Академический Документы
Профессиональный Документы
Культура Документы
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
ALGEBRAS,
LATTICES, VARIETIES
'7
Ralph N. McKenzie
University of California, Berkeley
George F. McNulty
University of South Carolina
Walter F. Taylor
University of Colorado
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
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 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
Bibliography
Table of Notation
Index of Names
Index of Terms
ALGEBRAS,
LATTICES, VARIETIES
%VOLkjiWE1
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
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
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
+ + +
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 - ,
--
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
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.
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).
+
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
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 ,
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
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
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
18 Chapter 1 Basic Concepts
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 ,
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
and
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
22 Chapter 1 Basic Concepts
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 ,
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
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.
24 Chapter 1 Basic Concepts
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.
al, -,
a,. E n S . For al1 B E S we know that a,, a,, - ., a,-, E B; but then
n~
S * . ,
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
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
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.
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.
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.
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.
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.
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.
32 Chapter 1 Basic Concepts
The routine proof of this theorem is left as an exercise. This theorem allows
us to proceed as we did with subuniverses.
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
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.
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
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
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
2.1 Fundamental Concepts
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.
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.
2.1 Fundamental Concepts
Fd+(3)
Figure 2.1
40 Chapter 2 Lattices
Figure 2.2
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 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.
42 Chapter 2 Lattices
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
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
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
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.
44 Chapter 2 Lattices
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.
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.
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.
THEOREM 2.12. Every complete lattice is isomorphic to the lattice of al1 closed
sets of some closed set system.
46 Chapter 2 Lattices
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.
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.
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.
Exercises 2.20 i!
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
52 Chapter 2 Lattices
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
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.
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
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,
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
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
2.3 Modular Lattices: The Rudiments 57
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.
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
58 Chapter 2 Lattices
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,.
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
So F preserves meets.
CLAIM 1: F is one-to-one.
Notice that x can be recovered from x v y as follows:
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
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,.
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).
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.
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
62 Chapter 2 Lattices
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
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
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
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.
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 - .
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.
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). •
66 Chapter 2 Lattices
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
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
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.
,
,
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) -
{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, 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 )
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
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 , .
72 Chapter 2 Lattices
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
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
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,
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'}).
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
76 Chapter 2 Lattices
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
78 Chapter 2 Lattices
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
= a v (b 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
Figure 2.13
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
82 Chapter 2 Lattices
THEOREM 2.52. The following statements are equivalent for any lattice L:
i. L is distributive.
ii. For any a EL, both da and $a are homomorphisms.
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,.
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.
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. ¤
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.
and
((b A a*) v c) A a = c.
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
2.5 Distributive Lattices 85
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].
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.
CLAIM O: F is a homomorphism.
Let a, b E L and c E J . Notice
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.
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.
CLAIM O: F is a homomorphism.
Let I be a prime ideal and a, b E L. Observe that
88 Chapter 2 Lattices
(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.
90 Chapter 2 Lattices
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.
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. ¤
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.
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
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
2.6 Congruence Relations on Lattices 93
and evidently
and so 8 E 4, as desired. m
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
2.6 Congruence Relations on Lattices 95
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.
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
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
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.
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. ¤
98 Chapter 2 Lattices
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:
i. L is distributive.
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
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
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 ,
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 ,
Chapter 3 Unary and Binary Operations
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
* ,
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
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
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
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
3.2 Unary Algebras 105
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), # ,
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
106 Chapter 3 Unary and Binary Operations
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
)
(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 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
110 Chapter 3 Unary and Binary Operations
$
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:
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
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].)
Chapter 3 Unary and Binary Operations
3.3 SEMIGROUPS
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
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.
= 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.
116 Chapter 3 Unary and Binary Operations
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 ,
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
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
118 Chapter 3 Unary and Binary Operations
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 .
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
3.4 Groups, Quasigroups, and Latin Squares
(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 )
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
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,
122 Chapter 3 Unary and Binary Operations
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
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
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,
Proof. One easily verifies that if (Q, ) is a Steiner quasigroup, then (Q, m, e, )
satisfies laws (1) and (2), which define quasigroups. m
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,). ¤
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
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
Chapter 3 Unary and Binary Operations
iff it is possible to define a group (G, -l, e) such that x/y = x - y-' and
S ,
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 ) ,
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
128 Chapter 3 Unary and Binary Operations
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?
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
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.
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.
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
-+
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
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
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 ,
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.
138 Chapter 3 Unary and Binary Operations
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 ,
Fundamental
Algebraic Results
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
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
Chapter 4 Fundamental Algebraic Results
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.
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.
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.
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
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
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
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-,),
148 Chapter 4 Fundamental Algebraic Results
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 ,
* 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
Clone A is the category defined near the end of $3.6,with objects the powers
A", then the set of maps of Clone A is identical with the centralizer of Clo A
in Clo A.
*14. Let C be the clone on A = (0,l) that is the clone of term operations of a
two-element lattice. Determine the centralizer C' (defined in the previous
exercise) and the double centralizer (C')'. 1s (C')' = C?
Figure 4.1
Proof. To prove (i) we notice that since f (x) = f (y) implies g(x) = g(y), and
since f is onto B, the relation
elements of B, and choose ai E A with f (a,) = b, for O < i < n. Then we have
which shows that h respects the interpretation of Q. It is clear from the definition
of h that h is one-to-one iff g(x) = g(y) -,f(x) = f(y).
To prove (ii), let f : A + A/q4 and g: A +A/y be the quotient homo-
morphisms. Now, the hypotheses of (i) are satisfied. The homomorphism h with
g = h o f has ker h = y/$. 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
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.
'"
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
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
".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
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.
154 Chapter 4 Fundamental Algebraic Results
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
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 ,
congruence above takes the form fi(ai) = fi(bi). Since the fi preserves p, we have
the desired conclusion.
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
156 Chapter 4 Fundamental Algebraic Results
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
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
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 ) .
"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
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 ,
(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 ,
*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.
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.
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.) ¤
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
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.
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.
When this holds, the relations p, and p, are called the factors of p. (They are
uniquely determined if p # @.)
( 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
4.4 Direct and Subdirect Representations 165
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.
166 Chapter 4 Fundamental Algebraic Results
$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
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.
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 ,
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
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
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.
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.
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.
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
172 Chapter 4 Fundamental Algebraic Results
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 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.
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.
Proof. To prove (i), we note first that if x and y are 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.
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.) •
where denotes congruence moda. The argument proves that A has only two
congruences 0, and 1,-i.e., that A is simple. ¤
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
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
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.
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.
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. •
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.
180 Chapter 4 Fundamental Algebraic Results
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
is a congruence.)
iv. Prove that the only simple Heyting algebra is the two-element one.
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
182 Chapter 4 Fundamental Algebraic Results
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
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.
and that {b,, ,b,-,) G C(X). If ai = bi for al1 i i n, then C({a,, - . a,-, 1) c e,
case, f,(b,, b,-,) = b,, which certainly belongs to C(X). This argument shows
m.,
where E = (c,, - ,ck-, ,c), implying that c E sgA(x).That concludes the proof.
M
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
184 Chapter 4 Fundamental Algebraic Results
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
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. ¤
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.
4.6 Algebraic Lattices 185
-
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. •
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).
186 Chapter 4 Fundamental Algebraic Results
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
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,
4.6 Algebraic Lattices
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 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.
188 Chapter 4 Fundamental Algebraic Results
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
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
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}.
Chapter 4 Fundamental Algebraic Results
(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. •
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
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
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
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.
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).
4.6 Algebraic Lattices 193
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-
194 Chapter 4 Fundamental Algebraic Results
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,
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
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.
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
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 Notice that the following equation holds for any binary relations p and T.
l Hence a o" p = p on a, ending the proof that (b) implies (a) for n even.
198 Chapter 4 Fundamental Algebraic Results
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
-
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 ,
¿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 ,
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
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.
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.
/ 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
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
,
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.
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
4.7 Permuting Congruentes 205
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.
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
206 Chapter 4 Fundamental Algebraic Results
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
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".
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.
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.
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.
4.8 Projective Geometries 211
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,
LEMMA 4.84.
-
as desired. This argument proves that u is a complement of u, and that concludes
our proof of (iv) (i).
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
212 Chapter 4 Fundamental Algebraic Results
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.
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.
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
4.8 Projective Geometries 213
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,].
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
4.8 Projective Geometries 215
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
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.
4.8 Projective Geometries 217
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 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.
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
220 Chapter 4 Fundamental Algebraic Results
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
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. •
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".
222 Chapter 4 Fundamental Algebraic Results
We introduce a useful class operator for the study of locally finite varieties.
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
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
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. ,
224 Chapter 4 Fundamental Algebraic Results
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.
Figure 4.8
4.10 Class Bperators and Varieties 225
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
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.)
226 Chapter 4 Fundamental Algebraic Results
*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 ,
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
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.
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
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.
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.
230 Chapter 4 Fundamental Algebraic Results
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
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 ,
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.
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
4.11 Free Algebras and the HSP Theorem 231
Proof. This is a corollary of Lemmas 4.110 and 4.1 12 and Theorem 4.1 17. i
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 } .
For every term p there is a unique smallest set of variables {x,, . xk-,} for S ,
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).
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.
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
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. ¤
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 .
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
+
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 ,
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
236 Chapter 4 Fundamental Algebraic Results
1
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. ¤
+
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.
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. 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.
238 Chapter 4 Fundamental Algebraic Results
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.
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 ,
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.
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 ,
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.
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
240 Chapter 4 Fundamental Algebraic Results
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
xn-, . We use + , and - for the basic operation symbols, instead of v , A ,and
S ,
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.
4.11 Free Algebras and the HSP Theorem 243
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 . . .
- ,
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.)
(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
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 ,
-
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 .
-
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.
(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].
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)
248 Chapter 4 Fundamental Algebraic Res&
Proof. Suppose first that for a certain n we have terms do, dn obeying the
S ,
,
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 ,
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.
250 Chapter 4 Fundamental Algebraic Results
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.
iff for every n 2 1, and for every term operation t E Clo,,, A and for al1 c,, -.m ,
en,di, EA EA
Thus an algebra is Abelian iff (29) holds for al1 term operations and al1
elements of the algebra.
+
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 ,
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,
+
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
252 Chapter 4 Fundamental Algebraic Results
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.
Proof. Straightforward. m
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
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 ,
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
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.
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,).
4.13 Commutator Theory 255
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).
256 Chapter 4 Fundamental Algebraic Results
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
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
4.13 Commutator Theory 257
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.
258 Chapter 4 Fundamental Algebraic Results
Unique Factorization
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
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
264 Chapter 5 Unique Factorization
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
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:
(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.)
266 Chapter 5 Unique Factorization
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:
given by (f (a)), = u/@.It turns out to be convenient to consider the slightly more
general situation of the natural map
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
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
270 Chapter 5 Unique Factorization
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.
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.
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
272 Chapter 5 Unique Factorization
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.
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,
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
5.2 Direct Factorization and Isotopy
(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
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
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 . ,
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
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
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:
-,
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).
5.2 Direct Factorization and Isotopy
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 ,
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.
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.
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
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
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.
-
then m = n, and, after renumbering the y,, we have Pi y, for each i.
280 Chapter 5 Unique Factorization
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
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
..T*v;5v; 5 0 .
Finally, let us call T (or V,) cyclic iff it is multicyclic with k = l.
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.)
282 Chapter 5 Unique Factorization
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:
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
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.
and
((B, Y):k e r f = P x Y}.
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
/ 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).
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)
286 Chapter 5 Unique Factorization
and finally,
a + (hfgf'b + c) = h[a + (hfgf'b + c)] + hf[a + (hfgf'b + c)]
= hy + hf(a + c)
= 6.
and
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
5.4 Algebras with a Zero Element 287
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 ,
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
l
and so
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
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
5.4 Algebras with a Zero Element 289
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.
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
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.
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
290 Chapter 5 Unique Factorization
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. ¤
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
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
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
292 Chapter 5 Unique Factorization
AxCrBxCrD
for D un algebra with zero and C finite, then A r B.
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.
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 )
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.
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
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'.
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.
+
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 - .
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.
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
5.5 The Center of an Algebra with Zero 297
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,)
,
298 Chapter 5 Unique Factorization
for every n > 1, for every term operation t E Clo,+,A, and for al1 c,, - c,, d,, s.,
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
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
5.5 The Center of an Algebra with Zero 299
+
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). ¤
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.
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
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.
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,
5.6 Some Refinement Theorems
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
302 Chapter 5 Unique Factorization
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
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.
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). ¤
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,
304 Chapter 5 Unique Factorization
-
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 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.
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.
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:
Chapter 5 Unique Factorization
-
(R" O R) 0.- o (R" o R), with n factors (R" o R).
-
then a R b-a1Rb'.
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
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.
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.
308 Chapter 5 Unique Factorization
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.
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
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
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
310 Chapter 5 Unique Factorization
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
-
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
5.6 Some Refinement Theorems 311
COROLLARY 2. Every connected ordered set has the strict refinement property.
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 ,
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.
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
5.6 Some Refinement Theorems 313
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
-
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.
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.
314 Chapter 5 Unique Factorization
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:
-
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
5.6 Some Refinement Theorems
-
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
316 Chapter 5 Unique Factorization
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
-
[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
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
320 Chapter 5 Unique Factorization
(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
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
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
5.7 Cancellation and Absorption 321
Proof. This is immediate from Corollary 1 and the fact that kth roots are unique
in (a,-,1). ¤
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
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
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
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.
5.7 Cancellation and Absorption 323
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
324 Chapter 5 Unique Factorization
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
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
5.7 Cancellation and Absorption 325
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.
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 ) ~ .
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,
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
326 Chapter 5 Unique Factorization
X must contain al1 the elements that we described, and so this is the smallest
one. ¤
for u,, - m , u,, ER and any k 2 0. And similarly, Ks is given by al1 products
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 ,
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
(with an arrow from m to n iff (m, n) E R). The directed graph C, is also known as
a p-circuit.
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
5.7 Cancellation and Absorption 329
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:
Chapter 5 Unique Factorization
(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. •
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.
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.
Proof. Let us take algebras A,, A,, A, as described in the above definition
of infinite divisibility. Thus we have homomorphisms
5.7 Cancellation and Absorption 333
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. •
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.
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.
336 Chapter 5 Unique Factorization
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
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.
Math. Soc., Providence, R.I.
[1967]. Lattice theory, Third Edition, Colloquium Publications, Vol. 25, Amer.
Math. Soc., Providence, R.I.
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.
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.
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.
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
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. Remak [1911]. Über die Zerlegung der endliche Gruppen in direkte unzerlegbare
Faktoren, J. reine angew. Math. 139, 293-308.
Bibliography
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.
-
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