooth par-
ticle hydrodynamics with a variable resolution scale carried by each fluid
element) applied to single formation events are preferable. Even so, the sub-
grid physics will always haunt us. SPH is being used to great effect in studies
of cluster and galaxy formation and merging, as Evrard describes. Gas cal-
culations typically have at most a few hundred thousand particles, although
massively parallel machines promise to give us a leap forwar~ i~ resolution.
(Warren et al. 1992 are running 20 million particle-tree code calculations of
galaxies on the INTEL Delta parallel computer at Caltech, and this tech-
nology can be applied to SPH.) .
In spite of the COBE detection, it remains unclear, for any mass scale,
at which epoch we should turn on our gastrophysical calculations of forming
cosmic structures; and even more so to decide what happens to the gas.
Data must be our guide. And a gathering of observers and theorists steeped
in the lore of the ISM and IGM, quasars and AGNs, galaxy morphology,
mergers and winds, how structure from little stars to great walls forms, and
how COBE fits in, is definitely a forward stride.

Acknowledgem.ents
In this overview of the current state of our cosmological art, referencing is
extremely sporadic to the many who have made important contributions.
Forgive my sins of omission; the rest of the book will do more justice. Col-
laborative work with George Efstathiou, with Ed Cheng, Steve Meyer and
Lyman Page, with Todd Gaier, Phil Lubin and Jeff Schuster, and with Steve
Myers led to Figures 2 and 3.

References
Abbott, L. & Wise, M.: 1984, Nucl. Phys. B244, 541
Arnaud, K. et al.: 1992, preprint
Bahcall, N. & West, M.: 1992, Ap. J. 270, 70
 25

Bardeen, J.M., Bond, J .R., & Efstathiou, G .: 1987, Astrophys. J . 321, 18; BBE
Bennett, D.P., Bouchet, F.R., & Stebbins, A.: 1992, preprint
Bennett, D.P . & Rhie, S.H.: 1992, preprint
Birklnshaw, M., Hughes, J.P., & Arnaud, K.A.: 1991, Astrophys. J. 379, 466
Bertschinger, E., Dekel, A., Faber, S.M., Dressler, A ., & Burstein, D.: 1990, Astrophys. J.
364,370
Bond, J .R. & Efstathiou, G;: 1987, Mon. Not. R . astr. Soc. 226, 665
Bond, ·J .R. & Myers, S.: 1991, Trends in Astroparticle Physi cs, ed. D. Cline & R. Peccei
(Singapore: World Scientific), p .262 '
Cen, R . and .Ostriker, J .P.: 1992, Astrophys. J. 393, 22
Couchman, H.M.P. & Carlberg, R.G .: 1992, Astrophys. J. 389, 453
Dalton, G.D ., Efstathiou, G ., Maddox, S.J., & Sutherland, W .J.: 1992, Astrophys. J . Lett.
390,Ll
Efstathiou, G ., Bond, J .R . , & White, S.D.M.: 1992, Mon . Not. R . astr. Soc . 258, IP;
EBW
Efstathiou, G .. : 1992, Proc. Nat. Acad. Sci. , in press
Fisher, K.B. Davis, M., Strauss, M.A., Yahil, A . & Huchra, J .P .: 1992, Astrophys. J . ,in
press
Freese, K. Frieman, J.A . & Olinto, A.V.: 1990, Phys. Rev. Lett. 65, 3233
Frenk, C .S. , White, S.D.M., Efstathiou, G. & Davis, M. : 1990, Astrophys. J . 351, 10
Ganga , K. et al. : 1992, Proc. Texas/PASCaS Conference
Gaier, T., Schuster, J., Gunderson, J., Koch, T., Seiffert , M., Meinhold, P . and Lubin, P.:
1992, Astrophys. J. Lett. 398, Ll
Kaiser, N., Efstathiou, G., Ellis, R.S ., Frenk, C .S., Lawrence, A., Rowan-Robinson, M., &
Saunders, W .: 1991, Mon. Not. R. astr.Soc. 252, 1
Kirshner, R. et al.: 1992, preprint
Mather, J.C . et al. : 1993, COBE preprint
Maddox, S.J., Efstathiou, G., & Sutherland, W.J.: 1990, Mon. Not. R. astr. Soc. 246, 433
Meinhold, P. & Lubin, P.: 1991, Astrophys. J . Lett 370, 11
Nichol, R.C., Collins, C.A., Guzzo, L . & Lumsden, S.L.: 1992, Mon . Not. R. astr. Soc .
255, ZIp
Olive, K.A., Schramm, D.N., Steigman, G ., & Walker, T.P.: 1990, Phys. Lett. B 236, 454
Page, L.A., Cheng, E.S., & Meyer, S.S.: 1990, Astrophys. J. Lett. 355, Ll
Park, C ., Spergel, D . & Turok, N. : 1991 , Astrophys. J. Lett. 372, L53
Pen, U., Spergel, D. & 1'urok, N. : 1993, preprint
Postman, M., Geller, M ., & Huchra, J.: 1992, Astrophys. J. 384, 404
Readhead, A.C .S. et al.: 1989, Astrophys. J. 346, 556
Renzini, A. : 1992 , Proc. Texas/PASCaS Conference
Sandage, A. et al.: 1992, preprint
Salopek, D.S., Bond, J.R. & Bardeen, J.M. : 1989, Phys. Rev. D . 40 , 1753
Schuster, J. et al. : 1993, preprint
Seljuk, U. & Bertschinger, E .: 1992, preprint
Smoot, G .F. et al.: 1992, Astrophys. J . Lett. 396, Ll
Starobinsky, A.A.: 1985, Soviet Astron. Lett 11, 133
Tonry, J .: 1992, Proc. Texas/PASCaS Conference
Tremaine, S.: 1990, in Dynamics and Interactions of Galaxies, ed . R. Wielen (Springer-
Verlag), p.394
van Dalen, A. & Schaeffer, R .K .: 1992, preprint
Vogeley, M.S ., Park, C ., Geller, M.J . & Huchra, J .P. : 1992, Astrophys. J . Lett. 391, L5
Warren, M.S., Quinn, P .J., Salmon, J .K., & Zurek, W .H.: 1992, preprint
White, S.D.M., Efstathiou, G. & Frenk, C.S .: 1992, preprint
Wright, E.L. et al. : 1992, Astrophys. J . Lett. 396, L13
 RECENT RESULTS FROM COBE
C. L. BENNETT, N. W. BOGGESS, M. G. HAUSER and J. C. MATHER
NASA Goddard Space Flight Center, Laboratory for Astronomy fj Solar Physics,
Greenbelt, MD 20771
G. F. SMOOT
Bldg 50-351, Lawrence Berkeley Labs, Berkeley, CA 94720
and
E. L. WRIGHT
Astronomy Department, UCLA, Los Angeles, CA 90024

Abstract. The Cosmic Background Explorer or GOBEl, NASA's first space miSSion
devoted primarily to cosmology, carries three scientific instruments to make precise mea-
surements of the spectrum and anisotropy of the cosmic microwave background (CMB)
radiation on angular scales glieater than 7 0 and to conduct a search for a diffuse cosmic
infrared background (CIB) radiation with 0.7 0 angular, resolution. The observing strategy
is designed to minimize and ' allow determination of systematic errors that could result
from spacecraft operations, the local environment of the spacecraft, and eIIlissions from
foreground astrophysical sources such as the Galaxy and the solar system. Data from the
Far-InfraRed Absolute Spectrophotometer (FIRAS) show that the spectrum of the CMB
is that of a blackbody of temperature T = 2.73 ± 0.06 K, with no deviation from a black-
body spectrum greater than 0.25% of the peak brightness. Data from the first year of the
Differential Microwave Radiometers (DMR) show statistically significant CMB anisotropy.
The anisotropy is consistent with a scale invariant primordial density fluctuation spectrum
and with the gravitational potential variations required to cause the observed present day
structure. Infrared sky brightness measurements from the Diffuse InfraRed Background
Experiment (DIRBE) provide new conservative upper limits to the CIB. Extensive mod-
eling of solar system and galactic infrared foregrounds is required for further improvement
in the CIB limits.

1. Introduction to the CO BE and Mission Objectives

The observables of modern cosmology include the Hubble expansion of the

universe; the ages of stars and clusters; the distribution and streaming mo-
tions of galaxies; the content of the universe (its mass density, composition,
and the abundances of the light elements); the existence, spectrum and
anisotropy of the cosmic microwave background radiation; and other po-
tential backgrounds in the infrared, ultraviolet, x-ray, gamma-ray, etc. The
purpose of the COBE mission is to make definitive measurements of two
of these observable cosmological fossils: the cosmic microwave background
( CMB) radiation and the cosmic infrared background (CIB) radiation. Since
the discovery of the CMB in 1964 (Penzias & Wilson 1965), many exper-
iments have been performed to measure the CMB spectrum and spatial
1 The National Aeronautics and Space Administration/Goddard Space Flight Center
(NASA/GSFC) is responsible for the design, development, and operation of the GOBE.
Scientific guidance is provided by the GOBE Science Working Group. GSFC is also respon-
sible for the development of the analysis software and for the production of the mission
data sets.

27
J, M Shull and H. k Thronson, Jr. (eds.), The Environment and Evolution a/Galaxies, 27-58.
© 1993 Kluwer Academic Publishers.
28
anisotropies over a wide range of wavelengths and angular scales. Fewer at-
tempts have been made to conduct a sensitive search for a CIB radiation,
expected to result from the cumulative emissions ofluminous objects formed
after the universe cooled sufficiently to permit the first stars and galaxies to
form.
In 1974 NASA issued Announcements of Opportunity (AO-6 and AO-
7) fot new Explorer class space missions. A proposal for a Cosmic Back-
ground Radiation Satellite was submitted by John Mather et al. (1974) from
N ASAj Goddard. The objectives of this mission were: (1) make "definitive
measurement of the spectrum (of the 2.7 K CBR) .. with precision of 10- 4
around the peak ... It will also look for the emission from cold dust clouds
and from infrared galaxies"; (2) "measure the large scale isotropy of the
background radiation ... to a precision of 10- 5 •• • Measurements at several
wavelengths are required in order to distinguish anisotropy in the back-
ground radiation itself from anisotropy due to discrete sources" ; and (3) " ...
search for diffuse radiation in the 5-30 micron wavelength range, expected
to arise from interplanetary dust, interstellar dust, and, in particular, from
the integrated luminosity of very early galaxies . The experiment is designed
to separate these contributions by their spectral and directional properties."
Additional proposals were also submitted for large angular scale microwave
isotropy experiments by Sam Gulkis et al. (1974) from JPL and by Luis
Alvarez et al. (1974) from UC Berkeley. NASA selected 6 investigators and
formed the core of the COBE Science Working Group:

Bennett, C. L. NASA-GSFC DMR Deputy Principal Investigator

Boggess, N. W. NASA-GSFC COBE Deputy Project Scientist
Cheng, E. S. NASA-GSFC COBE Deputy Project Scientist
Dwek, E . NASA-GSFC
Gulkis, S. NASA-JPL
Hauser, M . G . NASA-GSFC DIRBE Principal Investigator
Janssen, M. NASA-JPL
Kelsall, T . NASA-GSFC DIRBE Deputy Principal Investigator
Lubin, P. M . D.C .S.B.
Mather, J. C. NASA-GSFC FIRAS PI & COBE Proj . Scientist
Meyer, S. S. M.I.T.
Moseley, S. H. NASA-GSFC
Murdock , T. L. Gen. Res. Corp.
Shafer, R. A. NASA-GSFC FIRAS Deputy Principal Investigator
Silverberg, R . F. NASA-GSFC
Smoot, G. F . LBL & DCB DMR Principal Investigator
Weiss, R . M.I.T. COBE SWG Chairman
Wilkinson , D. T. Princeton
Wright, E . L. D.C.L.A .

To achieve the full benefit of space observations, a goal of the mission

and instrument design was that CO BE measurements would be limited ul-
 29

timately by our ability to identify and model the various components of the
astrophysical foreground emission, and discriminate between them and the
cosmological emission. This goal drove the design of the mission strategy,
the spacecraft and operations, and the choice of instruments. Basic elements
in the mission strategy were the requirements for highly redundant full sky
coverage and for sufficient time in orbit to achieve necessary sensitivity and
evaluate potential sources of systematic errors in the observations. The in-
struments were designed to measure specific attributes of the cosmological
backgrounds and also, through their complementary spectral coverage, to
enable the modeling and subtraction of foreground emissions. Early descrip-
tions of the mission concept have been given by Mather (1982, 1987) and
by GuIlds et al. (1990).
The three scientific instruments are the Far Infrared Absolute Spec-
trophotometer (FIRAS), the Differential Microwave Radiometers (DMR),
and the Diffuse Infrared Background Experiment (DIRBE). The FIRAS ob-
jective is to make a precision measurement of the spectrum of the CMB
from 1 cm to 100 JLm. The DMR objective is to search for CMB anisotropies
on angular scales larger than 7° at frequencies of 31.5,53, and 90 GHz. The
DIRBE objective is to search for a CIB by making absolute brightness mea-
surements of the diffuse infrared radiation in 10 photometric bands from 1 to
240 JLm and polarimetric measurements from 1 to 3.5 JLm. The FIRAS and
DIRBE instruments art! located inside a 650 liter superfluid liquid helium
dewar.

2. COBE Mission Design & Implementation

A full description of the COBE mission by Boggess et al. (1992) is sum-
marized here. Many papers giving overviews, implications, and additional
detailed information about the COBE have been presented by Mather et al.
(1990b), Mather et al. (1991b), Mather (1991), Janssen & GuIlds (1992),
Wright (1990), Wright (1991a), Hauser (1991a), Hauser (1991b), Smoot et
al. (1991c), Smoot (1991), Bennett (1991), and Boggess (1991).
The need to control and measure potential systematic errors led to the re-
quirements for an all-sky survey and a minimum time in orbit of six months.
The instruments required temperature stability to maintain gain and offset
stability, and a high level of cleanliness to reduce the entry of stray light
and thermal emission from particulates. The control of systematic errors in
the measurement of the CMB anisotropy and the need for measuring the
interplanetary dust cloud at different solar elongation angles for subsequent
modeling required that the satellite rotate.
In near-Earth orbit, the Sun and Earth are the primary continuous sources
of thermal emission and it was necessary to ensure that neither the in-
struments nor the dewar were exposed to their radiation. A circular Sun-
30

synchronous orbit satisfied these requirements. An inclination of 99° and

an altitude of 900 km were chosen so that the orbital plane precesses 360°
in one year due to the Earth's gravitational quadrupole moment. The 900
km altitude is a good compromise between contamination from the Earth's
residual atmosphere, which increases at lower altitude, and interference due
to charged particles in the Earth's radiation 'belts at higher altitudes. A 6
PM ascending node was chosen for the CGBE orbital plane; this node follows
the terminator (the boundary between sunlight and darkness on the Earth)
throughout the year. By maintaining the spacecraft spin axis at about 94°
from the Sun and close to the local zenith, it is possible to keep the Sun and
Earth below the plane of the instrument apertures for most of the year. How-
ever, since the Earth's axis is tilted 23.5° from the ecliptic pole, the angle
between the plane of the CGBEs orbit and the ecliptic plane varies through
the seasons from -14.5° to +32.5°. As a consequence, the combination of the
tilt of the Earth's axis, the orbit inclination, and the offset of the spacecraft
spin axis from the Sun brings the Earth limb above the instrument aper-
ture plane for up to 20 minutes per orbit near the June solstice. During this
period the limb of the Earth rises a few degrees above the aperture plane
for part of each orbit, while on the opposite side of the orbit the spacecraft
goes into the Earth's shadow. In the nominal CGBE orbit the spacecraft's
central axis scans the full sky, though not with uniform coverage, every six
months. The orbital period is 103 minutes, giving 14 orbits per day.
A 3-axis attitude control system was implemented by using a pair of iner-
tia wheels (yaw angular momentum wheels), with their axes oriented along
the spacecraft spin axis. These wheels carry an angular momentum opposite
that due to the spacecraft rotation to create a nearly zero net angular mo-
mentum system. The spacecraft orientation is controlled by three reaction
wheels with spin axes 120° apart in the plane perpendicular to the space-
craft spin axis and by electromagnetic coils (torquer bars) that interact with
the Earth's magnetic field. Earth and Sun sensors (one of each on each of
the three transverse control axes) provide control signals to point the spin
axis away from the Earth and at least 90° from the Sun. Rate damping and
fine resolution attitude sensing are provided by six gyros, one on each trans-
verse control axis and three on the spin axis. Coarse attitude parameters
are calculated by using telemetered data from the attitude control sensors
to produce attitude solutions good to 4 arcmin (1 0"). A fine aspect is de-
termined by using gyro data to interpolate between the positions of known
stars detected in the short wavelength bands of the DIRBE instrument. The
fine aspect solution has an accuracy of 1.5 arcmin (1 0") and is now used in
the analysis of data from all three instruments (Kumar et al. 1991).
The FIRAS instrument, located inside the dewar, points along the spin
axis with its 7° field of view. The three pairs of DMR receivers are spaced
120° apart around the aperture plane of the dewar. Each radiometer channel
 31

measures the difference in sky signal from a pair of horn antennas defining
7° fields of view separated by 60°, each beam being 30° from the spin axis.
The spin causes a short-term interchange of the two beams associated with
a single differential radiometer and thereby gives a modulation of the differ-
ential sky signal at the spin rate. The 0.8 rpm spin rate was chosen to be
fast enough to reduce the noise and systematic errors that could otherwise
arise from radiometer gain and offset instabilities. The DIRBE, also located
inside the dewar, views 30° from the spin axis. The spin allows DIRBE to
measure the emission and scattering by the interplanetary dust cloud over
a range of solar elongation angles for each celestial direction, which aids in
the discrimination and subsequent modelling of zodiacal radiation. DMR
and DIRBE trace out a pattern of epicycles that enable them to scan half
of the sky every day and obtain multiple measurements for each pixel of the
sky.
The dewar is a 650 liter superfluid helium cryostat that kept the FI-
RAS and DIRBE instruments cooled to '"'-'1.6 K. A deployable dewar aper-
ture cover protected the cryogen and permitted calibration and performance
testing of the cryogenic instruments prior to launch. A contamination shield
attached to the inside of the dewar cover protected the DIRBE primary
mirror from particulate or gaseous contamination until ejection of the de-
war cover in orbit. It also protected DIRBE from emission from warm parts
of the cryostat during ground testing. The deployed conical Sun-Earth shield
protects the scientific instruments from direct solar and terrestrial radiation
and provides thermal isolation for the dewar. The shield also provides the
instruments isolation from Earth-based radio frequency interference (RFI)
and from the spacecraft transmitting antenna. The shield was designed to
be flexible and was folded to fit within the Delta rocket fairing for launch.
Contamination covers attached to the Sun-Earth shield were placed over the
DMR horn antennas and were pulled away in orbit by the deployment of
the shield. The deployed solar arrays provide the nominal spacecraft and
instrument power load of 542 Watts.
The CGBE has two omnidirectional antennas, one to communicate with
the Tracking and Data Relay Satellite System (TDRSS), and the other to
transmit data stored on tape recorders directly to the ground. The anten-
nas are located on a mast at the bottom of the spacecraft deployed after
launch. The CGBE has a command and data handling system that stores
and decodes the commands received from the ground, collects data from
the instruments and spacecraft at the rate of 4 kbps, and prepares data
for transmission to the ground. The on-board tape recorders and data sys-
tem allow 24 hours of data to be transmitted to the Wallops Flight Facility
in a single 9 minute pass. The data rate allocations for DIRBE, FIRAS,
and DMR are 1716, 1362 and 250 bps, respectively. The remainder of the
telemetry is assigned to spacecraft subsystems.
32

The COBE, as initially proposed, was to have been launched by a Delta

rocket. However, once the design was underway, the Shuttle was adopted as
the NASA standard launch vehicle. After the Challenger accident occurred in
1986, ending plans for Shuttle launches from the West Coast, the spacecraft
was redesigned to fit within the weight and size constraints of the Delta. The
final CO BE satellite had a total mass of 2,270 kg, a length of 5.49 m, and
a diameter of 2.44 m with Sun-Earth shield and solar panels folded (8.53 m
with the solar panels deployed).
Ground testing of the COBE was necessary to demonstrate that the in-
dividual sub-systems, and ultimately the entire spacecraft and instrument
assembly in its flight configuration, could s~tisfy both the sensitivity and
systematic error requirements. System level tests were performed: to sim-
ulate the space environments, including vacuum and temperature; to de-
termine susceptibility to vibration, acoustic excitation, and acoustic shock;
to quantify electromagnetic interference (EMI) self-compatibility and RFI
susceptibility; to determine the interaction between instruments and space-
craft; to simulate the thermal and power conditions that would occur during
eclipse periods; to test the deployables and moving parts (Sun-Earth shield,
antenna boom, solar panels, dewar cover, FIRAS external calibrator and
moving mirror transport, and DIRBE shutter and chopper); and character-
ize and calibrate the instruments.
Technical papers on various aspects of the COBE have been published.
These include papers on contamination control (Barney 1991); test facility
requirements for thermal balance tests (Milam 1991); design of the dewar
(Hopkins & Castles 1985); optical alignments (Sampler 1990); thermal per-
formance of the dewar prior to launch and in orbit (Hopkins & Payne 1987,
Volz & Ryschkewitsch 1990a, Volz et al. 1990, Volz et al. 1991a & 1991b, Volz
& Dipirro 1992); thermal design of the cryogenic optical assembly (Mosier
1991a, 1991b); cryogenic cool-down tests (Coladonato et al. 1990), and at-
titude control (Bromberg & Croft 1985).

3. COBE Operations in Orbit

The COBEwas launched aboard Delta rocket No. 189 at 1434 UT on Novem-
ber 18, 1989 from the Western Space and Missile Center at Vandenberg Air
Force Base, California. The DMR receivers began operating the day after
launch. The dewar cover was ejected three days after launch, and the FI-
RAS and DIRBE instruments began obtaining data on the same day. During
the first month in orbit, various tests were undertaken to evaluate the per-
formance of the instruments and spacecraft, and to optimize instrument
parameters.
The CO BE operated in a routine survey mode. The three instruments
completed their first full sky coverage by mid-June 1990, and returned high
 33

quality data until the depletion of the liquid helium at 0936 UT September
21,1990. The FIRAS, which had surveyed the sky 1.6 times, ceased operating
when the helium ran out, but the DMR is still operating normally in all of
its six channels. By November 1991 (over one year after helium depletion)
the dewar temperature at the DIRBE detectors was about 50 K. The six
longest wavelength bands were turned off in September 1990, but the four
short wavelength bands of the DIRBE continue to acquire data at reduced
sensitivity. The detector system responsivity in the short wavelength bands
decreased about an order of magnitude following cryogen depletion (largely
due to the change in load resistance). However, sky maps of the large scale
interplanetary dust signals are of adequate quality to permit searching for
evidence of temporal changes on annual time scales.
In flight, the helium temperature inside the main cryogen tank was 1.40
K and the temperature of the inner surface of the Sun-Earth shield was 180
K. As expected, the Earth limb rose a few degrees above the Sun-Earth
shield for a part of every orbit during a three month period starting in May.
At these times, the Earth's radiation produced thermal transients in the
instruments and adversely affected data for a portion of each orbit. Some
of these data are still usable after careful calibration. One of the gyros for
a transverse control axis failed electrically on the fourth day after launch.
On September 7, 1991, one of the three gyros on the spin axis failed, but no
data were lost and satellite operations continue in the nominal orbit.

4. Spectral Results From FIRAS

4.1. THE SPECTRUM OF THE PRIMEVAL RADIATION

The discovery of the cosmic microwave background radiation by Penzias &

Wilson (1965) provided strong evidence for Big Bang cosmology. Radiation
produced in the very early universe was frequently scattered until about
300,000 years after the Big Bang. At this point, the "recombination", the
characteristic energy in the universe fell to the point where previously free
electrons could combine with nuclei to form neutral atoms. The 2.7 K ra-
diation we see today has been traveling to us unimpeded since that time.
The rapid production and destruction of photons within the first year after
the Big Bang forced the radiation to have a Planck (blackbody) spectrum.
Any mechanism that injected energy into the Universe (e.g. a particle de-
cay) between a year after the Big Bang and", 2000 years after the Big Bang
would give rise to a radiation spectrum characterized by a non-zero chemical
potential. Thus there would be a Bose-Einstein spectral distortion with the
photon occupation number,

N (E) '" -,---~1-=-=-_

e(e-J.L)/kT - 1 '
(1)
34

where J.L is the chemical potential and E is the photon energy. A Compton
distortion is usually parameterized in terms of a Compton y-parameter,

(2)

where O'T is the Thomson scattering cross-section and the integral is the elec-
tron pressure along the line of sight. A Compton distortion of the spectrum
can become important when (1 + z)dyjdz > 1, which occurs'" 2000 years
after the Big Bang. The thermodynamic temperature distortion observed at
a frequency v is
bT ~ y (a: em +1_4) (3)
T em - 1 '
where a:
= hvjkTcMB, h is the Planck constant, and k is the Boltzmann
constant (Sunyaev & Zeldovich 1980). After recombination it becomes nearly
impossible to distort the CMB spectrum short of reionizing the universe.
Thus a perfect Planck CMB spectrum would support the prediction of the
simplest Big Bang model of the universe, while spectral distortions would
indicate the existence of more complicated releases of energy.

4.2. THE FIRAS INSTRUMENT

The FIRAS instrument is a polarizing Michelson interferometer (Mather
1982, Mather et al. 1991a) with two separate spectral channels. The low
frequency channel, extending from 0.5 mm to 1 cm, was designed to obtain
a precision comparison between the' CMB spectrum and a Planckian cal-
ibration spectrum. The objective was to attain, in each 5% wide spectral
element and each 7° pixel, an accuracy and sensitivity of vI", ~ 10- 9 W
m- 2 sr-I, which is 0.1% of the peak brightness of a 2.7 K blackbody. The
high frequency channel, with a useful spectral range from 0.12 rnm to 0.5
mm, was designed to measure the emission from dust and gas in our galaxy
and to remove the effect of galactic radiation on the measurements of the
CMB made in the low frequency channel.
The FIRAS uses a multimode flared horn (Mather, Toral, & Hernmati
1986) with a 7° beam. The instrument directly measures the difference be-
tween the sky signal in its beam and that from a temperature-controlled
internal reference body. The best apodized spectral resolution is 0.2 cm- 1
(6 GHz). The in-orbit absolute calibration of FIRAS was accomplished by
inserting an external blackbody calibrator periodically into the mouth of the
horn. The calibrator is a precision temperature-controlled blackbody, with
an emissivity greater than 0.999, and bolometric detectors (Mather 1981,
1984a,b) are used in both bands.
In ten months of cryogenic operation the FIRAS obtained over two million
interferograms. This complete data set is now undergoing careful analysis.
 35

4.3. FIRAS RESULTS

Analysis of the FIRAS data to date confirm the prediction of the simplest
Big Bang model that the CMB must have a thermal spectrum. Initial results
based on only nine minutes of data showed that there is no deviation from a
blackbody spectrum Bv(T) as large as 1 % of the peak brightness (Mather et
a1. 1990a, 1991a) over the spectral range from 500 J-Lm to 1 cm. The temper-
ature of the CMB in the direction of the north galactic pole is 2.735 ± 0.060
K, where 60 mK is the initial conservative uncertainty in the calibration
of the thermometry of the absolute calibrator. These data also ruled out
the existence of a hot smooth intergalactic medium that could emit more
than 3% of the observed x-ray background. The thermal character of the
CMB spectrum was subsequently confirmed by Gush, Halpern, & Wishnow
(1990), who obtained virtually the same temperature over the spectral range
2-30 cm- l . Neither mean CMB temperature quotedabove are corrected for
the dipole distortion. These experiments found no submillimeter excess as
previously reported by Matsumoto et a1. (1988b).
More recently, Shafer et al. (1991) and Cheng et al. (1991a) have ex-
amined FIRAS spectra in a direction known previously to be very low in
interstellar material (1 = 142°, b = 55°). In this direction, known as Baade's
Hole, the temperatura is 2.730 ± 0.060 K and there is no deviation from a
blackbody spectrum greater than 0.25% of the peak brightness. The lack of
deviations from a Planck spectrum translate to a limit on a chemical poten-
tial (see eqn 1) of IJ-L/ kTI < 0.005 (95% CL) and a limit on the Compton
y-parameter (eqn 2) of y < 0.0004 (95% CL). These results rule out a hot
smooth intergalactic medium that could emit more than 1 % of the observed
x-ray background.
The dipole anisotropy of the CMB, presumed due to our peculiar motion
relative to the Hubble flow, can be seen clearly in the FIRAS data, and is
consistent with previous results (Cheng et a1. 1990). The FIRAS data show
for the first time that the difference in spectra between the poles of the dipole
is that expected from two Doppler-shifted blackbody curves. This result also
indicates that the stability of the FIRAS instrument is better than one part
in 5000 over long time scales. The dipole amplitude measured by FIRAS is
3.31 ± 0.05 mK in the direction (l,b )=(266° ± 1°, 47.5° ± 0.5°).
FIRAS results also include the first nearly all-sky, unbiased, far infrared
survey of the galactic emission at wavelengths greater than 120 JLm (Wright
et a1. 1991). Wright et a1. present a map of the dust emission across the
sky from the COBE FIRAS experiment. They write the absolute Galac-
tic emission intensity in the form J(v, I, b) = g(v)G(I, b), where g(v) is the
mean spectrum of the emission and G(l, b) is a dimensionless map. The dust
36

component of the mean spectrum is given as

gd(lI) = 0.00016 ( iI/30 cm- 1)1.65 B v (23.3 K) , (4)

or,

gd(lI) = 0.00022 (il/30 cm- 1 f [Bv(20.4 K) +6.7Bv (4.77 K)] , (5)

where iI is the frequency in units of cm- 1 , and Bv(T) is the Planck function.
The total far infrared luminosity of the Galaxy is inferred to be (1.8 ± 0.6) X
10 10 L 0 (Wright et al. 1991).
Wright et al. (1991) report that spectral lines from interstellar [C I],
[C II], [N II], and CO are detected in the mean galactic spectrum, g(lI). The
lines of [C II] at 158 JLm and [N II] at 205.3 JLm were sufficiently strong to
be mapped. This is the first observation of the 205.3 JLm line. Wright et al.
interpret the [C II] line as coming from photodissociation regions and the
[N II] lines as partially arising from a diffuse warm ionized medium and par-
tially arising from dense H II regions. Petuchowski & Bennett (1992) agree'
with this conclusion and further elaborate on it by apportioning the [C II]
and [N II] transition line intensities among various morphologies of the in-
terstellar medium. Petuchowski & Bennett (in preparation) have conducted
observations on N,ASA's Kuiper Airborne Observatory to measure the scale
height of the 205.3 JLm [N II] line with a much higher angular resolution (",1
arcmin) than FIRAS.

5. DMR: Microwave Anisotropy Measurem.ents & Interpretations

Primordial gravitational potential fluctuations at the surface of last scatter-
ing give rise to the distribution and motions of galaxies and to large angular
scale fluctuations in the CMB (Sachs & Wolfe 1967). ill inflationary mod-
els of cosmology (Guth 1981, Linde 1982, Albrecht & Steinhardt 1982) the
gravitational energy fluctuations arise from quantum mechanical fluctua-
tions from 10- 35 seconds after the Big Bang that inflate to become classical
fluctuations with a nearly scale invariant power spectrum (Bardeen, Stein-
hardt, & Turner 1983, Guth & Pi 1982, Hawking 1982, Starobinskii 1982).
fufiation theories are not yet sufficiently constrained to be able to make ac-
curate predictions of the amplitude of current energy fluctuations. Rather,
estimates of the energy fluctuations inferred from observations are a con-
straint on the theories.
The large angular scale CMB temperature anisotropy t1T and gravita-
tional potential fluctuations at the surface of last scattering t1 ~ are simply
related by 3t1T /T = t1 ~ / c 2 for adiabatic fluctuations in a universe with
no cosmological constant (Ao = 0). On much smaller scales than the DMR
measures (0 < 4°), i.e. on scales in causal contact with one another after the
 37

universe became matter dominated, the gravitational potential fluctuations

are affected by the growth of structures through gravitational instability
(e.g., Bond & Efstathiou 1987). Usually the mass density is written as

p(x, t) = p(l + 6(x, t)) , (6)

where 6(x, t) describes the spatial density fluctuations . Density fluctuations
are often considered in terms of their spectrum by comoving wavenumber,
k. The Fourier relations are

C()
uk = J dx
(21r )3 e
-ik-x 6 ( x ) , (7)

Newtonian gravitational potential fluctuations are related to density fluctua-

tions through the Poisson equation, V2 c) = 41rG p, and its Fourier transform
c)/e = -41rGpa 26/e/k 2, where a is the cosmological scale factor relating the
physical scale size r to the comoving scale size x, r = a(t)x.
A "transfer function", T( k), relates the initial primordial density fluctu-
ations at the epoch ti to those observed at the present epoch, to: 6(k, to) <X
T(k)6(k, ti). By convention, on large angular scales (small k) T(k) = 1 for
Ao = 0 and no = 1 cosmologies. That is, the fluctuations on the largest
angular scales are primordial and unaffected by any physical evolution since
there was never sufficient time for causal contact on this scale. The statistics
of the primordial density fluctuations, 6(k, ti), are described by a primor-
dial power spectrum, the Peebles-Harrison-Zeldovich (Peebles & Yu 1970;
Harrison 19~O; Zeldovich 1972) spectrum,

(8)

where the angle brackets represent a spatial average over a large volume of
the universe. For this spectrum the rms Sachs-Wolfe potential fluctuations
(and the resulting CMB temperature anisotropies) as a function of angle 0
are tlc),.m8/c2 = 3tlT,.rn8/T <X o(1-n)/2 for no = 1 and 0:::4 0 (Sachs & Wolfe
1967). Note that P(k, to) = T2(k)P(k,td = AknT2(k). The scale invariant
value n = 1 gives gravitational potential fluctuations with an rms amplitude
that is independent of scale size and a large angular scale CMB temperature
fluctuation spectrum that is approximately independent of the separation
angle. For no < 1 and Ao = 0 the above expression is still approximately
true for angles 0 < no/(l - n o)1/2.
Measurements of the abundances of the light elements together with nu-
cleosynthesis calculations imply that 0.011 :::; nBh 2 :::; .0.037 (Walker et a1.
1991; Olive et al. 199.0), where nB is the fraction of the critical mass density
(Pc = 3H;/81rG = 1.88h 2 x 1.0- 29 gm cm- 3 ) in baryons and h = H o /1.o.o km
s-1 Mpc- 1. Inflation requires that no + Ao/3HJ = 1 so that either Ao f; 0,
or inflation theory is incorrect, or most of the mass in the universe is yet
38

to be detected nonbaryonic material. It is useful to assume that this non-

baryonic material does not interact with light. This simultaneously explains
why it is not seen, and allows it to begin clustering while the universe was
radiation dominated, earlier than is possible for the baryonic matter. The
nonbaryonic material is broadly categorized as "hot" or "cold" dark matter,
depending on whether it was or was not relativistic when the universe be-
came matter-dominated. A neutrino with mass is a favorite hot dark matter
candidate.
Bond & Efstathiou (1984) calculated the transfer function of the "stan-
dard cold dark matter model" assuming TO,CMB = 2.7 K, OB ~ OCDM,
and three massless neutrinos with To,1/ = 1.9 K, giving

T(k) _ 1 (9)
- [1 + (ak + (bk )3/2 + (ck )2)1/]1/1/ '
where a = 6.4/(Ooh 2 ) Mpc, b = 3.0/(Ooh 2 ) Mpc, c = 1. 7/(Ooh 2 ) Mpc, and
v = 1.13. The scale size corresponding to the time when eDM and radiation
have equal energy densities is lO/(Ooh2) Mpc. Holtzman (1989) presents the
results of calculations of T( k) for 94 cosmological models.
A successful model of cosmology and the evolution of structure must
match the amplitude and spectrum of density fluctuations from the galaxy
scale to the horizon scale. Several observables have been derived from galaxy
surveys, including the two-point correlation function, the amplitude of its
integral, the rms mass fluctuation in a fixed radius sphere, and rms galaxy
streaming velocities. The two-point correlation functio'u is defined by ~(x) =
(bp(X' + x)bP(x')/p2), where ~(x) is simply the Fourier conjugate of the
power spectral density

() = f.
P k
dx -ikx (I I)
(21r)3 e ~ x , (10)

The integral of the two-point correlation function is J 3(R) == foR ~(x )x 2dx
(Peebles 1981). Based on the efA redshift survey Davis & Peebles (1983) find
J 3(10h- 1 Mpc)::::; 277h- 3 Mpc 3 and J 3(25h- 1 Mpc) ::::; 780h- 3Mpc 3. Here,
J 3 relates directly to the power spectral density of fluctuations according to

J 3(R) = 41r f P~k) [sinkR - krcoskR]dk, (11)

where the J 3 definition assumed a spatial top-hat sampling or window func-

tion that results in the weighting function 3( kR)-3 (sin kR - kR cos kR) in
k-space. Galaxies are not necessarily distributed in the same way as the
mass density fluctuations. In general a linear proportionality is assumed,
(b plp)g = b (bpi p)p where the constant b is the "biasing factor". The two
point correlation function then scales as ~g = b2 ~p. The rms density fluctua-
tion in a sphere of radius 8h- 1 Mpc is 0"8 ::::; lib (Peebles 1982). The radius
 39

8h- 1 Mpc is chosen because the rms fluctuation of the galaxy distribution is
unity at this radius. Note that, if the biasing factor is a constant (indepen-
dent of angular scale), ratios of the rms density fluctuations over different
scale sizes, such as Ci81 Ci25 are independent of b. Kaiser et a1. (1990) find
that blng· 6 = 1.16 ± 0.21 based on redshift surveys of IRAS galaxies so it
is likely that b < 2. The rms of the peculiar velocities of galaxies averaged
over a sphere of radius R is another observable that relates directly to the
presently observed spectral power density by

(v 2(R)) ~ 367rH~n~/7 JP(k)(sinkR (k~~6cOSkR)2 dk. (12)

Bertschinger et a1. (1990) derive average galaxy velocities in a sphere of

given radii centered on the local group of v(R=4000 km s-l)= 388 ± 67 km
s-l and v(R= 6000 km s-l)= 327 ± 82 km s- l Thus measurements of ~, J 3 ,
Ci8, and (v 2 ) help to determine P(k) today.
Peacock (1991) finds that the power spectrum that is a best-fit to several
independent observations of galaxy clustering is

P(k) __1_ (kjkoY' (13)

- 47rk31+(klkc )-f3'
where a = 1.6, f3 = 2.4, ko 0.19 h Mpc- 1 , and kc ~ 0.025 h Mpc - 1 .
More recently, in light of the new COBE results summarized below, Peacock
prefers kc ~ 0.033 h Mpc- 1 (private communication).
Hence, measurements of large scale (i.e., primordial) CMB anisotropies
can provide the observational link between the production of gravitational
potential fluctuations in the early universe and the observed galaxy dis-
tributions and velocities today. Large-scale CMB anisotropy measurements
provide both the amplitude and the power spectrum of the primordial fluctu-
ations. Large scale anisotropy measurements are usually expressed ill terms
of a multipole expansion and a correlation function. The multipole expansion
of the CMB temperature as a function of sky location is
I
T(O,</J) = L L alrnYirn(O,<p) , (14)
I m=-l
where Yim( 0, <p) are the spherical harmonic functions. Since DMR is a dif-
ferential experiment, as are almost all anisotropy experiments, the I = 0
monopole term is not observed. (It is observed by FIRAS.) The I = 1 dipole
term is also dropped since it is dominated by the Doppler effect due to our
local peculiar velocity and not by cosmic perturbations . Thus the I = 2
quadrupole term is the first term of interest . We are at liberty to select
any coordinate system we choose. Since galactic emission dominates the sky
signal, we choose galactic coordinates, with the usual galactic coordinate
40

angles 1 and b. We define Q(I, b) to be the 1 = 2 term of equation (14)

and rewrite the five Yi=2 ,m components:

Q ( 1, b) = Q 1 (3 sin2 b - 1) / 2 + Q2 sin 2b cos 1 + Q 3 sin 2b sin 1 +

Q4 cos 2 b cos 21 + Q5 cos 2 b sin 21 , (15)

where the rms quadrupole amplitude is

2 1 r 2 4(3 2 2 2 2 2
Qrrns = 411" J47r Q (l, b) dO = 15 "4Q1 + Q2 + Q3 + Q 4 + Q5) , (16)

There is a small kinematic quadrupole, Qrms = 1.2 JLK, from the second or-
der terms in the relativistic Doppler expansion (Peebles & Wilkinson 1968),
for which (Qb Q2, Q3, Q4, Q5)=(0.9, -0.2, -2 .0, -0.9, 0.2) JLK .
The measured correlation function determines the parameters of the fluc-
tuation power spectrum. The correlation function is

C(a) = L ~T?W(l)2 Pl(cos(a)) , (17)

l>l

where Pl are Legendre polynomials, and a 3.20 rms Gaussian beam gives a
weighting W(I) = exp[-~(l(l + 1)/17.8 2)]. Thus,

~T? = ~
411"
L lal m l2
1n
(18)

are the rotationally-invariant rms multipoie moments. As with the spherical

harmonic expansion, the 1 = 0 is excluded from the correlation function
since it is not measured by differential instruments, and the 1 = 1 term
is excluded because it is contaminated by the kinematic dipole. The 1 =
2 quadrupole term is sometimes excluded since the quadrupole has only
21 + 1 = 5 degrees of freedom, and thus has an intrinsically high statistical,
or "cosmic" variance, independent of the measurement . For a power law
primordial fluctuation spectrum the predicted moments, as a function of
spectral index n < 3, are given by Bond & Efstathiou (1987):

< ~T2 > = (Q )2(21 + 1) r(1 + (n - 1)/2)r((9 - n)/2) .

(19)
l rms 5 r(1 + (5 - n)/2)r((3 + n)/2)

For n = 1 this simplifies to

< ~T2 > _ (Q )2 ~ 21 + 1 (20)

l - rms 5 -=-1(:-::-1-+-1--:-)

Smoot et al. (1991b) presented preliminary DMR results based on six

months of data. Smoot et al. (1992) describe results based upon the first year
of DMR data, Bennett et al. (1992a) describe the calibration procedures,
 41

Kogut et al. (1992) discuss the treatment of systematic errors, and Bennett
et al. (1992b) discuss the separation of cosmic and Galactic signals. Wright
et al. (1992) compare these data to other measurements and to models of
structure formation through gravitational instability. Previously published
large-angular-scale anisotropy measlirements include Fixsen et al. (1983),
Lubin et al. (1985), Klypin et al. (1987), and Meyer et al. (1991). Some
excellent reviews of CMB anisotropy and cosmological perturbation theory
include Bertschinger (1992), Efstathiou (1990), Kolb & Turner (1990), Pee-
bles (1971, 1980), and Wilkinson (1986).

5.1. THE DMR INSTRUMENT & DATA PROCESSING

The CGBE DMR instrument is described by Smoot et al. (1990). DMR

operates at three frequencies: 31.5, 53 arid 90 GHz (wavelengths 9.5, 5.7,
and 3.3 mm), chosen to be near the minimum in Galactic emission and
near the CMB maximum. Wright et al. (1990) have used the FIRAS and
DMR data to show that the ratio of the galactic emission to that of the
CMB reaches a minimum between 60 and 90 GHz. There are two nearly
independent channels, A and B, at each frequency. The orbit and pointing
of the CGBE result in a complete survey of the sky every six months while
shielding the DMR from terrestrial and solar radiation (Boggess et al. 1992).
The DMR measures the difference in power received between regions of
the sky separated by 60°. For each radiometer channel a baseline is sub-
tracted and the data are calibrated. Data are rejected when the limb of the
Earth is higher than 1 ° below the Sun/Earth shield plane, when the Moon is
within 25° of a beam center, when any datum deviates from the daily mean
by more than 50", or when the spacecraft telemetry or attitude solution is
of poor quality. Small corrections are applied to remove the estimated emis-
sion from the Moon and Jupiter in the remaining data. CorrectionS' are also
applied to remove the Doppler effects from the spacecraft's velocity about
the Earth and the Earth's velocity about the solar system barycenter. A
least-squares minimization is used to fit the data to spherical harmonic ex-
pansions and to make sky maps with 6144 nearly equal area pixels using a
sparse matrix technique (Torres et al. 1989; Janssen & Gulkis 1992). The
DMR instrument is sensitive to external magnetic fields. Extra equations
are included in the sparse matrix to allow these magnetic susceptibilities to
be fit separately as a linear function of the Earth's field and the radiometer
orientation. The magnetic corrections are on the scale of 10 to 100 JlK in
the time-ordered data. Residual uncertainties in the individual radiometer
channel maps, after correction, are typically 2 J-LK and never more than 8.5
J-LK.
Kogut et al. (1992) have searched the DMR data for evidence of residual
systematic effects. The largest such effect is the instrument response to an
42

external magnetic field. Data binned by the position of the Earth relative to
the spacecraft show no evidence for contamination by the Earth's emission at
the noise limit (47 ILK at 95 % CL). The contribution of the Earth's emission
to the maps is estimated to be less than 2 ILK. The time-ordered data with
antenna beam centers more than 25° away from the Moon are corrected
to an estimated accuracy of 10% (4 ILK) of the lunar flux. The estimated
residual effect on the maps is less than 1 ILK. Kogut et al. list upper limits
for the effects of variations in calibration and instrument baselines, solar and
solar system emissions, RFI, and data analysis errors. The quadrature sum
of all systematic uncertainties in a typical map, after corrections, is < 8.5
ILK for rms sky fluctuations, < 3 ILK for the quadrupole and higher-order
multipole moments, and < 30 ~K2 for the correlation function (all limits
95% CL).

5 .2 . THE DMR ANISOTROPY

The DMR maps are dominated by the dipole anisotropy and the emission
from the Galactic plane. The dipole anisotropy (6.T jT ;:::: 10- 3 ) is seen
consistently in all channels with a thermodynamic temperature amplitude
3.36 ± 0.1 mK in the direction 1 = 264.7° ± 0.8°, b = 48.2° ± 0.5°, consistent
with the FIRAS results, above .. Our motion with respect to the CMB (a
blackbody radiation field) is assumed to produce the dipole anisotropy, so
the dipole and associated;:::: 1.2 ILK rms kinematic quadrupole are removed
.from the maps.
The DMR instrument noise and the intrinsic fluctuations on the sky are
independent and thus add in quadrature to give the total observed signal
variance
2 2 2
O"obs = O"DMR + O"Sky· (21)

The O"obs is estimated from the two channel (A+B)j2 sum maps, and the (A-
B )/2 difference maps provide an estimate of O"DMR, yielding the sky variance
O"sky(100) = 30 ± 5 ILK for Ibl > 20°. The observations are made with a 7°
beam, and the resulting maps are smoothed with an additional 7° Gaussian
function, resulting in the effective 10° angular resolution.
The correlation function, C ( lX), is the average product of temperatures
separated by angle lX. It is calculated for each map by rejecting all pixels
within the Galactic latitude band Ibl < 20°, removing the mean, dipole, and
quadrupole from the remaining pixels by a least squares fit, multiplying all
possible pixel pair temperatures, and averaging the results into 2.6° bins.
Bennett et al. (1992b) conclude that the galactic contribution to the corre-
lation signal is small for Ibl > 15°. This is consistent with the fact that the
correlation function and rms sky fluctuation are insensitive to the Galactic
latitude cut angles so long as Ibl < 15° is excluded. The DMR correlation
 43

functions exhibit temperature anisotropy on all angular scales greater than

the beam size (7°) and differ significantly (> 7(1) from the flat correlation
function due to receiver noise alone.
All six channels show a statistically significant quadrupole signal. A com-
parison of the fitted quadrupoles between channels and frequencies, and
between the first and second six months of data, shows that inllividual
quadrupole components, Qi, typically differ from map to map by ~ 10 ILK
with comparable uncertainty. Determination of the cosmic quadrupole is
linked to its separation from Galactic emission (Bennett et al. 1992b), sum-
marized below. Discrete extragalactic sources individually contribute less
than 2 ILK in the DMR beam and the expected temperature variations are
less than 1 ILK (Franceschini et al. 1989).

5.3 . SEPARATION OF GALACTIC SIGNALS & THE COSMIC QUADRUPOLE

The DMR anisotropy maps are sufficiently sensitive and free from system-
atic errors that our knowledge of Galactic emission is a limiting factor in
interpreting the measurements of the I-year DMR maps. The detected sig-
nals expressed in thermodynamic temperature are nearly constant ampli-
tude: the rms fluctuations on a 10° scale are proportional to V- O. 3 ±l and
the quadrupole and correlation functions IX v- O.2±l. The flat spectral index
of the DMR anisotropy, without correction for Galactic emissions, is con-
sistent wit~ a cosmic origin and inconsistent with an origin from a single
Galactic component . However, from this fact alone we are unable to rule
out a correlated superposition of dust, synchrotron, and free-free emission
and thus more detailed galactic emission models are required. Bennett et
al. (1992b) constructed preliminary models of microwave emission from our
Galaxy based on COBE and other data for the purpose of distinguishing
cosmic and Galactic signals.
Four emission components are important at microwave wavelengths. The
anisotropies in the CMB are assumed to produce differences in the mea-
sured antenna temperature according to !::l.TA =!::l.T x 2 ex j(e x _1)2, where
x = hv j kT, T is thermodynamic temperature. Synchrotron emission arises
from relativistic electrons accelerated by magnetic fields . Free-free emission
occurs when free electrons are accelerated by interactions with ions. Thermal
emission from dust is also important at microwave wavelengths.
The brightest pixels in the DMR maps are T A =5.9 ± 0.4 mK at 31.5
GHz, 1.9 ± 0.2 mK at 53 GHz, both at (l,b)=(337°, _1°), and 1.3 ± 0.2 mK
(348°,+1°) at 90 GHz. Galactic plane emission would have to be removed to
better than 1 % to reveal cosmologically interesting fluctuations in the CMB
at low Galactic latitudes, so our preliminary models concentrate on Ibl > 10° .
44

The intensity of synchrotron radiation is given by the integral

l(v) = JJ P(v, B, E) N(E, I) dE dl ,

where P(v, B, E) is the power emitted at a frequency v by a single electron of

energy E in a magnetic field B, and N(E, l)dE is the number of relativistic
electrons of energy E per unit volume at the position I along the line of
sight. Since N(E, 1) and B(I) are not known for every position in the Galaxy,
approximations must be made to model synchrotron emission. Bennett et
al. (1992b) use the local electron spectrum in conjunction with radio data
to approximate the synchrotron integral.
Free-free emission is characterized by an antenna temperature that de-
pends only on the emission measure and electron temperature along the line
of sight for each frequency, wifh a spectral index

f3f f = 2 + [10.48 + 1.5 In (Te/8000 K) - In VGHzt 1 ,

Here, f3 is the spectral ~ndex of the antenna temperature, TA ex: v-/3, which
relates to the flux density, S(v), by S(v) ex: 2kTA(V)v 2 /c 2 • With Te = 8000
K (Reynolds 1985), the ratio of the 53 GHz free-free antenna temperature to
Ha intensity can be expressed as TA(J-lK)/I(Rayleigh) = 0.83 J-lK/0.44 Rayleigh
or 2 J-lK /Rayleigh, almost independent of the electron temperature. Reynolds
(1984, 1992) has observed several high-latitude lines .of sight with a 0.8°
beam and reports that the high latitude Ha diffuse Galactic distribution is
I(R) :::::; 1.2 csc Ibl for Ibl > 15°. Deviations from a cosecant-law are larger
than the:::::; 15% Ha measurement uncertainties (Reynolds 1992).
Another estimate of the high latitude emission measure comes from the
COBE FIRAS measurement of the N+ ground state transition at 205 J-lm
(Wright et al. 1991). The observed intensity of this line is I(N+) :::::; (7 ±
2) X 10- 8 csc Ibl erg S-l cm- 2 sr- 1 for Ibl > 15° compared with I{N+) :::::;
2.4 X 1o- 8 csclbl erg s-l cm- 2 sr- 1 predicted by Reynolds (1992) for the
diffuse component. In doing Ha background measurements, Reynolds picks
locations that are free from discrete sources, while the CO BE observations
do not exclude sources. The COBE observations may also include emission
with velocities outside of Reynolds' Ha passband. If we assume that the
excess observed N+ arises from the full sky (unbiased) sampling by FIRAS
and any Ha bandwidth exclusions by Reynolds, then we can use the ratio
of measured-to-predicted N+ to correct the diffuse free-free predictions for
these effects. We deduce from the above that a correction factor of ",3 is
required; the free-free emission is then approximately TA(J-lK) :::::; 7 csc Ibl for
Ibl > 15° at 53 GHz. This prediction is only approximate since it depends
on assumptions of the chemical abundance of nitrogen and its fractional
population in the N+ state. Our factor of ",3 to convert from diffuse to full
 45

sky-averaged Ha intensity is consistent with the Reynolds (1992) a priori

estimate of a factor of "'2. Unfortunately, there exists no full sky survey
of free-free emission to the sensitivity required by the COBE DMR, so the
DMR 31.5 GHz map must serve this purpose. We compare this with the
cosecant-law predictions discussed above and find good agreement.
Along each line of sight the observed dust emission is the sum over the
emission from each dust grain. Since the grain temperatures, emissivities,
and spatial distributions are not known, it is not possible to make a priori full
sky dust emission models. We must rely, instead, on full sky measurements of
the dust intensity at higher frequencies and extrapolate these with empirical
spectral fits. The dust antenna temperature is TA = 33 G(1, b) R~ J-LK , where
R~=(0.43, 1.0, 2.3) at (31.5, 53, 90 GHz) and G(l, b) is the dimensionless
map of the dust distribution from FIRAS, discussed earlier. G(l, b) ~ 1
in the plane of the Galaxy and ranges from 0.03 to 0.1 at high Galactic
latitudes.
Bennett et al. (1992b) present three approaches to modeling the Galactic
emission signal in the DMR maps. A "subtraction technique" makes use of
external data to subtract Galactic emission maps from the DMR maps. The
dust and synchrotron emission models, described above, were subtracted
from the DMR maps. Since no map exists of the ionized component of our
Galaxy, the 31 GHz residual map is used to subtract the free-free emission
from the 53 and 90 GHz DMR maps. The angular autocorrelation functions
of the individual emission components show that the Galactic components
have different angular correlations than the residual cosmic signal. In fact,
the cosmic correlation function is largely unaffected by the Galactic model
subtraction. A "fitting technique" directly fits the DMR maps pixel-by-pixel.
The results of the fit are maps of free-free and Planckian emission. A "com-
bination technique" uses a linear combination of the DMR 31.5, 53, and 90
GHz maps to minimize the Galactic emission without recourse to Galactic
models or other data. The subtraction, fitting, and combination techniques
produce consistent results. The combination technique is independent of the
fitting and subtraction techniques, aside from the use of DMR data and
the assumed free-free spectral index. Bennett et al. conclude that no known
Galactic emission component or superposition of components can account
for most of the observed anisotropy signal. In the absence of significant ex-
tragalactic source signals or systematic errors, as argued above, this signal
must be intrinsic to the CMB radiation.
DMR maps, with the modeled Galactic emission removed, are fit for a
quadrupole distribution. Bennett et al. derive a cosmic quadrupole, corrected
for the expected kinematic quadrupole, of Qrm$ = 13 ± 4 ILK, (IlT /T)Q =
(4.8± 1.5) X 10- 6 , for Ibl > 10 0 • When Galactic emission is removed from the
DMR data, the residual fluctuations are virtually unaffected and therefore
they are not dominated by any known Galactic emission component(s).
46

5.4. INTERPRETATION OF THE DMR ANISOTROPY

The anisotropy detected by the DMR is interpreted as being a direct result

of primordial fluctuations in the gravitational potential. Assuming a power
spectral density of density fluctuations of the form P( k) = Ak n , the best-fit
results are n = 1.1 ± 0.5 with Qrm8- PS = 16 ± 4 JtK . Here, Qrm8-PS is the
rms quadrupole amplitude resulting from this power spectrum fit, i.e. making
use of fluctuation information from all observed angular scales,. as opposed
to the Qrm8 derived from a direct quadrupole fit . Forcing the spectral index
to n = 1 gives Qrm8-PS = 16.7±4 JtK and increases the X 2 from 79 to 81 for
68 degrees of freedom. Interpreted as a power-law spectrum of primordial
fluctuations with a Gaussian distribution, the !::J.Tl in each horizon have a
X2 distribution of 21 + 1 degrees of freedom, giving a cosmic variance for
observations within a single horizon volume in the universe of 2 < !::J.Tl >2
/(21 + 1). Best fit values are n = 1.15~g::~ and Qrm8-PS = 16.3 ± 4.6 JtK
including the cosmic variance, with a X2 of 53. Cross-correlation of the 53
GHz and 90 GHz maps are consistent with a power law spectra with index
n = 1 ± 0.6 and amplitude Qrms-PS = 17 ± 5JtK, including cosmic variance.
The observed cosmic quadrupole from the maps [Qrms = 13 ± 4JtK from
Bennett et al. 1992b (see above)] is slightly below the mean value pre-
dicted by the higher-order moments deduced from the correlation function
(Qrms-Ps = 16 ± 4JtK) . This is a likely consequence of cosmic variance:
the mode of the X2 distribution is lower than the mean. A map quadrupole
value of 13 JtK or lower would be expected to occur 35% of the time for
an n = 1 universe with Qrms-PS = 16 JtK. The results above exclude the
quadrupole before computing C(a). Including the quadrupole in computing
C(a) increases the X2 , raises n to 1.5, and decreases Qrms-PS to 14JtK.
The measured parameters [O"Sky(100), Qrms, Qrms-PS, C(a), and n] are
consistent with a Peebles-Harrison-Zeldovich (scale invariant) spectrum of
perturbations, which predicts Qrms = (1~g:~)Qrms-Ps and O"Sky(100) =
(2 .0 ± O.2)Qrms-PS . The theoretical 68% CL errors take into account the
cosmic variance due to the statistical fluctuations in perturbations for our
observable portion of the Universe. The minimum Qrm8 for models with an
initial Peebles-Harrison-Zel'dovich perturbations, normalized to the local
large-scale galaxy streaming velocities, is predicted to be 12 JtK, indepen-
dent of the Hubble constant and the nature of dark matter (Gorski 1991;
Schaefer 1991).
These observations are consistent with inflationary cosmology models.
The natural interpretation of the DMR signal is the observation of very large
(presently ~ 100 Mpc) structures in the Universe which are little changed
from their primordial state (t ~ 1 sec) . These structures are part of a power
law spectrum of small amplitude gravitational potential fluctuations that
on smaller length scales are sources of the large scale structure observed
 47

in the Universe today. The DMR data provide strong support for gravita-
tional instability theories (Wright et a1. 1992). Wright et al. compare the
94 cosmological models for which Holtzman (1989) has computed a trans-
fer function, T( k), with the DMR anisotropy results. None of the Holtzman
isocurvature models are compatible with the DMR anisotropy amplitude for
a biasing factor b < 4. Wright et al. find that three Holtzman models fit the
observational data (galaxy clustering, galaxy streaming velocity, and CMB
quadrupole amplitude) reasonably well. These models are described below.
A model with vacuum energy density with Ovae = A o/3HJ = 0.8, Ho =
100 km S-1 Mpc- 1, OB = 0.02, OCDM = 0.18 is an excellent fit to the
observational data (see, e.g. Gorski, Silk, & Vittorio 1992; Efstathiou, Bond,
& White 1992; and Peebles 1984, 1991).
A "mixed dark matter" (MDM) model that fits the data uses both hot
dark matter (a massive neutrino with OHDM = 0.3) and cold dark matter
(OCDM = 0.6) with baryonic dark matter OB = 0 .1 and Ho = 50 km
S-1 Mpc- 1. See, for example, Efstathiou, Bond, & White (1992), Davis,
Summers; & Schlegel (1992), and van Dalen & Schaefer (1992) for further
recent discussions of mixed dark matter models.
An open universe model with 0 0 = 0.2, OB = 0.02, and OCDM = 0.18
for Ho = 100 km s-1 Mpc- 1 satisfies the observations, except perhaps for
the galaxy rms peculiar velocities, but is in conflict with the inflation model
and theoretical prejudices for 0 0 = 1 (see Gouda & Sugiyama 1992; Peebles
1984, 1991).
The unbiased standard cold dark matter is in conflict with galaxy cluster-
ing data, even without the constraint of the COBE data (e.g. Vogeley et al.
1992; Loveday et a1. 1992). Hogan (1991, 1992, 1993) and Hoyle & Burbidge
(1992) interpret the COBE-DMR results in terms of models where the tem-
perature anisotropies do not arise from gravitational potential fluctuations
on the surface oflast scattering. Bennett & Rhie (1992) interpret the DMR
data in terms of global monopoles and textures .
In summary, the COBE detection of CMB temperature anisotropy has
added another important observational piece of knowledge to the cosmic
puzzle. There is not yet a clear favorite among the models that attempt to
account for all of the pieces, nor is there likely to be one without further
observational information.

6. The DIRBE Experiment

6 . 1. THE COSMIC INFRARED BACKGROUND RADIATION

The Diffuse Infrared Background Experiment (DIRBE) is the first space

experiment designed primarily to measure the CIB radiation. The aim of
the DIRBE is to conduct a definitive search for an isotropic CIB radiation,
48

within the constraints imposed by the local astrophysical foregrounds.

Cosmological motivations for searching for an extragalactic infrared back-
ground have:. been discussed in the literature for several decades (early pa-
pers include Partridge & Peebles 1967; Low & Tucker 1968; Peebles 1969;
Harwit 1970; Kaufman 1976). B~th the cosmic redshift and reprocessing of
short-wavelength radiation to longer wavelengths by dust act to shift the
short-wavelength emissions of cosmic sources toward or into the infrared.
Hence, the wide spectral range from 1 to 1000 JLm is expected to contain
much of the energy released since the formation of luminous objects, and
could potentially contain a total radiant energy density comparable to that
of the CMB.
The CIB radiation has received relatively little attention in the theo-
reticalliterature compared to that devoted to the CMB (Negroponte 1986),
which has a central significance to Big Bang cosmology and quite distinctive
and definite predictions as to its character. However, advances in infrared
instrumentation, and especially the introduction of cryogenically cooled in-
frared instruments on space missions, have stimulated increasing attention
to prediction of the character of the CIB radiation (Fabbri & Melchiorri
1979; Bond, Carr, & Hogan 1986; McDowell 1986; Fabbri et al. 1987; Fabbri
1988; Bond, Carr, & Hogan 1991). Measurement of the spectral intensity
and anisotropy of the CIB radiation would provide important new insights
into intriguing issues such as the amount of matter undergoing luminous
episodes in the pregalactic Universe, the nature and evolution of such lumi-
nosity sources, the nature and distribution of cosmic dust, and the density
and luminosity evolution of infrared-bright galaxies.
Observing the CIB radiation is a formidable task. Bright foregrounds from
the atmosphere of the Earth, from interplanetary dust scattering of sunlight
and emission of absorbed sunlight, and from stellar and interstellar emis-
sions of our own Galaxy dominate the diffuse sky brightness in the infrared.
Even when measurements are made from space with cryogenically cooled in-
struments, the local astrophysical foregrounds strongly constrain our ability
to measure and discriminate an extragalactic infrared background. Further-
more, since the absolute brightness of the CIB radiation is of paramount
interest for cosmology, such measurements must be done relative to a well
established absolute flux reference, with instruments which strongly excludf
or permit discrimination of all stray sources of radiation or offset signah
which could mimic a cosmic signal.
Hauser (1991b) lists recent experiments capable of making absolute sky
brightness measurements in the infrared (for a compilation including some
earlier measurements, see Negroponte 1986). Instruments or detector chan-
nels designed specifically to measure that part of the spectrum dominated
by the CMB radiation have been excluded. Murdock & Price (1985) flew an
absolute radiometer with strong stray light rejection on a sounding rocket
 49

in 1980 and 1981. Their primary objective was measuring scattering and
emission from interplanetary dust, and no attempt was made to extract
an extragalactic component. Matsumoto et al. (1988a) flew a near-infrared
experiment on a rocket in 1984. They have reported possible evidence for
an isotropic residual near 2 JLm, perhaps in a line feature, for which they
cannot account in their models of emission from the interplanetary medium
and the Galaxy. This gr~up has flown a modified instrument early in 1990
to investigate further this result (Noda et al. 1992). The IRAS sky sur-
vey instrument, though not specifically designed for absolute background
measur~ments, was, within the limits of long term stability, capable of good
relative total sky brightness measurements, and so is included in this list. Un-
certainties in the IRAS absolute calibration have impeded efforts to extract
an estimate of the CIB radiation (Rowan-Robinson 1986). The FIRAS-high
frequency channel (100 to 500 JLm), with its all-sky coverage, excellent stray
light rejection, absolute calibration, and high sensitivity, also promises to
be an important instrument for cm radiation studies. Quantitative com-
parison of the measurements from the experiments duscussed above, and a
summary of current CIB radiation limits are discussed further below.

6.2. THE DIRBE INSTRUMENT

The experimental approach is to obtain absolute brightness maps of the
full sky in 10 photometric bands (J[1.2 JLmj, K[2.3 JLmj, L[3.4 JLmj, and
M[4.9 JLmjj the four IRAS bands at 12, 25, 60, and 100 JLmj and 140 and
240 JLm bands). To facilitate discrimination of the bright foreground con-
tribution from interplanetary dust, linear polarization is also measured in
the J, K, and L bands, and all celestial directions are observed hundreds
of times at all accessible angles from the Sun in the range 64 0 to 124 0 •
The instrument rms sensitivity per field of view in 10 months is AI ( A) =
(1.0, 0.9, 0.6, 0.5, 0.3, 0.4, 0.4, 0.1, 11.0,4.0) x 10- 9 W m- 2 sr- 1 , respec-
tively for the ten wavelength bands listed above. These levels are generally
well below both estimated CIB radiation contributions (e.g., Bond, Carr, &
Hogan 1986) and the total infrared sky brightness.
The DIRBE instrument is an absolute radiometer, utilizing an off-axis
Gregorian telescope with a 19-cm diameter primary mirror. Since the DIRBE
was designed to make an absolute measurement of the spectrum and angu-
lar distribution of the diffuse infrared background it must have extremely
strong rejection of stray light. The optical configuration (Magner 1987) has
strong rejection of stray light from the Sun, Earth limb, Moon or other off-
axis celestial radiation, or parts of the COBE payload (Evans 1983j Evans
& Breault 1983). Stray light rejection features include both a secondary
field stop and a Lyot stop, super-polished primary and secondary mirrors, a
reflective forebaffle, extensive internal baffling, and a complete light-tight en-
50

closure of the· instrument within the COBE dewar. Additional protection is

provided by the Sun and Earth shade surrounding the COBE dewar, which
prevents direct 'illumination of the DIRBE aperture by these strong local
sources. The DIRBE instrument, which was maintained at a temperature
below 2 K within the dewar as long as the helium was present, measures ab-
solute brightness by chopping bE!tween the sky signal and a zero flux internal
reference at 32 Hz using a tuning fork chopper. .
The synchronously demodulated signal is averaged for 0.125 second. before
transmission to the ground. Instrumental offsets are measured by closing a
cold shutter located at the prime focus. All spectral bands view the same
instantaneous field-of-view, 0.7° X 0.7°, oriented at 30° from the spacecraft
spin axis. This allows the DIRBE to modulate the angle from the Sun by 60°
during each rotation, and to sample fully 50% of the celestial sphere each day.
Four highly-reproducible internal radiative reference sources can be used to
stimulate all detectors when the shutter is closed to monitor the stability
and linearity of the instrument response. The highly redundant sky sampling
and frequent response checks provide precise photometric closure over the
sky for the duration of the mission. Calibration of the photometric scale
is obtained from observations of isolated bright celestial sources. Careful
measurements of the beam shape in pre-flight system testing and during the
mission using scans across bright point sources allow conversion of point-
source calibrations to surface brightness calibrations.
The data obtained during the helium temperature phase of the mission
are of excellent photometric quality, showing good sensitivity, stability, lin-
earity, and stray light immunity. Few artifacts are apparent other than those
induced by energetic particles in the South Atlantic Anomaly and variations
in instrument temperature. Both of these effects will be removed in final data
processing. Strong rejection of off-axis radiation sources is confirmed by the
absence of response to the Moon (which saturates the response in all detec-
tors when in the field of view) until it comes within about 3° of the field of
view. The sensitivity per field of view, listed above, is based on noise mea-
sured with the shutter closed and response determined from measurements
of known celestial sources. The noise when the shutter was open is somewhat
above the shutter-closed values due to discrete source confusion.
The nuclear radiation environment in orbit caused very little response
change « 1 %) in all detectors except the Ge: Ga photo conductors used at
60 and 100 J1.m. Thermal and radiative annealing procedures applied to
these detectors following passages through the South Atlantic Anomaly will
allow response correction to about 1 % at these wavelengths. It is expected
that fully reduced DIRBE sky maps will have photometric consistency over
the sky better than 2% at each wavelength, nearest neighbor band-to-band
( color) brightness accuracy of 3% or better, and absolute intensity scale
accuracy better than 20%.
 51

6.3. DIRBE RESULTS

Preliminary results of the DIRBE experiment have been described by Hauser

et al (1991), Hauser (1991a, 1991b). Qualitatively, the initial DIRBE sky
maps show the expected character of the infrared sky. For example, at 1.2
JLm stellar emission from the galactic plane and from isolated high lati-
tude stars is prominent . Zodiacal scattered light from interplanetary dust
is also prominent. These two components continue to dominate out to 3.4
JLm, though both become fainter as wavelength increases. A composite of
the 1.2, 2.3, and 3.4 JLm images was presented by Mather et al. (1990b).
Because extinction at these wavelengths is far less than in visible light, the
disk and bulge stellar populations of the Milky Way are dramatically appar-
ent in this image. At 12 and 25 JLm, emission from the interplanetary dust
dominates the sky brightness. As with the scattered zodiacal light, the sky
brightness is strongly dependent upon ecliptic latitude and solar elongation
angle. At wavelengths of 60 JLm and longer, emission from the interstellar
medium dominates the galactic brightness, and the interplanetary dust emis-
sion becomes progressively less apparent . The patchy infrared cirrus noted
in IRAS data (Low et al. 1984) is evident at all wavelengths longer than
25 JLm. Weiland et al. (1991) gave a preliminary description of the emission
from the interplanetary dust bands, and Murdock et al. (1991) have used
the DIRBE data to carry out a preliminary examination of the ecliptic pole
emission. The DIRBE data will clearly be a valuable new resource for stud-
ies of the interplanetary medium and Galaxy as well as the search for the
CIB radiation.
In searching for the extragalactic infrared background, the most favorable
conditions are directions and wavelengths of least foreground brightness .
In general, because of the strong interplanetary dust foreground and the
relatively modest gradient of that foreground over the sky, the infrared sky
is faintest at high ecliptic latitude. A preliminary DIRBE spectrum of the
sky brightness toward the south ecliptic pole was presented by Hauser et al.
(1991), and is reproduced in Table I.
This table shows the strong foreground from starlight and scattered sunlight
at the shortest wavelengths, a relative minimum at 3.4 JLm, emission dom-
inated by interplanetary dust peaking around 12 JLm, and generally falling
brightness from there out to submillimeter wavelengths.
To meet the cosmological objective of measuring the CIB radiation, the
foreground light from interplanetary and galactic sources must be discrim-
inated from the total observed infrared sky brightness. This task requires
extensive careful correlation studies and modelling, which in the case of the
DIRBE investigation is in progress. A conservative upper limit on extra-
galactic light is the total observed brightness in a relatively dark direction.
The sky brightness at the south ecliptic pole is a fair representation of the
52

TABLE I
Cosmic Infrared Background Limits
Reference ,\ AI)..
(JLm) (10- 7 W m- 2 sr- 1 )
DIRBE 1.2 8.3 ± 3.3
(South Ecliptic Pole) 2.3 3.5 ± 1.4
3.4 1.5 ± 0.6
4.9 3.7 ± 1.5
12. 29. ± 12
22. 21. ± 8
55. 2.3 ± 1
96. 1.2 ± 0.5
151. 1.3 ± 0.7
241. 0.7 ± 0.4

best current limits from the DIRBE. The faintest foregrounds occur at 3.4
J-Lm, in the minimum between interplanetary dust scattering of sunlight and
re-emission of absorbed sunlight by the same dust, and longward of 100 J-Lm,
where interstellar dust emission begins to decrease. Through careful mod-
elling, we hope to be able to discriminate isotropic residuals at a level as
small as 1 percent of the foregrounds. These near-infrared and submillimeter
windows will allow the most sensitive search for, or limits upon, the elusive
cosmic infrared background.
These data are to be compared with the theoretical estimates of contri-
butions to the CIB radiation from pregalactic and protogalactic sources in
a dust free universe (Bond, Carr, and Hogan 1986, Carr 1988). The present
conservative observational limits are beginning to constrain some of the the-
oretical models at short infrared wavelengths, though in a dusty universe
energy from these sources can be redistributed farther into the infrared. If
the foreground components of emission can confidently be identified, the
current CO BE measurements will seriously constrain (or identify) the CIB
radiation across the infrared spectrum. However, the spectral decade from
about 6 to 60 J-Lm will have relatively weak limits until measurements are
made from outside the interplanetary dust cloud.
The CIB radiation promises to enhance our understanding of the epoch
between decoupling and galaxy formation. The high quality and extensive
new measurements of the absolute infrared sky brightness obtained with
the DIRBE and FIRAS experiments on the CO BE mission promise to al-
Iowa definitive search for this elusive background, limited primarily by the
difficulty of distinguishing it from bright astrophysical foregrounds.
 53

7. COBE Data Products and Plans

Extensive data products from the COBE mission consisting of calibrated
maps and spectra with associated documentation are planned. The CO BE
databases have been described by White & Mather (1991). An overview of
the CO BE software system has been given by Cheng (1991). All COBE data
processing and software development for analysis take place at the Cosmol-
ogy Data Analysis Center (CDAC) in Greenbelt, MD, a facility developed
by the COBE project for that purpose. This facility, and the software tools
developed there, will become available to the scientific community when the
data products are released.
Initial data products are planned for release in mid 1993. Galactic plane
maps, including the nuclear bulge, will be available at all 10 DIRBE wave-
lengths and the high frequency FIRAS band. Full sky maps from all six
DMR radiometers will also be available.
Full sky maps from all three COBE instruments, spaIUling four decades
of wavelength, are planned for release in mid-1994. These data gathered
by the COBE's three instruments will constitute a comprehensive data set
unprecedented in scope and sensitivity for studies of cosmology, and large
scale galactic and solar system science.

Acknowledgements
The authors gratefully acknowledge the contributions to this report by their
colleagues on the COBE Science Working Group and the other participants
in the CO BE Project. Many people have made essential contributions to
the success of COBE in all its stages, from conception and approval through
hardware and software development, launch, flight operations, and data pro-
cessing. To all these people, in government, universities, and industry, we
extend our thanks and gratitude. In particular, we thank the large number
of people at the GSFC who brought this challenging in-house project to
fruition.

References
Albrecht, A. & Steinhardt, P. J.: 1982, Phys.Rev.Lett. 48,1220
Alvarez, 1. W., Buffington, A., Gorenstein, M. V., Mast, T. S., Muller, R. A ., Orth, C .
D., Smoot, G . S., Thornton, D. D., & Welch, W. J . 1974, Observational Cosmology:
The Isotropy of the Primordial Black Body Radiation, UCBSSL 556/75 Proposal to
NASA
Bardeen, J. M., Steinhardt, P. J. & Turner, M. S.: 1983, Phys.Rev.D 28,679
Barney, R . D.: 1991, nluminating Eng. Soc. 34, 34
Bennett, C. L.: 1991, Highlights Astron. , in press
Bennett, C. L. et al.: 1992a, ApJ 391, 466
Bennett, C. L. et a1.: 1992b, ApJ 396, L7
Bennett, D. P. & Rhie, S. H. 1992, preprint
54

Bertschinger, E., Dekel, A., Faber, S. M., Dressler, A., & Burstein, D.: 1990, ApJ 364,
370
Bertschinger, E. 1992, in Current Topics in Astrofundamental Physics, eds. N. Sanchez &
Z. Zichini, in press and in New Insights Into The Univt;rse, eds. V. J. Martinez, M.
Portilla & D. Saez, in press
Boggess, N. W. : 1991, Highlights Astron. , in press
Boggess, N. et a!.: 1992, ApJ 397, 420
Bond, J. a. & Efstathiou, J.: 1984, ApJ 285, L45
Bond, J. R., Carr, B. J., & Hogan, C. J.: 1986, ApJ 306, 428
Bond, J. R. & Efstathiou, J.: 1987, MNRAS 226, 655
Bond, J. R., Carr, B. J., & Hogan, C . J. 1991, ApJ, 367, 420
Bromberg, B. W. & Croft, J.: 1985, Adv.Astron.Sci. 57,217
Carr, B. J. 1988, Comets to Cosmology, ed. A. Lawrence, (Springer Verlag:London), 265
Cheng, E. S. 1991, 1st Annual Conf. on A"tronomical Data Analysis Software and Systems,
NOAO, Tucson, AZ
Cheng, E. S. et a!. : 1990, Bull.Am.Phys .Soc. 35, 971
Cheng, E. S. et a!.: 1991, BAAS 23,896
Coladonato, R. J., Irish, S. M., & Mosier, C. L. 1990, Third Air Force/NASA Symp .
on Recent Advance" in Multidisciplinary Analy"is and Optimization, (Hayward, CA:
Anarnet), 370
Davis,M. & Peebles, P. J. E. : 1983, ApJ 267, 465
Davis, M., Summers, F. J. & Schlegel, D. 1992, Nature, in press
Efstathiou, G . 1990, in Physics of the Early Universe, eds. J . A. Peacock, A. F. Heavens,
& A. T . Davies (Edinburgh Univ. Press: Edinburgh, Scotland), p. 361
Efstathiou, G., Bond, J. R . & White, S. D. M . 1992, MNRAS, 258, IP
Evans, D. C.: 1983, SPIE Proc. 384, 82
Evans, D. C.& Breault, R. P.: 1983, SPIE Proc. 384, 90
Fabbri, R.: 1988, ApJ 334,6
Fabbri, R. & Melchiorri, F.: 1979, ABA 78, 376
Fabbri, R., Andreani, P., Melchiorri, F., & Nisini, B.: 1987, ApJ 315, 12
Fixsen, D. J., Cheng, E. S., & Wilkinson, D. T.: 1983, Phys .Rev.Lett. 50,620
Franceschini, A., Toffolatti, L., Danese, & De Zotti, G.: 1989, ApJ 344, 35
Gorski, K.: 1991, ApJ 370, L5
Gorski, K. M., Silk, J. & Vittorio, N. : 1992, Phy".Rev.Lett. 68,733
Gulkis, S., Carpenter, R . L., Estabrook, F. B., Janssen, M. A., Johnston, E. J., Reid, M.
S., Stelzried, C. T., & Wahlquist, H. D. 1974, Cosmic Microwave Background Radiation'
Proposal, NASA/JPL Proposal
Gulkis, S., Lubin, P . M., Meyer, S. S., & Silverberg, R. F. : 1990, Sci.Amer. 262, 132
Gush, H. P . , Halpern, M. & Wishnow, E. H.: 1990, Phy".Rev.Lett. 65,537
Guth, A.: 1981, Phys.Rev.D 23, 347
Guth, A. & Pi, Y-S.: 1982, Phys. Rev. Lett. 49, 1110
Harrison, E. R.: 1970, Phys.Rev. D 1,2726
Harwit, M. 1970,Rivista del Nuovo Cimento II, 253
Hauser, M. G. : 1991a, Proc. Infrared Astronomy and ISO, Les Houches, June in press
Hauser, M. G.: 1991b, Highlights Astron. , in press
Hauser, M. G. et al. 1991,After the First Three Minute8 eds. S. S. Holt, C . 1. Bennett, &
V. Trimble, (New York:AIP Conf. Proc. 222), 161
Hawking, S.: 1982, Phys.Lett. 115B, 295
Hogan, C . J .: 1991, Nature 350, 469
Hogan, C. J. 1992, submitted to ApJ Letters
Hogan, C. J . 1993, ApJ, in press
Holtzman, J . A.: 1989, ApJS 71, 1
Hopkins, R. A. & Castles, S. H.: 1985, Proc. SPIE 509, 207
Hopkins, R. A. & Payne, D. A.: 1987, Adv.Cryog.Eng. 33, 925
Hoyle, F & Burbidge, G. 1992, preprint
 55

Janssen, M. A. & Gulkis, S. 1992, The Infrared and Submillimetre Sky After COBE, eds .
M. Signore & C. Dupraz, (Dordrecht: Kluwer), 391
Kaiser, N., Efstathiou, G., Ellis, R., Frenk, C., Lawrence, A., Rowan-Robinson, M. &
Saunders, W .: 1990, MNRAS 252, 1 .
Kaufman, M.: 1976, Ap.Sp.Sci. 40, 369
Kogut, A., et a1.: 1992, ApJ , in press
Kolb, E., & Turner, M. S. 1987, The Early Universe, (Addison Wesley: Redwood City,
. CA)
Klypin, A. A., Sazhin, M . V., Strukov, I. A.,& Skulachev, D.P.: 1987, Sov.Astr.Letters 13,
104
Kumar, V. K ., Freedman, I., Wright, E. L., & Patt, F. S., 1991, Flight Mechanics and
Estimation Theory Symposium
Linde, A.: 1982, Phys.Lett. 108B, 389
Loveday, J ., Efstathiou, G., Peterson, B. A. & Maddox, S. J. 1992, preprint
Low, F . J. & Tucker, W. H.: 1968, Phys. Rev.Lett. 22, 1538
Low, F. J. et a1. : 1984, ApJ 278, L19
Lubin, P., Villela, T., Epstein, G., & Smoot, G.: 1985, ApJ 298, L1
Magner, T. J .: 1987, Opt. Eng. 26, 264
Mather, J . C.: 1981, Applied Optics 20, 3992
Mather, J. C.: 1982, ~pt . Eng. 21, 769
Mather, J. C.: 1984a, Applied Optics 23, 584
Mather, J. C.: 1984b, Applied Optics 23,3181
Mather, J. C. 1987, Proc. 13'h Texas Symp . Rei. Astrophys., (World Scientific Pub.: Sin-
gapore) p. 232
Mather, J. C. 1991, Highlights Astron., in press
Mather, J., Thaddeus, P ., Weiss, R., Muehlner, D., Wilkinson, D. T., Hauser, M. G., &
Silverberg, R. F. 1974,Cosmological Background Radiation Satellite, NASA/Goddard
. Proposal
Mather, J. C., Toral, M., & Hemmati, H.: 1986, Applied Optics 25,2826
Mather, J . C., et a1.: 1990a, ApJ 354, L37
Mather, J. C., et a1. 1990b, IA U Colloq. 1~3, Observatories in Earth Orbit and Beyond,
ed. Y. Kondo, (Boston: Kluwer), 9
Mather, J. C. et a1. 1991a, After the First Three Minutes, eds. S. S. Holt, C L . Bennett ,
& V. Trimble, (New York: AlP Conf Proc 222), 43
Mather, J. C. et a1.: 1991b, Adv.Sp.Res. 11,181
Matsumoto, T., Akiba, M. & Murakami, H.: 1988a, ApJ 332, 575
Matsumoto, T., Hayakawa, S., Matsuo, H., Murakami, H., Sato, S., Lange, A . E., &
Richards, P . L. : 1988b, ApJ 329, 567
McDowell, J . c.: 1986, MNRAS 223, 763
Meyer S. S., Page, L., & Cheng, E. S.: 1991, ApJ 371, L7
Milam, L. J. : 1991, fllum.Eng.Soc.J . 34,27
Mosier, C. L. 1991a, "Thermal Design, Testing, and Analysis of the Cosmic Background
Explorer's External Calibrator" in AIAA 29th Aerospace Sciences Conference
Mosier, C. L. 1991b, "Thermal Design of the Cosmic Background Explorer Cryogenic
Optical Assembly" in AlA A 29th Aerospace Sciences Conference
Murdock, T. L. & Price, S. D.: 1985, AJ 90,375
Murdock, T. L., Burdick, S. V ., Hauser, M . G ., Kelsall, T., Moseley, S. H., Silverberg, R.
F ., Toller, G . N., Stemwedel, S. W., & Freuderueich, H. T. : 1991, BAAS 23, 1313
Negroponte, J.: 1986, MNRAS 222, 19
Noda, M., Christov, V. V ., Matsuhara, H., Matsumoto, T., Matsuura, S., Noguchi, K.,
Sato, S. & Murakami, H.: 1992, ApJ 391, 456
Olive, K. A., Schramm, D . N., Steigman, G., & Walker, T. P. : 1990, Phys. Lett. B 236,
454
Partridge, R . B. & Peebles, P. J . E.: 1967, ApJ 148, 377
Peebles, P. J. E.: 1969, Phil. Trans. Royal Soc . London, A 264,279
56

Peebles, P. J. E. 1971, Physical Cosmology, (Princeton Univ. Press: Princeton, N J)

Peebles, P. J . E. 1980, The Large-Scale Structure of the Universe, (Princeton Univ. Press:
Princton, NJ)
Peebles, P. J . E.: 1981, ApJ 248,885
Peebles, P. J. E.: 1982, ApJ 263, Ll
Peebles, P. J . E.: 1984, ApJ 284, 439
Peebles, P. J. E. 1991, After the First Three Minutes, eds. S. S. Holt, C L. Bennett, & V.
Trimble, (New York: AlP Conf Proc 222), 3
Peebles, P. J . E. & Wilkinson, D. T. : 1968, Phys. Rev. 174,2168
Peebles, P . J. E. & Yu, J . T .: 1970, ApJ 162,815
Penzias, A. A ., & Wilson, R. W.: 1965, ApJ 142, 419
Petuchowski, S. J . & Bennett, C.L.: 1992, ApJ , in press
Reynolds, R. J. : 1984, ApJ 282, 191
Reynolds, R. J .: 1985, ApJ 294, 256
Reynolds, R. J. : 1992, ApJ 392, L35
Rowan-Robinson, M. : 1986, MNRAS 219, 737
Sachs, R. K. & Wolfe, A. M. : 1967, ApJ 147,73
Sampler, H. P.: 1990, Proc. SPIE 1340, 417
Shafer, R . A. et a1. : 1991, Bull.Am.Phys.Soc. 36, 1398
Schaefer, R . K . 1991, After the First Three Minutes, eds. S. S. Holt, C L. Bennett, & V.
Trimble, (New York: AlP Conf. Proc. 222), 119
Smoot, G. F.: 1991, Highlights Astron , in press
Smoot, G. F. et al.: 1990, ApJ 360, 685
Smoot, G. F. et a1.: 1991a, Adv.Space Res. 11, 193
Smoot, G. F. et al.: 1991b, ApJ 371, Ll
Smoot, G. F. et al. 1991c, After the First Three Minutes, eds. S. S. Holt, C L. Bennett, &
V . Trimble, (New York: AlP Conf. Proc. 222), 95
Smoot, G. F. et a1.: 1992, ApJ 396, Ll
Starobinskii, A. A.: 1982, Phys.Lett. 117B, 175
Sunyaev, R. A. & Zeldovich, Ya. B.: 1980, Ann. Rev. Astr. Ap. 18, 537
Torres, S., et a1. 1989, Data Analysis in Astronomy eds. V . di Gesu, L . Scarsi, & M . C .
Maccarone, Erice, June 20-27, Plenum Press
van Dalen, A. & Schaefer, R. K. 1992, ApJ, in press
Vogeley, M . S., Park, C., Geller, M. J . & Huchra, J. P.: 1992, ApJ 391, L5
Volz, S. M . & Ryschkewitsch, M. G. 1990a, Superfiuid Helium Heat Transfer, eds. J. P.
Kelly & W. J. Schneider, (New York: AME), 134,23
Volz, S. M., Dipirro, M. J ., Castles, S. H., Rhee, M . S., Ryschkewitsch, M. G., & Hopkins,
R. 1990, Proc. IntI. Symp. Optical and Opto-electronic Appl. Sci. and Eng., (San Diego:
SPIE),268
Volz, S. M ., Dipirro, M. J., Castles, S. H., Ryschkewitsch, M. G., & Hopkins, R.: 1991a,
Adv.Cry.Eng. 35B, 1703
Volz, S. M., Dipirro, M. J ., Ryschkewitsch, M. G ., & Hopkins, R. : 1991b, Adv.Cry.Eng.
37A,1183
Volz, S. M. & DiPirro, M . J.: 1992, Cryogenics 32, 77
Vrtilek, J. M . & Hauser, M. G. 1992, in preparation
Walker, T. P., Steigman, G., Schramm, D. N., Olive, K. A ., & Kang, H .-S.: 1991, ApJ
376,51
Weiland, J. L ., Hauser, M. G., Kelsall, T ., Moseley, S. H., Silverberg, R . F., Odegard, N.
P., Spiesman, W. J ., Stemwedel, S. W., & Freudenreich, H. T .: 1991, BAAS 23, 1313
White, R. A. & Mather, J. C . 1991, Astrophysics and Space Sciences Library, eds. M. A.
Albrecht & D. Egret, (Dordrecht: Kluwer ),171, 29
Wilkinson, D . T .: 1986, Science 232, 1517
Wright, E. L .: 1990, Ann. NY Acad. Sci. Proc. 647, 190
Wright, E. L ., 1991a, The Infrared and Submillimetre Sky After COBE, eds. M . Signore
& C . Dupraz, (Dordrecht:Kluwer), 231
 57

Wright, E. L.: 1991b, ApJ 375, 608

Wright, E. 1. et al.: 1.990, BAAS 22, 874
W~ight, E . L. et al.: 1991, ApJ 381, 200
Wright, E . TJ. et al.: 1992, ApJ 3,96, LI3
7eldovich, Ya B.: 1972, MNRAS 160, 1
 THE FORMATION OF GALAXIES

Structure Formation in a Hierarchical Universe

MARC DAVIS
University of California, Berkeley

Abstract. The past decade has witnessed remarkable progress in our understanding of
the development of structure in the Universe. The amplitude of matter fluctuations has
been measured over more than four decades of length scale, from the size of the observable
Universe itself to galactic scale sizes. Although cosmologists argue over a factor of two in
the relative normalization of large-scale and small-scal~ structure amplitudes, and whether
all the data is consistent with standard Cold Dark Matter (CDM), there is little question
about the basic hierarchical nature of structure formation. A few elementary considerations
of such theories are summarized in this brief presentation.
Key words: large scale structure - cosmology - galaxy formation

1. Power Spectra and Large Scale Sfructure

We have all been taught that the Universe is to lowest order smooth and ho-
mogeneous. Of course on small scale there are large amplitude fluctuations,
but the rms fluctuation b M / M, is observed to decrease with scale, approxi-
mat ely as M-l/3. Most of the participants of this conference are concerned
with the large amplitude fluctuations observed on small scale, namely in-
dividual galaxies and their internal structure. Although galaxy formation
involves gas dynamical processes, gravitational collapse is the major phys-
ical ingredient controlling their formation and evolution; therefore galaxy
formation must be considered an aspect of large scale structure formation,
which is almost certainly controlled by self-gravitational instability.
Fluctuations from the smooth background can be characterized by using
a number of different statistical tools, perhaps the most fundamental of
which is the power spectrum of density fluctuations. First, write the density
fluctuation at a point r as 1 + b (r) = p( r ) / p, with p the mean density. The
dimensionless density fluctuation b( r) can be written as a Fourier transform,
i.e. as a superposition of plane waves,

(1)

Given that there is no preferred location in the Universe, by symmetry the

expectation value of a given Fourier mode must be zero, (b(k») = O. However
the variance of its modulus need not be zero and is defined as the power
spectrum of density fluctuations,

(2)

59
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, 59-68.
© 1993 Kluwer Academic Publishers.
60

By isotropy in the cosmological context, P(k) = P(k) with k = Ikl. The

power spectrum has units of density and is the Fourier conjugate function
of the well-studied two-point correlation function ~(r),

~(r) = _1_
(211" )3
jd kP(k)e-
3 ik.r . (3)

Theories of large-scale structure based on gaussian random noise, such as

models in which the fluctuations arise as quantum noise in an inflationary
phase of the early Universe, are fully specified by their power spectrum of
density fluctuations because each Fourier mode has random phase. Mod-
els with structure forming from cosmic defects, e.g. cosmic strings, wakes of
strings, textures, or the like, have phase coherence between modes and there-
fore cannot be fully characterized by P(k), even for low-amplitude modes. In
a matter-dominated Universe, individual Fourier modes are decoupled and
all grow at the same rate, bk <X D(a), where D(a) = a in an Einstein - de
Sitter phase. Thus, the power spectrum preserves its initial shape on large
scale, where the fluctuations are small. However, on small scales, the large-
amplitude fluctuations lead to mode coupling and the initial power spectral
shape is modified.
For the past decade, the structure studies have focused on ~(r) because it
is relatively easy to measure on small scales, where the clustering amplitude
is large and nonlinear. However, in the large- scale linear regime, ~ (r) is
notoriously difficult to measure (Hamilton 1992), largely because 1 + ~ is
uncertain in proportion to the uncertainty in the catalog density, usually no
better than 5-10%. Direct measurement of P(k) from galaxy redshift survey
data offers a technical advantage because the power spectrum is a positive
definite quantity, whereas ~(r) is expected to be negative on large scale. In
properly designed studies, the uncertainty in the background density merely
scales the amplitude of P( k) and therefore is not the limiting noise source.
Within the past several years, direct power spectral analysis has been
applied to the major redshift survey catalogs. Figure 1, taken from Fisher et
ai. (1993), shows the results for the IRAS 1.2 Jy survey, as well as the recent
efA measurements (Vogeley et al. 1992), scaled to a common amplitude
reflecting the known difference in normalization of the IRAS versus optically
selected correlations. The measurements are seen to be quite consistent with
each other over the common scales. The largest mode of the IRAS analysis
is for a wavelength of 180h- 1 Mpc (h = H o /100), which is well beyond the
usual20h- 1 Mpc limit for reliable estimates of ~(r). Note that the measured
power for the smallest k's of the redshift survey data is beginning to flatten,
although the significance of this turndown is small. The recent COBE results
from the DMR experiment (Smoot et ai. 1992) can be converted to a power
spectrum measurement, interpreting their observed fluctuations as Sachs-
Wolfe potential fluctuations in an Einstein - de Sitter Universe. The COBE
 61

Rh in Mpc

.
10J~0~'nn~~,-_1~0~0~0~-..-.-_1~0~0~".-.-~lpOon~,,-.__~rg

a .
T o~fW-=O . 7

ar =1 .0
A
bIRAS.a~/a~w

bCfA.a~A/ag"
10'
........... .

1000 ~
.••....•..••.. / ..... .... f\. ~
i . .\;

COWPARlsOH
o COBE OMR (n-I. Q....
or OATA TO COM/NON
o-16.7:1:4ILK )
if,. ..
• IRAS (Fisher etal 1992. ar=0.7)
10
- - CrA (Vogeley etal 1992. ar m O.7)
- - COM (Holtzman. h-0 .5. a: -O.7)
M

..... NOM (Holtzman, H.-I. 0.-0.3, h-0.5, a: -0.7)

M

(RedahUt correcUons made

~o IRAS.CrA at large scales)
0.1 0.001 0.01 0.1 10
kh- I In Mpc- I

Fig. 1. The power spectum of density fluctuations, as inferred from large scale
redshift surveys and from the COBE DMR experiment.

results are plotted as the parallelogram showing the wavenumbers covered

(10 0 to 180°), as well as the reported error in amplitude. The COBE result
is plotted for spectral index n = 1 and EH = (5.4 ± 1.6) X 10- 6 (Smoot et
al. 1992; Wright et al. 1992).
The entire subject of large-scale structure is summarized by Figure 1. On
the scale of the entire observable Universe, COBB has detected fluctuations
that can be smoothly joined to the fluctuations measured in galaxy catalogs
on much smaller scales. The range in length scales shown in Figure 1 is four
decades, with a gap between COBE and the galaxy catalogs of only 2/3
decade. Hopefully within the next several years, this gap will be filled by
means of improved microwave fluctuation measurements on the 1-5 degree
scale, as well as measurements from larger redshift survey catalogs.
To fit theoretical models to Figure 1, one must be aware that the galaxy
distribution might be a biased tracer of mass, and that on small scales,
62

(k > 0.6h Mpc- 1 ), nonlinear effects can distort the spectral shape. Part of
the COBE fluctuations might be primordial gravitational waves (Davis et
al. 1992), but this contribution is likely to be small. Otherwise, the COBE
fluctuations presumably trace the mass fluctuations, and can be used as a
normalization point for theories of large scale structure. Shown in Figure 1
are two such theories, standard Cold Dark Matter (SCDM) for Oh = 0.5, and
Mixed Dark Matter (MDM) (Davis et al. 1992). The SCDM curve fails to fit
the shape of the observed power spectrum; if normalized to the COBE data
it predicts too much small-scale power in mass fluctuations, and if scaled
to the observed galaxy clustering, it underpredicts the large-scale power by
a factor of two. It is quite incredible that cosmology has advanced to the
degree that we are concerned with a mere factor of two disagreement over
such a large dynamic range. Yet such is the state of the field that the COBE
results might be the final straw that breaks the back of standard CDM.
Several modifications to SCDM have been suggested (Efstathiou et al.
1992). A simple modification, which fits the n = 1 mandate of inflation, is
a Universe with mostly cold dark matter, but with some hot dark ma,tter
as well. Massive neutrinos are the natural hot dark matter candidate. The
power spectrum of a mixed dark matter model with one type of hot dark
matter contributing Ov = 0.3, and OCDM = 0.69 (Holtzman 1989) is shown
in Figure 1. The fluctuation spectrum of this model is shaped slightly differ-
ently than SCDM, with more large-scale power and less small-scale power,
but the differences are not large. Not only does this slight modification pro-
vide an excellent fit to the power spectrum observations, but giving neutrinos
mass (nv = 0.3 implies mv = 7 eV for h = 0.5, presumably the T neutrino)
is astrophysically motivated by MSW mixing between Ve and v!-" a possible
explanation of the solar neutrino data (Bahcall & Bethe 1991).

2. Scaling laws
In terms of galaxy formation, there is little distinction between SCDM and
its possible successors. All are hierarchical models, implying that large struc-
tures are formed by the accretion and merging of smaller structures. Given
a normalization to the power spectrum, the formation time scale of galaxies
is completely specified. Because it is a statistical process, there is no single
"epoch" of galaxy formation. However, at any given epoch, there is always
some characteristic scale that is just becoming nonlinear, and this charac-
teristic mass scale increases with time. The nonlinear scale can be readily
derived from P( k) by noting that rms density fluctuations on a scale r can
be expressed as an integral over P( k ),

(4)
 63

where W(kr) is the Fourier space representation of the window. This de-
pends on the precise shape of the averaging volume, but for reasonable
windows can be crudely approximated as

if a: < 1
W(a:)={~ ifa:>1'

which is adequate for the scaling arguments below.

Prior to the detection of fluctuations with COBE, the usual method of
normalization ofthe fluctuations was at a scale of 8h- 1 Mpc, chosen because
a sphere of this radius is observed to have l'ms fluctuations in the optically
selected galaxy distribution (Davis & Peebles 1983). We can thus define
the l'ms mass fluctuations on this scale as (<5p/ P)lrm8 = b-t, where b is to
be interpreted as a bias factor by which the galaxy distribution is more
clustered than the matter distribution (e.g., Davis et a1. 1985).
Let P( k) ()( k n and consider small amplitudes. Then during the Einstein
- de-Sitter phase we have

(5)

with a: a co-moving length scale and a the cosmic scale factor. Since co-
moving mass scales as a: 3 , the mass fluctuations are given as

(6)

On the large COBE scale, we have n = 1 and (<5M/M)rm8 ()( M- 2 / 3 , but on

the nonlinear clustering scales observed today, the effective spectral slope is
n ~ -1, or (<5M/M)rm& ()( M- 1 /3. Thus, hierarchical clustering naturally
leads toward homogeneity in a large-scale Universe, on a rate slower than
expected for random Poisson processes (n = 0) on small scales, but faster
than random Poisson on the larger scales explored by COBE.
When a given fluctuation has small amplitude, linear theory implies that
it will continue to expand with the general expansion. However, when

(7)

the structure becomes self-gravitating and separates from the expansion,

collapsing typically a factor of two before coming into virial equilibrium.
The physical size of the structure at turnaround is

(8)
64

Given a scaling law between radius and mass, plus the virial theorem, all
other scaling relations follow immediately:

Vrm~OC [G~(r)r/2 oc M(1-n)/12 (9)

(10)

(11)
In CDM type models, neff ~ -2 on galaxy scales, thus leading to Vrm~ OC
M 1 / 4 ; if M/L is constant, then we expect L oc (~v)4.
An example of the formation of structure in a high-bias (b = 2.5) CDM
model is shown in Figure 2. Shown are slices from a small-scale simulation
with co-moving box size of only 2.5h- 1 Mpc on a side. Shown are slices 1/8
the thickness of the simulation cube at epochs corresponding to Z = 6 and
Z = 2. Note the degree of nonlinear structures at Z = 6 in this high bias
model; by red shift Z = 2 most of the particles (mass tracers) are located
within nonlinear lumps. These lumps grow with time by accretion of single
particles as well as merging with smaller lumps.

3. The Abundance of Lumps

A simple algorithm that can be used to estimate the number of lumps of
given size was given by Press & Schechter (1974). Imagine an initially gaus-
sian density field that is smoothed on scale roo Let .a(ro) be the rms density
fluctuation for this smoothing at a given epoch Z. Then the fraction of mass
within lumps of density contrast :2: ~c is simply given by an error integral

F(ro,Z) = i~ (2~r/2 .a(r~,Z)exp [2.a 2(::,Z)] d~. (12)

Suitable choices of ~c are 1.3 or 1.69, the linear overdensities at which self
gravitation in a nonlinear spherical model stops the expansion, or recollapses
and virializes. The differential abundance of lumps of size ro is then given
by
8F
f(ro, Z)dro = 28 - dro
ro
The factor 2 is a fudge to give mass conservation, but it has been fully
justified by subsequent work demonstrating the remarkable accuracy of this
procedure (Bond et al. 1991). To express the number density as a function
of mass, we have

N(M)dM=2-8F dM (13)
PaM M
 65

Fig. 2. Small-scale structure formation in a b = 2.5 CDM Universe. The different

panels are slices from a simulation of size 2.5h- 1 Mpc at Z = 6 and Z = 2. The
largest lumps to form in this model have masses of order 10 11 M 0 .

(14)

One can use the scaling laws above to relate M and ~(ro, Z), ~ ex: M- a .,
0 6 = (3 + n)/6. On small scales, ~(ro,O) is large so the exponential is unity
and we have N(M) ex: M-(2-a.). This is much steeper than the reported
faint-end slope of the galaxy luminosity function, 0 = -1, or -1.25, (Ef-
stathiou, Ellis, & Peterson 1988). Although there is some question about
how well the Press-Schechter formalism can work on the most nonlinear
small scales, there is also the possibility that the slope of the faint end of
the galaxy luminosity function is much steeper today than generally rec-
ognized (e.g., Bothun et al. 1991), or was steeper in the past (Maddox et
al. 1990). Furthermore, it is not obvious that M/L is constant, because the
simulations suggest that most of the mass in the Universe resides in very
extended halos of lumps, well beyond the optical extent of galaxies.
Numerical simulations (e.g., Efstathiou et al. 1988) demonstrate that the
Press-Schechter results evolve in a self-similar fashion for power-law initial
conditions. Narayan and White (1988) have shown how to recast the Press-
66

->"
I o

- c::

co
o W
_6~~-A~~~~~~-L-L~~ ~~-L~~~~~~~-L-L~

1.5 1 .5 1 .5 o
log(l +z) log(l+z)

Fig. 3. The abundance of halos of different potential well depth as measured by

circular velocity, versus epoch, for b = =
2.5 and b 1.5 CDM models (from White
&. Frenk 1991)

Schechter formula in terms of the abundance of objects versus potential

well depth. This is a nearly direct observable via rotation curves or velocity
dispersions, whereas the mass for the highly extended galaxies is not directly
measurable. Shown in Figure 3, taken from White and Frenk (1991), is the
abundance of halos as a function of circular velocity versus redshift for CDM
models with two different normalizations: b = 1.5 and b = 2.5. Note how
the abundance of lumps with circular velocity of 200 km S-1 (slightly less
than an L· galaxy) reaches its peak at Z = 4 in the b = 1.5 model, and
at Z = 1.2 in the b = 2.5 model. (The hot plus cold dark matter model
mentioned above behaves much like a b = 2.5 CDM model on small scale.)
The formation of lumps that could plausibly be associated with galaxies is
seen to be an extended affair, with deep potential wells forming rather late.

4. Physics Beyond Gravitation

Clearly the scaling arguments above are only a first step toward understand-
ing galaxy formation. Gaseous dissipation during and after the collapse of
galaxy-sized systems is clearly important for segregating the baryons from
the dark matter (White & Rees 1978). Press-Schechter methods can be ex-
 67

tended to include baryonic cooling (Cole 1991; White & Frenk 1991; Blan-
chard et al. 1992), but unless some type of feedback is included, the result is
that cooling is too efficient. All the baryons will cool in the shallow poten-
tial wells that form in the first levels of the hierarchy. Thus details of early
star formation and supernovae and other heating processes become very
important for any understanding of thenongravitational aspects of galaxy
formation. For example, one way to hide the predicted overabundance of
dwarf galaxies is to imagine that a first burst of star formation leads to
supernovae that blow the gas out of the shallow potential wells (Dekel &
Silk 1986). Perhaps during their one burst of star formation these objects
account for the excess of faint blue galaxies (Lilly et al. 1991).
A complementary technique is to study galaxy formation via simulation
using a code that includes hydrodynamic effects, such as Smooth Particle
Hydrodynamics. Gus Evrard (this volume) will give some details on recent
work in this direction.

5. Summary

From this brief discussion, one can extract an even briefer summary.
• The COBE detection of fluctuations shows that the models cosmologists
have been pushing for the past decade were a reasonably accurate guide
to a final model.of structure in the Universe. Cosmology has moved from
the domain of a near total absence of data to arguments over a factor
of two in the degree of power on a given scale. This is progress indeed!
• All of the models still under consideration behave on small scale like a
high-bias hierarchical CDM model. In hierarchical models of the growth
of structure, one must not consider galaxy formation in isolation. In
CDM-like models, galaxy formation is an ongoing process with most
shallow wells forming early, and deeper wells forming rather late.
• The gravitational aspects of galaxy formation are well in hand. The gas
processes are understood in principle, but details are very uncertain in
practice. The cooling is likely to lead to a multiphase media, and the
roles of star formation, energy feedback, and magnetic fields remain
to be understood. The large number of free parameters renders the
analytic treatments as very uncertain.
• Numerical simulations including gas are just getting underway. These
models should be considered as illustrative, but they will suffer the
same uncertainties as above, plus they will always be very limited in
resolution.

References
Bahcall, J. N., & Bethe, H. A.: 1991, Phys. Rev. D. 44,2962
Blanchard, A ., Valls-Gabaud, D., & Mamon, G. A.: 1992, Astron. fj Astrophys. 264, 365
68

Bond, J. R., Cole, S., Efstathiou, G., &; Kaiser, N. : 1991, Ap. J. 379, 440
Bothun, G. D., Irnpey, C. D., &; Malin, D. F.: 1991, A8tron. J . 376, 404
Cole, S: 1991, Ap. J. 367, 45
Davis, M., &; Peebles, P. J. E.: 1983, Ap. J. 267, 465
Davis, M ., Efstathiou, G., Frenk, C. S., &; White, S. D. M.: 1985, Ap. J. 292, 371
Davis, M., Schlegel, D., &; Summers, F.: 1992, Nature 359, 393 -
Davis, R. L., Hodges, H. M., Smoot, G. F. Stehiliardt, P. J. &; Turner, oM. : 1992, PhY8.
Rev. Lett 69, 1856
Dekel, A., &; Silk, J.: 1986, Ap. J. 303, 39
Efstathiou, G., Frenk, C. S., White, S. D. M., &; Davis, M: 1988, M.N.R.A.S. 235, 715
Efstathiou, G, Ellis, R. S., &; Peterson, B. A.: 1988, M.N.R.A.S. 232, 431
Efstathiou, G., Bond, J. R., &; White, S. D. M.: 1992, M.N.R.A.S. 256, 43P
Fisher, K. B., Davis, M., Strauss, M. A., Yahil, A., &; Huchra, J.: 1993, Ap. J. 402, 42
HanUlton, A.: 1992, Ap. J. 385, L5
Holtzman, J. A.: 1989, Ap. J. Supp. 71, 1
Lilly, S., Cowie, L. L, &; Gardner, J. P.: 1991, Ap. J. Lett. 369, L79
Maddox S. J., Sutherland, W. J., Efstathiou, G, Loveday, J. et al. : 1990, M.N.R.A.S.
247,IP
Narayan, R., &; White, S. D. M.: 1988, M.N.R.A.S. 231, 97p
Press, W. H., &; Schechter, P.: 1974, Ap. J. 187,425
Smoot, G. F. et al. : 1992, Ap. J. Lett. 396, Ll
White, S. D. M., &; Rees, M. J .: 1978, M.N.R.A .S. 183, 341
White, S. D. M., &; Frenk, C.: 1991, Ap. J. 379, 52
Vogeley, M. S., Park, C., Geller, M. J ., &; Huchra, J.: 1992, Ap. J . Lett. 395, L5
Wright, N. et al. : 1992, Ap. J. Lett. 396, L13
 EXPERIMENTAL GALAXY FORMATION

AUGUST E. EVRARD
Physics Department, University of Michigan

Abstract. We present results from two-fluid numerical simulations modelling the for-
mation of galaxies and larger structure in a 16 Mpc-periodic volume of a flat, cold dark
matter universe with baryon fraction fh =0.1. This simulation is the first to form galactic
disks in a cosmological environment. The Tully-Fisher relation for these systems is ex-
amined, as well as their merger frequency. Other results which depend crucially on the
dissipative nature of the baryonic component, namely the mass function, degree of biasing
and cluster mass estimates, are presented. The role of gas dynamics in large scale structure
simulations is briefly examined and future directions are outlined.
Key words: Cosmology-Theory; Galaxies-Formation

1. Introduction

The formation of galaxies is a long-standing problem in modern cosmol-

ogy. The most popular scenario nowadays postulates a universe dominated
by dark, non-baryonic matter which was seeded at some early (inflation-
ary, quantum gravity?) epoch with small density inhomogeneities. Those
small fluctuations were amplified by self-gravity over a Hubble time to the
point where they effectively separated from the universal expansion and then
collapsed to form quasi-hydrostatic equilibrium structures (halos is the buz-
zword needed here). Observational motivation for this scenario comes from
the r(Kent COBE DMR detection of fluctuations in the microwave back-
ground temperature at about the level expected theoretically (Bennett et
al., this volume). In a universe dominated by cold, dark matter (CDM; e.g.,
Blumenthal et al. 1986), small halos form first that then grow by merging
and accretion. This process has been extensively investigated by N-body ex-
periments (see Frenk 1991 for a review) which employ a single, collisionless
fluid to represent the dominant dark matter species.
Galaxy formation within such a scenario involves collapse and dissipation
(via radiative cooling) of baryons within the potential wells of the dark halos
(White &: Rees 1978). The characteristic maximum mass of galaxies falls out
of simple cooling time scale arguments (Rees & Ostriker 1977; Silk 1977;
see also Davis in this volume) . The ability for the baryons to dissipate via
radiative processes has long been recognized as an essential ingredient in
the recipe of galaxy formation, particularly in the formation of spiral disks
(Fall & Efstathiou 1980). Ironically, until recently, the baryons have largely
been ignored in numerical experiments. The reasons are mainly practical;
computer hardware and software were not yet up to the task. The N-body
studies of the mid-to-Iate eighties were largely intended to address the large-
scale distribution of mass. Galaxies were, at some level, an afterthought dealt

69
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, 69-90.
© 1993 Kluwer Academic Publishers.
70

with by introducing some particular scheme for "biasing" their distribution

with respect to the dark matter.
With increased computer power acting as fertilizer, there have sprouted
recently a number of independent codes designed to model the combined gas
dynamic/ gravitational evolution of baryons within a universe dominated by
collisionless dark matter (Evrard 1988; Hernquist & Katz 1989; Chiang,
Ryu & Vishniac 1989; Cen 1992). The purpose of this paper is to report
results from an application of one of these codes, P3MSPH (Evrard 1988),
to the problem of galaxy and cluster formation in a CDM universe. A more
thorough analysis of this work is presented in Evrard, Summers, & Davis
(1993; hereafter ESD) . The reader is referred to this paper for any details
missing in the discussion below.
The gas dynamics of the baryonic fluid is calculated using smoothed
particle hydrodynamics (SPH, see Monaghan 1992 for a recent review). The
physics incorporated for the baryons includes thermal pressure effects, shock
heating via an artificial viscosity, and radiative cooling using a primordial
composition cooling curve with an imposed minimum temperature of 10 4 K.
Star formation and associated feedback, expected to be important processes
in low mass galaxies, are not included.
We assume a relatively "standard" CDM cosmology with critical density
no = 1, present Hubble constant Ho = 50 km S-1 Mpc- 1 , 10% baryon frac-
tion and a spectrum normalized within a sphere of radius 16 Mpc to have
a variance (J' = 0.6, equivalent to a bias parameter b = 1/ (J' = 1.7. Initial
conditions within a co-moving 16 Mpc periodic cube were drawn from the
CDM spectrum constrained to form a small group of galaxies in the center
of the box (Bertschinger 1987) .. The applied constraint generated a density
perturbation of amplitude 3 times the variance on a mass scale 3.6 X 10 13 M0
near the center of the simulation volume.
The system was evolved in two phases. The Zel'dovich approximation
was used to perturb a grid of 262, 144 particles, consistent with the realized
initial density field at a starting redshift z = 49. This set of particles, rep-
resenting the dark matter, was evolved as a collisionless fluid to z = 9. At
this epoch, another set of 262,144 particles was added to follow the distinct
evolution of the baryons, assumed to be initially cold, T = 104 K, with their
density tracing that of the dark matter at this time. The coupled evolution
of baryons and dark matter was then followed until the final epoch z = 1. We
chose not to push the evolution to the present because the mass distribution
on the scale of the box size would be going non-linear; the periodic nature
of the simulation prevents accurate solution of the growth of structure on
these scales.
The total mass in the simulated volume is 2.8 X 10 14 M0 , implying that
a baryonic particle represents 1.08 X 10 8 M0 and each dark matter particle
9.72 X 108 M 0 • This mass resolution yields approximately 1000 particles
 71

Bary ons Dark MaU er

-.- ,"

:... ", ~. :. ;' .

', ,'

.......

Fig. 1. Projec ted particle distrib utions showin g the dark matter
and baryon s in
success ively magnif ied views: 2.8 Mpc (upper) , 400 kpc (middl e)
and 70 kpc (lower) .
Dissipa tion throug h radiati ve cooling leads to a distinc t two-ph
ase structu re in the
baryon s which is not possibl e in the dark matter .
72

of each species per massive galazy. The minimum spatial resolution of rv

13 kpc in the densest regions is comparable to the optically bright regions
of normal galaxies. The intent in choosing these parameters was to resolve
galaxies in as robust a manner possible within the cosmological environment.
More importantly, the intent was to avoid identifying galaxies with very
small numbers (~ 10) of particles - a practice often employed but poorly
understood and rarely trusted.
Figure 1 exhibits the degree of resolution attained in the experiment. On
large scales, the dark matter and baryons trace each other well. Blowing up
the largest dark matter halo reveals obvious differences. The dark matter is
fairly smoothly distributed on a scale ,of rv 400 kpc, while the baryons possess
a distinct, two-phase structure consisting of cold, dense blobs immersed in
a tenuous, hot intracluster medium. We use the term globs, for "galaxy-like
objects", when referring to these blobs of cold, dense baryons. These objects
are assumed to trace the sites of galaxy formation within the volume.

2. Galactic Disks and the Tully-Fisher Relation

The small glob seen in the close-up of Figure 1 is the twelfth largest found
at z = 1, with a baryon mass of 7.2 x 10 10 M 0 , or 671 particles. Closer
examination shows that it has an extended, centrifugally supported disk
with a peak rotation velocity of 210 km s-l. A systematic analysis of the
isolated, massive globs in the simulation found disk components in nearly
all of them. Figure 2 displays the rotation curves of this sample. All but one
of the 15 objects poss~ss extended, centrifugally supported disks. The one
exception, number 29, suffered a recent major merger. Globs 13 and 19 are
accreting mass in their outer parts, which affects their velocity profiles.
With the existence of rotationally supported disks, an obvious question
is whether they obey a relation akin to the observed Tully-Fisher relation,
which links rotation speed with band-limited luminosity. With no star for-
mation included, we do not have a direct estimate of the disk luminosity.
However, we can examine the relation between the baryon disk mass and
rotation speed and infer an appropriate L / M from the observations. Figure
3 shows the baryon disk mass plotted against the maximum measured rota-
tion velocity for isolated globs resolved by 100 or more particles. The open
circles show the raw data while the asterisks apply a correction factor enforc-
ing centrifugal equilibrium (see ESD for details). The correction is largest
for low mass systems because the disks in these globs are typically resolved
by only a few tens of particles. The moment analysis used to determine the
principal component frame in which the rotation velocities are measured is
less reliable for these objects.
The data in Figure 3 are well fitted by a power-law relation MB ()( V~ot
with slope a = 2.5 for the corrected data. Residuals of the fits, shown in
 73

-
,

..
O~.~.~:'~.~-~---i~~~..~.~~~.~t-~~~~~---+.~.7..~.~.~..~~~~~~~r-~.~
(13) "')" (20) [21]

radius (kpc)

Fig. 2. Measured velocity curves for isolated, massive globs. Data points and ~llor
bars show the mean rotational velocity and its dispersion binned in radii containing
20 particles per bin. Dotted and dashed lines show the mean radial and vertical
velocity components, respectively. Solid lines are the equilibrium estimate of the
rotation speed based on the internal mass distribution.
74

0.4

<J
G.2
0
00 p*

..c4

-0.2
-0.4

o raw ... ,/
* corrected .
~.:fS
00;
o,/~ •..i;
'/n •• ~
'0.'*

Fig. 3. Glob mass plotted against the maximum rotation velocity (the Tully-Fisher
relation) for isolated systems with more than half their kinetic energy in the form
of rotation. Open circle~ use the raw rotation velocity, asterisks employ a correction
enforcing centrifugal equilibrium. The best-fit lines of slopes 1.7 (dashed) and 2.5
(dotted) are shown. The upper panel shows the residuals in log MB about the best
fit relations. The rms residuals in dare 0.10 and 0.06 for the raw and corrected
estimates, respectively.

the upper panel of Figure 3, are extremely smalL The rms deviation in
loglO MB is 0.06, equivalent to 0.15 magnitudes of scatter. This is smaller
than the rv 0.3 mag of scatter in the observed Tully-Fisher relation. The
value of the slope arises from the fact that the baryon-to-total mass ratio
measured at the peak of the rotation curve increases with Vrot roughly as
MB / M tot oc v~~~. Larger globs are more dominated by baryons than smaller
globs at fixed radius.
Recovering the observed Tully-Fisher relation L oc V~ot with {3 '::::. 4 requires
the glob luminosity to scale with baryon mass as LocM~/O< rvM1·6. Luckily,
the small scatter in the mass-velocity relation allows extra scatter to be
added in, going from mass to luminosity. Had the scatter in Figure 3 been
larger than observed, one would have to postulate a star formation law that
 75

magically arranged to eliminate some of the scatter. Observational support

for the required trend of luminosity with mass comes from the fact that the
ratio of dynamical mass to light increases toward smaller galaxies (Persic
& Salucci 1988). Dwarf galaxies such as DDO 154 (Carignan & Freeman
1988) may be the extreme end of this trend. It is interesting to note that the
luminosity scaling required to "fix" the Tully-Fisher relation also flattens the
faint-end slope of glob luminosity function to a = -1.24, in better agreement
with observations.

3. Merger Rates

It has long been known that merging is the generic process through which
structures grow in "bottom-up" hierarchical clustering models. The merger
rate at a given epoch depends on the mass scale of interest, the assumed
shape of the fluctuation spectrum, and the value of the density parameter
o (for recent treatments, see Kauffmann & White 1992 and Cole & Lacey
1992). In low-fl models, the growth of fluctuations effectively stops at red-
shift Z'" (0- 1 - 1), while in flat, matter-dominated models, the merging
hierarchy continues to the present. T6th & Ostriker (1992) used this fact
to argue against a flat, matter-dominated universe (such as the "standard"
CDM model) based on the ubiquity of cold galactic disks. Their calculations
indicated that a disk galaxy like our Milky Way could not have accreted
more than 4% of its current mass inside the solar circle in the past third of a
Hubble time via satellite mergers. With more than this amount of accretion,
the galahic disk would be heated to a level in conflict with observations.
Since the majority of bright galaxies possess thin, cold disks, they argued
that cosmologies with high recent merger rates could be ruled out.
One caveat is the fact that, although high recent merger rates are ex-
pected for dark matter halos in 0 = 1 cosmologies, the rates for galaxies
within halos is poorly known. Calculation of the galactic merger rate in-
volves extra assumptions involving, for example, dynamical friction time
scales and dissipation factors, which are generally poorly constrained (Cole
& Lacey 1992). The existence of many galaxies within a common group or
cluster halo indicates that the merger rate for galaxies must be below that
of halos at some point.
The simulation provides the opportunity for direct measurement of the
relative merger frequencies of halos and galaxies. By tracing the history of
particles in a given halo or glob back in time, a "family tree" of the object
can be constructed. One can then answer the question "How long has it been
since the object was involved in a merger between two objects M1 > M2
with mass ratio M2/ M1 > t?" where t is a threshold setting the scale of the
mergers of interest. Figure 4 shows the results of that analysis for the halos
and globs at two values of the threshold, t = 0.3 and 0.1, corresponding to
76

25
20
15
10
rn
~
5
I::
0
=='
0
(J
25
20
15
10
5
, 00 2 4 0 2 4

6tmer, (lOg yr)

Fig. 4. Histograms of the time since last merger event ~trn.~g for the 35 most
massive globs (bottom} and halos (top) identified at z = 1. The left panels are
based on "major" mergers between objects with mass ratios M2/ Ml > 0.3, while
the right panels are based on M 2 /M1 > 0.1. The rightmost bin in each histogram
corresponds to no merger over the course of the simulation.

"major" and "minor" mergers. Objects incurring no merger that satisfies

the threshold condition are plotted in the rightmost bin, which is the age of
the simulation at z = 1.
Over the last one-third of the simulation, the theoretical expectation,
based on the model of Carlberg (1990), is that roughly 50% and 25% of all
galaxies should have suffered mergers involving progenitors with mass ratios
MdMl > 0.1 and 0.3, respectively. From Figure 4, we find that the merger
rates for halos were 45% and 23%, very close to Carlberg's expectations
for galaxies. However, the merger rates for globs, our galaxy tracers, were
suppressed by about a factor 2, being 23% and 11 % for the high and low
thresholds, respectively. Although a factor-of-two reduction in merger rate,
by itself, may not be enough to rescue n = 1 hierarchical models from the
 77

disk frailty argument, the result does call into question the reliability of
n determinations based on merger probabilities. Future experiments with
improved spatial resolution will help decide this issue.
Globs grow in mass via accretion as well as mergers. The dividing line be-
tween accretion and merging is generally fuzzy. However, in the simulation,
the demarcation can be made precise by defining accretion to be acquisition
of mass not collapsed into objects meeting the density and mass require-
ments imposed to define the halo and glob populations. Figure 5 shows the
rate of baryonic mass increase via mergers and accretion over the last 1.3
billion years of the calculation, from z = 1.5 to z = 1, plotted against local
glob density contrast A g • The latter is defined by using the distance to the
6th-nearest glob using the unbiased estimator of Casertano & Hut (1985).
Note that over 4 orders of magnitude in local density are covered in the sim-
ulation. All globs with local density contrast Ag > 10 2 are embedded in the
two largest dark matter halos. Globs that have experienced a major merger
(t> 0.3) at any time during the simulation are plotted as open circles. The
mass of the largest glob, which sits at the bottom of the cluster potential,
decreases because of the loss of a collision fragment that just misses being
linked onto the central object. Inclusion of the mass of the fragment changes
the accretion rate from -21 to 108 M0 yr-I, shown by the dotted line in
the figure. Globs that have not experienced a merger grow through accretion
at an average rate rv 10 M0 yr- 1 •
Two aspects of this diagram are relevant to the observed morphology-
density relation (Dressler 1980; Postman & Geller 1984). One is the fact
that the the accretion rate among the non-merger population is reduced in
the high-density environment of the cluster. The reduction is due to ram-
pressure stripping of the reservoir of material surrounding a galaxy once it
enters the cluster environment. This evolutionary history follows the sce-
nario proposed by Larson, Tinsley & Caldwell (1980) for the origin of SO
galaxies. Another relevant aspect is the modest evidence that major mergers
are more likely in high-density environments. Four of the nine globs (44%)
with density Ag > 10 2 have had a merger above the t = 0.3 threshold. Of the
globs outside the two largest groups (the "field"), only 21 % have experienced
a merger above this threshold. It is reasonable to assume that major merg-
ers produce predominantly early-type (E+SO) galaxies (see Hernquist in this
volume). Both of these results point to an enhanced fraction of early-type
galaxies in denser environments, in the sense of the observed morphology-
density relation.

4. Mass Functions

The number of globs in the volume above the 32-particle minimum grows
from 11 at z = 6.6 to 205 at z = 2 and remains close to 200 to z = 1. The
78

Fig. 5. The mean mass accretion mass rate for globs over the 1.3 Gyr from z == 1.5 to
z = 1 against the final local galaxy density. Circles show globs that have experienced
a merger with M2/ Ml > 0.3 over the course of the simulation.

largest glob at z = 1 has a baryon mass of 6 X 1011 M 0 • It resides in the

center of the largest dark halo, and its formation history involved mergers
("cannibalism") with over a dozen smaller objects. It is natural to think that
the stellar dynamical counterpart of this object would be a giant elliptical
or cD galaxy.
The mass functions for both globs and dark halos measured at z = 1 is
displayed in Figure 6. The halo mass function is nearly a power law with
slope - 2. The high mass "toe" arises from the constrained nature of the
initial conditions. The region simulated is not a typical 16 - Mpc cube of a
CDM universe, in the sense that it has a 30" density perturbation on a mass
scale of 10 13. 6 Mev imposed upon it. The glob mass function, on the other
hand, is not well described by a pure power but is well fitted by a Schechter
function,

~(M) dM = ~. (M/M.t' exp( -M/M.) dM/M. , (1)

with parameter values ~. = (0.0080 ± 0.0026~pc' M. = (7.2 ± 2.7) X 10 10 M0
and a = -1.39 ± 0.12 . The exponential cutoff in the glob mass function
arises from the inability of the baryons in the largest collapsed halos to
cool efficiently. The reason for the change in low-mass slope relative to the
halos is less obvious. It arises partly from a pile-up in the number of objects
around M. (due to the truncation of the merger hierarchy) and partly from
a decrease in the number oflow-mass objects due to less efficient dissipation.
 79

..-
~
I
CJ halos z=l

-
~-2
~

-----
APM z=O
(M/L=5)

-49~~~~1~O~~~~1~1~~~~1~2~~~~1~3~~~~14

Log M (Mo)
Fig. 6. Differential mass function for globs (dots) and halos (asterisks) at z = 1
along with the local Stromlo-APM luminosity function scaled assuming a baryon
mass-to-light ratio of 5.

The local luminosity function from the Stromlo-APM redshift survey

(Loveday et al. 1992) is also plotted, using a baryon mass-to-light ratio of 5.
The low-mass slope in the experiment is steeper than the faint end slope of
the luminosity function a = -0.97±0.015 in the Stromlo-APM redshift data.
That value of a is somewhat flatter than the canonical a = -1.25 derived
from previous redshift surveys (Sandage, Tammann & Yahi11979; Kirshner
et al. 1983). Observational complications due to the uncertain numbers of
dwarf and low surface brightness galaxies (Phillipps & Disney 1985) might
be a serious problem for the local studies. Bothun, Impey, & Malin (1991)
suggest that the faint-end slope may be closer to 0: ~ -1.6 when corrections
for low surface brightness galaxies are included.
A steeper faint-end slope is expected from analytic considerations such
as a Press-Schechter (1974) analysis. The steeper slope means that there
are many more potential galaxies in a standard CDM model than there are
bright galaxies locally. From Figure 6, at MB = 1010 M0' there are an order
of magnitude more globs in the simulation than faint nearby galaxies. Of
course, it is important to bear in mind that this is far from an "apples-to-
apples" comparison. It is not entirely clear whether an excess in the simula-
80

tion at z ~ 1· is a problem, since analyses of faint galaxy counts in the blue

suggest a ~ignificant evolution of the luminosity function at modest look back
times (Broadhurst, Ellis, & Shanks 1988; Colless et al. 1990; Lilly, Cowie, &
Gardner 1991). It may be that these small objects undergoa single, perhaps
self-destructive, burst of star formation in a manner that allows them to
contribute to the faint counts but "disappear" by the present. Then again,
this may be just wishful thinking. . .
The relation between globs and their parent halos is illuminated in Figure
7, which plots glob baryonic mass versus halo mass, where the latter is
defined as the mass in dark matter within a sphere defining an overdensity
of 500 centered on the site of highest binding energy in the halo. Halos
containing only one glob are plotted as points, while open circles are used
for globs in halos containing multiple globs. The solid line is the baryonic
mass associated with the halo mass under our assumed fib. There are 246
halos with 32 or more particles spanning 3.5 decades in mass. The largest
cluster has a dark mass of 3.5 X 1013 M0 and contains 19 globs within a
radius of 330 kpc defined by the factor 500 density contrast. The second
largest system contains 10 globs and is located 1.1 Mpc from the largest
cluster, infalling with a relative velocity of 400 km S-l. Both of these are
visible in Figure 1. After these two, there are 12 other halos containing more
than one glob. A further 135 halos each contain a single glob. Most of the
remaining halos without a glob contain "proto-globs" that do not meet our
critical mass and density criteria. A small number have had their gas content
modified by the largest cluster environment. All of these "empty" halos are
poorly resolved, having ;S 100 particles. More importantly, all well-resolved
halos contain at least one glob.
Some interesting trends can be seen in Figure 7. All halos with MDM >
10 12 M0 are multiple systems. Halos in the range 2.5 X 1011 M0 < MDM <
10 12 M0 may be either single or multiple. Below this range, down to about
1011 M 0 , one finds mostly singles. For single halos, it is noteworthy that
the mass of the cold baryons within a halo is well approximated by the
baryon equivalent of the halo mass within a density contrast of 500. For the
multiple systems, the mass of the largest glob increases roughly as Mli~ .
These systems show a range of masses down to our resolution limit, with a
relatively large mass ratio between the first and second ranked globs.
The sum of the baryon mass of globs in a given halo can be used to define
a "galaxy formation efficiency" .
MB,tot
€g = (2)
f!bMDM

The top panel of Figure 7 shows this behavior of €g with MDM. The efficiency
is seen to decrease with halo mass. Observational mass estimates based on
models of the intracluster medium in rich clusters indicate just such a trend
 81

.... 0
1 . o.
till 0
W 0 0

~
0

0.1

12
0

,-...,.
0 @

.6 11
0

~
0
~ 0 0 0
Q.(J
0
0
8 0
0 0 0
~ 0 0
0 0
0 0
@
0 0 0
10 0
0
0 0 0
~
8
0 0 0° 0
0
0

11 12 13 14
Log MDM (Mo)

Fig. 7. Glob masses within each halo as a function of halo mass. Circles indicate
halos containing more than one glob. The relation MB ~ ObMDM is a good approx-
imation for single-glob halos. The upper panel shows the galaxy formation efficiency
fg (defined in the text , equation 2) versus halo mass. The efficiency is seen to decline
in the most massive halos.
82

(David et ai. 1990; Arnaud et ai. 1991). Hotter, hence richer, clusters have
a fractionally higher baryon mass in the intracluster plasma, indicating less-
efficient galaxy formation in the largest clusters. Crude extrapolation of the
data in Figure 7 to clusters as massive as Coma or A2256 (MDM ~ 10 15 M 0 )
would predict efficiencies €g ~ 0.2. The inferred values for the ratio of stellar
mass to (stellar + ICM mass) 'in these clusters is 0.24 and 0.15, respectively
(David et ai. 1990). .
The physical explanation for the trend is the following. When viewed at
the same epoch, small halos, by virtue of their smaller virial temperatures,
can have cooling times short compared to both their internal dynamical time
and the Hubble time. Larger halos, on the other hand, will "break out" of
the efficient cooling regime when the virial temperature ' rises above some
critical value. During the progression of hierarchical clustering, the baryons
in these larger wells are allowed a finite time to cool and accrete into local
(galactic) minima of the potential well. For any finite time, this process will
be incomplete and the leftover gas will be shocked onto an adiabat with a
long cooling time. Given the conditions in our experiment, the hot fraction
should increase with halo mass since larger halos will spend less time in the
efficient cooling regime.
Overall, the baryon mass fraction in globs is 18%. This figure is sensi-
tive to the treatment of the physics on sub-galactic scales. For example,
increasing the mass per particle (decreasing the mass resolution) leads to
a smaller cooled fraction because of the inability to resolve cooling in low
mass halos. Adding extra physics (e.g., heating terms or increasing cooling
rates via metal enrichment) will also affect this quantity. Ultimately, the
value of ngal will be determined by a balance between cooling and heat-
ing processes, the latter containing contributions from photoionization and
supernova blastwaves in star forming regions.

5. Biasing
Bias has become an often-used and abused four letter word in the galaxy
formation literature. One of the prime motivations for performing two-fluid
simulations is to demystify the links between light and mass in the universe.
As any good therapist (or even a bad one) will tell you, there is rarely any
reason to expect a simple answer to a complicated question. So it seems to
be with biasing.
The simulation certainly provides a forum for direct investigation of bi-
asing. However, the results obtained in this experiment and in other studies
using independent methods (Katz, Hernquist, & Weinberg 1992; Cen & Os-
triker 1992) have so far not produced a consensus.
In Figure 8, the correlation functions for galaxies and dark matter are
given at two epochs, z = 3.0 and z = 1.0. Two mass thresholds are used for
 83

the galaxies, 32 and 256 particles, corresponding to MB > 3.5 X 10 9 M0 and

MB > 2.8 X 1010 M 0 , respectively. Also shown is the observed correlation
function scaled to redshift z assuming stable clustering ~ob,,(r) = (r /ro)--Y (1+
z)-3+"I where r is a co-moving separation, ro = (5 h- 1 ) Mpc and 'Y = 1.8. For
either mass threshold, the globs at z = 3 are more clustered ("positively"
biased) than the dark matter. A mass dependence in /;,gg is clearly apparent;
clustering of the high mass sample is enhanced by a factor'" 4 over the low
mass sample. By z = 1, the dark matter appears to have "caught up" with
the galaxies and, in fact, the low-mass globs are somewhat less clustered
than the dark matter ("negatively" biased) on scales ;c, 300 kpc. A counts-
in-cells analysis detailed in ESD indicates that merging of globs in the largest
cluster is responsible for the anti-bias signal.
Whether the degree of merging seen in the experiment is correct is cru-
cial to the interpretation of the results. It is likely that future generation
of experiments with enhanced spatial resolution will be necessary to settle
this issue. One possibility is that the degree of galaxy merging in clusters is
as high as the simulation predicts, in which case an anti-bias may exist for
real galaxies, as proposed by Couchman & Carlberg (1992). Alternatively,
the present experiments may be overestimating the amount of merging, in
which case galaxies may be positively-biased tracers of the mass at all times.
Finally, one must be cautious about interpreting results from a single, con-
strained realization. Experiments probing a range of environments must be
done to properly understand how galaxies trace mass throughout the uni-
verse.
Along with biasing in the spatial domain, galaxies may be biased rela-
tive to dark matter in velocity space. Figure 9 shows the one- dimensional
rms pairwise peculiar velocity 0'( r) as a function of separation at z = 3
and z = 1 for dark matter, low- and high-mass globs . The glob velocities are
clearly lower than the dark matter values on small scales, but become compa-
rable on co-moving scales ;c, 3 Mpc. At z = 1, the amplitude of the glob veloc-
ity dispersion increases weakly with radius and is '" 400 km s-1 at 1 Mpc.
This is somewhat larger than the present observed value of ", 300 km S-1
(Davis & Peebles 1983; Bean et al. 1983) . The velocity bias, bv ( r), defined
as the ratio of glob-to-dark-matter velocity dispersions, depends on spatial
scale but appears to be only mildly dependent on redshift and glob mass.
A rough power-law fit is given by bv(r) = 0.7(r/l Mpc)0.2. The lack of any
mass or redshift dependence seems to downplay the role of dynamical fric-
tion in generating this effect. Rather, we suspect (though it is difficult to
prove) that the process of non-violent relaxation (Zurek, Quinn, & Salmon
1988) is mainly responsible. Since galaxies are born "cold" in their parent
halos, they remain cooler than the dark matter in general throughout the
clustering hierarchy.
There have been several other estimates of the velocity bias that range
84

10 5

10·

1000
o4.V
100

10

comovlng r (Mpc)

Fig. 8. Two-point correlation function for the dark matter (solid line), show-
ing all globs with MB > 3.5 X 10 9 M0 (filled symbols) and massive globs with
MB > 2.8 X 10 10 M0 (open symbols) at z = 3 and z = 1. The heavy line in both
panels is the observed galaxy correlation function, scaled assuming stable clustering.

600
..-
I
fIl

~
E 400
..........
b .
200 .... . -,..:•

comoving r (Mpc)

Fig. 9. Mean one-dimensional pairwise peculiar velocity as a function of pair sepa~

ration. Symbols are as in Figure 8.
 85

TABLE I
Group Virial Analysis

View O'gl ( km s 1) Tgi ( kpc) Mdyn (l013 M 0 ) Est no

X axis 337 114 2.85 0.32

Yaxis 403 106 3.78 0.41
Z axis 330 82 1.96 0.21

Average 356 101 2.86 0.32

in magnitude from 0.3 to 1 (see Carlberg 1991 and references therein). The
main source of confusion in the various estimates of velocity bias is that
each arises from a different method of tracing galaxies in numerical simula-
tions. The strength of our approach over previous experiments is that our
galaxy tracer population is more physically reasonable than the single fluid
(DM only) schemes employed in many of the previous studies. Collisionless
schemes have not been shown to accurately trace galaxies through clustering
(Summers, Evrard, & Davis 1993). The final chapter of the book of biasing
is far from being written.

6. Cluster Mass Estimates

Using 26 globs within the virial radius of the central cluster (defined below),
we have applied a projected virial mass estimator of the form

(3)

where {Tgl and Tgl are the glob one-dimensional velocity dispersion and pro-
jected harmonic radius and Mdyn is the inferred binding mass of the group.
The results for three orthogonal projections of the simulation are summa-
rized in the following table.
The estimated mass should be compared to the true binding mass. There
is, unfortunately, considerable leeway in defining the latter, since halos do
not have well defined edges. The virial radius defined in the standard spher-
ical model (Peebles 1980) is that within which the mean density is a factor
of 170 above the background. Using this definition, we obtain a virial radius
of 530 kpc encompassing a total mass of 6.1 x 10 13 M 0 . On average, the
dynamical mass estimate underestimates the true binding mass by a factor
of two.
86

Observationally, an estimate of the mean mass density in the universe

is obtained by multiplying the mass-to-light ratio in groups with the mean
luminosity density generated by galaxies. Typically this leads to values of
20% of the closure value, i.e. no = 0.2. We can perform a similar procedure
f'-.J

with the simulation data by using glob mass in place of luminosity. The
inferred no estimate is

(4)
where Mgl is the net baryonic mass of globs within the group and fgl is the
global fraction of baryons in globs. The first term is the simulation equivalent
of a mass-to-light ratio and the product of the remaining terms takes the
place of a normalized luminosity density. We find Mgl = 1.63 X 10 12 M0 and
/gl = 0.18, which produc~s a mean estimate of no = 0.32. The true value in
the simulation is, of course, n = 1. Dynamical biases in clusters may be large
enough to reconcile observations with a closed universe. As with velocity
biasing above, we suspect that non-violent relaxation plays the key role in
generating this effect .

7. Future Directions
Inclusion of dissipation in a separate baryon component is essential in cre-
ating a two-phase structure with cold, dense objects (globs) having char-
acteristic dimensions of real galaxies. The resulting compact nature of the
baryonic globs reduces their merger cross section, allowing them to survive as
distinct, self-bound entities within the larger cluster potential. As is evident
in Figure 1, a two-phase structure is not a natural outcome of collisionless
clustering with no dissipation - halo substructure is efficiently erased dur-
ing mergers (the "overmerger problem"). Although the merger rate for globs
is reduced relative to that of dark halos, it still may overestimate that of
real galaxies, whose optical radii are generally somewhat smaller than the
13 kpc resolution in this experiment.
The formation of disk galaxies from hierarchical clustering of an initially
Gaussian random density field has now been verified by direct experiment.
Although this result is neither "new" nor surprising (in the sense that it
was the theoretical expectation), it is a milestone for numerical studies to
achieve this level of resolution in a cosmological (i .e., large-scale) setting. It
opens the way for further exploration of the systematic properties of galaxies
and their dependence on local environment. The tight power-law relation
between cold baryon mass and rotation velocity found for the simulated
globs is a precursory step toward understanding the observed Tully-Fisher
relation.
Future experiments with more physics (e.g., star formation and feedback)
and enhanced spatial resolution will provide increasingly reliable answers to
 87

the many questions still left regarding biasing, merger frequencies, and galac-
tic internal properties. The ultimate (and we're talking Ultimate here) goal
of numerical simulations of cosmic structure is to create "virtual universes"
with the same degree of realism as the one we live in. That means all the
principal mass components - (stars, gas, radiation, dark matter form( s)
- must be present, distributed in their respective phase spaces with nearly
unlimited dynamic range. Given such a bounty of information, synthetic ob-
servations at multi-wavelengths will be made possible. The purpose is not to
make pretty pictures (though it is an engaging by-product ... ) but rather to
increase the constraining power of observations, particularly of non-linear
structure, on cosmological theories as well as to improve our understanding
of the star formation history of galaxies and the evolution of multi-phase
media.
Clearly, we are just getting started and there is much work to be done.
One area ripe for development in the not-so-distant future is combined X-ray
and optical analysis of clusters of galaxies (see Metzler & Evrard' s poster
contribution to this meeting) . Figure 10 shows the central cluster in the
simulation as it might be seen in the ROSAT PSPC if it were placed at
a redshift z = 0.09. The X-ray morphology is similar to the poor group
centered on NGC 5044 imaged by David (1993). The temperature of the
intracluster gas is about 2 keY. Superposed on the X-ray contours is an
"optical" image of the globs within the cluster region. Not surprisingly, the
X-rays peak at the location of the central, dominant glob which itself sits
at the bottom of the dark matter group potential well. The asphericity in
the outer isophotes arises from material acquired in recent mergers that has
not yet settled into hydrostatic equilibrium. With full understanding of the
dynamical history of the cluster, one can explore systematic effects (e .g. ,
departure from hydrostatic equilibrium, temperature gradients) relevant to
observational estimates of cluster binding masses. The ability to view the
cluster both "optically" and in X-ray emission provides independent chan-
nels of analysis and effectively doubles the level of contact with observat ions.
This work is currently in progress.

Acknowledgements

I thank my collaborators, F . J. Summers and M. Davis, for giving me free

reign in presenting our collective results. I also thank the esteemed organizers
of this meeting, H. Thronson and M. Shull, for their skill in running such
a fine event and their patience in waiting for invited contributions. This
research was supported by NSF, in part through allocation of CPU time
on a CRAY Y-MP at the San Diego Supercomputer Center, and NASA, in
part through Theory Grant NAGW-2367.
88

200

100

()
0.
~ .
0

-100

-200

Fig. 10. Combined "optical" and X-ray view of the central cluster placed at a
viewing redshift z = 0.09. Contours show a simulated ROSAT PSPC observation
(0.1- 2.4 keY) with minimum contour 10- 3.8 cnts s-l arcmin - 2 and spacing 10°·3.
The points are particles belonging to "galaxies" within the region imaged. .

References
Arnaud, M., Rothenfiug, R ., Boulade, 0., Vigroux, L. & Vangioni-Flam, E . 1991, A&A,
254 , 49.
Bean, A.J., Efstathiou, G., Ellis, R .S., Peterson, B .A . & Shanks, T. 1983, MNRAS, 205,
605.
Bertschinger, E. 1987, ApJ, 323, L103 .
Blumenthal, G .R., Faber, S.M., Primack, J .R., & Rees, M .J . 1984, Nature, 311 , 517.
Bothun, G .D ., Impey, C .D . & Malin, D.F . 1991, ApJ, 376, 404.
Broadhurst, T.J ., Ellis, R .S. & Shanks, T . 1988, MNRAS, 235, 827.
Carignan, C . & Freeman, K.C . 1988, ApJ, 332, L33.
Carlberg, R . G. 1990, ApJ, 359, Ll.
Casertano, S. and Hut, P . 1985, ApJ, 298, 80.
Cen, R . & Ostriker, J. 1992, ApJ, 399, LIl3.
 89

Chiang, W.H., Ryu, D. & Vishniac, E.T. 1989, ApJ, 339, 603.
Cole, S. & Lacey, C. 1992, preprint.
Colless, M.M., Ellis, R.S., Taylor, K. '& Hook, R.N. 1989, MNRAS, 244,408.
Couchman, H.M.P. & Carlb~rg, R.G. 1992, ApJ, 389, 453 .
David, L.P 1993, preprint.
David, L.P., Arnaud, K.A., Forman, W. & Jones, C. 1990, ApJ, 356,12.
Davis, M. & Peebles, P.J.E. 1983, ApJ, 267, 465.
Dressler, A. 1980, ApJ, 236, 351.
Evrard, A.E. 1988, MNRAS, 235, 911.
Evrard, A.E., Summers, FJ & Davis, M. 1993, ApJ, submitted.
Fall, S.M. & Efstathiou, G. 1980, MNRAS, 193, 189.
Frenk, C.S. 1991, Physica Scripta, T36, 10.
Hernquist, L. & Katz, N. 1989, ApJS, 70, 419,.
Katz, N., Hernquist, L. & Weinberg, D.H. 1992, ApJ, 399, L109.
Kauffmann, G. & White, S.D.M. 1992, preprint.
Kirshner, R.P., Oemler, A., Schechter, P.L. & Schectman, S.A. 1983, AJ, 88, 1285.
Larson, R.B., Tinsley, B.M. and Caldwell, C.N. 1980, ApJ, 231, 692.
Lilly, S.J., Cowie, L.L. & Gardner, J.P. 1991, ApJ, 369,79.
Monaghan, J.J. 1992, ARA&A, 30, 394.
Peebles, P.J .E. 1980, The Large-Scale Structure of the .universe, (Princeton : Princeton
University Press).
Persic, M. & Salucci, P. 1988, MNRAS, 234, 131.
Phillipps, S. & Disney, M. 1985, A&A, 148, 234.
Postman, M. & Geller, M.J. 1984, ApJ, 281, 95.
Press, W.H. & Schechter, P. 1914, ApJ, 187,425.
Rees, M.J. & Ostriker, J.P. 1977, ApJ, 179, 541.
Sandage, A., Tammann, G.A. & Yahil, A. 1979, ApJ, 232, 352.
Silk, J.1. 1977, ApJ, 211, 638.
Summers, F J, Evrard, A.E. & Davis, M. 1992, in The Evolution of Galaxies and Their
Environment: Contributed Papers, eds. D. Hollenbach, J .M. Shull & H. Thronson,
(NASA conf. pub.), in press.
Tath, G. & Ostriker, J.P. 1992, ApJ, 389, 5.
White, S.D.M. & Rees, M.J . 1918, MNRAS, 183, 341.
Zurek, W., Quinn, P. & Salmon, J. 1988, ApJ, 330, 519.
 POPULATION SYNTHESIS AND SPECTRAL EVOLUTION
OF GALAXIES

GUSTAVO BRUZUAL A.
Gentro de Investigaciones de Astronomia (GIDA), A.P. 264, Merida, Venezuela

Abstract. The basic assumptions behind Population Synthesis and Spectral Evolution
Illodels are reviewed. The nUlllerical problems encountered by the standard population
synthesis technique, when applied to .I llodels with truncated star fonnation rates, are
described. The Isochrone Synthesis algorithm is introduced as a Illeans to circulllvent
these problems. A sununary of results from the application of this algorithm to Illodel
galaxy spectra by Bruzual & Charlot (1993) follows.
Key words: Stellar Evolution - Population Synthesis - Spectral Synthesis - Spectral
Evolution

1. The Population Synthesis Problem

Stellar evolution theory provides us with the functions Teff(m, Z, t) and

L( m, Z, t) which describe the behavior in time t of the effective temperature
Teff and luminosity L of a star of mass m and metal abundance Z. For
fixed m and Z, L(t) and Teff(t) describe parametrically in the H-R diagram
the evolutionary track for stars of this mass and metallicity. The Initial
Mass Function ¢>( m) indicates the number of stars of mass m born per unit
mass when a stellar population is formed. The Star Formation Rate 'l!(t)
gives the amount of mass transformed into stars per unit time according
to ¢>( m). The Population Synthesis Problem can then be stated as follows.
Given a complete set of evolutionary tracks and the functions ¢>( m) and
w(t), compute the number of stars present at each evolutionary stage in
the H-R diagram as a function of time. To solve this problem exactly we
need additional knowledge about the Metal Enrichment Function «( t), which
gives the time evolution of the chemical abundance of the gas from which
the successive generations of stars are formed. «( t) is also provided by the
theory of stellar evolution. For simplicity, it is commonly assumed that ¢>( m)
and 'l!(t) are decoupled from «(t), even though it is recognized that in real
stellar systems these three quantities are most likely closely interrelated.
The Spectral Evolution Problem can be solved trivially once the Population
Synthesis Problem is solved, provided that we know the Spectral Energy
Distribution at each point in the H-R diagram representing an evolutionary
stage in our set of tracks. The discussion that follows is based in the work
of Charlot & Bruzual (1991, hereafter CB91) and Bruzual & Charlot (1993,
hereafter BC93). We ignore chemical evolution and assume that at all epochs
stars form with solar metallicity and the same ¢( m).
Let Nt be the number of stars of mass Mi born when an instantaneous
burst of star formation occurs at t = o. This kind of burst population has

91
I M. Shull and H A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, 91-104.
© 1993 Kluwer Academic Publishers.
92

been called Simple Stellar Population (SSP, Renzini 1981). When we look
at this population at later times, we will see the stars traveling along the
corresponding evolutionary track. If the stars live in their kth evolutionary
stage from time ti,k-l to time ti,k, then at time t the number of stars of this
mass populating the kth stage is simply

N. (t) = {Nt, if ti,k-l. ~ t < ti,ki (1)

t,k 0, otherwIse.

For an arbitrary SFR, 'P(t), we compute the number of stars 1]i,k(t) of mass
Mi at the kth evolutionary stage from the following convolution integral

(2)

which in view of (1) can be written as

1]i,k(t) = Nt l min( t,t.,k)

t',k-l
'P(t - t')dt'. (3)

From (3) we see that the commonly heard statement that the. number of
stars expected in a stellar population at a given position in the H-R diagram
is proportional to the time spent by the stars at this position, i.e.

(4)

is accurate only for a constant 'P(t). For non-constant SFR's, the integral in
(3) assigns more weight to the epochs of higher star formation. For instance,
for an exponentially decaying SFR with e-folding time T, 'P(t) = exp( -tiT),
we have

1]i,k(t) ()( NNexp[-(t - ti,k)/Tj- exp[-(t - ti,k-l)/T]}. (5)

'P(t) was stronger at time (t - ti,k) than at time (t - ti,k-d, which is clearly
taken into account in (5).
A prerequisite for building trustworthy population synthesis and spectral
evolution models is an adequate algorithm to follow the evolution of con-
secutive generations of stars in the H-R diagram. This goal is accomplished
by the standard technique described above provided that the function 'P(t)
extends from t = 0 to t = 00. Special caution is required if 'P(t) becomes 0
at a finite age. As an illustration, let us consider the case of a burst of star
formation which lasts for a finite length of time T,

if 0 ~ t ~ Ti
(6)
otherwise,
 93

or equivalently,
if t - T ~ t' ~ t;
w(t - t') = { 0,wo , otherwise.
(7)

In this case, equation (2) reduces to

i k(t) = { > 0, if [ti,k-~' ti,k] n [t - T, t] :f. 0; (8)

"I , 0, otherWIse.

From (8) we see that "Ii,k(t) can be = °

depending on the value of T
and on the value of t chosen to sample the stellar population. The most
extreme case is that of the SSP (T = 0), for which w(t) = 15(t), and "Ii,k(t)
in (8) is identical to Ni,k(t) in (1). For a typical set of evolutionary tracks,
there is no grid of values tj of the time variable t for which all the stellar
evolutionary stages included in the tracks can be adequately sampled for
arbitrarily chosen values of T. This is obviously an undesirable property of
population synthesis models. The models should be capable of representing
galaxy properties in a continuous and well behaved form, independent of the
sampling time scale t. Consequently, standard population synthesis models
for truncated w(t) as given by (6) will reflect the coarseness of the set of
evolutionary tracks, and may miss stellar evolutionary phases depending on
our choice of t. This results in unwanted and unrealistic numerical noise in
the predicted properties of the stellar systems studied with the synthesis
code (see example in CB91).
A solution to this problem is to build a set of evolutionary tracks with a
resolution in mass which is high enough to guarantee that all evolutionary
stages, Le. all values of kin (2), can be populated for any choice of t or T in
(6). In other words, there must be tracks for so many stellar masses in this
ideal library that for any model age, we have at hand the position in the
H-R diagram of the star which at that age is in the kth evolutionary stage.
The realization of this library is not possible with present day computers.
This limitation can be circumvented by careful interpolation in a relatively
complete set of evolutionary tracks.

2. The Isochrone Synthesis Algorithm

The isochrone synthesis algorithm described in this section allows us to com-
pute continuous isochrones of any age from a carefully selected set of evolu-
tionary tracks. The isochrones are then used to build population synthesis
models for arbitrary SFR's w(t), without encountering any of the problems
mentioned above. This algorithm has been used by CB91 and BC93. The
details of the algorithm follow.
A relationship is built between the MS mass of a star M and the age t
of the star during the kth evolutionary stage (CB91). Ignoring mass loss is
justified because M is used only to label the tracks. For the set of tracks
94

used by BC93 (described in the next section) 221 different relationships

log M vs. log t are built, which can be visualized as 221 different curves
in the (log t, log M) plane. We derive by linear interpolation the MS mass
mk( t') of the star which will be at the kth evolutionary stage at age t', given
by
log mk(t') = Ak,i log(Mi) + (1 - Ak,i) log(Mi+l), (9)
where
log ti+l,k - log t'
Aki= . (10)
, log ti+l,k - log ti,k
ti,k represents the age of the star of mass Mi at the kth evolutionary stage,
and

and
Mi+l ~ mk(t') < Mi.
The procedure is performed for all the curves that intersect the log t = log t'
line. We thus obtain a series of values of mk which must now be assigned
values of log L and log Tel I in order to define the isochrone corresponding
to age t'.
To compute the integrated properties of the stellar population, we must
specify the number of stars of mass mk. This number is determined from
the IMF, which we can write generically as

(11)

Of this number of stars,

(12)
stars are assigned the observational properties of the star of mass Mi at the
kth evolutionary stage, and

(13)

stars, the observational properties of the star of mass Mi+l at the same kth
stage. This procedure is equivalent to interpolating the tracks for the stars of
mass Mi and Mi+l to derive the values oflog L and log TeJf to be assigned
to the star of mass mk, but this intermediate step is unnecessary. In (11)
mJ; and mt are computed from

(14)

In this case mk-l, mk, and mk+l represent the masses obtained by interpo-
lation at the given age in the segments representing the (k - 1)th, k t\ and
 95

(k + 1 )th stages, respectively. The IMF has been assumed to be a power law
of the form (Salpeter 1955)

(15)

For a given ~(t) equation (2) again gives the number of stars of mass Mi
at the kth evolutionary stage. If fi,k().) represents the SED corresponding to
the star of mass Mi during the kth stage, then the contribution of the TJk,i (t)
stars to the integrated evolving SED of the stellar population is simply given
by
Fi,k().) = TJi,k(t)fi,k().). (16)
The resulting SED for the stellar population is given by

F()., t) =L Fi,k()., t). (17)

i,k

3. Stellar Ingredients
The library of stellar evolutionary tracks used in the isochrone synthe-
sis models of BC93 was compiled in CB91, to which I refer for details.
The tracks are based largely on the computations of Maeder & Meynet
(1989), using their revised main sequence lifetimes for stars with masses be-
tween 1.3 M0 and 2.5 M0 (Ma~der 1991; BC93; Bertelli, Bressan, & Chiosi
1992). These authors, however, did not compute the evolution of low and
intermediate-mass stars through several important stages. Their tracks were
complemented in CB91 with tracks from other updated calculations, while
limiting the inconsistencies in the stellar input physics between the different
sets of tracks. The evolution of low and intermediate-mass stars through
the luminous double-sheIl-burning regime at the end of the AGB was added
semi-empirically. Magris & Bruzual (1993) refined the prescription adopted
in CB91 for the post-AGB evolution of low and intermediate-mass stars
to study the far- UV properties of early-type galaxies. They added the post-
AGB evolution of stars with initial masses in the range 2.5 M0 :::; m :::; 7 M0
using the tracks of Schonberner (1981, 1983). The connection between the
CB91 AGB tracks in this mass range and the Schonberner tracks was per-
formed by Magris & Bruzual by means of either the Then & Truran (1978)
or the Weidemann (1987) relationship between the initial mass of a star
and its core mass after ejection of the planetary nebula at the tip of the
AGB. The results reported in the next section were computed with the set
of tracks extended according to the Weidemann (1987) relationship. The fi-
nal tracks include the evolution of all stars initially less massive than 7 M0
as planetary nebulae (PN s), bare PN nuclei (PNN s, i.e. after dissipation of
the envelope), and through the white-dwarf cooling sequence.
96
The library of stellar spectra is described in detail in BC93. The spec-
tra in the visible are taken from the Gunn & Stryker (1983) atlas, which
includes 175 stars of most spectral types and luminosity classes. Persson
(1987, unpublished) supplemented the spectra of all but eight of these stars
in the near-IR (1.22 ::; ,\ ::; 2.56 JLm with 0.2 JLm resolution) with data from
the Kuiper Airborne Observatory. The resulting library of visible/near-IR
spectra covers the luminosity classes III, IV, V and all spectral types from
05 to M8. Observations from the International Ultraviolet Ezplorer (IUE)
were used to extend the spectra of the Gunn and Stryker stars in the range
1200 ::; ,\ ::; 3240 A. Each star is assigned a UV spectrum according to its
spectral type and luminosity class. We extend the spectra of earlier-type
stars in the range 240 ::; ,\ < 1200 A using Kurucz (1979) model atmo-
spheres. The procedure adopted to assign a stellar spectrum from the library
to each location on the evolutionary tracks is described in BC93.

4. Summary of Results
Some of the most relevant results obtained by BC93 and Magris & Bruzual
(1993) are presented in this section. The reader is referred to these two
papers for a detailed discussion. All the models shown were computed for a
power-law IMF with the Salpeter (1955) exponent x = 1.35. In all models
the IMF extends from mL = 0.1 M0 up to mu = 125 M 0 . The total mass of
the model galaxies is 1 M 0 . The constant c in (15) is determined from the
normalization condition 4>( mL, mu, 1 + z) = 1. Model SEDs were computed
at 221 unequally spaced time steps in the range from 0 to 20 Gyr. Each SED
covers the range from 5 A to 100 JLm in 1206 steps.
It is important to compare the properties of the population synthesis
models to observations of stellar systems whose age and metallicity is well
determined. This is a means to test to what extent the adopted relationships
between stellar color and magnitude and effective temperature and luminos-
ity (or surface gravity) introduce systematic shifts between the predicted and
observed isochrones in the C-M diagram. Figure 1 compares the (B, B - V)
observations of the clusters M67 and the Hyades with the isochrones ob-
tained from the BC93 models. These two clusters have nearly solar metallic-
ity: M67 ([Fe/H]~ 0.01), the Hyades ([Fe/H]~ 0.15). For M67 we adopted a
distance modulus of 9.5 mag and a color excess E(B - V) = 0.06 mag (Janes
1985). For the Hyades a distance modulus of 3.4 mag (Peterson & Solensky
1988) and E(B - V) ~ o. Estimates of the ages of these clusters vary from 4
to 4.3 Gyr for M67 and from 0.5 to 0.8 Gyr for the Hyades. The isochrones
are shown at 4 Gyr (M67) and 0.6 Gyr (Hyades). The data points for M67
are from Eggen & Sandage (1964, open circles), Racine (1971, filled circles),
Janes & Smith (1984, triangles), and Gilliland et al. (1991; crosses). For the
Hyades the observations are from Upgren (1974, triangles), Upgren & Weis
 97

-2
6
Hyades
M67 0

8
2

4
10

.....
0

.- 10 6

12
> > 8

. .
0. ,
. ....
,

Z··
~

14
10 .~.
....
•.. 12 .... ~.'. ~~\ .
. ........' . ":\
16 ~

to
~
14
· .,..,...
.
" t':t!
"'.
o 0 00
18 · ....-.
• t .. ..... 16
00 0
.. ! .':11:
. ~.. . .... :- 18
: .... - . -III

~LLLOLLLLL.L5LLLLLLLLLLLL~2
j
o .5 1.5 2 -.5
B-V
B- V

Fig. 1. Calibration of the models in the C-M diagram.

(1977, filled circles), and Micela et al. (1988, open circles). Despite some
discrepancies seen in Figure 1 (discussed in detail by BC93) the agreement
between the predicted isochrones and the loci in the C-M diagrams of these
clusters may be regarded as satisfactory.
Figure 2a shows the evolution in time of the SED for the SSP model. The
evolution is fast and is dominated by massive stars during the first Gyr in the
life of the SSP (6 top SEDs). The flux seen around 2000 A at 4 and 7 Gyr is
produced by the turn-off stars. The UV-rising branch (Burstein et al. 1988,
Greggio & Renzini 1990) seen after 10 Gyr is produced by the PAGB stars.
These stars are also responsible of the decrease in the amplitude of the 912
A discontinuity observed after 4 Gyr. The SSP model is the basic ingredient
which, together with the convolution integral (2), is used to compute models
with arbitrary SFRs and equal IMF. For illustration we show in Figure 2b
the evolution of a model with w(t) = exp( -tiT) for T = 3 Gyr. The UV to
optical spectrum remains roughly constant during the main episode of star
98

-2

,
"
-:::g

~
.::..
r.: -4
bO
.2

300 1000 3000 10000 300 1000

A (A)

Fig. 2. Evolving spectral energy distributions. FA in frame (b) has been multiplied
by 100 to use a common vertical scale.

formation because of the continuous input of young massive stars, but the
near-infrared light rises as evolved stars accumulate. When star formation
drops, the spectral characteristics at various wavelengths are determined by
stars in advanced stages of stellar evolution.
Using the quantitative technique described in BC93, we derive best fitting
models for the SED of an average Elliptical galaxy and the Irregular galaxy
N4449. Figure 3a shows the SED of a 1 Gyr burst model and that of an
"average" quiescent elliptical galaxy with a UV spectrum of the type of
NGC 4472 (Burstein et a1. 1988). The model is shown at the best-fitting
age of 13.5 Gyr. The observed spectrum between 3000 A and 1.06 J.Lm is
an average over five first-ranked (radio-quiet) elliptical galaxies by Yee &
Oke (1978). We added the three near-infrared fluxes in such a way that the
resulting spectrum has the mean (V - K) = 3.20, (J - K) = 0.90, and
(J - H) = 0.70 colors of a standard elliptical. The fit is quite good over
 99

eyr burs t at 13.5 eyr - - • Cons SFR at 1 Gyr - - •

-4. 5
Average Elliptical -- - - • NeC 4449 - - - •

-5
,
a
:::E

~
....l
--:: -5.5
~

Q()
o

-6

- 6.5

.1 .2 .5 2 .1 .2 .5 2
A ().tm)

Fig. 3. Best fit to Elliptical and Irregular galaxy SED's. F).. in frame (b) has been
divided by 100 to use a common vertical scale.

the whole spectral range. A similar comparison is performed in Figure 3b

for a model with constant star formation rate and the observed SED of the
Irregular galaxy NGC 4449. The best-fitting age in this case is 1 Gyr. The
observed spectrum from 1250 A to 2.2 11m was assembled from IUE, optical,
and near-IR data by Ellis & Bruzual (1983, unpublished). Blueward of 1500
A the model shows an excess of UV radiation, possibly due to internal dust
in the galaxy that extinguishes some of the ultraviolet light. The nebular
component, responsible for the observed emission lines, is not included in
BC93 models.
Figure 4 shows a comparison of models with different timescales of star
formation with the characteristic spectra of galaxies of various morphological
types obtained by Kennicutt (1992). The models shown were computed with
~(t) = exp( -tiT) for T ranging from 1 to 7 Gyr. Models with increasing
timescales of star formation can reproduce the spectra of galaxies from early
100

.8

.6

.4

'"
IC:
0
.2
0

+
r.:
tUl
.8
~

.6

.4

.2

0
4000 5000 6000 7000 4000
J... (l)

Fig. 4. Best fit to optical SED of galaxies of various morphologies.

to late morphological types. The main absorption features and the 4000 A
breaks of the models match reasonably well the observations. The models
selected here have the typical U - V and V - K colors observed in E, Sb, Sbc,
and Sc galaxies. This suggests that the spectral fits are likely to extend down
into the near-infrared. Hence, BC93 models can reproduce reasonably well
the observed spectra or photometric properties from the UV to the near-IR
of nearby galaxies of various Hubble types, in which young massive stars,
old low-mass stars, or mixtures of stars of all masses and ages produce the
light. These models constitute therefore a reliable mean to investigate the
spectral evolution of any stellar population dominated by solar metallicity
stars. However, the spectral energy distributions alone do not permit to
determine unambiguously the age of a stellar population. We note that the
best-fitting ages determined in the previous comparisons of models with
observed spectra are, in fact, very loosely constrained (see details in BC93).
Magris & Bruzual (1993) have applied the isochrone synthesis models of
 101

- 1
~

o
o
e
If")

t...'"
OIl

..2 -1.5
I

-2

- - 1 Gyr burst at 13.75 Gyr - - 1 Gyr burst at 12.5 Gyr

• N4472 (Burstein et aJ.) • N3379 (Burstein et aJ.)

1500 2000 2500 3000 1500 2000 2500 3000

A (..\.)

Fig. 5. UV of nearby ellipticals.

BC93 to the study of the UV upturn in the spectrum of Elliptical galax-

ies. Figure 5 shows a comparison of the best fitting BC93 models and the
Burstein et a1. (1988) data for N4472 and N3379. The UV flux, shown in
these models is due to the passive evolution of the solar metallicity old stel-
lar population in the 1 Gyr burst model reaching the PAGB evolutionary
phase (nuclei of planetary nebula). The observed level of UV flux can be
reached without invoking any amount of star formation beyond the initial
burst . In the case of M32, Magris & Bruzual conclude that there is definite
spectral evidence that a recent major event of star formation took place
in this galaxy about 4 Gyr ago. The spectrum of N4649 cannot be under-
stood unless star formation (in minor amounts) has been taking place in this
galaxy until recent epochs, or that a UV bright stellar component is missing
in the BC93 synthesis models (Greggio & Renzini 1990).
102

5. Conclusions

Considerable progress has been achieved in recent years in the field of popu-
lation synthesis thanks to a combination of different factors. First, complete
libraries of evolutionary tracks and stellar spectra have become available to
the general user. Second, it has been realized by model builders that more
sophisticated algorithms are needed in order to fully exploit the potential
of these libraries, and to produce reliable population synthesis and spec-
tral evolution models which represent stellar systems in a realistic fashion.
In particular, the isochrone synthesis models of BC93 discussed in this pa-
per can reproduce reasonably well the observed spectral and photometric
properties from the UV to the near-IR of nearby galaxies of various Hubble
types. These models constitute therefore a reliable mean to investigate the
spectral evolution of solar metallicity stellar populations for any w(t). The
possibility that some of the other evolved hot star candidates (Greggio &
Renzini 1990) contribute significantly to the UV flux is not excluded, but
for many ellipticals, the normal evolution of stars beyond the AGB produces
enough flux to account for their UV rising branch. Spectrophotometric mod-
els which also follow the chemical evolution of the stellar population need
further development.

References
Bertelli, G., Bressan, A., & Chiosi, C.: 1992, ApJ 392, 522
Bruzual A., G., & Charlot, S.: 1993, ApJ 405, 538,BC93
Burstein, D., Bertola, F., Buson, L.M., Faber, S.M., & Lauer, T.R. : 1988, ApJ 328,440
Charlot, S., &; Bruzual A ., G.: 1991, ApJ 367,126 (CB91)
Eggen, O.J., & Sandage, A.R.: 964, ApJ 140, 130
Gilliland, R .L., Brown, T .M ., Duncan, D.K., Suntzeff, N.B., Wesley Lockwood, G.,
Thompson, D.T ., Schild, R.E., Jeffrey, W .A., & Penprase, B.E.: 1991, AJ 101, 541
Greggio, L., & Renzini, A .: 1990, ApJ 364, 35
Gunn, J.E., & Stryker, L.L.: 1983, ApJS 52, 121
Then, I., Jr., & Truran, J.W .: 1978, ApJ 220, 980
Janes, K.A.: 1985, in Calibration of Fundamental Stellar Quantities, lAD Symposium No.
111, ed(s)., D.S. Hayes, L.E. Pasinetti, and A.G. Davis Philip, (Dordrecht: Reidel),
361
Janes, K.A ., & Smith, G.H .: 1984, AJ 89, 487
Kennicutt, R. C.: 1992, ApJS 79, 255
Kurucz, R .L .: 1979, ApJS 40, 1
Maeder, A., & Meynet, G.: 1989, Af!A 210, 155
Maeder, A.: 1991, ,(private communication)
Magris C., G., & Bruzual A ., G .: 1993, ApJ , (submitted)
Micela, G ., Sciortino, S., Vaiana, G.S., Schmitt, J .H.M.M., Stern, R.A., Harnden, F.R.,
Jr., and Rosner, R.: 1988, ApJ 325, 798
Racine, R.: 1971, ApJ 168, 393
Renzini, A. : 1981, Ann. Phys. Fr. 6, 87
Salpeter, E.E.: 1955, ApJ 121, 161
Schonberner, D.: 1981, Af!A 103, 119
Schonberner, D.: 1983, ApJ 272, 708
Dpgren, A.R.: 1974, ApJ 193, 359
 103

Upgren, A.R., & Weis, E .W.: 1977, AJ 82,978

Weidemann, V .: 1987, A8A 188,74
Yee, H.K.C., & Oke, J.B.: 1978, ApJ 226, 753
 SEARCHING FOR HIDDEN GALAXIES
J. I. DAVIES
Department of Physics and Astronomy, University of Wales College of Cardiff

Abstract. There is increasing evidence to suggest that there are many more galaxies than
we would have expected from extrapolation of observations of the nearby Universe. Deep
observations of the distant Universe reveal a much larger number density of galaxies than is
observed locally. The spectra of distant quasars show far too many absorption-line features
to be accounted for by the local galaxy population. Big Bang nucleosynthesis predicts a
factor of 5 - 10 times more baryonic matter in the Universe than is seen in galaxies. The
brightness of the Earthly night sky (about 23 blue magnitudes arcsec- 2 (BJ.&) at a dark
site) sets quite stringent limits on what we can hope to detect as Earth bound observers.
Is our perception of the Universe highly biased by the position from which we observe it?
Or more explicitly for this paper, how many galaxies may we be missing when we survey
the sky?

1. Introduction
Our knowledge of the local population of galaxies comes from observations
of the Luminosity Function (LF) of galaxies, that is the number of galaxies
per unit volume per unit interval ofluminosity or magnitude. The most com-
monly used functional form for the LF is the so-called Schechter function
(Schechter 1976). This function has a power-law form at low luminosities and
an exponential "choke" to give the necessary cut-off at higher luminosities.
Although LFs have proven to be useful with regard to our understanding
of stars, their usefulness with regard to galaxies, I feel, is much less clear.
This is because we view most galaxies as extended objects, so there is at
least one other parameter necessary in their description (size?). This lim-
itation, and the general reluctance to give up LFs, is probably inhibiting
our knowledge of how many galaxies there are; the LF alone does not entice
us into searching for galaxies that have "non-normal" surface brightnesses.
A potentially more useful function for galaxies is the Bivariate Brightness
Function (BBF). This function describes the joint distribution of luminosity
and size (or equivalently surface brightness). The BBF describes not only
the total range of galaxy luminosities, but also the range of galaxy sizes over
which this luminosity is possible.
To find the BBF we must first of all understand the ways in which galax-
ies are selected and the biases this selection process may introduce into the
data. For example, to be part of some "complete" sample, galaxies are typi-
cally selected by their apparent magnitude within some limiting isophote or
their apparent size at some limiting isophote (or possibly a combination of
both). For galaxies of the same surface brightness (SB) the volume of the
Universe sampled out to some predefined limit depends on the 3/2 power
of the intrinsic luminosity, or, for a diameter-limited sample, the cube of its
size (cosmological corrections apart). The corrections for completeness (for

105
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, 105-126.
© 1993 Kluwer Academic Publishers.
106

the different volumes sampled) for galaxies of various intrinsic luminosities

and sizes are thus well defined and straight forward. The problem is more
complicated when galaxies have different SBs. For example, galaxies of a
given intrinsic luminosity can be observed to be small (at a given isophotal
detection level) when they have either high or low SB (Disney 1976). Thus,
the volume of the Universe a galaxy can lie in and still be included in the
sample is a sensitive function of both its SB and its total intrinsic luminosity;
this is a problem I shall discuss in much more detail in §3.
Knowing the BBF rather than the LF would enable us to answer ques-
tions like:

1. Where does most of the luminosity of the Universe come from? From
faint galaxies, from bright galaxies, from LSB galaxies, or from High
Surface Brightness (HSB) galaxies?

2. Which kind of galaxies, along any arbitrary line of sight, are most likely
to cause absorption features in, say, a QSO spectrum?

In addition, a knowledge of the BBF could be an important constraint on

models of galaxy formation and evolution (see for example Dekel & Silk,
1986). Like the HR diagram for stars, the BBF may well be the Rosetta
stone that leads to a much clearer understanding of the nature of galaxies.
How can the BBF help us answer intriguing questions like 1 and 2 above?
Let us consider a simple example. The BBF, n(L,~) 1 [here I choose to use
luminosity (L) and surface brightness (~)l is the number of galaxies Mpc 3
per unit ~ and L. For the present purposes, I shall choose a parameterization
that is not ideal, but it does have the merit of touching familiar observational
ideas and measurements. Assume that the population in L and ~ can be
described by the distribution function
n(L,~) = NoL-OI.~-/3

and that L and ~ are independent (see §3.3). With f3 = 0 (no SB depen-
dence) this is a good approximation to the observed LF, i.e. the Schechter
function with no bright-end cutoff.
We can now ask the question: which galaxies contribute most light? Let
TL equal the total light from all galaxies so that

TL=No 1 ll:ma.,
LTna .,

Lrnin Emin
L-OI.+1~-/3 dL d~ , (1)

No L -0I.+2)(~-/3+1 ~-/3+1)
(-0 + 2)(-f3 + 1) (L ma:!:
-01.+2
- min ma:!: - min •
(2)

1 For the rest of this paper I will use", to represent SB in Blue magnitudes arcsec- 2
and :E to represent blue solar luminosities pc -2.
 107

If we take Lmax ~ Lmi n and ~max ~ ~min then

Case a > 2 - Light is dominated by low luminosity galaxies.

Case a < 2 - Light is dominated by high luminosity galaxies.
Case 13 > 1 - Light is dominated by LSB galaxies.
Case 13 < 1 - Light is dominated by HSB galaxies.

What about the cross sectional area (A) these galaxies subtend on the
sky? Since ~ = L / A, the total area subtended is

(3)

No (L-o:+2 L-9t+2)(~-.13 ~-.13) (4)

(-a + 2)( _13) max - min max - min ,

and as above,

Case a > 2 - Area is dominated by low luminosity galaxies.

Case a < 2 - Area is dominated by high luminosity galaxies.
Case 13 > 0 - Area is dominated by LSB galaxies.
Case 13 < 0 - Area is dominated by HSB galaxies.

The above discussion is summarised in Figure 1.

The above formulation clearly highlights the importance of knowing the
values of a and 13 if we want to answer questions 1 and 2 above. Consideration
of the LF alone does not supply the answer. What are the observed values of
a and j3? General observations of field galaxies (such as those in the RC2 or
the RSA) would put a:::::; 1.2 (see for example Felton 1985): Because the SB
distribution is observed to be sharply peaked (Freeman 1970;"Fish 1964), i.e.
approximately constant surface brightness, we can take 13 : : :; O. These values
imply that LSB galaxies only contribute a small fraetion of the total light
(mass?) and an approximately equivalent contribution to the cross sectional
area.
In the following sections we shall review what is known about Q (the
distribution of galaxy luminosities) and 13 (the distribution of galaxy surface
brightnesses) . In particular, we shall analyse the selection effects that may
influence our measurements of a and 13 and discuss how recent observations
affect our knowledge of hidden galaxy populations.

2. Which is the best measure of surface brightness?

It is pertinent at this point to briefly consider what we mean by SB and how
we measure it. Because galaxies have a brightness profile that varies with
108

B A
--------1 - - - - - - - - • - -

y '
_1 - _ _ _ _ _

-2 -1 2I
I

-1
c D
-2

Fig. 1. Possible Universes defined by the luminosity parameter a and the surface
brightness parameter {3. Region A - Universe dominated by LSB dwarf galaxies.
Region B - Universe dominated by LSB giant galaxies. Region C - Universe domi-
nated by HSB giant galaxies. Region D - Universe dominated by HSB dwarf galax-
ies. Dashed lines partition the plane with regard to total luminosity and total cross
sectional area on the sky (see text). Point ;e indicates the inferred values of a and
f3 derived from a "typical" galaxy catalogue. Point y is the value inferred after
accounting for surface brightness selection effects.

radius, there is no unique value of SB. Obtaining a consistent and meaning-

ful measure of SB is not easy. There are generally four measures of SB in
common use:

1. The central SB (JL:r) - a value that is not directly measurable but

instead is derived from an inward extrapolation of some intensity law.
The profile is fitted over what is hoped to be a high signal-to-noise region
away from the centre where seeing may smear out a bright nucleus.
Relating JL:r to a physical characteristic of the galaxy, i.e. the central
surface mass density, is problemmatic. The intensity law often varies
within the inner regions (nucleus, bulge) leading one to question whether
the disc actually exists as a separate entity within the central regions. In
 109

addition, fitting an exponential to a bulge-dominated galaxy may lead

to an approximately constant value for J-Lx irrespective of the central
mass density (Kormendy 1977; Phillipps & Disney 1983; Davies 1990) .
It is also difficult to compare galaxies with different types of light profile
(exponential, r 1 /4, or a combination of both).
2. The Olean SB within the effective radius (J-Le) - the radius that
contains half the light. Although J-Le is simply derived from J-Lx for pure
exponential or r 1 / 4 law galaxies, it does have the advantage of being a
mean over a large fraction of the galaxy and so reflects a gross property
of all galaxies irrespective oftheir profile shape (morphology). It has the
disadvantage that the total magnitude has to be measured independently
of a profile fit, i.e., aperture photometry. To do this, one must have an
accurate measure of the point at which the galaxy comes to an end
(accurate sky determination). For a pure exponential profile, J-Le is easily
derived as follows: (~)e = LT/27rR~, or J-Le = mT + 510g(Re) + 2.0 in
logarithmic units, where mT = J-Lx - 510g( ro) - 2.510g( 211'") where ro is
the exponential scale length. So J-Le = J-Lx+510g(Re/ro )-2.510g(211'" )+2.0.
For an exponential disc Re/ro ~ 1.7 so J-Le ~ J-Lx + 1.2.
3. The surface brightness at the effective radius (J-L!SO) - this is
the surface brightness at the isophote that contains half the total light;
J-L!SO ~ J-Lx + 1.8 for an exponential disc.
4. The Olean SB within SOOle other isophote (J-Lm) - which isophote
to use is rather arbitrary. LSB galaxies may be undefinable if the selected
isophote is too bright.
As I compare data from different sources and analyse different problems I
shall interchange between different measures of SB. Thus, the above relations
and definitions should be kept in mind.

3. ProbleOls associated with Oleasuring 0: and f3

Luminosity obviously acts as a strong discriminator when selecting galax-

ies . For example, in a flux-limited sample, two-thirds of the galaxies will lie
within 0.75 magnitudes of the flux limit; this can lead to strong biases in
the data, such as Malmquist bias. If the faint end of the LF is not steep,
then galaxy samples are expected to be dominated by galaxies of a char-
acteristic luminosity L *. Basically, this is because the number density of
high-luminosity galaxies drops off very quickly with increasing luminosity.
At the faint end of the LF, the volume sampled as a function of luminos-
ity (<x L 3 / 2 ) increases more quickly than the observed LF faint-end slope
(0: ~ 1.2?). The L* galaxies force their way into galaxy catalogues! Ideally
we would like to construct large volume-limited samples to some well-defined
limiting magnitude or size for galaxies of "all" surface brightnesses. Many
of the problems associated with constructing such samples are more easily
110

dealt with by looking at a nearby cluster where there is a high density of

galaxies at the same djstance, though there may be local environmental ef-
fects. Of most int~rest to us are the intrinsic values of a and {3, Le., those
values laid down hy the formation process or since changed by evolution in
a particular environment.
Let us now consider the SB selection effects in some detail. Samples of
galaxies in the general field suffer from selection effects in a very different
way to those that affect galaxies in one particular cluster. I will discuss
cluster and field samples separately.

3.1. THE MEASUREMENT OF {3 IN FIELD GALAXY SAMPLES

The strong SB selection effects operating on samples of galaxies have been
described in detail in Disney (1976), Disney & Phillipps (1983) and Davies
(1990). In essence, these can be described as a strong preference for the
selection of galaxies with a preferred surface brightness, J.L*, defined by the
sample selection criteria: mL the limiting magnitude; (h the limiting isopho-
tal size; and J.LL the surface brightness at (h. In essence, the fraction of a
galaxy's luminosity that is measured above some isophote and the size at
that isophote depends on its surface brightness. For a sample of field galax-
ies, the volume of space sampled varies not only as a function of luminosity,
but also as a function of SB. For clusters within this volume, the portion
of the SB absolute magnitude plane that is observable varies with distance
(Davies 1990). Figure 2 shows how this volume is related to the area of
the surface brightness plane that can be observed at different distances. At
constant SB the volume sampled is proportional to 10- 0 .6M ; at constant
luminosity the volume sampled is a function of surface brightness, VT(J.La:) -
this is the "visibility" curve which I will discuss in detail below.
Disney & Phillipps (1983) define the visibility of a galaxy as the maxi-
mum volume in which a galaxy can exist and still obey the selection criteria
defined above. The visibility is proportional to the likelihood of a galaxy
of luminosity L and radius R appearing in a catalogue with given selection
limits. At a given total luminosity, the luminosity within a given isophote
and the size at that isophote (Rapp) are a sensitive function of the galaxy's
SB. The visibility curve (Figure 2) in pc 3 is defined as the minimum of the
two curves;
v1 -- [L app /L Tot J3/2100.6(mL-M+5) (5)

(6)

(see also Davies 1990 for a discussion of the influence of a galaxy's inclination
to the line of sight and its morphological type). Here, VI defines the volume
sampled, as a function of SB, by a galaxy of absolute magnitude M selected
 III

Fig. 2. The volume sampled at each point in the bivariate brightness plane (M
vs 11.,). At constant absolute magnitude, the volume sampled depends on surface
brightness and is defined by the visibility curve VT (11.,). At constant surface bright-
ness the volume sampled is VT(M) ex: lO - o.6M. The hatched regions show the areas
of the bivariate brightness plane observable for clusters at two different distances.

by isophotal magnitude; V2 defines the volume sampled, as a function of SB,

by a galaxy of absolute magnitude M selected by its apparent isophotal size.
The functional forms of Lapp/ LTot and Rapp are;

(Lapp/ LTot)D = (1 + SD + O.5sb) exp( -SD) - (1 + PD) exp( -PD)

R = (J.LL- J.Lx)10[(M-tt",+23.6]/-5)
app 1.08

for the exponential disc, where SD = r,,/1'OD, PD = R app /1'OD, 1'" is the
radius at which saturation (mainly applicable to photographic data) occurs,
and rOD is the exponential scale length.
The visibility curve, VT(J.Lx), essentially defines the range of SBs one may
expect to observe in a sample of field galaxies. More importantly, if the
intrinsic SB distribution of galaxies is fiat and the range of luminosities is
112

the same at each SB, then the observed distribution would have the same
shape as the visibility curve. For an exponential profile, the Vl curve peaks
at about 2.510g(N) mag arcsec- 2 below P,L, where N is the dynamic range
of the detector; the V2 curve peaks at about 2.2 mag arcsec- 2 below P,L.
The peaks of the relevant curve, or the point where the curves cross, define
p,*, the sample's characteristic SB (see Davies 1990). For other intrinsic SB
distributions the observed distribution is that of the intrinsic multiplied by
the visibility (Davies et al. 1992). The observed distribution of numbers
against surface brightness is narrow (Freeman 1970; Fish 1964) and similar
to the predicted visibility curve. But is the observed narrow distribution of
surface brightness a selection effect or is it real?

3.2. MEASURING THE ACTUAL VISIBILITY OF FIELD GALAXIES

To try to answer the above question I have taken a new look at data from
the ESO-LV catalogue (Lauberts & Valentijn 1989). The galaxies in the
catalogue were selected to meet a well-defined size criterion (a diameter
larger than 1 arcmin at the 26 B JL isophote). I have selected all those spiral
galaxies (type 4 ~ T ~ 8) that have redshifts (918 galaxies). This large data
set is ideal for the task because each galaxy in the sample has a measure of
SB (p,!SO) and a total magnitude derived from a profile fit. Restricting the
sample to late-type spirals reduces the influence of a central bulge component
on the derived SB values (see references in §2). It also focuses attention
on the numbers of relatively bright (massive?) spiral galaxies that may be
missing from catalogues. Almost all of what we know about the spiral galaxy
LF has been derived either from galaxies in the above sample or from samples
selected in an almost identical way. As the sample is selected by size, we
might expect visibility (as described in §3.1) to act in such a way as to
severely limit the detection of both HSB and LSB galaxies.
U sing the redshift information (approximate distances) we can measure
the actual volume sampled as a function of SB and compare this with the
observed galaxy numbers (Figure 3). It is clear that the relative volume
sampled (solid line) is indeed a sensitive function of SB. Galaxies of both high
and low SB are sampled over much smaller volumes than those of normal
(in the peak of the distribution) surface brightness. This implies that the
true SB distribution, when measured in magnitudes, must be fairly broad
and flat, or equivalently a b- l distribution in numbers, i.e., f3 = 1. High and
low SB galaxies are only detected over a small volume (our local Universe)
and so they are underrepresented in catalogues, not because they are rare
but because they are highly selected against. The necessary corrections to
account for the different volumes sampled are obviously very SB dependent.
Applying the visibility correction would produce a much flatter distribution
over six magnitudes of SB, increasing the number of galaxies by about a
 113

ESO data

"
I'
o
ul -
I
,
,"
~

,,
I

, I

\
,
()
0 - --
,,
C',

/
,
,
, I

\
,
I
,
.... ,
\
\

n
,)

",
--\
,
'-
...-::;::--
,.-
l )
q)

"

2 1.
tl L f------L------+-----'--f-----L---------t==J
22 . 23. 24 . 25.

Fig. 3. The histogram shows the distribution of effective surface brightness (the
surface brightness measured at the half total light radius) for all spiral galaxies with
redshifts in the ESO catalogue of type T = 4 - 8. The solid line is proportional
to the median volume sampled by galaxies in each surface brightness interval in
the histogram. The curve has been normalised at its peak to the peak height of the
histogram. The error bars are obtained from the measured median value of randomly
selected sllbsets of the full data set. The dashed line is proportional to the volumes
expected, considering the mean absolute magnitudes of galaxies in each surface
brightness interval and normalised to the peak histogram value. The short/long
dash line shows the volumes predicted by the vcisibility calculation described in
the text.

factor of three. This ignores any galaxies whose visibilities are so low that
they are not detected at all.
One could, of course, ask whether the volumes sampled are, in fact, due
to a change in average luminosity across the SB distribution. In, essence is
there a relationship between luminosity and SB? The dashed curve in Figure
3 shows the predicted relative volumes sampled, calculated from the mean
absolute total magnitudes of galaxies in each SB interval. It is clear that
absolute magnitude is not strongly correlated with SB for late type spiral
galaxies (Disney & Phillipps 1985) and that there must be another factor
114

that strongly restricts the detectability of both high and low SB galaxies.
How well do the actual volumes sampled at each surface brightness cor-
respond to those predicted by visibility? At first sight, not very well. The
predicted visibility curve (selection by size) for the given selection criteria
peaks at J.L!so = 25.5, not at the observed value of J.L!so ::::: 22.9 (or J.L3: ::::: 21.1
which supports previous observational evidence - Freeman 1976). But is this
sample limited by size at all SBs? Might there be a hidden magnitude limit?
To test this, I have calculated an "effective limiting magnitude", m'L, for this
diameter-limited sample (the details of how I have done this are described
below). I find that m'L = 15.4 (a photographic dynamic range of 100 is as-
sumed). The VI and V2 visibility curves then cross to giv~ a predicted peak
in the surface brightness distribution at J.L!so ::::: 23.4 (see Figure 3). The
predicted theoretical visibility curve, VT(J.L3:), is then in excellent agreement
with the actual volumes measured.
What are the real selection criteria for this sample? Is it size, magnitude,
or a combination of both? Samples can be selected by size only if the V 2 curve
always falls below the VI curve, and vice versa for selection by magnitude
only. If the curves cross, then the minimum of the two at any particular
SB should be used. It is clear that the visibility curves do cross and that
this sample is limited on the LSB side not by size but by magnitude. This
illustrates a most important and oft neglected point: selection is always
by a combination of size and magnitude. For example, if you select by
diameter then there will be a magnitude at which a galaxy is of such LSB
that it can no longer be detected in the survey, not because it is too small
but because its signal to noise is so low (see §3.4). There must inevitably be
a hidden magnitude constraint.
I will now explain how I have calculated an effective magnitude limit for
a nominally size-limited sample. The method is a simple derivative of the
(VjVma 3:) test. For a magnitude-limited sample we can calculate (VjVma 3:)
using

(7)

Let us now define the mL value that gives (VjVma 3:) = 0.5 in this equation,

and call this the "maximum effective mL for the sample". Why a maximum?
Finding the mean value of lOO.6m (in equation 7) for the sample galaxies gives
a number Vave proportional to the average volume sampled. Using the same
assumptions implicit in the VjVmaz test, i.e., that the galaxies should on
 liS

average sample half the volume available, I set 100 . 6m l = 2Vave ' Thus, mi, is
the maximum limiting magnitude consistent with the observed magnitudes
of the galaxies in the sample and corresponds to twice the average volume
sampled by the galaxies. Whatever the "real" value of mL, it is no fainter
than this derived value. Therefore, mi, can be used to form an upper limit
to the Vl curve.
The derived mi, limit can be used to estimate an upper magnitude limit
to any sample allegedly selected by diameter, and then it can be used retro-
spectively on galaxy samples. The mi, limit is a simple way of obtaining an
upper limit to the true mL and the method can also be used on magnitude-
selected samples to derive Oi, an estimate of the size limit for a nominally
magnitude-complete sample.

3.3. THE MEASUREMENT OF f3 IN CLUSTER GALAXY SAMPLES

For cluster galaxies, apparent magnitudes relate directly to absolute mag-

nitudes. Thus, a unique SB/magnitude plane can be defined for the sample
without the worry of obtaining redshifts, though of course you have to faith-
fully believe that you can separate cluster from background galaxies on the
basis of appearance. As for the field sample I will first quantify the surface
brightness selection effects, since I wish to define that portion of the SB /mag
plane that is observable.
For selection by apparent magnitude (see Davies 1990) the selection
boundary is defined by,

m = mL + 2.5Iog[(1 + SD + 0.5sb)e-SD - (1 + PD )e- PD ]

for an exponential disc galaxy. Similarly, for selection by apparent size, the
selection boundary is defined by,

m = f.lx - 5Iog[OL/(1.09PD)] - 2.0 .

Definitions are as given in §3.1.

The intrinsic SB distribution (parameter (3) is dependent on whether the
selection boundaries allow you to sample all portions of the SB magnitude
plane that galaxies occupy. In addition, the data should also be well away
from any selection boundary before the existence of any SB/magnitude re-
lations is inferred.
For example Sandage & Perelmuter (1990) (henceforth SP) have con-
structed, from various sources, a plot of surface brightness against magnitude
for elliptical galaxies with a magnitude range of -12 > MBT > - 24. The
data show a distinct correlation in the sense of decreasing surface brightness
with decreasing luminosity for galaxies fainter than MBT ~ -19 and a trend
for galaxies brighter than this to decrease in surface brightness as they be-
come more luminous. The data used by SP comes from five separate sources,
116

each selected in a different way. Here I shall consider, as an example, the

data on Virgo cluster dwarf galaxies.
The Virgo data that SP use comes from a number of different sources,
though the majbrity of galaxies are drawn from the sample of Ichikawa et
al. (1986) . Ichikawa et a1. describe their selection criteria as (h > 5 arcsec
for galaxies with SB 27 < 26 where SB 27 is the mean SB within the 27 BJ-L
isophote. In intensity units (and ignoring saturation) the selectio,n boundary
(we assume that these LSB galaxies all have exponential profiles) is defined
as;

I 27 -_ LT[1 - (1 + P27) exp( -P27 )]

Ll 2 ,
7rUL

where P27 ;:::: (27 - J-L:e) . In logarithmic units, this can be written as

26 =m - 2.5log[1- (1 + P27)exp( -P27)] + 5log((}L) + 1.2,

where, for (}L = 5,
m = 19.8 + 2.5log[1 - (1 + P27) exp( -P27)] .
This is equivalent to a limit on apparent magnitude with mL = 19.8. The
selection boundary is drawn on Figure 4 (right hand curve) along with the
data. In addition Ichikawa et a1. (1986) state that they limit selection to
galaxies smaller than 100 arcsec (left hand curve) and that they were search-
ing for LSB dE galaxies of smooth appearance. Observational selection stops
galaxies from being detected in the two regions labeled D. Region B is ex-
cluded because only LSB galaxies were selected. I suggest that galaxies in
region A would also be excluded because they would resemble background
galaxies (higher SBs with lower luminosities). The galaxies in the sample
essentially fill those parts of the SB magnitude plane accessible to observa-
tion.
Both ourselves Phillipps et a1. (1988) and Impey et a1. (1988) have pre-
viously questioned the validity of the surface brightness magnitude relation
for low luminosity galaxies because the data lie very close to the selection
boundaries. From Figure 4 it is clear that this data set is almost totally
bound by the selection criteria. Plotting data like this without including
the selection boundaries can be misleading. To be fair to Sandage & Perel-
muter, some of the other data sets they use are not so tightly constrained by
observational selection as the one described above (see Davies et a1. 1992).

3.4. SIGNAL-TO-NOISE RATIO

It is also worth pointing out that in addition to the above observational

selection boundaries there are regions of the apparent magnitude SB plane
 117

Virg o data
0 . - - . - r . - - .__. - - '__~-.__~-'__' - - '__' - - ' - - '

-
IX) B

0
N
..
0

<V
ill
Vl N
N

t~ ,.!

..
C • ":' • : •

<.D
N-

- 24 -22 -2 0 - 18 -16 . - 14 -12

M

Fig. 4. The Virgo data along with the observational selection boundaries. A - ex-
cluded by morphological selection; B - excluded by surface brightness; C - the only
region where galaxies could be selected; D - excluded by observational selection.

that are inaccessible to observation because the signal-to-noise ratio (5) of

the objects wQuld be very low. The signal, in the case of an exponential disc,
is

LT[(1 + SD + 0. 5s b) exp (-SD) - (1 + PD) exp{ -PD)] ,

where LT is the total apparent luminosity of the object. If the noise per
pixel (O'n) is assumed to be constant from pixel to pixel then the total noise
in the image (O'T) is given by,

where a is the pixel size in arcsec and rL is the radius of the galaxy, in
arc sec, at the limiting isophote. Therefore,

s _ _L~T_a.::..:.[(_1_+_S_D_+_0_.5_s_D--.:.2)_e--:-x=.::p(C--_S_D..:...)_-_{'-1_+-'p=-D--.:.)_e_x=-p(.:....-..::.P-=D~)]
- 7r 1/ 2 rLO'n
118

After a little algebra, this leads to

m = 2JL6ky - JL:1! - Slog(1f'1/2Sa -lk) + 2.0+

(1 + SD + O.SSD2) - (1 + PD) exp( -PD)]

S I og [ ,
, PD

where JL6ky is the sky brightness in mag arcsec- 2 and k = U n /I6ky is the
noise per pixel as a fraction of the sky intensity (laky).
It is, of course, very important to ensure that the signal-to-noise con-
straint is less stringent than any chosen observational limits placed on the
data. For example the "hidden" magnitude limit discussed in §3.2 with re-
gard to the ESO - LV data may well be due to a signal-to-noise constraint
(see Davies et al. 1992 for more details).

4. Testing the LSB galaxy hypothesis

In §3, I have outlined our reasons for believing that observational selection
severely limits our knowledge of the true number of galaxies in the Universe.
Selection effects at other than optical wavelengths may be quite different
from those described above. Because one can hope that galaxies that are
elusive in the optical may nevertheless turn up elsewhere, it is necessary
to explore the constraints set on a large population of dark galaxies by
observations made at all wavelengths. We must remember, though, that
any search technique at any wavelength will have two quite distinct limits
to its sensitivity. One refers to total flux, the other to surface intensity or
surface brightness . In this section, I will review some previous observations
that have brought hidden galaxy populations to light and will suggest some
future observations that may reveal hidden galaxies if they exist.

4.1. OPTICAL SEARCHES FOR LSB GALAXIES

To search for LSB galaxies in the optical you ideally need;
1. a large telescope
2. a high-efficiency detector
3. a dark sky
4 . long integrations
5. and most of all a lot of telescope time
The game is all about collecting as many photons from the object and as
few from the sky as you can.
Most of what we know about LSB galaxies comes from surveys of local
clusters. As LSB galaxies can only be seen to small distances, these nearby
clusters were initially thought to be the most profitable areas to search.
Although many LSB galaxies were known prior to the survey of the Virgo
 119

cluster by Binggeli et al. (1984) - see, for example, Reaves (1983). Of of,
course the local LSB dwarf spheroidals have been studied for many years.
These studies showed just how important LSB galaxies were in terms of
numbers. About 80% of the catalogued galaxies were identified as dwarfs
and a large fraction of these were are of LSB (faintest with J1.e ~ 26BJ1.).
Impeyet al. (1988) followed up this work by going to even lower SBs using
Dave. Malin's unique photographic talents. They detect some 140 galaxies
with J1.e down to about 27 B J1.. •
We have also searched" nearby clusters for hidden LSB galaxies. In the
Fornax cluster we have identified some 300 galaxies (Davies et al. 1988; Irwin
et al. 1990) using photographic material (see also Caldwell & Bothun 1987;
Ferguson 1990). The galaxies identified so far have SBs down to J1.e ~ 27
BJ1.. We have carried out follow-up multi-colour CCD photometry of a sub-
sample of these galaxies (Davies et al. 1990) and find that they have a
wide range of intrinsic colours. We have also carried out a rather limited
CCD survey (limited because of the small sky area coverage of typical CCD
cameras) within the cluster A1367 (Davies et al. 1989) and a more extensive
study of A3574 using a larger-area, focal reducing CCD camera (Turner et
al. 1992). The surveys extend down to an equivalent of J1.e ~ 28.5BJ1.. Both
surveys again indicate large numbers of LSB galaxies within clusters (their
spatial distribution on the sky indicates that a substantial fraction of these
galaxies lie in the cluster). If they are cluster members then they are of low
lumin"osity. They will need (M/L)0 ~ 100 for them to contribute an equal
amount of mass to the cluster as the more visible normal SB galaxies (see
Phillipps 1991). After correcting for the area of the SB/magnitude plane
not sampled we find a steep faint end slope to the LF (a ~ 1.5) and a SB
distribution with f3 ~ 1.0 (broad and flat).
Are all LSB galaxies in clusters? Rather limited searches have been made
for LSB galaxies in the field, and the conclusion has generally been that
the LSB galaxies follow the distribution of the brighter galaxies (Bothun et
al. 1986, Thaun et al. 1991). If we want to detect large LSB galaxies then
clusters are probably the wrong place to look as it is unlikely that they
would be able to survive in a dense cluster environment. Malin 1 (Bothun
et al. 1987; Thuan et al. 1991) is "living proof" that large LSB galaxies do
exist in the field. We really do need to survey the sky to faint isophotal
limits over large areas if we are going to place constraints on the numbers
of similar objects. Schombert et al. (1992) have already started this most
important work.
Ideally, what we need are deep large area surveys of the general field,
though the huge amount of time needed on a large telescope to carry out
such a survey would be very difficult to obtain. Large telescopes are generally
not designed for survey work, yet there is a way that they can be adapted to
carry out such a survey. We have designed and are now operating a parallel
120

CCD camera, called Hitchhiker, on the 4.2m William Herschel Telescope.

By parallel, we mean that the camera images an off-axis field of view while
the scheduled observer carries out on-axis spectroscopy. In this way we are
able to collect a large number of deep CCD images of the general field. In
addition, the camera has a dichroic beam splitter so that we in fact image
the same field in two colours simultaneously. In this way we hope to create
a large, deep CCD survey of the sky to search for a population of large LSB
galaxies that may exist beyond and between galaxy clusters. The results of
this survey will be reported in future papers.

4.2. SEARCHES AT 21 CM

It has been argued that repeated failures to find 21 cm extragalactic clouds

in random pieces of sky set strong constraints on the number of dark or
very low surface brightness galaxies, or of intergalactic clouds, present in the
Universe, i.e. Fisher & Tully (1975). But we have previously found (Boyce
et al. 1992) that instrumental effects and the short integrations previously
used set a lower limit, presently around NHI = 1019 cm- 2 , to the column
densities detectable, irrespective of either the telescope diameter or the total
mass (MHI) of hydrogen involved. We argue that galaxies with 'NHI in
excess of 10 19 cm- 2 , and with normal [MHI/ LBJ ratios, must have surface
brightnesses high enough to render them easily detectable in the optical.
Therefore the existing 21 cm emission observations set few constraints on a
large population of lower surface brightness galaxies or hydrogen clouds.
So long as 21 cm observers, for good economic or technical reasons (e.g.,
transit telescope design), confine integrations per beam to an hour or less,
then system noise and baseline problems will limit them, irrespective of tele-
scope size, to detecting high hydrogen column densities (greater than about
1019 cm- 2 ), not those of a LSB galaxy with normal (MHI/LB)0 ratios.
Briggs (1990) has recently been even more trenchant. He says "The short
integrations that typify 21 cm line redshift surveys would not reliably detect
column densities much less than 1020 cm- 2 even if the emission filled the
beam ... " . It follows that there may be considerable amounts of intergalac-
tic hydrogen present, provided it lies at low column densities. The 21 cm
emission observations do not set strong constraints on its contribution to
the cosmic density.
It is possible, by carrying out very long integration per beam (~ 12
hours), to reach down a further factor of 10 in sensitivity to about 10 18
cm- 2 , where dim galaxies of low surface brightness would be picked up.
This is an observational programme we are about to start. Many of these
galaxies would be too dim to see on modern Schmidt material let alone
on Palomar Prints. It would therefore seem worthwhile to carry out such
a deep survey to search for intergalactic hydrogen clouds and for dim and
 121

dark galaxies.

4.3. QSO ABSORPTION-LINE STUDIES

QSO absorption lines (QSOALs) probe column densities of gas, and corre-
sponding' surface brightnesses of the accompanying stellar population down
to levels unreachable by other means. Observations of QSOALs, in partic-
ular the metal-line systems seen at relatively low redshifts (e.g., Sargent et
aL 1979) which extend down to column densities around 10 17 to 10 18 cm- 2 ,
could therefore place strong constraints on populations of hidden galaxies,
and may even rule out their existence altogether. Indeed, the recent inter-
esting claim by Bergeron (1988) and Bergeron & Boisse (1991) to identify
luminous giant galaxies fairly close to the line of sight as the absorbers in
three quarters of well studied MgII systems at redshifts c::= 0.4, though requir-
ing galaxy halos", 100 kpc across (for Ho = 50 km s-l Mpc- 1 ), apparently
leaves little room for a substantial population of LSB galaxies (or, at least,
gas-rich ones).
In a previous paper (Phillipps et aL 1992) we have reviewed some well-
known difficulties with the Large Galaxy Halos (LGH) hypothesis for the
origin of QSOALs. For instance, why do searches for bright galaxies near
QSOs with metal-line systems usually succeed while searches for metal-line
systems in QSOs near bright galaxies usually fail? We also discussed the
apparent problems with taking LSB galaxies to be a major cause of metal-
line systems. We showed first that the identifications claimed by Bergeron
are no more than one expects by chance on the LSB galaxy conjecture; if
the true absorbers are actually invisible LSB galaxies, it is straightforward
to calculate the probability that a luminous normal (i.e., "normal" surface
brightness) galaxy of the appropriate redshift will also appear close to the
line of sight to the QSO. By including the known clustering of galaxies, we
find that the majority of our hypothetical LSB galaxies would indeed have
giant neighbours close enough to them to be identified as the (presumed)
absorber, and therefore conclude that many of the claimed identifications
could be merely neighbours of the true absorber.
We went on to consider the other main arguments which have been raised
against LSB galaxies as the source of QSOALs: (1) that in extreme cases
there are too many separate systems in front of a single QSO to be explained
by galaxies of any kind (Pettini et al. 1983; York et al. 1986; York 1988);
(2) that the velocity structure observed within Mg II systems is too wide
and complex to be explained by absorption in a galaxy, particularly a dwarf
or LSB galaxy (York et al. 1986); and (3) that the line ratios seen are
atypical of either our Galactic disc or halo (York et al. 1986). When the latest
observations of LSBGs are taken into account, we find that none of these
arguments prove to be particularly persuasive, and that LSB galaxies do
122

present a preferable alternative to huge halos around large galaxies. Indeed,

the arguments against the LGH hypothesis may well be the more substantial.
Continued observations of QSOs, particularly behind nearby clusters (Davies
et al. 1988) may help resolve this matter. Recent observations by HST of
Lyo: clouds at low z (Bahcall et al. 1991) were a surprise to many. Will more
complex absorption-line systems, typical of galaxies, reveal themselves at
low z when more sight lines have been observed?

4.4. IMPLICATIONS FROM DEEP NUMBER COUNTS

Recent deep number counts have shown that there are many more galaxies
per unit volume in the distant Universe than we observe locally (Cowie et
al. 1991). Could these galaxies now be part of the local "hidden galaxy"
population? Deep studies of nearby clusters (§4.1 above) indicate a much
more steeply rising LF at the faint end than is observed in the field. The
general field has hardly been studied at all to the faint isophotallimits of the
nearby clusters. If the field population had a LF faint-end slope similar to
that found for clusters, then the number counts at faint isophotal detection
limits would quickly be dominated by rather nearby lower luminosity galax-
ies and their number density would be observed to be significantly higher
than the expectation for L· galaxies. SB selection effects can -also greatly
affect what we might expect to observe in the distant Universe (Phillipps
et al. 1990). With both the (1 + z)4 and K-correction effects dimming a
galaxy's SB, it is clear that only intrinsically high SB galaxies will find their
way into the selection window described by visibility. Deep number counts
may be revealing part of the hidden galaxy population we have conjectured
to exist in our local Universe. Again, cameras like Hitchhiker may help us
understand these problems.

4.5. OBSERVATIONS AT OTHER WAVELENGTHS

Is there an optimum wavelength to detect LSB galaxies? It seems, at least

from the IRAS data, that LSB dwarf galaxies are not great emitters in the
infrared. Although they have low total outputs, it is not clear that their
contrast against the background is any better or worse than that in the
optical. This is an area that needs to be explored, but it will probably have
to wait until a large-aperture IR space observatory is in operation.
In the ultraviolet (UV) the sky is very dark. O'Connell (1987) has al-
ready suggested that this may be the ideal wavelength at which to detect
LSB objects. Until the advent of a large UV observatory with high efficiency
detectors, this conjecture is difficult to test. UV observations with HST are
of course possible, but with a small telescope, a low efficiency in the UV,
and a small field of view, a survey is impossible. Observations of specific
 123

targets may be a more productive approach with HST. For example, obser-
vations of known LSB galaxies to see if they really do stand out against the
UV background and possibly observing quasars to identify absorption-line
galaxies. For example, observing quasars beyond their Lyman limit, where
the flux is dropping off, and the absorbers around Lya, where there may be
line emission, may enhance the chances of detection.

4.6. THE EXTRAGALACTIC BACKGROUND . LIGHT (EBL) IN THE OPTICAL

A direct measurement of the EBL is extremely difficult (see Mattila 1990

for a review). The signal to be detected is constant across the detector
and probably is at least 100 times fainter than the local background from
our solar system and Galaxy. Nevertheless, attempts have been made to
make a direct measurement. Methods that attempt to subtract off the large
local effects (Dube et al. 1977, 1979; Toller 1983) must, in my opinion,
suffer from large systematic effects that are very difficult to quantify. An
ingenious method first suggested and used by Mattila is to make a differential
measurement between a "dark" high galactic latitude cloud and a nearby
"clear" region. His measurements (see Mattila 1990) using this method have
been consistently higher than those of other workers in this field. His latest
value for the EBL (Mattila 1990) is about 7 times larger than that predicted
from galaxies in a non-evolving Universe (Sandage & Tammann 1965). The
problem, of course, is distinguishing between galaxy evolution and local
hidden galaxies as the source of the excess. I really feel that Mattila's dark
cloud method is potentially the best available at the moment to make a
decisive measurement. A more prolonged observational program to reduce
the inherent signal-to-noise problems and to assess short-period variability
in the night sky would give a much better understanding of any systematic
effects.

4.7. THE BARYONIC MASS DENSITY OF THE UNIVERSE

There are many missing-mass problems, some of which seem inescapably to

require exotic non-baryonic dark matter. One exception is the inference from
big bang nucleosynthesis that the baryonic mass content of the Universe is
some 5 - 10 times higher than that observed to be associated with known
galaxies (Pagel et al. 1992; Persic & Salucci 1992). Where can this missing
mass be? It appears that LSB galaxies can have very high mass-to-light
ratios (see for example Carignan & Freeman 1988), so considerable amounts
of baryonic matter could be tied up in hidden galaxies.
124

5. Conclusion
I have reviewed some calculations of the visibility of galaxies and have shown
using new data how strong this SB selection effect can be. The demonstration
in §3.2 that the measured volumes as a function of SB compare almost ex-
actly with those predicted by visibility should make astronomers concerned
as to whether we really do know the extent of the local galaxy population. It
strongly supports the hypothesis that there is a large population of hidden
galaxies. It is also important to understand the observational selection bi-
ases before making inferences about SB/magnitude correlations. I have also
reviewed some of the previous arguments that have been said to place low
limits on the LSB population, and I have indicated ways in which we can
probe the Universe more efficiently to lower SBs.
Has this analysis helped answer the two questions posed in §1? The impli-
cation from Figure 3 is that the true SB distribution amongst spiral galaxies
is flat and wide (<x ~-1). Therefore, f3 = 1 and the SB integral is impor-
tant, although any slope in the distribution would still be disguised by the
strong selection effects. Can we infer a similar result for ellipticals? Com-
bining galaxies of all types and searching to low surface brightnesses we
might expect, from our observations of nearby clusters, that a ~ 1.5. We
are then left with a most intriguing and still rather tentative result. For at
the moment my best guess is that the Universe is illuminated equally by
high-luminosity galaxies of both high and low SB, but that the filling factor
on the sky is weighted towards high luminosity lower SB galaxies (see Figure
1).

Acknowledgements
None of the work described in this paper would have been possible without
the help and advice of Mike Disney and Steve Phillipps.

References
Bahcall, J .N ., Jannuzi, B., Schneider, D., Hartig, G ., Bohlin, R ., & Junkkarinen, V. 1991,
ApJ, 377, L5
Bergeron, J . 1988, IAU Symposium No . 130, Evolution of Large Scale Structure in the
Universe, eds J . Audouze and A . Szalay (Reidel: Dordrecht), p .343
Bergeron, J. & Boisse P. 1991, A&A, 243, 344.
Binggeli, B., Sandage, A., & Tammann, G.A. 1985, AJ, 1681
Briggs, F. 1990, AJ , 100,999
Bothun, G ., Impey, C., & Malin, D. 1987, AJ, 94, 23
Bothun, G ., Beers, T., Mould, J., & Huchra, J . 1985, AJ, 90, 2487
Boyce, P., Davies, J ., Disney, M., Cohen, J., & Wright, A . 1992, MNRAS, submitted
Caldwell, N . & Bothun, G. 1987, AJ, 94, 1126
Carignan, C . & Freeman, K . 1988, ApJ, 332, 33
Cowie, L., Songailia, A. , & Hu, E . 1991, Nature, 354, 460
Davies, J .I. 1990, MNRAS, 244 , 8
 125

Davies, J .1., Disney M ;J., & Phillipps, S. 1988, A&A, 192, 90

Davies, J .I., Phillipps, S., Cawson, M., Disney M .J., & Kibblewhite, E. 1988, MNRAS,
232,239
Davies, J.I., Phillipps, Disney M.J ., Boyce, P., &; Evans, Rh. 1992, MNRAS, submitted
Disney, M. 1976, Nature, 263, 573
Disney, M.J. &; Phillipps S. 1983, MNRAS, 205, 1253
Disney, M.J. &; Phillipps S. 1985, MNRAS, 216, 53.
Dube, R., Wickes, W., &; Wilkinson, D. 1977, ApJ, 215, L51
Dube, R., Wickes, W., &; Wilkinson, D . 1979, ApJ, 232, 333
Felton, J. 1985, Comments on Astrophys., Vol. 11 , No.2, 53
Ferguson, H. 1990, Ph.D thesis, Johns Hopkins University
Fish, R. 1964, ApJ, 139, 284
Fisher, J.R. &; Tully R .B. 1975, A&;A, 44,151
Freeman, K.C. 1970, ApJ, 160, 811
Ichikawe, S.-I., Wakamatsu, K .-I., & Okamura, S. 1986, ApJS, 60, 475
Impey, C.D ., Bothun G .D., &; Malin, D. 1988, ApJ, 330, 634
Irwin, M.J., Davies J.I., Disney M.J., &; Phillipps S. 1990, MNRAS, 245, 289
Kormendy, J. 1977, ApJ, 218,333
Lauberts, A. &; Valentijn, E. 1989, ESO - Uppsala Catalogue, European Southern Obser-
vatory
Mattila, K. 1990, in "The Galactic and Extragalactic Background Radiation", eds. S.
Bowyer &; C. Leinert, p. 257
O'Connell, R. 1987, AJ, 94, 876
Pagel, B., Simonson, E., Terlevich, R., &; Edmunds, M., 1992, MNRAS, 255, 325
Persic, M. &; Salucd, S. 1992, MNRAS, 258, 14P
Pettini M., Hunstead R.W., Murdoch H.S., &; Blades J.C. 1983, ApJ, 273, 436.
Phillipps, S. &; Disney, M. 19~3, MNRAS, 203, 55
Phillipps, S., Davies, J., & Disney, M.J. 1988, MNRAS, 233, 485
Phillipps, S., Davies, J., &; Disney, M.J. 1990, MNRAS, 242, 235
Phillipps, S., Disney, M ., & Davies J. 1992, MNRAS, submitted
Reaves, G. 1983, ApJS, 53, 375
Sandage, A. &; Tammann, G . 1965, Ann. Rep. Mt. Wilson & Palomar Obs., p .35
Sandage, A. &; Pereirnuter , J. 1990, ApJ, 361,1
Sargent, W., Young, P., Boksenberg, A., Carswell, R., &; Whelan, J., 1979, ApJ, 230, 49
Schechter, P. 1976, ApJ, 203,297
Schombert, J., Bothun, G., Schneider, S., &; McGaugh, S, 1992, IPAC preprint No 0080
Thuan, T., Alimi, J., Gott, J., &; Schneider, S. 1991, ApJ 370, 25
Toller , G. 1983, ApJ, 266, L79
York, D., Dopita, M., Green, R. , &; Bechtold, J . 1986 , ApJ, 311, 610
York, D. 1988, QSO Absorption Lines: Probing the Universe, eds. J.C . Blades, D . A.
Turnshek, &; C. A. Norman (Cambridge Univ. Press)
 PANEL DISCUSSION: THE NATURE OF HIGH-REDSHIFT
OBJECTS
J. MICHAEL SHULL
University of Colorado, Boulder CO

Abstract. The panel discussion at the Tetons III conference was held July 6, 1992 on the
topic of "High-Redshift Objects". A number of questions were posed to the panelists, and
these foul astronomers and the audience spent two hoUls debating these and other issues.
The written versions of their presentations follow in the next articles.

1. Introduction
Traditionally at the Tetons Conferences, we devote one evening to a Panel
Discussion, in which four or five astronomers lead the discussion on a contro-
versial topic of current interest. The audience was encouraged to participate
in this discussion, through questions that attempted to "stump the panel".
At the 1992 meeting, the conference organizers chose the topic of "High-
red shift Objects". We requested brief presentations by four astronomers
(Richard Ellis, Simon Lilly, Hyron Spinrad, and Tim Heckman) working
in sub-areas of this field, and the panel discussion was moderated by J. M.
Shull. The written contributions which follow are based on the panelists'
oral presentations at the meeting, coupled with some wisdom of hindsight.

The topics chosen by the panelists were:

• Ellis: Galazy Evolution as a Function of Environment

• Lilly: Distant Galazies: Evolutionary Clues

• Spinrad: Old, Red, and Dead Radio Galazies at Large Redshift

• Heckman: Primeval Galazies, IGM, and QSO-Protogalazy Connection

2. Questions to be addressed by the panelists

To set the scope of the debate, I posed a set of questions on the nature
of high-redshift universe objects; the panelists were urged to address any
aspects that resonated with their interests. In any direct sense, most of
these questions remain largely unanswered. However, it is useful to assess
whether any progress has been made towards any of the major underlying
issues. The following list was culled from discussions with the panelists and
colleagues prior to July 1992.

127
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, 127-130.
© 1993 Kluwer Academic Publishers. .
128

Questions on High-Redshift Objects:

• What is a primeval galaxy (and have we seen one?)

• What is the era of galaxy formation?

• Do cooling flows build galaxies?

• What is the relation between quasars and protogalaxies?

• What is the luminosity function of dwarf galaxies (present and past)?

• Are the faint blue galaxies really dwarfs?

• What are the products (stars, metals, mergers) of these faint galaxies?

• What ionized the intergalactic medium?

• Does the Lya forest have anything to do with galaxies?

• What do red z ~ 1 galaxies say about the last epoch of star formation?

• What are the high-z radio galaxies, if not protogalaxies?

• Why are "post-starburst galaxies" abundant in high-z clusters?

3. SUDlDlary
At the close of the debate, it was my impression that many of these questions
remained unanswered. It wasn't that the panel and audience didn't try; part
of the reason is that much of the debate centered on semantic issues, while
other portions dealt with specific objects. As Hy Spinrad noted, many of
the high-redshift objects that are discovered are "the tail of a distribution",
a minority class of possibly pathological cases. This is clearly a field where
large-scale surveys will have an enormous effect, and will likely clarify which
galaxies are the dominant population and which are the special cases. In
the following paragraphs, I will give some summary impressions on a few of
these questions.
First, have we actually "seen" a primeval galaxy? The answer is probably
yes, but this is largely a semantic issue. The lack of an agreed-upon definition
for a "primeval galaxy" probably means that we wouldn't recognize one if
we saw it. In addition, the clear indications that the first (and last) epochs
 129

of star formation are spread over a range of redshifts make such definitions
inherently imprecise. The new generations of array detectors at 2 - 5 JLm
and 10 - 25 JLm will undoubtedly increase our knowledge dramatically in
this field. Surveys of ~ 104 near-infrared galaxies can be exploited through
number-count anslysis, followed by spectroscopic surveys. The possibility
that such arrays, with low backgrounds, might soon be flown in space
holds even greater promise. One expects to detect redshifted spectral energy
distributions from young galaxies, as well as dust-processed emission from
shrouded QSOs and merging dusty galaxies. Both of these surveys should
help to clarify whether z ~ 1 is the era of significant star formation, and
whether mergers and AGN activity at z ~ 3 are also important.
The ionization of the intergalactic medium was also an active topic of
study. At the meeting, we heard from Boyle, Bechtold, and Meiksin & Madau
that the gap may be narrowing between the required ionizing photon flux and
the QSO number counts at z ~ 3. These studies have relied on new surveys
of QSOs at high redshift, but there are still uncertainties over the extent
of luminosity evolution (and its luminosity dependence). The intriguing
possibility that star burst galaxies provide additional ionizing photons at
high redshift remains unanswered; perhaps spectral information on He I and
He II Gunn-Peterson tests may provide some guidance on the spectral shape
of the dominant ionizing continuum. However, the degree of spectral filtering
of this continuum by the intergalactic H I clouds adds further uncertainty,
since the number counts of Lya systems with column density 1015 cm- 2 :s;
N(H I) :s; 10 17 cm- 2 and the He IjHe II ionization fraction in these clouds
are both uncertain and controversial.
A third fascinating question is whether cooling flows build galaxies.
Musho • .ky gave cogent arguments that the 100 M0 yr- 1 infall rates inferred
from cluster X-ray emission are difficult to explain away. If the duration
of these flows is comparable to the Hubble time, then substantial mass is
deposited into the central galaxies of rich clusters. We do not know where this
mass goes; it cannot all go into stars with a normal initial mass function. Is
the ratio of gas (baryonic) to dark (non-baryonic) matter seen in clusters of
galaxies is a fair sample of matter in large-scale universe? For some clusters
(but not all) a significant fraction ofthe binding mass appears to be baryonic.
However, the possibility of baryon fractionation must be settled before one
can make definitive statements about no.
A related unanswered question is whether the "dark matter haloes" in
galaxies and clusters extend to the sub-galactic ("minihalo") scale. Could the
Lya forest clouds be confined by these minihaloes, as Rees and Ikeuchi have
suggested? Until we better understand the physical density and structure of
the QSO absorption clouds, we cannot critically assess whether Lya clouds
have anything to do with galaxies. Answering the cloud confinement question
may help, as may spectroscopic inferences about the ionization fraction of
130

these systems. I also have a feeling that we have more to learn about these
clouds and dwarf galaxies through careful two-point correlation studies.
Possibly the area in which most recent progress has been made are the
K-band surveys, which have turned up numerous faint blue galaxies at B
~ 23 - 24. The frequency of these galaxies appears to be inconsistent with
no-evolution predictions, implying a change in the star-formation rate, or
mergers, or even a new population of dwarf galaxies. The possibility that
these sub-L* galaxies might have strong galactic winds could provide a
relation to low-redshift Lya absorbers, and the star formation could also
explain the Butcher-Oemler effect in clusters of galaxies. Thus, although
the z ~ 1 epoch is hardly the high-redshift era that Partridge, Peebles, and
others had in mind when they first theorized about proto-galaxies, it does
appear to be an important epoch in the life of galaxies. Astronomy may be
redefining the definition of "primeval galaxy", as we learn more about the
universe at faint magnitudes.
 GALAXY EVOLUTION AS A FUNCTION OF
ENVIRONMENT
RICHARD ELLIS
Physics Department, University of Durham, Durham DHl 9LE, UK

1. Introduction

In the theoretician's ideal Universe, galaxies are formed at some character-

istic epoch, Zjorm, and evolve to their present forms and stellar mixtures
according to well-understood laws governing star formation and stellar evo-
lution. The diverse distribution of local properties might be explained via
a range of star formation histories and ages, originating from different con-
ditions at birth. Alternatively, since galaxies are not isolated objects, in-
teractions and other local processes may influence their evolution in com-
plex ways. In either case, morphologies and stellar populations would show
marked variations from place to place, as observed.
The distinction between 'nature' and 'nurture' has been a long-standing
problem in cosmology, and this is reflected in the title of this meeting. The
principle difficulty in distinguishing between them with local data is that
diverse evolutionary histories might simultaneously reproduce a given ob-
servation. One has to 'search for environmental processes in action or infer
them from the internal structures of selected galaxies, in either case gener-
alising the conclusions derived from individual examples. The difficulty is
apparent when it is realised that the properties of lenticular galaxies, long
considered valuable in this regard, have been used to argue independently
for both the nature and nurture hypotheses.
In contrast, distant sources offer a direct probe of evolution and, more-
over, can be observed systematically in large numbers. At various epochs,
populations can be compared across field, group and rich cluster environ-
ments. Recently, we have begun to see such data from a variety of techniques
and, it appears that significant differences are emerging in the evolutionary
trends as a function of environment. The challenge to observers and theo-
reticians alike is to interpret the new observations in a coherent picture of
galaxy formation. In this brief presentation, I will summarise the various
observations at large look-back times.

2. Rich Clusters
The logical place to begin is with early-type galaxies in rich clusters. Sandage
& Visvanathan (1978) and, more recently with greater precision, Bower et
al (1992) have demonstrated the remarkable homogeneity of the rest-frame

131
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, 131-142.
© 1993 Kluwer Academic Publishers.
132

u- V colours (~ ±3-5%) of early-type galaxies in nearby clusters; indeed, to

the accuracy of modern CCD photometry there is negligible scatter around
the trend with luminosity. This places very tight constraints on the amount
of recent star formation in the cores of rich clusters « 10% by mass since
z~0.5).
Distant clusters allow us to check this locally-derived conclusion directly,
as well as to extend the argument to earlier times. Aragon-Salamanca et
al (1992) recently completed a photometric survey of the 10 most distant
clusters (0.5< Z <0.9). By selecting galaxies i:rr the K-band, they minimise
bias to blue star forming systems. The mean colour of the early-type ridge
line or 'red envelope' (O'Connell 1988) moves systematically blueward at
high red shift (Figure 1) and,moreover, the trend is shared by all clusters at
a given redshift. This supports the notion that ellipticals are old systems:
formally, we can eliminate models where zJorm <2 unless the mass function
is highly non-standard. Moreover, the similarity in the evolutionary trends
across the clusters sampled suggests ellipticals are coeval, passively evolving
from a single burst. The absence of any detectable luminous phase in pri-
maeval galaxy searches (Djorgovski & Thompson 1992) indicates this burst
happened a long time ago, possibly zJorm > 3 - 4.
The emerging picture is thus one where the stars in giant ellipticals
formed mostly during a single burst at large redshift. Note that this dating
refers to the stellar populations and does not, necessarily, preclude some
merging of galaxies. Aragon-Salamanca et al find little evidence for any
change in the cluster K luminosity function since Z ~1, other than that
expected for the passive evolution implied by Figure 1. However, this con-
clusion only applies to the most massive galaxies in the richest environments.

3. Field Galaxies
These are equally exciting times for the studies of faint field galaxies, many
aspects of which are reviewed here by Simon Lilly. I would stress foremost
that we are entering a new era of faint redshift measurement made possible
by wide-field multislit devices such as the Low Dispersion Survey Spectro-
graph (Colless et a11990) and large format thinned CCDs of high quantum
efficiency and low read noise. Allington-Smith et al (1992a) demonstrate
how redshifts within 22.5< .B <24 can be acquired with high completeness
in 3-4 hour exposures on the 4.2m Herschel telescope. Similar results are
being gathered with MOSIS at CFHT by Cowie et al (1991), Lilly (1992)
and Tresse, Hammer & Le Fevre (1992). I remain optimistic that when 'the
stops are pulled out' in good conditions, B=25 surveys will be feasible on
4-m telescopes. The main limitation will be incompleteness arising from the
disappearance of the diagnostic [0 II] emission line from the optical regime
prior to the arrival of Lya in the UV. In this respect the instrumental chal-
 133

....... -1
~
I
>
......
<l

-2
(a)
-3
0 .2 .4 .6 .8 1
z

.5

.......

-
~
I
.<l
.....
- .5

-1 .,
(b)
-1.5
0 .2 .4 .6 .8 1
z

Fig. 1. Mean V -K and I -K properties for K-selected samples in 10 Z >0.5 clusters

from the survey of Aragon-Salamanca et al (1992), normalised so that zero colour
at any redshift represents that expected for a present day elliptical galaxy. The
data are binned in 3 redshift intervals and compared with Bruzual's (1983) passive
evolution (c-model with r=1 Gyr, z/orm=2 (solid), 5 (dotted) and 10 (dashed) for
qo=0.5 (thick) and 0.0 (thin)). The point at Z >1 comes from a single radio quiet
source identified close to 3C245 (Le Fevre & Hammer 1992), and that at z=0.37
refers to data for Abell 370 (Aragon-Salamanca et aI1991). Note the systematic
evolutionary trend, particularly in V - K .
134

lenge is now to design an efficient wide field instrument of extremely broad

wavelength coverage (320nm - 1JLm).
Lilly has described the 'faint blue galaxy conundrum': for a decade it has
been known that there are x4-6 too many galaxies to B=24 compared to the
no evolution prediction (c.f. Peterson et aI1979), yet statistically complete
surveys show the volumes probed are as expected in the same no evolution
model. Inescapably therefore, there appears to be some form of number evo-
lution and/or a dramatic increase in the star forming luminosity density.
Casting aside explanations involving a major misunderstanding in the basic
properties of galaxies or the cosmological model, the evolution appears to
involve star formation in low luminosity systems. 'Low luminosity' since the
dramatic excess is not so apparent in K -band light and thus, in terms of
the underlying older population, the galaxies cannot be that massive. 'Star
formation' not only because of the blue colours, but also because of increas-
ing proportion of strong [0 II] emission lines as one progresses to fainter
sources. The archetypal faint blue galaxy is essentially an extragalactic HII
region whose number density far exceeds local estimates.
Broadhurst et al (1992) present a self-consistent picture whereby low mass
sub-units decline in their star formation rate since z~1 and slowly merge.
Rather a lot of recent merging is required to explain the excess whereupon
it is then difficult to explain the abundance of thin disk structures seen
today (Ostriker 1990). Alternatively, Babul & Rees (1992) suggest a new
intermediate-redshift population of dwarfs may provide the excess and that
these have faded by today. Stellar fading in K -light cannot proceed faster
than passive evolution and, given the excess is particularly prominent at
z ~0.4, calculations suggest a sizeable Euclidean excess would be seen in
the faint K counts which is not the case. AdditIonally, the faintest B red-
shift surveys now reveal a lower redshift limit of ~0.1: few intrinsically faint
optical systems are found as might be expected if this enormous excess is
rapidly decaying down the luminosity function to faint absolute magnitudes.
To overcome these objections it appears that one must disrupt a star form-
ing dwarf altogether. This seems rather a contrived situation where sources
pile up in a specific time interval but are undetectable both before and
afterwards.
Despite theoretical objections to recent merging, high resolution HST
imaging of I <22 galaxies from the Medium Deep Survey key project is pro-
viding some supporting examples. Long integrations with HST have shown
that the narrow core of the point spread function emerges in tact, yielding
unique information on faint galaxies on ~0.1-0.2 arcsec scales (see Figure 2).
However it is difficult from images alone to differentiate between a genuine
merger and a lumpy irregular undergoing star formation. The HST images
also reveals a class of amorphous low surface brightness objects whose na-
ture remains unclear. Collectively the two classes could provide the missing
 135

excess population. At the present time few redshifts are available in the MDS
fields, but the sources are within range of available 4-m spectrographs.
Regardless of the explanation, it is clear that the field galaxy luminosity
function has changed shape in the past 5-8 Gyr. Although the number of
L B galaxies has increased 3-6 fold, the absence of a significant high z tail
severely constrains any evolution at the bright end. The question of whether
incompleteness in the survey redshifts might weaken this conclusion was
thoroughly discussed by Colless et al (1990, 1992). The next few years will
see a dramatic increase in the number of redshifts from fibre systems at
intermediate magnitudes 19< B <22 enabling a direct estimate of the galaxy
luminosity function with redshift e.g. as a function of the [0 II] strength, in
the manner discussed by Boyle (herein) for the faint QSO population.

4. Environmental Effects at Moderate Redshifts

Although the rich cluster data to z ~1 suggests the bulk of the star forma-
tion in giant ellipticals occurred long ago, there is mounting evidence that
elliptical galaxies in less rich environments may not be similarly explained.
O'Connell (1980), Pickles (1985) and Bower et al (1990) have presented con-
vincing cases of early-type systems with 'intermediate age' (~3-5 Gyr) stel-
lar populations. Indeed using line indices independent of metallicity changes,
Bower et al find a direct correlation between the mean stellar dwarf/giant
ratio and the environmental velocity dispersion (c.f Ellis 1992, Fig. 4) in the
s~nse that ellipticals in less rich systems are more likely to have undergone
recent star formation.
Is it possible to figure out what happened by directly observing similar
sources across various environments at modest redshifts? For ellipticals per
se, morphologies are strictly required since a burst of star formati6n would
transform the colour/spectral properties. Morphologies are now practical
with HST in rich clusters to z ~0.4 (c.f. impressive results discussed by
Couch et al 1992 and Dressler 1992), but to secure morphologies for indi-
vidual field ellipticals at various z would be a daunting task.
A start, however, has been made by examining the environmental varia-
tion in the 'Butcher-Oemler' effect (Butcher & Oemler 1978), whereby some
subset of the rich cluster population becomes noticeably bluer at z >0.3. Is
this a generic property of all galaxies of a certain kind, or a rich cluster phe-
nomenon indicative of strong environmental processes? By utilising the fact
that moderately luminous radio galaxies occur in groups, Allington-Smith
et al (1992b) have imaged a sample of almost 100 radio-selected groups in
two redshift intervals, 0< z <0.25 and 0.25< z <0.5. They compare the
field-subtracted colour distribution of the group members in the two red-
shift intervals in a manner virtually identical to the classical rich cluster
study, save for the fact that data from many groups must be combined to
136

Mergers LSB gals

R.A. (uoooo)

10 D 10

RA (oro ... ) R.A.(.....-)

10
R.A. (&rClleo)

Fig. 2. HST Wide Field Camera images of selected I <22 galaxies from the HST
Medium Deep Survey (Griffiths et alI992). Each panel covers a field 10 x 10 arcsec
in extent. Within this magnitude range, Lilly (1992) finds a mean redshift z ~0.5
and a number excess of x 3. The leftside shows examples of possibly merging or
interacting systems, and the rightside shows a class of low surface brightness blue
amorphous sources. Collectively, such systems are sufficiently numerous to provide
the excess count to I=22.
 137

80 , 80
o 2S < z < O· S o< z < 0·25

r
60 60

n
(i
~rl~J
40 40

l
20 20

0 L-J
U 0 w ~
nruJ L.
Ll

-20 -20
-1 -0.5 0 0.5 -1 -0.5 0 0.5
A(B-V)o A(B-V)o

Fig. 3. Coadded rest-frame B - V colours (normalised to zero for an elliptical galaxy

of that luminosity) for radio-selected groups within red shift slices 0< z <0.25 and
0.25< z <0.5 from the imaging survey of Allington-Smith et al (1992b). For each
group, members with MB <-19 are selected within r <0.5 Mpc (H o =50) of the
radio source with careful allowance for field contamination via offset images. No
increase in the blue fraction with redshift is seen, indicating the Butcher-Oemler
effect is primarily a rich cluster phenomenon.

yield a reliable histogram. Allington-Smith et al find no significant increase

with redshift in the blue fraction in groups indicating the B-O -effect must
be intrinsic to the rich clusters (Figure 3).
Recognising that the B-O effect is environmental in origin, we can ask
what process induces the star-formation cycle that links the 13-0 blue galax-
ies, the 'E+A' post-starburst galaxies (Dressler & Gunn 1983, Couch &
Sharples 1987) and the ultraviolet excess optically-red galaxies (MacLaren et
al1988)? Some possibilities are reviewed by Oemler (1992) and the prospects
for resolving this outstanding problem with HST are excellent. The first HST
images are discussed by Couch et al and Dressler (1992) and show the bluest
members are a mixture of normal spirals and interacting systems. The post-
starburst galaxies are the most crucial to understand since it is they which
138

demonstrate most clearly the sudden change in star formation history at

z~0.3-0.4. Couch et al have a fairly good selection of such examples in their
first HST images, the bluest of which are clearly interacting. However, few of
the redder 'E+A' galaxies are disk-like, suggesting the present day SO popu-
lation did not necessarily form from gas-stripped spirals associated with the
B-O phenomenon.
Placing these new results into a coherent jigsaw of galaxy evolution in
different environments is not easy. Evidently the passive evolution of the
massive clusterellipticals is correct only to first order and may not preclude
some secondary star formation associated with the 'E+A' phenomenon in
z~0.3 clusters. However, whatever drives the rapid decline in disk star for-
mation in rich clusters is peculiar to these systems so that as recently as
z ~0.5 the blue fraction defined by Butcher & Oemler and Allington-Smith
et al would have been largely independent of environment. Reconciling these
remarkably recent environmental variations and changes in the field galaxy
luminosity function with the much slower changes in the massive ellipticals
is the puzzle that now faces us.

5. Where Next?
We have seen how, in the redshift interval 0< z <1, high resolution imaging
from ground-based telescopes on excellent sites and, particularly from HST,
promises to reveal the morphological identity of samples of faint blue galaxies
to B ~23-24, as well as the origin of the Butcher-Oemler effect in z ~0.4
rich clusters. '
But what about more distant galaxies? If merging is rampant in the field
galaxy population there should be dramatic changes in the luminosity func-
tion by z~2. In terms of detecting distant sources, traditional techniques
start to fail beyond z~0.5-1 and we would be wise to consider alternatives.
Limitations include clusters which cannot be identified reliably by optical
means beyond z~0.5 because of projection effects, field redshift surveys con-
strained by the lack of optical features for 1< z <2, and the inherent diffi-
culty of much of the work described above which, after all, will only proceed
~4 times faster with the new generation of telescopes.
Two areas are becoming increasingly important in probing beyond z = 1
- gravitational lensing by foreground clusters, and linking QSO absorption
lines with normal galaxies. The latter is not a new idea, of course. Berg-
eron and Boisse (1990) pioneered the identification of z < 1 Mg II absorbers
with massive galaxies (for an update see Steidel's article herein). Arag6n-
Salamanca, Bergeron and I have commenced an equivalent K-band search
for the more distant z > 1 CIV absorbers. The near-zero k-correction at 21lm
for all Hubble types makes it feasible to detect a L* galaxy, regardless of its
spectral energy distribution, to z ~2 in a few hours of 4-m time. In time,
 139

we can expect constraints on the K luminosities and star formation rates

for representative absorbers as a function of look-back time to z~3.

Lensing offers two distinct prospects. First, there are the 'giant arcs'
which are readily recognisable by their unusual morphologies. Soucail (1992)
has reviewed the salient properties. About a dozen are known and some have
redshifts. Since lensing occurs only because a source lies serendipitously be-
hind an unrelated cluster, arcs should be representative of field galaxies at
large redshifts. However, arcs are found via their surface brightness, a quan-
tity conserved in lensing, and, furthermore, since all have been found at
rest-frame wavelengths in the far UV, we would be wise to consider selec-
tion effects carefully before assigning their blue colours to any primordial
phase of star formation in normal galaxies. Nonetheless, as lensing models
constrain the magnifications more tightly and K data is gathered to prop-
erly constrain the longer-lived stellar populations, one can hope to make
progress. Regrettably, the sample is small and it is exceeding difficult to get
reliable redshifts. The bulk appear to have 0.7< z < 1 which is only slightly
more distant than the mean B=24 field galaxy population.

The second lensing test is more fundamental and concerns statistical dis-
tortions of more numerous faint field galaxies selected to apparent magni-
tudes well beyond the limits of spectrographs on any projected telescope.
Tyson et al (1988) have pioneered this application. Clusters whose mass dis-
tributions are reasonably well-understood from dynamical, X-ray or other
data, may 9·ultimately be used as 'well-constrained' lenses to yield statis-
tical distances to field galaxies independently of the problems associated
with traditional techniques. Some new results in this area are discussed by
Smail (1992), who has combined deep imaging through 3 X-ray clusters with
z=0.22,0.54 and 0.89 attempting to get a self-consistent N (z) for field galax-
ies to I =25. He finds it is very difficult to place a substantial fraction of the
I =25 population beyond z=l unless each of the 3 clusters is considerably
less massive and concentrated that the existing dynamical and X-ray data
indicate.

In summary, we learn that the interval 0< z <1 was an exciting one for
galaxies: their properties were affected by their environment and high reso-
lution imaging, especially with HST, promises to put this in the context of
morphologies that we can directly connect to present day galaxies. To probe
beyond z=l, however, we will need to be more imaginative, but exciting op-
portunities are available via the absorbing effects of galactic halos detected
in QSO spectra, and rich clusters which act as natural lenses for distant field
galaxies.
140

Acknowledgements

I thank Alfonso Aragon-Salamanca, Jeremy Allington-Smith, Warrick Couch,

Karl Glazebrook, Gus Oemler, Ray Sharples and my fellow MDS workers
for allowing me to present results obtained in collaboration ahead of pub-
lication and to co-panelists Simon Lilly, Hy Spinrad and Tim Heckman for
a stimulating discussion. Joss Bland-Hawthorne is thanked for providing an
injection of caffeine prior to the panel debate essential to overcoming jet-lag.
Finally, I thank the organisers Mike Shull and Harley Thronson for their en-
thusiasm evident throughout this meeting, and for their generous financial
support.

References
Allington-Smith, J.R., et a11992a, Gemini RGO Newsletter, 36, 4.
Allington-Smith, J.R., Ellis, R.S., Zirbel, E. & Oemler, A. 1992b, Astrophys. J., in press.
Aragon-Salamanca, A., Ellis, R.S. & Sharples, R.M. 1991. Mon . Not. R. astr. Soc., 248,
126.
Aragon-Salamanca, A., Ellis, R.S., Couch, W.J. & Carter, D. 1992, Mon. Not. R. astr.
Soc., in press.
Babul, A. & Rees, M.J. 1992, Mon. Not. R. astr. Soc., 255, 346.
Bergeron, J. & Boisse, P. 1991 A8tron. Astr., 243,344.
Bower, R.G., Lucey, J.R. & Ellis, R.S. 1992 Mon. Not. R. astr. Soc., 254,601.
Broadhurst, T.J., Ellis, R.S. & Glazebrook, K. 1992 Nature, 355, 55.
Bruzual, G . 1983 Astrophys. J., 273, 105.
Butcher, H. & Oemler, A. 1978 AstrophY8. J., 219, 18.
Colless, M., Ellis, R .S., Taylor, K. & Hook"R.N. 1990 Mon. Not. R. a8tr. Soc., 244,408.
Couch, W .J., Ellis, R.S., Sharples, R.M. & Smail, 1. 1992 in Observational Cosmology, eds.
Chincarini, G. et ai, in press
Couch, W.J. & Sharples, R.M. 1987 Mon. Not. R. astr. Soc., 299, 423.
Cowie, L. L., Songaila, A. & Hu, E. 1991 Nature, 354, 460.
Djorgovski, G. & Thompson, D.J. 1992 in Stellar Populations in Galaxies, eds. Barbuy,
B. & Renzini, A., Kluwer, p337.
Dressler, A. & Gunn, J.E. 1983, Astrophys. J.,263, 553.
Dressler, A. 1992 STScl Newsletter,9, No.2.
Ellis, R.S. 1992 in Stellar Populations in Galaxies, eds. Barbuy, B. & Renzini, A., Kluwer,
p297.
Griffiths, R.E., Ratnatunga, K., Doxsey, R., Ellis, R.S., Glazebrook, K., Gilmore, G., El-
son, R., Schade, D., Green, R., Valdes, F., Huchra, J.P., Illingworth, G.D., Koo, D .C.,
Schmidt, M ., Tyson, A., Windhorst, R.A., Neuschaefer, L., Pascarelle, S., Schmidtke, D.
1992 in Observational Cosmology, ed. Chincarini, G. et ai, in press.
Le Fevre, O. & Hammer, F. 1992 Astr. A8trophys. Lett., in press.
Lilly, S.J. 1992 Astrophys. J., in press.
 141

MacLaren, 1., Ellis, R.S. & Couch, W.J. 1989 Mon. Not. R. astr. Soc., 230,249.
O'Connell, R .W. 1980 Astrophys. J., 236,340.
O'Connell, R.W . 1988 in Towards Understanding Galaxies at High Redshift, eds . Kron, R.
& Renzini, A ., Kluwer, p177.
Oernler, A. 1992 in Clusters iJ Superclusters of Galaxies, ed. Fabian, A., Kluwer, p29.
Ostriker, J.P. 1990 in Evolution of the Universe of Galaxies: Hubble Centennial Sympo-
sium, ed. Kron, R., p25.
Peterson, B.A., Ellis, R.S., Kibblewhite, E.J., Bridgeland, M.T., Hooley, A. & Horne, D.
1979 Astrophys. J. Lett., 233, LI09.
Pickles, A.J 1985 A&trophys. J., 296, 340.
Sandage, A. & Visvanathan, N. 1978 Astrophys. J., 223, 707.
Soucail, G. 1992 in Clusters iJ Superclusters of Galaxies, ed. Fabian, A., Kluwer, p199.
Smail, I. 1992 Ph.D. thesis, University of Durham.
Tresse, L., Hammer, F. & Le Fevre, O. 1992, preprint.
Tyson, A.J, Valdes, F. & Wenk, R.A. 1989 Astrophys. J.,349, L1.
 DISTANT GALAXIES: EVOLUTIONARY CLUES

SIMON LILLY
Department of Astronomy, University of Toronto; 60 St. George Street, Toronto,
Ontario MSS-1A 7, CANADA

May 16,1993

Abstract. In this short discussion, I will concentrate on what we are learning from studies
of distant galaxies, about how and when they are formed, and how they subsequently
evolved. I will concentrate primarily on studies based on photospheric emission of starlight
rather than on absorption.

1. Field Galaxies
The most general view of galaxies comes from field galaxies selected without
regard to their cluster environments or other peculiar properties. In the last
couple of years, we have begun to push spectroscopic surveys down to very
faint limits, B ~ 24, or equivalently lAB ~ 22.5. The results have been
rather interesting.
The B-band counts are steeper than expected for an unevolving popula-
tion of galaxies. This produces an excess of faint galaxies that is generally
estimated to be around a factor of five at B ~ 24 for no ~ 1. However, as
discussed by Lilly et al. (1991), this excess can be reduced: (a) by decreasing
qo to '" 0 or lower, by introducing a non-zero cosmological constant A; (b) by
making the population of galaxies rather bluer than the best local estimates,
e.g., by defining the non-evolving population to be as seen at z ~ 0.4; or (c)
by increasing the normalization 4>* or the faint-end slope 0: of the luminosity
function from the most recent local determinations.
When the redshifts are measured for faint samples selected down to
B ~ 24, the surprising result has been found (cf., Colless et a1. 1990; Lilly et
al. 1991; Cowie, Songaila, & Hu 1991; Ellis 1993) that, despite this excess,
the N (z) distributions look remarkably like that expected for no evolution,
with a median redshift (z) ~ 0.42 for B ~ 23.5. This result implies that the
galaxies causing the excess at z ~ 0.4 have slightly sub- L * luminosities at
3000 A. My own recent sample with lAB ~ 22.5 supports this conclusion
(Lilly 1993) . The redshift distribution (Figure 1) peaks at 0.4 ~ z ~ 0.5, at
which point the galaxies have Mv ~ -19.7 (for Ho = 50 km s-1 Mpc':" l and
no ~ 1), i.e. 1 - 2 magnitudes below L* at 5500 A and an apparent comoving
density about 2.5 times that of the present-day luminosity function at that
point.
A quite general problem that arises in the study of faint galaxies is how
to associate distant sources of light with any local property of the Universe.
In this regard, the production and return to the interstellar medium (ISM)

143
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution ()lGalaxies, 143-150.
© 1993 Kluwer Academic Publishers.
144

Mv Mv MB
-19.8 -21.0 -21.4
I 1 I
Unidentified

D
0.5 1 1.5
redshift

Fig. 1. Redshift distribution of faint galaxy sample (Lilly 1993) with hB :::; 22.5,
showing distribution similar to the no-evolution expectation at redshift z ~ 0.4-0.5.

of heavy elements seems to be a relatively robust accounting unit. The far-

ultraviolet luminosity of a blue stellar population is emitted by the same
massive stars that produce and return large quantities of metals to the
ISM, so a far-ultraviolet energy density can, in principle, be directly related
to a degree of enrichment of the Universe (cf., Songaila et al. 1990). The
main uncertainties in this estimate are the mass fraction of heavy elements
returned to the ISM as a function of mass and the form of the upper end
of the initial mass function (IMF). Using this formalism, one can compute
that the faint galaxy population is producing a present-day density of metals
of order 106 . 3 M0 Mpc- 3 • This estimate is formally independent of the
evolutionary timescales, the redshifts, and even no (Lilly & Cowie 1987)
and corresponds to ,. . ., 20hS'J% of the metals seen in present-day galaxies.
Of more interest, it is ,. . ., 10 times the density of metals (M0 Mpc- 3 ) seen
in the local populations in galaxies of type Sdm (10 5 •3 h 5o ), Irr (10 5 •o h 5o ),
 145

and dE (10 4 •3 h so ), which have comparable comoving number densities in the

present-day Universe, assuming a Binggeli et al. (1988) luminosity function
and solar metallicities.
The potential overproduction of metals has led to various evolutionary
scenarios for these objects, involving: (a) extensive mass loss of enriched gas
and an IMF biased toward high-mass stars; (b) the merging of these galaxies
into present-day L· galaxies where we see most of the metals today; or (c) a
"new population" of objects whose unidentified remnants would exist around
us today in large numbers. It is not clear which, if any, of these processes is
actually occurring. All of them, it should be noted, invalidate the traditional
closed-box, number-conserving models for the evolving galaxy population.
In particular, the extent of the problem of metal overproduction is reduced
if the implied comoving number density is reduced, either by altering the
cosmological model (decreasing qo) or by increasing either </>. or a from their
best-estimate "local" values. However, in retrospect, each of these processes
is quite reasonable and is, to a large degree, expected theoretically (cf. Dekel
& Silk 1986; Carlberg & Couchman 1989).
One interesting result found by both the Durham group (Ellis 1993) and
by myself (Lilly 1993) is that a large fraction (50% in my data) of the blue
galaxies at z ~ 0.4 show morphological evidence of interactions or other
disturbances. Other direct evidence for unusual activity comes from blue
colors and high equivalent widths of [0 II] 3727.
Finally, my own I-band sample is ideally suited to studying galaxies at
0.5 :::; z :::; 1, since the I-band corresponds to rest-frame B and V at these
redshifts. While the numbers of such galaxies seen is broadly as expected,
if normal L· galaxies have undergone luminosity evolution of 0 :::; MB :::; 1
magnitudes back to z ~ 1, an interesting result is that the color distribution
is distinctly bluer. All the identified galaxies (Figure 2) have roughly Irr
spectral energy distribitions (SEDs), whereas locally such galaxies have
generally Sbc or redder SEDs. Indeed, all of the objects that have the colors
to be quiescent galaxies at z ~ 1 are reliably identified with stars or red
galaxies at z :::; 0.7.

2. Cluster Galaxies
A great deal of work has been aimed at understanding the evolutionary
effects in clusters over the range 0 :::; z :::; 1. The Butcher-Oemler effect is
clearly due primarily to an increase in star-forming galaxies, many of which
appear to have undergone dramatic bursts to convert a large fraction of their
mass into stars (e.g., Dressler & Gunn 1983). In many cases, the absence
of strong emission lines indicates that the gas has been completely removed
from the objects. There is growing evidence that these objects have a high
incidence of disturbed morphologies, which in many cases are indicative of
 146

10 7
10 «I

10 6
...........
w
~
Cl)
10 •
~
............
~
1000
cO
S 100
............
Z
~
'-'" 10 •
0
Lilly et al 1991
Tyson 1988

- ... t.
...•
~
Metcalfe et al 1990
0 1 ... A
A B Maddox et al 1990
AA A Heydon-D et al 1989
0.1
A
A •
0
Cowie et al 1991
Glazebrook et al 1991

0.01 • Hall & Mackay 1984

15 20 25 30 35
KAB • IAB+4. BAB +7

Fig. 2. Number counts in the K, I, and B bands (see Lilly et al. 1991 for .references).

interactions with other galaxies. However, along with the increased fraction
of blue galaxies, the majority of the cluster members at z ~ 0.5 form a
homogeneous population of red, extremely quiescent galaxies that clearly
finished forming well before the epoch of observation. As we look farther
back in time, however, this red population becomes considerably bluer and
shows distinct signs of youth. By z ~ 1 it is much harder to discern any
truly quiescent population.

3 . High-redshift Radio Galaxies

I have, for several years, argued that the remarkably small dispersion in the
infrared K(z) relation for 3C and "1 Jansky" radio galaxies, out to z ~ 2
and beyond (Figure 3), and their generally quite-red colors indicated that
these were basically mature gE galaxies containing some additional blue
component. This component, of unknown origin, was subsequently found to
 147

21
/

20 • 3C sample (complete) qo =0 /
/
qo = 0.5
/
o 4C-Ievel (incomplete)
19 /

----
($I
/
() / . 0 00 0
Q)
18
.,;,.
/ .0. 0

...,."...-. .
rn • •/ .... 0
() 0
$-t ~
a:s 17 .~ 0
CO •
'-'"
16 /'
Q)
0
'U
15
....+>a
~
•
0

14
tlO
ro •
e •
•• •.·0•
13
:::.:::
12 -1

"-J
11

18.01 0.1 1 10
redshift

Fig. 3. Infrared K(z) relation (8 arcsec aperture magnitudes) for for 3C and 4C
samples of radio galaxies.

be aligned with the radio axis (McCarthy et a1. 1987); possible explanations
included young stars, scattered or anisotropically emitted nuclear emission,
or one of several mechanisms involving relativistic electrons (see Daly 1992
for a review).
Recently, this interpretation has faced two challenges. The first concerned
claims that the infrared morphologies were inconsistent with the infrared
light being dominated by gE galaxies (e.g., Eisenhardt & Chokshi 1990).
This has, I believe, been refuted, certainly at z ~ 1, by the quantita-
tive multi-wavelength analysis of Rigler et a1. (1992), who showed that
the infrared morphologies are indeed more symmetrical than at shorter
wavelengths, to the degree expected. Furthermore, Rigler & Lilly (1993)
have shown that the reddest galaxies at 1 ::; z ::; 1.5, which should be
least contaminated by the blue aligned component, have surface brightness
profiles well matched by de Vaucouleurs profiles, with characteristic size
148

scales and surface brightnesses remarkably similar to those of local gE radio

galaxies. However, Eales & Rawlings (1993) and Eisenhardt & Dickinson
(1992) have independently shown that in at least one of the very high redshift
systems (0902+34 at z = 3.4) most of the infrared light comes from line
emission ([0 IIIj4959, 5007 and H,8 4861) despite only a modest increase in
the scatter in the K(z) relation.
Thus, while there is still considerable uncertainty in what is going on
at the highest redshifts, the conclusion seems to be holding ":lP that radio
galaxies at z :::; 2 are well-formed stellar systems that underwent the bulk of
their star formation at epochs of order 1 Gyr earlier.

4. A Composite Picture of Galaxy Evolution

I personally take away from these studies two conclusions:

First, as we look back in time, I believe it is possible to discern a graded

sequence in the level of evolutionary activity. Locally, we have activity in
relatively low-mass systems. By z ~ 0.35 (7" ~ 0.667"0) we see extensive
activity, with the field populations dominated by blue galaxies with blue
Irr-type SEDs and Lv ~ 0.2L*. In the rich clusters, we have similar activity
producing the Butcher-Oemler effect, but still overall domination in these
richer environments by the very quiescent gE galaxies. By z ~ 1 (7" ~ 0.357"0)
we see sharply decreasing numbers of quiescent galaxies in both the field and
in clusters. Finally, at z ~ 2 (7" ~ 0.27"0) we may suppose, from the peak
in quasar activity and the phenomena seen in radio galaxies, that even the
most massive galaxies are undergoing fairly dramatic changes.
Secondly, it appears that, as one might have expected, galaxy evolution
is likely to be a complex process involving extensive feedback loops, both
within galaxies and between galaxies and their environments. These pro-
cesses may control, either through the mass of the galaxy or the richness of
the environment, the phased evolution described in the previous paragraph.
If these general ideas are correct, then we will encounter a great deal of
interesting astrophysics if we seek to understand how galaxies formed and
evolved. Unfortunately, we will also find it even harder to use galaxies in
cosmological studies of the Universe. The use of galaxies as unbiased tracers
of the large-scale structure and as direct indicators of the growth of structure
in the Universe will all be much more difficult .

References
Binggeli, B ., Sandage, A., & Tammann, G . 1988, ARAA, 26, 509
Carlberg, R. G., & Couchman, H. M. P. 1989, ApJ, 340, 47
Colless, M., Ellis, R . S., Taylor, K., & Hook, R. N. 1990, MNRAS, 244, 408
Cowie, L . L., Songaila, A., & Hu, E. M. 1991, Nature, 354, 460
 149

Daly, R. 1992, ApJ, 399, 426

Dekel, A., &; Silk, J. 1986, ApJ, 303, 39
Dressler, A., &; Gunn, J . E. 1983, ApJ, 270, 7
Eales, S. A., &; Rawlings, S. 1993, ApJ, in press
Eisenhardt, P., &; Chokshi, A. 1990, ApJ, 351, L9
Eisenhardt, P., &; Dickinson, M. 1992, ApJ, 399, 47
Ellis, R., 1993, these procedings
Lilly; S. 1993, ApJ, in press ,
Lilly, S., &; Cowie, L. L. 1987, in "Infrared Astronomy with Arrays", ed . C . G . Wynn-
Williams &; E. Becklin (Honolulu: Univ. of Hawaii), p 473
Lilly, S. J., Cowie, L. L., &; Gardner, J. P. 1991, ApJ, 369, 79
McCarthy, P. J., van Breugel, W., Spinrad, H., &; Djorgovski, S. 1987, ApJ, 321, L29
Rigler, M. A., &; Lilly, S. J. 1993, in preparation
Rigler, M. A., Lilly, S. J., Stockton, A., Hammer, F., Le Fevre, O. 1992, ApJ, 385, 61
Songaila, A, Cowie, L. L., &; Lilly, S. 1990, ApJ, 348, 371
 OLD, RED, AND DEAD RADIO GALAXIES AT LARGE
REDSHIFT
HYRON SPINRAD
Department of Astronomy, University of California Berkeley

Abstract. A minority of powerful radio galaxies at z ~ 0.7 are quite red, and they show
stellar spectroscopic signatures in their near- UV spectra. These galaxies must have formed
ana ceased star formation long ago. A few trial comput!:'tions of z f' the last star-formation
epoch, are presented for one red radio galaxy at z = 0.97.

1. The epoch of last significant star formation in old gE galaxies

While many of the peculiarities of radio galaxies arguably result from strong
and recent jet activity accompanied by star formation, there does exist a
tail of less-spectacular systems that are quite red in both observed and
rest-frame colors. These galaxies persist beyond z = 0.7, although they are
somewhat rare in complete redshift samples. This is partly due to the fact
that their redshifts are harder than normal to determine. They tend to show
low-excitation emission lines, weaker than other radio galaxies of bluer color
(Spinrad 1986; McCarthy 1990). In those respects they can be termed dead.
The red radio galaxies, while still powerful radio sources (FR II morpholo-
gies), are desynchronized with respect to the outbursts of star formation
which seem so frequent in the nuclei and outskirts of many other strong
radio sources. At the panel discussion of the 1992 Tetons Meeting, I concen-
trated my discussion on the stellar content of the distant red radio galaxy
MG 1054+073. This object is one of the optically reddest radio galaxies
~nown at high redshift, with z = 0.967 (Spinrad & Dickinson, unpublished
data). The redshift is well determined from a moderately-strong [0 II]
emission line, the 4000 A discontinuity, and another break at rest wavelength
2900 A (see Figure 1). Still, at r = 22, the galaxy is faint enough that long
spectroscopic integrations with both the Kitt Peak 4-m telescope and the
Shane 3-m reflector were needed to produce the composite spectrum shown.
The signal-to-noise ratio at A S; 6000 A is still fairly poor.
Besides MG 1054+073, two other red radio galaxies are worthy of note
in our working hypothesis that these are old stellar systems seen at large
redshift. One is 3C 65 (Lilly & Longair 1984; Rigler et al. 1992) at z =
1.175 (very red in r - K color), and the second is another MG galaxy,
MG 0139+1712 at z = 0.758.
Concentrating for the moment on MG 1054+073, my claim of an old-
age stellar popullation is fairly dependent on the recognition of the short-
wavelength continuum discontinuity seen in Galactic late F, G, and K
stars at Ao = 2900 A (Morton et al. 1977; Fanelli et al. 1992). Figure
1 shows the spectral discontinuity to be present in our galaxy spectrum,

151
J. M. ShuLL and H. A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, 151-154.
© 1993 Kluwer Academic Publishers.
152

15 MG 1054+073
z = 0.967
~ 10
3,

5000 6000 7000 8000

A (1)

20
MG 1054+073
Flux averaged in
M=20aA bins

22

24

5000 6000 7000 8000

A (1)

Fig. 1. Spectrum and spectral energy distribution of MG 1054+073, a z = 0.97

galaxy from our MG survey and a good example of the minority class of very-red,
z ~ 1 radio galaxies that we are occasionally finding. We plan to model the continua
of these galaxies in detail, using the overall shape as well as detailed features such
as the A2900 A discontinuity, to constrain the age of the last epoch of star formation
in these relatively quiescent galaxies.
 153

TABLE I
Turnoff Ages and the Epoch of Last Star Formation l

Time Since Last SF Total Age ZJ

(Gyr) (Gyr)

1.0 10.44 1.2

2.0 11.44 1.5
3.0 12.44 1.9
4.0 13.44 2.4
5.0 14.44 3.4
6.0 15.44 4.8

1 (Assuming Ho = 50, 110 = 0.1, 1"g = 9.44 Gyr)

but its amplitude is not well determined. That is unfortunate, since the
..\2900 discontinuity is very heavily weighted by the contributions of G
dwarfs and subgiants in an old population - the break is small in F stars
observed by the International Ultraviolet Ezplorer satellite. Based only upon
a detection of this discontinuity, we would conclude that GO V - G2 V
stars might be the hottest objects contributing to the composite near- UV
spectrum of MG 1054+073. That conclusion would be important, because
evolutionary isochrones with main-sequence turnoff points at GO V [or
(B - V)o ~ 0.60 - 0.63] imply ages of nearly 8 Gyr (vanden Bergh 1983). Of
course, my estimate, which is based only on main-sequence stars, is slightly
too simplistic; sub-giant G IV stars and even early-K giants contribute
something to the galaxy composite spectrum just longward of 2900 A and
the the 8 Gyr noted above is a likely upper bound to the age of this stellar
system.
If we also use the deeper- UV portion of the MG 1054+073 spectrum (that
below 2800 A), we note the spectral gradient to flatten; it is flatter than seen
in the spectra of GO V stars. So, we appear to have slightly contradictory
evidence with the 2900 A discontinuity and the overall spectral energy
distribution at shorter wavelengths. The age of our population is probably
more weakly determined than I suggested verbally at the Panel Discussion.
The postscript on this interpretation will soon be available; Dr. Stephane
Charlot is now employing the new Charlot - Bruzual evolutionary models
in their entirety to describe the theoretical spectrum of a metal-rich system
some N Gyr after a starburst. Initially, it seems that the proper age will be
somewhat in excess of 1 Gyr. So, let us contemplate in Tables 1 and 2 the
last epoch of significant star formation, arbitrarily made coincident with the
154

TABLE II
Turnoff Ages and the Epoch of Last Star Formation 1

Time Since Last SF Total Age ZJ

(Gyr) (Gyr)

1.0 9.31 1.30

2.0 10.31 1.85
3.0 11.31 2.9
4.0 12.31 5.9
5.0 13.31 00

1 (Assuming Ho = 50, no = 1.0, Tg = 8.31 Gyr)

dynamical collapse of a giant E galaxy in one conventional wisdom. Recall

that the look-back time to MG 1054+073 at z = 0.97 is 9.44 Gyr with
Ho = 50 and no = 0.1 (and 8.31 Gyr for no = 1). For historical reasons, we
denote the redshift of collapse by Z J.
So, we note again that some red galaxies apparently completed all their
star formation very early in their life, at correspondingly large redshifts. Are
these ideas compatible with the relatively late acumulation of dwarf galaxies
or other entities (with gas) that may eventually accumulate to become a gE
system? These conjectures are just that. Still, they make for pleasant and
controversial conversation at our meetings!

References
Fanelli, M. N., O'Connell, R. W., Burstein, D., & Wu, C.-C. 1992, ApJS, 82, 197
Lilly, S., & Longair, M. 1984
McCarthy, P. 1990
Morton, D. C., et al. 1977
Rigler, M. A., Lilly, S. J., Stockton, A., Hammer, F., & Le, O. 1992, ApJ, 385, 61
Spinrad, H. 1986
vanden Bergh, D. A. 1983, ApJS, 51, 29
 PRIMEVAL GALAXIES, THE IGM, AND THE
QSO-PROTOGALAXY CONNECTION
TIMOTHY M. HECKMAN
Rowland Department of Physics & Astronomy, The fohns Hopkins University;
and Space Telescope Science Institute

Abstract. The other members of this panel are far more knowledgable than I about
galaxy evolution and the nature of the faint galaxy population. Thus, rather than address
the main panel discussion topics head-on, I will instead briefly summarize some of the
things we may be able to learn about the universe at high redshift from the study of
two classes of objects that I know something about: AGNs at high redshift and starburst
galaxies at low redshift. I will organize my remarks around several of the questions the
panel has been asked to address.

1. What is a "primeval galaxy"?

It seems physically reasonable to posit that the maximum rate of star-
formation possible in a self-gravitating system is that corresponding to
turning all the gas in the system into stars within one dynamical time (e.g.,
a crossing time or freefall time) for the system. An upper bound on the
gas mass will be the total baryonic mass. It may also be reasonable to
call a "primeval galaxy" any galaxy meeting this benchmark. It certainly
corresponds to the old Eggen, Lynden-Bell, & Sandage idea about spheroid
formation.
Using the normal scaling laws for galaxies (e.g., taking the Tully-Fisher
relation and a flat rotation curve and assuming that the baryonic mass is
proportional to the total mass within the visible part of the galaxy), I find
that the maximum star-formation rate will then be about 1000 M0 yr- 1 for
an L* galaxy, and will scale as mass to the 3/4 power. A star-formation rate
of 1000 M0 yr- 1 corresponds to a bolometric luminosity of about 3 X 10 12
L0 (assuming a normal Salpeter IMF).
The moral of this is that the ultraluminous IRAS galaxies may therefore
be forming stars at close to the maximum rate possible, and at least in this
regard are reasonable local anologs to primeval galaxies at high redshift.

2. Whence the faint blue galaxies?

The dilemma posed by the existence of a plethora of faint, blue galaxies
at relatively moderate redshift ("Smurfs" according to Professor Kennicutt)
has been discussed at length by my fellow panel members. There are two
properties of starburst galaxies that may help alleviate some of the difficul-
ties with explaining where these faint blue galaxies have gone at z = O.
The first is the possibility that starbursts may form only massive stars
(Le. have "top-heavy" IMFs). To give a specific example, a burst that

155
J. M. Shull and H. A. Thronson. Jr. (eds.). The Environment and Evolution of Galaxies. 155-158.
© 1993 Kluwer Academic Publishers.
156

produced only stars more massive than 3 M0 (e.g. stars earlier than AO
on the main sequence) would completely vanish after only about 400 million
years (a look-back time corresponding to only redshift z = 0.04hlOO)!
The second is the capability of starbursts to drive galactic winds (as we
have discussed in detail in a review article elsewhere in this volume). Suppose
we had a (proto )galaxy in which all the baryons were initially in the gas
phase. Turning only 1 % of this gas into stars with a normal IMF would return
5 X 10 13 ergs in kinetic energy (from stellar winds and supernovae) per gram
of remaining gas. This value for the specific kinetic energy corresponds to a
velocity of 100 km s-l, which is at least comparable to the escape velocity for
a dwarf galaxy. Thus, in principle, a burst involving only 1 % of the baryons
could blow all the other baryons out of the galaxy. Forming stars with a
top-heavy IMF is even more efficient in turning mass into kinetic energy.
On the other hand, radiative losses will likely be important, so that not
all the kinetic energy supplied by the starburst will be converted into bulk
motion of the galaxy's ISM. If this mass were ejected impulsively and if more
than half the total mass were ejected, the galaxy would become unbound and
"dissolve". This is rather unlikely, but even if the galaxy were not destroyed
by the mass loss, the mass ejection of the ISM could effectively shut-off all
further star-formation, thereby allowing the galaxy to rapidly fade.

3. How was the inter-galactic InediuIn ionized?

The lack of a clear Gunn-Peterson trough in the spectra of high-z QSOs

means that the IGM was probably ionized already at the time the first
known QSOs appear. The source of ionization is by no means clear.
One possibility is photoionization of the IGM by massive stars in young
galaxies. If present-day starbursts are any guide, this idea suffers from the
difficulty in getting ionizing photons out of the starburst. Even longward
of the Lyman edge (where the opacity of the ISM is far smaller), and even
in a sample of starbursts observed with IUE (and hence a sample biased
in favor ofUV brightness), only about 10% of the far-UV photons (1100 to
1900 A) typically escape the starburst. Leitherer, Robert, and I find that the
escaping UV fraction does correlate inversely with metallicity (as expected if
the UV is being absorbed by dust). Perhaps a population of low-metallicity
starbursting dwarfs at high redshifts could photoionize the IGM if big holes
in their H I envelopes were blown by superwinds.
Starburst galaxies could also collisionally ionize the IGM via superwinds.
I refer readers to my review article elsewhere in this volume for further
discussion of this idea and references to papers which investigate this idea
is some detail.
 157

4. Where are the high-redshift primeval galaxies?

I have already alluded to the importance of dust in snuffing out UV con-

tinuum radiation from starbursts. The effect of the starburst's ISM on the
Lyman Alpha emission line can be even more dramatic. Thus, to the extent
that the ultraluminous IRAS galaxies (or even the more mundane sample
of star-forming galaxies observed with IUE) provide clues about actively-
star-forming galaxies in the early universe, it may not be surprising that
intensive optical searches have turned up precious few candidate "primeval
galaxies" .
I will venture two informal opinions. The first is that the single best
candidate for a true high-z "primeval galaxy" (in the sense of bullet #1
above) is the object IRAS F10214+4724, which emits nearly all of its
radiation in the far-IR and exhibits a negligible amount of UV continuum
or Lya emission. The second is that, based on my perusal of the impressive
IUE galaxy spectral atlas compiled by Kinney and colleagues, I surmise that
most (if not all) of the objects discovered at high-redshift by virtue of their
Lya emission will turn out to be AGNs rather than simple star-forming
galaxies.

5. What is the QSO-protogalaxy connection?

The strong evolution with cosmic time of the QSO / AGN population is one
of the most remarkable phenomena in all of astronomy (as has been ably
reviewed at this Summer School by Boyle). It is hard to resist speculating
that this evolution is connected in some way to galaxy formation or evolution
(especially since we believe that the AGN phenomenon at low redshifts is
probably triggered, or at least regulated, by the extra-nuclear environment).
It is also interesting and suggestive that the following fundamental timescales
are all similar (to within an order-of-magnitude) at the height of the AGN
Epoch at z = 3:

• The Eddington time for black hole growth (for reasonable radiant
efficiencies)

• The orbital/dynamical time in galaxy halos

• The e-folding time for the evolution of the QSO population

• The age of the universe

158

Observations of AGNs at high redshifts have by no means led to a

definitive picture of the causes or consequences of the AGN Epoch nor
its relation to galaxy formation/evolution. Nevertheless, some preliminary
results are:

• The properties of most high-z radio galaxies (and by extension,

the hosts ofradio-Ioud QSOs) are probably more consistent with the
rejuvenation of an existing (massive) galaxy than with the on-going
formation of a galaxy. There are some plausible counterexamples
though of radio galaxies that may be true protogalaxies undergoing
their initial burst of star formation. While the situation is evidently
more complicated than it appeared a few years ago, it still seems
very likely that powerful radio sources at high red shift are capable
of triggering titanic galaxy-wide starbursts in their wake.

• The systematic changes with redshift in the properties of both the

extended radio sources and gaseous emission-line nebulae associated
with radio-loud QSOs together strongly suggest that such AGNs
lived in galaxy-scale regions of substantially higher gas density and
pressure at high redshifts than they do today. It is tempting to link
this evolution in the QSO environment to the evolution of the QSO
populatio,n with redshift.

• Most of what we know about the environments of AGNs at high

redshift has come from studies of the most radio-loud subset of the
AGN population (I estimate that for every 4C-Ievel radio-loud QSO
at z = 2 there are about 1000 less-radio-bright QSOs with similar
absolute magnitudes at the same redshift). Preliminary work by my
colleagues and 1 indicates that the host galaxies of radio-quiet QSOs
at high-redshift are considerably less spectacular than radio galaxies
or hosts ofradio-Ioud QSOs (e.g., at least two magnitudes fainter at
K). 1 therefore worry about how generally applicable the spectacular
data on high-z radio-loud AGNs are to the AGN phenomenon as a
whole.
THE EXTRAGALACTIC X-RAY BACKGROUND (1 - 20 keY)

RICHARD E. GRIFFITHS
Dept. of Physics and Astronomy, Johns Hopkins University

Abstract. The extragalactic X-ray Backround (1 - 20 keY) comes from discrete sources,
the composition of which changes with energy. In the low energy range (1 - 3 ke V) where
the first generation X-ray telescopes have operated, at least half of the XRB arises in
Active Galactic Nuclei, with important contributions from X-ray luminous galaxies such
as star burst galaxies and early-type galaxies, with small contributions from clusters of
galaxies. Above 3 keY, the composition is less well understood because X-ray telescopes
are only just beginning to operate in that range, but the candidates for the bulk of the
XRB are again AGN (via Compton processes) and starburst galaxies.

1. Introduction
Solar X-rays were discovered in a simple rocket experiment in 1948, and
fourteen years later the first night-time rocket flight carrying X-ray sensitive
equipment was made by a group from American Science and Engineering
(Giacconi, Gursky, Paolini, & Rossi 1962). The "diffuse X-ray background"
was thus discovered a few years before the microwave background, and at
about the same time as the discovery of the first quasars~ in an era when
controversy still reigned over whether the universe "originated" in a Big
Bang or was in Steady State production. Because the microwave background
was immediately recognised to be of cosmological significance, there was a
tendency to associate the X-ray background with a cosmological source also,
rather than with objects in the relatively local universe.
Although there had been hints of extra-solar X-ray sources in unpublished
solar rocket-flight data, there had been no predictions of ubiquitous X-ray
emission in the universe, and the discovery came as a major surprise - see
Giacconi (1974) and Friedman (1990) for some insights into the events lead-
ing to the discovery. The initial rocket and satellite experiments consisted of
crude mechanical collimators in front of geiger counters or proportional gas
detectors, and they had angular resolutions of order several degrees. If the
atmosphere were completely opaque at optical wavelengths, one can only
surmise as to what might have been the conclusions from night-time rocket
experiments in the optical with similar angular resolution!
The early rocket experiments with their crude collimators detected the
brightest Galactic X-ray binaries: the unresolved X-ray counts were at-
tributed to an apparently "diffuse" background, and the first "bandwagon"
effect was the association of the "diffuse" emission with a hot, intergalactic
medium (IGM), a hypothesis which retained some popularity through the
late 1980's (Guilbert & Fabian 1986). For a full quarter-century, the idea of
a hot IGM thus persisted, fuelled by the apparent goodness of fit between a
bremsstrahlung X-ray spectrum and the measurements made over a limited

159
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution oj Galaxies, 159-190.
© 1993 Kluwer Academic Publishers.
160

energy range with proportional counters with low energy resolution. The
hot IGM model for the bulk of the XRB did not fall into general disfavor
until shortly after the launch ofthe Cosmic Background Explorer (COBE)in
1990; the limits on a hot IGM from COBE are far more stringent than those
from X-ray experiments themselves. The hot IGM model was thus eventually
dismissed by observations at microwave frequency with the COBE satellite
(see Mather et al. 1990; and Bennett et al., this volume).
The first-generation satellite experiments in cosmic X-ray astronomy (in-
cluding OSO-5, UHURU, Ariel V, HEAO-1) all carried mechanically colli-
mated, gas-filled proportional detectors which operated from'" 2 to '" 20-40
ke V. These instruments placed limits on the anisotropy of the "diffuse" emis-
sion at the level of a few percent and also catalogued hundreds of Galactic
and extragalactic sources. They did not make significant progress towards
an explanation for the great bulk of the detected X-rays, those which con-
stituted the. unresolved or "diffuse" flux. These early experiments detected
the brightest Seyferts and quasars (or at least 3C273), and it was pointed
out (Setti & Woltjer 1979) that AGN could be a major contributor to the
XRB.
Over the past decade, since the launch of the Einstein observatory in
1978, we have seen the real beginnings of extragalactic X-ray astronomy, i.e.
the study of galaxies, clusters of galaxies, and AGN at X-ray wavelengths:
all such objects contribute to what was previously known as the "diffuse
X-ray Background". It has become clear that the unravelling of the X-ray
Background can only be achieved through studies of these individual classes
of X-ray sources, by measuring their spectra, luminosities and space densities
(luminosity functions) and understanding their evolution, including possible
spectral evolution, to moderately large look-back times (i.e., to at least
z = 3).
Thirty years after the discovery of the XRB, the deepest surveys with
the ROSAT telescope have resolved as much as 60% of the XRB at '" 1
ke V into discrete sources, and optical identification programs have shown
that at least half of these sources are Active Galacic Nuclei (AGN). It is
still true today, however, that above 3 keY, where most of the XRB energy
lies, the "background" is unresolved into point sources, and may still de-
scribed as "diffuse" in that sense. The first x-ray telescope working above
3 keY has been the B:['oad-Band X-ray Telescope (BBXRT) developed by
Serlemitsos and colleagues at the Goddard Space Flight Center and flown
on a week-long Shuttle mission (Serlemitsos et al. 1992, also see Mushotzky,
these proceedings). Thi~__situation will change radically with the launch of
the Japanese satellite Astro-D (ASCA) in 1993, which includes BBXRT-type
mirrors, and even more so later in the decade with the launch of JET -X (part
of SPECTRUM-X), NASA's Advanced X-ray Astrophysics Facility (AXAF),
and ESA's X-ray Multi-Mirror Mission (XMM), each comprising telescopes
 161

which have good efficiencies up to energies of'" 8 - 10 keY.

It has been a common and fatal mistake to try to account for the whole
of the XRB spectrum from 0.1 ke V to at least 40 keV by invoking a single
type of source; papers published as late as 1990 were still attempting this.
As pointed out by Fabian (1988), this is like trying to account for all of the
extragalactic radiations from 1 JLm to the UV using a single source model.
The "X-ray background" is very complex - at any given energy above", 1
ke V, it is the superposition of emission from different classes of extragalactic
sources, and at energies below", 0.3 ke V it is dominated by diffuse Galactic
radiation from local hot gas (e.g., Wang & McCray 1993). Between 0.3 and 1
ke V the background is partly accounted for by extragalactic discrete sources
(about half) and partly by "diffuse" emission from a thermal component
at about 2 X 106 K, which may be either Galactic or extragalactic - if
extragalactic, it may be truly diffuse or consist of discrete sources, either
predominantly point-like or extended.
At each frequency, the sources have to satisfy the total energy content
and the isotropy. An integrated background model covering even a partial
decade of energy has to satisfy the overall spectral constraints. Owing to
the nature of X-ray astronomical technology, two of the most important
energy ranges have been: (i) the low-energy « 3 keY) range accessible to
the first generation oflarge X-ray telescopes (Einstein, ROSAT) and (ii) the
medium-energy ('" 3 4;0 '" 40 ke V) range accessible to proportional detectors
of a few degrees, with fields of view limited only by mechanical collimators.
The extragalactic component of the low-energy range, at least above 0.5
keY, is observed to be dominated by AGN, which probably cannot account
for such a large fraction of the flux at medium energy, unless there is a
population of "hard AGN". There is probably an overlap region in which
the AGN contribution is falling (above 2 ke V) and a second (or multiple)
component becomes increasingly important.
As X-ray astronomy has developed, then, classes of extragalactic sources
have been identified and their properties measured. One of the most impor-
tant source characteristics, as far as contributions to the XRB are concerned,
is the X-ray spectrum. Starting with the quasars and Seyferts, it was realised
that the individual spectra were softer than that of the (3 - 40 keY) XRB,
and that subtraction of their contribution to the XRB resulted in a residual
XRB spectrum that was even harder and more difficult to explain. As fur-
ther source classes have been identified, the tendency has been to find more
classes which have spectra softer than the XRB (especially with X-ray tele-
scopes working in the soft X-ray range!), thus exacerbating the problem even
further (Tanaka 1992). The residual sources must be Comptonised, heavily
absorbed at low energy, or both. The proposed classes of objects that might
make up this residual include the "bard AGN", starburst galaxies or nuclei,
or hybrids. At higher energies still, it seems likely that the background may
162

be partly explained in terms of the high-energy extensions of sources con-

tributing to the medium-energy range, especially AGN, but supernovae (in
starburst galaxies?) and radio-loud AGN may be important.
There have been two recent conferences on the subject of the extragalac-
tic X-ray background: the first at Laredo in September 1991 (eds. Barcons &
Fabian 1992) and the second, concentrating on the contribution from AGN,
at the Max Planck Institut in Garching in November 1991 (eds. Brinkmann
& Triimper 1992). The conference proceedings from these meetings give de-
tailed accounts of recent work on the X-ray background. In addition, Fabian
& Barcons (1992) have recently provided an ~xcellent and comprehensive
review. The present review will therefore attempt to put more emphasis
on the optical identification content of X-ray surveys, especially the deep
telescope surveys which try to resolve as much as possible of the XRB into
discrete sources. As we shall see, this technique has worked well over the
limited energy range in which the telescopes have good reflectivity, and may
be providing clues on the discrete sources contributing to the XRB at higher
energies.
In §2, we review the overall properties of the XRB, i.e., the energy con-
tent, the spectrum, and the isotropy. The contributions from discrete sources
are introduced in §3, with summaries of the work with the Einstein Observa-
tory and ROSAT in §4 and §5 respectively. The contributions from starburst
and early-type galaxies are reviewed in §6 and §7, with speculation on "X-ray
unique" AGN in §8 and conclusions in §9.

2. Overall Properties

2.1. ENERGY CONTENT

The overall XRB spectral energy distribution spans several decades of en-
ergy, and can conveniently be divided up into several energy bands: the soft
X-ray range below 1 keY, the Low Energy band (1 - 3 keY), the Medium
Energy range (3 - 40) keY, hard X-ray (40 - 200 keY) and the gamma-ray
region at higher energies.
The energy content of the LEXRB and MEXRB is at the level of 10- 4 -
10- 5 of the energy content of the microwave background - see Figure 1,
from Fabian & Barcons (1992). One might think that such a low power
density could easily be explained, but thirty years after the discovery, the
problem is by no means solved, primarily because of the technical difficulties
of focussing X-rays at the medium and high energies. Most ofthe energy is in
the ME range ('" 50% of it is between 30 and 80 ke V) where telescopes have
not yet reached - the LE contains only 10% of the total energy density from
0.1 to 100 keY. In finding sources for the origin of the LE, one therefore has
to be careful in making extrapolations into the ME; this applies especially to
 163

to lOCIII 1(11 1_ 100JI .,., 111 10001 100A 10A 1A 0.11

...
I
0
1fN lOfN lOOtV 1kfN 10kfN 100kfN 1MI\I 10 row.I
C)
I
....
0

OpUcal-UV

-...
f'o
I
_ 0
...
•
I
1/1 C)
I
"t ~
e
-- ...
~~

:; 2
~
o

XRB

...
I
o

-.....
I
o
•
•
-
C\I
I
~ 10 10 10" 1012101310141016 10 1e 10 17 10 18 10 18 1020 1021 1022
" (Hz)

Fig. 1. The energy density in the extragalactic radiations, from Fabian & Barcons
(1992), showing the relative energy in the XRB VI . the MWB).

known types of AGN. The fact that as much as 50% of the XRB at 1 keY may
originate in AGN does not mean that one can account for a similar fraction
of the XRB at 6 keY, 20 keY, or 100 keY. Indeed, the observed spectra of
AGN imply much smaller XRB contributions at the higher energies, and
"different" AGN are needed.
The energy content of the MEXRB has led to difficulties for some source
models, in the sense of both over-production or, more generally, under-
production. Just after the discovery of the XRB, Hoyle (1963) proposed
an origin from neutron decay in his hot Steady-State cosmological theory,
but this was shown to produce too much X-ray emission (Gould & Burbidge
1963). The origin of the XRB as bremsstrahlung from hot, intergalactic gas
was put forward by Cowsik & Kobetich (1972). The temperature ofthe IGM
needed is ,. . ., 40( 1 + z) ke V, where z is the redshift at which the bulk of the
gas is reheated after the initial cooling and production of the microwave
background. As explained by Field & Perrenod (1977), the required energy
density in the hot gas was a serious problem, leading to general unpopularity
with that model long before COBE results were known. Further, Guilbert
164

& Fabian (1986) showed much later that the baryonic density required in
the hot gas needed to be at least 23% of the critical density, four times that
required by nucleosynthesis. Further major problems appeared early for the
hot IGM - the apparently good fit to a bremmstrahlung spectrum was de-
stroyed by subtracting an AGN component (Leiter & Boldt 1982; Boldt &
Leiter 1987; Giacconi & Zamorani 1987; Fabian 1988).
The energy problem for compact sources can similarly be used to elim-
inate some models, and is less severe a problem for others. The energy ar-
guments cannot be taken alone and need to be carefully evaluated, but
they have some usefulness. Cowie (1989) showed that hot gas in the halos
of galaxies (arising via supernova winds) cannot contain sufficient energy
to power the XRB, given the observed energy in the optical emission from
galaxies, because the model predicts that Ex / Eopt ~ 1%, compared with
an observational requirement of Ex / Eopt '" 2%, from the observed optical
energy density in deep galaxy images, and the argument that most of the
optical light produced is actually observed. Cowie also argued on the appar-
ent marginality of making up the energy content of the ME using massive
X-ray binaries in starburst galaxies, if one assumed that their optical ap-
pearance is similar to the faint blue galaxies seen in deep optical surveys.
The calculation is much more favorable, however, if one assumes massive
X-ray binaries like SMC X-lor those in the LMC, for which Lx / Lopt '" 103
(Griffiths & Padovani 1990). Deep surveys with ROSAT have shown exam-
ples of X-ray selected galaxies with Lx / Lopt '" 103 times higher than the
optically-selected objects observed with the Einstein observatory (see §5,
below).
Daly (1991) argued against starburst galaxies as a prime contributor
because of the number of neutron stars that would be left behind. Daly's
arguments with respect to starburst galaxies were based on the assumption
of large (L·) galaxies and were flawed for that reason; massive X-ray binaries
(if they are the dominant source type) do not have to be contained within
large galaxies. Even ifthey were, the number of neutron stars is not excessive;
the number left in our own galaxy may be as many as '" 109 , and if the
neutron star Geminga is typical, they could be very important in terms of
their hard X-ray emission. Both Cowie and Daly assumed that the compact
sources are in regions that are bright in the optical band (blue galaxies).
However, if the sources are in dusty regions, then the X-rays can escape
while the optical light is absorbed; there should then be a correlation with
infrared emission, as observed (Griffiths & Padovani 1990).
The first generation of X-ray telescopes (Einstein, ROSAT) were respon-
sive in the LE range, and have helped to solve the energy content in that
range only. Indeed, the bulk of discrete sources resolved by ROSAT in the
deep imaging surveys (see §5) have explained the excess in the overall spec-
trum below 3 ke V, i.e. the excess above the extrapolation of the kT = 40
 165

keY spectrum to low energy ("LE ezcess"). The AGN in these samples have
a great variation in X-ray spectra, however, both in power law indices and in
absorbing column density (large variations are also seen in optically-selected
quasars - Wilkes & Elvis 1987). Nevertheless, at an effective energy of 1 ke V,
the deep surveys with ROSAT have shown that,....., 50% of the XRB at 1 ke V
comes from AGN (Boyle et al. 1993; Della Ceca et al. 1993), with a residual
from more populous objects (several thousand per sq. deg.). Arguments of
energy content thus become rather superfluous for the AGN contribution to
the LEXRB, and a "hard AGN" population would similarly not have serious
energy problems in making up the MEXRB.

2.2. OVERALL SPECTRUM

The overall XRB spectrum is, of course, another representation of the energy
plot in Figure 1, but it is more convenient in some respects to present specific
energy fluxes (keV cm- 2 s-l keV- 1 ) and to compare familiar spectral indices
of the various components such as AGN with that of the LE and MEXRB
spectrum.
In early studies of the LEXRB, it was initially found with rocket flights
and later confirmed with the Einstein observatory (Garmire et al. 1992) that
there is an excess over the spectral fits to the MEXRB; for example, excess
amounts to about 100% at 1 keY (see review by McCammon & Sanders
1990). This upturn in the spectrum below 3 keY has been more carefully
quantified by Hasinger et al. (1992) , who have used the ROSAT deep sur-
vers to confirm the earlier suspicions of an LEXRB excess, and who measure
an energy index of -1.0 over the range 0.5 to 2 keY. The X-ray telescopes,
if they resolve, say, half of the XRB at 1 - 2 keY, may therefore be largely
investigating the low-energy excess, and not necessarily addressing the bulk
of the low-energy extrapolation of the MEXRB spectrum (which could re-
main unresolved in all but the deepest telescope images). Since the MEXRB
is where most of the energy lies, this is a serious drawback to,the use of the
telescope results in interpreting the overall XRB. The BBXRT · is the first
X-ray telescope to span the energy range between the LEXRB and MEXRB
regions (BBXRT operates from'" 0.5 ke V to 8 ke V, using solid-state de-
tectors with energy resolution of", 15% at 1 keY and", 3% at 6 keY): the
spectrum is apparently well fitted by the 40 ke V shape from 8 ke V down to
'" 1 ke V, but the amount of data collected was insufficient to constrain the
previous measurements with any confidence (Jahoda et al. 1992).
Below", 1 ke V, and especially below 0.5 ke V, the XRB is highly anisotropic
and most of it probably originates locally within the Galaxy - see McCam-
mon & Sanders (1990) for a review of the soft X-ray background measure-
ments. Wang & McCray (1993) have shown, using ROSAT data, that the
soft diffuse component of the LEXRB below 1 keV can be represented by a
166

jl"1

-...
.:;
~ '"
>.,
_"'!
10-'
.,"
"i '" ,1 """"""
E
>.,"
.=.
10 .•
x: Marshall eta!. 1980 \~ij ifi
\'" ' f
,4
0 : Rothchild eta!. 1984
?;
'iii . : Kinzer etal. 1978 ~ l
c
.:; 10- 3 0 : Trombka eta!. 1977 \ ' If i
C
. : F'ichlel eta!. 1981 \ ' 1
T upper limits (rom Fab ian' etal. 198'9 ' , , , ~
10'" Solid line: Log Normal, 0 2 0 ,85. b-0.25. E,-80 tv -:!
Dotted line: PL+TB. p-0 .9. p=0.47. kT=90keV I 5
Dashed lines: Power Low . a =0.7. E,-200keV ~

10-'
10- ' 10' 10' 103 10'
Pholon Energy (keV)

Fig. 2. The spectrum of the X-ray Background, from Schwartz (1992)

two-component model, with the flux below 0.3 keY dominated by a 106 K
plasma in the Local Bubble, and about 40% of the XRB in the "M-band"
(0.5 - 0.9 keY) arising from a diffuse component with a temperature of
"" 2.2 X 106 K (the remaining M-band flux comes from discrete sources - see
sections below). Present data do not demonstrate conclusively whether the
hotter diffuse component is of Galactic or extragalactic origin. The bulk of
the emission below 0.3 ke V, arising in the local interstellar medium, must
be largely in the form of collisionally-excited emission lines of the heavy ele-
ments, such as 0 VII and 0 VIII. The investigation of this line emission, the
resulting plasma diagnostics, and the spatial structures that might be im-=
aged in these monochromatic lines are untouched observational areas which
await future experimenters. Potentially, these studies could lead to a much
improved undestanding of the local ISM and its history.
The overall XRB spectrum has been best measured for the MEXRB, us-
ing proportional gas detectors with energy resolution (the FWHM/ energy,
for a monochromatic energy input) of about 40% at 1 ke V, 20% at 6 ke V,
and 12% at 20 keV. There have been no satellite experiments carrying col-
 167

limated detectors of higher energy resolution to cover the medium-energy

range; large-area solid-state detectors or CCDs may be feasible, but the
silicon detectors developed thus far n"eed to be cooled and have not yet
been flown (CUBIC will be the first - see Burrows et a1. 1992). For the
MEXRB and HEXRB, the spectra obtained 15 years ago by the HEAO-
1 spacecraft are still the most precise measurements made to date; Ginga
and EXOSAT results have generally been consistent without improving the
quality of the spectral fits. The HEAO-A2 instrument gave the apparent 40
ke V bremsstrahlung fit over the decade of energy from about 3 to about 40
keY (Marshall et a1. 1980), although the spectrum did not fit below 3 keY,
or above 40 ke V:
dJ ( E ) -0.29 "
dE ~ (5.6) 3 keY exp( -E/kT)keV keV- 1 cm- 2 s-1 , (1)

where kT = 40 keY.
With detectors of higher energy resolution, it would be surprising if the
40 ke V spectrum were found to fit quite so well, or over such a broad range of
energy. Power laws (J(E) ex: E-O<) can also be made to fit over more limited
spectral regions (0: '" 1.4 from 2 to 20 ke V).
It has long been recognised that, having established a source component
of the XRB in a given range, the subtraction of the integrated spectra from
such sources will generally change the overall XRB spectrum, usually making
the residual XRB "harder". This has been apparent at least since 1978,
since the same HEAO-1 detectors which measured the MEXRB spectrum
were also used for measurement of the spectra of Seyfert 1 galaxies. The
Sy1 's were estimated to contribute about 20% of the XRB at a few ke V, and
subtraction of this component immediately destroyed the 40 ke V shape. This
was at once clear to the HEAO-l investigators but seems to have been lost
on much of the general astronomical community, some of whom persisted in
looking for explanations of the "40 ke V bremsstrahlung" spe'c trum, partly
because of the convenience of taking such a "perfect" observed quantity and
looking for theoretical interpretations. In reality, the XRB spectrum is the
superposition of several components, and is much harder to interpret.
The Seyferts and other AGN have negative spectral indices of 0.7 or
larger (2 - 20 keY), some with low-energy excesses. It is therefore no sur-
prise that thay can account for the excess in the LEXRB flux, over and
above the extrapolation of the 40 keY shape. Subtracting off the (0: = 0.7)
AGN component, one is left with a residual MEXRB spectrum of energy
index 0.2 or less, i.e. almost flat (cf. Fabian 1988). The finding or hypothesis
of other components with moderate or "soft" indices (e.g., 0: = 0.5 - 1.0)
further "hardens" the remaining XRB and serves to exacerbate an already
awkward problem. One is driven to look for sources which are heavily cut
off at low energy, and which have rising or flat spectra in the lower part
168

of the medium range. Heavily absorbed AGN are possible contributors, as

are heavily absorbed starburst or nuclear regions of galaxies. Some X-ray
binaries also have extremely hard spectra, arising from accretion onto neu-
tron stars. In any case, the hard component almost certainly arises from
Compton scattering, in plasmas formed by accretion onto compact objects,
whether supermassive or of stellar mass. In the case of X-ray binaries, the
hard spectra may arise from Compton up-scattering of the accretion-disk
photons on relativistic electrons. In the case of AGN, the hard spectra may
arise from Compton reflection of the power-law X-ray spectra on cold ma-
terial. Both processes may occur in either physical case, of course, and it is
not clear which one dominates in either case.
Above 40 ke V, and up to "'" 1 Me V, the overall spectrum can be described
by a power law of energy index"", 1.6. The overall XRB spectrum can then be
equally well described as a power law from 1 to 100 keY with a superposed
"hump" around 10 - 30 ke V, where the underlying power law could be
ascribed to AGN at high energy, and '" half to AGN at the LE end; the
"hump" is then possibly also attributable to another AGN component, viz.
a soft thermal (opticaJ- UV) flux Comptonized by mildly relativistic electrons
(see below) but could also come from other Comptonized sources such as
accreting neutron stars or ('" stellar-mass) black holes in binary systems in
starburst or post-starburst galaxies. The spectral "break" at about 40 ke V is
especially important for future spectral measurements, since it can be used
to constrain the AGN (or other) models (Zdziarski 1993a).

2.3. ISOTROPY

The second fundamental property of the XRB is the isotropy, which was
used very early on to claim that the bulk of the XRB was extragalactic and
even "cosmological". The measured isotropy of the XRB at all energies is
a powerful tool which can be used to constrain models of its origin. If we
propose, for example, that all of the MEXRB comes from AGN, then the
measured (an)isotropy should probably be consistent with the corresponding
(an)isotropy in the number counts of AGN determined at other wavelengths,
i.e consistent with the observed clustering of AGN (unless we propose a
new, ad hoc type of AGN which is visible at medium X-ray energies only).
Measured anisotropy in the LE and ME ranges can thus be be used: (i) to
put limits on the XRB contributions from certain source classes; and (ii) to
search for correlations with features observed at other wavelengths, such as
large-scale structure inferred from, say, optical galaxy counts, and thus to
infer contributions to the XRB.
Isotropy can be measured: (i) on the large angular scales afforded by the
all-sky survey collimated detectors (several sq. degrees), at all X-ray energies;
and (ii) small (arcminute) scales afforded by X-ray telescopes (low energy
 169

only, before ASTRO-D, JET-X, and later missions). These different scales
are independently powerful, and their combination even more so. ROSAT
has performed the first all-sky survey with a focussing X-ray telescope, and
can therefore be used to give results on the isotropy at both small and large
scales (see Snowden et al. 1992). The fluctuation analysis is often called a
P(D) analysis, in reference to the work done in radio astronomy in the 1950's
- P(D) was the radio astronomical term for the probability of obtaining a
given "deflection" or amplitude (of signal measured with a galvanometer).
The measurements of large-scale and small-scale fluctuations are considered
separately below.
Large-Scale: The HEAO-A2 instrument has been used to measure anisotropy
at the level of rv 0.5% on scales of 10 X 10 sq. degrees (Jahoda et al. 1992
and Mushotzky, this volume). Large-scale features are seen which coincide,
in some cases, with superclusters of galaxies and voids observed in the large-
scale optical surveys carried out over the past decade. (In other cases, the
X-ray results predict the presence of such structures). These large-scale fea-
tures are comparable in amplitude to the dipole anisotropy predicted from
the motion of the earth with respect to the very high-redshift universe (or
microwave background). The measured X-ray dipole (Warwick et al. 1980) is
roughly consistent with the microwave dipole, but has large uncertainties in
amplitude and direction, largely because of the effects of large-scale galaxy
structures. The observed signal from large-scale structure provides an im-
portant clue to the origin of the MEXRB; the observed signal from galaxies
in the local universe shows that there must be a significant contribution
from X-ra~ emitting galaxies distributed like those locally, whether they be
AGN or starburst or other X-ray galaxies. This minimum contribution has
been estimated to be about rv 13% (Lahav et al. 1992).
Small-Scale: On arc-minute angular scales, the data from the Einstein
and ROSAT telescopes have been used by several groups of investigators
to place constraints on source contributions and clustering, from the two-
point correlation coefficient, and from fluctuations in the residual counts
after individually-detected discrete sources have been removed from deep-
survey observations. The observed fluctuations in the Einstein deep surveys
(excess in the number of, say, 20- fluctuations above the mean background
per cell, with a cell size of about 1 arc minute) were used by Schaeffer &
Fabian (1983), Hamilton & Helfand (1987) and Barcons & Fabian (1990)
to show that the number counts (log(N) VS. 10g(S)) cannot continue to fol-
low the "Euclidean" (5-1.5) power law to X-ray fluxes much below those
corresponding to 40- source detections (the Einstein "Deep Survey Limit");
otherwise, the observed fluctuations would have been greater. This result
from the fluctuations analysis was consistent with the observation that the
individual source detections were dominated by AGN at typical magnitudes
of B rv 20 (Griffiths et al. 1983, 1992). The turnover in the corresponding
170

eso
ma

OIFFU~ .
COM~~E~~. of XRB (KT. 4OKeV)
,,
,,
., ....
~ ....
'1:J ~1.2 ~O"'SIaT
g-
... AGN? " ........

!. GR 88! GR 83
Uj
~ ! t GIACCONI ET AL 1979
Z 10'

UHURU
AAIELV
HEAQ-l

10·'· 10'"

S ergs em·z ., (0.3 - 3.5 KeY)

Fig. 3. The number-flux relation for X-ray sources, log(N) vs. log(S), adapted from
Giacconi & Zamorani (1987)

number counts of optically-selected AGN had already been established at

magnitudes fainter than B = 20. The turnover in the X-ray counts at the
faint end was confirmed by the ROSAT deep surveys (see below), with a
break in the log(N} /log( S} curve roughly at the Einstein Deep Survey limit
(2 x 10- 14 ergs cm- 2 s-t, 0.5 - 3.5 keY).
In turn, fluctuation analysis of the ROSAT deep-survey data has been
used to show that the unresolved sources are weakly clustered or unclustered
(Georgantopoulos et al. 1993), which may be consistent with a population
of "normal" or starburst galaxies, for example, but not with most classes
of known AGN such as Seyfert galaxies of types I or II, at least not those
selected at other wavelengths.
 171

o 5 10 15
8 (arcmin)

Fig. 4. The auto-correlation function of the 0.5 - 2 keV XRB (filled circles). The
solid triangle denotes the 10- upper limit of Carrera & Barcons (1992).

The fluctuation measurements thus provide extremely useful constraints

on the nature of the sources which make up the residual LEXRB, after
allowance for the contribution of "classical" AGN (50 - 70% of the XRB at
1 - 2 keY) (see Boyle et al. 1993). Whatever these sources are, however, they
do not necessarily contribute much to the MEXRB.

3. The All-sky Surveys ('" 2 - 40 ke V): Clues and Contributions

from Discrete Sources:

The first-generation all-sky surveys were carried out in the 1970's using satel-
lite experiments with collimated proportional detectors (including 050-5,
172

UHURU, Ariel V, HEAO-1), and similar experiments have been continued

in the 1980's with EXOSAT and Ginga. The X-ray counts from these instru-
ments were dominated by the "diffuse" emission; the resolved or detected
X-ray sources accounted for only a few % of the total observed emission
from the sky. These early experiments established the existence of several
classes of extragalactic X-ray sources (see e.g. Kellogg 1974) viz. groups and
clusters of galaxies, Seyfert galaxies, quasars, BL Lac objects, and narrow
emission-line galaxies. The major problem with each of these source classes
with respect to the origin of the ME XRB is that their spectra are softer
than that of the 40 keY XRB, and subtraction of an estimated XRB con-
tribution from anyone of them only makes the problem more intractable
for the others. This problem, especially with respect to the AGN, has be-
come known as the "X-ray spectral paradox". The paradox is simply that
the observed spectra of AGN have power laws of index 0.7, while the XRB
has a power law slope of 0.4 as measured by the same instrument over the
same energy range. Tanaka (1992) has presented spectra of AGN and radio
galaxies as observed by Ginga, and almost all are softer than the XRB (see
Figure 5). Before returning to AGN, the contributions from the other classes
should be briefly considered.
The discovery of X-ray emission from clusters of galaxies was one of the
highlights of the UHURU mission. Their measured temperatures are in the
range rv 1-10 keY, and their luminosity function and evolution has become
understood in the two decades since the UHURU satellite was operational. It
is now clear that clusters do not contribute to the MEXRB at more than the
'" 5% level, although their contribution to the LEXRB is probably at least
as great as '" 10 - 15%, especially if one includes hot gas retained in small
groups and the halos of ellipticals. Larger contributions (Schaeffer & Silk
1990) are probably ruled out by the observed negative evolution of clusters
in X-rays (Gioia et al. 1990; Edge et al. 1990), i.e. the observation that the
most luminous X-ray clusters are seen in the local universe and are seen in
decreasing numbers towards higher redshift. This is in exactly the opposite
sense to the evolution of AGN, which are observed in apparently increasing
numbers at high luminosity towards higher redshift (at least, up to redshifts
of 2 or 3). The negative evolution is reflected in the number counts at various
flux: levels. At the UHURU flux limit of 3 X 10- 11 ergs cm- 2 s-l (2 - 10)
keY, the X-ray counts consist of roughly half clusters and half AGN (Cooke
et al. 1978). In the Einstein Medium-Sensitivity Survey sample, however,
the fraction of clusters drops towards fainter flux levels, until it reaches the
level of about rv 10% in the Einstein Deep Surveys. The fraction of AGN
rises correspondingly, so that they dominate the Deep survey counts.
H AGN are to dominate the MEXRB, then their spectra must get harder
as the sources get fainter. There are at least three possibilities which would
produce this result: (a) the spectra evolve in the sense that more distant
 173

o QSO and Seyfert

• BL Lac
• Radio Galaxies
2.0
··· ..
•I
I
·.
1.5
•
I
1.0
+
I

f
I

It I
~
.I 4>
t
! ~ 1 i, f i +
__ ~ _~_ -: J ___ ~' __ ~t J __ tj _~L ____ _
I d>

t
0.5

~ Cosmic X-Ray Background I

0.0
0.001 0.01 0.1 10

Red Shift
Fig. 5. The x-ray spectral indices of AGN and other extragalactic sources VS . that
of the XRB (from Tanaka 1992)

AGN have harder X-ray spectra; or (b) AGN spectra are complex, and
gradually become harder at higher energy; or (c) there is a class of AGN,
yet unobserved at any wavelength, which have much harder spectra than
those observed so far. Along the lines of hypothesis (b), Schwartz & Tucker
(1987) proposed a universal power-law plus bremsstrahlung spectrum for
AGN which was consistent with all known spectral data on them, and which
resulted in the observed spectra appearing harder as the softer component
is redshifted out of the observed range, as a function of redshift. Hypothesis
(c) has recently gained great popularity (e.g., Fabian et al. (1990); Rogers
& Field (1991); and summary by Zdziarski et al. 1993a). In this hypothe-
sis, only "-' 10% of the observed X-rays above 3 keV are seen directly; the
174

bulk of the emission comes via reflection off cold material. One of the strong
attractions of this model is that a fundamental physical process, Compton
scattering, can be invoked to explain the downturn in the overall XRB spec-
trum around 40 keY (the X-ray spectral "break"). The class of sources most
likely to harbour such sources are the Seyfert 2 galaxies, which would have
to outnumber Seyfert 1 's at faint fluxes; although some Seyferts do show a
Compton-reflected component, they are in the minority, and Seyfert 2 spec-
tra taken as an ensemble do not resemble the MEXRB (Tanaka et al. 1992).
Further discussion of this point will be deferred to §8.

4. The Discrete Source Content of The Einstein Surveys

The free-standing collimated proportional counters with wide fields of view
can never be used alone to solve the XRB problem. They can be used to
gain information on the properties of certain, bright source classes., but their
comments on the source content of the XRB are .limited to extrapolations
from local luminosity functions, at best. Fluctuation analysis is useful, as
are correlations with catalogs of objects observed at other wavelengths, but
the results cannot be definitive with respect to the makeup of the XRB as
a whole. The ways in which different source classes can possibly contribute
to the 2-20 keY XRB are summarised by Holt (1990).
The only direct method of establishing the nature and red shift of dis-
crete sources contributing to the extragalactic XRB is to resolve as many
"background" sources as possible in a deep X-ray survey with an imaging
telescope, and to establish the nature of such sources by means of optical
spectroscopy of the corresponding counterparts.
The first attempt to resolve the X-ray background into discrete sources
by using an imaging X-ray telescope was made by Giacconi et al. (1979)
using the Einstein observatory (HEAO-B), which carried an imaging X-
ray telescope with position sensitive proportional detectors (IPC) and a
micro channel plate (HRI). The Einstein Observatory was used to make six
deep surveys, each of observation time exceeding 50,000 secs. Giacconi et
al (1979) detected 40 X-ray sources, giving a source density of about 20
per sq. degree at flux limits of 2.6 X 10- 14 ergs cm- 2 s-1 (0.5 - 3.5 keY).
Giacconi et. ale claimed that discrete sources contribute about a quarter of
the extragalactic XRB in the energy range 1-3 keY, where such sources were
detected at least five standard deviations above background in the IPC.
They assumed that the spectrum of the X -ray background in the 1 to 3 ke V
range was a smooth extrapolation of the 3 - 40 ke V background spectrum
measured by Marshall et. ale (1980) with HEAO-A2. These discrete source
detections were made largely with the Imaging Proportional Counter on the
Einstein Observatory, with resulting X-ray astrometric errors of 60 arcsec.
The large "error circles" and lack of full spectroscopy of all the possible
 175

candidates meant that the nature of most of the discrete sources remained
unknown. Giacconi et. al. identified four of their 43 sources with quasars.
An independent and careful re-analysis of the Einstein IPC Deep Survey
data by Hamilton, Helfand, & Wu (1991) has shown that two-thirds of the
33 IPC sources reported by Giacconi et. al. were probably spurious.
In contrast, the Pavo deep survey in the southern hemisphere was per-
formed predominantly with the micro channel plate High Resolution Imager
(HRI), with source positional errors of 10 arcsec or less. The HRI was used
for 4 overlapping exposures, each exceeding 70,000s, with an IPC exposure of
69,000s (Griffiths et al. 1983). The source identification content of this field
was determined using multi-object spectroscopy at the Anglo-Australian
Telescope and at the European Southern Observatory and showed that, at
the Einstein Deep Survey (EDS) limit, the X-ray source counts were domi-
nated by AGN at median redshifts approaching one (Griffiths et al. 1992).
This survey also demonstrated that the EDS sources were not dominated by
low-luminosity AGN (LLAGN); the X-ray luminosities were 1043 _10 44 ergs
s-l , at optical magnitudes of J = 20.3. Since, on the number-magnitude
diagram for quasars, the optical quasar counts start to turn down at this
magnitude, it was predictable that deeper surveys would find a fall-off in
the X-ray log(N)jlog(S) slope, in agreement with the fluctuations analysis
of Hamilton & Helfand (1987) and later surveys with ROSAT and other
telescopes.
Although the deep surveys were used to try to directly resolve as much
of the XRB as possible, there were also indirect ways of approaching the
problem. The Einstein Observatory images from long exposures (greater
than 1000 s) generally contained serendipitous field objects which could
be used to define a medium-depth survey. Eventually, hundreds of fields
covering nearly 1000 deg 2 were used in the Extended Medium-Sensitivity
Survey (EMSS) conducted by the Center for Astrophysics group (Gioia et
al. 1990); Stocke et al. (1991) undertook the extraordinary task ofidentifying
700 of the 800 sources in the EMSS. Of the identified sources, 487 were found
to be AGN, and were used to define X-ray luminosity functions in different
redshift shells, and thus to establish the evolutionary properties of AGN in
the 0.3 - 3.5 keY band (Della Ceca et al. 1992). From the best-fit models for
the evolution of the AGN, estimates were then made of the contribution to
the XRB, with a most-likely estimate of 40 ± 10%.
The Einstein Medium and Deep Surveys also contained a minority of
objects identified with starburst galaxies (Griffiths et al. 1992; Fruscione and
Griffiths 1991), some of which had been classified by Stocke et al. (1991) as
"normal" galaxies in the EMSS. These objects, and those in the ROSAT
deep surveys, give an important clue towards the nature of the next most
populous type of discrete source contributing to the LEXRB and possibly
the MEXRB (see §6).
176

10

This model
[),ISS model

5 I ,.,.... .
....... ..... .. .... ..

,,......... ,,
I .... . .
.... ..

i
,
.i " ... ... . .. ......
:,
:1 ... ... ~

...

----
o L - - J__ ~ __ ~ __ L-~ _ _- L_ _~_ _~~~~_ _~--J

o 1 2 3
z
Fig. 6. The redshift distribution of the AGN in the ROSAT QSO-survey fields QSF1
and QSF3 from Boyle et al. (1993).

5. The Discrete Source Content of The ROSAT Surveys

The medium and deep surveys were discontinued in 1980 with the demise of
the Einstein Observatory, and were not resumed until 1990 with the launch
of the Roentgen Satellite, ROSAT. ROSAT is a "softer" X-ray telescope than
the Einstein Observatory, and, as far as the extragalactic XRB is concerned,
is effective from 0.5 to 2 keY, with an effective detection energy for most
sources of about 1 ke V. There have been several teams who have performed
deep surveys with ROSAT: Hasinger et al. (1993), Shanks et al. (1991),
Anderson et al. (1992). The deepest surveys, and those which have resolved
the highest source density, have been the Hasinger et al. surveys, reaching
 177

source densities of rv 400 deg- 2 in the Lockman hole areas; the shallower
surveys of Shanks et al. (1991) and Anderson et al. (1992) have reached
source densities of rv 120 deg- 2 • Shanks et al. (1991) and Griffiths et al.
(1993) have used the multi-object spectrographs at the AAT to show that
the ROSAT surveys are dominated by AGN with median redshift of 1.5.
In the 30,000 s surveys of fields which had previously been surveyed for
UV-excess AGN by Boyle et al. (1990), the AGN account for two-thirds of
the sources, which in turn make up 30% of the XRB; AGN are observed
to account directly for at least 20% of the XRB at 1 keY. The AGN are
observed to have X-ray spectra slightly softer than the residual XRB in
the limited ROSAT range (Shanks et al. 1991), hinting towards a residual
source population that is harder than the AGN. Hasinger et al. (1993) have
confirmed this result, finding that the faintest sources in the deepest surveys
have summed spectra which are harder than the brighter sources which are
suspected of being predominantly AGN.
The fact that the Einstein deep surveys were dominated by AGN meant
that the ROSAT surveys at slightly deeper flux levels would probably con-
tain a large fraction of AGN also. The known quasars in the fields were of
enormous advantage as a means of providing ready identification to about
a quarter of the total source count, and as a means of refining the X-ray
astrometry for the rest of the sources in the fields, especially during the
early phase of the ROSAT mission. Results of the optical identifications of
ROSAT sources found in fields QSF1 and QSF3 of Boyle et. al. (1990) were
completed first. These fields each contained rv 20 known UV-excess quasars,
at median redshifts of rv 1.5. Initial results on the QSF3 field were presented
by Shanks et. al. (1991). Extended optical sources were eliminated during
the quasar survey of Boyle et. al., so that follow-up spectroscopy was needed
for all X-ray sources identified with galaxies, as well as "stellar" objects that
did not show UV-excess and objects fainter than the UV-excess survey mag-
nitude limits. The AGN in the ROSAT-QSF surveys have been combined
with those from the Einstein EMSS used to expand upon the evolutionary
models of Della Ceca et al. (1992) and to show an apparent evolution at 1
keY which appears to be faster than that seen in the Einstein data (Boyle
et al. 1993, and Boyle these proceedings).
One of the major problems in the calculation of the LF is that the spectral
index of an individual AGN is not determined; the data are insufficient to
measure anything but the summed index for all AGN in the fields, or all
AGN in two or three redshift ranges. Using the luminosity function from
the combined datasets, Boyle et al. (1990) obtain a best estimate of the
contribution of AGN to the 1 ke V XRB of between 30% and 70%.
Other discrete source detections have included the first examples of star-
burst and star-forming (narrow emission line) galaxies in ROSAT deep sur-
veys. A discussion of the contribution of these galaxies to the XRB is elab-
178

50rl~-r-r'-r-rl~-r'-~TirlJ/~r-rJ~-r'-'-rl-'-'~~
I
no evolution I
I
I
T = 0.4 I
40 I
---'T T = 0 .2 /
I
I
I
I
30 I
I
I
I
I
I
I
20 /
/
/
/

------ ---
/
/

/
/

-----
10

//~-
------~~-=---==~---
I¥ _---
o E~~/.~-:
~ · ~l§~~~-~-I-~~;c~jt~~~~~~~~=t~====~
_8 ----

o .5 1.5 2 2.5 3
redshift

Fig. 7. The predicted contribution to the XRB from starburst galaxies from Griffiths
& Padovani (1990)

orated upon in the following Sections.

6. The Contribution from Starburst Galaxies

The deep-survey fluctuations analyses have indicated a (residual) source
population having a surface density of at least 5,000 deg- 2 • It cannot be
ruled out that these sources are AGN (Seyfert 2's?), but another possibility,
supported by the optical identifications in the ROSAT deep surveys, is that
they are starburst or post-starburst galaxies, with X-ray emission possibly
dominated by populations of massive and low-mass X-ray binaries (Griffiths
& Padovani 1990).
 179

The Population I X-ray binaries were established as a class of X-ray emit-

ters in the late 1960s and 1970s (for a summary of their X-ray properties see
Rappaport & Joss 1983). MXRB have primaries of rv 20 M 0 , with accreting
secondaries of rv 1 - 7 M 0 . Their orbital periods are typically a few days,
and their optical properties have been summarised by van Paradijs (1983).
They start emitting X-rays about 10 7 years after the formation of the initial
massive binary, and the epoch of X-ray emission is rather uncertain, but
probably lasts for at least 10 5 to 106 years (see "an den Heuvel 1983 for a
review of the evolution of such systems). The presently active and known
MXRB in our Galaxy number about 20 (van den Heuvel 1983) with X-ray
luminosities of rv 1038 ergs S-1 and typical X-ray temperatures >15 keY
spectra make them excellent candidates as contributors to the all-sky back-
ground, provided that the low-metallicity counterparts of such systems are
sufficiently common in regions of star-formation at moderate redshifts.
Our Galaxy does not contain any low-metallicity MXRB, so that sources
like Cyg X-1 and Cen X-3 may be underluminous examples of those found
in star-forming galaxies. We have to look at least as far as the Local Group,
where we find SMC X-1 and LMC X-4. The low metallicity of the SMC has
been taken as the reason for the extraordinary X-ray output of the SMC,
rivalling the total X-ray output of the Galaxy, with only one-tenth the mass
(Clark et. al. 1978). The O-stars in the SMC are also more luminous in X-
rays than O-stars in the Galaxy. Generally, low metal content means lower
X-ray opacity in the accreting material, driving up the accretion rate and
enhancing the production of hard X-rays. The high-energy tail of binaries
like SMC X-1 may be especially important in considerations of the XRB.
Nearby galaxies include examples of super-luminous sources in M82 (Wat-
son, Stanger, & Griffiths 1984), M101 (Trinchieri, Fabbiano, & Romaine
1990), and perhaps NGC 253 (Fabbiano & Trinchieri 1984). Long & Van
Speybroeck (1983) drew attention to the compact binaries in "normal"
galaxies, where it is not unusual to find individual sources with luminosities
in excess of 1039 ergs s-1. M101 has four or five such sources, M100 has a
source emitting in excess of 1040 ergs s-I, and M82 likewise has a source in
an outlying arm with L:x rv 1039 ergs S-1, possibly variable (Schaaf et. al.
1989). Although not individually identified with massive X-ray binaries (but
see, e.g., Stocke, Wurtz, & Kuhr 1991), this this may be possible with HST
in some cases. We make the plausible assumption here that such sources are
members of the class typified by SMC X-l.
The X -ray spectra of the MXRB are typically flat in the 2 - 10 ke V range;
they can be characterised by bremsstrahlung temperatures in excess of 108
K, or kT > 15 keY. Examples are reviewed in Rappaport & Joss (1983),
and in Nagase (1989). Although individual X-ray spectra have not yet been
obtained for star-forming galaxies, Fabbiano (1989) has suggested that the
summed spectrum of a few spiral galaxies appears hard over the limited
180

energy range of the Einstein imaging proportional counter, consistent with

that expected if their emission is dominated by X-ray binaries. We note that
the total X-.ray binary contribution will generally be a composite of MXRB
and low-mass or intermediate-mass X-ray binaries. Some of the latter objects
also have very hard X-ray spectra and high-energy tails. The X-ray spectrum
of the starburst galaxy M82 has 'been reported by Schaaf et. al. (1989), who
combined the data from the EXOSAT and Einstein Observatories with the
result that the composite spectrum has an apparent temperature of kT '" 10
keY, but this is presumably a composite of the soft X-ray component from
the outfiowing wind (Watson, Stanger, & Griffiths 1984), combined with
a hard component from the X-ray binaries. We do not consider here the
contribution to the soft X-ray background from the' out flowing wind, but
concentrate instead on the binary component.
The relationship between X-ray and infrared luminosities can be derived
for both IRAS and starburstlinteracting (S/1) galaxies from Einstein/lRAS
data. For the IRAS sample the best linear fit, using the EM method, is given
by

10gLa:(O,5-3,OkeV) = (9.69 ± 5.03) + (0.70 ± 0.12) 10gL 6ol-'m , (2)

while for the S II sample it was

10gLa:(o.5-3.0keV) = (13.26 ± 6.12) + (0.62 ± 0.14) 10gL6ol-'m • (3)

We can then infer that the X-ray to far infrared ratio is similar for IRAS,
starburst and normal galaxies. Having established the correlation between
X -ray and infrared luminosities, which is independent of the attempt to
classify galaxies as actively "starburst" or not, we can use the two infrared
luminosity functions and the X -ray linfrared luminosity relationship for S II
galaxies to derive the corresponding X-ray luminosity functions for IRAS
and SII galaxies and hence the contributions to the XRB.
It is easy to show that, if La: <X LSOl-'m' the differential X-ray luminosity
function is <f>(L:z:) = <f>(L60l-'m) (L601-'ml L:z:)c 1 • The intensity of the XRB
due to star-forming galaxies can be calculated, for a Friedmann-Robertson-
Walker cosmology, from the equation

(4)

where 0a: is the X-ray spectral index. Assuming no evolution (i.e., constant
luminosity and number conservation) the fraction of the XRB at 2 keY due
to objects having Z ~ Zma:z: can be expressed as

F( 2 ke V) ~ 2.6 10- 48 p( L2 keV) [

1 (1 + Zma:z: )(-l-a",-qo)] ,
X - (5)
1 + 0a: + qo
 181

where P(L2 keV) is the luminosity density at 2 keY in units of ergs S-1 Gpc- 3
and qo can be 0 or 0.5. For the case of luminosity evolution of the type
L(z) = L(O) exp[T(z)/r], where T(z) is the look-back time and r is the time
scale of the evolution, equation (3) can be solved numerically.
It has also become apparent that nuclear activity may be related to star-
burst activity. Ward (1988) has shown a correlation between X-ray lumi-
nosity and Br-y emission in starburst nuclei, supporting the hypothesis of
X-ray emission arising from MXRB in the nuclear regions. It would not be
unreasonable for the evolution of such starburst nuclei to evolve as rapidly
as quasars. Some fraction of the active star-forming galaxies at high redshift
may evolve into AGN (Norman & Scoville 1988). This may well be the case
for the luminous IRAS galaxies, and may indeed be a common origin for
AGN, but the calculations performed here are for star-forming galaxies in
general, whether or not they develop AGN. .'
Figure 8 shows the fractional contribution of S /1 galaxies (curves a) and
IRAS galaxies (curves b) to the XRB at 2 ke V, as a function of redshift,
for the cases of no evolution, r = 0.4, and r = 0.2 assuming Qz = 0.3 and
qo = O. The value r '" 0.2 is typical of the X-ray evolution of AGN (Maccac-
aro, Gioia, & Stocke 1984). Star-forming galaxies with moderate evolution
contribute then between 4% - 26% of the XRB at 2 keY, depending on
the choice of the luminosity function: given our previous discussion, higher
values are preferred. It is also interesting to note that non-evolving IRAS
galaxies contribute about 10% of the XRB at 2 keY, while the same ob-
jects following a quasar-like evolution would exceed 50% of the background
when integrated to z '" 1.5. These values refer to log Lz > 39.3 and should
therefore be considered as lower limits. For example, an extension of the low
end of the IRAS luminosity function to a luminosity a factor of 3 smaller
would increase our estimate of the contribution to the XRB by about 60% .
We note in particular that the evolutionary models of Hacking, Condon, &
Houck (1987) - infrared evolution (1 + z )3-4 for qo = 0.5 - converted to an
X-ray evolution through equation (2), predict a contribution to the 2 keY
XRB in the range 22% - 34% with Qz = 0.3.
The expression which relates the differential counts to the luminosity
function, assuming a Friedmann-Robert son-Walker cosmology, is

N(S, z) dS dz = 411"c </>(L(S, z), z) Di(z) dS dz (6)

411" Ho (1 + z)4-a(1 + 2qoz)l/2
where c is the speed of light and DL(Z) is the luminosity distance. An inte-
gration between the appropriate redshifts gives the differential counts N(S)
and the integral counts are N( > S) = J N(S) dS.
Star-forming galaxies may reach a peak in the integral counts at 10- 16 -
10- 19 ergs cm- 2 s-I, with sufficient numbers in the range (5 - 10) X 10- 15
ergs cm- 2 s-1 that they start to appear in the Einstein deep surveys. They
 182

4
no evolulion
3 T = 0.4
T = 0.2
2

eo
I
.,
tIO
1
,,
,,
--""
(f) 0 ,
-
A

-
Z
tIO ,,
" , [ ..SS
0 -1
,,
"

-2 .
...
"

-3

-4
-16 -15 -14 -13 -12 -11
log S (0.3 - 3.5 keY)

Fig. 8. The predicted log(N)/log(S) for starburst galaxies from Griffiths & Padovani
(1990)

should also show up in the ROSAT deep surveys as individual detections

in slightly larger numbers. Figure 9 illustrates the difficulty in establishing
the X-ray evolution of star-forming galaxies; the evolving and non-evolving
cases do not differ significantly until large samples are observed at :f:I.ux limits
of", 10- 15 - 10- 16 ergs cm- 2 S-1, beyond the range of current instruments
and barely possible with AXAF. Detection in the Einstein and ROSAT deep
surveys is, of course, aided by X-ray selection; small numbers of objects will
be detected with unusually large X-ray /infrared luminosity ratios, in a sim-
ilar fashion to the X-ray selected AGN in the Einstein Medium Sensitivity
and deep surveys.
A composite spectrum for a star-forming galaxy may be constructed by
 183

averaging the spectra of MXRB in our own galaxy (e.g., Nagese 1989),
together with some fractional contribution from medium and low-mass x-
ray binaries. Although some MXRB have high-energy cutoffs at or above 20
keY, hard X-ray tails are by no means uncommon amongst these sources.
In summary, the (0.5 - 3.5 keY) X-ray fluxes from starburst galaxies have
been found to be correlated with their infrared fluxes. The infrared luminos-
ity function of starburst galaxies can then be used to form the X-ray LF, and
observed or estimated evolution of the infrared LF can be used to infer the
total contributions of starburst galaxies to the XRB. The estimates of Grif-
fiths & Padovani (1990) have beeen corroborated by calculations of Green et
al. (1989); David et al. (1991), and Treyer et al. (1992). On the assumption
that the starburst galaxies have hard X-ray spectra typical of massive X-ray
binaries, then the contribution to the 2 ke V XRB is of the order of 10 - 50%.
Some of the starburst galaxies which have been studied using ROSAT have
shown soft X-ray spectra, but ROSAT will, of course, q,etect the soft X-ray
excesses in such objects resulting from supernova remnants and ouflowing
winds. ROSAT is not very sensitive to the heavily-absorbed starburst nu-
clei that may be important for the XRB, although spiral galaxies have been
observed from a ROSATjIRAS galaxy correlation (Boller et al. 1991).
A difficulty with the prediction of X-ray emission from MXRB in star-
forming systems is that the metallicity is unknown. Much of the activity
may take place in dwarf galaxies or satellite galaxies, by direct analogy with
our own Galaxy and the MXRB in the LMCjSMC, or in galaxies of low
surface brightness.
Of the unidentified sources in the ROSAT deep surveys (e.g., Shanks
et al. 1991), some are certainly likely to be AGN fainter than the optical
identification limit (B = 21), but most are more likely to be X-ray selected
galaxies of various kinds. The initial non-AGN galaxy identifications are
predominantly early-type galaxies with summed X-ray spectra which are not
noticeably harder than the AGN in the same fields (Griffiths et al. 1993),
but initial estimates by Hasinger et al. (1993) of the summed spectra of faint
X-ray selected galaxies indicate that they may, indeed, represent a harder
class of sources. The optical properties and classification of the galaxies in
the deep Hasinger et al. fields is not yet established. If such a population
of hard-spectra galaxies does exist in the ROSAT data, then obviously they
may be extremely important for the XRB above 3 ke V.
The 4m-class telescopes equipped with multi-object couplers were a fairly
good match in the 1980's to the faintest sources seen in the Einstein deep
surveys (which had optical magnitudes B = '" 20 - 24). In contrast, the
faintest counterparts to the ROSAT deep survey sources are somewhat be-
yond the capabilities of most 4m telescopes in typical observing conditions,
and await the installation of multi-object spectrographs on the next gener-
ation of large telescopes (Keck, Gemini, the SAO j ASU 6.5m, etc.).
184

7. Early-type Galaxies in the ROSAT Deep Surveys

The next most populous class of extragalactic objects at 1 keY appears to be

early-type galaxies (Griffiths et al. 1993), where ROSAT may be detecting
the '" 1 ke V bremsstrahlung emission from the halos of the ellipticals. Such
objects may contribute 5 - 10 % of the XRB at 1 keY, but probably are
not important in the MEXRB. In the first deep surveys, many early-type
galaxies have been identified. Several of them lie at a common redshift of
0.312, although spread over 14 Mpc, with the following possibilities for the
X-ray emission:
(a) The galaxies may each be the brightest members of small groups of
galaxies, in turn members of a loose supercluster or uncollapsed structure
at z = 0.312. The weak X-ray sources could be at the centroid of extended
emission from each group, although we do not observe evidence for X-ray
extensions. Small groups have been observed to have X-ray luminosities (0.3
- 3.5 keY) of", 1043 ergs cm- 2 s-1 (Stocke et a1. 1991). Compact groups of
size'" 400 kpc would appear to have angular extent 80" at z = 0.3, and might
thus appear as unresolved sources in ROSAT PSPC observations. The X-ray
spectra of groups generally show kT '" 1 ke V, consistent with the observed
X-ray spectra. The EMSS source 1Ell11.9-3754 is a prototype for this class
of X-ray selected object: viz. a giant red elliptical galaxy (GREG) in a poor
group of galaxies at the lower redshift of 0.13 (Maccagni et al. 1988). The
X-ray luminosity is 1044 erg S-I, and the absolute visual magnitude is Mv =
-25. GREG is at the bottom of a potential well which confines the X-ray
emitting gas, similar to some of the Albert/Morgan groups (Kriss, Cioffi, &
Canizares 1983).
(b) The X-ray emission may arise in the active galactic nuclei of otherwise
passive early-type galaxies. The EMSS contains interesting examples of such
objects, such as E0002.5-4205 and E0007.1-0231, with about twenty other
similar objects (Stocke et a1. 1991). These objects generally do not have
[0 III] or H,L3 in emission, but can have broad emission lines of Mg II and
Ho superimposed on an early-type absorption spectrum. The LDSS spec-
troscopic coverage did not extend to redshifted Ho, and the sensitivity was
very low at redshifted Mg II. We cannot rule out the possibility that each
of the "passive" galaxies harbors an active nucleus. Considering the evolu-
tionary models of the interstellar medium and outflowing/inflowing gas in
ellipticals, it has been pointed out that, if and when a "mini-cooling flow" is
established in the inner regions of an elliptical with a de Vaucouleurs profile,
AGN activity may be triggered or re-triggered (Ciotti et al. 1991). In such
a scenario, L:e would reach a peak of a few times 10 42 ergs S-I, consistent
with the observed fluxes within a factor of a few. An extreme example of X-
ray selected AGN activity in an elliptit:al galaxy is provided by H1821+643
(with X-ray luminosity'" 1046 ergs S-I), albeit in a rich cluster of galax-
 185

ies, with the AGN activity possibly powered by a moderately high cooling
flow. Some or all of the objects described here may be weaker examples
of the same phenomenon. We also note that H1821+643 is radio-quiet, as
are the galaxies reported here. Further spectroscopy, outside the observed
wavelength range, is required in order to confirm this possibility.

(c)·The X-ray emission may be coming from !tot gas in the coronal halos
of the early-type galaxi~s, gravitationally confined by massive dark halos.
David, Jones, & Forman (1991) and Ciotti et al. (1991) have modelled the
evolution of the interstellar medium of elliptical galaxies, along with the
out flowing winds and the development of cooling inflows in the more mas-
sive systems. The X-ray spectra would be characterised by thermal emission
with kT '" 1 keY, consistent with the observed X-ray spectra. In these mod-
els, X-ray luminosities in excess of 10 42 erg S-1 can arise from hot coronal
gas around a galaxy with ME < -21, but the corresponding amount of gas
can only be confined if there is a massive dark halo. Such a hot corona may
be in the cooling phase, and a "cooling flow" may have been initiated, but
there is no evidence for an established flow in the galaxies reported here -
i.e., we do not see evidence in the optical spectra for emission lines from
cool filaments. The spectra did not cover redshifted Ha, however, and not
all X-ray luminous "cooling flows" show line emission from cool filaments.
In fact, only about half of them do so (Sarazin 1986), and cooling flows may
establish themselves first in the cores of the galaxies, without involving all of
the halo gas (Ciotti et al. 1991). Certainly, the observed X-ray luminosities
are not excessive for "cooling flow" galaxies or for galaxies which are still in
the outflow phase. The key factor in the explanation of the X-ray luminosity
is the containment by massive halos. Fully-developed "cooling flow" galaxies
are usually more luminous systems, viz. cD galaxies with absolute lumonosi-
ties Mv '" -22 or brighter (see Sarazin 1986), but the objects reported here
may be weaker examples.

In all three possibilities, we are still confronted by the discovery of X-

ray selected galaxies at a common redshift, spread over an angular range of
30 arcmin, or '" 10 Mpc in the plane of the sky. Other observations that
could potentially discover such common galaxy structures or redshifts are, of
course, the "pencil-beam surveys", such as those of Broadhurst et al.(1988) .
Pencil-beam surveys in directions at large angles to the Galactic pole axis
do not show the same periodic redshift distribution as those at the North
and South Galactic Poles (Koo et al. 1993), so that all observations thus far
are consistent with tesselated structures of characteristic size 50 - 100 Mpc.
186

8. A new class of A G N, or Seyfert lIs ?

Although Schwartz & Tucker (1988) showed that the MEXRB could be
a superposition of ad hoc spectra of AGN that were consistent with all
known AGN spectra, there was little physical reason for such spectra to
be ubiquitous. Cavaliere et al. (1979) had shown earlier that the MEXRB
could equally well be a superposition of power-law spectra of AGN, with
an exponential distribution in the cut-off energies. More recently, Fabian et
al. (1990) have discussed a theoretically more plausible AGN spectrum that
could give rise to the XRB, viz. a spectrum dominated by a reflected X-ray
component, in which the direct X-rays are heavily absorbed. The geometry
of such sources is a little hard to understand, but the model has gained
great popularity over the past few years (Rogers & Field 1991). The new
class of AGN has not been observed - but one or two objects in the Ginga
compilation have the requisite hard spectra (Tanaka et al. 1992). The optical
counterparts to these examples do not, however, form a large, separate class
of AGN,
Zdziarski et al. (1993a) has pointed out some major difficulties with the
reflection model as the dominant component of the MEXRB, and has pro-
posed instead that the bulk of the emission arises from Compton scattering
of soft thermal photons by a mildly relativistic plasma in a corona around
the AGN (Zdziarski et al. 1993b). The soft photons could be in the optical-
UV range and arise in an accretion disk. The model is based on observations
of 50 ke V thermal spectra in Seyfert galaxies by the Oriented Scintillation
Spectrometer Experiment (OSSE) on the Gamma Ray Observatory (GRO).

9. Conclusions
The X-ray background, although discovered 30 years ago, remains largely
unexplained because of the technical difficulties involved in focussing X-rays
in the appropriate energy range (3 - 40 keY). The super-polished mirrors of
the Einstein and ROSAT X-ray telescopes have had good reflectivities up
to 4 ke V and 2 ke V respectively. Above these energies, only BBXRT has
thus far been flown, although several missions are planned for the remaining
part of the decade: the joint Japanese-US mission ASTRO-D comes first,
in 1993, followed by the European mission JETX in 1995, the Japanese
DUET in 1997, the Advanced X-ray Astrophysics Facility (Imaging Mission),
AXAF(I) in 1998 and the European X-ray Multi-Mirror Mission XMM in
1999. All these missions carry X-ray telescopes which have large collecting
areas and good efficiencies up to energies of rv 8 ke V.
The Einstein and ROSAT missions have made great strides in explaining
the excess background counts below 3 keY (in excess of the "40 keV" XRB
observed at 3 - 40 keY), which is largely caused by AGN and other galaxies.
 187

The same missions, however, have made little progress towards explaining
the bulk of the problem above 3 keY. The AGN themselves appear to have
spectra which are too soft for the overall XRB, and can only help to explain it
if they have hard spectral components. The low-redshift samples fail to meet
this criterion, although we do not yet have> 3 ke V spectra of representative
AGN at redshifts of 1 - 3 ..
The hot IGM has been effectively eliminated as the source of the "40 ke V"
XRB, and a new class of hard AGN, unobserved at any other wavelength, is
currently the favored model. Other forms of active galaxy, such as starburst
nuclei in IRAS-type galaxies, are still viable candidates, however. Progress
in the study of the XRB can only come via a complete understanding of the
source content at each energy, together with measurement of their spectra
and the evolution of each class of object. Much more progress is expected
to be made during the remainder of the decade.

References
Anderson, S. F ., et al. 1992, in X . ray Emiuion from Active Galactic Nuclei and the X-ray
Background, Proc . Conf. at MPE, Nov. 1991 , Brinkmann, W ., and Trumper, J., eds.,
p. 227, MPE report 235
Avni, Y. & Tananbawn, H. 1986. ApJ, 305, 83
Barcons, X. & Fabian, A . C . 1990. MNRAS, 243 , 366
Baicons, X. & Fabian, A . C. eds., 1992. The X-ray Background, Proc . Laredo Workshop,
Cambridge University Press
Blanchard, A. , Wachter, K., Evrard, A. E., Silk, J. 1992, in The X-ray Background, Bar-
cons , X . , & Fabian, A . C., eds., p. 201, Cambridge University Press
Boldt, E. A. & Leiter, D. 1987. ApJ, 322, L1
Boller, Th. , Meurs, E . J . A ., Brinkmann, W ., Fink, H., Zimmermann, H .-U ., Adorf, H.-M.
1992, in X-ray Emission from Active Galactic Nuclei and the X-ray Background, Proc .
Conf. at MPE, Nov. 1991, Brinkmann, W ., and Trwnper, J ., eds., p. 231, MPE report
235
Boyle, B . J. , Fong, R. , Shanks, T ., & Peterson, B. A. 1990, MNRAS, 243, 1
Boyle, B. J ., Griffiths , R. E., Shanks, T ., Stewart, G. C. , & Georgantopoulos, I. 1993 ,
MNRAS, 260, 49
Brinkmann, W ., & TrUmper , J., eds. 1992, X-ray Emisllion from Active Galactic Nuclei
and the X-ray Background, Proc. Conf. at MPI, Nov. 1991, MPE report 235
Broadhurst, T. J. , Ellis, R. E., & Shanks, T. 1988, MNRAS , 235, 827
Burrows, D. N. , Garmire, G. P., Lwnb, D. H., & Nousek, J . A. 1992, in The X-ray Back-
ground, Barcons, X., and Fabian, A. C., eds., p. 277, Cambridge University Press
Carrera, F., & Barcons, X . 1992, MNRAS, 257,507
Cavaliere, A., Danese, L. , De Zotti, C ., & Franceschini, A. 1979, A&A, 79, 169
Ciotti, L., D'Ercole, A., Pellegrini, S., & Renzini, A. 1991, ApJ, 376, 380
Clark, G ., Doxsey, R., Li, F., Jernigan, J. G. , & van Paradijs, J . 1978, ApJ, 221, L37
Condon, J . J., 1987, in "Starbursts and Galaxy Evolution", eds. Trinh Xuan Thuan, T.
Montmerle, & J . Tran Thanh Van, Editions Frontieres, p . 425
Cooke, B. A., et al. 1978, MNRAS, 182, 455
Cowie, L. L., 1989, AGN and the X-ray Background, Proc. 23rd ESLAB SYInposium, p .
707, ESA SP-296
Cowsik, R ., Kobetich, E. J. 1972, ApJ, 177, 585
Daly, R. A . 1991, ApJ, 379, 37
188

Danese, L., de Zotti, G., Franceschini, A., & Toffolati, L. 1987, ApJ, 318, L15.
David, L. P." Forman, W., Jones, C. 1991, ApJ, 369, 121
Della Ceca, R., Maccacaro, T., Gioia, I. M., Wolter, A., Stocke, J. T. 1992, ApJ. 389, 491
de Zotti, G., Boldt, E. A ., Cavaliere, A ., Danese, L., Franceschini, A., Marshall, F. E.,
Swank, J. H., & Szymkoviak, A. E. 1982, ApJ, 253, 47
Edge, A .. C., Stewart, G. C., Fabian, A. C ., & Arnaud, K. A. 1990, MNRAS, 245, 559
Fabian, A. C. 1988. In The Post-Recombination Universe, NATO ASI Series C, vol. 240,
eds. N. Kaiser and A. N. Lasenby, p. 51
Fabian, A. C., & Barcons, X. 1992, ARAA, 30, 429
Fabian, A. C., George, I. M., Miyoshi, S., & Rees, M. J . 1990, MNRAS, 242, 14P
Fabbiano, G., & Trinchieri, G. 1984, ApJ, 286, 491
Fabbiano, G. 1989, ARAA, 27, 87
Field, G. B., & Perrenod, S. C. 1977, ApJ, 215,717
Friedman, H. 1990, The Astronomer's Universe, Norton, New York
Fruscione, A., & Griffiths, R. E. 1991. ApJ, 330, L13
Garmire, G. P., Nousek, J. A., Apparao, K. M. V., Burrows, D. N., Fink, R. L., & Kraft,
R. P. 1992, ApJ, 399, 694
Georgantopoulos, I., Stewart, G. C., Griffiths, R. E., Shanks, T., & Boyle, B. J. 1993,
MNRAS, in press.
Giacconi, R. 1974, in X-ray Astronomy, Giacconi, R., & Gursky, H., eds., p. 1, Astrophysics
and Space Science Library, D. Reidel
Giacconi, R., Bechtold, J., Branduardi, G., Forman, W., Henry, J. P., Jones, C., Kellogg,
E., van der Laan, H., Liller, W., Marshall, H., Murray, S. S., Pye, J. , Schreier, E.,
Sargent, W. L. W., Seward, F. D., & Tananbaum, H. 1979. ApJ, 234, Ll
Giacconi, R., Gursky, H., Paolini, F., & Rossi, B. 1962, Phys. Rev. Lett., 9, 439
Giacconi, R. & Zamorani, G. 1987, ApJ, 313, 20
Gioia, I. M., Maccacaro, T., Schild, R . E., Wolter, A ., Stocke, J. T ., Morris, S. L ., &
Henry, J. P. 1990. ApJS, 72, 567
Gioia, I. M., Henry, J. P., Maccacaro, T., Morris, S. L., Stocke, J. T., & Wolter, A. 1990,
ApJ, 356, L35
Gould, R., & Burbidge, G. 1963, ApJ, 138, 969
Green, P. J., Ward, M., Anderson, S. F., Margon, B., deGrijp, M. H. K., & Miley, G. K.,
1989, ApJ, 339, 93
Griffiths, R. E., 1989, in "The Epoch of Galaxy Formation", eds. C. S. Frenk et. at:.,
NATO ASI Series C, vol. 264, Kluwer Academic Publishers, p. 235.
Griffiths, R. E., & Padovani, P. 1990, ApJ, 360, 483
Griffiths, R. E., Murray, S. S., Giacconi, R., Bechtold, J., Murdin, P., Ward, M., Peterson,
B. A., Wright, A. E., & Malin, D. F. 1983. ApJ, 269, 375
Griffiths, R. E., Tuohy, I. R., Brissenden, R. J. V., & Ward, M. J. 1992, MNRAS, 255,
545
Griffiths, R. E., Georgantopoulos, I., Boyle, B . J ., Stewart, G. C., & Shanks, T. 1993,
MNRAS, in press
Guilbert, P. W., & Fabian, A. C. 1986. MNRAS, 220, 439
Hacking, P., Condon, J. J., & Houck, J. R. 1987 ApJ, 316, L15
Hamilton, T., 1992, in The X-ray Background, Barcons, X., & Fabian, A. C., eds., p. 138,
Cambridge University Press
Hamilton, T. T., & Helfand, D. J. 1987, ApJ, 318, 93
Hamilton, T. T ., Helfand, D. J., & Wu, X. 1991, ApJ, 379, 576
Hasinger, G., Burg, R., Giacconi, R., Hartner, G., Schmidt, M., Trumper, J., & Zamorani,
G., 1993, A&A, in press, MPE preprint 238.
Holt, S. S. in The X-ray Background, Barcons, X., & Fabian, A. C., eds., p. 29, Cambridge
University Press
Hoyle, F. 1963, ApJ, 137, 993
Jahoda, K., et a1. 1992, in The X-ray Background, Barcons, X., & Fabian, A. C., eds., p .
240, Cambridge University Press
 189

Kellogg, E. 1974, in X-ray A6tronomy, Giacconi, R., & Gursky, H., eds., p. 321, Astro-
physics and Space Science Library, D. Reidel
Khembavi, A. K. & Fabian, A. C. 1982, MNRAS, 198, 921
Kriss, G. A., Cioffi, D. F., & Canizares, C. R. 1983, ApJ, 272, 439
Lahav, O. 1992, in The X-ray Background, Barcons, X., & Fabian, A . C., eds., p. 102,
Cambridge University Press
Leiter, D. & Boldt, E. 1982, ApJ, 260,1
Long, K. S. & van Speybroeck, L. P. 1983, in "Accretion-driven Stellar X-ray Sources",
eds. W. H. G. Lewin & E. P. J. van den heuvel, p. 117
Maccacaro, T., Gioia, 1. M., & Stocke, J. T. 1984, ApJ, 283, 486
Maccagni, D, Garilli, B., Gioia, 1. M., Maccacaro, T., Vettolani, G., & Wolter, A. 1988,
ApJ, 334, L1
Marshall, F. E., Boldt, E. A., Holt, S. S., Miller, R. B., Mushotzky, R. F., Rose, L. A.,
Rothschild, R. E., & Serlemitsos, P. J. 1980, ApJ, 235, 4
Mather, J. et al. 1990, ApJ, 354, L37
McCammon, D., & Sanders, W. T. 1990, ARAA, 28, 657
Morisawa, K. & Takahara, F. 1989, PASJ, 41,873
Nagase, F. 1989, PASJ, 41, 1
Norman, C., & Scoville, N. 1988, ApJ, 332,124
Pierre, M., 1992, in X-ray Emission from Active Galactic Nuclei and the X-ray Back-
ground, Proc. Coni. at MPE, Nov. 1991, Brinkmann, W., & Triimper, J., eds., p. 341,
MPE report 235
Primini, F. A., Murray, S. S., Huchra, J., Schild, R., Burg, R :, & Giacconi, R. 1991, ApJ,
374, 440
Rappaport, S. A. & Joss, P. C. 1983, in "Accretion-Driven Stellar X-ray Sources", eds.
W. H. G. Lewin & E. P . J. van den Heuvel, p. 1
Rogers, R. D., & Field, G. B. 1991, ApJ, 366,22
Sarazin, C. 1986, Rev. Mod. Phys., 58, 1
Schaaf, R., Pietsch, W., Biermann, P. L., Kronberg, P. P., & Schmutzler, T. 1989, ApJ,
336,722
Schaeffer, R. P., & Silk, r. 1990, ApJ, 360, 483
Schaeffer, R., & Fabian, A. 1983. Early Evolution of the Univer6e and it6 Pre6ent Struc-
ture, Proc. [AU Symposium No. 104, p. 333, eds. Abell, G., & Chincarini, G., Reidel,
Dordrecht
Schmidt, M. & Green, R. F. 1986, ApJ, 305,68
Schwartz, D. A. 1992, in The X-ray Background, Barcons, X., & Fabian, A. C., eds., p.
149, CaInbridge University Press
Schwartz, D. A. & Tucker, W. H. 1988, ApJ, 332, 157
Serlemitsos, P. J., Marshall, F. E., Petre, R., Jahoda, K., Boldt, E. A., Holt, S. S.,
Mushotzky, R. F., Swank, J., Szymkoviak, A., Kelley, R., and Loewenstein, M., 1991,
Frontiers of X-ray A6tronomy, Proc. 28th Yamada Conf., Y. Tanaka, & K. KoyaIna,
eds., p. 221, Universal Academy Press, Tokyo
Setti, G., 1990. in Proc. IAU Symp. no. 139 "Galactic and Extragalactic Background
Radiations", eds. S. Bowyer & Ch. Leinert (Kluwer AcadeInic Publishers), p.345
Setti, G., & Woltjer, L. 1979, A&A, 76, L1
Setti, G., & Woltjer, L. 1989, A&A, 224, L21
Shanks, T., Georgantopoulos, 1., Stewart, G. C., Pounds, K. A., Boyle, B . J., & Griffiths,
R. E. 1991, Nature, 353, 315
Stewart, G. C., Fabian, A. C., Terlevich, R. J., & Hazard, C. 1982, MNRAS, 200, 61
Stocke, J. T., Wurtz, R., & Kuhr, H. 1991, AJ, 102, 1724
Stocke, J. T., Morris, S. L., Gioia, 1. M., Maccacaro, T., Schild, R., Wolter, A., Fleming,
T. A., & Henry, J. P . 1991, ApJS, 76, 813
Tanaka, Y. 1992, in X-ray Emission from Active Galactic Nuclei and the X-ray Back-
ground, Proc. Coni. at MPE, Nov. 1991, Brinkm.ann, W., & Triimper, J., eds., p. 203,
MPE report 235
190

Tananbaum, H., Avni, Y., Branduardi, G., Elvis, M., Fabbiano, G., Feigelson, E., Giacconi,
R., Henry, J. P., Pye, J. P., Soltan, A., &; Zamorani, G. 1979, ApJ, 234, L9
Trinchieri, G., Fabbiano, G., &; Romaine, S. 1990, ApJ, 356, 110
Turner, M. J. L., Williams, O. R., Saxton, R., Stewart, G. C., Courvoisier, T. J-L., Ohashi,
T., Makishima, K., Kii, T., &; Inoue, H. 1989, Two Topics in X-ray A6tronomy: 1101.2:
AGN and the X-ray Background, Proc. ~3rd ESLAB Sympo6ium, p. 769, eds. Hunt,
J., &; Battrick, B., ESA, Noordwijk
Treyer, M. A., Mouchet, M., Blanchard, A., &; Silk, J. 1992, A&;A, 264, 11
van den Heuvel, E. P. J. 1983, in "Accretion-driven Stellar X-ray Sources", eds. W. H. G.
Lewin &; E. P. J. van den Heuvel, p. 303
van Paradijs, J. 1983, in "Accretion-driven Stellar X-ray Sources", eds. W. H. G . Lewin
&; E. P. J. van den Heuvel, p. 189
Wang, Q., &; McCray, R. 1993, ApJ (Letters), in press
Ward, M. J. 1988, MNRAS, 231, 1
Warwick, R. S., Pye, J. P., &; Fabian, A . C. 1980, MNRAS, 190, 243
Watson, M. G. W., Stanger, V., &; Griffiths, R. E. 1984, ApJ, 286, 144
Weedman, D. W. 1986, in "Star Formation in Galaxies", NASA CP-2466, ed. Carol J .
Lonsdale, p. 351
Wilkes, B.J., &; Elvis, M. 1987, ApJ, 323, 243
Wu, X., Hamilton, T., Helfand, D. J., &; Wang, Q. 1991, ApJ, 379, 564
Zdziarski, A., Zycki, P. T., Svensson, R., &; Maciolek-Niedzwiecki, A. 1993a, ApJ, 405,
125
Zdziarski, A. A., Zycki, P. T., &; Krolik, J. 1993b, preprint
 THE EVOLUTION OF THE INTERGALACTIC MEDIUM
AND LYMAN-ALPHA CLOUDS: COMPARISON OF
THEORIES AND OBSERVATIONS

STANISLAW BAJTLIK
Joint Institute for Laboratory Astrophysics, University of Colorado and National
Institute of Standards and Technology, Boulder, CO 80309-0140

Abstract. The basic observational facts known about the intergalactic medium in both
smoothly distributed (IGM) and clumped forms (Lya clouds) are reviewed. Six candidate
theories of Lya clouds and the IGM are compared with the essential non-controversial
observational facts. Deficiencies of the theories and inconsistencies in the existing data sets
are stressed. The three most plausible candidate theories successfully explain some aspects
of the IGM, but fail to explain others. A new approach, employing hydrodynamic N-body
simulations is presented. In this approach elements of several theories are combined in an
effort to study the evolution of the IGM and the formation and properties of Lya clouds.
Preliminary results of such simulations are presented and q>mpared to observations.

1. Introduction

The main motivation for studying the intergalactic medium in both diffuse
(IGM) and clumpy (Lya clouds) forms comes from the increased volume
of observational data, and from the relevance of the properties of the IGM
and (Lya clouds) to the process of galaxy formation. Using remote, high-
redshift quasars as flashlights shining through the intergalactic medium, we
can probe the Universe on much larger scales (up to a redshift z ~ 5) than
allowed by existing catalogs of galaxies. Lya clouds are located at higher
redshifts than the observed galaxies and are much more numerous than
quasars themselves (Figure 1). They can be used both as tracers of the
matter distribution at high-z through vast regions of space, and as probes
of physical properties of the Universe at early epochs.
The last decade was a period of very rapid advancement in our knowledge
and understanding of the intergalactic medium, both in its smooth (IGM)
and clumpy (Lya clouds) forms. High-resolution quasar spectra revealed
properties of narrow absorption lines in unprecedented detail, and provided
the large and homogeneous data sets for studies of their statistical distri-
butions. The spectra have also consistently shown no Gunn- Peterson effect
(Gunn & Peterson 1965) up to the highest observed QSO redshifts, raising
an intriguing question as to when and how the Universe became reionized
(after it recombined for the first time at redshift z ~ 1500). Lack of de-
tectable neutral hydrogen in intergalactic space at redshifts corresponding
to the location of the most remote observed QSO suggests either a very
early and efficient formation of compact objects (even before formation of
the most remote QSO) with very little intergalactic material left, or an early

191
1. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution o/Galaxies, 191-212.
© 1993 Kluwer Academic Publishers.
192

Fig. 1. Quasar light passing through the IGM and Lya clouds probes the Universe
at high redshifts and on large scales.

release of a large amount of ionizing radiation, or both. This problem is still

unsolved.
Sargent et al. (1980, hereafter SYBT) presented solid evidence for the
interpretation of narrow absorption features located blueward of QSO Lya
emission line as caused by intervening clouds of intergalactic material of pri-
mordial (no detectable metal lines at corresponding redshifts) composition.
They also established that the spatial distribution of clouds was random (no
detectable spatial correlations). Other properties such as the distribution of
observed Doppler broadening parameters and corresponding column den-
sities of neutral hydrogen were studied at length. Spectra of lensed QSOs
(only two systems so far) have revealed the only direct information about
the physical characteristics of clouds - their size, or rather lower limits on
their size.
The basic theoretical problems that theories of Lya clouds face are:
the confinement mechanism, the observed range of column densities of HI,
Doppler broadening parameters, and rest frame equivalent widths, the lack
of spatial correlations and constraints on their sizes.
Attempts to model Lya clouds as self-gravitating spheres (Black 1981)
failed because clouds of the required range of column densities would be
unstable on too short time scales. Soon other models and interpretations
followed. Ostriker and Ikeuchi (Ostriker & Ikeuchi 1983; Ikeuchi & Ostriker
1986), following a suggestion by SYBT, created an elaborate theory of in-
tergalactic clouds stabilized by the pressure of an external, rarefied but hot
intergalactic gas (IGM). The virtue of this theory was that it provided sev-
eral observational tests and predictions about both the IGM and clouds
themselves. However, the theory lacks a clear picture of the process of for-
mation of Lya clouds and ionization of the IGM.
Early successes of the Cold Dark Matter theory (CDM) created an inter-
 193

est in embedding the theory of formation and evolution of the IGM and Lya
clouds into the broader theory of the formation of large scale structure of the
Universe. In this picture clouds are stabilized by the gravity of dark matter
mini-halos (Rees 1986, 1988), or are low mass objects developing from small
mass fluctuations in the tail of the CDM spectrum (Bond, Szalay, & Silk
1988). This scenario offers an explanation of the formation of clouds. The
large number of free parameters, however, made it difficult to compare this
theory to the observations.
Another solution of the confinement problem is no confinement at all.
Two classes of models have assumed clouds are not confined. The interest in
freely expanding clouds was revitalized after the publication of observations
of very narrow (less then 10 km s-1) absorption lines (Pettini et al. 1990,
hereafter PHSM). Attributing the line widths to thermal broadening leads
to very low cloud temperatures, below 104 K, impossible to attain by clouds
in thermal equilibrium with an external UV flux. However, freely expanding
clouds cooling adiabatically can stay highly ionized and attain very low tem-
peratures at the same time (Duncan, Vishniac, & Ostriker 1991, hereafter
DVO).
Chernomordik & Ozernoy (1983) proposed a model of dynamically evolv-
ing clouds as shells of expanding intergalactic matter. The theory made a
clear prediction that QSO spectra should show many pairs of lines separated
by a characteristic scale.
The existence of very narrow (::; 10 km S-1) lines stimulated the idea
of very small, very dense, very cool (::; 104 K), almost neutral clouds -
"Jupiters" (PHSM). Such a model allows clouds to be cool without invoking
an adiabatic cooling mechanism. However, the model has trouble explaining
the sizes of clouds inferred from lensed quasar spectra. This talk is not
intended as a systematic review of observations and theories of the IGM
and Lya clouds. Two recent reviews of the subject are by Ray Weymann
and Jill Bechtold in this volume. My intention is rather to summarize briefly
the most basic observational data and the most frequently discussed theories.
In §2, I will try to convince you that even the simplest observational
facts are still a subject of debate and controversy. Similarly, there exists
no clear theory explaining all the fundamental properties of the IGM and
intergalactic clouds, including the process of their formation and evolution.
In §3, I will list the most frequently discussed theories and their underlying
assumptions. Section 4 compares the predictive power of several (classes of)
theories with observations.
In §5 a new approach to the problem of formation, evolution and physical
properties of the IGM and Lya clouds will be proposed. I will present the
first preliminary results of hydrodynamic N-body simulations of the IGM
and intergalactic clouds. In this approach the IGM and clo!1ds are treated as
one medium. Only after the simulation is complete do we try to recreate the
194

observational procedure leading to the distinction between discrete clouds

(showing up in QSO spectra as Lya absorption lines) and smooth IGM
(detectable in QSO spectra through the Gunn-Peterson absorption trough).
In this approach elements of several of the theories mentioned above are
combined with no prejudice as to which is more plausible. Section 6 contains
a comparison of the results of these hydrodynamic N-body simUlations with
the observations. Finally, in §7 I summarize our conclusions.

2. Observations
In this section I will summarize briefly the essential observational facts about
the IGM and Lya clouds. Those facts are considered well established, but I
will discuss disturbing inconsistencies and contradictions. It is important to
realize the status of these observations since they are to be compared with
the models and theories.

2.1. REDSHIFT EVOLUTION OF THE NUMBER DENSITY OF CLOUDS

After years of study and debate since Lya clouds were discovered (Lynds
1971), the evolution of the number of absorbers per unit redshift has been
established as following a power law,
dN
-
dz
= Ao(l + z)"!, (1)

The main effort has been concentrated on determining the value of the evolu-
tion parameter I' The value of 1 corresponding to a nonevolving population
of absorbers is in the range between 0.5 (for cosmological deceleration pa-
rameter qo = 0.5) and 1 (for the empty Universe qo = 0). For a long time the
value of 1 was a subject of controversy, with different observers determining
(from different data sets) values as disparate as 1 = -2.1 and 1 = 2.36. At
least part of the problem was the prozimity effect, which will be discussed
later. For a review of this subject see Hunstead (1988). The exclusion of
regions affected by the proximity effect - a deficiency of clouds in the prox-
imity of a quasar - enabled a value of 1 to be determined consistently
(within the random errors) between different observers and data sets. This
"standard" value of "'I ::::: 2.3 ± 0.4 (Murdoch et ai. 1986; Tytler 1987; Bajt-
lik, Duncan, & Ostriker 1988, hereafter BDO) proved that there is strong
evolution of the population of intergalactic clouds, and predicted a very low
number of Lya clouds at the present time. This result was obtained from a
sample of lines complete to an equivalent width threshold of Wthr 2: 0.32 A.
The satisfaction in achieving agreement between different observers was
undermined however by recent determinations of "'I values as divergent as
"'I = 2.75 ± 0.29 (Lu, Wolfe, & Turnshek 1991) from a compilation of 38
QSOs and "'I = 1.89 ± 0.28 (Bechtold 1993) from a homogeneous sample
 195

of 34 QSOs. Both results were for strong lines with rest frame equivalent
widths exceeding Wthr = 0.32 A. Bechtold found an even more disturbing
result from a sample of lines complete to Wth,. = 0.16 A in her 34 QSO
spectra. This sample yielded ; = 1.32 ± 0.24, a striking result consistent
with almost no evolution.
A power-law extrapolation (2.1.1) with Bechtold's small value of; pre-
dicts more clouds at Z ~ 0 than previously thought, consistent with re-
cent results from the Hubble Space Telescope (Morris et al. 1991; Bahcall
et al. 1991, 1992). HST data for 3C273 (zem = 0.158) and H1821+643
(zem = 0.297) indicate dN / dz ~ 13 ± 5 lines per unit redshift at z ~ o.
The value ; = 1.89 predicts a density at z ~ 0 only '" 10" lower than
observed.

2.2. THE DISTRIBUTION OF COLUMN DENSITIES OF NEUTRAL HYDROGEN

One of the amazing properties of the population of intergalactic absorbers

is the wide range in their column densities of neutral hydrogen. The weakest
absorbers have NHI ~ 10 12 •8 cm- 2 , while the strongest metal line systems
show NHI ~ 10 22 cm- 2 (Smith, Cohen, & Bradley 1986). Approximately
99% of intergalactic absorption is due to the Lya systems (i.e. systems with
no metal lines). Only about 1% of intergalactic absorption systems show
metal lines. Systems with NHI ::; 10 17 cm- 2 are predominantly Lyo clouds
while higher column densities typically correspond to metal systems.
This 9 orders of magnitude span of values of N HI is fairly well described
by a simple power law,

dN N- f3 (2)
dNHI ex: HI'

first established by Carswell et al. (1984). High-resolution studies gave the

value j3 ~ 1.7 (Carswell et al. 1984; Atwood, Baldwin, & Carswell 1985;
Hunstead et al. 1987; Tytler 1987; Rauch et al. 1992). Tytler (1987) argued
on observational grounds in favor of an unbroken power law (2.2.1) and
interpreted it as evidence for a single population of intergalactic absorbers.
This clear picture of an unbroken power law (2.2.1) of column densities
over 9 decades was shaken by the question: why is there no feature in the
distribution for NHI 2:: 1017 cm- 2 ? At such column densities clouds become
optically thick and should form a (nearly) neutral core. This effect was
studied by Duncan & Ostriker (1988), and in more detail by Bajtlik, Duncan,
& Ostriker (unpublished). Their analysis of radiative transfer through a
spherical cloud of H and He confirms the formation of a neutral core. The
observational picture is complicated by the fact that a line of sight to the
background QSO may intercept a cloud either through its neutral core or
through its ionized mantle. Figure 2 presents the results of these studies for
196

HARD IONIZING
dN SPECTRUM
log--
dNHI
SOFT "
IONIZING __: >~' __-BELATED TO PIGM
SPECTRUM :,: " , -AND X(CLOUD
, NEUTRAL
FRACTION)
10 17
log N HI

Fig. 2. Theoretical distribution of column densities of HI. The feature around

NHr 2: 1017 cm- 2 arises as clouds became optically thick. The dotted line cor-
responds to a soft (black body) ionizing external flux and an area average approx-
imation to the geometry. The solid line represents a more realistic model with 'a
hard (power law, slope -1) spectrum and an exact treatment of the geometry of
spherical clouds. The upward shift ofthe power law at NHr 2: 10 18 . 5 cm- 2 relative
to that at N H 1 ~ 10 17 cm - 2 depends on the internal pressure of the cloud and the
neutral fraction in its mantle X.

both soft and hard external ionizing flux, and for both exact (integration
over a range of line of sight impact parameters) and approximate (area
average column density approximation) treatments of geometry,
Data published by Bechtold (1988) seemed to show a break, confirming
the theory of optically thick clouds. However, subsequent work by Sargent,
Steidel, & Boksenberg (1989) showed no break, reaffirming Tytler's (1987)
earlier study. Part of the problem is that the range of N H I of interest corre-
sponds to a saturated part of the curve of growth. The horizontal error bars
in this region of Figure 2 are very large and make comparison of theory and
observations ambiguous.

2.3. DISTRIBUTION OF VELOCITY BROADENING PARAMETERS

The distribution of velocity broadening parameters b is well concentrated
around 25 rv 30 km S-1. The width of this distribution is quoted to be '" 10
km s-1 for either Gaussian or top hat models of the distribution (PHSM).
For a long time researchers agreed on the lack of any significant correlation
between velocity broadening b and column density NHI (BDO).
Quite recently Pettini et aI. (PHSM) reported discovery of very narrow
 197

absorption lines in a 6 km S-l resolution echelle spectrum of Q2206-199N

(zem = 2.559, B = 17.5). The reported values of b in some cases were 10 :s
km s-1, implying temperatures Tc :s
104 K. PHSM also found a strong
b - NHI correlation, which can be described by a simple fit (Bajtlik &
Duncan 1991),

b = [11 + 14(logNHI - 13)J km s-1 , (3)

:s
The detection of lines with b 10 km S-l prompted interest in theoretical
models of small, cold, dense, neutral ("Jupiter") clouds (PHSM) and of
nonequilibrium, freely expanding clouds (DVO).
A disturbing fact is that other high-resolution echelle studies of QSO
:s
spectra have not confirmed the existence of lines with b 10 km S-l. Obser-
vations of Q1100-264 (zem = 2.14,B = 16.0) by Carswell et al. (1991) and
of QOOOO-263 (zem = 4.11,m r = 17.5) by D'Odorico, Molaro, & Savaglio
(1991) showed neither b:S 10 km s-1 lines nor a b - NHI correlation. They
also concluded that the mean value of b was close to the conventional value
b rv 30 km s-l. Even more disturbing is that the observations of Q2206-199N
and QOOOO-263 were made with the same instrument - the University Col-
lege London echelle spectrograph at the AAT. For a brief summary of this
controversy see Webb & Carswell (1991), Hunstead (1991) and D'Odorico,
Molaro, & Savaglio (1991).

2.4. DISTRIBUTION OF REST FRAME EQUIVALENT WIDTHS

The number of lines in a QSO spectrum depends on the limiting equivalent

width W irn the rest frame. This dependency is a strong one - exponential,
of the form,
dN ex: e -W/W.
- (4)
dW '
so that the distribution in Z and W takes the form

{J3N _ Ao -w/w·( )'Y (5)

8z8W - W. e 1+ z .

The value of the characteristic rest frame wavelength W· is found to be in the

range from about 0.25 A to 0.35 A (SYBT; Young, Sargent, & Boksenberg
1982; MHPB; Weymann, this volume).
Jenkins, Bajtlik, & Dobrzycki (see Jenkins 1991) and Barcons & Webb
(1991) pointed out that the value W· in the range 0.25 - 0.35 A derived
from the rest frame equivalent width limited samples is inconsistent with
the power law index j3 in equation (2) being in the range j3 ~ 1.5 - 1.9. For
an individual absorption line, values of Wand N HI must correspond to a
point on a curve of growth characterized by a single value of the parameter
198

b. Similarly, for a population of lines described by a distribution of N H I of

the form of Equation (2) and a distribution of W described by Equation 4,
a set of (W, NHI) points must be consistent with the set of curves of growth
corresponding to the known distribution of b values. The resolution of this
disturbing inconsistency is not known.

2.5. SIZE OF LYa CLOUDS

Analysis of lensed QSOs or of pairs of nearby QSOs provides a way to

estimate the physical size of intergalactic absorbers. Until now only two
systems were found to be bright enough and close enough to be useful for
this kind of study.
From high-resolution observations of Q2345+007 A & B (zero = 2.17,
images separated by () = 7".3), Foltz et al. (1984, hereafter FWRC) derived
a limit on cloud sizes of 5 :s; R :s; 25 kpc for clouds at Z ~ 2. FWRC
analyzed the coincidences of absorption in both A and B components of
the Q2345+007 system. Bajtlik & Duncan (1991) analyzed correlations not
only in the positions of absorbers but also in the relative strengths of the
lines in both spectra. This kind of analysis can be used (in principle) to
test particular theories of cloud internal structure. For pressure confined
spherical clouds Bajtlik & Duncan estimated the best fit typical cloud size
R to be R ~ 14 kpc (Ho = 75 km s-l Mpc- 1 ) .
. Steidel & Sargent (1990) have identified most of the absorption lines in
Q2345 + 007 to be metal lines, which if true would make the system useless
for determining cloud sizes. Such a high fraction of metal systems in the
spectrum of this QSO would make it highly unusual.
The only firm limit on cloud sizes comes from a high-resolution spectrum
of the gravitationally lensed object UM 673 at Zem = 2.727 (Smette et al.
1992). Unfortunately, the two lensed images are separated by only () = 2".2,
too close to provide a strong upper limit on cloud sizes R. Almost all - 65
out of 68 - absorption lines are common to both spectra. This translates
to a firm lower limit of R 2: 6 kpc and a weak upper limit of R :s; 80 kpc.

2.6. CORRELATIONS, VOIDS

One of the characteristics distinguishing the population of Lya absorbers

from metal line systems is their clustering properties. Metal line systems
have been proven to cluster on velocity scales of a few hundred km s-l
(SYBTj Sargent, Boksenberg, & Steidel 1988). Lya systems have been shown
by many researchers to show no clustering even down to comoving scales of
a few Mpc, somewhat smaller than the galaxy correlation length.
Extensive efforts to detect structure in the spatial distribution of clouds
have yielded some evidence of correlation but no conclusive confirmation of
 199

its existence. Webb (1987) claimed a positive cloud-cloud correlation func-

tion on velocity scales between 50 and 200 km s-l. Ostriker, Bajtlik, &
Duncan (1988) used a different statistic - the distribution of intervals be-
tween lines - to search for voids in the cloud distribution. They found an
upper limit on the void filling factor of less than 10 - 15% for voids with
comoving sizes in the range of between 5 and 50 Mpc. Crotts (1987, 1989)
studied a single region in the spectrum of Q0420-388 devoid of any absorp-
tion lines and interpreted it as a void of comoving size 40 Mpc. Dobrzycki &
Bechtold (1991) have found another example of a single void in the spectrum
of Q0302-003 (zem = 3.285). This void corresponds to a co moving size of 9
Mpc (Ho = 50 km s-l Mpc- 1 ).
The significance of the voids and correlations found so far remains an
open question requiring further analysis. The issue is of great importance for
relating intergalactic clouds to galaxies, and their formation to the formation
of large scale structure in the Universe.

2.7. PROXIMITY EFFECT, J",

Carswell et al. (1984) in a sample of 7 QSOs, and Murdoch et al. (1986)

and Tytler (1987) in a sample of 19 QSOs found a deficiency of lines in
the vicinity of QSOs. Bajtlik, Duncan, & Ostriker (BDO) attributed this
effect to the enhanced ionization of clouds in the proximity of QSOs (the
prozimz'-y effect). They modeled the ionization state of clouds as a function
of the ratio of the intensity of QSO flux (observable) to the intensity of
intergalactic background flux (unknown). This method effectively allows one
to determine the intensity of UV flux at high z. The resulting value is J", =
10- 21 erg cm- 2 s-l Hz- 1 sr- 1 , approximately 10 times more than expected
from extrapolating the luminosity function of observed QSOs.
This result was confirmed by Lu, Wolfe, & Turnshek (1991) but an in-
triguing puzzle was the lack of the expected correlation between the strength
of the proximity effect and QSO luminosity. In a recent study Bechtold
(1993), using a sample of QSOs covering a much wider range of QSO lumi-
nosity, not only showed that such a correlation exists but also found an even
higher value J", = 3 X 10- 21 erg cm- 2 s-l Hz- 1 sr- 1 •
The problem of the origin of the metagalactic ionizing flux is still un-
solved. The most recent developments in this field are, first, Sciama's hy-
pothesis of decaying neutrinos, as summarized in Sciama (1990) and elab-
orated in Sciama (1991), and, second, Teresawa's attempt to explain the
high-intensity UV background as coming from low luminosity AGNs (Tere-
sawa 1992).
200

2.8. IONIZATION OF THE IGM

The question of the origin of the extragalactic UV background flux at high-z

is closely related to the problem of the ionization of the intergalactic medium.
The Gunn-Peterson test (Gunn & Peterson 1965) provides a remarkably
sensitive test for the detection of diffuse neutral hydrogen in space. The
optical depth for extinction by scattering of UV photons off neutral hydrogen
atoms is,

(6)

where !lIGM is the IGM contribution to the total cosmological density pa-
rameter !l, h = (Ho/100) km s-l Mpc- 1 , f is a clumping factor (2: 1), a is
the spectral index of the ionizing radiation, z is the cosmological redshift and
J -21 is the background UV flux in units of 10- 21 erg cm- 2 s-l Hz- 1 sr- 1 .
At redshift z ~ 4.7 (eq. 6) translates to,

(7)

From the spectrum of Q1158+4635 at Zem = 4.73, Webb et al. (1992) have
found TGP ~ 0.04. This strong limit implies either a very small value of
{hGM (at z ~ 5 matter must already be in compact objects) or a very high
value of J v (again ----; there must exist compact objects to emit this radiation
unless there is some other mechanism to produce UV flux early enough, such
as in Sciama's neutrino decay hypothesis).

2.9. FORMATION OF LYa CLOUDS: EVOLUTION OF THE IGM

The reverse evolutionary trend in a (eq. 1) power law, i.e., a decrease in

the number of clouds per unit redshift would be a clear indication that we
are observing the epoch of the formation of intergalactic clouds. Since this
would mean material (presumably neutral) is left in space in a diffuse form
we would expect to detect a non-zero optical depth from the Gunn-Peterson
effect.
Nothing of this kind has ever been observed. The simple power law (eq. 1)
persists up to the highest observed redshifts (z 2: 4), and no positive Gunn-
Peterson effect has been detected. This is puzzling. Most galaxy formation
scenarios predict formation of galaxies at relatively late epochs (later than
z ~ 4). Moreover QSO counts show a decline in QSO comovingnumber den-
sity for z > 2.3. If, like galaxies and QSOs, intergalactic clouds are formed at
some particular epoch, we would expect to see their number density decline
at the corresponding redshift.
 201

3. Models
The confusing and often contradictorY_, character of the data on Lya clouds
has contributed to the plethora of theories on these objects. Some theories
aim to explain all properties of the population of intergalactic absorbers,
while others concentrate on specific aspects. In this section we list the most
frequently invoked theories and their basic assumptions and characteristics.
Ostriker (1988) has reviewed the fundamental physical problems that theo-
ries of intergalactic clouds must face.

3.1. SELF-GRAVITATING CLOUDS

Black (1981) proposed that Lya clouds are confined by their own self-gravity.
The problem with this model is that self-gravitating (gaseous) isothermal
spheres are unstable to collapse or expansion. Furthermore, the estimated
sizes of clouds suggest that gravity should be unimportant in most cases.

3.2. DARK MATTER CONFINEMENT

Rees (1986, 1988) proposed that the gravity of dark matter mini-halos is
what confines intergalactic clouds. This idea is closely related to a theory
proposed by Bond, Szalay, & Silk (1988), who suggested that Lya clouds
would form in the framework of the CDM galaxy formation scenario. In
this picture clouds condense in the gravitational potentials of small mass
fiuctuations (in the low mass tail of CDM density fiuctuations spectrum).
To account for the observed evolution of cloud density with redsb,ift, this
scenario requires tuning of the evolution of the UV background flux. ijeating
by UV photons would eventually cause clouds to expand and evaporate,
hopefully in agreement with the observed trend (eq. 1). In a~dition to this
fine tuning problem is the problem of the observed random distribution of
clouds. If clouds formed in CDM fiuctuations, one would expect to observe
cloud-cloud correlations, in much the same way that galaxies are correlated.

3 .3. PRESSURE CONFINED CLOUDS

Ostriker & Ikeuchi (1983) and Ikeuchi & Ostriker (1986) considered the
idea that intergalactic clouds are stabilized by the external pressure of a
hot intergalactic medium. In this picture clouds are stable, highly ionized
(neutral fraction X ::; 10- 4 ), about 100 times denser than the surrounding
medium, and cloud temperatures are Tc r-.J3 X 104 K. The free parameters
of the model can be adjusted so that it satisfies limits on the temperature
of the IGM from observations of the X-ray background. This theory still
provides the most elaborate and verifiable model of Lya clouds; its most
202

obvious deficiency is that it makes no attempt to explain the formation of

clouds .

3.4. FREELY EXPANDING CLOUDS

It is not clear whether clouds have to be confined at all. In principle clouds

could be expanding freely from an earlier epoch. Let us try to understand
the confinement problem, without necessarily worrying about the origin of
clouds. The interest in the theory of freely expanding clouds was revitalized
by the discovery of narrow absorption lines (s 10 km S-1). The implied
temperatures (Tc S 10 4 K) cannot be explained by clouds in thermal and
ionization equilibrium. Duncan, Vishniac, & Ostriker (DVO) have shown
that adiabatic cooling of expanding clouds can lead to very low cloud tem-
peratures and clouds could still maintain a high ionization state.
Duncan, Vishniac, & Ostriker's scenario is related to Bond, Szalay, &
Silk's theory. In both theories, clouds are stable at an earlier epoch but,
heated by UV background photons, and surrounded by an adiabatically
expanding external medium, they gradually go out of equilibrium and evap-
orate.
The viability of the theory rests on the reality of the b S 10 km S-1 lines
and the b - NHI correlation. The theory also predicts that the rest frame
equivalent widths of lines should evolve with redshift (narrow lines appear
predominantly at lower z).

3.5. EXPANDING SHELLS

Chernomordik & Ozernoy (1983) proposed that Lya absorption lines are
caused by shells of expanding intergalactic matter. This idea is reminiscent
of the explosive galaxy formation scenario proposed by Ostriker & Cowie
(1981) . The theory predicts that lines should be paired on velocity scales
corresponding to the size and bulk velocity of the shells. In practice, this
pattern is complicated by geometry - the distribution ofline-of-sight impact
parameters. Nevertheless extensive studies of the line distribution (see §2.6)
failed to reveal this sort of pattern.

3.6. "JUPITERS"

The discovery of very narrow Lya lines with b S 10 km s-1 (§§2.3 and
3.4) stimulated consideration of the possibility that clouds are actually very
small ("Jupiter" size), dense, cold (S 104 K), and almost completely neutral
(PHSM). In this model the proximity effect would be explained by the fact
that clouds close to the QSO would not be able to cover the emission region
of the QSO, and would disappear from the spectrum. The major defect of
 203

this model is that it conflicts with the limits on cloud sizes inferred from
observations of lensed QSOs (§2.5).

4. Comparison of Theori~s with Observations

Some of the theories mentioned above were intended to explain only some
of the observed properties of intergalactic clouds. Some failed basic obser-
vational tests and/or were never studied in detail.
Self-gravitating clouds fail to pass the stability test. Expanding shells of
intergalactic matter have not survived the extensive studies of the spatial
and velocity distribution of absorbers. The "Jupiters" model, although never
studied in detail, fails to pass the size limit inferred from gravitationally
lensed QSOs.
So, we are left with three candidate models of Lya clouds: pressure con-
fined clouds; clouds bound by gravity of dark matter haloes; and freely ex-
panding clouds. The three models have several properties in common: clouds
are formed at high z in a stable form, are later heated by UV background
photons, and undergo expansion, evaporation and eventual destruction. All
three models can be made to achieve agreement with the observed power-
law evolution (eq. 1) of cloud number density. Other common properties are
the high degree of ionization and large sizes (of order of galactic size).
An objection frequently raised against the pressure confined cloud model
is that to account for the nine decade range of observed column densities
NHI, clouds would have to cover a huge range of masses. This problem
can easily be solved by supposing that the intergalactic pressure varies by
approximately 2 orders of magnitude. The required range of cloud masses can
then be constrained to approximately 2.5 orders of magnitude, specifically
Me ~ 10 6 - 10 8 . 5 Mev (Bajtlik, Duncan, & Ostriker 1988).
The three theories differ however in important aspects. They invoke differ-
ent confinement mechanisms (pressure, gravity, or both), and clouds expand
and disappear at different rates and for different reasons (heating, adiabatic
evolution of external intergalactic medium, or both). The theories also of-
fer different observational tests (correlated or random spatial distribution,
range of sizes, distributions of column densities N HI, values of the Doppler
broadening parameter b, rest frame equivalent widths W, evolution of the
number density of absorbers with redshift, etc.).
In Table I, we summarize briefly and schematically the current status
of the comparison between theories and observations. The table lists the
six theories discussed in §3 and rates them them against 10 observational
tests (or rather classes of tests). A plus sign means the theory successfully
passes the test. A question mark can mean either that data are inconclusive,
contradictory, or not accurate enough (for instance: structure in the cloud
distribution or the existence of very narrow lines), or else that the theory
204

TABLE I
Theories of Intergalactic Clouds vs. Observations

OBSERVATIONS MODEL
Pressure CDM Freely Self Exp.
Confined Halos Exp. Grav. Shells Jupiters
Stability + + - +
dNjdz, low z lines + +? +
dNjdNHI, (3, NHI '" +? + +
10 13_10 17
b distribution +? +? +
dnjdw + +
R-size ? + + -
Correlations, voids, + ? ? -
struc-ture
Proximity effect J - 21 + +? + +
Ionization of IGM, G-P, + ?
X-ray background
Formation (origins) - +? -

fails some tests but not mortally, that it needs fine tuning of free parameters
(for example, special adjustment of the evolution of UV background flux in
CDM based theories, or IGM pressure fluctuations in models of pressure
confined clouds), or that the basic underlying assumptions are questionable
for reasons unrelated to either the IGM or Lya clouds themselves (such as
general problems with CDM - see articles by Bond, Davis, and Evrard in
this volume). Finally, a minus sign is reserved for those cases where either a
theory contradicts the data (stability of self-gravitating clouds, correlations
for expanding shells, size for "Jupiters"), or a theory makes no prediction
(the formation of clouds in the pressure confined and freely expanding cloud
models).

Three theories (pressure confinement, CDM halos, and freely expanding

clouds) pass successfully or relatively well nine direct comparisons with the
observations ("+" or "?" marks), although two of these theories do not
address the problem of the origin of clouds (bottom line in Table I). As we
do not yet have a complete theory of the formation of galaxies and of large
scale structure of the Universe, it would perhaps be unrealistic to demand
a clear theory of the formation of Lya clouds.
 LOS

5. Hydrodynamic N-body Simulations of the IGM and Lyo:

Clouds
The limited ability of analytic theories to account for the complexity and
wide range of observed properties of Lyo: clouds and the IGM, combined
with rapid progress in numerical techniques, have motivated attempts to
study the evolution of the intergalactic medium as a whole (in both clumpy
and diffuse forms) from first principles by means of hydrodynamic N -body
simulations.
The advantage of this approach is that there is no necessity to assume a
priori that clouds are all pressure confined, or all expanding, or all in the
potential wells of dark matter halos. Nor is there a need to constrain the
analysis to particular (e.g. spherical) cloud geometries.
In this and the next section I will present results of research done in
collaboration with Renyue Cen, Jordi Miralda-Escude and Jerry P. Ostriker.
We use the output of a hydrodynamic N-body simulation performed by Cen
and described in detail by Cen (1992), and Cen & Ostriker (1992a,b).

5.1. METHOD

The simulations used an Eulerian hydrodynamic code to evolve both dark

matter and baryonic (Le. observed) components of the Universe. The code
tracks the evolution of the temperature, velocity, and abundance of six dif-
ferent species: e-, HI, HII, Hel, Hell, Hell!.
The following heating processes were included:
photoionization heating (HI, Hel, Hell);
compton heating (for matter temperature lower than the background
radiation temperature);
shock heating (via artificial viscosity).
Cooling processes included:
collisional ionization cooling (H, He, Hell, He( 2 3 5»;
recombination cooling (HII, Hell, Helll);
dielectric recombination cooling (He);
collisional excitation cooling (H (all n), Hell (n = 2), Hel (n = 2,3,4
triplets );
bremsstrahlung cooling (all ions);
compton cooling (by microwave background photons and by diffuse X-
ray background photons).
The initial power spectrum of density fluctuations was taken to be the
CDM power spectrum, with h = 0.5 (Hubble constant in units of 100 km s-1 Mpc- 1 )
n = 1 (cosmological total density parameter), and fib = 0.06 (baryonic den-
sity parameter). The bias of galaxies relative to total mass was found to
206
a
*

Fig. 3. A cube of 200 x 200 x 200 cells used in the simulations. A line of sight to
the background 'QSO' was chosen at random. Density and velocity fields inside the
cube were used to generate an artificial observed QSO spectrum.

be b = (~N / N)gat/(~M / M)tot ~ 1.6 on the 8h- 1 Mpc scale. The output
corresponds to the epoch of redshift z = 3.
Two cubes of 200 X 200 X 200 cells were studied: one 80h- 1 Mpc on
a side, and the second a zoomed-up fragment 8h- 1 Mpc on a side (Fig-
ure 3). The input UV background flux was taken from Miralda-Escude &
Ostriker (1990) with supernova input parameter value eSN = 10- 4 . 5 • The
flux was roughly constant between redshifts of z = 5 - 3, with J", ~ 10- 21
erg cm- 2 s-1 Hz- 1 sr- 1 •

To model observations of Lya clouds and the IGM, we probed the 80 and
8 Mpc boxes with random lines of sight . Knowing the density, temperature,
and velocity of neutral hydrogen along the line of sight allowed us to create
artificial QSO spectra which could be compared to real spectra of QSOs
with Zem 2 3. Figure 4 illustrates typical results of this procedure. In the
next section we present preliminary results of this analysis and comparisons
with observations.
 207

12

10

----
0 6

0
0 1000 2000 3000 4000 5000
Av [km/s]

Fjg. 4. Examples of QSO spectra obtained from eight different lines of sight crossing
the 80 Mpc box.

6. N -body SiInulations vs. Observations

To achieve a more realistic comparison between the output of the hydro-

dynamic N-body simulations and real observations, we add random noise
to the synthetic spectra created as described above. We treat the resulting
spectra in the same way as observed QSO spectra. That is, we identify ab-
sorption lines, their widths, column densities, and velocity dispersions. The
results of this analysis are presented in Figures 5, 6, 7 and 8.
We also looked for the Gunn-Peterson effect. In our simulations the op-
tical depth TGP due to diffuse neutral hydrogen in space (outside clouds) is
:S 0.04 - again, in agreement with the non-detection in real data. We found
no evidence for Lya clouds being caustics in velocity space as suggested by
McGill (1990a,b). A more detailed determination of the shapes of clouds (or
in general of regions of neutral hydrogen overdensities) will require further
study.
208

1.0

N
"C
~
"C
Cl
o
- 0.5

o
12.0 12.5 13.0 13.5 14.0 14.5
log HI COLUMN DENSITY

Fig. 5. The distribution of column densities of Lya clouds identified in the numerical
simulation. The standard power law dN IdNHI ()( Nii: with {3 = 1.7 is shown. The
bins corresponding to log NHI ::=; 12.8 are (as in the observations) incomplete. Units
on the vertical axis are arbitrary.

1.5

a::
w 1.0
III
~
:::>
z
0.5

20 40 60 80 100
DOPPLER PARAMETER b (km/s)

Fig. 6. The distribution of velocity broadening parameters b. As in the real data,

a Gaussian with b =
28 km s-1 and ( j = 15 km s-1 fits the simulations quite
accurately.
 209

60

--
en ..
- -
-E 40
.::t! .. ..
- -
. - --
-
.0

-
..:

. .. -
. ----
-
_r'

. --
20
. .
--
12.5 13.0 13.5 14.0 14.5 15.0
log N(HI)

Fig. 7. b - NHI correlation . We do not find (in agreement with most observations)
any hint of a b - NHI correlation. Solid line represents b = 28 km S-1. Dashed line
is for b(NHI) in the form of (2.3.1).

7. Conclusions

In the last decade both the volume and the quality of the data on the IGM
and Lya clouds has grown impressively. Unfortunately, there remains much
controversy and inconsistency in the data published by different observers.
There are plenty of theories of the origin, structure and evolution of the
IGM and intergalactic clouds, but often the theories are not specific enough
or have so many free parameters that a conclusive confrontation with the
data is either difficult or impossible.
To make the comparison between theoretical predictions and data more
productive, we need more data to improve statistics, to solve inconsistencies,
to resolve the problem of the existence of spatial structure on small scales,
to investigate the possibility of a break near NHI = 10 17 cm- 2 , and to look
for the Gunn-Peterson effect at z 2: 4. We would also like to get hitherto
unavailable data, such as lensed QSOs with images separated by () ""' 7 - 8"
to infer the cloud sizes. Observations of close systems of bright QSOs would
allow studies of the two QSO proximity effect. Detection of helium lines
would allow a direct measurement of cloud temperatures. Other interesting
observations would be the detection of deuterium (to check for cosmological
210

1.5

-
:>1.0~--~--------~--~r-~+--r+----r---r~--1
oJ../'
+
0.5

o 500 1000 1500 2000 2500

VELOCITY (km/s)

Fig. 8. Correlation function in velocity space ~(v). In agreement with most observa-
tions we do not detect statistically significant correlations. Dashed line represents
10' errors.

abundance at high-z) and observations of emission radiation from clouds.

Theories need to be more specific. Hydrodynamic N-body simulations
should shed light on the complex problem of the formation and evolution
of clouds and the diffuse IGM. Pressure confinement, dark matter gravity,
and free expansion at late epochs probably all play an important role in
determining the structure and evolution of Lya clouds.
The most prominent unanswered questions are: how did clouds form, and
how and when did the intergalactic medium become ionized?

Acknowledgements
I am grateful to my collaborators Renyue Cen, Jordi Miralda-Escude and
Jerry P. Ostriker for allowing me to present results of joint research prior to
publication. I acknowledge stimulating discussions with Jill Bechtold, Mitch
Begelman, Andrew Hamilton and Mike Shull. Andrew Hamilton spent a lot
of time on improving the text of this paper. I thank the meeting organizers,
especially Harley A. Thronson and Mike Shull for providing support which
enabled me to participate in this conference. Special thanks go to the staff
of JILA Scientific Reports Office for assistance with the preparation of the
manuscript. This research was partially supported by the JILA Visiting
Fellowship Program at the University of Colorado in Boulder, the Polish
Committee for Scientific Research through Grant 2-1234-91-01, and NASA
 2tt

through Grant 2448.

The author is permanently at: Copernicus Astronomical Center, ul. Bar-
tycka 18, 00-716 Warsaw, Poland.

References
Atwood, B., Baldwin, J . 'A., & Carswell, R. F . 1985 , ApJ, 292,58 .
Bahcall, J. N ., Jannuzi, B. T., Schneider, D. P., Hartig, G. F., Bohlin, R ., & Junkkarinen,
V. 1991 , ApJ, 377, L5.
Bahcall, J. N ., Jannuzi, B. T., Schneider, D. P., Hartig, G . F. , & Green, R. F. 1992,
preprint.
Bajtlik, S., & Duncan, R . C . 1991, in Proceedings of the ESO Mini-Workshop on Quasar
Absorption Lines, ed. P . A. Shaver, E . J. Wampler & A. M. Wolfe (ESO Scientific
Report No.9 - February 1991),35.
Bajtlik, S., Duncan, R . C., & Ostriker, J. P. 1988, ApJ, 327, 570 (BDO).
Barcons, X ., & Webb, J . K . 1991, MNRAS, 253, 207.
Bechtold, J. 1988, Mount Wilson and Las Campanas, preprint.
Bechtold, J. 1993, ApJS, submitted.
Black, J. H. 1981 , MNRAS , 197, 553 .
Bond, J . R ., Szalay, A. S., & Silk, J. 1988, ApJ, 324, 627.
Carswell, R . F. , Morton, D . C ., Smith, M. G., Stockton, A . N., Turnshek, D . A ., &
Weymann, R . J . 1984, ApJ, 278, 486.
Carswell, R. F ., Lanzetta, K. M., Parnell, H. C., & Webb, J. K. 1991, ApJ, 371 , 36.
Cen, R . Y ., 1992, ApJS, 78, 341.
Cen, R . Y., & Ostriker, J. P . 1992a, ApJ, 393, 22 .
Cen, R . Y., & Ostriker, J. P. 1992b, ApJ , 399, Lll3.
Charlton, J. C., Salpeter, E. E., & Hogan, C . J. 1993, ApJ, 402, 493
Chernomordik, V. V., & Ozernoy, L. M. 1983, Nature, 303, 153.
Crotts, A. P . S. 1987, MNRAS, 228, 41P.
Crotts, A. P. S. 1989, ApJ, 336, 550.
Dobrzycki, A ., & Bechtold, J. 1991, ApJ , 377, L69.
D'Odorico, S., Molaro, P., & Savaglio, S. 1991, in Proceedings of the ESO Mini Workshop
on Quasar Absorption Lines, ed. P . A. Shaver, E . J. Wampler & A . M . Wolfe (ESO
Scientific Report No. 9 - February 1991), 15.
Duncan, R. C., & Ostriker , J. P. 1988, Princeton Observatory Preprint, POP-261.
Duncan, R . C. , Ostriker, J. P., & Bajtlik, S. 1988, ApJ, 345, 39 .
Duncan, R . C., Vishniac, E. T. , & Ostriker, J. P. 1991, ApJ, 368, Ll (DVO) .
Foltz, C. B. , Weymann, R . J., Roser, H. J., & Chaffee, F. H . 1984, ApJ . 281, Ll (FWRC).
Gunn, J. E., & Peterson, B . A . 1965, ApJ, 142, 1633.
Hunstead, R . W . 1988, in QSO Absorption Lines, ed. J . C. Blades, D. A. Turnshek & C .
A. Norma n (Cambridge University Press), 71.
Hunstead, R. W. 1991, in Proceedings of the ESO Mini Workshop on Quasar Absorption
Lines, ed. P. A. Shaver, E. J . Wampler & A. M. Wolfe (ESO Scientific Report No . 9 -
February 1991), 11 .
Hunstead, R. W., Pettini, M ., Blades, J . C., & Murdoch, H . S. 1987, in Observational
Cosmology, ed. A . Hewett, G. Burbidge, & L. Z. Fang (Dordrecht: Reidel), 799-802 .
Ikeuchi, S., & Ostriker, J . P. 1986, ApJ, 301, 522.
Jenkins, E . B. 1991, in Proceedings of the ESO Mini Workshop on Quasar Absorption
Lines, ed. P . A . Shaver, E. J . Wampler & A. M. Wolfe (ESO Scientific Report No . 9 -
February 1991),25.
Lu, L., Wolfe, A . M ., & Turnshek, D. A. 1991, ApJ, 367, 19 .
Lynds, R . 1971 , ApJ, 164, L73 .
McGill, C. 1990a, MNRAS, 242, 428.
McGill, C. 1990b, MNRAS, 242 , 544.
212

Miralda-Escude, J., & Ostriker, J. P. 1990, ApJ , 350, 1.

Morris, S. L., Weymann, R. J., Savage, B.D., &; Gilliland, R. L. 1991, ApJ, 377, L21.
Murdoch, H . S ., Hunstead, R. W., Pettini, M ., &; Blades, J. C . 1986, ApJ, 309, 19.
Ostriker, J.P. 1988, in QSO Absorption Lines, ed. J. C. Blades, D. A. Turnshek, &; C . A.
Norman (Cambridge University Press), 71.
Ostriker, J. P ., Bajtlik, S., & Duncan, R . C. 1988, ApJ, 327, L35.
Ostriker, J. P., & Cowie, L. L. 1981, ApJ, 243, L127.
Ostriker, J. P., &; Ikeuchi, S. 1983, ApJ, 268, L63.
Pettini, M., Hunstead, R. W., Smith, L., &; Mar, D. P. 1990, MNRAS, 246, 545 (PHSM).
Rauch, M., Carswell, R. F., Chaffee, F. H., Foltz, C. B., Webb, J . K., Weymann, R. J.,
Bechtold, J ., &; Green, R. F. 1992, ApJ, 390,387.
Rees, M. J. 1986, MNRAS, 218, 25P.
Rees, M. J. 1988, in QSO Absorption Lines.,ed. J. C. Blades, D . A. Turnshek, &; C . A.
Norman (Cambridge University Press).
Sargent, W. L. W., & Boksenberg, A., &; Steidel, C. C. 1988, ApJS, 68, 539.
Sargent, W. L. W., Steidel, C. C., & Boksenberg, A. 1989, ApJS, 69, 703.
Sargent, W. L. W., Young, P. J., Boksenberg, A., &; Tytler, D . 1980, ApJS, 42, 41 (SYBT).
Sciama, D. W. 1990, Comments Astrophys., 15,71.
Sciama, D. W. 1991, ApJ, 367, L39.
Smette, A., Surdej, J ., Shaver, P. A., Foltz, C. B., Chaffee, F. H., Weymann, R. J.,
Williams, R. E, & Magain, P. 1992, ApJ, 389, 39.
Smith, H. E., Cohen, R . D ., & Bradley, S. E. 1986, ApJ, 310, 583.
Steidel, C. C., & Sargent, W. L. W. 1990, AJ, 99, 1693.
Teresawa, N. 1992, ApJ, 392, L15.
Tytler, D. A. 1987, ApJ, 321, 69.
Webb, J. K. 1987, Ph.D. thesis, Cambridge University.
Webb, J. K ., Barcons, X., Carswell, R. F., &; Parnell, H. C. 1992, MNRAS, 255, 319.
Webb, J. K., &; Carswell, R. F. 1991, in Proceedings ofthe ESO Mini Workshop on Quasar
Absorption Lines, ed. P. A. Shaver, E. J. Wampler &; A . M. Wolfe (ESO Scientific
Report No.9 - February 1991), 3.
Young, P., Sargent, W. L. W., &; Boksenberg, A. 1982, ApJ, 252, 10.
 PROPERTIES OF THE LYMAN 0: FOREST AT HIGH AND
LOW REDSHIFTS

RAY J. WEYMANN
Carnegie Observatories, Pasadena, California

October, 1992

Abstract. We review the current state of observations of the Lya forest, both the ground-
based observations of high-redshift systems and recent observations with the Hubble Space
Telescope of low-redshift systems. We also describe follow-up ground-based observations
of galaxies near the line of sight through low-redshift Lyman a forest clouds, especially
towards 3C 273, and the implications of all these observations for the nature of the Lya
forest clouds and their relation to galaxies. While it is a priori obvious that some Lya
systems must arise from the disks and/or halos of ordinary galaxies (as confirmed by recent
observations) it also appears that not all Lya clouds are closely associated with ordinary
galaxies. We speculate on the structure of those very low redshift Lya clouds which would
escape detection at 21 cm.
Key words: Lya forest - Intergalactic medium - Galax.ies

1. Introductory Remarks

Since the discovery and correct identification of the Lyo: forest by Lynds over
two decades ago (Lynds 1971), it has been recognized that these ubiquitous
features in the spectra of high redshift quasars held enormous potential for
the study of the nature and evolution of the intergalactic medium as well
as the development of structure in the Universe at the low end of the mass
spectrum. The landmark paper by Sargent and collaborators (Sargent et al.
1980) provided the paradigm for interpreting the high redshift clouds. In
particular, they demonstrated that the high redshift Lya forest systems l
showed little, if any, clustering in velocity space, in contrast to the higher
column density metal-containing systems. These, and other considerations
led these authors to suggest that the high redshift Lya forest clouds consisted
of unprocessed hydrogen and helium clouds which were pressure-confined by
a hot intergalactic medium (IGM). Subsequent work has also demonstrated
that the evolution with red shift of the line density of the high redshift Lya
forest is different from that of the metal-containing systems. This picture
suggests that these clouds might have little, if anything, to do with what we
would consider "ordinary" high red shift galaxies, and indeed this question

1 The term "Lya forest" was first introduced (Weymann et al. 1981) to refer to Lya
absorption lines of low-to-moderate column densities (:s 10 15 cm- 2 ) which do not show
lines of the heavy elements. In this article we shall adhere to this usage, though whether the
Lya forest and the higher column density, heavy element-containing systen1S represent two
quite different populations or a continuous spectrum of properties is a major unresolved
issue.

213
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, 2l3-236.
© 1993 Kluwer Academic Publishers.
214

remains unclear today for both the high and low redshift Lya forest clouds
and is one of the most important to begin to answer.
It was widely appreciated that with the advent of the Hubble Space
Telescope (HST) it would be possible to study these clouds at low redshifts
and thus to study the evolution of the clouds over a very much longer span of
time than was possible from the ground, as well as to begin the study of the
possible association of these clouds with various types of galaxies. Until the
results for 3C 273 were obtained (Morris et al. 1991; Bahcall et al. 1991) it
was not clear that these low redshift Lya forest clouds would be numerous
enough for meaningful statistical studies to be carried out . The fact that
there are a moderate number of such clouds allows us to be optimistic that
at last the study of structure through the analysis of galaxy redshifts will
conjoin with QSO absorption line studies to provide us with a much clearer
idea of the evolution of the IGM and the development of structure than we
have had thus far.
The remainder of this Review is organized as follows: In §2 we give a
summary of the properties of the high-redshift Lya forest clouds as deduced
from ground-based observations. (A detailed discussion of models ofthe IGM
and Lya clouds will be found in the review by Bajt1ik in this volume.) In §3
we summarize the (still rather meagre) observations of the low red shift Lya
forest based upon HST observations, in particular results from the Quasar
Absorption Line Key Project. In §4 we describe the (even more meagre)
observations pertaining to the association between Lya forest clouds and
galaxies, and finally in §5 we close with a list of still unresolved questions
and suggestions for further work, as well as some speculations of ours and
others on the nature of the Lya forest.

2. The Properties of the Lya forest at Redshifts z ~ 1.6

An enormous amount of ground-based telescope time has been invested in

learning about the properties of that portion of the Lya forest accessible
from the ground. There are a number of fairly recent reviews and discussions
of the Lya forest properties. See, for example, (Bechtold 1987) and the
separate presentations by Carswell, Hunstead, Rees and Sargent in (Blades
et al. 1988) and we summarize only a few salient highlights. Implications of
these features for models will be found in Bajtlik's review in this volume. We
also do not discuss the use of the so-called proximity effect or the possible
existence of voids, which are addressed in the review paper by Bechtold
elsewhere in this volume.
 215

2.1. EVOLUTION OF THE LYa FOREST

It was first noted by Peterson that the number of Lya forest lines increases
with redshift (Peterson 1978). Until the last few years, very little data ex-
isted at resolutions adequate to resolve the Lya lines, so that only equivalent
widths were available. It has been customary to parameterize the distribu-
tion of lines as a function of redshift and rest equivalent width by (Murdoch
et al. 1986)

.o2N
ozoW
(:) 0 (W*)-l(l + zyrexp( - WjW*) (1)

where (dN j dz)o (the density of lines at zero red shift above some prescribed
limiting rest equivalent width), "I and W* are parameters to be determined.
There is some evidence that there may be a steepening in the rate of evolu-
tion for redshifts below about 2.3 (Lu et al. 1991, see especially their Figure
1).
Bechtold has provided a recent discussion of the evolution of the Lya for-
est over the range of redshifts accessible to ground-based telescopes (Bech-
told 1993). She combined new moderate resolution data with a careful se-
lection of previously published data to produce a sample with about 2800
Lya lines. Her main results may be summarized as follows. For lines with
rest equivalent width (REW) > 0.32 A she finds '"Y ~ 1.8 (considerably
smaller than the value found by Lu et al. ) and W· = 0.26 A, also slightly
smaller than previous values . No convincing evidence for a break in the rate
of evolution was found.
The simple parameterization of equation (1) may not be adequate since,
in addition to the possible steepening noted by Lu et al., there is some evi-
dence that the rate of evolution depends upon line strength, though this last
point remains controversial and there is not even general agreement yet on
whether the weaker or the stronger lines evolve more rapidly. For example,
for REW' s between 0.16 and 0.32 A Bechtold finds a much slower rate of
evolution, with '"Y ~ 0.5. However, her sample was based upon moderately
low resolution, and it has been suggested that line blending may artificially
cause the stronger lines to appear to evolve more rapidly (Trevese et al.
1992). But even analyses based on higher resolution data have reached dif-
fering conclusions (though some studies have been based on samples utilizing
rest equivalent width while others have utilized column densities, so direct
comparisons are not straightforward).
One such study found a marginally significant trend in the same sense
found by Bechtold (Webb & Larson 1987). A second study (Giallongo 1991)
found the opposite trend (i.e., the weaker lines evolve faster than the strong
lines), while two more recent studies found no statistically convincing evi-
dence for a dependence of'"Y upon line strength (Rauch et al. 1992), or, more
216

generally, no evidence that the colunm density distribution depends upon

redshift (Petitjean et al. 1993a).

2.2. DISTRIBUTION OF COLUMN DENSITIES AND DOPPLER PARAMETERS

When the resolution is adequate to measure the actual profiles, then it is

preferable, of course, to actually derive the column density and doppler pa-
rameters for the lines. To do this reliably, a resolution of ::::, 15 km S-1 is
probably required. Several high resolution investigations have been carried
out to derive colunm density and doppler parameter distributions. A con-
troversy has arisen over the result of one such investigation (Pettini et al.
1990a). These authors used a resolution of 6 km s-1 to examine the Lya
forest in Q2206-199 and reported the existence of Lya forest lines with some
doppler widths substantially less than 10 km S-1. They also found a strong
correlation between doppler width and colunm density. Both these results
were not found in previous high resolution studies of other objects (e.g.,
(Carswell et al. 1991)). This issue attracted considerable attention since the
very low tempertures implied by these low values were incompatible with
static clouds in thermal and ionization equilibrium for reasonable cloud sizes.
Expanding clouds can cool the gas adiabatically and lead to somewhat,lower
effective doppler parameters (see the review by Bajtlik) but it is doubtful if
clouds in which realistic heating input occurs can be cooled to values as low
as some of the doppler values reported by Pettini et al. (1990a). Donahue
& Shull (1991) have also pointed out that if the electron colunm density is
sufficiently low, photoelectrons can escape, which dramatically lowers the
heating rate. However, as these authors note, this requires the fraction of
neutral hydrogen to be much higher than estimates based upon the inter-
galactic radiation field and the transverse cloud size discussed in §2.3 below,
so that extreme aspect ratios for the clouds are implied.
To attempt to resolve the controversy, a completely independent analy-
sis of the same data for Q2206-199 upon which the Pettini et al. analysis
was based was carried out (Rauch et al. 1993). These authors performed a
number of simulations to explore the possible effects of sample selection and
reduction procedures. In the opinion of this (possibly prejudiced) reviewer,
this reanalysis makes a good case for the fact that the apparent correlation
between doppler width and colunm density is an artifact arising from using
a sample which is not complete to a given column density for all reasonable
doppler widths. Additionally, for low SIN, photon noise can cause systems
with true moderate doppler parameter to migrate into a domain with appar-
ently low doppler parameters so that some of the Lya lines with apparently
low doppler parameters may prove, at higher SIN, to have higher doppler
parameters. Moreover, some of the lines previously identified as Lya can be
identified as being metal lines (for which the narrow doppler parameters are
 217

not surprising) or, in a small number of instances may prove not to be real
lines.

However it · is certainly not possible with that data set to prove that
there are no Lya clouds with doppler parameters less than, say, 10 km s-l.
To settle this controversy once and for all requires higher SjN and more
extensive spectral coverage. Pending the final resolution of this controversy,
the Rauch et ai. analysis finds a median doppler parameter of'" 26 km S-l,
with a 20 th percentile value of rv 17 km s- l and an 80 th percentile value of
rv 38 km s-l, and no evidence for an intrinsic correlation between column

density and doppler parameter.

It is not clear what the interpretation of the larger values of the effective
doppler parameters found for this and other objects should be. They clearly
cannot be purely thermal motions-unrealistically high temperatures would
be required. Whether they represent unresolved blends or bulk motions only
very high signal-to-noise observations at high resolution can settle. There
are sufficient numbers of such cases that, if they were unresolved blends, a
substantial amount of clustering at very small velocities would be implied.
In this connection, it is interesting to note that Crotts found the excess
number of pairs in his cross-correlation analysis of the Lya forest to occur
entirely within the first 100 km S-l bin, as described below (Crotts 1989).

The Rauch et al. analysis finds a column density distribution which, over
a limited range in column density is a power law with index rv -1.6, slightly
flatter than previous estimates, but not significantly so. It is of considerable
interest to know how the column density distribution behaves at the very
lowest column densities. This question is of intrinsic interest in terms of
models of clouds (e.g., the possible role of thermal conduction, the presence
or absence of a high temperature confining medium, the role of magnetic
fields in mitigating the effects of thermal conduction, etc.). Additionally, it
is important in attempts to measure the Gunn-Peterson trough, since very
weak lines which are individually undetectable can contribute to the trough
(Webb et al. 1992).

By assembling data on column densities extending to high values (well

beyond what we have defined to be the Lya forest), it has recently been
shown that a single power law is a rather poor representation to this dis-
tribution (Petitjean et ai. 1993b); see also (Kulkarni and York 1992). The
distribution has substantial structure and this structure can be interpreted
in terms of models of the clouds (Petitjean et al. 1993c; Charlton et ai. 1993)
and also bears upon the question of whether the low column density Lya
clouds belong to the "same population" as the higher column density ones.
218

2.3. CHARACTERISTIC SIZES OF THE CLOUDS

An important parameter in deducing the properties of the Lya forest clouds

is their characteristic dimension. Since the clouds could be far from spheri-
cal, ideally one would like to measure the transverse and lateral dimensions
for a large number of clouds. Moreover, we would like to know how these
parameters vary with both column density and redshift. Such a possibility
is far in the future. Very close pairs (gravitational lenses or otherwise) al-
low estimates to be made (or limits to be set) of the transverse dimension
for (presumably) the individual cloud sizes and can also provide, in princi-
ple, structural information on individual clouds. Pairs with somewhat wider
separations provide information (presumably) on the clustering of the Lya
forest clouds rather than on their individual sizes, as discussed below.
The strongest lower limit on cloud sizes comes from the pair UM673
(Smette et al. 1992) and yields a value':::' 25 kpc. The characteristic size of
",,10 kpc based upon the pair 2345+00A,B (Foltz et al. 1984) is currently
of dubious value because of lingering doubts about the reality of its being
a gravitational lens, plus uncertainties in the lensing geometry even if it is
lensed (Steidel & Sargent 1990) and/or the fact that some of the Lya lines
may belong to metal-containing systems whose cloud sizes may be different
from the Lya forest lines (Steidel & Sargent 1991).
The observations of Smette et al. of UM673 are also important for an-
other reason: It is not a priori evident that absorption at the same redshift
along two neighboring lines of sight is really sampling the same cloud, as
opposed to a cluster of clouds of smaller dimensions (this latter assumption
is tacitly assumed in the discussion of clustering below.) However, the strong
correlation between equivalent widths along the lines of sight to UM673 is a
strong argument that the clouds are coherent and moderately homogeneous
over the dimension quoted above.
In principle, given knowledge of the radiation field beyond ",2 Rydbergs
at a given redshift, systems for which column densities for both He I and
He II can be measured would allow a direct estimate to be made of the
electron density, from which the dimension along the line of sight could be
estimated.

2.4. CLUSTERING PROPERTIES

As noted in the introduction, one of the striking properties of the Lya forest
is the absence of power in the two-point correlation function at all but the
smallest velocity separations. Some power does seem to exist at velocities
up to '" 300 km s-l(Webb 1987), see also (Rauch et al. 1993, Figure 5) and
(Kulkarni and York 1992), as measured directly by the velocity two-point
correlation function. Supporting evidence for the reality of such structure is
 219

provided by an independent statistical technique (Barcons & Webb 1991).

Information on clustering is provided not only by the two-point (auto)
correlation function in velocity space along individual lines of sight, but also
by the study of QSO pairs with moderately close separations. In the first
such study (Sargent et al. 1982), an upper limit of'" 1 - 2 Mpc (or ~ V '"
400 km S-I) on the clustering scale was set based on a study of a pair with
a separation of 173 arcsec . These authors also concluded from this study
that "Lya clouds cluster more weakly than galaxies". A study of a some-
what closer pair with a separation of 58 arcsec yielded "a marginal excess of
small velocity separations" with splittings ;S500 km S-I. (Shaver & Robert-
son 1983). Subsequently, Crotts reexamined the pair at 173 arcsec with the
significant addition of data from.a third fainter QSO which generates two
additional separations of 127 arcsec and 147 arcsec (Crotts 1989). Examin-
ing the two-point velocity cross-correlation function from these pairs, Crotts
found an apparently statistically significant excess of pairs, but somewhat
surprisingly, this excess was confined entirely to the first 100 km s-1 bin.
As Crotts has stressed, the use of cross-correlations between such pairs al-
lows the detection of velocity separations so small that they would be badly
blended along a single line of sight at most resolutions. Nevertheless, if such
clustering exists, it should show up in the autocorrelation function in data
of even higher resolution and S IN than that which suggests some possible
clustering on scales of ;S 300 km s-l, e.g., (Rauch et al. 1993), (Kulkarni
and York 1992). As suggested above, splitting at very small velocity scales
may be relevant to the interpretation of the anomalously high values of the
doppler parameter which are found.
The question of power at large velocity scales is of considerable interest
in connection with both "real" voids and voids produced by the radiation
field of a QSO near some clouds which will ionize and weaken the lines. See
the review by Bechtold in this volume.

2.5. METAL ABUNDANCE

The paradigm for the Lya forest originally proposed (Sargent et al. 1980)
envisions the clouds as pressure-confined objects in which stars never formed
and which were sufficiently far removed from early regions of star formation
that the gas in the Lya forest remained pure hydrogen and helium.
Detailed abundance determinations in QSO absorption line systems have
been limited mainly to H I column densities so high (,(, 10 20 cm- 2 ) that the
H I column density can be determined from the Lya damping wings. Because
of the heavy shielding, lower ionization species dominate and the ionization
corrections are far less uncertain than when the metals (and presumably
the hydrogen) are highly ionized. Additionally, comparison of species which
suffer very little and very significant dust depletion (e.g., zinc vs. chromium)
220

can be used to estimate how significant depletion by grain formation might

be. Typical metal abundances are of order 0.2 to 0.04 of the solar abundances
(York 1993) and (Meyer & York 1992). While there is likely a systematic
trend of decreasing metal abundance with increasing redshift at the higher
redshifts leading to depletions as low as -3.0 (Cooke 1992) there also appear
to be systems with comparably low metal abundances at lower redshifts
(Rauch et al. 1990) so that significant variations at a given epoch may exist.
These variations may reflect differences in the nature of the absorbers (Meyer
& York 1992; Pettini et al. 1990b). Attempts to determine, or set limits on,
metal line strengths in clouds having lower column density Lya lines (e.g.,
N(H I) ~ 1017 cm- 2 ) are more difficult. Unless one has high resolution data
of some of the higher members of the Lyman series as well as for Lya itself,
the column density estimates can be off by an order of magnitude, although
the absence of a Lyman limit at least sets an upper limit to the H I column
density. Additionally, the ionization corrections are likely to be very large
and uncertain. Nevertheless, there are now known many instances of metals
being detected in systems with log N(H I) in the range 15 - 17, and it
appears not unlikely that at these intermediate column densities there are
variations in the metal deficiencies comparable to those of the damped Lya
systems and spanning similar ranges (Chaffee et aI. 1986; Meyer & York
1987; Blades 1988).
At even lower column densities, the mere detection of metal lines is very
difficult for likely ionization conditions for even modest metal deficiencies.
Several authors have attempted such detection by coadding spectra which
are shifted successively to the rest frame of each Lya line. The most con-
vincing case for a detection using this technique is the recent work of Lu,
who coadded up to ",300 Lya lines from 14 lines of sight and claimed de-
tection of the Al548 line of C IV at the 99.99% confidence level, though the
weaker A1550 line was less apparent (Lu 1991). Lu then attempted to make
ionization corrections and deduced a value of [C jH] ~ -3.2. This interesting
and important result should be checked with a larger dataset as well as a
search for individual C IV absorption in systems with low-to-moderate H I
column density using very high S jN and very high resolution data.
All these detections imply metal deficiencies which are less than the most
metal-poor stars known in our galaxy, so that the characterization of the
composition of all the Lya forest clouds as truly "primordial" is no longer
tenable, though we do not yet know how extreme the metal deficiency in
some of the lower column density systems might be.

3. Preliminary HST Observations of the Lya forest at z ~ 1.6

As noted in the previous section, some recent ground-based work (Lu et aI.
1991) gave some suggestion that the decrease in density ofLya forest clouds
 221

TABLE I
Parameter Estimates for Lya Systems from Key Project Data

Dataset /Method (dN/dz) 10 l' W· (I)

HST(89), fix 1', W· 17.25±1.83 0.75 0.32
HST(89), fix W· 18.97±4.62 0.48±0.62 0.32
HST(89), fit 1', W· 17.67±4.26 0.31±0.61 0.216±0.021
HST(51/89)+Lu, fix W· 7.29±1.40 1.72±0.15 0.32

in redshift space with decreasing redshift is moderately steep at redshifts ;;::

2.0 and then decreases even more steeply for z ~ 2.0. If one extrapolated
this very steep slope to zero redshift then the Lya clouds would be predicted
to be very scarce. In retrospect it is obvious that even if this extrapolation
proves in some respects to be valid, there must be a non-negligible popula-
tion of absorbers associated with the disks and halos of ordinary galaxies.
Whether such absorbers are' intrinsically different from the high redshift
clouds remains a crucial unanswered question.
In fact, the first reliable detection of intervening Lya absorption systems
was made not with HST, but with IUE in the line of sight to the BL Lac
object PKS 2155-304 (Maraschi et al. 1988). However, the GHRS and FOS
observations of 3C 273 (Morris et al. 1991; Bahcall et al. 1991) clearly es-
tablished that the Lya forest clouds were present at low redshift in sufficient
numbers that either the putative Lya absorbing disks and halos of present
day galaxies are much larger than previously thought or there are numerous
Lya clouds which are not associated with ordinary spiral galaxies.
The most homogenous data set of low redshift Lya lines comes from the
HST Quasar Absorption Line Key Project. Since a preliminary account of
this work is now in press (Bahcall et al. 1993) we give here only a brief
summary.
As of this date, full analysis has been completed for 13 QSOs. The essen-
tial results for the Lya forest are summarized in Table 1, which is extracted
from Table 8 of Bahcall et al. (1993). A maximum likelihood technique was
used to estimate the parameters in Table 1, assuming that the density dis-
tribution of Lya forest clouds in equivalent width-redshift space could be
adequately represented by the simple form given by Equation 1 above.
The entry (dN / dz) 10 is the estimated total number of lines per unit red-
shift at zero redshift having rest equivalent widths above 0.32 A. In the next
two columns, values of I and W· without uncertainties are assumed values
222

for these parameters; those with uncertainties are fitted values. Except for
the last row of the table, the sample size consists of the 89 Lya lines from
the Key Project data base which were farther than 3000 km s-1 from the
adopted emission line redshift (to avoid both "associated systems" and the
"proximity effect") and which were "real" at the 4.50' level. The last row
consists only of a subset of 51 from the 89 HST lines which have rest equiv-
alent widths greater than 0.36 A to which has been adjoined an additional
905 lines satisfying the same conditions taken from the ground-based sample
of (Lu et al. 1991) (with redshifts z 2:: 1.64).
Using the currently available HST data, Bechtold suggests that a single
value of, ~ 1.8 may adequately represent the evolution of the lines with
REW >0.32 A over the entire redshift range from 0 to rv4. (Bechtold 1993).
(This is very similar to the global fit represented by the last row of Table
1). While such a fit somewhat underpredicts the number of observed low-
redshift systems, such a fit cannot be rejected at present. It is intriguing that
the third row of the table, which is the best fit for the parameters, and W·
based solely on the key project data gives a much flatter rate of evolution
than the global fit of the last line. However, the difference between these two
slopes is still well within the uncertainties due to the current very limited
redshift range for most of the key project Lya lines. Assuming that this key
project is completed as currently contemplated, the full HST red shift range
from 0 to rv1.6 should be reasonably well sampled, and, with the addition
of further ground-based data in the interesting 1.6 - 2.0 redshift range, we
should have a much clearer idea of the evolution of the Lya forest over the
entire redshift range from 0 to rv5. One might hope that this will shed some
light on the dominant mechanism influencing the evolution as a function of
redshift and whether this mechanism is the same at high redshifts as it is at
low redshifts.

4 . The Low Redshift Lya Clouds and their Relation to Galaxies

The existence of Lya forest clouds at very low redshift offers the oppor-
tunity to explore their immediate environment, which may provide a clue
to their nature and origin. Ideally, we would like to obtain images of the
clouds themselves. In practice, this is probably well beyond the reach of
current technology. The clouds we consider to be members of the Lya forest
have, virtually by definition, column densities of H I sufficiently small that
they are optically thin at the Lyman limit, with an H I column density of
1014 cm- 2 being typical. This implies an optical depth at the Lyman limit
of order 10- 3 or less. The order of the difficulty of detecting such clouds by
their Ha recombination radiation may be appreciated by noting that recent
measurements made with Fabry-Perot devices of Ha emission from clouds
which have large optical depths in the Lyman continuum have resulted in
 223

marginal detections (Kutyrev & Reynolds 1989; Songaila et al. 1989) - see
also Stocke et al. (1991). Additionally, these detections must be treated as
providing only upper limits to the actual metagalactic UV radiation field,
since there may be local sources of ionization affecting these measurements.
In fact, these detections are about an order of magnitude higher than current
theoretical estimates of the UV radiation field from QSOs (Madau 1992).
Thus, detection of Ha recombination radiation from a typical Lya forest
cloud would require an increase in sensitivity of the limiting surface bright-
ness of betw~en 3 and 4 orders of magnitude unless the cloud is near a
powerful source of radiation, or, unless there is a non-radiative source of
ionization (e.g., a shock front).
The situation is similarly discouraging for 21 cm imaging; measurements
of column densities much below 5 X 1018 cm- 2 are very difficult, so detection
of a cloud with H I column density'" 1014 cm- 2 is currently out of the
question. Despite this discouraging situation, it still seems worthwhile to
check for features in emission in Ha and to obtain 21 cm maps in the vicinity
of known Lya clouds, especially at low redshifts. As discussed below, it has
been suggested that some of the Lya clouds are the extensions of hydrogen
disks in galaxies beyond the point where they can remain neutral when
exposed to the metagalactic radiation field (Maloney 1992). While we have
just noted that direct detection of either the Ha recombination radiation or
21 cm radiation at the low H I column densities in question is not feasible, it
might be that substantially larger column densities are lurking a bit off the
line of sight . Furthermore, the parent "galaxy" whose flaring disk ultimately
leads to a low column density Lya system might not be a normal galaxy
at all. It might, for example, be a system in which only a few early type
stars have formed and which gives rise to a small amount of Ha emission
and/or a system in which the total H I content is so small that it has escaped
detection in a 21 cm survey, especially when hidden in the radio glare of a
powerful source like 3C 273.
With these considerations in mind, searches for Ha emission along the
sightline towards 3C 273 and in redshift ranges bracketing the lower redshift
clouds in the 3C 273 line of sight have been carried out (Morris et al. 1993,
hereafter "Mor93"). These authors used narrow band filters centered on the
redshifted Ha for the v ~ 1000, 1600 and 8600 km s-1 Lya clouds towards
3C 273. No Ha emission was found at these redshifts over a field of about
2.6 arcmin radius (about 12 kpc at an assumed distance of 16 Mpc) down to
a 30- flux level of", 6 X 10- 18 ergs cm- 2 S-1 arcsec- 2 (equivalent to an emis-
sion measure of about 9.) While this is about an order of magnitude larger
than the marginal detections mentioned above (for clouds with very large
optical depths at the Lyman limit, hence orders of magnitude larger than
the surface brightness expected from these low redshift Lya forest clouds
themselves) this limit nevertheless corresponds to an Ha surface brightness
224

produced by only about one Bl star per square arc second. Evidently, there
is very little star formation occuring within a few kiloparsecs of these Lya
clouds themselves. Measurements still more sensitive by about an order of
magnitude have been made using the CTIO Fabry-Perot (Williams 1992)
but the results are not yet available.
Concerning 21 cm results, a very preliminary description of a VLA search
for 21 cm emission in a large field around 3C 273 and over a velocity range
bracketing two of the low redshift Lya clouds was presented at this confer-
ence by van Gorkom. No obvious features down to H I surface densities of
order 1019 cm- 2 are present, but final results must await a detailed analysis
(van Gorkom et al. 1993).
In order to check for the presence of very low surface brightness dwarfs
whose outer edges might give rise to the Lya clouds, a mosaic of deep r-
band images were obtained over a region about 18 arcmin square centered
on 3C 273 (Mor93). Any dwarfs in this region which were morphologically
similar to those found in the Virgo cluster itself (Bingelli et al. 1985; Sandage
& Bingelli 1984) would have been seen if they had had absolute magnitudes
greater than about -13 or -14. Thus, if the outer envelopes of dwarf galaxies
are responsible for the 1000 km s-l and 1600 km s-l absorption features, the
galaxies themselves must be of very low luminosity or else their centers must
be located more than about 40 kpc from the line of sight intercepting these
clouds. Inspection of a new POSS-Il IlIaJ plate and an older IlIaJ plate
taken by Sandage also reveals no extended low surface brightness objects
closer to 3C 273 than the optical counterpart to the giant H I cloud H I
1225+01 (Giovanelli & Haynes 1989) which is visible on both these plates. 2
It is not clear how far away one can seek galaxies which match the ob-
served redshift of the Lya clouds to within some prescribed accuracy and
still have the match indicate any physical association. If some association
is indicated, how is that association to be interpreted? It has already been
pointed out (Salzer 1992) that U7549 (=N4420), with a redshift of 1685 km
S-l and a projected distance from the 3C 273 line of sight of 193 X (DMpc/16)
kpc may be "associated" with the ",1600 km s-l cloud3 while Mor93 note
a somewhat closer match of the 1600 km s-l cloud with U7612 which has
a velocity equal to the 1600 km s-:-l cloud to within the measurement er-
rors and a projected separation of 187 X (DMpc/16) kpc. As Salzer also
noted, the object currently known to have the closest projected distance to
the 3C 273 sight line is in fact H I 1225+01 with a projected distance of
2 I thank Neil Reid and Allan Sandage for very kindly allowing me to inspect these
plates. It should be noted that an earlier preliminary report of a dwarf galaxy about 11
arcminutes east and 2 arcmin north of 3C 273 was erroneous; the feature is a plate defect
on the new POSS-II plate.
3 Since the two lowest redshift clouds lie on a steep portion of the galactic Lya damp-
ing wing, their exact redshifts will remain uncertain until the medium resolution GHRS
observations have been obtained. We adopt here Vi = 1000 Ian S-l and \12 = 1572 Ian S-l .
 225

157 x (DMpc/16) kpc, but its velocity differs by about 300 km s-l from both
Lyo clouds. The galaxy with the smallest known projected distance to the
3C 273 sightline whose redshift agrees to within ",150 km s-l of the 1000
km s-l cloud is UGC 7694, also noted by Salzer, whose projected distance
is 598 X (DMpc/16)kpc. A second object (NGC 4580) matches the red shift of
the 1000 km S-l cloud to within the measurement errors but has a projected
distance of over 1 Mpc (Mor93).
In the absence of any direct observational evidence of a connection be-
tween putative identifications such as these (e.g., bridges of H I extending
towards the Lyo clouds) the question becomes a statistical one. However,
two or three cases along other lines of sight are already known where it
is obvious, without any sophisticated statistical analysis, that the agree-
ment in redshift between galaxies and Lyo clouds having small projected
separations is so close that they must be physically associated. The most
compelling such example to date is along the line of sight to H1821 +643
where a triplet of Lyo lines lies within a few hundred km S-l of a galaxy
whose projected separation is '" 90 kpc (Bahcall et al. 1992) .
Indeed, given what is known about the association between low redshift
QSO Mg II absorption and galaxies (see the review paper by Steidel at this
conference) it would be very surprising if such instances did not occur. It
seems inevitable that there will be instances under which gas in the outer
disk or halo of a galaxy, in addition to giving rise to Mg II absorption (and
corresponding high column density Lyo: clouds which are self-shielding) will
have the appropriate particle and column density to allow it to become
ionized, as described by (Maloney 1992), thus giving rise to lower H I column
density Lyo clouds. Such clouds may show metal lines of only high ionization
and then only with spectra of high resolution and high S /N. The question is
thus whether all the Lyo: clouds can be understood in this way, and if not,
what fraction of them can be. This question, as illustated by the ambiguity
in the case of the clouds at the Virgo distance, is inherently a statistical one.
Salzer attempted to carry out a statistical discussion of whether the Lyo:
clouds in the sight line towards 3C 273 were distributed like galaxies, but
lacked a suitable data base of galaxy redshifts to reach firm conclusions
(Salzer 1992). In an attempt to improve this situation, Mor93 carried out
fiber spectroscopy of candidate galaxies down to m ~ 19 over a 2.20 x 1.6 0
field centered on 3C 273. The redshifts obtained by this program were sup-
plemented by known redshifts in the literature. A full discussion of this work
will be found in Mor93, and we give only the main results here:

(i) Nearest neighbor simulations and cloud-galaxy pair counts to fixed linear
separations suggest fairly strongly that the distribution of the observed Lyo:
clouds along the 3C 273 sightline is incompatible with the hypothesis that
there is no correlation between these Lyo: clouds and the galaxies in the
226

2 pt correlation functions

----- Gal-Gal
0-------40 Cloud-Gal

10.0

1.0
-'-

0.1 L-~L...~~"""""-~~~~~0.i...O-~~~~10~0:-.0::-~~~-:-10~OO.0
0.1 1.0 1 .
Separation (Mpe)

Fig. 1. The galaxy-galaxy and Lya cloud-galaxy spatial two point correlation func-
tions from Morris et al. 1993. The correlation functions shown here converted the
observed velocity difference between each pair to a line-of-sight separation utiliz-
ing a statistical algorithm incorporating gravitational perturbations on a smooth
Hubble flow (Ho = 80) and are also corrected for the galaxy selection function.

Mor93 sample.

(ii) However, pair counts and comparison of the redshift distribution of the
clouds and galaxies (after the latter distribution is corrected for being a flux
limited sample defined over a fixed solid angle) also negate the hypothesis
that the correlation between the Lya clouds and galaxies is the same as that
between galaxies and galaxies.

Both these conclusions are evident by simply displaying the galaxy-galaxy

and the galaxy-cloud two-point correlation function, which is shown in Fig-
ure 1. In calculating the 3-dimensional spatial correlation functions, some
 227

assumption must be made about how to convert the observed velocity differ-
ence between each pair of objects to their actualline-of-sight separation. In
Mor93, two different assumption were made: (i) a pure unperturbed Hubble
flow; (ii) an algorithm based upon earlier work of Davis and Peebles which in-
corporates gravitational interactions and consequent modifications of a pure
Hubble flow (Davis & Peebles 1983). As expected, these two assumptions
produce significantly different results only at the smallest separations and
both support the two main conclusions above: (i) there is clearly power in the
cloud-galaxy correlation function which can be detected in the Mor93 sample
to several Mpc, but (ii) the cloud-galaxy correlation is clearly weaker than
the galaxy-galaxy correlation, except possibly at the very smallest scales.
In addition to these two general results, there are some particular ob-
jects of interest which illustrate result (ii), though in and of themselves they
may not be highly significant in a statistical sense. One of the Lya clouds,
at a redshift of z = 0.067, lies in a region in which the nearest galaxy in
the Mor93 sample is at a projected distance of 4.75 x(80/ Ho) Mpc and a
velocity difference of 710 km s-1 from the cloud. At this distance, a typical
galaxy at the flux limit of the Mor93 sample (m "-'19) would have an ab-
solute magnitude of M"-' -18 (bearing in mind, though, the stricture given
by Heavens at the conference that detection or non-detection of a galaxy
involves more than just its integrated flux.) This would imply that if every
Lya cloud was intimately associated with some galaxy (which result (ii)
suggests in any case is false) then the association is very tenuous indeed,
or else the assoCiated galaxies have absolute magnitudes less than M,,-, -18.
This latter possibility cannot be excluded and is not a priori unreasonable,
though as noted in connection with the imaging of the 3C 273 field, there
are no dwarf galaxies in close proximity to known Lya clouds at even very
low luminosities. Even if we adopt the view that low luminosity galaxies
give rise to all the Lya clouds which cannot be associated with luminous
galaxies, it is interesting that the situation with regard to the z = 0.067
cloud just described (which happens, incidentally to be one of the strongest
in the 3C 273 sightline) would imply that low luminosity field galaxies exist
without any nearby bright "parents". Apparently very little is known about
whether this does or does not occur at such low luminosities.
A second configuration of interest concerns a large swarm of galaxies
which is spread more or less uniformly across the Mor93 field between red-
shifts 0.075 - 0.082. This swarm may be part of a large structure including
the richness class-O Abell cluster A1564. The nearest Lya cloud in the Mor93
sample has a redshift of 0.088. A simple analysis based upon the hypoth-
esis that the distribution of Lya clouds follows the distribution of galaxies
suggests that the absence of absorbers at redshifts near the peak of this
swarm is a somewhat unlikely occurrence, but the result is not a strong one
because the density of clouds in redshift space is so small. Because the selec-
228

tion function for the galaxies at very low redshifts is so large and uncertain,
it is not clear if the clouds at the approximate Virgo distance are consistent
with this result or provide a counterexample to the absence of Lya clouds
in the swarm of galaxies associated with the A1564 cluster.

5. Discussion
What conclusions can one draw about the nature of the Lya clouds on
the basis of the currently available ground-based and HST data, and what
seem to be the most important questions to attack? In his review Bajtlik
describes in detail various models and simulations for the Lya clouds so I will
not review such material here. Instead, I will indulge in a few speculations
on the nature of the Lya clouds and comment on the issue of "populations"
and then summarize some observational programs that would seem worth
carrying out and which would help to further our understanding of the Lya
forest.

5.1. SOME SPECULATIONS

Regardless of what the nature of the Lya forest clouds proves to be, the
inescapable fact is, that even at low redshifts where the Lya forest is rather
sparse, the surface density of the clouds is very high. Ignoring the Lya clouds
with redshifts below 3000 km s-l for 3C 273 (on the grounds that the line
of sight passing through the southern extension of the Virgo cluster may
not be typical) the data for 3C 273 (Morris et al. 1991) suggest that a line
of sight encounters of order one cloud every 3000 km s-l. If instead the fit
for the third row of Table 1 is extrapolated to very weak lines using the
parameters for that fit and the representation given by Equation (1), then
a cloud is encountered every 4000 km s-l.
Using the latter, more conservative value one can then construct a simple
table (Table 2) giving the relation between the space densities of absorbers
and their effective cross sections (Maloney 1992, equation 4). Note however
that Table 2 is expressed in terms of the effective radii of the clouds, not
their diameters. Table 2 invites comparison of these required densities with
the space density of various types of galaxies. As many authors have re-
alized, if the Lya forest were to be identified with the outer disks of L*
spirals the effective cross sections become huge, and as noted by Salzer, it
would be much more reasonable to ascribe the absorption to dwarf satellites
rather than to the main galaxy. Thus, the outer envelopes of dwarfs might
be promising candidates for producing the Lya absorption if they are suf-
ficiently numerous. Since it is the gas content that is relevant, and since it
may be that at the low end of the dwarf irregular sequence very little star
formation may have occurred, the mass (of gas) function is of more interest
 229

TABLE II
Space Density of Low Redshift Lya Absorbers vs. Radius

Reu (kpc) Space Density (Mpc 3)

1000 (8 X 1O- 3 )h 1OO

300 (8 X 10- 2 )h lOO
100 (8 X 1O- 1 )h lOO
30 (8 X 1O- 0 )h lOO
10 (8 X 10+1)hlOO

than the luminosity function.

A recent 21 cm survey of very low mass dwarfs sensitive to H I masses
~108 M0 resulted in fewer detections than expected on the basis of extrap-
olation of a Schechter optical luminosity function, and "leaves no room for
a substantial population of H I dwarfs ... that could have been missed by
conventional, optical galaxy catalogs" (Weinberg et al. 1991). This survey
suggests that the H I mass function is truncated between 108 - 109 M 0 .
As Maloney suggests, it is conceivable that this truncation is related to
a transition betwj!en H I and H II clouds as the parameters characterizing
the formation and structure of the clouds vary (Maloney 1992). As a very
simple illustration of the parameter space required to produce very low red-
shift Lya clouds with undetectable amounts of H I, consider the following
crude model for clouds of gas devoid of stars:

1. Adopt a value of the zero redshift UV background photoionization rate

corresponding to J v = 1.0 X 10- 23 ergs cm- 2 S-1 Hz- 1 sr- 1 at 912 A with
a spectral index of -1. This is a value intermediate between Madau's recent
estimate (Madau 1992) and the possible detections (Kutyrev & Reynolds
1989; Songaila et al. 1989). Then, if there is no self-shielding, the fraction of
hydrogen which is neutral, x, is given by x :::::: 10n o, for no ~ 0.1 and x :::::: 1
for no ~ 0.1, where no is the total particle density.

2. Assume that the ionized gas has a temperature of 20,000K.

3. Idealize the structures as being spheres of radius R with constant (neutral

+ ionized) gas density no. (If gravity plays a significant role in the structure
of the cloud, then R should be identified as the scale length, and there may
be a large envelope of lower density and more highly ionized material.)
230

4. Consider only radial rays and provisionally assume that the ionization is
constant through the sphere at the value it has at the "surface" of the sphere
at radius R . Compute the optical depth at the Lyman limit, TLL, from R to
the center.

Under these assumptions the optical depth scales as TLL "" M2 / R5 . If TLL ~
1, then a substantial fraction of the hydrogen in the sphere will be mostly
neutral so that the mass of H I will be very nearly equal to the total mass of
gas. If TLL is of order unity, then the H I content will be a moderate fraction
of the total gas mass, whereas if TLL ~ 1, the neutral fraction will not rise
much above its value at the surface of the cloud, and the H I content may
be very much lower than the total gas content.
Characterize the clouds in terms of M s , the total gas mass in units of
lOs M 0 , and the radius Rkpc in kpc. Even neglecting shielding, the gas would
be mostly ionized only if log Rkpc ~ 0.33 (log Ms + 1.0). The more strin-
gent and interesting limit for clouds to be fully ionized (i. e., TLL ~ 1) is
log Rkpc ~ 1.06 + 0.4 log Ms. Provided this condition is satisfied for log Ms
~ 0 or provided log Rkpc ~ 1.0 + 0.67 log Ms when log Ms ~ 0, then the
total amount of H I will be less than 106 M 0 .
If such fully ionized clouds satisfying this last condition are indeed re-
sponsible for at least some of the low red shift Lya clouds - as the observa-
tions suggest may be the case-then they are too distended to be bound by
self-gravitation. Since the expansion time is short compared to the Hubble
time for all reasonable cloud sizes, they must be continuously regenerated
or else must be pressure-confined or bound by dark matter. In the latter
case, they need not neccessarily have reached equilibrium, but, considering
static clouds with a ratio of baryonic-to-dark matter of ~ 5% and with a
baryonic virial velocity comparable to that of the gas, they occupy a region
defined roughly by log Rkpc ~ log Ms + 0.8. These structures remain highly
ioniz~d and contain total H' I masses which are very small provided log Rkpc
~ 1.6. Structures this size would still produce quite strong Lya lines for rays
passing near their centers, but for log Rkpc ~ 2.3 produce only weak Lya
lines even for central rays. Structures much larger than this are unphysical
and uninteresting; among other things they imply central densities lower
than the background density. Thus, the interesting sizes for such structures
appear to be 40 ~ Rkpc ~ 200. These structures would be the low red shift
analogues of the "mini-halo" models for the high redshift Lya forest clouds
originally proposed by (Rees 1986).
Whether such low-density and late-forming structures could in fact form
and survive to the present epoch is another question. Their chances are
probably greatest in isolation from groups and clusters, suggesting that they
would be weakly correlated or perhaps even anti-correlated with ordinary
 231

galaxies.
For pressure-confined clouds the radii could be significantly smaller, but
both the required mass and radius for a given neutral hydrogren column
density depend sensitively upon the confining presssure. The pressures re-
quired are not unreasonably high. For example, to produce an H I column
density of 10 14 cm- 2 for a radius of 30 kpc requires a pressure which is
f'V f'V

4 to 5 orders of magnitude smaller than those typically found in the center

of X-ray clusters (Donahue 1992).
Perhaps it is too much to expect simple static models to describe all Lya
absorption. In the case of the clouds at the Virgo distance, for example, if
we were in fact able to image the gas responsible for their absorption, might
we see a very chaotic distribution of streamers, ,plumes or shocks produced
by tidal and ram pressure interactions?

5.2. REMARKS ON POPULATIONS AND FUTURE OBSERVATIONAL PROGRAMS

The work of (Sargent et al. 1980) envisaged "two populations" of clouds:

(i) a set of pressure-confined low H I column density structures uncontami-
nated by the products of stellar nucleosynthesis and showing no correlation
with galaxies or protogalaxies, and (ii) a set of high H I column density
structures with metal enhancements at least comparable to or greater than
the Population II stars in our own galaxy and with velocity splittings sug-
gestive of clouds in the disks and halos of galaxies and protogalaxies. Since
then, mimerous authors have raised the question of whether the structures
giving rise to Lya absorption spanning over nine orders of magnitude in
H I column density and epochs from the present to f'V7% of the age of t,he
universe can be thought of as constituting a "single population" in contrast
with the "two population" scenario just described. "Two populations" have
also been invoked in the context of the evolution of the low column density
systems alone as a function of redshift with, for example, one mechanism
(e.g., pressure-confined clouds) dominating at high redshifts and another
( e.g., gravitationally-confined clouds) dominating at low redshifts.
As the work on the evolution of our galaxy illustrates, the universe Qn
astronomical scales is generally fairly complex, and it would be seem appro-
priate to move beyond a debate couched in terms of one vs. two populations
and concentrate instead on elucidating the various complex distributions of
parameters characterizing QSO absorption lines.

Specifically, we should attempt to answer the following questions:

(1.) What is the distribution of abundances in all QSO absorption line sys-
tems as a function of total column density and redshift?
232

The intriguing result obtained by (Lu 1991) that small column density
Lya forest lines have C IV lines with [CjH] ~ 10- 3 should be confirmed and
extended to more sensitive levels.

(2 . ) How does the surface density of H I clouds depend upon H I column

density and redshiJt? .

If the QSO absorption line Key Project is completed as currently envis-

aged, and as more work is done in the lower red shift regime accessible from
the ground (z ~ 1.6 - 2.0), we should be able to see any structure that is
present in the function describing the density of lines in redshift space and
equivalent width space. Over most of the HST domain, blending of the Lya
forest is not so severe as to demand high resolution work, but as the redshifts
approach"'" 1.6, the second generation HST instrument STIS could be used
to advantage. Enough observations from the ground over all accessible red-
shifts should also be carried out at echelle resolutions to avoid any question
of biasing the results due to blending.

(3.) What are the correlation properties of all QSO absorption systems among
themselves (and with objects which we can characterize as galazies) as a
function of total column density, redshiJt, and, in terms of the cross corre-
lation with galazies, the cross correlation as a function of galazy mass and
morphology?

In the cross-correlation analysis of Mor93, for example, it was not pos-

sible to examine the galaxy-galaxy correlation and cloud-galaxy correlation
if, for example only spirals were considered. It would be of interest to do
this, as and when images with sufficient spatial resolution become available.
Such a correlation may more nearly resemble the galaxy-galaxy correlation
among spirals.

(4.) How does the characteristic scale of clouds, as deduced by studying pairs,
change as a function of column density and redshift?

Do the clouds really cluster at sizes of a fraction of a Mpc and velocities

~ 100 km s- 1? With the marked increases in the sensitivity of spectro-
graphs and larger telescope apertures now at hand, new searches should
be conducted for fainter pairs from a few arcsecond separations up to "'" 1
arcminute. One very promising existing pair at ",10 arcsec separation in-
vites study (Crampton et al. 1988). Both the pair studied by (Shaver &
Robertson 1983) and the triplet studied by (Crotts 1989) should be reexam-
ined at better resolution and S IN. The same is true of Q2345+00, originally
studied by (Foltz et al. 1984) and subsequently by Steidel & Sargent (1990,
 233

1991) as well as the very close pair studied by (Smette et al. 1992). It will
be fascinating to see if one can really discern differences between lines of
sight passing through two impact parameters of the same cloud on the one
hand and clusters of different clouds on the other. These studies would be
especially interesting to carry out at very low redshifts with HST. Cruelly,
nature has interposed Lyman limit systems between ourselves and the two
most promising candidates .....

(5.) Studies of the detailed line profiles of the high redshift Lya forest clouds
should be continued.

Specifically, high-resolution ('" 10 km s-l or less) and high signal-to-noise

(",30 or greater per pixel) studies of at least two or three objects which in-
clude Ly,8 as well as Lya and which also cover an extensive region longward
of the Lya emission should be carried out. Such observations should set-
tle once and for all the lingering .question about what the smallest doppler
widths encountered among the Lya forest lines really are. Extensive coverage
to the red is required to help resolve ambiguities about whether the narrow
lines are metal lines or Lya lines. With such high S IN and resolution, and
with information from the Ly,8 transition in some cases, it should also be
possible to detect departures from Voigt profiles. The apparent occurrence
of Doppler widths as high as 50 km s-l has received far less attention than
the (possibly) very narrow ones, but they are also of interest. Such veloci-
ties clearly cannot be thermal. Do they represent a small number of cooler
clumps or smooth expansion? What are the highest velocities one could rea-
sonablyexpect (either expansion or contraction) in various scenarios? These
observations should also provide new information about the behavior of the
column density distribution in the regime below log N :::::: 13.0, which is of
intrinsic interest and is also important in detecting or setting limits on a
Gunn-Peterson trough.
Even with the data just described in hand, we may still.not fully under-
stand the nature of the Lya forest clouds. Individual interesting cases will
warrant intensive individual study, especially those ~ow redshift clouds that
appear to be remote from galaxies.

Acknowledgements
I wish to thank the following for providing information andlor helpful dis-
cussions during the preparation of this review: A. Dressler, G. Helou, S.
Morris, M. Rauch, A. Sandage, J. van Gorkom, J. Webb, D. York, and D .
Zaritsky. I thank M. Shull for calling my attention to a serious numerical
error in a preliminary draft of this article. Support from NASA contract
NAS5-30101 and NSF grant AST-9005117 is gratefully acknowledged.
234

References
Bahcall, J. N., Jannuzi, B. T., Schneider, D. P., Hartig, G. F., Bohlin, R., & Junkkarinen,
V. 1991, ApJ Letters, 377, L5
Bahcall, J. N., Jannuzi, B. T., Schneider, D. P., Hartig, G. F., & Green, R.F. 1992, ApJ,
397,68.
Bahcall, J., Bergeron, J., Boksenberg, A., Hartig, G., Jannuzi, B., Kirhakos, S., Sargent,
W., Savage, B., Schneider, D., Turnshek, D., Weymann, R., & Wolfe, A. 1993, ApJ
Suppl (in press).
Barcons, X., & Webb, J.K. 1991, MNRAS, 253,207.
Bechtold, J. 1987, in Proc. of Third lAP Workshop: High Redshift and Primeval Galaxies,
ed. J. Bergeron, D. Kunth, B. Rocca-Volmerange, J. Tran Thanh Van (Frontieres, Gif
sur Yvette Cedex, France) p. 397.
Bechtold, J. 1993, ApJ (in press).
Bingelli, B., Sandage, A.R., & Tammann, G.A. 1985, AJ, 90, 1681.
Blades, J.C. 1988, in QSO Absorption Lines: Probing the Universe, ed: Blades, J.C., Turn-
shek, D., Norman, C.A. 1988, (Cambridge: Cambridge University Press), p. 147.
ed: Blades, J.C., Turnshek, D., Norman, C.A. 1988, QSO Absorption Lines: Probing the
Universe, (Cambridge: Cambridge University Press), pp. 91-99; 71-87; 107-117 and
1-9.
Carignan, C., Beaulieu, S., & Freeman, K.C. 1990, AJ, 99,178.
Carswell, R.F., Lanzetta, K.M, Parnell, H. C., & Webb, J.K. 1991, ApJ, 371, 36.
Chaffee, F.H., Foltz, C.B., Bechtold, J., & Weymann, R.J. 1986, ApJ, 301, 116.
Charlton, J.C., Salpeter, E.E., & Hogan, C.J. 1993, ApJ, 402, 493.
Cooke, A. 1992 (Poster Paper #6, Day 2: July 7, 1992; this conference)
Crampton, D., Cowley, A.P., Hickson, P., Kindl, E., Wagner, R.M., Tyson, J.A., & Gul-
lixson, C. 1988, ApJ, 330, 184.
Crotts, A. P. S. 1989, ApJ, 336, 550.
Davis, M., & Peebles, P.J.E. 1983, ApJ, 267, 465.
Donahue, M. 1992, Invited Paper at this conference: Evolution of Clusters and the lntra-
cluster Medium •
Donahue, M., & Shull, J. M. 1991, ApJ, 383, 511.
Foltz, C.B., Weymann, R.J., Roser, H.J., & Chaffee, F.H. 1974, ApJ Letters, 281, L1.
Giallongo, E. MNRAS, 251, 541.
Giovanelli, R., & Haynes, M.P. 1989, ApJ Letters, 346, L5.
Kulkarni, V.P., & York, D.G. 1992, (Poster Paper #23, Day 2: July 7,1992; this confer-
ence)
Kutyrev, A.R., & Reynolds, J.R. 1989, ApJ Letters, 344, L9.
Lake, G., Schommer, R.A., & van Gorkom. J.H. 1990, AJ, 99, 547.
Lu, L. 1991, ApJ, 379, 99.
Lu, L., Wolfe, A. M., & Turnshek, D.A. 1991, ApJ, 367, 19.
Lynds, R. 1971, ApJ Letters, 164, L73.
Madau, P. 1992, ApJ Letters, 389, L1.
Maloney, P. 1992, ApJ Letters, 398, L89.
Maloney, P. 1993, ApJ (1993 in press)
Maraschi, L., Blades, J.C., Calanchi, C., Tanzi, E.G., & Treves, A. 1988, ApJ, 333, 660.
Meyer, D.M., & York, D.G. 1987, ApJ Letters, 315, L5.
Meyer, D.M., & York, D.G. 1992, ApJ Letters, 399, L121 and references therein.
Morris, S. L., Weymann, R. J., Savage, B. D., & Gilliland, R. L. 1991, ApJ Letters, 377,
L21.
Morris, S. L., Weymann, R. J., Dressler, A., McCarthy, P.J., Smith, B.A., Terrile, R.J.,
Giovanelli, R., & Irwin, M. 1993, (submitted to The Astrophysical Journal).
Murdoch, H.S., Hunstead, R;W., Pettini, M., & Blades, J.C. 1986, ApJ, 309,19.
Peterson, B.A. 1978, in IA U Symposium 79: The Large Scale Structure of the Universe,
ed. M.S. Longair and J .Einasto (Dordrecht: Reidel), p. 389.
Petitjean, P., Rauch, M., Carswell, R.F., Webb, J.K., & Lanzetta, K. 1993, paper given
 235

at Conference on Observational Cosmology, Milan, September 1992; to be published

in A.S.P. Conference series, editors G. Chincarini, A. Iovino, T. Maccacaro and D.
Maccagni.
Petitjean, P., Webb, J.K., Rauch, M., Carswell, R.F., & Lanzetta, K. 1993, MNRAS, (1993
in press).
Petitjean, P., Bergeron, J., Carswell, R.F., & Puget, J.L. 1993, MNRAS (1993 in press)
Pettini, M., Hunstead, R., Smith, L., & Mau, D. 1990, MNRAS, 246, 545.
Pettini, M., Boksenberg, A., & Hunstead, R. 1990, ApJ, 348, 48.
Rauch, M., Carswell, R.F., Robertson, J.G., Shaver, P.A., & Webb, J.K. 1990, MNRAS,
242,698.
Rauch, M., Carswell, R.F., Chaffee, F.H., Foltz, C.B., Webb, J.K., Weymann, R.J., Bech-
told, J., & Green, R.F. 1992, ApJ, 390,387.
Rauch, M., Carswell, R.F., Webb, J.K., & Weymann, R.J. 1993, MNRAS, 260,589.
R~es, M.J. 1986, MNRAS, 218, 25P.
Salzer, J.J. 1992, AJ, 103, 385.
Sandage, A.R., & Bingelli, B. 1984, AJ, 89, 919.
Sargent, W.L.W., Young, P.J., Boksenberg, A., & Tytler, D. 1980, ApJ Suppl., 42, 4l.
Sargent, W.L.W., Young, P.J., & Schneider, D.P. 1982, ApJ, 256, 374.
Shaver, P.A., & Robertson, J.G. 1983, ApJ Letters, 268, L57.
Smette, A., Surdej, J., Shaver, P.A., Foltz, C.B., Chaffee, F.H., Weymann, R.J., Williams,
R.E., & Magain, P. 1992, ApJ, 389, 39.
Songaila, A., Bryant, W., & Cowie, L.1. 1989, ApJ Letters, 345, L71.
Steidel, C.C., & Sargent, W.L.W. 1990, AJ, 99, 1963.
Steidel, C.C., & Sargent, W.L.W. 1991, AJ, 102, 1610.
Stocke, J.T., Donahue, M., Case, J., Shull, J. M., & Snow, T. P. 1991, ApJ, 374, 72.
Trevese, D., Giallongo, E., & Camurani, 1. 1992, ApJ, 398, 491.
Van Gorkom, J.H., Bahcall, J.N., Jannuzi, B., & Schneider, D. 1993, (in preparation).
Webb, J.K. 1987 in IA U Symposium 124: Observational Cosmology ed. A. Hewitt., G.
Burbidge, & L.Z. Fang. (Reidel, Dordrecht) p. 803.
Webb, J.K., & Larsen, J.P. 1987, in Proc. of Third lAP Workshop: High Redshift and
Primeval Galaxies, ed. J. Bergeron, D. Kunth, B. Rocca- Volmerange, J. Tran Thanh
Van (Frontieres, Gif sur Yvette Cedex, France) p. 419.
Webb, J.K., Barcons, X., Carswell, R.F., & Parnell, H. C. 1992, MNRAS, 255, 319.
Weittberg, D.H., Szomoru, A., Guhathakurta, P., & van Gorkom, J.H. 1991, ApJ Letters,
372, L13.
Weymann, R.J., Carswell, R.F., & Smith, M.G. 1981, Ann Rev Astron fJ Astroph, 19, 41.
Williams, T. 1992, (private communication) ; see also Schommer, R.A., Williams, T., &
Bahcall, J.N. 1993, (in preparation).
York, D.G. 1993, (private communication)
 QSO ABSORPTION LINES: IMPLICATIONS FOR GALAXY
FORMATION AND EVOLUTION
KENNETH M . LANZETTA
Center for Astrophysics and Space Sciences
University of California, San Diego

Abstract. QSO metal-line absorption systems are reviewed in the context of galaxy
formation and evolution. Attention is focused on the high column density "damped Lyo"
absorption systems, which dominate the cosmological mass density of neutral gas in the
universe and hence trace the bulk of material available for star formation to high redshifts .
Recent evidence of evolution of the gaseous content of the universe is interpreted in terms of
the global star formation history of galaxies. Physical properties of the gaseous component
of high-redshift galaxies, including chemical and molecular abundances and the presence
and strength of magnetic fields , are discussed.

1. Introduction
Several developments have over recent years transformed the study of QSO
absorption lines from an esoteric topic of only passing relevance to cosmology
to a rapidly maturing subject capable of producing results that bear directly
on scenarios of galaxy formation and evolution. First, a number of system-
atic, unbiased surveys for absorption lines in the spectra of background
QSOs have established fundamental statistical properties of the absorption
systems, relating to the size and spatial number density of the absorbing ob-
jects, the cosmological mass density contributed by the absorbing material,
and the evolution of these quantities with redshift. Second, sensitive imaging
and spectroscopic observations of galaxies responsible for absorption systems
have demonstrated beyond doubt that many (at least) of the systems are
associated with galaxies and therefore that QSO absorption lines provide a
unique and direct probe of the properties of galaxies to very large redshifts.
Third, high signal-to-noise ratio ( ~ 50) or high spectral resolution ( ,:S 25
km s-l) observations of certain key transitions have established (for a small
number of absorption systems) important physical properties of the absorb-
ing gas, including, for example, the chemical and molecular abundances. On
the basis of these and similar developments, galaxies at redshifts inaccessi-
ble to other methods of investigation have been admitted into the realm of
direct observational inquiry.
The main advantage provided by QSO absorption lines over more tradi-
tional means of observing distant galaxies is straightforward: Galaxies may
be studied in absorption more or less independently of redshift, whereas
galaxies may be studied in emission only subject to the strong selection
biases that result from surface brightness effects and K corrections. Thus,
while the most distant galaxies detected in emission near the thresholds of
flux-limited samples are necessarily draw from the bright tip of the galaxy

237
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, 237-262.
© 1993 Kluwer Academic Publishers.
238

60
~
Q)
40
~z 20

0
60
~
Q)
40
~Z 20

0
0 1 2 3
z
Fig. 1. Comparison of the redshift distributions of objects detected in spectroscopic
surveys for metal-line absorption systems (lower panel) and in deep surveys for
faint (20 < B < 24) field galaxies (upper panel). It is quite common to detect QSO
absorption systems at z > 3, whereas to date deep surveys for faint field galaxies
have failed to reveal any galaxies with z > 1. Based on redshifts of metal-line
absorption systems from Junkkarinen, Hewitt, & Burbidge (1991); Sargent, Steidel,
& Boksenberg (1989); and Lanzetta (1991) and on red shifts of faint field galaxies
from Broadhurst, Ellis, & Shanks (1988); Colless et al. (1990); and Cowie, Songaila,
& Hu (1991).

luminosity function and are therefore not representative of the bulk of the
galaxy population, galaxies detected in absorption are drawn from a wide
range of luminosities and are therefore representative of the bulk of the
galaxy population. Practically, this advantage is dramatically illustrated by
noting that it is quite common to detect QSO absorption systems at z > 3,
whereas to date deep surveys for faint field galaxies have failed to reveal
any galaxies with z > 1 (see Figure 1). The important implication of this
result is clear: QSO absorption lines have so far provided the only means of
routinely studying normal galazies at redshifts z > 1.
The emphasis of this review is on results derived from spectroscopic ob-
servations of QSO absorption lines and on the implications of these results
for scenarios of galaxy formation and evolution. Specifically, the emphasis of
this review is on the so-called "metal-line" absorption systems, which have
neutral hydrogen column densities satisfying N(HI) .2: 10 16 cm- 2 , show ab-
sorption lines from common ions of abundant metal species, and for which
 239

the connection with galaxies is now firmly established. The emphasis of a

complementary review in this volume by Steidel is on results derived from
imaging and spectroscopic observations of galaxies responsible for metal-
line absorption systems, and the emphases of complementary reviews in this
volume by Bajtlik, Weymann, and Bechtold are on the lower column den-
sity "Lya-forest" absorption systems, for which the connection (or lack of
connection) with galaxies is not yet clear.

2. Metal-Line Absorption Systems: Overview

2.1. SPECTROSCOPIC SURVEYS FOR METAL-LINE ABSORPTION SYSTEMS

Metal-line absorption systems are usually detected on the basis of the strong
resonance-line doublets of C IV AA1548, 1550 or Mg II AA2796, 2800 or on the
basis of optically thick Lyman-limit discontinuities or Lya absorption lines
that have been broadened by radiation damping to large equivalent widths.
(Absorption systems detected on the basis of optically thick Lyman-limit
discontinuities or damped Lya absorption lines almost invariably show cor-
responding metal absorption lines and hence are considered as metal-line
absorption systems.) These absorption features are relatively strong, and so
can be readily detected even in data of only moderate spectral resolution
and signal-to-noise ratio, and these absorption features exhibit character-
istic spectral signatures that can be easily recognized and hence provide
unambiguous redshift identifications. Several of these absorption features
are illustrated in the spectra of the high-redshift QSOs shown in Figures 2
and 3.
It is, in general, quite difficult to determine column densities based on
measurements of metal-line absorption profiles. For this reason, surveys for
C IV and Mg II absorption lines are complete to some rest-frame equivalent
width threshold rather than to some column density threshold. Surveys for
C IV absorption lines are usually complete to rest-frame equivalent widths
(of the stronger member of the doublet) of W > 0 .15 A or W > 0.3 A,
whereas surveys for Mg II absorption lines are usually complete to W > 0.3
A or W > 0.6 A.
In contrast, it is in general quite straightforward to determine column
densities (or at least limits) based on measurements of Lyman-limit ab-
sorption discontinuities or damped Lya absorption lines. For this reason,
surveys for Lyman-limit absorption discontinuities and damped Lya absorp-
tion lines are complete to some neutral hydrogen column density threshold.
Surveys for Lyman-limit absorption discontinuities are usually complete to
N(HI) > 2 X 10 17 cm- 2 , which corresponds to Lyman-limit optical depths of
7"LL > 1, and surveys for damped Lya absorption lines are usually complete
to N(HI) > 2 X 10 20 cm- 2 , which corresponds to rest-frame Lya equivalent
 240

10
)fa n AA2796,280S C IV Ul648,
8 :6 - 0.6203 :6 - 1.9666

~
r..c
Q)
6 n-..aJj ... n
.:::
-
1\1
~
Q)
4

0
4100 4200 4300 4400 4500 4600
Wavelength (1)

Fig. 2. Spectrum of the z = 2.43 QSO 0149+336. The C IV 'u1548,1550 reso-

nance-line doublet at z = 1.9556 and the Mg II ,\,\2796, 2803 resonance-line doublet
at z = 0.5203 are indicated. Adapted from Wolfe et al. (1993).
10~~~~~~~~~--~~~~~~~~~~r-~

Lyman Limit HI U216

8 :6 - 2.994 :6 - 2.678

o
3500 4000 4500 5000
Wavelength (1)
Fig. 3. Spectrum of the z = 3.27 QSO 1159+01. The Lyman-limit discontinuity at
z = 2.994 and the Lya absorption line at z = 2.678 are indicated. Adapted from
Lanzetta et al. (1991).
 241

widths ofW > 10 AI.

Metal-line absorption systems are often referred to on the basis of the
absorption features by which the systems were discovered, hence systems
detected on the basis of C IV absorption lines are referred to as C IV -selected
absorption systems, systems detected on the basis of Mg II absorption lines
are referred to as Mg II-selected absorption systems, systems detected on the
basis of Lyman-limit absorption discontinuities are referred to as Lyman-
limit absorption systems, and systems detected on the basis of damped Lya
absorption lines are referred to as damped Lya absorption systems. It is im-
portant to bear in mind, however, that there is substantial overlap between
these empirical categories and that, for example, C IV-selected absorption
systems can (and often do) show Mg II absorption lines and Mg II-selected
absorption systems can (and usually do) show C IV absorption lines.
Detailed photoionization models (Wolfe 1986; Bergeron & Stasinska 1986;
Steidel 1990 ) indicate that a more physically meaningful classification scheme
is based on neutral hydrogen column density, with the important column
density threshold of N(HI) ~ 2 X 10 17 cm- 2 separating gas that is opti-
cally thin at the Lyman limit and gives rise only to high-ionization lines
like C IV from gas that is optically thick at the Lyman limit and gives rise
(at least potentially) to both high-ionization lines like C IV and to low-
ionization lines like Mg II. On this basis, absorption systems are sometimes
referred to as "optically thin" or "optically thick" systems depending on the
Lyman-limit optical depths and on the presence or absence oflow-ionization
absorption lines (Lanzetta, Turnshek, & Wolfe 1987). These detailed pho-
toionization models also indicate that most metal-line absorption systems
are predominantly ionized but that the higher column density damped Lya
abosrption-line systems are predominantly neutral.
Over the past ten years or so, a considerable amount of observational
effort has been devoted toward obtaining systematic, unbiased surveys for
absorption lines in the spectra of background QSOs, from which fundamen-
tal statistical properties of the absorption systems are derived. In this con-
text, systematic means that the surveys are conducted at moderate spectral
resolution toward a relatively large number of background QSOs (or equiv-
alently BL Lac objects or Seyfert galaxies) in order to maximize the total
redshift coverage, and unbiased means that the surveys are conducted with-
out a priori knowledge of the presence or absence of absorption toward the
background QSOs in order to derive homogeneous samples of absorption
1 For small enough doppler parameters, Lya lines of neutral hydrogen column density
satisfying N(HI) :::::: 5 x 10 18 cm- 2 or even less could be substantially broadened by radi-
ation damping and hence considered as damped Lya absorption lines. But in the present
context Lya absorption lines are considered as damEed only if they arise from neutral
hydrogen column densities satisfying N(HI) 2': 2 x 10 0 em -2. Historically, this threshold
was chosen to allow direct comparison with the 21 cm studies of nearby galaxies by Bosma
(1981), which was also sensitive to N(HI) 2': 2 x 10 20 cm- 2 •
242

TABLE I
Systematic, Unbiased Surveys for Metal-Line Absorption Systems

Redshift Background
Survey Type Range QSOs
C IV -Selected Absorption Systems
Young, Sargent, & Boksenberg {1982} .. . .. .. ... 1.2 - 2.7 30
Foltz et al. (1986) .... . : ....... .. ...... ..... ... 1.1 - 1.9 31
Sargent, Boksenberg, & Steidel (1988) . ...... .. . 1.2 - 3.5 52
Steidel (1990) .... ........ ............ .. .... .... 2.2 - 3.7 11
Mg II-Selected Absorption Systems
Tytler et al. {1987} ............................ 0.2 - 0.8 24
Lanzetta, Turnshek, & Wolfe (1987) ... ... ...... 1.3 - 2.2 32
Sargent, Steidel, & Boksenberg (1988) ....... . .. 0.6 - 1.4 9
Sargent, Steidel, & Boksenberg (1989) . . ... . . . . '. 0.7 - 1.5 41
Steidel & Sargent (1992) ...... .. .. .... .. ...... . 0.2 - 2.2 103
Lyman-Limit Absorption Systems
Tytler (1982) .. ........................ .. .. .. .. 0.4 - 3.5
Bechtold et al. (1984) .. .. .. ................. .. 0.4 - 2.2
Lanzetta (1988) .. .... .... ... .. .... ... .... .. .. . . 0.4 - 3.8
Sargent, Steidel, & Boksenberg (1989) . .. ... . ... 2.5 - 4.1
Lanzetta (1991) .. .. .. .. ..... .... .. ... .... ..... . 2.6 - 3.8
Damped Lya Absorption Systems
Wolfe et al. (1986) .. .......... .. ............. . 1.6 - 3.3 68
Lanzetta et al. (1991) ......................... 1.6 - 4.1 101
Lanzetta, Turnshek, & Wolfe (1993) ............ 0.0 - 1.6 167
aCompiled from the literature.

systems suitable for statistical analysis. The defining characteristics of the

major systematic, unbiased surveys for metal-line absorption systems are
summarized in Table I.
Analysis of these spectroscopic surveys yields the redshifts and "strengths"
(Le. rest-frame equivalent widths for C IV- or Mg II-selected absorption sys-
tems or neutral hydrogen column densities for Lyman-limit or damped Lyo:
absorption systems) of the detected absorption systems and an account of
the redshift path length over which the absorption systems were or could
have been detected. This information is used to derive quantities related to
the size and spatial number density of the absorbing objects and the cosmo-
logical mass density contributed by the absorbing material. These quantities
are discussed in §§ 2.2 and 2.3.

2 .2. SIZE AND SPATIAL NUMBER DENSITY OF ABSORBING OBJECTS

The rate of incidence of absorption systems is characterized by the system

number density n(z), which is defined in such a way that n(z)dz is the
 243

number of systems per line of sight with redshift in the interval z to z +

dz. Due to the effects of the expansion and deceleration of the universe,
the number density redshift distribution expected in a standard Friedmann
cosmology (with zero cosmological constant) for non-evol1ling objects that
are jized in co-moving coordinates is given by,

n(z) = no(1 + z)(l + 2qoz)-1/2 , (1)

where no is the system number density at z = 0 and qo is the deceleration
parameter. The observed number density redshift distribution is often fitted
as

n(z) = no(l + z)'Y , (2)

in which case intrinsic evolution of the absorbing objects is indicated by

·departures from the no-evolution values: 'Y = 1 for qo = 0 and 'Y = 0.5 for
qo = 0.5.
On the basis of absorption measurements alone, it is not possible to deter-
mine the size and spatial number density of absorbing objects independently
but rather only the product of the projected absorption cross section and
.the spatial number density of absorbing objects. It is therefore common
to set the spatial number density of absorbing objects by assuming that
the absorbers are associated with galaxies and that galaxies are fixed in
co-moving coordinates, thereby allowing the projected absorption cross sec-
tion, or equivalently the characteristic size of the absorbing objects, to be
determined.
In particular, assuming that the radii of the absorbing objects scale with
galaxy luminosity L according to some relationship R( L) and that some
fraction € of galaxies participate in absorption with "covering factor" K-, the
system number density is given by,

(3)

where c is the speed of light, Ho is the Hubble constant, C)(L) is the galaxy
luminosity function, and where K specifies the geometrical model and takes
on the value K = 1 for 51= 1 "rical geometry and the value K = 0.5 for disk
geometry. If one adopts a a "Schechter" (1976) galaxy luminosity function,

c) (fJ = c)* (fJ -6 exp ( - f*) , (4)

and a "Holmberg" (1975) correlation between galaxy luminosity and radius,

(5)
 244

- 0

+ ----
-!J,
1:1

-tIO
0
-1 ~~I

I t _________________________

.. --- ------
~-----

-_ .. ----- .. --- - - - - - - - - - - - - - --
--:;: -.::..:-:,.:.-- - - -
-2
0 1 2 3 4
z
Fig. 4. Logarithm ofthe system number density n(z) vs. redshift z forC IV-selected
absorption systems (filled squares), Mg II-selected absorption systems (open
squares), Lyman-limit absorption systems (filled circles), and damped Lya absorp-
tion systems (open circles). Vertical bars indicate 10- uncertainties and horizontal
bars indicate bin sizes. Data taken from the various spectroscopic surveys listed
in Table 1. Curves show the number density redshift distributions predicted from
Equation (3) for absorbing objects of radius equal to the Holmberg radius of nearby
spiral galaxies, assuming spherical geometry and that all galaxies participate in ab-
sorption with unit covering factor, for qo = 0 (dotted curve) and qo = 0.5 (dashed
curve) .

then from Equation (3) the characteristic radius R. (z) of the absorbing
objects is given by,

R • -- [ ~ 1 + z)(1 +
( 2qoz)- 1/2 .-.;. K r( )]-1/2
Ho
( )
n z
€'f.K 7r 1 + 2t _ s , (6)

where r(z) is the gamma function and where +., s, and t have values near
+. = 1.2 X 10- 2 h3 Mpc- 3 , S = 1.25 (Kirshner et ai. 1983), and t = 0.4
(Holmb€'rg 1975). Here, h is the dimensionless Hubble constant h in units
of 100 km S-1 Mpc- 3 •
The observed number density redshift distributions of the C IV-selected,
Mg II-selected, Lyman-limit, and damped Lya absorption systems are shown
in Figure 3 together with the number density redshift distributions predicted
from Equation (3) for absorbing objects of characteristic radius equal to the
optical (or "Holmberg") radius of nearby spiral galaxies, assuming spherical
geometry and that all galaxies participate in absorption with unit covering
 245

TABLE II
Characteristic Radii of Metal-Line Absorption Systems

R.
Absorption Systems (z) [(€K)-1/2h- 1 kpc]

C IV-selected . .. ............ . .. 3.0 66

1.5 98
Mg II-selected ................. 1.5 66
0.5 62
Lyman-limit ................... 3,0 82
0.5 74
Damped Lya .................. 3.0 37
0.5 28

factor (Le., assuming K = 1, € = 1, and K = 1). It is apparent that the

rate of incidence of all types of metal-line absorption systems exceeds the
predicted rate by a significant factor (and that for € < 1 or K < 1 this factor
is even larger). It is therefore usually concluded that the low column density
C IV-selected, Mg II-selected, and Lyman-limit absorption systems arise in
eztended galactic halos that are 5 - 10 times the optical extent of nearby
spiral galaxies, and that the high column density damped Lya absorption
systems arise in eztended galactic disks that are 2 - 3 times the optical extent
of nearby galaxies.
Characteristic radii of metal-line absorption systems derived from Figure
3 on the basis of Equation (6) are listed in Table II, assuming spherical
geometry (i.e. assuming K = 1) for the C IV-selected, Mg II-selected and
Lyman-limit absorption systems and assuming disk geometry (i.e assuming
K = 0.5) for the damped Lya absorption systems . The values of R .. listed
in Table II are appropriate for a deceleration parameter of qo = 0.5.
Sensitive imaging and spectroscopic observations of galaxies responsible
for Mg II-selected absorption systems at z ~ 0.5 (Bergeron & Boisse 1991;
Steidel, this volume) find a distribution of projected impact parameters be-
tween the absorbing galaxies and the background QSOs that is in remark-
able agreement with the characteristic radii indicated in Table II. This result
lends strong support to the interpretation that the Mg II-selected absorption
systems (at least) arise in extended galactic halos that are several times the
optical extent of nearby spiral galaxies.

2.3. COSMOLOGICAL MASS DENSITY OF ABSORBING MATERIAL

The cosmological mass density of absorbing material is determined directly
from spectroscopic surveys for absorption systems. This can be seen by con-
246

sidering a population of absorption systems of some given neutral hydrogen

colwnn density No. The proper mass density of neutral hydrogen contributed
by these absorption systems at epoch z is obtained by taking the product
n( z )dz-which gives the mean number of absorption systems intercepted by
a line of sight over the redshift interval z to z+dz, multiplying by the column
density No-which then gives the mean colwnn density traversed by a line
of sight over the redshift interval z to z + dz, multiplying by the mass of the
hydrogen atom-which then gives the mean mass colwnn density traversed
by a line of sight over the redshift interval z to z + dz, and dividing by cdt
(where dt is the cosmic time interval corresponding to dz )-which finally
yields the desired proper mass density.
It is usually more convenient to form the cosmological mass density pa-
rameter fig of neutral gas from the proper mass density of neutral hydrogen
by multiplying by a correction factor 1/(1 - Y) ~ 1.33 to allow for a he-
lium abundance Y = 0.25 by mass, dividing by (1 + z)3 to allow for the
expansion of the universe, and dividing by the current critical density to
yield a dimensionless number. The cosmological mass density parameter fig
thus gives the co-moving mass density of neutral gas in units of the current
critical density.
In practice, of course, the cosmological mass density parameter is deter-
mined not for absorption systems of some given neutral hydrogen column
density No but rather for absorption systems that span a range of neutral
hydrogen colwnn densities. The distribution of neutral hydrogen column
densities is characterized by the column density distribution function f(N),
which is defined in such a way that f(N)dN dX is the number of absorbers
per line of sight with neutral hydrogen column density in the interval N to
N + dN over an absorption distance 2 interval dX. Over the column den-
sity range of interest [N(HI) ~ 10 16 cm- 2 ] the column density distribution
function is observed to be well represented by a power-law form as,
f(N) = BN-{3 , (7)
with a power-law index f3 in the range 1.2 ~ f3 ~ 1.8 (Sargent, Steidel, &
Boksenberg 1989; Lanzetta 1991; Lanzetta et al. 1991). This functionalform
of the column density distribution function appears to extend to column
densities at least as large as N(HI) ~ 6 X 10 21 cm- 2 (Lanzetta et al. 1991).
In terms of the column density distribution function, the cosmological mass
density parameter is given by,

fig = Ho pm.H
c Pcrlt J
r
Nmax

Nmin
N f(N)dN , (8)

2 The absorption distance X is given by X(z) = (1/2)[(1 + Z)2 - 1] for qo = 0 and

X(z) = (2/3)[(1 + z?/2 - 1] for qo = 0.5 and is chosen in such a way that nonevolving
objects that are fixed in co-moving coordinates have a constant density per unit absorption
distance interval.
 247

where mH is the mass of the hydrogen atom, Pcrit is the current critical den-
sity, and Nmin and N max are the minimum and maximum neutral hydrogen
column densities of interest, respectively.
The cosmological mass density parameter defined by Equation (8) has
several interesting and important properties as follows:

1. The cosmological mass density parameter is determined without assump-

tion about the sizes, shapes, or covering factors of the absorbing objects.
The only inherent ambiguity in the value of the cosmological mass density
parameter is in the choice of cosmological model.

2. Due to the relatively shallow slope"of the column density distribution func-
tion, the cosmological mass density parameter is dominated by the highest
column density absorption systems under consideration. This can be seen
by substituting the functional form of f(N) given in Equation (7) into the
expression for flg given in Equation (8) to yield,

fl = no jlmH ~ (N2-f3 _ N 2-:f3) (9)

g C Pcrit 2 - f3 max mm'

which is valid for f3 i: 2. For f3 < 2, as is indicated by the observations, the

value of this expression is controlled by the upper rather than by the lower
column density limit. The high column density damped Lya absorption sys-
tems thus dominate the mass density of neutral gas in the universe.

3. Because surveys for damped Lya absorption systems sample all high col-
umn density neutral hydrogen in the universe, the cosmological mass density
parameter is sensitive without bias to the total mass density of neutral gas
in the universe. This includes neutral gas associated with spiral galaxies,
elliptical galaxies, dwarf galaxies, tidal debris, and intergalactic clouds; no
significant source of neutral gas is excluded.

The cosmological mass density parameter defined by Equation (8) is, of

course, insensitive to the mass density of ionized gas in the universe. Results
of detailed photoionization models suggest that the mass density of ionized
gas contributed by the metal-line absorption systems may be comparable to
the mass density of neutral gas contributed by the damped Lya absorption
systems (Steidel 1990). Nevertheless, the cosmological mass density param-
eter represents a fundamental cosmological quantity that may provide im-
portant constraints on scenarios of galaxy formation and evolution. These
constraints are discussed in § 3.
248

3. Evolution of the Gaseous Content of the Universe

3.1. THEORETICAL FRAMEWORK
The results of § 2.3 indicate that the damped Lyo: absorption systems dom-
inate the mass density of neutral gas in the universe. Because stars form
from neutral gas, this has the important implication that the damped Lyo:
absorption systems trace the bulk of material available for star formation to
large redshifts. In principle, observations of the damped Lyo: absorption sys-
tems can therefore constrain the history of gas consumption, star formation,
and chemical enrichment-not of anyone particular galaxy or even anyone
particular type of galaxy, but rather of the universe as a whole, averaged
over all galaxies in the universe.
This idea can be cast into a quantitative context by writing the equations
of "cosmic" chemical evolution, which are simply the standard equations of
galactic chemical evolution (e.g., Tinsley 1980) conceptually averaged over
all galaxies in the universe and expressed in terms of co-moving densities
divided by the current critical density:

d~. =(1 - R)1/1 , (10)

d!lg ( )
Tt=-l-R1/1+4>, (11)

dZ
!lgdi =y(l - R)1/1 - 4>Z , (12)
where !lg and !l .. are the cosmological mass density parameters of gas and
stars, respectively, t is the cosmic time, 1/1 is the co-moving star formation
rate density, or the rate at which mass density is consumed into stars, 4> is
the net co-moving "accretion" rate density, or the rate at which mass density
is "accreted" onto galaxies, R is the "returned fraction," or the mass fraction
of each generation of star formation that is returned as gas, y is the "yield,"
or the mass fraction of each generation of star formation that is returned
as metals, Z is the metallicity defined in such a way that the cosmological
mass density parameter !lm of metals is given by !lm = Z!lg, and where the
instantaneous recycling approximation has been adopted. These equations
are independent of the Hubble constant but depend on the deceleration
parameter through t.
The goal of the observations is to determine or constrain the quantities in
the equations of cosmic chemical evolution that relate directly to evolution of
the gaseous content of the universe: the cosmological mass density parameter
!lg( z) (which is determined based on spectroscopic surveys for damped Lyo:
absorption systems), the metallicity Z(z) (which is determined based on
measurements of metal abundances of damped Lyo: absorption systems),
 249

and the net accretion rate density 4>( z) (which is perhaps constrained based
on indirect arguments). In the context of Equations (10), (11), and (12),
these quantities may be combined to infer other important quantities that
are not directly accessible to observation: the cosmological mass density
parameter of stars n~(z), the star formation rate density ¢(z), and the
yield y. Together these quantities form the basis upon which it is possible
to address such questions as: When and over what time scale did the bulk
of star formation in the universe take place? When and over what time
scale were the bulk of metals in the universe produced? And how do these
quantities relate to scenarios of the global chemical enrichment history of
the universe?

3.2. OBSERVATIONAL RESULTS

Spectroscopic surveys for damped Lya absorption systems have been con-
ducted both at optical wavelengths (based on observations obtained with
ground-based telescopes) and at ultraviolet wavelengths (based on archival
observations obtained with the International Ultraviolet Ezplorer) , from
which evolution of the damped Lya absorption systems has been examined
over the redshift interval 0 :;., z :;., 3.5 (see Table I). The redshift path density
g( z), which gives the number of lines of sight that have been surveyed for
damped Lya absorption systems as a function of redshift, is shown for the
current surveys in Figure 5. Roughly 40 damped Lya absorption systems
have as yet been detected, from along 156 lines of sight surveyed at optical
wavelengths and 166 lines of sight surveyed at ultraviolet wavelengths.
Because damped Lya absorption systems occur infrequently compared
with other types of metal-line absorption systems (see Figure 4), it is rel-
atively difficult to determine the properties of the damped Lya absorption
systems as a function of redshift. But based on the most recent analysis of
the available data (Lanzetta, Turnshek, & Wolfe 1993), the following results
seem to be established with a reasonable degree of certainty:
1. The number density redshift distribution of the damped Lya ab-
sorption systems (shown in Figure 4) is consistent with the prediction for
nonevolving absorbers, indicating that the product of the absorption 9"oss
section and the co-moving spatial number density of the damped Lya ab-
sorption systems does not evolve strongly from z ~ 3.5 to z ~ o.
2. The cosmological mass density parameter of the damped Lya absorp-
tion systems (shown in Figure 6) decreases significantly from z ~ 3.5 to
z ~ 0, indicating that the co-moving mass density of neutral gas in the
universe decreases significantly from z ~ 3.5 to z ~ o.
3. The fraction of very high column density [N(HI) > 10 21 cm- 2 ] damped
Lya absorption systems (shown in Figure 7) decreases steadily from z ~ 3.5
to z ~ 0, indicating that evolution of the cosmological mass density con-
 250

150

-
~100
~

50

o~~~~~~~~~~~~~~~~~~--~~

o 1 2 3 4
z
Fig. 5. Redshift path density g(z) vs. redshift z for damped Lya absorption sys-
tems. Observations at optical wavelengths contribute to the red shift path density
at z > 1.6, and observations at ultraviolet wavelengths contribute to the redshift
path density at z < 1.6. Based on the spectroscopic surveys listed in Table 1.

,.
o
3

-
-' 2
X

z
Fig. 6. Cosmological mass density parameter Og(z) of damped Lya absorption
systems vs. red shift z. Vertical bars indicate 10' uncertainties and horizontal bars
indicate bin sizes. A deceleration parameter of qo = 0.5 has been assumed. Adapted
from Lanzetta, Turnshek, & Wolfe (1993).
 251

0.8

~
0 0.6
:fj
~
~ 0.4

0.2

0
0 1 2 3 4
z
Fig. 7. Fraction of high column density [N(HI) ::: 10 21 cm- 2 ) damped Lya absorp-
tion systems vs. red shift z. Adapted from Lanzetta, Thrnshek, & Wolfe (1993).

tributed by the damped Lya absorption systems results from a systematic

decrease in the incidence of high column density absorption systems with
decreasing redshift.
Together these results imply that the average mass of neutral gas per
absorption system decreases with time. This is qualitatively consistent with
the hypothesis that the damped Lya absorption systems trace the gas con-
sumption and hence star formation history of the universe over the redshift
interval 0 ~ z :s
3.5. This interpretation is also supported by two additional
results:
4. The cosmological mass density parameter of the damped Lya absorp-
tion systems at z :::::: 3.5 is roughly comparable to the mass density parameter
of stars in nearby galaxies. The mass density parameter 0.,( z = 0) of stars
in nearby galaxies can be obtained as,

Os(Z=O)=LB (LM) BPcrit'

-1
(13)
where LB is the B-band galaxy luminosity density and (MjL)B is the B-
band mass-to-light ratio of visible matter. Adopting the value
LB = 1.6 X 10 8 h L0B Mpc- 3 , (14)
recently advocated by Binggeli, Sandage, & Tammann (1988) and the value

(~) B = (5 ± 2) (~:) , (15)

252

recently advocated by Gnedin & Ostriker (1992), then the value of n.(z = 0)
is found to be

(16)
Comparison with Figure 6 shows that n.(z = 0) ~ ng(z ~ 3.5), which
strongly suggests that the gas observed at z ~ 3.5 forms the stars that
are observed in nearby galaxies. This result also indicates that to within
observational uncertainty there is no evidence for a net increase or decrease
of the total (gas plus stars) mass of galaxies from z ~ 3.5 to the present,
thus suggesting that the net accretion rate density ¢>( z) can be neglected
over the redshift interval 0 < z < 3.5.
5. Evolution of the fraction of very high column density damped Lya
absorption systems is consistent with what is expected if star formation in
the absorption systems proceeds as in nearby galaxies. Global star formation
in nearby galaxies can be described by a modified "Schmidt" (1959) law, in
which the star formation rate is proportional to some positive power m of the
total column density (Kennicutt 1989). Assuming a similar description holds
also for the damped Lya absorption systems, then due to star formation
column densities should evolve as

dN = -CN m (17)
dt '
where C is a parameter that characterizes the star formation rate. Adopt-
ing an initial column density distribution at some time t = 0 according to
the power-law form of Equation (7), it is straightforw<!-rd to show that the
column density distribution function at some later time t is given by

[1 - (m _ I)N
-m-{3
f(N, t) = BN-{3 m - 1 Ct] 1-= , (18)

from which it is seen that for m > 1 star formation preferentially affects the
high column density end of the column density distribution function. The
observed evolution of the column density distribution function (shown in
Figure 8) indicates that in fact the high column density end of the column
density distribution function decreases with time in a way that is qualita-
tively consistent with the prediction of Equation (18) and with the inter-
pretation that evolution of the cosmological mass density parameter results
from the effects of star formation in galaxies.
On the basis of these observations it is at least plausible that evolution of
the gaseous content of the universe has been detected and can be reasonably
attributed to the effects of star formation in galaxies. This result then has
implications bearing on the epoch and time scale of galaxy formation and
on the chemical evolution of the universe. These implications are discussed
in § 3.3 and § 4.1, respectively.
 253
-21 -21 -21

t;'

~
a
-22
ett -
"S
~
-22 r
+
+ "S
~
-22

t+ I
+-
~ ~ ~
..... -23 - ..... - 23 .... -23
1 1 toO
.2

-24 r T -24 r -24

20 20.5 21 21.5 22 20 20.5 21 21.5 22 20 20.5 21 21.5 22

log N (em.--> 10gN (em-l) 10gN (cm4

Fig. 8. Logarithm of the column density distribution function f(N) vs. logarithm
of the column density N evaluated at three different epochs. Left panel shows
o < z < 1.5, center panel shows 1.5 < z < 2.5, and right panel shows 2.5 < z < 3.5 .
Vertical bars indicate 10- uncertainties and horizontal bars indicate bin sizes.

3 .3 . EpOCH AND TIME SCALE OF GALAXY FORMATION

Under the interpretation that the observed evolution of the cosmological

mass density parameter results due to the effects of star formation in galax-
ies, then the observed evolution of H.( z) directly indicates both a character-
istic epoch and a characteristic time scale of star formation. The character-
istic epoch of star formation is expressed in terms of a quantity which is z. ,
defined as the redshift by which ha1f of the current stellar mass of galaxies
had formed (Peebles 1988). If the net accretion rate density </>(z) can be
neglected, then from Figure 6 it is seen that z. occurs roughly within the
redshift interval 2 ~ z. :::, 3. This measurement represents the first direct ob-
servational determination of z•. From Figure 6, it is furthermore seen that,
for z :::., 1.5, Hg( z ) appears to evolve on a very rapid time scale. In fact, for
z :::., 1.5 the evolution of Og( z) can be described by an exponential in cosmic
time, with a time scale T given by T ~ 0.5 X 109 h- 1 yr for qo = 0 .5 or
254

T ~ 1 X 109 h- 1 yr for qo = 0 (Lanzetta, Turnshek, & Wolfe 1993). This

time scale is much shorter than the time scales characteristic of disk evo-
lution (e.g., Ostriker & Thuan 1975), perhaps suggesting that the observed
evolution of !1 g ( z) may relate not only to disk formation and evolution but
also to spheroid formation and evolution. If the net accretion rate density
</>( z) cannot be neglected, then for 4>( z) > 0 the value of z. given above
corresponds to a lower limit to the characteristic epoch of star formation in
galaxies and the values of T given above correspond to upper limits to time
scale for star formation in galaxies.
Irrespective of the interpretation of the observed evolution of the cosmo-
logical mass density parameter, the rough correspondence between Og( z ~
3.5) and O,,(z = 0) indicates that a quantity of gas sufficient to form the
entire stellar content of nearby galaxies was already present in galaxies by
Z ~ 3.5. This result contrasts with the predictions of recent models of hi-
erarchical galaxy formation under cold dark matter scenarios (White 1990;
Cole 1991; White & Frenk 1991). A major result of these models-which
combine analytical descriptions of radiative cooling of gas with analytical
descriptions of evolving distributions of dark matter halos-is that gas can
cool very rapidly in the low binding energy dark matter halos that predomi-
nate at high redshifts (z ~ 2), leaving little gas to form large galaxies in the
high binding energy dark matter halos that predominate at lower redshifts
(z:;' 2). To suppress this rapid cooling of gas, these models invoke super-
nova driven "feedback" mechanisms to expel gas from low binding energy
dark matter halos. The observation that Og(z ~ 3.5) ~ O,,(z = 0) indicates
that in fact a significant quantity of gas had cooled into collapsed objects by
z > 2, perhaps suggesting either that the supernova driven feedback mech-
anisms operate with a low duty cycle or that the proportion of high binding
energy dark matter halos at high red shift is larger than what is predicted
under cold dark matter scenarios.

4. Physical Properties of the Gaseous Component of

High-Redshift Galaxies
4.1. CHEMICAL CONTENT OF HIGH-REDSHIFT GAS

Observations of metal-line absorption systems can, in principle, directly

trace the chemical evolution of gas in the universe. Combining these mea-
surements with the measurements ofthe gas consumption and star formation
histories probed by the cosmological mass density parameter will ultimately
allow a complete description of the chemical evolution of the universe.
For several reasons, measurement of the chemical abundances of absorp-
tion systems is a notoriously difficult problem. First, the absorption-line
profiles of metal-line absorption systems very often show complex structure
 255

on velocity scales ranging from several km S-l up to many hundreds of km

s-1, and the commonly observed ions of C, Mg, Si, and Fe are usually very
highly saturated, making it difficult or impossible to derive reliable column
densities based on a simple curve-of-growth analysis applied to observations
of low spectral resolution. Second, ionization corrections are potentially very
large, and attempts to model these corrections must contend with the un-
known strength and spectral shape of the ionizing radiation field, the un-
known geometry and density distribution of the absorbing material, and
(in some cases) poorly determined atomic parameters. Third, a significant
fraction of metals could be depleted onto dust grains, although observations
of selective extinction toward damped Lya absorption systems indicate a
dust-to-gas ratio significantly less than the Galactic value (Fall & Pei 1989;
Fall, Pei, & McMahon 1989; Pei, Fall, & Bechtold 1991).
These difficulties have been addressed most persuasively by means of high
signal-to-noise ratio ( ;::, 50) spectroscopic observations of Zn II AA2025, 2062
and Cr II AAA2055, 2061,2065 toward high column density damped Lya
absorption systems. First, both Zn and Cr have low solar abundances, so
the absorption lines are not expected to be highly saturated. Second, both
Zn and Cr are predominantly singly ionized in the neutral regions that are
expected to dominate high column density damped Lya absorption systems,
thus large ionization corrections are not required. Third, Zn is observed to
be not strongly depleted in the Galactic interstellar medium, whereas Cr is
observed to be quite heavily depleted in the Galactic interstellar medium,
indicating that the ratio of Zn to Cr provides a qualitative measure of the
dust content of the absorbing gas.
Sensitive observations of Zn and Cr have so far been obtained for four
damped Lya absorption systems, and the results are summarized in Table
III. On the basis of these measurements it is concluded that the damped
Lya absorption systems represent galaxies at an early stage of chemical
evolution-because the metalicities inferred on the basis of Zn are only :::::
10% of the solar value-and that the amount of depletion onto dust grains is
small-because Cr is less abundant than Zn by a significantly smaller factor
than what is found in the Galactic interstellar medium.
These observations may be connected with the equations of cosmic chem-
ical evolution by combining Equations (11) and (12) under the assumption
that the net accretion rate density <1>( z) can be neglected to yield

Z(z) = Z(z = 3.5) - yin [!lg(~g~z;.5)1 ' (19)

which relates the metallicity Z(z) to the cosmological mass density param-
eter !lg(z) given an assumption about the "initial" metallicity Z(z ::::: 3.5)
and the yield y. In Figure 9, the metallicity evolution predicted by Equa-
tion (19) is plotted together with the measurements summarized in Table
256

TABLE III
Cr and Zn Abundances of High-Redshift Gas

log N(HI)
QSO z (cm- Z ) [Cr/H] [Zn/H] Reference
2206-199 .. . ...... 1.921 20.8 -1.1 -0.7 1
0528-250 . ... ... . . 2.140 20.7 -1.4 < -0.9 2
0100+130 .. . . . .... 2.309 21.4 -1.9 -1.4 3
0528-250 .... . .... 2.811 21.3 -1.6 - 0.9 2
REFERENCES-(I) Meyer & Roth 1990; (2) Meyer, Welty, & York 1989;
(3) Pettini, Boksenberg, & Hunstead 1990.

III based on Zn II for various values of the initial metlalicity Z(z = 3.5),
where the yield y has been adjusted to satisfy the boundary condition that
Z(z = 0) gives solar metalicities. Although the data are as yet too limited
to provide strong constraints, it is clear that by integrating measurements
of the metallicity of damped Lya absorption systems with the history of gas
consumptjon and hence star formation traced by cosmological mass density
parameter it will ultimately be possible to develop a coherent picture of
the chemical evolutionary history of the universe, from which will be deter-
mined important information about the yield and hence about the initial
stellar mass function of high-redshift galaxies.

4.2. MOLECULAR CONTENT OF HIGH-REDSHIFT GAS

Molecular hydrogen plays an essential role in star formation and in the

thermal balance of the interstellar media of nearby galaxies. In addition,
molecular hydrogen serves as a sensitive tracer of dust and as an impor-
tant diagnostic of physical conditions, including temperature, density, and
the intensity of the ultraviolet ra(liation field. Because molecular hydrogen
bears so significantly on the physical state of nearby galaxies, it is clearly
important to determine the molecular content of high-redshift galaxies.
Over the past several years there have been several attempts to detect
and measure Lyman- and Werner-band molecular hydrogen absorption lines
from metal-line absorption systems. These measurements have been directed
toward high column density damped Lya absorption systems, for which the
neutral hydrogen column densities are similar to those of the Galactic clouds
that show appreciable molecular fractions and for which .constraints on the
molecular abundances are therefore most meaningful. Detection and mea-
surement of the Lyman- and Werner-band lines from metal-line absorption
systems is difficult because the lines occur at rest-frame wavelengths short-
ward of Lya and hence always occur in the "Lya forest," where the line den-
 257

-2
• •

-
!XI
taO
0
•
-4

-6~~~~~~~~~~~~~~~~~~~~~~~~

o 1 2 3 4 5
z
Fig. 9. Logarithm of metallicity Z(z) vs. redshift z. Curves show the predictions
of Equation (19), assuming (bottom to top) "initial" metallicities Z(z = 3.5) of 0,
0.001,0.01, and 0.1 times solar values and adjusting the yield y to satisfy the bound-
ary condition that Z(z = 0) gives solar metallicities. Points show the measurements
based on Zn II summarized in Table III.

sity is high and the prohability of chance coincidence is great. The Lyman-
and Werner-band molecular hydrogen lines are accessible from ground-based
telescopes for absorption systems wth redshifts satisfying z ;(, 2.5.
These observations have so far been obtained for three damped Lya ab-
sorption systems, and the results are summarized in Table IV, which shows
for each observation the measurement or limit to the fractional molecular
abundance F, which is defined by F = 2 N(H2)jN(H I). On the basis of these
measurements, it is concluded that the molecular content of high-redshift,
damped Lya absorption systems is generally much lower than for compara-
ble Galactic clouds, although it is not clear whether this is due to an absence
of an appreciable amount of dust or to the existence of an intense ultraviolet
radiation field, perhaps related to a high rate of star formation.
Actual detection of the Lyman- and Werner-band molecular hydrogen
lines is important in order to exploit the potential of these lines for de-
termining physical properties of the absorbing gas, and for this reason the
z = 2.811 absorption system toward 0528-250 is of particular interest. For
this purpose, high spectral resolution ( ~ 25 km s-1) observations are es-
sential, both in order to separate the H2 absorption lines from the many
unrelated "Lya-forest" absorption lines and to determine accurate column
densities of the various ground-state rotational levels. A small portion of
258

TABLE IV
H2 Abundances of High-Redshift Gas

log N{HI)
QSO z ( cm - 2) log F Reference
0100+130 ... . ..... 2.309 21.2 < -5.4 1
1337+113 ......... 2.796 20.9 < -3.9 2
0528-250 ......... 2.811 21.1 -2.7 3
REFERENCES-{I) Black, Chaffee, & Foltz 1987; (2)
Lanzetta, Wolfe, & Turnshek 1989; (3) Foltz, Chaffee, & Black
1988.

-
~
~
I i.
Q)
.....
:>
...., 0.5
-co
Q)
0::

o
3980 4000 4020
Wavelength (1)

Fig. 10. Spectrum of the QSO 0528-250 obtained with the echelle spectrograph of
the Cerro Tololo Inter-American Observatory 4.0 telescope together with a fit the
Lyman- and Werner-band molecular hydrogen lines of the z = 2.811 damped Lya
absorption system. The high quality of the spectrum together with an apparent lack
of velocity structure in the H2-absorbing components allow column densities of the
ground-state rotational levels to be precisely determined. Adapted from Lanzetta
et al. (1993).
 259

such a spectrum of the QSO 0528-250 recently obtained with the echelle
spectrograph of the Cerro Tololo Inter-American Observatory 4.0m telescope
(Lanzetta et al. 1993) is shown in Figure 10. In this case the identification
of the Lyman- and Werner-band molecular hydrogen lines is unambiguous,
and the high quality of the spectrum together with an apparent lack of veloc-
ity structure in the H 2 -absorbing components allow column densities of the
various ground-state rotational levels to be precisely determined. Detailed
fitting to the line profiles yields a doppler parameter of b = 3.5 ± 0.1 km S - 1
and an excitation temperature (that appears to hold to ground-state rota-
tionallevels up to and including J" = 5) of Tex = 212 ± 20 K. Because the
redshift of this absorption system is greater than the redshift of the back-
ground QSO, this absorption system is not representative of the damped
Lya absorption systems generally. Nevertheless, these observations indicate
what sorts of information can be extracted from high spectral resolution ob-
servations if molecular hydrogen absorption lines could be detected toward
more metal-line absorption systems.

4.3 . MAGNETIC FIELDS IN HIGH-REDSHIFT GAS

By combining surveys for metal-line absorption systems with measurements
of Faraday rotation of the polarized radio continua of the background QSOs,
it is possible to derive unique information about the presence and strength
of magnetic fields in high-redshift galaxies. Early progress in this direction
was obtained by Kronberg & Perry (1982) and Welter, Perry, & Kronberg
(1984), who demonstrated a link between the presence of metal-line absorp-
tion systems and the incidence of Faraday rotation. This result has recently
been extended by Wolfe, Lanzetta, & Oren (1992), who found evidence that
lines of sight toward damped Lya absorption systems are statistically more
likely to show significant Faraday rotation than are random lines of sight,
thus suggesting that the damped Lya absorption systems are the sites of
significant magnetic fields .
If the damped Lya absorption systems are in fact the sites of significant
magnetic fields, then the line-of-sight component of the magnetic field BII
can be inferred from the observed Faraday rotation measure RM as

RM=2.7(1+z) -2 ( Ne
19 2
) ( -BII
G ) radm -2 , (20)
10 cm- 1 JL
where z and Ne are the redshift and electron column density of the damped
Lya absorption system. By writing the electron column density as Ne =
xeN(HI), where Xe is the average electron fraction , then from the known
neutral hydrogen column densities of the absorption systems it is possible
to derive the values of xeBII . Values derived for the damped Lya that have
as yet been studied are listed in Table V. Because the high column density
260

TABLE V
Magnetic Field Strengths of High-Redshift Gas

RM log N(HI) zeBIi

QSO z (rad m- 2 ) (cm- Z } (JLG)
1229-020 ... .. .... 0.395 -56 21.0 0.46
0809+483 ......... 0.437 -145 21.1 0.95
1328+307 ......... 0.692 3 21.0 0.036
0957+561A ........ 1.391 -63 20.4 5.4
1331+170 ... ... ... 1.776 -23 21.2 0.26
0458-020 .. ....... 2.038 100 21.7 0.70

damped Lya absorption systems are mainly neutral, it is unlikely that the
electron fraction exceeds Ze ~ 0.3. The values of zeBU listed in Table V thus
suggest that magnetic field strengths of ~ 1 J..LG strength are representative
of the damped Lya absorption systems.
The existence of magnetic field strengths of '" 1 J..LG at z ~ 2 is difficult
to understand in the context of galactic dynamo models of magnetic field
generation. In these models, a small initial seed field of Bseed ~ 10- 15 J..LG
(Mishustin & Ruzmaikin 1972) grows exponentially on the time scale of a
galactic rotation period T as

B(t) ~ Bseed exp(tJT). (21)

The time available for dynamo amplification is bounded by the rotation pe-
riod (in order that the systems establish centrifugal equilibrium) and the
age of the universe at the absorption red shift z. Adopting a rotation period
T ~ 0.3 X 109 yr appropriate for the outer regions of ordinary spiral galax-
ies, the characteristic magnetic field strengths predicted from the galactic
dynamo model at z ~ 2 are of order B ~ 10- 10 J..LG. Comparison with the
results listed in Table V indicates that this value falls short by at some ten
orders of magnitude from the observed value. The galactic dynamo thus
cannot reproduce the magnetic fields inferred from Faraday rotation if the
damped Lya absorption systems are centrifugally supported disks with ro-
tation periods comparable to the periods characterizing the ourter regions
of ordinary spiral galaxies. This result seems to suggest that either a more
efficient magnetic field amplification mechanism is at work or that the uni-
verse at z ~ 2 was pervaded by a primordial magnetic field. Larger samples
of damped Lya absorption systems toward radio-loud QSOs will be needed
both to confirm this result and to determine the redshift evolution of mag-
netic fields in galaxies.
 261

AcknowledgeDlents
It is a pleasure to thank Harley Thronson, Mike Shull, and the orgaruzmg
committees of the Third Teton Summer School for a stimulating meeting
and my collaborators Jack Baldwin, Bob Carswell, Dave Turnshek, Gerry
Williger, and Art Wolfe for allowing me to discuss some of our results in
advance of publication. Much of the work reported here was supported by
NASA through grant HF-1021.01-91A, awarded by the Space Telescope
Science Institute, which is operated by the Association of Universities for
Research in Astronomy, Inc., for NASA under contract NAS5-26555.

References
Bechtold, J., Green, R. F., Weymann, R. J., Schmidt, M., Estabrook, F. B., Sherman, R.
D., Wahlquist, H. D., & Heckman, T. M. 1984, ApJ, 281, 76
Bergeron, J., & Boisse, P. 1991, AA, 243, 344
Bergeron, J., & Stasinska, G. 1986, AA, 169, 1
Binggeli, B., Sandage, A., & Tarnrnann, G. A. 1988, ARAA, 26, 509
Black, J. H., Chaffee, F. H., Jr., & Foltz, C. B. 1987, ApJ, 317, 442
Bosma, A. 1981, AJ, 86, 1825
Broadhurst, T. J., Ellis, R. S., & Shanks, T. 1988, MNRAS, 235, 827
Cole S. 1991, ApJ, 367, 45
Colless, M., Ellis, R. S., Taylor, K., & Hook, R. N. MNRAS, 244, 408
Cowie, L. L., Songaila, A., & Hu, E. M. 1991, Nature, 354, 460
Fall, &. -M., &Pei, Y. C. 1989, ApJ, 337,7
Fall, S. M., Pei, Y. C., and McMahon, R. G. 1989, ApJ, 341, L5
Foltz, C. B., Chaffee, F. H., Jr., & Black, J. H. 1988, ApJ, 324, 267
Foltz, C. B., Weymann, R. J., Peterson, B. M., Sun, L., Malkan, M. A., & Chaffee, F. H.,
Jr. 1986, ApJ, 307,504
Gnedin, N. Y., & Ostriker, J. P. 1992, ApJ, 400,1
Holmberg, E. 1975, in Galaxies and the Universe, ed A. Sandage, M. Sandage, & J. Kristian
(Chicago: University of Chicago Press), 123
Junkkarinen, Y., Hewitt, A., & Burbidge, G. 1991, ApJS, 77, 203
Kennicutt, R. C., Jr. 1990, in The Interstellar Medium in Galaxies, ed. H. A. Thronson,
Jr. & J. M. Shull (Dordrecht: Kluwer Academic Publishers), 405
Kirshner, R. P., Oemler, A., Schechter, P. L., & Shectman, S. A. 1983, AJ, 88, 1285
Kronberg, P. P., & Perry, J. J. 1982, ApJ, 263,518
Lanzetta, K. M. 1988, ApJ, 332, 96
Lanzetta, K . M. 1991, ApJ, 375, 1
Lanzetta, K. M., Baldwin, J. A., Williger, G. M., & Carswell, R. F. 1993, in preparation
Lanzetta, K. M., Turnshek, D. A., & Wolfe, A. M. 1987, ApJ, 322, 739
Lanzetta, K. M., Turnshek, D. A., & Wolfe, A. M. 1993, ApJ, submitted
Lanzetta, K. M., Wolfe, A. M., & Turnshek, D. A. 1989, ApJ, 344, 277
Lanzetta, K. M., Wolfe, A. M., Turnshek, D. A., Lu, L., McMahon, R. G., & Hazard, C.
1991, ApJS, 77, 1
Meyer, D. M., & Roth, K. C. 1990, ApJ, 363,57
Meyer, D. M., Welty, D. E., & York, D. G. 1989, ApJ, 343, L37
Mishustin, I. N., & Ruzmaikin, A. A. 1972, Soviet Phys.-JETP, 34, 233
Ostriker, J. P., & Thuan, T. X. 1975, ApJ, 202,353
Peebles, P . J. E. 1988, in The Epoch of Galaxy Formation, ed. C. S. Frenk, R. S. Ellis, T.
Shanks, A . F. Heavens, & J. A. Peacock (Dordrecht: Kluwer Academic Publishers), 1
Pei, Y. C., Fall, S. M. & Bechtold, J. 1991, ApJ, 378, 6
Pettini, M., Boksenberg, A., & Hunstead, R. W. 1990, ApJ, 348, 48
262

Sargent, W. L. W ., Boksenberg, A., &: Steidel, C. C. 1988, ApJS, 68, 539

Sargent, W. L. W., Steidel, C. C., &: Boksenberg, A. 1988, ApJ, 334, 22
Sargent, W. L. W., Steidel, C. C., &: Boksenberg, A. 1989, ApJS, 69, 703
Schechter, P . 1976, ApJ, 203, 297
Sdunidt, M. 1959, ApJ, 129, 243
Steidel, C. C. 1990, ApJS, 74, 37
Steidel, C . C., &: Sargent, W. L. W. 1992, ApJS, 80, 1
Tinsley, B. M., 1980, Fundamentals of Cosmic Physics, 5, 287
Tytler, D. 1982, Nature, 298, 427
Tytler, D ., Boksenberg, A., Sargent, W. L. W., Young, P ., &: Kunth, D. 1987, ApJS, 64,
667
Welter, G. L, Perry, J. J. &: Kronberg, P . P. 1984, ApJ, 279,19
White, S. D . M. 1990, in The Interstellar Medium in Galaxies, ed. H. A. Thronson, Jr. &:
J. M. Shull (Dordrecht: Kluwer Academic Publishers), 371
White, S. D. M ., &: Frenk, C . S. 1991, ApJ, 379, 52
Wolfe, A . M., 1986, in Proc. NRAO Conf. on Gaseous Halos of Galaxies, ed. J. Bregman
&: J. Lockman (NRAO SP)
Wolfe, A. M., Lanzetta, K. M., &: Oren, A. L. 1992, ApJ, 388, 17
Wolfe, A. M., Turnshek, D. A., Lanzetta, K. M., &: Lu, L. 1993, ApJ, 404,480
Wolfe, A. M., Turnshek, D. A., Smith, H . E., &: Cohen, R. D. 1986, ApJS, 61, 249
Young, P., Sargent, W. L. W., &: Boksenberg, A. 1982, ApJS, 48, 455
 THE PROPERTIES OF ABSORPTION-LINE SELECTED
HIGH-RED SHIFT GALAXIES
CHARLES C. STEIDEL
University of California, Berkeley, Astronomy Department

Abstract. The properties of the galaxies selected by the presence of metal-line absorption
systems in the spectra of background QSOs are reviewed. We first describe the properties of
these galaxies at redshifts where they can be compared to galaxies selected by other means;
selection of galaxies by gal croll lection picks out a population of relatively bright galaxies
which has not evolved dramatically in either number or luminosity since z '" 1. These are
probably the predecessors of present-day normal field galaxies. Intrinsically faint galaxies
or low surface brightness objects appear to contribute little to the total gas cross section
(as defined by gas which is just optically thick in the H I Lyman continuum), as essentially
no galaxies fainter than '" 1.5 magnitudes below L· are identified as QSO absorbers. The
implications for the evolution of normal galaxies are discussed. Absorption-line selection
is possibly the most powerful criterion for isolating examples of very high-redshift (z ~ 1)
galaxies. We review the status of searches for galaxies associated with QSO absorption
systems with z ~ 2.

1. Introduction

Our understanding of the nature of distant galaxies is one of very rapid

change due to continually improving data to fainter apparent magnitudes.
One now has galaxy counts and broad-band colors to m 27 (e.g., Tyson f'V

1988; Lilly et al. 1991; Metcalfe et al. 1991; Steidel & Hamilton 1993) which
imply very substantial evolution in the galaxy population in the past several
billion years. Recent spectroscopic surveys for field galaxies to B 24 (Lilly f'V

et al. 1991; Ellis, this conference) imply that the "excess" counts found in
the imaging surveys are dominated by faint galaxies (ME ~ -19 for Ho = 50
1m s-1 Mpc- 1 and qo = 0) at modest redshift (z '" 0.3). It is not at all clear
how these relatively faint galaxies (are they really "dwarfs"?) fit into the pic-
ture of the evolution of the galaxies observed at the present epoch. However,
it is evident from the redshift surveys that there are very few high-redshift
galaxies contributing to the counts (at least, to B '" 24), implying that there
has not been substantial luminosity evolution at the bright end of the galaxy
luminosity function. In fact, the surveys have not ruled out the existence of
a "passively" evolving population of bright galaxies (Le., with a luminosity
function comparable to that observed today) at high redshift. Instead, they
show that the excess numbers of faint galaxies at a given apparent magni-
tude cannot be ascribed to the appearance of massive galaxies which were
much brighter in the past. Thus, it is not clear whether the relatively faint
galaxies dominating the counts are the "building blocks" of the galaxy pop-
ulation observed today, or whether they have "disappeared" because they
have returned to a much fainter, quiescent phase in their star formation
history. One anticipates more progress in understanding the nature of the

263
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, 263-294.
© 1993 Kluwer Academic Publishers. Printed in the Netherlands.
264

faint field galaxies with more efficient, multi-object spectrographs and the
advent of 8m-class telescopes. However, realistically, it will be difficult to
go more than 1-1.5 magnitudes deeper, and it is not absolutely clear that
more will be revealed in doing so.
In this article, we summarize the results of distant galaxy studies which
are selected from the presence of metal-line absorption systems in the spectra
of background QSOs. That is, rather than selecting objects for study based
upon their apparent magnitude in a particular observed bandpass (e.g., B,
I, or K), the galaxies are selected by the cross section of gas which gives
rise to detectable absorption lines in the QSO spectra. There is no redshift-
dependent selection function involved in absorption line selection, and as
we will see in §3, it is possible to construct a statistically complete sample
defined solely on the basis of an appreciable optical depth of H I. Thus,
absorption-line selection provides an approach that is both independent of
and complementary to the magnitude-limited surveys, with the added bonus
that the evolutionary properties of the absorption systems (not to say the
understanding of those properties) are now reasonably well-established from
z rv 0.2 all the way to z rv 4, more than 90% of the age of the Universe. In
addition, as discussed in detail by Ken Lanzetta (this meeting), one of the
advantages of absorption-line-selection is that one, in principle, has access to
very detailed physical information on these same very distant galaxies from
the absorption-line studies (e.g., kinematics, chemical abundance, molecular
content). Such information will probably never be accessible for the bulk of
normal field galaxies simply because they are so faint in the stellar contin-
uum. Moreover, the stellar continuum and emission lines never will provide
as much information about the galaxy ISM as the absorption lines recorded
in the QSO spectra.
We divide the topic into two broad areas: first, the study of absorption-
line selected galaxies at moderate redshift (z ~ 1), where the galaxies can
be studied by other methods (Le., direct imaging and spectroscopy) using
current telescopes and instrumentation, and where the properties of the
absorbing galaxies might be compared with those selected by other methods.
Establishing the absorber/galaxy connection at moderate redshift is crucial
to understanding what the absorption systems at very high redshift are
telling us about the evolution of galaxies at early epochs, and to defining
how selection by gas cross section relates to other physical properties of the
galaxies (e.g., luminosity, color, structure, etc.) . Secondly, we give a status
report on the use of QSO absorption-line systems to "flag" sites of very
high-redshift galaxies (z ~ 2). It is already obvious that it is not particularly
efficient to find high-redshift galaxies simply by obtaining spectra of samples
of objects with faint apparent magnitudes. To locate the probable sites of
high-redshift objects for further study and to avoid the "confusion" of the
large numbers offoreground "dwarfs", absorption-line selection seems a very
 265

promising method. Moreover, one may be able to comment on the state of

fairly "typical" objects at high redshift (rather than the very brightest and
most unusual ones, as probed by the highly successful surveys ofhigh-z radlo
galaxies).

2. Background

Well before the absorber/galaxy connection was made directly, the metal-
line absorption systems were generally associated with galaxies because of
a number of less-direct lines of evidence established from systematic stud-
ies of the absorption systems themselves. Most obviously, the presence of
elements heavier than He indicated that some star formation had occurred,
even at very early epochs. Does gas enriched by star formation imply the ex-
istence of a galaxy by definition? More quantitatively, the metal-line systems
were first shown to have a uniform distribution in co-moving coordinates (as
expected for intervening objects) by Young et al. (1982) and confirmed sub-
sequently by many authors. When the metal-line systems are observed at
high spectral resolution, they almost always break up into sub-components
with a typical velocity spread of 150 - 200 km s-1; this was taken by some
to imply their association with galaxy-sized objects. The metal systems are
known to cluster on similar scales to galaxies (~v ::; 600 km S-1; Sargent et
al. 1988), and perhaps even on very large scales (e.g., Sargent et al. 1988;
Heisler et al. 1989; Steidel & Sargent 1992).
Nevertheless, until the galaxies actually producing the absorption could
be identified directly, it was not completely clear how the absorbers were
related to the types of objects which we would call galaxies at the present
epoch. For example, it became clear early on that, if normal galaxies were to
,
account for enough cross section to produce the observed number of metal-
line absorption systems, and if one were to more or less preserve the co-
moving number density of galaxies, then each galaxy would have to have a
gas cross section many times larger than the typical optical sizes of galaxies
today. From the statistics of the incidence of the metal-line absorbers alone,
if one insists on producing the cross section using normal galaxies, a picture
of a "typical" L· galaxy similar to the cartoon in Figure 1 is implied. The
very large gaseous sizes implied by the statistics are responsible for the
somewhat presumptuous term "halo gas" which often is used to refer to the
sites giving rise to the bulk of the metal-line absorption systems.
While we will return to a more extensive discussion of the gaseous sizes of
high-redshift galaxies in the next section, there are a few important points
to be made from the outset. Referring to Figure 1, each "contour" indicat-
ing the radial extent for the detection of a particular absorption-line species
also corresponds to a characteristic H I column density. By definition, the
damped Lya systems refer to systems with N(HI):2': 3 X 10 20 cm- 2 , compa-
266

I I I

CIV

50 - -

0- -
I
.c

-50 r- -

I I I
-50 o 50
h- 1 kpc
Fig. 1. Schematic diagram illustrating an over-simplified view of the structure of an
"L·" galaxy as deduced solely from the statistics of the various classes of metal-line
absorpti.on systems. Note that the Mg II-selected and Lyman-limit selected systems
have the same cross section, and that the damped Lya systems have a cross section
which is only a few percent of the total.

rable to the typical column densities in the disks of spiral galaxies (see Wolfe
et al. 1986; Lanzetta et al. 1991). The Lyman Limit systems, or LLSs, refer
to systems with N(H 1) '~ 3x 10 17 cm- 2 , or systems which are just optically
thick in the H I Lyman continuum. An optical depth of unity or greater in
the Lyman continuum is also necessary to have appreciable column densities
of Mg II (see, e.g., Bergeron & Stasinska 1986); in fact, as we will discuss in
§3.2, LLSs and Mg II-selected systems are essentially the same population,
and it is only because low-redshift Mg II doublets can be detected using
ground-based spectroscopy (whereas LLSs require observations in the vac-
uum UV for redshifts z < 2.5) that many more Mg II-selected systems are
known for z ~ 1.
The C IV-selected absorption systems, which can have a very large range
in N(H I), have by far the largest cross section, and have been found with
N(H I) as small as 1015 . 5 cm- 2 • It is very important to realize that more
 267

than 90% of th\) metal-line absorption systems have H I column densities

smaller than can be detected in the 21-cm emission line using current tech-
nology. The absorption systems are orders of magnitude more sensitive to
the detection of small columns of neutral gas than any other technique. As
a result, we are ill the rather ironic position of knowing much more about
the presence of relatively diffuse, highly ionized gas associated with galaxies
at high redshift than for nearby galaxies. It is not until the statistics of the
metal-line systems from HST observations are completely digested that we
will be able to confirm or rule out the persistence of this highly ionized com-
ponent to the present epoch. What follows from thi,s is that when we talk
about the sizes of galaxies, as measured from the 21-cm emission measure-
ments, we are talking about conservative lower limits on the extent of the
gas. It also should be noted that we will be discussing the state of the gas
in galaxies at early epochs, when the gaseous structure of galaxies was al-
most certainly different than at the present epoch. Although it is beyond the
scope of this article, we do in fact observe evidence for substantial changes
in the gas associated with galaxies from z '" 2 to z '" 0.3 from analyses of
the QSO absorption lines.

3. Galaxies at z ::; 1

3.1. PREVIOUS WORK

Until very recently, despite a large amount of effort establishing the many
circumstantial lines of evidence associating the metal-line absorption sys-
tems with galaxies briefly mentioned above, what was clearly lacking was
direct association with galaxies which could be imaged using standard tech-
niques and for which spectra could be obtained in order to confirm ' that
the gas in which the absorption arises is indeed associated with that galaxy.
The most practical types of QSO absorption systems for selecting moderate
redshift galaxies are those containing the easily recognizable Mg II AA2796,
2803 resonance doublet; with ground-based observations, such doublets can
be observed over the redshift range 0.2 ::; z ::; 2.2. The ground-breaking
work on the absorption-line/galaxy connection was recently completed by
Bergeron & Boisse (1991) using a sample of Mg II doublets taken from the
literature. Bergeron & Boisse were able to identify the absorbing galaxy in 11
of the 13 cases examined for (Zab.) = 0.4. Although the sample size was rel-
atively small, the high success rate enabled them to present evidence which
supported the long-hypothesized very large cross sections (Rgal rv 30h- l
kpc, where h == H o/I00 km S-1 Mpc- l ) for a population of what appeared
to be bright galaxies ( ~ L *) whose spectra showed evidence for considerable
ongoing star formation.
The distribution of "impact parameters" of the QSO sight line with the
268

putative absorbing galaxies was found to be much better fitted with the as-
sumption of a spherical distribution of absorbing gas than more "linear" ge-
ometrical configurations which one might expect, e.g., if tidal features were
contributing most of the absorbing cross section. The rather close agree-
ment of the implied gaseous sizes of the absorbing galaxies, as compared
to a calculation of the expected sizes obtained by assuming a galaxy lumi-
nosity function similar to that observed at the present epoch, and spherical
gaseous halos that scale in size with galaxy luminosity like the Holmberg
(1975) relation, led to the conclusion that essentially all bright galaxies at
z t"V0.4 must possess halos that are several times their Holmberg radii. Thus,
they not only established that at least Mg II-selected absorption systems are
associated with normal galaxies at moderate redshift, but they also appar-
ently confirmed the long-held belief that extended gaseous "halos" must
have existed at earlier epochs.
However, the Bergeron & Boisse results have not been uniformly accepted
as definitive. Most notably, York et al. (1986) proposed that the absorption
systems, rather than being produced in extended gas associated with a bright
galaxy, instead arise in more numerous compact star-forming dwarf galaxies.
Yanny, York, and collaborators (e.g., Yanny, York, & Williams 1990) used
narrow-band imaging at absorber redshifts in the fields of QSOs for which
moderate-redshift Mg II absorption is known in order to search for [011]
A3727 emission from objects associated with the absorbing material. They
apparently find a considerable over-density of emission-line objects in these
fields, some of which have been confirmed to have the same redshift as the
absorber through spectroscopy. Generally, the confirmed objects have been
at quite large projected distances from the QSO lines of sight, and so are
unlikely to be the actual absorbers. They suggest that the sites of absorp-
tion are associated with "groups" of objects at moderate redshift. However,
Yanny & York (1992) have suggested that there is always a compact star-
forming object superposed directly on the bright QSO image, which would
not have been seen by Bergeron & Boisse, who searched only beyond the
seeing disk of the QSOs for the absorbing objects. Yanny & York thus at-
tribute the necessarily large total cross sections for absorption to numerous
dwarf galaxies, which they suggest may eventually coalesce to form the L*
galaxies of the current epoch.
Clearly, these two viewpoints are diametrically opposed, although each
of the proponents would claim to have observational evidence supporting
their scenario. In addition, Phillipps, Disney, & Davies (1993) have recently
presented a case for absorption by low surface brightness galaxies, which
mayor may not be large objects. Phillipps et al. (1993) are uncomfortable
with the very large gaseous sizes implied by the Bergeron & Boisse work, and
instead explain the presence of a "bright" galaxy at a few Holmberg radii
from the line of sight as being a manifestation of galaxy clustering. Thus,
 269

both the Yanny et al. and Phillipps et al. pictures support "hidden", faint
galaxies rather than bright galaxies with large gaseous halos (see Davies,
this conference).
On the other hand, Lanzetta & Bowen (1990), on the basis of the Berg-
eron & Boisse sample, found that there is a correlation between the observed
impact parameter and the strength of the Mg II absorption feature which
would be difficult to explain under the "clustering" hypothesis. This seems
to be a strong indication that perhaps the bright galaxies are themselves
responsible for the absorption. Of course, at some point, the two apparently
opposite viewpoints merge into semantics, for if the "absorber" is always
within a particular characteristic distance of a bright galaxy, can it be said
that the "dwarf" absorber is associated with (or, perhaps, "in the halo of")
the' bright galaxy?

Some of the still-unanswered questions are the following:

1. What is the contribution of intrinsically faint galaxies to the total ab-

sorption cross section?
2. What is the luminosity function of absorption-line selected galaxies?
With a statistically homogeneous absorption-line sample, this should be
more straightforward to evaluate than the luminosity function obtained
from magnitude-limited surveys because there is no distance-dependent
selection function.
3. What are the optical and optical/infrared colors of galaxies selected in
this manner? What are the galaxies' spectroscopic types, and are they
similar to normal galaxies at zero redshift?
4. Is there a population of galaxies at moderate redshift which does not
contribute to the total absorption cross section?
5. Do the galaxy properties depend upon the observed absorption-line
properties? What parts of galaxies give rise to what types of charac-
teristic absorption signatures?
6. How do the galaxies selected by gas cross section compare to the galaxy
samples selected in the B/I/K magnitude limited samples? Are they the
same population of objects?
7. What are the implications of all of the above for the evolution of normal
galaxies with redshift?
In an effort further to address some or all of the above questions, we
(Steidel & Dickinson 1993) are in the process of completing a comprehensive
survey for absorption-selected galaxies in the redshift range 0.3 ~ z ~ 0.9.
We now describe the observational program and some of the preliminary
results. Some of the results of this program have been reported previously
in Steidel & Dickinson (1992) and Steidel (1992).
270

3.2. THE NEW SURVEY

If one wishes to construct a well-defined sample of galaxies selected by

QSO absorption, it is important to ensure that the absorption-line sample is
statistically unbiased. That is, if the galaxy properties depend in any way on
the characteristics of the absorption systems, then one introduces bias into
the galaxy sample if one does not select the absorption systems such that
the sample has the sa~e distribution of properties as would be encountered
in an ensemble of random lines of sight. In practice, until recently, this
was difficult because any individual sample of Mg II-selected absorption
systems at moderate redshift was very small, and many of the systems in
the literature were obtained in the course of other spectroscopic projects and
so were not really subject to any well-defined selection criterion. In general,
a statistically unbiased absorption-line survey is designed to detect lines to
a particular threshold in rest equivalent width, so that while the resulting
sample would be statistically unbiased, it would not be "complete" because
absorption lines which fall below the equivalent width threshold are not
included. Fortuitously, however, in the case of the Mg II absorption-line
sample which we have used to construct our galaxy survey, we believe that
the sample is both statistically unbiased and complete.
The sample is taken mainly from two large absorption-line surveys: those
of Sargent, Steidel, & Boksenberg (1988); and of Steidel & Sargent (1992).
Both surveys were designed to detect Mg II doublets to 0.3 A rest equivalent
width. Many of the spectra had sufficiently high SIN that weaker lines could
have been found; however, it appears that Mg II absorption lines weaker than
0 .3 A rest equivalent width are exceedingly rare (probably due at least in
part to the H I column density "threshold" near an optical depth of rv 1,
where the Mg II column density decreases by 2 orders of magnitude as the
H I column density goes from 10 17. 5 cm- 2 to 10 17 . 0 cm- 2 ; see §2 and below).
There are two other important points to make about the Mg II selected
absorption-line systems. First, as illustrated in Figure 2, the incidence of
Mg II per unit redshift range, dN I dz, is consistent with no evolution in
co-moving cross section from z = 0.2 to z = 2.2. Second, and very im-
portantly, it appears that selecting by the presence of a detectable Mg II
doublet is equivalent to selecting by the criterion that the H I optical depth
exceeds unity, and vice versa- when the Lyman continuum region is ob-
served corresponding to a known Mg II system, there is always a LLS, and
all LLSs known for z :s2 have corresponding Mg II doublets (see also the
article by Lanzetta in these proceedings). That is, the sample defined by
the criterion Wo(Mg 11)2 0.3 A is equivalent to a sample selected such that
N(H I) 2 3 X 10 17 cm- 2 - the Mg II-selected sample is a complete sample of
systems ezceeding a well-defined H I column density. There is therefore no
bias in the sample against gas with lower metallicity; given optically-thick
 27\

Mg II Evolution, W0 > 0.30 A

OL-~~~~'~~~~--7-I~~~~-T'~~~~~~2~~~
0.5 1 1.5
Absorption Redshift
Fig. 2. The red shift dependence of the redshift path density of the Mg II-selected
absorption systems. The plotted data are for the sample which is complete to a rest
equivalent width threshold of 0.3A. The region over which the absorbing galaxies
have been identified directly is marked. The expected behavior of the dN/ dz relation
for no evolution in the comoving cross section of the absorbers is shown with the
dash-dot (qO = 0) and long-dashed (qO = 0.5) curves, respectively. The maximum
likelihood fit to the data assuming the fiJrm dN/ dz oc: (1 +z)1' is shown with the
short-dashed curve, and corresponds to I = 0.8

H I, one essentially always detects Mg II. For this reason, we now refer to
the galaxy sample as being one which is selected by the cross section of H I
gas with THI :G 1 (or, equivalently, LLSs; see Figure 1) .
. The full survey consists of 52 absorber fields, including about '" 56 ab-
sorption systems, as well as a number of "control" fields, which are also
fields of QSOs for which very good spectra were obtained as part of the
same absorption line surveys defining the absorber sample, but which have
no detected Mg II absorption systems in the redshift range 0.2 ::; Zab6 ::; 0.9.
Deep images were obtained of each field to a depth designed to go at least
2 magnitudes fainter than L* at the appropriate redshift. For each field,
a combination of G[4870j800), n[6930j1500), and i[8100j1500) filters were
used such that we could always construct a rest-frame equivalent B mag-
nitude and a measurement of the characteristic 4000AjBalmer continuum
break at the redshift of the absorber. We have also begun to obtain K mag-
nitudes for all identified absorbing salaxies, since the K light is much less
sensitive to the current star formation rate (and more sensitive to the stars
272

dominating the stellar mass of the galaxy) than the rest frame U and B
magnitudes that are being measured in the optical passbands. Follow-up
spectroscopy, the most time-consuming part of the program (since many of
the candidate absorbing galaxies would have observed B magnitudes fainter
than 24), is still ongoing, but at the time of this writing we have succe'ssfully
confirmed more than half of the candidates, and have obtained redshifts for
many other galaxies in the same fields.
We have been very careful to perform accurate point-spread function
(PSF) subtraction of the relatively bright QSO image so that even galaxies
at very small angular separations from the QSO line of sight are revealed.
About 30% of the time, the candidate absorbing galaxy would not have
been discernible without the PSF subtraction, even with the typical seeing of
'" I". For the spectroscopy, in general, we have attempted to obtain redshifts
for all non-stellar objects within'" 15" of the QSO on the plane of the sky.
We are interested not only in confirming the redshift of the absorber, but
also in finding galaxies in the redshift range of interest which do not produce
detectable absorption in the QSO spectrum.
The sample to be discussed here (80% of the total sample) has a mean
redshift (z) = 0.63. Examples of galaxy identifications in 4 absorber fields
are shown in Figure 3; note that it is usually unambiguous to identify the
absorbing object, and that the PSF subtraction is rather successful in pro-
viding a view of the sky "underneath" the QSO image. Of course, it would
still be very difficult to detect a very faint galaxy which was exactly spatially
coincident with the QSO image on the plane of the sky. Fortunately, there
is the rare occurrence of a very high-redshift QSO spectrum which has both
an optically thick Lyman limit absorption system which completely absorbs
the QSO light shortward of 912A in the rest frame of the absorber, and
a moderate-redshift Mg II doublet, so that a broad ultraviolet filter can be
used unambiguously to image the region on the sky which is "contaminated"
by the QSO profile at longer wavelengths. An example of such a case is illus-
trated in Figure 4; note that the galaxy that would have been identified as
the absorber under ordinary circumstances is almost certainly the absorbing
object, as no additional objects are found "on top" of the QSO to Un '" 27.

3.2.1. Preliminary Results

From a combination of the deep images and the follow-up spectroscopy, we
have been able in every case ezamined to identify the absorbing galaxy, or
a candidate whose implied luminosity and broad-band colors place it well
within the range spanned by the confirmed galaxies. Moreover, from both the
absorber and the control fields, we have found that "interloper" galaxies, or
those galaxies which have angular separations from the QSO similar to those
 273

G~ 1148 + 384? z=O.551

jQ •/Gj'*': ··· ·.7~;

. . . . ~/~~ "- ' +~ . . . . .. .
~
+. . . .• . . . . . . . .
: G8 ;. C . ." \ : V. I

. .. ...... ~ .. . .. .... ...... ... : .... ........G4 ..:. .. ... . ..... . ~

; .4t G4,C8,G 13 3C 3!36
:' ~ z=OA72,O . 660 , O.~92 C:l 1623+ 2356 z = O.889

Fig. 3. Examples of identified absorbing galaxies in 4 representative fields. In each

case, the QSO profile has been subtracted from the image by modeling the PSF
using bright stars in the field. The two top panel are from 'R band frames, while
the two bottom panels are from i band frames. The field size shown is 30" by 30"
and the grid spacing is 10".

of identified absorbers, but whose redshifts do not correspond to any known

absorption system, are very rare. The galaxies identified as the absorbers
have apparent R magnitudes in the range 19.8 ::; R ::; 23.8, and would
have observed B magnitudes (estimated) in the range 21 ::; B ::; 25, i.e.
comparable to the deepest field-galaxy surveys to date. We have plotted
a two-color diagram in Figure 5 which shows (in our photometric system)
the expected colors of unevolved spectral energy distributions of various
"spectroscopic" types as a function of redshift over the range of our survey.
Generally speaking, the optical colors of the absorbing galaxies are consis-
tent with mid- to late-spiral galaxy spectroscopic types at the same redshift;
there are few very blue or very red galaxies in the sample. The preliminary
indication from the K-band measurements is that the optical/infrared colors
are consistent with the same range of spectroscopic type as implied by the
optical colors alone. This suggests that the observed-frame optical light is
274

:.
Q2239-3~3 : ~.
Q2239 - 383 un---:-'
:0 -
.~ .. . . .... ...•..........
~ ~ ... .. .... .... , .... .... .. ... ..... ~

~I$I .
. .. . . •
.i :.· ·t~ >~·r~~
~
----_ ... ..... --

~~
I.·
.e ;
.. fj'.. ···1 · .'

Fig. 4. Here the QSO (zern = 3.55) has both an optically-thick Lyman limit system
(Z[,L = 3.44) and a Mg II system at z = 1.033j the former results in no QSO flux
in the observed Un band (right panel) . Note that the galaxy indicated with the
arrow in the lefthand panel (1<. = 23.8), which would naturally be identified as the
z = 1.033 absorber, is still the closest object to the QSO on the plane of the sky,
even though the Un frame goes to Un '" 27 magnitude. The indicated galaxy would
have MB '" -20.8 at z = 1.033, placing it in the middle of the range of absorbing
galaxy luminosities for the sample at z '" 0.6. That is, there are no other candidate
absorbers "on top" of the QSO line of sight to about 3.5 magnitudes fainter than
L* .

not drastically biased by the current star formation rate. In the near future
we hope to construct a complete K-band luminosity function to compare
with the rest-frame B luminosity function to be discussed below.
In the context of the discussion in §3.1, it is of considerable interest to ex-
amine the distribution of observed "impact parameters" (hereafter denoted
as D) with the QSO line of sight. A histogram of observed D values is shown
in Figure 6; the median value is Dmed = 23h- 1 kpc, in very good agree-
ment with the much smaller sample of Bergeron & Boisse (1991). However,
whereas Bergeron & Boisse could not rule out that the gas cross section
was independent of galaxy luminosity, a non-parametric statistical test of
our sample shows that there is a correlation at the 99.9% confidence level
between the rest-frame B magnitude MB and D, in the sense that brighter
galaxies tend to be found with larger impact parameters (see Figure 6). Note
that the sense of this correlation is opposite to a possible selection effect that
would make it more difficult to observe fainter galaxies closer to the QSO
image. There is also a marginally significant correlation (98% confidence)
between the rest frame equivalent width of the Mg II absorption and D,
 275

Coleman ; Wu and We~dman SEDs

Points a1 llz =O.l from O.3 ~z~ O . 9
1 .................. ... -, ........................:

Sab
0 .8 ............. ....... ....... ............. ~ ... ;

Sbe
•
I 0.6 ········Scd.··~·······
~ :\
.......
0.4 .... rm ...... .
\."\
0.2 ..................,

~
.......... ;
°0~~~~~~--~~~~~2~L-~~~~3 ~--~

(G-S{)AB

Fig. 5. The expected colors in our AB-normalized Gni photometric system as a

function of redshift and of unevolved morphological type. The galaxy SEDs are
from Coleman, Wu, & Weedman 1980. The redshifts proceed from lowest (z = 0.3)
to highest (z = 0.9) from bottom left to upper right in steps of 0.1 in z. The
heavy parallelogram shows the range of colors occupied by the bulk of the identified
absorbing galaxies. The colors tend to be consistent with mid- to late-type spiral
galaxy SEDs.

in the sense that larger equivalent widths occur for smaller values of D, as
found by Lanzetta & Bowen (1990) from the Bergeron & Boisse sample. The
correlation appears to be produced by the fact that the very strongest Mg II
lines tend to be found at relatively small impact parameters, while beyond
D "-' (15 - 20)h- 1 kpc, there is no obvious trend. A natural interpretation
of this fact is that lines of sight at small impact parameters are traversing
both optically thick disk material and more diffuse "halo" gas (cf. Figure 1).
There is evidence that the strongest Mg II lines may arise in gas which is
involved in regular motion, as it would be if, e.g., it were co-rotating with
the circular velocity of a disk, giving a very large kinematic velocity spread
(Lanzetta & Bowen 1992). In any case, the two trends make it almost certain
that the galaxies which are identified as the absorbers are really responsible
for the absorption; there is no reason to expect these correlations if unseen
276

-24

- 23
-. -•
,- • -.••
10
... - 22

- •. :_.,....- -
V
.D
.'
.,
E :::e

. -- -•....
::I
Z - 21
5

-20
•
0 0 10 20 30 40 50
Impact Parameter (h- I kpc)

Fig. 6. The distribution of observed impact parameters with respect to the QSO
lines of sight for the identified Mg II absorbing galaxies (left) and absolute magni-
tude ofthe absorbing galaxy versus the observed impact parameter (right). The dot-
ted curve in the lefthand panel shows the distribution expected for spherical gaseous
halos which scale in size according the the relation R(L) = 35h- 1 kpc(L/L·)O.2 (see
text). There is a correlation at the 99.9% confidence level between MB and impact
parameter (D), in the sense that brighter galaxies tend to have larger value of D.

dwarf galaxies were the true absorbers, with the relatively bright galaxies
merely being interlopers.
This brings us to the luminosity function ofthe absorption-selected galax-
ies. As outlined above, the survey was specifically designed to allow measure-
ments of rest-frame B luminosities, so that the galaxies could be compared
directly with normal galaxies at zero redshift. The luminosity function his-
togram is shown in Figure 7; the mean and median luminosity of the sam-
ple is ME = -21.0 for the adopted cosmology of qo = 0 and Ho = 50
km S-l Mpc- 1 , or present-day L* . The faintest galaxy in the sample has
ME -19.4. The absence of faint galaxies from the absorption selected
"-J

samples was noted by Bergeron & Boisse (although we find many more
galaxies fainter than L* in the present sample), and appears to be borne out
by our larger sample.
This cannot really be a selection effect, since in most of the fields we
could easily have detected galaxies fainter than ME -19.5. We cannot
"-J

rule out the presence of galaxies 3 magnitudes fainter than L * at the higher
redshifts, but the important point is that it does not appear to be necessary
to go that faint in the luminosity function in order to find a galaxy at the
appropriate redshift, with a very regular distribution of impact parameter
 277

20

Fig. 7. The observed luminosity function of the absorption-selected galaxy sample

(histogram), together with the luminosity distribution and normalization expected
if the local luminosity function were moved to z ,...., 0.6 and galaxy cross sections
scale as observed in the absorption-selected sample. Shaded regions are plotted for
=
each of 3 assumed amounts ofluminosity evolution between z 0 and z = 0.6; each
shaded region is bounded by the local luminosity functions of Loveday et al. (1992)
(the lower normalization) and that of de Lapparent et al. (1989) (the higher normal-
ization). The dashed vertical line corresponds to present-day L· for the adopted
cosmology. The 3 shaded regions correspond to no evolution, 0.5 magnitudes of
luminosity evolution, and 1.0 magnitudes of luminosity evolution, going from left
to right on the diagram. The no-evolution case fails to match the observations in
both normalization and in the di~tribution ofluminosity; the 0.5 magnitude case is
marginally consistent, and 1.0 magnitude case is a good fit in both normalization
and in luminosity distribution (see §3.2.2). Note that the normalization for each
model is uncertain by about 25% because of uncertainties in dNjdz for the Mg II
absorption systems.
278

(see Figure 6, and below), in every case.

We know the luminosity distribution of the absorption-selected objects,
we know from above that brighter galaxies tend to have apparently larger
gaseous envelopes, and we know the distribution of impact parameters. Thus,
it is feasible to reconstruct some information about the gaseous geometry
of the galaxies. In particular, we would like to know what the dependence
of total gas cross section on luminosity really is, since that relationship will
prove crucial in obtaining the galaxy space density from the absorption line
statistics. As mentioned earlier, the usual assumption (for lack of a better
one) is that the gaseous radius scales like the Holmberg radius, so that the
total cross section for spherically-symmetric gas distributions would scale as
eTGAL(L) ex LO.8 • This means that, ifthe true galaxy luminosity function had
a Schechter function faint-end slope of a = -1, then the realized luminosity
function for a sample selected by gas cross section would have a faint-end
slope of close to zero. IT plotted on axes similar to those in Figure 7, the
function would have a rather strong peak at '" L * and would approximate
a Gaussian with a '" 1 magnitude. Thus, given the sense of the correlation
of impact parameter with MB, one expects the realized luminosity function
to be biased against the existence of lower-luminosity absorbers.
For the moment, we assume that the gaseous radius of a galaxy R depends
upon its luminosity L as

(1)

where L * is the value at the present epoch and R* is the corresponding

radius of the gas distribution (cf. Bergeron & Boisse 1991). Given the ob-
served luminosity distribution, it is straightforward to obtain the value of (3
which has the maximum likelihood of resulting in the observed distribution
of impact parameters. We find that (3 ~ 0.2, R* ~ 35h- 1 kpc, and that the
limiting assumptions of (3 = 0 (no dependence of gas radius on luminosity)
and (3 = 0.4 (the Holmberg relation) can be ruled out at the,...., 4eT and
'" 3eT levels, respectively. The expected distribution of D for the (3 = 0.2
case, spherical geometry, and a covering fraction of unity inside radius R is
shown together with the observed distribution in Figure 6. These assump-
tions provide a very good fit to both the observed distribution of D and to
the observed impact parameters on a case-by-case basis.
Can an extended disk-like geometry by excluded? The answer is "no", on
the basis of the distribution of impact parameters alone- randomly oriented
disks with the same (or slightly larger) value of R* result in a D distribu-
tion which, for the size of our sample, is statistically indistinguishable from
the spherical case. However, with a flattened geometry one should find, for
a given galaxy luminosity, that in only ,....,50% of the cases in which the
projected distance from the QSO is within R(L) should there be observed
 279

Mg II absorption. In other words, one should find as many "interlopers" as

absorbers- such occurrences must be very rare on the basis of the fields in
which we have done spectroscopy and from our "control" fields. For similar
reasons, it appears that the assumption of a covering fraction of unity is in
general a good one; in fact, since the typical Mg II absorption system consists
of several discrete components in the line profile, this may suggest that the
typical line of sight through a galaxy "halo" intersects several clouds. Thus,
we conclude that the assumption of spherical geometry and unity covering
fraction are consistent with the data.
What galaxies are not part of the absorber population? From above, we
have seen that galaxies fainter than MB rv -19.5 do not appear to be repre-
sented in the absorption-selected sample. We now have several examples of
intrinsically faint galaxies in our QSO fields which are within the characteris-
tic impact parameters discussed above, but do not give rise to absorption in
the QSO spectrum to quite sensitive equivalent-width limits. These galaxies
tend to be very blue and to have much stronger [011] A3727 line equivalent
widths than the typical absorbing galaxy in the sample. Thus far, we have
not found any bright galaxies within rv 35h- 1 kpc of a QSO sightline which
do not also give rise to detectable absorption. On the other hand, Bechtold
& Ellingson (1992) have recently approached this subject from a somewhat
different angle. Instead of searching for absorbing galaxies in absorber fields,
they have taken QSO fields in which galaxy redshifts are already known and
have obtained spectra of the QSOs in search of Mg II absorption features
corresponding to the "previously known" galaxies. They conclude that, in
fact, only about 25% of the galaxies produce detectable absorption, and
therefore that the assumption of unity filling factor cannot be correct.
However, it is possible that the two results can be reconciled. In the
Bechtold & Ellingson sample, there are 8 galaxies within 40h- 1 kpc of the
QSO lines of sight (recall that we find no impact parameters greater than
this value) . Of these eight, six show no Mg II absorption; however, 5 of
these are in the same cluster as the QSO, and the remaining one has a
rather large impact parameter (39h- 1 kpc) and has MB = -19.2, fainter
than the faintest absorber in our sample. The two galaxies that do produce
absorption are very much in line with what we have found . The implications
of the result may be that galaxies in clusters with QSOs do not possess the
extended gaseous envelopes that appear to be characteristic of field galaxies.
This might reasonably be explained by either stripping of the outer gas in
the galaxies because of tidal interactions within the cluster, or because the
outer gas is very highly ionized because of the proximity of the luminous
QSO. Along similar lines, it is uncommon to find the absorbing galaxies as
cluster members, although potential cluster member absorbers have been
identified by Bechtold & Ellingson (1992) and Steidel & Dickinson (1992).
280

3.2.2. Implications
From the observations of the absorption systems and the galaxies which pro-
duce them, we know the following:

1. That the total gas cross section does not evolve significantly as a function
of z (see Figure 2).
2. That the absorption-line selected galaxies have a fairly well known lu-
minosity function.
3. That the cross section (per galaxy) depends upon luminosity, with an
approximately known dependence.
4. That the cross section per galaxy is not correlated with redshift (i.e.,
there is no dependence of the impact parameter on redshift over the
range spanned by the survey).

Taken together, points 1 and 4 imply that the population of objects

selected by the Mg II absorption systems has a constant co-moving number
density over the observed redshift range. What do the absorbing galaxy
results imply about the nature of normal galaxies at moderate (z '" 0.6)
redshifts?
It has become customary to compare the observed redshift density dN / dz
(see Figure 2) obtained from an absorption-line survey with the ezpected
number of systems that would be encountered by assuming a luminosity
distribution and a scaling of gas cross section with luminosity. In other
words, one can calculate the gas cross section per galaxy which would be
required to produce the observed number of absorption systems, and then
compare those numbers to the actual observed gas cross sections at moderate
redshift (see, e.g., Bergeron & Boisse 1991). A comparison of these numbers
then gives one some idea of how well the assumed galaxy luminosity function
actually compares to the one which is inferred from the absorption- selected
galaxies. Of immediate interest is to estimate how well the population of
relatively luminous galaxies at (z) = 0.6 corresponds to the same galaxies
locally; from this, some sense of the evolution of relatively bright galaxies
over that range of time can be obtained.
As usual, we assume a Schechter (1976) luminosity function (measured
"locally") with faint end slope a and normalization 9*:

tt(L )dL = 9* (t* ) exp( -L/ L*)dL .

a (2)

We also use the observed scaling of the effective gaseous radius

R(L) = (35h- 1 kpc) (t*)/3 , (3)

 281

with f3 = 0.2. For no evolution in the red shift path density, with which the
Mg II observations are compatible (Steidel & Sargent 1992),
dN C<1'n
dz = Ho (1 -+- z )(1 + 2Qoz)-O.5 , (4)

where n is the (;amoving number density of galaxies and <1' is the cross
section per galaxy, 7r R2(L). We obtain the value of <1'n by integrating over
the luminosity function,

(5)

where Lmin is the luminosity at the apparent truncation in the absorbing

galaxy luminosity function, which we have seen corresponds to '" 0.2 times
the present- day L *. Thus, for Qo = 0, one obtains

dz = 11.5c} *r [a
dN + 1.4; Lmin / L *] , (6)

where r is the incomplete gamma function.

This value can then be compared directly to the observed dN / dz for the
Mg II absorption systems, which is equal to '" 0.8 at (Zab6) = 0.63. Note that
the galaxy radii scale relative to MB = -21.0 as observed at high redshift; if,
for example, there has been luminosity evolution between z '" 0.6 and z '" 0,
a zero redshift galaxy of luminosity L * would correspond to a brighter galaxy
at higher redshift, and thus to a radius larger than R* = 35h- 1 kpc. Simi-
larly, luminosity evolution will change the range of luminosities contributing
to the integral over the luminosity function, such that Lmin becomes smaller
if galaxies at z '" 0.6 are brighter than their z '" 0 counterparts. Figure 7
summarizes what the predicted observed absorber luminosity distribution
and normalization at (z) = 0.6 looks like for 3 different assumptions about
the extent of the galaxy luminosity evolution at rest frame B and two dif-
ferent local galaxy luminosity functions (representing the factor of '" 1.4
variation in c}* spanned by recent observations).
It is clear that, if there has been no luminosity evolution, one finds 2 to 3
times too many bright absorption-selected galaxies at z '" 0.6; in addition,
one can reject the null hypothesis of "no-evolution" based simply on the dis-
tribution of observed luminosities (ignoring normalization) at about the 3.5<1'
level. For L* = -21.5 at (z) = 0.6, both the shape of the luminosity distribu-
tion and the normalization are marginally consistent with the observations
(depending, in this case, on the chosen representation of the local luminos-
ity function); for a full magnitude ofluminosity evolution, the normalization
and luminosity distributions are fully consistent with the observations. With
more luminosity evolution than 1 magnitude, one would expect to find more
bright galaxies than are observed (this appears to be ruled out also by the
282

faint galaxy redshift surveys; e.g., Ellis, Lilly, this meeting ). Thus, with
0.5-1.0 magnitudes of luminosity evolution between z ""' 0.6 and z ""' 0, the
observations of the absorption selected galaxies are consistent in both lumi-
nosity distribution and in number density with the population of relatively
bright galaxies observed today. This implies that luminous galazies were al-
ready in place at z .<, 0.6, and that they have undergone only modest blue
luminosity evolution since that time. Consequently, it would be difficult to
accommodate a model in which the bright galaxies of today formed from the
mergers of the excess faint galaxies which are observed at moderate redshift
(cf. Broadhurst et al. 1991).
How are the absorption selected galaxies related to the galaxies which
dominate the faint counts (and which, by B rv 25, exceed the "no- evolution"
expectations by a factor of 5-10)? The answer seems to be that they cannot
be the same population. Indeed, although the distribution of [OIl].A3727
equivalent widths is not yet properly quantified for our survey, it is clear that
strong line emission is not a general characteristic of the absorption-selected
galaxies, although it is present in some cases. The largest [OIl] equivalent
width found thus far for an identified absorber is approximately equal to
the median value from the deep redshift survey of Colless et al. (1990) .
One's impression is that the number counts are dominated by galaxies which
are somewhat fainter than the faintest identified absorber. The fact that
extreme evolution in the faint galaxy population is a common ingredient for
any model which attempts to explain the excess numbers of blue galaxies
(see, e.g., Efstathiou et al. 1992, Lilly et al. 1991; Metcalfe et al. 1991 ;
Broadhurst et al. 1991; Colless et al. 1990) implies that galaxies fainter than
ME rv 19.5 must have completely different evolutionary histories than the
more luminous galaxies which are being picked up in the sample selected by
gas cross section.
Another implication of the survey results is that the cross section of'
optically-thick H I gas is dominated by luminous galaxies. The results seem
to argue strongly against a large abundance of galaxies having very large
H I sizes but very low surface brightness, such as Malin I (see, e.g., Impey
& Bothun 1989). That faint galaxies are very much underrepresented in the
absorption selected sample implies that their typical gaseous extents must
be on the order of a few kpc, rather than rv 30 - 40 kpc as found for the
luminous galaxies. It is interesting to note that the "transition" from large
to small gas cross sections appears to occur at approximately the luminosity
of the LMC, after accounting for the implied luminosity evolution in the
bright galaxies discussed above.
Perhaps the most important result from the absorption-selected galaxy
surveys is that selection by gas cross section appears to be a very efficient
means of finding what one would generally call "normal" galaxies at high
redshift . Since the absorption systems are well-studied to very high redshift,
 283

we can perhaps have some confidence on the basis of the moderate redshift
studies that metal-line systems are very good tracers of material which would
one day form the population of normal galaxies of the present epoch; as a
result, one can use the absorption-line studies in combination with standard
techniques to trace the history of galaxies similar to the Milky Way over
most of the age of the Universe.

4. Galaxies at z 2 .2: Current Status

Our current knowledge of the properties of galaxies at extreme redshifts is
limited to the highly unusual-Le., the radio galaxies and the QSOs. We can
of course "extrapolate" the association of moderate redshift absorption line
systems to apparently "normal" galaxies, and thus we can claim to trace,
e.g., the evolution of the chemical content of these presumably run-of-the-
mill objects beyond z = 4. However, it is a separate (and complementary)
question to ask what the stellar radiation was like for such young objects.
It is fundamentally important to our understanding of galaxy formation to
know what young galaxies looked like; e.g., were they diffuse, patchy, low
surface brightness objects, high surface brightness objects with very high
star formation rates, collections of small galaxies which would one day form
larger ones, or something else? Clearly the redshift surveys have only told us
that we ·haven't searched deep enough to find the high-redshift predecessors
of today's normal galaxies - but how deep does one have to go? Here again
the metal-line absorption systems are very useful because they essentially
tell us, within 10" or so, where there is a galaxy (or, protogalaxy) and
its precise redshift . One can then use standard techniques (Le., imaging or
spectroscopy) to attempt to identify the associated galaxy. Since we now
have some knowledge of the properties of absorption selected galaxies at
perhaps one half to two thirds of the look- back time of the very high-redshift
galaxies, one can, at the very least, test the null hypothesis that the galaxies
did not evolve over the intervening time.

4 . 1. SEARCHES FOR LYa EMISSION

Most of the effort aimed at identifying high-redshift absorption selected

galaxies has attempted to isolate Lya A1216 emission. The motivation be-
hind the method comes partly from the successes in using Lya emission to
study the morphology of emission line gas in very high-redshift radio galax-
ies, partly from the conventional wisdom that at early times galaxies were
undergoing large bursts of star formation (whereupon essentially each ioniz-
ing photon from young stars is reprocessed and emitted at Lya, so that the
line should be strong and thus more easily detected against the sky back-
ground than continuum emission), and partly because it is the strongest
284

spectroscopic feature expected at optical wavelengths for z ~ 2. The investi-

gations to date basically fall into two categories: narrow-band imaging, and
spectroscopy. Each method has its advantages; the narrow-band method
provides complete 2-dimensional information, so that the emitting object(s)
need not fall within a small spatial area to be detected. The spectroscopic
method may be preferred if the emitting regions are compact with intrin-
sically narrow lines, so that the added velocity resolution achievable with
long-slit spectroscopy greatly improves the contrast of any putative detec-
tion. The target absorbers have generally been the damped Lyman a sys-
tems, which is well-motivated in the case of the spectroscopic observations
because one knows only where the QSO is, not the absorber, and the regions
having large enough H I column densities to produce a damped profile are
likely to be concentrated toward the inner regions of the galaxy, making it
more likely that the galaxy resides near the QSO line of sight. It is likely
that the metal-line systems in general, particularly those selected because
they give rise to a Lyman limit absorption system, are sampling different
parts of the same objects which give rise to damped Lya absorption, so that
in the case of narrow-band imaging there really is no compelling reason to
confine oneself to damped systems. In fact, systems which are not damped
are more likely to be well-separated from the QSO line of sight, making the
contrast between object and background all the better.
The :first detection of Lya emission from a damped Lya system was re-
ported by Hunstead, Pettini, & Fletcher (1990), using long slit spectroscopy
of the QSO H0836+ 113. The line was detected as a narrow emission fea-
ture in the core of the damped Lya absorption line. The flux in the Lya
line implied a star formation rate of rv 1 Me yr- 1 using a standard initial
mass function and the assumption of "Case B" recombination. The emis-
sion all fell within 1 spatial pixel, corresponding to rv 2", thereby implying
that the emitting region is relatively small. Hunstead et al. (1990) favored
a star-forming dwarf galaxy, or a "naked HII region" as the origin of the
emission. There have been several narrow-band imaging surveys for Lya
emission from the damped systems; Smith et al. (1989) observed 4 DLA
systems with narrow band filters with FWHMrv 20 - 25 A, and placed up-
per limits on the star formation rates of 1 - 5 Me yr- 1 • Lowenthal (1991)
has used both narrow band imaging and long-slit spectroscopy for 7 DLA
systems, setting upper limits on the star formation rate of, again, 1 - 4 Me
yr- 1 • Interestingly, Lowenthal has not been able to confirm the detection of
Lya emission for the H0836+113 absorber. Wolfe et al. (1992) have made a
very detailed study of this same absorber using both narrow-band imaging
and spectroscopy; they too are unable to confirm the spectroscopic detec-
tion. However, from the deep narrow band images, they report a detection
of extended line emission, of very low surface brightness, spread over an
area corresponding to a size of 24 X 11h- 2 kpc, which they argue favors an
 285

identification with a large galaxy. The implied Lya flux is at about the level
of the upper limits in most of the cases of non-detections. However, the case
of H083S+113 is further complicated by the discovery (Lowenthal 1991) of
a z = 0.78 Mg II absorbing galaxy which is nearly spatially coincident with
the putative Lya emitter. Nevertheless, the implications of all of the sur-
veys, both reported detections and upper limits, are clear: the damped Lya
systems, which are believed to represent the population of "normal" objects
at z '" 2, have very weak Lyman a emission. In the absence of dust (assumed
for all of the estimates of star formation rates), this implies that typical ob-
jects at high redshift were not forming stars any more rapidly than typical
quiescent spiral galaxies at the present epoch.
However, the Lya emission can be subject to severe extinction by even
a relatively small quantity of dust, due to resonant scattering. Charlot &
Fall (1991) estimate that, given dust even at the relatively low level implied
by the observations of the damped systems ( ~ 0.1 times the dust/gas ratio
in the local interstellar medium; see, e.g., Pei, Fall, & Bechtold 1991), it
is conceivable that the Lya emission is attenuated by a factor as large as
100. Of course, depending on the geometry of the emitting regions and the
distribution of the dust, virtually any value between no extinction and 5
magnitudes of extinction is possible. Thus, while the observational situation
for Lya emission seems clear, the implications for the star formation rates
are still very uncertain.
A by-product of the narrow-band searches for emission from damped Lya
systems has been the search for companions to the absorbing galaxies - that
is, galaxies which have nearly the same redshift as the absorber, but are
not close enough to the line of sight that they could easily be producing
the observed absorption themselves. Lowenthal et al. (1991) discovered such
an object at z = 2.3 in the field of PHL 957, 48" from the QSO line of
sight and differing in redshift from the DLA absorption system by f'V 900
km S-l . The spectrum of this object, which has B '" 24, bears an uncanny
resemblance to the spectra of high-redshift radio galaxies (detectable He II
and C IV emission, together with Lya), yet it has been confirmed to be
radio-quiet (Lowenthal 1992). A second example of a "DLA companion"
was reported by Turnshek et al. (1991) in the field of QOOOO-263, at z =
3.408 and 19" from the QSO line of sight. This galaxy has V f'V 23.5 and
somewhat weaker line emission than the PHL 957 companion, and would
represent the highest redshift radio quiet galaxy yet identified. While further
observations have indicated that the object is probably at a much smaller
red shift (zf'V 0.44), where the line emission is attributed to [011] A3727
rather than Lya (Machetto et al. 1993; Steidel & Hamilton 1992; see also
§4.3), a new candidate in the same field and with similar properties to the
PHL 957 companion has been identified (Machetto et al. 1993).
Complementary observations appear to support the latter identification
286

(Steidel & Hamilton 1993; see §4.3). Both the QOOOO-263 and the PHL
957 DLA "companions" have Lya equivalent widths which are near the
maximum that can feasibly be produced (even in the dust-free case) by
stellar populations. Are these objects galaxies, or are they low-luminosity
AGNs? What is the source of the excitation of the line emission? Why are
their properties so different from the damped Lya absorbers, yet they appear
to be in some way associated with, and may in fact be significantly clustered
with, the sites of QSO absorption (see Lowenthal 1991; Wolfe 1993)? The
answers to these questions will, of course, require further observations.

4.2. SEARCHES IN THE INFRARED

Given the implications of the searches for Lya emission detailed above, i.e.,
that either star formation rates were rather low for systems selected on the
basis of damped Lya absorption at z i?, 2, or the star formation rates were
almost arbitrarily high and the Lya emission is extinguished by resonant
scattering, makes searches using infrared techniques particularly important.
For z i?, 2, lines such as [01l]A3727, H,8, [OIII]AA4959, 5007, and Ha are
redshifted into the various windows available to ground-based infrared de-
tectors. To date, the only published example of such observations are by
Elston et al. 1991, who reported the detection of lines of [01l]A3727 and H,8
from the damped Lyman a system at z = 1.998 toward Q1215+333. Amaz-
ingly, although no Lya emission is detected to flux limits of"" 3 X 10- 17
ergs s-1 cm- 2 (Lowenthal 1991), the IR observations of [011] imply a star
formation rate in excess of 100 Me yr- 1! If the result is confirmed, and if
similar observations of other absorption systems yield comparable results,
one would have to hide a population of exceedingly luminous galaxies from
detection in the faint spectroscopic samples. For example, for qo = 0, the
expected B magnitude of such a galaxy would be in the neighborhood of
'" 22 - 23, so that unless the UV continuum as well as the Lyman a line is
heavily suppressed, such objects should be found in the galaxy surveys to
B '" 24 if they are common.
Further IR spectroscopy of z > 2 absorbing galaxies is eagerly antici-
pated; it is a very promising technique, particularly if the rather pessimistic
expectations for Lya emission hold true. The one caveat in performing
"blind" spectroscopy of absorbing galaxies is that there is no guarantee that
the centroid of the absorber is coincident with the QSO position; depending
on qo, even for damped Lya systems (which probably represent the inner-
most regions of absorbing galaxies-see the cartoon in Figure 1), the angular
separation of the absorber from the QSO line of sight might be as large
as 3- 4". Unfortunately, without complementary imaging observations, this
fact makes it somewhat difficult to interpret non-detections of line emission
from the absorber.
 287

1.3. SEARCHES USING UV CONTINUUM PROPERTIES

Finally, we discuss the use of contin'uum observations in the observed-frame

optical (rest frame UV) of absorption-line selected systems. At sufficiently
high redshift (z ~ 3), the rest frame Lyman limit (912 A) is shifted into the
window accessible to ground- based observations. As emphasized by many
authors (see, e.g., Songaila et at. 1990; Guhathakurta et al. 1991; Steidel
& Hamilton 1992), one expects a pronounced discontinuity at the Lyman
limit for all but the most contrived initial mass function in a galaxy which
is undergoing a burst of star formation. In addition to the intrinsic spectral
energy distribution of the stars contributing the bulk of the UV photons,
however, even a very small quantity of H I gas in the galaxy will make the
effective discontinuity at the Lyman limit quite severe. It is possible that no
UV photons just above the Lyman edge actually escape from star-forming
galaxies, since in general the highest columns of H I will be coincident with
the sources of the UV photons. Even if some photons do escape the galaxy,
there is a substantial probability that an eztrinsic Lyman limit absorption
system will further attenuate the spectrum. In view of this, the "Lyman
edge" is probably the most reliable continuum spectroscopic feature for any
high-redshift object which is forming stars. This fact has been exploited by
Guhathakurta et al. (1991) to set a limit on the redshifts of the bulk of the
very faint blue galaxies in the deepest imaging surveys to date. The fact
that at most a few percent of the galaxies to B rv 27 are not detected in
the observed U band shows that in general the galaxies must have z < 3.
Thus, even the faintest population of galaxies is not dominated by very
high-redshift galaxies.
On the other hand, the observations have not shown that high-redshift
g<ilaxies do not exist at z > 3; rather, they show that most of the galaxies
which have faint apparent magnitudes are at lower redshift. For example,
suppose that galaxies at z ~ 3 had the same distribution of luminosities as
the absorption selected systems at z rv 0.6. We know the statistical proper-
ties of the Lyman limit absorption systems to z '" 4, and that they (at least
for z < 2) are the same population as the systems selected on the basis of
Mg II absorption (see §3.2). Adopting the same dependence of galaxy cross
section on luminosity as inferred from the moderate redshift systems, one
can calculate the number of galaxies expected in a particular redshift range
per area on the sky. It turns out that the surface density of galaxies in the
redshift range 3 ~ z :S 3.5 would be fairly high (of the order of 4.3 X 104
deg- 2 ), but unless there has been substantial luminosity evolution between
z rv 3 and z rv 0.7, only a few percent of these would have been detected
in the imaging surveys to B rv 26, and none would have B :::. 24. Note that
even if all of the galaxies were detected in the deep imaging surveys in that
redshift range, they would still represent only :::. 10% of the total surface
288

density of galaxies. We go through these rough arguments only to empha-

size that no observations to date have ruled out a population of moderately
luminous, star-forming galaxies at very high redshift. They really have only
ruled out a large population of radio-quiet radio galaxies!
In order to attempt to identify examples of "normal" high-redshift galax-
ies on the basis of continuum observations, we have recently begun a survey
designed to combine the redshift information from the study of QSO absorp-
tion line systems with the method of "ultra-deep" broad band imaging. The
first results of the survey are given in Steidel & Hamilton (1992, 1993). The
idea behind the deep continuum imaging of high-redshift QSO fields is sum-
marized in Figure 8. The Un and G filters are optimally placed to measure
any Lyman continuum break in objects at the redshift of the z = 3.390 ab-
sorption system. Basically, we are attempting to assign redshifts to galaxies
near the line of sight to a QSO whose absorption spectrum is well-studied,
so that perhaps 4 or 5 metal-line absorption systems in the redshift range
1 ~ z ~ il.5 are known.
The fields for study have been chosen such that the QSO has an optically-
thick Lyman limit absorption system at z > 3.1, so that a broad bandpass
can be placed completely below the Lyman edge of any galaxies which are
associated with the object producing the absorption. At the same time, of
course, the QSO will be completely black in the UV passband, so that any
galaxies with z < ZLL should be detectable even if they lie exactly at the
position of 'the QSO on the plane of the sky. Because the vast majority of
faint field galaxies are very blue in Un - G ((Un - G) = 0.35 for 'R > 24),
objects at the redshift of the Lyman limit should be well separated in color
from the general locus. Using this method, we believe that we have identified
the galaxy responsible for the Z = 3.390 DLA absorption system in the
spectrum of QOOOO-263. The object has G = 25.5 (B '" 25.7), 'R = 24.8,
and Un - G > 1.9 (see Figure 8). Note that "A" is not present in the Un
picture (nor is the QSO) to a limit of Un '" 27.4. The object labeled "G"
is the galaxy identified to have Z = 3.408 by Turnshek et ale (1991); it
has an essentially flat spectrum in Un - G, suggesting that it is probably
at lower redshift. If the object is at z = 3.4, then the projected distance
from the QSO line of sight is 10 - 20h- 1 kpc, and its absolute magnitude
is MB '" -22 - placing it well within the range of the z '" 0.6 absorption
selected galaxies! At the same time, the lack of a discontinuity between G
and Un for the object marked "G" (the galaxy identified by Turnshek et ale
(1991) as having z = 3.408) implies that, realistically, the identification as a
very high-redshift galaxy cannot be correct. The second z = 3.4 candidate,
identified by Machetto et ale (1993), is also in the field of the deep continuum
images, and has G = 25.2, 'R = 24.2, and Un - G > 1.7. The presence ofthe
expected Lyman continuum break would appear to confirm that the object
is indeed at high redshift (Steidel & Hamilton 1993).
 289

100
2 •• 1=3.390

80
~
(I)
0 60 5t
C
...,
<U
-'
·E 40
'"C<U
'-
E-

3000 4000 5000 6000 7000

Wavelength (J..)

QOOOO - 263 G- PSF'

~ ~

-g'" 20 -g'" 20
o o
u o
(I) (I)
III
'"
U
'-
<Il
...o<Il
o o
u u
(I) (I)
Q Q
(I) (I)

.::
-'
.:::
-'
!1Qj - 20 !1
(I)
- 20
0:: 0::

- 20 0 20
Relative RA (arc seconds)

Fig. 8. Top panel: A model spectrum of a star-forming galaxy at the red shift of the
Lyman limit/DLA absorption system is plotted on the same axes as the transmis-
sion curves of the broad band filters and a low-dispersion spectrum of the QSO.
Bottom panel: Contour diagrams of the G and Un frames in the I' square region
surrounding the QSO. The QSO profile has been subtracted in the G picture to
reveal galaxy "A", separated by'" 2.8/1 from the QSO sightline (see text).
290

Clearly, it will be interesting to see if the absorption selected galaxies at

z '" 3 are generally not overly luminous. If this turns out to be the case,
then all of the null results (and the faint detections) of Lyman a emission
may be telling us that high-redshift galaxies really do have only modest star
formation rates after all, and that in many ways they may resemble their
lower redshift counterparts.

5. Summary and Conclusions

The principal conclusions reached for the two different redshift regimes dis-
cussed are as follows:

5.1. Z < 1

1. The QSO absorption systems which are selected on the basis of the Mg II
AA2796, 280~ resonance doublet are the same population of systems defined
by having an optical depth THI > 1 in the H Lyman continuum (N(H I»
3 x 10 17 cm- 2 ); consequently, an unbiased sample of systems selected on
the basis of the Mg II doublet is equivalent to an H I-selected sample. The
objects giving rise to this absorption are therefore selected on the basis of
their gas cross section.
2. There is now considerable information on the nature of absorption-
selected galaxies at moderate redshifts, where the galaxies may be compared
to those selected by other criteria. For a statistically unbiased sample of
Mg II absorption systems in the red shift range 0.3 :s; z :s; 0.9, a confirmed
or candidate absorbing galazy can be identified in every case.
3. The optical and optical/infrared colors of the identified absorbing
galaxies are consistent with mid- to late-spiral spectroscopic types. Very
blue or very red galaxies are rare, and strong line emission, while present in
some cases, is not a general feature of the galaxies.
4. There is a correlation at the 99.9% confidence level of the observed
"impact parameter" of the QSO sight line with the centroid of the galaxy
image, in the sense that brighter galaxies tend to have larger linear separa-
tions from the QSO line of sight. The distribution of "impact parameters"
of the QSO sight lines with the centroids of the galaxy images is consistent
with spherical gas distributions and an approximate scaling of galaxy size
with luminosity, R(L) = 35h- 1 kpc(L/L*)O.2. The general absence of "inter-
loper" galaxies is consistent with a gas covering fraction near unity within
R( L), and is inconsistent with a flattened geometry and random angle of
inclination as would be the case if the absorbers were thin disks.
5. There are correlations of the absorption-line properties with the im-
pact parameters to the identified absorbing galaxies. That is, given the
absorption-line properties, one can predict a priori where the galaxy will
 291

be relative to the QSO line of sight. Together with point 4 above, the cor-
relations appear to be strong indications that the identified galaxies are
actually responsible for the absorption lines, and make it less likely that the
absorption can be ascribed to "unseen", fainter objects which would tend to
be hidden by the relatively bright QSO profile.
6. The rest-frame B magnitudes of the absorbing galaxies «(z) = 0.6 are
in the range -19.4 2: ME 2: -23.3 with a mean (and median) ME = -21.0;
thus, there is a conspicuous absence of intrinsically faint galaxies in the
sample selected by gas cross section. We have argued that this cannot be
due to incompleteness. Apparently, with few exceptions, all galaxies brighter
than ME '" -19.5 at z '" 0.6 produce Mg II absorption for a line of sight
within R( L) of the galaxy centroid. After accounting for modest luminosity
evolution at rest frame B (0.5-1.0 magnitudes) between z '" 0.6 and z '"
0, both the space density and the luminosity distribution of the galaxies
are consistent with the same population of relatively bright galaxies of the
present epoch. Selection by gas cross section therefore picks out a population
of bright galaxies that was already in place by z '" 0.9, and which has
remained largely stable over more than one half the age of the Universe. This
inference would tend to argue against hypotheses involving the assembly
of present-day bright galaxies from small galaxies at moderate redshifts. If
mergers were important to the formation of bright field galaxies, the "merger
epoch" is likely to have been prior to z '" 1.
7. The modest evolution of the bright galaxies, together with the absence
of faint galaxies from the absorption-selected sample, has several implica-
tions. First, the population of galaxies at B '" 24 - 26 which exceeds "no
evolution" expectations by a very large factor and whose brighter mem-
bers dominate the faint field galaxy redshift surveys cannot be the same as
the absorption-selected population. The fact that galaxies near ME '" -19
dominate the redshift surveys but make ahnost no contribution to the ab-
sorption selected sample implies that the gaseous sizes of these "faint blue
galaxies" must be on the order of a few kpc, rather than the 30-40 kpc radii
typifying the bright galaxies. A substantial population of very large (and gas
rich) but low surface brightness galaxies at moderate red shift also seems to
be ruled out . Generally speaking, completely different evolutionary histories
and gaseous sizes are indicated for galaxies brighter than and fainter than
'" LMC luminosities.

5.2. Z ~ 1

1. Strong Lyman 0: emission is not a common characteristic of very high-

redshift galaxies; whether the star formation rate in typical objects is high
or low is as yet inconclusive because of the possibility that Lyo: emission
is attenuated by the resonant scattering process. Taken .literally, though,
292

the limits and detections of Lya emission imply star formation rates which
are not much higher than in present-day spiral galaxies. Unlike the galaxies
responsible for the damped Lya absorption, there appears to be a population
of objects (AGN or galaxies?) with strong Lya emission which, while they
are not the QSO absorbers themselves, may be associated with or even
clustered with the absorbing galaxies. These line-emitting objects must be
much rarer than the objects responsible for the bulk of the QSO absorption
cross section.
2. There is probably a population of z ~ 3 galaxies in the very deep imag-
ing surveys, beginning at magnitude levels of B '" 25.5 - 26; however, they
do not dominate the counts at these magnitudes, and still-fainter samples of
field galaxies will probably prove inefficient in turning up a large number of
high-redshift objects. No observations thus far have ruled out the possibility
that the same population of galaxies may already be in place at z ~ 2 as
is identified with the Mg II absorbers at z '" 0.6. Concentrated efforts at
identifying the objects responsible for very high-redshift absorption systems
( through narrow-band imaging, infrared imaging and spectroscopy, and UV
continuum studies) are likely to be the most efficient means of uncovering
the nature of very early (normal) galaxies (or proto-galaxies). In combina-
tion with the detailed physical information on the objects which are obtained
from the QSO absorption lines themselves (some of which are summarized in
Ken Lanzetta's contribution to these proceedings), the possibility of putting
together a rather complete picture of the early history of normal galaxies
seems quite feasible.

Acknowledgements
I would like to thank the Local and Scientific organizing committees for a
very enlightening and enjoyable meeting, and for financial support so that
I could attend. My collaborators, Mark Dickinson, Don Hamilton, and Wal
Sargent, are thanked for allowing me to describe joint work prior to publica-
tion, and of course for innumerable discussions: I am grateful to Hy Spinrad
for reading an earlier draft of this manuscript. Some of the work presented
here has been supported by NASA through grant #HF-I008.01-90A awarded
by the Space Telescope Science Institute which is operated by AURA, Inc.
for NASA under contract NAS5-26555.

References
Bechtold, J. & Ellingson, E.: 1992, Ap.J. 396, 20
Bergeron, J., & Boisse, P.: 1991, A6tr. Ap. 243, 344
Bergeron, J., & Stasinska, G.: 1986, A6tr. Ap. 169, 1
Broadhurst, T.J., Ellis, R.S. , & Glazebrook, K.: 1992, Nature 355, 55
Charlot, S., & Fall, S.M. : 1991, Ap.J. 378, 471
Coleman, G.D., Wu, C.C., & Weedman, D.W.: 1980, Ap.J.S. 43, 393
 293

Coiless, M, Ellis, R.S., Taylor, K., & Hook, R.N.: 1990, MNRAS 244, 408
de Lapparent, V., Geiler, M. J., & Huchra, J. P.: 1989, Ap.J. 343, 1
Efstathiou, G. ,Bernstein, G., Katz, N., Tyson, J.A., & Guhathakurta, P.: 1991, Ap.J.L.
380,L47
Elston, R.J., Bechtold, J., Lowenthal, J.D., & Rieke, M.: 1991, Ap.J.L 373, L39
Guhathakurta, P., Tyson, J.A., &; Majewski, S.R.: 1990, Ap.J.L. 357, L9
Heisler, J., Hogan, C. J., & White, S.D.M.: 1989, Ap.J. 347, 52
Holmberg, E.: 1975, in Stars and Stellar Systems, 9, Galaxies and the Universe, ed(s)., A.
Sandage, M. Sandage, iJ J. Krilltian, (Chicago: Univ. Chicago Press), 123
Hunstead, R.W., Pettini, M., & Fletcher, A.B.: 1990, Ap.J. 365, 23
Impey, C.D., & Bothun, G .D.: 1989, Ap.J. 341, 89
Lanzetta, K.M., Wolfe, A. M., Turnshek, D.A., Lu, L., McMahon, R.G., and Hazard, C.:
1991, ApJS 77, 1
Lanzetta, K.M., & Bowen, D.: 1992, Ap.J. 391, 48
Lanzetta, K.M., & Bowen, D.: 1990, Ap.J. 357, 321
Loveday, J., Peterson, B. M., Efstathiou, G., & Maddox, S. J.: 1992, Ap.J. 390, 338
Lowenthal, J.D.: 1991, PhD Tht!llill , University of Arizona
Lowenthal, J. D., Hogan, C.J., Green, R.F., Caulet, 'A., & Woodgate, B.E.: 1991, Ap.J.L.
377,L73
Lilly, S. J., Cowie, L. L., & Gardner, J. P.: 1991, Ap.J. 369, 79
Machetto, F ., Lipari, S., Giavalisco, M., Turnshek, D., & Sparks, W.B.: 1993, Ap.J. 404,
511
Metcalfe, N., Shanks, T., Fong, R., & Jones, L. R.: 1991, MNRAS 249, 498
Pei, Y., Fall, S.M., & Bechtold, J.: 1991, Ap.J. 378, 6
Phillipps, S., Disney, M.J., & Davies, J.I.: 1993, MNRAS , in press
Sargent, W. L. W., Boksenberg, A., & Steidel, C. C.: 1988, Ap.J.S. 68, 539
Sargent, W.L.W., Steidel, C.C., & Boksenberg, A.: 1988, Ap.J. 334, 22
Schechter, P.: 1976, Ap.J. 293, 97
Smith, H.E., Cohen, R.D., Burns, J.E., Moore, D.J., & Uchida, B.A.: 1989, Ap.J. 347, 87
Songaila, A., Cowie, L.L., & Lilly, S.J.: 1990, Ap.J. 348, 371
Steidel, C.C.: 1992, in Galaxy Evolution: The Milky Way Perspective, ed(s)., S. Majewllki,
(San Francisco: ASP), in press
Steid'el, C. C., & Dickinson, M.: 1992, Ap.J. 394, 81
Steidel, C. C., & Hamilton, D.: 1992, A.J. 104, 941
Steidel, C. C., & Hamilton, D.: 1993, A.J. , in press, June 1993
Steidel, C. C., & Sargent, W. L. W.: 1992, Ap.J.Suppl. 80, 1
Turnshek, D.A., Machetto, F., Bencke, M.V., Hazard, C., Sparks, W.B., & McMahon,
R.G.: 1991, Ap.J. 382, 26
Tyson, A.: 1988, AJ 96, 1
Wolfe, A.M.: 1993, Ap.J. 402, 411
Wolfe, A. M., Turnshek, D. A., Lanzetta, K. M., &; Oke, J. B.: 1992, Ap.J. 385, 151
Wolfe, A.M., Thrnshek, D. A., Smith, H.E., &; Cohen, R.D.: 1986, Ap.J.S. 61, 249
Yanny, B., & York, D.G: 1992, Ap.J. 391,569
Yanny, B., York, D.G., & Williams, T.B.: 1990, Ap.J. 351, 377
York, D.G., Dopita, M., Green, R., & Bechtold, J.: 1986, Ap.J. 311, 610
Young, P., Sargent, W. L. W., &; Boksenberg, A.: 1992, Ap.J.S. 48, 455
MOLECULAR GAS AND THE YIELD OF YOUNG STARS IN
GALAXIES

JUDITH S. YOUNG
Department of Physics and Astronomy and
Five College Radio Astronomy Observatory
University of Massachusetts, Amherst, MA 01003

Abstract. Recent results regarding the global gas properties of spiral galaxies are re-
viewed, with special emphasis on the systematic trends in the ratio of molecular to atomic
gas and in the yield of high-mass stars per unit mass of molecular gas as a function of
galaxy morphology, bar type, and environment.

1. INTRODUCTION

The distribution and abundance of dense interstellar matter playa critical

role in determining the morphology and evolution in the disk of a galaxy,
since stars form in molecular clouds, and the most massive of these stars
produce a significant fraction of the optical luminosity. A major goal of
studies of spiral galaxies is to understand why stars form where they do
and to explain the differences in this process from galaxy to galaxy. Thus,
it is important to determine the mass of molecular clouds available to form
stars, and to compare this with the mass of atomic gas and with tracers of
star formation in order to isolate the most basic physical parameters which
..
influence current epoch star formation and the overall evolution of galaxies .
Such comparisons provide the only means to study the yield of high mass
stars per unit mass of molecular gas, or the efficiency of high mass star for-
mation within and among galaxies and to relate it to galactic properties such
as Hubble type, luminosity, and environment. The acquisition of molecular
line data in external galaxies is therefore an essential part of the study of
galactic evolution.
Using the 14m millimeter telescope of the Five College Radio Astronomy
Observatory (HPBW = 45"), we have now completed the observations of
the distribution and global content of CO J = 1 - 0 emission (A = 2.6 mm)
in 300 galaxies (Young et al. 1993). The galaxies observed as part of the
FCRAO Extragalactic CO Survey were selected to cover a wide range of
properties, including lenticulars, spirals, and irregulars; galaxies with lumi-
nosities between 106 and 10 12 L0 in the infrared; environments from pair to
cluster, and including isolated and merging systems. This sample of galaxies
is not complete in a flux-limited sense, nor is it volume-limited, but rather
the galaxies observed were selected to span a wide range of parameter space.
Among galaxies which are either brighter than 11th magnitude in the blue
or brighter than 15 Jy at 100 pm, the Survey is 80% complete.

295
1. M Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, 295-304.
© 1993 Kluwer Academic Publishers.
296

Because many of the galaxies in the Survey are at distances of 20 Mpc

or greater, the properties we are determining are averages over regions at
least a few kiloparsecs across. The most effective way to compare galaxies
which cover a wide range in size and luminosity is to investigate the ratios
of properties, where the size differences among galaxies are normalized out.
The ratio of molecular to atomic gas and the star formation rate per unit
mass of molecular gas are the two size-independent quantities which will be
discussed here; for a more complete review of the molecular gas properties
of galaxies, see Young & Scoville (1991).
The H2 masses used in this work were derived from major axis CO ob-
servations, assuming a constant CO/H 2 proportionality of 2.8 X 10 20 cm- 2
[K km s-1 ]-1 (Bloemen et al. 1986). The global H2 masses derived for lu-
minous spirals have been shown to be statistically accurate to 50% (Dev-
f'J

ereux & Young 1990), and the absolute value of the CO/H 2 proportionality
has been verified in M31 and M33 by observations that resolve individual
molecular clouds. This latter result is illustrated in Figure 1, where the
CO luminosities and virial masses of Galactic and extragalactic molecular
clouds are compared. The molecular clouds in the Milky Way (Scoville et al.
1987; Solomon et a.J.. 1987; Carpenter et al. 1990), M 31 (Vogel et al. 1987),
M 33 (Wilson & Scoville 1989, 1990), and IC 10 (Wilson & Reid 1991) are
apparently similar, even though they represent regions of lower metallicity
and are found in galaxies ranging from type Sb (M 31) to Scd (M 33) to
irregular (IC 10).

2. RATE OF STAR FORMATION PER UNIT MASS OF

MOLECULAR GAS
Perhaps one of the most fundamental parameters describing the current-
epoch star formation in a galaxy is the ratio of the star formation rate to
the mass of gas available to form stars. This yield of high mass stars per
unit mass of molecular gas is sometimes called the star formation efficiency.
It has been know since before IRAS that interacting galaxies exhibit ele-
vated rates of star formation (Larson & Tinsley 1978; Lonsdale, Persson, &
Matthews 1984; Josp.ph & Wright 1985) relative to non-interacting galaxies.
Additionally, with the advent of the IRAS Survey, it was shown that in-
teracting galaxies also exhibited elevated yields of high mass stars per unit
mass of molecular gas relative to the yield in isolated galaxies (Young et al.
1986; Sanders et al. 1986; Solomon & Sage 1988; Tinney et al. 1990). To
date, of all of the investigations of the star formation efficiency in galaxies,
the largest effect which has been found is the factor of '" 5 - 10 elevated
star formation efficiency in interacting galaxies versus isolated galaxies, an
effect which is present when using star formation rates traced by either the
Ha or IRAS emission (Young, Allen, & Kenney 1993). Given that the star
 297

t3 [ T - ----,-- I I I I I I I I I 1, -'-, I
Inner Galaxy Clouds
7 Outer Galaxy Clouds

o M33 Clouds
6 6 M31 Clouds

o Ie 10 Clouds . ..
.. :.;:.:. .. ... ~t
• • • • • : • • f'I.~• •

Fig. 1. Comparison of CO luminosities and virial masses of molecular clouds in the

Milky Way, M 31, M 33, and IC 10 (for references see text). In the Milky Way, the
CO lurtlinosity is closely correlated with the virial masses of the clouds, both with
and without high-mass star formation. This linear proportionality justifies the use
of CO as a tracer of the mass ofH 2; the best fit to these data for clouds with masses
between 10 5 and 2 x 10 6 M0 yields a constant of proportionality of 3 x 10 20 H2
cm- 2 [K km s-1]-I . The similarity of the clouds in M 31, M 33, and IC 10 to the
Milky Way justifies the use of the Galactic COjH2 proportionality in the external
galaxies.

formation efficiencies in galaxies cover '" 2 orders of magnitude, one order

of magnitude can be attributed to the effects of environment.
Figure 2 shows a plot of the ratio LIR/M(H2) versus dust temperature
for 150 galaxies in the Survey, in which the points are coded by the lo-
cal environment (isolated galaxies, group and cluster galaxies, merging and
interacting galaxies, and dwarf irregular galaxies). From this figure, it is
clear that isolated galaxies are characterized by low values of both the IR
luminosity to gas mass ratio and the dust temperature, while the strongly
interacting galaxies are characterized by high values of both quantities. If
the IR luminosity is powered by young high mass stars, these results indi-
cate that the interacting galaxies produce a higher yield of young stars per
298

• Isolated Galaxies
2.5 a Group and Cluster Galaxies
* Merging/Interacting Galaxies
.--...
::;s
0 . Irregular Galaxies *
2 0
'------
0 *
.....:I f:>
0
-------- *
**' *
0 0
.--... 1.5 lY> **0 *
o~
~t*o
N
~
0 0
::;s
--------
'------
0::
1 *
S
QLJ
0
........
.5 •

o .2 .4 .6 .8 1 1.2 1.4
Sao/SlOo
Fig. 2. Comparison of the ratio LIR/ M(H2) with the ratio of 60/100 J-tm flux den-
sities for 150 galaxies in the FCRAO Extragalactic CO Survey. The galaxies are
coded by environment as indicated. The merging/interacting galaxies have high
values of both LIR/ M(H2) and 8 60 /8100 relative to the isolated galaxies.

unit mass of molecular gas than isolated galaxies, and the resultant higher
energy density in the radiation field would be expected to heat the dust to
higher temperatures than in the isolated galaxies.
Several mechanisms have been suggested for the enhancement in the yield
of young stars per unit mass of molecular gas in interacting/merging galax-
ies. On the one hand, a galaxy interaction may cause a new star formation
mechanism to operate such that more stars form per unit molecular mass.
Alternatively, the physical processes that cause high-mass stars to form
in merging/interacting galaxies may be no different than those in isolated
galaxies, but the interaction may enhance their effectiveness. One mecha-
nism which is aheady present in a galaxy, and which probably becomes im-
portant during galaxy-galaxy interactions, is that of cloud-cloud collisions.
Numerical simulations of galaxy-galaxy interactions, including gas clouds as
well as stars (Noguchi & Ishibashi 1986; Olsen & Kwan 1990), indicate that
the cloud-cloud collision rate increases as bridges and tails develop during
 299

violent galaxy-galaxy encounters. If cloud-cloud collisions are responsible for

high mass star formation in galaxies, the enhanced rate of cloud-cloud colli-
sions in interacting galaxies should result in an increase in the star formation
rate per unit mass of H 2 •
Investigations of the mean value olthe star formation efficiency in galaxies
along the Hubble sequence have recently been conducted using the Extra-
galactic CO Survey. It is interesting that the mean value ofthe star formation
rate per unit mass of molecular gas, '" 5 L0/M0' is found to be typical of
spiral galaxies of all types and therefore independent of the global morphol-
ogy of the galaxy (using LIR to trace the star formation rate - Devereux &
Young 1991; using L(Ha) to trace the star formation rate - Young, Allen,
& Kenney 1993). Figure 3 shows the ratio LIR/M(H 2 ) for the spiral and
irregular galaxies in the Survey, with upper limits coded in black. It appears
that galaxy morphology has little to do with the factor of 10 spread in the
star formation efficiency which is present among galaxies of similar type and
environment.
Another global property has been uncovered which appears to correlate
with an elevated yield in high mass stars per unit mass of molecular gas -
the bar morphology of a galaxy. Among 219 of the Survey galaxies which
have bar classifications in the RC2 (de Vaucouleurs et a1. 1976), 141 are
barred or mixed, and 78 are unbarred. The mean value of the star formation
efficiency for the barred galaxies is a factor of '" 2 higher than for the
unbarred galaxies, a trend which is present in both the early-type and late-
type galaxies, independently. Figure 4 shows the histograms of the ratio
Lm/M(H2) for the barred and unbarred galaxies, illustrating the different
mean star formation efficiencies for these samples.
There is also a property of the nuclei of galaxies which has been found
to correlate with enhanced global star formation efficiencies. For a 'complete
sample of nearby starburst galaxies in which starbursts are occurrirtg in the
central 500 pc with an intensity comparable to the starburst in M82 (Dev-
ereux 1989), the mean global yield of young stars per unit ma.ss of molecular
gas is intermediate in value between isolated galaxies and interacting galax-
ies (Young & Devereux 1991). Future work will address the variations in
the star formation efficiency as a function of position within starburst and
"normal" galaxies.

3 . MOLECULAR AND ATOMIC GAS

Since stars form from molecular clouds, and molecular clouds form from
atomic gas, the mass of gas in molecular versus atomic form in galaxies
is of interest in assessing the parameters which lead to a high abundance
of gas available for star formation. The FCRAO Extragalactic CO Survey
provides a unique opportunity to determine the molecular to atomic gas
 300

til
.....
Q)
25=-~~~~~'-"'-"'-OT'-rT'-ro'3
20
>< 15
cd
....... 10
cd 5
t-' o E:L-J---.l-L..-L___=
...... - 1 0
o
I-..
Q)
.0
S
::I
Z

Fig. 3. Histograms of the distribution of the ratio LIR/M(H 2 ) for 257 spiral and
irregular galaxies in the FCRAO Extragalactic CO Survey showing the same mean
value of the star formation rate per unit mass of molecular gas for spiral galaxies of
a wide range of morphology, as discussed by Devereux & Young (1991). The shaded
portions of the histograms represent galaxies with lower limits to the LIR/M(H2)
ratio due to upper limits in the H2 mass determination.
 301

60

40

20

0
Vl - 1 0 2 3

·xte
Q)

-
(\1 40
t)
30
0 20
....
Q) 10
.D
0
E -1 0 2 3
;::l
Z
30

20

10

0
-1 0 2 3
log [L(IR)/M(H 2 ) (Lo/Mo)]

Fig. 4. Histograms of the distribution of the ratio L IR / M(H2) for 219 galaxies -
141 barred, and 78 unbarred - with bar classifications in the RC2. The lower filled
histogram represents the unbarred galaxies, and the hatched histogram represents
the barred galaxies (both barred and mixed bar morphology).

ratio within and among hundreds of galaxies. Based on an analysis of 160

Survey galaxies, Young & Knezek (1989) showed that the dominant neu-
tral gas phase in spiral galaxies changes smoothly in the mean along the
Hubble sequence, with the ratio of molecular to atomic gas ranging from
M(H2)/M(H I) ~ 4 ± 2 for the early types spirals, to 0.2 ± 0.1 for the late
type spirals. Analysis of the entire Survey confirms this trend, and recently
published data in the literature further supports this result, as shown in
Figure 5. This set of histograms illustrates the molecular to atomic gas ra-
tio along the Hubble sequence, where the dominant gas phase is molecular
in the early type galaxies, and atomic in the late type galaxies. The top
two panels represent the data for E, SO and Sa galaxies published by Lees
et al. (1991), indicating good agreement with the early-type data from the
Survey. Future work on the M(H 2)/M(H I) ratio will address the variations
in the dominant gas phase with position within galaxies as a function of
type, luminosity, and environment.
302
Ratio of Molecular to Atomic Gas
Along the Hubble Sequence
1,2
ID
6 25 E's
0
*
•
2
0
10

.,.,
8 N
8

~
0
10
8
8

III
(I) 10
'><
til
~
0 .0
..... JO
0 20
10
H
(I)
.D
E 20

~ 10
Z

xe 15 Sd- Sdm

log [M(Hz)/M(HI)]

Fig, 5. Histograms of the distribution of the ratio of molecular to atomic gas mass
as a function of morphological type. The lower 7 panels represent the data from
the FCRAO Extragalactic CO Survey, and the upper 2 panels represent the data
from Lees et al. (1991), indicating good agreement with the data for early-type
galaxies. The filled portion of the histogram represents galaxies with upper limits
to the molecular gas mass. Galaxies with upper limits to the atomic gas mass are
included, with the assumption that the H I mass is given by the 2 0- limit. The dot
in each histogram indicates the mean ratio M (H2) / M (H I) without the H2 limits,
and the cross marks indicate the mean ratio including upper limits.
 303

4. DISCUSSION AND SUMMARY

The results discussed here, with regard to the trends in the star formation
efficiency and the molecular to atomic gas ratio in galaxies along the Hubble
sequence and as a function of environment, provide constraints for theories
of the formation and evolution of galaxies.
For spiral galaxies of a range of morphology, the mean ratio of molecular
to atomic gas decreases by a factor of 10 along the Hubble sequence, while
the mean star formation efficiency does not change. Specifically, the mean
ratio of the mass of molecular to atomic gas decreaf:es along the Hubble
sequence, with more H2 than H I in the early type spirals, and less H2
than H I in the late type spirals (Young & Knezek 1989; Thronson et al.
1989; Wiklind & Henkel 1989). Young & Knezek (1989) also showed that the
mean total gas surface density (molecular plus atomic normalized by optical
area) is only a factor of '" 2 higher in the early type spirals compared to
the late type spirals; thus, the principal effect we find is that the dominant
phase of the gas changes along the Hubble sequence. On the other hand,
the mean rate of high mass star formation, relative to the mass of molecular
gas present, does not vary with spiral morphology (Devereux & Young 1991;
Young, Allen, & Kenney 1993).
These results can be simply understood in terms of the effects of grav-
ity in cloud and star formation, and the size scale on which gravity must
operate. In order for molecular clouds to form from atomic gas, it is not
surprising that gravity on the large scale should be important. This is re-
flected in the HdH I ratio changing with galaxy morphology, or equivalently,
with the mass distribution of the galaxy. In contrast, it appears that star
formation in unbarred, isolated, spiral galaxies occurs with the same mean
efficiency independent of morphological type. This suggests that once the
molecular clouds are formed, star formation within them is a local process,
due to gravity on the small scale. Only when a large-scale perturbation is
present, such as a bar or a tidal encounter with another galaxy, does the
star formation rate per unit mass of molecular gas appear elevated.

Acknow ledgeITlents

I would like to thank the Directors and Staff of the FCRAO and NOAO
for generous allocations of observing time and excellent support during the
CO and Ha Survey observations. I am also grateful to the partipants in the
FCRAO Extragalactic CO Survey for their help observing: Shuding Xie,
Lowell Tacconi-Garman, Linda Tacconi, Steve Schneider, Pete Schloerb,
Steve Lord, Pat Knezek, Jeff Kenney, Nick Devereux, James Case, John
Carpenter, and Lori Allen. Special thanks are due Shuding Xie for helping
make the publication of the FCRAO Extragalactic CO Survey a reality.
304

References
Bloemen, et al. 1986, A&A, 154, 25
de Vaucouleurs, G., de Vaucouleurs, A., & Corwin, H . 1976, Second Reference Catalogue
of Bright Galaxies , (Austin: Univ. of Texas Press) (RC2)
Carpenter, J., Snell, R ., & Schloerb, F .P. 1990, ApJ, 362, 147
Devereux, N . 1989, ApJ, 346, 126
Devereux, N., & Young, J .S. 1990, ApJ, 359, 42
- 1991, ApJ, 371, 515
Joseph, R., & Wright, G . 1985, MNRAS, 214, 87
Larson, R ., & Tinsley, B. 1978, ApJ, 219, 46
Lees, J ., et a1. 1991, ApJ, 379, 177
Lonsdale, C ., Persson, E., & Matthews, K. 1984, ApJ, 287, 95
Noguchi, M ., & Ishibashi, S. 1986, MNRAS, 219, 305
Olsen, K., & Kwan, J . 1990, ApJ, 361, 426
Sanders, D., et aI. 1986, ApJL, 305, L45
Scoville, N., Yun, M., Clemens, D. , Sanders, D. , & Waller, W. 1987, ApJS, 63, 821
Solomon, P., Rivolo, R., Barrett, J ., & Yahil, A. 1987, ApJ, 319, 730
Solomon, P. , & Sage, L . 1988, ApJ, 334, 613
Tinney, C., et aI. 1990, ApJ, 362, 473
Thronson, H ., et aI. 1989, ApJ, 344, 747
Vogel, S., Boulanger, F., & Ball, R~ 1987, ApJL, B316, 243
Wildind, T., & Henkel, C . 1989, A&A, 225, 1
Wilson, C ., & Reid, M . 1991, private communication
Wilson, C ., & Scoville, N . 1989, ApJ, 347, 743
- 1990, ApJ, 363, 435
Young, J ., Allen, L., & Kenney, J. 1993, Ap.J., submitted
Young, J ., & Devereux, N . 1991, ApJ, 373, 414
Young, J., et aI. 1993, in preparation
Young, J., & Knezek, P. 1989, ApJL, 347, L55
Young, J., & Scoville, N.Z. 1991, ARAA, 29, 581
 THE PHYSICAL INTERPRETATION OF MORPHOLOGY
ROSEMARY F. G. WYSE
Department of Physics and Astronomy, The Johns Hopkins University,
Baltimore, MD 21218

Abstract. The morphological type of a galaxy depends on many physical parameters, be-
ing correlated with, for example, environment, star formation rate and angular momentum
content. Which of these correlations is causal is obviously a crucial question. The initial
larger-scale environment determines the density profile around a ga},u:y-mass peak in the
primordial density fluctuation spectrum, and may affect the merger history of the proto-
galaxy. The major-merger rate and the growth of substructure is important for regulating
the star formation rate in the proto-galaxy, since star formation is necessarily inefficient
in potential wells below a critical depth to prevent star-burst driven winds. Star formation
is also inefficient in centrifugally-supported disks, where there is a maximum scale that
is gravitationally unstable. Angular momentum redistribution depends on substructure
mass and density, and on the star formation rate. Dynamical evolution and star formation
rate are related, leaving an imprint via chemical evolution on the stellar populations, and
in particular the pattern of element ratios of stars formed at different times and locations.
The Milky Way Galaxy provides an ideal testing ground for theories of the chemical and
dynamical evolution of galaxies.

1. The Hubble Sequence: Disks and Ellipticals

Various physical properties of galaxies correlate with their morphological

type and, if they reflect initial conditions, may be responsible for the final
appearance of the galaxy. The Hubble Sequence, from ellipticals to late-type
spirals and irregulars, is to a first approximation a sequence of decreasing
spheroid-to-disk ratio (note that the name 'bulge' will be reserved for the
very centrally-concentrated, metal-rich stellar population represented in our
Galaxy by the stars seen by the IRAS and COBE satellites). Thus under-
standing morphology can be reduced to understanding formation, evolution
and survival of disks.
This review will first discuss the differences among galaxies of different
Hubble type, and speculate on their origins. Particular emphasis is placed
on the extreme morphologies of the low surface brightness giant galaxies like
Malin I (section 2). The Milky Way is the typical disk galaxy that can be
studied in most detail, and section 3 contains a brief summary of the stellar
populations in the two most recently recognised morphological components,
the thick disk and central bulge.
Davis & Geller (1976) quantified the clustering behaviors of the differ-
ent Hubble types by a study of the angular correlation functions of spirals
and ellipticals taken separately. Elliptical galaxies clearly favor high-density
regions of the Universe, such as of rich clusters, while disk galaxies are gen-
erally found the field or the outer regions of unrelaxed clusters (cf. Dressler
1980). One can make the immediate deduction from the mere existence of
a thin, centrifugally-supported disk that most of the stars formed after the

305
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution (!l Galaxies, 305-326.
© 1993 Kluwer Academic Publishers.
306

gas cooled and settled into the disk - and that the disk was not disturbed
significantly since the bulk of the stars formed. This points to an explanation
for the morphological type-density relation: disks are fragile and disrupted
in encounters in a cluster environment (Ostriker & Thuan 1975). This is an
enduring concept, emphasised by Larson, Tinsley & Caldwell (1980), and in
the context of hierarchical-clustering theories of galaxy formation by, e.g.,
White & Frenk (1991). The question remains as to whether this suffices to
explain the disk sequence in detail (what are SOs?). Perhaps a more basic
question is whether the initial conditions in a proto-cluster prevent a disk
from forming, or whether evolution destroys disks (nature versus nurty,re).
Important trends along the Hubble sequence include (i) there is a much
narrower stellar age distribution in elliptical galaxies than in disk galaxies,
with the stars in general being old in ellipticals, which implies that ellipti-
cals formed stars early; (ii) there is little cold gas in ellipticals, and a much
larger gas fraction in disks, which implies that ellipticals formed stars with
high efficiency; (iii) angular momentum is less important in massive elliptical
galaxies than in disks. N-body simulations have suggested that all galaxies
started out with the same angular momentum content, as measured by the
parameter A = J E 1 / 2 G- 1 M- 5 / 2 , where J is the total angular momentum,
E is the binding energy, of the system of mass M, a measure of the bal-
ance between rotation and random motions to the figure shape of a system
(Barnes & Efstathiou 1987; Zurek, Quinn, & Salmon 1987). Further, the
value of A that is obtained through tidal torquing in the early Universe is
sufficient to explain the angular momentum content of disk galaxies (Fall &
Efstathiou 1980), so that there must have been angular momentum trans-
port away from the visible parts of ellipticals (c/. Quinn & Zurek 1988); and
(iv) elliptical galaxies favor denser environments.
It is likely that the Hubble Sequence arises from various feedback loops
that connect environment, merger rate, mass of substructure, star formation
rate, star formation efficiency and angular momentum (re )distribution (see
Silk & Wyse 1993 for a more detailed review).
Star formation and merger history are related since sustained star for-
mation requires a sufficiently deep potential well to balance the energy and
momentum output from massive stars and supernovae, allowing recycling of
gas through successive generations of stars. An estimate of this minimum
potential well depth for a system forming stars on a dynamical time can be
found from the condition that supernova remnants cool before they overlap,
and is (Wyse & Silk 1985; Dekel & Silk 1986)
<T3D ~ 100km s-1.

There is now supporting evidence (Meurer et al. 1991; A. Marlowe priv.

comm.) that some dwarf galaxies which are actively star-forming (Blue Com-
pact Dwarfs) are ejecting their interstellar medium in the form of a wind (and
 307

also that some extreme starburst galaxies are blowing 'superwinds'j Heck-
man, Armus & Miley 1990; M . Lehnert priv. comm.). The skewed metallicity
distributions of stars in the couple of dwarf spheroidal satellite galaxies to
the Milky Way in which such measurements are currently possible (Suntzeff
1993) are also suggestive of abrupt truncation of the chemical enrichment
process by a catastrophic event. Thus if galaxies were built by hierarchical
clustering of substructure, such as in Cold Dark Matter dominated cosmolo-
gies, there will not be efficient star formation until this critical potential well
depth is reached.
One could postulate that the stellar halo of disk galaxies formed its stars
in substructure below this threshold, while in elliptical galaxies star forma-
tion were efficient, due to a larger characteristic mass of the substructure.
Dynamical friction between the dense stellar lumps in proto-ellipticals can
then act to transport angular momentum from the lumps to the background
halo, as suggested by Zurek et al. (1987). A larger characteristic mass of sub-
structure for proto-ellipticals has many attractive consequences, apart from
allowing efficient star formation. As an example, let us propose that elliptical
galaxies form from fluctuations on a galaxy mass scale that have a charac-
teristic scale of substructure that is 1/30 of the mass of the galaxy, while
spirals form from galaxy-sized fluctuations that have significantly smaller-
scale substructure, say, 10- 3 Mgal . The Gaussian peaks formalism of Bardeen
et al. (1986) allows evaluation of the rate at which substructure of a given
mass goes non-linear, when embedded in a larger structure (the galaxy).
This rate will be proportional to a function of the star formation rate, with
a smaller constant of proportionality in the case of a proto-spiral (inefficient
star formation due to low mass substructure).
Figure 1 shows the fraction of the substructure that has gone non-linear
in galaxies with the substructure prescribed above, and which for conve-
nience of comparing the relative 'star formation' rates are normalised to
turn around at the same epoch, chosen to be z = 1, adopting a power law
spectrum of density perturbations with P(k) ()( knj n = -2 (and approx-
imating the fluctuation correlation matrix by its value at the origin) . In
this picture, there are some very old stars in spiral galaxies, while the star
formation in proto-elliptical galaxies evolves more rapidly than in spirals.
These effects are in accord with observations. One implication of this model
is that the oldest stars in a spiral galaxy may well be in the central regions,
where the over-density was highest - galaxies form "inside-out", which is a
natural feature of many models. Recent inferences from the metallicity dis-
tribution of RR Lyrae variables in Baade's window compared to that of the
field stars there, interpreted as an age effect, lend support to this viewpoint
(Lee 1992), but more direct determinations of the age distribution of the
central bulge of our Galaxy are as yet unreliable (the central bulge and its
relation to other populations in the Milky Way are discussed in more detail
308

.0

o~ -1
......
+l
()
Cd
.b
~
o
......... -2
star formation

-3
1 .8 .6 .4
log (l+z)

Fig. 1. A schematic model for the star formation rates in proto-spiral and in
proto-elliptical galaxies, where this rate is proportional to the rate at which
sub-structure turns around and collapses. The different Hubble types are assumed
to result from different characteristic scales of substructure. The star formation in
this mode ends when the galaxy scale itself goes non-linear . .As discussed in the
text, the star formation rate in proto-spirals will be inefficient, due to the inability
of the substructure to sustain star formation within its own potential well, while
that in proto-ellipticals will be efficient. Thus there is a very different constant of
proportionality between the non-linear mass in substructure and the mass in stars
in each of these two cases. The left-over gas in the proto-spiral will settle to a disk.

in section 3 below).
Note that this scheme is not intended to describe star formation in the
centrifugally-supported disk that is presumed to form from left-over gas (the
mass scale of substructure in the disk is more probably set by fragmenta-
tion) but indicates the duration of star formation in a 'proto-spheroid'. The
illustrative example here provides for a few gigayear spread in ages of trac-
ers of the stellar halo in a spiral (this is obviously sensitive to the chosen
turnaround redshift) as found for the Milky Way field stars (Schuster &
Nissen 1989) and globular clusters (reviewed by Bolte 1993).
Fragmentation, via gravitational instability, in a centrifugally-supported
disk has a mazimum unstable wavelength (unlike normal, non-rotating,
Jeans' instability), which for a present-day disk corresponds to the mass
scale of a giant molecular cloud complex, well below the above threshold,
 309

so that also here one expects inefficient star formation, as indeed observed.
The steep vertical density gradients in disks compound this, leading to erup-
tion through the disk by 'chimileys' etc. (Norman 1991). Thus disks, once
formed, should form stars slowly, at a fairly constant rate.
The formation of a system which can form stars with high efficiency (des-
ignated a proto-spheroid or proto-elliptical galaxy) necessitates the destruc-
tion of the disk and its organised differential rotation, which in turn requires
the injection of sufficient random energy. Thus one is drawn to the conclusion
that a merger with an approximately equal-mass system - a 'major merger'
- is required to form an elliptical, echoing early work (e.g., Toomre 1977).
Of course the total combined mass must be sufficient for the potential well
to be deeper than the minimum required for efficient star formation. The
spectrum of gravitationally unstable masses after a major merger is then set
by the simple Jeans' criterion, and enhanced angular·momentum transport
drives unconstrained agglomeration of gas clouds within the system to form
a massive dense central gas cloud (Barnes & Hernquist 1992). Star forma-
tion after a major merger then occurs within higher-mass substructure than
in disks, and can be more efficient. An elliptical galaxy may well form as a
result of a binary-hierarchy merging history, while disks result from many
smaller companions being added to a significantly bigger main proto-galaxy.
The environment enters since, almost by definition, equal mass companions
are more likely in a high density region (as are low mass companions, but
their effect is not as significant).
Carlberg (1990a,b) demonstrated that in a universe of global density
parameter n, and density perturbations with power spectrum P(k) <X k-1.5,
the time-dependence of the rate of major mergers can be expressed as

dn
dt <X
(
1 + z )m , with

The steep density dependence arises due to the velocity dependence of the
merging cross-section for galaxies. One can interpret this as meaning that
the rate of 'major mergers' is very sensitive to environment by identifying
proto-cluster regions as those of high effective n in a flat universe, and proto-
void regions as those of low effective n, when smoothed on a suitable scale
(Silk & Wyse 1993; c/. Zurek, Quinn, & Salmon 1988). Elliptical galaxies
would result from the high early major-merger rate in proto-clusters, while
spirals would form in the field.
In this picture of hierarchical clustering spiral galaxies would suffer re-
peated merger events with smaller companion galaxies, which are eventually
accreted. The effect of the mergers on the stellar population of the resulting
final galaxy depends in large part on the gas content of the merging entities,
since gas can radiate energy while stars cannot. Thus the star formation
rate again plays a crucial role. Should a thin stellar disk exist at the epoch
310

of most merging, then with maximal coupling of satellite orbital energy to

the random energies of the disk stars, the disk would be heated such that
(cf. Ostriker 1990)

This fragility of disks is no doubt central to the morphological type - lo-

cal galaxy density relationship. Further, T6th & Ostriker (1992) used the
observed thinness of the stellar disk of our Galaxy to limit the allowable ac-
cretion of stellar satellites since the birth of the Sun to only a few percent of
the mass of the disk. They then argued that this low accretion rate favoured
a low density CDM Universe, since the growth of perturbations would then
be truncated at a redshift of ,...., n- 1 • As discussed in section 3 below, the
thick disk component of our galaxy may be a tracer of the merging history
of the Milky Way.
However, as discussed by Quinn, Hernquist, & Fullager (1993), the disk
heating effect of an incoming satellite is quite uncertain, due in part to
the many parameters - both of the orbit and of the satellite structure. For
example, the orbital energy may rather be transferred to the internal degrees
of freedom of the satellite, puffing it up and allowing it to be tidally shredded
with little damage to the thin disk. Indeed, any gas present will most likely
be simply shock-heated and subsequently radiate away the orbital energy.
Indeed, smoothed particle hydrodynamic cosmological simulations of a CDM
universe by Navarro (1992) suggest that thin disks form despite the merging
inherent in such a model. As discussed in section 3 below, there is a thick
disk component to our Galaxy which may contain the fossil remnants of an
earlier merger between the Milky Way and a satellite galaxy.

2. Giant, Low Surface Brightness Disk Galaxies

Normal spiral galaxies obey 'Freeman's Law', which means that the blue
central disk surface brightnesses are approximately constant, at 21.65±OAO
magnitudes/square arcsec (,...., 170 L0 pc- 2 ; Freeman 1970; van der Kruit
1987). Such a narrow distribution could arise as a result of selection effects,
where the identification of galaxies for study is determined by the relative
brightness compared to sky background: indeed Disney (1976) could re-
produce Freeman's results with this hypothesis (see article by Davies, this
volume).
Van der Kruit (1987) repeated Freeman's analysis but with a statisti-
cally complete sample and confirmed his earlier result, for morphological
types SO-Sc. The narrow range of central surface brightnesses is not then
due to selection biases. The later Hubble types are mostly low luminosity
irregular galaxies and have central regions about a magnitude fainter, with
blue surface brightnesses of about 22.5 magnitudes/square arcsec. Phillipps
 311

et al. (1987) also concluded that 'the normal field galaxy population' has ap-
proximately constant surface brightness, and confirmed that dwarf galaxies,
common in clusters, tend to be of lower surface brightness.
Jura (1980) argued that the apparent constancy of surface brightnesses
arose because disk galaxies were optically thick in the blue. This has been
echoed recently by Valentijn (1990). However, it is hard to see how one could,
for example, observe galaxies behind others, or avoid substantial differences
in the optical and radio rotation curves if this were the case. Indeed, Bothun
& Rogers (1992) analyse far-infrared (60J-Lm and 100J-Lm) surface photometry
of a sample of nearby disk galaxies, and find that the optical and infrared
scalelengths are equal, implying that disks are optically thin.
Nevertheless, it is clear that some low surface brightness, but high mass,
disk galaxies could exist which would be missed by normal identification
techniques - "crouching giants" (cf. Disney 1976). As discussed by Davies
(this volume), searches for low surface brightness galaxies are now being
successful. I will here focus on the subset of massive, low surface brightness
galaxies that have normal, evolved central bulges, despite having unevolved,
low surface brightness disks.
Several such galaxies have been identified by a photographic enhancement
technique due to D. Malin and two have been studied in some detail -
Malin I, which was discovered serendipitously (Bothun, Impey, Malin, &
Mould 1987), and Malin II (Bothun, Schombert, Jmpey, & Schneider 1990).
Although the disk of Malin I has a high total luminosity, its central surface
brightness is some 4 magnitudes fainter than the 21.5 B-mag/ arc-sec 2 of
normal galaxies, and the scale-length is '" 55 kpc, a factor of ten larger
than typical disks (van der Kruit 1990). The total HI mass is '" 1011 M 0 ,
a factor of ten more than a typical spiral galaxy, but with a mean column
density of only 2 M0 pc- 2 (compared with", 10 M0 pc- 2 of gas and '"
35 M0 pc- 2 of stars in the solar neighborhood). As emphasized by Impey
& Bothun (1989), this column density is reminiscent of the damped Lya
systems seen in absorption in quasar spectra at high redshift and which
have been interpreted as the earliest stages of disk formation (Wolfe 1989),
and is only of order the gas surface density that is believed to be required for
local gravitational instability and star formation (c/. Quirk 1971; Kennicutt
1989). In contrast, the available spectroscopic and photometric data (Impey
& Bothun 1989) imply that the central regions of the bulge of Malin I are
composed of the same stellar population as normal evolved central bulges -
probably predominantly old and metal-rich as inferred for the Milky Way
(FrogeI1990). Indeed, the central bulge is normal in all its properties, while
the disk is extremely unevolved. Malin II has similar properties.
Hoffman, Silk, & Wyse (1992) demonstrated that the initial structure of
galaxies can be strongly affected by their larger-scale environments. In par-
ticular, rare ('" 30") massive galaxies in under-dense regions will have normal
312

central bulges, but unevolved, extended disks, which are just the characteris-
tics of Malin I and Malin II. Their model, which identifies 'crouching giants'
with 30" density peaks on a galaxy scale which are embedded in -30" density
troughs on a larger cluster/void scale, predicted that searches for more ex-
amples of 'crouching giants' should be fruitful, but that such galaxies do not
provide a substantial fraction of mass in the universe. The rare, '" 30", mas-
sive galaxies in clusters do nothave extended low surface density envelopes,
and are identified with high surface brightness ellipticals. The identifica-
tion of dwarf galaxies with rv 10" fluctuations suggests that the structure
of dwarf galaxies is relatively unaffected by their environment, so that low
surface brightness dwarfs should exist both in clusters and in the field. Their
results are shown in Figure 2. The adopted power spectrum of density per-
turbations is that of a CDM dominated cosmology. Galaxies are assumed to
be seeded by local density maxima on the scale of Ms = 1010 h- 1 M0 (in the
sense of Gaussian smoothing), while the background environment is deter-
mined by the amplitude of structure on the comoving scale corresponding to
the mass of a small cluster of galaxies, Mb = 5 X 10 14 h- 1 M 0 , which is also
the Lagrangian scale of voids. The typical local maximum for most power
spectra has v '" 2 (Peacock & Heavens 1985), and here Va = 3 and Vs = 1
have been adopted for high and low peaks respectively, to bracket the range
of amplitudes over which the formation of bound structure occurs. The 30"a
peaks will represent rare galaxies. The background amplitudes-are allowed
to vary over the entire range, -3 ::; Vb ::; +3. In all cases the size of the
object is defined as the radius enclosing half of the coherent -mass associated
with the peak.
The statistics of Gaussian random fields, applied to the density fluctua-
tions, gives the linear density profiles around local maxima. The virialized
mass distribution that results after non-linear evolution and collapse under
self gravity can be calculated from the 'secondary infall' paradigm (e.g.,
Gunn & Gott 1972) whereby successive shells of material are assumed to
be accreted smoothly by the mass already at the center, with no crossing
of shell orbits. The projected surface density, ~(r) '" M(r)/7rr 2 , can be cal-
culated immediately from the virialized mass distribution. It is clear from
Figure 2 that when Vs is fixed but the amplitude of the background varies,
the inner regions of the resulting objects have a very similar structure, in-
dependent of the environment. Hoffman, Silk, & Wyse (1992) proposed that
these regions be identified with galactic central bulges. For low peaks, the
mass coherence does not extend far away from the peak and is limited to
length scales such that the background field has a negligible effect on the
density profile. Note that for the structure seeded by the high, 30"s, peaks,
Figure 2(a), the coherence extends out to", 10 13 h- 1 M 0 . On this large scale
('" 1h- 1 Mpc in low density regions) the primordial overdensity is affected
by the background, resulting in a marked dependence of the final virialized
Fig. 2. (a) Projected surface density as a function of enclosed mass, out to the ra-
dius where the density perturbations runs into the noise, for galaxy-scale peaks with
V, = 1, assuming a flat CDM dominated Universe, for various assumed amplitudes
of fluctuations on a larger scale, Vb = -3, -2, -1,0,1,2,3. The curves correspond-
ing to different background amplitudes are all very similar.
(b) : Projected surface density, as in (a), but for high galaxy-scale peaks, with
v, = 3. In contrast to the v, = 1 case, here the profiles are rather sensitive to the
background amplitude and galaxy-scale peaks are more extended in low density
environments.
(c) : Local cooling times (solid lines) and dynamical times (dot-dashed lines) for the
10- peaks, in a variety of backgrounds, Vb = -3, -2, ... , 3. Dynamical and cooling
times increase as Vb decreases. All of the perturbation can cool efficiently, indepen-
dent of environment.
(d) : Local cooling and dynamical times, as in (c), but for the 30- galaxy-scale peaks.
Efficient cooling requires that the cooling time be shorter than the dynamical time,
and this only occurs for the inner regions.
314

structure on the environment.

The predicted structure of the luminous components of the galaxies which
form as described above depends on the rates of cooling and of star forma-
tion, and has been described in terms of the ratio of the cooling time to the
dynamical time (Rees & Ostriker 1977; Silk 1977). Assuming that the initial
gas density is simply one-tenth of the total density, that the gas at a given
radius is initially heated to the virial temperature of the system out to that
radius, and adopting a cooling function for primordial gas produces the run
of cooling times shown in Figure 2( c) and (d). Efficient cooling, in the sense
that the cooling time is shorter than the dynamical time, occurs throughout
galaxies seeded by 10' peaks, independently of environment. The correspond-
ing plot for the 30' galaxies is shown in Figure 2( d) ; here the cooling time is
shorter than the dynamical time only in the central regions, which we have
proposed be identified as the protobulge. A larger fraction of the coherent
mass can cool efficiently in a cluster environment than in a void. The mate-
rial which cools on a Hubble time rather than a dynamical time will provide
continuous infall, which may be associated with disk formation.
Ongoing surveys have provided several more examples of giant, but low
surface brightness galaxies (e.g., Schombert, Bothun, Schneider, & McGaugh
1992), and the sample is now large enough that preliminary statistical anal-
yses can be done. These galaxies indeed occupy low density regions of the
Universe, lacking close companions (Bothun et al. 1992; Knezek & Schneider
1992; Zaritsky and Lorrimer 1992). The region of low density surrounding
them is smaller than that used in the published calculations of Hoffman et
al., which means that the effect of the environment is even stronger than
that demonstrated (since 'crosstalk' is stronger if the two scales are closer to-
gether). However, low surface brightness giant galaxies with evolved central
bulges are a small subset of this sample, and cannot be analysed statistically.
The temporal evolution of the star formation rate in disk galaxies may
self-regulate, giving a value of Toomre's Q -parameter than remains close to
unity (c/. Silk 1993). One can imagine both positive and negative feedback
loops (e.g ., Dopita 1987), with a threshold surface gas density (Kennicutt
1989) or combination of density and time (Wyse & Silk 1989). Tidal inter-
actions may be important in driving star formation (e.g., Larson & Tinsley
1978; Lacey & Silk 1991), rather than internal processes, perhaps contribut-
ing to the slow evolution of isolated galaxies such as Malin I. However, the
normal central bulge of this and other galaxies, which presumably formed
by rapid and efficient star formation, suggests that initial structure may be
more important. In the context of the model suggested in the previous sec-
tion, the isolation of Malin I has allowed the extended disk to be sustained,
while the low density background allows the tracing of the galaxy out to
very large radii, without confusion with the density profiles around of other
similarly sized peaks (which would provide the major-mergers).
 315

3. The Milky Way as a Galaxy

The chemical evolution and the dynamical evolution of the Galaxy are re-
lated since, as discussed above, the efficiency of star formation depends on
the depth of the local potential well of the star-forming site. One is able
to constrain the dynamical history of the Milky Way by studying the old
stars which are tracers of the early epochs of Galaxy formation (c/. Eggen,
Lynden-Bell, & Sandage 1962). The stellar populations of the central bulge
and thick disk morphological components of the Galaxy are discussed as
illustrations.

3.1. STAR COUNTS

The morphology of the Milky Way, at least at high latitudes, may be de-
lineated by the classical technique of counting stars. The analyses of star-
count data depends on the "Fundamental Equation of Stellar Statistics",
which relates the number of stars, N, countable in a given solid angle to a
given magnitude limit, m , to the volume integral of the product of the stel-
lar luminosity function, ~(Mv, x) , and the stellar space density distribution,

f ~(Mv,x)D(Mv,x)d3x.
D(Mv,x),
N(m) =
The application of computer modelling to such datasets by van den Bergh
(1979) led to a considerable resurgence of interest, continuing to the present.
In general, the luminosity function is too broad to allow any solution for
both D(Mv, x) and ~(Mv, x). The situation can be eased by restricting the
data by color and/or spectral type, which is the technique usually followed.
In this case, for an assumed form of the distribution function ~(Mv, x) the
density function D(Mv,x) may be recovered from N(m, color) . This may
be done by inverting the data - the technique of photometric parallax - or
computer calculation, with subsequent comparison of data and model. These
techniques are clearly entirely equivalent, and should agree. They often do
not .
The fundamental problem with use of the fundamental equation is that
both the stellar luminosity function and the stellar density law are func-
tions of many parameters. Few of these are sufficiently well known to be
fixed. Consequently, a wide variety of combinations of ~ and D are allowed
mathematically. Other astrophysical constraints are necessary, whose choice
remains subjective.
The availability of high-efficiency, linear, 2-dimensional detectors (CCDs)
and fast automated photographic plate scanning microdensitometers (PDS,
COSMOS, APM) has revolutionised stellar statistics. Complete samples of
stars can be measured to useful precision in several wavebands over suffi-
ciently large areas of sky that random errors due to counting statistics are
316

unimportant. The minimisation of systematic errors still requires an enor-

mous effort, however (Gilmore 1984).
The general features of the high latitude sky are illustrated by the V /B - V
SGP data from Gilmore, Reid, & Hewett (1985). The important aspects of
these data are:

a) There is a sharp edge to the distribution near B - V = 0.4, with very

few stars being seen significantly bluer than this limit. This corresponds to
the main sequence turnoff color of an old, metal poor population, and has
two important consequences. The first is the absence of a younger turnoff,
showing that no substantial continuing star formation has taken place in
the spheroid. The second is the apparently small number of blue horizontal
branch stars. This may mean that the metal poor field spheroidal stars show
the "second parameter" effect, and have a red horizontal branch.

b) The distribution has two local maxima, at B - V "" 0.6 and a red-
der peak at B - V 1.5. The bluer peak is similar to the main sequence
f'V

turnoff color of metal-rich globular clusters, and corresponds to the thick

disk, discussed further below. The redder peak of the distribution contains
cool main sequence stars, and the pile-up at this color reflects the insensi-
tivity of B - V color to effective temperature in such stars. It is therefore
an artefact of the choice of photometric passbands, rather than evidence of
a structural property of the Galaxy.
Direct solution of the fundamental calculation of stellar statistics is a
straightforward computational exercise. Consequently, many attempts have
been made recently to explore parameter. space so that the uniqueness of
the results from the direct analysis of star count data can be determined. In
relevant form this equation is

N(V,B- V) =w J
+(Mv , [~] ,r,x, ... )D(r,Mv ,r)r2 dr.
The luminosity function (stars mag- 1 pc- 3 ) has been known for many
years to be a function of distance from the Galactic plane (e.g., Bok &
MacRae 1941). Similarly, the existence of age-velocity dispersion and age-
metallicity relations for old thin disk stars is well known (e.g., Stromgren
1987). This emphasises the crucial and irreducible limitation of analyses of
this type - both the luminosity function and density law are functions of
the other phase space parameters. A unique solution of the above equation
is therefore impossible. Instead, a large number of parameters must be fixed
on external astrophysical grounds. This illustrates the fundamentallimita-
tion of such modelling, which is a severe restriction on its value - too few
constraints are usually available to provide a unique model, and too few
consistency arguments are usually applied during its use.
 317

Photometric parallax provides the most straightforward analysis tech-

nique for stellar number-magnitude-color data. This involves use of the color-
absolute magnitude relation for a galactic or globular cluster of appropriate
chemical abundance. The absolute magnitude is read directly from this dia-
gram, at given color, and combined with the apparent magnitude to give a
photometrically-defined distance. A stellar density law is derivable directly
from a large set of distances, after appropriate corrections for Malmquist
bias (perhaps more familiar now in determinations of the Hubble constant,
but of course first identified in the selection of stellar samples; see Stobie,
Ishida, & Peacock 1989 for a recent discussion).
This technique allowed the identification of the thick disk as a distinct
structural component of the Milky Way Galaxy by Gilmore & Reid (1983),
from analysis of the stellar density profile at the SGP derived from a sample
of", 12,500 stars to I = 18 in 18.25 square degrees. The steep density profile
from 2 kpc to 4 kpc was identified by these authors as a Galactic thick disk,
with exponential scale height '" 1.3 kpc and local normalisation"" 2 percent
of the old disk stars. They emphasised that a flattened R 1 / 4 law was an
equally good fit to the data. Indeed, they suggested that the thick disk was
simply that part of the well-studied Rl/4 stellar halo that is responding to
the locally-dominant potential of the thin disk.
While the structural characteristics of the thick disk found by Gilmore
& Reid (1983) have been confirmed by various other star-count analyses in
several fields at both high and low Galactic latitudes (e.g., Yoshii, Ishida, &
Stobie 1987; Fenkart' 1989; Yamagata & Yoshii 1992) and by spectroscopic
studies (so that a better estimate of metallicity can be used; Kuijken &
Gilmore 1989), the metallicity and kinematic properties of thick disk stars
are not consistent with the view that the thick disk is part of the R 1 / 4
halo. This illustrates the limitation of star count analyses used in isolation.
Chemical and kinematic data are also required to make firm conclusions
about stellar populations, and inferences about Galaxy formation.

3.2 . THE THICK DISK

The characteristics of the thick disk at the solar neighborhood are reasonably
well defined, but the global properties of this stellar population are not.
Larger datasets are needed to decide the evolutionary status of the thick
disk: current data are such that it is difficult to distinguish, with statistical
confidence, between either a discrete thick disk or a continuous sequence of
disks (e.g ., Gilmore, Wyse, & Kuijken 1989; their Fig. 6a). Locally it can
be described by a predominantly old population (Gilmore & Wyse 1987;
Carney et ai. 1990; Schuster & Nissen 1989), with mean metallicity about
one-third that of the Sun, and a significant dispersion in metallicity, of ""
0.3 dex (Gilmore & Wyse 1985; Laird et ai. 1988; Friel 1987; Morrison,
318

Flynn, & Freeman 1990). It is a kinematically 'warm' component, with a

lag behind the Sun in rotational velocity (the asymmetric drift) in the range
35 - 90 km S-l (e.g., Wyse & Gilmore 1986; Ratnatunga & Freeman 1989),
perhaps dependent on height above the plane (Murray 1986; Majewski 1992).
The most recent determination of the asymmetric drift of the thick disk is
from a very large proper motion survey of the NGP (Soubiran 1992) from
Schmidt plates, complete to V = 17, covering 7 square degrees, with a
baseline of 40 years and 4400 proper motions accurate to 2 milliarcsec/yr.
The kinematics of the thick disk stars in this survey are most consistent
with a constant lag behind circular velocity of amplitude,....., 50 km s-1, out
to z = 2.5 kpc. The scaleheight of the thick disk is a factor of3-4larger than
that of the old thin disk, with a vertical velocity dispersion of,....., 50km s-l.
The importance of this is discussed below. The radial scalelength of the thick
disk can also be an useful discriminant for formation scenarios (see below)
but is poorly determined due to the small size of published samples far from
the Sun. The extant data are consistent with a scalelength equal to that of
the thin disk (Ratnatunga & Freeman 1989).
Heating will increase the scaleheight of the disk, and leads to the specu-
lation that a thick disk could be a tracer of the merging history of a galaxy.
Quinn, Hernquist, & and Fullager (1993) have carried out the most com-
plete set of simulations of stellar mergers between a spherical satellite and
a disk galaxy. They find that a generic heated disk will indeed have larger
velocity dispersions in each direction (vertical, radial and azimuthal), and
will have a lower rotational velocity than the initial disk configuration. Not
only is the vertical scalelength increased, but the radial scalelength is too.
It should be noted that the asymmetric drift - the lag behind circular veloc-
ity - depends on both the scalelength and amplitude of random motions, in
such a way that an increase in radial scalength offsets an increase in radial
velocity dispersion. Quinn et al. offer a 'standard' model which produces
a thick disk with properties very similar to .those of our Galaxy but with
radial scalelength a factor of ~ 1.5 greater than that of the thin disk. This
standard model is a merger with a satellite of mass 10% of that of the thin
disk, initially on a prograde circular orbit , inclined at 30°.
A somewhat related scenario for formation of the thick disk, in that it too
involves the relaxation into virial equilibrium of the Galaxy, was proposed
by Jones & Wyse (1983). This latter model appealed to star formation in
transient 'pancaking' in the early stages of collapse of gas in a non-spherical
dark halo. This disk will be thin initially, but fattens by violent relaxation
as the potential equilibrates. In each case, the present thin disk forms after
dynamical evolution ends, with the thick disk older and more metal-poor.
As demonstrated by White (1980), in the context of forming an elliptical
galaxy by the major merger of two disks, and by Quinn et al. explicitly
for the case of a thick disk, merging and violent relaxation do not erase a
 319

pre-existing metallicity gradient. Thus the presence of such a gradient in the

thick disk does it not imply that it formed by a predominantly dissipative
process.
Star formation during the dissipative settling of gas to the plane of a
steady potential can also provide a thick disk, perhaps smoothly connected
to the thin disk (Larson 1974; Norris 1987). A gradient in asymmetric drift,
as found by Murray (1986) and by Majewski (1992), may favor a continuum
of properties from thick to thin disk. The metallicity of the thick disk is
such that one might expect enhanced cooling and presumably star forma-
tion once this enrichment is reached, since then line radiation takes over from
bremsstrahlung as the major cooling mechanism from temperatures typical
of the virial temperatures of galaxies. Indeed, Burkert, Truran, & Hensler
(1992) analyse the 'chemodynamics' of disk collapse and find that a short-
lived phase of enhanced star formation may occur, driven by early super-
novae, leading to a distinct thick disk. Their calculation is one-dimensional
and thus the role of angular momentum is ignored.
It is difficult to distinguish a dissipationless heating origin for the thick
disk from a dissipational settling origin without a more detailed understand-
ing of the star formation rate and of the possible importance of angular
momentum transport.
One can, however, rule out two scenarios for the formation of the thick
disk, both involving scattering of thin disk stars. The thin disk stars show
a well-defined increase of random motions with age which can (mostly!) be
understood by scattering by potential fluctuations in the disk - giant molec-
ular clouds and bits of transient spiral structure (e.g., reviewed by Lacey
1991). This increase saturates well below the random motions characteristic
of the thick disk, essentially due to the fact that the scatterers are confined
to the plane, and lose influence once a star spends little time there. Thus
the thick disk cannot be due to this mechanism. Rare, close encounters with
high-velocity scatterers, such as may be expected if the dark halo consisted
of massive black holes, could heat a small fraction, '" 1 %, of the thin disk to
a velocity dispersion more typical ofthe thick disk (Lacey & Ostriker 1985).
However, this mechanism cannot provide an explanation for the thick disk,
since one would then expect the thick disk to be a random sample of the
thin disk, which it manifestly is not, having for example, a very different
metallicity distribution.
Understanding of the relationship between the thick disk and the other
stellar components of the Galaxy - the R 1 / 4 halo, central bulge and thin disk
- would be greatly facilitated by age and metallicity distributions for all. For
example, it is well known that the thin disk did not form stars as a closed
box from zero metallicity (the G-dwarf problem), and inflow of pre-enriched
gas from the R 1 / 4 halo and thick disk may be required (cf. Hartwick 1976;
Gilmore & Wyse 1986). However, the halo angular momentum distribution
320

suggests rather that it pre-enriched the bulge (Carney 1990; Wyse & Gilmore
1992) which leaves open the question of what pre-enriched the thick disk.
Late infall of primordial material offers another solution to the G-dwarf
problem, essentially since it skews the normalisations and hence the expected
number of oldest low metallicity stars (e.g., Audouze & Tinsley 1976). Infall
has also been appealed to to sustain star formation in some external galaxies,
where the detected gas is inferred to last only a fraction of a Hubble time
(e.g., Kennicutt 1983), and to help provide a suitably cold component for
long-lived spiral structure (e.g., Sellwood & Carlberg 1984). However, there
is no observational evidence to support the smooth infall usually assumed,
but rather the available observations favor somewhat disruptive accretion
of gas-rich companion galaxies (Sancisi 1990). A comprehensive survey for
thick disks in external galaxies is clearly required. Alternatively, the source
of the metal-poor gas may be the unevolved, outer regions of the disk. Radial
gas flows, driven by viscosity, are seen in hydrodynamic simulations of disk
formation (e.g., Katz & Gunn 1991), and occur due to other processes such
the passage of a spiral density wave. Clarke (1991) showed that the G-
dwarf distribution in the solar neighborhood could be explained provided
there were an edge to star formation in the disk, such as suggested by recent
observations (Kennicutt 1989 and this volume), and gas flowed inwards from
beyond this threshold radius.

3.3. THE BULGE

The observation that the mean iron abundance of the stellar halo is well
below the theoretical yield for a solar neighborhood IMF is most easily ex-
plained by gaseous outflow, truncating chemical evolution (Hartwick 1976;
sudden, rather than steady, outflow might be favored by the a.pparent asym-
metry in the halo metallicity distribution; Laird 1993). The angular mo-
mentum distributions of the present stellar components of the Galaxy offer
constraints on formation scenarios. Available data suggest that the R 1 / 4 halo
and central bulge have very similar angular momentum distributions, and it
is easy to envisage that gas ejected from the halo would flow to the central
regions to form the proto-bulge. The thick disk has a significantly higher
angular momentum content than the R 1 / 4 halo, but the relevant kinematic
and structural parameters are uncertain (Wyse & Gilmore 1992, their figure
1 ).
Several lines of evidence suggest that the stellar population of the inner
one kiloparsec or so of the Milky Way - hereafter called 'the bulge', perhaps
displayed most graphically in the COBE image (Hauser et al. 1990) - is
not smoothly related to that of the R 1 / 4 spheroid, represented at the solar
neighborhood by the high-velocity sub dwarfs.
The stellar density profile of the bulge is steeper than that of the Rl/ 4
 321

halo, with a vertical half-light radius of about 200 pc (Frogel et al. 1990),
an order of magnitude smaller than the value - ,...., 3 kpc - for the canonical
stellar halo. The bulge is flattened, with an axial ratio of ,...., 0.6, and its
shape is apparently supported by rotation (e.g., Minnitti et al. 1992) rather
than completely by random motions, which are believed responsible for the
shape of the R 1 / 4 halo (Wyse & Gilmore 1989). The bulge may indeed be
a rapidly-rotating, triaxial 'peanut' (Binney et al. 1991; see also Blitz &
Sperge11991).
The bulge is metal-rich, with representative K-giant samples having a
mean metallicity of one-and-a-half times solar (Rich 1988), and late M-
giants having a (more uncertain) mean metallicity of two-three times solar
(Terndrup, Frogel, & Whitford 1990; Sharples, Walker, & Cropper 1990). A
gradient in mean chemical abundance within the bulge has been detected
in a sample of M-giants, with the suggestion of perhaps a factor of three
decline -in iron abundance between fields at latitude b = - 30 to b = -12 0 ,
although there is poor agreement between optical and infrared data (Frogel
1990). There is no evidence from available spheroid samples for a metallicity
gradient of an amplitude sufficient to allow a smooth transition between R 1 / 4
halo and bulge. Further, there are no metallicity data on suitable scales for
external disk galaxies.
Some fraction of at least the luminous, red stars in the bulge is signifi-
cantly younger than the local R 1 / 4 halo - as found in analyses of the period
distribution of variable stars observed with IRAS (Harmon & Gilmore 1988;
Whitelock, Feast, & Catchpole 1991) and from photometry and spectroscopy
of M-giants (Sharples et al. 1990). There is continuing uncertainty however,
due to the lack of solid understanding of mass-loss processes on the giant
branch and Asymptotic Giant Branch, and of good stellar atmospheres for
'super-metal-rich', cool stars. The bulk of the stars in the bulge probably
has ages ~ 5 Gyr, but the ages are sufficiently uncertain that the bulge may
even be the same age as the metal-poor spheroid, ,...., 15 Gyr (Terndrup 1988)
- which translates into a plausible range of formation redshifts for the bulge
of 0.7 ~ zJorm ~ 7.
Wyse & Gilmore (1992) investigated a variety of models for the bulge,
and concluded that the best diagnostic of the source of the proto-bulge gas,
and of the star formation history of the bulge, is element ratio data for the
entire range of iron abundances in the bulge.
Element ratios are very important for understanding galactic evolution,
in part since different elements come from different sources, and are ejected
into the interstellar medium on different timescales (Tinsley 1979). For ex-
ample, oxygen is predominantly synthesized in short-lived, massive stars
that explode in Type II supernovae, while iron has an additional important
source in Type Ia supernovae. The best current model for these latter su-
pernovae invokes the eventual merging, driven by gravitational radiation,
322

of a pair of white dwarfs, with total mass greater than the Chandrasekhar
mass (Then & Tutukov 1987). Thus the explosion timescale is set not by
intrinsic stellar parameters, but rather by orbital parameters, and can be
a Hubble time (note that such time-delayed supernovae pose a potential
problem for advocates of white dwarfs as the dark matter in the haloes of
galaxies, unless one suppresses the binary fraction of the progenitor stars
by several orders of magnitude below that of even the old, metal-poor stars
in our Galaxy; Smecker & Wyse 1991). Detailed models suggest that half
of the possible progenitors (white dwarf binaries of suitable mass) from a
single burst of star formation explode in '" 109 yr (e.g., Smecker-Hane &
Wyse 1992). Current understanding of supernova rates and nucleosynthe-
sis implies that for a solar neighborhood IMF, Type II supernovae provide
a yield of about ~ Fe0' while Type la provide a yield of ;;:. Fe0 (c/. Rana
1991). Thus a stellar population whose mean iron abundance is above solar
metallicity either self-enriched sufficiently slowly that Type la supernovae
contributed fully, or had an IMF very biased towards massive stars.
The pattern of rOlFe] versus [Fe/H] for halo stars in our Galaxy - ,a
plateau at low [Fe/H] with a turndown at higher [Fe/H] (e.g., Wheeler,
Sneden, & Truran 1989) - is easily understood with the identification of a
long-lived source of iron (Tinsley 1979). It is very important to pin down, ob-
servationally, the iron abundance at which the element ratios reflect enrich-
ment by Type 1<,\ supernovae, since this allows identification of a timescale,
'" IGyr, for the formation of the halo that is represented in the sample (e.g.,
Matteucci 1987; Wyse & Gilmore 1988; Truran 1987). There is no detected
intrinsic scatter in the element ratios as a function of [Fe/H], which contrasts
strongly with the large intrinsic scatter in the age-iron abundance relation,
and places constraints on the star formation process (see Wyse & Gilmore
1993).
Models for the bulge can be characterised by (a) the source of the proto-
bulge gas (it is not likely that the bulge formed from accreted stellar satel-
lites since all those known are too metal-poor ct. Rich 1989). Since the
mean metallicity of the bulge - significantly higher than solar (Rich 1988)
- is much higher than plausible initial metallicities from any of the possible
sources (primordial gas, halo, thick disk or thin disk) the identification of
the source is relevant for only the early stages of bulge evolution; (b) the rate
of accumulation of this gas; (c) the rate of star formation in the bulge. The
important timescale relative to which these rates are scaled is the timescale
for the enrichment by Type la supernovae, since this produces observable
signatures in the element ratios (Wyse & Gilmore 1992). If the proto-bulge
material accumulated, and formed stars, more rapidly than this (of order a
Gyr), then only Type II supernovae are available for chemical enrichment. In
this case, the high mean metallicity of the bulge population requires an IMF
biased towards massive stars, when compared to that of the solar neighbor-
 323

hood. The IMF can be biased in one of two ways - either by modifying the
slope (c/. Matteucci & Brocato 1990), or by truncation. Truncation of the
IMF is limited by the requirement of forming sufficient low-mass stars to
populate the main sequence as observed in color-magnitude diagrams (cf.
Terndrup 1988). A conservative lower limit is then 0.5 M 0 . As tabulated
by Sandage (1986), a variation in lower mass limit from 0.1 M0 to 0.5 M0
decreases the locked-up fraction by at most a factor of 1.3. This is not suf-
ficient to increase the Type II yield ('" 0.25 of the solar iron abundance for
a normal IMF) high enough to provide the iron in bulge stars (mean well
above solar). The more plausible way to increase the nucleosynthetic yield
is to flatten the slope of the IMF, which provides a clear observational sig-
nature by predicting a plateau in [0 /Fe] at some value significantly higher
than that of the R 1 / 4 halo stars (assuming of course that the R 1 / 4 halo stars
formed with a normal IMF; Wyse & Gilmore 1992). Iron will eventually be
synthesised in Type I supernova and ejected into the bulge, releasing about
the same amount of iron as ejected by Type II supernovae, for a reasonable
range of IMFs and models of Type I supernovae (the Type I supernovae
from the R 1 / 4 halo background population that is also present are merely
perturbations to the present model, due to the much larger mass, '" x 10,
of the bulge) . This gas must be accounted for in the model. Such iron-rich
gas, which is not incorporated in the main bulge population, is produced in
any model for the bulge which has sufficiently rapid star formation that it
is essentially complete prior to Type I explosions. Should this gas sustain a
low level of star formation, then one may expect to find extremely iron-rich
stars which have very low oxygen abundances (c/. Gilmore & Wyse 1991).
If the star formation were slow, following rapid accumulation, allowing
the incorporation of the ejecta from Type Ia supernovae, then the IMF can
be normal, producing a pattern of element ratios like that seen in the solar
neighborhood, with a downturn at higher iron abundances (the location of
the downturn is set by the star formation rate).
If the accumulation of proto-bulge gas were due to the outflow of gas from
slow, chaotic halo formation, then the possibility of incorporation of iron
from Type Ia supernovae prior to star formation in the bulge can produce
stars with [Fe/H] ~ - 1 but with rOlFe] '" 0 (rather than the enhanced
oxygen seen in the solar neighborhood at this iron abundance).
The preliminary element ratios for bulge stars in situ presented by Rich
(1993) are at present limited in usefulness by small number statistics, but
this technique should provide unique insight into the history of the bulge.

Acknowledgements
I acknowledge partial support from NASA (grant NAGW-2892), from the
NSF (grant AST 90-16226) and from the Alfred P. Sloan Foundation.
324

References
Audouze, J. & Tinsley, B.M.: 1976, Ann. Rev. A6tr. Ap. 14, 43
Bardeen, J.M., Bond, J.R., Kaiser, N., & Szalay, A.S.: 1986, A6trophys. J . 304, 15
Barnes, J., & Efstathiou, G.P.: 1987, A6trophY6. J. 319, 575
Barnes, J . & Hernquist, L. : 1992, Ann. Rev. A6tr. Ap. 30, 705
Binney, J., Gerhard, O.E., Stark, A.A., Bally, J ., & Uchida, K.I.: 1991, M.N.R.A.S. 252,
210
Blitz, L. & Spergel, D.: 1991, AstrophY$. J. 379,631
Bok, B.J. & MacRae, D.A.: 1941, Ann. New York Acad. Sci. 42, (Art 2) 219
Bolte, M.: 1993, in J . Brodie & G . Smith, ed(s)., Globular Clu6ters in the Conte:et o/their
Parent Gala:eie6, ASP Conference Series, in press
Bothun, G.D ., Impey, C.D., Malin, D., & Mould, J. : 1987, A6tr. J. 94,23
Bothun, G .D. & Rogers, C.: 1992, A6tr. J. 103, 1484
Bothun, G.D., Schombert, J.M., Impey, C.D. & Schneider, S.E.: 1990, A6trophY6. J. 360,
427
Bothun, G.D., Schombert, J .M., Impey, C.D., Sprayberry, D., & McGaugh, S.S.: 1992,
A6trophys. J. , submitted
Burkert, A., Truran, J .W., & Hensler, G .: 1992, A6trophys. J. 391, 651
Carlberg, R.G.: 1990a, A6trophY6. J. 350, 505
Carlberg, R.G.: 1990b, A6trophY6. J. (Lett.) 359, Ll
Carney, B.W.:J990, in B. Jarvis & D. Terndrup, ed(s)., Bulge6 0/ Galaries, ESO , Garching,
p26
Carney, B.W., Latham, D.W., & Laird, J.B .: 1990, A6tr. J. 99, 572
Clarke, C.J.: 1991, M.N.R.A.S. 229,704
Davis, M . & Geller, M .J.: 1976, A6trophY6. J . 208, 13
Dekel, A. & Silk, J. : 1986, A6trophY6. J. 303, 39
Disney, M.J.: 1976, Nature 263, 573
Dopita, M. : 1987, in S.M. Faber, ed(s)., Nearly Normal Gala:eiu, Springer-Verlag, New
York, p144
Dressler, A .: 1980", A6trophys. J . 236, 351
Eggen, O.J., Lynden-Bell, D ., & Sandage, A.: 1962, A6trophY6. J. 136,748
Fall, S.M. & Efstathiou, G. : 1980, M.N.R.A.S. 193, 189
Fenkart, R.: 1989, A6tr. Ap. Suppl. 78, 217
Freeman, K .C .: 1970, A6trophY6. J. 160, 811
Friel, E .: 1987, A6tr. J. 93, 1388
Frogel, J.: 1990, in B. Jarvis & D. Terndrup, ed(s)., Bulge6 0/ Gala:eie6, ESO, Garching,
p180
Frogel, J.A., Terndrup, D.M. , Blanco, V.M., & Whitford, A .E .: 1990, A6trophys. J. 353,
494
Gilmore, G .,: 1984, in M. Capaccioli, ed(s)., A6tronomy with Schmidt-Type Telescopes,
Reidel, Dordrecht , p77
Gilmore, G. & Reid, LN.: 1983, M.N.R.A.S. 202, 1025
Gilmore, G ., Reid, LN. & Hewett, P.C.: 1985, M.N.R.A.S. 213, 257
Gilmore, G. & Wyse, R.F.G .: 1985, Astr. J. 90, 2015
Gilmore, G . & Wyse, R .F.G.: 1986, Nature 322, 806
Gilmore, G . & Wyse, R .F .G.: 1987, in G. Gilmore & B . Carswell, ed(s)., The Gala:ey,
Reidel, Dordrecht, p247
Gilmore, G., Wyse, R.F.G., & Kuijken, K.: 1989, Ann. Rev. A6tr. Ap. 27, 555
Gunn, J .E. & Gott, J .R .: 1972, A6trophY6. J. 176, 1
Harmon, R.T . & Gilmore, G .: 1988, M.N.R .A.S. 235,1025
Hartwick, F.D.A. : 1976, A6trophys. J. 209, 418
Hauser, M.G. et al. : 1990 NASA photograph G90-03046
Heckman, T.M., Armus, L. & Miley, G.K.: 1990, Astrophys. J. Suppl. 74, 833
Hoffman, Y ., Silk, J. & Wyse, R.F.G. : 1992, A6trophY6. J . (Lett.) 388, L13
Then, I. & Tutukov , A .V.: 1987, A6trophY6. J . 313, 727
 325

Impey, C.D. & Bothun, G.D.: 1989, Astrophys. J. 341, 89

Jones, B.J.T. & Wyse, R .F.G.: 1983, Ii 120, 165
Jura, M.: 1980, Astrophys. J. 238, 499
Katz, N. & Gunn, J.: 1991, Astrophys. J. 377,365
Kennicutt, R.C.: 1983, Astrophys. J. 272, 54
Kennicutt, R.C.: 1989, Astrophys. J. 344, 685
Knezek, P .M. & Schneider, S.E.:1992, poster paper at 3rd Tetons Summer School.
Lacey, c.: 1991, in B. Sundelius, ed(s)., The Dynamics of Disc Galaxies, University of
Gothenburg Press, p257
Lacey, C.G. & Ostriker, J .P.: 1985, A6trophys. J. 299, 633
Lacey, C. & Silk, J.: 1991, Astrophys. J. 381, 14
Laird, J.B.: 1993, in J. Brodie & G. Smith, ed(s)., Globular Clusters in the Context of
their Parent Galaxies, ASP Conference Series, in press
Laird, J.B., Rupen, M.P., Carney, B.W., & Latham, D.W.: 1988, Astr. J. 96, 1908
Larson, R.B.: 1974, M .N.R.A .S. 169, 229
Larson, R.B. & Tinsley, B.M.: 1978, Astrophys. J. 219, 46
Larson, R.B., Tinsley, B .M. & Caldwell, C.N.: 1980, Astrophys. J . 237, 692
Lee, Y.-W.: 1992, Astr. J. 104,1780
Majewski, S.R.: 1992, Astrophys. J . Suppl. 78, 87
Matteucci, F.: 1987, in M. Azzopardi, ed(s)., Stellar Evolution and Dynamics of the Outer
Halo of the Galaxy, ESO, Garching, p609
Matteucci, F. & Brocato, E.: 1990, A6trophys. J. 365, 539
Meurer, G.A., Freeman, K.C., Dopita, M.A., & Cacciara, C. : 1992, Astr. J . 103, 60
Minniti, D., White, S.D.M., Olszewski, E.W., & Hill, J .M.: 1992, Astrophys. J. (Lett.)
393,L47
Morrison, H., Flynn, C., & Freeman, K.C.: 1990, Astr. J. 100, 1191
Murray, C.A.: 1986, M.N.R.A.S. 223, 64
Navarro, J.:1992,poster paper at 3rd Tetons Summer School.
Norman, C.A.: 1991, in H . Bloemen, ed(s)., The Interstellar Disk-Halo Connection in
Galaxies, Reidel, Dordrecht, p337
Norris, J.: 1987, in G. Gilmore & B. Carswell, ed(s) ., The Galaxy, Reidel, Dordrecht, p297
Ostriker, J.P.: 1990, in R.G. Kron, ed(s)., Evolution of the Universe of Galaxies,
A.S.P. Conf. Series, p25
Ostriker, J.P. & Thuan, T.X.: 1975, Astrophys. J. 202, 353
Peacock, J.A. & Heavens, A .F .: 1985, M.N.R.A.S. 217,805
Phillipps, S., Disney, M.J., Kibblewhite, E.J., & Cawson, M.G .M.: 1987, M.N.R.A.S. 229,
505
Quinn, P .J., Hernquist, L., & Fullager, D.P.: 1993, Astrophys. J. 403,74
Quinn, P.J . & Zurek, W.H .: 1988, Astrophys. J. 331, 1
Quirk, W.J.: 1971, Astrophys. J . 167, 7
Rana, N.C .: 1991, Ann. Rev. Astr. Ap. 29, 129
Ratnatunga, K.U. & Freeman, K.C.: 1989, Astrophys. J. 339, 126
Rees, M.J. & Ostriker, J.P.: 1977, M.N.R.A.S. 179, 541
Rich, R.M.: 1988, Astr. J. 95, 828
Rich, R.M. : 1989, in M. Morris, ed(s)., The Center of the Galaxy, Kluwer, Dordrecht, p63
Rich, R.M.: 1993, in J. Brodie & G. Smith, ed(s)., Globular Clusters in the Context of
their Parent Galaxies, ASP Conference Series, in press
Sancisi, R.: 1990, in G. Fabbiano, J.S. Gallagher, & A. Renzini, ed(s)., Windows on Galax-
ies, Kluwer, Dordrecht, p199
Sandage, A. : 1986, Astr. Ap. 161, 89
Schombert, J.M., Bothun, G.D., Schneider, S.E., & McGaugh, S.S.: 1992, Astr. J. 103,
1107
Schuster, W.J. & Nissen, P.E.: 1989, Astr. Ap. 222, 69
Sellwood, J. & Carlberg, R.: 1984, Astrophys. J. 282, 61
Sharples, R., Walker, A., & Cropper, M.: 1990, M.N.R.A.S. 246, 54
326

Silk, J.: 1977, Astrophys. J. 211, 638

Silk, J.: 1993, Astrophys. J. ,in press
Silk, J. & Wyse, R.F.G.: 1993, Phys. Rep. ,in press
Smecker, T.A. & Wyse, R.F.G.: 1991, Astrophys. J. 372,448
Smecker-Hane, T.A. & Wyse, R.F.G.: 1992, Astr. J. 103, 1621
Soubiran, C ., 1992, PhD Thesis, Observatoire de Paris.
Stobie, R.S ., Ishida, K., & Peacock, J.A.: 1989, M.N.R.A.S. 238, 709
Stromgren, B.: 1987, in G. Gilmore & B. Carswell, ed(s)., The Galaxy, Reidel, Dordrecht ,
p229
Suntzeff, N.: 1993, in J. Brodie & G. Smith, ed(s)., Globular Clusters in the Context of
their Parent Galaxies, ASP Conference Series, in press
Terndrup, D.M.: 1988, Astr. J. 96, 884 .
Terndrup, D.M., Frogel, J.A., & Whitford, A.E.: 1990, Astrophys. J. 357,453
Tinsley, B.M.: 1979, Astrophys. J. 229, 1046
Toomre, A.: 1977, in B.M. Tinsley & R.B . Larson, ed(s)., The Evolution of Galaxies and
Stellar Populationl, Yale Univ. Obs, New Haven, p401
T6th, G. & Ostriker, J.P.: 1992, Astrophys. J. 389, 5
Truran, J.W.: 1987, in M.P. Ulmer, ed(s)., 13th Texas Symposium on Relativistic Astro-
physics, World Scientific, Singapore, p430
Valentijn, E.A.: 1990, Nature 346, 153
van den Bergh, S. : 1979, in M.S. Longair & J.W . Warner, ed(s)., Scientific Research with
the Space Telescope, NASA CP-2111, p151
van der Kruit, P.C.: 1987, Astr. Ap. 173, 59
van der Kruit, P. C.: 1990, The Milky Way as a Galaxy, G. Gilmore, I.R. King, P.C . van
der Kruit, University Science Books, Berkeley
White, S.D.M.: 1980, M.N.R.A.S. 191, IP
White, S.D.M. & Frenk, C.S.: 1991, Astrophys. J. 379, 52
Wolfe, A .: 1989, in C.S. Frenk et al., ed(s)., The Epoch of Galaxy Formation, Kluwer,
Dordrecht, p101
Whitelock, P., Feast, M., & Catchpole, R.: 1991, M.N.R.A.S. 248, 276
Wyse, R.F.G. & Gilmore, G .: 1986, Astr. J. 91, 855. (Erratum 1986, Astr. J., 92, 1215.)
Wyse, R.F.G. & Gilmore, G.: 1988, Astr. J. 95, 1404
Wyse, R.F.G. & Gilmoce, G.: 1989, Comments on Ap 13, 135
Wyse, R.F.G. & Gilmore, G. : 1991, Astrophys. J. 367, L55
Wyse, R.F.G. & Gilmore, G.: 1992, Astr. J. 104, 144
Wyse, R.F.G. & Silk, J. : 1985, Astrophys. J. (Lett.) 296, Ll
Wyse, R.F.G. & Silk, J.: 1989, Astrophys. J. 339, 700
Yamagata, T. & Yoshii, Y.: 1992, Astr. J. 103, 117
Yoshii, Y., Ishida, K ., & Stobie, R.S.: 1987, Astr. J. 93, 323
Zaritsky, D. & Lorrimer S.J., 1992, poster paper at 3rd Tetons Sununer School.
Zurek, W., Quinn, P.J., & Salmon, J.: 1988, Astrophys. J. 330, 519
 MERGERS AND THE ORIGIN OF EARLY-TYPE GALAXIES
LARS HERNQUIST
Lick Observatory, UC Santa Cruz, Santa Cruz, CA 95064

Abstract. The relevance of mergers to the structure of present-day galaxies is examined,

focusing on the proposal that most early-type galaxies originate by "major" mergers of
late-type progenitors. In particular, N-body simulations are used to predict the structure
of remnants formed by mergers of various types of progenitor galaxies. Observational
signatures of major mergers are discussed along with difficulties arising from evolutionary
effects.
Key words: galaxy evolution, galaxy formation, galaxy structure

1. Introduction
Contrary to views held by most astronomers during the first half of the twen-
tieth century, modern observations have demonstrated convincingly that
galaxies are not secluded "island universes," but interact with one another on
a regular basis. In fact, it is now believed that encounters and outright merg-
ers playa role in a vast array of galactic phenomena, including bridges and
tails, bars and grand-design spirals, rings in disks, "shells," "ripples," and
other fine structure, kinematic subsystems, and anomalous emission from
the nuclei of starburst and active galaxies (for recent reviews, see Barnes et
al. 1991; Barnes & Hernquist 1992a).
While it is generally accepted that collisions likely shape most peculiar
galaxies, such as those first surveyed by Arp (1966), their relevance to the
structure of "ordinary" galaxies is still much debated. The suggestion that
most earl~-type galaxies are remnants of mergers between spiral progenitors
remains especially controversial. Given that fully 20 and 15 years, respec-
tively, have elapsed since the seminal articles on this topic by Toomre &
Toomre (1972) and Toomre (1977) , it is appropriate to review the argu-
ments both for against this "merger hypothesis." For conciseness, the dis-
cussion here is restricted mostly to the origin of elliptical galaxies, although
many of the arguments likely pertain to 50's and even some early spirals.

2. Historical Perspective
The study of interacting and merging galaxies has a surprisingly long his-
tory considering that significant advances have been made mainly within the
past'" 15 - 20 years. As early as 1926, Lindblad suggested that encounters
of "nebulae" would likely be highly inelastic and " ... tend to convert trans-
lational into rotational kinetic energy, [leading to] a fusion of the respective
bodies." While it seems probable that Lindblad was thinking of collisions
between gaseous objects, it is now recognized that galaxy interactions are
indeed quite "sticky." The first serious attempt to simulate galaxy collisions

327
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution o( Galaxies, 327-344.
© 1993 Kluwer Academic Publi.ihers.
328

was made by Holmberg in 1941, using analog computations. Unfortunately,

his algorithms appear to have not been fully debugged as he concluded
that tidal effects do not lead to mergers! Notwithstanding Holmberg's as-
sessments, Zwicky (1959) put forward the notion that close encounters of
galaxies lead to " ...considerable disruption of both systems or total mutual
capture," in remarkable agreement with our present understanding of the
dynamics of galaxy interactions.

Following Holmberg's early attempts to model colliding galaxies, theoret-

ical progress remained at a standstill until the development of high-speed
computers in the 1960's and 1970's. Analytic work employing the impulse
approximation (e.g. Alladin 1965) could only begin to characterize the dy-
namical violence of slow, interpenetrating encounters between galaxies. In-
deed, not until Toomre & Toomre's (1972) study of galactic bridges and tails
did the true nature of tidal forces in such events become widely appreciated.
In addition to elucidating the response of rotating disks to tides, Toomre &
Toomre also conjectured that the attendant conversion of orbital into inter-
nal energy would lead to mergers whose remnants would" ... simply blend or
twnble into a single three-dimensional pile of stars [which may represent] a
scenario for nothing less than the delayed formation of elliptical galaxies."

The possibility that many, and perhaps even all ellipticals originate in
roughly this manner was formalized by Toomre (1977) five years later and
is embodied in what is now commonly termed the "merger hypothesis." Ac-
cording to this suggestion, early-type galaxies are nothing more than rem-
nants of "major mergers" between two or several comparable-mass spirals.
This view runs counter to the notion that ellipticals are older than spirals,
as might be inferred by comparing the stellar populations of different mem-
bers of the Hubble sequence. Although Toomre's statement of the merger
hypothesis is somewhat vague as regards the progenitor types which might
yield ellipticals, he evidently had in mind mergers of disks consisting largely
of stars. The ambiguity in the nature of the progenitors has apparently led
to some confusion over what constitutes a "merger" within the context of
Toomre's proposition. For example, when Ostriker (1980) argued that ellip-
ticals cannot be produced by mergers of spiral galaxies, it seems that he was
thinking mainly of progenitors whose luminous mass resides entirely in stel-
lar disks. In fact, it now appears that mergers of stellar disks by themselves
do not yield remnants sufficiently concentrated to be identified with ob-
served early-type galaxies. However, as discussed below, these concerns are
obviated by more reasonable interpretations of Toomre's conjecture, which
admit mergers between progenitors with less restricted properties.
 329

3. Statistics of Merging
Clearly, the merger hypothesis requirel!. that the average rate of mergers over
the history of the universe must be able to account for the observed num-
ber of ellipticals. From careful inspection of data available to him, Toomre
(1977) argued that roughly 11 out of the 4000 NGC galaxies are "good"
merger candidates, possessing double tidal tails and two density centers.
For a constant rate of merging, the number of remnants, Nremnant6' will be
related to the number of merger candidates, N candidate6 by
Tage
NremnantIJ ,...., NcandidatelJ (1)
Tmerger

where Tage is the typical age of the progenitor galaxies and T merger is the
time interval over which a merger remnant would be identifiable according to
Toomre's selection criteria. For Tage ,...., 12.5 X 109 years and Tmerger '" 5 X 108
years, say, NremnantlJ '" 11 X 12.5/0.5 = 275. But, approximately 20% of all
galaxies are ellipticals, or '" 800 out of the 4000 NGC galaxies. Thus, the
most simple-minded estimate of the number of merger remnants lurking
among the NGC galaxies falls short by a factor of ,...., 3.
Present scenarios for the evolution of structure in the universe suggest
that estimates such as these are compromised by the fact that the merger
rate is epoch-dependent. Let Nmerger(t) be the merger rate at time t. Then,
the number of remnants will crudely be

NremnantlJ
N candidate6
~l tnow

tgl
Nmerger d t / l
tnow
t notu -Tmerge'r
Nmerger dt, (2)

where tgf is the time at which the typical progenitor galaxy forms and t now
is the present time reckoned from the Big Bang. It is useful to parametrize
the merger rate by the redshift, Z, according to

Nmerger <X (1 + z)m • (3)

As an example, consider an Einstein-de Sitter universe, where time is .related

to redshift by
2 1 1
t - - - (4)
3 Ho (1 + Z)2/3 .
-,------:-_=_.:::
-
Combining equations (2-4) yields, after algebra

1 - (1 + Zgf )m-3/2
Nremnant6 = N candidate6 (3)
2 - m HOTmerger
, (5)

where Zgf is the typical red shift at which the progenitors form and it IS
assumed that HOTmerger ~ 1.
330

If galaxies merge from loosely bound orbits initially with a flat distri-
bution of binding energies, then Nmerger ex: C 5 / 3 (Toomre 1977), implying
'Tn = 5/2 and
Zgj
Nremnant~ = Ncandidate~ -H--=-- (6)
OTmerger

For Ho = 50, Tmerger = 5 X 108 years, and Zgj = 2, equation (6) predicts
Nremnant~ '" 11 X (2 X 20)/0.5 ~ 900, as required to account for the number
of NGC ellipticals. With our present understanding of galaxy formation and
merger dynamics, it now appears that the values required for Zgj and Tmerger
are perhaps overly-optimistic. Moreover, as noted by Toomre (1977), any
such determination of the number of remnants will be compromised by ob-
servational selection effects. In a magnitude-limited sample, bright pairs will
be over-represented relative to individual galaxies. From the redshifts of his
11 candidates and the typical redshift of an NGC galaxy, Toomre estimates
that the above argument overestimates the number of merger remnants by a
factor '" 3. On the other hand, Toomre selected only clear-cut cases where
a merger is almost certainly ongoing. Many lesser examples are now known;
in fact, Toomre proposed Cen A as an intermediate case and argued that
the actual number of candidates would be a factor of", 3 larger than his 11,
mostly cancelling the factor of three arising from observational bias.
Unfortunately, other difficulties tend to obscure estimates of the num-
ber of merger remnants even further. Cosmological models suggest that the
merger rate in a hierarchical universe is probably not of the form assumed by
Toomre (1977). For example, using a Press-Schechter formalism, Carlberg
(1990) shows that the merger rate is indeed described by equation (3), but
with m ~ 4.50 0 •42 , for 0 ~ 1 and CDM-like power spectra. For 0 == 1, the
number of merger remnants is then

Nremnanh = Ncandidate. H Zgf

OTmerger
[1 + Zgj + -31 Z;j] • (7)

With the same choices as above, Ho = 50, Tmerger = 5 X 108 years, and Zgj =
2, equation (7) predicts Nremnanta '" 11 x80x(13/3) ~ 4000, suggesting that
all galaxies should be ellipticals! Clearly, it is more sensible to take Ho = 50,
Tmerger = 109 years, and Zgj = 1, in which case Nremnanta '" 11 X 50 ~ 550.
The differences between these various estimates emphasize the inherent
problems in determining Nremnant. with any confidence. It is perhaps fairest
to say that the values thus obtained for Nremnant~ are not inconsistent with
the merger hypothesis but in themselves are not terribly illuminating. How-
ever, they reinforce the need for broadening our notions of what types of
"mergers" might produce early-type galaxies: Progenitors at large redshifts
Z ~ 1 were likely gas-rich objects whose disks may not yet have fully as-
sembled.
 331

4. Stellar Dynamics of Merging

In principle, many of the factors contributing to the expected number of

merger remnants can be determined theoretically. Owing to neglect of self-
gravity in their models, Toomre & Toomre (1972) were not able to estimate
the rate of orbital decay in major mergers, which partly determines the factor
Tmef'gef" Some progress along these lines was made in the late 1970's and early
1980's by, among others, van Albada & van Gorkom (1977), White (1978,
1979), and Villumsen (1982), who used N-body simulations to study mergers
of spherical, single-component galaxies. While crude by today's standards,
these investigations revealed some aspects of the dynamics of merging that
obtain even when more sophisticated models are employed. In particular,
they sufficed to demonstrate that mergers of stellar-dynamical systems are
"sticky," in the sense that interpenetrating collisions from parabolic orbits
generically lead to mergers within a few dynamical times. These short merger
times result from the efficient conversion of orbital into internal energy via
dynamical friction and tidal torques.
Of course, real galaxies consist of several distinct components, interacting
with one another gravitationally. Early work on mergers of multi-component
objects by e.g. Gerhard (1981), Farouki & Shapiro (1982), and Negroponte &
White (1983) was severely compromised by limitations of computing technol-
ogy. In particular, simulations with particle numbers N rv 1000 are plagued
by excessive numerical diffusion and do not adequately sample resonances
in phase-space. Moreover, the galaxy models employed in these calculations
were crude and did not include halos extending much further than the lu-
minous components.
As emphasized by the more recent studies of Barnes (1988, 1992) and
Hernquist (1992a,b), the rate of merging depends in detail on the mass
profiles of the progenitors, the orbital geometry of the collision, and the
interaction between the various components comprising each galaxy. It is
clearly quite hopeless to incorporate these various effects without appealing
to simulations employing models of multi-component galaxies realized with
particle numbers exceeding those of the previous studies by factors rv 100-
1000. Improvements in computing hardware along with the development of
efficient N-body algorithms, such as the tree-code originated by Barnes &
Hut (1986), and schemes for generating models of compound galaxies in a
systematic manner (Hernquist 1992c) have made calculations such as these
feasible.
An example is shown in Figure 1, where the merger of two galaxies con-
sisting of self-gravitating disks and halos is displayed. As is typical of such
encounters, the disks develop large-amplitude non-axisymmetric features
in response to their tidal forcing, resulting in extended tidal tails and inner
spirals and bars. The development of each of these features entails loss of
332

o 12.0 24.0

·..L~
.·"·'·"
',
': . '
- , ";,.
," j ••.•
- ". '-:~':'

Fig. 1. Time evolution of disk components in a merger of two disk-halo galaxies.

Time in dimensionless units is shown in the upper right corner of each frame (see
Hernquist 1992a) .
 333

halos
disks --total

.,...
6 ---total ••••• sp'in
........ spin 30 - - - orbital .' .
- - - - orbital
....... -
4 ... "\
:
:
"\
,,...........
20 \

......
,,
,.- -
2 10
\ ~
.... \
\

50 100 150
time time

Fig. 2. Time evolution of total (j), spin (8), and orbital (1) angular momenta for
disks and halos in simulation shown in Figure 1.

orbital energy, trapping the galaxies in a bound, decaying orbit. In this ex·
ample, the galaxies merge within a few dynamical times and relax to form
a smooth remnant possessing some irregular structure at large radii.
As noted by Barnes (1988), the interaction amongst the various compo-
nents in the progenitors mediates a transfer of energy and angular momen-
tum from dense regions to outlying ones. In the case of the model shown in
Figure 1, for example, this tendency leads to a nearly monotonic loss of or-
bital energy and angular momentum by the luminous disks and a "spin-up"
of the surrounding diffuse halos. This is illustrated in Figure 2 where the
spin (s), orbital (I), and total angular momenta (j) of the disks and halos are
displayed separately. Calculations such as this make it painfully obvious that
it is not meaningful to relate the structure of remnants in simulations to ac-
tual galaxies unless the models include self-gravity and a realistic treatment
of all components.

5. Structure of Remnants

The gross features of some aspects of merger remnant structure can be pre-
dicted by a simple energy argument (Hausman & Ostriker 1978). Consider
334

a merger of two identical spherical galaxies from a parabolic orbit in which

mass, luminosity, and energy are all conserved; i.e. all mass remains bound
to the remnant . Then, the remnant mass, luminosity, binding energy, and
gravitational radius will simply be equal to twice those of each progeni-
tor. The mean velocity dispersion and characteristic surface density of the
remnant will be equal to and half that of each progenitor, respectively. As
emphasized earlier, real progenitor galaxies comprise more than one compo-
nent; nevertheless, numerical simulations show that the a~ove scalings are,
for the most part, surprisingly accurate. An exception is suggested by Figure
2: The dense luminous components tend to transfer energy to the surround-
ing halos, resulting in remnants that are somewhat more tightly bound than
predicted by the energy argument (e .g. Hernquist & Bolte 1992).
Intuition gleaned from scalings such as these, hl:>wever, fails to predict
even the qualitative aspects of the kinematic structure of remnants. As
suggested by Figure 2, the transfer of angular momentum from the most
tightly bound regions to outlying ones yields remnants tha.t rotate slowly
in their inner regions, unlike mergers of single-component galaxies which
produce rapidly rotating remnants (see Barnes 1988). In particular, detailed
models have shown that the luminous remnants of mergers between multi-
component galaxies rotate slowly in regions of high surface density even
if the luminous progenitors comprise pure disks supported mainly by rota-
tion (Hernquist 1992a) or include dense, rotating bulges (Hernquist 1992b).
Orbital angular momentum is efficiently converted into intrinsic spin and
"hidden" in the halos, as in Figure 2, while the spin angular momentum
of the luminous material is largely carried off by loosely-bound debris and
shed by gravitational torques excited by non-axisymmetric structures such
as bars.
The inner, luminous remnants of mergers of multi-component progeni-
tors are characterized by values v / (F ,....., 0.2 near their centers, but possess
significant rotation at large radii. Figure 3 displays the "spin" parameter, A,
defined by Barnes (1992) for subsets of particles in four remnants produced
by mergers of disk-halo progenitors (Hernquist 1992a). This A is defined so
that it approaches zero for a system supported by dispersion and tends to
unity for an object in which all particles are on circular orbits. As implied by
Figure 3, most of the residual spin resides in the,....., 50% of the material least
tightly bound to the remnant. This appears to be a generic outcome of merg-
ers between objects resembling present-day spirals. Therefore, the merger
hypothesis will be tested by searches for rotation in the outer envelopes of
elliptical galaxies.
For the most part, mergers between spiral progenitors appear quite capa-
ble of accounting for the spatial structure of observed ellipticals. The lumi-
nous remnants are triaxial figures supported mainly by anisotropic velocity
dispersion whose density profiles are described by Hernquist (1990) models
 335

.6

/
Model 1
/
--------- Model2
----- Model3 /
.4 .I
- - - Model4

I
I "
.2

o
-.0002 -.00015 -.0001
energy

Fig. 3. Spin parameter A, as a function of dimensionless binding energy for groups

of particles in four remnants of mergers of disk-halo progenitors.

and whose projected light profiles are well-fitted by R 1 / 4 -laws (Barnes 1988,
1992; Hernquist 1992a,b). The models fail in at least two respects, however,
if the luminous progenitors comprise pure stellar disks. In this limit, the
remnants are significantly more diffuse than observed ellipticals (Hernquist
1992a). Figure 4 shows the projected surface density, S, of the remnant of
a merger between two disk-halo progenitors. While the surface density is
well-fitted by an R 1 / 4 law over a large range, the remnant has a prominent
core where S is nearly constant. In fact, the ratio of effective to core radii
for remnants such as these is typically '" 2 - 4, much smaller than values
inferred observationally for ellipticals. Second, the remnants of disk- halo
mergers tend to be more elongated than typical ellipticals. In projection,
the luminous remnants are characterized by Hubble types n '" 3 - 7, while
the distribution of observed ellipticities peaks for n '" 0 - 2 (Franx et al.
1991).
These difficulties are readily obviated by appealing to other types of pro-
genitors. The failure of mergers of purely stellar disks can be traced to the
dearth of material in each progenitor at phase-space densities as high as in
observed ellipticals (Carlberg 1986). One possible "fix" is to include suffi-
ciently dense bulges in the progenitors (e.g. Barnes 1988). Figure 5 shows
336

.5 1 1.5 2 2.5
a1/...

Fig. 4. Projected surface density, S, as a function of fourth root of ellipsoidal radius,

Q, for three orthogonal projections of the remnant produced by simulation in Figure
1.

the projected surface brightness .of a remnant formed by a merger similar to

that yielding the remnant in Figure 4, but now 25% of the progenitor mass
resides in compact bulges. The ratio of effective to core radii in the remnant
is greatly increased relative to the model shown in Figure 4 and is limited
in this case by numerical resolution. Moreover, the remnants of mergers be-
tween progenitors with compact bulges tend to be less highly elongated than
those produced from bulgeless progenitors (Hernquist 1992b).
Central phase-space constraints may also be evaded if the progenitors
are gas-rich. Simulations demonstrate that the torques operating in mergers
between disks containing gas act to remove angular momentum from the gas,
concentrating it in the inner regions of the remnant (Hernquist 1989; Barnes
 337

.5 1 1.5 2 2.5
01/4

Fig. 5. Projected surface density, S, as a function of fourth root of ellipsoidal radius,

Q, for three orthogonal projections of the remnant produced by merger of two
galaxies each comprising self-gravitating disks, bulges, and halos.

& Hernquist 1991; Hernquist & Barnes 1991). Depending on the details of
how this gas is converted into stars, the core radii of the remnants can be
reduced relative to those produced by bulgeless progenitors.

These considerations do suggest, however, that the formation of ellipticals

probably involved some dissipation, whether or not one accepts the merger
hypothesis. The first of the two possibilities noted above leaves unanswered
the process by which bulges originate, while the second relies explicitly on
dissipation to drive gas into the central regions of remnants.
338

6. Status of the Merger Hypothesis

A variety of arguments have been raised against the merger hypothesis

since it was first put forward by Toomre (for summaries, see Ostriker 1980;
Tremaine 1981). Most have been refuted, while those that linger are likely
made irrelevant by considering mergers between progenitors other than pure
stellar disks.

1. Observationally, it is well-known that the fractional abundance of el-

lip ticals is high in rich clusters, whose large velocity dispersion's inhibit
merging. As argued by White (1982) and Veeraraghavan & White (1985),
this difficulty is easily averted by assuming that merging takes place in
sub clusters which later gather together to form the clusters seen today.
In fact, this is precisely what is expected in hierarchical models where
increasingly large mass scales turn around and collapse with time (see,
e.g., Barnes 1989).

2. Ostriker (1980) has noted that ellipticals appear to satisfy both a color-
luminosity and a metallicity-Iuminosity relation, and that the most
metal-rich ellipticals are far more metal-rich than typical spirals. The
present status of these objections is in doubt . However, White (1982)
has suggested that the correlations in color, metallicity, and luminosity
may simply reflect similar properties of the progenitor galaxies. In ad-
dition, the metallicity may be boosted during a merger accompanied by
vigorous star formation.

3. Ellipticals often possess radial gradients that would be difficult to ex-

plain if mergers completely randomized the phase-space distribution of
material in the progenitors (White 1982). However, it is well-known that
violent relaxation is incomplete in mergers and that gradients present
in the original galaxies will be preserved at some level in the remnants.
Moreover, metallicity gradients, for example, can be enhanced by merg-
ers if gas concentrating in the inner regions of the remnant experiences
a burst of star formation.

4. As emphasized especially by van den Bergh (1990), ellipticals appear to

possess many more globular clusters per unit luminosity than do spirals.
In principle, this is a very serious difficulty for the merger hypothesis.
While our understanding of the origin of globular clusters is rudimen-
tary, it has been suggested that they form during mergers (e.g. Schweizer
1987; Ashman & Zepf 1992), a suggestion which may in fact have obser-
vational support (Lutz 1991; Holtzmann et al. 1992) .
 339

5. Dwarf ellipticals are probably not made by mergers of less-massive disks.

At present, a suitable class of progenitors is not known. Moreover, the
progenitors would likely represent a class of object distinct from present-
day galaxies, as they would necessarily be more centrally concentrated
than disks in local spirals to account for the high phase-space densities
of bulges and dwarf ellipticals. It is amusing, however, that recent mod-
eling suggests that some dwarfs may be by-products of major mergers,
forming out of debris in tidal tails (Barnes & Hernquist 1992b).

6. J>erhaps the most annoying objection to the merger hypothesis is the

fact that ellipticals are too concentrated to be built from mergers of
stellar disks, as indicated by the results presented in §5, and in accord
with arguments made by a number of workers on previous occasions
(Ostriker 1980; Carlberg 1986; Gunn 1987; Lake 1989). This difficulty
can be averted in principle if the progenitors are gas-rich and in practice
if sufficient progenitor mass resides already in dense spheroidal compo-
nents.

Of these various points, (1) is refuted by appealing to the growth of sub-

structure in hierarchical cosmogonies, (2) and (3) can be circumvented by
appealing to the incomplete nature of violent relaxation or perhaps by in-
voking star formation during mergers, ( 4) may suffer a similar fate if mergers
can form globular clusters, (5) has no clear refutation, and (6) rules out only
an extreme form of the merger hypothesis in which ellipticals are postulated
to originate from mergers of purely stellar disks. In retrospect, the resilience
of the merger hypothesis is quite remarkable considering the varied nature
of the attempts to overthrow it.

7. Observational Signatures
Given that major mergers represent a viable mechanism for producing mas-
sive ellipticals, depending on the nature of the progenitors, it is clearly im-
portant to determine observational means for identifying remnants. Tidal
tails are obviously "smoking guns," but these features do not persist long
and, in any event, they are not prominent in all orbital configurations. Still,
merger remnants should exhibit considerable debris at large radii, as sug-
gested by Figure 1. Therefore, observational searches for faint structures in
the outer envelopes of ellipticals would appear a fruitful approach for re-
vealing merger products. For example, Sansom et al. (1988) have argued
convincingly that the "Pentagon" galaxy, NGC 6776 is the result of an old
merger, on the basis of its diffuse tidal structures. The results in §5, es-
pecially as embodied in Figure 3, imply that remnants produced in major
mergers will possess significant rotation at large radii. Clearly, it would be
340

of interest to observe such rotation, although it has yet to be demonstrated

that this effect is a unique signature of a major merger where a significant
fraction of the progenitor mass resides in disks.
If attempts to identify merger products through careful analysis ofloosely
bound material are not viable, there is little alternative to observing the
internal structure of ellipticals. As most ellipticals are comprised of.old stars,
the presence of ongoing star formation and a disturbed morphology may
well imply a recent merger or accretion event, as in e.g. NGC 3597 (Lutz
1991). Schweizer (1980, 1986, 1992) has pioneered the use of fine str~cture
as an observational probe of galaxy interactions. In an interesting study,
Schweizer et ai. (1990) show how correlations between spectral features and
fine structure, such as shells and plumes, may identify intermediate-age
remnants older than '"" 109 years, but much younger than f'V 1010 years.
Presently, this technique appears quite promising, particularly in view of
recent modeling which demonstrates that shells are a natural consequence
of major mergers (Hernquist & Spergel 1992), contrary to the commonly-
accepted view that such features result only from minor mergers or accretion
events (e.g. Quinn 1984; Hernquist & Quinn 1988).
Similarly, Bender and his collaborators (Bender 1988, 1990a,b; Bender
et ai. 1989) have argued that correlations between "boxiness" in ellipticals
and radio and x-ray emission provides evidence for an intermediate-age
population of merger remnants. This is also an interesting proposal, although
it is not clear that boxy isophotes in themselves provide conclusive evidence
for a merger (Hernquist 1992a).
As argued above, it now seems certain that major mergers will yield ob-
jects resembling early-type galaxies only if the progenitors are comprised of
more than one component. Owing to the incomplete nature of violent relax-
ation in mergers (see Spergel & Hernquist 1992 for a recent discussion) the
components of the progenitors will not be fully mixed in the remnant. There-
fore, observational searches for radial gradients in ellipticals may reveal some
merger products. For example, the projected light profiles of remnants may
display weak "seams" in the transition between regions dominated by various
components from the progenitors. Likewise, star formation may boost color
and metallicity gradients, especially if much of the gas in the progenitors is
driven into the nuclei of remnants.
Finally, stellar kinematics may provide unique signatures of various sce-
narios for forming early-type galaxies. As demonstrated by Barnes (1992),
the orbital structure of remnants depends on the nature of the progenitors
and the geometry of the encounter. In particular, it appears that major merg-
ers often yield remnants having large misalignments between their spin and
minor axes. However, observations imply that ellipticals typically possess
small kinematic misalignments (see, e.g., de Zeeuw & Franx 1991). Clearly,
this result is potentially of great import, since it may provide a diagnostic
 341

of the nature of the progenitors which merged to shape most ellipticals.

8. Conclusions
The merger hypothesis was introduced by Toomre (1977) as an alternate
view to the then commonly accepted notion that spheroids formed early
in the universe through dissipational collapse and tHat disks formed later,
perhaps by secondary infall. ill the merger hypothesis, disks are the funda-
mental building blocks which form ellipticals through their collisions. The
most extreme interpretation of Toomre's hypothesis would have us believe
that primordial galaxy formation yields disks of stars and that ellipticals are
produced by dissipationless merging. This radical extreme, while appealing
in its simplicity, is not tenable: present-day stellar disks are too diffuse to
make elliptical galaxies. These problems are obviated by weakly dissipational
mergers or when the progenitors include modest spheroids.
These considerations imply that the the dissipational collapse and merger
hypotheses for the origin of elliptical galaxies are to some extent converging
(Kormendy 1989, 1990). Of course, even in the most extreme merger picture
it would be silly to pretend that elliptical galaxies could arise solely by
collisionless processes: Without much gaseous dissipation and settling prior
to star formation, there would be no way to build any of the disks destined to
merge. Similarly, recent modeling of galaxy formation via collapse of density
perturbations in the early universe suggests that any ellipticals formed in
this manner were not built by the smooth collapse of gas clouds (as in, e.g.,
Carlberg 1984), but rather through the coalescence of fragments on sub-
galactic scales (e.g. Dubinski & Carlberg 1991; Katz & Gunn 1991). It is
plausible that much of the gas would have been converted into stars before
the fragments coalesced. Thus, the dynamics of this process is quite similar
to, e.g., the formation of ellipticals by repeated merging in compact groups
(Barnes 1989). Indeed, it may be quite difficult to isolate the dominant
merger "channel" by which the majority of ellipticals were assembled.
To wit, consider three dynamical scenarios, all of which yield remnants
which share properties in common with observed ellipticals:

1. The formation of an elliptical by a major merger where the two progeni-

tors are of comparable mass, are dominated by disks, and possess either
compact bulges or sufficient amounts of gas to obviate phase-space con-
straints.

2. The formation of an elliptical via a sequence of major and minor merg-

ers in which the progenitors resemble present-day spirals and ellipticals
and have masses that are individually much smaller than the mass of
the remnant.
342

3. The formation of an elliptical through a series of merger events in which

the progenitors are fragments of a much more massive gas cloud, are
mainly stellar in nature, and again have masses that are small com-
pared to the ensuing remnant.

The first possibility is most similar to that envisaged originally by Toomre

(1977), the second may be relevant to the long-term evolution of groups
(Barnes 1989), while the third appears to be a generic outcome of frag-
mented collapse in an expanding universe. In terms of the dynamical pro-
cesses which ultimately shape the remnant, are these three scenarios really
distinct? In view of the fact that mergers appear invariably to produce rem-
nants having Rl/4_law light profiles over some range and considering that
galaxy formation in a hierarchical cosmogony likely leads to fragmentation
on sub-galactic scales, perhaps the relevant question is no longer "Did ellip-
ticals form by mergers?" but rather "When did the mergers which formed
ellipticals occur, and what was the nature of the progenitors?"
Answers to these questions will be provided through future theoretical
and observational research. From the point of view of modeling, the most
pressing need is to quantify the structure of remnants produced by merg-
ers in different limits, as in examples 1-3 above. In this manner, it will be
possible to predict which properties, if any, are unique to a particular sce-
nario. From the point of view of phenomenology, it is critical to elucidate the
physical processes underlying the formation of bulges, especially to the ex-
tent that these objects are similar in nature to early-type galaxies. Finally,
from an observational point of view, it is desirable to determine what, if
any, signatures are unique to merging in a particular epoch of the universe.
Kinematic misalignments may represent one such signature. Another issue
concerns the question of whether or not star formation during mergers will
yield "seamless" light profiles, and to what extent observations can be used
to interpret departures from a featureless luminosity distribution.
The preceding discussion should hardly be interpreted as signaling the
demise of the merger hypothesis. Richard Feynman was once overheard lec-
turing to a class of students explaining his views on how "good" and "bad"
theories are distinguished. According to his maxim, paraphrased slightly, the
deeper one probes with a weak theory, the more one is led to question the
fundamental assumptions and assertions underlying that theory, while the
deeper one probes with a good theory, the more physical phenomena that
theory ezplains. On this basis, it is quite obvious that the merger hypothesis
is a "good" theory, as demonstrated by the many aspects of galaxy evolu-
tion and formation which have been attributed to galaxy collisions. Indeed,
the merger hypothesis represents one of the most innovative and important
contributions to these fields.
 343

Acknowledgements

This work was supported in part by the Pittsburgh Supercomputing Center,

the Alfred P. Sloan Foundation, NASA Grant NAGW-2422, and the NSF
under Grants AST 90-18526 and the Presidential Faculty Fellows Program.

References
Alladin, S. M . 1965, ApJ 141, 768.
Arp, H . 1966, ApJS 14, 1.
AsluTIan, K.M. &: Zepf, S.E. 1992, ApJ 38:4, 50.
Barnes, J. E. 1988, ApJ 331, 699.
Barnes, J. E. 1989, Nature 338, 132.
Barnes, J. E. 1992, ApJ 393, 484.
Barnes, J. E. &: Hernquist, L. 1991, ApJ 370, L65.
Barnes, J . E. &: Hernquist, L. 1992a, ARA&:A 30, 705.
Barnes, J. E. &: Hernquist, L. 1992b, Nature, submitted.
Barnes, J. E., Hernquist, L. &: Schweizer, F. 1991, Sci. Am. 265,40.
Barnes, J. E. &: Hut, P. 1986, Nature 324, 446.
Bender, R. 1988, A&:A 193, L7 .
Bender, R. 1990a, in Dynamics and Interactions of Galaxies, ed. R. Wielen, p. 232 . Berlin:
Springer-Verlag.
Bender, R. 1990b, A&:A 229, 441.
Bender, R., Surma, P ., Dobereiner, S., Mollenhoff, C. &: Madejsky, R. 1989, A&:A 217, 35.
Carlberg, R. 1984, ApJ 286, 403.
Carlberg, R. 1986, ApJ 310, 593.
Carlberg, R. 1990, ApJ 350, 505.
de Zeeuw, T. &: Franx, M. 1991, ARA&:A 29, 239.
Dubinski, J . &: Carlberg, R.G. 1991, ApJ 378, 496.
Farouki, R.T. &: Shapiro, S. 1982, ApJ 259, 103 .
Franx, M., Dlingworth, G.D. &: de Zeeuw, T. 1991, ApJ 383, 12.
Gerhard, O. 1981, MNRAS 197, 179.
Gunn, J .E. 1987, in Nearly Normal Galaxies, ed. S. Faber, p. 455. New York: Springer-
Verlag.
Hausman, M. &: Ostriker, J.P. 1978, ApJ 224, 320.
Hernquist, L. 1989, Nature 340, 687 .
Hernquist, L . 1990, ApJ, 356, 359.
Hernquist, L . 1992a, ApJ, 400, 460
Hernquist, L. 1992b, ApJ, submitted.
Hernquist, L. 1992c, ApJ, submitted.
Hernquist, L. &: Barnes, J. E. 1991, Nature 354, 210.
Hernquist, L. &: Bolte, M. 1992, ApJ, submitted.
Hernquist, L. &: Quinn, P. J. 1988, ApJ 331, 682.
Hernquist, L. &: Spergel, D. N. 1992, ApJ, 399, L1l7.
Holmberg, E. 1941, ApJ 94, 385.
Holtzmann, J. A. et al. 1992, AJ 103, 691.
Katz, N. &; Gunn, J.E. 1991, ApJ 377, 365.
Kormendy, J. 1989, ApJ 342, L63.
Kormendy, J. 1990, in Dynamics and Interactions of Galaxies, ed. R . Wielen, p. 499.
Berlin: Springer-Verlag.
Lake, G. 1989, AJ 97, 1312.
Lindblad, B. 1926, Arkiv. Mat . Ask Fys. 19A, No. 35.
Lutz, D. 1991, A&:A 245, 31.
Negroponte, J . &: White, S.D.M. 1983, MNRAS 205, 1009.
344

Ostriker, J.P. 1980, Comments Astrophys. 8, 177.

Quinn, P. J. 1984, ApJ 279, 596.
Sansom, A.E., Reid, LN. & Boisson, C. 1988, MNRAS 234, 247.
Schweizer, F. 1980, ApJ 237, 303.
Schweizer, F. 1986, Science 231, 193.
Schweizer, F. 1987, in Nearly Normal Galaxies, ed. S. F~ber, p. 18. New York: Springer-
Verlag.
Schweizer, F. 1992, in Physics of Nearby Galaxies: Nature or Nurture, ed. T.X. Thuan,
C. Balkowski & J. Tran Thanh Van, in press.
Schweizer, F., Seitzer, P., Faber, S.M., Burstein, D., DaIle are, C.M. & Gonzalez, J.J.
1990, ApJ 364, L33.
Spergel, D. N . & Hernquist, L. 1992, ApJ, 397, L75
Toornre, A . & Toornre, J. 1972, ApJ 178,623.
Toornre, A. 1977, in The Evolution of Galaxies and Stellar Populations, ed. B. Tinsley &
R. Larson, p. 401. New Haven: Yale Univ. Obs.
Tremaine, S. 1981, in The Structure and Evolution of Normal Galaxies, ed. S.M. Fall &
D. Lynden-Bell, p. 67. Cambridge: Cambridge Univ. Press.
van Albada, T.S. & van Gorkom, J.H. 1977, A&A 54, 121.
van den Bergh, S. 1990, in Dynamics and Interactions of Galaxies, ed. R. Wielen, p. 492.
Berlin: Springer-Verlag.
Villurnsen, J.V. 1982, MNRAS 199, 493 .
Veeraraghavan, S. & White, S.D.M. 1985, ApJ 296, 336.
White, S.D.M. 1978, MNRAS 184, 185.
White, S.D.M. 1979, MNRAS 189, 831.
White, S.D.M. 1982, in The Morphology and Dynamics of Galaxies, ed. L. Martinet & M.
Mayor, p. 289. Sauverny: Geneva Obs.
Zwicky, F. 1959, Hdb. d. Phys. 53, 373.
 INTERGALACTIC AND EXTENDED ATOMIC GAS

JACQUELINE VAN GORKOM

Columbia University, Astronomy Department

Abstract. Observational results are discussed on the properties of the diffuse atomic
gas in the outer parts of galaxies and between galaxies. In the local universe, no H I in
emission is found at column densities less than about 1019 cm -2, possibly because the
lower column density gas is mostly ionized. The presence of large gas reservoirs around
galaxies strongly depends on the environment that they are in. Galaxies with low-surface-
brightness, huge H I disks are found in very low density environments. In small groups,
merging and interacting galaxies are often surrounded by large H I envelopes and many
gas-rich dwarfs. Results of unbiased search~ directly in the H I line are presented, both at
z = 0 and z = 3.3. At low redshift, H I rich dwarf galaxies follow the structure as defined
by the bright galaxies. The data leave no room for a substantial population of H I dwarfs
or gas-rich, low surface brightness giants that could have been missed by optical catalogs.
Few, if any, large neutral structures seem to have survived at redshifts z ~ 3.3, although
a tentative detection of a possible pancake is ·reported. H I imaging results of quasar -
galaxy pairs shows that at least the nearby metal absorbers arise in the outskirts of tidally
disturbed galaxies. A sensitive search for H I emission from the vicinity of nearby Lya
clouds failed to detect any H I.

1. Introduction

Once upon a time, galaxies, as we know them today, consisted entirely of

gas. Thus for a proper understanding of galaxy formation and evolution, we
will need to know what gas reservoirs exist around and between galaxies.
Whether truly intergalactic neutral hydrogen exists is still an open question.
The aim of this review is to discuss what we have learned in recent years
about the diffuse gas in the outer parts of galaxies. The focus will be on
where it is, how much of it there is, how this depends on environment,
and how these gas reservoirs evolve with time. Although one of the main
motivations for studying diffuse gas associated with galaxies is to trace the
galaxy potential, this will not be discussed in this paper. Rather, I would
like to leave you with an image of what the universe looks like in the 21
cm neutral hydrogen line, which is the easiest observable and most directly
interpretable probe of the diffuse gas.
First, I shall summarize what is currently thought to be a normal H I
distribution in isolated spiral galaxies. Ever increasing sensitivity does not
appear to reveal ever increasing H I extents, and the possible causes are
discussed. The typical gas extent of a galaxy depends on what environment
it is in. In high-density environments various gas removal mechanisms are at
work; detailed imaging allows us to discriminate between these. In very low-
density environments galaxies with huge H I reservoirs are found, possibly
examples of delayed disk formation. Most extended H I is found around
small groups and interacting galaxies, and the question arises whether all

345
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution (~fGalaxies, 345-368.
© 1993 Kluwer Academic publishers.
346

this gas was pulled out of the galaxies or whether we are seeing some gas
left over from formation.
To get a true H I image of the universe, unbiased searches directly in the
H I line are needed. A relevant question to ask is whether the large-scale
distribution of gaseous things follows that of the optically luminous galaxies.
Especially intriguing is the distribution of gas rich dwarf galaxies, which
perhaps form the best example of gas reservoirs around galaxies. Some of
them might have been formed when galaxies were merging and large amounts
of gas were swept into intergalactic space. Unbiased searches in the H I line
have not only been done in the local universe. Results on similar surveys at
redshifts of 3 are presented as well.
Several excellent reviews have recently appeared which are relevant to
this paper. A discussion on the properties of the different phases of the ISM
in our Galaxy is given by Kulkarni & Heiles (1988), the H I properties of
external galaxies are discussed by Giovanelli & Haynes (1988), gas in the
outer parts of galaxies is reviewed by Sancisi (1983, 1990), while the effect
of environment on the cool phases of the ISM in galaxies is discussed by
Kenney (1990). In this conference several reviews were presented on what
can be learned about the intergalactic medium by using quasar absorption
lines. I conclude by discussing some H I imaging results of nearby quasar
absorbers.

2. The H I extent of normal field spiral galaxies

Can we predict the H I size of a galaxy by looking at the optical image?
A first attempt to answer that question was made by Bosma (1978). Using
the then available H I imaging results of about twenty spirals, he concluded
that the ratio of the H I diameter (corrected to face-on at 1.82 X 10 20 cm- 2 )
to optical diameter (de Vaucouleurs et al. 1976) is on average 2.2 ± 1.1. No
significant dependence on morphological type was found. But the question
as to whether the scale length of the radial H I distribution remains constant
in the outer parts or whether it falls abruptly remained unanswered. Bosma
pointed out another puzzle: giant H I envelopes are found around 30% of
the spirals. It is not at all clear why these envelopes only occur around some
galaxies.
With the more sensitive receivers now used at synthesis arrays, results on
much larger samples of galaxies have become available. Broeils (1992), ana-
lyzing 50 galaxies, finds that at a level of 1 M0 pc- 2 the ratio DHI/ D( 0) =
1.8 ± 0.4. Figures 1 and 2 show the radial surface density distribution for
spirals of different morphological type using the samples of Broeils (1992)
and Warmels (1986) as compiled by Cayatte et al. (1993). The radial extent
is normalized to the optical radius of each of the galaxies. The solid curves
give the mean distribution for each morphological type. Although the scatter
 347

is large, there is a hint that the H I envelopes at very low colwnn densities
are more extended for the later-type galaxies than for the Sb spirals and
earlier types.
The first systematic search for H I envelopes around spiral galaxies was
done by Briggs et al. (1980). The Arecibo telescope was used to do a very
sensitive search (down to 3 x 10 18 cm- 2 ) for H I along the major axis of
13 spiral galaxies out to 3 Holmberg radii. Two important results were ob-
tained. Extended envelopes are rare. Despite the much better sensitivity of
the Arecibo telescope, a similar number to Bosma's was found; roughly 30%
of the spiral galaxies have extended H I envelopes. Almost all the detected
H I is at column densities of 10 19 cm- 2 or above.
A more complete picture was obtained for the prototypical nearby iso-
lated spiral NGC 3198. A deep mosaic was made with the VLA by van
Gorkom et al. (1993), leading to an almost unifoI;m sensitivity across a
square degree. The 30' colwnn density limit is 4 x 10 18 cm -2. This is ten
times more sensitive than most synthesis observations. Figure 3 shows the
total hydrogen image of the galaxy, as seen from different viewing angles.
The galaxy has a very sharp edge in H I; the surface density drops by an
order of magnitude from 2 x 1019 cm- 2 to less than 4 x 10 18 cm- 2 , within
one synthesized beam (2.7 kpc). Figure 3 illustrates both the sharp edge and
the rather asymmetric H I distribution.
As it turns out, this is not an isolated result. A very sensitive scan along
the major axis of M33 with the Arecibo telescope (Corbelli, Schneider, &
Salpeter 1989) gave a strikingly similar result. The H I disks of galaxies
appear to cut off very sharply at about 1019 cm- 2 , and searches for H I
emission at lower column densities at larger distances from galaxies have
been unsuccessful (Briggs et al. 1980). Several explanations are possible.
Sharp edges could simply be due to the fact that the galaxies are running
out of gas. The turbulent velocities of the H I clouds in radial direction
(typically 10 km s-1) would widen the sharp edges to less than a few kpc.
Thus, they would remain unresolved in the current observations. However
the usual surface-density distribution in H I is highly asymmetric (NGC 3198
is not unusual in this respect). Azimuthal smearing by galactic rotation
would make sharp edges in most galaxies a rather short-lived phenomemon.
An alternative, and more general explanation is that although the edges
are sharp in H I, they are not genuine edges. Below a certain column density
the hydrogen gets ionized by the intergalactic ultraviolet radiation field. In
this scenario galaxies would be enveloped by giant H II regions and should
be glowing faintly in Ha. This explanation is appealing because it would
explain why it is so general that no H I emission at low colwnn densities is
found. Observationally though it is far from being established. The idea has
been around for a long time. Sunyaev (1969) first suggested that H I in the
peripheries of galaxies could be used to set limits on the intergalactic UV
348

;
I
I

;.1 :
/ ~~l
.j ~ .
;.
f

"
I
1
};

-1
1

i
I -.- a
I

\ .'"
'~
: - ;

~ I
<':_W:lI'ewoW,.DI) -4p1J'O ~OJJnS t4

I
,
/
/
\ ....
"
I

I
I

0. • 0. •
(,,_u..o·.wow.~u) ~ !MItQ ex.JJI'tS H ~ _W!)'~rtorOl) .tt.u.o IOOVftS H

Fig. 1. Radial surface density distributions of the neutral hydrogen in field spirals
for eight morphological types (Cayatte et al. 1993). The radius is normalized by the
optical corrected radius at the 25 th magnitude arcsec- 2 isophote. The solid line is
the average distribution for each group. Continued on next page.
 349

1/

.r
I :
: I

., J
- ~

I
J; ~
J !
-1
j
~

,
N/fll
, ', ' .
01 1 01 1
(,;_UoQ',wowor-t) o4Ituec no".-.s 14 4_w~' IWoCIIV~') 1aI1UtrO .oo1olftS. ~

:! I
.)f i
J'

Ij
!

01 1 01 1
(c _ I,I,C'lIUIO'jyoeO~) .qJt~ 1UO.lM'lS ~ ~_WO'lWOWotOl) ~ IODJIIS 14

Fig. 2. Continuation ofradial surface density distributions (Fig. 1)

.
o
c
46 00

: ... 5

•t
~

I
o
~

(
B
1
9
5
o
)

Fig. 3. The total hydrogen distribution in NGC 3198 as seen from different viewing
angles. Note the sha.rp edge and the a.symmetric distribution.

field. Silk & Sunyaev (1976) pointed out that high resolution H I imaging
should resolve the transition region from H I to H II and that that would
provide useful information on the nature of the ionizing photon spectrum.
This very problem has been tackled recently by Maloney (1990, 1993) and
by Corbelli & Salpeter (1993). Maloney (1993) used the current best estimate
of the intergalactic UV field at z = 0 and tried to reproduce the observed
cutoff in NGC 3198. He treats the gaseous disk in plane-parallel geometry
and calculates the vertical equilibrium explicitly for a composite disk-halo
model, exploring a range of velocity dispersions and ionizing photon fluxes.
The models show in general that the ionization fraction increases sharply
at a characteristic column density of a few times 1019 cm- 2 • Figure 4b
shows, for a velocity dispersion of 9 km s-t, the total hydrogen column
density along the major axis needed to match the observed neutral hydrogen
fraction for three different values of the ionizing photon flux. The observed
neutral hydrogen column densities (filled dots) agree to within 5% with
 351

10"
lL,=l

6·.!..~_ ••
~
10" - 0 '- - .. - • •
c o 0 0'-" . . .-- -~-~-------- - - - . • •
'E o
~
o .0 0
OU 0
o
• 00
~=I
~=2
101t • - NE Major .A.xis • ~=5
o .- -- . S7f Wajor Axis __ Observed N,.{R)

20 25 30 35 40 20 25 30 35 40
Radius (kpc) Radius (kpc)

Fig. 4. Total hydrogen column densities that match the observed neutral hydrogen
column densities ofNGC 3198 (Maloney 1993). The circles are the observed neutral
hydrogen densities. At left are the derived total hydrogen column densities for the
two halves of the galaxy. Note that the total hydrogen is rather symmetric, despite
the asymmetries seen in the H I distribution. At right are the calculated total
hydrogen for three different values of the radiation field.

the calculated values. An interesting outcome of this calculation is that,

although the observed neutral hydrogen distribution in the galaxy is highly
asymmetric, the calculated total hydrogen distribution (neutral plus ionized)
is much more symmetric, i.e., as the ionized fraction becomes substantial,
small differences in the total column densities are magnified in the neutral
column densities. Figure 4a shows the total hydrogen and neutral fraction
for both sides of the galaxy.
The extragalactic ionizing photon flux in the energy range between 13.6
and 200 eV needed to reproduce the observed sharp cutoffs in R I must
be ~ 104 photons cm- 2 s-l . This is embarrassingly close to the observed
upper limits to the extragalactic ionizing photon flux of 104 - 10 5 photons
cm- 2 S-1 from searches for Ra emission from gas far removed from galaxies
(Reynolds 1987; Kutyrev & Reynolds 1989; Songaila, Bryant, & Cowie 1989;
Reynolds et al. 1986; Stocke et ai. 1991). Maloney (1993) also presents the
352

predicted emission measures from the ionized skins around spiral galaxies.
These range from 0.05 to 0.2 cm- 6 pc, a factor of 5 to 20 below the current
best limits.
The work of Maloney (1993) has a number of interesting implications. H
true, it would explain why it is so general that we do not see H I in emission
at column densities below 1019 cm- 2 • It also suggests that the ubiquitously
observed asymmetries in the H I surface density distribution of galaxies,
might be much less pronounced in the total H I distribution. Last but not
least, large amounts of ionized gas could escape detection around especially
low surface brightness galaxies and in dwarfs (Maloney 1992). The scenario
would become much more believable if this ionized gas could actually be
detected. It seems that that is where future efforts should go.

3". The effect of the enviromnent

The typical gas extent of a galaxy does depend on what environment it
is in. For the high-density end of the spectrum, the centers of clusters, an
abundance of data exist. An excellent review summarizing those results is
given by Kenney (1990). Here I would like to mostly emphasize recent results
that have been obtained since then. For the low-density end of the spectrum,
the so-called voids, hardly any data exist at all. The reason is obvious: it is
hard to find galaxies in these areas. The first H I survey of optically known
galaxies in the Bootes void has recently begun and I shall present some of
these data.

3 . 1. CLUSTERS OF GALAXIES

Single dish studies over the last decade have shown convincingly that clus-
ter galaxies are deficient in H I as compared with isolated galaxies of the
same morphological type and size (see for example Haynes, Giovanelli, &
Chincarini 1984; Huchtmeier & Richter 1989). The H I imaging surveys of
the Virgo Cluster (Cayatte et al. 1990; Warmels 1988a, 1988b) have shown
that the H I morphology of the galaxies in the very center of the cluster is
strikingly different from field spirals; the H I disks are much smaller than the
optical disks. Several mechanisms have been proposed to remove the atomic
gas from cluster spirals: ram pressure sweeping, turbulent viscous stripping,
and galaxy - galaxy interactions.
In a detailed analysis of the radial profiles of the H I distribution of
individual galaxies, Cayatte et al. (1993) noticed that different galaxies are
affected in distinctly different ways. The galaxies in the core of the Virgo
Cluster have central surface densities that are the same as, if not higher
than, those for comparable galaxies in the field. The H I disks, however, are
truncated well within the optical disk. These galaxies all have high velocities
 353

relative to the mean cluster velocity, and the most likely interpretation is
that ram pressure sweeping has removed the gas from the outer parts of the
galaxies. The galaxies that are only mildly deficient and which have an H I
disk size comparable to the optical disk, have a surface density distribution
that is low across the entire face of the disk as compared to field spirals of
the same morphological type. The authors suggest that turbulent viscous
stripping might be causing this.
In contrast to ram pressure sweeping, turbulent viscosity is expected to
affect the entire disk. Galaxies with normal H I sizes, show slightly de-
pressed surface density in the outer parts, possibly due to galaxy - galaxy
interactions. These results indicate that all suggested gas-removal mecha-
nisms may play some role in the Virgo Cluster. Ram pressure sweeping does
affect galaxies in a dramatic way, but only the few galaxies that pass at high
speeds through the very center of the cluster are affected. A recent survey of
the Hydra Cluster (McMahon 1992) shows that no severely stripped galaxies
are present, despite the fact that Hydra has a central X-ray source compa-
rable to the one in Virgo. The spiral galaxies are all mildly deficient. These
results and the velocity structure of the cluster, suggest that Hydra is not
a relaxed cluster and that most of the spirals are located in the outskirts of
the cluster. Thus, although clusters at low redshift show, in general, some
H I deficiency, how individual galaxies get affected probably depends rather
sensitively on the structure and evolutionary stage of the cluster.

3.2. H I PROPERTIES OF GALAXIES IN VOIDS

The existence of large voids in the distribution of optically bright galaxies

is a fairly recently discovered phenomenon. The first void to be recognized
as such was the Bootes void (Kirshner et al. 1981). Being the first, this void
has been the subject of a large number of investigations. By now, almost 30
galaxies have been identified in objective prism surveys and redshift surveys
of IRAS galaxies (e.g., Moody et al. 1987; Bothun & Aldering 1988; Dey et
al. 1990). Despite these findings the region is still under dense by a factor
of four. To determine whether these void galaxies are different from field
galaxies is important for theories of galaxy formation and evolution. The
absence of interactions between galaxies in such an extremely under dense
region might lead to retarded star formation and undisturbed, extended H I
disks. It has also been suggested (Hoffman et al. 1992) that initial conditions
can have a strong influence on the structure of galaxies, causing massive low
surface density galaxies like Malin 1 (§4) to form in voids.
The first results of a VLA H I survey of the optically known galaxies in
the void (Szomoru, Gregg, & Van Gorkom, in preparation) confirm some of
these expectations. Most of the galaxies turn out to have large H I masses .
They were easily detected despite the large distance to the Bootes void (the
354

void is centered at a redshift of 15,500 km S-I). In many of the observa-

tions, other (optically uncatalogued) galaxies were detected in H I as well.
High-resolution H I imaging of the first such H I discovered galaxy in the
Bootes void shows that the galaxy has an irregular, slowly-rotating H I disk,
extending as far out as 6.4R 2S • The rotation curve of the galaxy is flat; the
amplitude and shape can be explained by the presence of a dark halo with
a mass of 1.6 times the luminous mass (Szomoru et al. 1993). So, as ex-
pected, the galaxy is gas rich: MH I/ LB ~ 1.1 and its H I disk is enormous,
extending in places out to 45 kpc. However it is not a low surface brightness
galaxy and does not lie in the same mass range as Malin 1. Its H I mass is
5 X 10 9 M 0 , and its dynamical mass is seven times that.

3 .3 . GALAXIES IN SMALL GROUPS, MERGING AND INTERACTING GALAXIES

As was discussed by Kenney (1990), loose groups provide some of the best
hunting grounds for diffuse intergalactic H I. This turns out to be equally
true for compact groups of galaxies, such as HCG 31 which often are em-
bedded in enormous gaseous envelopes (Williams, McMahon, & van Gorkom
1991). At this conference some beautiful examples have been presented of
very extended gaseous envelopes of gas around interacting and merging
galaxies (e.g., Yun 1993; Hibbard 1993). One thing I find remarkable about
these results is that if one would only look at the neutral hydrogen around
galaxies like NGC 520 or M81, one would never guess what (optical) tur-
moil is going on in the center; the gas looks like a disk in rather regular
rotation. The other interesting finding is that often H I clumps with cur-
rent star formation are found around mergers. This has led several people
to suggest that dwarfs might actua.lly be formed during the merger process
(e.g., Barnes & Hernquist 1992; Kaufman et al. 1993). Figure 5 shows an
H I overlay (contours) on an optical photograph (grayscale) of N GC 520.
Note the small dwarf north west of the galaxy. At right the velocity field is
shown and the gas at the position of the dwarf fits in very smoothly. I cannot
help but notice that some of these mergers and some compact groups look
remarkably similar to some of the simulations by, for example, Katz (1992)
on galaxy formation.

4. Optically inconspicuous gaseous giants

From the previous discussion, we might conclude that we may not gain
very much by integrating deeper to find previously undetected H I. The
Bootes results, on the other hand, do suggest that there might be as yet
undetected H I in regions where we have never searched, because such voids
do not contain optically luminous galaxies. The best example of a gas-rich,
but optica.lly inconspicuous galaxy is still Malin 1 (Jmpey & Bothun 1989).
 355

NOUVNI1:ll0

M
o
(OS6~8) NOI.LVNIl030

Fig. 5. H I overlay (contours) on an optical image of the merger NGC 520. At

left the total hydrogen, at right the velocity field. Note the enormous H I extent
compared to the optical image. The gas associated with the tiny dwarf fits smoothly
in the velocity field (Hibbard 1993).
356

AnlOng its properties are an extremely low-surface-brightness disk, with B(O)

= 26 mag arcsec- 2 , an extremely large size (one scale length is 55 kpc),
and a very large H I mass (3 X 10 10 M0). High-resolution (15 arcsec) VLA
H I imaging (van Gorkom, Bothun, & Impey 1993, in preparation) shows
that the disk is highly warped, but in regular rotation. The optical (active)
nucleus does not coincide with the kinematic center of the H I disk. One of
the more remarkable results is that this is also a very low surface brightness
galaxy in H I; the column densities range from a few times 10 19 to a few
times 10 20 cm- 2 • The H I extent is similar to the optical radius. The results
suggest that everywhere in the disk the H I surface density is too low to
trigger star formation (Kennicutt 1989). Similar to the galaxy in the Bootes
void, a certain amount of dark matter is needed to explain the observed
rotation curve, but not an unusually large amount. In this respect, the other
kind of gas-rich galaxies, the dwarf irregulars with very large H I extents and
high ratios of MHI/ LB, are very different. These galaxies have mass-to-light
ratios from 50 to 100 M0 / L 0 . Examples are DDO 154 (Carignan & Freeman
1988) and DDO 170 (Lake, Schommer, & van Gorkom 1990). Malin 1, like
the galaxy in the Bootes void, resides in a low-density environment, while
dwarf irregulars are found in much higher density environments (§5.1).

5. Optically unbiased searches for H I

5.1. THE LOCAL UNIVERSE

In the previous section we have seen examples of optically inconspicuous,

but gas-rich galaxies. These galaxies were almost all found by accident and
almost all were a byproduct of some optical investigation. How many of
these can there be that we do not know about? Several lines of observa-
tional and theoretical evidence suggest the existence of a large population
of faint, gas-rich dwarf galaxies. Many observed dwarfs appear to be in the
midst of star formation bursts (Kunth 1988). Similar galaxies in a quies-
cent phase should be optically faint, but gas rich. Tyson & Scalo (1988),
using self-propagating star formation models conclude that the majority of
dwarf galaxies could fall below the magnitude of surface brightness limits of
existing optical catalogs. In their models, smaller galaxies are increasingly
subject to cycles oflong dormancy and sharp bursts, the dwarfs that are seen
optically only mark the tip of the iceberg. In addition to their total num-
bers, it is also the distribution of dwarf galaxies that is of interest. Theories
of biased galaxy formation suggest that bright galaxies form preferentially
in regions of high-background mass density, so that galaxy counts give an
exaggerated picture of the underlying mass fluctuations. Many models of
biased galaxy formation (White et al. 1987; Kaiser 1988) predict that low
luminosity galaxies should be distributed more uniformly than bright galax-
 357

ies . To address these questions, optically-unbiased, blind searches directly

in the H I line are needed.
In a seminal paper, Briggs (1990) discussed the space density of low-
profile gas-rich galaxies at the present epoch. Briggs critically evaluated the
results of all major H I surveys conducted before 1990. All these surveys
were done with large single-dish telescopes, mostly with the Arecibo tele-
scope, some with the telescopes at Greenbank and Effelsberg. Almost all
consist of observations targeted at known, optically selected galaxies for the
purpose of redshift determination. For each observation a long path through
space is sampled along the line of sight, and for most an additional line
of sight near the target is sampled to calibrate the spectrometer passband.
The total bandwidth probed by the combined surveys is 347 GHz. The short
integration times that typify these surveys would not reliably detect column
densities much less than 10 20 cm- 2 • 'rhe surface density of the giant H I disk
of Malin 1 is mostly just at or slightly below that level. No object like Malin
1 was found in the entire volume of 1.5 X 10 4 h- 3 Mpc 3 • Briggs concludes
that it is unlikely that such an object will be discovered within 15 h- 1 Mpc
of the Milky Way. On the other hand, there could be several thousand closer
than the prototype at a redshift of 25,000 km S-l.
The limits on the low-mass end of the H I luminosity function are less
well constrained by these surveys, since possible detections would closely
resemble the numerous instances of weak radio interference and would almost
certainly be rejected as spurious. The most suitable survey to constrain the
H I luminosity function in the H I mass range of 106 to 10 7 M0 turned
out to be the survey by Schneider et al. (1989), which covered 16 square
degrees centered on the Leo group with a velocity coverage of 1000 km s-1
and (for discovering dwarfs most importantly) a velocity resolution of 4 km
s-l. The space density of objects in this mass range (1 object was detected)
was found to be more than a factor 10 below the prediction .of Tyson and
Scalo.
More recently, several surveys were carried out to investigate the low mass
end of the H I luminosity function and the spatial distrilmtion of gas-rich,
low-mass galaxies. Weinberg et al. (1991) used the VLA to perform a truly
unbiased search for H I. Equal volumes were searched in the Pisces-Perseus
void and supercluster. Within the supercluster, some fields w~re centered at
bright galaxies and some were centered at optically blank sky. The main re-
sult of the survey is summarized in Figure 6, showing the volumes searched
down to a given mass sensitivity and the resulting H I mass luminosity func-
tion. No galaxies were detected in the void, while 9 previously uncatalogued
objects were found in the supercluster. The result is consistent with H I
dwarfs having a large-scale distribution like that of optically bright galaxies.
Even within the supercluster, dwarfs are preferentially found near bright
galaxies; six dwarfs were found in the five fields centered at luminous galax-
358

100
r- -
,
1
..
1
J ••••
.. ' ----
................
......... ...... -
'
0.3
I .....

Q) ,///

-
S
::s
o
:> 50
0.2

,- -,
1 1
0.1

r-- --,
o , , 1,- 0 .
8 8.5 9 9.5 10

log Mm
Fig. 6. Smooth curves show the volume (in Mpc 3 marked on the left axis) in the
supercluster (solid) and void (dotted) within which a source of mass MHI could
have been detected above 5 (J". Histograms show the H I mass function of the
objects detected in the supercluster, in numbers Mpc- 3 per decade of H I mass.
Solid histograms show the mass function for the previously uncatalogued detections
(Weinberg et al. 1991).

ies, while three dwarfs were found in the seven optically blank fields. Despite
the fact that the volume searched down to 108 M0 ofH I was 10 times larger
than previous surveys, no galaxies with very small H I masses were found. G.
Hoffman et al. (1992) searched for small gas-rich dwarf galaxies just outside
the local supercluster. Although the total volume searched was very smail,
the sensitivity was unprecedented, 5 X 106 M0 of H I could easily have been
detected, no such small galaxy was found. The combined result of these sur-
veys is that galaxies with MHI below 108 M0 are less common than galaxies
of greater mass, quite contrary to the predictions of Tyson & Scalo (1988).
This is illustrated in Figure 7.
It is tempting to look for a common explanation of the lack of galaxies
with H I masses much less than 108 M0 and the absence of H I emission
at column densities less than 1019 cm -2. The intergalactic UV field is an
 359

.....- 10
It!
:::g
v
-...-
Z

8 8.5 9
log MHI

Fig. 7. Comparison of survey results to model predictions. Histograms show the

cumulative number of objects found below the indicated H I mass. The dotted curves
show the number of detections predicted from a Schechter luminosity function, solid
curves show the counts predicted by Tyson & Scalo (1988). At the low H I mass
there are many fewer H I dwarfs than predicted by Tyson & Scalo. (Weinberg et
al: 1991).

obvious suspect. Alternatively very small galaxies might not be able to hold
onto their gas, since one supernova explosion could possibly remove th.e small
H I masses left in these galaxies.
The issue of the distribution of dwarf galaxies is not settled yet. There
is mounting evidence (Salzer et al. 1990; Weinberg et al. 1991) that they
do not fill in the voids. The lore is that they cluster around bright galaxies,
and indeed one hardly finds a bright galaxy without at least one dwarf.
For example, Weinberg et al. 1991 found dwarfs or optically cat·alogued
companions near all their bright galaxies. This is true in the Hydra Cluster
as well. McMahon (1992) finds an isolated galaxy in only three out of thirteen
pointings. A typical example of a Hydra cluster galaxy surrounded by dwarfs
is shown in Figure 8. Even for the Bootes void the preliminary results of
Szomoru, Gregg, & van Gorkom (1993) indicate that 7 out of 11 galaxies
have H I detected companions. The largest numbers of dwarfs have been
360

-H.

t
-"t
SS

05~ #
i
i
!

~:l
j
I

~ :~
-i
i
n~ \

" 36 00 3' 30 00 3' 30 00

J I

IUCHT ASCENS I ON
~ • .SIk ! l ux - 9 , 0691£.02 JY / a·H/ S
!..eV$ - 1. 0000&.02· { 1. 000. 2 . 000 . l . OOO.
< . 000. ' . 000. 6 . 000. 7 . 000 . 8 . 000 '

Fig. 8. H I image of a typical Hydra cluster galaxy surrounded by dwarfs (McMahon

1992). .

found in loose groups (see Hoffman et al. [1992] for a summary). The M81
- M82 group is a champion with 23 late-type dwarfs (Tully 1987; Kraan-
Korteweg 1986). Relatively isolated dwarfs do exist, although Weinberg et
al. (1991) found three dwarfs in blank supercluster fields.

5.2. SEARCHES AT REDSHIFTS AROUND z =3

I started out this talk by saying that galaxies as we know them today might
well have consisted predominantly of gas in the distant past. In the preceding
sections it was discussed how much gas is left over in the outer parts and
in the surroundings of galaxies in the local universe. Is there a hope of
detecting galaxies in H I at redshifts where they are still mostly gaseous, Le.,
protogalaxies or protoclusters? Quite surprisingly, the answer is yes, and in
fact one example might already have been discovered. Sunyaev & Zel'dovich
(1972,1974) pointed out that one of the predictions ofthe top-down scenario
 361

Constraints on the comoving number density of HI masses at z .. 3.3

I
o llorW(1)

... VU(2)
-400<<l.V<1200
2'<8<8'

v VU(6) -1>- - - . -. -. - - -. v' -. -. -. -. -. -. ~1?'!!.1.~.11.~1~. -. -.-

o 00ty(8) a
o
I> VUe?) b--&,
1
1

ol.-----------
·· · ·.. . . . -
•••••• •••. ' 4
• 1r.IIn'(HII+8)
,.l=~~.--,.-.- !b~!!. ~?Le!~._. - - ~
• 1r.IIn'(-10dD)

,
I

.----------------

0.- 0 . 1

Fig. 9. Constraints on the number density of H I masses at z = 3.3. Number

densities and H I masses to the right and above the dashed lines are excluded by
the observations. For a detailed description see Wieringa et al. (1992).

for galaxy formation is that gas clouds with masses of order 3 X 10 14 M0

and diameters of about 2 Mpc exist at redshifts z :S 10. At redshifts of 10
these clouds would overlap, while at redshifts of 3.3 their separation would
be of order 30 Mpc. The onset of quasar emission might ionize the gas and
prevent its detection in the neutral hydrogen line. Pancake-like structures
might be observable at redshifts between 3 and 5, but the predictions of size
and ionization state of these structures is highly uncertain (Wieringa, de
Bruyn, & Katgert 1992).
Many searches have been performed for such structures at redshifts be-
tween roughly 3.3 and 5, the range being dictated by the availability of
receivers at that frequency. The more sensitive searches have all beeen done
at 330 MHz, corresponding to a redshift of about 3.3. These observations are
362

extremely difficult and most measurements have been limited by interference

and systematic errors. The two most sensitive searches to date have been
performed with the VLA by Uson, Bagri, & Cornwell (1991a) and with the
WSRT by Wieringa et al. (1992). The negative results set interesting limits
on the H I content ofthe early universe. Uson et al. (1991a) rule out the exis-
tence of hydrogen masses greater than 10 14 M0 within the surveyed volume
of 1.6 X 106 Mpc 3 , while Wieringa et al. (1992) rule out the existence of neu-
tral hydrogen masses greater than 1.6 X 10 13 h- 2 M0 at comoving densities
exceeding 1000h3 Gpc- 3 . Wieringa et al. (1992) have combined all the pub-
lished observational limits around a redshift of 3.3 to calculate limits on the
number density of neutral hydrogen masses. The result is shown in Figure
9. Number densities and H I masses to the right and above the dashed lines
are excluded by the observations. Also indicated are the estimated number
densities for Abell clusters of two richness classes. The implication is that
the percentage of R ~ 1 clusters around z = 3.3 with more than 10% of
their mass in neutral hydrogen is less than 10%. Similarly, fewer than 40%
of the R ~ 2 clusters have a hydrogen content exceeding 10% to 50% at
z = 3.3. The results indicate that no or few large neutral structures have
survived to this epoch.
Recently, evidence has been found that at least some neutral hydrogen
can be found at those redshifts. Uson, Bagri, & Cornwell (1991b) pointed the
VLA toward a radio galaxy at a redshift of z = 3.395, which was known to be
a strong Lya emitter (Lilly 1988). They detected H I in absorption against
the radio source at'the velocity of the galaxy. The column density and width
of the absorption line are typical for an edge-on spiral galaxy. The authors
point out that it could equally well arise in a number of (proto )galaxies along
the line of sight, especially since the size of the unresolved dominant radio
component is not very well constrained. The close agreement in velocity
between the galaxy and the H I absorption would be fortuitous in such a
scenario. Even more exciting, these authors report the first detection of H I
in emission from a possible Zel'dovich pancake. The emission is seen in the
same velocity channel as the absorption at a projected distance of 7 to 24
Mpc from the optical galaxy. The emission is slightly resolved, the H I mass
is of order 10 14 M 0 , the velocity width is 180 km s-l, and its linear size
about 1h- 1 Mpc. These properties lead the authors to suggest that this
might be the first example of a Zel'dovich pancake. The narrow line width
would then imply that the system might not yet have become virialized and
we are observing it during its collapse.
In Figure 10 the absorption and emission spectrum are shown. Unfortu-
nately the detected emission is very weak. The signal is barely 50", and it is
above the noise in one channel only. Although the authors have been very
careful in ruling out instrumental effects, it still remains possible that the
pancake is a prosaic noise bump. Only additional VLA observations can tell.
 363

E
g",
.
JO-

..
"0
I
,
. N

tj : ; I ; vLH-i ; .
I
.
.,c
1''''
>-
'-
...,'"
>-
E-
r -
en
I

10 20 30 10 20 30
Chonne 1 • (98 kHz/ channe 1 )
'Channe 1 If: (98 kHz I chonne 1 )

Fig. 10. At left the first detection of an H I absorption line at a redshift of 3.3. The
absorption is seen toward the radio galaxy 0902+343. At right the emission seen
33 arcmin away from the radio galaxy. The emission occurs at the same velocity as
the absorption (Uson et al. 1991) .

These will have to wait until the source is up at night time during the most
compact VLA configuration, a few years from now.

6. Quasar absorption lines: the low-z connection

One of the remarkable results presented at this conference is the close agree-
ment between the H I density at the current epoch as derived from an ex-
trapolation of the damped Lya systems (Lanzetta 1993) and from modeling
the H I distribution in a sample of nearby spiral galaxies, which have been
observed with Arecibo in the 21 cm line (Rao & Briggs 1993). I think that
the agreement might be so good because, as Lanzetta has pointed out, the
neutral hydrogen mass density is dominated by the highest column density
systems. At those column densities the nearby universe has been probed
rather well. It is only at column densities less than 10 20 cm- 2 that our
knowledge is rather incomplete, and, as was discussed before, the hydrogen
mass in these very faint envelopes is a small fraction of the total hydrogen
mass.
There has been quite some debate whether quasar absorption lines are as-
sociated with galaxies, and if so where they arise in these galaxies. The above
mentioned agreement is but one more piece of evidence that the damped
Lya systems are associated with galaxies. It seems that the evidence that
in general the metal absorbers and the damped Lya systems are associated
364

with galaxies is fairly convincing. However where the absorbing material is

with respect to the galaxies and whether these galaxies are typical for the
galaxy population as a whole, is still an open question. One way to address
this issue is to study nearby galaxies, which happen to be close in projec-
tion to the line of sight to a quasar, the so-called quasar - galaxy pairs.
The advantage is that here one can actually see the galaxies. Carilli & van
Gorkom (1992) have made an H I study offour quasar - galaxy pairs. These
pairs were selected as the only known pairs for which Ca II or Na I ab-
sorption had been detected toward the background galaxy at the velocity of
the foreground galaxy. In three of the four pairs H I is detected in emission
from the galaxy and in absorption against the radio loud quasar. A number
of interesting, though in retrospect perhaps not surprising, conclusions can
be drawn from this small data set. In each case the H I absorption arises
in extended neutral gas associated with the galaxy. In each case the gas
morphology indicates a recent gravitational disturbance. As a particularly
pathetic example in 1327-206 and ESO 1327-2041, the quasar happens to
sit behind a recently accreted polar ring around the galaxy. In all cases,
the H I seen in emission and or absorption along the narrow line of sight
toward the quasar covers a large velocity range (from 50 km S-1 to 250 km
s-1 ).The large velocity gradients are similar to those seen in high-redshift,
heavy-element quasar absorption line systems (York et al. 1986). For the
low-redshift systems, these gradients are caused by tidal interactions. These
H I results are in agreement with optical work by Womble (1993) and Bowen
(1993), presented at this conference. Nearby Ca II or Na I quasar absorbers
are associated with tidally disturbed or starbursting (Womble 1992) galax-
ies. In retrospect, this is not too surprising. The sample must be dominated
by galaxies with a large cross section in neutral gas. The H I extent of galax-
ies is a strong function of environment: galaxies in mergers, pairs, or small
groups might have cross sections in H I that are two orders of magnitude
larger than the typical isolated field spiral. Whether this tells us anything
about quasar absorption systems at high redshifts remains to be seen.
York et al. (1986) have suggested that some quasar absorption lines may
arise in gas-rich dwarfs, clustered around galaxy-like potential wells. The
absorption systems at low redshift appear to be exactly that: star-forming
clumps of gas around tidally interacting and merging galaxies.
I am leaving this conference with the feeling that the nature of the Lya
clouds is as uncertain as ever. Recent HST observations have revealed the
existence of a surprisingly large number of Lya systems at very low redshift
seen in absorption against 3C 273 (Bahcall et al. 1991; Morris et al. 1991).
Although the observed column densities of these systems are low, between
1014 and 1016 cm- 2 , their nearness makes them ideal targets for searches for
H I and optical emission from the vicinity of these clouds, in hope of learning
more about the parent population of the unsaturated Lya absorbers. One
 365

such search was recently made with the VLA in H I (van Gorkom et ai.
1993). In an entire cube of 40 X 40 arcmin centered at 3C 273 and covering a
velocity range from 840 km s-1 to 1840 km s-1, which includes the velocities
of two Lya clouds, no H I was found at all. Small H I clouds down to a few
times 106 M0 could have been detected.

Acknowledgements

It is a pleasure to thank Harley Thronson and Mike Shull for a very stimu-
lating meeting. I thank Staszek Bajtlik, Albert Bosma, Frank Briggs, Chris
Cariili, John Hibbard, Ken Lanzetta, Phil Maloney, Rob Oiling and Sandya
Rao for illuminating discussions and Phil Maloney for letting me use results
in advance of publication. This work was supported through NSF grant
90-23254 to Columbia University.

References
Bahcall, J.N ., Jannuzi, B.T., Schneider, D.P., Hartig, G.F., Bohlin, R., & Junkkarinen, V.
1991, ApJL, 377, L5
Barnes, J .E., & Hernquist, L. 1992, Nature 360, 715
Bosma, A. 1978, Ph.D. thesis, University of Groningen
Bothun, G.D., & Aldering, G. 1988, BAAS, 20, 1087
Briggs, F.H. 1990, AJ, 100, 999
Briggs, F.H., Wolfe, A.M., Krumm, N., & Salpeter, E.E. 1980, ApJ, 238, 510
Broeils, A. 1992, Ph D thesis, University of Groningen
Carignan, C., & Freeman, K. 1988, ApJL, 332, L33
Carilli, C.L., & van Gorkom, J.H. 1992, ApJ, 399, 373
Cayatte, V., Kotanyi, C, Balkowski, C., & van Gorkom, J.H. 1993, AJ, in press
Cayatte, V. van Gorkom, J.H., Balkowski, C., & Kotanyi C. 1990, AJ, 100, 604
Corbelli, E., & Salpeter, E .E. 1993, preprint
Corbelli, E., Schneider, S.E., & Salpeter, E.E. 1989, AJ, 97,390
Dey, A., Strauss, M.A., & Huchra, J.P. 1990, AJ, 99, 463
de Vaucouleurs, G., de Vaucouleurs, A., & Corwin, H.G., Jr. 1976 Second Reference Cat-
alogue of Bright Galaxies (University of Texas, Austin)
Giovanelli, R ., & Haynes, M.P. 1988 in Galactic and Extragalactic Radio Astronomy, eds.
G.L. Verschuur and K.I. Kellermann (Springer-Verlag), p 522
Haynes, M.P., Giovanelli, Giovanelli, R., & Chincarini, G.L. 1984, ARAA, 22,445
Hibbard, J.E. 1993, in The Evolution of Galaxies and their Environment, Contributed
papers, eds D. J. Hollenbach, J .M. Shull & H.A Thronson (NASA-CP), in press
Hoffman, G. L., Lu, N.Y., & Salpeter, E.E., 1992, AJ, 104, 2086
Hoffman, Y., Silk, J., & Wyse, R.F.G. 1992, ApJL, 388, L13
Huchtmeier, W.K., & Richter, O.-G. 1989, AA, 210,1
Impey, C., & Bothun, G.D. 1989, ApJ, 341, 89
Kaiser, N. 1988, in Large Scale Structures of the Universe, eds J. Audouze & A. Szalay
(Reidel), p4
Katz, N. 1992, ApJ 391, 502
Kaufman, M., Elmegreen, B.G., & Thomasson, B. 1993, in The Evolution of Galaxies
and their Environment: Contributed papers, eds D. J. Hollenbach, J.M. Shull & H.A
Thronson (NASA-CP), in press
Kennicutt, R. 1989, ApJ, 344, 685
366

Kenney, J.P.D. 1990 in The Interstellar Medium in Galaxies, eds H.A Thronson & J.M .
Shull (Kluwer), p 151
Kirshner, R.P., Oemler, A., Schechter, P.L., & Shectman, S.A. 1987, ApJL, 248, L57
Kraan-Korteweg, R.C . 1986, AA Suppl, 66, 255
Kulkarni, S.R., & Heiles C. 1988 in Galactic and Extragalactic Radio Astronomy, eds. G.L.
Verschuur & K.I. Kellermann (Springer-Verlag), p 95
Kunth, D. 1988, in Evolutionary Phenomena in Galaxies, eds. J.E. Beckman & B.E.J.
Pagel (Cambridge University Press), p22
Kutyrev, A.S., & Reynolds, R.J. 1989, ApJL, 344, L9
Lake, G., Schommer, R.A., & van Gorkom, J.H. 1990, AJ, 99, 547
Lanzetta, K .M, 1993, this conference
Lilly, S.J. 1988, ApJ, 333, 161
Maloney, P. 1990, in The Interstellar Medium in Galaxies: Contributed Papers, eds D.J.
Hollenbach & H.A. Thronson, Jr. (NASA- CP3084), pI
Maloney, P. 1992, ApJL, 398, L89
Maloney, P. 1993, ApJ, 414, in press
McMahon, P.M. 1992, PhD Thesis, Columbia University
Moody, J.W., Kirshner, R.P., MacAlpine, G.M., & Gregory, S.A. 1987, ApJL, 314, L33
Morris, S.L., Weymann, R.J., Savage, B.D., & Gilliland, R.L. 1991, ApJL 377, 'L21
Rao, S., & Briggs, F.H. 1993, in The Evolution of Galaxies and their Environment, Con-
tributed papers, eds D. J. Hollenbach, J.M. Shull & H.A Thronson (NASA-CP), in
press
Reynolds, R.J. 1987, ApJ, 323, 553
Reynolds, R.J., Magee, K., Roesler, F.L., Scherb, F., & Harlander J . 1986, ApJL, 309, L9
Salzer, J.J, Hanson, M.M., & Gavazzi, G. 1990, ApJ, 353, 39
Sancisi, R. 1983, in Internal Kinematics and Dynamics of Galaxies, ed E. Athanassoula
(Reidel), p 55
Sancisi, R. 1990 in Windows on Galaxies, eds G. Fabbiano, J .S. Gallagher & A. Renzini
(Kluwer), p 199
Schneider, S.E., et al., 1989, AJ 97, 666
Silk, J ., & Sunyaev, R.A. 1976, Nature, 260, 508
Songaila, A., Bryant, W., & Cowie, L.L. 1989, ApJL, 345, L71
Stocke, J.T., Case, J., Donahue, M., Shull, J.M., & Snow, T.P. 1991, ApJ, 374, 72
Sunyaev, R.A. 1969, ApJL, 3, L33
Sunyaev, R.A., & Zel'dovich, Ya.B., 1972, AA, 20, 189
Sunyaev, R.A., & Zel'dovich, Ya.B., 1974, MNRAS, 171, 375
Szomoru, A., Gregg, M., & van Gorkom, J.H . 1993, in preparation
Szomoru, A. van Gorkom, J.H., Gregg, M., & de Jong, R.S. 1993, AJ, 105, 464
Tully, R.B. 1987, ApJ, 321, 280
Tyson, N.D.T., & Scalo, J .M. 1988, ApJ, 329, 618
Uson, J.M, Bagri, S., & Cornwell, T .J. 1991a, ApJL, 377, L65
Uson, J.M, Bagri, S., & Cornwell, T.J. 1991b, Phys Rev Lett, 67, 3328
van Gorkom, J.H., van Albada, T .S., Cornwell, T.J ., & Sancisi, R. 1993, in preparation
van Gorkom, J.H., Bahcall, J.N., Jannuzi, B.T., & Schneider, D.P. 1993, in preparation
Warmels, R.H. 1986, PhD Thesis, University of Groningen
Warmels, R .H . 1988a, AA Suppl, 72, 19
Warmels, R.H. 1988b, AA Suppl, 72, 57
Weinberg, D.H., Szomoru, A., Guhathakurta, P., & van Gorkom, J.H., 1991, ApJL, 372,
L13
White, S.M.D., Davis, M., Efstathiou, G., & Frenk, C.S. 1987, Nature, 330, 451
Wieringa, M.H ., de Bruyn, A.G., & Katgert, P. 1992, AA 256, 331
Williams, B.A., McMahon, P.M., & van Gorkom, J.H. 1991, AJ, 101, 1957
Womble, D.S., 1992, PhD Thesis, UCSD
Womble, D.S. 1993, in The Evolution of Galaxies and their Environment, Contributed
papers, eds D. J. Hollenbach, J.M. Shull & H.A Thronson (NASA-CP) , in press
 367

York, D .G., Dopita, M., Green, R., &. Bechtold, J. 1986, ApJ, 311, 610
Yun, M.S., Ho, P.T.P. Brouillet, N., &. Lo, K.Y. 1993, in The Evolution of Galaxies and
their Environment: Contributed papers, eds D. J. Hollenbach, J.M. Shull, &. H.A
Thronson (NASA-CP), in press
 ENERGY INPUT FROM AGNs

MITCHELL C. BEG ELMAN

Joint Institute for Laboratory Astrophysics, University of Colorado and National
Institute of Standards and Technology, Boulder, CO 80309-0440

Abstract. Active galactic nuclei pump substantial amounts of energy into their sur-
roundings, in several forms. The direct action of radiation pressure, radiative heating of
surrounding gas, and kinetic energy input through jets and winds may all play a role
in modifying the environments of AGNs. In this article, I review some of the important
effects of these energy inputs.
Key words: galaxies: intergalactic matter - galaxies: jets - galaxies: nuclei - galaxies:
Seyfert - interstellar: matter - quasars - radio sources: extended - hydrodynamics

1. Overview

Active galactic nuclei (AGNs) emit large quantities of energy in several dif-
ferent forms. In addition to producing an electromagnetic radiation output
which can be spread over eight or more decades in frequency, they are often
sources of powerful - and fast - winds and jets. Each of these forms of
energy output can have a distinctive effect on the matter surrounding the
AGN. In this article I will describe some of the important ways in which
these various forms of energy can modify an AGN's environment, and I
outline some possible observational consequences.

1.1. DEPENDENCE ON FORM OF ENERGY INPUT

The effects produced by the energy output from an AGN depend largely on
the mechanisms by which the energy couples to the gas. For example, radia-
tion pressure can exert a force on the gas via electron scattering, scattering
and absorption on dust, photoionization, or scattering in atomic resonance
lines (§ 2). The penetration depth of the radiation and the effective band-
width in which the force is exerted, and hence the force per unit mass acting
on the gas, depend on which mechanisms dominate. Similarly, the outcome
of radiative heating depends on value of the "ionization parameter" (Krolik,
McKee, & Tarter 1981) . If it is low, a thermostatic effect will keep the tem-
perature at ~ 10 5 K. A high ionization parameter will lead to runaway heat-
ing by the X-ray portion of the AGN spectrum, with very different thermal
and dynamical consequences (§ 3). Interestingly, the value of the ionization
parameter, and hence the outcome of radiative heating, can depend in part
on the effects of radiation pressure.
The energy carried by jets and winds couples to the circumnuclear gas
mainly by driving shocks into it (§ 4). Shocks convert a fraction of the in-
cident kinetic energy flux into internal energy, of either thermal gas alone
or a mixture of thermal gas and suprathermal "cosmic ray" particles (if

369
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, 369- 382.
© 1993 Kluwer Academic Publishers.
370

Fermi acceleration is efficient at the shock front). Many factors determine

what fraction ' of the "randomized" energy will be recovered as kinetic en-
ergy through adiabatic expansion, how much will be radiated away, and
how much will leak away through cosmic-ray diffusion. Agns may also emit
"exotic" particle outflows, consisting, e.g., of relativistic neutrons and neu-
trinos (Begelroan, Rudak, & Sikora 1990). These particles tend to deposit
their energy in a volume-distributed fashion, rather than in impulsive fash-
ion at a shock front. To predict the outcome of this form of e:nergy injection,
one needs to know its magnitude and spatial distribution. A discussion of
the specific effects of these speculative forms of energy release is beyond the
scope of this paper, except insofar as these particles drive winds of ordinary
plasma (e .. g., Begelman, de Kool, & Sikora 1991).

1.2. DEPENDENCE ON AMBIENT MEDIUM AND GEOMETRY

The effects of energy input also depend crucially on the properties of the
gas in which the energy is being deposited. A wide range of pressures and
densities characterize the regions into which AGN energy is propagating.
We may reasonably expect these conditions to be loosely correlated with
distance from the nucleus and with such observables as morphological type
of the host galaxy and galaxy cluster environment. Thus, we might guess that
radio-quiet AGNs such as Seyferts lie in otherwise typical spirals having an
interstellar medium (ISM) pressure rv 10- 12 _10- 13 dyne cm- 2 at a distance
of several kpc from the nucleus, as is the case for the ISM in the 'Milky Way.
Radio-loud AGNs, which apparently lie in giant elliptical galaxies, might
be expected to have the correspondingly higher ISM pressures associated
with the latter (rv 10- 10 dyne cm- 2 ), whereas sources lying at the centers of
cluster cooling flows would have higher pressures still (rv 10- 9 - 10- 8 dyne
cm- 2 ).
In each of these cases, it is likely that the pressure is considerably higher
towards the nucleus, although it is risky to speculate by how much. As a
further caution, it should be remembered that the central regions of lumi-
nous active galaxies may bear little resemblance to morphologically similar
galaxies closer to home. In particular, the most impressive of AGNs, such as
quasars and powerful radio galaxies, may reside in galaxies at a particularly
active stage or shortly after an encounter with another galaxy. It is harder
to characterize the density in each case, since the ISM is likely to be highly
inhomogeneous with distinct thermal "phases" (e.g., Jura 1987). Whereas
the mean density of ISM in the Milky Way is roughly (n) '" 0.1 - 1 cm- 3 ,
a more useful description would distinguish the hot halo «(n) '" 10- 2 cm- 3 )
from the atomic and molecular gas comprising the disk «(n) '" 1 cm- 3 ),
and further would note the concentration of a large fraction of ISM mass in
extremely dense giant molecular clouds «(n) rv 103 cm- 3 ) . In giant ellipti-
 371

cals and cluster cooling flows, the characteristic mean densities on few-kpc
scales might be '" 0.1 cm- 3 and 1 cm- 3 , respectively, but the phase struc-
tures are virtually unknown. Finally, the energetic influence of an A G N may
extend all the way out into the intergalactic medium (IGM), where results
from COBEhave yielded only upper limits on the pressure (;S 2 X 10- 15 dyne
cm- 2 : Mather et al. 1990; Bennett et al. 1993) and essentially no constraints
on the density.
The importance of geometric effects cannot be overemphasized. The speed
of a shock propagating through a medium with a "cloudy" phase structure
will be highest in the phase with the lowest density (the inter cloud medium).
Although clouds will be compressed after passage of the shock, and may be
subject to hydrodynamic instabilities which tear them apart (Klein, McKee,
& Colella 1990), they will basically be overrun and left behind by the shock
front. This effect was first noted by McKee & Ostriker (1977), in connection
with supernova blast waves propagating into the ISM (Figure la). As a
result, most of the energy carried by the blast wave goes into the gas which
has the lowest density (and is the hottest) to begin with. For radiative forms
of energy input, shadowing effects in a cloudy medium can also be important.
The global geometric structure of the ambient gas is important as well. Since
a wind or hot bubble emanating from an AGN will tend to follow the "path
of least resistance", a disk-like ISM structure (e.g., in a spiral galaxy) can
lead to a "blowout" of the hot gas along the disk axis (Figure Ib). A more
spherically symmetric gas distribution will tend to keep the AGN energy
confined in a "bubble".
To make matters more complicated, there is overwhelming evidence that
the energy output from many AGNs is anisotropic. We have known about
the existence of jets for many years; now there is strong evidence for colli-
mated UV radiation, from both statistical arguments and studies of individ-
ual objects. Such anisotropy is an essential part of schemes which attempt to
"unify" different classes of active galaxies, such as Seyfert 1 's with Seyfert
2's (Antonucci & Miller 1985), Fanaroff-Riley class II radio galaxies with
radio-loud quasars (Barthel 1989), and Fanaroff-Riley class I radio galaxies
with BL Lac objects (Padovani & Urry 1990,1991; Urry, Padovani, & Stickel
1991). A variety of observational arguments support the existence of radia-
tion anisotropy, including direct imaging of "conical" extended emission line
regions, "photon-counting" arguments, alignments of emission line regions
with large-scale radio structure, and spectropolarimetric evidence (Wilson
1993). The most suggestive examples are morphological, and include HST
imaging of the conical emission line region in NGC 1068 (Evans et al. 1991),
observations of photoionized filaments aligned with the radio lobes of Cen A
(Morganti et al. 1991), and numerous examples of the radio-optical "align-
ment effect" in high-redshift radio galaxies (McCarthy 1993). However, the
strongest evidence for anisotropy is provided by photon-counting and spec-
372

~ (a)

-BLOWOUT"
(b)

Fig. 1. (a) Depiction of a shock front propagating through a cloudy medium,

adapted from McKee & Ostriker (1977). The shock overruns and engulfs the dense
clouds, which are compressed in the postshock region. (b) The global geometry
of the ambient medium will determine whether the energy in the AGN wind is
confined to a "bubble" or "blows out" through the poles.
 373

tropolarimetric arguments. Interpretation of the radio-optical ali~nment ef-

fect remains highly controversial. Several competing theories to explain the
effect are in play, including star formation triggered by the radio source (Mc-
Carthyet al. 1987; Chambers, Miley, & van Breugel1987) and anisotropy of
the ambient gas pressure or density (Eales 1992), in addition to scattering of
anisotropic UV continuum (Tadhunter, Fosbury, & di Serego Alighieri 1988;
Fabian 1989).

2. Effects of Radiation Pressure

Having been forewarned of the complexitles of our subject in the preceding
section, let us now consider each of the important energy input mechanisms
in its simplest abstraction. We start with the radiation pressure force which,
as noted earlier, can couple to the gas through several mechanisms. The
simplest of these to treat is electron scattering, for which the cross section
per hydrogen atom is (0) H) '" 7 X 1O- 25 z cm 2; where z is the ionized
fraction. The maximum column density over which the force can be exerted
is given by NH,maz '" (qjH)-1 '" 2 X 10 24 z- 1 cm- 2.
If the radiation flux from the nucleus does not greatly exceed the Ed-
dington limit of the central black hole, we can generally assume that the
radiation force exerted through electron scattering will have a relatively
minor dynamical effect on the gas, compared to gravitational and thermal
pressure forces. In contrast, radiation pressure acting on dust can exert a
much larger force per H atom, though over a correspondingly smaller column
density. If we assume a dust-to-gas ratio (by mass) of 0.01, a typical grain
size a, and a total cross section per grain of the same order as the geometric
cross section (a reasonable assumption for UV and soft X-ray photons hit-
ting'" 0.1 J.lm grains), then the cross section per H atom exceeds that for
electron scattering in fully ionized gas by a factor of order 105( a/O.l J.lm)-1.
The force exerted per particle is higher by the same factor, but the column
density affected is only 2 X 10 19 (a/0.1 J.lm) cm- 2 •
The force exerted on ;S 10 5 K gas as a result of steady-state photoioniza-
tion and recombination is characterized by the mean cross section (q / H) rv
10- 18 (1 - z) cm 2 • Photoionization equilibrium predicts that (1 - x) rv
10- 4 Pgas/Prad, where Pgas and Prad are the gas pressure and pressure ofioniz-
ing radiation, respectively. (Strictly speaking, one should use 47rJ /e instead
of Prad, where J is the mean intensity. However, use of Prad is quantitatively
correct and promotes a more physically intuitive discussion.) Thus, we can
write (q / H) '" 1O- 22Pga./Prad cm2. Note that Pgas/Prad is just the reciprocal
of the "ionization parameter" B defined by Krolik et al. (1981). The force
per H atom exerted through photoionization is 200Pgall/Prad times greater
than that exerted through electron scattering.
The largest forces per H atom are possible through scattering in UV
374

resonance lines. For species i, the effective cross section is

(1)

where t::..VD is the Doppler width of the line and Ai, Xi, and fi are, re-
spectively, the abundance of element i, the fraction of the element in the
relevant ionization state, and the oscillator strength of the transition. For
important resonance lines such as those of C IV, Si IV, and N V, AiXdi can
attain values of order 10- 4 for cosmic abundances. The fractional Doppler
width, at T '" 10 4 K, is t::..VD/V '" 10- 4 ; hence, we find that the mean cross
section can be as large as 10- 17 cm2 and the corresponding force as much
as seven orders of magnitude larger than electron scattering. The draw-
back is that the bandwidth over which resonance-line scattering is effective
is extremely small, fractionally of order '" 10- 4 of the ionizing spectrum
for each strong resonance line. The aIllount of momentum available from
such a small bandwidth is very small. Therefore, in order for resonance-line
scattering to be dynamically important, the gas must accelerate, so that
new portions of the spectrum are continuously Doppler shifted into the line.
This is, of course, the essence of the Sobolev approximation, which has been
exploited to construct models of winds from hot stars (Castor, Abbott, &
Klein 1975). Resonance-line scattering may playa role in the acceleration
of the fast (v ----+ O.le) outflows in broad absorption line QSOs (BALQSOs);
however, a discussion of these models is beyond the scope of this review.
For AGN radiation acting on general interstellar matter, the radiation
pressure force will be exerted mainly through dust and photoionization.
Dynamical effects can be significant if Prad > PISM, where PISM is the
pressure in the undisturbed gas and

(2)

for gas situated Rkpc kpc from an isotropic source of ionizing radiation with
luminosity 1046 L46 erg s-1. The maximum distance over which radiation
pressure can be important, under various "typical" interstellar conditions,
is shown in Figure 2a as a function of L. To overcome the gravitational po-
tential of the galaxy requires a mean cross section per H atom which exceeds
that for electron scattering by a factor'" 10 3 L 461 (M/10 11 M 0 ), where Mis
the mass contained within radius R. Gas can be accelerated smoothly by
radiation pressure if the column density is smaller than the maximum pen-
etration depth for the radiation responsible for acceleration, N H < N H,maa:'
If N H exceeds N H,ma:J;, the radiation force will tend to drive shocks into the
circumnuclear gas, as described by Chang, Schiano, & Wolfe (1987).
 375

RADIATION FORCE ./
47
Prad> Pgas

46

45
~

'(I)

....
Cl
II>
44
~

...J
Cl
0
43

42 Prad < Pgas

41
lOpe 100pe 1 kpe 10kpe 100kpe
log R

47 WIND BUBBLE
Pbubble > PISM

46

~ 44
..J
Cl
o
- 43

42

41 L -_ _ _ _ _ _ _ _~ _ _ _ _ _ _ _ _~ _ _L __ _ _ _ L __ _~~~L__ __

lOpe 100pe 1 kpe 10kpe 100kpe

log R

Fig. 2. Conditions under which: (a) radiation pressure; and (b) kinetic energy output
are dynamically important, for various interstellar parameters. In panel (a), the
diagonal lines denote Prad = Pga. for three different values of interstellar pressure.
In panel (b), they denote Pb = PISM for typical combinations of ISM pressure and
density. For the same power output L, kinetic energy is more effective in disturbing
the ambient medium because it is trapped inside the bubble, and most of it is used
to do work on the surrounding gas. In contrast, only a small fraction of the radiation
energy does work on the ambient medium.
376

The maximum column density in a slab of gas that can be fully ionized
by AGN continuum is given by

.
N H,1on rv 1022ln (1 + P.,.ad) cm-2 , (3)
PISM
if dust absorption is neglected, and only slightly smaller if a cosmic dust
abundance is taken into account. If PISM < P.,.ad, then N H,ion gives the
depth to which an irradiated cloud will be "pressurized" by the radiation.
Differential pressure forces between the front and back of a cloud can lead to
a "pancake effect" (Mathews 1982) which tends to squash irradiated clouds
down to a column density of order N H,ion .

3. Radiative Heating
Gas exposed to ionizing radiation from an AGN tends to undergo an abrupt
transition from the typical H II region temperature rv 104 K to a higher
temperature and ionization state when PgaIJ/Prad falls below some critical
value, which lies in the range'"" 0.03-0.1 (McCray 1979; Krolik et al. 1981).
The "hot-phase" equilibrium temperature is close to the temperature at
which Compton cooling balances inverse Compton heating, which is typically
TIC ~ 10 7 K for AGN spectra (Figure 3a). Clouds with column densities
~ N H,ion will not heat up all at once. Only the surface layers will be ablated,
and the backpressure of the ablated gas will keep the cloud interior at a high
enough pressure to avoid immediate heating (Begelman, McKee, & Shields
1983; Begelman 1985). The structure of a cloud subject to runaway X-ray
heating is depicted in Fig. 3b. Depending on the size of the cloud and the
times cales involved, the heated gas may reach a temperature TIJ '"" 10 5 -106
K at the point at which it becomes supersonic with respect to the cloud
surface. A critical condition at the sonic point determines the mass flux per
unit area, which is proportional to Prad/TIJ1/2,

N'"" (6 X 106 - 6 X 10 7 ) ~:6 cm- 2 S-1 • (4)

kpc

The lifetime of a cloud against ablation by X-ray heating is given by

N 6 7 R~pc NH
-;- '"" (5 X 10 - 5 X 10 ) - L 22 _ 2 yr , (5)
N 46 10 cm
which corresponds to a global ablation rate of
• 1
Mabl '"" (20 - 200)CL 46 M0 yr- , (6)
for gas with a covering factor C. The effects of X-ray heating on the evolu-
tion of the ISM in a spiral galaxy were studied by Begelman (1985) and by
 377

(a)

,,
i
T
\
\
\
\
,, - STABLE

, --- UNSTABLE
----'"

-10 - 30 (DEPENDS ON
AGN SPECTRUM)

(b)

~ ~

X-RAY HEATED
~
T < 10 K/
ATOM~
~
T-10 5 -10 6 K /~OR/ :%
MOLECULAR
~ ~
W$fi
Fig. 3. (a) Schematic illustration of the thermal equilibrium curve for gas exposed
to unattenuated AGN radiation. The two stable temperature levels are separated by
a thermally unstable transition. The :l-axis represents the "ionization parameter"
::: used by Krolik et al. (1981). (b) Structure of a thick (NH > 10 22 cm- 2 ) cloud
undergoing runaway heating due to irradiation by AGN continuum. Ablation of the
X-ray heated gas pressurizes the interior, allowing an H II zone to exist in thermal
equilibrium.

Shanbhag & Kembhavi (1988). Possible observable consequences of X-ray

heating include: 1) elimination of cool ISM phases from the inner parts of
the galaxy, which would allow UV radiation to penetrate to much larger
distances than would otherwise be possible; 2) modification of ISM phase
structure by preferential destruction of small clouds. Only clouds (e.g., gi-
ant molecular clouds) with a large enough column density would survive
long enough to be observed; and 3) generation of peculiar cloud velocities
~ 100 km s-1, via the "rocket effect", with a random component intro-
duced through cloud-cloud shadowing. The importance of these effects is
highly uncertain as they depend critically on the geometric distribution of
the dense phases of the ISM, and on the replenishment of ISM through
stellar evolutionary and other processes.
378

4. Kinetic Energy Input

4.1. BUBBLES VS. CIGARS
The principal effect of a wind or a pair of jets from an AGN is to inflate a
"bubble" in the surrounding gas, similar to that produced by a hot stellar
wind propagating into the ISM (Castor, McCray, & Weaver 1975). To zeroth
order, the evolution of the bubble can be described by a self-similar model
in which the internal and kinetic energy are comparable, and share the
integrated energy output of the wind: Lwt PbR3 pv 2R 3, where Lw is
r'V r'V

the wind power, Pb is the gas pressure inside the bubble, p is the ambient
density, t is the elapsed time, v is the speed of the shock front bounding the
bubble, and R is the radius of the shock. These scaling relations imply that

R (L;)
r'V 1/5 t2/5 , (7)

and
2/3
Pb r'V
2/ 3P2/3R-4/3 ,...., 10-6 n 1/3
LW ( Lw,46 )
-2- dyne cm- 2 , (8)
R kpc
from which we conclude that

Pb
Prad r'V
(Lw)
Lrad
(C)
:;; ,
(9)

where Lrad is the power output in radiation. Equation (9) implies that the
pressure inside the bubble can exceed the radiation pressure if Lw f Lrad >
v f c. This has the immediate consequence that clouds overrun and com-
pressed by the bubble may be spared the effects of runaway X-ray heating,
even if the gas pressure is much lower than Prad outside the bubble. More-
over, the numerical estimate of equation (8) indicates that Pb could easily
exceed typical ISM pressures, even many kpc from the nucleus. This is il-
lustrated in Figure 2b.
The stellar wind bubble analogy has to be modified to treat kinetic energy
injection by jets. The jets in powerful (Fanaroff-Riley class II) radio sources
are thought to be highly supersonic and to have low densities compared to
the surrounding medium. This is sufficient to ensure that they will inflate a
large "cocoon" containing most of the energy output (Scheuer 1974), which is
analogous to the bubble described above. However, the surrounding medium
is also pushed aside by the momentum deposited by the jets. In the picture
originally developed by Scheuer (1974) and Blandford & Rees (1974), the
cocoon has a cigar-like shape. The rate at which it lengthens (Vi = dlfdt) is
determined by the momentum deposited at the ends of the jets, and is given
by Vi (Ljf pVjA)1/2, where Lj and Vj are the power and speed of the jet
r'V
 379

and A is the area over which momentum is deposited. The rate at which the
cocoon widens, VU} = dw / wt, is governed by the pressure according to
2 Ljt
pVU} '" Pcocoon rv w 2[ • (10)

The shape of the cocoon is determined mainly by the area over which the
jet momentum is deposited. If this were restricted to the size of the intense
"hotspots" which apparently mark the instantaneous impact points of the
jets, then the momentum deposition would be highly concentrated and the
cocoons would be long and narrow. However, recent observations (Laing
1989; Carilli, Perley, & Dreher 1988) support Scheuer's (1982) conjecture
that the ends of powerful jets actually "wiggle" inside the cocoon, spreading
the momentum (in a time-averaged sense) over an angle () ~ 0.1 rad. This
tends to make the cocoons resemble spherical bubbles more than cigars,
according to the criterion for sphericity derived by Begelman & Cioffi. (1989) :

Lj,46)1/lO( t
() > 0.1 ( - - --8--
)-2/5 rad, (11)
n-2 10 yr
where n-2 is the ambient particle density in units of 10- 2 cm- 3 • Observa-
tionally, there appears to be a wide range of cocoon shapes (Leahy & Perley
1991).
The high pressures and low densities predicted to exist inside cocoons
may have a number of observable consequences. First , the high pressure can
provide lateral confinement for the jets. In many cases, observations of syn-
chrotron surface brightness indicate minimum pressures which exceed that
of the static intergalactic medium (Potash & Wardle 1980; Burns et a1. 1984;
Perley, Dreher, & Cowan 1984). Second, the low-density bubble might insu-
late the jet against disruptive hydrodynamic instabilities. The presence or
absence of a protective cocoon might be crucial to the difference between
Fanaroff- Riley type I and type II radio sources. Third, the pressurization
of the circumgalactic medium by an elongated cocoon, and resulting star
formation , provides one promising mechanism for explaining the alignment
effect in high-redshift radio galaxies (Begelman & Cioffi. 1989; Rees 1989;
DeYoung 1989; Daly 1990).

4.2 . WINDS FROM RADIO-QUIET AGNs

To date, most work on large-scale outflows from AGNs has concentrated
on radio-loud objects. But there is strong evidence that radio-quiet AGNs,
particularly QSOs, also produce powerful outflows, despite the fact that
these do not create bright radio lobes and hotspots. The strongest evidence
comes from observations of blueshifted broad absorption lines (BALs), with
velocities ranging up to rv 0.2c relative to line center, in about 3-10% of
380

all QSOs (Turnshek 1988; Weymann, Turnshek, & Christiansen 1985). The
absence of redshifted P-Cygni emission indicates that the covering fraction
of the absorbing gas is typically;:; 10%, implying that virtually all radio-
quiet QSOs would show BALs when viewed from certain directions. The
kinetic energy flux carried by BAL winds is estimated to be at least a few
percent of the QSO luminosity.
Despite their apparent ubiquity, the place of BAL outflows in the phe-
nomenology of AGNs is unclear. There is some evidence that BAIJs occur
in lower-luminosity, radio-quiet AGNs, i.e., Seyfert galaxies (Shull & Sachs
1993), as well as in QSOs. There is a very strong anti-correlation between the
presence of BALs and high radio luminosity in QSOs (Stocke et al. 1992).
Yet X-ray broad absorption lines in highly ionized species (e.g., 0 VIII)
have been observed in some BL Lac objects, which constitute a radio-loud
class of AGNs (Canizares & Kruper 1984; Madejski et al. 1991). It is not
known whether these two types of BAL outflows arise through similar pro-
duction mechanisms. One problem is that there is still no good working
model for the acceleration of BAL winds. Remarkably, most of the acceler-
ation appears to take place outside the broad emission line region, i.e., at
radii ;<: 1 pc for a luminous QSO (Turnshek 1988; Turnshek et al. 1988).
The presence of numerous broad absorption troughs, including many lines
in the extreme ultraviolet band (Korista et al. 1992), makes acceleration by
radiation pressure due to resonance line scattering an attractive possibility
(see § 2). But there are strong arguments that the absorbing gas is confined
to clouds or filaments which occupy only a small fraction (;:; 10- 4 ) of the
volume in the outflow. Resonance-line acceleration has great difficulty ac-
counting for the high pressure in the intercloud region, which is necessary to
prevent the absorbing gas from becoming too highly photoionized. Evidently
some additional form of pressurization, which also presumably contributes
to the acceleration, must be operating. The high pressure must be attained
without too high a density in the intercloud medium, or else the mass and
energy fluxes become prohibitively high. Magnetic stresses provide a very
promising mechanism for producing a low-density, high-pressure, fast wind.
They are currently the leading mechanism for driving jets in extragalactic
radio sources (Blandford 1993), and have been invoked both to accelerate
and to confine broad emission line clouds (Emmering, Blandford, & Shlos-
man 1992). A more speculative mechanism, involving energy deposition by
ultrarelativistic free neutrons streaming away from the central engine, has
the attractive feature that it naturally produces the acceleration at the right
distance from the nucleus (Begelman, de Kool, & Sikora 1991).
Whatever the origins of BAL outflows, they are clearly carrying large
quantities of kinetic energy into the regions sutrounding the AGN, yet they
do not produce large, diffuse regions of radio emission. What becomes of
this energy is an open question. If BAL winds drive shocks into the ISM,
 381

one might be able to see the associated line emission. The "satellite" lines
of Mg II observed by Granados et al. (1993) may be an example of this.
Regions of diffuse (~ 1 kpc) X-ray emission around Seyfert nuclei (Elvis et
a1. 1990; Wilson et al. 1992) may also reflect the dissipation of BAL outflows.

Acknowledgements

This work was supported in part by NSF grants AST88-16140 and AST91-
20599, and NASA Astrophysical Theory Center grant NAGW-766.

References
Antonucci, R. R. J., &£ Miller, J. S. 1985, ApJ, 297, 621
Barthel, P. D. 1989, ApJ, 336, 606
Begelman, M. C. 1985, ApJ, 297,492
Begelman, M. C., &£ Cioffi, D. F. 1989, ApJ, 345, L21
Begelman, M. C., de Kool, M., &£ Sikora, M. 1991, ApJ, 382, 416
Begelman, M. C., McKee, C. F., &£ Shields, G. A. 1983, ApJ, 271, 70
Begelman, M . C., Rudak, B., &£ Sikora, M. 1990, ApJ, 362, 38 (Erratum: 370, 791 [1991])
Bennett, C. L. et a1. 1993, this volume
Blandford, R. D. 1993, in Space Telescope Sci. Inst. Symp. 6, Astrophysical Jets,
ed. D. Burgarella, M. Livio, &£ C. O'Dea (Cambridge: Cambridge Univ. Press), in
press
Blandford, R.D., &£ Rees, M. J. 1974, MNRAS, 169, 395
Burns, J. 0., Basart, J. P., DeYoung, D. S., &£ Ghiglia, D. C. 1984, ApJ, 283,515
Canizares, C. R., &£ Kruper, J. 1984, ApJ, 278, L99
Carilli, C. L., Perley, R. A., &£ Dreher, J. H. 1988, ApJ, 334, L73
Castor, J. I., Abbott, D. C., &£ Klein, R. I. 1975, ApJ, 195, 157
Castor, J., McCray, R., &£ Weaver, R. 1975, ApJ, 200, LI07
Chambers, K. C., Miley, G. K., &£ van Breugel, W. J. M. 1987, Nature, 329, 604
Chang, C. A., Schiano, A. V. R., &£ Wolfe, A. M. 1987, ApJ, 322, 180
Daly, R. A. 1990, ApJ, 355, 416
DeYoung, D. S. 1989, ApJ, 342, L59
Eales, S. A. 1992, ApJ, 397, 49
Elvis, M., Fassnacht, C., Wilson, A. S., &£ Briel, U. 1990, ApJ, 361, 459
Emmering, R. T., Blandford, R. D., &£ Shlosman, I. 1992, ApJ, 385, 460
Evans, N. I., Ford, H. C., Kinney, A. L., Antonucci, R . R. J., Armus, L., &£ Caganoff, S.
1991, ApJ, 369, L27
Fabian, A. C. 1989, MNRAS, 238, 41P
Granados, A., Sachs, E., Shull, J. M., &£ Stocke, J. 1993, in NASA Conference Publication,
The Evolution of Galaxies and their Environment: Summaries of Contributed Papers,
ed. D. Hollenbach, J. M. Shull, &£ H. A. Thronson, Jr. (Washington, D.C.: NASA), in
press
Jura, M. 1987, in Interstellar Processes, ed. D. J. Hollenbach &£ H. A. Thronson, Jr.
(Dordrecht: Reidel), 3
Klein, R. I., McKee, C. F., &£ Colella, P. 1990, in The Evolution of the Interstellar Medium,
ed. L. Blitz (San Francisco: PASP), 117
Korista, K. T. et a1. 1992, ApJ, 401, 529
Krolik, J. H., McKee, C. F., &£ Tarter, C. B. 1981, ApJ, 249, 422
Laing, R. A. 1989, in Hot Spots in Extragalactic Radio Sources, ed. K . Meisenheimer &£
H.-J. Roser (Berlin: Springer), 27
Leahy, J. P., &£ Perley, R. A. 1991, AJ, 102, 537
382

Madej ski , G . M., Mushotzky, R. F ., Weaver, K. A ., Arnaud, K . A., & Urry, C. M 1991,
ApJ, 370, 198
Mather, J. C., et aI. 1990, ApJ, 354, L37
Mathews, W. G. 1982, ApJ, 252, 39
McCarthy, P. J . 1993, in Connections Between AGNs and Starburst Galaxies, ed. A. V . Fil-
ippenko (San Francisco: PASP), in press
McCarthy, P. J., van Breugel, W., Spinrad, H., & Djorgovski, S. 1987, ApJ, 321, L29
McCray, R. 1979, in Active Galactic Nuclei, ed. C. Hazard & S. Mitton (Cambridge:
Cambridge Univ. Press), 227
McKee, C. F. , & Ostriker, J. P. 1977, ApJ., 218 , 148
Morganti, R., Robinson, A., Fosbury, R . A. E. , di Serego Alighieri, S., Tadhunter, C. N.,
& Malin, D. F. 1991, MNRAS, 249, 91
Padovani, P., & Urry, C. M. 1990, ApJ, 356, 75
=---=-_.1991, ApJ, 368, 373
Perley, R . A ., Dreher, J . W., & Cowan, J . J. 1984, ApJ, 285, L35
Potash, R . I., & Wardle; J . F . C. 1980, ApJ, 239, 42
Rees, M. J . 1989, MNRAS, 239, IP
Scheuer, P. A . G. 1974, MNRAS, 166, 513
_--:--:' 1982, in IAU Symp. 97, Extragalactic Radio Sources, ed. D. S. Heeschen &
C. M, Wade (Dordrecht: Reidel), 163
Shanbhag-, S., & Kembhavi, A. 1988, ApJ, 334, 34
Shull, J . M., & Sachs, E. R. 1993, ApJ, 416, in press (Oct . 10, 1993) .
Stocke, J. T., Morris, S. L., Weymann, R. J., & Foltz, C. B . 1992, ApJ, 396, 487
Tadhunter, C. N., Fosbury, R. A. E., & di Serego Alighieri, S. 1988, in BL Lac Objects:
Proceedings of the Como Conference 1988, ed. L. Maraschi, T. Maccacaro, & M.-
H. Ulrich (Berlin: Springer-Verlag), 79
Turnshek, D. A. 1988, in Space Telescope Sci. Inst . Symp. 2, QSO Absorption Lines:
Probing the Universe, ed. S. C. Blades, D. A. Turnshek, & C. A. Norman (Cambridge:
Cambridge Univ. Press), 17
Turnshek, D. A., Foltz, C. B., Grillmaier, C . J., & Weymann, R. J. 1988, ApJ , 325, 651
Urry, C . M., Padovani, P., & Stickel, M. 1991, ApJ, 382, 501
Weymann, R . J. , Turnshek, D. A., & Christiansen, W . A . 1985, in Astrophysics of Ac-
tive Galaxies and Quasi-Stellar Objects, ed . J . Miller (Mill Valley: University Science
Books), 333
Wilson, A. S. 1993, in Physics of Active Galactic Nuclei, ed. S. J . Wagner & W . J . Duschl
(Berlin: Springer-Verlag), in press
Wilson, A. S., Elvis, M., Lawrence, A., & Bland-Hawthorn, J. 1992, ApJ, 391, L75
 COOLING FLOWS IN CLUSTERS OF GALAXIES
RICHARD MUSHOTZKY
Laboratory for High Energy Astrophysics
Goddard Space Flight Center

Abstract. I review the observational evidence for the cooling flow phenomena in clusters
of galaxies. Clusters with cooling flows have high x-ray surface brightness and a decrease
in temperature of the x-ray emitting gas in the center. Only cooling flow clusters have
strong optical emission line filaments in the central 20 kpc. The central galaxies in these
clusters often have a relatively large amount of blue continuum light compared to "normal"
cluster ellipticals and frequently have large Faraday rotation measures. X-ray imaging
data show that the cooling time of the x-ray gas in the central 100 kpc of a majority of
clusters is shorter than 5 x 109 years. Thus, it is relatively common for clusters to possess
a cooling flow. Detailed spatially resolved x-ray spectra obtained with the BBXRT are
consistent with the cooling flow models derived primarily from x-ray imaging data and
are independent evidence for mass cooling out at a variety of radii. There is at present no
alternative theory which can explain the wide variety of observed phenomena despite the
theoretical problems that the presence of'" 100 Me yr- 1 of cooling material present.

1. Introduction

There have been several excellent recent reviews of cooling flows (cf. Fabian,
Nulsen, and Canizares 1991; Fabian 1992 and the proceedings, Cooling Flows
in Clusters of Galaxies 1988) and review articles on x-ray emission from
clusters of galaxies (Sarazin 1988; Mushotzky 1988; Edge and Stewart 1991;
Donahue 1993). I will not repeat much of the material in those references.
However, a short synopsis of the general properties of clusters is important
to the understanding of cooling flows.
Clusters of galaxies have a dynamical time less than the Hubble time
and are the largest gravitationally bound objects in the universe. However,
they are rather rare objects, having a space density of "" 1O-6h~o Mpc- 3 ,
where h50 is the Hubble constant in units of 50 km s-1 Mpc- 1. Clusters
represent a galaxy density enhancement of"" 100 over the field at a length
scale of"" 3h s01 Mpc. In the x-ray band, they are rather bright with a range
of observed luminosities from 1042 to "" 2 X 1045 ergs s-l. There is a strong
correlation of x-ray luminosity with temperature (Mushotzky 1984; Edge &
Stewart 1991) with Tz ~ (10 7 K)(L/10 43 )0.4. The most luminous of these
systems would be visible out to z ~ 3 with the ROSAT satellite. Their total
gravitational mass is on the order of 10 14 _10 15 M0 and is dominated by dark
matter (Mushotzky 1990). The major baryonic constituents are the stars in
the galaxies, with 2 - 10% of the total mass, and the x-ray emitting hot gas,
which contributes up to (0.30)h~03/2 of the total mass (Cowie, Henriksen &
Mushotzky 1987). Thus, clusters appear to have a larger fraction of their
mass in baryons than does the total universe, since 0.04 ~ nbh~o ~ 0.06
from nucleosynthesis arguments (Steigman et al. 1989).

383
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, 383-408.
© 1993 Kluwer Academic Publishers.
384

Recent ROSAT results have confirmed (Briel, Henry, & Boehringer 1992)
that the hot gas frequently extends to distances of 3 Mpc from the center
(Henriksen & Mushotzky 1985). When appropriate observations exist , they
indicate that the intracluster medium (ICM) is always enriched in heavy
elements, with a median Fe abundance of 1/3 solar (Edge & Stewart 1991).
Detailed analysis of the temperature profiles of the Coma (Watt et al. 1992)
and Perseus clusters (Eyles et al. 1992) confirm that the dark matter is
more centrally condensed than the x-ray emitting gas (Henriksen 1985).
Recent results (cf. Briel et al. 1991, Edge et al. 1991) show strong evidence
for cluster merging being a very important process, and direct evidence for
cluster mergers has been found for systems at z ~ 0.3 (Davis et al. 1992).
For additional material on the evolution of clusters and the implications of
the x-ray and optical data on models of structure in the universe, I refer the
reader to Donahue (1993).

2. What is a "Cooling Flow" ?

In the central regions of a large fraction of clusters of galaxies there is di-
rect evidence from x-ray imaging for high (n ~ 10- 3 cm- 3 ) central den-
sities (Edge, Stewart & Fabian, 1992, henceforth ESF). Since the cool-
ing time tcool ~ [5kT /neA(T)], where A(T) is the cooling function, the
maximum cooling time is reached at the minimum of the cooling func-
tion A f'.J2 X 10- 23 ergs cm3 s-l at T ~ 2 X 10 7 K (see Figure 1), and
tcool < (2.3 X 10 10 yr)(T /2 X 10 7 )/(ne /10- 3 ). Thus for those clusters with
high central densities and which are "old" enough, significant cooling must
be occurring if it is not balanced by heat input. The question of whether
sufficient heating is occurring to balance cooling is both an observational and
theoretical question. As has been stressed by Fabian, Nulsen & Canizares
(1991), it is very difficult to come up with a heating mechanism, such as
conduction or cosmic ray heating that accounts for the prevalence of cooling
flows in clusters and reproduces the observed surface brightness properties
in detail. A heating mechanism that balances cooling while maintaining
consistency with the observed short central cooling times and low central
temperatures (see below) requires either an ad hoc fine- tuning of the heating
parameters or some sort of feedback mechanism (Loewenstein, Zweibel, &
Begelman 1991). No such mechanism that successfully confronts all of the
observed data has yet been found (for a different point of view see Sparks
1992).
In addition to the direct evidence of high central densities, x-ray spec-
troscopy of the central regions (Mushotzky & Szymkowiak 1988; Canizares
et al. 1988) shows the presence of cooler gas in the central regions of the
same clusters that show the high central densities. This cooler gas has a
shorter cooling time than the hot (T ~ 108 K) ICM in these clusters and
 385

......... Line
........ Recombination
- . - Brem sstrahlung
.••• . Two Photon
- Total
.• - - ' Extrapolated .

,,
,,
,,
,
,
I/)

'" E
u
,,
,
. ...
"
,.
,"
10· n
I/)

e> \ , i
.. " )
I '.

~
0-
0>
0
, \\
-I '"
1\, I
• V,,, •

.d··..! . '\. .
i i ' ... ..../.'.c; ......
1
:
i
I
,/ \ '\'"
\ .... \
\,. tJ. .· \. \,
i :I \\ ," \\ '\~ \"

I.:
: ::

:
:
:
:
... \

! '
1 0. 25 '--_..L-_ - ' -_---"-l...o..----"_ _~_-'-_-L..._'_____'__ ____'

1 2 3 4 5 6 7 8 9 10
Log T

Fig. 1. The cooling function A(T) is shown. Note the minimum in the X-ray regime
at 2 x 10 7 K. Figure courtesy of Neil Gehrels.

provides direct evidence for cooler material in the central regions of clusters.
The cooling rates inferred from the x-ray spectroscopic evidence are in good
agreement with those from the imaging data (White et al. 1991) and thus
provide strong additional evidence for cooling in these objects.
There is no direct observational evidence for a flow from x-ray spec-
troscopy, primarily because of the limited spectral resolution and signal to
noise of x-ray spectrometers flown in the 1980s. However, (see § 6.1 below)
the variation of the width with distance of the optical emission lines often
associated with "CF clusters" is consistent with a flow of material into the
central regions. In the absence of any evidence inconsistent with the exis-
tence of cooling flows, I will call these objects in this review "CF clusters"
386

rather than the more cumbersome but more accurate name of "high central
gas density clusters with cool central temperatures".

3. X-ray Imaging Data

3.1. EVIDENCE FOR COOLING FLOWS

The high-density (n > 10- 3 cm- 3 ) gas in the centers of a large fraction of
nearby clusters (see Forman 1988 for a review) has been seen in data from
all the x-ray imaging experiments: the Einstein Observatory (EO) imaging
proportional counter (EO/IPC) (Stewart et al. 1984); the EO high resolu-
tion imager (EO /HRI) (Branduardi-Raymont et al. 1981, Thomas, Fabian
& Nulsen 1987); and the ROSAT PSPC (position sensitive proportional
counter) and HRI (Sarazin et al. 1992). For the EO and ROSAT imaging
data the transformation from x-ray surface brightness to density is roughly
independent of temperature. In the EO/IPC, 0.01 ct s-1 arcmin- 2 corre-
sponds to a density of 10- 3 cm- 3 (Mushotzky 1988) and thus the simplest
analysis (Figure 2) shows that", 30% of all clusters with this high a central
surface brightness have cooling times less than the Hubble time. A more
detailed analysis includes the effects of the limited spatial resolution of the
EO /IPC, which strongly limited the central densities one could observe for
clusters at z > 0.05. In a distance-limited, x-ray flux-limited sample, more
than 70% of clusters show evidence for the CF phenomenon if the "age" of
the cluster is greater than 5 x 109 yr (ESF). In this re-analysis of IPC and
HRI cluster data. ESF show that when the effective resolution of the IPC
was less than 100 kpc, 31/36 clusters have tcool < 2 X 1010 yr and 25/36 have
tcool < 5 X 10 9 yr. These authors conclude that CF's are both long-lived and
common, since if they were either rare or transient phenomena they would
not occur in such a large fraction of clusters. This high frequency of oc-
currence strongly constrains models which try to account for much of the
observed "CF" correlations as being due to other physical processes such as
galaxy mergers in the centers or heating by radio sources.
In addition to the presence of a high central density, CF clusters often
show cusps in their central surface brightness (Jones & Forman 1984). That
is, the surface brightness in the central regions exceeds by a factor of 2 - 5
that predicted by King model fits to the outer regions. The strong association
between high central densities and a surface brightness cusp in the center is
an indication that there is another component in addition to the "normal"
cluster atmosphere.
Because of the limited spatial resolution of the EO /IPC this central cusp
often appears as a "small" source in the central regions and can be well
fitted by a gaussian beam (Henriksen 1985). This central cusp only appears
in high-central-density clusters and thus is not a phenomenon which is due
 387

Cooling Clusters
5
c
.D
4
'--
Q)
0.
'--
Q) 3
.D
E
~
C 2

0 -+---
-J50 -JOO -2.50 -200 -1 .50 -100

log 5(0) cts/sec/arcmin2

Fig. 2. Distribution of central surface brightness for published clusters. Note that
approximately 300/9 of the clusters have a central surface brightness above 0.01 IPC
counts s-1, which indicates a high central density.

to other cluster physics. However, the presence of a cusp is not a definite

indication of cooling occurring.

3.2. PHYSICAL PARAMETERS FROM IMAGING DATA

In order to derive the detailed physical conditions in the high-central-density

gas one must either fit detailed models to the data or "deproject" the surface
brightness distribution (White & Sarazin 1987; Arnaud 1988). The depro-
jection analysis gives the run of density, temperature, and mass cooling
rate (1\1) which is consistent with the observed surface brightness. While
the deprojection technique is rather robust, it does involve certain detailed
assumptions. In particular, one must assume hydrostatic equilibrium, the
nature of the potential ( core radius, central velocity dispersion and the pres-
ence of a central galaxy), and use an external pressure boundary condition
which is constrained by non-imaging x-ray spectroscopy. The full range of
maximal systematic uncertainty gives a factor of 2 uncertainty in the derived
388

values of if. However, it is not clear that the full range of likely potentials
have been examined and it is possible that there are additional systematic
uncertainties in the values of if so derived.
In the deprojection analysis, if scales as H;;2, the gas density as H;;1/2,
and the ratio of the cooling time to the Hubble time as H;,/2 (Stewart et
al. 1984). Typically we assume Ho = 50 km s-1 Mpc- 1. If Ho = 100 then
the values of if have been overestimated by a factor of 4. The deprojec-
tion analysis also derives the distribution of if with radius. The best fits
(Thomas, Fabian & Nulsen 1987; Arnaud 1988) show that the observed sur-
face brightness of CF clusters is less peaked thea if all the gas were to flow
into the center and that the observed distribution of surface brightness is
most consistent with if oc r; that is, mass is dropping out of the cooling
flow over a large range of radii. In order to achieve this, the gas must be
inhomogenous so that some of it cools out at large radii and some contin-
ues to flow in. "This distribution of mass dropping out of the CF, if oc r,
is consistent with the distribution of dark matter in the central regions of.
clusters. Since this material is optically faint, (see below) the dark halos of
giant elliptical galaxies might be created in this way.
The distribution of if is quite large (ESF), with a broad peak at '"
100 M0 yr- 1 (se~ Figure 3). These authors estimate that at the present
time, the net cooling due to cooling flows is '" 10- 4 M0 Mpc- 3 or only 1%
of the mean star formation rate in all galaxies whether or not they are in
clusters. Since most of the cooling is in the centers of rich clusters, the total
implied star formation rate is considerable compared to the total mass of
the central galaxies. With 100 M0 yr- 1 inside a cooling radius of'" 200 kpc,
10 12 M0 of cooled material will form in 10 10 years. Thus, it is possible that
CFs make a major contribution to the total mass of the central galaxy in
CF clusters.

There have been very few model fits to the data. Jones & Forman (1984)
report "excess" emission above the King model which was fitted to the
cluster surface brightness profile at r 2: 100 - 300 kpc. The luminosity in
this .excess correlates moderately well (Stewart et al. 1984) with the values
of M inferred from the deprojection analysis, but the scatter is large and
the absolute values are too large for the observed if. This is probably due
to the strong coupling between the King model fits and the value of the
excess flux due to the limited angular coverage and resolution of the IPC.
Henriksen (1985) modeled the central excess as a gaussian and found a linear
relationship between the derived excess luminosity and if, and found good
agreement with the predicted luminosities from the CF model. Figure 4
shows the distribution of excess luminosity and mass accretion rate from
the available model fits.
 389

14

.5 12
,.D
..---- non -CF clusters
~
~
10
8
6
4

2
o
-0.5 o 0.5 1 1.5 2 2.5 3 3.5
log M (dot) solar masses per year

Fig. 3. Distribution of the values of the mass accretion rates from ESF. A depro-
jection analysis of imaging data indicates that most cooling accretion rates are
between 5 and 200 M0 yr- 1 , with a substantial fraction up to 500 M0 yr- 1 .

3.3. INHOMOGENEITIES

One of the most surprising discoveries with the ROSAT satellite has been
that the x-ray surface brightness in the very bright centers of at least two CF
clusters, A2029 and 0335+096 (Sarazin, O'Connell, & McNamara 1992) is
highly structured on small (() ~ 20") angular scales, showing knots and blobs
of emission and evidence for deficits of x-ray emission. The very high surface
brightness seen in these filaments implies gas densities on the order of 1 cm- 3
and thus cooling times of", 10 7 yr, vastly exacerbating the requirements of
any heating mechanism. These filaments cannot be supported in hydrostatic
equilibrium and should fall into the center of the cluster in a free-fall time
('" 108 yr). Thus, their mere existence is hard to understand.
390

1.5
A

A
c
0.5 0
0-
Q) IJ.
</l
A
~ 6.
0
A
~ 0
-0.5
~
0 A
.2:- A
'iii A
0
c: A
'e
..2
-1.5
A
Cl
..Q

-2.5 b.

0 0.4667 0.9333 1.4 1.867 2.333 2 .8

log accretion rate (Solar masses/year)

Fig. 4. Data from Jones & Forman (1984) (a) and Henriksen (1985) (0) show a
correlation between the central excess over a King profile and the mass accretion
rates from Stewart et a1. (1984) .

.4. X-ray Spectral Analysis

4.1. SUMMARY OF INSTRUMENT PROPERTIES

Much of the spectral data obtained has been rather dependent on the in-
strumental characteristics. We briefly summarize the salient aspects of the
primary detectors used in the last few years. The EO solid state spectrome-
ter (SSS) had both moderate spectral resolution (E/6E::::; 6 at 1 keY) and
collecting area in the 0.5 - 4 ke V band, and its non-imaging detector had a
3' field of view. The EO focal plane crystal spectrometer (FPCS) had better
spectral resolution, a 3 X 30' beam, and poorer signal to noise in the 0.4 - 3
ke V band. The Ginga non-imaging proportional counter was sensitive in the
2- 30 ke V band, had a 10 X 20 field of view, a very large collecting area, and
E /6 E ::::; 6 at 6 ke V. The ROSAT position sensitive proportional counter
(PSPC) is sensitive in the 0.15 - 2 keY band, has E /6E ::::; 2.5 at 1 keY,
a large (,....., 10 radius) field of view, an imaging FWHM of 25", a very low
background, and rather large collecting area. However, in the survey mode,
 391

the net total exposure was rather low ((t) ~ 500 s).

4.2. X - RAY SPECTRAL EVIDENCE FOR CF FROM THE EINSTEIN OBSER-

VATORY

Both the EO JSSS (Mushotzky & Szymkowiak 1988; White et al. 1991) and
the EO jFPCS (Canizares, Markert & Donahue 1988) saw emission from low
energy lines which arise primarily in cool (T S; 3 X 10 7 K) gas. The better
spectral resolution of the EO SSS, compared to the imaging detectors (pri-
marily the EO JIPC and EO jHRI, and more recently the ROSAT PSPC),
allowed it to measure the strongest emission lines from gas in the tempera-
ture range from 5 X 106 - 2 X 10 7 K (for a summary of the SSS results see
Mushotzky & Szymkowiak 1988). The main coolants in gas, in this temper-
ature range, are the Fe-L lines, in particular the lines due to Fe XX-XXII
(Figure 5). The presence of these lines in the SSS spectra of over 12 clusters
(White et al. 1991; Mushotzky & Szymkowiak 1988) is direct evidence for
the existence of gas at T S; 3 X 10 7 K, in clusters whose average tempera-
ture, derived from large-beam (more than 2 square degrees) experiments, is
T ~ 6 X 10 7 K.
The temperature and emission measures of the cool component were pri-
marily constrained by the relative strengths of the different emission lines.
Mushotzky et al. (1981) found that the fitted temperature of the cool com-
ponent is the temperature of peak emissivity for the strongest observed Fe-L
lines. Thus, the strong temperature dependence of the emissivity of these
lines, ~bove a critical temperature, makes direct measurement of the emis-
sion measure versus temperature distribution difficultm with the relatively
low resolution and signal to noise of the SSS. The SSS did not possess enough
bandpass or spectral resolution to detect the continuum emission from ga.l' in
the temperature range from 108 K to 2 X 10 7 K. In this temperature range,
the emission lines are relatively weak and the prime signature. of cooling
would be the detection of a bremsstrahlung continuum from gas cooler than
the bulk of the cluster gas seen at larger radii.
The 826 e V line of Fe XVII was seen by the FPCS in 3 clusters. Since its
emissivity is ex: T 6 , its mere presence shows the existence of T ~ 5 X 10-6 K
gas (Canizares, Markert & Donahue 1988). The luminosity in this line, or
any line whose strength is sensitive to temperature in the range of interest
(Cowie 1981), can be used to estimate M.
Alternatively, one can fit CF models to the overall spectra. Estimates of
M from the SSS spectra and the FPCS line luminosities agree quite well
with the M from the IPC data (White et al. 1991). Thus, in general, the
x-ray spectral data in are good agreement with the imaging data. However,
neither the SSS nor FPCS data were of high enough quality to actually
measure the distribution of emission measure with temperature, nor did
392

Data from Raymond Smith Code

1 0 3 c-----t-'~:__-------___,
-lr o H like
.:::
..... Si He like
..... -+- Si H like

s:.....
'0
-e- S He like
10 2 ""*" S H like
c -.- Fe xvii (826)
~ ~ Fe 1020 blend
C<l
:>
::l
0"
~ 10 1
6.4 6 .6 6.8 7.0 7.2

log T

HARD X-RAY LINES IN COLLISIONAL EQUILIBRIUM

1000

"...
>
Q)

:J:
I-
0
~
3: -;.\
I-
z
\\1""'b\~.
w r.,,\
..J
« ~\
O~,
>
::J
0
W

OATA FRO'"
RAVWOND AND St-i lT H
(19111

10
.4 4 10 40
kT (keV)

Fig. 5 . Top panel: The equivalent width versus temperature of some ofthe strongest
lines in the 0.6 - 3 keY band from a collisional plasma. Notice the very strong
temperature dependances and that the EW of some of the lines is greater than 1
ke V, thus making a low-T plasma dominated by lines. Bottom panel: The equivalent
width versus temperature of some of the strongest lines in the 1.8 - 7 ke V band. The
line "blends" are weighted by the energy that would be measured by a proportional
counter. Notice that the Fe (He-like) "6.7 keV" line is the strongest line in the
spectrum for kT 2: 2 ke V.
 393

they have sufficient spatial coverage to determine the distribution of gas

temperature with position. With the exception of a few clusters, the SSS and
FPCS observations were always centered on the cluster center. If the cool
gas were not concentrated in the cluster center, the implied flux of the lower
temperature component to fit the SSS and FPCS would have significantly
altered the spectrum seen by the large beam experiments from that which is
observed. Thus, indirectly, the EO spectrometer data indicate that the cool
gas was in the central regions of the cluster.

4.3. RECENT X-RAY SPECTRAL RESULTS

Since the flight of the Einstein Observatory, there have been several ex-
periments which have had x-ray imaging spectral capability but somewhat
limited spectral resolution. The Spartan 1 results on the Perseus cluster (Ul-
mer et al. 1987) found that inside of the nominal cooling radius ("'" 5', 0.16
Mpc) the effective temperature was significantly less (T :::::: 4 X 10 7 K) than
in the outer regions (T :::::: 7 X 10 7 K). A similar, but less significant result
was found by the SL2jXRT experiment (Eyles et al. 1991). More recently,
ROSAT all-sky survey imaging spectra of the Perseus cluster (Schwarz et
al. 1992) show a significant temperature gradient in the central 5', with the
projected average temperature decreasing from'" 6 X 10 7 K at r :::::: 6 - 20' to
'" 1.8 X 10 7 K at r :::::: 0.5', consistent with the EOjIPC deprojection analy-
sis. The relatively short exposure of the survey data combined with the low
spectral resolution did not allow unique fits to the data, but the indication
of cool gas in the center was unequivocal.
The Ginga spectra of CF clusters show residuals at E ::; 5 ke V (Hatsukade
1992). Recently Day, White & Hatsukade (1992) have fitted Ginga spectra
of 3 CF clusters jointly with EO JSSS data. They find that the combined
data sets require multiple temperatures and are consistent with lhe CF
hypothesis. While the Ginga data may also be fitted with a model having
a strong decrease of temperature with radius, the combination of the two
data sets cannot, giving more evidence for the spatial localization of the cool
component in the center of these clusters.

4.4. BBXRT OBSERVATIONS

General Results

The highest quality imaging x-ray spectra of clusters was obtained by the
BBXRT experiment (Serlemitsos et al. 1991; Mushotzky 1992). This experi-
ment combined good energy coverage (0.3-12 keY), good spectral resolution
(EjbE:::::: 10 at 1 keY and 40 at 6 keY), low detector background ("'" 1% of
Ginga), good collecting area, moderate angular resolution and solid angle
 394

R Fe XII-XXIII
./"

(J) ~
('V
::J
-0 Fe X II- XX I V
(J) S I XII I
G.>
'- Q
0l
~

- Oxygen edg e feature

0.5 1.0 1.5 2.0

Energy (kev)
Fig. 6. The X 2 residuals for an isothermal fit to the central 4'-diameter region of
the Perseus cluster from tile BBXRT data. Line blends due to Fe-L and Si-K which
arise in low temperature plasma are clearly seen, as is the absorption feature due
to cold oxygen.

coverage. It was the first experiment which had sufficient spectral resolution
and angular resolution to "map" the cooling flow.
BBXRT observed 5 clusters with adequate signal to noise to determine
spatially resolved temperature data; only for the CF clusters A262, A496
and A426 (Perseus) was significant emission from gas at temperatures con-
siderably less than the cluster average (Figure 6) seen in the central regions
(2' radius) of the cluster and not seen in the outer regions (3 - 8' radius).
While the data for A262 and A496 confirm the existence of cool material
whose emission measure is consistent with that indicated by a CF analysis of
the imaging data for these clusters, the relatively low signal to noise from the
very short exposures do not allow a detailed model independent confirma-
tion of the CF hypothesis. However because of the broader bandwidth of the
BBXRT experiment compared to the SSS, and its better spatial resolution
compared to big-beam proportional counters such as Ginga, the BBXRT
measurement of the temperature is not entirely dependent on the relative
 395

line strengths, and we can demonstrate that the central continuum temper-
atures are not consistent with the cluster average and definitively show cool
gas in the central regions, in qualitative agreement with the predictions of
the CF model.
In the case of A496, the BBXRT temperature in the central regions in an
isothermal model is kT = 2.66 ± 0.23 keY. This value is in excellent agree-
ment with the deprojection results of ESF and is significantly less than the
cluster average temperature of 3.97 ± 0.06 keY (Hatsukade 1992).

Perseus cluster

The high signal-to-noise multiple observations and the acceptable spatial

resolution allow a detailed measurement of the distribution of gas tempera-
tures in the central lS' (....., O.S Mpc) regions of the cluster by direct spatial
imaging. Simulations of cooling flow models, show that, with the BBXRT
spectral resolution, signal to noise and bandpass, spatially-dependent two-
temperature fits are an adequate representation of the CF model. Using
this decomposition of the data we find (Figure 7) that there is a significant
change in the ratio of cool to hot gas with distance from the central source,
falling off as r-l. The fraction of the emission due to hot gas is consistent
with the surface brightness profiles reported by previous experiments.
By integrating and projecting various cooling flow models onto the BBXRT
pixels, Loewenstein et al. (1992) show that the distribution of the cool com-
ponent is consistent with the deprojection analysis of the imaging data
(Fabian et al. 1981, ESF) that indicate if ex r, T ex r 1 / 2 , and a cooling
radius Rcool ~ S - 8'. The BBXRT data are not consistent with constant M
if Rcool ::; 10', thus confirming the occurrence of extended mass deposition.
Moreover, the value of the total estimated from the emission measure of the
cool component, if = [SkT(Rcool)/2JLmpl-l X EM ~ (200 - 300) M0 yr-I,
is consistent with previous results. These results are the first direct evidence
for detailed CF models and for material cooling over a wide range of radii
(2 - 15', 100 - 800hsl kpc).
However, the simple CF spectral models do not provide a good fit to the
BBXRT spectral data, giving apparently too much emission at temperatures
less than 1 keY. We are not certain of the origin of the discrepancy; possi-
bilities include the neglect of gravitational heating in the spectral models,
additional heating sources, edges due to highly ionized material, or perhaps
additional physics. Clearly a more detailed analysis is required.

Absorption in the Central Region

The BBXRT data for A496, A262 and Perseus confirm the surprIsmg
discovery of cool absorbing material in the central regions of these clusters.
396

Relative Distribution of Hot and Cold Material

in the Perseus Cluster
I
I

1+ BBXRT data
•
--:+-

•
2
W
0
o· • •
•
0.1 - -::r-
0) 0
>
......
co 0
8 •
0)
a: 0.01 -::- -
•0
EM hot
EM cold
I
0.001 I

0.01 0.1
distance from center (Mpc)
Central Part of Perseus cluster
(J)
BBXRT Spatially resolved spectra
co I
0) I
en
...... co
cr-
•• •
0
..r:
0
...... CI

•
C)
C)
0
0 :J.)
u 0.1 + --
10.0

T
....... 0
0 ~
2
w
.......
0
0 I
...... 0.01 I
co
'- 0.1
distance from center (Mpc)
Fig. 7. Spatially-resolved BBXRT spectra confirm the cooling flow hypothesis by
showing that the amount of cool gas projected along the line of sight decreases with
radius in the Perseus cluster. The top panel shows the distribution of the emission
measure of the cool T ~ 10 7 K component and the hot T ~ 7 X 10 7 K component.
The lower panel shows the ratio of these two components versus radius.
 397

The values ofthe column densities, typically N(H) ~ 5 X 10 20 cm- 2 , in addi-

tion to the Galactic absorption, are in agreement with the EO JSSS results
(White et a1. 1991). For the Perseus cluster the BBXRT spectral resolution
and signal to noise allowed a measurement of the redshift of the absorbing
material which is consistent with the Perseus cluster and inconsistent with
Milky Way material. Also, the material does not appear to be uniformly dis-
tributed in Perseus, with greater absorption to the northwest and southeast
over a scale of ~ 5'. There is little evidence in the BBXRT spectra (Arnaud
et al. 1992) for the strong emission from NGC 127'5 seen by the SL2jXRT
experiment, and thus the absorption cannot be attributed to this object.
The total amount of this cool material is (3 X 1011 M0)N21R~oo. Here,
N21 is the cool H column in units of 10 21 cm- 2 and RIOO is the intrinsic size
of the cold material in units of 100 kpc (White et al. 1991). One possible
explanation for this cool absorbing material is the accumulation from the
CF, since its mass is roughly the accumulation of the total if over the life of
the cluster (White et a1. 1991). If so, this obviates (see below) the need for
large amounts of star formation in CF clusters. The formation and survival
of this material is not understood at present (Loewenstein and Fabian 1991)
and much more theoretical and observational work is necessary.

5. Radio Astronomical Evidence for Cooling Flows

Recently it has become clear that cooling-flow clusters have very high Fara-
day rotation. In particular, Faraday rotation measures (RM) greater than
800 rad m- 2 have been found in 9 CF clusters: M87, Cyg-A, PKS 0745-191,
Perseus, Hydra-A, 3C295, A1795, A2199, and A2052. To quote Ge (1991),
"all sources in the center of strong cooling flows have high RM, and all
sources in the center of weak cooling flows or in non-cooling flow clusters
have much smaller (::; 100 rad m- 2 ) rotation measures". This pattern was
originally hypothesized by Dreher, Carilli & Perley (1987).
Ge (1991) shows that the high RM does not correlate with radio lumi-
nosity, the size of radio source, or features in the total intensity map, but
it does correlate well with x-ray measured central gas density and cooling
rate (see Figures 8a,b). The high RM seems to result from a screen in front
of the radio source (the data follow the ,\2 dependence very well), often has
sharp gradients, and its absolute value is about the same across the source.
Since the RM = (812 rad m- 2 ) J neBII(J.LG)dlkpc, one can easily estimate
the required magnetic fields if the RM is due to the cooling flow (Sarazin
1992). Using "typical" CF densities (ne) ~ 5 X 10- 3 cm- 3 and scales of 10
kpc for the coherence length of the B-field, we estimate BII ~ 10 - 100J.LG,
a factor of 10 stronger than in the outer regions of the cluster.
This general property of cooling flow clusters was not discovered un-
til recently because its measurement requires: (1) a small beam size (large
398

Correlation of RM and Cooling rate

1 1
'I 'I

104 -Fo-
• • -

• ••• • •
M87
01
<
E 1000-1:
'-

•I
--0
(1J
L

L
a:: 100-1:

10 I I
. 10 100 1000
M (solar mass/yr)

-
Densjt

10 4 inear relation

~
c.c::
1000
\ • M87

100

0.001 0.01 0.1

density (cm-3)
Fig. S. (Sa) The correlation of ladio lotation measure (RM) with x-lay infelled
cooling lates. The accletion lates of:::; lOM0 Yl-l ale uppel limits. Note that the
high RM seems to have an onset when the clustels have cooling flows. The only
exception is MS7 whele the high RM is plobably due to the velY high effective
spatial lesolution possible fOl this close object. (Sb) The correlation of X-ray de-
telmined density with rotation meaSUle. The lineal lelation shown corresponds to
J BII (j.LG)df(kpc) ~ 100 lad m- 1 . If dl ~ Rcoo!' this implies that B ~ 1J.LG
 399

RM gradients generate beam depolarization); (2) a high sensitivity; and (3)

closely spaced frequencies. Since RM oc A2 , the very high observed RM re-
quires measurement of the polarization at very closely spaced frequencies for
the rotation to be seen. One finds 6.(P.A.) = 7r for RM = 9500 rad m- 2 and
6.11 = 4995 - 4885 MHz. Without such close frequency spacing one would
just observe depolarization.
Using x-ray data to estimate the local gas density, Ge (1991) showed that
the observed RM is proportional to the gas density derived from a CF analy-
sis of the x-ray data; thus, it is the gas density that primarily determines the
RM. The B-field energy density, (B2 /8k7r) ~ (ne/O.Ol cm- 3 )(T /2 X 10 7 K),
is in rough equipartition with the gas energy density for ne ~ 0.01 cm -3 and
T ~ 2 X 10 7 K, the conditions appropriate for a cooling flow. Thus inside
radii of a few 10's of kpc the magnetic field must have a very important
effect on the physics of the cooling gas.
The size of the cluster radio source (typically ~ 20 kpc) limits the abil-
ity of this technique to probe the outer regions of the CF and study the
relationship of the magnetic field to the cooling gas over the whole dynamic
range of the CF phenomenon. It would be very interesting to find sufficiently
bright background radio sources in the outer regions of the CF and see if
the correlations seen in the central regions of the CF continue.

6. Optical Filaments and Cooling Flows

6.1. "EVIDENCE" FOR CFs

There has recently been an extensive review of this subject by Baum (1992)
to which I refer the reader. It has long been known (Heckman 1981, Cowie
et al. 1983) that strong, spatially-extended, Ha emission occurs in the cen-
tral regions of some, but not all, CF clusters. What is now clear from the
larger surveys of Heckman, Baum, van Breugel, & McCarthy (1989, here-
after denoted HBvBM) and Hu (1988) is that this emission occurs only in
CF clusters. While the upper limits for non-CF clusters have not always
been tabulated, it is clear that the presence of strong Ha always implies a
CF. However, it is not true that a strong CF always has strong Ha:.
The question of what qualifies as strong Ha emission can be quantified by
comparing CF clusters with the only other luminous extended Ha:-emitting
objects, low-redshift radio galaxies (HBvBM, Baum 1992). The range of Ha:
luminosities, L(H a), for CF clusters is 10 40 ~ L(H a) ~ 1042 . 5 ergs S-l
(with the lower limit being a selection effect), while no low-redshift radio
galaxy has L(Ha) 2: 1041 ergs S-l (HBvBM). While both CF clusters and
radio galaxies show a roughly linear dependence of L( H a) on radio power,
CF clusters have roughly 10 - 100 times the Ha luminosity per unit radio
luminosity as do radio galaxies.
400

The optical .line emitting gas in CF clusters has different kinematics

(Baum 1992; Crawford & Fabian 1992) and emission-line ratios from fil-
aments in radio sources. In particular, the range of line ratios of [N II], Ha,
[S II], and [0 I] are rather different from radio galaxies and show a smaller
range of variation. Thus, as Baum (1992) points out, CF nebulae exhibit
a characteristic spectrum. In contrast to radio source filaments, there is no
correlation between the radio structure and optical filaments in CF clusters,
except that they both occur in the central regions of the cluster. There is
a trend for L( H a) to be correlated with !VI, but the scatter is large. The
CF clusters form the class of "calm non-rotators", as described in Baum et
a1. 1992. In this class the gas shows little or no signs of rotation and has
relatively low total velocity dispersion, (T ::; 1000 km s~l (Balun et a1. 1992).
HBvBM (1989) point out that the velocity widths in this class decrease, in
general with distance from the center, dropping from 500 - 1000 km S-1
FWHM to 100 - 300 km s-1 at 10 kpc. This variation in line width with
distance is the signature of a radial flow. As Baum et al. (1992) note, neither
the dynamics nor the images of the emission line gas show evidence of merg-
ers and the kinematics of the gas is rather different from objects that are
clearly mergers based on their optical morphology (the "rotators"). Thus,
CF cluster optical filaments form a class that has distinct properties.

6 . 2 . WHAT Do THE OPTICAL FILAMENTS TELL Us ABOUT CFs?

While it is clear that CFs have something to do with the class of "calm
non-rotating" emission line nebulae, the direct relationship of the optical
filaments to the CF is difficult to understand (Baum 1992). The filaments
do not appear at the cooling radius but rather are strongly concentrated in
the inner 10 kpc, and the total emission has a much steeper surface bright-
ness profile than the !VI oc r that the x-ray data would suggest. Some optical
filaments (Sparks et a1. 1989) show reddening consistent with a "normal"
galactic reddening law, while one would not expect any dust for gas con-
densing out of a cooling flow. Finally, the observed line luminosities are, in
general, too large.
If the gas is condensing out of a cooling flow, one expects one Ha photon
per cooled proton (Hrec = 1). In most CF clusters one finds H rec ~ 10, and
in a few exceptional cases H rec ~ 100, which implies that the gas producing
the emission line filaments is receiving additional energy. This results in
more optical emission line flux than would be produced by a pure cooling
flow. The situation is actually more extreme, since it is clear that Ha only
carries a small fraction of the total emission in the optical and UV band.
The observed Lya fluxes are 3 - 10 times the Ha fluxes (Hu 1992), and the
theoretical prediction is 8 - 27 times as large. The total energy released in
the cooling flow, is 5kT /2 or 22 ke V per proton. Thus, if the observed Ha
 401

luminosity is too large, H rec 2: 1600, and if one assumes Lya/Ha ~ 10, the
luminosity in Lya would exceed the total available energy from cooling. For
those few clusters in which this situation holds, the cooling flow does not
have enough energy to power the optical emission line filaments. In addition,
since if <X r and the filaments lie well within the cooling radius, we have
overestimated the fraction of the cooling luminosity available to power the
lines.
Therefore, the emission-line filaments do not look like what one would
naively expect them to be if they were cooling out of a CF. These discrep-
ancies produced a virtual sea of theoretical papers whose conclusions range
from "we can easily produce the observed filaments if x, y and z are true"
to "the observed optical filaments disprove the CF hypothesis". I think the
jury is still out on the details of the physical connection of the filaments to
the CF but it is clear from the observational data that there is some sort of
strong connection between CFs and the optical emission line filaments .

7. Optical Colors
It has recently been shown (Hintzen & Romainshin 1988; Johnstone et al.
1988; McNamara & O'Connell 1992) that most of the central galaxies in
CF clusters exhibit extended color-profile anomalies which are correlated
with if and that no "non-CF" galaxies show these effects (Figure 9) . It is
thus clear that, at low redshift, CF clusters are the only cluster ellipticals
that show evidence for recent star formation. However, the radial extent of
the color anomalies is only 5 - 10% of -the cooling radius (5-20 fPC) and
thus similar to the extent of the emission-line filaments. The extent of the
"color excess" and its magnitude correlate with the x-ray inferred M such
that clusters with the largest M have the most dramatic excesses in both
magnitude and size. Thus, the optical colors and the total amount of "blue"
light is related to the if inferred from x-ray images.
In at least two objects (McNamara & O'Connell 1992) the morphology
of the excess blue light is rather different from the Ha emission, but oc-
curs in the same general area of the cluster. The relationship of the color
excess to the radio morphology and the presence of dust appears complex
and may vary from cluster to cluster. However, the amount of the "excess"
blue light observed is considerably less than that expected from the conver-
sion of the x-ray determined M into stars with a standard IMF (Sarazin &
O'Connell 1983; Fabian, Nulsen, & Canizares 1982). IDE spectra of low-z
cooling flows, DIT observations of Perseus (Smith et al. 1992) and ground
based spectroscopy of moderate redshift (z ~ 0.2 - 0.3) CF clusters (Allen
et al. 1992) all indicate the absence of 0 stars and, in most cases, of stars
earlier than BO in CF clusters. Detailed modeling of A2597 (Crawford et al.
1989) place very tight constraints on the form of the IMF and the fraction
402

Colors in CF vs non-CF central Galllxies

.. •
data from McNamara and O'Connell 1992
2~~~~~~rrrlh~,n. . . .~~~~~~

..--..
1.5
•
I

:r
~ 0.5

•
...........
19

•
0

-0.5
A2029
- 1
-100 0 100 200 300 400 500 600
M(solar masses/yr)

Fig. 9. Correlation of the "color excess", G(U-I), vs x-ray cooling rate. Only CF
galaxies show the blue color excess gradient. Note that one CF cluster A2029 does
not show a strong color gradient. This same object also does not have optical
filaments.

of massive stars that can form. However, as noted by these authors, the de-
tails are quite sensitive to the total amount of Balmer continuum emission
produced by the optical filaments and the reddening that was used.

8. What Happens to the Cooling Material?

Despite the strong optical evidence for star formation and the luminous opti-
cal filaments, in general the material formed in the cooling flow is optically
dark and thus qualifies as baryonic dark matter. There is as yet no firm
agreement as to the final state of the cooled material. In addition since the
"age" of the CF is not known with any precision, nor how long the cluster
has existed since the CF was established, estimates of the total amount of
cooled material are rather uncertain. Several possibilities discussed in the
literature for the final depository of the cooled material are: (1) a flow into
a central black hole; (2) low mass star formation; (3) the formation of very
 403

cold clouds.
While it is certainly possible that some fraction of the material flows into a
central engine there are several observations which indicate that only a minor
fraction does so. First of all, most of the CF central galaxies are not classical
active galaxies. That is, they do not show broad optical emission lines, a. non-
thermal continuum, or hard x-ray emission. There are clear exceptions to
this rule, the most extreme being NGC 1275 in the Perstms cluster. While
most of the central galaxies are radio emitters, very few of them show strong
jet emission (Cyg-A and A1795 being obvious counterexamples) or other
indications of recent activity. Secondly, if a fair fraction of the material
falls into the center, the implied luminosity of an AGN would be enormous,
L ~ (6 X 1047 ergs s-l)(M/lOO M0 yr- 1 )(1]/0.1), where 1] is the efficiency
of the conversion of rest mass into energy. Thus either the putative massive
object in the central galaxy has a very low efficiency or much less than
100 M0 yr - 1 falls into the center. Thirdly, the x-ray surface brightness
profiles indicate that a substantial fraction of the cooled material does not
make it into the center .
Because of the low ratio of blue light to it, low-mass star formation
(Sarazin & O'Connell 1983; Fabian, Nulsen, & Canizares 1982) seems to be
a favorable possibility. The obvious objections to it are: (1) there is no direct
evidence for such stars; and (2) there is no obvious mechanism by which low-
mass stars can be preferentially formed. However, star formation is not a
well understood process, and there is substantial evidence in our galaxy for
spatially-dependent bimodal star formation (Gusten & Mezger 1983) with
low-mass stars being preferentially formed in certain regions. If low-mass
stars are indeed the reservoir of the accreted mass , strong constraints can
be obtained on their mass distribution if the underlying IMF is assumed to
be a power law.
The observational indication of cold absorbing material along the line of
sight to CF clusters (White et al. 1991) gives impetus to the idea that much
of the cooling material has gone into cold dense clouds. The mass of cold
material inferred from x-ray absorption data are consistent with this idea.
However, confirmation of the existence of large amounts of cold material
from 21 cm and CO observations (McNamara, Bregman, & O ' Connell 1990)
is so far absent. If much of the cooling material ends up as cold clouds, they
must be rather small, have a projected covering factor of '" 0.5, and, from
stability /survival analysis of cool clouds, have M 2:: 1 M0 with an optical
depth greater than 1 at 21 cm.
What seems clear is that star formation is occurring in the central regions
of CF clusters, but it has rather different properties from star formation in
nearby spiral galaxies. This is perhaps not unexpected given the very dif-
ferent physical conditions. If the recent HST observations of bright knots
of star formation in the Perseus cluster (Holtzman et al. 1992) can be at-
404

tributed to the cooling flow we predict that similar bright blue knots will be
found in other nearby cooling flow clusters such as Centaurus and A262. If
confirmed, this would be direct evidence for the form of the star formation
and will allow a detailed examination of its relation to other properties of
the cooling flow.

9. Evolution and All That

It seems clear that many CF's are depositing 100hs02 M0 yr- 1 in the central
regions of a significant fraction of present day rich clusters. It has been
noted (Fabian et al. 1986) that CFs seem to have the following properties
important for galaxy and cluster evolution: (1) they produce dark matter,
e.g., matter with a high Mj L ratio; (2) if if <X T, this matter is distributed in
the same fashion as an "isothermal" halo; (3) there is strong direct evidence
for the existence of CF's at z ~ 0.3; (4) there is evidence (high-pressure
extended gas around some quasars) for CFs at high redshift (Forbes et al.
1990; Crawford et ale 1989), and indications that most high-redshift radio
galaxies are also surrounded by high pressure gas; (5) a large fraction (30
- 70%) of low-redshift clusters have CFs and thus CFs are a common and
persistent feature of clusters; (6) CF's provide an alternative (e.g. to the
Silk and Rees mechanism) method for creating cold stuff out of hotter stuff;
and (7) at present there is an indication, based on the Einstein Medium
Survey (Donahue, Stocke, & Gioia 1992), that CFs may be more numerous
at higher redshift.
The direct extension of the low-z results to higher redshifts is problem-
atic because of the strong effect of mergers on present-day clusters. Thus,
low-z clusters may be a poor guide to high-z ones. It is unclear how mergers
affect the cooling flows. Detailed comparison of the CF properties of obvi-
ous nearby merger candidates with non-mergers will only be possible with
ROSAT data. However, comparison of the Einstein data for bimodal versus
isolated clusters (Henriksen 1991) indicates that these "pre-merger candi-
dates" are less likely to harbor cooling flows. If this result is substantiated
by more accurate ROSAT data, it may indicate that cooling flows may be
initiated by mergers.
It is certainly possible that CFs are the origin of much of the mass of
central dominant galaxies in clusters. Thomas & Fabian (1990) have calcu-
lated the upper limit to the amount of material that could have been formed
in cooling flows using the flux in the soft x-ray background as a boundary
condition. This result depends critically on the initial redshift of formation
and the initial temperature of the gas. The density ranges from 0.01 of the
critical density, if log Tinit = 6.3 and Zform = 2, to 40% of critical density
if log Tinit = 7.7 and Zform = 10. In the hierarchical clustering scenario, it
is possible that CFs around M ~ 1013 M0 objects could be responsible for
 405

much of the baryonic dark matter.

If much of the mass of a present-day large galaxy were virialized, then
it would have had T ~ 10 7 K, R ~ 0.25 Mpc, and tage ~ tcool 2: tfreefall.
These are the conditions for a CF with a cooling time, tcool ~ 109 yr, a
mass accretion rate of M ~ 104 M0 yr- 1 , and a cooling luminosity of
Lcool ~ 1045 T 7 ergs s-l. Such an object would have an extended history
of star formation as well as a large extended emission-line nebula, with an
angular size of tV IS" at z ~ 2. The galaxies in these nebulae will have A-
type spectra indicative of large amounts of star formation with a truncated
mass fWlction. There would tend to be segregation of star types with massive
stars found only in the centers of the protogalaxies and thus an abundance
gradient.

10. Conclusion
There are strong observational indications from x-ray, optical and radio ob-
servations that the centers of many clusters are sites of unique phenomena.
The best unifying hypothesis which ties all the data together is the cooling
flow theory, which attributes the high x-ray surface brightness, the decrease
of temperature of the x-ray emitting gas in the center, the appearance of
optical emission line filaments, the relatively large amount of blue contin-
uum light, and the large Faraday rotation to the presence of high-density
gas which is undergoing significant cooling.
The prime physical implication of this theory, that in many clusters
'" 100 M0 yr- 1 of material is cooling out of the hot phase in a distributed
fashion, is unappealing to some observers. However, there is no real alterna-
tive. While some of the data are hard to understand in the framework of the
cooling flow hypothesis, such as the presence of dust in some of the optical
emission-line filaments, alternate ideas such as invoking mergers to explain
the line emission or heating by relativistic particles to prevent the cooling
of the gas have been shown to be either unstable or of insufficient generality
to explain most of the data. It seems clear to this author that this field is in
its infancy and that much more theoretical and observational work will be
necessary before these objects can be even qualitatively understood.
The launch in 1993 of the first imaging x-ray spectroscopic satellite,
Astro-D, will provide for the first time detailed spatially resolved x-ray spec-
tra of a reasonable sample of cooling flows. Based on this sample I expect a
significant jump in our observational knowledge of cooling flows.

Acknowledgements
I would like to thank Mike Shull and Harley Thronson for their kind in-
vitation to present this lecture at the Grand Teton Summer School and
406

for organizing an exciting meeting. I would like to thank Mark Henriksen

and Mike Loewenstein for a critical reading of the manuscript and Mark
Henriksen for help in preparing the text.

References
Allen, S.W., et a1. 1991, MNRAS, in press
Arnaud, K. 1985, PhD Thesis, University of Cambridge
Arnaud, K. 1988, in Cooling Flows in Clusters of Galaxies, ed. A. Fabian, Kluwer, 31
Arnaud, K. et. a1. 1992, in preparation
Bawn S. A. 1992, in Clusters and Superclusters, ed. A. Fabian., 171
Bawn, S. A., Heckman, T. M., & van Breugel, W. 1989, ApJ, 389, 208
Bechtold, J. & Ellingson, E. 1992, ApJ., 396, 20
Branduardi-Raymont, G., et aI. 1981, ApJ, 248,55
Briel, U., et a1. 1991, A&A, 246, L10
Briel, U., Henry, J.P., & Boehringer, H. 1992, A&A, 259, L31
Canizares, C. R., Markert, T.H., & Donahue, M. 1988, in Cooling Flows in Clusters of
Galaxies, ed A. Fabian, Kluwer, 63
Cowie, L.L. 1981, in X-ray Astronomy with the Einstein Satellite, ed R. Giacconi, Reidel,
227
Cowie, L.L., Hu, E.M., Jenkins, E.B., & York, D.G. 1983, ApJ, 272, 29
Cowie, L.L., Henriksen, M.J., & Mushotzky, R.F. 1987, ApJ, 317, 593
Crawford, C., Aranud, K., Fabian, A., & Johnstone, R. 1989, MNRAS, 236, 277
Crawford, C., & Fabian, A. 1992, MNRAS, in press
Davis D., et a1. 1992, in preparation
Day, C., White, R., & Hatsukade, I. 1992, in preparation
Donahue, M. 1993, these proceedings
Donahue, M., Stocke, J. T., & Gioia, I. M. 1992, ApJ, 385, 49
Dreher, J. W., Carilli, C. L., & Pedey, R. A. 1987, ApJ, 316, 611
Edge, A., & Stewart ,G. 1991, MNRAS, 252, 414
Edge, A., Stewart, G., & Fabian, A. 1992, MNRAS, in press
Eyles, C. J., et a1. 1991, ApJ, 376, 23
Fabian, A., Hu, E. M., Cowie, L. L., Grindlay, J. 1981, Ap J, 248, 47
Fabian, A., Nulsen, P. E. J., Canizares, C. 1982, MNRAS, 201,933
Fabian, A ., Nulsen, P. E. J., Canizares, C. 1984, Nature 311, 733
Fabian, A., Nulsen, P. E. J., Canizares, C. 1991, Astr. Ap. Rev., 2,191
Fabian, A., Arnaud, K., Nulsen, P. E. J., & Mushotzky, R. F. 1986, ApJ, 305, 9
Fabian, A. 1992, in Clusters and Superclusters of Galaxies, ed. A. C. Fabian, NATO ASI
series, 151
Forbes, D., Fabian, A., Crawford, C., Johnstone, R. 1990, MNRAS, 244, 680
Forman, W. R. 1988, in Cooling Flows in Clusters of Galaxies, ed. A. Fabian, Kluwer, 17
Ge, J.-P. 1991, PhD Thesis, New Mexico Inst. of Mining and Technology
Gusten, R., & Metzger, P. G., Vistas Astr., 26, 159
Hatsukade, I. 1992, PhD Thesis, University of Osaka
Heckman, T. M . 1981, ApJL, 250, L59
Heckman, T. M., Bawn, S., van Breugel, W., & McCarthy, P. 1989, ApJ, 338,48 (HBvBM)
Henriksen, M. J ., & Mushotzky, R. F., 1985, ApJ, 292, 441
Henriksen, M. J. 1985, PhD Thesis, University of Maryland
Henriksen, M. J. 1991, in Physical Cosmology, ed. A. Blanchard et aI., Edition Frontiers,
535
Hintzen, P., & Romanishin, W. 1988, ApJL, 324, L17
Holtzman, J. A. et aI., AJ, 103, 691
Hu, E. M. 1988, in Cooling Flows in Clusters of Galaxies ed A. Fabian, Kluwer, 73
Hu, E. M., 1992, ApJ, 391,608
 407

Johnstone, R . M. , Fabian, A. C . 1988, MNRAS, 233, 581

Jones C ., & Forman W. 1984, ApJ, 276, 38
Loewenstein, M., Zweibel, E ., & and Begelman, M. 1991, ApJ, 377, 392
Loewenstein, M., & Fabian, A ., 1990, MNRAS, 242, 120
Loewenstein, M ., et a1. 1992, in preparation
McNamara, B. R., & O ' Connell, R . W. 1992, ApJ, 393, 579
McNamara, B. R., & O'Connell, R. W. 1989, AJ, 98, 2010
McNamara, B . R ., Bregman, J . N., & O'Connell, R . W., 1990, ApJ, 360, 20
Mushotzky, R., et al. 1981, ApJL, 244, L47
Mushotzky, R. 1984, in Physica Scripta, T7
Mushotzky, R ., & Szy~owiak, A . 1987, in Cooling Flows in Clusters of Galaxies, ed A .
Fabian Kluwer, 47
Mushotzky R., 1988, in Hot Thin Astrophysical Plasmas, ed R. Pallovicini, NATO ASI,
273
Mushotzky, R . 1990, in After the First 3 Minutes, ed S. Holt , C. Bennett, V. Trimble , 395
Mushotzky, R., 1992, in Clusters and Superclusters of Galaxies, ed. A . Fabian, 91.
Sarazin, C. L ., & O'Connell, R. W. 1983, ApJ, 258, 552
Sarazin, C . 1988, X-Ray Emission from Clusters of Galaxies, Cambridge University Press
Sarazin, C. 1992, in Clusters and Superclusters of Galaxies, NATO ASI Series, ed. A. C.
Fabian, 131
Sarazin, C . L ., O'Connell, R . W. , & McNamara, B. R . 1992, ApJL, 389, L59
Schwarz, R., Edge , A., Voges, W., Bohringer, H., Ebeling, H. & Briel, U . 1992, A&A, 256 ,
Lll
Serlemitsos P ., et a1. 1991, in Frontiers of X-ray Astronomy, ed. Tanaka and Koyama, 221
Smith, E., et aI. , 1992, ApJL, 395, L49
Sparks, W .B., Macchetto, D., & Golombeck, D., ApJ, 348, 183
Sparks, W.B. 1992, ApJ, submitted
Steigman, G. 1989, in Cosmic Abundances of Matter, ed. J . Waddington, 310
Stewart, G ., Fabian, A ., Jones, C., & Forman, W. 1984, ApJ, 285, 1
Thomas, P ., Fabian, A ., & Nulsen, P . 1987, MNRAS, 228, 973
Thomas , P . A., Fabian, A. C. 1990, MNRAS, 246, 156
Ulmer, M., Cruddace, R., Fenmore, E ., Fritz, G., & Snyder, W. 1987, ApJ 319, 118
Watt, M., et a1. 1992, in press
White, D.A ., Fabian, A. C., Johnstone, R . M ., Mushotzky, R. F., & Arnaud, K. A. 1991,
MNRAS, 252, 72
White, R. E., & Sarazin, C. L. 1987, ApJ, 318, 612
White, R. E ., & Sarazin, C . L. 1987, ApJ, 318, 621
 EVOLUTION OF CLUSTERS AND THE INTRACLUSTER
MEDIUM
MEGAN DONAHUE
Carnegie Observatories, Pasadena CA

Abstract. I review observations and theoretical models of evolution of gas in clusters of

galaxies. The mass of hot gas is an important (:omponent of clusters of galaxies, equalling
or exceeding the total mass in galaxies in most clusters. The hot gas also traces the
gravitational potential of the cluster, giving better measurements of the total mass of
a cluster than those from velocity dispersions of the member galaxies. Clusters are still
evolving, as expected from estimates of their dynamical times, and as observed in X-
ray luminosity functions of clusters at different redshifts and in studies of subclustering.
Theoretical models of how the intracluster gas might respond to a changing potential
are important to interpretation of X-ray and optical data, particularly for clusters at
high redshift. Models based on the standard assumptions of cold dark matter scenarios
(and variations thereof) make predictions about cluster evolution. Finally I discuss the
implications of the high ratios of gas to galaxy masses and of the iron abundances for the
origin of the ICM.

1. Introduction

Astronomers have known since the 1930s that the amount of gravitational
matter in clusters of galaxies might far exceed the total mass in the mem-
ber galaxies, i.e. that the dynamical state of clusters of galaxies might be
dominated by intracluster matter (Zwicky 1933). Work in the last few years
shows that the galaxies in rich clusters comprise not even the dominant com-
ponent of luminous matter, and that that distinction belongs to hot, X-ray
emitting, intracluster gas (e.g., Hughes 1989; David et al. 1990b; Edge &
Stewart 1991ab). The evolution of the gas in clusters directly affects the en-
vironment of member galaxies. Therefore I have chosen to focus this paper
on the evolution and origin of the intracluster medium (ICM). I will briefly
touch upon the connection between this subject and other areas addressed
at this summer school.
Clusters are the most massive virialized systems in our Universe. In hier-
archical clustering models (e.g., Peebles 1980), clusters form from the rare
3 to 4 u peaks of the initial mass fluctuation spectrum. Therefore, knowing
the properties of clusters, particularly cluster masses, mass distributions,
and number densities of clusters as a function of mass, enables us to place
powerful constraints on theories of formation of large-scale structure in the
Universe. Further, the dynamical time in clusters is longer than the age of
the Universe, so we can expect to see signs of recent structure formation
in relatively nearby clusters. The X-rays emitted by hot gas confined in the
gravitational potentials of clusters provides a sensitive probe of the evolution
of those potentials. In this paper, I assume that Ho = 50 km s-1 Mpc-l,
unless otherwise indicated.

409
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, 409-432.
© 1993 Kluwer Academic Publishers.
410

2. Intracluster Medium in Clusters of Galaxies

The intraduster medium (ICM) in dusters of galaxies is a hot, diffuse plasma
characterized by temperatures of (1 - 10) X 10 7 K, and electron densities of
ne ,. . ., 10- 3 - 10- 4 cm- 3 • X-ray luminosities range from 1043 - 1045 erg S-I.
The number of dusters with luminosities at the upper end of this range are
rare; only about 1% of all dusters with luminosities greater than 1043 erg s-1
have luminosities greater than 1044 erg S-I . (See the luminosity function in
Edge et al. 1990.)
The X-ray spectrum of hot , diffuse gas is characterized by thermal bremsstrahl
emission when T > 3 X 10 7 K and line emission from collisionally excited
gas when T < 3 X 10 7 K. The iron abundance in this gas, usually found
by measuring the intensity of the 6.7 Ke V emission line blend of Fe+ 2 4,
Fe+2 5 , and Ni, is roughly 10- 5 iron atoms per hydrogen, or about 1/3 solar
abundance (e.g., Mushotzky 1984). The O/Fe abundance estimated from
high-resolution Einstein (X-ray) FPCS spectra of the 0 VIII and Fe XVII
lines is roughly 3 times solar (Canizares et al. 1982; Canizares, Markert, &
Donahue 1988).

2.1. THE ,8-MODEL AND ,8-PROBLEM

The hot gas in dusters of galaxies has a characteristic sound-crossing time

of:

(1)

If t. is less than the age of the Universe (tHubble) and also less than the
timescale for the evolution of the cluster potential, then one can assume
that the gas is in hydrostatic equilibrium.
If (1) the gas and the duster galaxies respond to the same potential, (2)
the gas is isothermal, and (3) the orbits of the galaxies are isotropic, then p,
the intracluster gas density is related to ngal' the number density of galaxies,
by the following expression:
f3 (2)
P <X ngal ,

where

,8 ==
J.l-m
k~
(1'2 = 0.76
( ( 1 ' ) 2TS-l.
-1 (3)
1000km s
(Cavaliere & Fusco-Femiano 1976; Sarazin & Bahcall 1977). Therefore, the
ratio of the optical velocity dispersion of the galaxies in a cluster and the
temperature of the ICM, known as f3. p ec in the literature, is an estimate of
this idealized ,8. Typical values of f3.pec range from 0.9 - 1.2 (Mushotzky
1984; Edge & Stewart 1991b; Henry & Arnaud 1992).
 411

All Fields
-2
10

(\j
I
C
E
u
..... -3
0 10
'(/)

- (/)

u
><
H -4
10

I05L-____- L__- L__L-~~~~_______ L_ _~_ _L_L_~~~_ _ _J

I 3 10 30 100

R (aremin)

Fig. 1. Plot from Hughes (1989) showing the X-ray surface brightness of the Coma
cluster, with the best fit ,B-model overlaid. Coma does not have a cooling flow, so
the central surface brightness points do not have to be excluded from the fit. The
angular scale at the redshift of Coma is 39h so kpc/arcminute.

If we further assume that the radial distribution of galaxies follows a

King model (and this is a reasonable assumption), then the X-ray surface
brightness distribution of the gas can be fit to the following two-parameter
radial function:

2]-3.8+1/2
Sa;(r) = So [1 + (~) , (4)

where So is the central X-ray surface brightness of the cluster, a is the core
radius, and J3 is known in the literature as J3/it or J3image. This function
accurately describes the radial surface brightness profiles of clusters without
cooling flows and of clusters for which the central 100-200 kpc are excluded
from the fit (e.g., Figure 1.) Typical values for J3image are 0.6 - 0.7 (Jones &
Forman 1984), significantly less than the values for J3~pec. This discrepancy
is known as the "J3-problem."
412

The ",6-problem" can be addressed theoretically by examining models

of duster evolution which include both the ICM and the galaxies and obser-
vationally by examining possible sources of experimental error, particularly
the estimation of the velocity dispersion. Work by Edge & Stewart ( t991b)
shows that ,63pec is a not strong function of temperature (although t he ex-
perimental error is so large that a correlation cannot be ruled out; David,
Forman & Jones 1990). However, i33pec is systematically higher for high ve-
locity dispersion clusters (Edge & Stewart 1991b). Edge and Stewart suggest
that the high velocity dispersions for some clusters may be over-estimated
due to contamination of cluster galaxy sample by foreground and back-
ground galaxies (see also Lucey 1983; Frenk et al. 1990). If cluster evolution
is dominated by mergers, merging subchunps along the line of sight which
have not yet virialized may produce high velocity dispersions that are not
(yet) representative of the cluster gravitational potential. I will address this
further in § 4.1.1.

2.2. CLUSTER GAS AND GRAVITATIONAL MASSES

Because X-ray (free-free) emissivity is proportional to n;To.s , we can es-

timate the amount of gas within a given radius with minimal sensitivity
to l' by integrating the surface brightness over r. However, recall that
,6image ~ 0.6 - 0.7 implies that 5:r(r) ex: r- 3 at large radii, while p ex: r- 2 so
that MgalJ diverges with r . Therefore, one must specify the r to which the gas
mass is measured, and one should beware gas masses estimated from data
which have been extrapolated to larger radii. Typically, the gas mass within
1 Mpc of the richest clusters, is Mga3 '" 10 14 M 0 , about 3 - 5 Mgal (Edge &
Stewart 1991aj David et al. 1990bj Arnaud et al. 1992a). The variation of 3
to 5 is due to different techniques of calculating masses of clusters. Edge &
Stewart use a metric radius and David et al. calculate masses within a set
number of core radii. In poor clusters, the gas mass is nearly that of the sum
of the cluster galaxies (David et al. 1990b; Figure 2). The gas mass in some
clusters turns out to be a substantial fraction (10-40%, depending on model
assumptions) of the virial mass (Hughes 1989; Briel, Henry, & Bohringer
1992; Eyles et al. 1991).
The virial mass can be estimated from the X-ray data using only the
assumption of hydrostatic equilibriwll:

M(r) = -kT(r) (dlogp + dlogT) , (5)

JLmpG d log r d log r

(e.g., Bahcall & Sarazin 1977; Mathews 1978). However T(r) is only known
for a handful of clusters, e.g. Virgo, Coma, and Perseus. Estimations of
T(r) lead to virial masses which are at least consistent (factors of two) with
 413

A2029
X
6 A2256
(j)
X
(j)
X
A401
<t
2: 5

a::
<t Hydro A Perseus
-1 4
-1
W
t-
(j) Como
...... 3 ~
(j)
(j)
<t MKW3s
2: 0
2 AWM7
(j) AI54 X
0
<t
<:> 0
AWM4
MKW9
0.. OMKW4

0 0 I M67

2 3 4 5 6 7 6 9 10

GAS TEMPERATURE ( keY )

Fig. 2. The ratio of gas mass to stellar mass within 5 core radii versus gas temper-
ature for clust~rs and poor groups, as derived by David et al. (1990b). The core
radii of the groups is assumed to be 100 kpc. The diamond represents a canonical
early-type galaxy.

estimates of virial masses from the galaxy orbits. The following relation
seems to hold for rich clusters (Sarazin 1992):

Mgal < MHotGa8 < MDarkMatter • (6)

Observations implying that the gas mass is a significant fraction of the to-
tal mass of the cluster, of order 20% (Hughes 1989; Briel, Henry, & Bohringer
1992), provide a significant challenge to at least two treasured notions in
structure formation, biased galaxy formation and WIMPy flat universes. As
I will discuss later (§ 6), some of the gas must be primordial, since the total
mass in stars in the cluster galaxies is too low for the stars to be the source of
all the hot intracluster gas. Therefore galaxy formation must not have been
very efficient in clusters of galaxies; indeed, galaxy formation may be even
less effident in clusters than in poor groups or the field (David & Blumenthal
1992). This is in contrast to the sense of models of biased galaxy formation
(Kaiser 1984; Bardeen et al. 1986) that predict that galaxy formation is more
414

efficient in dense regions of the universe like clusters of galaxies. Further-

more, as pointed out in audience discussion here and by Briel et al. (1992),
if the matter composition in such clusters of galaxies is representative of the
Universe, then nB/n o ~ 0.2h~03/2, where fiB is the ratio of the density of
baryons in the Universe to the closure density. Homogeneous models of pri-
mordial nucleosynthesis (Walker et al. 1991) require nBh~o = 0.05 ± 0.01 in
order to agree with observations of 4 He and 7Li abundances and limits on the
number of neutrino types; hence, one might conclude that no : : , 0.25h~~/2.
Therefore, if the primordial nucleosynthesis calculations are correct, the es-
timates of gas mass and vi rial cluster masses are correct, and clusters are
typical places in the universe, then exotic particles do not supply enough
"missing matter" to make the universe flat. 1

2.3. DISTRIBUTION OF CLUSTER MASS: Is THE DARK MATTER CEN-

TRJ\.LLY CONCENTRATED?

Sarazin (1988) shows that if the cluster gas temperature monotonically de-
creases with increasing radius, if the gas is in hydrostatic equilibri~ with
the cluster potential, and if the shape of the gas density distribution is
known 2 , then the dark matter must be more centrally concentrated than
the gas. Recent analyses of X-ray data (Hughes 1989) show that the dark
matter is at least as concentrated as the galaxies in the Coma cluster, con-
sistent with Merritt's (1987) analysis of the velocity dispersion of Coma.
Eyles et al. (1991) have analyzed X-ray data from a Spacelab 2 experiment,
which show a trend for the X-ray temperatures to decrease with radius, im-
plying that the dark matter is centrally concentrated with respect to the
gas (Figure 3). The dark matter in A2256, observed by ROSAT, seems to
be centrally concentrated, and the ratio of the binding mass to the gas
mass decreases with increasing radius, perhaps flattening asymptotically at
large radii (Briel, Henry, & Bohringer 1992). Buote & Canizares (1992) give
evidence that the dark matter, as traced by X-ray gas, has a rounder distri-
bution than the galaxies in 5 Abell clusters with flat galaxy distributions.

Gravitational lens studies have not conclusively determined the distribu-

tion of the dark matter relative to the galaxies and the hot gas. Prelim-
inary results are consistent with a hypothesis that the dark matter may
be centrally condensed (Bergmann, Petrosian, & Lynds 1990; Grossman &
Narayan 1988, 1989; Miralda-Escude 1992), but these results depend on

1 I do not address another issue, namely, that the known baryon content in the universe
might fall far short of the baryon density required by primordial nucleosynthesis, implying
that 90% of the baryons in the universe are dark (see Persic & Salucd 1992).
2 Recall that X-ray (free-free) emissivity is ex ne 2 T 1 / 2 , so derived electron density
distributions are minimally sensitive to T.
 415

o
o
-...
tv) en
I I
() 0
O-c~

:::g~
o
"--" I
o
~

~
L
o
~

o 0.5 1
Radius (Mpc)

Fig. 3. The density of mass in galaxies, in hot gas, in dark matter, and in total mass,
as derived by Eyles et al. (1991) from X-ray observations of the Perseus cluster.

multi-parameter models that are not yet well constrained by the lensing
data. A combination of lensing, galaxy velocity, and X-ray data would pro-
vide the most complete picture of the relative distribution of matter in
clusters of galaxies (Soucail 1992).
If the dark matter is more centrally concentrated than the gas or the
galaxies, at least some of the dark matter must be baryonic. The dark matter
must be able to cool to be more condensed than the hot gas, unless one can
inject non-gravitational energy into hot gas and puff it back out. Most of the
baryons in the ICM cannot cool quickly, but the baryons in the very central
regions of a cluster are dense enough to cool radiatively and form a cooling
flow. Most models of cluster evolution do not include cooling because this
central region is too small to be resolved by current simulations. Cooling
flows, even at the large mass cooling rates of several hundred solar masses
per year, cannot be a major source of cluster dark matter. The amount of
dark matter in a cluster exceeds the total gas mass by a factor of at least
3 (cf. Eyles et ai. 1991). A relatively strong cooling flow of several hundred
solar masses a year would only deposit (3 - 6) X 10 12 M0 within the cooling
416

radius of the cluster, typically a few hundred kiloparsecs. Cooling flows may
well supply a baryonic dark matter component of the central galaxy (Thomas
& Fabian 1990), but they would probably have had to be of order a factor
ten larger in the past to contribute substantially to the cluster dark matter
concentrations suggested by the X-ray observations.

2.4. SUBSTRUCTURE IN CLUSTERS

Many nearby clusters show evidence for substructure. Forty to fifty percent
of the X-ray clusters observed with the IPC on the Einstein Observatory
show substructure (Jones 1992; Jones & Forman 1984). 'Thirty to forty per-
cent of nearby clusters show sub clustering in radial galactic velocity and
galaxy spatial data (Dressler & Shectman 1988). Abell 2256, a relaxed rich
cluster, revealed its first hint of substructure in Einstein data (Fabricant et
al. 1989). Long ROSAT observations confirmed the presence of two merging
clusters, (Briel et al. 1991) in both temperature and surface brightness dis-
tributions. The existence of substructure implies that the clusters are still
in the process of formation, and that mergers are an important aspect of
cluster formation. (Roettiger et ale 1992; Mohr et ale 1992; Lucey 1983)

3. Evidence for Evolution

3.1. X-RAY

Clusters are still evolving according to two independent X-ray-based surveys.

Both of these surveys were reviewed by Henry (1992); I summarize their
results here. Such X-ray surveys are superior to optical surveys of similar
nature because they do not suffer the same types of selection problems.
Since X-ray emission is produced by gas in near-hydrostatic equilibrium
in a cluster's gravitational potential, two superposed low-luminosity X-ray
clusters do not sum to resemble a high-luminosity cluster, whereas in the
optical, two poor clusters along the line of sight can look like a rich cluster.
A high flux-limit, all-sky survey compiled primarily from HEAO-1 data
was analyzed by Edge et ale (1990). Log N-log S plots of high X-ray luminos-
ity clusters show a deficiency of low X-ray flux clusters with respect to the
low X-ray luminosity clusters. Since in a flux-limited survey, high luminosity
corresponds to high redshift, this comparison implies that fewer (a factor 3)
high-luminosity [L:I:(2 -10keV) > 6 X 1044 erg s-l] clusters exist at redshifts
of about 0.15.
Gioia et ale (1990) and, later, Henry et al. (1992), studied a low flux-limit,
small solid-angle survey, the Einstein Medium Sensitivity Survey (EMSS),
of 67 clusters. These clusters were selected from sources discovered serendip-
itously in pointed observations by Einstein. The X-ray luminosity func-
tions of clusters in this sample show fewer clusters at high redshift (z =
 417

0.3 - 0.6) with high X-ray luminosities when compared with nearer clusters
(z = 0.14 - 0.20), qualitatively the same trend seen by Edge et al ..
Both of these surveys are limited by the small numbers of clusters in
each survey. Edge et al. sample has only 3 clusters with redshifts greater
than 0.10, for example. The confidence levels of both results are about 3
sigma. Therefore, increasing the sizes of cluster samples with the ROSAT
All-Sky survey and the serendipitous sources found in the deep survey fields
of ROSAT will be important in verifying them.

3.2. OPTICAL

In contrast to X-ray data, the scant optical data available suggest that the
number density of rich clusters does not evolve. Gunn (1990) reports the
detection of five clusters with redshifts greater than 0.7 in the Gunn, Hoessel,
& Oemler (1986) Palomar CCD survey. Of these, two clusters have measured
velocity dispersion of order 1000 km s-1 based on velocity measurements
of about 25 galaxies. "Coma-like" clusters with high velocity dispersions
therefore exist at high redshift. While the existence of such high-velocity
dispersion clusters challenges biased cold dark matter theories (Evrard 1989;
Peebles, Daly, & Juszkiewicz 1989), there is only low statistical confidence
in the claim that they may be as common at high red shift as they are
today. With only one or two objects, one can only derive the probability
that one or more examples of high velocity dispersion objects would appear
given a space density and volume of the survey. Quantification of the BHO
survey selection effects is impossible, as the members were chosen by hand
for contrast against a background that becomes increasingly complex with
redshift. A survey for distant southern clusters (Couch et al. 1991) concludes
that galaxy evolution complicates cluster counting, even in optical surveys
that are machine-selected from uniform data.
How should we interpret high velocity dispersions at high red shift ? Are
these clusters virialized? Is the cluster velocity dispersion enhanced by merger
activity? How does contamination enhance the velocity dispersion? Dressler
& Gunn (1992) have recently published velocity dispersions and colors for
about 30 galaxies in each of 7 clusters at relatively high redshift (0.35 -
0.55) . Only one of these clusters appears to be a superposition of two clus-
ters. These clusters are indeed real. Zabludoff, Huchra, & Geller (1990) find
that the low-redshift clusters (z < 0.1) in their survey have systematically
lower velocity dispersions than somewhat higher redshift clusters. The veloc-
ity dispersions of high-redshift clusters may be affected more by merging and
contamination (Lucey 1983). A high velocity dispersion probably means that
a large mass is present, whether or not the velocities are virialized, because
infall velocities also depend on mass, but quantifying the mass is difficult
without knowing the dynamical state of the system.
418

N-body simulations may give some insight into how to interpret velocity
dispersions. But Couchman & Carlberg (1992), for instance, suggest that the
velocity dispersion of the galaxies underestimates the velocity dispersion of
the dark matter. In their simulations the galaxies are "cooler" than the dark
matter, a result that is contradicted by observations (see § 2.3).
If the number of rich clusters of galaxies with high velocity dispersions
indeed remains at least constant out to redshifts of 0.9 (and this velocity
dispersion reflects the amount of matter present and not an ongoing merger)
and the number of X-ray luminous clusters decreases, one might require an
evolutionary scenario where the galaxies respond to the deepening potential
wells of dark matter before the gas does. In this scenario, the gas must be
too hot to fall into the potential wells until these wells are deep enough to
confine the gas. One might then look for the ratio of hot gas to the total mass
to decrease as the redshift increases. The mass of hot gas can be estimated
from X-ray data, while the cluster mass could be estimated variously from
X-ray temperature, velocity dispersion, and gravitatioilallensing data.

3.3. RADIO

Indirect evidence that the intracluster medium is evolving is provided by

studies of radio source structure, which is affected by the density of the IeM.
Radio galaxies are classified into two types, Fanaroff-Riley I and II (Fanaroff
& Riley 1974). At low redshift, the FR I sources, which are centrally bright,
distorted structures with edge-darkened lobes, are found preferentially in
the first-ranked ellipticals in rich clusters. The FR II sources, the classical
double-lobed sources with hot spots on the lobes, are found in less luminous,
smaller, and more optically active galaxies. These galaxies are either isolated
or in poor groups (cf. Prestage & Peacock 1989, and references therein.)
One exception to this trend is the powerful radio galaxy, Cygnus A, which
is probably in a rich cluster (Spinrad & Stauffer .1982).
At redshifts near 0.5, the environments of FR I radio galaxies are similar
to those of low-redshift FR I radio galaxies, but the high-redshift FR II's
are often in a richer environment than are local FR II's (Hill & Lilly 1991;
Allington-Smith et ai. 1992). Similar trends are seen in other studies of radio
galaxies (Yates et ai. 1988) and the environments of radio-loud and radio-
quiet quasars (Yee & Green 1987; Ellingson et ai. 1991, 1992; Craven et ai.
1992.) One might speculate that the intracluster medium of rich clusters
began to increase in density at intermediate redshifts, after the galaxies had
formed a cluster core. The ICM could have originated in two ways: (1) the
intracluster gas could have been ejected from the cluster galaxies, or (2) that
the intracluster gas was initially largely intergalactic and somewhat hot and
thus did not fall into the cluster until the gravitational potential had grown
large enough to confine the gas. In the latter alternative the primordial gas
 419

is enriched by galactic winds. The large ratios of gas to total mass in clusters
support the ICM infall picture. In a delayed infall scenario, high-redshift rich
clusters with FR II's would be systematically less X-ray luminous than rich
clusters at low redshift.

4. Models for ICM Development

Here, I sketch the highlights of the history of models of the intracluster

medium. Gunn & Gott (1972) constructed analytic models of cluster col-
lapse and infall of primordial intergalactic medium. Perrenod (1978) set the
stage for all subsequent cluster evolution models, with his 1D simulation
of gas in a potential modeled by an N-body simulation of White (1976).
The simulated clusters evolved more rapidly than observed clusters because
the N-body simulation assumed the entire virial mass of a cluster was in the
galaxies (Sarazin 1988). Cavaliere et al. (1986) studied a collisionless N-body
model and assumed the gas remained in hydrostatic equilibrium with the
changing potential of the collisionless fluid. Soine of the most recent work
has been done by Evrard (1989, 1990) who studies a two-fluid simulation
of a collisionless fluid and a collisional fluid in clusters of galaxies. See also
Thomas & Couchman (1992), which appeared after this talk was given. The
initial conditions in these fluids are specified by the perturbations predicted
of a CDM-dominated universe.
Such models are important to test our assumptions about the intracluster
gas, particularly when we rely on such assumptions to interpret the observa-
tions. Such models seek to answer questions like: Can the X-ray data, when
combined with the assumption of hydrostatic equilibrium, reliably provide
cluster masses? Can galaxy velocities be "hotter" or "cooler" than the gas
temperature? How is collisionless dark matter distributed with respect to the
gas and the galaxies? Are these distributions consistent with observations?

4.1. EVRARD'S MODELS; A CASE STUDY OF A RICH CLUSTER

Evrard's models (1990) couple gas dynamics with N-body simulations to

follow the evolution of a single cluster as it collapses and merges with sub-
clusters. Radiative cooling is ignored, so shock heating and PdV work are
the only heat inputs to the ICM. The gas streams along filaments and walls
of collisionless matter and converges on their intersections. There the gas
shocks, and the shock wave propagates out through the accreted material,
heating it to the temperature of the cluster potential. The main results of
the models are:
1. Clusters grow via mergers of smaller groups and accretion of intergalactic
gas.
420

2. Mergers do not create shocks. Rather, a compression wave forms, heating

cluster gas between the two merging elements.
3. The cluster gas is nearly isothermal, as the temperature decreases only
20% in 1 Mpc in the simulation; but the gas is ,not polytropic. This
is consistent with X-ray observations (Hughes 1989; Hughes & Tanaka
1992; Arnaud et al. 1992b)
4. The gas and the dark matter do not have significantly distinct spatial
distributions. The gas is somewhat more spherical than the dark matter.
5. The X-ray luminosity of a cluster increases with time, but this trend is
punctuated by merger events, which create sudden increases in X-ray
luminosity.
6. Initial gas temperatures of 10 million degrees were required to reproduce
typical centr3l surface brightnesses of clusters. Other cluster character-
istics were well reproduced by the simulations.
The main result of the models is that the estimation of cluster masses
using the hydrostatic, isothermal models with measured {3 can be made
with about 10% accuracy if scaled upward by a factor of 30%. The virialized
regions of clusters of galaxies are within a radius of 3h si Mpc, so mass
estimates that approach this radius may be shaky.

4.1.1. A solution to the "{3 - problem"?

Evrard's simulations also have a "{3-problem" (§ 2.1)! In the simulations,
{3~pec > 1 and {3/it ~ 0.76. The discrepancy between {3/it and {3.pec has
been thought to arise from overestimations of the velocity dispersions of
clusters (Edge & Stewart 1991b; Frenk et al. 1990; Sarazin 1988; § 2.1) or
non-isothermal gas. Models and data show that most of the ICM gas is
nearly isothermal, and better determinations of velocity dispersions for X-
ray clusters (more galaxies) have reduced the velocity dispersions for some
clusters, but not all clusters. The discrepancy in the simulation is due to
several sources: incomplete thermalization (85 %) of the gas, leading to some
turbulent support in the ICM, anisotropic galaxy velocity dispersions, and
the deviation of the dark matter distribution at large radii from a King
model. The most important contributor is incomplete thermalization. The
effect is probably not numerical, because it persists when the number of
particles in the simulations is increased. Since the longest equilibrium time
scale in the gas for T = 108 K and electron density ne = 10- 3 cm- 3 is
only 6 x 108 yr (proton-electron equilibriwn), "incomplete thermalization"
may imply that the beta-discrepancy arises because the gas has not yet
thermalized from recent merger events. The bottom line from the simulation
is that the galaxy velocites are "hotter" than the gas temperature, because
not all of the kinetic energy in the gas has been thermalized.
But not all clusters, particularly poor clusters, have {3~pec > 1. The gas
 421

in these clusters appears to be "hotter" than the galaxy velocities. This

could be due to mechanical heating by galactic winds, which supplement
the ICM energy budget (deYoung 1976; David, Jones & Forman 1991), and
are unaccounted for in Evrard's models. Mechanical heating is important
particularly if the cluster is a low-temperature, poor cluster. This scenario
predicts that there should be a trend of lower {3,pec with lower cluster tem-
perature or richness. This prediction is consistent with current data (Edge &
Stewart 1991b), but improved limits on both velocity dispersion and cluster
temperatures over a large range of temperatures are required.
Future models of cluster evolution should include cooling processes and
heating and enrichment by galaxy winds. While some of the dark matter
component of clusters is dissipationless, some of it may not be. The distri-
bution and character of dark matter in clusters of galaxies is open to further
investigation.

5. Models for Evolution of Cluster Luminosity and Temperature

Functions

5.1. COMPARISON OF EVOLUTION MODELS AND DATA

In assessing models for the evolution of cluster populations, we must be able

to compare the observations with the models. Deriving the observable quan-
tities from the models often involves some non-trivial assumptions, which I
discuss here. Both the current distribution of the ICM temperatures, char-
acterized by the temperature function N(T), the number density of clusters
with temperature greater than T, and the evolution of cluster X-ray lumi-
nosity functions are sensitive to the initial density fluctuation spectrum that
produced structure in the universe (see the contributions of Bond, Bennett,
and Davis in these proceedings.) Several recent studies have attempted to
constrain this fluctuation spectrum by comparing various models' predic-
tions for N(T) and the X-ray luminosity function to the growing body of
X-ray data on clusters. This section briefly reviews the implications of these
comparisons.
Bits of jargon often appear in discussions about the evolution of the
cluster population, but as first-year graduate students soon discover, single
phrases of jargon occasionally replace paragraphs of explanatory text, but
sometimes serve only to link topics of investigation without saving trees. In
the context of cluster evolution, standard cold dark matter only implies
that the power-law index of the initial fluctuation spectrum is n ~ -1 at
the mass scale of M rv 10 15 M 0 . Self-similar evolution implies that all
cluster properties scale similarly as they evolve: clusters look alike and differ
only in size. Self-similar evolution, as Kaiser (1991) points out, occurs only
if the sole source of heat is shocks due to the gravitational collapse and if a
422

constant fraction of baryons wind up in clusters. Cool clusters, for example,

may not evolve self-similarly because galactic winds may affect the cluster
energy budget.
The Press-Schechter (1974) relation provides an analytic expression for
the number of collapsed objects of mass M per unit volume per unit mass.
This relationship is difficult to test observationally, but it gives a reason-
able approximation to the results of N-body simulations. For example, the
predictions of Press-Schechter were compared with N-body simulations in
Frenk et al. (1990) by Henry & Arnaud (1991). Henry & Arnaud found good
agreement, except at temperatures lower than about 3 KeV. This discrep-
ancy may be due to their assumption that f3 ~ 1.2 for all clusters. They
needed to assume a f3 to convert from the model output from mass to veloc-
ity dispersion. As mentioned above, f3 could be less for cool clusters because
galactic winds may become important contributors for the heating in those
clusters.
U sing the Press-Schechter formula, one can go from a description of
the density perturbation spectrum to a description N(M) of the number of
collapsed objects per unit volume per unit mass. From N(M) we can derive
the luminosity function N(L) or temperature function N(T) if we know the
relationship between the quantites L and M or T and M. The T - M relation
comes from assumptions of hydrostatic equilibrium or simulations such as
Evrard (1990). The L - T relation is derived empirically (e.g., Mushotzky
1984; Edge et al. 1990). This relation, illusrtated in Figure 4, shows about
a factor 3 scatter in temperature for a given luminosity.
In order to convert from observed luminosity to bolometric luminosity,
energy band and absorption corrections must be made. Energy corrections
("K-corrections") based on temperature are probably not critical for hot
clusters observed by Einstein and ROSAT observatories, since those satel-
lites are most sensitive to soft X-rays, but corrections could be important for
clusters observed by HEAO-1 and EXOSAT, as high energy cut-offs for the
spectra of cool clusters fall into the energy bandpasses of these telescopes.
Conversely, any variation in absorption by cool gas in the cluster (not ac-
counted for in current models) will affect the bandpasses of ROSAT and
Einstein, while corrections for HEAO-1 and EXOSAT bandpasses are rela-
tively insensitive to absorption. Considerations like these could be important
when comparing the results of surveys by different instruments.

5.2. A DISCUSSION OF SELF-SIMILAR MODELS AND CDM ASSUMPTIONS

5.2.1. The temperature function and the power-law indez n

If the Press-Schechter analytic relation is valid, the shape of the tempera-

ture function indicates that n ~ -1.7~g:~~ (Henry & Arnaud 1991; see also
 423

......---
III

CD
1-..
<lJ
10 45
>.
III
0
t:
E
::J
.....J
()

1-..
-'
<lJ 1044
E
0
0
OJ
<>
10°
Temperature (keV)

Fig. 4. Bolometric X-ray luminosity versus temperature, from Edge & Stewart
(1991) . Four of the brightest clusters (Perseus, Coma, Virgo, and Triangulum Aus-
tralis) are marked with diamonds. Note that the scatter is about a factor of 3.

Lilje 1992). This result implies one of two things: either the Press-Schechter
relation is wrong, or n t -1 at the cluster scale, in conflict with the CDM
standard model.

5.2.2. Self-similar Models

We now examine the self-similar models in light of the temperature function/
Press-Schechter result that n :::::: -2. In self-similar models gas begins' at
some initial temperature and falls into dark-matter potential wells. As these
wells merge, the gas shocks and heats as needed to remain in hydrostatic
equilibrium. Kaiser (1986) describes the evolution of clusters where the gas
starts off cool. These self-similar models predict that high velocity dispersion
clusters evolve dramatically from z '" 1 to now when n :::::: -2. The optical
data of Gunn and Dressler suggest that the numbers of rich clusters may
remain high out to redshifts of 0.75. More significantly, models with power-
law indices that are greater than -1, show little or no evolution, a result
424

which is inconsistent with the evolution seen in the X-ray luminous clusters.
Self-similar models do not fit the X-ray luminosity function evolution even
for n = -1 (Kaiser 1991). These two results prompted Kaiser (1991) to
conclude that the self-similar models he originally proposed were wrong (see
also Evrard & Henry 1991).

5.2.3. Can Self-similar Models be Saved or Do We Need Big Computers?

Kaiser (1991) proposes that if the proto-cluster gas were heated, and thus
shifted to a higher adiabat, the evolution of clusters would proceed differ-
ently. Although cluster cores with high velocity dispersions can form rel-
atively quickly, infall of most of the cluster mass occurs more gradually.
Pre-heated cluster gas accumulates in large quantities only after a large
fraction of the cluster's mass has fallen in and the gravitational potential
has become deep. This delays the evolution of the X-ray properties of the
cluster. One prediction of this theory, proposed to explain the evolution (or
lack thereof) observed by both the optical and the X-ray astronomers, the
rich clusters at high redshift should show systematically less X-ray emission
than their counterparts at low redshift. Preheating the IGM also seems to be
required for the N-body /3D hydro codes to fit the X-ray surface brightness
data (e.g., Evrard 1990), but it is not clear how seriously to take this result.
There are two major problems with the revised self-similar model: (1)
The model assumes that the gas is isentropic, but the gas is very nearly
isothermal over much of the cluster, dropping only in temperature 10-20%
at the outer radii, and (2) an'initial condition in the model is that the power-
law index n = -1, so the Press-Schechter relation as applied to the cluster
temperature function would still have to be incorrect.
The standard self-similar model was generalized to a scaling model by
Evrard & Henry (1991; see also Pierre 1992) who set:

(7)
where LBoL is the bolometric luminosity, L 15 is the luminosity of a 1015 Mev
cluster, M is the cluster mass, and z is the cluster redshift. They compare
such models to the EMSS luminosity functions at three different redshifts
(Gioia et al. 1990; see Figure 5) and find that they can also rule out the
self-similar model (CDMSS, in their terminology) with a high degree of
confidence (see Table 1 for model parameters.)
Two of their models fit the luminosity function data equally well; CDMP3,
which assumes n = -1 as in CDM, and XCDM, which assumes n = -2 as
consistent with the X-ray temperature function and Press-Schechter predic-
tions. The two model families both predict that typically 200 clusters per
steradian will be detected in a 600 second ROSAT PSPC exposure, if a 4.8'
by 4.8' detection cell is used. The North Ecliptic Pole deep survey should
 425

-6 <z> 0 . 17

-8
.....,....-....
I
,....-....
I
(/)
-10

ttO
h
(j) -6 <z> = 0 .24
~
0
,.....-i
............,
-8
M
I
C)
p..
::E
............, -10
,....-....
.........:I
............,
-6 <z> 0.33
~
ttO
0
.........:I -8

-10

44 44.5 45 45.5

Log Lx [0.3-3.5 keY] (erg S-1)

Fig. 5. The X-ray luminosity functions for clusters of galaxies in three redshift shells
from the EMSS sample. The best fitting scaling models (CMDP3 and XCDM, see
Table 1) are overlaid with solid lines. The standard self-similar model (CDMSS)
is overlaid with a dashed line, and is not a good fit (figure from Evrard & Henry
1991).
426

TABLE I
Parameters of Model Families for Fits to Luminosity Function Data

p 8

CDMSS -1 5.5 4/3 7/2

CDMP3 -1 5.5 3 7/2
XCDM -2 7 11/6 11/4

detect 330 clusters with Fz ~ 9 X 10- 14 ergs- 1 cm- 2 • Approximately 50% of

these clusters will have z ~ 0.2, and only 10% will have z ~ 0.4. These esti-
mates were made assuming that the galaxies do not play an important role
in cluster heating. One surprising inference from both Kaiser's and Evrard
& Henry's work is that X-ray observations may not be that helpful in iden-
tifying very distant clusters!
The two models can be distinguished observationally. XCDM predicts
more power at large scales than does CDM, which may be consistent with
angular clustering correlations (REF). A minimum core entropy is set in
the core by preheating the IGM at redshifts near 5. The gas distribution is
non-isentropic, and the predicted temperature function is consistent with
Press-Schechter. CDMP3 predicts that the intracluster gas fraction is a
strong function of the ,cluster mass or richness (heM ex: M5/6). CDMP3,
together with the observed cluster temperature function, implies that the
Press-Schechter relation is incorrect. One possible way to distinguish be-
tween the two models is to measure the fraction of clusters with temper-
atures greater than 4 keY in the NEP survey. XCDM predicts 10% and
CDMP3 predicts 40%.

6. Source and Enrichment of the ICM

The cluster gas mass is substantially greater than the amount of mass in
the galaxies. The ratios of Mgall/Mgal '" 1 - 5, increasing from groups to
rich clusters (David, Forman, & Jones 1990 = DFJj Arnaud et al. 1992a
= ARBVV). Therefore most of the ICM gas is primordial, left over after
the formation of galaxies. However, the [Fe/H) is 0.35 ± 0.2 solar and does
not vary much from cluster to cluster within observational uncertainties.
So some of the gas must have been processed through stars and ejected or
stripped from the member galaxies.
I review here the recent work of ARBVV and DFJ. ARBVV point out
that the ratio of mFe/m. (where m. is the mass in stars in the cluster)
corresponds to a parameter in galaxy evolution models known as the "yield."
 427

The yield is the amount of iron that can be produced by a population of

stars divided by the sum of the masses of the stars. This can be estimated
by multiplying two factors: the metallicity mFe/mga8 and the gas to stellar
mass ratio mgas/m•. Rich clusters have an implied yield of 0.5 solar times 6
(a typical gas-to-stellar mass ratio for rich clusters), but the implied yield of
the Galaxy is 1.0 solar times 0.15. Thus, the yield for clusters, or mFe/m., is
about a factor of 10 greater for clusters of galaxies than for our own Galaxy.
Both groups agree that a normal initial mass function (IMF) for star for-
mation does not reproduce this ratio. DFJ claim that a flat IMF is required
while ARBVV insist that a bimodal IMF must be assumed. The difference
between the two groups arises because ARBVV include recycling by subse-
quent stellar formation, and claim that the iron mass locked up in low-mass
stars and remnants is not negligible. Both groups agree that a maximum
of 10-40% of the mass in an IMF can be ejected by a galaxy, and that the
dominant ejection process is an early wind phase, in which Type II super-
novae provide most of the ejecta and energy. The dominance of Type II
supernovae is consistent with the suggestion that the ICM is overabundant
by a factor of 3 or so in oxygen (Canizares et al. 1982, 1988), since Type II
supernovae produce oxygen much more effectively than Type I supernovae.
Ram pressure stripping is not necessary to enrich the ICM.
These models make two important predictions: (1) that the iron abun-
dance should vary with X-ray temperature or richness, because the iron is
more dilute in the richest clusters, and (2) that galaxies are important source
of heat input, particularly for the poor clusters. The former prediction is con-
sisten~ with iron abundance data presented in DFJ. The latter prediction
implies that the ,B-parameter, the ratio between the energy in galaxy orbits
and the energy in the gas, should be smaller for poorer clusters, a trend that
is not excluded by current data.

7. Cold Gas in Clusters of Galaxies

No discussion of evolution of the ICM would be complete without at least
briefly reviewing aspects of cold gas in the ICM because this gas can affect
the appearance, luminosity, and the character of cooling of a cluster. Neutral
hydrogen (21-cm) observations show absorption towards radio sources in
clusters (McNamara et al. 1990). X-ray spectral observations of clusters
show neutral gas absorption above that expected from the neutral gas in
our own Galaxy (White et al. 1991; Wang & Stocke 1992). And, infrared
emission in clusters is detected in IRAS maps (Bregman et al. 1990; Wise et
al. 1992), which suggests that a substantial amount of dust may be present
in the ICM. Further evidence for dust is provided by extinction estimates
which compare the Lyand H,B emission from optical filament systems in
clusters of galaxies (Hu 1992, 1988).
428

Ram pressure stripping may remove cool gas and dust from galaxies,
although it is an understatement to say this process is not well understood.
Ram pressure of the ICM upon the interstellar medium of galaxies may
provoke the activity level of gas rich galaxies that are falling into the cluster
for the first time. This scenario of "first infall" is consistent (Evrard 1991)
with observations (Dressler 1987; Gunn 1989), although mergers and tidal
encounters also are important (e.g., Lavery, Pierce, & McClure 1992). It is
conceivable that some of the cool ISM may be removed at this time. Wise
et al. (1992) and Bregman et at. (1989) point out that the destruction rate
for dust, as observed in far-infrared radiation, in the ICM is approximately
equal to the production rate of dust by stars in the cluster galaxies. Coherent
patches of stripped ISM may provide low-entropy seeds for cooling flows
(Sparks et al. 1989).
If ram-pressure stripping is important for supplying dust and cool gas
to the ICM, then since this process is more efficient at high redshift (which
would be true if ram-pressure stripping accounts for the higher fraction of
active galaxies in clusters at higher redshifts), then one might expect to
see evolution in the signatures of cool gas in ICM. One example might be
that cooling flows, or the fraction of clusters with central galaxies that ex-
hibit optical filamentation systems, may increase with redshift, as suggested
by Donahue et al. 1992. Another trend may be that higher column densi-
ties of X-ray absorbing material may exist in distant clusters than in those
nearby (Wang & Stocke 1992). Finally, if clusters at high red shift have more
cool material, the galaxies in distant clusters would be more reddened than
galaxies in clusters nearby.

8. Conclusion

I leave you with several things to remember:

• X-ray observations of the best-studied clusters show that the hot gas in
clusters is the second-largest component of mass in clusters of galaxies
(behind the dark matter, but ahead of the galaxies.) Mass estimates
from X-ray observations are sensitive to the assumed shape of the dark
matter distribution, so X-ray astronomers should not assume that the
dark matter is distributed like the galaxies. Further, astronomers should
not rely on assumptions that the gas and the galaxies are virialized
beyond large radii of order 3 Mpc or so .
• X-ray observations suggest that the dark matter may be more centrally
concentrated than the gas or the galaxies. These suggestions rely on
assumptions that the gas is in hydrostatic equilibrium, an assumption
that may not hold for an entire cluster, and on observed temperature
distributions that show that the gas temperature is slowly decreasing at
large radii. Further observations are needed to confirm these suggestions.
 429

• There exists ample evidence that the ICM is evolving in the sense that
there are fewer high X-ray-Iuminosity clusters in the past than there are
now.
• Preheated ICM models fit the data the best (so far); but we do not yet
know what causes this pre-heating. (It is difficult enough find sources
of ionization for even a tenuous intergalactic medium - finding energy
sources sufficient to heat it to 107 K may prove even more challenging.)
• Standard Cold Dark Matter assumptions of n = -1 and Press-Schechter
predictions are inconsistent with temperature functions of clusters at the
cluster mass scale of 10 15 M 0 .
• Galaxy winds are a non-negligible contributor to the cluster gas energy
budget, particularly in cool or poor clusters.
• "The present is very much the epoch of cluster formation" (Gunn & Gott
1972, italics theirs.) " ... we may be observing clusters at a confusing stage
in their development, but ... we can feel very encouraged." (Rees 1992).

Acknow ledgements
I thank Harley Thronson and Mike Shull for organizing a lively meeting in
such a breathtakingly beautiful place. I had helpful discussions with Gus
Evrard, Pat Henry, and John Stocke. Figures were kindly provided by Gus
Evrard, Jack Hughes, Alistair Edge, Chris Eyles, and Larry David. I also
acknowledge Mark Voit for listening to and providing a critique of two earlier
versions of this talk.

References
Allington-Smith, J. R ., Ellis, R. S., Zirbel, E. L., & Oemler, Jr. A. 1992, ApJ, subIllitted.
Arnaud, M., Rothenfiug, R., Boulade, 0., Vigroux, L. & Vangioni-Flam, E. 1992a, AA,
254, 49. (ARBVV).
Arnaud, M., Hughes, J. P ., Forman, W., Jones, C., Lachieze-Rey, M., Yamashita, K., &
Hatsukade, I. 1992b, ApJ, 390, 345.
Bahcall, J. N. & Sarazin, C. L. 1977, ApJ, 213, L99.
Bardeen, J. M., Bond, J. R., Kaiser, N., Szalay, A. S. 1986, ApJ, 304, 15.
Bergmann, A. G., Petrosian , V ., & Lynds R. 1990, ApJ, 350, 23.
Bregman, J. N., McNamara, B. R., & O'Connell, R. W. 1990, ApJ, 351, 406 .
Briel, Henry, & Bohringer, 1992, AA, 259, L31.
Briel, U. G . et aI. 1991, AA, 246, LI0.
Buote, D. A. & Canizares, C. R . 1992, ApJ, submitted.
Couch, W.J, Ellis, R . S., Malin, D. F., MacLaren, I. 1991, MNRAS, 249, 606.
Couchman, H. M. P . & Carlberg, R . G . 1992, ApJ, 389, 453.
Canizares, C. R., Markert, T. H., & Donahue, M. 1988, in Cooling Flows in Clusters and
Galaxies, ed. A. C. Fabian (Dordrecht: Kluwer), p. 63 .
Canizares, C. R., Clark, G. W., Jernigan, J. G., Markert, T. H. 1982, ApJ , 262, 33.
Cavaliere, A. & Fusco-Femiano, R . 1976, AA, 49,137.
Cavaliere, A. , Santangelo, P., Tarquini, G., & Vittorio, N. 1986, ApJ, 305, 651.
Craven, S. E., Hickson, P., & Yee, H. K. C . 1992, in Contributed Papers to Evolution of
Galaxies and Their EnvirollIllent, in press.
430

David, L. P., Forman, W., & Jones, C. 1990a, ApJ, 359,29. (DFJ).
David, 1. P., Arnaud, K., Forman, W., & Jones, C. 1990b, ApJ, 356,32.
David, L. P. & Blumenthal, G. R. 1992, ApJ, 389, 510
deYoung, D. S. 1976, ApJ, 223, 47.
Donahue, M., Stocke, J. T., & Gioia, I. M. 1992, ApJ, 385, 49.
Dressler, A . & Shectmann, S. A. 1988, AJ, 95, 985.
Dressler, A. & Gunn, J. E. 1992, ApJS, 78, 1.
Dressler, A. 1987, ApJ, 317, 1.
Edge, A. C. & Stewart, G. C. 1991a, MNRAS, 252, 414.
Edge, A. C. & Stewart, G. C. 1991b, MNRAS, 252, 428.
Edge, A . C., Stewart, G. C., Fabian, A. C. & Arnaud, K. A. 1990, MNRAS, 245, 559.
Ellingson, E. & Yee, H. K. C. 1992, in Contributed Papers to The Evolutions of Galaxies
and Their Environment, in press.
Ellingson, E., Green, R . F ., & Yee, H. K. C. 1991, ApJ, 378, 476.
Evrard, A. E. 1991, MNRAS, 248, 8P
Evrard, A. E. 1990, ApJ, 363, 349.
Evrard, A. E. 1990, in Clusters of Galaxies, ed. M. Fitchett and W. Oegerle, p. 287.
Evrard, A . E . & Hemy, J. P. 1991, ApJ, 383, 95 .
Eyles, C. J., Watt, M. P ., Bertram, D., Church, M . J ., Ponman, T. J., Skinner, G. K . &
Willmore, A. P. 1991, ApJ, 376, 23.
Fabricant, D . G., Kent, S. M., & Kurtz, M. J. 1989, ApJ, 336, 77.
Fanaroff, B. L. & Riley, F. M. 1974, MNRAS, 167, 31p.
Frenk, C. S., White, S. D. M., Efstathiou, G., & Davis, M. 1990, ApJ, 351, 10.
Gioia, I. M., Henry, J. P., Maccacaro, T., Morris, S. L., Stocke, J. T., & Wolter, A. 1990,
ApJ, 356, L35.
Gunn, J. E . 1989 in The Epoch of Galaxy Formation, eds. C . Frenk, R. Ellis, T. Shanks,
A. F. Heavens, J. A. Peacock, (Kluwer: Dordrecht), p. 167.
Gunn, J. E., & Gott, J. R . 1972, ApJ, 209,1.
Gunn, J. E., Hoessel, J. G., & Oke, J. B. 1986, ApJ, 306, 30.
Gunn, J. 1990, in Clusters of Galaxies, ed. W. Oergerle, M. Fitchett, & L . Danly, (Cam-
bridge: Cambridge Univ. Press), p. 341.
Grossman, S. A. & Narayan, R., 1988, ApJ, 324, L37.
Grossman, S. A . & Narayan, R., 1989, ApJ, 344, 637.
Henry, J. P., 1992, in Clusters and Superclusters of Galaxies, (Kluwer: Dordrecht), ed. A.
C. Fabian, p . 311.
Henry, J. P., Gioia, I. M., Maccacaro, T., Morris, S. L ., Stocke, J. T., and Wolter, A. 1992,
ApJ, 386, 408.
Henry, J. P. & Arnaud, K . A. 1991, ApJ, 272, 410.
Hill, G. J. & Lilly, S. J. 1991, ApJ, 367,1.
Hu, E. M. 1988, in Cooling Flows in Clusters and Galaxies, ed. A. C. Fabian, (Dordrecht:
Kluwer), p . 73.
Hu, E. M. 1992, ApJ, 391, 608.
Hughes, J. P. 1989, ApJ, 337, 21.
Hughes, J .P. & Tanaka, T . 1992, ApJ , 398, 62.
Jones, C. & Forman, W. 1984, 1984, ApJ, 276, 38.
Jones, C. & Forman, W. 1992, in Clusters and Superclusters of Galaxies, (Kluwer: Dor-
drecht), ed. A. Fabian, p. 49.
Kaiser, N. 1984, ApJ, 284, L9.
Kaiser, N. 1986, ApJ, MNRAS, 219, 785 .
Kaiser, N. 1991, ApJ, 383,104.
Lavery, R . J., Pierce, M. J., & McClure, R. D. 1992, preprint.
Lilje, P. B. 1992, ApJ, 386, L33.
Lucey, J. R. 1983, MNRAS, 204 , 33.
Mathews, W. G. 1978, ApJ, 219, 413.
McNamara, B. R., Bregman, J. N., & O'Connell, R. W. 1990, ApJ, 360, 20.
 431

Merritt, D. 1987, ApJ, 313, 12l.

Miralda-Escude 1992 , ApJ, 390, L65.
Mohr, J. J., Fabricant, D. G ., & Geller, M. J., in Contributed Papers of The Evolution of
Galaxies and Their Environment , in press.
Mushotzky, R. F . 1984, Phys. Scripta, T7, 154.
Perrenod, S. C. 1978, ApJ, 224, 285 .
Persic, M. & Salucci, P. 1992, MNRAS, 258, 14p.
Peebles, P . J. E . 1980, The Large Scale Structure of the Universe, (Princeton University
Press: Princeton), p. 120.
Peebles, P . J . E ., Daly, R . A., & Juszkiewicz, R . 1989, ApJ, 347, 563.
Pierre, M . 1991, AA, L23 .
Press, W . H. & Schechter, P. 1974, ApJ, 187, 425.
Prestage , R . M . & Peacock, J. D . 1989, MNRAS, 236, 959.
Rees, M. J. 1992, in Clusters and Superclusters of Galaxies, (Kluwer: Dordrecht), ed. A.
Fabian, p. l.
Roettiger, K., Burns, J ., & Loken, C . 1992, in Contributed Papers of The Evolutions of
Galaxies and Their Environment, in press.
Sarazin, C . L . & Bahcall, J. N. 1977, ApJS, 34, 45l.
Sarazin, C . L . 1988 , "X-ray Emission from Clusters of Galaxies" , Cambridge: Cambridge
University Press.
Sarazin, C . L. 1992, in Clusters and Super clusters of Galaxies, ed. A. C. Fabian, (Dor-
drecht : Kluwer), 13l.
Soucail, G. 1992, in Clusters and Superclusters of Galaxies, (Kluwer: Dordrecht), ed. A.
C. Fabian, p . 199.
Sparks, W. B., Macchetto, F ., & Golombek, D. 1989, ApJ, 345,153.
Spinrad, H. & Stauffer, J. R . 1982, MNRAS, 200,153.
Thomas, P. A . & Fabian, A . C. 1990, MNRAS, 246, 156.
Walker, T. P ., Steigman, G ., Schramm, D . N ., Olive, K. A., & Kang, H . 1991, ApJ, 376,
5l.
Wang, Q. & Stocke, J. T . 1992, in Contributed Papers to Evolution of Galaxies and Their
Environment, in press .
White, D. A., Fabian, A . C., Johnstone, R. M ., Mushotzky, R . F., & Arnaud, K . A. 1991,
MNRAS, 252, 72.
White, S. D. M., 1976, MNRAS , 177, 717 .
Wise, M. W., O'Connell, R . W ., Bregman, J . N., Roberts, M. S. 1992, ApJ, in press.
Yahil, A . & Ostriker, J. P. 1973, ApJ, 185, 787.
Yates , M . G ., Miller, L., & Peacock, J. A. 1989, MNRAS, 240,129 .
Yee, H . K. C. & Green, R. F. 1987, ApJ, 319, 28 .
Zabludoff, A. I., Huchra, J . P., & Geller, M . J . 1990 ApJS , 74, l.
Zwicky, F. 1933, Helv. Phys, Acta, 6, 110.
 THE QUASAR LUMINOSITY FUNCTION
B.J.BOYLE
Institute of Astronomy, University of Cambridge, Madingley Road,
Cambridge CB3 OHA, U.K.

Abstract. Recent work on the quasar luminosity function at radio, optical and X-ray
wavelengths is reviewed. It is demonstrated that the evolution of the quasar luminosity
function in all three regimes is marked by a strong and approximately similar power-law
increase in luminosity, L oc (1 + z )3±O.S, between the present epoch and z :::::: 2. At z > 2, a
significant slow-down in the rate of quasar evolution is witnessed at all three frequencies
with possible evidence for a decrease in the space density of quasars being seen amongst
the radio population at z > 2 and amongst optically faint (MB > -27) QSOs at z > 3.5.

1. Introduction
The quasar luminosity function (LF) is one of the most fundamental statis-
tics relating to the quasar population. It can be employed in a wide variety of
studies, not only to investigate the quasar phenomenon itself (see e.g. Cava-
liere et al. 1985) but also to explore the wider implications of their relation-
ship to the early Universe. These include constraints on galaxy formation
models from the high redshift evolution of the quasar LF (Efstathiou & Rees
1988) and the predicted contribution from quasars to the X-ray background
(Maccacaro et ai. 1984) and UV ionising flux at high redshifts (Bechtold et
al. 1987).
The quasar LF, +(L, z), is simply the co-moving space density of quasars
expressed as a function of luminosity L at any given red shift z. The evo-
lution of the LF is directly obtained from its dependence on redshift. In
co-moving units, an unevolving quasar population will therefore be repre-
sented by a LF that remains constant with redshift. Estimates of the quasar
LF and its evolution are normally obtained from the statistical analysis of
large, unbiased quasar surveys with complete spectroscopic identification.
The rapid increase in the number of such surveys in recent years (particu-
larly in the optical and X-ray regimes) has led to a dramatic improvement
in our knowledge of the quasar LF and its evolution. The purpose of this
review is to describe the current observational status of the quasar LF in the
optical (4400 A), X-ray ('" 2 keY == 6.2 A) and radio (2.7 GHz == 1.1 GA)
regimes by summarising the recent survey work and the results obtained.
Further details on the optical LF and its evolution may be found in recent
reviews by Warren & Hewett (1990), Hartwick & Schade (1990), and Boyle
(1992).

2. The Optical Luminosity Function

In recent years there has been an order of magnitude increase in the number
of optically-selected quasars in samples with complete spectroscopic identi-

433
1. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, 433-454.
© 1993 Kluwer Academic Publishers.
434

fication. This is graphically illustrated by the absolute magnitude - redshift

(ME, z) diagram for such quasars plotted in Figure 1. Prior to 1985 (Figure
la), existing complete surveys contained fewer than 200 quasars, represent-
ing < ~O% of the total quasar population known at that time (Veron-Cetty
& Veron 1985). The surveys also poorly sampled the (ME, z) plane, con-
taining no quasars with B ? 20 or z > 2.2. With samples limited to B < 20
the derived optical luminosity function (OLF) was a featureless power law,
making it difficult to discriminate between the principal competing mod-
els of density and luminosity evolution (see Figure 2). The lack ·of suitable
quasar samples at z > 2 also restricted the study of quasar evolution at high
redshifts, with many results being primarily based on the absence of high
redshift quasars in surveys (Osmer 1982).
Luminosity (erg s - ') Luminos ity (e rg 5- ')
1045 10 4 • 10" 1048 10 45 10 4• 1047 10 48

1985 a 1992 b
12 '" rI ;:"
12
4 4 .' ........
'"e,'
.. p~.
~
r
0
0
;>;"
0 \' • .

'I'
o
o f!. ) •••" • 1
0-

.. ,. ..
. Q!' •• ~.:., • Il>
0:
N 3 _ _ __ __________________ . "j!._ 3 ,,-\;-':'
___________ .- _.., ..._ ~l
0
;>;"
\ " ~
, 11 11 311>
--- ---------.-----------
z _ 2.2 I

, .'
2
,
f
,. • ., 2
10
0 ..

.. . .
0 ' • 10 '<

,'
~'
,'
.
~:' ~."
. ,",
i"
'
9 9
2.-

~
, .,
.. - . '
. . .
,..
I
8
7
8
7
__ _.. ..,:_'.: _._ : _._~ tA'_ : '____ !.""9~ _ 5 5
O L..LLL.LL-'-1....I....1...L..J.'-'--.ILL.L.L.L.L-LJ O LLLl~~'-'--'ULLL.LL-'-1....I....1...L..J.

- 20 - 22 - 24 - 26 - 28 - 30 - 20 -22 -24 - 26 - 28 - 30
Absolute Magnitude (Mo) Abso lu t e Magnitude (MB)

Fig. 1. Absolute magnitude - red shift (MB , z) diagrams for optically-selected

quasars identified in complete samples published prior to 1986 (Figure la) and
published to June 1992 (Figure Ib). Quasars with z < 0.3 have been excluded
from these diagrams for all but the LBQS (Large Bright Quasar Survey) due to
possible incompleteness at these redshifts in the original samples. Also indicated
on the axes are the corresponding bolometric luminosities and look-back times for
H 0 = 50 km s-1 Mpc- 1, qo = 0.5. The dotted line in both Figures indicates the
apparent magnitude limit, B = 20 for z < 2.9 and R = 20 for z > 2.9.
 435

High
A B

$
Vi
.,c ~~~
z=o z= ! z=2
~ z=O z=! z= 2
.,
Q

()
C 0
"'

~\
0.
(f)

~~ z=O z= ! z=2 z=O

Low ~-------------------~--------------------~
z=! z= 2

Faint Bright
Abs olute Magnitud e

Fig. 2. A schematic representation of the quasar LF plotted as a function ofredshift.

With a featureless LF (panel A) it is impossible to distinguish whether the redshift
dependence (i.e., evolution) of the LF is given by density evolution (a vertical shift
in the LF) or luminosity evolution (a horizontal shift in the LF). When optical
surveys for quasars were limited to 13 < 20, the derived LF exhibited this featureless
form. With a feature visible in the LI<', it is, however, relatively straightforward to
discriminate between density evolution (panel B), luminosity evolution (panel C)
or a combination of these two evolutionary forms (panel D).

As of June 1992, '" 2000 quasars have been identified in complete samples
(see Figure 1b) and now comprise r-v 40% of all known quasars (Vcron-Cetty
& Veron 1991). Details of these samples are given in Table I. As can be seen
from Figure 1b, there has been a vast increase in the number of quasars,
particularly amongst those identified at faint magnitudes (B > 20) and at
high redshifts (z > 2.2).
As a result of the improvement in quasar statistics at B > 20, it has
become increasingly clear (Koo 1983; Marshall 1987, Boyle et al. 1987, 1988;
Koo & Kron 1988) that the low-redshift (z < 3) quasar OLF exhibits a
significant break in its power law slope at faint absolute magnitudes. The
redshift dependence of this "break" reveals that the OLF undergoes strong
luminosity evolution between z -,:, 0 and z ~ 2 with a significant cut-off
occurring in this evolution at z ::: 2. A qualitative picture of this result can
he obtained from the binned esthnate of the quasar OLF at z < 2.9 plotted
in Figure 3, derived by Boyle et al. (1991) using the 1jVa statistic (see
Schmidt & Green 1983) from a composite catalogue containing over 700
quasars selected from the surveys listed in Table I.
436

o 0.30 < z< 0.70

• 0.70 < z< 1.25
o 1.2 5< z<2 .00
• 2.00 < z<290

10 - 12 LL-L..l.-Ll.-L.L.JL.L-L.L.JL.L-L--'---J'--'-~.L.....J~....J
- 20 - 22 - 24 -26 -28 - 30
M.

Fig. 3. Binned estimate of the quasar OLF in four redshift intervals equally spaced
in 10g(1 + z) at z < 2_9 for a qo = 0.5 universe (Boyle et al. 1991).

TABLE I
Surveys of Optically-selected Quasars

Reference I Mag. Limit' I z limit" NQSO

Schmidf. & Green ,(1983) B < 16.2 z < 2.2 114
Marshall et al. (1983a) B < 18.5 z < 2.2 18
Marshall et al. (1984) B < 19.8 z < 2.2 35
Mitchell et al. (1984) B < 17.8 z < 2.2 32
Crampton et al. (1987) B < 20.5 z<3 117
Koo et al. (1988) B < 22.5 z < 3 20
Cristiani et al. (1989 ) B < 19.8 z < 2.2 99
Boyle et al. (1990) B < 21.0 z < 2.2 420
Boyle et al. (1991) B < 22.5 z<3 66
Zitelli et al. (1992) B < 22.5 z < 3 52
Warren et al. (1991a) R < 20 z < 4.5 130
Morris et al. (1991) B < 18.5 z < 3 10;31
Irwin et al. (1991}t 1<18 z < 5 20
Schmidt et al. (1991)t
-
R < 20.5 z <5 141

• Magnitude and redshift limits are approximate; see reference for full details.
t Full survey details yet to be published.
 437

Based on the appearance of the OLF in Figure 3, Boyle et ('.1. (1991)

fitted this data-set with a two-power-Iaw function for the OLF, c)(MB' z):

c)*
C)(MB' z)dMB = lO 0.4[MB -MB(z)](a+1) + lO 0.4[MB-MB(z)](/3+1) dMB ,
where a and f3 correspond to the bright-end and faint-end slopes of the
optical LF, respectively. The evolution of the LF is uniquely specified by
the redshift dependence of the break magnitude, MB(Z) :

ME(z) = ME - 2.5klog(1 + z) . z< Zmax

ME(z) = ME - 2.5klog(1 + Zmax} Z > Zmax,

corresponding to a power law evolution in optical huninosity, L!pt(z) ex

(1 + z)k, at low redshifts (z < Zmax) and with an unevolving population
at higher redshifts. Note that models in which the huhlnosity evolution is
modeled as an exponential in look-back time, r: L!pt(z) ex eieT , are strongly
rejected by the data (Marshall 1985; Boyle et al. 1988).
For qo = 0.5 and an assumed 1 optical spectral index aopt "-' 0.5 (Iv ex
v- aopt ) Boyle et al. (1991) obtained a satisfactory fit with values of a =
-3.9 ± 0.1, f3 = -1.5 ± 0.15, MB* = -22.4 ± 0.2, kL = 3.45 ± 0.1, Zmax =
1.9 ± 0.1 and c)* = 6.5 X 10- 7 mag- 1 Mpc- 3 •
This result implies that the OLF undergoes rapid power-law luminosity
evolution between the present epoch and Z '" 2, with the characteristic
optical luminosity of the quasar population increasing by a factor of more
than 30 over this period. As demonstrated by Marshall (1987) and Boyle
et al. (1988), the extrapolated quasar OLF at Z = 0 is consistent with the
Seyfert galaxy LF (see Figure 4), leading to the conclusion that Seyfert
galaxies are the present-day counterparts of quasars. However, the precise
role of Seyferts (as old evolved quasars or merely the present generation of
quasars) and, indeed, the physical significance of luminosity evolution, is
more difficult to establish.
Luminosity evolution can be interpreted as: (a) the evolution of a single
long-lived (10 10 yr) population of quasars; or (b) the evolution of successive
generations of short-lived (10 8 yr) quasars. The long-lived interpretation
runs into significant problems with observations, predicting larger remnant
black hole masses ("-' 109 M 0 ) and smaller Eddington ratios ('" 0.001) in
Seyferts and low-redshift quasars than are currently inferred from emission
line/continuum studies (Wandel & Mushotzky 1986; Padovani 1989). The
short-lived hypothesis also runs into problems explaining the observed con-
stant co-moving space density of "active" quasars at any given epoch. It is
1 For power-law luminosity evolution, any increase or decrease in the adopted value of
(lopt also increases or decreases the derived value of k by the SaJIle amount .
438

z= O QSO LF'

• Seyferl 1 LF'

-20 -22 - 24
M.

Fig. 4. The extrapolated quasar OLF at z = 0 (solid line) compared to observed

Seyfert LF (Cheng et al. 1985). The dotted line denotes the part of the LF that is
not currently sampled at higher redshifts.

possible that the real physical explanation of luminosity evolution lies some-
where between these two alternatives. Cavaliere & Padovani (1988) have
discussed a "recurrent" model in which short-lived quasar activity occurs at
intervals in a significant fraction of the galaxy population, possibly fueled
by recurrent episodes of interactions with companion galaxies (Smith et ai.
1986). The strong luminosity evolution in this case would be explained by
the progressive decrease in the amount of gas available to fuel the quasar at
each successive interaction.
At high redshifts, the derived fit to the OLF by Boyle et ai. (1991) sug-
gests that the strong luminosity evolution "cuts-off" at z = 1.9 and that
the distribution of quasars over the redshift range 1.9 < z < 2.9 is con-
sistent with an unevolving population. This result is qualitatively similar
to that derived by Kron et al. (1991) based on a sample of 150 quasars
with B < 22.5, although the detailed analysis of this sample has yet to
be published. Using a grism-selected sample of faint quasars, Schmidt et
al. (1991) also find that the co-moving space density of quasars with MB
::: -26 remains approximately constant over the range 2 < z < 3. Warren
et ai. (1991) find some evidence for continuing luminosity evolution beyond
z = 2, with the integrated space density of quasars with MB < -26 peaking
at z ~ 3. Nevertheless, this evolution is much slower than that derived at
 439

50

The Epoch of QSOs

40

eg. 30
o-l
"- L'(z) = L'(O) (1+Z)3<5 z < 1.9
~ = constant z> 1.9
g.
o-l 20

qo=0 .5
(')
II
10 N

o 4xl0· 6xl0· Bxl0· 10'0 1.2 x l0'o

Age of Universe (years)

Fig. 5. The evolution of the OLF, characterised by the change in the normalised
break luminosity L~pt(z)/ L~pt(O), expressed as a function of time (qo = 0.5, Ho = 50
km S-1 Mpc- 1 ). The dashed lines represent the uncertainty in the evolution of
quasars at z > 3.

lower redshifts. The Warren et ai. (1991) model predicts a 45% increase in
the value of L* over the range 2 < z < 3, corresponding to a power-law
evolution, L~pt(z) <X (1 + z)1.3. A preliminary analysis of the 1000 quasars
in the Large Bright Quasar Survey (see Morris et al. 1991) by Hewett et
ai. (1993) also suggests that the quasar population may continue to exhibit
some mild evolution beyond z = 2, but again this is much reduced from
the strong evolution derived at lower redshifts in this sample. Thus, there is
considerable agreement between surveys that the strong optical evolution of
quasars witnessed at z < 2 dramatically slows down at z ;:: 2, with little or
no evidence for any decline in the co-moving space density of quasars over
the redshift range 2 < z < 3. This evolution is represented in Figure 5 by the
dependence of the normalised break luminosity on cosmic time. This figure
emphasises the relatively short period of time during which the luminosity
of the quasar population was at its peak (see also Schmidt et ai. 1991).
At z > 3, the picture that emerges is somewhat contradictory. At bright
magnitudes, a number of surveys for high redshift quasars (Hazard et ai.
1986, Mitchell et ai. 1990; Irwin et ai. 1991) have established that the space
440

density of quasars with MB < -27 remains unchanged between z = 2 and

z = 4. In contrast, surveys for fainter quasars at high redshift (Warren et
al. 1991b; Schmidt et al.1991) indicate that the space density of quasars
with -27 < MB < -26 is a factor of 3-10 lower at z = 4 than at z = 2.
Schmidt et al. (1991) model this decline as an exponential decrease in the
co-moving space density of quasars with MB < - 26 beyond z = 3 by factor
of 3.2 per unit redshift. Although these two opposing results can be recon-
ciled straightforwardly by a model which invokes a luminosity dependence
in the space density evolution of quasars at high redshifts (i.e., faint quasars
undergo a rapid decline in their numbers at z > 3, while bright quasars ex-
hibit little evolution), the steep OLFs derived from each individual survey
are inconsistent with such a view (see discussion following Marshall 1991).
However, samples of quasars at z > 3 are still small, and the results
derived from them are particularly sensitive to large corrections for incom-
pleteness (see e.g. Warren et al. [1991bJ).1t is possible that a clearer picture
of quasar evolution at high redshift will only emerge with larger surveys for
quasars at z > 3, ideally selected at magnitudes R ;:: 22, thereby probing
below the break in the OLF at these redshifts.
The high-redshift evolution of quasars is particularly important to estab-
lish, as it can place fundamental constraints on galaxy Jormation models. In
particular, the lack of a significant decline in the co-moving number density
of quasars at high redshifts would pose significant problems for models of
galaxy formation, such as Cold Dark Matter (Davis et al. 1985), that pre-
dict a late epoch for galaxy formation. Efstathiou & Rees (1988) have shown
that a constant co-moving density of quasars over the range 2 < z < 4 is
consistent with Cold Dark Matter models, although the discovery of sig-
nificant numbers of quasars at higher redshifts would be difficult to recon-
cile in such models. The high-redshift OLF can also be used to establish
whether the integrated UV flux from quasars can contribute significantly to
the re-ionisation of the intergalactic medium (IGM) at high redshift (see e.g.
Bechtold et al. 1987). Boyle (1992) has demonstrated that, for a constant
co-moving space density of quasars over the range 2 < z < 4, the integrated
background intensity at 912 A (J912) is sufficient for re-ionisation at z . . . , 4;
J 912 = 7 X 10- 22 ergs S-1 cm- 2 sr- 1 Hz- 1 • Although this result was cal-
culated neglecting absorption due to galaxies, more recent work (Meiksin
& Madau 1992) suggests that the UV flux from quasars is still sufficient to
re-ionise the IGM when realistic amounts of galactic absorption are included.

3. The X-ray Luminosity Function

In common with the optical regime, the dramatic increase in the number
of X-ray selected quasars identified in recent years has also led to a much
improved understanding of the quasar X-ray (......, 2 keY) luminosity function
 441

3 3
I I
1964 1992
a b
..
...... . ....
2 r- - 2

.
0:.
.. 00· 'a 00

. . ...'., ;si 0°

Ir- .. -
,.. fi
••
(0
t:bo,,°Y
Oq) 0
0000
8

.
••• 00
00 CoO 00

0,0
• ROSAT
o o
001. !.ISS EMSS

o ~~ c» o
00 • e°lt.o 10 j
0
0
42 44 46 _. 46 42 44 46 46
lo g L(O.3-3.SkeV) (er gs s ) log L( O.3 - 3.SkeV) (ergs s ')

Fig. 6. The luminosity-redshift diagram (Lx,z) for X-ray selected quasars (qO = 0)
in complete samples: (a) published prior to 1991; and (b) published to date.

(XLF) and its evolution with redshift. This increase is illustrated by the
quasar X-ray (0.3-3.5 keY) luminosity - redshift (Lx, z) diagram plotted
in Figure 6. Prior to 1991, the only major published survey with com-
plete optical spectroscopic identification was the Medium Sensitivity Survey
(MSS, Gioia et al. 1984) which contained 56 quasars at a flux limit S(0.3-
3.5 keY) ~ 2 X 10- 13 ergs s-l cm -2 (Figure 6a). Since 1991, this data-set
has been greatly expanded with the publication of two X-ray surveys for
which optical spectroscopic identification has recently been completed; the
Extended Medium Sensitivity Survey (EMSS) compiled from X-ray obser-
vations made in the 0.3-3.5 keY passband with the Einstein satellite (Stocke
et al. 1991) and the ROSAT quasar survey (Shanks et ai. 1991, Griffiths et
ai. 1993, in preparation), also selected at soft (0.5-2 keY) X-ray energies.
The EMSS survey contains 427 quasars with fluxes S(0.3-3.5 keV)~ 10- 13
ergss- 1 cm- 2 , while the ROSAT survey contains 42 quasars identified to a
much deeper flux limit, S(0.5-2.0 keY) ;:: 6 X 10- 16 ergs s-l cm -2 • As can
be seen from Figure 6b, the surveys provide complementary coverage of the
X-ray luminosity-redshift (Lx, z) plane, with the EMSS providing detailed
information at low redshifts and the ROSAT survey extending the range in
lwninosity sampled at z 0.5 to '" 1046 ergs s-l.
Although a break in the XLF at low luminosities had first been identified
from the Medium Sensitivity Survey by Maccacaro et ai. (1984), a signif-
icantly more detailed description of the XLF and its evolution based on
the EMSS quasars was provided by Maccacaro et al. (1991). They demon-
442

10-'

10-'

_~ 10·-
,
:i'
~ 10-7

7
li 10--
3
J 10-' o 0 .0 <z <0.4
~ 0.4 <z < 1.0
~ 10- . 0 o 1.0<z < 1.8
• 1.8 <z<3.0
10- 11

10- 12
43 44 45 1 46 47
log Lx (ergs s - )

Fig. 7. A binned representation of the 0.3-3.5 ke V XLF derived from the EMSS and
ROSAT samples for a qo = 0.5 universe based on the analysis of Boyle et al. (1992).
Note that the XLF in this figure has been recomputed per logarithmic luminosity
interval and thus differs from the form presented by Boyle et al. (1992), in which
the XLF was expressed per linear 10 44 ergs s-1 bin. In this diagram, luminosity
evolution is therefore represented by a uniform horizontal shift of the XLF.

strated that the low-redshift (z < 0.2) XLF exhibited a pronounced break
to a flatter slope at low luminosities and that the evolution of the XLF was
consistent with luminosity evolution. However, the relative shallowness of
the EMSS prevented any detailed analysis of the XLF at anything other
than the very brightest luminosities (> 10 45 ergs s-l) even for relatively
moderate redshifts (z > 0.5). The ROSAT survey significantly extends the
coverage of the XLF to much lower luminosities at these redshifts. By com-
bining the ROSAT sample with the EMSS, Boyle et al. (1992) were able to
demonstrate the existence of a break in the XLF at higher redshifts (z > 0.2)
and to confirm that luminosity evolution was a satisfactory fit tothe XLF at
z;:; 2. Both these results are clearly illustrated by the binned representation
of the XLF plotted in Figure 7 obtained from the analysis of the combined
EMSS and ROSAT data set by Boyle et al. (1992).
Based on the appearance of the binned XLF in Figure 7, Boyle et al.
(1992) followed Maccacaro et al. (1991) in modeling the z = 0 XLF as a
two-power-Iaw function of the form:
 443

c)(Lx) = K 1 L-')'1
XH Lx < Lx(z = 0)
Lx > Lx(z = 0) ,
where Lxu is the 0.3-3.5 keY X-ray luminosity expressed in units of 1044
ergs s-1 The evolution of the XLF was parameterised as a power law in
(1 + z):
Lx(z) = Lx(O)(l + z)k ,
with exponential evolution models (Lx 0( ek-r) being strongly rejected by
the data (see also Della Ceca et al. 1992).
Assuming qo = 0 and an X-ray spectral index aa: = 1, Boyle et al. (1992)
derived an acceptable fit to the combined EMSS jROSAT data set with
the following parameter values for the XLF: 1'1 = 1.7 ± 0.2, 1'2 = 3.4 ± 0.1,
logLx = 43.84±0.I,K1 = K2L~::-,),2) = 0.57xl0- 6 Mpc- 3 (10 44 ergs s-1 )-1
and k = 2.75 ± 0.1. Note that this value of k implies a significantly faster
rate for the evolution of XLF than that derived from the EMSS sample by
Maccacaro et al. (1991),
Lx 0( (1 + z)2.45±O.1 •
However, as demonstrated by Boyle et al. (1992), the slow evolution rate
found by Maccacaro et al. (1991) fails to account for the large number of
z > 1 quasars identified in the ROSAT sample, predicting a median redshift
for this sample of (z) = 0.5, significantly different from the observed median
redshift (z) = 1.5. Given the dependence of k on the adopted spectral in-
dex (see above), Boyle et al. (1992) suggested that a softening of the X-ray
spectral index at ::: 1.5 keVin the quasar rest frame could account for the
discrepancy in the values of k obtained. This occurs because the predomi-
nantly low redshift ((z) :::::: 0.2) EMSS quasars are observed at softer energies
in the rest frame than the higher red shift (z ~ 0.5) ROSAT quasars. Indeed,
this is consistent with the observation by Marshall (1991) that the evolu-
tion rate derived by Maccacaro et al. (1991) from the EMSS is significantly
slower than that inferred from a quasar sample selected at much harder en-
ergies (2-10 keY) with the HEAO Al satellite. However, the difference in
the value of k obtained for the EMSS and ROSAT samples may also reflect
a real change in the rate of evolution at z :::::: 0.5. Unfortunately, larger sam-
ples of X-ray quasars will be required before such a model can be tested
rigourously.
Some evidence for a departure from the strong power-law evolution at
higher redshifts has already been found by Boyle et al. (1992) . They demon-
strated that the power-law evolution model in a qo = 0.5 universe (rejected
at the 95% confidence level) gave a significantly better fit if a cutoff in the
evolution similar to that required in the optical was introduced at z = 2:
444

Lx(z)= Lx(O)(l + Z)k z<2

Lx(z) = Lx(2) z > 2,

where k = 2.73 ± 0.1. This cut-off is evident from the binned XLF in Figure
7, with the XLF at 1.8 < z < 3 exhibiting little, if any, evolution from its
position at 1.0 < z < 1.8. This result is also consistent with Della Ceca &
Maccacaro (1991) who suggested that a cut-off in the power law evolution
was required at z ;:: 2 (albeit for qo = 0) in order to achieve consistency
with the deep ROSAT counts of Hasinger et al. (1991). Despite this mea-
sure of agreement, there is still significant uncertainty over the high-redshift
evolution of the XLF, and further samples of X -ray selected quasars at high
redshift will be required before the precise nature of this cutoff is established.
The uncertainty in the high-redshift evolution of the XLF, coupled with
the error on the faint-end slope of the XLF, dominate any attempt to deter-
mine the quasar contribution to the soft (2 keY) X-ray background (XRB)
from the observed XLF. Boyle et al. (1992) demonstrate that the uncer-
tainty in the XLF allows the actual quasar contribution to lie anywhere in
the range 30-90%, with a best estimate of "" 50%. This is consistent with
other estimates made from the EMSS quasars (Maccacaro et al. 1991), from
deep source count constraints (Shanks et al. 1991), and from the observed
fluctuations of the XRB (Carrera & Barcons 1992; Georgantopoulos et al.
1992).

4. The Radio Luminosity Function

The most recent and comprehensive study of the quasar radio luminosity
function (RLF) has been made by Dunlop & Peacock (1990) from a sample
of 171 flat-spectrum (OR < 0.5) radio sources selected at 2.7 GHz with
S > 0.1 Jy. Over 90% of the objects in the sample have been confirmed as
quasars from optical spectroscopy. The binned version of the RLF obtained
from this sample for qo = 0.5 is reproduced in Figure 8.
Dunlop & Peacock (1990) were able to model the RLF successfully as a
two-power-law function:

dN _ +(L z) - +*
d(logL)dV - , - [Ltfz)]a + [Ltrz)]fj ,

where LR denotes the monochromatic 2.7 GHz luminosity in WHz- 1 and

where luminosity evolution is expressed as a quadratic function of redshift:
 445

- • - .=0.5
o - 'Z - l .Q
,
.
x - z=2.50
:l.
~
",
.£ I

,
.,<l

u
Q.

~
-0
- 0
011 -
3 !

24 25 26 27 28 29
Log lO(P 2.?/ WHz -I sr - I)

Fig. 8. The 2.7 GHz RLF for flat-spectrum quasars (qO = 0.5) derived by Dunlop
& Peacock (1990).

Dunlop & Peacock (1990) obtained best-fit values of 0 = 0.83 and f3 = 1.96
for the slopes of the RLF with an overall normalisation factor given by
c)* = (7 X 1O-9)(log L )-1 Mpc 3 . The corresponding best-fit coefficients
for the quadratic luminosity evolution were ao = 25.26, a1 = 1.18 and
a2 = -0.28. At low redshifts (z ::: 2), this quadratic expression approximates
to a strong power'law luminosity evolution, LR(z) '" LR(O)(1 + z)3, similar
to that seen in the optical and X-ray regimes. At z ~ 2 the evolution also
rapidly "switches off", but unlike the optical and X-ray evolution, goes into
reverse for z > 2, predicting that by z = 4 the RLF has evolved back to its
position at the present epoch. This evolution corresponds to a decline in the
space density of quasars with LR = 1026 W Hz- 1 by a factor of 2 between
z = 2 and z = 3. Peacock & Dunlop (1990) also obtained a similar cutoff in
the evolution of steep-spectrum sources (OR;::: 0.5), a sample predominantly
composed of radio galaxies. However, in this case, the precise form of the
evolution is considerably more uncertain due to the larger number of objects
without spectroscopic redshifts in the sample.
Additional free-form modeling of the RLF by Dunlop & Peacock (1990)
suggests that some form of luminosity-dependent evolution may occur at
z > 2, As in the optical regime, high-luminosity radio quasars (LR =
1027 W Hz- 1 ) appear to exhibit a less marked decline in their co-moving
space density between z = 2 and z = 3 than low-luminosity quasars (LR =
10 26 W Hz- 1 ). Indeed, to within the errors of the fit, the space density of
LR = 10 27 W Hz- 1 quasars between z = 2 and z = 3 is consistent with an
unevolving population.
At z > 3, the model fits to the RLF derived by Dunlop & Peacock (1990)
are poorly constrained (the flat spectrum sample contains only one quasar
446

with a spectroscopic redshift at z > 3). Nevertheless, extrapolation of the

quadratic luminosity evolution derived by Dunlop & Peacock (1990) at lower
redshlfts yields good agreement with the number of radio-selected quasars
observed at z > 3 by McMahon (1991) based on an identification program of
5 GHz sources currently in progress. A:; yet, however, such surveys only con-
tain a few quasars (;:; 10) and larger samples will be required before a more
accurate picture of the RLF and its evolution at z > 3 can be constructed.

5. Discussion
5.1. THE LUMINOSITY FUNCTION

From the preceding sections it is clear that the quasar LF and its evolution
exhibits a number of striking similarities throughout the optical, X-ray, and
radio regimes. For the quasar LF we can adopt a single parameterisation for
its two power law form seen in each regime:

dN +*
d(logL)dV L ]0:-1 + [ L ]{3-1'
= CJ(L,z) = [LO(Z) LO(z)

where the co-moving space density of quasars (dN / dV) is quoted per unit
logarithmic luminosity interval in order to facilitate comparison between the
values obtained for the normalisation (+*) of the LF in the different regimes.
Although strictly only a measure of the space density of L * quasars, the
relative values of CJ* can be taken as an approzimate guide to relative total
space density of quasars in each regime. Note that the slopes of the LF
correspond to:
CJ(L)dLocL-O: L~L*

CJ( L ) dL oc L -{3 L ~.L * ,

when the LF is expressed per unit luminosity.
Table II summarises the best-fit parameters obtained for the monochromatic
OLF, XLF, and RLF at 4400 A, 2 keY and 2.7 GHz respectively. The value
of Lx(O) in the broad-band 0.3-3.5 keY XLF has been corrected to a mono-
chromatic 2 keY luminosity assuming ax = 1. Both the OLF and XLF
exhibit similar normalisations, in marked contrast to that obtained for the
RLF. Indeed, the values of the faint-end slopes and normalisations of the
XLF and OLF are consistent with an optical to X-ray luminosity relation
of the form Lx oc L~;i8±O.08 (Boyle et al. 1992), in good agreement with the
relation Lx oc L~~ derived by Avni & Tananbaum (1986). Furthermore, the
agreement in normalisation between the XLF and OLF lends additional sup-
port to the argument that there are few, if any, X-ray-quiet quasars (Avni
& Tananbaum 1986).
 447

TABLE II
Summary of Luminosity Function Parameters (qO = 0.5)

LF Parameters Optical (4400 A) X-ray (2 keY) Radio (2 .7 GHz)

(Boyle et a1. 91) (Boyle et ale 92) (Dunlop & Peacock 90)

a 3.6 ± 0.1 3.4 ± 0.1 3.0 ± 0 .1

{3 1.5 ± 0.2 1.7 ± 0.2 1.8 ± 0.2
L· (ergs s-lHz-1l 4.2 x 10 29 5.2 X 10 25 2.2 X 1033
~. (log L-1Mpc- ) 1.6 x 10- 6 2.8 X 10- 6 7 X 10- 9

In contrast, the difference in the normalisations for the OLF and the RLF
demonstrates that at the characteristic break luminosity, L * , almost 99% of
the optically-selected quasar population is radio-quiet (Le., exhibit radio
luminosities ~ L R)' This accounts for the extremely low fraction ('" 1%) of
radio-loud quasars observed in optically-selected samples at faint (B '" 20)
magnitudes (Peacock et ale 1986) which are predoininantly composed of L~pt
quasars. By the same token, the significantly flatter bright-end slope for the
RLF also naturally explains the increasing prominence ofradio-Ioud quasars
in samples at higher optical luminosities (see e.g. Kellerman et al. 1989).
The values of L* can be viewed as the luminosity of a "characteristic"
quasar in each regime. With values for the faint end slopes of f3 < 2 in
each regime, L * quasars dominate the quasar "luminosity density" (and
thus background intensity calculations) at any given redshift. The energies
associated with these break luminosities at z = 0 are:

Optical (4400A) : vL*(O) = 2.9 x 1044 ergs S-1

X - ray (2keV) : vL*(O) = 2.2 x 1043 ergs s-1
Radio (2.7GHz): vL*(O) = 5.9 x 1042 ergs S-1
These figures underline the dominance (at least in terms of energy) of the
UV /optical regime in quasars. Moreover, these "characteristic" energies for
L* quasars are broadly consistent with the relative X-ray/radi%ptical-UV
energies observed in individual (radio-loud) quasars (Sanders et ale 1989).

5.2. EVOLUTION

Figure 9 summarises the observed evolution of the OLF, XLF, and RLF in
a qo = 0.5 universe. As can been seen from this Figure, the evolution of all
LFs at z < 2 can be represented by power law luminosity evolution with
the X-ray (k = 2.75 ± 0.1), radio (k ~ 3) and optical (k = 3.45 ± 0.1) all
exhibiting very similar rates of evolution. While the similarity between the
448

evolution of the OLF and XLF is naturally explained by the close correlation
between the optical and X-ray luminosity for quasars Lx <X L~~8 (see §5.1),
it is perhaps surprising to see the RLF exhibit a similar rate of evolution. As
discussed above, radio-loud quasars only comprise rv 1% of the optical/X-
ray quasar population at L *, and thus there is no a priori reason to expect
them to exhibit a similar rate of evolution. This is particularly true given
the evidence that the strong evolution with redshift in the environment of
radio-loud quasars (Yee & Green 1987) is radically different to that seen
seen in optically-selected quasars (Boyle et al. 1988; Ellingson et al. 1991).
It should be noted that the differences between- the rates of evolution
obtained for the optical, radio and X-ray regimes are only significant at
the level of the statistical errors which have been quoted here. It is likely
that, with the ever-increasing numbers of quasars now used in the derivation
of the quasar LF, systematic rather than statistical errors will begin to
dominate the derivation of the LF parameters. For example, a decrease in the
mean adopted optical spectral index of 0.3 coupled with a similar increase
in the adopted X-ray spectral index would render the evolution derived
in the optical and X-ray regimes statistically indistinguishable from each
other. Although such a large change in the mean optical spectral index for
quasars is perhaps at the limits of current observational uncertainties (see
e.g. Francis et al. 1991), other systematic effects such as incompleteness
of existing optical surveys (Goldschmidt et al. 1992) or a spread in the
optical spectral index (Giallongo & Neushafer 1992) could reduce the rate
of evolution derived in the optical to the point at which all three regimes
exhibit nearly identical evolution. As such, the observation by Saunders
et al. (1991) that IRAS galaxies also exhibit an evolution consistent with
LIR <X (1 + z)3±1 (albeit only derived for z < 0.1), makes this "one plus zee
to the three" power-law evolution seem remarkably (but as yet inexplicably)
ubiquitous.
At z ::: 2, all three regimes exhibit some evidence for a cutoff in the strong
luminosity evolution witnessed at lower redshifts. In the optical and X-
ray, the distribution of quasars is consistent with an unevolving population,
L*(z) = constant, while the radio population undergoes negative luminosity
evolution at these redshifts. The corresponding evolution in terms of space
density is represented in the panel at the foot of Figure 9. The observation
that the X-ray and radio regimes exhibit a cutoff at approximately the same
redshift as that found in the optical provides additional evidence in favour
of the optical cut-off at z '" 2 being a real effect, rather than one artificially
introduced by the different techniques used to select optical quasars at z ::; 2
(ultraviolet excess) and at z ;:: 2 (multi-colour/prism).
The epoch prior to z : : : : 2 therefore corresponds to the period when quasar
activity in the optical, X-ray, and radio was at its maximum. If quasars un-
dergo a significant decline in their numbers beyond z : : : : 3, then this period is
 449

Optical/Radio/Xray Evolution

-------jI

--
?

-- ····1
.~
.......... 10
N 1
.~
1_ _ _ -

Optical (4400A)
Boyle el a1. (1991)
X-ray (2keV)
Boyle el a1. (1992)
Radio (2.7GHz)
Dunlop and Peacock (1991)
1~-L~~-L~~-L~~~~~~~
o 123 4
z

~
Optical L(z) <X (1 + Z)3.45±O.1
X-ray L(z) <X (1 + z)2.75±O.t
Radio L(z) <X (1 + Z)3
2 < z < 3.5
Optical ~(L',z) ~ constant
X-ray ~(L',z) ~ constant
Radio ~(L',z = 2) - O.5~(L·,z = 3)
LL..l&
Optical (MB < -27) ~(L,z) ~ constant
Optical (MB > -27) ~(L , Z = 4) - O.2~(L , z = 3)
X-ray ?????
Radio ?????

Fig. 9. The evolution of L·, normalised to the present-epoch break luminosity,

LO(O), plotted as function ofredshift for the optical, X-ray, and radio regimes_ The
statistical uncertainty in the derived rates of evolution in the optical, X-ray and
high red shift radio regimes are indicated by the line pairs plotted. A summary of
the functional forms used to describe the evolution of the quasar LF is given in the
panel at the foot of this figure.
450

quite short (rv 1 Gyr, qo = 0.5, Ho = 50 km S-l Mpc- 1 ) and sharply peaked
(Schmidt et ai. 1991). Unfortunately, there is still considerable uncertainty
over the nature of quasar evolution at z > 3. At these redshifts, little or no
data exist for the radio and X-ray regimes, and the results in the optical
regime lead to a somewhat ad hoc model (as discussed in §2) which invokes
markedly different rates of evolution for high- and low-luminosity quasars.
Nevertheless, whatever the nature of quasar evolution at z > 3, it is now
clear that quasars have a "brief but brilliant" career during the first 2 Gyr
of the Universe, followed by an equally dramatic burn-out at subsequent
epochs. Further surveys of quasars at high redshift should, in future, help
complete the picture of quasar evolution and, in turn, also provide valuable
clues as to physical conditions in the early Universe.

6. Conclusions

The current status of quasar LF and its evolution in the optical, X-ray and
radio regimes can be summarised as follows:

• The optical, X-ray, and radio LFs can all be represented by a two-
power-law function with a break from a steep to a flat slope at a character-
istic luminosity L *.

• At z < 2 the evolution of the LF is characterised by a strong evolution

in luminosity; L* ex: (1 + z)k. The exponent of evolution is remarkably sirni-
lar in each regime; k = 2.75 ± 0.1 (X-ray), k rv 3 (radio) and k = 3.4 ± 0.1
(optical). Further study of systematic errors may reduce the observed dis-
crepancy between these values.

• The strong luminosity evolution switches off at z ~ 2 in all regimes. In

the optical and X-ray regimes the space density of quasars (2 < z < 3) is
consistent with an unevolving population, while in the radio regime there is
some evidence for a decline in density by a factor of 2 over the same red shift
range.

• At z > 3, little information exists on the radio or X-ray LF. In the opti-
cal, the current results are somewhat contradictory, but may be reconciled by
luminosity-dependent density evolution models in which faint (MB > -27)
quasars exhibit a rapid decline (by a factor of 3-10) in their space density
between z = 3 and z = 4, while bright quasars (MB < -27) remain at a
constant co-moving space density.

• The optical and X-ray LFs can be used to provide limits on the UV
ionising flux and X-ray background due to quasars. Current estimates reveal
 451

that quasars can provide most of the UV flux required to re-ionise the IGM
at high redshifts and form a significant part (> 30%) of the soft (2 ke V)
X-ray background.

Acknowledgements
Much of the work on the quasar optical and X-ray LF reported in this paper
was carried out in collaboration with my colleagues Ioannis Georgantopou-
los, Richard Griffiths, Laurence Jones, Bruno Marano, Tom Shanks, Gordon
Stewart, Gianni Zamorani and Valentina Zitelli. I would like to thank all of
them for many stimulating and productive discussions. During this work I
was supported by a Royal Society University Research Fellowship.

References
Avni, Y. & Tananbaum, H.: 1986, ApJ 305,83
Bechtold, J., Weymann, R. J., Lin, Z., & Malkan, M. A.: 1987, ApJ 315, 180
Boyle,B.J .: 1992, in Texas-ESO/CERN symposium on Relativistic Astrophysics, Cosmol-
ogy and Particle Physics, ed(s)., J. Barrow, L. Me6tel, 8 P. Thoma6, Ann N.Y. Acad.
of Sci. No 647, 14
Boyle, B. J., Fong, R., Shanks, T., & Peterson, B. A.: 1987, MNRAS 227, 717
Boyle, B. J., Shanks, T ., & Peterson, B. A.: 1988, MNRAS 238, 957
Boyle, B. J., Shanks, T., & Yee, H . K. C .: 1988, in Large Scale Structures in the Universe:
IAU Symposium No 130, ed(s)., J. Audouze, M .. C. Pelletan, 8 A . Szalay, Kluwer:
Dordrecht, 577
Boyle, B. J., Fong, R., Shanks, T ., & Peterson, B. A: 1990, MNRAS 243, 1
Boyle, B. J., Jones, L. R., & Shanks, T .: 1991, MNRAS 251 , 482
Boyle, B. J., Jones, L. R., Shanks, T., Marano, B., Ziielli, V., & Zamorani, G.: 1991, in The
Space Distribution of Quasars, ASP Conference Series No. 21, ed(s)., D. Crampton,
Provo: Brigham Young, 191
Boyle, B. J., Griffiths, R. E., Shanks, T., Stewart, G. C., & Georgantopoulos, I.: 1992,
MNRAS in press
Carrera, F. & Barcons, X. : 1992, MNRAS 257, 507
Cavaliere, A., Giallongo, E., & Vagnetti, F.: 1985, ApJ 296, 402
Cavaliere, A. & Padovani, P.: 1988, ApJ 333, L33
Cheng, F. Z., Danese, J., De Zotti, G., & Franceschini, A.: 1985, MNRAS 212, 857
Crampton, D., Cowley, A. P., & Hartwick, F. D. A. : 1987, ApJ 314, 129
Cristiani, S., Barbieri, C., Iovino, A., La Franca, F., & Nota, A.: 1989, A8AS 89,77
Davis, M., Efstathiou , G., Frenk, C. S., & White, S. D. M.: 1985, ApJ 292, 371
Della Ceca, R., Maccacaro, T., Gioia, I. M., Wolter, A., & Stocke, J . T.: 1992, ApJ 389,
491
Della Ceca, R. & Maccacaro, T.: 1991 , in The Space Distribution of Quasars, ASP Con-
ference Series No. 21, ed(s)., D. Crampton, Provo: Brigham Young, 150
Dunlop, J. S. & Peacock, J. A.: 1990, MNRAS 247, 19
Efstathiou, G . & Rees, M. J .: 1988, MNRAS 230, 5P
Ellingson, E., Yee, H . K. C. & Green, R. F.: 1991, ApJ 371, 41
Francis, P. J., Hewett, P. C., Foltz, C. B., Chaffee, F. H., Weymann, R. J., & Morris, S .
L.: 1991, ApJ 373, 465
Georgantopoulos, I., Stewart, G. C., Shanks, T., Griffiths, R . E., & Boyle, B. J.: 1992,
MNRAS in press .
Giallongo, E . & Vagnetti, F.: 1992, ApJ 396, 411
452

Gioia, I. M., Maccacaro, T., Schild, R. E., Stocke, J. T., Liebert, J., Danziger, I., Kunth,
D., & Lub, J.: 1984, ApJ 283, 495
Goldschmidt, C. R., Miller, L., La Franca, F., & Cristiani, S.: 1992, MNRAS in press
Hartwick, F. D. A. & Schade, D.: 1990, ARAA 28, 437
Hasinger, G., Schmidt, M., Triimper, J.: 1991, AfJA 246, L2
Hazard, C., McMahon, R. G., & Sargent, W.: 1986, Nature 322, 38
Irwin, M. J., McMahon, R. G., & Hazard, C.: 1991, in The Space Distribution of Quasars,
ASP Conference Series No. 21, ed(s)., D. Crampton, Provo: Brigham Young, 117
Hewett, P. C., Foltz, C. B., & Chaffee, F. H.: 1993, ApJL in press
Kellerman, K. I., Sramek, R., Schmidt, M., Shaffer, D. B., & Green, R.: 1989, AJ 98,
1195
Koo, D. C.: 1983, Qua6ar6 and Gravitational Len6e6: 24th Liege A6trophY6ical Colloqium,
University of Liege, 240
Koo, D. C., Kron, R. G., & Cudworth, K. M.: 1986, PASP 98, 285
Koo, D . C. & Kron, R. G.: 1988, ApJ 325, 92
Kron, R. G., Bershady, M: A., Munn, J. A., Smetanka, J. J., Majewski, S. & Koo, D. C.:
1991, in The Space Distribution of Quasars, ASP Conference Series No. 21, ed(s)., D.
Crampton, Provo: Brigham Young, 32
Maccacaro, T., Gioia, I. M., & Stocke, J. T.: 1984, ApJ 283, 486
Maccacaro,··T., Della Ceca, R., Gioia, I., Morris, S. L., Stocke, J. T., & Wolter, A.: 1991,
ApJ 374,117
Marshall, H. L., Tananbaum, H., Zamorani, G., Huchra, J. P., Braccesi, A., & Zitelli, V.:
1983a, ApJ 269,42
Marshall, H. L., Avni, Y., Tananbaum, H., & and Zamorani, G.: 1983b, ApJ 269, 35
Marshall, H. L., Avni, Y., Braccesi, A., Huchra, J. P., Tananbaum, H., Zamorani, G., &
Zitelli, V.: 1984, ApJ 283, 50
Marshall, H. L.: 1985, ApJ 299, 109
Marshall, H. L.. : 1987, AJ 94, 628
Marshall, H. L.: 1991, in The Space Distribution of Quasars, ASP Conference Series No.
21, ed(s)., D. Crampton, Provo: Brigham Young, 184
McMahon, R. G.: 1991, in The Space Distribution of Quasars, ASP Conference Series No.
21, ed(s)., D. Crampton, Provo: Brigham Young, 129
Meiksin,A. & Madau, P.: 1992, in Evolution of Galaxies and Their Environment: The
Contributed Papers, ed(s)., D. J. Hollenbach, J. M. Shull, fJ H. A. Thron6on, NASA-
CP,99
Mitchell, P. S., Miller, L., & Boyle, B. J.: 1990, MNRAS 244, 1
Mitchell, K. J., Warnock, A., & Usher, P. D.: 1984, ApJ 287, L3
Morris, S. L., Weymann, R. J., Anderson, S. F., Hewett, P. C., Foltz, C. B., Chaffee, F.
H., Francis, P. J., & MacAlpine, G. M.: 1991, AJ 102, 1627
Osmer, P. S.: 1982, ApJ 263, 28
Padovani, P.: 1989, AfJA 209, 27
Peacock, J. A.: 1985, MNRAS 217, 601
Peacock, J. A., Miller,L., & Longair, M. S.: 1986, MNRAS 218, 265
Sanders, D. B., Phinney, E. S., Neugebauer, G., Soifer, B. T., Matthews, K.: 1989, ApJ
347,29
Saunders, W., Rowan-Robinson, M., Lawrence, A., Efstathiou, G., Kaiser, N., Ellis, R. S.,
& I<'renk, C. S.: 1990, MNRAS 242, 318
Schmidt, M. & Green, R. F.: 1983, ApJ 269, 352
Schmidt, M., Schneider, D., & Gunn, J.: 1991, in The Space Distribution of Quasars, ASP
Conference Series No. 21, ed(s)., D. Crampton, Provo: Brigham Young, 109
Shanks, T., Georgantopoulos, I., Stewart, G. C., Pounds, K . A., Boyle, B. J. & Griffiths,
R. E.: 1991, Nature 353, 315
Smith, E. P., Heckman, T. M., Bothun, G. D., Romanshin, W ., & Balick, B.: 1986, ApJ
306,64
Stocke, J. T., Morris, S. L., Gioia, I. M., Maccacaro, T., Schild, R., Wolter, A., Fleming,
 453

T. A., & Henry, J. P.: 1991 , ApJS 76, 813

Veron-Cetty, M.-P. & Veron, P. : 1985, ESO Scientific Publication No 4
Veron-Cetty, M.-P. & Veron,P.: 1991, ESO Scientific Publication No 10
Wandel, A . & Mushotzky, R.: 1986, ApJ 306, L61
Warren, S. J . & Hewett, P. C .: 1990, Rev. Mod. Phys. 53, 1093
Warren, S. J. & Hewett, P . C., & Osmer, P . S.: 1991a, ApJS 76,23
Warren, S. J., Hewett, P. C., & Osmer,P.S.: 1991b, in The Space Distribution of Quasars,
ASP Conference Series No. 21, ed(s)., D. Crampton, Provo: Brigham Young, 139
Yee, H. K. C. & Green, R. F.: 1987, ApJ 319,28
Zitelli, V., Mignoli, M., Marano, B., Zamorani, G., & Boyle, B . J .: 1991, MNRAS 256,
349
 GALACTIC SUPERWINDS

TIMOTHY M. HECKMAN and MATTHEW D. LEHNERT

Rowland Department of Physics ff Astronomy, The fohns Hopkins University;
and Space Telescope Science Institute
and
LEE ARMUS
Department of Physics, Caltech

Abstract. We review the galactic 'superwind' phenomenon: global outflows powered by

the collective effect of the supernova explosions and stellar winds associated with powerful,
compact starbursts. We briefly summarize the theoretical and astrophysical underpinnings
of the super wind concept, stressing how even simple idealized models can provide an
informative interpretational guide to the observations.
We then discuss the growing body of multi-waveband observations of superwinds.
X-ray emission from superwinds is probably produced by a combination of internal
shocks in the high-speed superwind and by interactions between the superwind and
inhomogeneous ambient gas ('douds'). Most of the available data on super winds comes
from optical emission-line imaging and spectroscopy. Such optical line-emission is most
likely produced as the wind shock-heats and accelerates ambient gas in the disk or halo of
the starburst galaxy. These optical data provide morphological, kinematic, and physical
evidence for the existence of superwinds, and even allow us to estimate some of the key
superwind parameters. The interaction between the wind and cool atomic gas has been
probed by both optical absorption-lines and radio observations of the H I 21cm line.
Evidence for outflowing molecular gas in starbursts is also summarized. The nonthermal
radio continuum halos associated with some starbursts probably represent relativistic
magnetized plasma convected out of the star burst by the superwind.
We then summarize our attempts to take a census of superwinds in the local universe.
We use this census to estimate the rates at which superwinds eject metals and energy at
the present epoch. Finally, we briefly consider the implications of superwinds for diverse
topics in extragalactic astronomy. Superwinds may play a crucial role in the chemo-
thermal evolution of the intergalactic medium, in the chemical evolution of galaxies, in
the evolution (destruction?) of dwarf galaxies, in the stimulation or suppression of galaxy
formation at early times, in the quasar absorption-line phenomenon (through their effect
on galactic gaseous halos), and in the production of part of the X-ray background at
E<10keV.

1. Introduction

'Superwinds' - galactic-scale mass outflows driven by the collective effect of

supernovae and winds from massive stars - have a rich legacy in theoretical
astrophysics. They figure prominently in models for the formation and
evolution of galaxies (e.g., Dekel & Silk 1986; White & Frenk 1991; Mathews
1989; Berman & Suchkov 1991) and of the intergalactic medium (e.g.,
DeYoung 1978; Ostriker & Cowie 1981), for the dynamics and structure of
the interstellar medium and Galactic Halo (e.g., MacLow & McCray 1988,
hereafter MM; Heiles 1990; Dettmar 1993), for the origin of the cosmic X-
ray background (e.g., Bookbinder et al. 1980), and for some types of QSO
absorption-lines (e.g., York et a1. 1986).

455
J. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and EvoLution of GaLaxies, 455-498.
© 1993 KLuwer Academic PubLishers.
456

Despite their potential importance, superwinds have remained largely in

the realm of theoretical deus ez machina until quite recently. This situation
has changed dramatically in the last several years, and our primary goal in
this paper is to review how X-ray, optical, IR, and radio observations of star-
burst galaxies have been used to discover superwinds and to quantify many
of their basic physical properties. This review is designed to complement two
other recent reviews by Dettmar (1993) and Begehnan (this volume). While
Dettmar's emphasis is on the 'disk-halo connection' in typical spiral galaxies,
the emphasis in the present review is on starburst galaxies (e.g., galaxies in
which the gas depletion time is much less than a Hubble time and in which
the region of star-formation is largely concentrated to the central kpc or so
of the galaxy). Likewise, we leave to Begehnan the discussion of outflows
driven by powerful AGNs. It is indeed possible that some of the outflows we
will discuss in this review are driven by AGNs rather than by starbursts. As
we have argued elsewhere (Heckman, Lehnert, & Armus 1993), this is not
likely to be true in the majority of cases however.
In order to place the importance of these multi-waveband observations
in context, we will begin our review in §2 with a brief summary of the basic
theoretical and astrophysical underpinnings of the superwind concept. Mter
extensively discussing the most relevant observations in §3, we will describe
in §4 how our recent 'census' of super winds can be used to estimate the
rate at which they eject mass and energy in the local universe. We will then
conclude the review in §5 with a summary of the potential implications
superwinds have for many of the topics considered during this Summer
School. We direct motivated readers to the paper by Heckman, Armus, &
Miley (1990 - hereafter HAM) for a more detailed discussion of some of the
material to follow.

2. The Basic Physics and Astrophysics of Superwinds

The theoretical astrophysics pertinent to superwinds has been extensively
discussed in the literature. A comprehensive review of models of the closely
related phenomenon of 'supershells' can be found in Tenario-Tagle & Bo-
denheimer (1988), while more specific theoretical aspects of the superwind
phenomenon are treated by - for example - Chevalier & Clegg (1985 -
hereafter CC), Tomisaka & Ikeuchi (1988), Norman & Ikeuchi (1989), Lei-
therer, Robert, & Drissent (1993 - hereafter LRD), and Balsara, Suchkov,
& Heckman (1993).
Briefly, we expect a superwind to occur when the kinetic energy in the
ejecta supplied by supernovae and winds from massive stars in a starburst is
efficiently thermalized. This means that the collisions between stellar ejecta
convert the kinetic energy of the ejecta into thermal energy via shocks, with
little energy subsequently lost to radiation. This situation is expected to
 457

arise when the (suitably normalized) kinetic energy input rate is so high
that most of the volume of the starburst's ISM is filled by hot, tenuous
supernova-heated gas (e.g., McKee & Ostriker 1977; MM). The collective
action of the supernovae and stellar winds 'then creates a 'bubble' of very
hot (T up to 108 K) gas with a pressure that is much larger than that of its
surroundings.
As the over-pressured bubble expands and sweeps up ambient gas, it
will rapidly enter the 'snowplow' or radiative phase, which occurs when
the radiative cooling time of swept-up and shock-heated ambient material
becomes shorter than the wind's expansion timescale. Should the bubble
be expanding inside a disk-like ISM (as expected in a typical starburst),
the bubble will expand most rapidly along the direction of the steepest
pressure gradient (the disk's minor axis). It is likely that once the bubble
has a diameter that is a few times the scale-height of the disk, it will evolve
from the 'snowplow' phase into the 'blow-out' phase as the swept-up shell
accelerates outward and fragments via Rayleigh-Taylor instabilities. The
wind can then propagate out into the intergalactic medium at velocities of
several thousand kms- 1 (provided that radiative losses have been small -
cf. CC).

2.1 . MOMENTUM & ENERGY INPUT IN STARBURSTS

Superwinds will be driven by the kinetic energy and momentum supplied by

the starburst. Supernovae (primarily types II and Ib) , winds from massive
stars, and :radiation pressure will all contribute significantly in this regard.
In very useful recent paper, LRD have calculated the predicted return
of mass, momentum, and kinetic energy from an ensemble of massive stars.
These calculations have been made for a range of different metallicities and
initial mass functions. The time-evolution of these injection rates has been
followed for two extreme starburst 'histories': 1. an instantaneous burst with
all stars formed at once, and 2. a constant star-formation rate lasting' up to
5 X 107 yr. LRD show that stellar winds are most important at early times,
for the most top-heavy IMF's, and for the highest metallicities. Their results
are relatively insensitive to the IMF and metallicity for starbursts older than
about 107 yr when supernovae start to dominate the injection rates. This is
fortunate from the standpoint of comparisons of the models with the data,
since most starbursts are believed to last 10 7 to 108 yr (e.g., Rieke et al.
1980; Bernlohr 1993a,b).
For concreteness and ease of comparison to the LRD models, we will
assume a constant star-formation rate lasting for 5 x 107 yr, a normal
Salpeter IMF extending up to 100 M0' and solar metallicity. These seem
reasonable approximations for the typical starbursts we will discuss later.
Scaling of the corresponding LRD model for M 82 then leads to the following
458

predicted relationships between the starburst bolometric luminosity and the

various injection rates:

Kinetic Energy : E = 8 X 1042 Lbol,ll erg s-1 , (1)

Mas s : M = 3LOOl,11 M0 yr- 1 , (2)

Momentum: p .
= 5 X
34
10 Lbol,l1 dyne, (3)

where the bolometric luminosity is in units of 1011 L 0 .

Several comments are in order. First, for the IR-selected starbursts that
will dominate most of the discussion in §§3 and 4 to follow, it is probably
a reasonable approximation to equate the starburst bolometric luminosity
to the total (1 to 1000JL) IR luminosity of the galaxy - as given (for
example) by the two-dust-temperature fits to the IRAS data in Rice et al.
(1988). Second, note that the momentum flux associated with the starburst's
radiation pressure (Lbot/c) is about 25% as large as the momentum flux
due to supernovae and stellar winds given in Equation 3. It is also worth
noting that the mechanical energy output of the starburst is about 2% of
the bolometric luminosity, and that the mass injection rate represents about
10% (25%) of the star-formation rate for a Salpeter IMF with a lower mass
cut-off of 0.1 M0 (1 M 0 ) respectively.
The above equations are meant to represent 'typical' starbursts with
durations of at least 10 7 yr. Clearly, the relationships between such quantities
as bolometric luminosity, Lyman continuum luminosity, and the input rates
of mass, momentum, and energy could vary substantially with time for an
instantaneous starburst. Since the kinetic energy input at times later than
10 7 yr in a starburst is dominated by supernovae, since a normal IMF rises
towards low mass, and since the kinetic energy per supernova is roughly
independent of progenitor mass (e.g., Woosley & Weaver 1986), the rate at
which an instantaneous starburst injects kinetic energy will actually peak
at about the time the least massive type II supernova progenitors explode.
A minimum supernova progenitor mass of 8 M0 corresponds to an implied
maximum kinetic energy injection rate at a time of about 5 X 10 7 yr (cf.
LRD). Thus, a starburst in which the duration of star-formation is much
shorter than this time-scale (e.g., Bernlohr 1993a,b), will have a peak in the
superwind's mechanical energy output at a time when no stars more massive
than about 8 M0 remain (i.e., at a main sequence turn-off at B4V).
Such a post-burst galaxy would have a significantly reduced bolometric
(IR) luminosity. The instantaneous burst models of Charlot & Bruzual
(1991) show that LOOl has dropped by a factor of about 40 at 5 X 107
 459

yr with respect the peak value. Even more importantly, the very steep
dependence of the Lyman continuum luminosity with stellar mass means
that this post-burst would have a truly negligible ionizing luminosity: the
models of Charlot & Bruzual (1991) and those of De-Gioia-Eastwood (1985)
together imply a drop in LLyc by a factor of about 30,000 with respect to
the peak value. Kornneef (1993) has dubbed such objects 'Postburst Infrared
Galaxies' ('PIGs'), and has proposed that the superwind system NGC 4945
(HAM) represents the prototype. Other superwind galaxies like Arp 220,
NGC 6240, and NGC 3079 in HAM which all have unusually small ratios of
Bq to bolometric luminosity (cf. Beck 1993) may be additional examples of
PIGs. The dwarf galaxy NGC 1705 seems to be an example of a low-mass
post-burst that is driving a global outflow (M~urer et al. 1992).

2.2. SIMPLE MODELS OF SUPERWIND HYDRODYNAMICS AND RADIATION

The simplest possible hydrodynamical model for a superwind would be
spherically-symmetric and would ignore the effects of gravity, radiative
cooling, and ambient gas (Le., a 'free wind'). CC have discussed such a
model in the context of M 82. They assume that there is a spherical region
of radius r. (the starburst) inside which mass and kinetic energy are injected
at a constant rate per unit volume. The result is a region of hot gas which
slowly expands through a sonic radius (located approximately at the radius
of the starburst) and is then transformed into a supersonic outflow.
HAM have found that this model is a reasonable fit to the radial pressure
profiles in a sample of well-studied starburst superwind systems (see §3.2
below) . In such a model, the gas pressure inside the starburst (and hence
inside the sonic radius of the wind) is just the static thermal pressure of
the hot fluid that has been produced by thermalization (shock-heating) in
high-speed collisions between material ejected in stellar winds and supernova
explosions. As discussed by CC, this pressure is essentially determined by the
size of the star burst and the rate at which mass and kinetic energy is injected.
Since the sound-crossing time inside the star burst ·is much shorter than the
outflow time (e.g., since the flow is subsonic inside the starburst itself), the
pressure is roughly constant throughout this central region. Following 'CC,
and using Equations 1 & 2 above to relate the injection rates to the starburst
luminosity, we expect the central pressure to be given by:

(4)

Thus, the smaller the starburst radius (r.), the higher the central pressure
for a given star burst luminosity.
For radii much larger than the starburst radius, the pressure of the
outflow is dominated by the wind's ram pressure. This will essentially be
460

set by the rate at which the starburst injects momentum into the flow, and
will drop like the inverse square of distance from the starburst. Following
CC and using Equation 3 above we get :

(5)

where r is the distance from the starburst, assuming r ~ r •.

The temperature inside the star burst is just set by the mean amount of
kinetic energy per gram in the material injected by supernovae and stellar
winds, and the models of LRD then imply that To = 1.2 X 108 K for the
assumptions discussed in §2.1. As this hot fluid expands out through the
sonic radius, it is adiabatically cooled (it is easy to show in the context of
the CC models that adiabatic cooling dominates radiative cooling for the
wind fluid). The CC models combined with the LRD models then imply
that for r ~ r.:

(6)

Finally, the terminal velocity of the outflow (velocity at r ~ r.) is just

the square root of twice the kinetic energy per gram in the material injected
by the supernovae and stellar winds (cf. CC). Equations 1 and 2 above imply
that the terminal velocity will be about 2900 km s-l.
The next level of complication is to retain spherical symmetry, but to put
in a homogeneous ambient medium that the wind must push against. This
is a well-understood problem first considered in the context of wind-blown
bubbles inflated by individual stars (e.g., Castor, McCray, & Weaver 1975).
Dyson (1989) has recently published an excellent review of the relevant
physics.
Briefly, in this situation one has an 'onion-skin' structure of five concentric
zones. From inside-out these are: 1. An innermost region inside which the
mass and energy is injected (the starburst), 2. A region of supersonic outflow,
3. A region of hot gas (wind material that has passed through an internal
shock that divides regions 2 and 3), 4. A thin, dense shell of ambient gas that
has been swept-up, shocked, compressed, and then radiatively cooled as the
'piston' of hot gas in zone 3 expands into the ambient medium, and 5. The
undisturbed ambient gas. Such an expanding bubble may be either energy-
conserving or momentum-conserving (depending on whether radiative losses
in zone 3 are negligible or not).
For concreteness, let us adopt an energy-conserving structure inflating in
a medium of constant density no (cm- 3 ). Two ofthe most relevant properties
of the gas in zone 3 (shocked wind material) are then given by (cf. MM):

T - 3
- X 10 7
no2/35 L 8/35 t- /
6 35
bol,l1 15
K
, (7)
 461

L X -- 2 X 10 40 37/35 t 13 / 35
no18/35 L 001,11 15 erg s -1 , (8)

where tIS is the age of the bubble in units of 1015 s (about 32 Myr, and
we have again used the relations from LRD in Equations 1 through 3 above
to express T and Lx in terms of the starburst luminosity. We therefore
expect zone three to be a strong source of X-rays (though note that the
X-ray luminosity is only about 10- 4 of the starburst lu.minosity). It is also
helpful to note that an ambient density of unity inside a sphere of radius r
corresponds to a total gas mass of 108 [7' jkpc]3 M 0 .
For the same model the chief parameters of zone 4 (the expanding shell
of shocked ambient gas) are given as (cf. Castor, McCray, & Weaver 1975):

(9)

= 100L 1/5 no-1/5 t-15 / km s -1 .

2 5
V 6 hell oo1 ,l1 (10)

Thus, the shell expands far more slowly than would a freely expanding
wind. The typical shock-speeds implied by Equation 10 yield post-shock
temperatures in zone 4 of order few times 105 K. This is near the peak of
the cooling curve (e.g., Spitzer 1978), and cooling times in zone 4 should be
much shorter than the bubble expansion time.
We would expect zone 4 to be a strong source of optical and UV line
emission, with a predicted Ha luminosity (from shocks alone) of-order:

LHo: = 1041 Lbol,ll

3/5 2/5 4/5
no t 15 erg s
-1
. (11)

Thus, the Ha luminosity of zone 4 is a few times larger than the X-ray
luminosity of zone 3, and is a few times 10- 4 of the starburst luminosity.
Since the Ha line typically represents 1 to 2% of the total emission-line
luminosity in a shock (e.g., Shull & McKee 1979), the total line luminosity
due to shock-heating of zone 4 will be of-order 1 to 2% of the starburst
bolometric luminosity. Since the total wind kinetic energy flux is about 2%
of the starburst bolometric luminosity (see Equation 1 above), this may be
regarded as an upper bound on the luminosity of gas that has been shock-
heated by the wind.
For the more general case in which an energy-conserving bubble is inflated
into an ambient medium in which the density declines with radius like n <X
rf3, the expression corresponding to Equation 10 above has v <X t-(f3+ 2 /f3+ 5 )
- cf. Dyson (1989). For values of fj < -2 the shell therefore accelerates
outward. Thus, it is expected that Rayleigh-Taylor instabilities will cause
462

the shell to fragment, allowing the the wind fluid in zones 2 and 3 to then
freely escape from the ruptured bubble. Such a situation is referred to as
'blow-out' in the jargon of the trade.
As a more geometrically realistic model, we can consider a bubble being
inflated in a plane-parallel stratified atmosphere (e.g., the gaseous disk of
a starburst). The structure and evolution of the bubble is very similar
to the spherical model above. However, in the disk case, the bubble will
expand most rapidly along the direction of the steepest pressure gradient
(e.g., normal to the plane of the disk). This process has been studied
by numerous authors (e.g., Schiano 1985; MM; Tomisaka & Ikeuchi 1988;
Balsara, Suchkov, & Heckman 1993; and see the review by Tenario-Tagle &
Bodenheimer 1988). For an exponential density law in the vertical direction,
the shell of shocked ambient gas will accelerate .and fragment (e.g., blow-
out will occur) when the bubble's radius reaches a few disk scale heights.
Thereafter, the hot gas injected inside the starburst could vent directly out
into the galactic halo (or beyond).
A schematic version of the evolution of this process is shown in Figure
1a,b. Figure 1a shows the radiative phase before 'blow-out'. The molecular
disk fueling the starburst is labeled 'MD', and the associated supernovae
and stellar winds are indicated by stylized stars. The starburst has partially
evacuated a central cavity (labeled 'C') that is filled with the hot (T rv 108 K)
thermalized supernova and stellar wind ejecta. This hot fluid passes through
a sonic radius and then expands outward at several thousand km s-1 as a
free wind (arrows labeled 'FW'). It passes through a:ll internal wind shock
(dashed line labeled 'WS') and is then reheated back to Trv10 8 K. This
shocked-wind material ('SW') can be an important source of thermal X-ray
emission. The pressure of the wind inflates a 'bubble' which is elongated
along the vertical axis, the direction of the maximum pressure gradient in
the gaseous halo ('GH'). The solid contours are isodensity contours of the
gaseous halo, which has a vertical scale-height labeled 'SH'. The shock driven
by the expanding hot bubble will sweep up the a:mbient gas, accelerating it
to velocities of a few hundred km s-1 (with a velocity field indicated by the
short arrows). The shocked gas then cools radiatively into a thin shell ('S').
This shock-heated shell will be an important source of optical line-emission.
Figure 1b shows the situation after the bubble expands more than a
couple of vertical scale-heights it can 'blow-out' of the halo. The bulk of
the interior of the region evacuated by the wind is now occupied by the
freely expanding wind fluid, with a small shell of shocked-wind fluid at
the leading edge of the outflow separated from the ambient halo gas by a
contact discontinuity (dot-dashed line labeled 'CD'). Bow shocks in the wind
('BS') will form around density inhomogeneities in the galaxy halo ('SC' for
'shocked clouds'). These clouds will be accelerated outward (short arrows),
and may radiate opticaljUV emission-lines and soft X-rays depending on
 463

GH

SW

5- __
......

FW "\ \

?\

Fig. 1. Schematic diagrams showing the essential features predicted for two
evolutionary phases of a superwind expanding into the gaseous halo of a disk galaxy.
464

the speed of the shocks driven into them.

2.3 . CONDITIONS FOR DRIVING SUPERWINDS

In this section, we will argue that there are two basic conditions to be met
if a starburst is to drive a superwind. First, the energy injection rate inside
the starburst must be hlgh enough to create a cavity of hot gas whose
radiative cooling time is much greater than its expan~ion time, and second,
the wind must be sufficiently powerful to ultimately blowout of the ISM of
the starburst (Le., not remain trapped as a hot bubble inside the galaxy).
We will show that these conditions are in fact met in typical starbursts
(see HAM for details). Norman & Ikeuchi (1989) have reached the same
conclusion using related arguments.
In order to drive a superwind, the kinetic energy supplied by supernovae
and stellar winds must be efficiently converted into the bulk kinetic energy
of the outflow - that is, very little of the energy should be lost in the form
of radiation from the hot gas at the base of the outflow. This condition will
be met if the energy injection rate is high enough to create a hot, tenuous
gas phase that dominates the ISM of the starburst (Le., has a volume-filling
factor near unity). In this case, supernova and stellar wind ejecta will expand
into a low-density medium and the radiative cooling time of the ejecta will
exceed the wind outflow time. This criterion can be expressed in terms of
the well-known McKee - Ostriker (1977) ' porosity' parameter. The volume
filling factor of the hot, tenuous phase is given by:

!!HOT =1- exp[-QsNRJ , (12)

where

QSNR S -0.14
= 3E511.28 -8 n O,100
p,-1.3
7 • (13)

In Equation 13, E51 is the kinetic energy per supernova in units of 10 51

ergs, S-8 is the supernova rate in units of 10- 8 SNe per year per pc3 , nO,100
is the mean gas density in starburst in units of 100 cm- 3 , and P7 is the
pressure in the starburst in units of 107 K cm- 3 • For typical parameters
appropriate to starbursts, HAM concluded that QSN R will be of-order ten,
so that ffHoT is near unity.
An alternative way of evaluating whether radiative losses from the hot
gas are important enough to stall the superwind is to use the expression
derived by MM for the ratio of the radiative cooling time to the bubble
expansion time for a supernova-driven bubble expanding in an disk with an
exponential vertical density law with scale-height H and mid-plane density
no:
 465

t cool It e:cp L O•61

= 80 n O,100
-1.1
bol,l1
B-1.7
lOOpc , (14)

where we have used Equations 1 - 3 from the LRD models to write Equation
14 in terms of the starburst luminosity. Again, HAM showed that tcool ~ t exp
for typical starburst parameters.
Finally, for the same 'bubble in an exponential disk' model, MM conclude
that blow-out will occur if the dimensionless wind power A > 100. Using our
Equations 1 - 3 to express this power in terms of the starburst bolometric
luminosity, and following MM, we find:

A = 24000 L 001,11 B 100pc

-2 p-3/2
7
1/2
nO,lOO • (15)

Typical parameters appropriate to starbursts then imply that blow-out is

likely (see HAM for details).

2.4. THE EFFECT OF CLOUDS

So far, we have considered the theory of superwinds only in the context of

a wind blowing into a homogeneous (single-phase) mediwn. In reality, the
gaseous halos of star-forming galaxies are clearly multi-phase 'cloudy' sys-
tems (cf. Dettmar 1993; Bloemen 1991; and the review of QSO absorption-
line systems by Steidel and Lanzetta in this volume). Such clouds could have
a profound impact on the thermal and dynamical evolution of a superwind,
and on the strength and character of the radiation they produce.
Let us then imagine a case in which clouds occupy only a small fraction
of the volume swept by the superwind, but contribute a significant fraction
of the gas mass within that volwne. The evolution of the wind-blown cavity
will still be qualitatively similar to the scenarios discussed above. The wind
will propagate through and shock-heat the tenuous intercloud mediwn,
inflating an expanding cavity. The clouds engulfed by the wind will suddenly
find themselves highly over-pressured (and hence shock-heated) as they are
subjected to the thermal pressure of the hot intercloud mediwn and/ or the
ram pressure of the wind itself (cf. Fig 1). The clouds will then preswnably
be destroyed by either hydrodynamical processes (cf. Hartquist & Dyson
1988; Stone & Norman 1992), by evaporation due to conductive heating
(e.g., Cowie & McKee 1977), or by a combination of these processes in the
'mixing layers' at the cloud-intercloud interface (e.g., Begelman & Fabian
1990; Slavin, Shull, & Begelman 1993).
This 'mass-loading' by clouds of the intercloud medium will have several
important consequences. First, the clouds will provide a source of cool, dense
material which (when mixed with the intercloud mediwn) could significantly
decrease the cooling time inside the wind-blown cavity. This could turn an
energy-conserving outflow into a momentum-conserving one, and could even
466

rob the bubble's interior of so much thermal energy that the bubble collapses
before blow-out occurs (cf. Hartquist & Dyson 1988; White & Long 1991;
McKee, Van Buren, & Lazarefi' 1984; Berman & Suchkov 1991). Second,
the increased radiative cooling inside the superwind cavity should make the
superwind's emission more readily detectable.
Since this review is primarily about the observational manifestations of
superwinds, we will address this second point in a bit more detail. It is
likely that engulfed clouds may in fact be the dominant sources of the
soft X-rays and optical emission-lines observed from superwinds (see ;§§3.1
and 3.2 below). For example, Begelman & Fabian (1990) and Slavin, Shull,
& Begelman (1993) have proposed that hydro dynamical processes at the
interface between a cloud and the intercloud medium through which it moves
will lead to the development of a 'mixing layer': a region containing gas from
both media whose temperature will be roughly equal to the logarithmic
aveage of the temperatures of the two media. These mixing layers and the
clouds they irradiate can be a copious source of optical, EUV, and soft X-ray
emission.
Clouds heated by shocks driven into them by the superwind will also be
sources of radiation. Balancing momentum across the shock implies that the
speed of the shock driven into the cloud will be related to the superwind
velocity by the inverse of the square root of the cloud-wind density contrast.
U sing Equation 2 above for the mass outflow rate in the wind, and adopting
a wind speed of 2900 km s-l for consistency (§2.2 above), implies:

1/2 -1 -1/2 k -1
(16)
V"hock = 166L bo /,1l r kpc ne/oud m s ,

for a cloud with a particle density ne/oud (cm -3) located a distance r (in kpc)
from the starburst.
Such a shock will heat the cloud to a temperature

Te/oud = 1.4 X 105 (Ve/oud/100 km s-1)2 K

=3.9 X 10 5 Lbo/,ll ri:;c nci!ud K , (17)

and we therefore see that such shocked clouds would produce soft X-rays,
ionizing radiation, and optical emission-lines (cf. Shull & McKee 1979;
Binette, Dopita, & Tuohy 1985).
The wind will also accelerate these clouds outward. Following CC, we
can estimate the terminal velocity reached by a cloud with a column density
typical of interstellar clouds in the Milky Way (N21 in units of 10 21 cm- 2 )
accelerated outward from an initial location rkpc (well outside the sonic
radius of the wind):
 467

k
Vcloud = 400 L bol,ll
1/ 2 -1/2N-1/2
r kpc 21 m s
-1
(18)

Here we have used Equations 1 and 2 above to relate the mass and energy
flux in the wind to the starburst luminosity. We therefore expect the shocked
clouds responsible for optical emission-lines to have typical outflow speeds
of a few hundred km s -1.

3. Observational Diagnostics of Superwinds

3.1. X-RAY OBSERVATIONS OF SUPERWINDS

The overall X-ray properties of starburst galaxies have been recently re-
viewed by Petre (1993), so we will focus on the interplay between the X-ray
data and the superwind phenomenon. X-ray observations of superwinds are
crucial to our understanding of the phenomenon, because (potentially at
least) they offer the most direct probe of the hot, tenuous superwind fluid
itself (rather than simply probing the interaction of the wind with ambient
gas, as is most likely the case for the optical line emission).
The characteristic temperature corresponding to the complete thermal-
ization of the kinetic energy from an ensemble of supernovae and winds
from massive stars is about kT = 10 keY (e.g., 10 51 ergs per 10 M 0 ) -
see §2 above. Thus we expect hard X-ray emission from a galaxy driving
a superwind to come from the hot and tenuous supernova/wind-heated gas
inside the starburst (i.e., inside the sonic radius of the wind). Since the
corresponding terminal velocity of the out flowing superwind fluid is several
thousand km S-1, hard X-rays will also arise from internal shocks in the
superwind (cf. Tomisaka & Ikeuchi 1988; Balsara, Suchkov, & Heckman
1993). In the idealized expanding bubble discussed in §2 above, these X-
rays will be coming from zone 3.
Soft X-rays will be produced by dense ambient material that is shock-
heated and/or evaporated by the superwind (cf. Watson, Stanger, & Griffiths
1984; White & Long 1991). In the idealized 'expanding bubble' model
discussed in §2, this would correspond to the thin shell of shocked ambient
gas at the interface with the undisturbed ambient gas (zone 4). In a more
realistic situation, the soft X-rays will arise as the wind collides with, shock-
heats, and evaporates density inhomogeneities ('clouds') in the galaxy halo.
H I observations of starbursts show that such cool, dense material indeed
exists in the environs of starbursts (see §3.4 below). Note that the typical
outflow velocity of dense, wind-accelerated ambient material is observed in
the optical to be several hundred km S-1 (see §3.2 below), corresponding to
post-shock temperatures of kT rv 200 eV.
This theoretical/phenomenological picture is in reasonable agreement
with the limited amount of X-ray data presently available for starburst
468

galaxies. The X-ray luminosity of starburst galaxies correlates rather well

with both the IR luminosity (e.g., Griffiths & Padovani 1990; David, Jones,
& Forman 1992; Green, Anderson, & Ward 1992) and the Lyman continuum
luminosity (Ward 1988) of the starburst. Moreover, the observed average
ratio of X-ray and bolometric starburst luminosities agrees roughly with
the predictions of simple theoretical models for star burst-driven superwinds
(HAM - and see the discussion in §2 above). X-ray spectra of starbursts
imply characteristic temperatures of kT = 6 to 9 keY for the global X-ray
emission (e.g., Kim, Fabbiano, & Trinchieri 1992; Ohashi et al. 1990; Petre
1993), again in agreement with simple models (see §2 above).
Spatially-resolved X-ray images of well-studied starburst galaxies whose
large-scale stellar disks are viewed nearly edge-on (M 82, NGC 253, NGC
2146, and NGC 3628) show striking X-ray 'halos' or 'plumes' that can extend
out to a radius of 10 kpc or more along the galaxy minor axis. These have
been interpreted as direct evidence for galactic superwinds (e.g., Watson,
Stanger, & Griffiths 1984; Fabbiano 1988; Fabbiano, Heckman, & Keel 1990;
Heckman & Fabbiano 1993; Armus, Heckman, & Weaver 1993). The trans-
galactic-scale X-ray nebulae associated with such 'ultraluminous' IR galaxies
as Arp 220, NGC 3690, and Mrk 266 are likely to be more energetic versions
of this phenomenon (Eales & Arnaud 1988; Armus, Heckman, & Weaver
1993).
In general, the estimated thermal energy content and mass of the X-ray
gas are consistent with the time-integrated kinetic energy and mass output
of the starburst (cf. HAM and references therein). However, these estimates
are based on several simplifying assumptions. The most important of these
are that all the observed X-rays arise from hot gas at a uniform - and high
- temperature eq1.:.al to that deduced on the basis of the integrated X-ray
spectrum, and that this gas has a volume filling factor of unity. As we now
discuss, the actual situation is likely to be considerably more complex.
The X-ray emission from the halo of M 82 is significantly softer than
that from the central starburst (Petre 1993), as is also the case in NGC
3628 (Fabbiano, Heckman, & Keel 1990; Heckman & Fabbiano 1993) and
NGC 3690 (Armus, Heckman, & Weaver 1993). Within the context of the
superwind model, this could imply that the wind has broken free of the
galaxy and has suffered severe adiabatic cooling on its way out (cf. CC
and Equation 6 above). However, this is unlikely to be the case, since the
X-ray nebulae are far too bright to simply be an adiabatically-cooled free-
wind (e.g., Mathews & Doane 1993). Our new ROSAT images of several
starbursts with superwinds also show a considerable amount of fine-scale
structure: X-ray-bright 'lumps' and 'filaments' with sizes of a few hundred
pc to a few kpc (Heckman & Fabbiano 1993; Armus, Heckman, & Weaver
1993). We believe it is most likely that the soft (kT < keY) X-ray emission
in the starburst halos arises predominantly from ambient clouds that are
 469

being shock-heated and/or evaporated by the superwind.

Petre (1993) emphasizes that BBXRT data on starbursts show clear
evidence for multiple X-ray spectral components, and Armus, Heckman,
& Weaver (1993) have reached similar conclusions based on ROSAT PSPC
data. X-ray images of the two nearest starburst galaxies (M 82 and NGC
253) also show that the X-ray emission is produced by both the spatially-
extended component discussed above and a collection of discrete sources
in or near the starburst (Watson, Stanger, & Griffiths 1984; Fabbiano &
Trinchieri 1984). Thus, it may well be that much of the hard X-ray emission
from starbursts is produced by the ensemble of X-ray binary systems in the
starburst (e.g., Griffiths & Padovani 1990). Inverse Compton scattering of
IR photons off the relativistic electrons responsible for the radio continuum
emission may also contribute substantially to the hard X-ray emission (e.g.,
Schaaf et al. 1989; but Seaquist & Odegard 1991 and Tsuru 1992 argue
otherwise). A composite thermal plus nonthermal origin for the hard X-rays
could explain the relative weakness of the 6.7 keY Fe K line in M 82 and
NGC 253 compared to expectations for optically thin thermal emission from
gas with solar Fe abundances and kT = 5 to 10 keY (Tsuru 1992; Ohashi et
al. 1990; Petre 1993).
One of the most interesting X-ray observations of a likely superwind is
the Ginga data on the center of our own Milky Way (Yamauchi et al. 1990).
Emission-line images using the 6.7 ke V Fe K line imply a region of gas with a
temperature of about 10 8 K, dimensions of about 270 x 150 pc, gas pressures
of P/ k = 8timesl0 6 K cm- 3 , and total thermal energy content of6times10 53
ergs. Since the sound speed in this gas (""' 1500 km S-1) greatly exceeds the
escape velocity from the Galactic Center, the gas should flow out as a high
speed wind (unless it is confined by some other mechanism). If this outflow
is a steady-state phenomenon, collisional heating with a supernova rate of
about 0.6 per century within this volume is required to balance adiabatic
cooling.
To summarize, while the current situation regarding X-ray emission from
superwinds is unclear, we can expect dramatic progress over the next few
years as the avalanche of new ROSAT, BBXRT, and ASTRO-D data on
starbursts is analyzed and digested. Certainly the data already in hand are
very tantalizing.

3.2. OPTICAL LINE EMISSION FROM SUPERWINDS

Most of the existing data on superwinds concern the optical line emission
associated with such outflows. Such data provide a whole array of diagnostics
of the dynamical and physical state of the outflow.
In the simple schematic pictures described in §2 above, the optical line
emission would not arise from the very hot and tenuous superwind fluid
470

itself (the cooling times are excessively long, and most of the relatively feeble
radiation that is produced there is in the form of hard X-rays). Instead, we
would expect the optical line emission to come from relatively dense ambient
material into which the wind's ram pressure drives slow « few hundred km
s-l) radiative shocks (e.g., McKee & Hollenbach 1980 and see §2.2 above).
In the wind-blown bubble 'onion-skin' model, this material would be the
thin, dense shell of compressed and shock-heated ambient gas at the leading
edge of the bubble. In a more realistic situation in which blow-out has
occurred, and/ or in which the wind is propagating through a inhomogeneous,
multi-phase medium, the optical line emission will come from clouds (e.g.,
fragments of the ruptured shell, pre-existing density inhomogeneities in the
halo, or material carried out of the galactic disk by the superwind).
We therefore expect that the optical data will provide a detailed - but
indirect - view of the superwind phenomenon. The largest compilations of
such data are found in the recent spectroscopic and narrow-band imaging
surveys by HAM, Armus, Heckman, & Miley (1989,1990- hereafter AHM89,
AHM90), Lehnert (1992 - hereafter L92), and Lehnert & Heckman (1993 -
hereafter LH).

3.2.1. Structure and Luminosity

Galaxies selected to be both luminous and warm in the far-IR have optical
emission-line nebulae whose size and luminosity correlate 'r easonably well
with the far-IR luminosity, and hence with the estimated formation rate
of massive stars (AHM90; L92; LH). The Ha luminosities of the extra-
nuclear portion of the nebulae (e.g., well outside the starburst region) are
just consistent with the predictions of the superwind model (see Equation
11 and the associated discussion in §2 and in HAM).
We (L92, LH) have recently completed an Ha imaging survey of a sample
of edge-on disk galaxies selected on the basis of far-IR flux and color, and
find many examples which show emission-line filaments, loops, or bubbles
extending out one to ten kpc along the optical minor axis of the galaxy
(four examples are shown in Figure 2). In general, we find a pronounced
excess of optical line emission along the optical minor axis compared to what
would be expected for ordinary emission-line gas confined to the galactic
disk. At the highest levels of IR luminosity (L > few X 1011 L 0 ), the Ha
nebulae approach trans-galactic dimensions (30-100 kpc), with the large-
scale morphology often dominated by filaments, loops, or bubbles (e.g.,
Heckman, Armus, & Miley 1987; AHM90) . Some examples of such nebulae
are shown in Figure 3.
HAM, L92, and LH have measured radial electron density profiles in the
emission-line nebulae associated with about 20 IR-selected starburst galax-
ies. Densities range from 103 to <10 2 cm- 3 • The observed Ha luminosities
 471

Fig. 2. Ha+[NII] images of IR-bright, edge-on disk galaxies exhibiting large-scale

optical emission-line filaments extending several kpc out of the galaxy disk and into
the halo. (a) NGC 660j (b) NGC 3079j (c) NGC 3628j and (d) NGC 4666 - see
AHM90j Fabbiano, Heckman, & Keel 1990j L92j LH).
472

Fig. 3. Ha+[NII] images of two galaxies with IR luminosities greater than 10 46 erg
s-1: (a) Arp 220 (from Heckman, Armus, & Miley 1987) and (b) Mrk 266 (from
AHM90) . These represent the superwind phenomenon at the high-luminosity end,
where the emission-line nebulae are tens of kpc in size.
 473

and the measured densities allow us to estimate the mass and the volume-
filling factor of optical emission-line gas. The derived masses range from 10 5
M0 to 10 7 M0' and the volume-filling factors are typically 10- 3 to 10- 4 •.
Thus, the emission-line gas represents a modest amount of relatively dense
material distributed in the form of clumps, sheets, or filaments that occupy
only a very small fraction of the volume of the halo of the starburst galaxy.
In the few starburst galaxies for which both Ha and X-ray images are
available, there is a clear correspondence between the two images (e.g.,
Watson, Stanger, & Griffiths 1984; Armus, Heckman, & Weaver 1993). In
some cases, the optical line emission is preferentially located along the outer
boundary of the X-ray nebula (Fabbiano, Heckman, & Keel 1990; Heckman
& Fabbiano 1993; McCarthy, Heckman, & van Breugel 1987 - and see
Figure 4) . This suggests that the optical line emission arises at the interface
between the hot superwind and the ambient interstellar gas in the halo of
the starburst galaxy (in good agreement with both the theoretical picture
sketched above and the kinematics of the optical emission-line gas in several
well-studied cases, as summarized below). New ROSAT images should allow
us to conduct many more such comparisons, in order to better test this idea.

3.2.2. Kinematics and Dynamics

The kinematics of the optical emission-line gas associated with IR-Iuminous
galaxies also suggest that outflows are common. Optical spectra of the
nuclei of IR-Iuminous galaxies show that they often have blue-asymmetric
emission-line profiles (e.g., AHM89; LH; Phillips 1992). Mirabel & Sanders
(1988) find that the optical emission-line velocities in the nuclei of high-
luminosity IR galaxies are blueshifted with respect to the galaxy systemic
velocity by an average of about 100 km S-1. Both the blue-asymmetric
profiles and blueshifts suggest an outflow of ionized gas whose redshifted
back side is obscured by dust.
For the edge-on star burst galaxies investigated by L92 and LH, in many
cases the line widths actually increase with increasing radius along the minor
axis. The emission-lines are also systematically broader along the minor than
along the major axis (with typical FWHM's of 150 to 350 km S-1 vs. < 100
to 200 km S-1 respectively). The line widths along the minor axis correlate
strongly with the star burst IR luminosity, but not with the galaxy rotation
speed, thus favoring a starburst-driven outflow over simple orbital motions
in the galaxy gravitational potential (see Figure 5). This interpretation is
further supported by the strong trend found by L92 and LH for the largest
measured velocity shears along the minor axis to occur in the galaxies viewed
more nearly face-on (as expected for a starburst-driven radial outflow along
the minor axis) and in the galaxies with the highest IR luminosities (see
Figure 6).
474

-25' 33 ' 0 0 "

-25 ' 33 ' 300"

0
II)

~
z
0
i= -25 '34 '0.0 "
<t
Z
....I
U
W
0 0 CJ
()
- 2 5'34'30.0"
0
CJ 0

RIGHT ASCENSION ( 1950)

Fig. 4. Overlaid Ho+[N II] image (greyscale) and Einstein HRI X-ray image
(contour plot) of the central few kpc of the prototypical IR starburst galaxy NGC
253. The starburst nucleus is located in the upper right corner. Note how the
filamentary optical line emission to the SE of the nucleus seems to enclose the
X-ray plume (cf. McCarthy, Heckman, & van Breugel 1987j Fabbiano & Trinchieri
1984j LHj L92).

The detailed kinematic properties of the gas located along the minor axes
of such nearby and well-studied edge-on starbursts as M 82 (e.g., Bland &
Tully 1988; HAM), NGC 253 (Ulrich 1978; HAM), NGC 3079 (HAM; Filip-
penko & Sargent 1992), and NGC 4945 (HAM) provide more direct evidence
for outflows. All four galaxies have bubble-like or filamentary emission-line
nebulae that protrude out about a kpc along the optical minor axis. The
kinematics of these structures (broad, double-peaked emission-line profiles
in the center of the bubble, narrowing to single-peaked profiles along the
bubble periphery) imply that they are either expanding bubbles or the walls
of cone-like or cylindrical outflows, with inferred outflow/expansion speeds
 475

3 3
Off - Nucleus (Irl >ZR.)
. "."
---.
~

> 2 R.
.
1 1
U
•
CIJ
<1l 2.5 -
Ol
E I" I "

..
I"
E 2.5
• ""
~
I I
I I
~ I I
" "
I
"" I
5' I "
~

Z I z 2
I
" I

""
~
I
-........ I
• I " '-"'

::::s ": I
" ~
::r: 2 I
I I
"I"
I
:r:
;;t: ~
~ ~ ~ 1.5
I • I

-
tlIl tlIl
0 0
.....l

1.5
9.5 10 10.5 11 11.5 12 2 2.2 2.4 2.6 2.8 3
log 4x (I.e) Log V...,tatlon (lan s -1)

Fig. 5. Summary of the kinematics of the emission-line gas located outside the
central starburst and along the optical minor axis for the IR-selected edge-on disk
galaxies investigated by L92 and LH. The x-axis in both plots is the FWHM of
the [N IIj6584 line. This line width correlates strongly with the IR luminosity
of the galaxy (left plot), but not with the rotation speed of the galaxy (right
plot), implying that the gas dynamics is influenced much more strongly by the
star burst-driven wind than by gravity.

ranging from about 200 km s-1 to nearly 1000 km s-1 (see Figure 7). The
sizes and expansion speeds of these structures are in good agreement with the
quantitative predictions of the simple wind-blown-bubble model discussed in
§2 (see equations 9 and 10 above, and HAM). Additional examples of similar
outflow structures are found in the composite Seyfert/starburst galaxies
NGC 1365 (Phillips et al. 1983a), NGC 5506 (Wilson, Baldwin, & Ulvestad
1985), NGC 7582 (Morris et al. 1985), and Mrk 509 (Phillips et al. 1983b).
Qualitatively similar, but even larger and more energetic examples of
such expanding structures, are found in several 'ultraluminous' IR galaxies
(HAM). These are the 'double-bubble' emission-line nebula in Arp 220, the
central 'hour-glass' structure in NGC 6240, and the extraordinary nebula
associated with IRAS 00182-7112 in which line-widths are nearly 1000 km
476

12
0.8
., .
III
........
0.6
. "]
'-"" 11
•
. . .-.
. ••
.0
..s
0: •••• I

-
tlD

... .
0 tlD

..·• •..
..s
~. 10
0.4

· .. ••
0.2 9~LJ.....J~-L..J.....1....L..L.L...l-I....L....L...l-I....L.L.J

0.5 1 1.5 2 2.5 0.5 1.5 2 2.5

log ItoVI-t.V2Imlnorott (Nil; Ian sec-I) log It.V I -t.V2Imlnoro!t (NIl; km sec-I)

Fig. 6. Summary of the kinematics of the emission-line gas located along the optical
minor axis for the IR-selected edge-on disk galaxies investigated by L92 and LH.
These two plots show that the velocity shear along the minor axis is largest in the
galaxies that are viewed more nearly fa.:e-on (left plot) and in the galaxies with the
largest IR luminosities (right plot).

S-1 over a region 30 kpc in extent. Colina, Lipari, & Macchetto (1991)
have also published kinematic evidence for a superwind in the ultraluminous
galaxy IRAS 19254-7245.
On the 'micro-starburst' level, Meurer et al. (1992) have found a kpc-
scale bubble of ionized gas expanding at about 50 km s-1 in the core of the
post-starburst dwarf galaxy NGC 1705. They show that this expansion was
probably powered by the kinetic energy supplied by a burst of star-formation
that occurred some 10 Myr ago. Royet al (1991) have found a kinematically
similar 200 pc-scale expanding bubble of ionized gas in the starbursting
dwarf irregular galaxy NGC 2366, while de Vaucouleurs, de Vaucouleurs, &
Pence (1974) found evidence for large-scale outflow in the similar system
NGC 1569. JHU graduate student A. Marlowe with T. Heckman, R. Wyse,
and R. Schommer have now found an additional half-dozen similar examples
in their echelle and Fabry-Perot Do: survey of UV-bright dwarf galaxies
 477

Fig. 7. Long-slit spectrum of (from left to right) the [N II]A6548, Ha, [N II]A6584
emission-lines in the central few kpc ofNGC 253. The slit was oriented in a position
angle of 135 degrees, along the minor axis of this edge-on galaxy. Note that each
of the three lines has a double-peaked profile shape throughout a region about 600
pc in extent to the SE of the nucleus. This line-splitting has a magnitude of about
300 to 400 km S-I, and suggests either an expanding bubble or outflow along the
surface of a hollow cone-like structure (cf. HAM) . This region of expanding gas also
corresponds to the bright X-ray plume seen in Fig. 4 (cf. Fabbiano & Trinchieri
1984; McCarthy, Heckman, & van Breugel 1987).

(Marlowe et a1. 1993). Because the potential wells of dwarfs are so shallow,
there is a broad theoretical consensus that star burst-driven mass loss plays
a crucial role in the evolution of such systems (e.g., Silk, Wyse, & Shields
1987). We will discuss related H I observations of several dwarf galaxies in
§3.4 below.
As a closing comment on the kinematics of the optical emission-line
gas, we note that the outflow velocities as deduced from all the evidence
summarized above are quite modest (typically < several hundred km S-l).
These velocities are much lower than the predicted outflow speed for the
hot, tenuous superwind fluid (a few thousand km s-l - see §2), but are
quite consistent with our expectations for ambient gas that is being 3hocked
478

and accelerated outward by the wind (see the discussion in §§2.2 and 2.4
above). This may be seen by considering Equations 10, 16, and 18 above,
or by noting that strong optical line emission will only be produced by
material into which radiative shocks are being driven (e.g., shocks in which
the radiative cooling time behind the shock is short compared to the shock
dynamical timescale). As reviewed by McKee & Hollenbach (1980), radiative
shocks in the ISM generally have shock speeds less than a few hundred km
S-l. Since we expect the velocity to which a shocked cloud is accelerated
by the superwind to be similar to the speed of the shock driven into it, the
observed outflow velocities therefore represent additional indirect evidence
for the schematic picture we have developed above.

3.2.3. Emission-Line Ratios and the Ionization Source

The relative emission-line intensities in the nebulae associated with super-
winds provide indirect evidence for a high-speed outflow. Key emission-line
ratios like [S IIjAA6716,6731/Ha, [N IIjA6584/Ha, and [0 IjA6300/Ha have
values that are more similar to those seen in gas that has been shock-
heated (e.g., old supernova remnants and Herbig-Haro objects - cf. Hartigan,
Raymond, & Hartmann 1987; Fesen, Blair, & Kirshner 1985) than to what is
observed in H II regions in spiral and irregular galaxies (e.g., McCall, Rybski,
& Shields 1985). We (L92; LH) find that all these ratios tend to increase
with increasing radius along the minor axis in edge-on star burst galaxies
(see also AHM89 and HAM) and are also systematically larger along the
minor axis than along the major axis.
The simplest picture is that the heating and ionization of the gas in
the central starburst and in the large-scale disk of the galaxy is dominated
by photoionization by hot stars. In contrast, shocking of ambient gas by
the superwind probably makes a significant contribution to heating the gas
located out in the galaxy halo and seen most clearly along the minor axis.
The transfer of energy from the hot superwind gas to imbedded cool, dense
material via the 'mixing layer' phenomenon (Begelman & Fabian 1990;
Slavin, Shull, & Begelman 1993) may also contribute to the heating and
ionization of the optical emission-line gas. It is also possible that the halo
gas is photoionized by an extremely dilute radiation field 'leaking' out of the
starburst (e.g., Sokolowski & Bland-Hawthorn 1993; see also Dettmar 1993
and references therein).
Perhaps the strongest indirect evidence for mechanical heating of the
gas by the superwind is the strong correlation found by HAM, L92, and
LH between the dynamical state of the gas (as measured by the emission-
line widths) and the thermal/ionization state of the gas (as probed by
the strengths of the [0 Ij, [S IIj, and [N IIj lines relative to Ha). This
strongly suggests that the same mechanism is responsible for both heating
 479

and accelerating the gas. The most decisive way of testing whether the gas
is collisionally ionized or photoionized would be to determine whether the
electron temperature in the 0 III zone is higher than can be explained by
photoionization (e.g., T > 20,000 K - cf. Osterbrock 1989). HAM attempted
such measurements using the ratio of the [0 IIIjA4363 and A5007 lines, but
the results were inconclusive (in some cases shock-heating was favored and
in other cases photoionization).

3.2.4. Densities and Pressures as Diagnostics

One of the most useful diagnostics of superwinds is to use the ratio of the
[5 IIjA6716 and A6731 lines to determine the radial run of density within the
superwind nebulae. In many cases, the average temperature of the nebula
can also be deduced from the ratio of the [N IIjA5755 and A6584 lines, so
that the radial run of pressure can be determined. According to superwind
models, the central pressure in the star burst is set by the starburst radius
and the rate at which momentum is injected by supernovae and stellar winds
(cf. CC; HAM; see Equation 4 above). HAM have used the measured central
pressures and the predicted momentum injection rates to estimate starburst
radii for 12 IR-selected galaxies. These radii agree surprisingly well (typically
within 0.1 dex) with radii measured directly from radio, IR, and optical
imaging of the starbursts (see also L92; LH).
The superwind model also predicts that once the wind 'blows out' of the
galactic disk and is in free expansion, the pressure within the wind is dom-
inated by ram pressu,re. The ram pressure will drop like the inverse square
of the radius and will be proportional to the rate at which the starburst
injects momentum into the wind (see Equation 5 above). HAM have used
the measured pressure profiles in 12 IR-selected galaxies to estimate the
momentum flux of the wind, and find excellent agreement between these
estimates and the predicted rate of momentum input from the starburst.
This analysis has been extended to an additional set of galaxies by L92 and
LH, and the good agreement between theory and observation verified over
a range of nearly three orders-of-magnitude in starburst luminosity.
It is worth emphasizing that an ultraluminous starburst/superwind can
produce pressures over galactic-scale volumes that are three to four orders-
of-magnitude higher than the general pressure in the ISM in the Milky Way:
e.g., P / k > few X 106 K cm -3 (see HAM, L92, and LH for details). The radius
of the over-pressured region (as measured out to some fixed isobar) should
scale as the square-root of the star-formation rate, and this dependence
has been confirmed by HAM, L92, and LH. HAM and LRD have further
shown that the P ~ V work represented by the measured pressure profiles
in the halos of starbursts with superwinds is in satisfactory agreement with
the time-integrated kinetic energy input predicted from the starburst (cf.
480

Equation 1 above).
The high measured pressures and the systematic dependence of the
radial pressure profile upon the starburst luminosity clearly establish that
starbursts are able to regulate the gas pressure over very large volumes (at
least one hundred times the volume of the starburst itself). But must this
regulation be due to a wind? Might it not be possible for the measured
pressure profiles to reflect the static thermal pressure in a hot volume-filling
phase that is gravitationally bound to the the starburst galaxy?
In fact, it is rather easy to rule out the above idea for those starbursts in
HAM that have X-ray data available. Since the X-ray luminosity depends
on density squared, and since the density (at fixed pressure) is inversely
proportional to the temperature, it follows that the hotter the volume-
filling phase (at fixed pressure), the lower its X-ray luminosity. Specifically,
we calculate that a volume-filling hot phase in pressure-balance with the
optical emission-line clouds would have an X-ray luminosity that exceeds
the measured X-ray luminosity unless it has a temperature in excess of 4
x 10 7 K. The sound-speed in gas at T > 4 X 10 7 K is Cs > 1000 km S-1,
which is in turn much greater than the escape velocity from the starburst
potential well. Thus, any such gas could not be bound to the starburst and
will flow out as a wind (QED!).
The shapes of the measured pressure profiles can also be used to argue
against an AGN as the source of the wind. If, as would be the case with an
AGN, the region of mass/energy injection is small compared to the seeing-
disk (which has a typical radius of 100 pc for the starbursts in HAM and
LH), then the radial pressure profile should follow an inverse square law
all the way in to the center. Such a r- 2 radial density/pressure profile is
directly measured in the classical AGN (type 1 Seyfert nucleus) NGC 4151
by Mediavilla et al. (1992) over a range in radii between about 70 and 400 pc
(e.g., into the approximate radius of the seeing disk). Pronik (1989) has used
the observed correlations between critical density and line width for many
different forbidden lines in the nucleus to conclude that the density in NGC
4151 continues to follow a r- 2 law all the way in to sub-pc scales. In contrast,
the observed radial pressure profiles in most IR-bright galaxies show flat
central zones (within the starburst, where mass and energy are injected) and
then steepen to approximately r- 2 at large radii (in reasonable agreement
with theoretical models - cf. CC; HAM; LH). This behavior can be clearly
seen in Figure 8 for the two galaxies in HAM for which the measured radial
pressure profiles are of the highest quality (M 82 and NGC 3256). The shapes
of the radial pressure profiles are remarkably similar in the two galaxies, and
the larger radial scale associated with the 8-times-more-luminous starburst
in NGC 3256 is exactly as predicted by the superwind model. Note that
both profiles are relatively flat in the central region (within the starburst
radius denoted r * in the plot), and then steepen to an r -2 dependence at
 481

log RN3256 (pc)

1.5 2.0 2.5 3.0 3.5

R
3.0 r·'------~O~ _ _=~
• R
,fCC
-8.5

C')

'E 2.5
~
Q) R -9.0
Z
01
.2
2.0
-9.5
o
r.
I-M82 "SUbble'-1

1.0 1.5 2.0 2.5 3.0

log RM82 (pc)

Fig. 8. A log-log plot of the electron density (left axis) and gas pressure (right axis)
vs. radius for two well-studied starburst galaxies (M 82 - solid dots; NGC 3256 -
hollow dots). Note that the radial scale for NGC 3256 (upper axis) has been shifted
by 0.45 dex (factor 2.8) with respect to that for M 82 (lower axis). The radial
pressure profile predicted for both galaxies by the simple spherically-symmetric
wind model of CC is indicated by the solid curve. We have also indicated estimates
of the pressures in the X-ray and radio plasma in M 82. See HAM for further details.

large radii.

3.2.5. The Role of Scattering

Finally, in closing our discussion of optical line emission, it must be empha-
sized that imaging polarimetry of the famous M 82 nebula shows that Ha
emission from a portion of the starburst nucleus is being scattered into our
line-of-sight by dust grains in the halo of M 82, and that this scattered light
makes a significant contribution to the observed Ha emission in the halo
(Scarrott, Eaton, & Axon 1991 and references therein). On the other hand,
the systematic spatial variations in the optical emission-line ratios measured
in the inner portion of the minor-axis nebula (that is, the gas located within
about 30 degrees of the minor axis and within about a kpc from the nucleus)
482

show that the majority of the Ha emission from this region must be intrinsic
to the filament system rather than scattered light (e.g., McCarthy, Heckman,
& van Breugel 1987; Bland & Tully 1988; HAM) . Indeed, the polarization
of the Ha within this region is typically only 4 to 10%, compared to values
of 10 to 30% at greater angles off the minor axis (Scarrott, Eaton, & Axon
1991). It is also this inner region with lower polarization in which the clear
kinematic signature of an outflow is seen (HAM; Bland & Tully 1988). We
suggest that scattered light contributes no more than about 30% of the Ha
flux in this inner region. A spectropolarimetric, multi-emission-line mapping
of the M 82 nebula with a spectral resolution of an Angstrom or better would
clearly be a magnificent Ph.D . thesis project!

3.3. THE RELATIVISTIC PLASMA ASSOCIATED WITH SUPERWINDS

Several star burst galaxies have been discovered to have large-scale nonther-
mal radio halos. These presumably arise as relativistic magnetized plasma
is convected by the superwind out of the central starburst, where it was
created by supernovae (e.g., Lerche & Schlickeiser 1982).
The galaXy M 82, studied most recently by Seaquist & Odegard (1991),
has a synchrotron-emitting halo extending out to a radius of at least 8
kpc. The spectrum of the radio emission steepens markedly with increasi~g
radius, which Seaquist & Odegard attribute to the more rapid cooling of
the most energetic electrons by inverse Compton scattering of the starburst
IR photons. They then derive an outflow speed for the relativistic plasma
of about 2000 km s-1 (consistent with the expected outflow speed of the
tenuous superwind fluid - cf. CC and the discussion in §2 above). They also
use the radio data to argue that inverse Compton radiation contributes
a negligible fraction of the X-ray emission from M 82 (see §3.1 above),
and they estimate that relativistic particles comprise only a few percent
of the energy content of the wind. They note that this ratio is similar to
that observed in supernova remnants, as expected in the superwind model.
Reuter et al. (1992) have obtained higher resolution maps of the M 82 halo,
and find a considerable amount of structure (radially-oriented filaments and
gaps). They interpret this as clear'evidence that magnetized plasma is being
convected out of the M 82 starburst, creating a bipolar (poloidal) magnetic
field geometry.
Hummel & Dettmar (1990) have studied the nonthermal radio halo of
the edge-on actively-star-forming galaxy NGC 4631, and have argued that
the radial spectral index gradient they measure for the radio emission is
consistent with models in which relativistic plasma is convected out of the
galaxy by a wind, and is then cooled by synchrotron, inverse Compton,
and/or adiabatic-expansion losses. As in M 82, the magnetic field geometry
in the halo ofNGC 4631 is inferred to be poloidal (field lines directed radially
 48 3

outward). Hummel & Dettmar suggest this is due to the operation of a strong
galactic wind in this galaxy (though a dynamo model for the field can not
be ruled out).
Most recently, Carilli et al. (1992) have discovered a radio halo around
the edge-on starburst galaxy NGC 253 which extends out to at least 9 kpc.
As in the case of M 82 and NGC 4631, the radio spectrum steepens with
increasing distance from the disk/starburst, and Carilli et al. interpret this
as evidence for an outflow from the galactic disk into the halo. The magnetic
field geometry in this case appears more nearly toroidal (field lines parallel to
the galactic disk), except along a bright radio spur that attaches to the inner
disk of the galaxy. Dettmar (1993) has also discussed the radio halo of the
IR-bright edge-on disk galaxy NGC 5775 (Hummel 1991), and emphasized
the close connection between the radio halo and the system of faint Ha
filaments extending many kpc out of the galactic disk.
There are also less complete data indicating the presence of thick ra-
dio disks and/or radio halos in the edge-on starburst systems NGC 1808
(Dahlem et al. 1990), NGC 3628 (Reuter et al. 1991), and NGC 4945
(Harnett et al. 1989). The smaller (few-kpc-scale) radio sources extending
out along the minor axes of some edge-on radio-bright disk galaxies (e.g.,
Duric et al. 1983; Hummel, van Gorkom, & Kotanyi 1983) may also be
related phenomena.

3.4. THE COOL GAS ASSOCIATED WITH SUPERWINDS

Studies of the interaction of superwinds with cool atomic and molecular

gas are potentially very important, since such gas is expected to comprise
the major fraction of the mass of the ISM inside starburst galaxies or in
their surrounding halos. One of the most important insights provided by
such studies is that star burst galaxies are often immersed in large (trans-
galactic-scale) HI envelopes or halos, presumably of tidal origin (e.g., Yun
& Ho 1992; Hibbard et al. 1993; Fisher & Tully 1977; Haynes et al. 1979;
Stanford 1990; Stanford & Wood 1989; Gottesman & Mahon 1990). This is
of crucial significance, because the interaction between the wind and this
comparatively dense halo gas probably accounts for the relative brightness
of superwinds in the optical and X-ray domain. Were such gas not present -
Le., if superwinds instead flowed into a very low density intergalactic medium
- they would quickly adiabatically cool, their density would rapidly drop,
and they would then become undetectably faint sources of either X-rays or
optical line emission (cf. Mathews & Doane 1993). Thus, it is likely that the
interaction of the superwind with the H I envelope/halo is fundamental in
determining what we observe at optical and X-ray wavelengths.
Studies of the cool gas in the environs of starbursts have also provided
direct evidence for outflows. Indeed, possibly the single most convincing
484

piece of evidence to date for an outflow from a star burst galaxy is the recent
analysis of the kinematics of the Na I D doublet seen in the well-known
starburst galaxy NGC 1808 (Phillips 1992). He has studied the kinematics
of the system of radial dust-filaments that extend several kpc out along the
minor axis ofthis galaxy. The Na D doublet is seen in absorption on the near
side of the outflow, and has a strongly blueshifted component (by up to 400
km s - 1 with respect to the systemic velocity). Since the kinematics al"e being
probed with absorption-lines in this-,Case, there is no possible ambiguity in
the sign of the observed (outward) radial motions. Assuming a standard gas-
to-dust ratio in the outbound dust filaments, the implied mass ofthe ejected
material is about 6 X 10 7 M 0 . The kinetic energy of this material is then
about 10 56 ergs, or about 20% of the time-integrated kinetic energy of the
starburst in NGC 1808 over a period of 10 7 yr (the dynamical lifetime of the
dust filament system). This is in good agreement with the prediction that
the kinetic energy in the shell of an energy-conserving wind-blown bubble
should be about 20% of the total time-integrated amount of injected kinetic
energy (cf. Dyson 1989).
There are also several pieces of evidence for outflowing molecular gas .in
starbursts. High-resolution 115GHz CO observations of M 82 by Nakai et
al. (1987) suggest that molecular gas is flowing out of the central starburst,
and Irwin & Sofue (1992) have reached similar conclusions for the CO in the
center of NGC 3079. Turner (1985) has found a molecular outflow traced
by the OH emission lines extending out from the nucleus to about 1.5 kpc
along the minor axis of NGC 253, and van del' Werf et al. (1993a) have
fotmd plumes of near-IR-emitting molecular hydrogen at the base of this
apparent outflow. In the most detailed near-IR study to-date of the famous
IR-luminous galaxy NGC 6240, van del' Werf et al. (1993b) have uncovered
convincing kinematic and morphological evidence that the relatively warm
molecular gas (as probed with the near-IR vibration-rotation lines of H 2 )
in t.his system is being ejected from the star burst at velocities of several
hundred to 1000 km S-1 (in agreement with the kinematics of the ionized
gas discussed by HAM).
Radio observations of H I in several starburst galaxies also reveal the
impact a superwind can have on the cool ISM. Yun, Ho, & Lo (1992) have
suggested that the wind from the star~urst in M 82 has carved a channel
through the inner part of that galaxy's H I envelope. Koribalski et a1. (1992)
have found clear kinematic evidence for H I flowing out along the minor axis
of the starburst galaxy NGC 1808 at a velocity of at least 150 km S-1. This
is the same outflow studied optically at much higher spatial resolution via
the Na I D line by Phillips (1992 - see above).
Recent VLA H I maps of several star-forming dwarf galaxies have revealed
the presence of kpc-scale H I shells expanding at about 10 km s-1 (e.g.,
Puche et a1. 1992; Westpfahl & Puche 1992). These authors argue that
 485

such expanding, hollow structures are the result of the collective effect
of multiple stellar winds and supernovae. Moreover, since the sizes and
expansion speeds of the bubbles are comparable to the total sizes and
internal velocity dispersions in the smallest dwarf galaxies, these structures
probably playa major role in the evolution of the ISM and of star-formation
in such galaxies (e.g., Westpfahl & Puche 1992).
Perhaps the most spectacular phenomenon ascribed to a superwind is the
ram pressure stripping of the H lout of the outer parts of the galaxy NGC
3073 by its starbul'sting, superwind-driving neighbor NGC 3079 (Irwin et a1.
1987). The ram pressure that Irwin et a1. estimate as being required to strip
the H I 'from NGC 3073 implies a superwind momentum flux that agrees
well with that estimated for NGC 3079 by HAM via three independent
techniques (from the kinematics of the expanding bubble, from the radial
pressure profile, and from the starburst luminosity using Equation 3 above).

4. A Local Census of Superwinds: Their Global Significance

HAM presented preliminary estimates of the rates at which star bursts inject
mass, metals, and energy into superwinds at the present epoch. These
estimates were based on the following assumptions (see HAM for details):
First, all galaxies in the local universe with large IR luminosities (L I R > 10 44
erg S-I), large IR excesses (LIR/ LOPT > 2), and/or warm far-IR colors
(flux density at 60 I'm greater than 50% of the flux density at 100 I'm)
drive superwinds. Second, The superwind outflow rates are proportional to
the far-IR luminosity, with the constants of proportionality determined from
well-studied examples like M 82 and NGC 253.
The recent survey of optical line emission associated with a well-defined
sample of nearly 50 IR-selected edge-on starbursting disk galaxies by L92
and LH have put these two assumptions on a much firmer observational
foundation. They find that such indicators of a superwind as excess optical
line-emission along the minor axis, broad emission-line widths and velocity
shears along the minor axis, and 'shock-type' optical emission-line ratios
(e.g., strong [0 I] and [S II] emission relative to Ha) along the minor axis
all correlate with such quantities as the IR luminosity, the IR/opticallumi-
nosity ratio, and the IR color-temperature. They conclude that superwinds
are probably ubiquitous in galaxies meeting the IR criteria listed in the
preceding paragraph.
Integrating over the local far-IR luminosity function for galaxies meeting
the above criteria (Bothun, Lonsdale, & Rice 1989) then yields the superwind
injection rates per unit volume in the local universe. The resulting injection
rates per unit volume can be normalized by multiplying them by the age
of the universe, and then dividing the result by the local space density of
galaxies. This calculation then implies that superwinds have carried out
486

approximately 5 X 108 M0 in metals and 10 59 ergs in kinetic plus thermal

energy per average (Schechter L·) galaxy over the history of the universe.
For reference, note that these two quantities are approximately equal to
the mass of metals contained inside an average galaxy and the gravitational
binding energy of an average galaxy respectively.
These values pertain to the conservative assumption that the superwind
rate has not evolved with look-back time. Since the star-formation rate was
presumably higher in the early universe, the true time-integrated values are
likely to be about an order-of-magnitude higher (cf. Ostriker & Cowie 1981;
HAM). On the other hand, it is unclear what fraction of the energy and
mass injected into a superwind by a starburst ultimately escapes the galaxy
and is delivered to the intergalactic medium (e.g., less energetic or more
strongly confined superwinds may conceivably cool and turn into galactic
fountains (cf. Norman & Ikeuchi 1989). In any case, it seems quite likely
that superwinds do play an important role in enriching and heating the
intergalactic medium (as we will discuss more fully below).

5. IlDplications of Superwinds
5.1. HEATING & ENRICHING THE IGM
The ejection of highly metal-enriched material by super winds offers a very
natural explanation for the long-standing puzzle that the intra-cluster medium
in clusters of galaxies contains at least as much mass in the form of metals as
do the galaxies in the cluster (e.g., Sarazin 1986). This idea has been explored
by many authors, including Larson & Dinerstein (1975), DeYoung (1978),
and (more recently) White (1991) and David, Forman, & Jones (1991) .
Superwinds can heat the ICM as well as enrich it. The combination of
gravitational and superwind heating of the ICM could readily account for the
result that the temperature of the ICM seems to exceed the 'temperature'
of the galaxies as they move within the cluster potential. That is, the
parameter 13 defined as j.LmHu;a,/kTICM may be systematically less than
unity in clusters (cf. David, Forman, & Jones 1991; White 1991). A good
quantitative match of White's model to the data on superwinds and the
ICM was found by HAM.
The idea that the ICM has been enriched and heated by superwinds
makes several specific and testable predictions (cf. David, Forman, & Jones
1991; White 1991; and Donahue's contribution to this volume):
.13should be correlated with the cluster's velocity dispersion (since
the relative effect of superwind heating will be greatest in the clusters
with low velocity dispersions) .
• The 0 jFe ratio in the ICM should be significantly above the solar
value (since this is true for type II supernova ejecta).
 487

.The 0 abundance in the ICM should be inversely correlated with

f3 and with the ratio of ICM gas mass to galaxy mass .
• f3 should be positively correlated with the ratio of ICM gas mass to
galaxy mass .
• The ratio of ICM gas mass to galaxy mass should be larger in
clusters with deeper potential wells (since more of the IeM can be
retained in the face of superwind heating in clusters with large escape
velocities ).
While direct evidence for a hot, all-pervasive intergalactic medium re-
mains scant, there are compelling theoretical reasons to believe it exists (cf.
the recent review by Shapiro 1989). In order that a cosmologically significant
IGM not violate Gunn-Peterson limits, it must be ionized at quite early
epochs (e.g., already by the time at which the first known quasars appear).
The mechanical energy input from superwinds driven by protogalaxies may
be capable of heating and collisionally ionizing the IGM in the early 'pre-
quasar' era of the universe (cf. Miralda-Escude & Ostriker 1991).

5.2. GALAXY FORMATION & EVOLUTION

The most remarkable and profound constraints on our ideas of how galaxies
actually form are the relatively tight relationships between such disparate
quantities as velocity dispersion, mass and luminosity, length-scale, sur-
face brightness, and chemical abundances (e.g., Djorgovski & Davis 1987;
Dressler et al. 1987; Franx & Illingworth 1990). Recent attempts to under-
stand the processes that regulated the conversion of gas into stars and the
subseque~t chemical enrichment in proto-galaxies have led to the concept
of 'self-regulated galaxy formation' or 'feedback' (e.g., Silk 1987; White &
Frenk 1991): the effect of the kinetic energy supplied by massive stars and
supernovae on the gas in a (proto )galaxy may lead naturally to the creation
of the specific properties of galaxies we see today.
For example, Dekel & Silk (1985), Vader (1986), and Silk, Wyse, &
Shields (1987) have argued that the striking differences between many of
the global properties of dwarf and massive galaxies may reflect the existence
of a critical escape velocity (potential well depth) below which galaxies
suffer catastrophic superwind-driven loss of their ISM following the initial
burst of star-formation that accompanies their formation. This process leads
naturally to the low surface brightnesses and very low metal abundances
of dwarf galaxies (see also DeYoung & Gallagher 1990). The catastrophic
destruction of dwarfs by starburst-driven winds might be related to the
apparently rapid disappearance of the 'faint, blue galaxies' (starbursting
low mass systems) during the last few Gyr of the history of the Universe (cf.
Broadhurst, Ellis, & Glazebrook 1992; Cowie et a1. 1992).
488

Wyse & Silk (1985) have examined the consequences of supernova heating
for the chemical enrichment of disk galaxies. Similarly, Larson & Dinerstein
(1975), Vader (1986), Franx & Dlingworth (1990), Lynden-Bell (1992) and
others have proposed that superwinds are the mechanism that underlies the
correlation between the metal abundance of the stellar population and the
local escape velocity in elliptical galaxies (Le., both the mass vs. metallicity
relation between galaxies and the metallicity vs. radius relation within
galaxies). That is, the deeper the local gravitational potential, the more
difficult to drive a wind and hence the greater the ability of that locale to
retain metal-rich processed material.
Mathews (1989) has explored the consequences of superwinds in normal
ellipticals. He argues that the observed mass-to-light ratio of the stellar
population is set by superwind-driven mass loss. Superwinds might also
play an important role in the evolution of the star-bursting product of the
merger of two major disk galaxies into a giant elliptical (see Hernquist's
contribution to this volume): superwinds could shock-heat the cool ISM up
to X-ray temperatures or even eject much of the ISM from the galaxy (e.g.,
Mathews & Baker 1971).
It has even been suggested that the superwinds ('explosions') associated
with the formation of galaxies may have played some role in the formation
of structure in the universe (Ostriker & Cowie 1981). They propose that
superwinds swept up the IGM into large shells of dense radiatively-cooled
material that then fragmented into Jeans-unstable hunps that became new
galaxies. If the process of amplification via 'stimulated galaxy formation' is
efficient enough, then superwinds may have helped induce structure in the
early universe.

5.3. QSO ABSORPTION LINES

The nature of the intervening clouds responsible for producing the absorption-
lines in the spectra of QSO's remains unclear (cf. the collection of reviews
in the volume compiled by Blades, Turnshek, & Norman 1988 - see also
the reviews by Steidel, Lanzetta, Bechtold, Weymann, and Bajtlik in this
volume). The sharp metal lines and Lyman-limit systems are believed to
arise in the gaseous halos of intervening galaxies, and the implied cross-
sectional area of absorbing material per galaxy significantly exceeds the
traditional 'optical' dimensions of galaxies (even if every galaxy in the
universe has such an extended gaseous halo).
While it is by no means obvious that superwinds are consistent with the
myriad properties observed for the various classes of QSO absorption-line
systems, at the very least, they offer a novel way to transport metals out
into a galactic halo (or even out into the general IGM), and could perhaps
help us understand some of the most salient properties of the absorption-line
 489

systems:
• This transport process could.s.ignificantly increase the effective
absorbing cross-sectional area available per galaxy.
• An evolution in the starburst rate (and hence superwind rate) with
cosmic time could help explain the observed evolution with redshift
of the sharp metal line and Lyman-limit systems.
• Superwinds might be responsible for some of the complex observed
velocity structure of the metal-line systems: the absorption-line pro-
files typically consist of several narrow subcomponents spanning a
velocity range of several tens to several hundred km s-l. In extreme
cases, velocity spreads of up to 2000 km S-1 are observed. Such
substructures and velocities are a natural consequence of a line-of-
sight that traverses a superwind.
• If such broad, multi-component absorption-lines originate in an
organized outflow (rather than being due to turbulent motions of
clouds in a galactic halo), it would explain why the halos are not de-
stroyed in a fraction of a halo-crossing time by cloud-cloud collisions.

• The apparent lack of gaseous halos associated with low-mass galax-

ies noted in this volume by Steidel might be due to the preferential
destruction/ejection of these loosely-bound halos by superwinds.
One obvious test of these ideas is to obtain far-UV spectra of QSO's
viewed through the halos of known starburst / superwind galaxies, and then
to compare these spectra to spectra of typical high-reshift QSO sharp metal-
line systems. Some of us have recently embarked on an HST program to do
that for a sample of QSOs seen behind the halos of the nearby starburst
galaxies NGC 253, NGC 520, and NGC 3079.

5.4. THE X-RAY BACKGROUND

Another enduring mystery concerns the origin of the cosmic X-ray back-
ground (CXRB). Bookbinder et al. (1980) first suggested that X-ray emission
from the hot thermal plasma in superwinds might make an important
contribution to the CXRB. More recently, Weedman (1987) and Griffiths
& Padovani (1990) have shown that the X-ray luminosity function of local
star burst galaxies could provide a significant fraction of the CXRB if they
evolve by only a modest amount with cosmic epoch.
It is worth emphasising however that even if superwinds do make a
non-negligible contribution to the CXRB, this will occur only at relatively
low energies. More specifically, superwinds are not expected to radiate
significantly above about 10 keY (see §2 above), while most of the overall
490

energy in the CXRB is in the range 10-40 ke V. If the superwinds contributing

to the CXRB are located at a significant mean redshift this will only
exacerbate this spectral deficiency at high energies.

5.5. THE DISK-HALO CONNECTION

One of the principal themes in the study of the interstellar medium (ISM) is
the relationship between massive stars and the thermal and dynamical state
of the ISM. Recently, there has been an explosive growth in the number
of papers on this topic, spurred in part by the discovery of expanding HI
'supershells' (Heiles 1990) and of a warm, ionized medium which has a
relatively large scale-height and contains a substantial fraction of the Milky
Way's total gas mass (Reynolds 1991). As reviewed by Dettmar (1993), there
also mounting evidence of a 'disk-halo connection' in external galaxies by
which star-formation in the disk ISM can propel mass and thermal/kinetic
energy up into the galactic halo.
Starbursts offer wonderful laboratories for studying these processes, with
the advantage that conditions are so extreme in these systems that the
number of relevant physical processes may be fewer than in a normal disk-
halo system. Even if it does not prove to be easier to understand the disk-
halo connection in starbursts, the spectacular superwind phenomena we have
described above demonstrate that is clearly easier to observe the disk-halo
connection there!
The ultimate goal of course will be to develop a comprehensive theory
of the feedback effect of the formation of massive stars that explains a
whole continuum of disk-halo systems from normal star-forming disks with
their galactic fountains or chimneys to powerful starbursts with global
superwinds. Some important steps in this direction have already been taken
(cf. Norman & Ikeuchi 1989).

5.6. THE STARBURST-AGN CONNECTION & ULTRALUMINOUS GALAXIES

One of the most fascinating issues in extragalactic astronomy concerns the

relationship between AGNs and starbursts (cf. Heckman 1991; Blandford
1992; Filippenko 1992). Roberto Terlevich and his colleagues have argued
that ultra-compact starbursts occurring in conditions of very high gas den-
sity can reproduce all the phenomena associated with classical radio-quiet
AGNs (e.g., Terlevich 1992). Our contention here is more limited in scope,
but similar in spirit: superwinds driven by starbursts can significantly blur
the phenomenological distinction between starbursts and AGNs.
First, superwinds are estimated to have terminal velocities of several
thousand km s-1, and so (at least in principle) can accelerate gas up to
velocities well in excess of normal galaxian orbital velocities. Thus, the mere
 491

presence of broad emission-lines doesnot necessarily imply that a galaxy

harbors an AGN.
Second, ambient material that has been shock-heated by the superwind
will produce LINER-type line ratios (which again are often taken as firm
evidence for the presence of an AGN). This classification problem is exacer-
bated by the much heavier obscuration suffered by the optical line emission
produced within the central starburst compared to the emission originating
in the superwind (thereby making it harder to clearly recognize the signature
of the OB-star-photoionized gas in the integrated optical spectra of distant
and powerful starbursts).
Third, superwinds are potential sources of X-rays (at least up to 10
keY or so) . Now X-rays comprise an energetically trivial fraction of the
bolometric luminosity (typically few times 10- 4 ) in IR-selected starburst
galaxies (e.g., Griffiths & Padovani 1990; Green, Anderson, & Ward 1992).
Since the fractional contribution of the X-ray luminosity is several orders-of-
magnitude larger than this in type 1 Seyfert nuclei and QSOs (e.g., Sanders
et a1. 1989), there is no difficulty in using X-ray data to differentiate between
these AGN classes and starbursts (cf. Rieke 1988). On the other hand, there
is a considerable amount of overlap in the distribution of Lx /Lbol) between
starbursts and type 2 Seyfert galaxies (e.g., Green, Anderson, & Ward 1992).
Indeed, Wilson et al. (1992) have suggested that much of the hard X-ray
emission from the type 2 Seyfert NGC 1068 is associated with the circum-
nuclear starburst rather than with the AGN itself.
Finally, superwinds are also capable of producing large-scale radio sources
that extend at least 10 kpc out into the halo of the starburst galaxy. Such
radio sources are very weak (the total core-plus-halo radio luminosity of M 82
is only about 10- 5 of the bolometric luminosity of the starburst), and show
only a weakly bipolar morphology. Thus, we suggest that spatially-extended
(> kpc-scale) radio emission should be taken as unambiguous evidence for the
presence of an AGN only if the radio source provides a much larger fraction
than this of the bolometric luminosity and/or exhibits a highly-collimated,
jet-type morphology.
Taking these points together, and noting that the observational mani-
festations of superwinds become more spectacular with increasing starburst
luminosity, we suggest that the evidence that most 'ultraluminous' IRAS
galaxies are powered by buried quasars (e.g., Sanders et a1. 1988) is not
entirely persuasive. Indeed - based especially on the near-IR colors and the
IR spectral energy distributions of the ultraluminous galaxies (see Sanders et
al1988 figures 14 & 15) - we suggest that only three of the ten ultraluminous
galaxies (namely IRAS 05189-2524,IRAS 08572+3915, and Mrk 231) clearly
have ultraluminous AGNs in their cores (see also Majewski et al. 1993).
492

6. Summary

We have reviewed the theoretical basis for, the observational evidence per-
taining to, and the astrophysical implications of star burst-driven galactic
superwinds: global outflows driven by the collective effect of the mechanical
energy supplied by supernovae and stellar winds in a nuclear or circum-
nuclear star burst.
Models of the superwind phenomenon imply that a starburst can create
a highly over-pressured bubble of hot gas which will expand preferentially
along the direction of the maximum pressure gradient in the ambient ISM of
the starburst galaxy (Le., probably along the galaxy minor axis). For typical
starburst conditions, it is expected that this expanding bubble will quickly
'blow-out' into the galactic halo or beyond to form a high-speed (several
thousand km s -1) wind.
Hard X-rays (kT = 5 - 10 keY) will be produced by the hot fluid inside
the wind's sonic radius and by internal shocks at larger radii. Much softer
X-rays can arise as the wind shock-heats and/or evaporates dense ambient
gas in the galactic halo. The available X-ray imaging and spectroscopic data
on starburst galaxies are broadly consistent with this simple picture, though
the weakness of the 6.7 ke V Fe K line is a puzzle.
Optical line emission will also be produced as dense halo gas is shock-
heated and/or evaporated by the wind. Optical data on starbursts spanning
nearly three orders-of-magnitude in luminosity provide highly suggestive ev-
idence for the ubiquity of superwinds. There is a marked excess of emission-
line gas along the minor axes of starburst galaxies viewed edge-on. The
gas along the minor axis is much more disturbed kinematically than that
along the major axis, the line widths along the minor axis correlate better
with the starburst luminosity than with the galaxy rotation speed, and the
largest velocity shears along the minor axis are observed in the galaxies
that are more face-on and have larger IR luminosities (all suggesting a
starburst-driven outflow rather than simple orbital motions). In the best-
studied (most-nearby) starbursts, the kinematics suggest an outflow along
the walls of a cone-like structure at velocities of a few hundred km s-l. The
emission-line ratios produced by this gas are consistent with shock-heating,
and the large measured pressures agree well with the predicted ram pressure
of a superwind. In one case, the outflowing gas has been detected in optical
absorption-lines (providing unambiguous evidence for an outflow).
Radio continuum data on a few starburst galaxies imply that relativistic
particles and magnetic fields are probably being convected out in the wind
at velocities of a few thousand km s-l. Molecular-line observations of su-
perwinds have been reported, and the HI morphology of a companion to the
well-known starburst/superwind galaxy NGC 3079 is strongly suggestive of
the ablation of the ISM in this galaxy by the ram pressure of the NGC
 493

3079 superwind. HI observations of starburst galaxies show that many such

systems are surrounded by large 'envelopes' of H I, presumably of tidal
origin. The interaction between the superwind and such relatively dense and
cool material probably plays a fundamental role in producing the observed
optical line emission and X-rays.
The results from our on-going survey ofIR-selected edge-on disk galaxies
imply that superwinds are probably ubiquitous in galaxies with warm far-
IR colors and total IR luminosities greater than ~Lbout 1044 erg S-l. The
local IR luminosity function of galaxies then allows us to estimate the local
injection rate of metals and energy via superwinds. Integrated over a Hubble
time (with no cosmic evolution) and normalized to the present space density
of all galaxies, we find that superwinds may have ejected a mass of metals
comparable to the mass of metals contained inside galaxies and supplied
an average of about 10 59 ergs per galaxy to intergalactic medium. These
figures could easily be an order-of-magnitude larger allowing for evolution
of the superwind rate with cosmic time.
Superwinds may play a key role in heating and chemically enriching
the intra-cluster and intergalactic media (especially during the epoch of
galaxy formation). The 'feedback' between star-formation and the gaseous
component of proto-galaxies (via the superwind phenomenon) may have
helped imprint the basic and specific properties of galaxies we see today.
Superwinds may also have stimulated or suppressed galaxy formation in
the environs of newly-formed galaxies. Superwinds may be implicated in the
QSO absorption-line phenomenon, as they offer a natural way to create a cos-
mically evolving population of large, metal-enriched, kinematically-complex
gaseous halos. Superwinds also represent an ideal laboratory to study the
'disk-halo connection' under relatively extreme conditions. Superwinds prob-
ably contribute non-negligibly to the X-ray background at E < 10 ke V.
Finally, the superwind phenomenon blurs the empirical distinction between
AGNs and starbursts: a starburst with a strong superwind can exhibit broad
emission-lines, LINER emission-line ratios, (weak) X-ray emission, and even
(weak) large-scale, (weakly) bipolar nonthermal radio emission (all of which
are commonly taken to indicate the presence of an AGN).

References
Armus, L., Heclanan, T.M., & Miley, G.K.: 1989 (AHM89), ApJ 347, 727
Armus, L., Heclanan, T .M., & Miley, G.K.: 1990 (AHM90), ApJ 364, 471
Armus, L ., Heclanan, T.M., & Weaver, K.: 1993, in preparation
Balsara, D., Suchkov, A., & Heclanan, T.: 1993, in preparation
Beck, S.: 1993, in "The Nearest Active Galaxies", ed(s). , J. Beckman, (Madrid: CSIC), in
press
Begelman, M.C., & Fabian, A.C.: 1990, MNRAS 244, 26P
Berman, V. , & Suchkov, A.: 1991, ApfJSS 184, 169
Bernlohr, K.: 1993a,b, AfJA, in press
494

Binette, L., Dopita, M., & Tuohy, 1.: 1985, ApJ 297, 476
Rlarles, C., Turnshek, D., & Norman, c.: 1988, "Quasar Absorption Lines: Probing the
Universe ", (Cambridge University Press: Cambridge)
Bland, J. & Tully, R.B.: 1988, Nature 334,43
Blandford, R.: 1992, in "Relationships Between Active Galactic Nuclei and Starburst
Galaxies", ed(s)., A. V. Filippenko, (San Francisco:ASP), in press
Bloemen, H.: 1991, "IAU Symp. #144: The Inter8tellar Disk-Halo Connection in Galax-
ies ", (Kluwer: Dordrecht)
Bookbinder, J., Cowie, L.L., Krolik, J.H., Ostriker, J.P., & Rees, M.: 1980, ApJ 237, 647
Bothun, G.D., Lonsdale, C.J., & Rice, W.A.: 1989), ApJ 341, 129
Broadhurst, T., Ellis, R. & Glazebrodk, K.: 1992, Nature 355, 55
Carilli, C., Holdaway, M., Ho, P., & De Pree, C.: 1992, ApJL 399, L59
Castor, J., McCray, R., & Weaver, R.: 1975, ApJL 200, LI07
Charlot, S., & Bruzual, G.: 1991, ApJ 367, 126
Chevalier, R.A., & Clegg, A.W.: 1985 (CC), Nature 317, 44
Colina, L., Lipari, S., & Macchetto, F.: 1991, ApJ 379, 113
Cowie, L., Gardner, J., Hu, E., Wainscoat, R., & Hodapp, K.: 1992, preprint
Cowie, L., & McKee, C.: 1977, ApJ 211, 135
Dahlem, M., AaIto, S., Klein, U., Booth, R., Mebold, U., Wielebinski, R., Lesch, H.: 1990,
A&A 240, 237
David, L., Forman, W., & Jones, C.: 1991, ApJ 369, 121
David, L., Jones, C., & Forman, W.: 1992, ApJ 388, 82
DeGioia-Eastwood, K.: 1985, ApJ 288, 175
Dekel, A., & Silk, J.: 1986, ApJ 303, 39
Dettmar, R.-J.: 1993, FundCosPhys, in press
de Vaucouleurs, G., de Vaucouleurs, A., & Pence, W.: 1974, ApJL 194, L119
DeYoung, D.S.: 1978, ApJ 223, 47
DeYoung, D., & Gallagher, J.: 1990, ApJL 356, L15
Djorgovski, S., & Davis, M.: 1987, ApJ 313, 59
Dressler, A., Lynden-Bell, D., Burstein, D., Davies, R., Faber, S., Terlevich, R., & Wegner,
G.: 1987, ApJ 313, 42
Durie, N., Seaquist, E., Crane, P., Bignell, R.C., & Davis, L.: 1983, ApJL 273, Lll
Dyson, J.E.: 1989, in IAU Colloq. 120, ed(s)., G. Tenario-Tagle, M.Moles,. & J. Melnick,
(Springer-Verlag: Berlin), p. 137
Eales, S., & Arnaud, K.: 1988, ApJ 324, 193
Fabbiano, G.: 1988, ApJ 330, 672
Fabbiano, G., & Trinchieri, G.: 1984, ApJ 286, 491
Fabbiano, G., Heckman, T.M., & Keel, W.C.: 1990, ApJ 355, 442
Fesen, R., Blair, W., & Kirshner, R.: 1985, ApJ 292,29
Filippenko, A.: 1993, in in "Physics of Active Galactic Nuclei", ed(s)., S. Wagner & W.
Duschl, (Berlin: Springer-Verlag), in press
Filippenko, A., & Sargent, W.: 1992, AJ 103,28
Fisher, J.R., & Tully, R.B.: 1976, A&A 53, 397
Franx, M., & lllingworth, G.: 1990, ApJL 359, L41
Gottesman, S., & Mahon, M.: 1990, in IAU Colloq. 124, ed(s)., J. Sulentic, W. Keel, and
C. Telesco, NASA Conf. Pub. 3098, p. 209
Green, P., Anderson, S., & Ward, M.: 1992, MNRAS, in press
Griffiths, R., & Padovani, P.: 1990, ApJ 360, 483
Harnett, J., Haynes, R., Klein, U., & Wielebinski, R.: 1989, A&A 216,39
Hartigan, P., Raymond, J., & Hartmann, L.: 1987, ApJ 316, 323
Hartquist, T., & Dyson, J.E.: 1988, Ap&SS 144, 615
Haynes, M., Giovanelli, R., & Roberts, M.: 1979, ApJ 229, 83
Heckman, T.M.: 1991, in "Massive Stars in Starbursts", ed(s)., C. Leitherer, T. Heckman,
C. Norman, (j N. Walborn, (Cambridge University Press:Cambridge), p. 289
Heckman, T.M., Armus, L., & Miley, G.K.: 1987, AJ 93, 276
 495

Heckmall, T., Armus, L., & Miley, G.: 1990 (HAM), ApJS 74, 833
Heckman, T., Lehnert, M., & Armus, L.: 1993, in "The Nearest Active Galaxies", ed(s).,
J. Beckman, (Madrid: CSIC), in press
Heckman, T., & Fabbiano, G.: 1993, in preparation
Heiles, C.: 1990, ApJ 354, 483
Hibbard, & van Gorkom, J.: 1993, in "Evolution of Galaxie! and Their Environment:
Contributed Papers", ed{s)., D. Hollenbach, J.M. Shull, ,lnd H. Throm on, (NASA
Conference Publication Series), in press
Hummel, K.: 1991, Af1A 251, 442
Hummel, K., van Gorkom, J., & Kotanyi, C.: 1983, ApJL 267, L5
Hummel, K., & Dettmar, R.-J.: 1990, Af1A 236, 33
Irwin, J., Seaquist, E., Taylor, A., & Duric, N.: 1987, ApJL 313, L91
Irwin, J., & Sofue, Y.: 1992, ApJL 396, L75
Kim, D.-W., Fabbiano, G., & Trinchieri, G.: 1992, ApJS 80, 531
Korihalski, B., Dahlem, M., Mebold, U., & Brinks, E.: 1993, Af1A, in press
Kornneef, J.: 1993, ApJ, in press
Larson, R.B., & Dinerstein, H.L.: 1975, PASP 87, 911
Lehnert, M.: 1992, Ph. D. thesis, The Johns Hopkins University
Lehnert, M., & Heckman, T.M.: 1993, in preparation
Leitherer, C., Robert, C., & Drissen, L.: 1993 (LRD), ApJ, in press
Lerche, I., & Schlickeiser, R.: 1982, Af1A 239, 1089
Lynden-Bell, D.: 1992, preprint
MacI,ow, M., & McCray, R.: 1988 (MM), ApJ 324, 776
Majewski, S., Hereld, M., Koo, D., Illingworth, G., & Heckman, T.: 1993, ApJ 402, 125
Marlowe, A., Heckman, T., Wyse, R., & Schommer, R.: 1993, in preparation
Mathews, W.: 1989, AJ 97, 42
Mathews, W., & Baker, J.: 1971, ApJ 170, 241
Mathews, W., & Doane, J.: 1993, preprint
McCall, M., Rybski, P., & Shields, G.: 1985, ApJS 57, 1
McCarthy, P., Heckman, T.M., & van Breugel, W.: 1987, AJ 93, 264
McKee, C., & Hollenbach, D.: 1980, AnnRevAA 18, 219
McKee, C., & Ostriker, J.: 1977, ApJ 218, 148
McKee, C., Van Buren, D., & Lazareff, B.: 1984, ApJL 278, L115
Mediavilla, E., Arribas, S., & Rasilla, J.: 1992, ApJ 396, 517
Meurer, G., Freeman, K., Dopita, M., & Cacciari, C.: 1992, AJ 103, 60
Mirabel, I.F., & Sanders, D.B.: 1988, ApJ 335, 104
Miralda-Escude, J., & Ostriker, J.: 1990, ApJ 350,1
Morris, S., Ward, M., Whittle, M., Wilson, A.S., & Tayler, K.: 1985, MNRAS 216, 193
Nakai, N., Hayashi, M., Handa, T., Sofue, Y., Hasegawa, T., & Sasaki, M.: 1987, PASJ
39, 685
Norman, C., & lkeuchi, S.: 1989, ApJ 345, 372
Ohashi, T., Makishima, K., Tsuru, T., Takano, S., Koyama, K., & Stewart, G.: 1990, ApJ
365, 180
Osterbrock, D.: 1989, "A8trophY8ics of Ga8eous Nebulae and Active Galactic Nuclei",
(University Science Books: Mill Valley, CA)
Ostriker, J., & Cowie, L.: 1981, ApJL 243, L127
Petre, R.: 1993, in "The Nearest Active Galaxies", ed(s)., J. Beckman, (Madrid: CSIC),
in press
Phillips, A.C.: 1992, preprint
Phillips, M.M., Turtle, A.J., Edmunds, M.G., & Pagel, B.E.J.: 1983a, MNRAS 203,759
Phillips, M.M., Baldwin, J.A., Atwood, B., & Carswell, R.F.: 1983b, ApJ 274, 558
Pronik, V.: 1989, in "IAU Symp. 134", ed(s)., D. Osterbrock and J. Miller, (Kluwer:
Dordrecht), p. 296
Puche, D., Westpfahl, D., Brinks, E., & Roy, J.-R.: 1992, AJ 103, 1841
Puche, D., & Westpfahl, D.: 1993, preprint
496

Reuter, IL-P., Krause, M., Wielebinski, R., & Lesch, H.: 1991, A&A 248, 12
Reuter, H.-P., Klein, U., Lesche, H., Wielebinski, (t., & Kronberg, P.: 1992, A&A 256, 10
Reynolds, R~,: .1991, in "IAU Symp. #144: The Interstellar Disk-Halo Connection in
Galaxies", ed(s)., H. Bloemen, (Kluwer: Dordrecht), p. 157
Rice, . . .V., Lonsdale, C., Soifer, B., Neugebauer, G., Koplan, E., Lloyd, L., de Jong, T., &
Habing, H.: 1988, ApJS 68, 91
Rieke, G.: 1988, ApJL 331, L5
Rieke, G.H., Lebofsky, M.J., Thompson, R.I., Low, F.J., & Tokunaga, A.T.: 1980, ApJ
238,24
Roy, J.-R., Boulesteix, J., Joncas, J., & Grundseth, B.: 1991, ApJ 367,141
Sanders, D.B., Soifer, B.T., Elias, J.H., Madore, B.F., Matthews, K., Neugebauer, G., &
Scoville, N.Z .: 1988, ApJ 325, 74
Sanders, D., Phinney, E.S., Neugebauer, G., Soifer, B.T., & Matthews, K.: 1989, ApJ 347,
29
Sarazin, C.: 1986, Rev. Mod. Phy& 58, 1
Scarrott, S., Eaton, N., & Axon, D.: 1991, MNRAS 252, 12P
Schaaf, R., Pietsch, P., Biermann, P., Kronberg, P., & Schmutzler, T.: 1989, ApJ 336,
732
Schiano, A.: 1985, ApJ 299, 24
Seaquist, E., & Odegard, N.: 1991, ApJ 369, 320
Shapiro, P.R.: 1989, in "Fourteenth Texas Symposium on Relativistic Astrophysics",
ed(s)., E.J. Fenyve&, Ann. NY Acad., p. 128
Shull, .I.M. & McKee, C.F.: 1979, ApJ 227,122
Silk, J.: 1987, in "Dark Matter in the Universe", ed(s)., J. Kormendy and G. Knapp, (D.
Reidel:Dordrecht), p. 335
Silk, J., Wyse, R., & Shields, G.: 1987, ApJL 322, L59
Slavin, J., Shull, J .. M., & Begelman, M.: 1993, ApJ 407,83
Sokolowski, J., & Bland-Hawthorn, J.: 1993, preprint
Spitzer, L.: 1978, "Phy&ical Proceues in the Interstellar Medium", (Wiley: New York)
Stanford, A.: 1990, ApJ 358, 153
Stanford, A., & Wood, D.: 1989, ApJ 346, 712
Stone, J., & Norman, M.: 1992, ApJL 390, L17
Tenario-Tagle, G., & Bodenheimer, P.: 1988, ARAA 26, 145
Terlevich, R.: 1992, in "Relationships Between Active Galactic Nuclei and Starburst
Galaxies", ed(s)., A. Filippenko, (ASP: San Francisco), p. 133
Tomisaka, K., & Ikeuchi, S.: 1988, ApJ 330, 695
Tsuru, T.: 1992, private communication to R. Petre
Turner, B.: 1985, ApJ 299, 312
Ulrich, M.-H.: 1978, ApJ 219, 424
Vader, P.: 1986, ApJ 305, 669
van der Werf, P., Cameron, M., Genzel, R., Blietz, M., Krabbe, A., Forbes, D., & Ward,
M.: 1993a, in "IAU ColI. #140: Astronomy with mm and sub-mm Interferometry",
ed(s)., M. I&hiquro, (ASP: San Francisco), in press
van der Werf, P., Genzel, R., Krabbe, A., Blietz, M., Lutz, D., Drapatz, S., Ward, M., &
Forbes, D.: 1993b, ApJ, in press
Ward, M.: 1988, MNRAS 231, IP
Watson, M., Stanger, V., & Griffiths, R.: 1984, ApJ 286, 144
Weedman, D.W.: 1987, in "Star Formation in Galaxies", ed(s)., C.J. Lon&dale - Peruon,
NASA Office of Space Sciences and Applications: Washington, D.C., p. 351
Westpfahl, D., & Puche, D.: 1993, preprint
White, R.E. III: 1991, ApJ 367, 69
White, R .L., & Long, K.: 1991, ApJ 373, 543
White, S., & Frenk, C.: 1991, ApJ 379, 52
Wilson, A.S., Baldwin, J., & Ulvestad, J.: 1985, ApJ 291, 627
Wilson, A., Elvis, M., Lawrence, A., & Bland-Hawthorn, J.: 1992, ApJL 391, L75
 497

Woosley, S., & Weaver, T.: 1986, AnnRevAA 24, 205

Wyse, R" & Silk, J.: 1985, ApJL 296, Ll
Yamauchi, S., Kawada, K, Koyama, K., Kunieda, Y., Tawara, Y., &: Hatsukade, I.: 1990,
ApJ 365,532
York, D.G., Dopita, M., Green, R., & Bechtold, J.: 1986, ApJ 311, 610
Yun, M., & Ho, P.: 1993, preprint .
Yun, M., Ho, P., & Lo, K.Y.: 1993, preprint
 THE FIRST STARS

FRED C. ADAMS
Physics Department, University of Michigan, Ann Arbor, MI48109

Abstract. In this review, we outline the current theory for the formation of stars in
present-day molecular clouds and then extrapolate this paradigm to earlier generations of
stars. We discuss the radiative signature of star formation, gravitational instabilities in
circumstellar disks, and the current status of the initial mass function.
Key words: Stars - Formation

1. Introduction

The problem of the formation and nature of the first stars is both inter-
esting in its own right and useful for understanding the overall evolution
of the galaxies in which they live. The problem of the first stars is on a
somewhat different footing than the formation of galaxies or the formation
of large scale structure in the universe in that multiple generations of stars
have formed during the age of the universe. In particular, stars are form-
ing today in nearby (d '"" 150 pc) molecular clouds, which provide a useful
laboratory in which to study the star formation process. Over the course of
the last decade, a fairly successful paradigm of present-day star formation
has been developed. Within this paradigm, the agreement between obser-
vations and theory is quite good, especially for the case of low-mass stars.
This paradigm thus provides the cornerstone of our current understanding
of the star formation process. We review this paradigm in §2. Disk accretion
plays an important role in the formation of stars, but a viable theory of
disk accretion remains lacking. One of the more promising mechanisms for
obtaining disk accretion is through gravitational instabilities; the current
status of this mechanism is reviewed in §3.
The fundamental unresolved issue of star formation theory is the question
of the initial mass function. A theory of the initial mass function is essential
for understanding the effects of star formation on galactic evolution, galaxy
formation, and other key issues addressed in this volume. We discuss the
initial mass function in §4.
Although we now have a reasonable understanding of star formation in
the present epoch, we are left with the challenge of extrapolating this picture
back to the first generation of stars. Although some work along these lines
has been done, particularly for star formation with low metallicity, much
work remains. We review the current status of work on primordial star for-
mation in §5 and speculate about the primordial initial mass function. We
conclude with a summary and a discussion of unresolved issues in §6.

499
J. M. Shull and H. A. Thronson, Jr. (eds. ), The Environment and Evolution of" Galaxies, 499- 532.
© 1993 Kluwer A cademic Publishers.
500

2. Present-Day Star Formation in Molecular Clouds

In this section, we review the current star formation paradigm which has
emerged in the last decade. In this paradigm, stars form within the cores
of molecular clouds. These core regions, which are small sub condensations
within the much larger molecular clouds, evolve in a quasi-static manner
through the process of ambipolar diffusion. In this process, magnetic field
lines slowly diffuse outward and the central regions of the core become in-
creasingly centrally concentrated. The magnetic contribution to the pressure
support decreases with time until the thermal pressure alone supports the
core against its self-gravity (at least in the central regions) . At this point,
the core is in an unstable quasi-equilibrium state which constitutes the initial
conditions for dynamic collapse.
When a core begins to collapse, a small hydrostatic object (the protostar)
forms at the center of the collapse flow and an accompanying circumstellar
disk collects around it. This phase of evolution - the protosteliar phase -
is thus characterized by a central star and disk, surrounded by an infalling
envelope of dust and gas. The characteristics of this infalling envelope largely
determine the spectral appearance of the object in this phase.
As a protostar evolves, both its mass and luminosity increase; the proto-
star eventually develops a strong stellar wind which breaks through the infall
at the rotational poles of the system and creates a bi-polar outflow. During
much of this bipolar outflow phase of evolution, the outflow is well colli-
mated (in angular extent) and in/ali takes place over most of the solid angle
centered on the star. The outflow gradually widens in angular extent and
the visual extinction to the central source gradually decreases. The outflow
eventually separa~es the newly formed star/disk system from its parental
core and the object enters the T Tauri Phase of evolution. The newly re-
vealed star then follows a pre-main-sequence track in the H-R diagram and
evolves toward the main-sequence.

2.1. MOLECULAR CLOUDS : THE BIRTHPLACE OF STARS

Molecular clouds are the birthplaces of stars in present-day galaxies. These

clouds are observed to be not collapsing as a whole, in spite of the fact that
their masses are generally much larger than a Jean's mass, i.e., the self-
gravity of these systems is generally much larger than the thermal pressure
support (see, e.g. , Zuckerman & Palmer 1974; Shu, Adams, & Lizano 1987).
As a result, these clouds must be supported by some nonthermal means.
The most viable candidates for providing this support are magnetic fields
and "turbulence". Magnetic fields with flux C) can support a cloud with a
 501

mass not exceeding the critical mass Me given by

Me
~
= 0.13 Gl/2 4 (B) (10pc
~ 2.5 x 10 Me 30JLG
R )2 ' (2.1)

where the numerical coefficient is from Mouschovias & Spitzer (1976).

Magnetic fields are observed in these clouds (see, e.g., Myers & Goodman
1988; Goodman et al. 1989; and the review of Heiles 1987) and typically have
field strengths in the range 10 - 30 JLG. In addition, maps of polarization
of background starlight seen through these clouds show very uniform (well
ordered) structure of the polarization vectors (e.g., Goodman 1990). Since
the polarization is believed to arise from scattering by dust grains which
are themselves aligned with the magnetic field, well-ordered structure in
the polarization maps is usually interpreted as well-ordered structure in the
magnetic field itself. In other words, the magnetic field is not extremely
tangled; it retains a more or less uniform component.
Turbulence in these clouds has begun to be studied, but definitive results
in this area are hard to come by (see, however, the review of Scalo 1987).
The formation of substructure within molecular clouds remains an im-
portant but largely unsolved problem. Molecular clouds are observed to
have highly nonlinear substructure on virtually all resolvable spatial scales
(e.g., Houlahan & Scalo 1992; Wood, Myers, & Daugherty 1992; Wiseman &
Adams 1992; see Figure 1). Cloud structure has been described in terms of
"sheets", "filaments", and quasi-spherical "clumps" of cloud material (see,
e.g., Blitz 1992; Myers 1991). At present, no definitive theory exists for the
formation of substructure within molecular clouds. However, if a cloud be-
gins to collapse, a wide spectrum of wave motions can be excited (Arons &
Max 1975). Nonlinear Alfven waves (Elmegreen 1990) and nonlinear solitary
waves (Adams & Fatuzzo 1993) have recently been studied and may provide
part of the explanation for the observed clumpy structure.

2.2. CORE FORMATION THROUGH AM BIPOLAR DIFFUSION

Molecular cloud cores represent the cloud structure on small size and mass
scales (closest to that of the forming stars). The formation of these cores
is thought to occur through the process of ambipolar diffusion (Shu 1983;
Lizano & Shu 1989; Nakano 1985; Mouschovias 1976, 1978). This process
occurs as follows: The cloud is very lightly ionized; only about one particle
per million carries a charge. The charged species - the ions - are very well
coupled to the magnetic field. However, the only interaction that the neutral
species has with the magnetic field is through interactions with the ions.
These interactions occur through a friction term of the form
(2.2)
502

Fig. 1. Column density map of Taurus Molecular Cloud. Map shows column density
contours as produced from lOOj.tm emission maps made by the [RAS satellite (from
Houlahan & Scato 1992; Wiseman & Adams 1992).

where Pn and Pi are the densities of neutrals and ions, respectively, and Un
and 11i are the corresponding velocities. The constant 'f/ is the drag coefficient,
which is usually taken to have the value 'f/ = 3.5 X 10 13 cm3 g-l S-l (Draine,
Roberge, & Dalgarno 1983; Shu 1992). Thus, the ions can exert a force on
the neutrals if and only if a drift velocity exists between the two species.
The physical meaning of this result is that the ions (and hence the magnetic
field) must be slowly diffusing outward relative to the neutral component,
which is slowly contracting because of gravity. This process takes place on a
timescale which is generally much slower than the free-fall collapse timescale
of the cloud. The ratio of the diffusion timescale to the dynamic timescale
is approximately given by

TdiJ = 'f/C
(2.3)
Tdyn (81rG)1/2'
where the constant C relates the ion density to the neutral density through
Pi = Cp~/2 (see Elmegreen 1979). Since the ratio given by equation [2.3]
 S03

typically has a value of rv 8 - 10, the loss of magnetic flux provides the
bottleneck in the star formation process and keeps the efficiency of star
formation fairly low (rv 1 %) in present-day molecular clouds.
We stress, however, that the mass scale defined by cores in still much
larger than that of the forming stars. Hence, the interstellar medium cannot
provide the mass scale for star formation. Instead, these core regions do
provide the initial conditions for protostellar collapse, as described in the
following sections.

2.3. MOLECULAR CLOUD CORES: INITIAL CONDITIONS FOR COLLAPSE

Stars form within molecular cloud cores (e.g., Myers et al. 1987) and hence
the observed core properties provide the initial conditions for protostellar
collapse. In nearby molecular cloud complexes which are forming low mass
stars, these cores are observed to be nearly isothermal and slowly rotating. In
our current (idealized) theory of star formation, the cloud cores (the initial
state) can be described by two physical variables: the isothermal sound speed
a and the (uniform) angular velocity n of the core. Current observations of
cloud cores indicate temperatures in the range 10 - 35 K (e.g., Myers 1985)
and corresponding sound speeds a = 0.20-0.35 km S-1. When rotation rates
are detectable, the measured angular velocity is of the order n = 1 km S-1
pc- 1 rv 3 X 10- 14 rad s-1 or larger. At spatial scales much smaller than the
radius ofrotational support, R = a/n rv 0.1- 1.0 pc, isothermal cloud cores
are highly centrally concentrated and are expected to have density profiles
of the form
(2.4)
(see Shu 1977; Terebey, Shu, & Cassen 1984).
Recent observational work (Myers and Fuller 1992) has suggested that
cores of higher mass have a slightly different structure than the pure isother-
mal model described above. For sufficiently large size scale (1' rv 1 pc) and
low density n < 104 cm-a, the observed linewidths ~v in cores have a sub-
stantial nonthermal component which scales with density according to
~V()(p-1/2. (2.5)
If this velocity is interpreted as a transport speed, then a "turbulent" or
"nonthermal" component to the pressure can be derived (Lizano & Shu
1989; Myers & Fuller 1992). The entire effective equation of state for the
molecular cloud fluid then becomes

p = a 2 P + Po log[p/ Po], (2.6)

where Po is determined by the amplitude of the nonthermal motions. When a
core with the above equation of state undergoes dynamic collapse, the initial
504

evolution is the same as for a purely isothermal core. However, as soon as

the expansion wave of the collapse grows into the spatial region where the
nonthermal component is substantial, the expansion wave accelerates and
the mass infall rate if increases with time (see the following section).

2.4. PROTOSTELLAR COLLAPSE

In this section we discuss protostars, objects still gaining mass through infall.
For simplicity, we focus our discussion on low-mass protostars (M ~ 2M0 ),
where the radiation field is not strong enough to affect the infalling envelope
and the dynamical collapse is thus decoupled from the radiation.
An isothermal cloud core wilL collapse from inside-out (Shu 1977), i.e., the
interior of the core will collapse first and successive outer layers will follow.
The collapse naturally produces a core/envelope structure, with a central
hydrostatic object (the forming star) surrounded by an infalling envelope
of dust and gas. This inside-out collapse progresses as an expansion wave
propagates outward at the sound speed. Outside the expansion wave radius
( r H = at) the cloud is static; inside this radius, the flow quickly approaches
free-fall velocities. The density distribution of the flow is nearly spherical
outside a centrifugal radius,

(2.7)

the position where the infalling material with the highest specific angular
momentum encounters a centrifugal barrier (Cassen & Moosman 1981; Tere-
bey, Shu, & Cassen 1984). Inside the radius Rc, the flow becomes highly
non-spherical as particles spiral inward on nearly ballistic trajectories. In
the region immediately surrounding the star, the temperature is too high
for dust grains to exist and an opacity-free cavity is produced (see Stahler,
Shu, & Taam 1980).
Notice that there is no mass scale in this collapse scenario. Instead, the
collapse flow feeds material onto the central star and disk at a well defined
mass infall rate
• 3
M = rnoa /G, (2.8)
where rno = 0.975 is a constant (see Shu 1977). The absence of a mass scale
suggests that the origin of stellar masses is determined in part by stellar
processes rather than only by the interstellar medium (e.g., Shu, Adams, &
Lizano 1987; Lada & Shu 1990; see also §4). Notice also that in a complete
theory, three physical variables specify the (idealized) problem: the sound
speed a and the rotation rate S1, which determine the initial conditions,
and the time t since the collapse began (or equivalently, the mass M = if t
that has fallen to the central star/disk system so far). As we discuss below,
 505

however, our current theory is incomplete and we must specify an additional

parameter which represents the effects of disk accretion.

2 .5 . PROTOSTELLAR RADIATION

The radiation from forming stars is important because it allows for both a
test of the underlying theory and for a means of identifying protostellar can-
didates . The radiation field for protostellar objects can be divided into three
separate components: the star, the disk, and the infalling dust envelope. The
stellar and disk components are highly attenuated by the extinction of the
dust envelope and most of the luminosity is absorbed by dust grains and
re-radiated at far-infrared wavelengths. Hence, the exact form of the initial
star / disk spectrum is unimportant as long as the total luminosity of the
system is properly determined. The protostellar luminosity has several com-
ponents. The infalling material produces shocks on the surfaces of both the
star and disk and thereby dissipates energy. Additional energy is dissipated
as the infalling interstellar material becomes adjusted to stellar and disk
conditions. Disk accretion produces a substantial amount of luminosity. The
main uncertainty in this picture is the disk accretion mechanism; however,
disk stability considerations place fairly tight constraints on the allowed disk
accretion activity (see §3). The total luminosity of the system is thus rea-
sonably well determined and is a substantial fraction of the total available
luminosity
L _ GMM
0- , (2.9)
R.
where R. is the stellar radius (typically R. '" few R0 - see Stahler, Shu,
and Taam 1980).
The diffuse radiation field of the dust envelope can be determined through
a self-consistent radiative transfer calculation (see Adams & Shu 1985, 1986).
The theoretical spectral energy distributions calculated from this proto-
stellar model are in good agreement with observed spectra of protostellar
candidates (see Figure 2 and Adams, Lada, & Shu 1987; hereafter ALS).
The spectral energy distributions generally have maxima at wavelengths of
60 -lOOJLm and have absorption features at 9.7 JLm (from silicates) and at 3.1
JLm (from water ice). The spectra of sources in the bipolar outflow stage of
evolution are also well described by protostellar in/all models (see ALS); this
result suggests that both infall and outflow are taking place simultaneously
in such objects.
Since the density distribution of the protostellar envelope is known from
the dynamic collapse solution and the temperature distribution is known
from radiative transfer calculations, the spatial distribution of emission from
protostars can be calculated theoretically. Recent observations (especially
506

Haro 6-10

11 12 13 14 lfj
Log[v]

Fig. 2. Observed and theoretical spectral energy distributions of protostellar can-

didate Haro 6-10 (adapted from Adams, Lada, & Shu 1987).

from the J CMT on Mauna Kea) have produced spatial emission maps of
protostars at millimeter and submillimeter wavelengths in the continuum.
These measurements show that protostellar emission is extended and that
the observed spatial profiles of emission are roughly consistent with the
current theory (see Figure 3; Walker, Adams, & Lada 1990; Butner et al.
1991; Ladd et al. 1991; Adams 1991).

2.6. THE L - Av DIAGRAM FOR PROTOSTARS

Protostars cannot be placed in the H-R diagram because they have no well-
defined photospheric temperature: protostars have extended atmospheres
and hence their spectral energy distributions are always much wider than
that of a blackbody. As a result, another approach must be invoked to de-
scribe this early phase of stellar evolution. The total system luminosity L
and the column density N of the infalling envelope (or, equivalently, the
visual extinction Av) are the two most appropriate physical quantities that
 507

I
I
,
\
I \
I \

I
I
\\
I
I \
I \
I \
I \
I \
I \
I \
I \
I \
I \
I \
I \
I \
\
I
I \
\
I \
I \
I \
I \
I \
I \
I \
I \
I \
/' ""
o
-2 2

Fig. 3. Spatial distribution of emission for protostellar candidate L1551-IRS5

(adapted from Walker, Adams, & Lada 1990).

describe the system; we therefore use the L - Av diagram to characterize

the protostellar phase of evolution. The total system luminosity constitutes
the vertical axis; the visual extinction Av = KN of the infalling envelope
constitutes the horizontal axis (which is plotted backwards). For a given
initial state (i.e., a given a and n) a protostar will trace out a well-defined
track in the diagram as time proceeds (Adams 1990).
Figure 4 shows the L - Av diagram for protostars in the nearby Taurus
molecular cloud. The solid lines in the diagram are the set of tracks calcu-
lated for a = 0.20 km s-l, which is appropriate for Taurus; the open symbols
represent observed sources. As time proceeds, protostars move toward the
upper right in the L - A V diagram. The visual extinction A V decreases
with time until the object moves off the right-hand side of the diagram and
becomes optically visible. The newly formed star will then appear on the
birthline in the H-R diagram (Stahler 1983; see also Shu 1985; Stahler 1988;
Palla & Stahler 1991) and enter the T Tauri evolutionary phase. The lower
dotted curve in Figure 4 shows the evolution of a protostar if the disk ac-
cretion is assumed to be zero. Notice that as time goes on, irifalling material
508

,
,,,
,
,,,
,
,,/
,
,,,
,
,,/
,,,'
.....
o
/
/

.0 ,
I
/

t---=I

10000 1000 100 10 1

Av
Fig. 4. The L - Av Diagram for protostars in the Taurus molecular cloud (adapted
from Adams 1990).

falls through an increasingly shallow potential well and hence the luminosity
of the system decreases to values much lower than those of observed sources.
Thus, disk accretion must occur.

2.7 . THE PROTOSTELLAR TO STELLAR TRANSITION

The transition from embedded source to optically revealed T Tauri star may
be accomplished through the action of a powerful stellar wind. Although the
presence of such winds (and related outflow phenomena) are well established
observationally (Lada 1985), the wind production mechanism is not yet un-
derstood theoretically. However, one promising approach is a centrifugally
driven magnetic wind mechanism (see, e.g., Shu et al. 1988 for further dis-
cussion; see also Najita 1992). Notice that although the wind mechanism
is not yet fully understood, these winds and the outflows they drive are
crucially important for understanding the final properties of the star/disk
system (and hence for understanding the initial mass function - see §4). We
 509

-.
.J
......
.J
-0
o
C'I
o

-1
TAURUS -AURIGA

Fig. 5. The stellar hirthline (adapted from Stahler 1988).

also note that these winds provide an energy source for the cloud (Norman
and Silk 1980) and can help support the cloud against collapse.

When pre-main-sequence stars (often T Tauri stars) first appear in the

H-R diagram, they seem to appear on a well defined locus denoted as the
"stellar birthline" (see Figure 5; Stahler 1983). The location of this birth-
line corresponds to stellar configurations which are capable of burning deu-
terium (Shu 1985; Stahler 1988). In other words, stars appear on the H-R
diagram as visible objects with just the right properties to burn deuterium.
Deuterium burning produces stellar convection which, in conjunction with
differential rotation in the star, can lead to dynamo action and the produc-
tion offairly large (100 - 1000 G) magnetic fields. Since these field strengths
are required for the centrifugally driven wind mentioned above, this entire
picture of stars determining their own mass through the production of winds
is consistent.
510

2.8. PRE- MAIN-SEQUENCE STARS AND CIRCUMSTELLAR DISKS

Young stars are often observed to have infrared excesses in their spectral
energy distributions (see, e.g., Rydgren & Zak 1987; Rucinski 1985; see also
the review of Appenzeller & Mundt 1989). The most likely explanation for
the infrared excess is the presence of circumstellar disks, which may be
either active and have appreciable luminosity or passive and only absorb
and re-radiate stellar photons.
We first discuss systems with passive disks. In these systems, all of the
luminosity is intrinsic to the star and the disk is assumed to be spatially
thin and optically thick. In the limit that the disk radius is large compared
to the stellar radius, the disk intercepts and re-radiates 25% of the stellar
luminosity. The resulting surface temperature profile of the disk approaches
the form TD '" w- 3 / 4 at large radii w (see Adams & Shu 1986). The disk thus
radiates like a classical Keplerian accretion disk with an effective luminosity
of L. /4 and adds an infrared excess to the spectral energy distribution of
the system. These passive disk models have no free parameters and produce
the correct infrared excess for some observed T Tauri systems (see ALS;
Kenyon & Hartmann 1987).
Next we discuss the class of star / disk systems with active disks which have
appreciable intrinsic luminosity LD in addition to the energy intercepted
from the star. This intrinsic energy source produces a radial temperature
gradient of the form TD '" w- q , which is generally much flatter than that
of a classical Keplerian accretion disk (where q = 3/4). In modelling these
systems, we allow for the disk to be partially optically thin by introducing a
surface density profile of the form 0" rv w- P , where the magnitude of the the
surface density profile is determined by the mass MD of the disk. By using
models of this type, we can fit the observed spectral energy distributions of
virtually all T Tauri stars with infrared excesses, from passive disks (in the
limit LD ---+ 0) to the extreme case of flat spectrum objects.
Through fitting the spectral energy distributions of active disk systems,
we can estimate the basic physical properties of the disks (see Figure 6;
Adams, Lada, & Shu 1988; Adams, Emerson, & Fuller 1990; Kenyon & Hart-
mann 1987). First, we find that the disks associated with the flat spectrum
sources must have unorthodox radial gradients of temperature and contain
intrinsic luminosity in addition to the energy intercepted and reprocessed
from the central star, i.e., the disks must be active. The minimum values
for the radii of these disks are ",100 AU. Estimates for the disk masses can
be obtained by measuring the dust continuum emission at low frequencies,
where the disks are likely to be optically thin. The currently available mass
estimates, which depend on the dust opacity at submillimeter wavelengths
(see Beckwith et al. 1990), lie in the range 0.1 - 1.0 M0 (e.g., Adams, Emer-
son, & Fuller 1990). Notice that the estimated disk properties (radial sizes,
 511

T Tau
~
C')

,.--,
~
~
~
t\l
~
t:::: N
C')
~
L--J
tlD
0
~

0
C')

11 12 13 14 15 16

Log[v]
Fig. 6. Observed and theoretical spectral energy distributions of the T Tauri
star/disk system (adapted from Adams, Emerson, & Fuller 1990). dashed curves
show effects of finite disk masses of 0.01, 0.1, and 1.0 Me

masses, and hence angular momenta) are in good agreement with the disk
properties predicted by the protostellar theory (see above and ALS).

2.9. SUMMARY OF PRESENT-DAY STAR FORMATION

We have outlined the current picture of star formation and shown that
the observed spectra of protostellar candidates are in good agreement with
spectra predicted by the theory. The spectral energy distributions calculated
from the protostellar theory fit objects in both the pure infall and the bipolar
outflow phases of evolution; hence, infall is probably still taking place in the
latter. The emission from protostellar candidates is observed to be extended
in a manner which is roughly consistent with theoretical expectations. In
addition, the spectral energy distributions of T Tauri stars with infrared
excesses can be understood as young stars surrounded by circumstellar disks.
Some systems have passive disks that reprocess stellar radiation but have
5t2

no intrinsic luminosity, whereas other systems require appreciable intrinsic

disk luminosity. For some sources, estimates for the disk masses (0.1 - 1.0
M 0 ) and the disk radii ('" 100 AU) can be obtained.
Two fund~ental problems in the current theory of star formation are
the disk accretion mechanism and the stellar wind mechanism. For the disk
accretion problem, dynamical studies of self-gravitating instabilities (see the
following section) are promising and may lead to the observed activity in
these systems. For the wind problem, the centrifugally driven magnetic wind
(see, e.g., Shu et al. 1988) is also promising, but more work is needed to
complete the theory. We also stress that the current theory works well for
the formation of single stars oflow mass (although most stars are members of
binary systems - see, e.g., Abt 1983). The extension ofthis theory to include
the formation of binary systems and stars of higher mass is currently being
studied.

3. Gravitational Instabilities in Disks

The picture of star formation described above is incomplete, especially our

current understanding of disk physics. During the protostellar phase of evo-
lution, disk accretion must occur: otherwise, the masses of the forming stars
would be unrealistically small, the luminosity would be much smaller than
that of observed sources, and the disk would become gravitationally unsta-
ble. During the T Tauri phase of evolution, disk accretion must also occur
in order to produce the observed intrinsic disk luminosities and flat radial
gradients of temperature. However, the disk accretion mechanism is not yet
understood, although gravitational instabilities may play an important role
in the disk accretion process.
In this section we discuss the stability of circumstellar disks to the growth
of self-gravitating modes (see Adams, Ruden, & Shu 1989; hereafter ARS,
and Shu, Tremaine, Adams, & Ruden 1990; hereafter STAR). These cal-
culations begin with an unperturbed system which consists of a star and
an accompanying gaseous disk. The growth of spiral modes is mainly deter-
mined by three elements: self-gravity, pressure, and differential rotation. The
gravitational forces are determined by the potential of the star and by the
disk's surface density distribution O"o(r), which is assumed to be a power-law
in radius r from the central star. The pressure is determined by the tempera-
ture distribution, or equivalently, the distribution of sound speed in the disk.
Observations indicate that T(r) and hence a(r) is also a power-law in ra-
dius. The rotation curve O( r ) in the disk is then determined self-consistently
from the potential of the star, the potential of the disk, and the pressure
gradients. Since the potential well of the star dominates that of the disk
everywhere except near the disk's outer edge, the rotation curve is nearly
 513

Keplerian throughout most of the disk's radial extent. Since observations

indicate that "typical" disk sizes are'" 100 AU (see §2), the linear stabil-
ity calculations consider disk sizes RD up to 0(10 4 ) times the radius R. of
the star (only the ratio RD / R. enters into the calculations). For the non-
linear SPH simulations discussed at the end of this section, computational
limitations necessitate smaller radius ratios of RD/ R. = 30.

3.1. WHY m=l PERTURBATIONS?

Recent work has focused on modes with azimuthal wavenumber m= 1. For

a typical circumstellar disk, the gravitational potential is dominated by the
star so that the disk rotation curve is nearly Keplerian. Hence, spiral modes
with azimuthal wave number m = 1 should arise naturally in these systems.
The particle orbits in a purely Keplerian potential will be closed ellipses with
foci at the stellar position. In the absence of other forces, the orbits will not
precess spatially and a purely kinematic mode can occur. If the potential is
not exactly Keplerian, a relatively small amount of self gravity in the disk
should be sufficient to sustain an m = 1 mode in a circumstellar disk.
Modes with m = 1 are especially interesting in the context of these
star / disk systems because the center of mass of an m = 1 perturbation does
not lie at the system's geometrical center (the star). Formally, this effect
gives rise to a noninertial contribution to the potential and plays an impor-
tant role in the amplification of these modes. Physically, this effect causes
the star to move from the system's center of mass, i.e., this coupling allows
the disk to transfer angular momentum to the stellar orbit. This indirect
potential is essential for the growth and maintainence of spiral modes with
azimuthal wavenumber m = 1. In fact, the interaction of this indirect po-
tential with the outer Lindblad resonance in the disk can be the dominant
amplification mechanism for these modes (see ARS and STAR).

3.2. DENSITY WAVE PHYSICS

In the simplest description of spiral instabilities, self-excited disturbances

(spiral modes) can grow through the feedback and amplification of spiral
density waves. In the asymptotic (WKBJ) limit, the dispersion relation for
spiral density waves in a gaseous disk has the form

(3.1)

where k is the radial wavenumber, K, is the epicyclic freqeuncy, and where

w is the complex eigenvalue of the system (see, e.g., Lin & Lau 1979). Since
514

the gravitational term is proportional to Ikl, this dispersion relation has four
branches,
(3.2)
where

k - 1I" GlTo
o=~,

Here, the quantity Q determines the stability of the system to azisymmetric

disturbances and is defined by

Q=~ (3.4)
- 1I"GlTo'

where Q > 1 implies axisymmetric stability (Toomre 1964). The overall sign
of k determines whether the waves are leading (k > 0) or trailing (k < 0);
the inner sign determines whether the waves are short [k Q( (ko + k1 ) J or long
[k ex (k o -- kdJ.
The quantity v is a dimensionless frequency of the spiral density :waves.
The radius in the disk where v = 0 (i.e., where 3l(w) = mn) is known
as the corotation resonance (CR); the energy and angular momentum of
the perturbation (and the action) are positive outside the corotation radius
and negative inside. Notice that for Q > 1, the wavenumber kl becomes
imaginary for radii sufficiently close to the CR, i.e., a classical turning point
exists for the density waves. The resulting "forbidden" region surrounding
the CR is known as the "Q-barrier". Notice also that for long waves, k -+ 0
at any radius where Ivl = 1. The radius in the disk where v = +1 is known
as the outer Lindblad resonance (0 LR) and plays an important role in the
physics of m = 1 modes.

3 .3. SLING AMPLIFICATION AND THE 4 WAVE CYCLE

Our analysis indicates that the dominant mechanism for modal amplifica-
tion arises from the indirect potential, which provides an effective forcing
term. The indirect term varies slowly with radius in the disk. Notice that a
slowly varying force can only couple to oscillatory disturbances at the disk
edges or at the Lindblad resonances; for the modes considered here, the main
coupling occurs at the outer Lindblad resonance. Thus, this amplification
mechanism differs substantially from the previously studied mechanisms,
which utilize the process of super-reflection across the corotation resonance
(super-reflection can still occur in these disks and is included in the numer-
ical treatment, but it does not dominate the amplification). In our analytic
treatment, we determine the growth rates for the modes under the assump-
tion that all of the amplification arises from this coupling of the indirect
 515

rD

Fig. 7. The Four Wave feedback cycle for m = 1 spiral modes in circumstellar disks
(from STAR - see text).

term to the outer Lindblad resonance in the disk (and that the indirect
term arises mostly near the outer disk edge). In other words, we concep-
tually regard the indirect potential as an external forcing term acting on
the disk and calculate the torque exerted on the disk at the OLR. Since
the long-range coupling of the star to the outer disk provides the essential
forcing, this new instability mechanism is called SLING: Stimulation by the
Long-range Interaction of Newtonian Gravity.
We now discuss the feedback cycle for m = 1 modes in gaseous disks
(from STAR). One interesting aspect of this feedback cycle is that all four
types of waves are utilized (see Figure 7).
[A] Begin (somewhat arbitrarily) with the excitation of a long trailing (LT)
spiral density wave at the outer Lindblad resonance (OLR) by the indirect
term. The LT wave propagates inward (its group velocity is negative) until
it encounters the outer edge of the Q-barrier.
[B] At the Q-barrier, the LT wave refracts into a short trailing (ST) spiral
density wave that propagates back outward, through the OLR to the outer
disk edge.
516

[C] The ST wave that propagates to the outer disk edge reflects there to
become a short leading (SL) wave. The SL wave then propagates back to
the interior, through the OLR, until it encounters the outer edge of the Q-
barrier, where it refracts into a long leading -(LL) spiral density wave that
propagates back toward the OLR.
[D] At the OLR, the LL wave reflects to become a LT wave. H the reflected
LT wave possesses the correct phase relative to the LT wave launched from
OLR by the indirect term in step [A] above, then we have constructive
reinforcement of the entire wave cycle, and the-hasis for the establishment
of a resonant wave cavity.
U sing a WKBJ analysis, we have derived a quantum condition on the
basis of the above four-wave cycle. This quantum condition accurately pre-
dicts the pattern speeds (i.e., the real part of the eigenfrequencies) for these
modes; for strongly growing modes, the analytical and numerical results
agree to within ",1 % (see STAR).
The combined numerical and analytical treatments indicate the depen-
dence of the growth rates (Le., the imaginary part of the eigenfrequencies)
on the parameters of the problem. Most importantly, a finite threshold ex-
ists for the SLING Amplification mechanism. When all other properties of
the star/disk system are held fixed, this effect corresponds to a threshold in
the ratio of disk mass MD to the total mass M. + MD. We find that the
growth rates are largest for the case of equal masses MD = M. and decrease
rapidly with decreasing relative disk mass. In the optimal case, MD = M.,
the growth rates can be comparable to the orbital frequency at the outer
disk edge, Le., the modes can grow on nearly a dynamical timescale. On the
other hand, the presence of the finite threshold implies a critical value of
the relative disk mass, Le., the maximum value of MD/(M. + MD) that is
stable to m = 1 disturbances; for the simplest case of a perfectly Keplerian
disk and Q(RD) = 1, this critical ratio has the value

MD 3
(3.5)
(M. +MD) 411"

3.4. NONLINEAR SPH SIMULATIONS OF STAR/DISK SYSTEMS

In order to study the nonlinear behavior of the gravitational instabilities

which are suggested by the linear analysis, we have performed smooth par-
ticle hydrodynamic (SPH) calculations of the behavior of these instabili-
ties in nearly Keplerian disks (Adams & Benz 1992). Previous studies have
performed N-body simulations of similar star/disk systems (e.g., Tomely,
Cassen, & Steiman-Cameron 1991) and have obtained qualitatively similar
results.
 517

The numerical simulations of star / disk systems were performed using the
smooth particle hydrodynamics code written by W . Benz (see Benz 1990
and Benz et al. 1990). All of these simulations use an isothermal equation of
state, i.e., p = a 2 p, where the sound speed a is a constant. From a physical
point of view, this assumption implies that the disk is able to radiate all
energy dissipated in shocks or adiabatic compression.
The linear stability analysis (ARS, STAR) suggests that modal growth is
relatively insensitive to the inner boundary condition, but rather sensitive to
the outer boundary condition (the effects of the outer boundary are quanti-
fied in Ostriker, Shu, & Adams 1992). In particular, the outgoing ST waves
must be able to reflect off of the outer edge. For the SPH simulations, we
adopted a "free" outer boundary condition, i.e., the particles on the outer
edge are free to wander according to the gravitational and pressure forces
exerted on them by the rest of the system.
Next, we specify the initial perturbations of the disk. Since we are con-
strained by computing resources to studying only the first few orbits of the
outer disk edge (the hence the first ",100 - 200 orbits of the inner edge),
we start the simulations with 1% amplitude perturbations in density with
azimuthal wavenumber m = 1.
We find that gravitational instabilities can grow strongly in these systems,
i.e., the grow rates are comparable to the dynamical timescale of the outer
disk edge. For disks which are not too far from the condition of stability to
axisymmetric modes, spiral instabilities with azimuthal wavenumbers m =
1,2,3,4 (and higher) are present. As stability is increased (Le., as the Toomre
Q values are raised or the mass ratio MD/M. is decreased), the relative
strength of the m = 1 disturbance increases, although the growth rates of
all modes decrease (as expected).
Finally, we find that a spiral arm can often collapse to form a "knot" of
bound gas when the equation of state of the disk is isothermal (see Figure 8).
These collapsed "knots" typically have masses of ",0.01 MD and travel on
elliptical orbits. The possibilty that these knots survive to form either giant
planets or binary companions is especially interesting. Notice, however, that
these knots are formed as a result of our assumption of isothermal evolution
of the disk. This assumption limits the local pressure support available. The
density enhancements found in our unstable modes are sufficient for pushing
small regions over the local Jeans mass and thus triggering the collapse.
This collapse would not occur for adiabatic evolution. Hence, the formation
of these bound knots of gas can occur in real physical systems only when
these systems can efficiently radiate away a large fraction of their energy.
518

... ' .
. ...
'. '.. . ..
'

.
'
.
: ... : . ... : ...:.... : ": .:,',
'

" .'.~" "

• • • ' .' •••••• " " . ' .' '0 ' ,:, . ' •• .'.

... ·?··: ~:;<iiC.,/'//·'. ,. . .• '. . .

. ..' .
" ,"

, ,,',
... ]:N;X}·'/ •.·.
. :. >.:.:./.~.?/:.:.~.: .: :,: :.:.:. . -":
: " .

. . . , '0
...
"
' ..... .

Fig. 8. Result of an SPH simulation of a star/disk system. The main spiral arm has
collapsed to form a gravitationally bound "knot" of gas (adapted from Adams and
Benz 1992).

3.5. SUMMARY OF DISK PROCESSES

The calculations done thus far show that self-gravitating instabilities can
grow on nearly a dynamical timescale (the orbital timescale at the outer
disk edge) which is much shorter than the evolutionary timescale of these
systems (10 5 - 106 yr). SPH calculations show that these instabilities can
grow well into the nonlinear regime and that small "knots" of gravitationally
bound gas can form out of the disk. Thus, these instabilities may playa role
in the disk accretion processes and/or the formation of binary companions
out of the disk. We stress, however, that much more work is needed to
explore these possibilities.
In addition to gravitational instabilities, which are the focus of this sec-
tion, other physical processes can lead to disk accretion. We expect that
self-gravitating instabilities will be most important early in the evolution of
the disk (when the disk mass is relatively large). At sufficiently late times,
the disk mass must eventually become small enough that the self-gravitating
 519

instabilities will shut off and some type of viscous accretion process will take
over (see the review of Adams and Lin 1990 for futher details) .

4. The Initial Mass Function

The initial mass function (IMF) is perhaps the most fundamental output
of the star formation process . As discussed throughout the other chapters
in this volume, a detailed knowledge of the initial mass function is required
to ~derstand the chemical evolution of galaxies and the interstellar (and
intergalactic) medium. Unfortunately, at the present time, we remain un-
able to calculate the initial mass function from a priori considerations (see,
e.g., Zinnecker, McCaughrean, & Wilking 1992 for a recent comprehensive
review).
We can, however, empirically determine the IMF today in our galaxy.
The first such determination (Salpeter 1955) shows that the number of stars
born with masses in the range m to m + dm is given by the relation

( 4.1)

where the index JL = 2.35 for stars in the mass range 0.4 - 10 M 0 . More recent
work (Miller & Scalo 1979; Scalo 1986) suggests that the mass distribution
may have two local maxima: a primary maximum at m '" 0.3 M0 and a
(weaker) secondary maximum at m '" 1.2 M 0 . In addition, the distribution
seems to flatten out (and perhaps turn over) at the lowest masses ('" 0.lM0 ).

4.1. THE IMF AS A Two PHASE PROCESS

In considering the calculation of the initial mass function, we can conceptu-

ally divide the process into two basic subproblems:
[A] The spectrum of initial conditions produced by the interstellar (or in-
tergalactic) medium, and
[B] The transformation between a given set of initial conditions and the
properties of the final (formed) star.
In the following subsections, we discuss the current status of each of these
two subproblems.

4.2. THE SPECTRUM OF INITIAL CONDITIONS

In order to understand item [A], we must know in detail all of the relevant
processes which occur in the formative medium (molecular clouds in the
case of present-day star formation) . In particular, we must understand the
520

co
o
o
C!

"<f'
.......-,. 0
0
><
"-""
C!
b
CIl
0
0
~

o
o 50 100 150 200

Fig. 9. A one-dimensional "cut" across the Perseus molecular cloud showing highly
nonuniform structure. Column density (in arbitrary units) is plotted against pixel
number (adapted from Wiseman and Adams 1992).

mechanism that selects out the distribution of properties for molecular cloud
cores (recall that these cores provide the initial conditions for protostellar
collapse - see §2). Unfortunately, however, our current understanding of
this aspect of the problem is rather poor. Much more work on the formation
of substructure within molecular clouds is needed before we can begin to
predict the distribution of core properties from a priori considerations.
Although we cannot calculate and predict the spectrum of initial condi-
tions in present-day clouds, we can measure these conditions. Unfortunately,
an in-depth discussion of molecular clouds as star forming environments is
beyond the scope of this present discussion. For our purposes here, however,
the most important property of these clouds is that they are highly non-
uniform (e.g., see Figure 9)j clumpiness and structure exist on all resolvable
scales (see also §2 and Figure 1). Thus, no characteristic density exists for
these clouds and hence no (single) Jeans mass exists. We stress that, at
least in the case of present day star formation, the Jeans mass has virtually
nothing to do with the masses of forming stars.
 521

As discussed in §2.1, these clouds exhibit generally quasi-static behavior.

Molecular clouds are not observed to be collapsing as a whole; on average, the
lifetime of a molecular cloud is an order of magnitude longer than the naive
free-fall time. Some additional support mechanism (in addition to thermal
pressure) must be present in these clouds; magnetic fields and turbulence
are generally believed to play an important role. Thus, it makes sense to
conceptually divide the process of determining stellar masses into the two
steps given above. We note that this state of affairs need not hold for the
formation of the first stellar generation.

4.3. SEMI-EMPIRICAL MASS FORMULA

In order to understand the transformation between initial conditions and

the final masses of the stars produced (item [B] above), we must know how
the stellar outflow stops the inflow and separates the star j disk system from
its molecular environment. Although this process has not been well studied,
in this section we obtain an order of magnitude estimate by adopting the
arguments first presented by Shu, Lizano, & Adams (1987) .
Since we do not yet understand how high velocity outflows are produced,
we must proceed in a semi-empirical manner. The kinetic energy Eout of the
outflow will generally be some fraction a of the binding energy of the star,
Le.,
GM 2
Eout = a--* . (4.2)
R*
The natural time scale associated with stellar processes is the Kelvin Helmholtz
timescale. We thus take the duration of the outflow (which is produced by
a stellar process) to be a fraction f3 of the Kelvin-Helmholtz time, Le.,

( 4.3)

Combining the above two equations then reproduces the observational cor-
relation that the flow luminosity Lout == Eout/ Tout is roughly a constant frac-
tion of the photon luminosity of the central source L* over several decades
of L., Le.,
Lout a
( 4.4)
L.-~'
where empirically aj f3 '" 10- 2 (see Bally & Lada 1983; Lada 1985).
We now derive a scaling relation for the masses of forming stars. The
strength of the infall can be measured by the rate M. at which matter falls
directly onto the star. When the stellar radius is small compared to the
centrifugal radius, R. ~ Rc, the direct infall rate M. is a small fraction of
522

the total mass infall rate (see equation [2.8]) and is given by (R./2R c )M
(from equation [24J of Adams & Shu 1986). Given the quantities M and Rc,
we thus have
(4.5)
where we have used equations [2.7] and [2.8] . Disk stability considerations
(§3) suggest that the stellar mass M. is some fraction "( of the total mass
M which has collapsed to the central star/disk system, i.e.,

M. =,,(M. ( 4.6)

The stellar wind must have a mass loss rate Mw with a terminal velocity
comparable to the escape velocity from the surface of the star. In order for
the outflow to be sufficiently strong to reverse the direct infall (with the rate
lIit.), the mass loss rate Mw of the wind must be be comparable to M.:
(4.7)

Notice that we should really compare the ram pressure ('" Mw vw) of the
wind with that of the infall ('" lIit. v.). However, both velocities are deter-
mined by the depth of the stellar potential well and are thus comparable
in magnitude; we thus divide out the velocities and incorporate any uncer-
tainties into the parameter ~ . Finally, if the winds roughly conserve energy
while driving bipolar outflows, we can write

(4.8)

where € is a constant.
The combination of the above results then implies that the final properties
of the newly formed star (at the point of evolution when its wind is capable
of reversing the direct infall) are given by

~(f3) all all

L.M; = 8mO"(3 ~ ~ G 3 S"P = X G 3 n2 ' (4.9)

where in the second equality we have collected all of the constants into a
single parameter x. Under most circumstances, we expect the parameters ,,(,
~, and € to be of order unity. For example, the disk stability arguments of the
previous section suggest that "( ~ 2/3; a smaller value would correspond to a
larger relative disk mass and hence an unstable system. We find empirically
that f3 /
a '" 102 • We thus expect that X '" 250 is reasonable estimate.
Equation [4.9] above provides us with a transformation between initial
conditions (the sound speed a and the rotation rate n) and the final prop-
erties of the star (the luminosity L. and the mass M.). If we use "typical"
 523

values for present day clouds (e.g., a = 0.2 km S-1 and n ,. . ., 3 X 10- 14 rad
S-I), we obtain stellar masses of approximately M. ~ 0.5 M0' which is the
typical mass of stars forming in regions with these properties. Thus, in spite
of its highly idealized nature, equation [4.9] provides reasonable order of
magntiude estimates for M. as a function of initial conditions (a, n).
In the interpretation of the transformation [4.9], several points must be
kept in mind. The right hand side of the equation depends rather sensitively
on the sound speed (,....., all) and hence the temperature (,....., T ll / 2). However,
on the left hand side of the equation, we must use the luminosity vs mass re-
lation for stars; for low mass stars on the main sequence, L. ,. . ., M;, whereas
for high mass stars L. ,. . ., M •. We thus obtain (very crude) scaling relations of
the form M. ,....., T ll / 12 for low mass stars and M. '" T ll / 3 for high mass stars.
Keep in mind, however, that much uncertainty has been encapsulated in the
parameter X, which should really be considered as a complicated function
of all the environmental parameters. We have also characterized the initial
conditions by only two physical variables (a and n) whereas much more
complicated initial states are possible. Finally, we have ignored radiation
pressure in this argument; for sufficiently massive stars (M. ~ 7 M 0 ) radi-
ation pressure will help the outflow reverse the infall (e.g., see Wolfire and
Cassinelli 1986, 1987).

5. Primordial Star Formation

In this section, we discuss the problem of primordial star formation - the

first stellar generation - as an extrapolation of the theory of present-day
star formation. We emphasize that this extrapolation is dangerous because
little or no data is available to directly constrain theories of primordial star
formation.

5.1. INITIAL CONDITIONS

One fundamental problem with understanding primordial star formation is

that we have no way of measuring the initial conditions. Even for the case of
present day star formation, where we can measure the initial conditions in
great detail, we are presently unable to calculate a definitive transformation
between a given set of initial conditions and the corresponding properties of
the final system. However, as discussed in the previous section (see especially
equation [4.9]), the masses of forming stars are expected to increase with
increasing temperature of the initial pre-collapse states (molecular cloud
cores in the case of present day star formation). In the primordial case,
however, we do not know the strength and structure of the magnetic field, the
density structure, or the rotation rates (angular momentum distributions) .
524

At first glance, it appears that star formation with zero (or very low)
mctallicity must proceed very differently from present day star formation.
hi the absence of heavy elements (and, in particular, dust grains), most of
the gas is expected to be atomic H I, which is a poor radiator. As a res ult, a
collapsing cloud will tend to heat up to very high temperatures. The Jeans
mass then becomes very large compared to a stellar mass and fragmentation
will not occur. In addition, the discussion of §4 suggests that only stars with
very high mass will be sufficiently powerful to separate themselves from their
environment.
An important process which prevents the above scenario from occlll"ring
is the formation of H2 molecules. In particular, the three body reactions

H + H + H ---+ H2 + H,
H + H + H2 ---+ H2 + H2,
are surprisingly efficient (Palla, Salpeter, & Stahler 1983; Lepp & Shull
198~). As a result, cloud temperatures are much closer to present day values.
These authors estimate that the temperatures can be as low as 100 K and
that the Jeans mass can be as low as 0.1 Me. Since this value of the Jeans
mass is lower than the expected masses of the forming stars, the Jeans mass
probably does not detemine stellar masses in primordial times.

5.2. PROTOSTELLAR EVOLUTION WITH ZERO METALLICITY

We now consider the formation of stars in an environment which is essentially

devoid of heavy elements. First let us consider the protostellar phase of
evolution where a protostar is deeply embedded in an infalling evelope of
primordial gas. The basic picture is the same as that for present day star
formation except that (1) the gas is of primordial composition, and (2)
the infall rate is expected to be higher. Stahler, Palla, & Salpeter (1986a)
have performed this calculation using a representative infall rate of if =
4.4 X 10- 3 Me yr- 1 • They have followed the evolution of the forming star
as its mass increases from 0.01 M0 up to 10.5 Me. The principal result
of this work is that the radial sizes of primordial stellar objects are much
larger than those of their present day counterparts during the protostellar
phase of evolution. In particular, a 1 Me stellar object will have a radius
of 47 Re, approximately 10 times larger than a Population I object of the
same mass (Stahler, Shu, & Taam 1980). This difference is relatively easy
to understand: The structure of a stellar object is determined by the usual
stellar structure problem with a highly unusual outer boundary condition -
the surface of the star is blasted by the infall. When the infall time scale
is short compared to the cooling time of the newly arriving material on
 525

the stellar surface, the star gains mass faster than the stellar structure can
adjust itself. As a result, the star swells up to a relatively large radius.
In summary, a large infall rate leads to a large stellar radius during
the protostellar phase. Thus, the main difference between primordial and
present-day star formation is the value of the infall rate it, which is highly
uncertain in the primordial case. We note, however, that if the infall rate
is higher for primordial star formation, then all indications suggest the for-
mation of more massive stars. Since the infall rate is larger, a more massive
star is required in order to reverse the infalL In addition, the larger infall
rate produces a larger stellar radius and hence a cooler central temperature.
This swelling of the star delays the onset of nuclear processes in the stellar
interior and thus helps further delay the reversal of the infall. In addition,
radiation pressure is unable to help reverse the infall in the primordial case
because of the low metallicity of the gas: the cross section of interaction be-
tween photons and the infalling envelope is smaller by orders of magnitude
than in the case of Population I abundances. Finally, we note that because
of the low metallicity (and hence zero dust opacity), the spectral appear-
ance of these objects should be quite different from that of their present day
counterparts.

5.3 . PRE- MAI N-SEQUENCE EVOLUTION WITH ZERO METALLICITY

We now consider the pre-main sequence phase of primordial stellar evolution

(see Stahler, Palla, & Salpeter 1986b). As discussed above, stars forming
under primordial conditions will generally be larger in radius (for a given
mass) than their present-day counterparts. However, these stars will contract
fairly quickly to stellar configurations such that they follow conventional
pre-main-sequence tracks in the H-R diagram. Basically, a swelled-up star as
produced by infall at a large rate will have a very short thermal timescale. As
a result, these stars quickly erase all memory of their initial conditions and
asymptotically follow conventional stellar evolution tracks (Stahler, Palla,
& Salpeter 1986b).

5.4. SUMMARY OF PRIMORDIAL STAR FORMATION

If primordial star formation proceeds in a qualitatively similar manner to

present day star formation, then the two cases can be concisely compared.
The initial conditions for star formation in primordial times are likely to be
quite different from the conditions in molecular clouds today. The temper-
ature is likely to be higher; the rest of the conditions are largely unknown.
Once collapse begins, however, the differences become somewhat smaller.
The most important physical variable during the collapse phase is the mass
526

infall rate M, which is likely to be higher in primordial times (provided that

the star formation scenario still has a well defined M as in present day star
formation). With a larger value of M, the stellar radius will be larger (for
a given mass) and the luminosity smaller, but the overall picture will be
similar to the present day case.
Pre-main-sequence stars in the primordial epoch quickly evolve onto (nearly)
standard PMS tracks in the H-R diagram. Thus, for a forming star of a given
mass, the differences between the primordial epoch and the present epoch
generally diminish with time. The main issue is therefore the spectrum of
masses of forming stars - the initial mass function - which we discuss in the
following section.

5.5. THE PRIMORDIAL INITIAL MASS FUNCTION

Since the primordial initial mass function is largely unknown, we must dis-
cuss many different possible scenarios. We stress, however, that any remarks
(including those presented here) about the primordial IMF are highly spec-
ulative.
We first consider the "minimal extension" of the theory of present-day
star formation. In this case, the process of star formation proceeds in a
qualitatively similar manner to the case of present day star formation. The
arguments presented above suggest that the IMF will also be qualitatively
similar, but skewed toward the formation of higher mass stars. In other
words, there will still be a roughly power-law distribution of stellar masses,
and higher mass stars will still be rarer than lower mass stars, but the
power-law index J.L of the distribution (see equation [4.1]) will be lower.
In addition to the "minimal extension hypothesis" outlined above, many
alternate viewpoints have been expressed in the literature. In what follows,
we briefly discuss the two limiting cases: (1) the initial stellar generation
was composed of very high mass stars (VMOs), and (2) the initial stellar
generation was composed of very low mass stars (brown dwarfs) . If either of
these limiting cases is realized, the star forming process is likely to be quite
different from that of today.
We first consider the interesting case in which the first stellar generation
consists of the so called "very massive objects" (VMOs) . These objects are
stars of very high mass, M. = 10 2 - 105 M0 (Bond, Arnett, & Carr 1984).
Since these objects are relatively efficient at converting their rest mass energy
into radiation through nuclear reactions, VMOs can have a rather large
impact on their immediate environment and on cosmology as a whole (Carr,
Bond, & Arnett 1984). These objects have been invoked to provide dark
matter, to reionize the universe, and to produce spectral distortions of the
cosmic microwave background (CMB). However, in light ofthe recent COBE
 527

observations of the CMB spectrum (see, e.g., the review by Bennett in this
volume), the abundance of these VMOs is now more tightly constrained (see
Bond, Carr, & Hogan 1986, 1991).
Finally, we discuss the possibility that the first stellar generation con-
sisted of brown dwarfs, i.e., very low mass stars which a.re not capable of
burning hydrogen. Under normal circumstances the minimum mass which
is capable of hydrogen burning is '" 0.08 Mev (e.g., Burrows, Hubbard, &
Lunine 1989), although this limit is somewhat modified for zero metallicity
objects (see Salpeter 1992). The possibility of a primOIdial generation of
brown dwarfs is of great interest because such objects can provide a bary-
onic dark matter for galactic halos (e.g., Adams & Walker 1990j Salpeter
1991, 1992). Notice that big bang nucleosynthesis implies that the allowed
range of OB (the ratio of the mass density in baryons to the critical density
of the universe) must be 0.04 ::s nBh 2 ::s 0.08 (see, e.g., Kawano, Schramm,
& Steigman 1988). Thus, the total mass density in baryons must be greater
than that of visible matter and, consequently, at least some baryonic dark
matter is required by big bang nucleosynthesisj brown dwarfs are a natu-
ral candidate. Although it is difficult to form extremely low mass objects
through the current star formation scenario (§2), recent work (see Lenzui,
Chernoff, & Salpeter 1992) suggests that if the primordial star forming en-
vironment was in a Jeans-stable configuration, then star formation could
occur with a very small mass infall rate M. Under these conditions, the
formation of large numbers of small stars (brown dwarfs) becomes possible.

6. Sum.m.ary and Discussion

In this review, we have outlined the current theory of star formation in

present day molecular clouds. The theory is successful in predicting the ra-
diative signature of forming stars oflow mass. In particular, the theory pre-
dicts both spectral distributions and spatial distributions of emission which
are in good agreement with observations. The theory suggests that stars, in
part, determine their own masses through the action of powerful outflows.
An important unresolved issue is the mechanism which produces these out-
flows. The presence of circumstellar disks, which often must have substantial
activity, can explain the spectral appearance of newly formed stars. Another
unresolved issue is the mechanism which leads to accretion through these
disks. In spite of its successes, this theory still must be extended to include
the formation of high mass stars and the formation of binary companions.
Finally, we note that we are (unfortunately) still quite far from being able to
calculate and predict the initial mass function for forming stars as a function
of environmental parameters.
528

We have also reviewed our current understanding of primordial star for-

mation. In this case, the initial conditions for the star formation process are
largely unknown, but are likely to be quite different from those of today.
Once stars begin to form, however, the differences between primordial ob-
jects and present-day stars (for a g-i'uen stellar mass) generally decrea:;e with
time. Thus, the key issue for primordial star formation is the IMF, which is
rather poorly understood and poorly constrained at present.
In the environments which produce the very first generation of stars, the
magnetic field could be very much smaller than in present day clouds . In
this case, star formation could proceed in a much different manner. With
no magnetic fields for support, clouds may be able to collapse as a whole
and fragment. An important question is then what happens to the angular
momentum in these environments, since the magnetic field is not strong
enough to provide efficient magnetic braking of the rotation.
For star formation in cooling flows, recent claims suggest that the IMF
favors the formation of low mass stars (relative to the usual IMF). On the
other hand, other recent claims suggest that in star burst galaxies the IMF
favors the formation of high mass stars. One possible (and highly specula-
tive) explanation for these puzzles could be the presence and behavior of
the magnetic field. If a substantially large magnetic field (from the galaxy)
is present in a cooling flow, then cloud material can be squeezed together
as it falls in. The magnetic flux to mass ratio (within the infalling material)
can become fairly large and the formation of low mass stars is then favored.
On the other hand, in the case of star burst galaxies, the magnetic field will
act quite differently. In these galaxies, collisions of molecular clouds are gen-
erally thought to explain the observed activity. If two clouds (each with an
embedded magnetic field) collide, then the product cloud will have a flux to
mass ratio which is smaller than (or, at best, equal to) the flux to mass ratio
of the original clouds (Shu 1987). Hence, the product cloud can suddenly
find itself with a rather large mass, a much reduced magnetic flux ratio,
and a somewhat elevated temperature (from energy dissipation in shocks).
These conditions are expected to favor the formation of high mass stars.
We conclude this discussion by noting the "coincidence" that almost all
stars observed today have masses within an order of magnitude of the Chan-

r( 2f/2
drasekhar mass MCh given by

MCh = 0.2 ( Z he mp ,
A Gmp
where Z and A are the atomic number and weight of the ions in the star.
This mass scale depends only on fundamental constants (h, G, etc.) and has
a value of", 1.4 M0 (Chandrasekhar 1939; see Lieb 1990 for a recent review).
This mass scale represents the maximum mass that can be supported against
gravity by quantum mechanical degeneracy pressure. The fact that observed
 529

stars have M. = O(MCh) can be either due to observational selection or to

formation limits. Stellar objects with masses much greater than MCh have
a large "excess" of gravity and are (relatively) short-lived. Objects with
masses much less than MCh have a large "excess" of pressure support and
evolve very slowly; they are relatively inactive and hence very faint. Thus,
observational selection favors M. ,. . ., MCh. On the other hand, stars help
determine their masses through stellar winds, which are the result of stellar
physics. Stars with M. ~ MCh are relatively inactive and have trouble
producing winds strong enough to halt the infall. Stars with higher masses
are so active that, except under rather special circumstances, we expect
winds to halt the infalliong before the condition M. ~ MCh is realized.
Thus, the star fomation process also favors stars with M. ,. . ., MCh.

Acknowledgements
I would like to thank F. Palla, S. Shore, S. Stahler, and R . Watkins for
helpful discussions regarding the preparation of this review. This work was
supported by NASA Grant Nos. NAGW-2802 and NAGW-3121 and in part
by funds from the Physics Department at the University of Michigan.

References
Abt, H. A. 1983, A R A & A, 21, 343
Adams, F. C. 1990, ApJ, 363, 578
Adams, F. C. 1991, ApJ, 382, 544
Adams , F. C., and Benz, W . 1992, in Complementary Approaches to Binary and Multiple
Star Research, IAU ColI. No. 135, ed. H. McAlister (Provo: Pub. A. S. P .), in press
Adams, F . C ., Emerson, J. P., and Fuller, G. A. 1990, ApJ, 357, 606
Adams , F . C., and Fatuzzo, M. 1993 , ApJ, in press
Adams, F. C., Lada, C . J., and Shu, F. H. 1987, ApJ, 321, 788 (ALS)
Adams, F. C., Lada, C. J. , and Shu, F . H. 1988, ApJ, 326,865
Adams, F. C., and Lin, D.N.C. 1992, in Protostars and Planets III, ed. E. Levy and M.
S. Mathews (Tucson: University of Arizona Press), in press
Adams, F . C., Ruden, S. P ., and Shu, F. H. 1989, ApJ, 347,959 (ARS)
Adams, F . C., and Shu, F. H . 1985, ApJ, 296, 655
Adams, F. C., and Shu, F. H. 1986, ApJ, 308, 836
Adams, F . C., and Walker, T. P. 1990, ApJ, 359, 57
Appenzeller, I., and Mundt, R . 1989, Astron. Astrophys. Rev., 1, 291
Arons, J., and Max, C. 1975, ApJ Letters, 196, L77
Bally, J ., and Lada, C. J . 1983, ApJ, 265 , 824
Beckwith, S., Sargent, A. I. , Chini, R., and Gusten, R. 1990, AJ, 99, 924
Benz, W. 1990, in Numerical Modeling of Nonlinear Stellar Pulsations: Problems and
Prospects (Dordrecht: Kluwer Academic Press)
Benz, W., Bowers, R. L., Cameron, A.G .W ., and Press, W. H. 1990, ApJ, 348, 647
Blitz, L. 1992, in Protostars and Planets III, ed. E. Levy and M. S. Mathews (Tucson:
University of Arizona Press), in press
Bond, J. R., Arnett, W . D., and Carr, B . J. 1984, ApJ, 280, 825
Bond, J . R., Carr, B . J., and Hogan, C . 1986, ApJ, 306,428
Bond, J. R., Carr, B. J., and Hogan, C. 1991, ApJ, 367, 420
Burrows, A., Hubbard, W. B., and Lunine, J. I. 1989, ApJ, 345, 939
530

Butner, H. M., Evans, N. J., Lester, D. F., Leneault, R. M., and Strom, S. E. 1991, Ap.
J., 376, 636
Carr, B. J., Bond, J. R., and Arnett, W. D. 1184, ApJ, 277,445
Cassen P., and Moosman, A. 1981, Icarus, 48,353
Chandrasekhar, S. 1939, An Introduction to the Study of Stellar Structure (New York:
Dover)
Draine, B. T., Roberge, W., and Dalgarno, A.1983, ApJ, 270, 519
EI:megreen, B. G. 1979, ApJ, 232, 729
El:megreen, B. G. 1990, ApJ, 361, L77
Goodman, A. A. 1990, PhD Thesis, Harvard University
Goodman, A. A., Crutcher R. M., Heiles, C., Myers, P. C., and Troland, T. H. 1989, ApJ,
338, 161
Heiles, C. H. 1987, in Physical Processes in Interstellar Clouds, ed. M. Scholer (Dordrecht:
Reidel), p. 429
Houlahan, P., and Scalo, J. 1992, ApJ, 393, 112
Kawano, L., Schramm, D. N., and Steigman, G. 1988, ApJ, 327, 750
Kenyon, S., and Hartmann, L. 1987, ApJ, 32S, 714
Lada, C. J. 1985, A R A & A, 23,267
Lada, C. J., and Shu, F. H. 1990, Science, 1111, 1222
Ladd, E. F., Adams, F. C., Casey, S., Davidson" J. A., Fuller, G. A., Harper, D. A., Myers,
P. C., and Padman, R. 1991, ApJ, 382, 555.
Lenzuni, P., Chernoff, D. F., and Salpeter, E. E. 1992, ApJ, 393, 232
Lepp, S., and Shull, J. M. 1983, ApJ, 270, 57.
Lieb, E. H. 1990, Bull. AMS, 22, 1
Lin, C. C., and Lau, Y. Y. 1979, Studies in Applied Math., 60, 97
Lizano, S., and Shu, F. H. 1989, ApJ, 342, 83t.
Miller, G. E., and Scalo, J. M. 1979, ApJ Sup~., 41, 513
Mouschovias, T. Ch. 1976, ApJ, 206, 753
Mouschovias, T. Ch. 1978, in Protostars and Planets, ed. T. Gehrels (Tucson: University
of Arizona Press), p. 209
Mouschovias, T. Ch. ,and Spitzer, L. 1976, A,J, 210, 326
Myers, P. C. 1985, in Protostars and Planets II, ed. D. C. Black and M. S. Mathews,
(Tucson: Univ. Ariz. Press), p. 81
Myers, P. C., Fuller, G. A., Mathieu, R. D., Bcichman, C. A., Benson, P. J., Schild, R. E.,
and Emerson, J. P. 1987, ApJ, 319, 340
Myers, P. C., and Goodman, A. A. 1988, ApI. 329, 392
Myers, P. C. 1991, in Fragmentation of Molec.ar Clouds and Star Formation (IAU Sym-
posium 147), eds. E. Falgarone and G. DU'IIert (Dordrecht: Kluwer), in press
Myers, P. C., and Fuller, G. A. 1992, ApJ, 391, 631
Najita, J. 1992, PhD Thesis, University of Caifornia, Berkeley
Nakano, T. 1985, Pub. A.S.J., 37,69
Norman, C., and Silk, J. 1980, ApJ, 238, 158
Ostriker, E. C., Shu, F. H., and Adams, F. C.1992, ApJ, in press
Palla, F., Salpeter, E. E., and Stahler, S. W. 1983, ApJ, 271, 632
Palla, F., and Stahler, S. W. 1991, ApJ, 375, %88
Rydgren, A. E., and Zak, D. S., 1987, Pub. A. S. P., 99, 141
Rucinski, S. M., 1985, A. J., 90, 2321
Salpeter, E. E. 1955, ApJ, 121, 161
Salpeter, E. E. 1991, Mem. S. A. It., 62, 897
Salpeter, E. E. 1992, ApJ, 393, 258
Scalo, J. M. 1986, Fund. Cosmic Phys., 11, 1
Scalo, J. M. 1987, in Interstellar Processes, «1. D. J. Hollenbach and H. A Thronson
(Dordrecht: Reidel), p. 349
Shu, F. H. 1977, ApJ, 214, 488
Shu, F. H. 1983, ApJ, 273, 202
 531

Shu, F. H. 1985, in The Milky Way: IAU Symposium No. 106, ed. H. van Woerden , W .
B. Burton, and R . J. Allen (Dordrecht: Reidel), 561
Shu, F. H. 1987, in Star Formation in Galaxies, ed . C. PeIsson (NASA CPo 2466), p . 743
Shu, F. H. 1992, Gas Dynamics (Mill Valley: University Science Books)
Shu, F. H., Adams, F. C ., and Lizano, S. 1987, A R A & A, 25, 23
Shu, F. H., Lizano, S., and Adams, F. C. 1987, in Star Forming Regions, IAU Symp. No.
115, ed. M . Peimbert and J. Jugaku (Dordrecht: Reidel), p. 417
Shu, F. H., Lizano, S., Ruden, S. P., and Nagita, J. 1988, ApJ, 328, L19
Shu, F. H., Tremaine, S., Adams, F . C., and Ruden, S. P. 1990, ApJ, 358, 495 (STAR)
Stahler, S. W. 1983, ApJ, 274, 822
Stahler, S. W . 1988, ApJ, 332, 804
Stahler, S. W . , Palla, F., and Salpeter, E. E. 1986a, ApJ, 302, 590
Stahler, S. W ., Palla, F., and Salpeter, E . E. 1986b, ApJ, 308, 697
Stahler, S. W ., Shu, F. H., and Taam, R. E., 1980, ApJ, 241,637
Terebey, S., Shu, F . H., and Cassen, P. 1984, ApJ, 286, 529
Tomley, L., Cassen, P ., and Steiman-Cameron, T. 1991, ApJ, 382, 530
Toomre, A. 1964, ApJ, 139, 1217
Walker, C . K ., Adams, F. C., and Lada, C . J. 1990, ApJ, 349,515
Wiseman, J . J ., and Adams, F. C. 1992, in preparation
Wolfue, M. G., and Cassinelli, J. P. 1986, ApJ, 310, 207
Wolfue, M . G ., and Cassinelli, J . P . 1987, ApJ, 319, 850
Wood, D.O.S., Myers, P. C., and Daugherty, D. A. 1992, in preparation
Zinnecker, H., McCaughrean, M. J., and Wilking, B . A. 1992, in Protostars and Planets
III, ed. E . Levy and M. S. Mathews (Tucson: University of Arizona Press), in press
Zuckerman, B., and Palmer, P. 1974, A R A & A, 12, 279
 THE BIRTH OF STARS IN GALAXIES

ROBERT C. KENNICUTT, JR.

Steward Observatory, University of Arizona

Abstract. Ow: empirical understanding of galaxy formation and evolution rests on a

combination of observations of the stellar populations in the nearest galaxies, direct ob-
servations of galaxies at cosmological lookback times, and integrated measurements of
galaxies at the present epoch. This review describes several examples of recent observa-
tions ,which offer important clues about the history of star formation in galaxies.

1. Introduction
The conference organizers have asked me to combine a talk on the evolution
of nearby galaxies with a r~view of conference highlights relating to galaxy
evolution in general (Jill Bechtold will review quasars and the intergalactic
medium). Rather than attempt a comprehensive conference review, I will
discuss what recent observations of nearby galaxies reveal about the history
of star formation in galaxies, and integrate that with some of the results
presented at this meeting, most of them pertaining to more distant galaxies.
Several approaches can be used to study the evolutionary histories of
galaxies. One is to study the resolved stellar populations in our own Galaxy
and Local Group, where in principle detailed fossil records of galaxy forma-
tion and evolution can be reconstructed. The distributions of stellar age and
abundance in the Galactic neighborhood offer the ultimate tests of the the-
ories of galaxy formation and evolution. Since this area was discussed only
briefly this week, I will discuss a few of the most interesting recent results
here.
Another direct approach is to observe galaxies at successive cosmological
lookback times, and reconstruct the history of the stellar birthrate directly.
This exciting field has been discussed extensively by several speakers, and
was the sole subject of the panel discussion. Consequently, I will limit my
own discussion to summarizing the main conclusions and drawing parallels
to what we see in nearby galaxies.
A third approach is to measure the star formation properties, stellar
populations, and star formation histories of nearby galaxies (z ~ 0.1). This
approach is less direct than the others, since we only observe galaxies at the
present epoch, and we obtain only rudimentary information on the birthrate
history of any single object. However, these observations provide several
important pieces of the puzzle. They eaable us to delineate the evolutionary
nature of the Hubble sequence, to study the systematic effects of galactic
environment on evolution, and to identify local analogs to primeval galaxies
of different types. Perhaps most importantly, these data provide information
on the physical relationships between the stellar birthrate and the physical

533
J, M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, 533-558.
© 1993 Kluwer Academic Publishers.
534

properties of galaxies and their interstellar media. Such observations offer

the hope, at least over the long term, of understanding not only how galaxies
evolve, but why they evolve as they do, in the context of an empirically based,
physically deterministic model of galactic evolution.
My approach in this review will be to present, in §3-5, a sampling of
recent results from each of these subfields. My objective is not to provide a
detailed review of any individual area, but rather to illustrate by example
the broad range of observational constraints which are available to us. The
results will also illustrate how these nearly independent approaches often
point to similar pictures of galaxy formation and evolution.

2. Theoretical Background and Expectations

At this school we were blessed with a comprehensive review of current theo-
retical ideas on galaxy formation and evolution, in talks by Mitch Begelman,
Ed Bertschinger, Dick Bond, Gustavo Bruzual, Marc Davis, Gus Evrard,
Lars Hernquist, and Rosemary Wyse. Before proceeding to the observations
of star formation in galaxies, it is useful to review what the star formation
rates and histories of galaxies are predicted to be in the different galaxy
formation/ evolution scenarios.
As recently as a few years ago, many searches for primeval galaxies and
starburst galaxies were predicated on a much simpler collapse model for
galaxy formation. In this model, similar to the "ELS scenario" o}ltlined in
the famous paper by Eggen, Lynden-Bell, & Sandage (1962), the formation
of galactic spheroids was a rapid process, with the bulk of the star formation
occurring on a dynamical time scale (of order 108 yr). From: a theoretical
point of view, the model offers an elegant simplicity, as all of the principal
steps involved in galaxy formation- the accretion of mass, the accretion of
gas, star formation, metal enrichment- occur almost simultaneously, over
the brief free-fall collapse time scale. From an observer's point of view, the
collapse model generates glorious protogalaxies. (Throughout this discussion
I will reference the SFRs to a roughly L* galaxy of mass M*j for lower
mass galaxies the SFRs will scale down proportionally.) If we consider, as
an example, an M* proto-spheroid with a gaseous mass of 2 x 1011 Me, a
characteristic star formation rate (SFR) during the formation phase would
be of order:

2 X 1011 Me 1
SFR '" 8 '" 1000 Me yr- .
2 X 10 yr
It is easy to demonstrate, using models of the type discussed by Bruzual,
that such a protogalaxy would have bolometric and Ha luminosities of Lbol '"
10 13 L e , MB ~ -25, and L(H a) ~ 1043 erg s-1. The possibility of such
protogalactic mega-starbursts motivated an entire generation of searches for
 535

x>-
(0 )
<l
...J
<t 30
Cl

0
I::;
0
-2
<l 10
a::
0 3
"-
a::
<t
w
>- sO'\
a:: 0
w
Cl.

-'"
a:: \oJ LARGER DISSIPATION (b)

'" '"
...=..~
~

a::
<t
...J
0
~
W
~ 0 .5
a::
:r:
....
a::
iii Sc
..J Sb
~ 0 So
0 tic
I- DISK fORMATION
8ULGE
I I
0 .... 109
TIME-

Fig. 1. Schematic depiction of the evolution of the stellar birthrate from Sandage
(1986) . Note the contrast in time scales and star formation rates for the spheroid
and disk components, and the changes in star formation history with Hubble type.

primeval galaxies in the 1970's and early 1980's. The failure to detect them
implies either that galaxies do not form so rapidly, or that the collapses
occur beyond the limiting redshifts of current surveys.
A more detailed depiction of the star formation histories predicted in this
picture, including later disk - star formation and the variation with Hubble
type, is shown in Figure 1, taken from Sandage (1986). On an absolute
scale, one expects peak SFRs of order 100-1000 M0 yr- 1 in the largest
proto-spheroids, and mean rates of order 1-10 M0 yr- 1 in spiral disks.
In the simplest form of this scenario, at least, the transition between a
rapidly collapsing spheroid and a steadily evolving disk is quite distinct,
and distinguishing between these different stages of galactic evolution at
high redshift should be relatively straightforward.
This collapse picture contrasts markedly, however, with what is expected
in the hierarchical "bottom-up" picture of galaxy ·formation presented at
this conference. In that scenario galaxy formation consists of several physi-
cally and temporally distinct steps. First the protogalactic mass fragments,
536

mostly dark matter, must condense; then the gas must cool and condense
further; finally stars can form , either from the collapse of clouds within in-
dividual fragments, or in mergers of fragments. These complications have
important consequences for observing stat' formation in primeval galaxies.
In many cases, the major epoch of star fo,rmation may be delayed to late
cosmological times and consequently lower redshifts. The formation of galac-
tic spheroids, rather than involving a single starburst of order 100-1000
M0 yr- 1 , may consist of several smaller collapse and merger events, OCClli'-
ing over a span of billions of years. In such cases the peak SFR associated
with spheroid formation will be of order the fragment mass (in stars) divided
by its formation time scale,

This may be barely distinguishable from the steady-state SFRs in disks (of
order several M0 yr- 1 ), especially when extrapolated back to early look-back
times. For example, in the recent numerical simulations of galaxy formation
by Katz (1992), a typical spiral does not reach peak luminosity until the disk
forms, at z f'V2. For more detailed discussions of the observable properties
of protogalaxies in this picture see Baron & White (1988) and Katz (1992).
The point of this admittedly sketchy comparison is to demonstrate the
wide range of star formation properties of galaxies expected in these different
theoretical scenarios. I have deliberately chosen two extreme versions of
the collapse and merger pictures to illustrate the range of possibilities; the
truth may well lie somewhere between these extremes. As observers, we are
interested in testing to what extent these scenarios apply to our universe,
and in isolating examples or analogs to the various galaxy formation and
evolution stages.

3. Resolved Stellar Populations in the Local Group (z :S 0.001)

As was emphasized by Rosemary W yse in her review, observations of the

ages, kinematics, and abundances of individual stars in the Milky Way (or
other galaxies) should allow us to construct the most complete picture pos-
sible for the formation and evolution of these galaxies. Several recent instru-
mental developments, most significantly the availability of CCD detectors,
multi-object spectrographs, and sophisticated PSF -fitting reduction software
have led to major advances in our ability to probe the fossil record in the
Galaxy and other Local Group members. Here I mention just a sampling of
recent results, to illustrate the variety of observations which are being used,
and to emphasize the fundamental character of the questions that are being
raised.
 537

3.1. THE GALAXY

Probably the most exciting (and controversial) work in the last few y~a.rs
concerns the distribution of stellar n.ges in the Galactic bulge and halo. ~tate­
of-the-art CCD photometry now makes it possible to obtain color-magnitude
diagrams for globular clusters with internal precision of 0.01 mag or better,
sufficient in principle to measure relative cluster ages with an accuracy of ",1
Gyr. This makes it possible to test whether the Galactic globular clusters
fOl"m a coeval system, and if not to measure the time scale for the formation
of the spheroid.
Several recent investigations show evidence for significant age differences
among the globular clusters. The most convincing results are based on mea-
surements of pairs of clusters with nearly identical compositions but strik-
ingly different horizontal branch morphologies- examples of the famous
"second-parameter problem." Figure 2 shows an example from one such
pair, NGC 288 and NGC 362, from Green & Norris (1990). The striking
difference near the main sequence turnoff (both clusters are fitted to the
same ridge line in Figure 2) corresponds to an age difference of 2-3 Gyr.
Age differences of 2-5 Gyr are seen in other cluster pairs (e.g., Buonanno
et al. 1990). Relative ages can also be measured from the turnoff vs. red gi-
ant branch color difference, or by fitting the horizontal branch luminosities
(e.g., VandenBerg et al. 1990; Chaboyer et al. 1992). These methods yield
age ranges of up to 5 Gyr, though some subsamples appear to be coeval.
There are some reports of systematic correlations of age with metallicity
and galactocentric radius (e.g., Chaboyer et al. 1992).
Taken at face value, these results suggest that there is ' a young tail to
the globular cluster age distribution, extending to 3-5 Gyr younger than
the oldest clusters, or perhaps even a dispersion in the formation time of
the entire cluster system. This would be consistent with the extended for-
mation picture for the outer halo by Searle & Zinn (1978). As emphasized
by Bolte (1992), however, there are several important caveats which must
be addressed. Most recent studies have concentrated on the most extreme
examples of second-parameter-effect clusters, which may not be representa-
tive of the entire cluster ensemble. Age comparisons for clusters with widely
differing compositions are subject to errors from variations in other param-
eters or to uncertainties in the theoretical models. It may be premature to
conclude from the existing data that "the spheroid formed over a period
of several Gyr," but it seems likely that observations over the next several
years will resolve this question once and for all. For a more detailed (and
competent) review of this field, I refer the reader to Bolte (1992).
Chemical nucleosynthesis offers an entirely independent constraint on the
formation time scales for the Galaxy. As an illustration, Figure 3 shows the
differential abundance distributions of oxygen and iron in nearby field stars,
538

~CC <BB NCC 362

15 15

16 . 16

17 17

B 18 B lB

19 19 ~. " .:..:", , ' "

':~'~~~. .
20

21 21

0 .2 0 .6 1 .0 1.4 1.8 2 .2 0 .2 0 .6 1. 0 1.4 1.8 2 .2

B-R B-R

Fig. 2. Color-magnitude diagrams (Green & Norris) for the Galactic globular clus-
ters NGC 288 and NGC 362. The empirical fit to the NGC 288 ridge line, shifted
slightly in color and magnitude to account for the differences in reddening and
distance to the clusters, is superimposed on both diagrams. The diagrams also il-
lustrate the precision that can be obtained with current CCD photometry and PSF
fitting techniques.

from a review, by Wheeler et a1. (1989). If °

and Fe were formed together
in the Galaxy, we would expect the [Fe/O] ratio to be identically zero (Le.,
the solar value) everywhere. Instead, Figure 3 shows a distinct transition
near [O/H] = -0.5 (or [Fe/H] = -I). Below that level, iron is consistently
underabundant with respect to the solar Fe/O ratio, but above [O/H] =
-0.5, [Fe/O] increases smoothly toward the solar value. The same trend is
seen when the abundances of the "alpha-addition" elements Mg, Si, Ca, and
Ti are compared to iron. Similar abundance patterns are also seen in the
Galactic globular cluster system (Minniti et a1. 1993).
Oxygen and the alpha-addition elements are thought to arise predomi-
nantly from Type II supernovae, with massive and short-lived progenitors,
while iron is preferentially produced in older Type Ia supernovae. Conse-
quently the transition in [Fe/O] vs [O/H] may represent the delayed Fe
enrichment of the Galaxy from the Type Ia supernovae, at [Fe/H] ~-1. If
this interpretation is correct, then the lifetimes of the supernova progenitors
provide an independent clock for the enrichment time scale for the metal-
poor halo. This time is sensitive to the nature and masses of the supernova
progenitors, as well as to the nucleosynthetic yields of the different supernova
types. Analyses by Wyse & Gilmore (1988) and Smecker-Hane & Wyse (this
meeting) suggest a time scale of ::;1 Gyr. This could imply that the bulk of
the metal-poor component of the spheroid formed within a few dynamical
time scales, or it may be tracing the enrichment times within individual
 539

.5
-
o --. -... . ... . .... - _._ .. . . -_..............• .......... - -_. - --• ~t-

..
.••,-.-.: .
"t'-.~ --

.. ..
A ........ .

.
• . .... :"t ... ",,,,
"' "' ..... '""'.
- '" MIt""
-. 5
•. .
- . ...'" '" '"
0
-
0
-• '"
·1
o Bessel and Norris 1987
0 o M~ain 1985
0 '" A arson at a11988
- Granon, Ortolani 1986
• Barbuy 1988
-1.5

·3 ·2 ·1 0
lOtH)

Fig, 3. Comparison of oxygen and iron abundances in field stars, with oxygen taken
as the independent metallicity indicator. Figure is taken from Wheeler, Sneden, &
Truran {1989}.

protogalactic fragments . Regardless of the final interpretation, the example

serves to illustrate how chemical abundance constraints can provide crucial
clues to galactic formation and evolution.
Observations of several stellar age tracers in the Galactic disk provide a
fairly self-consistent history. The most detailed histories come from isochrone
fitting of F dwarfs (Twarog 1980; Carlberg et al. 1985) and from chromo-
spheric dating of K giants (Barry 1988). Both data sets indicate that the
stellar birthrate in the disk, when averaged over 1 Gyr time scales, has been
nearly constant over its age, probably varying by no more than a factor
of 2-3 from its current value. Otherm less direct age tracers, including the
white dwarf luminosity function, lithium dating, and the continuity of the
local IMF near the main sequence turnoff, also indicate a relatively constant
disk birthrate. Important questions that remain unresolved include the con-
stancy of the disk birthrate over time scales shorter than a Gyr, and the
age of the oldest disk stars relative to the spheroid. See Scalo (1986) and
Kennicutt (1992c) for complete discussions of these observations.

3 .2 . OTHER LOCAL GROUP GALAXIES

It is now possible to extend the same types of stellar population studies to
the other galaxies in the Local Group. To date, the most complete studies
540

have been made of the Magellanic Clouds and several dwarf spheroidal galax-
ies. These observations reveal star formation histories which are strikingly
different from either the spheroid or disk of our own Galaxy.
The LMC has been the object of several investigations, as summarized
in Stryker (1984). Photometry of field stars in the main body of the LMC
indicates that most of the stars lie in the age range 0-5 Gyr. The stars in
the outer halo of the LMC appear to have a much broader distribution in
age, with a peak age of several Gyr, but with both older and younger stars
present as well. The LMC bar is a relatively young structure; star formation
was initiated there in the past 3 Gyr and continues to the present (Hardyet
al. 1984). Taken together, these studies indicate that star formation in the
LMC commenced at times comparable (but not necessarily identical) to the
formation of the Pop II component of the Galaxy, but the suJ>sequent star
formation was quite different from the Galaxy, peaking only in the past.few
Gyr.
The LMC contains a population of massive star clusters ranging in age
from a few million years to over 12 Gyr; these can be used to construct an
independent picture of the star formation and chemical enrichment history
of the LMC. Figure 4 shows the age and abundance distributions for the
populous clusters, as compiled by Da Costa (1991). Both distributions are
remarkably bimodal. There is a distinct population of old, metal-poor globu-
lar clusters which resemble in most respects Galactic globulars, and a second
population of clusters with ages less than about 3 Gyr, and abundances at
least an order of magnitude higher than the old clusters. Only one cluster
in Figure 4 (and no more than a handful of other candidates in the LMC)
lie in the gap between 3 and ",,15 Gyr.
The striking bimodality in Figure 4 could reflect a peculiarity unique to
the LMC population - for example a collision with the Galaxy 3 Gyr ago,
which disrupted all but the most massive halo clusters. It is interesting that
the SMC clusters do not show such a strong bimodality in age or composi-
tion (Da Costa 1991). However, the correspondence of the LMC cluster age
distribution with the results of the field star studies suggests that the 3-15
Gyr age gap is at least partly real. Whatever the interpretation, it is clear
that the LMC experienced a star formation history that was different from
any of the components of the Milky Way.
The dwarf spheroidal galaxies in the Local Group tell yet a different story,
as reviewed recently by Da Costa (1992). Studies of the Sculptor, Fornax,
Ursa Minor, Carina, and Andromeda II galaxies show evidence (excepting
perhaps And II) for a composite stellar population, with both a very old (T Z
U Gyr) metal-poor component, and an intermediate age (roughly 5-10 Gyr)
population. In Carina, the system best studied to date, the intermediate
age component comprises over 80% of the entire stellar population (Mighell
1990). Yet none of these galaxies is forming stars today, and none appears to
 541

'- t ES0121-SC03

~-,-

~I

N
I

o 4 6 8 10 16 18
Age (Gyn)

Fig. 4. Age-metallicity relation for populous clusters in the LMC, reproduced from
a review by Da Costa (1991). Sample is restricted to clusters with ages determined
from main sequence turnoff photometry. The ages of the oldest clusters have arbi-
trarily been set to a constant value of 15±2 Gyr.

possess enough interstellar gas to form stars anyway. Again, we see evidence
for a dramatic change in the stellar birthrate over the past ",5 Gyr.
With such a variation between the star formation histories of the individ-
ual members of the Local Group- it is probably a fair statement that no
two systems studied to date have the same history- one must be cautious
about drawing general inferences, but some important trends are apparent.
While the schematic evolutionary picture illustrated in Figure 1 appears to
provide a realistic description of the history of our own Galaxy (and pre-
sumably M31 and M33), it bears little resemblence to the evolution of the
lower-mass members of the Local Group. Galaxy mass appears to be at
least as important a variable as Hubble type in dictating evolution. This is
nowhere more apparent than in the dwarf spheroidal galaxies. These sys-
tems, formerly classified as "dwarf ellipticals", share more in common with
dwarf irregulars than with giant elliptical galaxies. In addition to having
distinct histories on long time scales, these systems show indirect evidence
for starbursts being a more important factor in the evolution of the low-mass
systems.
It is tempting to associate the rapid evolution of the Local Group dwarfs
over the past 5 Gyr with the evolution seen in faint galaxy counts and red-
shift surveys, as has been suggested recently by van den Bergh (1992). The
542

high space densities of the faint blue galaxies are difficult to associate with
any present-day population (e.g., Cowie et al. 1993), but the parallels be-
tween the fossil histories revealed in many the Local Group galaxies and what
is seen at cosmologicallookback times of a few Gyr are unlikely to be en-
tirely coincidental. As discussed later, present-day Magellanic irregulars are
spectroscopically indistinguhhable from the z = 0.4 blue galaxies. It seems
likely that many nearby low-mass galaxies comprise the fossil remnants of
part of this extraordinary blue galaxy population. Applying modern observ-
ing techniques to galaxies like the Magellanic Clouds (much of the work cited
above was based on photographic plates!) could clarify the relationship, if
any, between these populations.

4. Distant Galaxies and Clusters (z 2:: 0.3)

The current excitement in this field was reflected at this meeting, with talks
by Ellis, Heckman, Lilly, and Spinrad at Monday's panel discussion, reviews
of distant clusters by Donahue and Mushotzky, and roughly 20 posters de-
voted to studies of distant galaxies. I will restrict my discussion to those
results which are most relevant to understanding the star formation histo-
ries of galaxies. For proper reviews by the leading players in this field, see
those by Cowie & Lilly, Dressler & Gunn, and Ellis in Kron (1990).
Much of the discussion at this conference has focussed on the ongoing
multi-object spectroscopic surveys of faint (B = 20 - 25) galaxies, and the
remarkable intermediate redshift "blue dwarf" population revealed by these
surveys. Figure 5 shows a dated but nevertheless instructive example, the
redshift distribution at B = 20 - 22.5 from Colless et al. (1990). See Ellis'
paper in this volume and Colless et a1. (1992) for the most recent work by
this group.
Figure 5 is dominated by the enigmatic blue dwarf population, but from
the perspective of understanding star formation histories, what we don't see
in Figure 5 is nearly as significant as what we do see. For example, we do
not see a high-redshift tail of galaxies extending beyond the "no-evolution
model" fit. If most massive (L* and brighter) galaxies were significantly
brighter at z = 0.5 - 1, their apparent magnitudes would fall into the B =
20 - 22.5 window, and they would show up in Figure 5. Their absence rules
out rapid luminosity evolution of massive galaxies over the past several Gyr
(Colless et al. 1992). Nor do we see, among the hundreds of galaxies that
have been detected in these field surveys to date, even a single example of
a high-redshift mega-starburst galaxy, of the sort envisioned in the rapid
collapse galaxy formation scenarios. Given the large volume over which a
1000 M0 yr- 1 starburst should be visible, this result suggests that such
starbursts are either very rare at z ::; 3, or they are highly obscurred in
the rest-frame blue and UV. At the other extreme, the absence of very low-
 543

.2 20 . 0~bJ~22.5
- No-evolution model
.15
,.......,
N
..........
!:: .1

.05

0
0 .2 .4 .6 .8
z
Fig. 5. Redshift distribution of galaxies in a spectroscopically complete sample
(Colless et aI. [1990])at B = 20 - 22.5.

redshift galaxies at B 2': 22 constrains the nature and subsequent evolution

of the z = 0.3 - 0.4 population.
The galaxies that do appear in Figure 5 include the infamous blue dwarfs,
relatively unevolved normal galaxies, AGN, and some spectroscopically pe-
culiar objects, including flat-spectrum galaxies and post-burst "E + A"
galaxies. The blue galaxies are remarkable in terms of both their space den-
sity, roughly 2-5 times the unevolving density for B ~ 21-24, and their
photometric properties, which appear to be roughly comparable to nearby
Markarian galaxies. Speculation about the nature of the faint blue galaxies
appeared in several talks and posters during the meeting. Producing objects
which are so bright, blue, and numerous as recently as z = 0.3-0.5, yet evolve
to obscurity by z =0-0.1 without leaving much of a trace in light or heavy
elements, is utterly problemmatic. The list of possibilities includes mergers
(e.g., Guiderdoni & Rocca-Volmerange 1990; Broadhurst et al. 1992), burst-
ing dwarf galaxies (e.g., Babul & Rees 1992), or an entirely new population
of galaxies that evaporate a.t very recent epochs (e.g., Cowie et al. 1991,
1993). Adopting an open cosmology (or worse) helps reduce the space den-
sities at high redshift. I refer the reader to the reviews cited above for more
detailed discussions. As emphasized by Cowie et a1. (1993) these objects
may well dominate the global star formation, light, and metal production in
the universe, so understanding their relationship to present-day galaxies is
544

a crucial prerequisite for understanding galaxy formation and evolution in

a broader context.
ill many of the scenarios cited above, the faint blue galaxies have little or
no evolutionary connection to today's dominant population of giant galax-
ies. If that proves to be the case, then how do we penetrate this clutter of
low-redshift dwarfs to study galaxies like our own at large look-back times?
Several approaches were discussed during the meeting. fufrared-selected field
surveys (e.g., Elston et al. 1991; Glazebrook 1991; Cowie et al. 1993) are
effective probes for red galaxies out to at least z = 1, though at faint mag-
nitudes even the K counts appear to become dominated by the blue dwarf
clutter. Studying distant galaxy clusters is another powerful approach (e.g.,
Dressler & Gunn 1990), at least for understanding the evolution of present-
day cluster members. The Mg II absorber selected samples, as reviewed here
by Steidel, appear to offer a powerful probe of L· field spirals. Optical and
infrared identification of distant radio source fields, as discussed here by
Spinrad, provide a virtually unique probe at very high redshifts. Gravita-
tionally lensed galaxies may provide another window to z ~ 1.
The enigma presented by the faint blue dwarf population illustrates what
is perhaps the singular limitation of distant galaxy surveys for unraveling
galaxy evolution. In principle, there should be nothing more direct than
surveying the universe at successive redshifts, and tracking the evolution of
galaxy populations with increasing look-back time. However, this approach
relies on our being able to connect an individual galaxy at each redshift
(or photometric class of galaxies) to its evolutionary counterpart at other
epochs. This is proving to be very difficult, even over the modest range of
look-back time between z = 0-0.3. The availability of expanded spectro-
scopic data sets, combined with high-resolution images from HST eventu-
ally may resolve this particular difficulty, but we should not be surprised
to encounter similar ambiguities at higher redshift. The long-term solution
may require integrating the statistical information on high-redshift galaxy
populations with the more detailed data for nearby gala.xies, along with
theoretical constraints on photometric and chemical evolution.

5. Star Formation Properties of Nearby Galaxies (z < 0.1)

The development of observational techniques for measuring star formation
rates (SFRs) in galaxies has made it possible to characterize the evolutionary
nature of the Hubble sequence, and to begin probing the physical and envi-
ronmental parameters which regulate the SFR. This subject was reviewed
at length at the Second Teton Conference (Kennicutt 1990), and I will not
repeat that discussion here. Instead, I discuss three specific areas which are
especially relevant to galactic evolution in a broader context, the statistics
of global star formation rates and histories in nearby galaxies, the physical
 545

nature of the star formation law, and local analogs to primeval galaxies.

5.1. GLOBAL STAR FORMATION PROPERTIES

As noted earlier, several empirical techniques for measuring SFRs in galax-

ies have been developed over the past decade. These include calibrations of
the UV continuum fluxes (1500-2500 A), far-infrared fluxes (20-300 /Lm),
Balmer emission-line fluxes, and population synthesis modelling of the UV
and visible colors. See Kenrucutt (1990) for a details and references to the rel-
evant literature. The most complete surveys to date come from Ha emission-
line fluxes (e.g., Kennicutt 1983; Romanishin 1990; Gavazzi et al. 1991), and
balloon or satellite-based UV fluxes (e.g., Donas et al. 1987; Buat 1992).
Data are currently available for some 300-400 galaxies, spanning types SO
to Irr, and luminosities of MB = -12 to -22. Intercomparison of these stud-
ies shows that the SFRs for galaxies in common reproduce to ±25% when a
common set of extinction corrections and IMFs ar.e adopted. Variations in
extinction and the IMF remain as the largest sources of systematic error in
these global SFR determinations.
The publication of full optical emission-line spectra for nearby galaxies
has made it possible to calibrate the [01l]A3727 emission line as a quantita-
tive star formation tracer (Gallagher et al. 1989; Kenrucutt 1992a). Varia-
tions in excitation an,d reddening make [011] a less precise SFR tracer than
Ha, but its flux or equivalent width can provide a useful substitute, at least
for comparisons of large samples. This is especially valuable for high-redshift
galaxies, where Ha is unobservable. Applying [011] in this way, either alone
or in conjunction with blue-UV continuum fluxes, makes it possible to ex-
tend the techniques developed for nearby galaxies to z 2 1.
The total SFR, normalized to unit luminosity or area, is a very strong
function of Hubble type. For a galaxy of constant blue luminosity (MB =
-21), the total SFRs integrated over all stellar masses range from about
0.1-1 M0 yr- 1 in SO/a-Sa galaxies to an average of about 10 M0 yr- 1 in
Sc-Irr galaxies (Kenrucutt 1983, Caldwell et al. 1991). Overall, the total
SFRs can range from below 0.001 M0 yr- 1 in extreme dwarf galaxies to
over 100 M0 yr- 1 in luminous starburst galaxies. Spatially-resolved Ha
observations show that this trend in total SFR is due to both an increase
in the number of star forming regions per unit area, and an increase in the
characteristic masses and luminosities of the individual regions (Kennicutt et
al. 1989). Note that the total SFR per galazy is not necessarily a monotonic
function oftype. As pointed out by Devereux & Young (1991), the mean FIR
luminosity (and probably the total SFR) per galaxy peaks at intermediate
Hubble types, decreasing for both very early (Le., SO) and very late (Sd-Irr)
galaxies. The distinction appears to be mainly due to the systematic change
in mean galaxy mass along the Hubble sequence, but other factors, such as a
546

systematic change in extinction or IMF with galaxy type, cannot be entirely

ruled out.
The changes in current SFRs along the Hubble sequence imply similar
changes in star formation histories. Kennicutt (1983), Gallagheret al. (1984),
Sandage (1986), and Larson (1991) have investigated the systematic behav-
ior of the birthrate history for spirals and irregulars, where reliable global
SFRs can be obtained. A useful parametrization is the ratio of the current
SFR to the average past SFR, often denoted as the birthrate parameter b.
This parameter is a strong function of type, increasing from ::;0.05 in Sa-
Sab galaxies to '" 1 in Sc galaxies. Note that these results are consistent with
the measurements ofthe Galactic disk (§3.1), where b ~ 1 as well. Irregular
galaxies are characterized by b", 1 on average (Gallagher et al. 1984), but
with a considerable dispersion among individual objects. In many Sc and Irr
galaxies b ~ 1, i.e., their current SFRs are much higher than in the past.
Since these high SFRs cannot be sustained indefinitely, these galaxies must
be experiencing large bursts of star formation.
The observed trends in birthrate parameter with type form the basis of
the evolutionary picture of the Hubble sequence first outlined by Roberts
(1963), and which is illustrated schematically in Figure 1. While this general
trend has been quite firmly established, several other important questions
remain unanswered. These include the radial variation, if any, in the star-
formation history, the temporal variation of the SFR in early-type disks,
and the role of starbursts in the evolution of galaxies of different masses.
For elliptical galaxies and bulges, which have not been discussed at all here,
the most fundamental questions about the duration and variation in the
star-formation histories remain to be addressed (see Bruzual's review in this
volume).

5.2. THE STAR-FORMATION LAW

As Judy Young emphasized in her review, one of the most fundamental con-
tributions that these kinds of observations can offer is a better understand-
ing of the physical mechanisms that regulate the global stellar birthrate in
a galaxy. There has been considerable progress in this area in the past few
years, which was reviewed in detail in Kennicutt (1990). Here I summarize
and update that discussion.
Gas content must be a fundamental parameter that determines the SFR.
Since the seminal paper by Schmidt (1959), several dozen papers have been
written which attempt to parametrize the SFR as a power-law function
of the gas density. Recent work indicates that the form of the SFR vs.
gas density law is more complicated than a single Schmidt power law. Its
form depends on the physical scale one considers, and on the dynamical
properties of the gas, independent of its absolute density. This should not
 547

34

. ~

I
()
0..
i I
~---
() 32 I

'"
III /
/ -'
III /
,,
<>0
....
/
I
~

,,
I
c I
:z:
1-'1 30
I

'.2 ,I
I
I

,
I
I

,
I

28

0.5 1 1.5 2
-2
log u ... (Mo pc )

Fig. 6. Spatially-resolved relations between Ha surface brightness and gas surface

density for selected galaxies. Lowest values at bottom of plot represent zero Ha
emission.

be surprising, because the physical mechanisms that regulate the gas/star

conversion rate on the scale, say, of individual molecular clouds are likely
to be quite different from those which regulate the formation of interstellar
clouds on galactic scales.
The form of the star formation law on kpc scales is illustrated in Figure 6.
The plots show the relationships between Ha: surface brightness (assumed to
scale with the SFR) and total gas density (atomic and molecular) within sev-
eral spirals in the Virgo cluster, from Kennicutt & Martin (in preparation).
Similar relations, covering both field and Virgo galaxies, can be found in
Kennicutt (1989). These observations show that the star-formation law pos-
sesses three distinct physical regimes. At high gas densities the SFR follows
a Schmidt law, but below a critical threshold density the relation becomes
much steeper, and the SFR drops to very low values at densities well below
the threshold. The thresholds are often quite sharp, as illustrated in Figure
6. The thresholds are observed over the full range of galaxy types (e.g., Skill-
man 1986; Guiderdoni 1987; Kennicutt 1989; Taylor et al. 1992). They occur
at column densities of order 10 20 to 1021 H cm- 2 or ~ ~ 1 - 10 M0 pc- 2 ,
but the variation from object to object is real (Kennicutt 1989).
548

The observed thresholds appear to be physically associated with the grav-

itational stability of the gas. Application of the Toomre (1964) stability cri-
terion for an isothermal gas disk reproduces the threshold densities and radii
in spiral with excellent accuracy (Kennicutt 1989). The observation that the
gas lies near the stability limit at a11 radii (Toomre parameter Q ~ 1) is also
consistent with the gravitational model (Quirk 1972; Kennicutt 1989; Zasov
& Simikov 1989). The criterion outlined in Kennicutt (1989) has been ap-
plied subsequently to several other types of objects, including SO and Sa
disks, low surface brightness galaxies, ring galaxies, and spiral arms within
galaxies. I refer the reader to papers at this meeting by Hogg & Roberts,
Struck-Marcell, Taylor et al., van der Hulst, and van Gorkom for more de-
tails. The physical connection between the stability criterion, which is rele-
vant only to the large-scale gas component, and the star formation, which
ultimately is a local phenomenon, is not completely clear. I suspect that the
stability criterion actually regulates the formation of gravitationally bound
clouds (either atomic or molecular), but it is conceivable that we are actually
seeing a atomic/molecular threshold in some cases, which should occur 'at
column densities of the same order as the Q = 1 gravitational threshold. Fur-
ther observations, extending to lower CO limits than are currently available,
are needed for a empirical test of these hypotheses.
For gas dens~ties above the threshold, the star formation laws in different
galaxies are remarkably coincident (see Figure 6 and Kennicutt 1989). This
suggests that some other physical mechanism may control the SFR in that
regime. Self-regulating models, in which pressure and energy input from
stellar winds and supernovae limit the SFR, predict a Schmidt law similar
to that seen in Figure 6 (e.g., Silk 1987; Dopita 1990). Models of density-
wave induced star formation may also be able to account for the observed
gas and SFR distributions (e.g., Wyse & Silk 1989). So far, we have found
no direct empirical support for these models, insofar as they fit the data
no better than a pure Schmidt law, but this may reflect the limited range
of physical conditions that we can probe with our current data. Further
observational tests of the self-regulating models are currently under way,
both by our group and by B. Wang (private communication).,

5.3. LOCAL ANALOGS TO PRIMEVAL GALAXIES

There are several approaches to isolating potential analogs to primeval galax-

ies. One is to identify nearby galaxies with the most extreme star formation
bursts, as evidenced by their ultraviolet, infrared, or emission-line spectra.
Another is to search for local examples of genuinely young galaxies, i.e.,
those which are forming stars for the first time, and thus can be regarded
as bona fide primeval galaxies themselves. Yet another approach is to iden-
tify objects which by some peculiar circumstance are analogs to a particular
 549

stage in galaxy formation, e.g., mergers or galaxies with cooling flows. In

this section I describe examples of each type.

5.3.1. Blue Emission-Line Galazies

The most common type of nearby starburst galaxy is characterized by blue
colors and strong nebular emission lines. Figures 7a and 7b show the inte-
grated optical spectra for two examples, NGC 1569 and NGC 4449, from a
recently published spectrophotometric atlas of galaxies (Kennicutt 1992b).
These data were obtained by drift-scanning the galaxies across a long slit,
so that an integrated spectrum could be obtained. NGC 1569 is a true star-
burst galaxy, with a dominant young stellar population and a SFR which is
several times its average rate. NGC 4449 represents a milder level of star for-
mation, chosen because its luminosity and integrated spectrum are similar
to the intermediate redshift "blue dwarf" galaxies.
At visible wavelengths these galaxies are dominated by a blue contin-
uum, with Balmer absorption lines and strong nebular emission lines. In
many intermediate-luminosity systems, including NGC 4449, [OIl]..\3727 is
the strongest spectral feature, rivaling even Ha in flux. The anomalous
strength of [OIl] appears to be due to a combination of an intermediate
metal abundance, which increases forbidden line strengths in general, and a
strong contribution from diffuse interstellar gas, which is ionized by a dilute
radiation field and thereby has an abnormally high [OIl]/H,l3 ratio (Kenni-
cutt 1992j.L). This is relevant to the spectra of distant galaxies, which also
are characterized by strong [OIl] equivalent widths. The abnormal excita-
tion may leau to a modest overestimate of the SFR if only the [OIl] flux
is used, but this does not alter the qualitative interpretation of the distant
blue galaxies as starbursting objects (Kennicutt 1992a).
Figure 8a shows the integrated energy distribution of NGC 1569 over
the UV-FIR spectral range, as measured by Israel (1988). The peak in
the energy distribution extends well into the ultraviolet, so such galaxies
should be detectable to quite high redshifts. Identifying such galaxies be-
yond z '" 1 can be difficult, however, because their UV spectra consist of
a flat-spectrum continuum with only weak emission or absorption features
(e.g., Hartmann et a1. 1988). Even Lya emission is weak or absent in all but
the most metal-poor galaxies. Hartmann et al. attribute the Lya quenching
to absorption/ scattering from surrounding neutral gas, as well as to trapping
by dust in the H II regions themselves. Work presented at this meeting by
Valls-Gaboud indicates that superimposed stellar Lya absorption may be a
more important effect, especially during the late stages of the starburst. If
this is the case, the detectability of Lya may depend more on the age of the
burst than on metallicity, and the line may be a more useful diagnostic at
high redshift than was previously believed.
550

46 14
Nee 1680 Nce 4UQi
~o C.) (b)
12
36
10
30

15

l f.~ l 1
4000 4500 5000 5500 8000 8500 7000 4000 .500 6000 6600 8000 8500 7000

Nee 3034 (1182) 3.5 Nee 1276

(e) Cd)

1.5

0.5

0 0
~OOO 45 00 5000 5500 8000 0500 7 000 4000 4500 5000 6500 8000 8500 7000

I.~ I.~

uee 5101 ARP 220

(e) (,)
1.2 1.2

0-
:i:oa 0.0
~
.....
:cO.S 0.8
1;:

0.4 0.'

0.2 0.2

0 0
3800 4000 4.00 4eOO 5200 3800 0400
~000 4800 5200
Wevelenalh (A) We ... , len,Ul (A)

Fig. 7. Spectra for nearby galaxies, as described in the text. Note the different
wavelength scales for the bottom two spectra.
 551

l
100

NGe 1569
......
..,
III
.~
10
::I
Ql
-=-
r.-:
~ OJ

-<
0.1

0.1 10 100
A (microns)

100

M82
'S
.~
10
::I
Ql
-=-
r.-:
-<
0.1

0.1 10 100
A (microns)

Fig. 8. Spectral energy distributions for NGC 1569, a blue emission-line galaxy, and
M82, a prototype infrared starburst galaxy. The ordinate is AF),., to better indicate
the relative contributions to the bolometric luminosity.

Many dwarf emission-line galaxies are characterized by very large gas

fractions and low metal abundances. Thus, it is legitimate to ask whether
any of these might be genuinely young galaxies, for which we are seeing the
first generation of stars. The question has been debated since this class of
galaxies was first identified, and it still is not firmly settled. Infrared pho-
tometry and imagery show that most blue compact dwarfs corltain smooth
extended red envelopes, which presumably represent an older underlying
stellar population (e.g., Loose & Thuan 1986; Kunth et al. 1988; Salzer &
Elston 1991). Among the possible exceptions are I Zw 18 and the optical
counterpart to the Virgo H I cloud (Salzer & Elston 1991; Salzer et aI. i991).
For a review of the subject see Thuan (1991).

5.3.2. Far-Infrared Starburst Galazies

Many starburst galaxies emit the bulk of their radiation at far-infrared wave-
lengths (e.g. , Soifer et a1. 1987; Telesco 1988). The prototype of this class is
552

M82, and its integrated spectrum (Kennicutt 1992b) and energy distribution
(Telesco 1988) are shown in Figures 7c and 8b, respectively. The differences
which are most immediately apparent are the strong reddening in the optical
spectrum of M82, and the vastly increased fraction of bolometric luminosity
in the far-infrared. If such objects are common at high redshift, they will be
most readily detectable at FIR and submillimeter wavelengths, either in the
continuum or in fine-structure nebular lines.
Despite the high FIR luminosities and large optical depths in these ob-
jects, many appear quite blue at visible wavelengths. Charles Liu and I
recently obtained spectra for a complete sample of ultra-luminous IRAS
galaxies, and two examples are shown in Figures 7e and 7f. The blue spectra
are usually quite flat, and early-type stars dominate the absorption spectra;
they are recognizable as starburst or post-starburst galaxies even without in-
formation in the infrared. This combination of high extinction and relatively
low reddening can be understood if the dust is clumpy, and if significant star
formation occurs in regions of low extinction. These observations also con-
firm that star formation is responsible for at least a part of the extraordinary
infrared luminosities of these objects.
While some of the IR-luminous starburst galaxies may be nothing more
than dusty versions of the blue galaxies discussed earlier, several lines of
evidence suggest that the most extreme cases represent a physically distinct
phenomenon. The bolometric luminosities of the most luminous objects, if
powered predominantly by star formation, imply SFRs of up to several hun-
dred M0 yr- 1 , comparable to the primordial starhursts in rapid-collapse
models of galaxy formation. Many objects possess unusually massive cen-
tral gas complexes (e.g., Scoville et a1. 1991; Solomon et a1. 1992). Several
authors have demonstrated a strong statistical link between the most lumi-
nous IR starbursts and very close interactions or mergers of galaxies (e.g.,
Sanders et a1. 1988; Melnick & Mirabel 1990). It is conceivable that these
objects represent sites where galactic nuclei and/or spheroids are forming
today (e.g., Sanders et a1. 1988), hence their relevance to more distant proto-
galaxies. Modelling of M82 and similar galaxies also suggests that the local
properties of the starburst, such as the IMF and star formation efficiency,
may be different from those in normal disks (e.g., Rieke 1991).

5.3.3. Star Formation in Cooling Flows

Yet another star formation environment occurs in the dominant central
galaxies of cooling flow clusters (see Mushotzky's review in this vohune).
Figure 7 d shows the integrated optical spectrum of one of the best known
examples, NGC 1275 in the Perseus cluster (Kennicutt 1992b). This is easily
the most peculiar spectrum in the atlas. The famous double-velocity emis-
sion line system is readily apparent; this emission arises from gas associated
 553

with the cooling flow itself, together with nuclear activity in NGC 1275 itself.
The X-ray inferred mass deposition rates for typical cooling flows are
of order 10-1000 M0 yr-t, which implies utterly enormous SFRs, if this
gas is converted into stars at this rate. This issue has stimulated several
extensive surveys of the stellar populations and emission-line gas in cooling
flow cluster dominant galaxies (e.g., Romanishin 1986, 1987; McNamara &
O'Connell 1989; Heckman 1989; Caulet et al. 1992). Evidence for young
stars is seen in ",60% of the galaxies studied to date, but the mean inferred
SFR is only 7% of the mass deposition rate (20% in NGC 1275) if a normal
IMF is assumed (McNamara & O'Connell 1989). The discrepancy suggests
either that the mass accretion rates are in error, or that the stellar IMF in
the cooling flows is abnormally enriched in low-mass stars. Note that the
latter effect is quite the opposite of what is alleged to occur in IR-Iuminous
star burst galaxies. It is possible, but not yet firmly established, that the
cooling flows contain an entirely distinct mode of star formation from what
is observed in other types of starburst galaxies.

5.3.4. Young Analogs to Globular Clusters

A prominent signature of galactic spheroid formation should be the forma-
tion of globular star clusters. If objects comparable to our Galaxy's globular
clusters formed in regions with a normal IMF, they would have initiallumi-
nosities of order MB = -15, and ionization rates of order 10 53 photons S-1
(Kennicutt & Chu 1988). In the past year, there have been claims of detec-
tion of such clusters with the Hubble Space Telescope, in NGC 1275 (Holtz-
man et al. 1992), and in the infrared starburst galaxy Arp 220 (Dowling &
Shaya 1992).
The observation of massive, compact young clusters is not new. The Mag-
ellanic Clouds contain a large population of blue populous clusters with ages
of order 10 7 -10 9 yr. These clusters are structurally similar to Galactic globu-
lar clusters, although their masses are typically an order of magnitude lower.
Other examples of luminous, unresolved clusters with ages of order 10 7 yr
are seen in NGC 1569, NGC 1705, NGC 3597, NGC 5253, and possibly
NGC 5128 (references in Kennicutt & Chu [1988] and next section). Com-
pact core clusters are also observed in several nearby giant H II regions,
including 30 Doradus. There seems to be a strong correlation between those
galaxies which possess giant H II regions and those which contain populous
blue star clusters (Kennicutt & Chu 1988).
Whether these compact young clusters are destined to evolve into true
globular clusters is still open to question, in my view. Even with HST, the
spatial resolution at the distance of NGC 1275 or Arp 220 is tens of parsecs
at best, insufficient to resolve a true globular cluster from a more weakly
bound open cluster or compact association. The starlight in these objects
554

is dominated by 0, B, and early A-type stars, so it is nearly impossible to

detect a low~mass stellar population, even in nearby clusters. Probably the
strongest circumstantial evidence in favor of the formation of at least some
young globular clusters is the existence of intermediate age objects ('" 107 - 8
yr) in the LMC, which have roughly Salpeter IMFs ranging from below 1
M0 to nearly 10 M 0 • However the formation rate of such clusters is at least
an order of magnitude below the formation rate of ·of gi,ant H II regions of
the same primordial luminosity (Kennicutt & Chu 1988). This suggests that
bound globular clusters, if they form at all in such H II regions, only form
under rare circumstances.

5.3.5. Interacting Galazies and Mergers

Up to now we have only considered galaxies that bear some photometric
signature of a starburst. An alternative approach to identifying analogs
to primeval galaxies is to select objects that dynamically resemble proto-
galaxies, independent of their photometric properties. By studying the star
formation properties of these objects, one can avoid the selection biases in-
herent in studies of starburst galaxies. Here, I discuss an example of how this
approach can be applied to the study of interacting and merging galaxies.
The link between galaxy-galaxy interactions and star-formation bursts
is now well established. Close pairs of galaxies are known to have average
mean disk star formation rates (e.g., Bushouse 1987; KEmnicutt et al. 1987),
stronger FIR emission (Bushouse et al. 1988), and elevated levels of nuclear
star formation and nonthermal activity (e.g., Cutri & McAlary 1985; Keel
et al. 1985). N -body simulations of interacting systems can reproduce the
large-scale gas compression and angular momentum transport needed to fuel
this disk and nuclear activity, respectively (e.g., Barnes & Hernquist 1991;
Mihos et al. 1991).
The most spectacular activities appear to be associated with galactic
mergers. Observations of a few prototype mergers, such as NGC 7252 and
NGC 3597, show evidence for large fossil starbursts, in some cases with
spheroid-like r 1 / 4 luminosity profiles (Schweizer 1990; Lutz 1991). Lutz iden-
tifies several bright compact blue knots in NGC 3597, which he suggests may
be young globular clusters. Ashman Ie Zepf (1992) discuss the cluster for-
mation in mergers in general, and suggest that similar clusters have formed
in Cen A (NGC 5128). Recently, I began a more comprehensive spectropho-
tometric and imaging survey of ",40 systems, in collaboration with Charles
Liu and Pat Hall. Two examples of these objects (which also happen to be
ultraluminous IR galaxies) are shown in Figures 7e and 7f. Our preliminary
results show that the mergers display the full range of SFRs, ranging from
completely old stellar populations, to ultra-luminous starbursts. The most
common spectrum is that of a postburst population (UGC 5101, Figure 7e),
 555

3.6

2.6

0.5

4000 4500 5000 5500 6000 6500 7000 .000 4500 5000 5500 6000 6500 7000
Wavelength (A) Wavelength (A)

Fig. 9. Integrated spectra of NGC 1832 and NGC 3077, from Kennicutt (1992).
Insets are blue images of the galaxies.

often mixed with an older underlying population (Arp 220, Figure 7f). A
few of the latter resemble the "E+A" galaxies seen frequently in distant
clusters, although true analogs appear to be rare. Quantitative analysis of
the star formation rates and histories is underway.

6. Concluding Remarks
Observations of nearby galaxies offer a wide range of constraints on galaxy
formation and evolution, which in many respects complement the informa-
tion that is obtained from observations of cosmologically distant objects. In
all three of the areas discussed (local stellar populations, nearby galaxies,
and distant galaxies) recent advances in instrumentation have led to signif-
icant progress in our ability to reconstruct the stellar birthrate history of
the universe. The advent of new large telescopes promises to continue this
progress. Galactic evolution has become an observation-driven science, and
it is an exciting time to work in the field.
As a final illustration of the value of close-up observations, Figure 9 shows
integrated spectra of two nearby galaxies, NGC 1832 and NGC 3077, from
Kennicutt (1992b). These galaxies are as close to being spectroscopic twins,
at least at visible wavelengths, as any I encountered during the survey (ex-
cept that NGC 1832 is about 3 magnitudes brighter). If these galaxies were
observed at high redshift, one might reasonably conclude that they have sim-
ilar stellar populations and star formation histories. The insets show blue
photographs of NGC 1832 (SBb) and NGC 3077 (Irr II). Caveat emptor!
556

Acknowledgements
I wish to thank L. Calzetti, R. Ellis, and A. Kinney for providing unpub-
lished figures used in my talk, and G. Bernstein, M. Bolte, R. Rand, J.-R.
Roy, and D. Valls-Gabaud for valuable comments and suggestions. I am es-
pecially grateful to my graduate student collaborators at Arizona, P. Hall, C.
Liu, and C. Martin, for their contributions to the work presented here. My
own research in this area is supported by the National Science Foundation
through grant AST-9019150.

References
Ashman, K . M ., Zepf, S. E .: 1992, ApJ 384, 50
Babul, A., Rees, M. J.: 1992, MNRAS 255, 346
Barnes, J. E ., Hernquist, L.: 1991 , ApJ 370, L65
Baron, E., White, S. D. M.: 1987, ApJ 322, 585
Barry, D. C .: 1988, ApJ 334, 446
Bolte, M.: 1992, in J. Brodie, G. Smith, ed(s)., The Globular Clu6ter-Galaxy Co~mection,
(Astr. Soc. Pacific), in press
Broadhurst, T. J ., Ellis, R. S., Shanks, T.: 1988, MNRAS 235, 827
Broadhurst, T . J ., Ellis, R . S., Glazebrook, K.: 1992, Nature 355, 55
Buat, V.: 1992, A8A ,in press
Buat, V., Deharveng, J . M., Donas, J.: 1989, A8A 223, 42
Buonanno, R ., Buscema, G ., Fusi Pecci, F., Richer, H . B ., Fahlman, G . G.: 1990, AJ 100,
1811
Bushouse, H. A. : 1987, ApJ 320, 29
Bushouse, H. A., Lamb, S. A., Werner, M. W.: 1988, ApJ 335, 74
Carlberg , R. G., Dawson, P. C., Hsu, T., VandenBerg, D. A.: 1985, ApJ 294, 674
Caulet , A. , Woodgate, B. E., Brown, L. W., Gull, T . R., Hintzen, P., Lowenthal, J. D.,
Oliversen, R .J., Ziegler, M. M. : 1992, ApJ 388, 301
Chaboyer, B ., Sarajedini, A., Demarque, P.: 1992, ApJ 394, 515
Colless, M., Ellis, R . S., Taylor, K., Hook, R. N .: 1990, MNRAS 244, 408
Colless, M., Ellis, R . S., Broadhurst, T . J., Taylor, K ., Peterson, B. A.: 1992, MNRAS ,
in press
Cowie, L. L. , Songaila, A. , Hu, E . M.: 1991, Nature 354, 460
Cowie, L. L., Gardner, J . P., Hu, E. M., Wainscoat, R . J., Hodapp, K. W.: 1993, ApJ ,in
press
Cutri, R . M. , Me Alary, C . W. : 1985, ApJ 2'6, 90
Da Costa, G. S.: 1991, in R. Haynes, D. Milne, ed(s)., The Magellanic Cloud6, IAU Sym-
p06ium 148, (Dordrecht: Kluwer), 183
Da Costa, G. S.: 1992, in B . Barbuy, A. Rensini, ed(s) ., The Stellar Populations of Galax-
ies, IAU Symposium 149, (Kluwer), in press
Devereux, N. A., Young, J . S.: 1991, ApJ 3'11, 515
Donas, J. , Deharveng, J. M ., Laget, M., Milliard, B., Huguenin, D .: 1987, A8A 180, 12
Dopita, M . A .: 1990, in H . A. Thronson, J. M. Shull, ed(s)., The Interstellar Medium in
Galaxies, (Kluwer), 437
Dowling, D., Shaya, E .: 1992, BAAS 24, 728
Dressler, A., Gunn, J . E .: 1990, in R. G . Kron, ed(s) ., Evolution of the Universe of Galaxies,
(Astr. Soc. Pacific), 200
Eggen, O . J., Lynden-Bell, D., Sandage, A.: 1962, ApJ 136,748
Elston, R., Rieke, G., Rieke, M.: 1991, in R. Elston, ed(s)., Astrophysics With Infrared
Arrays, (Astron. Soc. Pacific), 3
Gallagher, J. S., Bushouse, H ., Hunter, D. A.: 1989 , AJ 97, 700
 557

Gallagher, J . S., Hunter, D. A.; Tutukov, A. V.: 1984, ApJ 284, 544
Gavazzi, G., Bosselli, A., Kennicutt, R.: 1991, AJ 101, 1207
Glazebrook, K.: 1991, Ph.D. thesis, Univ. of Edinburgh
Green, E. M., Norris, J. E.: 1990, ApJ 353, L17
Guiderdoni, G., Rocca-Volmerange, B.: 1990, A&A 227, 362
Hardy, E., Buonnanno, R., Corsi, C. E., Janes, K. A., Schommer, R. A.: 1984, ApJ 278,
592
Hartmann, L. W., Huchra, J. P., Geller, M. J., O'Brien, P., Wilson, R.: 1988, ApJ 326,
101
Heckman, T. M., Baum, S. A., van Breugel, W. J. M., McCarthy, P.: 1989, ApJ 338, 48
Holtzman, J. A., et al.: 1992, AJ 103,691
Israel, F. P.: 1988, A&A 194, 24
Katz, N.: 1992, ApJ 391, 502
Keel, W. C., Kennicutt, R. C., Hummel, E., van der Hulst, J. M.: 1985, AJ 90, 708
Kennicutt, R. C.: 1983, ApJ 272, 54
Kennicutt, R. C.: 1989, ApJ 344, 685
Kennicutt, R. C.: 1990, in H. A. Thronson, J. M. Shull, ed.(s)., The Interstellar Medium
in Galaxies, (Dordrecht: Kluwer), 405
Kennicutt, R. C.: 1992a, ApJ 388, 310
Kennicutt, R. C.: 1992b, ApJS 79, 255
Kennicutt, R. C.: 1992c, in G. Tenorio-Tagle, ed(s)., Star Formation in Stellar Systems,
(Cambridge U. Press), in press
Kennicutt, R. C., Chu, Y.-H.: 1988, AJ 95, 720
Kennicutt, R. C., Edgar, B. K., Hodge, P. W.: 1989, ApJ 337,761
Kennicutt, R. C., Keel, W. C., van der Hulst, J. M., Hummel, E., Roettiger, K. A.: 1987,
AJ 93,1011
Kennicutt, R. C., Kent, S. M.: 1983, AJ 88, 1094
Kron, R. G., ed.: 1990, Evolution of the Universe of Galaxies, San Francisco, Astr. Soc.
Pacific)
Kunth, D., Maurogordato, S., Vigroux, L.: 1988, A&A 204, 10
Larson, R. B.: 1991, in D. L. Lambert, ed(s)., Frontiers of Stellar Evolution, (Astr. Soc.
Pacific, 571
Loose, H. H., Thuan, T. X.: 1986, ApJ 309, 59
Lutz, D.: 1991, A&A 245, 31
McNamara, B. R" O'Connell, R. W.: 1989, AJ 98, 2018
Melnick, J., Mirabel, I. F.: 1990, A&A 231, L19
Mighell, K. J.: 1990, A&AS 82,1
Mihos, J . C., Richstone, D.O., Bothun, G. D.: 1991, ApJ 377, 72
Minniti, D., Peterson, R., Geisler, D., Clara, J.: 1992, in J. Brodie, G . Smith, ed(s)., The
Globular Cluster-Galazy Connection, (Astr. Soc. Pacific), in press
Quirk, W. J.: 1972, ApJ 176, L9
Rieke, G.: 1991, in C. Leitherer, N. R. Walborn, T. M. Heckman, C. A: Norman, ed(s) .,
Mas6ive Stars in Starbursts, (Cambridge Univ Press, 205
Roberts, M. S.: 1963, ARAOA 1, 149
Romanishin, W.: 1986, ApJ 301, 675
Romanishin, W.: 1987, ApJ 323, L113
Romanishin, W.: 1990, AJ 100, 373
Salzer, J. J., Alighieri, S., Matteucci, F., Giovanelli, R ., Haynes, M. P.: 1991, AJ 101,
1258
Salzer, J. J., Elston, R.: 1991, in R. Elston, ed(s)., A8trophysics With Infrared Arrays,
(Astr. Soc. Pacific, 41
Sandage, A.: 1986, A&A 161, 89
Sanders, D. B., Soifer, B. T., Elias, J. R., Madore, B. F., Matthews, K., Neugebauer, G.,
Scoville, N. Z.: 1988, ApJ 325, 74 '
Scalo, J. M.: 1986, Fund. Cosmo Phys 11, 1
558

Schmidt, M.: 1959, ApJ 129, 243

Schweizer, F .: 1990, in R. Wielen, ed(s )., Dynamic6 and Interaction6 oj Gala:cie6, (Springer-
Verlag), 60
Scoville, N. Z., Sargent, A. Z., Sanders, D. B., Soifer, B. T.: 1991, ApJ 366, L5
Searle, L., Zinn, R.: 1978, ApJ 225, 357
Silk, J.: 1987, in M. Peimbert, J. Jugaku, ed(s)., Star Forming'Region6, (Reidel), 663
Soifer, B. T., Houck, J. R., Neugebauer, G.: 1987, ARA8A 25, 187
Solomon, P. M., Downes, D., Radford, S. J.E.: 1992, ApJ 387, L55
Stryker, L. L.: 1984, ApJS 55, 127
Telesco, C. M.: 1988, ARA8A 26,343
Thuan, T. X.: 1991, in C. Leitherer, N. R. Walborn, T. M. Heckman, C. A. Norman,
ed(s)., Ma66ive Star6 in StarbtJr6t&, (Cambridge Univ. Press, 183
Toomre, A.: 1964, ApJ 139, 1217
Twarog, B. A.: 1980, ApJ 242, 242
VandenBerg, D. A., Bolte, M., Stetson, P. B.: 1990, AJ 100, 445
van den Bergh, S.: 1992, MNRAS 255, 29P
Wheeler, J. C., Sneden, C., Truran, J. W.: 1989, ARA8A 27, 279
Wyse, R. F. G., Gilmore, G.: 1988, AJ 95, 1404
Wyse, R. F. G., Silk, J.: 1989, ApJ 339, 700
Zasov, A., Simikov, S.: 1988, A6trophY6ica 29, 190
 THE FIRST QSOs AND THEIR INFLUENCE ON THE
INTERGALACTIC MEDIUM
JILL BECHTOLD
Steward Observatory, University of Arizona

Abstract. High-redshift QSOs are prodigious sources of ultraviolet photons which may
influence the ionization and thermal history of the diffuse intergalactic medium and the
Lya fo~est clouds. Recent results on the ambient, metagalactic ultraviolet radiation field
as a function of redshift are reviewed. At z = 0, searchell for Ha emission from 21-cm
emitting clouds, modeling of the sharp edges of H I clouds, and direct measurements of
the far ultraviolet flux at high Galactic latitude place limits on any diffuse extragalactic
component. At z ~ 3, the ambient background can be estimated through studies of the
Lya forest and other types of QSO absorption-line systems. Estimates of the ultraviolet
background obtained through modeling of the "proximity effect" in the Lya forest at z ~ 3
probably are in reasonable agreement with the expected contribution of luminous QSOs,
when the large uncertainties in both quantities are taken into account.

1. Introduction

The highest-redshift QSOs are of widespread interest because they tell us

about the Universe at early times. As Efstathiou & Rees (1988) and Turner
(1991) have emphasized, at redshift z ~ 4, the age of the Universe is only
about (6 X 108 yrs)h 101o for Ho = (100 km S-1 Mpc- 1 )h100 and qo = 0.5. This
is comparable to a typical galaxy dynamical time, as well as the Eddington
accretion time for a supermassive black hole. Thus formation of the first
QSOs had to be efficient; the most obvious problem is finding a plausible
mechanism for the rapid dissipation of angular momentum (e.g., Loeb 1992).
High-redshift quasars produce relatively hard ultraviolet radiation, which
may be an important source of energy in the early Universe. The integrated
extreme ultraviolet (EUV) radiation field from quasars and other high red-
shift objects provides an ambient background of ultraviolet radiation which
may contribute to the "reionization" of the intergalactic medium (e.g., Gunn
& Peterson 1965; Rees & Setti 1970; Arons & Wingert 1972; Ikeuchi &
Ostriker 1986; Donahue & Shull 1987; Shapiro & Giroux 1987; Meiksin &
Madau 1993; Cen & Ostriker 1993; Bajtlik, this volume). After the epoch of
reionization, this UV radiation photoionizes the Lya forest clouds (Sargent
et al. 1980; Atwood, Baldwin, & Carswell 1985; Weymann, this volume).
The redshift evolution of the EUV background determines the redshift evo-
lution of the number density of Lya forest clouds, a basic observed property
(Atwood, Baldwin, & Carswell 1985; Bechtold et al. 1987; Ikeuchi & Turner
1991; Charlton, Salpeter, & Hogan 1993). The ambient field may determine
the ionization of some quasar metal absorption-line systems (Bergeron and
Stasinska 1986; Steidel & Sargent 1989; Steidel, this volume), as well as
clouds in our own halo (York 1982; Fransson & Chevalier 1985; Bloemen

SS9
1. M. Shull and H. A. Thronson, Jr. (eds.), The Environment and Evolution of Galaxies, SS9-S78.
© 1993 Kluwer Academic Publishers.
560

1987; Sembach & Savage 1992). Finally, estimates of the radiation field at
z = 4 to 5 can be used to place limits on populations of yet-to-be-discovered
objects at higher redshift (Heisler & Ostriker 1988; Bajtlik, Duncan & Os-
triker 1988, hereafter BDO).
Here I review what we know about the ultraviolet radiation background
as a function of redshift from an observational point of view. The radiation
field can be described by its specific intensity, J",(z), which is a function of
frequency, v, and redshift, z, and has cgs units of ergs cm- 2 s-1 HZ-1 sr- 1 •
It is convenient to express the specific intensity at the Lyman limit (912
A) in units of J- 2b where J",(912 A) = J- 21 X 10- 21 ergs cm- 2 S-l Hz- 1
sr- 1. In the next section, calculations of the contribution of various sources
to J", are reviewed. Following that, the empirical measurements of J", are
discussed.

2. Sources of J",
The expected contribution of a population of objects to the mean ambient
radiation field can be described by the specific intensity J",(ZOO,,) seen by an
observer at redshift ZOO" at frequency v. This can be written as an integral
over redshift

J",(ZOO,,) 1
= -4
7r
1 (1 ++
00

%06. 1
Zoo,,) 3 £(v , 'Z)-d
z z
dl (exp( -r(v, Z, zob,,»)dz , (1)

where £(v', z) is the proper volume emissivity of the object's at redshift Z

and frequency v' is given by

v' = ( 1 1++Zob"
z )
11.
(2)

Here,
dl c ,1 1
(3)
dz = Ho (1 + z)2 y'1 + 2qoz
is the expression relating the proper length increment dl and redshift incre-
ment dz, and (exp( -r(v, z, zoo,,») is the mean transmission for a photon of
frequency v as it travels from redshift z to ZOO"' The proper volume emissiv-
ity is computed from the luminosity function of the objects and an assumed
spectral shape, or distribution of spectral shapes, for the individual objects.
Note that the expression for J", in (1) is independent of Ho except indirectly
through the derivation of £ and r.
Thus, one needs to know the luminosity function for the objects as a
function of redshift for all redshifts higher than ZOO"' the EUV spectrum of
individual objects, cosmological parameters (<<:lo 'and H o ), and the amount
 561

and distribution of neutral gas and dllst which attenuates the UV radiation.
Needless to say, the uncertainties in these quantities are large, and hence
the estimates of various source contributions have fairly large uncertainties.

2.1. QSOs

Ltuninous QSOs have long been thought to be major contributors to J -21

at z > 2, owing to their relatively hard spectral energy distributions in the
EUV (Rees & Setti 1970; Sargent et al. 1980), despite the fact that they
are rare objects. Early estimates were uncertain since the exact form of the
quasar luminosity function at z > 3 was not well known (e.g., Bechtold et al.
1987). However, recently the situation has greatly improved (see Schneider,
Schmidt, & Gunn 1989; Irwin, McMahon, & Hazard 1991; Boyle, Jones, &
Shanks 1991; Schmidt, Schneider, & Gunn 1991; Warren, Hewett, & Osmer
1991; Boyle, these proceedings), and there are now sufficient numbers of
quasars known with Zem = 3 to 5 that the luminosity function for very high
redshift quasars can be discussed. Madau (1992) and Meiksin & Madau
(1993) have recomputed the quasar contribution to J- 21 based on recent
estimates of the quasar luminosity function. They find that J- 21 ~ 0 .1
to 0.7, with relatively little dependence on z, for Z > 2. Attenuation by
intervening absorbers is potentially important, and can account for factors
of 4 to 6 diminution of the background.
If high-redshift quasars are obscured by dust, the J -21 calculated from
the quasars we see today is a lower limit to the actual radiation field that
say, an intergalactic cloud at Z = 3 "sees" (Ostriker & Heisler 1984; Wright
1986; Heisler & Ostriker 1988; BOO; Boisse & Bergeron 1988; Najita, Silk,
& Wachter 1990; Wright 1990; Ostriker, Vogeley, & York 1990; Fall & Pei
1993). Certainly a small E(B-V) from an intervening galaxy can cause a
very high redshift quasar to be missed in an optical survey, where the search
is based on the presence of relatively blue continuum and broad emission
lines at rest wavelengths A < 2000 A. The magnitude of the correction
for dust obscuration to the luminosity function is a matter of debate since
it depends on the amount of dust and its distribution within galaxies at
intermediate redshifts, which cannot be measured directly. However, given
plausible assumptions for the dust content of galaxies, Miralde-Escude &
Ostriker (1990) and Fall & Pei (1993) agree that J- 21 may be underestimated
by as much as a factor of 10 at Z = 3, without violating other observational
constraints.

2 .2. AGNs

At high redshifts, only the bright end of the quasar luminosity function is
observed directly, and the contribution of the dimmer, but more numerous
562

active galactic nuclei to JI/ may be important. At low redshifts, the relative
importance of the AGN is even greater, since the high luminosity objects are
less common. These trends may be offset by an increase of intrinsic absorp-
tion and dust with decreasing luminosity, as suggested by the large X-ray
absorbing columns seen for Seyferts, but not luminous quasars (however see
Elvis et a1. 1993 and references therein), and by evidence for obscuring tori
in Seyfert 2's (e.g., Antonucci & Miller 1985; Krolik 1990). Most estimates of
the contribution of bright QSOs to JI/ assume that UV photons are emitted
isotropically.
Fall & Pei (1993) show that a plausible extrapolation of the quasar lumi-
nosity function to low luminosity can increase J -21 by a factor of 2. Terasawa
(1992) has approached the question from the high energy side. Terasawa
argues that the low luminosity AGN make up most of the soft X-ray back-
ground and uses Comptonized accretion disk models to predict the EUV
flux of QSOs from the X-ray fluxes. In these models, J- 21 ~ 1 at z = 3.

2 .3. YOUNG STARS IN GALAXIES

Massive, young stars in high-redshift galaxies may contribute to J- 21 at

the Lyman limit even if quasars dominate at shorter wavelengths (Rees &
Setti 1970; Tinsley 1972; Code & Welch 1982; Bechtold et a1. 1987; Shapiro
& Giroux 1989; Songaila, Cowie, & Lilly 1990; Miralde-Es'cude & Ostriker
1990; Cen & Ostriker 1993). Rough estimates of the contribution of pas-
sively evolving ellipticals showed that the contribution of hot stars may be
substantial, if all UV photons escape the galaxies. In general, the contribu-
tion of young stars depends on the initial mass function, the mass cut-off
on the high end, the star formation rate as a function of time, the stellar
atmosphere models shortward of 912 A, and the distribution of dust and gas
with respect to the stars (see Bruzual, these proceedings). Miralde-Escude
& Ostriker (1990) have constructed the most detailed models, and show that
plausibly J -21 may be as high as 4 at z = 3, and thus exceed the contribution
of quasars.
An upper limit on the contribution of hot stars to JI/ can be made by
a different argument, namely by estimating the number of ionizing photons
emitted when the metals in the Universe were produced by nucleosynthesis.
Miralde-Escude & Ostriker (1990) estimate that J- 21 ~ 7(1 + z)3, assuming
that all the photons are emitted at a redshift z (in a burst of star-formation)
and that f!meta/" ~ 0.01. Obviously this is crude, but shows that the contri-
bution from stars is potentially significant.
 563

2.4. THE DECAY OF WEAKLY INTERACTING PARTICLES

Massive particles left over from the hot Big Bang subsequently decay and
produce line emission, which is smeared out by the expansion of the Uni-
verse into a background continuum radiation field (Cowsik 1977; DeRujula
& Glaskow 1980; Stecker 1980, Kimhle, Bowyer, & Jakobsen 1981; Melott
& Sciama 1981). The suggested particles include photinos, neutrinos and
gravitinos, but any particle with a lifetime,o f '" 10 23 s can result in suitable
ionizing photons, that is, photons with energies greater than 13.6 e V. Sciama
(1990abc, 1991) has pursued the astrophysical consequences of particle de-
cays, and estimates that they could have ionized the intergalactic medium,
and account for J- 21 ~ 100 at z = 3.
This model makes testable predictions. If such particles are also making
up the dark matter in local galaxies, then they might produce observable
line emission in rich galaxy clusters, which is not seen (Fabian, Naylor, &
Sciama 1991; Davidsen et al. 1991). Cold clouds might absorb such radiation,
however. Also, Miralde-Escude & Ostriker (1992) point out that decaying
neutrinos would produce a soft spectrum compared to the spectrum pro-
duced by quasars, so that He I, with an ionization potential of 24.6 e V, may
be the dominant ionization state of helium. Thus, the Lya forest clouds
should show He I .\584 absorption in HST spectra. Limits on the He I .\584
Gunn-Peterson test are probably already in conflict with this picture as well
(Tripp, Green, & Bechtold 1990; Reimers et al. 1992).

2.5. IGM EM.ISSION

A diffuse intergalactic medium may produce He II Lya .\304 and other line
emission which is smeared out with redshift (Field 1959; Kurt & Sunyaev
1967; Weymann 1967; Jakobsen 1980). Barcons, Fabian, & Rees (1991) have
reviewed IGM models. If the IGM is photoionized, then recombination He II
.\304 is probably negligible (see also Paresce, McKee, & Bowyer 1980), but
if the IGM is collisionally ionized and clumpy, then He II .\304 may be
substantial. This last scenario requires substantial kinetic energy input into
the IGM at early times, perhaps from shocks in collapsing protogalaxies (Cen
& Ostriker 1993) or from proto-QSOs. Such objects may also be significant
sources of EUV photons which have not been accounted for above.

3. Measurements of J v

Empirical measurements of J v have been made in a variety of ways. First I

discuss the limits at z = 0, then at high redshift.
564

3.1. THE INTENSITY J- 21 AT z =0

Measurements of J -21 locally not only are interesting by themselves, but also
place limits on J -21 at higher redshifts. If a source of UV photons turns off
at redshift z, then J", decays with J", ()( (1+z)4 in the ansence of absorption.

3.1.1. Ho. emission from 21-cm clouds

Reynolds (1987) and Songaila, Bryant, & Cowie (1989) have reported detec-
tions of Ho. emission from high velocity 21-cm clouds in the Milky Way. This
emission is hard to measure since it is weak compared to the much brighter
background of Ha emission from the warm, ionized ISM (Reynolds 1987).
The limit on J -21 depends on how far the high velocity clouds are above
the plane of the Milky Way. If they are far from the plane, Case B, and
have T = 104 K, then the extragalactic background at the Lyman limit is
J -21 = 0.06. If the clouds are close to the plane of the Galaxy, then photons
from hot 0 and B stars will contribute to the ionization of hydrogen, so this
is an upper limit to the extragalactic flux.
Similarly, limits have been placed on Ha emission from a few extragalactic
21 cm clouds which appear to be far from stars, so that one again expects
the metagalactic EUV photons to dominate. Reynolds et al. (1986) reported
a weak detection of Ha emission for the "intergalactic" H I Cloud in Leo
(Schneider et al. 1983), but concluded that the expected ambient J", was
inadequate by a factor of 3 to produce the Ha photons. Stocke et al. (1991)
put a limit on Ha emission for an H I "finger", which, analogous to the
Magellanic Stream, may have been tidally drawn out of NGC 3067 (Carilli,
van Gorkom, & Stocke 1989). They derive a 3 - ( j upper limit on J- 21 < 0.18
from their limits on Ha.

3.1.2. The Sharp cutoff in H I disks 01 Spiral Galazies

Silk & Sunyaev (1976) predicted that the H I disks of spirals should have a
sharp edge as the result of ionization by an extragalactic EUV background.
Such edges have been observed in 21-cm entission maps of M33 and NGC
3344 by Corbelli, Schneider, & Salpeter (1989); see also Van Gorkom, these
proceedings. Sharp edges have also been inferred from 21-cm absorption
studies of a number of galaxies (Corbelli & Schneider 1990) and high velocity
clouds in our own Galaxy (Colgan, Salpeter, & Terzian 1990). Assuming that
this is a result of photoionization (and not for example a tidal cut-off), the
limits for M 33 and NGC 3198 imply J -21 ~ 0.04 (Maloney 1993; Charlton,
Salpeter, & Hogan 1993). Making the apparently reasonable assumption that
the gas is far from other sources of ionizing radiation, this translates into
an estimate of the extragalactic ambient background locally. Observational
 565

confirmation that the cut-off is indeed the result of photoionization, e.g., the
detection of Ha emission, would be reassuring.

3.1.3. Direct Measurement

The extragalactic UV background can be measured directly just longward of
the Lyman limit. These observations are reviewed by Paresce (1990), Bowyer
(1991) and Henry (1991); early measurements are summarized by Paresce &
Jacobsen (1980). The observations of course are challenging (see particularly
the discussion in Henry 1991). However, the major uncertainty arises from
the difficulty in isolating the extragalactic component. From the observed
fluxes, one must subtract the contribution of stars, zodiacal light, and, most
importantly, scattered starlight from high latitude dust. One approach is to
measure the intercept of the plot of EUV flux versus N(H I) (e.g., Martin
& Bowyer 1989; Fix, Craven, & Frank 1989; Martin, Hurwitz, & Bowyer
1991). The assumption is that at "N(H I) = 0", the amount of dust is
negligible and hence the EUV scattered starlight is nil. However, whether
an actual "N(H I) = 0" line of sight exists in our Galaxy is questionable
(Lockman, Jakoda, & McCammon 1986). Nonetheless, the result is that the
extragalactic component is non-zero, but weak - only a few percent of the
observed flux. Wright (1992) has criticized this approach, arguing that the H
I alone does not trace the all the dust. Including H II and H2 in the total gas
budget implies lower estimates of the extragalactic component. Moreover,
using independent data) Murthy et al. (1990) do not find the correlation of
UV flux with N(H I) upon which this method is based.
The spectrum of angular fluctuations in the observed background inten-
sity also provides a constraint on the population of sources, provided one
understands their EUV spectral energy distributions and their clustering
properties (Shectman 1973; Code & Welch 1982; Martin & Bowyer 1989).
Martin & Bowyer (1989), for example, report fluctuations on 6' to 12' scales
of AI / I ~ 4% longward of the Lyman limit, at 1660 A. They plausibly
account for these fluctuations in terms of a small contribution to the total
observed background from galaxies at z = 0.1 to 0.6. A conservative upper
limit, J -21 < 0.5 can be derived from these observations.

3 .2. THE INTENSITY J II AT z = 2 TO 5

3.2.1. The Prozimity Effect
The number of Lya forest lines near high-redshift QSO~ is smaller than one
would expect from the observed line density far from the QSO (Weymann,
Carswell, & Smith 1981; Murdoch et al. 1986; Tytler 1987; Carswell et al.
1987; BDO; Lu, Wolfe, & Turnshek 1991; Bechtold 1993). Weymann, Car-
swell & Smith (1981) first suggested that this so-called "proximity effect"
566

may result from the fact that the clouds near the QSO are more ionized
than average by the hard EUV photons of the QSO itself. Since the Lya
lines measure the amount of H I, if more of the hydrogen is H II, the equiv-
alent width of Lya decreases, and the number of lines above some fixed
equivalent width threshold decreases. Alternatively, Tytler (1987) suggested
that perhaps the clouds are too small to cover the broad-line region, but
this is observationally ruled out by the shape of line profiles of lines with
Zab8 >=:::; Zem. Other possibilities involve more complex interactions of the QSO
with the IGM near it, which then affects the Lya forest clouds if they are
predominantly pressure confined. It is interesting, parenthetically, that the
cold (T < 10,000 K) Lya forest clouds found by Pettini et al. (1990) are
mostly at the QSO emission line redshift, or redward of it, suggesting that,
in fact, the physical conditions in the IGM near the QSO are pathological.
If, however, the simple picture is true, that the proximity effect is the re-
sult of photoionization of the clouds by the UV radiation of the QSO, then
the extent of the proximity effect can be used to put limits on J v (Carswell
et al. 1987; BDO). Qualitatively, the larger J v is, the smaller the "sphere
of influence" of a particular QSO will be, that is, the smaller the region
will be in whiCh the QSO's ionizing radiation dominates the ambient back-
ground. By estimating the number of ionizing photons from the QSO fro~
spectrophotometry, one can then use the observed extent of the proximity
effect to measure J",. One further expects that at a fixed redshift, where J v
is by definition fixed, brighter QSOs will have "stronger" proximity effects
than fainter ones. Even if photoionization is not the dominant cause of the
proximity effect,.one can still place a lower bound to J v since other processes
can only add to the ionization of clouds (BOO).
Carswell et al. (1987) first discussed using the proximity effect to estimate
J v and concluded that it was too large to be entirely from QSOs. They
observed the very luminous, Z = 3.78 quasar PKS 2000-330, and argued
that the ambient J", had to be large, or else a stronger effect would have
been seen. Bajtlik, Duncan, & Ostriker (1988) compiled a sample of Lya
lines of 19 QSOs from the literature, and constructed an elegant formalism
for analyzing the proximity effect. They also found that the J v derived was
larger than the estimates of the quasar background at Z >=:::; 3 of Bechtold et
al. (1987). Lu, Wolfe, & Turnshek (1991) found a similar result in a sample
of 38 QSOs. However, these estimates of the quasar background were based
on guesses for the QSO luminosity function at Z > 2, which turned out to
be at least an order of magnitude too low.
A larger sample has been constructed by Bechtold (1993) especially to
study the proximity effect. It includes new, moderate-resolution spectra of 34
QSOs with Zem > 2, most with Zem ~ 3, obtained at the MMT and Palomar
5m (see also Dobrzycki & Bechtold 1993). Many ofthe objects are relatively
faint (V < 18) in the continuum, and so had been neglected in previous
 567

absorption-line studies. However, the lines in the vicinity of the strong Ly-o
emission line can be easily measured in faint objects. Combined with spectra
from the literature, the sample contains 87 QSOs with 1.6 < Zem < 4.1, and
3117 lines. By comparison, earlier studies of the proximity effect were based
on many fewer lines: Murdoch et al. (1986) considered 11 QSOs with 277
lines; BDO included 19 QSOs with 470 lines; Lu, Wolfe, & Turnshek (1991)
included 38 QSOs and 950 lines. The low-redshift (z = 0 - 1) results from
the HST QSO Absorption line Key Project (Bahcall et al. 1992) are based
on 11 QSOs and 43 lines.
The expected number of clouds near each quasar can be estimated from
the statistics ofthe clouds far from the quasar. In the usual parameterization
of the statistics for Zabs <t: Zem, one can count the number of clouds per unit
redshift, dN /dz, with rest equivalent width larger than some threshold, and
write

-dN
dz
= Ao(1 + z)'Y (4)

The maximum likelihood technique of Murdoch et al. (1986) is used to

estimate '"Y, and a Kolmogorov-Smirnov test is used to verify the fit. For
o S; qo S; 0.5, if there is no evolution in cloud properties, then '"Y < 1.
Previous studies have found that '"Y ~ 2 to 3 for the Ly-o forest, indicating
a steep evolution in the sense there was more absorption in the past (Lu,
Wolfe, & Turnshek 1991, and references therein).
The results for the new sample are shown in Figure 1, where log( dN / dz) is
plotted as a function of log( 1 + z) for lines with rest equivalent width greater
than 0.32 A. The data have been binned for presentation only. The best fit
to the HST data alone (the 4 lowest redshift points) is '"Y = 0.0 ± 1.0, and is
shown by the dotted line. It is interesting that the high-redshift clouds do
not fall on this line, suggesting a different population of clouds at Z < 1 and
Z > 1 (Weymann, this volume). Formally, however, a single power law of the
form given in Equation (4) fits all the data adequately, with '"Y = 1.26±0.13,
with a K-S probability PKS = 0.61 that Equation (4) is a good fit. Curiously,
this is only 2<7 from the "no-evolution" value of'"Y = 1, and the probability
that'"Y = 1 for the whole sample is a marginally acceptable PKS = 0.089.
Getting back to the discussion of the proximity effect, I have fitted the
high-redshift data only (z > 1.6) in order to describe the statistics of the
clouds at Zab6 <t: Zem. The result is that '"Y = 1.69 ± 0.25 and PKS = 0.41.
This sample may be subject to subtle effects of line blending, although it is
more homogeneous, and contains generally higher signal-to-noise data than
the previous compilations of BDO or Lu, Wolfe, & Turnshek (1991). For a
detailed discussion of the sample selection and line blending, see Bechtold
(1993) and Dobrzycki & Bechtold (1993). However, the value of'"Y = 1.69 is a
568

Lya Forest Clouds

0.5
o 0.2 0.4 0 .6

log( 1 +z)

Fig. 1. Ground-based and HST results for the number of Lya forest lines per unit
redshift as a function of (1 + z), for lines with rest equivalent width W>. > 0.32 A
and Zab, ~ Zem. The dotted line indicates the best fit 'Y = 0.0 to the HST Key
Project data of Bahca.ll et at. (1993). The high redshift data are from Bechtold
(1993).

useful empirical description of the data which can then be used to investigate
the proximity effect; what the "true" 'Y may be is another issue.
The first question to ask is whether there is in fact a significant deficit
of lines with Zab. ~ Zem. Figure 2 shows a plot of (Npred - Nobs)jNpred
versus luminosity distance from the QSO in Mpc (Ho = 100 km s-1 Mpc- 1,
qo = 0.5), where Npred is the number of lines in each bin which should have
been detected, given the best ,fit to Equation (4), and Nobs is the actual
number of lines observed. Clearly, there is a strong deficit over scales of ",5
Mpc, with the first point being 7.6 (T different from zero.
Another way to plot the data is shown in Figure 3. Plots of Npred-Nobs for
individual QSOs versus their Lyman-limit continuum luminosity and emis-
 569

0.8

0.6

-0.2
o 5 10 15
ilR(Mpc)

Fig. 2. For the high-redshift Lya forest clouds with W A > 0.32 A, a plot of the
fractional deficit of lines versus distance from the QSO. A clear deficit of lines on
scales of 3 - 5 Mpc is seen.

sion line redshift are shown, where the lines within ~z = 0.15 of the emission
line redshift are used. There is a large scatter in the proximity effect from
object to object, owing to the small number of lines expected for each ob-
ject. Spearman rank coefficient tests on these plots show that (Npred - Nobs)
correlates slightly with luminosity but not redshift, which would support
the photoionization explanation for the proximity effect. However, the trend
is not particularly striking to the eye. These plots, however, do show that
the proximity effect does not depend on radio loudness, which one might
expect if the radio properties of QSO are related to the cluster environment
(Bechtold 1987; Ellingson, Yee, & Green 1991, and references therein).
There are 2 QSOs . in the sample which have recently been found to
be gravitationally lensed (UM 673 = 0142-100, Surd~j et al. 1987; and
1208+101, Magain et al. 1992 and Maoz et al. 1992). H lensing has bright-
 570

20
11'"" = 0.32 A
6z :> 0.15
o : Radio Loudness ~ 1
• : All Others
.. 10

+
D
6 : Lenses
0
Z

Z '"
..,..
.
0
++
-10 L-J--L~~_ _L-~~-L~~_ _L-L-~-L-L~_ _L-~~-L~__

30.5 31 31.5 32 32.5

log LLyman Umlt

20
11'.",. = 0.32 A
6% :> 0. 15
o : Radio Loudness ~ 1

+t
• : All Others
.
D
0
10
6 : Lenses
Z

..,..
Z '"
0

2 2.5 3 3.5 4

Fig. 3. The deficit of Lyo: forest lines as a function of Lyman-limit luminosity and
emission-line redshift for individual objects in the high redshift sample.
 571

ened the QSO continuum, then these two QSOs are intrinsically fainter than
they appear and should have corr~spondingly weaker proximity effects for
their apparent luminosity (BDO). They are shown in Figure 3 as filled trian-
gles; neither stands out in the large scatter. A larger sample of high-redshift
lenses, or better limits on the possibility of lensing for the objects in this
sample would be of interest. Fifteen QSOs in this sample have already been
the subject of searches for multiple images, with negative results (Crampton
et al. 1989; Bahcall et al. 1992; Maoz et al. 1992).
The BDO photoionization model can be applied to this sample to derive
limits on J -21 as a function of redshift, for the redshifts studied, 1.6 < z <
4.1. BDO considered the number of lines, X-y, per co-moving, co-evolving
redshift interval, where

X-y = f (1 + z)'Y dz , (5)

so that Equation (4) would become

dN
dX-y = Ao , (6)

if there were no proximity effect. BDO showed that for highly ionized clouds
and for a neutral hydrogen column density distribution that is a power law
of index f3 = 1.7, this becomes

dN -0.7
dX
-y
= Ao [1 + w( z)J , (7)

where
FQ
w=-- (8)
47rJ"

and FQ is the Lyman-limit flux of each QSO as seen by each cloud, and J"
is the specific intensity of the UV background at the Lyman limit at the
redshift of the cloud.
In other words, a plot of dN / dX-y versus log w is a plot of line density
versus distance from the QSO, scaled according to the ratio of the Lyman-
limit flux of the QSO at that distance, to J" at that redshift. The advantage
of this analysis is that the proximity effect should be the most apparent; the
disadvantage is that a form for J" as a function of redshift must be assumed
to plot the data. A clear description of how to actually make these plots is
given by BDO.
For the redshift dependence of J", I have taken the form

(9)
572

8 All QSOs
Wthr = 0.32. J..
?f
"-
6
4
J" ex (l+z)

Z
"0
2
0
-4 -2 0 2

8 All QSOs
Wthr = 0.32 J.. .,
?J
"-
6
4'
J,,(z) ex (l+zf

Z
"0
2
0
-4 -2 0 2
log (CJ)

Fig. 4. Plots of dN/dX., versus logw, for two acceptable forms for the redshift
dependence of J v, J v oc (1 + z) -7 and J v Q( (1 + Z)4. The points are the data, and
the solid line is the model prediction if photoionization by the QSOs is causing the
proximity effect.

where J vo and j are constants. This form can be related analytically to mod-
els for the Lya forest clouds - either models where the clouds are confined by
CDM mini-halos, or models where the clouds are confined by IGM pressure
(Ikeuchi & Turner 1991; Charlton, Salpeter, & Hogan 1993). In fact, given
1 of Equation (4) and j of Equation (9), one could in principle distinguish
between these very different physical pictures. Unfortunately, the proximity
effect data does not constrain the index j very well, and acceptable fits are
possible for -7 < j < 4 (see Figure 4).
The reason for the poor constraints on j is shown in Figure 5. The best fit
 573

..-..-18
s...
Q)
+>
rn
~19
:r:: ............

---- --
............
N"""- ............

"""-
8-20
C)
......
.....
------ -~-~

--.. --.--0 .-,--.--.--' ___ __-.,;

~.--.
C)
.-.,. ........ ---.------.

--.--_--0
Q)

~-21 <
rn
0.0
.-.-. a( 1 +Z)4
s...
Q)
'-"-22
~
"'""?

0.0
~-23
2 2.5 3 3 .5
z
Fig. 5. J v at the Lyman limit as a function of redshift, for 3 models which fit the
proximity effect data - see Equation (9) with j
~,
-7,0 and 4. =

J", is plotted for j = -7,0 and 4. The EUV background is well constrained
at z :::::; 3, but not at z :::::; 2. Constraining J", at z :::::; 2 is just a matter
of observing more quasars, and a program is underway, in collaboration
with Adam Dobrzycki, using the MMT Blue Spectrograph to do just that.
Unfortunately, the HST key project will probably not be able to observe
enough QSOs to put interesting limits on J", at lower redshifts in the near
future, but any limit on J", at lower redshifts is interesting.
There are several assumptions that went into the derivation of J", given
in Figure 5, and two are worth mentioning. First, the clouds were assumed
to be highly ionized, so that N(H J) = No(l + w)-l, where N(H I) is the
neutral hydrogen column density of the cloud, and No is the neutral hydro-
574

gen column density the cloud would have had in the absence of the QSO.
If the cloud is neutral, then the change in neutral fraction with increasing
UV flux is slower than this (see Figure 7 of Pettini et al. 1990). That is,
if the clouds are very neutral, they have to be blasted with UV photons to
have any observable proximity effect at all. This means that the ambient
flux must be very low, for the quasar flux to be relatively so important.
Bechtold & Wolfe (1993) derive the equivalent to Equation (7) for neutral
clouds. The interest in neutral clouds originates in the suggestion by Pettini
et al. (1990) that the kinetic temperatures of the Lya forest clouds are too
cool to be easily accounted for in the standard cloud models. In any event, if
for example the neutral fraction n(H l)jn(H) = 0.95 for the clouds, then the
new data sample implies J- 21 = 10- 2 (for J v constant with z). This is an
order of magnitude smaller than the contribution to the EUV background
for the QSOs we see, so it is unlikely that the clouds are this neutral.

The second interesting question is whether or not we know the actual sys-
temic redshifts of the QSOs. It has been known for some time from detailed
studies of QSO emission lines that there are systematic shifts between lines
of low-ionization and lines of high-ionization species (Gaskell 1982; Wilkes
1984). For the high-red shift QSOs used in the Lya forest studies, the red-
shifts are typically derived from optical spectra of Lya, C IV and Si IV
emission lines. However, recently, near-infrared spectra of Ha and Mg II
show that Lya can be shifted 1000 - 2000 km S-1 to the blue of the ac-
tual systemic redshift of the QSO (Espey 1989; Corbin 1990; Carswell et
al. 1991) - see Donahue & Shull (1991) for a calculation of how this effect
influences estimates of the ionizing radiation field incident on the putative
narrow Lya clouds towardPKS 2206-199 (Pettini et al. 1990). For Lya at
z = 3, Az = 0.05 corresponds to AR = 3.5 Mpc, or Av = 3800 km S-1,
so this is an important consideration. Note that if the real redshift were
redward of Lya then the entire proximity effect could be an artifact of the
systematic underestimate of Zem'

How would the systematic underestimate of Zem change the estimates of

J v from the proximity effect? If the QSO redshift is really higher, then the
clouds are farther away from the QSO than has been assumed, and so the
QSO influence is less at the cloud's position. Thus, the estimates of J v should
be lower. If I assume that all the QSOs in my sample have redshifts that are
1500 km s-1 too low, and that J v is constant with redshift, then J- 21 = 1,
approximately a factor of 3 lower than the J -21 derived neglecting this effect
(Figure 5). It would be of interest to obtain IR spectra of the QSOs in this
sample, or some large subset of this sample, to get redshifts from the Balmer
Series, to see if the trend is universal.
 575

3.2.2. Lya emission limits

Recombination Lya emission from the Lya forest clouds or metal-line sys-
tems can be used to place limits on the ionizing background (Hogan & Rees
1979; Hogan & Weymann 1987). Surveys for such emission from blank. sky
have so far been negative (Pritchet & Hartwick 1987; Cowie 1988; Lowen-
thal et al. 1990). Lowenthal et al. 1990 placed a limit on Lya emission
from one quasar absorption line cloud which implied an upper limit to
J", < 19 X 10- 2°,8-1 ergs cm- 2 s-1 sr- 1 Hz- 1 at z = 2.912 where ,8 is
the fraction of ionizing photons which are converted into Lya. This is some-
what higher than the values discussed in the previous section, but those are
lower limits to the flux. This method has the advantage of being relatively
free of assumptions compared to other techniques.

4. Summary

Table 1 gives a summary of the various calculations and empirical measures

of J -21 discussed above, with generous estimates of the uncertainties. At
high redshift, the estimates for J -21 from the proximity effect are now in
reasonable agreement with the expected contribution from luminous QSOs,
given recent discoveries of large numbers of QSOs with z = 3 to 5. A large
population of obscured QSOs at z = 3 to 5, or yet-to-be-discovered objects
aihigher redshift, is not required. Clearly better observational constraints
are of interest. For a number of the techniques discussed improvements in
the limits on J -21 are possible.

Acknowledgements
I thank Mike Shull and Harley Thronson for their gracious hospitality, and
for providing encouragement and financial support to the students whose
presence made the meeting so enjoyable. The written version of this review
was influenced by numerous conversations with colleagues at the meeting.
I also thank Adam Dobrzycki, Art Wolfe, Steve Shectman, and Wal Sar-
gent for their contributions, and the HST Key Project group for sharing re-
sults before publication. Financial support was provided by NSF grants RII-
8800660, AST-9058510, and INT-901OS83 to the University of Arizona. Ob-
servations were made at the Multiple Mirror Telescope Observatory, which is
a joint facility of the University of Arizona and the Smithsonian Institution.

References
Antonucci, R. R. J., & Miller, J. S. 1985, A,. J., 287, 621.
Arons, J., & Wingert, D. W. 1972, Ap. J., 1'17, 1.
Atwood, B., Baldwin, J. A., & Carswell, R. F. 1985, Ap. J., 292, 58.
576

. TABLE I
Summary of Metagalactic EUV Flux, J -21

Source z=o z=3

Calculated Source Contributions

QSOs 0.005 0.1 - 0.7

AGNs < 0.01 < 1.0
Young Galaxies < 0.1 <4
Decay of V e , etc. 100

Empirical Limits

Ho: from HVC's "'0.06

H I Edges "'0.04
Direct Measure < 0.5
Lyo: Emission < 200
Proximity Effect > 1-12

Bahcall, J. N., Maoz, D., Doxsey, R., Schneider, D. P., Bahcall, N. A., Lahav, 0., &: Yanny,
B. 1992, Ap. J., 387, 56.
Bahcall, J. N. et al. 1993, ApJS, in press.
Bajtlik, S., Duncan, R. C., &: Ostriker, J. P. 1988, Ap. J., 327, 570 (BDO).
Barcons, X., Fabian, A. C., &: Rees, M. J. 1991, Nature, 350, 685.
Bechtold, J. 1987, in Proc. 3rd lAP Workshop on High Redshift and Primeval Galaxies, Ed.
J. Bergeron, D. Kunth, B. Rocca-Volmerange, &: J. Tran Thanh Van (Paris: Editions
Frontieres), 397.
Bechtold, J. 1993, Ap. J. S., in press.
Bechtold, J., Weymann, R. J., Lin, Z. and Malkan, M. 1987, Ap. J., 315, 180.
Bechtold, J., &: Wolfe, A. M. 1993, in preparation.
Bergeron, J., &: Stasinska, G. 1986, A. Ap., 169, 1.
Bloemen, J. B. G. M. 1987, Ap. J., 322, 694.
Boisse, P., &: Bergeron, J. 1988, A. Ap., 192, 1.
Bowyer, S. 1991, ARA&:A, 29, 59.
Boyle, B. J., Jones, L. R., &: Shanks, T. 1991, MNRAS, 251, 482.
Carilli, C. L., van Gorkhom, J. H., &: Stocke, J. T. 1989, Nature, 338, 134.
Carswell, R. F., Webb, J. K., Baldwin, J. A., &: Atwood, B. 1987, Ap. J., 319, 709.
Carswell, R. F., Lanzetta, K. M., Parnell, H. C., &: Webb, J. K. 1991, Ap. J., 371, 36.
Cen, R., &: Ostriker, J. P. 1993, preprint.
Charlton, J. C., Salpeter, E. E., &: Hogan, C. J. 1993, Ap. J., 402, 493.
Code, A. D., &: Welch, G. A. 1982, Ap. J., 256, 1.
Colgan, S. W. J., Salpeter, E. E., &: Terzian, Y. 1990, Ap. J., 351, 503.
Corbelli, E., Schneider, S. E., &: Salpeter, E. E. 1989, A. J., 97, 390.
Corbelli, E., &: Schneider, S. E. 1990, Ap. J., 356, 14.
Corbin, M. R. 1990, Ap. J., 357, 346.
Cowie, L. L. 1988, in The Post-Recombination Universe, ed. N. Kaiser &: A. N. Lasenby,
(Dordrecht: Kluwer Academic), 1.
Cowsik, R. 1977, PhY6. Rev. Lett., 39, 784.
 577

Crampton, D., McClure, R. D., Fletcher, J. M., & Hutchings, J. B. 1989, A. J., 98, 1188.
Davidsen, A. F. et a1. 1991, Nature, 351,128.
De Rejula, A., & Glashow, S. 1980, Phys"--Rev. Lett., 45,942.
Dobrzycki, A., & Bechtold, J. 1993, in preparation.
Donahue, M., & Shull, J. M. 1987, Ap. J. (Letters), 323, L13.
Donahue, M., & Shull, J. M. 1991, Ap. J., 383, 511
Efstathiou, G., & Rees, M. J. 1988, MNRAS, 230, 5P.
Ellingson, E., Yee, H. K. C., & Green, R. F. Ap. J., 371,49.
Elvis, M., Fiore, F., Wilkes, B., McDowell, J. and Bechtold, J. 1993, preprint.
Espey, B. R., Carswell, R. F., Bailey, J. A., Smith, M. G., & Ward, M. J. 1989, Ap. J.,
342,666.
Fabian, A. C., Naylor, T., & Sciama, D. W. 1991, MNRAS, 249, 21P.
Fall, S. M., & Pei, Y. C. 1993, Ap. J., 402, 479 .
Field, G. B. 1959, Ap. J., 129,536.
Fix, J. D., Craven, J. D., & Frank, L. A. 1989, Ap. J., 345, 203.
Fransson, C., & Chevalier, R. A. 1985, Ap. J., 269, 35.
Gaskell, C. M. 1982, Ap. J., 263,483.
Gunn, J. E., & Peterson, B . A. 1965, Ap. J., 142, 1633.
Heisler, J. and Ostriker, J . P. 1988, Ap. J., 332,543.
Henry, R. C. 1991, ARA&A, 29, 89.
Hogan, C. J., & Rees, M. J. 1979, MNRAS, 188, 791.
Hogan, C. J ., & Weymann, R. J. 1987, MNRAS, 225, IP.
Ikeuchi, S., & Ostriker, J. P. 1986, Ap. J., 301,522.
Ikeuchi, S., & Turner, E. L. 1991, Ap. J., 381, Ll.
Irwin, M., McMahon, R. G., & Hazard, C. 1991, in ASP Conf. Ser. 21, The Space Distri-
bution of Quasars, ed. D. Crampton (San Francisco: ASP), 117.
Jakobsen, P. 1980, A. Ap., 81, 66.
Kimble, R., Bowyer, S., & Jakobsen, P. 1981, Phys Rev. Lett., 46, 80.
Krolik, J. H. 1990, in The Interstellar Medium in Galaxies, ed. H. A. Thronson and J. M .
. . Shull, (Dordrecht: Kluwer), 239.
Kurt, V. G., & Sunyaev, R. A. 1967, Cosmic Research, 5, 496.
Lockman, F. J., Jakoda, K., & McCammon, D. 1986, Ap. J., 302, 432.
Loeb, A. 1992, preprint.
Lowenthal, J . D., Hogan, C. J., Leach, R. W ., Schmidt, G. D., & Foltz, C. B. 11'90, Ap.J.,
357,3.
Lu, L., Wolfe, A. M., & Turnshek, D. A. 1991, Ap. J., 367, 19.
Madau, P. 1992, Ap. J. (Letters), 389, L1.
Magain, P., Surdej, J., Vanderriest, C., Pirenne, B., & Hutsemekers, D. 1992, A. Ap., 253,
L13.
Maloney, P. 1993, Ap. J., 414, in press.
Maoz, D., Bahcall, J . N., Schneider, D. P., Doxsey, R., Bahcall, N. A., Filippenko, A. V.,
Goss, W. M., Lahav, 0., & Yanny, B. 1992, Ap. J., 386, Ll.
Martin, C., & Bowyer, S. 1989, Ap. J., 338, 677.
Martin, C., Hurwitz, M., & Bowyer, S. 1991, Ap. J., 379, 549.
Meiksin, A., & Madau, P. 1993, preprint.
Melott, A. L., & Sciama, D. W . 1981, Phys. Rev. Lett., 46, 1369.
Miralde-Escude, J., & Ostriker, J . P. 1990, Ap. J., 350, 1.
Miralde-Escude, J., & Ostriker, J. P. 1992, Ap. J., 392,15.
Murdoch, H. S., Hunstead, R. W., Pettini, M., & Blades, J. C. 1986, Ap. J., 309, 19.
Murthy, J., Henry, R. C., Feldman, P. D., & Tennyson, P. D. 1990, A. Ap., 231, 187.
Najita, J ., Silk, J., & Wachter, K . 1990, Ap. J., 348, 383.
Ostriker, 1. P., & Heisler, J. 1984, Ap. J., 278, 1.
Ostriker, J. P., Vogeley, M. S., & York, D. G. 1990, Ap. J., 364, 405.
Paresce, F. 1990, in The Galactic and Extragalactic Background Radiation, ed. S. Bowyer
and C. Leinert, 307. -
578

Paresce, F., &; Jacobsen, P. 1980, Nature, 288,119.

Paresce, F., McKee, C., &; Bowyer, S. 1980, Ap. J., 240, 387.
Pettini, M., Hunstead, R . W., Smith, L., &; Mar, D. P. 1990, MNRAS, 246, 545.
Pritchet, C . J., &; Hartwick, F. D. A. 1987, Ap. J., 320, 464.
Rees, M . J., &; Setti, G. 1970, A. Ap., 8, 410.
Reimers, D., Vogel, S., Hagen, H.-J., Engels, D., Groote, D., Wamsteker,W., Clavel, J., &;
Rosa, M. R. 1992, Nature, 360 561.
Reynolds, R. J. 1987, Ap. J., 323, 553.
Reynolds, R. J., Magee, K ., Roesler, F. L., Scherb, F., &; Harlander, J. 1986 Ap. J., 309,
L9.
Sargent, W. L. W., Young, P. J., Boksenberg, A., &; Tytler, D. 1980, Ap. J. S., 42, 41.
Schmidt, M ., Schneider, D. P., &; Gunn, J. E. 1991, in ASP Conf. Ser. 21, The Space
Distribution of Quasars, Ibid., 109.
Schneider, D. P., Schmidt, M., &; Gunn, J. E. 1989, A. J., 98, 1507.
Schneider, S. E., Helou, G., Salpeter, E. E ., &; Terzian, Y. 1983, Ap.J. (Letter8), 273, Ll.
Sciama, D. W. 1990a, Ap. J., 364, 549.
Sciama, D. W. 1990b, PhY8. Rev. Lett., 65, 2839.
Sciama, D . W. 1990c, Nature, 348,617.
Sciama, D. W. 1991, Ap. J., 367, L39.
Sembach, K. L., &; Savage, B. D. 1992, Ap. J. S., 83, 147.
Shapiro, P. R., &; Giroux, M. L. 1987, Ap. J. (Letter8), 321, L107.
Shapiro, P. R., &; Giroux, M. L. 1989, in The Epoch of Galaxy Formation, ed. C. S. Frenk,
R. S. Ellis, T. Shankes, A. F. Heavens, &; J. A . Peacock (Dordrecht: Kluwer).
Shectman, S. A. 1973, Ap. J., 179, 681.
Silk, J., &; Sunyaev, R. A. 1976, Nature, 260, 508.
Songaila, A., Bryant, W., &; Cowie, L. 1. 1989, Ap. J. (Letter8), 345, L71.
Songaila, A., Cowie, L.L., &; Lilly, S. J. 1990, 348, 371.
Stecker, F. W. 1980, PhY8. Rev. Lett., 45, 1460.
Steidel, C. C., &; Sargent, W. L. W. 1989, Ap. J. (Letter8), 343, L33.
Stocke, J. T., Case, J., Donahue, M., Shull, J. M., &; Snow, T. P. 1991, Ap. J., 374,72.
Surdej, J. et al. 1988, A. Ap., 198, 49.
Terasawa, N . 1992, Ap. J. (Letter8), 392, 115.
Tinsley, B. M. 1972, Ap. J ., 178, 319.
Tripp, T. M., Green, R. F., &; Bechtold, J. 1990, Ap. J. (Lett), 364, L29.
Turner, E. L. 1991, A. J ., 101, 1.
Tytler, D. 1987, Ap. J., 321, 69.
Warren, S. J., Hewett, P. C., &; Osmer, P. S. 1991, in ASP Conf. Ser. 21, The Space
Distribution of Quasars, Ibid., 139.
Weymann, R. J. 1967, Ap. J., 887 .
Weymann, R. J., Carswell, R. F. , &; Smith, M. G. 1981, ARABA, 19, 41.
Wilkes, B. J. 1984, MNRAS, 207,73.
Wright, E. L. 1986, Ap. J., 311, 156.
Wright, E. L. 1990, Ap. J ., 353, 411.
Wright, E. L. 1992, Ap. J., 391, 34.
York, D. G. 1982, ARABA, 20, 221.
Index
3C 273:
Lya absorption 228
Absorption lines:
metal-line systems 238
redshift distribution 238
Absorption systems:
cross section 266
evolution 264
Abundances: 537
C, Mg, Si, Fe, Zn, Cr 254
Cr, Zn 255
Fe, 0 321,410
depletion 255
ion~ation correction 255
high-z gas 254
Active galactic nuclei (AGNs):
159-190,369-382,403,490,
561ff
hard spectra 168
radio luminosity function 444
Z > 3440
Advanced X-ray Astrophysics
Facility (AXAF): 160
Age dating: 153
Ambipolar diffusion: 501
ASTRO-D: 186
Atomic gas: 295-304, 345-367,
483ff
extent 346
Bars: ~ Galaxies: bars
Baryonic cooling: 67, 69
Baryonic matter: 230
Baryon density: 123
579
Baryons: 383ff
Biasing: 19, 82
Bivariate brightness function:
105-125
Broad-Band X-ray Telescope
(BBXRT): 160ff, 393ff
Butcher-Oemler effect: 130, 135ff,
145, 148
C IV absorption: 138,220, 239ff,
266
Chandrasekhar mass: 528
Chemical evolution: 537ff
Chemical potential (p): 33,35
Clustering amplitude: 60
Clusters:
beta-problem 410ff, 486
chemical enrichment 426
cold gas 427
evolution 409-431
hydrostatic equilibrium 412
masses 85, 412
optical filaments 399ff
radio sources 399, 418
substructure 416
velocity dispersion 417
x-ray surface brightness 411
x-rays 172
Cold dark matter (CDM): 3-
25,38,46,47,59-68,62,
69-89,193,201,307,310,
421,424
Coma cluster: 412ff
Compton y parameter: 33,35
580
Comptonization: 161, 168
Cooling flows: 128, 184, 383-
407, 552
evolution 404
high-z 404
mass flow rates 415
star formation 403ff
Correlation function: 11
galaxy-galaxy 226
galaxy-Lya cloud 226
Cosmic Background Explorer
(COBE): 27-57,61
Cosmic background radiation:
3-25,
Cosmic infrared background:
27-57
Cosmic microwave background
(CMB): 27-57
anisotropy 29, 35, 36
dipole 35
power spectrum 40
temperature 35 , Far-Infrared Absolute Spec-
quadrupole 40, 43, 45ff
Cosmic strings textures: 14
Cosmological mass density: 245
Cosmology: 3-25
Damped Lya systems: 247,248,
285
Dark matter: 129, 192, 230, 409-
431, 536
x-ray probes 414
Density waves: 513
Diffuse Ha emission: 44
Diffuse Infrared BackgrOlmd
Experiment (DIRBE):
29, 47ff
Diffuse Microwave Radiome-
ter (DMR): 29, 36ff
Disk accretion: 504
Disk-halo connection: 490
Disks, circumstellar: 516ff
Doppler parameter: 196
Dust:
distinction 285
polarization 481
scattering 481
Einstein Deep Survey (EDS):
175
Einstein Medium Sensitivity
Survey (EMSS): 175
Einstein observatory: 160ff,386,
410
Elliptical galaxies:
winds 488
Equivalent widths: 197
Evolution:
gas 237
number 251
Evolutionary tracks: 93
EXOSAT: 167,422
Extragalactic background: 123
Fanaroff-Riley clusters: 418
Faraday rotation: 259, 397ff
trometer (FIRAS): 29,
33ff
Fe K-lines: 469
Fe XVII: 391ff
Free-fall collapse type: 502
Galactic center: 469
Galactic fountains: 455-497
Galactic halo gas: 265
Galactic winds: 130, 306
Galaxies:
accretion 77
ages 23
barred 299
bars 334
bulge 108, 311ff
clusters 17, 19,69-89,82, 131,
145,305-326,345-367,352ff,
542
collisions 327-344
color evolution 275

disks 72, 305, 334, 345-367
dust emission 44, 552
dusty 129
dwarf spheroidal 540ff
dwarfs 23, 79, 224,269, 276ff,
307, 312, 345-367, 352,
354, 356ff,476, 484,487
dynamics 305-326
E & A galaxies 555
early-type 131, 327-344, 383-
407
efficiency 80
ellipticals 23,98,305-326,370
evolution 263ff, 487, 499-531
extent 346
far-infrared 551
field 132, 143-149
formation 21, 59-68, 69-89, 80,
128, 131-140, 132, 143-
149, 253, 305-326, 327-
344, 360ff
free-free emission 44
halo formation 66, 69-89
halos 185, 268, 333, 537
HI envelopes 347
Holmberg relation 268
hot stars 562
Hubble sequence 303
groups 268, 354
infrared emission 299
interactions 298, 554
interstellar medium 455-497
IR-luminous 473
irregular 98, 476, 484, 549
lenticular 77
low surface brightness 105-125,
131-140, 269, 310ff, 354
luminosity function 181
mass function 77ff
mergers 21, 75, 305-326, 309,
327-344, 438, 554
metallicity 487
581
morphology 135,295-304,305-
326
nucleus 108
number counts 131-140, 143-
149
optical colors 401ff
post-starburst 128, 135ff, 137
primeval 127-130, 151-154, 155-
158, 360ff, 534, 542ff, 548ff
radii 281
radio: see Radio galaxies
rotation 72
satellites 310
Seyfert 159-190,370,433-453
spheroid 537
spin 334
spiral 305-326
star formation 22,127-130,327-
344, 403ff, 455-497
starbursts 22, 164, 178, 455-
497,534, 549, 551
surface brightness 336
surveys 131-140, 143-149
synchrotron emission 44
tidal effects 327-344, 364
tidal torques 306
ultraluminous 455-497
UV spectra 95
virialization 409-431
winds 22
x-ray emission 340
z < 1 267
z > 2 283
z> 3 360
Galaxy clusters: 409-431
Galaxy formation: 413,487,533-
558
gas cooling 314
secondary infal! 312
virialization 312, 318
Galaxy surveys: 105-125
Galaxy voids: 353
582
Gamma Ray Observatory (GRO):
186
Gas: see also Molecular Gas,
Atomic Gas; Intergalac-
tic Medium
abundances 384
cooling 384
densities 479
heating 409-431
kinematics 47 4ff
pressures 479
thermal equilibrium 377
Gastrophysics: 3-25 ,
Giant molecular clouds: 319
Ginga: 18, 167, 393
"Globs": 69-89
Globular clusters: 537, 553
Gravitational lensing: 139,414,
569
Gunn-Peterson effect: 21 , 129,
156, 191,233
Halo: 370
H I disks: 352
Hierarchical clustering: 86,309
High-z galaxies: 263-293
Holmberg radius: 243
Hot dark matter (HDM): 13,
47, 62
Hubble constant : 6,
Hubble sequence: 295-304, 305-
326, 328
Hubble Space Telescope (HST) :
220, 553
Hydrodynamics: 205, 459
simulation 516
Impact parameter: 278
Inflation: 3-25,
Inflationary cosmology: 46 , 60
Infrared:
spectroscopic searches 286
Infrared cirrus: 51
Infrared emission:
[ell] reI] [NIl] eo 36
[NIl] 44
Initial Mass Function (for stars):
91,323,458, 519ff, 526ff,
545
Intergalactic medium (IGM) :
23, 128, 129, 155- 158, 159-
190,191-212,213-235,345-
367,371, 488, 559-578
chemical enrichment 486ff
expanding shells 202
heating 486ff
photoionization 156
Interstellar medium (ISM) : 370
cloudy 371
enrichment 143ff
Intracluster medium (ICM) :
383-407, 409-431
evolution 419ff
Ionization parameter: 369
Ionized gas: 247, 345-367
IRAS: 60
Isochrones: 93
IUE: 96, 249
Inverse compton scattering: 481
Jeans mass: 520, 524
Jets: 369
cocoons 379
J upiters: 202
King model: 388
Large Magellanic Cloud (LMC):
539
Large-scale structure: 3-25,59-
68
Lindblad resonance: 514
LINERs: 491
Local group: 539ff
Low surface brightness (LSB)
galaxies: see Galax-
ies, low surface bright-
ness
Luminosity functions: 105-125,

131-140, 243, 278, 280,
357
Lya clouds:
emission 575
gravitation confinement 231
pressure confinement 231
redshift evolution 194
Lya absorption: 237
Lya clouds size: 198
Lyman continuation searches:
287
Lyman a forest: 23, 128, 129,
191-212, 213-235, 559-578
21-cm searches 224
cloud sizes 218
clustering 218
column densities 216
correlation 217, 232
Doppler parameter 217
evolution 215 , 221
galaxies 222
HI columns 232
metal abundances 219
sizes 232
Lya emission: 283
Lyman limit: 230, 237-262, 266
Malmquist bias: 109
Magnetic fields: 259, 397, 483 ,
501
dynamos 260
seed fields 260
Mass density: 251
Merger:
hypothesis 338
statistics 329
Metal-line absorber: 239ff
Metallicity evolution: 248
Mg II absorption: 138,225 , 239ff,
263-293, 544
evolution 271, 280
"impact parameter" 274
Mg II systems: 121
583
Milky Way: 75, 315ff, 317ff, 537ff
bulge 320
spiral 320
structure 320
Minihalos: 192
Molecular clouds: 295, 500ff
CO/Hz 296
cores 503
Molecular gas: 483ff
Molecules:
formation 524
gas 295-304
Hz abundan ce 256
H2 Lyman and Werner bands
257
role in protostars 524
N-body simulation: 331,419
Non-violent relaxation: 83
Nunlber counts: 122
Number density: 243ff
Number evolution: 281
Optical surveys: 263-293
Pencil-beam surveys: 185
Perseus cluster: 395 , 412ff
Photoionization: 200 , 373ff, 571
Polarization: 501
Population synthesis: 91-102,
151-154
Pre-main sequence stars: 510
evolution 525
Press-Schechter formalism: 64,
330 , 422
Primordial star formation: 523
Protogalaxies: 52, 128
virializa tion 63
Protostars: 504ff, 524
bipolar outflows 500
disk accre tion 499
gravitational instability 512ff
infall models 505
TR emission 505 , 510
magnetic wind 508
584

Proximity effect: 194, 199, 565ff SLING amplification: 514

Quasar-galaxy pairs: 364 Small Magellanic Cloud (SMC):
Quasars: 128 539
dust 561 Smoothed particle hydrody-
high redshift 559 namics (SPH): 70
redshifts 574 Spectral evolution: 91-102
Quasi-stellar objects (QSOs): Spectroscopic surveys: 263-293
23,155-158,191-212,378, Spiral galaxies:
559-578 HI edges 347, 564
absorption lines 121, 213-235, Starbursts: 101
237-262, 264, 363, 488 Star clusters: 540
broad absorption lines 374, 379ff Star counts: 315
correlation with galaxies 279 Star formation: 151-154, 155-
evolution 433-453, 443 , 567 158, 248, 295-304, 305-
luminosity function 433-453, 326, 356, 533-558
440, 443, 560ff efficiency 296, 299
optical luminosity function 437 global properties 545
surveys 288 Ret emission 547
x-ray 440, 443 rate 534ff
Radiation field: 156, 191-212 timescale 253
extragalactic 223, 350, 559- yield 298
578 Star formation rates: 91-102
Radiation pressure: 373ff Stars:
Radiative heating: 376ff evolution 499-531
Radio galaxies: 146, 151-154, formation 499-531
370, 378 formation: see also protostars,
Radio sources: pre-main-sequence stars
flat-spectrum 445 Stellar evolution: 91-102
Ram pressure stripping: 428 Stellar populations: 309, 315,
Ram pressure sweeping: 353 536
Resonance-line scattering: 374 Stellar spectral atlas: 96
Reynolds layer: 44 Stellar wind bubbles: 378
ROSAT: 161, 383,386,422 Stellar winds: 460
Sachs-Wolfe effect: 11 Sunyaev-Zeldovich effect: 6,
Scale invariance: 15 18
Schechter function: 78,243,280 Supernova rate: 464
Schmidt law: 252, 546 Supernovae: 455-497
"Second parameter" problem: Supershells: 456
537 blowout 462
Selection effects: 105-125 bow shocks 462
Seyfert 2's: 186 Superwinds: 455-497
Shocks: 369,455-497,461 Surface brightness: 105-125
 585

Surveys: spectrum 165ff

atomic gas 345-367 Zel'dovich pancake: 362
extragalactic CO 295
HI 120
medium sensitivity survey 441
metal-line absorbers 242
optical 417
optical line emission 485
quasars 435, 439
ROSAT quasar survey 441
VLA HI 362
x-ray 171ff, 416
Synchrotron emission: 481
Taurus molecular cloud: 502
Topological defects: 9
Tully-Fisher relation: 72
Turbulent mixing layers: 466 ,
478
Turbulent pressure: 503
T-Tauri stars: 507, 509
Universe, age: 6
UIT: 401
UV background: 433
UV radiation field: 229
Velocity width: 196ff
Virgo cluster: 352, 412ff
Voids: 198ff
Weakly interacting particles:
563
XMM: 186
X-ray absorption: 395ff
X-ray background: 159- 190,433,
489
X-ray binaries: 179, 469
X-ray sources: 467
X-ray spectra: 390ff
X-rays: 409-431
XRB:
discrete sources 174
energy content 162ff
isotropy 168
spectral paradox 172