Вы находитесь на странице: 1из 511
' -A Comprehensiue Introduction to . DIFFERENTIAL GEOMETRY VOLUME ONE Third Edition MICHAEL SPIVAK PUBLISH
' -A
Comprehensiue Introduction
to
.
DIFFERENTIAL GEOMETRY
VOLUME ONE
Third Edition
MICHAEL SPIVAK
PUBLISH OR PERISH, INC.
Houston, Texas
1999
ACKNOWLEDGEMENTS 1 am greatly indebted to without his encouragement these volumes would have remained a
ACKNOWLEDGEMENTS
1 am greatly indebted to
without his encouragement
these volumes would have remained
a short set of mimeographed notes
and
without his TE/X program they would
never have become typeset books
PREFACE T he preface to the first edition, reprinted on the succeeding pages, excused this
PREFACE
T he preface to the first edition, reprinted on the succeeding pages, excused
this book's deficiencies on grounds that can hardly be justified now that
these "notes" tmily have become a book.
At one time I had optimisticaliy planned to completely revise ali this material
for the momentous occasion, but I soon realized the futility of such an under-
taking. As I examined these five volumes, written so many years ago, I could
scarcely believe that I had once had the energy to learn so much material, or
even recall how I had unearthed some of it.
So I have contented myself with the correction of errors brought to my atten-
tion by diligent readers, together with a few expository ameliorations; among
these is the inclusion of a translation of Gauss' paper in Volume 2.
Aside from that, this third and final edition diffiersfrom the previous ones only
in being typeset, and with figures redrawn. I have merely endeavored to typeset
these books in a manner befitting a subject of such importante and beauty
As a final note, it should be pointed out that since the first volumes of this
series made their appearance in 1970, references in the text to "recent" results
should be placed in context.
HOW THESE NOTES CAME TO BE md how they did not come to be a
HOW THESE NOTES CAME TO BE
md how they did not come to be a book
For many years 1 have wanted to write the Great American Differential
Geometry book. Today a dilema confronts any one intent on penetrat-
íng the mysteries of differential geometry. On the one hand, one can
consult numerous classical treatments of the subject in an attempt to
f orm some idea how the concepts within it developed. Unf ortunately ,
a modern mathemat ical education tends to make classical mathematical
works inaccessible, particularly those in differentialgeometry. Onthe
other hand, one can now find texts as modern in spirit, and as clean in
exposition, as Bourbaki's Algebra. But a thorough study of these books
usually leaves one unprepared to consult classical works, and entirely
ignorant of the relationship between elegant modern construct ions and
their classicalcounterparts. Most students eventually findthat this
ignorance of the roots of the subject has its price -- no one denies that
modern definitions are clear, elegant, andprecise; it's just that it's
impossible to comprehend how any one ever thought of them, And even after
one does master a modern treatment of dif f erent ial geometry , other modern
treatments often appear simply tobe abouttotally different subjects.
Of course, these remarks merely mean that no matter how well some of the
present day texts achieve their objective, 1nevertheless f eel that an
introductionto differentialgeometry ought to have quite different aims.
There are two main premises on which these notes are based. The first
premise isthat it is absurdlyinefficient to eschewthe modern language
of manifolds, bundles, forms, etc., whichwas developedprecisely in
order to rigorize the concepts of classicaldifferentialgeometry.
Rephrasing every thing in more element ary tems involves incredible
contortions which are not only unnecessary , but misleading. The work of Gauss , f
contortions which are not only unnecessary , but misleading. The work
of Gauss , f or example, which uses inf initesimals throughout , is most
naturally rephrased in terms of differentials, even if it is possible
to rewrite it in terms of derivatives. For this reason, the entire
first volume of these notes is devoted to the theory of differentiable
manifolds, the basic language of modern differential geometry. This
language is compared whenever possible with the classical lanpage, so
that classical works can then be read.
The second premise for these notes isthat in order for anintroduction
to differential geometry to expose the geometric aspect of the subject,
an historical approach is necessary; there is no point in introducing
the curvature tensor without explaining how it was invented and what it
has to do with curvature. 1 personally felt that 1 could never acquire
a satisf actory understanding of diff erentiable geometry until 1 read
the original works. The second volume of these notes gives a detailed
exposition of the fundamental papers of Gauss and Riemann. Gauss' work
is now available inEnglish (General Investigations of Curved Surfaces;
Raven Press). There are also two Englishtranslations of Riemann's work,
but 1 have provided a (very f ree) translation in the second volume .
Of course, 1 do not think that one should f ollow al1 the intricacies of
the historical process, with its inevitable duplications and falce leads
What is intended, rather, is a presentation of the subject along the
lines which its development might have followed; as Bernard Morin said
to me, there is no reason, in mathematics any more than in biology, why
ontogeny must recapitulate phylogeny. When modern terminology finally
is introduced, it should be as an outgrowth of this (mythical) historical
development, And al1 the major approaches have to be presented, for they
were al1 related to each other, and al1 still play an important role .
At this point 1 am reminded of a paper described in Littlewood' S Mathematician' S
At this point 1 am reminded of a paper described in Littlewood' S
Mathematician' S Mi scellany . The paper began "The aim of this paper is
"
to prove
and it transpired only much later that this aim was not
achieved (the author hadn't claimed that it was) . What 1 have outlined
above is the content of a book the realization of whose basic plan and the
incorporation of whose details would perhaps be impossible ; what 1 have
written is a second or third draft of a preliminary version of this book,
1 have had to restrict myself to what 1 could write and learn about within
the present academic year, and allrevisions and corrections have hadto
be made within this same period of time. Although 1 may some day be able
to devote toits completionthe time which suchanundertaking deserves,
at prssent 1 have no plans f or thi s. Consequent ly , 1 would like to make
these notes available now, despite their deficiencies, andwith al1 the
compromises 1 learned to make in the early hours of the morning.
These notes were written while 1 was teaching a year course in dif-
ferentialgeometry at Brandeis University, during the academic year
1969-70. The course was taken by six juniors and seniors, and audited by
a few graduate students. Most of them were familiar with the material in
Calculus -on Manifolds, which is essentially regarded as a prerequisite.
More precisely, the complete prerequisites are advanced calculus using
linear algebra and a basic knowledge of metric spaces. An acquaintance
with topological spaces is even better, since it allows one to avoid the
technicaltroubles which are sometimes relegatedto the Problems, but 1
tried hard to make everything work without it.
The materialinthepresentvolume was covered inthe firstterm, except
for Chapter 10, which occupied the f irst couple of weeks of the second
term, and Chapter 11, which was not covered in class at all. We f ound it
necessary to t ake rest cures of nearly a week af ter completing Caapters 2,
3, and 7. The same material could easily be expanded to a fu11 year course
in manifold theory with a pace that few would describe as excessively leisurely . 1
in manifold theory with a pace that few would describe as excessively
leisurely . 1 am gratef u1 to the class f úr keeping up wi th my accelerated
pace, f or otherwise the second half of these notes would not have been
written. 1 am also extremely grateful to Bichard Palais, whose expert
knowledge saved me innumerable hours of labor.
TABLE OF CONTENTS Although the chapters are not divided into sections. the listing for each
TABLE OF
CONTENTS
Although the chapters are not divided into sections.
the listing for each chapter @vessome indication
which topics are treated. and on what pages.
CHAPTER 1. MANIFOLDS
Elementary properties of manifolds
Exarnples of manifolds
Problems
1
6
20
CHAPTER 2. DIFFERENTIAL STRUCTURES
Cm structures
Cm functions
Partial derivatives
Critica1 points
Immersion theorems
Partitions of unity
Problems
27
31
35
40
42
50
53
CHAPTER 3. THE TANGENT BUNDLE
The tangent space of IRn
Tlle tangent space of an imbedded manifold .
Vector bundles
The tangent bundle of a manifold
Equivalence classes of curves. and derivations
Vector fields
Orieiltation
Addendum . Equivalence of Tangent Bundles
Problems
xiu CHAPTER 4. TENSORS The dual bundle The difkrential of a function Classical versus modern
xiu
CHAPTER 4. TENSORS
The dual bundle
The difkrential of a function
Classical versus modern terminology
Multilinear functions
Covariant and contravariant tensors
Mixed tensors. and contraction
Problems
107
109
111
115
117
121
127
CHAPTER 5. VECTOR FIELDS AND DIFFERENTIAL EQUATIONS
Integral curves
Existence and uniqueness theorems
The local flow
One-parameter groups of difkomorphisms
Lie derivatives
Brackets
Addendum 1. Dserential Equations
Addendum 2. Parameter Curves in Two Dimensions
Problems
135
139
143
148
150
153
164
167
169
CHAPTER 6. INTEGRAL MANIFOLDS
Prologue; classical integrability theorems
Local Theory; Frobenius integrability theorem
Global Theory
Problems
179
190
194
198
CHAPTER 7. DIFFERENTIAL FORMS
Alternating functions
The wedge product
Forms
DSerential of a form
Frobenius integrability theorem (secondversion)
Closed and exact forms
Tlle Poincaré Lemma
Problems
201
203
207
210
215
218
225
227
CHAPTER 8. INTEGRATION Classical line and surface integrals Integrals over singular k-cubes The boundary of
CHAPTER 8. INTEGRATION
Classical line and surface integrals
Integrals over singular k-cubes
The boundary of a chain
Stokes' Theorem
Integrals over manifolds
Volume elements
Stokes' Theorem
de Rliarn cohomology
Problems
CHAPTER 9. RIEMANNIAN METRICS
Inner products
Riemannian metrics
LRngth of curves
The calculus of variations
The First Variation Formula and geodesics
The exponential map
Geodesic completeness
Addendum. Tubular Neighborhoods
Problems
301
308
312
316
323
334
341
344
348
CHAPTER 1O. LIE GROUPS
Lie groups
Lft invariant vector fields
Lie algebras
Subgroups and su balgebras
Homomorphisms
One-parame ter subgroups
The exponential map
Closed subgroups
LRft invariant forms
Bi-invariant metrics
The equations of structure
Probleins
371
374
376
379
380
382
384
391
394
400
402
406
xuz Conlents CHAPTER 11. EXCURSION IN THE REALM OF ALGEBRAIC TOPOLOGY Complexes and exact sequences
xuz
Conlents
CHAPTER 11. EXCURSION IN THE REALM OF ALGEBRAIC
TOPOLOGY
Complexes and exact sequences
The Mayer-Vietoris sequence
Triangulations
The Euler characteristic
Mayer-Vietoris sequence for compact supports
The exact sequence of a pair
Poincaré Duality
The Thom class
Index oí'a \lector field
Poincaré-Hopf Theorem
Problems
419
424
426
428
430
432
439
442
446
450
453
APPENDIX A
To Chapter I
Problems
To Chapter 2
Problems
To Chapter 6
To Chapters 7.9. 10
Problem
459
467
471
-472
473
474
-475
NOTATlON INDEX
477
INDEX
481
A Comprehemive Introduction to DIFFERENTIAL GEOMETRY VOLUME ONE
A
Comprehemive Introduction
to
DIFFERENTIAL GEOMETRY
VOLUME ONE
CHAPTER 1 MANIFOLDS he nices t example of a metric space is Euclidean n-space R",
CHAPTER 1
MANIFOLDS
he nices t example of a metric space is Euclidean n-space R", consisting of
T al1 n-tuples x =
,Y) with each x'
E IR, where R is the set of real
numbers. Whenever we speak of R" as a metric space, we shd assume that it
has the "usual metric"
unless anotlier metric is explicitly suggested. For n = O we will interpret R0 as
the single point O E R.
A manifold is supposed to be "locally" like one of these exemplary metric
spaces R". To be precise, a manifold is a metric space M with the following
property:
If x E M, then there is some neighborhood U of x and some integer
n 2 O such that U is homeomorphic to R".
The simplest example of a manifold is, of course, just Rn itself; for each
x E R" we can take U to be ali of R". Clearly, R" supplied with an equiva-
lent metric (one which makes it homeomorphic to R" with the usual metric),
is also a manifold. Indeed, a hasty recollection of the definition shows that
anything l-iomeomorphic to a manifold is also a manifold-the specific met-
ric with which M is endowed plays almost no role, and we shd alrnost never
mention it.
[If you know anything about topological spaces, you can replace "metric
space7'by "topological space" in our definition; this new definition allows some
pathological creatures which are not metrizable and which fad to have other
properties one might carelessly assume must be possessed by spaces which are
locally so nice. Appendix A contains remarks, supplementing various chapters,
which should be consulted if one allows a manifold to be non-metrizable.]
The second simplest example of a manifold is an open bali in R"; in this
case we can take U to be the entire open ball since an open ball in R" is
homeomorphic to R". This example immediately suggests the next: any open
subset V of R" is a manifold-for each x E V we can choose U
subset V of R" is a manifold-for
each x E V we can choose U to be some open
ball with x E U c V. Exercising a mathematician's penchant for generalization,
we immediately announce a proposition whose proof is left to the reader: An
open subset of a manifold is also a manifold (calIed, quite naturally, an open
submanifold of the original manifold).
The open subsets of R" already provide many dXerent examples of manifolds
Cjusi 11ow maiiy is the subject of Problem 24), though by no means all. Before
proceeding to examine other examples, which constitute most of this chapter,
some preliminary remarks need to be made.
If x is a point of a manifold M, and U is a neighborhood of x (U contains
some open set V with x E V) which is homeomorphic to RR by a homeomor-
phism #: U + R", then #(V) c R" is an open set containing #(x).
Conse-
quently, there is an open ball W with #(x) E W c $(V). Thus x E #-'(w) c
V c U. Since # : V + R" is continuous, the set #-'(W) is open in V, and thus
open in M; it is, of course, homeomorphic to W, and thus to R". This compli-
cated little argumentjust shows tliat we can always choose the neighborhood U
in our definition to be an open neighborhood.
Witli a little tliought, it begns to appear that, ii-i fact, U mtwf be open.
Witli a little tliought, it begns to appear that, ii-i fact, U mtwf be open. But
to prove this, we need the following theorem, staied I-iere without proof.*
1. THEOREM. If U c Rn is open and f : U + Rn is one-one and continu-
ous, then f (U) c IR" is open. (It foUows that f (V) is open for any open V c U,
so f-' is continuous, and f is a homeomorphism.)
Theorem 1is called "Invariance of Domain", for it implies that the property of
being a "domain" (a connected open set) is invariant under one-one continuous
maps into R". The poof that the neighborhood U in our definition must be
open is a simple deduction from Invariance of Domain, left to the reader as ail
easy exercise (it is also easy to see that if Theorem 1 were false, then there would
be an example where the U in our definition was not open).
We next turn our attention to the integer n appearing in our definition. Notice
that n may depend on tlie point x. For example, if M c IR3 is
ihen we can clioose n = 2 for points in MI and ri = 1 for points in M2. Tliis
example, by the way, is an unnecessarily complicated device for ~roducingone
manifold from two. In general, given Mi and Ms, with metrics di and d2, we
can firsi replace each di witli an equivalent metric di such that di(x, y) .E 1 for
al1 X, y E Mi; for example, we can define
* Al1 proofs require some amount of macliiner)' The quickest routes arc probably pro-
vided by Vick, Hont o1o.g TJleory and Massey, Singular Hont ology Theoy. An old-fashioned ,
but pleasanily geometric, treátment may be found in Newan, %pologp ofPlenr Se&.
Then we can define a metric d on M = Mi U M2 by di(x,y)
Then we can define a metric d on M = Mi U M2 by
di(x,y)
d(x,v) =
if therc is some i such that x, y E Mi
otherwise
(we assume that Mi and M2 are disjoint; if not, they can be replaced by new sets
which are). In tlie new space M, both Mi and M2 are open sets. If MI and M2
are manifolds, M is clearly a manifold also. This coiistruction can be applied
to any number of spaces-even uncountably many; the resulting metric space is
called the disjoint union of the metric spaces Mi. A disjoint union of manifolds
is a manifold. In particular, since a space with one point is a manifold, so is any
discrete space M, defined by the metric
Although ddfeient 11's may be required at ddferent points of a manifold M,
it would seem tliat only one n can work at a giveii point x E M. For the proof
of this ii-ituitivelyobvious assertion we llave recourse once again to Invariance
of Domain. As a first step, we note that R" is not liomeomorphic to Rm wlien
11 # 111, for if 11 > 111, t1-iei-iiliei-e is a one-oiie continuous map from Rm into
a iloii-open subset of R". The furtlier deduction, tliai the n of our definition
is uilique at each s E M, is lefi to the reader. This unique n is called the
dimension of M at x. A manifold has dimension 11 01- is n-dimensional or is an
a-manifold if it lias dimension i~ at each point. It is convenient to refer to tlie
inaiiifold M as M" wlieii we want to indicaie tliai M has dimensioii n.
Consider once more a discrete space, whicli is a O-dimensional maiiifold.
The oiily compact subsets of such a space are finite subsets. Consequently, an
ui-icountablediscrete space is not o-compact (it cannot be written as a countable
uiiioi~of com11act sul~sets).Tlie same plieiiomenon occurs with higher-dime~i-
sional manifolds, as we see by taking a disjoint uilion of uncountably many
manifolds liomeomorpl~icto R". In tliese examples, however, he manifold is
not coi~i-iected.Mie \vil1 often ileed to know that tliis is the only way in wliicl-i
o-compactness can fail to hold.
2. THEOREM. If X is a connecied, Iocally coinpaci metric space, tl-ien X is
o-compact.
PR001;. Foi- each 1- E X cconsider tl-iose numbers r > O such that the closed
ball
(y E X: d(x-,y) 5 i')
is a compact set (there is at least one such r > O, since X
is a compact set (there is at least one such r > O, since X is locally compact).
The set of al1 such I. > O is an interval. If, for soine A-, tliis set includes all r > 0,
then X is o-compact, since
If not, then for each x E X define r(x) to be one-half the least upper bound of
al1 su& r.
The triangle inequality implies that
so that
which implies that
Jnterchanging xi and x2 gives
so tlle function r : X + IR is continuous. This has tlle following important
consequence. Suppose A c X is compact. Let A' be the union of all closed
balls of radius r (y) aiid cei-iter y, for all 11E A. Then A' is also compact. The
poof is as follows.
Let
be a sequence in A'. For eacli i there is a yi E A such
that zi is in the ball of radius r(yi)with center yi. Since A is compact, some
subsequence of the yi, which we might as well assume is the sequence itself,
converges to some point y E A. Now the closed ball B of radius and
center y is compact. Since yi :i y and since the function r is continuous,
center y is compact.
Since yi :i y and since the function r is continuous,
eventually the closed bds
are contained in B. So the sequence ri is eventually in the compact set B, and
consequently some subsequence converges. Moreover, the limit point is actually
in the closed ball of radius r(y) and center y (Problem 10). Thus A' is compact.
Now let AO E X and consider the compact sets
Their union A is clearly open. It is also closed. To see this, suppose that x is
a point in the dosure of A. Tlicn there is sorne y E A with d(x, y) c ir(x).
This shows tliat jf y E A,,
tlien x E A,',
so x
E A.
Siiice X is coi-ii-iected,and A # 0 is operi a~idclosed, it must be tliat X = A,
wliich is o-compact. 6
After this liassle with point-se t topology? we present the long-promised exarn-
ples of manifolds. Tlie oiily connected i-manifolds are tlie line IR and the circle,
or I -dimensional sphere, S', defined by
The function f : (0,2ir) + S' defined by f (O) = (cos O, sin
The function f : (0,2ir) + S' defined by f (O) = (cos O, sin O) is a home-
omorphism; it is even continuous, though not one-one, on [0,2n].
We will
orten denote the ~oint(cos O, sin O) E S' simply by O E [O, 2x1. (Of course, it
is always necessary to check that use of this notation is valid.) The function
g : (-s, s) + S', defined by the same formula, is also a homeom~r~hism;
together with f it shows that S' is indeed a manifold.
There is another way to prove this, better suited to generalization. The pro-
jection P from the point (O, 1) onto the line IR x (-1) c R x R, illustrated in
the above diagram, is a homeomorphism of S' - ((0, 1)) onto R x (- 1): this is
poved most simply by calculating P : S' - ((O,])) + R x (- 1) explicitly. The
point (O, 1) may be taken care of similarly, by projecting onto R x (I), or it suf-
fices to note that S' is "homogeneous"-there is a homeomorphism taking any
point into any other (namely, an appropriate rotation of IR2). Considerations
similar to these now show that the n-sphere
is an n-maiiifold. The 2-spliere S2, commonly known as "the sphere", is our
first example of a compact 2-manifold or surface.
From these few manifolds we can already construct many others by noting
tl-iat if Miare manifolds of dimension ir j (i = 1,2), then MI x M2 is an (ni + n2)-
manifold. In particular
n times
is called the n-torus, while S' x S' is commonly called "the torus". 1t is ob-
viously liomeomorphic to a subset of IR4, and it is also liomeomorphic to a
certain subset of IR3 which is what most people have in mind when they speak of
"the torus": This subset may be obtained by revolving the circle {(O,y,a) E IR3 :
"the torus": This subset may be obtained by revolving the circle
{(O,y,a) E IR3 : (JJ - 1)2+ z2 = 1/41
around the z-axis. The same construction may be applied to any 1-manifold
coiitaiiied ili {(O,jp,r) E IR3 : 1):> O). Tlie resul tiiig surface, called a surface of
revolution, lias coinpoiients lioineomorpl~iceither to tlie toi-usor to the cyliiidei-
Si x R, tlie latter of which is also homeomorphic to the annulus, the region of
tlie plai-ie contail-iedbetweeii two concentric circles.
l'hc i-iestsimplest compact 2-maiiifold is tlie 2-lioled torus. To provide a more
explicit description of tlie 2-lioled torus, it is easiest to begin witli a "l-iandle", a
space liomeomorphic to a torus witli a hole cut out; more precisely, we tlirow
away d the points on one side of a certain circle, wbich remains in our
away d the points on one side of a certain circle, wbich remains in our llandle,
and which will be referred to as the boundary of the handle.
The 2-holed
torus may be obtaiiied by piecing two of tliese together; it is also described as
the disjoint uiiion of two handles witli corresponding points on the boundaries
"identified".
Tlie >I-holedtorus may be obtained by re~eatedapplications of this proce-
dure. It is homeomorpliic to iIie space obtained by starting with the disjoint
union of i7 liandlcs and a spliere with n holes, and then identif9ng points on
the boundary of the ithIiandle with corresponding points on the ithboundary
piece of the sphere.
There is one 2-manifold of which most budding mathematicians make the
acquaintance wheli tliey still know more aboui paper and paste than about
metric spaces-ihe famous Mobim S@, which you "make" by giving a strip of paper a
metric spaces-ihe famous Mobim S@, which you "make" by giving a strip
of paper a half twist before pasting its ends together.
This can be described
analytically as the image in R3 of the function f : [O,2x1 x (- 1,l) + R3 defined
by
e
e
f(B,i)=
(2cosB+icos~cosB,2sinB+icosIsinB,
If we define: f on [O, 2x1 x [- I,1] instead, we obtain the Mobius strip with
a boundary; as iiivestigation of the paper model will show, this boundary is
homeomorphic to a circle, not to two disjoint circles. With our recently intro-
duced terminology, tlie Mobius strip caii also be described as [O, I] x (- 1,l)
with (O, i) and ( 1, -1) "identified".
We have not yet liad to make precise this notion of ccidentification'y,but our
next example will force tlie issue. We xvisli to identify eacli point x E s2witli
its aiitipodal poiiit -x E S2.The space whicli results, the projecúve plane, IP2, is a
its aiitipodal poiiit -x E S2.The space whicli results, the projecúve plane, IP2,
is a lot harder io visualize than previous examples; indeed, there is no subset
of IR3 which represents it adequately.
The precise definition of IP2 uses the same trick that mathematicians always
use when they want two things wliich are not equal to be equal. The p0in.h-
of P2 are defilied to be the sets (p,-p) for p E s2.We wilI denote this set
by [p] E IP2, so tliat [-p] = [p]. We thus have a map f : S2 + IP2 @ven
by f (p) = [p], for whicl-i f (p) = f (q) implies p = fq. We will postpone
for a while the problem of defining the metric giving the distance between two
points [p] and [q],but we can easily say wliat the open sets will turn out to be
(and this is all you need to know iil order to check that P2 is a surface). A subset
U c p2 will be open if aiid only if f-' (U) c S2 is open. This just means that
the open sets of P2 are of tlie form /(V) where V c S2 is an open set with
the addiiional important property that if it contains p it also contains -p.
In exacily ihe same wa): we could have defined ihe poinis of the Mohius sirip
In exacily ihe same wa): we could have defined ihe poinis of the Mohius
sirip M io be
al1 points (s,f) E (O, 1) x (-1,I)
iogeiher wiih
allseis {(O,r),(l,-111,
denotedby[(O,f)Ior[(I,-111.
(S, f)
[(s,f)]
if s # O, 1
ifs = O or 1,
and U
c
M is open if and only if
f -' (U) c
[O, 11 x
(- 1,l) is open, so tliat
tl-ie open seis of M are of the form f (V) ~rliereV is open and coniains (S, -1)
wlienevcr it coniains (S, f) for s = O or 1.
To gei an idea of wliat p2looks Iike, we can make tliings easier for ounelves
by first ilirowiiig away all poiiiis of s2helow tlie (x, y)-plane, siiice iliey are
ideniified with poinis ahmre ihe (x, y)-plane anyway. This leaves the upper
I-icniisplier-e(including tlie bounding ci~-clc), wliicli is Iiomeoinor~~liic io ilie disc
aiid wc must idcntify eacli p E Si wiili -p E S'. Squaring tliiiigs oKa
bit, tliis is ilie same as identifying points on ilie sides of a square according
to tlie sclienie shown below (poinis on sides wiil-i ihe same lahel are ideniified
in sucli a way iliat ille heads of tlie arrows are ideniified wiih each otlier). Tlie
dotied liiies iii ihis piciui-e are he key io undersiaiiding IP2. If we disiori ilie
region beiween tl-iem a bit we see tliat the froni pari of B followed by
region beiween tl-iem a bit we see tliat the froni pari of B followed by tl~e
I~ackpart of A, ai the upper left, is to be ideniified wiih the carne thing ai the
10wer righi, in reverse direction; in other wods, we obtain a Mobius sirip with
a boundary (narnely, ihe doited line, which is a single circle). If this h4obius
strip is removed, \ve are lefi wiih two pieces \vI~ichcan be rearranged to form
something homeomorphic to a disc. The projectiw plane is thus oh<ained from
tl-ie disjoini union of a disc and a Mobius sirip with a boundary, by identifying
points on the boundary and poinis on the boundary of the disc, both of which
are circles. Thus io make a model of iP2 we jusi have to sew a circular piece of
cloth and a clotlr Mobius strip iogci1ie1-along ifieir cdges. Unforiunately a litile
experimentaiion will convince you ilrai ihis cannot be done (without having tlle
iwo pieces of doth pass through each other}.
The subsei of IR3 obiained as the union of ilie Mobius sirip and a disc, al-
ihough noi hoiiicoiiioipliic to iP2, can siill be described mathematically in terms
of IP2. There is clearly a coiitinuous funciion /: iP2 + IR' wliose image is this
subsei; moreovei; althougli f is not one-one, ii is locally one-one, that is, every
point p E IP2 has a neigliborhood U on whicli f is one-one. Sucli a function f
is called a topological immersion (the single word "immersion" has a more ve- cialized meaiiiiig,
is called a topological immersion (the single word "immersion" has a more ve-
cialized meaiiiiig, explained in Chapter 2). IVe can thus say thai P2 can be
topologicaIIy immersed in W', al though noi topologically imbedded (ihere is no
liomeomoi-phism f fiom PZLO a subset of W3), In W4, liowever, wiih an extra
dimension to play around ivith, the disc can be added so as noi io interseci the
Mobius strip.
Another topological immersioii of P2in W3 can be obtained by first immers-
ing ihe Mobius sirip so iflat its boundary cirde lies iri a plane; ibis can be done
in the follo~vingway. The figures below show thai the Mobius sirip may be ob-
tairled from an annulus by identifying opposite points of the inner circle. (This
is also obvious from ihe fact thai ihe A4obius sirip is the projective plane wiih a
disc removed.} TIlis inner circle can l~ereplaced by a quadrilateral. When ihe
resuliing figure is di.a\vn up irito 3-space and the appropriatc ideniifications are
niade we ol~tairiibe "cross-cap". The cross-cap iogether wiih the disc ai tlie
bo~tomis a iopologically immersed P2.
'llic onc gap in ihe precedii-ig discussion is ilre definiiion of a metric for P2.
'llic onc gap in ihe precedii-ig discussion is ilre definiiion of a metric for P2.
'l'lir niissiiig nleir-ic can he supplied by an appeal to Problem 3-1, whicl~will
I;itri. l~eused quite often, and which the reader should peruse sometime hefoi-e
1-i.;icliiigCliapter 3. Rouglily speaking, ii sliows ihai iliiiigs llke PZ,whidi ouglir
ti) be inanifolds, are. (Those who know ahout iopological spaces will recognize
ii 21s a disguised case of ille Urysohn Metrizaiion Theorem.) For the preseiii,
Iio~*rver,we will obtain our metric by a irick thai simultaneously provides an
iiiibidding of P2 in JR4. Coiisider ihe fuiiciion f : s2+ JR4 defined by
f(x, y, 5) = (yz, xz, xy, x2 +2y2 + 32).
( :lr.asly f(p) = f (-p).
We mainiain ihai f (p) = f(q) implies ihat p = fy.
'li) p~-ovetl~is~suppose ihai f (xi y,~)= f (ai b, c). We have, firsi of ail
xy = ab.
I 1' a, b , c # 0, ihis leads to
r4
Now
lience
(3)
Using (2), ihis gives
so s = fa. Similarly, Mre obiain y = fb, z = fb, wiih ihe
so s = fa. Similarly, Mre obiain y = fb, z = fb, wiih ihe same sign (which
comes from (3))holding for dl three equations. In this case we have proved our
conteniion ~riihouteven using the fourth coordinate of f. Now suppose a = 0.
If x # 0; then (1) would immediately give y = z = O, so that
(x, y, z) = (fl,O, O).
But y = ; = O implies (by (1) again) that bc = O, so b = O or c = O and
These equations clearly contradict
Thus x = O aiso, and Mre have
But (6) implies ihat
2).
2 + 3z2 = zY2+ 3(1 - y2)
2
=3-y,
and similarly for b and c, so (5) gives
Now (4)$ves
(tlris holds eveil if y = b = 0, since then 1,c = f 1). Clearly, (4) also shows thai
thc same sign holds in (7) arid (8).which completes the proof.
Sinre f(p) = f(q) precisely wlieii p = fq, we can define j: P2 + R4 b!,
This nlap is ont-oi~eand wc can LISC it to define the meti-ic in IP2:
Then one can check that the open sets are indeed the ones described above. By
Then one can check that the open sets are indeed the ones described above.
By ihe way, the map g : P' + W3defined by tlie firsi 3 componenis of j;
is a topologicai immersion of P'
surface".
in W 3. The image in W3 is Steiner's "Romaii
Wiih the new surface P2 at our disposal, we can create other surfaces in
t he same way as the n-holed torus. For example, to a handle we can attacli
a projective space 14th a hole cut out, or, what amounts to the same thing, a
h4obius strip. The closesi we can come io picturing ihis is by drawing a cross-
cap sticking on a iorus. We can dso join together a pair of projective planes
with holes cut out, whicll amounts to sewing two Mobius strips together along
tlieir boundary. Although this can be pictured as two cross-caps joined together,
it has a nicer, and famous, representation. Consider the surface obtained from
ilre square with identifications indicated below; it may aiso be obtained from
tlie cylinder [O, l] x S' by identifying (O, x) E [O, l] x S' with (1, x'), where x'
is ihe reflection of x through a fixed diameter of ihe circle. Notice that ihe
identifications on the square force PI, P2,P3, and fi to be identified, so that the
identifications on the square force PI, P2,P3, and fi to be identified, so that
the set {Pl, P2, P3,P4) is a single point of our new space. The doited Iines
beloy dividing the sides into thirds, form a single circle, which scparates the
surface into two parts, one of which is shaded.
Rearrangemeni of ihe iwo parts shows thai ihis surface is precisely iwo
Mobius strips witb corresponding points o11 iheir boundary ideniified. The
description iti terms of [O, 11 x S' immediaiely suggests an immersion of the
surface. Turriing one ei~dof the cylinder around and pushing ii through itself
orients ihe left-hand boundary so that (O, x) is direcily opposite (1, x') , io which
it can then be joined, forming the "Klein bottle".
1;xamples of higher-dimensional manifolds ~~illnoi be treated in nearly such ifvtail, I~ui!in addition to the
1;xamples of higher-dimensional manifolds ~~illnoi be treated in nearly such
ifvtail, I~ui!in addition to the famiiy of 11-manifolds S", we wiIl mention ihe
i.i.lated fa mil^ of "projeciive spaces". Projective 11-space P" is defined as ihe
i.olleciion of al1 sets (p, -p) for p E S". The description of the open sets in P"
is ~x.eciselyanalogous to tlie descripiion for P2. Alihough these spaces seem to
li)i-ina family as regular as the family S", we ~cillsee Iater that the spaces P"
liii. cven n dXer in a very important way from the same spaces for odd n.
One further definition is needed io complete this introduction to manifolds.
\%. have dready discussed some spaces which are noi manifolds only because
tliey have a "boundarf"' for example, the h4obius strip and the disc. Points
riii these "boundaries" do not have neighborhoods homeomorphic to IRn, but
tlrey do llave neighhorhoods homeomorphic io an important subset of IR". Tlie
(closed) hdf-space W" is defined by
W" = {(xt,
,
x")
E IRn : x"
2 O}.
!'I manifold-with-boundary is a metric space M with the following property:
If x E M, ihen there is some neighborhood U of x and some integer
rr > O such ihai U is homeomorphic to either R" or W".
A poini in a manifold-with-boundary canriot have a neighborhood liomeo-
mor-phic to both IR" and W" (Invariance of Domain again); we can therefore
distirlguish ihose poinis x E M having a neighborhood homeomorphic to Hn.
.]'he sei of al1 such x is cdled ihe boundary of M arid is denoted by aM. If M is
;ictually a manifold, tlien 8M = 0. Notice that if M is a subset of IR", then aM
is noi necessarily the same as the boundary of M in the old sense (defined for
iiny subset of IRn); indeed, if M is a manifold-wiih-boundary of dimension < n,
tlien ali poinis of M wiil be boundary points of M.
If manifolds-~lith-boundaryare studied as frequently as manifolds, it becomes
I~oibersomeio use tl-iis long designaiion. Ofien, ihe word "manifold" is used
ior "manifold-~~ith-11oundary"-A manifold in our sense is then cdled "non-
I~ounded";a non-l~oundedcompact mailifold is cailed a "closed manifold". We
\vil1 stick to the other termiiiology, hui ~lillsometimes use "bounded manifold"
insiead of "manifold-with-boundary".
PRQBLEMS 1. Slio\v that if d is a metric on X, then both d =
PRQBLEMS
1. Slio\v that if
d
is
a metric on
X,
then both d
=
d/(l
+ d)
and d
=
min(1,d) are dso metrics and ihai tliey are equivdent to d (Le., the identiv
map 1 : (X, d) + (X, d) is a homeomorphism).
2. Jf (Xi, di) are metric spaces, for i E 1, with meirics di < 1, and Xi n Xj = 0
for i # j, tl~en(X, d) is a meiric space, where X = UiXj, and d(x, y) =
di(x, 1.) if A+, y E Xi for some i, while d(x, y) = 1 otherwise. Each Xi is an
UiYi
open subsei of
X, and Y is homeomorphic to X if and only if
Y
=
where the Yi are disjoint open sets and Yi is homeomorphic to Xi for each i.
Tl~espace (X, d) (or any space homeomorphic io it) is cdled the disjoint union
of ihe spaces Xi.
3. (a) Exre177 manifold is locally compaci.
(b) Every manifold is locdly pathwise connecied, and a connected manifold is
pathwise connected.
(c) A connecied manifold is arcwise connected. (A path is a continuous image
of [O, 11: but an arc is a une-ui~econtinuous image. A difficult theorem states
that every path contains an arc between iis end points, but a direct proof of
arcwise-conneciedness can be @ven for manifol ds.)
4. A space X is cdled locdl)~connected if for eacll x E X ii is ihe case that
every neighborhood of x contains a connected neighborhood.
(a) Conncctedness does iiot imply local conneciedness.
01) An open subset of a locally coi~nectedspace is locally connected.
(c) X is locally connected if and only if components of open sets are open, so
evcry iieighboi-liood of a poini in a Iocally connected space coniains an open
connected neighborhood.
(d) A locally connecied space is homcomorphic to the disjoint union of its com-
ponents.
(e) Every manifold is locally connected, and consequently homeornorpliic io
the disjoint unioii of its componeiits, wliich are open submanifolds.
5. (a) Tlie neigliborhood U in oui. definition of a manifold is always open.
(b) The integer n in our definition is unique for each x.
6. (a) A subset of aii n-manifold is an n-manifold if and only if it is open.
(h) lf M is connected, ihen ihe dimension of M at x is the same for al1 x E M.
7. (a) lf U c IR is an intenlal and f: U + IR is contiriuous and one-one,
ihen .f is eiiher increasing or decreasing.
(11) 'l'lie image f(U) is open. 'l'he inap f is a homeomorpl~ism. (f.) 8. Ikr
(11) 'l'lie image f(U) is open.
'l'he inap f is a homeomorpl~ism.
(f.)
8. Ikr t his problem, assume
(1) (Tlie Generalized Jordan Cunle Theorem) If A c Rn is homeomorphic
to S"-',
then IR" - A has 2 components, and A is the houndary of eacli.
(2) If
B
c
IR"
is homeomorphic to D" = (x E R"
: d(x,O) 5
l), then
R" - B is connecied.
(a) One componeni of R" - A (the "outside of A") is unhounded, and the other
(ihc "inside of A") is bounded.
(h) If U c W" is open, A c U is homeomorphic to SR-' and f: U +
Ri" is oi~e-oneaiid coniinuous (so ihai f is a honieomorphism on A), thei~
-/'(iiiside of A) = inside of f (A). (First prove c .)
(f.) Pi-ove Invariance of Domain.
9. (a) Give an elenientary proof tliat IRi is not liomeomorphic to R" for n > 1.
(h) l'rove directly from the Generalized Jordan Cunre Theorem thai IRm is no1
Iiorneomorphic to IR" for nr # n.
10. In ihe proof of TI~eoirin2, sliow that the limii of a convergent subsequence
oi'the zi is aciudly iii the closed ball of radius r(y) and cenier y.
11. Every connecied maiiifold (which is a metr-ic space) has a couniahle 11asc
li~i-its iopology, and a countahle dense subset.
P
12. (a) Compute tlie composiiion f
=
S'
- ((0, 1)) +
Ri
x
(-
1) +
Ri
i:spIiciily for ihe map P on page 7, and show that it is a homeomorphism.
(h) Do the same for f: S"-' - ((O,.
,O,
1)) + IRn-'.
13. (a) Tlie text describes tlie open subsets of iP2 as seis of ihe form f(V);
\\*liei-eV c S' is open aiid coiitaiiis - p wlienever it contains p. Show tliat this
last condition is aciually unnecessary.
(II) The andogous condiUon Zr necessary for the Mobius sirip, which is discussed
iinmediately afterwards. Explain how ihe two cases differ.
14. (a) Check tliat tlie inetric defined for IP2 gises tlie open sets described in
ihe texi.
(II) Check that P* is a surface.
15, (a) Sho~lthat P' is homeomorphic to S'. (b) Since we can consider S"-' c
15, (a) Sho~lthat P' is homeomorphic to S'.
(b) Since we can consider S"-' c S", and since aniipodal points in S"-' are
siill antipodal when considered as points in S", we can consider P"-' C P" in
an obvious way. Show that P" - iPn-' is Iiomeomorphic to interior D" = (x E
R" : d(x,O) -= 1).
16. A classical tlieorem of topology states that every compact surface otl~er
~liaiiS2 is obiained by gluiiig iogether a certain number of ton and projective
spaces, and that all compact surfaces-~4th-boundaryare ohtained from these
by cutting out a finite number of discs. To whích of these "standard" surfaces
are the following homeomorphic?
a hole in
a hole in
a hole
17. Lei C c R c IR2 be the Cantor sei. Show thai W2 - C is homeomorpliir
to tlie surface shown at the top of the next page.
illis uii.clc is tiol in tlic sur fa c.^ these circles are 18. A locdly
illis uii.clc is tiol in tlic sur fa c.^
these circles are
18. A locdly compact (but non-compact) space X "has one end" if for every
voinpact C c X there is a compact K such that C c K c X and X - K is
c-onnected.
(a) R" Iias one end if n >
1, but not if n = 1.
(13) Rn - (O) does not "have one end" so Rn - (O) is not homeomorphic to Rm.
l 9. This proaem is a seque1 to the previous one; it will be used in Problem 24.
Aii end of X is a function E which assigns io each compaci subset C c X a
iion-empty component E(C) of X - C, in sudi a way that Cl c C2 hplies
~(CZ)C E(C1).
(a) If C c R is compact, ihen R - C has exactly 2 unbounded components, the
"left" component coiltaining dl iiumbers < some N, the "right" one containing
al1 ilumbers > some N. If E is an end of R, show thai E(C) is either dways tlie
"lefi" componeni of IR - C, or dways the "rhiglli" one. Thus R has 2 ends.
(11) Show that R" has only one end E for n > 1. More generdly, X has exactly
oiie end E if and o~llyif X "has one end" in tlie sense of Problem 18.
(c) This part requires some knowledge of topological spaces. Let E(X ) be the
set of d ends of a coi-ii-iected, locally connected, locdly compact Hausdofl
sl~aceX. Define a topology on X U E(X) by choosing as neighborhoods Nc (EO)
of an end EO die sets
for- dl compact C. Sl-iow ihat X U G(X) is a compact HausdorK space. Whai
is R U E(R), and R" U E(Rn) for n > l?
20. Coilsider ihe folloiving tliree surfaces.
-
(A) The infiniie-lioled torus:
(B) The doubly infiniie-holed torus: (C)The infinite jail cell window: _kll~:~~lli~- J\t- Ji!+ - -
(B) The doubly infiniie-holed torus:
(C)The infinite jail cell window:
_kll~:~~lli~-
J\t-
Ji!+
-
-
-
-
-
-
-
e**
e*.
(a) Surfaces (A) and (C) liavc oiie end, wllile surface (B) does not.
(b) Surfaces (A) and (C) are homeomorphic! Hi?zt: Tl-ie region cut out by ihe
lines in tlie piciure below is a cylinder, wliicl-i occurs ai tlle left of (A). Now draw
in two more liiies enclosirlg more holes, aild consider ihe region between the
iwo pairs.
21. (a) The three open subsets of RZ shown below are Iiomeomorphic. (b) The points
21. (a) The three open subsets of RZ shown below are Iiomeomorphic.
(b) The points kside the three surfaces of Problem 20 are homeomorphic.
22, (a) Every open subset of ES is homeomorphic to the disjoint union ofinter-
vals.
(b) Ther-e are only countably many non-homeomorphic open subsets of R.
23. For the purposes of this problem we will use a consequence of the Urysohn
Metrization Theorem, that for any connected manifold M, there is a homeo-
morphism f from M to a subset of the countable product R x R x - .
(a) If M is a connected non-compact manifold, then there is a continuous
function f:M + R such that f "goes to ca at m", i-e., if (x,)is a sequence
which is eventually in the complement of every compact set, then f (x,) + m.
(Compare with Problem 2-30.)
(b) Given a homeomorphism f : M + R x R x - - - and a g : M + R which
goes to ca at m, define f: M + R x (R x R x - --)by f(x) = (g(x),f (x)).
Show that f(~)is closed.
(c) Thei-eare ai most c non-liomeomorphic coniiected manifolds (where c = ZN"
is the cardinality of ES).
24. (a) It is posible for IRZ- A and IR2 - B to be homeomorphic even though A
and B are non-liomeomorphic closed subsets.
(b) If A c IRz is closed arid totally disconnected (the only components of A are
points), then E(RZ - A) is liomeomorphic to A. Hence IR2 - A and IRZ - B are
non- homeomorph ic if A and B are non-homeomorphic closed to tally discon-
nected sets.
(c) The derived set A' of A is the set of al1 non-isolated points. We define A(")
inductively by A(') = A' and A(~+')= (A("))'.For each n there is a subset A,
of R such that A,(") consists of one poini.
*(d) There are c non-homeomorphic closed totally disconnected subsets of IRZ. Hint: Let C be
*(d) There are c non-homeomorphic closed totally disconnected subsets of IRZ.
Hint: Let C be the Cantor set, and ci C: cz C: c3 C: . -. a sequence of points
in C. For each sequence nl r m r - - - , one can add a set A,, such that its
n jth derived set is (ci).
(e) Tliere are c non-homeomorphic connected open subsets of IRZ.
25. (a) A manifold-with-boundary could be defined as a metric space M with
the property tliat for each x E M there is a neighborhood U of x and an
inieger n 2 O such thai U is homeomorphic to an open subset of W".
(h) If M is a manifold-with-boundary, then aM is a closed subset of M and
aM and M - aM are manifolds.
(c) If Cj, i E 2 are the components of aM, and Ir c I, then M - UiEl,Ci is
a manifold-with-boundary.
26. If M c Rn is a closed sei and an 11-dimensional manifold-with-boundary,
ilien the iopological boundary of M, as a subsei of IR", is aM. This is not
iiecessarily true if M is not a closed subset.
27. (a) Every poini (a, b, c) oii Steiner's surface satisfies bZc2+ ozcZ+ 02b2=
abc.
@) If (a, b,c) saiisfies tliis equation and O # D = JbZcZ + oZcZ + oZbZ,then
(a, b, c) is on Sieiner's suríiace. Hint: LRt x = bclD, etc.
(c) The sei {(a, b, e)
E IR3 : bZcZ+ ozcz + azbz = abc) is the union of tlie
Steiner surface and of the portions (-m, - 112) and (1 12, m) of each axis.
CHAPTER 2 DIFFERENTIABLE STRUCTURES e are now ready to apply analysis to the study of
CHAPTER 2
DIFFERENTIABLE STRUCTURES
e are now ready to apply analysis to the study of manifolds. The neces-
sary tools of "adva~lcedcalculus", which the reader should bring along
fieshly sharpened, are contained in Chapters 2 and 3 of Cizku1u.r on A4an8Ms.
We will use freely the notation and results of these chapters, including some
11roblems, notably 2-9, 2-15, 2-25, 2-26, 2-29, 3-32, and 3-35; however, we wül
denote the identity map from W" to Wn by 1, rather than by n (which will be
used oRen enough iii other contexts), so that li(x) = xi.
Ori a general manifold M the notion of a continuous function f : M + W
inakes sense, but tlie notiori of a differentiable function f:M + W does not.
This is the case despite the fact that M is locally like IR", where differentia-
bility of funclions can be defined. If U c M is ari open set and we choose
a homeomorpliism #: U + IRn, it would seem reasonable to define f to be
differentiable on U if f o #-' : R" + W is differeiitiable. Unfortunately, if
q: V + W" is anotl~erhomeomorphism, and U n V # 0, then it is not
necessarily true that f o SI-': W" + W is also dfirentiable. Indeed, since
we can expect f o VI-' to be differentiable for al1 f wliich make f o#-' diñer-
eiitiable only if #o q-' : Wn + W" is differeiitiable. This is certainly not always
tlie case; for example, one need merely choose # to be h o@,where h :
tlie case; for example, one need merely choose # to be h o@,where h : R" + R"
is a liomeomorphism that is not differentiable.
If we insist on defining differentiable functions on any manifold, there is no
way out of this impasse. It is necessary to adorn our manifolds with a little
additional structure, the precise nature of which is suggested by the previous
discussion.
Among all possible homeomorphisms from U c M onto Rn,we wish to select
a certain collection witli tlie properry that #o@-'
is differentiable whenever #, @
are in the collection. This is precisely whai we shall do, but a few refinements
will be introduced along the way.
First of all, we Mil1 be interested almost exclusively in functions f : R" + R"
wliich are Cm (that is, each component function fi possesses continuous partial
derivatives of all orders); sometimes we will use the words "dflerentiable" or
c'smoothy'to mean Cm.
Moreover, insiead of considering homeom~r~hismsfrom open subsets U
of M orlto R" , it will sufice to consider Iiomeomorphisms x : U + x(U) c R"
onto open subsets of R".
The use of tlie letters x, 11,etc., for tliese liomeoniorphisms, henceforth ad-
Iiered to almost religiously is meant to encourage the casual confusion of a point
p E M with x(p) E Rn: wliich has "coordinates" x' (p), .,xn(p). The only
time tliis notation will be confusing (and it will be) is when we are referring to
ihe maiiifold R", whcre ii is hard not to lapse back into the practice of denoting
points by x and y. IVe will often mention tlie pair (x, U), instead of x aloile,
just to provide a coii\rei~ieniname for the domain of x.
If U and V are opeii siibsets of M, two homeomorphisms x : U + x(U) c
R" arid 11:V + jc(V) c R" are called Cm-related if the maps
are Cm. This make seiise, since x(U nV) and y(U nV)are open subsets of R".
Also, it makes sense, aild is automatically true, if U n V = 0.
A family of mutually Cm-related homeomorpliisms whose domains cwer M
is called an atlas for M. A particular member (x, U) of an atlas 36 is called
a chart (for tlie atlas 36):. or a coordinate system on U, for the obvious reason
iliai it provides a way of assigiiing "coordinaies" to points on U, namely, tlic
coordinates x'
,xn(p) to the point p E U.
I4Je can even imagine a mesh of coordinate lines on U, by considerirlg the
l)2jintz/iable Structure~ iiiverse images uiider x of lines in Rn parallel to one of the
l)2jintz/iable Structure~
iiiverse images uiider x of lines in Rn parallel to one of the axes.
m
L
-
s
-
,
,
d
, .
,-,
.-
-
The simplest example of a manifold together ~rithan atlas consists of Rn with
ari atlas A of only one map, the identity 1 : R" + Rn. We can easily make the
atlas bigger; if U a~idV are llomeomorphic operi subsets of R", we can adjoin
any homeomorphism s:U + V with the property that x and x-l. are Cm.
Indeed, we can adjoiii as many such x's as we like-it is easy to check that
they are al1 Cm-relaied io each other. Tlie advantage of this bigger atlas U
is ihai the single word "cliart", wlieri applied to this atlas, denotes something
which must be described in cumbersome lariguage if one can refer only to 36.
Aside from this, U differ-sonly superficially from 36; one can easily construct U
from 36 (and one would be foolish not to do so once and for all). What has just
been said for ihe atlas (1)applies to any atlas:
1. LEMMA.
If 36 is an atlas of Cm-related cliarts on M, then 36 is contained
in a unique maximal atlas 36' for M.
PR001;. Let A' be il-ic set of al1 cliarts y wliicll are Cm-related to al1 charts
x E 36. It is easy to check that al1 charts in 36' are Cm-related, so 36' is an atlas,
and it is clearly tlie uriique maximal atlas coiitaining 36. *:*
Mle now dcfine a Cm manifold (ordifferentiablemanifold, or smoothmanifold)
to be a pair (M, A): wliere A is a niam'rna/ atlas for M. Thus, about the simplest
example of a Cm manifold is (R", U), wliere U (the "usual Cm-structure for
Rn") is tlie maxi~naladas containing (1).Arioiher example is (R, V) where V
contains the l~~rneornor~liismx H x3, wliosc inverse is no/ Cm, together with
al1 charts Cm-relaied io it. Although (R, U) and (R, V) are not tlie same, tliere
is a one-one onto function f : R + R such thai
E U
ifand only if
x o f E V,
A-
namely, the obvious map f(x) = x Thus (R, U) aiid (8, V) are the sort
of structures one would want to cal1 "isomorphic".
Tlie term actually used is
"diffe~mor~hic":two C* manifolds (M, A) and (N, a) are diffeomorphic if there is a one-one
"diffe~mor~hic":two C*
manifolds (M, A) and (N, a) are diffeomorphic if
there is a one-one onto function f:M + N such that
x E J? if and only if
A- o f E A.
Tl-ie map f is called a diffeomorphism, and f-' is clearly a difeomorphism
also. If we had not required our atlases to be maximal, the definition of diffeo-
mor.pliism would have had to be more complicated.
Normally, of course, we will suppress mention of tlie atlas for a dfireritiable
mariifold, and speak elliptically of "the difirentiable manifold M"; the atlas
for M is sometimes referred io as the dzjierentiabk structure for M. Ii wdl always
be undersiood that IR" refers to the pair (IR", U).
Ii is easy to see that a difkomor-phism must be coritinuous.
Consequently,
its inverse musi also be coniinuous, so ihat a difeomorpliism is automatically
a homeomorpl~ism.This r-aises the natural question whetlier, conversely, two
liomeomorpliic maiiifolds are necessarily diffeomor-phic. Later (Pi.oblem 9-24}
we will be able to prove easily that IR with any ailas is difkomorphic to (R, U).
A proof of tlie corresporiding assertion for is much harder, the proof for IR3
would ccrtainly be ioo difficult for-iriclusion llere, aiid the proof of the essential
uniqueness of C* structures on IR" for n 2 5 requires very difficuli techniques
from topology,
In ihe case of spl-ieres,tlie projections Pl and P2 from the points
,O, 1)
and (0,
,O, - 1) of S"-' are easily seen to be Cm-related. They iherefore
clctei-minean ailz5-tl-ie "usual C* structure for S"-'". T1-iis atlas may also bc
described in terms of the 2n homeomorpliisms
defined by J(s)= gi(x) =
,si-',xi+',
,xn),wbicli are Cm-related
to P1and P2.Tlrei-eare, up io diffeoniorpl-iism,unique differentiable struciures
on S" for 17 5 6. But tl-iere are 28 difeomorpl-iism classes of differeniiable
structures on S7. aiid over- 16 million on S3'. However, we shall not come
close to proving iliesc asser-tioi~s,whicli are pari of tlie field called "differential
topology", rather then diferential geometry. (hrhaps most astonishing of al1
is ilie quite receiit discovery that IR4 lias a difeirritiable structure that is not
cliffromorphic io tlie usual diffei-entiablesiructure!)
Otl-ier examples of differentiable manifolds will be @ven soon, but we cari
already describe a diffcrentiable structure A' 011 any open submarlifold N of
a differentiable manifold (M, A); tlie atlas A' consists of all (x, U) in 36
a differentiable manifold (M, A); tlie atlas A' consists of all (x, U) in 36 witli
U c N.
Just as dfiomor-phisms are analogues for C* manifolds of homeomor-
phisms, there are analogues of continuous maps. A function f: M + N is
called differentiable if for every coordinate system (x, U) for M and (y, V)
for N, the map y o f o x-' : IRn + IRm is differentiable. More particularly, f
is called differentiable at p
E
M
if
y o f
o x-'
is differentiable at x(p) for
coordinate systems (x, U) and (y, V) with p
E
U and f (p)
E
V.
If this is
true for one pair of coordinate systems, it is easily seen to be true for any otlier
pair. We can thus define differentiability of f on any open subset M'
C M;
as one would suspect, tliis coincides with differentiabilit~of the restricted map
f 1M': M' + N. Clearly, a differentiable map is continuous.
A differeniiable function f : M + IR refers, of course, to the usual dfiren-
tiable structure oii IR, and lience .f is differentiable if and only if f o x-'
is
diflerentiable foi- cach chart x. It is easy to see that
(1)
a function f : IR" + IR"
is diflerentiable as a map between C*
manifolds
if and only if it is difler-entiable in tlie usual sense;
(2) a fuiiction f : M + IRm is differentiable ifand only if each fi : M + Wm
is diferen tiable;
(3) a coordinate system (x, U) is a diffeomorphism from U to x(U);
(4) a fuiiction f : M
+
N is diffei-eiitiable if and only if each y'
o f
is
difirentiable for each coordinate system y of N;
(5) a difler-eritiable function f : M + N is a dfleomorphism if and only if
f is one-o~ieonto and fU1
: N + M is differentiable.
Tlie differe~itiablesiructui-eson inaiiy manifolds are designed to make ccrtain
fuilciions differ-eniiable. Consider first the product MI x M2 of two dfleren-
iial~lemanifolds Mi, arid the two "projections" ni : MI x M2 + Mi defined by
iial~lemanifolds Mi, arid the two "projections" ni : MI x M2 + Mi defined
by ni(pl, p2) == pi. It is easy to define a differentiable structure on MI x M2
which makes eac1-ini difirentiable. For each pair (xi, Ui) of coordinate systems
on Mi, we constmct the homeomorphism
defined by
Tlien we cxiend tliis atlas to a maximal one.
Similarly, ihere is a difirentiable structure on Pn which makes the map
.f: Sn + Pn (defiried by f(p) = [p] = (p,-p)) difirentiable. Consider
any coordinate system (x, U) for S",where U does not contain -p if it con-
tair~sp, so tllat f 1U is o~i,e-one.Tlie map x o (f 1U)-" is a homeomorp1~ism
on J'(U) c Pn,and any two such are Cm-related. The collectiori, of tliese
lioineoi~iorpliismscan then be extended to a maximal atlas.
To obtain differentiable structures on other surfaces, we first note that a Cm
manifold-witli-bour~darycan be defined in an obvious way. It is only necessary
to know wlien a map f : Wn+ Rn is to be considered differentiable; we cal1 f
differentiable den it can be extended to a dflerentiable function on an open
i-ieiglibor-hoodof W".A "Iiandle" is then a Cm manifold-with-boundary.
A diffei.eriiiablestructure on tlie 2-holed torus can be obtained by "matcliiiig"
tRe dflereiitiable structure on two l-iandles. Tl-ie details involved in tliis process
are reserved for Pro blem 14.
To deal witl-i Cm fuilctions effectively,one needs to know that there are lots of
ilicm. Tlie esistcrice of COQfurictions oii a manifold depends oii tlie existence of
Cm funciions on R" ~Iiicliare O ouiside of a compact set. We briefly recall liere
ilie necessat-y facis aboui such Cm fuiictions (c.f. Cakulus on MaliifoMs, pg. 29).
D@ben/iabL Slructures (1) The function h: R + R defined by is Cm, and h(")(0)
D@ben/iabL Slructures
(1) The function h: R + R defined by
is Cm, and h(")(0) = O for al1 n.
(2) The function j: IR + R defined by
= {
e-(x-r)-2
e-(x+~)-2
x E (-1,
1)
o
x
(-41)
-1
I
I
is Cm.
Similarly, there is a Cm function k : W + R which is positive on (0,S) and O
elsewhere.
I
S
(3)The fuiiction I : R + R defined by
is Cm; ii is O for x 5 O, increasing on (O, S), and I for x 2 S.
(4)Tlie function g: R" + R defined by
is Cm; ii is positive on (-E, E) x . - - x (-E, E) and O elsewliere.
On a Cm manifold M we can now produce many non-constant Cm func-
tions. The closure (x : f(m)
# O) is called tlie support of f,and denoted simply
by support f (or sometimes supp f ).
2. LEMMA. Let C c U c M witli C conipact aild U open. Then there
is a Cm function f: M +
[O, 11 such tliat
f =
1 on C
and support f c
U.
(Compare he 2 of the proof of Theorem 15,)
PROOF. For each p E C, choose a coordinate system (x, V) with c U
PROOF. For each p E C, choose a coordinate system (x, V) with c U and
x(p) = O. Then x(V) 3 (-E,&) x --- x (-E,&) for some E 3 O. The function
g o x (wliere g is defined in (4))is CDgon V. Clearly it remains Cm if we extend
it to be O outside of V. IRt Sp be the extended function. The function fp can be
coiisir~uctedfor each p, aiid is positive on a neighborhood of p whose closure is
coniained iii U. Since C is compact, finitely many such neighborhoods cover C,
aiid the sum, fp, + - - + -fp,, of the correcponding functions has support c U.
011 C it is positive, so on C it is 2 S for some 8 3 O. Let f = Io ( fp, + - . -+fpt,),
where I is defined in (3).e:*
By tlie way, we could have defined Cr rnanifolds for each 1. 1, not just for
6 c i- = m". (A function f : IRn + IR is Cr if it Iias continuous partid derivatives
up to order i-). A "CO function" is just a continuous function, so a CO manifold is
jusi a maiiifold in tlie sense of Chapter l. We caii also define analytic manifolds
(a function f : Wn + W is analytic at u E IRn if f can be expressed as a
power series in the (xi - íii) which converges in some neighborhood of U). Tlie
symbol CW staiids for analytic, and it is conveiiient to agree that r .: w c: o
for each iiiteger i. > O. If a < B, tlien tlie cliarts of a maximal CB atlas are al1
Ca-related, bui ihis atlas can always be extended to a btggu atlas of Ca-related
cliaris, as in Lemma 1. Tlius, a CB structure on M can always be extended
io a C" structure ii~a uiiique way; tlie smaller siructui-e is tlie "stronger" one,
tlie CO structure (coiisistiiig of al1 homeomoi~pliismsx: U + W") beiiig tlie
lüigest. The coiiverse of this trivial remark is a hard tlieorem: For a 2 1,
ewry C" structurc roiitains a CB structurc for eacli /3 3 a;it is not unique, of
c,ourse, but it is ui~iqueup to diffeomorpliism. Tliis will not be proved liere.*
In fact, C" maiiifolds for a # w will hardly ever be mentioned again. One
irinark is iii ordei- 130~~;tlie proof of hmma 2 produces an appropriate C"
fiinciion f oii a C" inaiiifold, for O 5 a 5 w. Of course, for a = o the proof
* hi.a p1-00fsee h4unkrcs, EIentefztayDitpereniial %/~olog)l.
fails completely (aiid ihe result is fahe-an analytic function which is O on an open
fails completely (aiid ihe result is fahe-an
analytic function which is O on an
open set is O everywhere).
With difirentiable functions now at our disposal, it is fitting that we begin
differentiating them. What we S hall define are the partid derivatives of a dif-
ferentiable function f : M + R, with respeci to a coordinate system (x, U). At
this point classical notation for partid derivatives is systematically introduced,
so
it is wortli recalling a logicai notation for the partial derivatives of a function
f
: R" + R. We denote by Di f (u) the number
f(u'y*.*yui+h,-.-,u")- f(u)
lim
h-O
I?
The Chaiii Rule states that if g : Rm + R" and f : R" + R, then
n
Dj (f
0 g) (u) =
Di f (g(u)) - ~j gi (u).
i=21
Now, for a function f : M + R and a coordinate system (x, U) we define
af
(or simply -
= Di(f o x-') o r,as an equation behveen functions). If we
axl
define tlie curve ci : (-E, E) + M by
then this partial derivative is just
so it measures the rate change of f along tlie curve c;;in fact it is just (f oci)'(0).
Notice tliat
ax3
1
ifi= j
-
(p) = 6;
-
aA-J
1f x liappens to be the identity map of Rn, then Dtf(p) = 8f/axi(p), whidi
is the classical synibol for this partid derivative.
Anot her classical instance of this notation, often not completely clarified, is thc use of
Anot her classical instance of this notation, often not completely clarified, is
thc use of the syrnbols a/ar and a/a9 in connection with "polar coordinates".
On the subset A of JRZ defined by
we can iii troduce a "coordinate system" P : A + IW'by
where r(x, y) = m
and 9(x, y) is the unique numbcr in (O, 2rr) with
Tliis really is a coordinate system on A in our sense, with its image being the
set (r
: r
> O) x (0, k).(Of course, the polar coordinate system is often
not restricted to tlie set A. One can delete any ray other than L if 9(x, 1))
is restricted to lie in the appropriate interval (60, 90 + 2~);many results are
essentially independent of which line is deleted, and this sometimes justifies the
slopl~inessinvolved in the definition of the polar coordinate system.)
IVe liavc 1-eallydefined P as an inverse function, whose inverse P-' is defined
simply by
P-'(r,9)
= (rcos9,r sino).
From this formula we can compute fiar explicitly: (f o P-')(r, O) = f (r
From this formula we can compute
fiar explicitly:
(f o P-')(r,
O) = f (r cos O, i- sin O),
af
-(x? ar
Y) = Di (f 0 p-'1 (P(x9 Y))
= DIf (P-' (~(x,Y)) Di [P-'11 (P(x, Y))
+ ~zf(p-'(p(x,
Y))- Di [P-'I~(P(X~Y))
by the Chain Rule
= DIf (x, y)
tos O(x, y) + Dzf (x, Y) sino(x, Y).
Tliis formula just gives tlie value of the directional derivative of f at (x, JI),
along a uiiit \lector v = (cos O(x, y), sin O(x, y)) pointing outwards from the
origin to (x, y). Tliis is to be expected, because cl, the inverse image under P
sin B(x,y)
of a culve along tl~e"1.-axis", is just a line in tliis direction.
A similar computation gives
The lector w = (- siii O(x, y), cos 0(x, y)) is
to v, and thus tlle
direction, at tlie point (A-, y), of the curve cz ivliicli is tlie inverse image under P
of a cunfe along tlle "O-axis". Tlie factoi- i. (x, y) appears because this cuive
goes around a circle of that radius as 8 goes from O to 2n,so it
goes around a circle of that radius as 8 goes from O to 2n,so it is going i-(x,y)
times as fast as it should go in order to be used to compute the directional
derivative of f in the direction w, Note that af/aO is independent of which
line is deleted from the plai~ein order to define the function 8 unarnbiguously.
Using the notation af/ax for Di f,etc., and suppressing the argument (x, 11)
evenutliere (thus writing an equation about functions), we can write the above
equations as
af
af
af
cos 8 + - ay sin 8
-c=-
ax
ay
- 8.f
?f
a.f
=
cos 8.
ae
-(-)e
ax
sin 8) + -r
ay
111 particular, these formulas also te11 us what ax/ai- etc., are, where (x, y)
rlrnotes ilie identit~coordinate system of RZ.We have ax/ar = cose, etc., so
oui- formulas can be put in the form
af
afax
- = --
+-- aya~
al- a~-al-
In classical notation, the Cl~ainRule would always be written in this way. It is
a pleasui-e to repoi-t thai Iienceforih this may always be done:
3. PROPOSITION.
If (S, U) and (y, V) are coordinate systems on M, and
f : M + R is differentiable, then on U n V we have
PROOF.
It's tlie Chaii~Rule, of course, if you just keep your cool:
At this point we could introduce the "Einstein summation convention". No- tice that the summation
At this point we could introduce the "Einstein summation convention". No-
tice that the summation in this formula occurs for the index j, which appears
both "above" fin axj/ayi) and "below" (in af/axj). There are scads of for-
mulas in which this happens, often with hoards of indices being summed over,
and the convention is to omit the C sign completely-double indices (wliicli
by luck, tlie nature of t hings, and felicitous choice of notation, almost always
occur above and below) being summed over. 1 won't use this notation because
wheiiever 1 do, 1 sooii forget I'm supposed to be summing, and because by do-
ing things "riglit", one can avoid what Élie Cartan lias called the "debauch of
indices".
We will often write formula (1) in the form
here alay' is considered as aii operator taking the function f
operator taking f to af /ayi (p) is denoted by
to af/ayi. The
For later use we record a property of f = a/axil,:
it is a "point-derivation".
4. PROPOSITION. For any diflerentiable f, g : M + IR,and any coordinate
system (A-,U) with p E U, the operator l = a/axilPsatisfies
Kfg) = f (p)f(g) + f(f )g(p).
PROOF.
Lft to the reader.
If (A-,U) and (x', U') are two coordinate systems on M, the n x n matrix
is just the Jacobian matrix of x1 o A--' at x(p). It is non-singular; in
is just the Jacobian matrix of x1 o A--'
at x(p).
It is non-singular; in fact, its
inverse is clearly
Now if f: M" + Nm is Cm and (y, V) is a coordinate system around f(p),
t he rank of the m x 11 matrix
clearly does not depeiid on the coordinate system (x, U) or (y, V). It is called
tlie rank of f at p. The point p is called a critical point of f if the rank of f
at p is 4 i?7 (the dimension of the image N); if p is not a critical point of f,
it is called a regular point of f. If p is a critical point of f,the value f (p) is
called a cñtical value of f. Other points in N are regular values; thus y E N
is a regular value if and only if p is a regular point of f for ewry p E f -' (y).
Tliis is true, in particular, if q $ f (M)-a non-value of f is still a "regular
value".
If f: W + W, tlien x is a critica] point of f if and only if ft(x) = O. It
is ~ossiblefor al1 points of the intenral [u, b] to be critical points, although this
can happen only if f is constant on [u, b]. If f : WZ + IR has al1 points as
critical values, tlien Dl f = Dzf = O everywhere, so f is again constant. On
tlie other hand, a function f : WZ + IRZ may have al1 points as critical points
without being constant, for example, f (x, y) = A. In this case, Iiowever, the
iinage f (WZ)= W x {O) c WZ is still a "small" subset of IRZ. The most importan1
theorem about critical points generalizes this fact. To state it, we will need some
terminology.
Recall tliat a set A c W" has "measure zero" if for every E > 0 tliere is a
sequeilce Bi, Bz,
of (closed or open) rectangles with
and
wliere u(&) is the volume of B,l. We want to define the same concept for a
subset of a manifold. To do this we need a lemma, wliich in turn depends on
a Icmma from Caku/us on Adanfolds, which we merely state.
5. LEMMA. Let A C IR" be a rectangle and let f: A + Rn
5. LEMMA.
Let A
C
IR"
be a rectangle and let f: A +
Rn be a function
such that ~~~f'l5 K on A for i, j = 1,
,
n. Then
for al1 x, y E A.
6. LEh4MA. If f : IR" + IR" is C' and A c IR" has measure O, then f (A) has
measure O.
PROOF. We can assume that A is contained in a compact set C (since IR" is a
countable union of compact sets}. Lemma 5 implies that there is some K such
that
If (x) - f (r)l 5 n2Klx - yl
for al1 x,y
E
C.
Tllus f
takes rectangles of diameter d into sets of diameter
- >I' Kd. This clearly implies that f (A) has measure O if A does. +3
A subset A of a Cm 71-manifold M Iias measure zero if there is a sequence
of charts (xi, Ui), with A c Ui Ui, SUC~that each set xi(A n Ui) c IR" has
measure O. Using Lmma 6, it is easy to see tliat if A c M has measure O, then
x(A n U) c IR" Iias measure O for any coordinate system (x, U). Conversely,
if this condition is satisfied and M is connected, or has only countably many
compoiieiits, then it follows easily from Theorem 1-2 that A has measure O. (But
if M is the disjoint union of uncountably many copies of IR, and A consists of
one poirit from each compoiient, then A does not have measure O). Lemma 6
thus implies another result:
7. COROLLARY.
If f: M + N is a C' fu~ictionbetween two n-manifolds
and A c M has measure O, then f (A) c N has measure O.
PR001;. Tliere is a sequelice of charts (xi, Ui) with A c Ui U; and each set
xi(A n Ui) of measure O. If (y, V) is a cliart on N, then f (A) n V = Ui f (A n
Ui)n V. Each set
Iias measure O, by Lemma 6. Thus y( f (A)n V) has nieasure O. Since f (UiUi)
is contained in tlie uiiion of at most countably many components of N, it follows
that f (A) has measure O. e+
8. THEOREM (SARD'S THEOREM). If f : M + N is a Ci map between
8. THEOREM (SARD'S THEOREM). If f : M + N is a Ci map between
n-manifolds, and M has at most countably many components, then the critical
values of f form a set of measure O in N.
PROOI;. It clearly suffices to consider the case where M and N are Rn. But
this case is just Theorem 3.14 of Calculw on Manfob.
Tlie stronger version of Sard's Theorem, which we will never use (except once,
in Problem 8-24), states* tliat the critical values of a C' map f : M" + N"'
are a set of measure O if k 2 1 + max(n - in,O). Theorem 8 is the easy case,
and tlie case nz > n is the trivial case (Problem 20). Although Theorem 8 will
be very important later on, for the present we are more interested in knowing
wliat tlie image of f: M + N looks like locally, in terms of the rank k of f
at p E M. More exact information can be @ven wlien f actually has rank k
in a neighborhood of p. It should be noted that f must have rank 2 k in
some iieiglihor-hood of p, because some k x k submatrix of (a(yi o f)/axi)
has non-zero determinant at p, and hence in a neighborhood of p.
9. THEOREM.
(1) If f : M" + Nm has rank k at p, then tliere is some coor-
dinate system (x, U) around p and some coordinate system (y, V) around f(p)
with y o f ox-'
in the form
Moreover, given any coordinate system J?, the appropriate coordinar e system
ori N can be obtained merely by pei-muting the component functions of y.
(2) If f has rank k in a rieiglil~orlioodof p, theii tliere are coordinate systems
(A-,U) and (y, V) such that
Ren~ark:Tlie special case M = Rn, N = Rm is equivalent to the ge~ieraltlieo-
rem, wliicli @\res only local results. If y is tlie identity of R", part (1) says thai
ly first pei-formiiig a difleoniorpliism on Rn, arid tlieri permuting the coordi-
iiates in Rm,we can irisure that f keeps tlie first k components of a poiiit fixed.
Tliesc difTcornoiphisms on Rn and Rm are clearly necessary, since f may not
everi be oiie-oiic on Rk x {O) c RW aad itr image could, for example, contain
oiily ~ointswith first coordinate O.
* Foi-a proof, scc hqiliroi; To/)ologilhtn tlic Ditp~~'nl&inblcI4cuy)oi)iror Sternberg, LEcllrres ott
Dflirenliol Geonzelv.
In part (2) we must clearly allow more leeway in the choice of y, since
In part (2) we must clearly allow more leeway in the choice of y, since f (IRn)
rnay not be contained in any k-dimensional subspace of IRn.
I'ROOF. (1) Choose some coordinate system u around p. By a permutation of
ilie coordinate functions u' and y' we can arrange that
Define
Condition (1) implies that
This shows thai A- = (x o u-') o u is a coordinate system in some neighbor-
I~oodof p, since (2) and the Inverse Function Theorem show that x o u-'
dfleomorphism in a ncighborhood of u(p). Now
is a
1
q = x-'(a',
.,an)
means
x(q) = (a ,
,an),
i
hence
x'(~)= a ,
SO
Y~fox-l(al,
,aJ1)=yo
f(q)
for q = x-l
(a1 ,
,an)
1
=(a
,
,a
k
,
).
(2) Choose coordinate systems x and v so that v o f o x-' has the form in (1).
Since rank f = k in a neigliborhood of p, the lower square in the matrix
must vanish in a neighborhood of p. Thus we can wnte Define ya = va
must vanish in a neighborhood of p. Thus we can wnte
Define
ya = va
y'=
vr -$ro(~',
,V
k ).
Since
has non-zero determinant, so y is a coordinate system in a neighborhood
of f (p). Moreover,
Theorem 9 acquires a special form when the rank of f is n or nz:
10. THEOREM.
(1)
If
m 5 ,I
and f
: M" -t
Nm lias rank m at p, then for
any coordinate sysiem (y, V) around f (p), there is some coordinate system
(x, U) around p with
y o f ox-'(a',
.,an) =
,am).
(2) If 11 5 m and $: Mn + N"' has rank ir ai 11,
(2) If 11 5 m and $: Mn + N"' has rank ir ai 11, tlien for any coordinate
system (x, U) around p, there is a coordinate system (y, V) around $(p) with
i
1
n
yof~x-~(a,
,an)
= (a
,
,
a
,O,
,0).
PROOF. (1) This is practicdly a specid case of (1) in Theorem 9; it is only
rlecessary to observe that when k = rn, it is clearly unnecessary, in the proof of
this case, to permute ihe yi in order to arrange that
det (a(y~u~f)(p))f 0
a,p =
,m;
only the u' need be permuted.
(2) Since the rank of f at any point must be 5 n, the rank of $ equals n
in some neighborhood of p.
It is convenient to think of the case M - R"
and N = RJn and produce the coordinate system y for Rm when we are Siven
ihe identity coordinate system for R". Par-t (2) of Theorem 9 yields coordinate
systems # for Rn and S, for Rm such that
1
@o f o#-'(a
,
1
,a")
= (a
,
-,an,O,
,O),
Even if we clo rioi pei-form #-'. firsi, tl-ie map f still takes EXn into the subset
(R") wliich S, takes to R" x (O) c Rm-the
points of Rn just get moved to
ihe wrong place in Rn x (O). This can be corr-ected by another map on R".
Define h Ily
h(bi ,
,bJn)=
(4-' (bl,
,b"),
b"+',
,bm).
Then
hoS,o$(ai,
,an)=hoS,o$o#-'(b1
,
,b")
for (b',
,bn)
= #(a)
= ,ón,0,
.,O)
= ($-'(bl,
,b"),O,.
.
.,O)
i
R
= (a
,
,a
,O,
. ,O),
so h o S, is tl-ie desired
If we are Siven a coordinate system x on R" other
)J.
than the identify, we just define
h(bl,
,b"')
= (~(4-'(bi,-
.
.,b')),
,bm);
it is easily cliecked tliat y = h o S, is rlow the desired y. +3
Aliliough p is a regular point of f in case (1) of Theorem 10 and
Aliliough p is a regular point of f in case (1) of Theorem 10 and a critica1
point in case (2) (if n < m), it is case (2)\vhich most interests us. A dflerentiable
funciion f:Mn + Nm is called an immersion if the rank of f is n, the
dimension of the domain M, at al1 points of M. Of course, it is necessary that
In > n, and ii is clear from Theorem 10(2) that an immersion is locally one-one
(so it is a iopological immersion, as defiiied in Chapter 1). On the other hand,
a differentiable map f need not be an immersion ewn if it is globally one-one.
The simplest example is ihe function f:IR + IR defined by f(x) = x3, with
f (O) = O. Another example is
A more illun~inatingexample is the function h : IR + IR2 defined by
alihough its image is tlie graph of a non-dflerentiable function, the curve itself
manages to be differenkiable by slowing down to velocity O at the point (0,O).
One can easily define a similar curve whose image looks like the picture below.
Tliree immersions of IR in IR2 are shown below. Althougll the second and
iliii-d immei-sions Bi and B2 are one-one, iheir iinages are not homeomorphic
to R. Of course, even if tl-ie one-one immersion f : P + M is
to R. Of course, even if tl-ie one-one immersion f : P + M is not a homeo-
niorpliism onto iis image, tliere is certainly some metric and some dflerei~tiable
sti-ucture on f (P) which makes the inclusion map i: f (P) + M an inlmer-
sion. In general, a subset MI c M, with a dflerentiable structure (iiot nec-
cssarily compatible with the metric Mi inherits as a subset of M), is called an
immersed submanifoId of M if the inclusion map i : MI + M is an immersion.
l'he follo\ving picture, indicating the image of an immersion Bp: R + S' x S],
rS3
first time around
second time around /
sl-iowsthat Mi may even be a dense subset of M.
Despite these complications, if Mi is a k-dimensional immersed submanifold
of Mn and Ui is a neighborhood in M' of a point p E MI, tllen there is a
coordinate sysiem (y, V) of M around p, such that
tliis is an immediaie consequerice of Tlieorcm 10(2), with f = i. Tlius, if
g ; Mi + N is Cm (co~isideredas a function oii tlie manifold Mi) in a neigli-
borhood of a point p E MI, ihen ihere is a Cm furiction g on a rieigliboi-liood
V c M
of
p sucli that g = g O i on V n Mi-we
can define
~'~(q')= ~'"(4)
a= 1,
,k
S(4) = g (qi),
where
r = k+
l,.s.,iz.
On the oilier hand, even if g is Cm on al1 of MI we may
On the oilier hand, even if g is Cm on al1 of MI we may not be able to
define g on M. For example, this cannot be done if g is one of the functions
B;' jj(M) + R.
One other complicatioi-i arises with immersed submanifolds. If MI c M is
an immersed submailifold, aild f : P + M is a Cm function with f (P) c Mi,
ii is not necessarily true that $ is Cm when considered as a map into MI, with its
Cm structure. The following figure shows that f might not even be continuous
as a map jnto MI. Actually, tliis is the only thing that can go wrong:
11. PROPOSITION.
If MI c M is an immersed manifold, f: P +
M
is
a Cm function witli f (P)c MI, and f is coniinuous considered as a map
irlto MI, then f is also Cm coilsidered as a map into MI.
PROOF.
Let i : MI + M be tlie inclusioil n-iap. We want to show that iP1o j'
is Cm if it is continuous. Given p E P, choose a coordiilate system (y, V) for M
around f (p) such that
is a iieigliboi-hood of f (p) in MI and (yi1U],
of MI on U].
,yk 1U]) is a coordiliate system
By assumption, i-' o f is continuous, so f -' o i(open set) is an
By assumption, i-'
o f is continuous, so
f -' o i(open set) is an open set.
Since Ui is open in MI, this means that f -' (Ul) c P is open. Thus f takes
some neighborhood of p E P into U,. Since al1 yj o f are Cm, and y',
,y
k
are a coordinate system on U1, the function f is Cm considered as a map
into MI. 43
Most of these difficulties disappear wlien we consider one-one immersions
f : P + M which are liomeomorphisms onto their image. Such an immersion
is called an imbedding ("embedding" for the English). An immersed subman-
ifold Mi c M is called siinply a (Cm) submanifold of M if the inclusion map
i; MI + M is an imbeddirlg; it is called a closed submanifold of M if MI is
also a closed subset of M.
Tliere is one way of getiiilg submanifolds wl-iich is very important, aild @ves
the sphei-e S"-'
C W" - {O) c JR", defined as (x : 1x12 = 11, as a special case.
12. PROPOSITION.
If f : M" + N lias constani railk k on a neighborhood
of f -' (J!), then f -' (y) is a closed submanifold of M of dimension n - k (or
is empty).
Iii particular, if y is a regular value of f : M" + Nm, then f -' (y)
is an (11 - 111)-dimensionalsubmanifold of M (or is empty).
PROOF. Left to the reader. 43
It is to be hoped that however abstract the noiion of Cm manifolds may
appear, sul~manifoldsof JRN will scem like faii-ly concrete objects. Now it turns
out tliat euqr (connected) Cm manifold can be imbedded in some JRN, so that
manifolds caii be pictured as subsets of Euclidean space (though this picture
is iiot always die nlost uscful one). We will prove tliis fact oilly for compaci
mailifolds, but we first develop some of the macl~iileiywhich would be used in
the general case, since we xrtil1 need it later on anyway. Unfortunately, there are many
the general case, since we xrtil1 need it later on anyway. Unfortunately, there are
many definitions and theorems involved.
If
O
is a
cover of a space M,
a cover O'
of
M is a refinement of O (or
"refines O ") if for every U in O' there is some V in O with U c V (the sets of O'
are "smaller" than those of O)-a subcover is a very special case of a refining
cover. A cover O is called locally finite if every p E M has a neighborhood W
which intersects only finitely many sets in O.
13. THEOREM. If O is an open cover of a manifold M, then there is an open
cover O' of M whicli is locally finite and which refines O. Moreover, we can
choose al1 members of O' to be open sets dfleomorphic to R".
PROOF. \Ve can obviously assume that M is connected. By Theorem 1.2, there
are compact seis
- with M = CI U C2 U C3 U -- - - Clearly Ci has
an open neighborliood U] with compact closure. Then U Ci has an open
neighborhood U2 ~4thcompact closure. Continuing in this way, we obtain
open sets Ui, witli compact and c Ui+i, whose union contains aH C;,
and lience is M. Let U-] = U. = 0.
-
Now M is ilic union for i > 1 of the "aniiular" regons Ai = Ui - Ui-]. Since
eacli Aiis compact, we can obviously cover Ai by a finite number of open sets,
eacli coiiiaiiied in some membcr of O, and eacli coiitained in Vi = Ui+, - Ui-2.
\Ve can also clioose these open sets to be diffeomoi.phic to RR.In this way we
obiain a cover O' wliich refines O aild wliicli is locally finite, since a point in Ui
is noi in 6 for j 3 2 + i. /f,
Notice that if O is an open locally finite cover of a space M and
Notice that if O is an open locally finite cover of a space M and C C M is
compact, then C intersects o~ilyfinitely many members of O. This shows that
aii open IocaHy finite cover of a connected manifold must be countable (like the
cover constructed in the proof of Theorem 13).
14. THEOREM (THE SHRINKING LEMMA). Let O be an open locally
firiiie co\rer of a manifold M. Then it is possible to choose, for each U in O, an
open set U' wiui c U in such a way that the collection of al1 U' is also an
open cover of M.
PROOF. 1442 can clearly assume that M is connected. Let O = (Ui, U2,
Then
.).
c1= u, - (U2UU3U.--)
is a closed set co~itainedin U], and M
open set witli Ci c U; c c U1 Now
=
Ci U U2 U U3
Lt
U{ be an
is a closed set contained in U2, and M = U; U C2 U U3 U
-
.
Let
U;
be an
open set wiili C2 c U; c
c U2. Continue in this way.
For aliy p
Now
E
M tliere is a largest n witli p
E
U,,
because O is locally finite.
p E u; UU;
U
-S U u; U (u.+, U UN+2U.-.);
it follows that
P E U/UU;Ua.*,
since replacing U,+í by Un+i cannot possibly eliminate p. 34
15. THEOREL4. Let O be an open IocaHy firiite cover of a manifold M. Then
ihere is a collection of Cm functioiis #u: M +
such tllat
[O,11, one for each U in O,
(1) support $U C U for each U,
$u(/)) =
1 for al1 p
E M (this sum is really a finite sum in some
(2)
U
iieigliborliood of p, by (1)).
PROOF. Cme 1. Each U in O Rar ~omjia~tclosure. Choose the U' as in Theorem
PROOF. Cme 1. Each U in O Rar ~omjia~tclosure. Choose the U' as in Theorem
14. Apply Lemma 2 to
c U c M to obtain a Cm function Jru : M
+ [O, 11
wliicli is 1 on aiid has support c U. Since the U' cover M,clearly
Define
Cme 2. Gerie~alcae. Tliis case can be proved in tlie same way, provided diat
Lemma 2 is irue for C c U c M with C closed (but not necessarily compact)
and U open. But this is a consequence of Case 1:
For eacli y E C choose an open set Up c U with compact closure. Cwer
M - C with open sets V. haviiig compact closure and contained in M - C.
The open cover (Up,V
applies. Let
)lias
an open locally fiilite refinement O to which Case I
Tliis sum js Cm,silice it is a fi~iitesum iii a iieigliborliood of each point. Since
xu#u (p) = 1 for al1 p, arid #u (p) = O when U c V., clearly J(p) = 1
fol. al1 p E C. Usirig tlic faci tliat O is locally finite, it is easy to see tliat
support f c U.+:+
16. COROLLARY.
If O is aiiy open cover of a n~aiiifoldM, then ihere is a
collection of Cm functions #i : M
+ [O, I] such that
(1) tlie collection of sets (p : #i (p) # O) js locally finite,
(2) ti#i(p)= I for al1 p E M,
(3) for eacli i tliere is a U E O such that support #i C U.
(A collccti~ri(#i : M
+ [O, 11: satisfying (1) and (2) is called a partition of unity;
if it satisfics (3), ii is called subordinate to O.)
It is riow fairly easy to prove tlie last theorem of tliis cliapter.
17. THE0REh.I.
If M" is a conipact Cm mailifold, tlien tliere is aii iml~ed-
cliiig /:
M + IRN for some N.
1'1200F. TIiere are a £nite number of coordinaie systems (xl, Ul), , (xk,Uk) witl-i M
1'1200F.
TIiere are a £nite number of coordinaie systems (xl, Ul),
, (xk,Uk)
witl-i M = Ul U - - -UUk. Choose U; as in Tlieorem 14, and functions @i : M +
[O, 11 whicli are
N=nk+k,by
1 on
and have support c Vi. Define f: M + RN, where
Tl-iisis ai-i inlmersion, because aiiy poiiit p is in U; for some i, and on U,', where
qfj = 1, ihe N x it Jacobian matrix
It
is also one-one. For suppose that f(p) = f(q).
There is some i such that
p
E U;. T1-iei-i @i(p) = 1, SO also @i(q) = 1- This shows that we musi llave
y E Ui. Moreover,
@i ' xi (P) = @i- Xi (Y),
Problem 3-33 sl-iows that, in fact, we can always choose N = 2rz + 1.
1, (a) Sliow tliat beir-ig Cm-r-elated is no.! a1.i equivalente relation.
(b) 111 the pl-oof of Leninia 1, show ihat al1 charis in A'
claimed.
are Cm-related, as
2. (a) If M is a 1-ileti.i~space togeiber wiil-i a collection of homeomorphisms
x: U + Rn whose domains cover M and which are Cm-related, show that
tl-ie tt at each yoiiit is ui-iique rtrillrouf using Irivariai-ice of Domain.
(b) SI-iowsin-iilarlythai aM is well-defined for a Cm manifold-with-boundary M.
3. (a) Al1 Cm fui-ictionsare continuous, ai-id il-ie coinposition of Cm functions
is Cm.
(b) A funciion f : M + N is Cm if al-id only if g o f is Cm for every Cm
function g : N + R.
4. How inaiiy clistii-ict Cm stl-ucturesare thel-e 011 R? (Tlleie is only one up to
diffeon-iorpliisin;il-iat is i-ioi the question beil-ig asked.)
5. (a) If N c M is open and A' consists of al1 (x, U)
5. (a) If N c M is open and A' consists of al1 (x, U) in A with U c N, show
thai A' is maximal for N if A is maximal for M.
(b) Show that A' can also be described as ihe set of al1 (xlV íl N, V íl N) for
(x, V) in A.
(c) SIlow ihai ihe inclusion i: N + M is Cm, and that A' is the unique atlas
with this property.
6. Clieck thai tlie iwo projections Pl and P2 on S"-] are Cm related to the 2n
l~omeomor~hisms_I;: and gj.
7. (a) If M is a connected Cm marlifold and p,q E M, tl-ien ihere is a Cm
cum c: [O, I] +
M with c(0) = p and c(I) = q.
(b) It is even possible to clioose c to be one-one.
8. (a) Sliow ihat (MI x M2) x M3 is dfleomorphic io MI x (M2 x M3) and
iI-iai MI x M2 is dfleomorphic to M2 x MI.
(b) The dflcrentiable structure on MI x M2 makes the "slice" maps
of MI, M2 + M1 x M2 diffel-entiablefor a11 pl E Mi, p2 E M2.
(c) Pl4ore gerierally, a map f : N + MI x M2 is Cm if and only if the compo-
siiions rrl o f: N + Mi and rr2 o f : N + M2 are Cm. Moreover, ihe Cm
structure \ve llave defined for Mi x M2 is the only one wiih this property.
(d) If fj : N + Mi are Cm (i = 1,2), can one determine ihe rank of ($1,$2) : N
+ Mi x M2 (71 p in terms of the ranks of _f;: at p? For fj : Ni + Mi, show thal
Si x f2: Ni x N2 + Mi x M2, defined by ji x fi(pi, p2) = (Si(pi1, Sz(p2)),
is Cm and determine its rank in terms of the ranks of ji.
9.
Let g : S"
+
P"be tlie map p H
[p]. SIiow that f : P" + M is Cm if and
only if j' o g : S" + M is Cm. Compare the rank of f
and the rank of f o g .
10. (a) If
U c
R"
is open and f
: U
+
R is locally Cm (every point has a
iieiglil~orlioodori wllich f is Cm), tllen f is Cm. (Obvious.)
(b) If f: W" + R is locally Cm, then f is Cm, i-e-,f can be exieilded to a
Cm function on a ileighborhood of M". (Not so obvious.)
11. If f : W" + R Ilas hvo extensions g, h to Cm functions in a neighborhood
of M", illeli Qg and Qh are ihe same at points of R"-' x (O) (so we can speak
of Dif at these points).
12. If M is a Cm rnanifold-wiil-i-boundary, tlien there is a unique Cm struciure
ori aM sucll iliai il-ie inclusion map i: aM + M is an imbeddiilg.
13. (a) Let U C M" be an open set sucb ihat boundary U is
13. (a) Let U C M" be an open set sucb ihat boundary U is an (n - 1)-dimen-
sional (dflereiitiable) submanifold. Show tliat U is aii 11-dimensionalmanifold-
with-boundary (It is well to bear in mind the following example: if U = {x E
EX' : d(x,O) < 1 or 1 < d(x,O) < 21, then Ü is a manifold-with-boundary, but
aÜ # boundary U.)
(b) Consider the figure shown below. This figure may be extended by putting
srnaller copies of tlie two parts of S' into the regions indicated by arrows, and
ihen repeatirig tliis coristi-uction iudefinitely. The closure S of the final resulting
figure is known as Alexnnder's Houied SSplnr. Sliow tliat S is liomeomorphic to s2.
(Hinl: Tlie additional poinis in tbe closure are bomeomorphic to the Cantor
sei.) If U is tlie uiibourided component of W3 - S, ihen S = boundary U, but
Ü is not a 2-dimensional manifold-with-boundary, so part (a) is true only for
differentiable submanifolds.
14. (a) Therc is a map f : EX2 + EX2 SUC~Itliat
(1) f(x,O) = (x,O) for al1 x,
(2) f(x,y) c EI2 for y > O,
(3) f (~,y)c W2 - IN2 for y < O,
(4) f restricted to the upper half-plane or the lower half-plane is Cm, but f
(4) f restricted to the upper half-plane or the lower half-plane is Cm, but f
itself is not Cm.
(b) Suppose M and N are Cm manifolds-with-bounda~yand f : aM + aN
is a difiomorphism. Let P = M U1 N be obtained from the disjoint union
of M and N by identifying x E aM with f (x) E aN. If (x, U) is a coordinate
system around p E aM and (y, V) a coordinate system around f(p), with
f(U naM) = V naN, and (yo f)IU naM = xlU naM, we can define a
homeomorphism from U U V c P to IRn by sending U to Wn by x and V to
tlle Iower half-plane by the reflection of y. Show that this procedure does not
define a CbO structure on P.
(c) Now suppose t11at there is a neighborhood U of aM in M and a dfleo-
morphism a: U + 8M x [O, l), such that a(p) = (p,O) for al1 p E aM, and a
similar diffeomorphism p : V + aN x [O, 1). (M'e will be able to prove later that
such diffeomorpllisms always exist). Show that tllere is a unique CbO structure
on P sucl-itl-iattl~eii~clusionsof M and N are CbOand such that the map from
U U V to aM x (- 1,l) induced by a and B is a dfleomorphism.
(d) By usiilg two difTei-entpairs (a,p), define two dflerent Cm structures on EX2,
collsidcred as thc union of two copies of W2 witli corresponding points on aW2
iclei~tified.SIlow that the resulting CbO manifolds are dfleomorphic, but that
the difKeomorphism ca?~nolbe chosen arbitrai-iIy close io the identity map.
15. (a) Find a Cm structure on W' x M' which makes the úidusion into
15. (a) Find a Cm structure on W' x M' which makes the úidusion into EX2
a Cm map. Can the inclusioi-i be an imbedding? Are the projections on each
factor Cm maps?
(b) If M and N are manifolds-with-boundary, construct a Cm structure on
M x N such that ali the "slice maps" (defined in Problem 8) are Cm.
16. Show that the fui-iction f : IR -+ IR defined by
is Cm (the formula e-'ix2
is used just to get a function which is > O for x < 0,
and e-'/lXl could be used just as weil).
17. Lemma 2 (as addended by the proof of Theorem 15) shows that if Cl and C2
are disjoint closed subsets of M, then there is a Cm function f: M + [O, 11
such that Ci c f -'(O) and C2 c f -'(I). Actually, we can even find f with
Cl = /-'(O) and C2 = f-' (1). The proof turns out to be quite easy, once you
know the trick.
(a) It suffices to find, for any closed C c M, a Cm function f with C = /-'(O).
(b) Let (Ui) be a countable cover of M - C, where each Ui is of the form
for some coordinate system x taking an open subset of M - C onto EXn. Let
Si : M +
[O, 11 be a Cm function with _I;: > O on Ui and j;. = O on M - Ui.
Functions like
-
axi'
axíaxk5
"'
will be called mixed partials of A, of order
Let
ai = sup of all mixed partials of Si,
,f;.
of al1 orders 5 i.
Show that
f=C-
" Si
ai2'
i=I
is Cm, and C = f -l
(O).
18. Consider the coordinate system (y1,y2) for IR2 defined by
y' (a,b) = a
y2(a,b) = a +b.
(a) Compute f/ayr(a, b) from the definition. (b) Also compute it from Proposition 3 (to
(a)
Compute
f/ayr(a, b) from the definition.
(b)
Also compute it from Proposition 3 (to find aIi/ayj, write each 1' in terms
of y' and y2).
Notice that af/ayl # 8//alr
on y and i, not just on y'.
even though
= Ir;the operator alay' depends
19. Compute the "Laplacian"
in terms of polar coordinates. (First compute a/ax in terms of alar and a/a9;
ra
a
a
ra
then compute a2/ax2 from this). Amwer: ;
a;)+ae(,=)J.
20. If
: M"
+
Nm is C1 and m > n, then /(M)
has measure O (provided
that M has only countably many components).
21. The following pictures show, for 11 = 1,2, and 3, a subdivision of [O, 11 x
[O, 11 into 22n squares,
, An,2b; square An,k is labeled simply k. The
numbering is determined by the following conditions:
(a) The lower left square is A,,].
(b) The upper left square is An1221f.
(c)
Squares A,,k
and AR,k+lhave a common side.
(d)
Squares An,41+1, An,41+2, An,41+3i An,41+4 are contained in A,-, ,[+l.
Define f: [O, 11 +
[O, 11 x [O, 11 by the condition
k-1
k
f(t) E An,k
for al1
-
- <ts-.
22"
22"
Show that f is continuous, onto [O, 11 x [O, 11, and not one-one.
22. For p/2" E [O, 11, define f (p/2") E lR2 as shown below. (a)
22. For p/2" E [O, 11, define f (p/2") E lR2 as shown below.
(a) Show that f is uniformly continuous, so that it has a continuous extension
g: [O, 11 + lR2. Show that g is one-one, and that its image wiil not have
measure O if tlie shaded triangles are chosen correctiy.
(b) Consider tlie homeomorphic image of S' obtained by adding, below the
image of g,a semi-circle with diameter tlie line segment AB, JVhat does the
inside of this curve look like?
23. Let c : [O, 11 + Rn be continuous.
For eacli partition P = {to,
. , fk) of
[O, 11, define
k
The curve c is rectifiable if {l(c, P)) is bounded above (with length equal to
sup{l(c, P))). Show tliat the image of a rectifiable curve has measure O.
24. (a) If M is a Cm manifold, a set Mi c M can be made into a k-dirnen-
sional submanifold of M if and only if al-ound each point in MI there is a
coordinate system (x,U) on M such tliat Mi n U = {p: xk+'(p) = =
x"(p) = O).
(b) Tlie subset Mican be made into a closed submanifold if and only if such
coordinate systems exist around every point of M.
25.
Tlie sei {(x, 1x1); x E R) is noi tlie image of aiy immersion of R into IR2.
26.
(a) If
U
c
Rk is open
and f: U
IRn-'
-t
is Cm, then the graph of
f = {(p, f(p)) E R": p E U) is a submanifold of Rn.
(b) Every submaliifold of R" is locally of tiiis form, after renumbering coor-
dinates. (Neitlier Theorem 9 nor 10 is quite strong enough. You wiil need
the implicit function theorem (Gilculw un Manzjbolds, pg. 44)).Theorem 10 is es- sentially Theorem 2-13
the implicit function theorem (Gilculw un Manzjbolds, pg. 44)).Theorem 10 is es-
sentially Theorem 2-13 of Celculw on ManifoHs; comparison with the implicit
function theorem will show liow some information has been allowed to escape.)
27. (a) An immersion from one n-manifold to another is an open map (the
irnage of an open set is open).
(b) If M and N are n-manifolds with M compact and N connected, and
f: M + N is an immersion, then f
is onto.
28. kove Proposition 12: If f: M" +
N has constant rank k on a neigh-
borhood of f-'(y), then f -' (y) is a (closed) submanifold of M of dimension
n - k (or is empty).
29. Let f: P'
+ W3 be the map
defined in Chapter 1, whose image is the Steiner surface. Show that g fails to
be an immersion at 6 points (tlie image points are the points at distance &1/2
on each axis). There js a way of immersing P2 in IR3, known as Boy's Surface.
See Hilbert and Col1n-Vossen, Geomeiy and !he Imagination, pp. 3 17-32 1.
30. A continuous f~inctionf: X + Y is proper if f-'(C) is compact for
evcry compact C c Y. The limit set L(f) of f is the set of al1 y E Y such
ihat y = lim f(x,) for some sequence
subsequence.
E X with no convergent
(a) L( f) = 0 if and only if f is proper.
(b) f (X) c Y is closed if and only if L(f) c f (X).
(c)
Tliere is a continuous f: W + IWZ with f(W) closed, but L(f) # 0.
(d)
A onc-one continuous function f: X + Y is a homeomorphism (onto its
image) if and only if L(f) n f(Y) = 0.
(e) A submanifold Mi c M is a closed submanifold if and only ifthe inclusion
map i: MI + M is proper.
(f) If M is a manifold, there is a proper map f: M + IW; the function f can
be made Cm if M is a Cm manifold.
31. (a) Fiiid a cover of [O, 11 which is not locally finite but wliich is "point-
finite": every point of [O, 11 is in only finitely many members of the cover.
(b) Prove the Shrinking Lemma when the cover O is point-finite and countable
(notice that local-finiteness is not really used).
(c) Prove the Shrinking Lemma when O is a (not necessarily countable) point-
finite cover of any space. (You will need Zorn's Lemma; consider collections C
of ~airs(U, U') where U E O, U' c U, and t he union of
of ~airs(U, U') where U E O, U' c U, and t he union of all U' for (U, U') E C,
togetlier with all other U E O covers the space.)
32. (a) If MI c M is a closed submanifold, U 2 Mi is any neighborhood,
aiid f: Mi + R is Cm, then there is a Cm function f:M + W with / = f
on Mi, and with support J c U.
(b) This is false if M = W and MI = (0,l).
(c) This is false if R is replaced by a disconnected manifold N.
Remark: It is also false X M = RZ,MI = N = S', and f = identity; in fact, in
tliis case, f has no contiiiuous extension to a map from R2 to S', but the proof
requires some topology. However, f can always be extended to a CbOfunction
in a neighborhood of MI (extend locally, and use partitions of unity).
33. (a) The set of aH non-siiigular n x n matrices with real entries is called
GL(i1, R), tlie general linear group. It is a CbO manifold, since it is an open
subset of W" . The special linear group SL(n, R), or unimodular group, is the
subgroup of al1 matrices witli det = 1. Usirig tlic formula for D(det) in CahuEw
on A4aa@ds, pg. 24, show tliat SL(n, R) is a closed submanifold of GL(n, R) of
dimension n2 - 1.
2
(b) The symmetric n x 11 matrices may be thought of as R"("+')/~. Define
$: GL(n, W) + (symmetric matrices) by $(A) = A . A', wliere A' is tlie
traiispose of A. Tke subgroup $-'(I) of GL(J~,W)is called the orthogonal
group O(ri). Sliow tliat A E O(n) if arid only if tlie rows [or columns] of A are
orthonormal.
(c)
Show that O(n) is compact,
(d)
For any A E GL(JI,R), define RA : GL(!r, R) + GL(n, R) by RA(B) = BA.
Show tliat RA is a dfleomorphism, aiid ihat $ o RA = @ for ali A E O(n). By
applying tlie chain rule, show that for A E O(n) the matrix
a@ij
lias the same rank as (-(1))
.
axkl
(Here xk' are the coordinate functions in WnL, and @'j the n(n+ 1)/2 component
functions of @.) Conclude from Proposition 12 that O(n) is a submanifold of
GL(n, R).
(e) Using the formula
62 show that \ O otherwise. Show that the rank of this matrix is n(n
62
show that
\ O
otherwise.
Show that the rank of this matrix is n(n + 1)/2 at I (and hence at A for al1
A E O(n).) Conclude that O(n) has dimension n (n - 1)/2.
(fj Show that det A - &1 for al1 A E O@), The group O(n) nSL(n, R) is called
the special orthogonal group SO@),or the rotation pup R(n).
34. Let M(nr, 11) denote the set of al1 n.i x n matrices, and M(n2,n; k) the set
of al1 n? x n matrices of rank k.
(a) For every Xo E M(m,n; k) there are permutation matrices P and Q such
that
PXoQ =
Bo) ,
where A. is k x k and non-singular.
co
Do
(b) There is some E > O SUC~that A is non-singular whenever al1 entries of
A - A.
are cl E.
(4 If
AB
PXQ=(C
D)
wbere the entries of A - A.
are
< E, then
X
has rank k if and only if
D =
CA-'B. Hinl: If Ik denotes the k x k identity matrix, then
(d) M (112,n; k) c M (m,n) is a submanifold of dimension k (m + n - k) for a11
k 5 in,n.
CHAPTER 3 THE TANGENT BUNDLE A point v E R" is frequentiy pictured as an
CHAPTER 3
THE TANGENT BUNDLE
A point v E R" is frequentiy pictured as an arrow from O to V. But there are
many situations where we would like to picture this same arrow as starting
at a dXerent poini p E Rn:
For example, suppose c : EX + R" is a differentiable curve. Then cr(t) =
(cl'(t),
-,cnr(t))is just a point of R", but the line bemeen c(t) and c(t)+ cr(t)
is tangent to the curve, aiid the "velocity vector" or "tangent vector" cr(i) of
the curve c is customarily pictured as the arrow from c(i) to c(t) + cr(t).
To give tliis picture mathematical substance, we simply describe the "arrow" from p to p
To give tliis picture mathematical substance, we simply describe the "arrow"
from p to p + v by the pair (p, u). The set of al1 such pairs is just R" x R",
wliich we will also denote by T R", the "tangent space of R" "; elements of T R"
are called "tangent vectors" of R". We will often denote (p, u) E TR" by U,
("the vector v at p"); in conformjty with this notation, we will denote the set
of al1 (p,v) for v E R" by Rnp. At times, it is more convenient to denote a
member of TR" by a single letter, like v. To recover the first member of a pair
2i E TRn, we define tlie "projection" map rr : R" x IR" + R" by ~(a,b) = a.
For any tangent vector u, the point X(V)is "where it's at".
The set rr-l (p) may be pictured as al1 arrows starting at p. Alternately,
ii can be pictur-ed more geometi-ically as a particular subset of R" x R", the
oiie visualizable case occurring when 11 = l. This picture gives rise to some
tei.minology-we
call rr-'(p)
ihe fibre over p. Tliis fibre can be made into a
The Tazgeni Bundle vector space in an obvious way: we define (p, u) (p, w)
The Tazgeni Bundle
vector space in an obvious way: we define
(p, u)
(p, w) = (PY u + w)
a (p, u) = (p,a
u)-
(The operations $ and * should really be thought of as defined on
()x~(~),and
R x
T R",
res~ectively.
,€E"
Usually we will just use ordinary + and . instead of $ and .)
If f : Rrl + Rm is a differentiable map, and p E Rn, then the linear trans-
formation Df(p): Rn + Rm may be used to produce a linear map from
Rnp + Rni (,, defined by
This map, whose appareiitly anomalous features will soon be justified, is de-
noted by S,,; tlle symbol denotes tlie map f* : TR" + TRm which is the
union of al1 f*,. Sincc j;,(u) is defined to be a vector E Rmftpl, the follow-
iiig diagram "commutes" (the two possible compositions from TR" to Rm are
equal),
f*
TR" -
TRm
J
IR" -
Rm
Tlius, $, has the map f, as well as al1 maps Df(p), built into it.
This is noi the only reason for defining f* in this particular way, however.
Suppose that g: Rm + IRk is another dzerentiable function, so that, by the
chain rule,
By our definition,
This looks horribly complicated, but, using (11, it can be written
thus we have g*of* = (g of)*. Thjs relation would clearly fall apart completely if
thus we have
g*of* = (g of)*.
Thjs relation would clearly fall apart completely if f*(vp) were not in RT(pi;
with our present definition of f*, it is merely an eiegant restatement of the
chain rule.
Henceforth, we will state almost al1 concepts about Jacobian matrices, like
rank or singularity, in terms of f,, rather than Df. The "tangent vector" of a
curve c: EX + R" can be defined in terms of this concept, also. The tangent
vector of c at t may be defined as
[If c happens to be of the form
then
tllis \lector lies along the tangent line to tlie graph of f at (1, f (l)).] Notice
thai the tangeni vector of c at 1 is the same as
c*(l*)= [Dc(t)(UIc(t) = (cl/(t), .-,cns(l))c(*),
wliere lt = (1,I) is iIie "unit" tangent vector of R at 1.
If g: Rn + EXm is differentiable, then g o c is a curve in Rm. The tangent
Th 7angent Bundle vector of g o c at 1 is (g O c)*(lr)= g*(c*(lr))
Th 7angent Bundle
vector of g o c at 1 is
(g O c)*(lr)= g*(c*(lr))
= g,(tangent vector of c at t).
Consider now an n-dimensional manifold M and an imbedding i : M + WN.
Suppose we take a coordinate system (x, U) around p. Then i o x-' is a map
from EXn to WN with rank n. Consequently, (i o x-'), (W\(,)) is an n-dimen-
sional subspace of WNi(,). This subspace doesn't depend on the coordinate
system x., for if y is another coordinate system, then
(i o y-'),
= (i O x-'
ox O y-1),
= (i o x-'),o
(X O y-1)*
and
(x 0 y-')*y(,) :
Wnx(P)
+
is an isomorphism (with inverse (y o x-'),~(,)).
There is another way to see this, which justifies the picture we have drawn. If
There is another way to see this, which justifies the picture we have drawn.
If c: (-&,E) + W" is a curve with c(0) = x(p), then a = i o x-' o c is a
curve in EXN which lics in i(M), and every dxerentiable curve in i(M) is of this
form (Proof?). Now
ar(lO) = (i O x-')*
o lo),
so the tangent vector of every a is in (i ox-'), (EXn,(,)). Moreover, every vector
in this subspace is the tangent vector of some a, since every vector in EXn,(,) is
tlie tangent vector of some curve c. Thus, our n-dimensional subspace is just
the set of al1 tangent vectors at i(p) to dxerentiable curves in i(M). We will
denote this n-dimensional subspace by (M, i),.
We now want to look at the (disjointj union
We can define a "projection" map
As in the case of TWn, each "fibre" x-'(p) has a vector space structure also.
Beyond this we have to look a little more carefully at some specific examples.
Coiicider first the maiiifold M = S' and tlie inclusion i: S' + EX2.
curve c(O) = (cos @,sinO) passes through every point of S', and
Tlie
cr(0) = (-
sin O, cos O) # O.
For eacli p =
(cos O, sin O) E S', let up = (-
sin O, cos O),
(it clearly doesn't
matter wliich of the infinitely many possible 0's we choose). Then (S', i),
con-
The Ta?gel2tBundle 69 sists of al1 multiples of the vector u,,. We can therefore define
The Ta?gel2tBundle
69
sists of al1 multiples of the vector u,,.
We can therefore define a homeomorphism
h : T(S1, i) + S' x IR1 by h (hu,,) = (p, A), which makes the following dia-
gram commute.
If we define tlie "Abres" of n' to be the sets n'-' (p), then each fibre has a vector
space structure in a natural way. Commutativity of the diagram means that $i
takes fibres into fibres; clearly f] restricted to a fibre is a linear isomorphism
onto the image.
Now consider the manifold M = s2and tlie inclusion i: s2c R3. In thjs
case tliere is «o map f2 : T(s~,i) + s2x R2 with the properties of the map h.
If there were, then, for a fixed vector u # O in R2, the set of vectors
would be a collection of non-zero tangent vectors, one at each point of s2,
which varied coiiiiriuously. It is a well-known (hard) theorem of topology that
this is impossible (you can't comb the hair on a sphere).
There is another example where we can prove that no appropriate homeo- morphism T(M,i) -t
There is another example where we can prove that no appropriate homeo-
morphism T(M,i) -t M x EX2 exists, without appealing to a hard theorem of
topology. The rnap i will just be the inclusion M +- where M is a Mobius
strip, to be precise, the particular subset of IR3 defined in Chapter 1-M is the
image of the map f : [O, 2x1 x (- 1,l) + R3 defined by
At each point p = (2cos 8,2 sin 8, O) of M, the vector
u,, = (-2 sin 8,2 cos 0, O)p -- f*((lf0)(e,0))
is a tangent vector. The same is true for al1 multiples of f,((O, I)(~,~)),shown
as dashed arrows in the picture. Notice that
while
This means tliat we can never pick non-zero dashed vectors continuowEy on the
set of al1 points (2cos 8,2 sin @,O); If we could, then each vector would be
for some continuous function A: [0,2rr] + R. This function would have to be
non-zero everywhere and also satisfy h(2rr) == -h(O), which it can't (by an easy
theorem of topology). The impossibility of choosing non-zero dashed vectors
contiiiuously clearly sliows that there is no way to map T(M,i), fibre by fibre,
The 7azge~Bundle 71. homeomorphically onto M x IR2. We thus have another case where T
The 7azge~Bundle
71.
homeomorphically onto M x IR2. We thus have another case where T (M, i)
does not "look like" a product M x Rn.
For any imbedding i ; M + RN , however, the stmc ture of T (M, i) is always
simple locab: if (x,U) is a coordinate system on M, then n-'(U), the part
of T(M,i) over U, can always be mapped, fibre by fibre, homeomorphically
onto U x Rn. In fact, for each p E U, the fibre
(M,i),
equals
(i o x-')*~(,)(WnX(,)) = mp (Rnx(,)) ,
where the abbreviation 172,
define
lias been introduced temporarily; we can therefore
/: n-'(U) + U x R"
by
/ (m, (U,(,)))
= (p,u)-
In standard jargon, T(M, i) is "locally trivial". This additional feature qualifies
T(M,i) to be included among an extremely important class of structures:
An ~r-dimensionalvector bundle (or n-plane bundle) is a five-tuple
where
(1) E and 3 are spaces (tlie "total space" and "base space" of 6,
respectively),
(2) n ; E + B is a continuous map onlo B,
(3) $ and O are maps
with $ (n-'(p) x n-'(p)) c n-'(p) and @(W x n-'(p)) c
n-' (p), which make each fibre ir-' (p) into an n-dimensional
vector space over IR,
such that the following "local triviality" condition is satisfied:
Foi. each p E B, there is a neighborhood U of p and a homeomor-
phism t : n -' (U) + U x 8" which is a vec tor space isomorphism
from each n-'(q) onto q x Rn, for al1 q E U.
Because tliis local triviality condition really is a local condition, each bundle 6 = (E,n,
Because tliis local triviality condition really is a local condition, each bundle
6 = (E,n, B, $,O) automatically gives rise to a bundle 6lA over any subset
A c B; to be precise,
Notation as cumbersome as al1 tliis invites abuse, and we shall usually refer
simply to a bundle rr ; E + B, or aren denote the bundle by E alone. For
vectors u, w E n-' (p) and a E R, we wiil denote $(u, w) and @(a, v) by v + w,
and a .v or av, respectively.
The simplest example of an n-plane bundle is just X x Rn with x : X x R" +
X the projection on tlie first factor, arid tlie obvious vector space structure on
each fibre. This is called the trivial H-plane bundle over X and will be denoted
by E"(X). Tlie "tangent bundle" TR" is just E"(R").
The bundle T(S' , i) considered before is equivalent to E' (S'). Equivalente
is here a tecl~iiicalte1.m: Two vector bundles 61 = XI : El + B and 62 =
n2 : E2 + B are equivalent (tih., C2) if there is a homeomorphism h : El + E2
which takes each fibre ni-' (p) isomorphically onto n2-' (p). The map h is
called an equivalence. A bundle cquivalent to E" (B) is called trivial. (The
local triviality condition for a bundle 6 just says that ,$[Uis trivial for some
neighborhood U of p.)
The bundles T(s~,i) aiid T(M, i) are not trivial, but tliere is an even simpler
example of a non-trivial buiidle, The Mobius strip itsey (not T(M, i)) can be
i.,ir~sideredas a 1-dimeiisional vector bundle over S', for M can be obtained
1iom [O, I] x R by identifying (0,a) with (1, -a), wliile S' can be obtained from
The Tangen.?Bundle 73 [O, 11 by identifying O with 1; the map rr is defined
The Tangen.?Bundle
73
[O, 11 by identifying O with 1; the map rr is defined by x(1.a) = t for O cr r cr 1
aild ~(((0,a), (1, -a))) = {O, 1). The diagram above illustrates local triviality
iiear the point {O, 1) of S'. SupPose that S: S' + M is a continuous function
with rr o S = identity of M (such a function is called a section). Such a map
corresponds to a continuous function S : [O, 11 + R with S(0) = -S(l). Since S
must be O soinewliere, the section s must be O somewliere (that is, S(@)E rr-l(8)
must be the O vector for some 0 E S'1. This surely sliows that M is not a trivial
bundle.
An equivalente is obviously the analogue of an isomorphism. The analogue
of a homomorphism is the following.* A bundle map from c1 to c2 is a pair of
continuous maps (7,f), with f: El + E2 and f : Bl + Bz, such that
(1) the following diagram commutes
(2) /: nl-'(p) + rrz-' (f (p)) is a linear map.
Tlie pair (f., f) is a bundle map kom TRk to TR' for any dxerentiable
f: IRk + IR'. If MW c Rk and Nm c IR1 are submanifolds. i: .M + IRk and
j: N + IR1 are tlie inclusions, and the rnap f satisfies f (M) C N, then f,
* Tliere are actually severa1 possjble choices, depending o11 whether one is consider-
ing al1 buridles at once, fixed bundles over various spaces, or a fwed base space with
varyiiig buiidles. Thus f may be restricted to be an isomorphism on fibres and f to
be tlie identity or a Iion~eomorphism.The relations between some of these cases are
considered in the problems.
takes T(M,i) to T(N,j); to see this, just remember that v E T(M,i)pis the tangent
takes T(M,i) to T(N,j); to see this, just remember that v E T(M,i)pis the
tangent vector of a curve c in M, so &(u) is the tangent vector of the cum
f o c in N, and consequently f+(u) E T(N,j). In this way we obtain a bundle
map from T(M,i) to T(N,j). Actually, it would have sufficed to begin with
a Cm function f: M + N, since f can be extended to IRk locally. In fact,
this construction could be generalized much further, to the case where i and j
are merely imbeddings of two abstract manifolds M and N, and f:M + N
is Cm; we just consider the function j o f o i-' : i(M) + i(N) and extend it
locally to 3.The case which we want to examine most carefully is the simplest:
where M = N and f is the identity, while i and j are two imbeddings of M
in IRk and dR',respectively. Elements of T (M,i)pare of the form (i o x-' ).(tu)
for w E dnxnx(p),while elements of T (M,j)p are of the form (j Q x-').
(w) for
w E IRnx(,).
If we map
(i O x-'), (w) H (jo X-'), (w)
we obtain a bundle map from T (M,i)lU to T(M,j)lU, which is obviously an
equivalence. The map (M,i)p+ (M,j)p iiiduced on fibres is independent of
the coordinate system x, for if (y,V) is another coordinate system, then
We can therefore put al1 tliese maps together, and obtain an equivaEence from
T (M,í) to T (M,j). In otlier words, the dependence of T(M,i) on i is al-
most illusory; we could abbreviate T(M,i) to TM, if we agreed that TM really
denotes an equivalence class of bundles, rather than one bundle. That is the
The 7angenl Bundle 75 sort of thing an algebraist might do, and it is undoubtedly
The 7angenl Bundle
75
sort of thing an algebraist might do, and it is undoubtedly ugly. What we would
like to do is to get a single bundle for each M, in some natural way, which has
al1 the properties any one of these particular bundles T (M, i) has. Can we do
ihis? Yes, we can. When we do, TRn will be dflerent from our old defini-
tion (namely, &"(IRn)),and so will f, for f : Rn + Rm, so in stating our result
precisely we will write "old f," when necessary.
f. THEOREM. It is possible to assign to each n-manifold M an n-plane bun-
dle TM over M, and to each CbOrnap f : M + N a bundle map (f,, f), such
that:
(1)
If
1: M
+
M is the identity then 1,:
TM +
TM is the identity.
If
g : N +
P,then (g o f),
= g, o f,.
(2) There are equivalentes t" : T Rn + E" (R") such that for every Cm func-
tion f : Rn + Rm the following commutes.
(3) If U C M is an open submanifold, then TU is equivalent to (TM) 1U,
and for f: M + N the map (f IU), : TU + TN is just the restriction
of f,. More precisely, there is an equivalence TU 2 (TM)IU such that
the following diagrams commute, where i: U + M is the inclusion.*
PR00F.
The construction of TM is an iilgenious, tllough quite natural, sub-
terfuge. We will obtain a single bundle for TM, but the ehents of TM will eacli
be large equivalence classes.
* M7henusiiig the notation $*,it must be understood that the symbol 'y"realiy refers to
a triple (f, M, N) wliere f : M + N. The identitymap 1 of U to itself and the inclusion
map i: U + M have to be considered as different, since the maps 1, : TU + TU and
i, : TU + TM are certainly dfleren t (they map TU into two different sets).
The construction is much easier to understand if we first imagine that we al- ready
The construction is much easier to understand if we first imagine that we al-
ready had our bundles TM. Then if (x, U) is a coordinate system, we would
have a map x,: TU
+
T(x(U)),and this would be an equivalence (with
inverse (x-').). Since TU should be essentially (TM)IU3and T (x(U)) should
essentially be x(U) x Rn, a point e E n-'(p) would be taken by x. to some
(x(p),u). Here v is just an element of Rn (and every v would occur, since x*
maps a-'(p)isomorphically onto {p)x R"). If y is another coordinate system,
then y,(e) would be (y(p),w) for some w E Rn. 'We can easily figure out what
tlle relationship between v and w would be; since (x(p),u) is taken to (y@),w)
by y,
o x,-'
= (y o x-'),, and (y o x-'),
is supposed to be the old (y o x-').,
we would have
This condition makes perfect sense without any mention of bundles. It is the
clue wliicli enables us to now define TM.
If x and y are coordinate systems whose domains contain p, and u, w E R",
we define
(x, U) ;(y,w)
if (a) is satisfied.
It is easy to check (using the cbain rule) that - is an equivalence relation; the
equivalence class of (x,u) will be denoted by [x,u], . These equivalence classes
will be called tangent vectors at p3 and TM is defined to be the set of al1 tangent
vectors at al1 points p E M; the map n takes - equivalence classes to p. We
define a vector space stmcture on n-' (p) by the formulas
P
P
tliis definition is independent of the particular coordinate system x or y, because
D(y o x-')(x (p))is an isomorphism from RR to R".
Our definition of TM provides a one-one onto map
(b>
tx : n-' (U) -t
U x Y,
namely
[x,vJqH
(q,u).
We want tliis to be a homeomorphism, so we want tx-' (A) to be open for every
open A C U x RR3and thus we want any union of such sets to be open. There
is a metric witli exactly these sets as open sets, but it is a little ticklish to produce,
so wc leave this one part of the proof to Problem 1.
We now have a bundle n : TM
+
M.
We will denote the fibre n-'(p)
by Mp, iii conformity with the notation Rnp, tbough TMp might be better. If
The Znge~dBundle 77 J': M + N, and (x, U) and (y, V) are coordinate
The Znge~dBundle
77
J': M + N, and (x, U) and (y, V) are coordinate systems around p arid f (p),
r-espectively, we define
Of course, it must be cbecked that this definition is independent of x and y
(the chain rule again),
Condition (1) of our theorem is obvious.
To prove (2), we define tn to be tl, where I is the identity map of R" and t,
is defined in (b); it is trivial, though perhaps confusing to the novice, to prove
commutativity of the diagram.
Condition (3) is pi4acticallyobvious also. In fact, the fibre of TU over p E U
is almost exactly the same as the fibre of TM over p; the only difirence is that
each equivalence class for M contains some extra members, since in M there
are more coordinate systems around p than there are in U c M. *3
Henceforth, the bundle rr : TM + M will be called the tangent bundle of M.
If i : M + Rk is an imbedding, then TM is equivalent to T(M, i). In fact, if
(x,U) is a coordinate system around p, and I is tbe identity coordinate system
of IRk, then
the composition tnic is easily seen to be an equivalence. But T (M, i) will play
no further rolc in this story-the abstract substitute TM will always be used
instead.
Having succeeded in producing a bundle over each M, which is equivalent to
T(M, i), we next ask how fortuitous this was. Can one find other bundles with
the same properties? Tlie answer is yes, and we proceed to define two different
such bundles.
For tbe first example, we consider curves c: (-&,E) + M, each defined
on some interval around 0, witb c(0) = p. If (x, U) is a coordinate system
around p, we define
cl 7 c2
if and only if:
x ocl and x oc2, mapping IR to IR",
have the same derivative at O.
The equivalence classes, for al1 p E M, will be the elements of our new bun-
dle, T'M. For f: M + N tliere is a map S# taking the % equivalence class
P
of c to the Jz)equivalence class of f o c. Without bothering to check details,
of c to the Jz)equivalence class of f
o c. Without bothering to check details,
we can already see that this example is "redly the same" as TM -
the 7 equivdence class of x-'
o y,
[x,vJp
corresponds to:
where y is a cum in Rn with y'(0) = u;
under this correspondence, $# corresponds to f,.
In the second example, things are not so simple. We define a tangent vector
at p to be a linear operator which operates on al1 CbO functions f and which
is a "derivation at p":
We have already seen that the operators = a/axilp have this property For
these operators, clearly t (f) = l(g) if f = g in a neighborhood of p. This
condition is actually tme for any derivation t. For, suppose that f = O in a
neighborhood of p. There is a Cm function h: M + R with h(p) = 1 and
support h c f-' (O). Then
Thus, if
f
= g in a neighborhood of O, then O = e(f -g) = e(f) - l(g). If f
is defined only in a neighborhood of p, we may use this trick to define e(f):
choose h to be 1 on a neigliborhood of p, with support h c f -l (O), and define
[(S)as [(fh).
Tlie set of al1 sucli operators is a vector space, but it is not a jjriori clear what
its dimension is. This comes out of the following.
2. LEMMA.
Let f
be a CbOfunction in a convex open neighborhood U of O
in IRn, with f(0) = O. Then there are CbO functions gi : U + R with
(Tlie second condition actually follows from the first.)
PROOF.
For x
E U, let Ax (t) = f
(tx); this is defined for O 5 t 5
1, since U
is
convex. Then
Therefore we can let g (x)= J: Dif(tx) dt. 9
The Tangeni Bundle 79 3. THEOREM. The set of al1 linear derivations at p E
The Tangeni Bundle
79
3. THEOREM. The set of al1 linear derivations at p E M" is an n-dimen-
sional vector space. In fact, if (x,U) is a coordinate system around p, then
span tbis vector space, and any derivation l can be written
(so e is determined by the numbers l(xi)).
PROOE
Notice that
so l(1) = O. Hence l(c) = c l(1) = O for any constant function c on U.
Consider the case where M = R" and p = 0. Assume U is convex. Given f
on U, choose gi as in Lemma 2, for the function f - f (O). Then
n
l(f)=e(f -f(o))=e
(lidenotes the ith
coordinate function)
This sliows tliat a/aliiospan the vector space; they are clearly linearly inde-
pendent. It is a simple exercise to use the coordinate system x to transfer this
result fmm R" to M. 43
Froni Tbeorcm 3 we caii see tbat, once again, a bundle constructed from al1
derivations at al1 points of M is "really tlie same" as TM. We can let
the formula
a
P
j=]
derived in Chapter 2, shows that n a n a ayj xai -1 = bi
derived in Chapter 2, shows that
n
a
n
a
ayj
xai -1
= bi -1
ifand only d
bj =
axi
axi
i= 1
i=~
ayi p
i=l
and this is precisely the equation which says that (x,a) -, (y,b). It is easily
checked that under this correspondence, the map which corresponds to f, can
be defined as foilows:
Ef* (01(g) I= e(g o f
n
Notice that if x denotes the identity coordinate system on Rn, then
.al
-
a'
,
axl
corresponds to ap when we identi@ TRn with E" (Rn).
Wc will usually make no distinction whatsoever between a tangent vector
i=l
v E MP and the linear derivation it corresponds to, that is, between Ex, a], and
consequently, we will not hesitate to write u(f) for a differentiable function f
defined in a neighborhood of p. In fact, a tangent vector is often most easiiy
described by tclliiig what derivation it corresponds to, and the map f, is often
most easiiy analyzed from the relation
11 is customary io denote the identity coordinate system on IR'
write
by
r, and to
11 ;
ior
dt
at
2,
1,
this is a basis for Rz,,. If c : R + M is a differentiable curve, then
is called tlie tangent vector to c at lo. We will denote it by the suggestive symbol
The 7angnl Bundle 81 This symbol will be subjected to the standard abuses one finds
The 7angnl Bundle
81
This symbol will be subjected to the standard abuses one finds (unexplained)in
calculus textbooks: the symbol
-dC
will often stand for
1,
dt
dt
l
the subscript "t" now denoting a particular number t E R, as well as the identity
coordinate system.
As you might well expect, it is no accident that our second and third exarnples
turned out to be "really the same" as TM. There is a general theorem that a11
"reasonable" examples will have this property, but it is a littie delicate to state,
and quite a mes5 to prove, so it has been quarantined in an Addendum to this
chapter.
The tangent bundle TM of a Cm manifold has a little more stmcture than
an arbitrary n-plane bundle. Since TM locally looks like U x Rn, clearly TM
is itself a manifold; there is, moreover, a natural way to put a Cm stmcture
on TM. If x: U + IRn is a chart on M, then every element u E (TM)IU is
uniquely of the form
Let us denote a' by Xi(u). Then the map
H (,Y1 (H (U)),
.,xn (ir (U)),x1
.,Xn (U)) E IR2"
U
is a homeomorphism from (TM)IU to x (U) x R". This map, (x o x, X), is
simply the map x, when we identify TU with U x Rn in the standard way If
(y, V) is anotlier coordinate system, and
then, as we have already seen,
This shows that if (t,a) =
.,tn,al,
,an)
E IR2", then
Y* 0 (x*)-I (1, a)
=(yox - 1 (t), xFZlai Di(y1 0 x-l) (t)> - - - ,
ai ~i
( Y" 0 x-l) (t))-
This expression sliows that y* o (x*)-' is Cm.
\/Ve thus have a collection of Cm-related charts on TM, which can be ex- tended
\/Ve thus have a collection of Cm-related charts on TM, which can be ex-
tended to a maximal atlas.
With this Cm structure, the local trivializations x, are Cm. In general, a
vector bundle rr: E + B is called a Cm vector bundle if E and B are Cm
manifolds and there are Cm local trivializations in a neighborhood of each
point. It follows that rr : E + B is Cm.
Recall that a section of a bundle rr : E + B is a continuous function s : B + E
such that rr OS = identity of B; for Cm \lector bundles we can also speak of Cm
sections. A section of TM is called a vector field on M; for submanifolds M
of Rn, a vector field may be pictured as a continuous selection of arrows tangent
to M. The theorem that you can't comb the hair on a sphere just states that
there is no vector field on s2which is werywhere non-zero. We have shown that
there do not exist two vector fields on the Mobius strip which are everywhere
linearly independent.
Vector fields are customarily denoted by symbols like X, Y, or 2, and the
\rector X(p) is often denoted by X, (sometimes X may be used to denote a
single vector, i~isome M,). If we think of TM as the set of derivations, then for
any coordinate system (x, U), we have
n
X(P) = C~'(P)
for al1 p E U.
i=l
The functions a' are continuous or Cm if and only if X: U + TM is contin-
uous or Cm.
If X and Y are two vector fields, we define a new vector field X + Y by
Similarly, if f : M + R, we define the vector field JX
by
The 7angeni Bundle 83 Clearly X + Y and f X are Cm if X,
The 7angeni Bundle
83
Clearly X + Y and f X are Cm if X, Y, and f
are Cm. On U we can write
the symbol a/axi now denoting the vector field
If f : M + R is a Cm function, and X is a \lector field, then we can define
a newfwidion 2 (f): M + B by letting X operate on f
at each point:
It is not hard to check that if X is a Cm vector field, then f( f) is Cm for
every Cm function f; indeed, iflocally
then
which is a sum of products of Cm functions. Conversely, if y(f) is Cm for
euq Cm function f, then X is a Cm vector field (since 2 (xi)= a').
Let F denote the set of al1 Cm functions on M. We have just seen that a Cm
vector field X @ves rise to a function 9 : P + P. Clearly,
tlius 2 is a "derivation" of the ring F. Ofien, a Cm vector field X is identified
witli the derivation f. Tlie reason for this is that if A : P + 7 is any deriva-
tion, then A = 2 for a unique Cm vector field X. In fact, we clearly rnust
define
X,(f> = A(f )(P),
and the operator XP thus defined is a derivation at p.
The tangent bundle is the tme beginning of the study of dflerentiable mani- fold~,and you
The tangent bundle is the tme beginning of the study of dflerentiable mani-
fold~,and you should not read further until you grok it.* The next few chapters
constitute a detailed study of this bundle. One basic theme in all these chap-
ters is that any structure one can put on a vector space leads to a structure on
any vector bundle, in particular on the tangent bundle of a manifold. For the
present, we will discussjust one new concept about manifolds, which arises in
this very way from the notion of "orientation" in a vector space.
The non-singular linear maps f: V + V from a finite dimensional vector
space to itself fall into two groups, those with det f > 0, and those with det f < 0;
linear transformations in the first group are called orientation preserving and
the others are called dentation reversing. A simple example of the latter is
1
the map f : Rn + R" defined by f (x) = (x ,
,xn-l, -xn) (reRection in the
llyperplane xn = O). There is no way to pass continuously between these two
groups: if we identify linear maps Rn + R" with n x n matrices, and thus
with IRn2, then the orientation presewing and orien tation reversing maps are
disjoint open subsets of the set of al1 non-singular maps (those with det # 0).
The terniinology "orientation presewing" is a bit strange, since we have not yet
defined anything called "orientation", which is being preserved. The problem
becomes more acute if we want to define orientation preseMng isomorphisms
between two dflerent (but isomorphic) vector spaces V and W; this clearly
makes no sense unless we supply V and W with more structure.
To provide tliis extra structure, we note that two ordered bases
, u,)
and (vf1, . ,vfn)for V determine an isomorphism f : V + V with f(vi) = vfi;
the matrix A = (aij) of f is given by the equations
We cal1
,u,) and (vfl,
.,u',)
equally oriented if det A > O (i.e., if f is
orientation preserving) and oppositely oriented if det A < 0.
Tlle relation of being equally oriented is clearly an equivalence relation, divid-
ing tlie collection of al1 ordered bases into just two equivalence classes. Either of
iliese two equivalence classes is called an orientation for V. The class to which
,un) belongs will be denoted by
of V, then
,un) E ,u if and only if
,vn], so that if p is an orientation
.,un] = p. If ,u denotes one
*A cult word of the sixties, "grok" was coined, purportedly as a word from the Martian
language, by Robert A. Heirilein in his pop ccience fiction novel Shnger in a Slrange
Lnttd. lts sense is nicely coiiveyed by the defiriitjoii in The Anzen'can Hdhge Diclio?tag*:
"To understand profoundly through intuition or empathy".
The 7angeni Bundle Examples of equally oriented ordered bases in R, IR2, and IR3. orientation
The 7angeni Bundle
Examples of equally oriented ordered bases in R, IR2, and IR3.
orientation of V, tbe other will be denoted by -p, and the orientation [e],., .,e,]
for R" will be called the "standard orientation".
Now if (V, p) aild (W, u) are two ri-dimensional vector spaces, together with
orientations, an isomorphism f : V + W is called orientation preserving (with
respect to p and u) if [ f (vi),-.-,f(v,)] == v whenever
, u,] = p; if this
holds for any one (vi, .,u,), it clearly holds for all.
For tlie trivial bundle P(X) = X x R" we can put the "standard orientation"
, (x,e,)] on each fibre {x)x Rn. If f : en(X)+ en(X)is an equiva-
lente, aiid X is connected, tlien f is either orientation presewing or orientation
reversing on eacli fibre, for if we define the functions aij : X + R by
then det (aij): X + R is coniinuous and never O. If rr : E + B is a non-
trivial 11-plane bundle, an orientation p of E is defined to be a collection of
orientatioiis pp for rr-I (p)wllicll satisfy the following "~ompatibilit~condition"
for any open connected set U C B:
lf t : x-' (U)+ U x IRn is an equivalence, and the fibres of U x IR" are
given tlie standard orientation, tlien t is either orientation presewing
or orientation reversing on all fibres.
Notice tliat if tliis condition is satisfied for a certain t, and t' : n-' (U) + U x Rn
is another equivalence, then t' automatically satisfies the same condition, since
t'o t -' : U x 1"+ U x 1"is an equivalente. This shows that
t'o t -' : U x 1"+ U x 1"is an equivalente. This shows that the orientations pp
define an orientation of E if the compatibility condition holds for a collection
of sets U which cover B.
If a bundle E has orientation p = {,up), it has another orientation -p =
{-,up), but not every bundle has an orientation. For example, the Mobius
strip, considered as a 1-dimensional bundle over S', has no orientation. For,
although the Mobius strip has no non-zero section, we can pick two vectors
from each fibre so that the totality A looks like two sections. For example,
we can let A be [O, i] x {- 1, 1) with (O, a) identified with (1, -a); then A just
looks like the boundary of the Mobius strip obtained from [O, i] x [- 1, 11. If
we had compatible orientations p,,, we could define a section S: S' -, M by
choosing s(p) to be the unique vector s(p) E A nn-'(p) with [s(p)] = pp.
A bundle is called orientable if it has an orientation, and non-orientable oth-
erwise; an oriented bundle is just a pair (6, p) where p is an orientation for 6.
Tliis definition can be applied, in particular, to the tangent bundle TM of a
Cm manifold M. In this case, we cal1 M itself orientable or non-orientable de-
pending on whether TM is orientable or non-orientable; an orientation of TM
is also called an orientation of M, and an oriented manifold is a pair (M,p)
where p is an orientation for TM.
Tlle manifold Rn is orientable, since TW 2 E" (Rn), on which we have the
standard orientation. The sphere S"-' c Rn is also orientable. To see this we
note that for each p E S"-' the vector w = pp E en(Rn) 2
note that for each p E S"-' the vector w = pp E en(Rn) 2 TR" is nol in
.i (Sn-',,) E T Rn,, (Roblem 2 l), so for VI,
,v.-
,un-')
E
,up if and only if
(w, i,(vi),
,i,
1 E Sn-lp we can define
is in the standard
orientation of Rnp. The orientation p = {/L, : p E S"-' ) thus defined is called
the "standard orientation" of S"-'.
The iorus S' x S' is aiiotlier example of an orientable manifold. This can
be seen by noting that for any two manifolds Mi and M2 the fibre (Ml x M2)p
of T (MI x M2) can be written as VI $ V2p where (ni)*: K,,+ (Mi),, is an
isomorphism and the subspaces Vi,vary continuously (Problem 26). Since Ts'
is trivial, this shows that T (S' x S') is also trivial, and consequently orientable.
Any n-holed torus is also orientable-the proof is presented in Problem 16,
which also discusses the tangent bundle of a manifold-with-boundary.
The Mobius strip M is the simplest example of a non-orientable 2-manifold.
For tlie imbedding of M considered previously we have aiready seen that on the
subset S = ((2cos 8,2sin 8, O)) c M there are continuously varying vectors u,,,
but tliat it is impossiWe io clioose continuously from among the dashed vectors
wp = f*((O, I)(~,~))and their negatives. If we had orientations /l, for p E S,
then we could simply choose wp if [up, wp] = pp and -wp otherwise.
The projective plane IP2 must be non-oiieniable also, since it contains the
Mobius strip (for any orientable bundle 6 = rr : E +
By the restriction 6IB'
to aiiy subset B' c B is also orientable). Non-orientability of IP2 can be seen
in aiiother way, by considering the "antipodal map" A: S2 -t S2 defined by
A(p) = -p. This rnap is just the restriction of a linear map A: R3 + R3
defined by tlie same formula. The rnap A*: S2,, -t s~~(~)is just (p,~)H
A(v)), wlien S2,, is identified with a subspace of {p) x R3. The map A
is orientation reversing, so if vi = (p, ui) E SZP,the bases
(ui,u2, p)
and
(A(ui),A(u2), A(P))
are oppositely oriented. Tliis shows that if ,u is the standard orientation of S2
and [ul,vi] E ,up, then [A,v1, A,v2] E -/,LA(,,). Thus the map A: S2 + s2is
"orientation reversing" (the notion of an orientation preserving or orientation reversing map f: M +
"orientation reversing" (the notion of an orientation preserving or orientation
reversing map f: M + N makes sense for any imbedding f of one oriented
manifold into another oriented manifold of the same dimension). From this fact
it follows easily that IP2 is not orientable: If IP2 had an orientation v = {v[,l)
and g: s2 -t IP2 is the map p H [p], then we could define an orientation
{ji,) on S" by requiring g to be orientation preserving; the map A would then
be orientation preserving with respect to fi, which is impossible, since k. = p
or -p.
For projective 3-space IP3 the situation is just tlle opposite. In this case, the
antipodal map A : S' + s3ir orientation presening. If g : S3 + IP3 is the
map p H [P],we obviously can define orientations v, for IP3 by requiring g
to be orientation presewing. In general, these same arguments show that IP" is
orientable for n odd and non-orientable for n even.
There is a more "elementary" definition of orientability which does not use
the tangent bundle of M at all. According to this definition, M is orientable if
there is a subset A' of the atlas A for M such that
(1) tlle domains of al1 (x,U) E A' cover M,
(2) for al1 (x, U) and
(y, V) E A',
An orientation p of TM allows us to distinguish the subset A' as the collection
of al1 (x, U) for which x, : TM lU + T(x(U)) i x(U) x IRn is orientation
preserving (when x(U) x IR" is given the standard orientation). Condition (2)
liolds, because it is just the condition that (y o x-'),: T(x(U)) + T(x(U))
is orientation preserving. Conversely, given A' we can orient the fibres of
TM IU in such a way that x, is orientation preserving, and obtain an orientation
of TM. Although our original definition is easier to picture geometrically, the
determinant condition will be very important later on.
The 7angeniBundle ADDENDUM EQUIVALEMCE OF TANGENT BUNDLES The fact that al1 reasonable candidates for the
The 7angeniBundle
ADDENDUM
EQUIVALEMCE OF TANGENT BUNDLES
The fact that al1 reasonable candidates for the tangent bundle of M turn out
to be essentially the same is stated precisely as foilows.
4. THEOREM*.
If we have a bundle T'M over M for ea& M, and a bundle
map (h,f) for each Cm map f : M + N satisfying
(1) of Theorem 1,
(2) of Theorem 1, for certain equivalences t'",
(3) of Theorem 1, for certain equivalences T'U h: (TfM)IU,
then there are equivalences
eM : TM + T'M
such that the following diagram commutes for every Cm map f: M + N.
PROOF. The details of this proof are so horrible that you should probably skip
it (and you should definitely quit when you get bogged down); the welcome
symbol 44 occurs quite a ways on. Nevertheless, the idea behind the proof is
simple eilough. If (x, U) is a chart on M, tllen both (TM)I U and (T'M) 1U
"look like" x(U) x Rn,so there ought to be a map taking the fibres of one to
the fibres of the other. Wliat we have to hope is that our conditions on TM and
T'M make them "look alike" in a sufficiently strong way for this idea to really
work out. Those who have been through this sort of rigamarole before know
(i-e.,have faith) that it's going to work out; those for whom this sort of proof is
a new expei-ienccshould find it painful and instructive,
* Functorites will iiotice that Theorems 1 and 4 say that there is, up to natural equiv-
aleiice, a uiiique functor from the category of Cm manifolds and Cm maps to the
category of bundles and bundle maps which is naturally equi\~alentto (E", old f,) on
Euclidean spaces, and to the restriction of the functor on open submanifolds.
Let (x, U) be a coordinate system on M. Then we have the following string
Let (x, U) be a coordinate system on M. Then we have the following string
of equivalences. Two of them, which are denoted by the same symbol 2 , are
the equivalences mentioned in condition (3). Let a, denote the composition
a, = (fnlx(u))0-
0 x* 0 (-)-l.
Similarly, using equivalence 2' for TI, we can define p,.
Then
Bx-' o ax : (TM)IU -t (T'M)IU
is an equivalence, so it takes tlie fibre of TM over p isomorphically to the fibre
of TIM over p for ea& p E U. Our main task is to show that this isomorphism
between the fibres over p is independent of the coordinate system (x,U). This
will be done in three stages.
(1) Suppose V c U is open and y = xlV. We will need to name all the inclusion
maps
To compare a, and ay,consider the following diagram.
TIze 7angeniBundle 91 Each of the four squares in this diagam commutes. To see this
TIze 7angeniBundle
91
Each of the four squares in this diagam commutes. To see this for square O,
we enlarge it, as sllown below. The two triangles on the left commute by con-
dition (3) for TM, and the one on the right commutes because i o j = r.
-
Square @ commutes because k o y = x o j. Square @ commutes for the
same reason as square @; tlie inclusions x(U) + Rn and y(V) + IRn come
into play. Square (4) ob~iouslycommutes. Chasing through diagram (1) now
shows tllat the follo-ving commutes.
-
This means that for p E V, the isomorphism ay bemeen the fibres over p
is the same as a,. Clearly the same is true for B, and By, since our proof
used only properties (l), (21, and (3), not the explicit construction of TM. Thus
By-' o (Y,, = Bx-l O ax on the fibres over p, for every p E V.
(11)We now need a Lcmma which applics to both TM and TIM. Again, it will
be proved for TM (where it is actually obvious), using only properties (l), (2),
and (3),so that it is also true for T'M.
LEMMA. If A c IRn and B c IRm are open, and f : A
LEMMA. If A c IRn and B c IRm are open, and f : A + B is Cm, then the
following diagram commutes.
PROOF.
Case 1.
%e
ir a map /: Wn + Rm with 7 = f on A. Consider the
following diagram, where i : A + IRn and j : B + IRm are the incIusion maps.
Everythirlg in tliis diagram obviously commutes.
compositions
This implies tliat the two
Ev
tnlA
(TR")I A d
C
old /;
TA 2
&"(Rn)IA d
E" (IRn) -
.em(Rm)
and
are equal and this proves the Lemma in Case 1, since the maps "old T*'' and
"old f," are equal on A.
&e 2. General' cae. For each p E A, we want to show that mo maps are the
same on the fibre over p. NOWthere is a map j: IRn + IRm with j' = f on an
open set A', wl~erep E A' C A. We tben have tlie following diagram, wliere
every i comes from the fact tliat some set is an open submanifold of another,
The Gngeni Bundle and i: A' + A is the inciusion map. Boxes @, 0,and
The Gngeni Bundle
and i: A' + A is the inciusion map.
Boxes @, 0,and @ obviously commute, and commutes by &e l. To
see that square
@ (whidi has a triangle witliin it) commutes, we imbed it in
a larger diagram, in which j: A + R" is the incIusion map, and other maps
have also been named, for ease of reference.
T Rn
(u)
h/
E
TA
(TRn)IA
i*]
E
TA1
+(TP)IAr
To prove that h o i, = p o K, it suffices to prove that
since v is one-one. Thus it suffices to prove j* o i* = v o p o o, which amounts
to proving commutativi ty of the foilowing diagram.
Since j o i is just the inclusion of A' in Rn,this does commute.
Commutativity of diagram (2) shows that the composition coincides, on the subsel (TA)IA', with the
Commutativity of diagram (2) shows that the composition
coincides, on the subsel (TA)IA', with the composition
and on A' we can replace "old /," by "old f,". In other words, the two com-
positions are equal in a neighborhood of any p E A, and are thus equal, which
proves the hmma.
(111) Now suppose (x, U) and (y, V) are any two coordinate systems with p E
U n V. To prove that By-' 0 a, and Bx-' o a, induce the same isomorphism
un the fibre of TM at p, we can assume without loss of generality that U = V,
because part (1) applies to x and xl U n V, as well as to y and y1 U n V.
Assuming U = V, we have the following diagram.
ry
(3)
(TM)lU 4 TU
old (y o x-'),
a. -
(y 0 h.-1)*
O-R")ly(U)
Tlle triangle obviously commutes, and tlle rectangle commutes by part (11).
Diagram (3) thus shows tbat
Exactly the same result llolds for T':
The desired result B,-' o a, = B,-' o a, follows immediately.
Now that we have a well-defined bundle map TM + T'M (the union of al1
Bx-' 0 a,), it is dearly an equilraience eM. The proof that eni o f, = ffi o eM is
left as a masochisiic exercise for the reader. *:*
nle Zngeni Bundle PROBLEMS 1. Let M be any set, and {(xi,Ui)) a sequence ofone-one
nle Zngeni Bundle
PROBLEMS
1. Let M be any set, and {(xi,Ui)) a sequence ofone-one functions xi : Ui + Rn
with Ui c M and x (Ui) open in Rn, su& that each
is continuous.
It would seem that M ought to have a metric which makes
each Ui open and each xi a homeomorphism. Actually, this is not quite true:
(a) Let M
=
R U {s), where s
6 R.
Let
U]
= R
and xl: Ul
+
R be the
identity, and let U2 = R - {O) U {s), with x2: U2 + R defined by
Show that there is no metric on M of the required sort, by showing that every
neighborhood of O would have to intersect every neighborhood of s. Never-
theless, we can find on M a pseudometric p (a function p: M x M + R with
a11 properties for a metric except that p (p,q) may be O for p S;q) such that p
is a metric on each Ui and each xj is a homeomorphism:
(b) If A C IRn is open, then there is a sequence Al, A2, A j,
of open subsets
of A such that every open subset of A is a union of certain Ai's.
(c) There is a sequence of continuous functions 1;.: A + [O, l], with support J
c A, which "separates points and closed sets": if C is closed and p E A - C,
then there is some j;: with fi(p) 6 (A n C). Hint: First arrange in a sequence
all pairs (Ai, Aj) of part (b) with
C Al.
(d) Let Alj, j =
be such a sequence for each open set xi (Ui)- Define
gi,j : M + [o, 11 by
Arrarlge a11 gil] in a single sequence Gl, G2,
on R, and define p on M by
, let d be a bounded metric
Show that p is the required pseudometric.
(e) Suppose that for every p,q E M there is a Ui and Uj wit.h p E Ui and
q
E
Uj and open sets Bi
c xi(Ui) and Bj
c xj(Uj) SO that p
E xi-'(Bi),
q
E xj-'(l?j), and xi-'(Bi) n xj-'(Bj) = 0. Sliow that p is actually a metric
on M.
Tle TangeniBundle 97 9. (a) Show that the correspondence between TM and equivalence classes of
Tle TangeniBundle
97
9. (a) Show that the correspondence between TM and equivalence classes of
curves under which [x,u], corresponds to the 7 equivalence clau of xdl o y,
for y a curve in R" with y'(0) = u, makes f, correspond to $#.
(b) Show that under the correspondence [x,a],n Ciaia/axilp,the map f,
can be defined by
Ef*re)l(g)= [(g
f).
10. If V is a finite dimensional vector space over R, define a Cm structure on V
and a homeomorphism from V x V to T V which is independent of choice of
bases. As in the case of R", for u, w E V we will denote by u, E Vw the vector
corresponding to (w,u).
11. If g ; R +
R is Cm show that
for some Cm function h : R + R.
12, (a) Let F, be the set of al Cm functions f : M + R with f (p) = 0, and
1et e: F,
+
R be a linear operator with e(fg) = O for al1 f,g
E
F,.
Show
that l has a unique extension to a derivation.
(b)
F,.
Let W be the vector subspace of 3, generated by al1 products fg for f, g E
Show that the \rector space of al1 derivations at p is isomorphic to the dual
space (.7;p/ W)*.
(c) Since (F,/ W)* has dimension n = dimension of M, the same must be true
of F,/ W. If x is a coordinate system with x(p) = O, show that x1+ W,
x" + W is a basis for F,/ W (use Lemma 2). The situation is quite different for
,
C1 functions, as the next problem shows.
13, (a) Let V be the \lector space ofall C1 functions f : R + R with f(0) = 0,
and let W be the subspace generated by ail products. Show that lim f (x)/x2
exists for al1 f E W.
x-o
(b) Foro < E < 1, jet
Show that al1 $, are in V, and that they rcpresent linearly independent elements
of V/ W.
(c) Conclude that (V/ W)* has dimension cC= 2'.
14, If
f
; M
+
N aild f,
is the O map on each fibre, then f
is constant on
each component of M.
15, (a) A map f: M + N is an immersion if and only if
15, (a) A map f: M + N is an immersion if and only if f* is one-one on
each fibre of TM. More generally, the rank of $ at p E M is the rank of the
linear transformation f, : Mp + Nf (pl.
@) If f o g = f, where g is a difkomorphism, then the rank of f o g at a
equals the rank of f at g(a). (Compare with Problem 2-33(d).)
16. (a) If M is a manifold-with-boundary, the tangent bundle TM is defined
exactly as for M; elements of Mp are y equivalente classes of pain (x,u).
Although x takes a neighborhood of p E aM onto Wn, rather than R", the
ueciors v still run through IR", so Mp still has tangeilt vectors "pointing in ail
directions". If p E aM and x: U + W" is a coordinate system around p, then
x,-'
(Rn-',(,))
c M, is a subspace. Show that tliis subspace does not depend
on the choice of x; in fact, it is i,(aM),,, where i: aM + M is the inclusion.
(b) Let a E IRn-' x {O) c Wn. A tangent vector in W", is said to point "in-
ward" if, under the identification of TWn with &"(Mn), the vector is (a,u) where
un > O. A vector v E Mp which is not in i,(aM)p is said to point "inward" if
x*(u) E Wnqp1 points inward. Show that this definition does not depend on
the coordinate system x.
(c) Show tllat if M has an orientation p, then aM has a unique orientation
ap such that
- , = (ap),, if and only if
,i,~,-~]=
pp for
every outward pointing w E Mp.
(d) If p is the usual orientation of W",show that a,u is (- l)n times the usual
orientation of IRn-' = aWn. (The reason for this choice will become clear in
Chapter 8.)
(e) Suppose we are in the setup of Problem 2-14,
Define g: 8M x
[O, 1)
+
aN x [O, 1) by g(p, t) = ($(p), 1). Show that TP is obtained from TM U TN
Tlze Zqmi Bundle (f) If M and N have orientations p and u and f:
Tlze Zqmi Bundle
(f) If M and N have orientations p and u and f: (aM,ap) + (aN,au) is
orientation-reuer~rg,show that P has an orientation which agrees with p and u
on M c P and N c P.
(g) Suppose M is s2 with two holes cut out, and N is [O, 11 x S'. Let f be
a dfiomorphism from M to N which is orientation preserving on one copy
of S' aiid orientation reversing on the other. Wliat is the resulting manifold P?
17. Sliow that TP~is homeomorphic to the space obtained from T(s~,~)by
identifying (p,u) E (s2,i), with (-p, -u)
E (s2,i)-,.
18. Although there is no everywhere non-zero \rector field on s2,there is one
on s2- {(O, 0, l)), which is dfiomorphic to IR2. Show that such a vector field
can be picked so that near (O, 0,l) the vector field looks like the following picture
(a "magnetic dipole"):
19. Suppose we have a "multiplication" map (a,b) H a - b from IR" x R" to R"
that inakes R" into a (non-associative)division algebra. That is,
(al + a2)
b = al
b + a2
b
a (bi +b2) = a .bl +a .b2
h(a - b) = (ha) b =a
(hb)
for h E R
a (l,O,.
,.,O)
= a
and there are no zero divisors:
a,b#O
ab#O.
(For example, for 12 = 1, we can use ordinary multiplication, and for n =
(For example, for 12 = 1, we can use ordinary multiplication, and for n = 2
we can use "complex multiplication", (a, b) (c, d) = (ac - bd, ad + bc).) Let
,e,
be the standard basis of Rn,
(a)
Every point in S"-' is a - el for a unique a E Rn.
@)
If a # O, then a
el,
, ,a
en are linearl~independent.
(c)
If p = a + el E S"-', then the projection of a -
,a
e, on (Sn-', i),
are linearly independent.
(d)
Mu~tiplicationby a is continuous,
(e)
TSn-'
is trivial.
(f) TP"-' is trivial.
The tangent bundles TS3 and TS' are both trivial. Multiplications with
the required properties on R4 and R8 are provided by the "quaternions" and
"Cayley numbers", respectively; the quaternions are not commutative and the
Cayley numbers are not even associative, It is a classical theorem that the
reals, complexes, and quaternions are the only associative examples. For a
simple proof, see R. S, Palais, TLv Classi;cation oJReal DiuisiDn Akebrdi, Amer,
Math. Monthly 75 (1968), 366-368. J. E Adams has proved, using methods of
algebraic topology, that n = 1, 2, 4, or 8.
pncidentally, non-existence of zero divisors immediately implies that for a # O
there is some b with ab =
,O)
and b' with b'a
=
,O).
If the
multiplication is associative it follows easily that b = b', so that we always have
multiplicative inverses. Conversely, tliis condition implies that there are no zero
divisors if the multiplication is associative; otherwise it suffices to assume the
existence of a unique b with a b = b a = (1, O,, .,,O).]
20. (a) Consider the space obtained from [O, 11 x Rfl by identifying (0, u) with
(1, Tu), where T : R" + Rn is a vector space isomorphism. Show that this can
be made into the total space of a vector bundle over S' (a generalized Mobius
strip),
@) S11ow that the resulting bundle is orientable if and only if T is orientation
preserving.
21. Show that for p E s2, the \rector pp E R3, is not in i.(s2,)
by showing that
the inner product (p, c'(0)) = O for al1 curves c with c(0) = p and Ic(l)l = 1
for al1 t. (Recall that
wliere denotes the traiispose; see Cahb on Man@Idr, pg. 23-1
22, Let M be a Cm manifold, Suppose that (TM)I A is trivial whenever A c M
is homeomorphic to S'.
Show that M is orientable.
Hinl:
An arc c from
The hgmt Bundle 101 po E M to p E M is contained in some
The hgmt Bundle
101
po E M to p E M is contained in some cuch A so (TM)lc is trivial. Thus one
can ('transportJithe orientation of Mpo to Mp. It must be checked that this is
independent of the choice of c. First consider pairs c, c' which meet only at p0
and p, The general, possibly quite messy, case can be treated by breaking up c
into smd pieces contained in coordinate neighborhoods,
Remark: Using results from the Addendum to Chapter 9, together with Prob-
lem 29, we can conclude that a neighborhood of some S' c M is non-orientable
if M is non-orientable,
The next two problems deal with important constructions associated with
vector bundles.
- x : E
23.
+
X
is a bundle and f
: Y
+
X is a continu-
ous map.
(a) Suppose 6
Let E'
c Y x E be the set of al1 (y,e) with f(y) = x(e), define
n': E' + Y by xr(y,e)= y, and define f:E' + E by j(y,e) = e. A vector
space structure can be defined on
by using the vector space structure on x-'( f (y)). Show that x': E' + Y is a
bundle, aiid (j,f) a bundle map which is an isomorphism on each fibre. This
bundle is denoted by f *(C), and is called the bundle induced (from 6) by f.
(b) Suppose we have anotlier bundle 6" = n": E" + Y and a bundle map
-
(Y,f) from 6" to 6 wliich is an isomorphism on each fibre. Show that 6" 2
-
'
= f *) Hid: Map e E E" to (+"'e), ?(e)) E E'.
(c)
If g:
y, then (f og)*(6)
8*(f*(6)).
+
(d)
If A C X and i : A + X is the inclusion map, then i*(6) 2 61A.
(e)
If 6 is orientable, then f *(e)is also orientable.
(f) Give an example where 6 is non-orientable, but f *(6)is orientable.
(g) Let e = n : E + B be a vector bundle, Since n : E + 3 is a continuous
map from a space to the base space B of 6, the symbol n*(() makes sense,
SIiow that if 6 is not orientable, then x*(6)is not orientable.
24, (a) Given an n-plane bundle 6 = x : E + B and an m-plane bundle 7 =
rr': E' + 3, 1et E" C E x E' be the set of al1 pairs (e,ef) with x(e) = x1(e').
Let x"(e,el) = x(e) = nl(e'). Show that n": E" + B is an (n + m)-plane
I~undle.It js called the Whitncy sum 6 $ 7 of 6 and 7; the fibre of 6 $7 over p
is the direct sum n-' (p)$ n/-' (p).
(b) If f: Y + 3, show that f4(6$ 7) 2r S*(.$)$ f4(7).
(c) Given bundles ti = ni: Ei + Bi, define X: El x E2 +
(c) Given bundles ti = ni: Ei +
Bi, define X: El x
E2 +
B1 x
B2 by
rr(el,e2) = (rr1(ei),rr2(e2)). Show that this is a bundle
x c2 over Bl x
B2.
(d) If A: B + B x B is the "diagonal map", A(x) = (x,x), show that e$z~EZ
A*(6
ZJ).
(e) If 6 and q are orientable, show that e $ z~ is orientable.
(f) If 6 is orientable, and ZJ is non-orientable, show that $ ZJ is also non-
orientable.
(g) Define a "natural" orientation on V $ V for any vector space V, and use
this to show that 6 $ 6 is always orientable.
(11) If X is a "figure eight" (c.L Problem 5), find two non-orientable 1-plane
bundles 6 and q over X such that 6 $ q js also non-orientable.
25, (a) If rr r E + M is a Cm vector bundle, then rr, has maximal rank at
ixch point, and each fibre n-' (p) is a Cm submanifold of E.
(b) The O-section of E is a submanifold, carried dfiomorphically onto B by rr.
26. (a) If M and N are Cm manifolds, and li~[or rrN] : M x N + M [or N]
is the projection 017 M [or N], then T(M x N) h. XM*(TM)$ rrhr*(TN).
(b) If M and N are orientable, then M x N is orientable.
(c) If M x N is orientable, then both M and N are orientable,
27. Show that the Jacobian matrix of y, o (x.)-'
is of the form
Tliis shows that the manifold TM is alz~rqsork~ttabk,i,e., the bundle T(TM) is
orientable. (Here is a more coilceptual formulation: for v E TM, the orientation
for (TM), can be defined as
thc form of y, o (x,)-'
shows that this 01-ientation is iildependeni of the choice
of A-.) A differeilt proof that TM is orieiltable is $ven in Problem 29.
28, (a) Let (x, U) be a coordinate system on M with x(p) = O and let v E Mp
he Cy=la' a/axi lp . Consider the curve c in TM defined by
Tlze 7ageni Bundle Show that (b) Find a curve whose tangent vector at O is
Tlze 7ageni Bundle
Show that
(b) Find a curve whose tangent vector at O is a/a(xi o rr)I,.
29.
This problem requires some familiarity with the notion of exact sequences
-
f
2
(c.f. Chapter 11). A sequence ofbundle maps El -+ E2 -+ E3 With $ = g =
identity of B is exact if at each fibre it is exact as a sequence of vector space
maps.
(a)
If
,$
=
rr : E
+
B is a CbO vector bundle, show that there is an exact
sequence
O + rr*(.$) + TE + rr*(TB) + O.
fit:
(1) An element of the total space of ~"(6)is a pair of points in the same
fibre, which determines a tangent vector of the fibre. (2) Map X E (TE)e to
(e,GX).
(b)
If O + El + Ez + E3 + O is exact, then each bundle Ei is orientable if
the other two are.
(c)
T(TM) is always orientable,
(d)
If rr r E + M is not orientable, then the manifold E is not orientable. (This
is why the proof that the Mobius strip is a non-oriei-itable manifold is so similar
to the proof that the Mobius bundle over S' is not orientable.)
The ilext two Problems contain more information about the groups intro-
duced in Problem 2-33. 111addition to being used in Problem 32, this informa-
tion wiH al1 be important in Chapter 10,
30.
(a) Let po E S"-' be the point
,O,
1). For n 3 2 define / : SO(n) +
S"-'
by /(A) = A(po). Show that / is continuous and open.
Show that
/-' (po) is homeomorphic to SO(n - I), and then show that j-'(p) is home-
omorphic to SO(n - 1) for ali p E S"-'.
(b)
SO(1) is a point, so it is connected. Using part (a), and induction on n,
prove that SO(n) is connected for ali n > 1.
(c)
Show that O(tz) has exactly two components.
31, (a) If T: IRn + IR" is a linear transformation, T*: IRn + Rn, the adjoint
of T, is defined by (T*v, w) = (u, Tw) (for each u, the map w H (u, Tw) is
linear, so it is w H
(T*v, w) for a unique T*v). If A is the matrix of T with
respect to the usual basis, show that the matrh of T* is the transpose A'.
(b) A linear transformation T: R" + Rn is self-adjoint if T = T*, so
(b) A linear transformation T: R" + Rn is self-adjoint if T = T*, so that
(Tu, w) = (u, Tw) for al u, w E R". If A is the matrix of T dth respect to the
standard basis, then T is self-adjoint if and only if A is symmetric, A~ = A. It
is a standard theorem that a symmetric A can be written as CDC-1 for some
diagonal matiix D (for an analy& proof, see CaEculw on Man@ldr, pg. 122).
Show that C can be chosen orthogonal, by showing that eigenvectors for distinc t
eigenvalues are orthogonal.
(c) A self-adjoint T (or the corresponding symmetric A) is called positive semi-
ddnite if (Tu, u) > O for al1 u E Rn, and positive definite if (Tu, u) > O for al1
v # O. Show that a positive definite A is non-singular. Hznt: Use the Schwarz
inequality.
(d)
Show that A' - A is always positive semi-definite.
(e)
Sliow that a positive semi-definite A can be written as A = B2 for some B.
(Remember that A is symmetrjc.)
(f) Show tliat every A E GL(n,R) can be written uniquely as A = Al .A2 where
Ai E O(n) and A2 is positive definite. Hht: Consider A' . A, and use part (e).
(g) The matrices Al and A2 are continuous functions of A. Hint: If A(") + A
and A(") = A(")]
~("12,then some subsequence of (A(")~)converges.
) GL(i1, R) is homeomorpliic to O(n) x R"("("*')/~and has exactly two com-
ponents, (A : det A > 0) and (A : det A < O). (Notice that this also gives us
anotller way of fiilding the dimension of O(n).)
32, Two continuous fuilctioils So,5 : X + Y are called homotopic if there is
a coiltjiluous function H: X x [O, i] + Y such that
The functions H; : X + Y defined by Hl (x) = H(x, t) may be thought of as a
path of fuilctions fi-om Ho = fa to HI = J. The map H is called a homotopy
between fo and J.
The i-iotation f: (X, A) + (Y, B), for A c X and B c Y, means that
f : X + Y and f (A) c B. We cal1 fl~,J : (X, A) + (Y, 3) homotopic (as maps
from (X, A) to (Y, B)) if there is an H as above such that each Hr : (X, A) +
(Y, B)-
(a) If A : [O, 11 + GL(n, R) is continuous aild H : R" x [O, 11 + R" is defiiled by
H(x, 1) = A(r)(x),show that H is continuous, so tllat Ijg and Hl are homotopic
as nlaps from (R" , R" - (O)) lo (Rn,R" - (O)). Conclude that a non-singular
linear trailsformation T: (R", R" - (0)) + (R", R" - (O)) with det T > O is
homotopic to ihe identity map.
(b) Suppose f : R" + R" is Cm and f(0) = 0, while f (R" - (0)) c R" - (O).
If Df(0) is non-singular, sl-iow that f : (R", Rn - (O)) + (R", R" - (O)) is
homotopic to Df (O): (Rn,Rn - (0)) + (Rn,Rn - (O)). Hint; Define H(x,I) =
homotopic to Df (O): (Rn,Rn - (0)) + (Rn,Rn - (O)). Hint; Define H(x,I) =
f (tx) for O < 1 5 1 and H(x, O) = Df (O) (x). To prove continuity at points
(x, O), use Lemma 2.
(c) Let U be a neighborhood of O E R" and f:U -4 Rn a homeomorphism
with f(O) = O. Let B, C V be the open ball with center O and radius r, and let
h : Rn + B, be the homeomorphism
then
(R",rw" - (O)) + (Rn,Rn- (O)).
f 0,z:
MTe will say that f is orientation preserving at O if f o h is homotopic to the
identity map 1 : (Rn,Rn - (0)) + (Rn,Rn - (O)). Check that this does not
depend on the choice of B, c V.
(d) For p E R", Jet Tp: R" + Rn be T'(q) = p+q. If f:U 4 V is a
l-iomeomorphism, where U, V c Rn are open, we will say that f is orientation
preserving at p if T-J(p) O f O T,, is orientation preserving at O. Show that if M
is orientable, then there is a collection C of charts ~lhosedomains cover M such
that for every (x, U) and (y, V) in e, the map y ox-' is orientation presening
at x(~)for a11 p E U n V.
(e) Noiice that the condition on y o xd' in part (d) makes sense even if y ox-'
is not djffereniiable. Thuus, if M is any (not necessarily dxerentiable) manifold,
we can define M to be orientable if there is a collection C of homeomorphisms
x: U + R" whose domains cover M, such that C satisfies the condition in
part (d), To prove that this definition agrees with the old one we need a fact
from algebraic iopology: If f : Rn + Rn is a homeomorphism with f(O) = O
and T; Rn S'> R" is
,x") = (x',
,xn-l, -xn), then precisely one
of f and T o f is orientation presen~ii-igat O. Assuming this result, show that
if M has such a collection C of l-iomeomorphisms, then for any Cm structure
on M the tailgeiit bundle TM is orientable.
33. Let M" c RN be a Cm n-dimensional submanifold. By a chord of M we
meai-i a point of IRN of the form p - q for p,q E M.
(a) Prove tliat if N > 212 + 1, then there is a vector v E sN-lsuch that
(i) no chord of M is parallel to u,
(ii) no tangent plane M,,
contains v.
Hint; Coilsider ceriain n-iaps from appropriate open subsets of M x M and
TM to S"-'.
(b) Let IRN-l c IRN be the subspace perpendicular to U, and n: IRN +
(b) Let IRN-l
c IRN be the subspace perpendicular to U, and n: IRN + IRN-l
the correspoilding projection. Show that nlM is a one-one immersion. In
particular, if M is compact, then rrl M is an imbedding.
(c) Every compact Cm rl-dimensional manifold can be imbedded in IR2"+'.
Note: This is the easy case of M'hitney's classical theorem, which gives the
same result even for non-compact manifolds (H. Mfhitney,fizermtzabk man@ldí,
Ann. of Math. 37 (1935), 645-680). Proofs may be found in Auslander and
MacKenzie, Intmduction ¿o Dzffmentiabk A4un@lds and Sternberg, Latures on fiy
fmential Geotne~tIn Munkres, Eln~tentaryDflia~tialZpology, there is a dflerent
sort of argumerlt to prove that a not-necessarily-compact n-manifold M can be
imbedded in some IRN (in fact, With N = (n+ I)~).Thal we may show that M
imbeds in IR2"+' using essentiaily the argument above, together with the exis-
tence of a proper map f : M + IR, given by Problem '2-30 (compare Guillemin
,:lid Pollack, fizmnlial Topology). A much harder result of Whitn y shows that
M" can actually be imbedded in IR2" (H. Wllitney, nie selj-intersectioonroja mooU1
u-t~ranfoldin 2n-space, Ai111. of h4ath. 45 (1944), 220-246).
CHAPTER 4 TENSORS A II the constructions on vector buildles carried out in this chapter
CHAPTER 4
TENSORS
A II the constructions on vector buildles carried out in this chapter have a
common feature. In each case, we replace each fibre x-'(p) by soirie
other vector space, and then fit al1 these new vector spaces together to form a
new vector bundle over the same base space.
Tlie simplest case arises when we replace each fibre V by its dual space V*.
Recal1 that V* deilotes the vector space of al1 linear functions A: V + R. If
f r V + W is a linear transformation, then there is a linear transformation
f * : W* + V* defined by
It is clear that if
1 v : V +
V is the ideiitity, tlien 1 v* is the identity map of V*
and if
g : U
+
V, then
(f
o g)*
= g* o f *.
These simple remarks already
sliow tliat f * is an isomorphism if f: V + W is, for (f -' o f)* = 1 v* and
(f 0 f-l)*
= 1W+.
Tlie dimensjon of V* is tlie same as that of V, fol. finite dimeilsional V. In
fact, if
, v,
is a basis for V, theii the elements
E V*, defined by
are easily cliecked to be a basis for V*. The linear fuiiction v*i depeilds on tlle
eiltire set
.,v,, not just on vi alone, aiid tlle isomorphism from V to V*
obtained by sending vi to is not indepeildeilt of the choice of basis (consider
what happens if vi is replaced by 2vl),
On the othei. hand, if v E V, we caii define u** E V** = (V*)* unambigu-
ously by
If v**(h) = O for every h E V*, then h(v) = O for al1 h E V*, which implies that
v = O. Thus the map v w v** is an isomorphism from V io
v = O. Thus the map v w v** is an isomorphism from V io V*'.
It is called
the natural isomorphism from V to V**.
(Problem 6 gives a precise meaning to the word "natural", formulated only
after the term had long been in use. Once the meaning is made precise, we can
prove that there is no natural isomorphism from V to V*.)
Now let 6 = n: E + B be any vector bundle. kt
and define the function x': E' + B to take each [n-'(p)]*to p- If U c B,
lind I : ?r-'
(U)+ U x R" is a trivialization, then we can define a function
I' : nf-'(U)+ U x (Rn)*
in the obvious way: sincc the map r restricted to a fibre,
tP : H-'(~)+ {p)x R",
is an isomorphism, it gives us an jsomorphism
(g*)-': [ir-'(p)]*+ {p}x (R")'
We can makc n':
E' +
B into a vector bundle, tl~edud bundle '6
of 6,by
requiring that al1 such ir be local trivializations. (We first pick an isomorphism
from (Rn)*to R",once and for all.)
At fii-st jt n-iigllt appear tliat e* 2 6, since eacli n-'(p) is isomorphic to
x (p. However, tliis is true merely because tl~etwo vector spaces havc tlie
same dime~ision. Tl-ie lack of a natural isomorpl-iism from V to V* prevents
us from constructing an equivalence between '6 and 6. Actually, we will see
later that in "most" cases e* Zr equivalei-it to 6;for the present, readers may
ponder tl-iis question for tliemselves. In contrast, the bundle e** = (t*)' is
nlu~oysequivalcnt to 6. We consti-uct the cquivalence by mapping the fibre V
of 6 over p to the fibre V** of t'*
oler p by the natural isomorphism. If you
tl-iink about how e* is constructed, it will appear obvious that this map is indeed
tl-iink about how e* is constructed, it will appear obvious that this map is indeed
an equivalence.
Even if 6 can be pictured geometrically (e.g., if 6 is TM), there is seldom a
geometric picture for C*. Rather, 6* operates on 6: If s is a section of 6 and o
is a section of t*,then we can define a function from B to R by
s(p) E fi (p)
P G(P)(S(P))
o(p) E x'-l (p) = ,-'
(p)".
This function wjll be denoted simply by o(s).
When this construction is applied to the tangent bundle TM of M, the re-
sulting bundle, denoted by T*M, is called the cotangent bundle of M; the fibre
of T*M over p is (M,)". Like TM, the cotangent bundle T*M is actuaiiy a
Cm
vector bundle: since two trivializatjons X* and y, of TM are Cm-related,
the
same is clearly true for x.' and y,' (in fact, y.' o (x.')-' = y, o (x,)-').
1% can thus define Cm, as wcll as continuous, sections of T*M. If w is a Cm
section of T*M and X is a Cm lector field, then w(X) js the Cm function
P l+ ~(P)(X(P)).
If
f: M
+
R is a Cm function, then a Cm section df
of T*M can be
defined by
df(p)(X)= X(f)
for X E M,.
The sectio~idf is called tlie differentd of f. Suppose, in particular, that X is
dcldt 11,,, where c(io) = p. Recall that
This means that
Adopting the elliptical notations this equation takes the nice form If (x,U) is a coordinate
Adopting the elliptical notations
this equation takes the nice form
If (x,U) is a coordinate system, tlien tlie dxi are sections of T*M over U.
Applyiiig the definition, we see tliat
Tlius dxl(p),
,dxn(p) is just tlie basis of M',
dual to tlie basis a/axl l,,
.,
a/axnIP ~f M,.
Tliis means tl~atevery section w can be expressed u~~iquelyon U as
for certaiii functio~~sw; o11 U. Tlie section w is coiitiiluous or Coa if aiid only if
the functions wi are.
We can also write
if wc define sums of sections and products of functions and scctioiis in the
olwious way ("poii~rwise"additioii ai-id multiplication).
Tlie sectioii <f must liave some sucli expressioii. In fact, we obtain a classical
formula:
1. THEOREM. If (x, U) is a coordiiiate system and f is a Cm function,
1. THEOREM.
If (x, U) is a coordiiiate system and f is a Cm function, then
tlien
Thus
Classical differential geometers (and classical aiialysts) did not liesitate to talk
about "infiiijtely small" clianges dxi of the coordinates xi ,just as Leibnitz had.
No one wanted to admit tliat tliis was nonsense, because true results were ob-
taincd wlien tliesc iiifinitely small quantities wcre divided into each other (pro-
vided one did it in tlie right way).
Eventually it was rcalized that the closest one can come to describing an
infinitely sinall cliange is to describe a directiori in wliicli tliis change is supposed
to occur, Le., a tangent vector. Since df is supposed to be the infinitesimal
changc of f uiidei- aii iiifiiiitesimal chaiige of tlic poirit, df must be a function
of this chaiige, whicli means that df should be a function on tangent vectors.
The dxi tliemselves tlien metamorpliosed into fuiictions, and it became clear
tliat tliey must be distinguislied from tlie tangent vectors a/dxi.
Oiice this realization came, it was only a matter of making new definitions,
which presenled tlie oki notation, and waitiiig for everybody to catch up. In
sliort, al1 classical iiorioiis involviiig iilfinitely small quantitics became functions
on tangent vectors, likc df, eexcept for- quotieiits of infinitely smaü quantities,
which became taiigent vectors, like dc/dr.
Looking back at tlic classical works fi-omour modern vantage point, one can
usually see that, 110 niatter liow obscui*elyexpressed, this point of view was in
some sense the one always taken by classical geometers. In fact, the differential df was
some sense the one always taken by classical geometers. In fact, the differential
df was usually introduced in the following way:
CLASSICAL FORMULATION
MODERN FORMULATION
LRt f
be a function of the x',
,x",
kt j be a function on M, and x
a
say f
= f
coordinate system (so that f
,x").
= fox
for some function f on W", namely
j= f
0.-'1.
Let xi be functions of r, say x'
Let c: W -. M
be a curve. Then
=
xi(l). Then f
becomes a function
f
o c : W +
R, where
of 1, f (t) = f
,-Y"(¡)).
f
O c(r)= j(xl o c(I),
,xn
O e(¡)).
We now have
We now have
df
"
af
(f O c)I(r)
-=EGd;.dr
n
i=I
= E D;j(x(~(t)))m (xio c)((I)
i=I
(Tlle classical notation, which
suppresses tlie curve c, is still used
by physicists, as we shall point out
once again in Chapter 7.)
"
af
= E ,(c(r))
- (xi0 c)I(r)
a~
f=I
or
"
af
d(f(c(r)))
dxi(c(r))
dr
=Ea;i(c(0).i= 1
di
Multiplying by d r gives
Consequently,
"
a.f
df = E -
dxi.
axl
i=l
1= 1
(Tl-iisequation signifies that true
results are obtained by dividing by
Since every tangent vector at c(r) is
of the form dcldr, we have
dr again, 710 mal& what ~e functions
xi(t) are. It is the closest approach
in classical analysis to the realiza-
tion of df as a function on tangent
vectors.)
"
af
df = E -
dxi.
a~i
i= 1
In preparation for our reading of Gauss and Riemann, we will continually examine tlie classical
In preparation for our reading of Gauss and Riemann, we will continually
examine tlie classical way of expressilig al1 concepts which we introduce. After
a whiie, the "translation" of classical terminology becomes only a little more
difficult than tlie translation of the German in which it was written.
Recall that if f: M + N is Cm, then there is a map f,: TM + TN;
for each p E M, we have a map f*p : Mp + Nf (p). Since fZp is a linear
transformation between t~7ovector spaces, it gives rise to a map
Strict notational propriety would dictate that this map be denoted by (f,,)*,
but everyone denotes it simply by
Notice tliat we cannot put al1 4 together to obtain a bundle map from T*N
to T*-M;in fact, tlle same q E N may be f (pi)for more than one pi E M,
and there is no reason why fWp, should equal f,p,. On the other hand, we can
do sometl~ingwith tl~ecotaiigent bundle that we could not do with the tangent
bundle. Suppose w is a section of T*N. Then we can define a section rj of T*M
as follows:
V(P)= w(f(p)) 0 f*p,
(The complex symbolism tends to hide the simple idea: to operate on a vector,
we push it over to N by f*,and then operate on it by u.) This section q is
denoted, naturally enough, by $*m. There is no corresponding way of trans-
ferring a vector field X on M wer to a vector field on N.
Despite these differences, we can say, roughly, that a map f: M + N pro-
duces a map f* going in tlie same direction on tlle tangent bundle and a map
f * going in the opposite direction on tlle cotangent bundle. Nowadays such
situations are always distinguished by calling the things which go in the same
direction "covariant" and the tliings which go in the opposite direction "con-
travariant". Classical terminology used these same words, and it just happens
to llave reversed tllis: a vector field is called a contravariant vector fieId, whiie
a section of T*M is called a covariant vector fieId. And no one has had the
gall or autllority to reverse terminology so sanctified by years of usage. So
it's very easy to remember which kii-id of vector field is covariant, and which
contravariant -itJs j ust tl~eopposite of wliat it logically ought to be.
The rationale behind the classical terminology can be seen by considering coordinate systems x on
The rationale behind the classical terminology can be seen by considering
coordinate systems x on R" which are linear transformations. In this case, if
x(vi) = el, then
1
1
x(a vi +.v.+
a%,)=
(a ,
,a"),
so he x coordinate system is just an "oblique Cartesian coordinate system".
If x' is aiiotlier such coordinate system, then x'j = =yzI
aijxi for certain aij.
Clearly aij = ax"/axi,
so
ilib can be seeii directly from die fact that he matrix (ax'J/axi) is the constant
inatrix D(xt o -Y-' ) = xt o x-' . Comparing (*) with
fi-om Tlieorem 1, we see tliat tlie differeiitials dxi "cliange in tlie same way" as
tlic coo~diiiatcsxt, lience tliey are "covai-iant". Coiisequently, any combination
is also called "covariant". Notice that if we also llave
o =
otidx",
tl~en\ve can express the wti in terms of the wt. Substituting n axi dx' =
tl~en\ve can express the wti in terms of the wt. Substituting
n
axi
dx' = E
dxtJ
j==l
into tlle first expression for w and comparing coeffrcients with the second, we
find that
1
Wj=
t-1
On the otller hand, given two expressjons
for a vector field, tlie functions atj must satisfy
Tllese expressioiis can always be remembered by iioting that indices which are
summed over always appear once "above" and once "below". (Coordinate func-
tions x',
,x" used to be denoted by xl,
,A- Tliis suggested subscripts wi
for covariaiit vector fields and superscripis o' íor coiiti.avariant vector fields. M-
ter tllis was fiimly established, the indices oli the x's were shifted upstairs again
to inake tlle summation convention work out.)
Covariar-it and contravariant vector fields, i.e., sections of T*M and TM,
respectively, are also called covariant ar-id coi1travarian t tensors (or tensor fields)
of order 1, which is a warning that worse things are to come. We begin with
some worse algebra.
If VI,
, Vm are vector spaces, a function
is limar for each choice of VI,
is clearly a vcctoi- space. If VI ,
. ,vk-l,vk+lr
.,um. The set of aii such T
, Vm = V, this vector space will be denoted
by Tm(V). Notice that 7' (V) = V*. If f : V + W is
by Tm(V). Notice that 7' (V) = V*. If f : V + W is a linear transfor-
mation, then there is a linear transformation $" : Tm( W) + Tm(V), defined
completely analogously to the case m = 1:
For T E Tk(V), and S E 7' (v) we can define the "tensor product" T @SE
T'+'(v)
by
Of course, T @ S is not S @ T. On the otl-ier lland, (S @ T) @ U = S @ (T @ U),
so we can define n-fold tensor products unambiguously; this tensor product
operation is itself multilinear, (Sl + S2) 8 T = SI 8 T + S28 T, etc. In
particular,
if VI,.
, vn is a
basis for
V and
,v*,
is the dual basis for
V* = T'(V), then the elemen ts
are easily seen to be a basis for T~(V),which ihus has dimension nk.
We can use this new algebraic construction to obtain a new bundle from any
\rector bundle 6 = ?r : E + B. We let
E' = uTk(?r-l(p)),
P@
and let
ni: E' +
3
take
T~(s-' (P)) to p.
If
U c B and
r:n-I(u)+
U xwn
is a ti+i\4alization7then the isomorphisms
ip : n-I
(p) + {p} x Rn
yield isomorpllisms
(rpL)-' : Tk(n-' (p)) + {p}x 7"~").
If we choose an isomorphism Tk(wn)+ Rnk once and for all, these maps can
bc put together to give a map
\Ve rnake kerr>:E' + B into a vector bundle T~ (6) by requiring that al1
\Ve rnake kerr>:E' + B into a vector bundle T~ (6) by requiring that al1 such 'i
1)e local trivializations. The bundle 6* is the special case k = 1.
For the case of TM, he bundle T~(TM)is called the bundle of covariant
tensors of order k, and a section is called a cwariant tensor fidd of order k. If
(x,U) is a coordinate system, so that
is a basis for (Mp)*,tlien tlie k-fold tensor products
are a basis for Tk(M,). Thus, on U every covariant tensor feld A of order k
can be written
A(P)=
Ai,
i,
(p)dxi((p)@
- . @ dxik (p),
il ,
,ik
or simply
dxi18 - - . @ dxfk,
A
=
E Ai,
wllere dxil 8 - - @ dx-'k riow denotes a section of T~(TM).If we also have
then
(tlie pi-oducts are just 01-diiiaryproducts of fuilctions). To denve this equation,
we just use equation (**) on page 114, and multilinearity of B. The section A
is continuous or Coo if and only if the functions Ai,
;,
are.
A covariant tensor feld A of order k can just be tliought of as an operation 2
on k vector fields XI,
, Xk which yields a function:
Notice that A is rnultilinear on he set V of Cm vector fields:
Moreover, because A is defined 'pointwise", fuízdions F; ie., if f is Coo,then it is
Moreover, because A is defined 'pointwise",
fuízdions F; ie., if f is Coo,then
it is actually linear ouer he Cm
for we have
\Ve are finally ready for anotlier theorem, one that is used over and over.
is liiiear over Y,theii tliel-e is a unique tensor field A with A = A.
PXOOF. Note first that if v E Mp is any tangent \lector, then there is a vector
Geld X E V witll X(p) = v. In fact, if (x, U) is a coordinate system and
then we can define
(0
outside U,
~~liereeacli af now denotes a constant function and f is a Cm function with
f (p) = 1 and support f c U.
Now if v1,
, vk E Mp are extended to vector fields Xf,
, Xk E V we
clearly must define
The problem is to prove tl-iat tl-iis is well-defined: If Xr(p) = Yi(p) for eacl-i i,
we claim that
d(x1,- -,Xk)(p) = A(Y1 3 - ., Yk)(p)*
(The map A "lives at points", to use tlie in terminology.) For simplicity, take the
(The map A "lives at points", to use tlie in terminology.) For simplicity, take the
case k = 1 (tlle general case is exactly ailalogous). The proof that A(X)(p)=
A(Y)(p)when X(p) = Y(p) is in two steps.
(1) Suppose first that X
=
Y in a neighborhood U of p.
Let
f
be a
Cm
function witli f (p) = I and support f c U. Then fX = f Y, so
evaluating at p gives
~(X)(P)= &(Y)(P).
(2) To prove tlle result, it obviously suffrces to show that A(X)(p) = O if
X(p) = O.
kt (x, U) be a coordinate system around p, so that on U we
can write
n
m
x = ~~~dwhere al1 b'(p) = O.
axi
f-l
If g is 1 i~ia iieigliboi-liood V of p, and support g c U, tlien
is a well-defined Cm \rector field on al1 of M wliich equals X on V, so that
Now
= O,
since bf(~)= 0. +3
Because of Tlieoi-em 2, we will ilevei- distinguisli between tlie tensor field A
and tlie ol~csationA, ilor will we use the symbol A any longer. Note that
Tlieorem 2 applies, in particular, to tlie case k = 1, wliere 7'(TM) = T'M,
tlie cotangent bui-idle: a fiiilctioii fi-om V + 9 wliicll is linear over 9 comes
fi-om a covariaiit vectoi- field u.Just as witli covariant vector fields, a Cm map
f : M + N @ves a map f* taking covariant tensor fields A of order k on N
to covariaiit tei-isor fields f*A of 01-der k on M:
Moreover, if A and B are covariani tensor fields of orders k and I, respectively,
Moreover, if A and B are covariani tensor fields of orders k and I, respectively,
then we can define a new covariant tensor field A @ 3 of order k + I:
.
(A @ B)(p) = A(p) @ B(p)
(operating on
Mp x
x Mp
k + I times).
Although covariant tensor fields will be our main concern, if only for the
sake of completeness we should define coiitravariant tensor fields. Recall that
a contravariant vector field is a seciion X of TM. So each Xp E Mp. Now an
element v of a vector space V can be thought of as a linear function v : V' + R;
we just define v(h) to be h(v). A contravariant tensor field of order k is just
a section A of the bundle T~(T*M);thus, each A(p) is a k-linear function
on Mp*. We could also use the notation Tk(TM),if we use Tk (V) io denote al1
k-linear functioiis on V'. In local coordinates we can write
(remember that each a/axj 1,
operates on Mp*), or simply
If we have another such expression,
tl-ien we easiiy compute that
A contravariant tensor field A of order k can be considered as an operator A
iaking k covariant vector fields wl ,
,wk
into a function:
Naturally, tliere is aii analogue of Tlieorem 2, proved exactly the same way, that
allows us to dispense with the notation A,and to identify contravariant tensor
fields of order k with operators on k covariant vector fields that are linear over
,!he Coofunciioni F.
Finally, we are ready to introduce "mixed" tensor fields. To make the intro- duction less
Finally, we are ready to introduce "mixed" tensor fields. To make the intro-
duction less painful, we consider a special case first. If V is a vector space, let
q' ( V) denote al1 bilinear functions
A
vector bundle e = a : E + B gives rise to a vector bundle q' (e),obtained by
replacing each fibre K-' (p) by q' (a-' (p)). In particular, sections of (TM)
are called tensor fields, covariant of order 1 and contravariant of order 1.
There are al1 sorts of algebraic tricks one can play with 7,' (V);although they
should be kept to a minimum, certain ones are quite important. kt End(V)
denote the vector space of al1 linear transformations T: V + V ("endomor-
phisms" of V), Notice that each S E End( V) @vesrise to a bilinear S E II;'( V),
by the formula
Moreover, the correspondence S I+ S from End(V) to q' (V) is linear and one-
one, for S = O implies that A(S(v)) = O for al1 A, which irnplies that S(v) = O,
for al1 v. Since both End(V) and (V) have dimension n2, this map is an
isomorphisrn. The inverse, however, is not so easy to describe. Given 3, for
each v the vector S(v) E V is merely determined by describing the action of a h
on it according to (*). It is not hard to check tl-iat this isomorphism of End(V)
and T1(V)makes the identity map 1: V + V in End(V) correspond to the
cc evaluation" map e: V x V*+ W in T'''(v)
given by
e(v,A) = A(v).
Generally speaking, our isomorpliism can be used to transfer any operation
from End(V) to ?''(V). In particular, given a bilinear
we can take the trace of the corresponding S: V + V; this number is called
the contraction of T. If
.,v, is a basis of V and
then we can find the matrix A = (aij) of S, defined by in terms
then we can find the matrix A = (aij) of S, defined by
in terms of the T,' ; in fact,
contraction of T =
T.:
(The term "contraction" comes from the fact that the number of indices is con-
tracted from 2 to O by setting tl-ie upper and lower indices equal and summing.)
Tl-ieseidentifications and operations can be carried out, fibre by fibre, in any
fibre bundle T,' (5). Thus, a section A of 7'' (5) can just as well be considered
as a section of tlie bundle End(h), obtained by replacing each fibre a-' (p) by
~nd(a-'(p)).In tliis case, each A(p) is an endomorphism of a-'@).
over, each section A gives rise to a function
More-
(contraction of A) : B + R
defined by
contraction of A (p)
p I+
= trace A (p)
if we consider A (p) E ~nd(n-'@))
In particular, given a tensor field A, covariant of order 1 and contravariant
of order 1, wliich is a section of (TM), we can consider each A (p) as an
endomorpllism of M,,
coordinate system
and we obtain a function "contraction of A".
If in a
tl-ien
(contraction of A) = y A:.
Tlle general notion of a mixed tensor field is a straightfonvard generalization.
Define 7jk( V) to be the set of al1 (k + !)-linear
k times
l times
Every bundle 5 gives rise to a bundle ~'(5).Sections of q(TM) are called tensor fields,
Every bundle 5 gives rise to a bundle ~'(5).Sections of q(TM) are called
tensor fields, covariant of order k and contravariant of order i, or simply of type
(f), an abbreviation that also saves everybody embarrassment about the use of
the words "covariant" and "coníravarian t". Locally, a tensor field A of type (f)
can be expressed as
and if
then
Classical differentia1 geometry books are filled with monstrosities like this
equation.
111fact, the classical definition of a tensor field is: an assignment
of nk+* functions to every coordinate system so that (*) holds between the nk+'
functions assigned to any two coordinate systems x and x'. (!) Or even, "a set
of ,rk+' functions which changes according to (*)" . Consequently, in classical
differential geonletry, al1 important tensors are actually defined by defining the
functions A;, in terms of the coordinate system x, and then checking that (*)
holds.
Here is an important example. In every classical dxerential geometry book,
one will find the following assertion: 'The Kronecker delta 6; is a tensor." In
other words, it is asserted that if one cliooses the same n2 functions 6; for each
coordinate system, then (*) holds, i.e.,
this is certainly true, for
From our point of view, what this equation shows is that a = C6; dx'
From our point of view, what this equation shows is that
a
= C6; dx' @ -
A
i,j
axj
is a certain tensor field, independent of the choice of the coordinate system x
To identify the mysterious map
we consider v E Mp and h E Mp* with the expressions
then
Thus A(p) is just tl-ie evaluation map Mp x Mp* + R; considered as an ei-ido-
morphism of M,,
it is just the identity map.
Tl-ie conti.actjon of a tensor is defined, classically, in a similar manner. Given
a tensos, Le., a collection of functions A!,
isfying
one for each coordinate system, sat-
we note that n =EA/E-- axi axiUaxj i,j a-l so tliat tliis sum is a
we note that
n
=EA/E--
axi
axiUaxj
i,j
a-l
so tliat tliis sum is a well-defined function. Tliis calculation tends to obscure
the one part which is really necessary-verification of the fact that the trace of
a linear transformation, defined as the sum of the diagonal entries of its matrix,
is indcpendent of the basis with respect to which the matrix is written.
Incideiitally, a teiisor of type (f) can be contracted with respect to any pair
of upper and lower indices. For example, the functions
a-l
"transform correctly" if the AaPY do. If we consider each A (p) E T
we are taking B(p) E qZ(Mp)to be
(M~),then
~(p)(v~,vz, Al,Az) = contraction of: (v,A) H A(p)(v, vi, vz7Al7hz, A).
Wlliie a corltravai-iant vector feld is classically a set of n functions which
"transforms in a certain way", a vector at a single point p is classically just an
assigiiment of n tiumbns u',
,an to each coordinate system x, such that the
numbers a",
,a'" assigned to x' satisfy
Tllis is preciseb the defi~iitionwe adopted when we defined tangent vectors as
equivalente classcs [A-,a],. The revolutioil iil tlle modern approach is tliat the
set of al1 vectors is made into a bundle, so tl~at\lector fields can be defined as
sections, ratller tlian as equivalente classes of sets offunctions, and that al1other
types of tensors are constructed from this bundle. The tangent bundle itselfwas almost a victim
types of tensors are constructed from this bundle. The tangent bundle itselfwas
almost a victim of the excesses of revolutionary zeal. For a long time, the party
line held tliat TM must be defined either as derivations, or as equivalence classes
of curves; the return to the old definition was influenced by the "functorial"
point of view of Theorems 3-1 and 3-4.
The modern revolt against the classical point of view has been so complete
in certain quarters that some matliematicians will give a three page proof that
avoids coordinates in preference to a three line proof that uses them. We won't
go quite that far, but we will give an "invariant" definition (one that does not
use a coordinate system) of any tensors that are defined. Unlike the "Kronecker
delta" and contractions, such invariant definitions are usually not so easy to
come by. As we shall see, invariant definitions of al1 the important tensors in
differential geometry are made by means of Theorem 2. We seldom define A(p)
directly; instead we define a function A on vector fields, which miraculously
turns out to be linear over the Coo functions Y,and hence must come from
some A. At tl~eappropriate time we will discuss wliether or not this is al1 a big
cheat.
PROBLEMS l. LRt f: M" + Nm, and suppose that (x, U) and (y, V)
PROBLEMS
l. LRt f: M"
+
Nm, and suppose that (x, U) and (y, V) are coordinate
systems around p and f (p), respectively.
(a) If g: N +
R, then
(Proposition 2-3 is tlie special case f
= identity.)
(b) Show that
and, more generally, express j;(C;,,
aia/axi1P) in terms of the a/ayjIp.
(c)
Show that
(d)
Express
in terms of the dx'.
2. If f, g: M + N are Cm, sl~owtllat
3.
LRt f: M + R be Cm. For v E Mp, show tllat
4. (a) Sllow that if the ordered bases vi,
, v, aiid wl ,
, W, for V are equaiiy
oriented, tllen tlle same is true ofthe bases v*I,
, v*, and
,u*, for V*.
(b) Show that a bundle 6 is orientable if and only if e* is orientable.
5. The following statements and problems are al1 taken from Eisenhart's clas- sical work Riemalznian
5. The following statements and problems are al1 taken from Eisenhart's clas-
sical work Riemalznian Geometgt. In each case, check them, using the classicai
methods, and then translate the problem and solution into modern terms. An
"invariant" is just a (well-defined) function. Remember that the summation
convention is always used, so hipi means x7=,hipi. Hints and answers are
given at the end, after (xiii).
(i) If the quantity hipi is an invariant and either hi or pi are the components
of an arbitrary [covariant or coiitravariant] vector field, tlie otlier sets are com-
ponents of a vector field.
(ii) If haIiare the components of n vector fields [in an n-manifold] , where i
for i =
,IZ indicates the component and a for a =
,n
the vector, and
tbese vectors are independent, that is, det(hUii)# O, then any vector-field h' is
expressible in the form
h' = ~~h,~',
where the a's are invariants.
(iii) If p; are the componeiits of a given vector-field, any vector-field hi satisfy-
iiig hipi = O is expressible linearly in terms of ,z - 1 independent vector fields
h,lr for a =
,n - 1 wliicli satisfy the equation.
(i) If uij = aji for the components of a tensor field in one coordinate system,
then U"J = arJ' for the coordinates in any other coordinate system.
(1) If uij and bi; are compoiieii ts of a teiisor field, so are aij + b*~. If aiJ aiid
bki are components of a tensor field, so are aijbki.
(vi) If uohihJ is an invariant for 1' an arbitiary vector, then uij + aji are the
componeiits of a teiisor; in particular, if uUhihj = O, then aij + aji = O.
(vii) If uijhihJ = O for al1 vectors hi sucli tliat hipi = O, wliere pi is a given
covariant vector, ir vi is defined [c.f. (iii)] by aijhUii~j= 0, a =
,n - 1 and
pivi # 0, and by definition
1
tlieii (~'j- 7pio1)eiej
= O is satisfied by every vector field Pi, and consequently
(viii) If a,, are tlie componeiits of a tensor aiid b aiid c are invariants, sl~ow
tliat if bu,, + ea,, = O, tlien either b = -c
u,, is skew-symmetric.
and u,, is symmetric, or b = c and
(ix) By delinition tlie rnlzk of a tensor of the second order a;j is tlie rank of
the matrix (utj). Show tliat the rank is invariant under al1 transformations of
coordinates.
(x) Show tliat the rank of the tensor of components aibj, where ai and bj
(x) Show tliat the rank of the tensor of components aibj, where ai and bj
are the components of two vectors, is one; show that for the syrnmetric tensor
aibj + ajbi the rank is two.
(xi) Show that the tensor equation aijhi = ahj, where a is an invariant, can
be written in the form (aij - aGij)hi = O. Show also that aij = Gija, if the
equation is to hold for an arbitrary vector hi.
(xii) If aijh< = ahj holds for al1 vectors hi SUC~that pihi = O, where ,ui is a
@ven vector, then
a
i j = 0161 + oi.pi.
if j,
= j~ for some a #
or i,
= ip for some a # fi
( O
.
.
or if {jl,
. ,jp} # {il,
. , ip}
~JI
JI>-
il
i,
- 1
if jl,
, jp is an even permutation of il,
,ip
if
jl ,
,jp is an odd permutation of ii,
,ip
(-1
then 6tA.'!'
are tlie components of a tensor in al1 coordinate systems.
I,
HINTS AND ANSWERS.
(i) w is determined if w (X) is known for al1 X, and uice versa.
(iii) Given w (with w(p) # O for al1 p], there are everywhere linearly indepen-
dent vector fields X1,
,X,-1
which span ker w at each point. (This is true
only locally. For example, on x IR ihere is an w such that ker w(p, í) consists
of vectors tangent to S2x {í}.)
(vi) For T: V x V + IR, let Tt(v, w) = T(w, u). Tlien T + T' is determined by
S(v) = T(v, u). For, T(u + w, u +w) = T(v, u) + T(u, w) +T(w, v) +T(w, w).
Similarly, T(v, u) = O for al1 v implies that T + T' = 0.
(vii) Give~iw (with w(p) # O for al1 p], clioose Y complementary to ker w at
al1 points. If o (2) = T(Y, Z), then T (2,2) = w (Z)o(Z)/w(Y) for all vector
fields 2.
(ix) T : V x V + W corresponds to T : V + V* [where T(v)(w) = T (v, w)].
The rank of T may be defined as the rank of T (coiisider the matrix of T with
respect to bases U,,
.,un and v*i,
,u",).
(xii) Let V = Mp*. If T:
V +
V and p E V* and T(v) = av for aii u E kerp,
there is a y complementary to ker ,u such that
(Begin by clloosing yo complementary to ker ,u and writing v uniquely as vo+cyo
for u0 E ker p.)
(xiii) Define -- S: vx xvxv*x--~xv*+R p times p times 6. (a) Let iv :
(xiii) Define
--
S:
vx
xvxv*x--~xv*+R
p times
p times
6. (a) Let iv : V + V** be the "natural isomorphism" iv (v)(l) = l (v). Show
that for aily linear traiisformation f : V + W, tlie following diagram com-
mutes:
v
iv
)y**
h) Show tliat there do not exist isomorphisms iv : V +
lowing diagram always commutcs.
V* such that the fol-
v-v*
2 v
Hint: There docs not even exist an isomorphism i : R + R* which makes tlie
diagram commute for al1 linear f : R + R.
7. A covariantfuílctol- from (finite dimensional) vector spaces to vector spaces is a
function F wliich assigns to every vector spacc V a vector space F ( V) and to ev-
cry linear transformation f : V + W a limar uansformation F(f) : F( V) +
F(W), sucli that F(lv) = 1~(v)and F(g 0 f) = F(g) o F(v).
(a) Tlie "identity functor", F(V) = V, F(f) = f is a functor.
(b) The "double dual functor", F ( V) = V**, F (f) = f ** is a functor.
(e) The "Tk fuiictor", F(V) = &(V) = Tk(v*),
is a functor.
(d) If
F
is any fuiictor and f
: V +
W is an isomorphisrn, tlien F(f) is