Algebras Lattices Varieties Volume 1

ALGEBRAS,
LATTICES, VARIETIES
The Wadsworth & Brooks/Cole Mathematics Series
Series Editors
Raoul H. Bott, Harvard University
David Eisenbud, Brandeis University
Hugh L. Montgomery, University of Michigan
Paul J. Sally, Jr., University of Chicago
Barry Simon, California Institute of Technology
Richard P. Stanley, Massachusetts Institute of Technology
M. Adams, V. Guillemin, Measure Theory and Probability

W . Beckner, A. Calderón, R. Fefferman, P. Jones, Conference on Harmonic Analysis in
Honor of Antoni Zygmund
G. Chartrand and L. Lesniak, Graphs and Digraphs, Second Edition
J. Cochran, Applied Mathematics: Principies, Echniques. and Applications
W . Derrick, Complex Analysis and Applications, Second Edition
J. Dieudonné, History of Algebraic Geometry
R. Durrett, Brownian Motion and Martingales in Analysis
S. Fisher, Complex Variables
A. Garsia, Topics in Almost Everywhere Convergence
R. McKenzie, G. McNulty, W. Taylor, Algebras, Lattices, Varieties, Volume I
E. Mendelson, Introduction to Mathematical Logic, Third Edition
R. Salem, Algebraic Numbers and Fourier Analysis, and L. Carleson, Selected Problems on
Exceptional Sets
R. Stanley, Enumerative Combinatorics, Volume I
K. Stromberg, An Introduction to Classical Real Analysis
-
R .- 5 2 5 cesa y
ALGEBRAS,
LATTICES, VARIETIES
'7
Ralph N. McKenzie
University of California, Berkeley
George F. McNulty
University of South Carolina
Walter F. Taylor
University of Colorado
Wadsworth & Brooks/Cole

Advanced Books & Software
Monterey, California
Wadsworth & Brooks/Cole Advanced Books & Software
A Division of Wadsworth, Inc.
O 1987 by Wadsworth, Inc., Belmont, California 94002. Al1 rights

reserved. No part of this book may be reproduced, stored in a retrieval
system, or transcribed, in any form or by any means-electronic,
mechanical, photocopying, recording or otherwise-without the prior
written permission of the publisher, Brooks/Cole Publishing Company,
Monterey, California 93940, a division of Wadsworth, Inc.
Printed in the United States of America
Library of Congress Cataloging-in-PublicationData

McKenzie, Ralph, [date]
Algebras, lattices, varieties.
Bibliography: v. 1, p.
Includes index.
l . Algebra. 2. Lattice theory . 3. ~ a r i e t k s(Universal
algebra). 1. McNulty, George F., [date].
11. Taylor, Walter F., [date]. 111. Title.
QA51.M43 1987 512 86-23239
ISBN 0-534-07651-3 (V. 1)
Sponsoring Editor: John Kimmel

Editorial Assistantf Maria Rosillo Alsadi
Production Editor: S.M. Bailey
Manuscript Editor: David Hoyt
Art Coordinator: Lisa Torri
Interior Illustration: Lori Heckelman
Typesetting: Asco Trade Typesetting Ltd., Hong Kong
Printing and Binding: The Maple-Vail Book Manufacturing Group,
York, Pennsylvania
This volume is dedicated to our parents-
Annie Laurie and Milton, Helen and George, Portia and John
Preface
This is the first of four volumes devoted to the general theory of algebras and
the closely related subject of lattice theory. This area of mathematics has grown
very rapidly in the past twenty years. Not only has the literature expanded
rapidly, but also the problems have become more sophisticated and the results
deeper. The tendency toward specialization and fragmentation accompanying
this growth has been countered by the emergence of new research themes (such
as congruence class geometry, Maltsev classification, and congruence classifica-
tion of varieties) and powerful new theories (such as general commutator theory
and tame congruence theory), giving the field a degree of unity it never had before.
Young mathematicians entering this field today are indeed fortunate, for there
are hard and interesting problems to be attacked and sophisticated tools to be
used. Even a casual reader of these volumes should gain an insight into the
present-day vigor of general algebra.
We regard an algebra as a nonempty set equipped with a system of finitary
operations. This concept is broad enough to embrace many familiar mathemat-
ical structures yet retains a concrete character. The general theory of algebras
borrows techniques and ideas from lattice theory, logic, and category theory
and derives inspiration from older, more specialized branches of algebra such as
the theories of groups, rings, and modules. The connections between lattice
theory and the general theory of algebras are particularly strong. The most
productive avenues to understanding the structure of algebras, in al1 their diver-
sit y, generally involve the study of appropriate lattices. The lattice of congruence
relations and the lattice of subalgebras of an individual algebra often contain (in
a highly distilled form) much information about the interna1 structure of the
algebra and the essential relations among its elements. In order to compare
algebras, it is very useful to group them into varieties, which are classes defined
by equations. Varieties can in turn be organized in various ways into lattices (e.g.,
the lattice of varieties, the lattice of interpretability types). The study of such
lattices reveals an extraordinarily rich structure in varieties and helps to organize
our knowledge about individual algebras and important families of algebras.
Varieties themselves are elementary classes in the sense of logic, which affords
an entry to model-theoretic ideas and techniques.
viii Preface
Volume 1 is a leisurely paced introduction to general algebra and lattice

theory. Besides the fundamental concepts and elementary results, it contains
severa1 harder (but basic) results that will be required in later volumes and a final
chapter on the beautiful topic of unique factorization. This volume is essentially
self-contained.We sometimes omit proofs, but-except in rare cases-only those
we believe the reader can easily supply with the lemmas and other materials that
are readily at hand. It is explicitly stated when a proof has been omitted for other
reasons, such as being outside the scope of the book. We believe that this volume
can be used in severa1 ways as the text for a course. The first three chapters
introduce basic concepts, giving numerous examples. They can serve as the text
for a one-semester undergraduate course in abstract algebra for honors students.
(The instructor will probably wish to supplement the text by supplying more
detail on groups and rings than we have done.) A talented graduate student of
mathematics with no prior exposure to our subject should find these chapters
easy reading. Stiff resistance will be encountered only in $2.4-the proof of the
Direct Join Decomposition Theorem for modular lattices of finite height-a
tightly reasoned argument occupying severa1 pages.
In Chapter 4, the exposition becomes more abstract and the pace somewhat
faster. Al1 the basic results of the general theory of algebras are proved in this
chapter. (There is one exception: The Homomorphism Theorem can be found in
Chapter 1.) An important nonelementary result, the decomposition of a com-
plemented modular algebraic lattice into a product of projective geometries, is
proved in $4.8. Chapter 4 can stand by itself as the basis for a one-semester
graduate course. (Nevertheless, we would advise spending severa1 weeks in the
earlier chapters at the beginning of the course.) The reader who has mastered
Chapters 1-4 can confidently go on to Volume 2 without further preliminaries,
since the mastery of Chapter 5 is not a requirement for the later material.
Chapter 5 deals with the possible uniqueness of the factorization of an
algebra into a direct product of directly indecomposable algebras. As examples,
integers, finite groups, and finite lattices admit a unique factorization. The Jordan
normal form of a matrix results from the unique decomposition of the representa- l
tion module of the matrix. This chapter contains many deep and beautiful results.
Our favorite is Bjarni Jónsson's theorem giving the unique factorization of finite
algebras having a modular congruence lattice and a one-element subalgebra
(Theorem 5.4). Since this chapter is essentially self-contained, relying only on the
!
Direct Join Decomposition Theorem in Chapter 2, a one-semester graduate
course could be based upon it. We believe that it would be possible to get through
the whole volume in a year's course at the graduate level, although none of us
t
has yet had the opportunity to try this experiment.
Volume 2 contains an introduction to first-order logic and model theory (al1
that is needed for our subject) and extensive treatments of equational logic,
equational theories, and the theory of clones. Also included in Volume 2 are
many of the deepest results about finite algebras and a very extensive survey of
the results on classifying varieties by their Maltsev properties. Later volumes will
deal with such advanced topics as commutator theory for congruence modular
varieties, tame congruence theory for locally finite varieties, and the fine structure
of lattices of equational theories.
Preface ix
Within each volume, chapters and sections within a chapter are numbered
by arabic numerals; thus, g4.6 is the sixth section of Chapter 4 (in Volume 1).
Important results, definitions, and exercise sets are numbered in one sequence
throughout a chapter; for example, Lemma 4.50, Theorem 4.51, and Definition
4.52 occur consecutively in Chapter 4 (actually in g4.6). A major theorem may
have satellite lemmas, corollaries, and examples clustered around it and numbered
1, 2, 3, . . . . A second sequence of numbers, set in the left-hand margins, is used
for a catch-al1 category of statements, claims, minor definitions, equations, etc
(with the counter reset to 1 at the start of each chapter). Exercises that we regard
as difficult are marked with an asterisk. (Difficult exercises are sometimes ac-
companied by copious hints, which may make them much easier.)
The beautiful edifice that we strive to portray in these volumes is the product
of many hundreds of workers who, for over fifty years, have been tirelessly
striving to uncover and understand the fundamental structures of general algebra.
In the course of our writing, we have returned again and again to the literature,
especially to the books of Birkhoff [1967], Burris and Sankappanavar [1981],
Crawley and Dilworth [1972], Gratzer [1978, 19791, Jónsson [1972], Maltsev
[1973], and Pierce [1968].
We wish to thank al1 of our friends, colleagues, and students who offered
support, encouragement, and constructive criticism during the years when this
volume was taking shape. It is our pleasure to specifically thank Clifford Bergman,
Joel Berman, Stanley Burris, Wanda Van Buskirk, Ralph Freese, Tom Harrison,
David Hobby, Bjarni Jónsson, Keith Kearnes, Renato Lewin, Jan Mycielski,
Richard Pierce, Ivo Rosenberg, and Constantin Tsinakis. Thanks to Deberah
Craig and Burt Rashbaum for their excellent typing. Our editor at Wadsworth
& Brooks/Cole, John Kimmel, and Production Editor S. M. Bailey, Designer
Victoria Van Deventer, and Art Coordinator Lisa Torri at Brooks/Cole have al1
taken friendly care of the authors and the manuscript and contributed greatly
to the quality of the book. Don Pigozzi's contribution to the many long sessions
in which the plan for these volumes was forged is greatly appreciated. We regret
that he was not able to join us when it came time to write the first volume;
nevertheless, his collaboration in the task of bringing this work to press has been
extremely valuable to us.
We gladly acknowledge the support given us during the writing of this
volume by the National Science Foundation, the Alexander von Humboldt
Foundation, and the Philippine-American Educational Foundation through a
Fulbright-Hays grant. Apart from our home institutions, the University
of Hawaii, the University of the Philippines, and die Technische Hochschule
Darmstadt have each provided facilities and hospitality while this project was
underway. Finally, we are deeply grateful for the solid support offered by our
wives and children over the past five years.
Ralph N. McKenzie
George F. McNulty
Walter F. Taylor
Contents
Introduction
Preliminaries
Chapter 1 Basic Concepts

1.1 Algebras and Operations
1.2 Subalgebras, Homomorphisms, and Direct Products
1.3 Generation of Subalgebras
1.4 Congruence Relations and Quotient Algebras
Chapter 2 Lattices
2.1 Fundamental Concepts
2.2 Complete Lattices and Closure Systems
2.3 Modular Lattices: The Rudiments
2.4 Modular Lattices with the Finite Chain Condition
2.5 Distributive Lattices
2.6 Congruence Relations on Lattices
Chapter 3 Unary and Binary Operations

3.1 Introduction
3.2 Unary Algebras
3.3 Semigroups
3.4 Groups, Quasigroups, and Latin Squares
3.5 Representations in End A and Sym A
3.6 Categories
xii Contents
Chapter 4 Fundamental Algebraic Results

4.1 Algebras and Clones
4.2 Isomorphism Theorems
4.3 Congruences
4.4 Direct and Subdirect Representations
4.5 The Subdirect Representation Theorem
4.6 Algebraic Lattices
4.7 Permuting Congruences
4.8 Projective Geometries
4.9 Distributive Congruence Lattices
4.10 Class Operators and Varieties
4.1 1 Free Algebras and the HSP Theorem
4.12 Equivalence and Interpretation of Varieties
4.13 Commutator Theory
Chapter 5 Unique Factorization

5.1 Introduction and Examples
5.2 Direct Factorization and Isotopy
5.3 Consequences of Ore's Theorem
5.4 Algebras with a Zero Element
5.5 The Center of an Algebra with Zero
5.6 Some Refinement Theorems
5.7 Cancellation and Absorption
Bibliography
Table of Notation
Index of Names
Index of Terms
ALGEBRAS,
LATTICES, VARIETIES
%VOLkjiWE1
Introduction
The mathematician at the beginning of the twentieth century was confronted

with a host of algebraic systems ranging from the system of natural numbers
(which had only recently been given an axiomatic presentation by Giuseppe
Peano) to Schroder's algebra of binary relations. Included among these algebraic
systems were groups (with a rapidly developing theory), Hamilton's quatern-
ions, Lie algebras, number rings, algebraic number fields, vector spaces, Grass-
mann's calculus of extensions, Boolean algebras, and others, most destined for
vigorous futures in the twentieth century. A. N. Whitehead [1898] undertook
the task of placing these diverse algebraic systems within a common framework.
Whitehead chose "A Treatise on Universal Algebra" as the title of his work.
Whitehead's attention was focused on systems of formal axiomatic reasoning
about equations, and it was to such formal systems that he applied the term
"universal algebra." In content, Whitehead's treatise is a survey of algebra and
geometry tied together by this focus on formal deductive systems.
B. L. van der Waerden's influential Moderne Algebra [1931] brought the
axiomatic method of modern algebra into instruction at the graduate level. It
deals with a number of algebraic systems, such as groups, rings, vector spaces,
and fields, within a common framework. Some of the basic concepts of general
algebra are implicit in van der Waerden's book, but at the time of its publication
there were no known results that belong, in the proper sense, to the general theory
of algebras. So, despite their broad conception, both Whitehead's treatise and
van der Waerden's book appeared before the true beginnings of our subject.
The origins of the concept of lattice can be traced back to the work on the
formalization of the logic of propositions by Augustus DeMorgan [1847] and
George Boole [1854]. The concept of a lattice, as separate from that of a Boolean
algebra, was enunciated in the work of C. S. Peirce [1880] and Ernst Schroder
[1890- 19051. Independently, Richard Dedekind ([1897] and [1900]) arrived at
the concept by way of ideal theory. Dedekind also introduced the modular law
for lattices, a weakened form of the distributive law for Boolean algebras. A series
of papers by Garrett Birkhoff, V. Glivenko, Karl Menger, John von Neumann,
and Oystein Ore, published in the mid-thirties (shortly after the appearance of
2 Introduction
van der Waerden's book), showed that lattices had fundamental applications to
modern algebra, projective geometry, point-set theory, and functional analysis.
Birkhoff s papers, appearing in 1933-1935, also marked the beginning of the
general theory of algebras.
In Birkhoff [1933], one finds both the notion of an algebra and the notion of
subalgebra explicitly stated-though somewhat more broadly than the notions
we shall use. Birkhoff [1935a] delineates the concepts of a congruence, a free
algebra, and a variety and presents severa1 of the most fundamental theorems
concerning algebras in general. In these papers, the insight that lattices offer an
important tool for understanding mathematical (especially algebraic) structures
is put forward and amply illustrated. Indeed, the general theory of algebras is
proposed in these papers as a highly appropriate arena in which to exploit the
concepts and theorems of lattice theory. Birkhoff's further development of this
theme can be followed in the successive and very different editions of his book
Lattice Theory ([1940], [1948], and [1967]). In the preface to the 1967 edition,
Birkhoff expressed his view in these words: "lattice-theoretic concepts pervade
the whole of modern algebra, though many textbooks on algebra fail to make
this apparent.. . . thus lattices and groups provide two of the most basic tools of
'universal algebra,' and in particular the structure of algebraic systems is usually
most clearly revealed through the analysis of appropriate lattices." Since these
words were written, lattices have come to play an ever more central role in the
analysis of general algebrajc systems.
In the same edition of his book (p. 132), Birkhoff defined universal algebra
as that field of mathematics that "provides general theorems about algebras with
single-valued, universally defined, finitary operations." This excellent definition
of our subject is not to be taken too literally, for in order to analyze algebras,
the researcher is inevitably led to the study of other systems that are not
algebras-ordered sets, partial algebras, and relational structures, for instance.
Neither should the dictum to prove general theorems be regarded as binding.
The concept of an algebra is broad, and truly important results applying to al1
algebras are rare. In mathematics, deep theorems are usually established under
strong hypotheses, and the general theory of algebras is no exception. In fact,
much effort has been expended in the last fifty years on the discovery of appropri-
ate conditions (such as modularity of the congruence lattice) under which far-
reaching conclusions can be drawn.
The evolution of the general theory of algebras, from its origin in the early
writings of G. Birkhoff up to the present day, divides naturally into four periods.
The first era, which lasted until about 1950, saw the publication of a handful of
papers working out the first ramifications of the ideas Birkhoff had introduced.
Free algebras, the isomorphism theorems, congruence lattices, and subalgebra
lattices were discussed. Birkhoff [1944] established the subdirect representation
theorem, destined to become a principal result in the subject. Ore ([1935] and
[1936]) gave a purely lattice-t heoretic account of the Krull-Schmidt direct
decomposition theorem from group theory and noted that this considerably
broadened the scope of the theorem. Post [1941] presented a detailed analysis
of the lattice of clones on a two-element set. The notion of a relational structure
Introduction 3
emerged during this time in investigations of mathematical logic (cf. Tarski

[1931] and [1935]). The initial work of Jónsson and Tarski [1947] on unique
direct factorization appeared. This period also saw the flowering of lattice theory
in the works of Birkhoff, Dilworth, Frink, von Neumann, Ore, and Whitman. It
was also in this period that M. H. Stone published his representation and duality
theory for Boolean algebras and A. 1. Maltsev E19361 established the compact-
ness theorem (for arbitrary first-order languages). Both of these results led to
significant later developments in general algebra.
The second era, which was to last until about 1963, was distinguished by
the predominance of ideas and methods derived from mathematical logic, espe-
cially model theory. The period began with Alfred Tarski's 1950 address to the
International Congress of Mathematicians [1952]. In his lecture, Tarski defined
very clearly a new branch of mathematics (which he called the theory of arith-
metical classes), outlined its early results, and described its goals, methods, and
prospects. Today this branch is called the model theory of first-order languages,
or simply model theory. Tarski's address marked the recognition of model theory
as a legitimate mathematical theory with potentially important applications. His
fame attracted talented young mathematicians to the field, and it developed very
rapidly during the next fifteen years. Significantly, Tarski conceived the theory
of arithmetical classes to be "a chapter of universal algebra," and he credited
Birkhoff for an important early result in this theory. In [1935a], Birkhoff had
characterized varieties by the form of the first-order sentences axiomatizing them
(i.e., equations), and this result now became a paradigm for numerous results in
model theory, of a type called "preservation theorems." During the 1950s, Tarski
influenced severa1 generations of logicians at Berkeley in the direction of the
general theory of algebras; in Russia, Maltsev exerted a comparable influence.
One of the many fruits of this era was the concept of ultraproduct, put forward
by LOS [1955] and extensively developed in the Berkeley school. Independently
of these schools, in the late fifties a group of Polish algebraists (most prominently,
E. Marczewski) published more than fifty papers on the algebraic theory of free
algebras. It was also during this period that A. L. Foster at Berkeley focused the
attention of an active group of students on the possibility of placing Stone's
representation theorem for Boolean algebras in a general setting.
By 1960, the focus of research in model theory had moved far from
the general theory of algebras. At the same time, universal algebraists had
rediscovered an interest in problems that could not be handled by purely
model-theoretic methods. There now began a third era, which was to last into
the late 1970s. An early result of this period, which still exerts considerable
influence, appeared in a paper by G. Gratzer and E. T. Schmidt [1963]. The
authors presented an abstract characterization of congruence lattices of arbitrary
algebras, thus solving a difficult problem of Birkhoff left open from the first
period. This was followed shortly by a paper of B. Jónsson [1967], which
provided elegant and powerful tools for the study of a diverse family of
varieties-the congruence-distributive varieties. A new theme of classifying
varieties according to the behavior of congruences in their algebras, and more
generally by Maltsev properties, received much attention in the 1970s. A 1954
4 Introduction
paper by Maltsev had shown the first connection of the type that was now to be
intensively studied, between a congruence property and a pair of equations
holding in a variety. This third period also witnessed the formulation of the
notion of Boolean product, the first deep results concerning the lattices of
varieties, and a resumption of efforts toward understanding the lattices of clones
over finite sets.
The quantity of published papers increased astronomically during the 1960s
and 1970s, and three of the hardest long-standing open problems in lattice theory
were solved near the end of the period. P. Pudlák and J. Tiima [1980] proved
that every finite lattice embeds into a finite partition lattice (which Whitman had
conjectured in 1946). R. Freese [1980] proved that the word problem for free
modular lattices is recursively unsolvable. J. B. Nation [1982] proved, as Jónsson
had conjectured in 1960, that the finite sublattices of free lattices are identical
with finite lattices satisfying three of the elementary properties that characterize
free lattices.
In the midst of this period appeared four books devoted to the exposition
of the general theory of algebras. Cohn [1965] was the first to appear. The books
of Gratzer and Pierce came out in 1968, but the publication of Maltsev [1973]
was delayed by the author's death in 1967. In essence, each of these books gives
an account of the findings of the two earlier periods. Of the four, Gratzer's book
is the most comprehensive and soon became the standard reference. More
sharply focused books, like Jónsson [1972], were also published. A very readable
(but al1 too slender) book by Burris and Sankappanavar E19811 appeared at the
close of this period. This period also saw the founding of Algebra Uniuersalis (a
journal devoted to lattice theory and the general theory of algebras), the growth
of strong centers of activity in Winnipeg, Darmstadt, Szeged, Honolulu, and
elsewhere, and the regular organization of conferences at the national and the
international levels, emphasizing universal algebra and lattice theory.
The appearance of general commutator theory (in the papers by J. D. H.
Smith 119761 and J. Hagemann and C. Herrmann [1979]) heralded the start of
the present era. In just a few years, the application of commutator theory has
greatly transformed and deepened the theory of varieties. Commutator theory
finds its strongest applications in varieties that satisfy rather strong Maltsev
properties, such as congruence modularity. Another product of this era, tame
congruence theory, has found broad applications in the study of locally finite
varieties without Maltsev conditions. In another direction, D. Pigozzi (with W.
Blok and P. Kohler [1982], [1984]) has shown that the congruence distributive
varieties fa11 naturally into an interesting hierarchy with many levels. One of the
principal features of this period will certainly be a much deeper understanding
of the significant natural families of varieties and the different approaches suit-
able for uncovering the properties of these families.
Preliminaries
We assume that the reader has a working knowledge of the most basic notions
of set theory and has had a modest exposure to classical algebraic structures such
as groups and rings. The approach to set theory is informal, and no particular
set of axioms need be specified. (If the reader is unfamiliar with the material
reviewed below, we recommend consulting Halmos [1960] or Vaught [1985],
both of which contain excellent elementary introductions to set theory.) We use
classes as well as sets. Informally speaking, a class is a collection so large that
subjecting it to the operations admisible for sets would lead to logical contradic-
tions. For example, we speak of the set of al1 sets of natural numbers but the
class of al1 Abelian groups. We often use the term family in reference to a set
whose members are sets. (In pure formal set theory, every member of any set is
itself a set.)
In dealing with sets we use these standard notations: membership (E),
set-builder notation ({-: -)), the empty set (a),inclusion (S), proper inclusion
(c), union (U and U),intersection ( í l and o),
difference (-), (ordered) n-tuples
((x,, . ,xn-, )), direct (or Cartesian) products of sets (A x B, ni,,
Ai), direct
powers of sets (A'). We shall not distinguish between (ordered) pairs and 2-tuples.
The ordered pair of x and y will be denoted by (x, y) and sometimes by
(x, y). There follows a series of reinarks introducing further notations and basic
definitions.
1. The power set of a set A is the set {B: B S A) of al1 subsets of A. It is denoted
by Y ( A ) .
2. An is the set ( (x,, xn-, ): {x,, - xn-, ) G A) of al1 n-tuples each of
e , m ,
whose terms belongs to A. For the denotation of n-tuples we use bars over
letters. Thus
A" = {X: X = ( X ~ , . . ~ , X ~where
-~) { x O , ~ ~ ~ , xS
n -A).
,)
3. Concerning relations:
a. An n-ary relation on a set A is a subset of A".
b. A 2-ary relation on A is called a binary relation.
Preliminaries
c. The converse r" of a binary relation r on A is given by (a, b) E ru iff

(b, a) E r. (L'iff" is an abbreviation for "if and only if".) r" may also be
denoted r - l .
d. The relational product of two binary relations r, S on A is defined by:
(a, b) E r o S iff for some c, (a, c) E r and (c, b) E S.
e. The transitive closure of a binary relation r on A is the binary relation
r U (r o r) U (r o r o r) U
m. Thus (x, y) belongs to this transitive closure iff
for some integer n 2 1 there exist elements x, = x, x, , - x, = y satisfy-
e,
ing ( x , , x , + , ) ~ rfor a11 i < n.

4. Concerning functions:
a. A function f from a set A to a set B is a subset of B x A such that for
each a E A there is exactly one b E B with (b, a) f. Synonyms for function
are mapping, map, and system. If f is a function from A to B, then we
write f : A -+ B; and if (b, a) EL then we write f(a) = b, or fa = b, or
f, = b,or f : a ~ b . T h u s i f f :A h B , then f = {(f(a),a):a~A).
b. The set of al1 functions from A to B is denoted by B ~ .
c. If f E BAand g E CB,then f and g are relations on A U B U C. We write gf
for their relational product g o f ; thus gf E CAand gf (a) = g(f(a)).
d. If f E BA,then ker f, the kernel of f, is the binary relation {(a,, al ) E A2:
1
I f (a,) = f (a,)}. f is called injective (or one-to-one) iff (x, y) E ker f implies
I
1
x = y (for al1 x, y E A).
e. If f E BA, X E A, and Y G B, then f (X) = { f (x): x E X) (the f-image of
X) and f -'(Y) = {x E A: f (x)E Y} (the f-inverse image of Y). f : A -+B is
called surjective (or said to map A onto B) iff f (A) = B.
I f. The function f E BAis bijective iff it is both injective and surjective.
l g. If f eBA,then we say that the domain of f is A, the co-domain of f is B,
1
and the range of f is the set f (A).
ll 1 h. Function-builder notation (-: -) is quite analogous to set-builder
notation. If X is a set and r(x) is an expression that completely prescribes
1
I an object for each x EX, then (r(x): x E X ) designates the function f with
I
domain X satisfying f (x) = r(x) for al1 x E X. For example, if g E Y'
1 and h e z Y ,then f = hg (see paragraph (4c)) is the same function as
(h(g(x)): x EX). If the domain of a function that we wish to introduce is
11
clearly implied by the context (say it is the set of real numbers), then we
l
I may specify the function with a simple phrase like "let f be the function
X H x2."
1 i. An (indexed) system (F,: i E 1), indexed by a set I, is just a function whose
i domain is I.
j. If (A,: iEI) is a system of sets (A, is a set for al1 iEI), then
I
l
l
n (A,: i~ I), or ni,, Ai, denotes the set of al1 functions f with domain
I such that f (i) E A, for al1 i E I. It is called the (direct or Cartesian) product
1
1
I l
of the A,, i~ l. For any sets A and I, the set A' of al1 functions from I to
I A is a direct power of A.
I
5. The Axiom of Choice is the statement that if (A,: i E I) is any system of sets
1
l
Preliminaries 7
with A, # @ for al1 iE 1, then

axiorn.
ni,, A, # @. We assume the validity of this
+ + +
Z, Q, R, C denote respectively the set of al1 integers O, 1, 2, 3, s., the
set of al1 rational numbers, the set of al1 real numbers, and the set of al1
complex numbers.
The union of a family (or set) F of sets, UF, is defined by x E UF iff x E B for
some B EF. The intersection of a family F c 9(A), written OF, is defined
(dually to the union) to be the set { X EA: X E B for al1 B EF}. (If F is non-
empty (F # @), then this is independent of A.)
An order over a set A is a binary relation I on A such that
a. I is reflexive over A; Le., (x, x) E I for al1 x E A.
b. I is anti-symmetric; i.e., if (x, y) E I and (y, x) E 1, then x = y.
c. I is transitive; i.e., (x, y) E I and (y, z) E 5 always imply (x, Z) E < .
For orders (and binary relations more generally), we often prefer to write
x 4 y in place of (x, y) E I. Given an order < over a nonempty set A, the
pair (A, I) is called an ordered set. (Ordered sets are frequently called
"partially ordered" sets in the literature.)
By a chain in an ordered set (A, 5 ) is meant a set C G A such that for al1
x, y E C either x < y or y I x. An upper bound of C is an element u E A for
which c i u for al1 c E C.
Zorn's Lemma is the statement that if (A, 5) is an ordered set in which
every chain has an upper bound, then (A, I ) has a maximal element m (i.e.,
m~ A and m 5 X'E A implies m = x). We take this statement as an axiom.
The Hausdorff Maximality Principle is the statement that every ordered set
has a maximal chain. More precisely, let (A, I )be an ordered set and let
L denote the family of al1 chains in (A, 5 ) . The principle states that the
ordered set (L, S) has a maximal element. The Hausdorff Maximality
Principle is equivalent to Zorn's Lemma.
A linearly ordered set (also called a chain) is an ordered set (A, I ) in which
every pair of elements x and y satisfy either x I y or y i x. A well-ordered
set is a linearly ordered set (A, I ) such that every nonempty subset B c A
has a least element 1 (i.e., 1 E B and 1 I x for al1 x E B).
Concerning ordinals:
a. The ordinals are generated from the empty set @ using the operations
of successor (the successor of x is S(x) = x U {x}) and union (the union
of any set of ordinals is an ordinal).
b. O = @ (the empty set), 1 = S(O), 2 = S(l), . The finite ordinals are
0, 1,2, - , also called natural numbers or non-negative integers.
c. Every set of ordinals is well-ordered by setting a I /3 (a and /3 are ordinals)
iff a = /3 or a ~ / 3Then
. O 5 1I 2 and we have n = {O, 1;--,n - 1)
- m - ,
for each finite ordinal n 2 1.

d. The least infinite ordinal is w = {O, 1,2, a},which is the set of al1 finite
ordinals.
Preliminaries
e. It is useful to know that n-tuples are functions having domain

(0, l , . . . , n - l}.
14. Concerning cardinals:
a. Two sets A and B have the same cardinality iff there is a bijection from
A to B.
b. The cardinals are those ordinals K such that no ordinal j< IC has the
same cardinality as K.The finite cardinals are just the finite ordinals, and
o is the smallest infinite cardinal.
c. The Well Ordering Theorem is the statement that every set has the same
cardinality as some ordinal. We take it as an axiom. (Actually, the Well
Ordering Theorem, the Axiom of Choice, Zorn's Lemma, and the Haus-
dorff Maximality Principle-see (5), (10) and (11)-can be proved mutu-
ally equivalent in a rather direct fashion.)
d. The cardinality of a set A is the (unique) cardinal K such that A and K
have the same cardinality. The cardinality of A is written 1 Al.
e. The power set P(A) of a set A has the same cardinality as 2* (hence the
term "power set").
f. Operations of addition, rnultiplication, and exponentiation of cardinals
are defined so that for any sets A and B, (Al-IBI = (A x B ( ,( A (+ ( B J=
\A U BI if A and B are disjoint (i.e., if A n B = a), and j ~ l l =
~ l /AB/.
Addition and multiplication are rather trivial where infinite cardinals are
involved. For instance, if K, A are cardinals, O < K < A, and co < A, then
K + A = K - A = A. The cardinal 2" is the cardinality of the set of real
numbers.
g. There is a unique one-to-one order-preserving function (denoted by the
Hebrew letter aleph) from the class of al1 ordinal numbers onto the class
of al1 infinite cardinal numbers. The first few infinite cardinal numbers,
in ascending order, are K,(=co), N,, N,,
S ,
etc. The least cardinal
larger than an infinite cardinal K is denoted by K'.
15. Concerning equivalence relations:
a. An equivalence relation over a set A is a binary relation - -
on A that is
reflexive over A, transitive (see (8)), and symmetric (i.e., (x, y) E iff
(Y,x) E 4-
--
b. For an equivalence relation over A and for x E A, the equivalence class
- -
of x modulo is the set x/ = (y E A: x y). The factor set of A modulo
-5s Al- = (x/-: XEA).
c. For an equivalence relation
- - over A, Al- is a partition of A. That is,
A/ is a set of nonempty subsets of A, A = UA/-, and each pair of
-
distinct sets U and V in A/ are disjoint. Every partition of A arises in
this fashion from an equivalence relation (which is uniquely determined
by the partition).
d. The set of al1 equivalence relations over A is denoted by Eqv A. Individual
equivalence relations are usually denoted by lowercase Greek letters a,
p, p, etc. Instead of (x, y) E p (where p is an equivalence relation) we may
write, variously, xpy, x = y (mod p), or x E,y. (In diagrams, the fact that
Preliminaries 9
(x, y ) ~p may be indicated by connecting points labeled x and y with a

line segment and labeling the segment with p.)
e. (Eqv A, c.) is an ordered set having greatest lower bounds and least
upper bounds for any subset of its elements. The greatest lower bound
of a set S c Eqv A is OS. The least upper bound of S is the transitive
closure of U S (see (3e)).
16. The equality symbol (=) is used to assert that two expressions name the
same object. The formal equality symbol (Ñ) is used to build equations,
such as the commutative law x y Ñ y x, which can only become true or
false after we assign specific values to their symbols and ask whether the
two sides name the same object. (See $4.11 in Chapter 4.)
C H A P T E R
Basic Concepts
1.1 ALGEBRAS AND OPERATIONS
An algebra is a set endowed with operations. Algebras are the fundamental

objects with which we shall deal, so our first step is to make the preceding
sentence precise. Let A be a set and n be a natural number. An operation of rank
n on A is a function from A" into A. Here we have used A" to denote the n-fold
direct power of A-the set of al1 n-tuples of elements of A. By a (finitary)operation
on A we mean an operation of rank n on A for some natural number n. Because
virtually every operation taken up in this book will be finitary, we will generally
omit the word "finitary" and use "operation" to mean finitary operation. If A is
nonempty, then each operation on A has a unique rank. Operations of rank O
on a nonempty set are functions that have only one value; that is, they are
constant functions of a rather special kind. We call operations of rank O constants
and identify them with their unique values. Similarly, we call operations of rank
1 on A unary operations and identify them with the functions from A into A.
Binary and ternary operations are operations of rank 2 and 3, respectively. We
use n-ary operation, operation of rank n, and operation of arity n interchangeably.
It is important to realize that the domain of an operation of rank n on A is the
whole set A". Functions from a subset of A" into A are called partial operations
of rank n on A. Subtraction is a binary operation on the set Z of integers, but it
is only a partial operation on the set m of natural numbers.
The fundamental operations most frequently encountered in mathematics
have very small ranks. A list of these important operations certainly includes
addition, subtraction, multiplication, division, exponentiation, negation, con-
jugation, etc., on appropriate sets (usually sets of numbers, vectors, or matrices).
This list should also include such operations as forming the greatest common
divisor of two natural numbers, the composition of two functions, and the union
of two sets. Of course, one is almost immediately confronted with operations of
higher rank that are compounded from these. Operations of higher finite rank
whose mathematical significance does not depend on how they are built up from
operations of smaller rank seem, at first, to be uncommon. Such operations will
emerge later in this work, especially in Chapter 4 and in later volumes. However,
12 Chapter 1 Basic Concepts
even then most of the operations have ranks no larger than 5. While there is
some evidence that operations of such small rank provide adequate scope for the
development of a general theory of algebras, why this might be so remains a
puzzle.
To form algebras, we plan to endow sets with operations. There are severa1
ways to accomplish this. We have selected the one that, for most of our purposes,
leads to clear and elegant formulations of concepts and theorems.
DEFINITION 1.1. An algebra is an ordered pair (A, F ) such that A is a

nonempty set and F = (F,: i E 1) where Fi is a finitary operation on A for each
i E I. A is called the universe of (A, F), Fi is referred to as a fundamental or basic
operation of (A, F) for each i E I, and I is called the index set or the set of operation
symbols of (A, F).
The reason we have endowed our algebras with indexed systems of opera-
tions rather than with mere sets of operations is so that we have a built-in means
to keep the operations straight. From the customary viewpoint, rings have two
basic binary operations labeled "addition" (or +) and "multiplication" (or S).
For the development of ring theory, it is essential to distinguish these operations

from each other. In effect, most expositions do this by consistent use of the
+
symbols and . : The actual binary operations in any given ring are indexed by
these symbols. This is why we have chosen to cal1 the index set of ( A , F ) its set
of operation symbols. The distinction between operation symbols and operations
is important, and we will have much to say regarding it in 94.11.
The notation implicit in the definition above is unwieldly in most situations.
Quite often the set of operation symbols is small. For example, ring theory is
accommodated by the operation symbols +, and - (this last symbol is
e ,
intended to name the unary operation of negation). But surely
is an uncomfortable way to display the ring of integers. In this situation and

others like it, we find
+
much more acceptable. Notice that in this last display , and - are no longer
S ,
operation symbols but operations; exactly which operations they are is clear
from context.
As a general convention, we use uppercase boldface letters A, B, C, to
denote algebras and the corresponding uppercase letters A, B, C, to denote
their universes, attaching subscripts as needed. Thus, in most uses, A is the
universe of A, B, is the universe of B,, and so on. If Q is an operation symbol of
A, then we use Q* to stand for the fundamental operation of A indexed by Q; we
e
say that denotes Q* or that Q* is the interpretation of Q in A. Whenever the
cause of clarity or the momentum of customary usage dictates, we will depart
from these conventions.
Given an algebra A with index set 1, there is a function p called the rank
function from I into the set cc, of natural numbers defined by:
1.1 Algebras and Operations
p(Q) is the rank of QAfor al1 Q E 1.

The rank function of an algebra is also referred to as its similarity type or, more
briefly, its type. Algebras A and B are said to be similar if and only if they have
the same rank function. The similarity relation between algebras is an equiva-
lente relation whose equivalence classes will be called similarity classes. Most of
the time (with some important exceptions), only algebras of the same similarity
type will be under consideration. In fact, this hypothesis that al1 algebras at hand
are similar is so prevalent that we have left it unsaid, even in the statement of
some theorems.
The rank functions are partially ordered by set inclusion (that is, by exten-
sion of functions). This ordering can be imposed on the similarity classes as well.
For individual algebras, we say that A is a reduct of B (and that B is an expansion
of A) if and only if A and B have the same universe, the rank function of A is a
subset of the rank function of B, and eA
= QBfor al1 operation symbols Q of A.
In essence, this means that B can be obtained by adjoining more basic operations
to A. For example, each ring is an expansion of some Abelian group.
We close this section with a series of examples of algebras. Besides illus-
trating the notions just introduced, these examples specify how we formalize
various familiar kinds of algebras and serve as resources for later reference. In
formulating the examples, we use the following operation symbols:
Constant symbols: e, 1, 0, and 1'
Unary operation symbols: -, -',
-, ", and f, for each r E R
Binary operation symbols: + , A , v . m ,
Semigroups
A semigroup is an algebra A = (A, m A ) such that:
(a b) c
aA =a eA (b aAc) for a11 a, b, c E A.
Thus a semigroup is a nonempty set endowed with an associative binary opera-
tion. A typical example of a semigroup is the collection of al1 functions from X
into X, where X is any set, with the operation being composition of functions.
A more sophisticated example is the collection of al1 n x n matrices of integers
endowed with matrix multiplication.
Monoids
A monoid is an algebra A = (A, eA,eA) such that ( A , sA) is a semigroup and
To obtain some concrete examples, we can let e denote the identity function in
our first example of a semigroup and the identity matrix in the second example.
Although every monoid is an expansion of a semigroup, not every semigroup
can be expanded to a monoid.
Chapter 1 Basic Concepts
Groups
A group is an algebra A = (A, S , -', e) such that ( A , e) is a monoid and
S ,
"
(The reader will have observed that the superscript has grown tiresome and
been dropped.) A typical example of a group is the collection of al1 one-to-one
functions from X onto X (such functions are called permutations of X), where X
is an arbitrary set and the operations are composition of functions, inversion of
functions, and the identity function. A more intricate example is the collection
of isometries (also called distance-preserving functions) of the surface of the unit
ball in ordinary Euclidean three-dimensional space endowed with the same sorts
of basic operations. Groups have been construed as algebras of severa1 different
similarity types. Most of the popular renditions of group theory define groups
as certain special kinds of semigroups. Our choice of fundamental operations
was motivated by the desire to have the class of groups turn out to be a variety.
Still, there are a number of quite satisfactory ways to present the intuitive notion
of a group. For instance, there is no real need to devote a basic operation to the
unit element. It is even possible to make do with a single binary operation, though
this cannot be . The sense in which such formulations are equivalent, not only
for groups but for algebras in general, is made precise in 54.12. Some interesting
aspects of serdigroups and groups are explored in Chapter 3.
Rings
A ring is an algebra (A, +, ., -, O) such that (A, + ,-, O) is an Abelian group,
( A , - ) is a semigroup, and
and
A ring with unit is an algebra (A, +, -, 0 , l ) such that (A, +, -,O) is a ring
S , S ,
and ( A , 1) is a monoid. A familiar example of a ring (with unit) is the integers

S ,
endowed with the familiar operations. Another example is the set of n x n

matrices with real entries endowed with the obvious operations. We regard fields
as special kinds of rings.
Vector Spaces and Modules

In the familiar treatments, vector spaces and modules are equipped with a binary
operation called addition and a scalar multiplication subject to certain condi-
tions. As ordinarily conceived, the scalar multiplication is not what we have
called an operation. An easy way around this trouble is to regard scalar multi-
plication as a schema of unary operations, one for each scalar. Actually, this
1.1 Algebras and Operations 15
is in accord with geometric intuition by which these operations amount to

stretchings or shrinkings. Let
R = (R, +, ., -,o, 1)
be a ring with unit. An R-module (sometimes called a left unitary R-module) is
an algebra (M, +, -, O,f,),,. such that (M, +, -,O) is an Abelian group and
for al1 a, b E M and for al1 r, S, and t E R the following equalities hold:
f,(f,(a)) = ft(a) where r - s = t in R
f,(a + b) = f,(a) + L(b)
f,(a) + fs(a) =&(a) where r + s = t in R
f d a ) = a.
In essence, what these conditions say is that f,is an endomorphism of the Abelian
group (M, + , - ,O) for each r E R, that this collection of endomorphisms is itself
a ring with unit, and that the mapping r wf, is a homomorphism from R onto
this ring. Although we will soon formulate such notions as homomorphism in
our general setting, this last sentence is meaningful as it stands in the special
context of ring theory. Part of the importance of modules lies in the fact that
every ring is, up to isomorphism, a ring of endomorphisms of some Abelian
group. This fact is analogous to the more familiar theorem of Cayley to the effect
that every group is isomorphic to a group of permutations of some set. In the
event that R is a field, we cal1 the R-modules vector spaces over R.
Bilinear Algebras over a Field

+
Let F = (F, + , - ,O, 1) be a field. An algebra A = (A, , -, O, f,), , is a
m, m , ,
bilinear algebra over F if (A, + , - ,O,f,),., is a vector space over F and for al1
a, b, c ~ A a n da l l r ~ F :
(a + b).c = (aqc) + (b-c)
c (a + b) = (c . a) + (c b)
f,(a b) = (f,(a)). b = a -f,(b).

If, in addition, (a b) c = a (b c) for al1 a, b, c E A, then A is called an associative
algebra over F. Thus an associative algebra over a field has both a vector space
reduct and a ring reduct. An example of an associative algebra can be constructed
from the linear transformations of any vector space into itself. A concrete exam-
ple of this kind is obtained by letting A be the set of al1 2 x 2 matrices over the
field of real numbers and taking the natural matrix operations. Lie algebras,
Jordan algebras, and alternative algebras provide important examples of non-
associative bilinear algebras that have arisen in connection with physics and
analysis. A Lie algebra is a bilinear algebra that satisfies two further equalities:
a . a = O forallaEA
((a - b) c) + ((b . c) - a) + ((c a). b) = O for al1 a, b, c E A.
Suppose (A, + , ., -,O,f,),EFis an associative algebra over F. Define
It is not difficult to verify that (A, +, *, -, O,f,),,, is a Lie algebra. A good but
brief introduction to bilinear algebras is available in Jacobson [1985]. Pierce
[1982] offers an excellent account of associative algebras over fields. We remark
that a common usage of the word "algebra" in the mathematical literature is to
refer to those mathematical structures we have called "bilinear algebras over
fields." The objects we have called "algebras" are then referred to as "universal
algebras" (although there is nothing especially universal about, say, the three-
element group, which is one of these objects) or as "a-algebras" (perhaps because
R is a kind of Greek abbreviation for "operation").
The establishment of the theories of groups, rings, fields, vector spaces,
modules, and various kinds of bilinear algebras over fields is a sterling accom-
plishment of nineteenth-century mathematics. This line of mathematical research
can be said to have reached its maturity at the hands of such mathematicians as
Hilbert, Burnside, Frobenius, Wedderburn, Noether, van der Waerden, and E.
Artin by the 1930s. It has continued to grow in depth and beauty, being today
one of the most vigorous mathematical enterprises.
There is another important series of examples of algebras, different in
character from those described above.
Semilattices
A semilattice is a semigroup (A, A ) with the properties
and
A typical example of a semilattice is formed by taking A to be the collection of

al1 subsets of an arbitrary set with the operation being intersection. Another
example is formed by taking A to be the compact convex sets on the Euclidean
plane and the operation to be the formation of the closed convex hull of the
union of two compact convex sets.
Lattices
A lattice is an algebra (A, A , v ) such that both (A, A ) and ( A , v ) are semi-
lattices and the following two equalities hold:
a v (a A b) = a for al1 a, b E A
and
a A (a v b) = a for al1 a, b E A.
1.1 Algebras and Operations 17
A typical example of a lattice is formed by taking A to be the collection of al1

equivalence relations on an arbitrary set, A to be intersection, and v to be the
transitive closure of the union of two given equivalence relations. Another
example is formed by taking A to be the set of natural numbers, v to be the
formation of least common multiples, and A to be the formation of greatest
common divisors. Lattices have a fundamental role to play in our work. Chapter
2 is devoted to the elements of lattice theory. The operation A is referred to as
meet, and the operation v is called join.
Boolean Algebras
A Boolean algebra is an algebra (A, A , v , -) such that ( A , A , v ) is a lattice
and for al1 a, b, c E A the following equalities hold:
a A (b v c) = (a A b) v (a A c)
a v (b A c) = (a v b) A (a v c)
(a v b)- = a- A b-
(a A b)- = a- v b-
--
a =a
(a- A a) v b = b
(a- v a) A b = b.
Thus a Boolean algebra is a distributive lattice with a unary operation of
complementation (denoted here by -) adjoined. As an example, take A to be the
collection of al1 subsets of an arbitrary set X, let the join v be union, the meet
A be intersection, and the complementation - be set complementation relative
to X. Another example is afforded by the clopen (simultaneously open and
closed) subsets of a topological space under the same operations as above.
Relation Algebras
A relation algebra is an algebra (A, A , v , -,", 1') such that (A, A , v , -) is a
m,
Boolean algebra, ( A , 1') is a monoid, and the following equalities hold for al1
a ,
a, b, c E A:
a (b v c) = (a b) v (a. c)
(a b)" = bu a"
(aU)" = a
(a-)" = (au)-
( i 7 y= 1'
(a v b)u = a" v bu
(au (a b)-) A b = a- A a.
An example of a relation algebra can be formed by taking A to be the collection
of al1 binary relations on an arbitrary set X, giving A the Boolean operations by

regarding A as the power set of X2,and defining the remaining operations so that
R . S = { ( x , y ) : ( x , z ) ~ R and (z,y)~S forsomez~x)
The relation algebra obtained in this way from the set X is sometimes denoted
by RelX. Jónsson [1982] provides a thorough and very readable overview.
Relation algebras have a rich theory, indeed rich enough to offer a reasonable
algebraic context for the investigation of set theory. The essential idea for such
an investigation is to regard set theory as the theory of membership. Membership
is a binary relation. Adjoining an additional constant (nullary operation) e to the
type of relation algebras to stand for membership opens the possibility of
developing set theory by distinguishing e from other binary relations by means
of the algebraic apparatus of relation algebras. It turns out to be possible, for
example, to render the content of the Zermelo-Fraenkel axioms for set theory
entirely as equations in this setting. The deep connections these algebras have
with the foundations of mathematics emerges in the monograph of Tarski and
Givant [1987].
Our list of examples of algebras ends here, having merely touched on a small
selection. Further kinds of algebras will be introduced from time to time to serve
as examples and counterexamples.
Exercises 1.2
1. Let A be a nonempty set and Q be a finitary operation on A. Prove that the
rank of Q is unique.
2. Let A be a nonernpty set. Describe the operations on A of rank O in set-
theoretic terms.
3. Construct a semigroup that cannot be expanded to a monoid.
4. Construct a semigroup that is not the multiplicative semigroup of any ring.
5. Prove that every ring can be embedded in a ring that can be expanded to a
ring with unit.
.,
6. Let (A, + , -,O, f,), be an associative algebra over the field P and define
m ,
Prove that (A, +, *, -, O,f,),., is a Lie algebra.

7. Let A be a set and denote by Eqv A the set of al1 equivalence relations on A.
For R, S E Eqv A define
R A S = R ~ S
R v S = R U RoSU RoSoRU R O S O R O S U . . .
where o stands for relational product (that is, a(R O S)b means that there is
some c such that both aRc and cSb). Prove that (Eqv A, A , v ) is a lattice.
1.2 Subalgebras, Homomorphisms, Direct Products
1.2 SUBALGEBRAS, HOMOMORPHISMS, AND

DIRECT PRODUCTS
One of the hallmarks of algebraic practice is the prominent role played by

relationships holding among algebras. Some of the subtleties of complicated
algebras can be more readily understood if some tractable way can be found to
regard them as having been assembled from less complicated, more thoroughly
understood algebras. The chief tools we will use to assemble new algebras from
those already on hand are the formation of subalgebras, the formation of homo-
morphic images, and the formation of direct products. The reader is probably
familiar with these notions in the settings of groups, rings, and vector spaces.
They fit comfortably into our general setting.
Let F be an operation of rank r on the nonempty set A, and let X be a subset
of A. We say that X is closed with respect to F (also that F preserves X and that
X is invariant under F) if and only if
F(a,,a,,...,a,-,)EX for al1 a,, a,,.-.,a,-, E X .
In the event that F is a constant, this means that X is closed with respect to F if
and only if F EX. Thus the empty set is closed with respect to every operation
on A of positive rank, but it is not closed with respect to any operation of rank O.
Taking A to be the set of integers, we see that the set of odd integers is closed
with respect to multiplication but not with respect to addition.
DEFINITION 1.3. Let A be an algebra. A subset of the universe A of A, which

is closed with respect to each fundamental operation of A, is called a subuniverse
of A. The algebra B is said to be a subalgebra of A if and only if A and B are
similar, the universe B of B is a subuniverse of A, and QBis the restriction to B
of QA, for each operation symbol Q of A. SubA denotes the set of al1 sub-
universes of A.
The ring of integers is a subalgebra of the ring of complex numbers.

"B is an extension of A" means that A is a subalgebra of B; we render this
in symbols as A c B. This convenient abuse of symbols should not lead to
ambiguity-if nothing else, the boldface characters convey the algebraic intent.
Our system of conventions exposes us, from time to time, to the minor
annoyance of subuniverses that are not universes of subalgebras. We insisted
that the universes of algebras be nonempty, but we have also insisted on empty
subuniverses (exactly when there are no operations of rank O). By accepting this
incongruity we avoid the need to single out many special cases in the statements
of definitions and theorems.
The notion of subalgebra defined above occasionally conflicts, at least in
spirit, with common usage. Consider the case of fields. We have regarded fields
as rather special sorts of rings, but the possibility of putting the function that
sends each nonzero element to its multiplicative inverse on the same distin-
guished footing as the ring op~rationsis certainly enticing. We have not taken
this step, since the function involved is only a partial operation, and the resulting
mathematical system would not fa11 within our definition of an algebra. This
deviation from our definition may seem small-from many viewpoints it is-but
to widen the definition so as to allow partial operations would result in a havoc
of technical complications and substantially alter the character of the ensuing
mathematics. We have the option of declaring that O-' = O in order to force the
operation to be defined everywhere, but this invalidates many equalities one
ordinarily thinks of as holding in fields. For example,
would not hold when x = O unless the field had only one element. In any case,
fields generally have subalgebras that are not fields. The integers form a sub-
algebra of the field of complex numbers that is not a subfield. In connection with
the examples of algebras that concluded the previous section, the situation is
very pleasant: Every subalgebra of a group, a ring, a vector space, etc., is again
an algebra of the same sort.
Now consider similar algebras A and B and let Q be an operation symbol
of rank r. A function h from A into B is said to respect the interpretation of Q if
and only if
for al1 a,, . - ,a,-, E A.

DEFINITION 1.4. Let A and B be similar algebras. A function h from A into
B is called a homomorphism from A into B if and only if h respects the interpre-
tation of every operation symbol of A. hom(A,B) denotes the set of al1 homo-
morphisms from A into B.
We distinguish severa1 kinds of homomorphisms and employ notation for

them as follows. Let A and B be similar algebras. Each of
h:A-+B
ASB
h E hom(A, B)
denotes that h is a homomorphism from A into B. By attaching a tail to the
arrow, we express the condition of one-to-oneness of h; by attaching a second
head to the arrow, we express the condition that h is onto B. Thus both
and
denote that h is a one-to-one homomorphism from A into B. We cal1 such

homomorphisms embeddings. Likewise, both
1.2 Subalgebras,Homomorphisms, Direct Products
and
ALB
denote that h is a homomorphism from A onto B, and in this case we say that
B is the homomorphic image of A under h. Further, each of
and
h
A r B
denotes that h is a one-to-one homomorphism from A onto B. We cal1 such
homomorphisms isomorphisms. A and B are said to be isomorphic, which we
denote by A E B, iff there is an isomorphism from A onto B. A homomorphism
from A into A is called an endornorphism of A, and an isomorphism from A onto
A is called an autornorphism of A. End A and Aut A denote, respectively, the set
of al1 endomorphisms of A and the set of al1 automorphisms of A. The identity
rnap 1, belongs to each of these sets; moreover, each of these sets is closed with
respect to composition of functions. In addition, each autornorphism of A is an
invertible function and its inverse is also an automorphisrn. Thus (End A, o, 1,)
is a monoid, which we shall designate by End A, and (Aut A, o, -l, 1,) is a group,
which we shall designate by Aut A.
(R', . ) and (R, + ) are isomorphic, where R is the set of real numbers and
R+ is the set of positive real numbers. Indeed, the natural logarithm function is
an isornorphism that illustrates this fact.
An isomorphism is a one-to-one correspondence between the elements of
two algebras that respects the interpretation of each operation symbol. This
rneans that with regard to a host of properties, isomorphic algebras are indistin-
guishable from each other. This applies to most of the properties with which we
shall deal; if they are true in a given algebra, then they are true for al1 isornorphic
images of that algebra as well. Such properties have been called "algebraic
properties."
On the other hand, algebras that are isornorphic can be quite different from
each other. For example, the set of al1 twice continuously differentiable real-
valued functions of a real variable that are solutions to the differential equation
can be given the structure of a vector space over the field of real numbers. This
vector space is isomorphic to the two-dimensional space familiar from Euclidean
plane geometry. Roughly speaking, the distinction between these two isoqorphic
vector spaces can be traced to the "internal" structure of their elements: on the
one hand, functions, and on the other, geometric points. The notion that func-
tions can be regarded as points with a geometric character is a key insight, not
only for differential equations but also for functional analysis. Because the
interna1 structure of the elements of an algebra can be used to establish algebraic

properties, some of the most subtle and powerful theorems of algebra are those
that assert the existence of isomorphisms.
Isomorphism is an equivalence relation between algebras, and the equiva-
lente classes are called isomorphism types. Isomorphism is a finer equivalence
relation than similarity, in the sense that if two algebras are isomorphic, then
they are also similar. In fact, among equivalence relations holding between
algebras, isomorphism is probably the finest we will encounter; similarity is one
of the coarsest.
The formation of subalgebras and of homomorphic images seems to lead to
algebras that are no more complicated than those with which the constructions
started. By themselves, these constructions do not offer the means to form larger,
more elaborate algebras. The direct product construction allows us to construct
seemingly more elaborate and certainly larger algebras from systems of smaller
ones.
Let I be any set and let A, be a set for each i E I. The system A = (A,: i E 1)
is called a system of sets indexed by I. By a choice function for A we mean a
function f with domain I such that f(i) E Ai for al1 i E I. The direct product of the
system A is the set of al1 choice functions for A. The direct product of A can be
designated in any of the following ways:
Each set Ai for i E I is called a factor of the direct product. For each i rs 1, the ith
projection functiori, denoted by pi, is the function with domain n A such that
pi(f)= f(i), for al1 f E n A . Sometimes we refer to members of n A as I-tuples
from A, and we write fi in place of f(i). Observe that if A, is empty for some i E I ,
then n~ is empty. Also note that if I is empty, then n A has exactly one element:
the empty function. If A, = B, for al1 iEI, then n~
is also denoted by B' and
referred to as a direct power of B. In the event that I = (0,1), we use A, x A, to
denote n A .
Now let I be a set and let A, be an algebra for each i E 1. Moreover, suppose
that A, and A, are similar whenever i, j E I . SO (Ai: i E I ) is a system of similar
algebras indexed by 1. We create the direct product of this system of algebras by .
imposing operations on n A coordinatewise. This is the unique choice of opera-
tions on the product set for which each projection function is a homomorphism.
DEFINITION 1.5. Let A = (A,: i~ I ) be a system of similar algebras. The
direct product of (A,: i E 1 ) is the algebra, denoted by n A , with the same simi-
larity type, with universe n A such that for each operation symbol Q and al1
f O, f ', f'-'E n A , where r is the rank of Q,
m ,
( e n A (f O, f l,. ,f '-l)), = ~ * i ( ~ .,~Jr-l)

, ~ ~ ,
for al1 i E I .
Here are some alternatives for denoting direct products:
1.2 Subalgebras, Homomorphisms, Direct Products 23
If B = A, for al1 i E 1, we write Br for the direct product and call it a direct power
of B. In case I = {O, 11, we write A, x A, in place of n A .
Throughout this book we have frequent need to write expressions like
where Q is an operation symbol (or a more complicated expression) of the algebi-a

A with rank r and a,, a,, - a,-, E A. Very often the exact rank of Q is of little
m.,
significance and the expression above is needlessly complex. We replace it by
where ü stands for a tuple of elements of A of the correct length.

The formation of homomorphic images, of subalgebras, and of direct prod-
ucts are the principal tools we will use to manipulate algebras. Frequently, these
tools are used in conjunction with each other. For example, let R denote the ring
of real numbers and let I denote the unit interval. Then IW1 is the ring of al1 real-
valued functions on the unit interval. Going a step further, we can obtain the
ring of al1 continuous real-valued functions on the unit interval as a subalgebra
of m'.
Let X be a class of similar algebras. We use the following notation:
H ( X ) is the class of al1 homomorphic images of members of X
S(%) is the class of al1 isomorphic images of subalgebras of members of X
P ( X ) is the class of al1 isomorphic images of direct products of systems of
algebras belonging to X
We say that X is closed under the formation of homomorphic images, under the
formation of subalgebras, and under the formation of direct products-provided,
respectively, that W(X) E X , S ( X ) c X , and P ( X ) E jy: Observe that if X is
closed with respect to direct products, then X contains al1 the one-element
algebras of the similarity type, since, in particular, X must contain the direct
product of the empty system of algebras.
Let X be a class of similar algebras. We call X a variety if and only if X
is closed under the formation of homomorphic images, of subalgebras, and of
direct products (i.e., H ( X ) E X , S ( X ) c X , and P ( X ) c S ) . Al1 of the classes
described at the close of the last section are varieties. Varieties offer us a means
to classify algebras (that is, to organize them into classes) that is compatible with
our chief means for manipulating algebras. The notion of a variety will become
one of the central themes of these volumes.
Exercises 1.6
1. Let A and B be algebras. Prove that
hom(A, B) = (Sub B x A) í l {h: h is a function from A into B).
2. Let A = (A,: i~ I) be a system of similar algebras. Prove that pi is a
homomorphism from n A onto Ai for each i E l.
3. Let A*= ( A i : i E 1) be a system of similar algebras and assume that B is an

algebra of the same type with B = n~.
from B onto Ai for each i E l , then B = n A .
Prove that if pi is a homomorphism
4. Let A = ( A i : i E I ) be a system of similar algebras. Let B be an algebra of the

same type and hi be a homomorphism from B into Ai for each i~ l . Prove
that there is a homomorphism g from B into n A such that hi = pi 0 g for each
i ~ l .
5. Prove that every semigroup is isomorphic to a semigroup of functions from
X into X, where the operation is composition of functions and X is some set.
6. Prove that every ring is isomorphic to a ring of endomorphisms of some
Abelian group. [Hint: Let A be an Abelian group. The sum h of endomor-
phisms f and g of A is defined so that
h(a) = f(a) + g(a) for al1 ~ E A ,
where + is the basic binary operation of A. The product of a pair of endo-
morphisms is their composition.]
7. Describe al1 the three-element homomorphic images of (m, +), where m is
the set of natural numbers.
1.3 GENERATION OF SUBALGEBRAS
Let A be an algebra and let X be an arbitrary subset of the universe A of A. X

is unlikely to be a subuniverse of A, since quite possibly there is a basic operation
that, when applied to certain elements of X, produces a value outside X. So X
may fail to be a subuniverse because it lacks certain elements. As a first step
toward extending X to a subuniverse, one might gather into a set Y al1 those
elements that result from applying the operations to elements of X. Then X U Y
is no longer deficient in the way X was. The new elements from Y, however, may
now be taken as arguments for the basic operations, and X U Y may not be closed
under al1 these operations. But by repeating this process, perhaps countably
infinitely many times, X can be extended to a subuniverse of A. With respect
to the subset relation, this subuniverse will be the smallest subuniverse of A
that includes X, since only those elements required by the closure conditions in
the definition of subuniverses are introduced in the process. The subuniverse
obtained in this way must be included in every subuniverse of which X is a subset.
Thus it may be obtained as the intersection of al1 such subuniverses. The finitary
character of the fundamental operations of the algebra ensures that this iterative
process succeeds after only countably many steps.
THEOREM 1.7. Let A 'be un algebra and let S be any nonempty collection of
subuniverses of A. Then O S is a subuniverse of A.
Proof. Evidently OS is a subset of A. Let F be any basic operation of A and

suppose that r is the rank of F. To see that OS is closed under F, pick any a,,
1.3 Generation of Subalgebras 25
al, -,
a,. E n S . For al1 B E S we know that a,, a,, - ., a,-, E B; but then
n~
S * . ,
F(a0, a,,.. a,-,) E B, since B is a subuniverse. Therefore F(ao,a,, . ,a,-,) E

m ,
and ( ) S is closed under F. m
DEFINITION 1.8. Let A be an algebra and let X E A. The subuniverse of A

generated by X is the set O{B: X c B and B is a subuniverse of A}. sgA(x)
denotes the subuniverse of A generated by X.
Since X E A and A is a subuniverse of A, Theorem 1.7 justifies calling

sgA(x)a subuniverse of A.
Now we can formalize the discussion that opened this section.
THEOREM 1.9. Let A be un algebra and X G A. Define X,, by the following

recursion:
X, =X
X,,,, = X,, U {F(¿i):.F is a basic operation of A and a is a tuple from X,,}.
Then s$(x) = U{X,,: n E w}
Proof. The proof consists of two claims.

CLAIM O: sgA(x)
E U{X,,: n ~ w ) .
Since X c U{X,,: n E m}, we need only show that U{X,: n~ w)is a subuniverse.
Let F be a basic operation and let ¿be ia tuple from U{X,: n E a ) . Since F has
some finite rank and Xo c X, E - , we can easily see that a is a tuple from X,
for some large enough m. But then F(ü) E X,,, c U{X,,: n E o}. So this latter set
is a subuniverse, as desired.
CLAIM 1: U{X,,: n E w} E sgA(x).
Since sgA(x)is the intersection of al1 subuniverses that include X, it suffices to
show that X,, c B for every subuniverse that includes X and for every natural
number n. This can be immediately accomplished by induction on n. m
COROLLARY 1.10. Let A be un algebra and X G A. If a E sgA(x),then there

is a finite set Y such that Y G X and a E sgA(y).
Proof. We will prove by induction on n that

* If a EX,,, then a E sgA(y)for some finite Y c X.
Initial Step: n = O. Take Y = {a).
Inductive Step: n = m + 1, and we assume without loss of generality that

a = ~ ( bwhere
) F is a basic operation and b is a tuple from X,. Letting Y be the
union of the finite sets obtained by applying the inductive hypothesis to each
element of 6, we see that b is a tuple from s$(Y). Thus a E sgA(y)as desired.
C O R O L L A R Y 1.11. Let A be un algebra and X, Y E A. Then

i. X E sgA(x).
ii. sgA(sgA(x)) = sgA(x).
iii. If X S Y , then sgA(x)S sgA(y).
iv. sgA(x)= U{sgA(z):Z E X and Z is finite).
The properties of sgA,considered as a unary operation on the power set of

A , which have been gathered in this last corollary, are so frequently used that
usually no reference will be given. Subuniverses of the form sgA(Z),where Z is
finite, are said to be finitely generated. Part (iv) of this corollary entails that the
universe of any algebra is the union of its finitely generated subuniverses.
The set-inclusion relation is a partial order on the collection of al1 sub-
universes of A. This order induces lattice operations of join and meet on the
collection of al1 subuniverses. Some of the fundamental facts concerning this
order are easily deduced. They have been gathered in the next corollary.
C O R O L L A R Y 1.12. Let A be any algebra, S be any nonempty collection of

subuniverses of A , and B be any subuniverse of A. Then
i. With respect to set-inclusion, n S is the largest subuniverse included in each ,
member of S.
ii. With respect to set-inclusion, s g A ( U s )is the smallest subuniverse including
each member of S.
iii. B is finitely generated if and only if whenever B c sgA(U T) for any set T
of subuniverses of A , then B c s g A ( U ~ for' ) some finite T' c T.
iv. Suppose that for al1 B, C E S there is D E S such that B U C E D. Then U S is
a subuniverse of A. E
Parts (i) and (ii) describe, respectively, the meet and join in the lattice of
subuniverses. In fact, they describe how to form meets and joins of arbitrary
collections of subuniverses rather than just the meets and joins of subuniverses
two at a time. The import of (iii) is that the notion of finite generation, which on
its face appears to be something "internal" to a subuniverse, can be characterized
in terms of the order-theoretic properties of the set of al1 subuniverses. This last
corollary can be deduced from the preceding material with the help of only the
following fact:
The subuniverses of A are precisely those subsets X of A such that
X = sgA(x).
Suppose B and C are any two subuniverses of A. We define the join of B and C
(denoted B v C) by
and the meet of B and C (denoted B A C) by

1.4 Congruence Relations and Quotient Algebras 27
It is not hard to prove, using the last corollary, that the collection of al1
subuniverses of A endowed with these two operations is a lattice. We cal1 this
lattice the lattice of subuniverses of A and denote it by Sub A.
Exercises 1.13
l. Prove that every subuniverse of (o, +) is finitely generated, where o is the
set {O, 1,2, - - 1 of natural numbers.
2. Supply proofs for Corollaries 1.11 and 1.12.
3. A collection C of sets is said to be directed iff for al1 B, D E C there is E E C
such that B c E and D E E. C is called a chain of sets provided c is a linear
ordering of C. Let A be an algebra. Prove that the following statements are
equivalent:
i. B is a finitely generated subuniverse of A.
ii. If C is any nonempty directed collection of subuniverses of A and B c UC,
then there is D E C such that B E D.
iii. If C is any nonempty chain of subuniverses of A and B E UC, then there
is D E C such that B S D.
iv. If C is any nonempty directed collection of subuniverses of A and B = C, U
then B E C.
v. If C is any nonempty chain of subuniverses of A and B = UC, then B E C.
(HINT: It may help to prove first that every infinite set M is the union of a
chain of its subsets, each of which has cardinality less than the cardinality 1 MI
of M. Zorn's Lemma or some other variant of the Axiom of Choice would be
useful at this point.)
4. An algebra A is called mono-unary if it has only one basic operation and that
operation is unary. Prove that any infinite mono-unary algebra has a proper
subalgebra.
1.4 CONGRUENCE RELATIONS AND

QUOTIENT ALGEBRAS
Unlike the formation of subalgebras, the formation of homomorphic images of
an algebra apparently involves externa1 considerations. But there is a notion
of quotient algebra that captures al1 homomorphic images, at least up to iso-
morphism. The constructions using normal subgroups and ideals familiar from
the theories of groups and rings provide a clue as to how to proceed in the general
setting.
Let h be a homomorphism from A onto B. Define
O = ((a, a'): a, a ' € A and h(a) = h(af)).
So 8 is a binary relation on the universe of A. It is convenient to write a O a' in
place of (a, a') E O. Now 8 is easily seen to be an equivalence relation on A, since
h is a function with domain A. Because h is a homomorphism, O has an additional
property called the substitution property for A:
Let F be any basic operation of A and let a,, b,, a,, b,, E A.
If a, 6 bi for al1 i less than the rank of F, then F(a,, a,, - .) 6 F(b,, b, , S).
We cal1 the relation 6 just described the kernel of h and denote it by ker h.
DEFINITION 1.14. Let A be an algebra. By a congruence relation on A we

mean an equivalence relation on the universe of A that has the substitution
property for A. Con A denotes the set of al1 congruence relations on A.
The kernels of homomorphisms are always congruence relations.

Now congruence relations on A, being special kinds of equivalence relations,
induce partitions of A. Let 6 be a congruence relation on the algebra A. We use
the following notation:
ale= { b : a 6 b a n d b ~ A }f o r a l l a ~ A .
A/6 = {a/6: a € A}.
For a E A, the set a16 is called the congruence class of a modulo 6. A/6 is the
partition of A into congruence classes modulo 6. There is a natural map g, called
the quotient rnap, from A onto A16 defined by
g(a) = a/6 for al1 a E A.
We can impose basic operations on A/0 in such a way that the quotient map
becomes a homomorphism. Let F be a basic operation of A and let r be its rank.
We will define an operation F, on A16 that will correspond to F. The following
condition is crucial if g is to be a homomorphism:
for al1 a,, a,, . . - EA. This looks very much like an adequate definition of F,,
except that the elements a,, a,, seem to play a rather privileged role on
the left-hand side, whereas g(a,) = a,/@, g(a,) = a1/6, - - of the right-hand side
are congruence classes that are represented, as it were, accidently by a,, a,, s.
But the substitution property is exactly the statement that any other choice of
representatives would lead to the same value of either side. Thus the equation
displayed above can be used as a definition of an operation on A/6.
DEFINITION 1.15. Let A be an algebra and 6 be a congruence relation on

A. The quotient algebra A/6 is the algebra similar to A with universe Al6 in which
Q t is the interpretation of Q, for each operation symbol Q.
Since the congruence 6 is obviously the kernel of the quotient map from A
onto A/6, we conclude that the congruence relations on A are exactly the kernels
of the homomorphisms with domain A. This begs the question of the connection
between B and A/6 when 6 = ker h and h: A -+, B. The answer is contained in
the next theorem, where we record the result of this discussion as well.
THE HOMOMORPHISM THEOREM (1.16). Let A and R be similar alge-

bras, let h be a homornorphism from A onto B, and let 6 be a congruence relation
1.4 Congruence Relations and Quotient Algebras
on A and g be the quotient map from A onto A/8. Then

i. The kernel of h is a congruence relation on A.
ii. The quotient map g: A -» A/8 is a homomorphism from A onto A/8.
iii. If 8 = ker h, then the unique function f from A/tl onto B satisfying f o g =h
is an isomorphism from A/8 onto B.
A
ker h is
represented
as a partition
\
the quotient
Figure 1.1
Proof. As the various definitions were virtually designed to make (i) and (ii)
true, we will look only at (iii). Since we want f o g = h and since g is the quotient
map, the only option is to define f by
f (a/8) = h(a) for al1 a E A.
To see that this definition is sound, suppose that a 8 a'. Then h(a) = h(af),since
8 = ker h. .Thus f(a/8) = h(a) = h ( a f )= f(af/8),as desired. f is one-to-one, since
f (ale) = f (a'/O) implies h(a) = h(a' )
and, as ker h = 0, we have a/8 = a'/@. Finally, to demonstrate that f is a
homomorphism, let Q be an operation symbol and ü be a tuple from A. Then
f ( ~ e ( a i e=
) )f ( ~ e ( ~ ( ü ) ) )
=f (g(eA(a3))
=h(eA(4)
= eB(h(a))
= eB(f( ~ ( 3 ) ) )
= QB(f (ale)).
Thus, f is a homomorphism and hence an isomorphism from A/B onto B. U
1.4 Congruence Relations and Quotient AIgebras 31
So the congruence relations on the ring of integers are exactly the relations =,
where q is a non-negative integer. The homomorphic images of Z are, up to
isomorphism, Z itself, the one element ring, and the rings of residues modulo
integers greater than l.
2. Let R = (R, A , v ) where R is the set of real numbers and
R is a lattice. We wish to describe al1 the congruence relations on R. So let 0 be

a congruence relation. Suppose r 0 s and r 5 t _< s. Then r = (r A t) 0 (S A t) = t,
and so r 0 t. This means that the congruence classes of 0 are convex-that is, they
are intervals, perhaps infinite or even degenerate. Pick an arbitrary element from
each congruence class. It is evident that the set of selected elements forms a
subalgebra of IW isomorphic to R/O.
Now let O be any equivalence relation on R such that each O-equivalence
class is a convex set of real numbers. To verify the substitution property, let r 0 r'
and S 0 S'.
CASE O: r Os.
The substitution property is immediate, since r A S,r v S,r' A S', and r' v S'
al1 belong to {r, S,r', S') E rl0.
CASE 1: r and s lie in different equivalence classes modulo 0.
The two 0-classes, which are intervals, cannot overlap. Thus, without loss
of generality, we assume that every element of 1-16is less than every element of
s/o. Hence
Therefore (r A S)0 (r' A S') and (r v S)O (r' v S'), as desired, so 0 is a congruence
relation.
Thus the congruence relations of R are exactly those equivalence relations
whose equivalence classes are convex sets of real numbers. Since any proper
convex subset of [W is a congruence class of infinitely many congruence relations,
IW provides a strong contrast with the behavior of congruence relations on
groups.
For most algebras, the task of describing al1 the congruence relations is
hopelessly difficult, so these two examples have a rather special character. The
collection of al1 congruence relations of an algebra is a rich source of information
about the algebra; discovering the properties of this collection often leads to a
deeper understanding of the algebra.
Just as the subuniverses of an algebra form a lattice, so do the congruence

relations. Roughly the same analysis can be used. Let A be an algebra and let X
be a binary relation on A. Now X may fail to be a congruence relation, either
because it is not an equivalence relation on A or because it does not have the
substitution property. In either case, the failure can be traced to the existence of
an ordered pair that fails to belong to X but that must belong to any congruence
relation that includes X. Al1 such necessary ordered pairs can be gathered into
a set Y, and X U Y is at least not subject to the same deficiencies as X. Yet X U Y
may fail transitivity or the substitution property. But by repeating the process,
perhaps countably infinitely often, a congruence relation will be built; with
respect to set inclusion, it will be the smallest congruence relation on A that
includes X.
THEOREM 1.18. Let A be un algebra and C be any nonempty collection of

congruence relations on A. Then n C is a congruence relation on A. ¤
The routine proof of this theorem is left as an exercise. This theorem allows
us to proceed as we did with subuniverses.
DEFINITION 1.19. Let A be an algebra and X EA x A. The congruence

relation on A generated by X is the set
(-){O:X S 0 and 0 is a congruence relation on A}.
c g A ( x )denotes the congruence relation on A generated by X.
As we did with subuniverses, we can formalize the discussion above to obtain

a description of how to extend a binary relation to obtain the congruence relation
it generates. This is complicated by the necessity of arriving at an equivalence
relation. For the purposes of convenience in dealing with congruences, both here
and in general, we introduce some notation. Let A be an algebra and 0 be a binary
relation on A. Let ü and a' be tuples from A of the same length, say r. So
ü = (a,,a,,.-.,a,-,) and ü' = (aó,a;,--.,a:-,).
We use
¿i0 zf
in place of the more elaborate
Thus ü 0 ü' stands for "ai 0 a: for al1 i < r." Using this notation, we can rephrase
the substitution property as:
ü0 2' implies F (a)0 F (a'),for al1 basic operations F and al1
tuples ü and a'.
1.4 Congruence Reiations and Quotient Algebras 33
THEOREM 1.20. Let A be un algebra and X E A x A. Define X, by the

following ~ecursion:
X, = X U {(a, a'): (a', a) E X ) U {(a, a): a € A )
xn+,= xnu T,'JQ,,
where Q, = {(F(ü), F(Ü1)):F is a basic operation and Zand ü' are tuples such that
üX,Ü') and T, = {(a,c): aX,bX,,c for some ~ E A J .
Then c g A ( x )= U
X,. ¤
nEm
The proof of this theorem is much like the proof of Theorem 1.9, the only
new element being an easy argument about transitive closures.
COROLLARY 1.21. Let A be un algebra and X G A x A. If (a, a') E cgA(x),

then there is a finite set Y E X such that (a, a') E cgA(y). ¤
COROLLARY 1.22. Let A be un algebra and X, Y A x A. Then

i. X c c$(x).
ii. cgA(cgA(x))= cgA(x).
iii. If X E then c g A ( x )E cgA(y).
iv. c g A ( x )= U{cgA(Z):Z E X and Z is finite).
Congruente relations of the form cgA(2), where Z is finite, are said to

be finitely generated. Those of the form ~ g ~ ( { ( a , a ' ) ) are
) called principal
congruence relations. This last piece of notation is obviously awkwaid. We
replace it by cgA(a,a'). Evidently, cgA(a,a') is the smallest congruence relation
that places a and a' in the same congruence class-or, as we shall say more
suggestively, cgA(a,a') is the smallest congruence collapsing (a, a').
COROLLARY 1.23. Let A be un algebra, C be any nonernpty collection of

congruence relations on A, and 8 be any congruence relation on A. Then
i. With respect to set-inclusion, n C is the largest congruence relation on A
included in each member of C.
ii. With respect to set-inclusion, c g A ( U c ) is the smallest congruence relation
including each member of C.
iii. 8 is finitely generated fi and only if whenever D is a set of congruences on A
and 8 i c g A ( UD), then 8 E c$(U D') for some finite D' E D.
iv. Suppose that for each #, $ E C there is q E C such that # U S, E q. Then UC
I is a congruence relation on A.
, The proofs of these three corollaries do not differ significantly from the
proofs of the three corollaries to Theorem 1.9.
Suppose 4 and $ are congruence relations on the algebra A. We can define
l
the join (designated by v ) and the meet (designated by A ) of # and S/ so that
i
34 Chapter 1 Basic concepts
4 v $ = CgA(4U $) and 4 A $ = 4 í? $. With these operations, the collection

of al1 congruence relations on A becomes a lattice, which we shall cal1 the
congruence lattice of A and denote by Con A. Every congruence relation of A is
an equivalence relation on A and, as we saw in Exercise 1.2(7), the equivalence
relations on A constitute a lattice Eqv A. In fact, Con A is a sublattice of Eqv A.
The meet operation in both Con A and Eqv A is just set-theoretic intersection.
To establish the contention that Con A is a sublattice of Eqv A, it is necessary to
prove that the join operation in Con A is the restriction of the join operation in
Eqv A. ~ h i is
s the import of the next theorem, and it even applies to joins of
arbitrary subsets of Con A. Note that we write Con A for the set of al1 congruence
relations on A-the universe of the lattice Con A. By the same token, we write
Sub A for the set of al1 subuniverses of A, and Eqv A for the set of al1 equivalence
relations over the set A.
THEOREM 1.24. Let A be un algebra and let C G Con A.

i. Con A = (Sub A x A) í l Eqv A.
ii. c g A ( U c ) =~ { R : U C G R U ~ ~ R E E ~ V A ) .
Proof.
i. This is just a restatement of the definition of a congruence relation, using
the language of subuniverses and direct products.
n
ii. Let 0 = {R: U C c R and R E Eqv A). Thus 0 is the smallest equivalence
relation on A that includes UC. Since it is clear that U C c 0 c c g A ( U c ) ,we
need only establish that 0 is a congruence on A. In view of part (i) of this theorem,
it only remains to establish that 0 is a subuniverse of A x A.
The transitive closure of relations was described in the Preliminaries. A more
detailed analysis is used here. Let D be any collection of relations on A-that
is, subsets of A x A. Let D* denote the set of al1 those relations that can
be obtained as compositions (i.e., relational products) of finite nonempty
sequences of relations belonging to D. Thus 4, o 4, o - - o &,, where 4i E D
for each i < n is a typical element of D*. Checking that UD* is actually the
transitive closure of UD presents no difficulties. This analysis of the transi-
tive closure leads immediately to the following conclusion: If D consists
entirely of symmetric reflexive relations on A, then the transitive closure of
UD is also symmetric and reflexive and is, therefore, the smallest equivalence
U U
relation including D. In particular, 0 is the transitive closure C* of UC.
Now consider any two subuniverses 4 and $ of A x A. # o $ must be
a subuniverse of A x A as well, since if F is any basic operation of A and n
is the rank of F and if ai4bi$ci for al1 i < n, then
By the obvious inductive extension of this fact, if D consists entirely of sub-

universes of A x A, then so does D*. Moreover, if every member of D is
1.4 Congruence Relations and Quotient Algebras 35
reflexive, then D* is directed upward by set-inclusion (see Exercise 1.13(3)).

In particular, this means that C* is a collection of subuniverses of A x A
and that for any # and S in C* there is y in C* such that # U c y. Hence,
by Corollary 1 . 2 ( v ) U C * is a subuniverse of A x A. That is, 8 is a
subuniverse of A x A, as desired. •
Let A be any algebra. With A, we can associate four other algebras: End A,
which is the monoid of al1 endomorphisms of A; Aut A, which is the group of al1
automorphisms of A; Sub A, which is the lattice of al1 subuniverses of A; and
Con A, which is the lattice of al1 congruence relations on A. Chapter 2 is devoted
to the rudiments of the abstract theory of lattices, and Chapter 3 takes up some
aspects of the theories of monoids and of groups. These four algebras related to
A contain significant information about A.
Exercises 1.25
1. Let h E hom(A, 13). Prove that ker h is a congruence relation on A.
2. Let 0 E Con A. Prove that the quotient map from A onto A10 is a homo-
morphism from A onto A/8 and that its kernel is O.
3. Let G be a group, 0 E Con G, and N be a normal subgroup of G. Prove that
el8 is a normal subgroup of G, where e is the unit of the group. Prove that
({a, b): a . b-' E N) is a congruence relation on G. Finally, prove that if
4E Con G, then q,5 = 8 iff el4 = e/8.
4. Verify that =, is a congruence relation on the ring H of integers for every
natural number q.
5. Suppose 0 E Con A. Prove that 8 = U{cgA(a,a'): a 8 a'}. 1s every sub-
universe the join of 1-generated subuniverses?
6. Let A be an algebra and h ~ h o m ( A , A ) Prove
. that h is one-to-one iff
ker h = O*, where OA denotes the least congruence relation on A, namely the
identity relation ((a, a): a E A}.
7. Let A be an algebra. Define F to be the function with domain End A such that
F(h) = h-' o h for al1 h E End A.
Prove that F maps End A into Con A.
"8. (Burris and Kwatinetz) Let A be an algebra that is countable (i.e., finite or
countably infinite) and has only countably many basic operations. Prove
each of the following:
i. IAutAl < o or IAutAl = 2 "
ii. JSubAJ_ < o or ISubA( = 2 "
iii. IEndAl < o or IEndA( = 2 "
iv. JConAl < o or IConAl=2"
where is the cardinality of the set of natural numbers and 2" is the
cardinality of the set of real numbers.
C H A P T E R T W O
Lattices
2.1 FUNDAMENTAL CONCEPTS
Lattices arise often in algebraic investigations. We have already seen that Sub A
and Con A are lattices for every algebra A. Knowing the significance of normal
subgroups in group theory and ideals in ring theory, we should not be surprised
that lattices of congruences have an important role to play. By itself, this would
justify a detailed development of lattice theory. It turns out that, in addition to
congruence lattices, many other lattices prove useful in developing a general
theory of algebras.
This chapter is an introduction to lattice theory, focusing on results that will
be put to use in these volumes. 594.6 and 4.8 further elaborate some aspects of
lattice theory introduced here. Lattice theory is a rich subject in its own right.
We can highly recommend Birkhoff [1967], Crawley and Dilworth [1973], and
Gratzer [1978] for fuller expositions of lattice theory.
Lattices were defined in Chapter 1 as algebras of the form (L, A , v ), with
two binary operations called meet (designated by A ) and join (designated by v ),
for which the following equalities are true for al1 a, b, c E L :
The equalities on the first line express commutativity, those on the second line
associativity, those on the third line idempotency, and those on the last line
absorption. These equalities are called the axioms of lattice theory. An alternative
system of notation in common use denotes join by " +" and meet by "." (or
simply by juxtaposition).
Lattices can also be viewed as special ordered sets. Let L be a nonempty set
and 5 be a binary relation on L. The relation < is said to be an order on A if
and only if for al1 a, b, c E L
i. (Reflexivity) a 5 a;
ii. (Anti-symmetry) If a i b and b 5 a, then a = b;
iii. (Transitivity) If a I b and b I c, then a i c.
Orders have frequently been referred to as partial orders in the literature. Let I
be an order on L and let X c L. An element a E L is called an upper (lower) bound
of X if and only if x I a (a I x) for al1 x E X ; a is called a least upper bound of X
if and only if a is an upper bound of X and a I b for al1 upper bounds b of X.
Dually, a is called a greatest lower bound of X if and only if a is a lower bound
of X and b 5 a for al1lower bounds b of X. Since I is anti-symmetric, least upper
bounds and greatest lower bounds are unique, when they exist. The least upper
bound of X is also called the supremum of X and is denoted by sup X;the greatest
lower bound of X is also called the infimum of X and is denoted by inf X.
DEFINITION 2.1. Let L be a nonempty set. A lattice order on L is an order

on L with respect to which every subset of L with exactly two elements has both
a least upper bound and a greatest lower bound. The relational structure (L, _< )
is called a lattice ordered set if i is a lattice order on L.
Each lattice has an underlying lattice order, from which the lattice operations
of join and meet can be recovered. More precisely, let L be the lattice (L, A , v )
and define Lo to be (L, 1) where i is the binary relation on L specified, for
al1 a, b EL, by
a i b if and only if a = a A b.
Routine calculations reveal that 5 is a lattice order on L. Now suppose L is
some lattice ordered set (L, 5 ) and define La to be (L, A , v ) where the two
binary operations are specified, for al1 a, b E A, by
a A b = inf {a, b )
a v b = sup{a, b).
Again, routine calculations reveal that La is, indeed, a lattice. Moreover, Loa= L
for every lattice L and La"= L for every lattice ordered set L.
It is useful to gain some ski11 at manipulating lattice equations and inclusions.
Some of the most frequently used manipulations come up in the following
exercises.
Exercises 2.2
l. Let L be a lattice. Prove that for al1 a, b EL, a = a A b if and only if b =a v b.
2. Verify the claims made above.
i. If L is a lattice, then Lo is a lattice ordered set.
ii. If L is a lattice ordered set, then La is a lattice.
iii. If L is a lattice, then Loa= L.
iv. If L is a lattice ordered set, then La"= L.
38 Chapter 2 Lattices
3. Let L be a lattice and let a, b, c, d E L. Prove that if a < b and c < d, then
a~c<bndandavc<bvd.
4. Let L be a lattice and let a, b, c E L. Prove that
i. a s c a n d b ~ c i f a n d o n l y i f av b < c .
ii. c ~ a a n d bci f~a n d o n l y i f c < a A b.
iii. a A b < a v b.
5. Prove that the two axioms of lattice theory that express idempotency are
derivable from the other six axioms.
6. Let L be a lattice and let a, b, c E L. Prove that
i. (a A b) v (a A c) < a A (b v c).
ii. a v (b A c) 5 (a v b) A (a v c).
iii. (a A b) v (b A c) v (c A a) 5 (a v b) A (b v c) A (c v a).
iv. (a A b) v (a A c) 5 a A (b v (a A e)).
7. Let L = (L, A , v ) be a lattice. Prove that (L, A ,*) is a lattice if and only if
a*b = a v bforalla,b~L.
Let L be a lattice, or more generally, an ordered set. For a, b EL, we say that
b covers a (and that b is an upper cover of a and that a is a lower cover of b), and
we write a < b if and only if a < b and (c: a < c < b, c EL} is empty. L is said
to be bounded provided that L has both a greatest element and a least element.
We use 1 to denote the greatest element of L and O to denote the least element
of L, whenever they exist. If L has a least element O, then the upper covers of O
are called the atoms of L. Dually, if L has a greatest element 1, then the lower
covers of 1are called the coatoms of L. Elements a, b E L are said to be comparable
if a < b or b S a; a[lb denotes that a and b are incomparable. A subset of L in
which any two elements are comparable is called a chain, whereas a subset in
which no two elements are comparable is called an antichain.I [a, b] denotes the
interval from a to b-that is, the set (c: c E L and a 5 c < b}. We also use I(a]
to denote (c: c E L and c < a} and I[a) to denote (c: c E L and a S c}.
By using the covering relation, it is possible to draw diagrams of finite
lattices and of certain infinite lattices. The practica1 usefulness of these diagrams
is great, and the reader is encouraged to draw lattice diagrams whenever that
may seem helpful. A Hasse diagram of the lattice L is obtained by arranging the
elements of L as points on a plane in such a way that if a < b, then the point
representing b is above the point representing a. Then a line segment is drawn
between any two points that constitute a covering in L. For those lattices for
which the lattice ordering is the transitive closure of the covering relation (this
includes al1 finite lattices), a Hasse diagram completely determines the lattice.
The ability to draw revealing Hasse diagrams of lattices and other ordered sets
is an acquired skill. A lattice does not have to be very large before many strikingly
different ways of drawing its Hasse diagrams become available. Generally speak-
ing, a diagram is most useful when it is spread out and reduces line crossings
to a minimum. Figure 2.1 consists of just a few of the lattice diagrams we will
use at various places in these volumes. One of the most elaborate lattice diagrams
in this work is given in the second volume, where the bulk of two sections is
essentially devoted to providing instructions for the drawing of the diagram.
Fd+(3)
Figure 2.1
Some caution needs to be exercised with Hasse diagrams. Figure 2.2 is a

diagram that looks very much like a lattice diagram but is not:
Figure 2.2
It is apparent that when a lattice diagram is turned upside down, another

lattice diagram is obtained. Considering the definitions, we see that I becomes
2 and that A and v have been interchanged. Of course, this is a consequence
of the obvious symmetry of the axioms of lattice theory. More formally, we have
the principle of duality for lattices:
If (L, A , v ) is a lattice, then (L, v , A ) is a lattice.
If (L, 5 ) is a lattice ordered set, then (L, 2: ) is a lattice ordered set.
As a consequence, if o is a statement that is true in every lattice and a ' is
the statement obtained from a by interchanging 5 and 2: and interchanging
A and v , then o' is true in every lattice.
If L = (L, A , v ) is a lattice, then L~is the lattice (L, v , A ),and it is called
the dual of L. Similar notation applies to lattice ordered sets.
Now let L = (L, 5 ) and Lr = (L', 2') be two lattice ordered sets. A
function f from L into L' is said to be isotone or order preserving if and only
if, for al1 a, b E L, a 5 b implies f(a) sff(b). It is easy to check that every
homomorphism between two lattices is isotone. But not every isotone map is a
homomorplnism, as Figure 2.3 reveals.
Figure 2.3
2.1 Fundamental Concepts 41
On the other hand, if h is a one-to-one isotone map from L onto L', and h-l
is also isotone, then h is an isomorphism from the lattice L onto the lattice L'.
It is also important to realize that while I may be a lattice order on L and
L' may be a subset of L on which I induces a lattice order, it can happen that
L' is not a subuniverse of the lattice (L, A , v ). This phenomenom can be traced
to the global nature of the definition of join and meet in terms of the ordering.
Figure 2.4 is an example. L' consists of the points denoted by *.
Figure 2.4
Exercises 2.3
1. Draw a Hasse diagram for the lattice of al1 subgroups of the symmetric group
on (O, 1,2).
2. Draw the Hasse diagram of the lattice of al1 subsets of the set {O, 1,2,3).
3. Prove that every isotone one-to-one function from a lattice L onto a lattice
L' that preserves incomparability is an isomorphism. Provide an example to
show that an isotone one-to-one function from a lattice L into a lattice L'
need not be a homomorphism.
In the study of lattices, it is helpful to single out individual elements of

lattices that have special properties with respect to the ordering or the basic
operations.
DEFINITION 2.4. Let L be a lattice and a E L. a is join irreducible iff a = b v c

always implies a = b or a = c. Dually, a is said to be meet irreducible iff a = b A c
always implies a = b or a = c. J(L) denotes the set of al1join irreducible elements
of L and M(L) denotes the set of al1 meet irreducible elements of L. a is said to
be join prime iff a Ib v c always implies a I b or a I c. Dually, a is meet prime
iff a 2 b A c always implies a 2 b or a 2 c.
In the lattice N, (see Figure 2.1), every element is either join prime or meet
prime. In M, only 1 is meet prime and only O is join prime, but every element is
either join irreducible or meet irreducible. These four properties of an element
are preserved in passing to a sublattice that contains the element. Every join
prime element is join irreducible and every meet prime element is meet irreducible.
THEOREM 2.5. Let L be afinite lattice. The following conditions are equivalent:
i. The two-element lattice is a homomorphic image of L.
ii. L has a join prime element differentfrom O.
iii. L has a meet prime element differentfrom 1.
Proof. Since (i) is a selfdual property and (ii) is the dual of (iii), we need
only show that (i) and (ii) are equivalent. Let C, be the two-element lattice
with C, = (0,l) and O < 1. Suppose that (i) holds and that h: L ++ C,. Let
h-'(1) = ( a o , a l , . . . , a n )and set
a = a, A a, A - S - A a,.
So h(a) = h(a,) A h(a,) A A h(an)= l. To see that a is join prime, suppose
a 5 b v c. Then h(a) 5 h(b) v h(c) and so 1 = h(b) v h(c). Since h(b), h(c) E (O, 11,
we see that either h(b) = 1 or h(c) = 1. Thus either a < b or a 5 c, and a is join
prime. a # O, since h is onto C,.
For the converse, suppose that a is a nonzero join prime element of L.
Define h: L ++ C, by: h(u) = 1 iff a 5 u. Then, since a is join prime, h-'(O)
is closed under join. It is clear that h-'(1) is closed under meet. Moreover,
if h(u) = O and h(v) = 1, then a I u v v and a $ u A v, so h(u v u) = 1 and
h(u A u) = O. Thus h is a homomorphism from L onto C,, as desired. E
The condition in the previous theorem that L be finite cannot be omitted.

In the demonstration that (i) 3 (ii), it played a crucial role. The condition can
be weakened. The ascending chain condition holds for the lattice L provided
L has no sublattice isomorphic to the lattice of natural numbers under their
usual ordering-that is, provided that every ascending chain in L is finite.
The dual property is referred to as the descending chain condition. These con-
ditions have been applied to the ideal lattices of rings.
THEOREM 2.6. The following conditions are equivalent for any lattice L.
i. Every nonempty subset of L has a maximal element.
ii. L satisfies the ascending chain condition.
Dually, the following conditions are equivalent for any lattice L.
i'. Every nonempty subset of L has a minirnal element.
1
ii'. L satisfies the descending chain condition. 1
Proof. In view of the duality, we need only prove that (ir)is equivalent to (ii').
( i ) ( i ) To argue the contrapositive, just notice that any sublattice of L
ordered like the negative numbers has no smallest element.
( i r ) (ir) Again, we argue the contrapositive. Let X be a nonempty subset
of L without minimal elements. According to the Axiom of Choice, there is a
choice function F for the collection of nonempty subsets of X (i.e., F(Y) E Y for
every nonempty Y c X). Since X has no minimal elements, {x: x EX,and x < z }
2.1 Fundamental Concepts 43
is nonempty, for al1 z E X . This allows us to define a function g from the set of
natural numbers into X by the following recursion:
g(O) = F ( X )
g(n + 1 ) = F ( ( x :x E X and x ig ( n ) ) )for al1 natural numbers n.
Evidently, (g(n):n is a natura-l number) is a subset of L ordered like the negative
integers. m
The theorem above, which is actually very straightforward, nevertheless

invokes the Axiom of Choice. One more or less immediate consequence of this
theorem is that the descending chain condition is sufficient for the representation
of elements of a lattice as joins of finitely many join irreducible elements.
T H E O R E M 2.7. If L is a lattice with the ascending chain condition, then every

element of L is a meet of finitely many meet irreducible elements; dually, if L is a
lattice with the descending chain condition, then every element of L is a join of
finitely many join irreducible elements.
Proof. Suppose that L has the descending chain condition. Let X be the set of
al1elements of L that cannot be written as the join of finitely many join irreducible
elements. If X is nonempty, then it would have a minimal element x. In this case,
[
x cannot be join irreducible, so there are elements y and z, each properly less
than x, such that x = y v z.Since y and z are both properly less than x, they are
not in X. Consequently, y and z can be expressed as joins of jo,in irreducible
elements. Thus x can be expressed in the same way. But this means that x cannot
belong to X. Thus the supposition that X is nonempty is not tenable. So every
element of L can be expressed as the join of join irreducible elements. E
This theorem resembles the familiar theorem of arithmetic that every natural
number can be written as the product of prime numbers. Here, multiplication is
replaced by join and primeness is reeaced by join irreducibility. Actually, the
connection is closer than it appears at first. The set of natural numbers, endowed
with the operations of forming greatest common divisors and least common
multiples, is a lattice with the descending chain condition. The join irreducible
elements in this lattice are the powers of prime numbers. Of course, a powerful
aspect of factorization of numbers into primes is the uniqueness of the factori-
zation. For lattices in general, there may be elements that can be expressed as
the join of join irreducible elements in many different ways. Later in this chapter,
we will return to this topic and demonstrate that uniqueness can be obtained for
some interesting classes of lattices.
Exercises 2.8
l. Construct a lattice that has the two-element lattice as a homomorphic image
but has no join prime elements.
2. Prove that a lattice with the ascending chain condition has a meet prime
element different from 1 iff it has the two-element lattice as a homomorphic
image.
3. Construct a finite lattice that has an element that can be expressed as the join
of join irreducible elements in two distinct ways.
4. Prove that if L satisfies the ascending chain condition, then every chain in L
has a largest element.
2.2 COMPLETE LATTICES AND CLOSURE SYSTEMS
The lattices that prove most significant in the development of the general theory
of algebras are congruence lattices, lattices of clones, subuniverse lattices, lattices
of equational theories (and their duals, the lattices of varieties), and interpretability
lattices. We have already introduced congruence lattices and subuniverse lattices;
the reader will meet the other lattices later in this work. Al1 these lattices have
important properties in common that do not hold for al1 lattices.
A lattice L is said to be complete if and only if every subset of L has both a
least upper bound and a greatest lower bound.
A complete lattice L always has a largest element, usually designated by 1,
which is the least upper bound of L, and a smallest element, usually designated
by O, which is the greatest lower bound of L. If L is a complete lattice and X c L,
we use A X to denote the greatest lower bound of X and VX to denote the least
upper bound of X. (Thus, in the notation introduced just prior to Definition 2.1,
sup X = VX and inf X = Ax.) If X = (xi:i E I } we also write
I
xifor A X and V xifor VX.
I
Theorem 1.7 and Theorem 1.18 have corollaries that assert that Sub A and Con A
are complete lattices, for every algebra A. These two conclusions were established
in virtually the same manner, which we now formalize.
DEFINITION 2.9. F is a closed set system on the set A if and only if

i. F is a collection of subsets of A,
ii. A EF, and
iii. G E F for every nonempty G G F.
The collection of al1 subuniverses of an algebra, the collection of al1 con-
gruence relations of an algebra, the collection of al1 equivalence relations on a
set, and the set of al1 closed subsets of a topological space are al1 examples of
closed set systems.
DEFINITION 2.10. Let A be a set. C is a closure operator on A if and only if

C is a function from the power set of A into the power set of A such that
i. X e C(X) for al1 X c A,

ii. C(C(X)) = C(X) for al1 X c A, and
iii. If X E Y c A, then C(X) C(Y).
sgAand cgA,for any algebra A, are examples of closure operators, as are
the familiar operations of forming topological closure and of forming the closed
convex hull in a topological vector space.
The connection between closed set systems and closure operators is much
like the connection between lattice orderings and lattices, discussed in the
previous section. Given a closed set system on A, one may define a closure
operator on A; given a closure operator on A, one may define a closed set system
on A. Moreover, these two processes are inverses of each other. More precisely,
let F be a closed set system on A. Define the function C on the power set of A by
c K and K E F )
C(X) = ~ { K : X
for al1 X c: A. C turns out to be a closure operator on A. For the reverse
definition, let C be any closure operator on A. Define
F = {C(X): X E A).
In the next set of exercises, the reader is asked to check that F is a closed set
system on A and that the two procedures just described reverse each other. The
distinction between closed set systems and closure operators is one of viewpoint,
not essence.
THEOREM 2.11. Let C be a closure operator on the set A and let F be its closed
set system. Then set-inclusion on F is a lattice ordering with respect to which F
becomes a complete lattice, with the lattice operations defined so that for al1 G c F,
/\G = A, if G is empty,
AG = OG, if G is not empty, and
VG = C ( ~ G ) .
Proof. Evidently, AG is the greatest lower bound of G, with respect to the set-
inclusion relation.
U U
V G is an upper bound of G, since if K E G, then K c G and G c C ( UG),
since C is a closure operator. Now suppose H is a closed set such that K c H
for al1 K E G. Then UG c H. This implies that C(UG) c C ( H ) = H, since C is
a closure operator and H is a closed set. Thus C(UG) is the least upper bound of
G. So set-inclusion is a complete lattice ordering on F, and the lattice operations
are as described.
The converse of this theorem is true as well, at least up to isomorphism, as

shown in Theorem 2.12.
THEOREM 2.12. Every complete lattice is isomorphic to the lattice of al1 closed
sets of some closed set system.
Proof. Let L be a complete lattice. Take F to be the collection of al1 principal

ideals of L-that is, al1 sets of the form:
( a : EL and a Ib ) = D,.
It is easy to check that F is a closed set system and that the function that takes
b to D, for al1 b E L is an isomorphism. w
This theorem is a representation theorem for complete lattices, since it

renders every abstract complete lattice isomorphic to some concrete complete
lattice of closed sets.
The closure operators sgAand cgA,where A is an algebra, were seen in
Corollaries 1.11 and 1.22 to have an additional property: The closure of a set is
the union of the closures of its finite subsets. This property, which is distinctively
algebraic, is not enjoyed by the familiar topological closure operation on the real
line.
DEFINITION 2.13. Let C be a closure operator on the set A. C is said to be

algebraic if and only if C(X) = U{C(Z): Z c X and Z is finite), for al1 X A.
Let be an ordering of the set A. A subset B of A is said to be directed

upward by 5 iff for al1 a, b E B, there is c E B such that a Ic and b c.
THEOREM 2.14. A closure operator is algebraic if and only if the union of any
collection of closed sets that is directed upward by is itself a closed set.
Knowing the close connection between closure operators and the corre-
sponding complete lattices of closed sets, we can hope for an order-theoretic
property corresponding to the property of being algebraic for closure operators.
This property is suggested by the last corollary to Theorem 1.20.
DEFINITION 2.15. Let L be a complete lattice. An element a E L is called

compact if and only if for al1 X c L, if a I VX, then a I V Y for some finite
Y E X. L is said to be algebraic if and only if every element of L is the join of a
set of compact elements of L.
If C is a closure operator and K is a closed set, we say K is finitely generated

iff K = C(Z) for some finite set Z. In the lattice of closed sets arising from an
algebraic closure operator, the formation of the join of an arbitrary collection
M of closed sets can be reduced to the union of the collection of the joins of al1
the finite subcollections of M.
THEOREM 2.16. I f C is un algebraic closure operator, then its lattice of closed

sets is un algebraic lattice, and the compact elements of this lattice are exactly the
finitely generated closed sets. Conversely, every algebraic lattice is isomorphic to
.i '
the lattice of closed sets for some algebraic closure operator.
2.2 Complete Lattices and Closure Systems 47
Proof. Let C be an algebraic closure operator on A. We first argue that the

compact elements of the lattice of closed sets coincide with the finitely generated
closed sets. So suppose that Z is a finite subset of A and let H = C(Z). Let G be
any collection of closed sets such that H c VG. Since C is algebraic and Z is
finite, there is a finite set Y c UGsuch that Z S C(Y). Thus H 5 C(Y) c
C ( ~ G=) V G . Now pick G' c G so that G' is finite and Y c UG'. So H c VG'.
Therefore, every finitely generated closed set is compact. Now let H be any
compact member of the lattice of closed sets. Evidently,
H c V{C(Z): Z c H and Z is finite).
So there are finitely many finite subsets Zo7Z,, m , Zk of H such that
By letting Y = Zo U Z, U U Zk7we see that
Since H is a closed set, we conclude that H = C(Y). Therefore, every compact

member of the lattice of closed sets is finitely generated.
The lattice of closed sets is algebraic, since (for any closure operator) every
closed set is the union of the closures of its finite subsets.
For the converse, suppose that L is an algebraic lattice, and let A be the set
of al1 compact elements of L. For each b E L define
D, = {a: a ~ A a n ad Ib}.
It is easy to check that F = {D,: b E L ) is a closed set system and that the map f
from L onto P defined by f(b) = D,, for al1 b, is an isomorphism of L onto the
lattice F of closed sets (that f is one-to-one follows from the fact that L is
algebraic). To see that the associated closure operator is algebraic, we apply
Theorem 2.14.
Let G = {D,: b EX}be a collection of closed sets directed by E. Notice that
for any elements b, c~ L, we have D, E D, iff c 5 b. So the set X is directed
by 5 . Let g = V X . Now observe
a E Dgiff a is compact and a VX
VX'
iff a is compact and a I for some finite X' cX
iff a is compact and a Ib for some b E X
iff aED, for some EX
iff a E UG.
l So D, = UG, making UG a closed set. Thus the closure operator associated

with F is algebraic. ¤
1
Chapter 2 Lattices
On any lattice L, sgLand c g L are algebraic closure operators. There are

three other algebraic closure operators connected with L, which we introduce
next.
DEFINITION 2.17. Let L be a lattice and U c L. U is said to be an ideal of

L iff U is nonempty, b < a E U implies b E U, and a, b E U implies a v b E U.
Dually, U is a filter of L iff U is nonempty, b 2 a E U implies b E U, and a, b E U
implies a A b E U. U is called convex iff whenever a, c E U and a 5 b < c, then
b~ U.
This definition of ideal has much in common with the definition in ring
theory. However, the correspondence between ideals in rings and congruences
in rings does not carry over to lattices. For example, M, (see Figure 2.1) has five
ideals but only two congruence relations. One intuition about ideals in lattices
is that an ideal specifies a notion of "small element." The members of the ideal
are "small," whereas the members of the lattice that lie outside the ideal are not
"small." For example, in the lattice of al1 subsets of the unit interval, the sets with
Lebesgue measure zero constitute an ideal. Ideals of the form I(a] = (b: b 5 a}
are called principal ideals;dually, filters of the form I [a) are called principal filters.
It is easy to verify that the intersection of any nonempty collection of ideals
of a lattice L is once more an ideal of L or it is empty. Thus the collection
consisting of the empty set and al1 the ideals of L is a closed set system. With
respect to G , we obtain a complete lattice. Actually, the collection of ideals of L
(without the empty set) constitutes a sublattice that can fail to be complete. If L
has a least element, then the ideals of L form a complete lattice. Conversely, if
the ideals of L form a complete lattice, then L will have a smallest ideal, which
is easily seen to be a singleton set whose element must be the least element in
the lattice. We adopt the convention that Id1 L, which we cal1 the lattice of ideals
of L, is the lattice of al1 ideals of L if L has a least element and is the lattice
consisting of the empty set and al1 the ideals of L otherwise. The situation for
.
filters is dual to that for ideals. We adopt a similar convention for Fil L, the lattice
of filters of L. CvxL denotes the lattice of convex sets in L; this lattice always
includes the empty set. I ~ F ~ ~ and , ~ c, v L denote the corresponding closure
operators.
THEOREM 2.18. Let L be a lattice. Id1 L, Fil L, and Cvx L are algebraic lattices
and L is isomorphic to a sublattice of both Id1 L and Fil L ~Moreover,
. if L is finite
then Id1 L E L Fil L ~ The. intersection of any filter on L with any ideal on L is
a convex subuniverse of L, and every convex subuniverse of L arises in this way.
Proof. The following descriptions of I ~ F ~ ~ and

, ~ cvL
, reveal that they are
algebraic closure operators; in view of Theorem 2.16, the corresponding lattices
are algebraic lattices. Let X c L. Then
I ~ ~ (= VY
x )(a: a I for some finite Y X }
F ~ ~ ( x=)(a: /\Y < a for some finite Y X}
c v L ( x )= (c: a 5 c 5 b for some a, b EX}.
I
The reader can easily supply the proofs that these sets are, respectively, an ideal,
a filter, and a convex set.
The two maps
a H (b: b 5 a} = I(a]
a H(b: a b} = Ira)
are the desired embeddings, and in case L is finite they are easily seen to be
surjective. Since filters and ideals are convex subuniverses, the intersection of a
filter with an ideal is again a convex subuniverse. Finally, let B be a convex
subuniverse. Set
F =(a:biaforsomeb~B)
and
I = (a: a 5 b for some b E B).
F is a filter and I is an ideal, since B is a subuniverse. B = F í l I, since B is
convex.
As a consequence, every lattice is embeddable in an algebraic lattice. On the

other hand, some infinite joins or meets that exist in L may not be preserved by
these embeddings. For example, the integers under their usual ordering, with top
and bottom elements adjoined, comprise a complete lattice. Neither of the
embeddings described above preserves both infinite joins and meets. In this case,
the embeddings can be chosen differently so that arbitrary meets and joins are
preserved. In general, this is not possible (see Exercise 2.20(4) below). But, as
discussed in Example 2.22(6) below, every lattice is embeddable in a complete
lattice in such a way that whatever infinite joins and meets exist will be preserved.
Unfortunately, the complete lattices one obtains are no longer algebraic.
In finite lattices, every element is the meet of a set of meet irreducible
elements; more generally, as we saw in Theorem 2.7, the same holds for lattices
with the ascending chain condition. A version of this statement holds in algebraic
lattices as well. Let L be a complete lattice and a 6 L. The element a is called
strictly meet irreducible iff a = r\X implies that EX for every subset X of L.
Strictly join irreducible elements are defined dually.
THEQREM 2.19. In un algebraic lattice, every element is the meet of a set of

strictly meet irreducible elements.
Proof. Let a E L and let b = /\x where

X = (d: a 5 d and d is strictly meet irreducible}.
It is clear that a L b. Since L is algebraic, in order to prove that b L a, al1 we
need to do is show that c 5 a for al1 compact c 5 b. For the sake of contradiction,
suppose c is compact and c L b but c $ a. Let F = (y: a L y but c $ y}. Plainly
a E F, so F is not empty. Since c is compact, the join of any chain in F is again a
member of F. Hence every chain in F has an upper bound in F. By Zorn's Lemma,
let m be maximal in F. Now m is strictly meet irreducible, since m < d implies

m v c < d by the maximality of m. Thus mEX, and so b 5 m, contrary to the
choice of m E F, since c < b. M
This theorem is a lattice-theoretic rendition of Birkhoff's Subdirect Repre-

sentation Theorem (Theorem 4.44), which is one of the theorems we will find
most useful.
Exercises 2.20 i!
l. a. Let F be a closed set system on A. Define C, on the power set of A by

C,(X) = n ( K : X E K and K E F}. Prove that C, is a closure operator
on A.
b. Let C be a closure operator on A. Define Fc = (C(X): X E A}. Prove that
Fc is a closed set system on A.
c. For C, defined as in (a), prove that F = (C,(X): X G A}.
d. For Fc defined as in (b), prove that C(X) = n ( K : X c K and K E Fc} for
al1 X c A.
2. a. Prove that if L is a lattice in which every set has a least upper bound, then
L is complete.
b. Prove that if L is a lattice in which every chain has a least upper bound,
then L is complete.
c. Prove that if L is a lattice in which every well ordered chain has a least
upper bound, then L is complete.
3. Prove Theorem 2.14: A closure operator is algebraic iff the union of any
collection of closed sets that is directed upward by c is itself closed.
4. Give an example of a complete lattice that cannot be embedded into an
algebraic lattice in such a way that arbitrary (infinite) joins and meets are
preserved.
5. (A. Tarski) Prove that if L is a complete lattice and f : L -,L is an isotone
map, then f (a) = a for some a 6 L. (Such an element a is called a fixed point
of f.)
6. Let C be an algebraic closure operator on A. We say that X c A is C-
independent iff x 4 C(X - (x}) for al1 x E X.
a. Prove that the following are equivalent:
i. For every subset X E A and al1 u, v E A, if U E C(X U (u}) and u 4 C(X),
then v E C(X U (U}).
ii. For every X G A and u E A, if X is C-independent and u C(X), then
X U (u} is C-independent.
iii. For every X e A, if Y is a maximal C-independent subset of X, then
C(X) = C(Y).
iv. For every Y and X with Y E X c A, if Y is C-independent, then there
is a C-independent set Z with Y e Z E X and C(X) = C(Z).
b. Suppose that one of the equivalent conditions of (a) is fulfilled. Prove that
if X and Y are C-independent and C(X) = C(Y), then 1x1= 1 Y[.
2.2 Complete Lattices and Closure Systems 51
It is possible to associate with an arbitrary binary relation two closely

connected closure operators. Perhaps for this reason, closure operators and
complete lattices are quite commonly met in mathematics and quite frequently
useful. Here is how we make this association.
Let A and B be any two classes and let R be a binary relation from A to B
(that is, R E A x B). We are going to define two functions, one from the power
set of A into the power set of B and the other from the power set of B into the
power set of A. These functions are called the polarities of R. Let X c A and
U c B. By definition we take
X' = ( b : xRb for al1 x E X)
U' = ( a : aRu for al1 u E U}.
X' is read "X polar," and U' is read "U polar." The fundamental properties of
polarities are gathered in the next theorem. The proof of this theorem is left
as an exercise.
THEOREM 2.21. Let A and B be classes and R A x B. Let 'and 'be the
polarities of R. Then
i. X E X* and U U' for al1 X E A and al1 U E B.
ii. If Y E X c A, then X' c Y.'
ii'. If V S U S B, then U' c V'.
iii. X' = X- and U' = U ' jbr al1 X A and al1 U E B.
iv. * is a closure operator on A whose closed sets are exactly the polars of
subsets of B.
iv'. " is a closure operator on B, whose closed sets are exactly the polars of
subsets of A.
v. The lattice of closed subsets of A is isomorphic with the dual of the lattice of
closed subsets of B by the map induced by .' 'induces the inverse isornorphism.
m
The polarities are said to establish a Galois connection between the two
closed set systems described in this theorem. Part of the usefulness of such Galois
connections resides in the possibility of drawing conclusions concerning one of
the closed set systems on the basis of information about the other system. Galois
connections also offer a means of analyzing the underlying relation R. We close
this section with a list of examples of Galois connections.
EXAMPLES 2.22.
1. Let q(x) be a polynomial with rational coefficients and let A be the
splitting field of q(x).Let B be the group of automorphisms of A. Define R by
aRg iff g(a) = a.
The resulting closed subsets of A are the subfields of A, and the resulting closed
subsets of B are the subgroups of B. This is essentially the connection brought

to light by Galois in his investigation of the roots of polynomials.
2. Let A be the n-dimensional affine space Cn and let B be the ring
C [x,, x,, ,x,_,] of polynomials in n variables with complex coefficients.
Define R by
ERp(%)iff p(E) = O in C.
The resulting closed subsets of C" are known as affine algebraic varieties, and
the resulting closed subsets of the polynomial ring are the nilradical ideals. This
latter statement is a formulation of Hilbert's Nullstellensatz. This Galois con-
nection is a starting point for the development of algebraic geometry.
3. Let A be a principal ideal domain and let B be an A-module. Define R by
aRb iff ab = 0.
The closed subsets of A turn out to be certain ideals of A called annihilators,

and the closed subsets of B are certain submodules. This Galois connection is a
key to understanding the structure theory of finitely generated modules over
principal ideal domains. This theory in turn comprehends the Fundamental
Theorem of Finitely Generated Abelian Groups and the theorems concerning
the existence and uniqueness of the Jordan and rational canonical forms of
matrices.
4. Let U be a set, let Q be a finitary operation on U, and let S be a finitary
relation on U. We say that S is closed under Q (and that Q preserves S)iff S is a sub-
universe of (U, Q)", where n is the rank of S. The Galois connection established
by the relation R defined by
S R Q iff S is closed under Q
is of considerable importance to us. The resulting closed sets of operations are I
known as clones. Clones will be more carefully introduced in Chapter 4. A chapter 1
in Volume 2 elaborates the theory of clones over finite sets U. On the other side
of the duality, notice that if (U, F) is an algebra, then the unary relations in F'
are precisely the subuniverses of the algebra, whereas the equivalence relations
that belong to F' are precisely the congruence relations of the algebra.
5. Fix a similarity type. Let A be the class of al1 algebras of this type and
let B be the set of al1 equations that can be expressed using variables and the I
operation symbols of the type. Define R by I
While the precise definitions of the concepts of equation and truth are deferred
to $4.11, our intent here should be clear. (Associativity and commutativity are
both expressed by equations; associativity is true in the multiplicative semigroup
of 2 x 2 matrices over the reals, but commutativity is not.) The closed sets of
algebras turn out to be exactly the' varieties, and the closed sets of equations are
the sets closed with respect to logical consequence. This Galois connection is
2.3 Modular Lattices: The Rudiments 53
central for our subject. Its fundamental properties are among the chief concerns
of Chapter 4 and will be fully developed in later volumes.
6. Let L be a lattice and take A = L = B. Let R be the binary relation S
on L. The complete lattice of closed sets of the form'X where X c L is called
the Dedekind-MacNeille completion of L. The Dedekind-MacNeille completion
of the ordered set of rational numbers is (isomorphic to) the ordered set of real
numbers. The map that assigns {a)+ to a for every a~ L turns out to be an
embedding of L into its Dedekind-MacNeille completion, justifying the word
"completion." This map also preserves whatever infinite joins and meets exist
in L.
Exereises 2.23
l. Provide a proof for Theorem 2.21.
2. Prove that the Dedekind-MacNeille completion of the rationals with their
usual order is isomorphic to the reals with their usual order.
3. Prove the contention in Example 2.22(6) that the map described embeds L
into its Dedekind-MacNeille completion and that this embedding preserves
whatever joins and meets exist in L.
2.3 MODULAR LATTICES: THE RUDIMENTS
The study of congruence lattices is central in the development of our subject.

Such lattices must be algebraic, but they need have no other property. It turns
out, however, that most of the intensively investigated kinds of algebras, such as
I
groups, rings, modules, Boolean algebras, and lattices themselves, always have
1
congruence lattices with the following property:
For any elements a, 6, and c, if c a, then a A (b v c ) = ( a A b) v c.
This statement is called the modular law, and lattices for which it holds are called
modular lattices. The significance of the modular law, and indeed of lattices
generally, was first realized by Richard Dedekind. In this section, we present the
I rudiments of the theory of modular lattices.
I
l
l
THEOREM 2.24. (Dedekind [1900_7.) The congruence lattice of any group is
modular.
Proof. Let G = (G, a , -',e) be a group and let p(x, y,z) denote the group-
theoretic expression
For al1 a, b, c E G, the following equalities hold:

P
Chapter 2 Lattices
These two equalities allow us to express the join in Con G in terms of the
composition of relations:
4 v S/ = $OS/ for any 4, S/€ConG.
Indeed, 4 U S / e $OS/ e 4 v S/ is clear. To see that 4 v S/ E 4oS/, it is only
necessary to prove that 4 0 S/ is an equivalence relation. The reflexivity of
4 o S/ follows directly from the reflexivity of 4 and S/. To see the transitivity,
suppose a$oS/b and b4oS/c. Pick u and v in G so that a 4 u S / b and b4vS/c.
Observe that a = p(a, b, b)4p(u, b, u) since a 4 u and b 4 u; since u S/ b and v S/ c,
p(u, b, v) S/ ~ ( bb,, c) = c.
Written more briefly,
and so 4 o S/ is transitive. But now notice that S/ o 4 c 4 o S/ o 4 o S/ c 4 o S/, where

the last inclusion is just the transitivity of 4 o S/. The symmetry of 4 o S/ follows
easily from the symmetry of 4 and S/ and from the inclusion S/ o 4 c 4 o S/. In this
way we have verified that 4 v S/ = 4 o S/.
Now we can easily show that Con G is modular. Let 4, S/, and 0 be con-
gruence relations on G such that 4 < 0. We will deduce
or what is the same in this context in view of our reasoning above:
So let a and b be elements of G such that a (4 o S/) fl0 b. Hence a 4 o S/ b and a 0 b.

Pick c E G so that a 4 c and c S/ b. Since 4 c 0, we obtain a 8 c. So c 8 b, since 8 is
symmetric and transitive. Thus c (S/ íl0) b, and since a 4 c, we can conclude that
a 4 o (S/ íl0) b. So Con G is modular (consult Exercise 2.2(6) for the apparently
missing reverse inclusion). ¤
This proof applies to a much wider class of algebras than groups. Indeed,
the only property of groups used in this proof was the existence of an expression
p(x, y, z), for which two particular equations were satisfied. Any algebra that
allows the construction of such an expression p(x, y, z) from its basic operations
will have a modular congruence lattice. For these algebras, an even stronger
property holds: The join of congruences coincides with the composition of
relations. This theme will be talcen up again in 54.7 and will be explored in some
depth in later volumes.
P
2.3 Modular Lattices: The Rudiments
Figure 2.5
Not every lattice is modular. Consider Figure 2.5. Notice that a 5 c, but
a v (b A c) = a v O = a, whereas (a v b) A c = 1 A c = c. So N, is not modular.
There are many statements equivalent to the modular law. Some are included
in the next theorem, but others can be found in the next set of exercises.
THEOREM 2.25. (Dedekind [1900].) For any lattice L the following state-
ments are equivalent:
i. L is modular.
ii. For any a, b, c E L, if c < a, then a A (b v c) 5 (a A b) v c.
iii. ((a A C) v b) A c = (a A c) v (b A c) for all a, b, c E L.
iv. Foranya,b,c~L,ifa<c,a~b=c~b,andavb=cvb,thena=c.
v. L has no sublattice isomorphic to N,.
Proof.
(i)-(ii) In every lattice, if c < a, then (a A b) v c < a A (b v c). So the
equivalence of (i) and (ii) is clear.
(i) --S(iii) Since a A c 2 c is true in every lattice, the equation displayed in
(iii) must hold in every modular lattice. Conversely, suppose the equation in (iii)
holds in L and let a, b, and c be elements of L such that a < c. Then a = a A c, so
1 (a v b) A c = ((a A c) v b) A c = (a A c) v (b A ec) = a v (b A c).
(i) *(iv) According to the following equations, every modular lattice
satisfies (iv):
a = a v (a A b ) = a v (c A b) = a v (b A c ) = ( a v b) A c = ( c v b) A c = c.
t T T
absorption modularity absorption
(iv) (v) Evidently, lattices that satisfy (iv) cannot have sublattices iso-
morphic to N,.
(v) 3 (i) We argue the contrapositive. Suppose L is not modular. Pick a,
b, and c from L so that a I c but a v (b A c) # (a v b) A c. The elements of L
in Figure 2.6 constitute a sublattice of L isomorphic to N,:
Chapter 2 Lattices
a v b
Figure 2.6
It is necessary to prove that these elements are actually distinct and that the joins
and meets work as indicated in the diagram. First, observe that
where the middle < follows since a i c, and al1 the inequalities must be strict
since a v (b A c) # (a v b) A c. (Collapsing the strict inequality at either end
collapses the whole chain.) Second, observe that
where the strictness once more follows from a v (b A c) # (a v b) A c. Also note

that a v (b A c) $ b and that b $ (a v b) A c. Thus, the Hasse diagram drawn
above is correct. Finally, to see that the joins and meets are correct, just observe
that
and
((a v b) A c) A b = ((a v b) A b) A c =b A c.
This theorem reveals severa1 things about the class of al1 modular lattices.
Part (v) allows us to determine the modularity of a lattice by referring to its Hasse 1
diagram, at least if the lattice is relatively small. Actually, this works best in
proving a lattice nonmodular through the discovery of a copy of N5. The task
of discovering a copy of N5 might be exhausting, but once it is in hand, the task
is basically finished. Of course, one must be careful to verify that the joins and
meets are correct, since the diagram may present only the order structure clearly. I
As a means for concluding that a lattice is modular, part (v) may not be very !
helpful, since al1 of the possible five-element subsets must be examined. Part (iii)
guarantees that subalgebras, homomorphic images, and direct products of mod-
ular lattices are once more modular. Modular lattices can be characterized in
another very useful manner.
DEFINITION 2.26. Let L be a lattice and let a E L. Let #a and Sl, be the maps
from L into L described by
#a(x) = x A a for al1 EL
S/,(x) = x v a for al1 X E L.
Now let a, a', b, b' E L such that a Ib and a' i b'. The interval I [a, b] transposes
up to I [a', b'] iff b' = b v a' and a = b A a'. Dually, I[a, b] transposes down to
I [a', b'] iff b = a v b' and a' = a A b'. We use I [a, b] r I [a', b'] to mean I [a, b]
transposes up to I [a', b']. I [a, b] L I [a', b'] is used to mean I [a, b] transposes
down tol[al, b'].
We call I [a, b] and I [a', b'] transposes if either of these relations hold, and
we call the appropriate rnap (either S/,/or $,?)between the intervals a perspectivity
rnap. Finally, we say that I [a, b] and I [a', b'] are projective iff there is a finite
sequence
such that I[ci, di] and I[ci+,, di+,] are transposes for i < n. The rnap that
results from composing the perspectivity maps associated with this sequence of
transposes is called a projectivity map.
Note that, in general, intervals can be projective by way of many sequences

of transposed intervals. The basic facts about perspectivity maps were realized
by R. Dedekind [1900].
DEDEKIND'S TRANSPOSITION PRINCIPLE (2.27). Let M be a modular

lattice and let a and b be elements of M. The rnap induces an isomorphism from
I[b,a v b] onto I[a A b,a], and S/, induces the inverse isomorphism. Moreover,
the image under either of these maps of a subinterval is a transpose of that
subinterval.
Proof. In view of modularity, for al1 x E I [b, a v b]

S/b(Q)a(~))
= (X A a) v b = x A (a v b) = x
and for al1 x E I [a A b, a]
Hence, S/, o 4, induces the identity function on I [b, a v b], and 4a0 $, induces the
identity function on I [a A b, a]. Consequently, 4a induces a one-to-one function
from I[b, a v b] onto I[a A b, a], and S/, induces its inverse. To conclude that
i these functions are isomorphisms, it is only necessary to note that is isotone,
! since x 5 y implies x A a i y A a is true in every lattice.
To see that subintervals are mapped onto transposes, pick x and y so that
b Ix Iy I a v b. da induces a one-to-one rnap from I[x, y] onto I[x A a, y A a].
To see that these two intervals are transposes, let y' = y A a. We need to verify
that x A a = x A y' and that y = x v y'. This is straightforward:
and
modularity
-1
Applied to the congruence lattice of a group, Dedekind's Transposition

Principle is an abstraction of one of the familiar "Isomorphism" theorems. In
fact, this is the first of severa1 results that were first established about normal
subgroups of a group but which are related to general results for modular lattices.
Other group theoretic results that have led to theorems for modular lattices are
the Jordan-Holder Theorem (see Theorem 2.37) and the Krull-Schmidt Theorem
(see the Direct Join Decomposition Theorem).
COROLLARY 2.28. Projective intervals in a modular lattice are isomorphic.

w
COROLLARY 2.29. Let L be a modular lattice and a, b, c E L with a # b.

i. If a and b both cover c, then a v b covers both a and b.
ii. If c covers both a and b, then a and b both cover a A b. ¤
THEOREM 2.30. The following statements are equivalent for any lattice L:
i. L is modular.
ii. q5a and S/, induce inverse isomorphisms between I [b, a v b] and I [a A b, a] for
al1 a and b in L.
Proof. That (i) implies (ii) is just part of Dedekind's Transposition Principle.
For the converse, suppose L is not modular. Inside L, find a copy of N, and label
it as in Figure 2.7. Since q5a(c) = a A b = q5a(b), we see that q5a is not one-to-one.
w
a v b
Figure 2.7
The proof of the next theorem illustrates another use of Dedekind's Trans- I
position Principle.
THEOREM 2.31. (Birkhoff [1948].) Let L be a modular lattice and let a and b
be members of L. Set Lo = I [ a A b, a] and L , = I [ a A b, b] and let L, be the
sublattice of L generated by Lo U L,. Then L, r Lo x L,.
Proof. We define the desired isomorphism F: Lo x L , + L, for al1 x E I [a A b, a]

and y E I [ a A b,b] by:
Before showing that F is an isomorphism onto L,, we note the following simple
facts:
i. I[a A b , a ] r I [ b , a v b]andí[a A b,b]rI[a,a v b].
ii. If a A b i x i a, then x = $,($,(x)) = a A (x v b).
iii. If a A bI yI b, then y = $,($,(y)) = b A (y v a).
iv. If a A b i x i a, and a A b 5 y i b, then
Thus F(x, y) = x v y = $,(x) A $,(y) for al1 (x, y) E Lo x L,.

CLAIM O: F is a homomorphism.
Pick (x, y) and (x', y') E Lo x L, .
F((x, y) v (x', y')) = F(x v x', Y v Y')
= (xv x') v (y v y')
= F(x, y) v F(xf,y').
Thus F preserves joins.

F((x, Y) A (x', Y')) = F(x x', y A y')
= $b(x A x') A $ a ( ~A Y')
= Sb(x) A $b(xl) A $,(Y) A $,(Y')
= $b(x) A S/,(Y)A $ b ( ~ ' )A $a(yl)
= F(x, y) A F(xl,y').
So F preserves meets.
CLAIM 1: F is one-to-one.
Notice that x can be recovered from x v y as follows:
In a similar way, y can be recovered. Hence, F(x, y) = F(xl,y') implies x = x'

and y = y'.
CLAIM 2: F is onto L,.
Just note that the image of Lo x L , under F is a lattice, and it is comprised of
al1 joins x v y where x E [a A b, a] and y E [a A b, b]. ¤
The lattice M, diagrammed in Figure 2.8 illustrates that, even for finite
modular lattices, there may be severa1 distinct ways to represent an element as
the join of join irreducible elements. But the next theorem, due to Kurosh [1935]
and Ore [1936], asserts some uniqueness in such representations.
Chapter 2 Lattices
Figure 2.8
DEFINITION 2.32. Let L be a lattice and let M be a finite subset of L. M is

called join irredundant iff for al1 proper subsets N of M,
VN < V M .
Meet irredundance is the dual notion.
THE KUROSH-ORE THEOREM (2.33). Let L be a modular lattice and a E L.

Suppose that
where {ai:O < i 5 n} and {b,: O < j 5 m) are join irredundant sets of join irre-
ducibles and the si's are distinct, as are the bjk. Then n = m, and, after
renumbering,
= b, v b, v v b,-, v a,.
Proof. We will first establish that a = bj v a, v a, v v a, for some j. Let

c = a, v a, v v a,. Then I[c,a] transposes down to I[c A a,,a,], so these
intervals areisomorphic by Dedekind's Transposition Principle. Since a, is join
irreducible, a must also be join irreducible in I[c, a]. But clearly
a = (b, v c) v (b, v c) v -v (b, v c),
so a = bj v c for some j, as promised.
Suppose, for the moment, that n < m. Continuing the above process will
ultimately yield a as a join of only n of the bis, in contradiction to irredundancy.
By symmetry, m < n is also contradictory, so we have n = m. The desired
equalities now follow easily by iterating the above method. m
We conclude this section with a simple result concerning complementation

in modular lattices. Let L be a bounded lattice with largest element 1 and smallest
element O. Let a E L. The element c E L is called a complement of a iff a A c = O
and a v c = 1. L is said to be a cornplernented lattice iff every element of L has
a complement; L is called relatively cornplernented provided every interval I [a, b]
in L, when construed as a sublattice, is a complemented lattice. The lattice of al1
subspaces of a finite dimensional vector space is easily seen to be a complemented
lattice.
2.4 Modular Lattices with Finite Chain Condition 61
THEOREM 2.34. Every complemented modular lattice is relatively comple-

mented.
Proof. Let M be a complemented modular lattice and let a I x 5 b hold in M .

Let y be a complement of x in M. So x v y = 1 and x A y 5 O. But just notice:
a = O v a = ( O ~ b ) v a = ( ( x ~ y ) ~ b ) v a = ( x ~ ( y ~ b ) ) v a
= x A ( ( y A b) v a)
and dually,
b = l ~ b = ( vl a ) ~ b = ( ( x v y ) v ab )=~( x v ( y v a ) ) b~
= x v ( ( y v a) A b).
But since(y v a) A b = (a v y) A b = a v ( y A b) = ( y A b) v a andsince

a<(y~b)va<b
we conclude that ( y A b) v a is a complement of x in I[a, b].
Exercises 2.35
1. Let L be a finite lattice. Prove that L is modular iff I [a A b, a] r I [b, a v b]
for al1 a, b E L.
2. Construct a lattice that is not modular such that I [ a A b, a] r I[b, a v b] for
al1 a, b E L.
3. Prove that in a modular lattice no element can have two distinct complements
that are comparable to each other.
4. A lattice is said to satisfy the upper covering property (or to be semimodular)
iff given a, b, and c, if a 4 b, then either a v c = b v c or a v c 4 b v c.
The lower covering property (or lower semimodularity) is the dual notion.
a. Prove that every modular lattice has both the upper and lower covering
property,
b. Construct a nonmodular lattice with both the upper and lower covering
property.
c. Let L be a lattice. Prove that L is semimodular iff for al1 a, b E L , if
a A b 4 a, then b < a v b.
5. Prove that the join irreducible elements of a complemented modular lattice
are exactly the atoms of the lattice.
2.4 MODULAR LATTICES WITH THE FINITE

CHAIN CONDITION
Some very fruitful directions in algebra were opened by the observation that
infinite algebras satisfying various "finiteness conditions" were amenable to an
almost combinatorial analysis. Often these "finiteness conditions" amount to
restrictions on ascending or descending chains in the lattice of congruence

relations on the algebras. Noetherian and Artinian rings are specified by just
such conditions. Moreover, among vector spaces, the finite dimensional ones
are exactly those that have congruence lattices in which every chain is finite.
The structure theorems for a very broad range of algebras, which emerge in
later chapters (especially Chapter 5), flow from some of the principal theorems
concerning modular lattices in which every chain is finite. The length of a finite
+
chain with n 1 elements is n.
DEFINITION 2.36. A lattice L satisfies the finite chain condition iff every chain
in L is finite.
Every lattice with the finite chain condition has a greatest element and a
least element. According to the Hausdorff Maximality Principle, every lattice has
a maximal chain. Hence, lattices with the finite chain condition must have finite
maximal chains. A finite maximal chain is one in which each "link" is a covering
and in which the top and bottom elements are the 1 and O of the lattice. Even
so, it is not possible to bound the lengths of chains in such a lattice, as Figure 2.9
reveals. This sort of pathology does not happen among modular lattices, as
the next theorem confirms. Perhaps this was first realized in the context of
congruence lattices of finite groups (the familiar Jordan-Holder Theorem), and
it may have been one of the clues that led Dedekind to formulate the concept of
modularity. Ore [1935] has also been credited with the following result. An
alternative proof based on one of the most familiar proofs of the group result is
sketched in the exercises.
Figure 2.9
THEOREM 2.37. (Dedekind [1900], Birkhoff 119331.) Let L be a modular

lattice and let a < b in L. If there is a finite maximal chain from a to b, then every
chain from a to b is finite, and al1 the maximal ones have the sume length. If
and
then the intervals I[ai,aic1] and I [ c j ,cj+,] can be matched in such a way that
matching intervals are projective.
i
l 2.4 Modular Lattices with Finite Chain Condition 63
Proof. Let us cal1 two finite chains equivalent iff they have the same length and
have the property described in the final sentence of the theorem (applied to
arbitrary chains, not just maximal chains). This specifies an equivalence relation
on finite chains between two elements of L.
We argue by induction on the length of a finite maximal chain from a to b.
Let a = a, < a, -< -< a , = b be a finite maximal chain from a to b.
l
INITIAL STEP: n = 1. The definition of covering leaves nothing to prove.
: INDUCTIVE STEP: n > 1. Our induction hypothesis is that the conclusions
l
of the theorem hold for any two elements linked by a maximal chain of length
less than n. Let C be a maximal chain from a to b. If a, E C, then C - {a,} is
a maximal chain from a, to b, and the induction hypothesis applied to a, and b
yields al1 the desired conclusions. In the remaining case, pick c E C such that c
and a, are incomparable. Hence a, A c = a,. Let d = a, v c. Thus
-T[ao,c]~I[al,d] and I[a,,a,]~I[c,d].
,
Let C, = (C'EC: cf < C) and let C, = {C'EC: c I ;cf}.Let Do be the image of C,
under the first perspectivity map and let D, be any maximal chain from d to b.
(Such a chain must exist according to the Hausdorff Maximality Principle
applied to I[d, b].) Figure 2.10 suggests an arrangement of these chains. By
Dedekind's Transposition Principle, Do U {d} is a maximal chain from a, to d,
since C, U (c} is a maximal chain from a, to c. Thus Do U D, is a maximal chain
l from a, to b.
,
a0
Figure 2.10
According to the induction hypothesis, Do U D, and a, < a, < -< b

!i are equivalent. By the Dedekind Transposition Principle, C, U D, U {c) and
1l
are equivalent. The induction hypothesis also yields {c} U D, equivalent to C,,
j so C is equivalent to C, U D, U {c). Therefore, C is equivalent to
Chapter 2 Lattices f
DEFINITION 2.38. A lattice L is said to be of finite height iff there is a finite

upper bound to the length of chains in L. The least such upper bound is called
the height of L. A lattice L is said to be sectionally of finite height iff L has a least
element 0, and for every a E L ,the interval I[O, a] is of finite height. In this case,
the height of I[O, a] will be denoted by h(a) and called the height of a.
Every lattice of finite height satisfies the finite chain condition. From
Theorem 2.37, it follows that every modular lattice with the finite chain condition
is a lattice of finite height. In modular lattices sectionally of finite height, the
height function is very well behaved, as the next theorem reveals. The proof is
left as an exercise.
THEOREM 2.39. Let L be a modular lattice sectionally of finite height. Then

for al1 a, b EL,
i. h(0) = O
ii. If a < b, then h(a) < h(b)
iii. h(a) + h(b) = h(a v b) + h(a A b).
THEOREM 2.40. Every bounded modular lattice in which 1 is the join of a

finite set of atoms is a relatively complemented modular lattice of finite height.
Proof. Let L be a bounded modular lattice in which 1 is the join of a finite set
A of atoms. We first argue that L is of finite height. In view of Theorem 2.37 +
and the fact that every chain can be extended to a maximal chain, we need only ,
find one finite maximal chain from O to 1. We construct this chain recursively
as follows. Let c, = O. If ci # 1, then pick a,€ A such that ai ci and let \
ci+, = ci v ai. l
i
Since A is finite, this construction stops after finitely many steps and produces
a chain O = c, < c, < < c, 7 l. But observe that ci A ai = O, since ai is an
m - .
atom not less than ci. Consequently, IIO,ai] transposes up to I[C,,C,+~]. By

Dedekind's Transposition Principle, we conclude that ci -< ci+, for al1 i < n, so
there is a finite maximal chain from O to l.
According to Theorem 2.34, we need only show that L is complemented. l
The construction we just used actually produces complements. Indeed, let x E L 1

and proceed as follows. Let do = x. If di # 1, then pick ai E A such that ai $ di
and let di+, = di v ai. ,
Since A is finite, this construction stops after finitely many steps, producing,
let us say a,, a,, a,. Take y = a, v a, v
m . . , v a,. By the construction we
have x v y = 1. By Theorem 2.39, x A y = O iff h(x A y) = O. First observe that
for each i < n,
i
Gonsequently,
h(x v y) = h(x) + h(a,) + - + h(a,),
and, by the same reasoning applied to the sequence of a,)s, we have
1
I
i 2.4 Modular Lattices with Finite Chain Condition
Therefore h(x v y) = h(x) + h(y). By Theorem 2.39,
The three conditions listed in the Theorem 2.39 above are familiar properties
of the dimension function applied to subspaces of a finite dimensional vector
space.
DEFINITION 2.41. Let L be a lattice with least element O. A function d: L -+ co

with the following properties, for al1 a, b E L,
i. d(0) = O.
ii. If a < b, then d(a) < d(b).
iii. d(a) + d (b) = d (a v b) + d (a A b).
is called a dimension function on L. A lattice is said to be finite dimensional iff it
is bounded and there is a dimension function on it.
I
l
The height function might be regarded as the natural dimension function
on a modular lattice that is sectionally of finite height. This is the case, for
example, with the lattice of finite dimensional subspaces of a vector space. The
height function h puts h(a) = 1 for every atom a in the lattice, a condition we did
not include in the definition of dimension function but which will be of use in
Chapter 4.
,
THEOREM 2.42. For any lattice L the following are equivalent:
1 i. L is finite dimensional.
l
i
ii. L is a modular lattice with the finite chain condition.
1 iii. L is a modular lattice of finite height.
Proof. As we remarked after Theorem 2.37, parts (ii) and (iii) are equivalent.
Part (i) follows from (iii) by Theorem 2.39, since lattices of finite height are
bounded.
To prove that (i) implies (ii), let L be a bounded lattice with dimension
function d. By properties (i) and (ii) of the dimension function, no chain in L can
have length greater than d(1). Therefore, L has the finite chain condition. Finally,
to verify the modularity of L, pick a, b, c E L with a < c. Since L is a lattice, we
know that a v (b A c) < (a v b) A c. In view of property (ii) of d, equality will
hold iff d(a v (b A c)) = d((a v b) A c)). But observe by property (iii) of d:
d(a v (b A c)) = d(a) + d(b A c) - d(a A b A c)
= d(a) - d(a A b) + d(b A c)
= d(a v b) - d(b)+ d(b A c)
= d(a v b) - d(b v c)+ d(c)
= d(a v b) - d(a v b v c) + d(c)
= d(a v b) + d((a v b) A c) - d(a v b)
= d((a v b) A e). •
The Kurosh-Ore Theorem is a step toward a unique join decomposition

theorem for modular lattices, but it falls short. It is not difficult to devise finite
modular lattices in which there are elements that can be written as joins of many
different finite sets of join irreducibles. But it is possible, for modular lattices
of finite height, to obtain a stronger decomposition theorem, the Direct Join
Decomposition Theorem. This stronger result concerns not arbitrary joins of
finite sets, but rather joins of sets of a rather restricted kind that we will cal1
directly join independent sets. Moreover, the uniqueness obtained is really
"uniqueness up to direct join isotopy," where direct join isotopy is a certain
equivalence relation between the elements of the lattice. Just as Theorem 2.37
was inspired by the Jordan-Holder Theorem from group theory, the Direct Join
Decomposition Theorem can be traced to some well-known theorems in group
theory. In Kronecker [1870] it was shown that any finite Abelian group can be
written as a direct product of directly indecomposable Abelian groups in an
essentially unique way. This result has been extended in various ways. Perhaps
the best known goes under the name of the Krull-Schmidt Theorem, which
asserts that every group whose normal subgroup lattice satisfies the finite chain
condition can be decomposed into a direct product of directly indecomposable
groups in essentially only one way (see also Wedderburn [1909].) It was Ore
[1935, 19361 who realized that this unique factorization property could really
be traced to a purely lattice-theoretic property of the congruence lattice of
the group. It has turned out to be possible to use the resulting Direct Join
Decomposition Theorem to obtain unique direct factorization results for much
wider classes of algebras than just finite groups. This point is taken up again in
Chapter 4 and more extensively in Chapter 5, which is devoted to unique direct
factorizations. Our development of the Direct Join Decomposition Theorem
relies heavily on Jónsson [1966].
DEFINITION 2.43. Let L be a lattice with least element O. A subset M G L

is directly join independent iff whenever N is a finite subset of M and a E M - N,
then a A VN = O. An element a E Lis called directly join irreducible iff O < a and
if a = b v c with {b, c) directly join independent and b # c, then b = O or c = 0.
IND(L) denotes the collection of directly join independent subsets of L.
For notational convenience, we introduce a partial operation @, referred

to as direct join, on the lattice L. a @ b is defined whenever a A b = 0, and it
takes the value a v b in that case. Thus, a @ b = c is equivalent to the assertion
of the following three conditions:
i. {a, b) is directly join independent,
ii. a v b = c, and
iii. a # b or a = 0 = b.
DEFINITION 2.44. Let L be a lattice with least element O and let a, b E L. The
elements a and b are directly join isotopic in one step iff there is c E L such that
a @ c = b @ c. a and b are said to be directly join isotopic iff there is a finite
f 2.4 Modular Lattices with Finite Chain Condition
sequence do, d,, - . d, of elements of L such that a = do, d, = b, and di is directly

S ,
/ join isotopic with di+, in one step, for each i = 0, 1, . . . , n - l .

Notice that if a and b are directly join isotopic in a lattice L, then I [O, a]
and I[O, b] are projective intervals in L; the associated projectivity map will be
referred to as a join isotopy map. One-step direct join isotopy is not a transitive
relation, even in finite modular lattices; see Exercise 2.49(10).
The notions dual to direct join independence, direct join irreducibility, and
direct join isotopy apply to lattices with a greatest element and are referred to,
I respectively, as direct meet independence, direct meet irreducibility, and direct
meet isotopy. Actually, these dual notions (and variants of them) are the ones
used in Chapters 4 and 5 to obtain unique direct factorization results for algebras.
But it has been traditional in lattice theory to approach this material from the
direct join viewpoint. For the remainder of this section, we will be concerned
exclusively with the selfdual class of modular lattices of finite height.
1
To illustrate these concepts, let L be the lattice of al1 subsets of some set X.
1 It is easy to see that a collection of subsets of X will be directly join independent
1! iff it is a collection of pairwise disjoint sets. The only directly join irreducibles
1 are the singleton sets, and two sets are directly join isotopic iff they are equal.
Taking L to be the lattice of subspaces of the three-dimensional real vector space,
, it is not very hard to classify the directly join independent subsets of L. Accom-
l
plishing the same task for the four-dimensional real vector space is more time-
1 consuming but may provide a better intuitive feel for these notions.
In the setting of finite dimensional lattices, direct join independence takes
on an especially simple form.
THEOREM 2.45. Let L be a finite dimensional lattice with dimension function
d and let M S L. M is directly join independent 8M is finite and = ~(VM)
C
Proof. First suppose that M is directly join independent and let N be any finite
subset of M. We will prove by induction on 1 N 1 that
lI This will entail that M is finite, since d(a) 2 1 provided that a # O and since d(1)
is an upper bound on d ( V N ) for al1 finite N . In the initial step of the induction,
N is empty and the conclusion is immediate. For the inductive step, let a E N and
set N' = N - ( a ) . Then
1 = d(a) + ~ ( V N '-) d(0) since M is directly join independent
= d(a) + d(a') by the induction hypothesis

a' E N'
= 1 d(b).
b€N
Chapter 2 Lattices
For the converse, we need the following extension of the dimension formula
r
occurring as (iii) in Definition 2.41: For any n distinct elements a,, a,, a,-, m ,
of L
d(a, v a, v v a,,-,) = d(a,) + d(a,) + - - + d(an-,)
- [d(a, A (a, v v a,-,))
pI + d(a, A (a, v v a,-,)) + -
+ d(an-2 A a,-,)l.
This formula can be established by induction. Now observe that the formula
above depends on the order in which the aiYshave been indexed. Plainly, we have
one such formula for each of the n! ways of indexing available.
Suppose that M is a set with n elements such that
Let N be any subset of M, say with k elements, and pick c E N, setting N' = N -
(c}. We must argue that c A VN'
= O. Now let M = (a,, ai;-.,an-,} SO that
c = un-, and N' = {an-,+, , ,un-,}. According to the extended dimension
formula above and the condition just imposed on ~ ( V M )we , conclude that
d(a, A (a, v - v un-,)) + + d(c A VN') + + d(an-, A a,,-,) = O.
Since d only produces non-negative values, we conclude that al1 the terms of this
sum are 0. In particular, d(c A V N ' ) = O. But this implies that c A = O, as VN'
desired. Hence, M is directly join independent.
Before turning to the Direct Join Decomposition Theorem, we gather in the

next theorem the fundamental properties of directly join independent sets in
modular lattices of finite height that we shall use. Most of these properties follow
very easily from the definitions and Theorem 2.45. However, the ten properties
listed are more than useful tools. In fact, they constitute al1 the conditions on the
family IND(L) necessary to establish the Direct Join Decomposition Theorem
for the finite dimensional lattice L. As a consequence, any family I of subsets
of L that has al1 the properties attributed below to IND(L) will give rise to a
variant of the Direct Join Decomposition Theorem. We could have introduced
an abstract concept of "join independence family," using the ten properties below
as a definition, and then established a more general theorem. Observe that the
notion of direct join irreducibility depends on the notion of direct join indepen-
dence. Direct join isotopy and the direct join operation are also derivative
notions. To obtain a variant of the Direct Join Decomposition Theorem for a
"join independence family" I, these notions must both be modified by referring
them to I in place of IND(L). The specific notion of direct join independence
introduced above would then be one example of a "join independence family."
In 55.3, we will invoke the Direct Join Decomposition Theorem for a slightly
different notion of join independence. That notion and the one defined in 2.43
are the only kinds of "join independence families" in this volume.
F 2.4 Modular Lattices with Finite Chain Condition
,
,
THEOREM 2.46. Let L be a modular lattice of finite height.
1 i. If N s M E IND(L), then N E IND(L).
ii. If M E IND(L),then M U {O)E IND(L).
iii. If a @ b E M E IND(L),then ( M - ( a @ b ) )U {a,b ) E IND(L).
iv. If a, b E M E I N D ( L ) and a # b, then ( M - {a,b )) U {a @ b ) E IND(L).
v. If M E I N D ( L ) and f : M -+ L such that f ( x ) 5 x for al1 x E M , then
{ f ( x ) :x E M ) E IND(L).
vi. If a @ a' = b @ b' = a v b' = a' v b, then a @ b' = a' b = a a'.
vii. If {a,a ' ) E I N D ( L ) with a # a' and b < a, then b @ a' < a @ a'.
viii. If b 5 a @ a', b $ a, and ( a A (a' v b),b ) E IND(L),then {a,b ) E IND(L).
ix. If a = a @ b, then b = 0.
x. If a @ b is directly join isotopic with c, then there are a' and b' such that
c = a' @ b', and a is directly join isotopic with a' and b is directly join isotopic
t
1
with b'.
Proof.
i. This is completely straightforward.
ii. This is also immediate.
iii. Suppose that a O b, a,, , a,-, is a list of al1 the distinct elements of M.
To see that (M - {a O b } )U (a,b ) is directly join independent, we invoke
Theorem 2.45.
iv. The argument just given for (iii) can be easily rearranged to prove this part.
v. This follows easily from the definition of direct join independence.
vi. We assume that none of a, a', b, b' is O, since otherwise the desired result is
immediate from the definition of direct join independence. Hence a # a' and
b # b'. The hypotheses now give the following dimension equations:
d(a) + d ( a f )= d(b) + d(bl)
d(a v b') = d(af v b)
d(a) + d(a') = d(a v b').
In turn, these equation yield
+ d ( a f )= d(a) + d(bl) d(a A b')
d(a) -
+ d ( b f )= d ( a f )+ d(b) d(af A b).

d(b) -
From these equations we obtain d(a A b') + d(af A b) = O. Therefore both

d(a A b') = O and d(af A b) = O. Hence a A b' = O = a' A b. Thus both
Chapter 2 Lattices
{a, b') and {a', b) are directly join independent sets of cardinality two and
so a @ b' = a' @ b = a @ a' as desired.
vii. According to (v) {b, a') is directly join independent and d(b) < d(a) since
b < a. So
d(b @ a') = d (b)+ d (a')
< d(a) + d(af')
= d(a @ a').
Thus b @ a' S a @ a' and b @ a' # a @ a'.

viii. Just notice that a A b = a A b A b I a A (a' v b) A b = O. (Here the hy-
pothesis that b < a @ a' is unnecessary.)
+
ix. Since d(a) = d(a) d(b), we conclude that d(b) = O and so b = 0.
x. It suffices to prove this when a @ b is directly join isotopic to c in one step.
So pick d with a @ b # d # c such that
Thus, I [O, a v b] and I [O, c] are projective, and the isotopy map that takes
x €110, a v b] to (x v d) A c is a lattice isomorphism by the Dedekind
Transposition Principle. Now let a' = (a v d) A c and b' = (b v d) A c.
a' A b' = O, since a A b = 0, and a' v b' = c, since a v b = a v b, by the
isomorphism. Hence a' @ b' = c. T o see that a and a' are directly join
isotopic, just observe:
= ((a v d) A c) v d
= (a v d) A (c v d) by modularity
= ( a v d) A ( c @ d )
= ( a v d) A ((a v b)@d)
= ( a v d ) ~ ( a bv v d )
The properties attributed to IND(L) and to the partial operation of direct

join by the first four parts of Theorem 2.46 make the direct join easy to manipu-
late. For example, they entail that direct join is associative in a strong sense:
Parentheses can be rearranged without the worry of whether the operations are
defined (a concern when dealing with partial operations). O's can be inserted or
deleted without trouble. Also note that direct join is commutative, as a con-
sequence of the definition itself. In the proofs below, we have mostly neglected
to point out such uses of Theorem 2.46. The reader should note that a @ a is only
defined when a = 0.
The next lemma is the key to our proof of the Direct Join Decomposition
Theorem. This lemma is taken from Jónsson [1966].
LEMMA. Let L be a modular lattice of finite height and let a, a', b, b', d E L such
t hat
a@af@d=b@b'@d.
Then there are c and c' with c 5 b and c' Ib' such that
Proof. Let e = a @ a' @ d = b @ b' @ d. We view the lemma as an assertion

about 6-tuples (e, d, a, a', b, b') of elements of L. The following three cases are
exhaustive and mutually exclusive:
i. e = a v b ' v d = a f v b v d .
ii. a v b f v d < e .
iii. a v b' v d = e but a' v b v d < e.
CASE 1: e = a v b' v d = a' v b v d.
In view of Theorem 2.46 (vi) (and with the help of parts (i), (ii), and (iii) as well),
putting c = b and c' = O settles this case.
The two remaining cases are settled by indirect proof. So we now assume
the lemma is false. A counterexample to the lemma is a 6-tuple (e, d, a, a', b, b')
such that
e = a @ a f @ d and e = b @ b f @ d ,
but no choice of c and c' will fulfill the lemma. Since L is finite dimensional, it
has the finite chain condition (Theorerh 2.42), so every nonempty subset of L has
both minimal and maximal members (Theorem 2.6). Fix e so that it is minimal
among al1 first entries of counterexamples. Next, fix d so that it is maximal among
al1 second entries of counterexamples with first entry e. Of course, there may exist
6-tuples with first entry e and second entry d that are not counterexamples.
In fact, we have already observed that 6-tuples falling into Case 1 cannot be
counterexamples. We will prove the lemma by showing that the same applies to
the remaining cases.
CASE 11: a v b' v d < e.
Let e, = a v b' v d. So e, < e; this means that e, is not the first entry of any
counterexample. Let
a,= a a; = a' A e,
b, = b A e, b; = b'.
By Theorem 2.46 (v), (a,, a;, d} and (b, ,b; ,d} are both directly join independent.
Moreover, modularity yields
a, v a; v d =av (a' A e,) v d = (a v d) v (a' A e,)
= ( a v d v a') A e, = e A e, = e , .
b, v b; v d = (b A e,) v (b' v d) = (b' v d) v (b A e,)
= ( b v d v b') A e, = e A e, = e , .
Thus, (e,, d, a,, a;, b,, b;) is a 6-tuple fulfilling the hypotheses of the lemma. It is
not a counterexample. So pick c < b, and c' < b; such that e, = c @ c' @ a; @ d.
NOW ~ ~ ~ ' ~ a ' v d = c v c ' v a ~ v a ' vv ad ' == ae v, b f v d v a ' = e .
Al1 that remains for Case 11 is to show that {c, c', a', d ) is directly join indepen-
dent. But observe that by Theorem 2.46 (iv),
(a;, c @ c' @ d ) is directly join independent
and
{a @ d, a') is directly join independent.
Another way to write the first of these two sets is
(a' A (a v b' v d), c @ c' @ d).
With the help of Theorem 2.46 (v), we deduce that
(a' A ((a v d) v (c @ c' @ d)), c @ c' @ d ) is directly join independent.
Now, Theorem 2.46 (viii) entails that
(a', c @ c' @ d ) is directly join independent,
since c @ c' @ d < e, < e = a' @ (a v d). By Theorem 2.46 (i) and (iii), we con-
clude that (c, c', a', d ) is directly join independent, as desired. So no 6-tuple
beginning with e that falls into Case 11 is a counterexample to the lemma.
Moreover, in Case 11, for our fixed e, we can conclude that
for otherwise c = b, and since c I b, < b, we obtain b = b, = b A e,. In turn,

this implies that b < e, and so e, = e, v b = a v b' v d v b = e, contradicting
e, < e. Case 11 is settled.
The following claim, which is easily establishing using modularity and
Theorem 2.46 (v), is used severa1 times in Case 111 and also in the proof of the
Direct Join Decomposition Theorem.
CLAIM: Let x, y, z E L and define x # to be x A z. If y 5 x < y @ z, then
x=x#oy. M
CASE 111: a' v b v d < a v br v d = e.
Interchanging a with b and a' with b', we obtain the 6-tuple (e, d, b, b', a, a') for
our fixed e and d. This 6-tuple falls into Case 11. Since we have already verified
Case 11, we pick c, < a and c; < a' so that
By Theorem 2.46 (iv), (c;, c, @ b' @ d) is directly join independent. Now we

invoke the claim, with a' as x, c; as y, c, @ b' @ d as z, and a # = a' A (c, v b' v d).
Hence a' = a # @c;. Further, we have
e=a@a#@(c;@d)
(*>
e = c, @ b' @ (c; @ d)
according to Theorem 2.46 (iii) and (iv).
2.4 Modular Lattices with Finite Chain Condition
SUBCASE IIIA: O < c;

In this subcase, Theorem 2.46 (vii) yields d < c; O d. Thus the lemma holds at
the 6-tuple (e, c; O d, a, a # , c,, b'), in view of (*). So pick c, I c, and c; 5 b' so
that
e = c, O c; O a" @ (c; O d).
Observe that
e = c2 O c; O (a# c;) d
=c20c;@ar@d
e = c2 O a' O (c; d).
We apply the claim again, this time taking b' as x, c; as y, and c, a' d as z.
Thus for b# = b' A (c, v a' v d), the claim gives us
b' = b# Oc;.
Hence
(***) e = b O b# O (ci d).
But c; > 0, for otherwise e = c, O a' O d < a O a' O d = e, where the strict
inequality comes from c, c, < a (by Theorem 2.46 (vii)). Hence d < c; d,
again by Theorem 2.46 (vii). In view of (**) and (***), the 6-tuple
(e, c; d, c27 U',b7 b # )

fulfills al1 the hypotheses of the lernrna. Since d < c; d, the lemma holds at this
6-tuple. So pick c, I b and c j 5 b# so that
e = c, O c3 a' (c; @ d).
Note that by Theorem 2.46 (iii) and (iv), we have
e = c, O (c3 ci) @ a' O d.
Since c, 5 b and ci c; < b# v b' 5 b' v b' = b', this subcase is settled.
SUBCASE IIIB: c; = 0.
In this case, we have c, < a and

e a @ a ' O d and
=
e = b ' O c, O d.
In the event that a v c, v d = e, we get e = a v d. So Theorem 2.46 (v) and (vii)

yield a' = O. Then choosing c = b and c' = b' demonstrates the lemma. So for the
remainder of this case, we take a v c, v d < e. Thus the 6-tuple (e, d, a, a', b', c, )
falls into Case 11. So pick c, _< c, and c; b' such that
(**> e = c, @ a' O (c; O d).
Now the same reasoning, word for word, used to complete Subcase IIIA, from
the point labeled (**) in that case, can be used to complete this subcase. •
THE DIRECT JOIN DECOMPOSITION THEOREM (2.47). Let L be a

modular lattice of finite height. Every element of L is the join of a finite directly
join independent set of directly join irreducible elements. If M and N are finite
directly join independent sets of directly join irreducible elements of L such that
VM andVN are directly join isotopic, then there is a one-to-one function f from
M onto N such that x is directly join isotopic with f(x) for each x E M.
Proof. There are two parts to the theorem: the existence of direct join de-
compositions and their uniqueness up to direct join isotopy. The existence
follows by an easy induction on the dimension of elements. We omit the details
except to say that Theorem 2.46 (ix) has a role to play. The uniqueness is
established by induction on the smaller of 1 M 1 and 1 N 1. Without loss of generality,
suppose 1 NI 5 1 MI.
INITIAL STEP: 1 N 1 = 0.
In this case N is empty, so VN = O. It follows from the definition of direct join
isotopy and Theorem 2.46 (ix) that VM = O. Thus M is empty, as desired, or
M = (O}. The last alternative is excluded because O is not directly join irreducible.
INDUCTIVE STEP:
M is nonempty, since 1 N 1 I 1 M 1 and N is nonempty. It follows from Theorem
2.46 (ix) that only O can be directly join isotopic to O. From this and the obvious
inductive extension of Theorem 2.46 (x) to arbitrary finite direct joins, pick g to
be a one-to-one function from M into I[O, VN] such that x is directly join
isotopic with g(x) for al1 x E M, and V{g(x): x E M} = V N .
Pick a E {g(x):x E M} and b ' N. ~ Let a' = V({g(x): x E M} - { a } ) and
b = V ( N - {b'}).
Thus a @ a' = b @ b'. Letting d = O in the lemma, pick c and c' such that
a @ a' = c @ c' @ a', c 5 b, and c' 5 b'. Hence a and c @ c' are directly join
isotopic. From Theorem 2.46 (ix) and (x), it follows that any element directly
join isotopic to a directly join irreducible element must itself be directly join
irreducible. Now, c @ c' is directly join isotopic to an element of M (by way of a),
so c @ c' is directly join irreducible. Thus either c = O or c' = 0.
CASE 1: c =O and a is directly join isotopic with c'.
Using c' 2 b' 5 c' @ a', pick c# so that b' = c' @ c # . (This can be done using
the claim that was isolated during the proof of the lemma.) Since b' is directly
join irreducible and c' # O, it follows that b' = c'. Therefore
In particular,
so a' and b are directly join isotopic. But

a' = V({g(x): x E M} - { a } )
b = V ( N - {b'}).
According to the induction hypothesis, pick a one-to-one function h' from {g(x):
x E M } - { a } onto N - { b ' } such that y is directly join isotopic with h f ( y )for al1
y E {g(x):x E M } - {a). Extend h' to a function h: {g(x);x E M } + N by setting
h(a) = b'. Then hog is a one-to-one function from M onto N such that h(g(x))
is directly join isotopic with x for al1 x E M .
CASE 11: c' = O and a is directly join isotopic with c.
Again using the claim established in the proof of the lemma and c I b I c @ a',
pick c# with b = c @ c#. By the existence part of this theorem, let C be a finite
directly join independent set of directly join irreducible elements such that
c# = V c . So
V ( C U { c } )= b = V(N
- {b'}).
By the induction hypothesis, let h' be a one-to-one function from C U { c } onto

N - { b ' } such that h l ( y )is directly join isotopic with y for al1 y E C U {c}.Extend
h' to h: C U { c }U { b ' } + N by setting h ( b f )= b'. Now
and in particular c# @ b' is directly join isotopic with a'. But

c# @ b' = V(CU { b ' ) ) and
a' = V ( { g ( x ) x: E M ) - (a}).
The induction hypothesis supplies a one-to-one map f ' from {g(x):x E M } - { a }
onto C U {b' ) such that y and f ' ( y )are directly join isotopic, for al1 y E ( g(x):
x E M } - {a}. Extend f ' to f from {g(x):x E M } onto C U { b ' }U {c} by setting
f (a) = c. The desired one-to-one function from M onto N is h o f o g.
Ore's formulation of a direct join decomposition theorem for modular

lattices of finite height can now be drawn as a corollary.
COROLLARY 2.48. (Ore 119351, 119361.) Let L be a modular lattice of finite

height. Every element of L is the join of a finite directly join independent set of
directly join irreducible elements. If M and N are finite directly join independent
sets of directly join irreducible elements such that =VM VN, then there is a
one-to-one function from M onto N such that I [O, x ] and I [O, f ( x ) ]are projective
for al1 x E M .
Exercises 2.49
l. a. Prove that every element of a relatively complemented lattice of finite
height is the join of finitely many atoms.
b. Prove that if L is a bounded modular lattice and 1 is the join of finitely
many atoms, then every element of L is the join of finitely many atoms.
2. Prove Theorem 2.39-i.e., prove that the height function is a dimension
function.
3. The idea of this exercise is to reprove Theorem 2.37, following the lines of
the Zassenhaus-Schreier approach to the Jordan-Holder Theorem of group
theory. Thus, we need an analog of the Zassenhaus "Butterfly" Lemma

and of the Schreier Refinement Theorem. Let L be a modular lattice. Recall
from the proof of Theorem 2.37 that two chains a, < a, < <a,-, and
b, < b, < < b,-, are equivalent provided that n = m and the intervals
I[ai,ai+,] and I[bj,bj+,] can be matched in such a way that matching
intervals are projective. Finally, we say that the chain C' is a refinement of
the chain C in L iff C c C'.
a. Suppose that a, I a, and b, b, in L. Prove that I[a, v (a, A b,),
a, v (a, A b,)] and I[b, v (b, A a,), b, v (b, A a,)] are projective in-
tervals. [Pictorial hint: Draw a diagram of a lattice meeting al1 the require-
ments of this statement. There should be some resemblance to a butterfly
in this diagram.]
b. Let a 5 b in L. Using (a), prove that any two chains from a to b in L
have equivalent refinements.
c. Deduce Theorem 2.37 from (a) and (b).
4. Prove that if L is a semimodular lattice of finite height, then any two maximal
chains in L have the same length (i.e., Theorem 2.37 can be established, in
part, for semimodular lattices).
5. Let L be a semimodular lattice with a least element O. Let a and b be atoms
ofLandc~L.Provethatifc<avc<bvc,thenavc=bvc.
6. Let L be a modular lattice with least element O and let M be any set of
elements of L - {O}. Prove that M is directly join independent iff (VN) A
(VP) = V ( N í l P), for al1 finite N, P c M.
7. Let L be a modular lattice with least element O and let a,, a,, . , a, be n 1 +
distinct elements of L - {O}. Prove that {a,, a,, - , a,} is directly join
independent iff
(a, v a, v v ai) A ai+, = O
for al1 i < n.
8. Let L be the congruence lattice of the three-dimensional vector space over
the real numbers. Describe the directly join independent subsets of L and the
directly join irreducible elements of L. Do the same for the four-dimensional
vector space over the real numbers.
*9. In this exercise, we sketch an alternate approach to Corollary 2.48 that
essentially follows Ore's original path to the result. Let E be a modular lattice
of finite height and suppose that
For each i < n, let ¿ii= a, v v a,-, v a,+, v a -v a,-,.

.
a. Prove that (ai: i < n} is a directly meet independent set with n elements
and that O is the meet of this set, if a = l.
b. Let b E L. Prove V((b v a,) A a,) = (l\(b v Üi)) A (Va,).
Now suppose further that a = b, @ . @ b,-, and that al1 the si's and bis
are directly join irreducible. The definition of bj is similar to that for üi. The
goal is to prove that m = n and that there is a permutation f of {O, 1, m ,
n - 1) such that
1 2.4 Modular Lattices with Finite Chain Condition 77
1I for al1 i < n. This is accomplished by induction on the dimension d(a) of the
I element a. Check the initial step and then do the next steps to establish the
inductive step of the argument.
j
I c. Fix i < n. Prove that if there is k < m such that ai v 6, < a, then there is
I j # k such that
[Here is a hint: For each r < m, define c, = (ai v 6,) A b,. Argue that the
c,'s are directly join independent and let their direct join be e. Note that
e = a, O (e A si), as in the claim used in the lemma for the Direct Join
Decomposition Theorem (use (b)).Prove that e < a, and use the induction
hypothesis on the two direct join decompositions of e. Now, using the
dimension function and Theorem 2.46, finish this part (c).]
d. Now dispense with the hypothesis in (c) that ai v b, < a for some k. In
view of the interchangeability of the a,'s and b;s, this amounts to eliminat-
ing the possibility that ai v 6, = a and bk v ¿ii< a for al1 k < m.
e. Now complete the inductive proof of the assertion made before part (c)
above, and deduce from it Corollary 2.48.
10. (R. Freese) Prove that the lattice in Figure 2.1 1 is modular and that a and
c are directly join isotopic but not in one step.
Figure 2.11
2.5 DISTRIBUTIVE LATTICES
Just as modularity emerges as a fundamental property of the congruence lattices

of groups and algebras closely connected with groups, it turns out that the
congruence lattices of lattices themselves and of algebras closely connected with
them have the following stronger property:
a A (b v c) = (a A b) v (a A c ) for al1 a, b, and c.
This statement is called the distributive law, and lattices for which it holds are
called distributive lattices. The earliest lattices to be investigated were distributive
lattices; the obvious analogy with the distributive law for multiplication over
addition made this statement appealing. Indeed, some early writers in lattice
theory considered al1 lattices to be distributive. Every chain is easily seen to be
distributive, as is the lattice of al1 subsets of any given set. It is also clear that
every distributive lattice is modular, so al1 the conclusions of the last two sections
apply to distributive lattices. We shall see here that most of these results hold in
a much sharper form.
THEOREM 2.50. (Funayama and Nakayama [1942].) The congruence lattice

of any lattice is distributive.
Proof. Let L be a lattice. First observe that (4 A S/) v (4 A 0) I 4 A (S/ v 0)

holds in ConL, since it holds in every lattice. We will establish the reverse
inclusion. Our argument shares some features of the proof that the congruence
lattice of a group must be modular (Theorem 2.24). Let M(x, y, z) be the lattice
theoretic expression
Straightforward lattice arguments show that for al1 a, b E L
Now suppose that a(4 A (S/ v 0))b, where 4, S/, 0 E Con L. This means that
a4b and a($ v 0)b. According to Theorem 1.24 (ii), S/ v 0 is the smallest equiva-
lente relation of L that includes both S/ and 0. This equivalence relation is
obviously S / U S / ~ ~ U S / O ~ ~ S / U S / ~ ~ O Thus
$~Othere
U . -must
- . be a finite
sequence c,, c,, c,, . - , c, such that a = c,, b = c,, and
cii,hci+, if i is even and i < n
ciOci+, if i is odd and i < n.
Notice that for al1 i In, a = M(a, a, ci)4M(a,b, ci). This implies that
M(a, b, ci)(4 A $)M(a, b, ci+,) if i is even and i < n
M(a,b,ci)(4 A O)M(a,b,ci+,) i f i i s o d d a n d i < n.
Since a = M ( a , b,a) = M(a, b,c,) and b = M ( a , b, b) = M ( a , b,c,), we conclude
that a ( ( 4 A $) v (4 A O))b,as desired. Therefore, Con L is distributive.
This proof applies to a wider class of algebras than lattices. In fact, it applies
to any algebra for which there is a term M ( x ,y, z)that can be built up from the
basic operations and variables so that the three equations mentioned in the proof
are satisfied. This line of investigation will be taken up in 54.12 and pursued in
greater depth in our second volume.
Although every distributive lattice is necessarily modular, there are modular
lattices that fail to be distributive. The smallest such lattice is M,, diagramed in
Figure 2.12. I n M , we havea A b = Oanda A c = O but a A ( b v c) = a A 1 = a .
Figure 2.12
There are many statements equivalent to the distributive law. Some are
contained in the next theorem, while others can be found in the exercises below.
For the history of this theorem, consult pages 133-134 of Birkhoff [1948].
1 THEOREM 2.51. For any lattice L the following statements are equivalent:
i. L is distributive.
ii. a A ( b v c) 5 (a A b) v (a A c) for al1 a, b, c E L.
iii. a v ( b A c) = (a v b) A ( a v c) for al1 a, b, c E L.
iv. (a A b) v ( b A c) v (c A a ) = (a v b ) A ( b v c) A (c v a) for al1 a, b, c E L.
v. Foralla,b,c~L,ifa~c=b~candavc=bvc,thena=b.
vi. L has no sublattice isornorphic with either N , or M,.
I Proof.
I i. e ii.
i. v .
This follows easily, since (a A b) v (a A c) 5 a A ( b v c) in al1 lattices.
First, assume that L is distributive. Obtain (iv) as follows:
(a v b) A ( b v c) A (c v a)
= ( ( ( a v b ) A b) v ((a v b) A c ) ) A (c v a)
= ( b v ( ( a A c) v ( b A c))) A (c v a)
= ( b v (a A c ) ) A (c v a )
= ( b A (c v a ) ) v ( ( a A c) A (c v a ) )
= ( ( b A c) v (b A a ) ) v ( a A c)
= ( a A b) v ( b A c) v (c A a).
Chapter 2' Lattices
Now assume that (iv) holds. Then L is modular, since if a 2 c, then

(a v b) A c = (a v b) A ((b v c) A c)
= (a v b) A (b v c) A (c v a)
= (a A b) v (b A c) v (c A a)
= ( U Ab ) v ( b A c ) v a
= ((a A b) v a) v (b A c)
= a v (b A c).
Using the modularity of L, the distributive law can be deduced as

follows:
a A (b v c)
= (a A (a v b)) A (b v c)
= ((a A (C v a)) A (a v b)) A (b v c)
= U A (a v b ) ~ ( b c)
v ~ ( c v a )
= a A ((a A b) v (b A c) v (c A a))
= (a A ((a A b) v (b A c))) v (c A a) by modularity
= (a A (b A c)) v (a A b) v (c A a) by modularity
= (a A b) v (a A c).
i. e iii. A direct proof of this is left as an exercise. However, observe that (iii)
is the dual of (i). On the other hand, (iv)is its own dual. Since (i)e (iv),
we know that the class of distributive lattices is selfdual. So (i) and
(iii) are equivalent.
i. => v. Suppose that a A c = b A c and a v c = b v c. Then
v. vi. It is easy to find violations of (v) in both N, and M,. Thus any lattice
satisfying (v) cannot have a sublattice isomorphic to either of these
lattices.
vi. * iv. Since N, is excluded as a sublattice, we know that L is modular. We
argue the contrapositive. So suppose L fails to satisfy (iv). Since
(a A b) v (b A c) v (c A a) 2 (a v b) A (b v c) A (c v a) is true in
every lattice, we can pick elements a, b, c E L such that
(a A b) v (b A c) v (c A a) < (a v b) A (b v c) A (c v a).
Let d = (a A b) v (b A c) v (c A a) and u = (a v b) A (b v c) A
(c v a) and define
2.5 Distributive Lattices 81
. We contend that Figure 2.13 is a sublattice of L. That is, these five

elements are distinct and
Figure 2.13
Actually, verifying these equalities suffices, because they imply that

the five elements are distinct, in view of d < u. Moreover, from
modularity we have
a' = d v (a A u)
b' = d v (b A u)
c' = d v (c A u).
Because of al1 this symmetry, it suffices to establish, say,
Reason as follows:
a' A c'
= ((d v a) A u) A ((d v c) A u)
= ( d v a) ~ ( vdC ) A u
= ((a A b) v ( b A c) v (c A a) v a) A (d v c) A u
= ( ( b A c) v a) A (d v c) A u
= ((b A c) v a) A ((a A b) v c) A u
= ( ( b A c) v a ) A ((a A b) v c) A (a v b) A (b v c)
A ( C v a)
= ((b A c) v a) A ((a A b) v c)
= (b A c) v (a A ((a A b) v c)) b,y modularity
= (b A c) v ( ( ( aA b) v c) A a)
= (b A c) v ( ( a A b) v (c A a)) by modularity
= d. m
The import of this theorem is like that of the analogous Theorem 2.25
concerning modular lattices. Since the class of distributive lattices is specified by
a set of equations, it is a variety. Either (iii) or (iv) above tells us that the dual
of a distributive lattice is again distributive. Part (vi) characterizes the class of

distributive lattices by forbidding certain sublattices, offering some prospect for
using a Hasse diagram to check whether a lattice is distributive.
Let L be a lattice, a E L, and recall the maps da(x) = a A x and $,(x) = a v x.
Note that $a:L + I(a] and S/,: L -,I [a). These maps are always isotone, but
even in the modular lattice M, they fail to be homomorphisms. In distributive
lattices, the situation is nicer.
THEOREM 2.52. The following statements are equivalent for any lattice L:
ii. For any a EL, both da and $a are homomorphisms.
In modular lattices, projective intervals are isomorphic by a composition of

a sequence of such maps associated to a sequence of transposed intervals. In
general, these sequences can be arbitrarily long, with no way to shorten them.
In distributive lattices, the situation is nicer.
THEOREM 2.53. Let L be a distributive lattice. If I[a, b] and I[c, d] are

projective in L, then either these two intervals are transposes or there are intervals
I [u, u] and I [u', u'] such that
and
Proof. First, observe that in any lattice, if I [a,, bol transposes up to I [a,, b,],
which transposes up to I[a,, b,], then I[a,, bol transposes up to I[a,, b,] and
that a similar phenomenon happens for transposing down. The conclusion of
the theorem will follow if we can show how to "reverse the kinks" in a chain
of perspectivity maps. More precisely, we want to show that if I[a,, bol r
I [a,, b,] L I [a,, b,], then there are a, and b, such that I [a,, bol L I [a,, b,] r
I [a,, b,]. Just define
a, = a, A a,
and
b, = bo A b,.
From symmetry considerations, it is enough to demonstrate that
and
a, v b, = b,.
For the first equation, we have:
2.5 Distributive Lattices
a, A b, = a, A b, A b,
= a, A b,
= bO A a , A b,
= b, A a,
= a , A b, A b, A a,
= a, A a,
= a,.
For the second equation, we have:

a, v b, = a, v (b, A b,)
= b, A (a, v b,) by modularity
= bo A ( P o A a,) v b2)
= (b, A a,) v (b, A b,) by modularity
= b, A ( a , v b,) by distributivity
= b, A b,
= b,. ¤
Another useful property that holds for distributive lattices but not for
modular lattices in general is presented in the next lemma. Recall the notions of
join prime and meet prime from Definition 2.4.
LEMMA 2.54. In any distributive lattice, un element is join irreducible iff it is

join prime; it is meet irreducible $it is meet prime.
Proof. We concern ourselves only with the "join" aspects of the theorem. The
"meet" statement will follow, since the class of distributive lattices is selfdual. In
any lattice, join prime elements are always join irreducible. So let a be a join
irreducible element in the distributive lattice L. To see that a is join prime,
suppose that a i b v c. So a = a A (b v c) = (a A b) v (a A c),by distributivity.
Because a is join irreducible, either a = a A b or a = a A c. Hence, either a I b
or a i c.
THEOREM 2.55. Let L be a distributive lattice and let N, M c L where both

N and M are finite join irredundant sets of join irreducible elements. If VN
= VM,
then N = M .
Proof. Let a E N. Then a is join irreducible, and so, by Lemma 2.54, a is join
prime. Now a i VN = V M . Hence we can pick b E M with a I b, since a is join
prime. Likewise, we can pick c E N such that b i c. Therefore a i c. Since N is
join irredundant, we obtain a = c. Hence, a = b E M . Consequently, N E M. The
reverse inclusion is obtained by a similar argument. ¤
THEOREM 2.56. Let L be a distributive lattice with the descending chain

condition. For every element a EL, there is a unique join irredundant finite set N
of join irreducible elements such that a = V N .
Proof. The existence of the set N is guaranteed by Theorem 2.7, and uniqueness
is just a restatement of Theorem 2.55.
Theorem 2.56 has a much stronger conclusion than the Kurosh-Ore Theorem
(2.33), which holds more generally for al1 modular lattices. This simple result has
some use in algebraic geometry, and we will use it in the investigation of subdirect
representations of algebras with distributive congruence lattices.
As shown by M,, elements of modular lattices can have severa1 comple-
ments. This cannot happen in distributive lattices. In view of Theorem 2.51(v),
an element of a distributive lattice can have at most one complement relative to
any bounded interval. Thus complements and relative complements in distribu-
tive lattices are unique whenever they exist. According to Dilworth [1945], there
are nonmodular lattices in which every element has a unique complement. The
construction is very elaborate and is not included here. On the other hand, every
lattice in which relative complements are unique must be distributive, by Theorem
2.51. See Exercise 2.63(7) regarding uniquely complemented modular lattices.
The complemented elements in a distributive lattice can be used to decompose
the lattice.
THEOREM 2.57. Let L be a bounded distributive lattice and let a, a* E L where

a and a* are cornplements of each other. L r I(a] x I[a).
Proof. Define f : L -,I(a] x I [a):hby f (x) = (x A a, x v a) for al1 x E L. This f is

the desired isomorphism. Given distributivity, the demonstration that f is a
homomorphism presents no difficulty, so we omit it. To see that f is one-to-one,
suppose f(c) = f(b). This means that c A a = b A a and c v a = b v a. But then,
by Theorem 2.51 (v), c = b. Finally, f is onto I(a] x I [a): suppose c 5 a 5 b. Just
observe that
((b A a*) v c) v a =b
and
((b A a*) v c) A a = c.
So f((b A a*) v c) = (c, b). •
The converse of this theorem holds in the following sense. Suppose that
L = Lo x L, where Lo and L, are bounded lattices. Then the elements (1,O)
and (0,l) are complements of each other in L and Lo r I((1, O)] while L, r
IC(1,O)). Distributivity plays no role here. Even in the proof of the theorem,
the full power of distributivity is not needed to obtain the decomposition of L
into the direct product of other lattices. Later we will see how to decompose
relatively complemented lattices of finite length.
A complemented distributive lattice is called a Boolean lattice. In Chapter
1, we defined Boolean algebras in such a way that complementation was a basic
unary operation. Thus the relation between Boolean lattices and Boolean algebras
is like that between groups treated as algebras with one operation-the group
multiplication-and groups as we have introduced thein in Chapter 1. Homo-

morphic images of Boolean lattices are again Boolean lattices and direct products
of systems of Boolean lattices are also Boolean lattices. However, subalgebras of
Boolean lattices are not generally Boolean lattices. For example, the only chains
that are Boolean are those with one or two elements, but long chains are quite
common in most Boolean lattices. Theorem 2.57 can obviously be applied to
finite Boolean lattices to obtain the following result.
COROLLARY 2.58. Every finite Boolean lattice is isomorphic to a direct power

of the two-element chain. ¤
Another way to formulate this corollary is the following: Every finite Boolean
lattice is isomorphic to the lattice of al1 subsets of some set, where the join of two
subsets is just their union and the meet of two subsets is just their intersection.
This reformulation of the corollary is a simple consequence of the connec-
tion between sets and characteristic functions. Indeed, the members of the kth
direct power of the two-element chain with elements O and 1 can be regarded as
the characteristic functions defined on a k-element set. The correlation of subsets
with their characteristic functions is an isomorphism from the lattice of subsets
onto the direct power.
Neither the corollary nor its reformulation hold for arbitrary finite distribu-
tive lattices or for arbitrary infinite Boolean lattices in place of finite Boolean
lattices. But it is possible to accommodate these lattices by giving up only a little
of the power of the conclusion. It turns out that every finite distributive lattice
is isomorphic, in a rather special way, to a sublattice of a direct power of the
two-element chain (or, in the language of the reformulation, to a sublattice of the
lattice of al1 subsets of some set, consisting of certain kinds of subsets.)
Let J = (J, 5 ) be any ordered set. An order ideal of J is just a subset of J
that is closed downward. That is, I c J is an order ideal of J iff for al1 a and b
in J , if a E I and b 2 a, then b E 1. Let L be a lattice and let J(L) be the set of al1
nonzero join irreducible elements of L. The ordered set obtained by restricting
the ordering of L to J(L) is denoted by J(L), and the set of order ideals of J(L)
is denoted by Ord J(L). Evidently, (Ord J(L), fl, U ) is a distributive lattice.
Ord J(L) will denote this lattice. Also let I S O ( J ~ , C,) stand for the set of al1 isotone
maps from the ordered set J~into the two element chain C,. (Recall that the
superscript a indicates the operation of forming the dual of an ordered set.)
Evidently ISO(J~,C,) is a sublattice of the direct power C: of the two-element
chain. The connection between Ord J(L) and ISO(J~,C,), where J = J(L), is that
the characteristic functions of the order ideals are exactly the isotone maps, and
this correlation establishes an isomorphism.
The next theorem is due to Birkhoff [1933].
THE REPRESENTATION THEOREM FOR FINITE DISTRIBUTIVE

LATTICES (2.59). Let L be a finite distributive lattice and let J be the ordered
set of nonzero join irreducible elements of L. Then L E Ord J(L) lso(Ja, C,).
Moreover, each projection function on C: maps ISO(J~,C,) onto C,.
Proof. Our proof focuses on isotone maps rather than order ideals, so that the
last sentence of the theorem can be easily handled. In view of the remarks
preceding the theorem, the whole proof could be reformulated in terms of order
ideals.
Foreacha~L,define fa: J+(O,l)by.f,(b) = l i f b 5 aandf,(b) = Oifb$ a
for al1 b E J . For each a EL, f , is easily seen to be isotone from Jainto C,. Define
F: L -+ ISO(J~,C,) by
F(a) = f , for al1 a E L.
Let a, b E L and c E J . Notice
because c is join prime, and
With the help of these two observations, it is straightforward to argue that F is

a homomorphism.
Suppose fa = f,.Then the set of join irreducibles below a is the same as the set
of join irreducibles below b. Since every element of a finite lattice is the join of
the set of join irreducibles below it, we conclude that a = b.
CLAIM 2: F is onto Iso(Ja, C,).
Let h: J -+ {O, 11 be isotone (for Ja!). So h is order reversing for the order Ion
L. Let a = V{c: h(c) = 1). To see that h = fa, observe that for d~ J , d I a iff
h(d) = 1, because d is join prime and h is order reversing.
CLAIM 3: Let a E J and let pa be the associated projection function. Then pa
maps 1so(Ja,C,) onto C,.
Notice that, for b E L, pa(h) = pa(fb) = &(a) = 1 or O, depending on whether
aI b. Since a > O, we conclude that pa(fa) = 1 while pa(f,) = O. Hence pa is onto
C2. m
The last sentence of the theorem is an assertion that the given embedding
of L into the direct power of the two-element chain links L closely to each factor
of the power. In the language to be introduced in Chapter 4, this sentence asserts
that every finite distributive lattice is a subdirect power of the two-element chain.
Next, we extend this result to infinite distributive lattices. Let L be a lattice
and I be an ideal of L. I is said to be a prime ideal provided a A b E I implies a E I
or b E I for al1 a, b E L. This is equivalent to saying that I is a meet prime element
of Idl(L). In the proof above (join) primeness played a crucial role. We will
replace it with primeness of ideals. In place of the fact that every element in a
finite distributive lattice is the join of join irreducible elements, we will need to
know that in any distributive lattice, every ideal is the intersection of prime ideals.
What we need is gathered in the next theorem, due to M. H. Stone [1936]

and A. Tarski [1930].
THE PRIME IDEAL THEOREM FOR DISTRIBUTIVE LATTICES (2.60).

Let L be a distributive lattice.
i. If I is an ideal of L and F is a filter of L such that I and F are disjoint, then
there is a prime ideal J of L such that I E J with J disjoint from F.
ii. Every ideal of L is the intersection of the prime ideals that include it.
Proof. To prove (i), we let H be the collection of ideals containing I but disjoint
from F. By Zorn's Lemma, pick a maximal member J of H. We need to demon-
strate that J is prime. For the sake of obtaining a contradiction, suppose not.
Pick a, b E L so that a A b E J but neither a nor b belongs to J . By the maximality
of J , ( J v I(a]) í l F and (J v I(b]) í l F are both nonempty. So pick c, d E J such
that a v c and b v d belong to F. Since F is a filter, (a v c) A (b v d ) F.~ By
distributivity, (a v c) A (b v d) = (a A b) v (a A d) v (c A b) v (c A d), so this
element belongs to J as well. This contradicts the fact that J and F are disjoint.
To prove (ii), we let I be any ideal of L. If I = L, there is little to prove. So
we suppose that I is a proper ideal. According to (i), for each a $1, there is a prime
ideal Jasuch that I E Jaand a $ Ja.Clearly, I is the intersection of al1 these prime
ideals. •
Let L be a distributive lattice and P(L) be the set of al1 proper prime ideals
of L. P(L) is ordered by set inclusion, so let P(L) denote this ordered set. The
collection of al1 order ideals of P(L) is evidently a'distributive lattice, with set
union taken for the join and set intersection taken for the meet. Cal1 this lattice
Ord P(L). (Notice that the elements of Ord P(L) are collections of prime ideals
of L.) Once more, the connection between order ideals and their characteristic
functions yields OrdP(L) E lso(pa, C,), where P is taken as P(L). This means
that the theorem below, due to G. Birkhoff [1933] and M. H. Stone [1936], can
be reformulated to give a representation of any distributive lattice L by sets under
the operations of intersection and union.
THE REPRESENTATION THEOREM FOR DISTRIBUTIVE LATTICES

(2.61). Let L be any distributive lattice and let P be the set qf proper prime ideals
of L ordered by set inclusion. L can be embedded into lso(pa, C,) in such a way
that the projection functions, restricted to the image of L, are onto C,.
Proof. For each a E L, define f,: P -t (O, 1) byf,(I) = O iff a E I. It is easy to check
that f ,E lso(pd, C2).Define F: L-t lso(pa, C,) by F(a) = f, for al1 a E L.
Let I be a prime ideal and a, b E L. Observe that
and, because I is prime,

a A b ~ I i f~
f Eor I~ E I .
These two observations lead immediately to the conclusion that F is a
homomorphism.
Suppose F(a) = F(b). So f,(I) = f,(I) for al1 proper prime ideals I. This means
that a and b belong to exactly the same prime ideals. By the Prime Ideal Theorem
(ii), I(a] = I(b]. Therefore a = b, so F is one-to-one.
CLAIM 2: Let I belong to P and let p, be the associated projection function.
Then p, o F maps L onto C,.
Just note that (p, o F)(a) = p,(F(a)) = p,(f,) = f,(I) = O or 1, depending on
whether a belongs to I or not. a
As with the finite case, we also obtain the result that every distributive lattice
is a subdirect power of the two-element chain. The line of reasoning yielding the
Representation Theorem for Distributive Lattices can easily be modified for
Boolean algebras, yielding the following theorem due to M. H. Stone [1936].
THE REPRESENTATIQN THEOREM FOR BOOLEAN ALGEBRAS (2.62).

Every finite Boolean algebra is isomorphic to a direct power of the two-element
Boolean algebra. Every Boolean algebra is embeddable into a direct power of the
two-element Boolean algebra in such a way that each projection function of the
direct power maps the image of the Boolean algebra onto the two-element Boolean
algebra. U
Of course, another version of this theorem is the statement that every

Boolean algebra is isomorphic to a Boolean algebra composed of subsets of a
set with the operations of set intersection, set union, and set complementation.
The story of representation of distributive lattices and of Boolean algebras
does not end here. In fact, underlying these representation theorems are more
powerful results. Notice that, in the proofs above, each distributive lattice was
correlated, in a natural way, with an ordered set. (For finite distributive lattices,
this was the set of al1 nonzero join irreducible elements, whereas for infinite
distributive lattices, this was the collection of al1 proper prime ideals.) In turn,
each ordered set was correlated with a distributive lattice-its lattice of order
ideals. Unfortunately, for infinite distributive lattices L, we obtain only the
conclusion that L can be embedded into OrdP(L). In general, P(L) will not
determine L up to isomorphism. But it turns out that P(L) can be given a
topology, and the resulting topological ordered set does determine L up to
isomorphism. Indeed, not only will there be a natural one-to-one correlation of
distributive lattices with certain kinds of ordered topological spaces, but, roughly
speaking, this correlation matches lattice homomorphisms with continuous maps
(in the reverse direction), lattice ideals with open subsets, and lattice filters with
closed subsets. The reader may well imagine that many results abo-utdistributive
lattices and Boolean algebras can now be deduced using the concrete setting of
sets. The deeper correlation with topological ordered sets allows topological
tools to be brought to bear on lattice-theoretic problems. These correlations,
known as the Stone and Priestley dualities, are taken up in a later volume.
Exercises 2.63
1. Let L be the lattice of natural numbers with the meet of two numbers taken
to be their greatest common divisor and their join taken to be their least
common multiple. Prove that L is a distributive lattice.
2. Prove that the complemented elements of any distributive lattice comprise
a subuniverse of that lattice.
3. Prove that in any distributive lattice with a O, direct join isotopy coincides
with equality and that a set is directly join independent iff every pair of
distinct elements of it meet to O. Finally, prove that a finite directly join
independent set of which O is not a member must be join irredundant. Thus
the Direct Join Decomposition Theorem for distributive lattices is a corollary
of Theorem 2.55. .
4. a. Let L be a finite distributive lattice and let J(L) be the set of join irreducible
a
elements of L. Prove that the length of any maximal chain in L is 1 J(L)I.
b. Prove that in any finite distributive lattice, the number of join irreducible
elements and the number of meet irreducible elements is the same.
5. (Nachbin [1947]) Let L be a distributive lattice. Prove that L is relatively
complemented iff every proper prime ideal of L is a maximal ideal. [Hint:
The Prime Ideal Theorem is useful in establishing both implications.]
*6. Prove that every finitely generated distributive lattice is finite.
7. Prove that every uniquely complemented modular lattice is distributive.
[Hint: Pick x, y, Z E L such that x A z = y A z and x v z = y v z. Pick a
complement u of x A z in I [O, z ] and then a complement u of x v z in I [u, 11.
Prove that u is a complement of both x and y.]
8. Prove that in a distributive lattice, no interval can be projective with a proper
subinterval of itself.
9. a. Construct two countably infinite nonisomorphic Boolean lattices Lo and
L, such that OrdP(L,) z OrdP(L,).
b. Prove that if Lo and L, are finite distributive lattices such that J(L,)
J(L,), then Lo r L,.
10. Let Lo and L, be finite distributive lattices and let J, and J, be their respec-
tive ordered sets of nonzero join irreducible elements. Let h be a homomor-
phism from Lo into L, such that h preserves top and bottom elements. Let
J(h) be the map with domain J, defined by J(h)(b) = A(x: h(x) 2 b} for al1
b€J1.
a. Prove that J(h) is an isotone map from J, into J,.
b. Prove that J(h) is onto J, iff h is one-to-one.
2.6 CONGRUENGE RELATIONS O N LATTPCES
For every lattice L the lattice Con L is a distributive algebraic lattice, according
to Theorem 2.50 and Corollary 1.23. In this section we will examine the con-
gruences of lattices in general and investigate ConL for particular kinds of
lattices L.
Let L be a lattice and let 8 E Con L. It is a simple but useful observation that
the congruence classes modulo 8 are convex sublattices of L. It is also useful to
notice that a 8 b iff a A b 8 a v b for al1 a, b E L. These remarks mean that 8 is
determined by the pairs (a, b) belonging to 8 with a I b.
THEOREM 2.64. Let L be a lattice and 8 be a reflexive binary relation on L.

8 is a congruence relation of L $for al1 a, b, c E L,
i. a e b $ a A b 8 a v b.
ii. If a b I c, a 8 b, and b 8 c, then a 8 c.
iii. I f a < b a n d a 8 b , t h e n a ~ c 8 b n c a n d a v c 8 b v c .
Proof. Al1 three of these properties are clearly possessed by congruence relations,
so we will only deal with the converse. Property (i) and the commutativity of A
and v yield the symmetry of 8. According to (iii), if a c 5 b and a 8 b, then
a 8 c and c 8 b-a kind of convexness. To see the transitivity of 8, suppose a 8 b 8 c.
It will follow from this convexness that a e c , provided we first prove that
(a A b A c) 8 (a v b v c). Observe that we have the following string of inclusions
and that the formulas resulting from replacing I by 8 in these inclusions are
also true:
Therefore, two applications of (ii) yield the desired conclusion. Hence 8 is transitive
and, thus, an equivalence relation.
In checking the Substitution Property, we will only deal with v ,leaving the
similar argument for A aside. Since transitivity is already verified, we need only
show that a 8 b implies a v c 8 b v c. Property (i) tells us that a A b 8 a v b.
From property (iii), we get (a A b) v c 8 a v b v c. But a v c and b v c belong
to I[(a A b) v c, a v b v c]. So the convexity yields a v c 8 b v c. ¤
Intuitively, L/@is formed by collapsing the congruence classes, which are

always convex, to points. In a lattice, any congruence relation that collapses a
given interval may of necessity collapse others. Consider the transposed intervals
I[a A b,a] and I[b,a v b]. I f a 8 a A b, then b = b v (a A b)8b v a, so collapsing
I [a A b, a] to a point forces I [b, a v b] to collapse to a point. More generally,
if 8 collapses an interval I to a point, then it must also collapse to points al1
intervals projective with I; it is also clear that the congruence must collapse al1
intervals projective with subintervals of the original interval. Generally, however,
this does not give a satisfactory account of the situation. The reader 'should easily
verify that collapsing any nontrivial interval of M, collapses the whole lattice to
2.6 Congruence Relations on Lattices 91
a point. To obtain a characterization of congruences in lattices, we modify the

notion of projectivity.
DEFINITION 2.65. Let L be a lattice and let I [a, b] and I [c, d] be intervals
in L. We say that I [a, b] transposes weakly down into I [c, d] iff b = a v d and
c i a. We say that I[a, b] transposes weakly up into I[c, d] iff a = b A c and
b d. We denote these weak transposes, respectively, by I[a, b] L, I[c, d] and
I [a, b] rM'I [c, d]. (See Figure 2.14.) We say that I [a, b] transposes weakly into
I [c, d] whenever I [a, b] transposes weakly either up into I [c, d] or down into
I [c, d]. Finally, we say that I [a, b] is weakly projective into I [c, d] iff there is
a finite sequence I[a,, bol = I[a, b], I[a,, b,], I[a,, b,] = I[c, d] such that
. a - ,
I [a,, b,] transposes weakly into I [a,,, ,b,,,] for al1 i < n.
I [a, hl ,,,I [c, d l I[a,h] r"'I[c,d]

Figure 2.14
The relation "weakly projective into" between intervals is transitive and

reflexive, but it is not symmetric. Recall that cgL(c,d) denotes the principal
congruence of L generated by the pair (c, d).
THEOREM 2.66. (R. P. Dilworth [1950].) Let L be a lattice with a 5 b and

c S d in L. Then (a, b) E cgL(c,d) iff there is a finite sequence
a = e, i e, i e, i 5 e, = b
such that I[e,, e,,,] is weakly projective into I[c, d] for al1 i < n.
Proof. Let 8 be the binary relation defined on L by u 8 v iff there is a finite

sequence as described above, where a = u A u and b = u v v. It is clear from the
definition of weak projectivity that
(c, d ) E e G cgL(c,d).
We shall show that 8 = cgL(c,d). It suffices to show that 8 is a congruence
relation. Since 8 is4obviously reflexive, we will do this by verifying the three
properties listed in Theorem 2.64. Properties (i) and (ii) are immediate. For the
purposes of property (iii), let a = e, I . I e, = b be as described above and
let f ~ L . N o t e t h a t a ~ f = e , ~ f i - - . ~ e , ~ af n=d bt h~a ft I [ e , ~ f ,
e,,, A f ] is weakly projective into I[ei, e,,,] and hence into I[c, d]. Joins work
similarly. m
While this characterization of principal congruences in lattices is useful, it

can be sharpened in certain classes of lattices. A lattice L is said to have the
projectivity property iff whenever I [a, b] is weakly projective into I [c, d], then
I [a, b] is actually projective with a subinterval of I [c, d]. For lattices with this
property, principal congruences can be characterized as in Theorem 2.66, except
that each I [ei, e,,,] is projective with a subinterval of I [c, d].
EXAMPLE 2.67. A lattice without the projectivity property.
Proof. Let L be the lattice diagrammed in Figure 2.15. The interval I [b, a]
transposes weakly down to I [O, c], which transposes up to I [f, e]. So I [b, a] is
weakly projective into I [f, e]. But b is meet irreducible, so it is impossible for
I[b,a] to transpose up to any other interval. Since c is the only element that
gives a when joined with b, I [d, c] is the only interval to which I [b, a] transposes.
Similarly, since c is join irreducible and b is the only element that gives d when
met with c, we find that I[b, a] is the only interval to which I[d, c] transposes.
Hence I[b, a] is not projective with I[ f, e] (or any of its subintervals), even
though it is weakly projective into I [f, e]. •
Figure 2.15
THEOREM 2.68. Every modular lattice and every relatively cornplemented lattice
has the projectivity property.
Proof. First let L be relatively complemented. Suppose that I [a', b'] transposes
up to I [c, d] weakly and that a' Ia bI b'. We argue that I [a, b] is projective
with a subinterval of I[c,d]. Let a* be a complement of a in I[al, b]. Now just
note that
so I [a, b] L I [a', a*] and
since a' = b' A c and a' I a* I b'. So I [a', a*] r I[c, a* v c]. Therefore I [a, b]
is projective with I[c, a* v c], a subinterval of I[c, d]. The proof can now be
completed by a straightforward induction on the length of the chain of weak

projectivity.
The case when L is modular is an easy consequence of Dedekind's Trans-
position Principle. m
In distributive lattices, principal congruence relations have an especially

simple form.
THEOREM 2.69. Let L be a distributive lattice and let I[c, d ] be an interval

in L. (a,b)€cgL(c,d) iffa A c = b A c and a v d = b v d, for al1 a, EL.
Proof. Let 8 = ((a, b): a A c = b A c and a v d = b v d). First we prove that

8 is a congruence relation. 8 is clearly an equivalence relation on L. Suppose
a 8 b. Then
and evidently
So (a v e) 8 (b v e) for al1 e E L. Meets behave in a similar fashion. Therefore, 8

is a congruence relation.
Since (c, d ) é 8, al1 that remains is to prove that 8 S 4 for every congruence
4 which collapses c and d. So suppose a 8 b. Then
and so 8 E 4, as desired. m
The next theorem presents a distinctive property of the congruence lattices

of lattices. Roughly speaking, it says that the congruences on the direct product
of two lattices can be taken apart into congruences on the factor lattices.
THEOREM 2.70. If Lo and L, are lattices, then

Con (Lo x L,) r ConL, x Con L,.
Proof. Actually, part of this theorem is true for algebras in general:
Con Lo x Con L, is embeddable in Con (Lo x L,).
Define h: Con Lo x Con L , -+Con ( L o x L,) by

( a , , a l ) h ( ~ o , ~ l ) ~ b o ,iffa,0,b,
bl) and a, @lb,.
We leave it to the reader to verify that h is actually an embedding. For algebras
in general, this embedding can fail to be onto Con ( L o x L,), but for lattices it
is always onto. To see this, let m(x,y, z) stand for the expression ( x v y) A z,
p(x, y) stand for x A y, and q(x,y) stand for x v y. In any lattice, the following
equalities then hold for al1 x and y:
To see that h is onto Con(L, x L,), let 0 ~ C o n ( L x, L,). Define 8, on Lo by

a 8, b iff ( a , c ) 0 ( b , c ) for some c E L ,.
CLAIM: a O o b i f f ( a , c ) 8 ( b , c ) f o r a l l c ~ L l .
Suppose ( a , d ) 0 ( b , d ) and that c E L , . Now, ( p ( a ,b),c ) 0 ( p ( a ,b),c ) and
(q(a,b),c ) 0 ( d a , b),c ) . Hence
( a , c> = (m(a7 p(a7 b),q(a7 b)),m(d7 c, 4 )
= m((a,4,(p(a, b),c ) , (q(a7 b),c ) )
e m(@, 4,( p ( a ,b),c ) , (q(a7 b),c ) )
= (m(b7 p(a, b),i ( a ,b)),m(d,c7 4)
= P(b,a),q(b7 a)),m(d7 c, 4 )
= (b,c).
Using the claim, it is easy to prove that 8, E Con Lo. In a similar way, we define
8, E Con L,. Now to see that h(0,, 0,) = 0, just follow the implications below:
(a,, a1 )h(0070,) (b07b l ) * a, 0, b, and a , 8, b1
* ('07 c ) 0 (b,, c ) and ( d , a , ) (d7 b, )
for al1 d E Lo and al1 c E L ,
* (a07a1> 8 (b07a1 ) 8 (b07bi )
* ('07 ) 0 @o, bl ).
Hence h(O,, O,) E 0.
Now suppose (a,, a , ) 0 (b,, b, ). So
= b1 )7 ( ~ ( ~ 0 7 ),(q(b07 al ))
= ~ ( ~ 0 q(bo>
, U,)),
= (bo,a,).
Thus a, 0, b,. Sirnilarly, a, 8, b, . Therefore h(8,, 0,) = 8. E
This theorem holds for a wider class of algebras than lattices. In fact, the
crucial requirement is the existence of severa1 expressions [m(x, y, z), p(x, y), and
q(x, y)] built from variables and the fundamental operations for which certain
equalities hold in both the factor algebras. This theorem supplies us with a
strategy for describing ConL for certain lattices L. In the first step, we try to
write L, up to isomorphism, as a direct product of lattices that cannot themselves
be written as direct products of other lattices. The second step then consists of
analyzing Con L in the case that L cannot be factored further. Let A be an algebra.
A is directly indecomposable iff A has more than one element and if A B x C
then either B has only one element or C has only one element.
Carrying out the first step in our strategy is easy, if we impose some kind of
finiteness condition on L.
THEOREM 2.71. Every lattice of finite height is isomorphic to a direct product

of finitely many directly indecomposable lattices. •
This theorem can be proven by a straightforward induction on height. The

proof is left in the hands of the reader.
The second step in the strategy requires more work and additional hypotheses
to obtain useful conclusions. We cal1 an algebra A simple iff Con A has exactly
two elements. Every simple algebra is directly indecomposable (just consider the
kernels of the projection functions), but it is not hard to invent finite modular
lattices that are directly indecomposable (by virtue of having a prime number of
elements) but fail to be simple.
THEOREM 2.72. (R. P. Dilworth [1950].) Every directly indecomposable rela-

tively complemented lattice of finite height is simple.
Proof. Let L be a directly indecomposable relatively complemented lattice of

finite height. Let 8 be a congruence on L different from the identity relation. Pick
u < v so that u 8 v. Let u* be a complement of u in ICO, u]. Now observe that
o=uAu*~vAu*=(uVU*)AU*=U*#O.
Thus, ( x : O 8 x # O} is not empty. Since L has finite length, in view of Theorem

2.6 pick a to be a maximal element of (x: 0 ex}. Since this set is an ideal, it is
easy to see that a is the largest element in the ideal. Theorem 2.57 suggests that
L might be decomposable as I(a] x I[a). That theorem requires that a have a
complement and that L be distributive. We are still able to carry out roughly the
same argument, the difficult point being to establish enough "distributivity."
CLAIM O: x 8 y iff (x A y) v a' = x v y for some a' I a.
Suppose that x 8 y. Let a' be a complement of x A y in I[O, x v y]. Then a' =
a' A (x v y) e a ' A (x A y) = O. So a ' < a by the maximality of a. The converse
is immediate.
CLATM 1: a v (x A y ) = ( a v x) A (a v y ) f o r a l l x , y ~ L .
Observe that
Chapter 2 Lattices
and so
x A y 8 (a v x) A (a v y).
By Claim O, pick a' < a so that
((x A y) A (a v x) A (a v y)) v a' = (x A y) v ((a v x) A (a v y)).
Using the absorption axioms, we obtain
(x A y) v a ' = (a v x) A (a v y).
Finally, using this equation, we obtain
and therefore a v (x A y) = (a v x) A (a v y). Thus, a possesses some distribu-

tivity in L. We need to know that a satisfies the dual of Claim 1 as well. To
accomplish this, we will characterize the property in Claim 1 in a selfdual way.
Cal1 an element b distributive iff b v (x A y) = (b v x) A (b v y) for al1 x, y E L.
CLAIM 2: b is distributive iff no nontrivial subinterval of I(b] is projective
with a nontrivial subinterval of I[b).
Suppose first that b is distributive and that I[x, y] is a subinterval of I(b] and
I [u, u] is a subinterval of I [b) such that I [x, y] and I [u, u] are projective. Define
(b = {(z, w): z v b = w v b). Because b is distributive, (b is a congruence relation
on L. Now x (b y, so u (b v. But this means that u = u v b = u v b = v. Therefore,
I [u, u] is trivial and since I [x, y] is projective with I [u, u], it follows that I [x, y]
is trivial, too.
For the converse, suppose that b is not distributive. Pick x and y so that
b v (x A y) < (b v x) A (b v y). Let $ = cgL(O,b).Then b v (x A y)$(b v x) A
(b v y). Use Theorem 2.66 to pick a sequence
such that Ice,, e,,,] is weakly projective into I(b] for each i < n. Since b v
(x A y) < (b v x) A (b v y), choose j < n with e, < e,,, . So I[e,, e,,,] is a non-
trivial subinterval of I[b) that is weakly projective into I(b]. By Theorem 2.68,
I[ej, e,+,] is projective with a subinterval of I(b], as desired.
CLAIM 3: a ~ ( x v y ) = ( a ~ x ) v ( a ~ y ) f o r a l l x , y ~ L .
CLAIM 4: If a A x =a A y and a v x = a v y, then x = y.
Observe that
2.6 Congruence Relations on Lattices
By Claim O, pick a' L a so that
Therefore a' S y and a' i y A a = x A a i x. This means that a' < x A y. Con-
sequently, x A y = y. By a similar argument, x A y = x. Therefore x = y.
Now define h: L -+ I(a] x I [a) by h(x) = (x A a, x v a). Easy computations

using Claims 1 and 3 reveal that h is a homomorphism. Claim 4 is virtually the
statement that h is one-to-one. The fact that h is onto I(a] x Ira) follows as in
Theorem 2.57, using Claims 1 and 3 above in place of the distributivity of L.
Since L is directly indecomposable and a # O, we deduce that Ira) has only one
element. Therefore a = 1 and 8 is L x L. Hence L has just two congruences; it
is simple. ¤
COROLLARY 2.73. Every relatively complemented lattice of finite height is

isomorphic to a direct product of simple relatively complernented lattices of finite
height. ¤
G. Birkhoff [1935b] and K. Menger [1936] had earlier established that

every complemented finite dimensional lattice is isomorphic to a direct product
of simple lattices. The complemented finite dimensional simple modular lattices
turn out to be the two-element chain and the subspace lattices of nondegenerate
finite dimensional projective geometries. See 54.8 for an account of this important
result.
Combined with Theorem 2.70, Corollary 2.73 yields the following corollary.
COROLLARY 2.74. (R. P. Dilworth [1950].) If L is a relatively complemented

lattice of finite height, then Con L is a Boolean lattice. •
The corollary can be drawn as well from the following theorem.
THEOREM 2.75. If L is a lattice with the finite chain condition and the pro-
jectivity property, then Con L is a Boolean lattice.
Proof. First suppose a < b in L. Let c < d and (c, d) E cgL(a,b). According to
Theorem 2.66, pick a finite sequence c = e, el . 5 e, = d such that I [ei,
is projective with a subinterval of I [a, b]. Since a < b, there are no proper
subintervals, and therefore Ira, b] is projective with some subinterval I [ei,
of I [c, a]. Again by Theorem 2.66, we have (a, b) €cgL(c,d), SO cgL(a,b) is
an atom in ConL. Since L has the finite chain condition, there is a finite
maximal chain from O to 1; let O = do 4 d, < < d, = 1 be one such chain.
Apparently,
cgL(O,1) = cgL(d0,d,) v cgL(d,,d,) v v cgL(d ,-,,d,).
This means that Con L is a bounded distributive lattice in which the top element
is the join of finitely many atoms. By Theorem 2.40, Con L is relatively comple-
mented. Therefore Con L is a Boolean lattice. ¤
COROLLARY 2.76. If L is a modular lattice of finite height, then ConL is

Boolean. ¤
Exercises 2.77
1. Prove that if a -< b in the lattice L and 8~ Con L, then either a/8 = b/8 or
a/8 -< b/8 in L/O.
2. Prove that every lattice of finite height is isomorphic to a direct product of
finitely many directly indecomposable lattices of finite height. Does a similar
assertion hold for lattices satisfying the ascending chain condition?
3. Give an example of a finite directly indecomposable modular lattice that is
not simple.
4. Let L be a lattice. For each ideal I of L , define
O ( I ) = { ( a , b ) : a v c = b v c for some C E I }
and for each congruence 8 E Con L , define
I ( 8 ) = { a : a/8 Ib/O in L/B for al1 b E L ) .
a. Prove that if L is distributive and I is an ideal of L , then O ( I )E Con L .
b. Prove that I ( 0 )E Id1 L , for al1 8 E Con L.
c. Prove that the following are equivalent:
ii. For each ideal I of L , O ( I )E Con L and I ( O ( I ) )= I.
iii. Every ideal of L has the form I ( 0 ) for some 8 E Con L.
5. Let L be a lattice and a E L . a is said to be standard iff
The element a is said to be neutral iff
Recall that a is said to be distributive iff
Thus an element is called distributive, standard, or neutral, depending on

whether a certain part of the distributive law is true when that element is
present.
a. Prove that a is distributive iff { ( b , c ) : a v b = a v c} is a congruence
relation on L.
b. Prove that a is standard iff a is distributive and for al1 b, c E L, if a A b = a A c
and a v b = a v c, then b = c.
c. Prove that a is neutral iff for any b, c E L the sublattice generated by {a,b, c}
is distributive.
d. Prove that every neutral element is standard and that every standard
element is distributive.
e. Prove that if L is modular, then every distributive element is neutral.

f. Prove that the set of neutral elements of L is the intersection of the maximal
distributive sublattices of L.
g. Prove that if L is a complemented modular lattice then a is neutral iff a has
a unique complement.
C H A P T E R T H R E E
Unary and Binary

Operations
3.1 INTRODUCTION
In the two previous chapters, we introduced the reader to some of the fundamen-
tal kinds of algebras with which we will be dealing throughout this work, such
as lattices, semilattices, and Boolean algebras. Before turning to a formal presen-
tation of the basics in Chapter 4, we devote this chapter to some other kinds of
naturally occurring algebras. Such examples are central and important, because
they help provide the diversity that is necessary for the discovery of general
results in algebra.
The idea of an algebra allows arbitrarily many operations of arbitrary rank,
and yet, as we remarked at the beginning of Chapter 1, the surprising fact is that
al1 of the classical algebras are built with unary and binary operations, especially
binary ones. Moreover, except for R-modules, the classical algebras require only
one or two binary operations and usually no unary operations. (For eac'h r in
the ring R of scalars, R-modules have one operation f, of multiplication by r.)
The same is true for the lesser-known algebras we will describe in this chapter.
Experience has thus shown that almost al1 of the diversity of the theory of
algebras already occurs for one or two binary operations. In this chapter, we
provide a representative sampling of this diversity and at the same time describe
some special kinds of algebras that will be useful for our later work.
Before leaving this introduction, we will briefly address the question, which
we mentioned in Chapter 1, of whether there is any mathematical basis for
thinking of binary operations as special. (We will address this question more
systematically when we come to study the algebraic representation of lattices,
monoids, and groups in later volumes.) We may especially ask whether a single
binary operation is inherently different from some other finite collection of
operations, such as a single ternary operation or two binary operations, and
whether a finite collection of operations is inherently different from an infinite
collection.
To understand the relationships between operations of various arities, it
helps to have the notion of a term operation of an algebra A = ( A , F,, F,, . ),
3.1 Introduction 101
which we now present somewhat informally. (For a more formal presentation,

see Definition 4.2 in 54.1.) he set of term operations is the smallest set that
contains each Fi, that contains, for O 5 i < n, the coordinate projection operation
and that is closed under composition. (Another way of saying this is that every
operation formed by composition from projections and A-operations is a term
operation of A, and nothing else is a term operation of A.) Thus, for example, if
G is an n-ary term operation and H,, Hn are m-ary term operations, and if
S ,
H is defined by H ( x ) = G(H,(F), . H,(F)), then H is also an m-ary term opera-

S ,
tion of A. A term operation F of A may also be called a derived operation of A,

and we sometimes say that F is derived from the operations F,, F,, and so on.
The easiest distinction is that between unary operations and those of higher
rank. Obviously, composition of unary operations yields only unary operations.
It is not much more difficult to check that every operation F derived from unary
operations depends on only one variable-i.e., for each such F there exists j < k
and a unary operation f such that F(x,,. xk-,) = f (xj) for al1 x,,
S , , xk-, .
Obviously, a cancellative binary operation (such as a group operation) on a set
of more than one element cannot depend on only one variable and hence cannot
be derived from unary operations.
For another intrinsic difference between unary operations and those of
higher rank, we mention the obvious fact that the subuniverse lattice of every
unary algebra is distributive (see Exercise 1 below). On the other hand, the
subgroups of even a very small group such as Z, x Z, form a nondistributive
lattice.
When we consider how operations are derived from binary operations,
however, the facts are altogether different, for we have the following Theorem of
W. Sierpinski [1945].
THEOREM 3.1. For every set A, every operation F on A is derivable from

( A , F,, F,, - ) for some binary operations F,, F,, on A.
Proof. Given a set A and F : Ak -+A, we need to show how to build F from
binary operations. There are two constructions of F, depending on whether A is
finite or infinite. The one for infinite A is straightforward, and we give the reader
some hints for finding it in Exercise 2 below.
For finite A, a number of proofs are known. It is even possible to show that
for each finite set A there exists a single binary operation B such that every F
can be built from B (Exercise 3 below), but as far as we know the proof of this
fact is fairly complicated. At the cost of using two binary operations and 1 Al
unary operations instead of a single binary operation, we will provide an entirely
elementary proof of Sierpinski's Theorem.
Obviously, if the result holds for one set of N elements, then it holds for al1
sets of N elements, and so we may assume that A is the set N = (0,1, s., N - 11,
with + and defined on A by ordinary addition and multiplication modulo N.
We then take our algebra to be (A, +, t,, (a E A)) where, for each a E N, t, is
S ,
the unary operation defined by
In other words, our algebra is just a ring with extra unary operations. Now, given
an arbitrary k-ary operation F, we simply observe that
for al1 x,, a x, E N. (Here ü abbreviates the vector (a,, - ,a,). The formula
* ,
appears to go beyond our definition of derived operation, since it involves

multiplication by F(ü), but certainly operations of the form m . x can be replaced
+
by appropriate sums. Thus 2 - x is x + x, 3 x is x + (x x), and so on.) This
construction of F completes the proof of the theorem.
Sierpinski's Theorem does not take into consideration certain important

questions about algebras whose fundamental operations have rank 2, 3, and so
on. For example, if we have an algebra (A, F), with F ternary, we may use
Sierpinski's Theorem to build F from binary operations B,, B,, . . . , but we must
realize that the algebra (A, B,, B,;. a may not be very closely related to the
)
original algebra (A,F). It is clear that we will have Con(A, B,, B,;..) c
Con(A, F), but the opposite inclusion will hold only exceptionally. (Similar
remarks apply to the subuniverse lattice and other constructs important in
algebra.) As one indication of the limitations of a single operation acting alone,
we mention a recent result of R. McKenzie that certain finite lattices are not
isomorphic to the congruence lattice of any finite algebra (A, F) that has a single
operation F (of any arity); for example, this holds for the lattice
(Nevertheless, according to some recent results of W. A. Lampe and W. Feit, such

an (A, F) exists if we allow A to be infinite, and there exist F,, F,, on a set A
of 3 1 elements such that (A, F, ,F, , ) has this congruence lattice.)
On the other hand, we might mention a result of McKenzie's, from the 1960s,
that if F,, F,, - . F, are operations on A, of arbitrary ranks, then there exists a
unary operation f and a binary operation B defined on a set A' such that
Con(A1,f, B) is isomorphic to Con(A, F,, - .,F,,).(See Exercise 19 at the end of
$4.4 and Exercise 10 at the end of $4.6. It is unknown whether f is really essential
here.) Moreover, similar statements hold for subuniverse lattices, and so on; we
will consider results of this kind in detail in a later volume. As far as infinitely
many Fi's are concerned, the situation is certainly more complicated, and again
we will address such questions in later volumes; suffice it to say that an infinite
set of operations is not generally replaceable by a finite set of operations.
3.2 Unary Algebras 103
For one more example of a (possible) intrinsic difference between groupoids

(algebras of type (2)) and algebras of a more general finite type, we look ahead
to 55.1, where we will introduce the notion of a prime algebra in a class X (usually
taken to be closed under the formation of isomorphic algebras and of products).
An algebra A is said to be prime in a class Y of algebras iff 1 Al > 1 and A satisfies:
if B, C E X and A divides B x C, then A divides B or A divides C. Here, "A divides
B" means that B r A x D for some D. According to a result of R. McKenzie
[1968], the class of al1 finite groupoids contains prime algebras-notably, it
contains every finite group that is directly indecomposable. On the other hand,
it remains open whether there is any prime in the class of al1 finite algebras of
any fixed type that has two or more operation symbols of rank greater than zero.
Exercises 3.2
l. Prove the assertion in the text above that if f,, f2, - are unary operations
on A, then the subuniverses of (A, f,, f,, ) form a distributive lattice. (Hint:
Prove that the union of two subuniverses is again a subuniverse.)
2. Prove Sierpinski7sTheorem for an infinite set A. (Hint: Use the fact that every
infinite set A is in one-to-one correspondence with A2. That is, there exist a
binary operation B: A2 + A and unary operations 4 and $ such that the
maps
are mutually inverse. Thus, given a ternary operation G on A, we may define

a binary operation F via F(x, y) = G(x, 4(y), $(y)).)
"3. (D. L. Webb [1936]) Prove that the binary operation
x + 1 (modn) i f x = y
ifx#y
where + denotes addition modulo n, will generate al1 operations on
(O;--,n -1).
*4. (R. W. Quackenbush [1971]) Define an operation F: A" -+A to be idem-
potent iff F obeys the identity F(x, x, - ..,x) Ñ x. Prove that if 1 Al 2 3, then
every idempotent operation on A can be built from idempotent binary
operations on A.
5. (R. W. Quackenbush [1971]) Prove that there exists an idempotent ternary
operation on a two-element set that cannot be built from idempotent binary
operations on that set.
3.2 UNARY ALGEBRAS
As we saw in 53.1, unary operations are fundamentally different from those of

higher rank, and as we shall see throughout this work, the more substantial and
104 Chapter 3 Unary and Binary Operations
more "algebraic" questions always seem to revolve around operations of rank

two or more. (Consider, for instance, the limitations involved in writing an
equation with unary operations. The only possible forms are flf2 fm(x)Ñ
g g2 . gn(x),and the same thing with the second x changed to y.) Nevertheless,
unary algebras merit a brief study in their own right, for many important features
of an algebra A (such as Con A) are determined by its family of unary polynomial
operations. (See Theorem 4.18.) As we will see, algebras with one unary operation
are much simpler than those with more than one, so we will begin our study with
these.
Algebras with one unary operation and no other operations, otherwise known
as mono-unary algebras, are particularly easy to describe informally. We simply
malce a diagram which has an arrow going from a to f(a) for each a E A. For
example,
denotes the successor function on o.In the finite case, such a diagram can in
principle contain al1 the points of A and can therefore constitute a complete and
precise definition of f : A -+ A. If A is infinite, however, or even finite but large,
such a diagram can supplant a formal description of f only if there is a clear
pattern discernible in the fragment that we are able to draw. A typical function
on a finite set might be given by the diagram
In attempting to classify such diagrams, the first significant thing to notice

is that A obviously can be divided into disjoint component diagrams, each
corresponding, of course, to a subalgebra of (A, f ). For an official definition, we
say that x and y are connected under f, or in the same component of (A, f ) iff
f "(x) = f "(y) for some m, n 2 O (where we take f O(x) to mean x). It is not hard
to check that this defines an equivalence relation on A; we will refer to the blocks
(equivalence classes) of this relation as connected components of (A, f ). Again,
it is not difficult to see that each connected component is a subuniverse of A on
which the diagram of f is connected, in the sense that each two points x and y
are connected by a path of arrows. (That is, there exist a,, - a, such that x = a,,
S ,
y = a,, and for each i < n, either f(a,) = or f(a,+,) = a,.) Equivalently, as
the reader may verify, a mono-unary algebra is connected iff any two nonempty
subuniverses have a point in common. Now since each mono-unary algebra is
seen to be a disjoint union of connected subuniverses, our description will be
complete if we describe al1 connected mono-unary algebras.
Obviously, every one-generated mono-unary algebra is connected, and
things will be easiey for us if we first describe this easy special case. If a generates
(A, f ), then A = { f "(a): m E o).If f "(a) # f "(a) for m # n, then (A, f ) has the
form
that is, it is isomorphic to (u, a), where a is the successor operation. Otherwise,
we may assume that f '"(a) = f "+,(a) for some m and some k > O. Let us suppose
that in fact m is the smallest integer for which such an equation holds, and that,
for this value of m, k is the smallest positive integer for which the equation holds.
Now one easily observes that, when restricted to the set {a,f(a), f 2(a), # ,
f"+'-'(a)}, the operation f has the form
Evidently this diagram describes a subuniverse of (A, f ), which must be al1 of

(A, f ), since this algebra is generated by a. Thus we see that every one-generated
mono-unary algebra has the form (A,) or (A,,,). The algebras that are depicted
as (A,) and (A,,,) will be denoted A, and A,,,.
We now define a core of a connected mono-unary algebra (A, f ) to be any
nonempty subalgebra on which f is a one-to-one function. Examining the forms
(A,) and (A,,,) above, we see that every one-generated algebra has a core
isomorphic to A, or A,,,, for some k. (It is entirely possible to have k = 1, in
which case the core is simply a fixed point of f.) Hence every connected mono-
unary algebra has at least one core of type A, or A,,,.
LEMMA. Assume that (A, f ) is connected. If (A, f ) has any finite core C, then
C must be of type (A,,,), that is, a finite cycle for f, and, moreover, C is the only
core of ( A ,f ).
Proof. Let C be a finite core of (A, f ). By the above, (A, f ) has some core D
of type (A,) or of type (A,,,). C ílD is a finite nonempty subuniverse of the core
D, and hence type (A,) is impossible. We will now prove that every core E of
( A , f ),whether finite or infinite, is equal to D. It is evident that this will complete
the proof of the lemma.
Since D is a finite cycle that has E í l D as a nonempty subuniverse, it follows
that E n D = D (i.e., that D c E). We will now prove by contradiction that E = D.
By connectedness, if D were a proper subuniverse of E, we would have f "(e) =
d E D for some e E E - D and some n. Taking n as small as possible will assure
us that in fact f "-'(e) $ D; replacing e by f "-'(e) will yield an e E E such that e 4 D
but f (e) = d E D. Now it is easily seen that f fails to be one-to-one on the subset
{e,f'-'(d)} of E, in contradiction to the fact that E is a core. This contradiction
completes the proof that E = D. M
Of course, the lemma implies that if one core is infinite, then they are al1
infinite, but the reader may easily observe that (@,a) has many subalgebras
isomorphic to itself, so the core is not unique (although any two cores intersect
in a core).
Now for our description of al1 connected mono-unary algebras, we let C be
any core of (A, f ), and observe that (by the lemma) there are no finite f-cycles
in A - C. Therefore A - C may be ordered by defining a < b iff f "(b) = a for
some m. The set M of minimal elements in this order consists of those a 6C such
that f (a)E C. Since (by connectedness) each one-generated subuniverse ( b ,f (b),
e meets the core C , it is clear that each element of A - C must lie above some
)
a E M. Now obviously, for each b E A - C, the elements of A - C that are Ib

form a finite chain, b 2 f (b) 2 f 2(b)2 2 f m(b)= a E M . Let us define a rooted
tree to be an ordered set T with a least element O (called the root of T) such that
the interval ITIO, t ] is a finite chain for each t E T. Now it should be clear that
the ordered set A - C is a disjoint union of rooted trees (whose roots form the
set M).
We have now supplied every detail of our description of connected mono-
unary algebras. We collect these details into the following result.
THEOREM 3.3. Every mono-unary algebra is a disjoint union of connected

mono-unary algebras, in a unique way. Every connected mono-unary algebra (A, f )
has a subuniverse C with (C, f ) isomorphic either to ajinite cycle, or to co with
the successor function. The set A - C can be given the order of a disjoint union of
rooted trees so that f acts on A - C by mapping the root of each tree into C and
mapping each other element of A - C into its unique predecessor.
Two examples of connected mono-unary algebras are depicted below. A z

can be formally described as (Z, f ), where Z is the set of integers and f ( n ) =
n + l . The other connected mono-unary algebra depicted here has a core
consisting of a single (fixed)point and thus consists wholly of a single rooted tree.
In case f happens to be a permutation rc of A, the description given in

Theorem 3.3 above takes a particularly simple form. The finite components of
(A, rc) obviously consist of various algebras A,,, (i.e., finite cycles), since the addi-
tion of any rooted tree would result in n's not being one-to-one. As the reader
may easily check, an infinite component of (A,n) must be isomorphic to the
algebra A z introduced just above. This algebra may be thought of as the core
algebra A , to which has been appended an infinite linear rooted tree. On the
other hand, we may also observe that A z is a core of itself. In fact, with the
introduction of A z , we now have al1 isomorphism types of cores. (In Exercise 4
below, it is established that every core is isomorphic either to Az or to A, or to

A,,, for some k.) Therefore, we have the following corollary to Theorem 3.3.
COROLLARY. If n is a permutation of A, then (A,n) is uniquely a disjoint

union of algebras isomorphic to AZ or to A,,, for some k. 4 .
The individual components of (A, n), for n a permutation of A, are closely

related to the well-known cyclic decomposition of n. Before stating this connec-
tion in detail, let us recall a few things about permutations. We cal1 a permutation
n of A cyclic iff (A, n) has one component isomorphic to Az or to some A,,,,
and al1 other components trivial @.e.,consisting of fixed points). In the finite case,
this is equivalent to saying that the elements of A that are moved by 7c can be
arranged in a finite sequence a,, a,, a,-, such that n(aj) = aj+, for (O 5 j I
k - 2), and n(a,-,) = a,. We will follow the usual practice of denoting such a
permutation as
(This sequence denoting n is in fact uniquely determined except for the possibility
of starting with a different a,. For example, the above permutation could just as
well be denoted (a,c-,,a,, a,, . . . ,a,-,).) A cyclic permutation may be called a
cycle for short. Cycles (a,, - ,a,-,) and (b,, - - ,b,.-,) are called disjoint if no ai is
equal to any bj. We multiply permutations by composing them as functions-
i.e., np(x) = n(p(x)). It is not difficult to check that two cycles permute with each
other if and only if they are disjoint or one is a power of the other.
For the cyclic decomposition of a permutation n, we will assume that (A, n )
is the disjoint union of its connected subalgebras (A,, nlAo), m(A,-, ,nlAs- ).
,
We then define nj (O Ij < S) to be the permutation of A that acts like n on Aj,

and like the identity elsewhere-i.e.,
( ) if x E Aj
rcj(X) =
otherwise.
It is easy to check that each 9is a cyclic permutation of A, that the permutations
nj commute with one another, and that their product is n. Moreover, these
conditions determine the permutations 3 uniquely up to a permutation of their
indices j. We will in fact follow the common and useful practice of writing
permutations as products of disjoint cycles. For instance, if p = (24531) and
a = (256) (as, say, permutations of a),then we may express the product po by
its cyclic decomposition as (123)(456).
Turning now to unary algebras with more than one operation, we may
observe that the above unique decomposition into connected algebras remains
valid, if we take "(A, f,,,f,, a is connected" to mean that A is not a disjoint
)
union of proper subuniverses. Nevertheless, there is no way to capture anything

like the remainder of Theorem 3.3, even for algebras with just two unary opera-
tions. Our attention will therefore be focused on one special lcind of unary algebra
with more than one unary operation.
If each unary operation fi is a permutation of A, then (A, f,, f,, admits

S )
a description in terms of a group of permutations on A, and thus our understand-

ing of such unary algebras is complete, modulo our understanding of permuta-
tion groups. For such algebras, the family of al1 operations generates a subgroup
S of the group Sym A of al1 permutations of A. Unless our similarity type is
already dictated by other considerations, the nicest way to describe such an
algebra will be to take each permutation f E S as a fundamental operation.
Nevertheless we will find it useful to consider such algebras from a slightly more
general point of view, which we will now describe.
Let G be an arbitrary but fixed group. A G-set is a unary algebra
(A, fg(g E G)), which obeys the following axioms:
i. f,(x) Ñ x for e the unit element of G;
ii* fg(fh(4) f g h W for 9, h f G.
Evidently, if the axioms are satisfied, each fg has the two-sided inverse fg- l , and
hence is a permutation of A. Thus the map g Hfg is a homomorphism from G
to the group S of permutations of A, which was mentioned in the previous
paragraph; in fact, this assertion is equivalent to axioms (i) and (ii). (In the
previous paragraph, the map g t+ fg was one-to-one; the "greater generality" to
which we alluded just above lies in allowing this map to be an arbitrary homo-
morphism.) Although we asserted above that unary identities can take on only
a limited number of forms, let us note here that they are certainly interesting in
the case at hand, since equations (i) and (ii) will serve as a definition of (a group
isomorphic to) G.
As with mono-unary algebras, our description of a G-set begins with its
decomposition into components. The connected components of a G-set are often
called its orbits; it is not hard to check that x and y are in the same orbit if and
only if y = &(x) for some g E G. A connected G-set (i.e., one consisting of a single
orbit) is often called a transitive G-set.
To complete our description of G-sets, we will presently prove that every
transitive G-set is isomorphic to a special kind of G-set that is determined by a
subgroup of G. Here, the description is not quite so completely pictorial as the
one for mono-unary algebras, but it is algebraically easier to work with. For any
subgroup K of G, we define the quotient G-set G/K to be the G-set ({hK: h E G),
fg (y E G)), whose universe consists of al1 left cosets hK and whose operations are
defined via f,(hK) = (gh)K. We leave it to the reader (Exercise 9 below) to check
that the conjugacy class of the subgroup K is determined by the isomorphism
class of the quotient G-set GIK.
If (A, fg (g E G)) is any transitive G-set, then, for some b E A (in fact for any
b E A) we have A = {f,(b): g E G}. For each b E A, we define the stabilizer or
isotropy group of b to be the subgroup
Hb = { ~ E Gfg(b)
: = b}.
of G. To establish an isomorphism of (A,& ( g G))
~ with the quotient G-set
G/Hb, we may define 4: G/Hb -,A via 4(gHb)= g(b). We leave it to the reader
to check that 4 is well defined, one-to-one, onto, and a homomorphism of G-sets;

thus 4 establishes an isomorphism of our given G-set with GIH,.
We now collect the facts we have proved about G-sets.
THEOREM 3.4. Every G-set is isomorphic to a disjoint union of subalgebras,

each of which is isomorphic to the quotient G-set G/K for some subgroup K of G.
The partition into such subalgebras is unique, and for each such subalgebra the
subgroup K is unique up to an inner automorphism of G.
COROLLARY. If G is a Jinite group and A is a transitive G-set, then 1 Al is a

divisor of ( G 1. m
We remark that for each monoid M (semigroup with unit), there is a variety
of M-sets that is defined similarly to that of G-sets for a group G; nevertheless,
there is no analog of Theorem 3.4 for M-sets. In fact every unary algebra can be
conceived of as an M-set for an appropriate M. (See Exercise 3.8(13) of $3.3
below.)
We close this section on unary algebras with an indication of the kind of
unusual behavior that can be found in algebras with only two unary operations
f and g. An algebra A is called a Jónsson algebra iff it is infinite, has only
countably many operations, and every proper subalgebra of A has power less
than 1 Al. Jónsson algebras of power N, are not very difficult to come by, but in
higher powers they are relatively rare. Here we present a simple example, due to
F. Galvin, of a bi-unary Jónsson algebra of power N,. (In Exercise 13 below, we
will see that one unary operation will not suffice for a Jónsson algebra of power
N1
For the construction, we first recall from the Preliminaries that col denotes
the first uncountable ordinal, which may be construed as the collection of al1
countable ordinals. We define two unary operations f and g on col as follows. g
is the successor operation on o , . For any countable limit ordinal A, f maps the
set
{A + 2k: k E o}
+ +
onto the (countable) set A + o [ = {y: y < A + o}], and f(A 2k 1) = A for al1
k E o . It is easy to check that the subuniverses of (o,, f, g ) are precisely the limit
ordinals A < o, (i.e., the sets of the form {x: x < A}), and thus that this is a
Jónsson algebra.
Since no other opportunity arises naturally in this volume, we will briefly
discuss the subject of (not necessarily unary) Jónsson algebras. By and large, this
subject remains mysterious and caught up in axiomatic questions of set theory.
For instance, the set-theoretic axiom V = L implies that Jónsson algebras exist
in al1 powers, but, on the other hand, no Jónsson algebra can have measurable
cardinal. In Exercise 16 below, we will outline a result (of Galvin, Rowbottom,
Erdos, Hajnal, and Chang) that, for each cardinal K, if there exists a Jónsson
groupoid of power K,then there also exists a Jónsson groupoid of power K'. For
$
more detailed references and more results, see either pages 469-470 of Chang
and Keisler [1973] or pages 120- 135 of Jónsson [1972].
Very little is known about what Jónsson algebras can occur in most of the
familiar varieties, although, for example, T. Whaley proved [1969] that no lattice
is a Jónsson algebra. There exist Jónsson groups, but for years the only known
examples were the groups ZPm.(For each prime p, Z,, is the group isomorphic
2ni
to the group of al1 complex numbers e 7 under multiplication.) 0. J. Johnson
realized in 1930 that these are the only commutative Jónsson groups. In [1980]
A. Ju. Ol'shanskiT proved the existence of a countably infinite torsion group G
with the following subuniverse lattice:
The atoms of this lattice correspond to the one-generated subgroups of G, which

are al1 finite, and hence it is obvious that G is a Jónsson group. (We have heard
that E. Rips constructed a group with these properties plus the additional prop-
erty that every element of G has order p for a fixed prime p. This is a spectacular
(but not the first) contradiction to the conjecture of Burnside, which held that
finitely generated groups obeying a law x n Ñ e (n 2 1) are always finite. We will
return briefly to the subject of Burnside's conjecture in the chapter on equational
theories and free algebras in Volume 2.) A Jónsson group of power N, was
constructed by Shelah [1980].
Exercises 3.5
1. For any algebra A, prove that there exists a smallest equivalence relation
8 on A such that each block of 8 is a subuniverse of A. Moreover, if A is
unary, then the blocks of 8 are the connected components of A.
2. Prove that a unary algebra is connected iff it is not a disjoint union of two
proper subuniverses.
3. Prove that for each unary algebra A, there is a unique decomposition of A
into a disjoint union of connected subuniverses.
4. Prove that every core is isomorphic to Az or to A, or to A,., for some k.
5. Describe, in terms of our pictorial representations, how many elements are
required to generate a mono-unary algebra A.
,
6. Give pictorial representations for the product algebras A,, x Al,,, A,, x ,
and Al,, x A,.
7. Show that if a unary algebra A has at least n components that have more
than one element (and possibly some one point components as well), then its
congruence lattice contains a sublattice isomorphic to the Boolean algebra
of 2" elements.
3.2 Unary Algebras
8. Establish a natural one-to-one correspondence between mono-unary alge-

bras and Z+-sets, where Z+ stands for the monoid of non-negative integers
under addition.
9. Prove that if H and K are subgroups of G, then the quotient G-sets GIH
and G / K are isomorphic iff H and K are conjugate-Le., iff K = gW1 Hg for
some g E G.
10. Prove that if A is the largest quotient G-set, namely G/{e), then Con A is
isomorphic to the lattice of al1 subuniverses of G. What about Con G/K for
K an arbitrary subgroup of G?
11. Prove that for every unary algebra A there exists an infinite algebra B
in HSP(A) with al1 operations trivial (i.e., each operation f of B satisfies
either f (x) Ñ x or f (x) Ñ f (y)). What is the congruence lattice of such
a B?
12. Prove the following assertion, which is a little stronger than an assertion
made in the text of this section. If a and p are cyclic permutations on X and
a p = pa, then either a and fl are disjoint or a = pk for some k.
13. Prove that there is no mono-unary Jónsson algebra of power N,.
"14. Let rc be a cardinal. Prove that if (rc, F) is a Jónsson algebra, with F
binary, then the ternary operation G defined as follows on rc' makes
( K + , G) into a Jónsson algebra. The ordinal rc' has the well-known prop-
erty that if rc il < rc', then 111= 1 lcl (rc and A have the same cardinality).
Therefore, for each such A, there exists a binary operation F, such that
(A, F,) is a Jónsson algebra. Now for (A, a, P) E ( K + ) ~ ,we define G(A, a, 8 )
to be F,(a, P) if this is defined (i.e., if rc < A and a, /? < A), and to be O
otherwise.
"15. Prove that the term z(x, y, z) =
1 is universal in the following sense. For every infinite set A and for every
ternary operation G defined on A, there exists a binary operation + on A
that satisfies the equation G(x, y,z) Ñ z(x, y,z). (It is possible to base an
elementary proof of this fact on naive set theory; in Volume 2 we will return
to the subject of universality.)
16. Prove that for every ternary operation G on an infinite set A, there exists a
binary operation F on A such that Sub(A, F) G Sub(A, G). Conclude that
if ( A , G) is a Jónsson algebra, then so is (A, F). (It thus follows from
Exercise 14 that if (K,F) is a Jónsson algebra of type (2), then there exists
a binary operation F' on rc' such that (rcf, F') is a Jónsson algebra.)
*17. Generalize the result of Exercise 16 to algebras of arbitrary countable
similarity type. That is, if ( A , F,, F,, .
e ) is a Jónsson algebra with opera-
tions Fi for i E CO, then there exists a single binary operation F defined on A
such that (A, F) is a Jónsson algebra. (For this, one will need to develop
the analog, for an infinite collection of terms z(x,, x,, . . of the universality
e),
result of Exercise 15. This subject will be covered in Volume 2. For the
moment, for a proof of the necessary universality result, we refer the reader
to J. Loi [1950].)
3.3 SEMIGROUPS
As we mentioned in Chapter 1, a semigroup is an algebra S = (S, consisting

S )
of a set S together with an associative multiplication on S. We follow tradition

in writing multiplication in semigroups (and in groups, and so forth) with infix
notation; that is, we normally write x y or simply xy instead of (x, y). An
associative operation is, of course, one that obeys the associative law
Our main purpose in this section will be to introduce a few special kinds of
semigroups that are of interest in the study of universal algebra.
A semigroup with unit is a semigroup S obeying the axiom
3yvx x - y Ñ y - x N x.
Even without associativity, it follows easily that if such a y exists, it is unique.
The unique such y is called the unit of S and is usually denoted e or 1. A monoid
is a semigroup that has a unit element explicitly mentioned in the similarity
type-i.e., an algebra (M, .,e) of type (2,l) that obeys the associative law
for and moreover obeys the equations
X - eN e - x Ñ x.
For a single algebra, the notion of monoid is more or less equivalent to that of
semigroup with unit, but as soon as we consider two or more algebras together,
there are important distinctions between the two notions. For instance, a sub-
algebra of a monoid is always a.monoid, but the same is not true for semigroups
with unit.
The prototypical example of a monoid is of course the collection of al1
selfmaps of a set A (i.e., al1 functions f:A + A, with e the identity function and
f - g the composition of functions f o g). We will denote this monoid by End A,
since its universe is the set End A introduced in 51.2 (if we think of A as an algebra
with no operations). It is almost trivial to verify the associative law for the
composition of functions; we leave this to the reader. Interestingly, this fact can
be seen as the source of associativity for al1 semigroups, since, as we will see in
53.5, every monoid is isomorphic to a submonoid of End A for some A.
The submonoids of End A most important to our subject are those of the
form End A, for A an algebra with universe A. As we said in Chapter 1, the monoid
EndA, also known as the endomorphism monoid of A, is defined to be the
submonoid of End A whose universe is the set End A consisting of al1 homo-
morphisms from A to A. An important part of our subject has been concerned
with the possibility of representing an arbitrary monoid as isomorphic to an
endomorphism monoid End A, with A to be found in certain restricted classes
of algebras. We will present one elementary result on this topic in 53.5, saving a
more systematic study for a later volume.
A related monoid, equally important but less frequently encountered, is the
monoid Part A of al1 partial functions from A to A (i.e., al1 functions f:B -+ A
with B G A). (Here, f .g is the partial function defined as follows. If x is in the
3.3 Semigroups 113
domain of g and g(x) is in the domain of f, then we include x in the domain of

f.g and define f g(x) to be f (g(x)). Moreover, the domain of f g consists of
+
these x and no others.) Clearly, we have End A E Part A. An even larger monoid
is the monoid Bin A of al1 binary relations on A-i.e., al1 subsets of A x A. Here
R S is defined to be the relational product
R o S = ((x, 2): 3y[(x, y ) R~ and (y, z) ES]).
(To embed Part A in Bin A, we need to map each partial function f to the set of
ordered pairs { (f (x), x): x E A}; the more traditional representation of f via pairs
(x, f(x)) causes o to take on different meanings for functions and relations.)
Notice that Bin A is a reduct of the relation algebra Re1 A that we described near
the end of 51.1. The relational product will be introduced again in 54.4; see
Definition 4.28.
Another easily defined and useful semigroup is the free semigroup on a set
A. We will in fact define the free monoid on A, denoted F,(A), observing that the
free semigroup is formed from the free monoid by simply deleting the unit
element. The elements of FA(A) are al1 finite sequences of elements of A, including
the empty sequence, which we denote u. We define the product a p of sequences
a = (a,, . . ,a,-,) and p = (b,, - .. ,bm-,) to be the sequence y = (e,, ,cm+,-,),
where
Usually, we drop the formal apparatus of finite sequences and write members of
the free monoid as "words" a,a, - Thus, for example, the product of abc and
a .
bcd is abcbcd. This form of writing two words together is often referred to as
concatenation of two words, and the product we defined above for sequences is
called the concatenation of two sequences. It is obvious that for 0,the empty
word, both concatenations a n and u a are equal to a, and so is the unit element
of the free monoid. The associativity of the free monoid is more or less obvious
from the informal idea of concatenation of words. This intuition can easily be
made into a formal proof that invokes our definition for the product of two
sequences; we leave the details to the reader.
In Chapter 4 we will present the general notion of free algebra. Here we
only mention that the free monoid, as we have described it, satisfies one of the
defining properties of free algebras. (This is the next theorem.) It is in fact rare
for a free algebra to be so simply describable as is the free monoid.
THEOREM 3.6. If M = (M, .,e) is any monoid and f : A -+ M is any function,

then there exists a homomorphism 4: F,(A) -+ M such that $(u) = e and 4(a) =
f (a) for al1 a E A.
Proof. Ifa=(a,,.-.,a,-,),wesimplydefine~(a)tobe f(a,)-f(a,)..-S-f(a,-,).

We leave to the reader the details of checking that this definition yields a
homomorphism with the desired properties. m
We go on to mention a few of the many interesting varieties of semigroups.

Perhaps the most interesting of these for our purposes is the variety of semi-
lattices, which was already mentioned briefly in Chapter 1. A semilattice is a
semigroup that is commutative (xy Ñ yx) and idempotent (xx Ñ x), and a semi-
lattice with unit is a monoid that obeys the two axioms. Quite often (e.g., in
Chapter 1 of this book), to emphasize the connection with lattice theory, the
binary operation of a semilattice is written A or v . In fact, as the reader may
easily check from $1.1 above, if (L, A , v ) is a lattice, then (L, A ) and (L, v )
are both semilattices (most likely not isomorphic to each other). Moreover, the
natural one-to-one correspondence, described in $2.1,between lattices and lattice
orders extends easily to a natural correspondence between semilattices and
semilattice orders. Unfortunately, there are two ways to do this, depending on
whether we conceive of the semilattice operation as meet ( A ) or as join ( v ) in
the partial ordering. The reader is warned that in much of the literature on this
subject there occur the expressions "meet-semilattice" and "join-semilattice."
These expressions refer to the writer's having selected one of the two possible
ways of defining an order in a semilattice. From the point of view of pure algebra,
there is no meaningful distinction between a meet-semilattice and a join-
semilattice-they are defined by the same axioms, which we stated above. We
will in fact allow ourselves to use these two expressions when it seems useful or
appropriate.
From the "meet" point of view, the correspondence between semilattices
and order relations goes as follows. (The other version is dual to this one and
hence omitted.) A semilattice order on a set A is an order with the additional
property that every two-element subset {a,b> of A has a greatest lower bound
in A. As in $2.1, if a A b is defined to be this greatest lower bound for al1 a and
b, then A becomes a binary operation that is easily verified to be a semilattice
operation on A. For the reverse correspondence, we may start with a semilattice
operation on A and define a I b for a, b E A to mean that a A b = a. Now it is
very easy to establish, along the lines of $2.1, that these definitions establish a
one-to-one correspondence between semilattice operations on a set A and semi-
lattice orders on A; we leave the details to the reader.
As we remarked in $1.1, one of the simplest examples of a semilattice is the
collection P ( X ) of al1 subsets of a set X, with the operation í l of set-intersection.
In our next result, we will see that every semilattice is isomorphic to a sub-
semilattice of a semilattice of this type. In this regard, semilattices are much
simpler than lattices. (For another viewpoint on this result, see Corollary 2 to
Theorem 4.44.)
THEOREM 3.7. Every semilattice S = ( S , A ) is isomorphic to a subalgebra of

(P(S),n >-
Proof. For each s G S, we define 4(s) to be {x E Slx I S}, where I is as defined
above. To see that 4 is one-to-one, we will assume that 4(s) = 4(t) and prove
that s = t. We first observe that s E #(S), and so S E 4(t), which means that s It.
Similar reasoning shows that t I S, and hence s = t. To see that is a homo-
3.3 Semigroups 115
morphism, we first observe that, in any meet-semilattice S, by definition of

"greatest lower bound" we have, for al1 x E S,
From this we readily calculate:

+(S A t ) = {XESIX< S A t)
= {XESIXI S and x I t)
= {XESIX S ) ~ ) ( X E ~ ( X t)
= n
If L = (L, A , v ) is a lattice, then one can obviously use the order relation
<
- to infer the lattice structure from the semilattice (L, A ). Nevertheless, from
the point of view of considering more than one algebra at a time, the semilattice
structure (L, A ) cannot be regarded as an adequate substitute for the full lattice
L. (Technically, as we will put it in Volume 2, "the variety of lattices is not
interpretable in the variety of semilattices.") For instance, as we will see in
Exercises 8 and 9 below, the H and S operators act very differently on (L, A )
and (L, A , v ).
The reader may at this time be interested in writing his own version of
Theorem 3.6 for the free semilattice FiP(A)on a set A. Suffice it for now to say
that the statement and proof are very much like those above, except that F9(A)
is the set of al1 cofinite proper subsets of A, and the operation is intersection.
Each of the two laws defining semilattices among semigroups, namely the
commutativity and idempotency laws, is interesting in its own right. For some
interesting information about commutative semigroups, see Exercises 16 and 17
of $4.5 and Exercise 4 of $5.1. Idempotent semigroups are sometimes also known
as bands. The very simplest bands are those obeying the law xy Ñ y (right-zero
semigroups) or its dual, xy Ñ x (left-zero semigroups). These semigroups are
really completely trivial, since the "multiplication" is nothing more than either
the first or second coordinate projection. Somewhat less trivial are the rectangu-
lar bands, which are idempotent semigroups obeying the additional law xyz Ñ xz.
It is not hard to prove that every product of a left-zero semigroup with a right-zero
semigroup is a rectangular band, and in Exercise 10 we will ask the reader to prove
the converse, namely that every rectangular band is isomorphic to a product of
two semigroups, one of which is a right-zero semigroup and the other a left-zero
semigroup. (This also follows from-and is a special case of-the idea of a
decomposition operation, which will be presented in $4.4; see Definition 4.32 and
the lemma that follows.) In particular, a rectangular band of prime order must
be either a left-zero or a right-zero semigroup.
As one further example of a variety of semigroups, we consider a variety
Y described by J. A. Gerhard in [1970] and [1971]. It is defined by the laws
x(y-4 Ñ (xy)z
XX ÑX
xyx Ñ xy.
It is not difficult to check that -tr contains the three-element semigroup S, that
has the following multiplication table:
The semigroup S, is useful to us, in that it can show us how terms act in V. The
semigroup terms (i.e., the formal expressions in semigroup theory) are of course
the finite formal products xilxi, ximof variables. The identities of Y allow us
to shorten any term that has a repeated variable, as follows. If two occurrences
of a variable stand together, one of them can be removed by applying the law
xx Ñ x. If two occurrences are separated, as in xu, u,x, the second occurrence
can be removed with the law xyx Ñ xy, by taking y to be u, u,. Thus every
semigroup term can be reduced, using the laws that define Y , to a product of
distinct variables. Now the semigroup S, can be used to show that no further
reductions of terms are possible for this variety. We ask the reader to show, in
Exercise 11 below, that if a is a semigroup word that contains no variable twice,
and p is also such a word, then either a = ,!?or it is possible to substitute values
0,1, and 2 for the variables appearing in a and j such that upon this substitution,
a and p evaluate in S, to two different elements of (O, 1,2). For another look at
this variety, see Exercise 28 of $4.11.
The product of two algebras can be thought of as defining a binary operation
on the class of al1 algebras of a given similarity type. If we identify two algebras
that are isomorphic to each other, then this operation is commutative and
associative, since A x (B x C) is isomorphic to (A x B) x C, and A x B is
isomorphic to B x A. The one-element algebra clearly serves as a unit element,
so we have a commutative monoid of isomorphism classes of algebras under direct
product. The structure of this monoid is actually very complicated; it contains
every countable commutative monoid as a submonoid. (See 65.1 for a reference
to a proof of this result of J. Ketonen.) Nevertheless, some simplifications are
possible (this will be a major focus of Chapter 5) if we look at some special
submonoids of the monoid of isomorphism classes. For instance, the submonoid
of isomorphism classes of finite algebras of a given type is isomorphic to a
submonoid of (a, 1)". For some even smaller classes of algebras, including
S ,
some very natural generalizations of finite groups to universal algebra, the

monoid of isomorphism types has unique factorization and is thus isomorphic
to a submonoid of ( m , 1).
a ,
Exereises 3.8
1. Describe al1 one-generated semigroups and al1 one-generated monoids.
"2. Prove that for finite n, End n is generated by the following set of three selfmaps
of n: first, the permutation (2-cycle) (0, l), which interchanges O and 1; next,
3.3 Semigroups 117
the cyclic permutation (O, 1, n - 1);and finally, the map f (x) = x v 1-i.e.,
f(0) = 1 and f(x) = x for x 2 l .
3. Prove that for 3 I n < co, End n is not generated by any two of its elements.
4. Prove that for infinite A, End A satisfies
(where f ,g2fg is of course an abbreviation for f o f O g O g O f O g).

5. Prove that End co does not satisfy
(Hint: Try h = the successor function on m.)

6. Find a distributive lattice (D, A , v ) for which the semilattices (D, A ) and
(D, v ) are not isomorphic.
7. Prove that if (B, A , v ,-, 0,1) is a Boolean algebra, then (B, A ) and (B, v )
are isomorphic semilattices.
8. Give an example of two distributive lattices D, = (D,, A,, v, ) and D, =
(D,, A,, v,), such that (D,, A , ) is a subsemilattice of (D,, A,) but
(DI, v,) is not a subsemilattice of (D,, v,). (Thus DI is not a sublattice
of D,. Hint: There is a solution with a very small finite number of elements.)
Prove that in this situation we will at least have x v, y 2 x v, y for al1
x, y E DI, and that equality holds whenever we happen to have x v, y E D,.
9. With the notation of the previous exercise, find distributive lattices D, and
D, such that D, is not a homomorphic image of DI, but (D,, A,) is a
homomorphic image of (DI, A, ).
10. Prove that every rectangular band A is isomorphic to a product B x C, with
B a left-zero semigroup and C a right-zero semigroup. (Hint: Define a relation
0, on A by saying a, 8, a, iff a, x = a,x for every x E A. Prove that 8, E Con A,
define a symmetrically related congruence O,, and try to establish directly an
isomorphism between A/8, x A/B, and A.)
11. Let a and p be distinct semigroup terms with no variable occurring twice in
either a or p. Show how to substitute either O, 1, or 2 for each of the variables
so that a and p evaluate differently in the three-element semigroup S, defined
on the previous page.
12. S o get a preview of some of the pathology that can occur in the monoid of
isomorphism classes under direct product, find four directly indecomposable
mono-unary algebras A, B, C, and D with A x B E C x D, such that A is
isomorphic neither to C nor to D. (An algebra A is directly indecomposable
iff [Al > 1 and A r E x F implies [El = 1 or IFJ= l. Hint: The pictorial
representation of mono-unary algebras that we developed in 53.2 should be
helpful here. The reader may also look ahead to 55.1, where we present this
exercise again with some more detailed suggestions. See, for example, Exer- '
cises 3 and 5 of 55.1. In that same exercise set, we also present some failures
of unique factorization that involve G-sets and commutative semigroups.)
13. Every unary algebra can be expanded to an F-set for an appropriate free
monoid F. More precisely, let ( A , Fi)ie, be any unary algebra, and define P
to be the free monoid on (xi: i~ I}. Show how to define an F-set ( A , f,),,,
such that fxi = Fifor al1 i E I .
3.4 GROUPS, QUASIGROUPS, AND LATIN SQUARES
Groups are frequently thought of as special kinds of semigroups. This familiar

conception of group theory differs from our main definition (introduced in
Chapter 1)in ways that are subtle but important. For this alternate presentation
of group theory, we define a multiplication group to be a semigroup (G, that m )
has unique solutions: When any two of the variables in the equation x .y = z are
given fixed values, there is exactly one value of the third variable that satisfies
the equation. One easy consequence of the definition comes if we take x and z
to be equal; we thus have e such that x . e = x. It is a routine exercise to check
that e does not depend on x, and that e x = x as well. Moreover, for each x there
is a unique element x-l such that x - x-' = x-l x = e.
In our work, the primary meaning of "group" is the one defined in Chapter
l. Just as we formed monoids from semigroups with unit by promoting the unit
element e to the status of a nullary operation, we may turn any multiplication
group into a group by defining e and x-l as above. Thus multiplication groups
are precisely the semigroups that can be formed by taking a group and deleting
the operations e and -l. Thus, as for monoids and semigroups with unit, the two
group notions are more or less interchangeable, especially if only one group is
under consideration at a time.
As before, however, distinctions arise when we consider two or more alge-
bras together. For instance, a subalgebra of a group is always a group, but the
same is not true for multiplication groups. Traditionally, writers on group theory
have blurred this distinction by using the expression "subgroup": ( H ; ) is
defined to be a subgroup of (G, - )iff the corresponding ( H , -,-',e) is a sub-
algebra (in our sense) of (G, -, e ) . Not wishing to burden our readers with
fussy terminology, we will often allow ourselves to refer to a "group" and let the
context provide the distinction between a group and a multiplication group. (Our
primary terminology is of course that of Chapter l.) "Subgroup," however, will
always take the meaning we described just above, and in keeping with this usage,
when we refer to the subgroup generated by a set X c G or write Sg(X), we always
mean the subuniverse sgG(x)that is generated by X in the algebra (G, -,-', e).
For congruence relations on groups, however, this distinction between group
theories makes no difference. As the reader will verify in Exercise 1 below, if
(G, -l, e) is a group, then Con(G, -,-l, e) = Con(G,
a , Instead of considering
e).
0 E Con(G, e),it is traditional in group theory to consider in its place the normal
subgroup
N, = (x E G: (e, x) E O } .
We leave to the reader the proof that this is indeed a normal subgroup of G.
I
(Recall that a subgroup N of G is called normal iff g n g-l E N for al1 n E N and
al1 g E G.) Conversely, if N is any normal subgroup of G, then, as the reader may
check, the relation
x8,y iff x - Y - ~ E N
defines a congruence relation 8, on G. These two correspondences are inverse to
each other @.e., 8 = 8, iff N = N,), and it is easily seen that 8 c 4 implies
N, S N$. Thus we have a one-to-one correspondence between congruentes on
(G,.) and normal subgroups of G, which is in fact an isomorphism between
Con<G, and the lattice of normal subgroups of G. We will frequently use this
a )
correspondence, implicitly and without further mention, in discussing con-

gruence relations on groups.
As with monoids, the best-known groups are free groups and groups of
mappings. Into the latter category fa11 most of the classical examples of group
theory and Lie group theory, such as the group of al1 rotations of Rn or al1 unitary
transformations of a Hilbert space. As the reader may easily check, for any set
X, if G is a submonoid of EndX that is a group, then G consists entirely of
permutations of X. For fixed X, the largest such group is the group of al1
permutations of X, or the symmetric group on X, and may be denoted either
Sym X or Aut X. Notice that (if we consider Sym X as a multiplication group)
we have the following extension of the inclusion relations of the previous section:
I1 Sym X G End X c Part X c Bin X.

In the next section, we will briefly consider Cayley's Theorem, which says that
every group is isomorphic to a subgroup of Sym X for some X.
The construction of the free group Fg(X) over a set X turns out to be more
complicated than that of free semigroups in the previous section, since a free
group is a quotient of the obvious "word construction. By this we mean that
Fg(X) is a quotient of FA(X U X-l), where X-l is a set formed by taking one
new element x-l for each X E X . Informally, we may say that two words in
FA(X U X-l) are equivalent iff they represent the same expression in any group,
and then we can say that Fg(X) is the quotient of FA(X U X-l) that results from
identifying equivalent words-that is, Fg(X) = FA(X U XV1)/8,where 8 is the
congruence relation of equivalence. Another way to describe 8 is that it consists
of al1 equations, such as x - x-l y Ñ y, that are identities of the theory of groups.
We will see in Chapter 4 that in fact every free algebra can be constructed as a
quotient of an algebra of "words" by a congruence relation consisting essentially
of identities. Although the 8 under consideration here probably seems clear
enough to those readers who have even a modest experience in group theory, we
will now present a self-contained definition of it.
Taking X, X-l, and FA(X U X-l) as above, we define R to be the set of
reduced group words on X. That is, R is the subset of FA(X U X-l) consisting of
words in which no x is adjacent to the corresponding x-l. By Theorem 3.6, there
exists a unique homomorphism ,u: FA(X U X-l) + End R such that ,u(x) and
,u(x-l) are the following two selfmaps of R:
if r does not begin with x-'

if r = x-'-S
r if r does not begin with x
ifr = x-s.
It is evident that y(x) o y(x-') = y(x-') o y(x) = 1, and so that y[FA(X U X-l)]
is a subgroup of Sym R. Now 9 is defined to be the kernel of y (i.e., {(w,, w,):
y(wl) = y(w2)}),and, as we said before, F9(X) is defined to be FA(X U X-')/O. By
the Homomorphism Theorem (1.16), F9(X) is isomorphic to the above men-
tioned subgroup of Sym R, so F9(X) is itself a group. (This procedure has defined
Fg(X) as a multiplication group. We leave it to the reader to state the proper
definition of (~10)-'for w E FA(X U X-' ).)
From now on throughout our discussion of the free group F9(X), we will
let F stand for the free monoid Fd(X U X-'). For each w E F, G will stand for a
unary operation; in fact ( A , W),,, will stand for an F-set. In particular, when
we write (R,@)w,F we mean the F-set that arises naturally on the R defined
above by taking W(r) to be [y(w)] (r).
LEMMA. I f i is any homomorphism from F to a group G such that i(x-') =

[i(x)]-' for each x E X , then ker i 2 9. Thus 9 is the smallest congruence on F
such that F/9 is a group and (e, x x-') E 9 for each x E X .
Proof. Given such a 1, we construct another F-set (G, W),,, by defining

@(g)= A(w) g for each g E G. Letting i stand also for the restriction of i to R,
we claim that
n: (R, W) + (G, W)
is a homomorphism of F-sets. For this we observe that if w' is formed from w E F
by deleting an adjacent pair of letters x x-l or x-l x, then A(wf)= i(w), since
g-'g = gg-' = e for any g E G. Letting w' now stand for a word formed from w
by any number of such deletions, we have, for arbitrary w E F and r E R,
which establishes that A is a homomorphism of F-sets.

Now to prove the lemma, let us suppose that (w,, w,) E 9 and prove that
A(wl) = i(w2).The hypothesis means that w1 = i?, in (R, 17); we calculate now
that
For each x E X , we will let X stand for the image of x in Fg(X), i.e., the class
x/0 in F/9. In the terminology we shall develop in Chapter 4, the next theorem
states that Fg(X) is the group that is freely generated by {x:x EX}.
3.4 Groups, Quasigroups, and Latin Squares 121
T H E O R E M 3.9. If G = (G, . ) is any group, and i f f : X + G is any function,

then there exists a unique homomorphism 4: F9(X) + G such that 4(%)= f ( x )for
al1 x é X .
Proof. The uniqueness of 4 is immediate from the fact that G is generated by

(%: X E X ) . For its existence, first use Theorem 3.6 to get a homomorphism
1:F + G such that A(x) = f ( x ) and L(x-') = [f ( x ) ] - l for each x E X . By the
lemma, 8 E ker A, and hence for each word w E F, we may define $(w/8) to be
A(w). It is easy to check that this 4 is a homomorphism such that $(Y) = f ( x )for
each x. m
Although it contained a few details special to group theory, our construction

of F9(X) mainly followed the general pattern that will be described in Chapter
4. Let us look briefly at another, more special, description of the free group Fg(X),
which relies on the following lemma. Recall that the collection R of reduced
words consists of al1 words in F = FA(X U X - l ) in which no x is adjacent to the
corresponding x-l.
LEMMA. Each word of F is equivalent to a unique word in R. I.e., for each w E F

there exists a unique r E R such that (r,w) E 8.
Proof. It is obvious that every w can be made equivalent to a reduced r by

deleting contiguous pairs x x-l and x-' x until no more such pairs remain. For
uniqueness, let us suppose that (r,, r,) E 8 fl R~ and prove that r1 = r,. Examining
the definitions of R and ,u(r), we see that if r is reduced, then [,u(r)](e) = r. Now
(r,, r,) E 8 implies by definition that ,u(rl) = ,u(r2),and so we have
Therefore, we may represent each equivalence class w/8 by the unique r that
is 8-equivalent to w. Using this representation, we obtain the following indepen-
dent description of the free group on X .
FREE G R O U P S . For any set X , the set R of reduced words in x and x-' (with
x ranging over X ) forms a free group over X i f multiplication is defined as follows.
T o multiply reduced words r, and r,, we write the word r1 . r, and then convert r1 r,
to a reduced word by removing pairs x x-l and x-' x until no such pair remains.
The inverse, in this group, of a word w is formed by writing w in the reverse order
and replacing each occurrence of each x in w by x-l, and vice-versa. m
The class Y of al1 groups is one of the varieties that, like lattices and Boolean
algebras, motivated the general study of varieties. One well-known subclass of
Y, which is perhaps less well known as a variety, is the class Nnof al1 n-nilpotent
groups. Its definition begins with the comrnutator operation
[ x , y] = x-l y-lxy,
which, of course, is well defined in every group G and is a constant operation on

G if and only if G is commutative. We also like to think of [m,.] as a group-
theoretic term (symbolic operation), which can be used as a constituent of
equations. Although technically required, parentheses are dispensable in group
terms (by associativity), and we will follow tradition in leaving them out of words
like x-l y-lxy. To recall the definition of 4, we first define an increasing
sequence of subgroups of an arbitrary group G:
zo = {e}
Zi+, = {x E G: (b'z E G) [x, z] E Zi).
It is clear that {e} c Z, c Z1 G , and it is in fact possible to have either
a - .
Zi < Zi+, for every i or Zi = Zi+, # G for some i. But if Z, = G, then we say that
G is n-nilpotent. The class of al1 n-nilpotent groups is denoted M,. Obviously,
Jlr, is the class of al1 commutative groups and thus is the variety defined by the
law xy Ñ yx (or, equivalently, [x, y] Ñ e). In order to build some equations that
will define 4, we first give the following recursive definition of some group-
theoretic terms for n 2 2:
THEOREM 3.10. M, is the class of all groups that obey the law
[x,, - a - , x,] Ñ e.
Proof. Reasoning inductively, we see that the following is true for al1 x, E G:
xo ~ ~ , * ( b 'E~G)Cxo,x,l
l €2,-1
@ ( b ' ~ E2 G ) ( b ' ~El G) CCxo,xll,x21~~,-2
@ ( b ' ~ ," ' x1 EG)[x,, ..., x,] €2,

*(b'x, xl EG)[x,, , x,] = e.
Therefore, we have Z, = G iff these equivalent conditions hold for every x, E G,
which is to say that the equation
[x,, , x,] =e
holds for al1 x,, , x, E G. m
Another familiar class of varieties of groups is that of the Burnside varieties

a , . For each n, a, is defined by the laws of group theory together with the law
x n Ñ e. Notice that for this variety the laws x x-l Ñ x-l x Ñ e are in a sense
redundant, as is the very operation of forming inverses, since it is clear that xn-l
is an inverse of x. At the end of 83.2, we mentioned a group (constructed by E.
Rips) that satisfies x n Ñ e for some n and is generated by a two-element subset
but is not finite.
A quasigroup is an algebra (Q, \, /) that obeys the following laws:
m,
The equations (1) imply that, given x and y, the equations y . z = x and z y = x
have solutions given by z = y\x and z = x/y, respectively. Equations (2) imply
that in each case, such z is unique. (If we had, for example, x = y z, = y z,, then
the first equation of (2) would yield z, = y\(y - z,) = y\(y z,) = z,.) In other
words, if (Q, ., \,/) is any quasigroup, then its reduct (Q,. ) has the unique
solution property, which we mentioned above relative to multiplication groups.
We will use the term multiplication quasigroup for any groupoid satisfying the
unique solution property. According to what we just said, the multiplication
quasigroups are just the reducts of quasigroups to type (2). Therefore, especially
in the finite case, a quasigroup can usefully be thought of as given by a multi-
plication table in which each element appears exactly once in each row and once
in each column, such as
An n x n array of elements selected from some n-element set S is called a Latin

square iff each element of S occurs once in each column and once in each row.
From this point of view, therefore, a multiplication quasigroup is nothing more
than a Latin square whose rows have been labeled by (in one-to-one correspon-
dence with) the elements of S, and likewise the columns.
A loop is an algebra (L, \, /, e) such that (L, \, /) is a quasigroup and e
e , m,
is a unit for (L, ). As before, one may define a multiplicationloop to be an algebra

(L, e) that can be expanded to a loop (L, - ,\, /, e). Obviously, every Latin
a ,
square can be made into a loop by selecting one element e to be the unit element,
and then labeling the rows and columns in such a way that e turns out to be a
unit element.
The conversion of a quasigroup (Q, \, /) into the corresponding multi-
m,
plication quasigroup (Q,.) is not so well behaved as is the corresponding

procedure for groups. There exist a quasigroup (Q, \, /) and an equivalence
m,
relation 0 on Q such that 0 is a congruence relation of (Q,.), but 8 is not a

congruence relation of (Q, -,\, 1). (See Exercise 3.12(9) below.)
For an interesting special sort of quasigroup, we turn to the variety of Steiner
quasigroups. These are algebras (S, - ) that obey the equations
The following lemma justifies the "quasigroup" terminology.
LEMMA. Every Steiner quasigroup is a multiplication quasigroup.

Proof. One easily verifies that if (Q, ) is a Steiner quasigroup, then (Q, m, e, )
satisfies laws (1) and (2), which define quasigroups. m
Steiner quasigroups are intimately connected with the notion of Steiner

triple system (STS). An STS on a set A is a family S of three-element subsets of
A such that each two-element subset of A lies in a unique member of S. (Some-
times the pair (A, S) is itself referred to as an STS.) The notion is named after
J. Steiner, who observed in 1853 that if A is finite and possesses an STS, then
1 Al E 1 or 3 (mod 6) (see Exercise 11 below); he asked whether, conversely, every
such finite A has an STS. In fact this question had already been raised and
answered (affirmatively) by T. Kirkman in 1847. (Steiner quasigroups were
invented by H. Bruck in 1963, more than 100 years later.)
Now if (A,S) is an STS, we define a Steiner quasigroup (A;,) as fol-
lows: if x = y, then x ., y = x, and otherwise x ., y is the unique z E A such that
(x, y, z) E S. Reciprocally, given a Steiner quasigroup (A, ), we define an STS
(A, S.) via
The following theorem is easily proved from the definitions involved.
THEOREM 3.1 l. The maps
form a one-to-one correspondence, and its inverse, between Steiner quasigroups and
-,
Steiner triple systems. Moreover (A,, ) is isomorphic to (A,, m,)if and only if
the corresponding (A,, S, ) is isomorphic to (A,, S,). ¤
In Exercise 12 below, we will indicate how quasigroups of another kind,

called Steiner loops, can be put into a one-to-one correspondence with STS7s.An
interesting corollary of Theorem 3.11 is that the cardinalities of finite Steiner
quasigroups are precisely the cardinalities of finite STS's. Thus it follows from
Theorem 3.11 (and our remarks about the cardinalities of finite Steiner systems)
that there is an n-element Steiner quasigroup if and only if n E 1 or 3 (mod 6).
That is, for Y the variety of Steiner quasigroups, the spectrum
SpecY = ([Al:A E Y ,A finite}
consists precisely of non-negative integers n such that n 1 or 3 (mod 6). (We
will return to a more general study of spectra in a later volume of this work.)
Exercises 3.12
1. Show that if (G, - , -',e) is a group, then Con(G, ) = Con(G, e).
S , -',
2. Describe another form of group theory that does not require the constant e.
(Its axioms are to be laws based on and -', and the groups of this type are
3.4 Groups, Quasigroups, and Latin Squares 125
to be interchangeable with groups and multiplication groups in the same

way that those two are interchangeable with each other. If we permitted
empty algebras, this notion would be intrinsically different from the group
notion defined above. A group (in the usual sense) must contain the element
e and hence cannot be empty, but one can conceive of an empty group in the
sense of this exercise. Nevertheless, we have explicitly excluded empty alge-
bras from consideration, and so this subtle difference does not affect our
work.)
*3. (G. Higman, B. H. Neumann, A. Tarski) Let V be the variety that consists
of al1 algebras (A, /) that obey the law
x/( C ((x/x)/Y)/~IIC((XIX)IX)/~~
) Ñ Y.
Prove that if (G, -l, e) is a group, and if we define x/y to be x . y-l, then
m ,
(G, /) E Y, and conversely, if (A, /) E V , then the following definitions yield

a group (A, -l, e):
e,
x Y = x/( (x/x)/Y>
e = X/X
x-l = (x/x)/x.
In the words of 54.12, this says that V is equivalent to the variety of groups.
In fact the "conversely" part is rather difficult, and so we supply the following
sequence of hints. First, define t (x, y, z)to be the /-term
C ((x/x>/Y)lzlC
l ((xlx)/x)/zl
Thus the defining identity is x/t(x, y, z) Ñ y. Now evaluate
xlt(x, x, t((xlx)lx,Y, 4 )
in two different ways to obtain
(a) xl(vlv) x.
The next milestone is to prove that u/u does not depend upon u, i.e.,
U/U Ñ 21/21.
To obtain this, substitute u/u for y in the axiom and then use (a) to eliminate
u from the left-hand side.
Letting e stand for the constant u/u, notice that (a) then reads x/e Ñ x.
Substituting e for x and z in the original axiom will lead to
el(ely) Ñ Y.
Upon defining
x Y xl(ely)
y-l = ely,
it is apparent from the above considerations that x y-' Ñ x/y, e-' Ñ e,
e x Ñ x - e Ñ x and x x-' Ñ x-l x Ñ e. Now the original axiom evidently
reduces to
x-' - [(yz) - (xz)-l]-l = y-l.

.Now three easy substitutions into (P) will yield the equations x .(x-lz) Ñ z,
(yz) z-1 Ñ y, and x-'y-' Ñ (yx)-l. At this point, the associative law for is
just around the corner, and the remainder of the proof is straightforward.
4. Prove that the congruence relation 8 appearing in the definition of free
groups is the same as the relation O', described as follows. First, define N to
be the smallest submonoid (i.e., subsemigroup containing O ) of F&(X U X-l)
s u c h t h a t i f y ~ ~ a n dt hx e~n~x ,- ' . ~ . X E Na n d x - y - x - l ~ N . T h e n d e f i n e
O' = {(wl,w,): w1 wyl E N).
(Hint: It is not difficult to prove O' c 8. For the reverse inclusion, consider
(w,, w,) E 8. Form a reduced word K, from w1 by successively deleting ap-
propriate parts of w,; likewise form % from w2. Use the lemma after 3.9 to
establish that w, = W,,and so W, q1 E N. Follow the direct reduction of wi
to Wi to establish ultimately that w, wyl E N.)
5. Prove that the sets Zi defined prior to Theorem 3.10 are in fact normal
subgroups of G.
6. Prove that the following two equations hold for x, y elements of any quasi-
group:
7. (G. Birkhoff) A quasigroup (G, e , \, /) satisfies the equation
iff it is possible to define a group (G, -l, e) such that x/y = x - y-' and
S ,
y\x = y-' x for al1 x, y E G.

"8. Suppose that G = (G;) is any groupoid that has (possibly non-unique)
solutions-that is, when any two of the variables in the equation x y = z
are given fixed values, there is at least one value of the third variable that
satisfies the equation. Prove that there exists a multiplication quasigroup Q
and a homomorphism from Q onto G. (Hint: The easiest case occurs when
G is countable. In this case, define the universe Q of Q to be G x co. Place
the elements of Q2 in a sequence and define the product on elements of this
sequence one after another, making sure that the equations x - y = z are al1
uniquely solvable and that (g, m) - (h,n) is always equal to (g h, p) for some p.
The generalization to higher cardinalities will be apparent to those readers
who are familiar with ordinal sequences.)
"9. Prove that there exist a multiplication quasigroup Q = (Q, and con- m )
gruence relations #, $ on Q such that # o $ # $ o #. Prove that such a

quasigroup must be infinite, and cannot be associative. Note that, since
quasigroups are congruence permutable (see Exercise 3 of §4.7), either 4 or
$ must be an example of a congruence on a multiplication quasigroup that
is not a congruence of the corresponding quasigroup. (Hint: By Exercise 8,
it is enough to find such nonpermuting congruences on a G that has (possibly
non-unique) solutions. Start with any countable groupoid G that has con-
gruences q!~and $ that do not permute. Now progressively enlarge G so as
to adjoin solutions of equations without disrupting $ and $ too much. A
way to perform one stage of this construction is to let f be a one-to-one
correspondence between G and a set H that is disjoint from G. Then extend
to G U H by taking g h = f -'(h), h. g = f -l(h) and h, h, = h,. The corre-
sponding enlargement of $ and $ proceeds by merely adjoining al1 pairs (h, h)
with h E H. Check that an iteration of this construction through m steps will
yield the desired algebra G.)
10. Let A = (3, with defined by the following table:
S ) ,
Prove that A is a Steiner quasigroup that occurs as a subalgebra of every

nontrivial Steiner quasigroup. (In fact, every two distinct elements of a
Steiner quasigroup generate a subalgebra isomorphic to A). Prove that A
also satisfies the equation
In a later volume, we will see that these equations (the Steiner quasigroup
laws together with this last equation) have at most one model in every
cardinality, and in particular, every finite model C of the identities is a power
A" for some n E m). There is also an elementary inductive proof of this result
for finite models, which is due to J. T. Baldwin and A. Lachlan. Like every
finite algebra, such a C has a (possibly empty) proper subuniverse B and an
element c such that A is generated by B U (c). Fix b, E B. It is not hard to
check, using the given identities, that each element of C is uniquely of the
form b, b c or (b c) b, for some b E B. Now, knowing that B r A"-', one
can use the given identities to check how the various elements b, b c and
(b c) b, multiply in order to deduce that C E A". The reader is asked to fill
-
in the details of this argument. (It helps to first establish that the equation
x(y(zw)) N w(y(zx)) holds for al1 x, y, z, w E C.)
11. Prove that if A is finite and possesses a Steiner triple system, then 1 Al 1 or
3 (mod 6). (Hint: First, fix a point a of A, and find a formula for 1 Al in terms
of the number of "lines" through a. Secondly, write an equation relating the
number of "lines" to the number of two-element subsets of A. Then compare
the results of these two calculations.)
12. The variety of Steiner loops is defined by the identities
I x - x N e, x . y Ñ y - x , x.(x.y) Ñ y.
1 Prove that every Steiner loop is a loop. Discover and prove a one-to-one
correspondence between STS's and Steiner loops, which goes much like
Theorem 3.11, with one notable deviation. If (A, S) is an STS, then the
associated Steiner loop has universe A U (e), where e is a new element that
is taken to be the unit of the Steiner loop. What is the spectrum of the variety
of Steiner loops?
3.5 REPRESENTATIONS IN End A AND Sym A
Very generally, a representation of a group G or a semigroup S means a homo-

morphism into some other group G' or semigroup S' (or a family of such
homomorphisms). Since the object is to increase our understanding of G or S,
we usually restrict G' and S' to belong to some class of (semi)groups that even
if not perfectly understood, is at least manageable. (For instance, G' is most often
taken as the automorphism group of some vector space; such representations are
called "linear.") In both group theory and semigroup theory, the investigation
of representations has become a highly developed and specialized subject in its
own right. Almost al1 of this material (including al1 of the linear theory) lies
outside the scope of this book. The following paragraphs are concerned with just
the basics of representation in the broader sense. In particular, we will concen-
trate on the case G' = Sym A and S' = End A for some A.
Our first theorein will display the close connection between associativity
and the existence of such representations. We first observe that for an arbitrary
groupoid (A,. ), there is a very natural map, called the (left) regular representa-
tion L: A + End A, which is defined as follows:
L(a) = La E End A,
where
THEOREM 3.13. A = (A,. ) is

i. a (multiplication) group iff L is un isomorphism from A onto a subgroup
of Sym A;
ii. a sernigroup iffL is a homomorphism,fromA into End A;
iii. a quasigroup iff L maps A one-to-one into Sym A and the range of L
partitions A x A (i.e., for each (b, c) E A x A, there is a unique a E A such that
La(b) = c).
Proof. The statements are nearly trivial; we shall deal only with (i). If L is an
isomorphism onto a subgroup of Sym A, then of course A is a group. Conversely,
if A is a group, then the unique solutions property easily implies that La is a
permutation for each a E A, and that La # Lb for a # b. (In fact La(c) # Lb(c)for
al1 c EA.)The following simple calculation shows that L is a homomorphism:
(La o Lb)(c)= La(Lb(c))= a . (b c)
= ( a . b ) - c = La.,(c). ¤
3.5 Representations in End A and Syrn A 129
Narrowing somewhat the general notion mentioned above, we will define a

representation of a semigroup S on a set T to be a homomorphism 4: S -+ End T.
The set T need not be named; a representation of S means a representation of S
on some set T. A representation 4: S -+ End T is called faithful iff 4 is one-to-one.
Recall from Theorem 3.13 that every semigroup has one representation called
the regular representation (denoted L above).
By a representation of a monoid M let us mean a homomorphism (M, e) -+ a ,
End T = (End T, o, l,), where 1, is the identity map of T. Not every represen-
tation of a semigroup (M, ) with unit is a monoid representation of (M, e); S ,
one can be sure only that e maps to some retraction of T onto one of its subsets
(see Exercise 2 below). Nevertheless, the regular representation is obviously a
monoid representation. When we refer to a representation of a group G, we will
always understand a representation of the monoid (G, e). (We need not make
a ,
express demands on the behavior of -', for if e maps to l,, then x and x-'
automatically map to mutually inverse permutations of T.) Thus every represen-
tation of a group in our sense is a representation by permutations of some set T.
The following theorem is named after Arthur Cayley, who first formulated
the notion of abstract group in 1854. He also invented the regular representation
(sometimes called the Cayley representation)and proved Theorem 3.13 and the
next theorern, both for the case of groups.
CAYLEY'S THEOREM. The regular representation of a semigroup with unit

is faithful; hence every group has a faithful regular representation. Every semigroup
has a faithful representation.
Proof. The regular representation of a semigroup S is faithful iff whenever x # y

in S, there is z E S with x z # y z. If S has a unit 1, one can always take z = 1.
Even if S has no unit, the following lemma yields a monoid S' -> S, and it is clear
that the regular representation of S', when restricted to S, is a faithful representa-
tion of S. m
LEMMA. Every semigroup is embeddable in a monoid.
Proof. See Exercise 1 below. ¤
It is advantageous to view the collection of al1 representations of one fixed

group G in terms of the variety of G-sets, introduced on page 108 of 53.2 above.
If R: G + End T is a representation, then clearly ( T , f,), , is a G-set if we define
f, by the equation f,(t) = [R(g)] (t) for g E G and t E T. Moreover, the same
equation obviously defines a representation R for each G-set, and in this way we
establish a one-to-one correspondence between G-sets and representations of G .
Thus, for example, we will say that two representations R: G -+ End T and
R': G -+ End T' are isomorphic iff the two associated G-sets are isomorphic.
Similarly, we may define a homomorphism from one representation to another,
the product of two or more representations, and so on. Thus, for instance, a
representation R: G + End T is transitive iff for each x, y E T there exist g E G such
that R(g) maps x to y. In this case, we may also say that G acts transitively on T
via R.
Translating Theorem 3.4 (which is about G-sets) into the language of repre-
sentations of G, we see that for every representation R: G -+ Sym T there is a
unique partition of T into sets (called orbits of R) on which G acts transitively.
Moreover if G acts transitively on T via R, then there exists a subgroup K of G
(unique up to an inner automorphism of G) such that R is isomorphic to the
representation M of G on the left coset space G/K by left multiplication (where
CM(s)l thK) =
According to Cayley's Theorem, the regular representation L embeds an
arbitrary group G as a subgroup L(G) of the full symmetric group Sym G. For
any X, the most natural subgroups of Sym X are those that express the symmetry
of some structure on X (such as the subgroup consisting of al1 rigid motions, in
case X happens to be a plane). A subgroup expressing symmetry is one that has
the form Aut (X, F,, F,, ) for some operations Fi on X. In our next theorem,
we provide operations Fi on G so that the image L(G) of the regular representa-
tion is indeed an automorphism group. This is our first theorem yielding an
isomorphism of an arbitrary group with Aut A for some algebra A. In a later
volume, we will return to this topic and prove a number of further results where
the algebra A is constructed in some more special classes of algebras.
In fact, we prove the next theorem both for monoids and for groups.
By analogy with our definition of the (left) regular representation, for each b E M
we define the right regular representation of b to be the map R,: M -,M de-
fined by R,(a) = a . b. Let us say, for #, $ E EndX, that # commutes with $ iff
# o $ = $04.
LEMMA. A map # E End M commutes with every R, if and only if # = La for

some a E M.
Proof. If # = La, then #(R,(x)) = La(Rb(x))= La(x b) = a - (x b) = (a x) b =

R,(a x) = Rb(La(x))= R,(#(x)), and so # commutes with every R,. Conversely,
if $ commutes with every R,, we will prove that # = La, where a = #(e). To see
this, we simply calculate #(b) = #(e. b) = #(R,(e)) = R,($(e)) = R,(a) = a . b =
La(b).Since these equations hold for every b, we obviously have $ La. ,=
THEOREM 3.14. (G. Birkhoff [1946].) Let M be a monoid and G a group. The
image L(M) of the regular representation in End M is the endomorphism monoid
of (M, R,),,,. The image L(G) of the regular representation in Sym G is the
automorphism group of (G, Rh),,,. Thus every monoid is isomorphic to the
endomorphism monoid of some unary algebra and every group is isomorphic to the
automorphism group of some unary algebra.
In fact, as we shall see in Exercise 3 below, the unary algebra of this last
theorem is itself a transitive G-set for a certain group G closely related to G.
The general subject of permutation groups (subgroups of Sym X) is a vast
one, about which we will say little more in this volume. We do ask, however, that
3.5 Representations in End A and Sym A 131
the reader be aware of the alternating subgroup A, of S, = Sym {O, - e., n - 11,
which consists of al1 permutations in S, that leave invariant the form
That is to say, a permutation a is in A, iff this term is equal (modulo the axioms
of ring theory) to
As is well known, no single permutation that is a two-cycle (i.e., its only non-
trivial orbit is a two-element set) is in A,, but every product of an even number
of two-cycles is in A,. Conversely, every element of A, is equal to a product of
an even number of two-cycles. (Recall from $3.2 that a two-cycle is a permutation
that interchanges two points and leaves al1 other points fixed. Recall also that
we multiply two permutations by composing them as functions: a p is defined so
that ap(x) = a(p(x).)
Since simple algebras form a major point of interest in this work, we ask the
reader, in Exercises 6-10 below, to review the proof that, for n 2 5, A, is a simple
group. From the usual viewpoint of group theory, this statement means that A,
has no normal subgroup other than (e} and A,. According to the translation
between normal subgroups and congruence relations (pages 118- 119), simplicity
means for us that A, has no congruence relations besides A, x A, and the
equality relation. This latter form turns out to be the appropriate notion of
simplicity for algebras in general, as we will see in Chapter 4.
Exercises 3.15
l. Prove that every semigroup is embeddable in a monoid. (Hint: The only
element that needs to be added is a unit element.)
2. If G is a group, and if 4: (G, e) (End T,o) is a representation of the
-+
semigroup (G,.) on T, then there exists a unique retraction 0 of T onto

a subset S of T, and a unique (monoid) representation S/: (G,.,e) +
(End S, o, 1,) such that 4(g) = S/(g)o 8 for al1 g E G. Conversely, for every
such 8 and S, these formulas define a representation of (G, ) on T. (By a
retraction of T onto S, we mean 8: T -+ T such that 8 o 0 = 8 and 0(T) = S.)
3. If G = (G, is a group (or even any groupoid), define the opposite group
S )
(groupoid) to be G = (G, *), where a * b is b - a for al1 a, b E G. Recalling the

right regular representation Rh, which was defined before the lemma to
Theorem 3.14, prove that R;, = Lh for al1 h E G. Deduce that the multi-unary
algebra (G, Rh) of Theorem 3.14 is a transitive G-set; in fact it is the quo-
tient G-set G / K (notation of $3.2) where K = {e). (We remark that the
asymmetry seen here-l is a homomorphism, but R is only an anti-
homomorphism (homomorphism from G)-is only apparent and has no
intrinsic meaning; it stems from our arbitrary decision to write f (x) rather
than (x)f for the result of applying a function f to an element x, which has
the consequence that we use f o g rather than g o f to denote the composite
function that first does g and then does f.)
4. For every transitive representation $ of a group G, there exists a subgroup

H of G such that $ is isomorphic to $H. (Here $ H represents G on the set T
of left cosets x H by the formula [$H(g)] (x - H) = (g . x) H. (Hint: Trans-
late a result from 53.2) H is unique up to an inner automorphism of G;
moreover, SHis a faithful representation iff
s-l-~ . = g{e}.
gsG
5. Define a subgroup K of Sym T to be transitive iff for al1 S, t E T there exists

a E K such that a(s) = t, and to be regular iff k(t) # t for al1 t E T and al1 k E K
with k # e. Prove that a representation $ of G on T is isomorphic to the
regular representation iff $ is faithful and $(K) is a regular and transitive
subgroup of Sym T.
6. If n 2 5, then every normal subgroup N of the alternating group A, (other
than {e)) is transitive. (Hint: Clearly, we can get a(i) = j # i for some a E N
and some i < n. By carefully selecting P E A,, we can get P-' 0 a 0 P(i) = k for
any k.)
In $3.2 we defined the stabilizer subgroup of an element t in a K-set
(T,f,),,,. We will now recall this definition for the special case that K is a
subgroup of Sym T and fa = a for al1 a E K. Modifying the notation to suit the
case at hand, we say that Stab,(t), the stabilizer of t in K, is the subgroup of K
consisting of al1 a~ K such that a(t) = t. Notice that if we disregard the point
n - 1, which does not move, then StabAn(n- 1) may be thought of as A,-,; we
will take this point of view in Exercise 8. The next exercise repeats some ideas
of $3.2 in the present context.
7. 1 I<1= IStab,(t)l 1 K-orbit of tl.
8. If n and N are as in Exercise 6, then Stab,(n - 1) is a nontrivial normal
subgroup of the alternating group A,-, .
9. The only proper normal subgroup of A, is the four-element subgroup
{e,(O1)(23),(02)(13),(03)(12)).
10. The alternating group A, is simple for n 2 5. (Hint: If N is a nontrivial
normal subgroup of A,, then induction and Exercise 8 yield Stab,(n - 1) =
A,_, (if n is 5, then a special argument is required, which involves Exercise
9). Now Exercises 6 and 7 allow us to evaluate 1 N 1.)
Let us call a group G directly indecomposable iff 1 GI > 1 and it is impossible
to write G E H x K with IHI > 1 and IKI > 1. Also, let us call G Zdivisible
iff for each a E G there exists b E G such that b2 = a.
11. If A is a unary algebra with Aut A 2-divisible, then no two distinct con-
nected components of A are isomorphic to each other. If AutA is also
directly indecomposable, then al1 but one of the connected components of
A have trivial automorphism groups.
*12. (G. Fuhrken [1973]) The group (Q, + ) of rational numbers is not isomor-
phic to Aut (A, f ) for any mono-unary algebra ( A , f ). {Hint: By the last
exercise, we may assume that (A, f ) is connected. Turning to the descrip-
3.6 Categories 133
tions in 83.2, first determine the action of a rational number q on the core
of (A, f ). Then look at the action of q on whatever trees may be attached
to the core. Make a second use of 2-divisibility.)
3.6 CATEGORIES
A category is a special kind of partial semigroup. More precisely, a category C

consists of a class C together with an 0peratio.n such that x -y is defined for
some but not necessarily al1 pairs (x, y) E C2, and such that the following axioms
hold for In Axioms (iv) and (v), a unit element of C means an element e of C
a .
such that e . x = x whenever e x is defined and x e = x whenever x - e is defined.
ii. If y - z and x (y z) are defined, then so are x y and (x . y) z, and x (y z) =

(x . y) ' Z.
iii. If x . y and y z are defined, then so is x - (y z).
iv. For each x there exists a unit e such that x e is defined.
v. For each x there exists a unit f such that f - x is defined.
It is easy to prove (see Exercise 3.16(1) below) that the unit elements e and f of
(iv) and (v) are unique; e is called the domain of x, and f is called its codomain.
For historical and motivational reasons (see the discussion of homomorphism
categories below), the elements of a category C are sometimes called morphisms,
and unit elements are called objects of C. If A and B are objects of C, then
hom(A, B) (or homc(A, B)) stands for the collection of al1 X E C such that the
domain of x is A and the codomain of x is B. Such an x is sometimes called a
morphism from A to B. We sometimes write x: A -+ B or A$ B to indicate that
x is a morphism from A to B. We will use the terminology of morphisms and
objects whenever it seems useful or appropriate. There exists an alternative
axiomatization of category theory in terms of objects and morphisms (see
Exercise 3. P6(2) below).
For categories C and D, a functor from C to D is a function f from C to D
such that if x . y is defined in C then f(x). f(y) is defined in D and f(x). f(y) =
f ( x Sy), and such that f(e) is a unit of D for each unit e of C. This notion is a
natural generalization of the notion of homomorphism to this class of "algebras"
whose operations are not everywhere defined. We may write F: C -+ D to denote
the fact that F is a functor from C to D. If C c D and the inclusion map
(restriction to C of the identity map on D) is a functor from C to D, then we say
that C is a subcategory of D. We say that C is a full subcategory of D iff C is a
subcategory that satisfies:
Vd E D [(domain(d) E C) & (codomain(d) E C) -+ d E C].
Thus, a full subcategory is completely determined by a subclass X of the objects

of D, and we may unambiguously refer to the full subcategory determined by X .
Categories C and D are isomorphic, written C r D, iff there exists a one-to-
one functor F mapping C onto D. We also write F : C 5 D and say that F is an
isomorphism between C and D. It is not hard to check that a functor F : C 4 D
is an isomorphism iff there exists a functor G: D -+ C such that F o G is the
identity map on D and G 0 F is the identity map on C. A faithful functor from C
to D is a functor that is one-to-one on each subclass homc(A, B). In other words,
a functor F : C D is faithful iff it satisfies:
-+
Vc, d E C[dom(c) = dom(d) & codom(c) = codom(d) & f(c) = f (d) -+ c = d].
It is not hard to see that if (M, -,e) is a monoid, then (M, - ) is a category
in which al1 products are defined and e is the only unit. Conversely, if C is a
category such that either C has a unique unit e or the multiplication operation
is defined over al1 of C2, then (C, - ,e) is a monoid. Thus if X consists of a single
unit element of a category C, then the full subcategory of C determined by X is
a monoid. If C and D are monoids, then a functor F: C -,D is the same thing
as a homomorphism.
Our interest in category theory stems from the fact that (just as for monoids)
its most natural models are collections of functions, such as the following, which
is known as the category of sets. Let SETS be the class of al1 (B,f, A) where A
and B are arbitrary sets and f is a function from A to B. Then SETS = (SETS, )
is a category if we declare (B,f, D) - (E, g, A) to be defined only for D = E, and
equal in this case to (B,f og, A). In this category, the unit elements (i.e., the
objects) have the form (A, l,, A) for A an arbitrary set and 1, the identity function
on A; the domain of (B,f, A) is (A, l,, A); and hom(A, B) is the collection of al1
(B,f, A) for f ranging over al1 functions A -+ B-i.e., hom(A, B) is in one-to-one
correspondence with B,. There is usually so little danger of confusion that we
may dispense with the official formula (A, l,, A) and allow ourselves to speak of
a set A as being an object in SETS. The full subcategory of SETS determined by
a single object A is naturally isomorphic to the monoid End A via the mapping
(A, f,A) -f.
In the category SETS, as well as in many of its important subcategories,
there is no natural bound on the cardinality of the sets A associated with the
objects (A, l,, A). Therefore, it often turns out that (the universe of) a category
of interest is a proper class. A small category is a category whose universe is a set
rather than a proper class. A straightforward extension of the technique used to
prove Cayley's Theorem of the previous section will establish that every small
category is isomorphic to a subcategory of SETS.
The mappings in which we are most interested are, of course, homo-
morphisms, and these form a category by a mild extension of the above construc-
tion. We use the same definition as in SETS to define a partial operation on the
collection of al1 triples (B,f, A) where B and A are algebras of a fixed similarity
type p, and f is a homomorphism from A to B. In this way we build a category
Alg,, whose objects are the triples (A, l,, A). The full subcategory of Alg, deter-
mined by a single object (A, lA,A) is isomorphic to End A. Every variety of
3.6 Categories 135
algebras of type p can be viewed as a class of objects in Alg, and as such

determines a full subcategory CAT Y of Alg,. The universe functor U: Alg, -+
SETS takes (B,f, A) to (By f, A)-that is, an algebra goes to its universe, and a
homomorphism goes to its underlying set-map. The universe functor is easily
seen to be faithful, and thus the pair (Alg,, U) is an example of a concrete
category-i.e., a pair (C, W) with C a category and W a faithful functor from C
to SETS. Many of the categories of general importance in mathematics are
concrete categories (with an obvious W), such as the category of continuous
functions between topological spaces, monotone maps between ordered sets, and
so on. There also exist important categories for which there is no such W, such
as the category of homotopy classes of continuous maps between topological
spaces.
Many important aspects of set theory can be described within the category
of sets, in terms of pure category theory. When mathematics is translated in this
way into category theory, the elements of sets are ignored, and even the sets
themselves play a minor role; the basic notions are those of functions and their
composition. In principle, set theory can indeed be translated into this language
with very little loss of detail, although many basic notions become considerably
more complicated to define. This book is not the place to pursue this path, but
we do wish to mention a few cases where the category-theoretic version of a
set-theoretic concept turns out to be simple to state and understand, and some
other cases where the abstraction of a set-theoretic idea in categorical terms
yields a new notion of significant interest. We will put some of these ideas to use
in $5.7, when we prove the Lovász Isotopy Lemma (5.24). That proof refers to
the notions of product and monomorphism in a special category of relational
structures.
A morphism f of a category C (i.e., an element of C) is called a mono-
morphism iff it is left cancellable-if f .g and f . h are defined and equal, then
g = h. A morphism f of C is called an epimorphism iff it is right cancellable. (In
SETS, for example, epimorphisms and monomorphisms are the onto and one-to-
one functions, respectively.) An isomorphism is a morphism that has a two-sided
inverse-i.e., an element f E hom(A, B) such that f g = B and g f = A for some
g E hom(B, A). Objects A and B of C are isomorphic, written A r B, iff there exists
an isomorphism in homc(A, B). An object D is called a product of objects A, and
A, in C, with associated projection morphisms no and n,, provided that n i €
hom(D, A,) (i = 0, l), and for every object-E and every pair of morphisms LE
hom(E, A,) (i = 0,1), there is a unique h E hom(E, D) such that ni h = (i = 0,l).
The product D is unique up to isomorphism (see Exercise 3.16(13) below), but
only up to isomorphism, since D may clearly be replaced by any isomorphic copy
of D. For any index set I and any family (A,: i~ 1 ) of objects of C, there is an
analogous definition of the product of the objects Ai. The reader may easily check
that in the category Alg, definedjust above, or even in one of its full subcategories
CAT Y arising from a variety Y, the usual product A x B is a product in the
sense of this paragraph.
Categories C and D are called equivalent iff there exists a functor F: C -+D
(called an equivalente) such that F : homc(A, B) -+ homD(F(A),F(B))is a bijection
for every pair of objects A, B E C , and such that for every object E of D there
exists an object A of C with F(A) r E. In fact, equivalences between categories
seem to occur naturally more often than isomorphisms. For instance, the double-
dual operation defines an equivalence from the category of finite-dimensional
vector spaces over a field K to itself. When we study duality in later volumes, we
will see a number of category equivalences.
We saw above that in Alg,, the full subcategory determined by a single
object (A, lA,A) is essentially the same as the endomorphism monoid End A.
Therefore, for A an object of an arbitrary category C, we will use the notation
Endc A to stand for the monoid that is the full subcategory of C determined by
A. (The universe of this monoid is of course homc(A, A).) We can likewise extend
the notion of automorphism group to such an object A by defining Autc A to be
the (multiplication) group of al1 invertible elements in the monoid Endc A. This
monoid Endc A has a natural extension to a larger category called the clone of A,
which has infinitely many objects and hence cannot be described simply as a
group or monoid. Clonec A (or simply Clone A when the context permits) is
defined to consist of the full subcategory of C determined by A and its finite
powers (and thus the definition requires al1 finite powers of A to exist in C). More
formally, Clone A is a full subcategory of C with distinct distinguished objects A,
and distinguished morphisms 1L,'E hom(A,, A) (i 2 1,O I j < i) such that each A,
is a product of i copies of A via the morphisms ni, - , ni-, and such that there
are no objects besides the objects A,. (We remark that Clone A is determined up
to isomorphism by this definition and does not really depend on the choice of
the objects A, and the morphisms n,'.)
An (abstract) clone (in the sense of category theory) is defined to be a category
C with distinguished objects A, and morphisms írj (indexed as above) such that,
with these objects and morphisms C = Clonec A,. A subclone of a clone C is a
subcategory D of C such that D contains al1 the objects A, and morphisms ni',
and such that each A, is the product in D of i copies of A, via the morphisms
ni, - ni-, . For one natural example, consider an arbitrary set A as an object
m ,
of SETS. Then ClonesETsA is most readily constructed by taking each A, to be

the usual set-theoretic product A'-i.e., A x A x x A (i factors). Then for
each n 2 1, homsETs(An,A) consists of al1 n-ary operations on the set A. One can
easily check that if C is any subclone of ClonesETsA, then
is a family of functions that is closed under composition and contains the

coordinate projection functions. Conversely, for every such family F of opera-
tions on a set A, there exists a subclone C of ClonesETsA such that the above
equation holds. Such families of operations are known as concrete clones, or
clones of operations, or if the context permits, simply as clones.
Starting at the beginning of the next chapter, we will consider this concrete
notion of clone in detail, reserving our study of abstract clones for Volume 2 (but
see Exercises 16- 18 and 21 -23 of 3.16 below). In Volume 2 we will say something
more about the history of and relation between these notions. Suffice it for the
moment to say that the word "clone" was coined in the 1940s by P. Hall, and
3.6 Categories 137
that the category-theoretic definition was invented in 1963 by F. W. Lawvere,

who used the term "algebraic theory" for what we are calling a clone. As we will
see in Volume 2, there is an alternative axiomatization of a clone as an algebra
whose universe is a disjoint union A = A , U A , U and whose m-ary operations
(for various m) are defined, not on al1 of A", but on certain subsets of the form
Ail x Ai2 x x Ai, (A typical model of these axioms is the set F of operations
mentioned just above, which has an obvious natural division into sets A,; for
each non-negative k and n, this model has an operation corresponding to
the construction of an n-ary operation from one k-ary operation and k n-ary
operations, as described below at the beginning of $4.1.) The two definitions are
equivalent in a sense that we will make precise in Volume 2. We treat them both
on an equal footing, since they are easily interchangeable and since each has its
own conceptual advantages.
Our working foundation for algebra is (an applied form of) one of the
standard set theories; nevertheless we find category theory useful as an alternative
point of view. Let us close this chapter with three illustrations of how category
theory provides a valuable viewpoint for general algebra. Firstly, notice that the
monoid of isomorphism types (under direct product), which we mentioned near
the end of $3.3, can be constructed directly from the category Alg, of al1 algebras
of the appropriate type. Secondly, Clone A and its subcategories End A, Aut A,
and so on, are al1 definable up to isomorphism as subcategories of Alg,. (For
Clone A, the direct powers Ai are not definable within category theory except up
to isomorphism, but nevertheless any choice of direct powers in the sense of
category theory yields the same CloneA up to isomorphism.) Finally, we will
devote a good bit of space in later volumes to a study of representation problems
for groups and monoids: given a group (or monoid), is it isomorphic to Aut A
(or End A) for some A in a given class X of algebras. (For instance, in Theorem
3.14 we solved this for al1 groups and al1 monoids with X the class of multi-unary
algebras.) Of course, these problems have the following more general form: Given
a monoid M and a category C, find an object A of C such that Endc A r M.
From this perspective, it is apparent that representation problems really
apply to al1 categories and need not be limited to monoids and groups. That is,
given a small category C and another category D, we may ask whether C embeds
as a full subcategory of D. The book of A. Pultr and V. Trnková [1980] is devoted
to this and closely related topics. For instance, as a generalization of Cayley's
Representation Theorem, along the lines of our Theorem 3.14, they prove that
if D = Alg, for p a unary type with at least 1 + 1 CI operations, then C does embed
as a full subcategory of D. In fact, there exist categories D into which every small
C can be fully embedded regardless of 1 CI, such as the category of al1 rings with
unit (i.e., CAT 2 for 2 the variety of rings with unit). See A. Pultr and V. Trnková
[1980] for details.
Exercises 3.16
l. The domain of x is unique. That is, if C is a category, x, e, f E C, e
and fare units, and e. x and f - x are both defined, then e = f.
2. An alternative axiomatization of category theory. Consider a structure

(X, Hom, e, m) consisting of the following things:
(1) A class X, called the class of objects.
(2) For each A, B E X, a class Hom(A, B).
(3) For each A E X, an element e, E Hom(A, A).
(4) For each A, B, C E X, a function
m,,: Hom(B, C) x Hom(A, B) -,Hom(A, C).
We now define a CATEGORY to be a structure of this type that obeys the
following axioms:
(a) rf ( A ,B) # (C, D), then Hom(A, B) f l Hom(C, D) = D.

(p) If z E Hom(A, B), y E Hom(B, C) and x E Hom(C, D), then
The exercise here is to prove that the theory of categories is equivalent to

the theory of CATEGORIES in the following sense. There are mutually
inverse definitions (to supply these fairly obvious definitions is part of the
exercise) of a CATEGORY in terms of a category, and of a category in terms
of a CATEGORY. (The definition here of "CATEGORY" is more fre-
quently encountered than is the definition of "category" that we have
adopted in this chapter. Usually one finds one extra postulate, namely that
Hom(A, B) should always be a set. This postulate in fact holds for every
category that we plan to discuss in this and future volumes.)
3. Prove that the notion of concrete category as we defined it in this section
is equivalent to the following alternative notion. A CONCRETE CATE-
GORY consists of a class P (whose elements are called objects), together
with a function U from P to the class of al1 sets, and a function HOM from
P x P to the class of al1 sets, satisfying the following axioms.
(a) H OM (x, y) E U (y)'(").
(p) The identity function 1,,(, E HOM(x, x).
(y) If f €HOM(x,y) and g ~ H o M ( y , z )then

, g o f€HOM(x,z).
That is, there exist mutually inverse definitions of a CONCRETE CATE-
GORY in terms of a concrete category, and of a concrete category in terms
of a CONCRETE CATEGORY.
4. Let 5 be an order relation on a set X (i.e., a reflexive, transitive, and
anti-symmetric subset of x2). Then 5 forms a small category if we take
(a, b) (c, d) to be (a, d) if b = c, and undefined otherwise. Conversely, if C is
a small category in which there is at most one morphism between any two
objects, and if no two distinct objects of C are isomorphic, then C derives
in this way from an order relation on its set of objects.
5. In the previous exercise, the order is that of a v -semilattice (i.e., least upper
bounds exist for S) if and only if, in the corresponding category, every two
objects have a product.
6. Every small category is isomorphic to a subcategory of SETS.
7. Prove that if f is an isomorphism in any category, then f is both a mono-
morphism and an epimorphism. (The converse is false; see Exercise 9 below.)
8. A morphism (B,f, A) of SETS is a monomorphism iff f is one-to-one, and
an epimorphism iff f maps A onto B. For a morphism (B,f, A) in Alg, (or
in one of its full subcategories CAT Y whose objects are determined by a
variety Y ) ,we have that (B,f,A) is a monomorphism iff f is one-to-one,
and if f maps A onto B, then (B, f,A) is an epimorphism.
9. Let DIST be C A T 9 for 9 the variety of al1 distributive lattices (i.e., the
category whose elements consist of al1 homomorphisms between two dis-
. tributive lattices). Prove that there exist a distributive lattice D and a proper
sublattice Do of D such that the embedding a : Do -+D is an epimorphism
in DIST. This a is therefore a morphism that is a monomorphism and an
epimorphism but not an isomorphism.
10. Although we are generally using a different meaning in this work, a groupoid
is often taken to be a category in which every morphism is an isomorphism.
A connected groupoid is a groupoid in which every two objects are iso-
morphic. One natural example is the path groupoid P(A) of a topological
space A. The elements of P(A) are homotopy classes [ y ] of continuous maps
y : [O, 1 1 + A with respect to homotopies that fix the endpoints. (I.e., [ y ]
-
denotes a --equivalence class, where y y' iff there exists a continuous
F: [O, 11 + A such that F(x, O) = y (x), F (x, 1) = y' (x), F(0, t) = y (O), and
F(l, t) = y ( l ) . ) If [ y ] and [6] are such homotopy classes, then [ y ] [ a ] is
defined iff y(1) = 6(0), in which case [ y ] [ 6 ] is defined to be the homotopy
class of y - 6 where
Prove that this is a groupoid, and a connected groupoid iff A is a path-

connected topological space. What are the objects of this category? If a is
an object in this category, what is End a? (It is something very familiar.)
11. A group can be thought of as a groupoid (previous exercise) that has only
one object.
12. Prove that the notion of product of objects A, actually makes sense for i
ranging over any index set I. What happens to your definition if I is the
empty set?
13. Prove the assertion in the text that the product (in the sense of category
theory) is uniquely defined up to isomorphism. More precisely, let A, (i E I)
be some family of objects of a category C. Suppose that D is a product of
the objects A, with projection morphisms ni, and that E is a product of the
objects A, with projection morphisms y,. Prove that there exists an iso-
morphism iZ E hom(D, E) such that ni = y, - A for each i. Conversely, if E is
isomorphism in hom(D, E), then the formula ni = vi ;1 defines projection

morphisms that make D into a product of the objects A,.
14. Prove the assertion in the text that the isomorphism type of Clonec A does
not depend on the choice of a full subcategory containing al1 finite products
of A. More precisely, let A be an object of G , and let D and E be full
subcategories of C with the following properties. The objects of D are
precisely A,, A,, . and for each i 2 1 there exist morphisms rr; (O 5 j < i)
m ,
from A, to A such that A, is a product of i copies of A via the projection

morphisms rr6, - nf-l. Similarly, the objects of E are B,, and each B, is a
e ,
product of i copies of A via the morphisms rl,! Then there exists a category
isomorphism A from D to E such that A(4)
= for al1 i and j.
15. Prove the assertion in the text that for every clone of operations F on a set
A, there exists a subclone C of ClonesETsA such that
F= U homc(An,A).
n 21
*16. Let C be the category of ordered sets and monotone functions. (Its elements
are triples (B,f, A), where A and B are ordered sets and f : A + B is a
function such that a 5 a' implies f(a) 5 f(af).) Let A be the object corre-
sponding to the ordered set (0,l) with the usual ordering. Describe Clone A.
*17. Let GROUPS be CAT $ for $ the variety of groups (i.e., the category whose
morphisms consist of al1 homomorphisms between two groups). Describe
CloneGRoupsA for A an object that corresponds to an Abelian group and
for A an object that corresponds to a finite simple group.
18. If D = Clonec A exists, then CloneDA exists and CloneDA = D.
As in our definition of the opposite of a groupoid in Exercise 3.15(3), the
opposite of a category (C, ) is the category (C, *), where x * y is defined iff y x
is defined, in which case its value is defined to be y x. The opposite of C may be
denoted CoP.
19. If A is a product of objects A, ( i 1)~ in COP,what does this mean in
terms of the category C? (I.e., what morphisms have to exist in C, and what
properties must they have?) In this situation, A is said to be a coproduct of
the objects A, in C.
20. In GROUPS (defined in Exercise 17 above), the free group on n generators,
F9(x1, - .- ,xn), is the coproduct of n groups, each isomorphic to the free
group on one generator. (It is most convenient to represent these n groups
as F9(x1), Fg(x2),and so on. We will be using this notation in Exercise 22
below.)
21. Let G be the opposite of GROUPS (defined in Exercise 17 above), and let
FG(1) be the object of G corresponding to the free group on one generator.
Prove that CloneGFG(1) exists, and describe it.
If C = (C, - ,A,, nl) and D = (D, B,, o,!) are clones, define a clone-map
m ,
from C to D to be a functor F: C -+ D such that &'(A,) = Bi and ~ ( n , ' =

) o,! for
al1 i and j.
3.6 Categories 141
22. Take CloneGFG(1) to be as in Exercise 21 above, and let p be the member

of this clone that is the group homomorphism p: FG(x,) -+FG(x,,x,),
which is determined by the stipulation that p(x,) = x,.x,. Let X be
any set, regarded as an object in SETS. Show that every clone-map
F: CloneGFG(1) + ClonesETsX determines a (multiplication) group struc-
ture on the set X by defining x y to be the binary operation that is the image
of p under F. Give the details of how the corresponding x-l can be similarly
defined. Prove that, conversely, if (X, ., -') is any group with universe X,
-'
then . and arise in this way from a clone-map F.
23. Let TOP be the category of topological spaces and continuous functions.
For A an object of TOP @.e.,for A a topological space), prove that there
exists a clone map CloneGFG(1) + CloneTopA iff A is the underlying space
of some topological group. (This means that there exist continuous func-
tions A, -+A and -l: A + A, and an element e E A such that (A,
a: S, -',
e) is
a group.) For every variety -Ir there exists a clone constructed analogously
to the clone CloneGFG(1) of these last three exercises. It is known as the
clone of Y ,and results analogous to those stated here for groups hold for
C H A P T E R F O U R
Fundamental
Algebraic Results
This chapter is a complete, in-depth presentation of basic universal algebra

including concepts, results, perspectives, and intuitions shared by workers in the
field. Much of the material implicitly underlies such specialized and highly
developed branches of algebra as group theory, ring theory, algebraic geometry,
and number theory. It is the starting point for research in universal algebra.
4.1 AEGEBRAS AND CLONES
An algebra is a set endowed with operations. The concept has been formally
defined already (see the first definition in Chapter 1).According to our definition,
an algebra is an ordered pair A = (A, F) in which A is a nonempty set and
F = (F,: i ~ l is ) a system of finitary operations on the set A. Algebras will
frequently be denoted in the form A = (A, Fi(iE 1)), or A = (A, QA(QE 1)), or
more simply as A = (A, - ), B = (B, . ), and so on. The listed operations are
called basic operations of the algebra, to distinguish them from other, derived,
operations to be discussed in this section. The set of elements of an algebra is
called its universe, or sometimes its base set. The cardinality of an algebra is the
same as that of its base set, so a finite algebra is one with a finite universe. (It
may have an infinite set of basic operations.) An algebra is said to have finite
type if its index set is finite. (In the examples above, the set I is the index set of A.)
By composition of operations is meant the construction of an n-ary operation
h from k given n-ary operations f,, - - - ,f,-, and a k-ary operation g, by the rule
(Al1 of these must be operations on the same set A. The non-negative integers k
and n can be arbitrary.) We shall write h = g(fo, .,A-,) for the above defined
operation h. The projection operations on a set A are the trivial operations pf
(with O 5 i < n) satisfying pf(xO,- x ~ - =
S ~ x,.
, )
4.1 Algebras and Clones 143
DEFINITION 4.1. A clone on a nonvoid set A is a set of operations on A that

contains the projection operations and is closed under al1 compositions. The
clone of al1 operations on A will be denoted by Clo A, while the set of al1 n-ary
operations on A (n = 0,1,2, .) will be written as Clo, A.
Clones furnish important examples of what is called a partial algebra-that

is, a set endowed with partial (or partially defined) operations. The composition
of a binary operation with two ternary operations, for example, is a 3-ary partial
operation on the set Clo A, defined not for al1 ordered triples of elements but
only for those triples (f,, f,, f2) where f, belongs to Clo, A, and f, and f, belong
to Clo, A.
Given any algebra, we can construct, from its basic operations, many more
by forming compositions. These derived operations constitute a clone. They
are frequently more useful than the given basic operations for purposes of
understanding the qualities of an algebra and how it is put together.
DEFINITION 4.2. The clone of term operations of an algebra A, denoted by

Clo A, is the smallest clone on the base set A that contains the basic operations
of A. The set of n-ary operations in Clo A is denoted by Clo, A.
The members of Clo A are called term operations of the algebra A..This usage
is borrowed from logic. The term operations of an algebra are the same as the
operations determined by terms built from variables and symbols denoting the
+
basic operations of the algebra. For example, x2 xy and x(x + y) are terms
that determine binary operations in any ring; in fact, in each ring they determine
the same binary term operation. In $4.11, we shall deal systematically with terms
and their connection with term operations.
Our definition of the clone of an algebra is worthless from a computational
standpoint. For example, the set of binary term operations of a finite algebra is
a finite set, and we ought to be able to determine that set, given the algebra. But
in order to apply the definition directly, we would have to construct the (infinite)
set of all term operations before we could be sure which binary operations are
term operations. The next theorem supplies an effective process for determining
the n-ary term operations of any finite algebra with finitely many basic opera-
tions, and the process is effective for many other algebras as well. We will need
this definition: If f is any k-ary operation on a set A and is any set of n-ary
operations on A, we say that C is closed under composition by f iff g,, . ,gk-, E C
always implies f (g,, .- ,gk-,) E C.
THEOREM 4.3. Let A = (A, FA(FE 1)) be un algebra and let n > O. Then
Clo, A is the smallest set r, of n-ary operations on A including the n-ary projections
and closed under compositionby every basic operation FAof A.
Proof. For each n >: O let T, be the set defined in the statement of this theorem.
Define
I
We can prove the theorem by proving that r = Clo A. It is clear from Definition
l
l
4.2 that Clo,A includes the n-ary projection operations and is closed under
composition by each basic operation of A; i.e., we have that Clo, A i T, for each
l
n, and thus Clo A r> T.
To prove that Clo A c T, it will obviously suffice to establish this statement:
Let r' be the set of al1 operations g on A such that for al1 n, r, is
I
closed under composition by g. Then T' is a clone, and Clo A c T' c T.
We first establish that I" is a clone. Let k, m, and n be arbitrary, and let h
be a k-ary operation, h,, hk-, be m-ary operations, and f,, - ,fm-, be n-ary
e ,
operations on A. We ask the reader to verify the associative law for compositions: I
Now (1) easily implies that T' is closed under compositions. (Choose any k-ary
h in T' and any m-ary h,, , hk-, E r'. If fO, -,fmvl E r,, then h(h,, ,hk-,)
I
,fm-l) E T, by (l).) Since r' trivially contains al1 the projection operations,
I
(f,,
it is a clone. 1
By the definition of the T,, we see that T' contains the basic operations of
our algebra, so we have that Clo A c T'. To see that T' E T, let g be any member
of r', say m-ary. Note that
From this it follows that g E Tm,since Tmis closed under composition by g. The
proof is complete. E
+
To illustrate the concept of the clone of an algebra, let A = ( {O, 11, ), the
two-element group, with 1 + 1 = O. Using Theorem 4.3, it is easy to show that
Clo, A consists of the four operations
f(x,y)=x,y,x+y, and O ( = x + x ) .
It is a simple matter to characterize al1 the term operations of this algebra.
For another example, let U2 = (Q, +, ., -, 0 , l ) be the ring (or field)
of rational numbers. The term operations of U2 are easily determined, using
Theorem 4.3. They are just the operations defined by the familiar ring polyno-
mials with integer coefficients. Thus, a binary operation f (x, y) on rationals is a
term operation of U 2 iff it can be expressed in the form of a finite sum of integers
times powers of x times powers of y. The polynomials with rational coefficients
define a larger class of operations on Q. This larger class of operations is an
example for the following definition.
DEFINITION 4.4. The clone of polynomial operations of an algebra A,

denoted by Po1 A, is the smallest clone on the universe A that contains the basic
1 4.1 Algebras and Clones 145
operations of A and al1 the constant O-ary operations. The set of n-ary members
1
of Po1 A (n-ary polynomial operations of A) is denoted by Pol, A.
1 We remark that our terminology for the derived operations of an algebra

is not universally used, although it is gaining wide acceptance. Be aware that in
the literature of universal algebra prior to about 19.82, the term operations and
11 polynomial operations of an algebra were often called "polynomial functions"
and "algebraic functions," respectively.
THEOREM 4.5. Let A = (A, FA(FE 1)) be un algebra and let n 2 O. Then
Pol, A is the smallest set T, of n-ary operations on A including the n-ary projections
pr (O 5 i < n) and al1 the n-ary constant operations, and closed under composition
by al1 the basic operations of A.
Proof. Comparing Definitions 4.2 and 4.4, we see that Po1 A is identical with
Clo B where B has the same base set as A and its basic operations are those of
I A together with al1 the constant O-ary operations. Then Theorem 4.3 can easily
1 be applied to yield this theorem.
There is a direct way to construct al1 the polynomial operations of an algebra

from its term operations. We first give an example and then state the result as a
theorem. The function defined by
is a polynomial operation of the ring Q. To see this, let 1 = p; be the (unary)

identity function, let S and M be the (binary) addition and multiplication opera-
tions of the ring, let a,, a,, a, be the O-ary operations with values 1, $, and $,
and let b,, b, ,b, be the unary constant operations with the same respective values.
Then we have bi = ai(4) (i = 0,1,2), where 4 is the empty sequence of unary
operations. The function f defined by (2) is obtained from the basic operations
of 0 , and the projection operations and constants through compositions in this
way:
The operation f can also be obtained by substituting constants directly into a

term operation of Q. Namely, we can use
Clearly, g is a 4-ary term operation of Q, and f (x) = g(1, $, 2,x) for al1 x, or
what amounts to the same thing, f = g(b,, b,, b,, 1) where b,, b,, b, are the unary
constant operations defined above.
146 Chapter 4 Fundamental Algebraic Results
THEOREM 4.6. Let A be any algebra. If t(x,,. ,xm-,, u,, Su,_,) is un

,
m + n-ary term operation of A and ü = (a,,. ,a,-,) is an m-tuple of elements

of A , then the formula
(3) P(UO,. un-,) = t(ao, a,-, ,uo, ,un-,)
= t(ü,U)(where ü = (u,, u,-, ))S ,
defines an n-ary polynomial operation p of A. Conversely, if p is an n-ary polynomial

operation of A (for some n ) then there exists m 2 O and there exists an m + n-ary
term operation t of A and un m-tuple ü of elements of A satisfying (3). -
Proof. The proof is quite easy, so we merely describe it. Let l? be the set of al1
operations on A (n-ary for any n) for which t E Clo A and ü exist satisfying (3).
Show that r c Po1 A (using that Po1 A is a clone containing Clo A andathe con-
stant operations). Show that r is closed under compositions and contains the
term operations of A and the constant operations. So l? = Po1 A. m
DEFINITION 4.7. Let f (x,, ,xn-,) be any operation on a set A, and let
i < n. We say that f depends on xi iff there exist n-tuples ü = (a,,
- ,a,-,), a@
b = (b,,..., bn-,) in A" such that aj = bj for al1j # i, O 5 j < n while f ( ü ) # f(b).
We say that f is independent of xi iff f does not depend on xi. We cal1f essentially
k-ary iff f depends on exactly k of its variables; and we say that f is essentially
at most k-ary iff it depends on at most k variables.
THEOREM 4.8. Let k, n 2 1 and let f be any n-ary operation on a set A. Then
f is essentially at most k-ary 8there is a k-ary operation g on A and a sequence
of n-ary projection operations g,, . gk-, such that f = g(g,, - .. ,y,-,). More-
S ,
over, if f is essentially at most k-ary, then we can take g = f ( f,, - - - ,fn-,) for some
k-ary projection operations f,, - ,fn-, .
Proof. First suppose that f = g(g,, - gk-,) and the g,'s are projections. There
a ,
are at most k non-negative integers i < n such that for some j < k, gj = pr. If
i < n is not among these, then each gj ( j < k ) is independent of x i , and from this
it clearly follows that f is independent of xi. Hence f is essentially at most k-ary.
Now suppose that f is essentially at most k-ary. Let i, < i, < < i,-, < n
( E < k ) be a list of non-negative integers including al1 the i such that f depends
on xi. We define
f; = for al1 i in, i$(i,,-..,i,-,).

Now we put g = f (f,, . . . ,fn-,). Otherwise expressed, we have
where xj is occurring at the ij-th place in f (for O < j < 1 ) and x , occupies al1 other
places in f. Since f depends on at most xiO, , x ~ ~it -is~easy , to see that
4.1 Algebras and Clones
In other words, if we take
then we have f = g(g,;--,gk-,) and g = f(f,;.-, f,-,). •
The second volume of this work contains much material on the clones
of finite algebras and on the abstract clones, which constitute an important
generalization of the notion of a semigroup.
The set of al1 clones on a set is lattice-ordered by inclusion. The associated
lattice of clones is an algebraic lattice (see Definition 2.15.) For a finite set of three
or more elements, the cardinality of this lattice of clones is equal to that of the
continuum. (A proof of this fact is outlined in the ninth exercise below.) But for
A = (0,1), there are just countably many clones on A, and in fact every clone
on A is the clone of term operations of a two-element algebra with finitely many
basic operations. In principle, therefore, one can arrange al1 of the two-element
algebras in a denumerable list (algebras whose sets of term operations are
identical being regarded as equal). These results were proved in E. Post [1941].
Using modern results of the theory of finite algebras, we shall present a stream-
lined version of Post's results in Volume 2.
The concept of the Galois connection associated to a binary relation was
introduced in 52.2 (Theorem 2.21 and Examples 2.22). The most basic Galois
connection in algebra is the one associated to the binary relation of preservation
between operations and relations. (Nearly al1 of the most basic concepts in
algebra can be defined in terms of this relation.) When Q is an n-ary operation
on a set A, and R is a k-ary relation over A (for some n and k), then Q preserves
R just signifies that R is a subuniverse of the direct power algebra (A, Q ) ~ The .
polarities associated with the relation of preservation (see $2.2) are denoted by
and *. Thus, if Z is a set of relations over A, then CAis the set of al1 operations
on A that preserve each and every relation of C. And if r is a set of operations
on A, then TVis the set of al1 subuniverses of finite direct powers of the algebra
(A, r).Every set of the form CAis a clone.
Observe that the automorphisms, endomorphisms, subuniverses, and con-
gruences of an algebra are defined by restricting the preservation relation to
special types of relations. The congruences of an algebra, for example, are the
equivalence relations that are preserved by the basic operations of the algebra.
Exercises 4.9
l. Let A be an algebra and n be a non-negative integer. Let B = AA"be the
direct power of the algebra A (defined in the paragraph following Definition
1.5).Show that B = Clo, A (defined in Definition 4.1). Let F be a k-ary basic
operation symbol of A. Show that for any g,, . . gk-, E B, we have
a ,
A
~ ~ ( 9 0.,,gk-1)
. =F (90, . . - ,gk-,),
that is, it is the composition of these operations. Now sharpen Theorem 4.3
by proving that Clo, A is identical to the subuniverse of B generated by the
n projections p;f,pq, . - pn-, . Thus the n-ary term operations of A form an
S ,
algebra, which we shall denote by Clo,,A.

2. In the setting of the last exercise, show that Po1,A is identical to the
subuniverse of AA" generated by the projections and the constant n-ary
operations on A.
3. Supply a detailed proof of Theorem 4.6.
4. Let R be any ring with unit. Verify that the term operations of R are precisely
the operations that can be defined by a polynomial (in the sense of ring
theory) with integer coefficients. Verify that the polynomial operations of
R (our Definition 4.4) are the operations on R defined by polynomials with
coefficients from R.
5. Construct an algebra with one essentially binary basic operation that has
no term operation that depends on more than two variables.
6. Construct an algebra al1 of whose term operations of rank less than 9 are
projections, and which has a 9-ary basic operation that is not a projection.
7. Prove that every subuniverse of an algebra is closed under each of the term
operations of the algebra. Then show that the subuniverse generated by a
subset S of an algebra A is identical to the set of values of term operations
of A applied to elements of S. Show, in particular, that if S = {x, , ,x,-, }
then
. SgA (S) = {t(x,,- s., x,-,): t E Clo, A}.
8. Prove that every term operation of an algebra preserves al1 of its sub-
universes and congruences. (See Definitions 1.3 and 1.14.)
"9. This exercise is to construct a one-to-one mapping from sets of positive
integers to clones on the set N = {O, 1,2}. (The existence of such a mapping
was proved by Ju. 1. Janov and A. A. Mucnik [1959] and independently by
A. Hulanicki and S. ~wierczkowski[1960].) For any n z 1 let fn be the n-ary
operation defined by letting f,(x,, .,x,-, ) be 1 if xi 2 1 for al1 i and xi = 1
for precisely one value of i, and letting the value of the operation be O in al1
other cases. For every set S of positive integers define Cs to be the clone
generated by { f,: n E S}. Now show that Cs = C,, iff S = S'; in fact, for
any positive k, f,,, E CS iff k E S.
*lo. (Every algebra can be constructed from a semigroup.) Let A = (A, Fi(iE 1))
be any algebra. Prove that there exists a semigroup S = (S, ) with A G S
and having elements qi for i E I such that the following equation holds (for
al1 i E I and, if Fi is n-ary, for al1 a,, . a,-, in A):
S ,
* l l . An algebra A is called primal iff Clo, A = Clo, A for al1 n 2 1-Le., iff every
n-ary operation on the universe is a term operation of A for al1 n 2 1. If an
algebra is primal, then the subuniverses of its direct powers are quite special.
In particular, the algebra can have no congruences, subuniverses, automor-
phisms, or endomorphisms other than the trivial ones possessed by every
algebra. Prove that the n-valued Post algebra ({O, 1,. . . ,n - 11, A , v ,O, 1,')
4.2 Isomorphism Theorems 149
is primal. ((0,1, ,n - 11, A , v ) is the lattice determined by the ordering

O < n - 1 < n - 2 < . - - < l,andxt=x+ lforx#n-l,(n-1)'=0.
12. Let f : A" -+A and g: A" -+ A be operations. Show that f is a homo-
morphism (A, g)" -+ (A, g) iff g is a homomorphism (A, f )"-+ (A, f ) iff f
+
preserves g. (Recall that an n-ary operation is an n 1-ary relation.) f and
g are said to commute if these equivalent conditions hold.
13. Let C be a clone on A (see Definition 4.1). Define C' to be the set of al1
operations on A that commute with al1 members of C. (See the previous
exercise.) Show that C' is a clone and that (C')' i C. C' is called the
centralizer of C in Clo A. Show that if A = (A, S is an algebra, and if
)
Clone A is the category defined near the end of $3.6,with objects the powers
A", then the set of maps of Clone A is identical with the centralizer of Clo A
in Clo A.
*14. Let C be the clone on A = (0,l) that is the clone of term operations of a
two-element lattice. Determine the centralizer C' (defined in the previous
exercise) and the double centralizer (C')'. 1s (C')' = C?
4.2 ISOMORPHISM THEOREMS
A type (or similarity type) of algebras is a set I and a mapping p of I into

the set o of non-negative integers. The elements of I are called operation symbols,
of type p. An algebra of type p is an algebra A = (A, QA(QE 1)) in which QA
is a p(Q)-ary operation for every Q EI. The class of al1 algebras of type p
(called a similarity class and denoted by jyp) is closed under the constructions of
subalgebras, homomorphic images, and direct products introduced in Chapter
1. Whenever these constructions are discussed, it is always assumed that al1
algebras mentioned belong to the same similarity class. A similarity class is a
category. If we ignore the elements and the operations of algebras, then we
are left with algebras as "objects'@and homomorphisms as "morphisms," and
these constitute a category as defined in 93.6. Most of the concepts and results
presented in this section can be formulated in purely categorical language, but
we prefer to use algebraic language in most situations.
The Homomorphism Theorem of $1.4 (Theorem 1.16) is the first of severa1
results relating homomorphisms, congruences, and subalgebras. Collectively,
they are called the Isomorphism Theorems. When f : A -+ B and g: A -, C, the
existence of a homomorphism h: B -,C satisfying g = h o f implies that ker f E
ker g. (Whenever f(x) = f(y) we must have g(x) = g(y).) The converse of this,
which holds when f is surjective, is a most useful result. It implies, for instance,
that if 4 E y are congruences of an algebra A, then there is a natural homo-
morphism of A/d onto Aly.
THE SECOND ISOMORPHISM THEOREM (4.10).

i. Let f : A -,B and g: A -+ C be homomorphisms such that ker f E ker g and f
is onto B. Then there is a unique homornorphism h: B -+ C satisfying g(x) =
h(f (x)) for al1 x in A. h is un embedding fl ker f = ker g.
ii. Let 4 and y be congruentes of A with 4 G y. Then

: Y> E Y1
y14 = ( ( ~ 1 4~, 1 4 ) (x7
is a congruente of A/$, and the formula
S / ( ( ~ I ~ ) / ( Y=I ~
aly
)>
defines un isomorphism S/: (Al$)/(y/b) r A/y. (See Figure 4.1 .)
Solid lines for y classes

Dashed lines for 9 classes
Figure 4.1
Proof. To prove (i) we notice that since f (x) = f (y) implies g(x) = g(y), and
since f is onto B, the relation
is a function from B to C. Clearly g = h of. To see that h is a homomorphism,

let Q be one of the operation symbols, say it is n-ary. Let bo, bn-, be s..,
elements of B, and choose ai E A with f (a,) = b, for O < i < n. Then we have
which shows that h respects the interpretation of Q. It is clear from the definition
of h that h is one-to-one iff g(x) = g(y) -,f(x) = f(y).
To prove (ii), let f : A + A/q4 and g: A +A/y be the quotient homo-
morphisms. Now, the hypotheses of (i) are satisfied. The homomorphism h with
g = h o f has ker h = y/$. Moreover, for any a, b E A, we have (al#, b/$) E y/# iff
(a, b) E y. By the Homomorphism Theorem of 91.4, (A/$)/(y/#) r A/y, and it is
easy to check that S/ is the mapping that accomplishes the isomorphism. i
COROLLARY. Every homomorphism f : A -+ B factors as f = i o n where rc is

un onto homomorphism and i is un embedding.
4.2 Isomorphism Theorems 151
Proof. Let n be the quotient homomorphism of A onto Alker f. Then ker f =

ker n, so this eorollary follows from statement (i) in the theorem. ¤
The next theorem conveys a precise image of the relation between the
congruence lattice of an algebra and the congruence lattices of its quotient
algebras.
DEFINITION 4.11. The least and largest congruences of an algebra (A, - )

are denoted by 0, and 1,. (Occasionally, they are denoted simply by O and l.)
When elements of a lattice L satisfy a 5 b, we write I[a, b] for the interval
sublattice of L formed of al1 the elements in the closed interval I[a, b] =
{x: a S x 5 b).
THE CORRESPONDENCE THEOREM (4.12). Let A be an algebra and let

[ E Con A. The rnapping F defined on I [[,l,] by F(y) = y/[ is an isornorphisrn of
I[c, l,] with Con A/c. (See Figure 4.2.)
'"
8
Con A
----------
-- u ------
Con A/C
Figure 4.2
Proof. Let ngdenote the natural homomorphism of A onto Alc. This homo-
morphism is onto, and its kernel is 5. From this, one can easily show that the
map defined on Con A/c by a + nil(a) where
(4) n;l(a) = { (x, Y> E A2: (n4(x),nc<y)>E a}
is a one-to-one order-preserving mapping of Con A/C into the interval I[[, l,].
We observed in the proof of Theorem 4.10 that for any y E I [ ~ l,] , we have
y = n;'(~(y)). It is also easy to see that F(n;'(a)) = cc for al1 a ~ C o n A / [ .The
two maps are mutually inverse order-preserving bijections between the lattices
in question. Thus each of them is an isomorphism of lattices. w
DEFINITION 4.13. Suppose that B is a subset of A and 8 is a congruence of

A. Denote by B' the set {x E A: x 8 y for some y E B). Also define 81, to be 8 í l B ~ ,
the restriction of 8 to B.
We ask the reader to verify that if B is a subalgebra of A and 8 E Con A; then

Be is a subuniverse of A and 81, is a congruence on B. We denote by Be the
subalgebra of A with universe Be.
THE THIRD ISOMORPHISM THEOREM (4.14). If B is a subalgebra of

A and 6 E Con A, then Bl(6 1,) r Be/(8lB0).
Proof. Verify that f (b/6 1,) = b/0 l B 0 defines the required isomorphism. m
When f is a mapping from a set A into a set B, there are associated mappings
from relations over A to relations over B and, conversely, from relations over B
to relations over A. The image under f of an n-ary relation R over A is the relation
The inverse image under f of an n-ary relation S over B is the relation
".i
Let f be a homomorphism A -,B. The image and inverse image under
-'
subuniverses are subuniverses. The inverse image map f from Sub B to ub A
is order preserving; it preserves the meet operation of the lattices but oes not
in general preserve the join. (See the exercises.) The direct image map fr m Sub A
to SubB is order preserving; it preserves the join operation of the lattices but
does not in general preserve the meet. The inverse image under f of a congruence
is a congruence; the direct image of a congruence need not be a congruence. In
Theorem 4.12, we proved that if f : A -» B is surjective, then the inverse image
map and the direct image map, restricted to congruences above the kernel of f,
are mutually inverse lattice isomorphisms between Con B and a principal filter
in Con A.
The notations s g A ( x )(for the subuniverse of A generated by X G A) and
c g A ( x ) (for the congruence on A generated by X c A2) were introduced in
Definitions 1.8 and 1.19.
THEOREM 4.15.
i. Suppose that f : A -+ B and that X c A. Then f(sgA(x)) = sgB(f(k)).
ii. Suppose that f : A -++ B is surjective and that X E A2. Then
f (kerf v c g A ( x ) )= c g B ( f( ~ 1 ) .
, Proof. To prove (i), note that f(sgA(x)) is a subuniverse of B including
-'
f (X), and thus sgB(f (0)) f (sgA(x)).On the other hand, f (sgB(f (x)) is a
subuniverse of A including X, and thus also sgA(x).So f (sgA(x))c sgB(f (x)).
To prove (ii), let a = (ker f ) v c g A ( x )and P = cgB(f ( ~ ) )Since
. f -'(P) 2
X U lcer f and is a congruence, we have that f -l(P) 2 a. On the other hand, in
the isomorphism between I[ker f, l.,] and Con B, a corresponds to a congruence
f (a) that includes f (X), and so f (a) 2 P. •
Exercises 4.16
1. Let X be a subset of an algebra A such that s g A ( x )= A. (See Exercise 4.9(7).)
Let f : A -+ B and g: A -+ B be homomorphisms. Prove that if f (x) = g(x) for
a l l x ~ xthen
, f = g.
4.3 Congruences 153
2. Fínd an example of a homomorphism for which the direct image map does
not preserve intersections of subuniverses, and an example for which the
inverse image map does not preserve joins of subuniverses.
3. Find an example of an onto homomorphism f:A -+ B and an a E Con A such
that f(a)# Con B.Show that A cannot be a group or a ring. (See Theorem 4.68
in $4.7.)
4. Suppose that f:A r B and B s C (subalgebra). Prove that there exists an
algebra D i A and an isomorphism g:D E C with f c g.
5. In a category, an epimorphism is a map f such that for any two maps g and
h, g o f = h O f -+ g = h. Prove that in the category of al1 algebras of a type,
epimorphisms are onto homomorphisms.
6. Let A be an algebra and n be a natural number. According to Exercises
4.9(1-2), the sets of n-ary term operations and polynomial operations of A
constitute subalgebras Clo, A and Pol, A of AA".Let f : A -+ B be an onto
homomorphism. Prove that f induces onto homomorphisms f ':Clo, A -+
Clo, B and f ": Pol, A -+ Pol, B.
I 4.3 CONGRUENCES
-
In every branch of algebra, the congruences demand attention, and the deepest
results often require a systematic study of congruences. The first congruences
you met in the study of algebra were most likely the congruences (mod n) on
the ring of integers. In groups, congruences are conveniently replaced by normal
subgroups, the congruence classes of the unit element. In rings, they are re-
placed by ideals. In more general algebras, there is no possibility of such a
replacement.
Congruences of an algebra A are at once subuniverses of A2 and equivalence
relations over the universe A of the algebra. More exactly, we have the equation
(7) Con A = Sub A2 í l Eqv A
in which Eqv A denotes the set of al1 equivalence relations on A. The sets Con A,
Sub A2, and Eqv A are lattice ordered by set-inclusion, and we have here three
algebraic lattices. (Algebraic lattices were introduced in 52.2, Definition 2.15, and
will be discussed at length in $4.6.) The lattice ConA is a sublattice of the
full lattice of equivalence relations, Eqv A. It is a complete sublattice of
Eqv A-infinite joins and meets of elements in ConA are the same as those
computed in Eqv A. The join is the transitive closure of the union, and the
meet is the set intersection. In general, ConA fails to be a sublattice of
SubA2. (This holds, however, if A is a group or a ring.) Nevertheless, the con-
gruence lattice of any algebra is equal to the subuniverse lattice of another
algebra.
THEOREM 4.17. Let A be un algebra and let B = (A x A, be un algebrae )
whose basic operations are those of A2 plus O-ary operations C, = (a, a) for each
a E A, and a 1-ary operation S and a 2-ary operation T satisfying
Then Con A = Sub B.
Proof. A subset of A2 is a subuniverse of B iff it is a subuniverse of A2, is a re-

flexive and symmetric relation, and is transitive. This theorem follows from (7). m
The two following theorems supply useful tools for dealing with congru-
entes. This is an appropriate place to introduce nonindexed algebras. Algebras,
as we have known them so far, are indexed; basic operations are given as the
values of a function mapping operation symbols to operations. A nonindexed
algebra is an ordered pair (A, T) in which A, the universe, is a nonempty set and
T is a set of operations over A, called the basic operations. The definitions of
congruences, subuniverses, automorphisms, and endomorphisms of an algebra
go over without change to nonindexed algebras. When the context permits, we
will allow ourselves to omit the adjective "nonindexed." The next theorem shows
that the unary polynomial operations of an algebra determine its congruences.
THEOREM 4.18. For any algebra A, the algebras A, (A, Clo A), (A, Po1 A),
and (A, Pol, A) al1 possess exactly the same congruences.
Proof. We work with the polarities and A between sets of operations on A and
sets of equivalence relations over A. (See the final paragraphs of 44.1. For a set T
of operations, Tv is the set of al1 equivalence relations that are preserved by al1
members of T. For a set C of equivalence relations, CAis the set of al1 operations
that preserve al1 members of C.) Let r be the set of basic operations of A, We
have Con A = TV;moreover, rVA is a clone of operations on A containing the
basic operations of A and the constant operations. Therefore Po1 A (Con A)A,
which means that Con A c Con(A, Po1 A). But
Con A 2 Con (A, Clo A) 3 Can (A, Po1 A)
because of the reverse inclusions between the sets of basic operations of these
algebras. These three algebras are thus shown to have the same congruences.
-
Clearly, Con A S Con(A, Pol, A). To obtain the reverse inclusion, let P be
any equivalence relation belonging to (Pol, A)'. For an n-ary operation symbol
Q and (a,, a,-, ), (b,,
m , b,-, ) E An satisfying ai bi (mod P) (O _< i < n),
m ,
we have to show that eA(aO, un-,) z eA(b,, - - .,bn-,) (mod p). This would
e ,
follow from the transitivity of P if modulo P the following congruences hold:

~ * ( a , , . . - , abi+,;.-,
~, bn-,) = QA (a,;-.,a,-,, b i , - . -bn-,)
, (0 _< i < n).
There are clearly unary polynomials of A, fi (O L: i < n), such that the ith
4.3 Congruentes 155
congruence above takes the form fi(ai) = fi(bi). Since the fi preserves p, we have
the desired conclusion.
THE CONGRUENCE GENERATION THEOREM (4.19). Let X E A2

where A = (A, - m is un algebra. The congruence generated by X, or c g A ( x ) ,is
)
identical to the binary relation O defined by (a, b) E 0 iff

for some elements z,, , znE A and (x,, y,), - (x,-, ,y,-,)
a , EX
and for some f,, a ,f,_,E Poli A we have a = z,, b = z,, and
i ) }each i < n.
{zi,Zi+l}= { f i ( ~ i ) , f i ( ~for
Proof. On one side, it is trivial that c g A ( x )2 O, since c g A ( x )includes X and

is closed under unary polynomials (by Theorem 4.18). On the other side, it is
trivial that 0 includes X and is an equivalence relation. To see that 0 is closed
under the unary polynomials, suppose that (a, b) E 0 by the definition via
f,,..., f,-,.Thenif f €Pol,A, wehavethat (f(a), f(b))~Oviaff,,-a-,&-,.Thus
0 is a congruence of A by Theorem 4.18, and it follows that O 2 cgA(x).
We have seen in Theorem 4.18 that the congruences of an algebra are

determined by its monoid of unary polynomial operations. Algebras for which
this monoid is essentially a group of permutations are quite interesting to
consider. We give two results and some exercises, anticipating a detailed analysis
of congruences in finite algebras (the tame congruence theory) that will be
presented in a later volume.
Recall from 53.2 (just prior to Theorem 3.4) that a G-set, where G is a group,
is an algebra B = (B, fg(g E G)) such that g -+& is a homomorphism of G into
the group of al1 permutations of B. A G-set B is called transitive iff for every pair
of elements x and y of B there is g E G with &(x) = y. By the stabilizer of an
element b E B, written StabB(b),we mean the set {g E G: &(b) = b), a subgroup
of G.
Let us cal1 an algebra A permutational iff every nonconstant unary poly-
nomial operation of A is a permutation of the universe of A. An equivalent
condition is that there exists a G-set (A, fg(g E G)) (for some group G) such that
{ f,: g E G} is identical with the set of nonconstant unary polynomial operations
of A. For example, a vector space is a permutational algebra. The congruences
of a permutational algebra are the same as the congruences of the corresponding
G-set; if this G-set is transitive, its congruence lattice is determined by the next
lemma.
LEMMA 4.20. If A is a transitive G-set then for any a E A, Con A is isomorphic

to the interval IIStabA(a),G] in the subgroup lattice of G.
Proof. Let r be the G-set (G, l,(g E G)) where lg(h)= gh. Fix an element a E A.
Now the map from G to A defined by n(g) = &(a) is a homomorphism from r
to A, and it is onto, since A is a transitive G-set. By Theorem 4.12 (and the
Homomorphism Theorem of 51.4) we have Con A r I[ker n,l,], an interval in
Conr. We leave it to the reader to verify that the congruences of r are the
equivalence relations corresponding to partitioning G into left cosets of a
subgroup; that this sets up an isomorphism of Con I'with Sub G; and that under
this isomorphism, kern corresponds to the subgroup StabA(a). The lemma
follows from these facts. m
Our next lemma appeared in a paper of P. P. Pálfy and P. Pudlák [1980],

where it was used in a proof of the equivalence of two statements: (1) every finite
lattice is isomorphic to an interval in the subgroup lattice of a finite group; (2)
every finite lattice is isomorphic to the congruence lattice of some finite algebra.
At the time of this writing, it is unknown whether these equivalent statements
are actually true. A major portion of the proof of their equivalence is sketched in
Exercises 7, 8, and 9, where we ask the reader to supply the details.
DEFINITION 4.21. Let A be an algebra and X be a nonvoid subset of A. If f

is an n-ary operation on A such that X is closed under f then by f 1, we denote
the restriction of f to Xn, an n-ary operation on X. We denote by (Po1A)/, the
set of al1 restrictions to X of the polynomial operations of A under which X is
closed. We write A 1 for the nonindexed algebra (X, (Po1A)\.), and cal1 it the
algebra induced by A on X.
LEMMA 4.22. Let A be any algebra, let e be a unary polynomial operation of

A satisfying e = e o e, and set U = e(A). Then the restriction to U constitutes a
complete lattice homomorphism of Con A onto Con Alu.
lU
ConA -» ConAl,
Proof. The restriction mapping is clearly a mapping into the designated lattice,
and it trivially preserves al1 intersections. Let y be any congruence of A 1 ., Define
The reader can verify that y^ is a congruence of A. (It is preserved by al1 unary
polynomial operations of A-use Theorem 4.18.) It is also easy to check that
91, = y, and that for every congruence a on A we have a c y^ al, G y. From
these facts, it follows in particular that 1, maps Con A onto Con Al,, and that y^
is the largest congruence mapping to y, for each y E Con Al,.
To see that 1, preserves complete joins, let Os (SE S) be a family of congruences
on A and put
(where the second join is computed in Con Al,). From the above remarks, we
deduce that Os G y^ for al1 s E S. Thus 0 G y^,and this means that O 1, c y. Trivially
y O / , , and so we have shown that
which ends the proof.

4.3 Congruentes 157
The congruence lattice of an algebra determines an important property,

which we now define.
DEFINITION 4.23. An algebra A is called simple iff it has exactly two

congruences, 0, and 1,.
EXAMPLES
l. The lattice M, is a simple algebra. Every congruence other than O must
contain a pair of elements, one covering the other. By taking meets and joins
of these elements with suitably chosen third elements, we can successively
derive that every such covering pair belongs to the congruence. Thus al1
,
elements are identified in the congruence. The lattice M,, is also simple.
Con M, :
2. If R is the ring of al1 k x k-matrices with entries from a division ring F, then
R is a simple algebra.
Exercises 4.24
l. Let U be a direct product of two nontrivial unary algebras. (The basic
operations are l-ary; each of the two factor algebras has at least two
elements.) Prove that Con U is not a sublattice of Sub U2.
2. Prove that if R is a group or a ring, then Con R is a sublattice of Sub R2.
3. If G = (G, -',e) is a group, then Con G = Con(G,
S , S ) .
4. If A = (A, A , v , -) is a Boolean algebra, then Con A = Con(A, A , v ).

5. Let S = (S, A ) be a semilattice and S E S. The relation R defined by
( x , y ) ~ R i f f xA S = S - y A S =sisacongruenceofS.
6. Let L be any lattice and suppose that a, b, c, d E L satisfy a < b (i.e., b covers
a) and c 4 d. Prove that (c, d) E cgL(a,b) iff there exists f E Pol, L such
that { f(a),f (b)} = {c,d}.
7. Prove the converse of Lemma 4.20, that every interval in a subgroup lattice
is isomorphic to the congruence lattice of a transitive G-set for some group
G. (See Exercise 10 in $3.2.)
"8. (Pálfy and Pudlák [1980].) Let F be a finite algebra whose congruence
lattice L has the following properties: L is a simple lattice, and the atoms
of L join to 1. (An atom in a lattice is an element that covers the zero element.
This assumes, of course, that the lattice has a least, or zero, element.) Show
that L is isomorphic to the congruence lattice of a fínite G-set for some
group G. Here are some hints for working the exercise. Define C. to be the
collection of al1 sets p(F) with p any nonconstant unary polynomial of P.
Let M be a minimal under inclusion member of E. Next prove that there
exists a unary polynomial e of F satisfying e(F) = M and e = e o e. To
accomplish this, define a binary relation p on F by
Show that p is a congruence of F and p # 1, (the largest congruence). Since

the atoms of L join to l,, there is an atom a with a $ p, and so a í l p = 0,.
Now use this to show that there are p, q E PollF with p(F) = M = q(F) =
pq(F). Then the desired e can be found as a power of p. Now Lemma 4.22
implies that L r Con F 1., Show that every nonconstant unary polynomial
of FIMis a permutation of M.
"9. Find some special property of a lattice such that if the lattice L in the
previous exercise has the property, then this implies that the group of
nonconstant unary polynomials of FI, acts transitively on M.
10. Verify Example 2.
11. Let X be a (possibly infinite) set of at least 5 elements. Define Ax to be the
group of al1 permutations o of X such that for some finite set F c X,
o(x) = x for x E X - F and o is the product of a finite even number of
2-cycles. Prove that Ax is a simple group. (See the result of Exercise 10 at
the end of $3.5.)
12. Prove that if A is a set of at least two elements, then the lattice Eqv A of al1
equivalence relations on A is a simple algebra. Here are hints. Let R be a
congruence on Eqv A different from the identity relation. It must be shown
that R = (Eqv A) x (Eqv A). First show that OARafor some atom a. Then
if IA( 2 3, use intervals at the bottom of Eqv A isomorphic to Eqv{O, 1,2)
to transfer this result to al1 atoms; conclude that OARPfor every atom P.
The proof is essentially finished if A is finite. Assume that A is infinite. Parti-
tion A into two disjoint sets X and Y of equal cardinality, and let f be a
one-to-one function of X onto Y. Let ,u be the equivalence relation such that
X is the only equivalence class of ,u having more than one element, and let
v be the similar equivalence relation corresponding to Y. Let y be the
equivalence relation defined by (x, y) E y iff x = y or f (x) = y or f(y) = x.
Show that (,u v v)RIA,then consider the sublattice generated by ,u, v, and y.
*13. Prove that if A is finite, then Eqv A is generated by a set of four of its
elements. Show that the two-element lattice, and M,, are the only simple
lattices generated by three or fewer elements.
"14. (Maurer and Rhodes [1965].) Let G be a finite non-Abelian simple group.
Then Po1 G = Clo G-Le., every finitary operation on G is a polynomial
operation of G. Here is a sketch of a proof. The exercise is to supply the
4.4 Direct and Subdirect Representations 159
details. Write [x, y] for the commutator x-'y-'xy, put xY= y-lxy, and use
e to denote the unit element of G.
a. For every u # e and v in G there are y,, y, in G (for some n) such
m ,
that v = uY1uYz.. . uYn.

b. For every u # e # v in G there exists y such that [u, vY] # e.
c. Let b # e. For every n there is a polynomial f (x,, ,x,) such that
f (b, b, - ,b) = b and f (y,, - ,y,) = e whenever for some i, y, = e.
(You can take f (x, , - - ,x,) = n,,, hui for some elements a,, where
h(x,, ,x,) = [x, ,x?, ,xin] for some elements cj.)
d. Let b # e. There is a unary polynomial g such that g(e) = b and g(u) = e
for al1 u # e. (Say u,, u, are al1 the elements different from e. Choose
S ,
(by (a)) unary polynomials f , such that f,(e) = e and f,(ui)= b. Choose
f (x,, ,x,) as in statement (c). Take g(x) = f (bf,(x)-l, ,bf,(x)-l).)
e. Let n 2 1 and a,, a,, b E G. There exists f E Pol, G such that
m ,
f (a,, a,) = b and f (x,, ,x,) = e whenever xi # a, for some i.

m ,
*15. (R. Wille [1977].) If L is a finite simple lattice whose atoms join to 1, then
every finitary operation on L that preserves the lattice ordering of L is a
polynomial operation of L.
4.4 DIRECT AND SUBDIRECT REPRESENTATIONS
We introduced the direct product construction in 51.2 and remarked that it can
be used to build complicated algebras from simpler ones. Conversely, one can
sometimes represent big and seemingly complicated algebras as products of
simpler ones. A related construction, the subdirect product, is frequently more
useful in this respect than the direct product. In this section we discuss the
representations of an algebra as direct or subdirect product, focusing mainly on
the direct product. The first thing to note is that homomorphisms into a direct
product take a particularly simple form.
LEMMA. Let Bi (i E 1)and A be similar algebras and B =

natural bgection between hom(A, B) and ni,, ni,, B,. There is a
hom(A, B,). In fact, for every system
(f,: i E 1)E JJ i,, hom(A, B,) there is a unique homomorphism f E hom(A, B) satis-
fying f , = p, o f for a11 i E I, where p, is the projection of B onto Bi.
Proof. The function f +t (p, o f : i E 1) is a mapping of hom(A, B) into

ni,, hom(A, B,). Verify that it is one-to-one. For any systemf, hom(A, B,) (i I)
let n (A: iE 1) f denote the function from A to B defined by f(x)
=
E E
=
(f,(x): i E I). Verify that f is a homomorphism and pi o f = fi for al1 i E I.
DEFINITION 4.25. An isomorphism f : A z ni,,

Ai (or the associated sys-
tem of homomorphisms (f,: i~ 1)) is called a direct representation of A with
factors A,. By a subdirect representation of A with factors A, we mean an
embedding f : A 4 ni,,
A, (or the associated system (A: i E 1)) such that f , is
onto Ai for al1 iEI. By a subdirect product of (A,: ~ E ' Iwe

) mean an algebra
BE ni.,A, such that p, maps B onto A, for al1 i E l . (We say that B projects
onto each factor.)
We shall cal1 two subdirect representations A --* ni,,

A, and A 5 ni
Bi
isomorphic if there exist isomorphisms h,: A, r B, (i E I) such that hif, = g, for al1
i~ 1. By Theorem 4.10, f and g are isomorphic iff kerf, = kerg, for al1 i. A
subdirect representation f is determined within isomorphism by the system of
congruences (kerf,: iEI). Therefore every subdirect representation is iso-
morphic to one in the form
(8) f : ~ + n ~ / 4 ,'
isI
with f(x) = ( ~ 1 4~
~ :E Ifor
) a11 x E A.
In al1 work with subdirect products, we shall feel free to deal directly either with
an embedding f, with the correlated homomorphisms f,,or with the congruences
4i = kerf,, whichever is most convenient for the work at hand.
DEFINITION 4.26. A system (4i: i E 1) of congruences of an algebra A (or a
system (f,: i E 1) of homomorphisms from A) is said to separate points of A iff
ni,, ni
4i = O,, the identity relation on A (or E ker f, = O,).
TWEOREM 4.27. A system (qbi: i E I) of congruences of un algebra A (or system
(f,: i E 1) of homomorphisms from A) gives a subdirect representation $ it sepa-
rates the points of A.
Proof. The kernel of the mapping (8) is precisely O{#,: i E I}, hence this map-
ping is an embedding iff the system of congruences separates points. (Note that
if (A: i E I) separates points and f,:A + Ai, then we have a subdirect representa-
tion with factors f,(A), but it may be that f,(A) # Ai.) ¤
Theorem 4.27 shows that the subdirect representations of an algebra can be

determined (up to isomorphic representations) by examining its congruence
lattice. In this sense, the concept of subdirect representation is purely lattice-
theoretic. Every subset of the congruence lattice that intersects to zero determines
a subdirect representation. Thus an algebra is likely to have many rather different
subdirect representations, with entirely different sets of congruences involved.
The concept of direct representation, on the other hand, is not a purely lattice-
theoretic one; it requires congruences that permute with one another. To see
what is involved, we focus on representations with two factors.
DEFINITION 4.28.
l. The relational product of two binary relations a and P is the relation
a o /3 = ((x, y): 3z((x, z) E a and (z, y) E B)}.
Two binary relations a and P are said to permute if a o P = P o a.
I 4.4 Direct and Subdirect Representations 161
2. Let a and /3 be equivalence relations on a set A. We write O, = a x P and

cal1 (a, p ) a pair of complementary factor relations iff
1 and moreover a and /3 permute.

3. Let a and /3 be congruences of an algebra A. Then (a, P ) is a pair of comple-
mentar~factor congruences of A iff 0, = a x p.
4. ' By a factor congruence of A we mean a congruence a such that there exists a
congruence p satisfying O, = a x P.
Thus if a'and P are congruences of A, then O, =a x P signifies that a and

/3 are complements of one another in Con A and that they permute. We remark
that equivalence relations a and P permute iff a v /3 = a o P.
THEOREM 4.29. A pair (4,, 4, ) of congruences of un algebra gives a direct
representation iff 0, = 4, x 4,.
Proof. The reader may easily verify that the homomorphism x i-, (xld,, x/d, )
is injective iff 4, A 4, = O, and maps onto A/b, x A/b1 iff 4, o q51 = 1, =
4, o 4,. (Either one of these equalities implies the other.) m
It is easy to see that a finite system (4,, - 4n-, ) of congruences of A is a
m ,
direct representation iff 4i are pairwise permuting factor congruences and for
each i < n, (4i, j # i ) ) is a pair of complementary factor congruences.
Here is a characterization of arbitrary direct representations in terms of con-
gruence relations.
DEFINITION 4.30. For any system (@i:i E I ) of congruences of an algebra A

n
we write O, = i,, 4i iff 0, = {bi:i E I } and for every Z = (x,: i E I ) E A' there
is an element x E A such that (x, xi) E 4i for al1 i E 1.
THEOREM 4.31. A system (4i: i E I ) of congruences of un algebra A constitutes
a direct representation $ OA = , &. n
Proof. It is easily verified that the two conditions of Definition 4.30 are equiva-
lent to the injective and surjective properties of the mapping defined by formula
(8). m
To every direct representation of an algebra with n factors is naturally
associated a certain n-ary operation on the universe. The factor congruences of
the representation and this operation are interchangeable, each definable directly
from the other. The operations associated with direct representations, called
decomposition operations, play a significant role in clone theory and in the theory
of varieties; they will become quite familiar to the reader of these volumes. We
deal here only with binary decomposition operations, saving the extension to n
factors for an exercise.
DEFINITION 4.32. A (binary) decomposition operation of an algebra A is a

homomorphism f : -t A satisfying the following equations (for al1 x, y, zE A).
Any operation f on a set A satisfying (9) will be called a decomposition operation

on A.
Given an algebra A = B x C, we define the canonical decomposition opera-

tion by setting f ((b, c), (b', c')) = (b, c'). Clearly, this is a homomorphism of
A2 onto A, and it satisfies equations (9). Conversely, every decomposition opera-
tion on a set gives rise to a direct representation in which the operation takes
this simple form.
LEMMA. Suppose that f : A2 -+ A satisfies (9).

i. For al1 x, y, u, v E A we ,have
f(x,u) = f(y7 u)++f(x,v) = f(y,v),
f(u,x) = f(u,~)-f(v,x) = f(v,y).
ii. The relations
are equivalence relations on A satisfying O, = yo x yl. For al1 (x, y) E A2,

f (x, y) is the unique element z E A satisfying
X Z y.
iii. (yo7y1 ) is a pair of complementary factor relations of the algebra (A, f ). We
have
(A,f) <A,f )/yo x <A,f )/VI = A0 Al
, x and f
where f A O (y)~ = y) = y.
Proof. The straightforward proof is Exercise 8.
THEOREM 4.33. Let A be any algebra. There is a bijective correspondence

between pairs of complementary factor congruences of A, and decomposition
operations of A, such that a decomposition operation f and the corresponding pair
(q,, q ) of factor congruences satisfy statement (ii) of the Lemma.
Proof. Let f be a decomposition operation of A (Definition 4.32). For any

x, y E A we define functions f and fy by f "(u) = f (x, u) and fy(u) = f (u, y). By
the lemma, ker f a = ker f = yl, and ker f, = ker f, = y, (for al1 a, b E A). The
homomorphism property of a decomposition operation of A translates into the
4.4 Direct and Subdirect Representations
following. For every basic operation g of A (say g is n-ary) we have

(10) f(g(x0, - .,xn-l),g(Yo, .. ,Y,-,)) = g(f(xo,yo), .. ,f(xn-,, Y,-,))
for al1 x,, yo, , xn-,, y,,-, E A.
Combining equations (9) and (lo), we easily derive that if p is any one of the
functions f a or f,and g is a basic operation, then
Equation (11) falls just short of implying that p = f a is an endomorphism of A,

but it does easily imply that ker f a = y, is preserved by the operation g. Since g
is any basic operation of A, we can conclude that y,, and likewise q,, are
congruences of A. By the lemma, 0, = y, x vi, and so (yo, y,) is a pair of
complementary factor congruences of A.
To see that every pair of complementary factor congruences of A arises in
this way from a decomposition operation of A, let ( q o , q, ) be such a pair. Either
take the natural decomposition operation of Aly, x A/ql and pul1 it over into
A via the isomorphism, or define f directly as the unique solution to
The two procedures give the same operation f, and it is quite elementary to see
that f is a decomposition operation of A related to (y,, y,) as in statement (ii)
of the lemma.
The equation (10) holds for operations f and g iff f is a homomorphism of

) ~ (A,g) iff g is a homomorphism of (A, f )" into (A, f ). This is a
( A , ~ into
special case of an important symmetric relation between operations, which we
now define for operations of arbitrary rank.
DEFINITION 4.34. Operations f : A" 4 A and g: A" + A are said to commute

iff for al1 x,,, x,-~,-, EA,
m - . ,
Let f be the natural decomposition operation of an algebra A = B x C.

(This discussion will apply to any decomposition operation of an algebra, but
what we wish to say is intuitively more clear in this case.) Thus f commutes with
each basic operation of A. The set of al1 operations on A that commute with f
is a clone, called the centralizer of f in Clo A. This clone is precisely the set of
operations that act coordinatewise in the product set, so it contains Po1 A as well
as f. (But we cannot expect the decomposition operation f to be a polynomial
operation of A, although it will be, for example, if A is a ring with unit or a lattice
with O and 1.) Recall from the last paragraphs of $4.1 that { f 1' denotes the set of
al1 relations over A that are preserved by f. Then the centralizer of f in Clo A is
precisely the set of operations belonging to { f j v (where we identify an n-ary
operation with an n + 1-ary relation).
DEFINITION 4.35. Let p be an n-ary relation on a product set B = B, x B,,

p will be called factorable iff there exist n-ary relations pi on Bi (i = 0,l) such
that
When this holds, the relations p, and p, are called the factors of p. (They are
uniquely determined if p # @.)
THEOREM 4.36. Let f be the decomposition operation for un algebra A =

B x C.
i. (f }' is the set of al1 factorable relations on A.
ii. The set of factorable congruence relations of A constitutes a sublattice of
Con A isomorphic to Con B x Con C. (Similar facts hold with automorphisms
or endomorphisms in place of congruence relations.)
iii. The clone on A generated by Clo A U ( f ) is identical with the set of factorable
operations on A whose respective factors belong to Clo B and Clo C.
Proof. It is convenient to use ( b , ~ )to denote the n-tuple ((b,, c,), S ,
( b n ~ , , c n ~ , ) ) i n A n w h e r e b(bo,--.,bn-,)€BnandF=
= <c,,~~~,c,~,)~C".
To prove (i), first let p be a factorable relation on A with factors p, and p,.
Thus (6, c) E p iff b E p, and FE p, . To see that f preserves p, we must deal with
the operation f of (A, f )". (We must show that p is a subuniverse of (A, f )".)
Now it is easily verified that f((b, e), ( g ,e' )) = (5, e' ). Thus, clearly if (6, e),
(b:2') E p then (6, F') E p. (Because p is factorable.) We have proved that p E
(f lv,as desired.
The proof that every relation preserved by f is factorable is left to the reader.
It is quite easy, using the above description of the operation f on A". If p is a
nonempty relation preserved by f, just choose (6, F ) E p and define
po = (z:( ? , C ) E ~ } and p, = (y: ( 5 , y ) ~ p ) .
To prove (ii), note that for every (p, ( ) E: Con B x Con C there is a unique
factorable binary relation r(P, c) on A having p and 5 for factors. (((b, c), (b', c')) E
r(P, C) iff (b, b') E /3 and (c, c') E c.) This relation r(P, () is clearly a congruence
of A. Verify that the map r is one-to-one and preserves meets and joins. Thus r
is an isomorphism of Con B x Con C with the lattice of factorable congruence
relations of A.
To prove (iii), let E be the clone generated by Clo A U (f }. Let F be the set
of factorable operations whose factors belong to Clo B and to Clo C. Verify that
F is a clone. Since it includes the generators of E, we have that F includes E. To
get the other inclusion, let h be any n-ary member of F, with factors hB and hC.
Now we need to know that every term operation of B is a factor of some term
operation of A, and a similar fact about the term operations of C. The verification
of these facts is left to the reader. (Later, when we have our hands on "terms" as
well as term operations, these facts will be obvious.) So let g, and g, be n-ary
term operations of A such that hB is a factor of g, and hC is a factor of y,. Then

g = f (g,, g,) is a member of E and has the factors hB and hC. From this it follows
that g = h, which ends our proof that E = F. m
DEFINITION 4.37. An algebra A is called directly indecomposable iff it is
nontrivial and is not isomorphic to a product of two nontrivial algebras.
Every finite algebra of prime cardinality is directly indecornposable. Every

simple algebra is directly indecomposable. Directly indecomposable algebras can
be characterized as those that have precisely two decomposition operations,
f (x, y) = x and f(x, y) = y, or as those that have precisely two factor congruences,
O, and 1,.
In severa1important classes of algebras, the factor congruences are very well
behaved, and it is posible, by examining them, to prove that an algebra has an
essentially unique direct representation with directly indecomposable factors
(just as every positive integer can be uniquely factored into a product of primes).
Finite groups and rings, for instance, have this unique factorization property.
Many results of this type are proved in Chapter 5. The strict refinement
property for direct decompositions is likewise studied there. If an algebra has
this property and has a factorization into a product of directly indecomposable
algebras, then the set of factor congruences involved in such a factorization is
uniquely determined. The strict refinement property has severa1 equivalent for-
mulations. One is that the set of factor congruences of the algebra constitutes a
Boolean sublattice of the congruence lattice. (In general, this set will not even be
a subuniverse of the congruence lattice.) Another is that every pair of decompo-
sition operations of the algebra commute. Therefore, any algebra whose de-
composition operations are polynomial operations (for example, any ring with
unit) has the strict refinement property. The point of Exercise 16 below is essen-
tially that every algebra with a distributive congruence lattice has the strict
refinement property.
Exercises 4.38
l. Let {e) be a 1-element subuniverse of an algebra A, and 4, be comple-
m e n t a r ~factor congruences of A. The subalgebra of A whose universe is el4
is isomorphic to A/4*.
2. Every finite algebra is isomorphic to a product of directly indecomposable
algebras.
3. Find examples of similar algebras A and B for which the set of factorable
subuniverses of A x B is not a subuniverse of the lattice Sub A x B. If the
natural decomposition operation of A x B is a term operation, however,
then every subuniverse of A x B is factorable. Even then, the formula
SubA x B r SubA x SubB may fail. (The empty subuniverse may be
factorable in severa1 ways.) Verify that the isomorphism holds if the de-
composition operation is a term operation and these algebras have a O-ary
operation symbol.
4. Find a pair of similar algebras A and B, neither of which can be embedded

into their product.
5. By Theorem 4.29, the direct representations of an algebra whose con-
gruences permute can be read off from the picture of its congruence lattice.
This applies to groups. Thus, for instance, a nontrivial group is directly
indecomposable iff its congruence lattice has no complemented element
other than O and 1. In some cases, the picture of a congruence lattice will
te11 us much more. Prove that a group G has this picture of its congruence
lattice
iff G is isomorphic to Z$ (the direct square of the 5-element cyclic group).

6. Verify that Con Z5 $ (ConZ,)'. Compare this result to the second state-
ment in Theorem 4.36.
7. A relational structure is a nonvoid set paired with a system of finitary
relations over that set, usually written in the form (A, Ri(i E 1)). The notion
of similarity for relational structures is defined in just the same way as
for algebras. We define the direct product of similar relational structures
(A, Ri(iE 1)) and (B, Si(iE 1)) to be the structure ( A x B, T,(iE 1))in which
T, is the factorable relation that factors into Ri and Si. Show that if Lo
and L, are lattices whose lattice orders are 1 and 5 ',and if 1 is the
lattice order of Lo x L,, then (Lo, 1O ) x (L,, < ') = (Lo x L,, 5 ).
8. Prove the lemma following Definition 4.32.
9. An algebra A has factorable congruences iff whenever A r B x C, every
congruence of B x C is factorable. Prove that A has factorable congruences
iff for every pair 4, 4* of complementary factor congruences of A, and for
every congruence p, this equation holds: P = (P v 4) A (P v 4").
10. If Con A is a distributive lattice, then A has factorable congruences. (Note
that by Theorem 2.50 in $2.5, the condition holds if A is a lattice.)
11. Let R be a ring with unit. Discover the bijective correspondence between
pairs of complementary factor congruences of R and pairs (e, e') of ele-
ments of R satisfying: (i) e e' = O; (ii) e + e' = 1; (iii) e . e = e; (iv) e' . e' =
e'; and (u) each of the elements e, e' = x satisfies x - y = y x for al1 y E R.
Show that a binary operation f(x, y) on R is a decomposition operation of
R iff there exists a pair (e, e') satisfying (i-v), and (vi) f(x, y) = xe + ye'
for al1 x, y E R. Moreover, (i-vi) imply that e = f(1,O) and e' = f (O, 1).
12. Let L = (L, A , v ,O, 1) be a lattice with O and 1. Let f(x, y) be a decompo-
sition operation of L. Show that the elements e = f(1,O) and e' = f(0,l)
satisfy: (i) f(x, y) = (x A e) v (y A e'); (ii) e v e' = 1 and e A e' = O; (iii)
each of z = e, e' satisfies, for al1 u, V E L, (z A u) v (u A u) v (u A z) =
(Z v U) A (U v v) A (v v z). Elements satisfying (iii) are called neutral (see
$2.6, Exercise 5). Show that whenever e, e' are neutral and satisfy (ii)
then the operation defined by (i) is a decomposition operation of L.
13. Let A = (A, A , v , - ) be a Boolean algebra. Prove that every element of the
lattice L = (A, A , v ) is neutral. (See the preceding exercise.) Conclude
that for every a E A, f (x, y) = (a A x) v (a- A y) defines a decomposition
operation of L. With Exercise 4.24(4), conclude that f (x, y) is a decomposi-
tion operation of A. Thus A is directly indecomposable iff it has precisely
two elements.
14. Let A be an algebra and n be a positive integer. An n-ary operation f on the
universe of A is called a decomposition operation iff it obeys the equations
(for al1 choices of x, x,, , - of course); and called a decomposition operation

a ,
of A if, in addition, f commutes with the operations of A. Develop a theory

of n-ary decomposition operations, including our results of this section as
the special case n = 2.
15. Let f and g be an m-ary and an n-ary decomposition operation of an algebra
A. Prove that f and g commute iff f (g(x, ,, - x,,),
e , ,g(xm,, ,xmn))is a
decomposition operation of A.
16. Suppose that ConA is a distributive lattice. Show that every factor con-
gruence of A permutes with every congruence of A. Prove that the factor
congruences form a sublattice of Con A.
17. Let A c n,, B, be a subdirect product (See Definition 4.25.) For t E T let
y, be the kernel of the projection at t, restricted to A. Show that y, and
n { y . : S E T and S # t) are complementary factor congruences of A for al1
t E T iff the following holds: for every XEA, if 7 belongs to the product
algebra and x(t) = y(t) for al1 but finitely many t E T, then FEA.
18. ([nl-th powers) For any nonvoid set A and positive integer n we define
an n-ary operation eA on A" as follows. Let Fi = (xil, xin)E An for
S ,
1 5 i 5 n. Then we define e A ( F , , . . . , ~=) (xnn,xl,,x,,;~~,xn-,n-, ).

Find a set of equations, analogous to those defining decomposition opera-
tions, such that an n-ary operation F on a set B obeys the equations iff
(B, F) is isomorphic to (An, e*) for some set A. (It may be useful to observe
that the natural n-ary decomposition operation of A" is a term operation
of (An, en).)
19. Suppose that A is an algebra of finite type, and without losing much
generality, suppose that A has precisely k basic operations, and they are al1
k-ary, k > 1. Let F,, ,F, be a list of these operations. Set B = Ak and
define a unary operation f on B. Namely, for any F EB put f(x) =
(F,(X), . ,F,(x)). Let B = (B, f, eA) where en is the operation from the
last exercise. Now show that every congruence p of the algebra B is a fac-
torable equivalence relation on the set Ak, and its factors at al1 coordinates
are the same, a congruence a of A. Using this, prove that Con B r Con A.
"20. Let A be any algebra and T be a set. A homomorphism f : AT -+C will be
called finitely determined if there exists a finite subset T' c: T such that if x
and y i n agree on T' then f(3) = f(y). Now let B be the additive group
of integers and T be a countable set. Prove that every homomorphism from
BT to B is finitely determined.
4.5 THE SUBDIRECT REPRESENTATION THEOREM
We shall have a good deal to say in Chapter 5 about the situations in which an
algebra has only one (essentially unique) representation as a direct product with
directly indecomposable factors. When this occurs, it can be decisive for the
algebraic theory, as we see in the following examples. Every finite Boolean
algebra is isomorphic to a direct power of the two-element algebra, and almost
anything you might want to know about finite Boolean algebras can be deter-
mined using this fact. (See Theorem 2.62 or Exercise 4.38 (13).) Every finitely
generated Abelian group is isomorphic to a direct product of finitely many cyclic
groups. Every finitely generated ring obeying a law xn Ñ x (n > 1) is isomorphic
to a direct product of finitely many finite fields. (See Example 4 and Corollary
4 to Theorem 4.44.) Every semisimple ring satisfying the descending chain
condition for left ideals is isomorphic to a direct product of finitely many full
matrix rings over division rings. Every complemented modular lattice of finite
height is isomorphic to a direct product of a finite Boolean lattice and finitely
many projective geometries. (See Theorem 4.88.)
Every finite group G is isomorphic to a direct product G, x x Gnwhere
the Gi are directly indecomposable, and the sequence (G,, - G,) is uniquely
S ,
determined up to a permutation of the sequence and replacement of the Gi by

isomorphic groups. (See Theorem 5.3.) This example is quite different from those
above in that it comes close to being practically useless. Directly indecomposable
groups include simple groups and many other groups that we have little knowl-
edge of. Just for this reason, the study of finite groups has never relied on their
unique factorizations.
The usefulness of direct product representations is generally limited by two
facts. Direct indecomposability is a rather weak property of most algebras
(despite the impression made by some of the examples above); and many algebras
fail to have any decomposition into indecomposable factors. For example, a
denumerable Boolean algebra and a vector space of denumerable dimension over
the field of rational numbers both fail to have such a decomposition.
In 1944, Garrett Birkhoff introduced subdirect praducts (which he called
subdirect unions) and proved the important Subdirect Representation Theorem.
This theorem states that (unlike the case for direct products) every algebra has
a subdirect representation with subdirectly irreducible factors.
We defined subdirect products in the last section and discussed them briefly
(Definition 4.25, Theorem 4.27). A subdirect product of a system of algebras
(A,: i E I ) is simply a subalgebra A of their product with the added property that
each Ai is a homomorphic image of A via the projection map. A subdirect
4.5 The Subdirect Representation Theorem 169
representation of an algebra A is an embedding of A into a product n i s I A i ,

given by a system of homomorphisms of A onto the A,. Subdirect representations
will also be called subdirect embeddings.
Some algebras have only trivial subdirect representations. Let us define
precisely what we mean by this.
DEFINITION 4.39. An algebra A is called subdirectly irreducible iff 1 Al > 1 and

for every subdirect embedding f : A + ni,,A, with associated homomorphisms
fi: A + A,, there exists an i for which fi: A r A,.
Here are two useful intrinsic characterizations of subdirectly irreducible

algebras.
THEOREM 4.40. Each of these statements is equivalent to the subdirect irredu-

cibility of A:
i. Con A has a smallest nonzero member, that is, a congruence p such that for
every congruence a, p 5 a iff a # O,.
ii. There exist elements a, b E A such that for every pair (c, d) of elements of
A, (a, b) E cgA(c,d) Iff c # d.
Proof. To see that (i) implies (ii), suppose that P is the smallest nonzero
congruence of A. Choose any pair (a, b) E P - O,. If c and d are distinct elements
of A, then Cg(c, d) # O, hence P c Cg(c, d) and (a, b) E Cg(c, d). On the other
hand, if c = d, then the congruence generated by (c, d) is just O, and (a, b) does
not belong to this congruence. Thus a and b satisfy (ii).
Now suppose that (ii) holds. We show that A is subdirectly irreducible.
Let a, b be such a pair of elements. Thus a # b. Let (fi: i € 1 ) be a system of
homomorphisms from A constituting a subdirect representation. Since this sys-
tem separates points (Definition 4.26, Theorem 4.27) there must be iE I such
that &(a) # fi(b). Thus (a, b) 4 ker fi, and it follows from (ii) that the congruence
kerh can be none other than O,. Since fi is one-to-one, it is an isomorphism.
(These homomorphisms are assumed to be onto maps.)
Now assume that A is subdirectly irreducible. Let be the set of al1
nonzero congruences of A, and fi = n Z . If p # O,, then P is clearly the least
nonzero congruence of A, so (i) holds. Assume that P = O,. Then (o:o E C)
gives a subdirect representation of A (by Theorem 4.27). None of the associated
homomorphisms f, is one-to-one, because ker f, = a. This contradicts the sub-
direct irreducibility of A. Thus indeed (i) holds if A is subdirectly irreducible. 4
DEFINITION 4.41. The least nonzero congruence of a subdirectly irreducible

algebra is called its monolith. When elements a, b of A satisfy 4.40(ii), we say that
A ik (a, b)-irreducible.
Note that A is (a, b)-irreducible iff A is subdirectly irreducible, (a, b)

belongs to its monolith, and a # b.
We defined simple, directly indecomposable, and subdirectly irreducible

algebras in Definitions 4.23, 4.37, and 4.39. These three properties of algebras
are obviously related as follows:
simplicity =. subdirect irreducibility + direct indecomposability.
The first two properties are purely lattice-theoretic, in the sense that a glance
at a congruence lattice will te11 us whether an algebra has the property. These
are the pictures we look for:
Con A
Simple Subdirectly irreducible

Figure 4.3
EXAMPLES
a. The two-element Boolean algebra is the only subdirectly irreducible
Boolean algebra.
b. A vector space is subdirectly irreducible iff its dimension is equal to 1.
c. The lattice N, is a subdirectly irreducible algebra. Its monolith identifies the
two middle elements of the four-element chain, and no other elements.
d. Those two middle elements in N, can be replaced by a copy of N, to get

another subdirectly irreducible lattice N,. This process can be iterated.
4.5 The Subdirect Representation Theorern
Con N,
e. The cyclic group Z,, is subdirectly irreducible for each prime p and n 2 1.
The infinite quasi-cyclic group Z,, (p a prime), which is the multiplicative
group of complex numbers a satisfying apk= 1 for some k 2 O, is subdirectly
irreducible. These cyclic and quasi-cyclic groups are the only subdirectly
irreducible Abelian groups. (Exercises 2 and 7 below.)
To prove the Subdirect Representation Theorem, we require the concept of
a strictly meet irreducible element of a lattice.
DEFINITION 4.42. Let L be a complete lattice and u be one of its elements.

We cal1 u strictly meet irreducible iff u # 1 and for al1 S c L, if u = AS, then
u E S. (The notion of a strictly join irreducible element is defined dually.)
LEMMA 4.43. Let 8 be a congruence of the algebra A. The subdirect irreducibility

of A/8 is equivalent to each of these statements.
i. 8 is a strictly meet irreducible member of Con A.
ii. There exists a pair ( a , b ) of elements of A such that 8 is a maximal member of
the set of congruentes not containing ( a , b).
Proof. The proof uses Theorem 4.40 and Theorem 4.12 (the Correspondence
Theorem). The congruence lattice of A/B is isomorphic to the interval lattice
I[8, l,] contained in Con A (by 4.12), and so (by 4.40) A/8 is subdirectly irreduc-
ible iff (6 E Con A: 8 < 6 ) has a least member. Clearly, such a least member exists
iff 8 is strictly meet irreducible. Thus (i) is equivalent to the subdirect irreduci-
bility of A/8. The equivalence of (ii) with either of the other conditions can be
proved easily.
Note that elements a, b of A satisfy statement (ii) of the lemma iff A/8 is
(a/8, b/8)-irreducible.
THE SUBDIRECT REPRESENTATION THEOREM (4.44). Every algebra

A has a subdirect representation with subdirectly irreducible factors that are
quotient algebras of A.
Proof. Let A be any algebra and define I = { ( a , b ) E A2: a # b}. For each
i = ( a , b ) E I , choose a maximal member 8, in the set ( 4 E Con A: (a, b ) 4 4 ) . (The
existence of such a maximal member follows easily by Zorn's Lemma.) Now A/Oi
i E I } = O,.
is subdirectly irreducible for al1 i E I , by Lemma 4.43, and clearly (-){Oi:
By Theorem 4.27, this system of congruences gives a subdirect representation, as
desired. (This theorem is also an immediate corollary of Theorem 2.19-i.e., of
the fact that every element in an algebraic lattice i~ the meet of a set of strictly
meet irreducible elements.) ¤
COROLLARY l . Every finite algebra can be subdirectly embedded into a

product of finitely many finite subdirectly irreducible algebras.
EXAMPLE 1: ({O, 11, A , v , -) is the only subdirectly irreducible Boolean

algebra (up to isomorphism, of course). The direct powers of this algebra are the
full algebras of subsets of sets (again up to isomorphism). If we apply Theorem
4.44 to a Boolean algebra A, the subdirectly irreducible factors in the representa-
tion are Boolean algebras, since they are homomorphic images of A. By Exercise
4.38(13), a subdirectly irreducible Boolean algebra can only be a two-element
algebra. This analysis leads again to the result in Theorem 2.62 that every
Boolean algebra is isomorphic to an algebra of sets.
EXAMPLE 2: Representations of semilattices and distributive lattices in alge-
bras of sets were produced in Theorems 2.61 and 3.7. These results can also be
derived from the Subdirect Representation Theorem.
COROLLARY 2.
i. Every subdirectly irreducible semilattice is isomorphic to ({O, 1}, A ). Every
semilatticeis isomorphic to an algebra of sets under the operation of intersection.
ii. Every subdirectly irreducible distributive lattice is isomorphic to the lattice
({O, 1}, A , v ). Every distributive lattice is isomorphic to un algebra of sets
under the operations of union and intersection.
Proof. Let S = (S, A ) be a subdirectly irreducible semilattice. Assume that it

is (a, b)-irreducible. (See Definition 4.41.) We can assume that b # a A b (or else
a # b A a). Let O, denote the congruence defined in Exercise 4.24(5).Then (a, b) 81:
O, and thus we must have O, = O,. Since O, has precisely two equivalence classes,
it follows that S has precisely two elements. Now S = {a,b} and a A b = a. From
the equations defining semilattices, we also have b A a = a and a A a = a and
b A b = b. These equations imply that a H O, b I+ 1 constitutes an isomorphism
of S with ({O, 1}, A ).
Let D = (D, A , v ) be a subdirectly irreducible distributive lattice. Assume
that it is (a, b)-irreducible. We can assume that a < b (or else replace one of the
two elements by a A b). By Theorem 2.69 (characterizing principal congruences
in distributive lattices) and the (a, b)-irreducibility, whenever c < d, we have
a A c = b A c a n d a v d = b v d. From this we easily conclude that b < x holds
for no x, or else we could take c = b and obtain that a = b. Likewise, x < a holds
for no x. Thus a A x = a and b v x = b for al1 x, and we have that every member
4.5 The Subdirect Representation Theorern 173
of D lies in the interval I [a, b]. Now if a Ix < b, then a A x = b A x, implying

that a = x. We conclude that D = (a, b}. The remainder of the argument is
obvious. ¤
It was many years after the appearance of Birkhoff's paper before the
usefulness of his concepts of subdirect representation and subdirectly irreducible
algebra became fully apparent. They find extensive applications in the study of
varieties, especially the congruence distributive varieties. Lattices, and most of
the algebras that arise in logic, form congruence distributive varieties. The
example of monadic algebras, which we now present in detail, is illustrative of
such applications.
EXAMPLE 3: A monadic algebra is an algebra A = (A, A , v , -,c, O, 1) such

that (A, A , v , -, 0,1) is a Boolean algebra (called the underlying Boolean algebra
of A) and these equations are valid in A:
The 1-ary operation c is called a closure operation. An element of A is called

closed iff a = c(a). A is called Boolean iff c(x) = x for al1 x, and called special iff
O # 1 and
O iff x = O
c(x) =
1 iff x # O.
Monadic algebras are the 1-dimensional case of the a-dimensional cylindric
algebras, which include the algebras of formulas in first-order logic. (See Henkin,
Monk, and Tarski [1971].)
LEMMA 1. Let A be a monadic algebra.

i. The set of closed elements of A is a subuniverse C of the underlying Boolean
algebra of A.
ii. For every x E A, c(x) is the least element u E C satisfying x I u.
iii. If u E C then the formula f (x, y) = (x A u) v (y A u-) defines a decomposition
operation of A.
Proof. To prove (i), we note first that if x and y are closed, then
so x v y is closed. Note that x = x A c(x) Ic(x) for al1 x. If x is closed, then

so c(x-) 5 x-, implying that x- is closed. Now O E C, and C is closed under joins
and complements. Thus C is a subuniverse of the underlying Boolean algebra
of A.
To prove (ii), we note first that c(c(x)) = c(x) for al1 x (i.e., C = c(A)). This
is proved as follows.
We note also that c is a join endomorphism of A-i.e., it commutes with v -and

thus c is an order-preserving mapping. Now if x E A, u E C and x 5 u, then
x 5 c(x) < c(u) = u. The truth of statement (ii)follows easily from these formulas.
The most complex of the equations we wrote as a definition of monadic
algebras assumes-a simpler form, in view of (i) and (ii). We shall need this
statement for proving (iii):
if u is closed, then c(x A u) = c(x) A u.
To prove (iii), let u E C and let f (x, y) be the operation defined in (iii). From
Definition 4.32 and the proof of Theorem 4.33, we know that what we have to
prove is the following: f satisfies the equations (9), and for each of the basic
operations g of A, f and g satisfy the equation (10). This calculation shows that
(10) holds for f and c:
In the third of the four equations we used that u = c(u) and u- = c(u-).
The other equations involved in showing that f(x, y) is a decomposi-
tion operation require nothing more than straightforward deductions from
the equations defining Boolean algebras. (Exercise 4.38(13) required these
deductions.) •
LEMMA 2. For monadic algebras the properties of simplicity,subdirect irreduci-

bility, and direct indecomposability coincide. A monadic algebra is subdirectly
irreducible iff it is special.
Proof. Suppose that a monadic algebra A is directly indecomposable. Then A

has precisely the two trivial decomposition operations, f(x, y) = x and y. By
Lemma 1 (iii),A must have precisely two closed elements, O and 1. Then we must
have c(x) = 1 for al1 x # O. Thus A is special.
We complete the proof by showing that every special algebra is simple. Let
a # O, be any nonzero congruence of a special algebra A. Let (a, b) E a, a # b.
By forming meets with either a- or b-, we obtain an element u = a A b- or
u = b A a- with u # O, ( u , O) E a. 'Then (O, 1) = (c(O), c(u)) E a. For every x E A

we have
where denotes congruence moda. The argument proves that A has only two
congruences 0, and 1,-i.e., that A is simple. ¤
COROLLARY 3. Every monadic algebra can be subdirectly embedded into a

product of simple monadic algebras. Every finite monadic algebra is isomorphic to
a product of special monadic algebras.
EXAMPLE 4: A celebrated theorem of N. Jacobson [1945] states that every

ring obeying an equation xn Ñ x (where n > 1) is commutative and can be
subdirectly embedded into a product of finite fields that obey the equation. The
Subdirect Representation Theorem easily implies that Jacobson's theorem is
equivalent to the assertion that every subdirectly irreducible ring obeying xn Ñ x
(where n > 1) is a finite field. Each finite ring obeying the equation is directly
decomposable as the product of fields. (This follows from Jacobson's theorem
together with a lemma that precedes Theorem 4.76 in $4.7.)
Jacobson's theorem follows from three assertions, two of which we shall
prove. Let n be a fixed integer greater than 1. By a division ring we mean a ring
having a unit element whose nonzero elements form a group under multiplica-
tion. Note that a commutative division ring is a field. (Division rings are some-
times called skew fields.) Here are the assertions. (1) Every subdirectly irreducible
ring obeying xn Ñ x is a division ring. (2) A noncommutative division ring
obeying xn Ñ x contains a finite noncommutative division ring. (3) Every finite
division ring is commutative. If the three assertions are true, then every sub-
directly irreducible ring obeying xn Ñ x is a field. The finiteness of such a field
is due to the fact that the polynomial xn - x can have at most n roots in a
field. The first two assertions are proved in the lemmas and corollary below. The
third is a well-known theorem of Wedderburn [1905], which is proved in most
textbooks on modern algebra above the elementary level. (For example, a proof
of Wedderburn's theorem will be found in Jacobson [1985], p. 453.)
DEFINITION 4.45. By an ideal in a ring (R, +, m, -,O) is meant a subset J
of R satisfying:
l. +
O E J and for al1 x, y E J , the elements x y and - x belong to J . ( J is a
subgroup of (R, + ).)
2. Forallx~Jandy~R,bothx~yandy~xbelongtoJ.
LEMMA 1. For any ring R there is a one-to-one order-preserving correspon-

dence between the ideals of R and the congruences of R. An ideal J and the
corresponding congruence ( satisfy:
i. XEJ++(O,X)E[.
ii. (x,y)~(++x- EJ.
Proof. The correspondence between homomorphisms from a ring and ideals

in the ring is a basic result proved in any undergraduate course in modern
algebra. The correspondence between homomorphisms and congruences in
general algebras has been worked out in $1.4 and throughout this chapter.
Clearly, these two lines of thought must link up; there must be some correspon-
dence between ideals and congruences in a ring. We leave it to the reader to verify
that this is it. ¤
LEMMA 2. Every subdirectly irreducible ring in which xn = x holds for al1 x,

for some integer n > 1, is a division ring.
Proof. Let R be a subdirectly irreducible ring in which xn = x holds for al1 x,

where n is a fixed integer greater than 1.
The set of al1 elements x E R such that x y = y x for al1 y E R is called the
center of R. Elements of the center are also called central elements. An element
e E R is said to be idempotent iff e e = e. We prove a series of statements.
-i. For each x E R, e = xn-l is an idempotent and e x =x e = x.
ii. F o r a l l x , y ~ R , x . y = O i f f y - x = 0 .
The proof of (i) is easy. We have e2 = x ~ = ~XXn-2- -~e and ex = xe = xn.
To prove (ii), suppose that xy = O. Then yx = (yx)" = y (xy)"-l x = 0.
iii. Every idempotent of R is central.
To prove (iii),let e be idempotent and x be any element. Now e (x - ex) = 0.
Hence (by (ii)),xe - exe = (x - ex) e = O. Thus xe = exe, and we can obviously
prove in the same way that ex = exe.
Since R is subdirectly irreducible, it is (a, b)-irreducible for some a # b

in R. (See Definition 4.41, Theorem 4.40.) Taking u = a - b, we easily see that R
is (0,u)-irreducible. Then replacing u by un-l = e, we find that R is (O, e)-
irreducible. (Each of these pairs of elements generates the same congruence.)
Translating this concept frorn congruences to ideals, using the lemma, we have:
iv. e # O, ee = e, and e E J whenever J # {O) is a nonzero ideal.
Now since e lies in the center of R by (iii), the set ann(e) = (x: xe = 0) can
I
be checked to be an ideal of R. Clearly, e&ann(e),so ann(e) = {O) by (iv). In
particular we must have ex - x = O = xe - x for al1 x. Thus
v. e is a two-sided unit element of R.
Let us redenote this element by 1.
We conclude by proving the following statement, showing that x " - ~is a
multiplicative inverse of x for al1 x # O in R, and hence that R is, in fact, a division
ring.
vi. For al1 x # O in R, xn-' = 1.
To prove it, let x # O and f = xn-l. Since f = f, f (1 - f ) = O. Thus ann(1 - f )
1 4.5 The Subdirect Representation Theorern 177
is a nonzero ideal of R (it contains f ). By (iv), 1 (=e) belongs to this ideal. So we

have O = 1 (1 - f ), or equivalently, 1 = f. ¤
COROLLARY 4. Every ring obeying un equation xn Ñ x for un integer n > 1

is a subdirect product of finite fields that obey the equation.
Proof. Every homomorphic image of a ring that obeys the equation must also
obey it. Theorem 4.44 reduces the work to proving that every subdirectly irre-
ducible ring obeying the equation is a finite field. By the last lemma, we can
assume that R is a division ring obeying xn Ñ x, where n is a fixed integer greater
than 1. We wish to prove that R is commutative. (By an earlier remark, if R is
commutative, then it must be a finite field.) Our strategy will be to show that if
R is not commutative, then it contains a finite noncommutative division ring,
which contradicts Wedderburn's theorem. In order to read this proof, the reader
needs to know a modest amount of the elementary theory of fields.
We shall assume that n > 2, for if n = 2, then it is elementary to prove that
R is commutative (see Exercise 12 below).
Observe that every subring S of R that contains a nonzero element is a
division ring, because x " - ~is the multiplicative inverse of x when x # O. Also, if
S is commutative, then S has at most n elements, because every element of S is
a root of the polynomial xn - x. Now let A be a maximal commutative subring
I
i
of R. Thus A is a field and 1 Al I n. Also, for every x E R - A there is an element
a € A such that xa # ax, since A is a maximal commutative subring. We shall
now assume that A # R and derive from this assumption a contradiction.
Since A is a finite field, there is a prime integer p and an integer k 2 1 such
that [Al = pk and pl = O in A. Then it follows that px = O for al1 X E R. The
multiplicative group of the field A is cyclic; thus let a be an element satisfying
A = {O, 1, a', SapkW2}.
, Notice that R is naturally a vector space over A. The
product ux of u E A and x E R is just the product of these elements in the ring R.
Let EndAR denote the ring of linear transformations of this vector space. We
define a mapping D: R -+R, namely, for x E R we put D(x) = xa - ax. Then it is
clear that D E EndAR. Let also I denote the identity map-i.e., the unit element
of the ring EndAR. Now EndAR is naturally a vector space over A, and we have
that the element D commutes with u1 for al1 U E A. Let A[z] be the ring of
polynomials over A in the indeterminate z. Thus there exists a homomorphism
p mapping A[z] onto the commutative subring of EndAR generated by D and
the uI, with p(z) = D and p(u) = uI.
We shall need the following expansions for the powers of D, the truth of
which can easily be verified by induction. If m 2 1 and x E R, then
Taking m =p in the above, and noting that

(P) is divisible by p for O < i < p,
we get that DP(x)= xaP - apx, and then DP'(X)= xap' - ap'x.
Since apk= a, it follows from the above that DP"= D. In other words, D is
a zero of the polynomial f (2) = zpk - z E A[z]. Now f (z) has pk distinct roots in
A (in fact every elenient of A is a root) and so has the factorization
in A[z]. It follows that in End, R we have
Notice that ker D = {x E R: D(x) = O} is precisely A (because A is a maximal

commutative subring and is generated by a). Now let us choose any x E R - A
and define x, = D(x), x, = (D - 1)(x,), x, = (D - al) (x,), and so on, running
through al1 the maps D - ul (u E A). Since x $ A, we have x, # O. Then the
preceding displayed equation implies that there must exist j 2 O such that ( =
xj # O and xj+, = O = (D - uI)(c) for a certain u E A - {O}. This means that
where c E A. Notice that since u # O and # 0, the above equation implies that
D(() # O-i.e., $A. Now every nonzero element of A is of the form w = aj, and
then from the equation above we also have cw = cj(. These facts imply that the
set
T = {g(c):g(z) E A [z] has degree < n}
is a subring of R. Since A U 15) c T, then T is a noncommutative division ring.
But T is obviously finite. This contradicts Wedderburn's theorem.
The concept defined below helps to illuminate al1 of the preceding

examples.
DEFINITION 4.46.
1. Let A be an algebra and K be a cardinal number. We say that A is residually
less than K ifffor each pair of distinct elements a and b in A, there is an algebra
B satisfying JB]< ~cand a homomorphism f : A -,B with f(a) # f(b). We
say that A is residually finite iff for every pair of elements a # b in A there
exists a homomorphism f of A into a finite algebra with f (a) # f (b).
2. A class X of algebras is called residually less than K,or residually finite, iff
every member of X has the same property.
COROLLARY 5. Suppose that X = H(X). Then X is residually less than ~c

$ every subdirectly irreducible member of X ' h a s a cardinality less than K,and X
is residually finite # every subdirectly irreducible member of X is finite.
Proof. We assume that X = H(X)-Le., that every homomorphic image of an

algebra in X belongs to X . Thus by the Subdirect Representation Theorem,
every algebra in X is a subdirect product of subdirectly irreducible algebras
in X . Thus in each of the assertions, the sufficiency of the condition (in order
that X have the residual property) follows from the Subdirect Representation
Theorem. Now for the necessity, suppose first that X is residually less than
K and let A be a subdirectly irreducible member of X . Choose a and b such
that A is (a, b)-irreducible. There exists a homomorphism f of A into an alge-
bra of cardinality less than K satisfying f (a) # f(b). This inequality implies that
ker f = O,, since A is (a, b)-irreducible. Thus f is one-to-one, and it follows that
the cardinality of A is likewise less than K , as desired. The proof of the statement
regarding residual finiteness is just the same. •
DEFINITION 4.47. A class of algebras is residually small iff it is residually less

than some cardinal. A class of algebras is residually large iff it is not residually
small.
The classes of Boolean algebras, semilattices, and distributive lattices are

residually < 3 (Examples 1 and 2 and Corollary 2). The class of rings satisfying
x5 Ñ x is residually < 6, since every field satisfying this identity has just 2, 3, or
5 elements (Example 4 and Corollary 4). The class of monadic algebras is
residually large, since there exist simple monadic algebras having any infinite
cardinality (Example 3 and Lemma 2). The class of finite lattices is not residu-
ally less than any finite cardinal. (This follows by Example (d) after Definition
4.41.) The class of al1 lattices is residually large, for there exist simple lattices of
cardinality larger than any given cardinal number. (See Exercise 4.24(12) for an
example.) The class of al1 groups is residually large, for the same reason. (See
Exercise 4.24(1l).)
One of our main concerns throughout these volumes will be the study of
varieties (classes of similar algebras closed under the formation of homomorphic
images, subalgebras, and direct products). Residually small varieties tend to have
a rather nice structure theory for their members, which residually large varieties
usually do not possess. The class of monadic algebras supplies an example of the
phenomenon. On the face of it, monadic'algebras would appear to be not much
more complicated than Boolean algebras, since every monadic algebra is a
subdirect product of special algebras, which are just Boolean algebras with an
additional, but trivial, operation. Yet it is possible to build any finite graph into
an infinite monadic algebra, in such a way that the graph (its vertices and edges)
can be recovered from the algebra through algebraic definitions. Such construc-
tions are impossible in Boolean algebras, indicating a considerable difference in
the complexity of these two varieties.
1 Exercises 4.48
l. An algebra A is called semi-simple iff it can be subdirectly embedded into
a product of simple algebras. Prove that an algebra is semi-simple iff 0, is
the intersection of a set of maximal members in Con A.
2. Verify that the quasi-cyclic groups Z, (Example (e) following Definition
4.41) are subdirectly irreducible. Determine the structure of their con-
gruence lattices.
3. (D. Monk) An algebra A is called pseudo-simple iff ]Al > 1 and for every
8 < 1, in Con A, A/8 r A. (The quasi-cyclic groups are pseudo-simple.)
Prove that if A is pseudo-simple, then Con A is a well ordered chain.
4. Verify that the lattices N, and N, (Examples (c) and (d) following Defini-
tion 4.41) do have the congruence lattices that were pictured.
5. Represent the three-element lattice as a subdirect product of two two-
element lattices.
6. Show that any linearly ordered set of at least two elements is a directly
indecomposable lattice.
*7. Prove that every subdirectly irreducible Abelian group is cyclic or quasi-
cyclic.
8. Let A be the set of functions from co to (O, 1). Define A = (A, f, g) to be
the bi-unary algebra with
+
f(a)(i) = a(i 1)
d a ) (9 = a(O>.
Show that A is subdirectly irreducible.
9. For each infinite cardinal rc, construct a subdirectly irreducible algebra with
rc unary operations having cardinality 2".
10. Prove that every algebra having at most rc basic operations, al1 of them
unary, is residually 5 2 " (where K is an infinite cardinal).
11. Describe al1 subdirectly irreducible algebras with one unary operation.
(They are al1 countable.)
12. Prove directly (without invoking Corollary 4) that a ring obeying the
equation x 2 Ñ x obeys also x + x Ñ O and xy Ñ yx. (This class of rings is
discussed in $4.12.)
*13. Prove directly that a ring obeying x3 Ñ x obeys also xy Ñ yx.
14. Let A be the algebra ((O, 1,2,3), o) in which a o b = a if 1 I a, b 5 3 and
la - bl < 1, and al1 other products are 0. Show that A is simple. Show that
there exist simple homomorphic images of subdirect powers of A having
any prescribed cardinality 2 4. (The variety generated by A is not residually
small.)
15. Let S be the semigroup ( (O, 1,2}, o) in which 2 0 2 = 1 and al1 other pro-
ducts are 0. Show that there exist subdirectly irreducible homomorphic
images of subdirect powers of S having any prescribed cardinality 2 3.
16. Describe al1 simple commutative semigroups and prove that they are al1
finite.
17. Prove that no commutative semigroup has the congruence lattice pictured
below.
4.6 Algebraic Lattices 181
18. A Heyting algebra is an algebra (H, A , v , -+,O,1) such that (H, A , v ) is

a lattice, O and 1 are the least and largest elements of this lattice, and this
condition holds: x -+ y 2 z iff z A x I y for al1 x,y, and z.
i. Prove that the lattice in a Heyting algebra is distributive.
ii. Prove that Heyting algebras can be defined by a finite set of equations.
iii. Prove that a Heyting algebra is subdirectly irreducible iff it has a largest
element b < 1. (Show that for each a E H,
is a congruence.)
iv. Prove that the only simple Heyting algebra is the two-element one.
4.6 ALGEBRAIC LATTICES
The class of algebraic lattices derives its name from the fact that these are
precisely the lattices normally encountered in algebra. A lattice is algebraic if and
only if it is isomorphic to the subuniverse lattice of some algebra. It is a much
deeper fact that a lattice is algebraic if and only if it is isomorphic to the
congruence lattice of some algebra. We prove the easier fact in this section. We
begin by reviewing the definition of algebraic lattices, from $2.2. An element c of
a lattice L is said to be compact iff for al1 X S L, ifVX exists and c 5 V X , then
cI VX' for some finite set X' c X. A lattice L is said to be algebraic iff L is
complete and every element of L is the join of a set of compact elements of L.
The congruence lattice of any algebra is an algebraic lattice. In 1963, G.
Gratzer and E. T. Schmidt proved that the converse is also true: Every algebraic
lattice is isomorphic to the congruence lattice of a suitably constructed algebra.
A proof of the Gratzer-Schmidt Theorem along the lines of the original argument
will be found in Gratzer's book [1979] (Chapter 2). A streamlined proof has been
given by P. Pudlák [19'76]. Beyond the Gratzer-Schmidt Theorem, the natural
question of which algebraic lattices may be represented as congruence lattices in
specific classes of algebras has produced some deep results, although many hard
and interesting problems remain unsolved. (We discuss this subject thoroughly
in a later volume.) We remark that although every finite lattice is algebraic, it
remains unknown whether every finite lattice is isomorphic to the congruence
lattice of a finite algebra. (This question about the congruence lattices of finite
algebras has been discussed before, just prior to Definition 4.21.)
In this section we introduce a number of interesting facts about algebraic
lattices, whose proofs do not require a lot of effort or sophisticated machinery.
An important class of modular algebraic lattices will be examined in $4.8.
Our first lemma supplies some basic properties of algebraic lattices that are
often useful. Recall that for elements u and v in a lattice L, I[u, v] designates the
interval {x: u i x I u) and IEu, v] designates the sublattice of L having the
interval as its universe. We say that v covers u, written u -< v, iff I [ u , v] is a

two-element set.
LEMMA 4.49. Let L be any algebraic lattice.

i. If u and c are elements of L such that u < c and c is compact, then there exists
an element m such that u < m -< c.
ii. If u < v in L, then the interval sublattice I[u, v] is un algebraic lattice.
iii. If x and y are elements of L such that x < y, then there exist a, b E L satisfy-
ingx < a < b I .y.
Proof. To prove (i), we use Zorn's Lemma. The set ( x : u 2 x < c } is closed
under the formation of joins of chains, since c is compact. Therefore it has a
maximal element m. Such an element will be covered by c.
To prove (ii), note that I[u,vJ is complete, since L is, and that for every
compact element c < v of L, u v c satisfies the condition to be a compact element
of I[u, u]. Then it is clear that every element of I[u, u] is a join of compact elements
of I[u, u].
To prove (iii), apply (ii) and (i). The lattice I [ x , y] must have a compact
element b > x. By (i), this lattice has an elemeilt a covered by b. w
There is a close connection between algebraic lattices and algebraic closure

operators. In fact (Theorem 2.16), a lattice is algebraic iff it is isomorphic to the
lattice of closed sets for some algebraic closure operator. The next easy lemma
gives us an opportunity to recall the definition of an algebraic closure operator.
LEMMA 4.50. Let A be any algebra. sgA is un algebraic closure operator whose
lattice of closed sets is identical with Sub A, the lattice of subuniverses of A.
Proof. According to Definition 1.8, sgA(x)is the smallest subuniverse of A

containing the set X. Since a set X c A is a subuniverse of A iff X = sgA(x),the
subuniverses of A are precisely the closed sets for sgA. That C = sgAis an
algebraic closure operator is equivalent to the satisfaction of the following
properties (by Definitions 2.10 and 2.13):
(12i) X c C(X) for al1 X E A;
(12ii) C(C(X)) = C(X) for al1 X E A;
(12iii) if X E Y,then C(X) c C(Y); and
(12iv) C(X) = U(C(Z): Z c X and Z is finite} for al1 X c A,
The verifications are entirely straightforward. (This result is the same as Corol-
lary 1-11.) ¤
THEOREM 4.51. If C is un algebraic closure operator on a set A then there is

un algebra A = ( A , . ) such that C = sgA.
Proof. Let C be an algebraic closure operator on A. We define

-
a: for some n 2 0,
T={
ü = (a,;..,a,-,,a,,) and a , ~ C ( { a ~ , - - - , a , - ~ ) )
We endow A with an operation ji corresponding to each ü = (a,, a * , a,,) in T.
fE is the n-ary operation defined by
a, if xi = 'ai for al1 i < n,
f a ( ~ o , . . . , ~ n -=i )
x, otherwise
Let A = (A,&(ÜE T)). Note that A has O-ary operations iff C(@) # @.
Now let X c A. To prove that sgA(x)c C(X), we recall that X c C(X)
and we show that C(X) is a subuniverse of A. Suppose that ü = (a,, a,) E T, a ,
and that {b,, ,b,-,) G C(X). If ai = bi for al1 i i n, then C({a,, - . a,-, 1) c e,
C(C(X)) = C(X), implying that fa(b,, b,-,) = a, E C(X). In the remaining

a ,
case, f,(b,, b,-,) = b,, which certainly belongs to C(X). This argument shows
m.,
that C(X) is a subuniverse of A.

To prove that C(X) E sgA(x),we use the fact that s$(x) is closed under
the operations of A, and the fact that C satisfies (12iv).Thus it will suffice to show
that C(Z) c sgA(x)for every finite subset Z c X . Let Z be such a finite set. Say
Z = (c,, . ,ck-,), and let c be any element of C(Z). Clearly, c = fc(c0,- . ck-,)
- m S ,
where E = (c,, - ,ck-, ,c), implying that c E sgA(x).That concludes the proof.
M
COROLLARY. (Birkhoff and Frink [1948].) For every algebraic lattice L

there exists un algebra A with L Sub A.
Proof. It is immediate from Theorems 2.16 and 4.51. ¤
If the operations of an algebra are of bounded rank (or arity), what, if

anything, can be inferred about its lattice of subuniverses? This question leads
immediately to the definition of an n-ary closure operator, and to severa1 results.
DEFINITION 4.52. A closure operator C on a set A is of rank n + 1, or n-ary,

(where n 2 O) iff for every X G A, if C(Z) 5 X for every Z c X with 1 ZI 4 n, then
C(X) = X.
THEOREM 4.53. Let C be a closure operator on a set A. C is of rank n + 1 iff

there exists un algebra A such that C = sgA
and euery basic operation of A is of
an arity 5 n.
Proof. It is quite easy to see that sgAis of rank n 1, where A is any algebra +
whose operations are of arity I n ; we leave this verification to the reader. On
the other hand, suppose that C is a closure operator of rank n + 1 on a set A.
Let A be the algebra corresponding to C constructed in the proof of Theorem
4.51. Let A' be the reduct of A obtained by removing al1 of the operations of arity
greater than n. We can now follow through the proof of Theorem 4.51, using the
fact that C has rank n + 1, which implies it is algebraic, to show that C = sgA'.
m
Although being of rank k is an interesting property of an algebraic closure

operator, it is not reflected in any lattice-theoretical property of the lattice of
closed sets for the closure operator (unless k < 2). That is the content of the next
theorem.
THEOREM 4.54. Every algebraic lattice is isomorphic to the lattice of sub-

universes of an algebra whose basic operations have arity at most 2.
Proof. Let L be any algebraic lattice, let A be the set of compact elements of L,
and let C be the closure operator on A defined i n the proof of Theorem 2.16.
Thus for any X c A,
C(X) = (EA: c 5 VX}.
We know that L is isomorphic to the lattice of C-closed sets. We shall show that
C is of rank 3, so that this theorem follows by invoking Theorem 4.53.
To do so, let X be a subset of A that includes the C-closure of any subset
of two or less of its elements. We are to show that X = C(X). Let b = VX. Let
c be any element of A such that c < b. We must show that c E X. Now it is obvious
that the join of any two compact elements is compact. Thus if u, V EA, then
u v v E C((U,U}).From this and our assumption about X, it follows that the join
of any two, and hence of any finite number of, elements of X belongs to X. (The
join of the empty set must be considered separately, but it is trivial to do so.)
Since c is compact, we have c < u for an element u that is the join of a finite
subset of X (i.e., u E X). Then c E C((u}), so it follows that c E X. ¤
Although there is nothing special about the subuniverse lattices of algebras

that possess no n-ary operations for n 2 3, the subuniverse lattices of algebras
having only unary and nullary operations are quite special indeed. In such a
lattice, the join and meet of any subset coincide with the union and intersection
of the subset; consequently, the lattice is distributive. (In fact, it must satisfy
every version of the distributive law formulated with infinite joins and meets.)
THEOREM 4.55. Let L be any algebraic lattice. Al1 of the following statements
are mutually equivalent.
i. L is isomorphic to the lattice of closed sets for a closure operator of rank 2.
ii. L is isomorphic to the lattice of subuniverses of un algebra whose operations
are unary and nullary.
iii. y v (/\x)= /\{y v x : x ~ X f}o r a l l y ~ L a n d XG L.
iv. L is a distributive lattice, and every element of L is the join of a set of strictly
join irreducible elements.
-
Proof. Theorem 4.53 supplies the fact that (i) (ii). As we mentioned, in an
algebra with only unary and nullary operations, the joins in the subuniverse
lattice are identical with set unions; thus (iii) holds for such a lattice. Therefore,
(ii) (iii).
Now suppose that L is an algebraic lattice satisfying (iii).Taking X equal to
an arbitrary two-element set in (iii),we see that L is distributive. To conclude that
(iv) holds for L, we need only show that every compact element of L is the join
of a set of strictly join irreducible elements of L. So let c be a compact element
greater than O. (Note that O is the join of the empty set of strictly join irreducible
elements.) Our desired conclusion is equivalent to the following: For every x < c,
there is a strictly join irreducible element u with u < c and u x. So suppose
that b < c. By Lemma 4.49(i), we can choose an element d such that b I d < c.
Now take
T=(x:x<candx$d) and u = l \ T .
Notice that by the choice of d, for any element x in L, c = d v x is equivalent to
x E T. By (iii), it now follows that u E T, and u is the least member of T. Thus if
x < u, then x I d, so we have that u is strictly join irreducible. Since u $ d, then
also u $ b, which is precisely the conclusion we are after. This finishes the proof
that (iii) * (iv).
Finally, suppose that L satisfies (iv). To prove that L also satisfies (i), let A
be the set of strictly join irreducible elements of L. For X A put
The proof that C is a closure operator, and that L is isomorphic to the associated
closed set lattice, follows the lines of the argument for Theorem 2.16. The reader
is asked to supply the details of this argument. It remains now to prove that C
has rank 2. (This turns out to be equivalent to the assertion that every strictly
join irreducible element of L is strictly join prime.) Here is the argument.
Let X be a subset of A such that X contains the C-closure of each of its
subsets of at most one element. It is to be shown that C(X) = X. So let u E C(X)-
i.e., u E A and u I V X . Since u is a join of compact elements and is strictly join
irreducible, u must be compact. Thus there exists {x,, ,xm_,) E X with u 5
x, v x, v v x,-, . Since L is distributive, we have
Then the join-irreducibility of u implies that for some i < m we have u = u A xi.
Thus u E C((xi)), implying that u E X, as desired. •
If an algebra has at most K basic operations (where K is a cardinal number),

what inferences can be drawn about its lattice of subuniverses? The next lemma
supplies one such inference, and Theorem 4.57 and Exercise 5 below will demon-
strate that the inference in the lemma is essentially the only one possible.
LEMMA 4.56. Let B be un algebra such that the number of basic operations of
B is at most K (where K is an inJinite cardinal number).
i. If X G B, then 1sgB(x)[5 K if 1x15 K ; and 1sgB(x)l= 1x1if 1x12 K .

ii. Every compact member of SubB contains at most K compact members of
Sub B.
Proof. Let X c B. By Theorem 1.9, sgB(x) is identical with the union

U{Xn:n < w } where X, = X and X,,, is the union of X, and sets f(X,"),
taken over al1 basic operations f of B. Let 3L denote the larger of the two cardinals
K and (X 1. If 1 XnI 5 A, then Xn+,is the union of at most 3L sets, each of cardinality
at most 2, and so IXn+,J1 A. Inductively, we have that JXnJ1 ;1 for al1 n < o.
From this it follows that sgB(x)= U{Xn: n < w ) has cardinality at most
A -o = A. This conclusion is precisely what is asserted in the first statement of
the lemma.
Now suppose that S is a compact element of SubB--i.e., S = sgB(x) for
some finite subset X of B. From (i), the cardinality of S is at most K. Thus the
set of finite subsets of S has cardinality at most K. Since every compact element
of Sub B contained in S is generated by one of these finite subsets, we can conclude
that this set of compact elements has its cardinality bounded by K. That proves
statement (ii). ¤
THEOREM 4.57. Let E be un algebraic lattice, having at least two elements, in

which every compact element has only countably many compact elements below it.
There exists un algebra B = (B, ) with one binary operation such that L r Sub B.
Proof. This proof is a refinement of our proof of Theorem 4.54. We take for the
universe of our algebra the set
B = (c EL: c is compact and c # 0).
Just as before, we have that L is isomorphic to the closed set system
under the lattice operations determined by the set-inclusion ordering of F. Our

goal is to construct a binary operation on B for which we will have F =
Sub(B, m ) .
We use our special hypothesis about L in this way. There exists a function
a: B -+ B" such that for every b E B, o(b) is a function from the set of natural
numbers onto the set {c E B: c I b}. (Here we have used the axiom of choice twice,
once to conclude that for every b E B there exists a function from w onto the set
of nonzero compact elements 5 b, and again to get a function that chooses, for
each b, one of these functions.) We denote a(b)(n) by o(b, n). We can further
assume that for each b E B
(13) a(b, O) = b; o(b) is one-to-one if its range is infinite; and if the range
of o(b) is finite, then there exists k < o such that i < j 5 k implies
o(b, i) # a(b,j), and k < i < co implies ~ ( bi), = a(b, k).
We now define, for any elements b, c E B,
ifc$b
b.c =
a(b, n + 1) if c = a(b, n). '
The condition (13) implies that b c is well defined, and we have constructed a
binary operation, on B.
S ,
Since b c < b v c always holds, it is clear that every S E F is a subuniverse

of the algebra B = (B, ). Now let, conversely, S E Sub(B,. ); put u = V S . We
wish to show that S EF, or what amounts to the same thing, that c E B, c 5 u
imply CES.So let c be a nonzero compact element of L with c 5 u. Note that
the two clauses of the definition of ensure that S is closed under finite joins.
Therefore c, being compact, is I some member b of S. By (13) and our definition,
we have
o(b,O) = b
Since there exists n with c = o(b, n), it clearly follows that c E S as desired. This
concludes the proof. ¤
Recall that vector spaces have the property that al1 minimal generating sets
have the same cardinality, equal to the dimension of the space. Unary algebras
have this property in common with vector spaces (see Exercise 7 below); but it
is rarely satisfied by other kinds of algebras. The Abelian group Z,,for example,
has minimal generating sets (1) and (2,3). We next present two general results
about the minimal generating sets of algebras. One is phrased as a subtle property
of closure operators of rank n, and the other has a purely lattice-theoretical
component.
DEFINITION 4.58. Suppose that C is a closure operator on a set A. If
B S X c A then B is called a base (or generating set) of X with respect to C iff
X = C(B). B is called an irredundant base (or basis) of X with respect to C iff
X = C(B) and for every set B' properly contained in B, X # C(Bf).(In other
words, B is a minimal generating set of X.) By B'(x) (or B*(x) if C = sgA for
an algebra A) we denote the set
{n E u:X has a basis of n elements).
We will say that a set K of natural numbers is k-convex, where k is a natural
+
number, iff for e;ery a, b E K such that a k < b, there is c E K with a < c < b.
Thus K is O-convex iff K has at most one member; K is 1-convex iff K is an
interval in the set of natural numbers (i.e., K is convex in the customary sense).
Here is our first result on irredundant bases.
TARSKI'S INTERPOLATION THEOREM (4.59). If C is a closure operator
of rank k + 2 on a set A, where k is a natural number, then BC(X)is k-convex for
al1 X E A.
To prove Tarski's Theorem, we require a definition and two lemmas.
DEFINITION 4.60. Let C be a closure operator on A and k be a positive

integer. Let C,,, be the function defined on subsets of A by
C,,(X) =X U ( U {C(Z):Z c X, 121 < k}).
Let C;,,(X) = X, C;~;'(X) = CU)(Cyk)(X)),
and Ck(X)= U {Cyk,(X):1 S n < o}.
LEMMA l. If C is a closure operator on A, then for each k 2 1, C, is a closure
operator of rank k on A, C , ( X ) E C(X) for al1 X, and C = C, iff C has rank k.
Proof. This is a lengthy but straightforward sequence of verifications, based on

Definitions 4.52 and 4.60. We make it an exercise. ¤
LEMMA 2. Let U be any set, k be any natural number, and f : U -+o .

{ f (u): u E U} is k-convex iff there is a function p: U x U -+ o such that for every
u, v E U with f (u) < f (u), there is z E U such that
i. f(z) r f(u)+ kand
ii. p (u, z) < p(v, u).
Proof. Define p so that p(u, u) = 1 f(v) - f(u)l. The z required above exists,
since { f (u): u E U) is k-convex.
e Suppose f (u) + k < f (v). Take x E U such that p(v, x) is as small as
possible subject to the condition f (x) < f (v). Now pick z E U using the condition
above, so that f (z) I f (x) + k and p(v, z) < p(v, x). By the minimality of p(v, x),
it follows that f (v) 5 f (z). So f (u) If (x) + k. Since f (u) + k < f (v), we conclude
that
and therefore that { f (u): u E U) is k-convex. ¤
The function p introduced in this lemma is something like a distance

function telling how far apart u and v are.
Proof of Theorem 4.59. Let C be a closure operator of rank k + 2 on A and let

X be a subset of A. In order to use Lemma 2, we take U = {B: B is a finite basis
of X} and define f (B) = 1 BI for al1 B E U. Thus BC(X)= { f (B): B E U}, and al1 we
need do is find a function p to satisfy the lemma.
Consider a basis B of X and an element x EX. Since B is a generating set for
X, by Lemma 1 we can pick n as small as possible subject to the condition
XEC;,+,,(B). n is a measure of how many steps it takes to get from B to x. In
each step, at most k + 1 elements are combined. Thus x is really in the closure
of a subset of B with at most (k + 1)" elements. We define
where n is the least natural .number such that x E C&+2)(B).

Now let B and D be
any pair of finite irredundant bases of X. Define
To verify that this p has the required properties, let B and D be members of U
such that 1 D 1 < 1 BI. Since B is a minimal generating set, we cannot have D c B.
+
Pick d E D - B. Let (k 2)"" = n(B, d) and pick E c C?,+,,(B) such that d E C(E)
+
and / E (< k 2. Now (D - {d}) U E is a generating set of X and is finite. Let
D' c (D - (d}) U E such that D' is a basis of X. Then 1 D'I 5 1 D 1 + k and
Therefore al1 the conditions of Lemma 2 are fulfilled; BC(X) is k-convex, as

claimed. m
Tarski proved Theorem 4.59 in the course of a systematic study of bases of

equational theories. Equational theories are defined in $4.11and studied at length
in Volume 2, where we shall find that they are the closed sets for a closure
operator of rank 3.
Our next theorem has a special relevance in the study of equational theories
of lattices. It has been conjectured that the equational theory of a finite lattice
(i.e., the set of al1 equations that are true in the lattice) can cover only a finite
number of equational theories of lattices. To prove or disprove this conjecture
is one of the hardest unsolved problems in contemporary lattice theory.
THEOREM 4.61. Let L be the lattice of closed sets for un algebraic closure
operator C on A, and suppose that L is distributive. For a Jinitely generated closed
set X and any positive integer n the following are equivalent:
i. For some k 2 n, k E BC(X).
ii. In L the element X covers at least n elements.
Proof. To prove that (i) * (ii), we assume that B is a k-element basis of X and
show that X covers at least k closed sets. Let B = (b,, b,_,}. For each i < k
S ,
choose, by Lemma 4.49(i),a closed set Xi with C({bj:j # i}) c Xi and Xi covered
by X. Clearly, bj E Xi iff j # i (for al1 i,j < k). Thus X,, Xbl are distinct closed
S ,
sets covered by X.
To prove that (ii) (i), let us assume that X,, - ., Xn-, are distinct closed
sets covered by X. For each i < n let
x = n{xj:j<nandj# i}.
Since L is distributive, we have
(One can prove this easily for al1 n and for any n elements of L by induction on
n.) The above equation implies that the Y,join to X; i.e.,
X = C(Y) where Y= U (&: i < n).
Since C is algebraic and X is finitely generated, Y must contain a finite basis B
for X. Notice that there is a function f : B --+ (O, - ,n - 1) with b E q(H for al1
b E B. If 1 BI < n then f fails to be onto, and so there is an io < n (not in the range
of f ) such that
B G U ( & : i # io) c Xio.
This is impossible, as C(B) = X > Xio;therefore f is onto, and B has at least n
elements. •
The operator cgA of congruence generation, for an algebra A, is a closure

operator of rank 3. By Theorem 4.59, a finitely generated congruence has, for
the set of cardinalities of its finite irredundant bases, an interval of natural
numbers. If Con A is a distributive lattice and 13is a finitely generated congruence
that covers precisely k < co congruences, then by Theorem 4.61 this interval is
of the form [k', k] for some O 5 k' 5 k.
We discussed at the beginning of this section the Gratzer-Schmidt Theorem,
according to which every algebraic lattice is isomorphic to the congruence lattice
of some algebra. It is also true (as proved in Theorem 2.18) that every lattice can
be embedded into an algebraic lattice. Putting these two results together, we can
deduce a result discovered by P. M. Whitman [1946]: Every lattice can be
embedded into the lattice of al1 equivalence relations on a set. (Of course,
Whitman did not arrive at the result in this way.) B. Jónsson [1953] found a
streamlined proof of Whitman's result, which we shall now present. The proof
requires two lemmas.
By a meet homomorphism of L into L', where L and L' are lattices, we shall
always mean a meet-preserving mapping from Linto L'. Eqv Y denotes the lattice
of al1 equivalence relations on a set Y.
LEMMA 1. Suppose that L is a lattice and that f is a rneet hornomorphisrn of L

into Eqv Y. Let a, b E L and (x, y) E f (a v b). There exist Y' 3 Y and a rneet
homornorphisrn f ' of L into Eqv Y' satisfying:
i. For atE u E L, f (u) = f '(u) f l Y 2 .
ii. (x, y ) E f '(a) O f '(b) o f '(a) o f '(b) (a subset of f '(a) v f '(b)).
iii. \ Y f - Y1 = 3.
Proof. Let x,, x,, X, be three distinct elements outside of Y; set Y' = YU
(x,, x,, x,). For u EL, define f '(u) to be the smallest equivalence relation of Y'
containing f (u) on Y and such that:
Figure 4.4
I
(See Figure 4.4.) We leave it to the reader to verify that this works. It helps to
show that for t E Y and u E L, the following hold: ( t , x , ) E f '(u) iff a I u and
) av bI
( t , x ) E f ( U ) ; ( t , x 2 ) E f ' ( U iff u and ( t , x ) E f (u);( t , x , ) E f '(u)iff b 5 u
and ( 4 Y )Ef(U).
8.
LEMMA
- 2. Let f be a meet homomorphism of L into Eqv Y. There exists a set
Y 2 Y and a meet homomorphismf of L into Eqv Y satisfying:
i. For al1 u E L ,f(u) í l Y 2 = f (U).
ii. For al1 u, V E L f(u)
, v f(v) 3 f ( u v U).
iii. The cardinality of I is no greater than max((Ll,IY / ,o).
Proof. Let A be the cardinal number defined in (iii)-i.e., the least infinite
cardinal not less than 1 LI and not less than 1 YI. Let T be the set of al1 quadruples
( a , b , x , y ) with a, EL, X , ~ E Yand
, ( x , y ) ~ f ( va b). We have that [TI < A .
Therefore we can choose a function from A onto T, and write
i T = (tu:a < A).

We now put Yo = Y,fo = f and define by transfinite recursion a sequence
of sets Y, (a < A) and meet homomorphisms fa: L -+ Eqv Y, (a < A). These map-
pings will satisfy
i'. If a I a' < A, then Y, S Y,. and for al1 u E L ,f,(u) = & ( U ) íl K2.
ii'. If a < A and 5, = ( a , b, x, y ) , then ( x , y ) E fa+l(a)v fa+,@).
iii'. If a < A, then 1 Y,I I A.
For the recursive definition, suppose that P < A and that Y, and fa have been
defined for al1 a < P such that (ir)-(iii') are satisfied for al1 a 2 a' < P. If P is a
limit ordinal (i.e., has no immediate predecessor), then we define
If p =y + 1, then we define % and fp by Lemma 1, taking (a, b, x, y) = t, and

(f,,Y,) for the (f, Y) of that lemma. It is easy to verify that in either case,
(ir)-(iii') are now satisfied for al1 a 5 a' < B + 1. Thus the recursive definition is
complete
Finally, we take
Y = U Y, and ?(u) = U fa(u).
a<A a<A
Now (if)-(iii') ensure that (i) and (iii) are satisfied. (Yhas cardinality at most A
since it is the union of A sets, each having cardinality at most 2.) For every a,
b E Land pair (x, y) E f (a v b), there is an a < A for which we have
Thus (ii) is satisfied likewise.
Now we can prove the theorem.
THEOREM 4.62. Let L be any lattice. There exists a set Z, whose cardinality
is the greater of ILI and o , such that L is isornorphic to a sublattice of Eqv Z, the
lattice of al1 equivalence relations on Z.
Proof. The idea is first to construct a one-to-one meet homomorphism (i.e., a

meet embedding) of L into a lattice of equivalence relations. Starting with such
a function, we can apply Lemma 2 denumerably many times and construct a
chain of sets and meet homomorphisms whose union will be the required lattice
embedding.
We begin in this way. Put Z, = L and define g,: L -+ Eqv Z, by
With very little thought, it should be clear that g, is a one-to-one meet homo-
morphism of L into Eqv 2,.
We obtain Z1 = and g1 = from 2, and g, via Lemma 2. In the
same way, we obtain Z, and g, from Z, and g,, and we continue in this fashion,
generating Zn and gn for al1 n < o . Then of course we put
Now we have IZO(= ILI, and by(iii)in Lemma 2, IZn+,)5 max(lZnI,ILI,o). Thus
it follows that every Zn, and Z itself, has cardinality at most max(1LI, o).
The construction ensures that g is one-to-one; in fact g(u) n2,2 = gO(u)for

al1 u E L. g is, of course, a meet homomorphism. It is also a join homomorphism
(and hence is a lattice embedding). To see this, suppose that a, b E L. Since g is a
meet homomorphism, it is order preserving, so g(a) v g(b) <_ g(a v b). On the
other hand, if (x, y) E g(a v b), then for some n, (x, y) E g,(a v b). Then (x, y) E
gn+i(a) v g,,, (b) by construction (Lemma 2), and hence (x, y) E g(a) v g(b). The
proof is complete.
For a finite lattice L, the proof just concluded produces an injinite set Z.
For a long time, it was not known whether a representation must exist with Z
finite. We now know that every finite lattice is, indeed, isomorphic to a sublattice
of Eqv Z for some finite Z. The only known proof of this result (by Pudlák and
TUma [1980]) is quite difficult.
We conclude this section by proving that every lattice is embeddable into
the lattice of subgroups of a group. This result implies that there are no equations
true in al1 lattices of subgroups except equations true in al1 lattices. Contrast this
with the fact that the congruence lattice of any group (isomorphic to the lattice
of normal subgroups) is a modular lattice. Not even al1 modular lattices can be
embedded into congruence lattices of groups, for these congruence lattices are
known to obey an equation (the Arguesian equation) that is stronger than the
modular law.
Let L be any lattice. Let Z be a set such that L is isomorphic to a sublattice
of Eqv Z. The proof of the next theorem will represent L as a lattice of subgroups
of Sym Z, the group of al1 permutations of Z. The support of a permutation o of
Z is the set supp a = (x E Z: o(x) # x}. A permutation is said to be of finite
support iff its support is a finite set. The set of al1 permutations of finite support
on Z is a subgroup of Syrn 2. In the proof of the theorem, we work within this
group.
THEOREM 4.63. Let L be any lattice. There exists a set Z, whose cardinality
is the greater of 1 LI and w, such that L is isomorphic to a sublattice of the lattice
of subgroups of Sym Z.
Proof. Let Z be a nonvoid set. We shall show that Eqv Z can be embedded into
the lattice of subgroups of Sym Z, and thus obtain Theorem 4.63 as a corollary
of Theorem 4.62.
Correlated with each equivalence relation a on Z there is a subgroup f(a)
of Sym 2, defined as follows.
f(a) = j o ~ S y m Z IsuppoI
: < w and ( V z ~ Z ) ( ( z , a ( z ) ) ~ a ) } .
It is clear that we have defined a one-to-one mapping of Eqv Z into Sub(Sym 2).
Note that if a and fi are two equivalence relations on Z, and if a # P and, say,
(x, y) E p - a, then (xy) belongs to f(P) - f (a) where (xy) is the permutation (or
transposition) o such that a(x) = y, o(y) = x and o(z) = z if x # z # y.
Now let a, p E Eqv Z. Since it is obvious that f (a í l P) = f (a) n f (P), we shall
concentrate on showing that f(a v P) = f(a) v f(/?).It is clear from the defini-
tion tl-iat f(a) v f(P) c f(a v P). To get the other inclusion, suppose that o~
f(a v p). Then o restricted to its support set is a permutation of a fiiiite set. o
can be written as the composition of a finite number of disjoint finite cyclic
perinutations. (See the discussion of permutations in $3.2.) Each one of these
cyclic permutations, say (z,z2 - z,-,) (where zi+, = o(zi)and 2, = o(z,-,)), can
be written as a composition of transpositions. For example,
where each transposition (zOzi)belongs to f ( a v P). (The elements z,, - ,z,-,

are equivalent modulo a v P.) If we can show that every transposition in f (a v P)
belongs to f (a) v f (P), then the proof will be finished.
So let z = (xy) be any transposition belonging to f(a v P). Then x G y
(moda v 0).Tlius there exists a finite sequence x,, x,, x, where x = x,,
e..,
y = x,, and for every i < k either (xi, xi+l) E a or (xi, xi+,) E p. The elements
x,, - , x , can be assumed to be distinct. Now observe that
In this formula, every transposition occurring in the product belongs either to

f (a) or to ,f(P), and so z belongs to f(a) v f (P).
Exercises 4.64
l. Let n be a positive integer. Construct an n + 1-element algebra A, for which
the closure operator sgAnfails to be of rank n. (See Definition 4.52 and
Theorem 4.53.)
2. Prove that if a lattice L = (L, A , v ) is algebraic and satisfies the equivalent
conditions of Theorem 4.55, then the dual lattice L' = (L, v , A ) is algebraic
and also satisfies t hese conditions.
3. Using the result of the last exercise, prove that this staternent is equivalent
to each of the statements in Theorem 4.55, for an algebraic lattice: L is
distributive and the dual of L is algebraic.
4. By a retract of an algebra A we mean any subalgebra B S A for which there
exists a homomorphism f : A -H B such that f , 1 is the identity map on B.
The retracts of algebraic lattices constitute the class of continuous lattices, on
which there exists an extensive literature. Show that if K is a retract of an
algebraic lattice L, then K is join continuous; that is, for every x E K and
up-directed subset Y c K we have the equality
X A (VY) = V ( x A y: y € Y}.
5. Prove this generalization of Theorem 4.57: Let K be any infinite cardinal
number. Let L be any algebraic lattice with at least two elehents such that
every compact member of L has at most K elements below it. L is isornorphic
to the subuniverse lattice of an algebra with rc basic operations.
6. Prove Lemma 1, used in the proof of Theorem 4.59.
4.7 Permuting Congruentes 195
7. Theorem 4.59 implies that if a unary algebra has any irredundant base, then
al1 of its irredundant bases are of equal cardinality. Find a direct proof for
this fact that avoids invoking the theorem.
8. Let C be an algebraic closure operator on a set X and suppose that X =
C ( Y )= C(Z) with Z finite. Then show that Y contains an irredundant base
of X, and that every irredundant base of X is finite.
9. Show that the unary algebra (co,pd) with pd(x + 1) = x for x > O and
pd(0) = O has no irredundant base.
10. Starting from the result in Exercise 19 of 54.4, show that if L Con A where
A has finitely many basic operations, then there exists an algebra B with two
basic operations (unary and binary) such that L Con B.
4.7 PERMUTING CONGRUENCES
We shall say that an algebra A has permuting congruences, or that the congruences
of A permute, iff a o /3 = /3 o a for every two congruences a and P of A, where a 0 /3
is the product of relations defined in 4.28. The congruences of A permute if A is
a group, a ring, a module, or more generally if A has the operations of a group
I
among its term (or polynomial) operations. In each of these cases,
(14) A has a term operation t(x, y, z) obeying the laws
t(x7 X, Y) Y, t(x, Y, Y) 2:-
(For example, we can let t(x, y, z ) = xy-'z if A is a group.) When an operation
t(x, y, z) on a set A obeys the above laws and preserves the equivalence relations
a and B, then a and p permute. We have seen this fact before while proving
Theorem 2.24; let us review the proof. Assume that (a, b) ~a 0 P, so that there
exists an element c with (a, c) ~a and (c, b) EP. (See Figure 4.5.) Then we
have a = t(a, b, b) r t(a, c, b) (mod B), and b = t(a, a, b) I t(a, c, b) (moda). Thus
the element d = t(a, c, b) satisfies (a, d ) E P, (d, b) E a, demonstrating that
(a, b ) ~ p o a .
Figure 4.5
We remark that any relatively complemented lattice of at least two elements

has permuting congruences but does not satisfy (14). (See Exercises 2 and 12
below.) Considered over an entire variety, however, (14) turns out to be a
necessary and sufficient condition for al1 congruences to permute. A variety has
the property that al1 its algebras have permuting congruences iff there is a single
term defining in every algebra of the variety an operation that fulfills (14). This
fact is contained in Theorem 4.141 of $4.12.
It is quite a strong property of an algebra to have permuting congruences.

In this section, we examine some consequences of the property. The most impor-
tant consequence is that the congruence lattice of the algebra is a modular lattice.
An algebra whose congruencelattice is modular will be called congruence
modular. The congruence modular algebras include groups, rings, modules, other
algebras whose congruences permute, and, of course, every algebra whose con-
gruence lattice is distributive. The principal examples of congruence distributive
algebras are lattices. Included also are algebras-such as monadic algebras,
Boolean algebras, and Heyting algebras-that have the operations of a lattice
among their term operations.
Many of the deepest results in universal algebra involve algebras that are
either congruence distributive or whose congruences permute. But outside olf
these two important subclasses of the congruence modular algebras, relatively
little was known until recently. We had no general theory capable of dealing
equally with both kinds of algebras and putting the known results into a common
framework. Commutator theory for congruence modular varieties was created
in 1979 and quickly ended that state of affairs. In Volume 3, we shall present
this beautiful theory. (A brief introduction to it is contained in 54.13.) It deals
uniformly with the algebras in congruence modular varieties and acts somewhat
like a prism, placing the algebras into a spectrum ranging from congruence
distributive to Abelian. The algebras in varieties with permuting congruences are
distributed throughout this spectrum.
DEFINITION 4.65. Let a and fi be binary relations on a set A.
l. By a o" P (where n 2 1) we mean the relation a o P o a , which is a relation
product of n factors (alternating between a and 8). Thus a o' P = a, a o2 fi =
a o p, and a o"+' p = a o (p on a).
2. We say that a and p n-permute (where n 2 2) iff a on p = /?on a.
3. We say that the congruences of A n-permute iff every pair of congruences of
A n-permute.
Let us note that 2-permuting is the same thing as permuting. This concept
of n-permuting congruences turns out to be of interest in the classification of
varieties. Here, we are mainly interested in the case n = 2. We have introduced
the general concept just so that we can observe that the most memorable
consequence of congruence permutability, the modularity of the congruence
lattice, is already a consequence of congruence 3-permutability. We shall prove
this fact after first deriving some easy consequences of the definition. '
LEMMA 4.66. Let a and /? be equivalence relations on a set A.
i. (aonj?)U(pona)c(aon+'P)n(pon+'a)foralln2 1.
ii. a v p = U{aon/?:n r 1).
iii. For each n 2 1, statements (b) and (c) below are equivalent; they are implied
by (a), and imply (d). If n is even then (a), (b), and (c) are equivalent.
a. a and p n-permute.
b. p on a !E a on p.
c. a o" p = a v p, the join in the lattice of equivalence relations.
d. a and p n + 1-permute.
Proof. Since a and /3 are reflexive over A, for every binary relation p on A we
have p E (a o p) í l (p o a), and a similar inclusion holds with a replaced by P.
Statement (i) follows easily from this.
We know that a v /3 is the transitive closure of aU p. Since a and p are
reflexive, symmetric, and transitive, this means that (x, y ) E a v p iff there is a
. finite sequence x = x,, x,, . , xk = y such that (x,, xi+,) E a for al1 even i < k,
and (x,,xi+,) E /? for al1 odd i < k. This immediately implies statement (ii).
In statement (iii), the implication (a) => (b) is trivial. Now let n 2 1 be fixed
and assume that (b) holds. Then
since a o a = a. NOW it follows that a o"+' p = a o" j3 since (by (i)) a 0" P G a o"+'8.
In a similar fashion, we can deduce that P o n + l a = a o" p. Hence it follows that a
and p n + 1-permute (Le., (d) holds). Moreover, starting from the equalities
I which we just proved, an easy inductive argument establishes that

a ok p = a O"fl =
l ok a for al1 k > n,
and these equalities combined with statements (i) and (ii) yield (c). Thus (b) does
imply (c) and (d). The implication (c) * (b) is immediate from (ii).
Now suppose that n is an even integer and that (b) holds. By the converse
of a binary relation p we mean the relation
I Notice that the following equation holds for any binary relations p and T.
I From this equation, we immediately derive others by an easy inductive argument.

) al1 even k > 0;
(p ok z ) =~ (zu) ok ( P ~for
(p ok z)" = (pu)ok (zu) for al1 odd k > 0.
I Now since (b) holds and n is even, we can calculate
l Hence a o" p = p on a, ending the proof that (b) implies (a) for n even.
If a and p are congruence relations of a finite algebra A, then the expression

for their join given by part (ii) of the lemma involves only finitely many distinct
relations a o" p. Therefore a and P certainly n-permute if n is sufficiently large. In
fact, it can easily be shown that al1 congruences of an n-element algebra n-
permute. For a simple example of an n-element algebra and two congruences
that fail to n - 1-permute, see Exercise 1 below.
Suppose that L is a sublattice of Eqv A for some set A (i.e., a lattice of
equivalence relations on A). Part (iii) of the lemma implies that al1 members of L
n-permute iff a v p = a on P for al1 a, p E L. For this reason, lattices of n-permuting
equivalence relations are sometimes said to have type n - 1 joins. We remark
that our proof of Theorem 4.62 shows that every lattice is isomorphic to a lattice
of 4-permuting equivalence relations. That this is a sharp result is a consequence
of the next theorem.
THEOREM 4.67. Every lattice of 3-permuting equivalence relations is modular.

If un algebra A has 3-permuting congruences, then Con A is a modular lattice.
Proof. The proof uses the following version of the modular law of relational
arithmetic.
(15) Suppose that a, fl, and y are binary relations on the set A, and that
a" = a i (aoa)UP. Then a n ( p o y o p ) = /?0(any)0/3.
The law is proved by considering Figure 4.6. Suppose that (x, y) E a í l (P o y o P).
Thus there exist elements a and b that complete the picture-that is, (a, b) E y
and (x, a), (b, y) E p. Then, by considering the path (a, x), (x, y), (y, b), we see
that (a, b) E bu o a o pu. Since p c a and a is assumed to be symmetric and
transitive, we can deduce that (a, b) E a. It follows that (x, y) E /3 o (a í l y) o P, as
desired. The other inclusion implied in (15 ) is trivial.
Figure 4.6
Now suppose that L is a lattice of 3-permuting equivalence relations and

that a, p, y E L with a i p. It must be shown that a A (p v y) = /3 v (a A y). (See
$2.3.)But the meet operation in Lis identical with set intersection, and by Lemma
4.66(iii), the join operation is identical with the operation u * v = u o u 0 u. Thus
the modular law of relational arithmetic implies the modularity of L. ¤
In 54.2, we observed that the image of a congruence of an algebra A under

an onto homomorphism f : A -,B need not be a congruence of B. This will be
true, however, if A has 3-permuting congruences. In fact, the next theorem shows
that the property is equivalent to congruence 3-permutability.
THEOREM 4.68. Suppose that f : A -,B is an onto homomorphism. Let a =

ker f, let p be any congruence of A, and let f (P) = ( (f (x),f (y)): (x, y) E P). Then
f(P) is a congruence of B iff Poaop E aoPoa.
Proof. f (P) is obviously a subuniverse of B2, and it is a reflexive and symmetric

relation. Thus it must be shown that f (P) is transitive iff P o a o P G a o P o a. The
proof takes the form of a sequence of bi-implications. We leave it to the reader
to verify that each successive pair of the following statements are equivalent.
f (P) is transitive
for al1 x, Y, ZEB if (X,Y), (y,z) €f(P) then ( x , z ) Ef(P)-
for al1 (a, b), (c, d) E p, if f(b) = f(c) then there exists (r, S) E p
-
with f(r) = f(a) and f(s) = f(d)-
for al1 a, b, c, d E A if (a, b) E P, (b, C) E a and (c, d) E p, then
there exists (r, S ) E P with (a, r) E a and (S, d) E a
PoaoP G aopoa. M
We have seen that any algebra possessing a term operation that obeys the
equations in statement (14) has permuting congruences; that permuting con-
gruences 3-permute; and that algebras with 3-permuting congruences possess
modular congruence lattices. Before presenting some consequences of congru-
ente permutability that cannot be derived from congruence 3-permutability, we
present an important property of congruences that is implied by (14) and cannot
be derived from congruence permutability. We require a simple auxiliary result,
which will find some applications in later chapters.
THEOREM 4.69. Let A be any algebra and Y be any subuniverse of A" (where
n 2 1). The following are equivalent.
i. Y is preserved by every polynomial operation of A.
ii. ( a , . . . , a ) ~ Y f o r a l E a ~.A
Proof. To see that (i) * (ii), let Y satisfy (i) and let a E A. The constant O-ary
operation c, with value a is a polynomial operation of A. To assert that c,
preserves Y is to assert that when we apply c, coordinatewise to the empty
sequence of members of Y, the resulting n-tuple belongs to Y. This resulting
n-tuple is none other than the constant sequence (a, ,a). Thus this sequence
belongs to Y.
To see that (ii) * (i), assume that Y satisfies (ii) and let f be any polynomial
operation of A, say f is k-ary. By Theorem 4.6, for some m 2 O there is a
k + m-ary term operation t(xo,- xm-, ,u,, S , u,-,) of A and some elements
S ,
a,, , a,-, of A such that f is defined by the equality

. m .
To finish showing that f preserves Y, we let yo, yk-l be any members of Y

s.,
with yi = , ). Denote by ¿ i j ( j = 0, ,m - 1) the constant sequence

¿ij = (aj, ,aj) belonging to A" (and to Y since Y satisfies (ii)).Now f applied
coordinatewise to yo, . . yk-l gives the same result as t applied coordinatewise
S ,
to the elements ¿iO,, ¿im-l, yo, . , yk-l of Y. Since Y is a subuniverse, it is

preserved under the coordinatewise application of any operation derived by
composing the basic operations of A (i.e., by any term operation). In particular,
Y is preserved by t, and hence it is preserved by f. m
The subuniverses of A" satisfying 4.69(ii) are called diagonal subuniverses.
THEOREM 4.70. Suppose that the algebra A has a polynornial operation

d(x, y, z ) satisfying d(a, b, b) = a and d(a, a, b) = b for al1 a, b E A.
i. For every subuniverse p of A2 the following are equivalent.
a. p is a reflexive relation over A.
b. p is a congruence relation of A.
ii. For al1 a, b, c, d E A we have (a, b) E cgA(c,d) if there exists a unary poly-
nomial f (x) of A with f(c) = a and f (d) = b.
iii. Let X G A2 and (a, b) E A2. We have (a, b) E c g A ( x )if there is a finite
sequence of elements (x,, y,), (x,-, ,y,-, ) E X and a polynomial f E
S ,
Pol, A such that f (x,, - x,-,) = a and f (y,, . y,-,) = b.

a , S ,
Proof. Recall that a relation p is a congruence of A iff it is a reflexive, sym-

metric, and' transitive subuniverse of A2. (This is formula (7).) By the theorem
just proved, every reflexive subuniverse of A2 is preserved by the polynomial
operation d.
In order to prove statement (i), let p be any subuniverse of A2. Clearly, what
we have to do is prove that if p is reflexive, then it is also symmetric and transitive.
So assume that p is reflexive. Thus d preserves p (by 4.69). Let (a, b) E p. Then
we apply d to the triple (a, a), (a, b), (b, b) of members of p, obtaining
Thus (b, a) ~ p and , it follows that p is symmetric. To see that p is transitive,

assume that (a, b), (b, c) E p and apply d to the triple (a, b), (b, b), (b, c). The
resulting pair is just (a, c), showing that p is transitive.
Notice that statement (ii) is the special case of (iii) that results when X =
{ (c, d)}. The argument for (iii) (with most of the details left for the reader to
supply) goes like this. The relation p = cgA(x),which is just the smallest con-
gruence relation of A that includes the set X, is by (i) the smallest reflexive
subuniverse of A2 that includes X. In other words, p is the subuniverse of A2
generated by X U O, (where O, = ((a, a): a E A)). Using Theorem 4.69, argue, in
parallel with Exercise 4.9(7), that p is identical with the set of values of poly-
nomial operations of A applied coordinatewise in A2 to finite sequences of
elements of X. Verify that this statement is precisely the required result. m
4.7 Permuting Congruences 201
COROLLARY. Among these properties of un algebra, ( i ) => ( i i ) * (iii).

i. A satisfies (1 4).
ii. ConA is asublattice of SubA2.
iii. The congruences of A permute.
Proof. Assume that (i) holds. Then by Theorem 4.70(i), congruences of A are
the same as diagonal subuniverses of A2. These diagonal subuniverses constitute
the principal filter in SubA2 consisting of the subuniverses that contain the
identity relation. Thus (ii) holds.
Now assume that (ii) holds. Let a and P be abitrary congruence relations of
A. We have
where a v p is the equivalence relation join of a and P. By (ii), a v P is identical

with the join of a and /? in Sub A2. Moreover, we can easily check that n 0 B is
a subuniverse of A2. Thus the displayed inclusions imply that a o p = a v 8.
Reversing the roles of a and P in the argument, we get that P o a = P v a = a o b,
so (iii) holds.
When the congruences of A permute, the concept of a finite direct repre-

sentation of A becomes equivalent to a purely lattice-theoretical concept. This
permits the application of results in modular lattice theory to derive results about
direct representations. Recall from 54.4 (Definition 4.30, Theorem 4.31) that a
system of congruences (g5i: i~ 1 ) is a direct representation of A iff the mapping
f : A -+ni,, n
A/4i is an isomorphism, and that we write O, = i,, 4i to denote
that the system is a direct representation. The concept of a directly join indepen-
dent subset of a lattice, and that of a directly join irreducible element, were defined
in $2.4, Definition 2.43. The dual concepts of a directly meet independent subset
and a directly meet irreducible element are the relevant ones for the lattice-
theoretical discussion of direct products.
LEMMA. Let a, B and &, 4,, - #, be congruences of un algebra A with 4i #

S , Zj
(for i # j). If al1 congruences of A permute, then
i. O, = a x P iff a and P are complements of one another in Con A;
n mi
ii. O, = i,,, iff {#i: i 5 n ) is a directly meet independent subset of Con A
whose meet is O,;
iii. Ala is directly indecomposable iff a is a directly meet irreducible element of
Con A.
Proof. We assume that A has permuting congruences. Then statement (i)

follows immediately from Definition 4.28 and Theorem 4.29.
n
To prove the second statement, assume first that O, = i5n gbi. Thus O, =
A{q5i: i n). To conclude that {bi: i 5 n} is directly meet independent, we have
to show that for each i S n we have $i v v{bj: j # i } = 1,. Without losing
generality, we can assume that i = n. Let (a, b) be any pair of elements of A. Let
then (xj: j 5 n) be the sequence with xj = b for al1 j < n and x, = a. Since
0, = n,(, #,, by Definition 4.30 there is an element x E A with x = xj (mod #j)
for al1 j I n-in other words, with (a, x) E 4, and (x, b) E {#j:j < n}. Since
(a, b) was arbitrarily chosen, this proves what we desired: #, and j < n)
join to 1, in Con A.
Continuing with the proof of (ii), let us now assume that the 6,constitute a
directly meet independent set of congruences that meet to O,. Let x,, S , x, be
any elements of A. We wish to show that there exists an element y such that
y E xi (mod 4,) for every i 5 n. For i I n let Ji = A(#j: j # i]. Define elements
y, (i 5 n) by induction on i. Let y, = x,. Suppose that y, has been defined and
that y, 7 x,(mod 4,) for al1 k 5 i. If i < n define y,+, as follows. Since Ji+,0 #,+, =
1, and #i+l n$+, = O, there is a uniquely determined element a E A satisfying
y, a (mod #i+l) and a = x,+, (mod bi+,). Let y,+, be this element a. Clearly
y,+, E y, (mod 4,) for each k I i (since Ji+,c &) and thus y,+, S x, (mod A).
Therefore y,+, r x, (mod #,) for al1 k I i + 1. This inductive definition produces
an element y = y, that satisfies al1 of the required congruences. The proof of (ii)
is now finished.
The proof of (iii) is left as an exercise.
Direct representations with an infinite number of factors admit no charac-

terization in purely lattice theoretical language, even if we assume that congru-
entes permute. Exercise 4.38(17) shows why this is so.
THEOREM 4.71. Let A be any algebra whose congruences permute and whose
congruence lattice has jinite height. A is isomorphic to a product of jinitely many
directly indecomposable algebras. Moreover, for every two direct representations
ni,, Ci E A E n j , , D j with directly indecomposable factors, the sets I and J are
jinite and 111 = IJI.
Proof. The congruence lattice of A is a modular lattice by Theorem 4.67. A has

no infinite direct representations because Con A has finite height. According to
the lemma, the finite systems of congruences (#,, . . ,4,) giving direct representa-
tions of A are the same as the finite directly meet independent systems that meet
to O, and A/#, is directly indecomposable iff 4, is directly meet irreducible. This
theorem follows therefore from the Direct Meet Decomposition Theorem (i.e.,
Theorem 2.47 applied to the dual lattice of Con A).
Theorem 4.71 is a rather weak application of Theorem 2.47. With slight

changes in the hypotheses of the theorem, it is possible to prove that A is a direct
product of directly indecomposable algebras in essentially only one way. (The
algebras occurring as the factors in a direct decomposition of A into indecom-
posable factors are determined up to isomorphism and rearrangement of their
order.) This conclusion holds for example if ConA is a finite height lattice of
permuting equivalence relations and A has a 1-element subalgebra, or if Con A
is modular, A is finite, and A has a one-element subalgebra. These results are
proved in Chapter 5 using a variant of Theorem 2.47.
For subdirect representations there is a result somewhat analogous to
Theorem 4.71. It is a corollary of the Kurosh-Ore Theorem (Theorem 2.33) and
requires only that the algebra have a modular congruence lattice. As we saw in
$4.4, there is no essential difference between a subdirect representation of an
algebra and a system of congruences that intersect to zero. Systems of congru-
entes that intersect irredundantly to zero correspond to irredundant subdirect
representations (in which no factor can be removed without losing the one-to-one
property of the subdirect embedding). Let us define these notions more precisely.
/ DEFINITION 4.72.
l l.
2.
A subset M in a complete lattice L is called meet irredundant iff for al1 proper
subsets N of M we have /\M < &v.
A subdirect representation f : A -* ni,,
A, is called irredundant iff for every
proper subset J of 1 the natural map
on the f-image of A.
ni,, n,, ,
Ai -+ Aj is not one-to-one
THEOREM 4.73. Let A be un algebra whose congruence lattice is modular. Then

al1 irredundant subdirect representations of A with a Jinite number of subdirectly
irreducible factors have the sume number of factors. For any two such subdirect
representations given by systems of congruences (4,, bn) and (S/, . . S/,),
m , S ,
it is possible to renumber the $'S so that for every k 2 n the system

(S/, S/2, - Sk,4k+l,. #,,) giues un irredundant subdirect representation of A.
S , S ,
Proof. Suppose that (4,, m , 4,) and (S/, , ,SI,) are finite systems of congru-
entes associated with irredundant subdirect representations f : A -+ B, x - x B,
and g : A -. C, x x C, where B, and Cj are subdirectly irreducible. Then
clearly the 4's are distinct meet irreducible elements of ConA and {4i: i =
1, Sm ) is meet irredundant, and the same conclusion holds for the $'s. The
,
desired conclusion immediately follows upon an application of Theorem 2.33 to

the dual lattice of L = Con A. ¤
A subdirect product of two algebras in which the kernels of the projections

permute has an especially simple structure. In the terminology of category theory,
such an algebra (or its inclusion map into the full product) is an equalizer of two
morphisms.
THEOREM 4.74. (Fleischer J19.551.) Suppose that A S A, x A, is a subdirect

product and that yo o y, = y, o y, where y, is the kernel of the map A: A -+ A, for
i = 0, 1. There exists un algebra C and surjective homomorphisms a,: A, -+C such
t hat
Proof. We put y = y, v y, (=yo oy,), put C = Aly, and use n to denote the
quotient map from A onto C. Since the A are surjective, by the Second Iso-
morphism Theorem (Theorem 4.10) there are unique surjective homomorphisms
a,: A, -+ C satisfying aiJ = n (i = 0,l). For x = (x,, x,) E A we have
This means that A is contained in the equalizer of a, and a,. Conversely, let
(a,, a, ) E A, x A, be such that a,(a,) = a,(a,). Choose elements Zi E A such that
= a,. Thus n ( G ) = n(Z, ), or equivalently, (Z,, x, ) E y. We can now choose
an element F E A with Z , y , ~ , ~ (since
, y = y, oy,). Thus &(F) = fi(Zi) = ai
(i = 0,l). From these equations, it follows that Z = (a,,a,), so (a,,a,) does
belong to A, as required.
The next lemma, definition, and theorem concern a frequently encountered

situation in which irredundant subdirect products are direct.
LEMMA. Let f : A + A, x x A, be an irredundant subdirect representation

where A has permuting congruences and A,, - , A, are simple algebras. Then
f : A r A, x x A,.
Proof. Let 4, denote the kernel of the correlated homomorphism of A onto A,.
By the lemma preceeding Theorem 4.71, we just have to show that the 4's con-
stitute a directly meet independent set of congruences (and that they intersect to
0,-but that is one of our hypotheses). Letting i be a fixed but arbitrary index and
4, = Aj+,d>i, we have to show that 4, v 6, = 1,. Now 4, 4 1,-i.e., 4, is a
coatom of Con A, since A/4, r A, is a simple algebra (Definition 4.23, Theorem
4.12). Moreover, the meet irredundance of idj: 1 I j S k ) implies that 4i A6, < Ji,
vi
equivalently, q5i v > 4,. Since 4i 4 lA, there is only one possible value for
4, v vi, namely 1,. This concludes the proof.
DEFINITION 4.75. An algebra is called semisimple iff it is isomorphic to a
subdirect product of simple algebras.
THEOWEM 4.74. Let A be any algebra such that ConA is modular. These
statements are equivalent.
i. Con A is a complemented modular lattice of finite height.
ii. A is isomorphic to a subdirect product of a Jinite system of simple algebras.
iii. A is semi-simple, and Con A has finite height.
If the statements hold and the congruences of A permute, then there is un integer
n such that every irredundant subdirect representation of A with subdirectly irre-
ducible factors is un isomorphism of A with a product of n simple algebras.
Proof. A modular lattice L is complemented and of finite height iff the zero
element in L is the meet of a finite set of coatoms. (This follows by an easy
application of Theorem 2.40 to the dual of L.) Therefore, it is clear that (i) and
(ii) are equivalent and that (ii) implies (iii).
In a modular lattice of finite height, every meet of a set of elements is equal
to the meet of a finite subset of the set (Exercise 9 below). If (iii) holds, then O, is
equal to the intersection of a set of coatoms of Con A and so is equal to the meet
of a finite set of coatoms; thus (ii) holds. Therefore, the three statements are
equivalent.
Now suppose that A has permuting congruences and that (i) holds. Let n
be an integer such that O, is the meet of a meet irredundant set of n coatoms of
Con A. Let M be any meet irredundant set of meet irreducible congruences whose
intersection is O,. Since Con A has finite height, there is a finite subset of M that
intersects to O,; hence M is finite. By Theorem 4.73, we have that 1 MI = n. By
Theorem 2.40 (applied to the dual lattice), Con A is relatively complemented. This
implies that every meet irreducible element # 1, is a coatom of the lattice. Thus
M gives an irredundant subdirect representation of A as a subdirect product of
n simple algebras. By the lemma above, this representation is direct.
The lattices mentioned in the theorem, complemented modular lattices of

finite height, have been extensively studied. They belong to the broader class of
complemented modular algebraic lattices. We shall prove in the next section an
important representation theorem for lattices in this broader class.
Exercises 4.77
l. Let k be an integer greater than 1, and let C, denote a k-element chain
construed as a lattice. Prove that C, has k-permuting congruences but does
not have k - 1-permuting congruences. Find two congruences a and fl of
C, such that a ok-l fl < fl ok-l a.
2. A lattice of more that one element cannot satisfy (14). (Use Theorem 4.70.)
3. Recall that a quasigroup was defined in 53.4 to be an algebra (Q, -,\, /) with
three binary operations obeying the laws: y (y\x) Ñ x Ñ (xly) y and
y\(y z) Ñ z Ñ (z - y)/y. Prove that every quasigroup satisfies (14). (Try the
term operation t(x, y, z) = [x/(y\y)] (y\z).)
4. Heyting algebras were defined in Exercise 4.48(18). Prove that Heyting
algebras satisfy (14). Conclude that Boolean algebras satisfy (14).
5. Let n be a positive integer and A be an algebra having 3-ary term operations
t,, - . S, tn-, satisfying these laws: x Ñ t,(x, y, y), t,(x, x, y) Ñ t, (x, y, y), S ,
t n - 2 ( ~X,, y) Ñ tn-, (x, y, y), tn-, (x, X,y) Ñ y. Prove that A has n + 1-permuting
congruences.
6. An implication algebra is an algebra (A, -+)with one binary operation satis-
fying the laws: (x -+ y) + x Ñ x, (x + y) -+ y Ñ (y + x) + x, x --+ (y -+ z) Ñ
y -+(x -+ z). For example, if (A, A , v ,-) is a Boolean algebra, then the
operation x -+ y = x- v y gives an implication algebra on the base set A.
Show that implication algebras possess term operations t,(x, y, z), t, (x, y, z)
that obey the equations in the last exercise with n = 2, and so implication
algebras are congruence 3-permutable. Prove that the implication algebra

derived as above from a 2-element Boolean algebra does not satisfy (14).
7. Suppose that the algebra A has n term operations obeying the equations in
Exercise 5. Prove that if p is any reflexive subuniverse of A2,then pu G p on p.
8. Prove statement (iii) in the lemma preceding Theorem 4.71.
9. Let L be a lattice in which al1 chains are finite. Prove that L and L~ are
algebraic lattices in which every element is compact. Equivalently, every set
X S L contains a finite set X' with VX VX' = and /\X = /\x'.
10. Prove that a distributive lattice without infinite chains is a finite lattice.
11. Prove that every subdirectly irreducible modular lattice of finite height is
simple. (See Corollary 2.76.)
12. Prove that every relatively complemented lattice has permuting congruences.
(Theorem 2.68 may be useful.)
13. We sketch an alternate proof of Theorem 2.72 in 52.6, which states that every
directly indecomposable relatively complemented lattice of finite height is
simple. The exercise is to supply the missing details for the argument. Let L
be any relatively complemented lattice without infinite chains. Show that
every congruence of L is determined by the pairs of congruent elements (a, b)
such that a -< b. Then show that every congruence is determined by the
set of atoms that are congruent to O, or by the largest element congruent
to O. Using Theorem 2.50 and Exercise 10, conclude that Con L is finite and
that L has a finite irredundant subdirect representation with factors Li that
are subdirectly irreducible, relatively complemented lattices without infinite
chains. L, is (O, b,)-irreducible for some atom b,. Using Theorem 2.68, show
that every two-element interval I[c, d] in L, is projective to I[O, b,], and
conclude that Li is simple. By the result of Exercise 12 and by the lemma
preceding Theorem 4.76, L is isomorphic to the direct product of the Li's.
Thus if L is directly indecomposable, then L is simple.
4.8 PROJECTIVE GEOMETRIES
The principal theorems of this section, 4.86 and 4.87, are important results in the
theory of modular lattices. These theorems assert that every complemented
modular algebraic lattice is a direct product of directly indecomposable lattices,
in fact, a direct product of two-element Boolean lattices and projective geom-
etries. A notable corollary asserts that the lattices occurring in Theorem 4.76,
complemented modular lattices of finite height, are decomposable into products
of finite Boolean lattices and finite dimensional projective geometries. Our route
to these results is reasonably short, given the machinery and results already
at hand in Chapter 2. The first task is to define precisely the concept of a
projective geometry, then to show how these objects can be identified with certain
complemented modular lattices. Our presentation of this topic is closely modeled
on the one in Jónsson [1959]. We begin by giving a classical definition of
4.8 Projective Geometries 207
two-dimensional projective geometries, and by showing that these objects can

be identified with certain lattices of height three.
DEFINITION 4.78. A projective plane is a pair n = (P, A) such th@ P is a

nonvoid set, A is a collection of subsets of P, and the following aAoms are
satisfied. The elements of P are called points, and the elements of A are called lines.
l. Any two distinct points belong to one and only one line.
2. Any two distinct lines contain precisely one point in common.
3. Every line has at least three points.
4. There exist three distinct points such that no line contains al1 of them.
Given a projective plane n = (P, A) we define
a set of subsets of P. Axiom (2) implies that L" is a closed set system. We
define Ln to be its lattice of closed sets ( L E ,A , v ). (See Definition 2.9 and
Theorem 2.11.) There is an obvious bijection between the points of n and the
atoms of L"-the elements of height 1 in this lattice. The lines of n are precisely
the coatoms of L"-the elements of height 2. The elements of L" are the null set
(the least element), atoms, coatoms, and P (the greatest element). For a point p
and line 1 we have p E 1 iff (p) I1 in L".
DEFINITIQN 4.79. Let L be any lattice with O. By a point of L we mean an

element of height 1 in L (or an atom); by a line of L we mean an element of height
2 in L.
THEQREM 4.80. Let L be a lattice. L is isomorphic to L" for some (necessarily

unique up to isomorphism) projective plane n iff L is a complemented modular
lattice of height three and every line of L contains at least three points.
Proof. The uniqueness of n up to isomorphism, if it exists, is made clear by the

preceding remarks.
Suppose that L = L". Note that the join of any two points (atoms) of L" is
a line, and that the meet of any two lines is a point (by axioms (1) and (2) for n).
To see that L" is modular, we use the characterization of modular lattices in
Theorem 2.25 [(i) e (iv)]. We assume that L" contains elements x, u, u, with u < u,
x v U = x v u, and x A u = x A u, and proceed quickly to a contradiction. In
this situation, neither u nor v can be the O or the 1 of L"; thus u can only be a
point {p), and v must be a line. Moreover, x must be incomparable to both u
and u, so we have x v v = 1 and x A u = O. If x = (q) is a point, then x v u is
the line containing p and q, contradicting that x v u = x v v = 1. If x is a line,
then x A v is the point of intersection of the lines x and u, contradicting that
x A v = x A u = O. These contradictions prove that L" is modular.
In a projective plane, every line fails to contain some point, and every point
fails to lie on some line (Exercise 3 below). These facts easily yield that L" is
complemented. It is clearly a lattice of height three. (Any maximal chain has the
form < (p} < 1 < P.) That every line in Ln contains at least three points
is equivalent to axiom (3). This completes the proof that Ln has the stated
properties.
Now let L be any complemented modular lattice of height three in which
every line contains three or more points. Note that L is relatively complemented
by Theorem 2.34. Let P be the set of points, and C be the set of lines, of L. For
each C E Clet C = ( ~ E Pp: 5 c} and let A = (F: CEC}.Then define n = <P,A).
Since al1 maximal chains in L have length three and take the form O < p < c < 1
where p is a point and c a line (this follows from Theorem 2.37), P and C are
disjoint sets, and we have L = {O} U P U C U (1). Thus, also, every line contains
a point. Let C E C and choose p E P, p < c. Since L is relatively complemented,
there exists q such that q v p = c and q A p = O. Clearly, q is a point. We
have just established that every line is the join of two points, from which it
follows that c -,¿? is a one-to-one function from C onto A. This function extends
obviously to a one-to-one map from L onto L", which is an order isomorphism.
Al1 that remains is to prove that n is a projective plane. Let h be the height
function of L. Thus h(0,) = O, h(p) = 1 for p E P, h(c) = 2 for c E C, and h(1,) = 3.
(See Theorem 2.39 for further properties of h.)
Axiom (1):Let p and q be distinct points of n. Then h(p v q) = h(p) + h(q) =
2, since p A q = O. Thus p v q = c, a line of L, and p and q belong to the line C
of n. Let u be any other line such that p and q belong to E. Then u 2 c, and we
have u = c, since both elements have height 2.
Axiom (2): Let ü and E be distinct lines of n. From the equations
u vu = l,,
we conclude that h(u A u) = 1-i.e., u A u is a point p. The proof that p is the
unique point contained in both u and u is analogous, but dual, to the argument
of the last paragraph.
Axiom (3): This is equivalent to one of our assumptions about L.
Axiom (4): Pick any line c in L and let the element p be chosen as a
complement of c. Thus p is a point. Let q and r be distinct points below c. The
points p, q, r do not al1 belong to any one line ü of n because if they did, then u
could only be c, while p $ c. •
EXAMPLE
The Fano Plane is a projective plane with seven points and seven lines. The
lines are represented by the straight-line segments and the circle in Figure 4.7.
The lattice associated with this plane is isomorphic to the lattice of vector
subspaces of three-dimensional vector space over the two-element field. This
means that the plane can be represented so that the points are the one-
1 4.8 Projective Geometries
dimensional subspaces of this vector space, the lines are the two-dimensional
subspaces, and the relation of a point lying on a line becomes the inclusion
relation between subspaces. In the same way, every field yields a projective plane,
which is said to be coordinatizable over that field.
The Fano plane

Figure 4.7
Just as in Euclidean geometry, we consider that a projective plane is a

projective geometric object of dimension 2; a line in the plane has dimension 1;
and a point has dimension O. Our definition of a general projective geometry has
for primitive notions just those of "pointy' and "line," but we have to introduce
subspaces of higher dimensions (a concept to be defined in terms of points and
lines) in order to formulate al1 of the axioms.
DEFINITION 4.81. By a projective geometry we mean a pair I' = (P, A) such

that P is a set of at least three elements (called the points of r ) , A is a collection
of subsets of P (called the lines of r ) , and the following three axioms are satisfied:
1. Any two distinct points belong to one and only one line.
2. Every line contains at least three points.
For every set X E P, denote by C(X) the smallest subset of P containing X, which
contains every line with which it has two distinct points in common.
3. Suppose that X U (p, q) E P, p # q, and p E C(X U ( q ) ) .Tl~ereexists r E C(X)
such that p E C ( ( r ,q)).
Let l? = (P, A) be a projective geometry. The function C defined above is
obviously an algebraic closure operator on P. We denote by Lr the set of closed
sets for this operator and cal1 the members of Lr the (linearly closed) subspaces of
l?. The null set, the sets {p) ( ~ E P )and, the lines are subspaces; every other
subspace contains at least two distinct lines. The lattice of closed sets for C,
Lr = ( L ~A, , v ),is called the lattice of subspaces of T. The least cardinal K such
that P = C(X)for some set X with 1 X 1 = K + 1is called the (projective)dimension
of T. We shall prove in Theorem 4.87 that L~ is a modular lattice. We remark
that when r has finite dimension, the lattice-theoretic dimension (or height) of
+
Lr is 1 the projective dimension of I'.
EXAMPLES
1. If P is a set of 2 3 elements, then r = (P, (P) ) is a projective geometry of
dimension 1, called a projective line. L~ is the modular lattice of height 2 having
rc = 1 P ( points.
2. The projective geometries of dimension 2 are precisely the projective

planes.
3. Let V be a vector space of dimension 2 2 over a division ring D. Let
P be the set of al1 one-dimensional subspaces of V. For every two-dimensional
subspace u let ü = ( p E P : p c u ) and let A be the collection of al1 of these sets ü.
Then I' = (P, A) is a projective geometry (said to be coordinatizable over D) and
1 + dim r = dim V. L' is isomorphic to the lattice of al1 vector subspaces of V.
It is known that the examples of the last paragraph include al1 the projective
geometries of dimension greater than 2 but not do not include al1 the projective
planes. In other words, every projective geometry whose dimension exceeds 2 is
coordinatizable over a division ring, but there exist projective planes (even finite
ones) that are not coordinatizable over any division ring.
DEFINITION 4.82. A complete lattice L is called atomistic iff every element

of L is a join of atoms of L.
Our aim is to characterize the lattices of subspaces of projective geometries

as the directly indecomposable, atomistic, modular, algebraic lattices that possess
at least three elements. We require three lemmas.
LEMMA 4.83. For a modular algebraic lattice L the following are equivalent.
i. L is cornplemented.
ii. L is relatively cornplemented.
iii. L is atomistic.
iv. The join of the atoms of L is 1.
Proof. We have that (i) * (ii) (for any modular lattice) by Theorem 2.34. (See
also Theorem 2.40.)
Now suppose that L is relatively complemented. Let u be any element of L.
We wish to show that u is the join of a set of atoms. Since every element of L is
the join of a set of compact elements, it will suffice to consider just the case when
u is compact. Let u' be the join of al1 the atoms 5 u. Assume that u' < u, in order
to get a contradiction. By Lemma 4.49(i) in $4.6, we can choose a maximal
element m below u in the interval I [u', u]. Since L is relatively complemented, we
can choose an element m' such that m v m' = u and m A m' = O. By Dedekind's
Transposition Principle (Theorem 2.27), the intervals I [m, u] and I [O, m'] are
isomorphic, which means that m' is an atom. This implies that m' I u'-but that
is impossible, because u' I m and m A m' = O. The contradiction proves that
(ii) S-(iii),
That (iii) a (iv) is contained in the definition of atomistic lattice. Finally, to
see that (iv) a (i), assume that (iv) holds and let u be any element of L. We wish
to show that u has a complement in L. The join-continuity of L (Exercise 4.64(4))
and an application of Zorn's Lemma yields that there exists a maximal element
u in the set {x: u A x = O}. To prove that u v u = 1, we will show that u v u
includes every atom of L. Let a be an atom. Without losing generality, we
can assume that a $ u and a $ u. We have then that a A u = O and ICO, a] is
isomorphic to I[v, a v u], implying that u -< a v v. By our choice of v we have
that b = u A (a v u) # O; thus b v. Then since u -< a v u, it follows that
b v u = a v v. By the modular law,
From this we derive that
LEMMA 4.84.
-
as desired. This argument proves that u is a complement of u, and that concludes
our proof of (iv) (i).
Let L be a modular lattice with O.

M
i. Suppose that a, b, c are three distinct atoms of L. Then a Ib v c iff b Ia v c

$favb=avc=bvc.
ii. Suppose that p and q are atoms of L, p # q, and p 4 x v q. Then there exists
an atom r < x such that p i r v q.
iii. Suppose that p and q are atoms of L and p $ x. If p x v q, then q i x v p.
Proof. To prove (i), we replace L by its sublattice L' = I[O, a v b v c], which
contains al1 the elements involved. L' has finite height; let h be the height
function of L'. Then h(a v b) = h(a v c) = h(b v c) = 2. Thus if a 5 b v c, then
b v c = a v b v c 2 a v c, implying that b v c = a v c, since these elements
have the same height. The proof of (i) is easily concluded, essentially by repeating
O this argument.
To prove (ii), suppose that p 5 x v q, p # q, and p and q are atoms. Then
by the modular law. Since I[O, p v q] is a modular lattice of height 2, the nonzero
element (p v q) A x must contain an atom a. If a, p, q are distinct, then p 2 a v q

holds by (i), and we can take r = a. If a = p, then we can take r = p. Suppose
that a = q. Then q < x, so p < x v q = x, and again we can take r = p.
To prove (iii), suppose that p and q are atoms and that p < x v q, p x. If
p = q, then the desired conclusion is obvious. Assume that p # q, and (by (ii))
choose an atom r < x such that p 5 r v q. Clearly, r # p # q # r. The desired
conclusion is an application of (i). m
Note that statement (ii) in the lemma is a lattice version of the third axiom
of projective geometry. Statement (iii)is a lattice version of the exchange property
involved in proofs that every two bases of a vector space have the same cardinality.
(For some equivalents of the exchange property, see Exercise 2.20(6).)
LEMMA 4.85. Let L be un atomistic lattice such that the join of any two distinct
atoms of L contains a third atom and ILI 2 2. Then L is directly indecomposable.
Proof. Assume that L is atomistic and that L z L' = Lo x L, where each of

the lattices L, has at least two elements. Each Li must contain an atom ai.
Then we have an element (a,,a,) in L' that is the join of two distinct atoms
and contains only those two atoms. L must also have an element with those
properties.
The next three theorems are the principal results of this section. Our proof
of the first shows that a complemented modular algebraic lattice is directly
indecomposable iff it satisfies the condition of the last lemma.
THEOREM 4.86. Let L be a complemented modular algebraic lattice. There

exists a system (Li: i E I ) of directly indecomposable complemented modular alge-
braic lattices such that L g n i , , L i .
Proof. We denote by P the set of al1 atoms in L. By Lemma 4.83, every element
of L is the join of a subset of P; it easily follows that the compact elements of
L are exactly the joins of the finite subsets of P. We define a binary relation
on P by
-
(18) -
p q iff p = q or there exists an atom a satisfying
aIpvq,p#a#q.
This relation is reflexive and symmetric. We claim that it is transitive. Suppose
-- -
that p q r. We wish to show that p r. We can assume that p, q, r are three
distinct atoms. Let a be an atom below p v q and distinct from p and q; let b
be an atom below q v r and distinct from q and r. If a = b, then by severa1
applications of Lemma 4.84(i) we can derive that p v q = p v r, from which we
-
can conclude that p r, as desired. Suppose that a # b. Considering the height
function h of the lattice I[O,p v q v r], we see that h(p v q v r) < 3, while
h(a v b) = 2 = h(p v r). Thus we have (a v b) A (p v r) # O. Let c be an atom
below this nonzero element. If p # c # r, then it follows from our definition
of
p I
- -
that p r, so we are done in this case. Now assume that c = p. Thus
a v b. We already know that p # a # b; thus b i p v a (true if b = p, and
true by 4.84(i) if b # p). Then, with severa1 applications of 4.84(i), we get that
b i p v a = p v q, p < q v b = q v r, q < p v r. The last inclusion implies that
-
p r, as desired. If instead c = r, then the same argument (with p and r inter-
-
changed) shows that r p. Thus -
is an equivalence relation on P.
We next prove this statement.
Let p E P and X E P with p S V X . Then X has a finite subset
(19)
-
(x,,--.,x,) such that p xi for al1 i I n and p < x, v v x,.
In proving it, we can first replace X by a finite subset (since p is compact). In
fact, we assume that X = {x,, - x,> and X is a minimal finite subset of P for
a ,
- ,-,.
which p 2 VX. We must show that p xi for al1 i. By symmetry, it will suffice
--
to show that p x,. Put x = x, v vx Thus p i x v x,, while p $ x. If
p = x, then p x,, as desired. If p # x,, then choose an atom r i x such that
p i r v x, (by Lemma 4.84(ii)). Since p $ x, it is easy to see that p, r, x, are
-
distinct. Thus r < p v x, by Lemma 4.84(i), and so p x,. This proves (19).
We can now define the indecomposable factors of L and the required
isomorphism. We take I = {pl- : p E P), the set of equivalence classes of atoms.
For ~ Ewe Iput
1, = V{p: p ~ p ) and L, = I[O,l,].
We define a mapping F from L to the product of this system of lattices:
To see that the L, are directly indecomposable, we need
(20) For p E I, Lp is an atomistic lattice and p is its set of atoms.

Indeed it is clear that L, is atomistic, and its atoms, which are the atoms of L
that lie below l,, include the members of p. It also follows by an easy application
of (19) that every atom below 1, belongs to p. Now Lemma 4.85 implies that L,
is directly indecomposable, since for every two atoms p and q of L, we have p q.
It remains to be shown that F: L z n,,,L,. It is clear that F preserves
-
meets. This argument shows that F is one-to-one: If u # u, then the atomicity of
L implies that some atom p lies below one or the other of these elements and not
below both. We have that p < 1, where ( = pl-, and therefore (F(u))<# (F(v)),.
In order to prove that F maps L onto the product, let (u,: v E 1) be a system
of elements with u, I 1, for al1 v. Put u = V{u,}. Clearly, F(u) 2 (u,: v E I). Let
II E I. TOprove that (F(u)), ( = u A 1,) is identical with u,, we invoke the atomicity
of L. Let p E P with p I u A 1,. By (20), we have p E A, and there are subsets
X, c v for v E 1, with u, = VX,. We now put X = U,,,X, so that á u = V X .
By (19), p lies below the join of a finite subset of X í l 1;i.e., p 5 VX, = u,. We
have proved that every atom below (F(u)), is below u,; and since obviously
u, i (F(u)),, it follows that these two elements are equal. This holds for every A,
so F(u) = (u,: v E 1).
The proof that F is an isomorphism is concluded by observing that every

meet-preserving, one-to-one, surjective mapping between two lattices is an
isomorphism. m
THEOREM 4.87. For every lattice L these statements are equivalent.

i. L E Lr for a (uniquely determined up to isomorphism) projective geometry r.
ii. L is a directly indecomposable complemented modular algebraic lattice and
ILI 2 3.
Proof. In this proof, we return to the practice of calling the elements of heights
1 and 2 in a lattice points and lines, respectively.
Let r = (P,A) be a projective geometry. This object satisfies the axioms
laid down in Definition 4.81. Its lattice of subspaces Lr has been defined, in the
paragraph following that definition, as the lattice of closed sets for a certain
closure operator C. Obviously, the points of I' can be identified with the points of
Lr, and the lines of I' can be identified with the lines of L ~in, such a way that the
incidence relation in I' (p is incident with 1 iff p E 1) becomes identified with the
order relation of Lr restricted to points and lines. These remarks establish the
assertion of uniqueness in the theorem.
Furthermore, it is clear from our definition of L' that C is an algebraic
closure operator of rank 3 (Definition 4.52) on P and that every one-element
subset of P is closed. These facts imply that Lr is an algebraic, atomistic lattice.
Thus Lemma 4.83 will yield that Lr is complemented, as soon as we have proved
that it is modular. Since IPI 2 3, then 1Lr/ t 3. The join of two distinct points
of L~ is a line, and every line contains three or more points (by 4.81(2)). Thus
Lemma 4.85 implies that Lr is directly indecomposable. Al1 that remains is to
prove that Lr is modular.
In order to do that, we require this explicit description of the join operation
of the lattice.
(21) ForX, ~ i n ~ ~ a n d r ~ ~ , w ev hYai fvf r ce Xr U~Y~
or there exist p EX, q E Y such that p # q
and r, p, q lie on one line.
To prove (21), we define X V Y to be the set of al1 points that belong to X
or to Y or to a line joining distinct points of X and Y. We assume that X = C(X)
and Y = C(Y), and it must be shown that C(X U Y) = X V Y. It is clear that
X U Y c X V Y c C(X U Y), because C(X U Y) is just the smallest set of points
that includes X and Y and includes every line on which it has two distinct points.
We note that every member r of C(X U Y) belongs to C(XfU Y') for some finite
sets X' E X and Y' c Y, because C is algebraic; and we define n(r) to be the
smallest of the numbers IX'I 1 Y'I such that these conditions hold. Notice that
n(x) 5 1 iff x E X V Y iff x E C({p, q)) for some {p, q) G X U Y.
Continuing with the proof of (21), let us suppose that it fails and choose
r E C(X U Y) - X V Y with n(r) as small as possible. Then we can choose finite
sets X' c X and Y' c Y such that r eC(XfU Y') and 1 x1 1 Y'/ = n(r). Since
I
n(r) 2 2, we have Y' # a, so we can write Y' = Y" U {q) with q $ Y". Letting
U = X'U Y", we have that P E C ( UU {q}). Clearly, r # q. Now by axiom (3)
for projective geometries, there exists a point p E C(U) such that r E C({p, q}).
Since, obviously, n(p) < n(r), we have ~ E X YV by the choice of r. Thus let
{a,b} E X U Y satisfy p E C({a, b}). Now r E C({a, b, q)), and r equals no one of a,
b, q. If b E Y, then we employ axiom (3) to get an element w E C({b, q}) with
r E C({a, w}). Now b, q E Y, implying that w E Y and {a,w} c X U Y Then, since
r E C({a, w}), we have that r E X V Y after all-a contradiction. The same argu-
ment works if a E Y. In the remaining case, {a, b} E X, and since r E C({a, b, q})
and q E Y, the argument can be concluded just as above. Our proof that (21)holds
is complete.
We are now ready to prove that L' is modular. Let X, Y, Z be elements of
this lattice such that X 2 Y. We have to show that X A (Y v Z) I Y v (X A 2).
Let p be any point belonging to X A (Y v 2). We can assume that p $ Y U 2 , else
the required conclusion is trivial. By (21), there are points q E Y and r E Z with
p E C( {q,r} ). We have that q # r (since p $ Y U Z), so C( {q,r}) is the unique line 1
containing q and r. (Here we are using axiom (l).) Since p # q, 1 can also be
characterized as the unique line containing p and q. Thus 1 c X, since {p, q) E X
and X is a subspace. Thus r E X í l Z, so p E C((q, r}) S Y v (X A Z), as required.
We have proved that L~ is a modular lattice.
The proof that (i)3 (ii) in Theorem 4.87 is now finished, and we turn to the
proof that (ii) =+ (i).
Let L be a directly indecomposable complemented modular algebraic lattice
with at least three elements. Let P be the set of points (atoms) of L, and let A be
the set of al1 sets of the form ¡ = {p E P: p I1) with 1 a line in L. Put r = (P, A).
We shall prove that r is a projective geometry and that L E L'. The function 4
that will turn out to be an isomorphism is defined thusly:
1 (22) 4(x) = { ~ E Pp :S x) for X E L .
We begin our argument by recalling that L is atomistic and relatively
complemented (Lemma 4.83). From this it follows that every line of L is the join
of two points. From the proof of Theorem 4.86, the direct indecomposability of
L now implies that each of its lines contains at least three points; this clearly
yields that r satisfies 4.8 l(2). Axiom 4.8 l(1) follows easily from considerations of
the height of elements. In fact, p v q is the unique line in L containing distinct
points p and q. This fact also implies that #(x) is a subspace of r (i.e., a member
of L ~for) every x E L. TOget 4.81(3), we need
l (23) For any X S P, the subspace C(X) spanned by X

is identical with ~ ( V X ) .
Indeed, it is clear from the last remark that C(X) c ~ ( V X )For
. the reverse
inclusion, note that every element of ~ ( V X is ) compact in L and must be
dominated by a join of finitely many elements of X. If there exists such an element
lying outside of C(X), we can pick p E P, p S x , v v x, (with x,, - - ,x, EX),
p$C(X) where n is as small as possible. Clearly, n 2 3. Then we can take
x = X, v m v x,-~SO that p x v x,, p # x,, and it follows from Lemma 4.84(ii)
that there exists a point q satisfying q x, p 5 q v x,. By the choice of p

(minimality of n), we have that q E C(X); certainly x, E C(X). This implies that
p€C(X), since this set is linearly closed, which gives us a contradiction. Thus
(23) is established.
Using (23),we find that 4.81(3) for r is directly equivalent to the second part
of Lemma 4.84 applied to L. Thus 4.81(3) holds, and the proof that r is a
projective geometry is finished.
Using (23) again, we easily see that 4 is a meet-preserving mapping of L onto
L'. 4 is one-to-one because L is atomistic. Thus 4 is an isomorphism. •
Lattice theorists frequently use the term projective geometry to denote a

lattice satisfying the equivalent conditions of Theorem 4.87. It follows from
Theorem 2.72 that the finite dimensional projective geometries are simple lattices.
It is a good exercise to look for a more direct proof of this fact.
THEOREM 4.88. Every complemented modular lattice of finite height is iso-

morphic to a direct product of a finite Boolean lattice and finitely many finite
dimensional projective geometries.
Proof. Let L be a complemented modular lattice of finite height. L is certainly

algebraic. (See Exercise 4.77(9).) By the previous two theorems, L is isomorphic
to a product of two-element lattices and projective geometries. The number of
factors in this product and the dimensions of the geometries are bounded by the
height of L. The two-element lattices in the product can be grouped together;
their product is a finite Boolean lattice. ¤
The direct product representations provided by Theorems 4.86 and 4.88 are
essentially unique. (See the paragraph following the theorem in the next section.)
Exercises 4.89
1. Verify that projective geometries of projective dimension 1 are the same as
the lattices M,, rc > 3. (See the first example following Definition 4.81.)
2. Let n: = (P, A) be a projective plane. For every p E P let p = (1 E A: p E 1). Let
pa = A and = {p:p E P). Show that na = (pa,
(It is called the dual plane of n). Show that
2
E L' .
is a projective plane.
3. Show that in a projective plane, every point fails to lie on some line, and
every line fails to contain some point.
4. Let n: be a projective plane. Prove that there is a cardinal rc 2 2 (called the
order of n) such that every line of n: contains precisely rc + 1 points; every
point of n: lies on precisely rc + 1 lines; and n: has precisely rc2 + rc + 1 points
and the same number of lines.
5. Prove that, up to isomorphism, the Fano plane of Figure 4.7 is the only
projective plane of order 2.
1
I
6. Show that projective planes can be defined by the axioms 4.78(1), 4.78(2),
and this axiom: There exist four points, no three of which are collinear.
7. Prove that every projective geometry satisfies Pasch's Axiom: Let lo, l,, 1,
be three distinct lines, each pair of which intersect. Let y be a point on 1, that
does not lie on 1, or l,, and let q be a point on 1, that does not lie on 1, or
1,. Then the line joining p and q intersects 1,. (In words, every line intersecting
two sides of a triangle and not passing through a vertex must intersect the
third side of the triangle.)
8. Let r be a projective geometry and let S be a linearly closed subspace con-
taining at least two points. Let A' denote the set of lines of I'included in S.
Prove that (S, A') is a projective geometry.
9. Let r be a projective geometry. Prove that the number of points on a line in .
r is a constant, independent of the line, and that the number of lines through
a point is a constant. Show that these constants need not be equal.
10. Let V be a finite dimensional vector space over a field f. Let L = SubV be
the lattice of subspaces of V. Verify that L is a finite dimensional projective
geometry. Prove that L S La. (We outline a proof and ask the reader to fill
in the details. Let vabe hom(V, f), where f is regarded as a one-dimensional
vector space and the hom set is given the structure of a vector space in the
obvious fashion. For u E V and or E va, define vRor to mean that a(u) = O. The
polarities associated with the relation R are isomorphisms between SubV
and (Sub Prove that dim V = dim va and conclude that V E va and
.-
L La.)
11. Does the result of Exercise 10 hold if f is only assumed to be a division ring?
12. Prove that every finite dimensional projective geometry is a simple lattice,
without using Theorem 2.72.
13. Let L = Sub V where V is an infinite dimensional vector space over a division
ring. Show that L is not simple. Show that L' is not an algebraic lattice, and
so L $ La.
14. Show that the lattice M,,, is a simple, noncomplemented modular lattice.
(See the examples at the end of $4.3.)
15. By a spanning n-frame (n 2 3) in a bounded lattice L is meant a system
(a,, - a,) of n + 1 members of L such that:
m ,
i. Vi,jai = 1 for a l l j 5 n;
ii* aj A (Viij,kai) =O for al1j # k (j, k 5 n).
Show that a spanning 3-frame in a projective plane is just a set of four points,
no three of which are collinear. Prove that every n-dimensional projective
geometry has a spanning n + 1-frame. (The concept of an n-frame has become
quite important in the study of general modular lattices.)
16. Let (a,, a,) be a spanning n-frame in a modular lattice L. Show that
S ,
a,, , a,-, generate a 2"-element Boolean sublattice of L.

17. Let R be any ring with unit and M = &-i.e., Rn construed as a left R-
module. Show that L = Sub M is a modular algebraic lattice with a spanning
n-frame. Find an example of R such that Sub RR3fails to be complemented.
4.9 DISTRIBUTIVE CONGRUENCE LATTICES
An algebra is called congruence distributive iff its congruence lattice is a distributive

lattice. For example, if
(24) A has a term operation t(x, y, z) that obeys the laws
t(x, X, Y)25 t(x, Y , x) t(y, x, x) x
then A is congruence distributive. In $2.5,Theorem 2.50, we saw that every lattice
is congruence distributive. The proof given there, based on the fact that the
operation M(x, y,z) = (x A y) v (y A z) v (z A x) obeys the above laws, really
demonstrates that every algebra satisfying (24) is congruence distributive.
A simple algebra is congruence distributive, therefore a congruence distri-
butive algebra need not satisfy (24). But most of the classes of congruence
distributive algebras we shall meet do consist of algebras satisfying (24)-in fact,
algebras that have the operations of a lattice among their term operations.
With regard to direct and subdirect representations, congruence distributive
algebras are extremely well-behaved. Such an algebra has essentially at most one
irredundant representation as a subdirect product of a finite system of subdirectly
irreducible algebras, and essentially at most one representation as a direct
product of directly indecomposable algebras. Here is the precise result on
subdirect representations.
THEOREM 4.90. Let A be un álgebra whose congruence lattice is distributive.

For any two irredundant subdirect representations of A with a finite number of
subdirectly irreducible factors, given by systems of congruences (4, ,. 4,) and
S ,
(S1,...,S/n), we have m = n and, after renumbering, 4i = S/i for 1 5 i 5 n.

Proof. The proof is the same as the proof of Theorem 4.73, relying on Theorem
2.55 instead of Theorem 2.33. M
The precise resült on direct representations is surprisingly strong (it requires

no finiteness assumption): A congruence distributive algebra has at most one
set of factor congruences yielding a direct representation with directly in-
decomposable factor algebras. This result is Corollary 1 to Theorem 5.17 in $5.6.
Although the result is not difficult to prove, it lies somewhat deeper than
Theorem 4.90. This result becomes even more surprising when we observe that
congruence distributive algebras in general admit no purely lattice-theoretical
characterization for the systems of factor congruences that yield direct repre-
sentations. For example, the three-element lattice is directly indecomposable, yet
its congruence lattice is a four-element Boolean lattice.
The varieties in which al1 algebras are congruence distributive played an
important role in the evolution of the theory of varieties and are still being
intensively studied. Lattices and most of the varieties that arise out of logic,
such as Heyting algebras and monadic algebras, have this property. In $4.12,
Theorem 4.144 furnishes a characterization of congruence distributive varieties
in terms of equations. Much space is devoted to this family of varieties in
Volumes 2 and 3.
Exercises 10 and 16 in $4.4, and Theorem 4.104 and Exercise 1 in $4.10, are
relevant to this section.
l 4.10 CLASS OPERATORS AND VARIETIES
A function mapping classes of algebras (al1of the same type) to classes of algebras
(of the same type) will be called a class operator. Three important class operators
were defined in $1.2. We repeat those definitions and add two more.
1 DEFINITION 4.91. Let X be a class of similar algebras.
I A E Z(X) iff A is isomorphic to some member of X .

A E H ( X ) iff A is a homomorphic image of some member of X .
A E S ( X ) iff A is isomorphic to a subalgebra of some member of X .
A E P ( X ) iff A is isomorphic to a product of a system of algebras in X .
I A E P,(X) iff A can be subdirectly embedded into a product of a system of

algebras in X .
Each of the class operators Q just introduced, when restricted to classes of
algebras of one similarity type, is order increasing (that is, Q ( X ) 2 X for al1
X ) , order preserving (that is, Q(Xo) c Q(Xl) if Xo G X,), and idempotent (that
is, Q(Q(X)) = Q ( X ) for al1 classes X ) .Thus the operator can be regarded as a
closure operator on the class of al1 algebras of one type.
If 0, and 0, are class operators, we write O, 0,for the composition of the
operators. We use < to denote this partial ordering of class>operators:0,I 0,
iff O, ( X ) s 0 2 ( X )for al1 classes X . Notice that Q ( X ) is always an abstract class
of algebras (if X is a nonvoid class of similar algebras and Q is one of the five
operators introduced above). In other words, if an algebra A belongs to the class,
then every algebra isomorphic to A also belongs. This fact can be written
symbolically in the form Q = ZQ. It is also true that Q = QI. Notice that P ( X )
and P,(X) always include the one-element algebras of the type of algebras in X ,
since we allow the product and subdirect product of an empty system of algebras.
(This implies that, strictly speaking, we have not actually defined either of these
classes if X is empty, unless a type is specified.)
LEMMA 4.92. The class operators HS, SP, HP, and HP, are closure operators
on the class of algebras of a type. We have: SH < HS, PS < SP, PH 5 HP and
P , H < H P , , P , P = P , = PP,,P,S=SP=SP,.
Proof. It is readily seen that the class of order increasing and order preserving
class operators is closed under composition. Moreover (restricted to this class of
operators), composition is an associative operation and preserves the order 5 ,

The idempotence of the four listed composite operators will follow easily from
these facts and from the listed inclusions. For example, given that SH i HS, we
can calculate:
(HS)(HS) = H(SH)S i H(HS)S = HHSS = HS;
while on the other hand,
HS = (HI)(IS) 5 (HS)(HS).
We shall prove one of the inclusions in our list, namely PH 4 HP, and leave
the others to the reader. To do so, let X be a class of similar algebras and suppose
that A E PH(X). Thus there exists a system of algebras and an isomorphism
a:ni,, A, E A where the algebras A, al1 belong to H(X). By the Axiom of
Choice, there is a system of algebras B , E X and onto homomorphisms fi:
B, ++ A,. Then the formula
f((b,:i E 1)) = (f,(b,):i E 1)
defines a mapping f from ni.,
Bi into ni,,
A,. It is easy to see that f is an onto
homomorphism of the product algebras. (Here again the Axiom of Choice must
be invoked.) Then af is a homomorphism from n i , , B i onto A, showing that
A E HP(X). ' m
We say that a class X is closed under a class operator Q iff Q ( X ) c X .
DEFINITION 4.93. A class X of algebras of type p is called a variety iff X

is closed under H, S, and P.
A class X is a variety iff the basic algebraic constructions of forming

homomorphic images, subalgebras, and products can always be carried out
within X . We shall often be interested in the smallest variety containing a given
class of algebras or containing a given algebra. Since the class of al1 algebras of
a type p is a variety, and since the intersection of any family of varieties of algebras
of type p is again a variety, it is intuitively clear that there does exist a smallest
variety containing a given class of algebras of type p. (We say "intuitively"
because dealing rigorously with families of classes and operators on classes, in
axiomatic set theory, requires special care. Ignoring the requirements of formal
rigor, we now introduce a special symbol for an important closure operator. The
next result shows that it can be defined in a quite legitimate way.)
DEFINITION 4.94. Let X be a class of similar algebras. V(X) denotes the

smallest variety containing X , called the variety generated by X . If X consists
of a single algebra A, or of finitely many algebras A,, A,, - , A,, then we write
V(A) or V(A, , ,A,), respectively, for this variety.
THEOREM 4.95. V = HSP.

4.10 Class Operators and Varieties 221
Proof. It follows readily from Lemma 4.92 that HHSP = SHSP = PHSP =
HSP. Thus H S P ( X ) is a variety containing X , for every class X . On the other
hand, if V is a variety containing X , then HSP(X) c HSP(V) = V. Thus
HSP(X) is the smallest variety containing X . m
Most of the classes of algebras we deal with are varieties. An exception is

the class of algebraic lattices, which is not closed under H or S. The classification
of algebras into varieties and the comparative study of varieties are basic themes
in these volumes. It can readily be seen that any class of similar algebras defined
by equations is a variety. This remark applies to al1 the classes introduced in
§§1.1 and 4.5. Conversely, every variety is the class of al1 models of a set of equa-
tions (the HSP Theorem of G. Birkhoff). This we prove in the next section.
A good part of basic algebra can be developed within any variety. In
developing the elementary theory of groups, for example, one rarely has occasion
to consider any algebra that does not belong to the variety of groups. The next
result, a variety-specific version of Theorem 4.44 (Birkhoff's Subdirect Repre-
sentation Theorem) illustrates this fact.
THEOREM 4.96. If X is a variety, every member of X is isomorphic to a

subdirect product of subdirectly irreducible members of X . In symbols, X = P,(Xsi)
where Xsiis the class of subdirectly irreducible rnernbers of X .
Proof. This follows immediately from Theorem 4.44, since in a subdirect repre-
sentation of A, al1 the factors are homomorphic images of A and must belong to
any variety to which A belongs. •
We have already seen severa1 applications of Theorem 4.96 in 54.5. The

subdirectly irreducible members of the varieties of Boolean algebras, semilattices,
distributive lattices, monadic algebras, and rings obeying a power identity were
determined and found to be quite special, yielding useful structural representation
theorems for al1 the algebras in these varieties. Each of these varieties is locally
finite, according to the next definition.
DEFINITION 4.97. An algebra A is locally finite iff every finitely generated

subalgebra of A is finite. A class of algebras is locally finite iff each of its members
is a locally finite algebra. A variety V is finitely generated iff V = V ( F )for some
finite set 9of finite algebras.
The theory of varieties demands a familiarity with free algebras and Birkhoff's
HSP Theorem, which we present in the next section. Some important results,
however, can be proved without the use of these tools. For example, the next
lemma permits us to prove that every finitely generated variety is locally finite.
Recall from 54.1 (specifically Exercise 4.9(1)) that the set of n-ary term operations
of an algebra A forms a subalgebra Clo, A of AA".
LEMMA 4.98. Suppose that A is un algebra, B E V(A), and B is generated by n

or fewer elements, where n is a positive integer. Then B is a homomorphic image
of Clo, A.
Proof. By Theorem 4.95, B is a homomorphic image of a subalgebra of AXfor

some set X . Choosing elements v,, v,-, in
m.., that map onto a set of
generators of B, we find that B is a homomorphic image of the subalgebra C
generated by these elements. It suffices to show that C is a homomorphic image
of Clo, A. To this end, we consider a mapping of the set of al1 n-ary operations
on the set A, into defined like this:
For g E AAn,F(g) = (g(vo(x), .,v,-, (x)): x EX).
It is trivial to verify that F is a homomorphism of AA"into AX,and that where
pf is the i + 1-st n-ary projection operation on A (i < n) we have F(pf) = u,. From
these facts, Theorem 4.15(i), and Exercise 4.9(1), it follows that F restricted to
Clo, A maps Clo, A onto C. m
THEOREM 4.99. A finitely generated variety is locally finite.
Proof. Suppose that Y'" = V(A,,. . A,) where A,,

S , A, are finite. Letting
S ,
A = A, x x A,, we see that Y'" = V(A). By the lemma, an algebra in Y

generated by n elements has its cardinality bounded by the cardinality of Clo, A.
Thus Y'" is locally finite, by Definition 4.97. ¤
We introduce a useful class operator for the study of locally finite varieties.
DEFINITION 4.100. Let X be a class of similar algebras. Pf ,,(%) is the class

of algebras isomorphic to a product of a finite system of members of %. Kinis
the class of finite members of X .
LEMMA 4.101. (SP(X)),,, c SPf,(A?).
Proof. Suppose that A is a finite algebra in SP(X). Thus we can assume that
Ac ni,, A, with A, E X . There clearly exists a finite set of indices I, c I such
that whenever f, g E A and f # g, then for some i~ 10,we have f (i) # g(i). For f
homomorphism of the full product onto the restricted product ni.,

in the product algebra, let d(f) be the restriction f lIo of f to I,. Then q5 is a
A,. We have
that 41, embeds A into the product of the finite system (A,: i E I,). This proves
that A E SPf,,(X). m
COROLLARU. Suppose that X is a finite set of finite algebras of the sume

type and Y'" = V(X). Then "ifin = HSPfin(X).
Proof. Let A be any finite algebra in Y'". There exists an algebra B in S P ( X )

and a homomorphism 4 of B onto A. By choosing preimages of the elements of
A, we arrive at a finitely generated algebra C c B such that $(C) = A. Now C is
finite (Theorem 4.99), so C E SPf,(X), by Lemma 4.101. This gives the desired
conclusion, that A E HSPf,(X). ¤
4.10 Class Operators and Varieties 223
The next lemma is implicitly used quite often in the study of varieties.
LEMMA 4.102. Let X be a class of algebras of type p and A be un algebra of
the sume type.
i. A E S P ( X ) 8 there exists a system of congruences pi (i E I) of A such that:
a. OA = n i E I P i ,and
b. A/P,ES(X)for al1 i.
ii. A E HSP(K) iff there exists un algebra B and congruences P, Pi (i E I) of B
such that:
a. BIP r A;
b. fl 2 niG1Bi; and
c. BIDi E S ( X )for al1 i E I.
iii. A E HSPfin(X)g t h e r e exists B, P, Di (i E I) satisfying (ii(a))-(ii(c)) and 1
is finite.
Proof. We ask the reader to supply the proofs. They are al1 easy or trivial. Use
the isomorphism theorems of 54.2 and the observations on the first two pages
of 54.4. m
DEFINITION 4.103. A variety is called congruence distributive (or congruence

modular)iff each algebra in the variety has a distributive (or modular) congruence
lattice.
We briefly discussed congruence modular and congruence distributive alge-

bras in $54.7 and 4.9. Varieties composed entirely of such algebras are frequently
encountered, and the systematic study of these varieties has led to a substantial
body of results. These results will be developed especially in the next two volumes,
but interesting results will be proved from time to time as they become accessible.
Here are two such results.
THEOREM 4.104. If V(X) is congruence distributive, then HSPfin(X)=
S e i n H S ( X ) If
. V(X) is congruence distributive and locally finite, then every finite
subdirectly irreducible algebra in V(X) belongs tu HS(X).
Proof. Let Y = V(X) and assume that Y is congruence distributive. It is easy

to see that PfinH2 HPfinand &,S 5 SPfin,which, combined with Lemma 4.92,
yields SqinHS5 HSPfi. To see that the other inclusion holds for this X , let
AE HSPfin(X).Choose B and a congruence fl and finite set of congruences
n
(Di: i E 1) of B to satisfy Lemma 4.102(iii). Let 6 = (Di: i E 1). The algebra B/6
belongs to Y , so by Theorem 4.12 the interval lattice I[6, l,] contained in Con B
is distributive. Then, using the distributivity of this lattice, we can calculate that
since the meet is finite. Then by the isomorphism theorems of 54.2, A r BID
can be subdirectly e e d d e d into ni.
B/(P v Bi), and the algebras B/(B v Bi)
belong to HS(X). This concludes the proof of the first assertion.
Now suppose that Y is also locally finite as well as congruence distributive. ,
Let A be a finite subdirectly irreducible algebra in V .Then A is a homomorphic

image of some finitely generated algebra B E SP(X). B must be finite, and thus
by Lemma 4.101 we have that B ESP~,,(X).From the statement proved above,
we can now conclude that A ESPHS(X); and then the subdirect irred~cibilit~
of A yields A E HS(X), as required. ¤
THEOREM 4.105. Suppose that A E HS(A, x A,) where V(A,, A,) is con-
gruence modular and A is subdirectly irreducible. There are algebras Bi E HS(Ai)
such that A is a homomorphic image of a subdirect product of B, and B, and each
B, is a one-element algebra or subdirectly irreducible.
Proof. There is an algebra B A, x A, and a congruence /? of B such that

A z BIP. We use /?, and /?, to denote the kernels of the projection homo-
morphisms of the product, restricted to B. Thus B, A /?, = 0, and B/pi E S(Ai).
The lattice L = ConB is modular, and since it is algebraic, it is join con-
tinuous. (See Exercise 4.64(4) at the end of $4.6.) Two applications of Zorn's
Lemma, using the join-continuity, yield the existence of a congruence y, that is
a maximal member of
(x con B: x 2 /?,
and x A /?,
5 /?),
and then a congruence y, that is a maximal member of
x :2 B, and y,
{xECO~B A x i p).
Thuswehavethaty,~y,1~,andfori~{0,1)if~>q~,then~~y,-~$~
Now Bi = B/yi E HS(Ai)since /?¡i y¡. A is a quotient of the algebra B/(y, A y ,)
(since y, A y, 5 /?), and this algebra is isomorphic to a subdirect product of
B, and B,. What remains to be proved is that each of the algebras B/yi is a
one-element algebra or is subdirectly irreducible-in other words, that y, = 1,
or vi is a strictly meet irreducible element of L. (See Lemma 4.43.) If y, = l,,
then y, 5 P, SO q1 = /? (by the maximality), and since /? is strictly meet irreducible,
we are done. Likewise, if q, = 1,, we are done. So we shall assume that neither
of these equalities holds, or equivalently, that y, $ fl (i = 0,l). Now observe .
that y, A (y, v (/? A y,)) 5 fl by modularity. Thus the maximality gives y, =
y, v (p A y,), so that p A y, 5 y,. This argument, which works also with y,
and y, interchanged, shows that y, A y, = /? A q, = /? A y,. The situation is
pictured in Figure 4.8.
Figure 4.8
4.10 Class Bperators and Varieties 225
Let b be the smallest congruence strictly larger than 8. Since y, $ P, we have

b a B v y,, and then by modularity,
Thus the interval I [P, b] is perspective with I [P A q,, fi A y ,], and it follows
that B A y l < b A y,. Now we put q0 = yo v (b A y,). Since q, A (b /\ y , ) =
p A y,, it follows by perspectivity that yo < q,.
We shall prove that 4, is the smallest congruence strictly above y,. Let
us suppose that x > y,. Then x A y, $ P; just as above, we conclude that
= p v ( ) A x A yl),andb A x A 47, $ P.Thus
yielding that b A y, = b A x A y,, because of the covering noted previously.

Thus it follows that x 2 b A y, and then x 2 q,, as claimed. We have now proved
that y, is strictly meet irreducible. Clearly, we can prove the same thing for y,.
The proof is finished.
The class operators 1,H, S, and P generate an ordered semigroup (actually

a monoid) under composition. The structure of this semigroup was determined
by Pigozzi [1972]. Figure 4.9 reveals that it just barely fails to be a lattice-ordered
semigroup.
HSP
1
The ordered semigroup of class operators
Figure 4.9
Exercises 4.106
1. Prove that if A and B are finite subdirectly irreducible algebras and V(A) is
congruence distributive, then V(A)= V(B) iff A r B. (Use Theorem 4.104.)
*2. Prove that the variety of monadic algebras is congruence distributive and
locally finite. (Use Corollary 3 of Theorem 4.44,94.5.)
3. Prove that the variety of monadic algebras is not finitely generated. (Use
the description of subdirectly irreducible monadic algebras in 94.5, and
Theorem 4.104.)
4. Show that the variety of semilattices is not congruence modular.
5. Show that every congruence modular variety -tr of semigroups is a variety
of groups, that is, for some n > 1 the equations xny Ñ y Ñ yxnare valid in V .
6. Complete the proof of Lemma 4.92.
7. Prove Lemma 4.102.
8. Suppose that P, y,, y, E ConB and y, A y, I P. Prove that B / ~ E
HE(l31yo B/v 1
3
9. Prove this fact about modular algebraic lattices. If b > b, A A bnin L and
b is strictly meet irreducible, then there exist elementc ci E I [b,, 11 such that
b > c, A A c,, and this inclusion fails if any ci is replaced by a larger
element. Moreover, when this holds, then every ci satisfies: ci = 1 or ci is
strictly meet irreducible. Use the fact to formulate and prove a version of
Theorem 4.105 valid for n factors.
10. Show that SH # HS, PS # SP, and PH # HP.
11. Show that restricted to classes of commutative semigroups, the operators
SPHS, SHPS, and HSP are distinct. In fact, if X is the class of finite cyclic
groups considered as semigroups (i.e., multiplication groups), then the three
operators applied to X give different classes.
12. An ordered monoid is a system (M, ,e, I ) such that (M, e) is a monoid,
S ,
(M, I ) is an ordered set, and I is a subuniverse of (M, -,e)'. Prove that

if M is an ordered monoid generated by elements h, S,p satisfying e I h = h2,
eI s = s2, e I p = p2, sh I hs, ps I sp, and ph I hp, then M is a homo-
morphic image of the ordered monoid of class operators diagrammed in
Figure 4.9. If also hsp # shps and hp $ sphs, then show that M is isomorphic
to the ordered monoid of class operators.
13. Verify that if X is a class of Abelian groups then HS(X) = SH(X). For-
mulate a property of varieties involving the behavior of congruences such
that if -tr has the property and X E -Ir, then H S ( X ) = SH(X).
14. Prove that if X is a class of Boolean algebras, or distributive lattices, then
HSP(X) = SP(X).
4.11 FREE ALGEBRAS AND THE HSP THEOREM
In this section we introduce free algebras, terms, algebras of terms, and equations,
and prove that every variety of algebras is defined by a set of equations (the HSP
Theorem). These are essential tools for the study of varieties.
We begin with a definition of free algebras and a series of easy results about
them.
4.11 Free Algebras and the HSP Theorem 227
DEFINITION 4.107. Let X be a class of algebras of one type and let U be an

algebra of the same type. Let X be any subset of U. We say that U has the
universal mapping property for X over X iff for every A E X and for every
mapping a: X -,A, there is a homomorphism P : U -,A that extends a (i.e.,
P(x) = a(x) for x EX). We say that U is free for X over X iff U is generated by
X and U has the universal mapping property for X over X. We say that U is
free in X over X iff U E X and U is free for X over X. If U is free in X over X,
then X is called a free generating set for U, and U is said to be freely generated
by X.
LEMMA 4.108. Suppose that U is free for X over X and that A E X . Then
hom(U, A) is in one-to-one correspondence with the set AXof mappings. For every
a: X + A, there is a unique extension of a to a homomorphism P: U -+A.
Proof. A homomorphism is completely determined by how it maps a set of

generators; see Exercise 4.16(1). m
LEMMA 4.109.
i. Let & c X,. If U is free for X, over X, then U is free for Xoover X .
ii. If U is free for X over X, then U is free for HSP(X) over X.
Proof. Statement (i) is trivial. Ta prove (ii), one shows that when U is free for
X over X, then U is also free over X for each of the classes H(X), S(X), and
P(X). We prove this now for H(X), and leave the rest for the reader to work out.
Suppose that U is free for X over X. Let A E H(X); say B E X , and we have
a surjective homomorphism 6: B + A. Let a be any mapping of X into A. We
l
have to show that a extends to a homomorphism of U into A. By the Axiom of
Choice, there is a mapping &: X -,B such that &(x)E 6-' {a(x)} for al1 x E X. By
the universal mapping property of U for X over X, there is a homomorphism
):U -+B extending 2. Let P = 6 0 /?.This is a homomorphism of U into A. For
X E Xwe have
1 Thus P is the required extension of a. i

1 LEMMA 4.110. Suppose that U, and U, are free in X over X, and X,,
I respectively. If (X,1 = 1 X, 1 then U, U,.
Proof. Let 1 X, 1 = 1 X, 1, say f is a bijective mapping of X, onto X,.Then there

exist homomorphisms P: U, -+ U, and y : U, -,U, extending f, and f - l . y o
extends f - ' o f = idxl; by Lemma 4.108, it can only be the identity homo-
morphism of U, onto itself. Likewise, we conclude that P o y is the identity on
U,. It follows that P is an isomorphism of the two algebras and y is the inverse
isomorphism. m
DEFINITION 4.111. Let X be a family of algebras and A be an algebra, al1

of the same type. We define the congruence O ( X ) on A by
LEMMA 4.112. Suppose that U is free for X over X. Then U = U/O(X) is

free for X over 2 = {x/O(X): x EX}, and belongs to SP(X). Thus Ü is free in
V(X) over 2.
Proof. That U belongs to S P ( X ) follows directly from Lemma 4.102(i) of

the last section and the isomorphism theorems of 54.2. Let fi be the quotient
homomorphism of U onto U/@(%). Clearly, n(X) ( = X) is a generating set of Ü.
To see that U has the universal mapping property for X over X, let a: 2 -+ A
be any mapping, where A E X . Let /? be the unique extension of an to a homo-
morphism of U into A. Thus
and so we have that O ( X ) E ker p by the definition of O ( X ) . By the Second

Isomorphism Theorem (Theorem 4.10), there is a homomorphism E: U --+ A for
which P o n = p. This is the desired homomorphism, for if % = x/O(X) belongs
to x (where x E X), then
B ( 3 = B(x(x)) = P(x)
= a(n(x)) = a(%).
Now by Lemma 4.109(ii), U has the universal mapping property for V ( X )

over X. Since it belongs to V(X), it is free in V(X) over X.
One of our goals in this section is to show that there exists an algebra free
in V over X where X is any nonvoid set and V is any variety that has a nontrivial
member. To show that these algebras exist, we shall construct them. By Lemma
4.110, when we succeed in constructing one algebra free in V over a set of a
certain cardinality, we can be assured that every algebra free in Y over a set of
the same cardinality is isomorphic to it. Our first task will be to construct an
algebra free in X over X where X is the class of al1 algebras having the type of
algebras in V. Such an algebra is called absolutely free and will automatically
be free for V . Suppose that we succeed in constructing an absolutely free algebra
U of the given type, freely generated by X. By Lemma 4.112, U/O(Y) will be free
in V over X. Since V has an algebra with more than one element, it follows
that x 4 y/O(V) when x and y are distinct elements of X. Thus we have an
algebra free in V over a set 2 of cardinality equal to that of X. A trivial set-
theoretical argument then produces an algebra free in V over X. (See the proof
of Theorem 4.117.)
Lemma 4.98 of the last section will motivate our procedure. The proof of
that lemma shows that for any nontrivial algebra A and positive integer n, the
algebra Clo,A of n-ary term operations of A is a free algebra in V(A) over an
n-element free generating set. Underlying the concept of term operation is the
concept of term, which we now define for any type of algebra.
l 4.11 Free Algebras and the HSP Theorem
In the next few paragraphs, o denotes a similarity type (as defined a t the
beginning of §4.2), 1 is the correlated set of operation symbols, and X, is the
class of al1 algebras of type o. Let X be a set disjoint from I. (Later we can
remove this assumption, but in the beginning it is crucial.) For O < n < o,let
I, = {Q E 1: o(Q) = n) (the set of n-ary operation symbols in I).
By a sequence from X U I we shall mean a finite sequence (S,, ,S,-,}
whose ith member si belongs to X U I for every i < n. Such a sequence will also
be called a word on the alphabet X U I and written as s,s, . S,-, . The product of
two words a = a,. . a,-, and b = b, - . bm-, is the word ab = a, a,-, b, . bm-,
(i.e., the sequence (e,, e,-,) in which k = n + m and ci = ai for i < n, and
e ,
= bi for i < m). Note that in this usage, when we write "the word u" (where
u E X U 1), we mean the sequence (u).
DEFINITION 4.113. The set T,(X) of terms of type o over X is the smallest
set T of words on the alphabet X U I such 'that
l. XUI,ET.
2. If p,, m , p,-, E T and Q E I,, then the word Qp,p, p,-, E T.
We adopt some conventions for informally denoting terms in print. If
p,, e ,p,-, are terms (of type o over X) and Q is an n-ary operation symbol
(of type rr), then we often denote the term Qp, p,-, by Q(p,,. ,p,-,). If is a
binary operation symbol, we often denote (p,, p,) by p, p, . Parentheses will
be used freely to make terms more readable. For example, if x, y, z E X and E I,,
then x yz, xyz, and * xy zx are terms that we may write as x (y z), (x y) z,
(x * y) (z x), respectively. Note that T,(X) is the empty set iff X U I, is empty.
DEFINITION 4.114. If T,(X) # @, then the term algebra of type o over X,

denoted T,(X), is the algebra of type o that has for its universe the set T,(X)
and whose fundamental operations satisfy
I for Q E 1, and p, E T,(X), O 5 i < n.
The term algebra T,(X) is free in jya over the set X of words (or more
precisely, over the set ((x): x EX)). This will follow from the fact that given any
term p there is exactly one way to reach p by starting with elements of X and
applying repeatedly the operations. The fact is called the unique readability of
terms. The next lemma gives a precise formulation of this fact.
I LEMMA 4.115.
i. T,(X) is generated by the set X of words.
ii. For Q E I,, ~ ' 0 ' ~ ) is a one-to-one function of (T,(X))" onto a subset of
T,(X) - X.
Proof. Statement (i) is immediate from Definition 4.113. From (i) it follows
that every member of the algebra either belongs to X or belongs to the range of
at least one of its basic operations. It is also immediately clear that the range of
each basic operation is disjoint from X. (Here we need the assumption that
x n I, = a.)
What now remains to be shown is equivalent to this statement:
= 1
(26) If Q,, Q, E I and Q > ( ~ ) ( ~- ,.,, (qo, ., qm-l), then
~ ~ 0 " )
e
Q, = Q, (so m = n) and pi = qi for al1 i < n.

A detailed proof of (26), and a simple algorithm to recognize whether a word is
a term are outlined in Exercises 1-4 at the end of this section. •
LEMMA 4.116. T,(X) has the universal mapping property for Xoover X.
Proof. Given a term p, define l(p) to be the length of p. (Recall that a term is
a sequence. The length of a sequence (S,, - S,-,) is n.)
S ,
Let A be any algebra of type o and a be any mapping of X into A. Define

P(p) for p~ T,(X) by recursion on l(p). If EX, then put P(p) = a(p). Now if
p E T,(X) - X, then, by the previous theorem, there are Q E I and p,, . ,pn-, E
T,(X) (where n = o(Q)) such that p = Q(p,, ,p.-,); moreover, Q, p,, p,-,S ,
are al1 uniquely determined. We define P(p) = Q*(D(~,), P(pn-,)), and note S ,
that since I(p,) < l(p), this serves to define P(p) recursively for al1 p. P is clearly a
homomorphism extending a. ¤
We can now state our principal result on the existence of free algebras.
THEOREM 4.117. Let o be a similarity type of algebras and I be the associated

set of operation symbols. Let X be a set such that 1x1> 1 if 1, = @. Suppose that
Y is a variety of algebras of type o.
i. If X í l I = @, then the term algebra T,(X) is an absolutely free algebra of
type a, freely generated by the set ((x): x E X).
ii. If X f l I = @, then T,(X)/O(Y") is free in Y" over the set { (x)/O(Y): x E X).
iii. If Y has a nontrivial member, then there exists an algebra free in Y over the
set X.
Proof. Statement (i) follows from the preceding lemma. Then statement (ii)
follows from Lemmas 4.109 and 4.112. Now suppose that Y" has a nontrivial
member. Then the map x H (x)/O(Y") is one-to-one. In set theory, we can find
a set F and a one-to-one map f of F onto T,(X)/O(Y") such that X E F and
f (x) = (x)/O(Y) for x EX. Since f is a bijection, there are unique operations on
F giving us an algebra F isomorphic to T,(X)/O(Y") under f. Then F is free
in Y" over X. ¤
DEFINITION 4.118. F'(X) denotes a free algebra in V(X) with free gener-
ating set X. Fx(rc), where rc is a cardinal number, denotes an algebra Fx(X)
where 1x1= rc.
I
Note that Fx(X), if it exists, is determined only up to an isomorphism that

is the identity on X. Fx(ic) is determined only up to isomorphism. Fx(0) exists
iff the type has O-ary operations. If TC # O, then Fx(lc) exists iff X has nontrivial
members or rc = l.
COROLLARY 4.119. Let X be a set and X be a class of algebras of type a

such that F,(X) exists. Then &(X) belongs to S P ( X ) and FUx)(X) z Fx(X) r
T,(X)/@ ( X ) .
Proof. This is a corollary of Lemmas 4.110 and 4.1 12 and Theorem 4.1 17. i
Absolutely free algebras (isomorphic to term algebras) have a rather trans-

parent structure, as we have seen. Explicit constructions of free algebras in the
variety of al1 groups and in the variety of al1 monoids are given in Chapter 3
(553.3 and 3.4). Free groups and free monoids are not very complicated; their
elements and operations are easily visualized. Free algebras in some classes,
especially in certain varieties of groups or of lattices, can be quite complicated,
or impossible, to describe. Severa1 free algebras that are relatively straight-
forward to construct are presented in examples and exercises at the end of this
l
1
section. Free algebras can be very useful, even where they cannot be readily
described. Many applications require only that they exist, satisfy the universal
mapping property, and bear a certain relation to the equations holding in a
variety, which we shall examine below.
We can now discuss the correspondence between terms and term operations.
Every term of type a can be regarded as a name for one or more term operations
in every algebra of type a. We could define term quite broadly, to mean any
element of an absolutely free algebra. There are, however, good and traditional
reasons for adopting a more restrictive definition, according to which the terms
of type o are the elements of one fixed absolutely free algebra with a denumerably
infinite free generating set. We wish to choose once and for al1 the set of free
generators, independently of the type. To this end we adopt a technical convention:
Henceforth we assume that the set 1of operation symbols of any type a is chosen
so that I í l co = @ (where co = {O, 1, ,n, S)).Then the following definition
makes sense for al1 types.
DEFINITION 4.120. Let a be a type. By a term (of type o)is meant an element
of the term algebra T,(co). We put v, = (n), and the terms un (n E co) are called
variables.
Terms of type o are elements of the absolutely free algebra T,(co) generated
by the set of variables {u,, v,, - ,u,, For each n 2 1, the term algebra T,(n)
a } .
is the subalgebra of T,(co) generated by v,, ,un-, . Note that

T,(w) = U T,(n) and
l<n<w
T,(l) c T,(2) .
232 Chapter 4 Fundamental Algebraic Res&
For every term p there is a unique smallest set of variables {x,, . xk-,} for S ,
which p belongs to the subuniverse of T,(o) generated by {x,, ,x,-,). These

are the variables vi such that i occurs in p. We say that p depends on vi or that vi
occurs in p iff vi belongs to the set {x,, - -. ,-xk-,). Then p E T,(n) iff every variable
p depends on is included among u,, a u,-, .
* ,
DEFINITION 4.121. Let A be an algebra of type o and p~T,(n).We define

an n-ary operation pA over A by induction on the length of p. If p = vi, then we put
that is, pA is the n-ary projection operation p: in this case. If qA has been defined
for al1 terms q E T,(n) of length less than that of p, then if p = Q(p,, - ,p,-,),
where Q is an m-ary operation symbol, we put
Recall that the n-ary term operations of an algebra A, and the algebra Clo, A
of n-ary term operations of A, were defined in 54.1. The next lemma contains
the fact that an n-ary operation f on A is a term operation iff f = PA for some
P E T,(n).
LEMMA 4.122. Let A be un algebra of type o and n E co be such that n > O if

o supplies no O-ary operations. The mapping p -+pA is a homomorphism of T,(n)
onto Clo, A.
Proof. The mapping can easily be seen to be a homomorphism of T,(n) into the
direct power AA"of A. Since v,, . . , v.-, generate T,(n), this homomorphism is
onto the subuniverse of AA" generated by the projections p$, p,"-,. The
m - . ,
theorem follows directly from Theorem 4.15 and Exercise 4.9(1) in 54.1. ¤
The members of T,(n) are called n-ary terms of type o, or n-ary a-terms.
Their usefulness should be apparent. The lemma tells us that an n-ary operation
f on the universe of an algebra A is a term operation of A iff f = pA for
some n-ary term p. We remark that according to our conventions, a term like
q = Q(v,, v,) (where Q is a binary operation symbol) can be considered to be
n-ary for every n 2 4 (since it belongs to T,(4)) but is not binary. This q is
essentially binary, however, since it depends only on the two variables u, and v,.
Thus we may by a simple substitution of variables (automorphism of T,(co))
convert it into the term Q(v,, u,), which names the expected term operation of
an algebra A (just the basic operation eA).Here are some easy and frequently
used facts.
LEMMA 4.123. Let A, B, Bi (i E I) be algebras of type o, and let p be un n-ary

o-term.
i. If a,, a , - , ~ A a n d f:T,(n)+A(or f:T,(co)+A)satisfies
s.., f(vi) = ai for
i < n, then pA(ao,---,an-l) = f(p).
4.11 Pree Algebras and the HSP Theorem
ii. p = pTu(")
( ~ 0 7 7 un-,).
iii. Iff : A 4 B and a,, un-, E A, then f (pA(ao,
s., a * , a,-,)) =
pB(f('O), f (U.-1 1)'
9
iv. Every subuniverse of A is closed under pA.

v. Let S be a subset of A. Then
A
Sg (S) = jqA(so, ., S,-,): S,, - 7 S,-, E S, q E 7-34, m < 0).
vi. If B = ni€B,, then (B, pB) ni€(B,, pBi).
=
Proof. To prove (i), suppose first that f : T,(n) + A. The projection of AA"onto
the (a,, a,-,)th coordinate gives a homomorphism q4 of Clo, A into A. We
S ,
have that q4(pr) = a, for al1 i < n. Thus 4 composed with the homomorphism
p-pA is a homomorphism of T,(n) + A that maps vi to a,, i < n-i.e., it is f.
This proves (i). (If instead f : T,(co) + A, then a consideration of the restriction
of f to T,(n) yields the desired result.)
Statement (ii) follows trivially from (i) through consideration of the identity
homomorphism on the term algebra.
To prove (iii) let g: T,(n) + A be the homomorphism with g(v,) = a,. Then
by (i)
Statements (iv) and (v) are immediate consequences of Exercise 4.9(7) and
Lemma 4.122.
To prove (vi), define a mapping of T,(n) into BBnby putting q* equal to the
operation of the algebra ni.,
{B,, qBi)for every n-ary term q. Show that q t+ q*
is a homomorphism of T,(n) into BBnthat takes vi to the ith projection operation
(for al1 i < n). Since the homomorphism q-qB has the same effect on the
variables, it follows that q* = qB for al1 q, which is the desired result. ¤
REMARKS 4.124. We shall frequently omit superscripts, writing p(a,, ,a,-,)

for pA(a0, ,a,-,). Thus if p is an n-ary term, we shall have p = pT~(n)(vo,~~~,vn-l)=
p(vo, ,un-,). Note that if x,, - ,xn-, are variables, then the o-terms that depend
on no variables other than these are precisely those of the form p(x,, ,xn-,),
p an n-ary a-term. (In this connection, see again Exercise 4.9(7).) We emphasize
that a term written as p(x,,x,,x,) (with the unwritten assumption that p =
p(v,, u,, v,) is a 3-ary term) may fail to be a 3-ary term, for instance, if p depends
on u, and x, = u,. Any term written as p(q,, - - - ,qn-, ) (where p is n-ary and
q,, a ,qn-, are terms) is the image of p under any endomorphism of T,(m) that
sends vi to q, for al1 i < n.
The most important application of our formal notion of term is in the

formalization of the concept of an equation holding in an algebra. We proceed
to define this concept. The associative law, x (y z) Ñ (x y) z, is an example of
what we have in mind. The associative law is said to hold in an algebra ( A , )
iff a (b c) = (a b) c for al1 a, b, c E A. One may observe that nearly every variety
we have thus far met was defined as the class of algebras in which some list of
specified laws, similar in kind to the associative law, hold. Notice that the
associative law is simply a formal statement (string of symbols) and cannot be
said to be true or false. It will hold (be true) in some algebras and will fail to hold
(be false) in others. The associative law is written in the form p N q where p and
q are terms, and the law holds in A iff pA = qA.Just as terms are taken to be
certain words over an alphabet, we shall regard equations as words over an
expanded alphabet, enriched by the addition of one letter N .
DEFINITION 4.125. An equation of type a is a word of the form p N q where

p and q are a-terms. Let A be an algebra and p N q be an equation (both of
type a) and suppose that {p, q} c T,(n). If Ü E A" then we say that ü satisfies p N q
in A iff pA(¿i)= qA(¿i).This relation is written
The equation p N q is said to be true in A iff pA = qA.This important relation

between an algebra and an equation is written
There are many ways to denote it in written and spoken language, for example:
A obeys the equation p N q; p N q holds (or holds identically) in A; p N q is
an identity of A; p N q is valid in A; etc. The equation p N q is said to be true in
a class X G X,, written
iff it is true in every member of X . If C is a set of equations of type a, we say that

C is true in X , written
2-l= E,
iff every member of C is true in X .
LEMMA 4.126. Let A be un algebra and p Ñ q be un equation (both of type a).

The following are equivalent.
i. A k p ~ q .
+
ii. If p, q E T,(n) then A, ¿piN q for al1 Ü E A".
iii. For al1 f ~hom(T,(co),A),f(p) = f(q).
iv. If p, q E T,(n) and x,, x,, xn-, are n distinct variables,
S ,
+
then A p(xo, xnV1)N q(xo, - ,x"-~).
a ,
Proof. Let n be any positive integer such that p, q E T,(n). The n-ary operations
PA and qA are equal iff f(p) = f(q) for al1 f : T,(n) + A (by Lemma 4.123(i)).
Since T,(co) is free, every homomorphism from T,(n) into A is the restriction of
a homomorphism from T,(o) into A; and so pA = qA iff f(p) = f(q) for al1
f : T,(co) -+ A. (This shows, moreover, that the condition is independent of the
specific choice of n, thereby putting our definition of the validity of an equation
1 4.11 Free Algebras and the HSP Theorem 235
on a sound footing.) The equivalence of the first three statements follows from
these remarks.
Now with n as above, suppose that x,, - . xn-, are n distinct variables.
m ,
There is an automorphism v of T,(co) satisfying v(p) = p(xo, - .. ,xn-,) and v(q) =

q(x,, - - ,x,-,). (Permute the variables, mapping vi to xi for i < n.) Thus p Ñ q
satisfies (iii) iff p(x,, ,x,-, ) Ñ q(x,, ,xn-, ) does. This proves the equivalence
of (i) and (iv), via the equivalence of (i) and (iii). •
The congruence O(%) on an algebra A determined by a family X of similar

algebras was defined in Definition 4.111. Here we use this concept to establish a
precise connection between the set of equations true in a class X and the free
algebras F'(X).
THEOREM 4.127. Given a class X of algebras of type o, terms p and q in T,(n),

a set X and distinct elements x,, . ,xn-, EX, the following are equivalent.
i. X k p ~ q .
ii. (p, q) belongs to the congruence O ( X ) on T,(co).
iii. If Fx(X) exists, then Fx(X) t= p Ñ q.
iv. If Fx(X) exists, then
Proof. Suppose that X t= p Ñ q. We wish to show that (p,q) belongs to

O(X). So let 8 E Con T,(co) be such that T,(co)/B E S(X). Thus 9 = ker f where
f : T,(co) A for some A EX.Since A k p Ñ q, we have that f (p) = f (q) by
-+
Lemma 4.126(iii), implying that (p, q) E 8. This proves that (p, q) E O(X); see
Definition 4.1 11. Thus (i) = (ii). That (ii) * (i) can be proved by reversing the
argument.
Now suppose that (i) and (ii) are true. We wish to prove (iii). By Corollary
4.119, it will be sufficient to show that p Ñ q is valid in T,(X)/O(X). Let
f : T,(co) -+ T,(X)/O(X). Let n denote the quotient homomorphism from T,(X)
onto T,(X)/O(X). Choose elements ti E T,(X) such that f (vi) = n(ti), and denote
by f the homomorphism T,(co) -r T,(X) defined by {(vi) = ti. Thus we have that
n 0 f = f. To prove that f(p) = f(q), it will suffice to show that <f(p),f(q)) E
O(%), the kernel of n. So let 8 E Con T,(X) be such that T,(X)/8 E S(X). Thus
8 = ker h where h is a homomorphism into some algebra in X. Since X p Ñ q,
we have that hf (p) = hf (o-i.e., that (&), f (q)) E 8, as desired. This concludes
the proof that (ii) => (iii).
The implication (iii) = (iv) is trivial.
Lastly, suppose that (iv) holds. We can suppose that Fx(X) exists. (Reader:
Supply the proof that (iv) * (i) in case Fy(X) does not exist.) Let f : T,(co) -+ A
with A E Y .We have to show that f (p) = f (q). Since Fz(X) has the universal
mapping property for X over X, there is a homomorphism g: Fx(X) -+ A with
g(xi) = f(vi) for i < n. Let h be any homomorphism T,(co) -+ Fx(X) satisfying
h(vi) = xi for i < n. We have that f = g o h on T,(n) since these mappings agree
1
on the generators of T,(n). We calculate that
This completes our proof that (iv) * (i). •
The basic class operators preserve the validity of equations (as the next
lemma asserts). We give a slick proof below, using the results already accumulated,
but it will be worthwhile for the reader to look for a direct proof of these facts.
LEMMA 4.128. For any class X of algebras of one type, al1 the classes X ,
P(X), S(X), H(X), V(X), and < ( S ) have precisely the sume valid equations.
Proof. We can assume that X possesses an algebra with at least two elements.
Let X be an infinite set, and let Y be any one of the six classes. It is easily seen
from our remarks following Definition 4.118 that Fx(X) and Fy(X) exist. Then
Lemma 4.109 implies that F'(X) r Fy(X). Thus an equation p Ñ q holds in X
iff it holds in Y , by the equivalence of (i) and (iii) in Theorem 4.127. ¤
DEFINITION 4.129. Let X be a class of algebras of type a and C be a set of

equations of type o. We put
O ( X ) = ( p Ñ q: 3-+ p Ñ q ) and
Mod(E) = (A: A t= E}.
X is called an equational class iff X = Mod(T) for some set I' of equations of
type a. We say that X is defined by, or axiomatized by, r when this holds.
C is called an equational theory iff C = @(Y)for some class 9of algebras of type
o. When this holds, C is called the equational theory of Y .
+
O and Mod are the polarities of the binary relation A E between algebras
and equations of a given type. (See 52.2, Theorem 2.21, and Example 2.22(5).)
Thus Mod O and O Mod are closure operators, on the class of algebras, and on
the set of equations, of a type. Lemma 4.128 obviously implies that V(X) G
Mod O ( X ) for al1 X . The important HSP Theorem of G. Birkhoff [1935a] is
the assertion that the operators V and Mod O are equal. Thus we have a one-to-
one correspondence between varieties of algebras of a type, and equational
theories of that type. Before proving the HSP Theorem, we need another lemma.
LEMMA 4.130. Let X be a class of algebras of type o, A be un algebra of the

sume type, and X be a set such that 1x1
2 1 Al.
4.11 Free Algebras and the HSP Theorem
i. +
If X consists of trivial algebras, then A O ( X ) iff 1 Al = 1.
ii. +
If X has a nontrivial algebra, then A O ( X ) iff A E H(Fx(X)).
Proof. To prove (i),we observe that if X consists entirely of trivial, one-element

algebras, then x Ñ y is a valid equation of X . This equation holds in an algebra
A iff A is a one-element algebra; moreover, every equation is valid in a one-
element algebra. Statement (i) follows from these observations.
To prove (ii), assume that X has a nontrivial member, so that Fx(X) exists.
+
Since F'(X) E V(X), if A E H(F'(X)) then A E V(X) and A O(%) (by Lemma
4.128 and Definition 4.129). For the converse, suppose that A O(X). Let +
F = Fsn(X) be an absolutely free algebra over the set X of free generators.
Choose any surjective mapping a of X onto A and let P be the homomorphism
of F onto A that extends a. Let 6: F + F'(X) be the homomorphism that extends
the identity map on X. We intend to use the Second Isomorphism Theorem (4.10)
to conclude that A is a homomorphic image of F'(X). Al1 we need do is show
that ker(6) c ker(P).
Suppose that 6(u) = 6(v) where u, v E F. Let x,, e, x,-, be a finite system
of distinct members of X such that u and u belong to the subuniverse generated
by these elements. Then there exist n-ary a-terms p and q such that u =
p F( ~ ~ ; ~ ~ , ~ ~ = ~ q, F) (a~no ,d- -v. , ~ nTheequalityJ(u)
-l). = 6(v)isequivalent to
Thus p Ñ q E O ( X ) (by Theorem 4.127), and it follows that A +p Ñ q. Now the

desired equality P(u) = P(v) is equivalent to
which is a consequence of A +p Ñ q. This concludes the proof that ker(6) c

ker(P). m
THE HSP THEOREM (4.131). For any class X of similar algebras,

HSP(X) = Mod O(X). Thus X is a variety ijjf X is un equational class.
+
Proof. Let C = O(X). Now V(X) C by Lemma 4.128; hence V(X) S
ModO(X). On the other hand, Lemma 4.130 implies that every member of
Mod O ( X ) is a homomorphic image of a member of V(X), and thus Mod O ( X ) c
V(X). This establishes the statement of equality in the theorem. If X is a variety,
t hen
X = V(X) = Mod O(X),
and so X is an equational class. If X is an equational class, then it follows
directly from Lemma 4.128 that X is a variety. ¤
An important corollary of the theorem is that for every type a, the number
of varieties of algebras of the type is bounded by 2", where rc = 11)+ co is the
cardinality of T,(co). ( I is the set of operation symbols of type o.) This is true
because varieties are in one-to-one correspondence with equational theories.
Until now, we could not be sure that the varieties of algebras of a given type
could be put in one-to-one correspondence with any set. Another corollary is
that every variety is generated by one of its algebras.
COROLLARY 4.132. If "Y is a variety having a nontrivial member and 1x12

then "Y = V(Fv(X)).
Proof. By Theorem 4.127, we have @(Y)= @(FV(X)).Thus by the HSP

Theorem, "Y = HSP("Y) = HSP(Er(X)). •
The basic facts about varieties, free algebras, and equations can now be
summarized. Let "Y be a nontrivial variety (Le., one with a nontrivial member)
of the type o. "Y is generated by its free algebra 'r(~), which can be constructed
as a quotient of T,(o). Two terms p and q are identified by the quotient map
iff the equation p~ q is valid in V. In this way the equational theory of 9'-
determines 13r(~),and "Y itself can be defined as the class of al1 models of its
equational theory. The finitely generated free algebra Fv(n) (where n E O) is, in
like manner, a quotient of the term algebra T,(n). Two n-ary terms p and q are
identified in Fv(n) iff the term operations pA and q A are equal for every A E V .
DEFINITION 4.133. Let "Y be a variety and A E V . Suppose that n < CO,
and let x,, x,, xn-, be the free generators of Fv(n). If q ~ F v ( n )and
. . S ,
q = p(xo, . .,xn-,) with p E T,(n), then we put q A = pA.
REMARK 4.134. It follows by Lemma 4.126 and Theorem 4.127 that q A is well
defined. The mapping q t+ q A is the natural homomorphism of Fy(n) onto ClonA.
The elements of Far(n) will be sometimes called n-ary terms of "Y (although we
normally reserve this expression for the members of T,(n); o being the type of
"Y). We reserve the expression "n-ary V-term" for a slightly different concept.
DEFINITION 4.135. Eet V be a variety and n be a non-negative integer. By

an n-ary Y-term we mean a mapping t whose domain is the class Y ,such that
1. For every A E V , tAis an n-ary operation over the universe of A.
2. For al1 A, B E "Y and f E hom(A, B), we have f E hom((A, tA), (B, tB)).
THEOREM 4.136. Let V be a nontrivial variety of algebras of type o and n be

a positive integer. A mapping t with domain "Y is un n-ary Y-term iffthere exists
a term q E Fv(n) (equivalently, a term p E T,(n)) with tA = q A (or tA = pA)for al1
AEY.
Proof. For one half of the theorem, we know that every term induces a "Y-term;
see Lemma 4.123(iii). For the other half, let t be an n-ary Y-term. Since n > O
and "Y is nontrivial, Fv(n) exists. Let x,, . xn-, be the free generators of Fv(n)
S ,
and choose a term p such that

4.11 Pree Algebras and the HSP Theorem 239
Then for every A E "tr and a,, , a,-, E A, where f : Fv(n) -+A satisfies f (x,) =
ai (i < n), we have
tA(a0;..,a,-,) =f(tF~(n)(xo,-~~,~,-l))
= f (pF~'n'(xo7., x,-,))
A
= P (a,, ' 9 a,-,). 4
Free algebras in the variety of al1 algebras of a type are called absolutely
free. Free algebras in other varieties are termed relatively free. The elements of
a free generating set are called variously letters, variables, indeterminates, or free
generators. A constructive description of the free algebras in a variety Y is the
same thing as an effective description of the congruence @(Y)on T,(o) for which
Fv(o) r T,(o)/@(V), or of the equational theory of "tr, since (p, q) E @ ( Yiff)
p z q E @(Y).One approach to getting such a description is to look for a subset
of T,(co)comprised of precisely one element from each @(Y)-equivalenceclass.
The elements of such a set may be called reduced terms or normal forms. One
hopes to find an algorithm for producing from each term the unique reduced
term equivalent to it. If such a n algorithm exists, it will automatically yield a
description of the operations (usually nonstandard) on the set of reduced terms
under which this set becomes an algebra isomorphic to FV(o). This approach is
illustrated in severa1 of the examples below. The first three examples were already
worked out in much greater detail in $53.3-3.4.
EXAMPLES OF FREE ALGEBRAS
1. Let "tr = &O, the variety of monoids. In a first college algebra course,
we learn to express the elements of a semigroup generated by a,, a,, - .in the
form x,x, - . x, where the x's range over the a's. The expression is meant to
designate an n-fold product of the a's. Formally, it is just a word over the alphabet
{a,, a,, e). The set of al1 words over an alphabet X is a free monoid (i.e., a free
algebra in the variety A & )over the set X, under the operation of concatenation,
with the unit element (the empty word). To justify this assertion, let us write
W(X)for this algebra of words. W(X) is clearly a monoid. Consider the absolutely
free algebra T(X) of the type of monoids. Its elements are certain words (or terms)
over the alphabet X U e). Using the associative law and the other two equa-
{ S ,
tions defining monoids, we see that every element of T(X) is equivalent modulo
@(A&) to a "reduced term": either e, or a term having no occurrence of e in which
al1 the occurrences of come at the beginning (a fully left associated term). The
natural homomorphism 6 from T(X) onto the monoid W(X) takes a very simple
form: 6(p) is the word over X obtained from p by removing al1 occurrences of
and e. Now it can easily be seen that the number of occurrences of in any term
is one less than the total number of occurrences of al1 other letters in the term.
Hence it follows that 6 restricted to the set of reduced terms is a one-to-one
function. Now we have @(A&)c ker 6, since W(X) is a monoid. But the above
observations imply that every term is @(&O)-equivalentto a reduced term, and
distinct reduced terms are inequivalent even in ker 6. From these facts it follows
that @(&O) = ker 6. Thus FAD(X)z W(X). The free semigroup over X is also
easily found in this example. It is the algebra composed of the nonvoid words in
W(X) under the operation of concatenation of words.
2. Let Y = 2%, the variety of groups. Free groups are slightly more com-
plicated to construct than free monoids. Let *: X -+ X* be a one-to-one map
where X and X* are disjoint sets. Let G be the set of al1 words in W(X U X*)
that contain no subwords xx* or x*x with x E X. From each u E W(X U X*) we
may obtain an element r ( u ) ~ Gby repeatedly removing occurrences of the
forbidden subwords xx* and x*x until none remain. It was shown in 53.4 that
r(u) is uniquely determined by u. Define a binary operation on G: u . v is the
word r(uv). Define a unary operation -' on G: u-' is the word obtained from u
by replacing every x by x* and every x* by x, and then reversing the order of
occurrence of the letters. For example, (y*xyx*)-l is xy*x*y. (G, -,-', 0)is a
free group over X. The not entirely trivial proof of this fact is given in 53.4.
Usually, we write x-l in place of x* and regard G as a subset of the set of words
over an alphabet X U X-l. We remark that there exist varieties of groups for
which the free algebras cannot be described constructively (i.e., for which the
equational theory is not recursive).
3. Let Y = d,, the variety of commutative groups defined by the equation
xn Ñ e where n is a positive integer. Finitely generated free algebras in dnare
particularly easy to describe, being finite and in fact isomorphic to direct powers
of the cyclic group Z,. Fdn(3), or Fdn(x,y, z), for example, is a quotient of the free
group F9*(x,y, z). Since F9&(x,y, z) is free for d , , Fdn(x,y, z) Fg4(x,y, z)/@(dn).
It is easily seen that every element of F9*(x,y, Z)is equivalent modulo @ ( d n )to
xiyyjzkfor some non-negative integers i, j, k < n. From this it follows that
Fdn(3) E zi.
4. Let Y = Y/, the variety of semilattices. Let W'(X) be the free semigroup
over X consisting of al1 nonvoid words over the alphabet X. Let F = FiDL(X)be
a free semilattice over X. Since W'(X) is free over X for the variety of semilattices,
F r W' (X)/@(Y/). Since semilattices satisfy the commutative and idempotent
laws, the order of occurrence of letters in a word of Wt(X) can be rearranged,
and repeated occurrences of letters can be deleted, without affecting the equiva-
lente class modulo @(Y/) of the word. In other words, w/@(Y/) depends only
on the set of l e y appearing in the word w. Let S(X) be the semilattice
consisting of a-13 nonvoid finite subsets of X under the operation of set-union.
Let 6 be the homornorphism of Wf(X)onto S(X) that takes x to (x) for x EX.
Clearly, 6(w) is the set of letters occurring in a word w. Now @(Y/) s ker 6, and
the previous observations imply that the inclusion is equality. Thus FiDl(X) z
S(X). If we wish to have the elements of S(X) represented by unique words in
Wt(X),then we introduce a linear ordering 2 of X. Define "reduced word" to
mean a word of the form x, x, with x, < - < x,. Under the quotient map,
each element of S ( X ) is the image of a unique reduced word.
5. Let F9(X) be the free ring with unit, and T,(X) be the absolutely free
term algebra of the type of rings with unit, over the set X. Every element of T,(X)
I'
4.11 Free Algebras and the HSP Theorem
is equivalent modulo O ( 9 )to a finite sum of elements +a where a is a finite

product of generating elements, or a = l. F9(X) is isomorphic to the ring Z[[X]]
of formal polynomials with integer coefficients constructed from non-commuting
indeterminates X E X . The ring Z[X], consisting of formal polynomials with
integer coefficients constructed from commuting indeterminates x EX, is a free
/ algebra over X in the variety of commutative rings with unit.
6. Let "y. = 9 designate the variety of distributive lattices. In dealing with
varieties of lattices, and lattice terms, it is convenient to replace the operation
symbols v and A by + and . L = F9(X) is a semilattice under each of its
operations + and -.Using the equations defining distributive lattices, one can
a
see that every member of L can be written as a finite sum of finite products of
generating elements. (The set of elements having such an expression contains the
members of X and is closed under + and so must be equal to L.) If u = x,x2
S ,
- - x, and v = y,y, y, are two products of generators, then u I v in L iff

(yi: 1 2 i 5 E} c (xi: 1 S i S k}. For example, we cannot have xyz 2 yu in L
(x, y, z, u are distinct free generators) because the equation xyzyu Ñ xyz fails to
1,
.,,,
hold in the two-element lattice. If u = sisnpi where the pi are finite products
of free generators, then we can remove from the expression any p, that is 2 some
pj, j # i. After severa1 deletions we arrive at an expression u = x,
sum of products of generators in which no two of the summands are comparable.
qi of u as
If also v = E, qi is such an expression for v, then u 5 v iff for every i there

is a j with q, 5 qf. (Suppose that for a certain i, for everyj there is a free generator
xj that occurs in the product expression for qf but not in the expression for q,.
Then mapping xj -0, 1 S j 5 k' and al1 other generators to 1 establishes a
homomorphism f to the two-element lattice such that f(u) = 1-since f(qi) =
1-while f(v) = O.) A Hasse diagram of F9(3) is drawn in Figure 2.1 on p. 39.
The function f9(n) = JF9(n)J,called the free spectrum of 9, has been computed
only for small values of n. From the above analysis,fg(n) is equal to the number
of nonvoid antichains'in the ordered set of nonvoid subsets of an n-element set.
7. Let B = Fgd(n) be a free Boolean algebra freely generated by x,, x,, S ,
xn-, . We use + , and - for the basic operation symbols, instead of v , A ,and
S ,
-. Every element of B can be expressed either as xx-(= O) or as a nonvoid sum

+
of elements p of the form x$xil x i q , a product in which ci = 1 and xT1 = xi
and x l l = x;. There are 2" of these formal products. It is easy to show that no
two of them represent the same element of B, and in fact that B is isomorphic to
the Boolean algebra of al1 subsets of a 2"-element set. The free spectrum function
of Boolean algebras is easy to calculate: fad(n) = 22n.
8. The free modular lattice FA(@)cannot be described constructively. In
fact, the set of equations in four variables valid in the class of modular lattices
is a nonrecursive set, as was proved by C. Herrmann [1983], following R. Freese
[1980], who proved this fact for equations in five variables. FA(4) is a very
complicated (and nonrecursive) infinite lattice that we can never hope to draw.
Nevertheless, FA(3) has just 28 elements. It is pictured in Figure 2.1 on p. 39.
9. Let 9be the class of al1 lattices. F9(X) can be described constructively,
but among al1 the free algebras of these examples, it is the hardest to describe.
For instance, it seems to be practically impossible to draw a picture of F,(3).

The structure of free lattices was worked out in P. M. Whitman [1941] and
[1942]. Whitman's algorithm effectively determines when two terms of the type
of lattices represent the same element of a free lattice, and it leads to a set of
reduced terms, one for each element of the free lattice. Unfortunately, these
reduced terms cannot be easily defined and are not easy to work with. Fortunately,
Whitman's results can nevertheless be summarized quite simply: Fy(X) is, up to
isomorphism, the unique lattice L having these properties:
WO. L is generated by X.
W1. If x, x,, ,xn-, are distinct elements of X then ni<,
xi $ x $ x,. Ii,,
W2. If a, b, c, d E L satisfy ab I c + d, then the interval I [ab, c + d] contains
one of the elements a, b, c, d.
Exercises 4.137
l . Define a function from nonvoid words on the alphabet X U I to integers:
L(w) = - l i f w ~ X , A ( w ) = o(w) - l i f w ~ ~ , a n d L ( w=) ~ , s i 5 , A ( a i ) i f w=
a, a, . .a, with a, E X U I. Prove that w is a term (i.e., w E T,(X))iff A(w)= - 1
and L(wt) 2 O for al1 nonvoid proper initial segments w' of w. (Cal1 w' a
proper initial segment of w iff w = w'u for some nonvoid word u.)
2. (Notation as in the previous exercise) Let w = ala2 a, with ai E X U I.
+
Prove that if A(w) = - 1 then there is a unique cyclic variant = a, - a,ala,
. . . a,-,of w that is a term.
3. (Notation as above) Prove that if w is a term and if w' is a proper initial
segment of w then w' is not a term.
4. Now prove that statement (26) in the proof of Lemma 4.1 15 is indeed valid
for the algebra of terms.
5. Show that the statements of Lemma 4.115 characterize the absolutely free
algebras. If A is generated by X, each of the basic operations of A is a
one-to-one function onto a set disjoint from X, and operations of A cor-
related with distinct operation symbols have disjoint ranges, then A is
isomorphic to the algebra of terms over X of its type.
6. Complete the proof of Lemma 4.109.
7. Prove that if F = Fy(n) for a variety V then F r Clo, F.
8. Prove that al1 the identities of A in n variables are identities of B iff C10,B
is a homomorphic image of Clo,A, and that B and A have precisely the
same identities in n variables iff Clo, A Clo, B.
9. Suppose that a variety V is defined by a set of equations in n variables.
S
Does it follow that V = HSP(Fy(n))?
10. Show that V = V((Fv(n): n < w))holds for every variety V.
11. Prove that the equality V = HP, holds for the class operators. (Hint: If
V = V(A) and X is sufficiently large, then Fy(X) is a subdirect power of A.)
12. If .V, c W, are varieties and X c Y then F%(X) is a homomorphic image
of EG(X) in a natural way, and EG(X) is isomorphic to the subalgebra of
FaG(Y)generated by X in a natural way.
13. Describe the free algebra on one generator in the variety of al1 algebras with
two unary operations.
14. Show that a variety V is locally finite iff its finitely generated free algebras
are al1 finite iff Ef(o) is a locally finite algebra.
15. The variety -tr of al1 monadic algebras is a locally finite variety. Calculate
its free spectrum Jy(n) = )FV(n)1.(The facts presented in $4.5 are al1 you
need.)
16. Find al1 subvarieties of the variety of monadic algebras.
17. A variety -tr is called minimal (or equationally complete) iff Y is nontrivial
but every variety properly contained in Y is trivial. As examples, the variety
of Boolean algebras and the variety of distributive lattices are minimal. (If
Y is a nontrivial subvariety of the variety of Boolean algebras, for instance,
then Y contains a nontrivial member A, and A contains a two-element
subalgebra Q; but V(Q) is the class of al1 Boolean algebras.) Find al1
minimal varieties of groups.
18. Find al1 minimal varieties of rings with unit.
19. Show that the variety 9 of distributive lattices is the only minimal variety
of lattices.
20. Prove that the free groups F9#(2)and F9*(co)are not isomorphic, but that
nevertheless F9*(2)contains a subgroup isomorphic to Fg4(o).(If a, b are the
free generators of F9*(2),then the elements a, b-' ab, b-2 ab2, ab-"abn . . .
- ,
freely generate a subgroup.)

21. Prove that if is a nontrivial variety, then in a free algebra Fv(X), the set
X is a minimal generating set.
22. Prove that if Y is any nontrivial variety of lattices, then in L = Fv(X) the
elements of X are both join prime and meet prime, and hence join and meet
irreducible. Thus every generating set of L includes X. (If X E X , every
member of L either lies below x or above a meet of finitely many generators
other than x. Prove this and the dual fact, and then show that every member
of X is both join prime and meet prime in L.) Note that this implies the
validity of a strengthened form of Whitman's condition (Wl) in Example 9.
*23. Show that the free lattice F2(3) contains a sublattice isomorphic to F9(o).
(Hint: Show first that F9(3) contains a sublattice isomorphic to F9(4).
Suppose that x, y, z are the free generators of F9(3), and consider the lattice
elements u, = xz + y(x + z(x + y)), u, = xz + y(z + x(y + z)), and the ele-
ments u, and u, obtained from u, and u, by exchanging x and y. Show that
no ui is above the meet of the other uj or below their join. Thus (W0)-(W2)
hold for the sublattice of F2(3) generated by X = (u,, u,). The proof
S ,
that F9(4) contains a sublattice isomorphic to F9(co) is a little easier.)

24. Compute the free lattice on two generators.
25. Show that the ordered set pictured in Figure 2.1 on p. 39 is indeed the free
modular lattice Fd(3).
26. Letting M, designate once again the five-element lattice with three atoms,
put "tr = V(M,) and calculate Fv(3). (Hint: It is a homomorphic image of
FA?(3).)
27. Letting N, designate the five-element nonmodular lattice and Y = V(N,),
show that in L = Fv(x, y,z), the subset L - {x,y,z) is the universe of a

distributive sublattice.
28. Let S, be the three-element semigroup introduced in the last paragraphs of
53.3. From the result of Exercise 11 of 53.3 deduce that V = V(S,) is
axiomatized by the three equations (xy)z x(yz), xx Ñ x, and (xy)x Ñ xy.
Now prove that V is not residually small (Definition 4.47). (Choose any
infinite set X including {O, 1) and for each x E X define mappings a,, b,,
c,: X + X as follows:
Check that {a,, a,} U {a,, b,, c,: x E X, x # O, 1) is closed under composi-
tion and constitutes a semigroup S in V. Show that if 4 is any congruence
of S with a, 4 a, (x # y), then a, 4 a,. Thus if 4 is a maximal congruence
separating a, and a,, then S/@is a subdirectly irreducible algebra in V and
/SI41= 1x1.)
4.12 EQUIVALENCE AND INTERPRETATION OF VARIETIES
In order to guide research and organize knowledge, we group algebras into

varieties. This way of classification has been so successful that it has no serious
competitor. We can also group varieties into families of varieties that share
interesting properties. A large number of these families will be introduced and
studied in Volume 2. Most of the schemes for classifying varieties are based on
the concepts of equivalence and interpretation we will introduce in this section.
The choice of multiplication and inversion to be the fundamental operations
of a group is determined by tradition and convenience, but other choices are
posible. For example, define a division group to be an algebra A = (A,/) with
one binary operation obeying the laws
(See the third exercise in 53.4.) It can be shown that if A = (A,/) is a division
group, then x/x = y/y holds in A, and by defining e = y/y, x-' = e/x, and x .y =
x/y-l, we obtain a group A" = (A, ., -l, e). Of course, if G = (G, .,-l, e) is a
group, then Gd = (G, /) is a division group, where x/y = x -yv1.Moreover, in
this setting, Amd= A and Gdm= G. The varieties of groups and of division
groups are really equivalent. In fact, the mapping from groups to division
groups defines a functor from the category (or variety) of groups onto the
category of division groups, and " is its inverse. (Functors between categories
were defined in 53.6.) We say that these varieties are equivalent, a term to be
defined below.
4.12 Equivalente and Interpretation of Varieties 245
The presentation of groups as division groups is economical (two equations

with one operation instead of five equations with three). For ease of presenta-
tion of the theory of groups, however, the standard set of three operations is
unbeatable. Here is an example where it is harder to decide which of two
equivalent varieties is the preferred object of study: the variety of Boolean
algebras and the variety of rings with unit obeying the law x2 Ñ x, called Boolean
rings. If A = ( A , A , v , -) is a Boolean algebra, and if we put
O =x A x-; and 1 = x v x-
then A' = (A, + ,., -, 0 , l ) is a Boolean ring. (Here x + y is the symmetric
+
difference, or Boolean sum of x and y.) If R = (R, , -, 0 , l ) is a Boolean ring,
S ,
then Rb = (R, A , v , -) is a Boolean algebra, where

x ~ y = x . y ;x - = 1 + x ; and x v y = x + y + x y .
Moreover, Arb = A and Rbr = R.
In these examples, the corresponding algebras in the two varieties are bi-
interpretable: The basic operations of one algebra are defined as term operations
in the other algebra, and vice versa. An important example of an interpretation
running in one direction only, between two varieties, is supplied by associative
algebras over a field F, and Lie algebras over the same field (see 41.1).
We proceed to define in full generality the concept behind these examples.
Let Y and W be varieties of respective similarity types a and p, and let I be the
set of operation symbols of type o. By an interpretation of Y in W is meant a
mapping D with domain I satisfying:
DI. If a(Q) = n > O, then D(Q) = Q, is an n-ary p-term.
D2. If o(Q) = O, then D(Q)= QD is a 1-ary p-term such that the equation
QD(v0)Ñ QD(v1)is valid in W.
D3. For every algebra A E W , the algebra AD = (A, Q$(QE 1)) belongs to Y.
(For nullary Q, the operation eAD
is taken to be the nullary operation
corresponding to the constant unary operation Q;, in order that ADbe an
algebra of type o.)
Note that when D is an interpretation of in W the mapping is a functor
from W to Y. (Strictly speaking, we must define a D when a: A 4 B. Clearly,
hom(A, B) c hom(AD,BD)if A, B E W, SO we can take a D= a.) The definition of
interpretation was made slightly complicated by our desire that groups be
interpretable into division groups. Since the type of division groups has no O-ary
operations or terms, the O-ary operation e of groups must be interpreted as the
constant 1-ary term operation of division groups.
We remark that this notion of interpretation becomes rather transparent in
clone theory, where it is the same thing as a homomorphism from the clone of
Y" (the clone of term operations of the free algebra Fv(co) with nullary term
operations deleted) into the clone of W. The notion of equivalence we now define
is the same thing as an isomorphism of the clones of the two varieties.
DEFINITION 4.138. By an equivalence of varieties and W is meant a pair

of interpretations, D of V in W, and E of W in V , such that ADE= A for a11
AE W, and B~~ = B for al1 B E V . We say that Y is equivalent to W iff there
-
exists an equivalence between them, and that V is interpretable into W iff there
exists an interpretation of Y in W . The relation of equivalence between varieties
is written V W .
Equivalence is of course an equivalence relation on varieties. There are

severa1 equivalence relations on algebras that are more or less closely related
to it.
-
DEFINITION 4.139. Two algebras A and B are called equivalent (or term
-
equivalent), written A B, iff A and B have the same universe and precisely the
same n-ary term operations for every n > O. A is said to be weakly isomorphic to
B iff there exists an algebra C that satisfies A C r B. A and B are called
polynomially equivalent iff they have the same universe and precisely the same
polynomial operations (i.e., Po1 A = Po1 B).
Our readers are encouraged to work out a proof of the theorem below. We
shall prove it in the chapter on clone theory in Volume 2.
THEOREM 4.140. For varieties V and W the following statements are

equivalent.
i. V S W .
ii. There exist algebras A and B such that A S B, V = V(A), and W = V(B).
iii. The free algebras Fy(m) and Fw(m) are weakly isomorphic.
iv. There exists a byective functor @ of V onto W that commutes with the
forgetful functor from algebras to sets. (A and @(A)have the same universe
for every A E ~ and , @(a)= a for every homomorphism a between two
algebras of Y.) ,
The question of whether V is interpretable into W can be quite subtle in

particular instances. Moreover, some very interesting families of varieties can
be defined by the interpretability of specific varieties. The classic result in this
area is that of A. 1. Maltsev [1954]. It asserts that a variety V has permuting
congruences iff V interprets the variety of al1 algebras (A, M ) in which M is
ternary and obeys the laws M(x, y, y) Ñ x and M(x, x, y) Ñ y.
THEOREM 4.141. Al1 the algebras in a variety V have permuting congruences

$ there is a 3-ary term p of the type of V such that the equations p(x, y, y) Ñ x,
p(x, x, y) Ñ y are valid in V .
Proof. If there is a term p such that these equations are identities of V , then
every algebra A in Y has a term operation pA obeying the equations. Thus every
algebra in V has permuting congruences, by the argument in the first paragraph
of 54.7.
Let us suppose, conversely, that the congruences permute in every algebra

of V .We shall see that this is equivalent to the permutability of congruences in
the free algebra of "f with three generators. Let A = Fv(3), freely generated by
x, y, z. (We can assume that Fv(3) exists, for otherwise Y" is a trivial variety of
one-element algebras, and any 3-ary term will satisfy our requirements.) Let
a = cgA(x,y) and fi = c ~ ~z).(Now ~ , (x, z) E a o p; consequently, (x, z) E /3 0 a,
so there exists an element u E A with (x, u) E p and (u, z) E a. Since (x, y, z)
generates A, there is a 3-ary term p such that u = pA(x,y, z). To see that p
has the required properties, we define f as the endomorphism of A satisfying
f (x) = f (y) = x, f (z) = y, and we let g be the endomorphism defined by g(x) = x,
g(y) = g(z) = y. Clearly a c ker f and p E ker g. (It can be shown that a = ker f
and /3 = ker g.) Since (x, u) E ker g, we have
x = g(x) = s(u) = g(pA(x,y7z))= pA(gx,gy7gz)
= pA(x,Y, Y).
(We are using Lemma 4.123.) In a similar fashion it follows that y = pA(x,x, y).
By Theorem 4.127, these equalities in Fy(x, y, z) are equivalent to the desired
conclusion.
In recognition of the great influence that this theorem has had in variety
theory, varieties with permuting congruences are often called Maltsev varieties.
Terms and operations obeying the equations in the theorem, and algebras having
a term operation obeying the equations, are also frequently called Maltsev terms,
Maltsev operations, and Maltsev algebras. The reader is referred to 54.7 for further
facts about the important class of Maltsev algebras, including al1 groups, rings,
modules, and quasigroups.
Another early result, showing that the combination of congruence per-
mutability and distributivity can be characterized in an analogous way, was
obtained by A. F. Pixley [1963].
DEFINITION 4.142. If the congruence lattice of A is a distributive lattice

of permuting equivalence relations, we say that A is arithmetical. A variety
composed of algebras with this property is called an arithmetical variety.
THEOREM 4.143. A variety V is arithmetical iff there is a 3-ary term p such

that the equations p(x, y, y) Ñ x, p(x, x, y) Ñ y, p(x, y, x) Ñ x are valid in Y".
Proof. Suppose first that the equations are valid in Y" for a certain term p. Let
A E Y". Then by the previous theorem and Lemma 4.66, we have that the congru-
entes of A permute and the join in Con A coincides with relational product. To see
that Con A is distributive, let a, p, y E Con A and suppose that (a, c) E y A (a v P).
We must show that (a, c) E (y A a) v (y A p). Now there exists an element b E A
with (a, b) E a and (b, c) E p. Modulo y we have the congruences
a =, c =, p A(a, b, c)
since c = pA(c,b, c) and the operation pA preserves y. Moreover,

a = p A (a, b, b) =p pA(a,b, c), and
c = pA(b,b, c) = a pA(a,b, c).
Thus (a, u) E y A P and (u, c) E y A a where u = pA(a,b, c), showing that (a, c) E
(Y A a) v (Y A P).
For the converse, assume that Y is arithmetical and as before, let A = F 4 3 )
be a free algebra in Y freely generated by x, y, and z. We consider these con-
gruences of A: a = cgA(x,y), P = c ~ * ( z),
~ , and y = cgA(x,z). Clearly, (x, z) E
y A (a v p), and so there exists an element u E A with
(since A is arithmetical). Arguing as in the proof of the last theorem, we have a

term p such that u = pA(x,y, z), and p obeys Maltsev's equations. That the third
equation holds in Y (i.e., pA(x,y, x) = x) follows in the same way from the fact
that <x,pA(x,Y,2)) E y-
A term obeying the equations of Theorem 4.143 in a variety Y is called a

Pixley term for Y . The varieties of Boolean algebras, Heyting algebras, and
monadic algebras are arithmetical; indeed, so is any variety of Boolean algebras
with extra operations. A rather different example is supplied by the rings that
obey the equation xn Ñ x, where n > 2. The ring of integers is an arithmetical
algebra (although the variety it generates is not). The Chinese Remainder Theorem
of elementary number theory is equivalent to this fact.
Properties of a variety Y that are determined by which varieties interpret
ifito Y , such as congruence permutability or arithmeticity, are called Maltsev
properties. Congruence distributivit y and congruence modulari t y are Maltsev
properties. The latter will be proved in Volume 2; the former is seen in the
very useful result of B. Jónsson [1967], reproduced below. Terms obeying the
equations in the theorem over a variety Y are called Jónsson terms for Y.
THEOREM 4.144. A variety Y is congruence distributive iff there exists a

natural number n and 3-ary terms do, - ,d, such that the following equations hold
in V .
i. d,(x, y, z) Ñ x.
ii. di(x,y, x) Ñ x for O < i 2 n.
iii. di(x,y, y) Ñ di+l(x, y, y) for al1 even i < n.
iv. di(x,x, y) Ñ d,+,(x,x, y) for al1 odd i < n.
V. d,(x, y, z) Ñ z.
Proof. Suppose first that for a certain n we have terms do, dn obeying the
S ,
equations (i)-(v) in Y . Let A EY . To see that ConA is distributive, let a, P,

y E Con A and suppose that (a, c) E y A (a v P). Then there exist elements xo = a,
xl, , x, = c in A with (xi, x , + )~E a U p for al1 i < k. We wish to prove that
I
= (y A a) v (y A p). For a fixed j In and for any i < k we do have

c), df(a, x,,, ,c)) EX,since this pair belongs to a U fl and since, by (ii),
r dp(a, x, a) = a for al1 x. Thus by transitivity of x
,
df(a, a, c) = df(a, x,, c) df(a, x,, c) = df(a, c, c)
for al1 O 5 j < n.
-,
Now this and the equations (iii) and (iv) yield that df(a, c, c) d$,(a, c, c) for
al1 j < n. Thus dt(a, c, c) =, db(a, c, c), which by (i) and (v) is equivalent to
(a, c) EX,the desired result. These calculations demonstrate that ConA is a
distributive lattice.
For the converse, we assume that the algebra A = FV (3) freely generated
in V by three elements x, y, z has a distributive congruence lattice. We consider
once again the congruences a = cgA(x,y), p = c ~ * ( z),~ ,and y = cgA(x,z). Since
(x, z) E y A (a v p) and Con A is distributive, there must exist a natural number
n and elements u, = x, u,, u, = z such that
m ,
ui -,x for al1 O S i 4 n;

u, =B ui+, for al1 even i < n; and
u, r,u,+, for al1 odd i < n.
Choosing terms do, , d, such that ui = dp(x, y, z), we see, in just the same way
as in the proofs of Theorems 4.141 and 4.143, that the equations (i)-(v) are valid
in V. m
Exercises 4.145
1. Verify that equivalent algebras are polynomially equivalent and have the
same nonvoid subuniverses, and that polynomially equivalent algebras have
the same congruences.
2. Prove that Z and (Z,+) are not equivalent, where Z is the additive group
of integers.
3. Prove that for any group G = (G, -l, e) these statements are equivalent:
S ,
(i) G (G, );(ii)xn Ñ e isan identity of G for some positive n; (iii)Z 6 SP(G).
4. Show that if * is a binary operation such that (Z, *) = (Z,+ ) then * = +.
5. Prove the equivalence of groups and division groups, as outlined in this
section.
6. Prove the equivalence of Boolean algebras and Boolean rings.
7. Show that if A = ( A , A , v ,-) is a Boolean algebra, then A is equivalent to
( A , 1) where xl y = x- A y- (the Sheffer's stroke operation). Conclude by
Theorem 4.140 that the variety of Boolean algebras is equivalent to a variety
of groupoids.
8. Prove that the variety of lattices is not equivalent to any variety of groupoids,
or of algebras with one 3-ary operation. (Consider the clone of term opera-
tions of the two-element lattice.) Define a variety of algebras with one 4-ary
operation that is equivalent to the variety of lattices.
9. Find two inequivalent varieties, each of which is interpretable into the other.
Can you find two inequivalent varieties, each of which is equivalent to a
subvariety of the other?
10. Prove that an algebra A is arithmetical iff this version of the Chinese Re-
mainder Theorem holds in A: For every finite sequence a,, . . . , a,, $, . - ., $n
of elements and congruences of A, if (a,, aj) E $, v S;. for a11 O 5 i, j 5 n, then
there exists an element x E A satisfying (x, a, ) E $, for al1 O 5 i 5 n.
11. Prove that a variety "tr is arithmetical iff there is a term p obeying the
equations of Theorem 4.141 and a term q obeying the equations q(x, x, y) z
4x7 Y, x) q(y, x, x) N x, in y.
12. Prove that for any integer n > 1 the variety of rings obeying the law xn z x
is arithmetical. (See Example 4 and Corollary 4 in 94.5.)
13. Show that the variety of lattices has Jónsson terms for congruence distri-
butivity (Theorem 4.144) with n = 2.
4.13 COMMUTATOR THEORY
In the theory of groups, the important concepts of Abelian (commutative) group,

the center of a group, the centralizer of a normal subgroup, solvable group,
and nilpotent group can al1 be defined in terms of the commutator operation
[x, y] = x-'y-'xy. Alternatively, they can be defined in terms of the operation
[M, N] = Sg({[x, y] : x E M, y E N)) on normal subgroups, also called the com-
mutator. Analogous concepts, based on the multiplication of ideals, are important
in ring theory. An extension of these concepts to algebras other than groups and
rings is behind one of the most exciting new directions of research in general
algebra.
General commutator theory has to do with a binary operation, the com-
mutator, that can be defined on the set of congruences of any algebra. The
operation is very well-behaved in congruence modular varieties, much less so
in most other varieties. This section is an introduction to the basic concepts
of general commutator theory. The topic will be covered more thoroughly in
Volume 3. For more details, we refer the reader to the pioneering papers of
J. D. H. Smith [1976] and J. Hagemann and C. Herrmann [1979], and the
more recent works by H. P. Gumm [1983] and R. Freese and R. McKenzie
[forthcoming].
DEFINITION 4.146. Let A be any algebra. The center of A is the binary

relation Z(A) defined by:
iff for every n 2 1, and for every term operation t E Clo,,, A and for al1 c,, -.m ,
en,di, EA EA
A is called Abelian iff Z(A) = A x A.

4.13 Commutator Theory 251
Thus an algebra is Abelian iff (29) holds for al1 term operations and al1
elements of the algebra.
LEMMA 4.147. The center of A is a congruence on A.
Proof. Certainly Z(A) is an equivalence relation on A. Next, let f be a basic

n-ary operation of A, and let (ai, bi) E Z(A) for O 5 i < n. To see that (f (ao, . S ,
+
a,-,), f (b,, ,bn-,)) E Z(A), let t be a k 1-ary term operation of A and let
c,, . . - , c,, d,, - d, E A. Applying the definition of Z(A) consecutively to each
m ,
member in a sequence of k + n-ary term operations, we have

t(f (ao, ,a,-,), e) = t(f (a,, 7 a,-, ), 4 9
t ( f ( b O , a , , ~ ~ ~ , a n - , )=, ~t(f(bo,a,,...,a,-,)
) ,d)*
EXAMPLES
l. Let G = (G, .,-',e) be a group. If (a, b) EZ(G), then with the term
t(x, y, z)= z-lxy-'z we have t(b, b, c) = e = t(b, b, e) for any c E G, hence t(a, b, c) =
t(a, b, e) = ab-l. Thus (a, b) EZ(G) implies that c-lab-lc = ab-l for al1 c, and
consequently that (a, b) belongs to the congruence associated with the usual
group theoretic center of G, the normal subgroup N = (g E G: gh = hg for al1 h).
Conversely, suppose that ab-l E N. If t is a term operation, there is an integer m
(possibly negative) such that t(cx, x, , . ,x,) = cmt(x,x,, ,x,) for al1 c E N and
x, x,, . a X,E G. If t(a,C) = t(a,d) (i.e., if t((ab-')b,~) = t((ab-')b,q), then it
- ,
b , = ( ~ b - ' ) ~ t ( bd)
follows that ( ~ b - ' ) ~ t (E) , and so t(b, F) = t(b, d). Thus
Z(G) = {(a, b): (ab-')c = c(abel) for al1 c E G).
2. Let R = (R, +, -, O) be a ring, and let (r, S) E Z(R). If t E R, then we
m,
have (S - S) t = (S - S) 0, and replacing the first s on each side of the equality

by r yields (r - S) t = O. We can likewise show that t (r - S) = O. Thus (r, S ) E
Z(R) implies r - s E ann(R), the annihilator of R. Conversely, let r - s E ann(R).
It is easy to show that for any term operation t there is an integer m such that
+
t(c X,xl, +
x,) = mc t(x, x,, ,x,) whenever c E ann(R). Thus if we have,
m ,
+
say, t (r, F) = t (r, d), then also m(r - S) t (S,F) = m(r - S) + t (S,d),implying that
t(s, F) = t(s, d).Consequently,
Z(R) = ((r, S): (r - S) t = t (r - S) = O for al1 t E R).
3. Let M = (M, +, -, O,f,(r E R)) be a module over a ring R. Suppose
that t is a k + 1-ary term operation of M. There exist r, r,, - , r, E R such
that t(x, x,, - ,x,,) = h x f,,xl + + +
fTkxk.Thus t(y, x,, - x,) = &(y - X) e , +
t(x, x,, . x,), and it readily follows that M is Abelian.
m ,
From these examples, we see that a group is Abelian (in the sense of
Definition 4.146) iff it is commutative, and that a ring is Abelian in this sense iff
it is a zero ring (i.e., satisfies the law xy Ñ O). Every module over a ring is an
Abelian algebra. Conversely, it follows from what has been said that every
Abelian algebra in the variety of groups, or the variety of rings, is polynomially
equivalent to a module. This result holds more generally for any congruente
modular variety. (See the concluding remark of this section.)
Our definition of commutators of congruences involves a certain relativiza-
tion of formula (29) to any three congruences.
DEFINITION 4.148. Let a, P, y be congruences of an algebra A. We say that

a centralizes p modulo y, written
C(a, P; Y),
iff for al1 n 2 1, and for every t E Clo,,, A, (a, b) E a, and (c, ,d, ), , (c,, d,) E P
we have
LEMMA 4.149. Let a, P, y, a, (SE S) and y, (t E T) be congruences of un algebra A.

i- C(a, P; a A P), C(OA, P; y), and C(a, O,; y).
,
ii. C(a,, P; y) for all s E S implies C(V,, a,, p; y).
iii. C(a, P; y,) for all t E T implies C(a, P; A, , y,).
Proof. Straightforward. m
DEFINITION 4.150. For congruences a and fi of A we define their com-

mutator, denoted [a,P], to be the smallest congruence y of A for which a
centralizes p modulo y. The centralizer of p modulo a, denoted (a : P), is the largest
congruence y of A such that y centralizes P modulo a.
According to the lemma, the commutator and the centralizer of any pair of
congruences are well defined. Notice that where 0, and 1, are the least and largest
congruences of A, we have Z(A) = (O, : l,), and that [l,, lA]= O, iff Z(A) = 1,
iff A is Abelian. Moreover [a, P] 5 a A P is always true.
The commutator has proved to be a powerful tool for investigating con-
gruence modular varieties. This is largely due to the fact that for any algebra in
a congruence modular variety, the commutator is a commutative and completely
join-preserving operation on congruences. We cannot prove this fact in this
volume, but for varieties with permuting congruences a proof is outlined in
Exercise 13. In groups, or in rings, our commutator is a familiar operation; see
Exercises 11 and 12.
We conclude this chapter with a brief study of Abelian algebras.
DEFINITION 4.151. A(A) is the congruence on A2, and S(A)is the subuniverse
of A4, defined by
4.13 Commutator Theory
THEOREM 4.152. These statements are equivalent for any algebra A.

i. A is Abelian.
ii. {(a, a): a E A} is a coset of a congruence on A2.
iii.((x, x), (y, z)) E A(A) iff y = z for al1 x, y, z E A.
iv. A4 has a subuniverse S such that
a. (x, y, x, y) E S for al1 x, y E A.
b. ( x , y , u , v ) ~ S - ( y , x , v , u ) ~ S + + ( u , v , x , y ) ~ forx,
S y,u,u~A.
C. ( x , x , y , z ) ~ S - y = z forx, y , z ~ A .
v. ( x , x , y , z ) ~ S ( A ) i f f y = z f o r a l l x , y , z ~ A .
Proof. We first show that (ii) and (iii) are equivalent. If (iii) holds, then O, =
((a, a): a E A} is a coset (equivalence class) of the congruence A(A), hence (ii)
holds. On the other hand, if O, is a coset of a congruence p on A2, then A(A) c p,
since its generators belong to p. Thus O, is a coset of A(A), and (iii) holds.
The equivalence of (iv) and (v) is proved in a similar fashion. The relation
S(A) automatically satisfies (iva) and (ivb), because its generating set is invariant
under the automorphisms (x, y, u, v) -, (y, x, u, u) and (x, y, u, u) --+ (u, u, x, y)
of A4.
The equivalence of (iii) and (v) is also quite easily seen. Let R = {((x, y),
(u, u)): (x, y, u, v) E S(A)}. Then R is a reflexive and symmetric binary relation
on A2 (since S(A) satisfies (iva) and (ivb)) and is a subuniverse of (A2) x (A2).
Therefore the transitive closure of R is a congruence on A2, and this congruence
is clearly just A(A). This makes it obvious that (iii) is equivalent to (v). We have
now seen that al1 of the statements but the first are mutually equivalent.
To prove that (iv)implies (i), let S be a subuniverse of A4 satisfying (iv a-b-c).
Let p be a k + 1-ary term of the type of A for some integer k, and let a, b, c,, S ,
ck,d,, . dkE A. Define elements of A4:

S ,
ii = (a, a, b, b),
-
v1 = <~l,dl,cl,dl),
-
Vk = ( ~ kdk,
, Ck, dk).
By (iva) and (ivc), these elements belong to S. We have
and this element belongs to S. Then by (ivb) and (ivc),

tA(a,C) = tA(a,d)tt tA(b,C) = tA(b,d).
Thus A is Abelian.
We can now complete the proof by demonstrating that (i) implies (v).
Basically, we can just reverse the preceding argument. Suppose that A is Abelian,
and let x, y, z E A be such that ( x , x, y, z) E S(A).By Lemma 4.123(v),there is a

term t in the type of A and some elements ü,, - . , ümin the generating set of S@),
such that
We can assume that for some k 5 m we have ü, = (a,, a,, bi, b,) for 1 2 i Ik, and
ü, = (ci,di,ci,d i ) for k < i 2 m. Thus
A
t (al,...,ak,c,+,,.-.,~,) = x = while
tA(al,-..,ak,d,+,,---,dm),
tA(bl,".,bk,~k+l,"',~m =)y and = z.
tA(bl,---,bk,dk+l,-..,dm)
So it follows that y = z, as desired, since A is Abelian. (Argue just as in the proof
of Lemma 4.147, using that (a,, b,) E Z ( A ) ,or see Exercise 2.) ¤
LEMMA 4.153. Suppose that Con A has a sublattice (O,, S,, S , ,$, ,1, ) con-
sisting of permuting congruences, isomorphic to M,. Then A is Abelian.
Proof. We have the sublattice pictured in Figure 4.10.
Figure 4.10
Suppose that t is a k + 1-ary term operation of A and that a, c,, , c,, d,,
- d, E A and t(a,F ) = t(a,d).Consider first an element b such that ( b , a ) E S,.
S ,
Choose ei with (c,, e,) E $, and (e,,d i ) E for 1 Ii 5 k (using that S/, O S, = 1,).
Now
-
t(a,E) t(a,d) = t(a,F) and
t(a,e> =$1 t(a,C),
giving us that t(a,E) = t(a,F ) = t(a,d)since t,h, í l $, = O,. Then
t(b,E) =$ot(a,é) = t(a,F) =$ot(b,C) and
t (b,e) =*1 t (b,F),
giving us that t(b,E) = t(b,F). Finally,
t(b,d) =$o t(a,d) = t(a,F) =$ot(b,C) = t(b,E),
and t(b,d)=,b2 t(b,é),implying that t(b,d) = t(b,Z) = t(b,C) (since S, í l S, = O,).
4
We have established the implication t(a,Z) = t(a, d) + t(b,F) = t(b, when (a, b) E
S/,. The same result obviously follows with S/, replaced by S/,. Now if a and b
are any two elements of A, there exists u with (a, u) E SOand (u, b) E S/1. Then
t(a, C) = t (a, d) -+ t(u, E) = t(u, d) + t(b, 2) = t(b, d), and that concludes the proof.
m
LEMMA 4.154. If A generates a nontrivial variety of algebras with permuting
congruences, then the following are equivalent.
i. A is Abelian.
ii. A2 has a sublattice {O,, S/, S/, ,S/,, 1,) isomorphic to M,.
iii. In Con A2,A(A) is a complement of the kernel of each projection homomorphism.
Proof. Assume that V(A) has permuting congruences. If (iii) holds, then {O,, y,,
A(A), y,, 1,) is a sublattice isomorphic to M,, where yi = {((xo,xl), (y,, y,)):
xi = y,) are the projection kernels, so (iii) implies (ii). If (ii) holds, then A2 is
Abelian by Lemma 4.153, so A is Abelian by the result of Exercise 9. Thus (ii)
implies (i).
Now suppose that A is Abelian. We shall prove that A(A) (defined in
Definition 4.151) is a complement of the first projection kernel y,. (The proof for
y1 would be completely analogous.) That y, v A(A) = i A 2 follows from this
calculation (for any (x, y), (u, u) E A2):
To see that 17, í l A(A) = OA2) let p(x, y, z) be a term obeying Maltsev7sequations
in A. (See Theorem 4.141.) Assume that ((x, y), (x, z)) E r ) , í7 A(A)-i.e., that
(x, y) -A(*) (x, 2). We have to show that y = z. This calculation does the trick:
Thus ((x, x), (y, z)) E A(A), implying that y = z by Theorem 4.152, since A is
Abelian. m
THEOREM 4.155. Suppose that the variety V has permuting congruences. For
A E "ir the following are equivalent.
i. A is Abelian.
ii. If the term p(x, y,z) obeys Maltsev's equations in Y , then pA is a homo-
morphism A3 + A.
iii. A is polynomially equivalent to a module over a ring.
Proof. We begin by choosing a Maltsev term p(x, y, z) for V. Now suppose that
A is Abelian. According to Lemma 4.154, A(A) is a complement of y, in Con A ~ ,
and of course A(A) and y, permute. This means that for given x, y, z E A there is
a unique element u such that (x, u) (y, z). We claim that u = p A ( ~y,,z).
In fact,
which establishes the claim. Now to see that (ii) holds, let q be any n-ary
term of V . Let (x,, y,, 2,) E be given for i < n, and let X = (x,, - ,xn-, ),
Y = (y,,---,yn-,), Y = (zo,---,zn-,), and ü = (uo,-*.,un-,) where u, =
pA(xi,yi,zi).For i < n we have (xi,ui) =A<Al (yi,zi) by the claim. An applica-
tion of the term operation q A 2 then yields
which, again by the claim, implies that
This equality, which can be written as a formal equation valid in A, is just the
condition for pA to commute with the term operation qA.Since pA commutes
with al1 term operations of A, it is a homomorphism A3 +-A, as (ii) asserts.
To see that (ii)implies (iii),we now assume that pAis a homomorphism. Our
task is to construct a module polynomially equivalent to A. First observe that if
A is polynomially equivalent to a module M = (A, +, - , O , f , ( r ~R)) (i.e., if
Po1 A = Po1 M), then pA(x,y, z) expressed as a polynomial of M can be nothing
+
other than x - y z. This observation will motivate our argument. Notice that
the polynomial operations of M will be precisely the operations that can be
expressed in the form f (x,, S , xn)= L1x, + . + Lnxn+ c with r, E R and c E A.
We choose an arbitrary element of A, write it as 0, and hold it fixed
throughout the argument. Then we put x + y = pA(x,O, y), and -x = pA(O,x, O).
In order to establish that we have defined a commutative group, we use severa1
times the facts that pA commutes with itself and obeys Maltsev's equations. Let
x, y, zE A. Then we have
Moreover,
P(X,Y, 0) = P(P(X,0, O), ~ ( 0o,, Y),~ ( 0o,, 0))
+
=p(x,o,p(O,y,O)) = x -y; and
P(X,Y, 2) = P(P(X,0, O), P(Y,0, O), ~ ( 0o,, 2))
= P(P(X,Y,O), o, z) = x - y + 2.
The ring we need is easily defined. We put
R = (r E Pol, A: r(0) = 0).

Note that Theorem 4.6 (in 54.1) easily implies that pA commutes with al1 the
polynomial operations of A. Thus for r E R we have r(x + y) = r(p(x, 0, y)) =
+
p(r(x), r(O),r ( y ) ) = r(x) r(y), implying that R is a subset of the ring of endo-
morphisms of the Abelian group (A, +). If r, SER, then r + s defined by
(r + s)(x) = r(x) + S(X)is a member of R. Moreover, R is trivially closed under
composition (the multiplication operation of the ring of endomorphisms). The
zero and unit elements of the ring of endomorphisms, defined by O(x) = O and
+
l(x) = x, belong to R. Thus we have a ring with unit R = (R, ,o, -, O, l), which
is a subring of the ring of endomorphisms of (A, +). Naturally, we have an
R-module M = (A, + ,- ,O, r(r E R)). It is clear that Po1 M S Po1 A because the
basic operations of M are polynomial operations of A.
To finish the proof that (ii) implies (iii), we have to show that every poly-
nomial operation of A is a polynomial operation of M. Let f €PolnA. Then
g(z) = f (z) - f (O, O,. O) defines a polynomial operation of A, and it suffices to
S ,
show that g E PolnM. Since g commutes with pA and g ( ~ = ) O, we have
for al1 xi, y, E A. An obvious inductive argument then gives that

~ ( ~ 0 7xn-1) = ro(x0) + + rn-,(xn-1)
where
ith
ri(x) = g(0;-.,O, x,O, -..,O)
and ri E R. This concludes the proof that (ii) implies (iii). It is quite easy to see
that (iii) implies (i). (See Example 3.) ¤
Theorem 4.155 was discovered in the mid-1970s. It is a corollary of a

much deeper result of C. Herrmann [1979], which states that every Abelian
algebra in a congruence modular variety is polynomially equivalent to a module.
(Herrmann's theorem will be presented in the chapter on classification of varieties
in Volume 2.)
Exercises 4.156
1. Assume that congruences a,, a,, Po, and p, of an algebra A satisfy a, I a,,
and Po < P,. Show that C(a,, P,; y) implies C(a,, Po; y), and conclude that
[ao,Pol 5 Ca1,P11.
2. Suppose that a, P, and y are congruences of A. Prove that C(a, j ; y) holds iff
for al1 m, n 2 1, and for every t E PO^,+^ A and (a,, b, ), (a- bm)E a and
S ,
(c,, d,), , (c., d.) EP, the implication t(ü,c) E,t(ü, d)-+ t(b,c) E,t(b, d)
is valid. (This formulation differs from Definition 4.148 in two respects: It
involves polynomial operations in place of term operations and (m, n)-tuples
of variables in place of (1, n)-tuples.)
3. Suppose that ,u < v and the interval I[,u, V] in Con A has a sublattice (,u, $o,
$,, v) isomorphic to M,. Show that [v, t,hi] < p for i = 0, 1, 2, and that
C$i $ i l 5 P.
4. Suppose that A has a polynomial operation m satisfying m(x,x, y) =

m(x, y, x) = m(y, x, x) = x for al1 x, y E A (a 3-ary majority operation). Prove
that [a, P] = a í l P for al1 a, P E Con A. (Corollary: [a, P] = a í l P if A is a
lattice.)
5. Suppose that A has a polynomial operation such that ( A , ) is a semilattice.
Prove that [a, P] = a í l P for al1 a, P E Con A.
6. An algebra A is called solvable iff it has a finite chain of congruences
O, = a, 5 a, 5 . - . 5 a, = 1, satisfying a,,,] 5 a, for i < n. Suppose
that A has a binary polynomial operation and two elements a and b such
thata # b a n d a . a = a , a . b = b . a = b - b = b.(I.e.,((a,b},-I{,,,))isasemi-
lattice.) Prove that A is not solvable.
"7. Assume that every subalgebra of A2 has a distributive congruence lattice.
Prove that then [a, p] = a í l P for al1 a, /3 E Con A. (Suppose that C(a, P; y)
where y < a í l p. Thus C(6,6; y) where y < 6 = a í l P. Take A(6) to be the
subalgebra of A2 with universe 6. Define p to be the congruence on A(6)
generated by (((x, x), (y, y)): (x, y) E 6). By modifying the proof of Theorem
4.152, show that (x,x) E,(y,z) only if (y,z) ~ y Let . y, and y+ be the
kernels of the projection homomorphisms of A(6) onto A. Note that y, v p =
y, v p (= A, say). By distributivity, p = p v (yo A yl) = A. Conclude from
this that 6 = y, which gives a contradiction.)
8. The previous exercise implies that a congruence distributive variety has only
trivial Abelian algebras. Prove this more directly, using Theorem 4.144.
9. Show that the class of Abelian algebras of a fixed similarity type is closed
under the formation of products and subalgebras, and that a direct product
is Abelian iff the factors are Abelian.
10. Let G be a group whose lattice of normal subgroups is isomorphic to M,
(the projective line with n points) for an n > 2. Show that n - 1 = p is a prime
integer and that G E Z :.
11. Let G be a group and a and P be congruences of G, with corresponding
normal subgroups N, = ela and Np = e l 6 Show that is the (normal)
subgroup N' generated by ( [x, y] : x E N,, y E Np}. (Prove that N' 5 NL,,BI.
Prove also that for any algebra A, if 6 is a congruence and if 6 < a í l P í l y,
then C(a, p;v) holds iff C(a16, P/6; y/6) holds in A/6. Using these results, show
that the claim will follow if it can be proved in the case where N' = (e}; then
prove it.)
12. Let R be a ring and a and p be congruences of R with corresponding ideals
J, = Ola and Jp= OIP. Show that JL.,pl is the ideal J,. JB+ JBJ, consisting
of al1 finite sums of products xy or yx with x E J, and y E Jp.
13. Let A be an algebra having a 3-ary polynomial operation f that obeys the
Maltsev equations of Theorem 4.141. Prove that if a, P, y, y' E Con A, y 5 y',
and C(a, B; y), then also C(P, a; y) and C(a, P; q'). Conclude that [a, P] =
[P, a] and, using Lemma 4.149, that
a, E Con A (i E I).
[vi E Vi
ai,P] = E [a,, P] whenever
14. Let A be an algebra having a binary polynomial operation . and elements u

and e such that u - x = u and e x = x for al1 elements x. Show that [l,, a] = a
for al1 a E Con A. \
C H A P T E R F I V E
Unique Factorization
5.1 INTRODUCTION AND EXAMPLES
The idea of analyzing a mathematical structure into a product of simpler struc-

tures goes a long way back to the beginnings of modern abstract algebra, and
even before that to the idea of independent events in probability theory and to
the work of Descartes in geometry. There was the Kummer-Dedekind theory of
products of ideals, the Wedderburn theory of finite-dimensional linear algebras,
and of course the early realization that a finite Boolean algebra is isomorphic to
a product of copies of the two-element algebra. Of course, al1 this came some
2000 years after the ancient Greeks had discovered the unique factorization of
the simplest structures (finite sets, or simply numbers) into prime factors.
The major utility of such representations of an algebra as isomorphic to a
product of simpler factors lies in the possible uniqueness of the representation.
For example, if G is a finite group and
with each factor indecomposable in the sense that it is nontrivial and cannot
be nontrivially factored, then n = m, and after reordering the Ki7s, we have
K, r H,, . K, r H,. (This was known to J. H. M. Wedderburn and R. Remak
m ,
early in this century; the Abelian case had already been taken care of by L.
Kronecker in [1870]. In the 1920s, Krull and Schmidt extended this result to
infinite groups with chain conditions.) Uniqueness of a factorization must always
be understood in this light (i.e., allowing rearrangements and isomorphisms of
the factors). The dominant theme of this chapter will be an account of the
attempts of various mathematicians (above al1 Bjarni Jónsson) to extend the
above group theoretical result to other kinds of algebras.
Before proceeding to the formal definitions, let us mention three places in
classical algebra where unique factorization holds and where the unique factors
can be classified in a simple way. The first is the theory of finitely generated
Abelian groups, where every directly indecomposable factor is cyclic. The next
260 Chapter 5 Unique Factorization
is the theory of finitely generated modules over a principal ideal domain; this
includes, of course, the aforementioned theory of Abelian groups, but also the
theory of one endomorphism F of a vector space. Here again the indecomposable
factors are cyclic, and one obtains the Jordan normal form for F (see Exercises
5.7(2-8) in 55.3). The third example is that of complemented modular lattices of
finite height; the indecomposable factors are the two-element lattice and certain
lattices derived from projective spaces. (For a precise statement, see Theorem
4.88 above.)
Let us say that an algebra A is directly indecomposable iff [Al > 1, and
A r B x C implies IBI = 1 or 1 CI = 1. We say, moreover, that A has the unique
factorization property (UFP) or is uniquely factorable iff
i. A is isomorphic to a product of directly indecomposable algebras, and
ii. this representation is unique in the above sense; that is, if
- -
for some index sets 1 and J and for directly indecomposable algebras Bi and
C,, then there is a bijection 4: I J such that B, Cm<,>for al1 i.
Untils5.5, we will be dealing with algebras that either are finite or at least have
a congruence lattice of finite height. For such algebras, point (i) of the defini-
tion of U F P holds automatically, and al1 our attention will be on point (ii)
(uniqueness).
Not al1 finite algebras have the unique factorization property: see 1-4 of
Exercises 5.1 below. These exercises form an essential part of this chapter in that
they place some important limits on what we may expect to prove, even for finite
algebras. The example presented in Exercises 6 and 7 is also very important in
that it puts essential limitations on what we will be able to prove for infinite
algebras. The reader is urged to begin any serious study of this chapter by doing
some or al1 of these six exercises.
For an overview of factorization theory in general, and of this chapter
in particular, it is helpful to speak in terms of the commutatii~emonoid of iso-
morphism classes of algebras under direct product, which we introduced at the
end of 53.3. In fact, for X any class of algebras of the same type that is closed
under the formation of products, the class X / E of isomorphism types of alge-
bras in X forms a submonoid of that monoid, which we will denote Mx =
( X , x )/E. For the study of direct factorization, it generally makes sense to
assume that X is also closed under the formation of direct factors, and this
will hold for al1 the examples of interest in this chapter. (In Exercise 5.1(3)
below, we examine such a X that is so small that Mx can be completely
seen and understood.) Occasionally we take a passing glance at an infinitary
operation (product of an infinite family of algebras) and at another binary
operation (disjoint union of relational structures; see 55.7). For the moment,
let us consider the class X of al1 finite algebras of a given similarity type. Of
course, the first few exercises of this section te11 us that, in general, Mx is not
a unique factorization monoid, but, according to Lovász's Theorem (Corol-
lary 2 to Theorem 5.23, described below), Mx does have unique kth roots. In
fact, the first corollary to Theorem 5.23 even tells us that this M, is
isomorphic to a submonoid of (m, S)", where denotes ordinary multiplica-
tion of natural numbers. As we will see in this chapter, there are many
interesting classes X of uniquely factorable algebras, such as the class of finite
algebras with O. For such X , the monoid Mx is obviously isomorphic to
(m, (Assign a different prime number to each indecomposable isomorphism
S ) .
type.)
For X containing infinite algebras, M,%is much less well behaved. The
non-uniqueness of square roots is illustrated in Exercises 5 and 7 below. The
extent of the possible pathology in M, is perhaps best revealed by the 119781
result of J. Ketonen, which concerns the class X of al1 countable Boolean
algebras. He proved that in this case, every countable commutative semigroup
can be embedded in M,. By a method of A. Tarski and B. Jónsson, which will
be described in Volume 2 in our section on Boolean powers, this result extends
to groups, lattices, etc. Thus we have, for example, countable groups, lattices, and
Boolean algebras A such that A z A3 but A A2. (A. L. S. Corner found such
an Abelian group in [1969].)
The group theoretical result mentioned at the start of this introduction can
be generalized in two different directions. The first such generalization assumes
congruence modularity, which we discussed briefly in $4.7 (see, e.g., Theorems
4.73 and 4.76). The main result (B. Jónsson [1966]) is presented in $5.3: every
finite algebra with modular congruence lattice and a one-element subuniverse
is uniquely factorable. This and some companion results are straightforward
corollaries of the Direct Join Decomposition Theorem (2.47) of $2.4. Our other
generalization of the unique factorization of finite groups is the one given by
B. Jónsson and A. Tarski in [1947]: ifA = (A, + ,O, F,, F,, - ) is a finite algebra,
+
with (O) a subuniverse, and obeying O + x Ñ x O Ñ x, then A is uniquely factor-
able. We develop this theory in $85.4 and 5.5. This is an elegant theory of
+
the algebras (A, +,O) that obey x + O Ñ O x Ñ x, with the other operations
F,, F2, getting more or less a free ride. These algebras ( A , +, O) have
been important in algebraic topology, where (when equipped with a topology
making + continuous) they are known as H-spaces.
One important open question about U F P for finite algebras is the following:
Does every finite idempotent algebra have UFP? (An algebra A is idempotent iff
each basic operation F satisfies the equation F(x, S ,x) Ñ x.) This question goes
back to A. Tarski some thirty years ago. R. McKenzie proved in [1972] (as a
corollary of a more general result) that every finite idempotent semigroup has
UFP.
We begin dealing seriously with infinite algebras in $5.5. It is not hard to
see that (i) often fails for infinite algebras (see, e.g., Exercises 2 and 7 of tj5.6), and
that in such cases (ii) holds vacuously. For this reason, point (ii) is not very
interesting in isolation. Therefore, we will introduce in $5.6 the closely related
refinement property, which is stronger than (ii) but equivalent to U F P for
algebras A such that every direct factor of A satisfies (i). (See Exercises 4 and 5 of
$5.6.) An algebra A is said to have the refinement property iff the following
262 Ghapter 5 Unique Factorization
holds for al1 index sets I and J and al1 algebras Bi and Cj:
then there exist algebras Dij (i E 1,j E J ) such that, for al1 i and j,
and
In 55.6 we will give various sufficient conditions (of Chang, Jónsson and Tarski)
for the refinement property.
Since it involves products over infinite index sets, the refinement property
cannot be construed as a property of the monoid Mx of isomorphism types. In
55.7, however, we return to the study of Mx, again mainly for X the class of
finite algebras of a given type. The highlights of that section are two results of
Lovász on kth roots and cancellation:
if Ak r Bk, then A r B, and
i f A x C r B x C, t h e n A ~ B
(for A, B, and C finite, and for C having a one-element subalgebra). The

proofs involve a simple and elegant counting argument of a style not usually
encountered in our subject.
Toward the end of 55.6, we include relational structures in our considerations
as well as algebras. (These are defined in Exercise 7 of Exercises 4.38 at the end
of 54.4.) This permits us to expand the monoid Mx of isomorphism types to
a semiring Sx = ( X , U, x ) / E by taking U as the disjoint union of structures.
Generally speaking, Sx is pretty complicated, but in some cases a simplification
is possible. Every finite structure is a disjoint union of connected structures in a
unique way (as we saw for unary algebras in 53.2). In Theorem 5.18 of $5.6, we
will see that every finite connected ordered set is uniquely factorable. In such a
case as this, it is evident that Sx will be isomorphic to the semiring P = (P, , ) +
of al1 polynomials with non-negative integer coefficients in N, commuting in-
determinates. (One indeterminate corresponds to each connected indecompos-
able ordered set.) Notice that P does not itself have unique factorization-e.g.,
Both of these factorizations may be refined to

(1 + x + x2) (1 - x + x2)- (1 + x),
but no such refinement is possible without using negative coefficients. Thus one
may obtain a (disconnected) finite ordered set without UFP by simply replacing
x in the above polynomial equation by a finite indecomposable connected
ordered set, such as a chain of two or more elements.
In any commutative monoid M (with unit element denoted 1) we can
distinguish between an indecomposable element a (a # 1, and a = b . c implies
b = 1 or c = 1) and a prime element (i.e., an element a # 1 satisfying: if a divides
b c, then a divides b or a divides c, where "a divides b" means that b = a d for
some d). The definition is motivated by the fact that, in the monoid of positive
integers, it clearly corresponds to one of the possible definitions of prime number.
(In monoids where 1 has factorizations other than 1 1, a more sophisticated
definition of "prime" is required, but this does not concern us here.) Following
A. Tarski, we say that an algebra A E X is X-prime, or prime in X , iff the
isomorphism class of A is a prime element of Mx. We will not prove any deep
results about prime elements in M%, but we do include three exercises on them
(11 and 12 below, and Exercise 10 of 953, and we will conclude this introduction
with a brief summary of results about them and some open problems.
In contrast to the situation for indecomposability, the notion of X-primality
depends very strongly on the class X, as can easily be seen from Exercises 11
and 12 below. In some rather small classes, such as finite groups, the prime
algebras are precisely the indecomposable algebras. We obviously have
x S x1 + x n 934ime4x1)c P r ~ i m ~ a ( X ) ,
and very often the inclusion turns out to be strict. For instante, R. McKenzie
proved in [1968] that if X' is the class of al1 algebras of a given type (with at
least one operation of arity 2 1), then no algebra is prime in X ' . On the other
hand, McKenzie showed in [1968] that equality holds in the above inclusion if
X is the class of finite groups and X' is the class of al1 finite groupoids-i.e.
every indecomposable finite group is prime in the class of al1 finite groupoids. In
[1971], R. Seifert proved that there are no primes in the class of al1 graphs, the
class of al1 finite graphs, or the class of al1 finite mono-unary algebras. Here are
some open problems about primes in classes of algebras:
Is there any prime in the class of al1 finite algebras of type (1,l) (or in any
type that has more than one operation of arity 2 l)?
Is there any prime in the class of al1 semigroups?
Is there any prime in the class of al1 idempotent semigroups?
Is Z prime in the class of al1 Abelian groups?
Exercises 5.1
REMARKS O N T HE FIRST SEVEN EXERCISES: The first two exercises
provide examples showing that the hypothesis of a one-element subuniverse
cannot be removed from the results of 55.3. The example provided by the second
exercise also shows that in the theorem of B. Jónsson and A. Tarski ($55.4-54,
we cannot remove the hypothesis that (O) should be a subuniverse of A. Nor
can we remove the hypothesis of a modular congruence lattice from Jónsson's
Theorem of 95.3, as is shown by the examples in Exercises 3 and 4. These examples
also show that the cardinalities of indecomposable factors are not invariant for
a finite algebra, and the second example in Exercise 3 shows that the number of
indecomposable factors is lilcewise not invariant. These examples therefore also
show that the hypothesis of congruence permutability cannot be removed from
Theorem 5.5 in 95.3. Exercise 4 also shows that, although we have interesting
extensions of the Krull-Schmidt Theorem to classes of finite algebras that are

not groups, it does not extend to finite commutative semigroups.
Exercises 1 and 2 also provide failures of the cancellation "law" A x C 2
B x C -+A r B, and these examples show that some hypothesis besides finite-
ness will be required in 95.7 to make cancellation work. Exercises 5, 6, and 7
provide examples of failures of the "unique square root principle" A2 r B2 -+
A r B. By one of the main results in $5.7, this principle does hold for finite
algebras, so the counterexamples here are necessarily infinite. The example of
Exercises 6 and 7 shows that the hypothesis of finite height cannot be removed
from the Direct Join Decomposition Theorem (2.47) and its corollaries in this
chapter, Theorems 5.3 and 5.4 below. This example also shows that in $5.6 we
cannot expect to prove a refinement theorem directly generalizing the results of
Ore, Birkhoff, and Jónsson.
1. Let A = (4, f, g) be the four-element algebra with two operations f and g,
which is graphically depicted as follows:
(i.e., f(O) = 1,f (1) = O, etc.). Prove that Con A is
for three proper congruences el, 8, and e,, and that each Alei has two
elements. It immediately follows that A has permuting congruences and that
Prove that no two of the algebras A/ei are isomorphic.

2. Let Z2 denote the two-element group, 1 the identity permutation on Z2, and
f the (unique) non-identity permutation of Z,. For A, = (Z,, +,f ) and
A, = (Z,, +,l), prove that A, x A, E A, x A,, but A, $ A,.
3. (F. Galvin.) Let A, denote the three-element mono-unary algebra (3, f),
where f is a cyclic permutation (i.e., A,,, in the notation of 53.2). Similarly,
A, will denote a mono-unary algebra of one element. This exercise concerns
the algebras A,,,, = k A, U A, (a disjoint union of A, with k copies of A,).
One easily checks that the product A,,,, x A,,+, is again 'of this type
(namely, it is isomorphic to A,,+,, where n = 3km + k + m). Prove that in
5.1 Introduction and Examples 265
fact the algebra A3,+, can only be factored in this manner, via a factori-
+
zation of 3n + 1 as (3k + 1)(3m + 1). Thus if n > O and 3n 1 cannot be
nontrivially factored with integral factors each congruent to 1 modulo 3, then
A,,+, is directly indecomposable. So, fcr example, A,, A,,, and A,, are al1
directly indecomposable. Thus the following is another failure of unique
factorization:
In other words, in this exercise we see a class X of finite algebras that is

closed under the formation of products and direct factors, such that the
monoid Mx of isomorphism types is isomorphic to the monoid (under
multiplication) of positive integers congruent to 1 (mod 3). The failure of
unique factorization in this monoid is immediately reflected in the above
failure of unique factorization for the algebra A,,,.
One may modify this example by replacing the 3-cycle A, by a 5-
cycle A,, thereby forming the algebras B,,,, = k. A, U A,. Prove that the
following is another failure of unique factorization:
B16 x B8, r B, x B, x B, x B,.
4. (R. McKenzie.) A commutative semigroup. For each n, let A, denote the
n-element semigroup with constant multiplication, and for n 2 2, let B,,
denote the 2n-element groupoid with this multiplication table:
(there are three square blocks of O's, and one square block of 1's). Clearly,
each B,, satisfies the equation x(yz) = (uv)w and thus is a semigroup. Prove
that
A,, r A, x A,,
B2mn B2rn x An,
and that these are the only factorizations possible within isomorphism. Thus
for p prime, both A, and B,, are directly indecomposable, so we have another
failure of unique factorization:
B, x A, z B4 x A,.
5. Find two (infinite)mono-unary algebras A, B such that A2 N B,, but A B.
(Use the representation of mono-unary algebras described in $3.2.)
6. A torsion-free Abelian group. (From articles of Bjarni Jónsson during the

years 1945-1957.) Here we will describe nonisomorphic Abelian groups G
and J , with G2 r J2.To begin, we let V be a 4-dimensional vector space over
the field of rational numbers, with (x, y, z, u} a basis for V. We then define
Prove that {x', y', z', u'} is also a basis for V and that
Let P and Q be infinite disjoint sets of prime numbers with 5 $ ( P U Q), and
define R (respectively S ) to be the set of al1 square-free positive integers al1
of whose prime factors belong to P (respectively Q). We now define four
subsets of V:
a
r
a
b
s
b
c
-x+-y+-(x+Y)
5
c
1
-z+-U+-(z+u)
r s 5
a b c
J=
{ -x'+-y'+-(x'+2yf)
S 5
K =
{
a b c
1
-2' + -U' + -(zl + 2 ~ ' ),
S 5
where a, b, and c range over integers, r ranges over R, and s ranges over S.
It is obvious that each of these four sets is a subgroup of ( V , +), as are
G + H and J + K. Thus G = (G, + ) is a group; the groups W,J, K, G H, +
and J + K are defined similarly. It is obvious that G í l H = J í l K = (O), and
that G E H and J r K. Therefore, elementary group theory tells us that
+
G2 E G H and J2 E J + K. The remainder of the exercise has two parts.
The first part is to show that G + H = J + K, which of course immediately
implies that G2 r J2.The second part is to show that G $ J. (Hints: For the
first part, one need only express +(x + y) and $(z + u) as sums of members
+
of J U K; likewise for $(xl 2yf)and +(zf + 2u1.)For the second part, assume
we had 4: G -+J , an isomorphism. Show that {x, -x} can be characterized
in G by the set of integers m such that x = mv for some u, and from this fact
deduce that d(x) = x' or -x' and d(y) = y' or -yt.)
7. Prove that the groups G and J of the previous exercise are directly
indecomposable.
8. (W. R. Scott; see E. A. Walker [1956].) Let G be the group given by the
presentation:
and let H be the group presented by

5.1 Introduction and Examples 267
(We will not take up presentations until Volume 2; nevertheless only a

rudimentary understanding of presentations is required for this exercise.)
Prove that G $ H, but that in fact Z x G z Z x H via isomorphisms that
we will now describe. Fix a generator of Z and denote it c. For the iso-
morphism from Z x G to Z x H, define
and for its inverse define
(Thus cancellation fails for finitely generated groups. As we will see in

Theorem 5.9 of 95.4, cancellation does hold (for arbitrary groups G and H) if
Z is replaced by any finite group. Here we see that "finite" cannot be replaced
by "finitely generated" in that theorem.)
9. Let B, denote the two-element Boolean algebra ({O, 11, A , v ,O, 1, -), and
let A be the subalgebra of BT with universe
A = (OE{O, l)m:(3n)(Vk 2 n)a(2k) = a(2k + 1)).
The exercise is to show that B, x B, x A r A, but B, x A $ A. (In the
language of 95.7, A absorbs B, x B,, but A does not absorb B,.) As an easy
corollary, we have A2 z (A x B2), but A $ A x B,. (The first construction
of such Boolean algebras was given by W. A. Hanf in [1957].) According to
a result of R. Vaught, this A must be uncountable. (For a more general result,
see the first corollary to Theorem 5.27 in 95.7 below.) Hint: The more difficult
part is to show that B, x A $ A. To do this, first establish that A is atomic
(and that its atoms are easily described). Now, if there were an isomorphism
4: A -,B, x A, then 4 would effect a one-to-one correspondence between
the set of atoms of A and those of B, x A. Moreover, the atoms are arranged
l
l naturally in pairs, and 4 must respect al1 but finitely many of these pairs.
10. Prove the assertion in the text that every finite algebra is isomorphic to a
product of directly indecomposable algebras.
11. If X is the class of al1 finite algebras of a given type, then every X-prime
algebra is indecomposable. For X the class of finite algebras with no
operations (i.e., finite sets), primality and indecomposability coincide. It
would now be wise to review or rediscover an elementary proof of unique
factorization of natural numbers. (This last part of the exercise is apropos
because such proofs usually begin by establishing that indecomposable =
prime for finite sets.) Next, prove that if the class X of finite algebras has
UFP, then, conversely, every indecomposable algebra in X is X-prime.
Finally, find a class X of finite algebras and an indecomposable algebra in

X that is not X-prime.
12. (B. Jónsson.) Every infinite Boolean group is prime in the class of al1 Boolean
groups. (By a Boolean group we mean a group obeying the law x x Ñ e, which
happens to entail commutativity. It may be helpful to observe that the variety
of Boolean groups is equivalent to the variety of vector spaces over a
two-element field.)
5.2 DIRECT FACTORIZATION AND ISOTOPY
This entire chapter concerns direct representations-Le., isomorphisms
(and some closely connected notions). In 54.4 we saw that an isomorphisrn

f : A -+ n B i is determined within isomorphism by certain congruences a, on A,
namely a, = ker pi o f (i E 1), where pi is the projection of flBi onto B,. (Likewise
for nCj.) This rather elementary observation will be useful and important
throughout this chapter, since it gives us the obvious advantage of working with
one single algebra A and its congruences. In this section we prepare for the
theorems of 55.3 (Birkhoff-Ore, Jónsson), first by reviewing what we require of
54.4 and then by introducing the notion of isotopy, both for algebras and for
congruences, and proving a few lemmas about these notions.
If a, is defined as above, then clearly each Bi is isomorphic, to Ala, by
the map fi : Ala, + B,, where fi(a/a,) = (p, o f ) (a). Thus we can make the blanket
assumption for unique factorization results that we are speaking of a single
algebra A and that factorizations have their factors taken from the collection
{Ale: 0 E Con A}. Now given any r c Con A, we first observe that there is a
natural map
given by (f (a)), = u/@.It turns out to be convenient to consider the slightly more
general situation of the natural map
for some @ E Con A and some family r of congruences 0 2 @.
LEMMA 1. f is one-to-one if and only if nr = $. I t is onto @ r satisfies
If the conditions of Lemma 1 hold-i.e., if f is one-to-one and onto-then

we will say that $ is the product of the congruences 0 E T,and we will write
This notation extends that introduced in Definition 4.30 (for the case of S/ = 0).
For a (finite or infinite) collection enumerated in one-to-one fashion by ordinals,
r = {8,,8,;.-},we will write
Even for two congruences 9, and O,, the direct product 9, x 9, will not usually
exist; if it does exist it will of course equal 8, A O,, but in order for it to exist,
condition (1) of Lemma 1 must hold. In the case at hand, clearly, that condition
states that 9,. 9, = 1 in Con A-i.e., that 9, and 8, permute, and that their join
is 1. We emphasize that whenever we write, e.g., 9, x 9, = S/, we are asserting
both that this product exists and that its value is S/. =
Thus, as we said in $4.4, to each direct product representation f : A G
ni,, Ai there corresponds a product decomposition
in Con A, where ai is the kernel of pi o f and pi is the ith coordinate projection

homomorphism, with isomorphic representations of A yielding the same product
decomposition of O. Under this correspondence, direct product decompositions of
O in ConA are in one-to-one correspondence with isomorphism classes of direct
product representations of A. Recall that an algebra A is directly indecomposable
iff IAl > 1,andA r B x Cimplies IBI = 1 or ICI = 1.
LEMMA 2. For al1 9 E Con A, Al9 is directly indecomposable 8 9 # 1, and

9 = 9, x 9,
implies 9, = 1 or 9, = 1.
Thus we will cal1 a congruence 0 indecomposable iff it satisfies these condi-

tions: 9 # 1, and 9 = 9, x 9, implies 9, = 1 or 9, = 1. Lemma 2 will, of course,
be our key to results involving indecomposable factors.
We continue with a lemma that contains some useful results on product
congruences. In omitting the proofs, we emphasize that the relation = 8, x 9,
is not lattice-theoretical, since the existence of 9, x 8, cannot be determined in
the (abstract) lattice Con A. (See Exercise 1 below.)
LEMMA 3.
n n
i. If r exists, then ,?l exists for all r, c r.
ii. Let us be given un index set I and index sets Jifor each i E I; moreover let us
be given O,, E Con A for each i E I and j E Ji.U neijexists and equals 9, for
each i E I , and if ni
9, = 9, then we also have
iii. The product is completely commutative and associative-i.e., both its

existence and its value are independent of any bracketings or permutations of
the factors, including infinite permutations.
n n
iv. If 6,exists and bi 2 6,for each i, then dialso exists. •
Our next lemma permits a special form of reasoning about x for congruences
on finite algebras. It is a direct analog for finite algebras of assertion (vi) of
Theorem 2.46 in $2.4, and it will be used in a similar way. It will form assertion
(vi)of Lemma 16 in 55.3 below, which plays a role similar to that of Theorem 2.46.
LEMMA 4. If A is finite and $

axal=PxP'=a~P'=a'~P
in Con A, then also a x p' exists and equals a x a'.
Proof. Since we may replace A by A/(a x a'), it is enough to prove the lemma
for the case that a x a' = O. Let m = IAlal, n = IAla'l, p = IAIPI, and q = IAIP'I.
Our assumptions te11 us that A is isomorphic to Ala x Ala' and to AIP x AIF,
and moreover embeddable both in Ala x AIP' and in Ala' x AIP. From these
isomorphisms and embeddings, we obviously have the following equalities and
inequalities:
IAJ 5 np.
These inequalities easily imply that (Al = mq, so the given embedding of A into
Ala x Al/?' is an isomorphism. •
Closely associated with the notions of direct product and factor congruence
are the notions of isotopic algebras and isotopic congruences; we begin with
the first of these. Algebras A and B (of the same similarity type) are said to be
isotopic over the algebra C (of the same type), written A -c B, iff there exists an
isomorphism
such that the second coordinate of b(a, c ) is c for al1 a E A and for al1 c E C. We
-
say that A and B are isotopic, and write A B iff A -c B for some C. Obviously,
isomorphic algebras are isotopic, but the first two exercises of $5.1 give simple
examples of isotopic algebras that are not isomorphic. (In Exercise 2 there, we
have A, - A, but A, $ A,, even though A, has a one-element subuniverse and
is congruence modular.) In 55.7 we will prove the surprising and nontrivial result
(of L. Lovász) that if A x C E B x C, with A, B, and C al1 finite (with no
restriction on the form taken by the isomorphism between A x C and B x C),
then A -cB. (Exercise 8 of 55.1 shows that this is false without the finiteness
condition.)
5.2 Direct Factorization and Isotopy 271
i
LEMMA 5. Isotopy is un equivalence relation on the class of al1 algebras of a
given similarity type.
Proof. It is obvious that --

is reflexive and symmetric, so we need only prove
transitivity. Suppose that A B via $: A x D S B x D, and that B C -
via $ : B X E ~ C X E Define . $ : A x D x E + B x D x E via $(a,d,e)=
(F(a,d), d, e), where $(a, d) = (F(a,d), d), and define $: B x D x E -+C x D x E
via $(b, d, e) = (G(b,e), d, e), where $(b, e) = (G(b,e), e). It is clear that $ and are
isomorphisms, so $ 0 $ is an isomorphism that effects the isotopy of A with C
over D x E. w
LEMMA 6. If A -B,thenIAl = IBI.
We cal1 congruences a and fl on an algebra E isotopic over the congruence y

on E, written a -, P, iff a x y = P x y. We say that a and fi are isotopic in one
step, written a -, p, iff a -, -
P for some y. Finally, we define to be the transitive
--,
closure of -,; that is, we say that a and fl are isotopic, written a p, iff there
exist congruences dl, - 6, such that a = 6,, p = 6,, and 6,
e , -, 6, - 6,. More
specifically, if 6, 6, -- ¿&,then we will say that a and P are isotopic
over y,, y,, . ., y,-,. The connection between isotopy of algebras and isotopy of
congruences is as follows.
LEMMA 7. Algebras A and B are isotopic JfL there exist un algebra E and
isotopic congruences a and fl on E such that A r E/a and B E E/P. In more
detail, f A -c B, then there are congruences a, P, and y on E = A x C such that
E/a E A, E/P E B, E/y E C, and a -,P. And f a, P, y E Con E, with a -,
P, then
E/a "Ely ElP.
Proof. Suppose first that we are given A -c B via an isomorphism $: A x C +

B x C. Take E to be A x C, and let p,, p, be the projection maps from this
product onto A and C. Now define a = ker p,, P = ker(p, o $), and y = ker p, =
ker(pl o 4). We leave it to the reader to check that a, P, and y are as required.
For the final statement of the theorem, we take a, p, and y as given there,
and define $: E/a x E/y -+E/P x E/y as follows: $(x/a, yly) is (w/P, wly), where
a
x-w- Y y. We leave it to the reader to prove that this $ is a well-defined
isomorphism of E/a x E/y with E//? x Ely, which commutes with the second
coordinate projection.
The first sentence follows immediately from the other things proved here,
together with the fact (Lemma 5) that isotopy of algebras is a transitive
relation.
Given congruences on E that are one-step isotopic over y, we will define the
y-projectivity rnap from I [a, 11 c Con E to I [P, 11 G Con E to be the projectivity
map (see Definition 2.26 in $2.3)defined by 6 I+ (6 A y) v P. It is the composition
of two perspectivities: meet with y, and then join with P. More generally, if a and
p are isotopic over y,, - , y,-,, then by the associated projectivity map from
I [a, 11 to I [p, 11 we mean the composition of the y-projectivity maps for y = y,,
y,, . . , y,-, . Of course, if Con E is modular, then the associated projectivity map
is an isomorphism between T[a, 11 and I[P, 11, but in general al1 we know is that
it is an order-preserving map. As in the Correspondence Theorem (4.12), for
6 E I [a, 11, we will take 6/a to denote the congruence on Ela that corresponds
naturally to 6 via the stipulation that (ala, a'la) E 6/a iff (a, a') E 6.
LEMMA 8. If a and /3 are isotopic congruences on E (over y,, . - ,y,-,), then

there exists a bijection 41,: Ela -+ E//? such that
for al1 a, a' E E, for al1 6 E I [a, 11 and for f the associated projectivity rnap from
I [a, 11 to I [P, 11. Similarly,
(blP7b'IP) E LlP (41,-l (blP),

-+ 4-l (b11P))E g(A)la
for al1 b, b' E E, for al1 L E I [P, 11 and for g the associated projectivity rnap from
I [P, 11 to,I [a, 11.
Proof. We will assume that k = 2-i.e., that a and P are one-step isotopic (over
y), that f is the y-projectivity rnap from I[a, 11 to I[P, 11, and g is the y-projectivity
rnap from I [p, 11 to I [a, 11. Clearly the general case follows easily from this one.
Fix an element e of E. Define 41,(a/a)to be b/P for any b such that
The existence of such an element b comes from the fact that a x y is defined, and
hence that a o y = 1. Moreover, if a is changed to a' in the same a-class (i.e.,
(a, a') E a), and if b' is any element satisfying
a' b' Y e,
then we clearly have (b, b') E a A y < P, and so 4 is well defined as a rnap from
Ela to EIP. To see that 41, is a bijection, we will show that the rnap S/ constructed
symmetrically to 41, is a two-sided inverse to 4. That is, we define S/: ElD -+ Ela
by defining S/(b/P)to be the a-class of a, where
(It follows as before that ala is determined uniquely by the class blP.) There is
an obvious symmetry to this situation. Therefore, to see that S/ is a two-sided
inverse of 4, it will be enough to show that if d(a/a) = b/P and S/(b/P) = a'la, then
a a a'. Since by definition we have
and
P
b-afY e,
we clearly also have
/
1
(a, a') E a v ( p A y) = a v (a A y) = a,
as was to be proved.
1 Now to verify the implication involving the y-projectivity map f, we will
assume that (ala, a'la) E ñ/a, and prove that (b/P, b'/P) E f (6)/P, where b and b' are
defined via
1 From this diagram, and the fact that a 5 ñ, it is evident that (b, b') ~ñ A y, and
thus that (b, b') E (6 A y) v p = f (6.).Therefore, (b/P, b'/P) E f (6)/P, as desired. This
1 proves that if (ala, a'la) E 6/a, then (4(a/a), 4(a1/a))E f (6)lP. The final sentence of
the lemma is proved similarly. m
LEMMA 9. In the situation of Lemma 8, if the algebra E-has a one-element

subuniverse, then 4 can be taken to be an isomorphism 4: Ela 3 EIP.
Proof. We continue the proof of Lemma 8, while assuming in addition that (e}
is a subuniverse of E. To show that 4 is a homomorphism, we consider an n-ary
operation F of E and elements a,, - a,-, E E. Let us further suppose that
a ,
so that 4(ai/a) = bi/P for each i. Now, since a and y are congruences, we have
1 which reduces, since {e} is a subuniverse, to
These last congruences imply that 4 maps F(a,, - ,a,-,)/a to F(b,, . ,b,-,)/P.
Therefore, since a and P are congruences on E, we also know that maps
F(aola, ., a,-,la) to F(b,IP, - ., bn-,lb) = F(4(a,la), ., 4(a,-,la)), and thus 4
is a homomorphism. m
We are able to make the most effective use of isotopy when it occurs in
combination with the modular law. We therefore define algebras A and B to be
modular-isotopic over C, and write A -ydB, iff A -C B and Con(A x C) is a
modular lattice. We say that A and B are modular-isotopic in one step, and write
A -yod B, iff A -Eod B for some C. Finally, we define -"Od to be the transitive
closure of -yd.That is, we say that algebras A and B are modular-isotopic, and
write A -"Od B, iff there exist algebras D I , D, such that A = D I , B = D,, and
-
D, lod -
D2 ?Od . . -lod
m . ,
Dk. We have the following counterpart to Lemma 7.

LEMMA 10. Algebras A and B are modular-isotopic in one step $ there exist
a congruence modular algebra E and one-step isotopic congruences a and P on E
such that A r E/a and B r E/P. In more detail, ifA -FdB, then E = A x C has
modular congruence lattice, and there are congruences a, P, and y on E such that
E/a z A, E/P z B, E/y g C, and a P. Conversely, $Con E is modular, and a, P,
y E Con E, with a N,
P, then E/a E/P. m
LEMMA 1l. If A -"Od B, then Con A r Con B.
Proof. For one-step modular isotopy, this follows immediately from the fact
(Dedekind's Transposition Principle, Theorem 2.27) that, under modularity, the
y-projectivity map is an isomorphism between I[a, 11 and I[P, 11. The general
case is then immediate from the transitivity of r . m
LEMMA 12. If A -ydB, then A -FdB for some C E HS(A x B).
Proof. By Lemma 10, there are a congruence modular algebra E and congru-
entes a, p, and y on E satisfying a x y = P x y, and such that E/a z A and .
E/P r B. We claim that, in addition, a x 7 = /? x 7, where 7 = y v (a A P). The
existence of a x 7 and p x 7 is immediate from Lemma 3(iv) and the obvious fact
that 7 2 y, so we need only show that a A 7 = P A 7.For this we have the
following calculation, which uses the modular law:
Thus the claim is established. Another application of Lemma 10 tells us that

A - p d B , where C is E/7. It remains only to show that CEHS(A x B). The
natural maps of E/(a A p) onto E/a and E/P obviously separate points of
E/(a A p), so E/(a A p) is isomorphic to a subalgebra of E/a x E/P, hence to a
subalgebra of A x B. Finally, C = E/? is a quotient of E/(a A P), and hence is in
HS(A x B). m
-,
LEMMA 13. Let a, P, y E Con E, with Con E modular. If a P, then the y-
projectivity map f : I[a, 11 -% I [P, 11 respects al1finite factorization relations. In
particular, for 6, do, , E I [a, 11 S Con E, 6 is directly indecomposable ifand
only if f (6) is directly indecomposable; 6 = 6, x x Jk-, if and only if f (6) =
f (6,) x . x f(dk-,); and 6 is isotopic to f (6).
Proof. We will base our proof on the facts that f is an isomorphism (the
Dedekind Transposition Principle, Theorem 2.27) and that there exists a bijec-
tion 4: E/a 4 E/P such that (ala, bla) E 6/a t,(4(a/a),b(b/a)) E f (6)lPfor al1 a, b E E
and for al1 6 E I [a, 11 (Lemma 8).
We first prove the simple fact, for A, p 2 a, that if A o p = 1, then f (A)o f (p) =
1. To see this, we will assume that A 0 p = 1 and take arbitrary c, d E E. We need
to show that c -f - (A) f ( ~ )
- y --- d for some y E E. Choose a, b E E such that 4(a/a) =
c/P and d(b/ct) = d/P. Since A o p = 1, we know that a x Y b for some x.
Choose y so that d(x/a) = y/P. Now, by the definition of the relation v a , we have
(ala, xla) E Ala, so by our assumption on 4, we have (4(a/a),4(x/a))E f (A)/P. By
our previous choices of a and y, this means that (c/P, y/P) E f (A)/P, and so, by the
definition of the relation f(A)/P we have (c,y) E f(A). A similar argument shows
that (y, d) E f (p), SO y is as required-that is, we have shown that if A O p = 1,
then f(A) o f(p) = 1. In fact, the reverse implication holds as well, by an almost
identical argument, so we have A o p = 1t,f (A) o f (p) = 1.
The conclusions of the lemma are now almost immediate. For instance, if
6, x S, = 6,, then we have f (6,) A f (6,) = f (6,) from the fact that f is an
isomorphism of lattices, and f (6,) o f (6,) = f (6,) o f (6,) = 1, from what we have
just proved. These two facts together te11 us that f(6,) x f(6,) = f(6,). M
As a final preparation for the results of $5.3, we ask the reader to take a
second look at the lemma before Theorem 4.71 and the remarks immediately
preceding that lemma.
Exercises 5.2
1. Find an algebra A with congruence lattice
such that A is not isomorphic to A/O, x A/O,. (This supports our claim that
the relation 4 = O, x O, is not definable within lattice theory.)
2. Prove Lemma 3.
3. Let O,, O, be congruences on a finite algebra A. O, x - - - x O, exists iff, for
m ,
each i = 2, - . n, (O, A - - - A Oi-,) permutes with Oi and (O1 A . - A

S , v
Oi = 1. (For an analog in modular lattice theory, see the next exercise.)
4. In a modular lattice L of finite height ($2.4), {a,, a,) is a directly join
S ,
independent subset of L iff, for each i = 2, - n, (a, v - -. v a,-,) A ai = 0.

a,
5. Let A be a five-element algebra with no operations. Find congruences O,, O,

O, on A such that each B, permutes with each Oj, each quotient A/Oi has exactly
two elements, 6, A 13, A 83 = 0, and
(O1 A 8,) v O, = (O2 A O,) v O1 = (O3 A 8,) v O, = 1.
(Clearly, we cannot have O = O, x O, x 83, so this exercise places some limits
on possible weakenings of the hypotheses for Exercise 3.)
In this section, we will prove the following two unique factorization results, which
follow from Ore's Theorem and its generalization, the Direct Join Decomposi-
tion Theorem (2.47). The first of these comes from the second [1948] edition of
Birkhoff's Lattice Theory; since it relies so heavily on Ore's Theorem, it is
generally called the Birkhoff-Ore Theorem.
THEOREM 5.3. (G. Birlzhoff[l948].) If A has permuting congruence relations,

ConA has finite height, and A has a one-element subuniverse, then A is uniquely
factorable.
Our other unique factorization result for this section is Bjarni Jónsson's
modification of the Birkhoff-Ore Theorem to hold for finite congruence modular
algebras.
TNEOREM 5.4. (B. Jónsson [1966].) If A is finite, Con A is modular, and A

has a one-element subuniverse, then A is uniquely factorable.
It remains open whether the natural common generalization of Theorems

5.3 and 5.4 is true. 1
PROBLEM. If ConA is a modular lattice of finite height, and A has a one-
element subalgebra, then must A be uniquely factorable?
REMARK: Our proof of Theorem 5.4 requires finiteness of A in one place only,
namely for proving Lemma 4 above, which enters into our proof as assertion (vi)
of Lemma 16 below. If Lemma 4 should turn out to hold for Con A modular and
of finite height, then we would have a positive answer to this problem. Exercises
1-6 of 55.1 show that none of the three hypotheses (congruence modularity, finite
height, one-element subalgebra) can be completely removed, either from the
problem or from 5.3 or 5.4. Even in the absence of a one-element subalgebra, we
still have modular-isotopy versions of Theorems 5.3 and 5.4. (Recall that modular
isotopy of algebras, denoted by -"Od, was introduced in $5.2.)
THEOREM 5.5. Suppose that Con A has finite height and the congruences of A
permute. Then A has a product representation with directly indecomposablefactors,
and al1 suchfactorizations have only finitely many factors. Moreover, if
A r B, x x B, r C, x x C,,
with al1 .factors directly indecomposable, then m = n and, after renumbering,
Bi -"Od Ci for 1 5 i 5 n. Consequently, for each i, 1 Bil = 1 Gil and Con Bi r
Con Ci.
THEOREM 5.6. If A is finite and Con A is modular, and if

5.3 Consequences of Ore's Theorern
with al1 factors directly indecomposable, then m = n and, after renumbering,

Bi wmodCi for 1 5 i 5 n. Consequently, for each i, 1 Bil = 1 C, 1 and Con B,
Con C,.
REMARK ON THE PROOFS: Since Theorem 5.3 is an obvious corollary of
some remarks occurring in the proof of Theorem 5.5, we will proceed as follows.
We will first prove 5.5 and then give a brief proof of 5.3 that refers to the proof
of 5.5. Likewise, Theorem 5.4 follows easily from the proof of Theorem 5.6, so
we will follow a similar strategy for proving those two theorems.
Notice that there is also an analog for modular isotopy to the problem stated
just after Theorem 5.4. The proof of Theorem 5.6 will require two more lemmas
about congruence relations on A for Con A modular, but we can prove Theorem
5.5 immediately. Al1 our lemmas refer to the sequence of lemmas begun in the
previous section. This proof may be thought of as a continuation of the proof of
Theorem 4.71 of $4.7.
Proof of Theorem 5.5. We already established in Theorem 4.71 that A has a

product representation with directly indecomposable factors, that al1 such fac-
torizations have only finitely many factors, and that the number of factors is
invariant (i.e., m = n in the theorem). By Lemmas 6 and 11, the final sentence of
the theorem follows immediately from the modular isotopy of Bi and Ci, and so
we need only prove the main conclusion, namely that Bi -"Od Ci for 1 I i I n.
As noted at the beginning of 55.2, we may assume that each Bi is A/&, with
Pl x x /3, = 0, and that each Ci is A/yj, with y, x - x y, = O. According to
Theorem 4.67, ConA and its dual lattice are finite dimensional modular lat-
tices, and by the lemma before Theorem 4.71, we know that P1 0- - @ P, =
y, 0 O y, = 1 in the dual of Con A. Therefore, by the Direct Join Decomposi-
tion Theorem (2.47), the yi (and, correspondingly, the Ci) may be renumbered in
such a way that, for each i, y, is directly join isotopic to Pi in the dual of Con A.
That is, for each i, we have congruences 6, = Di, al, ,6, = y, (for some k) such
that for each j, aj is one-step directly join isotopic with dj+, in the dual of
Con A. By the definition of this concept, we have 6, v Oj = a,+, v 9, and 6, A 9, =
Jj+, A Oj = O in the dual of ConA for each j. In other words, 6, A Oj = ¿ij+i A Oj
and 6, v 0, = 6j+, v 9, = 1 in Con A. Since Oj permutes with 6,, this last equation
tells us that 6, O 9, = 4 0 S, = 6j+, 0 9, = Bj 0 S,+, = 1, which in turn yields 6, x 9, =
bj+i x gj for each j, and hence Ci -, - -
Ci+, . Therefore, we have 6, 6,, that is Pi yi.
It now follows from the last sentence of Lemma 10 that Bi -"o~C,, thereby
completing the proof of the theorem.
Proof of Theorem 5.3. We continue the last paragraph of the previous proof,
-
under the assumption that A has a one-element subuniverse. From the fact that
p, y, for each i, together with Lemma 9, we deduce that A/Pi r A/yi for each i.
In other words, B, r Ci for each i. M
Without permutability, we cannot characterize the relation a = P x y in

pure lattice theory as we did before (by the two equations a = P A y and 1 =
p v y). There is one special case, however, in which modularity permits us to

deduce that two congruences permute. We present this special case in the next
lemma. In Lemma 15 we will apply it to obtain a product decomposition for one
special case, which will enter the main argument stream as part (viii) of Lemma
16 below.
LEMMA 14. (B. Jónsson.) If ConA is modular and a x a' =8 I P in Con A,

then a A p permutes with a'.
Proof. Suppose (a, b) E (a A P) o a',-that is, (a, x) E (a A P) and (x, b) E a' for
some x E A. Thus we have the following diagram
where the existence of y follows from condition (1)in Lemma 1 of $5.2. To obtain
(a, b) E a' o (a A p), and thus complete the proof, it will be enough to prove that
(y, b) E p. To see this, we note that (y, b) E a A (a' v (a A p)), and then use modu-
larity to compute
a A (a' v (a A p)) = (a A a') v (a A p)
= O v ( a ~ p)IPv(a~p)=P;
and hence (y, b) E p. ¤
LEMMA 15. Let a, a', 8, and y be elernents of Con A, and assume that Con A is
modular. Suppose that a x d and (d v (a A P)) x y exist, and that y 2 a A p and
p 2 a A a'. Then a' x y also exists.
Proof. We need to prove that a' o y = 1 and y o a' = 1 in Con A. The two proofs
are very similar, and so we present only the first. Given any a, b E A, we need to
find x E A such that (a, x) E a' and (x, b) E y. By hypothesis, (a' v (a A P)) 0 y = 1,
so there exists y E A such that
Now a' permutes with a A p, by Lemma 14, so we have
Since a A p y, we readily obtain

a'
a x - b,
which is the desired conclusion.
i
5.3 Consequences of Ore's Theorem
1 In order to prove Theorem 5.6, we will modify slightly the proof used for
Theorem 5.5. The main change is that we need to invoke a version of the Direct
i
1
Join Decomposition Theorem (2.47) that applies to the isotopy relation
congruences, introduced in $5.2, rather than to isotopy in the sense of lattice
-
for
theory, introduced in $2.4. Let us recall that in proving the Direct Join Decom-
j position Theorem, we based our argument on only those facts about (lattice-
1 theoretic) join independence, and its associated notion of isotopy, that are
contained in the ten conditions of Theorem 2.46. Therefore, to have a version of
the Direct Join Decomposition Theorem that applies to isotopy of congruences,
we need only prove the counterpart of Theorem 2.46 for the direct product notion
of independence and its associated notion of isotopy. We now define a set
r G Con A to be independent iff the product of r exists, in the sense of $5.2, that
1
is, iff condition (1) of Lemma 1 ($5.2)holds. We will let IND(A) denote the family
I
I
of al1 independent sets of congruences on A. We now prove such a counterpart,
which, for convenience, we have stated in a form dual to that of 2.46. The
finiteness assumption is actually required only for assertion (vi).
! LEMMA 16. Let A be a finite algebra with Con A modular.
i. If N E M E IND(A), then N E IND(A).
ii. If M E IND(A), then M U (1) E IND(A).
iii. If a x B E M E IND(A), then (M - {a x P) ) U {a,P) E IND(A).
iv. If a, p E M E IND(A) and a # /?, then (M - {a,P)) U {a x /?) E IND(A).
v. If M E IND(A) and f : M -+ Con A such that f (x) 2 x for al1 x, then
{ f (x): x E M) E IND(A).
vi. I f a x a ' = p x p l = a r \ p l = a f ~ p , t h e n ax p = a ' x / ? = a x a'.
vii. If {a,a') E IND(A) with a # a' and P > a, then fi x a' > a x d.
viii. If p 2 a x a', p 2 a, and (a v (a' A p), p) E IND(A), then {a,P) E IND(A).
ix. If a = a x p, then p = 1.
X. - - -
If a x P y, then there are a' and P' such that y = a' x j', a a', and P P'.
Proof. The ninth assertion, and the first four, are obvious from the definitions
involved and from the elementary remarks in $5.2. The fifth assertion is a restate-
ment of clause (iv) of Lemma 3, and the sixth is immediate from Lemma 4. The
seventh assertion holds in al1 algebras (see Exercise 1 below). The eighth asser-
tion is immediate from Lemma 15, and the tenth follows immediately from
Lemma 13.
THE DIRECT JOIN DECOMPOSITION THEOREM. (Second version.)

Let A be a finite algebra with Con A modular. If P,, - Pm,y,, - . , y,, E Con A and
S , ,
8, x - x pm y, x .- x y,,, with each pi and each yj directly indecomposable,

N
-
then m = n, and, after renumbering the y,, we have Pi y, for each i.
Broof. The proof is exactly the same as the one we gave for the original Direct
Join Decomposition Theorem (2.47),with the following modifications. The orig-
inal proof must be dualized, the notions of IND and "isotopic" from Chapter
2 must be replaced by the corresponding notions from this chapter, and finally,
each allusion in the original proof to one of the ten assertions of Theorem 2.46
must be changed to refer to the corresponding assertion of Lemma 16 here.
m
Proof of Theorem 5.6. By Lemmas 6 and 11, the final sentence of the theorem
follows immediately from the modular isotopy of B, and Ci, so we need only
prove the main conclusion, namely that B, wmodC, for each i.
By definition of wmod,we have a chain of algebras E, -,"Od E, -yd . -,"Od
E,, where E, is B, x - x B, and E, is C, x x C,. By Lemma 11, we know
that Con E, z Con A for each i, and by Lemma 6, we know that each Ei is finite.
Therefore, we may assume that each E, is a finite algebra with a modular con-
gruence lattice, and has been written as a product of indecomposables. Clearly,
if we prove the conclusions of the theorem for each pair E,, E,,,, we will then
have the full conclusion for Bi and C,. In other words, it will suffice to prove the
theorem under the simpler assumption that B, x x B, -ydC, x x C,.
Under this assumption we have, by Lemmas 10 and 12, a finite congruence
and E/y r C, x - a - -
modular algebra E and congruences P, y on E such that E/P z B, x
x C, and P y. By Lemma 1, we have P = P, x
x B,
x P,,
with Bi r E//&for each i, and likewise y = y, x - - x y,, with Cj z E/yj for each
j. We therefore have P, x - x P, - y, x x y,, so the conclusion of the
theorem is immediate from the Direct Join Decomposition Theorem (and
Lemma 10). m
Proof of Theorem 5.4. We look over the previous proof under two assumptions:
that A has a one-element subuniverse, and that we actually have A r B, x - x
B, r C, x x C,. In this case, we may take E = A, and p = y = O. Thus, as
-
in the previous proof, we have yi Pi for each i. Therefore, A/Pi r Aly, for each
i, by Lemma 9, which is to say that B, r Ci for each i. m
REMARK ON THE PROOF: We have perhaps made it look as if Theorem

5.4 (of B. Jónsson) were a simple corollary of the lattice theoretic results of
O. Ore. If we have given such an impression, it is a misleading one. In fact,
Jónsson's proof required a very careful treatment of Ore's Theorem, including
some substantial improvements. In order to present a unified development, we
decided to incorporate those improvements into our statement of the Direct Join
Decomposition Theorem in Chapter 2.
Exercises 5.7
l. Prove assertion (vii) of Lemma 16: If a, a', and P are congruences on an algebra
A such that a x a' exists and /? > a, then fl A a' > a A a'. (Lemma 16 assumes
finiteness and modularity, but they are not needed for this part, which holds
5.3 Consequences of Ore's Theorem 281
for any algebra A, whether finite or infinite, and with no assumption of

modularity.)
In Exercises 2-8 below, we consider a finite dimensional vector space

V = ( V; + ,r,(L E F) ) over an algebraically closed field F, together with a linear
transformation T: V -+ V. We analyze T by considering the algebra V, =
( V ; +, rA(LE F), T). Clearly, every finite product of such algebras is again an
algebra of the same type (i.e., a finite dimensional vector space W over F,
expanded by a single linear endomorphism of W), and a similar result holds for
direct factors. These algebras V, clearly satisfy al1 hypotheses of the Birkhoff-
Ore Theorem, so each V, has a product representation with directly indecom-
posable factor algebras, which are unique up to rearrangement and isomorphism.
As we will see in these exercises, the indecomposable factors of V, correspond
exactly to the elementary Jordan matrices into which the matrix of T can be
decomposed.
Let us call T (or the corresponding algebra V,) m-nilpotent iff Tm= 0,
and nilpotent iff it is m-nilpotent for some m. We call T (or V,) multicyclic iff
V has a basis {u!: 1 i i 5 k, 1 5 j 5 m(i)} such that T has the following form
when restricted to elements of this basis:
...r,v:5,; 5 0
...1,v;r,v;r,o
..T*v;5v; 5 0 .
Finally, let us call T (or V,) cyclic iff it is multicyclic with k = l.
2. If T is multicyclic, then k and m(i) (1 i i i k) are determined by the isomor-

phism class of V,, and conversely, these integers determine the isomorphism
type of V,.
3. The product of two multicyclic algebras is multicyclic (with the path lengths
m(i) of the product easily derivable from those in the two factors).
4. Every multicyclic algebra is a product of cyclic algebras.
5. A cyclic algebra of more than one element is directly indecomposable. Hence
a multicyclic algebra V, is directly indecomposable iff 1 V )# 1 and V, is
cyclic.
*6. V, is nilpotent iff V, is multicyclic (and, in fact, the least m such that Tm= O
is the largest of the m,). (Hint: Multicyclic -+ nilpotent, almost trivially.
For the converse, suppose that Tm= 0, but T"-' # O. Start by choosing
u,: vy, - so that vl/ker(Tm-'), v2/ker(Tm-l), form a basis of the quotient
m
space V/ker(~"-l). Define u:-' = T(vy), u,"-' = T(vy), and so on, and
prove that v;"-'/ker(~"-~),vY-'/ker(~"-~),. are linearly independent in
k e r ( ~ " - ' ) / k e r ( ~ " - ~Take
). as many further elements u?-' as are required
to make v ? - ' / k e r ( ~ ~ - ~~,"-'/ker(T"-~),
), form a basis of the space
k e r ( ~ " - ' ) / k e r ( ~ " - ~ )Continue
. this procedure to define al1 vi. It is n
necessary to prove that the collection of al1 vf forms a basis of V.)
7. If VT is directly indecomposable, then T = S + 1for some cyclic S and some

A E F. (Hint: By finite dimensionality, some nontrivial linear relation must
hold between the various powers I = TO, 7; T2, T 3 , . Let p(x) be a nonzero
polynomial of smallest degree in F[x] such that p(T) = O. Taking A to be a
zero of p(x) in F, we have p(x) = (x - A)Mq(~) with M 2 1 and with A not
a zero of q(x). Since F[x] is a principal ideal domain, we have 1 =
r(x)(x - A)M+ s(x)q(x). From this we easily deduce that V, is isomorphic
to (ker(T - x (ker q(T)),. Since q(x) has degree smaller than that of
p(x), we know that q(T) # O (i.e., ker q(T) # V). Hence, if indecomposable,
V, is isomorphic to (ker ( T - A)M),.)
Therefore, every finite dimensional VT is isomorphic to a direct product of

algebras on which T is A plus a cyclic transformation (for some A). From this it is
easy to deduce the Jordan normal form for T.
8. (Jordan normal form.) If V is a finite dimensional vector space, and T is an

endomorphism of V, then there exists a linear basis for V under which T
takes the form
where each A iis a square matrix of the form
5.4 ALGEBRAS WITH A ZERO ELEMENT
Earlier in this chapter, we saw how some hypotheses on the congruence lattice
could yield unique factorization results. In this section, we consider instead the
endomorphism monoid. If A is an Abelian group, then EndA has structure
beyond that of an Abelian group, as mentioned in Exercise 1.6(6);it is in fact a
ring. It turns out that a useful fragment of this ring structure persists under much
weaker assumptions on A: it suffices to assume that A has a binary operation +
and a one-element subuniverse (O) such that the equations O + x Ñ x + O Ñ x
hold in A. Throughout this section and the next, we will consider only algebras
5.4 Algebras with a Zero Element 283
A = (A, +,O, - - ) of this type. For short, we will cal1 such an algebra A an
algebra with zero.
Our principal result in this section will be the theorem ofSB.Jónsson and A.
Tarski that finite algebras with zero are uniquely factorable. (Some elaborations
of this result will come in the next section.) Their starting point was a simple way
to identify direct factorizations within the endomorphism "ring" of an algebra A
with zero; this is a straightforward generalization of the well-known technique
of viewing a direct product G x H of two Abelian groups as the "direct sum" of
its two subgroups G x {O) and (0) x H. To see the general case, let us be given
an algebra A with zero and a direct product decomposition
in Con A. Now, for i = 1, n, define fi: A -+A as follows: f,(a) is the unique
S ,
+
u, E A with (u,, a) E ai and (ui,O) E aj for each j i. The reader should establish (see
Exercises 1 and 2 below) that each fi is an endomorphism of A, and that together
the fi's satisfy the following conditions:
(Here, multiplication of functions means composition, and functions are added

+
pointwise: (f + g)(x) is by definition equal to f(x) g(x). The identity function
is denoted 1, and O denotes the constantly O function.) Conversely, given f,, - m ,
f, satisfying (2), then, as the reader may establish,

1 (3) O = kerf, x - - - x ker f,,
and our two procedures are reciprocal, thus yielding a bijection
Since the order of occurrence of f,, ,f,makes no difference in (3), the same
must be true for (2). By analogy with Abelian group theory, we will abbreviate
(2) as
l HISTORICAL NOTE: The original 1947 treatment by Jónsson and Tarski

was concerned with the ranges Bi = L(A) of the endomorphisms L. Here A is the
inner direct sum of its subalgebras Bi, written
in the sense that each a E A is uniquely expressible as a sum ( . (b, + b,) + +e )
b, with b, E B, for each i. We leave it to the reader to examine the equivalence of

this notion with our notion concerning endomorphisms. (We outline the ideas
and precise definitions in Exercises 4-6 below.) For our purposes, it will be more
convenient to deal with the endomorphisms directly. (This apparently was also
true for the later writings of Jónsson on this subject; by 1964 he was writing in
terms of the endomorphisms.)
284 Chapter 5 Unique factor iza ti^^
We may also ask for conditions in terms of endomorphisms that are equiva-
lent to indecomposability of factors. Now the general notion of the decomposa-
bility of a, namely a = P x y from 55.2, is not amenable to a treatment via
endomorphisms, since a may not be the kernel of any endomorphism. But if a is
itself a direct factor (i.e., if a = ker f, where f @ f ' = l), then we do have an
endomorphism theoretical characterization of decompositions of a.
LEMMA l. Let f @ f 1 = 1 . If 1 = q 5 @ $ @ f 1 and # + S / = f, then kerf =

ker 4 x ker $. In fact, the correspondence (4, $) ++ (ker 4, ker $) is a bijection
between
and
((B, Y):k e r f = P x Y}.
Proof. See Exercise 7 below. ¤
Our ringlike structure on EndA arises from the obvious binary opera-
tion on endomorphisms formed by pointwise addition (as we had it above):
(f + g)(x) = f(x) + g(x). Unfortunately, End A is not usually closed under this
addition. Moreover, + does not usually obey any laws like associativity or
commutativity. Our next lemma will describe some special situations where the
sum of two endomorphisms is again an endomorphism, and some situations
where associativity and commutativity hold. The lemma is phrased in terms of
functions A -+ A, so that one can see these results as mirroring a fragment of ring
theory. In many applications, we will apply them to situations where a, b, and c
are constant functions. Thus we will often think of a, b, and c in the next lemma
as elements of A. The next three lemmas will be in this ring theoretical spirit.
Remember that in this case we are writing f 0 g simply as fg.
As usual in unique factorization work, we ultimately intend to compare two
direct factorizations of a given algebra A (with zero). Thus, for the remainder of
this section, we will assume that
LEMMA 2. Let a, b, c and d be any functions A + A, and let S, t, f, f ', g, g', h, h'
be endomorphisms of A with 1 = f @ f ' = g @ g' = h O h'. Then
(4) a(bc)=(ab)c, al=la=a,
Oa = O, (a + b)c = ac + bc,
a+O=O+a=a, s(a+b)=sa+sb,
so = o.
(5) fa = fb and f 'a = f 'b imply a = b.
(6) fs + f 't is an endomorphism of A.
5.4 Algebras with a Zero Element
(7) s + hfgf ' t is an endomorphism of A.

(8) a + hfgf ' b = hfgf'b + a.
(9) (a+hfgf'b)+c=a+(hfgf'b+c).
(10) There exists a function x : A -+A such that x hfgf ' b = 0. +
(11) If c = hfgf 'd, or if c is any finite sum of maps of that form, then c is
cancellative-i.e., a + c = b + c implies that a = b.
(12) fg'fg = f g f f g .
(13) fgfg'f = fg'fgf, i.e., fgf commutes with fg'f.
/ Proof. Assertions (4) and (5) are obvious. For (6),let F be any n-ary A-operation,
j and select x,, x,EA. We need to see that a = ( f s f't)(F(x,;..,x,)) and
. . S , +
+
/3 = F( fs(x,) f ' t ( x , ) , ,fs(x,) +
f 't(x,)) are equal to each other. For this, we
make the following calculation, using the fact that f and s are endomorphisms
1
1
/
of A:
f (P>= Jz'(fs(x1) + f 't(x,), ,fs(xn) + f 't(xn))
= F ( f 2 ~ ( ~+1( f f ' t ( x i ) ., f 2 ~ ( ~ +n )(ffrt(xn))
= F(fs(x,),. ..,fs(xn)) = fsF(x,,.. ,x,)
1I +
= ( f 2 s ( f f f t ) F ( x 1- ,., x,) = f(a).
A similar calculation yields f ' ( a )= f ' ( j ) ,and so a = P, by (5).

We now give an argument that will yield (7), (8), and (9);it begins with a
long calculation. (Here a is introduced as an abbreviation for the expression that
appears in the previous line; similarly, when we introduce j a few lines below, it
is meant as an abbreviation for the expression in the line that precedes it, and
likewise for y and 6 farther on.)
fa + ( f ' b + f e ) = fCfa + ( f ' b + fc)l + f'Cfa + ( f ' b + fc)l
= + + + +
[ f ' a ((ff'b f2c)] [ f ' f a ( f t 2 b f'fc)] +
= + + + +
[ f a (O fc)] [O ( f ' b O)] +
+ +
= ( f a fc) f'b
= a.
Similar1y,
fa + ( g f ' b + fc) = gCfa + ( g f ' b + fc)l + s'Cfa + (gf ' b + fc)l
= ga + g'f(a + c)
i = P,
I and
1
1
a + ( f g f ' b + c) = f [a + ( f g f ' b + 4 1 + f ' [ a + (fgf'b + 4 1
I =f j + ff(a+c)
and finally,
a + (hfgf'b + c) = h[a + (hfgf'b + c)] + hf[a + (hfgf'b + c)]
= hy + hf(a + c)
= 6.
Now to prove (7), we examine these calculations for a = S , b = t, and c = 0.

Successive applications of (6) yield first that a is an endomorphism, then that P
is an endomorphism, then y, and finally that 6 is an endomorphism, thereby
proving (7). To prove (8) and (9), we first note that almost identical calculations
will establish that
(f'b + fa) + fc = a
(9f 'b + fa) + f c = P
(fgf'b + +
a) c = y
(hfgf'b + a) + c = 6.
Our two different formulas for 6 yield 1

a + (hfgf'b + c) = (hfgf'b + a) + c.
Taking c to be O yields the special case (8) of commutativity. Finally, (8) can be
applied to this last equation to yield equation (9).
For (lo), we simply take x = hfg'f'b and calculate
x + hfgf 'b = hfg'f'b + hfgf 'b(g + g')f 'b
= hf
= hf lf'b = hff'b = hOb = 0.
For (ll), we will assume that c = E;==,

hihgiLfbi(and, by (8) and (9) it does
not matter precisely how this "sum" is associated). By (lo), for each i we have an
additive inverse xi for hifigif,'bi. Taking y = (. (x, + x,) + - - - ) + x,, one easily
checks, using (8) and (9), that
for al1 a. It is now immediate that c is cancellative.

For (12), we calculate
and
Since fg'f'g is cancellative by (1l), we obtain fg'fg = fgf'g.

Finally, (13) follows easily from (12):
The next lemma was proved for groups by H. Fitting in 1934 and later 1
j
extended to algebras with zero by J. LOS.Recall that we are still assuming that
J
l
f and g are endomorphisms of A such that 1 = f @ f ' = g @ g'. A function
4: A + A is called idempotent iff 42 = 4.
LEMMA 3. ( n 2 1) If 4 = ( f g f ) " is idempotent, then there exists an endo-
morphismSof Asuchthat f = 4 + S a n d 1 = $ @ S @ f r .
Proof. Define ,
(Notice that by (9) we do not need further parentheses.) By (7), q is an endo-

morphism. We now claim that q + 4 = f. In the following calculation, we make
free use of (8) and (9):
This is the same expression we began with, except that the top exponent has been
reduced by l . We may continue this procedure until we obtain
A similar calculation yields 4 + q = f. We now take the desired to be q2 and

claim that bS/ = $4 = O. To see this, we first observe that dS/ = $4 follows from
4q = q4, which in turn follows from (13). Then for 4S/ = O we calculate
Now 4q is cancellative by ( l l ) ,so we obtain #S/ = 0.

Now q42 = 4 by hypothesis. To obtain S/2 = S/, we first calculate
q3 = o + q3 = 4q2 + qq2 = (4 + q)q2 = fq2 = q2.
Hence,
$2 = q4 = q2 = S/.
j It is now easy to obtain 4f' = S/f' = f ' 4 = f 'S/ = O. For example, 4f' = ( 4 f )f ' =
,1 4 ( f f 1 )= 4 0 = 0, and similarly in the other cases.
Al1 that remains to be proved is that 4 + S/ = f. We first observe that
l
and so
l 4 + S/ = 4 + q2 = (4 + $4) + q2 = 4 + (4q + q 2 ) (by (9))

= (4 + 44) + (44 + q 2 ) = (4 + d 2 = f 2 = f . m
1 LEMMA 4. 1 = f @ f ' = g @ g'; ( n > 1). If (fgf)" = f , then there is an endo-

1 morphism S/ such that (gfg)" + S/ = g, and the following two equations hold:
Chapter S Unique Factorization
1 = (gfg)" o $ o 9'
1 = (fg)" o C(f9f )"-'9' + 1If 'a
Proof. One immediately calculates that (gfg)" is idempotent, so the facts about
$ follow from the previous lemma (with the roles of f and g reversed). For the
final equation, the idempotence of (fg)" is immediate from (fgf)" = f, and the
idempotence of the second summand is easily calculated from the fact that it has
the form (f h + 1)f '; we omit the details. For the products of the two summands,
we have first the obvious equation
and then we calculate
Finally, we need to see that the two summands in the second equation really do
add to 1:
(fs)"+ [(fsf )"-'9' + llf'
= C(f9)" + (fsf )"-'slf'l + f ' (by (9))
= (fg)"-'Cfs + fslf'l + f '
= (f9)"-' C(f9f + fsf ') + fslf'l + f '
= (fg)"-'Cfsf + (fsf' + f9lf')l + f ' (by (9))
= (f!J)"-'Cfgf + 01 + f '
1
=f + f1=1. l
1
i
We now turn away from the pure "ring" theory of endomorphisms and look
instead at how endomorphisms interact with the algebra A and its congruence
relations. Thus, we will continue to have 1 = f f ' = g @gr. Moreover, we
adopt the notational convention that a = ker f, a' = ker f ', P = ker g, and P' =
ker g'.
LEMMA 5. If (fgf )" = f,then there are congruentes y and y # on A such that
P = y x y # , and
O =y x a'.
In fact, we muy take y to be ker fg.
Proof. This will follow immediately from Lemma 4 (using Lemrna l), as soon as
we establish that ker fg = ker(fg)" = ker(gfg)" and that ker[( fgf)"-'g' + 11f ' =
a'. The first assertions are immediate from (fgf)" = f. For the final assertion, we
ask the reader to verify the simple result that ker(f h + 1)f ' = ker f ' for any
endomorphism h (regardless of whether f h + 1 is an endomorphism).
Our next lemma is not particularly related to the theory of algebras with
zero; it is rather a general set-theoretical result. We have not had much occasion
to mention it in this book, but, in fact, ker4 makes sense (as an equivalence
relation on the domain of 4) for any function 4 whatever. Among equivalence
relations, O as usual denotes the smallest one, the kernel of the identity function.
LEMMA 6. Let A be any set, and suppose that 4, $: A + A with ker 4 í l

ker $ = O. If 4 o S/ = $ o 4, then ker 4" ílker $" = O for every positive integer n.
Proof. The proof is by induction on n, with the case n = 1 of course being one
of our assumptions on 4 and $. Assuming inductively that ker 4"-' í l ker $"-' =
O, let us suppose that a, b E A, with bn(a)= #"(b) and $"(a) = $"(b). To see that
"-'
1
a = b, let us first consider u = 4"-' ($ (a)) and v = 4"-' ($"-' (b)).Since 4 o $ =

l
l $ o 4, our assumptions on a and b easily yield #(u) = #(v) and $(u) = $(u),
and so u = u by our hypothesis that ker 4 í l ker $ = O. In other words,
($"-'(a), $"-'(b))~ker 4"-', and clearly our assumption on a and b yields
($"-'(a), $"-'(b)) E ker $"-' By induction, we therefore have $"-'(a) = $"-'(b).
A similar argument yields $"-'(a) = $"-l(b). Finally, one more use of induction
yields a = b.
LEMMA7. For al1 n 2 1, ker(fgf)"ílker(fg'f)" = kerf.
Proof. Clearly, ker f c ker(fgf)" 17 ker(fg'f)". For the reverse inclusion, we will
take 4 = fgf and $ = fg'f, considered as maps of f(A) into itself. Since $ + 4 = 1
on f(A), we know that ker 4 í l ker S/ = 0; moreover, 4 commutes with S/, by (13).
1
Therefore, Lemma 6 tells us that ker 4" ker S/" = O. Thus if (x, y) E ker(fgf)" í l
1 ker(fg'f)", then (f (x),f (y))E ker 4" ílker S/" = O, which means that f (x) = f (y),
1 or (x, y) E ker f.
The next lemma is the counterpart, for this theory, to the difficult lemma
that appeared in 92.4, just before the Direct Join Decomposition Theorem. It
will play a role in the proof of the Jónsson-Tarski Theorem much like the role
of that lemma in the proof of the Direct Join Decomposition Theorem.
LEMMA 8. If A is un algebra with zero, and O = a x a' = P x P' for a, a', P,

p' E Con A, with a indecomposable and Ala finite, then there exist y, y E Con A
such that O = y x a' and either /3 = y x y # or = y x y#.
Proof. Of course, as in the rest of this section, we will take 1 = f @ f ' = g @ g',
with a = kerf, and so on. The range of f is in one-to-one correspondence with
Ala, and so by hypothesis is finite. Therefore, as maps from the range of f
into itself, the sequence of maps (( fgf)": n = 1,2, m) must contain a repetition.
But in fact, fgf maps A into the range of f, and so the sequence of maps
(( fgf)"": n = 1,2, ...) contains a repetition, even when we consider these as
maps defíned on al1 of A. Therefore we have (fgf )" = (fgf)m+kfor some m and
k with k > 1. A similar argument yields (fg'f)' = (fg'f)'" for some r and s with
s 2 1. Taking n larger than both m and r, and divisible by both k and S, we
obviously have (fgf )2n= (fgf )" and (fg'f)2" = (fs'f)" Since a < 1,we know from
Lemma 7 that either ker(fgf )" < 1 or ker(fg'f)" < 1. These two alternatives give
rise to the two alternative conclusions of the lemma. Since the arguments are
similar, we shall continue our proof only for the case that ker(fgf)" < 1.
By Lemma 3, we have 1 = (fgf )" @ $ @ f ' for some $ with (fgf)" + $ = f,
and therefore a = ker f = ker(fgf)" x ker $, by Lemma 1. Since a is indecom-
posable and ker(fgf )" < 1, we must have a = ker(fgf )". We next claim that
(fgf >" = f. To see this, consider any x E A, and observe that, by idempotence,
(x, (fgf )"(x))E ker(fgf )" = ker f. By definition of ker f, we immediately have
f (x) = f (fgf )"(x), and this last term is equal to (fgf )"(x), by the idempotence of
f. Thus we have established our claim that (fgf )" = f. Now the conclusion of the
lemma is immediate from Lemma 5. ¤
THE JÓNSSON-TARSKI UNIQUE FACTORIZATION THEOREM (5.8).

Every finite algebra with zero is uniquely factorable.
Proof. We will prove, by induction on the smaller of m and n, that if A is a finite

algebra with zero, and
with each Ai and each Bj directly indecomposable, then m = n and, after re-
numbering, A, r Bi for each i. We may assume, without loss of generality, that
in fact n 5 m. The theorem obviously holds (by the definition of "directly in-
decomposable") for n = 1. Therefore we move to the inductive step and assume
that unique factorization holds for al1 smaller values of n. By 55.2, there exist
directly indecomposable congruences a,, . , a, and P,, . Pn on A such that
S ,
Ala, z Ai for each i and A/Pj r Bj for each j, and such that
We apply Lemma 8 to-the algebra A and the congruences

a = a,, a' = a, x m - . X a,,
P=Pl x ... x Pn-1, P'=Pn,
thereby obtaining congruences y, y # on A such that y x a' = 0, and such that
either /?= y x or = y x We will consider these two alternatives sepa-
rately. Before attending to those two cases, let us first note that from y x a' =
-
a, x a' we have y a,, and hence Aly r A/a,, by Lemma 9 of $5.2. Therefore,
we also know that y is directly indecomposable.
CASE 1: fl = y x y #. Let y # = y, x - x y,-, be a factorization of y # into a
product of directly indecomposable congruences. We have
with al1 factors directly indecomposable. Correspondingly, we have
with al1 factors directly indecomposable. Therefore, by induction, s = n, and we

5.4 Algebras with a Zero Elernent 291
1
may permute the indices of the algebras Bj so that B, r A/y and Bi r A/yi for
2Si1n-1.
Now we also know that y x a' = O. In other words,
and so
with al1 factors directly indecomposable, which implies (by Lemma 9 of $5.2) that
,
1 A, x -- x A, 2 A/y, x - x A/y,-, x B,,
I
with al1 factors directly indecomposable. By induction, m = n and we may

permute the indices of the algebras A,, - A, so that A, r B, and Ai r A/yi for
a,
2 5 i < n - 1. To see the full conclusions of the theorem, we note that A, S

A/y r B,, that for 2 5 i 5 n - 1 we have A, r A/yi r B,, and finally, as we
remarked above, that A, r B,. This completes the proof of the theorem in
Case 1.
CASE 2: jY = y x y # (i.e., pn = y x y#). The equation O = y x a' tells us that
y cannot be 1; therefore the direct indecomposability of 8, tells us that y # = 1
and p, = y. Therefore A, r Aly = A/p, r B,. Moreover, we have
and so p - a'. In other words,

I
l
Pl x --- x Pn-, N a, x m .- x a,,
1
with al1 factors directly indecomposable, which implies (by Lemma 9 of $5.2) that
1
Bl x x Bn-l z A2 x - x A,,
I
with al1 factors directly indecomposable. By induction, m = n, and we may
permute the indices of the algebras Ai so that B, r A, and Bi r Ai for 2 5 i 5
n - 1. Finally, if we interchange the indices of A, and A,, then we will have
Ai Bi for al1 i. •
The next theorem looks ahead to the theory of cancellation, which will be
developed in $5.7, although in a different framework. In 55.7 we will be able to
prove this result for A, B, and C arbitrary finite algebras, so long as C has a
l one-element subuniverse. Here we allow A and B to be infinite, but they (and C)
I
must be algebras with O. Exercise 8 of $5.1 shows that we cannot omit the
1
finiteness condition on C.
I
THEOREM 5.9. (Jónsson and Tarski.) Every finite algebra with zero is cancel-
lable in the class of al1 algebras with zero (of the sume type). In other words, if
AxCrBxCrD
for D un algebra with zero and C finite, then A r B.
Proof. We may assume that C is directly indecomposable, for in the general

case indecomposable factors of C may be cancelled one after another. Now D
has congruences a, a', p, p' with O = a x a' = p x fi' and
Obviously a is directly indecomposable and D/a is finite, and so by Lemma 8

there exist congruences y and y # on D such that O = y x a' and either P =
y x y # or P' = y x y # .
CASE 1: P = y x y#. Notice that the theorem is trivial for a' = 0, and so we
may assume that a' > O; hence y < 1. Since /?is directly indecomposable, we have
-
p = y, and so O = /? x a' = P x pl. Therefore, a' P', and so by Lemma 9 of 55.2
we have A r D/af r D/Pt r B.
CASE 2: P'= y x y#. Therefore, a' x y = 0 = P x B'=P x y x y # , and so
-
a' P x y#, which entails
On the other hand, y x a'

p' = yx y # , yields
=O =a x a' and so y - a, which, together with
Now our isomorphisms for A and B show that A r B, thereby completing the
proof of this case, and hence completing the proof of the theorem. ¤
Exercises 5.10
The first six exercises use the notation of the beginning of this section without
further notice. In particular, they al1 concern A = (A, + ,O, ),an algebra with
zero.
l. Prove that fi defined at the beginning of this section is an endomorphism.

2. Verify equations (2) from the beginning of this section.
3. Verify equation (3) and verify that the bijection mentioned there is really a
bijection.
Exercises 4, 5, and 6 al1 concern the connection between direct product

decompositions A r C, x x C, and inner direct sum decompositions A =
B, @ @ B,. As intermediate links for this connection, we have chosen to use
both product decompositions of congruences (e.g., O = a, x x a,) and direct
sum decompositions of functions (e.g., 1 = f, @ @ f,). But al1 four of the
subjects are closely interrelated, so the reader may choose instead to find a
different path connecting direct products and inner direct sums. The next three
exercises are therefore only a suggestion of a possible way to proceed.
4. Prove that if 1 = f1@ - @ f,, and Bi = &(A) (1 < i 5 n), then each a E A
can be written uniquely as a sum -(bl + b,) +
( e + b,,, with each bi E Bi.
a )
This last sum is equal to any of its rearrangements by commuting or asso-

ciating its summands. Finally, if F is an m-ary operation of A and b i j Bi~
(1 < i < n , l < j < m ) , t h e n
5. Conversely, if the subuniverses Bi have the property that each a E A can be

written uniquely as a sum -(bl + b,) +
( m - m..) +
bn, and if (14) holds for al1
F and al1 b(, then there exist endomorphisms fi such that 1 = f , @ - @ f,
and B, = f,(A) for each i. In such a case, we may write
where Bi is the subalgebra of A with universe Bi.

6. If A = B l @ - . . @ B n , then A E B , x x B,. Conversely, if A r
C1 x x C,, then A has subalgebras Bi such that Bi E Ci (1 < i < n) and
A = B1 @ - . . @ B , .
7. Prove Lemma 1.
8. Prove that Lemma 6 fails if we remove the assumption that 4 o S/ = $ o 4.
9. Prove that the congruence y = ker fg appearing in Lemma 5 is in fact equal
t o p v (a A P').
We now present some exercises that will allow the reader to step through
the main proofs in this section with some matrix calculations that go smoothly
but are nonetheless not completely trivial. In particular, these examples show
that there is no way to limit the exponent n appearing in Lemmas 3 and 4 and
in the proof of Lemma 8. These exercises ask the reader for some of the high
points, but in fact each detail of the proofs in this section can be examined by
seeing what it means for these examples.
10. An illustration of Lemma 3. Let R be a commutative ring with unit, and

define A to be a three-dimensional free module over R. Thinking of elements
of A as column vectors, 3 x 3 matrices over R act as endomorphisms of A.
Thus we have the following endomorphisms of A:
Prove that 1 = f @ f ' = g O 9'. Evaluate fgf, and notice that fgf = 2h,
where h is idempotent. (Thus, in fact, the behavior of the powers (fgf)"
depends on the behavior of the sequence (2") in the ring R. In particular, if
al1 powers 2" are distinct, then no (fgf)" will be idempotent, which shows
that Lemma 3 is not generally applicable (e.g., as in proving Lemma 8),
without some finiteness assumption.) Examine the special case R = Z,
distinguishing between three cases: m = 2k,m odd, and the mixed case-m =
2kq with k 2 1 and q an odd number 2 3. In these various cases, find the 4
and $ of Lemma 3. For example, in the ring h,, we have
Verify directly that 1 = 4 @ S/ O f '.

11. Verify that equations (12)and (13)hold for f,f ',g, and g' taken from Exercise
1o.
12. An illustration of the proof of Lemma 8. Let A be the algebra of Exercise 10,
with one extra unary operation: multiplication by the matrix
(Equivalently, we consider our algebra to be a module over the ring Q of al1

matrices of the form
with a, b E R.) Certainly, every matrix which commutes with E acts as an

endomorphism of A. First prove that f, f', g, and g' (defined below) are
endomorphisms of A:
and then prove that 1 = f @ f ' = g @ g'. Next prove under certain assump-
tions on R (say, R is a field) that a = kerf, a' = ker f ', P = ker g, and P' =
ker g' are al1 directly indecomposable. Calculate fgf and work out a simple
formula for (fgf )" (as matrices). Prove that the smallest n for which (fgf )" is
l
idempotent is precisely the characteristic of the ring R. (If R has characteristic
zero, then no (fgf)" will be idempotent.) Verify for this smallest n that 1I
(fgf )" = f, and also check the assertions of Exercise 9 for this example.
f
j
i
5.5 The Center of an Algebra with Zero 295
13. Obtain al1 the details of Lemmas 4 and 5 for the example given in Exercise
12. For instance, evaluate (fg)" and [( fgf)"-'g' + 11f ' as matrices. Get a
direct calculation of S/ by applying Lemma 3.
14. In the example of Exercise 12, describe the Q-submodules B = f(A), B' =
fl(A), C = g(A) and C' = gl(A). In the framework of Exercises 4-6, prove
that A = B @ B' = C @ C' = C @ B'.
5.5 THE CENTER OF AN ALGEBRA WITH ZERO
In the preceding section of this chapter, our only assumption about the binary
operation + was that it obeys the laws x + O Ñ O + x Ñ x. Thus, perhaps the
most surprising single feature of the proofs in that section was the discovery of
a family of elements y such that x + y = y + x and x + (y + z) = (x + y) + z for
al1 x and z. In particular, (8) and (9) te11 us that any y of the form fgf '(w) has this
property for any w and any endomorphisms f,f ',g, and g' such that 1 = f @ f ' =
g @ gr. (Of course, it is possible that al1 such y are 0, as happens, for instance, in
a centerless group, but in that case we obtain the law fgf '(w) Ñ O, which has very
strong consequences, as we shall see in Corollary 2 to Theorem 5.17 in $5.6
below.) In group theory, the set of al1 such y is easily seen to form a normal
subgroup, commonly known as the center. With a little more care, we can define
the center C of any algebra A with zero. Every J E C will have the properties
stated above (and more); fgf' (as above) will map A into C; and C will be a
subuniverse of A. In fact, for a finite algebra A, C is the O-block of O for 0 the
center congruence Z(A), which we defined in $4.13.(We will show in a later volume
that this also holds if A lies in a congruence modular variety.)
Our objective in this section will be to define this center and then strengthen
the Jónsson-Tarski Unique Factorization Theorem (5.8) to hold for any algebra
whose center is finite (Theorem 5.14 below). This stronger result was also achieved
by Jónsson and Tarski [1947]; they used the center right from the start, rather
than giving our development of the preceding section. We will also prepare for
a proof in $5.6 that if D is an algebra with center = (O), then its direct factoriza-
tions are unique in the sense that if O = a, x - x a, = p, x - x P,, with al1
factors directly indecomposable, then m = n and, renumbering if necessary,
al = P1, a2 = P2, and so on. For this result and a parallel result about algebras
with distributive congruence lattice, see Corollaries 1 and 2 to Theorem 5.17 of
$5.6.
We remarked above that a little more care is required in defining the center
in the general case than was required to define the center of a group. The main
reason for this is that there seems to be no simple criterion for a single element
c to be central. Instead, we must work with centrality of an entire subuniverse.
i DEFINITION 5.11. B is a central subuniverse of A iff B is a subuniverse and

for every A-operation F. An element b~ A is central iff b lies in some central

subuniverse.
Let d be a central element and a an arbitrary element of A. By (16),we have

+ d ) + (a + 0 ) = (O + a) + (d + O),
(O
from which we see that d + a = a + d-i.e., a central element commutes with every
element of A. The reader may similarly show that a central element d associates
with every two elements of A, in the sense that (a + d ) + c = a + (d + c) for
arbitrary a, c E A. (See Exercise 6 below.)
The next two lemmas follow readily from the definitions, and the reader
may work out the details.
L E M M A l . Any two central subuniverses together generate a central sub-

universe, so the set of al1 central elements forms a central subuniverse, which is in
fact the largest central subuniverse. U
DEFINITION 5.12. The center of A, denoted C(A),or simply C, is this largest

central subuniverse.
+
L E M M A 2. Every central element c is cancellable in the sense that a c = b c +
implies a = b, for any a, b E A. The center is an afine subuniverse in the sense that
( C , +,O) is un Abelian group and each A-operation F acts linearly on C. I.e.,
F(c,,c,,.-.) = al(cl) + a,(c,) + a - .
for some endomorphisms a,, a,, - of ( C , + ,O).
The next lemma shows the relevance of the center to direct factorization
theory. In a sense, it contains a large part of the information of Lemma 2 in 55.4.
Here f , f ', g, and g' al1 refer of course to endomorphisms of an algebra A with
zero.
L E M M A 3. l f 1 = f @ f' = g @ g', then f maps the center of A into itself and

fgf' maps al1 of A into the center of A.
Proof. Clearly f ( C )is a subuniverse, so for the first assertion it will be enough
to check that f (b)satisfies (15)for b E C, and also that (16)holds for arbitrary a,,
a,, . . E A and with b, ,b, . replaced by f (b,),f (b,), for arbitrary b,, b,, . . E C.
For the first of these, we know by the centrality of b that b + c = O for some
+
c, so f (b) f (c) = O. For the second one, we must prove the equality of a =
+
F(a1 + f(b,),a, + f(b,), - - . ) with p = F(a,,a,, ...) F(f(b,),f(b,), ...). The
centrality of the b,'s yields f(a) = f(P), and clearly f ' ( a )= f ' ( P ) (without any
centrality assumption). Therefore a = P.
Next let us show that the range of fgf' is contained in the center of A. We
will first show that the subuniverse B = fgf '(A) satisfies condition (16), which is
to say that
for al1 a,, b,, a,, b2, . E A. Denoting the two sides of this equation by y and 6,
we note that obviously f '(y) = f1(6), and so it only remains to prove that
f(y) = f (6). Working toward this equation, we first get
since the two sides are equal under application of f and of f '. We next observe
that
since both sides are equal under application of g and of g'. Finally, application
of f to both sides of this equation yields the desired equation f(y) = f (a), which
leads, as we remarked above, to the conclusion that B = fgf'(A) satisfies condi-
tion (16). Likewise, B' = fg'ff(A) satisfies condition (16), and so one easily checks
that D = B + B' is a subuniverse satisfying (16). Finally, since
each member of B has an additive inverse in B', and vice versa. Then one easily
checks that D obeys (15) and thus is a central subuniverse.
Before going on to obtain unique factorization for algebras with finite center,
we will pause to study the connection between this center C = C(A) (of Jónsson
and Tarski) and the center congruence Z(A) of $4.13 above, which is defined for
al1 algebras A (not just for algebras with zero). If A is an algebra with zero, then
the congruence block O/Z(A) is obviously a subuniverse of A. If A is a group,
then C(A) = O/Z(A), since both are equal to the center of A in the usual sense of
group theory. When we return in Volume 3 to the study of the commutator in
congruence modular varieties, we shall see that the equation C(A) = O/Z(A)
holds for A an algebra with zero in any congruence modular variety. For now,
we prove a related result that does not require modularity. As we will see in
Exercise 1 below, we cannot remove the finiteness assumption, since, e.g., if
A = ( o , +) (usual addition of non-negative integers), then C(A) = {O} and
O/Z(A) = o .
THEOREM 5.13. If A is un algebra with zero, then C(A) O/Z(A), with equality
holding if O/Z(A) is finite.
Proof. We first need to see, quite simply, that if b is a central element of A (in
the sense of this section), then (O, b) E Z(A). Since b is central, we have
t ( b , c 1 , . - - , ~ ,= +
) t(b,O;-.,O) t(O,cl;--,c,)
t(b, dl;..,dn) = t(b, O,... ,O) + t(0, d l , - a *d,)
,
for every n > 1, for every term operation t E Clo,+,A, and for al1 c,, - c,, d,, s.,
- d, E A. Since t(b, O,. O) is central, and hence cancellable, it is clear that

m , m ,
and hence, by Definition 4.146, that (O, b) E Z(A).

We will next prove that, regardless whether O/Z(A) is finite or infinite, it
forms a subuniverse B that satisfies condition (16) in the definition of centrality.
To see this, we first take any A-operation F, of arity n, and define t to be the
3n-ary term operation
Now we easily check that
for arbitrary a,, - - a, E A. Now let b,, b,, . be arbitrary elements of O/Z(A).
a ,
By definition of Z(A), we may change the final O on both sides of our last equation
to b,, yielding
Continuing to make changes of this sort, we ultimately obtain
Referring back to our definition of the term t, we may rewrite this last equation
as
F(O, ..,O) + F(a1 + bl, - ,a, + b,) = F(a,,. - - ,a,) + F(O + b,, ,O + b,),
which in turn is easily reduced to
Since a,, - ,a, were arbitrary elements of A and b, , ,b, were arbitrary elements
of B = O/Z(A), we have now established condition (16) for the subuniverse B.
As we remarked above, it is not hard to prove from condition (16) that +
is a commutative and associative operation on B. We will next observe that for
b E B, and c, d arbitrary elements of A, the cancellation law b + c = b + d -,c = d
holds. This follows immediately from
which is a special case of the defining condition (17) for (0, b) EZ(A),which we
stated in the first paragraph of this proof (and which carne originally from
Definition 4.146). Therefore, in particular, + obeys the cancellation law when
restricted to B.
We now invoke the hypothesis that B = O/Z(A) is finite. It is well known
+
that if is an associative cancellative operation on a finite set B, then (B, + )
is a (multiplication) group, so this must be true for the case at hand. Now
condition (15) of the definition of centrality is immediate, since it obviously holds
for groups. This completes the proof that B = O/Z(A) is a central subuniverse of
A, and hence that O/Z(A) C(A). ¤
Now that we have established this connection between the Tarski-Jónsson

center C(A) and our other center Z(A), we will continue, in this chapter, to refer
to C simply as the center of A. We will not need to mention the other center,
although the reader may compare the examples in Exercises 2-5 below with
various examples of Z(A) presented in 84.13.
We conclude this section with its main result, which is a strengthening of
the main result of the previous section.
THEOREM 5.14. (Jónsson and Tarski.) Every algebra with zero whose center
is finite is uniquelyfactorable, provided that each of its directfactors is decompos-
able into a finite product of indecomposable algebras.
Proof. If we look over the proof of the Jónsson-Tarski Unique Factorization

Theorem (5.8) and its lemmas, in the previous section, we see that finiteness was
invoked in one place only. In the proof of Lemma 8 (just before Theorem 5.8),
we used the finiteness of A to establish that (fgf )" = (fgf )m+kfor some m and k
with k 2 1 (and to establish a similar result for fg'f). Here we will prove this
same fact under the weaker assumption that the center is finite, and then we may
invoke the entire earlier proof of Theorem 5.8.
Of course, we adopt al1 the notation that was in force in the earlier proof,
especially 1 = f @ f ' = g Q gr. For m 2 3, we define
which, by Lemma 3, may be considered as a map of the center into itself. Since
the center is finite, we must have h, = h,+, (as maps defined on the center) for
some m and some k 2 1. Since the range of gfg' is contained in the center (again
by Lemma 3), we have
hm!?f~'f
=hm+k~f~'f
on the entire algebra. Now, we calculate
Iterating this calculation k times (with k as determined above), we obtain
Since hmgfg'f = h,+,gfg'f, and since this is moreover a cancellable element, we

finally obtain (fgf)" = (fgf)"+" A similar calculation yields (fg'f)' = (fg'f)'+"
for some r and S with s 2 l . From this point on, we simply follow the remainder
of the proof of Lemma 8 of the previous section and then follow the proof of the
Jónsson-Tarski Unique Factorization Theorem (5.8). m
Exercises 5.15
+
l. Let A be (co, )- that is, the cancellative commutative monoid consisting
of al1 non-negative integers under the operation of addition. Prove that
C(A) = (O}, but O/Z(A) = co (Le., A is Abelian, in the sense of 54.13).
2. The center of a group C is its center in the classical sense, namely (xE G:
xy = yx for al1 y E G}.
3. The center of a ring R is the set of x E R such that xy = yx = O for al1 y E R.
*4. If A is an algebra with zero that lies in some congruence distributive variety,
then C = (O}. Thus, of course, a lattice with O has a trivial center (although
we will see a stronger fact in the next exercise). (For some hints, refer to
Exercise 4.156(7) in 54.13.)
5. If A is an algebra with zero such that the implication
holds for al1 a and b in A, then C = {O}. This applies, for instance, to A =
(S, v ,O), a join semilattice with zero. (In the next section we will see that
the hypothesis C = (O} has very strong consequences, which go beyond
unique factorization theorems as we have known them up until now.)
6. If d is central, then (a + d) + c = a + (d + c) = d + (a + c) for al1 elements
a and c.
7. Prove Lemma 1 and Lemma 2.
8. If A is a subdirect product of algebras Bi(with zero), then the center of A
is contained in the product of the centers of the algebras B,.In particular,
if each Bihas center (O}, then the same is true of A.
9. Prove that Theorem 5.9 extends to the case where C is a finite product of
indecomposable algebras, each with finite center.
*lo. Let X be the class of algebras with zero of a given similarity type. Prove
that every indecomposable algebra in Y with finite center is prime in X .
(For the definition of X-primality, see page 263,§5.1.)
11. The previous exercise is false without the assumption of finite center.
5.6 SOME REFINEMENT THEOREMS
We will say that an algebra A has the refinement property (R) (for direct factor-
izations) iff
implies the existence of algebras Dij(i E 1,j E J) such that, for al1 i and j,
Bi z n
ja J
Dij and Cj g n Dij.
ieI
It is not hard to see that if al1 Bi and Cj are directly indecomposable, then almost
al1 D, will be singleton algebras. In fact 4 = {(i,j): 1 D,I > 1) will be a one-to-one
correspondence between J and I such that Bi Dij r Cj for each (i,j) E 4. Thus
the refinement property implies condition (ii)of the unique factorization property
from $5.1 (any two factorizations into indecomposables have isomorphic factors
after rearrangement). But refinement does not imply condition (i) of unique
factorization (the existence of at least one factorization into indecomposables),
and in fact we can deduce no relationship among (i), (ii), and refinement (R) that
is not already deducible from R -+(ii) (together with the obvious tautology that
(i) or (ii)must hold). (See Exercise 9 below.) On the other hand, for finite algebras,
(i) always holds and (ii) is equivalent to refinement (see Exercise 4). For a weak
generalization of this last fact to infinite algebras, see Exercise 5.
Even though the refinement property seems a natural modification of unique
factorization for infinite algebras, the methods we have developed so far will not
extend to prove the refinement property for infinite algebras, as we already saw
in Exercises 5.1(6 and 7) in 85.1. At least in the simpler cases where refinement
is known to hold for some infinite algebras, one must invoke some rather strong
hypotheses that also entail a stronger form of refinement, namely the strict
refinement property for A. This property asserts that if
for congruences pi, yj on A, then there exist congruences 6, (i E I,j E J) such that,
for al1 i and j,
Pi = n 6,
je J
and yj = n 8,.
~ E I
Many of our refinement results in this section will be based on the first assertion
of the following easy theorem, whose proof we leave to the reader.
THEOREM 5.16. If A has the strict refinement property, then A has the refine-
ment property. If, moreover, O = ni,, n
ai = j,J P; in Con A, with each ai and each
m
j?, indecomposable, then (a,: i E 1) = {pj:j E J).
COROLLARY. If ni,,Bi has the strict refinement property, al1 Bi and al1 Cj
are directly indecomposable and f : n B i -+n C j is un isomorphism, then there
exists a bgection 1:J -+ I and a family of isomorphisms fj: Ba(j)+ Cj, such that
f (b) = <f,(b,(j)):jEJ )
for al1 b = <bi:i~ I ) E ~ B , .
While the strict refinement property implies the refinement property, there
are many easy examples showing that the converse is false, such as the square of
the two-element group, or even a four-element set. While it is true that Theorem
5.16 does have a strong hypothesis, let us also note that it will provide us (via
later results of this section) with our first unique factorization results that do not
require a one-element subuniverse.
Our main tool in refining decompositions is the notion of binary decomposi-
tion operation, which was defined in Definition 4.32 in 54.4. A decomposition
operation of A is a homomorphism f : A2 -+ A that satisfies the equations
If f is a decomposition operation, then for each u, u E A we define functions f,

and f::A + A as follows: fv(x) = f (x, u), and f,'(y) = f (u,y). (In 54.4, we wrote f u
where we now have f:.) If a x a' = O for congruences a, a' on A, we define
f.,..:A2 -+ A as follows:f.,,.(x, y) is the unique w such that (x, w)E a and (w, y) E a'.
The following lemma recapitulates various results from 54.4.
LEMMA 1. ker fv and kerf,' are congruences on A that do not depend on the
choice of u and v. The correspondences
form a bijection, and its inuerse, between the set of pairs (a, a') E (Con A)2 with
a x a' = O and the set of decomposition operations on A. ¤
In case A is an algebra with zero and u = v = O, then f, and f:are the f and
f ' of the previous two sections. More generally, if A happens to have a one-
element subuniverse (e}, then fe and fé will be endomorphisms of A, but in the
general case f, and f: will not be endomorphisms.
Recall from Definition 4.28 in 54.4 that a factor congruence of A is a congru-
ente a on A such that a x a' = O for some a' E Con A. As we will see in our next
theorem, the strict refinement property is equivalent to the assertion that the
factor congruences form a Boolean sublattice of Con A. First, a lemma showing
that certain subsets of the factor congruences always form a Boolean lattice.
LEMMA2. If O = n i & , then themap

KH nai
I-K
is a (O, 1)-homomorphismfrom the (Boolean) lattice of subsets of I into Con A.
Proof. It is enough to prove that if O = a x Pxy x 6, then

( a x P ) ~ ( a x y ) = a x p x yand
(a x P) v (a x y) = a.
The first equation is immediate, and for the second, we certainly have a 2
(a x p) v (a x y). For the opposite inclusion, we will take (x, y) E a, and show that
(x, y) E (a x p) o (a x y). Since a x p x y exists, we know (from Lemma 1 of 55.2)
5.6 Some Refinement Theorems 303
that there exists u E A with (u, x) E a, (U,X)E p, and (u, y) E y. Since (u, x), (x, y) E a,
we have (u, y) E a by transitivity; thus (u, y) E a A y. Combined with (u, x) E a A P,
this yields (x, y) E (a A P) o (a A y). ¤
Our next theorem gives various characterizations of the strict refinement

property. This is the theorem we promised in the last paragraph of 54.4. For our
applications, the most useful conditions will be those involving the functions f,.
Notice that if f is a decomposition operation, and we define h(x, y) to be f (y, x),
then h is also a decomposition operation, and fi = h, for every u. Therefore,
conditions (v) and (vi) below also apply to f: and g:.
THEOREM 5.17. For any algebra A, the following conditions are equivalent:
i. A has the strict refinement property.
ii. A has the strict refinement property for finite index sets I and J.
iii. The set of factor congruentes of A forms a Boolean lattice (i.e., it is a sub-
lattice of Con A, and the distributive laws hold on this sublattice).
iv. If O = a x a' = p x pl for a, a', P, p' E Con A, then (a v p) A a' r P.
v. f,g, = g, f, for al1 decomposition operations f, g, and al1 v E A.
vi. There exists V EA such that f,g, = g, f, for al1 decomposition operations f
and g.
REMARK: In (iii), we did not have to mention complementation explicitly,
since, of course, every factor congruence has a complement that is itself a factor
congruence.
Proof. Lemma 2 immediately yields (ii) --+ (iii), and the implications (i) --+ (ii),
(iii) + (iv), and (v) -+(vi) hold a fortiori. It remains to show (iv) -+(v) and (vi) -+ (i);
n
we begin with the latter. Assuming (vi), and given O = nBi = y,, we define S,
to be the congruence Pi v y j for each i and j. To prove (i), it remains to show that
the 6,'s are as required for strict refinement, i.e., that n
di, = Di for each i, and
n 6, = y, for each j.
By Lemma 1, for each i there is a map fi: A + A such that Di = kerfi and fi
has the form f, for some decomposition function f. Likewise, we have y, = ker gj
for some g, of the same form. By (vi), every fi commutes with every We wil¡
now prove that 6, = kerfigj = kergjfi for each i and j. It is immediate that
ker fi 5 ker gjfi = ker figj and that ker gj 2 ker figj, and thus that 6, = ker A v
ker gj 5 kerfig,. For the reverse inclusion, it is clear that for any (x, y) E kerfigj
we have
ker gj ker fi ker fi
x ---- gj(x) f i ~ j ( x=
) fi~j(~) gj(y) JE!x- Y
and so kerfigj 5 kerfi v kerg, = dij.
Certainly A(x) = fi(y) iff gjfi(x) = gjfi(y) for al1 j, and so ,. 6, = Pi for
each i. To verify that this intersection is actually a product, we need only show
(for fixed i) that for any a E A' there exists b E A with fig,(b) = figj(aj) for al1j E J.
This follows immediately from the existence of b E A with gj(b) = gj(aj) for al1j,
which in turn is immediate from the existence of n y , . Therefore, the product

n, 8, exists and is equal to Pi. A similar argument shows that ni
6, exists and
-
is equal to yj. This completes the proof of (i) from (vi).
For (iv) (v), we will assume (iv) and then prove f,g,(x) = g, f,(x) for any
two decomposition operations f, g, and any u, x E A. In order to apply (iv), we
will take a = kerf,, a' = kerf:, p = kerg,, and p' = kergú; thus O = a x a' =
p x P'. We will prove the equality fug,(x) = guf,(x) in two steps: first,
(fug,(x), g, f,(x)) E p, and then the same for p'. To begin, we have
and so (fu(x),f,g,(x)) E (a v p) A a'. Therefore, by (iv) we have (f,(x), fugu(x))E P.

Moreover, (f,(x), g,f,(x)) E p, and so by transitivity (f,g,(x), g,f,(x)) E P.
For the corresponding p'-relation, we have
and so by (iv)(with p' in place of P) we have (u,fugu(x))E P'. Moreover, (u, g, fu(x))E
I(', and so by transitivity (fugu(x),gufu(x))E P'. Thus fig,(x) = g, f,(x), and the
proof of (v) is complete. M
This theorem has two especially important corollaries closely related to the
main results of 555.3-5.5 (the unique factorization theorems of Ore, Birkhoff,
and Jónsson, and of Jónsson and Tarski). Notice that they do not require
finiteness and that the first one does not require a one-element subuniverse. The
-
first corollary is immediate from (iv) + (i) of the theorem but can also be seen
from (iii) (i) together with Exercise 16 of $4.4.
COROLLARY l. If Cona is distributive, then A has the strict refinement

property. m
COROLLARY 2. If A is un algebra with zero and the center of A is {O}, then

A has the strict refinement property.
Proof. In the notation of 5$5.4 and 5.5, we need only see that fg = gf whenever I
1 = f @ f ' = g @ gr. By Lemma 3 in 55.5, fgf' and f 'gf both map al1 of A into
the center of A, so we must have fgf' = f 'gf. Hence,
fg = f s ( f + f ' ) = fsf + fsf'
= fsf + f 'sf = ( f + f ')sf = sf.
As we remarked above, it follows from these corollaries that an algebra with

distributive congruences or with zero center has the refinement property. Hence,
if it has any factorization into indecomposables, then it has the unique factoriza-
tion property, and even more: by Theorem 5.16, if
with each Diand yj indecomposable, then the 13,'s are exactly the same as the y~s.
Corollary 1 tells us that al1 lattices have the strict refinement property, and
Corollary 2 tells us that the same is true of any join semilattice S with O. Let us
look at another particularly simple proof for a lattice L with O and 1, which men-
tions neither the center nor congruence distributivity. If L = M x M', and if we
let f denote the decomposition operation associated to this factorization of L,
then fo(a, a') = (a, O) = (a, a') A (1, O). This general form for fo is isomorphism-
invariant, so we may conclude that for each decomposition operation f there
exists b~ L such that f,(y) = y A b. Now the semilattice laws te11 us that
fo (go(x)) = go(fo(x)), so condition (vi) of the theorem is immediately verified for
this example. One interesting feature of this alternate line of reasoning is that it
made almost no use of the fact that L is a lattice, only that (x, x') A (y, y') exists
in any direct product ordering when x I y and y' x'. Thus the argument
extends automatically to ordered sets with O and 1: every ordered set with O and
1 has the strict refinement property.
As we remarked in the introduction ($5.1), unique factorization does not
hold for al1 finite ordered sets. Therefore, we cannot completely remove our
assumption, in this last result, of the existence of O and 1. Nevertheless, as J.
Hashimoto proved in [1951] by a more sophisticated argument, refinement holds
for every connected ordered set. (See below for a definition of connectedness.)
In our treatment of the subject, Hashimoto's Theorem will follow readily
from an important lemma proved by R. McKenzie in [1971]. As we shall see,
McKenzie's Lemma permits the deduction of many interesting conclusions,
including some that are significantly stronger than Hashimoto's Theorem. For
instance, as an easy corollary of McKenzie's Lemma we have C. C. Chang's
theorem that if A = (A, R) is any connected reflexive binary relational structure
with at least one antisymmetric element u (i.e., u satisfying Vx[(R(x, u) & R(u, x)) +
x = u]), then A has the refinement property (Corollary to Theorem 5.19 below).
McKenzie's work followed a small but important group of articles by C. C.
Chang, B. Jónsson, and A. Tarski during 1961-1967. These works pioneered the
notions of refinement and strict refinement and developed the techniques of
calculating with the maps f,and fd. (Our Theorem 5.17 was extracted from those
works.) They also formed a foundation for the lemma of McKenzie, from which
some of the results of Chang, Jónsson, and Tarski can easily be derived.
McKenzie's Lemma will be stated and proved for a relational structure
(Exercise 7 of 54.4) A = (A, R) with a single relation R, which is binary. For
short, we will cal1 such a structure a binary (relational) structure, and we will
abbreviate (a, b ) R~ by a R b. We say that (A, R) is reflexive iff it satisfies
Vx[R(x, x)], and connected iff the conditions
BUC=A, BiiC=@, R E B ~ U C ~
entail that B = A or C = A. To every binary relational structure (A, R) we
.,
associate four binary relations I , I 5 , and E,which are defined as follows:
We now begin a sequence of seven lemmas, leading up to McKenzie7s

Lemma, which refer to a binary relational structure A = (A, R). Recall, from
our construction in 51.1 of a relation algebra consisting of al1 binary relations R
on A, that Ru denotes the relation converse to R, which is given by the formula
RU = ((Y,x): (x, Y) E R).
LEMMA l. Let A = (A, R) be rejexive. A is connected iff, for,each a, b E A

there exist wo, z, , - zn-, , wn-, E A such that
a R u w oRz, R u w l R.-.Rz,-, Ruwn-, Rb.
In other words, it is connected iff A2 = R1 U R2 U R3 U e , where Rn =
-
(R" O R) 0.- o (R" o R), with n factors (R" o R).
LEMMA 2. is un equivalence relation on A such that, if a a' and b = b',
-
then a R b-a1Rb'.
Equivalently, A is thin iff A r A/ . -

Therefore, we may define a quotient structure A/ = (Al=, S) where
(a/ = ,b/ r ) E S iff (a, b) E R. We cal1 A thin iff r is the equality relation on A.
The theory of products for relational structures is similar in almost al1 details
to the corresponding theory for algebras, so we will summarize what we need of
it, with very little in the way of proof. For simplicity, we will state our definitions
and lemmas only for the case of major interest, namely for binary structures. The
general case can easily be guessed from this case.
The product of binary structures A = (A, R) and B = (B,S), denoted
A x B, is the binary structure ( A x B, T) where, by definition, (a, b) T(a', b') iff
a R a' and b S b'. The product ni,, Ai of any family (A,: i E I ) of binary structures
is defined analogously. A homomorphism f : A -+ B is a function f : A -+ B such
that if a R b, then f(a) S f(b). If A = (A, R) is a binary structure and 8 is an
equivalence relation on A, then the quotient structure A/0 = (Ale, R') is de-
fined so that R' is the smallest relation rnaking the natural map A -+ A/8 a
homomorphism.
If A = (A, R) is a binary structure, then we may define a decomposition
operation of A to be a homomorphism f : A2 -t A that obeys conditions (18)
and (19) near the beginning of this section. A decomposition operation arises
naturally on each product P x Q via f((po, qo),(p,, q,)) = (p,, q,), and, within
isomorphism, al1 decomposition operations arise in this way. This fact is con-
tained (and made precise) in the following lemma. As with algebras, if f is a
decomposition operation of A, and u, v E A, then f,and f: are the selfmaps of A
defined by f,(x) = f (x, v) and fd(y) = f (u, y).
5.6 Sorne Refinernent Theorerns 307
LEMMA 3. If f is a decomposition operation of A = (A, R), then ker f, is

independent of u E A and ker fi is independent of u E A. Moreover, the natural map
#: A -,Alker fu x Alker fi is an isomorphism of A with Alker fu x Alker fi. Given
any isomorphism #: A + B x C, one muy define a decomposition operation f on A
via f (a, a') = #-'((p,(#(a)), p1 (#(a1)))),where p, and p , are the coordinate projec-
tions of B x C onto B and onto C. Moreouer, for this f and #, we have B z A/ker fu
and C z Alker fi, provided that the binary relations of B and C are both nonempty.
m
Regarding the final proviso of Lemma 3, the reader may easily observe that
if, say, the binary relation of C happens to be empty, then the binary relation of
A will also be empty, regardless of what is the binary relation of B. Therefore,
under these circumstances it will generally not be true that B z Alker f,. In view
of Lemma 3, we will define a factor relation of A to be any equivalence relation
ker f, for f any decomposition function and f, defined as above. The connection
between factor relations and direct factorizations goes exactly as it did in the
purely algebraic case (55.2).In particular, we have the following corollary to the
last lemma. Let us first observe that the relation n pi = O was defined (both
in 554.4 and 5.2) purely in terms of the equivalence relations pi, with no real
regard to the fact that in that context the Di were congruence relations. Therefore,
we may use that concept and notation in the present context with no change.
COROLLARY. If A = (A, R) is a binary relational structure with R # @, and

if A S then there exist factor relations B, of A such that B, S A/Oi for
each i and such that O = nGI
a. m
Now refinement and strict refinernent may 'be defined for binary structures
A exactly as they were for algebras. Strict refinement implies refinement as it did
before, and Theorem 5.17 continues to hold in this new context (after rewording
(iii) and (iv)so as not to mention Con A). The next lemma states and reviews some
elementary facts about a decomposition operation J: in succeeding lemmas, g is
a decomposition operator as well.
LEMMA 4.
i* f (f,(x),fú(y)) = f (x,Y).
ii* fv(fw(x)) = f,(x).
iii. fú( f,(x)) = f,( fi(x)) = fUf(v)= f,(u). In other words, for each u and u, f: com-
mutes with f,, and f; o f, is a constant function.
iv. If x R y and u R u, then f (x, u) R f (y, u).
v. If w R y and x R f,(y), then f (x, w) R y.
,
vi. If (A, R) is reflexive, and if a 5, b and a' 5 b', then f (a, a') 5 f (b, b'), and
similarly for 5, 5 , and E .
-
vii. If (A, R) is reflexive, and if f,(x) = f,(y) and fi(x) = fi(y), then x y.
Proof. The first four are straightforward consequences of the defining condi-
tions (18)and (19).(v)follows immediately from (iv) and the fact that f ( f,(y), y) =
y, which holds by (ii). Assertion (vii) follows readily from (vi), which we now
prove. To prove the conclusion of (vi), namely that f(a, a') 5 , f(b, b'), we will
assume that w R f(a, a') and prove that w k f(b, b'). By reflexivity, we have a R a
and a' R a'; therefore f (w, a) R f (f(a, a'), a) = a, and f (a', w) R f (a', f (a, a')) = a'.
Since a 5, b and a' 5, b', we have f (w, a) R b and f (a', w) R b'. Therefore, w =
f (f (w, a),f (a', w))R f(b, b'), as required.
Proof. By Lemma 4(v), we have f(x, w) R y and f(x, w) R z. Therefore

f (x, w) R g(y, z). Again using x R f,(y), we have
and this obviously reduces to x R f,(g(y, 2)).
LEMMA 6. If (A, R ) is rejlexive and connected, and if x, y, z, v E A, then
Proof. This assertion is trivial if y = z. Otherwise, by connectedness and

Lemma 1, there exist y; = y, wó, y;, w;, , y;-,, wn-, , y; = z E A such that
wf R yf, wf R y;,, (i < n).
Now define wi = f(x, w,')for i < n, y, = f(z, yf)for 1 5 i < n, yo = y, and y, = z.
By Lemma 4(v), it follows from w&R y and x R f,(y), that wo R y. It likewise
follows from Lemma 4(v) and x R f,(z) that w,-, R y, = z. It is easy to check that
w, R y, for (1 5 i < n) and that wi R y,+, for (O 5 i < n - 1). Finally, it is not hard
to see (say by Lemma 4(iii)) that f (w,, X)= x for each i and that f (y,, f,(z)) = f,(yi)
for 1 5 i < n. Therefore, from w, R y, and x R f,(z) we deduce x R f,(yi) for
(1 5 i < n). (Moreover, this last relation was already known to hold for i = O and
i = n.) In summary, what we now have is:
To complete the proof of this lemma, we will prove, by induction on n > O, that
(21) implies the conclusion of (20), namely, that x R f,(g(y, 2)).
The case n = 1 is covered by Lemma 5 (with w, taken for w). For the
inductive step, we now assume that n > 1 in (21), and that the corresponding
statement with n replaced by n - 1 implies x R f,(g(y, y,-,)). Two applications
of this inductive assumption give us that
Setting y = g(y, y,-,), T = g(yl, z), and w, = g(wo,w,-,), it is not difficult to

check that y, Z and w0 satisfy (21) (with n = 1), so by induction we have
x R f,(g(y, 5)).Now one easily checks that g(y, T ) = g(y, z). Hence x R f,(g(y, z)),
thereby completing the proof of the lemma. E
LEMMA 7. Suppose that ( A , R) is connected and reflexive, and let x, y, u, w E A.

If f,(x) = f,(y), then f,(gw(x)) = f,(g,(y)).
Proof. We first recall that the relation
,
-is the intersection of four relations,
namely 5, and 5 and their converse relations 5; and 5:. By the symmetry
of the assumptions, it will be enough to deduce only that f,(g,(x)) 5, f,(g,(y)),
since the corresponding fact for 5, <,U, and 5,V can be obtained with similar
arguments.
From the fact that (A, R) is connected and reflexive, it follows immediately
that (A, RUo R) is also connected. Then it should be obvious that an easy
induction will yield f,(g,(x)) 2, f,(g,(y)) from the following two assertions:
First we prove (23). We rewrite (23) as f,(x) <, f,(gx(y)) and then observe
-
that, according to the definition of S,, we can prove this by assuming that
z R f,(x) and then deducing that z R f,(g,(y)). Thus we are given z R f,(x); since
,
we have assumed f,(x) f,(y) (hence f,(x) 5 f,(y)), we also know that z R f,(y).
Therefore, by Lemma 6 we know that z R f,(g(y, x)), which is to say z R f,(gx(y)).
The proof of (23) is thus complete, so by the above remarks we know that
everything hinges on the proof of (24).
We will assume the hypothesis of (24) and prove its conclusion, namely that
,
fv(gb(x))-< fv(gb(y)),by assuming that z R f,(g,(x)) and then proving that
z R f,(gb(y)). Since (a, b) E Ru o R, there exists c E A with c R a and c R b. We define
1 It is not difficult to calculate that
and the remainder of the proof will consist of a long series of calculations.
From c R a and (25) we easily derive u, R g(x, a) = ga(x). This and similar
calculations yield
From the second of these, together with our assumption that z R f,(gb(x)), we
readily deduce that = f(z, u,) R f( f,(gb(x)), gb(x))= gb(x).In short,
From (29) and the first equation of (28), we deduce that z, = g(Z, u,) is R-related
N
I
to g(gb(x),ga(x)),which obviously reduces to ga(x). This, together with a similar
1 argument invoking the second equation of (28), yields
From (30) and our assumption that z R fv(gb(x)),we deduce that f(zo,z)R
f(ga(x),f(gb(x),u)) = f(ga(x),v) (and similarly with a replaced by b). In other
words, we have
From (31) and our assumption that fv(ga(x))5 , fv(ga(y)),we immediately have
(32) fz(z0)R f v ( ~ a ( ~ ) ) .
If we apply Lemma 6 to (32) and the second part of (31), we obtain fz(zo)R
fv(g(ga(y),gb(x))), which is easily reduced to
(33) f z ( ~ 0Rfv(gb(~))-
)
By an argument analogous to that used for (30) we have
and an argument analogous to that used for (31) yields
We now apply Lemma 6 to the converse binary relational structure ( A , Ru).

From that lemma and relations (33) and (35), we obviously have
(36) fz(g(z0,zd)Rf,(g,(y))-
By (27), the left-hand part of (36) is z, so we have z R fv(gb(y)),as was required to
,
prove that fv(gb(x))Ifv(gb(y)).This completes the proof of (24), and hence also
the proof of Lemma 7. m
-
THEOREM 5.18. (McKenzie's Lemma.) If A = (A, R) is a connected reflexive
binary relational structure, f and g are decomposition operations of A, and x, v E A,
then fvgv(x) gvfv(x).
Proof. We will base the proof on Lemmas 1-7, which al1 hold under the
hypotheses of this theorem. We certainly have fv(fv(x)) fv(x)(in fact, the two
sides are equal). Therefore, Lemma 7 yields fu(gv(fv(x)))r f,(g,(x)), which is
easily revised to read
(37) -
fv(gv(fv(x))) fv(fv(gv(x))).
On the other hand, f;(fv(x)) v r fi(v) by Lemma 4(iii), and therefore, by
Lemma 7,
f:(gv(fv(x))) - f:(gv(v)).
The right-hand side here is clearly equal to f;(f,(g,(x))) (since both are equal to
u), so we have
Now the desired relation, gv(fv(x)) = f,(g,(x)), is immediate from relations (37)
and (38) and Lemma 4(vii). m
Recall that we defined a binary relational structure A to be thin iff reduces

to the equality relation on A. Our first corollary is immediate from McKenzie's
Lemma and Theorem 5.17, which characterized strict refinement via f, o g, =
S, O fu.
COROLLARY l. If A = (A, R) is a connected reflexive binary relational struc-
ture, and if A is thin, then A has the strict refinement property. ¤
Our next corollary of McKenzie's Lemma is our version of the theorem of

Hashimoto, mentioned above. It is stronger than what Hashimoto proved, in
that he did not discuss strict refinement. This version was certainly known to C.
C. Chang, B. Jónsson, and A. Tarski during the period mentioned above.
COROLLARY 2. Every connected ordered set has the strict refinement property.
on A, then ( A , R) is thin. One may successively prove that 5, I 5 , and r .,

Proof. By Corollary 1, it will be enough to verify that if R is an order relation
are al1 the same as the equality relation; we leave the straightforward details to
the reader. m
The lemma of McKenzie will yield the refinement property under much
weaker hypotheses than thinness, but for the general results we must forego
obtaining strict refinement. One way to obtain refinement without the need for
first proving strict refinement is through the intermediate refinement property.
It is so named not because it describes any sort of refinement (it does not), but
because it is implied by strict refinement, and in turn implies the refinement
property. It is most convenient to define this new property for a relational
structure (or algebra) A relative to an element v of A. The pair (A, u) is said to
have the intermediate refinement property iff the two formulas f,(gv(x)) = u and
gv(fv(x))= u are equivalent for al1 decomposition operations f and g of A and
for al1 elements x of A. It is obvious from Theorem 5.17 that if A has the strict
refinement property, then (A, u) has the intermediate refinement property for
every element v of A. The other implication mentioned just above is contained
in Theorem 5.19 below. We cal1 an element v of a relational structure A =
(A, R,), reflexive iff each R, contains the tuple (u,. . u) (with the appropriate
E m ,
number of places). If this A is an algebra (that is, if each R, is really an operation),

then a reflexive element of A is simply a one-element subuniverse of A. The
following lemma was proved for algebras in Exercise 5.10(1) at the end of 55.4.
An endomorphism of a binary structure A is simply a homomorphism A -+ A.
LEMMA 8. If v is a reflexive element of a binary structure A, and f is a

decomposition operation of A, then f, is an endomorphism of A. Moreover, for any
u, the range of fv is the congruente block B = ulker fi, and the binary structure
B = (B, R í l B2) is isomorphic to Alker f,.
Proof. To see that f, is a homomorphism, let us be given (a, b) E R. Since

(u, u) E R, we easily obtain f (a, u) R f (b, u) (i.e., f,(a) R f,(b)), as required of a homo-
morphism. It follows easily from Lemma 4 that the range of f, is B, and that the
quotient homomorphism n: A -+ A/ker f, is a bijection when restricted to B. Thus
nl,: B -+ Alker f, is a one-to-one homomorphism. To see that n/,is an iso-
morphism, we let alker f, and blker f, be R-related in A/ker f, for some a, b E B =
range(f,); it will certainly suffice for us to prove that (a, b) E R. Notice first of al1
that since f, is idempotent and a, ~ E Bwe , have f,(a) = a and f,(b) = b. The
binary relation of Alker f, was defined to be the smallest relation such that n is
a homomorphism, and therefore there exists (a', b') E R such that f,(af) = f,(a) =
a and f,(bf) = f,(b) = b. Since f, is a homomorphism, we know that R contains
(fv(af),fv(b'))= (a, b). m
The following theorem is true for arbitrary relational structures, but for
simplicity, our proof will assume that A is a binary structure. In any event, this
is the case of interest for our applications.
THEOREM 5.19. If (A, u) has the intermediate refinement property, and v is a

reflexive element of A, then A has the refinement property.
Proof. We are given A = ( A , R) " ni,, nj,,

Bi E Cj, and we need to find
binary structures D,, as specified in the definition of the refinement property at
the beginning of this section. That is, we need each Bi isomorphic to
and each Cj isomorphic to ni E Dij.
m, ,D,,
Since A has a reflexive element, its relation R is nonempty, so by Lemma 3

and its corollary there exist equivalence relations fli (i E I) and yj ( jE J) such that
each B, A/fli, each Cj z A/yj, and O = ni,,fli n = j,J yj. For each i~ 1, let us
write for n,, , p,, and yj for n,, y,, and let us use fi and fi' to denote f, and
f:,where f is the decomposition function associated (via Lemma 1 near the
beginning of this section) with the direct product decomposition O = Di x 8,. It
follows from Lemma 8 that fi and fi' are endomorphisms of A. Endomorphisms
gj and g/ are defined analogously for j E J.
We will first show how to represent each B, as a product of algebras Dij for
EJ. The definition of Dii being symmetric in the i's and the j's, it will then be
apparent that we also have each Cj isomorphic to the product of the D, for i E l .
For convenience, we will denote a substructure of A simply by its universe;
in particular, we are interested in the substructures v/% and some closely related
substructures. Let us define g: A -+ n,, v / x via g(a) = (gj(a): j~ J). Then, ac-
cording to Lemma 8, g is a homomorphism, and we moreover have the following
homomorphisms:
where the left-hand vertical isomorphism was given, the lower isomorphism
comes from Cj E A/yj (stated above), and the right-hand (vertical) isomorphism
comes from Lemma 8. We leave it to the reader to sort out that al1 these
isomorphisms are compatible with one another and with g (i.e., that this is a
commutative diagram), and hence that g is an isomorphism.
Fix an i E l. We claim that, given the intermediate refinement property for
(A, v), we have
(thought of, for the moment, as merely the equality of two sets). To prove this,
we simply observe that
a E v/& uf;(a) = v
gj(5(a)) = v for al1j
-f;(gj(a)) = v for al1j
(by the intermediate refinement property)
+-+ gj(a)E v/& for al1j
thus verifying the claim.

Since g is an isomorphism, the substructure of A with universe v/& must be
isomorphic to the substructure of n,,,
v/j$ with universe n,,(1113 ílVIL).It is
easy to check from the definitions that this last substructure is equal to the
-
product of the substructures with universes (v/x í l v/&) In other words, if we
define Dij to be the substructure of A with universe v/(xíl&), then Biis iso-
morphic to nj, As we remarked above, the symmetry of the definition
of Dij allows us to conclude immediately that we also have, for each j, Cj
ni,, D,, thereby completing the proof of the refinement property for A.
Our first application of the intermediate refinement property is for a result

originally proved by C. C . Chang in 1966. In the present context, it is an
immediate corollary of Theorem 5.18 and 5.19. It has weaker assumptions than
those of the first corollary to Theorem 5.18 (here we need only a single anti-
symmetric element), but also a weaker conclusion. An element u of the binary
structure (A, R) is called antisymmetric iff x R u and u R x together imply x = u,
for each x in A.
COROLLARY. Every connected reflexive binary relational structure with an

antisyrnrnetric elernent has the refinement property.
Proof. If u is antisymmetric for A = (A, R), then it is not hard to check that
the = -class of u consists of u alone. Therefore (A, u) has the intermediate refine-
ment property by Theorem 5.18, so Theorem 5.19 completes the proof. w
In our last proof, we saw how Theorems 5.18 and 5.19 immediately yield
the refinement property whenever there exists a singleton = -class for the (reflex-
ive) binary structure (A, R). In fact McKenzie was also able to obtain a refine-
ment theorem for the case of a finite = -class.
THEOREM 5.20. If A is a connected reflexive binary structure, and f one

r -class of A is finite, then A has the refinement property.
COROLLARY. Every finite connected reflexive binary structure is uniquely

factorable. ¤
Proof of the Theorem. Let us be given isomorphisms 4: A -+

nj, ni,,
B~and $: A -+
,Sj. As a special notation inside this proof, we will for any binary structure
E let E denote the quotient structure of E by the special equivalence relation =
defined in this section. The natural map 9: E -+E is a homomorphism and also
has the property that if a l = R b/= in E/-, then a R b in E. It is also not hard
6:
to show that there are isomorphisms A -+ ni,,Biand $: A + njEJ
cj
such
that
is a commutative diagram, where 9' is the homomorphism that acts as Oi in the

ith coordinate, and likewise regarding the corresponding diagram for $ and $.
For each i, we let E be the factor relation of A that is the kernel of pi 0 4 (with pi
the projection of ni,, Bi onto B,), and for each j, we let Y; be the factor relation
that is the kernel of pj 0 @ (similar notation). Clearly, O = i,I n n x.
= j,
the thinness of A and Corollary 1to Theorem 5.18, we know that strict refinement
By
holds for A, so there exist congruences 6,on A such that = ai nj, ,6;, for each
i, and Y; = ni,, ---
&. Letting Dij denote the quotient stru-ure - A-,, we know
that there are isomorphisms
- - njEJ
Ai: Bi 5 nij and 6: ni,,
-
Cj -+ D, such that,
for each i and j, pj 0 Ai o pi 0 4 = pi O & 0 pj O $, where pj denotes projection onto the
jth factor.
In other words, the following two maps from A to ni, D, are the same:
where by 1we mean the map that takes (6,: i E 1) to (4:

i E 1,j E J), where d,, is
defined by Ii(b,) = (4:
j~ J); similarly for OJ and p. Now for x E ni,,bij, we
define
-
i.e., the number of preimages of x either under 109' or under OJ. (The equality
of the two maps just above tells us that these two cardinals are equal to each
other.) Thus the cardinals F(x) are just the cardinalities of the -classes of our
original structure A. Then for y E B, and z E q,
we put
Now it follows from the definition of the maps 1and ,üthat
where, for each j, x, is the element of

each i, xi, is the element of njE,
ni,, defined by x,(i) = x(i,j), and for
defined by xi*(j) = x(i,j). It is now possible
to prove, as a lemma of pure set theory (mostly cardinal arithmetic) that-under
our hypothesis of the existence of a finite E-class (i.e., the assumption that F(x)
is finite for some x)-there exist cardinal functions
-
Kij: Dij + Card,
for i E I and j E J such that
for each i and each y E nje,Di, and

for each j and each z E ni,, DijeThe only proof that we know for this result is
long and tedious, but nevertheless the individual steps are straightforward. We
therefore ask the reader to prove this in Exercise 15 below.
Given the cardinal functions Kij, we can build the structures D,, which we
require for the refinement property, by a procedure we now describe for a fixed
i~ I and j~ J . Let Z = (2,: x~ D ~ be ) a disjoint family of sets such that JZ,] =
Kij(x) for each X E & . We define the universe Dij of Di,ito be UZ, and define
0,: Dij + 4 via Oij(z) = x, where Z, is the unique set in Z that contains z. To
complete the definition of Di,i,we define the relation R of Dij to be the largest
-
binary relation on Dij such that 0, is a homomorphism. That is to say, if x R X '
in Dij, then we put z R z' in D, for each z E 2, and each z' E Z,.. (By the way, it is
now not hard to see that the kernel of 0, is the -relation of Dij, thereby belatedly
-
justifying the --notation for D, and the notation 0, for the natural map D, + D,.)
To complete the proof of the refinement property for A, we need only show that
Bi nj,, ni
Dij and Cj z E , Dij; by symmetry, it will suffice for us to show only
the first of these.
Let 0;: LE
Given y = (yj: j E J) E
- nj,,
nj, be defined by &(<yj:j~ J}) = (oij(&): j~ J } .
Dij, we calculate that
In other words, if &(x) = y, then 0 ~ ' ( x )is a subset of Bi that is equipotent with
njE,oij
the subset (?:-'(y). Therefore, if we now define Li: Bi -+ to be any function
which maps (?r1(x)bijectively to 0:-'(;ii(x)) for each X E Bi, then Li will be a
bijection of Bi with njEJ

Dij such that 13;o Ai = & o Oi.It follows readily, as the
reader may check, that Ai maps the binary structure Bi isomorphically onto
nj.J D,. m
Finally, we close this section with one application of McKenzie's Lemma
that takes us back to the consideration of ordinary algebras and illustrates a
method of interpreting one relation in another. In its statement (but not in its
proof), this theorem has some features in common with Theorem 5.14 above.
THEOREM 5.21. Let A = (A, + ) be un algebra with a single binary operation

+
+. If A has exactly one element u such that x + u = u x for al1 x E A, and if
u + u = u, then A has the refinement property.
Proof. Define the binary relation R on A via
It is obvious that {u) is a singleton =-block of the reflexive binary structure

(A, R). Connectedness of (A, R) is obvious from the fact that u R x for every
x E A. Therefore, ((A, R), u) has the intermediate refinement property, by
Theorem 5.18. From this we will deduce that (A, u) has the intermediate refine-
ment property, thereby proving the refinement property for A.
To see that (A, u) has the intermediate refinement property, we will show
that if f is any decomposition operation of A, then f is a decomposition operation
of (A, R). For (18) and (19), there is nothing to prove, for these already hold by
virtue of the fact that f is a decomposition operation of A. Al1 that remains is to
see is that f is a homomorphism from (A, R)2 to (A, R). This fact is obvious
from the definition of R, so we have proved our claim that every decomposition
operation of A is also a decomposition operation of (A, R). With this observa-
tion, the intermediate refinement property for (A, u) follows immediately from
the corresponding property for ((A, R),u), together with the definition of
intermediate refinement. Since u is a reflexive element of A, Theorem 5.19 yields
the refinement property for A. ¤
Exercises 5.22
1. The strict refinement property for an algebra A is equivalent to the following
condition. If O = a, x a , = ni., yi, then
2. Let A be a countable algebra with no operations. Then A has the

refinement property but not the strict refinement property. A cannot be
written as a product of directly indecomposable algebras.
3. More generally, we may regard any cardinal K as being an algebra with no 1
operations. Prove that, as an algebra, a cardinal rc has the refinement

property if K < 2'0, but 2'0 itself does not. Any strong limit cardinal has
the refinement property. (Note that, for rc = N,, the question whether
rc < 2"o or rc = 2"o-otherwise known as the continuum hypothesis-
cannot be resolved under any currently accepted scheme of mathematical
reasoning. (I.e., unless they already contain inconsistencies, the axiom sys-
tems ZFC, GB, etc., yield neither a proof nor a disproof of the continuum
hypothesis.) Thus N, is an example of an algebra for which the refinement
problem cannot be resolved.)
4. A finite algebra has the refinement property iff it has the unique factoriza-
tion property.
5. More generally, suppose that A has the UFP, and moreover satisfies the
following strengthening of part (i) of UFP: every direct factor of A has a
factorization into a product of indecomposables. Prove that A has the
refinement property.
6. Prove that the No-fold direct power (Z,, + )" of the two-element group is
uniquely factorable, but that the refinement property fails for (Z,, +)".
(Unique factorability requires the set-theoretic assumption that if 2" = 2"
for cardinals rc and A, then rc = A. This assumption follows, for instance,
from the Generalized Continuum Hypothesis (GCH).)
7. Prove that the No-fold weak direct power (B,, +)(") has the refinement
property but has no factorization into directly indecomposable algebras.
((Z,, + )(") is the subgroup of (Z,, + )" consisting of al1 w-sequences that
take only finitely many nonzero values.)
8. Prove that (Z,, +)" x (Z,, +)(") has no factorization into indecom-
posables, and that the refinement property fails for this algebra.
9. Prove the assertion in the text that except for the implication R -,(ii) and
for the tautology (i) v (ii),the two parts of UFP and the refinement property
are independent. (Of the eight possible Boolean combinations of (i), (ii), and
R, three are ruled out by the given relationships. To give examples for the
five remaining implications is mainly a matter of looking over the examples
we have already developed. The example of Exercise 6 may be used here,
since there exists a model of set theory in which GCH is true.)
10. If B is any Boolean algebra, then the factor congruences of B form a Boolean
algebra that is isomorphic to B.
11. (J. M. G. Fe11 and A. Tarski [1952].) Every perfect group has the strict
refinement property. (A group is called perfect iff every element is a product
of commutators.)
12. More generally, cal1 an algebra A in a congruence permutable variety
perfect iff [l, 11 = 1 holds in ConA for [m, -1the commutator operation
introduced in $4.13. Prove that if A is perfect, then A has the strict refine-
ment property.
13. If A lies in a congruence permutable variety, and the center congruence of
A (see $4.13) is O, then A has the strict refinement property.
14. (J. M. G. Fe11 and A. Tarski [1952].) For an algebra A with zero, the follow-
ing condition is equivalent to the strict refinement property for A: if O =

a x /3 = a x y (for a, p, y E Con A), then p = y. (By the way, a more or less
parallel assertion in pure lattice theory was conjectured in 1904 by E. V.
Huntington: Every uniquely complemented lattice with O and 1is a Boolean
algebra. This conjecture was, however, disproved by R. P. Dilworth in
-
[1945].)
*15. Prove the set-theoretical lemma used in the proof of Theorem 5.20. Let Dij
be sets indexed by i E I and j E J . For each i E I, let Gi:
ni,
for each j E J , let Hj: Dij + Card, in such a way that
n,,
Dij Card, and
ni,
for every y E Dij.(Here yi, denotes the element of
y(i,j), and y,, denotes the element of niér
nj,
D, with y,( j) =
Dij with ~ , ~ (=i )y(i,j).) The final
ni
-
assumption is that the cardinal Gi(yi,) is a finite number N for at least
one y E ni,, D ~The
~ . needed lemma then asserts that for each i E I and j E J ,
there exists a function Kij: Dij Card such that
for each i E I and each y E n jeJ Dij, and such that
for each j E J and each y E nicI

Dij. (Hint: An induction can be based on the
total number of prime factors dividing N (counting repeated factors).)
*16. Prove the following theorem of C. C. Chang, B. Jónsson, and A. Tarski.
Suppose that A has a one-element subuniverse (u) such that, for al1 x,
for al1 elements x, y of A. Then A has the refinement property.

17. Prove that Theorem 5.18 continues to hold if the assumptions of reflexivity
and connectedness are replaced by the assumptions that (A, R" o R) and
(A, R 0 RU) are both connected.
18. Prove that the strict refinement property (for an algebra A) is equivalent to
the seemingly weaker assertion that if
for congruences A,yj on A, then there exist equivalence relations 6, on A

(for i E I, j E J ) such that
pi = n 6, and
jé J
for al1 i E I and al1j E J.

5.7 Cancellation and Absorption 319
5.7 CANCELLATION AND ABSORPTION
In this section, we will consider some aspects of direct factorization theory related
to kth roots and cancellation. We will focus our attention on three important
implications:
(39) Ak N Bk -+ A r B (k finite and nonzero)
(40) AxCrBxC-ArB
(41) ArAxBxC-ArAxB.
These implications do not generally hold for infinite algebras, so we will be
examining them under some hypotheses of finiteness. The third one is in fact
trivially true for A finite but, surprisingly, also holds under the special assumption
of A countably infinite and either B or C finite. In the first two of these implica-
tions, we will be assuming that A, B, and C are al1 finite. Notice that both of these
implications can easily be proved if we have unique factorization results available
for Ak and for A x C (e.g., either for an algebra with modular congruences and
a one-element subuniverse or for an algebra with zero). (In the latter case, (40)
holds even with A and B infinite; see Theorem 5.9 in 55.4.) It thus came as a sur-
prise in 1966 to learn that (39) holds for al1 finite algebras A and B, and with a
proof much easier than the unique factorization proofs we have seen. The cancel-
lation law (40), while not universally valid, does hold for finite algebras if C has
a one-element subuniverse; the proof is again simple. In most of this section we
will be examining this beautiful method for (39) and (40), which is due to L.
Lovász. Differing sharply from the other methods in this book, it proceeds by
simply counting homomorphisms.
The results of this section apply not only to algebras, but also to relational
structures (as defined in Exercise 7 of 54.4 and discussed in 95.6). Moreover, the
proofs we present also apply in this broader context. We will not explicitly
mention this natural extension to relational structures until, at a certain point,
the proof methods themselves force us to include relational structures. This
situation begins with the proof of the Lovász Isotopy Lemma (Theorem 5.24),
which involves an expansion of the original similarity type. (In that proof, by
the way, we take exception to the definition of relational structure in Exercise
7 of 94.4 and permit structures to be empty.) Some of the later results of this
section (notably Theorems 5.25 and 5.26 and their corollaries) refer essentially
to the collection of al1 structures of a given type. Theorem 5.25, for example,
characterizes the failure of cancellation in the class of al1 finite structures of a
given type. The failures of cancellation described in the proof of that result
involve factors that are not algebras and cannot be made to look like algebras.
Until we come to consider (41), we will be taking al1 algebras (and struc-
tures) to be finite. We will let h(D, A) stand for the number of homomorphisms
D -+ A and m(D,A) stand for the number of these homomorphisms that are
monomorphisms (i.e.,one-to-one). We begin our considerations by recalling from
the corollary to Theorem 4.10 that every homomorphism f : D -+ A factors as
where 8 = kerf, 7~ is the natural homomorphism determined by 8, and z is an

embedding. It follows easily that
(regardless of any finiteness assumptions). Now (42) tells us that, for fixed A, the
system of cardinals (h(D, A) : D finite) is determined by the system of cardinals
(m(D, A) : D finite). In fact, for finite A, it is also true that (m(D, A) : D finite)
is determined by (h(D, A) : D finite). To see this, we use the finiteness of al1
cardinals involved to rewrite (42) in terms of subtraction, thereby obtaining
We are now ready to prove the following lemma.
LEMMA. For each finite algebra D there exist integers pe (8 E Con D) such that
for al1 finite algebras A (of the sume type),
m(D, A) = p,h(D/8, A).
e~ConD
Proof. By induction on ID 1, using (43). E
Thus of course the system (m(D, A): D finite) is determined by the system
(h(D, A): D finite). It is interesting to observe that the integers pe depend only
on the lattice Con D. The procedure of obtaining an equation like (44) from one
like (42)can be carried out for lattices in general (not just for congruence lattices).
This procedure is called Mobius inversion in the general context of number
theory, combinatorics, and lattice theory. Exercises 2-12 below contain a brief
development of this theory.
In order to apply (44) to a proof of (39) and (40), let us fix a similarity
type and take Di (i < K) to be finite algebras of this type that include, up to
isomorphism, every finite algebra of this type. Now for any finite algebra A of
this type, we define
and we let h(A) stand for the entire sequence (hi(A): i < K).
THEOREM 5.23.
i. h(A) = h(B) #A z B.
ii. hi(A x B) = hi(A)hi(B)for each i < K.
iii. If C has a one-element subuniverse, then each hi(C) > 0.
Proof. (iii) is obvious, and (ii) is immediate from the opening lemma of 54.4
(equivalently, from the categorical description of products on page 135 in
Chapter 3). For (i), it is certainly clear that h is an isomorphism invariant-i.e.,
that A r B implies h(A) = h(B). Conversely, if h(A) = h(B), then in particular we
have h(A/8, A) = h(A/8, B) for every quotient A/8 of A. By (44) we immediately

have m(A, B) = m(A, A), which is of course a positive integer. Thus m(A, B) 2 1,
so there exists an embedding 4 of A into B. Likewise, there exists an embedding
of B into A, so A and B have the same number of elements. Therefore, since B is
finite, the above embedding 4 is an isomorphism between A and B. ¤
The next corollary refers to the (commutative) monoid Mx of isomorphism

classes of algebras under direct product, introduced at the end of 53.3 and
discussed again in 55.1.
COROLLARY 1. If X is the class of al1 finite algebras of a given similarity

type, then Mx is isomorphic to a submonoid of a direct power of ( w , -,1) (the
monoid of non-negative integers under multiplication).
Proof. The isomorphism is given by the h of Theorem 5.23. •
COROLLARY 2. (L. Lovász [1967].) If A and B are finite algebras, and if

Ak r Bkfor somefinite k > O, then A r B. (In other words, (39) holdsfor al1finite
algebras.)
Proof. This is immediate from Corollary 1 and the fact that kth roots are unique
in (a,-,1). ¤
COROLLARY 3. (L. Lovász [1967].) If A, B, and C are finite algebras, fi C

has a one-element subuniverse, and if A x C r B x C, then A z B. (In other
words, (40) holds for al1 finite algebras A, B, and C such that C has a one-element
subuniverse.)
Proof. By part .(ii)of Theorem 5.23, we know that hi(A).hi(C) = hi(B) hi(C) for
each i. By part (iii) of that theorem, each hi(C) is nonzero, so hi(A) = hi(B) for
each i. Therefore A E B by part (i). M
COROLLARY 4. If A, B, C, and D are finite algebras such that there exists a

homomorphism D -, C, and ifA x C r B x C, then A x D r B x D.
Proof. By part (ii) of Theorem 5.23, we know that hi(A).hi(C) = hi(B) hi(C)for
each i. The existence of a homomorphism D -,C implies that if hi(C) = O, then
hi(D) = O. From this it easily follows that hi(A).hi(D) = hi(B) hi(D) for each i.
Therefore, A x D z B x D by parts (ii) and (i) of that theorem. H
Let us call an algebra C cancellable in a class X of algebras iff C E X and C

has the following property: for al1 A, B E X ,if A x C z B x C, then A S B. (For
one example, see Theorem 5.9 of 55.4.) We call C cancellable among finite algebras
iff C is cancellable in X for X the class of al1 finite algebras of its similarity
type. We begin a brief study of cancellation with an easy result that follows
immediately from Corollary 4.
COROLLARY 5. If D is cancellable among finite algebras, and if there exists

a homomorphism D -+ C, then C is cancellable among finite algebras. ¤
We saw in Corollary 3 (of Lovász) that every finite algebra with a one-
element subuniverse is cancellable among finite algebras. The converse is false:
there also exist algebras without one-element subuniverses that are cancellable
among finite algebras, as we shall see in the second corollary to the next theorem.
Before continuing our study of cancellation, however, it will be convenient to
return for a while to the study of isotopy we began in 55.2. Recall that A -cB
means that there exists an isomorphism 4: A x C -+ B x C such that the second
-
coordinate of 4(a, c) is c for al1 a and c, and A B means that A wcB for some
C. If this holds with C finite, then we will write A w f i nB. For relational structures
A, B, and C, the basic notion A -cB is defined exactly as for algebras. If the
relations of C are al1 empty, however, then the relation -c is trivial; structures
A and B (of the same type as C) satisfy A -c B iff 1 Al = 1 BI. For this reason, we
instead to define A -"

-
do not define an unrestricted relation A B for structures A and B, but prefer
B, for X a class of structures, to mean that A -c B for
some C E X .
Suitably strengthened, the Lovász method of proof, which led to Corollary
3, also yields the following theorem, in which A, B, and C may be either
finite algebras or finite structures, al1 of the same similarity type. Recall that a
homomorphism from a structure ( A , R,),, to another structure (B, S,),, is a ,
function f : A -+ B such that, for al1 t, if (a,, a,, E R,, then (f(a,), f(a,),
S ) E S,.
m )
THE-LOVÁSZ ISOTOPY LEMMA (5.24). (A, B, C .finite algebras.) If

A x C E B x C,thenAwcB.IfA x C N ~ ' " Bx c,thenAWfinB.
(A, B, C finite structures, X a class of finite relational structures closed
under finite products.) If A x C 2 B x C, then A -c B. If A x C -"
B x C and
C E X , then A wJ6'B.
REMARK: The reader may easily check that Corollary 3 above (the Lovász
cancellation theorem for algebras with a one-element subuniverse)is an immediate
consequence of the Isotopy Lemma.
COROLLARY l. w f i nis the smallest equivalence relation on the class of finite

algebras that contains the isomorphism relation and is cancellative over x . For X
a class of similar finite structures that is closed under finite products, -"
is the
smallest equivalence relation on X that contains the isomorphism relation and is
cancellative over x . E
Proof of the Isotopy Lemma. We will give the proof for structures only. The
idea is simply to use the same reasoning that led to Theorem 5.23 and its
corollaries, but in a different category. We fix a finite structure C and consider
the category U(C) of structures over C: an object of U(C) consists of a finite
structure A and a homomorphism 4: A -+ C. (For the purposes of this proof, we
permit structures to be empty.) A morphism in U(C), between objects (A, 4) and
(B, S/), consists of a homomorphism f : A + B such that 4 = S/ o f.
If A is any finite structure, then (A x C,p,) is an object of U(C) (with p ,

denoting the second coordinate projection), which we will denote Ac. The
structure Cc will play a special role; it is simply C2 with the second coordinate
projection onto C. Al1 U(C)-objects we will explicitly describe are of the form
Ac, but one should bear in mind that the proof proceeds inductively by looking
at homomorphic images, and these are not usually of the form Ac.
It is not difficult to check that products exist in the category U(C). In fact,
for (A, 4) x (B, S/) we may take (D,A), where D is the structure with universe
D = ((a, b): 4(a) = S/(b)) c A x B, and with relations defined as usual for a
substructure of a product, and where A = 4 o (p,) , 1 = S/ o ( p ,). ,1 In particular,
(A x B)c is the product of Ac and Bc.
Just as we did for Theorem 5.23, we may let (A,, 4,), (A,, 4,), be a listing
of al1 the objects in U(C) (up to isomorphism inside the category U(C)), and then
define h,(A, 4) to be the number of U(C)-morphisms from (A,, &)to (A, 4). In other
words, hi(A,4)= 1 hom,(C) ((Ai,$i), (A, $))l. (Here the index i ranges over some
appropriate ordinal a. In the case of a finite number of relations, we may take
a = m-i.e., we may speak of an ordinary sequence of invariants h,(A, 4).) As in
the proof of Theorem 5.23, h,((A,4) x (B, S/)) = hi(A,4)hi(B,S/) for each i, a ~ d
(A, 4) r (B, S/) iff hi(A,4) = h,(B, S/) for each i. But in the present context, there is
one bit of information we did not have available in our previous work. Notice
that for each i there exists a U(C)-morphism f : (A,,4,) + Cc given by f(a) =
(@,(a), 4,(a)), and therefore hi(Cc) # O.
Now, in order to prove the first assertion of the lemma, we begin by
supposing that A x C r B x C. Then in U(C) we have (A x C)c r (B x C)c,
and so, for each i, hi(Ac) - hi(Cc) = hi(Bc).hi(Cc). Since the numbers hi(Cc) are
nonzero, we have hi(Ac) = hi(Bc) for each i. It follows that Ac g Bc in the
category U (C). Finally, it is easily seen that Ac r Bc is equivalent to A -c B,
which is the desired conclusion for the first part of the Isotopy Lemma.
As for the final part, we are given A x C wXB x C and C E X . Thus
A x C x D r B x C x D for some D E X . The part of the Isotopy Lemma we
already proved now yields A B. Since C x D E X , we have A B.
The following corollary yields a large family of examples of algebras C that

are cancellable among finite algebras but do not have one-element subuniverses.
As one example, we might let B = (B, A , v , -) be a Boolean algebra, and then
define C = (B, F) where F(x, y) = (x A y)-.
COROLLARY 2. If C = ( C , F) is a finite groupoid with elements c, d such that

F(c, d) = F(d, d) = F(d, c) = c, then C is cancellable among finite algebras.
Proof. Let us suppose that A = (A, F*) and B = (B, F ~ are ) finite algebras
such that A x C r B x C. By the Lovász Isotopy Lemma, A -c B, which means
that there exist bijections $u:A + B (u€ C) such that the following map is an
isomorphism:
S/:AxC+BxC
We will make use of only 4c and q5d. We claim that 4c maps A isomorphically
onto B.
To begin our calculations, we make use of the fact that F(c, d) = c. Taking
any a, a' E A, we calculate that
= $(FA c((a, 4 , (a', 4))= FB C($(a,c), $(a1,4 )

- FB C((4c(a),c), (4d(a'),d))
Now comparing first coordinates, we have
Similar calculations also yield
Now the following calculations establish that 4c is an isomorphism.
This example may be interpreted as saying that the cancellation law (40)
holds whenever the multiplication table of the finite groupoid C contains a
fragment that looks like this:
The reader may enjoy devising some other fragmentary multiplication tables that
imply (40). We will give one example in Exercise 13.
Returning now to our study of cancellation, we may note that Corollary 5
to Theorem 5.23 suggests a possible form for a characterization of algebras (or
structures) that are cancellable (among finite algebras or structures). In this
context, it is convenient to cal1 an algebra (or structure) a zero-divisor iff it is not
cancellable among finite algebras (or structures). For every similarity type p there
exists a class 3 of algebras (or structures) of type p such that the zero-divisors
of type p are precisely those algebras (or structures) that have a homomorphism
into some element of S. (By Corollary 5, 2 may be taken to be the class of al1
zero divisors of type p.) For a nice result, such a 3 should be a fairly restricted
class of simple and familiar algebraic objects. We do not know any such 3 for
the class of finite algebras of a given type, but for finite relational structures, L.
Lovász and R. R. Appleson discovered a very nice 3,which we will now describe.
Although our main result (Theorem 5.25 below) holds for al1 structures, we
will for simplicity limit ourselves to discussing structures A = (A, R, S) with R
m-ary and S n-ary. Our characterization of algebras cancellable among finite
ones of this type will be stated in terms of some special structures, which we now
define. A coset structure is one of the form (G, Hr, Ks), where G is the universe
of a finite group (G, -,-l), Hr is a right coset defined by some subgroup H of
(G, , and some r = (r,, ,rm)E Gm,and similarly, Ks is a right coset of a
subgroup of (G, , -l)". A reflexive element of a structure ( A , R, S) is any a E A
such that (a, ,a) E R and (a,. a) E S. A structure with no reflexive element will
m ,
be called irreflexive.
LEMMA 1. Let (G, S , -l)be a finite group, H a subgroup of (G,

e, and K
a subgroup of (G, S , Then (G, H, K) E (G, Hr, Ks) fi (G, Hr, Ks) has a
reflexive element.
Proof. Suppose first that g E G is a reflexive element of (G, Hr, Ks). Then the
map g t-,x . g is an isomorphign from (G, H, K ) to (G, Hr, Ks). For the con-
verse, it is enough to check that (G, H, K) itself has a reflexive element. Obviously,
the unit element e will do. m
LEMMA 2.
(G, H, K ) x (G, Hr, Ks) E (G, Hr, K S ) ~ .
Proof. The isomorphism is easily established by mapping (x, y) E G2 to (xy, y).

m
Before proving the main theorem about coset structures, it will be helpful
to have available the following elementary lemma about cosets.
LEMMA 3. Every right coset Hr of any group (G, -') is closed under the
a,
ternary operation (x, y, z) it x y-' z. If X is any nonempty subset of G, then the

smallest right coset containing X is the set of al1 elements of G that have the form
for any finite k and any x,, S , x2, EX.
Proof. The first assertion is trivial. For the second, we ask the reader first to
check that the set we have described is in fact a right coset that contains X. (It
is a right coset over the group generated by al1 x, - x l l for x,, x1 EX.) On the
other hand, by the first assertion of the lemma, every right coset that contains
X must contain al1 the elements that we described, and so this is the smallest
one. ¤
THEOREM 5.25. A finite structure C is a zero-divisor @ there exists a homo-

morphism from C into un irrejlexive coset structure.
Proof. As we mentioned above, we will state our proof in terms of structures

of type (m, n). Lemmas 1 and 2 establish that every irreflexive coset structure is
a zero-divisor. Hence, by Corollary 5 to Theorem 5.23, the same is true of every
C admitting a homomorphism into an irreflexive coset structure.
Conversely, let us be given a (finite) zero-divisor C = (C, R, S) and prove
that there exists a homomorphism #: C -+ (G, Hr, Ks) with (G, Hr, Ks) ir-
reflexive. We are given that A, x C r A, x C for nonisomorphic finite struc-
tures A, and A,. Since 1 A, 1 = 1 Al 1, we may suppose that A, and A, have the
same universe: A, = (A, R,, S,) and A, = (A, R,, S,). By the Lovász Isotopy
Lemma (5.24), there exist permutations #c: A -+ A (for c E C) such that
defines an isomorphism of A, x C with A, x C. We now define the required G

to be the group of al1 permutations of A, and define #: C -+ G via #(e) = #c.
To complete the proof, we need to find cosets Hr and Ks such that (G, Hr,
Ks) is irreflexive and # is a homomorphism from C to (G, Hr, Ks). Well, we
make # into a homomorphism by fiat; that is, we define Hr to be the smallest
subset of Gmthat is a right coset of (G, -')" and that contains #(u) = (#(u1),. ,
a ,
#(um))for each u = (u1,.. ,um)E R. KS is defined similarly. By Lemma 3, Hr

consists precisely of al1 products
for u,, - m , u,, ER and any k 2 0. And similarly, Ks is given by al1 products
for U,, VZk E S.

m ,
Let us say, for this proof, that A = (A,, A,) maps R, to R, iff (A, (a,), - - ,
m ,
A,(a,)) E R, for every (a,, - a,) E R,. The fact that (45) defines an isomorphism
m ,
tells us that, for each U E R, #(U)maps R, to R, and [#(u)]-' maps R, to R,. It

follows from the above description of Hr that every A in Hr maps R, to R,.
Similarly, every element of Ks maps S, to S,.
At this point, al1 we need to see is that (G, Hr, Ks) is irreflexive. Suppose,
by way of contradiction, that (G, Hr, Ks) has a reflexive element p. Thus the m-
tuple (p, .. . ,p) is in Hr and hence maps R, to R,; likewise, the n-tuple (p, . - . ,p)
maps S, to S,. It follows that p is an isomorphism of A, and A,. Since A, and Al
are not isomorphic, we have a contradiction, which establishes that (G, Hr, Ks)
has no reflexive element. The proof of the theorem is now complete. •
REMARK: In Corollary 2 to the Lovász Isotopy Lemma (5.24), we gave an

example of a groupoid C = (C, F) that is cancellable in the class of al1 finite
groupoids. Let us see how this fact can also be deduced from Theorem 5.25.
Let (C, F) be as defined there, and let R be the ternary relation {(x,y, z) E C3:
F(x,y) = z). One may easily check that it is enough to prove that (C, R) is
cancellable among finite structures. Hence, by Theorem 5.25, it will be enough
to prove that if 4: (C, R) -+ (G, Hr) is a homomorphism with (G, Hr) a coset
structure, then (G, Hr) has a reflexive element. To see this, we recall that there
are elements c and d of C such that F(c, d) = F(d, d) = F(d, c) = c. In other words,
R contains the triples (c, d, c), (d, d, c), and (d, c, c). Since 4 is a homomorphism,
the coset Hr must contain the triples (E, d ~ )(d; , &E), and (d;F,F) (where ?? and d
abbreviate $(c) and 4(d)). By Lemma 3, Hr also contains
which is the reflexive element we require.

Given two similar structures A = (A, R,, R,, ) and B = (B, S,, S,, )
with A í l B = @, we define their disjoint union, denoted A U B, to be the structure
( A U B, R, U S,, R, U S,, ). Notice that the isomorphism type of A U B depends
only on the isomorphism types of A and B. This observation allows us to refer
to A U B even in the case where A and B are not disjoint. In this case, we allow
A U B to refer to the disjoint union of any two structures A' and B' that have
disjoint universes and satisfy A' z A and B' r B. In this more general case, A U B
is defined only up to isomorphism; since the next two results involve properties
of structures that are isomorphism-invariant, this uncertainty in the definition
of A U B is harmless. The interested reader may check that A U B is a coproduct
of A and B in the sense that it is a product of A and B in the dual of the category
of homomorphisms between relational structures (of a fixed type). (See 83.6 for
the notions of product in a category, and the dual of a category.) It is somewhat
surprising that we cannot seem to prove the next corollary without using
Theorem 5.25.
COROLLARY. If structures A and B are zero-divisors, then A U B is also a

zero-divisor.
Proof. Let C, and C, be zero-divisors; by Theorem 5.25 each Ci maps

homomorphically into an irreflexive coset structure A,. By Lemma 2 before
the theorem, there are structures B,, each with a reflexive element, such that
A, x Bi r A, x A, (i = 1,2). By Corollary 4 to Theorem 5.23, we have Ai x Ci z
B, x Ci (i = 1,2), from which one easily obtains
D x (C, U C,) r D' x (C, U C,),
where
D = [(Al x A,) U (Bl x B2)I
and
D' = [(A, x B,) U (A, x B,)].
But D and D' are not isomorphic, since D contains a reflexive element and D'
does not; thus C, U C, is a zero-divisor.
For binary relational structures, we can obtain a tighter characterization of

the zero-divisors. That is, we can find a class 2 of zero-divisors that is simpler
and more familiar than the class of al1 irreflexive coset structures and nevertheless
permits a characterization of zero-divisors in the style of Theorem 5.25. Consider
the binary coset structure (Z,,Hr), where Z, is the group of integers under
addition modulo p, H is the diagonal subgroup of Z;, and r = (O, 1). Clearly,
(E,, Hr) is the same as C, = ({O, - p - l}, R), where (m, n) E R iff n m + 1
S , -
(mod p). The structure C, is a directed graph, which can be depicted as follows:
(with an arrow from m to n iff (m, n) E R). The directed graph C, is also known as
a p-circuit.
THEOREM 5.26. A binary structure is a zero-divisor $ it has a homomorphism

hto Cpl U U C,, for some primes p,, p,. a ,
Proof. Each Cpi is an irreflexive coset structure, and so CplU... U C,, is a

zero-divisor by Theorem 5.25 and its corollary. It then follows from Corollary 5
to Theorem 5.23 that if C has a homomorphism into CplU . . - U C,,, then C is a
zero-divisor.
Conversely, let us suppose that the binary structure C is a zero-divisor and
then prove that C has a homomorphism into some CplU U Cpn.By Theorem
5.25, we may assume that C is itself an irreflexive coset structure-Le., that
C = (G, Hr) for some finite group G, some subgroup H of G2, and some r E G2.
We will think of G as a directed graph, with a (directed) edge from a to b iff
(a, b) E Hr. For a E G, we define a labeled path in G starting at a to consist of a
sequence of "points" a = x,, x,, - X, E G and a sequence of "edges" e,, e,, . .,
e,-, E Hr such that
for i < n. If e, = (x,, xi+,),then we call e, a forward edge of this labeled path, and
in the other case we call ei a reverse edge. (Clearly, a single edge cannot be both
a forward and a reverse edge.) This labeled path is called a path from a to b iff
x, = b. A circuit at a is a labeled path from a to a. The connected component of
a is the set of al1 b E A such that there exists a labeled path from a to b. It is well
known and easy to check that for al1 a, a' E G, the connected components of a
and a' are either identical or disjoint, so there is a partition of G into its connected
components. Equivalently, a connected component of G is a minimal nonempty
subset E of G that is closed in the sense that if a E E and either (a, b) E Hr or
(b, a ) €Hr, then b~ E. Since the relation (a, b) E Hr is impossible for a and b
in distinct connected components, we may proceed to define the desired homo-
morphism separately on each connected component. Therefore, for each con-
nected component G, of G, we will define a homomorphism 4: G, -+ C, for some
appropriate prime p; the required structure CpIU - U Cpnwill then be the disjoint
union of al1 these individual C,'s.
To define 4 on one component G,, we first fix a E G, (and thus G, is the
connected component of a). For each labeled path s starting at a, we define h(s)
to be m - q, where m is the number of forward edges in this path and q is the
number of reverse edges. It is easy to see that a circuit can be reversed and that
one circuit at a can be followed by another one, so
S = (h(c): c a circuit at a}
is a subgroup of the group Z of integers.
If G, is a singleton, then the relation Hr does not meet Gt, and so any
function 4: G, -+ C, will be a homomorphism. Otherwise, since Go is connected,
for each b E G, there is a labeled path s from a to b; we define bf(b)to be h(s).
Now #(b) clearly depends on the choice of the path S, but it should be clear that
if S' is another path from a to b, then h(s) - h(sf)= h(c) for some circuit c. In
other words, h(s) - h(sf)E S, SO 8 is well defined as a map from G to Z/S. It
is also apparent from the definition of h that 4' is a homomorphism from
(G,, Hr í l G;) to the directed graph C,, where n is the smallest positive member
of S. If p is a prime factor of n, there is a homomorphism 1,: C, -t C,, so we may
take 4 = 3L O 4'. Thus the proof is complete, except for the possibility that n = l.
Thus the remainder of the proof will be devoted to showing that n > 1, or in
other words, that 1 9 S.
Thus we need to prove that no circuit c has h(c) = l. Our proof will be by
induction on the number n of edges in c and will refer to circuits starting anywhere
in G, not just those starting at a. For an inductive argument, we suppose by way
of contradiction that h(c) = 1 for some circuit c, and then we choose such a c so
that its number n of edges is as small as possible. Clearly, n # 1, by the irreflexivity
of (G, Hr).
Obviously n is odd, so we may suppose that n 2 3. We will obtain our
contradiction by finding a circuit c' with fewer than n edges such that h(cf)= 1.
By definition of h, the circuit c must have exactly q + 1 forward edges and q
reverse edges (hence n = 2q + 1). We may group these edges into maximal
connected blocks of forward and reverse edges (with, of course, the last edge e,,
considered as adjacent to the first edge e,). Obviously, there are at least two
blocks; let us now prove that there cannot be exactly two blocks. If there were
exactly two blocks, then the circuit c would define elements of G on which the
relation Hr holds at least as often as indicated in the following diagram:
Now in the group G2 we form the alternating product

w = ( 4 , b1) ( 4 , d2)-' . (bl, b,) . (b,-, ,b,) (d,, e)-' (b,, e).
Since Hr is a coset, and w is an alternating product of elements of Hr, we have
w E Hr by Lemma 3. But this is impossible, since Hr is irreflexive and both
components of w are the same, namely
b, d;' b, - di1 . b,.
This contradiction shows the impossibility of having exactly two blocks, so we
may suppose that there are at least three blocks.
In the case of three or more blocks, we consider a block of smallest size. It
will be flanked by blocks at least as long. Without loss of generality, we may
suppose that we have a block of k reverse edges flanked by at least k forward
edges on either side. That is to say, there are elements of G on which the relation
Hr holds at least as often as indicated in the following diagram:
(Neither in this nor in our former diagram are we claiming that al1 displayed
elements are distinct from one another.) By Lemma 3, Hr contains (co,cl).
(e, b1)-l (e, a,) = (c,, c, b;' a,), (c, bll a,, c, by1 a,), and so on. We may
therefore replace the above depicted portion of c by
thereby obtaining a shorter circuit c'. It is clear that c' differs from c in having k
fewer forward edges and also in having k fewer reverse edges. It therefore follows
that h(cf)is also 1.This reduction of the supposedly minimal length of c completes
our inductive argument. •
We now turn briefly to a consideration of zero-divisors in graph theory. By

an (undirected) graph, we mean a finite binary structure (G, R) with R symmetric
and irreflexive. When graphs are conceived in this irreflexive way, the product
of structures (G, R) x (H, S ) yields what is known in graph theory as the weak
direct product of these two graphs. A graph (G, R) is called bipartite iff there
exist disjoint subsets A, B E G such that R S (A x B) U (B x A).
COROLLARY. Among finite graphs, the zero-divisors are precisely the bipartite
graphs.
Proof. If (G, R) is a zero-divisor, then by Theorem 5.26, for each connected

component G, of (G, R), there exists a homomorphism d from (G,, R í l Gi) to
C,, for some prime p. If R í l G; # a, then the symmetry of R implies that p = 2.
It is now easily seen that (G,, R í l G;) is bipartite-the sets A and B required
for the definition are simply $-'(a) and $-'(b) for a and b the two elements of
C,. On the other hand, if R í l Gg is empty, then that component consists of a
single point and is obviously bipartite. Thus every connected component of
(G, R) is bipartite, from which it readily follows that (G, R) itself is bipartite.
Conversely, one may easily check that if G is bipartite, then G x G, r
G x G,, where G, and G, are these two six-element graphs:
This last result concludes our examination of the Lovász methods for
cancellation of finite factors. We now turn briefly to the superficiallysimilar topic
of absorption. We say that A absorbs B iff A x B r A. Implication (41), stated
at the beginning of this section, holds that if A absorbs B x C, then A absorbs
B alone. In Exercise 9 of $5.1, we saw, in an example of W. Hanf, that (41) fails
for A a certain uncountable Boolean algebra, and for B, C both taken to be the
two-element Boolean algebra. In contrast to this, we have the surprising result
of R. McKenzie [1971] that (41) holds whenever A is countable and either B or
C is finite. (R. L. Vaught had previously proved the special case where A is a
countable Boolean algebra and B and C are finite Boolean algebras-see Hanf
[1957].)
We will in fact prove a slightly more general result. Let us say that an algebra
or structure A is infinitely divisible by B iff there exist algebras or structures A,
( n W)~ such that A r B x A, and A, r B x A,,, for each n ~ c o Although. we
will phrase the proof in terms of algebras, the following theorem and proof are
valid for both algebras and relational structures.
THEOREM 5.27. If A is infinitely divisible by B, with B finite and A countable,

then A r A x B.
Proof. Let us take algebras A,, A,, A, as described in the above definition
of infinite divisibility. Thus we have homomorphisms
countable and B is finite. Let (a,: i ~ c o )enumerate the elements of A, with

a' = (ci,(b;, bi,. )) in the above representation of A as a subalgebra of C x Bu.
Since B is finite, there exists an infinite set JoE co such that b; = b: for al1j, k E Jo.
Continuing in this manner, we see the existence of a nested sequence of infinite
sets, Jo2 J1 2 J22 such that, for each i,
b,' = bl for allj, k€Ji.
We now select natural numbers ni such that O < no < n, < and such that ni E Ji
for each i, and use these numbers to define A, as follows. If n does not occur in
the sequence O, no, n,, n,, . then l(n) = n. The other values of 3L are defined
m ,
by A(0) = no and 3L(ni)= ni+, for i E co. It is obvious that this 3L maps co bijectively
to co - (O), and hence that fa maps C x B" isomorphically to B x (C x Bu).
Consider an arbitrary element a' = (ci, (b;, bf , of A. If n 2 ni, then
S ) )
either A(n) = n or n and l(n) both lie in Ji, and so bi(,, = bi, either trivially or by
(47). In other words, for each a' there exists N such that b&,,. = bj for al1 n 2 N.
It is now obvious from the defining equation for fa that fa(al) = (b, a'), where a'
differs from a' only in finitely many B-coordinates, and hence that f,(a) E B x A.
In other words, fa maps A to B x A. To see that fa maps A onto B x A, consider
a typical element (b, a ') = (b, (c', (b;, bf ,. . ))) of B x A, and take N so that
bj(,, = bi for n 2 N. Now define a' = (c', (b;, b;, - - )) E C x B" as follows: c' =
ci, b; = b, b&,, = bi for n 5 N, and b; = bk for al1 other m. Since a' differs from
a' in only finitely many B-coordinates, a' E A. Since bi(,, = bj for n 2 N, it follows
easily that fa(a1)= (b, a'), and hence that fa maps A onto B x A. •
COROLLARY l. If A is countable and either B or C is finite, then A E

A x B x C implies A r A x B. (Thus (41) holds under these assumptions.)
Proof. First assume that B is finite. We have

A r B x (A x C),
AxCrBx(AxCxC),
AxCxCzBx(AxCxCxC),
and so on. In other words, A is infinitely divisible by B, and so by the theorem,

ArAxB.
In the other case, namely C finite, a parallel argument yields A r A x C,
from which we deduce that A x B r A x C x B A. w
COROLLARY 2. If A r B x C and B r A x D, with B countable and C

finite, then A r B.
Proof. A r A x C x D, and so, by Corollary 1, A r A x D r B. ¤
Corollary 2 may be likened to the Cantor-Bernstein Theorem of preaxio-

matic set theory: if A t,B U C and B t,A U D, then A t,B. (Here, t, means
merely the existence of a bijection.) Analogs of this theorem in categories other

than sets are extremely rare, and even in the present context we need our
restrictive cardinality assumptions. Hanf's example, mentioned above, obviously
provides a counterexample here too.
Exercises 5.28
l. Let C,, C,, . C, be finite structures, al1 of the same similarity type,
- a ,
and assume that Ci has a reflexive element for at least one i 2 1. For any
finite structure A of this type, put F(A) = C, U (C, x A) U (C, x A2)U U
(C, x A"). Prove that F(A) r F(B) implies A r B.
In exercises 2-11, we develop the theory of Mobius inversion for finite

ordered sets. Formula (44) of the lemma on page 320 is a special case of Mobius
inversion, with the coefficients p, there being special cases of the Mobius coeffi-
cients p(P),which we are about to introduce.
2. Prove that there exists an integer-valued function p defined on the collection

of al1 finite ordered sets with smallest and large5t elements, such that the
following assertion holds for al1 finite ordered sets P = (P, 5 )and al1 integer-
valued functions f and g defined on P :
(48)
a l b
iff
- -
a l b
(One may define p(P) by recursion on 1 P 1.)Prove that p is unique and depends
only on the isomorphism type of P. Prove also that the equivalence between
(48) and (49) holds for any ordered set P such that, for al1 b~ P, the set
{a E P : a 5 b ) is finite.
3. Taking p as in Exercise 2, prove that if D is a finite algebra, and the integers
p, are defined as in this section (i.e., so as to make (44) true), then in fact, for
each 0 E Con D, p, is p(P) for P the dual of the interval I[O, 01 in Con D. (The
correspondence is dual because tradition in the theory of ordered sets does
not correspond to the natural way of viewing the congruence lattice.)
4. Prove that ~(singleton)= 1, and that for any finite ordered set P with least
element O and greatest element 1, if O # 1 then
(Hint: Apply (48) and (49) to the function g such that g(0) = 1 and g(x) = O
for al1 other x. This exercise affords an easy inductive way to obtain p of any
finite ordered set.)
5. Evaluate p of the following ordered sets.
5.7 Cancellation and Absorption
6. If D is a finite algebra with more than one element, and p, is defined as above,
then
Fe = 0-
0 E Con D
7. Use Exercise 6, together with the fact (from Exercise 3) that p, depends only
on the isomorphism type of I[O, 81 in Con D to verify the indicated values of
,u, in the various finite congruence lattices depicted below.
8- P(P x Q) = P(P).AQ).
*9. If L is a finite lattice with more than one element, then p(L) = O unless L is
complemented.
10. 1f P is a finite distributive lattice with O and 1, then p(P) E ( - 1,0,1). (Hint:
Use Exercises 8 and 9.)
11. Let P be the set of positive integers ordered by the divisibility relation. Give
a general description of p(I[m, n]) for al1 m and n. (This is the Mobius function
of number theory.)
12. For functions a and p defined on the set co of natural numbers, with values
in an Abelian group, the two conditions
and
are equivalent. To prove this via the theory of Mobius inversions, let P be
the ordered set of al1 finite subsets of an infinite set, and then evaluate
p(I[F, K]) for F, K E P. Define g(F) to be a([FI) and f(F) to be P(I FI) for
F EP, and apply the equivalence of (48) and (49). (Here (i ) the
denotes
binomial coefficient-the number of m-element subsets of an n-element set.
This exercise is relevant to algebra, for we have the important case of
a(n) = the number of operations in a variety that depend on exactly n
variables, and P(n) = the cardinality of the Y-free algebra on n generators.)
13. Prove that the cancellation law (40) holds for finite A, B, and C if C is a
groupoid whose multiplication table has a fragment that looks like this:
a b c
Formulate some other results of this style.

Bibliography
R. R. Appleson and L. Lovász [1975]. A characterization of cancellable k-ary structures,

Period. Math. Hungar. 6, 17-19.
G. Birkhoff [1933]. On the combination of subalgebras, Proc. Cambridge Philos. Soc. 29,
44 1-464.
.- [1935a]. On the structure of abstract algebras, Proc. Cambridge Philos. Soc. 31,
433-454.
[1935b]. Combinatorial relations in projective geometries, Ann. Math. 47, 743-
748.
[1940]. Lattice theory, First Edition, Colloquium Publications, Vol. 25, Amer.
Math. Soc., Providence, R.I.
[1944]. Subdirect unions in universal algebra, Bull. Amer. Math. Soc. 50,764-768.
[1946]. Sobre grupos de automorfismos, Rev. Un. Math. Argentina 11,155-157.
[1948]. Lattice theory, Second Edition, Colloquium Publications, Vol. 25, Amer.
[1967]. Lattice theory, Third Edition, Colloquium Publications, Vol. 25, Amer.
C. Birkhoff and 0 . Frink [1948]. Representations of lattices by sets, Trans. Amer. Math.
SOC.64,299-316.
W. Blok and D. Pigozzi [1982]. On the structure of varieties with equationally definable
principal congruences 1, Algebra Universalis 15, 195-227.
W. Blok, P. Kohler, and D. Pigozzi [1984]. O n the structure of varieties with equationally
definable principal congruences 11, Algebra Universalis 18, 334-379.
G. Boole [1854]. An investigation of the laws of thought, Walton and Maberly, London.
S. Burris and H. P. Sankappanavar [1981]. A course in universal algebra. Graduate Texts
in Mathematics, Springer-Verlag, New York.
C. C. Chang, B. Jónsson, and A. Tarski [1964]. Refinement properties for relational

structures, Fund. Math. 55, 249-28 1.
C. C. Chang and H. J. Keisler [1973]. Model theory, Studies in Logic and the Foundations
of Mathematics, Vol. 73, North-Holland Publ. Co., Amsterdam-London.
P. M. Cohn [1965]. Universal algebra, Harper and Row, New York.
A. L. S. Corner [1969]. Additive categories and a theorem of Leavitt, Bull. Amer. Math.
SOC.7578-82.
P. Crawley and R. P. Dilworth [1973]. Algebraic theory of lattices, Prentice-Hall, Engle-
wood Cliffs.
338 Bibliography
A. DeMorgan [1847]. Formal logic, Taylor and Walton, London.

R. Dedekind [1897]. Über Zerlegungen von Zahlen durch ihre grossten gemeinsamen
Teiler, Festschrifte Techn. Hoch. Braunschweig.
[1900]. Über die von drei Moduln erzeugte Dualgruppe, Math. Ann. 53,371-403.
R. P. Dilworth [1945]. Lattices with unique complements, Trans. Amer. Math. Soc. 57,
123-154.
[1950]. The structure of relatively complemented lattices, Ann. Math. 51 (2), 348-
359.
(See also P. Crawley and R. P. Dilworth)
J. M. G. Fe11 and A. Tarski [1952]. On algebras whose factor algebras are Boolean, Pacific
J. Math. 2,297-318.
1. Fleischer [1955]. A note on subdirect products, Acta Math. Acad. Sci. Hungar. 6,
463-465.
R. Freese [1980]. Free modular lattices, Trans. Amer. Math. Soc. 261, 81-91.
R. Freese and R. McKenzie [forthcoming]. Commutator theory .for congruence modular
varieties. London Mathematical Society Lecture-Note Series.
O. Frink. (See G. Birkhoff and 0 . Frink)
Frobenius and Stickelberger [1879]. Ueber Gruppen von vertauschbaren Elementen,
J . reine angew. Math. 89,217-262.
G. Fuhrken [1973]. On automorphisms of algebras with a single binary operation,
Portugal. Math. 32,49-52.
N. Funayama and T. Nakayama [1942]. On the distributivity of a lattice of lattice-
congruences, Proc. Imp. Acad. Tokyo 18,553-554.
J. A. Gerhard [1970]. The lattice of equational classes of idempotent semigroups,

J. Algebra 15,195-224.
[1971]. Subdirectly irreducible idempotent semigroups, Pacfic J. Math. 39,
669-676.
S. Givant (See A. Tarski and S. Givant)
G. Gratzer [1968]. Universal algebra, First Edition, D. Van Nostrand, Princeton, N. J.
119781. General lattice theory, Pure and Applied Mathematics, Vol. 75, Academic
Press, New York; Mathematische Reihe, Band 52, Birkhauser Verlag, Basel; Aka-
demie Verlag, Berlin.
[1979]. Universal algebra, Second Edition, Springer-Verlag, New York.
G. Gratzer and E. T. Schmidt [1963]. Characterizations of congruence lattices of abstract
algebras, Acta Sci. Math. (Szeged) 24, 34-59.
H. Peter Gumm [1983]. Geometrical methods in congruence modular algebras, Memoirs
of the American Mathematical Society, Vol. 45, no. 286.
J. Hagemann and C. Herrmann [1979]. A concrete ideal multiplication for algebraic

systems and its relation to congruence distributivity, Arch. Math. (Basel) 32,234-245.
P. Halmos [1960]. Naive set theory, D. Van Nostrand, Princeton.
W. Hanf [1957]. On some fundamental problems concerning isomorphism of Boolean
algebras, Math. Scand. 5,205-217.
J. Hashimoto [1951]. On direct product decomposition of partially ordered sets, Ann.
Math. 54, 315-318.
L. Henkin, D. Monk, and A. Tarski [1971]. Cylindric algebras, Part I , North-Holland,
Amsterdam-New York-Oxford.
[1985]. Cylindric algebras, Part II, North-Holland, Amsterdam-New York-
Oxford.
C. Herrmann [1978]. Affine algebras in congruence modular varieties, Acta Sci. Math.
(Szeged) 41,119-125.
[1983]. On the word problem for the modular lattice with four free generators,
Math. Ann. 265, 513-527.
(See also. J. Hagemann and C. Herrmann)
Bibliography 339
D. Hobby (See R. McKenzie and D. Hobby)

A. Hulanicki and S. ~wierczkowski[1960]. Number of algebras with a given set of
elements, Bull. Acad. Sci. Polon., Ser. Sci., Math., Phys. et Astron. 8, 283-284.
N. Jacobson [1945]. Structure theory for algebraic algebras of bounded degree, Ann.
Math. 46,695-707.
[1980]. Basic algebra II, W. H. Freeman, San Francisco.
[1985]. Basic algebra I, Second Edition, W. H. Freeman, New York.
Ju. 1.Janov and A. A. Mucnik [1959]. Existence of k-valued closed classes without a finite
basis (Russian), Doklady Alcad. Naulc. SSSR 127,44-46.
B. Jónsson [1953]. On the representation of lattices, Math. Scand. 1, 193-206.
[1957]. On direct decompositions of torsion-free Abelian groups, Math. Scand.
5,230-235.
[1959]. Lattice-theoretic approach to projective and affine geometry, The axio-
matic method with special reference to geometry and physics (Henkin, Suppes, and
Tarski, Eds.), North-Holland, Amsterdam, pp. 188-203.
[1966]. The unique factorization problem for finite relational structures, Colloq.
Math. 14, 1-32.
[1967]. Algebras whose congruence lattices are distributive, Math. Scand. 21,
110-121.
[1972]. Topics in universal algebra, Lecture Notes in Mathematics, Vol. 250,
Springer-Verlag. New York.
[1982]. Varieties of relation algebras, Algebra universalis 15,273-298.
(See also C. C. Chang, B. Jónsson, and A. Tarski)
B. Jónsson and A. Tarski [1947]. Direct decompositions of 'finite algebraic systems. Notre
Dame Mathematical Lectures, no. 5. South Bend, IN.
H. J. Keisler (See C. C. Chang and H. J. Keisler)

J. Ketonen [1978]. The structure of countable Boolean algebras, Ann. Math. 108,41-89.
P. Kohler (See W. Blok, P. Kohler, and D. Pigozzi)
L. Kronecker [1870]. Auseinandersetzung einiger Eigenschaften der Klassenzahl idealer
komplexer Zahlen, Monatsh. Preuss. Alcad. Wiss., Berlin, 881-889.
W. Krull [1925]. Verallgemeinerte Abelsche Gruppen, Math. 2. 23, 161- 196.
A. G. Kurosh [1935]. Durchschnittsdarstellungen mit irreduziblen Komponenten in
Ringen und in sogenannten Dualgruppen, Mat. Sb. 42,613-616. .
F. W. Lawvere [1963]. Algebraic theories, algebraic categories, and algebraic functors,

The theory of models, Proc. 1963 int. symp. at Berkeley (Addison, Henkin, and Tarski,
Eds.), North-Holland, Amsterdam, pp. 413-41 8.
J. LOS[1950]. Un théorkme sur les superpositions des fonctions définies dans les ensembles
l
arbitraires, Fund. Math. 37, 84-86.
l [1955]. Quelques remarques théoremes et problemes sur les classes définissables
d'algkbres, Mathematical interpretation of formal systems. North-Holland, Amster-
dam, pp. 98-113.
L. Lovász [1967]. Operations with structures, Acta Math. Acad. Sci. Hungar. 18,321-328.
[1971]. On the cancellation law among finite relational structures, Period. Math.
I Hungar. 1, no. 2, 145-156.
I (See also R. R. Appleson and L. Lovász)
i A. 1. Maltsev [1936]. Untersuchungen aus dem Gebiete der mathematische Logik, Mat.
Sb. 1,323-336.
[1954]. On the general theory of algebraic systems (Russian), Mat. Sb. (N.S.) 35
(77), 3-20.
[1970]. The metamathematics of algebraic systems, collected papers 1936-1 967
(Translated and edited by B. F. Wells 111),North-Holland, Amsterdam.
[1973]. Algebraic systems. Grundlehren der mathematischen Wiisenschaften,
340 Bibliography
Vol. 192, Springer-Verlag, New York.

W. D. Maurer and G. L. Rhodes [1965]. A property of finite simple nonabelian groups,
Proc. Amer. Math. Soc. 16, 552-554.
R. McKenzie [1968]. On finite groupoids and K-prime algebras, Trans. Amer. Math. Soc.
133,115-129.
[1971]. Cardinal multiplication of structures with a reflexive relation, Fund. Math.
70,59-101.
[1972]. A method for obtaining refinement theorems, with an application to direct
products of semigroups, Algebra Universalis 2, 324-338.
(See also R. Freese and R. McKenzie)
R. McKenzie and D. Hobby [forthcoming]. The structure of .finite algebras (tame con-
gruence theory). Contemporary Mathematics, American Mathematical Society,
Providence, RI.
K. Menger [1936]. New foundations of projective and affine geometry, Ann. Math. (2) 37,
456-482.
D. Monk (See L. Henkin, D. Monk, and A. Tarski)
A. A. Mucnik (See Ju 1. Janov and A. A. Mucnik)
L. Nachbin [1947]. Une propriété charactéristique des algebres booléiennes, Portugal.

Math. 6, 115-118.
T. Nakayama (See N. Funayama and T. Nakayama)
J. B. Nation [1982]. Finite sublattices of a free lattice, Trans. Amer. Math. Soc. 269,
311-337.
A. Ju. Ol'shanskil [1980]. An infinite group with subgroups of prime orders (Russian),
Izv. Akad. Nauk SSSR Ser. Mat. 44, No. 2,309-321,479.
O. Ore [1935]. On the foundations of abstract algebra, 1, Ann. Math. 36,406-437.
[1936]. On the foundations of abstract algebra, 11, Ann. Math. 37,265-292.
P. P. Pálfy and P. Pudlák [1980]. Congruence lattices of finite algebras and intervals in
subgroup lattices of finite groups, Algebra Universalis 11,22-27.
C. S. Peirce [1880]. On the algebra of logic, Amer. J. Math. 3, 15-57.
R. S. Pierce [1968]. Introduction to the theory of abstract algebra, Holt, ~ i n e h a r tand
Winston, New York.
[1982]. Associative algebras, Graduate Texts in Mathematics, Vol. 88, Springer-
Verlag, New York.
D. Pigozzi [1972]. On some operations on classes of algebras, Algebra Universalis 2,
346-353.
(See also W. Blok and D. Pigozzi; W. Blok, P. Kohler, and D. Pigozzi)
A. F. Pixley [1963]. Distributivity and permutability of congruences in equational classes
of algebras, Proc. Amer. Math. Soc. 14, 105-109.
E. L. Post [1941]. The two-valued iterative systems of mathematical logic. Annals of Math.
Studies, no. 5. Princeton University Press, N.J.
P. Pudlák [1976]. A new proof of the congruence lattice characterization theorem, Algebra
Universalis 6,269-275.
(See also P. P. Pálfy and P. Pudlák)
P. Pudlák and J. Ttima [1980]. Every finite lattice can be embedded into a finite partition
lattice, Algebra Universalis 10, 74-95.
A. Pultr and V. Trnková [1980]. Combinatorial, algebraic and topological representations
of categories, North-Holland Publ. Co., Amsterdam.
R. W. Quackenbush [1971]. On the composition of idempotent functions, Algebra Uni-

versalis 1, 7-12.
R. Remak [1911]. Über die Zerlegung der endliche Gruppen in direkte unzerlegbare
Faktoren, J. reine angew. Math. 139, 293-308.
Bibliography
G. L. Rhodes (See W. D. Maurer and G. L. Rhodes)
H. P. Sankappanavar (See S. Burris and H. P. Sankappanavar)

E. T. Schmidt (See G. Gratzer and E. T. Schmidt)
E. Schroder [1890]. Vorlesungen über die Algebra der Logik (exakte Logik), edited in part
by E. Müller and B. G. Teubner, Leipzig. Published in 4 volumes. Second edition in 3
volumes, Chelsea Publ. Co., Bronx, New York, 1966.
R. Seifert, Jr. [1971]. On prime binary relational structures, Fund. Math. 70, 187-203.
S. Shelah [1980]. On a problem of Kurosh, Jónsson groups, and applications, Word
problems II (Adian, Boone, and Higman, Eds.), North-Holland, Amsterdam, pp. 373-
394.
W. Sierpinski [1934]. Sur les suites infinies de fonctions définies dans les ensembles
quelconques, Fund. Math. 24, 209-212.
[1945]. Sur les fonctions de plusieurs variables, Fund. Math. 33, 169-173.
J. D. H. Smith [1976]. Mal'cev varieties, Lecture Notes in Mathematics, Vol. 554,
Springer-Verlag, Berlin-New York.
M. H. Stone [1936]. The theory of representations for Boolean algebras, Trans. Amer.
Math. Soc. 40, 37-111.
S. ~wierczkowski(See A. Hulanicki and S. ~wierczkowski)
A. Tarski [1930]. Une contribution a la théorie de la mesure, Fund. Math. 15,42-50.

[1931]. Sur les ensembles définissables de nombres réels 1, Fund. Math. 17,
210-239.
[1935]. Der Wahrheitsbegriff in den formalisierten Sprachen, Studia Philosophica
1,261-405.
[1952]. Some notions and methods on the borderline of algebra and meta-
mathematics, Proceedings of the International Congress of Mathematicians, Cam-
bridge, Massachussetts, 1950, Amer. Math. Soc., Providence, R.I., pp. 705-720.
(See also C. C. Chang, B. Jónsson and A. Tarski; J. M. G. Fe11 and A. Tarski;
L. Henkin, D. Monk, and A. Tarski; B. Jónsson and A. Tarski)
A. Tarski and S. Givant [1987]. A formalization of set theory without variables, Collo-
quium Publications, Vol. 41, Amer. Math. Soc., Providence, R.I.
V. Trnková (See A. Pultr and V. Trnková)
J. Tfima (See P. Pudlák and J. TUma)
R. L. Vaught [1985]. Set theory, Birkhauser, Boston.

B. L. van der Waerden [1931]. Moderne algebra, Julius Springer, Berlin.
E. A. Walker [1956]. Cancellation in direct sums of groups, Proc. Amer. Math. Soc. 7,
898-902.
D. L. Webb [1936]. Definition of Post's generalized negative and maximum in terms of
one binary operation, Amer. J. Math. 58, 193-194.
J. H. M. Wedderburn [1905]. A theorem on finite algebras, Trans. Amer. Math. Soc. 6,
349-352.
[1909]. On the direct product in the theory of finite groups, Ann. Math. 10, 173-
176.
T. P. Whaley [1969]. Algebras satisfying the descending chain condition for subalgebras,
Pacific J. Math. 28,217-223.
A. N. Whitehead [1898]. A treatise on universal algebra, Cambridge University Press, U.K.
P. M. Whitman [1941]. Free lattices, Ann. Math. 42 (2), 325-330.
[1942] Free lattices, 11, Ann. Math. 43 (2), 104-115.
[1946]. Lattices, equivalence relations, and subgroups, Bull. Amer. Math. Soc. 52,
507-522.
R. Wille [1977]. Eine Charakterisierung endlicher, ordnungspolynomvollstiindiger Ver-
bande, Arch. Math. (Basel) 28, 557-560.
Table of Notation
Set-Theoretical Notation
Our elementary set-theoretical notation is standard. It is described fully in the Preliminaries
and only the less commonly encountered features are recalled here.
NOTATION PAGES DESCRIPTION

7 the set (0,1,2,. . .) of natural numbers.
7 the set of integers.
7 the set of rational numbers.
7 the set of real numbers.
7 the set of complex numbers.
5-6 sequence or function-builder notation
5 the sequence (xo, x,, . ) where the length
is specified by the context.
f : ~ - B ,A L B f is a function from A into B.
f:A++B f is a function from A onto B.
f:AHB f is a one-to-one function from A into B.
L f(a) the value of f at a.
ker f the kernel of f.
awh designates f by assignment-e.g., k w k2
designates the squaring function.
the ith projection function.
the ith n-ary projection function.
the direct product of the system A =
( A i :i E 1 ) .
BA the set of al1 functions from A into B.
End A the set of functions from A into A.
End A the monoid of functions from A into A.
Aut A, Sym A the set of permutations of A.
Aut A, Syrn A the group of permutations of A.
Part A the set of partial functions on A.
Part A the monoid of partial functions on A.
Table of Notation 343

Bin A 112-113 the set of binary relations on A.
Bin A 112-113 the monoid of binary relations on A.
Eqv A 8 the set of equivalence relations on A.
Eqv A 34 the lattice of equivalence relations on A.
aoP 6 the relational product (composition) of a
and B.
the n-fold relational product of a with B.
the converse of the relation a.
the factor set of A modulo the equivalence
relation a.
ala the equivalence class of a modulo a.
apb, a E, b, a b (mod p) a and b are related by p.
0, {(a, a) : a E A)-i.e., the identity relation on
A. This is the smallest equivalence rela-
tion on A.
((a, b): a, b E A)-i.e., the universal rela-
tion on A. This is the largest equivalence
relation on A.
Card the class of cardinals.
General Algebraic Notation

Generally, we have used boldface to indicate algebras, or sets equipped with additional
structure. For example, Aut A is the set of al1 permutations of the set A, Aut A is the group
of al1 permutations of A, and Aut A is the set of al1 automorphisms of the algebra A.

A = (A, F) the algebra A with universe A and system
= (A, Fi(iE 1)) F = (F,: i E I) of basic operations.
(AY r) a nonindexed algebra.
QA the interpretation of the operation symbol
Q in the algebra A.
AcB A is a subalgebra of B.
SgA(x) the subuniverse of A generated by X.
Sub A the set of subuniverses of A.
Sub A the lattice of subuniverses of A.
S(A) a special subuniverse of A4.
C(A) the center (subuniverse) af A, where A is an
algebra with zero.
~ : A + BA, ~ B h is a homomorphism from A into B.
~:AHBA , AB h embeds A into B.
h:~-B, ALB h is a homomorphism from A onto B.
h:A-By h : A g B h is an isomorphism from A onto B.
hom(A, B) the set of homomorphisms from A into B.
End A the set of endomorphisms of A.
Table of Notation

End A 21 the monoid of endomorphisms of A.
Aut A 21 the set of automorphisms of A.
Aut A 21 the group of automorphisms of A.
h(DY A) 319-320 the number of homomorphisms from D
into A.
the number of embeddings (= monomor-
phisms) from D into A.
the direct product of the system A =
(Ai: i E I ) of algebras.
the quotient algebra of A modulo the con-
gruence relation O.
Y 14 the quotient of two congruence relations.
B0 {x E A :xOy for some y E B).
01, the restriction of O to B.
Con A the set of congruence relations on A.
Con A the lattice of congruence relations on A.
CgA(x) the congruence relation on A generated
by X.
the (principal) congruence relation on A
generated by {(a, b ) ) .
a centralizes p modulo Y.
the commutator of the congruences a
and p.
the centralizer of P modulo a.
the center (congruence) of A.
a special congruence on A2.
the composition of a k-ary operation g with
k n-ary operations fo, . fk-l.
e ,
Clo A the clone of term operations of A.

Clo, A the clone of n-ary term operations of A.
Clo, A the algebra of n-ary term operations of A.
Po1 A the clone of polynomial operations of A.
Pol, A the clone of n-ary polynomial operations
of A.
A set of restrictions to X of polynomial
operations of A.
the algebra induced by A on X.
the product of a system r = (y,: i~ 1 ) of
congruences.
the product of the congruences a and P.
the family of al1 independent sets of con-
gruences on A.
A-B the algebras A and B are isotopic.
A-,B the algebras A and B are isotopic over the
algebra C.
A and B are modular-isotopic over C.
A and B are modular-isotopic in one step.
Table of Notation

A and B are modular-isotopic.
A and B are isotopic over a finite C.
A and B are isotopic over a member of X .
the monoid of isomorphism classes of alge-
bras in X under direct product.
the semiring of isomorphism classes of rela-
tional structures in X under direct pro-
duct and disjoint union.
-
unary operations derived from binary de-
composition operations.
sL,IR,5 , and special relations derived from a binary rela-
tional structure.
the class of al1 isomorphic images of mem-
bers of X .
the class of al1 homomorphic images of
members of X .
the class of al1 isomorphic images of sub-
algebras of members of X .
the class of al1 isomorphic images of direct
products of systems of algebras from X .
the class of al1 isomorphic images of direct
products of finite systems of algebras
from X .
the class of al1 isomorphic images of sub-
direct products of systems of algebras
from X .
the variety generated by X .
the class of al1 subdirectly irreducible mem-
bers of X .
the class of al1 finite algebras belonging
to X .
Spec X the spectrum of X
fx(n> the free spectrum function of X
@(X) for a given algebra A, the meet of al1 the
kernels of homomorphisms from A into
algebras belonging to X .
the set of terms of type o over X.
the term algebra of type o over X.
a free algebra in V(X)with free generating
set X.
an algebra F',(x) where 1x1 = rc.
the interpretation of the term p in the alge-
bra A.
the sequence ü satisfies the equation p Ñ q
in the algebra A.
the equation p Ñ q is true in A.
the equation p Ñ q is true in every algebra
in X .
346 Table of Notation

x+x 234 every equation in the set C is true in every
algebra in X.
O(X) 236 the equational theory of X .
Mod (C) 236 the class of al1 models of X.
W(X) 239 the set of al1 words over X.
O 113 the empty word.
AD 245 the application of the interpretation opera-
tor to the algebra A.
y- w 246 Y and W are equivalent varieties.
A r B 246 A and B are (term) equivalent algebras.
A,, A,,,, and A, 104-106 Certain mono-unary algebras.
SETS 134 The category of sets.
homc(A, B) 133 The collection of morphisms in the cate-
gory C from object A to object B.
Akp 134 The category of al1 algebras of type p.
CAT Y 135 The category of al1 algebras belonging to
the variety V .
136 The clone of an object A in a category C.
Notation for Lattices and Ordered Sets

NOTATION PACES DESCRIPTION
the lattice operations of meet and join.
the join (least upper bound) of the set X =
{xi:i E I ) of elements of an ordered set.
inf X, A X , A, xi the meet (greater lower bound) of the set
X = {x,:i E I ) of elements of an ordered
set.
a is covered by b (in an ordered set).
the dual of the lattice L.
the interval from a to b in an ordered set.
the sublattice with universe I [a, b].
the principal ideal determined by an ele-
menta. .
the principal filter determined by an ele-
ment a.
the ordered set of nonzero join irreducible
elements of the lattice L.
the ordered set of al1 proper prime ideals
of L.
Iso (L, L') the lattice of al1 isotone maps from the lat-
tice L into the lattice L'.
Ord (J) the lattice 'of order ideals of the ordered
set J.
Id1 L the lattice of ideals of L.
Fil L the lattice of filters of L.
Cvx L the lattice of convex subsets of L.
Table of Notation

IND(L) 66 the family of al1 directly join independent
subsets of L.
a@b the direct join of a and b.
40 and $, the perspectivity maps.
I [a, b] I [a', b'] I [a, b] transposes up to I [a', b'].
I [a, b] L I [a', b'] I [a, b] transposes down to Ira', b'].
I [a, b] r " I [a', b'] I [a, b] transposes weakly up into I [a', b'].
I [a, b] L , I [a', b'] I [a, b] transposes weakly down into
I [a', b'].
h (a) the height of an element a of a lattice.
d (a) the dimension of an element a of a lattice
with dimension function d.
the five element nondistributive modular
lattice.
the five element nonmodular lattice.
the free distributive lattice on three free
generators.
the free modular lattice on three free gener-
ators.
7I = (P, A), L" a projective plane and its associated lattice
of subspaces.
a projective geometry and its associated
lattice of subspaces.
dim r the dimension of a projective geometry T.
Bc(x) the set of cardinalities of finite bases of X
with respect to the closure operator C.
the set of cardinalities of finite bases of X
with respect to the closure operator sgA,
where X c A.
Notation for Groups and Rings

NOTATTON PAGES DESCRIPTION
No 118 the normal subgroup that is the coset of the
unit element of a group modulo the con-
gruence 8 on the group.
the commutator of elements x and y of a
group.
the stabilizer of t in K.
the infinite quasi-cyclic group.
the free group generated by the free generat-
ing set X.
the variety of n-nilpotent groups.
the (Burnside) variety of groups of expo-
nent n.
the annihilator of the ring R; the annihila-
tor of the element x of R.
Index of Names
Appleson, R. R., 325 Galois, E., 51-52

Artin, E., 16 Galvin, F., 109,264
Gerhard, J. A., 115
Baldwin, J. T., 127 Givant, S., 18
Bernstein, F., 333 Glivenko, V., 1
Birkhoff, G., 1-2, 36, 50, 58, 62, 79, 85, 87, Gratzer, G., 3,4, 36, 181, 190
97, 126, 130, 168, 173, 183,221,236, Grassmann, R., 1
264,276 Gumm, H. P., 250
Blok, W., 4
Boole, G., 1 Hagemann, J., 4,250
Bruck, H., 124 Hajnal, A., 109
Burnside, W., 16,110, 122 Hall, P., 136
Burris, S., 4, 35 Halmos, P., 5
Hamilton, W. R., 1
Cantor, G., 333 Hanf, W. A., 267,331
Cayley, A., 119, 129 Hashimoto, J., 305
Chang, C. C., 109-110,262,305, 313, Hasse, H., 38
318 Hausdorff, F., 7
Cohn, P. M., 4 Henkin, L., 173
Corner, A. L. S., 261 Herrmann, C., 4,241,250,257
Crawley, P., 36 Heyting, A., 181
Higman, G., 125
Dedekind, R., 1, 53,55,57-58,62,259 Hilbert, D., 16, 52
DeMorgan, A., 1 Holder, O., 58, 62, 75
Descartes, R., 259 Hulanicki, A., 148
Dilworth, R. P., 3, 36, 84, 91, 95, 97, 318 Huntington, E. V., 318
Erdos, P., 109 Jónsson, B., 3, 18,66,70, 109-110, 190,206,248,

259,261,262, 263,264, 266, 268,276,
Fano, G., 208 278, 280,283, 290, 291, 295, 299, 305,
Feit, W., 102 318
Fell, J. M. G., 317, 318 Jacobson, N., 16,175
Fitting, H., 286 Janov, Ju. I., 148
Fleischer, I., 203 Johnson, O. J., 110
Foster, A. L., 3 Jordan, C., 15,52,58,62,75,281-282
Fraenkel, A., 18
Freese, R., 4, 77,241,250 Kohler, P., 4
Frink, O., 3, 183 Keisler, H. J., 110
Frobenius, G., 16 Ketonen, J., 116,261
Fuhrken, G., 132 Kirkman, T., 124
Funayama, N., 78 Kronecker, L., 66,259
Index of Names
Krull, W., 2, 66,259 Quackenbush, R. W., 103

Kummer, E., 259
Kurosh, A. G., 59,60,66 Remak, R., 259
Kwatinetz, M., 35 Rhodes, G. L., 158
Rips, E., 110
Lachlan, A., 127 Rowbottom, F., 109
Lampe, W., 102
Lawvere, F. W., 137 Sankappanavar, H. P., 4
Lie, S., 1, 15 Schmidt, E. T., 3,181, 190
Los, J., 3, 111,286 Schmidt, O., 2,66,259
Lovász, L., 260,262,270,319,321,322,325,331 Schreier, O., 75-76
Schroder, E., 1
MacNeille, H. M., 53 Scott, W. R., 266
Maltsev, A. I., 3,4,246,247,248 Seifert, R., 263
Marczewski, E., 3 Shelah, S., 110
Maurer, W. D., 158 Sierpinski, W., 101-102
McKenzie, R., 102, 103, 250, 261,263, 265, 305, Smith, J. D. H., 4, 250
310,313,331 Steiner, J., 123-124
Menger, K., 1,97 Stone, M. H., 3, 87, 88,89
Monk, D., 173,180 ~wierczkowski,S., 148
Mucnik, A. A., 148
Tarski, A., 3, 18, 50, 87, 125, 173, 187, 189, 261,
Nachbin, L., 89 262, 263,283, 290, 291,295,299, 305,
Nakayama, T., 78 317,318
Nation, J. B., 4 Trnková, V., 137
Neumann, B. H., 125 TUma, J., 4, 193
Noether, E., 16
Vaught, R. L., 5,267,331
Ol'shanskiI, A. Ju., 110 van der Waerden, B. L., 1-2,16
Ore, O., 1,2, 59, 60, 62, 66, 75, 264,276,280 von Neumann, J., 1,3
Pálfy, P. P., 156, 158 Walker, E. A., 266

Pasch, M., 217 Webb, D. L., 103
Peano, G., 1 Wedderburn, J. H. M., 16,66,175,259
Peirce, C. S., 1 Whaley, T. P., 110
Pierce, R. S., 4, 16 Whitehead, A. N., 1
Pigozzi, D., 4, 225 Whitman, P. M., 3, 190, 242
Pixley, A. F., 247, 248 Wille, R., 159
Post, E. L., 2, 147
Priestley, H., 89 Zassenhaus, H., 75-76
Pudlák, P., 4, 156, 158, 181, 193 , Zermelo, E., 18
Pultr, A., 137 Zorn, M., 7
Index of Terms
Abelian algebra, 250 finite, 142

absorption, under direct product, 331 finitely generated, 26
absorption laws, in lattice theory, 36 of finite type, 142
abstract class of algebras, 219 free (see free algebras)
abstract clone, 136 homomorphic images of, 21
addition, of cardinals, 8 homomorphisms of (see homomorphisms)
afine algebraic variety, 52 idempotent, 261
aleph, 8 index set of, 12
algebraic: induced by A on X, 156
closure operator, 46 infinitely divisible, 331
function, 145 irreducible:
, geometry, 52, 84 (U,b) -, 169
property, 21 subdirectly, 169
algebraic lattices: isomorphic, 21
compact elements in, 46 isotopy of, 270-273,322
complemented modular, $4.8 Jónsson, 109
covering relation in, 182 locally finite, 221
defined, 46,181 Maltsev, 247
join continuity of [Exercise 41, 194 monolith of, 169
representation of, $4.6 nonindexed, 154
retracts of [Exercise 41, 194 nontrivial:
(see ulso closure operators (algebraic), an algebra with more than one element
Gratzer-Schmidt Theorem, lattices of partial, 143
(congruence~,ideals, etc.)) perfect, 317
algebra(s): polynomially equivalent, 246
Abelian, 250,252,257 polynomial operations of, 145
arithmetical, 247 primal, 148
basic (fundamental) operations of, 12, 142 prime in a class, 103,263 [Exercise 101, 300
cancellable in a class, 321 pseudo-simple, 180
cardinality of, 142 quotient of, 28
centers of, 250,296 rank function of, 12
congruence distributive, 196,218, 304 reduct of, 13
congruence modular, 196,276 residually finite, 178
congruence permutable, 94.7 residually less than rc, 178
congruences of, 28 retraction of, 194
defined, 12,142 similarity of, 13
directly indecomposable, 95, 165,260 semi-simple, 179,204
division between, 103 simple, 95, 157
(term) equivalent, 246 solvable [Exercise 61,258
expansion of, 13 subalgebra of, 19
extension of, 19 subdirectly irreducible, 169, 221
Index of Terms
(a, b)-irreducible, 169 bounded lattice, 38

monolith, 169 Burnside varieties (of groups), 122
subuniverses of, 19
term operations of, 143 cancellation, for direct products, 262,291, 319
that are zero divisors, 324 (see also isotopic (algebras))
(similarity) type of, 13, 149 Cantor-Bernstein Theorem, 333
unary (see unary algebras) cardinality, 8
universe of, 12 cardinals, 8
weak isomorphism of, 246 Cartesian (or direct) products, 6,22
with zero, 282, Gg5.4-5.5 categories:
Algebra Universalis, a journal, 4 clone, a certain kind of category, 136
alphabet, 229 clone-map, 140
alternating groups: clone of a variety, 141,245
defined, 131 subclone, 136
infinite [Exercise 11],158 concrete, 135
simplicity of [Exercises 6-10], 132 coproducts in, 140
annihilator, in rings, 52,251 defined, 133
anti-: domain (codomain) of an element, 133
chain, 38 equivalent, 135- 136
homomorphism, 131 functors between, 133
symmetric element, 305, 313 equivalence functor, 135- 136
Arguesian equation, for lattices, 193 faithful functor, 134
ari ty (rank): isomorphism, 134
of a closure operator, 183 universe functor, 135
of an operation, 11 groupoid, 139
of a relation, 5 isomorphic, 134
associative algebras, over a fíeld, 15, 245 morphisms, or elements, 133
associative law, 112 domains (codomains) of, 133
atom, of an ordered set, 38 epimorphisms, 135
atomistic lattice, 210 equalizer of morphisms, 203
Aut A, the automorphism group of an algebra, hom,(A, B), 133
21,35 isomorphisms, 135
automorphism, of an algebra, 21 monomorphisms, 135
axiomatized by, 236 objects (unit elements), '133
Axiom of Choice, 6-7,27,43 opposite category, 140
Axiom of Constructibility, 109 products in, 135
axioms of lattice theory, 36 representation problems, 137
small, 134
band, or idempotent semigroup, 115 subcategories, 133
base (see generating set) Aut, A, 136
base set (universe) of an algebra, 12 Clone, A, 136
basic (fundamental) operation, 12, 142 End, A, 136
basis (see generating set) full subcategories, 133
bilinear algebras: unit elements, 133
associative algebras, 15,245 category of:
Lie algebras, 15,245 algebras of type p : Alg,, 134-135
Bin A, the monoid of binary relations on A, 113, algebras in Y :CAT Y ,135
119 distributive Iattices, 139
Birkhoff-Ore Theorem [Theorem 531,276 groups, 140
block, Le., equivalence class, 8 sets, 134
Boolean: structures over C, 322
algebras, 17, 331 topological s-paces, 141
free, 241 Cayley represen tation:
representation of, 85-88 of a group or semigroup, 129
Stone duality for (see Stone duality) Cayley's Theorem, 129, 137
groups, 268 center:
lattices, 84, 97 of an algebra with zero, 295
powers, 261 congruence of an algebra, 250,297
products, 4 of a ring, 176
rings, 245 central element:
sum (symmetric difference), 245 of an algebra with zero, 296
352 Index of Terms
central element (continued) in groups, 121-122

of a ring, 176 commutator theory, 4,196,250
centralizer: commuting:
of a congruence, 252 functions, 130
of a subset of a clone, 163, [Exercise 131 149 operations, 149, 163 [Exercise 121
centralizes: relations, i.e., permuting binary relations,
a centralizes p modulo y, 252 196
central subuniverse of an algebra with zero, compact:
295-296 closed set, i.e., finitely generated [Theorem
chain (see lattices, ordered sets) 2.16],46
chain conditions (see lattices) element of a lattice, 46
Chinese Remainder Theorem, 248 comparable elements of an ordered set, 38
chain of sets, [Exercise 31 27, [Exercise 101 250 complement of an element in a lattice, 60
choice functions, 22 complementary factor relations and con-
class, in set theory, 5 gruences, 161
class operators: complementation operation, 17
1,H, S, P, P,, Pf,,, 23,219,222 complete lattices:
order monoid of [Exercise 121,225-226 compact elements, 46
Clo, A, the algebra of n-ary term operations connection to closure opeqators, 42.2
of A: continuous lattices [Exercise 41, 194
defined [Exercise 11,148 infimum (greatest lower bound), 37
induced homomorphisms of [Exercise 61,153 supremum (least upper bound), 37
quotients of [Lemma 4.98],222 (see also closure operators, algebraic lattices,
is a relatively free algebra [Exercises 7 and 81, Dedekind-MacNeille completion)
242 components (see connected components)
clones (certain categories) (see categories) composition:
clones (sets of operations closed under of binary relations, 6
composition): of functions, 6
of al1 operations on A: of morphisms in a category, 133
Clo A, Clo, A, 143 of operations, 142
centralizers of [Exercise 131, 149 (see also relational product)
lattices of, 4, 147 Con A, the lattice of al1 congruences of A, 34
of polynomial operations of A: concatenation of sequences, 113
Po1 A, Pol, A, 144-145 congruence distributive and congruence
of term operations of A: modular (see algebras, varieties)
Clo A, Clo, A, 143 congruence Generation Theorem [Theorem
clopen set, 17 4.191, 155
closed: congruence lattice of an algebra, i.e., the lattice
under a class operator, 23,220 of al1 congruences, 34
under composition by f, 143 congruence permutable variety (see Maltsev
under an operation, 52, [Exercise 71 148 variety)
closed set system, 44 congruence relations, i.e. congruences
closure operator(s): congruence(s):
algebraic, 46 centralizers (annihilators) of, 252
connection to algebraic lattices classes (blocks or cosets) of, 28
[Theorem 2.16],46 commutators of, 252
representation as SgA[Theorem 4.511, 182 complementary factor, 161
base of a subset for, 187 defined, 28
basis (or irredundant base), 187 factor, 161, 302
closed set system, related, 45 factorable [Exercise 91, 166
connection to complete lattices, 42.2 finitely generated, 33
defined, 44-45 generated by a set of ordered pairs, 32, 152,
finitely generated closed set for, 46 155
independent subset for [Exercise 6],50 indecomposable, 269
lattice of closed sets of, 45 isotopic, 271
+
n-ary (of rank n l), 183 joins (meets) of, 33-34
(see also Galois connections, Tarski's Inter- n-permuting, 196
polation Theorem) permuting, 195,s4.7
commutators: principal, 33
in general algebras, 252 (inner direct) products of, 268
Index of Terms
strictly meet irreducible [Lemma 4.431, 171 irreducible element, 66-67

substitution property of, 27-28 isotopic elements, 66-67
type n - 1 joins of, 198 direct products:
connected: decomposition operations of:
components: binary, 115, 162
of a graph, 328-329 [applied], 302-3 16
of a relational structure, 305 n-ary [Exercises 14-15], 167
of a unary algebra, 104, 107, [Exercise defined, 22
1-31 110 factor congruences of, 161,302
ordered set, 305 factors of, 22
relational structure, 305 projections to the factors, 22
unary algebra, 107, [Exercises 1-31 110 refinement properties for:
constants, 11 intermediate refinement property, 311
Continuum Hypothesis (CH), 317 refinement property, 261,300
convex: strict refinement property, 165, 301
(k-convex)set of natural numbers, 187 unique factorization property for, 116, 165,
subset of an ordered set, 16, 31,48 260, Ch. 5
coproduct, in categories, 140 (see also absorption, cancellation, division
Correspondence Theorem [Theorem 4.121, 151 between algebras, factorable con-
cosets (see congruence (classes of)) gruences, isotopic algebras, kth roots,
coset: prime algebras, zero divisors)
structure, 325 distributive:
of a subgroup, 108 element of a lattice, 96
cover (upper, lower), of an element, 38 lattices:
covering relation: defined, 78
in algebraic lattices, 182 free, 39, 241-242
in lattices, 38, 61, 62 representation of, 85-87
cyclic decomposition of a permutation, 107 Priestley duality for (see Priestley duality)
cylindric algebras, 173 law, 78
division:
Dedekind-MacNeille completion, 53 between algebras:
Dedekind's Transposition ~ r i n c i ~[Theorem
le A is:
2.271, 57 divisible by B, 103
defined (or axiomatized) by a set of equations, infinitely divisible by B, 33 1
236 groups:
diagonal subuniverse of A", 200 defined, 244
dimension: single equational axiom for, 125
function of a lattice, 65 rings, 157, 175,210
(or height) of a lattice, 65 domain and codomain:
of a projective geometry, 209 of a function, 6
direct: of a morphism (see categories)
decompositions of algebras, i.e., direct duality:
representations, 159 for lattices, 40
factor of an algebra, 22 of Priestley (see Priestley duality)
join of elements in a lattice, 66 for projective planes [Exercise 21,216
powers of algebras, 6,23 of Stone (see Stone duality)
weak [Exercise 71,317
products of algebras (see direct products) embeddings:
representations, 159 defined, 20,
sum (inner) of subalgebras, 283 meet, 192
directed: subdirect, 169
collection of sets [Exercise 31, 27 End A, the endomorphism monoid of an algebra,
subset of an ordered set, 46 21,35
Direct Join Decomposition Theorem End A, the monoid of selfmaps of A, 112, 119
[Theorem 2.471, 74, 279 equalizer of two homomorphisms, 203
directly: equations:
indecomposable algebra, 95, 165 defined, 234
indecomposable congruence, 269 synonyms: identity, law, 234
join (meet): true in A (or valid in A, or obeyed by A, or
independent subset, 66-67 holds in A), 234
354 Index of Terms
equational: free generators are called:

class, i.e., a variety, 236 letters, variables, indeterminates, 227,239
theory: freely generated by X, 227
base of, 189 reduced terms, i.e., normal forms for elements,
of a class, 236 239
defined, 236 relatively free algebras, 239
nonrecursive, 241 universal mapping property of, 227
equationally complete variety (see varieties (see also terms. For free groups, e.g., see
(minimal)) groups (free))
equivalence relations: function(s):
complementary factor relations, 161 and allied concepts, 6, $3.2
defined, 8 choice, 22
equivalence classes (blocks, cosets), 8 commuting, 130
factor set of A modulo, 8 continuous, 21,23, 141
joins (meets) of, 17, 33-34 fixed point of [Exercise 5],50
partitions, correlated to, 8 idempotent, 287
relational product of, 6, 160 order-preserving, 40,85
(see also congruences, lattices of (equivalence retraction [Exercise 21, 131
relations), relations (binary)) synonyms: map, mapping, system
equivalent: (see also unary algebras)
algebras, 246 functor (see categories)
categories, 135-136 fundamental (basic) operations, 12,142
varieties, 244-245 Fundamental Theorem of Finitely Generated
Eqv A, the lattice of al1 equivalence relations Abelian Groups [mentioned], 52
on A, 34, [Exercise 71 18 (see also lattices
of (equivalencerelations)) Galois connections (correspondences):
expansion, of an algebra, 13 established by polarities of a binary relation,
exponentiation, of cardinals, 8 51
extension, of an algebra, 19 examples of, 51-53,147
Galois field, i.e., a finite field
factor: Generalized Continuum Hypothesis (GCH), 317
congruences of an algebra, 161,302 generating set:
direct, of an algebra, 22 i.e., base, with respect to a closure operator,
of a factorable relation, 164 187
relations: basis, i.e., an irredundant (independent or
of a relational structure, 307 minimal) base, 187, [Exercise 61 50
of a set, 161 of a congruence, 32,152,155
subdirect, of an algebra, 159 the exchange property for bases, [Exercise 61
factorable: 50
congruences, 164, [Exercise 91 166 free, of an algebra, 227
relations, 164 of a subuniverse, 25,152
factorization, unique, 259 geometry (see algebraic, projective)
family, or set, 5 graphs:
Fano plane, 208 bipartite, 331
fields, 14, 175,208-209 directed, 328
filter, of a lattice, 48 weak direct product of, 330-33 1
finitary, 11 Gratzer-Schmidt Theorem [discussed], 3, 181,
finitely generated: 190
closed set for a closure operator, 46 greatest lower bound, 37,44
congruence, 33 groupoid, an algebra with one binary operation,
subuniverse, 26 103,186
variety, 221 groupoid, a category with invertible maps,
fixed point, of an isotone map [Exercise 51, [Exercise 101 139
50 group(s):
formal equality symbol (Ñ), 9 alternating (see alternating groups)
used with equations, 234 commutators in, 121,250
free algebras: congruence lattices of, [Theorem 2.241 53
absolutely free algebras, 228,239 conjugate subgroups of, [Exercise 91 111
the term algebra of type g, 229 defined, 14,118
free algebras in familiar classes, 239-242 directly indecomposable, 132,168,260
free for (or in) X over X, 227 finitely generated Abelian, 259
Index of Terms
free: subdirect embeddings, 169

compared, [Exercise 201 243 surjective (onto) homomorphisms, 21
constructed, 119-120, [Exercise 41 126 Homomorphism Theorem [Theorem 1.161,
revisited, [Examples 2-32 240 28-29
inner automorphisms of, 132 H-spaces, 261
nilpotent, 122 HSP Theorem (of G. Birkhoff)[Theorem 4.1311,
normal subgroups of, 30,36,118-119 237
perfect [Exercise 111, 317
permutation, 130 ideals:
quasi-cyclic: of a lattice, 48
as Jónsson groups, 110 order ideals, 85
as pseudo-simple algebras, [Exercise 31 180 prime ideals, 86
as subdirectly irreducible Abelian groups, principal ideals, 48
[Example (e)] 171, [Exercises 2-71 179- of a ring, 30, 175
180 idempotent:
representations of, 128-130 (see also G-sets) algebra, 261
simple, [Exercise 141 158- 159 element of a ring, 176
subgroups of, 118 function, 287
symmetric (see symmetric groups) operation, 103
torsion-free, 266 semigroup, 115
words (reduced), 119 identities (see equations)
(see also Boolean groups, Burnside variety, image, of a set or relation under a mapping,
division groups, multiplication groups) 6,152
G-sets: implication algebras, 205
congruence lattices of, 155- 156 incomparable elements of an ordered set, 38
defined, 108 independent (irredundant) set (see generating
quotients of, 108, 111 set)
orbits of, 108 index set, of an algebra, 12
stabilizer (or isotropy) subgroups of, 108,132 infimum, i.e., greatest lower bound, 37,44
transitive, 108 integers:
finite ordinals, 7
Hasse diagrams, 38, 56 modulo n, 30
Hausdorff Maximality Principle, 7 ring of, 30
height of: interpretation:
an element, 64 of an operation symbol in an algebra, 12
a lattice, 64 of varieties, 245, $4.12
Heyting algebras: interval(s):
an arithmetical variety [remark], 248 defined, 38
defined [Exercise 181, 181 perspectivity map between, 57
simple and subdirectly irreducible [Exercise projective, 57 (see also Dedekind's Trans-
18],181 position Principle, isotopy)
Higman-Neumann-Tarski and a single equa- projectivity map between, 57
tional axiom for group theory, 125 sublattice, 151
Hilbert's Nullstellensatz [mentioned], 52 transposed, 57
hom(A, B), 20 transposes down to an interval, 57
hom,(A, B), where A and B are objects transposes up to an interval, 56-57
in a category C, 133 transposes weakly down into an interval,
homomorphisms: 91
anti-, 131 transposes weakly into an interval, 91
automorphisms, 21 transposes weakly up into an interval, 91
defined, 20 weakly projective, 91
endomorphisms, 21,283 invariant, Le., preserved by an operation, 19
equalizer of two, 203 inverse image, of a set or relation under a
finitely determined, 168 mapping, 6, 152
hom(A, B), 20 irredundant (independent) set (see generating
injective (one-one) homomorphisms, i.e., set)
embeddings, 20 irreflexive structure, 325
isomorphisms, 21 isomorphic:
kernels of, 28 algebras, 21
retractions [Exercise 41, 194 categories, 134
separating points, 160 representations, 129, 160
Index of Terms
isomorphism, 21 complemented, 60, 84

Isomorphism Theorems, 58,149 complete, 44
isomorphism type (or class) of algebras, 22 congruences of, 52.6
isotone map (or function) (see ordered sets) congruence lattices of [Theorem 2.501, 78
isotopic: continuous [Exercise 41, 194
algebras, 270 covering relation of, 38
relative to a class, 322 directly join (meet) independent subset of,
congruences, 271 66-67
elements of a lattice (i.e., directly join isotopic), directly join (meet) irreducible element of,
66-67 66-67
(see also isotopy) directly join isotopic elements of, 66-67
isotopy: distributive, 78
direct join (or meet) isotopy, 66-67 distributive element of, 96
join (or meet) isotopy map, 66-67 duality for:
modular isotopy, 273 dual of a lattice, 40
Isotopy Lemma (of Lovász), 322 dual of a property, 42
isotropy subgroup (see G-sets) principal of duality for lattices, 40
selfdual class (or property), 42
Jacobson (theorem of), 175, 177 filters of, 48
join: finite dimensional, 65
continuous lattice [Exercise 41, 194 of finite height, 64
homomorphism (see meet (homomorphism)) free, 241-242
irreducible element, 41 Hasse diagrams of, 38
strictly, 49 height of, 64
irredundant set, 60 height of an element of, 64
operation of a lattice, 17, 36 ideals of, 48
direct, 66 intervals of, 38
prime element, 41 interval sublattices of, 151
semilattice, 114 lower covering property for [Exercise 41,
join independence family, 68 61
Jónsson algebras, 109 lower semimodular [Exercise 4],61
Jónsson-Tarski Unique Factorization Theorem meet embedding of, 192
[Theorem 5.81, 290 modular, 53
Jónsson terms, 248 most significant, 44
Jordan-Holder Theorem [discussed], 58,62,66, neutral element of, 98, [Exercise 121 166
75 point, line:
Jordan normal form: elements of a lattice, 207
[established, in Exercises 2-81? 281-282 projective geometry:
[mentioned], 52 a lattice, 216
relatively complemented, 60, 205, [Exercises
kernel of a function, 6,28 12-13] 206
Krull-Schmidt Theorem [discussed], 58,66, representation:
264 of finite lattices, 4
kth roots, for direct products, 261, 321 of lattices, 192-193
Kurosh-Ore Theorem: sectionally of finite height, 64
[discussed], 66,84,203 semimodular [Exercise 4],61
[Theorem 2.331, 60 standard element of, 98
uniquely complemented, 318
Latin square, 123 upper covering property for [Exercise 4],61
lattice diagram, i.e., Hasse diagram of a lattice, 38 (see also algebraic lattices, complete lattices,
lattice order, 37 distributive lattices, intervals, join, lattices
lattice ordered set, 37 of, meet, modular lattices)
lat tice(s): lattice(s) of:
algebraic, 46 clones, 147
atomistic, 210 closed sets for a closure operator, 45
atom of (coatom of), 38 congruences, 33-34
Boolean, 84 of algebras of various types:
bounded, 38 [discussed], 54,79, 196
chain (or antichain) in, 38 [Exercise 191,167
chain conditions for: [Exercise 101, 195
ascending (or descending), 42 of finite algebras:
finite, 62 [Exercises 8-91? 158
lndex of Terms
convex subsets of a lattice, 48 Mobius:

equivalence relations, 17, 34 function of an ordered set, 334
generation of [Exercise 131, 158 inversion formula, 334
representations in, 192 Mod(C), the variety axiomatized by C, 236
simplicity of [Exercise 121, 158 modular lattice(s):
filters of a lattice, 48 complemented algebraic, 210
ideals of a lattice, 48 defined, 53
permuting equivalence relations, 196-197 dimension function of, 65
positive integers ordered by divisibility, 43 of finite height [Exercise 41,275
submodules [Exercise 171,217 free, 39, 241
subspaces of a projective geometry, 209 projective geometry (a modular lattice),
subuniverses, 27 216
varieties, 4 spanning n-frame in, [Exercise 151 217
defined, 16 modular law:
law (see equation) of lattices, 53
least upper bound, 37,44 of relational arithmetic, 198
Lie algebras (see bilinear algebras) module (see rings)
limitations of a single operation, 102 monadic algebras:
line: an arithmetical variety [remark], 248
element of height two, 207 defined, 173
projective line (a lattice), 210 a locally finite variety [Exercise 151,243
locally finite: a nonfinitely generated variety [Exercise 31,
algebras, 221 226
varieties, 221 simple and subdirectly irreducible [lemmas
logic and model theory, 3 and corollary], 173- 175 '
loops (see quasigroups) special, 173

Lovász Isotopy Lemma [Theorem 5.241, 322 monoids:
lower covering property, 61 of class operators, 225
defined, 13
M, (a lattice), 39,41,48, 59-60, 79-82, 84, 157, free, 113,239-240
243,254 indecomposable (prime) elements of, 262
Maltsev: of isomorphism classes (or types), 116,137,
algebra, 247 260,321
operation, 247 ordered [~xercise121,226
property, 3,4, 248 (see also M-sets)
term, 247 monolith, i.e., the minimal congruence of a
variety, i.e., a congruence permutable variety, subdirectly irreducible algebra, 169
247 -morphisms (see homomorphisms)
.
map or mapping, i.e., a function, 6 morphisms, elements of a category:
matrices: codomain of, 133
elementary Jordan, 28 1 domain of, 133
Jordan normal form of, 281-282 epimorphisms, 135
monoids of, 13 isomorphisms, 135
multi-cyclic, 28 1 monomorphisms, 135
nilpotent, 281 units, 133
rings of, 14,293-294 M-sets, 109
maximal: multiplication:
element of an ordered set, 7 of cardinals, 8, [Exercises 2-31 316-3 17,
chain, 7 [Exercise 151 318
measurable cardinals, 109 group, 118
meet: loop, 123
homomorphism, 190 quasigroup, 123
irreducible element, 41 nonpermuting congruences in [Exercises
strictly, 49, 171 8-92, 126
operation of a lattice, 17, 36
direct meet, 67 N, (a lattice), 39,41, 55, 56, 58, 79-80, 170,
prime element, 41 243-244
semilattice, 114 n-nilpotent groups, 122
minimal: n-permuting congruences, 196
element of an ordered set (see maximal) [n] th power of an algebra [Exercise 181,
generating set, i.e., a basis, 187 167
variety [Exercise 171, 243 n-tuples (elements of A"),5
Index of Terms
natural numbers: defined, 7

a distributive lattice under divisibility, 43, directed subsets of, 46
[Exercise 11 89 Hasse diagram of, 38
finite ordinals, 7 interval in, 38
neutral element of a lattice, 98, [Exercise 121 166 isotone (order preserving) functions between,
nonindexed algebras, 154 40
normal forms, for elements of a relatively free lattice ordered, 37
algebra, 239 (see also Jordan normal least (greatest) element of, 7
form) linearly ordered (i.e., a chain), 7
normal subgroups, 30,36, 118-119 maximal (minimal) elements of, 7
Mobius function of, 334
obeys the law, i.e., the equation is valid in, 234 tree (an ordered set), 106
objects, of a category, 133 upper (lower) bound of a subset in, 7
o (the set of natural numbers), 7 upper (lower) cover of an element in, 38
one-to-one (injective) map, 6 well-ordered, 7
onto (surjective)map, 6 (see also Hausdorff Maximality Principle,
operation(s): lattices, semilattices, Zorn's Lemma)
acting coordinatewise, 22, 163 ordinals, 7-8
arity of (i.e., rank), 11 Ore's Theorem [Corollary 2.481, 75,276
basic (fundamental), 12, 142
binary, 11, Ch. 3 Part A, the monoid of partial functions from A
commuting, 163, [Exercise 121 149 into A, 112,119
composition of, 142 partial order, i.e., an order relation, 7
constants, 11 partition:
decomposition operations, 115, 162,302,306 correlated to an equivalence relation, 8
[Exercise 141, 167 Pasch's axiom for projective geometry
defined, 11 [Exercise 71, 217
depends on a variable, 101,146 permutational algebra, 155
derived (i.e., term) operations, 101 permutation group, 130 (see also groups, G-sets)
essentially at most k-ary, 146 permutation(s):
finitary, 11 cyclic (a cycle), 107
fundamental (basic), 12, 142 cyclic decomposition of, 107
idempotent [Exercise 41, 103 defined, 14
Maltsev, 247 of finite support, 193
n-ary, 11 notation for, 107
partial, 11 product of, 107
polynomial, 144-145 support of, 193
preserves a relation, 19 transposition (a 2-cycle) [Exercise 21, 116
projection, 142 (see also groups, G-sets)
rank of, 11 permuting:
term, 100, 143,231 binary relations, 160
ternary, 11 congruences, $4.7
unary, 11, Ch. 3 perspectivity map between intervals, 57
(see also algebras, clones) Pixley term for an arithmetical variety, 248
operation symbols, 12, 149,229 point, i.e., an atom, 207
orbit: Pol, A, the algebra of n-ary polynomial
connected component of a G-set, 108 operations of A [Exercise 61,153
order (or ordering relation) over a set, 7 polarities, of a binary relation, 51
order: polynomial:
ideal, of an ordered set, 85 equivalence, 246
lattice-, 37 operation, 144-145
partial- (i.e., an order), 7 Post algebras [Exercise 111, 148-149
preserving map, 40 power set, 5
of a projective plane [Exercise 41,216 preservation [discussed], 52, 147
semilattice-, 114 Priestley duality, between distributive lattices
ordered monoid of class operators [Exercise 121, and certain ordered topological spaces
226 [mentioned], 89
ordered set(s): primal algebra, 148
atoms (or coatoms) of, 38 prime:
bounded, 38 algebra (direct products), 103,263, [Exercise
chain (or antichain) in, 38 lo] 300
connected, 305 element of a lattice, 41
Index of Terms
element of a monoid, 262 relational structure(s):

ideal (or filter) of a lattice, 86 antisymmetric element of, 313
Prime Ideal Theorem for Distributive Lattices binary, 305
[Theorem 2.60],87 connected, 305
principal: defined [Exercise 71, 166
congruence, 33 homomorphisms and products of, 306
filter, 48 irreflexive, 325
ideal, 48 reflexive, [defined] 305,3 11
principal ideal domain, 52, 260 reflexive element of, 311, 325
product: thin, 306
in categories, 135 relation(s):
direct (see direct products) binary, 5
relational, or relative, 6, 113, 160 anti-symmetric, 7
subdirect, 160 converse of, 6
projection: n-permuting, 196
function, 22 reflexive, 7
operation, 142 relational product of, 6, 160
projective: symmetric, 8
geometry: transitive, 7
a collection of points and lines, 209 transitive closure of, 6, 34
coordinatizable, 209,210 (see also equivalence relations, functions,
dimension of, 209 graphs, ordered sets)
a lattice, 216 factorable, 164
intervals, 57 factors of, 164
line, 210 n-ary, 5
plane, 207 preserved by (or closed, or invariant, under)
projectivity map between intervals, 57 an operation, 52, 147
projectivity property of a lattice,' 92 relatively complemented lattice, 60,205,
proper class, i.e., a class that is not a set, 134 [Exercises 12-13] 206
proper (congruence, filter, ideal, subuniverse), representation(s):
i.e., not the largest direct (or direct product decompositions),
159,201,268 (see also direct products)
quasi-cyclic group, 110, 171, 180 faithful, 129
quasigroups: isomorphic, 129, 160
defined, 122- 123 problems, 137
loops, 123 regular, Le., the Cayley representation, 128-
permuting congruences in [Exercise 31,205 129
. Steiner loops, 127 subdirect, 160
Steiner quasigroups, 123 irredundan t, 203
Steiner triple systems, 124 (for representations of groups, e.g., see groups
quotient: (representation of))
algebra, 28 Representatíon Theorem:
map, 28 for Boolean Algebras [Theorem 2.62],88
for Distributive Lattices [Theorem 2.611, 87
range (of a function), 6 for Finite Distributive Lattices [Theorem
rank (arity): 2.591, 85
of an operation, 11 residual properties (see algebras, varieties)
of a relation, 5 restriction:
rank function, of an algebra, 12 of congruences, 151,156
reduced terms, 239 of operations, 156
reduct, of an algebra, 13 retract of an algebra [Exercise 41, 194
refinement, of a chain [Exercise 3],76 retraction [Exercise 21, 131
refinement properties (see direct products) ring (S);
reflexive: annihilator in, 52,251
binary relation, 7 Artinian, 62
element of a structure, 311, 325 Boolean, 245
regular: center of, 176
permutation group, 130 congruences of [Lemma 11,175
representation (left or right), 128-129 defined, 14
relation algebras, 17-18 free, 240-241
relational (or relation, or relative) product of ideals of, 30, 175
binary relations, 6, 113, 160 idempotents in, 176
Index of Terms
ring(s) (continued) type, of an algebra, 13, 149

of integers, 30 simple algebra, 95, 157
modules over: spectrum:
Abelian property of [Example 31,251 free spectrum of a variety, 241
defined, 14-1 5 spectrum of a variety, 124
direct product decompositions of stabilizer subgroup (see 6-sets)
[Exercises 10-141,293-295 standard element of a lattice, 98
finitely generated, over a principal ideal Steiner (see quasigroups)
domain, 52,260 Stone duality, between Boolean algebras and
representation module of a linear trans- Boolean topological spaces [mentioned],
formation, 281-282 89
Noetherian, 62 structures (see relational structures)
polynomials of, 144 Sub A, the subalgebra lattice of A, 27, 35
representation of [Exercise 6],24 subalgebra, 19
with unit, 14 subalgebra lattice, i.e., the lattice of al1 sub-
zero, 252 universes of an algebra, 27, 35
(see also associative algebras, division rings, subdirect:
principal ideal domains) embedding, 169
root: factors, of an algebra, 159
kth root of an algebra (direct products), products, 160
261,321 representation, 159-160
of a rooted tree, 106 unions (i.e., subdirect products), 169
subdirectly irreducible algebra, 169
Schreier Refinement Theorem, 76 Subdirect Representation Theorem (of
Second Isomorphism Theorem [Theorem 4.101, G. Birkhoff) [discussed], 50,168,221,
149 [Theorem 4.441 171
semigroups: substitution property, 27-28
commutative: subuniverse(s):
behavior of class operators on [Exercise defined, 19
111,226 diagonal, 200
finite, without unique factorization finitely generated, 26
[Exercise 41, 265 generated by a subset, 25, 152,182, 183,
forbidden congruence lattice of [Exercise [Exercise 71 148
171, 180 lattice of, 27
simple [Exercise 161, 180 supremum, i.e., least upper bound, 37,44
defined, 13 symmetric difference, 245
free, 113 symmetric groups:
idempotent, i.e., bands, 115 defined, 119
left (right) zero, 115 representations in, $3.5
rectangular, 115 universality of the lattices of subgroups
representation of, 128-130 of symmetric groups [Theorem 4.631,
residually large, finitely generated variety of 193
[Exercise 281, 244 symmetry, 130
with unit, 112 Sym X, the symmetric group on X, 119
(see also semilattices, monoids) system, Le., a function, 6
semilattices:
defined, 16 tame congruence theory [mentioned], 4, 155
free, 115, 240 Tarski's Interpolation Theorem [Theorem 4.591,
join (meet)-, 114 187
representation of [Theorem 3.71, 114 term algebra of type a over X, 229
semilattice order, 114 term operation, 143
semiring, 262 defined by a term, 232
semi-simple algebra, 179, 204 term(s):
separation of points by homomorphisms, 160 defined, 231
set theory: depends on a variable, 232
fundamentals and notation, 5-9 Jónsson, 248
Sheffer's stroke operation, of Boolean algebras, Maltsev, 247
249 n-ary term of type a (or n-ary a-term), 232
similar algebras, 13 n-ary term of Y , 238
similarity: n-ary Y-term, 238
class, of algebras, 13, 149 operations, 143
Index of Terms
Pixley, 248 clone of a variety [Exercise 231, 141

reduced, 239 congruence distributive, 223
unique readability of, 229 centers of algebras in, [Exercises 7-82 258,
universal [Exercise 151, 111 [Exercise 41 300
Third Isomorphism Theorem [Theorem 4.141, characterized by a Maltsev property
151-152 [Theorem 4.144],248
torsion free Abelian group with non-unique discussed, 4, 196, 218
square root [Exercise 61,266 congruence modular, 4,223
transitive: Abelian algebras in, 257
binary relation, 7 commutator theory for, 196
closure: congruence permutable, i.e., Maltsev, 247
of a binary relation, 6 defined, 23,220
G-set, 108 equationally complete, i.e., minimal, 243
permutation group [Exercise 51, 132 equational theories of, 236
representation of a group, 129 equivalence of, 246
transpose (see intervals) finitely generated, 221
tree, an ordered set, 106 free spectra of, 241
true in A, 234 generated by a class of algebras, 220
type of an algebra, i.e., similarity type, 13, 149 interpretation of, 245, $4.12
locally finite, 4,221
ultraproducts, 3 Maltsev, 247
unary algebras (see 93.2): minimal [Exercise 171, 243
automorphism groups of [Theorem 3.141, nontrivial:
130 possessing an algebra that has more than
connected, 107, [Exercises 1 and 21 110 one element
direct products of [Exercises 1-31, 264-265 residually finite, 178
endomorphism monoids,of [Theorem 3.141, residually large, 179
130 residually less than rc, 178
lattices of subuniverses of [Exercise 4],27, residually small, 179
[Theorem 4.551 184 spectrum of, 124
mono-: vector space(s):
automorphism groups of [Exercise 121, bases of, 187
132-133 defined, 15
connected components of, 104 duality for finite dimensional, 217
cores of, 105 Jordan normal form of a linear transformation
decomposition into connected components [Exercises 2-81,281-282
[Theorem 3.31, 106 lattice of subspaces of, 209-210
graphical depiction of, 104-106
(see also G-sets, M-sets) weakly isomorphic algebras, 246
unique: Wedderburn's theorem [applied], 175
complements, lattices with, 318 well-ordered set, 7
factorization property (see direct products) Well-Ordering Theorem, 8
readability of terms, 229 Whitman's Theorem, 190
solution property, 118,123 word(s):
universal: concatenation of, 113
mapping property (see free algebras) problem for free modular lattices [negative
term [Exercise 151, 111 solution discussed], 4,241
universe: of an algebra, 12 reduced group words, 119
or sequences, 113,229
valid in A, 234
variables, 23 1 Zassenhaus Butterfly Lemma [Exercise 3],76
varieties: Zermelo-Fraenkel axioms of set theory, 18
arithmetical, 247 zero, algebras with, 282, $$5.4-5.5
axiomatized (or defined) by a set of equations, zero divisor in a class (direct products), 324
236 zero ring, 252
classification of, 244 Zorn's Lemma, 7, [Exercise 31 27

Algebras Lattices Varieties Volume 1

Загружено:

Сведения о документе

Исходное описание:

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Algebras Lattices Varieties Volume 1

Загружено:

Авторское право:

Доступные форматы

ALGEBRAS,

M. Adams, V. Guillemin, Measure Theory and Probability

Wadsworth & Brooks/Cole

O 1987 by Wadsworth, Inc., Belmont, California 94002. Al1 rights

Printed in the United States of America

Library of Congress Cataloging-in-PublicationData

Sponsoring Editor: John Kimmel

Volume 1 is a leisurely paced introduction to general algebra and lattice

Chapter 1 Basic Concepts

Chapter 3 Unary and Binary Operations

Chapter 4 Fundamental Algebraic Results

Chapter 5 Unique Factorization

The mathematician at the beginning of the twentieth century was confronted

emerged during this time in investigations of mathematical logic (cf. Tarski

c. The converse r" of a binary relation r on A is given by (a, b) E ru iff

ing ( x , , x , + , ) ~ rfor a11 i < n.

with A, # @ for al1 iE 1, then

for each finite ordinal n 2 1.

e. It is useful to know that n-tuples are functions having domain

(x, y ) ~p may be indicated by connecting points labeled x and y with a

1.1 ALGEBRAS AND OPERATIONS

An algebra is a set endowed with operations. Algebras are the fundamental

DEFINITION 1.1. An algebra is an ordered pair (A, F ) such that A is a

For the development of ring theory, it is essential to distinguish these operations

intended to name the unary operation of negation). But surely

is an uncomfortable way to display the ring of integers. In this situation and

p(Q) is the rank of QAfor al1 Q E 1.

and ( A , 1) is a monoid. A familiar example of a ring (with unit) is the integers

endowed with the familiar operations. Another example is the set of n x n

Vector Spaces and Modules

is in accord with geometric intuition by which these operations amount to

Bilinear Algebras over a Field

f,(a b) = (f,(a)). b = a -f,(b).

Suppose (A, + , ., -,O,f,),EFis an associative algebra over F. Define

A typical example of a semilattice is formed by taking A to be the collection of

A typical example of a lattice is formed by taking A to be the collection of al1

of al1 binary relations on an arbitrary set X, giving A the Boolean operations by

Prove that (A, +, *, -, O,f,),., is a Lie algebra.

1.2 SUBALGEBRAS, HOMOMORPHISMS, AND

One of the hallmarks of algebraic practice is the prominent role played by

DEFINITION 1.3. Let A be an algebra. A subset of the universe A of A, which

The ring of integers is a subalgebra of the ring of complex numbers.

for al1 a,, . - ,a,-, E A.

We distinguish severa1 kinds of homomorphisms and employ notation for

denote that h is a one-to-one homomorphism from A into B. We cal1 such

interna1 structure of the elements of an algebra can be used to establish algebraic

( e n A (f O, f l,. ,f '-l)), = ~ * i ( ~ .,~Jr-l)

where Q is an operation symbol (or a more complicated expression) of the algebi-a

significance and the expression above is needlessly complex. We replace it by

where ü stands for a tuple of elements of A of the correct length.

3. Let A*= ( A i : i E 1) be a system of similar algebras and assume that B is an

4. Let A = ( A i : i E I ) be a system of similar algebras. Let B be an algebra of the

1.3 GENERATION OF SUBALGEBRAS

Let A be an algebra and let X be an arbitrary subset of the universe A of A. X

Proof. Evidently OS is a subset of A. Let F be any basic operation of A and

F(a0, a,,.. a,-,) E B, since B is a subuniverse. Therefore F(ao,a,, . ,a,-,) E

and ( ) S is closed under F. m

DEFINITION 1.8. Let A be an algebra and let X E A. The subuniverse of A

Since X E A and A is a subuniverse of A, Theorem 1.7 justifies calling

THEOREM 1.9. Let A be un algebra and X G A. Define X,, by the following

Proof. The proof consists of two claims.

COROLLARY 1.10. Let A be un algebra and X G A. If a E sgA(x),then there

Proof. We will prove by induction on n that