Libro - (Advances in Mathematics 2) P. A. Grillet (Auth.) - Commutative Semigroups-Springer US (2001) PDF

Commutative Semigroups
Advances in Mathematics
VOLUME2
Series Editor:
J. Szep, Budapest University of Economics, Hungary
Advisory Board:
G. Erjaee, Shiraz University, Iran
W. Fouche, University of South Africa, South Africa
P. Grillet, Tulane University, U.S.A.
H.J. Hoehnke, Germany
F. Szidarovszky, University of Arizona, U.S.A.
P. Zecca, Universita di Firenze, Italy
The titles published in this series are listed at the end of this volume.
Commutative
Semigroups
by
P.A. GRILLET
Tulane University, New Orleans, U.S.A.
SPRINGER-SCIENCE+BUSINESS MEDIA. B.V.

A C.I.P. Catalogue record for this book is available from the Library of Congress.
ISBN 978-1-4419-4857-1 ISBN 978-1-4757-3389-1 (eBook)

DOI 10.1007/978-1-4757-3389-1
Printed on acid-free paper
All Rights Reserved

© 2001 Springer Science+Business Media Dordrecht
Originally published by K.luwer Academic Publishers in 2001
Softcover reprint of the hardcover 1st edition 2001
No part of the material protected by this copyright notice may be reproduced or
utilized in any form or by any means, electronic or mechanical,
including photocopying, recording or by any information storage and
retrieval system, without written permission from the copyright owner
A ma mere
a qui je dois tout
CONTENTS
Preface . . . . . . . . . . . . . . • • . . XI
General structure theory

Chapter I. Elementary properties
1. First results . . . . . I
2. Homomorphisms and congruences 6
3. Ideals . . . . . . . . . . 13
4. Divisibility . . . . . . . I7
5. Free commutative semigroups 20
6. Presentations . . . . . . 25
Chapter II. Cancellative semigroups 29
I. Semigroups of fractions 29
2. Universal groups 32
3. Cancellative semigroups 36
4. Numerical semigroups 39
5. General structure 44
6. Faces . . . . 50
7. Free embedding 54
8. Krull monoids 57
Chapter III. Semilattice decompositions 69
I. General results . . . . . . . 69
2. Clifford semigroups . . . . . 73
3. Complete archimedean semigroups 78
4. N-semigroups . . . . . . . . 82
5. Subcompiete archimedean semigroups 86
6. Power-joined semigroups . . 90
Chapter IV. Subdirect decompositions 95
I. Subdirect products 95
2. Separative semigroups IOI
3. N ilsemigroups I04
4. Ponizovsky decompositions 107
5. Elementary semigroups 112
Chapter V. Group coextensions . 115
1. Dividing by J{ . . . . 115
2. SchOtzenberger functors 117
3. Coextensions . . . . . 120
4. Group coextensions I25
5. Subdirectly irreducible semigroups 133
vii
viii CONTENTS.
Chapter VI. Finitely generated semigroups 141

1. Redei's Theorem . . . . 141
2. Subdirect decompositions . 144
3. Subelementary semigroups 148
4. The Completion Theorem 149
5. Irreducible semigroups . . 154
6. Archimedean semigroups . 158
7. The a.c.c. on subsemigroups 160
Chapter VII. Subcomplete semigroups 165
1. Completions . . . . . . . 165
2. Ponizovsky families . . . . 169
3. Another Completion Theorem 173
4. Properties . . . . . . 179
5. Schutzenberger functors 181
Chapter VIII. Other results 187
1. Examples . . . . 187
2. Products and subsets 190
3. Homomorphisms and congruences 194
4. Other topics . . . . . . 198
Congruences
Chapter IX. Nilsemigroups 203
1. Free commutative monoids 204
2. The zero class 208
3. Comer points 212
4. Nests . . . . 215
5. Examples 221
Chapter X. Group-free semi groups 227
1. Semilattice congruences 227
2. Direction sets 232
3. Extent cells 236
4. Trace congruences 244
5. Main result 250
6. Partially free semigroups 254
Chapter XL Subcomplete semigroups 259
1. Direction sets 259
2. Extent cells 261
3. Trace congruences 268
4. Strand groups 271
CONTENTS. ix
5. Strand bases 278

6. Main result 281
7. Subelementary congruences 288
Cohomology
Chapter XII. Commutative semigroup cohomology 295
1. Triple cohomology 295
2. Abelian group objects 301
3. Beck extensions 305
4. Commutative semigroup cohomology 309
5. Symmetric cochains 318
Chapter XIII. The overpath method 327
1. Overpaths . . . . 327
2. Main result 332
3. Proof of main result 338
4. Defining vectors 345
5. Partially free semigroups 355
Chapter XIV. Semigroups with zero cohomology 361
1. Group-free monoids 362
2. The zero group 366
3. Nilmonoids 369
References . 379
Author Index 417
Notation 425
Index 427
PREFACE.
The first book on commutative semigroups was Redei's The theory of.finitely
generated commutative semigroups, published in Budapest in 1956.
Subsequent years have brought much progress. By 1975 the structure of
finite commutative semigroups was fairly well understood. Recent results have
perfected this understanding and extended it to finitely generated semigroups.
Today's coherent and powerful structure theory is the central subject of the present
book.
1. Commutative semigroups are more important than is suggested by the stan-
dard examples ofsemigroups, which consist ofvarious kinds oftransformations or
arise from finite automata, and are usually quite noncommutative. Commutative
semigroups provide a natural setting and a useful tool for the study of factoriza-
tion in rings. Additive subsemigroups of N and Nn have close ties to algebraic
geometry. Commutative rings are constructed from commutative semigroups as
semigroup algebras or power series rings. These areas are all subjects of active
research and together account for about half of all current papers on commutative
semi groups.
Commutative results also invite generalization to larger classes of semigroups.
Archimedean decompositions, a comparatively small part oftoday's arsenal, have
been generalized extensively, as shown for instance in the upcoming books by
Nagy [2001] and Ciric [2002].
In the past few years commutative semigroups have appeared in lattice the-
ory (Wehrung [ 1998]), global fields (Geroldinger [ 1997]), functional equations
(Brillouet-Belluot [ 1996], Taylor [ 1999]), Banach algebras (Esterle [ 1997]), C*-
algebras (Murphy [1996], Ressel & Ricker [1998]), harmonic analysis (Youssfi
[1998]), numerical analysis (Markov [1995]), theory of computation (Abdali
[ 1993]), and other unexpected places (see Chapter VIII).
Commutative semigroups thus have a wide range of potential applications.
Unfortunately, they are also quite complex and what has been known of their
xi
xii PREFACE.
structure has been either insufficient for applications or not readily available from
existing books. The present account of current structure theory fills a large gap
and will, we hope, ameliorate this situation. Recent results and many older results,
now scattered among a number of articles, should also benefit from a systematic
exposition.
2. Research on commutative semi groups has a long history. Lawson [ 1996]
makes a good case that the earliest article which would currently receive a semi-
group classification is an 1826 paper by Abel which clearly contains cancellative
commutative semigroups. It has indeed long been known that these semigroups
can be retrieved from abelian groups; some aspects of this difficult process are
explored in Chapter II (after the inevitable basic definitions and results in Chap-
ter 1).
The "classical" structure theory is more recent and includes notable contribu-
tions by Gluskin, McAlister, Petrich, Preston, Redei, Schein, S. Schwarz, Shevrin,
Tamura, Yamada, and others. It provides a solid foundation for later results and is
based upon three main ideas. Semilattice decompositions, discovered by Schwartz
[1953] and Tamura & Kimura [1954], and extended to all semigroup by Yamada
[ 1955], provided for many years the only structural information on commutative
semigroups. Subdirect decompositions of finite commutative semigroups were
found by Ponizovsky [1962]. Group coextensions were developed independently
by Grillet [ 1974] and Leech [ 197 5]. These three tools are explained in Chapters
III, IV, and V, and together provide a good first grasp of finite commutative semi-
groups (more generally, of complete semi groups) as explained in Grillet [ 1995].
The elaborate theory by Redei [ 1956], intended to produce numerical in-
variants for finitely generated commutative semigroups, brought little structural
insight but uncovered the important property, known today as Redei 's Theorem,
that congruences on these semigroups satisfY the ascending chain condition. The
shorter proof of this theorem given by Herzog [ 1970] and Preston [ 1975] led to
a first understanding of finitely generated commutative semigroups, by means of
subdirect decompositions (Grillet [ 1975P]) and a fourth tool, completions (Grillet
[1975C]). This occupies most of Chapter VI.
Very recent developments complete the basic structure theory of finitely gen-
erated semigroups (more generally, of subcomplete semigroups) by means of a
fifth tool, Ponizovsky families (Grillet [200 1S]). This constitutes Chapter VII
and completes the first part of current structure theory. The main results in this
part build semigroups from cancellative semigroups and nilsemigroups, arranged
along semilattices. Nilsemigroups must then be constructed by other methods.
3. The present structure theory has two other parts. The second part, which
PREFACE. xiii
constitutes Chapters IX, X, and XI, contains constructions of subcomplete semi-

groups, including nilsemigroups and finitely generated semigroups, due to the
author [ 1991 N], [ 1993], [ 1996C], [200 1C]. These semi groups are constructed
indirectly, as quotients of free semigroups by subcomplete congruences. Unlike
Redei's short and elegant description, this construction of congruences is rather
long but is highly detailed and geometric. It includes (and is based on) all the
structural information acquired in the first part and has connections to the co-
homology in the third part. Such a global construction bypasses all difficulties
that arise when a commutative semigroup is reassembled from its archimedean
components or Ponizovsky factors. Chapters IX and X highlight two important
particular cases, nilsemigroups and group-free semigroups; the general case oc-
cupies Chapter XI.
The third part of the current theory, which constitutes Chapters XII, XIII, and
XIV, is the cohomology theory for commutative semigroups introduced by the
author in [ 1974] and [ 1991 C] and defined in Chapter XII. A similar cohomology
theory for monoids in general was developed independently by Leech [1975].
Both theories arise naturally from group coextensions and are particular instances
of triple cohomology. They use abelian group valued functors for coefficients
and their second group classifies group coextensions by such functors. The ear-
lier Eilenberg-MacLane cohomology of monoids is a particular case of Leech
cohomology in which the abelian group valued functor is constant. The functor
Ext for abelian groups is a similar particular case of commutative semigroup
cohomology.
Commutative semigroup cohomology is especially interesting because it clas-
sifies semigroups and not just coextensions: if S is complete group-free and <G
is Schiitzenberger, then H 2 ( S, <G) classifies complete commutative semi groups
T with Tj'){ ~Sand Schiitzenberger functor isomorphic to <G (Grillet [1974]).
Cohomology is therefore an integral part of the structure theory. Applications had
to wait until the overpath method in Chapter XIII provided an efficient way to
compute H 2 (S, <G) (Grillet [1995M]). A number of results can be proved by this
method, using the knowledge of congruences acquired in the second part, and
will be found in Chapters XIII and XIV.
4. A very large amount of work has been done on commutative semigroups,
enough for several volumes. This book covers the structure theory in some detail
but surveys other worthwhile topics only briefly. Left out altogether are five areas
which are covered elsewhere, or for which short surveys would be inappropriate:
factorization theory; semigroup rings; varieties and pseudo-varieties of commu-
tative semigroups (recently determined by Kisielewicz [1991], [1994]); ordered
xiv PREFACE.
semigroups; and most algorithmic aspects of the theory. All other topics are sur-
veyed very sketchily, mostly in Chapter VIII, and for these we have tried to provide
as many references as possible (about 850 altogether). Such comprehensiveness
probably precludes perfection, if only because there is no clear boundary between
general and commutative semigroup theory. The author extends his apologies to
all authors whose work should have received greater mention or has been inad-
vertently left out.
5. I wanted this book to be published, like Redei's, by the Akademiai Kiado
in Budapest in its "Pure and Applied Mathematics" series, which has recently
become the series "Advances in Mathematics", published by Kluwer Academic
Publishing. I owe heartfelt thanks to my Hungarian friends, Prof. Szep Jeno,
Marki Laszlo, and Szendrei Maria, for their unfailing advice and support, and
particularly to Prof. Szep who also sponsored this book. KOszOnOm, barataim!
Thanks also go to Prof. Scott Chapman, Franz Halter-Koch, and Boris Schein for
helpful communications, and to the folk at Kluwer who have consistently been a
pleasure to deal with.
Last, but not least, special thanks are due my beloved wife, for her love,
patience, and understanding.
Slidell, Louisiana, March 2001.

Chapter I.
ELEMENTARY PROPERTIES.
This chapter contains basic first properties of commutative semigroups: idem-

patents, subsemigroups, homomorphisms and congruences, ideals, ideal exten-
sions, JC-classes and Schiitzenberger groups, free commutative semigroups, pre-
sentations. It is written for readers who are not familiar with semigroups and can
therefore expect a few surprises.
Many results in this chapter are stated for commutative semigroups but extend
to all semigroups (with suitable modifications); interested readers should consult
Clifford & Preston [1961], Howie [1976], [1995], or Grillet [1995].
1. FIRST RESULTS.
1. A semigroup is a set S together with an associative binary operation on

S. (This is the only axiom.)
The name "semigroup" suggests a generalization of groups. To disabuse the
reader of this erroneous belief we compare the number of distinct (not isomorphic)
groups of order n with the number of distinct (not isomorphic or antiisomorphic)
semigroups of order n, which has been determined for n :£ 8 (JOrgensen & Wick
[1977]; Sato, Yama, & Tokizawa [1991]):
All Semigroups
n Groups Semi groups
2 1 4
3 1 18
4 2 126
5 1 1160
6 2 15,973
7 1 836,021
8 5 1,843, 120,128
1
2 I. ELEMENTARY PROPERTIES.
The number of semigroups of order 9 has not been determined; it is known to

exceed 52,952,000,000,000, but probably not by much (JUrgensen, Migliorini, &
Szep [1991]).
A semigroup is commutative when its operation is commutative. (Commu-
tative groups are called abelian.) Commutative semigroups are not as numerous
as general semigroups but are still much more numerous than abelian groups:
Commutative Semigroups
n Groups Semigroups
2 1 3
3 1 12
4 2 58
5 1 325
6 1 2143
7 1 17,291
8 3 221,805
9 2 11,545,843
The number of commutative semigroups of order 9 is from Grillet [ 1996N].

These large numbers suggest that even commutative semigroups are quite
unlike groups and constitute a wholly different kind of algebraic object.
2. There are some natural examples of commutative semigroups.
We denote by N = {0, 1, 2, ... , n, ... } the additive semigroup of natural
numbers. N is a commutative semigroup under multiplication as well. We also
denote by N+ = {1, 2, ... , n, .. .} the additive semigroup of all positive natural
numbers, by Q+ the additive semigroup of all positive rational numbers, and by
JR+ the additive semigroup of all positive real numbers.
Abelian groups are commutative semigroups.
A semilattice (short for lower semilattice) is a partially ordered set in which
any two elements a and b have a greatest lower bound (or meet, or infimum)
a 1\ b. Every lower semilattice is a commutative semigroup under the binary
operation 1\.
Groups and semilattices lie wholly outside semigroup theory. By well-estab-
lished tradition, we regard as solved any problem which can be stated in terms of
groups or semilattices (we dump it onto other unsuspecting mathematicians).
Every commutative ring or algebra is a commutative semigroup under multi-
plication. In fact, many arithmetic properties of rings are now studied in purely
1. FIRST RESULTS. 3
multiplicative terms. (Constructing rings from semigroups is even more fruitful;

semigroup algebras are a major employer of commutative semigroups.)
3. The operation S x S --+ S on a semi group S is usually written like
a multiplication (x,y) 1-----+ xy; sometimes (for instance, in N) like an addition
(x,y) 1-----+ x + y. In the multiplicative notation, associativity is the property
x (y z) = (xy) z for all x, y, z E S;
commutativity is the property
xy = y z for all x, y E S.
Associativity and commutativity have a number of consequences which we
hope are well known to the reader. Associativity allows products of three or more
elements to be written without parentheses. Commutativity implies that a product
does not depend on the order of its terms:
for every permutation a- of {1, 2, ... , n}.

Powers are particular cases of products. When a is an element of a semi group
S and n is a positive integer, then an is the product of n elements of S all
equal to a (with an = a if n = 1 ). If S has an identity element 1 one can also
define a 0 = 1. In a commutative semigroup S, the equalities
hold for all x, y E S and m, n > 0 ( m, n ~ 0 if S has an identity element).

In the additive notation, products x 1 x 2 · · · xn become sums x 1 + x 2 + ··· + xn
and powers xn become positive integer multiples nx.
4. An identity element of a semigroup S is an element e of S such that
ea = ae = a for all a E S. When an identity element exists, it is unique, and
we normally denote it by 1 (by 0 in the additive notation). A semigroup with an
identity element is a monoid.
In a monoid, empty products (= products a 1 a 2 · · · an of elements of S in
which n = 0) are, by definition, equal to the identity element; for instance, zero
powers a 0 = 1.
An identity element can be adjoined to any semigroup S: given 1 ¢: S,
define an operation on S U {1} as follows: 1 is an identity element and xy is
the same in S and S U {1} for all x, y E S. Then S U {1} is a monoid, which
is commutative if S is commutative.
The identity element of S U {1} is not the same as the identity element of S
in case the latter exists. Hence the following construction:

{ S if S has an identity element,
s1 =
S U { 1} if S does not have an identity element,
which adjoins an identity element only when necessary.
A zero element of a semigroup S is an element z of S such that za = az = z
for all a E S. When a zero element exists, it is unique, and we normally denote
it by 0 (by oo in the additive notation).
A zero element can be adjoined to any semigroup S: given 0 ~ S, define
an operation on S U { 0} as follows: 0 is a zero element and xy is the same in
S and S U {0} for all x, y E S. Then S U {0} is a semigroup with zero, which
is commutative if S is commutative.
5. More generally, an idempotent of a semigroup S is an element e of S
such that e 2 = e; then en= e for all n > 0. We denote by E(S) the set of all
idempotents of S.
When S is commutative the Rees order on E(S) is defined for all e,J E
E(S) by:
e ~ f if and only if ef =e.
If an identity element (a zero element) exists, then it is the greatest (the least)
element of E(S) under the Rees order.
Proposition 1.1. When S is a commutative semigroup, then E(S) is a
(possibly empty) semilattice under the Rees order, in which e A f = ef for all e
and f.
Proof. First the Rees order is a partial order relation: for all e,J,g E E(S),
e ~ e since e is idempotent; e ~ f ~ e implies e = ef = fe = f; and e ~ f ~ g
implies e = ef = efg = eg and e ~g. Also e(e!) = ef and (e!)f = ef, so
that ef ~ e,J. If conversely g ~ e and g ~ f, then efg = ef and g ~ ef.
Thus ef is the infimum of e and f in E(S). 0
The Rees order can be defined in any semigroup S and is always a par-
tial order relation on E(S) (Rees [1940]); but then E(S) is not necessarily a
semilattice. It was generalized to arbitrary elements by Mitsch [1986], [1994].
Corollary 1.2. Let S be a commutative semigroup. If every element of S
is idempotent, then S is a semilattice under the Rees order, in which a A b = ab
for all a and b. If conversely Y is a semilattice, then (Y, A) is a commutative
semigroup in which every element is idempotent, and the Rees order on (Y, A)
is the given partial order on Y.
1. FIRST RESULTS. 5
Accordingly, commutative semigroups in which every element is idempotent

may be identified with (lower) semilattices, and are, in fact, called semilattices.
6. Subsets A, B ~ S of a semigroup S are multiplied by:
AB = {ab I a E A, b E B}.
In particular,
Ac = { ac I a E A} and cB = { cb I b E B}
for all A, B ~ S and c E S. Multiplication of subsets inherits associativity and
commutativity from S.
A subsemigroup of a semi group S is a subset T of S which is closed under
the operation on S ( xy E T for all x, y E T ); equivalently, such that TT ~ T.
For instance, S and the empty set are subsemigroups of S.
Every subsemigroup T of S inherits a semigroup operation T x T --*
T from S; this semigroup T is also called a subsemigroup of S. If S is
commutative, then so is T.
Every intersection of subsemigroups of S is a subsemigroup of S. Hence
there is for every subset X of S a smallest subsemigroup T of S which contains
X ; T is the intersection of all the subsemigroups of S which contain X, and is
the subsemigroup (sometimes denoted by (X) or by X*) generated by X.
Proposition 1.3. The subsemigroup generated by a subset X is the set
of all products of one or more elements of X. In a commutative semigroup,
the subsemigroup generated by a subset X is the set of all products of positive
powers of one or more distinct elements of X.
Proof. A subsemigroup which contains X must by induction contain all
nonempty products of elements of X. Conversely the set T of all nonempty
products of elements of X is closed under multiplication and contains all products
of one element of X, i.e. contains X .
Every sequence x 1 , x 2 , ... , xn of elements of X can be permuted into a
sequence y 1 , ... , y1 , y2 , ... , y 2 , ... , yk, ... , yk, where y 1 , y 2 , ... , Yk are the
distinct elements of { x 1 , x 2 , ... , xn}. In a commutative semigroup, the products
x 1 x 2 · · · xn and y 1 · · · y1 y2 · · · y2 · · · Yk · · · Yk are equal; the latter is a product
of positive powers of distinct elements of X. D
For example, the cyclic subsemigroup generated by X = {x} consists of all
the positive powers of x.
If the subsemigroup generated by X is S itself, then X generates S and
the elements of X are generators of S; this means that every element of S is
the product of one or more elements of X. A semigroup is finitely generated

when it is generated by a finite subset, cyclic when it is generated by a single
element.
Proposition 1.3 has an analogue for monoids. When S is a monoid, a sub-
monoid of S is a subsemigroup T of S which contains the identity element of
S; then T is a monoid in its own right, with the same identity element as S.
Proposition 1.4. In a monoid, the submonoid generated by a subset X is the
set of all products of elements of X. In a commutative monoid, the sub monoid
generated by a subset X is the set of all products of positive powers ofdistinct
elements of X.
This works since empty products yield the identity element.
2. HOMOMORPHISMS AND CONGRUENCES.
1. Let S and T be semigroups. A homomorphism of semigroups of S into

T is a mapping c.p : S ---+ T such that c.p (ab) = c.p( a) c.p(b) for all a, b E S.
Semigroup homomorphisms preserve all nonempty products:
c.p (a 1 a2 ···an) = c.p(a1 ) c.p(a 2 ) ··· c.p(an)
and preserve positive powers: c.p (an) = c.p( a )n.
The identity mapping 18 on a semigroup S is a homomorphism of S onto
S. If c.p : S ---+ T and '1/J : T ---+ U are homomorphisms, then so is '1/J o c.p :
S ---+ U. An isomorphism of semigroups is a bijective homomorphism; the
inverse bijection is also an isomorphism.
When T is commutative, the pointwise product
(c.p.'lj;)(a) = c.p(a) 'lj;(a)
of two homomorphisms c.p, '1/J : S ---+ T is a homomorphism c.p. '1/J : S ---+ T. With
this operation the set Hom ( S, T) of all homomorphisms of S into T becomes
a commutative semigroup.
2. Semigroup homomorphisms share a number of basic properties with map-
pings and with homomorphisms of algebraic systems in general.
When S and T are sets, a mapping c.p : S ---+ T has a range or image
im c.p = c.p(S) ~ T and induces an equivalence relation ker c.p on S,
ker c.p = {(a, b) E S x S I c.p(a) = c.p(b)}.

2. HOMOMORPHISMS AND CONGRUENCES. 7
This provides a quotient set S /ker cp (the set of all equivalence classes) and a
projection or canonical mapping S --+ S /ker cp, which sends x E S to its
equivalence class. Then cp induces a bijection S/ker cp --+ im cp, which sends
the equivalence class of x E S to cp( x), and cp can be reconstructed by composing
the projection S --+ S /ker cp, the bijection S /ker cp --+ im cp, and the inclusion
mapping im cp --+ T.
S~T
1 r
S /ker cp -----t im cp
Homomorphisms of groups have similar properties, with the important differ-
ence that quotient groups are constructed from subgroups. As we shall see, semi-
groups are more like sets than like groups in that, in general, quotient semigroups
cannot be constructed from subsets and must be constructed from equivalence
relations.
3. First, given a semigroup S and an equivalence relation c on S, how can
we induce an operation on the quotient set S/ c? The answer is:
Proposition 2.1. Let S be a semigroup and c be an equivalence relation
on S. The following conditions are equivalent:
(1) there exists an associative operation on Sf£ such that the projection
S --+ S / c is a homomorphism;
(2) for all a,b,c,d E S, if a c c and b c d, then abc cd.
When either condition holds, there is only one associative operation on S / c such
that the projection a 1---1- Ea is a homomorphism; the product of Ea and Eb
in Sf£ is the equivalence class which contains their product as subsets of S,
namely Eab· If S is commutative, then so is Sf£.
Ea denotes the £-class of a (=the equivalence class of a modulo c).
Proof. If a Ea is a homomorphism, then Ea = Eb, Ec = Ed implies
1----t
Eac = Ea.Ec = Eb.Ed = Ebd; thus (1) implies (2).
Conversely let (2) hold. By (2), c E Ea, d E Eb implies cd E Eab; thus
the product Ea Eb of Ea and Eb as subsets of S is contained in the single
equivalence class E ab . If the projection a 1---1- E a is a homomorphism, then the
product Ea.Eb = Eab of Ea and Eb in Sf£ is the equivalence class which
contains their product E a Eb as subsets of S; there is only one operation on S / c
with this property, and it is the operation described in the statement. With this
operation, Ea.Eb = Eab holds in Sf£; hence Sf£ is a semigroup:
Ea.(Eb.Ec) = Ea.Ebc = Ea(bc) = E(ab)c = Eab·Ec = (Ea.Eb).Ec
and the projection s

-----+ sI e is a homomorphism. Thus (2) implies ( 1). If s
is commutative, then
Eb.Ea = Eba = Eab = Ea .Eb
and S I e is commutative. 0
A congruence on a semigroup S is an equivalence relation e
on S which
satisfies condition (2) in Proposition 2.1; then the quotient semigroup of S by
e se
is the semi group I in Proposition 2.1' such that the projection s
-----+ I se
is a homomorphism. The equivalence relation on S induced by the projection
s se e
-----+ 1 is itself.
A congruence on a group is completely determined by the equivalence class
of the identity element, so that quotient groups can be constructed from normal
subgroups. This nice property does not extend to semigroups; not even to com-
mutative monoids with a zero element. For instance let S be the semilattice (also
a monoid) S = { 0, e, 1} in which 0 < e < 1. The equivalence relation whose
classes are { e, 0} and {1} is a congruence; so is the equality (whose classes are
{0}, {e}, and { 1} ); thus a congruence on S is not determined by the class of
the identity element. Similarly, the equivalence relation whose classes are { 1, e}
and {0} is a congruence; hence a congruence on S is not determined by the class
of the zero element.
4. Armed with quotient semigroups we can now state the Homomorphism
Theorem (also known as the First Isomorphism Theorem):
Theorem 2.2. When <p : S -----+ T is a homomorphism of semigroups:
( 1) Im <p = <p ( S) is a subsemigroup ofT;
( 2) ker <p is a congruence on S;
(3) there exists an isomorphism Slker <p -----+ Im <p such that the diagram
S~T
1
Slker <p ~ Im <p
r
commutes; in particular Slker <p ~ Im <p. If S and T are commutative, then
so are Im <p and S lker <p.
Proof. When <p : S -----+ T is a homomorphism, Im <p = <p ( S) is a sub-
semigroup ofT, since <p(a) <p(b) = <p(ab) for all a,b E S; e
= ker<p is a
congruence on S, since <p(a) = <p(b), <p(c) = <p(d) implies <p(ac) = <p(a) <p(c) =
<p(b) <p(d) = <p (bd); and the bijection Ea ~ <p(a) is a homomorphism, since
it sends Ea.Eb = Eab to <p(ab) = <p(a) <p(b). 0
Let S and T be semigroups. By Theorem 2.2, S is isomorphic to a subsemi-
group ofT if and only if there exists an injective homomorphism (an embedding)
of S into T; then S can be embedded into T. Similarly, T is isomorphic to
a quotient semigroup of S if and only if there exists a surjective homomorphism
of S onto T; then T is a homomorphic image of S.
5. Theorem 2.2 can be deduced from more general results which allow one
homomorphism to factor through another and help construct diagrams of semi-
groups and homomorphisms.
Proposition 2.3. Let <p : S ----+ T and 'ljJ : U ----+ T be homomorphisms
of semigroups. If <p is injective, then '1/J factors through <p ( '1/J = <p 0 e for some
homomorphism e:
U ----+ S) if and only if Im 'ljJ ~ Im <p; and then '1jJ factors
uniquely through <p (e is unique). If <p and '1jJ are injective and Im 'ljJ = Im <p,
e
then is an isomorphism.
This is clear.
Proposition 2.4. Let <p : S ----+ T and 'ljJ : S ----+ U be homomorphisms of
semigroups. If <p is surjective, then '1/J factors through <p ( '1/J = e 0 <p for some
homomorphism e:
T ----+ U) if and only if ker <p ~ ker '1jJ; and then '1jJ factors
uniquely through <p (t;, is unique). If <p and 'ljJ are surjective and ker <p = ker 'ljJ,
then t;, is an isomorphism.
S~T
~1 ~
u
Proof. If 'ljJ = eo<p, then <p(a) = <p(b) implies 'tj;(a) = e(<p(a)) = e(<p(b)) =
'tj;(b), and ker <p ~ ker 'ljJ.
Conversely, assume that <p is surjective and that ker <p ~ ker 'ljJ. Let e be
the set of ordered pairs
e = { (<p(a), 'tj;(a)) E T Xu I a E S}.
e,
For every t E T, there exists u E U such that (t, u) E since <p is surjective; if
moreover (t,u) E ~' (t 1 ,u1) E e,
and t = t', then u = u', since ker<p ~ ker'lj;.
Thus ~ is a mapping ofT into U. Also ~ (cp(a)) = 1/J(a) for all a E S by

definition and
~(cp(a) cp(b)) = ~(cp(ab)) = 1/J(ab) = 1/J(a) 1/J(b) = ~(cp(a)) ~(cp(b))
for all a, b E S, so that ~ is a homomorphism. Thus 1/J factors through cp; 1/J
factors uniquely through cp since any mapping x such that 1/J = x o cp must contain
all ordered pairs (cp(a), 1/J(a)) and must coincide with ~· If moreover 1/J is
surjective and ker cp = ker 1/J, then ~ is injective, since ~ (cp( a)) = ~ (cp(b)) implies
1/J(a) = 1/J(b) and cp(a) = cp(b), is surjective, since Im~ = Im(~ocp) = Im'ljJ,
and is an isomorphism. D
Analogues of the (other) two Isomorphism Theorems also hold for semigroups.
The most useful employ the following constructions. Let cp : S --+ T be a
semigroup homomorphism. The direct image under cp of a subsemigroup S' of
S is the subset cp( S') = { cp( x) I s E S'} of T. The inverse image under cp of
a subsemigroup T 1 of T is
cp- 1 (T 1) = {xES I cp(x) E T}

Proposition 2.5. Let cp : S --+ T be a homomorphism of semigroups and
e = ker cp.
If S' is a subsemigroup of S, then cp( S') is a subsemigroup of T.
lfT' is a subsemigroup ofT, then cp- 1 (T 1 ) is a subsemigroup of Sand a
union of e-classes.
If cp is surjective this defines an order preserving one-to-one correspondence
between subsemigroups ofT and subsemigroups of S that are unions of e-clas-
ses.
Proof. If S' is a subsemigroup of S, then cp( S') is a subsemigroup of T,
since cp(x) cp(y) = cp(xy) E cp(S') for all x,y E S'.
If T' is a subsemigroup of T, then cp -l (T') is a union of e-classes and is

a subsemigroup of S since cp( x), cp(y) E T' implies cp (xy) = cp( x) cp(y) E T'.
If cp is surjective, then cp(cp- 1 (T')) = T' for every T' ~ T. Also S' ~
cp- 1 (cp(S')) for every S' ~ S; conversely, x E cp- 1 (cp(S')) implies cp(x) = cp(s)
for some s E S' and x E S' if S' ~ S is a union of e-classes. D
Similarly, the direct image under a semigroup homomorphism cp : S --+ T
of a congruence e on S is the binary relation
cp(e) = { (cp(a), cp(b)) E TxT I (a, b) E e };
equivalently, the direct image of c ~ S x S under cp x cp : S x S ---+ T x T.

The inverse image under cp of a congruence :r on T is the binary relation
cp- 1 (:!) = {(a, b) E S x S I (cp(a), cp(b)) E :!};
equivalently, the inverse image of :r under cp x cp . We also say that cp (c) ,
cp - 1 ( :J) are induced by c and :J.
Proposition 2.6. Let cp : S ---+ T be a homomorphism of semigroups and
e= ker cp.
If :r
is a congruence on T, then cp- 1 (:!) is a congruence on S which
contains e; if cp is surjective, then Slcp- 1 (:!) ~ TI:J.
If cp
is surjective and c is a congruence on S which contains e, then cp (c)
is a congruence on T, and Tlcp(c) ~Sic.
If cp
is surjective this defines an order preserving one-to-one correspondence
between congruences on T and congruences on S that contain e.
Proof. Let :r be a congruence on T and p : T ---+ T I:J be the projection,
so that :r = ker p. We see that cp - 1 ( :J) = ker (p o cp) . Therefore cp - 1 ( :J) is a
congruence on S. If cp is surjective, then S I cp - 1 ( :J) ~ Im (p o cp) = Imp = T I :r
by Theorem 2.2.
Now let cp be surjective and c be a congruence on S which contains e.
Let p : S ---+ SIc be the projection. By Proposition 2.4, p factors through cp:
p = ~ o cp for some homomorphism ~ : T ---+ SIc . We have cp (c) = ker ~ : indeed
(a, b) E c implies ~(cp(a)) = p(a) = p(b) = ~(cp(b)) and (cp(a), cp(b)) E ker~; if
conversely (t,u) E ker~, then t = cp(a), u = cp(b) for some a,b E S, (a,b) E c
since p(a) = ~(t) = ~(u) = p(b), and (t,u) = (cp(a), cp(b)) E cp(c). Therefore
cp (c) is a congruence; by Theorem 2.2, T I cp (c) ~ Im ~ = Imp = SIc, since
cp is surjective.
If cp is surjective, then cp x cp is surjective; therefore cp (cp- 1 (:!)) = :r for
all :r ~ TxT. Similarly c ~ cp- 1 (cp(c)) for all c ~ S x S. If c is a
congruence on Sande~ c, and (a,b) E cp- 1 (cp(c)), then (cp(a), cp(b)) E
cp(c), (cp(a), cp(b)) = (cp(c), cp(d)) for some (c,d) E c, and, as above, (a, c) E c
and (b,d) E c; hence (a,b) E c, so that cp- 1 (cp(c)) =c. D
It follows from Propositions 2.5, 2.6 that the subsemigroups of a quotient
e
semigroup SI are precisely the sets of e-classes whose unions are subsemigroups
of S; and that the congruences on Sle are precisely the congruences induced on
se s
1 by congruences on that contain e.
6. Similar results hold for monoids. When S and T are monoids, a

homomorphism of monoids of S. into T is a homomorphism of semigroups
<p: S-----+ T such that <p(l) = 1. Then <p preserves all products and nonnegative
powers.
When S is a monoid and C. is a congruence on S, then
El.Ea = Ela = Ea = Eal = Ea.El
for all a E S, so that SIC. is a monoid and the projection S -----+ SIC. is
a homomorphism of monoids. If therefore <p : S -----+ T is a homomorphism
of monoids, then, as in Theorem 2.2, Im <p is a submonoid of T, ker <p is a
congruence on S, and there is an isomorphism S lker <p -----+ lm <p such that the
diagram
S~T
1 r
S lker <p ----+ Im <p
commutes; in particular S lker <p 9:! Im <p.
Results similar to Propositions 2.3, 2.4, 2.5, and 2.6 also hold for monoids;
this is left to the reader.
7. We complete this section with some properties of congruences.
Proposition 2. 7. An equivalence relation e on a commutative semigroup S
is a congruence if and only if, for all a, b, e E S, a e b implies ae e be.
Proof. If this condition holds, then a e b, e e d implies ae e be= eb e db=
bd, and e is a congruence. The converse is clear. D
For instance the equality = on a semigroup S is a congruence, and so is the
universal congruence U, of which S is the only equivalence class; SI= ~ S,
whereas S IUis trivial.
Since congruences on a given semigroup S are subsets of S x S, we can form
their unions and intersections in S x S. The following result is straightforward.
Proposition 2.8. Let S be a semigroup. Every intersection of congruences
on S is a congruence on S. The union of a chain of congruences on S is a
congruence on S.
In particular, the empty intersection niE0 ei of congruences on S can
be defined as the universal congruence on S; the empty union UiE0 ei of
congruences on S can be defined as the equality on S.
By Proposition 2.8 there is for every binary relation ::R ~ S x S a smallest
3. IDEALS. 13
congruence e on S which contains ~; e is the intersection of all the congruences

which contain ~ and is the congruence generated by ~.
Proposition 2.9. Let S be a commutative semigroup. The congruence e
generated by ~ ~ S x S can be constructed as follows. Let
s = {(xu, yu) I x,y E s, u E 8 1 ' and X~ y or y ~X }.
Then a e b if and only if there exist n ~ 1 and s 1 , ... , sn E S such that a = s 1 ,

sn = b, and si S si+l for all 1 ~ i < n.
Proof. We see that S contains ~ (let u = 1 E 8 1 in the definition of
S ), is symmetric (a S b implies b S a), and admits multiplication (a S b implies
ac S be). Hence e is symmetric (reverse the sequence s 1 , ... , sn in the definition
of e) and admits multiplication. Moreover e contains the equality on S (let
n = 1 in the definition), contains S (let n = 2), and is transitive. Thus e is a
congruence and contains ~.
Conversely, a congruence which contains ~ must contain S, since a congru-
ence is symmetric and admits multiplication; and a congruence which contains S
must contain e, since a congruence is reflexive and transitive. 0
Proposition 2.9 can be stated more simply as follows: every relation a e b
follows from relations x ~ y by finitely many applications of the following
inference rules: a~ b implies a e b; a e b implies au e bu, for all u E S;
a E 8 implies a e a; a e b implies be a; a e b, be c implies a e c.
3. IDEALS.
1. An ideal of a semigroup S is a subset I of S such that a E I implies

ax E I and xa E I for all x E S; equivalently, such that IS ~ I and S I ~ I.
For instance, S and the empty set are ideals of S. If S is commutative, the
condition SI ~ I is sufficient.
Proposition 3.1. Every union of ideals of S is an ideal of S. Every
intersection of ideals of S is an ideal of S.
By Proposition 3.1 there exists, for every subset X of S, an ideal of S which
contains X and is contained in every ideal of S which contains X; this is the
ideal of S generated by X .
Proposition 3.2. In a commutative semigroup S, the ideal generated by a
subset X is the set 8 1 X of all multiples of elements of X.
Proof. 8 1 X is the product in 8 1 , which is contained in S since either

S 1 X = SX or 8 1 X = SX U lX = SX U X ~ S; contains lX = X; and
is an ideal of S since SS 1 X ~ 8 1 X. Conversely, an ideal which contains X
also contains SX and 8 1 X. 0
In particular (when S is commutative) the ideal generated by one element
a E S is the set S 1 a of all multiples of a; such ideals are called principal.
Proposition 3.3. Let S be a commutative semigroup. If K is a minimal
nonempty ideal of S, then K is a smallest nonempty ideal of S, and K is a
group.
Proof. Let I be a nonempty ideal. Since I and K are ideals, I K ~ In K
and In K is a nonempty ideal. Since InK ~ K it follows that K ~ I.
When a E K, then K a ~ K is an ideal of S; hence K a = K. In particular
ea = a for some e E K. Since every element of K has the form ax for some
x E K it follows that e is an identity element of K. Then every element a of
K has an inverse in K, since ab = e for some b E K, and K is a group. 0
The smallest nonempty ideal of S, when it exists, is the kernel of S. Every
finite commutative semigroup has a kernel; N does not.
Proposition 3.4. Let !.p : S -----+ T be a homomorphism of semigroups and
e = ker !.p.
lf cp is surjective and I is an ideal of S, then cp(I) is an ideal ofT.
If J is an ideal ofT, then cp - 1 ( J) is an ideal of S and a union of e-classes.
If !.p is surjective this defines an order preserving one-to-one correspondence
between ideals of T and ideals of S that are unions of e-classes.
Proof. If !.p is surjective and I is an ideal of S, then ~.p(I) is an ideal ofT,
since ~.p(x) ~.p(y) = ~.p(xy) E ~.p(I) for all xES andy E I.
If J is an ideal of T, then !.p - 1 ( J) is a union of e-classes and is an ideal
of S since ~.p(y) E J implies ~.p(xy) = ~.p(x) ~.p(y) E J for all xES.
If !.p is surjective, then !.p ( !.p - 1 ( J)) = J for every J ~ T, and tp - 1 ( ~.p( I)) = I
for every I~ S that is a union of e-classes, as in the proof of Proposition 2.5. 0
2. Congruences on a group are determined by normal subgroups. In a semi~
group, congruences are most easily constructed from ideals. The resulting quotient
semigroups, discovered by Rees [ 1940], are peculiarly different from quotient
groups and from quotient rings.
Proposition 3.5. When I is an ideal of a semigroup S, the relation J'
3. IDEALS. 15
defined by
a :J b -¢:::::::} a = b or a, b E I
is a congruence on S, the Rees congruence of the ideal I.
Proof. :J is an equivalence relation, and is a congruence since a = b and
c, d E I implies ac, bd E I; a, b E I and c = d implies ac, bd E I; and
a,b,c,d E I implies ac,bd E I. 0
The quotient semigroup S I I = S I:J is the Rees quotient of S by I. It is
standard practice to identifY the :J -class { x} E S I I of each x ¢ I with x E S.
If I= 0, then SII = S. If I i= 0, the :J-class IE SII is a zero element and
is denoted by 0; then S I I = (S\I) U {0} with the multiplication . in which 0
is a zero element and
if xy ¢I
x.y = {~yES
if xy E I
for all x, y E S\I. Thus the Rees quotient is obtained by squeezing I to a zero
element (if I i= 0) and leaving S\I untouched.
3. The Rees quotient can be viewed as the completion of a partial semi group
into an authentic semigroup. In general a partial binary operation on a set P
is a mapping f.L : D ----+ P whose domain D is a subset of P x P : when
x, y E P, p(x, y) is defined when (x, y) E D and is undefined otherwise. In
the multiplicative notation, p( x, y) is denoted by xy. A partial semigroup is
a set P together with a partial binary operation on P which is associative in
the sense that x(yz) = (xy)z holds whenever x, y, z E P and both x(yz) and
(xy )z are defined. (Other associativity conditions have been considered; see the
book by Lyapin & Evseev [1997].)
When P and Q are partial semigroups, a partial homomorphism of P into
Q is a mapping cp : P ----+ Q which preserves existing products: cp (xy) =
cp( x) cp(y) whenever xy is defined in P. If Q is a commutative semi group, then
the set PRom (P, Q) of all partial homomorphisms of P into Q is closed under
pointwise addition and is a commutative semigroup; if P is an actual semigroup,
then PHom(P,Q) = Hom(P,Q).
Every subset A of a semigroup S is a partial semigroup for the partial
operation . induced by S in the obvious way: when x, y E A, then x. y is
defined in A if and only if xy E A, and then x . y = xy. When I is a non empty
ideal of S, the Rees quotient S I I is obtained from the partial semi group S\I
by adjoining a zero element and setting all undefined products to 0.
4. An ideal extension of a semigroup S by a semigroup Q with zero is a
semigroup E such that S is an ideal of E and Q is the Rees quotient Q = E / S.

Ideal extensions were first studied by Clifford [ 1950].
The ideal extension problem, first considered by Clifford [ 1950], consists
in constructing all ideal extensions of a given semigroup S by a given semigroup
Q with zero; one may assume S n Q = 0. This difficult problem is discussed in
some detail in Clifford & Preston [ 1961 ], Grillet [ 1995], and especially Petrich
[ 1973]. The particular case of mono ids has a very nice solution, due to Clifford
[ 1950]; another case will be seen in Chapter II. More general results are known
but have had few applications to commutative semigroups.
When S is a subsemigroup of E, a retraction of E onto S is a homomor-
phism of E into S which is the identity on S.
Proposition 3.6. Every ideal extension of a monoid S has a retraction
a r-----+ ea = ae, where e is the identity element of S.
Proof. Let e be the identity element of S. In E we have ea = (ea) e =
e (ae) = ae for all a E E, since ea and ae are in S. Let
'!fJ(a) = ea = eae = ae E S
for all a E E. Then '1/J(a) =a when a E S, and '1/J(ab) = eabe = '1/J(a) '1/J(b)
for all a, b E E. D
An ideal extension E of S by Q is a retract ideal extension when there
exists a retraction 'ljJ of E onto S. Then the restriction <.p : Q\ 0 ---+ S of 'ljJ to
Q\0 = E\S is a partial homomorphism, and the operation on E is determined
as follows by the operation on S, the partial operation on Q\ 0, and the partial
homomorphism <.p. If a,b E Q\0 and ab =F 0 in Q, then ab is the same in Q
and E. If a,b E Q\0 and ab = 0 in Q = EjS, then, in E, abE Sand ab =
'ljJ (ab) = <.p(a) <.p(b). If a E Q\0 and x E S, then, in E, ax= 'ljJ (ax) = <.p(a) x
and xa = 'ljJ (xa) = x <.p(a). If x, y E S, then xy is the same in S and E.
Lemma 3.7. Let S be a semigroup and Q be a semigroup with zero such
that S n Q = 0. If <.p is a partial homomorphism of Q\0 into S, then the
disjoint union E = S U (Q\0), with the multiplication * defined by
a*b abE Q ifab # 0 in Q,

a*b <.p( a) <.p(b) ifab = 0 in Q,
a*y <.p(a) y,
X*b x<.p(b),
X*Y xy E S
4. DIVISIBILITY 17
for all a, bE Q\ 0 and x, y E S, is a retract ideal extension of S by Q, and every

retract ideal extension of S by Q can be constructed in this fashion. Moreover,
E is commutative if and only if S and Q are commutative.
Proof. Associativity in E follows in a long but straightforward manner from
associativity in S, associativity in Q, and the hypothesis that <p is a partial
homomorphism. Then it is clear that S is an ideal of E and that E / S = Q.
Moreover the mapping '1/J : E --+ S defined by
'1/J(a) = <p(a) for all a E Q\{0}, '1/J(x) = x for all xES,
is a retraction of E onto S. Conversely we saw that every retract ideal extension
can be constructed as in the statement. D
The ideal extension constructed in Lemma 3. 7 is determined by the partial
homomorphism <p. Thus, an ideal extension is a retract ideal extension if and
only if it is determined by a partial homomorphism:
Proposition 3.8. Every ideal extension of a monoid S by a semigroup Q
with zero is a retract ideal extension and is therefore determined by a partial
homomorphism of Q\ {0} into S, namely a f---.+ ea = ae, where e is the identity
element of S.
This follows from Lemma 3.7 and Proposition 3.6.
4. DIVISIBILITY.
1. A preorder (also called quasiorder) is a binary relation which is reflexive

and transitive; thus, a preorder which is also antisymmetric is an order relation.
On a commutative semigroup S the Green's preorder ~:J-C (also called the
divisibility preorder) is defined by
a ~:J-C b ~ a= tb for some t E S 1 ~ S 1a ~ S 1 b.

if for example e and f are idempotents, then e ~:J-C f if and only if e ~ f in
the Rees order: indeed ef = e implies e ~:J-C f; conversely, e = tf implies
ef = tf f = tf = e.
Green's relation JC is one of several relations introduced for semigroups in
general by Green [1951]:
a JC b ~ a ~:J-C b and b ~:J-C a ~ S 1 a = S 1 b.

Proposition 4.1. In a commutative semigroup, JC is a congruence.
Proof. If a JC b, then a = tb and b = ua for some t, u E S 1 , ae = tbe and

be = uae for some t, u E S 1 , and ae JC be. 0
In a sense, JC measures the extent of group-like behavior in semigroups. Thus,
multiplication by elements of a semigroup induces permutations of its JC-classes:
Lemma 4.2. Let H be an JC -class and t E 8 1 . If tH s;;; H, then the
mapping 9t : x 1-----t tx of H into H is bijective.
Proof. Let hE H. Then thE H, h = uth for some u E 8 1 , and uH s;;; H,

since JC is a congruence. If a E H, then a = hv for some v E S 1 and
tua = uta= uthv = hv = a; thus 9t and 9u are mutually inverse bijections of
H onto H. 0
2. Actual groups arise from JC in two ways. A subgroup of a semigroup S
is a subsemigroup G of S which happens to be a group. Necessarily the identity
element of G is an idempotent of S. Conversely, every idempotent e of S yields
a trivial subgroup {e} of S. Less trivially:
Proposition 4.3. For an JC-class H of a commutative semigroup S the
following conditions are equivalent:
(1) abE H for some a,b E H;
(2) H is a subsemigroup of S;
( 3) H contains an idempotent;
(4) H is a subgroup of S.
Proof. (1) implies (2) since JC is a congruence: if a, b, and abE H, then
x,y E H implies x JC a, y JC b, xy JC ab, and xy E H.
Assume that H is a subsemigroup and let a E H. Then aH s;;; H; by
Lemma 4.2, ga : x 1-----t ax is a permutation of H. In particular ae = a for
some e E H. Then a = ae = aee and e2 = e, since ga is injective. Thus (2)
implies (3).
Now assume that H contains an idempotent e. Then a, b E H implies
a JC e, b JC e, ab JC ee = e since JC is a congruence, and ab E H; thus H is
a subsemigroup of S. For every a E H we have a= te for some t E 8 1 and
ae = tee = te = a. Moreover aH s;;; H, ga : x 1-----t ax is a permutation of H
by Lemma 4.2, and ab = e for some b E H. Hence H is a group. Thus (3)
implies (4); and (4) implies (1). 0
If for instance S is a monoid, then the elements of H 1 are the units of S
and H 1 is the group of units of S.
4. DIVISIBILITY 19
Proposition 4.4. In a commutative monoid S, S\H1 is an ideal.

Proof. If y E S is not a unit, then there cannot exist u E S such that uxy = 1,
and xy is not a unit. D
In general:
Corollary 4.5. The maximal subgroups of a commutative semigroup S
coincide with the 1-C -classes of S which contain idempotents. They are pairwise
disjoint. Every subgroup of S is contained in exactly one maximal subgroup.
Proof. If G is a subgroup of S and e is the identity element of G, then
every x E G satisfies ex= x and xy = e for some y = x- 1 E G ~ S; hence
G ~He. D
The history of Corollary 4.5 goes back to Schwarz [1943] for torsion semi-
groups and to Wallace [1953] and Kimura [1954] for semigroups in general.
3. In fact Lemma 4.2 yields a group for every 1-C-class H. Let
st (H) = { t E s 1 1 tH ~ H}
denote the (left) stabilizer of H. For every t E St (H), Lemma 4.2 provides a
bijection gt : H ---t H, x t----+ tx.
Proposition 4.6. For every 1-C-class H, r(H) = {gt I t E St (H)} is a

simply transitive group ofpermutations of H, and t t----+ gt is a homomorphism
ofSt (H) onto r(H). If H =He is a maximal subgroup of S, then r(H) ~H.
Proof. First gt(gu(x)) =tux= gtu(x) for all t,u E St(H) and x E H;
thus t t----+ gt is a homomorphism and r(H) is a semigroup (under composition).
Also 1 E St (H) ~ 8 1 and g 1 = 1H is the identity mapping on H.
Let gt E r(H) and a E H. As in the proof of Lemma 4.2, a 1-C ta and
a = uta for some u E 8 1 ; in fact u E St (H), since 1-C is a congruence. For
every x E H we now have x = av and utx = utav = av = x for some v E 8 1 ;
thus gt has an inverse in r(H), namely gu.
If a, b E H, then b = ta for some t E 8 1 , t E St (H) since 1-C is a

congruence, and gt(a) = b; thus r(H) is transitive. In fact r(H) is simply
transitive: if gt (a) = gu (a), then ta = ua, tx = tav = uav = ux for every
x = av E H, and gt = gu.
If finally H = He is a maximal subgroup of S, with identity element e, then
H ~ St (H) and the homomorphism h t-----t gh of H into r(H) is bijective,
since gh(e) = h and r(H) is simply transitive. 0
r(H) is the (left) SchOtzenberger group of H; it was discovered by Schiit-

zenberger [1957].
5. FREE COMMUTATIVE SEMIGROUPS.
1. When a commutative semigroup S is generated by a subset X, every

element of S is a product of positive powers of one or more distinct elements of
X (Proposition 1.3) but can in general be written in this form in several ways.
For example X = S generates S, and then every equality ab = c in S equates
two distinct products of positive powers of one or more distinct elements of S.
A commutative semigroup S is free on a subset X when every element of
S can be written uniquely (up to the order of the terms) as a product of positive
powers of one or more distinct elements of X. For example, the multiplicative
semigroup { 2, 3, ... , n, . .. } ~ N is free (as a commutative semigroup) on the set
of all prime numbers. In the additive notation, powers become positive integer
multiples; the additive semigroup N+ is free on {1}.
2. For every set X we now construct a commutative semigroup Fx which
is free on X.
Fx is one of the few commutative semigroups that we prefer to denote addi-
tively. Then products of positive powers of distinct elements of X become sums
of positive integer multiples of distinct elements of X, that is, (finite) linear
combinations of elements of X with coefficients in N+ . This suggests that we
retrieve Fx from the free abelian group Gx on X, which consists of all linear
combinations a = l::xEX ax x with integer coefficients ax E Z that are almost
all zero (that is, { x E X I ax i= 0} is finite). (Linear combinations l::xEX ax x
can be defined more formally as suitable families (ax)xEX of integers.) Addition
on Gx is coordinatewise:
l::xEX axx + l::xEX bxx = l::xEX (ax+ bx)x ·

Gx is a partially ordered group, as the coordinatewise partial order
'ExEX ax x ~ 'ExEX bx x if and only if ax ~ bx for all x E X

is compatible with the operation (if a ~ b, then a+ c ~ b +c).
Fx is the positive cone of Gx, which is a subsemigroup of Gx:
Fx={aEGxla>O};
5. FREE COMMUTATIVE SEMIGROUPS. 21
equivalently, Fx is the set of all linear combinations a = l:xEX ax x with

integer coefficients ax such that ax = 0 for almost all x, ax ~ 0 for all x,
and ax > 0 for some x. Note that a ~:J-C b in Fx if and only if a ~ b in the
coordinatewise partial order.
Every y E X can be written as a linear combination y = I:xEX ax x E Fx
in which ay = 1 and ax = 0 for all x #- y; thus X ~ Fx . Now every element
of Fx can be written uniquely (up to the order of the terms) as a nonempty sum
of positive integer multiples of distinct elements of X; hence
Proposition 5.1. For every set X, Fx is a commutative semigroup which
is free on X.
Sometimes it is better to denote Fx multiplicatively; then every element of
Fx is uniquely (up to the order of the terms) a non empty product a = ITxEX xax
of positive integer powers of distinct elements of X (with ax = 0 for almost all
x, ax ~ 0 for all x, and ax > 0 for some x).
3. The most important property of Fx is its universal property:
Theorem 5.2. Every mapping f of X into a commutative semigroup S
extends uniquely to a homomorphism <p of Fx into S, namely
'P(l:xEX axx) = ITxEX f(xtx ·

The image of <p is the subsemigroup of S generated by f (X). If S is generated
by f (X), then <p is surjective. If S is free on X, then S is isomorphic to Fx.
x-SFX
~ 1~
s
If Fx is denoted multiplicatively, then <p (ITxEX xax) = ITxEX f(x )ax.
Proof. A homomorphism <p transforms sums into products and transforms
linear combinations into products of powers:
<p (al xl + a2x2 + ... + anxn) = c.p(xltl c.p(x2t2 ... c.p(xn)an .

If <p : Fx --+ S extends f (if c.p(x) = f(x) for all i), then, for every a =
l:xEX axx E Fx,
'P(l:xEX axx) = ITxEX c.p(xtx = ITxEX f(xtx

(= ITxEX,ax#O f(xtx, which is a finite product). Hence <p is unique.
Conversely define a mapping t.p : Fx ~ S by
t.p(L:xEX axx) = flxEX f(xtx

Then t.p extends f and is a homomorphism:
t.p ( ~
uxEX ax X +~uxEX bx X ) = flxEX j(x)ax+bx
= ITxEX f(xtx f(x)bx (ITxEX f(xtx) (ITxEX f(x)bx).
By Proposition 1.3,
Im t.p = {flxEX f(xtx I a E Fx}

is the subsemigroup of S generated by f (X) . If S is free on X and f :X ~ S
is the inclusion mapping, then t.p is an isomorphism. 0
It follows from Theorem 5.2 that all commutative semigroups that are gener-
ated by a set X are homomorphic images of Fx . Since every semigroup S is
generated by some subset X S:: S (for instance, by X = S ), we have:
Corollary 5.3. Every commutative semigroup is a homomorphic image of a
free commutative semigroup. Every finitely generated commutative semigroup is
a homomorphic image of a finitely generated free commutative semigroup.
Commutative semigroups can thus be explored by means of congruences on
free commutative semigroups. This approach was pioneered by Redei [ 1956] and
will be explored in later chapters, and in Proposition 5.8 below.
4. Free commutative semigroups have certain finiteness properties:
Proposition 5.4. Every free commutative semigroup F satisfies the de-
scending chain condition. lf F is finitely generated, then every antichain of F
is finite.
An antichain is a subset A which does not contain elements a < b.
Proof. By the last part of Theorem 5.2 it suffices to prove these properties
for Fx.
When a = 2:xEX ax x E F the positive integer Ia I = 2:xEX ax is the
length of a. If a < b in F, then ax ;£ bx for all x E X, ax < bx for
some x E X, and Ia I < lbl. There cannot exist an infinite descending sequence
a1 > a2 > · · · >an > an+l > ·· · of elements ofF, for then la 1 1> la2 1>
· · · > Ian I > Ian+ 1 1> · · · would be an infinite descending sequence of positive
integers.
Now assume that X is finite. We prove by induction on the number of
5. FREE COMMUTATIVE SEMIGROUPS. 23
elements of X that every antichain of Fx is finite. If X is empty, then Fx is

empty and so is every anti chain of Fx. If X has just one element, then Fx ~ N+
is a chain and an antichain of Fx has at most one element.
Let X have more than one element and A be an antichain of Fx. For every
y E X and n ~ 0 let Ay,n = {a E A I ay = n}. Then
r~=xEX\{y} axx I a E Ay,n}
is an antichain of F X\ {y} and Ay,n is finite by the induction hypothesis.
For every x EX let m(x) =min (ax I a E A) and Mx = {a E A I ax =

m(x)}. By the above, M = UxEX Mx is finite. Let n(x) =max (ax I a EM).
Then n(x) ~ m(x), since ax= m(x) for some a EM. If a E A, then ax~ m(x)
for all x E X and ax ~ n(x) for some x E X, otherwise ax > n(x) for all
x E X, a > b for all b E M ~ A, and A is not an antichain. Hence
A ~ UxEX, m(x);Sn;Sn(x) Ax,n is finite. D
The second half of Proposition 5.4 is known as Dickson's Theorem, after
Dickson [ 1913] who proved it for the free multiplicative subsemigroups of N
generated by finitely many primes. A different proof will be given in Chapter VI
along with additional finiteness properties.
5. A commutative monoid S is free on a subset X (as a monoid) when
every element of S can be written uniquely (up to the order of the terms) as a
product of positive powers of distinct elements of X (Proposition 1.4).
The nonnegative cone of Gx is
Fx U { 0} = {a E Gx Ia ~ 0};
equivalently, Fx U {0} is the set of all linear combinations a= l:xEX axx with
integer coefficients ax such that ax = 0 for almost all x and ax ~ 0 for all x.
Every element of Fx U {0} can be written uniquely (up to the order of the terms)
as a sum of positive integer multiples of distinct elements of X; hence Fx U { 0}
is a free commutative monoid on X. If X is finite, with n elements, then Fx
is isomorphic to the direct product Nn. The universal property of Fx U {0} is:
Proposition 5.5. Every mapping f of X into a commutative monoid S
extends uniquely to a monoid homomorphism cp of Fx U {0} into S.
Corollary 5.6. Every (finitely generated) commutative monoid is a homo-
morphic image of a (finitely generated) free commutative monoid.
In later chapters it will be more convenient to denote the free commutative
monoid by Fx; then the free commutative semigroup on X is Fx \ {0}.

Similarly we call Fx U { oo} the free commutative semigroup with zero on the
set X, since every element of Fx U { oo} is either the zero element oo or uniquely
a nonempty sum of positive integer multiples of distinct elements of X. Every
mapping f of X into a commutative semigroup S with zero extends uniquely
to a semigroup homomorphism cp : Fx U { oo} -----+ S such that c.p( oo) = 0.
In the multiplicative notation, Fx U {0} and Fx U { oo} become Fx U {1}
and Fx U {0} , respectively.
6. As an application of free commutative semigroups we construct all cyclic
semigroups. By Proposition 1.3, a cyclic semigroup S consists of all the powers
of its generator x, and is necessarily commutative; hence S is isomorphic to
the quotient of F{x} by some congruence. Now every element of F{x} can be
written uniquely in the form nx with n E w+ ; hence F{ X} ~ w+ . Thus a cyclic
semigroup is isomorphic to the quotient of w+ by some congruence.
Let e be a congruence on w+ . If e is not the equality on w+ , the least
integer r > 0 such that r e t for some t of= r is the index of e. Then the least
integer s > 0 such that r e r + s is the period of e.
Lemma 5.7. When e is a congruence on w+ of index r and period s,
then a e b if and only if either a = b < r, or a, b ~ r and a = b mod s.
Proof. Since e is a congruence, r e r + s implies r e r + s e r +
2s e ... e r + ks for all k > 0' u + r e u + r + ks for all k > 0' u ~ 0'
and a e b whenever r ;;:; a ;;:; b and a = b mod s .
Conversely assume a e b with a < b. Then a ~ r by the choice of r. There
is an integer u ~ 0 such that u + a = r mod s, and an integer k ~ 0 such
that a + ks < b ;;:; a + ks + s. Then t = b - a - ks satisfies 0 < t ;;:; s and
a + ks e a e
b = a + ks + t; hence
r e u + a e u + a+ ks e u + a + ks + t e r + t.
Since 0 < t ;;:; s it follows from the choice of s that t = s. Then b = a+ ks + s =
a mods. 0
Proposition 5.8. Let S be a cyclic semigroup, generated by x E S. Either
S ~ w+, or S is finite and there exist integers r, s > 0 (the index and period
of x) such that xi = xj if and only if either i = j < r, or i,j ~ r and
i = j mod s; then every element of S can be written uniquely in the form xi
with 1 ;;:; i < r + s and
6. PRESENTATIONS. 25
if i + j < r + s,
if i + j ~r + s, where
r ~ k < r + s and k =i +j mod s;
and {xr, xr+ 1 , ... , xr+s- 1 } is a cyclic subgroup of S.
Proof. We have S ~ N+ je for some congruence e on w+. If e is the
equality, then S ~ w+. Now assume that e
is not the equality. As before, e
has index r > 0 and periods> 0. By Lemma 5.7, thee-class of a< r is {a};
thee-classes of r,r + l, ... ,r + s -1 are distinct (and infinite); and these are
all the e-classes. Since a = b mod s implies a e b when a, b ~ r the operation
on S is as described in the statement.
Finally, G = {xr, xr+ 1 , ... , xr+s- 1 } is a subsemigroup of S and we see
from the multiplication on S that G ~ Z/ s7l., the additive group of integers
modulo s. 0
Proposition 5.8 was first stated (for cyclic semigroups of subsets of a group)
by Frobenius [1895], and its Corollary 5.9 below, in its present form, by Moore
[1902]. Lemma 5.7 was rediscovered by Chacron [1982]. Tamura [1963] deter-
mined all congruences on Q+ .
Corollary 5.9. Every nonempty finite semigroup contains an idempotent.
Proof. If S is finite nonempty, then S contains a finite cyclic subsemigroup,
which by Proposition 5.8 contains a subgroup and its identity element. 0
6. PRESENTATIONS.
Corollary 5.4 suggests that commutative semigroups can be constructed by

presentations (= by generators and relations).
1. For this it is more convenient to denote free commutative semigroups
multiplicatively. Formally, a commutative semigroup relation between elements
of a set X is an ordered pair (u, v), normally written as an equality u = v,
of elements of Fx . (Relations are readily distinguished from actual equalities in
Fx , since the latter are all trivial.)
When f is a mapping of X into a commutative semigroup S, we say that
the relation u = v holds in S via f in case the equality cp( u) = cp( v) holds in
S, where cp : Fx ---t S is the homomorphism which extends f.
These somewhat abstract definitions make most sense when X is a subset of

S and f : X ---+ S is the inclusion mapping; then cp sends a product I1xEX xax
of elements of X as calculated in Fx to the same product I1xEX xax calculated
in S; hence the relation I1xEX xux = I1xEX xvx holds in S if and only if the
products I1xEX xux = I1xEX xvx are equal in S.
2. When X is a set and ::R ~ Fx x Fx is a set of relations between the
elements of X, we denote by (X I ::R) the quotient of the free commutative
semigroup Fx by the congruence e generated by ::R. By Proposition 2.9, e
consists of all the "obvious consequences" of the relations in ::R.
(X I::R) comes with a canonical mapping L : X ---+ (X I::R) which is the
composition
~,: X ~ Fx ---+ Fxle =(X I::R).

Proposition 6.1. (X I::R) is generated by L(X) and every relation (u, v) E ::R
holds in (X I::R) via L.
Proof. (X I::R) is generated by L( X), since Fx is generated by X. Moreover
the projection Fx ---7 Fx I e = (X I::R) is the only homomorphism which extends
L; since ::R ~ e, every relation (u, v) E ::R holds in (X I::R) via L. D
Accordingly (X I ::R) is known as the commutative semigroup generated

by X subject to ::R. This is somewhat misleading since X is not a subset of
(X I::R) (in fact, L need not even be injective) and every homomorphic image of
(X I::R) has the properties in Proposition 6.1. However, (X I::R) is the "largest"
semigroup with these properties:
Proposition 6.2. Let X be a set and ::R be a set of relations between the
elements of X. Let S be a commutative semigroup and f : X ---+ S be a
mapping such that every relation u = v in ::R holds in S via f. There is a
unique homomorphism cp : (X I::R) ---+ S such that f = cp o L If S is generated
by f(X), then cp is surjective.
Proof. Let e be the congruence on Fx generated by ::R and 1r : Fx ---+

Fx I e = (X I ::R) be the projection. Let 'ljJ : Fx ---7 s be the homomorphism
which extends f. Then '1/J(u) = '1/J(v) for every relation u = v in ::R, since
u = v holds in S via f; hence ::R ~ ker 'ljJ and e ~ ker 'ljJ. By Proposition 2.4,
'ljJ factors through 1r : 'ljJ = cp o 1r for some homomorphism cp : (X I::R) ---+ S.
6. PRESENTATIONS. 27
Then r.p o t = f.
Let r.p 1 : (X I ~) --+ S be another homomorphism such that r.p 1 o t = f. Then

T = {a E (X I~) I r.p( a) = r.p1 (a)} is a subsemigroup of (X I~) which contains
t(X); since t(X) generates (XI~) it follows that r.p = r.p1 • 0
3. A presentation of a commutative semi group S consists of a set X, a
set ~ of commutative semigroup relations between the elements of X, and an
isomorphism S ~ (X I~). By Corollary 5.4, every commutative semigroup S
has a presentation, in which X can be any subset of S which generates S, and ~
can be any binary relation which generates the congruence induced by Fx --+ S.
For example let e be the congruence on N+ of index r and period s. By
Lemma 5.7, e is generated by (r, r + s). Therefore a finite cyclic semigroup S
of index r and period s has the presentation S ~ ( x I xr = xr+s ) .
Presentations are associated with a number of logical and computational prob-
lems: the word problem (deciding when two products of generators are equal);
the isomorphism problem (deciding when two presentations yield isomorphic
semigroups); and recognition problems (recognizing additional properties from a
presentation). Algorithms in Rosales & Garcia-Sanchez [ 1999] solve a number
of these problems.
4. Similar definitions apply to commutative monoids and to commutative
semigroups with zero. A commutative monoid relation between the elements of
a set X is an ordered pair ( u, v) (normally written as an equality u = v) of
elements of the free commutative monoid Fx U {1} (written multiplicatively); the
identity element of Fx U { 1} may appear as u or v. When S is a commutative
monoid, a commutative monoid presentation of S consists of a set X, a set ~ of
commutative monoid relations between the elements of X, and an isomorphism
S ~ (X I~), where (X I~) now denotes the quotient of Fx U {1} by the
congruence generated by ~.
A commutative relation with zero between the elements of a set X is an
ordered pair (u, v) (normally written as an equality u = v) of elements of the
free commutative semigroup with zero Fx U {0} (written multiplicatively); the
zero element of Fx U {0} may appear as u or v. When S is a commutative
semigroup with zero, a presentation of S as a commutative semigroup with zero
consists of a set X, a set ~ of commutative relations with zero between the
elements of X, and an isomorphism S ~ (X I~) , where (X I~) now denotes

the quotient of Fx U {0} by the congruence generated by ~-
Chapter II.
CANCELLATIVE SEMIGROUPS.
One of the oldest results in semigroup theory embeds cancellative commutative

semigroups into abelian groups, by a construction which also completes N into
Z and embeds integral domains into fields.
This chapter begins with two generalizations of this construction, semigroups
of fractions and universal groups. Then we tum to cancellative commutative
semigroups, which later chapters will show are a basic building block for more
general commutative semigroups.
For the sake of brevity we often abbreviate "commutative semigroup" as c.s.
and "commutative monoid" as c.m. in what follows.
1. SEMIGROUPS OF FRACTIONS.
The semigroups of fractions in this section are reminiscent of localization in

commutative rings and were already known to Vandiver [1940]. They will be
used in later chapters to construct completions. More general constructions are
noted at the end of this section.
1. Let S be a c.s., written multiplicatively, and C be a nonempty subsemi-
group of S. Under the componentwise multiplication (x, a) (y, b) = (xy, ab) ,
S X C is a C.S. ( 8 1 and C 1 may be used here instead of S and C, and yield
the same fractions.) Let = be the binary relation on S ~ C defined by:
(x,a) =(y, b) if and only if cay = cxy for some c E C.
Since S is commutative, it is immediate that = is a congruence on S x C. The
fraction xja is the :=-class of (x,a). The semigroup of fractions 1c- s
is
the quotient semi group ( S x C)/= . By definition,
xja = yjb if and only if cay= cxy for some c E C
and (xja)(yjb) = xyjab, for all x,y E Sand a,b E C.
29
30 II. CANCELLATIVE SEMIGROUPS.
Proposition 1.1. Let S be a commutative semigroup and C be a nonempty

subsemigroup of S. The fractions x /a with x E S, a E C constitute a com-
mutative monoid c-Is, in which a/a= 1 for every a E C. A canonical ho-
momorphism a: S - - t c-Is is well defined by o:(x) =ax/a for any a E C.
For every xES and a E C, a( a) is a unit of c-Is and xja = o:(x) o:(a)-I.
Proof. First aja = b/b for all a,b E C, and (x/a)(b/b) = xb/ab = x/b for
all x E S. Hence c-IS has an identity element 1 and 1 = a/ a for all a E C.
Similarly ax I a = bx /b for all a, b E c, so that o:( X) E c-Is is well defined
by o:(x) = ax/a, for all x E S. For every a E C, a( a) = a 2 /a is a unit
since (a 2 ja)(aja 2 ) = a 3 ja 3 = 1, and o:(a)-I = aja 2 ; hence o:(x) o:(a)-I =
(axja)(a/a 2 ) = a2 xja3 = xja. D
An element e of a c.s. S is cancellative in S when ex = ey implies x = y,
for all x, yES. A c.s. S is cancellative when every element of S is cancellative
in S (when ex= ey implies x = y, for all x,y,e E S).
Proposition 1.2. Let S be a commutative semigroup and C be a nonempty
subsemigroup of S. The canonical homomorphism a : S - - t c-IS is injective
if and only if every element of C is cancellative in S.
Proof. If o: (x) = o: (y) , then ax/ a = by/ b and eaby = eabx for some
a, b, e E C; if every element of C is cancellative in S, then x = y. Conversely
assume that a is injective. If e E C and ex = ey, then e3 x = e3 y, o:(x) =
exje = eyje = o:(y), and x = y. D
If for instance S is cancellative, then the canonical homomorphism a : S - - t
s-I S is injective; this is the familiar embedding of a cancellative c.s. into an
abelian group, which is studied in more detail in the next section.
2. The semigroup of fractions c-IS and its canonical homomorphism enjoy
a universal property:
Proposition 1.3. Let T be a commutative semigroup and <p : S - - t T be
a homomorphism. If T is a monoid and <p(e) is a unit ofT for every e E C,
then <p factors uniquely through a: there exists a unique monoid homomorphism
~: c-Is - - t T such that~ o a= <p, namely, ~ (xja) = <p(x) <p(a)-I for all
xES and a E C.
1. 8EMIGROUPS OF FRACTIONS. 31
Proof. A monoid homomorphism ~ sends units to units and preserves inverses

of units: if u is a unit with inverse v = u- 1 , then uv = 1, ~(u)~(v) = ~(1) = 1,
and ~ (u) is a unit with inverse ~ (u - 1 ) . If therefore ~ : c- 1 S ----+ T is a monoid
homomorphism such that ~ o a = rp, then necessarily
~(xla) = ~(a(x) a(a)- 1 ) = ~(a(x)) ~(a(a))- 1 = rp(x) rp(a)- 1

whenever a E C, by Proposition 1.1.
If conversely xI a = y lb in c- 1S, then cbx = cay for some c E C and

rp(c) rp(b) rp(x) = rp(c) rp(a) rp(y); multiplying by rp(ar- 1 rp(b)- 1 rp(c)- 1 yields
rp(a)- 1 rp(x) = rp(b)- 1 rp(y). Therefore a mapping~: c- 1 s----+ T is well
defined by ~ (xla) = rp(x) rp(a)- 1 . Since rp (ab)- 1 = rp (a)- 1 rp (b)- 1 , it is
immediate that ~ is a monoid homomorphism. The first part of the proof then
shows that ~ is the only monoid homomorphism ~ : c- 1 S ----+ T such that
~oa = rp. 0
As a first application of Proposition 1.3 we note:
Proposition 1.4. lf S is a subsemigroup ofT, then c- 1 s is isomorphic

to the subsemigroup {xla E c- r
1 I XES} ofC- 1T.
Proof. Let rp : S ~ c- 1 T be the restriction to S of the canonical ho-
momorphism j3 : T ----+ c- 1T. Proposition 1.3 yields a homomorphism 1/J :
c- 1 S----+ c- 1T which sends xla E c- 1 8 to rp(x) rp(a)- 1 = j3(x) j3(a)- 1 =
xla E c- 1 T. If x,y E S and xla = ylb in c- 1T, then cay = cbx for
some c E C and xI a = y lb in c- 1S; thus 1/J is injective. We see that
Im 1/J = {X I a E c- 1T I X E S, a E c}. 0
If more generally S and C I 0 are subsemigroups of T, we denote the
subsemigroup {X I a E c-
1 T I X E S, a E c} by c- 1 s.
By Proposition 1.4,
this subsemigroup may be identified with the semigroup of fractions c- 1 in s
case C ~ S.
Proposition 1.5. lf I is an ideal of S which contains C, then c- 1 I =
c- 1 s.
Proof. Let a E C. For every xla E c- 1 s we have xla = xclac E c- 1I,
since xc E I and ac E C. 0
3. Semigroups of fractions have been studied, mostly following the analogy
with localization, by Maury [ 1958], Bouvier [ 1969], [ 1970], and Bouvier &
Faisant [1970].
Much more general semigroups of quotients have been defined for all semi-
groups, most notably by Berthiaume [1964], [1971], McMorris [1971], and Lue-
demann [1976], following similar constructions in ring theory (Johnson [1951];
Utumi [ 1956]; Findlay & Lambek [ 1958]). The author likes the very general
construction in Almkvist [ 1968] and StenstrOm [ 1970]. The ingredients are a
monoid S and a set J' of nonempty right ideals of S, such that: S E J'; I E J'
implies I: s = { x E S I sx E I} E J' for all s E S; and J' contains every
right ideal I such that I : s E J' for all s in some J E J'. Then J' is a filter
( J 2 I E J' implies J E J', and I, J E J' implies I n J E J'). A congruence
'J on S is defined by
x 'J y if and only if there exists I E J' such that xs = ys for all s E I.
The elements of 8'3' are equivalence classes of mappings f : I ---+ S /'J such that
IE J' and f(xs) = f(x)r(s) for all x E I and s E S, where T: S---+ S/'J
is the projection; f : I ---+ S and 9 : J ---+ S are equivalent if and only if f
and 9 agree on some K E J', K ~ I n J.
The semigroups c- 1 S are a particular case. When S is a commutative
semigroup and C is a subsemigroup of S, one may let J' be the set of all
nonempty ideals I of S 1 such that I : X intersects c
for every X E S 1 . Then
n = {s E s 1 s 1s E J"}
1 = {s E s 1 s 1s n c i= 0}
1
is a subsemigroup of S 1 (in fact D is the face of S 1 generated by C). Propo-

sition 6.8 of StenstrOm [1970] shows that S 1 /3=' 9:! n- 1s 1 ; it is readily verified
that n- 1s 1 ~ c- s.
1 ·
Other particular cases are more relevant for semigroups in general and have
been studied in commutative cases by McMorris [1972], Hinkle [1972], [1974],
Berthiaume [1973], Rompke [1975], and Johnson & McMorris [1977]. Semi-
groups of quotients were also constructed by Toea [1977], Fountain & Petrich
[1986], and Easdown & Gould [1996].
2. UNIVERSAL GROUPS.
Universal groups now dangle from the previous section, ripe for picking.
1. A universal group of a semigroup S is a group G(S), together with a
homomorphism 'Y : S ---+ G (S) , such that every homomorphism cp : S ---+ G
of S into a group G factors uniquely through 'Y (=there exists a unique group
2. UNIVERSAL GROUPS. 33
homomorphism ~ : G(S) ---+ G such that <p = ~ o "f).
S~ G(S)
~1~
G
This universal property implies that G(S) is unique up to isomorphism if it exists.
That every semigroup S has a universal group follows from the Adjoint
Functor Theorem. When S is commutative, Proposition 1.3 provides a simple
construction of G(S):
Proposition 2.1. When S is a nonempty commutative semigroup, then the
abelian group s- 1 S, together with the canonical homomorphism a : S ---+
s- 1 s, is a universal group of S: for every homomorphism <p : S ---+ G of
S into a group G there exists a unique group homomorphism ~ : s-
1 S ---+ G
such that <p = ~ o a, which sends xja E s- s

1 to <p(x) <p(a)- 1 E G.
Proof. The c.m. s- 1 S is a group since every x j a = a( x) a( a) - 1 is a

product of units. If <p : S ---+ G is a homomorphism of S into a group G,
then <p(S) generates a commutative subgroup H of G, in which every <p(x) is
a unit, and Proposition 1.3 yields a unique monoid homomorphism (also a group
homomorphism) ~ : s- 1 S---+ H ~ G such that <p = ~ o a, as in the statement. D
Because of Proposition 2.1 we use the group of fractions s- 1 S of S as its
universal group G(S) in what follows. In Section lll.2 we shall see that G(S)
has the following presentation: when written additively, G(S) is the abelian
group generated by the elements of S, subject to all relations a + b = ab with
a,b E S.
Proposition 2.2. Every homomorphism <p : S ---+ T of commutative semi-
groups induces a unique homomorphism G(<p) : G(S) ---+ G(T) such that the
square
S ~T
al
G(S) ----+ G(T)
r
G('P)
commutes, where a : S ---+ G(S) and {3 : T ---+ G(T) are the canonical
homomorphisms; G(<p) sends xja E G(S) to <p(x)j<p(a) E G(T).
Proof. By the universal property of G(S) there is a unique homomorphism
G(<p) such that G(<p) oa = {3o<p; from Proposition 2.1, G(<p) sends xja E G(S)
to f3(cp(x)) f3(cp(a))- 1 = cp(x)lcp(a). 0

2. Underlying the construction of universal groups is the following result. A
congruence e
on a c.s. S is cancellative when the quotient semi group SI is e
e
cancellative; equivalently, when ae be implies a b. e
Proposition 2.3. On every commutative semigroup S there is a smallest
cancellative congruence e, namely
x ey if and only if ex = ey for some e E S;
then Sle, together with the projection s ----t Sle, is a universal cancellative
semigroup of S, and G(Sie) is a universal group of S.
Proof. e is the congruence induced by the canonical homomorphism 'Y : S ----t
G(S); it is a cancellative congruence since Sle ~ Im 'Y <;;;: G(S) is cancellative.
If conversely e is a cancellative congruence on S, then the homomorphism
s ----t sI e ----t
G (sI e) factors through 'Y; hence e <;;;: e. Thus e is the
smallest cancellative congruence on S. (This can also be proved directly.)
Let 7f : s ----t sI e be the projection. If <p : s ----t T is a homomorphism of s
into a cancellative semigroup T, then Im cp is cancellative, ker cp is a cancellative
congruence, e
<;;;: ker cp, and cp factors uniquely through 7f (Proposition 1.2.4).
Thus sI e' together with the projection s ----t sI e' is a universal cancellative
semigroup of S.
Let G be a group and cp : S ----t G be a homomorphism. Since G is
e
cancellative, there is a unique homomorphism 1/J : SI ----t G such that cp = 1/J o 7f,
where 7f : s
----t se
I is the projection. Then there is a unique homomorphism
x : G(Sie) ----t G such that 1/J = X o {3, where {3 : Sle ----t G(Sie) is the
canonical homomorphisms. Therefore cp factors uniquely through {3 o 7f. Thus
G (sI e)' together with the homomorphism {3 0 7f' is a universal group of s. 0
r
s ~ s1e
~lY
G *-x G(Sie)
3. By Proposition 1.2, 'Y : S ----t G(S) is injective if and only if S is

cancellative. Semigroups for which 'Y is surjective were studied by McAlister &
O'Carroll [ 1971] and Tamura & Hamilton [ 1971]. The latter proved:
Proposition 2.4. 'Y: S ----t G(S) is surjective if and only iffor every x E S
there exist y, t E S such that xyt = t.
Proof. Assume that this condition holds and let xla E G(S). Then ayt = t
2. UNIVERSAL GROUPS. 35
for some y,t E S. Hence axyt = xt and xla = xytlt = 1(xy). Thus 1 is
surjective.
If conversely 1 is surjective and x E S, then 1( x) - 1 = I(Y) for some

yES, xlx 2= xylx, x2u = x 3 yu for some y,u E S, and xyt = t for some
y and t = x 2 u E S. D
If for instance S is finite, then 1 : S --+ G (S) is surjective: for every x E S,
some xn = e is idempotent, and xye = e, where y = xn- 1 if n ~ 2, y = e if
n = 1; then Proposition 2.5 below implies that G(S) is isomorphic to the kernel
of S. If on the other hand S = N, or more generally if S is cancellative but not
a group, then 1 is not surjective, since it is injective but not bijective.
4. For future reference we note the following properties.
Proposition 2.5. When I is a nonempty ideal of a commutative semigroup
S, then the inclusion homomorphism I--+ S induces an isomorphism G(I) ~
G(S).
Proof. By Proposition 2.2 the homomorphism ~ : G(I) --+ G(S) induced
by the inclusion I--+ S sends xla E G(I) to xla E G(S). Let a,b,x,y E I.
If xla = ylb in G(S), then tay = tbx for some t E S, atay = atbx with
at E I, and xla = ylb in G(I); thus ~ is injective. Also let a,x E S and
c E I. Then xI a = xcl ac with ac, xc E I; hence ~ is surjective. D
Proposition 2.5 is essentially due to Brameret [1962].
Proposition 2.6. G(S x T) ~ G(S) x G(T) for all nonempty commutative
semigroups S and T.
Proof. The canonical homomorphisms a: S--+ G(S) and j3: T--+ G(T)
induce a homomorphism
<p =ax j3 : S x T --+ G(S) x G(T), (x,y) ~----+ (a(x), j3(y)),
which by Proposition 1.3 induces a homomorphism
~: G(S X T) = (S X T)- 1 (8 X T) --+ G(S) X G(T)
that sends ( x, y) I (a, b) E (S x T) - 1 ( S x T) to
<p(x,y) <p(a,b)- 1 (a(x), j3(y)) (a( a), j3(b)) - 1

(a(x) a(a)-1, j3(y) j3(b)- 1 ) = (xla, ylb).
It is clear that~ is surjective. If ~((x,y)l(a,b)) = ~((z,t)l(c,d)), then xla =

zlc and ylb =tid, uaz = ucx and vbt = vdy for some u E S and v E T,
(u,v)(a,b)(z,t) = (u,v)(c,d)(x,y) inS x T, and (x,y)j(a,b) = (z,t)j(c,d);

thus ~ is injective. 0
3. CANCELLATIVE SEMIGROUPS.
Cancellative commutative semigroups provide natural structures for abstract

arithmetic and for general ideal theory. Some also have long been investigated
extensively in the guise of lattice ordered abelian groups. Examples include
abelian groups and free commutative semigroups. This section contains some
general properties and examples. Later sections give a few structural properties.
1. First we show:
Proposition 3.1. A cancellative c.s. which is finite and nonempty is a group.
Proof. In any c.s. S, multiplication by a E S is a mapping >.a : x f------+ ax
of S into S. If S is cancellative, then every >.a is injective; if S is also finite,
then every >.a is surjective. This implies that S is a single :J-C-class; then S is a
group by Proposition 1.4.3. 0
Proposition 3.2. A commutative semigroup S can be embedded into a group
if and only if it is cancellative; then the canonical homomorphism 1: S-----+ G(S)
is injective.
Proof. Subsemigroups of groups are cancellative. If conversely S is can-
cellative, then the canonical homomorphism 1 : S -----+ G(S) is injective by
Proposition 1.2, and S can be embedded into the abelian group G(S). 0
This provides the familiar embedding of S into an abelian group, namely, the
group of fractions or universal group G(S) of S. When S is cancellative it is
standard practice to identify x E S and o:(x) E G(S); then xja becomes xa- 1
and S becomes a subsemigroup of G(S).
Universal groups of cancellative c.s. enjoy additional properties.
Proposition 3.3. Let S be a nonempty cancellative c.s. An abelian group
G is isomorphic to the universal group of S if and only if S is isomorphic
to a subsemigroup T of G such that every g E G can be written in the form
g = ab- 1 for some a,b E T.
Proof. We just saw that every element of G(S) can be written in the form
xa- 1 for some x, a E S. Conversely let G be an abelian group and T be a
subsemigroup of G such that every g E G can be written in the form g = ab - 1
3. CANCELLATIVE 8EMIGROUPS. 37
for some a, b E T (in particular, T #0 ). Proposition 2.1, applied to the inclusion

homomorphism T ----+ G, provides a homomorphism ~ : G(T) ----+ G, which
sends x /a to xa - 1 for all a, x E T. Then ~ is surjective by the hypothesis,
and is injective, since ~ (xfa) = ~ (y/b) implies xa- 1 = yb- 1 , bx = ay, and
xfa = yjb. D
By Proposition 3.3 we may view cancellative c.s. (up to isomorphism) as
generating subsemigroups of abelian groups.
Proposition 3.4. Let S and T be cancellative c.s. If S is a subsemigroup
ofT, then G(S) is isomorphic to a subgroup of G(T).
Proof. By Proposition 2.2 the inclusion homomorphism S ----+ T induces a
homomorphism~: s-
1 8----+ T- 1 T which sends xja E s-
1 8 to xja E T- 1 T.
Let a,b,x,y E S. If xfa = yjb in r- 1T, then ay = bx since Tis cancellative,

and x/a = yjb in s- s; thus
1 ~ is injective. D
Alternately we may regard T as a subsemigroup of G(T), and then the
subgroup {ab- 1 I a,b E S} of G(T) generated by Sis isomorphic to G(S)
by Proposition 3.3, and may be identified with G(S).
Proposition 3.4 implies that G(S) is, up to isomorphism, the smallest abelian
group G of which S is a subsemigroup.
These results may lull readers into thinking that cancellative semigroups are
easily retrieved from abelian groups and are therefore not terribly complicated. A
cure for this particular delusion may be found in the next section, which shows
just how easy it is to retrieve all subsemigroups of Z.
2. Let S be a cancellative c.s. If E is a commutative ideal extension of S
by Q, then G(E) ~ G(S) by Proposition 2.3, and the canonical homomorphism
E ----+ G(S) induces a partial homomorphism of Q\0 into G(S). The ideal
extension E can be reconstructed from this partial homomorphism much as in
Lemma 1.3.7; the details are given below. More general results similarly construct
ideal extensions of a weakly reductive semigroup S from partial homomorphisms
into the translational hull of S (Clifford [ 1950]), which when S is a cancellative
c.s. is isomorphic to the subsemigroup {g E G(S) I gS ~ S} of G(S) (Petrich
[1973]).
Lemma 3.5. Let E be an ideal extension of a nonempty cancellative c.s.
S. A homomorphism T: E----+ G(S) which is the identity on S is well defined
by T(a) = asfs = safs for any s E S.
Proof. Let s, t E S and a E E. In G(S) we have as, ta E S and
asfs =taft, since s(ta) = (ta)s = t(as). Hence T(a) = asfs =taft depends
neither on s nor on t. For all a, b E E and s E S we have sbs = bs 2 , since

s, bs E S, and
7(a) 7(b) = (as/s) (bs/s) = asbsjs 2 = abs 2 js 2 = 7(ab);
hence 7 is a homomorphism. Also 7(a) = asjs =a if a E S. D
If E is commutative, then, as noted above, 7 can also be constructed by
composing the canonical homomorphism E ~ G(E) and the isomorphism
G(E) ';;:=! G(S) in Proposition 2.3.
In general, let S be a nonempty cancellative c.s. and E be any ideal extension
of S by a semigroup Q with zero. We may assume that Q n G(S) = 0. The
homomorphism 7: E ~ G(S) in Lemma 3.5 is the canonical homomorphism
of the ideal extension E of S. (A more general canonical homomorphism sends
an ideal extension of any semigroup S into the translational hull of S .)
Let r.p be the restriction to E\S of the canonical homomorphism. Then r.p
is a partial homomorphism of Q\0 = E\S into G(S). Moreover, x r.p(a) =
r.p(a) x = 7(a) 7(x) = 7 (ax) = ax E S for all a E Q\0 and x E S, and
r.p(a) r.p(b) = 7(a) 7(b) = 7 (ab) = ab E S whenever ab = 0 in Q. Hence
the operation on E is determined by the operation on S, the partial operation on
Q\0, and the partial homomorphism cp; in fact E is a subsemigroup of the ideal
extension of G(S) determined by r.p as in Lemma 1.3.7.
The converse is clear:
Lemma 3.6. Let S be a nonempty cancellative c.s. and Q be a semigroup
with zero such that Q n G(S) = 0. Let r.p be a partial homomorphism of Q\0
into G(S), such that r.p(a) x E S for all a E Q\0 and x E S and r.p(a) r.p(b) E S
whenever ab = 0 in Q. Let E = S U (Q\0), with multiplication * defined by
a*b abE Q if ab =/: 0 in Q,
a*b r.p(a) r.p(b) if ab = 0 in Q,
a*y r.p(a) y,
X*b xr.p(b),
X*Y xy E S
for all a, b E Q\0 and x, y E S. Then E is an ideal extension of S by Q,
and a subsemigroup of the ideal extension of G(S) determined by the partial
homomorphism r.p. Moreover, E is commutative if and only if Q is commutative.
If r.p sends Q\0 into S ~ G(S), then the ideal extension in Lemma 3.6
coincides with the ideal extension in Lemma 1.3.7 which is determined by the
4. NUMERICAL SEMIGROUPS. 39
partial homomorphism <.p. In general we say that the ideal extension E in Lemma
3.6 is determined (as an ideal extension of S) by the partial homomorphism
<.p. The remarks before Lemma 3.6 show that all ideal extensions of S can be
constructed in this fashion. Thus:
Proposition 3.7. Every ideal extension of a nonempty cancellative c.s.
S by a semigroup Q with zero is determined by a partial homomorphism of
Q\ {0} into G(S), namely the restriction to Q\0 ofthe canonical homomorphism
a f----7 axjx = xajx, where x is any element of S.
Stronger results were obtained by Heuer & Miller [ 1966] and Heuer [ 1971]
in case Q is a group with a zero element adjoined.
3. Cancellative c.s. have been studied in various ways. Sankaran [ 1961]
showed that every cancellative c.s. S has a (topological) uniform structure. Schein
[ 197 5] showed that the divisibility preorder ;£J{ on S induces a compatible pre-
order ;£ on G(S) and showed that every subsemigroup of S is densely embedded
in its idealizer if and only if G (S) is archimedean under this preorder. Hall [ 1971]
constructed all cancellative c.s. with two generators.
Cancellative c.s. with stronger properties are considered in the next sections.
Properties weaker than cancellativity have also been studied. A c.s. S is sep-
arative when a 2 = ab = b2 implies a = b; these semigroups are studied in
Section III.2. A c.s. S is reductive when ax= bx for all x E S implies a= b;
equivalently, when it is weakly reductive. Clifford [1950] showed that this is a
helpful property when studying ideal extensions of S. Gluskin [1983] studied
the weaker property that ax = bx for all x E S implies a = b when a E 8 2 .
4. NUMERICAL SEMIGROUPS.
In this section we consider subsemigroups of Z.

1. First we look at positive subsemigroups.
Proposition 4.1. When S is a nonempty subsemigroup ofN+, there exists an
integer d > 0 such that S consists of multiples of d and contains all sufficiently
large multiples of d. Hence S is finitely generated.
Proof. By Proposition 3.4, G(S) s;;; G(N+) = Z and G(S) = Zd for some
d > 0. Then every element of S s;;; Zd is a multiple of d. Also d = n - m for
some m, n E S by Proposition 3.3, and m +dE S for some m E S. (Thus d
is the greatest common divisor of the elements of S.)
Let m = kd. If l ~ k 2 , then l = kq + r for some 0 ~ r < k and q ~ k,

l = (q - r) k + r ( k + 1) , and ld = (q - r) m + r (m + d) E S. In particular S
is generated by m, m + d, and every other element s < k 2 d of S. D
Proposition 4.1 has been known for some time. A proof was published by
Jensen & Miller [1968].
Corollary 4.2. A nonempty subsemigroup of Z either contains only non-
negative integers, or contains only nonpositive integers, or is a subgroup of Z.
Proof. If S is not trivial or empty, then as above G(S) = Zd for some
d > 0. Assume that S contains both positive and negative integers. Then S
contain arbitrarily large positive integers and arbitrarily large negative integers.
Since d is the difference of two elements of S, d is then the difference of two
positive elements of S and the difference of two negative elements of S; hence
both parts of S generate the same subgroup Zd of Z. By Proposition 4.1, S
contains all sufficiently large positive multiples of d and all sufficiently large
negative multiples of d. Hence S contains both d and -d and S = Zd 9:! Z. D
2. A numerical semigroup is a subsemigroup of :N+ which contains all
sufficiently large natural numbers; numerical mono ids also contain 0 (and are
usually called numerical semigroups in the literature.) By Proposition 4.1 every
nonempty subsemigroup of :N+ is isomorphic to a numerical semigroup, and
numerical semigroups are finitely generated. The subsemigroup of :N+ generated
by a 1 , a 2 , ... , an is numerical if and only if a 1 , a 2 , ... , an are relatively
prime. Presentations of numerical semigroups have been studied by Gastinger,
Kunz, & Waldi [ 1991]; the minimal number of defining relations has been studied
by Rosales [ 1996A], [ 1996N] and Rosales & Garcia-Sanchez [ 1998N].
Numerical semigroups have been studied extensively, partly because of anum-
ber of open questions, partly because of their connections to algebraic geometry.
Higgins [ 1972] and Kunz [ 1987] are good surveys of early results.
Numerical semigroups first appear in problems posed by Sylvester [1884]
and Frobenius. Sylvester's problem consists in determining the largest integer
not in (a, b) , where a and b are relatively prime. He showed that the answer is
ab - a - b. According to Brauer [ 1942], Frobenius "mentioned occasionally in
his lectures" the more general problem of determining the largest integer not in
( a 1 , a 2 , ... , an) . This is also called the money-changing problem: given an
unlimited supply of coins with relatively prime denominations a 1 , a 2 , ... , an,
what is the largest amount that cannot be formed by means of these coins. (We
prefer to think of this as the fried chicken problem: given that fried chicken
can be ordered in boxes of a 1 , a 2 , ... , an pieces, where a 1 , a 2 , ... , an are
4. NUMERICAL SEMIGROUPS. 41
relatively prime, what is the largest number of pieces that cannot be ordered.)
Generally the Frobenius number of a numerical semigroup S, usually de-
noted by g(S), is the largest integer n tt. S. (The conductor of S is g(S) + 1.)
No general formula is known that computes g(S) from the generators of S. Cur-
tis [ 1990] showed that there is no finite set of polynomial formulas for g( S) .
Upper bounds and a few exact formulas for g(S) have been found in particular
cases (Sathaye [ 1977], Selmer [ 1977], Rod seth [ 1978] for 3-generator semi groups,
Froberg [1994] for 4-generator symmetric semigroups). See also Deddens [1979].
A numerical semigroup is symmetric when g(S) is odd and n > 0, n tt. S
implies g(S)- n E S; equivalently, when g(S) is odd and S is maximal for its
Frobenius number. Froberg, Gottlieb, & Haggkvist [1987] showed that there are
at least 2 Ln/BJ symmetric semigroups with g( S) = n. Backe lin [ 1990] showed
that there are C.2n/ 2 numerical semigroups and C' .2n/6 symmetric semigroups
with g(S) = n, where C and C' vary within finite bounds.
When S is generated by relatively prime numbers a 1 < a2 < ... < an, g.c.d.' s
d 1 , d 2 , ... , dn are defined by induction by d 1 = a 1 and di = g.c.d. (di_ 1 ,ai)
if i > 1 (then dn = 1 ). Bertin & Carbonne [ 1975], [ 1977] called a numerical
semigroup free when every element of S can be written in the form l::i ni ai
with ni < dddi+l· They showed that free numerical semigroups are symmetric.
Numerical semigroups generated by finite arithmetic progressions were stud-
ied by Juan [ 1993]. For numerical semigroups generated by intervals, explicit
formulas give the elements, Frobenius number, and minimum number of defining
relations, and decide whether the semigroup is symmetric or a complete intersec-
tion (Garda-Sanchez & Rosales [ 1999]). Patil & Singh [ 1990] studied numerical
semigroups generated by almost arithmetic progressions. Other interesting classes
are considered by Barucci, Dobbs, & Fontana [ 1997].
The multiplicity m(S) of a numerical semigroup S is the smallest (positive)
element m of S. A semigroup of multiplicity m is generated by at most m el-
ements, and has maximal embedding dimension when it requires m generators
(see below). Rosales [1996N] constructed a one-to-one correspondence between
numerical semigroups with Frobenius number g and multiplicity m, and numer-
ical semigroups with Frobenius number g + m, multiplicity m, next positive
element > 2m, and maximal embedding dimension. A similar one-to-one corre-
spondence is constructed in Rosales [ 1996S]. See also Rosales & Garcia-Sanchez
[ 1999].
3. Numerical semigroups also arise in algebraic geometry. At every point
P of a projective curve C, the rational functions on C which are regular on

C\ {P} but not at P have pole orders at P which constitute a subsemigroup
of .N+, the Weierstrass semigroup of C at P. (This concept was generalized
by Delgado [1990]; see also Kim [1994] and Homma [1996].) The Weierstrass
Gap Theorem states that S contains all but g elements of _N+ , where g is the
genus of C; in particular, S is a numerical semigroup. At an ordinary point,
S = { g + 1 , g + 2, ... } , but other configurations are possible.
In general, the genus of a numerical semigroup S ~ _N+ , generally denoted
by g or 'Y, is the number of elements of N+\S (the number of gaps). Selmer
[ 1977] gave a general formula for the genus.
It was long conjectured that every numerical semigroup occurs as a Weier-
strass semigroup of some curve (Hurwitz [1893]). The first counterexample was
given by Buchweitz [1980]. More were found by Torres [1994], [1995], Kim
[ 1996], and Komeda [ 1997]; Komeda [ 1998] constructed counterexamples of ar-
bitrarily high genus. All these counterexamples have genus ~ 16. It is still not
known precisely which numerical semigroups are Weierstrass. Komeda showed
that Weierstrass semigroups include every numerical semigroup with 2 or 3 gen-
erators, every symmetric or almost symmetric (g(S) = 2g- 2) semigroup with
4 generators [ 1982], every numerical semigroup of multiplicity 4 [ 1983] or 5
[1992], every numerical semigroup of genus at most 7, and every numerical
semigroup such that g(S) < 2m(S) [1994]. Weierstrass semigroups were also
constructed by Maclachlan [1971], Rim & Vitulli [1977], Waldi [1980], Knebl
[1984], Tutalar [1987], Coppens [1988], and Komeda [1991].
4. Ring theory provides a more exact relationship between numerical semi-
groups and certain valuation rings. When R is a discrete valuation ring, and
v : Q ---+ Z is the corresponding valuation on the quotient field Q of R,
then the set of values v(R) of v on R is a submonoid of N, which one may
assume is a numerical monoid. Conversely let k be a field and k [X] be the
polynomial ring with one indeterminate. If S = ( a 1 , a 2 , ... , an ) is a numer-
ical monoid, then the semigroup ring of S is (up to isomorphism) the subring
k[SJ = k [Xa 1 , Xa 2 , ... , Xan J of k[X]; moreover there is a valuation v on the
field of quotients k(X) such that v(k[S]) = S. (The multiplicity of S should
thus be interpreted in terms of polynomials in k[S] and not in terms of points on
algebraic curves.)
The finitely generated domain k[S] = k[Xa 1 , Xa 2 , ... , Xan] is isomorphic
to a quotient ring k[X1 , X 2 , ... , Xn]/P by a prime ideal p, which in tum
corresponds to an n-dimensional affine algebraic variety V whose coordinate
4. NUMERICAL 8EMIGROUPS. 43
ring is k[S]. In fact V is a curve, since k[S] has Krull dimension 1. The ideal
p was studied for generators by Bresinsky [1975], [1988], Patil [1993], Campillo
& Pis6n [1993], and Herzinger [1999]. The value semigroup S of V has been
studied extensively since Waldi [ 1972] showed that two curves are equisingular
if and only if they have the same value semigroup.
The embedding dimension e(S) of S is the minimal number of generators
of S (the smallest n such that S = ( a 1 , a 2 , ... , an) for some relatively prime
a 1 , a 2 , ... , an); equivalently, e( S) is the dimension of the smallest affine space
into which an algebraic variety can be embedded when its coordinate ring is k[S].
Since S contains all multiples of m( S), e( S) ~ m( S) always.
Numerical semigroups have been studied for properties of the corresponding
variety, such as being a complete intersection, and for various ring properties of
k[S], such as Gorenstein, Buchsbaum, Cohen-Macaulay, that are of interest for
the corresponding variety. Kunz [ 1973] showed that k [S] is Goren stein if and
only if S is symmetric; for a different proof see Huang [ 1995]. Cavaliere &
Niesi [1983] and Bresinsky [1984] gave general characterisations of Buchsbaum
and Macaulay semigroups. Herzog [ 1970], Delorme [ 197 6], and Cavaliere &
Niesi [1984] characterized complete intersection semigroups (see also Fischer
& Shapiro [ 1996]). Herzog [ 1970] also showed that a numerical semigroup
with three generators is complete intersection if and only if it is symmetric; this
also holds for semi groups with four generators (Bresinsky [ 1979]). See also
Kato [1979]; Ruiz [1985]; Brown & Curtis [1991]; Pfister & Steenbrink [1992];
Campillo & Marijuan [1991]; Barucci, Dobbs, & Fontana [1994]; Torres [1994],
[1997]; Anderson & Scherpenisse [1995]; Barucci & Froberg [1997].
The type t( S) of S is the number of positive integers n rt. S such that
n + s E S for all 0 < s E S. Froberg [ 1994] showed that the type of S is
also the Cohen-Macaulay type of k[S]. Cavaliere & Niesi [1983] showed that
there are 4-generator semigroups of arbitrary type. Froberg, Gottlieb, & Haggkvist
[1987] showed that (t(S) + l)(g(S)- g) ~ g(S) + 1. "Type" sequences that
begin with t(S) were studied by D'Anna & Delfino [1997] and D'Anna [1998].
5. Numerical semigroups arise from algebraic curves in other ways. The
LUroth semigroup of a curve, and that of its rational function field K = k( t),
was defined by Moh & Heinzer [1982] as the set of all finite degrees [K: k(x)]
(where x E K is transcendental over k ). It has been studied by Greco [ 1990],
Greco & Raciti [ 1991], Paxia, Raciti, & Ragusa [ 1992 ], Coppens [ 199 5], and
Tokunaga & Yoshihara [1995].
A value semigroup can more generally be assigned to every branch of an
algebraic curve (Apery [ 1946]). Two branches are equisingular if and only if
they have the same value semigroup (see e.g. Zariski [ 1973 ]). The resulting
numerical semigroups are symmetric (Apery [ 1946]) and were characterized by
Bresinsky [1972] and Teissier [1973]. Bertin & Carbonne [1975], [1977] gave
an alternate construction and showed that these value semigroups are free. See
also Angermiiller [ 1991].
5. GENERAL STRUCTURE.
The results in this section analyze cancellative c.s. by means of cancellative

congruences that eliminate units, then eliminate torsion; geometric properties of
convex sets can be brought to bear on the resulting semigroups. These techniques
are used in the next sections.
1. A congruence e on a c.s. S is cancellative in case the quotient semigroup
s I e is cancellative; equivalently, when ac e be implies a e b' for all a, b, c E s.
Proposition 5.1. Let S be a nonempty cancellative c.s. Every subgroup H
of G(S) induces a cancellative congruence e on S, for which x e y in S if
and only if xy- 1 E H in G(S); then G(Sie) ~ G(S)IH. Moreover, every
cancellative congruence e
on S is induced in this fashion by a subgroup of
G(S), namely R(e) = { xy- 1 E G(S) I x,y E S, x e y }.
R(e) is the Redei group of e, after Redei [1956].
Proof. When H is a subgroup of G(S), the partition of G(S) into cosets
of H is a congruence £ on G (S) ; G (S) I£ is the quotient group G (S) I H,
and g £ h if and only if gh - 1 E H, for all g, h E G(S). In particular £ is
a cancellative congruence on G(S). Hence the restriction e of £ to S is a
cancellative congruence on S. The e -class of x E S is H x n S, and SI e
is isomorphic to the subsemigroup { H x I x E S} of G (S) I H. Moreover
every coset of H can be written in the form H xy - 1 = (H x) (H y - 1) for some
x,y E S; hence G(Sie) ~ G(S)IH, by Proposition 3.3.
Conversely let e be a cancellative congruence on S. Then R = R( e) is
a subgroup of G(S). Let a, b E S. If ab- 1 E R, then ab- 1 = xy- 1 , where
x, y E s and X e y' ax e ay = bx' and a e b, since e is cancellative. If
conversely a e b, then ab- 1 E R. Thus e is the congruence induced by R. 0
When eis the cancellative congruence on S induced by a subgroup H
of G (S), we denote the quotient semigroup SI e by SI H and regard it as a
subsemigroup of G(S)IH; then G(SIH) = G(S)IH.
5. GENERAL STRUCTURE. 45
2. A cancellative c.s. S has at most one idempotent, which is its identity

element: indeed e 2 = e in S implies e = 1 in G(S). We call a cancellative c.s.
S reduced when either S has no identity element, or S has an identity element
and its group of units is trivial. These semigroups are also called group-free and
a variety of other names. Hamilton, Nordahl, & Tamura [1975] extended to these
semigroups Tamura's construction ofN-semigroups in Section 111.4.
The next result shows that reduced cancellative c.s. are really just another
name for partially ordered abelian groups. There is a well developed theory of
lattice ordered groups; see the recent books by Anderson & Feil [1988], Kopytov
& Medvedev [1994], Darnel [1995]. Partially ordered abelian groups in general
have not been studied as extensively; the book by Goodearl [ 1986] concerns
partially ordered abelian groups with the Riesz interpolation property.
Proposition 5.2. A c.s. is cancellative without identity element if and only if
it is the positive cone of a partially ordered abelian group. A c.m. is cancellative
and reduced if and only if it is the nonnegative cone of a partially ordered abelian
group.
Proof. When S is a reduced cancellative c.m., then the relation
g ~ h if and only if g -l h E S
is antisymmetric since S is reduced and is a partial order relation on G(S), which
is compatible (if g ~ h, then gk ~ hk for all k E G(S)). Then (G(S), ~) is a
partially ordered abelian group, and S = { g E G (S) I g ~ 1} is the nonnegative
cone of G(S). Conversely, the nonnegative cone of a partially ordered abelian
group (G, ~) is a cancellative c.m. S, which is reduced since ~ is antisymmetric.
(However, it may happen that G ( S) ~ G.)
If now S is a cancellative c.s. without identity element, then the monoid
S1 is cancellative, commutative, and reduced; hence S 1 is the nonnegative cone
of a partially ordered abelian group, whose positive cone is S. Conversely, the
positive cone of a partially ordered abelian group is a cancellative c.s. 0
A cancellative c.s. S with a group of units U is an ideal extension of S\ U
by U U {0}, as studied by Heuer & Miller [1966] and Heuer [1971]. The next
result provides another approach which anticipates Chapter V.
Proposition 5.3. Let S be a cancellative c.m. and U be the group of units
of S. In G ( S), S is a union of cosets of U, which coincide with the '}{-classes
of S; moreover S /U is reduced.
Proof. S is a union of cosets of U since Us ~ S for all s E S; S /U ~
G(S)/U is reduced since s,t E S and (Us)(Ut) = U in G(S)/U implies
46 II. CANCELLATIVE 8EMIGROUPS.
st E U, stu = 1 for some u E U, and s, t E U.

Also s J{ t implies s = ut, t = vs for some u,v E S, s = uvs, uv = 1,
u,v E U, and s E Ut. Conversely, s = ut for some u E U implies t = u- 1 s
and s J{ t. Thus Ht = Ut. D
By Proposition 5.3, a cancellative c.m. S is determined by a subgroup U of
G(S) and a reduced subsemigroup of G(S)/U. If S is uniquely divisible (and
cancellative) then S ~ U x S/U (Brown & LaTorre [1966]). In general, S/U
is a universal cancellative reduced c.m. of S, since every homomorphism of S
into a reduced semigroup must send U to the identity element. The rank of the
cancellative c.m. S is that of G(S); the reduced rank of S is the rank of its
reduced semigroup S /U, equivalently, the rank of G(S) /U.
3. A semigroup S is power cancellative in case xn = yn implies x = y,
for all n > 0 and x, y E S; c.s. with this property are also called torsion free.
Free c.s. and torsion free abelian groups are power cancellative. The author's
earliest reference to cancellative and power cancellative c.s. is Abellanas [ 1965].
Proposition 5.4. When S is a cancellative c.s., then G(S) is torsion free
if and only if S is power cancellative.
Proof. G(S) is torsion free if and only if xn y-n = (xy- 1)n = 1 implies
xy - 1 = 1 , for all n > 0 and x, y E S. D
A congruence e is power cancellative when the corresponding quotient semi-
e e
group is power cancellative; equivalently, when Xn yn implies X y, for all
n > 0.
Proposition 5.5. Let S be a c.s. There is a smallest power cancellative
congruence 'J on S, namely
x 'J y if and only if xn = yn for some n > 0.
Moreover, S j'J has no idempotent if and only if S has no idempotent.
Proof. 'J is reflexive, symmetric, and transitive (if xm = ym and yn =zn,
then xmn = ymn = zmn), and is a congruence (if xn = yn, then (xz)n = (yz)n).
'J is a power cancellative congruence since xm 'J ym implies xmn = ymn for
some n > 0 and x 'J y. If finally e is a power cancellative congruence on S,
then x 'J y implies xn = yn for some n > 0, xn e yn for some n > 0, and
x e y; thus 'J is the smallest power cancellative congruence on S.
If S has an idempotent, then so does S j'J. If conversely the 'J-class Tx of
x E S is idempotent in S j'J, then x 2 'J x, x 2n = xn for some n > 0, and
xn E S is idempotent. D
The quotient semigroup Sj'J in Proposition 5.5 is the universal power can-
cellative semi group of S: if <p : S ----+ T is a homomorphism of S into a
power cancellative semigroup T, then Im <p is power cancellative, ker <p is a
power cancellative congruence, and <p factors uniquely through the projection
S ----+ S j'J by Proposition 1.2.4.
Proposition 5.6. Let S be a cancellative c.s. The smallest power cancella-
tive congruence 'J on S is induced by the torsion part of G(S). Moreover Sj'J
is cancellative; S j'J has no idempotent if and only if S has no idempotent; if S
is reduced, then S j'J is reduced.
Proof. Let T = {t E G(S) I tn = 1 for some n > 0} be the torsion

subgroup of G (S) . 'J is the congmence induced by T, since xy -l E T if and
only if xn = yn for some n > 0. Hence S j'J = S /T is cancellative. If S is
reduced without an identity element, then S j'J has no idempotent, by Proposition
5.5, and is reduced. If S is reduced with an identity element, then Sj'J has an
identity element and is reduced: if the 'J-class Tx of x E S is a unit of S j'J,
then xy 'J 1 for some y E S, xn yn = 1 for some n > 0, x is a unit of S, and
X= 1. D
By Proposition 5.6, every commutative semigroup S has a universal cancella-

tive, power cancellative, and reduced c.s., which is obtained from the universal
cancellative semigroup C(S) by dividing C(S)/U by its smallest power can-
cellative congruence 'J. These universal semigroups were constructed by Head
[ 1979] using monoids of quotients. In what follows we abbreviate the mouthful
"cancellative, power cancellative, and reduced" as c.pc.r.; c.pc.r. semigroups are
the positive or nonegative cones of partially ordered torsion free abelian groups,
and have been called by a variety of other names. Laplaza [ 1964] showed that
every c.pc.r. c.s. can be embedded into the nonnegative cone of a totally ordered
abelian group.
4. The following constructions let basic geometric properties of convex sets
be applied to c.pc.r. semigroups.
When S is c.pc.r. we regard the torsion free abelian group G = G(S) as a
= Q 0z G, which is a vector space over Q, and
subgroup of its divisible hull G
which we in turn regard as a divisible subgroup of its real hull G = JR. 0z G,
which is a vector space over JR.. The rank of S is also the dimension of G over
Q and the dimension of G over JR.. If S is finitely generated, then G is a free
abelian group and every basis of G is also a basis of G over Q and a basis of
G over JR..
In a real vector space a convex cone is an additive submonoid C such that

c E C implies .Ac E C for all >. E IR, >. ~ 0. Similarly a convex cone in a rational
vector space is an additive submonoid C such that c E C implies qc E C for
all q E IR, q ~ 0. A convex cone is generally called reduced when it contains
no nonzero subspace; equivalently, when it is reduced as an additive semigroup.
Every subset X of G generates a convex cone X of G, namely
X = { q1 X1 + ··· + qk xk I X1' · · ·' xk EX, q1' · · · 'qk E IQ, q1' · · · 'qk ~ 0 }.
Similarly, every subset X of G generates a convex cone X of G, namely
X = { >. 1 x 1 + · ··+ >.k xk I x 1 , ... , xk EX, >. 1 , ... , >.k E IR, >. 1 , ... , >.k ~ 0 }.
The convex cone S of G generated by S is the divisible hull of S; we call
S the real hull of S. Embeddings into divisible semigroups have been set up by
enough authors to gain the status of folk theorems. Minimal divisible em beddings
were studied by Tamura [ 1963].
Proposition 5.7. When T is a subsemigroup of G, then
T = { tjn E G It E T, n > 0} U {0}
and T is a uniquely divisible monoid (for every a E T and m > 0 there exists
a unique x E T such that mx =a). If T is reduced, then so is T.
Proof. Every X is a submonoid of G. Let a = q1 t 1 + ··· + qk tk E T,
where t 1 , ... , t k E T, q 1 , ... , qk E iQ, q 1 , ... , qk ~ 0. In iQ, q 1 , ... , qk
have a common denominator n > 0; then t = nq 1 t 1 + ··· + nqk tk E T, since
T is a semigroup and nq 1 , ... , nqk EN, and a= tjn. Conversely tjn E T
for all t E T and n > 0.
Let a = t/ n E T, where s E T and n > 0. Then na E T. If m > 0 and
x E G, then mx =a if and only if x = tjmn, and then x E T. Thus T is
uniquely divisible.
Let a= sjm and b = tjn E T, with s,t E T and m,n > 0. Then
a+ b = (ns + mt)jmn. If T is reduced, then a+ b = 0 implies ns + mt = 0,
s = t = 0, and a = b = 0; hence T is reduced. D
Proposition 5.8. When T is a reduced subsemigroup of G, then T is a

- -
reduced cone of G. If T is a cone, then T n G = T.
Proof. Assume that T is not reduced. Then a + b = 0 for some a, b E T,
a,b =I- 0. We have a= >. 1 t 1 + ·· · + >.k tk, b = >.k+ 1 tk+ 1 + ··· + .Aztz for some
t 1 , ... , t 1 E T and .>. 1 , ... , >.1 E ~, .>. 1 , ... , >.1 ~ 0. We may assume that
=f. 0 and .>. 1 , ... ,
t 1 , ... , t 1 >.1 > 0 and still have l ~ 2. In the subspace of G
spanned by t 1 , ... , t 1 , the equality
,\1 t1 + ... + .Az tz = 0
can then be viewed as a (finite) homogeneous system of linear equations with
unknowns .>. 1 , ... , >.1, which has rational coefficients since t 1 , ... , t 1 E G
and has a solution in which all unknowns are positive. By Lemma 5.9 below
this system has a solution in which all unknowns are positive rational numbers
q1 , ... , q1 . The equality
ql t1 + ··· + qz tz = 0
then shows that T is not reduced. Hence T is not reduced.
Now let T be a cone in G. Let a E T n G, a =f. 0. Then a has rational
coordinates in G, and a is a linear combination a = .>. 1 t 1 + ··· + >.1 t 1 of
t 1 , ... , t 1 E T with real coefficients .>. 1 , ... , >.1 that may be assumed positive.
As above the equality
>. 0 a - .>. 1 t 1 - ··· - >.1t 1 = 0
can be viewed as a (finite) homogeneous system oflinear equations with unknowns
>.0 , .>. 1 , ... , >.1, which has rational coefficients since a, t 1 , ... , t 1 E G and has
a solution in which all unknowns are positive (including .>.0 = 1 ). By Lemma 5.9
this system has a solution in which all unknowns are positive rational numbers
q0 , q1 , ... , q1 . The equality
qo a - q1 t1 - · · · - qz tz = 0
then shows that q0 a E T. Hence a E T. Hence T n G = T. D
Lemma 5.9. A homogeneous system of linear equations which has rational
coefficients and has a solution in which all unknowns are positive real numbers
also has a solution in which all unknowns are positive rational numbers.
Proof. Applying Gauss-Jordan reduction to the given system (S) yields a
simpler equivalent system (R). The unknowns can be numbered so that (R) reads
X1- q1,k+1 xk+1- · · ·- q1,n xn 0,
X2- q2,k+1 xk+1 - · · ·- q2,n xn 0,

equivalently, xi = 'PI (xk+I, ... , xn), ... , xk = 'Pk(xk+I, ... , xn), where
'Pi(xk+I, · · ·, xn) = qi,k+I xk+I + ··· + qi,nxn ·

Here qi,k+I, ... , qk,n are rational since they are obtained from the rational
coefficients of (S) by Gauss-Jordan reduction. Now (S) and (R) have a solution
ai, ... , an in which ai, ... , an > 0. Since 'PI, ... , 'Pk are continuous there exist
neighborhoods uk+I> ... , un of ak+I> ... , an such that 'Pi(xk+I> ... , xn) > 0
for all i ;;;; k and all xk+I E Uk+l, ... , xn E Un. There exist positive
rational numbers bk+l E Uk+I, ... , bn E Un; then bi = 'PI (bk+l, ... , bn),
... , bk = 'Pk (bk+I , ... , bn) are positive rational numbers and bi, ... , bn is a
solution of (R) and of (S) in which all unknowns are positive rational numbers. 0
6. FACES.
In this section we use the techniques of Section 5 to study the faces of

cancellative semigroups.
I. A face of a commutative semigroup S is a nonempty subsemigroup A of
S such that ab E A implies a, b E A; equivalently, a nonempty subsemigroup
of S whose complement is an ideal of S. The name is due to Petrich [1963F].
Faces of semigroups were first considered by Thierrin [1953] but faces of cones
go back to antiquity.
When A is a face of S, the ideal P = S\A ~ S is a (completely) prime
ideal of S (abE P implies a E P or bE P). If conversely P ~ S is a proper
prime ideal of S, then S\P is a face of S. The following results can therefore
be stated in terms of proper prime ideals rather than faces.
Proposition 6.1. A finitely generated c.s. has finitely many faces.
Proof. Let X be a generating subset of S and A be a face. Every a E A
is a finite product a = xi · · · xn of elements of S; since A is a face, xi E A
for all i. Thus A is generated by X n A. If X is finite, then there are finitely
many subsets X n A and finitely many faces of S. D
Proposition 6.2. Let S be a cancellative c.s.
(1) If S is a monoid with group of units U, then a subset of S is a face of
S if and only if it is a union of cosets of U that constitute a face of S /U.
(2) If S is reduced, then a subset of S is a face of S if and only if it is a
union of'J-classes that constitute a face of Sj'J.
6. FACES. 51
Proof. (1). Let A be a face of S. If a E A and u E U, then u- 1 ua =

a E A and ua E A. Hence A is a union of cosets of U. Now AIU is a
nonempty subsemigroup of SIU. If Ua,Ub E SIU and (Ua)(Ub) E AIU, then
Uab = (Ua)(Ub) = Uc for some c E A, ab E Uc ~ A, a,b E A, and Ua,
Ub E AIU; thus AIU is a face of SIU. If conversely B is a face of SIU,
then A = {a E S I U a E B} is a nonempty subsemigroup of S (Proposition
1.2.5) and abE A implies (Ua)(Ub) = Uab E B, Ua, Ub E B, and a,b EA.
Thus A is a face of S. We see that B = AIU.
(2). LetT= SI'J and T: S----+ SI'J be the projection. If A is a face of S
and x 'J y E A, then xn = yn E A for some n > 0 and x E A; thus A is a union
of'J-classes. Also T(A) is a nonempty subsemigroup ofT. If T(a) T(b) E T(A),
then ab 'J c for some c E A, ab E A since A is a union of 'J-classes, a, b E A,
and T(a), T(b) E T(A); thus T(A) is a face ofT. If conversely B is a face of
T, then A= T- 1 (B) is a face of S, and B = T(A). D
Proposition 6.2 constructs an order-preserving one-to-one correspondence be-
tween the faces of a cancellative semigroup and the faces of its universal c.pc.r.
sem1group.
2. When S is a c.pc.r. semigroup, we write S additively and let G = G(S)
and S ~ G be the divisible hull of S, as in Section 5.
Proposition 6.3. Let S be a cancellative, power cancellative, and reduced
c.s. If T is a face of S, then T is a face of S and T = T n S. If conversely
A is a face ofS, then G(A) is a subspace of G, T =An S = G(A) n S is a
face of S, and T =A.
By Lemma 5.9, a similar result holds for the real hull S.

Proof. Let T be a face of S. Let a = s lm E S, with s E S and m > 0 .
If a E T, then a = tIn for some t E T and n > 0, ns = mt E T, and s E T;
hence s I m E T if and only if s E T. With m = 1 this yields T n S = T.
Let a= slm and b =tinES, with s,t E S and m,n > 0. If a,b E T,
then s,t E T and a+ b = (ns + mt)lmn E T. If conversely a+ bET, then
ns + mt E T, s,t E T since m,n > 0, and a,b E T. Thus Tis a face of S.
Conversely let A be a face of S. Then T = A n S is a face of S. If
a = sIn E A, with s E S and n > 0, then s = na E A n S = T and a E T.
If conversely a= tim E T, with t E T and m > 0, then ma = t E A and
a E A. Thus A = T.
Let q = rjs E Q+, with r,s > 0. Then a E A implies sqa = ra E A

and qa E A. Hence a- b E G(A) implies q (a- b) E G(A). Since G(A)
is a group this holds for all q E Q and G(A) is a subspace of G. Moreover
G(A) n S =A: if c =a-bE G(A) n S, where a,b E A, then b + c =a E A
with b,c E S, and c EA. Hence G(A) nS = AnS = T. D
Proposition 6.4. In a cancellative c.s. offinite reduced rank n, every chain
ofproper prime ideals has length at most n.
Proof. First let S be c.pc.r. of rank n. When T is a face of S, then G(T)

is a subspac.e of G and T = G(T) n S, by Proposition 6.4. Hence T1 ~ T2
implies G(T1 ) ~ G(T2 ). A chain of faces of S generates a chain of subspaces
of G of equal length and therefore has length at most n.
Now let S be any cancellative c.s. of finite reduced rank n, so that the
universal c.pc.r. c.s. T of S has rank n. By Proposition 6.2, a chain of faces
of S induces a chain of faces of T of equal length and has length at most n.
Hence a chain of proper prime ideals of S also has length at most n. D
If S is a finitely generated c.m. in Proposition 6.4, then S has a chain of
proper prime ideals of length n, since, in ffi.n, a polyhedral set (= the intersec-
tion of finitely many closed half spaces) has faces of every dimension (see e.g.
Br¢ndsted [1983], Corollary 8.6). This property does not extend to infinitely
generated semigroups, as the next example shows. Matsuda [1998A] proved
additional height properties of prime ideals in cancellative c.s.
Example 6.5. Let X = { x, y} and
S = {ax x + ay y E Fx I ax , ay > 0}.

The only face of S is S itself: if indeed A is a face of S and a= ax x + ay yEA,
then for any b = bx x +by y E S we have b ~ na if n > 0 is large enough
(namely, if nax ~ bx and nay ~ by), and then b + c = ( n + 1) a E A for some
c E S and b E A, for every b E S. D
3. Next we show:
Proposition 6.6. Let S be a finitely generated, cancellative c.s. If S has
reduced rank n, then S has at least n faces of reduced rank 1.
Proof. Let S be c.pc.r. of rank n with a finite generating subset X. Then
X = S. Let Y ~ X be minimal such that Y = S. Then Y spans G and has at
least n elements. Also x fj: Y\ { x} when x E Y, otherwise Y\ { x} = S. For
6. FACES. 53
every x E Y we show that the subsemigroup

C = {qx Iq E Q, q ~ 0}
of S is a face of S; hence S has at least n faces of rank 1, and so does S by

Proposition 6.3; then so does every finitely generated cancellative c.s. of reduced
rank n, by Proposition 6.2.
Assume that e = a+ b E C, where a, b E S. If e = 0, then a = b = 0 E C,
since S is reduced. Assume e -=f 0. We have a = l:yEY ay y, b = l:yEY by y,
e = l:yEY eyy, with ay, byE Q, ay, by~ 0, ey = ay +by ~ 0, for every
y E Y; also e = qx, where q > 0. If ex < q, then ( q - ex) x = e - ex x E
Y\ {x} and x E Y\ {x}, a contradiction by the choice of Y. If ex > q, then
0 = e- qx = (ex - q) x + l:y#x ey y, a contradiction since S is reduced.
Therefore ex = q. Then 0 = e - qx = l:y#x ey y and ey = 0 for all y -=f x,
since Sis reduced. Hence a= axx E C and b = bxx E C. D
S can be replaced by S in this argument, and then the result follows from
Minkowski's Theorem (see e.g. Bnzmdsted [1983], Theorem 5.10). Example 6.5
shows that Proposition 6.6 does not extend to infinitely generated semigroups.
If S is finitely generated of reduced rank n ~ 2, then it is immediate that S
has exactly n faces of reduced rank 1. The next example shows that a finitely
generated semigroup of reduced rank n ~ 3 may have more than n faces of
reduced rank 1; similar examples show that there can be any (finite) number of
faces of reduced rank 1 .
Example 6.7. Let X= {x,y,z} and S <;;:; Fx be generated by z, x + z,
y + z, and x + y + z. Then S has rank 3, and a = ax x + ay y + az z E S if and
only if 0 ~ ax ~ az and 0 ~ ay ~ az. It is straightforward that z, x + z, y + z,
and x + y + z all generate faces of S of rank 1. (This also follows from the
previous proof.) For instance assume that a+ b E (x + y + z), where a, b E S.
Then a+b= n(x+y+z) for some n ~ 0, and ax +bx = ay +by= az +bz = n.
Since 0 ~ax~ az, 0 ~ bx ~ bz and 0 ~ ay ~ az, 0 ~by~ bz, this implies
ax = az, bx = bz and ay = az, by = bz; hence a = az (x + y + z) and
b = bz(x+y+z) E (x+y+z). D
7. FREE EMBEDDING.
In this section we show that a finitely generated commutative semigroup can

be embedded into a free commutative monoid if and only if it is cancellative,
reduced, and power cancellative. This was proved by the author [ 1970E] and
Hochster [1972]; a third proof is given in Rosales & Garcia-Sanchez [1999].
We give yet another proof, which uses geometry as in Section 6, and shows that
there is an embedding into a free commutative monoid with the same rank. This
stronger Free Embedding Theorem appears to be a folk theorem.
I. First we note:
Lemma 7.1. Let C be a reduced convex cone in a finite dimensional real
vector space V. For any inner product on V, there exists p E V such that
p • s > 0 for all s E C\ 0.
Proof. C\0 is convex, since C is reduced. Hence there is a hyperplane H
such that 0 E H and H n (C\0) = 0 (see e.g. Bnzmdsted [1983], Lemma 4.4).
Then H = { s E V I p • s = 0} for some p -1- 0 and p • s -1- 0 for all s E C\ 0.
Since C\0 is convex, either p. s < 0 for all s E C\0, or p. s > 0 for all
s E C\ 0. We can replace p by -p if necessary and arrange that p • s > 0 for
all s E C\0. D
Lemma 7.1 implies that finitely generated reduced cones can be truncated to
polytopes and inherit thereby a number of geometric properties, for which the
reader is generally referred to Bnzmdsted [ 1983].
Corollary 7 .2. Let C be a finitely generated reduced convex cone in a
finite dimensional real vector space V. There exists a closed halfspace K, of
which 0 is an interior point, such that C n K is a polytope. If C spans V,
then C is the intersection of closed halfspaces bounded by the hyperplanes that
are spanned by the facets of C.
Proof. Let H = { v E V I p • v = 1 } and K = { v E V I p • v ~ 1},
where p • s > 0 for all s E C\ 0 as in Lemma 7 .1. Then 0 lies in the interior
I
{ v E V p. v < 1} of K. Every nonzero generator c of C has a projection in H,
which is the intersection of H and the ray from 0 through c; these projections,
together with 0, generate C n K as a convex set. Hence C n K is a polytope.
If C spans V, then so does C n K. Hence C n K is the intersection of
closed halfspaces bounded by the hyperplanes that contain the facets of C n K
(see e.g. Brendsted [1983], Corollary 9.6). One of these facets is C n H and
the corresponding halfspace in K; the remaining halfspaces are bounded by the
7. FREE EMBEDDING. 55
hyperplanes that are spanned by the facets of C, and their intersection is C. D

2. The main result in this section is:
Theorem 7.3. For a finitely generated commutative monoid S the following
conditions are equivalent:
(1) S is cancellative, power cancellative, and reduced;
(2) S can be embedded into a finitely generated free commutative monoid;
(3) S can be embedded into a finitely generated free commutative monoid
with the same rank as S.
Proof. (3) implies (2), and (2) implies (1). Now let S be a c.m. which is
finitely generated by s 1 , ... , sm E S and c.pc.r. Let G = G(S), X be a basis
of G, and G, G, and S be as in Section 5. G has an inner product
(L:xEX ax X) • (L:xEX bx X) = L:xEX ax bx ·
S is a reduced cone by Proposition 5.8; by Lemma 7.1 there exists p E G such
that p. s > 0 for all s E S\0; equivalently, p. s > 0 for all s E S\0.
We improve on Lemma 7.1 as follows. We may assume that s 1 , ... , sm i= 0.
In G, the non empty set
{p E G I p. s > 0 for all s E S\0} = {p E G I p. si > 0 for all i}

is open and contains a point with rational coordinates. Therefore we may assume
that p E G. Let
H = {hEG!p.h=O}.
We show that there exists k E H such that every coordinate of q = p + k is
positive. Indeed p = L:xEX Px x has at least one positive coordinate Pt, since
p. s 1 > 0. If Py ~ 0, then IPylt + PtY E H; hence
h = l:yEX,py;:::;o (IPylt+PtY) E H
has coordinates hx ~ 0 for all x E X and hy = Pt > 0 whenever Py ~ 0. If
A E Q, A > 0 is large enough, then Ah E H and every coordinate of q = p + Ah
is positive.
The semigroup S can now be embedded into the positive cone of G, as
follows. Since p has rational coordinates, G is the direct sum G = H E9 Qp and
every g E G can be written uniquely in the form
g = 7r(g) + tp(g) p
with 1r(g) E H and r.p(g) E Q. Then p. si = r.p(si) (p. p) and r.p(si) > 0 for
all i. Let
T(g) = 7r(g) + r.p(g) >.q,
where >. E Q, >. > 0. Since q fj. H, T is an injective linear transformation.
If >. is large enough, then every coordinate of T(si) is positive, since every
coordinate of r.p(si) q is positive; hence we can choose >. so that every coordinate
of T(s 1), ... , T(sm) is positive. Then T(S) ~Sis contained in the positive cone
of G. The coup de grace is that the positive coordinates of T(s 1 ), ... , T(sm) all
have a common denominator n > 0; then nT(s 1 ), ... , nT(sm) all have positive
integer coordinates, and generate a subsemigroup or submonoid nT(S) ~ S of
the free commutative monoid on X, which has the same rank as S. D
The hypothesis that S is finitely generated is very necessary in Theorem
7.3. The additive semigroup Q+ of all positive rationals is cancellative, power
cancellative, and reduced, but not finitely generated, and cannot be embedded into
a free commutative monoid, let alone one of rank 1.
Corollary 7.4. Let S be a finitely generated c.pc.r. c.m. The elements of S
are separated by finitely many homomorphisms of S into N.
Proof. When S is a submonoid of Fx, there is for each x E X an x-th coor-
dinate homomorphism L:xEX ax x ~------+ ax of S into N; these homomorphisms
separate the elements of S. D
3. A finitely generated c.pc.r. c.m. S can also be embedded into a free c.m.
F with the quasi-universal property that every homomorphism of S into a free
c.m. factors (though perhaps not uniquely) through the embedding S ----+ F;
we say that F (together with S ----+ F) is a quasi-universal free c.m. of S.
This was proved by the author [ 1970E] using properties of orthogonality in free
abelian groups. Examples show that a true universal property cannot be required
in general, and that the quasi-universal free c.m. of S may all have greater rank
than S. Consequently, the embedding constructed in Theorem 7.3 does not usually
have a universal or quasi-universal property.
4. Finitely generated submonoids of finitely generated free c.m. are known
as affine semigroups. By the Free Embedding Theorem, these semigroups can
be defined abstractly as finitely generated c.pc.r. c.s.
Affine semigroups have been studied from various viewpoints. The pa-
pers by Rosales [1997] and Fischer, Morris, & Shapiro [1997] contain (among
other things) interesting constructions; Bruns & Gubeladze [1996] and Gube-
ladze [ 1998] are especially penetrating. See also Ewald & Wessels [ 1991], Alt-
8. KRULL MONOIDS. 57
man [ 1996], and Thoma [ 1996]. Rosales & Garcia-Sanchez developed a "glu-
ing" construction [ 1995] and gave algorithms that recognize normal affine semi-
groups, or recognize normality in affine semigroups [1999N]. See also Ljubenova
& Cvetkova [ 1967] and Hoa [ 1991].
Like numerical semigroups, affine semigroups appear in algebraic geometry.
They are related to toric varieties (see e.g. Hochster [1972], Neeb [1992], and the
book by Oda [ 1988]). The semigroup ring k[S] of an affine semigroup S is also
the coordinate ring of an algebraic variety and has been studied for various ring
properties that are relevant to that variety. Affine semigroups whose semigroup
ring is Goren stein or Cohen-Macaulay were characterized by Trung & Hoa [ 1986]
and studied by Rosales, Garcia-Sanchez, & Urbano-Blanco [ 1998] and Rosales
& Garcia-Sanchez [ 1998C], [ 1999F]. Rosales & Garcia-Sanchez [ 1995] studied
complete intersection semigroups. Affine semigroups also appear as Weierstrass
semigroups of sets of points (Delgado [ 1990]) and as semigroups of values for
several valuations (Delgado [ 1988]). See also Bouvier & Gonzales-Springberg
[1992], [1995]. This list does not include a number of papers which focus on
algebraic geometry rather than semigroups.
8. KRULL MONOIDS.
In this section we show that several constructions yield (up to isomorphism)

the same class of semigroups. These semigroups, called Krull monoids, are the
central concept in factorization theory and its applications to rings and have been
studied by Chapman, Geroldinger, Halter-Koch, Krause, Lett!, and others.
1. A Krull monoid is a commutative and cancellative monoid S such that
there exists a free c.m. F and a homomorphism & : S ----+ F (called a divisor
theory for S) such that
(Dl) &a ~J-C &b in F implies a ~J-C b in S;
(D2) for every c E F there exist a 1 , ... , an E S such that &a 1 1\ .. . I\ &an = c.
This concept has roots in the ideal theory of commutative rings. Its first abstract
formulation is due to Clifford [ 193 8].
Proposition 8.1. A cancellative c.m. S with group of units U is a Krull
monoid if and only if S /U is a Krull monoid In fact, there is a one-to-one
correspondence between the divisor theories for S and the divisor theories for
S /U; the latter are injections.
Proof. Let S be a Krull monoid and {) : S ---+ F be a divisor theory for

S. Since a ~J{ b in S implies aa ~J{ {)b in F, and J{ is the equality on
F, (D1) implies that aa = {)b if and only if a J{ b. By Proposition 5.3, {) is
injective if S is reduced, and in general factors uniquely through the projection
S---+ S/U ~ Sj9i. Since a ~J{ binS if and only if Ua ~J{ Ub in SjU, the
induced homomorphism S/U ---+ F is a divisor theory for S/U. If conversely
{): SjU---+ F is a divisor theory for S/U, then S---+ S/U---+ F is a divisor
theory for S. 0
Thus, divisor theories are basically a matter of c.pc.r. semigroups.
When S has a divisor theory {) : S ---+ F, {) and F are unique up to
isomorphism (Clifford [1938]). With regard to condition (D2) we note
Lemma 8.2. Let S be a submonoid of a free c. m. F. The set
{c E F I there exist a 1, ... , an E S such that a 1 1\ ... 1\ an = c}
is a submonoid of F.
Proof. If c = a 1 1\ ... 1\ am and d = b1 1\ ... 1\ bm, then c + d = (a 1 +
d) 1\ ... 1\ (am+ d) and every ai + d = (ai + b1 ) 1\ ... 1\ (ai + bm); hence
c + d = 1\.2,). (a.2 + bJ.) • 0
2. When F is a free c.m., the intersection ofF and a subgroup of G(F) is
a submonoid of F. The resulting submonoids can be characterized as follows.
A subsemigroup T of a commutative semigroup S is unitary when a E T,
b E S, and ab E T implies b E T. This terminology goes back to Dubreil [ 1941].
A submonoid S of a free c.m. F is unitary in F if and only if a, b E S and
a ~ b in F implies a - b E S.
Proposition 8.3. For a submonoid Sofa free c.m. F the following condi-
tions are equivalent:
(1) S is unitary (if a,b E Sand a~ b in F, then a-bE S);
(2) if a,b E Sand a ~J{ b in F, then a ~J{ b in S;
(3) S = G (S) n F;
(4) S = H n F for some subgroup H of G(F).
IfF is finitely generated, then a unitary sub monoid ofF has only finitely many
irreducible elements, and is generated by its irreducible elements.
Proof. First, S is unitary if and only if (2) holds. Then t E G( S) n F implies
t =a- b for some a,b E Sandt E S; hence S = G(S) n F. If conversely
S = H n F for some subgroup H of G(F), then a,b E S and a~ b in F

implies a - b E H n F = S; hence S is unitary.
Let S be unitary. A minimal (nonzero) element m of S (in the usual partial
order on F) is irreducible, since m of. 0 and m = a+ b with a, b E S implies
0 ~ a, b ~ m in F and 0 < a, b < m is not possible. Conversely an irreducible
element c of S is minimal, since 0 < a < c with a E S would imply c =
a + (c - a) with c - a, a E S and c - a, a of. 0.
We show that every nonzero element of S is a nonempty sum of minimal
(nonzero) elements of S. If this is false, there is, by the descending chain
condition (Proposition 1.5.4), a nonzero element a of S which is minimal such
that a is not a sum of one or more minimal elements of S. In particular a is
not a minimal element of S and there is a minimal element m < a of S. Then
a - m E S, a - m < a, a - m is a sum of minimal elements of S by the choice
of a, and so is a = (a - m) + m; this is the required contradiction.
The minimal elements of S constitute an antichain of F. If F is finitely
generated, then, by Dickson's Theorem (Proposition 1.5.4), this antichain is finite
and S has only finitely many minimal elements. 0
3. Up to isomorphism, a finitely generated free c.m. of rank n is the direct
product F = Nn and consists of all families a = (a 1 , ... , an) of nonnegative
integers, under pointwise addition. Then G(F) = 7l.P consists of all families of
n integers, under pointwise addition and G(F) = Qn. In zn the solutions of a
finite homogeneous system of linear equations with integer coefficients
ci 1 x 1 + ... +cinxn =0 (i = 1,2, ... ,m)
constitute a subgroup H. The nonnegative solutions of the same system constitute
a submonoid H n F of F. These submonoids are sometimes called Diophantine,
as they are solution sets of systems of homogeneous linear Diophantine equations.
Proposition 8.4. Let F = Nn. For a submonoid S of F the following
(1) S is unitary and ma E S, m > 0, and a E F implies a E S;
(2) S = G (S) n F;
(3) S = H nF for some subspace H of G(F);
(4) S = K nF for some pure subgroup K of G(F);
(5) Sis the nonnegative solution set of a finite homogeneous system of linear
equations with integer coefficients.
As before, G ( S) is the subspace of G (F) generated by G (S) . A subgroup

K of G(S) is pure when maE K, m > 0, and a E G(F) implies a E K;
equivalently, if G(F) I K is torsion-free.
Proof. (1) ====? (2). By Proposition 5.7, every element of G(S) can be written
in the form glm for some g E G(S) and m > 0, and every element of G(F)
can be written in the form g I m for some g E G (F) and m > 0. If S is unitary
and maE S, m > 0, and a E F implies a E S, then t E G(S) n F implies
mt E G (S) n F for some m > 0, mt E S, and t E S; hence S = G (S) n F.
(2) ====? (3) since G(S) is a subspace of G(F).
(3) ====? (4) since K = H n G(F) is a pure subgroup of G(F) when H is
a subspace of G (F) .
(4) ====? (3). When K is a pure subgroup of G(F), then K is a subspace

of G(F), and K = K n G(F), since t E K n G(F) implies mt E K for some
m > 0 and t E K. Hence K n F = K n F.
(3) ====? (5). In the rational vector space G(F) every subspace H is the
solution set of a finite homogeneous system of linear equations with rational
coefficients. Multiplying every equation by a common denominator of its coef-
ficients yields an equivalent system with integer coefficients, whose nonnegative
solution set is H n F.
(5) ====? (3). The nonnegative solution set S of a homogeneous system of
linear equations with integer coefficients is the intersection of F and its solution
set in G(F), which is a subspace H of G(F).
(3) ====? (1 ). If H is a subspace of G (H), then S is unitary, as a ~ b,

a, b E S implies a - b E H n F = S; and mt E S, m > 0, and t E F implies
mt E H and t E H n F = S. D
4. More generally, the integer solutions of a homogeneous system of linear
inequalities with integer coefficients
cil x 1 + ... + cin xn ~ 0 (i = 1, 2, ... , m)
constitute a submonoid of F = Nn.
Proposition 8.5. LetS be a reduced subsemigroup of 71P such that G(S) =
zn. The following conditions are equivalent:
(1) S is finitely generated and g E G(S), m > 0, and mg E S implies
g E S;
(2) S is finitely generated and S = G(S) n S;

(3) S is finitely generated and S = G(S) n S;
(4) S = G(S) n C for some finitely generated reduced convex cone C of
Qn;
(5) S = G(S) n C for some reduced convex cone C ~ IR.n that is generated
by finitely many elements of Qn;
(6) S is the set of integer solutions of a finite homogeneous system of linear
inequalities with integer coefficients.
Then S has only finitely many irreducible elements, and is generated by its
irreducible elements.
As in Section 5, S ~ Qn and S ~ IR.n are the convex cones generated by S.
Proof. (1) ~ (2). If (1) holds, then g E G(S) nS implies mg E S for

some m > 0 and g E S; hence G(S) n S = S.
(2) ~ (3). Let g E G(S) n S. Then there exist a, b, a 1 , ... , ak E S such that
g =a- b = .\ 1 a 1 + ··· + ,\k ak for some .\ 1 , ... , ,\k E JR., .\ 1 , ... , ,\k > 0. Since
a 1 , ... , ak E zn it follows from Lemma 5.9 that q0 (-g)+ q1 a 1 + ··· + qk ak = 0
for some q0 , q1, ... , qk E Q, q0 , q1, ... , qk > 0. Hence g E S and (2) implies
(3 ); the converse holds since S ~ G (S) n S ~ G (S) n S.
(2) ~ (4) since S is a reduced cone by Proposition 5.7 and is generated as
a convex cone of Qn by the generators of S.
(4) ~ (5). Let C be a finitely generated reduced convex cone of Qn. Then
C is a convex cone of IR.n that is generated by finitely many elements of Qn. By
- - -
Proposition 5.8, C is reduced and C n Qn = C. Hence G(S) n C = G(S) n C.
(5) ~ (6). Let S = G(S) n C for some reduced convex cone C ~ IR.n
that is generated by finitely many elements of Qn. Then C spans IR.n, since
G (S) = zn . By Corollary 7.2, C is the intersection of finitely many closed
halfspaces, bounded by the hyperplanes spanned by the facets of C. Thus C is
the solution set of a finite homogeneous system of linear inequalities
cilx 1 + ... +cinxn ~ 0 (i = 1,2, ... ,m).
Hence S = G(S) E C is the set of integer solutions of a finite homogeneous
system of linear inequalities
ci 1 x 1 + ... + cinxn ~ 0 (i = 1,2, ... ,m).
62 II. CANCELLATIVE 8EMIGROUPS.
The hyperplanes spanned by the facets of C have equations with rational coeffi-
cients, since they are spanned by generators of C which have rational coordinates
in IRn . Hence we can arrange in the above that all coefficients cij are rational.
Multiplying each inequality by a common denominator then yields a finite homo-
geneous system of linear inequalities with integer coefficients, of which S is the
set of integer solutions.
(6) ===? ( 1). Let S be the set of integer solutions of a finite homogeneous
system of linear inequalities
!'i(x) = cilxl + ... +cinxn ~ 0 (i = 1,2, ... ,m).
with integer coefficients. If g E G(S) and mg E S for some m > 0, then

l'i ( mg) ~ 0 for all i, l'i (g) ~ 0 for all i, and g E S. Also { x E 7!.P I l'i (x) = 0
for all i} is a subgroup of S; since S is reduced, l'i ( x) = 0 for all i implies
x = 0 . Therefore the homomorphism
/' : X f-----1- (1'1 (x), ... , f'm(x)), zn ----+ zm
IS Injective. We see that f'(S) ~ F = wm. As in the proof of Theorem 7.3,
f'(S) is a unitary submonoid ofF: if !'(a)~ f'(b) in F, then !'i(a) ~ f'i(b) for
all i, !'i(a-b) ~ 0 for all i, a-bE S, and !'(a) -{'(b)= !'(a-b) E I'(S).
By Proposition 8.3, S ';;::! !'(S) has only finitely many irreducible elements, and
is generated by its irreducible elements; in particular, S is finitely generated. D
A c.pc.r. semigroup S is normal when ng E S implies g E S when n > 0
and g E G(S); equivalently, when a+ nb = nc with a, b, c E S and n > 0
implies a = nd for some d E S. Unitary submonoids are normal, but normal
submonoids need not be unitary. For instance the (free) submonoid S of F{x,y}
generated by x + 2y and 2x + y is normal, but is not unitary, since 2x + 4y,
2x + y E S but 3y rJ. S; also, 3x + 3y E S but x + y rJ. S.
Normal c.pc.r. semigroups are also called integrally closed and root closed;
the integral closure or root closure of S is
{ g E G ( S) I ng E S for some n > 0}.
(A larger complete integral closure {g E G(S) I ng +a E S for some n > 0
and a E S} has been used e.g. by Clifford [ 193 8] and Geroldinger [ 1996].)
Hochster [ 1972] showed that a c.pc.r. c.m. is finitely generated and normal if and
only if its semigroup algebra (over any field K) is a normal domain (is noetherian
and integrally closed), and then it is a Cohen-Macaulay ring. A cancellative and
power cancellative c.m. is normal if and only if its semi group algebra (over any
field K) is integrally closed (Gilmer [1984]).
5. Preliminary to our main result we show:

Lemma 8.6. Let S be a reduced subsemigroup of zn such that G ( S) = zn
and S is the set of integer solutions of a finite homogeneous system of linear
inequalities with integer coefficients. There exist homomorphisms <'\, ... , 8r :
zn --+ z with the following properties:
(1) Im 8i = z and s = {x = (x1, ... , xn) E zn I 8ix ~ 0 for all
i = 1,2, ... ,r};
(2) there exist ajk E S (j, k = 1, 2, ... , r) such that 8k akk = 1 and 8kajk >
0, 8j ajk = 0 whenever j # k;
(3) 8 1 , ... , 8r span Hom (zn, Q);
(4) the convex coneD~ Hom(zn,Q) generated by 8 1 , ... , 8r is reduced;
(5) C = D n Hom (zn, Z) has finitely many irreducible elements, and is
generated by its irreducible elements;
(6) 8 1 , ... , 8r are irreducible in C;
(7) the group Hom (zn, Z) is generated by the irreducible elements of C.
Proof. Let S be the set of integer solutions of a finite homogeneous system
of linear inequalities
ci 1 x 1 + ... + cin xn ~ 0 ( i = 1, 2, ... , m)
with integer coefficients. Then S = zn n P, where P is the convex cone

P = { (x 1, ... , xn) E :!Rn I ci 1 x 1 + ... + cinxn ~ 0 for all i = 1,2, ... ,m}.
P is reduced since S is reduced, and spans JRn since G(S) = zn. By Corollary
7.2, P is the intersection of finitely many distinct closed halfspaces
Ki = {(x 1 , ... ,xn) E :!Rn I di 1 x 1 + ... +dinxn ~ 0} (i = 1,2, ... ,r)
such that every hyperplane
Hi = {(x1, ... ,xn) E ]Rn I di1x1 + ... +dinxn = 0}
contains a facet of B. Since Hi contains a facet of S and is spanned by

elements of S which have rational coordinates in JRn, Hi has an equation with
rational coefficients, and we can arrange that all dij are rational. Multiplying by a
common denominator we can then arrange that di 1, ... , din are integers; dividing
by gcd (di 1, ... , din) we can further arrange that di 1, ... , din are relatively prime.
Then
is a homomorphism of zn into Z and Im 8i = Z. Also
S = {x = (x 1 , ... ,xn) E zn ioix ~ 0 for all i = 1,2, ... ,r}

since S = zn nP and P = K 1 n ... n Kr, and (1) holds.
(2). If n = 1, then P ~ lR is a closed halfspace, r = 1, 8 1 is an isomorphism,
and (2) holds. Now assume n ~ 2.
If j,k ~ r and j =/= k, then 8kajk > 0, 8jajk = 0 for some ajk E S:
indeed S n Hj rz_ H k, since Hj is spanned by S n Hj and Hj rz_ H k .
Now let k ~ r. We have 8kg = 1 for some g E G(S). Since S CZ- Hk we

have 8ka 1 > 0 for some a 1 E S. Since Hk n P is a facet of P and is spanned
by elements of S, Hk contains linearly independent elements a 2 , ... , an of S.
Then a 1, a 2 , ... , an are linearly independent and g = q1a 1 + q2a 2 + ··· + qn an
for some q1 , ... , qn E ()). Since a 2 , ... , an E Hk we have 8k(q 1 a 1 ) = 8kg = 1
and q1 > 0. Let Pj be the greatest integer Pj ~ qj and
a= q1a1 +(q2-p2)a2+ .. ·+(qn -pn)an = g-p2a2- ... -pnan E G(S).

Then 8ia 0 for all i, since q1 > 0 and qj- Pj ~ 0 for all j > 1, and a E S;
~
also aka= akg = 1, since a 2, ... , an E Hi. Thus 8akk = 1 for some akk E S.
(3). Every homomorphism 1 : zn ----+ ()) extends to a unique linear trans-
formation "( : Qn ----+ Q. Let V = Qn and W be the subspace of the dual
space V* spanned by 81 , ... , 8r. If W ~ V*, then { u E Q)n I w(u) = 0 for
all w E W} has dimension at least 1 and there exists u = ( u 1 , ... , un) E Qn,
u =/= 0 such that 8iu = 0 for all i. Multiplying u by a common denominator
of u 1, ... , un yields a E zn, a =/= 0 such that 8ia = 8ia = 0 for all i. Then
a E S and -a E S, a contradiction since S is reduced. Therefore 81 , ... , 8r
span V* ; every linear transformation "( : Q)n ----+ ()) is a linear combination of
81 , ... , &r with rational coefficients; and every homomorphism 1 : zn ----+ ()) is
a linear combination of 8 1 , ... , or with rational coefficients.
(4). D consists of all linear combinations of 8 1 , ... , or with nonnegative
rational coefficients. If q1 8 1 + ··· + qr or = 0 with q1, ... , qr E ()), q1, ... , qr ~
0, then q1 8 1 a + ... + qrara = 0 for all a E S, qJ)ia = 0 for all a E Sand
all i since qi 8ia ~ 0 for all i, and qi = 0 for all i since 8ia > 0 for
some a E S. If now 1 E D and -1 E D, then 1 = q1 8 1 + ·· · + qr or for
some q1, ... , qr ~ 0, -1 = p 1 8 1 + ··· + Pr or for some p 1, ... , Pr ~ 0,
(p1 + q1)81 + ... + (Pr + qr)ar = 0, Pi+ qi = 0 for all i, qi = 0 for all i,
and 1 = 0.
(5). C is the set of all homomorphisms 1 : zn ----+ Z that can be written as
o
linear combinations I_ = q 1 1 + ··· + qr or with q1, ... , qr E Q and q1, ... , qr ~
0. Then G( C) ~ Hom (zn, Z) is, like Hom (zn, Z), a finitely generated free
abelian group. Also C is the intersection of G(C) and the convex cone of G(C)
generated by o1 , ... , or. By Proposition 8.5, C has finitely many irreducible
elements, and is generated by its irreducible elements.
(6). We show that oj = q 1 o1 + ··· + qr or, with q 1 , ... , qr E Q and
q1 , ... , qr ~ 0, implies qj = 1 and qi = 0 for all i =J j; hence oj cannot be
written as a sum oj = 1 1 + 12 with 1 1 , 1 2 E C and 'h, 1 2 =J 0, oj, and is
irreducible in C.
Let oj = q1 o1 + ··· + qr or as above. By (2) there is for every k =J j some
ajk E S such that okajk > 0 = ojajk. Then q1 1 ajk o + ··· + qr or ajk =
ojajk = 0, qioiajk = 0 for all i, qkokajk = 0, and qk = 0. Hence qi =0
for all i =J j; then qj = 1, since oja > 0 for some a E S.
(7). Let o1, ... , or, or+ 1, ... , 0

be all the irreducible elements of C. By
8
(3) every homomorphism 1 : zn ----+ Z is a linear combination of o1 , ... , or

with rational coefficients. Therefore there exist rational numbers q1 , ... , qr ~ 0
such that 1 + q 1 o1 + ··· + qror ED. For any integers p 1 ~ q 1 , ... , Pr ~
qr we now have 1 + p 1 o1 + ··· + PrOr E D n Hom(zn,z) = C. Hence
1 + p 1 o1 + ··· + Pr or is a linear combination of o1 , ... , o 8 with nonnegative
integer coefficients, and 1 is a linear combination of o1 , ... , 08 with integer
o
coefficients. Thus o1 , ... , 8 generate Hom (zn, Z). D
6. The main result in this section is:
Theorem 8.7. Let S be a reduced subsemigroup of zn such that G(S) =
zn. The following conditions are equivalent:
(1) S is a Krull monoid(= S has a divisor theory);
(2) Sis finitely generated and normal (if g E G(S), m > 0, and mg E S,
then g E S);
(3) S is the set of integer solutions of a finite homogeneous system of linear
inequalities with integer coefficients;
(4) S is isomorphic to the nonnegative solution set of a finite homogeneous
system of linear equations with integer coefficients.
(5) S is isomorphic to a unitary sub monoid of a finitely generated free com-
mutative monoid.
Proof. (1) ==? (2). Let {) : S --+ F be a divisor theory for S. Let
g E G (S) , so that g = a - b for some a, b E S. If mg E S for some m > 0,
then ma = mb + mg ~']{ mb in S, {) (ma) ~ {) ( mb) in F, oa ~ {)b, a = b + c
for some c E S by (Dl), and g = c E S. Thus S is normal.
Since G(S) ='liP there exist a 1 , ... ,an E S that are linearly independent in
Qn . Let a = a 1 + ··· + an . Every b E S is a linear combination of a 1 , ... , an
with rational coefficients. If m E N+ is large enough, then ma - b E S,
ma- bE G(S) n s = s' and o(b) ~ o(ma) in F. By (D2), every generator X
of F can be written in the form x = 8b 1 1\ ... 1\ {)bk for some b1 , ... , bk E S;
hence x ~ 8( ma) for some m > 0 and x ~ oa. Therefore F is finitely
generated. Then {)S is finitely generated by Proposition 8.3 (since a, b E {)S and
a ~'Ji b in F implies a ~'Ji b in oS), and so is S ~ as.
(2) ==? (3) by Proposition 8.5.
(3) ==? (1). Let S be the set of integer solutions of a finite homogeneous
system of linear inequalities with integer coefficients. Lemma 8.6 provides ho-
momorphisms 8 1 , ... , {)r : 'liP --+ Z such that Im {)i = Z,
S = {x=(x 1 , ... ,xn)EZnl8ix~Oforalli=1,2, ... ,r},

and other wonderful properties. We show that the homomorphism
{) : X f----+ (81 X, ... , OrX)
of S into the free c.m. F = Nr is a divisor theory for S.
Let a, b E S. If oa ~ ob in F, then {)i (a - b) ~ 0 for all i and a - b E S;
hence (D 1) holds.
Part (2) of Lemma 8.6 provides ajk E S such that oakk = 1 and okajk > 0,
ojajk = 0 whenever j -=/= k. Then
ek = oa 1k 1\ ... 1\ oark E F
has the following properties. The k-th coordinate of ek is
min ( aka1k 1\ ... 1\ akark) = 1,

since okakk = 1 and okajk > 0 for all j -=/= k. But, for all j -=/= k, the j-th
coordinate of ek is
min ( aja1k 1\ ... 1\ ajark) = 0,

since {) .a .k 0. Thus ek is the k -th standard generator of F, and every
J J
generator of F is an infimum of elements of 88. Then (D2) follows from

Lemma 8.2.
(3) ~ (4). Let 8 1 , ... , or be as above, D be the convex cone generated
by 81 , ... ,or in Hom(zn,Q), and C = D n Hom(zn,z). By Lemma 8.6,
c has finitely many irreducible elements 81' ar+1' 0 including
0 0' ar, 0 0 0' as'
81 , ... , or. The homomorphism
8:a !-----+ (81 a, 0 0 0' {)sa)' zn --+ zs
is injective on S: if a,b E S and 8a = 8b, then 8(a- b) = 0, 8i(a- b) = 0
for all i ~ r, and a - b E S by part (I) of Lemma 8.6; similarly b - a E S,
and a = b since S is reduced.
We show that Im 8 is a pure subgroup of zs. Indeed assume that me= 8a,
where a = (a 1, ... , an) E zn, e = (e1, ... , es) E zs, and m > 0. Then m
divides 8ia for all i. Then m divides 1a for every homomorphism 1 : zn --+ Z,
since 8 1 , ... , {)s generate Hom(zn,z) by Lemma 8.6. In particular m divides
every aj. Hence a= mb for some bE zn, me= 8(mb), and e = 8b.
Finally we show that 88 = Im 8 n Ns: if e = (e1 , ... , es) E zs and ei ~ 0

for all i, and e = 8a for some a E zn, then 8ia ~ 0 for all i and a E S.
Hence it follows from Proposition 8.4 that 88 ~ S is the nonnegative solution
set of a finite homogeneous system of linear equations with integer coefficients.
(4) ~ (5) by Proposition 8.4.
(5) ~ (2). Let T be a unitary submonoid of a finitely generated c.m. F.
Then T is finitely generated by Proposition 8.3. If g E G(T), m > 0, and
mg E T, then g = a - b for some a, b E T, ma ~ mb in F, a ~ b, and
g = a - b E T. Thus T is normal. 0
Theorem 8.7 reached its present shape in Chapman, Krause, & Oeljeklaus
[2001]. Its proof is due to several authors. The equivalence of (2) and (5) was
proved by Hochster [1972]. The equivalence of (I), (2), and (3) is due to Halter-
Koch [1993] and Lettl [1988]. The proof that (3) implies (4) is from Kainrath
& Lettl [2000]. The semigroup C in this proof is isomorphic to Hom ( S, N), so
that 818 : S --+ F is the quasi-universal free c.m. of S in Grillet [1970E]. The
purity of Im 8 was shown by John [1974].
Chapter III.
SEMILATTICE DECOMPOSITIONS.
Semilattice decompositions were introduced in the 1950's and provide the

earliest structural insight into commutative semigroups in general, showing, for
instance, that finite commutative semigroups are assembled from groups and
nilsemigroups arranged along a semilattice. They have been the mainstay of
commutative semigroup theory for many years (see e.g. Petrich [ 1977]).
It turns out that every commutative semigroup has a semilattice decomposition
into semigroups called archimedean that lack proper decompositions. There are
two cases, Clifford semigroups and separative semigroups, where a semigroup
can be effectively reconstructed from its archimedean components. Unfortunately
this reassembly problem is in general very difficult. Hence much of this chapter
is about some classes of archimedean semigroups that are comparatively simple
in structure. Finitely generated archimedean semigroups will be considered in
Chapter VI.
Some of these results extend to more general semigroups; interested readers
are referred to Clifford & Preston [1961], Petrich [1973], Grillet [1995], Nagy
[2001], and Ciric [2002].
As before we abbreviate "commutative semigroup" into "c.s." and "commu-
tative monoid" into "c.m.".
1. GENERAL RESULTS.
I. This chapter studies c.s. (=commutative semigroups) through their homo-

morphisms into semi lattices. When cp is a homomorphism of a semigroup S into
a semilattice Y, then Sa = { x E S I cp( x) = a} is a subsemigroup of S for
every a E Y, and Sa Sb = Sab (= SaAb) for all a,b E Y; this gives an overall
picture of the multiplication on S.
In general, S is a semilattice of semigroups when there exists a semilattice
Y and a partition S = UaEY Sa of S into subsemigroups Sa (one for every
69
70 Ill. 8EMILATTICE DECOMPOSITIONS.
a E Y) such that Sa Sb ~ S ab for all a, b E Y; equivalently, when there exists

a congruence e on S (a semilattice congruence, or semilattice decomposition)
such that s;e is a semilattice (then y = s;e and the subsemigroups sa are the
e-classes). This concept is due to Clifford [1954].
Other decompositions into direct unions have been studied, most notably
by Croisot [1953], Poyatos [1966], [1967], Szep [1969], [1970], [1973], and
Migliorini & Szep [1979], [1981] (see also Jurgensen, Migliorini, & Szep [1991]).
These decompositions are typically more detailed, at some cost in multiplicative
properties.
2. A semilattice congruence provides the most detailed picture of the multi-
plication on S when it has the most classes; this also tends to make the classes
smaller and simpler in structure. It turns out that there is always a smallest
semilattice congruence and that its classes have no further semilattice decom-
positions. This result was first obtained for finite (more generally torsion) c.s.
by Schwarz [1953] and Thierrin [1954] and extended to all c.s. by Tamura and
Kimura [1954] and to all semigroups by Yamada [1955]. For other proofs, see
Clifford and Preston [1961]; Petrich [1963F], [1973]; Grillet [1995].
When S is commutative, a clue to the smallest semilattice congruence is
provided by the following result, which constructs a preorder ~N that contains
the divisibility preorder ~!}{. Recall that a ~!]{ b if and only if a E S 1 b, if
and only if every ideal which contains b also contains a. An ideal P of S is
(completely) prime when xy E P implies x E P or y E P.
Lemma 1.1. For elements a and b of a c.s. S the following conditions are
equivalent:
(1) every prime ideal which contains b also contains a;
(2) an = bt for some n > 0 and some t E S1 ;
(3) an = bt for some n > 0 and some t E S;
(4) cp(a) ~ cp(b) for every homomorphism <p of S into a semilattice.

We write a ~N b when a and b satisfy these conditions; ~N is a preorder
and contains ~!}{ (if a ~!]{ b, then a ~N b).
Proof. (1) ===} (2). If(2) does not hold, then the principal ideal S 1 b contains
no positive power an of a. Let P be the union of all ideals of S which contain
no positive power of a. Then P is an ideal of S, by Proposition 1.3 .1, and P
contains b E S 1b but contains no positive power an of a. If x ~ P and y ~ P,
then S 1x must contain some positive power am of a, otherwise we would have
1. GENERAL RESULTS. 71
S 1 x ~ P; S 1 y must contain some positive power an of a, otherwise we would

have S 1 y ~ P; hence S 1 xy = S 1 x S 1 y contains a positive power am+n of a,
S 1 xy rt_ P, and xy 1. P. Thus P is a prime ideal. Since P contains b but not
a, (1) does not hold. (The similar argument in commutative rings uses Zorn's
Lemma; for semigroups, Proposition 1.3.1 makes this unnecessary.)
(2) ===} (3). If an = bt for some t E S 1 , then an+l = bat with at E S.
(3) ===} (4). If an= bt for some n > 0 and t E S, and r.p is a homomorphism
of S into a semilattice Y, then r.p(a) = r.p(a)n = r.p(b) A cp(t) ;£ r.p(b) in Y.
(4) ===} ( 1). Let P be a prime ideal of S. Then Q = S\ P is a subsemigroup
of S and QQ ~ Q, PQ ~ P, P P ~ P. Hence there is a homomorphism r.p
of S into the semilattice {0, 1} which sends every x E P to 0 and every y E Q
to 1. Then bE P implies cp(a) ;£ r.p(b) = 0 by (4) and a E P. 0
3. A c.s. S is archimedean when a ;£N b holds for every a, b E S;
equivalently, when there exist for every a, b E S some n > 0 and t E S such
that an = tb. (This terminology originated with ordered abelian groups: the
positive cone of a totally ordered abelian group G is archimedean as a semigroup
if and only if G is archimedean as a totally ordered abelian group.)
Theorem 1.2. On every commutative semigroup S there is a smallest
semilattice congruence N, under which
aN b <¢:=:::} a ;£N b and b ;£N a
<¢:=:::} am = tb and bn = ua for some m, n > 0 and t, u E S.
Every homomorphism of S into a semilattice factors uniquely through the pro-
jection S ~ S jN. Moreover, every N -class is archimedean.
Proof. a ;£N b implies ac ;£N be, since cp(a) ;£ r.p(b) implies cp(a) cp(c) ;£
r.p(b) cp( c) for every homomorphism r.p of S into a semilattice. Hence N is a
congruence. Every element of S jN is idempotent, since a 2 ;£N a and a ;£N a 2
for every a E S; hence N is a semi lattice congruence.
N is contained in every semilattice congruence, by ( 4) in Lemma 1.1. When
r.p is a homomorphism of S into a semi lattice Y, then the subsemigroup im r.p
of Y is a semilattice and ker r.p is a semilattice congruence. Hence N ~ ker r.p
and r.p factors uniquely through the projection S ~ S jN by Proposition 1.2.4.
Finally, let A be an N -class of S. Let a, b E A. Then an = tb for some
n > 0 and t E S, and an+l = (at)b, with atE A since a ;£Nat and at ;£N a.
Thus A is archimedean. 0
By Theorem 1.2, N is a semilattice decomposition of S into semigroups with
72 Ill. SEMILATTICE DECOMPOSITIONS.
no further semilattice decompositions. The N -class of x is the archimedean

component of x, which we denote by A(x) (rather than Nx). Since ::1{ s;;; N,
every archimedean component of S is a union of ::1{ -classes.
Proposition 1.3. An archimedean semigroup contains at most one idempo-
tent.
Proof. When e and f are idempotent, e ~:N f implies e = en ~JC f and
e ~ f. Hence e N f implies e = f. D
Archimedean semigroups are considered in more detail in the next sections.
4. The semilattice Y(S) = S/N of archimedean components of S is also
called the universal semilattice of S, owing to its universal property in Theo-
rem 1.2.
Corollary 1.4. For every c.s. S there is an injective homomorphism
E(S) -+ Y(S), which is an isomorphism when S is finite.
Proof. Restricting the projection S -+ Y(S) = S/N to E(S) s;;; S yields
a homomorphism E(S) -+ Y(S) which is injective by Proposition 1.3. If S
is finite, then every archimedean component of S contains an idempotent, by
Corollary 1.5.9, and the homomorphism E(S) -+ Y(S) is surjective as well. D
Petrich [ 1962] showed that Y (S x T) ~ Y (S) x Y (T) for all c.s. S and T.
When <p : S -+ T is a homomorphism, the homomorphism S -+ T -+
Y(T) factors uniquely through Y(S):
S~T
1 1
Y(S) -----+ Y(T)
Y(r,o)
Thus <p induces a homomorphism Y(<p) : Y(S) -+ Y(T) such that the diagram
above commutes; when A is an archimedean component of S, then <p(A) is
archimedean and Y (<p) sends A to the archimedean component of T which
contains <p( A) .
5. The semilattice composition problem is the problem of reconstructing a
semigroup S from a semi lattice decomposition S = UaEY Sa of S. This amounts
to constructing the multiplication maps Sa X Sb -+ Sab which determine the
operation on S.
The semilattice composition problem is discussed in some detail in Petrich
[ 1973] and Grillet [ 1995]. It is even more difficult that the ideal extension
problem: indeed it requires the simultaneous solution of many ideal extension
2. CLIFFORD SEMIGROUPS. 73
problems, since, in a semilattice of semigroups (Sa)aEY, every Sab is an ideal

of Sa U Sb U Sab. Even the reassembly of a commutative semigroup from its
archimedean components seems quite difficult, and has been solved only in highly
particular cases: when the archimedean components are groups (Clifford [ 1941 ]),
cancellative (Hewitt & Zuckerman [ 1956]), or cyclic (Arendt and Stuth [ 1970]).
The first two cases are considered in the next section. See also Yamada [1965]
and Tamura [1971].
Archimedean semigroups appear in every commutative semigroup but can
be expected to be simpler to some extent. The extent however is not so large
as to have borne many general results. Maximal cancellative subsemigroups of
archimedean semigroups were studied by Putcha [1975], [1980], Tamura [1976],
[1977S] and Gale & Tamura [1976]. See also Putcha [1973], [1976], Lord [1978],
and the survey by Tamura [1978]. A general construction in terms of trees and
integer-valued factor sets was given by Tamura [ 1966], [ 1968] (see also Grillet
[ 1995]). But the study of archimedean semi groups has generally been limited to
smaller classes.
In this chapter we construct three such classes. Other classes were constructed
by Brown & LaTorre [1965] (uniquely divisible archimedean semigroups) and
Tully [1966], [1974] (archimedean semigroups on which ~9-C is a tree order);
nilsemigroups will be considered in Chapter IX.
2. CLIFFORD SEMIGROUPS.
Propositions 1.3.8 and 11.3.7 suggest that the semilattice composition problem
can be solved when the components are monoids or are cancellative.
1. A semilattice of groups, also called a Clifford semigroup, is a semigroup
with a semilattice decomposition S = UaEY Ga in which every semi group Ga is
a group. Commutative Clifford semigroups are also called commutative regular
semigroups and commutative inverse semigroups.
Clifford [ 1941] showed that the multiplication on a semi lattice of groups is
determined by a functor. A semilattice Y can be regarded as a small category,
whose objects are the elements of Y and whose morphisms are the ordered pairs
(a, b): a--+ b such that a~ bin Y; composition is given by (b, c) o (a, b)= (a, c)
and the identity morphism on a is (a, a) . A group valued functor G = (G, 'Y)
on a semilattice Y is a functor on this category: it assigns a group Ga to every
element a of Y, and assigns a homomorphism 'Yb : Ga --+ Gb to every pair
74 III. 8EMILATTICE DECOMPOSITIONS.
(a, b) of elements of Y such that a ~ b, so that ')'~ is the identity on Ga and

')'~ o 1'/: = ')'~ whenever a ~ b ~ c. The following result is due to Clifford [ 1941].
Petrich [ 1973] proved a similar result for semi lattices of monoids in general.
Theorem 2.1 (Clifford). Let G = (G, 'Y) be a group valued functor on
a semilattice Y. Let S be the disjoint union S = UaEY ( G a x {a}) with
multiplication
(g,a)(h,b) = ('Y~b(g) 'Y~b(h), ab)

for all a, b E Y, g E G a, h E Gb. Then S is a semilattice of groups, and, up
to isomorphism, every semilattice of groups can be constructed in this fashion.
Moreover, S is commutative if and only if G a is abelian for every a E Y.
Proof. The multiplication in the statement is associative:
(g,a)((h,b)(k,c)) (g' a) ('Yge (h) 'Ybe ( k) , be)
(/'~be (g) 'Y~be hie (h) 'Ybe ( k)), abc)

(/'~be (g) 'Y~be (h) 'Y~be ( k)' abc)
((g,a)(h,b))(k,c).
Also, the subsemigroup Sa = { (g, a) E S I g E G a } of S is isomorphic to G a,

and Sa Sb ~ Sab. Thus S is a semilattice of groups.
Conversely let S be a semilattice S = UaEY G a of groups Ga. Let ea be
the identity element of Ga. When a > b in Y, G a U Gb is an ideal extension
of Gb and is by Proposition 1.3 determined by a (partial) homomorphism 'Y/: :
Ga----+ Gb, namely f'/:(g) = geb = ebg for all g EGa. In particular eaeb =
ebea = f'/:(ea) = eb. Also ')'~ is the identity on Ga and ')'~(g) = gea = eag
for all g EGa. If a~ b ~ c, then ')'~(/'/:(g))= gebee =gee= f'~(g). Now
(G, 'Y) is a group valued functor on Y.
When g E G a and h E Gb, then the products gh and heab in S are in Gab,
so that g h = g he ab = ge ab he ab = 'Y~b (g) f'~b (h) as calculated in Gab . Therefore
S is isomorphic to the semigroup UaEY ( Ga x {a}) in the statement. D
A semilattice is directed downward; a group valued functor on a semilattice
can be regarded as a direct system. Tamura & Hamilton [1971] showed that the
direct limit lim G a is the universal group of the corresponding Clifford semi group.
---+
Congruences on Clifford semigroups were considered by Jarek [1964] and others.
Chapter VIII lists a number of additional results: due to Theorem 2.1, Clifford
2. CLIFFORD 8EMIGROUPS. 75
semigroups have been studied extensively and have provided a testing ground for
many ideas and problems.
2. A semilattice of cancellative semigroups is a semigroup with a semi1attice
decomposition S = UaEY C a in which every semi group C a is cancellative.
These semigroups can be characterized as follows:
Proposition 2.2. For a c.s. S the following are equivalent:
(1) S is a semilattice of cancellative semigroups;
(2) every archimedean component of S is cancellative;
(3) in S, x 2 = xy = y 2 implies x = y;
(4) for every n > 0, xyn = yn+l and xny = xn+l implies x = y.
Proof. (2) ===:;. ( 1) is clear.
(1) ====? (3). Let S = UaEY Ca be a semilattice of cancellative semigroups
Ca . Let x 2 = xy = y 2 in S. If x E Ca and y E Cb, then x 2 E Ca , y 2 E Cb,
a = b, and x = y since C a is cancellative.
(3) ===:;. (4). This is clear if n = 1. For n > 1 we proceed by induction on
n. Let abn = bn+ 1 and ban = an+ 1 , where n > 1. Then (with b0 = 1 E S 1
if n = 2)
(abn-1 )2 = abn-2 abn abn-2bn+1 = (abn-1)bn,
bn+1 bn-1 = (bn)2,
and abn- 1 = bn by (3). Exchanging a and b yields an- 1b =an. Then a= b

by the induction hypothesis.
(4) ===:;. (2). Let A be an archimedean component of S. Assume that
a, b, e E A satisfy ae = be. Since A is archimedean we have am = ex and
bn = ey for some m,n > 0 and x,y EA. We may assume that m = n: if, say,
m < n, then an = exan-m, with xan-m E A. Then
abn = aey =bey= bn+l and ban= bex = aex = an+l
and a = b by (4 ). 0
Proposition 2.2 implies a result of Kimura & Tsai [1972] that power cancella-
tive archimedean semigroups are cancellative, which was first showed by Brown
& LaTorre [1966] for uniquely divisible semigroups.
A c.s. S is separative when it satisfies the equivalent conditions in Propo-
sition 2.2. (We will show in the next chapter that the characters of S then
separate the elements of S .) Separative semigroups were introduced by Hewitt &

Zuckerman [ 1956] and Schwarz [ 1956]. Gluskin [ 1971] determined their dense
extensions (such that every homomorphism which is injective on the semigroup
is injective on the extension), which are also separative. Rompke [ 1975] studied
the semigroups of fractions of reductive separative c.s.
3. The following result is due to Hewitt & Zuckerman [ 1956].
Proposition 2.3. Every commutative semilattice S = UaEY C a of cancella-
tive semigroups C a can be embedded into a semilattice T = UaEY G (Ca) of
abelian groups. If every C a is archimedean, then every homomorphism of S
into a semilattice of groups extends uniquely to T.
Proof. We may assume that the groups G( Ca) are pairwise disjoint. If xjy =
x' jy' in G( Ca) and zjt = z' jt' in G( Cb) (with a, bE Y and x, y, x', y1 E Ca,
z,t,z ' ,t' E cb,) ten
h xz,y t ,x' z ' ,y't' E cab• xy ' = x ' y, zt' = z 't, xzy 't ' =
x'z'yt in Cab• and xzjyt = x'z'/y't' in G(Cab). Therefore a multiplication
on T = UaEY G(Ca) is well defined by (xjy)(zjt) = xzjyt. This makes T
into a c.s., in which G( Ca) G( Cb) ~ G( Cab); thus T is a semilattice of abelian
groups. sis a subsemigroup ofT ifwe identify X E ca and x 2 jx E G(Ca),
sothat (x 2 jx)(y 2 jy) = (xy) 2 jxy.
Now assume that every Ca is archimedean and let cp : S --+ U be a
homomorphism of S into a semi lattice of groups U = UcEZ G c. Every cp( C a)
is archimedean and is contained in a single group Gc. The restriction of cp to
Ca then extends uniquely to a homomorphism G(Ca) --+ Gc, which sends
xjy E G(Ca) to cp(x) cp(y)- 1 , where cp(y)- 1 is the inverse in Gc (Proposition
II.2.1). This provides a mapping 'ljJ: xjy 1---t cp(x) cp(y)- 1 ofT into U.
We show that 'ljJ is a homomorphism; then 'ljJ is the only homomorphism
which extends cp. Let x,y E Ca and z,t E Cb. Let cp(Ca) ~ Gc, cp(Cb) ~ Gd
and ec denote the identity element of Gc. Then cp(y) and cp(t) have inverses
in Gc and Gd and commute with each other in Gcd, since S is commutative.
Also cp(y) commutes with ed in the Clifford semigroup U. Hence
cp(y) cp(t)- 1 cp(t)ecd = cp(y) cp(t)- 1 cp(t) = cp(y) ed = ed cp(y)

= cp(t)- 1 cp(t) cp(y) = cp(t)- 1 cp(y) cp(t) = cp(t)- 1 cp(y) cp(t) ecd'
which in Gcd implies cp(y) cp(t)- 1 = cp(t)- 1 cp(y). The same calculation with
cp(y) replaced by cp(y)- 1 (which commutes with cp(t)) shows that cp(y)- 1 com-
mutes with cp(t)- 1 . Hence
2. CLIFFORD SEMIGROUPS. 77
and cp(yt)- 1 = cp(y)- 1 cp(t)- 1 in Gcd· Then
'1/J(xzjyt) cp(xz) cp(yt)- 1 = cp(x) cp(z) cp(y)- 1 cp(t)- 1

cp(x) cp(y)- 1 cp(z) cp(t)- 1 = '1/J(x/y) '1/J(zjt)
and 'ljJ is a homomorphism. D
By Clifford's Theorem the operation on T is determined by homomorphisms
l'b : G(Ca) ---+ G(Cb) (a ~ b), namely, l'b(xjy) = (xjy)(z/z) = xzjyz,
for all x, y E C a and z E Cb. We see that l'b is induced by the canonical
homomorphism T : ca ---+ G(Cb) of the ideal extension ca u cb of cb in
Lemma 11.3 .5.
4. When S is separative, Proposition 2.4 provides a universal semilattice of
groups T = UAEY(S) G(A) of S; that is, T is a semilattice of groups, S is a
subsemigroup of T, and every homomorphism of S into a semilattice of groups
extends uniquely to T. In fact, every c.s. has a universal semilattice of groups.
A congruence e on a c.s. S is separative when S/ e is separative; equiva-
lently, when x 2 e xy e y 2 implies X e y.
Proposition 2.4. On every c.s. there is a smallest separative congruence S,
for which
x S y if and only if xny = xn+ 1 and xyn = yn+l for some n > 0.
Every homomorphism of S into a separative semigroup factors uniquely through
the projection s
---+ s;s.
Proof. If S is separative, then S is the equality on S by Proposition 2.2. In
general, S is reflexive and symmetric. Let x S y and y S z. Then xmy = xm+ 1 ,
xym = ym+ 1 , ynz = yn+ 1, yzn = zn+ 1 for some m,n > 0, and
xzmn+m zn = x(yzn)m zn = xym zmn+n
ym+1 zmn+n = (yzn)m+1 = zmn+m+n+1.
Exchanging x and z yields zxmn+n+m = xmn+n+m+ 1 . Hence x S z. Thus

S is transitive.
Let x S y and z E S. Then xny = xn+ 1, xyn = yn+l for some n > 0
and
Exchanging x and y yields (yz)(xz)n = (xz)n+l. Hence xz S yz. Thus S is

a congruence.
Let x 2 S xy and xy S y2 . Then x 2m+ly = (x 2)m(xy) = (x 2 )m+l =
x 2m+ 2 and xy 2n+l = (xy)(y 2 )n = (y 2 t + 1 = y 2n+ 2 for some m,n > 0. We
may assume that m = n: if, say, m < n, then multiplying the first equality by
x 2n- 2m yields x 2n+ly = x 2n+ 2 . Hence x S y. Thus S is separative.
Finally let <p : S ---+ T be a homomorphism, where T is separative. If x S y
in S, then xny = xn+l and xyn = yn+l for some n > 0, <p(x)n <p(y) = <p(x )n+l
and <p(x) <p(y)n = <p(y)n+l, and <p(x) = <p(y). Hence S ~ ker <p and <p factors
uniquely through the projection S ---+ SIS by Proposition 1.2.4. If in particular
e is a separative congruence on s s
and <p : se
---+ 1 is the projection, then
S ~ ker <p = e. D
In Proposition 2.4, SIS is the universal separative semigroup of S. It
should be denoted by S(S), but this would not improve the notation.
Corollary 2.5. Every c.s. has a universal semilattice of groups, which is
also the universal semilattice of groups of its universal separative semigroup.
Proof. Semilattice of groups are separative. Hence every homomorphism <p
of a c.s. S into a semilattice of groups factors uniquely through the projection
S ---+ SIS, and thence extends uniquely to the universal semi lattice of groups
T of SIS. Hence <p factors uniquely through S ---+ SIS ---+ T. D
The universal separative and Clifford semigroups of a c.s. have been known
implicitly since Hewitt & Zuckerman [1956] and Schwarz [1956]. C.s. that map
onto their universal Clifford semi group were investigated by Head [ 1971 H] and
McAlister & O'Carroll [1971].
3. COMPLETE ARCHIMEDEAN SEMIGROUPS.
By Proposition 1.3, an archimedean semigroup has at most one idempotent.

We call an archimedean semigroup complete when it has an idempotent. For
instance, finite archimedean semigroups are complete; an interesting smaller class
was studied by Trueman [ 1979], [ 1980].
Complete archimedean semigroups are comparatively simple in structure.
1. A nilsemigroup is a semigroup S with a zero element in which every
element x is nilpotent: xn = 0 for some n > 0. Commutative nilsemigroups
are archimedean and contain an idempotent; so do abelian groups. Conversely:
3. COMPLETE ARCHIMEDEAN 8EMIGROUPS 79
Proposition 3.1. A c.s. S which contains an idempotent e (for instance, a

finite c.s.) is archimedean if and only if it is an ideal extension of a group by a
nilsemigroup; then S has a kernel K = He = eS and S / K is a nilsemigroup.
Proof. Let S be archimedean. Let a E He and x E S. Since S is archime-
dean we have e = en = axy for some n > 0 and y E S. Also ax = eax. Hence
ax E He. Thus the maximal subgroup He is an ideal of S; then He = eS. For
every x E S we now have xn = ey E He; hence the Rees quotient Sj He is a
nilsemigroup. Every nonempty ideal of S contains all powers of some element,
contains an element of He, and contains He; thus He is the kernel of S.
Conversely let the c.s. S be an ideal extension of a group G by a nilsemigroup
N. If a E S\G, then an= 0 for some n > 0 in S/G = N, and an E G. If
x is the inverse of an in G, then the equalities an = ean, e = anx show that
a ;;;N e and e ;;;N a. Hence S is archimedean. D
If in the above G is trivial, then S ~ N is a nilsemigroup; if N = 0 is
trivial, then S = G is a group.
As a first application of Proposition 3.1, we prove:
Corollary 3.2. For a c.s. S the following conditions are equivalent:
(1) every archimedean component of S contains an idempotent;
(2) every element of S has a power in some subgroup of S;
then the homomorphism E(S) ----+ Y(S) is an isomorphism.
Proof. If (1) holds, then the archimedean component A of x E S contains an
idempotent f, and xm = jt, fn = f = xmu for some m,n > 0 and t,u E S;
hence xm belongs to the subgroup H f of G, which proves (2). Conversely, (2)
implies that every archimedean component A contains an element of a subgroup
of S; since a subgroup is contained in a single JC -class, which is contained
in a single N -class, A contains the entire subgroup and contains its identity
element. Condition (1) implies that the injective homomorphism E(S)----+ Y(S)
in Proposition 1.4 is surjective. D
Semigroups with property (2) have been called ?T-regular, pseudo-invertible,
and epigroups; groups, nilsemigroups, and finite semigroups have this property.
2. The ideal extensions in Proposition 3.1 are determined by partial homo-
morphisms, as in Proposition 1.3.8. Our next result constructs the partial homo-
morphisms themselves from homomorphisms of abelian groups; this constructs
all archimedean semigroups with idempotent from nilsemigroups and groups.
Proposition 3.3. Let P be a commutative partial semigroup. There exist an
abelian group G(P) and a partial homomorphism 'Y : P ----+ G(P) such that
every partial homomorphism r.p of P into an abelian group G factors uniquely
through 'Y (r.p = f, o 'Y for some unique group homomorphism f,: G(P) ----+G);
P~ G(P)
~1€
G
in the additive notation, G(P) is the abelian group generated by the elements
of P, subject to one relation a + b = ab for every a, b E P such that ab is
defined in P. Then the group PRom (P, G) of all partial homomorphisms of P
into G is isomorphic to the group Hom(G(P), G) of all group homomorphisms
of G(P) into G.
Because of its universal property we call G(P) the universal (abelian) group
of P. If Pis an actual semigroup, then G(P) is the universal group of P defined
in Section 11.2, and Proposition 3.3 provides a presentation of G(P).
Proof. Let G(P) be the abelian group generated by P, subject to a+ b = ab
whenever ab is defined in P. Let 'Y : P ----+ G(P) be the canonical mapping.
Every defining relation a+ b = ab holds in G(P) via"(; that is, 'Y(a) + "f(b) =
'Y (ab) whenever ab is defined in P. Thus 'Y is a partial homomorphism.
Let G be any abelian group, written additively, and r.p : P ----+ G be a partial
homomorphism. Then r.p(ab) = r.p(a) + r.p(b) whenever ab is defined in P and
every defining relation holds in G via r.p. Therefore r.p factors uniquely through
'Y ( r.p = f, o 'Y for some unique group homomorphism f, : G(P) ----+ G.)
Since G is abelian, PHom(P,G) and Hom(G(P), G) are abelian groups
under pointwise addition. Iff, : G(P) ----+ G is a group homomorphism, then
f, o 'Y : P ----+ G is a partial homomorphism. This yields a homomorphism
f, f----7 f, o 'Y of Hom (G (P), G) into PHom (P, G), which is bijective by the
above. D
3. When S is a semigroup with zero, presentations of S readily yield pre-
sentations of G(S\0):
Proposition 3.4. If S is generated, as a commutative semigroup or monoid
with zero, by a set X subject to relations ui =vi (i E I) and wj = 0 (j E J),
where ui = vi =/= 0 and x =/= 0 in S for all x E X and i E I, then G(S\0) is
isomorphic to the abelian group generated by X subject to all relations ui =vi
(i E I).
Proof. Let F = Fx be the free c.s. (or c.m.) on the set X, written multi-
3. COMPLETE ARCHIMEDEAN SEMIGROUPS 81
plicatively. Up to isomorphism, S = (F U {0}) je, where e is the congruence on

FU{O} generatedbyallpairs (ui,vi) and (wj,0). The e-classofO isanideal
of F U {0} and is the union of { 0} and an ideal Z of F. By the hypothesis,
Z does not contain ui, vi, or any x E X; in particular the canonical mapping
t :X --+ S sends X into S\0.
There is a smaller congruence 13 on F which is generated by the pairs

(ui, vi); the quotient semigroup T = F/13 is the c.s. (or c.m.) generated by
X subject only to the relations ui = vi ( i E J). When a, b E F\Z we show
that a e b if and only if a 13 b; in particular F\ Z is a union of 13 -classes. If
indeed a e b, then by Proposition 1.2.9 there is a sequence x 1, ... ,xn E FU {0}
such that a = x 1 , xn = b, and every pair (xk, xk+l) has the form (tui, tvi),
(tvi,tui), (twj,0), or (O,twj), where t E F 1 U{O}. Since x 1, ... ,xn are
in the equivalence class of a and b it i not possible that xk = 0 or xk E Z.
Therefore every pair (xk, ik+l) has the form (tui, tvi) or (tvi, tui), and a 13 b.
Now S\0, which is the set of all e-classes contained in F\Z, is a partial
subsemigroup of T, and the canonical mappings t : X --+ S\ 0, ,., : X --+ T
agree on X.
S\0
/
X---+ T
1~
K,
Let G be the abelian group (written multiplicatively) generated by X subject

to all relations ui = vi ( i E I), and 'f/ : X --+ G be the canonical mapping.
Since the defining relations of T hold in G via 'f/, there is a homomorphism
T : T --+ G such that the following diagram commutes:
The restriction of T to S\ 0 is a partial homomorphism 1 of S\ 0 into G.

When <p : S\ 0 --+ A is a partial homomorphism of S\ 0 into an abelian
group A, then every relation ui = vi holds in A via r.p o t; hence there is a
unique homomorphism ~ : G --+ A such that r.p o t = ~ o 'f/.
82 III. SEMILATTICE DECOMPOSITIONS.
X~S\0 ~A
~~1 ~
G
Since ~,(X) generates S we have eo 'Y = <p. Thus G (together with 'Y) is
isomorphic to the universal group of S\ 0. 0
Example 3.5. Let N be the commutative nilsemigroup
N = ( a, b I a8 = a 5 b4 = b6 = 0, a 2 b4 = a 5 b2 ).
By Proposition 3.4, G(N\0) is the abelian group
(a, b I a 2 b4 = a 5 b2 ) = ( a, b I b2 = a3 )
or (in the additive notation) (Z EB Z) / K, where K is the subgroup generated by

(3,-2). Now (3,-2), (-1,1) isabasisofZEBZ(asshownb yitsdeterminant);
hence G(N\0) ~ Z.
The partial homomorphism 'Y : N\ 0 ---+ Z is found as follows. The images of
a and b in Z EB Z are ( 1, 0) = (3, -2) + 2 ( -1, 1) and (0, 1) = (3, -2) + 3 ( -1, 1) ;
in Z ~ (Z EB Z)/K, the images of a and bare 2 and 3. Thus 'Y(a) = 2 and
"f(b) = 3. For any abelian group G, PHom(N\0, G) ~ Hom(Z,G) ~ G; a
partial homomorphism <p : N\ 0 ---+ G sends a and b to 9 2 and 9 3 for some
unique 9 E G. 0
N = (a,bl a3 =a2 b=ab2 =b4 =0, a 2 =ab=b3 ).
This example was communicated to the author by Prof. Volkov. By Proposition

3.4, G(N\0) is the abelian group
( a, b I a2 = ab = b3 ) .
We see that a= b holds in G, and then a 2 = b3 yields a= b = 1. Thus G(N\0)
is trivial (even though N\ 0 itself is not trivial). For any abelian group G, the
only partial homomorphism <p : N\ 0 ---+ G is the trivial partial homomorphism
<p(a) = <p(b) = 1. 0
4. N-SEMIGROUPS.
An N-semigroup is an archimedean semigroup without idempotent which is

4. N-SEMIGROUPS 83
cancellative. N-semigroups have also been called N -semigroups, 17-semigroups,

and SJt-semigroups. This section contains a structure theorem, due to Tamura
[ 1957], which constructs N-semigroups in terms of abelian groups and factor sets
and started numerous papers on N-semigroups. A similar construction applies to
all reduced cancellative c.s. (Hamilton, Nordahl, & Tamura [1975]).
1. First we prove two general properties. By Proposition 11.5.5, there is on
any c.s. S a smallest power cancellative congruence 'J, given by
The following result is due to Kimura & Tsai [1972].
Proposition 4.1. When S is archimedean, then S j'J is archimedean and
cance/lative; moreover S has no idempotent if and only if S j'J has no idempo-
tent.
Proof. When S is archimedean, then T = S j'J is archimedean, power can-
cellative, separative (since x 2 = y 2 implies x = y ), and cancellative by Propo-
sition 2.2. By Proposition 11.5.5, S has no idempotent if and only if T has no
idempotent. D
N-semigroups thus provide a basic framework for all archimedean semigroups
without idempotent.
Corollary 4.2. When S is archimedean without idempotent, then x # xy
for all x,y E S.
Proof. If x = xy, then xy = xy 2 , xy 'J xy 2 , y 'J y 2 since S j'J is can-
cellative, S j'J contains an idempotent, and S contains an idempotent by Propo-
sition 4.1. 0
2. Now let S be an N-semigroup. We regard S as a subsemigroup of its
group of fractions G (S) = s- 1 S, whose elements can then be written in the
form xy- 1 with x,y E S.
Lemma 4.3. LetS be an N-semigroup and a E S. Every element of G(S)
can be written in the form anp for some unique n E Z and p E S\aS. Every
element of G(S) can be written in the form anp for some unique n E Z and
p E S\aS.
Proof. Let s E S. Since S is archimedean, ak = st for some k > 0 and
t E S. Then s (j. ak S, otherwise ak = aktu for some u E S, contradicting
Corollary 4.2. Thus s is not "infinitely divisible" by a: there is a greatest n ;:; 0
such that s = anp for some p E S, and then p (j. aS by the choice of n.
Now let g = xy- 1 E G(S), where x, y E S. Since S is archimedean
84 III. 8EMILATTICE DECOMPOSITIO NS.
we have an = yt for some n > 0 and t E S and xy- 1 = xta-n in G(S).

Therefore g = amp for some m E Z and p E S\aS.
Assume that amp= anq holds in G(S), where m,n E Z and p,q E S\aS.
If m < n, then p = an-mq E aS, a contradiction. If similarly m > n, then
q = am-np E aS, a contradiction. Therefore m = n, and then p = q. D
Let a E S and A be the cyclic subgroup of G(S) generated by a. Corollary
4.2 implies that a has infinite order in G(S), so that A~ Z. The abelian group
G = G(S)/A is the structure group of S relative to a. By Schreier's Theorem
on group extensions, G ( S) can be reconstructed from G = G ( S) /A by means
of an integer valued factor set on G. Lemma 4.3 implies that this construction
also yields S. (The role of group extensions in his 1957 construction was known
to Tamura [ 1970F].) The details are as follows.
By Lemma 4.3 the projection 1r: G(S) --+ G(S)/A = G induces a bijection
S\aS --+ G. Let u f-------1- Pu be the inverse bijection. For instance p 1 = a,
since a ~ aS by Corollary 4.2 and 1r sends a E A to the identity element of G.
By Lemma 4.3, every element of S can be written in the form anpu for some
unique n ~ 0 and u E G; then 1r(anpu) = 1r(pu) = u.
Let u,v E G. Since 1r(pu) = u, 1r(pv) = v we have 1r(pupv) = uv and
Pu Pv = aO'u,v p uv
for some unique nonnegative integer 0'u v. (Tamura denotes 0'u v by I( a, {3) .)
' '
Then
(ampu)(anpv) = am+nPuPv = am+n+au,v Puv.
Thus S is isomorphic to N x G when multiplication on the latter is
(m, u) (n, v) = (m + n + 0'u v , uv).
'
(More generally, Tamura [ 1970F) showed that every factor set on G with values
in Z is equivalent to a nonnegative factor set; Nordahl [1977] showed that all
additive subgroups of ffi. have this property.)
Lemma 4.4. In an N-semigroup the following holds for all u, v, w E G:
(1) O'lu=l;
'
(2) O'v,u = O'u,v;
(3) (J uv
'
+ (J uv w =
'
(J u vw
'
+ (J v w;
'
(4) for every u E G there exists k > 0 such that u,u k
0' > 0.
4. N-SEMIGROUPS 85
Proof. (1) holds since p 1pu = apu; (2) holds since PvPu = PuPv; (3) holds
smce
(PuPv)Pw aO"u,vp p aO"u,v+O"uv,w p

uv w (uv)w '
Pu (PvPw) aO"v,wp p aO"u,vw+O"v,w p
u vw u(vw) ·
In fact, properties (2) and (3) hold in every abelian group extension.
AI so Pu2 -a
- O"u,u Pu2, Pu3 -- PuPu2 -a
- O"u,u+O"u ,u2 Pu3, an d , by m
. d uc t"1on,
Pun -_ aSn Pun, w here sn -_ ~

~O<k<n (} u,uk.
Since S is archimedean we have p~ E aS for some n > 0. Then sn > 0 and
u,u k > 0 for some 0 < k < n. D
(}
Theorem 4.5 (Tamura). Let G be an abelian group and (} : G x G --+ .N

satisfy the conditions in Lemma 4.4. Let S = .N x G with multiplication given
by
(m,u)(n,v) = (m+n+(Juv ' uv).

'
Then S is an N-semigroup and, up to isomorphism, every N-semigroup can be
constructed in this fashion.
Proof. S is a semigroup by (3) since
( (m, u)(n,v)) (p,w) (m+n+(Juv , uv) (p,w)

'
(m+n+p+(J uv +(Juvw' uvw),
' '
(m, u) ( (n, v)(p, w)) (m,u) (n+p+(Jvw , vw)
'
(m + n + p + (} u vw + (} v w ' uvw);
' '
S is commutative by (2); and S is cancellative: if (m, u) (n, v) = ( m, u) (p, w),
then (m + n + (} u v , uv) = (m + p + (}u w , uw), uv = uw, and v = w; then
' '
m + n + (Ju v = m + p + (Ju w yields n = p. If (n, u) E S is idempotent, then
' '
(n+n+(Juu ' u 2 ) = (n,u) 2 = (n,u), u 2 = u, u = 1, and n+n+(Juu =
'
n implies n = 0 and (} 1 1 = 0, contradicting ( 1); therefore S contains' no
'
idempotent. We show that S is archimedean.
Let a= (0, 1) E S. We have a (m,u) = (m+ 1, u) by (1) and, by induction,
an (m, u) = (m + n, u). In particular an+ 1 = (n, 1) and (m, u)(1, u- 1 ) =
(n, 1) = an+ 1 for some n > 0, so that a ~N (m, u). On the other hand,
(n, u) = a (n - 1, u) ~N a whenever n > 0 . If n = 0, then p = (0, u) satisfies
by (4), sn > 0 for some n > 0, and then pn ~N a and p ~N a. Thus

(m, u) N a for all (m, u) E S and S is archimedean. Hence S is an N-
semigroup. Conversely we saw that, up to isomorphism, every N-semigroup can
be constructed as in the statement. D
Completing Theorem 4.5, Biggs, Sasaki, & Tamura [ 1965] determined all
factor sets that satisfy the conditions in Lemma 4.3 when G is finitely generated.
Sasaki [1965], [1966], Higgins [1966], [19691], and Lord [1979] investigated
when two N-semigroups are isomorphic. Tamura [1974B], [1975] also proved a
second structure theorem for N-semigroups, in terms of one abelian group G and
a positive mapping from G to lR. He also obtained descriptions as pullbacks
(Tamura [1975], [1977T]). More precise constructions exist when the structure
group is Z (Tamura [1977N]) or zn (Gale & Tamura [1980]).
N-semigroups have been studied extensively. Congruences on N-semigroups
(especially, N-semigroup congruences) were investigated by Dickinson [1972],
Tamura [1973C], and Hamilton [1978N]. Hall [1972] and Kobayashi [1982] stud-
ied ideal extensions ofN-semigroups. Kobayashi [1974] also studied their homo-
morphisms into lR. Homomorphisms of N-semigroups were studied by Tamura
[1974B], [1977T]. Simons [1963] proved a Mackey-like homomorphism theorem.
Some smaller classes have also been studied. Power joined N-semigroups
are considered in Section 6 and finitely generated N-semigroups in Section VI.6.
Tamura [ 1973C] called an N-semigroup irreducible when it has no proper non-
trivial N-semigroup congruence; equivalently, when it is isomorphic to a sub-
semigroup of JR+; a construction is given in Tamura [19741]. Tamura [1973S]
also studied N-semigroups that cannot be embedded as proper ideals in other N-
semigroups. Arendt [1975] studied N-semigroups in which the intersection of all
maximal congruences is the equality.
N-semigroups have been generalized by Dickinson [1976] and Sasaki [1978],
[1982], [1983], [1984].
5. SUBCOMPLETE ARCHIMEDEAN SEMIGROUPS.
We call an archimedean semigroup S subcomplete when it is isomorphic to a

non empty subsemigroup of a complete archimedean semigroup. These semigroups
5. SUBCOMPLETE ARCHIMEDEAN SEMIGROUPS. 87
are somewhat simpler in structure than archimedean semigroups in general. The

results in this section are due to Grillet [1975C].
1. First we show:
Proposition 5.1. For a nonempty archimedean semigroup S the following
(1) S is subcomplete (S can be embedded into a complete archimedean
semigroup);
(2) S contains a nonempty cancellative ideal;
(3) Sis an ideal extension ofa nonempty cancellative semigroup by a nilsemi-
group;
(4) S can be embedded into the direct product of an abelian group and a
nilsemigroup.
Proof. (1) ===> (2). Let S be a subsemigroup of an archimedean semigroup
T that contains an idempotent e. By Proposition 3 .I, eT is a group. If a E S,
then an = et for some n > 0 and t E T, and S n eT =!= 0 is cancellative
semigroup and an ideal of S.
(2) ===> (3). Let C be a nonempty cancellative ideal of S. Let c E C.
If a E S, then an = d E C for some n > 0 and t E S; hence S/C is a
nilsemigroup.
(3) ===> (4). When S is an ideal extension of a nonempty cancellative
semigroup C, Lemma 11.3.5 provides a homomorphism T : S--+ G( C) which is
the identity on C. Let 1r : S --+ S / C = N be the projection. The homomorphism
s 1-----t (T(s), 1r(s)) of S into G(C) x N is injective, since Tis injective on C
and 1r is injective on S\ C.
(4) ===> (I). The direct product of an abelian group G and a nilsemigroup
N is archimedean, since both G and N are archimedean, and contains an idem-
potent (1, 0). D
2. Not all archimedean semigroups are subcomplete.
Example 5.2 Let T = 1:~+ x N with the associative multiplication
(m,p) (n,q) = (m + n, p V q)
where p V q =max (p,q) is the max operation. In particular, (m,p)k = (km,p);
thus T consists of all powers of all tp = (1, p) E T, with t; t~ = t;t~n .
Define (m,p) = (n,q) if and only if (m,p) = (n,q) or m = n > p V q; =

is an equivalence relation on T since m = n > p V q, n = k > q V r implies
m = k > max (p V q , q V r) p V r. In fact,

~ =
is a congruence on T: if
m = n > p V q, then, for any (k,r) E T, either r ~ p V q, in which case
p V r = r = q V r and (m,p)(k, r) = (n,q)(k,r), or r < p V q, in which case
m + k = n + k > (p V r) V (q V r).
Let 8 = T /=; 8 consists of all positive powers s;, where p ~ 0 and sp
. th e -=-Casso
IS 1 f ( 1,p) ,. th en sPm sqn -- spVq
m+n an d sPn -- sqn w hen n > p,q.
Hence 8 is power-joined: any two elements x and y of 8 have a common
power xm = yn. In particular 8 is archimedean.
Let I be a nonempty ideal of 8. Some s; E I; hence s~- 1 E I if n
is
0 0
large enough. Then s E I and s~ = s~- 1 sn E I. But s~ s = s~ s~, whereas
s0 i- s~ in 8. Thus I is not cancellative, and 8 has no nonempty cancellative
ideal. D
3. On the other hand, a subcomplete archimedean semi group has what may
be called a completion, in spite of certain shortcomings, for instance that the
completion depends on the choice of a cancellative ideal.
Proposition 5.3 Let 8 be an archimedean semigroup with a nonempty
cance/lative ideal C. The restriction to 8\ C of the canonical homomorphism
T : 8 ---+ G(C) is a partial homomorphism and determines an ideal extension
T of G(C) by 8/C which is a complete archimedean semigroup and contains
S as a subsemigroup.
Let cp : S ---+ U be a homomorphism. If U is archimedean with an idem-
potent e and cp( C) ~ He, then cp extends uniquely to a homomorphism of T
into U.
Proof. By definition, the multiplication . on T is as follows: for all s, t E
8\C and g E G(C), g.s = s.g = T(s) g and s.t = T(s) T(t), in case st rJ. 8\C
in 8. In the latter case, T(s) T(t) = T(st) = st since Tis a homomorphism and
is the identity on C; if c E C, then s.c = T(s) c = (sc) c- 1 c = sc. Hence 8 is
a subsemigroup ofT (cf. Lemma 11.3.6). The latter is a complete archimedean
semigroup by Proposition 3.1, since 8/C is a nilsemigroup.
Let U be an archimedean semigroup with an idempotent e and <p : 8 ---+ U
be a homomorphism such that <p( C) ~ He. The restriction of <p to C extends
uniquely to a homomorphism~: G(C)---+ He, and ~(ab- 1 ) = cp(a) cp(b)- 1 E
He for all a, b E C (Proposition 11.2.1 ). Let '1/J : T ---+ U be defined by:
'1/J(g) =~(g) for all g E G(C), '1/J(s) = cp(s) for all s E 8\C. Then '1/J extends
<p. That '1/J is a homomorphism is tedious but straightforward: for all g, h E G (C)
and s,t E 8\C,
5. SUBCOMPLETE ARCHIMEDEAN SEMIGROUPS. 89
'1/J(g.h) = 'lj;(g) 'lj;(h) and 'lj;(s.t) = 'lj;(s) 'lj;(t) are clear when st E S\C;
if s.t E C, then s.t =stand 'lj;(s) 'lj;(t) = r.p(st) = r.p(s.t) = 'lj;(s.t);
if g = ab- 1 , with a,b E C, then s.g = T(s) g = (sa) a- 1 g = sab- 1 and

'1/J(s) 'lj;(g) = r.p(s) ~(g)= r.p(s) r.p(a) r.p(b)- 1 = ~(sab- 1 ) = 'lj;(s.g).
Any homomorphism x : T -----+ U which extends r.p must send G( C) into
the only subgroup He of U; therefore the restriction of x to G( C) must be ~,
whereas the restriction of X to T\ G (C) = S\ C is that of r.p; hence X = 'ljJ. D
4. One could wish for a better completion than Proposition 5.3 provides.
Alas, examples show that this may be impossible, that is, a universal complete
archimedean semigroup may not exist: when S is subcomplete, there may be no
complete archimedean semigroup T 2 S such that every homomorphism of S
into a complete archimedean semigroup extends uniquely to T.
Example 5.4. Let S ~ ~+ be an infinite cyclic semigroup S = {a, a 2 , ... ,

an, ... } . This mischievous semigroup is archimedean and has a nonempty can-
cellative ideal em = {am' am+ 1 ' ... } for every m > 0.
For every m > 0, Proposition 5.3 provides a complete archimedean semigroup
Tm = (S\Cm) U G(Cm) 2 S. Let G = {bn InEZ}~ Z be an infinite
cyclic group. If we identify an and bn for every n ~ m, then Cm becomes
a subsemigroup Cm = { bm, bm+ 1, ... } of G and Proposition ll.2.5 yields
G(Cm) =G. In Lemma II.3.5, T(ak) = (ak am)(am)- 1 = bk+m (bm)-1 = bk.
Hence Tm = {a, a 2 , ... , am- 1 } U { bn InEZ} with multiplication ai aj = ai+j
if i + j < m, ai aj = ai+j = bi+j if i +j ~ m, ai bl = bi+j.
Now assume that there is an embedding, or, more generally, a homomorphism
r.p : S -----+ T of S into a complete archimedean semigroup T, such that every
homomorphism of S into a complete archimedean semigroup factors uniquely
through r.p. Let e be the idempotent of T. Since T is archimedean we have
r.p(ak) = et for some k > 0 and t E T.
Let m > k. If the inclusion homomorphism S -----+ Tm factors through r.p,
that is, if there is a homomorphism 'lj; : T -----+ Tm such that 'lj; o r.p is the identity
on S:
90 III. 8EMILATTICE DECOMPOSITIONS.
then 1/J ( eT) is a subgroup of T m, since eT is a subgroup of T, 1/J ( eT) ~ G,

and ak = 1jJ (<p( ak)) = 1/J ( et) E G. This is not possible when m > k. 0
6. POWER JOINED SEMIGROUPS.
A commutative semigroup S is power joined when, for every a, b E S there

exist positive integers m, n such that am = bn; equivalently, when every two
cyclic subsemigroups of S intersect. These semigroups were first considered by
Abellanas [1965] (for cancellative semigroups only) and McAlister [1968], who
called then rational, due to Corollary 6.4 below.
1. Power joined semigroups are archimedean; the abelian group Z is archi-
medean but not power joined. In fact, an abelian group is power joined if and
only if it is torsion. More generally,
Proposition 6.1. An archimedean semigroup with an idempotent is power
joined if and only if its kernel is torsion.
Proof. Let S be archimedean with an idempotent e. If S is power joined,
then He is power joined and is torsion. If conversely He is torsion, then, for every
x, yES, we have xm, yn E He for some m, n > 0, and then (xm)k = e = (yn)l
for some k, l > 0; thus S is power joined. 0
For semigroups without idempotent we first look at N-semigroups. The fol-
lowing result is due to Levin [ 1968].
Proposition 6.2. An N-semigroup is power joined if and only if its structure
group (relative to any element) is torsion.
Proof. The structure group of an N-semigroup S, relative to a E S, is the
quotient G = G(S)/A of G(S) by the cyclic subgroup A~ Z generated by a.
Assume that S is power joined. For any x, y E S we have xm = aP
and yn = aq for some m, n, p, q > 0; then xmn = anp, ymn = amq, and
(xy- 1 )mn = anp-mq in G(S); hence G = G(S)/A is torsion.
If conversely G is torsion, then for any x, y E S we have (xa- 1 )m E A,
xm = aP for some integers m > 0 and p, and yn = aq for some integers n > 0
and q. Then p, q > 0, otherwise S contains an idempotent 1 = xm a -p or
1 = yn a -q, and xmq = aPq = ynp. Thus S is power joined. 0
For the general case we use Proposition 11.5.5, which provides a smallest
6. POWER JOINED 8EMIGROUPS. 91
power cancellative congruence 'J, for which

When S is archimedean without idempotent, then S I'J an N-semigroup, by
Proposition 4.1. Also, the universal group of S I'J is torsion-free.
Proposition 6.3. An archimedean semigroup S without idempotent is power
joined if and only if the universal group of S I'J has rank 1.
Proof. T = SI'J is an N-semigroup and G(T) has a subgroup A~ Z such
that G = G(T)IA is the structure group ofT. If S is power joined, then
T = SI'J is power joined, G is torsion by Proposition 6.2, and G(T) has rank
1. If conversely G(T) has rank 1, then G is torsion and T is power joined,
by Proposition 6.2; for all x, y E S, we have xm 'J yn for some m, n > 0 and
xkm = ykn for some k > 0; hence S is power joined. D
Corollary 6.4. A c.s. S without idempotent is power joined if and only if

S I'J is isomorphic to a subsemigroup of Q+.
Proof. If S is power joined without idempotent, then, by Proposition 6.3,
G(SI'J) is a torsion free abelian group of rank 1 and can be embedded into
the additive group Ql; hence S I'J is isomorphic to an additive subsemigroup of
Ql. This subsemigroup cannot contain positive numbers and negative numbers,
since S I'J contains no idempotent. Hence S I'J is isomorphic to an additive
subsemigroup of Ql+ .
If conversely S I'J is isomorphic to a subsemigroup of Q+ , then S I'J is
power joined without idempotent. Hence S has no idempotent, by Proposition
11.5.5. Also S is power joined: if x,y E S, then xk 'J y 1 for some k,l > 0,
since S I'J is power joined, and xkn = y 1n for some n > 0. D
Corollary 6.4 was obtained by McAlister [ 1968] and Levin [ 1968]. Other
characterizations of power joined semigroups were given by Levin & Tamura
[1970].
2. Another characterization of power joined N-semigroups was given by
McAlister & O'Carroll [1970] and Sasaki & Tamura [1971].
Lemma 6.5. Let S be a commutative semigroup. For every a E S a
congruence A on S is defined by
x A y if and only if am x = any for some m, n > 0.
If S is archimedean, then S I A is a group. If S is an N-semigroup, then S I A
is the structure group of S.
Proof. It is immediate that A is a congruence; in fact, A is one of the

reversible congruences of Dubreil [ 1941]. We have x A ax for all x E S, so
the A-class of a is the identity element of S I A. If S is archimedean, then for
every x E S there exist t E S and m > 0 such that xt = am A a; therefore
SIA is a group.
If S is an N-semigroup, then A is also the congruence induced on S by the
partition of G (S) into cosets of A = (a) . Hence S I A is a subsemigroup of the
structure group G = G (S) I A. In fact S I A is all of G: for all x, y E S there
exist t E Sand m > 0 such that yt =am, and then xy- 1 A= xta-mA = xtA;
thus every coset of A contains an element of S. D
The following result anticipates Chapter IV.
Proposition 6.6. An N-semigroup S is a subdirect product of its structure
group and SI'J; hence an N-semigroup is power joined if and only if it is a
subdirect product of a torsion group and a subsemigroup of Q+.
Proof. Let a E S and A be the congruence in Lemma 6.5, so that S I A = G,
the structure group of S. If x A y and x 'J y, then xn = yn and ar x = a 8 y
for some n, r, s > 0, anr xn = ans yn, nr = ns by Corollary 4.2, r = s, and
x = y since S is cancellative. Thus A n 'J is the equality on S. Hence the
projections S ----t S I A and S ----t S I'J separate the elements of S and S is a
subd irect product of G = S I A and S I'J, by Proposition IV. 1.1.
If S is power joined, then G is torsion by Proposition 6.2 and S I'J is
isomorphic to a subsemigroup of Q+ by Corollary 6.4. If conversely S is
contained in the direct product G x Q of a torsion group G and a subsemigroup
Q of Q+, then every element of S has a power in {1} x Q; since Q is power
joined, any two elements of S have a common power and S is power joined. D
Power joined N-semigroups were first considered by Levin [1968] and Ta-
mura [ 1970F]. Tamura [ 1970C] showed that every N-semigroup is a subdirect
product of an abelian group and a positive additive subsemigroup of ffi.. Tamura
[ 1973C] also showed that every N-semigroup is a subdirect product of an abelian
group and an irreducible N-semigroup; this leads to the second structure theorem
(Tamura [1974B]).
3. Every c.s. has power joined components, that are smaller than but similar
to its archimedean components.
Proposition 6.7. Every c.s. is a disjoint union of power joined subsemi-
groups.
Proof. Let S be a c.s. When x, y E S, let
6. POWER JOINED SEMIGROUPS. 93
x P y if and only if xm = yn for some m, n > 0.

Then P is reflexive, symmetric, and transitive (if xm = yn and yP = zq, then
xmp = ynp = znq ). Moreover every P-class is a subsemigroup of S: if indeed
xm = yn, then (xy)n = xm+n. 0
The P-classes are the power joined components of S. They were discovered
by Tamura [1970] (and rediscovered by Radelecki [1986]). They are contained
in but generally smaller than the archimedean components of S. Moreover P is
not in general a congruence.
Example 6.8. Let S = {an I n E Z} ~ Z be an infinite cyclic group.
S is archimedean but has three power joined components: U = {1}, P =
{an In > 0}, and N = {an I
n < 0}. P is not a congruence since
PN=S=UUNUP.D
I
Example 6.9. Let X= {x,y} and S = {axx + ayy ax, ay > 0} ~ Fx;
in fact S is an archimedean component of Fx . In S, which is written additively,
axx + ayy P bxx + byy if and only if m(axx + ayy) = n(bxx + byy) for
some m, n > 0, if and only if ax fay = bx/by. Therefore S has infinitely many
power joined components, one component { kpx + kqy I k > 0} for every pair
(p, q) of relatively prime positive integers. D
Tamura [ 1970P] proved that an archimedean semigroup has either 1 or 3
or infinitely many power joined components. McAlister [ 1968] proved that a
commutative semigroup S has the character extension property (all characters
of subsemigroups of S can be extended to characters of S) if and only if its
archimedean components are power joined.
Chapter IV.
SUBDIRECT DECOMPOSITIONS.
Subdirect decomposition is a general construction method, which goes back

to Birkhoff [1944] and was applied to semigroups by Thierrin [1956], Malcev
[1958], Ponizovsky [1962], and Lesohin ([1963A] and later articles). Ponizovsky
decompositions provide another way to assemble finite commutative semigroups
from groups and nilsemigroups arranged along a semilattice, and are particularly
fruitful for later chapters.
Finding all subdirect products of a given family of semigroups is generally
not an easy task. Hence subdirect decompositions, like semilattice decompo-
sitions, yield general descriptions rather than precise structure theorems. They
also uncover two other important classes of commutative semigroups: complete
semigroups and elementary semigroups.
This chapter contains basic properties of subdirect products (Section 1), sep-
arative semigroups (Section 2), nilsemigroups (Section 3), complete semigroups
(Section 4) and elementary semigroups (Section 5). Ponizovsky decompositions
are in Section 4. The study of subdirectly irreducible semigroups, begun in Sec-
tions 1, 3, and 4, requires additional techniques and is continued in the next
chapters.
1. SUBDIRECT PRODUCTS.
1. The direct product of a set (Si)iEI of semigroups is their cartesian

product f1iEJ Si, with componentwise multiplication:
(xi)iEI (yi)iEI = (xi Yi)iEI ·

This is the only operation on rriEJ si such that the projection 7fi : rriEJ si ---t
Si, (xi)iEI f----1- xi is a homomorphism for every i E I. If I = 0, the empty
product f1iEJ Si contains only the empty family and is trivial (f1iEJ Si = {0} ).
95
96 IV. SUBDIRECT DECOMPOSITIONS.
If in general Si ::/= 0 for all i, then IJiEI Si ::/= 0 and the projections 1ri are all
surjective.
A subdirect product of a set (Si)iEI of semigroups is a subsemigroup P
of the cartesian product IJiEI Si such that 1ri ( P) = Si for all i E I. Any
semigroup which is isomorphic to such a subsemigroup is also called a subdirect
product of the semigroups (Si)iEI. For instance, a direct product of nonempty
semigroups is a subdirect product of these semigroups; there usually are many
others.
A subdirect decomposition of a semigroup S describes S as a subdirect
product of a set (Si)iEI of semigroups, the components of the decomposition.
For example the empty semigroup and the trivial semigroup {0} have trivial
subdirect decompositions as subdirect products of the empty family. We are
mostly interested in other subdirect decompositions.
When S is isomorphic to a subdirect product P ~ IJiEI Si, the projections
9! ~ 'lfi
Pi : S ----+ p ---=-t IJiEJ Si ---=--+ Si
are surjective homomorphisms and separate the elements of S in the sense that
x ::/= y in S implies Pi(x) ::/= pi(y) for some i E I. The converse holds:
Proposition 1.1. A semigroup S is a subdirect product of semigroups
(Si)iEI if and only if there exist surjective homomorphisms <pi : S---+ Si which
separate the elements of S.
Proof. If the homomorphisms <pi : S ---+ Si separate the elements of S,
then the homomorphism cp(x) = (<pi(x))iEI of S into IJiEJ Si is injective, and
S ~ Im <p. If every 'Pi is surjective, then 1ri(Im <p) = Im 'Pi = Si for every i
and S ~ Im <p is a subdirect product of the semigroups (Si)iEI. D
2. A semigroup S is subdirectly irreducible in case S has at least two
elements and, whenever S is (isomorphic to) a subdirect product of semigroups
Si (i E I), at least one ofthe projections S---+ Si is an isomorphism. (Thus S
is not a subdirect product of "smaller" semigroups.) Every semigroup with two
elements is subdirectly irreducible.
Proposition 1.2. A semilattice is subdirectly irreducible if and only if it has
two elements.
Proof. Let Y be a semilattice. For every a E Y, the mapping <pa : Y ---+
{0, 1} defined by
<p a ( x) = 1 if x ~ a, <pa (X) = 0 if X t a
1. 8UBDIRECT PRODUCTS 97
is a homomorphism. The homomorphisms <p a separate the elements of Y: if

a =I b in Y, then, say, b ~ a, since a ~ b and b ~ a are not both possible, and
then c.p a (a) = 1, <p a (b) = 0. Hence Y is a subdirect product of two-element
semilattices. If Y is subdirectly irreducible, then Y has just two elements. D
Subdirectly irreducible semigroups in general can be characterized by their
congruences. Call a congruence proper when it is not the equality.
Proposition 1.3. A semigroup S with at least two elements is subdirectly
irreducible if and only if the equality on S is not the intersection of proper
congruences, if and only if S has a smallest proper congruence.
Proof. This follows from Proposition 1.1. When S is a subdirect product of
semigroups Si ( i E I), then the projections 1ri : S ---t Si, which separate the
elements of S, induce congruences on S whose intersection is the equality on S.
If S is not subdirectly irreducible, then one can arrange that none of the 1ri is an
isomorphism; then the equality on S is the intersection of proper congruences. If
conversely the equality on S is the intersection of proper congruences ei ( i E I),
then S is a subdirect product of the quotient semigroups S I ei, but none of the
projections s ---t Slei is an isomorphism, so s is not subdirectly irreducible.
If the equality on S is not the intersection of proper congruences, then the
intersection of all the proper congruences on S is not the equality, and is the
smallest proper congruence on S. If conversely S has a smallest proper con-
gruence e, then the equality on S is not the intersection of proper congruences,
since the latter contains e.D
3. Subdirectly irreducible semigroups are of interest because of a general
property of algebraic systems, due to B irkhoff [ 1944].
Theorem 1.4 (Birkhoff). Every semigroup is a subdirect product of subdi-
rectly irreducible semigroups; every commutative semigroup is a subdirect prod-
uct of subdirectly irreducible commutative semigroups.
Proof. The union of a chain of congruences on a semigroup S is a congruence
on S (Proposition 1.2.8). For every a,b E S with a =I b, there exists, by Zorn's
Lemma, a congruence ea b on S which is maximal such that a ea b b does not
' '
hold. By definition, a e b for every congruence e ~ ea,b; hence ea,b is not the
intersection of congruences e ~ e a,b. Then the equality on the quotient sI e a,b
is not the intersection of proper congruences, by Proposition 1.2.6, and sI ea b is
'
subdirectly irreducible. The intersection of all ea b is the equality on S; hence
'
the projections s ---t Slea,b separate the elements of s and s is a subdirect
product of the semigroups Sfea b. If S is commutative, then every Sfea b is

' '
commutative. 0
If for instance S is a semilattice, then Birkhoff's Theorem states that S is a
subdirect product of two element semilattices, and can be embedded into the lower
semilattice of all subsets of a set (under intersection). The proof of Proposition
1.3 provides a ready embedding x 1-----1- ('Pa(x))aES of S into ITaES {0,1}.
4. Birkhoff's Theorem provides subdirect decompositions whose components
are as simple as possible. Decompositions with fewer but more general compo-
nents can also be valuable. Thus we call a semigroup S finitely subdirectly
irreducible in case S has at least two elements and, whenever S is (isomorphic
to) a subdirect product of finitely many semi groups Si ( i E I), at least one of
the projections s ---+ si is an isomorphism.
Every subdirectly irreducible semigroup is finitely subdirectly irreducible. The
converse is false; Propositions 1.7 and 1.8 below provides a simple counterexam-
ple, w+.
Finitely subdirectly irreducible semigroups can be characterized by their con-
gruences:
Proposition 1.5. A semigroup S with at least two elements is finitely
subdirectly irreducible if and only if the equality on S is not the intersection of
finitely many proper congruences.
This is proved like Proposition 1.3.
Corollary 1.6. A finite semigroup which is finitely subdirectly irreducible is
in fact subdirectly irreducible.
5. A look at abelian groups and cancellative c.s. will illustrate the various
concepts in this section.
Proposition 1.7. An abelian group is subdirectly irreducible if and only if
it is a nontrivial cyclic or quasicyclic p-group for some prime p. A cancellative
c.s. is subdirectly irreducible if and only if it is a nontrivial cyclic or quasicyclic
p -group for some prime p.
An abelian group is quasicyclic when it is isomorphic to Z(p=) for some
prime p. The second part of the statement is due to Schein [1965].
Proof. If the abelian group G is subdirectly irreducible, then G has a smallest
nontrivial subgroup K, since G has a smallest proper congruence. If c E K,
c i= 1 , then every group homomorphism 'P : G ---+ H such that c ~ Ker 'P is
injective. Thus G is cocyclic; by Theorem 3.1 of Fuchs [1970], G is a nontrivial
1. SUBDIRECT PRODUCTS 99
cyclic or quasicyclic p-group. Conversely, these groups have a smallest nontrivial

subgroup and are subdirectly irreducible.
By the above, a subdirectly irreducible abelian group G is torsion; hence
every nonempty subsemigroup of G is a subgroup of G and is itself a cyclic or
quasicyclic p-group. Now let S be a cancellative c.s. By Birkhoff's Theorem,
G(S) is a subdirect product of subdirectly irreducible abelian groups. Hence S
is a subdirect product of subsemigroups of subdire~tly irreducible abelian groups,
and is a subdirect product of subdirectly irreducible abelian groups. If therefore
S is subdirectly irreducible, then S is a nontrivial cyclic or quasicyclic p-group;
conversely we saw that nontrivial cyclic or quasicyclic p-groups are subdirectly
irreducible. D
Proposition 1.8. A finitely generated abelian group is finitely subdirectly
irreducible if and only if it is either a nontrivial cyclic p-group for some prime
p, or an infinite cyclic group. A finitely generated cancel/alive c.s. is finitely
subdirectly irreducible if and only if it is either a nontrivial cyclic p-group for
some prime p, or isomorphic to Z or to a subsemigroup of N.
Proof. Let G be a finitely generated abelian group which is finitely subdirectly
irreducible. Then G is a direct product of finitely many cyclic groups. Therefore
G is cyclic. If G is finite, then G is subdirectly irreducible by Corollary 1.6, and
is a nontrivial cyclic p-group for some prime p, by Proposition 1.7. Conversely,
nontrivial cyclic p-groups are subdirectly irreducible by Proposition 1. 7, and Z is
finitely subdirectly irreducible by Proposition 1.5, since the intersection of finitely
many nontrivial subgroups of Z is nontrivial.
Now let S be a finitely generated cancellative c.s. which is finitely subdirectly
irreducible. Then G(S) is finitely generated and is a direct product of finitely
many cyclic p-groups (for various primes p) and infinite cyclic groups, and S
is a subdirect product of finitely many cyclic p-groups and subsemigroups of Z.
Therefore S is either a nontrivial cyclic p-group or isomorphic to a subsemigroup
of Z; by Corollary 11.4.2, a subsemigroup of Z is isomorphic to Z or to a
subsemigroup of N.
Conversely we saw that nontrivial cyclic p-groups and infinite cyclic groups
are finitely subdirectly irreducible. Now let S be a subsemigroup of N. If e is
a congruence on S, then
K (e) = {a - b E z I a, b E s and a e b}
is a subgroup of Z. Moreover, K(e) = {0} if and only if e is the equality on
S, and K(niEI ei) = niEI K(ei) if I is finite. If now e is the intersection of
finitely many proper congruences on S, then K (e) is the intersection of finitely
many nonzero subgroups of Z, K(e) I= {0}, and e is not the equality on S;

thus S is finitely subdirectly irreducible. D
By Corollary II. 7.4, a finitely generated c.s. which is cancellative, power
cancellative, and reduced is a subdirect product of finitely many subsemigroups
of N; that is, affine semigroups are subdirect products of finitely many numerical
semigroups or monoids. The proof of Proposition 1.8 also shows that every
finitely generated cancellative c.s. is a subdirect product of finitely many, finitely
subdirectly irreducible semigroups; we shall see in Chapter VI that this property
extends to all finitely generated c.s.
6. The next sections contain well-known results on subdirect decomposi-
tions of separative semigroups, nilsemigroups, and finite c.s.; decompositions of
finitely generated and subcomplete c.s., including Malcev's Theorem [1958] that
finitely generated c.s. which are subdirectly irreducible are finite, will be found
in Chapters VI and VII.
Results on subdirectly irreducible c.s. in general have been published by
Schein [1962], [1966] and McNeil [1971], [1972], but the proofs have some
gaps.
Lesohin [1963A] called a semigroup S approximable by a class e of semi-
groups in case homomorphisms of S into members of e separate the elements
of S; equivalently, S is a subdirect product of subsemigroups of members of
e. Semigroups that are approximable by finite semigroups are finitely approx-
imable; these semigroups are more usually called residually finite (or profinite,
in the case of groups). Lesohin [1963A] also called a semigroup S separable
by a class e of semigroups in case homomorphisms of S into members of e
separate the elements of S from its subsemigroups: if T is a subsemigroup of S
and a E S\T, then there is a homomorphism cp of S into a member of e such
that cp( a) ~ cp(T) . A number of papers by Lesohin and others investigate these
concepts and variations thereof: Lesohin [1963A], [1966A], [1967A], [1968A],
[1969], [1971A], [1972], [1974A], [19741], [1976], [1983], Lesohin & Golubov
[1966], Golubov [1969], [1970A], [1970S], Kublanovsky & Lesohin [1976],
Popyrin [1986]. Residually finite semigroups are studied in Lesohin [1963A],
Lesohin & Golubov [1966], Lesohin [1966A], [1968A], [19741], [1976], and Gol-
ubov [1969], [1970A]. Approximability by various other classes, including char-
acter semigroups, is considered in Lesohin [1963A], [1966A], [1967A], [1969],
[1974A], Lomadze [1992A], [1992B], and lgnat'eva [1996].
2. SEPARATIVE SEMIGROUPS. 101
2. SEPARATIVE SEMIGROUPS.
The main results in this section were discovered independently by Schwarz

[1954A], [1954B] and Hewitt & Zuckerman [1955], [1956].
1. In what follows, C is the multiplicative semigroup of all complex numbers
and § is the subgroup of C of all complex numbers of modulus 1 (the circle
group or !-sphere).
In group theory, a (complex irreducible) character of an abelian group G
is a homomorphism of G into §; these mappings arise as traces of irreducible
representations of abelian groups by complex matrices.
A character of a c.s. S is variously defined as a nonzero homomorphism x
of S into either § U {0} or C. We use the former definition and call a nonzero
homomorphism of S into C a semicharacter of S. If S is a group, then the
ideal x- 1 (0) of S is empty, otherwise x = 0; hence a character of an abelian
group is a character in the group theory sense.
As it turns out, semi groups don't much care whether we use characters or
semicharacters. There is a "radial" homomorphism p : 0 1-----t 0, z 1-----t z/lzl
if z i= 0, of C onto § U {0}. Hence every semi character x yields a character
x o p; conversely, every character is a semi character.
2. Characters are basically a matter of Clifford semigroups. We saw (Corol-
lary IV.2.5) that every c.s. S has a universal semi lattice of groups T, which
comes with a homomorphism T : S ---t T through which factors uniquely every
homomorphism of S into a semilattice of groups. Since § U {0} (and C) are
semilattices of groups, every character (or semicharacter) of S factors uniquely
through T and has the form x o T for some unique character (or semicharacter)
X of T. Thus the characters and semicharacters of S are determined by those
of its universal Clifford semigroup.
Proposition 2.1. The characters of a Clifford semigroup separate its ele-
ments.
Proof. Let G be an abelian group. By Proposition 1.7, the homomorphisms
of G into nontrivial cyclic and quasicyclic p-groups separate the elements of
G. Now § notoriously contains a copy of every Z(p00 ) , which consists of all
complex pk -th roots of unity with k ~ 0, and contains a copy of every cyclic
p-group. Therefore the characters of G separate its elements.
Now let S = UaEY ( G a x {a}) be a semilattice of groups, where Y is a
semi lattice, (G, 1) is an abelian group valued functor on Y, and the multiplication
on S is given, as in Clifford's Theorem 111.2.1, by

(x,a)(y,b) = h:bx ~~by, ab).
For any two elements (t,c) f- (u,d) of S we construct a character x of S such
that x(t,c) f- x(u,d).
If c f- d, then, as in the proof of Proposition 1.2, there is a homomorphism
c.p of Y onto {0, 1} such that c.p(c) f- c.p(d). Then (x,a) f----7 c.p(a) is a homo-
morphism of S into {0, 1} ~ § U {0} and x(t,c) =!= x(u,d).
If c = d, then there is a homomorphism c.p: Gc--+ §such that c.p(t) =/= c.p(u).
Define x: S--+ § U {0} by:
x(x,a) = 0 if a~ c, x(x,a) = c.p('y~(x)) if a~ c.
Then x is a homomorphism: if a,b ~ c, then
x( (x, a)(y, b)) = c.p(T~b (T:b(x )T~b(y)))

= 'P(i~(x)) c.p(T~(y)) = x(x,a) x(y,b);
otherwise x((x,a)(y,b)) = x(x,a) x(y,c) = 0. Also x(t,c) = c.p(t) f- c.p(u) =
x(u,d). D
3. A group with zero is a group with a zero element adjoined. For instance,
C and § U { 0} are groups with zero. A cancellative semigroup with zero is a
cancellative semigroup with a zero element adjoined. The following result is due
to Hewitt & Zuckerman [1956] and Schein [1965]. Its proof catches subdirect
decompositions smuggling structural information into semigroups.
Proposition 2.2. For a c.s. S the following properties are equivalent:
( 1) the characters of S separate the elements of S;
(2) the semicharacters of S separate the elements of S;
(3) S is a subdirect product of groups and/or groups with zero;
(4) S is a subdirect product of cancellative semigroups and/or cancellative
semigroups with zero;
(5) S is separative.
Proof. (1) ===? (2) since every character is a semicharacter.
(2) ===? (3). (2) implies that S is a subdirect product of the images Imx of its
semicharacters. Every Im X ~ C is either cancellative or cancellative with zero;
by Proposition 1. 7 there are homomorphisms of Im x onto groups or onto groups
with zero that separate the elements of Im x. This provides homomorphisms of
S onto groups or groups with zero that separate the elements of S.
2. SEPARATIVE SEMIGROUPS. 103
(3) ===} (4) since groups are cancellative.

(4) ===} (5). When S is a subdirect product of cancellative semigroups Ci
and/or cancellative semigroups with zero Ci U {0}, then the projection of S onto
Ci or Ci U {0} sends an archimedean component A of S into an archimedean
component of ci or ci u {0}' that is, into a cancellative semigroup; therefore
every archimedean component of S is cancellative.
(5) ===} (1). By Proposition IV.2.3, a separative semigroup S can be embed-
ded into its universal semilattice of groups T. If a i- b in S, then by Proposition
2.1 there is a character x ofT which separates a and b (x(a) i- x(b) ); then
the restriction of x to S is a character of S which separates a and b. D
4. Additional properties of semigroup characters and semicharacters can be
found in Clifford & Preston [1961] and in the literature. A bounded semicharacter
satisfies lx(s)l ~ M for all s E S; then lx(s)l ~ 1 for all s E S, since lx(s)l > 1
implies that lx( sn) I = lx( s) In is unbounded. Thus a c.s. S has several character
and semicharacter semigroups, all usually denoted by S* .
Schwarz [1954C] and Iseki [1957] related semiprime ideals of S (such that
an E I implies a E J) to ideals of its character semigroup S*. Comfort [1960]
investigated semicharacters that are isolated in the pointwise convergence topol-
ogy. Hewitt & Zuckerman [1960] and Parizek & Schwarz [1961] determined all
characters of the multiplicative semi group Zn. Petrich [ 1962], [ 1963C] studied
bounded semicharacters, showing in particular that, for bounded semicharacters,
(S x T)* ~ B* x T*. Lesohin [1968A], [1969], [1974S] (and other papers) in-
vestigated approximability and separability by characters and by character semi-
groups. All semigroups S* are semi1attices of groups; the actual groups and
semilattices were described by Warne & Williams [ 1961 ], Fulp [ 1966], McAlis-
ter [1968], and Anderson [1979A]. Lesohin [1970] gave an abstract characteri-
zation of character semigroups. Lesohin [1969], [1971C] also investigated when
two semigroups have isomorphic character semigroups, and when S* is finite or
torsion. Dress & Grabmeier [ 1991] studied sums of characters.
The problem of extending characters or semicharacters of a subsemigroup to
the entire semigroup was studied by Ross [1959], Comfort & Hill [1966], Hill
[ 1966] for bounded semi characters; Kobayashi & Tamura [ 1977] for bounded
semicharacters that preserve suitable preorders; Ross [1961] and Fulp [1967]
for characters; Hill [1967] and McAlister [1968], [1970] for semicharacters in
general.
There is a canonical homomorphism S ----+ S**. Necessary and sufficient
conditions that it be an isomorphism were found by Lesohin [ 1965) for characters,
by Fulp & Hill [ 1965] and Pondelicek [ 1966] for semi characters. More general
forms of duality were investigated by Lesohin [ l966D], [ l968D].
3. NILSEMIGROUPS.
This section contains general properties of commutative nilsemigroups, and a

nice result of Schein [ 1966] which determines when the finite ones are subdirectly
irreducible.
1. Proposition 3.1. In a commutative nilsemigroup, xy -=f. x whenever
x -=f. 0; hence Green 's preorder ;;;:;9-C is a partial order relation, and Green's
relation J{ is the equality.
Proof. If xy = x, then x = xy = xy 2 = · · · = xyn = 0; hence xy -=f. x
whenever x -=f. 0. If now a J{ b and a -=f. b, then a = ub, b = va for some
u, v EN and then a= uva implies a= 0 and b = va = 0 =a, a contradiction.
Hence J{ is the equality on N and ;;;:;9-C is antisymmetric and is a partial order
relation. 0
In a commutative nilsemigroup we denote ~9-C by just ~. By Proposition
3.1, xy < x whenever x,y EN and x -=f. 0.
A congruence e on a nilsemigroup is pure when {0} is a e-class. The next
result goes back to Teissier [1951] and Croisot [1953].
Proposition 3.2. On any commutative nilsemigroup N there is a greatest
pure congruence P, namely
x Py -¢:=::} (Vt E N 1 )(xt = 0 -¢:=::} yt = 0).
Equivalently, x P y -¢:=::} 0 : x = 0 : y, where 0 : x = { t E N 1 I xt = 0}.

Proof. P is an equivalence relation. Since (0: x): z = 0: xz, P is in fact
a congruence: if 0 : x = 0 : y, then 0 : x z = (0 : x) : z = (0 : y) : z = 0 : y z.
Also {0} is a P-class, as x = 0 if and only if 0: x = N 1 .
Conversely let e be a congruence on N such that {0} is a e-class. If X e y,
e
then xt yt and xt = 0 if and only if yt = 0; thus ~ p. 0 e
N = ( a, b I a 3 = a 2 b = a b2 = b3 = 0 ) .
We haveN= {a, b, a 2 , ab, b2 , 0}, and

3. NILSEMIGROUPS. 105
O:a {a 2 , ab, b2 , 0};

O:b {a 2 , ab, b2 , 0};
0 : ab = 0: b2 = N·
'
0:0 Nl.
Therefore the 'Y-classes of N are: {a, b}, { a 2 , ab, b2 } , and { 0} . We see that
N /'Y is cyclic. D
N = (a,bl a3 =a2 b=ab2 =b4 =0).

We have N = {a, b, a 2 , ab, b2 , b3 , 0} and
O:a {a2 , ab, b2 , b3 , 0};
{a 2 , ab, b3 , 0};
{a, a 2 , ab, b2 , b3 , 0};
0 : ab = 0 : b3 = N ·
'
0:0
Therefore the 'Y-classes are: {a}, { b}, { b2 }, { a 2 , ab, b3 } , and { 0}. We see
that N /'Y is the Volkov nilsemigroup Example IV.3 .6. D
2. We now look at some finiteness conditions for nilsemigroups. A semigroup
N with zero is nilpotent if Nk = 0 for some k > 0 (equivalently, if every product
of k or more elements is zero, for some k > 0 ).
An element m of a nilsemigroup N is minimal if m > 0 and there is no
m > x > 0. Say that N has enough minimal elements if every nonzero element
of N lies above some minimal element (if for every a > 0 in N there exists a
minimal element m ;;;;: a). Finite nilsemigroups have enough minimal elements.
Proposition 3.5. Let N be a commutative nilsemigroup.
(1) If N is finitely generated, then N is finite.
(2) If N is finite, then N is nilpotent.
(3) If N is nilpotent, then N has enough minimal elements.
Proof. (1 ). If N is generated by x 1 , x 2 , ... , xn, then every element of N
is a product of positive powers of x 1 , x 2 , ... , xn. Since N is a nilsemigroup,
x 1 , x 2 , ... , xn have only finitely many distinct positive powers; therefore N is
finite.
(2). Let N = { x 1 , x 2 , ... , xn}. Let k = k 1 + k 2 + ··· + kn, where .

k . . .
xi i = 0. In any product x{1 x~2 ••• x~n of k or more elements of N we have
.Ei ji ~ k, ji ~ ki for some i, and x{1 ~2 •· · x~n = 0. Thus Nk = 0.
(3). Let N be nilpotent with Nk = 0. In N every strictly descending chain
x1 > x 2 > ··· > xn has length at most k: indeed xi+l = xiui+l for some
ui+l E N, so that xk = x 1 u 2 · · · uk = 0 and xk > xk+l is not possible. In
particular the descending chain condition holds in N and N has enough minimal
elements. 0
3. Schein's result applies to nilsemigroups with enough minimal elements.
Proposition 3.6 (Schein). For a commutative nilsemigroup N # 0 with
enough minimal elements the following conditions are equivalent:
(1) N is subdirectly irreducible;
(2) N is finitely subdirectly irreducible;
(3) the greatest pure congruence P on N is the equality.
Proof. Already (1) ===> (2).
(2) ===> (3). Since N # 0, the set M of all minimal elements of N is not
empty. Assume that N is finitely subdirectly irreducible. By Proposition 1.5, the
equality on S is not the intersection of finitely many proper congruences. For
every m E M let !Rm be the Rees congruence of the ideal {0, m} of N. If M
had at least two elements m and n, then !Rm n !Rn would be the equality on N;
therefore M has just one element m. Next, 0 and m are not equivalent modulo
P; hence P n !Rm is the equality on N and P is the equality on N.
(3) ===> (1). Assume that P is the equality on N. Since M is a :P-class we
again have M = {m} . Let e
be a proper congruence on N, so that a b for e
some a, b E N, a # b. Then a P b does not hold and there exists u E N 1 such
that, say, ua # 0 and ub = 0. Since N has enough minimal elements we have
ua ~ m. Then vua = m for some v E N 1 and vub = 0. Therefore me 0.
Thus every proper congruence on N contains the Rees congruence !Rm and N
is subdirectly irreducible by Proposition 1.3. 0
When N # 0 is a commutative nilsemigroup with enough
Corollary 3. 7.
minimal elements, then N /P is subdirectly irreducible with enough minimal
elements.
Proof. N j:P is a commutative nilsemigroup with one minimal element, the
:P-class M of all minimal elements of N; since N has enough minimal elements,
4. PONIZOVSKY DECOMPOSITIONS 107
P ~ M for every nonzero P-class P, and N /P has enough minimal elements.

If Pa, Pb are nonzero P-classes, then Pa Px = 0 if and only if ax = 0, and
Pb Px = 0 if and only if bx = 0; hence Pa P Pb in N /P implies a P b in N
and Pa = Pb. Thus P is the equality on N /P. D
Example 3.4 shows that the Volkov nilsemigroup Example IV.3.6 is subdirectly
irreducible.
4. PONIZOVSKY DECOMPOSITIONS.
The Ponizovsky decomposition of a finite commutative semigroup S is an

explicit subdirect decomposition of S with remarkable properties, due to Poni-
zovsky [ 1962].
1. The Ponizovsky decomposition results from two properties of finite c.s.
First, every archimedean component contains an idempotent. Second:
Lemma 4.1. Let S be a finite c.s. and x E S. When e E E(S 1 ) is

idempotent, then x ~9-C e in S 1 if and only if ex = x, and there exists a least
such idempotent E( x).
Proof. ex =x implies x ~9-C e; conversely, x ~9-C e implies x = et for
some t E 5 1 and ex = eet = et = x.
Let el ' e2 '0 0 0en be the idempotents ei of 5 1 such that eix =
' X 0 The
product e 1 e 2 · · · en is the least such idempotent. 0
The Ponizovsky decomposition extends readily to any semigroup with these
two properties. Formally we call a c.s. S complete when
(1) every archimedean component of S contains an idempotent; equivalently,
by Corollary IV.3 .2, every element of S has a power in a subgroup of S;
(2) for every x E S there exists a least idempotent e of 5 1 such that x ~9-C e;
equivalently, a least idempotent e of 5 1 such that ex = x; we denote it by E( x).
Condition (2) is trivial when S has only finitely many idempotents: then, as in
the proof of Lemma 4.1, 5 1 has only finitely many idempotents e 1 , e 2 , ... , en
such that eix = x, and then the product e = e 1 e 2 · · · en is the least such
idempotent.
Finite c.s. are complete, by Lemma 4.1.
108 IV. SuBDIRECT DECOMPOSITIONS.
2. Complete semigroups are incorrectly defined in Grillet [1995], where

condition (2) above was replaced by
(3) the semilattice E(S) is complete (every nonempty subset A of E(S) has
an infimum /\eEA e)
in the mistaken belief that (1) and (3) imply (2). A counterexample is provided
by Example 4.2 below. The author is highly embarrassed by this error and hopes
that no such oversight has found unwanted refuge in the present text. The correct
definition and counterexample are from Grillet [2001S].
Example 4.2. Let
S = { e 1, e 2, ... , en, ... , a, e}
with commutative multiplication
e. a e
J
ei eiVj a e
a a e e
e e e e
where i V j = max (i,j). Associativity is straightforward: (xy) z = x (yz)
holds in S whenever x, y, or z equals e, and when x,y,z = ei, ej, ek; also
ei (eja) =a= (eiej)a, ei (aej) =a= (eia)ej, ei (aa) = e = (eia)a,
a (ej a) = e = (a ej) a, and commutativity yields the remaining cases.
We see that S = S 1 and that E(S) is the chain e 1 ~ e 2 ~ ... ~en ~ ... e;
it is a complete semilattice in which e = 1\n>O en. The archimedean components
of S are {a,e} and all {en}. Also en a= a for all n, but ea-=/= a; thus (2)
does not hold for a. D
3. When S is a complete semigroup, the Ponizovsky factors of S are the
semigroups
~ = Sef(UJEE(S),f<e Sf),
one for every idempotent e E E(S 1 ). In particular, if 1 tf. S (if S is not a
monoid), then .fl = Sf S E(S); if there is a least idempotent e0 (if for instance
S is finite), then ~o = Se0 .
The reader will observe that the kernel He of the archimedean component of
e is the group of units of its Ponizovsky factor.
Composing multiplication by e, which is a homomorphism S ---+ Se, with
the projection Se ---+ ~ to the Rees quotient yields a canonical surjective ho-
momorphism, the projection 1re : S -----+ ~. By definition,

e (x) = ex if E (ex) = e,
1r 1r e (x) = 0 otherwise,
since x E Se\(UjEE(S), f<e Sf) if and only if e = E(x) is the least idempotent
e ~9-C x.
Theorem 4.3. Every complete (for instance, finite) commutative semigroup
is a subdirect product of its Ponizovsky factors.
Proof. We show that the canonical projections 1re (e
E(S 1)) separate
E
the elements of S. Assume that 1re(x) = 1re(y) for every e E E(S 1 ). Let
E = E(S). If x ¢. SE, then 1 ¢. S, 1r1 (x) =/= 0, 1r1 (y) =/= 0, y ¢. SE, and
1r1 (x) = 1r1 (y) yields x = y. Now let x E SE. As above, y E SE. Let
e = E(x) and f = E(y). If ef < e, then 1re(y) = 0 in~, since ey ~ ef < e,

whereas 1re(x) = x, since e = E(x). Therefore ef = e. Similarly ef = f.
Hence e = f, and x = 1re ( x) = 1rf (y) = y. D
The Ponizovsky decomposition is unique among subdirect decompositions in
being also a partition. The partial Ponizovsky factors of a complete semigroup
S are the partial semigroups
~* = Se\(UfEE(S), f<e Sf),
one for every idempotent e E E(S 1 ). In particular, if 1 ¢. S, then 11* =
S\ S E (S) may be empty; if there is a least idempotent e0 , then ~: = S e0 .
In general, the ideals Se and u!EE(S), f<e Sf are unions of JC-classes; hence
every partial Ponizovsky factor is a union of JC-classes.
Proposition 4.4. Let S be a complete (for instance, finite) c.s. and e E
E(S 1 ). Then x E ~* if and only if e = E(x). Hence S is the union of its
partial Ponizovsky factors, which are pairwise disjoint. If S is a monoid, then
(~*)eEE(S) is a partition of S.
4. A c.s. S is elementary in case it is the disjoint union S = G U N of

a group G and a nilsemigroup N, in which N is an ideal of S, the identity
element of G is the identity element of S, and the zero element of N is the
zero element of S. (This last condition may be omitted.) In particular S has an
identity element, a zero element, and no other idempotent.
Proposition 4.5. In a complete (for instance, finite) c.s. S: if there is a
least idempotent e0 , then ~o = Se 0 is a group and the kernel of S; if 1 ¢. S,
then 11 is a nilsemigroup; all other Ponizovsky factors are elementary.
Proof. If there is a least idempotent e0 , then every nonempty ideal of S

contains some x E S, contains some xn in a subgroup H f of S, contains H f
and the idempotent f, contains e0 ~ f, and contains Se 0 ; thus ~o = Se 0 is
the kernel of S.
Let xES and e = E(x) E E(S 1), so that x E ~*.Since Sis complete, x
has an power xn in a subgroup H f of S. If x rt- He (for instance, if e = 1 rt- S ),
then f < e, since f ~J-C x <J-C e, so that xn E UtEE(S), f<e Sf and xn = 0
in ~. Therefore:
if e = 1 rt- S, then ~ is a nilsemigroup;
if e is the least idempotent of S (if such exists), then f <e is not possible,
x E He, and ~ = ~* = He;
otherwise e E S, UtEE(S), f<e Sf #- 0, ~ has a zero element, and an
element of ~ is either in He or nilpotent. Then ~ is the disjoint union ~ =
G U N of a group G = He and a nilsemigroup N which consists of all nilpotent
elements of ~ ; N is an ideal of ~, since y ~J-C x and xn = 0 implies yn = 0;
the identity element e of G = He is the identity element of ~; and the zero
element of N is the zero element of ~ . Thus ~ is elementary. 0
Corollary 4.6. A complete (for instance, finite) subdirectly irreducible c.s.
is either a group or a nilsemigroup or elementary.
Elementary semigroups are studied in more detail in the next section. The
subdirectly irreducible semigroups in Corollary 4.6 will be determined in the next
chapter.
5. The Ponizovsky factors of a complete c.s. are easier to reassemble than
its archimedean components; this is another remarkable feature of the Ponizovsky
decomposition. The author [ 1972] showed that a finite c.s. can be reconstructed
from its Ponizovsky factors by means of partial homomorphisms. This extends
immediately to complete semigroups.
For all e,f E E(S 1) let pj : ~* ----+ If be the restriction of 1r f to ~*; pj

is a partial homomorphism ( pj (xy) = pj (x) pj (y) whenever xy #- 0 in ~ ).
Lemma 4.7. The following holds:

(1) pj(1re(x)) = 1r1 (x) whenever e ~ f and 1re(x) #- 0;
(2) for every xES there exists a greatest e E E(S 1 ) such that 1re(x) #- 0
in~·
Proof. Let e ~ f and 1re(x)# 0, so that 1re(x) =ex and t:(ex) =e. Then
pj(1re(x)) = 1r1 (ex) = 1r1 (x), since fex = fx. This proves (1).
Let e = t:(x). Then 1re(x) = x # 0 in
If conversely 1r1 (x) # 0,
~·
then t:(fx) = f and f ~ e, since fx ~:J-C fe ~:J-C e. Thus (2) holds (with
e = t:(x)). 0
Proposition 4.8. Every complete (for instance, finite) c.s. S is determined
up to isomorphism by its Ponizovsky factors and the partial homomorphisms
pj : ~* -----+ If with e,J E E(S 1) and e ~ f. Namely, S is isomorphic to the
semigroup P of all (xe)eEE(Sl) E TieEE(Sl) ~ such that
(A) pj(xe) = x f whenever e ~ f and xe # 0;

(B) there exists a greatest e E E(S 1 ) such that xe # 0.
Proof. When x E S let B(x) = (1re(x) )eEE(Sl). Then B(x) E P by Lemma
4.7, and e : S -----+ P is an injective homomorphism by Theorem 4.3.
Conversely let y = (yf)fEE(Sl) E P; let e E E(S 1) be greatest such

that Ye # 0. Let x = Ye E ~*. Then 1re(x) = x = Ye· Iff ~ e, then
1rf(x) = pj(1re(x)) = pj(Ye) = Yj by Lemma 4.7. Iff~ e, then Yj = 0,
fx ~ fe, t:(fx) ~ fe < f, and 1r1 (x) = 0 = YJ. Thus y = B(x), which shows
that e is surjective. 0
The families (pj )e,JEE(Sl ), e?;J of partial homomorphisms that arise from
complete semigroups can be characterized by the following properties (Grillet
[1972]):
(i) iff~ e E S, then pj(e) = f;
(ii) p~(x) = x whenever x E ~*,and pt(pj(x)) = p~(x) whenever e ~ f ~ g

and pj(x) # 0;
(iii) for every x E ~* and y E If* there exists a greatest g ~ e, f such that
p~ ( x) p~ (y) # 0 in ~ .
This does not quite solve the subdirect product retrieval problem, since, even
with good constructions of the semigroups ~, it is not immediately clear how
to construct all families of partial homomorphisms with properties (i), (ii), and
(iii). Much less, however, can be said about the reassembly of archimedean
components.
5. ELEMENTARY SEMIGROUPS.
In this section we construct elementary semigroups from groups and nilsemi-

groups. Another construction is given in the next chapter.
1. An elementary semi group S = G U N is an ideal extension of the nilsemi-
group N by a group with zero G U { 0} . In this case the ideal extension problem
has a reasonable solution.
When S = G U N is elementary, the group G acts on the set S by left
multiplication (g. x = gx ).
Proposition 5.1. In an elementary semigroup S = G U N, the orbits under
the action of G coincide with the '}{-classes of S.
Proof. If a'){ b holds in T, then either a= b = 0, orb= ta -1 0, a= bu -10
for some t, u E 8 1 = S. If t E N or u E N, then a '){ b in N and a = b by
Proposition 3 .1. Otherwise b = ga for some g E G; this also holds if a = b. If
conversely b = ga, then a = g- 1 b and a'){ b. D
The semigroup S j'J{ is the monoid of orbits of S; it is a commutative
nilmonoid, that is, a commutative nilsemigroup with an identity element adjoined.
2. When S = G U N is elementary, the multiplication on S = G U N is
completely determined by the multiplications on G and N and by the action of
G on N. The latter satisfies g(xy) = (gx)y for all g E G and x,y EN.
Conversely, any such group action extends to an elementary semigroup multipli-
cation on GUN: indeed gO = g (00) = (gO) 0 = 0 and all other conditions are
satisfied.
Lemma 5.2. In an elementary semigroup S = GUN, every g E G permutes
every 'Y-class of N.
Proof. 'Y is the largest pure congruence on N (Proposition 3.2). When
x EN, then xt = 0 implies (gx) t = 0, and (gx) t = 0 implies xt = g- 1 gxt = 0;
thus gx 'Y x, and multiplication by g permutes the 'Y-class of x. D
Proposition 5.3. Let N be a commutative nilsemigroup. Let U (N) be the
group, under pointwise multiplication, of all families s = (s p) PEN/'Y such that
each s p is a permutation of the 'Y-class P and
(spx)y = sQ(xy) whenever P,Q E N/'Y, x E P, xy E Q. (1)
For every abelian group G disjoint from N, there are one-to-one correspon-
dences between: (i) elementary semigroups G U N; (ii) actions of G of N such
5. ELEMENTARY 8EMIGROUPS 113
that g (xy) = (gx) y for all g E G and x, y E N; (iii) homomorphisms of G

into U(N).
U(N) is the group of units of the semigroup of left translations of N as
defined in, say, Clifford and Preston [1961].
Proof. The one-to-one correspondence between (i) and (ii) was noted above.
We construct the one-to-one correspondence between (i) and (iii). Let U(N) be
defined as in the statement, with the pointwise operation ( st) pX = s p (t pX)
for all x E P, P E N /P. Then U(N) is a monoid, with identity element
(1 p) PEN/P. Moreover, U(N) is closed under pointwise inverses. Indeed let
s = (sp)PEN/P E U(N) and t = (s[})PENjP· Let x E P, xy E Q and
z = sj} x. Then z E P, zy E Q, xy = (spz)y = sQ(zy) by (1), and
(sj}x)y = zy = s(/(xy). Thus t E U(N). Therefore U(N) is a group, in
which t is the inverse of s.
When S = G U N is elementary, left multiplication by g E G induces a
permutation g p of every P-class P, namely g pX = gx for all x E P (Lemma
5.2). Moreover (gpx)y = gxy = gQ(xy) whenever x,y E N and x E P,
xy E Q, and (gh)p = gp hp (that is, (gh)px = gp(hpx) for all x E P) for all
g, h E G. To an elementary semigroup GUN thus corresponds a homomorphism
g 1---t (gp)PEN/P of G into U(N).
If conversely <p : G ---+ U (N) is a homomorphism, then for every g E G
and every P-class P of N we have a permutation cp(g) p of P, and an action
of g E G on x E N is defined by: gx = cp(g) p x, whenever x E P. This is
a group action, since <p is a homomorphism, and (gx) y = g ( xy) holds for all
x, y E N by definition of U(N). With this action we now have an elementary
semigroup multiplication on GUN. It is immediate that g 1---t (gp)PEN/P is
the given homomorphism <p. D
3. We conclude this section with two examples.
Example 5.4. Let N be the commutative nilsemigroup (Example 3.3)
N = ( a, b I a 3 = a 2 b = a b2 = b3 = 0 ) ,
N = {a, b, a 2 , ab, b2 , 0}. We saw that the P-classes are A = {a, b},
B = { a 2 , ab, b2 }, and { 0}. If sA is the identity on A, then the condition
(spx)y = sQ(xy), whenever x E P, xy E Q (1)
implies that s B is the identity on B. If sA is the transposition (a b), then
114 IV. SuBDIRECT DECOMPOSITIONS.
condition (1) yields s 3 (ab) = (sA a) b = b2 and s 3 (ab) = (sA b) a = a 2 ; this

contradiction shows that sA is the identity on A. Therefore U ( N) is trivial. In
any elementary semigroup G U N, gx = x for all g E G and x E N. 0
N = (a,bj a3 =a 2 b=ab2 =b3 =0,a 2 =b2 ),
N = {a, b, a 2 , ab, 0}. The P-classes are A = {a, b}, B = { a 2 , ab}, and
{0} . Hence U (N) has two elements t, u: u A, u B are identity permutations,
and t A, t B are transpositions. If for instance G = { 1, g} is cyclic of order 2,
then there are two elementary semigroups G U N: one in which gx = x for all
x E N, and one in which ga = b, gb = a, ga 2 = ab, and gab = a 2 = b2 . 0
Chapter V.
GROUP COEXTENSIONS.
Group coextensions were developed independently by Grillet [1974] and

Leech [ 1975] for semigroups in general. They yield precise constructions of
complete commutative semigroups in terms of abelian groups and group-free
semigroups. This leads to the semigroup cohomology which is studied in the
last chapters of this book.
This chapter contains general properties and, as an application, constructs all
finite commutative semigroups that are subdirectly irreducible.
1. DIVIDING BY JC.
This section reveals an intimate relationship, first explored by Kolibiarova

[ 1958] and others, between a c.s. S and the quotient semigroup S j'){.
1. Proposition 1.1. In a commutative semigroup S, a ~:K b holds in S if
and only if Ha ~:H Hb holds in Sj'J{. Hence ']{ is the equality on Sj'J{.
Proof. Let H 1 = 1 E (S/'){) 1 in case S =I S 1 does not have an identity

element. If a= bt for some t E S 1 , then Ha = HbHt in (S/'){) 1 and Ha ~:K Hb
in Sj'){. If conversely Ha ~:H Hb holds in Sj'){, then Ha = HbHt = Hbt for
some t E S 1 , a '){ bt ~:K b holds in S, and a ~:K b.
Hence Ha '){ Hb in S j'){ implies a '){ b in S and Ha = Hb. 0
Proposition 1.2. Let S be a commutative semigroup. The projection S ---+
Sj'){ induces an isomorphism of E(S) onto E(Sj'){).
Proof. If e is idempotent in S, then He is idempotent in S j'){, since '){
is a congruence; hence the projection S ---+ Sj'){ induces a semigroup homo-
morphism (} : E(S) ---+ E(Sj'){). If H is idempotent in Sj'){, then H is a
subsemigroup of S, is a subgroup of S by Proposition 1.4.3, and contains exactly
one idempotent of S; therefore (} is bijective. 0
115
116 V. GROUP COEXTENSIONS.
Corollary 1.3. A commutative semigroup S is a monoid if and only if S j'J{

is a monoid; hence S 1 j'J{ = (S/']{) 1 for every commutative semigroup S.
Proof. If S has an identity element 1, then S j'J{ has an identity element
H 1 . If conversely Sj'J{ has an identity element He, then, for every x E S,
Hx ~:H He, x ~:H e by Proposition 1.2, x = et for some t E S 1 , and
ex = eet = et = x; thus e is an identity element of S.
If S is a monoid, then S1 j'J{ = S j'J{ = (S j'J{) 1 . If S is not a monoid, then

{1} is an '}{-class of S1 and S1 j'J{ = (Sj'J{) u {1} = (S/']{) 1 , since Sj'J{ is
not a monoid. D
2. The universal semilattice and archimedean components of S can be lifted
from Sj'J{:
Proposition 1.4. When S is a commutative semigroup: a ~:N b in S if and
only if Ha ~:N Hb in Sj'J{; aN b in S if and only if HaN Hb in Sj'J{; and
the projection S ---+ Sj'J{ induces an isomorphism of Y(S) onto Y(Sj'J{).
Proof. If Ha ~:N Hb holds in Sj'J{, then Han ~:H Hb for some n > 0,
an ~:H b for some n > 0 by Proposition 1.1, and a ~:N b holds in S. The
converse is clear. Hence a N b holds in S if and only if Ha N Hb holds in S j'J{.
Thus the surjective homomorphisms S---+ Y(S) and S---+ Sj'J{---+ Y(Sj'J{)
induce the same congruence on S, and Proposition 1.2.4 yields an isomorphism
Y(S) ~ Y(Sj'J{) which makes the following square commute:
s ----t sI']{
1
Y(S) ~
1
Y(Sj'J{) D
Corollary 1.5. A commutative semigroup S is a Clifford semigroup if and

only if S j'J{ is a semilattice.
Corollary 1.6. A commutative semigroup S is a complete archimedean
semigroup if and only if S j'J{ is a nilsemigroup.
Proof. By Propositions 1.4, 1.2, S is archimedean if and only if S j'J{ is
archimedean, and S contains an idempotent if and only if S j'J{ contains an
idempotent. Since ']{ is the equality on S j'J{, S j'){ is complete and archimedean
if and only if it is a nilsemigroup, by Proposition 111.3.1. D
3. The Ponizovsky decomposition of S can also be lifted from S j'){:
Proposition 1.7. A commutative semigroup S is complete if and only if
2. SCHUTZENBERGER FUNCTORS. 117
Sj'){ is complete; and then, for all xES and e E E(S 1 ), E(Hx) = HE(x)' so
that x E ~* in S if and only if Hx E P_He in Sj'){.
In this statement we regard H 1 as 1 E 5 1 j'){ in case 1 ¢. S.

Proof. By Corollary 1.3, (S/'){) 1 = S 1 j'){. The projectionS----* Sj'){
induces an isomorphism E(S 1 ) 9:! E((S/'){) 1 ),
by Proposition 1.2, and sends
the archimedean components of S onto the archimedean components of S j'){,
by Proposition 1.4. Hence every archimedean component of Sj'){ contains an
idempotent if and only if every archimedean component of S contains an idem-
potent. By Propositions 1.1, 1.2, 8 1 has a least idempotent e ~3-C x if and only
if (S/'){) 1 has a least idempotent He ~3-C Hx; and then E(Hx) = HE(x). Thus
S is complete if and only if S j'){ is complete. The last part of the statement
now holds since x E ~* if and only if e = E(x) (Proposition IV.4.4). D
Corollary 1.8. A commutative semigroup S is elementary if and only if S
has a zero element and S j'){ is a nilmonoid
Proof. If S is elementary, then Sj'){ is a nilmonoid by Proposition IV.5.1.
Conversely assume that S has a zero element and that S j'){ is a nilmonoid.
Then S has two idempotents e > f, by Proposition 1.2; e is an identity element
of S by Corollary 1.3, so f is the zero element. Now G = He is a subgroup
of S, the identity element e of G is the identity element of S, and N = S\ G
is an ideal of S, since its projection (Sj'){)\1 in Sj'){ is an ideal of SjSH.
Every x E N is nilpotent: since Hx is nilpotent in Sj'){, we have xn '){ 0 for
some n > 0 and xn = 0. Thus S is elementary. D
2. SCHfrTZENBERGER FUNCTORS.
Continuing the previous section we show that the Schiitzenberger groups of a

commutative semigroup can be arranged into a functor. The results in this section
are from Grillet [1974] (see also Grillet [1995]).
I. Recall that the SchOtzenberger group of an '}{-class H is r(H) = { gfll t E
St (H)}, where
st (H) = { t E S 1 1 tH ~ H}
and gfl a = ta for all t E St (H) and a E H; it is a simply transitive group of
permutations of H.
Lemma 2.1. Let A and B be '}{-classes of a c.s. S. If A ~9-C B in Sj'J{,

then St (A) ~ St (B) and there is a unique homomorphism f~ : r(A) ---+ r(B)
such that the square
St(A) ~ St{B)
gAl lgB
r(A) 7a r(B)
commutes; then
(ga) s = (r~g)(as)
whenever g E r(A), a E A, s E T 1 , and as E B. Moreover fi is the identity

on r(A), and ffJ of~= f~ whenever A ~9-C B ~9-C C in Sj'J{.
Proof. Assume A ~9-C B in Sj'){ and let a E A, bE B. Then a ~9-C b in S
and b =as for some s E S 1 . Since '){ is a congruence, As~ B. If tESt (A),
then ta E A, tb = tas E B, and t E St (B) since '){ is a congruence. Thus
St(A) ~ St(B).
If t, u E St (A) and gf = g~, then ta = ua, tas = uas, and gf = g~
(since r(B) is simply transitive). Therefore there is a unique homomorphism
fjj : r(A) ---+ r(B) such that the square in the statement commutes. The
uniqueness of f~ with this property yields the last part of the statement. If
finally g E r(A), a E A, s E S 1 , and as E B, then g = gf for some
t E St(A) ~ St(B), f~g = gf, and (ga)s = tas = (r~g)(as). 0
Let H(S) be the category whose objects are the elements of S and where
a morphism from a to b is an ordered pair (a, t) with t E S 1 and at = b
(necessarily b ~9-C a). With morphisms "written on the left", composition is
given by (at,u) o (a,t) =(a, tu). The identity morphism on a is (a, 1). H(S)
is the commutative version of the categories L(S) and R(S) in Leech [1975].
An abelian group valued functor on a c.s. S is a functor G = (G,f) from
H(S) to the category of abelian groups. Thus G assigns an abelian group Ga
to every element a of S and a homomorphism fat : Ga ---+ Gat to each pair
'
s
(a, t) E X S 1 ' such that fa 1 is the identity on G a and fat u 0 fat = fa tu.
' ' ' '
An abelian group valued functor G = (G, f) on S is thin in case fa t = fa u
' '
whenever at= au; then G provides only one homomorphism Ga---+ Gb when
2. SCHUTZENBERG ER FUNCTORS. 119
a ~J-C b, which we will generally denote by l'b. A thin functor G can be

viewed as a functor from the simpler category arising from the preordered set S,
preordered by ~J-C. Thin functors suffice in this chapter, but the more general
functors defined above are needed later.
The SchOtzenberger functor of a c.s. S is the abelian group valued functor
Ir = (r,/') on Sj'){ in which r(A) is the Schiitzenberger group of A and
I'A T = ·-r:tris provided by Lemma 2.1.
'
2. The Schiitzenberger functor of a c.s. S has additional properties when S
is finite (Grillet [1974]) or, more generally, complete.
Lemma 2.2. Let S be a complete c.s., T = Sj'J{, and A E T be an
'}{-class of S.
(1) If A tJ_ T E(T), then r(A) is trivial.
(2) If A E T E(T) and E = E(A) holds in T, then /'~ is surjective.
If S is finite, then r(A) is finite.
Proof. T is complete by Proposition 1.7. Let a E A and t E St (A),
t E S. Since S is complete, some power tn of t belongs to a subgroup H f
of S. Then tn A ~ A and fA ~ A (since f '){ tn ), and FA = A in T,
where F = H f E E(T). If A tJ_ T E(T), then this is impossible; therefore
St(A) = {1} and r(A) is trivial, which proves (1).
Now assume A E T E(T). Then a E S E(S), by Propositions 1.1 and
1.2. Let e = E(a), so that E = He = E(A) (Proposition 1.7). In the above,
fA ~ A implies a ~J-C f and e ~J-C f. Now e E St (A), since ea = a, and
et E St (A); moreover g;d = gf and et E He, since e = ef ~J-C et ~J-C e.
Hence gf = g;d = /'~ g~; this proves (2). 0
Let G = ( G, /') be an abelian group valued functor on a c.s. S. We call G
finite when every G a is finite. When S is complete, we call G surjecting if it
has properties (1) and (2) in Lemma 2.2: that is, Ga is trivial if a tJ_ S E(S)
and l'e a is surjective whenever e = E(a) E S. (Surjecting functors are called
'
surjective in the author's papers, which is not very felicitous.) Thus the Schiitzen-
berger functor of a complete semigroup is thin and surjecting; the Schiitzenberger
functor of a finite semigroup is thin, surjecting, and finite.
We note some other consequences of Lemma 2.2.
Proposition 2.3. For a complete (for instance, finite) c.s. S the following
(1) all subgroups of S are trivial;

(2) JC is the equality on S;
(3) ~:J-C is an order relation on S.
This follows from Lemma 2.2 and previous results. A c.s. is group-free when
all its subgroups are trivial; these semigroups have also been called (in the finite
case) combinatorial and aperiodic. Commutative semigroups on which J( is
the equality, so that ~:J-C is an order relation, are also called naturally partially
ordered and holoids. Lemma 2.2 and Proposition 1.1 yield:
Proposition 2.4. When S is a complete (for instance, finite) c.s., then SjJC
is group-free, and the Schiitzenberger functor of S is thin and surjecting.
3. COEXTENSIONS.
Continuing the previous sections we now define JC-coextensions and group

coextensions and study the relationships between these two concepts. Group
coextensions will be constructed in the next section. The results in this section
are from Grillet [1974] (see also Grillet [1995]).
1. Generally, an extension of a semigroup S is a semigroup E of which S
is a subsemigroup; or, more generally, an ordered pair (E, 1-l) of a semigroup E
and an injective homomorphism 1-l : S ---+ E.
A (commutative) coextension of a (commutative) semi group S is an ordered
pair (E, 1r) of a (commutative) semigroup E and a surjective homomorphism
1r : E ---+ S; or just the semigroup E, if it is clear what 1r is. Then the sets
Ea = 1r- 1 a = {x E E I 7rX = a}
with a E S, which are the equivalence classes of the congruence ker 1r induced
by 1r, constitute a partition of E; moreover, E a Eb ~ E ab for all a, b E S. For
example, every c.s. T is a commutative coextension of its universal semilattice
Y (T); then the sets Ta are the archimedean components of T.
Commutative coextensions in which 1r : E ---+ S is a retraction were consid-
ered by Schmidt [1975], [1977]. Nguyen [1981] studied monoid coextensions of
S by B = 7r- 1 (1), including Baer sums and a commutative monoid Ext(S,B)
when B is central. This chapter deals with two related types of coextensions,
defined as follows.
A commutative JC-coextension of a (necessarily commutative) semigroup S
3. COEXTENSIONS 121
is a commutative coextension ( E, 1r) of S such that ker 1r ~ JC, and is exact if

ker 1r = JC, in which case S ~ E jJC. Thus every c.s. T is an exact JC-coexten-
sion of T jJC; then the sets Ta are the JC-classes of T. Clifford semigroups are
exact JC-coextensions of semilattices (Corollary 1.5). Archimedean semigroups
with an idempotent are exact JC-coextensions of nilsemigroups (Corollary 1.6).
Elementary semigroups are exact JC-coextensions of nilmonoids (Corollary 1.8).
When £ = (E, 1r) is an exact commutative JC-coextension of a c.s. S, then
S ~ EjJC and the categories H(S), H(EjJC) are isomorphic; composing the
Schiitzenberger functor of E with the isomorphism H(S) ---+ H(E jJC) yields
an abelian group valued functor on S, the SchUtzenberger functor Ir = (r, 1)
of£. Thus r(a) = r(A) and 'Ya,u = 'YA,U for every a E S and u E S 1 , where
A= Ea = and U = Eu if u E S, U = 1 E (E/Jt) 1 if u = 1 E S 1 .
1r- 1a
Then r(a) acts simply and transitively on Ea, and
(g • X) Y = (!a b9) • xy
'
whenever x,y E E, 1rx = a, 1ry = b, and g E Ga, by Lemma 2.1. These
properties suggest the next definition.
2. A commutative group coextension of a c.s. S by an abelian group valued
functor G = ( G, 1) on S is an ordered triple £ = ( E, 1r,.) of a c.s. E, a surjective
homomorphism 1r : E ---+ S and, for every a E S, a simply transitive group
I
action • of Ga on the set E a = { x E E 1rx = a}, such that
(g • X) Y = (!a b9) • xy
'
whenever x, y E E, 1rx = a, 1ry = b, and g E G a . We also call the semigroup
E a commutative group coextension of S by G when 1r and . are clear.
By Lemma 2.1, every c.s. T is a commutative group coextension ofT jJC by
its Schutzenberger functor; more generally:
Proposition 3.1. An exact commutative JC-coextension is a commutative
group coextension by the Schatzenberger functor of that coextension.
For another example, let B be an abelian group and A be a subgroup of B,
both written multiplicatively. Then the projection B ---+ B /A is a surjective
homomorphism, A acts by left multiplication (a • x = ax) on every one of its
cosets, this action is simply transitive, and (a • x) y = a. xy holds for all a E A
and x,y E B. Hence
Proposition 3.2. Every abelian group extension of an abelian group A by
an abelian group C is a commutative group coextension of C by the constant
functor A on C.
3. The converse of Proposition 3.1 does not hold in general: not every
commutative group coextension (E, 1r,.) of S by G is an exact J{-coextension,
even when G is the Schotzenberger functor of some c.s.; that is, the congruence
induced by 1r : E ----t S on E does not necessarily coincide with J{.
Example 3.3. Let T be the semigroup
T = {ai,bi,OiiEZ}
in which
aiaj = ai+j, aibj = bjai = bi+j, bibj = 0,
and 0 is a zero element. T is a c.s.: indeed ai (ajak) = ai+j+k = (aiaj)ak,
ai(ajbk) = bi+j+k = (aiaj)bk, and all other products ofthree elements are
0. The J{-classes ofT are all {ad, B = {bi I i E Z}, and {0}; r(B) 9:! Z
and the other SchOtzenberger groups are trivial. This specifies the Schiltzenberger
functor H.'.
Now let E be the semigroup
E = {ai,bi,OiiEZ}
in which
aiaj = ai+j, aibj = bjai = bj, bibj = 0,
and 0 is a zero element. We see that E is a group coextension of S = T j']{ by H.' :
r( B) 9:! Z acts simply and transitively on B as before, and (g. x) y = (1a bg) • xy
'
holds in E since either g = 1, or xy = 0 in E, or x,xy E B and Ia bg =g.
'
But ']{ is the equality on E, so E is not an J{-coextension of T jJ{ by H.' , let
alone an exact 'J{-coextension. 0
4. However, the SchOtzenberger functors of complete semigroups have addi-
tional properties which force group coextensions to be exact :K-coextensions.
Theorem 3.4. If T is a complete (for instance, finite) commutative semi-
group, then T j:K is complete and group-free and the Schatzenberger functor of
T is thin and surjecting.
Conversely let S be a complete group-free commutative semigroup, G be a
thin and surjecting abelian group valued functor on S, and £ = ( E, 1r, • ) be
a group coextension of S by G. Then E is complete, ker 1r = :K, and the
SchUtzenberger functor of £ is naturally isomorphic to G.
Proof. The direct part follows from Propositions 1. 7 and 2.3 and Lemma 2.2.
Conversely let S be a complete group-free c.s., G be a thin and surjecting abelian
group valued functor on S, and £ = (E, 1r,.) be a group coextension of S by G.
3. COEXTENSIONS 123
Anticipating on later results we note that
(g. X )(h. y) ba,bg) · (x(h.y)) = ba,bg) • ((h.y)x)

ba,bg) · (bb,ah) .yx) = (ba,bg)('"rb,ah)) .xy
whenever x E Ea, y E Eb, g EGa, hE Gb.
If x JC y holds in E, then 1rx JC 1ry holds in S and JrX = 1ry. Conversely,

assume 1rx = 1ry = a E S, so that y = g . x for some g E G a . If a ~ S E ( S) ,
then Ga = {1} since G is surjecting, g = 1, x = y, and x JC y. Now let
a E SE(S), so that e = E(x) E S. Take any p E Ee. Since ea =a we have
px E Ea and px = k. x for some k E Ga. Since G is surjecting, 'Yea is
surjective, and g = he a h) k for some h E Ge. Then '
'
(h.p)x = heah).px = g.x = y;
'
thus y ~9-C x. Similarly x ~9-C y. Thus ker 1r = JC. Then E is complete, by
Proposition 1. 7.
Now every set Ea is an JC-class. Note that t E St (Ea) if and only if either
t = 1 E E 1 or t E Ec with ac = a. Since G is thin, "fa c9 ="fa 1g = g for all
g E G a, and (g . x) t = g . xt for all x E E a and t E E c . Thus (g' . x) t = g . xt
'
for all x E Ea and t E St (Ea).
Let x E Ea. If 9t E r(Ea), then gtx = tx = g.x for some unique g EGa.
Moreover g does not depend on the choice of x: if y E Ea, then y = xu for
some u E St (Ea), and
g •y = g • xu = (g . x) u = txu = ty = 9t y
by the above. Hence there is a mapping ()a : r(Ea) --+ Ga such that gtx =
()a9t. X for all X E Ea and t E St(Ea). Ift,u E St(Ea), then, for any x E Ea,
()a(9t9u) • X = 9t9uX = ()agt • 9uX = ()a9t • (()agu • x),
and ()a(9t9u) = (()agt)(()agu). Thus ()a is a homomorphism.

The homomorphism ()a is injective, since 9t Pa = 9u Pa implies 9t = 9u. It
is surjective since both G a and r(Ea) act transitively on Ea: if g E G a, then
g. Pa = 9tPa = ()agt . Pa for some 9t E r(Ea), and g = ()a9t.
Finally we show that ()a is a natural isomorphism: the square

124 V. GROUP CO EXTENSIONS.
r(a) ~ Ga
'Ybl l'Ya,c
r(b) ----o;: Gb
commutes whenever b = ac ~9-C a. This is trivial if c = 1 E 8 1 . If c E S, take
any x E Ea and z E Ec, so that xz E Eb. Then
(gtx) z = (Oagt. x)z = 'Yac()a9t. xz,

'
for all 9t E r(Ea), whereas, by Proposition 3.1,
(gtx)z = ('Yf:gt)xz = ()b'Yb9t·xz.D

Thus, when S is complete (or finite) and group-free, exact JC-coextensions of
S coincide with commutative group coextensions of S by thin surjecting functors.
Group coextensions are therefore especially interesting in this case. We saw that
Theorem 3.4 does not extend to arbitrary commutative semigroups; nor are there
similar results for finite semigroups in general.
Corollary 3.5. Let T be a complete group-free semigroup. An abelian group
valued functor on T is (naturally isomorphic to) the Schutzenberger functor of
a c.s. if and only if it is thin and surjecting. If T is finite, then an abelian group
valued functor on T is (naturally isomorphic to) the Schutzenberger functor of
a finite c.s. if and only if it is thin, finite, and surjecting.
When T is complete group-free, we call an abelian group valued functor on
T Schntzenberger when it is thin and surjecting.
Corollary 3.6. A c.s. is a complete archimedean semigroup if and only
if it is a commutative group coextension of a nilsemigroup by a thin surjecting
functor.
When G is a thin surjecting functor on a nilsemigroup N, then Ga is trivial
for every a -I= 0 in N, and only G0 can be nontrivial.
Proof. If S is a complete archimedean semigroup, then S is an ideal extension
of a group by a ni1semigroup N and S jJC is a nilsemigroup. If conversely S
is a commutative group coextension of a nilsemigroup N by a thin surjecting
functor, then S is complete by Theorem 3.4 and archimedean by Proposition 1.5,
since S jJC ~ N is archimedean. D
Corollary 3.7. A c.s. S is elementary if and only if S is a commutative
group coextension of a nilmonoid by a thin surjecting functor G such that G 0
is trivial.
4. GROUP COEXTENSIONS 125
Proof. The Schiitzenberger functor of an elementary semigroup has these

properties. Conversely let S is a commutative group coextension of a nilmonoid
by a thin surjecting functor G such that G0 is trivial. By Theorem 3.4, S is
complete and S j']{ is a nilmonoid. By Proposition 1.2, S has two idempotents
e > f. Now H 1 = {!}, since r(H1 ) ~ G0 is trivial, and {!} is an ideal of
S since it is the inverse image of {0} under the projection S ---t Sj']{; thus f
is a zero element of S and S is elementary by Corollary 1.8. D
When G is a thin surjecting functor on a nilmonoid N 1 , then~~: G 1 ---t Ga

is surjective for every a-:/: 0 in N, so that every group G a -:/: G0 is a homomorphic
image of G 1 .
4. GROUP COEXTENSIONS.
Group coextensions are very similar to group extensions and are constructed
in much the same way; this completes the results in Section 3. The results in this
section are from Grillet [1974]; similar results for semigroups in general were
obtained independently by Leech [1975] and the author [1974] (see also Grillet
[ 1995]). Schreier-like extensions were first considered, for semi groups in general,
by Redei [1952] and Strecker [1969]; for commutative semigroups, by Inasaridze
[ 1965], [ 1967] and Lugowski [ 1966].
1. Two commutative group coextensions c = (E,1r,.) and c' = (E 1,1r1 ,.)
of S by G are equivalent in case there is an isomorphism 0 : E ---t E' which
preserves projection to S and the action of G: that is, 1r1Ox = 1rx for all x E E,
so that OEa = E~ for all a, and O(g.x) = g.Ox whenever x E Ea and g EGa.
Group coextensions need only be constructed up to equivalence.
When c = (E, 1r,.) is a commutative group coextension of S by G = (G, r),
it is convenient to denote ra ,t (g) by l; then
(g • x) y = i .xy (1)
when x E E a, y E Eb, and g E G a . Also
1a = 1, (gh)a = ga ha, gl = g, and (ga)b =gab,

whenever defined, since r a ,t is a homomorphism and (G, r) is a functor.
Group coextensions are constructed, like group extensions, by Schreier's
method [ 1926]. A cross section of c (actually, of the congruence ker 1r induced
by 1r) is a family (p a) aES of elements of E such that p a E E a ( 1rpa = a) for

all a E S. Since G a acts simply and transitively on Ea, every element x of E
can be written uniquely in the form x = g • p a with a = 1rx E S and g E Ga .
For every a,b E S we have PaPb E Eab and
(2)
for some unique sa,b E Gab· The family s = (sa,b)a,bES is the factor set ofthe
group coextension C. relative to the cross section p. For all g • p a and h. Pb E E,
(I) and (2) yield:
Pa (h • Pb) = (h • Pb) Pa = ha • (Pb Pa) = ha • (pa Pb) = (ha Sa ,b) • Pab
and
for all a, b E S and g E G a, h E G b • Thus the operation on E is determined

by the factor set and we obtain the following construction, which is similar to
Schreier's Theorem.
Theorem 4.1. Let S be a commutative semigroup and G = (G,1) be an

abelian group valued functor on S.
Let s = (sa,b)a,bES be a family such that sa,b E Gab and
(C)
(A)
for all a,b,c E S. Let E.(s) = (E(s),1r,.), where E(s) = UaES Ga x {a} is
the set of all ordered pairs (g, a) with a E S and g E G a, with multiplication
defined for all a,b E S and g E Ga, hE Gb by:
(3)
and 1r(g,a) = a, g. (h,a) = (gh,a). Then C.(s) is a commutative group

coextension of S by G.
q conversely C. is a commutative group coextension of S by G and s is the

factor set of C. relative to any cross section, then (A) and (C) hold and C. is
equivalent to C. (s) .
Proof. In E( s) we have
((g,a)(h,b))(k,c)
(g,a)((h,b)(k,c))
(g bc s a,bc hac sab,c kab ' a (be)) ·
Hence associativity in E(s) is equivalent to (A). Commutativity follows from

(C). We see that Ga acts simply and transitively on E(s)a = { (g,a) J g EGa},
and that
(g.(h,a))(k,b) = (gh,a)(k,b) = (gbhbsa,bka,ab) =gb.((h,a)(k,b));
thus (I) holds and C:(s) is a commutative group coextension of S by G.

Conversely let C: = (E, 1r, .) be any commutative group coextension of S by
G and s be its factor set relative to a cross section p. We already proved (2)
and (M). By (M),
a
sa,bc sb,c • Pabc '
s~ b sab c • Pabc ;
' '
hence (A) holds. (C) holds since E is commutative. Every element of E can
be written in the form g • Pa for some unique a E S and g E G a; hence the
mapping(}: E(s)-+ E, (g,a) I---+ 9·Pa is bijective. Then (M) and (3) show
that (} is an isomorphism, which we see preserves projection to S and the action
of G. Hence the group coextensions C: and C: (s) are equivalent. D
A commutative factor set on S with values in G is a family s = (sa,b)a,bES
such that sa bE Gab for all a,b E Sand (A) and (C) hold. Relative to Pa = (l,a),
'
the factor set of C:(s) is s, since (l,a)(l,b) = (sa b• ab) = sa b. (l,ab). By
' '
Theorem 4.1, s is a commutative factor set if and only if it is the factor set of a
group coextension relative to some cross section.
Commutative factor sets sometimes turn up in unexpected places. Factor sets
with values in lR (more precisely, in the constant abelian group valued functor
G on S with Ga = lR for all a E S) determine nonprobabilistic information
measures on sn (Ebanks [1979]). Factor sets also turn up in the construction
of certain semigroup algebras. Normally the product a.b of a, b E S in K[S]
is their product ab in S. One may define a different multiplication on K[S] by
a.b = sa b ab, where sa b E K; this yields a commutative ring if and only if
' '
sb a = sa b and sa b sab c = sa be sb c for all a, b, c E S. This construction has
' ' ' ' ' '
been used by Kunz [ 1966] and others.

2. Equivalence of group coextensions can be detected from their factor sets:
Proposition 4.2. Two group coextensions of S by G are equivalent if and
only if there exists a family u = (ua)aES with ua EGa for all a E S such that
their factor sets s and t (relative to any cross sections) satisfy
(E)
for all a, b E S; then both s and t are factor sets of either extension relative to
different cross sections.
Proof. Let(): E(t) --7 E(s) be an equivalence of group coextensions. Since
() preserves projection to S we have () ( 1, a) = (ua, a) for every a E S, where
ua E Ga; then
()(g,a) = (gua, a),
since() preserves the action of Ga. In E(t), (1,a)(1,b) =(tab> ab); since()
'
is a homomorphism,
() ( t a b, ab) = () ( ( 1, a) (1, b))
'
for all a, b E S. Thus (E) holds. By Theorem 4.1, (E) then holds whenever s
and t are the factor sets of two equivalent coextensions of S by G, relative to
any cross sections.
The calculation above also shows that (ua, a) (ub, b) = ta,b • (uab' ab) in
£ (s), so that t is the factor set of £ (s) relative to the cross section qa = (ua, a).
By the same calculation, if s is the factor set of a group coextension relative to
a cross section p, then t is the factor set of the same coextension relative to the
cross section qa = ua. Pa.
If conversely (E) holds, then () : E(t) --1- E(s), defined by () (g, a) =
(gua, a), is a bijection which preserves projection to S and the action of G.
Moreover,
() (g, a) () (h, b)
and () is an equivalence of coextensions. By Theorem 4.1, any two coextensions

with these factor sets must then be equivalent. 0
Two commutative factor sets s and t on S with values in G are equivalent
in case (E) holds, i.e. there exists a family u = ( ua)aES with ua E G a for all
a E S such that
for all a, b E S. By Proposition 4.2, two factor sets are equivalent if and only
if they are the factor sets of two equivalent group coextensions (relative to any
cross sections), if and only if they arise from two cross sections of the same group
coextension.
3. A commutative group coextension splits when it satisfies the equivalent
conditions in the next result.
Corollary 4.3. For a commutative group coextension E of S by G the
following conditions are equivalent:
(1) the projection 1r: E----+ S splits (there exists a homomorphism 11: S----+
E such that 1r o f1 = 18 );
(2) there exists a cross section of E which is a subsemigroup;
(3) there exists a cross section of E relative to which sa b= 1 for all a, bE S;
'
(4) relative to any cross section of E the factor set satisfies
(B)
for some family u = ( ua)aES with ua E G a for all a.

Proof. If 1r o f1 = 18 , then Im f1 = (f1a)aES is a cross section and a
subsemigroup, so (1) ====? (2). If a cross section p = (Pa)aES is a subsemigroup,
then Pa Pb = Pab' since Pa Pb = Pc for some c and 1r(pa Pb) = ab, so that
(2) ===} (1 ), and then sa,b = 1, so that (2) ===} (3); conversely (3) ===} (2).
Finally, (3) -<==:::} (4) by Proposition 4.2. D
A commutative factor set splits when (B) holds, i.e. there exists a family
u = (ua)aES such that ua EGa for all a E Sand
b -1 a
sa,b = ua uab ub
for all a, b E S; equivalently, when it is the factor set of a split coextension.
For any c.s. S and abelian group valued functor G on S, a split group
coextension of S by G always exists: indeed the trivial factor set sa b = 1
'
satisfies conditions (A) and (C). By Proposition 4.2, for any family u = (ua)aES
such that ua E G a for all a E S, the formula s a,b = u~ u;:b1 uf; then defines a
commutative factor set; that is, every family ua E G a gives rise to a split factor
set. This can also be verified directly.
Proposition 4.4. Every commutative group coextension of a semilattice by
a thin functor is a semilattice of groups and hence splits.
Proof. Let e = (E, 1r,.) be a commutative group coextension of a semilat-
tice S by a thin functor. Every thin functor on a semilattice is surjecting; by
Theorem 3.4, E is an exact JC-coextension of the semilattice S and is a semi-
lattice of groups by Corollary 1.5. Then Clifford's Theorem 111.2.1 shows that E
splits; more directly, if pe is the idempotent of the group E e , then pe p f is the
idempotent of Eef and Pe Pf = Pef. D
Free c.s. S also have the property that every commutative group coexten-
sion of S splits. Semigroups with this property are studied in more detail in
Chapter XIV.
4. A commutative factor sets is normalized when sea= sa e = 1 whenever
' '
e 2 = e and ea = a. This property may always be assumed if G is thin:
Proposition 4.5. Every commutative factor set with values in a thin functor
is equivalent to a normalized factor set.
Two normalized commutative factor sets s and t on S with values in a thin
functor G are equivalent if and only if there exists a family u = (ua) aES such
that ua E G a for all a E S, ue = 1 whenever e is idempotent, and
(E)
for all a,b E S.
Proof. Let E be a group coextension of S by some thin abelian group valued
functor G. As in Proposition 1.2, every idempotent e of S lifts to an idempotent
of E: indeed ge = g 1 = g for all g E G e , since ee = e = el, so that, relative
to any cross section p, s;; • Pe E Ee is idempotent by (M). Hence there is a
'
cross section q so that qe is idempotent whenever e is idempotent. The factor
set t relative to q is then equivalent to s and satisfies tee = 1 whenever e is
'
idempotent. Then t is normalized: if ea = a, then ge = g for all g E Ga , since
G is thin; hence (A), applied to e, e, and a, yields t~ e tee a = te eat: a and
' ' ' '
tea= 1.
'
Now let s and t be normalized and satisfy (E). When e is idempotent, then
1 = tee = see u: u; 1 u: = ue, since G is thin; hence the last part of the
' '
statement follows from Proposition 4.2. D
5. Theorem 4.1 and Proposition 4.2 imply that the equivalence classes of com-
mutative group coextensions of S by G constitute an abelian group Ext (S, G) .
Indeed commutative factor sets constitute an abelian group Z under pointwise
multiplication (st)a b = sa b ta b• since this operation preserves (A) and (C).
' ' '
Split factor sets (of the form u~ u;;} ub) constitute a subgroup B of Z. By
Proposition 4.2 there is a one-to-one correspondence between the cosets of B in
Z and the equivalence classes of commutative group coextensions of S by G.
Therefore the latter are the elements of an abelian group, the extension group
Ext(S,G) ~ Z/B.
Ext (S, G) generalizes the usual Ext for abelian groups:
Proposition 4.6. If C is an abelian group, then every abelian group valued
functor G on C is naturally isomorphic to a constant functor A, and then
Ext (C, G) (in the semigroup sense) coincides with Ext (C, A) (in the abelian
group sense).
Naturally we identify the constant functor A and its constant value A.
Proof. C has an identity element 1, so that C = C 1 . By definition, Ia 1 is
'
the identity on G a , and then 1 at u and 1a t are mutually inverse isomorphisms
' '
whenever u = c 1 inC. Hence 1 1 a: G 1 ---+ Ga is a natural isomorphism from
'
G to the constant functor G 1 . The result then follows from Proposition 3.2. 0
When S is not an abelian group, Ext (S, G) is most interesting when S
is group-free and G is thin and surjecting, for then Ext (S, G) classifies c.s.
rather than just coextensions; that is, Ext (S, G) classifies the commutative semi-
groups T with (up to isomorphisms) T jJC = S and Schiltzenberger functor G
(Theorem 3.4 ).
6. Results in the last chapters will simplify the generally daunting task of
finding all factor sets on a given c.s. with values in a given functor, or at least
their equivalence classes. For now we give one example where all factor sets are
readily found.
By Corollary 3.6, a complete archimedean semigroup is a commutative group
coextension of a nilsemigroup N by a thin surjecting functor G; then G a is
trivial for every a i= 0 in N, and only G0 can be nontrivial.
More generally, an abelian group valued functor G on a semigroup with zero
S is almost null when G is thin and G a is trivial for every a E S\ 0; then G
is surjecting as well.
Proposition 4. 7. Let S be a semigroup with zero and G be an almost
null abelian group valued fonctor on S. Up to isomorphism, a c.s. is a group

coextension of S by <G if and only if it is an ideal extension of G 0 by S,
and then there is a one-to-one correspondence between normalized factor sets
on S with values in <G and partial homomorphisms from S\0 to G 0 . Hence
Ext(S,<G) ~ PHom(S\0, G 0 ) ~ Hom(G(S\0), G 0 ).
Proof. Let s be a normalized factor set on S with values in <G. In E( s),
H = { (g, 0) I g E G 0 } is a subgroup of which is isomorphic to G 0 , since
so,o = 1, and an ideal. The Rees quotient E (s) / H is isomorphic to S, since <G
is almost null. Every commutative group coextension of S by <G is isomorphic
to some E( s) and is likewise an ideal extension of G 0 by S.
Conversely, every ideal extension T of G0 by S comes with a projection
1r : T ---t T jG 0 ~ S. Also G 0 acts on itself by left multiplication, and every
trivial G a acts trivially (how else) on 1r -l (a) = {a}. This makes T a group
coextension of S by <G.
Let s be a normalized factor set on S with values in <G. Then sa,b E G 0
when a, b E S\0 and ab = 0, and sa,O E G 0 ; also s 0 ,0 = 1 and sa,b = 1
whenever ab =/: 0 in S, for then Gab= {1}. When a,b E S\0, (A) reads:
if ab =/: 0, since <G is thin, and

or
if ab = 0. Thus
<.p(a) = sa 0
'
is a partial homomorphism of S\0 into G 0 . Moreover, s is completely deter-
mined by <.p: indeed sa,O = <.p(a) for all a =/: 0; sa,b = <.p(a) <.p(b) whenever
a, b =/: 0 and ab = 0; all other values of s are 1.
Conversely let <.p : S\0 ---t G 0 be a partial homomorphism. Let sa 0 =
'
s 0 a= <.p(a) for all a=/: 0, sa b = <.p(a) <.p(b) whenever a,b =/: 0 and ab = 0, and
' '
sa b = 1 whenever ab =/: 0. The calculations above show that s is a commutative
'
factor set.
We now have a one-to-one correspondence between commutative factor sets
on S with values in <G, and partial homomorphisms S\ 0 ---t G 0 , which we
see preserves pointwise multiplication. By Proposition 4.4, two normalized factor
sets on S with values in <G are equivalent if and only if they are equal. Hence
5. 8UBDIRECTLY IRREDUCIBLE 8EMIGROUPS 133
Ext (S, G) ~ PRom (S\0, G0 ), and PRom (S\0, G0 ) ~ Hom (G(S\0), G0 )

by Proposition III.3.3. 0
In the above the coextension E( s) can be constructed as follows. In E( s)
we can identify ( 1, a) and a for every a E S\ 0, and identify (g, 0) and g for
every g E G 0 . Then E(s) = (S\0) U G 0 , with multiplication * given for all
a,bES\0 and g,hEG 0 by: a*b=ab ifab#O inS; a*b=sab=c.p(a)c.p(b)
'
if ab = 0 inS; g*a = a*g = sao9 = c.p(a)g; 9* h = gh. This is precisely
'
the construction in Lemma 1.3.7. Thus, group coextensions and ideal extensions
give equivalent constructions in this case.
7. Summary. The results obtained so far construct every complete (for
instance, finite) commutative semigroup S in three ways.
(l ). S is a semi lattice of archimedean semi groups containing idempotents.
Such an archimedean semigroup is an ideal extension of an abelian group G by
a commutative nilsemigroup N and is determined by a partial homomorphism of
N\0 into G, or by a homomorphism of G(N\0) into G. Group coextensions
give essentially the same construction. However, S is not in general readily
reassembled from its archimedean components.
(2). S is a subdirect product of its Ponizovsky factors, which consist of an
abelian group, a commutative nilsemigroup, and elementary semigroups. An ele-
mentary semigroup GUN is an ideal extension of a commutative nilsemigroup N
and is determined by a suitable action of the abelian group G on N; equivalently,
G U N is a group coextension of its monoid of orbits, which is a commutative
nilmonoid. Then S can be reassembled from its Ponizovsky factors by means of
partial homomorphisms, which may, however, be difficult to construct.
(3). S is a commutative group coextension of the complete group-free semi-
group S jJ-{ by a thin surjecting functor, which Theorem 4.1 then constructs from
factor sets on S jJ-{. The group-free semigroup S jJ-{ can be assembled, with the
usual difficulties, from nilsemigroups: its archimedean components are nilsemi-
groups; its Ponizovsky factors are the trivial group, a commutative nilsemigroup,
and nilmonoids. Finite nilsemigroups or group-free semigroups have also been
constructed, by induction or from trees, by Yamada [1964], Tamura [1968], John
[ 1973], and others.
5. SUBDIRECTLY IRREDUCIBLE SEMIGROUPS.
As an application of the results in this chapter we determine all subdirectly

irreducible finite commutative semigroups. The results are from Grillet [ 1977].
1. From Chapter IV we know:
A complete c.s. which is subdirectly irreducible is either a group, or a nilsemi-
group, or an elementary semigroup (Corollary IV.4.6).
An abelian group is subdirectly irreducible if and only if it is a nontrivial
cyclic or quasicyclic p-group (Proposition IV.l.7).
A commutative nilsemigroup N with enough minimal elements is subdirectly
irreducible if and only if N -=f. 0 and its greatest pure congruence P is the equality
on N (Proposition IV.3 .6). This result can be restated as follows. Let M denote
the set of all minimal elements of N. Define
a JY( b -{=:::::> (VuE N 1 )(ua EM -{=:::::> ub EM ~ ua = ub).
Then JY( is a congruence on N. In fact, M is a pure congruence, so that JY( ~ P.
Call N weakly irreducible in case JY( is the equality.
Proposition 5.1. A commutative nilsemigroup N with enough minimal
elements is subdirectly irreducible if and only if it is weakly irreducible and has
just one minimal element.
Proof. Both conditions are necessary, since M is a P-class and JY( ~ P, and P
is the equality when N is subdirectly irreducible (Proposition IV.3.6). Conversely
assume that N is weakly irreducible and has just one minimal element m. Then
N has at least two elements. Let a, b E N, a -=f. b. Since a JY( b does not hold
there exists u E N 1 such that, say, ua = m, ub -=f. m. If ub = 0, then a P b
does not hold. If ub -=f. 0, then ub > m, m = vub for some v E N, whereas
vua < ua = m and vua = 0; hence again a P b does not hold. Thus P is the
equality and N is subdirectly irreducible. D
2. We now tum to elementary semigroups S = G U N. The cases where G
or N are trivial are covered by:
Proposition 5.2. If S
has at least two elements, then S is (finitely) sub-
directly irreducible if and only if S 0 is (finitely) subdirectly irreducible, if and
only if S 1 is (finitely) subdirectly irreducible.
When S is trivial, then S U {0} and S U {1} have two elements and are
subdirectly irreducible .
. Proof. Assume that S does not have a zero element. Every congruence e
on S extends to a congruence e0 on S 0 , whose classes are {0} and all the
e-classes. If e is proper then e0 is proper.
Conversely let e be a congruence on S 0 whose restriction to S is the equality

on S. If e is not the equality on S 0 , then 0 e z for some z E S, and then
0 e xz and xz = z for all x E S, contradicting the hypothesis. Therefore a
proper congruence on S 0 restricts to a proper congruence on S.
If S 0 is not subdirectly irreducible, then the equality on S 0 is the intersection

of proper congruences on S 0 , and the equality on S is the intersection of their
restrictions to S, which are proper congruences; hence S is not subdirectly
irreducible. If conversely S is not subdirectly irreducible, then the equality on
S is the intersection of proper congruences on S, and the equality on S 0 is the
intersection of their extensions to S 0 , which are proper congruences; hence S 0
is not subdirectly irreducible. Similarly, S is finitely subdirectly irreducible if
and only if S 0 is finitely subdirectly irreducible.
The proof is similar for S 1 . Assume that S does not have an identity
element. Every proper congruence e on S extends to a proper congruence e 1
on S 1 , whose classes are { 1} and all the e-classes. Conversely let e be a
congruence on S 1 whose restriction to S is the equality on S. If e is not the
equality on S 1 , then 1 e e for some e E S, and then X e xe and xe = X for
all x E S, contradicting the hypothesis. Therefore a proper congruence on S 1
restricts to a proper congruence on S. As above it follows that S is (finitely)
subdirectly irreducible if and only if S 1 is (finitely) subdirectly irreducible. 0
3. We now let S = G U N be an elementary semi group in which G and
N have at least two elements and N has enough minimal elements. We saw in
Section IV.S that the group G acts on S by left multiplication; under this action,
the orbits are precisely the J-C-classes of S (Proposition IV.5.1 ); they constitute a
commutative nilmonoid D1 = S jJ-C.
Lemma 5.3. .if S is finitely subdirectly irreducible, then N is weakly
irreducible and M is an orbit.
Proof. The definition of )V( shows that {0} and every {m} <;;;; M is an
M-class. Extend )V( to a congruence M' on S as follows: a M' b if and only
if either a = b E G or a, b E N and a )V( b. The intersection of M' and the
Rees congruence of the ideal M U { 0} is the equality on S; therefore M' is the
equality on S, and N is weakly irreducible.
Every P-class is a union of orbits (Lemma IV.5.2); hence M is a union of
orbits. If M contains two distinct orbits M 1 and M 2 , then M 1 U {0} and
M 2 U {0} are ideals of S, whose Rees congruences have the equality on S as
intersection. When S is finitely subdirectly irreducible this cannot happen, and

M is a single orbit. D
An elementary semigroup S = G U N is homogeneous when G acts simply
on N\ 0; equivalently, when gx = hx of 0 implies g = h when g, h E G;
equivalently, when gx = x of 0 implies g = 1 when g E G; equivalently, when,
in the Schtitzenberger functor of S, 1'1
is an isomorphism for every nonzero
orbit A.
Lemma 5.4. lf S is finitely subdirectly irreducible, then it is homogeneous.
Proof. Let g E G and e E N satisfy ge = e of 0. Since N has enough
minimal elements there is a minimal element n :::; e, n = ue for some u E N 1 .
For every mE M we now have m = hn for some hE G by Lemma 5.3 and
gm = huge= hue= m.
Let e be the congruence on S defined by: a e b if and only if gi a = b for
some a E Z; the e-classes are the orbits under the action of (g) ~ G. We see
that e is the equality on M u { 0}' since gx = X for all X E M u { 0} Hence the
0
intersection of e and the Rees congruence of the ideal M U {0} is the equality
on S. Since S is finitely subdirectly irreducible, it follows that e is the equality
on S. But g e 1, so g = 1. D
Lemma 5.5. lf S is subdirectly irreducible, or finitely subdirectly irre-
ducible, then so is G.
Proof. For every (normal) subgroup H of G, let H* be the congruence on S
defined by: a H* b if and only if b E H a. We see that H* is a pure congruence
on S, and that g H* h if and only if g- 1 h E H (so that H and H* induce the
same congruence on G).
Assume that {1} is the intersection of subgroups (Hi)iEI of G. If a Hi b
for all i, then either a = b = 0, or for each i E I there is some hi E Hi such
that b =hi a of 0. By Lemma 5.4 there is only one g E G such that b = ga of 0,
and then g = hi E Hi for all i, g = 1, and a = b. Thus niEI Hi is the
equality on S. If S is subdirectly irreducible, then some Hi is the equality on
S, and then Hi = {1}. This shows that G is subdirectly irreducible. If similarly
S is finitely subdirectly irreducible, then G is finitely subdirectly irreducible. D
Theorem 5.6. Let S = G U N be an elementary semigroup, in which G
and N have at least two elements and N has enough minimal elements. Then
S is (finitely) subdirectly irreducible if and only if G is (finitely) subdirectly
irreducible, N is weakly irreducible, the minimal elements of N form an orbit,
and S is homogeneous.
5. SUBDIRECTLY IRREDUCIBLE SEMIGROUPS 137
Proof. These conditions are necessary by Lemmas 5.3, 5.4, 5.5. Conversely
assume that S has all four properties.
Let e be a proper congruence on S, so that a e b for some a, b E S, a i- b.
We show that e is not the equality on M. If a, b E G, then for any m E M
we have am, bm E M, am e bm, and am i- bm since S is homogeneous. If
a E G and b E N, then for any m E M we have am E M, bm = 0, and n e 0
for some n E M; since M is an orbit it follows that n e 0 for all n E M; since
S is homogeneous and G is nontrivial we have m e n for some m i- n in M.
Finally assume a, b E N. Since N is weakly irreducible there exists u E N 1
such that, say, ua E M and either ub E M, ub i- ua, or ub rt- M. In the
first case, e is not the equality on M. Now assume ub rt- M. If ub = 0, then
ua e 0 with ua E M, and as above e is not the equality on M. If ub i- 0,
then ub > m for some mE M, m = vub for some v EN, vua = 0 since
ua E M, m = vub e vua = 0, and again e is not the equality on M.
Choose any m E M. For every congruence e on S define a congruence e'
on G by: g e' h if and only if gm e hm. If e is a proper congruence, then
by the above e is not the equality on M and e' is a proper congruence. Also
e = niEJ ei implies e' = niEJ e~: indeed g e~ h for all i is equivalent to
gm ei hm for all i, to gm e hm, and tog e' h. Since G is (finitely) subdirectly
irreducible, the equality on S cannot be the intersection of (finitely many) proper
congruences on S, for then the equality on G would be the intersection of (finitely
many) proper congruences on G. Thus S is (finitely) subdirectly irreducible. 0
Corollary 5.7. A .finite c.s. is subdirectly irreducible if and only if it is one
of the following:
a nontrivial cyclic p-group;
a finite nilsemigroup on which P is the equality;
{0, 1}; a nontrivial cyclic p-group with a zero element adjoined; a finite
nilmonoid on which P is the equality; a homogeneous elementary semigroup
S = G U N in which G and N have at least two elements, G is a cyclic
p-group, N is weakly irreducible, and the minimal elements of N form an orbit.
4. The elementary semigroup S = G U N is a group coextension of its
monoid of orbits 0 1 , as in Corollary 3.7. We note the following:
N has enough minimal elements if and only if 0 does;
M is an orbit if and only 0 has only one minimal element (namely, M); then
0 has enough minimal elements if and only if M is the least nonzero element
of 0;
S is homogeneous if and only if its Schiltzenberger functor is naturally iso-

morphic to the almost constant functor ( G, "(), in which G A = G for every
nonzero orbit A, G 0 = {1}, and "fA,T is the identity on G if AT# 0.
Proposition 5.8. Let S = GUN be a homogeneous elementary semigroup,
in which G and N have at least two elements, N has enough minimal elements,
and the minimal elements of N form an orbit M. Let s be any factor set of S as
a group coextension of its monoid of orbits 0 1 . Then N is weakly irreducible
if and only if, for every A, B E 0 with A M B and A # B, the jUnction
f(T) = sA,T sB~T EGis not constant on 'J = {T E 0 I AT= M} = {T E
0 I BT = M}.
Proof. As in Theorem 4.1 we may assume that S = E( s). Then S consists
of 0 and all pairs (g,A) with g E G, A E 0 1 , A -=f 0, with multiplication
(g,A)(h,B) = (ghsAB' AB) if AB # 0, (g,A)(h,B) = 0 if AB = 0. Also
'
(g,A) EM if and only if A= M, (g,A) E G if and only if A= 1 (this
identifies (g, 1) and g).
Assume that N is weakly irreducible. Let A, B E 0 satisfy A M B and
A# B; in particular, A,B # 0, 1,M, A,B > M, and 'J = {T E 0 I AT=
M} = {T E 0 I BT = M} # 0. Take any T E 'J and let a = (sB,T' A)
and b = (sAT' , B). Then a# b and aM b does not hold. However, au E
M
if and only if bu E M, for all u E N 1 , by the choice of A and B. Therefore
there exists u E N 1 such that au, bu E M and au -=f bu. In fact u E N, since
A,B > M. Hence u = (g,U) for some U E 0; then AU= BU = M and
(sr,B sA,U g, M) = au # bu = (sr,A sB,U g, M);
therefore s A,U sB~U # s A,T sB~T and f is not constant on 'J.

Conversely, assume that f is not constant on 'J, whenever A, B E 0, A M B,
and A -=f B. Let a, b E N, a -=f b. If either of a or b is zero or minimal, then
a M b cannot hold in N. Let a and b be neither zero nor minimal, so that
a= (g,A), b = (h,B), with A,B > M. If A= B, then g -=f h, at -=f bt
whenever t # 0, and aM b does not hold. Now assume A -=f B. If AM B
does not hold in 0, then a M b does not hold in N. Let A M B.
Let T E 'J and t = (1, T) . Then at = (g sA T, M) , bt = (h s B T, M) ,
' '
and at, bt E M. If at -=f bt, then a M b does not hold. If at = bt, then we use
the hypothesis that f is not constant: s A,U s B :u # s A,T s B :r for some U E 'J.
Then g s A,U # h s B,U and u = (1, U) satisfies au # bu and au, bu E M;
hence a JY( b does not hold. Thus JY( is the equality on N. 0

By Proposition 5.8, if 0 is weakly irreducible then so is N. In fact, if S
splits, then s can be chosen so that sA T s B1T = 1 for all T, and N is weakly
' '
irreducible if and only if 0 is. But Proposition 5.8 does not in general imply that
n is weakly irreducible.
Example 5.9. Let 0 = {A,B,M,O} with A 2 = AB = B 2 = M and all
other products equal to 0. In particular, all products of three or more elements
of n are zero. We have A JY( B, with 'J = {A, B}, so that 0 is not weakly
irreducible. On the other hand, a normalized factor set orr 0 1 consists of sA A,
'
sA B , s B A , and s B B E G such that s B A = sA B . If G is not trivial there is
' ' ' ' '
a factor set s such that s A,A sJ3~A =f s A,B sJ3~B. Then N is weakly irreducible
in the corresponding group coextension S = GUN, by Proposition 5.8. 0
Chapter VI.
FINITELY GENERATED SEMIGROUPS.
This chapter contains basic properties of finitely generated commutative semi-

groups: Redei's Theorem, various finiteness properties, and the Completion The-
orem. The results indicate that these semigroups are assembled somehow from
cancellative semigroups and nilsemigroups, arranged along a finite semilattice.
This similarity to finite semigroups is pursued much further in the next chapter.
This chapter also studies subelementary semigroups, subdirectly irreducible
semigroups, and other smaller classes.
For the sake of brevity we often abbreviate "finitely generated" as f.g. m
what follows.
1. REDEl'S THEOREM.
Redei's Theorem states that the congruences on a finitely generated free c.s.
are themselves finitely generated. This section contains Preston's proof and some
consequences of the Theorem. Other proofs were devised by Drbohlav [1963]
and Budach [ 1964] and still another proof will be found in Chapter XIII.
1. Let X be a set. To every element a = l:xEX ax x of the free c.s. Fx
corresponds a monomial Xa = ITxEX xax in the polynomial ring Z[X] with
commuting indeterminates in X. We see that Xa Xb = Xa+b for all a, b E Fx .
Freyd [ 1968) and Preston [ 1975] observed that every ideal I of Z[X] induces
a congruence e on Fx, for which a e b if and only if Xa - Xb E I; and that,
conversely, every congruence on Fx is induced by an ideal of Z[X]:
Lemma 1.1. Let e be a congruence on Fx. Let I(e) be the ideal of Z[X]
generated by all Xa - Xb with a e b. Then a e b if and only if Xa - Xb E
I(e).
141
142 VI. FINITELY GENERATED SEMIGROUPS.
Proof. Since e is a congruence, I(e) is the set of all linear combinations

2: 1;£j;£r mj (Xa 1 - Xb1 ) in which r ~ 0, mj is a positive integer, aj -=f. bj
in Fx, and aj e bj, for all j. If Xa- Xb E J(e), where a -=f. b in Fx, then
b·)
X
a
- X
b
= 2: 1;£j;£r mj ( X a·1 - X 1 ,
with mj > 0, aj -=f. bj, and aj e bj, for all j. Then r > 0, since Xa -=f. Xb.
We prove by induction on m = 2: 1;£j;£r mj > 0 that (*) implies a e b.
If m = 1, then r = 1, m 1 = 1, a = a 1 , b = b1 , and a e b. Now let
m > 1. The coefficient of Xa in the right hand side of (*) is I.:a 1=a mj -
I.:bj=a mj = 1; therefore ak = a for some k. If bk = b, then a e b and we
are done. If bk -=f. b, then the coefficient of Xbk in the right hand side of ( *) Is
2:a1=bk mj - 2:b1=bk mj = 0, since bk -=1- a= ak and bk -=f. b. Now
I.:aj=bk mj = I.:bj=bk mj ~ mk > 0;

hence bk = a1 for some l. Then ak e b1 • In the right hand side of (*) we can
now replace (Xak - Xbk) + (Xa 1 - Xb 1 ) by (Xak - Xb 1 ) if ak -=f. b1 , by
0 if ak = bz . This decreases m by 1 or by 2. The induction hypothesis then
yields a e b. Conversely, a e b implies Xa- Xb E J(e) by definition. D
Theorem 1.2 (Redei). Congruences on a finitely generated free commutative
semigroup satisfy the ascending chain condition.
Proof. If
e1 :;t:c e2 :;t:c . . . c:;t: en c:;t: ...
is a strictly ascending sequence of congruences on Fx, then
J(e 1 ) ~ I(e 2 ) ~ ··· ~ I(en) ~ · · ·
is a strictly ascending sequence of ideals of Z[X], by Lemma 1.1. This contradicts

the Hilbert Basis Theorem if X is finite. 0
Corollary 1.3. Every ideal of a fg. free c.s. is finitely generated Every
antichain of a fg. free c.s. is finite.
An antichain of a partially ordered set is a subset A such that a < b never
happens when a,b EA. The second part of Corollary 1.3 is Dickson's Theorem
(Proposition 1.5.4).
1. REDEl'S THEOREM. 143
Proof. When Fx is finitely generated, the Rees congruences on Fx satisfy

the ascending chain condition by Redei's Theorem; therefore the ideals of Fx
satisfy the ascending chain condition, and are finitely generated.
Now let A be an antichain of Fx. The ideal of Fx generated by A is
I = { t E Fx It ~ a for some a E A}.

Since A is an antichain, every element of A is a minimal element of I, and must
belong to any subset which generates I as an ideal; therefore A is finite. D
Corollary 1.4. When S is a fg.c.s., then the congruences on S and the
ideals of S satisfy the ascending chain condition.
Proof. By Corollary 1.5.6 there is a surjective homomorphism 1r : F --+ S
where F is finitely generated and free. If there was a strictly ascending sequence
of congruences (or of ideals) of S, their inverse images under 1r would constitute
a strictly ascending sequence of congruences (or of ideals) of F, contradicting
Theorem 1.2 (or Corollary 1.3). D
By Corollary 1.4, every congruence (every ideal) of a f.g.c.s. is finitely gen-
erated as a congruence (as an ideal). The converse holds: a c.s. all of whose
congruences are finitely generated is itself finitely generated (Budach [ 1964]); a
c.s. all of whose ideals are finitely generated is itself finitely generated (Satya-
narayana [ 1977]).
On the other hand, the subsemigroups of a f.g.c.s. do not in general satisfy
the ascending chain condition, and are not in general finitely generated. F.g.c.s.
whose every subsemigroup is finitely generated are studied in Section 7.
Example 1.5. Let F be the free c.s. on X = {a, b}, written multiplicatively.
F has three archimedean components: A = {am I m > 0}, B = { bn I n > 0},
and c = {ambn I m,n > 0}. Then C 2 = {ambn I m,n ~ 2}. Every subset
of C which generates C must contain
C\ C 2 = {am bn I m = 1' n ~ 2, or m ~ 2' n = 1}

and is infinite; hence C is not finitely generated.
2. By Redei's Theorem, every finitely generated c.s. is finitely presented
(= has a presentation with finitely many generators and finitely many defining
relations). Algorithms which recognize from a finite presentation of S whether
S has an identity element, is cancellative, or is a group, were given by Emelicev
[1962], [1966] and Birjukov [1966], [1967]. Algorithmic solutions of the word
problem (deciding when two products of generators are equal) were given by
Halezov [ 1966] for cancellative c.s. and Birjukov [ 1967] for c.s. in general. The
isomorphism problem (deciding when two finite presentations yield isomorphic

semigroups) was solved by Taiclin [1974], [1980]. See also Shleifer [1981]. A
number of efficient algorithms for these and other properties can be found in
Rosales & Garcia-S{mchez[ 1999], along with additional references.
2. SUBDIRECT DECOMPOSITIONS.
Redei's Theorem yields subdirect decompositions of finitely generated com-

mutative semigroups which are like Ponizovsky decompositions, with groups and
elementary semigroups replaced by cancellative and subelementary semigroups.
This result, due to Grillet [ 1975P], opens the door to the inner structure of f.g.c.s.,
beginning in this section with their archimedean components.
1. Call a congruence e intersection irreducible if it has more than one class
and is not the intersection of finitely many congruences ei ~ e. For simplicity's
sake, we call a c.s. S irreducible when it is finitely subdirectly irreducible.
Proposition 2.1. Let S be afg.c.s. Every congruence on S is the intersec-
tion offinitely many irreducible congruences. Hence S is a subdirect product of
finitely many fg. irreducible semigroups.
Proof. The first part is proved by noetherian induction. We regard the univer-
sal congruence tL on S (of which S is the only equivalence class) as the empty
intersection of congrences on S, and the intersection of just one congruence as
that congruence itself. If there exists a congruence on S which is not the in-
tersection of finitely many irreducible congruences, then, by the ascending chain
condition, there exists a congruence JY( on S which is maximal with this property.
In particular JY( i= tL and JY( is not irreducible. Therefore JY( is the intersection
JY( = niEJ ei of finitely many congruences ei ~ JY(. By the maximality of JY(
every ei is the intersection of finitely many irreducible congruences. But then so
is JY( = niEJ ei; this is the required contradiction.
In particular the equality on S is the intersection of finitely many irreducible
congruences ei. Then S is a subdirect product of the finitely many semigroups
Sjei. Since ei is irreducible, the equality on Sjei is not the intersection of
finitely many proper congruences, by Proposition 1.2.5, and every Sjei is finitely
subdirectly irreducible. Also Sjei is finitely generated like S. D
2. A subelementary semigroup is a c.s. which is the disjoint union S =
C U N of two non empty subsemigroups C and N, such that N is a nilsemigroup
and an ideal of S, the zero element of N is the zero element of S, and every
2. SUBDIRECT DECOMPOSITIONS. 145
element of Cis cancellative in S (ex= cy implies x = y, for all x,y E Sand

c E C). In particular C is cancellative.
Theorem 2.2. A finitely generated irreducible semigroup is either can-
cellative or nil or subelementary. Hence every finitely generated commutative
semigroup is a subdirect product of a cancellative semigroup, a nilsemigroup,
and finitely many subelementary semigroups.
Proof. Let S be a f.g. irreducible semigroup. There exists a finite set X
and a surjective homomorphism 1r : Fx ---+ S. The congruence e = ker 1r is
irreducible: indeed e has more than one class and is not the intersection of finitely
many congruences ei ~ e, otherwise the equality on S would be the intersection
of finitely many proper congruences, by Proposition 1.2.5, and S would not be
irreducible.
By Lemma 1.1, e is induced by an ideal I of Z[X]: 1r(a) = 1r(b) if and
only if Xa - Xb E I. We show that e is induced by a primary ideal Q (if
f g E Q and f (j. Q, then gn E Q for some n > 0 ). Indeed, in the noetherian
ring Z[X], I is the intersection I = Q 1 n Q 2 n · · · n Qr of finitely many primary
ideals Ql' ... ' QT. Then e is the intersection of the congruences Ql' ... ' QT
induced by Ql' ... ' QT. Since e is irreducible, we have e = Qj for some j
and e is induced by a primary ideal.
Now that e is induced by a primary ideal Q, let
P = {f E Z[X] I fn E Q for some n > 0}
be the radical of Q, which is a prime ideal of Z[X] (if f g E P and f (j. P, then
a a b a
g E P). Let a E F. If X E Q, then X X -X E Q and 1r(a)1r(b) = 1r(a)
for all bE F, so that 1r(a) is a zero element of S. If Xa E P, then (Xa) n E Q
for some n > 0, 1r(a)n is the zero element of S, and 1r(a) is nilpotent inS. If
Xa (j_ P, then 1r(a)1r(b) = 1r(a)1r(c) implies Xa (Xb -Xc) E Q, Xb -Xc E Q
(since Q is primary), and 1r(b) = 1r(c); thus 1r(a) is cancellative inS. Thus S
is the disjoint union S = C UN, where
C = {1r(a)ESjXa (j.P} and N = {1r(a)ESjXaEP}.
Either set may be empty. If N = 0, then S = C is cancellative. If C = 0, then
S = N is a nilsemigroup. If C, N "I- 0, then S has a zero element 0 E N, and
S = C U N is subelementary. This proves the first part of the Theorem.
Now let S be any f.g.c.s. By Proposition 2.1 and the above, S is a subdirect
product of finitely many cancellative semigroups cl ' ... ' ck' nilsemigroups
N 1 , ... , N 1, and subelementary semigroups S 1 , ... , Sm. We may assume only
146 VI. FINITELY GENERATED 8EMIGROUPS.
one cancellative component (which may be trivial): if indeed k > 1, then the
projection C of 8 in Cl X ···X Ck is cancellative, and 8 is a subdirect product
of C and the remaining components, since S --+ C and the remaining projections
still separate the elements of S. Similarly, we may assume only one nilsemigroup
component (which may also be trivial), since a nonempty subsemigroup of the
finite product N 1 x ... x Nz is a nilsemigroup. D
It follows from this proof that every congruence on Fx is the intersection
of finitely many "primary" congruences Q (such that Fx jQ is subelementary).
Drbohlav [1963], [1964] proved this directly, for congruences on any f.g.c.s.
3. Theorem 2.2 implies finiteness properties of archimedean components.
First we show:
Lemma 2.3. In a fg. subelementary semigroup S = C U N, C is fg. and
N is nilpotent (Nm = 0 for some m > 0).
Proof. Let X = { x 1 , ... , xn} generate S. We may assume that X n N =
{x 1 , ... , xk}, where k ~ n. Then Cis f.g., by X n C = {xk+l, ... , xn}.
Every element of N can now be written in the form xi 1 ··· x~k c where c E C,
t 1 , ... , tk ~ 0, and ti > 0 for some i.
Since N is a nil semi group every xi E N has a power x":i = 0. If m ~

m1 + ··· + mk, then every product xi1 · · · x~k c of m elements of N satisfies
t 1 + ··· +tk ~ m, ti ~ mi for some i, and x~1 ··· x~k c = 0. Thus Nm = 0.
The same argument shows that a nilsemigroup N which is finitely generated
(equivalently, finite) is nilpotent (Proposition IV.3.5). D
Proposition 2.4. When S is a fg.c.s., then Y(S) is finite; every archime-
dean component A of S has a cancellative power Am; and G(A) is fg.
In Proposition 2.4, Am is a nonempty cancellative ideal of A, and A is
an ideal extension of the cancellative semigroup Am by the nilpotent semi-
group A/Am.
Proof. First Y(S) is finite since it is a f.g. semilattice: if { A 1 , ... , An} gen-
erate Y(S), then every element of Y(S) can be written in the form 1 · · · A~nAi
with ti = 0 or ti = 1 for all i since A 1 , ... , An are idempotent; therefore
Y(S) is finite (with at most 2n elements).
By Theorem 2.2, S is a subdirect product of a cancellative c.s. C, a nilsemi-
group N, and finitely many subelementary semigroups Si = Ci u Ni , all of
which are homomorphic images of S and are finitely generated. By Lemma 2.3
there exists m > 0 such that Nm = NJ:' = 0 for all i. Let A be an ar-
2. SUBDIRECT DECOMPOSITIONS. 147
chimedean component of S. The projection of A in Si is archimedean and is

either contained in Ci or contained in Ni . Hence A is contained in the direct
product of C, N, some of the Ci 's, and some of the Ni 's. Then Am is con-
tained in a direct product of cancellative semigroups and trivial semigroups, and
is cancellative. Thus Am is a cancellative ideal of A.
By Propositions 11.2.6 and 11.3.4, G(Am) is (up to isomorphism) contained in
the direct product of G (C) and some of the G (Ci) 's, all of which are f. g. since
C and all Ci are f.g. by Lemma 2.3. Therefore G(Am) is f.g. By Proposition
11.2.5, G(A) ~ G(Am) is f.g. 0
In particular, a f. g. archimedean semigroup A has a cancellative ideal Am,
and G(A) is f.g. These semigroups are studied in Section 6. But Example 1.5
shows that the archimedean components of a f.g.c.s. need not themselves be f.g.
So does the next example.
Example 2.5. = G U N be the elementary semigroup in which
Let S
G ={an InEZ}~ Z, N = {bn In E Z}U{O}, and aibj = bi+i' bibj = 0
for all i,j. We see that S is finitely generated (by a, a- 1 , and b0 ) and that the
archimedean component N of S is not finitely generated. 0
Let S be a c.s. If there is an integer m > 0 such that Am is cancellative
for every archimedean component A E Y (S) of S, then the archimedean can-
cellativity index of S is the smallest such integer; otherwise the index of S is
infinite. Thus, when m > 0 ( m i- oo ), S has index ~ m if and only if Am is
cancellative for every A E Y (S) .
Lemma 2.6. Let A be archimedean with an idempotent e. If Am is
cancellative, then Am = eA =He.
Proof. eA = He by Proposition III.3.2, and He ~ Am since x = emx for
every x E He . Conversely, y E Am implies ey = e (ey) and y = ey E eA,
since Am is cancellative. 0
For instance, Example III.5.2, which is archimedean but has no nonempty
cancellative ideal, has infinite index. If on the other hand S is finitely generated,
then, by Proposition 2.4, Y(S) is finite and every archimedean component A E
Y (S) of S has a cancellative power Am ; therefore
Proposition 2.7. Every fg.c.s. has finite archimedean cancellativity index.
3. SUBELEMENTARY SEMIGROUPS.
This section contains basic properties of subelementary semigroups.

1. Every subelementary semigroup can be completed to an elementary semi-
group:
Let S = C U N be a subelementary semigroup. Then
Proposition 3.1.
c- = G (C) u c- is an elementary semigroup, and the canonical homo-
1S 1N
morphism t : s --+ c- 1 s is injective.

Proof. In c-Is, Oja = 0/b
for all a,b E C, and this provides the zero
element of c- 1 s, since (xje)(Oja) = Ojae for all xje E c-Is.
Let xja E c-Is, where a E C. If x E C, then xja = t(x) t(a)-I is
a unit of c-Is. If x E N, then (xja)n = xn jan = 0 for some n > 0 and
xja is nilpotent. Also eja # xjb with e E C, x E N, since ax E N and
beE C. Hence c-Is is the disjoint union c-Is= c-Ic U c-IN of a
group G(C) = c-Ic and a nilsemigroup c-I N. Then 1 c- s
is elementary,
1
since c-I N is an ideal of c- S, the identity element of c-I C is the identity
element of c-IS, and the zero element of c-I N is the zero element of c-IS;
t is injective by Proposition 11.1.2. 0
Since t: S--+ c- 1 S is injective we identifY xES and t(x) E c- 1 S, SO

that S becomes a subsemigroup of c-Is. Subelementary semigroups can thus
be retrieved from elementary semigroups, much as cancellative semigroups can
be retrieved from abelian groups.
2. When S is subelementary, the orbits of c- 1 S induce a partition of S.
Proposition 3.2. Let S = C U N be a subelementary semigroup. Every
']{-class of c- 1 S intersects S. The restriction to S of the congruence ']{ on
c-Is is the congruence
x <9 y if and only if ex= dy for some e,d E C
on S, and S/<9 ~ (c-IS)j']{. Moreover no two elements of an <9-c/ass are
comparable in N.
Proof. Let x,y E S. If x ']{yin c-Is, then y = gx for some g = ejd E
c-I C (by Proposition IV.5 .I) and ex = dy holds in S for some e, d E C.
Conversely, ex = dy in s implies y = (ej d) X and X ']{ y in c-Is. Thus (9
is the restriction of ']{ to S and is therefore a congruence on S.
4. THE COMPLETION THEOREM. 149
For every xla E c-Is' X :J{ xla holds in c-Is' since xla = L(x) L(a)-I
and L(a) is a unit of c-Is; thus every :J-C-class of c-Is contains an element
of s. Therefore sI() 9:! (c-IS) I:H; the isomorphism (c-IS) I:H ---+ sI()
sends an :J-C-class of c-IS to its intersection with S, which is an {9-class.
Assume that ex = dy for some c, d E C. If x > y in N, then y = ux for
some u EN, there is a greatest n ~ 0 such that un x # 0, and cunx = duny =
dun+ I x = 0 = cO, a contradiction since c E C is cancellative in S. Hence no
two elements of an {9-class are comparable in N. D
When S = C UN is subelementary, the {9-class Ox of x E S is the orbit
of x. For instance, C and {0} are orbits. The nilmonoid Sl() is the nilmonoid
of orbits of S. By Proposition 3.2, the nilmonoid of orbits of S is that of the
elementary semigroup c-Is.
Proposition 3.3. When S = C U N is finitely generated and subelementary,
then C is finitely generated; c-Is is finitely generated; Sl() is finite; N is
nilpotent; if N # 0, the set M of all minimal elements of N is the union of all
the minimal orbits.
Proof. When S = C UN is f.g., then C is f.g. by Lemma 2.3 and c-Is
is f. g., by the generators of S and the inverses of the generators of C. The
nilmonoid of orbits is f.g. and is finite by Proposition IV.3.5. (Hence N I() is
nilpotent and so is N, as in Lemma 2.3.) If x > y in N, then Ox ~ Oy in
Sl(), in fact Ox> OY, otherwise Ox= OY contains two comparable elements of
N. In particular m is minimal in N if and only if Om is minimal in N I(). D
4. THE COMPLETION THEOREM.
The Completion Theorem, due to Grillet [1975C], states that every finitely
generated commutative semigroup can be embedded into a finitely generated com-
plete semigroup with the same universal semilattice and a certain universal prop-
erty. This completes all the archimedean components at the same time. A simpler
version (without the universal property) is given in Grillet [ 1995].
I. We begin with a simpler result. First, a f.g.c.s. S is ~omplete if and only if
every archimedean component of S contains an idempotent (since S has finitely
many archimedean components and therefore finitely many idempotents).
Proposition 4.1. Every fg.c.s. can be embedded into a complete fg.c.s.
Proof. By Theorem 2.2, a f.g.c.s. S is a subdirect product of a cancellative

semigroup C0 , a nilsemigroup N 1 , and subelementary semigroups Si = Ci U Ni
( i = 2, 3, ... , r ). Then S can be embedded into the direct product T of the
abelian group with zero T0 = G(C0 ) U {0}, the nilmonoid T1 = Nl, and the
finitely many elementary semigroups Ti = c;
1 Si ( i = 2, 3, ... , r ). Since S is
f.g., N 1 , C0 , Si, Ci, and every Ti are f.g. Since all Ti are monoids, every
element t of T can be written as a product t = t 0 t 1 · · · tr with ti E Ti for all
i . Therefore T is finitely generated.
All Ti are elementary. Let Ti = Gi U Nf. Then an archimedean component
of T is a (finite) direct product of subsemigroups Ai ~ Ti, where Ai = Gi
or Ai = Nf for every i: indeed these semigroups are archimedean and consti-
tute a partition of T. Hence every archimedean component of T contains an
idempotent. D
2. Next we construct semigroups of fractions, using the following subsets.
Lemma 4.2. LetS be fg., A E Y(S), and m > 0. Let K =Am and
L be the ideal of S generated by all Bm with B E Y(S) and B ~A. Then
L n A = K. If cp is a homomorphism of S into a complete semigroup T of
index ~ m, and g is the idempotent of the archimedean component C ofT
which contains cp(A), then cp(K) ~ gC = H 9 and cp(L) ~ gT.
Proof. First K =Am~ L n A. Let X E L n A, so that X E A and X= tb
for some t E 8 1 and b E Bm, where B ~ A in Y(S). Since x = tb ~J£ b
we also have A ~ B and B =A. If t E S and C = A(t) is the archimedean
component oft, then A~ c since X= tb ~J{ t, CA ~A, cAm~ Am' and
x = tb E CAm ~ K. If t = 1 E 8 1 , then x =bE Bm = K. Thus L n A= K.
Now let cp, T, g, and C be as in the statement. Then gC = H 9 = em
by Lemma 2.6. Hence cp(K) = cp(A)m ~ em = H 9 • If similarly B E Y(S),
B ~A, cp(B) ~DE Y(T), and his the idempotent of D, then cp(Bm) ~ hD.
But D ~ C, since B ~A, and h ~ g, since E(T) 9:! Y(T) (Corollary 111.3.2);
hence cp(Bm) ~ hD ~ gT. Therefore cp(L) E gT. D
Lemma 4.3. LetS'= (S\L) U K- 1 £, where K and L are as in Lemma

4.2, with multiplication • given for all s,t E S\L, x,y E L, a,b E K by
s.t = st ifst ¢. L, s.t = a(s)a(t) = a(st) = astja ifst E L;
{ s. (xja) = (xja). s = sxja;
(xja). (yjb) = (xja)(yjb) = xyjab.
Then S' is a fg.c.s. The canonical mapping (3 : S ----+ S' defined for all
s E 8\L, x E L, a E K by
f3(s) = s if s E S\L, (3(x) = a(x) =ax/a if x E L
is a homomorphism. Every homomorphism of S into a complete semigroup T
of index ~ m factors uniquely through (3.
S~S'
~lx
T
Proof. In Lemma 4.2, L is an ideal of S which contains K. Hence K- 1 L =

s
K- 1 s: if x/a E K- 1 (with xES and a E K), then xja = axja 2 E K- 1 L
since ax E L. Then the restriction to S\L of the canonical homomorphism
a : S ----+ K- 1 s is a partial homomorphism S\L ----+ K- 1L. We see that
S' is the retract ideal extension of K- 1 L by S / L determined by this partial
homomorphism. Hence S' is a c.s.
K is an ideal of A since K = L n A; hence G(K) ~ G(A) by Proposition
11.2.5 and is finitely generated as an abelian group by Proposition 2.4. Hence
G(K) is finitely generated as a c.s. and S', which is generated by S and by
G(K), is finitely generated.
For all s,t E S\L, x,y E L, a E K we have f3(s) .(3(t) = s. t = f3(st);
f3(s) .(3(x) = s. (ax/a)= asxja = (3(sx); and f3(x) .(3(y) = a(x) .a(y) =
a(xy) = (3(xy), since a is a homomorphism. Thus (3 is a homomorphism.
Let c.p be a homomorphism of S into a complete semigroup T of index~ m,
and g be the idempotent of the archimedean component C of T which contains
c.p(A). By Lemma 4.2, c.p(K) ~ gC = H9 and c.p(L) ~ gT. Hence c.p(a) is
a unit of gT for every a E K. The universal property of K- 1L (Proposition
11.1.3), applied to the restriction L ----+ gT of c.p, begets a unique homomorphism
'lj; : K- 1 L ----+ gT such that 'lj; (a(x)) = c.p(x) for every x E L; namely
'lj;(x/a) = c.p(x) c.p(a)- 1 for all x ELand a E K, where c.p(a)- 1 is the inverse
in H9 . Then the mapping x: S'----+ T defined for all s E S\L, x E L, a E K
by
x(s) = c.p(s), x(xja) = 'lj;(x/a) = c.p(x) c.p(a)- 1

is a homomorphism, and is the only homomorphism such that x o (3 = c.p. 0
Lemma 4.4. When C is an archimedean component of S, the archimedean
component C' of S' which contains /3( C) is as follows: if C t A, then C' = C;

if C ;£A, then C' = (C\L) U K- 1 (C n L). Also Y(/3) : Y(S) ---+ Y(S') is
an isomorphism. If S has index ;£ m, then S' has index ;£ m.
Proof. Let 1r : S ---+ Y (S) and 1r1 : S' ---+ Y (S') be the projections. The
homomorphism {3 : S ---+ S' induces a homomorphism 8 = Y(/3) : Y(S) ---+
Y ( S') such that 8 o 1r = 1r1 o {3; 8 sends an archimedean component C E Y ( S)
of S to the archimedean component C' of S' which contains /3( C).
lfxjaEK- 1 L, then, in K- 1 £, a(a)- 1 }{ 1, xja=a(x)a( a)- 1 }{ a(x),

and xja N a(x). Therefore every archimedean component of S' contains an
element of {3(8) and 8 is surjective. To prove that 8 is injective we use the
universal property of S' in Lemma 4.3: since Y(S) is a complete semigroup
of index 1 ;£ m, there is a unique homomorphism x : S' ---+ Y(S) such that
1r = x o /3; namely,
x(s) = 1r(s), x(xja) = 1r(x) 1r(a)- 1 = 1r(x) 1\ A
for all s E S\L, x E L, a E K. Ifnow s,t E Sand /3(s) N {3(t) inS', then
x(/3(s)) N x(/3(t)), which in the semilattice Y(S) implies 1r(s) = x(/3(s)) =
x(/3(t)) = 1r(t) and s N tinS; hence 8 is injective. Thus Y(/3) = 8: Y(S)---+
Y ( S') is an isomorphism.
In particulars N tinS if and only if {3(s) N {3(t) inS'. Since xja N {3(x)
for all xja E K- 1 L, it follows that xja N yjb in S' if and only if x N y in
S. Similarly xja N sinS' if and only if x N sinS, for all s E S\L. When
C E Y(S), the archimedean component C' of S' which contains /3( C) is now
as follows. If C t A, then C n L = 0, since L ~ UBEY(S), B~A B, and
C' =C. If C ;£A, then, for all s E S\L and xja E K- 1 L, sEC' if and only
if sEC, and xja E C' if and only if x E C; hence C' = (C\L) uK- 1 (CnL).
Finally assume that S has index ;£ m, so that em is cancellative for every
C E Y(S). If C t
A, then (C')m =em is cancellative. Now let C ;£A.
Then (C')m = K- 1 cm, since em~ L. Also AC ~ C and ACm ~em. If
(xja). (zjc) = (yjb). (zjc), where x,y,z E em and a,b,c E K ~A, then
xzjac = yzjbc, bcdxz = acdyz for some a E K, bx = ay since bx, ay,
cdz E ACm ~ em, and xja = yjb. Thus (C')m is again cancellative. D
Lemma 4.5. Let S have index ~ m. Let n be the number of archimedean
components of S. For every 0 ~ k ~ n there exists a homomorphism 1: S--+ T
with the following properties:
( 1) T is finitely generated and has index ~ m;
(2) Y(1) : Y(S) --+ Y(T) is an isomorphism;

(3) every homomorphism of S into a complete semigroup of index < m
factors uniquely through 1;
(4) T has at least k idempotents.
Proof. The proof is by induction on k. If k = 0 we can let T = S and
1 be the identity on S. Now let k < n and 1 : S --+ T have properties ( 1)
through (4 ). If T does not have at least k + 1 idempotents, then there is an
archimedean component A of T which contains no idempotent. Construct K
and L from T as in Lemma 4.2 and (3: T--+ T' as in Lemma 4.3. By Lemmas
4.3 and 4.4, 1' = (3 o 1 : S --+ T' has properties (1), (2), and (3). Also there
is for every archimedean component C' ofT' a unique archimedean component
C of T such that (3( C) ~ C'. If C contains an idempotent, then so does C'.
In addition the archimedean component A' which contains (3(A) contains the
identity element of K- 1 L. Hence T' has at least k + 1 idempotents. D
3. We can now prove:
Theorem 4.6. Let S be a finitely generated commutative semigroup of index
~ m. If m > 0 is large enough, there exists a homomorphism 1 : S --+ T
of S into a finitely generated complete semigroup of index ~ m, such that
Y(r) : Y(S) --+ Y(T) is an isomorphism, and every homomorphism of S into
a complete semigroup of index ~ m factors uniquely through I·
Proof. With k = n, Lemma 4.5 yields a homomorphism 1 : S --+ T
such that T is f.g., has index ~ m, has n archimedean components by (2),
has the universal property (3), and has at least n idempotents. Then T has n
idempotents, every archimedean component of T contains an idempotent, and T
is complete.
By Proposition 4.1 there is an embedding <p : S --+ C of S into a complete
f.g.c.s. C. Now C has finite archimedean cancellativity index, by Proposition
2. 7. If m is sufficiently large (if C has index ~ m ), then <p factors through 1
by (3), and 1 is injective. D
154 VI. FINITELY GENERATED SEMIGROUP S.
4. It would be nice if there would exist a universal complete semigroup of

S; that is, a homomorphism 'Y : S --t U of S into a complete f.g.c.s. U, with
the stronger universal property that every homomorphism of S into a complete
semigroup factors uniquely through 'Y. Alas the following example shows that
this does not exist in general, even when S is archimedean.
Example 4.7. This is Example 111.5.4: S is an infinite cyclic semigroup S =
{a, a 2 , ... , an, ... } . Then S is archimedean and has index 1. We saw that there
does not exist a homomorphism 'Y : S --t T of S into a complete archimedean
semigroup T, with the stronger universal property that every homomorphism of
S into a complete archimedean semigroup factors uniquely through 'Y.
Let m > 0. The subsets in Lemma 4.2 are K = L = sm = {am,
I
am+l, ... }. Then K- 1L is an infinite cyclic group G = { bn n E Z} ~ Z,
with a(an) = bn for all n ~ m. Hence the semigroup S' in Lemma 4.3 is
I
S' = (S\K) U G = {a, a 2 , ... , am- 1 } U { bn n E Z} with multiplication
ai ai = ai+i if i + j < m, ai ai = ai+j = bi+j if i + j ~ m, ai hi = bi+j;
this is the semigroup Tm in Example III.5.4. We see that S' has index m.
Also j3(ai) = ai if i < m, j3(ai) = bi if i ~ m. By Lemma 4.3, every
homomorphism of S into a complete semigroup of index ~ m factors uniquely
through j3 : S --t S'. Moreover S' coincides with the complete semigroup
constructed in the proof of Theorem 4.6.
Assume that there is a homomorphism 'Y : S --t U of S into a complete
semigroup U, with the stronger universal property that every homomorphism of
S into a complete semigroup factors uniquely through 'Y. As in Example 111.5.4,
'Y(S) is archimedean and is contained in an archimedean component T of U,
which contains an idempotent e. Since T is archimedean we have 'Y( ak) = et for
some k > 0 and t E T. Let m > k. If j3 : S --t S' factors through 'Y (if j3 = c.p o 'Y
for some homomorphism c.p : U --t S' ), then c.p (eT) is a subgroup of S', since
eT is a subgroup ofT, c.p(eT) ~ G, and j3(ak) = c.p('Y(ak)) = c.p(et) E G,
which does not happen when m > k. D
5. IRREDUCIBLE SEMIGROUPS.
In this section we determine all finitely generated commutative semigroups

that are finitely subdirectly irreducible. The results are from Grillet [ 1977] and
imply Malcev' s Theorem [ 1958].
5. IRREDUCIBLE 8EMIGROUPS. 155
I. Let S be a f.g.c.s. If S is irreducible, then S is either cancellative or nil

or subelementary, by Theorem 2.2.
If S is cancellative, then, by Proposition IV.l.8, S is irreducible if and only if
S is either a nontrivial cyclic p-group, or isomorphic to Z or to a subsemigroup
of N.
If S is a nilsemigroup, then S is finite; by Propositions IV.3 .6 and V.S .1,
S is irreducible if and only if S is subdirectly irreducible, if and only if S is
weakly irreducible and has just one minimal element.
2. We now let S = C U N be subelementary. The cases where C or N are
trivial are covered by Proposition V.5.2: C U {0} is irreducible if and only if C
is irreducible; N U {1} is irreducible if and only if N is irreducible; and {0, 1}
is irreducible. In general:
Proposition 5.1. When S = C UN is subelementary, then S is (finitely)
subdirectly irreducible if and only if c-IS is (finitely) subdirectly irreducible.
Proof. First we show that the equality is the only congruence e on c-IS
which is the equality on S. When indeed s,t E c-Is, then cs,ct E S for some
c E C; hence set implies cs e ct, cs = ct, and s = t.
Now assume that S is irreducible. If the equality on c-IS is the intersection
of finitely many congruences ei, then the equality on S is the intersection of the
finitely many congruences on S induced by the congruences ei; therefore some
ei is the equality on S, and then ei is the equality on c-IS. Thus c-IS is
irreducible. If similarly S is subdirectly irreducible, then c-IS is subdirectly
irreducible.
To prove the converse we construct for every congruence e on S a congruence
e* on c-Is as follows:
s e* t if and only if cs,ct E S and cs e ct for some c E CI.
It is immediate that e* is a congruence. Also e* is the equality on c-IS if and
only if e is the equality on S. Moreover let e = niEI ei. Then e ~ ei and
e* ~ e; for all i' so that e* ~ niEJ e; . Conversely assume that s e; t for all
i. For every i E I we have cis, c/ E S and cis ei cit for some ci E ci. If
I is finite, then c = IJiEI ci E CI, cs, ct E S, cs ei ct for all i, and s e* t.
Thus (niEJ ei) * = niEJ e;' provided that I is finite.
Now assume that c-IS is irreducible. If the equality on S is the intersection
of finitely many congruences ei' then by the above the equality on c-Is is the
intersection of the finitely many congruences e;,

some e;is the equality on
c- 1 S, and ;orne ei is the equality on c- 1 S. Thus S is irreducible.
Finally assume that 1 c- s
is subdirectly irreducible. By Theorem V.5.6 and
Proposition IV.l.7, G(C) is either trivial or a cyclic or quasicyclic p-group.
Therefore the subsemigroups of G (C) are subgroups of G (C) , C = G (C) , and
s is elementary; hence s = c- 1 s is subdirectly irreducible. 0
Lemma 5.2. When S = C U N is subelementary, then N is weakly
irreducible if and only if c- 1 N is weakly irreducible.
Proof. Let M be the set of all minimal elements of N. By Proposition 3.3,
X I a E c- 1 is minimal in c- 1 N if and only if X is minimal in N; thus
s
the set of all minimal elements of c- 1 N is c- 1 M. Recall that N is weakly
irreducible if and only if its congruence
x Jv( y ~ (VuE N 1 )(ux EM ~ uy EM =:::::;. ux = uy)
is the equality. For any e E C, we have x E M if and only if ex E M; hence
x Jv( y implies ex Jv( ey.
We show that xla Jv( ylb in c- 1 N if and only if bx Jv( ay in N. Indeed
assume xI a Jv( y lb and let u E N 1 . If ubx E M, then ux I a= ubx I ab E c- 1M,
uylb = uxla E c- 1 M since xla Jv( ylb, and uay = ubx. Similarly uay EM
implies ubx = uay. Thus bx Jv( ay. Conversely assume bx Jv( ay and let
u EN and e E C. If (ule)(xla) E c- 1 M, then ubxlabe = uxlae E c- 1 M,
ubx EM, uay = ubx since bx Jv( ay, and (ule)(ylb) = uaylabe = ubxlabe =
(ule)(xla). Similarly (ule)(ylb) E c- 1M implies (ule)(xla) = (ule)(ylb);
X I a E c- 1M implies bx I ab = X I a E c- 1M' bx E M' ay = bx' and
ylb = xla; and ylb E c- 1M implies xla = ylb. Thus xla Jv( ylb.
If now N is weakly irreducible, so that Jv( is the equality on N, then
xla Jv( ylb in c-
1N implies bx Jv( ay in N, bx = ay, and xla = ylb; hence
c- 1 N is weakly irreducible. If conversely c- 1N is weakly irreducible, then

X Jv( y in N implies, for any e E c' ex Jv( ey in N' xle Jv( yle in c-
1N'
xle = yle, and x = y; thus N is weakly irreducible. 0
3. We call a subelementary semigroup S = C UN homogeneous when
c- 1 s is homogeneous; equivalently, if ex = dx ~ 0 implies e = d when
e, d E C. Combining Proposition 5.1 and results from Section V.5 (Propositions
V.5.1 and V.5.2, Theorem V.5.6) we obtain:
Theorem 5.3. A finitely generated commutative semigroup is finitely subdi-
6. ARCHIMEDEAN SEMIGROUPS. 157
rectly irreducible if and only if it is one of the following:

a nontrivial cyclic p-group; an infinite cyclic group; isomorphic to a sub-
semigroup of N;
a finite nilsemigroup on which Pis the equality (equivalently, which is weakly
irreducible with one minimal element);
{0, 1}; a finitely subdirectly irreducible cancellative semigroup with a zero
element adjoined; a finite nilmonoid on which P is the equality; a homoge-
neous subelementary semigroup S = C U N in which C is finitely subdirectly
irreducible, N is weakly irreducible, and the minimal elements of N form an
orbit.
Proof. This follows from Propositions V.5.1 and V.5.2 except in the case where
S = C U N is subelementary with C and N nontrivial. Then S is irreducible
if and only if c- 1 S is irreducible (Proposition 5.1 ); if and only if G( C) is irre-
ducible, c- 1 N is weakly irreducible, the minimal elements of c- 1 N form an
orbit, and c- 1 s is homogeneous (Theorem V.5.6); if and only if Cis irreducible
(Proposition IV.1.8), the minimal elements of N form an orbit (Proposition 3.3),
N is weakly irreducible (Lemma 5.2), and S is homogeneous. D
4. From Theorem 5.3 we can deduce Malcev's Theorem that f.g.c.s. that
are subdirectly irreducible are finite (Malcev [ 1958]); other proofs were given by
Carlisle [ 1971] and Lallement [ 1971]. Since Z and subsemigroups of N are not
subdirectly irreducible, a cancellative f.g.c.s. is subdirectly irreducible if and only
if it is a nontrivial cyclic p-group. Hence a f.g.c.s. is subdirectly irreducible if
and only if it is one of the following:
a nontrivial cyclic p-group;
a finite nilsemigroup on which P is the equality (equivalently, which is weakly
irreducible with one minimal element);
{0, 1}; a nontrivial cyclic p-group with a zero element adjoined; a finite
nilmonoid on which P is the equality; a homogeneous elementary semigroup
S = C UN in which C is a nontrivial cyclic p-group, N is weakly irreducible,
and the minimal elements of N form an orbit.
All these semigroups are finite; in the last case S is elementary with finitely
many finite orbits. Thus:
Corollary 5.4 (Malcev). Every finitely generated c.s. which is subdirectly

irreducible is finite.
6. ARCHIMEDEAN SEMIGROUPS.
This section studies finitely generated archimedean semigroups. Most of the

results are from MacAlister & O'Carroll [1970].
1. By Proposition 2.4, a finitely generated archimedean semigroup S has
a cancellative power sm and its universal group G(S) ~ G(Sm) is finitely
generated.
Proposition 6.1. A complete archimedean semigroup S is jg. if and only if
its kernel K is jg. and S I K is finite.
Proof. If S is f. g., then the f. g. nilsemigroup S I K is finite and K = G (K) 9:!
G(S) is f.g. (as an abelian group). If conversely K is f.g. as an abelian group,
then K is f.g. as a c.s. (by the generators of K and their inverses); if moreover
Sl K is finite, then S is f.g. (by S\K and the generators of K). D
2. We now assume that S has no idempotent. Proposition II.5.5 provides a
smallest power cancellative congruence 'J on S, for which
If S is archimedean without idempotent, then T = S I'J is archimedean without
idempotent and cancellative (Proposition III.4 .1 ). Also x =f. xy for all x, y E S
(Corollary III.4.2).
Proposition 6.2. A finitely generated archimedean semigroup without idem-
potent is power joined
Proof. In the above T = S I'J is f. g. archimedean without idempotent, power
cancellative, and cancellative. Since T is archimedean, T has only one face (T
itself) and has dimension 1 by Proposition 11.6.7. Thus G(T) has rank 1 and S
is power joined by Proposition 111.6.3. D
Proposition 6.2 was proved by Tamura [1968], Levin [1968] and, in case S
is cancellative, by Petrich [ 1964]. Chris lock [ 1969] showed that an N-semigroup
is f.g. if and only if its structure group is finite. F.g. N-semigroups were also
investigated by Higgins [1966]. Higgins & Tamura [1973] showed that anN-
semigroup S is f.g. if and only if G(S) is f.g. of rank 1.
Corollary 6.3. When Sis ajg. archimedean semigroup without idempotent,
then S I'J is isomorphic to a subsemigroup of N+.
Proof. By Corollary 111.6.4, S I'J is isomorphic to a subsemigroup Q of Q+ .
But Q is f. g. like S; if ptf q1 , p 2 1q2 , ... , Pnl qn generate Q, then every
element of Q is an integer multiple of 11 q1 .. · qn; hence Q is isomorphic to a
6. ARCHIMEDEAN SEMIGROUPS. 159
subsemigroup of N+ . 0
3. To sharpen Corollary 6.3 we use McAlister & O'Carroll's congruence Ma,
defined for any element a of S by
x May if and only if x = y, or x =au, y = av for some u,v E S 1 such
that aPu = aqv for some p,q > 0
The proof of Lemma 6.4 below shows that Ma is the smallest congruence e on
s such that a e a2 .
Lemma 6.4. Ma is a congruence on S; if S is archimedean without
idempotent, sm
is cancellative, and a E sm,
then Ma n 'J is the equality
on S.
Proof. If X= au, y = av, where u, v E S 1 and aPu = aqv, and y = aw,
z =at, where w,t E 8 1 and arw = a 8 t, then av = aw and aP+ru = aq+rv =
aq+r w = aq+sw. Hence Ma is transitive; then Ma is clearly a congruence.
Now let a E sm' where sm is cancellative. Assume that X Ma y' with
X= au, y = av, where U,V E 8 1 and aPu = aqv, and that X 'J y, SO that
xn = yn. We may arrange that n ~ m. Then a nun = anvn and anp un =
anqvn = anqun. Since x -=f xy for all x,y E S we have np = nq and p = q.
Then aPu = aqv implies x =au= av = y, since either p = q = 1 or a, aP~ 1 u,
v sm
aP- 1 E and sm
is cancellative. Thus Ma n 'J is the equality on S. 0
Lemma 6.5. When S is afg. archimedean without idempotent, then S/Ma
is finite.
Proof. M = S /Ma is f. g. and power joined like S, and has an idempotent
since a Ma a 2 . Since M is power joined, every element of M has an idempotent
power and has only finitely many distinct powers. Then the elements of M are
products of powers of finitely many generators and M is finite. 0
Now sm is cancellative for some m and 'J n Ma is the equality on S when
a E sm by Lemma 6.4; hence S is a subdirect product of Sj'J, which by
Proposition 6.2 is isomorphic to a subsemigroup of N+, and S /Ma, which by
Lemma 6.5 is finite and archimedean. This proves:
Proposition 6.6. A fg. archimedean semigroup without idempotent is a
subdirect product of a finite archimedean semigroup and a subsemigroup of w+.
If conversely F is a finite archimedean semigroup, then F and F x w+
are power joined, and every subsemigroup of F x w+ is power joined, and is
archimedean without idempotent. Higgins [ 1966], [ 1969S] proved Proposition
6.6 in the case of N-semigroups (in which case S is f.g. if and only if it is a
subdirect product of a finite abelian group and a subsemigroup of N+ ).
Proposition 6. 7. When m, n > 0,
S = (a, b I am = bn )
is an N-semigroup with two generators, and, up to isomorphism, every N-semi-
group with two generators can be constructed in this fashion.
Proof. Let F be the free c.s. on {a, b}. The congruence e
on F generated
by (am, bn) consists of all pairs (ai+km lJ, ai lJ+kn) and (ai lJ+kn, ai+km EJ)
with i,j, k ~ 0, i + j > 0, and coincides with the congruence induced by the
subgroup (am b-n) of G(F). Hence S = (a, b I am = bn ) is cancellative
by Proposition 11.5.1. Also S has no idempotent, since e contains no pair
(ai lJ, a 2i b2i). Finally (ai EJ)n = ani+mj; therefore S is power joined, and is
an N-semigroup.
Conversely let S be an N-semigroup with two generators a and b. Then S is
power joined by Proposition 6.2 and ar = b8 for some r, s > 0. By Proposition
III.6.6, S is a subdirect product of S j'J and a torsion group G, namely its
structure group G = G(S)/A, where A ~ Z is the subgroup generated by a.
Now G is generated by b + A and is a finite cyclic group. Let m, n be the least
r,s > 0 such that ar = b8 • Then am= bn and G has order n. If ai lJ = ak bl,
with, say, j ;;::; l, then l = j + tn for some t ~ 0, since G has order n, and
then ai lJ = ak EJ+tn = ak+tm lJ and i = k + tm, since S is cancellative and
a has infinite order. Thus S ~ ( a, b I am = bn ) . 0
Proposition 6.7 is due to Petrich [1964]; the above proof is from McAlister
& O'Carroll [1970]. More generally, Hall [1971] constructed all cancellative c.s.
with two generators.
7. THE A.C.C. ON SUBSEMIGROUPS.
This section looks at commutative semigroups whose subsemigroups satisfY

the ascending chain condition. These semigroups were studied by MacAlister &
O'Carroll [ 1970], Shevrin & Ovsyannikov [ 1990], and Rosales & Garcia-Gar-
cfa[l999].
I. We call a c.s. S strongly noetherian when its subsemigroups satisfY the
ascending chain condition; equivalently, when every subsemigroup of S is finitely
generated. A strongly noetherian c.s. is f.g.; hence its ideals and congruences also
7. THE A.c.C. ON 8UBSEMIGROUPS. 161
satisfY the ascending chain condition. Rosales & Garda-Garda[ 1999] call these
semigroups hereditarily finitely generated.
Example 2.5 shows a f.g.c.s. is not necessarily strongly noetherian. So does
Example 1.5, which will be used below, in which S is the free c.s. on { x, y} but
contains a subsemigroup { xm yn I m, n > 0} which is not f. g.
The following results are from MacA lister & O'Carroll [ 1970].
Proposition 7.1. A fg.c.s. is strongly noetherian if and only if its archime-
dean components are strongly noetherian.
Proof. Let S be a f.g.c.s. If S is strongly noetherian, then every archi-
medean component of S is strongly noetherian. Conversely assume that every
archimedean component of S is strongly noetherian. Let T be a subsemigroup
of S. Since S is f.g., Y(S) is finite and T is the union of the finitely many
subsemigroups T n A, A E Y (S) . Every T n A S.: A is f. g.; therefore T is f. g. D
Proposition 7.2. Every fg. archimedean semigroup without idempotent is
strongly noetherian.
Proof. Let S be a f.g. archimedean semigroup without idempotent and T be
a nonempty subsemigroup of S. By Proposition 6.2, S is power joined and so
is T. Also sm is cancellative for some m > 0; since S is archimedean, sm
contains an element a of T.
Since x =/= xy for every x,y E S, we have a E T\Ta. Lett E T. Since T
is power joined we have ak = t 1 for some k, l > 0. Then t ~ Tak, otherwise
ak = ak s for some s E S. Therefore there is a greatest n ~ 0 such that t E Tan
( = T if n = 0). Then t ~ Tan+ I and t = uan for some u E T\Ta. Thus T
is generated by T\Ta.
Let s, t E T\Ta. Assume that s JV(a t, with s = au, y = av, where
u,v E S 1 and aPu = aqv for some p,q > 0. Let r =min (p,q) and p = r + i,
q = r + j, so that i,j ~ 0, one of i, j is 0, and ar+iu = ar+jv. If r = 1, then
ais = ar+iu = ar+jv = ajt (with say ais = s if i = 0) If r > 1 then ar-l
' ' . ' '
ai+ 1u, aj+lv E sm and ar+iu = ar+jv implies ais = ai+ 1u = aj+lv = ajt,
since sm is cancellative. In either case ais = ajt. This contradicts s,t ~ Ta
if only one of i and j is 0. Therefore i = j = 0 and s = t. Thus JV(a is the
equality on T\Ta, and T\Ta is finite by Lemma 6.5. D
3. We now tum to archimedean semi groups with idempotents.
Recall that a f.g. abelian group A is the direct sum A = F EB G of a free
abelian group F and a finite abelian group G. The rank of A is the rank of F.
Thus a f. g. abelian group has rank 1 if and only if it is isomorphic to Z EB G for

some finite abelian group G; a f. g. abelian group has rank at most 1 if and only
if it is either finite or isomorphic to Z EB G for some finite abelian group G.
Lemma 7.3. An abelian group is strongly noetherian (as a commutative
semigroup) if and only if it is finitely generated of rank at most 1.
Proof. An abelian group A which is strongly noetherian (as a c.s.) is f.g. as
a c.s. and f.g. as an abelian group. Hence A is the direct sum A = F EB G of
a free abelian group F and a finite abelian group G. If F has rank 2 or more,
then F has a subsemigroup which is free with two generators, which we saw in
Example 1.5 contains a subsemigroup which is not f.g. Therefore A has rank at
most 1.
Conversely let A be a f.g. abelian group of rank at most 1. If A is finite,
then A is strongly noetherian. Otherwise A ~ Z EB G, where G is finite. Then
N+ x G ~ A is power joined and f.g. (by {1} x G), and is strongly noetherian
by Proposition 7 .2. Then every subsemigroup T of Z EB G is a union
T = (Tn ({0} X G)) u (Tn u~+ X G)) u (Tn (-N+ X G))

in which the first term is finite and the remaining two terms are isomorphic to
subsemigroups of N+ x G and are f.g.; therefore T is f.g. Thus A is strongly
noetherian. D
Proposition 7.4. LetS be an archimedean semigroup S with an idempotent,
and K be the kernel of S. Then S is strongly noetherian if and only if S I K is
finite and K is finitely generated of rank at most 1.
Proof. If S is strongly noetherian, then S I K is finite and K is f. g. by
Proposition 6.1 and has rank at most 1 by Lemma 7.3. Conversely assume that
S I K is finite and K is finitely generated of rank at most 1. Every subsemigroup
T of S is a union
T = (Tn (S\K)) U(TnK)
in which the first term is finite and the second term is a f.g. subsemigroup by
Lemma 7.3; therefore T is f.g. D
4. The next results, due to Rosales & Garcia-Garcia[1999], are global char-
acterizations of strongly noetherian semigroups.
Proposition 7.5. A cancel/alive commutative semigroup S is strongly
noetherian if and only if it finitely generated and G (S) has rank at most 1.
Proof. If S is strongly noetherian, then S is f.g. and has a least archime-
dean component C which is f.g. and an ideal of S. If C does not contain an
7. THE A.C.C. ON SUBSEMIGROUPS. 163
idempotent, then C is power joined by Proposition 6.2 and G( C) is f.g. of rank

1 by Proposition III.6.3. If C contains an idempotent, then G( C), which is
isomorphic to the kernel of C, is f.g. of rank at most 1 by Proposition 7.4. In
either case G (S) ~ G (C) is f. g. of rank at most 1.
If conversely S is f.g. and G(S) has rank at most 1, then G(S), which is
f.g., is strongly noetherian by Lemma 7.3, and so is S ~ G(S). 0
Proposition 7.5 appears as Proposition 3.1 in Rosales & Garcia-Garcia[ 1999],
with a somewhat different statement and a different proof.
The following example shows that Proposition 7.5 does not extend to all
semi groups.
Example 7.6. Let
S = ( x,y,z I xz = yz = z ).
S is the disjoint union S = F U C of F = (x, y), which is free on { x, y}, and
C = (z) ~ N+. S is not strongly noetherian since F, as in Example 1.5, is
not strongly noetherian. But C is an ideal of S and G(S) ~ G(C) ~ Z has
rank 1. 0
Proposition 7.7. For a commutative semigroup S with a finite generating
subset X the following conditions are equivalent:
(1) S is strongly noetherian;
( 2) the universal group of (a, b) ~ S has rank at most 1 for every a, b E S;
(3) the universal group of (x, y) ~ S has rank at most 1 for every x, y E X;
(4) for every x =f. y in X a relation xP yq = xr y with q > 1 holds in S.
This is Theorem 1.8 of Rosales & Garcia-Garcia[1999], to which the reader
is referred for a proof. Condition (3) holds if and only if x, y satisfy a nontrivial
relation xP yq = xr y 8 • In a strongly noetherian semi group one may assume that
this relation has the form xP yq = xr y with q > 1: indeed the subsemigroup
T = {xi~ I i ~ O,j > 0}

of S must be finitely generated; then some xr y is not a generator and is a product
of two or more generators.
Chapter VII.
SUBCOMPLETE SEMIGROUPS.
Subcomplete semigroups are a natural generalization of finitely generated

commutative semigroups. This chapter contains their main properties. Most
notable are Ponizovsky families, with transport a number of properties of finite
semigroups to all subcomplete semigroups. This brings additional properties of
finitely generated commutative semigroups and reveals a deep similarity to their
finite relatives.
1. COMPLETIONS.
In this section we define completions and subcomplete semigroups, and prove

their first properties. The results are from Grillet [200 1S).
1. A completion of a commutative semigroup S is a complete semigroup
T of which S is a subsemigroup such that every archimedean component of T
contains an element of S; a sharp completion of S is a complete semigroup
T of which S is a subsemigroup such that every archimedean component of T
contains a unique archimedean component of S.
The embedding of a cancellative semigroup into its group of fractions shows
that sharp completions are not necessarily the "best" completions.
Other examples of completions include the embedding of a cancellative semi-
group into its group of fractions; the embedding of an archimedean semigroup with
a nonempty cancellative ideal into a complete archimedean semigroup (Proposi-
tion III.4.4); the embedding of a separative semigroup into a semilattice of abelian
groups (Proposition IV.2.1 ); the embedding of a subelementary semigroup into an
elementary semigroup (Proposition Vl.2.1 ). The Completion Theorem VI.3 .6 im-
plies that every f.g.c.s. has a sharp completion.
Most of these examples have very nice universal properties. However, Exam-
ple Vl.4.7 shows that a c.s. Sneed not have a minimal completion T in the sense
that every embedding of S into a complete semigroup extends uniquely to T.
165
166 VII. SUBCOMPLETE SEMIGROUPS.
Completions are related to Shevrin's problem 3.58 [1989] of embedding into

epigroups. Recall that a c.s. S is an epigroup when every archimedean com-
ponent of S contains an idempotent. Thus, complete semigroups are epigroups;
conversely, a commutative epigroup S is complete if and only if there also exists
for every a E S a least idempotent e of 8 1 such that ea = a. For instance,
commutative epigroups with finitely many idempotents are complete.
Complete semigroups have properties, such as Theorems IV.4.3 and V.3.4,
which are not shared by all epigroups. Hence the author feels that completions
are more valuable than embeddings into epigroups.
2. A subcomplete semigroup is a commutative semigroup which has a com-
pletion. For instance, an archimedean semigroup is subcomplete if and only if it
is subcomplete as defined in Section III.4. Example 111.4.3 shows that not every
c.s. is subcomplete. By Theorem VI.3.6,
Proposition 1.1. Every finitely generated c.s. has a finitely generated sharp
completion and is subcomplete.
Every archimedean component of a subcomplete semigroup is contained in
a complete archimedean semigroup and has a nonempty cancellative ideal by
Proposition III.4.1. Corollary IV.4.6 provides a stronger necessary condition:
Proposition 1.2. A c.s. which can be embedded into a complete semigroup
is a subdirect product of nilsemigroups, cancellative c.s., and subelementary
semigroups.
Proof. By Corollary IV.4.6 a complete semigroup T is a subdirect product
of a nilsemigroup T 1 , a group T 0 , and elementary semi groups Ti = Gi U Ni .
A subsemigroup S of T is a subdirect product of subsemigroups Si of the
semigroups Ti; S 1 ~ T1 is a nilsemigroup, S 0 ~ T 0 is cancellative, and Si ~
Gi U Ni is a nilsemigroup if Si ~ Ni, cancellative if Si ~ Gi, or subelementary
if Si intersects both G i and Ni . D
The converse of Proposition 1.2 does not hold:
Example 1.3. Let S be Example 111.4.3, which consists of all positive powers
o s 0 , s 1 , ... , sp, ... ,WI"th sp-
f n _ sqn whenever n > p,q an d sp
m sqn -_ srm+n ,
where r = max (p,q). This nefarious semigroup has no nonempty cancellative

ideal and cannot be embedded into a complete semigroup. Nevertheless S is a
subdirect product of nilsemigroups. Indeed let
Ik = { s; I p ~ k or n~ k}.
Then I k is an ideal of S and S /Ik is a nilsemigroup since xk E I k for all
1. COMPLETIONS. 167
xES. Also nk~O Ik = 0, since s~ tt. Ik when k > n,p. Hence the equality
on S is the intersection of the Rees congruences of the ideals Ik, and S is a
subdirect product of the nilsemigroups S /Ik. 0
3. Nevertheless Proposition 1.2 has a partial converse, which is similar to
Proposition VI.3.1 but more general:
Proposition 1.4. For a c.s. S the following conditions are equivalent:
(1) S can be embedded into a complete semigroup with finitely many idem-
patents;
(2) S has a completion withfinitely many idempotents;
(3) S is a subdirect product offinitely many nilsemigroups, cancellative semi-
groups, and subelementary semigroups.
Proof. (1) ===::::;. (2). Let S be a subsemigroup of a complete semigroup
T with finitely many idempotents. Let U be the union of all the archimedean
components of T which intersect S. U is a subsemigroup of T since N is a
congruence. Every archimedean component of U is an archimedean component
of T and contains an idempotent. Also U, like T, has finitely many idempotents.
Hence U is complete.
(2) ===::::;. (3). By Theorem IV.4.3 and Proposition IV.4.5, a completion T
of S which has finitely many idempotents is a subdirect product of a group, a
nilsemigroup, and finitely many elementary semigroups. Then, as in the proof
of Proposition 1.2, S is a subdirect product of finitely many nilsemigroups, can-
cellative semigroups, and subelementary semigroups.
(3) LetS be a subdirect product of finitely many nilsemigroups Nj,
====? (1).
cancellative semi groups C k, and subelementary semigroups S 1 = C1 U N 1 . Then
S can be embedded into the direct product T of the finitely many nilsemigroups
Tj = Nj = Nj, abelian groups Tk = G k = G (Ck), and elementary semi groups
Tz = C 1- 1 S 1 = G1 UN{. An archimedean component of T is a (finite) direct
product of subseniigroups Ai ~ Ti, where Ai = G i or Ai = NI,
for every i:
indeed these semigroups are archimedean and constitute a partition of T. Hence
every archimedean component of T contains an idempotent. Also T has finitely
many idempotents; hence T is complete. 0
Proposition 1.4 requires some finiteness condition, since, in general, a sub-
semigroup of a complete semigroup T need not have a completion U ~ T:
Example 1.5. Let T be the commutative monoid
T = { e 1 , e 2 , ... , en, ... , e, a, f}
in which e 1 ~ e 2 ~ · · · ~ en ~ · · · ~ e ~ f are idempotents and
ena = ea = a, a 2 = af = f.
Associativity is readily verified. The archimedean components of T are {a, f},
{e}, and every {en}; all contain idempotents. Also there is for every x E T
a least idempotent g of T such that gx = x: if x is idempotent then g = x,
otherwise x = a and g = e. Thus T is complete. Next let
S = {e 1 ,e2 , ... ,en, ... ,a,f} = T\{e}.
Then S is a submonoid ofT. However, S (seen before as Example IV.4.2) is not
complete, since there is no least idempotent g of T such that ga = a. Moreover
T is not a completion of S, since the archimedean component {e} of T does
not intersect S. Thus S has no completion S <;;;; U <;;;; T. 0
4. When S has an identity element, a monoid completion of S is a com-
pletion T of S in which the identity element of S is also the identity element
ofT.
Proposition 1.6. A c.s. S has a completion if and only if S 1 has a monoid
completion. Every fg.c.m. has a sharp fg. monoid completion.
Proof. Assume that S does not have an identity element. If T is a completion
of S, then T U { 1} (with an identity element adjoined even if T already has one)
is a monoid completion of 8 1 , which is f.g. if T is f.g. Conversely let T be a
monoid completion of 8 1 . If the group of units U of T does not intersect S (if
U n S 1 = {1} ), then T\U is a completion of S; otherwise T is a completion
of S. Thus S has a completion if and only if 8 1 has a monoid completion.
Now let S have an identity element u. Let T be a completion of S. Then
S uT. We show that the c.m. uT is a monoid completion of S.
<;;;;
First, the idempotents of uT coincide with the idempotents e ~ u ofT. If

x E uT and e is the least idempotent of T 1 such that ex = x, then e ~ u and
e is the least idempotent of uT such that ex = x.
If a, b E uT and a ~N b in T, then an = bt for some n > 0 and t E T,
an= anu = b(tu) with tu E uT, and a ~N b in uT. Conversely a ~N bin
uT implies a ~N b in T. Hence the archimedean components of uT are the
intersections of uT and the archimedean components of T. In particular every
archimedean component of uT contains an element of S. For every x E uT
there is an idempotent e of T such that x N e in T; then x = ux A ue in
T and x N ue in uT. Thus every archimedean component of uT contains an
idempotent, and uT is a monoid completion of S.
2. PONIZOVSKY FAMILIES. 169
If S is f.g., then we can arrange that T has finitely many generators t 1 , · · · , tn

and that every archimedean component of T contains one and only one archime-
dean component of S (Proposition 1.1 ). Then uT is f. g., by u and ut 1 , · · · , utn,
and every archimedean component of uT contains one and only one archimedean
component of S, so that uT is a sharp monoid completion of S. D
2. PONIZOVSKY FAMILIES.
Ponizovsky families are a basic property of finitely generated commutative

semigroups and, more generally, of subcomplete semigroups. The results in this
section are from Grillet [200 1S].
1. When S has a completion T, the congruence N on T induces a congru-
ence ~ on S; by definition, a ~ b in S if and only if a N b in T.
Proposition 2.1. When T is a completion of S, then ~ is a semilattice
congruence on S, in particular N ~ ~. and S/~ ~ Y(T). If T is a sharp
completion of S, then ~=Nand Y(S) ~ Y(T).
Proof. ~ is the congruence induced by the homomorphism S <:;;; >T ----*
Y(T), which is surjective since the archimedean components ofT all intersect
S. Hence S /~ ~ Y (T). The isomorphism sends an ~-class to the archimedean
component of T that contains it. In particular ~ is a semilattice congruence and
contains the least semilattice congruence N on S.
Let T be a sharp completion of S, so that an archimedean component of
T contains exactly one archimedean component of S. When a, b E S, then
a N b in T if and only if a N b in S; hence ~ = N in S and Y (S) =
S/~ ~ Y (T) ; the isomorphism sends an archimedean component of S to the
archimedean component of T that contains it. D
If T is a sharp completion of S, then S/~ = Y(S). In general, S/~ ~ Y(T)
is a homomorphic image of Y ( S) = S/N, since N ~ ~ . Proposition 1.1 implies
an isomorphism S/~ ----* E(T) which sends A E Sj~ to the idempotent of
the archimedean component of T which contains A. We denote the ~-class of
xES by B(x).
2. When S has a completion T, the partial Ponizovsky factors of T induce
another interesting partition of S. For technical reasons we assume that S is a
monoid and that T is a monoid completion of S; by Proposition 1.6, this entails
no loss of generality.
Proposition 2.2. LetS be a c.m. and T be a monoid completion of S. For
every 'B-class A of S, let PA = S n ~CA)• where e(A) is the idempotent of the

archimedean component of T which contains A. Then
(Pl) P = (PA)AES/:B is a partition of S
and, for every 'B-class A:
(P2) LA= U (PB I BE Sj'B, B ~A) is an ideal of S;
(P3) PA ~ U (B E S/'B IB ~ A) and KA = PAn A i= 0;
(P4) if A~ B in Sj'B, a E LA, and k E KA, then ka E PB if and only if

a E PB;
(P5) KA is cancellative in LA (if k E KA, a,b E LA, and ka = kb, then
a= b).
(P6) (\Ia E A)(::!n > 0) an E KA.
Moreover KA = S n He and LA = S neT, where e = e(A) as above.
Proof. (PI) By Proposition IV.4.4, the partial Ponizovsky factors ofT con-
stitute a partition (~*)eEE(T) ofT. Since A~ e(A) is bijective, the sets PA
are pairwise disjoint and cover S. Let A E Sj'B and e = e(A).
The archimedean component of e in T contains some s E A, and then
sn E He in T for some n > 0, by Proposition VI.1.2, and sn E HenS ~ ~* n S.
Therefore PA i= 0.
(P2)InT,
U (Pj I f E E(T), f ~ e) U(JT I f E E(T), f ~ e) eT
is an ideal; therefore
LA U(PB IB E Sj'B, B ~ A) SneT

is an ideal of S.
(P3) Let C be the archimedean component ofT which contains A and e.
Then A = S n C and
KA = PA n A = S n ~* n C = S n He .
The proof of (PI) shows that S n He i= 0. When x E S n ~* and k E S n He,
then x ~N k in T and B(x) ~ B(k) =A in Sj'B; hence PA is contained in
u(B E Sj'B I B ~A).
(P4) Let A~ Bin Sj'B, a E LA, and k E KA. Then e = e(A) ~ e(B) = J,
a E eT, and k E He has a group inverse l E He in T. If ka E PB inS, then
2. PONIZOVSKY FAMILIES. 171
ka E Pj and E(ka) = f; hence fka = ka, fa= fea = fkla = kla = ea =a,
and ga = a implies gka = ka and g ~ f, so that f = E( a) and a E Fj* n S = PB.
If conversely a E PB, then E(a) = f; hence fa= a, fka = ka, and gka = ka
implies gea = gkla = kla =a and g ~ ge ~ f, so that E(ka) = f and ka E PB.
(PS) Let e = e(A). If k E KA, a,b E LA~ eT, and ka = kb, then, in T,
k E He, k has a group inverse l E He, and ka = kb implies a = kla = klb = b.
(P6) When a E A, then an E He in T for some n > 0, by Proposition
VI.l.2, and an E S n He = KA. 0
If T is a sharp completion of S, then every 13-class A is archimedean and
(P6) follows from (Pl) through (PS). Indeed let a E A. Since KA # 0 by (P3)
there exists k E KA. Then k E A, an = kt for some n > 0 and t E A, and
an E KA since KA is an ideal of A by Lemma 2.3 below.
A Ponizovsky family of a commutative monoid S, relative to a semi lattice
congruence 'B on S, is a family P = (PA)AES/'B with properties (PI) through
(P6) in Proposition 2.2; if 'B = N, then P is a sharp Ponizovsky family. When
T is a monoid completion of S, the family (~(A) n S)AES/'B in Proposition
2.2 is the Ponizovsky family of S induced by T.
A Ponizovsky family of a commutative semigroup S # S1 is a Ponizovsky
family of S1 .
Lemma 2.3. Let S be a c.m. When P is a Ponizovsky family:
(P7) LA ~ U (B E S/'B IB ~ A) and KA =LAnA;
(P8) LALB ~ LAI\B and KAKB ~ KAI\B;
(P9) 1 E KB(l) and LB(l) = S.
Moreover, P is a sharp Ponizovsky family if and only if every KA is archimedean.
Proof. (P7) If x E LA, then x E PB for some B ~A and C = B(x) ~ B
by(P3); hence LA~ U(C E S/'B I C ~A). Next, KA = PAnA ~LAnA.
If conversely x E LAnA, then x E PB n A for some B ~ A in Sj'B, and
B =A since A= B(x) ~ B by (P3); therefore x EPA n A= KA.
(P8) A 1\ B is the product of A and B in the semilattice Sj'B. Let a E LA
and b E LB. By (PI), ab E P0 for some C E Sj'B. By (P2), ab E LA, so
C ~A. Similarly abE LB and C ~B. Hence C ~A 1\ B and abE LAI\B.
Then KAKB ~ LALB ~ LAI\B' KAKB ~ AB ~A 1\ B, and KAKB C
LAI\B n (A 1\ B) = KAI\B by (P7).
172 VII. SUBCOMPLETE SEMfGROUPS.
(P9) We have 1 E PA for some A E S/13 by (PI); then 1 E B for some

B ~A, by (P3); since U = B(1) is the greatest element of S/13 this implies
B = A = U and 1 E Pu n U = Ku. By (Pl ),
Lu = U(Pn I BE S/13, B ~ U) = U(PB I BE S/13) = S.

If finally :P is a sharp Ponizovsky family, then every 13-class A is archime-
dean; if k,l E KA, then kn = lt for some n > 0 and tEA, kn+l = l(kt)
with kt E KA, since KA is an ideal of A by (P7), and KA is archimedean. If
conversely every KA is archimedean, then a, b E A implies am, bn E KA by
(P6) and aN am N bn N b; thus every 13-class A is archimedean, 13 = N, and
:P is a sharp Ponizovsky family. 0
3. We now look at some particular cases.
Proposition 2.4. A complete c.m. has only one Ponizovsky family, namely
its partition into partial Ponizovsky factors.
Proof. If S is complete, then T = S is a monoid completion of S and
(P;(A)) AEY(S) is a sharp Ponizovsky family, by Proposition 2.2. Conversely let
:P = (PA)AEs; 13 be a Ponizovsky family relative to a semilattice congruence 13.
We show that 13 = N. Let e,f E E(S). Assume that e 13 f and let
. A= B(e) = B(f). Then e =en E KA by (P6), f E KA, and e = J, since
KA is cancellative by (PS), (P7). Thus a 13-class contains only one idempotent
of S. Since S is complete, a 13-class contains only one archimedean component
of S, and 13 = N.
Let e E E(S) and A= A(e). Then He~ LAnA= KA- Ifconversely
a E KA, then ea E LA, ea = e(ea), and a= ea EAneS= He by (PS). Thus
KA =He.
Next, eS ~ LA, since e E He ~ KA. If conversely a E LA, then as above
ea E LA, ea = e(ea), and a= ea E eS by (PS). Thus LA= eS.
Finally P: ~ eS =LA- Let a E LA' Iff< e in E(S), then a E JS
implies a E LA(!)• a E PB for some B ~ A(f) <A, and a ~ PA by (PI). If
conversely a¢ PA, then a E PB ~ LB for some B <A in Y(S), and a E JS
for some f < e in E(S). Hence
PA = LA\ U(Pn I BE S/13, B <A) = eS \ (U/<e JS) = P;.

Thus :P is the partition of S into its partial Ponizovsky factors. 0
On the other hand, a c.m. may have many Ponizovsky families:
3. ANOTHER COMPLETION THEOREM. 173
Example 2.5. Let S = { xn I n ~ 0} be the free c.m. with one generator

x (cf. Example Vl.3.7). There are two archimedean components, U = {1} and
C = {xn In> 0}. For any given integer m > 0, let
Pu = { xn I n < m} and Pc = { xn I n ~ m};

then Ku = {1}, Lu = S, and K 0 = L 0 = P0 . Then P is a partition of
S; Lu and L 0 are ideals of S; Pu ~ U U C, P0 ~ C, and Ku, Kc are
nonempty; (P4) (ka E PB if and only if a E PB when a E LA and k E KA)
is trivial when A = U (then k = 1) and when A = C (then B = A); (P5)
holds since S is cancellative; and (P6) holds since am E Kc for all a "I= 1 in
S. Thus P is a sharp Ponizovsky family. Different values of m yield different
Ponizovsky families of S.
Proposition 2.6. Every subelementary monoid S = C UN has a Ponizovsky
family in which Sj'B = { C,N}, P0 = S\0, Kc = C, Lc = S, and PN =
KN = LN = {0}.
Proof. T = c- s1 is a monoid completion of S with two idempotents 1
and 0 and two partial Ponizovsky factors Pt = T\ {0} and P0* = {0}; this
yields two 'B-classes, C and N, and a Ponizovsky family in which P0 = S\0,
Lc = S, Kc = C, and PN = LN = KN = {0}. 0
When S is subelementary, the Ponizovsky family in Proposition 2.6 is the
standard Ponizovsky family of S.
By Proposition 1.6, every f.g.c.s. has a Ponizovsky family. In fact:
Proposition 2.7. When S is finitely generated and m is sufficiently large,
then S has a sharp Ponizovsky family in which Am ~ KA for all A E Y(S).
Proof. If m is sufficiently large, then S 1 has a sharp completion T in which

Am ~ He( A) for all A E Y(S 1), by Lemma VI.4.2. D
3. ANOTHER COMPLETION THEOREM.
In this section we prove that every Ponizovsky family is induced by a com-

pletion, and that there is a smallest such completion (Grillet [2001S]).
1. The main result constructs a very tight completion.
Theorem 3.1. A commutative monoid S has a completion if and only if it
has a Ponizovsky family P, and then S has a monoid completion in which every
']{-class intersects S, and which induces the given Ponizovsky family.
One such completion is the monoid S'Y of all fractions ajk with a E LA
and k E KA for some A E Sj'B, in which (a/k)(b/l) = abjkl and ajk = b/l
if and only if a,b E P0 for some C E S/'B and al = bk.
Like the completion in Theorem VI.3.6, S'Y consists of fractions, but it is
constructed in one step and much more simply. The proof of Theorem 3.1 is
given below. The Theorem implies:
Corollary 3.2. A c.m. is subcomplete if and only if it has a Ponizovsky
family.
Corollary 3.3. A family of subsets of a c. m. S is a Ponizovsky family if and
only if it is induced by the partition of a monoid completion of S into its partial
Ponizovsky factors.
Theorem 3.1 is related to the following result of Tretjakova [ 1993]: a c.s. S
is embeddable into an epigroup if and only if it embeddable into a commutative
epigroup, if and only if it contains a subset K such that every element of S has
a power in K and ak 2 = bkl, akl = bl 2 implies ak = bl, for all a, b E S 1
and k,l E K. In Theorem 3.1, K = UAES/'B KA has these properties, by (PS)
and (P6).
The index of an epigroup S is the least m > 0 such that am belongs to a
subgroup for every a E S (or is infinite); for a commutative epigroup, this equals
the archimedean cancellativity index. Tretjakova [ 1993] also proved that a c.s.
S is embeddable into an epigroup of index ;£ m if and only if it embeddable
into a commutative epigroup of index ;£ m, if and only if ax 2m = bxmym,
axmym = by 2m implies axm = bym, for all a, b, x, y E S 1 . This amounts to
using K = UAEY(S) Am, as in Proposition 2.7.
2. When a c.m. S has a completion, then S has a monoid completion by

Proposition 1.6 and has a Ponizovsky family by Proposition 2.2. We now assume
that S has a Ponizovsky family P relative to some semi lattice congruence 'B,
and construct S'Y as in Theorem 3 .1. Let
1!' = { (a,k) Ia E LA, k E KA for some A E Sj'B }.
1!' is closed under componentwise multiplication: if a E LA, k E KA, b E LB,

l E K B, then ab E L A/\B and kl E K A/\B by (P8), and (ab, kl) E 1!'. Thus 1!'
is a c.s.; 1!' has an identity element, since (1, 1) E 1!' by (P9).
Lemma 3.4. The relation
(a,k) = (b,l) if and only if a,b E Pe for some C E Sj'B and al = bk

is a congruence on 1I'.
Proof. = is reflexive and symmetric. If (a, k) = (b, l) = (e, m), with, say,
a E LA, k E KA, bE LB, l E KB, e E Le, mE Ke, a,b E PD, and b,e E PE,
then D = E ~ A,B,C by (Pl), al = bk, bm = cl, alm = bkm = ekl, and
am= ek since am,ek E PD ~ LB by (P4) and l E KB is cancellative in LB
by (P5). Hence = is an equivalence relation.
Let (a,k) := (b,l), with, say, a E LA, k E KA, bE LB, l E KB; then
a,b E PD for some DE Sj'B, D ~ A,B, and al = bk. Let (e,m) E 'JI', with
e E Le, mE Ke. Then ae, beE LeAD by (P8). Let ae E PE, so that E ~ D 1\ C.
By (P4), acl E PE since ae E LeAD ~ LB; bek = acl E PE; and be E PE,
since be E LeAD ~ LA. Finally aclm = bekm. Thus (ae, km) = (be, lm).
Hence = is a congruence on 1I'. D
3. Sp is the quotient monoid T = 1I' / =. We denote the =-class of (a, k) E 1I'

by ajk; thus ajk is defined when a E LA and k E KA for some A E Sj'B,
and
ajk = b/l if and only if a,b E Pe for some C E Sj'B and al = bk.
In particular a/1 is defined for all a E S = Lu by (P9) (where U = B(1))
and a/1 = b/1 if and only if a = b. Hence L : a 1----1- a/1 is an injective
homomorphism of S into T. It is convenient to identifY a E S and a/1 E T,
so that S is a submonoid of T.
Lemma 3.5. In T = Sp: a}( ajk in T for every ajk E T; in particular
aN ajk in T for every ajk E T; and ajk N b/l in T if and only if a 'B b
in S.
Proof. Let ajk E T, with, say, a E LA, k E KA, and a E PB, where
B ~ A. In T, (k/1)(a/k) = ak/k = a/1, since a,ak E PB by (P4) and
(ak) 1 = ak. Similarly (k/k 2 )(a/1) = akjk 2 = ajk, since a,ak E PB and
(ak) k = ak 2 . Thus a :J{ ajk. In particular aN ajk.
Next we show that ajk = bjl implies a 'B b. Let a E LA, k E KA, bE LB,
l E KB, and a,b E Pe, so that C ~ A,B. Then B(a), B(b) ~ C ~ A,B by
(P3),
B(a) = B(a) 1\ B(l) = B(al) = B(bk) = B(b) 1\ B(k) = B(b),
and a 'B b.
If now aN b holds in T, then an= b(ejm) = bejm for some n > 0 and
cjm E T, an 13 be inS, and B(a) ~ B(b); similarly B(b) ~ B(a) and a 13 b

in S. To prove the converse we note that k JC l in T for all k, l E KA: indeed
k(l/k) = kljk = l/1 = l, since kl,l E KA by (P8), and similarly l(k/l) = k.
If now a 13 bin S, then am, bn E KA for some m,n > 0 by (P6), am JC bn
in T, and a N b in T. 0
By Lemma 3.5, every JC-class ofT intersects S; hence every archimedean
component of T intersects S.
Lemma 3.6. In T = Sp. eA = k/k is idempotent whenever k E KA. Every

idempotent ofT is of this form; moreover eA ~ eB in T if and only if A ~ B
in S/13.
Let a E LA and k E KA. When a E PB, then eB is the least idempotent
e ofT such that e (ajk) = ajk. Moreover ajk = (ajl)(kjk 2 ), and kjk 2 is
the inverse of k in HeA.
= ljl for every k,l E KA, so that eA =

Proof. First k/k k/k depends only
on A, and eA = k/k = k 2 jk 2 = (k/k) 2 is idempotent.
If ajk is idempotent, with, say, a E LA, k E KA, and a E PB, then
C = B(a) ~ B ~A and ajk =an /kn for every n > 0. Since an E Kc E Pc
for some n > 0 by (P6), this implies a E P0 n C = Kc. Then a 2 jk 2 = ajk
yields ka 2 = k 2 a, a 2 = ka by (PS) since a 2 ,ka E LA, and a/k = aja = e 0 .
Thus every idempotent ofT has the form eA for some A E S/13.
If A~ Bin S/13, then (kjk)(ljl) = kljkl = kjk for all k E KA, l E KB,
since kl E KA by (P8). If conversely k E KA, l E KB, and kljkl = (k/k)(l/l) =
kjk, then kl E KA and A= A 1\ B ~ B by (P8). Thus eA ~ eB in T if and
only if A~ B in Sj13.
Let ajk E T, where a E PB, a E LA, B ~ A, and k E KA" Let
l E KB. Then la E PB by (P4) and (l/l)(ajk) = lajlk = ajk. If conversely
(mjm)(ajk) = ajk, where mE K 0 , then am E PB, B ~ C since am E Lc by
(P2), and ljl ~ mjm in T. Thus eB = ljl is the least idempotent e ofT such
that e(ajk) = ajk. Finally, (ajl)(kjk 2 ) = akjk 2 = ajk, since a,ak E PB
and (ak) k = ak 2 , and (k/l)(kjk 2 ) = k 2 jk 2 = eA, so that kjk 2 is the inverse
of kin HeA. D
We can now complete the proof of Theorem 3.1. By Lemmas 3.5, 3.6, every
archimedean component ofT contains an 13-class A of S, contains some a E A,
contains k = an E KA by (P6), and contains the idempotent kjk. Moreover
there is for every x = aj k E T a least idempotent e of T such that ex = x, by

Lemma 3.6. Thus T is complete.
By Lemma 3.5 the congruence induced on S by the congruence N on T
is the given semilattice congruence 23. When e = e0 , Lemma 3.6 implies that
E(ajk) = e if and only if a E P0 ; hence ajk E Pe* in T if and only if a E P0 ,
P0 = Pe* n S, and 'Y is the Ponizovsky family induced by T. D
4. We illustrate the construction of S'Y with two examples.

Proposition 3.7. Let S = C UN be a sube/ementary monoid. If 'Y is the
standard Ponizovsky family of S, then S'Y = c- 1 s.
Proof. In the standard Ponizovsky family 'Y, P0 = S\0, L0 = S, K 0 = C,
and PN = LN = KN = {0} (Proposition 2.6). By definition, T = S'Y consists
of all fractions a/ k where either a E S\ 0 and k E C, or a = k = 0, with
(ajk)(b/l) = abjkl and ajk = b/l if and only if al = bk and either a,b E S\0
or a = b = 0. The fraction 0/0 E T is the zero element of T and may be
identified with the zero element 0I k of c-l s; then T = c-l s. D
x. As in Example 2.5, let m > 0 and
Pu = { xn I n < m} and Pc = { xn In ~ m};
then 'Y is a sharp Ponizovsky family and Ku = {1}, Lu = S, K 0 = L0 = P0 .
S'Y consists of all fractions ajk where either a E S, k = 1 or a, k E C,

with ajk = bjl if and only if al = bk and either a,b E Pu or a,b E K 0 . The
fractions k / l with k, l E K 0 constitute the group of fractions of Kc, which is an
infinite cyclic group { yn I n E Z}, where xn j xP = yn-p whenever n, p ~ m.
We see that xn /1 = yn for all n ~ m, but xn = xn /1 equals no other element
of S'Y when n < m. Thus
s'Y = { xn I 0 ~ n < m} u { yn I n E z}'

where xi yj = yi+j and xn E S is identified with yn E S'Y if n ~ m. Thus
S'Y is the completion of index m constructed in Example Vl.3.7. D
5. Finally we show that S'Y is the smallest monoid completion of S which
induces 'Y:
Proposition 3.9. If Tis a monoid completion of S which induces the
Ponizovsky family 'Y, then there exists a unique homomorphism of S'Y into T
178 VII. 8UBCOMPLETE 8EMIGROUPS.
which is the identity on S, and it is injective.

Proof. Since T induces P, 'B is induced by the congruence N on T, there
is for every 'B-class A an idempotent e(A) in the archimedean component ofT
which contains A, and A~----+ e(A) is an isomorphism of Sj'B ~ E(T). When
e = e(A), then KA ~He and LA~ eT, by Proposition 2.2; let k- 1 E T denote
the inverse of k E KA in He .
We show that ajk = b/l in S'Y if and only if ak- 1 = bl- 1 in T. Let
ajk, bjl E T, with a E LA, k E KA, b E LB, and l E KB. Assume that
ajk = bjl, so that al = bk and a,b E P0 for some C E Sj'B. Then C ;£ A,B
and e = e(A), f = e(B), g = e(C) satisfy g ;£ e,f. Also E(a) = E(b) = g
in T, since a,b E P0 = Sn~*; hence ga =a, gb = b, ea =ega= ga =a,
and similarly fa= a, eb = b, and fb =b. Now k E He, l E H 1 , and
ak- 1 = afk- 1 = all- 1 k- 1 = bkk- 11- 1 = bel- 1 = bl- 1 .

If conversely ak- 1 = bl- 1 in T, then a= ae ]{ ak- 1 = bl- 1 ]{ bf = b and
E(a) = E(b). If g = E(a) = E(b), then g ;£ e, f, since ea = fa = a, and
a,b E Sn~* = P0 for some C E Sj'B. Then ajk = bjl, since
al = ael = alkk- 1 = blkl- 1 = bfk = bk.

By the above a mapping <p: S'Y--+ Tis well defined by: <p(a/k) = ak- 1 ,
and it is injective. Moreover <p is the identity on S, since 1- 1 = 1 in T and
<p(s) = <p(s/1) = s for all s E S = LB( 1) by (P9). Let ajk, b/l E S'Y,
where a E LA, k E KA, b E LB, and l E KB. By (P8), ab E Lc and
kl E K 0 , where C =AI\ B. Let e = e(A), f = e(B), and g = e(C). Then
g = ef, kl E He H 1 ~ H9 , k- 1 z- 1 E H9 , (kl)(k- 1 1- 1 ) = ef = g, and
k- 1 z- 1 = (kl)- 1 . Hence
<p ((ajk)(b/l)) = ab (kl)- 1 = (ak- 1 )(bl- 1 ) = <p(ajk) <p(b/l).

Thus <p is a homomorphism.
Conversely let 'ljJ : S'Y --+ T be a homomorphism which is the identity on

S. For every A E S j'B, 'ljJ sends the archimedean component of S'Y which
contains A into the archimedean component ofT which contains A, and sends
the idempotent of the former to the idempotent of the latter. Thus 'ljJ (eA) = e(A).
By Lemma 3.6, kjk 2 is the inverse of k E KA in HeA, so that 'ljJ (kjk 2 ) =
'lj;(k)- 1 = k- 1 , and
4. PROPERTIES. 179
'ljJ (ajk) a k -1 .
Thus 'ljJ = r.p. D
4. PROPERTIES.
In this section we use Ponizovsky families to transfer properties of complete

semigroups to subcomplete semigroups; this includes explicit subdirect decompo-
sitions and a congruence which is similar to but larger than J{.
1. By Proposition Vl.2.2, a subcomplete semigroup S is a subdirect product
of nilsemigroups, cancellative semigroups, and subelementary semigroups. With
a Ponizovsky family we can construct an explicit decomposition. In the complete
c.m. Sy, the projection Sy ---t ~ induces a congruence on Scy; the intersection
of these congruences is the equality on Sy, since the projections Sy ---t ~
separate the elements of Sy. Restriction to S yields:

Proposition 4.1. Let S be a c.m. and P be a Ponizovsky family on S.
When C E Sj'B and m E K 0 , the congruence on S induced by the projection
Sy ---t ~ is the relation
a Pc b if and only if either am= bm E P0 , or am,bm ~ P0 ;
hence Pc is a congruence and does not depend on the choice of m; S jPc is
cancellative if C is the least 'B-class of S, otherwise SfPc is subelementary;
and ncES/'13 Pc is the equality on S.
Proof. LetT= Sy and e E E(T). By definition, the projection 7fe: T ---t ~

takes x E T to ex if ex E ~*, to 0 if ex E Ut<e JT; since ex E eT,
ex E Ut<e JT = Uf<e Pj if and only if ex ~ ~*. Hence 1re(x) = 7re(Y) if
and only if either ex = ey E ~*, or ex, ey ~ ~* .
Let mE Kc and e = e0 = m/m E E(T). When a,b E S, then ea = eb in
T if and only if amjm = (m/m)(a/1) = (m/m)(b/1) = bmjm, if and only
if am,bm E PD for some DE Sj'B and am 2 = bm 2 ; since am,bm E Lc we
have D ~ C and am 2 = bm 2 implies am = bm by (P5). Thus ea = eb if and
only if am = bm. Moreover ea, eb E ~* if and only if D = C in the above,
by Lemma 3.6. Thus 1re(a) = 1re(b) in T if and only if a Pc b.
It follows that Pc is a congruence and does not depend on m. Furthermore
S j'Yc is isomorphic to a subsemigroup of ~, which intersects He since 1re (k) E

He when k E Kc . Hence S /'Yc is cancellative if C is the least 23-class of
S, otherwise S/'Yc is subelementary. Finally the intersection of all 'Yc is the
equality on S, since the projections 1re separate the elements of T. 0
2. The partial Ponizovsky factors of S.y already induce subsets PA of S

with similar properties. We now look at the congruence on S induced by the
congruence JC on S'Y. First we show:
Lemma 4.2. Let e = ec. In s'Y' g E He if and only if g = k I l for some

k,l E K 0 ; hence He = G(K0 ).
Proof. If k,l E K 0 , then (k/l)(l/k) = kl/kl = e0 and e0 (k/l) = kk/kl =
k/l; hence k/l E He.
Conversely let a/ k E He, with, say, a E LA, k E KA. Let m E Kc. Then
amjkm = (m/m)(a/k) = ajk and abjkl = (a/k)(b/l) = m/m for some
bjl E S'Y. By Lemma 3.5, am 23 a and ab 23 m, so that a 23 m and a E C.
Hence C ~ A by (P4). Then a E Pc n C = Kc and ajk = amjkm, with
am, km E Kc by (P8).
When k,l,m,n E K 0 , then k/l = m/n if and only if kn = lm. Thus
He= K 0 1 Kc = G(Kc). 0
Proposition 4.3. Let S be a c.m. with a Ponizovsky family 'Y. The relation
a X b if and only if a,b EPA and ak = bl, for some A E S/23 and k,l E KA
is a congruence on S; a X b in S if and only if a JC b in S'Y; and S /X~ S'Y jJC

is complete groupfree. If S is finitely generated, then S /X is finite.
~
Proof. Let a, b E S, a E PA. By Lemma V.2.2, a JC b in T = S'Y if and

only if b = ga for some g E He ~ T, where e = t:(a). By Lemmas 3.6 and
4.2, e = eA and g = k/l for some k, l E KA" Now ak E PA by (P4), so
b/1 = b = ga = (k/l)(a/1) = akjl if and only if b E PA and ak = bl. Thus
a JC b in T if and only if a X b in S. Since every JC-class of T intersects S it
follows that S /X ~ T jJC, and S /X is complete group free by Corollary V.1.6.
If S is f.g., then S/X is f.g. and the Ponizovsky factors of SjX, which are
nilmonoids, are f. g. and therefore finite. Hence S/X is a subdirect product of
finitely many finite monoids and is finite. 0
For example every KA is aX-class: if a, bE KA, then a X b, since a, bE PA
and ab = ba; if conversely a E KA and a X b, then a 23 b by Lemma 3.5 and
5. SCHU"TZENBERGER FUNCTORS. 181
bE AnPA = KA"
X is similar to the orbit congruence of a subelementary semigroup. In fact:
Proposition 4.4. If S = S 1 = C UN is subelementary, then, relative to the
standard Ponizovsky family, X is the orbit congruence on S.
Proof. Sp = c-l s' by Proposition 3. 7. By Proposition 4.3, X is induced
by Green's relation J{ on c- 1S and is the orbit congruence on S. 0
X is also similar to J{, as the next section will show. Subelementary semi-
groups show that it may be larger than J{. For another example:
x, and P consist of Pu = {xn In< m} and P0 = {xn In~ m}, where
m > 0, so that Ku = {1}, Lu = S, and K 0 = L0 = P0 (as in Example 2.5).
Then J{ is the equality on S, whereas X is the Rees congruence of Lc . 0
5. SCHOTZENBERGER FUNCTORS.
Continuing Section 4 we construct from the congruence X on S an expanded

SchUtzenberger-like functor. This actually provides a direct construction of the
SchUtzenberger functor of the complete semi group Sp, which may in tum be
used to reconstruct Sp as a group coextension of Sp/JC ~ SjX.
I. When K is a X-class,
St (K) = {a E S I aK ~ K} = {a E S I ac E K for some c E K},
since X is a congruence, and St (K) is a submonoid of S. Also:
Lemma 5.1. St (K) = St (Kc) for every X-c/ass K ~ P0 .
Proof. Let c E K and k E K 0 . If a E St(K0 ), then ak = l E K 0 . Hence
ack = cl. Also acE L0 , ack = cl E P0 by (P4), and acE P0 by (P4); hence
ac X c and a E St (K) .
Conversely let a ESt (K) and c E K, so that acE K. Let A= B(a) E Sj'B.
By (P6), an E KA for some n > 0. In Sp we have ac J{ c by Proposition
4.3, anc J{ c since J{ is a congruence, and an JC eA by Lemma 4.2; hence
eAc J{ c, c = eAc, eA ~ E(c), and A ~ C by Lemma 3.6. Hence k E K 0
implies ak E A/\C = C, ak E Cn£0 = K 0 , and a E St(Kc). 0
For every a E St (K) a mapping O"K(a) : K ---+ K is well defined by:

O"K(a)(c) = ac for all c E K.
Proposition 5.2. Let S be a c.m. with a Ponizovsky family P. For every
X-class K,
E(K) = {O"K(a) I a E St (K)}
is a cancellative c. m. of injective transformations of K; O"K : St (K) ---+ E (K)
is a homomorphism; and E(K) acts simply on K (if O", T E E(K) and O"C = rc
for some c E K, then O" = T ).
Proof. O"K(l) is the identity on K and O"K(ab) = O"K(a) o O"K(b) for all
a, b E St (K); hence O"K is a homomorphism of monoids and E(K) is a c.m.
Let K ~ Pc. If a E St(K) and ac =ad, with c,d E K ~ Pc, then, for
any k E Kc, we have akc = akd and c = d by (P5), since ak E Kc by Lemma
5.1. Hence every O"K(a) E E(K) is injective. Therefore E(K) is cancellative.
Finally let a, b E St (K) and ac = be for some c E K. Let d E K, so that
d E Pc and ck = dl for some k, l E Kc. Then adl = ack = bck = bdl and
ad = bd by (P5), since ad, bd E Lc and l E Kc. Hence E(K) acts simply
on K.D
E(K) is the SchUtzenberger monoid of the X-class K. It is, in general,
larger than the Schtitzenberger groups of the ~-classes H <;:;; K, but not transitive
or a group.
Since E(K) is cancellative it has a group of fractions r(K) = G(E(K));
this is the SchUtzenberger group of K.
Example 5.3. = { xn I n ~ 0} be
As in Examples 2.5 and 4.5, let S
the free c.m. with one generator x, and P consist of Pu = { xn I n < m}
and Pc = {xn In~ m}, where m > 0, so that Ku = {1}, Lu = S, and
Kc = Lc = Pc. Then St (Kc) = S, O"Kc is injective since S is cancellative,
and E(Kc) ~ S is larger than the Schtitzenberger groups of the JC-classes
H ~ Kc, which are trivial; but E(Kc) is not a group. When j ~ m, so that
xj E Kc, we have O"Kc(xi)(xj) = xixj = xi+j; E(Kc) is not transitive since
there is no O" E E(Kc) such that, say, O" xm+l = xm. Similarly r(Kc) ~ G(S)
is an infinite cyclic group; on the other hand the Schtitzenberger groups of the
JC-classes of S are all trivial. 0
Example 5.3 shows that the action of E(K) on K does not in general extend
to an action of r(K) on K. A partial action of r(K) on K is defined below.
5. ScHi.iTzENBERGER FUNCTORS. 183
2. The Schiitzenberger groups of the X-classes of S are naturally related to

the Schiitzenberger groups of S'Y .
Proposition 5.4. Let S be a c.m. with a Ponizovsky family P. Let H be
an '}{-class of S'Y and K = H n S. Then r(K) ~ r(H).
By Proposition 4.3 every X-class of S has the form H nS for some '}{-class
H of S'Y.
= S'Y . Recall that the Schiitzenberger group of an '}{-class H
Proof. Let T
ofT is r(H) = {'YH(t) It ESt (H)}, where St (H)= { t E T I tH ~ H} and
'YH (t) is well defined by: 'YH (t) (h)
= th for all h E H; it is a simply transitive
group of permutations of H. Proposition 5.4 follows from a more detailed result:
Lemma 5.5. Let H be an ']{-class of S'Y and K = H n S. Then St (K) ~

St (H) and there is an isomorphism ()K : r(K) ~ r(H), which is the unique
homomorphism such that ()K o CYK = 'YK; if CY E E (K), then ()K (CY) is the
unique ')' E r(H) which extends CY.
Proof. For any c E K, ajk E St (H) (in T = S'Y) is equivalent to

(a/k)c ']{ c, ac ']{ c in T, ac X c in S, and a E St(K) (in S), by Lemma
3.5 and Proposition 4.3. In particular St (K) ~ St (H).
When a, b E St (K), then CYK(a) = CYK(b) if and only if, for any c E K,
ac =be and 'YH(a) = 'YH(b). Therefore there is a unique monoid homomorphism
t: E(K)---+ r(H) such that t(CYK(a)) = 'YH(a) for all a ESt (K). Moreover,
" is injective. Therefore " extends to an injective homomorphism () = ()K :
r(K) ---+ r(H), defined by O(CY/T) = t(CY) 0 t(r)- 1 for all CY,T E E(K). The
square
St(K) ~ St(H)
aKl l~H
r(K) o;:+ r(H)
commutes; since r( K) is generated (as a group) by Im CYK, () is the only homo-

morphism with this property. For every a E St(K), O(CYK(a)) = 'YH(a) extends
CYK(a), since 'YH(a)(c) = ac = CYK(a)(c) for every c E K ~ H; it is the unique
')' E r(H) which extends CYK(a) since r(H) is simply transitive.
It remains to show that () is surjective. Let ')' E r(H), so that ')'C ']{ c in T
for any c E K = H n S. By Lemma V.2.2, ')' c = gc for some g E He, where
184 VII. 8UBCOMPLETE 8EMIGROUPS.
e = c(c); by Lemma 3.6, e = e0 , where c E P0 , and then g = m/n for some

m,n E K 0 , by Lemma 4.2. Thus "(C = (mjn)c and"(= "fH(m/n) for some
m,n E K 0 . In T, mjn = m(njn 2 ) and njn 2 is the inverse of n in He. Now
He ~ St (H) , since J{ is a congruence on T, and the restriction of "fH to He
is a group homomorphism; hence
"( = "fH(mjn) = "fH(m) "fH(n)- 1 = O(aK(m)faK(n)).

Thus OK is surjective. D
A partial action of f(K) on K can be defined as follows: if "( = a jT E
f(K), where a,T E 'E(K), and c E K, then "f.C is defined if and only ifthere
exists d E K such that Td = ac, and then "( . c = d. That this is well defined
follows from Lemma 5.5:
Proposition 5.6. Let H be an J<:-class of Sy and K = H n S. Up to the
isomorphism r (K) ~ r (H), the partial action of r (K) on K is the restriction
to K of the action of f(H) on K. For every a, b E K there is a unique
"f E f(K) such that "(.a= b.
Proof. Denote by "f -----r '7 the isomorphism () = OK : r(K) -----r f(H) in
Lemma 5.5. When"(= afT E f(K), where a,T E 'E(K), then '7 = 7- 1 o(f,
and a, T are the restrictions to K of (j and 7. If c, d E K and Td = ac, then
rd = (fc and '7c = d; thus, if "f • c is defined, then '7c E K and "f • c = '7c. If
conversely d = 'fc E K, then, whenever"(= a jT E r(K) (with a, T E 'E(K) ),
rd = (fc, Td = ac, and "f. c is defined, and then "( . c = '7c. In this sense the
partial action of "f on K is exactly the restriction to K of the action of '7.
For every a,b E K there is a unique '7 E f(H) such that '7a = b; therefore
there is a unique "f E f(K) such that "f. a = b. D
3. In S, the Schiitzenberger monoids and groups of X-classes organize
themselves into functors. By Proposition 4.3, s;x ~ SpjJ<: is partially ordered
by :£J{; in other words, K ~ L in S /X if and only if K M = L in S /X for
some ME SjX, if and only if Ks ~ L for somes E S.
Proposition 5.7. Let S be a c.m. with a Ponizovsky family 'Y. When K ~ L
in SjX, then St (K) ~ St (L) and there exists a unique monoid homomorphism
af : 'E(K) -----7 'E(L) such that the square
St(K) ~ St(L)
aKl laL
'E(K)
5. SCHUTZENBERGER FUNCTORS. 185
commutes. The c.m. ~(K) and homomorphisms O"f constitute a monoid valued
functor E on S /X.
As in Section V.2, E = (~, O") is actually a thin functor on H(S/X), in
which ~ (K) is the Schiitzenberger monoid of K and O"K T = O"§T is provided
'
by Proposition 5.7; E is the Schfltzenberger monoid functor of S relative
toP.
Proof. We have K s ~ L for some s E S. Hence aK ~ K implies aK s ~ K s
and ad E L for some d E L, and U(K) ~ U(L). Also O"K(a) = O"K(b) in
U(K) implies, for any c E K, ac =be, acs = bcs, and O"L(a) = O"L(b), since
cs E L. Therefore there is a unique monoid homomorphism O"f : ~ ( K) ---+ ~ ( L)
such that O"f ( O"K (a)) = O"L (a) for all a E U (K) . Since O"f is unique with
this property, O"§ is the identity on ~(K), and O"ft o O"f = O"~ whenever
K ~ L ~ M in SjX. D
When K ~ L, the monoid homomorphism O"f : ~(K) ---+ ~(L) induces a
homomorphism 'Yf : r(K) ---+ r(L) of abelian groups such that the square
aK
~(K) ~ ~(L)
~1 1~
r(K) ~ r(L)
rf
commutes. Since 'Yf is unique with this property, 'Y§ is the identity on r(K),
and 'Yft o 'Y[ = 'Y~ whenever K ~ L ~ M in S /X. Hence:
Proposition 5.8. The groups r(K) and homomorphisms 'Yf constitute an

abelian group valuedfunctor ll"' on SjX.
ll"' is the extended Schfltzenberger functor of S relative to P and is actually
a thin abelian group valued functor on H(S/X).
Proposition 5.9. Let S be a c.m. with a Ponizovsky family P. Up to the
isomorphism S/X ~ Spj'J{, the extended Schatzenberger functor of S relative
to P is isomorphic to the Schatzenberger functor of Sp.
Proof. Let T = Sp. The isomorphism S/X ~ Sp j'J{ in Proposition 4.3

sends a X-class K of S to the ']{-class H of T which contains K.
Let H and J be ']{-classes ofT and K = H n S, L = J n S. Since
S/X ~ Tj'Ji, we have H ~ J in Tj'){ if and only if K ~ L in S/X;

then K s ~ L for some s E S, and H s ~ J. By Lemma V.2.1, the inclusion
St (H) ~ St (J) induces a homomorphism 'Ylf : r(H) -+ r(J) of abelian
groups such that the square
st (H) 'H > r(H)
~1
St(L) ---+
IJ
commutes. Now Lemma 5.5 provides isomorphisms (}K : r(K) -+ r(H) and
(}L : r(L) -+ r(J). We show that the square
r(K) ~ r(H)
~tJ l~Y
r(L) --e;:+ r(J)
commutes. For every a E St (K) we have
'YlJ(eK(aK(a))) = "flf('YH(a)) = 'YJ(a) = (}L(aL(a)) = (}L('Yf(aK(a)));

since lm aK generates r K, this implies "flj o (}K = (}L o "'f. Thus, up to
the isomorphism S/X ~ Sp/'Ji, () is a natural isomorphism from the extended
Schiltzenberger functor of S to the Schiltzenberger functor of T. 0
Chapter VIII.
OTHER RESULTS.
Far too much worthwhile work has been done on commutative semigroups to
be exposed in some detail in one book. This chapter is a brief survey of topics
which are not covered by other chapters. As noted in the Preface, some subjects
have been omitted: partially ordered semigroups; varieties and pseudovarieties;
factorization theory; and semigroup rings.
1. EXAMPLES.
Commutative semigroups arise in various and sundry ways.

1. Finite semigroups with small numbers of elements have been an early
source of abundant examples. Lists of all distinct semigroups with n elements
have been computed for n ~ 8, incidentally providing all c.s. with n ~ 8 elements;
c.s. with n = 9 elements have been computed separately (Grillet [1996N]). Early
lists were compiled by hand (Tamura [1953], n ~ 3; Tamura [1954], n = 4;
Tetsuya, Hashimoto, Akazawa, Shibata, lnui, & Tamura [ 1955], n = 5 ). Other
early hand-drawn lists include all group-free c.s. of order n ~ 6 (Sagastume Berra
[ 1957]) and all archimedean semi groups of order n ~ 8 (Tamura, Dehara, Iwata,
Saito, & Tsukomo [ 1960]); see also Takahashi [ 1985]. Computers were soon
brought to bear on the problem (Forsythe [1955], n = 4; Motzkin & Selfridge
[ 1956], n = 5; Plemmons [ 1966], n = 6; Jurgensen & Wick [ 1976], n = 7;
Sato, Yama, & Tokizawa [ 1991 ], n = 8 ), using backtrack algorithms (JOrgensen,
Pallas, & Wick [1976]; see also Grillet [1996N]). Actual tables of semigroups
with n ~ 6 elements were published by Plemmons [1966] and Jorgensen [1990].
These computations suggest the Kleitman-Rothschild-Spencer conjecture,
which states that, on a statistical basis, almost all finite semigroups are nilpotent
of index ~ 3 ( N 3 = 0 ). This conjecture is a statement in Kleitman, Rothschild,
& Spencer [1976]; there is considerable evidence that it is true, but gaps in the
original proof have remained unfilled. There is some evidence that the commu-
187
188 VIII. OTHER RESULTS.
tative version of the conjecture is also true (almost all finite c.s. are nilpotent of
index ~ 3).
2. C.s. also arise from number systems. Additive subsemigroups of N and
their ties to algebraic geometry have been mentioned in Section 11.4. On a lighter
note, Sasaki [ 1988] showed that they are determined up to isomorphism by their
power semi groups; Ch 'ilan & Lee [ 1982] showed that the identity on N is the
only permutation of N under which all direct images and inverse images of
subsemigroups are subsemigroups. Various other explicit c.s. operations on N and
Z have been constructed e.g. by Fraenkel, Porta, & Stolarsky [ 1989]. Hippisley-
Coz [1992] studied n-point compactifications of N. See also Painter [1966].
Integers modulo n constitute a c.s. Zn under multiplication. Parizek [ 1957]
showed that Zn is Clifford if and only if n is a product of distinct primes.
Hewitt & Zuckerman [ 1960] and Parizek & Schwarz [ 1961] found all its char-
acters. Painter [ 1967] studied its groups of units. Lenzi [ 1980] studied which of
its JC-classes is 0-cancellative in ZnfJC. Abrgan [ 1984] studied its prime and
semi prime ideals. Dyadchenko & Shokuev [ 1985] determined its idempotents.
Jones & Ligh [ 1980] determined all ring additions on zn.
Free multiplicative semigroups of real numbers > 1 have been studied for
density properties by Cudakov & Pavlyucuk [1951], Bredihin [1953], [1954],
[1958A], [1958B], Cudakov & Bredihin [1956], Wegmann [1966], and Cibul's-
kite [1970]; many of these results extend to real valued homomorphisms of count-
ably generated free c.s.
Closed intervals in lR form c.s. under additions and under multiplication.
These semi groups were studied by Ratschek [ 1972] and Mitrovic [ 1976], [ 1977]
with an eye toward numerical analysis.
3. Free commutative semi groups are seen in Chapters VI, IX, and XIII; their
finitely generated subsemigroups have ties to algebraic geometry and have been
mentioned in Section II. 7. In what follows F denotes a free c.s. Tamura &
Sasaki [1963] and Sasaki [1964] studied subsemigroups ofF assuming that F
is countably generated. Decompositions of F (finding subsets (Ai\EI of F
such that every element of F can be written uniquely as a product rriEI ai
with ai E Ai) were investigated by de Bruijn [ 1956], Hansen [ 1969], Niven
[1971], Marcus [1974], and Lepetit [1973]. (Hansen [1969] studied similar de-
compositions of Zn.) Grillet [ 1969F] gave a characterization, using the Riesz
interpolation property, and showed that the fixed points of any set of endomor-
phisms of F constitute a free c.s.; another interesting characterization, in terms of
semigroup algebras, was given by Bruns & Gubeladze [ 1999]. Bredihin [ 1960]
studied the asymptotic behavior of v( x) , the number of elements of norm ~ x,
1. EXAMPLES. 189
for multiplicative norms on F such that v( x) is always finite. The retractions of

F were studied by Stambolieva [1973] when F is finitely generated. The maxi-
mal size of independent systems of equations in F was studied by Karhumaki &
Plandowski [ 1996]. Commutative languages (sets of subsets of F, where F is
finitely generated) were considered by Perrot [ 1965], Eilenberg & Schiitzenberger
[ 1969], and Butzbach [ 1972].
4. Other examples are provided by commutative subsemigroups of noncom-
mutative semigroups, such as full transformation semigroups (Rankin, Reis, &
Wang [ 1976]). C.s. of polynomials under composition were studied by Lidl &
Muller [1986], Nobauer [1986], [ 1987], Tumwald [ 1986], Eigenthaler, Nobauer,
& Wiesen bauer [ 1987], and Eigenthaler & Woracek [ 1997].
5. Characterizing multiplicative semigroups of commutative rings remains
an open problem. The class of c.s. with this property is not axiomatizable (Ko-
galowski [ 1961]) and its finitely generated members are finite (Isbell [ 1959],
Anderson & Stickles [2000]). Aull [1967] observed that a semigroup ideal which
is directed upward under ~9-C must be a ring ideal. Satyanarayana [1973] and
Isbell [2000] obtained various necessary properties.
6. As noted in the preface, commutative semigroups turn up in divers places
besides algebraic geometry. Examples include the monoid of equivalence classes
of Boolean lattices (Dobbertin [ 1982], [ 1983]); the c.s. of family graphs (Far-
rell [ 1992]); certain continuous lattices (Wehrung [ 19980); distributive lattices
(Berg [ 1985]); semi groups of circulant matrices (Schwarz [ 1977], Gurican [ 1984],
Huang [1996]); semigroups of subvarieties of varieties of associative algebras
(Gonchigdorzh & Mal'tsev [ 1982]); bilinear spaces over a scheme, under orthog-
onal sum (Knebusch [ 1969]); the semigroup of generalized divisors on certain
schemes (Hartshorne [ 1994]); the semi group of stable equivalence classes of al-
gebraic tori (Cistov [ 1978]); the semigroup ofNewton polyhedra in rings of formal
power series (Lipkovski [ 1988]); the branch multiplicity semigroup of a formal
power series (Kulkarni [ 1990]); semi groups of convergent power series (Groza
[1995]). No doubt a few others have been omitted from this list.
Bazzoni & Salce [ 1996] showed that the class semigroup of a valuation do-
main is a Clifford semigroup; domains with this property were studied by Zanardo
& Zannier [1994]. Kunze [1978] studied the class semigroups of ordered local
rings. Shirota [ 1952] showed that a locally compact Hausdorff space is deter-
mined up to homeomorphism by its multiplicative semigroup of continuous real
valued functions; this result was generalized by Thanh [ 1992].
2. PRODUCTS AND SUBSETS.
This section surveys properties of commutative semigroups related to products,

subsemigroups, ideals, and other subsets.
1. A number of interesting combinatorial properties have been established
for c.s. Some have applications to graph theory (Pus [ 1991 ]). Burgess [ 1969]
proved that, in an archimedean semigroup with n elements, every sequence of n
elements contains a subsequence whose product is idempotent. De Luca & Restivo
[1984] proved (for semigroups in general) that a finitely generated semigroup S
is finite if and only if there exists an integer m > 0 such that every sequence
s 1 , ... , sm of m elements of S satisfies s 1 · · · sj = (s 1 · · · sj)(si · · · sj) for
some 1 ;;;:; i ;;;:; j ;;;:; m. See also Niculescu [ 1995], [ 1996].
Khovansky [1992] showed that IAnl has polynomial growth in n when S
is commutative and A ~ S is finite. Nathanson [2000] generalized this to
IB A 1 1 · · · Arnr I· In a finitely generated c.s. S, an element a which is infinitely
n
divisible ((\In > 0) (:3x E S) xn = a) is idempotent (Clark, Holland, & Szekely

[1998]). Hindman & Woan [1993] studied systems oflinearequations AX= BX
in c.s., with coefficients in N.
2. C.s. with various additional elementary properties have been considered (for
instance, archimedean semigroups and separative semigroups). Tamura [1971]
studied c.s. in which xyz E {xy,xz,yz} for all x,y,z. Hora & Kimura [1972]
considered c.s. in which (Vx)(:3y) xmyp = xnyq (where m,n,p,q ~ 0 are
given). Blagojevic [1982] investigated c.s. in which (Vx)(:3y) x = xy 2 , y =
yx 2 . Cancellativity properties are mentioned in Chapter II. Other properties were
studied by Galbiati & Veronesi [ 1986].
The Riesz interpolation property states that ab = cd implies a = xy,
b = zt, c = xz, d = yt for some x,y,z,t. C.s. with this property (also called
refinement semi groups) were studied by Dobbertin [ 1983], [ 1984], and Shortt &
Rao [1989]; see also Goodearl [1986] and (for cancellative c.s.) Grillet [1970C],
[19701], and Caillot & Wehrung [2000]. Embeddings into refinement semigroups
were constructed by Wehrung [1998E].
Krob [ 1987] defined a complete c.s. as a set S with commutative and asso-
ciative product mappings rriEJ s ~ s' one for every set I.
3. The subsemigroups of a semigroup constitute a complete lattice :E(S).
According to the lucid and well organized survey by Shevrin & Ovsyannikov
[ 1983 ], the study of this lattice consists of three principal areas: lattice isomor-
2. PRODUCTS AND SUBSETS. 191
phisms, and determinacy (when is S determined up to isomorphism by :E(S) );

properties of :E (S) ; semigroups with certain types of lattices. Early work rele-
vant to commutative semigroups includes the following. Semigroups S such that
:E(S) is distributive, complemented, or a chain, were characterized by Shevrin
[1961]; distributive, by Ego [1961]; modular or semimodular, by Ego [1962A];
semidistributive, by Shiryaev [1985]. Ego [1962B], [1962C], [1963] also con-
sidered various weakened forms of modularity. Laplaza [1974] showed that a
c.s. whose subsemigroup lattice has finite length must be finite. Kacman [ 1979]
studied c.s. whose subsemigroup lattice is self-dual (more generally, pairs of
c.s. with anti-isomorphic subsemigroup lattices). See also Morel [1979]. Rep-
nitsky & Kacman [1988] and Repnitsky [1994] determined semigroups whose
lattice of subsemigroups satisfies a nontrivial identity. Shevrin [ 1966] showed
that, in a number of classes of commutative semigroups, subsemigroup lattices
can be characterized by finitely many axioms. Isomorphisms and determinacy
were studied by Mogiljanskaja [1972], Baransky [1975], Ovsyannikov [1977A],
[19778], [1981], Liberman [1980], and Borisov [1982], [1984], [1986], [1992].
Ovsyannikov [ 1977A] showed that every c.s. can be embedded into a c.s. which is
determined by subsemigroup lattice. Jorgensen & Wick [ 1973] gave an efficient
algorithm which computes :E(S) when S is a finite c.s.
Subsemigroups give rise to other structures. The lower semilattice of sub-
semigroups was considered by Evseev [ 1965]. The subsemigroups of a semigroup
and their isomorphisms constitute an inverse semigroup; the determinacy problem
for this inverse semigroup was studied by Libih [1974] for certain commutative
Clifford semigroups. The subsemigroups of a c.s. also constitute a c.s. under
multiplication; this was studied by Mogiljanskaja [ 1971]. See also Etayo [ 1965].
Various concepts of rank have been defined for a c.s. S. Shevrin [1963]
defined the rank of S as the smallest r such that every finitely generated sub-
semigroup of S is generated by at most r elements; he showed that S has finite
rank if and only if its archimedean components have finite rank. Okninski [ 1988]
defined rank by means of 0-cancellative homomorphic images, equivalently, by
means of free subsemigroups; this is useful in studying the Krull dimension of
semigroup rings.
4. Evseev [ 1991] defined a closing element c by the property that every
subset which contains c is a subsemigroup, and showed that such exists in a c.s.
S if and only if every partition of S has a class which is a subsemigroup.
A number of conditions on subsemigroups have been considered. Commuta-
tive semi groups all of whose subsemigroups are finitely generated have been seen
in Section VI. 7. Jensen & Miller [ 1968] showed than the only infinite c.s., all of
whose proper subsemigroups are finite, are the groups Z(p00 ) . Gluskin [1971],
[ 1973 ], [ 1976] studied dense extensions for subsemigroups and for normal sub-
semigroups. Nordahl [ 1973] found all c.s. whose every proper subsemigroup is
power joined. Schein [ 197 5] studied cancellative c.s. in which every subsemi-
group is dense in its idealizer. Morel [ 1979] constructed locally cyclic semigroups.
Hmelnitsky [ 1985] studied c.s. whose every proper subsemigroup is different from
its idealizer, and c.s. whose every proper subsemigroup appears in an ideal series.
Krivenko [ 1985] studied c.s. whose every cyclic subsemigroup is normal. Aucoin
[ 1999] characterized c.s. in which all ideals of every subsemigroup are induced
by ideals of the semigroup.
The possibility of direct summands was explored by Shershin & Moore

[1965]. Hora [1973] studied more general decompositions S = AB where A
and B are proper subsemigroups. Hule [ 1976] and Toffalori [ 1980] considered
analogues of purity. Poyatos [ 1968] derived a formula for the length of maximal
chains of unitary subsemigroups of a finite c.s. Chen [ 1988] studied semigroups
in which every subsemigroup is unitary. Pales [ 1989] studied subsemigroups
admitting a pair of commuting endomorph isms. See also Simon [ 1979].
5. We now tum to ideals. Fruitful similarity with ideals of commutative rings,

particularly in cancellative c.s., has led to Noether-like intersections of primary
ideals (Ward & Dilworth [ 1939], Aubert [ 1962], MacKenzie [ 1954], Johnson
[ 1986], Zhang & Qi [ 1997]) or irreducible ideals (Lesieur [ 1955]), Krull decom-
positions (Perel [ 1957]), a hull-kernel topology for prime ideals (Kist [ 1963]), var-
ious radicals (Aubert [1962], Satyanarayana [1975], Lal [1972R], Ro!z & Schein
[ 1978]), integrally closed semi groups (Maury [ 1958], [ 1959]), Dedekind semi-
groups (Aubert [ 1962], Dorofeeva [ 1972], Mannepalli & Satyanarayana [ 1974],
Jain [ 1982], [ 1983], [ 1989]), hereditary and semihereditary semi groups (Doro-
feeva [ 1972]), and other topics (Peri6 [ 1966], Murata [ 1980], Abrgan [ 1984],
Rachunek [ 1987]). Properties of commutative rings, domains, and valuation rings
were systematically extended to cancellative and power cancellative c.s. by Mat-
suda [1997], [1998A], [1999], Tanabe & Matsuda [1999], Kanemitsu & Ban-
sho [ 1999]; see also Matsuda [ 19988], Kanemitsu, Okabe, & Matsuda [ 1999],
Hirabuki & Matsuda [1999], Kitsuneda, Matsuda, & Sugatani [1999].
Commutative semigroups provide a fine setting for abstract ideal theory, as

carried out by Lorenzen [ 1939] in cancellative c.s. (later generalized in Lorenzen
[1949], [1952]) and by Aubert [1962] in arbitrary c.s. Aubert's theory of x-
ideals includes a Kruii-Stone theorem, a hull-kernel topology for maximal ideals,
intersections of primary ideals, and analogues of Dedekind rings; it has been
continued by Holme [1966], Skula [1975], [1976], Porubsky [1977], and Karasek
2. PRODUCTS AND SUBSETS. 193
[1985].
See also the survey of ideal theory by Anderson & Johnson [1984].
6. Ideal extensions of commutative semigroups of various kinds were studied
by Heuer & Miller [1966] and Heuer [1971], for cancellative semigroups; Ya-
mada [ 1965], [ 1968] and McNeil [ 1971] for null semi groups; Petrich [ 1973 E] for
Clifford semigroups; Arendt & Stuth [1972] for 0-free nilsemigroups. For more
general results, see Putcha & Weissglass [1973], Hildebrant [1984], Takahashi
[1984A], [1985]. Ideal extensions and dense ideal extensions were studied for
commutative semigroups by Gluskin [1971], [1973], [19760], [1976E], [1979],
[19831].
7. The ideals of a c.s. S form a complete distributive lattice which was first
studied by Aubert [1953] and was characterized by Anderson & Johnson [1984].
This lattice is trivial ( = { 0, S}) if and only if S is a group. Porubsky [ 1977],
[ 1978] showed that it is a chain if and only if every ideal of S is prime, if and only
if every ideal of S is primary, if and only if every ideal of S is semi primary. These
semigroups were studied by Lesohin [1964G], Satyanarayana [1971A], [1972],
[1978], Lal [1975], Jain [1988], [1989], and Geroldinger [1996]. Megyesi &
Pollak [ 1968], [ 1977] studied and constructed semigroups in which every one-
sided ideal is principal; when a c.s. S has this property, then S j'J{ is a eo-well
ordered semilattice of infinite cyclic semigroups and cyclic nilsemigroups; Jain
[ 1988] related this property to valuations on the universal group.
An ideal I is categorical when xyz E I implies xy E I or yz E I. (If is e
a small category, then the zero element of the semi group of morphisms U {0} e
is an ideal with this property.) Commutative semigroups in which every ideal is
categorical were studied by McMorris & Satyanarayana [1972], Monzo [1973]
Other conditions have been considered: every primary ideal is prime (Satya-
narayana [ 1971 8], Lal [ 1972P]); every nonzero ideal is prime (Satyanarayana
[19718]); A= B(A: B) for all ideals A,B (Mannepalli [1976A], [19768]);
every ideal is finitely generated (Satyanarayana [1977]); every ideal is a retract
(Tully [ 1969]); every ideal is a finitely generated disjunctive semigroup (Johnson
& McMorris [ 1977]).
8. Other subsets of c.s. have been studied. Rational sets were defined by
Eilenberg & SchOtzenberger [1969]. Iwanik & Plonka [1975] studied linearly
independent subsets (= which satisfy only trivial relations).
The subsets of a commutative semigroup S constitute a commutative semi-
group (under multiplication of subsets), the power semigroup or global of S
(Tamura & Shafer [ 1967]). Power semi groups of various c.s. have been stud-
ied by Tamura & Shafer [ 1967] (abelian groups and chains), Szimtenings [ 1970]
(cancellative c.s.), Mogi1janskaja [1972] (determinacy), Byrd, Lloyd, Pedersen,
& Stepp [1977] (finite cyclic groups), Pedersen & Sizer [1978] (torsion free
abelian groups), Tamura [1984C] (chains), [1984Z], [1985) (Z), Spake [1986A],
[ 1986F], [ 1988Z] ( Z ), [ 1988Q] ( Q ), Sasaki [ 1988] (determinacy), Sasaki, Spake,
& Tamura [1987] (Q), Spake & Hanlon [1993] (Z). Subsemigroups ofthe power
semigroup can be used to construct semi groups of quotients (Toea [ 1977]).
3. HOMOMORPHISMS AND CONGRUENCES.
This section surveys properties of commutative semigroups related to homo-

morphisms and congruences.
1. We begin with homomorphisms into particular c.s.; homomorphisms into
semilattices are an example.
Homomorphisms into N were used by Abe IIanas [ 1965] and Grillet [ 1970E]
to study cancellative, power cancellative, and reduced c.s. Homomorphisms into
Q+ were used similarly by Abe IIanas [ 1965], and in Section III.6 to study power
joined semigroups.
Homomorphisms into m;.+ or m;.+ U { oo} have been studied for existence
(Kaufman [ 1966B], [ 1967], Kobayashi [ 1977C], [ 1977E], Kobayashi & Tamura
[1977]) or for extension from subsemigroups (Kaufman [1966A], Kranz [1974],
[ 1979], Putcha & Tamura [ 1976], Kobayashi & Tamura [ 1977]). Krause [ 1989]
used homomorphisms into m;.+ in factorization theory. Homomorphisms of N-
semigroups into m;.+ were studied by Kobayashi [1974] and Tamura [1982] and
were used by Tamura [1974B] to construct N-semigroups and by Kobayashi
[ 1978] to characterize their Rees quotients. Homomorphisms that preserve a
given compatible preorder (which could be ~JC) were considered by Putcha
[1976], Kobayashi [1980], and Plappert [1995]. Kominek [1990] studied subad-
ditive mappings to JR..
Homomorphisms into groups have been studied by Dubreil [ 1941] and many
of his disciples. Putcha [ 1971] determined c.s. S such that Im 'P is a subgroup
of G for every group G and homomorphism 'P of S into G.
Representations are homomorphisms into more explicit semigroups. Matrix
representations of finite c.s. were studied by Ponizovsky [ 1970], who determined,
in terms of Ponizovsky factors, when there are only finitely many inequivalent
irreducible representations. Increasingly refined representations by products in
various categories were studied by Tmkova [1975A], [1975B], [1976], [1978],

[1981] and Adamek & Koubek [1977]. A commutative Clifford semigroup S
has faithful representations by partial bijections of sets; the minimal size of such
a set was determined for finite S by Byleveld & Easdown [1992]. C.s. were
also represented by compact convex subsets (Ratschek & Schr5der [ 1977]), by
endomorphisms of boolean algebras (Nasirov [1980]), by products of complete
0-dimensional metric spaces (Vinarek [ 1982]), by continuous linear operators on
Hilbert spaces (Ressel & Ricker [ 1998]), and by subsets of simpler c.s. (Lau
[1979], Gould & Iskra [1984R], Gould, Iskra, & Palfy [1986]).
2. When A and B are c.s., we saw that Hom (A, B) is a c.s. under the
pointwise operation. The semigroup Hom (A, B) was studied by Brameret [ 1962]
and, when A and B are commutative Clifford semigroups, by Fulp [1966],
Anderson [1979A], and Kupcov [1979H], [1980], [1981H].
Endomorph isms of a commutative semigroup constitute a monoid under com-
position, a commutative semigroup under the pointwise operation, and a semiring
under both. The monoid of endomorphisms of commutative semigroups was stud-
ied by Lesohin [ 1966H] and described, in terms of archimedean decomposition,
by Nakajima [1977] (see also Manukjanc [1974] for commutative Clifford semi-
groups). Kupcov [ 1979E] showed that a commutative Clifford semi group has only
the obvious endomorphisms x ~ xn if and only if it is a finite cyclic group.
Kublanovsky [1983] showed that every semigroup with at least 4 elements has at
least 4 endomorphisms, and that there are only finitely many nonisimorphic finite
c.s. with exactly 4 endomorphisms.
The determination problem for endomorphisms consists in detecting c.s. S
such that End (S) ~ End (T) implies S ~ T when T is limited to some
class of semigroups. This problem was considered by Trepetin [1971], [1974]
and Boguta [1976]. If for instance S and T are nilpotent and S has only
homogeneous defining relations, then End (S) ~End (T) (as semirings) implies
S ~ T (Trepetin [ 1971 ]).
Bredihin [1976] proposed a generalization of endomorphisms and automor-
phisms. Kozlov [1976] studied rigid cells. Kim & Neumann [1991] studied the
operation x * y = (ax) ((3y) , where a and (3 are commuting endomorphisms.
3. Given commutative semigroups A, B, and C, a bihomomorphism of
Ax B into C is a mapping (3: Ax B----+ C such that (3(aa 1, b)= (3(a,b) (3(a 1,b)
and (3( a, bb') = (3( a, b) (3( a, bb1) for all a, a' E A and b, b' E B. Bicharacters
are bihomomorphisms into the multiplicative submonoid § U {0} of C. Hi-
homomorphisms have been called bilinear mappings and other names but they
are not gay. Under pointwise operation they constitute a commutative semi-
group BiHom (A, B, C) . The latter is in fact a semigroup of homomorphisms:

indeed bihomomorphisms of A x B into C may be regarded as homomor-
phisms of A into Hom ( B, C) and as homomorphisms of B into Hom (A, C) .
Since Head [ 1967] and Grillet [ 1969T] showed that commutative semigroups
have a tensor product, there is also a natural isomorphism BiHom (A, B, C) ~
Hom (A® B, C) ; for instance, bicharacters of A and B are characters of A® B.
Lesohin gave conditions under which BiHom (A, B, C) is nonempty [19670]

(see also Kupcov [1981B], Popyrin [1988]), finite [1967£], has a nonidempotent
element [1969], or is finitely approximable [1976]. Lesohin also studied the par-
ticular case of semiring multiplications on a cancellative semigroup or a group
with zero [ 1964M] and the structure of BiHom (A, B, C) in the case of bichar-
acters [1971B]. Manukjanc [1976A], [1976B] studied multiplications on certain
commutative Clifford semigroups and on certain c.s. with cyclic archimedean com-
ponents. Bicharacters and bicharacter semigroups were investigated by Lesohin
[1969], [1971B] and Popyrin [1986]; see also the survey by Lesohin & Popyrin
[1987].
Any bihomomorphism f3 : A x B ~ C leads to separability questions: does
f3 separate the elements of A (if a I a' in A, is /3( a, b) I /3( a', b) for some
b E B)? and if so, is A maximal with this property, given B and C? Under
the names of regularity and completeness, these questions were investigated by
Lesohin (1958], [1961], [1963M], [1963S], (1964G], [1967£], [1967S], [1969]
and Popyrin [ 1984], [ 1985], [ 1986]. Lesohin also studied duality relative to a
bihomomorphism A x B ~ C [19660] and gave conditions under which the
natural homomorphism A ~ Hom (Hom (A, C), C) is an isomorphism for all or
some commutative semigroups C [19680], [1980] or under which f3: Ax B ~
C induces isomorphisms A~ Hom(B,C) and B ~ Hom(A,C) (Lesohin &
Kupcov [ 1980], Lesohin [ 1992]); weaker conditions were considered by Lesohin
[ 1980], Lesohin & Kupcov [ 1980], Popyrin [ 1985], Kupcov & Lesohin [ 1986],
Ermolina [ 1988], and Bogacheva [ 1988], [ 1989], [ 1990A], [ 1990B]. See also the
survey by Lesohin & Popyrin [ 1987].
4. Other mappings have also been considered. Putcha [ 1976] found good
use for positive mappings (mappings rp into a partially ordered set, such that
rp(xy) ~ rp(x), rp(y) for all x,y). Kowol & Mitsch [1976] studied monomial
mappings x f---+ axn of a c.s.; Tichy [ 1979], [ 1981] studied polynomial mappings
a x~ 1 x~ 2 ... x~k . Hule [ 1976] showed that a system of polynomial equations
which has a solution in a larger c.s. does not necessarily have more than one
solution in some larger c.s.; in which respect c.s. differ from groups and abelian
groups. See also Grossman [ 1982].
5. We now tum to congruences which, on a c.s., have some similarity to

ideals. In fact, an ideal theory of congruences, complete with Noether primary
intersections, was developed by Drbohlav [1963], [1964] and Hoehnke [1964].
Hoehnke [ 1965] also defined the radical of a commutative semigroup as the
congruence s 9< t if and only if sn = tn for all sufficiently large n > 0;
equivalently, s - t is nilpotent in Z[S]. Arendt [ 1975] studied a Jacobson like
radical, the intersection of all maximal modular congruences.
Conditions on congruences have been considered since Lyapin [ 1950S] de-

termined c.s. without proper nontrivial congruences (they are cyclic groups of
prime order or have at most two elements) and c.s. on which all congruences
are Rees congruences [ 1950R] (revisited by Aucoin [ 1995]). Jensen & Miller
[ 1968] characterized c.s. whose every proper homomorphic image is finite; these
are precisely the subsemigroups of Z(p00 ) U {0}. Tamura [ 1969] and Schein
[ 1969] determined all c.s. whose congruences form a chain; these are precisely
the subgroups of Z(p00 ) and the nilsemigroups and nilmonoids that are totally
ordered by divisibility. Hamilton [ 1975] determined all c.s. S whose congruences
commute: either congruences on S form a chain, or S is an elementary semi-
group whose monoid of orbits is totally ordered by divisibility. Schein [ 1981]
determined all c.s. that are isomorphic to all their nontrivial quotients.
Group congruences have attracted attention since Dubreil [ 1941]. Tamura

& Hamilton [ 1972] showed that a minimal group congruence 9 on a c.s. is a
smallest group congruence (then S/9 is the universal group of S and S ---+
G(S) is surjective); a shorter proof was given by Woan [1975]. Szasz [1976]
used congruences similar to principal equivalences (also considered by Tamura &
Hamilton [ 1971]) to study semilattices.
6. A special place belongs to the difficult study of congruence lattices. Hamil-

ton & Nordahl [ 1978] characterized c.s. whose lattice of congruences is boolean.
Hamilton [ 1978] characterized cancellative c.s. whose lattice of congruences is
modular (which consist of abelian groups and of the positive or nonnegative parts
of subgroups of Q) or distributive (which consist of locally cyclic abelian groups
and locally cyclic abelian groups and of the positive or nonnegative parts of sub-
groups of Q ). Tamura & Etterbeek [ 1966] proved distributivity for all locally
cyclic semigroups. The survey by Mitsch [ 1983] has a section on commutative
semigroups; some errors are pointed out in its review MR 84d:20064. Eberhart
[ 1982] studied c.s. whose lattice of congruences is simple; these are either finite
p-groups or nilsemigroups. Hamilton [ 1982] studied archimedean semigroups
whose lattice of congruences is distributive, and separative c.s. whose lattice of
congruences is modular. Trueman [ 1983] studied direct products of finite cyclic
semigroups, whose lattice of congruences is always upper semimodular, but is

lower semi modular only for groups. Trueman [ 1985] also gave examples, includ-
ing free c.s. with at least two generators, whose lattice of congruences is neither
upper semimodular nor lower semimodular.
Affine completeness is the property that every congruence preserving mapping
is polynomial; this was studied by Kaarli, Marki, & Schmidt [ 1985] for semi lat-
tices and by Kaarli & Marki [ 1992], [ 1995] for commutative Clifford semigroups.
7. A tolerance relation on a semi group S is, like a congruence, a reflexive
and symmetric subsemigroup of S x S, but is not necessarily transitive. They
were introduced by Zelinka [1975] who showed that all tolerances on a group are
in fact congruences [ 1975], and that groups are the only periodic c.s. with this
property [ 1977]. Tolerance lattices of c.s. were studied by Pondelfcek [ 1983],
[ 1985A], [ 1985B], [ 1988A], [ 1988B] for various lattice properties; Pondelicek
[ 1988C] also characterized c.s. in which every singly generated tolerance relation
is a congruence. In a similar vein, the reflexive and symmetric binary relations 'J
on S such that a 'J b implies xay 'J xby for all x, y E 8 1 constitute a complete
distributive lattice; Sitnikov [ 1983] characterized c.s. for which this lattice is
boolean. See also Chandran [1987] and Pondelicek [1988D].
Other relations that have been studied include positive preorders (Putcha
[ 1973]) and compatible partial or total order relations, which have been investi-
gated extensively and deserve a survey of their own.
4. OTHER TOPICS.
"Other topics" include extensions, certain constructions, coextensions, cate-

gories, S-sets, and miscellany.
1. Extensions (other than ideal extensions) have been studied by a number
of authors. Gluskin studied dense extensions (such that homomorphisms that are
injective on S are injective on E), normal extensions (such that S is unitary
in E in the sense that ab E S implies b E S when a E S and b E E), and
other kinds [1971], [1973], [1976D], [1976E], [1979], [19831]. A cancellative
semigroup S has a maximal dense normal extension if and only if abn E S
implies b E S when a E S, n > 0, and b E G(S); equivalently, S is its own
complete integral closure (Giuskin [ 1985]).
Embedding into Zn was studied by Parker [1954] and Burgess [1968].
A Stone semigroup is a c.m. S with a zero element but no other nilpotent
4. OTHER TOPICS. 199
element, in which for every annihilator ideal A = Ann (T) = { x E S I xT = 0}

there is an isomorph ism S ~ A x Ann (A) given by x f---t (ex ,fx) , where
e E A, f E Ann (A) are idempotents. Keimel [ 1971] showed that every c.m.
with a zero element but no other nilpotent element has a minimal embedding into
a Stone semigroup.
2. A tensor product for commutative semigroups was developed indepen-
dently by Head [1967A], Grillet [1969], and Nikolaev [1971]. Head [1967A]
showed that tensor products yield the universal semi lattice of S (as S 0 {1} ),
the universal Clifford semigroup of S (as S 0 Z), and the universal uniquely
divisible semigroup of S (as S 0 Q+ ). Tensor products of Clifford semigroups
were studied by Head [1967B], Nikolaev [1971], Kuroki [1973], and Anderson
[1974]; their structure was finally elucidated by Anderson & Kimura [1978],
[ 1979]. Anderson [ 1997], [ 1998] also determined tensor products of Clifford
semigroups generated by partially ordered sets of groups. The archimedean com-
ponents of tensor products were considered by Head [ 1968] and determined by
Anderson [ 1978]. Other results include functor properties (Head [ 19670], Gril-
let [ 1969]), kernels (Head [ 1967C]), maximal subgroups (Head [ 1969], Fulp
[ 1970]), faces (Head [ 1969], Grillet [ 1969]), universal cancellative semigroup
(Kuroki [1971]), and universal group (Head & Kuroki [1974], Li [1986]). Flat-
ness was considered by Grillet [19701], Bulman-Fieming & McDowell [1980],
Head & Anderson [ 1980], and Galanova [ 1982].
Cancellative c.s. and other classes of c.s. also have tensor products; see Gril-
let [1970T], Li [1986], [1987], Galanova [1987], Li & Zhang [1993]. For can-
cellative c.s. and their tensor product, Zhang, Li, & Wang [ 1993] showed that
G (S) ~ Z 0 S; Wehrung [ 1996] showed that S 0 T has the Riesz interpolation
property if S and T do. Fraser & Albert [1984] applied the tensor product of
lattices to semilattices.
3. Semigroups and their homomorphisms constitute a category, and similarly
for commutative semigroups.
Semigroups resemble rings in that monomorphisms of semigroups are injec-
tive, but epimorphisms of semigroups are not necessarily surjective; this also
holds for commutative semigroups. The dominion of a subsemigroup is the in-
tersection of all the equalizers that contain it. This was defined and described
by Isbell [ 1965] for semigroups in general; the same construction applies to c.s.
(Howie & Isbell [ 1967]); the original proofs were improved by Stenstrom [ 1971],
Philip [ 1974], Storrer [ 1976], and especially Higgins [ 1990]. Howie & Isbell
[ 1967] also studied saturated c.s. (= which cannot be epimorphically embedded
in a larger semigroup ), showing, among other results, that finite c.s. are satu-
rated (so that epimorphisms of finite c.s. are surjective). Dominions are useful
in studies of free products with amalgamation (Howie [1968], lmaoka [1976],
Shoji [ 1990]). Khan [ 1982], Hsieh [ 1982], Higgins [ 1983], [ 1985], and the sur-
vey by Higgins [ 1984] contain interesting examples; for instance Hsieh [ 1982]
showed that commutative semigroups which satisfy an identity of a certain kind
are saturated.
A projective commutative semigroup P is usually defined by the property
that every homomorphism P --+ S can be factored through every surjective
homomorphism T --+ S. Free commutative semigroups have this property. A
result of Grillet [ 1969F] implies that a commutative semigroup is projective if
and only if it is free. Projectives have been determined in the smaller category of
commutative Clifford semigroups (Bulman-Fleming & McDowell [1980]).
An injective commutative semigroup I can be defined by the property that
every homomorphism S --+ I can be extended to a homomorphism T --+ I
whenever S is a subsemigroup of T. Schein [ 1981 I] showed that all such semi-
groups are trivial (thereby correcting earlier results by Schein [ 1976] and Gluskin
[ 1976E]). The search for injectives has been more fruitful in smaller categories.
Hancock [ 1960] showed that injective cancellative c.s. coincide with injective
abelian groups (thereby correcting an earlier result of Wiegandt [ 1958]); this
was rediscovered by Li & Liu [ 1992]. Injective separative semigroups coincide
with injective (commutative) Clifford semigroups and have been determined by
Schein [1974].
A number of other categories were considered by Takahashi [ 1987].
4. Rings make a good living acting on abelian groups. Actions of semigroups
on sets (which can be regarded as representations by transformations) have been
studied fairly extensively; few of these results, however, apply specifically to
commutative semi groups. Kil'p [ 1973] showed that every principal ideal of a c.s.
S is projective if and only if S is separative. Murty [1983] extended noetherian
decompositions to S -sets. Poyatos [ 1988] devised an archimedean decomposition
for S -sets. Khovansky [ 1995] studied orbits of free c.s. See also Bogomolov &
Mustafin [ 1989]; Ahsan, Khan, Shabir & Takahashi [ 1991].
Semimodules are commutative semigroups on which a semiring acts in a suit-
able manner (the word has also been applied to plain commutative semigroups).
Their archimedean decompositions were studied by Poyatos [ 1985]; their endo-
morphisms, by Wang [ 1988].
5. We conclude with a number of topics which do not fit easily in any of the
above.
4. OTHER TOPICS. 201
Jezek & Kepka [1975] used uniquely 2-divisible semigroups to construct cer-
tain (non-associative) groupoids. Cho [1990] used c.s. to construct certain medial
algebras. Tamura [ 1977S] studied c.s. with a least archimedean component that
contains no idempotent. Batbedat [ 1978A] [ 1978B] used commutative semigroups
to construct schemes. Measures on c.s. have been studied e.g. by Ebanks [ 1979]
and Bisgaard [ 1998]. Toader [ 1990] studied various kinds of infinite sequences
of elements of c.s.
A holoid, also called naturally totally ordered c.s., is a c.s. on which Green's
preorder ~9-C is a total order relation. Then the ideals form a chain. Holoids have
long been studied as totally ordered c.s. Markov [ 1995] used them for abstract
interval arithmetic. Borisov [ 1992] characterized c.s. which can be embedded
into holoids.
Commutative semigroups have been applied to other areas of Mathematics.
Seitz & Blickle [1974] and Seitz, Blickle & Grega [1975] found that certain
types of semilattices are useful in Chemical Engineering. Duske [1976] studied
commutative automata and there are a number of papers on commutative Thue sys-
tems (see e.g. Simmons [ 1976], Kapur & Narendran [ 1983], [ 1985], Narendran,
O'Dunlaing, & Rolletschek [ 1985], Huynh [ 1986], Narendran & O'Dunlaing
[1989], Yap [1991]). Commutative semigroups have been used to solve func-
tional equations (see e.g. Brillouet-Belluot [ 1996] and Taylor [ 1999]). See also
Warner [1960], Lal [1967], Speed [1968], Moszner [1979], Thron [1981], Mush-
taq & Yusuf [1987], [1988], and Kisielewicz & Newrly [1993].
Chapter IX.
NILSEMIGROUPS.
This chapter gives a construction of finite nilmonoids, which is due to the

author [1991N], [2001N] and expands earlier ideas of Arendt & Stuth [1970A];
a shorter account is given in Grillet [1995]. Unlike previous constructions for
these semigroups, this is a global construction with a very geometric character, in
which nilmonoids are obtained as quotient of free commutative monoids by suit-
able congruences. It accounts well for various structural features of nilmonoids,
such as the greatest pure congruence and the universal group of N\ 0. A more
general construction was given by Grillet [2001N] and applied to fully invariant
congruences in [2001F].
Finite commutative nilsemigroups are very abundant. The construction in
this chapter quickly generates large numbers ofnilsemigroups but without enough
control to, say, prove the commutative Kleitman-Rothschild-Spencer conjecture
(that, on a statistical basis, almost all finite c.s. are nilpotent of index 3). Finding
a more powerful construction would be a major advance.
Inductive constructions of nilsemigroups were devised by Yamada [1964],
[1965], [1968], Yamada & Tamura [1969], and John [1973]. Tamura [1966],
[ 1968] constructed nilsemigroups from trees. Some special classes have also been
constructed by various methods. Tree-ordered nilsemigroups were constructed by
Tully [1966], [1974] and Grillet [1991 T]. Rees quotients of N-semigroups were
characterized by Kobayashi [1978], [1982] and Clarke [1980], [1984]. Locally
cyclic nilsemigroups were constructed by Buzashi & Vishnyakova [1986].
Due perhaps to lack of structural information, commutative nilsemigroups
have not been studied extensively, in spite of their importance. Ideal extensions
of 0-free nilsemigroups were studied by Arendt & Stuth [1970B]. The nilpotency
indices of the elements of a nilsemigroup form a set of integers which was in-
vestigated by Tamura [ 1982N]. Stratification in nilsemigroups was studied by the
author [1995S] and used in an unsuccessful attempt to prove the commutative
Kleitman-Rothschild-Spencer conjecture [1995N].
203
204 IX. NILSEMIGROUPS
1. FREE COMMUTATIVE MONOIDS.
Constructing congruences requires a number of additional properties of free

c.m. (= commutative monoids). This section also contains Redei's general de-
scription of congruences by kernel functions [ 1956].
In what follows, we denote by F or by Fx the free c.m. on a set X, and by
G or Gx the free abelian group on the same set. We write both additively. As in
Chapter I we write the elements ofF as linear combinations a= L:xEX ax x with
nonnegative integer coefficients ax that are almost all 0; we write the elements
of G as linear combinations L:xEX ax x with integer coefficients that are almost
all 0.
1. A face of a c.s. S is a nonempty subsemigroup T of S such that xy E T
implies x E T and yET; equivalently, the complement T = S\P of a proper
prime ideal P of S.
Proposition 1.1. The faces of the free c.m. on a set X are the free c.m.
generated by subsets of X.
Proof. Let Y be a subset of X . Then a = L:xEX ax x E F is in Fy if and
only if ax = 0 for all x E X\ Y. Hence Fy is a submonoid of F. In fact Fy
is a face of F: if a, b E F and a + b E Fy , then ax + bx = 0 for all x E X\ Y,
ax= bx = 0 for all x E X\Y, and a,b E Fy.
Conversely let T be a face of F. Then 0 + t E T for some t E T, since
T-=/::. 0, and 0 E T. Then Y = X n T ~ X and Fy ~ T. If conversely
a = LxEX ax x E T, then ax > 0 implies a - x E F, a = x + (a - x) E T
and x E T n X = Y; therefore a E Fy. Thus T = Fy. 0
By Lemma III.l.l, aN b holds in any c.s. S if and only if a and b belong
to the same prime ideals of S, if and only if a and b belong to the same faces
of S. Hence Proposition 1.1 yields the archimedean components of F:
Proposition 1.2. Y (Fx) is isomorphic to the semilattice of all finite subsets
of X under union; the archimedean component of Fx that corresponds to Y ~ X
is {a = L:xEY ax x E Fy I ax > 0 for all x E Y}.
I
Proof. Let a= L:xEX axx E F and S(a) = {x EX ax> 0}. For every
subset Y of X we have a E Fy if and only if S(a) ~ Y. Hence that a and
b E F belong to the same faces if and only if S(a) = S(b). 0
Y(Fx) is the free semilattice on the set X; every mapping of X into a
semilattice Y extends uniquely to a homomorphism of Y (Fx) into Y.
1. FREE COMMUTATIVE MONOIDS. 205
When A = Fy is a face of F, with Y ~ X, then A' = FX\Y is the

complementary face of Fy. The corresponding abelian groups Gy and G X\Y
provide a direct sum decomposition G = Gy EB G X\Y of G. There is a very
similar decomposition of F:
Proposition 1.3. Let A= Fy be aface of Fx. Every element a of Fx can
be written uniquely in the form a= pAa+pA_a with pAa E A, pA_a E A'= FX\Y;
namely, PA a = L:xEY ax x and PA a = L:xEX\ y ax x.
PA and pA_ are the projections of a on the faces A and A'. It follows from
Proposition 1.3 that F ~ A x A' .
2. There is a natural partial order on G: when a = L:xEX ax x and
b = L:xEX bxx E G, then
a ~ b if and only if ax ~ bx for all x, if and only if b - a E Fx.
If a, b E F, then a ~9-C b in F if and only if a ~ b.
Proposition 1.4. Gx is a lattice ordered abelian group.
Proof. Suprema and infima in G are pointwise: for all a = L:xEX ax x and
b = L:xEX bxx E G,
aVb = L:xEXmax(ax,bx)x and a/\b = L:xExmin(ax,bx)x.
The equalities
(avb)+c = (a+c)V(b+c), (a/\b)+c (a+c)/\(b+c)
hold for all a,b,c E G. D
We see that F is a sub lattice of G. Note the following useful properties:
Lemma 1.5. For all a, b, c E Gx:
(avb)+c = (a+c)V(b+c), (a/\b)+c = (a+c)/\(b+c)

-(a 1\ b) = (-a)= (-a) 1\ (-b),
V (-b), - (a V b)
c-(a/\b) = (c-a)V(c-b), c-(avb) = (c-a)/\(c-b).
The positive and negative parts of a E G are:
a+= aVO and a- = (-a)VO.
By Lemma 1.5,
Lemma 1.6. For all a E Gx, a+,a- ~ 0, a+ /\a-= 0, and a= a+ -a-.
3. We complete this section with the description of all congruences on F,

due to Redei [ 1956], which arises from the following observations.
Lemma 1.7. Let e be a congruence on Fx. Let
R = R(e) = {rEGxlr=a-bf orsomea,bEF xsuchthataeb }.

For every r E R, let
f(r) = Je(r) = {c E F I c+r+ e c+r-}.
Then R is a subgroup of Gx ; every f (r) is a nonempty ideal of Fx ; and
ae b if and only if a - b E R and a 1\ b E f (a - b).
Proof. Since e is a congruence it is immediate that R is a subgroup of G
and that f (r) is an ideal of F.
Let a, b E F and r = a - b E G. By Lemma 1.5,
a-((a-b)VO) b 1\ a, and
b-((b-a)VO) a 1\ b.
Hence a e b implies r =a-bE R, (a 1\ b)+ r+ =a e b = (a 1\ b)+ r-, and

a 1\ bE f(r). In particular, f(r) =f 0 when r E R. If conversely r =a-bE R
and a 1\ b E f (r) , then a = (a 1\ b) + r + e (a 1\ b) + r- = b. 0
In Lemma 1.7, R is the Redei group of e and f is the kernel function of
e; f is a mapping of R into the set of all non-empty ideals of F. By Lemma
1.7, e is completely determined by its Redei group and kernel function.
Lemma 1.8. The kernel function of a congruence e on F has the following
properties:
(1) f(O) = F;
(2) f( -r) = f(r) for all r E R;
(3) (r+ + f(r)) n (s+ + f(s)) ~ (r V s) + f(r- s), for all r,s E R.
Proof. Properties (I) and (2) are immediate since e is reflexive and symmet-
ric. Property (3) expresses the transitivity of e. Let c E (r+ + f(r)) n (s+ +
f (s)) . Then c - r +, c - s + E F, c - r = c - r+ + r- E F, and c - s E F.
By definition of f(r), c = c- r+ + r+ e c- r+ + r- = c-r. Similarly
c e c - s . Hence c - r e c - s and
c-(rVs) = (c-s)/\(c-r) E f(r-s)
by Lemma I. 7, which proves (3 ). 0
1. FREE COMMUTATIVE MONOIDS. 207
Theorem 1.9. Let Fx be the free commutative monoid on a set X, R be

a subgroup of the free abelian group on X, and f assign to each r E R a
non-empty ideal f(r) of Fx, so that properties (1), (2), and (3) in Lemma 1.8
hold. Then the binary relation e on Fx defined by
a eb -{=:=:} a-bE R and a 1\ b E f(a- b) (C)
is a congruence on Fx with Redei group R and kernel function f. Moreover
every congruence on Fx can be constructed in this fashion.
Proof. It follows from Lemmas 1. 7 and 1.8 that every congruence on F can
be constructed as in the statement. Now let R and f have properties (1), (2),
and (3); define e by (C).
e is reflexive by (1) and symmetric by (2). Now assume a e band be c, so
that r = b- a E R, s = b- c E R, a 1\ b E f (r) = f ( -r), and b 1\ c E f ( s) . Then
a - c = s - r E R. As in the proof of Lemma 1. 7, b - r + = b - ( (b - a) V 0) = a 1\ b
and b - s + = b - ( (b - c) V 0) = c 1\ b, so that b E (r + + f (r)) n (s + + f (s)) .
By (3), bE (r V s) + f(r- s). But r V s = (b- a) V (b- c)= b- (a 1\ c).
Hence a 1\ c = b - (r V s) E f (r - s) = f (a - c) , and a e c. Thus e is
transitive.
If a e b, then, for all c E F, (a + c) - (b + c) = a - b E R,
(a+c)l\(b+c) = (al\b)+c E f(a-b)+c s;;; f(a-b)
since f(a- b) is an ideal ofF, and a+ c e b +c. Thus e is a congruence.
By definition a e b implies a-bE R. Conversely let r E R. Since f(r) of 0
there exists c E f (r) , and for any such c we have a = c + r + e c + r- = b,
since a- b = r+- r- =rand al\b = c+ (r+ 1\ r-) = c E f(r). This shows that
R(e) = R, and that f(r) s;;; fe(r) for all r E R. If conversely c E fe(r), then
c + r + e c + r- and c = ( c + r +) 1\ ( c + r-) E f ( (c + r +) - (c + r-)) = f (r) .
Thus f = f e is the kernel function of e. 0
For an example we calculate the kernel function of a Rees congruence. First
define
S: a = {x E F I a+ xES}
for any subset S of F and a E F. We see that S is an ideal of F ( x E S
implies x + y E S for all y E F) if and only if S s;;; S: a for all a E F.
Let :J be the Rees congruence of a nonempty ideal I ofF. Let a E I. For
any r E G we have a + r +, a + r- E I and r = (a + r +) - (a + r-) E R.
Thus the Redei group of :J is G. If r of 0, then r+ #r-and c E f(r) if and
only if e + r+ :J e + r-, if and only if e + r+, e + r- E I. Thus the kernel

function of :J is
f(r) = (I: r+) n (I: r-)
for all r E G\0, with f(O) = G.
Much more information on kernel functions is given in Redei [ 1956]. This
includes a classification of congruences by various ranks, and the first proof of
Redei's Theorem that congruences on a free f.g.c.s. satisfy the ascending chain
condition.
2. THE ZERO CLASS.
A nilmonoid congruence is a congruence whose quotient semigroup is a

nilmonoid. The first step in the construction of these congruences determines the
equivalence classes which yield the identity and zero elements of the quotient.
I. In a commutative nilsemigroup N, Green's preorder ~J{ is a partial
order relation on N, which we denote by just ~, and xy < x whenever x =/=
0 (Proposition IV.3.1 ). These properties extend to any commutative nilmonoid
N 1 =NU{1}.
When S is a monoid we abuse the language by calling a surjective homo-
morphism 1r : Fx ----+ S a presentation of S; then 1r(X) generates S and an
actual presentation can be obtained by finding generating relations for the con-
gruence induced by 1r . Conversely every generating subset X of S gives rise to
a presentation 1r : Fx ----+ S of S.
For finite nilmonoids there exists a most economical presentation. An element
a of a semigroup S is irreducible when a= xy implies x =a or y =a; if S
is a monoid we also require a =/= 1.
Proposition 2.1. Let N 1 be a finite commutative nilmonoid An element
a of N 1 is irreducible if and only if a is maximal in N, if and only if a E
N\N 2 . Moreover, N 1 is generated (as a monoid) by its irreducible elements,
and therefore has a smallest generating subset N\ N 2 .
Proof. Let a E N. If a is not irreducible in N 1 , then a = xy for some
x, y E N 1 , x, y =/= a, and a < x =/= 1 is not maximal. If a is not maximal, then
a < b E N, a = be for some e E N, and a E N 2 . If finally a E N 2 , then
a = be for some b, e E N, which in the nilsemigroup N implies a < b, c and
2. THE ZERO CLASS 209
b, c -=/= a; thus a is not irreducible. This proves the first part of the statement.
An element of N\N 2 cannot be written as a product of other elements of

N and must belong to every generating subset of N 1 . It remains to show that
N\N 2 generates N 1 . This is proved by noetherian induction. If N 1 is not
generated (as a monoid) by its irreducible elements, then N 1 has an element
which is not a product of irreducible elements and has an element m which is
maximal with this property. This includes empty products and one term products,
so that m -=/= 1 and m is not irreducible. Therefore m E N 2 and m = xy for
some x,y EN, x,y-=/= m. Then m < x,y; by the choice of m, x andy are
products of irreducible elements of N, and then so is m = xy, which is the
required contradiction. D
When N is finite, the set X= N\N 2 provides the most economical homo-
morphism 1r : Fx --+ N 1 ; we refer to this as the standard presentation of N 1 .
The results in Grillet [ 1991 N] are stated in terms of this standard presentation.
This restriction is unnecessary; in fact we do not even assume that F is finitely
generated in what follows.
2. A surjective homomorphism 1r ofF onto a nilmonoid N 1 sends 0 E F

onto 1 E N 1 and generally sends a= L.xEXaxx E F to 1r(a) = fixEX 1r(xtx E
N1 . A congruence e on F = Fx is a nilmonoid congruence on F when FIe
is a nilmonoid; equivalently, when e is induced by a surjective homomorphism
1r ofF onto a nilmonoid N 1 . Then Fie~ N 1 ; the isomorphism sends thee-
class Ca of a E F to 1r(a) E N 1 . If N 1 is finite, then e is a finite nilmonoid

congruence. If F is finitely generated (if X is finite), then every nilmonoid
congruence on F is a finite nilmonoid congruence.
The identity class of a nilmonoid congruence e is the e-class U = C0 of
0 E F; if e is induced by 1r: F--+ then U =
N 1, 1r- 1 (1). We call a nilmonoid
congruence reduced when its identity class is trivial.
Proposition 2.2. When e is a nilmonoid congruence Fx, then the
on
identity class U of e is a face of Fx ; the restriction eu of e to U' is a reduced
nilmonoid congruence; a e b if and only ifr/ua eu rJub; and Fie~ U'le[;-.
Proof. {1} is a face of N 1 , since xy -=/= 1 when x E N or y E N. Hence

U = 1r- 1 (1) is a face of F. Moreover, 1r(a) = 1r(r/ua) for all a E F, since
1r(c) = 1 for all c E U; hence a e rJua and a e b if and only if rJua eu rJub.
In particular, FIe ~ U' I eu- ' and e[;- is reduced, since a eu- 0 with a E U'
implies a E U n U' and a = 0. D

Thus we may assume when constructing nilmonoids that e is reduced.
3. An ideal J of F is a nilmonoid ideal when the Rees quotient F I J is a
nilmonoid, in particular J =1- F is a proper ideal. An ideal J of F is cofinite
when its complement F\J is finite.
The zero class of a nilmonoid congruence e on F is the zero element J of
FIe; if e is induced by 7r : F ---7 N 1 ' then J = 7r - 1 ( 0) .
Proposition 2.3. Let ebe a reduced nilmonoid congruence on Fx. The
e
zero class J of is a nilmonoid ideal of Fx ; if Fx is finitely generated, then
J is a proper cofinite ideal of Fx .
Proof. J = 1r - 1 ( 0) is a non empty ideal of F, since {0} is an ideal of N 1 ,
and J =f. F, since 0 r:f. J.
For every x EX we have 1r(x) EN, since e
is reduced, 1r(x)kx = 0 in
N for some kx > 0, and kxx E J. Let a = I:xEX ax x E F. If a =1- 0, then
ax > 0 for some x and kxa E J. Hence F I J is a nilmonoid.
Since a E J whenever ax ~ kx for some x, we have
F\ J s;;; {a E F f ax < kx for all x}.

If X is finite, then so is F\ J. D
4. Sometimes the ideal J suffices to construct N 1 . A commutative nilmonoid
is 0-free when it is isomorphic to the Rees quotient of a free commutative monoid
by a nilmonoid ideal. 0-free nilmonoids are also called unique factorization
nilmonoids, due to the following result:
Proposition 2.4. A finite commutative nilmonoid N 1 is 0-.free if and only
if there exists a subset S of N such that every nonzero element of N 1 can be
written uniquely as a product of nonnegative powers of distinct elements of S.
Proof. Let N 1 = F I J be 0-free and 1r : F---+ N 1 be the projection. Every
element ofF can be written uniquely as a linear combination a = I:xEX ax x of
elements of X. If ax > 0 for some x E J, then a E J; hence every element of
F\ J can be written uniquely as a linear combination a = I:xEY ax x of elements
of Y = X\J and every nonzero element of N 1 can be written uniquely as a
product of nonnegative powers 1r(a) = ITxEY 1r(xtx of elements of S = 1r(Y).
Conversely assume that there exists a subset X of N such that every nonzero
element of N 1 can be written uniquely as a product of nonnegative powers of
2. THE ZERO CLASS 211
distinct elements of X. Let 1r : F = Fx --+ N 1 send L:xEX ax x E F to

ITxEX x ax ; let J = 1r - 1 ( 0) . The hypothesis states that 1r is injective on F\ J,
so that N 1 is isomorphic to the Rees quotient F I J. D
A homomorphism <p between nilsemigroups or nilmonoids is pure in case
= {0} ; equivalently, in case ker <p is a pure congruence.
<p - 1 ( 0)
Corollary 2.5. Every (finite) commutative nilmonoid is a pure homomorphic
image of a (finite) 0-:free nilmonoid
Proof. By Proposition 2.2 a commutative nilmonoid N 1 has a presentation
e
1r: F--+ N 1 for which = ker1r is reduced. Then the zero class J = 1r- 1 (0)
of e is a nilmonoid ideal ofF by Proposition 2.3, and F I J is a 0-free nilmonoid.
If N 1 is finite, then we may use the standard presentation, and then J is a proper
cofinite ideal of F and F I J is finite.
Since the Rees congruence of J is contained in e,
1r : F --+ N 1 factors
through the projection p : F --+ F I J: there is a surjective homomorphism
<p : FIJ --+ N 1 such that 1r = <pop. If a E F\J, then 1r(a) I= 0 and
<p(a) = 1r(a) I= 0; thus <p is pure. D
Corollary 2.5 is due to Arendt & Stuth [ 1970A].
5. When J is an ideal ofF, a J-congruence is a congruence on F modulo
which J is a class.
Proposition 2.6. When J is a nilmonoid ideal of a .free c.m. F, then every
J-congruence is a reduced nilmonoid congruence. Conversely every commutative
nilmonoid is isomorphic to Fje for some .free c.m. F, nilmonoid ideal J of F.
and J-congruence e.
Proof. As before a commutative nilmonoid N 1 has a presentation 1r : F --+
N 1 such that e = ker 1r is reduced. Then the zero class J = 1r- 1 (0) of is a e
nilmonoid ideal ofF by Proposition 2.3, and e is a J-congruence. Conversely
let J be a nilmonoid ideal of F and e
be a J-congruence on F. Since J
is a non empty ideal, the e-class J is a zero element of FIe. When a E F,
a I= 0, then na E J for some n > 0, since F I J is a nilmonoid; hence FIe is
a nilmonoid. Moreover e is reduced: if a I= 0 and a e 0, then na e 0 for all
n > 0 and 0 E J, a contradiction since the nilmonoid ideal J is proper. D
Proposition 2.7. A proper cofinite ideal J of a .free c.m. F is a nilmonoid
ideal of F, and every J -congruence is a finite reduced nilmonoid congruence.
Conversely every finite commutative nilmonoid is isomorphic to FIe for some
free c.m. Fx, proper cofinite ideal J of Fx, and J-congruence e; moreover
it may be assumed that Fx is finitely generated and that {x} is a e-class for
every x E X, and then Fx, J, and e are unique up to isomorphism.
Proof. A proper cofinite ideal J of F is a nilmonoid ideal: if a E F, a i- 0,
then a, 2a, ... , na, . . . are all distinct and cannot all be contained in the finite set
F\ J; hence na E J for some n > 0, and F I J is a nilmonoid. By Proposition
2.6, every J -congruence e is a reduced nilmonoid congruence; moreover FIe
is finite, since a e-class is either contained in F\J or equal to J.
Conversely every finite commutative nilmonoid N 1 has a standard presenta-
tion N 1 ~ Fx 1e. Then X = N\N 2 is finite and the zero class J of e is a
proper cofinite ideal of Fx by Proposition 2.3. Also the projection 1r: F---+ N 1
is injective on X and {x} is a e-class for every x E X.
Let Fx be any finitely generated free c.m., J be a proper cofinite ideal of
Fx , and e be a J -congruence such that Fx I e ~ N 1 . If every {X} is a e-
class, then 1r is injective on X and N\N 2 contains every n(x), since equalities
n(x) = n(a)n(b) with a,b i- 0 are not possible. Since n(X) generates N 1 it
follows from Proposition 2.1 that n(X) = N\N 2 . Thus 1r induces a bijection
X ---+ N\N 2 and an isomorphism Fx ---+ F N\N2, which takes Fx and e to
the standard presentation of N 1 . 0
3. CORNER POINTS.
This section investigates how J-congruences are determined by the ideal J.

1. The Rees congruence of a non empty ideal I is the smallest congruence for
which I is an equivalence class. The largest such congruence was first described
by Teissier [1951].
Proposition 3.1. Let I be a nonempty ideal of a c.s. S. There is a largest
congruence e on S such that I is a e -class; for all a, b E S,
ae bif and only if I : a = I : b.
Proof. Recall that I: a = {X E 8 1 I ax E I} ~ 8 1 . e is a congruence
since I: a = I: b implies I: ac = (I: a) : c = (I: b) : c = I: be. Also a E I
if and only if I: a = 8 1 , since I is an ideal; therefore I is a e-class.
Conversely let e be a congruence on S such that I is an e-class. Let a e b.
e
For every x E 8 1 , ax and bx are in the same -class; hence ax E I if and
3. CORNER POINTS. 213
only if bx E I, and I : a = I : b. Thus £ ~ e. 0

Proposition 3.1 provides a greatest J-congruence 'D (not to be confused with
Green's relation 'D) for every ideal J of a free c.m. F.
Proposition 3.2. Let J be a nilmonoid ideal ofF and e be a J -congruence
on F. The greatest pure congruence on FIe is the congruence induced by the
greatest J -congruence 'D.
Proof. Let P = 1r('D) be the congruence on Fie induced by 'D :2 e (where
1r : F ---+ F 1e is the projection); then 1r-l (P) = 'D (Proposition 1.2.6). Let
j E J. P is a pure congruence since 'D and e
are J -congruences: indeed
1r( a) P 1r(j) = 0 if and only if a 'D j, if and only if a E J, if and only if
1r(a) = 0. If A is a pure congruence on Fie, then 1r- 1 (A) is a J-congruence
on F, 1r- 1 (A) ~ 'D = 1r- 1 (P), and A~ P. 0
By Propositions IV.3.6 and V.5.2, a finite commutative nilmonoid F 1e is
subdirectly irreducible if and only if the greatest pure congruence on FIe is the
e
equality. By Propositions 3.2 this is equivalent to = 'D. Thus finite subdirectly
irreducible nilmonoids are determined by proper cofinite ideals.
2. The next result gives a sharper description of the greatest J -congruence
'D when J is a proper cofinite ideal of F. Then H = F\J is finite and the
set K = K(J) of all maximal elements of H is nonempty; in fact, for every
a E H there exists some k E K such that a ~ k. The elements of K are
the corner points of H, and also of J and of every J -congruence. The corner
points determine
H = {a E F Ia ~ k for some k E K},
J = F\H, and the greatest J-congruence:
Proposition 3.3. Let J be a proper cofinite ideal of F. The greatest
J -congruence 'D on F is given by
a 'D b -¢:::::::? J :a = J :b -¢:::::::? H :a = H :b -¢:::::::? K : a = K : b,
where H = F\J and K is the set of all maximal elements of H.
In particular, K is a 'D-class, the 'D-class of all a E F such that K: a= {0}.
Proof. Since H is the complement of J, H : a is the complement of J : a;
hence J : a = J : b if and only if H : a = H : b.
Assume H : a = H : b. If a + t E K, so that a + t is a maximal element of
H, then b + t E H and b + t is maximal in H, otherwise b + t + u E H for
some u > 0, a + t + u E H, and a + t is not maximal in H; hence b + t E K.
Similarly b + t E K implies a+ t E K. Thus K: a = K: b. Conversely assume
K : a = K : b. If a + t E H, then since H is finite we have a + t ~ k for some

k E K, a + t + u E K for some u E F, b + t + u E K, and b + t E H, since
b + t ~ b + t + u. Similarly b + t E H implies a+ t E H. Thus H :a = H : b. D
Corollary 3.4. If IKI = 1, then the Rees congruence of J is the only
J -congruence.
Proof. If K = {k}, then H = { t E F I t ~ k} and K : a = {k - a} if
a E H, otherwise K : a = 0 . Therefore 'D is the Rees congruence of J, and
this the only J -congruence. D
3. We illustrate the above with some examples.
Example 3.5. Let X = {x, y} and J be the ideal generated by 4x and
3y. Then a = ax x + ay y E F is in J when ax ~ 4 or ay ~ 3, and
a= axx+ayy E H = F\1 when ax~ 3 and ay ~ 2. Thus H = H(k),
where k = 3x + 2y, and K = {k}. By Corollary 2.8, 'D is the Rees congruence
of J; F/'D is the 0-free nilmonoid ( x,y I x 4 = y 3 = 0). D
When X= {x, y}, an element a= ax x + ay y ofF can be represented by
a point (ax, ay) with nonnegative integer coordinates. In the examples we draw
only the elements of H. Thus example 3.5 is pictured by:
.k
Example 3.5
Example 3.6. Let X = {x, y} and J be the ideal generated by 8x, 5x + 4y,
and 6y. Then a = ax x + ay y E F is in J when ax ~ 8, or ay ~ 6, or ax ~ 5
and ay ~ 4; and a E H = F\ J when ax ~ 7, ay ~ 5, and (ax ~ 4 or ay ~ 3 ).
Thus H has two maximal elements k = 4x + 5y and l = 7x + 3y. We see

that a = 2x + 4y 'D 5x + 2y = b: indeed a ~ k, a ~ l, b ~ k, b ~ l, and
k - a = l - b, so that K : a = {k - a} = K : b. But neither a - x 'D b - x nor
a - y 'D b - y hold. Moreover, x + 5y = a - x + y 'D b - x + y = 4x + 3y
does not hold: indeed x + 5y ~ k, x + 5y ~ l, 4x + 3y ~ k, and 4x + 3y ~ l,
so that IK: (x + 5y)l = 1 but IK: (4x + 3y)l = 2. Similarly, 3x + 3y =
4. NESTS 215
.k
Example 3.6
a- y + x 1) b-y+ x = 6x + y does not hold. Hence 1) is generated (as a

]-congruence) by (2x + 4y, 5x + 2y) and F/D is the commutative nilmonoid
F/D ~ ( x,y I xs = x5y4 = y6 = 0, x2y4 = x5y2 ). D
Example 3.7
Example 3. 7. Let X = { x, y} and J be the ideal of all a E F such that

II all ~ 8, where Ilax x + ay Yll = i + j. Then a E H if and only if llall ~ 7,
a E K if and only if II all = 7, and there are 8 comer points. When a E H
we have H : a = { t E F I II til ~ 7- llall}; hence a D b if and only if either
llall = llbll or a, b E J, and F /D is the cyclic nilmonoid ( x I x8 = 0). D
4. NESTS.
This section constructs nilmonoid congruences in terms of nests of equivalence

relations, as in Grillet [2001N], and brings the chapter's main result.

1. Let J be a proper cofinite ideal of a free c.m. F. Proposition 3.3 shows
that the corner points of a J -congruence e put very strict upper limits on the
size of e. If indeed a e b and a, b r:J. J, then a ~ m for some m E K and
K: a= K: b implies that there exists t E F such that a+ t, b + t E K; in
other words (a, b) has the form (k- t, l- t) for some k,l E K and t E F and
is translated down from a pair (k, l) of corner points.
Thus a J-congruence e is determined by: (I) K (which determines H and
J); (2) which pairs of corner points are in e; and (3) how far in e can each
pair of corner points be translated down. This is the main idea behind the present
construction of nilmonoid congruences.
2. Let J be a proper cofinite ideal of a free c.m. F. The support of a
J -congruence e is the restriction of e to its set K of corner points; it specifies
which pairs of corner points are in e.
Proposition 4.1. When J is a proper cofinite ideal ofF, every equivalence
relation on K ( J) is the support of some J -congruence on F.
Proof. Given an equivalence relation e on K, define e by:
a e b if and only if a = b, or a, b E K, a e b, or a, b E J.
e is a congruence, since a, bE K implies either a+ c =a, b + c = b, or a+ c,
b + c E J. We see that e is a J-congruence with support e. 0
The support of e determines the universal group of (Fie)\0. This in turn
determines partial homomorphisms into abelian groups. The following result was
also anticipated by Arendt & Stuth [ 1970B].
Proposition 4.2. Let J be a proper cofinite ideal of Fx disjoint from X and
e be a reduced J -congruence on Fx with support e. The universal group of
( Fx I e)\ 0 is isomorphic to GX I R, where R is the subgroup of GX generated
by all k - l with k l.e
R is similar to the Redei group of e except that we only use elements of K
to construct R. Note that N\0 and N 1 \0 have the same universal abelian group,
and that when the finite nilsemigroup N is given it can always be arranged that
J be disjoint from X, by Proposition 2.7.
Proof. We use Proposition III.3.4. Let N 1 = Fie and 1r : F --+ N 1 be
the projection. We have 1r(X) ~ N since eis reduced (Proposition 2.7). Let
e' = { (a, b) E H X H I a e b} be the restriction of e to H. As a monoid
with zero, N 1 is generated by X subject to all relations a = b ( (a, b) E e') and
4. NESTS 217
t = 0 (t E J). We have a= b -=J 0 in N 1 for all (a,b) E e'.

Also x -=J 0
in N1 for all x E X, since X n J = 0. By Proposition III.3.4, G(N 1 \0) is
isomorphic to the abelian group generated by X subject to all relations a = b
((a,b) E e'). Thus G(N 1\0) ~GIS, where Sis the subgroup of G generated
by all a- b with (a, b) E e'.
When (a,b) E e', then a e b, a,b E H, a~ k for some k E K, and
a = k- s for some s E F. Then s E K: a; by Proposition 3.3, s E K: b and
l = b + s E K. Also k e l, since a e b, and k c l. Thus a - b = k - l E R.
Therefore S ~ R. If conversely k, l E K and k c l, then k e l, (k, l) E e',
and k - l E S. Therefore R ~ S. Thus R = S. 0
Example 4.3. Let N 1 be the Volkov nilmonoid (Example 111.3.6)
N = ( x,y I x 3 = x 2 y = xy 2 = y4 = 0, x 2 = xy = y 3 ).
In the standard presentation N 1 ~ Fx I e,

X = {x, y} and the zero class of
e is the ideal J of Fx generated by 3x' 2x + y' X + 2y' and 4y. Thus H
consists of 0, x, 2x, y, x +y, 2y, and 3y; and K = {k,l,m}, where k = 3y,
l = x + y, and m = 2x.
.k
.m
Example 4.3
The presentation N 1 = Fx 1e shows that the support of e is the univer-

sal equivalence relation on K. By Proposition 4.2, G (N 1 \0) is isomorphic
to GxI R, where R is the subgroup of Gx generated by k - l = 2y - x,
k - m = 3y - 2x, and l - m = y - x. Then y = (2y - x) - (y - x) E R,
x = y- (y- x) E R, R = Gx, and G(N 1 \0) is trivial (as in Example IV.3.6).
We also have K : 0 = K;
K: x = { l - x, m- x} = { x, 2y };
K: y = { k- y, l - y} = {X+ y, 2y };
K:(x+y) = {l-x-y}={y};
K:2y = {k-2y,l-2y} = {x,y};
and K: k = K: l = K: m = {0}. Hence 'D = e, confirming that the Volkov

nilmonoid is subdirectly irreducible, as noted after Corollary IV.3.7. D
3. We call a subset H of F a coideal when its complement F\H is an
ideal; equivalently, when x ~ y E H in F implies x E H. These subsets are
called filters in Grillet [ 1991 N] and have also been called order ideals, lower
sections, and other possibly insulting names.
As noted above, a J -congruence e is determined by: (I) its corner points
(which determine J); (2) which pairs of corner points are in e; and (3) how
far in e can each pair of corner points be translated down. Classifying (3) by
pairs of corner points yields the co ideal sets in Grillet [ 1991 N]. The author has
come to prefer classification by translations, as in Grillet [2001N], even though
[ 1991 N] has additional properties of coideal sets.
Let e be a J -congruence, where J is a proper cofinite ideal of a free c.m.
F. We saw that every pair (a, b) of elements of H with a e b has the form
(k- t, l - t) for some k,l E Kandt E F; actually, t E H, since t ~ k,l. For
every t E H define an equivalence relation et on K by:
k et l if and only if k = l' or k, l ~ t and k - t e l - t.
Thus et specifies which pairs of corner points can be translated down by t
(within e). For instance e0 is the support of e, but et is the equality on K
whenever t E K. In general:
Lemma 4.4. When e is a J -congruence, then a e b if and only if either
a, b E J, or a, b E H and a + t et b + t for some t E H.
The family N(e) = (et)tEH is the nest of equivalence relations of the
J-congruence e. Similarly N('D) = ('Dt)tEH is the nest of equivalence relations
of the greatest J -congruence 'D .
Lemma 4.5. When e is a J -congruence, then every et is an equivalence
relation on K and, for every t, u E H:
(1) t ~ u implies eu ~ et;
(2) et ~ 'Dt;
(3) if k,l,m,n E K and k -l = m- n, then, for all t E H,
k ep+t l if and only if m eq+t n,

where p = k - k 1\ m and q = m - k 1\ m.
The proof shows that k, l ~ p + t if and only if m, n ~ q + t in (3).
Proof. et is an equivalence relation like e.
4. NESTS 219
(I) Ift ;£ u, k eu l, and k oj:.l, then k,l ~ u ~ t, k-u e l-u,

k-t = k-u+(u-t) e l-u+(u-t) = l-t
since e is a congruence, and k et l.
(2) follows from e ~ 'D.
(3) Let k,l,m,n,p,q be as in (3). Then k- p = k 1\ m = m- q and
l-p=n-q (since k-l=m-n). Hencep+t;£k,l ifandonlyifq+t;£m,n,
and then k- p- t e l- p- t if and only if m- q- t e n- q-t. Thus
k ep+t l if and only if m eq+t n. 0
4. A coherent nest of equivalence relations on K is a family (et)tEH of
equivalence relations on K with properties (1 ), (2), and (3) in Lemma 4.5.
Theorem 4.6. Let F be a free commutative monoid, J be a proper cofinite
ideal of F, and ( et) tEH be a coherent nest of equivalence relations on K (J).
Then the binary relation e on F defined by: a e b if and only if
a,b E J, or a,b E Hand a+t et b+tforsome t E H
is a J-congruence on F whose nest of equivalence relations is (et)tEH· More-
over, every J -congruence on F can be constructed in that fashion.
Proof. By Lemma 4.5, every J-congruence on F can be constructed as in
the statement. For the converse we prove:
Lemma 4.7. Let (et)tEH be a coherent nest of equivalence relations on
K = K(J). ljt,u E H, b+u eu c+u, and b+t E K, then c+t E K and
b + t et c + t.
Proof. Let t,u E H, b+u eu c+u, and b+t E K. Then b+u 'Du c+u
by (2) and b 'D c, by Lemma 4.5 applied to 'D. Hence b + t E K implies
c + t E K. We may now assume that b i- c. Then b + t, c + t, b + u,
c + u E K and (b + t)- (c + t) = (b + u)- (c + u) i- 0; by (3),
b + t ep+s c + t <¢:::=::} b + u eq+s c + u
for every s E H, where p = (b + t) - (b + t) 1\ (b + u) = t - t 1\ u and
q= (b+u)-(b+t)l\(b+u) =u-tl\u. With s=tl\u, b+u eu c+u
then yields b + t et c + t. 0
Now let (et)tEH be a coherent nest of equivalence relations on K = K(J).
The relation e in the statement is reflexive and symmetric, since every et is an
equivalence relation. Now assume a e b e c. We may assume that a, b, c E H
(otherwise a e c already holds). Then a+ t et b + t and b + u eu c + u for
some t, u E H, b + t et c + t by Lemma 4.7, a+ t et c + t, and a e c.

Thus e is transitive.
Assume a e b and let c E F. If a + c E J, then b + c E J (since e ~ 'D
by (2)) and a + c e b + c. Now assume a + c, b + c E H. Then a, b E H and
a+ t et b + t for some t E H. Since b + c E H we also have b + c ~ k for some
k E K and b + c + u E K for some u E H. Then a+ c + u ec+u b + c + u
by Lemma 4.7, a+ c + u eu b + c + u by (1), and a+ c e b +c. Thus e is
a J -congruence.
Let k, l E K, t E H, t ~ k, l. Assume k - t e l- t. Then k - t +
u eu l- t + u for some u E H. By Lemma 4.7,
k = k - t + t ep+s l - t + t = l ~ k- t +u eq+s l - t + u
for every s E H, where p = k - (k 1\ ( k - t + u)) and q = ( k - t + u) - ( k 1\
(k-t+u)). Now k-t ~ kl\(k-t+u),
u = (k-t+u)-(k-t) ~ (k-t+u)-(kl\(k-t+u)) = q,
and u - q = (k 1\ ( k - t + u)) - (k - t) = t - p. With s = t - p

u- q, k- t + u eu l- t + u and (3) yield k et l. If conversely k et l, then
k - t e l - t by definition of e. Thus (et)tEH is the nest of equivalence
relations of e. D
5. Theorem 4.6 readily yields generators of any ]-congruence e, and a
presentation of the corresponding nilmonoid F ;e.
Proposition 4.8. When J is a proper cofinite ideal ofF, a ]-congruence
e on F is generated by all pairs (r, r + x) where x E X and r is a minimal
element of J, and all pairs (k- s, l- s) where k,l E K(J), k i= l, and s tJ. J
is maximal such that s ~ k,l and k es l.
Consequently the quotient N 1 = Fje ofF by a reduced ]-congruence e is

generated (as a c.m. with zero) by X subject to all relations r = 0, where r is a
minimal element of J, and all relations k - s = l - s where k, l E K ( J), k i= l,
e
and s is maximal such that k 8 l. This is a finite presentation, by Dickson's
Theorem (Proposition VI.l.S).
6. Finally we find the kernel function of a ]-congruence e. Let
fJ(g) = (J: g+) n (J: g-);
iJ is the kernel function of the Rees congruence of J. For all t E H, let
ft(g) = { (k (\ l)- t I k,l E K, t ~ k,l, k et l, and k -l = g }.

5. EXAMPLES. 221
For instance, ! 0 (0) = K.

Proposition 4.9. In Theorem 4.6, the Redei group of e is G = G(F) and
the kernel JUnction of e is given for all g E G by
J(g) = iJ(g) u UtEH ft(g).

Proof. By definition, c E f(g) if and only if c + g+ e c + g-, if and only
if either c + g+, c + g- E J, or c + g+ + t et c + g- + t for some t E H.
In the first case, c + g+, c + g- E J if and only if c E iJ(g). In the second
case, k = c+g+ +t and l = c+g- +t satisfY k,l E K, t ~ k,l, k et l, and
k -l = g; hence c = (c + g+) 1\ (c +g-) = (k- t) 1\ (l- t) = (k 1\ l)- t E
ft(g). Conversely assume c E ft(g), so that c = (k 1\ l) - t, where k, l E K,
t ~ k,l, k et l, and k -l = g; then g+ = (k -l) v (k- k) = k- (k (\ l),
g- = (l - k) v (l - l) = l - (k (\ l)' c + g+ = k- t, c + g- = l - t,
c + g+ + t et c + g- + t for some t E H, and c E f (g) . 0
5. EXAMPLES.
In this section we look at some examples, which indicate that conditions ( 1)

through (3) in Lemma 4.5 allow large numbers of J-congruences even when 'D
is relatively small.
Example 5.1. This is Example 3.6, in which X= {x,y} and J is the ideal
generated by 8x, 5x + 4y, and 6y. We saw that K = {k, l}, where k = 4x + 5y
and l = 7x + 3y, and that 'D is generated as a J -congruence by (a, b), where
a = 2x + 4y and b = 5x + 2y.
.k
Example 5.1
There are but two equivalence relations on K, the equality e and the universal
relation U. We have k 'Dt l if and only if t ~ k,l and k- t 'D l- t; hence
'Do = 'Dx = 'D2x = 'DY = 'Dx+y = 'D 2x+y = U, since k- (2x + y) =a and
l- (2x + y) = b, and 'Dt = c for all other t E H.
Conditions (1), (2), and (3) in Lemma 4.5 are as follows. By (1), C =
{t E H I et = U} is a coideal of F, which by (2) is contained in D =
{ 0, x, 2x, y, x + y, 2x + y}. Condition (3) is trivial. Hence there is one
coherent nest of equivalence relations on K for every coideal C ~ D . There are
ten such coideals: 0, {0}, {O,x}, {O,x,2x}, {O,y}, {O,x,y}, {O,x,2x,y},
{O,x,y,x+y}, {O,x,2x,y,x+y}, and Dkz· This yields 10 J-congruences,
'
ranging from the Rees congruence ( C = 0) to 'D ( C = D). 0
Example 5.2. Let X = {x, y} and J be the ideal generated by 8x, 6x + 2y,
3x + 4y, and 5y. Then K = {k,l,m}, where k = 2x + 4y, l = 5x + 3y,
m = 7x + y. There are five equivalence relations on K: the equality c; the
universal relation U; the equivalence relation X whose classes are {k} and
{ l, m}; the equivalence relation £ whose classes are { l} and { k, m}; and the
equivalence relation M whose classes are { m} and { k, l} .
.k
.f
.m
Example 5.2
It is readily verified that k 'Dt l if and only if t ~ 2x, k 'Dt m if and only
if t ~ x, and l 'Dt m if and only if t ~ x + y. Hence N('D) is specified by
'Do 'Dx 'DY 'Dx+y 'D2x

U U X X M
with 'Dt = c for all other t E H.
Since condition (3) is again trivial, a coherent nest on K consists of equiva-
lence relations (et)tEH such that t ~ U implies eu et,
~ and et ~ 'Dt for all
t. In particular et = c unless t = 0, x, y, x + y, or 2x. The resulting gaggle
of coherent nests can be classified by 0 . e
5. EXAMPLES. 223
If e0 = e, then et = e for all t; e is the Rees congruence of J.

If e0 =X, then et ~ X n 'Dt for all t; hence e2x = e and et is either e
or X for all t. Thus e is determined by the co ideal c
= { t E H I et = X}.
Note that 0 E C ~ { 0, x, y, x + y}. There are five such coideals: {0}, { 0, x},
{0, y }, {0, x, y }, {0, x, y, x + y}. This yields five nests:
eo ex ey ex+y e2x
X e e e e
X X e e e
X e X e e
X X X e e
X X X X e
If e0 = ,C, then, similarly, ey = ex+y = e2x = e and ex is either e or
,C. This yields two nests:
eo ex ey ex+y e2x
,c e e e e
,c ,c e e e
If e0 = M, then ey = ex+y = e and et is either e or M for all t. Thus
e is determined by the nonempty coideal c = {t E H I et = JV(} ~ { 0, x, 2x}.
This yields three nests:
Nests with e0 = U can be classified by ex. If ex = e, then ex+y = e2x = e

and ey ~ X; this yields two nests:
If ex = X, then e2x = e and et ~ X for all t # 0. Thus e is determined by the

co ideal c
= { t E H I et 2 X}' which satisfies { 0, X} ~ c ~ {0, X' y' X + y}.
This yields four nests:
eo ex ey ex+y e2x
u X e e e
u X X e e
u X e X e
u X X X e
If ex = .C, then ex+y = e 2x = e and ey ~ X; this yields two nests:
eo ex ey ex+y e2x
u .c e e e
u .c X e e
If ex = M, then ex+y = e, ey ~ X, ex+y = e (since ex+y ~ ex n ey),
and e 2x ~ M. This yields four nests:
eo ex ey ex+y e2x
u M e e e
u M X e e
u M e e M
u M X e M
If finally ex = U, then ex+y ~ ey ~X and e 2x ~ M. This yields six nests:
eo ex ey ex+y e2x
'U 'U c. c. c.
u u X e e
u u X X e
u u c. c M
u u X c M
u u X X M
This example has 1 + 5 + 2 + 3 + (2 + 4 + 2 + 4 + 6) = 29 J-congruences. 0
Example 5.3. In this example condition (3) is not trivial. Let X = {x, y} and
J be the ideal generated by 5x, 3x + y, x + 3y, and 5y. Then K = {k,l,m},
where k = 4y, l = 2x + 2y, m = 4x. We see that k - l = l - m. As in
the previous example there are five equivalence relations on K: the equality e;
the universal relation U; the equivalence relation X whose classes are { k} and
{l,m}; the equivalence relation .C whose classes are {l} and {k,m}; and the
equivalence relation M whose classes are { m} and { k, l}.
It is readily seen that k 'D t l if and only if t ~ 2y; k 'D t m if and only if
t = 0; and l 'Dt m if and only if t ~ 2x. Hence N('D) is specified by
5. EXAMPLES . 225
.k
.e
. .m
Example 5.3
1lo 'Dx 1l2x ny 1l2y

U X X M M
with 1lt = c for all other t E H.
Let (~t)tEH be a coherent nest of equivalence relations on K. Conditions
(1) and (2) state that ~ 2 x s;;;; ~x s;;;; ~ 0 , ~x s;;;; X, ~ 2 Y s;;;; ~Y s;;;; ~ 0 , ~Y s;;;; M, and
~t = c for all t 1: 0, x, 2x, y, 2y.
Since k - l = l - m, with k - k 1\ l = 2y, l - k 1\ l = 2x, condition (3)

reads: k ~ 2 y+t l if and only if l ~ 2 x+t m, for all t E H; equivalently, ~ 2 Y = Jv(
if and only if ~ 2 x = X.
As before, the resulting nests can be classified by ~ 0 . If ~ 0 = c, then ~t = c
for all t; ~ is the Rees congruence of J.
If ~ 0 =X, then ~t s;;;; X n 1lt for all t; hence ~y = ~ 2y = c, ~2x = c by
(3), and ~x s;;;; X. This yields two nests:
~0 ~x ~2x
X c c
X X c
If ~ 0 =£,then ~t s;;;; X n 1lt for all t; hence ~t = c for all t 1: 0. This
yields one nest.
If ~ 0 = M, then ~x = ~2x = c, ~2Y = c by (3), and ~Y s;;;; M. This yields
two nests:
Nests with ~0 = U can be classified by ~x s;;;; X and ~Y s;;;; M. If ~x =

~y = c' then ~t = c for all t 1: 0. If ~X = X and ~y = c' then e2y = c
and ~2x = c by (3). If ~X = c and ~y = M, then e2x = c and e2y = c by
(3). If ~X =X and ~y = M, then, by (3), either ~2x = ~2y = e or ~2x = X,

~ 2 y = JY(. This yields five nests:
~0 ~X ~2x ~y ~2y
'U e e e e
'U X e JY( e
'U e e JY( e
'U X e JY( e
'U X X JY( JY(
This example has 1 + 2 + 1 + 2 + 5 = 11 ]-congruences. D

In each of these examples there is for every pair k, l of comer points some
h E H such that k 'D t l if and only if t ~ h. This is not true in general but
holds whenever F has two generators (Grillet [1991N]).
Chapter X.
GROUP-FREE SEMIGROUPS.
This chapter contains a global construction of complete group-free semi-

groups, as quotients of free commutative monoids; finite group-free semigroups
are a particular case. This construction bypasses the difficulties, noted earlier, in
reassembling archimedean components and Ponizovsky factors, and accounts well
for the main structural features of these semigroups (idempotents, JC-classes, ar-
chimedean components, and Ponizovsky factors); its relation to extension groups
will be noted in Chapter XIII. It was first obtained by the author in the case of
finite congruences [ 1993], and generalized to complete group-free congruences in
[2001C]. The class of partially free complete semigroups arises as an application.
A different construction of finite group-free semigroups (by induction) was
devised by John [ 1973].
For brevity's sake we often abbreviate "complete group-free" as c.g.f. and
"complete group-free monoid" as c.g.f.m. in what follows.
1. SEMILATTICE CONGRUENCES.
In this section we show that semilattice congruences on a free commutative

monoid F are determined by certain sets of faces of F. This preliminary result
expresses a basic property of semilattice congruences, that their classes run parallel
to faces ofF. The results, due to the author [2001C], extend to any commutative
monoid S but will be used later only when S is free.
1. Let Z! be a semilattice congruence on a free c.m. F. The direction face
of a E F is the set
DZ!(a) = D(a) = {t E F I a+t Z! a}.
Thus t E D( a) if and only if, in the semi lattice (F /Zl, 1\), Ya+t = ~ 1\ ~ = ~,
if and only if~ ;:;; ~. The zj-class ~ of a extends "in the direction of' D(a)
in the sense that is contains a + t for every t E D( a).
227
228 X. GROUP-FREE SEMIGROUPS.
Lemma 1.1. Let 'a be a semilattice congruence on F. Every direction

face is a union of 'a-classes and a face of F. Moreover, DlJ (a) is the smallest
direction face which contains a.
Proof. We saw that t E D( a) if and only if y;_ ~ r:;,
in the semi lattice F /'a.
Hence D(a) is a union of'a-classes; D(a) i= 0, since a E D(a); and D(a) is
a face of F since, in F /'a, s 1\ t ~ r:;,
if and only if s ~ and t ~ r:;,
If r:;,.
a E D(b), then r:;,
~ Jb, ~ ~ r:;,
implies ~ ~ Jb, and D(a) ~ D(b). D
2. When 'a is a semilattice congruence on a free c.m. F, the direction set
of 'a is the set
D('a) = {DlJ(a) Ia E F}
of all the direction faces.
Proposition 1.2. Let 'a be a semilattice congruence on a free c.m. F. The
direction set D('a), partially ordered by inclusion, is an upper semilattice, in
which DlJ(a) V DlJ(b) = DlJ(a +b).
Proof. For every a,b E F we have ab E D(ab), a,b E D(ab), since
D(ab) is a face, and D(a), D(b) ~ D(ab) by Lemma 1.1. If moreover D(a),
D(b) ~ D(c), then similarly a,b E D(c), abE D(c), and D(ab) ~ D(c). Thus
D(ab) = D(a) V D(b) in D('a). D .
Proposition 1.3. A semilattice congruence 'a on afree c.m. F is determined
as follows by its direction set :D = D(Y). For every C E :D.
C* = {a E F I D( a) = C}
= {c E C I c rf. B when B E :D and B ~ C}
is a 'a-class; every 'a-class can be constructed in this fashion; and C f---+ C* is
an isomorphism (D('a), V) ~ F/13.
Proof. Let C* = { c E C I c rf. B when B E D and B ~ C}. When
a E C*, then D(a) ~ C by Lemma 1.1, and D(a) ~ C is not possible since
a E D(a); hence D(a) =C. If conversely D(a) = C, then a E C, and a rf. B
when B E D and B ~ C, otherwise C = D( a) ~ B by Lemma 1.1. Thus
C* = {a E F I D (a) = C} .
By Lemma 1.1, D(a) ~ D(b) if and only if a E D(b), if and only if ~ Jb r:;,
in F /'a. Hence D( a) = D(b) if and only if r:;,
= Jb, and C* is the 'a-class of all
a E F such that D(a) = C; then C f---+ C* is an isomorphism (D('a), v) ~ F/13
by Proposition 1.2.4. D
Corollary 1.4. C* ~ D* in F /'a if and only if C ~ D; and C* + D* ~
1. SEMILATTICE CONGRUENCES. 229
(CVD)*.
Proof. If c E C* and d E D* , then D (c) = C, D (d) = D, D (c + d) =
D(c) V D(d) = C V D, and c +dE (C V D)*. 0
3. Direction sets of semilattice congruences can be characterized as follows.
Proposition 1.5. Let F be a free c.m. A set 1) offaces ofF is the direction
set of a semilattice congruence on F if and only if
(Dl) for every a E F there is a smallest C E 1) such that a E C;
(D2) no element of 1) is the union of smaller elements of 1).
Then there is only one semilattice congruence ~ on F whose direction set is 1);
when a,b E F, then a~ b if and only if ('vC E 1))(a E C ~ bE C).
Proof. When 1) = D(~), then D(a) is the smallest C E 1) which contains
a, by Lemma 1. 1, and when C = D (a) then a does not belong to any direction
face B ~ C, since C is the smallest direction face which contains a. Thus (0 1)
and (02) hold.
Conversely assume that (D I) and (D2) hold. Define ~ as in the statement.
By (01), a ~ b if and only if the smallest C E 1) which contains a coincides
with the smallest C E 1) which contains b. Hence ~ is an equivalence relation.
If a ~ b and c E F, then, for every C E 1), a + c E C if and only if a E C
and c E C (since C is a face), if and only if b E C and c E C, if and only if
b + c E C; hence a + c ~ b + c. Moreover a ~ 2a. Thus ~ is a semilattice
congruence. We show that D(~) = 1).
Let C E 1). By (02) there exists c E C \ U (B E 1) and B ~ C). By
(D I) there is a smallest D E 1) which contains c. Then D ~ C and the choice
of c rules out D ~ C. Therefore C is the smallest D E 1) which contains c.
We show that C = D}J (c) . When x E C, then x E C and c + x rt B when
B E 1) and B ~ C, since c rt B. Hence C is the smallest D E 1) which
contains c + x . Therefore c + x ~ c and x E D}J (c) . Conversely x E D}J (c)
implies c ~ c+x, c+x E C, and x E C. Thus C = D}J(c) ED(~).
Conversely let a E F. By (0 1) there is a smallest C E 1) which contains
a. Then a rt B when BE 1) and B ~C. By the above D'zl(a) = C E 1).
Thus D(~) = 1); by Proposition 1.3, ~ is the only semilattice congruence with
this property. 0
A direction set on a free c.m. F is the direction set of a semilattice congru-
ence on F; equivalently, a set of faces of S with properties (0 1) and (D2).
Proposition 1.6. When F is finitely generated, then every direction set on
F is finite.
Proof. A direction set 'D on F is the direction set 'D = D(;j) of a semilattice
congruence Y on F; if F is finitely generated, then F /Y is finitely generated,
F /Y is finite, and 'D ~ F /Y is finite by Proposition 1.3. 0
4. Finite direction sets have a simpler characterization.
Proposition 1.7. Let F be a .free c.m. A .finite set offaces is a direction set
of F if and only if it contains F and is closed under intersections.
Proof. Let Y be a semilattice congruence whose direction set 'D is finite. For
every nonempty subset S of F we show that there is a smallest C E 'D which
contains S. Since 'D is a finite upper semi lattice, the finite set { D( s) I s E S}
has a least upper bound D in 'D. Then s E D( s) ~ D and S ~ D. If conversely
S ~ C E 'D, then s E C for all s E S, D ( s) ~ C by Lemma 1.1, and D ~ C.
Thus D is the smallest C E 'D which contains S.
In particular F is contained in some C E 'D and F E 'D. Let S = niEI Di
be an intersection of elements of 'D . Then 0 E Di for all i and 0 E S i= 0 . By
the above there is a smallest C E 'D which contains S. Then C is contained in
every Di and C = S. Thus 'D is closed under intersections.
Conversely let 'D be a finite set of faces of F which contains F and is
closed under intersections. For every a E F the intersection of all D E 'D
which contain a is the smallest C E 'D which contain a; thus (Dl) holds. Next
we show that a face C ofF is not the union C = B 1 U · · · U Bn of finitely
many faces B 1 , ... , Bn ~ C: if indeed b1 E C\B 1 , ... , bn E C\Bn, then
b1 + ··· + bn E C but b1 + ··· + bn tt B 1 , ... , b1 + ··· + bn tt Bn. Therefore
no C E 'D is the union of the finitely many faces B E 'D such that B ~ C, and
(D2) holds. 0
5. The following examples shows that not every direction set contains F
or is closed under intersections, and that, in general, not every subset of F is
contained in a smallest direction face.
Example 1.8. Let F = Fx be an infinitely generated free c.m. The support
of a = 2:xEX ax x E F is the finite set
S(a) = { x E X I ax i= 0 },
The smallest semilattice congruence N on F is given by:
aN b if and only if S(a) = S(b).
Then t E D(a) if and only if S(a) = S(a + t), if and only if S(t) ~ S(a); hence
D(a) is the face generated (as a submonoid of F) by S(a). Thus the direction
1. SEMILATTICE CONGRUENCES. 231
e
faces of are the finitely generated faces. The direction set of N is closed under
intersections but is not finite and does not contain F. No infinitely generated
face is contained in a direction face, let alone a smallest direction face. D
Example 1.9. Let S be the semilattice { e 1 , e 2 , en, ... , ! 1 , ! 2 , fn, ... ,
r 1 , r 2 , rn, ... , 8 1 , 82 , 8n, ... , 1} in which: 1 is the greatest element;
em = em+1 1\ rm, fn+1 = fn 1\ 8n, em ~ fn for all m,n; but no two of
r 1 , r 2 , r n, ... , 8 1 , 82 , 8n, ... are comparable. We see that S is not a complete
semi lattice.
Let F be the free c.m. with an infinite basis x 1 , x 2 , xn, ... , y 1 , y2 , Yn, ... ,
z 1 , z 2 , zn , ... , w 1 , w 2 , wn , . . . . Let 7f : F ---+ S be the homomorphism such
that 1fXn =en, 1fYn = fn, 1fZn = rn, 1fWn = 8n for all n. The congruence
~ induced by 7f is a semi lattice congruence on F, and F /~ ~ S.
InS the upper sections U(8) = {u E S Iu ~ 8} are: U(1) = {1};

U(rn) = {rn,1}; U(8n) = {8n,1};
U(fn) = {1, !1' f2, ... 'fn, 81' 82, ... '8n_d; and U(en) =
{ 1, en' en+1' · · · 'f1' f2' · · · 'fn' · · · 'r n' r n+1' · · · ' 8 1' · · · ' 8n' · · ·} ·
In F, t E D(a) if and only if 1r(t) ~ 1r(a). Hence D(a) = 1r- 1 (U(1ra)) and
the direction faces of ~ are: D(O) = {0};
D(zn) is the face generated by {zn};
D( wn) is the face generated by { wn};
D(yn) is the face generated by { y 1 , y 2 , ... , Yn, w 1 , w 2 , ... , wn_ 1 };
D(xn) is the face generated by { xn, xn+ 1 , ... , y1 , Y2, ... , Yn, .. · , zn,
zn+1' · · · ' W1' · · · ' wn' · · ·} ·
These are all the direction faces. Then nn>O D(xn) is the face generated by
{ y 1 , y 2 , ... , Yn , ... , w 1 , w 2 , ... , wn , ... } and is not a direction face; thus
the direction set of Y is not closed under intersections. Also there does not exist
a smallest direction face which contains nn>O D(xn) or { y 1 , y 2 , ... , Yn, ... ,
w 1 , w 2 , ... , wn , ... } . 0
2. DIRECTION SETS.
We now begin in earnest the study of complete group-free congruences on a

free commutative monoid F. In Section 1 we saw that semilattice congruences
on F have classes that run parallel to faces of F. This property extends to every
complete group-free congruence; the resulting faces constitute a direction set as
defined in Section 1, and determine the archimedean components of the quotient
monoid.
1. First we note some properties of complete group-free monoids. When S is
c.g.f., all subgroups of S are trivial; JC is the equality on S; and Green's preorder
~:rc is a partial order relation on S (Proposition V.2.3), which we denote by ~.
Proposition 2.1. Let S be a complete group-free monoid. Every element of

S has an idempotent power. Jf S is finitely generated, then S is finite.
Proof. Since S is complete, every element of S has a power in a subgroup of
S and has an idempotent power. In particular an element of S has only finitely
many distinct powers. If S is finitely generated, then every element of S is a
product of powers of finitely many generators, and S is finite. 0
By Proposition 2.1 we can arrange that S is finite simply by letting F be
finitely generated. In this case there is a standard presentation, just as in the case
of nilmonoids. Recall that an element a of a monoid S is irreducible when
a =f. 1, and a = be implies b = a or c = a.
Proposition 2.2. A finite group-free monoid is generated (as a monoid) by
its irreducible elements, and therefore has a smallest generating subset.
Proof. An irreducible element of a monoid S is not the product of other
elements of S and must therefore be included in every generating subset of S.
We prove by noetherian induction on a that every element a of a finite group-
free monoid S is a (possibly empty) product of irreducible elements of S. If
this is not true then S has an element m which is not a product of irreducible
elements of S and is maximal (under ~) with this property. In particular, m is
2. DIRECTION SETS. 233
not an empty or one-term product of irreducible elements, so m i= 1 and m is

not irreducible. Therefore m = be for some b, e E S, b, e i= m. Then b, e > m;
by the choice of m, b and e are products of irreducible elements of S; hence
so is m = be, which is the required contradiction. D
By Proposition 2.2, a finite group-free monoid has a standard presentation
1r : F -----+ S, where F is free on the set X of all irreducible elements of S and
1r is the identity on X .
Lemma 2.3. Let S be a complete group-free monoid For every s E S,
U(s) = { u E S I us= s}
is aface of S.
U (s) is the stabilizer of the trivial JC-class {s}.
Proof. U(s) is a submonoid of S since ls = s, and us= vs = s implies
uvs = s. If conversely uvs = s, then s = uvs ~ us ~ s and us = s; similarly
vs = s. Thus U(s) is a face of S. 0
2. In what follows, F is a free commutative monoid; e is a complete
group-free congruence on F (a congruence e
on F such that the quotient
monoid Fje is complete group-free; equivalently, the congruence induced by a
surjective homomorphism ofF onto a c.g.f.m.); S is a c.g.f.m. and 1r : F-----+ S
is a surjective homomorphism which induces e.
The direction of the e-class C (a) of a E F is measured as in Section 1 by
its direction face, which is the set
De(a) = D(a) = {t E F I a+ t e a};
C(a) extends "in the direction of' D(a) in the sense that is contains a+ t
for every t E D( a) . Since e
= ker 1r, we have t E D( a) if and only if
1r(t) E U(1r(a)).
Lemma 2.4. De (a) is a face of F; moreover, a e b implies De (a) =
De(b), and a~ b implies De(a) ~ De(b).
Proof. D(a) = 1r- 1 (U(1r(a))) is a face ofF since U(1r(a)) is a face of
S. If a e b, then a+ t e b + t and a+ t e a if and only if b + t e b; thus
D(a) = D(b). If a ~ b, then b = a+ e, and a+ t e
a implies b + t =
a+t+e e a+e=b;thusD(a)~D(b).D
3. By Proposition 1.2.6, the smallest semilattice congruence N on S induces

a semi lattice congruence ~ = 1r -l (N) on F, which is the smallest semi lattice
congruence which contains e. Moreover, F /~ ~ S /N; the isomorphism sends
the ~-class ~ of a E F to the archimedean component A(1r( a)) of 1r( a) .
The main result of this section is that e and 'a have the same set of direction
faces.
Lemma 2.5. D'd(a) = {t E F 17r(t) ~N 7r(a)} = {t E F Ina e t+u
for some n > 0 and u E F}.
Proof. t E D'd(a) is equivalent to: yt ~ ~ in Fj'a; A(1r(t)) ~ A(1r(a)) in
SjN; 1r(t) ~N 1r(a) in S; 1r(a)n = 1r(t) s for some n > 0 and some s E S;
and na e t + u for some n > 0 and some u E F. D
By Lemma 2.5, De(a) ~ D'd (a). This may be a strict inclusion.
Example 2.6. Let F = N be free on one generator and e be the Rees
congruence of the ideal J = { m , m + 1 , ... } , where m > 1 . Then F / e
is a cyclic nilmonoid and is complete group-free. For every t E F we have
n1 = t + u for some n = t + 1 > 0 and u = 1 E F; hence D'd (1) = F. On
the other hand, { 1} is a e-class, since m > 1 , and De (1) = {0} . D
Nevertheless e and 'a have the same direction faces. This follows from
certain properties of c.g.f.m.
Lemma 2.7. Let S be a c.gjm. If e E S is idempotent, then e ~N s if
and only if es = e, if and only if s E U(e); if e = t:(s), then U(s) = U(e).
Proof. If e ~N s, then e =en= st ~9{ s, e = e 2 ~9{ es ~9{ e, es JC e,

es = e since JC is the equality on S, and s E U(e); conversely se = e implies
e = es ~N s.
Now let e = t:(s). Then es = s, and ue = e implies us= ues = es = s.
Conversely assume us= s. Since S is complete group-free, un = f is idempotent
for some n > 0, and then f ~9{ u and fu = f by the above; also fs = uns = s,
e ~ f, and uf = f yields ue =e. Thus U(s) = U(e). D
Proposition 2.8. When e is a c.g.f congruence on F, then
De (a) = D'd (a)
whenever 1r(a) is idempotent, and {De(a) I a E F} = {D'd(a) I a E F}.
Proof. When 1r( a) = e is idempotent, then s ~N e if and only if s E U (e),
by Lemma 2.7. Now t E 'D e(a) if and only if 1r(t) E U(e), and t E 'D 1-)(a)
if and only if a+ t 'a a, if and only if 1r(a) 1r(t) N 1r(a), if and only if
1r(t) ~N e; hence De(a) = D'd (a).
Now take any a E F. Since S is c.g.f., the archimedean component of
1r(a) contains an idempotent 1r(c). Then a 'a c. By Lemma 2.4 and the above,
D'd(a) = D'd(c) = De(c). There also exists 1r(d) = t:(1r(a)); then U(1r(a)) =
U(1r(d)) by Lemma 2.7, and the above yields De( a) =De( d) = D'd(d). Thus
2. DIRECTION SETS. 235
{De (a) I a E F} = { ~ (a) I a E F}. 0

The proof of Proposition 2.8 shows that ~ (a) depends only on the archi-
medean component of 1l' (a) , whereas De( a) depends only on E ( 1l' (a)) and on
the Ponizovsky factor of 7r( a) . In either case the direction faces are the direction
faces of idempotents.
4. The direction set of a complete group-free congruence e on F is the set
D( e) = {De (a) I a E F} of its direction faces. By Proposition 2.8, D( e) is
also the direction set of the semi lattice congruence 1J = 1l' -l (N) and is a direction
set as defined in Section 1. Propositions 1.2 and 1.3 yield:
Proposition 2.9. Let e be the c.gj congruence on F induced by a surjective
homomorphism 7r : F ---+ S. Then 'D = D( e) is a direction set and 'D, partially
ordered by inclusion, is an upper semilattice. For every C E 'D let
C* = { c E C Ic rJ. B when B E 'D and B ~ C};
I
Then C* = {a E F D}}(a) = C}; C* is a union ofe-classes; 7r(C*) is
an archimedean component of S; every archimedean component of S can be
constructed in this fashion; and CI---t 7r(C*) is an isomorphism ('D, V) 9:! S/N.
Corollary 2.10. Let A E 'D and a E A*. Then 7r(na) is idempotent for
some n > 0, and then na E A* and D(na) =A.
Proof. By Proposition 2.9, D}} (a) = A and na E A* for all n > 0, since
7r(A*) is an archimedean component of S. Also 7r(na) is idempotent for some
n > 0, since S is c.g.f.; then De(na) = ~(na) =A by Proposition 2.8. 0
5. We complete this section with two examples, which will be seen again in
later sections. When X = {x, y} we represent every element a = ax x + ay y of
F by a point (ax, ay) with nonnegative integer coordinates. F has four faces:
{0}, A = F{ x} = {a E F I ay = 0}, B = F{y} = {a E F I ax = 0}, and
F. Straight lines in the figures indicate the partition of F into e-classes.
Example A. As shown e is the nilmonoid congruence with identity class
B = {a E FI ax = 0 } , zero class {a E F I ax ~ 3 } , and two other classes
{ a E F I ax = 1 } and { a E F I ax = 2 } . Then S is the cyclic nilmonoid
I
( x x 3 = 0). We see that D(O) = D(x) = D(2x) = B and D(3x) =F.
Thus D(e) = {B,F}. 0
Example B. This example is more typical. The congruence is shown e
below. Slanted lines indicate two more e-classes, { 2x + 6y, 5x + 4y} and
{3x + 6y, 6x + 4y}. We see that D(O) = {0} = D({2x + 6y, 5x + 4y}),
D(9x) = D(7x+ 2y) = F{x} =A, and D(7y) = D(4x+5y) = D(7x+4y) =F.
Thus D(e) = {O,A,F}. The archimedean components of Fje in Example B
are shown by the thicker lines below.
·I·I·
·I ·I ·
Example A Example B
·I ·I ·
·I ·I ·
Example B: archimedean components
3. EXTENT CELLS.
In this section we construct the extent cells of a complete group-free congru-

ence and show how they determine the idempotents and Ponizovsky factors of the
quotient monoid.
3. EXTENT CELLS. 237
1. As before F is a free commutative monoid and e

is the complete group-
free congruence on F induced by a surjective homomorphism 1r : F ---+ S,
where S is a complete group-free monoid; 'D = D( e) is the direction set of e.
The extent cells of e are the sets
EA = {a E F I De(a) =A},
one for every A E 'D. The family e= (EA) AE'D is the extent cell family of
e; it is a partition of F. (The similar sets A* = {a E F I D}J (a) = A} yield
the archimedean components of S in Proposition 2.9.)
The extent cells in Examples A and B (introduced in Section 2) are shown
by thick lines:
·I ·I·
·I· I·
Example A: extent cells Example B: extent cells
2. The extent cells of e determine the idempotents of S:

Proposition 3.1. For every A E D(e), IA = An EA ~ A* is a e-class
and an ideal of A; eA = 1r(IA) is idempotent; this yields every idempotent of S.
Proof. For any a E A*, Corollary 2.10 shows that na E A* and D(na) =A
for some n > 0. Thus IA =f. 0, in fact IA intersects A* .
If a,b E IA = AnEA, then a,b E A= D(a) = D(b) and a+b ea,
a +b e b. Hence IA is a e-class; then IA ~ A* , since A* is a union of e-clas-
e
ses. If similarly a E IA and c E A, then c E D(a), a+ c a, D(a +c) = A
by Lemma 2.4, and a + c E IA; thus IA is an ideal of A.
Let a E F. Then a E EA for some A= D(a) E 'D, and 1r(a) is idempotent
if and only if a+ a e a, if and only if a E D(a) =A. Thus 1r(a) is idempotent
if and only if a E IA for some A E 'D. Hence 1r(IA) is idempotent, and every
idempotent of S is obtained in this fashion. 0
Corollary 3.2. 1r(a) ~J{ eA if and only if a E A.
Proof. If a E A, then a+ c E IA for any c E IA and 1r(a) ~J{ 1r(a+ c)= eA.
If conversely 1r(a) ~J{ eA, then a+t E IA ~A for some t E F and a EA. 0
homomorphism 1r : F ------7 S, and 'D be the direction set of e. Then eA ~ eE
in E(S) if and only if B ~ A in 'D; hence E(S) ~ ('D, v).
Proof. Let a E IA, bE IE, so that 1r(a) = eA, 1r(b) = eE. IfeA ~ eE,
then eA eE = eA and a+ b e a; hence b + t e b implies a+ t e a+ b +
t e a+b e
a, and B = D(b) ~ D(a) =A. If conversely B ~A, then bE A,
a+ bE IA since IA is an ideal of A, and eA eE = 1r(a +b) = 1r(a) =eA. 0
The isomorphism E(S) ~ ('D, v) also follows from Proposition 2.9, since S
is complete and E(S) ~ Y(S).
As noted after Proposition 2.8, the extent cells of e also determine the Poni-
zovsky factors of S:
homomorphism 1r : F ------7 S. Then a E EA if and only if eA = E(1r(a)); EA
is a union ofe-classes; 1r(EA) is the partial Ponizovsky factor of eA in S.
Proof. Let a E F and c E IA, so that 1r(c) = eA' If eA = E(1r(a)), then
U(1r(c)) = U(1r(a)) by Lemma 2.7, D(c) = D(a) =A, and a E EA. Conversely
let a E EA' Then a+ c e a, since c E A= D(a), and eA 1r(a) = 1r(a).
If eE 1r(a) = 1r(a), then eE = 1r(b) for some b E IE, a+ b e a, D(b) ~
D(a +b) = D(a) by Lemma 2.4, B ~ A, and eA ~ eE by Proposition 3.3.
Thus eA = E(1r(a)). Thus a E EA ifand only if eA = E(1r(a)). 0
In Example B the partial Ponizovsky factors ofF je are shown below by the
thicker lines; the idempotent classes are clear.
3. Extent cells also enjoy the following properties:
Lemma 3.5. Let e be a c.gj congruence on F with direction set 'D. The
extent cells of e constitute a partition ofF with the following properties:
(El) if a E EA, bE EE, and a~ b, then A~ B;
(E2) (\fa E A*)(:Jn > 0) na E EA;
(E3) for every A E 'D, the projections p~EE with A ~ B E 'D form a
' '
·I ·I·
Example B: partial Ponizovsky factors
partition of A';
(E4) every EA is a union oje-c/asses;
(E5) if t E A, a, b E EA, and a+ t e b + t, then a e b;
(E6) if a,b E EA and p~a = p~b, then a e b.
(E7) EA +A~ EA.
Proof. This can be deduced from Proposition 3 .4, as in the next chapter, but
we give a direct proof.
(El). If a E EA, b E EB, and a ~ b, then A = D(a) ~ D(b) = B by
Lemma 2.6.
(E2) follows from Corollary 2.1 0.
(E3). Let t E A'. For any a E A* we have na E EA for some n > 0 by
(E2) and na E A. Let c = na + t and C = D(c). Then p~c = t and A~ C
by (El) since c ~ na and D(na) = A. Thus t E p~Ec and the projections
p~EB with A~ BE 'D cover A'.
Assume t E p~EB, where A~ B, so that t = p~b for some bE EB. Now

b = PA b + t, c = na + t with PA b E A ~ C = D( c), na E A ~ B = D(b) ;
henceb e b+na=c+pAb e candB=D(b)=D(c)=CbyLemma2.4.
Thus and the projections p~EB with A ~ B E 'D are pairwise disjoint.
(E4). If a e bE EA, then D(a) = D(b) =A by Lemma 2.4 and a E EA.

(E5). 1ft E A, a,b E EA, and a+t e b+t, then A= D(a) = D(b) and
a e a+ t e b + t e b.
(E6). If a,b E EA and p~a = p~b, then pAa, pAb E A= D(a) = D(b) and
a e a+p~b = b+p~a e b.
(E7). If a E EA and tEA, then D(a) =A, a+t e a, and a+t E EA
by (E4). 0
If F is f.g., then every face is contained in a direction face by Proposition
1.7 and (E3) can be strengthened as follows:
Lemma3.6. IfF isfg., then,foreveryface A ofF, the projections p~EB
with A ~ B E 'D form a partition of A'.
Proof. Let A be a face of F. For every x E X, n(x)kx is idempotent

for some kx > 0, and then n(x)kx+l = n(x)kx by Lemma 2.7. Then d =
"ExEXnA kxx E A satisfies de 2d. Also d +X e d for every X E X n A,
since ( kx + 1) X e kx X; hence A ~ D( d) . In fact D( d) is the smallest B E '])
which contains A: if A~ D(b), then b+d e band D(d) ~ D(b+d) = D(b)
by Lemma 2.4.
For every t E A' we now have a = d +t E EB for some B E 'D,
A~ D(d) ~ D(d + t) = B by Lemma 2.4, and t = p~(d + t) E p~EB. Thus
the projections p~ EB with A ~ B E 'D cover A'.
Assume t E p~ Ec, where A ~ C E 'D. As above, d + t E EB, where
A ~ B E 2J. Also t = p~c for some c E E 0 . Since dE A ~ D(c) and
c = PA c + p~ c = PA c + t with PA c E A ~ D( d + t), we have
d+t e d+t+pAc = d+c e c
and C = D(c) = D(d + t) =B. Thus t belongs to only one of the sets p~EB
with A ~ B E 'D . 0
4. Properties (El) (E2) (E3) characterize extent cell families.
Theorem 3.7. Let F be a free commutative monoid and 'D be a direction
set on F. A partition e = (EA)AETI ofF is the extent eel/family of a complete
group-free congruence on F with direction set 'D if and only if it has properties
(El), (E2), and (E3); the smallest such congruence S is given by
aS b if and only if a,b E EA and p~a = p~b for some A E 'D.
Proof. Conditions (El), (E2), and (E3) are necessary by Lemma 3.5. Now
assume that (E 1), (E2), and (E3) hold. Define S as in the statement; S is an
equivalence relation since e is a partition of F.
Assume a S b, so that a, b E EA and p~ a = p~ b for some A E 'D, and let
c E F. Then a+c E En, b+c E E0 for some B,C E 'D, with A~ B,C
by (El). Moreover PA (a+ c) = PA (b +c); hence B = C by (E3). Then a+ c,
b+ c E En and
Pk(a +c) = Pk(PA (a+ c)) = Pk(PA.(b +c)) = Pk(b +c)
since A ~ B, and a + c S b + c. Thus S is a congruence. To show that S is
a complete group-free congruence we retrace the proof of Proposition 3.1 to find
the idempotents of S = F /S.
Let IA = An EA. Then IA =/= 0 by (E2). If a E IA and t E A, then
a+ t E A; also a+ t E En for some A ~ B E 'D by (El), and B =A by
(E3) since PA (a + t) = PA a E PA EA" Thus IA is an ideal of A. In particular
IA is a subsemigroup of F.
Moreover IA is the S-class of all a E EA with PA a = 0. Therefore IA is
an idempotent of S. This yields every idempotent of S: if indeed an S-class Sa
is idempotent, then a, 2a E EA and p_Aa = p_A(2a) for some A E 'D, and then
p_Aa = 0, a E AnEA, and Sa= IA"
Property (E2) then shows that every element of S has an idempotent power;
hence every archimedean component of S contains an idempotent.
Let a E EA. If c E IA, then as above a+c E En for some A~ BE 'D by(El)
andB=A by(E3)sincep_A(a+c)=p_AaEp_AEA;hencea+c Sa. Conversely
assume a+ d S a, where dE In. Then a+ dE EA and PA.(a +d)= p.Aa;
hence d E A and c + d E IA, since IA is an ideal of A. Therefore IA ~ In in
P/S. (Note that B ~A by (El).) Thus there is for every elements of Sa least
idempotent e of S such that es = s; namely, IA , when s = Sa with a E EA .
Hence S is complete.
Let e = I0 be an idempotent of S and Sa E He, where a E EA. Let c E Ic.
Since Sa E He we have a + c S a and a + b S c for some b E F. Then
c ~a+ c E EA and a~ a+ bE E0 ; by (El), C ~ A~ C and A= C.
Hence a+ b S c yields p_Aa + p_Ab = p_Ac = 0, p_Aa = 0, a E An EA, and
Sa = IA = e. Thus S is group-free.
Let a E EA" If a+t Sa, then a+t E EA, p_Aa+p_At = p_Aa, p_At = 0,
and t E A. If conversely t E A, then a+ t E En for some A ~ B E 'D by
(El), in fact B =A by (E3) since p_A(a + t) = p_Aa E p_AEA; hence a+ t S a.
Thus D8 (a) = A for all a E EA" Therefore 'D is the direction set of S and
£ is the extent cell family of S. By (E6), S is the smallest complete group-free
congruence with this property. D
When 1> is a direction set on F we call e = (EA)AEn an extent cell

family over 1> when it is a partition ofF with properties (El), (E2), and (E3);
equivalently, when it is the extent cell family of a c.g.f. congruence on F.
5. We illustrate properties (El), (E2), (E3) with an example.
Example 3.8. Let F be free on X = {x,y}. F has four faces: 0,
A= F{x}' B = F{y}' and F.
By Proposition 1. 7, a direction set 1> on F contains F and is closed under
intersections. In particular 1> has a least element C. Then 0 E E0 by (E 1). If
c E C, then pbc = 0 E pbEc and c E E0 by (E3); hence C ~ E 0 . Then C
is a e-class, by (E6), and is the e-class of the identity element.
Let e be a c.g.f. congruence on F with direction set 1>. If 1> = {F}, then
e has only one class, by (E6). If 1> = {C, F} , where C = 0, A, or B, then
EF is an ideal of F by (E7), and every a E F\ C = F* has an integer multiple
in EF. Therefore e is a nilmonoid congruence with identity class C and zero
class EF.
Example 3.8: EF
Let 1> = {0, A, F} (the case 1> = {0, B, F} is similar). Since F is the
only D E 1> which contains B, it follows from Lemma 3.6 that {p~EF} is a
partition of B', i.e. pAEF =A. Thus EF contains a multiple of y and extends
all the way to B. On the other hand, EF does not extend all the way to A:
indeed {p~EA, p~EF} is a partition of A'= B and p~EF is a proper ideal of
B. Let p~EA = {O,y, ... ,qy}. Then ay ~ q+ 1 for all a= axx+ayy E EF,
and ay = q + 1 for some a E EF. The first figure shows a typical shape of EF.
Next, EA U EF is an ideal ofF by (El), and then E0 = F\(EA U EF).

Example 3.8: Ep, EA, and E0
Noway ~ q for all a E EA, since p_AEA = {O,y, ... ,qy }. Let p ~ 0 be least
such that px + (q + 1) y E EF. Then ax ~ p for all a E EA: if indeed ax < p
for some a E EA, then b = axx + (q + 1)y ~a since ay ~ q, but b t/:. EA,
b tf:. EF, contradicting (E 1). The second figure shows typical shapes of Ep, EA,
and E0 .
The third figure then shows the S-classes.
·I ·I·
·I ·I·
Example 3.8: S-classes
Finally let 'D = {0, A, B, F}. Then Ep is an ideal of F but does not
extend all the way to A or B. As above let p~ EA = { 0, y, ... , qy} and
p~EB = {0, x, ... , rx}. Then ax ~ r + 1 and ay ~ q + 1 for every a E Ep,
and ax = r + 1, ay = q + 1 occur. Let p ~ 0 and s ~ 0 be least such that
px + (q + 1) y E Ep and (r + 1) x + sy E Ep. Then ax ~ p for all a E EA,
and ay ~ s for all a E EB . Our last figures show typical shapes of Ep, EA,
En, and E0 .
6. When e = (EA)AeD is an extent cell family, a strand is an S-class,

where S is the congruence in Theorem 3.7; S is the strand congruence of e
(and of e, when e is the extent cell family of e).
Every strand is obtained by translating an ideal of a face A E 1) along the
complementary face A'. Indeed let a E F, so that a E EA for some unique
A E 1), and let t = p~ a E A' . The strand of a E F is Sa = { b E EA I p~ b = t} .
The set
I = {a E A I a+t E Sa} = {a E A I a+t E EA}
is an ideal of A: if indeed a E I and bE A, then a+t E EA, a+t+b E EA

by (E3 ), and a + b E I. Then Sa = I + t is obtained by translating the ideal I
of A bytE A'.
e
Now let be a c.g.f. congruence on F and e be its extent cell family. Then
S ~ e by Theorem 3.7. Hence every e-class is a union of strands. The last
step in the construction of e is the assembling of strands into e-classes, which
is described in the next section.
4. TRACE CONGRUENCES.
The trace congruences of a complete group-free congruence e are nilmonoid

congruences which determine which strands are assembled into e-classes. Hence
4. TRACE CoNGRUENCEs. 245
e is determined by its direction set, extent cells, and trace congruences. This
construction of e will be completed in the next section.
I. As before, e is a c.g.f. congruence on F, induced by 1r : F -----+ S, with
direction set 1> and extent cell family E .
For every A E 1>, let
HA = pAEA ~ A' and JA = A'\HA.
Lemma 4.1. HA is a coideal and JA is a nilmonoid ideal of A'. IfF is
fg., then HA is finite. Moreover HA + IA ~ EA.
Proof. By (EI), the union I of all EB with A~ BE 1> is an ideal ofF;
hence JA =pAl is an ideal of A' and HA = A'\JA is a coideal of A'.
Let h E HA, h of. 0. Then h = pA.a for some a E EA, a rj:. A. Now
1r( na) is idempotent for some n > 0, by Proposition 2.1. Then na E I B for
some BE 1> by Proposition 3.1, A~ B by (El), and A~ B since na rj:. A.
Hence nh = PA (na) E JA. Therefore A' I JA is a nilmonoid. IfF is f.g., then
A' I JA is f. g., A' I JA is finite, and HA is finite.
Finally let h E HA and a E IA . Let B = D (a + h) , so that a + h E E B .
Then A= D(a) ~ B by Lemma 2.4; since PA.(a+ h)= hE pAEA, (E3) yields
B = A and a + h E EA" D
When A E 1> the trace congruence of e on A' is the congruence eA on
A' defined by: t eA u if and only if
t, u E JA' or a + t e a + u for some a E A.
In the above we may assume that a E IA, since IA is an ideal of A; then a + t,
a+u E EA if t,u E HA, by Lemma 4.1. The trace of e is the family (eA)AE'D.
If for instance e is the congruence S in Theorem 3.7, then eA is the Rees
congruence of JA. In general,
Lemma 4.2. eA is a JA -congruence.
Proof. First eA is a congruence on A', since JA is an ideal of A' and e is a
congruence. Also t eA u when t, u E JA. If h eA t with h E HA, t E A', then
a + h e a + t for some a E IA, a + h E EA by Proposition 4.1, a + t E EA
by (E4), and t = PA(a + t) E HA. Hence JA is a eA -class. D
Corollary 4.3. IfF is finitely generated, then A' 1eA is finite.
2. The main properties of trace congruences are:
Proposition 4.4. When e is a c.gj congruence on F, then a e b if and

only if there exists A E 1) such that a, b E EA and pA_ a eA pA_ b.
Proof. If a e b, then a and b are in the same extent cell EA, so that a' =
pA_a E HA, b1 = pA_b E HA" Also a e a+ c, b e b + c for all c E A =
D(a) = D(b). Hence
a1 +pAa+pAb = a+pAb e a e be b+pAa = b1 +pAb+pAa
and a1 eA b1 •
Conversely let a,b E EA. Assume that a' = pA_a eA pA_b = b1 , so that
a'+ c e b1 + c for some c EA. We may assume that c E IA, so that a'+ c,
b'+cEEAbyLemma4.l.Thena e a'+c,b e b'+cby(E6),andaeb.D
e
Proposition 4.5. Let be the c.gj congruence on F induced by a surjective
homomorphism 1r : F -----+ S. For every A E D, eA is a reduced nilmonoid
congruence on A', and A' I eA is isomorphic to the Ponizovsky factor of eA in
S; the isomorphism sends the eA -class of pA_ a to 1r( a), for every a E EA.
Proof. By Propositions 3.3 and 3.4, the Ponizovsky factor of eA is
PA = SeA I UAcBE'D
~
SeB
and a E EA if and only if 1r(a) is in the partial Ponizovsky factor
Pl = SeA\ UAcBE'D
~
SeB
of eA. Thus 1r induces a surjection 1r : EA -----+ Pl. Let p : EA -----+ HA be
the projection a 1-----t pA_ a. Then ker p ~ ker 1r by (E6) and there is a mapping
c.p of A' onto Pl such that c.p(pA_a) = 1r(a) for all a E EA, which is a partial
homomorphism, since 1r and p are homomorphisms.
EA~HA
~ 1~
p*
A
Moreover c.p(t) = c.p(u) if and only if teA u, by Proposition 4.4.
Extend c.p to '1/J : A' -----+ PA, so that 'lj;(t) = c.p(t) for all t E HA and
c.p(t) = 0 E PA for all t E JA" Then ker c.p = eA" We show that '1/J is
a homomorphism; then ker '1/J = eA, A' I eA PA, and eA is a nilmonoid
S:!
congruence, which is reduced by Proposition 4.4 since IA = p- 1{0} is a e-class.
4. TRACE CoNGRUENCEs. 247
The equality 1j; (t + u) = 'tj;(t) 'lj;( u) is clear if t + u E HA or if t E JA

or u E JA, since <p is a partial homomorphism. Now assume t, u E HA,
t + u E JA. Then t = p~a, u = p~b with a, b E EA, a+ b ¢:. EA, and
a+ bE EB for some A~ BE 'D by (E2). If c E IB, so that 1r(c) = eB, then
a+ b + c e a+ b, 1r(a +b)= eB 1r(a +b), and <p(t) <p(u) = 1r(a) 1r(b) = 0
in PA; hence 'tj;(t + u) = 'tj;(t) 'tj;(u) holds in this case also. D
Since S is a subdirect product of its Ponizovsky factors, the equality on S is
the intersection of the congruences induced by the projections S--+ SeA--+ PA.
Therefore:
Corollary 4.6. a e b if and only if p~ a eA p~ b for all A E 'D.
This can also be proved directly. If a e b, then for any A E 'D and c E IA
we have a+ c e b + c, a+ c, b + c E EB for some A ~ B E 'D, and
p~a = p~(a +c) eA p~(b +c) = p~b. Conversely assume that p~a eA p~b
for all A E 'D. We have a E EA, b E EB for some A,B E 'D. For any
c E IA we have b + c E Ec for some A ~ C E 'D by Lemma 2.4, and
p~(b+ c)= p~b eA p~a E HA; hence p~(b+c) E HA, b+c E EA by (E3), and
B ~ A. Exchanging a and b yields A ~ B, and then a e b by Proposition 4.4.
Since c.g.f.m. are subdirect products ofnilmonoids, every c.g.f. congruence on
F is an intersection of nilmonoid congruences; Corollary 4.6 provides an explicit
intersection.
4. Proposition 4.4 provides a ready set of generating pairs for e; equivalently,
a presentation of F je. When t E HA let
SA(t) = {aEEAip~a=t}
denote the strand which lies over t.
Proposition 4.7. A c.g.f congruence e on F = Fx is generated by:
(1) one pair (m+ x, m) for every A E 'D, x EX n A, t E HA, and minimal
element m of SA (t); and
(2) for every A E 1) and generator (t, u) of eA with t, u E HA, one pair
(m, n) in which m is a minimal element of SA(t) and n is a minimal element
of SA(u).
When F is finitely generated, then the above is a finite generating set, by
Dickson's Theorem (Corollary VI.l.3). Indeed SA (t) has only finitely many
minimal elements; since 1) and every HA are finite, (I) yields only finitely many
pairs and so does (2).
Proof. Let ~ be the smallest congruence on F containing all pairs (1) and
(2). Then~ s:;:; e.
We show that every strand SA (t) is contained in a ~-class (i.e. S s:;:; ~ ).
Every a E SA (t) has the form a= m + c for some minimal element m of SA (t)
and some c = LxEXnA cxx E A; since m + x ~ m for all x EX n A,
we have a~ m. Let m and n be any two minimal elements of SA (t); then
a = m V n E SA (t), since pA_ a = pA_ m V pA_ n = t, a - m E A = D( m), and
me a; by the above, m ~a~ n. Hence SA (t) is contained in a single ~-class.
When t,u E HA and (t,u) is a generator of eA, it follows from (2) and the
above that SA (t) and SA (u) are contained in the same ~-class. Hence SA (t)
and SA (u) are contained in the same ~-class whenever t, u E HA and t eA u.
Then Proposition 4.4 yields e s:;:; ~. 0
Proposition 4.4 also yields the kernel function of e. For every A E 'D and
g E G let fi be the kernel function of eA and
fA(g) = PA -l(fi(pA_g)) = {c E F I pA_c E fi(p~g)};

the proof of Proposition 4.8 below shows that fA is the kernel function of the
nilmonoid congruence induced by the projection F ---+ A' ---+ A' jeA. IfF is
finitely generated (more generally, if JA is cofinite in A' for all A E 'D), then
fi can be constructed as in Proposition IX.4.9.
Proposition 4.8. Let e be a c.gj congruence on F. The Redei group of
e is G = G(F). The kernel .function f of e is given by:
c E f(g) if and only if c E (EA: g+) n (EA: g-) n fA (g) for some A E 'D.
Equivalently, f(g) = nACD fA(g).
Proof. For every a E F we have a E A* for some A E 'D, na E EA for
some n > 0 by (E2), and na e (n + l)a, since a E D(na), so that a belongs
to the Redei group of e . Hence the latter is all of G = G (F) .
LetcEFandgEG. ThencEf(g) ifandonlyifc+g+ e c+g-,ifand
onlyifc+g+, c+g- EEA andp~(c+g+) eA p~(c+g-) forsome(unique)
A E 'D, by Proposition 4.4. Now (p~g)+ = p~ (g+) and (p~g)- = p~ (g-); hence
pA_(c+g+) eA pA_(c+g-) ifandonlyifpA_c+(pA_g)+ eA pA_c+(pA_g)-, if
and only if pA_c E fi(p~g). Thus c E f(g) if and only if c E (EA :g+) n (EA:
g-) n fA (g) for some A E 'D (and then A is unique).
ByCorollary4.6, c+g+ e c+g- ifandonlyifp~(c+g+) eA pA_(c+
5. MAIN RESULT. 249
g-) for all A E 1), if and only if (as above) c E fA (g) for all A E 1). D
5. An example will illuminate the results in this section.
·I ·I ·
·I ·I ·
Example B
Example B. As in Section 2 let X= {x,y}, 1) consist of 0 = {0}, A=

F{x}' and F, and e
be the congruence in the figure. We see that Sp(O) = EF
has three minimal elements 7y, 4x + 5y, and 7x + 4y; SA(O) has one minimal
element 9x; SA(y) has one minimal element 9x + y; SA(2y) has one minimal
element 7x + 2y; SA(3y) has one minimal element 7x + 3y; the remaining
strands SD(t) have D = 0. Hence (1) yields the following pairs:
(x + 7y, 7y), (8y, 7y), (5x + 5y, 4x + 5y), (4x + 6y, 4x + 5y), (8x +
4y, 7x+4y), (7x+5y, 7x+4y), (lOx, 9x), (10x+y, 9x+y), (8x+2y, 7x+
2y) , (8x + 3y, 7x + 3y) ,
some of which may be omitted; and (2) yields the single pair
(2x+6y, 5x+4y).
By Proposition 4.7, S = Fje has the presentation (as a commutative monoid)
S = 1 ablab 7 =b7 =b8 a 5 b5 =a4 b5 =a4 b6

\ ' ' '
Since Sp(O) is the zero element of S, it is more economical to present S as a

commutative monoid with zero:
5. MAIN RESULT.
We now complete the construction of complete group-free congruences.

I. By Proposition 4.4, a c.g.f. congruence e on a free c.m. F is determined
by its direction set, extent cell family, and trace. The latter can be characterized
as follows.
Lemma 5.1. The trace of a c.gf congruence has the following properties:
for every A E 1J,
(Tl) eA is a JA -congruence on A';
(T2) ift,u E HA, teA u, and A~ BE 1J, then (pAEB): t = (pAEB): u
in A' and Pk (t + v) eB Pk (u + v) for all v E (PA EB) : t = (PA EB) : u ~ A';
(T3) {0} is a eA -class.
Proof. (Tl) holds by Lemma 4.2 and (T3) holds by Proposition 4.5. Let
t,u E HA, teA u, and A~ BE 1J. For any a E IA we have a+t,
a + u E EA and a + t e a + u by Proposition 4.4. Let v E A'. If t + v E
PA EB, then b + t + v E EB for some b E A, a + b + t + v E EB by (E3 ),
a + b + t + v e a + b + u + v, a + b + u + v E EB' and u + v E PA EB .
Conversely u + v E PAEB implies t + v E pAEB. Thus (pAEB): t = (pAEB): u
holds in A'; and in the above t+v EpAEB implies a+b+t+v e a+b+u+v,
with a+b+t+v, a+b+u+v E EB, and
Pk (t + v) = Pk (a + b + t + v) eB PA (a + b + u + v) = Pk ( u + v). 0
Theorem 5.2. Let F be a free commutative monoid, 1J be a direction set

on F, and £ be an extent cell family over 1J. A family T = (eA) AETI is the
trace of a complete group-free congruence e on F with direction set 1J and
extent cell family £, if and only if it has properties (Tl) and (T2); then
a e b if and only if a,b E EA and pAa eA PAb for some A E 1J.

Moreover, every complete group-free congruence on F can be constructed in
this fashion.
Proof. Conditions (Tl) and (T2) are necessary by Lemma 5.1, and Proposition
4.4 shows that every c.g.f. congruence on F can be constructed as in the statement.
Now let 1J, £, and T be given, so that (Tl) and (T2) hold, and let a e b if
and only if a,b E EA and pAa eA pAb for some A E 1J. By Theorem 3.7 there
is a c.g.f. congruence S on F with direction set 1J and extent cell family £;
5. MAIN RESULT. 251
namely, aS b if and only if a,b E EA and p~a = p~b for some A E 'D. We
see that s s;;; e.
e is an equivalence relation on F, since E is a partition ofF and every eA
is an equivalence relation. Assume that a e b and let t E F. Then a, b E EA
for some A E 'D and p~a eA p~b. Also a+ t E EE and b + t E E 0 for
some B,C E 'D, and As;;; B,C by (El). Since p~a + p~t E p~EE, we have
p~b+p~t E p~EE by (T2), B = C by (E3), and b+t E EE. Since As;;; B,
(T2) also yields
pka + Pkt = Pk(p~a + p~t) eE Pk(p~b + p~t) = Pkb + Pkt,

and a + t e b + t. Thus e is a congruence.
To prove that e is a c.g.f. congruence we prove:
Lemma 5.3. If t + v eA t and t, t +v E HA, then v = 0. In particular,
(T3) holds.
Proof. t +v eA t implies
t eA t +v eA t + 2v eA · · · eA t + nv
for all n > 0. Since JA is a nilmonoid ideal, v =1- 0 would imply nv E JA
and t + nv E JA for some n > 0, and t E JA since eA is a JA -congruence,
contradicting t E HA ; therefore v = 0. 0
LetS= Fje. By Proposition 3.1, applied to S, IA = AnEA is an ideal of
A and an S-class. In fact IA is a e-class by (T3) and IA is idempotent in S, since
it is a subsemigroup of F. Conversely assume that the e-class Ca of a E F is
idempotent. Then there exists A E 'D such that a, 2a E EA and p~a eA p~ (2a)
for all n > 0; by Lemma 5.3, p~a = 0, a E AnEA, and Ca = IA" Thus every
idempotent e-class has the form IA for some unique A E 'D. By (E2), every
element of S has an idempotent power; hence every archimedean component of
S contains an idempotent.
Let a E EA. If c E IA, then a+ c E EE for some As;;; BE 'D by (El) and
B =A by (E3), since p~ (a+ c) = p~a E p~EA; hence a+ c e a. Conversely
assume a+d e a, where dE IE. Then a+d E EA and p~(a+d) eA p~a.
By Lemma 5.3, p~d = 0 and dE A. Hence c +dE IA, since IA is an ideal
of A, and IA ~IE in S. (Note that B s;;; A by (El).) Thus there is for every
element s of S a least idempotent e of S such that es = s; namely, e = IA,
when s = Ca with a E EA. Thus S is complete.
Let e = Ic be an idempotent of S and Ca E He . Let c E I0 . Then
a + c e a and a + b e c for some b E F. Now c ~ a + c E EA and

a ~ a + b E Ec; by (E 1), C ~ A ~ C and A = C. Hence a + b e c yields
p~a+p~b eA p~c = 0, p~a = 0, a E AnEA = IA, and Ca =e. Thus Sis
group-free, and e is a c.g.f. congruence.
We show that e has the given direction set 1) and extent cell family f.. Let
a E EA. Since S has direction set f> and extent cell family f., and S ~ e,
we have De(a) 2 Ds(a) = A. If conversely a+ t e a, then a+ t E EA,
p~a + p~t eA p~a, p~t = 0 by Lemma 5.3, and tEA. Thus De(a) =A for
all a E EA. Therefore e has the given direction set 1), and the given extent cell
family f..
Finally we show that e has the given trace. Let A E 1) and t, u E HA. If
a+ t e a+ u for some a E A, then a+ t, a+ u E EB for some A~ BE 1)
by (El), t = p~(a+t) E p~EB, B =A by (E3), a+t, a+u E EA, and teA u
by definition of e. If conversely t eA u, then for any a E IA we have a+ t,
a + u E EA by Lemma 4.1, and a + t e a + u, since p~ (a + t) = t eA u =
p~ (a + u) . Thus eA is the trace congruence of e on A' . D
2. When F is f.g., every trace congruence eA is a finite nilmonoid congruence

and is determined as in Chapter IX by its zero class JA and its nest of equivalence
relations. Examples show that condition (T2) puts serious restrictions on the latter.
Example 5.4. let X = {x,y}; let f> consist of {0}, A = F{x}' and
F; and let Ep, EA, and E{o} be as in the figure (and as in Example B).
Necessarily eF is the equality on F' = {0} and eA is the Rees congruence
of JA = {4y, 5y, · · ·}; but e{O} is more interesting. We see that H = H{o}
has three maximal elements (comer points) k = 3x + 6y, l = 6x + 4y, and
m = 8x + y. For the nest of equivalence relations (f>t)tEH of the largest ]-
congruence on F (where J = F\H), it is readily verified that k f>t l if and
only if t ~ 2x + y; k f> t m if and only if t ~ x + y; and l 1) t m if and only
if t ~ x + y. Let K = { k, l, m} and JY( be the equivalence relation on K with
classes { k} and {l, m} .
Since P{o} is the identity on F, condition (T2), applied to e{O}, reads: if

t,u E H and t e{O} u, then
(1) H:t=H:u (since e{O} isacongruencewealreadyha ve t+ve{O} u+v

for all v );
(2) EA:t=EA:uandp~(t+v) eA p~(u+v) forallvEEA:t=EA:u,
5. MAIN RESULT. 253
Example 5.4
equivalently ty = uy or ty + vy, uy + vy ~ 4 for all v E EA: t = EA: u;
(3) EF: t = EF: u (then p~(t + v) eF p~(u + v), since F' = 0).
It suffices to consider (2): since E{o}, EA, EF is a partition of F, (3) follows
from (1) and (2), and we saw in Chapter IX that (I) holds for every J-congruence
on F. By (2), k eA m and l eA m cannot hold, since EA : k = EA : l = 0 and
EA: m of: 0. Thus the support of e{O} has at most one nontrivial class {k,l},
and Ct ~ M for all t E H (whereas 'D x+y = U ).
Hence e {0} and e are determined by the co ideal c = {t E H I et = M}.
Now k and l satisfy (2), since EA: k = EA: l = 0 and p_A. (k + v) eA p_A. (l + v)
holds vacuously for all v E EA : k = EA : l. If more generally t E F, t ~ k, l,
then k- t and l - t satisfy (2) if and only if EA : (k- t) and EA : (l - t) are
both empty, that is, ty = 0; thus k et l implies t ~ 2x. Thus C ~ { 0, x, 2x}.
Hence there are four c.g.f. congruences with the given direction set and extent
cells. In Example B, C = {0, x}.
3. In Example 5.4 there is a largest c.g.f. congruence with the given direction
set and extent cells. This is always true:
Proposition 5.5. Let 'D be a direction set on F and £ be an extent cell
family over 'D. There exists a greatest c.gf congruence .C with direction set 'D
and extent cell family £; its trace is given for every A E 'D by: t £A u if and
only ift,u E JA or p_A.EB: t = p_A.EB: u holds in A' for every A~ BE 'D.
Proposition 5.5 generalizes Proposition IXJ .1.
Proof. We show that 'J = (£A) AE'D satisfies (Tl) and (T2). First, £A is
an equivalence relation, and t £Au for all t,u E JA. If p_A.EB: t = p_A.EB: u
holds in A' for every A ~ B E 1), and u E JA, then u E p~ EB for some
A ~ B E D, 0 E p~EB: u, 0 E p~EB: t, t E p~EB, and t E JA. Thus
t !:.-A u E JA implies t E JA, and JA is a £-A -class.
Assume t !:.-A u and let v E A'. If t, u E JA, then t + v, u + v E JA and
t + v £-A u + v. Otherwise t,u E HA, p~EB: t = p~EB: u holds in A' for
every A~ BED, and
p~EB:(t+v) = (p~EB:t):v = (p~EB:u):v = p~EB:(u+v)
holds in A' for every A ~ B E 1) . Hence t + v E JA if and only if u + v E JA ,
and t +v !:.-A u + v. Thus £-A is a JA -congruence on A'.
Next we prove the following: if A ~ B ~ C in 1) and u E A', then
pku E pkEc if and only if u E p~Ec. Indeed u E p~Ec implies u + v E E 0
for some v E A and pku = Pk(u + v) E pkE0 . The converse holds by (E3).
Now assume t,u E HA, t £-Au, and A ~ B E D. Then p~EB: t =
p~ EB : u holds in A'. Let v E p~ EB : t = p~ EB : u ~ A', so that t + v,
u + v E p~ EB and Pk (t + v), Pk (u + v) E PkEB = H B . If B ~ C E 1) and
w E B' ~ A', then Pk(t + v) + w E pkEc implies Pk(t + v + w) E fJkE0 ,
t + v + w E p~ E 0 , v + w E p~ E 0 : t, v + w E p~ Ec : u, u + v + w E
p~E0 , and Pk(u + v) + w = Pk(u + v + w) E pkE0 ; and conversely. Thus
piJE0 :pk(t+v) = pkEc :piJ(v+u), which shows Pk(t+v) J:.-B Pk(u+v),
and proves (T2).
By Theorem 5.2 there is a c.g.f. congruence !:.- on F with direction set D,
extent cell family e, and trace 'J = (£-A) AED; namely, a!:.- b if and only if
a, b E EA and p~ a !:.-A p~ b for some A E 1). If e is a c.g.f. congruence on
F with direction set 1) and extent cell family e, then the trace (eA)AED of e
satisfies eA ~£-A for all A ED, by Lemma 5.1, and e ~ !:.-. 0
Theorem 3.7 and Proposition 5.5 provide inequalities S ~ e ~ !:.- for any
c.g.f. congruence with given direction set and extent cell family.
6. PARTIALLY FREE SEMIGROUPS.
The results in Section 4 suggest a class of complete group-free semigroups,

which are of interest partly because of Proposition 6.2 below, partly because of a
formula in Chapter XIII for their extension groups Ext (S, G) .
6. PARTIALLY FREE SEMIGROUPS 255
1. A partially free semi group is a complete group-free semigroup S such

that there is a free commutative monoid F and a surjective homomorphism 1r :
F ----t S 1 such that the trace of e = ker 1r consists of Rees congruences;
equivalently, in case the strand congruence S of e is e itself; equivalently, every
e-class consists of a single strand.
Proposition 6.1. A commutative nilsemigroup or nilmonoid is partially free
if and only if it is 0-.free.
Proof. Let e be a nilmonoid congruence on a free commutative monoid
F. Since a nilmonoid has just one nontrivial Ponizovsky factor, Proposition 4.5
implies that there are only two direction faces, F and the identity class U. Then
Ep is the zero class J, HF = 0, whereas Ju = P[;J and Hu = U'\P[;J. The
trace of e consists of the trivial congruence eF and the Ju -congruence eu. If
e = s' then FIe 9:! U' I eu = U' I Ju is a 0-free nilmonoid.
If conversely S is a 0-free nilmonoid, then there is a free commutative monoid
F and a surjective homomorphism 1r : F ----t S such that e = ker 1r is the Rees
congruence of a nilmonoid ideal J of F. Then the direction set of e consist of
0 = {0} and F. We see that HF = 0 and eF is trivial, whereas H0 = F\J
and e 0 = e is the Rees congruence of J. Thus S is partially free. D
Theorem 3.7 implies that partially free semigroups are not rare. Call a surjec-
tive homomorphism c.p : S ----t T pure in case the inverse image c.p -l (e) of every
idempotent e ofT consists of a single element of S (necessarily an idempotent).
This is consistent with the definition of a pure congruence in Chapter III. The fol-
lowing result generalizes the result of Arendt and Stuth [ 1970] (Corollary IX.2.5)
that every nilsemigroup is a pure homomorphic image of a 0-free nilsemigroup.
Proposition 6.2. Every c.gf semigroup (or monoid) is a pure homomorphic
image of a partially free semigroup (or monoid).
Proof. Let S be a c.g.f.m. There is a free commutative monoid F and a
surjective homomorphism 1r : F ----t S 1 . Let e = ker 1r and S be the strand
congruence of e. Since S ~ e, 1r factors through the projection p : F IS:
1r = c.p o p for some homomorphism c.p : F IS ----t S.
F~FIS
~1~
s
Then F IS is partially free and c.p is surjective like 1r. Since e and S have the
same direction set and extent cell family, they have the same idempotent classes
IA = EA n A (where A E 'D). Hence <p is pure: if t = p(a) E F/S and

<p(t) = eA = 1r(IA) is idempotent in S, then 1r(a) = 1r(IA), a E IA, and
t = p(IA) is idempotent in F /S. D
2. By Proposition 4.5, the Ponizovsky factors of a partially free semigroup
are 0-free. Example 6.5 below shows that this condition is not sufficient. Never-
theless, finite partially free semigroups can be characterized by their Ponizovsky
factors, due to additional properties by which the irreducible elements of a par-
tially free semi group determine the irreducible elements of its Ponizovsky factors.
Let Irr (S) denote the set of all irreducible elements of S.
Lemma 6.3. Given afinite group-free monoid Sande E E(S), e i 0, let
Qe = {q E Irr (S) I eq < e and E(eq) = e}.
(1) lf r E Irr (~), then r = eq for some q E Qe.
(2) lf 1r : Fx --+ S is surjective, x EX, and e = eA, then 1r(x) E Qe if

and only if 1r(x) E Irr (S) and x E HA"
Proof. (1). Let p : S --+ eS --+ ~ = eS / UtEE(S), f<e JS be the

projection: p(s) = es if E(es) = e (if es 1. fS when f <e), p(s) = 0 E Pe if
E( es) < e (if es E f S for some f < e). Since Irr ( S) generates S, p (Irr ( S))
generates ~ and contains every r E Irr (~): thus r = p(q) for some q E Irr (S),
and then q E Qe, since p(q) i 0 in ~ yields E(eq) = e, and eq = r <e.
(2). Assume that 1r(x) = q Qe. Then q E Irr (S). Also x 1. A, since
E
x E A would imply 1r(x) ~ eA (Corollary 3.2) and qe ~ e. Hence x E A'.
Assume that x 1. HA. Then x E p~EB for some A~ BE 'D and x + t E EB
for some t E A. Hence s = 1r(t) E S and f = eB E E(S) have the following
properties: s ~ e by Corollary 3.2; f < e by Corollary 3.3; and E(qs) = f, in
particular qs ~ f, by Proposition 3.4. Hence eq ~ sq ~ f < e, contradicting
E(eq) = e and q E Qe. Therefore x E HA"
Conversely assume that x E HA and q = 1r(x) E Irr (S). Then q te by

Corollary 3.2; hence eq < e. Let p E IA. Then x + p E EA by Lemma 4.1,
1r(p) = eA = e, 1r(x + p) = qe, E(eq) = e by Proposition 3.4, and q E Qe. D
Theorem 6.4. For a finite group-free commutative monoid S the following
(1) S is partially free (there is a .free commutative monoid F and a surjective
homomorphism 1r : F --+ S such that the trace of ~ = ker 1r consists of Rees
congruences);
6. PARTIALLY FREE SEMIGROUPS 257
(2) in the standard presentation 1r : F ---+ S of S, the trace of e= ker 1r

consists of Rees congruences;
(3) all Ponizovsky factors of S are 0-.free and, for every e E E(S), e =/= 0,
the mapping q f------+ eq is a bijection of
Qe = { q E Irr ( S) I eq < e and E ( eq) = e}
onto Irr (Pe).
Proof. (2) ===} ( 1) is clear.
(1) ===} (3). Let 1r : Fx ---+ S be a surjective homomorphism such that the
trace of e = ker 1r consists of Rees congruences. The Ponizovsky factors of S
are 0-free by Proposition 4.5.
Let e E E (S) , e =/= 0; then e = eA for some A E 'D, A =/= X . If q E Qe , then
q E Irr (S), q = 1r(x) for some x EX, x E HA by Lemma 6.3, and x + p E EA
for every p E IA by Lemma 4.1. Then x is irreducible in A' and 1r(p) = e. The
isomorphism () : A' jeA ---+ ~ in Proposition 4.5 sends the eA -class of p~a to
1r(a), when a E EA, and sends the eA -class of x = p~(x+p) to 1r(x+p) = eq.
Therefore eq E Irr (Pe). Thus e: q f------+ eq is a mapping of Qe into Irr (~).
Now e is surjective by Lemma 6.3. If e(q) = e(r), then r = 1r(y) for some
yEX, yEHA byLemma6.3, 7r(x+p)=eq=er=7r(y+p), x+p e y+p,
X eA y' X = y since eA is the equality on HA' and q = r. Thus e is bijective.
(3) ===} (2). Assume that ~ is 0-free and that e: q f------+ eq is a bijection of
Qe onto Irr ( ~) , for every e E E (S) , e =/= 0 . Let 1r : Fx ---+ S be the standard
presentation of S. (Thus X = Irr (S) and 1r is the identity on X.) Let A E 'D,
A =!= X' e = e A' and p : A' ---+ A' I eA ~ ~ be the projection. By Proposition
4.5, p(p~a) = 1r(a) for all a E EA. If e E A' and p(e) is irreducible in ~'
then e is irreducible in A' and e E X\A, e E HA. Thus Irr (Pe) ~ p(Y),
where Y = X n HA.
Let p E IA. When x E Y, then q = 1r(x) E Qe by Lemma 6.3 and
x + p E EA by Lemma 4.1, so that p(x) = 1r(x + p) = eq = e(q). In particular
p(Y) ~ Irr (~). If y E Y, 1r(y) = r, and p(x) = p(y), then e(q) = e(r),
q = r, 1r(x) = 1r(y), and x = y, since 1r: F---+ Sis the standard presentation
of S. Thus p induces a bijection of Y onto Irr ( ~) .
Let e = L:xEX\A ex x E HA. Since HA is a coideal of A', ex > 0 implies

x E HA; hence e = L:xEY ex x. If now e, d E HA, then p( e) = I1xEY p( x) ex =/= 0
and p(d) = I1xEY p(x)dx =/= 0. Since ~ is 0-free, a nonzero element of~ can
be written uniquely as a product of nonnegative powers of irreducible elements of

~ , and by the above can be written uniquely as a product of nonnegative powers
of elements of p(Y). Hence p(c) = p(d) -/= 0 implies ex = dx for all x E Y
and c =d. Therefore eA = ker p is the Rees congruence of JA. D
3. The following example is a semigroup which is not partially free even
though all its Ponizovsky factors are 0-free. This example also shows that A'---+
A' jeA ~ PA need not be the standard presentation of PA; and that, when PA is
0-free, eA need not be the Rees congruence of JA.
Example 6.5. Let X= {x,y,z} and 1) consist of 0 = {0}, A= F{x,y}•
B = F{z} =A', and F. Let E0 = {0, x, 2x, y}; EA = A\E0 , which is the ideal
of A generated by 3x, x+y, and 2y; EB = {kz, x+kz, 2x+kz, y+kz k > I
0}; and Ep = F\ (E0 U EA U EB), which is the ideal of F generated by 3x + z,
X + y + Z, and 2y + Z. It is immediate that 1) is a direction set and that C is
an extent cell family; we see that H 0 = E0 = {0, x, 2x, y}, 10 = 0; HA = 0,
IA = EA; HB = H 0 = {0, x, 2x, y}, IB = {kz Ik > 0}; and HF = 0,
IF= EF.
Let eD = Sn be the Rees congruence of JD for all D E 1), except for eB,
for which we also have 2x eB y. We see that (Tl) and (T2) hold, so Theorem
5.2 provides a congruence e on F with the given direction set, extent cell family,
and trace, in which a e b if and only if a, b E D and p'na eB p'nb for some
DE 1). LetS= Fje and 1r: F---+ S be the projection.
By Proposition 4.7, S has a presentation, as a commutative monoid with zero,
whereby Sis generated by x,y,z subject to: x 3 z = xyz = y 2 z = 0 (from Ep);
x3 = x 4 = x 3 y, xy = x 2 y = xy2 , y2 = y 3 = xy 2 (from EA); z = z 2 (from
EB ); and x 2 z = yz (from eB ).
By Proposition 4.5, the Ponizovsky factors of S are the nilmonoids D' jeD.
The choice of eD shows that all are 0-free except perhaps for B' jeB. Since
2x eB y, the latter is a cyclic nilmonoid {1, c, c2 , 0}, where c is the eB -class of
x, and is 0-free. Thus all Ponizovsky factors of S are 0-free. In particular, PB
is 0-free, even though eB is not a Rees congruence, and B' ---+ B' jeB ~ PB
is not the standard presentation of PB.
We have 0, x, 2x, y E E0 = H0 ; since Ci> is the Rees congruence of J0 ,
{0}, {x }, {2x} and {y} are e-classes. Hence 1r(x), 1r(y) E Irr (S). Similarly
IB is a e-class, and 1r(z) E lrr (S). Thus 1r: F---+ S is the standard presentation
of S. But eB is not a Rees congruence and S is not partially free. D
Chapter XI.
SUBCOMPLETE SEMIGROUPS.
This chapter contains a global construction of subcomplete semigroups, as

quotients of free commutative monoids; complete semigroups, finite semigroups,
subelementary semigroups, and finitely generated semigroups are particular cases.
In particular, this constructs all congruences on finitely generated free commuta-
tive monoids. The construction uses Ponizovsky families to generalize the results
in Chapter X and relates smoothly to related structural features including archime-
dean components, subdirect decompositions, X-classes, and extended Schtitzen-
berger functors. Its relationship to extension groups is less obvious and is shown
in Section XIII.2. A similar construction was obtained by the author for finite con-
gruences [1996C], then generalized to complete group-free congruences [2001C).
Subcomplete congruences are built from direction sets, extent cells, trace con-
gruences, strand groups, and strand bases. Explicit representations of subcomplete
congruences as intersections of subelementary congruences complete the chapter.
1. DIRECTION SETS.
This section defines direction sets for arbitrary congruences on a free com-
mutative monoid.
1. In what follows, F is a free c.m. and e is any congruence on F; for
instance, the congruence induced by a surjective homomorphism 7f : F ---+ S.
Directionality is a property of complete group-free congruences but does not
much extend to more general congruences, as later results will show. Accordingly
the general definition of direction sets uses semilattice congruences rather than
e-classes. Later sections also require a definition which is not tied to archimedean
components.
Proposition 1.1. Let e be a congruence on a free c.m. F, For a direction
set 1) on F the following conditions are equivalent:
259
260 XI. SUBCOMPLETE SEMIGROUPS.
(1) every A E 'D is a union ofe-classes;

(2) A* is a union of e-classes for every A E 'D;
(3) 'D = D(~) for some semilattice congruence ~ 2 e on F;
(4) 'D = D('rr- 1 ('B)) for some semilattice congruence 'B on S.
Proof. (1) implies (2) since A* =A\ (UAcBETI B). By Proposition X.1.3,
"#
the sets A* with A E 'D are the classes of the semi lattice congruence ~ on F
such that 'D = D(~); hence (2) implies that every ~-class is a union of e-classes,
and (2) implies (3). By Lemma X.l.l, every A E D(~) is a union of ~-classes
e ;
and is a union of e-classes if ~ ~ thus (3) implies (1 ).
(3) and (4) are equivalent by Proposition 1.2.6. A semilattice congruence 'B
on S induces a congruence ~ = 1r- 1 ('B) on F which contains e,
under which
a~ b if and only if 1r(a) 'B 1r(b);
~ is a semi lattice congruence since F /~ S=' S /'B. Conversely, a semilattice
congruence ~ 2 e on F induces a congruence 'B = 1r(~) on S, under which
1r(a) 'B 1r(b) if and only if a~ b;
'B is a semilattice congruence on S, since Sj'B s::- Fj~, and~= 1r- 1 ('B). 0
e
A direction set 'D on F is a direction set of when either condition holds;
if 'B is known, then 'D is the direction set 'D = D'B (actually, D(1r- 1 ('B))) of
e over 'B.
e
The archimedean direction set of is its direction set DN over the smallest
semi lattice congruence N on S. By Proposition X.2.8, the direction set of a c.g.f.
congruence as defined in Chapter X is its archimedean direction set.
In general, let B8 denote the 'B-class of s E S. Then B8 ~ Bt in the
semilattice S /'B if and only if B8 1\ Bt = B8 in S /'B, if and only if B8 Bt ~ B8
in S, if and only if st 'B s in S. When 'D = D'B, then t E D'B (a) if and only
if B1r(t) ~ B1r(a) in Sj'B; thus
D'B(a) = {t E F I a~ a+t} = {t E F l1r(a) 'B 1r(a+t)}.

From Section X.1 we obtain the following properties. By Proposition 1.1 and
Lemma X.1.1:
Proposition 1.2. Let e be a congruence on F and 'D be a direction set of
e. Every A E 'D is a union of e-classes and a face of F.
When a,b E F, then a E D(b) if and only if D(a) ~ D(b), if and only if
B1r(a) ~ B1r(b) in Sj'B; in particular, D(a) is the smallest direction face which
contains a.
By Proposition X.1.3 and Corollary X.1.4:
Proposition 1.3. Let e be a congruence on F and 'D = D'B be the direction
set of e over a semilattice congruence 'B on S = F /e. Then 'D, partially
ordered by inclusion, is an upper semilattice, in which D( a) V D(b) = D( a+ b).
For every C E 'D let
C* = { c E C I c ~ B when B E 'D and B ~ C};
equivalently, C* = {a E F I D( a) = c}. Then C* is a union of e-classes;

1r(C*) is a 'B-e/ass of S (an archimedean component of S, if'B = N) and every
'B-e/ass of Scan be constructed in this fashion; C f----7 1r(C*) is an isomorphism
('D, V) ~ Sj'B, with 1r(C*) ~ 1r(D*) in Sj'B if and only if C ~ D in F, and
C* + D* ~ (C V D)*, for all C,D E 'D.
Corollary 1.4. If S is finitely generated, then every direction set of e is
finite, contains F, and is closed under intersections.
Proof. If S is finitely generated, then S/'B is finite and ('D, V) ~ S/'B is
finite. D
2. EXTENT CELLS.
The central result of this chapter is that all subcomplete congruences have
extent cells.
1. Let F be a free c.m. and e be a subcomplete congruence on F; for
instance, the congruence induced by a surjective homomorphism 1r : F ---+ S,
where S is a subcomplete c.m.
When e is a c.g.f. congruence we saw in Chapter X that e has extent cells
{aEFID(a)=A},
one for every face A in the direction set, which correspond to the partial Poni-
zovsky factors of S (Proposition X.3.4). This definition cannot be used in general,
since, by Proposition 1.3, it would merely yield the Y-classes; rather, extent cell
families for e are obtained from Ponizovsky families of S.
Proposition 2.1. Let e be the congruence induced on F by a surjective
homomorphism 1r : F ---+ S. When S is subcomplete, then there exists a
semilattice congruence 13 on S and a Ponizovsky family P = (Pc)CES/'B

of S relative to 13, and then 'D = D13 is a direction set of e and the family
e= 1r- 1(P) = (EA)AE'D of subsets EA = 7r-l (P1r(A*)) ofF has the following
properties:
(Xl) EA is a union ofe-classes;
(X2) e is a partition ofF;
(X3) LA= U(EB I BE 'D, As;;; B) is an ideal ofF;
(X4) EA s;;; U (B* I As;;; B E 'D) and IA = EA n A* =I 0;
(X5) if As;;; B, a E LA, and c E IA, then a+c E EB if and only if a E EB;
(X6) ifcEIA, a,bELA,anda+c e b+c,thenaeb;
(X7) (\Ia E A*)(::Jn > 0) na E IA"
These properties imply:
(X8) IA = EA n A= LAnA is a nonempty ideal of A;
(X9) LA+ LB s;;; LAVB and IA + IB s;;; IAvB·
lf S is finitely generated and 13 = N, then in the above we may assume that
(XM) a 1 + ··· + am E IA for all a 1 , ... , am E A* and all A E 'D, when
m is sufficiently large.
Proof. By Proposition VII.2.2, S has a Ponizovsky family P ::= (Pc )cES/'13
relative to some semi lattice congruence 13, with the following properties:
(P 1) P is a partition of S;
(P2) Lc = U (PD I D E S/13, D ~ C) is an ideal of S;
(P3) Pc s;;; U (D E S/131 D s;;; C) and Kc = Pc n C =/: 0;

(P4) if C s;;; D, c E L0 , and k E K 0 , then ck E PD if and only if c E PD;
(P5) Kc is cancellative in Lc (if k E K 0 , c,d E L 0 , and ck = dk, then

c =d);
(P6) (Vc E C)(::Jn > 0) an E K0 .
By Proposition 1.3, A f----7- 1r(A*) is an isomorphism ('D, v) 9:! S/13, and
1r(A*) ~ 1r(B*) in S/13 if and only if A s;;; B in F. When A E 'D, then
C = 1r(A*) E S/13 and
EA = 7r-l(Pc) = UsEPc 7r-l(s)
is a union of t!-classes. Then (X2)-(X7) follow from (Pl)-(P6) since inverse

image under 1r preserves unions and intersections: thus E = (EA) AE'D is a
partition of F by (P 1);
U(EB I B E D, A ~ B)
7r-l (U(PD I D E Sj'B, c ~ D))
is an ideal ofF by (P2);
EA = 7r- 1 (Pc) ~ 7r-l (U (D E S/'B Ic ~ D)) = u(B* I A ~ B E D)

and
IA = EA n A* = 1r- 1 (PsC n C) = 1r- 1 (Kc) =/= 0
by (P3); then (X5), (X6), and (X7) follow from (P4), (P5), and (P6).
Now let D be a direction set of e and E be a family with properties (XI)
through (X7). If a E LA, then a E EB for some A~ BED and a E C* for
some B ~ C E D by (X4). If a E LAnA, then in the above C = D(a) ~ A
by Propositions 1.3 and 1.2, so that A = B = C and a E EA n A* = IA . Thus
IA = EAnA* ~ EAnA ~LAnA~ IA. This proves (X8).
(X9). Let a E LA, bE LB. By (X2), a+ b E E0 for some C E D.
Then A ~ C, since a+ b E LA by (X3), B ~ C, since a+ b E LB by
(X3), A VB ~ C, and a+ b E LAvE· Thus LA+ LB ~ LAvE· Then
IA + IB ~LA+ LB ~LAVE' IA + IB ~A*+ B* ~(A VB)* by Proposition
1.3, and IA + IB ~LAvEn (A VB)* = IAvB. This proves (X9).
If finally S is finitely generated and m is sufficiently large, then S has a
Ponizovsky family in which 'B = N and em
~ K 0 for all C E Y (S) (Proposition
VII.2.7); then a 1 +···+am E IA for all a 1 , ... , am E A* and all A ED. 0
2. When e is a congruence on F and D is a direction set of e, an extent cell
family of e over 'D is a family E = (EA) AE'D with properties (XI) through (X7)
in Proposition 2.1; the sets EA are the extent cells of e. We shall show that this
agrees with the definition of extent cells in Chapter X when e is complete group-
free. A sharp extent cell family is an extent cell family over the archimedean
direction set DN.
By Proposition 2.1, a finitely generated monoid has a sharp extent cell family
which also satisfies (XM). Example VII.2.5 shows that a subcomplete monoid
(in fact, finitely generated) may have more than one Ponizovsky family; hence a
subcomplete congruence may have more than one extent cell family.
In general:
Proposition 2.2. e be the congruence induced on F by a surjective

Let
homomorphism 1r : F ---+ S. Then e has an extent cell family e if and only if
it is subcomplete, and then e = 1r- 1 (P) for some Ponizovsky family P of S.
Proof. Let e = (EA)AED be an extent cell family of e over a direction set
']) = D'B of e, where ~ is a semilattice congruence on S. By Proposition 1.3,
A f-..--+ 1r(A*) is an isomorphism ('D, v) ~Sf~, and 1r(A*) ~ 1r(B*) in Sf~ if
and only if A ~ B in F. For every ~-class C let P0 = 1r(EA), where A E 'D
and 1r(A*) = C. We show that P = (Pc)CES/'B is a Ponizovsky family of S
relative to ~; then S is subcomplete by Corollary VII.3 .2, and e= 1r - 1 (P),
since in the above EA = 1r- 1 (Pc) by (XI).

Properties (PI) through (P6) follow from (X2) through (X7) since every EA
is a union of e-classes by (X I) and direct image under 1r preserves unions and
intersections of unions of e-classes. Thus P is a partition of S, since A f-..--+
1r( A*) is a bijection and e is a partition of F;
Lc = U(Pn IDE Sf~, D ~C)= 1r(U(EB I BE 'D, A~ B))= 1r(LA)

is an ideal of S, since LA is an ideal of F;
P0 = 1r(EA) ~ 1r (U (B* I A ~ B E 'D)) = U (D E Sf~ ID ~ C)

and Kc = Pc n C = 1r (EA n A*) = 1r(IA) =/= 0; then (P4), (PS), and (P6)
follow from (X5), (X6), and (X7). D
3. Propositions 2.I and 2.2 have a number of interesting particular cases.
Proposition 2.3. Let e be the congruence induced on
F by a surjective
homomorphism 1r : F ---+ S. If S = C U N is subelementary, then e has a
direction set ']) = {A, F}, where A = 1r - 1 (C), and an extent cell family e
over 'D in which EF = 1r- 1 (0) = Z is the zero class of e and EA = F\Z.
Then LF =IF= Z, LA= F, and IA =A.
Proof. C and N are the classes of a semilattice congruence ~ on S. By
Proposition VII.2.6, S has a standard Ponizovsky family P over ~ in which
Pc = S\0, Kc = C, Lc = S, and PN = KN = LN = {0}. Then e = 1r- 1 (P)
is as above. D
Proposition 2.4. Let e be a congruence on F and e = (EA) AED be an
extent cell family of e. Then e is a sharp extent cell family of e if and only if
IAfe is archimedeanfor every A E 'D.
Proof. Let e
be induced by 1r : F ---+ S. By Proposition 2.2, e = 1r- 1 (P)
for some Ponizovsky family P of S. By Lemma VII.2.3, Pis a sharp Ponizovsky
family if and only if every K 0 is archimedean. Hence E is a sharp extent cell

family if and only if IAje ~ 1r(IA) = K 0 is archimedean for every A E 'D,
where C = 1r(A*). D
By Proposition VII.2.4, a complete monoid has only one Ponizovsky family,
namely its partition into partial Ponizovsky factors. Hence:
Proposition 2.5. A complete congruence has only one extent cell family.
The extent cell family of a complete group-free congruence coincides with the
extent cell family constructed in Chapter X
The last part of the statement follows from Proposition X.3 .4.
homomorphism 1r: F---+ S, and E = (EA)AE'D be an extent cell family of e.
Then S is complete if and only if 1r(IA) is a group for every A E 'D, and then
the groups 1r(IA) are the maximal subgroups of S.
Proof. Let 'D = D'B , where ':B is a semilattice congruence on S. If S is
complete, then, by Proposition VII.2.4, S has only one Ponizovsky family, namely
its partition into partial Ponizovsky factors. Hence ':B = N and K 0 = He, where
e is the idempotent in the archimedean component C of S. If now A E 'D, then
C = 1r(A*) E Sj'B and 1r(IA) = K 0 = He.
Conversely assume that every 1r(IA) is a group. Then ':B = N, by Proposition
2.4. Moreover every ':B-class 1r(A*) contains an idempotent, the identity element
eA of 1r(IA). When a E EA, so that s = 1r(a) E P0 , we show that eA is the
least idempotent e of S such that es = s. Hence S is complete.
We have eA = 1r(c) for some c E IA. Then c e 2c. If a E EA, then
a + c e a + c + c and a e a + c by (X6), since a, a + c E LA by (X3);
hence eA s = s. Now assume that f s = s, where f E S is idempotent. Let
f = 1r(d), where dE DE 'D. Then a e a+d, a+d E EA by(Xl), and
D s;;: A by (X3). Hence c + d E IA by (X9), j eA E 1r(IA), f eA = eA since
f eA is idempotent, and f ~ eA. D
Proposition 2.2 also implies that a congruence which is not subcomplete does
not have an extent cell family. Any monoid which is not subcomplete, such as
Example VII.1.3, readily provides an example.
Example 2.7. Let S be Example III.5.2 (also Example VII.1.3), which
consists of all positive powers of s 0 , s 1 , ... , sP, ... , with s; = s~ whenever
n > p, q and s"; s~ = s~+n, where r = max (p, q) . This villainous c.s. has
no cancellative ideal and is not subcomplete.
Let F be the free c.m. with basis em,p (m,p EN, m > 0), and e be the
congruence on F generated by all pairs (em,p, en,q) such that m = n > p, q.
Then Fje ~ 8 1 is not subcomplete; by Proposition 2.2, e has no extent cell
family. This can also be verified directly.
S 1 has two archimedean components, {1} and S, so either 23 = N or 8 1
is the only 23-class. Let 23 = N. By Proposition 1.3, 'D = ~ has two elements
A ~ B, and B = F by Proposition X.l.7. Assume that e = (Ec)CE'D is an
extent cell family of e over 'D. By (X3), EF = IF is a nonempty ideal ofF.
Then 0 E EA, otherwise EF would contain 0 and EA would be empty. Hence
n(IF) s;;; n(EF) s;;; S. By (XS), IF is a nonempty ideal of F. By (X6) the
restriction of e to IF is a cancellative congruence. Then n(IF) is a cancellative
ideal of S, contradicting the choice of S.
If S 1 is the only 23-class, then similarly Dp, has only one element, namely
F, EF = F, and n(IF) is a cancellative ideal of S, which again contradicts
the choice of S. 0
3. We now give additional properties and a simpler characterization of extent
cells.
Lemma 2.8. Every extent cell family e= (EA) AETI satisfies:
(XlO) 0 E ID(O);
(Xll) the projections p~ E B with A s;;; B E 'D constitute a partition of the
complementary face A';
(X12) A+ EA s;;; EA;
(X13) if a E A, b,c E LA, and a+ b e a+ c, then be c.
Proof. (XlO). We have 0 E EA for some A E 'D by (X2). Then 0 E B*
for some A s;;; B E 'D by (X4). Now D(O) is the smallest direction face by
Proposition 1.2, since 0 belongs to every direction face; hence A = B = D(O)
and 0 E EAnA* = IA-
(Xll). Let t E A'. By (X4) there exists c E IA = LAnA*, and then

a = c + t E LA and p~a = t. Now a E E 8 for some A s;;; B E 'D, since
a E LA, and t E p~E8 .
To show that B is unique in the above, assume that t = p~ b for some b E LA.
Then p~(b+c) = p~b = t, b+c E LA by (X3), and PA(c+b) = c+pAb E IA,
since IA is an ideal of A by (XS). Hence we may assume that PA b E IA . Then
pAb+a = pAb+c+t = b+c; by (X5), a E EB implies pAb+a E EB,

b + c E EB, and b E EB . Thus b E EB whenever b E LA has p.A_ b = t.
(Xl2). Let a E EA and c EA. Then a+ c E LA by (X3) and a+ c E EB
for some A ~ B E 'D. Now p.A_ (a+ c) = p.A_a E p.A_EB n p.A_EA. By (Xll),
B = A, and a + c E EA"
(X13). Assume that a E A, b, c E LA, and a+ b e a+ c. Since IA is
a non empty ideal of A by (XS), there exists p E IA, and a + p E IA . With
b, c E LA and a+ b + p e a + c + p this implies b e c, by (X6). 0
Proposition 2.9. Let e be a congruence on F and 'D be a direction set of
e. A partition c = (EA)AE'D ofF is an extentcellfamilyfor e over 'D if and
only if:
(El) if a E EA, bE EB, and a~ b, then A~ B;
(E2) (Va E A*)(:Jn > 0) na E EA;
(E3) for every A E 'D, the projections PA EB with A ~ B E 'D constitute a
partition of A';
(E4) every EA is a union ofe-classes;
(E5) if c E A E 'D, a,b E EA, and a+ c e b + c, then a e b.
Properties (El)-(E5) are the same as in Lemma X.3.5.
Proof. When c is an extent cell family, (El), (E2), (E3), (E4), and (E5)
follow from (X3), (X7), (X11), (Xl), and (X13), respectively.
Conversely let c be a partition ofF with properties (El)-(E5). We show
that (Xl)-(X7) hold. First, (X2) holds; (Xl) and (X3) follow from (E4) and (E5);
(X7) follows from (E2), since a E A* implies na E A* .
(X4). Let a E LA and D(a) = B. Then a E B* and na E IB for some
n > 0, by (E4). Then na E LA by (X3) and A ~ B. (X7) also implies that
IA =!=- 0, since A* =!=- 0.
(X5). If A~ B E 'D, a E LA, and c E IA, then a+ c E LA; by (E3),
a + c E EB if and only if p~ a = p~ (a + c) E p~ EB, if and only if a E EB .
(X6). Let a, b E LA and c E IA. Assume that a+ c e b +c. Then a E E 0
for some A ~ C E 'D and b E En for some A ~ D E 'D. By (X5), a + c E E0
and b + c E En; hence C = D by (XI). Now a+ c e
b + c with c E A~ C
e
and a,b E E 0 , and (ES) yields a b. 0
Corollary 2.10. Let 'D be a direction set on a free commutative monoid
F. A partition c = (EA)AE'D ofF is an extent cell family of a subcomplete
268 XI. SuscoMPLETE SEMIGROUPS.
congruence if and only if it has properties (El ), (E2), and (E3). Then E is an
extent cell family of the complete group-free congruence S defined by
aS b if and only if a,b E EA and pA_a = pA_b for some A ED.
Proof. (El}-(E3) are necessary by Proposition 2.9. The converse follows
from Theorem X.3.7: if (El}-(E3) hold, then S is a c.g.f. congruence with
direction set 2) and extent cell family £ (by Proposition 2.5, the extent cells of
S defined in Chapter X coincide with the extent cells defined in this chapter). D
An extent cell family on F is a family £ = (EA) AE'D such that 2) is a
direction set on F and (El), (E2), (E3) hold; equivalently, an extent cell family
of a subcomplete congruence on F; equivalently, the extent cell family of a c.g.f.
congruence on F.
3. TRACE CONGRUENCES.
Now that extent cells are available, the construction of subcomplete congru-
ences continues, much as in Chapter X, with trace congruences.
1. As before e
is the subcomplete congruence on a free c.m. F induced by
a surjective homomorphism 1r: F---+ S; £ = (EA)AE'D is an extent cell family
of e.
For every direction face A E 2), let
HA = pA_EA ~A' and JA = A'\HA = U (pA_EB I BED, A~ B),
where the last equality follows from (E3). By Lemma X.4.1, HA is a coideal of
A' and JA is a nilmonoid ideal of A'. If F is finitely generated, then HA is
finite. Moreover HA + IA ~ EA.
Given a subcomplete congruence e
on F, or, more generally, an extent cell
family £ = (EA) AE'D, the strands are the S-classes, where S is the congruence
in Corollary 2.10:
a S b if and only if a, b E EA and p~ a = p~ b for some A E 2).
There is one strand for every A E 2) and t E HA = p~ EA, namely
SA(t) = {a E EA I p~a = t}.

Every extent cell has a partition into strands. A e-class is contained in a
single extent cell and typically intersects several strands; hence a e-class C is
3. TRACE CONGRUENCES. 269
determined by: (I) the strands which intersect it; (2) the partition induced on
these strands by e; and (3) which parts of various strands constitute C.
The trace of e specifies the strands which intersect a e-class. For every
A E 2) the trace congruence of e on A is the binary relation eA on A' defined
by:
teA u if and only if either t, u E JA, or t = pA_a, u = pA_b for some
a,b E EA such that a e b.
Thus two strands SA(t) and SB(u) contain elements that are equivalent modulo
e if and only if A = B and t, u are equivalent modulo eA. The trace of e
(relative to the extent cell family c, which determines the strands) is the family
'J = (eA)AE'D ·
Lemma 3.1. The trace of a subcomplete congruence has the following
properties:
(Tl) eA is a Jxcongruence on A';
(T2) ifAED, t,uEHA, teA u,andA~BED,thenpA_EB:t=pA_EB:u
in A' and pk(t + v) eB Pk(u + v) for every v E pA_EB: t S: A';
(T3) {0} is a exclass;
(T4) when t, u E HA, then teA u if and only if a+ t e b + u for some
a,b E IA"
Proof. (T4). Let t,u E HA. If teA u, then a+t e b+u for some a,b E A
such that a+ t, b + u E EA. Since IA is a nonempty ideal of A by (X8), adding
any c E IA to a and b yields a + c, b + c E IA and a + c + t e b + c + u.
If conversely a+ t e b + u for some a,b E IA, then a+ t, b + u E EA by
Lemma X.4.1 and teA u.
(Tl) Assume teA u and u eA v, where t,u,v E A'. If t, u, or vis in JA,
then all three are in JA and t eA v. Otherwise (T 4) yields a, b, c, d E IA such that
a+t e b+u and c+u e d+v. Then a+c, b+c, b+d E IA; a+c+t,
b+c+u, b+d+v E EA byLemmaX.4.1; a+c+t e b+c+u e b+d+v;
and t eA v. Thus eA is an equivalence relation.
Assume t eA u and let v E A'. If t, u E JA, then t + v, u + v E JA

and t + v eA u + v. Now let t,u E HA, so that t = pA_a, u = pA_b, where
a,b E EA and aeb. If t+v E JA, then a+v E LA, a+v ~ EA since
pA_(a + v) = t + v ~ HA, b + v ~ EA since EA is a union of e-classes, and
u +v = PA (b + v) E JA" Similarly u +v E JA implies t +v E JA" In either
case t+v eA u+v. Iffinally t+v, u+v tj. JA, then a+v, b+v E EA by
(E3) and a+ v e b + v, and again t + v eA u + v. Thus eA is a congruence.
It is clear that JA is a exclass.
(T2) Assume t = p_Aa, u = p_Ab, where a, b E EA and a e b, and let
A~ BE 'D. Ift+v Ep_AEB ~A', then p_A(a+v) Ep_AEB, a+v E EB by(E3),
b+ v EB since EB is a union of e-classes, and u + v = p_A (b + v) E p_AEB.
E
Then Pk(t+v) eB Pk(u+v), since a+v e b+v. Similarly u+v E p_AEB
implies t + v E p_AEB. Thus p_AEB: t = p_AEB: u.
(T3) Let 0 =1- t E HA. Then nt E JA for some n > 0 since JA is a nilmonoid
ideal of A'' and 0 eAt would imply 0 eA nt, 0 E JA by (Tl), JA = A'' and
HA = 0, contradicting EA =1- 0. 0
In (Tl) the quotient semigroup A' jeA is a nilmonoid, since JA is a nilmonoid
ideal of A' by Lemma X.4.1.
2. Conditions (Tl) and (T2) in Lemma 3.1 are identical to conditions (Tl)
and (T2) in Lemma X.S.l. Hence every family of congruences which satisfies
(Tl) and (T2) is the trace of a c.f.g. congruence: by Theorem X.5.2,
Proposition 3.2. Let£= (EA)AE'D be an extent cell family on F. Afamily
'J = (eA)AE'D of congruences is the trace of a subcomplete congruence on F
with direction set 'D and extent cell family £ if and only if it satisfies (Tl) and
(T2), and then it is the trace of the complete group-free congruence 9 defined
by:
a 9 b if and only if a,b E EA and p_Aa eA p_Ab for some A E 'D;
in fact 9 is the largest subcomplete congruence with direction set 'D, extent cell
family £, and trace 'J.
The last part of the statement follows from (T4).
A trace for an extent cell family £ = (EA)AE'D is a family 'J = (eA)AE'D
which enjoys properties (Tl) and (T2).
e
3. When 'J is the trace of a subcomplete congruence on F (relative to an
extent cell family £ of e), the congruence 9 in Proposition 3.2 is the complete
group-free hull of e relative to £; we denote it by e* . By definition,
a e* b if and only if a,b E EA and p_Aa eA p_Ab for some A E 'D.
By Proposition 3.2 and (T4) (applied to e*, which has the same trace as e),
Corollary 3.3. e* is the largest congruence on F with the same direction
set, extent cell family, and trace as e,
and it is a complete group-free congruence;
4. STRAND GROUPS. 271
moreover a e* b if and only if

a, b E EA and a +c e b + d for some A E '.D and c, d E !A-
Since e is induced by the surjective homomorphism 1r : F ---+ S, e* ~ e
induces a congruence 1r(e*) on S. We show that 1r(e*) is none other than the
congruence X in Proposition VII.4.3, induced by the corresponding Ponizovsky
family, under which s X t if and only if
s,t E P0 and ks = lt for some C E Sj'B and k,l E K 0 .
Proposition 3.4. Let e be the subcomplete congruence on F induced by a

surjective homomorphism 1f: F---+ Sand£= 1r- 1 (P) be an extentcellfamily
of e, where P is a Ponizovsky family of S. Relative to £, e* = 1r- 1 (X).
Proof. Let P = (Pc)CES/'.B, where 'B is a semilattice congruence on S.
When £ = (EA)AE'D = 1r- 1 (P), then, as in Proposition 2.1, '.D = D'.B, so
that A 1--------t 1r(A*) is an isomorphism ('.D, v) ---+ Sj'B, and EA = 1r- 1 (P0 ),
IA = 1r- 1 (Kc) whenever C = 1r(A*). Hence 1r(a) X 1r(b) if and only if
a,b E EA and a+ c e b + d for some A E '.D and c,d E IA, if and only if
a e* b. 0
Proposition 3.5. If s is finitely generated, then A' I eA is finite and FIe*

is finite.
Proof. If s is f. g., then FIe* ~ s/X is finite by Proposition VII.4.3.
Let eA_ be the congruence induced by the projection F ---+ A' ---+ A' jeA;
in other words, a eA_ b if and only if p~a eA p~b. Then FjeA_ ~A' jeA- Also
e ~ eA_ since a e b implies p~aeAp~b. Hence the nilmonoid A' jeA ~ FjeA_
is a homomorphic image of FIe ~ s' is f. g., and is finite. 0
4. STRAND GROUPS.
Strand groups constitute the next step in the construction of subcomplete con-
gruences; they determine the partitions induced by the congruence on its strands,
and the Schtitzenberger monoids of the quotient monoid.
1. As before e
is the subcomplete congruence on a free c.m. F induced by
a surjective homomorphism 1r: F---+ S; £ = (EA)AE'D is an extent cell family
of e. We denote by G the universal group of F, which is the free abelian group
with the same basis as F; if A is a face of F, then GA is the universal group
of A, which is the free abelian group with the same basis as A, and which we
regard as a subgroup of G.
When t E HA, let
IA(t) = {cEAic+tEEA} = {cEAic+tESA(t)}

and let et be the equivalence relation on IA (t) defined by:
c et d if and only if c + t e d + t.
Then IA = IA (0) and every element of the strand SA (t) = {a E EA I p~a = t}

can be written uniquely in the form a = c + t with c = PA a E IA (t) ; and et is
a copy of the restriction of e to SA (t).
Proposition 4.1. Let A E :D and t E Hk Then IA (t) is an ideal of A;
et is a cancellative congruence on IA (t);
~(t) {a-bEGAia,bESA(t) and aeb}
{ c - d E GA I c, d E IA and c + t e d + t}
is a subgroup of GA; when a,b E SA (t), then a e b if and only if a-bE~ (t);
and G(IA(t)fet) ~ GA/~(t).
Proof. This follows from Proposition 11.5 .I, according to which cancellative
congruences on a cancellative semigroup are induced by subgroups of its universal
group.
First, if c + t E EA and c;;;; dE A, then d +tELA since LA is an ideal of
F, and d+t E EA by (E3) since p~(d+t) = t E HA = p~EA- Thus SA(t) +A~
SA (t) and IA (t) is an ideal of A. Hence G(IA (t)) = G(A) = GA, by Proposition
11.2.5. Also et is a congruence on IA(t) since e is a congruence, and IA(t)fet
iscancellative: ifc,d,eEIA(t) and c+e et d+e,then c+e+t e d+e+t
and c + t e d + t by (E5).
By Proposition 11.5.1,
R = ~(t) = {c-d E G(IA(t)) I c,d E IA(t) and c+t et d+t}
is a subgroup of G(IA (t)); c + t et

d + t if and only if c- d E R, for all
c,d E IA(t); and G(IA(t)fet) ~ G(IA(t))/RA(t). Hence
RA (t) = {a-bE GA I a, b E SA (t) and a e b};
R is a subgroup of GA; when a, bE SA (t), then a e b if and only if a-bE R;
and G(IA(t)fet) ~ GA/~(t).
. 4. STRAND GROUPS. 273
Finally let c,d E IA (t). Since SA (t) +A ~ SA (t) and IA is an ideal of A,

we may add any e E IA to c and d, to the effect that c - d = ( c + e) - (d + e) ,
with c + e and d + e E IA and c + e + t e d + e + t. If conversely r = c - d
with c,d E IA and c + t e d + t, then c,d E IA(t) since HA + IA ~ EA
(Lemma X.4.1) andrE R.A(t). Thus
R.A(t) = {c-d EGA I c,d E IA and c+t e d+t}. D

R.A (t) is similar to but smaller than the Redei group of e; in fact, the Redei
group of e is the union of all R.A (t) (Proposition 5.4).
R.A(t) is the strand group oft E HA; the family~= (R.A(t))AE'D,tEHA
is the strand group family of e relative to its extent cell family e.
InProposition4.1 we denote the quotient IA(t)fet by IA(t)fRA(t). The et-
class of a E IA (t) is the intersection of IA (t) and the coset a+ R.A (t) of R.A (t)
in GA. It is convenient to view IA (t)j R.A (t) as a submonoid of GAl R.A (t); then
G(IA(t)fR.A(t)) = GAfR.A(t).
Similarly, when R is a subgroup of GA, we denote by A/ R the quotient
of A by the cancellative congruence = under which c = d if and only if
c - d E R. The equivalence class of c E A is the intersection of A and the coset
c + R of c in GA; hence we may view A/ R as a submonoid of GA/ Ra. Then
G(A/R) = GA/R, by Proposition 11.5.1.
2. Lemma 4.2. The strand groups of a subcomplete congruence e have the
following properties:
(Rl) R.A(t) is a subgroup of GA;
(R2) ijt,u E HA and teA u, then R.A(t) = R.A(u);
(R3) if A~ B E 'D, t E HA, v E A', and t + v E p~EB, then R.A(t) ~
RB (pk (
t + v)). (In particular, R.A (t) ~ RA (u) whenever t ~ u in HA.)
Proof. (Rl) follows from Proposition 4.1.
(R2). Let t,u E HA and teA u. If r E R.A(t), then r =a- b, where
a, b E IA and a + t e b + t. By (T4) we also have c + t e d + u for some
c,d E IA. Then
a + d + u e a + c + t e b + c + t e b + d + u,
a + u e b + u by (X6), and r = a - b E R.A (u) . Similarly, r E RA (u) implies
rER.A(t).
(R3). Let A~BE'D, tEHA, vEA1 , t+vEp~EB,and rER.A(t). Then
274 XI. 8UBCOMPLETE SEMIGROUPS.
r = c - d, where c, d E IA and c + t e d + t. Then c + t + v e d + t + v,

with c + t + v, d + t + v E E B by (E3) since PA (c + t + v) = PA (d + t + v) =
t + v E PAEB. Hence r E RB(p~(t + v)). D
In the next section we shall see that properties (Rl)--{R3) characterize strand
group families (Corollary 5.4).
3. We illustrate properties (R 1)--{R3) by two examples.
Example A. This is Example A from Chapter X: X = {x, y}, so that F
has four faces, of which two, B = F{y} and F, constitute the direction set in
Example A. The extent cells are
EB = {a E F I ax ~ 2} and Ep = {a E F I ax ~ 3};
hence HB = {0, x, 2x} and HF = {0}. The figure shows S. There are four
strands: SB(O) = B, SB(x), SB(2x), and Sp(O) = EF. The only possible trace
congruences are Rees congruences; hence S = e* for every subcomplete congru-
ence e with the given extent cells; FIe* is the cyclic nilmonoid { 1' X' x 2 ' 0}.
By (Rl), RB(O), RB(x), RB(2x) are subgroups of GB = 7/..,y, and Rp(O)
is a subgroup of G = 7/..,x E9 7/..,y. (R2) is trivial since all trace congruences are
Rees congruences. (R3) states that RB(O) ~ RB(x) ~ RB(2x) ~ Rp(O). D
Example A Example C
Example C. This example, like Example A, will be seen again in later

sections. Let X= {x,y,z} and 1) consists of A= F{x}' B = F{x,y}' and F.
This example is best visualized from its projection to A' above.
We have HA = { 0, y, z, y + z, 2z} and EA = {a E F I ay y + az z E HA};
EB ={a E F I ay
~ 2, az = 0} and HB = {0}; EF = F \ (EA U EB) and
H F = {0} . There is a trace relative to £ in which k eA l, where k = 2z and
l = y + z are the comer points of HA. Then F /9 = { 1, y, z, yz, e, 0}, where
1 is the identity element, yz = z 2 , y2 = e = ye = e2 , and all other products
are 0.
By (Rl), R:A.(t) is a subgroup of GA = Zx for every t E HA; R 3 (0) is a
subgroup of GB = Zx EB Zy; and Rp (0) is a subgroup of G = Zx EB Zy EB Zz.
By (R2), RA (k) = R:A (l). By (R3), RA (0) ~ R:A (y), RA (z) and RA (y), RA (z) ~
R:A (k) = R:A (l), RA (y) ~ R 3 (0), and R:A (k), R3 (0) ~ Rp(O). D
4. In some cases properties stronger than (Rl) hold, to the effect that the
groups RA (0) (and therefore all groups ~ (x)) cannot be too small.
Proposition 4.3. When e is a subcomplete congruence on F, e is a sharp
extent cell family if and only iffor every A E 'D and every a, b E IA, there exist
n > 0 and r E RA (0) such that na + r ~ b.
Proof. By Proposition 2.4, e is a sharp extent cell family if and only if every
IA/ e is archimedean. If IAj e is archimedean, then for every a, b E IA there
exist n > 0 and c E IA such that na e b + c; then r = b + c - na E ~ (0)
and na + r = b + c ~b. If conversely na + r ~ b for some r E ~(0), then
na + r E IA, since c E A and c ~ b E IA and na + r = b + c for some c E A,
na eb+c,and(n+l)a e
b+(a+c)witha+cEJA;ifthisholdsforall
a and b, then IAje is archimedean. D
homomorphism 7r: F----+ S, and e = (EA)AE'D be an extent eel/family of e.
Then S is complete if and only if every IA/ RA (0) is a group, if and only iffor
every A E 'D and g E GA there exists r E ~ (0) such that g + r E IA'
Proof. By Proposition 2.6, S is complete if and only if every IAje =
IA/ R:A (0) is a group. Since IA/ R:A (0) is cancellative, it is a group if and only
if IA/R:A(O) = G(IA/R:A(O)) = GA/RA(O), if and only if every coset of R:A.(O)
in GA intersects IA . D
In particular, S is complete group-free if and only if ~ (0) = GA for all
A E 'D.

homomorphism 1r : F ----+ S. If S is finite, then ~ (0) has finite index in GA
for every A E 'D.
Proof. If S is finite, then it is complete and every coset of ~ ( 0) in GA
intersects IA by Proposition 4.4. These intersections are the e-classes contained
in IA . Therefore ~ (0) has only finitely many cosets in GA . D
5. When a E EA, then A is uniquely determined by a by (X2) and it is
convenient to denote GA by Ga' SA (p~ a) by sa' and RA (p~ a) by Ra. Thus
Sa is the S-class of a and a e b implies Ra = Rn, by (R2); if aS b, then a e b

if and only if a - b E Ra = Rn .
Armed with this notation we show that the SchUtzenberger monoids and func-
tor of S can be reconstructed from the strand groups of e.
Proposition 4.6. When a E EA. ~(K1r(a)) ~ A/ Ra and r(K1r(a)) ~
GafRa.
Proof. As before, e is an extent cell family of e, and e = 1r - 1 (P), where
P is a Ponizovsky family of S. By Proposition 3 .4, e* = 1r - 1 (X) .
Recall that the Schiitzenberger monoid :E ( K) of a X-class K, relative to P,

is constructed as follows (Section VII.S). Let
U(K) = { u E S I uK ~ K }.
For every u E U(K) an injective mapping aK(u) : K --+ K well defined by
aK(u)(k) = uk
for all k E K. Then
~(K) = { aK(u) I u E U(K) };
~(K) is a commutative cancellative monoid which acts simply on K (if a, T E
~(K) and ak = Tk for some k E K, then a = T) (Proposition VII.5.2). The
Schutzenberger group of K is the universal group r(K) = G(~(K)).
Let K be the X-class of 1r(a). Then 1r(u) E U(K) if and only if 1r(u) 1r(a) E
K, if and only if a + u e* a. For the c.g.f. congruence e* this is equivalent
to u E D(a). Since e* has the same direction set and extent cell family as e
(Corollary 3.3), we have D(a) =A, since a E EA, and 1r(u) E U(K) if and
only if u EA.
If 1r(u), 1r(v) E U(K), then u,v E A, a+ u, a+ v E EA by (X12), and
a+ u, a+ v E Sa; hence aK(1r(u)) = aK(1r(v)) if and only if 1r(u) 1r(a) =
1r(v) 1r(a), if and only if a+ u e a+ v, if and only if u- v E Ra. Therefore
:E(K) ~ AfRa; the isomorphism sends aK(1r(u)) to 1ra(u), where 1fa: A--+
A/ Ra is the projection. This extends to an isomorphism r(K) ~ G(A/ Ra) =
GafRa. 0
Let 1fa : A --+ A/ Ra be the projection. When a ~ b in F, with, say, a E EA
and bE EB, then A~ B by (El) and Ra ~ Rn by (R3); therefore 1ra(c) = 1ra(d)
implies 1fb (c) = 1fb (d), and there is a homomorphism <pf; : A/ Ra --+ B / Rn which
sends 1ra(c) to 1rb(c) for every c E A. This in turn induces a homomorphism
'l/Jl;: Ga/Ra = G(A/Ra)---+ G(B/~) = Gb/Rb which sends g+Ra to g+Rb

for every g E Ga.
Lemma 4.7. Let a ;£ b in F, with a E EA, b E EB, and K = K1r(a)•
L = K1r(b). The following square commutes:
L.(L) ~ B/Rb
Proof. When K ~ L in SjX, then U(K) ~ U(L) and there is a unique
homomorphism L.f:
L.(K)---+ L.(L) which sends aK(u) to aL(u) (Proposition
VII.5.7). Now a;£ b implies 1r(a) ~ 1r(b) and K ~ L in SjX. For all u E A,
v E B, the isomorphisms L.(K) ~ A/Ra, L.(L) ~ B/~ send aK(u), aL(v)
to 1ra (u), 1rb (v) respectively, whereas z:.f
sends a K ( u) to a L ( v) and rpf; sends
1ra (u) to 1rb (v) . Hence the square above commutes. 0
Let C~ denote the e*-class of a E F. When a e* b, with a E EA, b E EB,
then A= B, p~a eA p~b, and Ra = Rb by (R2). Hence A/Ra and Ga/Ra
depend only on C~. Similarly, rpf; and 1/Jf; depend only on C~ and c; .
If a ;£ b in F, then C~ ~ Cb in F je*. If conversely C~ ~ Cb in F je*,
then Cb = C~ c; = C~+t for some t E F, with a+ t ~a. Thus C~ ~ Cb in
FIe* if and only if c; = c~, where a ;£ c in F. Therefore the monoids A/ Ra
and homomorphisms <pf; induce a monoid valued functor on F je*, the strand
monoid functor of e relative to the extent cell family E; and the groups Gal Ra
and homomorphisms 'l/Jl; induce an abelian group valued functor on F / e*, the
e
strand group functor of relative to E.
If S is complete, then A/ Ra = GA/ Ra for every a E A, by Proposition
e
4.4, since R.4 (0) ~ Ra; then the strand monoid functor of (relative to the only
extent cell family of e) is an abelian group valued functor and coincides with the
strand group functor of e.
In general Lemma 4. 7 yields:
Proposition 4.8. Let e be the subcomplete congruence on F induced by
a surjective homomorphism 1r : F ---+ S and E = 1r- 1 (P) be an extent cell
family of e, where P is a Ponizovsky family of S. Up to the isomorphism
Fje* ~ SjX, the strand monoid functor of e (relative to E) is isomorphic to
the Schutz en berger monoid functor of S (relative to P ), and the strand group
278 XI. SUBCOMPLETE 8EMIGROUPS.
functor of e is isomorphic to the extended Schatzenberger functor of s. lf s is

complete, then, up to the isomorphism FIe* ~ s j']{' the strand group functor
of e is isomorphic to the Schatzen berger .functor of S.
5. STRAND BASES.
When e is a subcomplete congruence, a strand base determines when two

elements of different strands lie in the same e -class; this is the final step in the
construction of e .
1. Given an extent cell family E: of e, a strand base of e is a cross-
section s = (sA (t)) AE'D, tEHA of S such that sA (t) E SA (t) for all t E HA and
sA(t) e sA(u) whenever t,u E HA and teA u.
A cross-section of S with these properties always exists: for every A E 1), we
can, for one t in every exclass, choose sA (t) arbitrarily in SA (t); then every
other sA (u) is unique modulo RA (u), by Proposition 4.1. This construction
suggests that the typical subcomplete congruence has many strand bases.
When a E EA it is convenient to denote sA(p~a) by sa; then sa E Sa, and
sa e sb whenever a e b (which implies a,b E EA and p~a eA p~b).
A subcomplete congruence is completely determined by its extent cells, trace,
strand groups, and a strand base:
Proposition 5.1. a e b if and only if a,b E EA, p~a eA p~b, and a-
sA (p~ a) - b + sA (p~ b) E R:A (p~ a), for some A E 1); equivalently, a e* b and
a - sa - b + sb E Ra.
Note that p~a eA p~b implies R:A(p~a) = R:A(p~b), by (R2).
Proof. Let a,b E EA; let a= c+t, b = d+u, where c,d E A, t = p~a E HA,
and u = p~b E HA. Let sa= q+t and sb = r+u, where q,r EA. Then
a- sa = c + t- sa = c- q. Also, c + sb, q + b E EA by (Xl2); in fact c + sb,
q+b E sb.
If a e b, then a, b E EA for some A E 1), and in the above t eA u by
definition of eA and sa = SA (t) e SA (u) = sb. Hence
c + sb e c + sa = q+a e q+b
and a - sa - b + sb = (c + sb) - (q + b) E En by Proposition 4.1, since c + sb,
q+bESb.
5. STRAND BASES. 279
If conversely teA u and a- sa- b + sb E Ra, then (c + sb)- (q +b) =

a- sa - b + sb E ~; c + sb e q + b by Proposition 4.1, since c + sb, q + b E Sb;
q+a = c + sa e c + sb e q + b;
and a e b, by (E5). D
Two strand bases (for the same strand group family) are equivalent when
they define the same congruence. By Proposition 5.1,
Proposition 5.2. Two strand bases s and t are equivalent if and only if
a e* b implies sa- sb- ta + tb ERa(=~).
2. From Proposition 5.1 a set of generating pairs for e (a presentation of
S ~ Fje) can be set up as follows. For every A E 'D and t E HA, let
IA(t) = {aEAia+tESA(t)} = {aEAia+tEEA}

and :RA (t) be the intersection of the Rees congruence of IA (t) and the cancellative
congruence on A induced by R.4 (t); thus
a :RA(t) b if and only if a= b, or a,b E IA(t) and a-bE R.4(t);
if a,b E IA(t), then a :RA(t) b if and only if a+t e b+t.
Proposition 5.3. A subcomplete congruence e on F = Fx is generated
by:
(1) for every A E 'D and t E HA, one pair (a+ t, b + t) for every generator
(a,b) of:RA(t); and
(2) for every A E 'D and t f
u E HA such that teA u, every pair (a, b)
which is minimal such that a E SA (t), b E SA (u), and a- sA (t) - b +sA (u) E
RA(t) = RA(u).
Proof. Let 13 be the smallest congruence on F containing all pairs ( 1) and
(2). Then 13 ~ e.
Let QA (t) be the binary relation on A defined by:
a QA(t) b if and only if a= b, or a,b E IA(t) and a+t 13 b+t.
Then QA (t) is a congruence on A, since IA is an ideal of A and 13 ts a
congruence; also, QA (t) ~ 1\A (t), since 13 ~ e. On the other hand QA (t)
contains every generator (a, b) of :RA ( t) , since 13 contains every pair in (1 ).
Therefore QA(t) = 1\A(t). Hence 13 and e induce the same partition of each
strand.
Now let a e b. By Proposition 5.1 we have a,b E EA, teA u, and a-
sA(t)- b+sA(u) = r E R.4(t) = RA(u), where t = p~a E HA and u = p~b E HA"
By the descending chain condition in F there is a pair (c, d) ~ (a, b) which is

minimal such that c E SA(t), dE SA(u), and c-sA(t) -d+sA(u) E Il:A(t) =
Il:A (u) . Then c 13 d. Since c ~ a and d ~ b, we have e = d + (a - c) E SA (u)
and a 13 e. Also
a-sA(t)-e+sA(u) = c-sA(t)-d+sA(u) E RA(u).
Since a- sA (t)- b +sA (u) = r E Il:A (u), it follows that b-eE Il:A (u). Hence
b e e. Then b 13 e, since 13 and e induce the same partition of SA (u), and
a 13 b. Thus 13 = e. D
If e is c.g.f., then Proposition 5.3 yields a presentation which is similar to but
less economical than the presentation in Proposition X.4.7. Indeed :RA (t) is the
Rees congruence of IA (t) and is generated by all pairs (m + x, m) where m is
a minimal element of RA (t) and x E X n A; and m is minimal in IA (t) if and
only if m + t is minimal in SA (t). Also one can construct a strand base merely
by picking one minimal element in each strand. With these choices, Proposition
5.3 yields the same pairs as Proposition X.4.7, except that we now have one pair
for every (t,u) with teA u, t # u, rather than one pair for every generator
of eA.
When F is finitely generated, then Proposition 5.3 provides a finite generating
set: Indeed (1) yields only finitely many pairs, since :RA (t) is finitely generated
by Redei's Theorem, and 'D and every HA are finite. Similarly, pairs (a, b)
which are minimal such that a E SA (t), bE SA (u), and a- sA (t)- b +sA (u) E
Il:A (t) = Il:A (u), constitute an anti chain in F x F, which is finite by Dickson's
Theorem (Corollary VI.l.3); since 'D and every HA are finite, (2) yields only
finitely many pairs.
3. Proposition 5.1 also yields the kernel function of e. By Proposition X.4.8,
the kernel function !* of e* is given by
c E f*(g) if and only if c E (EA :g+) n (EA :g-) njA(g) for some A E 'D,
where fA (g) p~ -l (fl (p~g)) and fi is the kernel function of

= eA" If F
is finitely generated (more generally, if JA is cofinite in A'), then Jl can be
constructed more precisely as in Proposition IX.4.9.
Proposition 5.4. Let e be a c.gj congruence on F. The Redei group of
e is
R = UAETI, tEHA Il:A (t) = UAETI RA (0)

The kernel function f of e is given by:
6. MAIN RESULT. 281
c E f(g) if and only if c E (EA: g+) n (EA: g-) n fA(g) and PA (sA (t)-
sA(u)) E pAg + R.A(t) for some A E 1>, where t = pA_(c + g+) and u =
pA_(c+ g-).
Proof. Every R.A (t) is contained in the Redei group R = {a-bEG I a e b}
of e, by Proposition 4.1.Conversely let r =a-bE R, where a e b. We have
a+ bE A for some A E 1) by Proposition 1.2. Then a,b E A, a+ c, b+ c E IA
for any c E IA, and r =(a+ c)- (b+ c) E R.4(0).
Let c E F and g E G. Then c E f (g) if and only if c + g+ e c + g- , if
andonlyifc+g+, c+g- EEA,PA(c+g+) eA pA_(c+g-),and
(c+g+)-sA(t)-(c+g-)+sA(u) E R.4(t)
for some unique A E 1>, by Proposition 5.1, where t = pA_(c + g+) and u =
PA (c + g-) . As in the proof of Proposition X.4.8, (pA_g) + = PA (g +) and
(pA_g)- =pA_(g-). When c+g+, c+g- EEA, thenpA_(c+g+) eA pA_(c+g-)
if and only if pA_c + (pA_g)+ eA pA_c + (pA_g)-, if and only if pA_c E Jl (pA_g).
Then t = pA_(c+ g+) E HA, u = pA_(c+ g-) E HA, t- u = pA_g, teA u,
RA (t) = R.A (u) , and (c + g+) - sA (t) - (c + g-) + sA (u) E R.A (t) if and only
if sA (t) - sA (u) E g + R.A (t) , if and only if
PA(sA(t)-sA(u)) EpAg+R.A(t),
since PA (sA (t)- sA(u)) = t- u = pA_g EpA_ (g + RA(t)) always. D
6. MAIN RESULT.
1. The final step in the construction of subcomplete congruences is the char-

acterization of strand bases. This leads to the main result in this chapter.
Lemma 6.1. Every strand base has the following properties.
(8+) if a e* b, then sa- sb- sa+c + sb+c E Ra+c for every c E F.
In particular,
(S) if A~ BE 1>, t,u E HA, teA u, v E A', and t+v E pA_EB, then
sA (t) -sA(u)- sB(p) +sB(q) E RB(p), where p= Pk(t+v) and q= Pk(u+v).
Proof. (S+) Let a e* b, so that a,b E EA and t = pA_a eA pA_b = u. By (T4),
d+t e e+u for some d,e E /A" Then Ra = Rn by(R2), d+t, c+e E EA
by Lemma X.4.1, and

(d + t)- sa- (e + u) + sb E Ra
by Proposition 5.1. Also c + d + t e c + e + u, Ra+c = Rb+c by (R2), and .
(d+t)- sa+c- (e+u) + sb+c = (c+d+t)- sa+c- (c+ e+u) + sb+c E Ra+c
by Proposition 5.1. Since Ra ~ Ra+c by (R3), subtraction yields (S+):
sa- sb- sa+c + sb+c E Ra+c·
Now assume A~ BE 'D, t,u E HA, teA u, v E A', and t+v E pA_E8 .
By (T4), c + t e d + u for some c,d E !A' Then a= c + t, b = d + u E EA,
pA_a = teA u = pA_b, and sa - sb - sa+v + sb+v E Ra+v, by (S+). Now
a+ v E E8 by (E3), since a+ v E LA and pA_ (a+ v) = t + v E pA_E8 ,
and b + v E E 8 since E8 is a union of e-classes. Hence sa+c = s8 (p) and
sb+c = s8 (q), where p = p~(a+v) = p~(t+v) and q = p~(b+v) = p~(u+v),
just as sa = sA (t) and sb = sA (u). This proves (S). 0
2. The main result in this chapter is:
Theorem 6.2. Let F be a free commutative monoid. Given a direction set,
an extent cell family, a trace, groups RA (t), and a strand base s such that (Rl ),
(R2), (R3), and (S) hold, define
a e b if and only if a,b E EA, pA_a eA pA_b, and a- sa- b+ sb ERa(=~).
Then e is a subcomplete congruence on F with the given direction set, extent
cell family, trace, strand groups, and strand base, and is the only such congru-
ence. Conversely every subcomplete congruence on F can be constructed in this
fashion.
Proof. The uniqueness and converse follow from Proposition 5.1. For the
direct part let 'D be a direction set on F, £ = (EA) AeD be an extent cell family
over 'D, 'J = (eA)AE1) be a trace for £, ~(t) (A E 'D, t E HA) be groups
such that (Rl), (R2), and (R3) hold, and s be a cross section of S such that (S)
holds. Define e as in the statement. As before, a, b E EA, PA PA
a eA b implies
e
Ra =~,by (R2). Then is contained in the congruence 9 in Proposition 3.2:
a 9 b if and only if a,b E EA and pA_a eA pA_b for some A E 'D,
which is the largest congruence with direction set 'D, extent cell family £, and
trace 'J; in fact
a e b if and only if a 9 b and a - sa - b + sb E Ra .
6. MAIN RESULT. 283
e is an equivalence relation since a 9 b implies Ra = ~ by (R2).

We show that (S+) holds for 9. Let a 9 b, so that a,b E EA and p~a eA p~b
for some A E 'D, and let c E F. Then a+ c 9 b + c, and a+ c, b + c E EB,
pka + Pkb eB pkb + pkc for some B E 'D, and A ~ B by (El). Hence
t = p~ a, u = p~ b E HA, v = p~ c E A', and t + v = p~ (a+ c) E p~ EB ; by (S),
sa- sb- sa+c + sb+c = sA(t)- sA(u)- sB(P) + sB(q) E RB(P) = Ra+c'
where p = Pk (t + v) = Pk (a+ c) and q = Pk (u + v) = Pk (b +c) since A ~ B.
Now assume a e b, so that a 9 b and a- sa- b + sb E Ra = Rb, and let
c E F. Then a+ c 9 b + c since 9 is a congruence, sa- sb- sa+c + sb+c E
Ra+c by (S+), Ra ~ Ra+c by (R3),
(a+ c)-sa+c- (b +c)+ sb+c

= (a- sa- b + sb) +(sa- sb- sa+c + sb+c) E Ra+c,
and a + c e b + c. Thus e is a congruence.
'D is a direction set of e: indeed 'D is a direction set of 9, so that 'D = D('~)
for some semilattice congruence ~ 2 9 2 e.
We show that E is an extent cell family for e. By Proposition 3.2, E is an
extent cell family for 9, so that (E 1), (E2), and (E3) hold. Every EA is a union
of e-classes, since every EA is a union of 9-classes. If finally c E A E 'D,
a,b E EA, and a+ c e b + c, then a+ c, b + c E EA by (Xl2), p~a =
p~(a +c) eA p~(b +c)= p~b, sa+c =sa, sb+c = sb,
(a+ c)- sa+c- (b +c)+ sb+c = a- sa- b + sb E Ra = Ra+c,
and a e b. Thus (E5) holds and E is an extent cell family of e, by Proposition
2.9. In particular e is subcomplete, by Proposition 2.2.
We show that 'J is the trace of e relative to E. Let t, u E HA. If c +
t e d + u for some c,d E IA, then c + t, d + u E EA by Lemma X.4.1 and
t = p~(c + t) eA p~(d + u) = u. Conversely assume that teA u. We have
sA (t) = q + t and sA (u) = r + u for some q, r E A. For any c E IA , we have
a= c + q + t, b = c + r + u E EA by Lemma X.4.1, sa= q + t, sb = r + u,
a - sa = c = b - sb, a - sa - b + sb = 0 E Ra, and c + q + t = a e b = c + r + u
with c + q, c + r E IA . Hence eA is the trace congruence of e, by (T4) in
Lemma 3.1, and T is the trace of e. Moreover, 9 = e* .
Finally lett E HA and c,d E IA. Then c+t, d+t E EA and c+t 9 d+t;
hence c + t e d + t if and only if c- dE RA (t). By Proposition 4.1, RA (t) is
284 XI. 8UBCOMPLETE 8EMIGROUPS.
the strand group of t. Thus the strand groups of e are the given strand groups.
Also sA(t) e sA(u) whenever t,u E HA and teA u, since S 8A(t) = sA(t) and
sA(t)- sA(t)- sA(u) + sA(u) E RA(t); hence s is a strand base for e. 0
Corollary 6.3. Let £ be an extent cell family on F. A family :R =
(RA(t))AE:D,tEHA is the strand group family of a subcomplete congruence e
with extent cell family £ if and only if (Rl ), (R2), and (R3) hold Then a cross
section of S is the strand base of a subcomplete congruence e with extent cell
family £ if and only if (S) holds.
Let £ be an extent cell family on F and 'J be a trace for £ . A family
:R = (JSt(t))AE:D,tEHA is a strand group family over£ and 'J when it sat-
isfies (Rl ), (R2), and (R3); equivalently, when it is the strand group family of
a subcomplete congruence with extent cell family £ and trace 'J; then a cross
section of S is a strand base for :R when it satisfies (S); equivalently, when it
is the strand base of a subcomplete congruence with extent cell family £, trace
'J, and strand group family :R.
3. Theorem 6.2 has some noteworthy particular cases.
Corollary 6.4. In Theorem 6. 2, FIe is cancellative if and only if 'D = {F};
FIe is subelementary if and only if 'D = {A, F} for some face A #- F of F
and RF(O) = G.
Proof. A cancellative monoid S has a Ponizovsky family consisting of just
S. Hence a cancellative congruence e on F has an extent cell consisting only of
EF = F. If conversely 'D = {F}, then IF = EF = LF and e is cancellative
by (ES).
Now let F 1e
be subelementary. By Proposition 2.3, e
has a direction set
'D = {A, F}, where A = 1r - l (C), and an extent cell family £ over 'D in which
EF =IF is the zero class of e; then Proposition 4.1 yields RF(O) = GF =G.
Conversely assume that 'D = {A, F} for some face A of. F of F and
RF(O) = G. Then LF = IF = EF, LA = F, and IA = A. In Fie,
C = Ale = !AI 1St (0) is a cancellative submonoid; the e-class EF is a zero
element, since EF = LF is an ideal ofF; N = (F\A)Ie is a nilsemigroup,
since for every a E F\A = F* we have na E EF for some n > 0 by (E2);
and Fie is the disjoint union Fie= C UN since A and F\A are unions of
e-classes. 0
Subelementary congruences are studied in greater detail in Section 7.
Corollary 6.5. Let F be a finitely generated free commutative monoid
6. MAIN RESULT. 285
Given a direction set, an extent cell family, a trace, groups ~ ( x), and a strand
base s such that (Rl), (R2), (R3), and (S) hold, define
a~ b if and only if a,b E EA, p~a ~A p~b, and a- sa- b + sb ERa(= J?n).
Then ~ is a congruence on F with the given direction set, extent cell family,
trace, strand groups, and strand base, and is the only such congruence. Con-
versely every congruence on F can be constructed in this fashion.
This holds since finitely generated commutative semigroups are subcomplete,
by Proposition VII. 1.1.
Corollary 6.6. In Theorem 6.2, c is sharp if and only if
(R4) for every A E 1) and every a, b E IA, there exist n > 0 and r E ~ ( 0)
such that na + r ~ b.
This follows from Proposition 4.3. We call ~ = (~ (t)) AE'D, tEHA sharp
when (R4) holds.
Corollary 6.7. In Theorem 6.2, F/~ is complete if and only if
(R4+) every !A/~ ( 0) is a group; equivalently, for every A E 1) and g E GA
there exists r E RA (0) such that g + r E /A-
This follows from Proposition 4.4. We call ~ = (~ (t)) AE'D, tEHA complete
when (R4+) holds.
Corollary 6.8. In Theorem 6.3, F / ~ is finite if and only if 1) is finite,
A' /~A is finite, and~ (0) has finite index in GA, for every A E fl.
Proof. These conditions are necessary by Corollary 1.4 and Propositions 3.4
and 4.5. If conversely 1) is finite, A' /~A is finite, and ~ (0) has finite index
in GA , for every A E 1), then every GA/~ ( x) is finite, a strand intersects only
finitely many ~-classes, there are only finitely many ~-classes contained in EA
that project to a ~xclass, and EA contains only finitely many ~-classes; hence
F /~ is finite. D
4. Examples will illuminate the results in this section.
Example A. In this example, X= {x,y}, 1) = {B,F}, where B = F{Y}'
EB = {a E F I ax :;::; 2}, and Ep = {a E F I ax ~ 3} . We saw that there
are four strands: SB(O) = B, SB(x), SB(2x), and Sp(O) = Ep. By Corollary
6.3, the strand groups are any subgroups RB(O), RB(x), RB(2x) ~ GB = Zy,
and Rp(O) ~ G = Zx ffJ Zy such that RB(O) ~ RB(x) ~ RB(2x) ~ Rp(O).
Condition (S) is trivial; hence a strand base consists of any s 0 E S B ( 0),
286 XI. SuBCOMPLETE SEMIGROUPS.
Example A
sx E SB(x), s 2 x E SB(2x), and s 3 x E Sp(O). Proposition 5.2 shows that all

strand bases define the same congruence e.
For instance let RB(O) = RB(x) = 6yZ, RB(2x) = 3yZ, and Rp(O) be
generated by x - y and 3y. We see that :RB (0) = :RB (x) is generated by (0, 6y)
and :RB (2x) is generated by ( 0, 3y); inspection shows that :Rp (0) is generated by
(3x, 3x + 3y) and (4x, 3x + y). Proposition 5.3 yields the following generators
fore: (0,6y), (x, 6x+y) (which maybe omitted), (2x, 2x+3y), (3x, 3x+3y)
(which may also be omitted), and (4x, 3x + y) .
Returning to the general case, we saw that e* = S. By Proposition 3 .4, the
quotient semigroup FIe has four X-classes: the cancellative monoid B I RB (0),
which is isomorphic toN if RB(O) = 0 and is a finite cyclic group otherwise; two
X-classes SB(x)le and SB(2x)le, which are similar though not subsemigroups;
and the cancellative semigroup Ep I Rp (0) . Thus FIe is subelementary if and
only if Epl Rp(O) is trivial, if and only if Rp(O) = G, as in Corollary 6.4.
By Corollary 6.6, c is a sharp extent cell family if and only if
(R4( A)) for every a, b E IA there exist n > 0 and r E .1\4 (0) such that
na + r "i;;b
holds for every A E 2). Here (R4(B)) holds since IB = B is archimedean. We
show that (R4(F)) holds if and only if r Y > 0 for some r = r x x + rY y E R =
Rp(O). If indeed (R4(F)) holds, then for a = 3x and b = 3x + y E IF = Ep
there exist n > 0 and r E R such that na + r ~ b, and then r Y > 0. If conversely
ry > 0 for some r E R, and a,b E Ep, then ax ~ 3 and na + kr ~ b when
k ~ by and 3n ~ bx - kr x, and (R4(F)) holds.
By Corollary 6.7, Fie is complete if and only if
(R4+(A)) IAI.l\4(0) is a group; equivalently, for every g EGA there exists
r E .1\4 (0) such that g + r E IA
holds for every A E 2). Here (R4+(B)) holds if and only if RB (0) -::f. 0. We show
6. MAIN RESULT. 287
that (R4+(F)) holds if and only if R = Rp(O) has an element r = r x x + r Y y

with positive coordinates rx > 0, ry > 0. If indeed (R4+(F)) holds, then
-y + r E lp = Ep for some r E R, and then r has positive coordinates. If
conversely r E R has positive coordinates and g E G, then kr + g E Ep when
k ~ 3 - gx and k ~ -gy, and (R4+(F)) holds.
By Corollary 6.8, F If- is finite if and only if GAl RA (0) is finite for all
A E 'D. Since GA is finitely generated, this happens if and only if RA (0) has
the same rank as GA, for every A E 'D . Thus FIe is finite if and only if
RB(O) "I 0 (equivalently, RB(O) ~ Zy has rank 1) and Rp(O) ~ Zx EB Zy has
rank 2. D
Example C. In this example, X = {x, y, z} and 'D consists of A = F{ x},
B = F{ x,y}, and F; HA = { 0, y, z, y + z, 2z} and EA = {a E F I ay y + az z E

I
HA}; EB = {a E F ay ~ 2, az = 0} and HB = {0}; Ep = F \ (EA U EB)
and HF = {0}. The figure shows the projections of the strands to A'.
Example C
We saw that there is a trace relative to e in which k f-A l, where k = 2z and

l = y + z are the comer points of HA. By Corollary 6.3, the strand groups are any
subgroups RA(t) ~ GA = Zx (one for every t E HA), RB(O) ~ GB = ZxEBZy,
and Rp(O) ~ G = Zx EB Zy EB Zz, such that ~(0) ~ ~(y), ~(z) and
~(y), ~(z) ~ RA(k) = ~(l), ~(y) ~ RB(O), and RA(k), RB(O) ~ Rp(O).
Condition (S) states that sA (k)- sA (l) E Rp(O). Thus a strand base consists
ofany sA(O) E SA(O), sA(y) E SA(y), sA(z) E SA(z), sA(k) E SA(k), sA(l) E
SA(l), sB(O) E SB(O), and sp(O) E Sp(O), such that sA(k)- sA(l) E Rp(O).
By Proposition 5.2, two strand bases r and s define the same congruence if
and only if rA(k)- rA(l)- sA(k) + sA(l) E RA(k) = RA(l).
For instance let ~(0) = RA(Y) = ~(z) = ~(k) = RA(l) = RB(O) = 3xZ,
Rp(O) = xZ and sA(k) = x + k, sA(l) = 2x + l. Proposition 5.3 yields the

following generators for e. Pairs in (1) are (0, 3x), (y, 3x + y), (z, 3x + z),
(k, 3x+k), (l, 3x+l), (2y, 2y+3x), and, from ::Rp(O), (2y+z, x+2y+z),
(y + 2z, x + y + 2z), and (3z, x + 3z); some of these may be omitted. In (2),
(t,u) = (k,l); when a= mx + k E SA(k) and b = nx + l E SA(l), then
a-sA(k)-b+sA(l) = (m-1-n+2)x E RA(k)
if and only if m = n- 1 (mod. 3); minimal such pairs are (k, x + l) and
(2x + k, l). Thus e is generated by (0, 3x), (2y + z, X+ 2y + z), (y +
2z, x+y+2z), (3z, x+3z), (k, x+l), and (2x+k, l).
In general, it follows from Corollary 6.6 that £ is a sharp extent cell family
if and only if (R4(D)) holds for every D E 1). As in Example A, (R4(A)) holds
since IA =A is archimedean, and (R4(B)) holds if and only if RB(O) r.J;_ Zy (if
r x -=1 0 for some r E RB(O)). Similarly, (R4(F)) holds if and only if r x > 0 and
r Y > 0 for some r E R = Rp(O) ). If indeed (R4(F)) holds, then for a= 3z and
b = x+y+3z E Ip = Ep there exist n > 0 andrE R such that na+r ~ b, and
then rx > 0, ry > 0. If conversely r E R has rx > 0, ry > 0, and a,b E Ep,
then az > 0 and na + kr ~ b when k ~ bx, k ~ by, and n ~ bz - kr z.
By Corollary 6.7, Fje is complete if and only if (R4+(D)) holds for every
D E 1). As in Example A, (R4+(A)) holds if and only if RA (0) -=1 0; (R4+(B))
holds if and only if r x > 0 and r Y > 0 for some r E R = RB(O) ); and (R4+(F))
holds if and only if Rp(O) has an element with positive coordinates.
By Corollary 6.8, F je is finite if and only if RA (0) s;:; Zx has rank
RB(O) s;:; Zx EB Zy has rank 2, and Rp(O) s;:; Zx EB Zy EB Zz has rank 3. D
7. SUBELEMENTARY CONGRUENCES.
Every subcomplete congruence e is the intersection of subelementary and

cancellative congruences. We now use the results from previous sections to write
e as an explicit intersection of subelementary and cancellative congruences, and
show how these congruences give rise to the direction set, extent cells, trace, and
strand groups of e.
1. First we inspect the construction of cancellative and subelementary con-
gruences provided by Theorem 6.2.
By Corollary 6.4, a subcomplete congruence e on F is subelementary if and
7. 8UBELEMENTARY CONGRUENCES. 289
only if it has a direction set 2) = {D, F}, where D =/= F, and RF ( 0) = G.

Let D =J F be a face of F . Then 2) = {D, F} is a direction set (by
Proposition X.1.7). An extent cell family E over 2) consists of EF = Z (the
zero class of e in Proposition 2.3) and ED = F\Z. Then LF = IF = Z,
LD = F, and ID =D. (El) is trivial when A= D and states that Z is an
ideal ofF when A = F. (E2) is trivial when A = D and states that
(Va E F\D)(3n > 0) na E Z
when A = F. (E3) is trivial when A = F and states that
H = Pb (F\ Z) and J = Pb Z constitute a partition of D'
when A= D. Then J is a proper nilmonoid ideal of D', by (E2), and Z = D + J

by (E3). If conversely J is a proper nilmonoid ideal of D' and Z = D + J,
then (E2) and (E3) hold (with H = D'\J).
We see that a S b if and only if either a, b E Z, or a, b E F\ Z and pba =
pbb; thus Sp(O) = Z and SD(t) = t +D for all t E H.
A trace over E consists of eF, which we can ignore since F' = {0}, and
'J = eD. Then (Tl) is trivial when A = F and states that 'J is a ]-congruence
on D' when A = D; then 'J is a nilmonoid congruence, since J is a nilmonoid
ideal. (T2) is vacuous when A = F and trivial when A = D, for then B = F.
A strand group family over 'J consists of groups R(t) = RD(t), one for every
t E H, and Rp(O). To obtain subelementary congruences we let Rp(O) = G
(Corollary 6.4 ). Then (R 1) is trivial when A = F and states that
(RID) R(t) is a subgroup of CD for every t E H
when A = D. (R2) is trivial when A = F and states that
(R2D) if t,u E Hand t 'J u, then R(t) = R(u)
when A = D. (R3) is trivial when A = F or B = F and states that
(R3D) if t,u E Hand t ~ u, then R(t) ~ R(u)
when A= B =D.
A strand base s consists of sp(O) E Z and s(t) = sD(t) E t + D, one for
every t E H. We may identify s with the family (s(t))tEH· Condition (S) is
trivial when A = F or B = F; when A = B = F, (S) states that
(SD) if t, u E H, t 'J u, v E D', and t + v E H, then s(t) - s( u) - s(t +
v)+s(u+v) E R(t+v);
in the above we have t + v 'J u + v and u + v E H, since 'J is J-congruence;

then R(t + v) = R(u + v) by (R2D). Corollary 6.4 now yields:
Proposition 7.1. Let: D I- F be a face ofF; J be a proper nilmonoid
ideal of D'; Z = D + J and H = D'\ J; 'J be a J-congruence on D';
R = (R(t))tEH be a family of groups such that (RlD), (R2D), and (R3D)
hold; and s = (s(t))tEH satisfy (SD) and s(t) E t + D for all t E H. Define
a eb if and only if either a,b E Z, or a,b E F\Z, p'va 'J p'vb, and a-
s (pba)- b + s (pbb) E R(pba) = R(pbb).
Then e is a subelementary congruence on F with direction set { D, F}, extent
eel/family {F\Z, Z}, trace 'J, strand groups ::R, and strand bases, and is the
only such congruence.
Conversely every subelementary congruence e on F can be constructed in
this fashion, with D = 1r - 1 C and Z = 1r - 1 ( 0) if e is induced by a surjective
homomorphism 1r : F ----+ S where S = C U N is subelementary.
In particular a subelementary congruence e has one nontrivial direction face
and one nontrivial trace congruence. In Proposition 7.1 we call D the direction
face of e, Z = EF the zero class of e, and H = PbEn = D'\J the trace
coideal of e.
Proposition 7.2. In Proposition 7.1, D' j'J is isomorphic to the monoid of
orbits of S.
Proof. We have a e* b if and only if pba 'J pbb. By Proposition 3.4,
D' j'J ~ SjX. Now X arises from the Ponizovsky family P in which Pc =
n(En) = S\0 and PN = n(EF) = {0}, which is the standard Ponizovsky family
of S, and is the orbit congruence on S, by Proposition VII.4.4. 0
2. Cancellative congruences on F are similar to subelementary congruences
but simpler. By Corollary 6.4, a subcomplete congruence e on F is cancellative
if and only if it has a direction set 1) = {F}. Then e has one strand SF(O) = F
and one strand group R = RF(O). Theorem 6.2 yields a e b if and only if
a- b E R; this is according to Proposition 11.5.1.
If in Proposition 7.1 we let D = F and J = 0, then H = D' = {0}, Z = 0, 'J
is the equality on D', ::R consists of one subgroup R(O) of G, conditions (R2D),
(R3D), and (SD) are trivial, and a e b if and only if a- b E R(O), as above.
When e is a cancellative congruence, it is convenient to let the direction face
of e be F and its trace co ideal be { 0} .
3. By Proposition VII.1.2, a subcomplete monoid S is a subdirect product
7. SUBELEMENTARY CONGRUENCES. 291
of cancellative semigroups and subelementary semigroups. From any Ponizovsky

family :P = (Pc)CES/'B of S, Proposition VII.4.1 constructs an explicit subdirect
decomposition of S: when C E S j'B and m E K 0 , the relation
a :P0 b if and only if either am= bm E Pc or am,bm tj. Pc
is a congruence on S and does not depend on the choice of m; S j:Pc is

cancellative if C is the least 'B-class of S, otherwise S j:Pc is subelemen-
tary; ncES/'B :Pc is the equality on S; and s
is a subdirect product of the
monoids S j:Pc .
If now e
is the subcomplete congruence on F induced by a surjective ho-
momorphism 1r : F ---t S, then e
is the intersection of congruences 1r -l (:Pc)
which can be constructed explicitly from the extent cell family c = n- 1 (:P).
Proposition 7.3. Let e
be a subcomplete congruence on F with direction
set 'D and extent cell family c. For any c E IA, the relation
a QA b if and only if either a + c e b+c E EA or a + c, b + c E MA =
UAcBEB
o;C
is a congruence on F and does not depend on the choice of c; QA is cancellative

if A is the greatest element of 'D, otherwise QA is subelementary; and = e
nAE'D QA.
If 'D has a greatest element A, then A = F: indeed every a E F belongs
to some B E 'D by (01), and then a E B ~A. Then MA = 0 and a QA b if
and only if a+ c e b + c, if and only if b- a E RF(O).
Proof. By Proposition 2.2, c= n- 1(:P) for some Ponizovsky family :P =

(P0 )cES/'B on S; when C = n(A *),then EA = n- 1 (Pc) and IA = n- 1 (Kc).
Thus c E IA if and only if n(c) E K 0 ; a+ c E EA if and only if n(a) n(c) E P0 ;
and a+ c E MA if and only if a+ c tj. EA, if and only if n(a) n(c) tj. P0
(since a+ c E LA). Hence a QA b if and only if n(a) :P0 n(b). Therefore QA
is a congruence on F and does not depend on the choice of c; FjQA 9:! Sf:Pc
is cancellative if A is the greatest element of 'D, in which case C is the least
element of Sj'B by Proposition 2.3, otherwise FjQA is subelementary; and = e
nAE'D QA, since ncES/'B :J>C is the equality on 8. 0
It is instructive to verify directly that QA is cancellative if A is the greatest
element of 'D, subelementary otherwise. Let 1rA: F ---t FjQA be the projection.
Lemma 7.4. A is a union of Qxclasses; nA(a) is cancellative in FjQA
for every a E A; if A =/= F, then the zero class of QA is ZA = MA: c; 1rA (b)
is nilpotent in FIQA for every bE F\A.
Proof. If a QA b E A, then b + c E IA by (X8), a+ c E IA since IA is a
union of e-classes, and a E A; hence A is a union of Qxclasses.
Assume that a+ u QA b + u, where u EA. If a+ c + u E EB for some
A ~ B, then a+ c E EB by (E3), since PB(a + c + u) = PB(a +c), and
a+ c, b + c E MA. If a+ c + u E EA, then a+ c E EA by (E3), since
p~(a+c+u) =p~(a+c),and a+c+u e b+c+u implies a+c e b+c
by (X 13) and a QA b. Hence 1rA (u) is cancellative in F I QA for every u E A.
Let ZA = MA : c = { z E F I z + c E MA } . If A =/= F, then A is not
the greatest element of 'D, the ideal MA of F is not empty, and ZA =/= 0. If
z,t E ZA, then z+c, t+c E MA and z QAt. If t QA z E ZA, then z+c E MA,
t + c E MA, and t E ZA. Thus ZA is an Qxclass. Also ZA is an ideal of F,
since MA is an ideal ofF by (El). Hence ZA is the zero element of FIQA.
If finally bE F\A, then bE B* for some BE 'D; nb E IB for some n > 0
by (E2); nb + c E lAvE by (X9); nb + c E MA, since nb + c E (A V B)\A
and A VB~ A; nb E ZA; and 1rA(b) is nilpotent in FIQA" 0
4. We now let e be any subcomplete congruence and show that the direction
faces, extent cells, trace, and strand groups of e are directly related to the direction
faces, extent cells, trace congruences, and strand groups of the congruences QA.
Proposition 7.5. When e is a subcomplete congruence on F: the direction
face of QA is A; the extent cells of e are the intersections of the extent cells of
all QB; the trace coideal of QA is HA; the trace congruence of QA is eA; the
strand group R(x) of QA is ~(x); and sA is a strand base of QA-
Proof. Let A E 'D and 1rA : F -----t F IQA be the projection. If A =/= F, then,
by Lemma 7.4, F I QA = C U N is subelementary and A = 1rA1 (C) ; hence A
is the direction face of QA. This also holds if A = F.
If A =/= F, then, by Lemma 7.4, z E F belongs to the zero ideal Z = ZA
of QA if and only if z + c E MA = UAcBE1> EB, if and only if p~z E JA =
7:
UAcBE1> p~EB, by (E3); hence J = p~ Z = JA and the trace coideal of QA is
7:
H = A'\J = HA- This also holds if A = F.
The extent cells of QB are its zero class ZB and F\ZB (or just F, if
B = F). Let A,B E 'D and d E lB. If A rt, B, then a E EA implies
a + d E LA v B by (X9), a + d E M B since B ~ A V B, and a E Z B ; thus
7. 8UBELEMENTARY CONGRUENCES. 293
EA ~ ZB. But if A B, then a E EA implies a + d E EB by (E3), since

~
p~(a+d) =p~dEp~EB, so that a+d ~ME and a~ ZB; thus EA ~ F\ZB.

In particular, for any a E EA, A is the smallest B E 'D such that a E F\ZB.
It follows that
EA = n(zB I EA ~ ZB) n n(F\ZB I EA ~ F\ZB) :

if a belongs to the right hand side, then A is the smallest B E ']) such that
a E F\ZB and necessarily a E EA.
Let 'J be the trace congruence of QA. Let t, u E A'. By (Tl), (T4),
t 'J u if and only if either t, u E JA or t, u E HA and a + t QA b + u for
some a,b E A. If t,u E HA and a+ t QA b + u for some a,b E A, then
a+ c + t e b + c + u E EA and teA u by (T4), since a+ c, b + c E IA.
If conversely t, u E HA and t eA u, then a+ t e b + u for some a, b E IA,
a+ c + t e b + c + u E EA by Lemma 4.1, a+ t QA b + u, and t 'J u. Thus
'J =eA.
Let t E HA. By definition, g E R(t) if and only if g = a- b for some
a,bEAsuchthata+t QA b+t;thenc+a+t e c+b+tEEA,withc+a,
c +bE IA, and g = (c +a)- (c +b) E RA(t). If conversely g E R.4(t), then
g=a-bforsomea,bEIAs uchthata+t e b+t,c+a+t e c+b+tEEA
by Lemma 4.1, a+ t QA b + t, and g =a-bE R(t). Thus R(t) = RA(t).
Finally let s be a strand base of e. Then sA (t) E SA (t) for every t E HA,
and t,u E H, teAu implies sA(t)esA(u) and sA(t)QAsA(u). Thus sA
serves as a strand base of QA. D
Proposition 7.5 suggests an explanation for Theorem 6.2. As noted above,
Proposition VII.4.1 represents any subcomplete congruence e on F as an inter-
section of subelementary and cancellative congruences. Proposition 7.1 can be
established directly. The direction set, extent cells, trace, and strand groups of e
then arise as in Proposition 7.5.
Chapter XII.
COMMUTATIVE SEMIGROUP COHOMOLOGY.
Commutative semi group cohomology assigns abelian groups Hn (S, G) to a

commutative semigroup S and an abelian group valued functor G on S.
Other cohomology theories have been considered for commutative semigroups
(see the introduction to Section 4). The theory we call commutative semigroup
cohomology is of particular interest because H 2 ( S, G) classifies commutative
group coextensions of S by G, that is, H 2 ( S, G) coincides with the extension
group Ext (S, G) in Chapter V; moreover, if S is complete group-free and G
is Schtitzenberger, then, as we saw in Chapter V, H 2 (S, G) classifies complete
commutative semigroups T with T j'J{ ~ S and Schtitzenberger functor isomor-
phic to G. These results make cohomology an important part of the structure
theory of commutative semigroups.
Commutative semigroup cohomology is an instance of triple cohomology,
which provides a definition in dimensions n ~ 3 as well as valuable properties.
After a brief account of triple cohomology and two sections of preliminary re-
sults, this chapter defines commutative semigroup cohomology, and gives simpler
definitions in low dimensions.
1. TRIPLE COHOMOLOGY.
This section gives, without proofs, the definition and main properties of triple
cohomology. We follow Beck [ 1967] and Barr & Beck [ 1969] but have renum-
bered cohomology groups in the more traditional fashion. We assume a general
knowledge of category theory and triples, from, say, MacLane [ 1971]; Grillet
[ 1999] also has a short account of triples and the tripleability of varieties.
1. The minimal requirements for cohomology are: a category e; a functor
e
V : ---+ e (normally denoted by G, but we use G for abelian group valued
functors); a natural transformation E : V ---+ le; and a contravariant functor A
295
296 XII. COMMUTATIVE SEMIGROUP COHOMOLOGY.
from e to the category Ab of abelian groups and homomorphisms.

For every 0 ~ i ~ n the natural transformation E : V ---+ le induces a
natural transformation En,i = yn-i Eyi : yn+l ---+ vn. To each object c of
e can then be assigned a complex of abelian groups
o ---t AVC ---t ... ---t Avnc ~ Avn+lc ---t ...
where 8n = "Lo~i~n (-l)iA~:2:,i: Avnc---+ Avn+lc. That 8n+l o8n = 0
for all n ~ 1 follows by a standard argument from the equalities En,j o En+l, i =
En,i o ~:n+l, j+l, which hold for all 0 ~ i ~ j ~ n. The n-th cohomology group
of C with coefficients in A is
Hn(C,A) = Ker8n /Im8n-l if n ~ 2, and H 1 (C,A) = Ker8 1 .
(In Beck [1967], Barr & Beck [1969], these groups are Hn-l and H 0 .)
In what follows V and E arise from an adjunction (JF, 1U, 'TJ, ~:) : U ---+ e,
where lF : u ---+ e is a left adjoint of 1U : e ---+ u, and 'T] : lu ---+ 1UJF,
E : lF1U ---+ le are the corresponding natural transformations. Then 11.' = 1UlF, 'TJ,
and 1-l = 1U~:lF constitute a triple on U; V = lF1U, E, and v = lF7]1U constitute a
cotriple on e 0
If for example e is the category of commutative semigroups and homomor-

phisms, and U is the category of Sets of sets and mappings, then the free c.s.
functor lF : Sets ---+ e is a left adjoint of the forgetful or underlying set functor
1U : e ---+ Sets; "lx embeds a set X into the free c.s. on X; E is described by
Lemma 4.1 below. Like all varieties, e is tripleable over Sets.
2. To obtain the Beck cohomology groups of an object S of e, one applies
the above to the category e = ets of objects over s; abelian group objects of
~ provide coefficient functors (Beck [ 1967]). The details are as follows.
Recall that an object over S in e is an ordered pair (C, 1r) of an object

C of e and a morphism 1r : C ---+ S. A morphism 'Y : ( C, 1r) ---+ (D, p) of
objects over S is a morphism 'Y : C ---+ D in e such that p o 'Y = 1r.
C ___]__, D
~l
s
/e
Every adjunction (JF, 1U, 'T], E) : u ---+ e lifts to an adjunction (iF, 1U, 7], E)
U = U-l-1US---+ etS =~;namely,
1. TRIPLE COHOMOLOGY. 297
JF(X,1r) = (lFX,1f), 1I.J(C,p) = (1I.JC, 1I.Jp), fj(X,-rr) = 'T/x, "E(C,p) = Ec,

where 1f : lF X ----+ S is the morphism such that 1l.J7f o 'T/ x = 1r. In particular,
the cotriple (V = lF1l.J, E, v = lF'T!1l.J) induced by (JF,1l.J,'T/,E) lifts to the cotriple
(V = lF1I.J, "E, v = 1Ffj1I.J) induced by (lF, 1U, fj, "E) on e
= e-J,.S. By definition,
V( C, 1r) = lF(1I.JC, 1l.J1r) = (VC, 1f), where 1l.J7f o 'Tluc = 1l.J1r;
"E(C,-rr) = c:: 0 ; and z;(C,-rr) = lFfj(lUC,lU-rr) = lF'T!uc = v0 . In particular, "E(6,-rr) =

c"~/. If e is tripleable over 1L (if the adjunction (JF, 1l.J, 'T/, E) is tripleable ), then
e is tripleable over 1L ( (lF, 1l.J, fj, "E) is tripleable ).
3. An abelian group object of a category is an object G of e
such that e
every Home (C, G) is a set, together with an abelian group operation + on every
set Home (C, G), such that A = Home (-,G) is a (contravariant) abelian group
valued functor on e; equivalently, such that
(g + h) 0 'Y = (g 0 'Y) + (h 0 'Y)
for all morphisms g, h : D ----+ G and 'Y : C ----+ D of e.
A morphism <p : G ----+ G' of abelian group objects of is a morphism of e
e such that the mapping Home (C, <p) : Home ( C, G) ----+ Home (C, G') ts a
homomorphism of abelian groups for every object C of e; equivalently,
<p 0 (g + h) = ( <p 0 g) + (<p 0 h)
for all C and g, h : C ----+ G. Then Home (-, <p) is a natural transformation of
abelian group valued functors.
When e
has finite products, abelian group objects of can also be defined e
by means of suitable morphisms m : G x G ----+ G, e : 0 ----+ G, i : G ----+ G,
as in MacLane [1971, first edition] (see also Lemma 2.2 below).
If for example G is an abelian group in the usual sense, then every Homsets
(C, G) is an abelian group under pointwise addition, and this makes G an abelian
group object of Sets since (g +h) o 'Y =(go 'Y) + (h o 'Y) for all g, h: C----+ G
and 'Y : D ----+ C. If conversely G is an abelian group object of Sets, then
HomSets({l}, G) is an abelian group, and the bijection Homsets({l}, G)----+ G
is readily used to make G an abelian group so that pointwise addition on every
Homsets (C, G) is the given operation.
In general, an action . of an abelian group object G of e on an object A of
e assigns to every object c e
of a group action
Home(C,G) x Home(C,A)----+ Home(C,A), (g,a) 1---t g.a
298 XII. CoMMUTATIVE SEMIGROUP CoHOMOLOGY.
of the abelian group Home (C, G) on Home (C, A) which is natural m C;

equivalently,
(g • a) o ry = (g o ry) • (a o ry)
for all ry : C ---+ D, g : D ---+ G, and a : D ---+ A.
If for instance G is an abelian group which acts on a set X in the usual
sense, then, for every set C, the abel ian group Hom8 ets (C, G) acts pointwise
on Homsets (X, G), and this is an abelian group object action in Sets since
(g • a) o ry = (g o ry) • (a o ry) for all g : C ---+ G, a : C ---+ X, and
ry:D--+C.
4. The ingredients for Beck cohomology are: an adjunction (JF, 1U, TJ, t:) :
U ---+ e; an object S of e; and an abelian group object G = (G,p) of
e e,
= e..!.S. For any object T = (T, 1r) of the Beck cohomology groups or
triple cohomology groups ofT with coefficients in G are its cohomology groups
calculated from the cotriple V above and coefficient functor A = Home (-,G) .
The triple cohomology groups of S are those of (S, 18 ).
For a more detailed definition, let Tn = (Tn, 1rn) = vnr. Then T0 = T,

7ro = 1r, and (Tn+l' 7rn+l) = V(Tn,7rn), so that Tn+I = lF1UTn and 7rn+l:
Tn+I ---+ S is the morphism such that 1U1rn+l o TfvTn = 1U1rn.
An n-cochain is an element of
that is, a morphism u : Tn ---+ G of e such that p o u = 1rn. Next, "E~i = t:~'i :
n,i n,i - . n - - n+l - -
Tn+I ---+ Tn and At:r = Hom-e(t:r , G) . C (T,G) ---+ C (T,G) sends
u : Tn ---+ G to u o t:~'i . Hence
_ i n,i . n - - n+I--
8n - L:o~i~n ( -1) At:r . C (T,G)---+ C (T,G)
sends u : Tn ---+ G to
8nu L:o~i~n (-l)i(uot:~'i).

By definition, Hn(T, G) is the n-th homology group of the cochain complex
0---+ C 1(-
T,G-) ---+ ... ---+ C n(-
T,G-) -----'8n-'--)- C n+I(-
T,G-) ---+ ...
An n-cocycle is an element of zn(T, G) = Ker 8n ~ cn(T, G). An n-

coboundary is an element of Bn(T, G) = Im 8n-I ~ zn(T, G) if n ;:; 2,
1. TRIPLE COHOMOLOGY. 299
B 1 (T,G) = 0 if n = 1. The Beck cohomology groups ofT are the groups

Hn(T,G) = zn(T,G)/ Bn(T,G).
The Beck cohomology groups of S are Hn ( S, G) = Hn ( ( S, 18 ), G) .
A morphism T : T --+ T 1 in e induces a homomorphism

Tn = Home-c~rT,G): cn(T1,G)--+ cn(T,G)
which sends u E Cn(T,G) to u o vnT. Since composition with VnT pre-

serves sums and f~,i is natural in T, we have Tn+l (8nu) = 8n (Tnu) for all
u E cn(T1,G); that is, (Tn)n:::::l is a chain transformation. Hence Tn takes
zn (T 1 , G) into zn (T, G) , takes Bn (T 1 , G) into Bn (T, G) , and induces a ho-
- -/- -- -
momorphism Hn(T,G) : Hn(T ,G) --+ Hn(T,G). Thus Hn(-,G) is a
contravariant abelian group valued functor on e.
Similarly, a morphism c.p : G --+ c' of abelian group objects induces a
- -- --1
homomorphism c.pn = Home-(Tn,c.p) : cn(T,G) --+ cn(T,G) which sends
u E cn(T,G) to c.p o u. Again c.pn+l (8nu) = 8n (c.pnu) for all u E cn(T,G),
- -- --1
and c.p induces a homomorphism Hn(T,c.p): Hn(T,G)--+ Hn(T,G ). Thus
Hn(T,-) is a functor. In fact Hn(T,c.p) is natural in T, so Hn(-,-) IS a
bifunctor.
5. The main properties of Beck cohomology are as follows.
Theorem 1.1. When T = lF X for some object X of'U, then Hn(T,G) = 0
for all n ~ 2; also H 1 (VC,G) ~ C 1 (C,G) for every object C of e.
- -1 -II
A sequence G --+ G --+ G of abelian group objects and morphisms of
e is short V-exact in case
0--+ Home-(VC,G) --+ Home-(VC,c') --+ Home-(VC,G 11 ) --+ 0
is a short exact sequence (in Ab) for every object C of e.

Theorem 1.2. Every short V -exact sequence e : G --+ G --+ G
- -1 -II
of
abelian group objects of q.s induces an exact sequence
.. · Hn(T,G)--+ Hn(T,c')--+ Hn(T,G 11 )--+ Hn+l(T,G)
which is natural in e.
Theorems 1.1 and 1.2 constitute Theorem 2 of Beck [ 1967].
Up to natural isomorphisms, Hn(T,-) is the only sequence for which The-

orems 1.1 and 1.2 hold (Barr and Beck [ 1969), Theorem 3 .3). Another useful
characterization of Hn (T, G) was given by Barr and Beck [ 1969] (Proposition
11.2). We give the contravariant version in Wells [ 1978]:
Theorem 1.3. Let

e : 0 ----+ Home (-,G) ----+ e1 ----+ ... ----+ en ----+ en+1 ----+ ...
be a complex of abelian group valued contravariant functors on and IHin bee

the n-th homology jUnctor of e. Assume that IHin (V T) = 0 for all T and n ~ 2,
that IHI1 (V T) is naturally isomorphic to C 1 ( T, G), and that there is for each
n ~ 1 a natural transformation Tn : en O V ----7 en such that Tn O enE = lcn.
Then IHin is naturally isomorphic to Hn (-, G) .
6. The fourth property of Beck cohomology requires additional definitions.

As above, let (JF,lU,7],E): u----+ e be an adjunction, s be an object of e, and
G be an abelian group object of e.,J,.S. A Beck extension of G by S (called a
G-module in Beck [1967], Definition 6) is an object E = (E,n) of = e.,J,.S e
together with an action . of G on E such that
(BE 1) lUn ocr = lus for some cr : US ----+ lUE;
(BE2) for every object C of e, the action of Home(C,G) on Home(C,E)

preserves projection to S: 1r o (g . e) = 1r o e whenever g : C ----+ G and
e: C----+ E in e;
(BE3) for every object C of e, Home(C,G) acts simply and transitively
on Home (C, E) : for every e,f : C ----+ E, there exists a unique g : C ----+ G
such that g • e = f .
A morphism t.p : E ----+ F of Beck extensions of G by S is a morphism in

e which preserves the action of G: t.p o (g • e) = g • (t.p o e) for all g : C ----+ G
and e : C ----+ E .
Beck ([ 1967], Theorem 6) proved that H 2 ( S, G) classifies Beck extensions

of G by S in the following sense:
e
Theorem 1.4. When is tripleable over U, there is a one-to-one correspon-
dence between elements of H 2 (S,G) and isomorphy classes of Beck extensions
of G by S, which is natural in G.
2. ABELIAN GROUP OBJECTS. 301
2. ABELIAN GROUP OBJECTS.
For a more concrete definition of triple cohomology for commutative semi-

groups we investigate abelian group objects in the category of commutative semi-
groups over a given commutative semigroup. The main result is from Grillet
[1991C], [1995C]; a similar result for monoids was proved by Wells [1978].
1. Let S be a commutative semigroup.
Proposition 2.1. Let e be the category of commutative semigroups. An
abelian group object of e_j_S is a split commutative group coextension of S. The
category (9 of abelian group objects of e-J_S is equivalent to the category A of
abelian group valued functors on H(S).
Proof. First we show:
Lemma 2.2. Let G be an abelian group object of a category e with finite
products. Let p 1 , p 2 : G x G ---+ G be the projections and
m = p1 + p2 : G x G ---+ G.
Then g + h = m o (g,h) whenever g,h: C---+ G.
Proof. (g, h) : C ---+ G x G is the morphism such that p 1 o (g, h) = g and
p 2 o (g, h) = h. Then g + h = (p 1 o (g, h)) + (p 2 o (g, h)) = (p 1 + p 2 ) o (g, h) =
m o (g,h). D
To probe c.s. over S we use the additive semigroup N+ and the following
construction. For every a E S let 1ra : N+ ---+ S be the unique homomorphism
-+ =(N + ,1ra) isac.s.overS.
suchthat7ra(l)=a,namely 7ra(n)=an;thenNa
Lemma 2.3. Let T = (T, T) be a commutative semigroup over S. For every

a E S let Ta = {t E T I Ti =a}. Evaluation at 1 is a bijection <p f------+ zp(l)
of Home (N~, T) onto Ta.
-+ -
Proof. If <p: Na ---+ T is a morphism, then To <p = 1ra and zp(l) ETa.
Conversely there is for every t E Ta a unique semigroup homomorphism t :
N+ ---+ T such that t (1) = t, namely, t f------+ tn ; then T o t = 1ra and t is a
morphism N~ --+ T in e. This defines mutually inverse bijections. D
2. Now let e be the category of c.s. and e

= e-J_S. Let G = (G,p) be
an abelian group object of e = e-J_S; in particular, G is a c.s., which we write
multiplicatively, and p is a multiplicative homomorphism.
We use Lemma 2.2 to construct a partial addition on G such that addition on

every Home-(C,G) is pointwise. In e,
the direct product p = (P,7r) = G X G
and its projections p 1 , p 2 : P ----+ G are given by the pullback
P~G
P2l f
G -----+ S
p
where P = {(x,y) E G x G J px = py}, p 1 (x,y) = x, and p 2 (x,y) = y.

Then m = p 1 + p 2 : P----+ G is a partial addition on G; x + y = m(x,y) is
defined if and only if px = py, if and only if x and y belong to the same set
G a = p - 1a. Thus m provides an addition on every G a .
Lemma 2.4. When G = (G,p) is an abelian group object ofe:

(x+y)(z+w) = xz+yw (1)
whenever px = py and pz = pw; addition on Home-(C,G) is pointwise; every
Ga is an abelian group under addition; for every a E S, g 1-----+ g(1) is an
additive isomorphism of Home (N~, G) onto Ga; and p is surjective.
Proof. Since m is a multiplicative homomorphism, (x + y)(z + w)

m(x,y) m(z,w) = m(xz,yw) = xz + yw whenever px = py and pz = pw.
Let C = (C,1r) and g,h: C----+ G. By Lemma2.2, g+h = mo (g,h). Now

(g, h) (c) = (g( c), h( c)) E P for every c E C, and (g + h) (c) = m (g( c), h( c)) =
g(c) + h(c). Thus, addition on Home-(C,G) is pointwise. Then the bijection
Home-(N~,G) ----+ Ga, <p ~-----+ <p(1) in Lemma 2.3 is an additive isomorphism.
Hence Ga is an abelian group under addition. In particular, Ga =/= 0 and p is
surjective. D
Lemma 2.5. Let Pa be the identity element of Ga under addition. An
abelian group valued functor AG = (G,')') on H(S) is defined by
1'a,t9 = 9Pt (2)
for every a E S, t E 8 1 , g EGa, where Pt = 1 E G 1 ift = 1 E 8 1 . Moreover,

G is a split group coextension of S by AG.
Proof. If t = 1 E S 1 , then ')'a t is the identity on G a ; otherwise ')'a t9 E Gat,

' '
since p is a homomorphism, and 'Ya.t is an additive homomorphism, since
2. ABELIAN GROUP OBJECTS. 303
9Pt+hpt = (g+h)(Pt+Pt) = (g+h)pt

for all g,h EGa, by (1). For all t,u E S we have
PtPu = (Pt + Pt)(Pu + Pu) = PtPu + PtPu

by (1), which in the abelian group Gtu implies PtPu = Ptu· This also holds if
1 -
t = 1 or u = 1. Hence '"Yat u O')'a t = '"Ya tu for all t,u E S . Thus AG = (G,')')
' ' '
is an abelian group valued functor on H ( S) .
The abelian group Ga acts simply and transitively on itself by left addition:
g • X = g + X. By (1 ),
(g • x) y = (g + x) (Pb + y) = gpb + xy = (1'a bg) • xy
'
for all g,x EGa andy E Gb. Thus (G,p,.) is a commutative group coextension
of S by AG, which splits since PaPb = Pab for all a,b E S. D
Lemma 2.6. Let <p : G --+ H = (H,CT) be a morphism of abelian group
objects. Then <p( Ga) ~ Ha for every a E S and A<p = ( 'PiGJaES is a natural
transformation from AG to AH.
Proof. By definition, <p is a multiplicative homomorphism, CT o <p = p, and
<p* = Hom-e(C,<p) is a homomorphism for every C. In particular <p(Ga) ~ Ha
for every a E S. Let 'Pa = 'PIGa : Ga --+ Ha be the restriction of <p to Ga. For
every a E S, Lemma 2.4 provides additive isomorphisms Home (N~, G) 9:! Ga,
-+-
Home (Na , H)
I""V
= Ha. The square
-+-
Hom-e(Na ,H) ~ Ha
commutes, since (<p*(g))(1) = (<p o g)(1) = <p(g(1)) = 'Pa(g(1)) for all g :

N~ --+ G. Since <p* is an additive homomorphism, it follows that 'Pa is an
additive homomorphism. In particular <pa (p a) = qa, the identity element of
the abelian group Ha. Since <p is a multiplicative homomorphism, we have
'Pat(9Pt) = (<pag) qt for all a,t E Sand g EGa; thus A<p = ('Pa)aES IS a
natural transformation from AG to AH. D
We now have a functor A : (') --+ A.
3. Conversely let ( G, 1') be an abelian group valued functor on H ( S) . Let
l denote 'Ya,t9 when g E Ga and t E 8 1 . As in Theorem V.4.1, there is a

split commutative group coextension (E, p, •) of S by (G, 'Y) , in which E is the
disjoint union UaES ( Ga x {a}) with multiplication
(g,a)(h,b) = (gb + ha, ab),

projection p : (g,a) a to S, and action g. (h,a) = (g + h, a) of Ga on
f----t
Ea = Ga x {a}. Then G = (E,p) is a c.s. overS.

Let C = (C,1r) be an object of e and g,fi: C----+ G be morphisms in e.
Then p o g = p o h = 7f. For every c E C there exist unique ge, he E Ga such
that g(c) = (ge,a) and h(c) = (he,a), where a= 1rc. Define
(g +h)( c) = (ge +he, a).

Lemma 2.7. With the addition defined above, G is an abelian group object
over S. Moreover, AG ~ (G, "f).
Proof. Let g, h : C ----+ G be morphisms in e. Since g and h are multi-

plicative homomorphisms, we have
(ged' ab) = g(cd) = g(c)g(d) = (ge,a)(gd,b) = (g~ + gd, ab),
where a = 1rc and b = 1rd, and 9ed = g~ + gd; similarly hed = h~ + hd. Hence
(g + h)(c) (g + h)(d) (ge +he, a)(gd + hd, b)

(g~+h~+gd+hd, ab)
(ged + hed' ab) = (g + h)(cd).
and g + h is a multiplicative homomorphism. We now have an addition on
Home (C, A) . It is immediate that Home (C, A) is an abelian group, in which
the identity element is c f----t ( 0, 1rc) and the opposite of g : c f----t (ge, 1rc) is
c f----t ( -ge, 1rc). Moreover (g +h) o 8 = (go 8) + (h o 8) for every morphism
8 : D ----+ C. Thus G is an abelian group object of e.
We show that AG ~ (G, 'Y) . Since G is an abelian group object, it induces, as
in the first part of the proof, an abelian group addition on every Ea. For every a E
S, Lemma 2.4 provides an additive isomorphism g g( 1) of Home (N~, G)
f----t
onto Ea. For every (g, a), (h, a) E Ea we now have homomorphisms g, h :
N~ ----+ G such that g(l) = (g,a) and h(l) = (h,a); since addition on
3. BECK EXTENSIONS. 305
Home (N~, G) is pointwise,
(g,a) + (h,a) = g(l) + h(l) (g + h)(l) (g+h,a).

Thus g ~-----+ (g,a) is an isomorphism of Ga onto Ea.
In particular, the identity element of Ea is pa = ( 0, a) . The homomorphism
Eat in the functor AG = (E,E) is then given by (2): Eat(g,a) = (g,a)(O,t) =
' '
(l, at) for all t E S. Thus AG is naturally isomorphic to (G, 'Y) . 0
Lemma 2.8. Let (cpa)aES be a natural transformation from (G,"f) to
(H,8). Then <p : G ---t H, defined by <p(g,a) = (<pag, a), is a morphism of
abelian group objects.
Proof. Since (<p a) aE s is a natural transformation, we have ( <p a g) t = <pat (gt)
whenever g E Ga. Hence <p is a multiplicative homomorphism:
<p ((g,a)(h,b)) cp(gb + ha, ab) = ('Pabl + 'Pabha, ab)

((<pag)b + (cpbh)a, ab) = <p(g,a) cp(h,b)
and a morphism in e.
Moreover, for any g, h : C ---t G, we have, with the
notation as above, cp(g(c)) =(<page, a), cp(h(c)) =(cpa he, a), and
(cpog+cpoh)(c) = ('Pa9c+'Pahc,a) = <p(gc+hc,a) = cp((g+h)(c));

thus <p is a morphism of abelian group objects. 0
We now have a functor([)): A---t 0; ([)l(G,"f) is G in Lemma 2.7, and
Lemma 2.8 constructs ([)) (<p a) aES = <p ·
4. We saw that A([))G is isomorphic to G. If conversely G = (G, 1r) is any
abelian group object of e,
then G is a split group coextension of S by AG and
G is isomorphic to ([)lAG as a semigroup over S and as an abelian group object.
If not exhausted our reader will verifY that these isomorphisms are natural, which
completes the proof of Proposition 2.1. 0
3. BECK EXTENSIONS.
Continuing Section 2 we now investigate Beck extensions in the category of

commutative semigroups over a given commutative semigroup. The main result
is from Grillet [ 1991 C), [ 1995C]; a similar result for monoids was proved by
Wells [ 1978].
1. Let S be a c.s. and ~ be the category of c.s. and homomorphisms.

Proposition 3.1. Let G = (G,p) be an abelian group object of ~.j..S and
G = AG = (G,"f) be the corresponding abelian group valuedfunctor on H(S).
A Beck extension of G by S is a commutative group coextension of S by G.
The category 'B of Beck extensions of G by S is isomorphic to the category e
of commutative group coextensions of S by G.
A morphism in e (necessarily an isomorphism) is an equivalence of commu-
tative group coextensions.
Proof. The proof is rather similar to that of Proposition 2.1. First we show:
Lemma 3.2. Let G be an abelian group object of a category ~ with finite
products and . be an action of G on an object E of ~. Let p 1 : G x E --+ G,
p2 : G x E --+ E be the projections and
q = p 1 • p 2 : G x E --+ E.
Then g . e = n o (g, e) for all g : C --+ G and e : C --+ E.
Proof. (g, e) : C --+ G x E is the morphism such that p 1 o (g, e) = g
and p2 o (g,e) = e. Since . is an action, g. e = (p 1 o (g,e)). (p 2 o (g,e)) =
(p 1 • p2 ) o (g, e) = n o (g, e) . D
2. Now let G be an abelian group object over a c.s. S and E = (E, 1r) be
a Beck extension of G by S, so that there is an action of G on E and (BE 1),
(BE2), (BE3) hold. In particular, E is a c.s., which we write multiplicatively,
and 1r is a multiplicative homomorphism. By (BE 1), 1r o CT = lus for some
mapping CT : S --+ E; hence 1r is surjective, and (E, 1r) is a coextension of S.
We use Lemma 3.2 to construct a partial action of G on E such that
Home (C, G) acts pointwise on Home (C, E) for every C E ~. In ~, the direct
product P = (P, r) = G x E and its projections p 1 : P --+ G, p 2 : P --+ E
are given by the pullback
where P = { (g,x) E G x E I pg = 1rx }, p 1 (g,x) = g, and p2 (g,x) = x. Then

q = p 1 • p 2 : P --+ E is a partial action of G on E, for which g • x = q (g, x) is
defined if and only if pg = 1rx, if and only if g EGa= p- 1 a and x E Ea = 1r- 1a
for some a E S; thus q provides a set action of Ga on Ea, for every a E S.
3. BECK EXTENSIONS. 307
Lemma 3.3. When E = (E, 1r:) is a Beck extension of S by G = (G, p) :

(g.x)(h.y) = gh.xy (3)
whenever pg = 1r:x and ph = 1ry; Home (C, G) acts on Home (C, E) pointwise;
the action of Ga on Ea is a simply transitive group action; and lEE = (E, 1r:, .)
is a commutative group coextension of S by G.
Proof. Since q is a multiplicative homomorphism, we have (g. x)(h. y) =
q(g,x) q(h,y) = q(gh,xy) = gh. xy whenever pg = 7r:X and ph= 1r:y.
Let C = (C,cr) and g : C ~ G, e : C ~ E be morphisms in e.
Then (g,e)(c) = (g(c),e(c)) E P for every c E C, and Lemma 3.2 yields
g • e = n o (g, e) and
(g. e)(c) = q(g(c),e(c)) = g(c). h(c).
Thus Home(C,G) acts on Home(C,E) pointwise.
For every a E S, Lemmas 2.4 and 2.3 provide an additive isomorphism g 1-----+
g(l) of Home (N;, G) onto Ga and a bijection e ~------+ e(l) of Home (N;, E)
onto Ea. Moreover (g. e)(l) = g(l) . e(l), since Home (C, G) acts on
Home (C, E) pointwise. Therefore the action of Ga on Ea is a group action,
since the action of Home (C, G) on Home (C, E) is a group action; and Ga
acts simply and transitively on Ea, by (BE3).
Let Pa be the identity element of Ga. Then Pa • x = x for all x E Ea, and
(3), (2) yield
(g • X) Y = (g • X) (pb • Y) = 9Pb • xy = '"Ya b9 • xy )
whenever g EGa, x E Ea, andy E Eb. Thus lEE= (E,1r,.) is a commutative

group coextension of S by G. D
Lemma 3.4. Every morphism cp: E ~ E 1 = (E 1,1r1 ) of Beck extensions
of S by G is an equivalence of commutative group coextensions of lEE to lEE'.
Proof. By definition, cp is a multiplicative homomorphism of E into E' ;
1r:1 o cp = 1r:; and cp o (g . e) = g . (cp o e) for all g : C ~ G and e : C ~ E.
In particular, cp(Ea) ~ E~ for every a E S. For every g EGa and x E Ea,
-+ -
Lemma 2.3 yields morphisms g : Na ~ G and x : Na ~ E such that
-+ -
g(l) = g and x(l) = x. Then cp o (g. x) = g. (cp ox) and evaluation at 1 yields
cp(g • x) = g • cp( x). Thus cp preserves the action of Ga. Then cp is a bijection
of Ea onto E~, since Ga acts simply and transitively on Ea and on E~, and
cp is an equivalence of commutative group coextensions. 0

We now have a functor lE : '.B --+ C. .
3. Conversely let (E, 1r,.) be a commutative group coextension of S by G.
Then E = (E,1r) is a c.s. overS. By (1), (2),
gh = (g+pa)(Pb+h) = 9Pb+Pah = tab9+tbah

, , = l+ha
for all g EGa and hE Gb; then
(g.x)(h.y) = gb.(x(h.y)) = (gb+ha).xy = gh.xy (4)

for all g E Ga, h E Gb, x E Ea, and y E Gb, since E is a group coextension.
Let C = (C,T) be an object of e and g : C --+ G, e : C --+ E be

morphisms in e.
Then p 0 g = 1r 0 e = T and g(c). e(c) is defined in E for
every c E C. Define
(g. e)(c) = g(c). e(c).
Lemma 3.5. With the action defined above, B(E, 1r, .) = E is a Beck
extension of S by G. Moreover, lEE is the given coextension (E,1r,.).
Proof. Since g and e are multiplicative homomorphisms, we have, by (4),
(g(c). e(c)) (g(d). e(d)) = g(c)g(d) . e(c)e(d) = g(cd). e(cd)
where a = TC, b = Td; hence g. e is a multiplicative homomorphism. Now
Home ( C, G) acts on Home (C, E) ; this is a group action since addition on
Home (C, G) is pointwise. (BE I) holds since 1r is surjective. Also
1r((g.e)(c)) = 1r((g(c).e(c)) = 1r(e(c))

for all c E C and (BE2) holds.
We show that Home (C, G) acts simply and transitively on Home (C, E) .
Let e,f : C --+ E. For every c E C there exists a unique g(c) E Ga such
that ](c) = g(c) . e(c), where a = TC. Since e and f are multiplicative
homomorphisms, we have
g( cd) . e( cd) J(cd) = ](c) ](d)

(g(c). e(c)) (g(d). e(d)) = g(c)g(d) • e(cd)
for all c, d E C, by (4 ). Hence g is a homomorphism. Then g is the unique
morphism C --+ G such that g. e = f. Thus E (with the action . of G on
E) is a Beck extension of S by G.
4. COMMUTATIVE SEMIGROUP COHOMOLOGY. 309
Since E is a Beck extension, it induces, as in the first part of the proof,

a simply transitive group action of Ga on Ea, which makes E a commutative
group coextension lEE of S by G. Let a E S. For every g E Ga and e E Ea ,
Lemma 2.3 yields morphisms g: Ca---+ G and e: Ca---+ G such that g(l) = g
and e(l) = e. Then the action of Ga on Ea in lEE satisfies
g.e = g(l).e(l) = (g.e)(l);
so does the action of Ga on Ea in the given coextension (E, 1r,.) . Hence the
two actions coincide and lEE = (E, 1r,.) . 0
Lemma 3.6. Every equivalence of commutative group coextensions from
(E, 1r, .) to (E', 1r1 , .) is a morphism of Beck extensions from llll(E, 1r, .) to
Jill (E'' Jr'' . ) .
Proof. Let <.p be an equivalence of commutative group coextensions from
(E, 1r,.) to (E', 1r 1,.) • Then <.p is a multiplicative homomorphism and preserves
projection to S and action of G. For every g : C ---+ G and e: C ---+ E we
then have ~..p(g(c).e(c)) =g(c).<.p(e(c)) and ~..po(g.e) =g.(<.poe). Thus <.p
is a morphism of Beck extensions. 0
4. We now have a functor Jill: E---+ 'B. We saw (Lemma 3.5) that lEllll = lc:.
If conversely E is a Beck extension of G by S, then the action of Home (C, G)
on Home ( C, E) is the pointwise action induced by the action of G on lEE
and coincides with the action of Home (C, G) on Home (C, JilllEE) . Hence
lllllEE = E. Thus Jill and lE are mutually inverse isomorphisms. 0
4. COMMUTATIVE SEMIGROUP COHOMOLOGY.
Triple cohomology in the category e of commutative semigroups provides

a good definition of commutative semigroup cohomology. This section brings a
more concrete definition, based on the results in Sections 2 and 3. The results
are from Grillet [1991C], [1995C].
Other cohomology theories have been considered for commutative semigroups.
Inasaridze extended the construction of n-extensions and Extn from abelian
groups to commutative cancellative mono ids [ 1964], [ 1965] and to commutative
Clifford semigroups [ 1964], [ 1967]. Kruming [ 1982] characterized finite com-
mutative semigroups whose Eilenberg-MacLane cohomology vanishes; Novikov
[ 1990] showed that cancellative c.s. with this property are subsemigroups of N.
See also Carbonne [ 1983]. For a survey of semigroup cohomology in general, see
Grillet & Novikov [2002].
1. Let Fx be the free c.s. on a set X, which we write multiplicatively. For
what follows it is best to regard the elements of Fx as commutative words in
X, which are non empty unordered sequences [x 1 , ... , x ml of elements of X ;
unordered means
[xa1'oo·,xaml = [x1,.oo,xmJ
for every permutation (}'. It is customary to write [x 1 , ... , xml as a product
x 1 · · · xm, but this would quickly become very confusing in what follows. Mul-
tiplication in Fx is by concatenation:
[x1,oo.,xm] [y1,.oo,Ynl = [x1,.oo,xm, Y1,.oo,Ynl·

The injection 'T/x :X ~ Fx takes x EX to [x] E Fx. Every mapping of a
set X into a c.s. factors uniquely through 'T/x.
The free c.s. functor lF: Sets~~ assigns to a set X the free c.s. lFX = Fx
on X, and assigns to a mapping f : X ~ Y the unique homomorphism
lFf : lFX ~ lFY such that lFf o 'Tlx = 'T/y of; lFf sends [x 1 , x 2 ,. 00, xml =
[x 1] [x 2 ]oo· [xml E lFX to [jt 1 , ft 2 , oo•, ftml E lFY. lF is a left adjoint ofthe
forgetful functor 1U : ~ ~ Sets.
Lemma 4.1. In the adjunction (JF,1U,'T/,E): Sets~~' Er: lF1UT ~ T
sends [t 1 , t 2 , ... , tmJ E IF1IJT to t 1 t 2 · · · tm E T, for every c.s. T.
Proof. Since (JF, 1U, 'T/, E) :Sets~~ is an adjunction, we have 1UEr o '17vT =
lur (see e.g. Theorem IV. I in Mac Lane [ 1971 ]). Thus Er : lF1UT ~ T is the
homomorphism such that Er [t] = Er ('17vr t) = t for all t E T. Hence Er sends
[t 1 , t 2 , ... , tmJ = [t 1] [t 2 ] · · · [tmJ E lF1UT to t 1 t 2 · · · tm E T. D
In the cotriple (V,E,v) induced by the adjunction (lF,1U,'T/,E): Sets~~'
V = lF1U sends a c.s. T to the free c.s. VT = lF1UT on the set T. If f : T ~ T'
is a homomorphism, then V f : [x 1 ,. 00, xml 1-----7 [fx 1 , 00. ,Jxml· Lemma 4.1
describes Er : VT ~ T; v will not be used.
For every c.s. S the cotriple (V,E,v) lifts to a cotriple (V,€,'D) one= ~.j..S;
if T = (T, T) is a c.s. over S, then VT = (VT, 7'), where 1U7' o '17vr = 1UT;
that is, r[t] = Tt for every t E T and
7'[x1,oo·,xml = TX1oo•TXm = T(X1·"Xm)
for all m > 0 and x 1 ,oo.,xm E T.
2. The next Lemma describes Home-(VT, G) when G is an abelian group

object of e.
When G = (G, r) is an abelian group valued functor on H ( S),
and T = (T, T) is a c.s. over S, let
C(T,G) = IltET GTt;

C(T,G) consists of all families u = (ut)tET such that ut E G 7 t for all t E T,
under pointwise addition (that is, 1-cochains on T with values in G).
Lemma 4.2. Let G = (G, p) be an abelian group object of e and G be
the corresponding abelian group valued functor on H(S). For every object T
oje there is a natural isomorphism Home-CVT,G) ~ C(T,G).
Proof. By Proposition 2.1 we may assume that G is the split commutative
group coextension UaES Ga x {a} of S by the corresponding abelian group
valued functor G. Then addition on Hom-e(C,G) is as follows: if J,g: C---+ G
and f(c) = (!c,a), g(c) = (gc,a), then (! + g)(c) = Uc + Yc, a).
We have V T = (VT, 7), where VT is the free c.s. on the set T and
7[x 1, ... ,xml = T(x 1 ···xm) for all m > 0 and x 1 , ... ,xm E T, in particular
r[t] = Tt for all t E T. Iff: VT---+ G is a morphism in then p of= 7 e,
and there is for every t E T a unique ut E G::r[t] = G Tt such that f [t] = ( ut, Tt).
Since VT is free on T there is for every u = (ut)tET E C(T,G) a unique

semigroup homomorphism f : VT ---+ G such that f[t] = (ut, Tt). Then
f [x 1, ... ,xmJ = (ux 1 ,Tx 1 )···(uxm'Txm) = (L; 1 ~i~m u~;j, Tx),
where x.J = . 1 x J.+ 1 · · · x m and x

x 1 · · · x J- = x 1 · · · x m' for all m > 0 and
x 1, ... ,xm E T. (If m = 1, then xj is an empty product and xj = 1 E T 1 .)
This provides a bijection 8: Home-(VT,G) ---+ C(T,G). If f[t] = (ut,Tt)

and g[t] = (vt, Tt), then (!+g) [t] = (ut + vt, Tt); hence 8 is an isomorphism.
It is immediate that 8 is natural in T and G. D
3. The triple cohomology ofT= (T,T) E e,
and the triple cohomology of
S, which is that of ( S, 18 ) , can now be obtained as follows. Let G = (G, p) be
an abelian group object of e.
Let Tn = (Tn, 7rn) = vnT. Then To= T, 1ro = T (To= S, 1ro = 1s for
the cohomology of S); Tn+ 1 = VTn is the semigroup of all commutative words
[x 1, ... ,xmJ with m > 0 and x 1 , ... ,xm E Tn, and
7rn+l [x1, · · · ,xmJ = 1rn (x1 · · ·xm) E S. (1r)
For all 0 ~ i ~ n the morphisms ETi = E~,i : T n+l -+ T n m e,

E~,i = -yn-i EyiT may be defined by induction:
En,n [x 1, ... ,xm] = Eynr [x 1 , ... ,xm] = x 1 ·"Xm E Tn (E1)
for all [x 1, ... ,xml E Tn+ 1 , by Lemma 4.1; when i < n, E~,i = V E~- 1 , i and
En,i [x1, ... ,xm] = [En-1,ix1' ... 'En-1,ixm] E Tn (E")
for all [x 1, ... , xmJ E Tn+l.

The equality
7rn o En,i = 7rn+1 (7rE)
is proved by induction on n: for every [x 1, ... ,xml E Tn+ 1 ,
1rn En,n [x1, ... ,xm] = 1rn (x1···Xm) = 7rn+1 [x1, ... ,xm]
and
= ,.,.
"n '-
["'n-1,ix 1' ... , .cn-1,ix
'- m
]
7rn-1 (En-1,ix1 ... En-1,ixm) = 7rn-1 En-1,i (x1 .. ·xm)
1rn (x1"'xm) = 7rn+1 [x1, ... ,xm]

for all i < n, since En,i is a homomorphism.
An n-cochain is an element of cn(T, G) =Home (T n' G). The coboundary
homomorphism
i n,i - . n - - n+1 - -
on = L:o~i~n (-1) Hom-e(ET ,G). C (T,G)-+ C (T,G)
sends v : T n -+ G to
OnV L:o~i~n (-l)i(voE~'i).

An n-cocycle is an element of zn(T,G) = Keron ~ cn(T,G); if n ~ 2, an
n-coboundary is an element of Bn(T,G) = Imon_ 1 ~ zn(T,G). The Beck
cohomology groups of T are the groups
Hn(T,G) = zn(T,G)/Bn(T,G)
where n ~ 2, and H 1(T,G) = Z 1(T,G).
By Proposition 2.1, G is a split commutative group coextension of S by the
corresponding abelian group valued functor G. By Lemma 4.2, there is a natural
isomorphism 8 of cn(T,G) = Hom-e(Tn, G) onto
cn(S,G) = C(Tn-1•G) = IltETn-1 G-rrn-1t;
when v : T n ----+ G, then 8v = u = (ut)tETn_ 1 E cn(S,G) is given by
v[t] = (ut,1rn_ 1 t).

When u = (ut)tETn_ 1 E cn(S,G), then e- 1u = v: Tn----+ G is given by
where xJ· = · 1 x J.+ 1 · · · x m and x = x 1 · · · x m' for all m > 0 and

x 1 · .. x J-
x 1 , ... ,xm E Tn_ 1 ; then 8~u = 8(8nv) E cn+ 1(S,G) is given for all t =
[x 1, ... ,xmJ E Tn by
(8nv) [t] = l:o~i~n (-l)i (vEn,i [t])

l:o~i~n-1 ( -l)i (v [En-1,it]) + (-l)n vt
l:o~i~n-1 ( -l)i (uf:n-1,i t• 7rn_ 1 En- 1,it)
~
+ (-l)n (2:1~j~m u;;-1Xj' 7rn-1x),
l:o~i~n-1 ( -l)i (uf:n-1,i t• 1rnt)
+ (-l)n (2:1~j~m u;;-1Xj' 7rn-1x),
since 1rn- 1 oEn- 1,i = 1rn' where x.J = x 1 · ··X.J- 1 x J.+1 · · ·xm and x = x 1 · · ·xm'
so that 7rn_ 1x = 1rnt. Thus
I ,;
(8nu)t = l:o~i~n-1 (-1)• UEn-1,it + (-l)n l:1~j~m u;;- 1xj
for all t = [x 1, ... , xmJ E Tn and we have proved:
Theorem 4.3. Let S be a commutative semigroup. Up to natural isomor-

phisms, the Beck cohomology of S has coefficients in an abelian group valued
functor G on H (S), and Hn (S, G) is the n-th homology group Ker 8n / Im 8n_ 1
of the complex
0 -+ C 1 (S,G) -+ ·· · ~ cn(S,G) ~ cn+ 1 (S,G) -+ · ··

where cn(S,G) = IltETn-1 G7rn-1t and
314 XII. COMMUTATIVE 8EMIGROUP COHOMOLOGY.
(8)
for all t = [x1, ... ,xml E Tn, where xj = X1"'Xj-1 xj+1'''Xm.
If m = 1, then xj is an empty product and xj = 1 E T~_ 1 .
4. Theorem 4.3 describes H 1 (S,G) and H 2 (S,G) as follows.
A 1-cochain u E C 1 (S,G) is a family u = (ua)aES such that ua EGa for

all a E S (u E f1sES Ga). When t = [a 1, ... ,am] E T1 , so that m > 0 and
a 1, ... ,am E S, and a= a 1 · .. am, then 1r 1t = 1r0 (a 1 .. ·am) =a by (1r) and
c0 •0 t = a 1 .. ·am= a= t by (c1 ). Thus (8) reads
1r 1
(81 u)t = ua - 2::1;Sj;Sm u~~
for all t = [a 1 , ... , am] E T 1 , where a = a 1 .. · am and

aj = a 1 .. · aj _ 1 aj+ 1 .. ·am. Hence u is a 1-cocycle if and only if
for all a 1 , ... , am E S; equivalently, if uab = u~ + ub for all a, b E S.
A 2-cochain u E C 2 ( s, G) = ITxETl G 7rl X is a family u = (ux) xETl such

that ux E G1r 1 x for all x E T 1 . When A= [x 1 , ... ,xm] E T 2 , so that m > 0
and x 1, ... ,xm E T 1 , and x = x 1 · .. cm, then 1r2 t = 7r 1 (x 1 .. ·xm) = 1r 1x,
c1•1t = x 1 ... xm = x E T 1 by (c'), and
E1,0t = [co,ox1, ... ,Eo,oxml

by ( E11 ). Thus ( 8) reads
(82u) [x1, ... ,xm] -- u[7rX1, ... ,7rXm l - u Xl· .. Xm + 61;Sj;Sm

~ u1rxj
Xj
for all x1, ... ,xm E T1, where 1f = 7r1 and xj = X1"'Xj-1 xj+1"'Xm. Hence
u is a 2-cocycle if and only if
for all m > 0 and x 1 , ... , xm E T 1 ; u is a 2-coboundary if and only if u = 8v

for some 1-cochain v :
4. COMMUTATIVE 8EMIGROUP COHOMOLOGY. 315
In the above, m is unbounded; computing H 2 (S, G) by Theorem 4.3 is

therefore an infinite task, even when S is finite. Fortunately, more effective
techniques are just around the comer.
5. Commutative semigroup cohomology inherits a number of properties from
triple cohomology in general. Theorems 1.1, 1.2, and 4.3 yield:
Theorem 4.4. When S is a free commutative semigroup, then Hn ( S, G) =0
for all n ~ 2.
Theorem 4.5. Every short exact sequence 9 : G ---+ G 1 ---+ G" of abelian
group valued functors on H(S) induces an exact sequence
· · · Hn(S,G) ---+ Hn(S,G 1) ---+ Hn(S,G") ---+ Hn+ 1 (S,G)
which is natural in 9.
Proposition 3.1 and Theorem 1.4 yield
Theorem 4.6. There is a one-to-one correspondence between elements of
H 2 ( S, G)
and equivalence classes of commutative group coextensions of S by
G, which is natural in G.
Theorem 4.6 provides a bijection between H 2 (S, G) and the abelian group
Ext ( S, G) from Chapter V. In the next section we shall prove that the two groups
are in fact isomorphic. Hence H 2 ( S, G) can be calculated from factor sets and
split factor sets, which one may assume are normalized.
6. We conclude this section by constructing a projective complex of which
Hn(s, G) is the cohomology; this takes place in the abelian category A of abelian
group valued functors on H ( S) .
LetT= (T,1r) be a c.s. overS. For every a E S let Ka = Ka(T) be the

free abelian group generated by the set
xa = Xa(T) = {(t,z) E T X S1 I (7rt)z =a}.
When v E S 1 , (t,z) E Xa implies (t,zv) E Xav and there is a unique homo-

morphism Ka ,v : Ka ---+ Kav such that .
Ka ,v(t,z) = (t, zv)

whenever (t,z) E Xa. We see that Ka ,1 is the identity on Ka and that Kav ,w o
Ka ,v = Ka ,vw. Thus OCT= (K, K) is an abelian group valued functor on H(S).
If f : T ---+ T 1 is a morphism in e, then (t, z) E X a(T) implies (ft, z) E

-1 - -1
X a(T) and f induces a homomorphism (IK.f)a : Ka (T) ---+ Ka (T). It is
immediate that OCJ : OCT ---+OCT' is a natural transformation. This constructs a

e
functor OC : ---+ A .
Lemma 4.7. For every object T = (T, 1r) of e and abelian group valued
functor G on H(S) there is an isomorphism HomA (OCT, G) ~ C(T, G) which
is natural in T and G. Hence OCT is projective.
Lemma 4.7 shows that the functor C(T,-) is representable. Hence the
isomorphism HomA (OCT, G) ~ C (T, G) determines OC uniquely up to natural
isomorphism.
Proof. Let 'P : K ---+ G be a natural transformation, where K = OCT. For

every t E T, (t, 1) E Xrrt and
if5t = 'P1rt(t, 1) E G1rt.

This constructs <p E C (T, G) = IltET G1rt . Since 'P is natural we have
'Pa(t,z) = 'Pa ~'Trt,z(t,1) = 'Y1rt,z 'P'Trt(t,1) = "11rt,z lf5t
for all (t, z) E X a. Thus t.p is uniquely determined by <p.
Conversely let u E C (T, G) . Define t.p a : K a ---+ Ga by:
'Pa(t,z) = 'Y'Trt ,z ut
for all (t, z) E X a. Then
"'a v 'Pa(t,z) = 'Ya v "'1rt z ut = "'1rt vz ut = 'Pav(t,vz) = 'Pav ~a v(t,z)

' ' ' ' '
for all (t, z) E X a. Hence 'P : K ---+ G is a natural transformation. We see that
<p = u. Thus HomA ( K, G) ~ C (T, G) . It is immediate that this isomorphism
is natural in T and G.
If CT : G ---+ G' is an epimorphism in A, then every CTa : Ga ---+ G~ is
surjective, C(T,CT) = IltET CT1rt : IltET G1rt ---+ IltET G~t is surjective, and
HomA (K,CT) is an epimorphism. Thus K is projective in A. D
Proposition 4.8. Hn (S, G) is the n-th cohomology group of the projective

complex
0 ~ C 1 (S) ~ ··· ~ Cn(S) ~ Cn+ 1 (S) ~ ···
where Cn(S) = OCTn-l and 8: Cn+l(S)---+ Cn(S) is given by
fJa(t,z) = l::o~i~n-1 ( -1)i (En-l,i t, z) + (-l)n l:l~j~m (xj, (7rn_lxj) z)

for all (t,z) E Xa, where t = [x1, ... ,xml E Tn and xj = X1···Xj-1 xj+1
···Xm.
Proof. By Lemma 4.7 there are natural isomorphisms
HomA (Cn(S),G) =HomA (OCT n+ 1,G) ~ e(Tn+ 1,G) = cn(S,G).

We show that these isomorphisms take the coboundary homomorphism
HomA (&,G) :HomA (Cn,G) -----+ HomA (Cn+ 1,G),
c.p f------+ c.p o 8, to the co boundary homomorphism 8n in Theorem 4.3.
Let u E en ( S, G) . To u corresponds the natural transformation c.p from

OCTn- 1 to G defined by
<fJa(t,z) = l1rt,z ut = ut
for all (t, z) E X a ( T n- 1) . Let v E en+ 1 ( S, G) correspond to c.p o 8. For every

t = [x 1, ... ,xmJ E Tn,
vt <fJ1rt( 01rt(t,l))
c.pTrt(L:o~i~n-1 (-l)i (t:n-1,i t, 1)
+ (-l)n L:1~j~m (xj, (7rn-1xj)l))
~
L...-O~i~n-1
( - l)i Ul':n-l,i t + (-l)n ~ . uTrn-lXj
L...-1~J~m Xj
(8nu)t·
Thus v = 8nu. 0
The complex in Proposition 4.8 is not very barlike, since the generators (t, v)
of Cn ( S) include sequences t E Tn_ 1 of unbounded length when n ~ 2. It is
not known in general whether there is a commutative "bar" complex in which t
is replaced by a sequence of length n. Results in the next section indicate how
the first groups of such a complex might be constructed.
7. Simpler chains can be used when coefficient functors are constant, or nearly
constant. We call an abelian group valued functor G = (G, 1) on S constant
when there is an abelian group A such that Ga =A and Ia ,t = lA for all a E S
and t E 8 1 . Then G and A may be identified, and we denote cn(s, G) by
en ( S, A) , and similarly for Bn , zn , and Hn . Constant functors are thin and
surjecting. Cohomology with constant coefficients is the commutative analogue
of the Eilenberg-MacLane cohomology for monoids.
When T = (T,1r) is a c.s. over S let K(T) be the free abelian group gen-
erated by the set T. For every abelian group G we have Hom (K(T), G) ~
ITtET G = C(T, G). Hence Hn(s, G) is, as in Proposition 4.8, the n-th coho-
mology group of the complex
C(S): 0 ~ C1 (S) ~ ··· ~ Cn(S) ~ Cn+ 1 (S) ~ ···
where Cn(S) = K (T n-l) and 8: Cn+I (S) ~ Cn(S) is given by
Ot = l:o;£i;£n-1 ( -l)i En-l,i t + (-l)n 2::1;£j;£m xj

for all t = [x 1, ... ,xml E Tn. Since C(S) is a complex of free abelian groups,
there is for every abelian group G a Universal Coefficient Theorem
Hn(S,G) ~ Ext(Hn_ 1 (S),G) EBHom(Hn(S),G)
where Hn(S) is the n-th homology group of C(S), with H 1(S) = C 1 (S)/Im8 1
and H0 (S) = 0 (MacLane [1963]). We leave the details to our tireless reader.
When S has a zero element, then Hn(S,G) = 0 for every abelian group
G, at least when n ~ 3. This can be remedied by using functors that are not
quite constant. When S has a zero element, an abelian group valued functor
G on S is almost constant when there exists an abelian group A such that
Ga =A for all a# 0, G0 = 0, and la,t = lA whenever at# 0. Then G may
be identified with the abelian group A, and we denote the cohomology groups
Hn (S, G) by H[) (S, A), and similarly for cochains, cocycles, and coboundaries.
Almost constant functors are thin and surjecting.
As noted in Grillet [1974], almost constant functors arise naturally in the
construction of homogeneous elementary semigroups. The Universal Coefficient
Theorem can be saved if in the above we replace K (T) by the almost constant
functor JK0 (T) in which the abelian group is the free abelian group generated by
{ t E T I 1rt # 0}. Again we leave the details to our reader.
5. SYMMETRIC COCHAINS.
Commutative cohomology cries out for an equivalent description in which

n-cochains are functions of n variables. This has been found only for n ~ 4.
1-cochains already are functions u = (ua)aES of one variable a E S.
1. In dimension 2, Theorem 4.6 provides a one-to-one correspondence be-
5. SYMMETRIC COCHAINS. 319
tween the elements of H 2 ( S, G) and the elements of the abelian group Ext ( S, G)
of all equivalence classes of commutative group coextensions of S by G, which
is also the abelian group of all equivalence classes of commutative factor sets on
S with values in G.
We now construct a more direct connection between factor sets and 2-cocycles,
which induces an isomorphism H 2 ( S, G) ~ Ext ( S, G) .
Let T 1 be the free commutative semigroup on the set S and 1r = 1r1 : T 1 ---t
S, so that 1r[a1 , a 2 , ... ,am]= a 1 a 2 .. ·am for all a 1, ... ,am E S. As we saw
in Section 4, a 2-cochain u E C 2 ( s, G) = nxETl G1rX is a family u = (ux) xETl
such that ux E G1rx for all x E T1 ; a 2-cocycle is a 2-cochain u such that
(Z)
for all m > 0 and x1, ... ,xm E T1, where xj = X1···Xj-1 xj+1···xm; a
2-coboundary is a 2-cochain u (necessarily a 2-cocycle) of the form u = 8v,
(B)
for some 1-cochain v ' where a. =~

a1 ... a.z- 1 a ~"+1 ... al .
A symmetric 2-cochain on S with values in G is a family s = (sa,b)a,bES

such that sa b E Gab and
'
for all a, b E S. Under pointwise addition, symmetric 2-cochains constitute an

abelian group SC 2 ( S, G) = lla,bES Gab. A symmetric 2-cocycle on S with
values in G is a symmetric 2-cochain s such that
(A)
for all a, b, c E S; equivalently, a commutative factor set on S with values in G
as defined in Section V.4. A symmetric 2-coboundary on S with values in G
is a symmetric 2-cochain (necessarily, a symmetric 2-cocycle) s ofthe form
8 a,b = vab + vba - vab
for some 1-cochain v; equivalently, a split factor set. Under pointwise addition,
symmetric 2-coboundaries and 2-cocycles form abelian groups SB 2 (S, G) ~
sz2 (S,G) ~ sc2 (S,G).
We saw in Section V.4 that
Ext(S,G) ~ SZ 2 (S,G) 1SB 2 (S,G).
2. Restriction to X s syields a trimming homomorphism 2: of C 2 ( S, G)

into SC 2 (S, G); when u is a 2-cochain, 2-:u is given by
(2-:u) a,b = u[a,b]
for all a,b E S; 2-:u is symmetric since [b,a] = [a,b] in T 1 .

Lemma 5.1. When u is a 2-cocycle, then u[a] = 0 for all a E S and
(Z')
for alll > 0 and a 1 , ... ,az E S, where a~= a 1 .. ·ai and a~'-t 1 = ai+ 2 ... az.
Proof. When m = 1 and x 1 = [a], then 1rx 1 = [a], (Z) reads: ux 1
ux 1 + uL , and ux 1 = 0.
Now let i ~ 1 and a 1 , ... ,ai+l E S. With m = 2, x 1 = [a 1 , ... ,ai], and

x 2 = [ai+ 1 ], then 1rx 1 = a~, 1rx 2 = ai+ 1 , ux 2 = 0, and (Z) reads
(Z")
(Z') is proved by induction on l, using (Z"). First, (Z') is trivial when l = 1

(then the right hand side is the empty sum 0) and when l = 2 (then a~ is the
empty product and a~ = 1 E S 1 ). If l = 3, then (Z") yields
u [at,az,a3] -- ua[at,az]
3 + '
u [az,a3]
and (Z') holds. If l > 3, then, with b~'-t 1 = ai+ 2 · .. az_ 1 , (Z") and the induction
hypothesis yield
Lemma 5.1 shows that 2:: Z 2 (S,G) ---+ SC 2 (S,G) is injective.
Lemma 5.2. Im 2: = SZ 2 (S,G).

Proof. Let u be a 2-cocycle. With m = 2, x 1 =[a], and x 2 = [b,c], (Z)
reads
u[a,b,c] = u[a, be] + u[b,c]

(since u[a] = 0). With m = 2, x 1 = [a,b], and x 2 = [c], (Z) reads
u[a,b,c] = u[ab,c] + u[a,b] ·
Therefore ~u E S Z 2 ( S, G) .
Conversely lets E SZ 2 (S,G). Since ~ is injective, there is at most one

u E Z 2 ( S, G) such that ~u = s, and it is given by (Z'). Accordingly, define
t = (tal, ... ,a)l>O, a1, ... ,alES by
II
ai+l
= L:l~i~l-1
- -
sa'i' a·z+l
for all l > 0 and a 1 , ... ,a1 E S, where a~= a 1 ···ai and a~'+. 1 = ai+ 2 ···a1 •
In particular, ta = 0 (if l = 1, then the right hand side is the empty sum 0) and
tab= sab' for all a,b E S. Also
' '
for all l > 0 and a 1 , ... , a 1, b E S; thus t satisfies (Z 11 ).
We show by induction on l > 0 that

t aal' ... ,aal = t a1, ... ,al (P)
for every a 1 , . .. , a 1 E S and permutation (}". This is trivial if l = 1 and follows

from tab= sa b if l = 2. When l > 2, (}"is a product of transpositions of the
' '
form (j j + 1) and it suffices to prove (P) when (}" = (j j + 1) . If j < l - 1,
then (P) follows from the induction hypothesis and (Z 11 ). If j = l - 1, then, with
af_ 2 = b, a 1_ 1 = c, a 1 = d, we have af_
1 = be, af'_
1 = d, and
II
sai+l d
t al, ... ,al = L:l~i~l-3 a'i' a·z+l + sb,c + sbc,d'
- -
II
sai+l c
t aa1, ... ,aal = L:l~i~l-3
- -
a'i' a z+l + sb,d + sbd,c'
and (P) holds, since s E S Z 2 ( S, G) yields

By (P), a 2-cochain u E C 2 ( S, G) is well defined by

u (al, ... ,azl -- t al, ... ,al
for alll > 0 and a1 , ... ,az E S. We show that u E Z 2 (S,G); that is,
(Z)
holds for all m > 0 and x1, 0 0 0 ,xm E T1' with xj = x1 .. ·Xj-1 xj+l .. ·Xm.
(Z) is trivial if m = 1; for m > 1 we proceed by induction on m. Assume
that (Z) holds form and let y 1, ... , Ym' z E T1 . Let x = y 1 · · ·Ym = [a 1, ... ,az],
z= [c1, ... ,cn], nyj =bj, nx=a=a1 ···an =b1 ···bm,and nz=c=c1 ···Cn.
By the induction hypothesis,
uX = u (bl, ... ,bzl + """"'

w1~j~m ubi
Yj
where b.=
J
b1 ···b.J- 1 b.+
J 1
···bm . Hence
+ """"'
L...1~k~n-1
(
-
Ck+l
8 a,c~ +
by the induction hypothesis and (A). Hence
c"
- """"'
w1::;k::;n-1
--
s akc'
'k
u (bl, ... ,bz,c] + """"' < ·< ubic

L...1=J=m Yj
+ uaz
by (Z"), and (Z) holds for m + 1. Thus u E Z 2 (S, G). Then Eu = s. 0

Lemma 5.3. E (B 2 (S,G)) = SB 2 (S,G).
5. SYMMETRIC CocHAINS. 323
Proof. When v E C 1 (S,<G), then (B) yields (8v)[a,b] = vab- v~- vb'. D
3. Lemmas 5.1, 5.2, and 5.3 provide an isomorphism Z 2 ( S, <G) 9,! Sz 2 ( S, <G)
which is natural in G, sends B 2 ( S, G) to SB 2 ( S, G), and induces an isomor-
phism Z 2 (S,G) I B 2 (S,G) 9,! SZ 2 (S,<G) I SB 2 (S,<G). This proves:
Theorem 5.4. For every commutative semigroup S there is an isomorphism
H 2 (S,G) 9,! Ext(S,<G) which is natural in <G.
The isomorphism H 2 ( S, <G) 9,! SZ 2 ( S, <G) I SB 2 (S, <G) can be refined, using
Proposition V.4.5, into an isomorphism
H 2 (S,<G) 9,! NSZ 2 (S,<G) 1NSB 2 (S,<G),
where N SZ 2 ( S, <G) and N SB 2 ( S, <G) are the groups of symmetric 2-cocycles
and 2-coboundaries s that are normalized ( s e ,a = 0 whenever e 2 = e ~J{ a).
Normalization can be confined to a single idempotent; when applied to the identity
element, it yields:
Corollary 5.5. When the commutative semigroup S has no identity element,
then H 2 (S, <G) 9,! H 2 (S 1 , <G'), where <G' extends <G to S 1 so that G~ = 0.
The study of H 2 (S,<G) may therefore be limited to monoids. Theorems 5.6

and 5.7 below have similar corollaries for dimensions 3 and 4. It is probable that
Corollary 5.5 extends to all dimensions.
When S is finite, the computation of H 2 (S, <G) using Theorem 5.4 is a
finite task, since a symmetric 2-cocycle consists of finitely many group elements
subject to finitely many conditions, and a symmetric 2-coboundary is determined
by finitely many group elements. Thus Theorem 5.4 is a marked improvement
upon Theorem 4.3. But further improvement would not hurt.
For example, let S be the Volkov nilsemigroup
S = ( a,b I a 3 = a2b = ab 2 = b4 = 0, a 2 = ab = b3 ).
(Example 111.3.6); the elements of S are a, b, c = b2 , d = a 2 = ab = b3 , and

0. A normalized symmetric 2-cochain s on S consists of sa a, sa b, sa c, sa d,
' ' ' '
sao' sbb' sbc' sbd' sb0' sec' sed' sc0' sdd' sd0' and sao= 0 (since s
' ' ' ' ' ' ' ' ' ' '
is normalized).
Then s is a symmetric 2-cocycle if and only if
s~,y + sxy,z = sx,yz + s~,z A(x,y,z)
holds in Gxyz for all x,y,z E S. We note that A(x,y,z) is trivial if x = z and
remains unchanged if x and z are interchanged; moreover, A(x, y, z) follows

from
8 ~,x + 8 xy,z 8 y,xz + s~.z A(y,x,z)

and
s~,z + 8 yz,x = 8 y,zx + s~.x A(y,z,x).

Hence it suffices to state A(x,y,z) when x ~ y and x < z in the lexicographic
order a < b < c < d < 0 on S. If x = a, there are 5 choices for y and 4
choices for z; if x = b, there are 4 choices for y and 3 choices for z; if x = c,
there are 3 choices for y and 2 choices for z; if x = d, there are 2 choices for
y and 1 choice for z; this yields 20 + 12 + 6 + 2 = 40 conditions.
We leave the actual conditions and their solutions to northern readers, who
are blessed with long winter evenings. H 2 ( S, G) will be computed in the next
chapter when G is thin, using more efficient methods.
4. A symmetric 3-cochain on S with values in G is a family
t = (ta,b,c)a,b,cES such that ta,b,c E Gabc and
tc b a= -ta b c and tab c + tb c a + tc a b = 0

'' '' '' '' ' '
for all a, b, c E S. These conditions are satisfied by the coboundary
of every symmetric 2-cochain s E SC 2 ( S, G). Under pointwise addition, sym-

metric 3-cochains constitute an abelian group SC 3 ( s, G) ~ ITa,b,cES G abc.
A symmetric 3-cocycle on S with values in G is a symmetric 3-cochain t
such that
td
a,b,c - ta,b,cd + ta,bc,d - tab,c,d +ta
b,c,d =0
for all a, b, c, d E S. This condition is satisfied by the coboundary t = 8s of

every symmetric 2-cochain s; such a coboundary is a symmetric 3-coboundary
on S with values in G. Under pointwise addition, symmetric 3-coboundaries and
3-cocycles form abelian groups S B 3 ( S, G) ~ S z 3 ( S, G) ~ SC3 ( S, G) . The
following result is due to Grillet [ 1991 C], [ 1997C]:
H 3 (S,G) ~ SZ 3 (S,G) I SB 3 (S,G) which is natural in G.
A symmetric 4-cochain on S with values in G is a family
u -- ( t a,b,c,d )a,b,c,dE 8 such that ua ,b,c ,d E Gabcd and
ua ,b,b,a = 0, ud ,c, b,a = -u a, b,c, d '
c, ,a, b-u d,a,b,c

u a,b,c,d - ub,c,d,a +ud =0 '
u a,b,c,d - u b,a,c,d +u b,c,a,d - u b,c,d,a -0
-
for all a, b, c, d E S. These conditions are satisfied by the co boundary
(8t) a,b,c,d = tda,b,c - t a,b,cd + t a,bc,d - t ab,c,d + tab,c,d
of every symmetric 3-cochain s E SC 3 (S, G) (this is shown in the next sec-

tion). Under pointwise addition, symmetric 4-cochains constitute an abelian group
SC4 (S,G) ~ ITa,b,c,dES Gabcd·
A symmetric 4-cocycle on S with values in G is a symmetric 4-cochain u

such that
uea,b,c,d - u a,b,c,de + u a, b,cd,e - u a, bc, d ,e + u a b,c, d ,e - uab,c, d ,e -- 0
for all a, b, c, d, e E S. A symmetric 4-coboundary on S with values in G is

the coboundary u = 8t of a symmetric 3-cochain t. Under pointwise addition,
symmetric 4-coboundaries and 4-cocycles form abelian groups SB 4 (S, G) C
SZ 4 (S,G) ~ SC4 (S,G). The following result is due to Grillet [2001H]:
H 4 (S,G) ~ SZ 4 (S,G) I SB 4 (S,G) which is natural in G.
The author's proofs of Theorem 5.6 and 5.7 are computational like the proof
of Theorem 5.4, but much longer, and very likely to tax the reader's patience (even
during long winter evenings). Better proofs would rely on Theorem 1.3, or on the
underlying spectral sequence argument, to show that symmetric cochains define
the same cohomology. This requires a general definition of symmetric cochains
and a proof that S zn (S, G) = S Bn (S, G) when S is free. The author has proofs
of this last fact when n ;£ 3, which are, unfortunately, entirely comparable in
length and spirit to the proofs of Theorems 5.4 and 5.6.
Theorems 5.6 and 5.7 strongly suggest that symmetric cochains, cocycles,
and coboundaries can be defined in every dimension n so that Hn(S, G) ~
szn(s, G) I SBn(s, G). How to do this is still an open problem. In the
above, the coboundary homomorphisms for symmetric cochains are essentially
the same as in Leech cohomology and can be defined in all dimensions. The
symmetry conditions in dimension n + 1 can be defined by induction as all the
symmetry conditions inherited by coboundaries of symmetric n-cochains. The
difficulty lies in proving that Hn(s, G) ~ szn(s, G) I sBn(s, G) (or that

Szn (S, G) = SBn (S, G) when S is free). There the matter rests, for now.
Chapter XIII.
THE OVERPATH METHOD.
We saw in Chapter XII that the cohomology group H 2 ( S, G) can be computed

by means of functions of two variables in S, namely, symmetric 2-cocycles. The
overpath method computes H 2 ( S, G) by means of functions of one variable, one
for every defining relation of S in any suitable presentation of S. This makes
the computation of H 2 ( S, G) a finite task whenever S is finitely generated.
Applications compute H 2 (S, G) when S is cyclic or, more generally, has
only one defining relation, and when S is partially free. This last application
depends rather heavily on the construction of group-free congruences in Chapter
X. We also show that strand bases give rise to minimal cocycles. The main results
are from Grillet [ 1992], [ 1995F], [ 1996C], [ 1995P], [2000Z].
1. OVERPATHS.
The overpath method depends on certain properties of free commutative semi-

groups and congruences on these semigroups.
1. In what follows F = Fx is the free commutative monoid on a set X. We
return to the additive notation for F and write the elements of F as finite linear
combinations a = I::xEX ax x of elements of X, with the usual order:
a~ b if and only if ax~ bx for all x EX;
the length of a = I::xEX ax x is Ia I = I::xEX ax.
Proposition 1.1. On every free commutative monoid F there exists an order

relation [;;; such that:
(1) (F, [;;;) is well-ordered;
(2) if a ~ b in F, then a [;;; b;
(3) if a[;;; bin F, then a+c C b+cforall c E F.
327
328 XIII. THE OVERPATH METHOD.
Proof. Well-order X, then define [;;;; as follows: a C b if and only if either

lal < lbl, or Ia I = lbl, a -1- b and the least x E X such that ax -j. bx satisfies
ax > bx. (This is the degree lexicographic order on F.)
An element ofF of length l is the sum x 1 + ... + xl of l elements x 1 [;;;;
x 2 [;;;; ... [;;;; xl of X. Each set Fl = {a E F I lal = l} is a subset of the
lexicographic product xl = X X 0 0 0 X X and is well-ordered by [;;;; Then F is 0
the ordinal sum of F0 , F 1 , ... , Fz, ... and is well-ordered by !;;;; •

If a ~ b in F, then either a = < lbl; in either case a [;;;; b.
b or lal
Finally let a C b and e E F; then either Ia I < lbl, or Ia I = lbl and the
least x such that ax -j. bx satisfies ax > bx. In the first case, Ia + el < lb + el.
In the second case, the least x such that ax -j. bx is also the least x such
that ax + ex -j. bx + ex, and satisfies ax + ex > bx + ex. In either case
a+e [;;;; b+e. 0
We call an order relation [;;;; on F a compatible well order when it has
properties (1), (2), and (3) in Proposition 1.1. Explicit compatible well orders can
be constructed in various ways, besides the degree lexicographic order, particularly
if X = { x 1 , x 2 , ... , x n } is finite (see e.g. Adams & Loustaunau [ 1994]); if for
instance p 1 , ... , Pn are the first n prime numbers, then
a1 an < b1 bn
alxl +···+anxn [;;;; b1X1 +···+ bnxn {=::} P1 ···Pn =Pl ···Pn
is a compatible well order on F, the prime order on F of Rosales [ 1995].
A lexicographic order [;;;; on F is defined from a well order ~ on X by:
a C b if and only if a -j. b and the least x EX such that ax -1- bx has ax < bx.
Then a < b implies a c b. Also, x -< y in X implies x :::J y in F. (The usual
definition requires ax > bx, but then a< b does not imply a :::J b.)
Proposition 1.2. If F is finitely generated, then every lexicographic order
on F is a compatible well order.
Proof. If a C b and e E F, then, in the above, the least x such that ax -j. bx
is also the least x such that ax +ex -j. bx +ex, so that ax +ex > bx +ex and
a+ e C b +c. If X is finite, then (F, !;;;;) is a finite lexicographic product of
copies of N+ and is well ordered. D
When F is finitely generated, every total order !;;;; on F with property (2)
is a well order: if indeed there is an infinite sequence a 1 :::J a 2 :::J • • · :::J an :::J
an+l :::J ···,then An= {t E F I t;;;;) an} is an ideal ofF for every n,
by (2), and A 1 ~ A 2 ~ · · · ~ An ~ An+l ~ · · ·, a flagrant contradiction of
Corollary VI.l.3; therefore the totally ordered set (F, [;;;;) satisfies the descending
1. 0VERPATHS. 329
chain condition and is well ordered. Compatible well orders are then also known
as a linear admissible orders.
If on the other hand X is infinite, then the first generators x 1 -< x 2 -< · · · -<
xn -< · · · of X yield a nonempty subset x 1 =:J x 2 =:J • • • =:J xn =:J • • • of F with no
least element and F is not well ordered by its lexicographic orders, even though
they satisfy (2) and (3).
2. In what follows, !;;; is any compatible well order on F.
Let e be a congruence on F. Under !;;; the e-class C a of a E F has a least
element q( a) (the function minimum of Rosales [ 1995]). By definition,
a e q(a); ce a implies c;;;;) q(a); and a e b ¢:::::} q(a) = q(b).
Then F is the disjoint union F = P U Q, where Q = Q(e) = { q(a) I a E F} =
{ q E F I a e q ====? a ;;;;) q} is the set of all least elements of all e-classes, and
P = P(e) = F\Q = {a E F I a =:1 q(a) }.

If e is the equality, then P = 0.
Lemma 1.3. P is an ideal ofF.
Proof. When a E P and c E F, then a =:J q(a), a+ c e q(a) + c,
a+c =:J q(a) +c ;;;;) q(a+c), and a+c E P. 0
In what follows, M = M (e) is the set of all minimal elements of P, under
the usual order ~; since (F, ~) satisfies the descending chain condition, P is
generated as an ideal of F by M.
Proposition 1.4. The congruence e is generated by all pairs (m, q(m))
with mE M(e).
Proof. Let JY( be the congruence on F generated by all pairs (m, q( m)) with
mE M. Then JY( ~ e, since me q(m) for all mE M. We show by artinian
induction that a JY( q(a) for every a E F (this also follows from Proposition 1.5
below); then a e b implies aM q(a) = q(b) M b, so that e = M.
We have a JY( q(a) for all a E Q (then a = q(a)) and for all a E M (by
definition of M). Let a E P. Then a ~ m for some m E M and a= m + t
for some t E F. Let b = q(m) + t. Then aM b, since m M q(m), a e b
since me q(m), and a =:J b, since m =:J q(m). Then b M q(b) by the induction
hypothesis, and a JY( b JY( q(b) = q(a). 0
Proposition 1.4 implies Redei's Theorem. If indeed F is finitely generated,
then M is finite by Dickson's Theorem (Corollary VI.l.3), and Proposition 1.4
shows that e is finitely generated (Grillet [ 1993R]).
330 XIII. THE 0VERPATH METHOD.
Conversely Rosales [ 1995] devised an algorithm which constructs q from any

finite set of generators of e; this provides an explicit algorithm for the solution
of the word problem in any finite commutative presentation.
3. Given the congruence e, we now regard the free c.m. F as a directed
graph with labeled edges, in which the vertices are the elements of F and an
edge a ~ b from a to b, labeled by m, is an ordered pair (a, m) such that
mEM(e), m~a,and a-b=m-q(m). Then a=m+t and b=q(m)+t,
where t =a-m= b- q(m) E F; hence q(m) ~ b, a e b (since me q(m) ),
and a =:J b (since m =:J q(m) ).
A descending path from a E F to b E F is a sequence a = p0 , ... ,pk = b

of elements of F and edges
where k ~ 0. (We index sequences of elements of F by superscripts, to keep

subscripts for coordinates.) Equivalently, a path from a to b consists of a se-
quence a = p0 , ... ,pk = b of elements of F and a sequence m 1, ... , mk of
elements of M(e) such that mi ~ pi- 1 and pi- 1 -pi = mi- q(mi) for all
1<= z. <
= k . Th en q (m i) <
= pi , po, ... , p k E Ca an d po =:J ... =:J p k ; m . lar,
. parttcu
a e b and a~ b (with a= b if k = 0). Also a- b = 2:: 1::;i::;k (pi- 1 - pi) =

"El~i~k (mi- q(mi)).
An overpath from a E F to b E F is the sequence p: ml, ... , mk E M(e)

of labels in a path
mk k
----t p = b,
from a to b. A path from a to b is determined by p0 = a and its overpath, since
in the above the relation pi- 1 - pi = mi- q(mi) determines pi from pi- 1 and
mi. In particular,
a-b = "2:1~i~k(mi-q(mi)).
The empty sequence is an overpath from any a E F to itself. If
0 ml 1 m2
a = p ----t p ----t
is a path from a to b, and

1. OvERPATHS. 331
is a path from b to c, then
a=po ~ ~pk=b=qo ~--· ~qk=c

is a path from a to c. Hence if p : m 1 , ... , mk is an overpath from a to b, and
q : n 1 , ... , n l ·IS an overpathfrom b to c, t hen p + q : m 1 , ... , m k , n 1 , ... , n l ·IS
an overpath from a to c.
Let c E F. If (a, m) is an edge from a to b, then (a+ c, m) is an edge
from a + c to b + c. Hence if
is a path from a to b, then there is a path with the same labels from a + c to
b +c. Thus if m 1 , ... , mk is an overpath from a to b, then m 1 , ... , mk is an
overpath from a + c to b + c. ·
The following result is a well-ordered version of Proposition 1.2.9, and shows
how e is generated by all pairs (m, q( m)) with m E M.
Proposition 1.5. For every a E F, there exist a path from a to q(a) and
an overpath from a to q( a).
Proof. This is proved by artinian induction on a. If a = q( a) E Q, then
there is an empty path from a to q( a) . Now let a E P. Then a ~ m for some
m E M. Let b = q( m) + t, where a = m + t. Then (a, m) is an edge from a
to b. Hence a e b, a :::J b, and the induction hypothesis yields a path from b to
q(b). Adding a ~ b yields a path from a to q(a) = q(b). 0
4. The process of well ordering F to select "minimal" generators of e (as
in Proposition 1.4) is reminiscent of Gr5bner bases. Indeed let K be a field
and K[X] be the polynomial ring with the set X of commuting indeterminates.
Ordering F also orders the monomials Xa = flxEX xax E K[X] (where
a = L::xEX ax x E F).
Proposition 1.6. Let e be a congruence on F and I(e) be the ideal of

K[X] generated by all Xa- Xb with a e b. The set
G(M) = {Xm- xq(m) I mE M}

is a Grabner basis of I(e).
Proof. First we note that I = I(e) is generated by all Xa - Xq(a), since

a e b implies q(a) = q(b) and Xa- Xb = (Xa- Xq(a)) - (Xb- Xq(b)).
We show that the ideal L(I) generated by the leading tenns of polynomials in
I coincides with the ideal L( G) generated by the leading tenns of polynomials in
G(M); this is one of the criteria for Grobner bases (see e.g. Adams & Loustaunau
[ 1994], Theorem 1.6.2).
When a E P, then a :::J q(a) and the leading tenn of Xa- Xq(a) is Xa.
Since P is an ideal of F by Lemma 1.3, L(I) is generated by all Xa with
a E P. Now a E P implies, as above, a ~ m for some m E M, a = m + t for
a m t
some t E F, and X = X X E L( G) . Therefore L( I) ~ L( G); conversely
L(G) ~ L(I) since G(M) ~ I. Thus L(G) = L(I). 0
We give a direct proof that I(e) is generated by G(M). Let J be the ideal
of K[X] generated by G(M). We show by induction on a that Xa- Xq(a) E J
forallaEF. WhenaEQ,thena=q(a) andXa-Xq(a) EJ. LetaEP. As
in the proof of Proposition 1.4, a ~ m for some m E M and a = m + t for some
t E F. Let b = q( m) + t. Then a :::J b since m :::J q( m). Xb - Xq(b) E J by the
induction hypothesis, Xa - Xb = Xt ( Xm - Xq( m)) E J, and Xa - Xq( a) =
(Xa- Xb)- (Xb- Xq(b)) E J. Thus Xa- Xq(a) E J for all a E F;
therefore I = J.
2. MAIN RESULT.
The main result in this chapter is the computation of H 2 ( S, G) by the overpath

method. As a first application we find H 2 ( S, G) when S has a presentation
with only one defining relation; for instance, when S is cyclic. We also relate
H 2 ( S, G) to the strand bases in Chapter XI.
1. When S is a commutative semigroup which does not have an identity
element, we saw that H 2 (S,G) ~ H 2 (S 1 ,G'), where G' extends G to H(S 1 )
so that G~ = 0 (Corollary XII.4.5). Hence we may as well start with a monoid S.
In what follows S is a commutative monoid and G = (G,"f) is an abelian
group valued functor on H(S); 1r : F ---+ S is a surjective homomorphism,
e
where F is the free c.m. on some set X, and = ker 1r; G is any compatible
e
well order on F; M and q are as in Section 1. By Proposition 1.4, is generated
by all (m, q( m)) with m E M (e); this provides a presentation of S as the c.m.
generated by X subject to all relations m = q( m) with m E M.
2. MAIN RESULT. 333
A minimal cochain on S with values in G (short for minimal 1-cochain) is

a family u = (um)mEM such that um E G1rm for all mE M.
Let u be a minimal cochain. Let a E F, p : m 1 , ... , mk be an overpath

from a to b, and
. wh"1ch p 0 , ... , p k E
. path c.1rom a to b , m
be t he correspon d mg ca, so that
1rpi = 1ra for all i. Define
where ti =pi- q(mi) = p i - l - mi. A minimal cocycle on S with values in G

is a minimal cochain u such that ua;p = ua;q whenever p and q are overpaths
from a to q(a) (so that ua;p does not depend on p).
Let g = (gx)xEX E ITxEX G1rx be a family such that 9x E G1rx for every
generator x E X ofF. A minimal cochain 8g is defined by
(8g )m -_ "' 1r(m-x) _ "'
wxEX, x;;;m mx 9x
( ) 1r(q(m)-x)
wxEX, x;;;q(m) q m X 9x
for every m = l:xEX mx x E M. A minimal cochain constructed in this fashion

is a minimal coboundary. Under pointwise addition, minimal coboundaries,
minimal cocycles, and minimal cochains constitute abelian groups
MB 1 (S,G) ~ MZ 1 (S,G) ~ MC 1 (S,G) = ITmEM G1rm ·

The main result in this chapter is:
Theorem 2.1. For every commutative monoid S there is an isomorphism
which is natural in G.
2. The proof of Theorem 2.1 occupies the next section. First we consider
an example: when S has a commutative presentation (as a semigroup or as a
monoid)
with a single defining relation, in which we assume, not unreasonably, that ri +

si > 0 for all i ~ n and that ri =f si for some i. Other examples are given in
Grillet [2000T] and in Sections 4 and 5.
We can set up the surjective homomorphism 1r : F = Fx ---+ S so that X
contains distinct elements x 1 , ... , xn such that 1rx 1 = a 1 , ... , 1rxn = an. Then
e = ker 1r is the congruence on F generated by the single pair (r, s), where
r = 2:1:-::;i:<S;n ri xi and 8 = 2:1:-::;i:<S;n 8i xi ·
The congruence e is readily described:

Lemma 2.2. a e b if and only if there
exists a sequence p0 , ... , pk of
elements ofF such that k ~ 0, a = p0 , pk = b, and either
pi- 1 - r =pi-s ~ 0 for all i ~ 1 (A)
or
pi- 1 - s =pi - r ~ 0 for all i ~ 1 (B)
Proof. By Proposition 1.2.9, a eb if and only if there exists a sequence
p0 , ... , pk of elements of F such that k ~ 0, a = p0 , pk = b, and, for every
i ~ 1, either
(a)
or
(b)
If (a) holds for i < k and (b) holds for i + 1, then pi- 1 - r =pi-s= pi+ 1 - r
and pi, pi+ 1 may be deleted from the sequence. Similarly if (b) holds for i < k
and (a) holds fori+ 1, then pi- 1 - s =pi- r = pi+ 1 - s and again pi, pi+ 1
may be deleted from the sequence. After all such deletions, either (a) holds for
all i, or (b) holds for all i . 0
Now let ~ be any compatible well order on F. We may assume that r :::J s.
When a e b and (A) holds, then in Lemma 2.2 pi- 1 - r =pi-s ~ 0 implies
pi- 1 = r+t and pi= s+t, where t = pi- 1 -r =pi-s~ 0, so that pi- 1 :::Jpi
for all i ~ 1 and a :::J b; if (B) holds, then similarly a c:: b.
If therefore a E P,
so that a e q(a) and a :::J q(a), then (A) holds, a -r = p - r ~ 0, and a~ r.
0
On the other hand, r e s and r :::J s, so that r E P. This proves:
Lemma 2.3. M(e) has just one element, namely r; and q(r) = s.
Let
d = a r1 r2
1 a2 ··· anTn = a 81 82
1 a2 ··· an8n = 1rr = 1r s E S.
By Lemma 2.3, a minimal cochain consists of u E Gd, and MC 1 (S,G) = Gd.
Moreover there is only one overpath from any c E F to q(c), which is a sequence
of r 's. Hence every minimal cochain is a minimal cocycle and MZ 1 (S, G) = Gd.
2. MAIN RESULT. 335
A minimal cochain u E G d is a minimal coboundary if and only if there

exists a family g = (gx)xEX such that 9x E G1rx for every x E X and
u "' r 9 1r(r-x)
wxEX, x~r X X
Let 'Yi : Ga.t --+ G d be defined by:
where
d'·t a r11 ···
ri-1 ri-1 ri+1
ai-l ai r
ai+l .. · ann (when ri > 0), and
d2~l 81 Bi-1 Si-1 Bi+1 Bn ( h
a1 .. · ai-l ai ai+l .. · an w en si > 0).
Then u is a minimal coboundary if and only if there exist gi = 9x. E Ga. such
t t
that u = 2: 1 ~i~n 'Yi9i· Hence MB 1 (S,G) = L: 1 ~i~nlm'Yi and:

Proposition 2.4. When S has a commutative semigroup or monoid pre-
sentation with a single nontrivial defining relation, then, with the notation as
above,
H 2 (S,G) ~ Gd/(L:l~i~n Im1J
Corollary 2.5. When S = ( a I ar = ar+p ) is cyclic with index r and

period p, then H 2 (S,G) ~Gar/ plm 'Ya ar-1·
'
Proposition 2.4 becomes simpler when G is thin. Then d ~!J-C ai for all i
(since ri + si > 0) and 'Yai,d~t = 1:i when ri > 0, 'Ya.,d~'
• t
= 1:i when si > 0,
and 'Yi = (ri - si) 1:i for all i. Hence

Corollary 2.6. When S has a commutative semigroup or monoid presen-
tation with a single nontrivial defining relation and G is thin, then, with the
notation as above,
H 2 (S,G) 9E Gd/(2: 1 ~i~n (ri- si) Im1:i).
If for instance S = ( a I ar = ar+p ) is cyclic with index r and period p
and G =A is constant (G 8 =A and 'Ys.t = lA for all s E Sandt E 8 1 ), then
336 XIII. THE OvERPATH METHOD.
H 2 (S, G) ~ AjpA; thus H 2 (S, G) ~ Ext (H, A), where H is the subgroup
{ ak Ik ~ r} ~ ZP of S.
3. Finally we show that strand bases in Chapter XI give rise to minimal
cocycles. This result is from Grillet [ 1996C], [200 1C].
In what follows, e is a subcomplete congruence on a free commutative
monoid F and e* is its group-free hull; 1r : F -----t S and 7r* : F -----t S*
are surjective homomorphisms which induce e and e* respectively. If e is
complete, then S* ~ Sj'){ and one expects the cohomology of S* to show
up somewhere in the construction of S and e. Minimal cocycles provide this
connection.
The direction set, extent cells, strand groups, strand bases, and notation are
as in Chapter XI. Also ~ is a compatible well order on F; the mapping q and
sets M and Q are those of e* , not of e.
Lemma 2.7. Let s be a strand base of e. For every m E M(e*) let
sm = sm- m- sq(m) + q(m) E Gm.
If m 1 , m 2 , ... , mk is an overpath from a to b, then

sa - sb -a+ b- Sml - · · · - Smk E Ra = f?v.
Proof. Let
be a path from a to b, so that pi- 1 = mi + ti, pi = q(mi) + ti for some ti E F

and pi- 1 - pi= mi- q(mi), for every 1 ~ i ~ k. Then Ra = Rpo = RP1 =
··· = RPk = Rv by (R2), since a = p0 , p 1 , ... , pk = b are all in the same
e* -class, and
by (S+) in Lemma XI.6.l. Adding these equalities yields sa- sb- a+ b- sm1-
···-skER
m a· D
With b = q( a), Lemma 2. 7 implies that a strand base of e is completely

determined modulo strand groups by its values on M U Q.
2. The strand group functor lK = (K, 'lj;) of e is the thin abelian group
valued functor on F je* defined as follows (Section XI.4). To every e* -class
3. PROOF OF MAIN RESULT. 337
C*, K assigns the group Ka = Gal Ra, which does not depend on the choice
of a E C* . When C* ~:H D* in FIe*, then a :_;;: b for some a E C* , b E D*,
Ga s:;; Gb, Ra s:;; Rv by (R3), and 7/J/; : Ka --+ Kb sends g + Ra to g + Rv
and does not depend on the choice of a E C* and b E D* (as long as a :_;;: b).
Since S* ~ FIe* we may regard lK as a thin abelian group valued functor on
S* ; then lK is isomorphic to the extended Schlltzenberger functor of S, which
is the usual Schlltzenberger functor if S is complete (Proposition XI.4.8).
Proposition 2.8. Let e be a subcomplete congruence on F and s be a
strand base of e. For every m E M(e*) let
s:n = sm+J\n = sm-m-sq(m)+q(m)+f\n E GmiJ\n.
Then s* is a minimal 1-cocycle on FIe* with values in the strand group functor
lK of e. Moreover, two strand bases s and t define the same congruence if and
only if s* = t*.
. . I coc h.
Proof. s *.IS a m1mma am. Wh en p : m 1 , ... , m k.IS an overpathfirom
a to b and
0 ml 1 m2
a = p ----+ p ----+
is the corresponding path, then
by Lemma 2.7. Hence s~·p·b is independent of path and s* is a minimal cocycle.

' '
By Proposition XI.5.2, two strand bases s and t define the same congruence
if and only if a e* b implies sa- sb- ta + tb ERa(= Ilv). Since me* q(m)
this implies
(sm- m- sq(m) + q(m))- (tm- m- tq(m) + q(m)) E J\n
for all mE M and s* = t*. Conversely assume that s* = t*. Then s~·p·b = t~·p·b
' ' ' '
and
sa - sb - a + b + Ra = ta - tb - a + b + Ra
whenever p is an overpath from a to b. Hence sa- sq(a) - ta + tq(a) E Ra
for all a. If a e* b, then q(a) = q(b), sb- sq(a)- tb + tq(a) E Rb = Ra, and
sa - sb - ta + tb E Ra . Thus s and t are equivalent. D
Proposition 2.8 embeds the set of equivalence classes of strand bases (and the
set of all subcomplete congruence with the given strand groups) into the abelian
group MZ 1 (Fie* ,JK).
3. PROOF OF MAIN RESULT.
I . In what follows, S is a commutative monoid, <G = (G, 1') is an abelian

group valued functor on H(S), F = Fx is the free c.s. on a set X, and
1!' : F ---+ S is a surjective homomorphism; we prove Theorem 2.1.
We begin the proof by lifting 1- and 2-cochains from S to F.
The homomorphism 11' : F ---+ S extends to a functor 1!' : H (F) ---+ H (S) .
Hence <G = (G, 1') lifts to an abelian group valued functor <G11' = (<G1!', ')'11') = <Go 1!'
on H(F); <G1!' assigns G1ra to a E F and 1'1ra1rt to (a,t): a---+ at. Thus
'
l = g1rt, where g E G1ra, l
is provided by <G11', and g1rt is provided by <G.
Note that <G11' is thin, since F is cancellative.
Every 1-cochain u = (ua)aES E C 1 (S,<G) lifts to a 1-cochain 1!'*u = uo1!' E
C 1 ( F, <G1!') defined by
(11'*u)a = u'Tra E G7ra
for all a E F. If u is a 1-cocycle, so that uab = u~ + ub' for all a, b E S, then

( 11' * u ) ab 1rb + u1rb
= u(1ra)(1rb) = u1ra 1ra = (11' * u )ba+ (11' * u )ab
and 11'* u is a 1-cocycle; thus 11'* Z 1 ( S, <G) ~ Z 1 ( F, !G-11') •

Similarly every symmetric 2-cochain u = (ua) aES E SC 2 ( S, <G) lifts to a
symmetric 2-cochain 11'* u = u o 1!' E C 2 ( F, <G1!') defined by
(11'*u)a,b = u1ra,1rb E G1r(ab)
for all a, b E F. If u is a symmetric 2-cocycle, so that u~ b + uab c = ua be + ub' c

' ' ' '
for all a, b, c E S, then
(11'*u)~,b + (11'*u)ab,c = u;~,1rb + u(1ra)(1rb),7rc

= u1ra, (1rb)(1rc) + u;b,1rc = (11'*u)a,bc + (11'*u)b,c
for all a, b, c E F, and 11'*u is a symmetric 2-cocycle. If u = 8v is a symmetric
2-coboundary, so that ua b = v~ - vab + vb' for all a, b E S, then
'
(11' *u )a,b = v1ra
1rb - v(1ra)(1rb) + v1rb 5: * )
1ra = (u1!' v a,b
for all a, b E S, and 1!'*u is a symmetric 2-coboundary. Thus
11'* SZ 2 (S,<G) ~ SZ 2 (F,<G11') and 11'* SB 2 (S,G) ~ SB 2 (F,<G1!').

Since H 2 ( F, !G1r) = 0 (Theorem XII.3 .4), symmetric 2-cocycles on S lift to

symmetric 2-coboundaries on F and can therefore be constructed by projecting
the coboundaries of 1-cochains on F. This marks the birthplace of Theorem 2.1.
2. A 1-cochain u E C 1 ( F, !G1r) is consistent (relative to 1r) when
1ra = 1rb implies ua+c - ub+c = u~ - ub for all c E F.
Under pointwise addition consistent 1-cochains form a subgroup K 1 ( F, !G1r) of
C 1 ( F, !G1r). We shall see that consistent 1-cochains on F are precisely those
whose coboundaries project to symmetric 2-cocycles on S. First we show:
Lemma 3.1. K 1 (F,!G1r) contains Z 1 (F,!G1r) and 1r*C1 (S,!G).
Proof. If u E Z 1 (F,IG1r), then 1ra = 1rb implies l1ra 1rc = l1rb 1rc and
c a b c
' '
ua - ua+c = uc = uc = ub - ub+c ·
If u = 1r*v, then 1ra = 1rb implies ua = v1ra = v1rb = ub, l1ra,1rc = l1rb,1rc'
1r(a+c) = 1r(b+c), uac = ubc' and u~ -ua+c = ub-ub+c· D
When u E K 1 (F,IG1r), then 1ra = 1rb, 1rc = 1rd imply l1rc 1ra = l1rc 1rb'
' '
Hence a homomorphism .6. : K 1 (F,!G1r) ----+ SC 2 (S,!G) is well defined by
(.6.u)1ra 7rc = u~- ua+c + u~ = (8u)a c

' '
for all a, c E F. It is immediate that .6. is natural in !G.
Lemma 3.2. Im .6. = SZ 2 ( S, !G) .
Proof. .6.u is a symmetric 2-cocycle, since 8u is a cocycle and
(.6.u);~,1rb + (.6.u)c1ra)(1rb), 1rc = (8u)~,b + (8u)a+b, c

= (8u)a, b+c + (8u)b,c = (.6.u)1ra, (1rb)(1rc) + (.6.u);b,7rc"
Conversely lets E SZ 2 (S,!G). Then 1r*s E SZ 2 (F,!G1r). Since H 2 (F,!G1r)
= 0 we have 1r* s = 8u for some u E C 1 (F, !G1r), so that
8 1ra,1rc c
= ua- ua+c + uca
for all a,c E F. If 1ra = 1rb, then
'"'f1rc, 1rb = '"'f1rc, 1ra, u~ = u~, and u~- ua+c = ub- ub+c. Thus u E K 1 (F, <G?T).
We see that .6.u = s. D
3. Lemma 3.2 shows that H 2 ( S, <G) is determined by consistent cochains.
Lemma 3.3. When u E K 1 ( F, <G?T), then .6. u E SB 2 ( S, <G) if and only if
u E Z 1 (F,<G1r) + 1r*C1(S,<G).
Proof. If u = v + 1r*w, where v E Z 1 (F,<G1r) and w E C 1 (S,<G), then
u E K 1 (F,<G1r) by Lemma 3.1 and
(.6.u)1ra,1rb = v~- va+b + vb' + w;~- w(1ra)(1rb) + w;g = (8w)1ra,1rb
for all a,b E F, since v E Z 1 (F,<G1r), so that .6.u = 8w E SB 2 (S,<G).

Conversely assume .6. u = 8w E SB 2 ( S, <G), where w E C 1 ( S, <G) . Then
1rb w(1ra)(1rb)
b ua+b + uba = w1ra-
ua- 1ra
+ w1rb =
b (7T *w )a+b + (7T *w )ab
(7T *w )a-
for all a,b E F, and v = U-?T*w E Z 1 (F,<G7T). D

Corollary 3.4. There is an isomorphism
H 2 (s, <G) ~ K 1 (F, <G1r) 1(Z 1 (F,<G1r) + 7T*C 1 (S, <G))
which is natural in <G.
Proof. The isomorphism follows from Lemmas 3.2 and 3.3, since H 2 (S,<G)
~ SZ 2 ( s, <G) I SB 2 ( S, <G) ; it is natural in <G since .6. is natural in <G. D
4. Now let [;;;; be any compatible well order on F and ~ = ker 7T. We use
P, Q, M, q from Section 1 to trim consistent cochains.
A partiall-cochain on F with values in <G is a 1-cochain u = (ua)aEF E
C 1 (F,<G1r) such that uc = 0 for all c E Q. (Thus u is, in effect, a cochain
on P only.) Under pointwise addition, partial 1-cochains constitute a subgroup
P 1 ( F, <G?T) ~ ITaEP GJra of C 1 ( F, <G?T) . Consistent partial 1-cochains constitute
an abelian group KP 1 (F,<G1r) = K 1 (F,<G1r) n P 1 (F,<G1r).
When u E C 1(F,<G1r), define IIu by
(IIu)a = ua - uq(a) E G1ra
for all a E F. If a E Q, then a = q(a) and (IIu)a = 0. Thus II is a

homomorphism of C 1 (F, <G1r) into P 1 (F, <G1r). In fact Im II = P 1 (F, <G1r),
since every partial cochain v satisfies vq(a) = 0 and IIv = v. We see that II is
natural in <G .
Lemma 3.5. When u E C 1 ( F, Gn), then u E K 1 ( F, Gn) if and only if

IIu E KP 1 (F,Gn).
Proof. Let u E C 1 (F,Gn) and v = IIu. Let a,b,c E F satisfy na = nb.
Let r = q(a) = q(b) and s = q(a+c) = q(b+c). Then
(ua+c- u~) - (us- u~), and

(ub+c- ug) - (us- u~).
. consistent,
If u IS . t hen ua+c- uac -- ub+c- ubc an d va+c- vac -- vb+c- vb, c.
hence v is consistent. If conversely v is consistent, then va+c - v~ = vb+c - vg
c h
an d ua+c - uac = ub+c - ub; ence u .IS consistent.
. 0
5. Given a family g = (gx)xEX E IlxEX G1rx (with gx E G1rx for all

X EX), define Dg E C 1 (F,Gn) by
(D g )a -_ """
L.JxEX, x~a axgxa-x _ """
L.JxEX, x~q(a) q (a) x 9xq(a)-x
for all a= L:xEX axx E F. We see that (Dg)a E G1ra = G1rq(a)• and that D
is a homomorphism of IlxEX G1rX into C 1 (F,Gn) and is natural in G.
Lemma 3.6. Im D ~ KP 1 (F,Gn). When u E C 1 (F,Gn), then u E

Z 1 (F,Gn) + n*C 1 (S,G) if and only if IIu E Im D.
Proof. If a E Q, then a= q(a) and (Dg)a = 0; thus Dg E P 1 (F,Gn).

That Dg E K 1 (F,Gn) can be proved directly but follows from Lemma 3.1 and
the rest of the statement, as u E Im D implies IIu = u and u E Z 1 ( F, Gn) +
n*C 1 (S,G) ~ K 1 (F,Gn).
Let u = z + n*w E Z 1 (F,Gn) + n*C 1 (S,G), where z E Z 1 (F,Gn) and
wE C 1 (S,G), so that ua = za + w1ra for all a E F. Since z is a 1-cocycle
we have za+b = z~ + zf: for all a,b E F; hence
for every a = L:xEX ax x E F. Since 1l' q( a) = na,

(IIu)a = za + w1ra - zq(a) - w1rq(a)
= L:xEX, x~a ax Z~-x - L:xEX, x~q(a) q(a)x z~(a)-x

where g = (zx)xEX. Thus IIu E Im D.
Conversely assume IIu E Im D, so that there exists g = (gx)xEX E

f1xEX G1rx such that
ua - uq(a) = l:xEX,x~a axg~-x - l:xEX,x~q(a) q(a)xg~(a)-x

for all a E F. For every a E F let
za = "'
~xEX, x:S;a ax 9xa-x E G1ra ·
If a,b E F, and x ~a+ b (equivalently, (a+ b)x > 0), then x ~a or x ~ b

(or both); hence
"' a ga+b-x
za+b = ~xEX, x~a, x~b x x
+ 2: xEX, x~a, b ga+b-x
x~b x x
+ "'
~xEX, x:S;a, x:S;b (ax+ bx) 9xa+b-x = zab + zba;
thus z E Z 1 (F, G1r). Also ua- uq(a) = za- zq(a) for all a. Let va = ua- za.
Then v E C 1 (F,G1r) and va = vq(a) for all a. Hence 1ra = 1rb implies va = vb
and v = 7r*w, where w E C 1 (S,G) is well defined by WJra = va. Thus
u = z+7r*w E Z 1 (F,G7r) +7r*C 1 (S,G). 0
Corollary 3.7. There is an isomorphism
H 2 (S,G) ~ KP 1 (F,G1r) lim D

Proof. We saw that II: C 1 (F,G1r) ~ P 1 (F,G1r) is a surjective homomor-
phism. Now K 1 (F,G1r) = II- 1 KP 1 (F,G1r) by Lemma 3.5 and Z 1 (F,G1r) +
1r*C1 (S,G) = II- 1 Im D by Lemma 3.6; therefore
K 1 (F,G1r) 1(Z 1 (F,G1r) + 1r*C1 (S,G)) ~ KP 1 (F,G1r) lim D.

This isomorphism is natural in G since II and D are natural in G. The natural
isomorphism H 2 ( S, G) ~ KP 1 ( F, G1r) I Im D then follows from Corollary 3 .4. 0
Recall that a minimal cochain on S with values in G is a family
6.
u = (um)mEM such that um E G1rm for all m E M. Under pointwise
addition, minimal cochains constitute a subgroup M 1 ( F, G1r) ~ ITmEM G 1rm
of C 1 (F,G1r).
Every partial 1-cochain u = (ua)aEF has a restriction Ru = (um)mEM
to M, which is a minimal cochain. This defines a restriction homomorphism
R: P 1 (F,G1r) ---t M 1 (F,G1r) which is natural in G.

Lemma 3.8. R is injective on KP 1 (F,G1r).
Proof. Let u = (ua)aEF be a consistent partial cochain such that Ru = 0
(such that um = 0 for all m E M). We use artinian induction on a to prove
that ua = 0 for all a E P (so that u = 0). Already ua = 0 for all a E Q and
for all a E M. Let a E P\M. Then a > m for some m E M, a = m + c
for some c E F, and b = q( m) + c satisfies 1rb = 1ra and b c a. Since u is
consistent we have um+c- uq(m)+c = u~- u~(m), with um = 0, uq(m) = 0,
and uq(m)+c = ub = 0 by the induction hypothesis; hence ua = um+c = 0. D
7. Lemma 3.8 shows that a consistent partial cochain is determined by its

restriction to M. Therefore Corollary 3.7 can be restated in terms of minimal
cochains; this will yield the main result. First we reconstruct consistent partial
cochains from their restrictions.
Let u be a minimal cochain. When a ;;;;) b in F and p : m 1 , ... , mk is an
overpath from a to b, let
E G1ra,
where ti is obtained from the corresponding path
by ti =pi- q(mi) = pi- 1 - mi. Recall that p 0 , ... ,pk E Ca, so that 1rpi = 1ra
ti .
and umi E G1ra for all z.
We denote ua;p;q(a) by ua;p. If a E Q, then p is empty and ua;p = 0. If

a = m E M, then p = {m} and um;p = um .
When p is an overpath from a to b, then p is an overpath from a+ c to

b + c for any c
E F; the corresponding path is
0 ml 1 m2 mk k
a +c = p +c -----t p +c -----t . . . -----t p + c = b + c,
with pi+ c- q(mi) = pi- 1 + c- mi = ti + c, and
If a b, then p is empty and ua·p·b = 0. If p : m 1 , ... , mk is an overpath

=
' '
from a to b, and q : n 1 , ... , nl is an overpath from b to c, then p + q :
m 1 , ... , mk, n 1 , ... , nl is an overpath from a to c, and
ua· p+q· c
' '
=
' '
ua·p·b
,'
+ ub·q·c ·
In particular, ua;p;b = ua;p+q- ub;q when q is an overpath from b to q(b) = q(a).
8. Independence of path for ua·p·b means that ua·p·b = ua·q·b whenever p

' ' ,' ''
and q are overpaths from a to b; and similarly for ua;p· These properties are
equivalent and characterize the restrictions of consistent partial cochains:
Lemma 3.9. When u E M 1 ( F, G1r), then u E R ( KP 1 ( F, G1r)) if and only

if ua·p, is independent of path; and then ua·p·b
, , is independent of path.
Proof. First let v E KP 1 (F,G1r) and u = Rv. Let p: m 1 , ... ,mk be an
overpath from a to b; let
be the corresponding path and ti = pi - q(mi) = pi- 1 - mi. Since v E

KP 1 (F,G1r) and 1rmi = 1rq(mi) we have
ti
and u m i = v p i-1 - vpi . Therefore
Hence ua·p·b
, , is independent of path. In particular ua·p
, is independent of path.
Conversely let u E M 1 ( F, G1r) . Assume that ua;p is independent of path.
Then v E C 1 ( F, G1r) is well defined by
whenever p is an overpath from a to q( a) . If a E Q, then p is empty and

va= ua;p = 0; thus v E P 1 (F,G1r). If a= m E M, then p = {m} and
vm = um;p = um; thus u = Rv. It remains to show that v is consistent:
va+c- vb+c = v~- vg whenever 1ra = 1rb and c E F.
First let b = q( a) . Let p be an overpath from a to b and q be an overpath

from b + c to q(b +c) = q(a +c). Then p is an overpath from a+ c to b + c
and p + q is an overpath from a+ c to q(a +c). Hence
4. DEFINING VECTORS. 345
va+c ua+c; p+q
ua+c; p; b+c + ub+c; q = u~;p + ub+c; q = v~ + Vb.f.c
and v~ - va+c = -vb+c = -vq(a)+c. If now we assume only 1ra = 1rb, then
q(a) = q(b) and
-vq(a)+c -vq(b)+c
Thus v is consistent. D
9. Finally, recall that a minimal cocycle is a minimal cochain u such that ua;p
is independent of path. Under pointwise addition minimal cocycles constitute a
subgroup MZ 1 (F,G1r) of MC 1 (F,G1r).
A minimal coboundary is a minimal cochain u for which there exists g =
(gx)xEX E flxEX G1rx such that u = RDg; equivalently,
_ """"' m-x _ """"' ( ) q(m)-x
um - uxEX, x~m mxgx uxEX, x~q(m) q m X 9x
for all m E M . Under pointwise addition minimal coboundaries constitute a

subgroup MB 1 (F,G1r) of MC 1 (F,G1r).
Lemma 3.9 shows that R induces an isomorphism of KP 1 (F, G1r) onto
MZ 1 (F, G1r). Since Im D ~ KP 1 (F, G1r) it follows that MB 1 (F, G1r)
Im RD ~ MZ 1 (F,G1r). Then Corollary 3.7 yields
H 2 (S,G) ~ KP 1 (F,G1r)jlmD ~ MZ 1 (F,G1r) / MB 1 (F,G1r)

which is natural in G since R is natural in G. This proves Theorem 2.1. D
4. DEFINING VECTORS.
In this section we show that minimal cocycles are determined by relations

between certain integer vectors, and that the computation of H 2 ( S, G) is a finite
task when S is finitely generated and G is thin.
1. As before, S is a commutative monoid, F = Fx is the free commutative
semigroup on a set X, and 1r: F---+ S is a surjective homomorphism; G = Gx
is the free abelian group on X, whose elements are finite linear combinations
a = l:xEX ax x with integer coefficients and can be regarded as integer vectors.
By Proposition 1.4, e is generated by all pairs (m, q( m)) with m E M,

which may be regarded as defining relations of S. The defining vectors of e
(or of S) are the integer vectors

v(m) = m - q(m) E G
with mE M.
Proposition 4.1. The subgroup of G generated by the defining vectors is
the Redei group R of e; the universal group of s is isomorphic to G I R.
Proof. Recall that the Redei group of e is
R = {a- b E G I a e b}.
Since e is generated by all pairs (m, q( m)) with m E M, it follows from
Proposition 1.2.9 that a- b is a sum of differences m- q( m) and q( m) - m when
a e b, so that a- b belongs to the subgroup K of G generated by the defining
vectors. Hence R s;;; K. Conversely every defining vector v(m) = m- q(m) is
in R, since me q(m); hence K s;;; R.
Since e is generated by all pairs (m, q(m)) with m E M, S U {0} is
generated, as a commutative monoid with zero, by the set X subject to all relations
m = q(m) (mE M), with m, q(m) =/= 0 in S U {0}. By Proposition 111.3.4,
G(S) is the abelian group generated by X subject to all relations m = q(m);
that is, G(S) ~ GIK. D
Proposition 4.2. Let S have a zero element and Z = 7T -l 0 s;;; F be the
zero class. Let K be the subgroup of G generated by all defining vectors v(m)
with m f:. Z. Then G I K is the universal abelian group G ( S\ 0) of the partial
semigroup S\ 0.
Proof. Since Z is a e-class, mE Z implies q(m) E Z. Since e is generated
by all pairs (m, q( m)) with m E M, S is generated, as a commutative monoid
with zero, by the set X subject to all relations m = q( m) ( m E M\ Z) and m = 0
( m E M n Z ). By Proposition Ill.3 .4, G (S\ 0) is the abelian group generated by
X subject to all relations m = q(m) (mE M\Z); that is, G(S\0) ~ GIK. D
2. We now consider relations between defining vectors. We distinguish vector
relations
R(r) :
in which every r m is an integer and r m = 0 for almost all m, and positive
relations
R(r,s): L:":mEM r m v(m) = L:":mEM 8 m v(m),
in which every r m and every sm is a nonnegative integer and r m = sm = 0
for almost all m. The two types are essentially equivalent.
Lemma 4.3. Every positive relation is a trivial consequence of a finite sum

of minimal positive relations; if X is finite, then there are only finitely many
minimal positive relations.
Proof. When R = R(r, s) : I:mEM r m v(m) = I:mEM sm v(m) is a
positive relation, let the weight of R be I:mEM (rm + sm); call R nontrivial
when r m -1- sm for some m, and essential when it is nontrivial but there is
no m such that r m > 0 and sm > 0 (so that no cancellation is possible in
R(r, s) ). Every nontrivial positive relation can be simplified by cancellation in G
into an essential positive relation; hence every nontrivial positive relation R(r, s)
is a trivial consequence of an essential positive relation R( e, f) ( r m - em
sm -fm ~ 0 for all m).
Positive relations are ordered coefficientwise: R(p, q) :;:; R(r, s) if and only
if Pm :;:; r m and qm :;:; sm for all m. A minimal positive relation is a minimal
nontrivial positive relation. Minimal positive relations are essential.
When R(p,q) < R(r,s), then "L:mEM Pm v(m) = I:mEM qm v(m) and
"L:mEM r m v(m) = I:mEM sm v(m) imply
"L:mEM (rm- Pm) v(m) = "L:mEM (sm- qm) v(m),

so that R( r - p, s - q) is a positive relation; then R( r, s) is the sum of R(p, q)
and R(r - p, s - q). If R(r, s) is essential, then so are R(p, q) and R(r -
p, s - q). Thus an essential positive relation which is not minimal is a sum
of essential positive relations of Jesser weight. Therefore every essential positive
relation is a finite sum of minimal positive relations, and every positive relation
is a consequence of a finite sum of minimal positive relations.
Under pointwise addition, pairs (r, s) of families r = (r m)mEM, s =
(sm)mEM of nonnegative integers constitute a finitely generated free commu-
tative monoid F'. Minimal positive relations constitute an antichain of F'. If
X is finite, then, by Dickson's Theorem, M is finite, all antichains of F 1 are
finite, and there are only finitely many minimal positive relations. D
3. Relations between defining vectors arise when there is more than one path
from an element ofF to another. When p : m 1 , ... , mk is an overpath from a
to b, we saw that
a-b
We write this equality as
a- b L:mEM Pm v(m),
where Pm is the number of appearances of m in the sequence p : m 1 , ... , mk.

When p and q are two overpaths from a to b, then
R(p,q) : L:mEM Pm v(m) =a- b = L:mEM qm v(m)
is a positive relation and there is a vector relation
R(p- q) : L:mEM (Pm- qm) v(m) = 0.
A relation between defining vectors is realized at a E F when it arises in this
fashion from a pair of overpaths from a to some b. Thus a vector relation R(r) :
L:mEM r m v(m) = 0 is realized at a when there exist bE F and overpaths p
and q from a to b such that r m = Pm- qm for all m (so that R(r) = R(p- q) );
a positive relation R(r,s): L:mEM rm v(m) = L:mEM sm v(m) is realized at
a E F when there exist b E F and overpaths p and q from a to b such that
r m - Pm = sm - qm ~ 0 for all m (so that R( r, s) is a trivial consequence
of R(p,q)).
Lemma 4.4. When R(r, s) is realized at a, then r m = sm whenever
1rm ~']{ 1ra.
Proof. Let bE F and p, q be overpaths from a to b such that r m- Pm =
sm - qm ~ 0 for all m E M. If Pm > 0 or qm > 0 (if m appears in p or
in q ), then m ~ c for some c E Ca and 1rm ~']{ 1ra. Therefore 1rm ~']{ 1ra
implies Pm = qm = 0 and rm =sm. D
A relation of either kind is realized in a e-class C when it is realized at
some a E C (then bE C in the above). In the case of a vector relation it may be
assumed that b = q(a), since an overpath from b to q(a) = q(b) can be added
to p and q if necessary, without changing Pm - qm.
The trivial relation 0 = 0 is realized at every a E F. A relation which is
realized at a is realized at a + c for every c E F, since overpaths from a to b
are also overpaths from a + c to b + c. More surprisingly:
Proposition 4.5. Every relation between defining vectors is realized in some
e-class (in the zero class, if S has a zero element).
Proof. First we prove the following: for every family r = (r m)mEM of
nonnegative integers (with r m = 0 for almost all m E M) there exists an
overpath p such that Pm = r m for all m. This is shown by induction on
lrl = "EmEM rm. If lrl = 0, then the empty ovepath from any a to a serves.
If lrl > 0, then r n > 0 for some n EM, and the induction hypothesis yields an
overpath p : m 1, ... , mk from some a E F to some b E F such that Pm = r m
for all m "/= n and Pn = r n - 1. In F there is an element c such that c ~ b

and c ~ n (for instance, b V n). Then p : ml, ... , mk is an overpath from
a + (c - b) to b + (c - b) = c, {n} is an overpath from n + (c - n) = c to
q( n) + (c- n), and q = p + {n} : m 1 , ... , mk, n is an overpath from a+ (c- b)
to q(n) + (c- n). We see that qm = r m for all m.
If now R(r, s) is a positive relation, then the above provides an overpath p
from some a to some b such that Pm = r m for all m and an overpath q from
some c to some d such that qm = sm for all m. Then
a-b .L:mEM Pm v(m) .L:mEM r m v(m)
.L:mEM sm v(m) .L:mEM qm v(m) = c- d .
In F there is an element e such that e ~ a and e ~ c; if S has a zero element,

then the zero class Z is an ideal of F and we can arrange that e E Z. Let
f = (e - a) + b = (e - c) + d. Then p and q are overpaths from e to f, and
Pm = r m, qm = sm for all m. In particular R( r, s) is realized in C e. Thus
every positive relation is realized in some ~-class (in the zero class, if S has a
zero element). Then so is every vector relation. D
4. We now show that minimal cocycles are determined by relations between
the defining vectors, when the coefficient functor is thin.
Let G = (G,')') be an abelian group valued functor on H(S). When u =
(um )mEM . IS a. . l coc h'
mmima am an d p : m 1 , ... , m k ·IS as overpathe:trom a to
b, then
where ti =pi- q(mi) = pi- 1 - mi is provided by the corresponding path
If G is thin, then 1'1rmi 1rti depends only on mi and 1r(mi + ti) = 1rpi- 1 = 1ra
'
and is denoted by 1';;:i ; hence
where
Ma {mE M l1rm ~9{ 1ra}.
Recall that nm ~9-C na when m appears in p (when Pm > 0). Thus u is a

minimal cocycle if and only if
Z(p,q,a) :
whenever p and q are overpaths from a E F to q( a) .

Proposition 4.6. When G is thin, a minimal cochain u is a minimal cocycle
if and only if
Z(r,s,a) :
holds whenever a E F and the positive relation R(r, s) is realized at a.

Proof. If R(r, s) is realized at a, then R(r, s) is a trivial consequence of
R(p,q) for some overpaths p and q from a to some bE Ca; then Z(r,s,a) is
a trivial consequence of Z (p, q, a), since Pm = qm = 0 and r m = sm when
m tj. Ma by Lemma 4.5. By cancellation in GTra' Z(r,s,a) holds if and only
if Z(p,q,a) holds. Similarly, when an overpath from b to q(b) = q(a) is added
to p and q (to obtain overpaths from a to q(a) ), then Z(p, q, a) is replaced
by an equivalent condition. Hence u is a minimal cocycle if and only if every
Z(r,s,a) holds. D
On the other hand, minimal coboundaries satisfy relations between defining
vectors regardless of whether they are realized. Call a positive relation R( r, s)
verifiable at a E F when r m = sm whenever m tj. Ma (whenever nm ~9-C na ).
By Lemma 4.4, every positive relation which is realized at a is verifiable at a.
Proposition 4. 7. When G is thin and u is a minimal coboundary, then
Z(r,s,a) :
holds whenever a E F and the positive relation R(r, s) is verifiable at a.

Proof. Let u be a minimal coboundary, so that
um = L:xEX, x~m mx g;(m-x) - L:xEX, x~q(m) q(m)x g;(q(m)-x)
for all m = L:xEX mx x E M, where 9x E G1rx for all x E X. Since G is

thin and nm = nq(m),
um = "" 7rX
L..JxEX, x~m mx f1rm9x - ""
L..JxEX, x~q(m) q (m ) x f7rm9x
7rX
for all mE M.
Let R(r,s): L:mEM r m v(m) = L:mEM sm v(m) be verifiable at a. Then
L:m~Ma rm v(m) = L:m~Ma sm v(m), since rm = sm when m tj. Ma;
hence LmEMa r m v(m) = LmEMa sm v(m) and
LmEMa r m v(m)x = LmEMa 8 m v(m)x

for every X E X. Let xa = {X E X I JrX ~9-C Jra}. Then X ~ m E Ma implies
JrX ~9-C 1rm ~9-C Jra and X E Xa; X ~ q(m) E Ma implies X E Xa; and
'\" 7rm
umEMa r m f1ra um
LmEMa LxEX, x~m r m mx l;~gx
- LmEMa LxEX, x~q(m) r m q(m)x l;~gx
LmEMa LxEXa rm mx l;~gx

- LmEMa LxEXa rm q(m)x l;~gx'
since mx = 0 if x ~ m and q(m)x = 0 if x ~ q(m),
LmEMa LxEXa r m v(m)x l;~gx

LmEMa LxEXa 8m v(m)x l;~gx
LmEMa LxEX, x~m 8m mx l;~gx
- LmEMa LxEX, x~q(m) 8m q(m)x l;~gx
'\" 7rm
umEMa r m f1ra um '
since mx = 0 if x ~ m and q(m)x = 0 if x ~ q(m). Thus u satisfies
Z(r,s,a). D
5. Computing H 2 ( S, G) with Theorem 2.1 still looks like an infinite task
even when S is finite, since independence of path must be established at every
a E F. When F is finitely generated and G is thin, we show that minimal
cocycles are characterized by finitely many conditions ua;p = ua;q; hence the
computation of H 2 ( S, G) a finite task. It seems likely that this result holds even
if G is not thin.
Proposition 4.8. When G is thin and X is finite, a minimal cochain u is
a minimal cocycle if and only if if satisfies finitely many conditions Z (r, s, a), in
which a E F and the positive relation (r, s) is realized at a.
Proof. Let F" be the set of all ordered pairs (r, a) where r = (r m)mEM is
a family of nonnegative integers and a E F. Under pointwise addition, F" ts a
free commutative monoid F" 9:! FM x F.
Realizability yields a binary relation ::R on F":
(r,a) ~ (s,b) -¢:::=:;> a= band R(r,s) is realized at a.

We see that ~ is reflexive and symmetric. If moreover (r, a) ~ ( s, b), so that
R( r, s) is realized at a, then, for any t, R( r + t, s + t) is a trivial consequence
of R(r, s) and is realized at a, R(r + t, s + t) is realized at a+ c for any
c E F, and (r + t, a+ c) ~ (s + t, b +c); thus ~ admits addition.
By Proposition 1.2.9, the congruence ~ on F" generated by ~ is given by:
(r,a) ~ (s,b) if and only ifthere exist k ~ 0 and (r 0,a0), ... ,(rk,ak) E F"
such that (r,a) = (r 0,a0), (ri- 1 ,ai- 1 )~(ri,ai) foralll ~ i ~ k, and
(rk,ak) = (s,b). Then a= a0 = ... = ak =band the equalities
L:mEM r~ v(m) = L:mEM r~ v(m) = ... = L:mEM r~ v(m)

show that R(r, s) is a consequence of R(r 0, r 1 ), R(r 1, r 2 ), ... , R(rk-1, rk).
By Proposition 4.6, a minimal cochain u is a minimal cocycle if and only if
it satisfies Z(r, s, a) whenever (r, a)~ ( s, a). If (r, a) ~ (s, a), then in the above
Z(r 0, r 1 , a), Z(r 1, r 2 , a), ... , Z(rk- 1, rk,a) hold in G1ra; by the equalities
Z(r,s,a) is a consequence of Z(r 0, r 1 , a), Z(r 1, r 2 , a), ... , Z(rk-1, rk, a)

and holds in G1ra if u is a minimal cocycle. Hence a minimal cochain u is a
minimal cocycle if and only if it satisfies Z(r,s,a) whenever (r,a) ~ (s,a).
Since M is finite it follows from Redei's Theorem that~ is finitely generated.
Therefore a minimal cochain u is a minimal cocycle if and only if it satisfies
finitely many conditions Z(r,s,a), with a E F and (r,a) ~ (s,a), each of
which is a consequence of finitely many conditions Z(r,s,a), with a E F and
(r, s) realized at a. D
A more explicit choice of conditions to verifY is given in Grillet [ 1995F] but
no longer seems particularly helpful.
6. We conclude this section with an example. More general examples are
given in Grillet [2000T].
Example 4.9. Let S be the commutative nilmonoid
S = ( c,d I c3 = c2 d = cd2 = d 4 = 0, c2 = cd = d 3 );
S is the Volkov semigroup (Example III.3.6) with an identity adjoined.

Let X = {X' y} and 1rX = c' 1rY = d. Then e = ker 1r has four one element
classes, one three element class C = { 2x, x + y, 3y} = 1r-l ( cd), and one infinite
class J = 1r- 1 o which is the ideal ofF generated by { 3x, 2x + y, x + 2y, 4y}.
The lexicographic order ~ on F
ix + jy C kx + ly ¢::::::? i < k, or i = k, j < l
is a compatible well order on F. Under ~ the least element of C is 3y; the
least element of J is 4y.
.e
Example 4.9 Q and M
Thus Q is the coideal generated by 4y and x; M and the defining vectors are
given by the table
m q(m) v(m)
l = 5y 4y y
m=x+y 3y X- 2y
n = 2x 3y 2x - 3y
The defining vectors v(m) with m tJ. J are v(m) and v(n). They constitute
a basis of G since 1 1 - 2 1 = 1. Hence the universal abelian group of S\ 0 is

2 -3
trivial, by Proposition 4.2.
We see that v(n) = 2v(m) + v(l). The only e-class in which nontrivial
positive relations are realized is J; by Proposition 4.5, every positive relation is
realized in J. (More sophisticated examples are given in the next chapter.)
Let G be a thin abelian group valued functor on H(S). A minimal cochain
u consists of Uz E G0 , um E Gcd' and un E Gcd·
n m m l
In J, 2x + y - + 4y and 2x + y - + x + 3y - + 5y - + 4y are
paths from a = 2x + y to q( a) = 4y. Hence p : n and q : m, m, l are overpaths
from a to q(a) (and the relation v(n) = 2v(m) + v(l) is realized at 2x + y).
Therefore minimal cocycles satisfy

0
un = ua;p 2u~ + uz;
this is according to Proposition 4.6.
Conversely let u be a minimal cochain such that u~ = 2u~ + uz . Let
a E F and let p and q be overpaths from a to q( a) . We may assume that a E J
(otherwise p = q ). We have
a -q(a) PzY + Pm (x- 2y) + Pn (2x- 3y)
Qz Y + qm (X - 2y) + qn ( 2x - 3y);
hence Pm + 2pn = qm + 2qn and Pz - 2pm - 3pn = Qz - 2qm - 3qn . Adding
twice the first equality to the second yields Pz + Pn = Qz + qn . Hence
0 0
Pz Uz + Pm um + Pn un
(Pz + Pn) Uz + (Pm + 2pn) u~
(qz + qn)uz + (qm + 2qn)u~
0 0
Qz Uz + qm um + qn un = ua;q
Thus minimal cocycles are characterized by the single condition u~ = 2u~ + uz ;

this is according to Proposition 4.8. Hence there is an isomorphism u f---.+
(um, un) of MZ 1 (F,G1r) onto Gcd EfJ Ged' A peek at minimal coboundaries
suggests that um and un are uniquely determined by g = un - 2um and h =
2un - 3um (namely, um = h - 2g and un = 2h - 3g ), and provides a more
useful isomorphism e : u f---.+ (un- 2um, 2un- 3um) of MZ 1 (F,G1r) onto
Ged EB Ged·
Next, u is a minimal coboundary if and only if there exist g E Ge and
h E G d such that uz = 5g 0 - 4g 0 = g0 , um = (gd + he) - 3gd = he - 2gd,
and un = 2he- 3gd. Then e(u) = (gd, he) and e sends MB 1 (F,G1r) onto
Im l'e ,dEB Im I'd ,e <;;; Ged EB Ged· Hence
If G is surjecting (as well as thin), then ')'1,e, ')' 1,d, and ')'1,ed are surjective;
hence /'e ,d, and /'d ,e are surjective and H 2 ( S, G) = 0.
5. PARTIALLY FREE SEMIGROUPS. 355
5. PARTIALLY FREE SEMIGROUPS.
Partially free semigroups were defined in Section X.6. At this time they
constitute the only large class of finite commutative semigroups with a formula
for H 2 ( S, G) : namely,
H 2 (S, G) ~ fficEirr (S) Ge(c)/lm 'Y~(c) '

where G is thin, Irr ( S) is the set of all irreducible elements of S, and e( c)
is the idempotent in the archimedean component of c. This result is from Grit-
let [ 1995P].
Other formulas yield H 2 ( S, G) when S has one defining relation (Proposition
2.4) or is cyclic (Corollary 2.5). But no such formula seems to exist for semigroups
with two generators (Grillet [2000T]).
1. Let S be finite and partially free. By Corollary XII.5.5 we may assume
that S is a monoid. Since S is group-free, ~:H: is a partial order relation on S,
which we denote by ~ .
By Proposition X.2.2, S is generated by Irr (S) and has a standard pre-
sentation 1r : F = Fx ----+ S, where X is finite, 1r is injective on X, and
1r(X) = Irr (S). The direction set 1), extent cells EA, co ideals HA, and trace
congruences eA of the congruence e induced by 1r are as in Chapter X. The
idempotents of S are all eA = 1r(IA) with A E 1), where IA = EA n A. Since
S is partially free, all trace congruences are Rees congruences and
a e b if and only if a,b E EA and p~a = p~b for some A E 1).
Put any total order ~ on X and order F lexicographically: L:xEX ax x C
L:xEX bx x if and only if there exists t E X such that ax = bx for all x -< t
and at < bt. (Then x c y in F if and only if x >- y in X.) Since X is finite,
G.:; is a compatible well order on F.
We show that the defining vectors contain a basis of G.

Lemma 5.1. Let x E X, c = 1rx, and D be the smallest element of
1) that contains x. There exists m(x) E M such that: m(x) E En n D;
1rm(x) = e(c); ifx E A E 1), then 1rm(x) ~ 1rajor all a E EA; and
v(m(x)) = x + L:yEX tx,y Y
with tx,y = 0 unless y -< x and y E D.

Proof. First take any m E ID, so that 1rm = e D.
If A E 'D, A CD, then x ~ A by the choice of D. By (E2), kx E ED

for some k > 0. Then ck = 1r (kx) = eD and 1rm = e(c).
If x E A E 'D, then D ~ A, E(1ra) = eA for all a E EA by Proposition
X.3 .4, and 1rm = e D ~ e A ~ 1ra by Proposition X.3 .3.
We now choose m( x) = m E I D as follows. Let w be the least element
of thee-class ID under~- Since XED= D(w) we have w +X E ID. Let
m(x) = m be least under ~ such that mE ID and mx = wx + 1. In particular,
wcm~w+x.
We show that m E M. First m ~ Q = {a E F I q( a) = a}, since w e m

and w C m. To prove that m is minimal with this property (under ~) it suffices
to show that m - y E Q for every y E X, y ~ m. Note that y ~ m implies
y ED.
Assume m-y E ED. If y i x, then (m- y)x = mx = wx + 1; since

m - y E I D and m - y C m this contradicts the choice of m. Therefore
y = x. Hence m - x E ID, m - x ~ w, m ~ w + x, m = w + x, and
m-y=m-x=wEQ.
Now let m - y ~ ED. Then y of x. Also m - y E EB for some B E ']),
B c D. Then D %B and x ~ B. Let q = q( m - y) . Then q ~ m - y
and q e m - y; hence q E EB, and q E D since D is a union of e-clas-
ses. Since S is partially free we also have = p'eq P'e (
m - y) . In particular
qx = (m - y) x = mx . Hence q + y has the following properties: q + y E D,
q+y E ED (since q+y e m), q+y ~ m, and (q+y)x = qx = mx = wx + 1.
By the choice of m, q + y = m, and m - y = q E Q.
This proves mE M. We have w = q(m). Hence v(m)x = mx- wx = 1.
Also v(m)y =my- wy = 0 if y ~ D, since m,w ED. Since w c m there
exists t E X such that wy = my for all y -< t and wt < mt; in particular
t ~ x. If t -< x, then wy = my for all y -< t and wt< mt implies w + x c m,
whereas m ~ w + x; therefore t = x, and v(m)y = my- wy = 0 for all
y-<x.D
Lemma 5.1 implies that (v(m(x)))xEX is a basis of G. In fact:
Corollary 5.2. For every A E 'D, (v(m(x)))xEXnA is a basis of GA"

Proof. GA = G(A) ~ G is the free abelian group on X n A. When
mE M n EA, then q(m) E EA and me q(m) implies p~m = p~ q(m) since
S is partially free; hence v(m) E GA. In particular v(m(x)) E GA for all
5. PARTIALLY FREE 8EMIGROUPS. 357
X E X n A. When X E X n A' then D ~ A in Lemma 5.1 and

v(m(x)) = X + I:yEXnA, y-<x tx,y Y·
Hence the defining vectors v(m(x)) with x EX n A constitute a basis of GA. 0

2. By Corollary 5.2 there is for every m E M n EA an equality
v(m) = I:xEA km,x v(m(x)) E GA
with integer coefficients km x .

'
Lemma 5.3. For every m E M n EA the vector relation
v(m) = I:xEA km,x v(m(x))
is realized in the e-class em of m.

Proof. If m = m( x) for some x E A, then (*) is trivial, and is realized in
em. Hence we may assume that m # m(x) for all x E A.
We have m +A ~ em, since m E EA. Let p : m 1 , ... , mk consist of m
and of -km x copies of m(x) for every x EX n A with km x < 0, arranged in
' '
any order. Let q : n 1 , ... , n 1 similarly consist of km x copies of m( x) for every
'
x EX n A with km,x > 0, arranged in any order. Then I:i v(mi) = I:j v(ni),
by(*). Also Pm- qm = 1 and qm(x)- Pm(x) = km,x for every x EX n A.
If we can show that p and q are overpaths from some E to some b E a em em ,
then they will realize ( *).
We have m(x) E A for all x E X n A, and mi E A for all i > 1. Also
v(mi) EGA by Corollary 5.2. If a Em+ A has sufficiently large coordinates, we
can arrange by induction on i that pi- 1 ~ mi and pi- m =pi- 1 -m+ v(mi) E
A for all i. Then p is an overpath from a to b = a+ I:i v(mi) E m +A.
Similarly, if a E m +A has sufficiently large coordinates, then q is an overpath
from a to a+ I:j v(ni), and I:j v(ni) = b since I:i v(mi) = I:j v(ni) ). 0
3. Now let G be a thin abelian group valued functor on H(S).
Proposition 5.4. When S is partially free and G is thin, a minimal cochain
u = (um)mEM is a minimal cocycle if and only if
whenever A E 'D and mE M n EA.

Proof. By Proposition 4.6, a minimal cocycle u = (um)mEM must satisfy
( **) for all A E 'D and m E M n EA, since (*) is realized at Cm by Lemma
5.3. (Note that 1rm(x) ~ 1rm for all x EX n A, by Lemma 5.1, since mE EA")
Conversely assume that ( **) holds for all A E 'D and m E M n EA.
Let p and q be overpaths from a to b. Then l:mEM Pm v(m) = b- a =
l:mEM qm v(m), with mE Ma = {mE M I 1rm ~ 1ra} whenever Pm > 0
or qm > 0. Also a e band a,b E EA for some A E 'D.
If m appears in either p or q, then m ~ c for some c E C a ~ EA and
m E En for some B ~ A by (E2). Then
v(m) = l:xEXnn km,x v(m(x)) and um = L:xEXnn km,x "Y;:;:(x) um(x) ·

(By Lemma 5.1, 1rm(x) ~ 1rm when x EX n B and mE En). Let km ,x = 0
when m E En and x E X\B. Then v(m) = L:xEXnA km,x v(m(x)) and
L:mEM Pm v(m) L:mEM,xEXnA Pm km,x v(m(x)),

L:mEM qm v(m) L:mEM,xEXnA qm km,x v(m(x)).
Since the defining vectors v(m(x)) with x E X n A constitute a basis of GA
(Corollary 5.2), this implies
L:mEM Pm km,x = L:mEM qm km,x
for all x E X nA and
l:mEMa Pm km,x = L:mEMa qm km,x

for all x E X n A, since m E Ma whenever Pm > 0 or qm > 0. Since
1rm(x) ~ 1ra when x E A by Lemma 5.1, we now have
" Jrm
ua;p;b umEMa Pm "'!Ira um
L:n~A L:mEEsnMa Pm "Y;;:um
1rm(x) um(x)
"un~A "umEEsnMa "uxEXnn Pm k m,x "YJra
1rm(x) um(x)
"un~A "umEEsnMa "uxEXnA Pm k m,x "YJra
1rm(x) um(x)
"umEMa "uxEXnA Pm k m,x "'!Ira
1rm(x) um(x)
"umEMa "uxEXnA qm k m,x "'i1ra
ua;q;b ·
Thus ua;p;b is independent of path. 0

5. PARTIALLY FREE SEMIGROUPS. 359
4. Proposition 5.4 shows that a minimal cocycle is determined by its values

on all m( x) . Hence
Mzl(S,G) ~ EBxEX G7rm(x);
the isomorphism W : MZ 1 (S,G) ---+ EBxEX G1rm(x) sends u = (um)mEM E

MZ 1 (S,G) to (um(x))xEX· We now compute MB 1 (S,G).
Lemma 5.5. w(MB 1 (S, G)) = 8( EBxEX Im 'Y;~(x))' where 8 is the

automorphism of EBxEX G1rm(x) defined for all v = (vx)xEX E EBxEX G1rm(x)
by
( 8v ) X "'""' 7rm(y)
= vx + L.JyEX, tx,yoiO tx,y 'Y7rm(x) vy .
Proof. Let D be the smallest element of 'D containing x. By Lemma 5.1,

1fX ~ 1rm(x) and v(m(x)) = x + L:yEX tx,y y, where tx,y-=/= 0 implies y-< x,
y ED, and (by Lemma 5.1 applied toy) 1rm(y) ~ 1rm(x), since m(x) E ED.
Hence 8 is well defined. Since tx,y I= 0 implies y -< x, the matrix of 8 ts
triangular with 1 's on the diagonal and 8 is an isomorphism.
Let u = 8g E MB 1 (S,G), where g = (gx)xEX E EBxEX G1rx· Then
um = L:xEX, x~m mx 'Y;~gx - L:xEX, x~q(m) q(m)x 'Y;~(m) ·
Let x E X, D be the smallest element of 'D containing x, y E X, and m = m( x) .

By Lemma 5.1, mE ID =ED n D, y ~ m implies y ED, q(m) E ID, and
y ~ q(m) implies y E D. Since 1ry ~ 1rm(y) ~ 1rm(x) for all y E D and
m- q(m) = v(m) = x + 2:yEX tx,y y, we have
2:yEX, y~m my 'Y;lfngy - 2:yEX, y~q(m) q(m)y 'Y;~(m)
L:yEXnD my 'Y;~9y - L:yEXnD q(m)y 'Y;~(m)

"'""'
6yEXnD V
(m ) y 'Y1rmgy
1ry -_ 'Y1rmgx
1rx + "'""'
6yED 1ry
t x,y 'Y1rmgy
'Y;::ngx + 2:yEX, tx,yoiO tx,y 'Y;~9y = (8v)x'
where v = ('Y;~(x)9x)xEX" Thus Wu = 8v. Hence w(MB 1 (S,G))

8 (EBxEX Im 'Y;~(x)). 0
We can now prove
Theorem 5.6. When S is a finite partialy free commutative semigroup and
G is thin, there is an isomorphism
H 2 (S,G) 9'! E9cEirr(S) Ge(c}ilm')'~(c)'

Proof. We may assume that S is a monoid. Since 'l1 and 8 are isomorphisms,
H 2 (S,G) C:,! MZIMB 9'! 8- 1 wMz 18- 1 wMB
( ffixEX G1rm(x)) I (ffixEX lm 'Y;~(x))
ffixEX ( G1rm(x) I lm 'Y;~(x))
fficEirr (S) Ge(c}ilm 'Y~(c)'
since 1r : F ---+ S induces a bijection of X onto lrr (S) and x E X implies

c = nx E Irr (S) and 1r m( x) = e( c), by Lemma 5.1. The isomorphism is natural
in G, since 'l1 and 8 are natural in G. 0
Chapter XIV.
SEMIGROUPS WITH ZERO COHOMOLOGY.
Like other cohomology theories, commutative semigroup cohomology gives

rise to the following problem:
(1) For which commutative semi groups S does Hn ( S, IG) = 0 for all n ~ 2
and all IG?
By Theorem XII.4.4, free commutative semigroups have this property.
The special role of H 2 suggests two additional problems:
(2) For which commutative semigroups S does H 2 (S,IG) = 0 for all IG?
(3) For which finite (more generally, complete) group-free semigroups S does
H 2 ( S, IG) = 0 whenever IG is thin and surjecting (or thin, finite, and surjecting)?
Since H 2 (S,IG) ~ Ext(S,IG), H 2 (S,IG) vanishes for all IG if and only if

every commutative group coextension of S splits. Free commutative semigroups
and free commutative monoids have this property; so do semilattices, by Propo-
sition V.4.4, and free abelian groups, by Proposition V.4.6. Problem (2) asks if
there are any other semigroups with this property.
When Sis complete group-free, then H 2 (S,IG) vanishes whenever IG is thin

and surjecting if and only if every exact J-C-coextension of S splits; equivalently,
if every complete semigroup T with T jJ-C ~ S splits as a coextension of S. By
Proposition V.4.4, semilattices have this property. Problem (3) asks what other
complete group-free semigroups have this property.
From the point of view of the structure and construction of commutative
semigroups, problem (2) is at this time more interesting than problem (1), and
problem (3) is most interesting of all.
Problem (1) is still unsolved, but problems (2) and (3) have been solved in
some major particular cases. In this chapter we solve problem (2) when S is finite
group-free and, after some preliminary results, we solve problem (3) when S is
a finite nilmonoid. The results are due to the author [1997Z], [2001Z]. Problem
361
362 XIV. SEMIGROUPS WITH ZERO CoHOMOLOGY.
(3) was also solved by the author [2000T] for semigroups with two generators,
in which case the solutions are nilmonoids or semilattices.
1. GROUP-FREE MONOIDS.
In this section, S is a finite commutative group-free monoid. We show that

H 2 (S,G) = 0 for all G if and only if S is a semilattice. This was proved by
the author [1997Z].
1. If S is a semilattice, then H 2 ( S, G) = 0 for all G by Proposition V.4.4.
We now let S be a finite commutative group-free semigroup, but not a semilattice,
and cook up a functor G such that H 2 ( S, G) =I= 0. By Corollary XII.5 .5 we may
assume that S is a monoid.
Call an abelian group valued functor G on H(S) selective ifthere exists an
element c of S such that Gc =/= 0 and G8 = 0 for all s =/= c. Then Is t = 0
'
unless s = st = c, in which case Is t is an endomorphism of Gc; and G is thin
'
if and only if Is t is the identity on Gc when s = st = c.
'
Lemma 1.1. If S is partially free, but not a semilattice, then H 2 ( S, G) =/= 0
for some thin finite surjecting and selective functor G.
Proof. By Theorem XIII.5.6 there is for every thin functor G = (G,1) an
isomorphism
H 2 (S,G) 9:! fficEirr(S) (Ge(c)/ lml~(c)),

where Irr (S) is the set of all irreducible elements of S and e (c) is the idempotent
in the archimedean component of c. If S is not a semilattice, then Irr (S), which
generates S, contains an element c which is not idempotent. Then e = e( c) =I= c.
Let G be the thin selective functor in which Ge =/= 0 is any nontrivial finite abelian
group and G8 = 0 for all s =/= e. G is also finite and surjecting. Moreover
H 2 (S,G) =/= 0, since its direct summand Ge(c)/ Im1~(c) 9:! Ge =/= 0. D
2. We now let the finite monoid S be group-free but not partially free.
Green's preorder ~9i: is a partial order relation on S, which we denote as before
by just ~. Let 7f : F = Fx ----+ S be the standard presentation of S and
e = ker 1r. Then 1r induces a bijection of X onto Irr (S), and F is finitely
generated. The direction set 1), extent cells EA, and trace congruences eA of e
are as in Chapter X.
1. GROUP-FREE MONOIDS. 363
Since S is not partially free, one of the trace congruences e B is not a Rees
congruence and has a nontrivial class other than the ideal B\HB ; then there is
a e-class C ~ EB whose projection pkC ~ HB is not trivial. Let c E S be
maximal (under ;£) such that the e-class C = n-1 c has a nontrivial projection
pkC ~ H B (where B E 'D is determined by C ~ EB ).
Lemma 1.2. C does not contain elements a, b such that pka < pkb;
hence 0 't pkC.
Proof. If a,b E C and p~a < p~b, then p~a eB p~b = p~a + t for some
t E B', t > 0; hence
p~a e B p~a + t e B p~a + 2t e B · · · e B p~a + kt
for all k > 0 and p~ C ~ HB contains p~ a + kt for all k > 0. This contradicts
the finiteness of H B (Lemma X.4.1 ).
Assume 0 E p~C. Since p~C is nontrivial, C contains elements a, b such

that p~a = 0 =F p~b; then p~a < p~b, which we just saw is impossible. D
Lemma 1.3. The element c is not idempotent and is not irreducible; hence
cnx = 0.
Proof. If c is idempotent, then c, 2c E C with either p~c < p~(2c) or
p~c = 0, which is impossible by Lemma 1.2.
Since 1r : F ----+ S is the standard presentation, c is irreducible if and only if
c = ny for some y E X, if and only if C n X =1- 0 . Assume that C contains some
y E X. Then y 't B, otherwise 0 = p~y E pkC. Since p~C is nontrivial there
is some a = LxEX ax x E C with p~a =1- y. By Lemma 1.2 we cannot have
y = p~y < p~a; therefore y ~ p~a, y ~a, ay = 0, and a= LxEX,x¥yaxx.

Also a =1- 0 by Lemma 1.2. Since 1r is injective on X this makes c = na a
product of irreducible elements nx =1- ny. If Ia I = LxEX ax > 1, then c = ny
is not irreducible; otherwise Ia I = 1, a = x =1- y, and 1r is not injective on X;
this is the required contradiction. D
3. We now call upon the overpath method. Let :;< be any total order on X
in which X\B precedes X n B (x -< y for all x E X\B and y E X n B).
Order G = Gx lexicographically: let I:xEX ax x C:: I:xEX bx x if and only
if there exists t E X such that ax = bx for all x -< t but at < bt. (Then
x c:: y in G if and only if x ~ y in X.) Then !;;;; is a compatible total order on
G, and induces a compatible well order on F. Since X\B precedes X n B,
pka C p~b implies a C b.
364 XIV. SEMIGROUPS WITH ZERO COHOMOLOGY.
Let E( c) be the least idempotent e ~J-C c of S.

Lemma 1.4. If mE M
and m ~a E C, then either pkm = pkq(m), or
mE C and pkm ::::Jpkq(m). Jfm EM and m ~a E F, where c < 1ra ~ E(c)
in S, then pkm = Pkq(m).
Proof. In either case E(c) = e(1ra) and a E EB (Proposition X.3.4). Assume
pkm =/= pkq(m). Then pkm ::::J pkq(m) (otherwise m C q(m) ). Since m ~a E
EB, we have m E Ev for some D E 'D, D ~ B by (E2). Then X\B ~ X\D;
since X\B precedes X n B, pkm ::::J pkq(m) implies Pbm ::::J pbq(m). Thus
the e-class Cm ~ Ev has a nontrivial projection pbC. Since 1rm ~ c the
choice of c implies 1rm = c, and mE C. Then c <j.1ra, since 1ra ~ 1rm. D
In what follows
MB {mE M I pkm = pkq(m)} and
Me = {mE MImE C and pkm ::::J pkq(m)}.
By Lemma 1.4, when mE M and m ~a, then mE MB UMe if a E C, and
m E MB if c < 1ra ~ E(c).
Lemma 1.5. Let p be an overpathfrom a E C to b. If pka = pkb,
then
p consists solely of elements of MB. If pka =/= pkb, then p consists of elements
of MB and one element m of Me such that p~v(m) = p~(a- b).
Proof. Let p : m 1 , ... , mk be an overpath from a E C to b and
0 ml 1 m2 mk k
a= p -----+ p -----+ .. . -----+ p = b
be the corresponding path, so that pi- 1 ~ mi and pi - pi- 1 = v(mi) for all
i > 0. Since a E C, then mi ~ pi- 1 E C and mi E MB UMe for all i, by
Lemma 1.4. Also
a- b = Ei v(mi);
since pkv(mi) = 0 when mi E MB,
I I I i
PEa- PBb = EmiEMc PBV(m) •
with pkv(mi) ::::J 0 in G since pkmi ::::J pkq(mi) for all mi E Me.
If pka = Pkb, then 0 = pkv(mi) is a sum of positive elements

EmiEMc
of G (ordered by ~ ), which is not possible unless the sum is empty; hence p
consists solely of elements of MB.
1. GROUP-FREE MONOIDS. 365
If pka # pkb, then I:miEMc pkv(mi) # 0 and there is some mi E Me. Let
j be the least i such that mi E Me. For all i < j we have mi E M 8 for all i < j
and PkPi -PkPi-1 = pkv(mi) = 0. Therefore pka = PkP0 = pkr)-l ~pkmj.
Since mj E C this implies pka = pkmj, by Lemma 1.2. Also q(mj) = q(a) ~ b,
so that pkq(mj) !::; pkb (otherwise q(mj) ::J b) and pkq(mj) = Pkb by Lemma
1.2. (In fact, pkq(mj) = pkq(a) .) Hence pka- Pkb = pkv(mj) and
I:miEMc, mi=f.rrJ pkv(mi) = 0.

As before, this sum must be empty. Hence mj is the only element of Me which
appears in p. D
Corollary 1.6. Me i 0.
Proof. pka # pkq(a) for some a E C, since pkC is nontrivial; there exists
an overpath from a to q( a), which by Lemma 1.5 includes some m E Me. D
4. Now let G be the thin selective functor in which Gc is any finite abelian
group and G8 = 0 for all s # c. Then G is finite, but not surjecting since
Im 'Y~(c) = 0. By Lemma 1.3, G8 = 0 when s is idempotent and when s is
irreducible. Hence IlxEX G-rrx = 0 and MB 1 (S,G) = 0. We use Lemma 1.5
to construct nontrivial minimal cocycles.
Lemma 1.7. Given 9x E Gc for every x E X\B let
0 E G-rrm ifm ~ M 0
{
um = Z:xEX\B v(m)x9x E Gc ifm E Me.
Then u is a minimal cocycle.
Proof. Let a E F and p be an overpath from a to b. We show that
is independent of path (where Pm is the number of appearances of m in p ).

If a ~ C, then G-rra = 0 and ua·p·b = 0. Now let a E C. Since um = 0
' '
when m ~ Me and nm = na when m E Me we have
We now invoke Lemma 1.5. If pka = pkb, then p consists solely of elements
of M 8 and ua;p;b = 0. If pka =/= pkb, then p consists of elements of M 8 and
one element n of Me such that pkv(n) = Pk(a- b). Then v(n)x = ax- bx
366 XIV. 8EMIGROUPS WITH ZERO COHOMOLOGY.
for all x E X\B and
ua;p;b = Un = L:xEX\B (ax- bx) 9x ·

In either case depends only on a and b. D
ua·p·b
' '
5. Since Me i= 0 (Corollary 1.6) it is possible to choose the finite abelian
group Gc and 9x E Gc so that um i= 0 for some m E Me. (For instance take
any n E Me; then v(n)x i= 0 for some x E X\B; let Gc be cyclic of order
p, where p does not divide v(n)x, and let 9x i= 0, 9y = 0 for all y i= x; then
un = v(n)x 9x i= 0.) Then MZ 1 (S,G) i= 0; since we saw that MB 1 (S,G)
= 0
it follows that H 2 ( S, G) i= 0, and we have proved that H 2 ( S, G) i= 0 for some
G if S is not partially free. Since G is thin finite and selective we have in fact
proved:
Theorem 1.8. For a finite group-free commutative semigroup S the follow-
ing conditions are equivalent:
(1) H 2 (S, G) = 0 for all G;
( 2) H 2 ( S, G) = 0 whenever G is thin, finite, and selective;
(3) S is a semilattice.
2. THE ZERO GROUP.
In this section we assume that S has a zero element; for instance, that S is
finite group-free. We study how H 2 ( S, G) depends on the zero group G0 . This
yields necessary conditions that H 2 ( S, G) vanish when G is Schtitzenberger.
1. When S has a zero element, an abelian group valued functor G on H (S)
is almost null if Ga = 0 for all a i= 0 and reduced if G0 = 0.
When G is thin and almost null, Proposition V.4. 7 provides isomorphisms
H 2 (S,G) ~ PHom(S\0, G0 ) ~ Hom(G(S\0), G0 ); the partial homomorphism
<p which corresponds to the cohomology class cis s of s E S Z 2 ( S, G) sends
a E S\0 to c.p(a) = sa,O.
Proposition 2.1. LetS have a zero element. For every abelian group valued
fonctor G on H ( S) there is a short exact sequence
0 ----+ G' ----+ G ----+ G" ----+ 0

2. THE ZERO GROUP. 367
which is natural in CG, in which CG' is almost null and CG" is reduced. If CG is
thin, then CG' and CG" are thin. If CG is finite, then CG' and CG" are finite. If S
is complete group-free and CG is thin and surjecting, then CG' and CG" are thin
and surjecting.
Proof. CG' = (G', 1') and CG" = (G", 1") are defined as follows. Let G~ = 0
for all a 1- 0 and Gb = G0 ; let l~,t = 0 if a f. 0 and 1b,t = lo,t. Then CG'
is almost null, and CG' is thin if CG is thin (so that lo,t is the identity on G0 for
all t ), and is finite if CG is finite. If S is complete group-free and CG is thin and
surjecting, then CG' is thin and surjecting.
Let G~ = Ga for all a f. 0 and G~ = 0; let ~~ t = 0 if at = 0, ~~ t = 1 a t
' ' '
if at f. 0. Then CG" is reduced, and is thin (finite, surjecting) if CG is thin (finite,
surjecting).
The exact sequence CG' ~ CG ~ CG" is defined as follows: if a f. 0,

then aa = 0 and f3a = lea; a 0 = le0 and {30 = 0. The following diagrams
commute:
G'a = 0 ----+ Ga G"a
l~,tl la,tl l~~,t
G~t = 0 ----+ Gat = G~t
whenever a, at f. 0;
1,,a,t
G~ = 0 ----+ Ga G"a
l~,tl la,tl
G~t Go ----+ 0 = G~
whenever a f. 0 and at = 0; and
Gb=Go ----+ 0 = G~
lb,tl IO,tl l~~.t
Gb=Go ----+ 0 = G~.
Thus a and {3 are natural transformations. Naturality in CG is similar. D
2. By Theorem XII.4.5 there is an exact sequence
H 2 ( S, CG') -----* H 2 ( S, CG) -----* H 2 ( S, CG")

which is natural in CG, in which the homomorphisms are induced by a and {3.
Proposition 2.2. If S has a zero element, then H 2 (S, G) = 0 whenever G

is thin and surjecting if and only if
(N) H 2 (S,G) = 0 whenever G is thin and almost null, and
(R) H 2 (S,G) = 0 whenever G is thin, surjecting, and reduced;
also, H 2 ( S, G) = 0 whenever G is thin finite and surjecting if and only if
(Nf) H 2 (S,G) = 0 whenever G is thin finite and almost null, and
(Rf) H 2 ( S, G) = 0 whenever G is thin finite surjecting and reduced.
Proof. In the exact sequence
0 --+ G' --+ G --+ G" --+ 0
in Proposition 2.1, if G is thin (finite, surjecting), then G' and G" are thin (finite,
surjecting). If therefore H 2 ( S, G) = 0 whenever G is thin and surjecting, then
(N) and (R) hold; if H 2 (S, G) = 0 whenever G is thin finite and surjecting,
then (Nf) and (Rf) hold.
If conversely (N) and (R) hold, and G is thin and surjecting, then H 2 (S, G')
= 0, H 2 ( S, G") = 0, and the exact sequence
H 2 (S,G') --+ H 2 (S,G) --+ H 2 (S,G")
shows that H 2 (S,G) = 0. Conditions (Nf) and (Rf) similarly imply H 2 (S,G) = 0
whenever G is thin finite and surjecting. 0
Conditions (N) and (Nf) are easily settled when S is finite. Let 1r : F =
Fx --+ S be a presentation of S, e be the congruence induced by 1r, and
Z = 1r- 1 o be the zero class. In the following result, K is the subgroup of
G generated by all differences a - b with a e b and a, b 1. Z; relative to any
compatible well order on F, K is also generated by all defining vectors v( m)
with m 1. Z.
Proposition 2.3. Let S be a finite commutative semigroup with a zero
element. The following conditions on S are equivalent:
(N) H 2 ( S, G) = 0 whenever G is thin and almost null;
(Nf) H 2 (S,G) = 0 whenever G is thin finite and almost null;
(K) there is a presentation of S in which K = G;
(K +) K = G in every presentation of S.
Proof. By Proposition V.4.7, H 2 (S,G) ~ Hom(G(S\0), G 0 ) whenever G
3. NILMONOIDS. 369
is thin and almost null. Now G(S\0) ~ G I K by Proposition XIII.4.2, in any

presentation of S. Hence (K) implies (N), which in tum implies (Nf). If on the
other hand S is finite and G I K f. 0, then G I K ~ G ( S\ 0) is finitely generated,
there exists a finite abelian group A such that Hom (G I K, A) f. 0, and (Nf)
does not hold; therefore (Nf) implies (K+). D
3. NILMONOIDS.
In this section, we characterize finite commutative nilmonoids S such that

H 2 (S, G) = 0 whenever G is thin finite and surjecting; equivalently, every
elementary semigroup T such that T I'J{ ~ S splits as a coextension of S. This
result is due to the author [200 1Z].
1. To explore nilmonoids we use certain simple coefficient functors. When
J is an ideal of a commutative semigroup S, an abelian group valued functor
G = (G, "Y) on H ( S) is semiconstant over J when there is an abelian group
A such that
G8 = A for all s E S\J, G8 = 0 for all s E J,
"'( 8 t = 0 if st E J. If J = 0, then G is constant.

lA if st ~ J, and "'(8 t =
' '
If S has a zero element, then almost constant functors are semiconstant over 0;
conversely one can view a semiconstant functor G on S over an ideal J f. 0
as an almost constant functor on S I J. In general, a semiconstant functor is thin,
surjecting, and (if S has a zero element and J f. 0) reduced.
Let 1r : F ----+ S be any presentation of S, ~ = ker 1r be the congruence
induced by 1r, and ~ be a compatible well order on F. Let J be an ideal of
S. Let MJ and X J be the sets
MJ = {mE M J1rm ~ J}, XJ = {x EX J1rx ~ J}.
Lemma 3.1. When G is semiconstant over J, then:

(1) a minimal cochain u is determined by its values (um)mEMJ on MJ;
(2) u is a minimal cocycle if and only if
for every positive relation L:mEMJ r m v(m) = L:mEMJ sm v(m) which is

realized in a ~-class C ~ 1r-l J;
(3) u is a minimal coboundary if and only if there exists g = (gx)xEXJ such

that 9x E A for all x E XJ and
um = l:xEXJ v(m)x 9x
for all mE MJ;
(4) when u is a minimal coboundary, then
l:mEMJ rm um = 0
for every vector relation l:mEMJ r m v(m) = 0 which holds in G.
Proof. ( 1) is clear since um = 0 whenever 1rm 1. J.
(2) follows from Proposition XIII.4.6: u is a minimal cocycle if and only if
whenever a E S and the positive relation l:mEM r m v(m) = l:mEM sm v(m)

is realized in ca' where Ma = { m E M I 7rm ~9{ 7ra} . If 7ra E J' then
Z(a,r,s) is trivial. If 1ra 1. J, then Ma ~ MJ and Z(a,r,s) is equivalent to
since r m > 0 implies 1rm ~9-C 1ra and similarly for sm > 0.
(3) follows from the definition of 8g: when g = (gx)xEX, then u = 8g is
given by
um = l:xEX, x~m mx 1;::n9x - l:xEX, x~q(m) q(m)x l;::n9x

for all mE M. Again um = 0 if m 1. MJ. If mE MJ, then x ~ m implies
x E XJ, and so does x ~ q(m), and
um l:xEXJ, x~m mx 9x - l:xEXJ, x~q(m) q(m)x 9x
l:xEXJ (mx- q(m)x) 9x = l:xEXJ v(m)x 9x'

since mx > 0 implies x ~ m and similarly for q(m)x > 0.
(4) follows from (3). Assume that l:mEMJ r m v(m) = 0 holds in G. Then
l:mEMJ r m v(m)x = 0 for every x E X. If u is a minimal coboundary, then
l:mEMJ r mum = l:mEMJ l:xEXJ r m v(m)x 9x = 0. D

2. As a consequence of Lemma 3 .1, we prove:
Lemma 3.2. Let J be a nonempty ideal of S which contains every element
s E S such that a nontrivial positive relation is realized in C 8 = 1r -l s. Jf (Rf)
3. NILMONOIDS. 371
holds, then the defining vectors v(m) with m E MJ are distinct and linearly
independent.
Proof. Let A be any finite abelian group and G be the corresponding semi-
constant functor over J, which is thin finite surjecting and reduced.
Every minimal cochain u = (um)mEM is a minimal cocycle: if p and q
are overpaths from a to b, then either p and q consist of the same elements of
M, in which case ua;p;b = ua;q;b' or l:mEM Pm v(m) = l:mEM qm v(m) is
a nontrivial relation which is realized in Ca = C b, in which case 1ra = 1rb E J,
G1ra = 0, and ua·p·b = ua·q·b = 0. Hence MZ 1 (S,G) = MC 1 (S,G) ~
, ' ' '
ITmEMJ A.
If there is a nontrivial vector relation r: l:mEMJ r m v(m) = 0 between the

vectors v(m) with mE MJ, then
l:mEMJ r m um = 0
for every minimal coboundary u, by Lemma 3 .1. If A is a cyclic group of suitable
prime order p, then p does not divide every nonzero r m and there is a minimal
cochain u such that l:mEMJ r mum i- 0; for instance, let um i- 0, where p
does not divide r m, and un = 0 for all n i- m ). Then u is a minimal cocycle
but not a minimal coboundary, H 2 (S,G) ~ MZ 1 (S,G) / MB 1 (S,G) i- 0, and
(Rf) does not hold. D
3. We now let S be a finite nilmonoid and assume that F is finitely generated.
A thin abelian group valued functor G on S is surjecting if and only every
homomorphism /'; is surjective.
A vector relation l:mEM rm v( m) = 0 is reachable in a e-class C when it
follows from relations that are realized in C (when it is a linear combination with
integer coefficients of relations that are realized in C). By Proposition XIII.4.5,
every vector relation is reachable in some e-class and is reachable in the zero
class Z = 1r- 1o.
Let J be a nonempty ideal of S. Let KJ be the subgroup of G generated by
all defining vectors v (m) with m E M J . G is a finitely generated free abelian
group and so is KJ ~ G. A defining basis of KJ (relative to J) is a subset B
of MJ such that
(1) the defining vectors v(m) with m E B are distinct and constitute a
basis of KJ, so that for every m E MJ \B there is a unique vector relation
v(m) = l:nEB r n v(n) (with integer coefficients); and
(2) for every m E MJ \B the relation v(m) = I:nEB r n v(n) is reachable

in em.
If m tt B and the relation v(m) = I:nEB r n v(n) is reachable in a e-
class C8 = then some vector relation containing m is realized in C8 and
1r-l s,
1rm ~:K s; thus (2) states that the relation v( m) = I:nEB r n v( n) is reachable
in the highest possible e-class.
Lemma 3.2 shows that in some cases MJ itself is a defining basis of KJ.
Our main lemma is:
Lemma 3.3. Let S be a finite nilmonoid and J be a non empty ideal of S.
lf (Rf) holds, then KJ has a defining basis.
Proof. We assume (Rt) and proceed by induction on IS\JI. Let J0 be the
set of all s E S such that a nontrivial positive relation is realized in C8 = 1r- 1 s.
Then J 0 is an ideal of S, since a relation which is realized in C8 is realized in
every Cst. By Lemma 3.2, KJ has a defining basis whenever J contains J 0 .
This kickstarts the induction.
For the general case we expand S\J from the bottom, which matches what
writing this book is doing to the author. Let J be a nonempty ideal of S. Assume
that KJ has a defining basis B (relative to J) and that J o:l 0. Let s be a
maximal element of J (under ~:K ), so that J' = J\ { s} is an ideal of S and s
is a minimal element of S\J' = (S\J) U { s}. We construct a defining basis of
KJ'.
We have MJ' = MJ U M 8 , where
M8 = M nC8 = { m E M l1rm = s}.

Since S is a nilmonoid, the e-class C8 , which is not the zero class, cannot
contain comparable elements a< b. Therefore an overpath p: m 1 , ... , mk from
a E C8 to b E C8 contains at most one element of M 8 which must be its last
element mk: if a = p0 , ... ,pk = b is the corresponding path and mj E M 8 ,
then ~- 1 ~ mj implies ~- 1 = mj in C8 , and then~= q(mj), so that the
path p 0 , ... ,~ has reached the least element of C8 and continues no further.
Therefore the positive relations which are realized in C8 contain at most two
elements of M 8 and are of three types:
(a) relations I:nEMJ r n v(n) = I:mEMJ sm v(m) containing no element

of M 8 ;
3. NILMONOIDS. 373
(b) relations v(m 1 ) + LnEMJ r n v(n) = LmEMJ sm v(m) containing one

element m 1 of M 8 , with coefficient 1;
(c) relations v(m 1 ) + .L:nEMJ r n v(n) = v(m 2 ) + LmEMJ sm v(m) con-
taining two elements m 1 =/= m 2 of M 8 , with coefficients 1.
From B U M8 we extract a defining basis BUD of KJ'. Starting from
D = M8 we trim D, one element at a time, as follows. If m 1 E D appears in
a relation v( m 1 ) = LnEMJ r n v( n) of type (b), then remove m 1 from D and
replace v(m 1 ) by LnEMJ r n v(n) in every other relation (of type (b) or (c)) in
which v(m 1 ) appears; this yields relations of type (a) or (b) which are reachable
in C8 • If m 1 E D appears in a relation v(m 1) = v(m 2 ) + LnEMJ r n v(n)
of type (c) (with m 1 i= m 2 ), then remove m 1 from D and replace v( m 1 ) by
v(m 2 ) + LnEMJ r n v(n) in every other relation (of type (b) or (c)) in which
v (m 1 ) appears; this yields relations of type (b) or (c) which are reachable in C8 •
This process terminates since M 8 is finite. Then all relations of type (b) or (c)
have been used and D has the following properties:
(A) no relation v(m 1 ) + LnEMJ rn v(n) = LmEMJ sm v(m) with m 1 ED,
or v(m 1 ) + LnEMJ rn v(n) = v(m 2 ) + LnEMJ sn v(n) with m 1 , m 2 E D
and m 1 i= m 2 , can be realized in C8 ;
(B) for every m E M 8 \D there is a relation v(m) = LnEMJUD r n v(n)
which is reachable in C8 ; in particular,
(C) the defining vectors v(m) with m E MJ U D generate KJ'.
These properties imply:
(A*) If 1ra rt. J', then no nontrivial relation
R: LmEBUD r m v(m) = .L:mEBUD 8 m v(m)
can be realized in Ca. Indeed assume that R can be realized in Ca. If r m i= sm
for some mE D, then 1rm ~9-C 1ra, 1ra = s since s is minimal in S\J', the given
relation R is realized in C 8 , and R is ofthe form v(m 1 ) + LnEMJ rn v(n) =
LnEMJ sn v(n) with m 1 E D, or of the form v(m 1 ) + LnEMJ r n v(n) =
v(m 2 ) + LnEMJ sn v(n) with m 1 , m 2 ED and m 1 i= m 2 , which contradicts
(A). Therefore r m = sm for all m E D; then r m = sm for all n E BUD,
since the vectors v( n) with n E B are distinct and linearly independent, and R
is trivial.
(B*) for every m E MJ' \(BUD) there is a relation v(m) = LnEBUD
374 XIV. SEMIGROUPS WITH ZERO CoHoMOLOGY.
r n v(n) which is reachable in em. This follows from (B) if mE M 8 ; if mE M 1

there is a relation v(m) = l:nEB rn v(n) which is reachable in em since B is
a defining basis of K 1 .
(C*) the defining vectors v(m) with m E BUD generate KJ'.
We show that the defining vectors v (m) with m E B U D are distinct and
linearly independent; then (B*) and (C*) show that BUD is a defining basis of
KJ' (relative to J').
As in the proof of Lemma 3.2, let A be any finite abelian group and G be
the corresponding semiconstant functor over J', which is thin finite surjecting
and reduced.
By Lemma 3.1, a minimal cochain u = (um)mEM is determined by its
values (um)mEMJ' on MJ', and u = (um)mEM is a minimal cocycle if and
only if L:mEMJ' r mum = L:mEMJ' sm um whenever the positive relation
l:mEMJ' r m v(m) = L:mEMJ' sm v(m) is realized in a e-class e 1;. 1r-l J'.
Hence u is a minimal cocycle if and only if l:mEMJ' r mum = 0 whenever the

vector relation l:mEMJ' r m v(m) = 0 is reachable in a e-class e 1;. 1r-l J'.
For every m E MJ' there is by (B*) a relation v(m) = l:nEBUD r n v(n)

which is reachable in em. Hence every minimal cocycle u satisfies um =
L:nEBUD r n un. Therefore a minimal cocycle u is uniquely determined by its
values (un)nEBUD on BUD, which can be chosen arbitrarily since no nontrivial
vector relation L:nEBUD rn v(n) = 0 can be realized in any e-class e 1;. 1r-l J',
by (A*). Thus MZ 1 (S,G) ~ ITnEBUD A has IAIIBUDI elements.
Ifthere is a nontrivial vector relation r: l:nEBUD r n v(n) = 0 between the
vectors v(n) with nEB U D, then
l:nEBUD r n un = 0
holds for every minimal coboundary u, by Lemma 2.3. If A is a cyclic group of
suitable prime order p, then p does not divide every r n and there is a minimal
cocycle u such that L:nEBUD r nun i= 0; for instance, let um i= 0, where p does
not divide r m, and un = 0 for all n i= m, n E BUD. Then u is not a minimal
coboundary and H 2 (S, G) = MZ 1 (S, G)/ M B 1 (S, G) i= 0. If therefore (Rf)
holds, then there can be no nontrivial relation L:nEBUD r n v(n) = 0. Hence
BUD is a defining basis of KJ'. D
4. We can now prove:
3. NILMONOIDS. 375
Theorem 3.4. For a finite commutative nilmonoid S the following condi-

tions are equivalent:
(1) H 2 (S,G) = 0 whenever G is thin and surjecting;
(2) H 2 (S,G) = 0 whenever G is thin finite and surjecting;
(3) in some presentation 1r : F ---t S (with F finitely generated), G has a
defining basis (relative to 0);
(4) in every presentation 1r : F ---t S (with F finitely generated), G has a
defining basis (relative to 0).
If for example S is the Volkov nilmonoid (Example XII.4.9), we saw in
Section XIII.4 that G has a defining basis; we also saw that H 2 ( S, G) = 0
whenever G is thin finite and surjecting.
Proof. If (2) holds, then G = K = K 0 by Lemma 3.1 and G = K 0 has a
defining basis by Lemma 3.2, applied to the ideal J = {0}. Thus (2) implies
(4). It remains to show that (3) implies (1).
Assume that G has a defining basis B relative to 0, in some presentation
1r : F ---t S where F is finitely generated. Then K = G and H 2 ( S, G) = 0
whenever G is thin and almost null, by Proposition 2.2. Now let G be thin,
surjecting, and reduced ( G0 = 0 ).
Since B is a defining basis of G there is for every m E M 0 \B a relation
v(m) = l:nEB r n v(n) = l:nEB, rn#O r n v(n)

which is reachable in Cm; in particular 1rn ~:J-C 1rm when r n #- 0. Then
~ 7rn
um = wnEB, Tn#O r n f7rm un
for every minimal cocycle u. Also um = 0 whenever 1rm = 0 since G is reduced.
Thus a minimal cocycle is determined by its values (un)nEB on B. (The latter
can be chosen arbitrarily, as readily shown, so that M Z 1 ( S, G) ~ ITnEB G1rn.)
Since the vectors v(n) with n E B constitute a basis of G, their co-
ordinate matrix (v(n)x)nEB, xEX has an inverse, which is an integer matrix
(tn,x)nEB, xEX such that
l:xEX v(m)x tn,x = 1 if m = n, 0 if m =/= n

for all m, n E B. Let u be any minimal cocycle. Since G is surjecting there is
for every n E B some hn E G 1 such that un = 1J;.nhm. For every x EX let
9x = 2:::-:nEB tn,x 'Y!xhn E G1rx ·
Since v(m)x-=/= 0 implies x ~ m or x ~ q(m), and 1rx ~:H 1rm, we have
2:::-:xEX, 1rx~1rm v(m)x 'Y;::ngx
2:::-:xEX, 1rx~1rm 2:::-:nEB v(m)x tn,x 'Y!mhn
2:::-:xEX 2:::-:nEB v(m)x tn,x 'Y!mhn = 'Y!mhm = um

for every m E B. Since u and 8g are minimal cocycles, this implies u = 8g.
Thus H 2 (S,G) = 0 whenever G is thin, surjecting, and reduced. By Proposition
2.1, H 2 ( S, G) = 0 whenever G is thin and surjecting. 0
Theorem 3.4 does not extend immediately to every finite group-free semigroup
S. If indeed S has two generators and H 2 ( S, G) = 0 whenever G is thin finite
and surjecting, then S is a either a semilattice or a nilmonoid (Grillet [2000T]).
Theorem XIII.5.6 also shows that a partially free semigroup S does not in general
satisfy H 2 ( S, G) = 0 whenever G is thin finite and surjecting, even though G
always has a defining basis, by Lemmas XIII.5.l and XIII.5.3.
5. We conclude this section with some examples. All examples have two
generators c and d; in their standard presentation we let X = {x, y}, with
1rx = c, 1ry = d, and order F lexicographically, with ix + jy C kx + ly if and
only if either i < k, or i = k and j < l.
Example 3.5. Let S be the nilmonoid with the presentation
S S::! ( c,d I c3 d = r? = 0 ).
d 5 , c4 = c2 d 3 , c5 = c3 d 2 = c2 d 5 =
The nontrivial e-classes (other than the zero class) are { 5y, 3x + y}, { 2x +
3y, 4x }, and { x + 5y, 2x + 4y, 4x + y}. Hence M 0 consists of m 1 = 3x + y,
m 2 = 4x, m 3 = 2x+4y. We have q1 = q(m 1 ) = 5y, q2 = q(m 2 ) = 2x+3y,
q3 = q(m3 ) = x + 5y, and v(m 1 ) = 3x- 4y, v(m 2 ) = 2x- 3y, v(m 3 ) = x- y.
Thus v(m 1 ) and v(m 2 ) constitute a basis of G (since I~ =~I = -1).
There are two paths from 4x +y to x + 5y :

4x +y ~ x + 5y and 4x + y ~ 2x + 4y ~ x + 5y;
the corresponding overpaths are m 1 and m 2 , m 3 . Thus the relation v(m 3 )
v(m 1 ) - v(m 2 ) is realized in Cm 3 , and {m 1 , m 2 } is a defining basis of G.
Therefore H 2 ( S, G) = 0 whenever G is thin finite and surjecting. 0
3. NILMONOIDS. 377
5
" ,,
ql q3
0
:~
. . . 1
• • • • m2
0 4
Example 3.5
S ~ ( c,d I c3 = c2d 2 = d3 , c4 = c3d = cd3 = d4 = 0 ).

There is one nontrivial e-class (other than the zero class): { 3y, 2x + 2y, 3x}.
Hence M0 consists of m 1 = 2x + 2y and m 2 = 3x, with q = q( m 1 ) = q( m 2 ) =
3y and v(m 1 ) = 2x- y and v(m 2 ) = 3x- 3y.
3 ~1
0 . : : \ m2
0 3
Example 3.6
-l, -3 ).
The defining vectors v(m 1 ) and v(m 2 ) are linearly independent but do not
constitute a basis of G (since 1

2 = Hence K =/=- G and (Nf)
3 -3
does not hold: there exists an almost null functor G such that H 2 ( S, G) =/=- 0.
Since GjK ~ Z 3 the almost null functor with G 0 = Z 3 has H 2 (S,G) ~
Hom (Z3 , Z3 ) ~ Z3 . However, { m 1 , m 2 } is a defining basis of K. D
S ~ ( c,d I c6 = cd7 , c5 d = c3 ~ = d8 , c7 = c5 d 2 = c2 d 5 = rP = 0 ).
The nontrivial e-classes (other than the zero class) are { 8y, 3x + 4y, 5x + y },
{ x + 7y, 6x }, and { x + 8y, 4x + 4y, 6x + y}. Hence M0 consists of m 1 =
3x + 4y, m 2 = 5x + y, and m 3 = 6x. We have q = q(m 1 ) = q(m 2 ) = 8y,
q3 = q(m3 ) = x + 7y, and v(m 1 ) = 3x- 4y, v(m 2 ) = 5x- 7y, v(m3 ) =
5x- 7y; v(m 1 ) and v(m 2 ) constitute a basis of G, since 1 3
5 -7
-4 ~ = -1.
8 q"
5 ;~
.~
• • 1
0
0
Example 3.7
The only e-class with two overpaths (other than the zero class) is C =
{ x + 8y, 4x + 4y, 6x + y}, which does not contain m 2 or m 3 . The relation
v(m 3 ) = v(m 2 ) is realized in C but it is not reachable in Cm 3 . Therefore
{ m 1 , m 2 } is not a defining basis of G. Similarly { m 1 , m 3 } is not a defining
basis of G. Therefore G does not have a defining basis and (Rf) does not hold;
H 2 (S,G) i 0 for some functor G which is thin finite surjecting and reduced.
The proof of Lemma 3 .2 provides such a functor. Let t = c 5 d = c3 d 4 = d 8 ,

so that J 0 = {O,t} and 1rm 1 , 1rm2 , 1rm3 ~ J0 . Since v(m 1 ), v(m 2 ),
v(m 3 ) are not linearly independent, there is a finite abelian group A such that
H 2 (S,G) i 0 when G is semiconstant with G 0 = Gt = 0 and G 8 =A for
all other s E S. The proof of Lemma 3.2 shows that every minimal cochain is
a minimal cocycle, whereas a minimal coboundary u must satisfy um 2 = um 3 ;
accordingly H 2 ( S, G) ~ A and any finite abelian group A i 0 serves. D
Example 3.8. If we delete the relation c6 = cd7 from the presentation of S

in Example 3.7, we obtain a nilmonoid
S ~ ( c, d I c5 d = c3 d 4 = d8 , c7 = c5 d 2 = c 2 d5 = d9 = 0 )
for which M0 consists only of m 1 = 3x + 4y and m 2 = 5x + y, and is a defining
basis of G. Then H 2 ( S, G) = 0 whenever G is thin finite and surjecting.
REFERENCES.
Abdali, S.K. [1993] Parallel computations in *-semirings,

Computational Algebra (Fairfax VA, 1993), 1-16. Marcel Dekker, New York 1994.
Abel, N.H. [ 1826] Untersuchung der Funktionen zweier unabhanging veranderlicher
Grossen x und y, wie f(x,y), welche die Eigenschaft haben, dass f(z,f(x,y)) eine
symmetrische Funktion von z,x und y ist, J. Reine Angew. Math. 1 (1826) 11-15.
Abrgan, I. [ 1984] Prime ideals in groupo ids and multiplicative semigroups of residue
classes (mod m) [Russian], Math. Slovaca 34 (1984) 121-133.
Abellanas, P. [1965] Structure ofcommutative semi-groups [Spanish],
Rev. Mat. Hisp.-Amer. (4) 25 (1965) 3-44. '
Adamek, J. & Koubek, V. [ 1977] On a representation of semigroups by products
of algebras and relations, Colloq. Math. 38 (1977) 7-25.
Adams, W.W. & Loustaunau, P. [1994] An introduction to Grabner bases,
Amer. Math. Soc., Providence RI, 1994.
Ahsan, J., Khan, M.F., Shabir, M., & Takahashi, M. [1991] Characterizations
ofmonoids by p-injective and normalS-systems, Kobe J. Math. 8 (1991) 173-190.
Almqvist, G. [1968] Fractional categories, Arkiv fbr Math. 7 (1968) 449-475.
Altmann, K. [1996] Torsion of differentials on toric varieties,
Semigroup Forum 53 (1996) 89-97.
Anderson, D.D. & Johnson, E.W. [1984] Ideal theory in commutative semigroups,
Anderson, D.D. & Stickles, J. [2000] Commutative rings with finitely generated
multiplicative semigroups, Semigroup Forum 60 (2000) 436-443.
Anderson, D.F. & Scherpenisse, C. [1995] Factorization in K[S],
Commutative ring theory (Fes 1995), 45-56. Marcel Dekker, New York 1997.
Anderson, J.A. [1974] The tensor product of a semigroup with a union of groups,
[ 1978] Tensor product ofcommutative semigroups, Semigroup Forum 16 (1978) 153-164.
[1979A] Characters of commutative semigroups. I,
Math. Sem. Notes Kobe Univ. 7 (1979) 301-308.
[ 1997] The structure of the tensor product of commutative regular semigroups generated
by posets of groups, Math. Japon. 46 (1997) 57-62.
[ 1998] Tensor product ofcommutative regular semigroups generated by posets ofgroups,
Math. Japon. 48 (1998) 181-186.
Anderson, J .A. & Kimura, N. [ 1978] The tensor product of semilattices,
[ 1978] Tensor product of commutative unions of groups,
Proc. Amer. Math. Soc. 73 (1979) 149-153.
[ 1979B] Characters of commutative semigroups, II. Topological semigroups,
[ 1979C] Characters of commutative semigroups, Ill Bicharacters,
Anderson, M. & Feil, T. [ 1988] Lattice-ordered groups,
D. Reidel, Dordrecht-Boston 1988.
379
380 COMMUTATIVE SEMIGROUPS.
Angermiiller, G. [1977] Die Wertehalbgruppe einer ebenen irreduziblen algebroiden

Kurve, Math. Z. 153 (1977) 267-282.
Apery, R. [ 1946] Sur les branches superlineaires des courbes algebriques,
C. R. Acad. Sci. Paris 222 (1946) 1198-1200.
Arendt, B.D. [1975] On semisimple commutative semigroups,
Trans. Amer. Math. Soc. 208 (1975) 341-351.
Arendt, B.D. & Stuth, C.J. [1970A] On the structure of commutative periodic
semigroups, Pacific J. Math. 35 (1970) 1-6.
[19708] On partial homomorphisms ofsemigroups, Pacific J. Math. 35 (1970) 7-9.
[1972] Translations of commutative unique factorization semigroups,
Aubert, K.E. [ 1953] On the ideal theory of commutative semigroups,
Math. Scand. 1 (1953) 39-54.
[1962] Theory of x-ideals, Acta Math. 107 (1962) 1-52.
Aucoin, K.D. [1995] The structure of commutative ideal semigroups,
[1999] The structure of commutative semigroups with the ideal extension property,
Aull, C.E. [1966] Ideals andfilters, Compositio Math. 18 (1966) 79-86.
Backelin, J. [1990] On the number ofsemigroups ofnatural numbers,
Math. Scand. 66 (1990) 197-215.
Baransky, V .A. [ 1975] Structural isomorphism of commutative Archimedean semi-
groups without idempotents [Russian],
Ural. Gos. Univ. Mat. Zap. 9, no.3, 8-13, 133 (1975).
Barr, M. & Beck, J. [1969] Homology and standard constructions,
Seminar on Triples and Categorical Homology Theory, 245-335.
Lecture Notes in Math. no.80, Springer, Berlin-NewYork 1969.
Barucci, V., Dobbs, D.E., & Fontana, M., [1997] Maximality properties in
numerical semigroups and applications to one-dimensional analytically irreducible local
domains, Mem. Amer. Math. Soc. 125 (1997) no. 598.
Barucci, V. & Froberg, R. [1997] One-dimensional almost Gorenstein rings,
J. Algebra 188 (1997) 418-442.
Batbedat, A. [1978A] Le language des schemas pour les demi-groupes commutatifs,
[1978B] Des schemas en demi-groupes commutatifs,
Bazzoni, S. & Salce, L. [1996] Groups in the class semigroups ofvaluation domains,
Israel J. Math. 95 (1996) 135-155.
Beck, J. [1967] Triples, algebras and cohomology, Doct. Diss., Columbia Univ. 1967.
Berg, L. [1985] Distributive Verbtinde als relativ invertierbare Halbgruppen,
Rostock. Math. Kolloq. No.28 (1985) 66-74.
Berthiaume, P. [1964] Rational completions ofmonoids,
Doct. Diss., McGill Univ. 1964.
[1971] Generalized semigroups of quotients, Glasgow Math. J. 12 (1971) 150-161.
REFERENCES. 381
[1973] Une remarque sur les monoU/es de quotients generalises,

Canad. Math. Bull. 16 (1973) 23-25.
Bertin, J. & Carbonne, P. [1975] Sur Ia structure des semi-groupes d'entiers et
application au:x: branches,C. R. Acad. Sci. Paris Ser. A-B 280 (1975) A1745-A1748.
[1977] Semi-groupes d'entiers et application au:x: branches,}. Algebra 49 (1977) 81-95.
Biggs, R., Sasaki M., & Tamura, T. [1965] Non-negative integer valued functions
on commutative groups, I., Proc. Japan Acad. 41 (1965) 564-569.
Birjukov, A.P. [ 1966] Solution of certain algorithmic problems for finitely determined
commutative semigroups [Russian], Sibirsk. Math. Zh. 7 (1966) 523-530.
[1967] Certain algorithmic problems for finitely determined commutative semigroups
[Russian], Sibirsk. Math. Zh. 8 (1967) 525-534.
Birkhoff, G. [ 1944] Subdirect unions in universal algebra,
Bull. Amer. Math. Soc. 50 (1944) 764-768.
Bisgaard, T .M. [ 1998] On perfect semigroups, Acta Math. Hungar. 79 (1998) 269-294.
Blagojevic, D. [1982] More on anti-inverse semigroups,
Pub!. lnst. Math. (Beograd) (N.S.) 31 (45) (1982) 9-13.
Bogacheva, E.V. [1988] Duality of commutative semigroups [Russian],
Homomorphisms [Russian], 3-9. Leningrad. Gos. Ped. Inst., Leningrad 1988.
[1989] Absolute approximation of semigroups by bicharacters [Russian], Semigroup
studies of mappings [Russian], 3-7. Leningrad. Gos. Ped. Inst., Leningrad 1989.
[ 1990A] On the absolute approximability of commutative semigroups by bicharacters
[Russian], Algorithmic problems in the theory of groups and semigroups [Russian], 43-
46. Tul'sk. Gos. Ped. Inst., Tula 1990.
[19900] Weak duality of commutative semigroups [Russian],
Studies ofsemigroups [Russian], 3-9. Leningrad. Gos. Ped. Inst., Leningrad 1990.
Bogomolov, V.S. & Mustafin, T.G. [1988] Description of commutative monoids
over which all polygons are w-stable [Russian], Algebra and Logic 28 (1989) 239-247.
Boguta, B.K. [ 1976] The determination of commutative semigroups by their endomor-
phism semigroups [Russian], Modem algebra no. 4 [Russian] 26-30,
Leningrad. Gos. Ped. Inst., Leningrad 1976.
Borisov, A.A. [ 1982] Lattice determinacy of a class of commutative semigroups without
idempotents [Russian], Izv. Vyss. Ucebn. Zaved. Mat. 1982 no.12, 11-16.
[ 1984] Commutative semigroups that are lattice-isomorphic to commutative Archimedean
semigroups without idempotents [Russian],
Izv. Vyss. Ucebn. Zaved. Mat. 1984 no.9, 14-20.
[1986] Lattice isomorphisms of commutative semigroups without idempotents [Russian],
Izv. Vyss. Ucebn. Zaved. Mat. 1986 no.2, 54-56, 84.
[ 1992] Lattice isomorphisms of commutative holoidal semigroups without idempotents
[Russian], Partitions and holomorphic mappings of semigroups [Russian], 3-14.
"Obrazovanie", St. Petersburg 1992.
Bouvier, A. [1969] Ideau:x: d'un demi-groupe de fractions, Seminaire P. Lefebvre,
1968/69, no.IO. Dep. Math., Univ. Claude Bernard, Villeurbanne 1969.
[ 1970] Sur Ia factorisation dans les demi-groupes commutatift,
Publ. Dep. Math. (Lyon) 7 (1970) 187-216.
Bouvier, A. & Faisant, A. [1970] Proprietes des demi-groupes de fractions. IIL
Seminaire P. Lefebvre, 1969/70, no.11. Dep. Math., Univ. Lyon I, Villeurbanne 1970.
382 COMMUTATIVE 8EMIGROUPS.
Bouvier, C. & Gonzales-Springberg, G. [1992] G-desingularisations de varietes

toriques, C. R. Acad. Sci. Paris Ser. I Math. 315 (1992) 817-820.
[ 1995] Systeme generateur minimal, diviseurs essentiels et G-desingularisations de varie-
tes toriques, T6koku Math. J. (2) 47 (1995) 125-149.
Brameret, M.P. [1962] Surles demi-groupes d'homomorphismes,
C. R. Acad. Sci. Paris 254 (1962) 2115-2116.
Brauer, A. [ 1942] On a problem of partitions, Amer. J. Math. 64 (1942) 299-312.
Bredihin, B.M. [ 1953] On characters of numerical semigroups with a sufficiently
sparse base [Russian], Dokl. Akad. Nauk. SSSR 90 (1953) 707-710.
[1954] On sum functions of characters of numerical semigroups [Russian],
Dokl. Akad. Nauk. SSSR 94 (1954) 609-612.
[1958A] Free numerical semigroups with power densities [Russian],
[19588] Free numerical semigroups with power densities [Russian],
Mat. Sb. (N.S.) 46 (88) (1958) 143-158.
[ 1960] Elementary solutions of inverse problems on bases offree semigroups [Russian],
Mat. Sb. (N.S.) 50 (92) (1960) 221-232.
Bredihin, D.A. [1976] Bands of congruences ofsemigroups [Russian],
Modern algebra no. 4 [Russian] 31-47, Leningrad. Gos. Ped. Inst., Leningrad 1976.
Bresinsky, H. [ 1972] Semigroups corresponding to algebroid branches in the plane,
[ 1975] Symmetric semigroups of integers generated by 4 elements,
Manuscripta Math. 17 (1975) 205-219.
[ 1979] Monomial Gorenstein curves in A 4 as set-theoretic complete intersections,
[1984] Monomial Buchsbaum ideals in P', Manuscripta Math. 47 (1984) 105-132.
[ 1988] Binomial generating sets for monomial curves, with applications in A 4 ,
Rend. Sem. Mat. Univ. Politec. Torino 46 (1988) 353-370.
Briales, E., Campillo, A., Marijuan, C., & Pison, P. [1998] Minimal systems
of generators for ideals of semigroups, J. Pure Appl. Algebra 124 (1998) 7-30.
Brillouet-Belluot, N. [ 1996] Multiplicative symmetry and reated functional equations,
Aequationes Math. 33 (I 996) 187-191.
Bnzmdsted, A. [1983] An introduction to convex polytopes, Springer-Verlag 1983.
Brown, D.R. & LaTorre, J.G. [1966] A characterization of uniquely divisible semi-
groups, Pacific J. Math. 18 (1966) 57-60.
Brown, W.C. & Curtis, F. [1991] Numerical semigroups of maximal and almost
maxima/length, Semigroup Forum 42 (1991) 219-235.
Bruns, W. & Gubeladze, J. [1996] Rectangular simplicial semigroups, Commu-
tative algebra, algebraic geometry, and computational methods (Hanoi 1996), 201-213.
Springer, Singapore 1999.
[1999] A regularity criterion for semigroup rings, Georgian Math. J. 6 (1999) 259-262.
Buchweitz, R.O. [ 1980] On Zariski 's criterion for equisingularity and nonsmoothable
monomial curves, Preprint 113, Univ. of Hannover, 1980.
Budach, L. [1964] Struktur Noethersche kommutativer Halbgruppen,
Monatsb. Deutsch. Akad. Wiss. Berlin 6 (1964) 85-88.
REFERENCES. 383
Bulman-Fleming, S. & McDowell, K. [ 1980] Flatness and projectivity in commu-

tative regular semigroups, Semigroup Forum 19 (1980) 247-259.
Burgess, D.A. [1968] On a problem of Baayen and Kruyswijk,
Proc. Edinburgh Math. Soc. (2) 15 (1968/69) 145-149.
(1969] A problem on semi-groups, Studia Sci. Math. Hungar. 4 (1969) 9-11.
Butz bach, P. [ 1972] Les 1-convexes du monoi:de commutatif Nn,
J. Comput. System Sci. 6 (1972) 283-289.
Buzashi, K. & Vishnyakova, N.I. [1986] Characterization of locally cyclic nilsemi-
groups [Russian], Publ. Math. Debrecen 33 (1986) 99-106.
Byeveld, S. & Easdown, D. [1992] The minima/faithful degree of a finite commuta-
tive inverse semigroup, Computational algebra and number theory (Sidney 1992), 19-27.
Kluwer Acad. Publ., Dordrecht 1995.
Byrd, R.D., Lloyd, J.T., Pedersen, F.D., & Stepp, J.W. [1977] Automor-
phisms of the semigroup offinite complexes of a periodic locally cyclic group,
Pacific J. Math. 72 (1977) 27-39.
Caillot, J .F. & Wehrung, F. [2000] Finitely presented, coherent, and ultrasimplicial
ordered abelian groups, Semigroup Forum 61 (2000) 116-137.
Campillo, A. & Marijuan, C. [ 1991] Higher order relations for a numerical semi-
group, Sem. Theor. Nombres Bordeaux (2) 3 (1991) 249-260.
Campillo, A. & Pison, P. [ 1993] Generators of a monomial curve and graphs for
the associated semigroup, Bull. Soc. Math. Belg. Ser. A 45 (1993) 45-58.
Carbonne, P. [1983] Cohomologie des semi-groupes d'entiers,
J. Algebra 84 (1983) 1-13.
Carlisle, W .H. [ 1971] Residual finiteness offinitely generated commutative semigroups,
Pacific J. Math. 36 (1971) 99-101.
Cavaliere, M.P. & Niesi, G. [ 1983] On monomial curves and Cohen-Macaulay type,
[1984] Complete intersection projective monomial curves [Italian],
Boll. Un. Mat. Ital. D (6) 3 (1984) 189-200.
Chacron, J. [ 1981] Toutes les congruences du demigroupe additif des entiers naturels,
Proc. 106th National Congress of Learned Societies (Perpignan 1981 ), 63-70,
Bib. Nat. Paris 1982.
Chandran, V .R. [ 1987] On a problem of Zelinka, I, II,
Czechoslovak Math. J. 37 (112) (1987) 124, 125-127.
Chapman, S., Krause, U., & Oeljeklaus, E. [2001] On Diophantine monoids and
their class groups, to appear in Pacific J. Math.
Chen, Q.H. [1988] On well-semimodules,
Fujian Shifan Daxue Xiebao Ziran Kexue Ban 4 (1988) no.3, 12-17.
Cho, J .R. [ 1990] Representations of certain medial algebras,
J. Korean Math. Soc. 27 (1990) 69-76.
Chrislock, J.L. [1969] On medial semigroups, J. Algebra 12 (1969) 1-9.
Ch'iian, J.C. & Lee, W.S. [1982] Permuation of N that preserves semigroups,
Math. Japon. 27 (1982) 597-598.
Cibul'skite, D. [ 1970] The distribution of generating elements in free numerical semi-
groups. /,II, lll [Russian], Litovsk. Mat. Sb. 10 (1970) 397-415, 593-610, 835-844.
Ciric, M. [2002] Decompositions of semigroups, Kluwer A cad. Pub I., Dordrecht 2002.
Cistov, A.L. [ 1978] The number of generators of the semigroup of stable equivalence
classes of algebraic tori [Russian], Dokl. Akad. Nauk. SSSR 242 (1978) 1027-1029.
Clark, W.E., Holland, W.C., & Szekely, G.J. [1998] Decompositions in discrete
semigroups, Studia Sci. Math. Hungar. 34 (1998) 15-23.
Clarke, J .L. [ 1980] A study of commutative cancellative idempotent-free semigroups,
Doct. Diss., Univ. California Davis, 1980.
[1984] Rees quotients of N -semigroups, Semigroup Forum 29 (1984) 259-270.
Clifford, A.H. [1938] Arithmetic and ideal theory of commutative semigroups,
Annals of Math. 39 (1938) 594-610.
[1941] Semigroups admitting relative inverses, Annals of Math. 42 (1941) 1037-1049.
[ 1950] Extensions of semigroups, Trans. Amer. Math. Soc. 68 (1950) 165-173.
[ 1954] Bands of semigroups, Proc. Amer. Math. Soc. 5 (1954) 499-504.
Clifford, A.H., & Preston, G.B. [ 1961] The algebraic theory of semigroups, val.!,
[ 1967] The algebraic theory of semigroups, val. II,
Comfort, W. W. [ 1959] The isolated points in the dual of a commutative semigroup,
Comfort, W.W. & Hill, P. [1966] On extending nonvanishing semicharacters,
Coppens, M. [1988] Weierstrass points with two prescribed nongaps,
Pacific J. Math. 131 (1988) 71-104.
[ 1995] The existence of base point free linear systems on smooth plane curves,
J. Algebraic Geom. 4 (1995) 1-15.
Croisot, R. [ 1953] Demi-groupes inversif.s et demi-groupes reunions de demi-groupes
simples, Ann. Sci. Ecole Norm. Sup. (3) 70 (1953) 361-379.
[ 1957] Equivalences principales bilateres dans les demi-groupes,
J. Math. Pures Appl. (9) 36 (1957) 373-417.
Cudakov, N.G. & Bredihin, B.M. [1956] Application of Parseval's equality for
the estimation of sum functions of characters of numerical semigroups [Russian],
Ukrain. Mat. Zh. 8 (1956) 347-360.
Cudakov, N.G. & Pavlyucuk, A.K. [1951] On summationfunctions of characters
of numerical semigroups with a finite basis [Russian],
Trudy Mat. Inst. Steklov., v.38, 366-381. Izdat. Akad. Nauk. SSSR, Moscow 1951.
Curtis, F. [ 1990] On formulas for the Frobenius number of a numerical semigroup,
Math. Scand. 67 (1990) 190-192.
D' Anna, M. [ 1997] Type sequences of numerical semigroups,
Semi group Forum 56 (1998) 1-31.
D' Anna, M. & Delfino, D. [1997] Integrally closed ideals and type sequences in
one-dimensional local rings, Rocky Mountain J. Math. 27 (1997) no.4, 1065-1073.
Darnel, M.R. [1995] Theory oflattice-ordred groups, Marcel Dekker, New York 1995.
de Bruijn, N.G. [1956] On number systems, Nieuw Arch. Wisk. (3) 4 (1956) 15-17.
REFERENCES. 385
Deddens, J .A. [ 1979] A combinatorial identity involving relatively prime integers,

J. Combin. Theory Ser. A 26 (1979) 189-192.
de Luca, A. & Restivo, A. [1984] A finiteness condition for finitely generated
semigroups, Semigroup Forum 28 (1984) 123-134.
Delgado, F. [ 1988] Gorenstein curves and symmetry of the semigroup of values,
[ 1990] The symmetry of the Weierstrass generalized semigroups and affine em beddings,
Proc. Amer. Math. Soc. 108 (1990) no.3, 627-631.
Delorme, C. [ 1976] Sous-monoUles d'intersection complete de N,
Ann. Sci. Ecole Norm. Sup. (4) 9 (1976) 145-154.
Dickinson, R.P., Jr. [1972] Noncancellative congruences on N-semigroups,
Dickson, L.E. [1913] Finiteness of the odd perfect and primitive abundant numbers
with n distinct prime factors, Amer. J. Math. 35 (1913) 413-422.
Dobbertin, H. [ 1982] On Vaught's criterion for isomorphisms of countable Boolean
algebras, Algebra Universalis 15 (1982) 95-114.
[ 1983] Refinement mono ids, Vaught monoids, and Boolean algebras,
Math. Ann. 265 (1983) 473-487.
[1984] Primely generated regular refinement monoids, J. Algebra 91 (1984) 166-175.
Dorofeeva, M.P. [1972] Herediary and semi-hereditary monoids,
Drbohlav, K. [ 1963] Concerning congruence relations on commutative semigroups,
Comment. Math. Univ. Carolinae 4 (1963) 11-13.
[ 1964] Zur Theorie der Kongruenzrelationen auf kommutativen Halbgruppen,
Math. Nachr. 26 (1963/64) 233-245.
Dress, A. & Grabmeier, J. [ 1991] The interpolation problem for k-sparse polyno-
mials and character sums, Adv. in Appl. Math. 12 (1991) 57-75.
Dubreil, P. [ 1941] Contribution a Ia theorie des demi-groupes,
Mem. Acad. Sci. lnst. France (2) 63, no.3 (1941).
Duske, J. [1976] Zur Theorie kommutativer Automaten, Acta Cybemet. 3 (1976) 15-23.
Dyadchenko, G.G. & Shokuev, V.N. [1985] Idempotents of a multiplicative semi-
group ofthe ring Zn [Russian], Algebraic systems with one action and relation [Russian],
18-20. Leningrad. Gos. Ped. Inst., Leningrad 1985.
Easdown, D. & Gould, V. [1996] Commutative orders,
Proc. Roy. Soc. Edinburgh Sect. A, 126 (1996) no.6, 1201-1216.
Ebanks, B.R. [1979] Branching measures of information on strings,
Houston J. Math. 8 (1982) 323-332.
Eberhart, C. [ 1982] On abelian semigroups whose congruences form a simple lattice,
Houston J. Math. 8 (1982) 323-332.
Eigenthaler, G., Nobau.~r, W. & Wiesenbauer, J. [1987] Ober Halbgruppen
vertauschbarer Polynome, Osterreich. Akad. Wiss. Math.-Natur. Kl. Sitzungsber. II, 196
(1987) 227-247.
Eigenthaler, G. & Woracek, H. [1997] A remark on permutable polynomials, Con-
tributions to general algebra, 10 (Klagenfurt 1997), 139-142. Heyn, Klagenfurt 1998.
Eilenberg, S. & Schiitzenberger, M.P. [1969] Rational sets in commutative mon-

oids, J. Algebra 13 (1969) 173-191.
Ego, M. [ 1961] Structure des demi-groupes dont le treillis des sous-demi-groupes est
distributif, C. R. Acad. Sci. Paris 252 (1961) 2490-2492.
[1962A] Structure des demi-groupes dont le treillis des sous-demi-groupes est modulaire
ou semi-modulaire, C. R. Acad. Sci. Paris 254 (1962) 1723-1725.
[1962B] Structure des demi-groupes dont le treillis des sous-demi-groupes satisfait aux
conditions c 2 , C2, C1, C1, a Ia semi-modularite affaiblie, ou a Ia modularite affaiblie,
C. R. Acad. Sci. Paris 255 (1962) 1840-1842.
[ 1962C] Structure des demi-groupes dont le treillis des sous-demi-groupes satisfait aux
conditions C2, C2, C1, C1, a Ia semi-modularite affaiblie, ou a Ia modularite affaiblie,
C. R. A cad. Sci. Paris 255 (1962) 2699-2701.
[1963] Structure des demi-groupes dont le treillis des sous-demi-groupes satisfait a cer-
taines conditions, Bull. Soc. Math. France 91 (1963) 137-201.
Emelicev, V .A. [ 1962] The solution of certain algorithmic problems for commutative
semigroups [Russian], Dokl. Akad. Nauk. SSSR 144 (1962) 261-263.
[ 1966] Regularity of a finitely determined commutative semigroup [Russian],
Vesci Akad. Navuk. BSSR Ser. Fiz.-Mat. Navuk 1966, no.2, 23-30.
Ermolina, N.G. [1988] Approximation of commutative semigroups by real and finite
bicharacters [Russian], Homomorphisms [Russian], 27-43.
Leningrad. Gos. Ped. lnst., Leningrad 1988.
Esterle, J. [1997] Embedding semigroups into commutative radical Banach algebras,
Banach algebras '97 (Blaubeuren 1997), 153-168. de Gruyter, Berlin 1998.
Etayo, J.J. [1965] A category ofsemigroups, Fourth Annual Meeting of Spanish Math-
ematicians [Spanish], 79-88. Univ. Salamanca, Salamanca 1965.
Evseev, A.E. [ 1965] Semigroups with ordinally decomposable semi-lattices of subsemi-
groups [Russian], lzv. Vyss. Ucebn. Zaved. Mat. 1965 no.6, 74-84.
[ 1991] Commutative semigroups, every partition ofwhich contains a subsemigroup [Rus-
sian], Semigroups and their homomorphisms [Russian], 8-11.
Ross. Gos. Ped. Univ., Leningrad 1991.
Ewald, G. & Wessels, U. [1991] On the ampleness of invertible sheaves in complete
projective toric varieties, Results in Math. 15 (1991) 275-278.
Farrell, E.J. [ 1992] An ·introduction to F-graphs, a graph-theoretic representation of
natural numbers, Intemat. J. Math. Math. Sci. 15 (1992) 313-317.
Findlay, G.D. & Lambek, J. [ 1958] A generalized ring of quotients. I, IL
Canad. Math. Bull. 1 (1958) 77-85.
Fischer, K.G., Morris, W., & Shapiro, J. [1997] Affine semigroup rings that are
complete intersections, Proc. Amer. Math. Soc. 125 (1997) no.ll, 3137-3145.
Fischer, K.G. & Shapiro, J. [1996] Mixed matrices and binomial ideals,
J. Pure Appl. Algebra 113 (1996) 39-54.
Forsythe, G.E. [1955] SWAC computes 126 distinct semigroups of order 4,
Fountain, J .B. & Petrich, M. [ 1986] Completely 0-simple semigroups of quotients,
J. Algebra I 01 ( 1986) 365-402.
Fraenkel, A.S., Porta, H., & Stolarsky, K.B. [1989] Some arithmetical semi-
groups, Analytic number theory (Allerton park 1989), 255-264.
REFERENCES. 387
Progr. Math. 85, Birkhauser, Boston 1990.

Fraser, G.A. & Albert, J.P. [1984] Associativity ofthe tensor product ofsemilattices,
Proc. Edinburgh Math. Soc. (2) 27 (1984) 337-340.
Freyd, P. [ 1968] Redei 's finiteness theorem for commutative semigroups,
Proc. Amer. Math. Soc. 19 (1968) 1003.
Frobenius, F.G. [1895] Ober endliche Gruppen,
Sitzungsber. Preuss. Akad. Wiss. Berlin (1895) 163-194.
Froberg, R. [ 1994] The Frobenius number of some semigroups,
Comm. Algebra 22 (1994) 6021-6024.
Froberg, R., Gottlieb, C., & Haggkvist, R. [1987] On numerical semigroups,
Fuchs, L. [ 1970] Infinite abelian groups, Academic Press, New York, 1970.
Fulp, R.O. [ 1966] On homomorphisms of commutative inverse semigroups,
Czechoslovak Math. J. 16 (91) (1966) 72-75.
[ 1967] On extending semigroup characters,
[ 1970] Tensor and torsion products of semigroups,
Pacific J. Math. 32 (1970) 685-696. correction, ibid. 39 (1971) 827.
Fulp, R.O. & Hill, P. [1965] Reflexive semigroups,
Galanova, J. [ 1982] The tensor product of a semigroup and semigroups of idempotents
[Russian], Math. Slovaca 32 (1982) 55-60.
[ 1987] Some properties of tensor products of semigroups,
Demonstratio Math. 20 (1987) 57-65.
Galbiati, J .L. & Veronesi, M.L. [ 1983] Permutable commutative semigroups [Ital-
ian], Istit. Lombardo Accad. Sci. Lett. Rend. A 117 (1983) 43-56.
Gale, D. & Tamura, T. [ 1976] Maximal cancellative subsemigroups and sub-archime-
deanness, Algebraic theory ofsemigroups (Szeged, 1976), 147-154.
Colloq. Math. Soc. Janos Bo1yai, North-Holland, Amsterdam, 1979.
[ 1980] A theorem on commutative cancel/ative Archimedean idempotent-free semigroups,
Simon Stevin 54 (1980) 233-240.
Garcia-Sanchez, P.A. & Rosales, J.C. [1999] Numerical semigroups generated by
intervals, Pacific J. Math. 191 (1999) 75-83.
Gastinger, W., Kunz, E., & Waldi, R. [1991] Relation matrices of monomial
curves, An. Univ. Bucuresti Mat. 40 (1991) 41-53.
Geroldinger, A. [ 1996] On the structure and arithmetic offinitely primary mono ids,
[ 1997] Chains offactorizations in orders of global fields,
Colloq. Math. 72 (1997) 83-102.
Gilmer, R. [ 1984] Commutative semigroup rings Univ. of Chicago Press, 1984.
Gluskin, L.M. [19710] Dense extensions [Russian],
Dokl. Akad. Nauk SSSR 196 (1971) 270-273.
[1971S] Separative semigroups [Russian],
Gluskin, L.M. (cont'd) [1973] Dense extensions ofsemigroups [Russian], Mini-Confe-

rence on Semigroup Theory (Szeged, 1972) 1-6. Joszef Attila Univ., Szeged 1973.
[1976D] Dense extensions of commutative semigroups [Russian],
Algebraic theory of semigroups (Szeged, 1976), 155-171. Colloq. Math. Soc. Janos
Bolyai, North-Holland, Amsterdam, 1979. Correction, Semigroups (Szeged, 1981), 489-
490. Colloq. Math. Soc. Janos Bolyai, North-Holland, Amsterdam, 1985.
[1976E] Extensions of commutative semigroups [Russian],
[1979] Extensions of commutative semigroups [Russian],
Izv. Vyss. Ucebn. Zaved. Mat. 1979 no.ll, 3-12. correction, ibid. 1982 no.7, 88.
[19831] Ideals of commutative semigroups [Russian],
Izv. Vyss. U cebn. Zaved. Mat. 1983 no.3, 31-3 9.
[1983S] Subreductive commutative semigroups [Russian],
Associative actions [Russian] 18-25, Leningrad. Gos. Ped. lnst., Leningrad 1983.
[1985] Normal extensions of commutative semigroups [Russian],
Golubov, E.G. [1969] Semigroups with certain finitely separable subsemigroups [Rus-
sian], Ural. Gos. Univ. Mat. Zap. 7 (1969/70) tetrad' 1, 35-51.
[ 1970A] Conditions for the finite approximability of naturally totally ordered commuta-
tive semigroups [Russian], lzv. Vyss. Ucebn. Zaved. Mat. 1970 no.2, 10-18.
[1970S) Finite_ separability in semigroups [Russian],
Sibirsk. Mat. Z. 11 (1970) 1247-1263.
Gonchigdorzh, R. & Mal'tsev, Yu.N. [ 1982) On a variety of universal algebras
for which the groupoid of subvarieties is a commutative semigroup [Russian],
Sibirsk. Mat. Z. 23 (1982) no.2, 31-38, 215.
Goodearl, K.R. [1986] Partially ordered abelian groups with interpolation,
Gould, M. & Iskra, J .A. [ 1984G] Globally determined classes of semigroups,
[1984R] Embedding in globals of groups and semilattices,
Gould, M., Iskra, J.A. & Palfy, P.P. [1986] Embedding in globals offinite semi-
lattices, Czechoslovak Math. J. 36 (Ill) (1986) 87-92.
Greco, S. [1990) Orders ofrationalfunctions on algebraic curves,
Proc. 21st. Annual Iranian Math. Conf. (Isfahan 1990), 159-165. Univ. Isfahan, Isfahan.
Greco, S. & Raciti, G. [1991] The Liiroth semigroup of plane algebraic curves,
Pacific J. Math. 151 (1991) 43-56.
Green, J.A. [1951] On the structure ofsemigroups Annals of Math. 54 (1951) 163-172.
Grillet, P.A. [1969F] On free commutative semigroups,
J. Natur. Sci. and Math. 9 (1969) 71-78.
[1969T] The tensor product of commutative semigroups,
[1970C] Cancellative commutative semigroups, Semigroup Forum 1 (1970) 249-253.
[ 1970E] The free envelope of a finitely generated commutative semigroup,
REFERENCES. 389
[ 19701] Interpolation properties and tensor product of semigroups,

[ 1970T] The tensor product of cancellative semigroups,
[ 1972] Subdirect decompositions offinite commutative semigroups,
[ 1974] Left coset extensions, Semigroup Forum 7 (1974) 200-263.
[1975C] A completion theorem for finitely generated commutative semigroups,
J. Algebra 34 (1975) 25-53.
[ 1975P Primary semigroups, Michigan Math. J. 22 (1975) 321-326.
[ 1976] Directed colimits offree commutative semigroups,
J. Pure Appl. Algebra 9 (1976/77) 73-87.
[ 1977] On subdirectly irreducible commutative semigroups,
Pacific J. Math. 69 (1977) 55-71.
[1991C] Commutative semigroup cohomology, Semigroup Forum 43 (1991) 247-252.
[ 1991 N] A construction offinite commutative nilsemigroups,
Corum. Algebra 19 (1991) no.ll, 3145-3172. errata, ibid. 20 (1992) no.9, 2789.
[1991T] Nilsemigroups on trees, Semigroup Forum 43 (1991) 187-201.
[ 1992] The commutative cohomology of nilsemigroups,
[ 1993] Some congruences on free commutative semigroups,
Acta Sci. Math. (Szeged) 58 (1993) 3-23.
[1993R] A short proof of Redei's Theorem, Semigroup Forum 46 (1993) 126-127.
[ 1995] Semigroups: an introduction to the structure theory,
Marcel Dekker, New York 1995.
[ 1995C] Commutative semigroup cohomology,
Corum. Algebra 23 (1995) no.IO, 3573-3587.
[ 1995F] The commutative cohomology offinite semigroups,
[ 1995N] The number of commutative semigroups of order n,
[1995P] Partially free commutative semigroups,
Acta Sci. Math. (Szeged) 61 ( 1995) 155-169.
[ 1995S] Stratified semigroups, Semigroup Forum 50 (1995) 25-36.
[ 1996C] Finite congruences on free commutative semigroups,
[ 1996N] Computing finite commutative semigroups,
[ 1997C] Cocycles in commutative semigroup cohomology,
Corum. Algebra 25 (I 997) no.11, 3427-3462.
[1997Z] Commutative semigroups with zero cohomology,
[1999] Algebra, Wiley, New York 1999.
Grillet, P.A. (cont'd) [2000T] The commutative cohomology of two-generator semi-

groups, Semigroup Forum 60 (2000) 4-15.
[200 I C] Congruences on free commutative semigroups,
submitted to Acta Sci. Math. (Szeged)
[200 IF] Fully invariant congruences on free commutative semigroups,
Acta Sci. Math. (Szeged) 67 (200 I) 173-20 I.
[200 I H] 4-Cocyc/es in commutative semigroup cohomology,
submitted to Comm. Algebra
[200 1N] A construction of commutative nilsemigroups,
to appear in Period. Math. Hungar.
[2001S] Subcomplete commutative nilsemigroups, submitted to Acta Sci. Math. (Szeged)
[2001Z] Commutative nilsemigroups with zero cohomology,
Grillet, P.A. & Novikov, B.V. [2002] Cohomology ofsemigroups,
to appear in Semigroup Forum.
Grossman, P.A. [ 1982] A note on local polynomial functions on commutative semi-
groups, J. Austral. Math. Soc. Ser. A 33 (1982) 50-53.
Groza, G h. [ 1995] Semigroups of integral functions in valued fields, Algebra, logic &
discrete mathematics (Nis 1995). Filomat No.9, part 3 (1995) 675-687.
Gubeladze, J. [ 1998] The isomorphism problem for commutative monoid rings,
Gurican, J. [ 1984] The semigroup of general circulant matrices,
Acta Math. Univ. Comenian. 44/45 (1984) 13-21.
Halezov, V .A. [ 1966] Algorithmic solution of certain problems in the theory offinitely
defined commutative semigroups with cancellation [Russian],
Sibirsk. Math. Zh. 7 (1966) 419-436.
Hall, R.E. [1971] Commutative cancellative semigroups with two generators,
Czechoslovak Math. J. 21 (1971) 449-452.
[1972] The translational hull of an N-semigroup, Pacific J. Math. 41 (1972) 379-389.
Halter-Koch, F. [ 1993] The integral closure of a finitely generated monoid and the
Frobenius problem in higher dimensions, Semigroups. Algebraic Theory and applications
to formal languages and codes, 86-93, World Scientific, 1993.
Hamilton, H. [1975] Permutability of congruences on commutative semigroups,
Semigroup Forum 10 (1975/76) 55-66.
[ 1978L] Modularity of the congruence lattice of a commutative cancellative semigroup,
Pacific J. Math. 79 (1978) 469-485.
[1978N] Congruences on N-semigroups, Pacific J. Math. 75 (1978) 423-448.
[ 1982] Modularity and distributivity of the congruence lattice of a commutative separ-
ative semigroup, Math. Japon. 27 (1982) 581-589.
Hamilton, H. & Nordahl, T. [1978] Semigroups whose congruence lattice is Boo-
lean, Pacific J. Math. 77 (1978) 131-143.
Hamilton, H., Nordahl, T. & Tamura, T. [1975] Commutative cancellative semi-
groups without idempotents, Pacific J. Math. 61 (1975) 441-456.
Hancock, R. [ 1960] On complete semimodules,
REFERENCES. 391
Hansen, R. T. [ 1969] Complementing pairs of subsets of the plane,

Duke Math. J. 36 (1969) 441-449.
Hartshorne, R. [ 1992] Generalized divisors on Gorenstein rings,
K-Theory 8 (1994) 287-339.
Head, T. [1967A] Homomorphisms of commutative semigroups as tensor maps,
[ 1967B] The tensor product of a group with a semigroup,
[ 1967C] The tensor product of semigroups with minimal ideals,
[ 1967D] Functor properties of semigroup tensor products,
[ 1968] Problems and conjectures concerning tensor products ofcommutative semigroups,
[ 1969] Tensor products and maximal subgroups of commutative semigroups,
Pub!. Math. Debrecen 16 (1969) 145-147.
[1971G] A semigroup-theoretic view of projective class groups,
Proc. Japan Acad. 47 (1971) suppl. I, 831-836.
[1971H] Commutative semigroups having greatest regular images,
[ 1979] Quotient monoids and greatest commutative monoid images of several types,
Head, T. & Anderson, J .A. [ 1980] On injective and flat commutative regular semi-
Head, T. & Kuroki, N. [1974] Greatest regular images oftensor products of com-
mutative semigroups, Kodai Math. Sem. Rep. 26 (1974/75) 132-136.
Herzinger, K. [1999] The number ofgenerators for an ideal and its dual in a numerical
semigroup, Comm. Algebra 27 (1999) no.l 0, 4673-4690.
Herzog, J. [ 1970] Generators and relations of abelian semigroups and semigroup rings,
Heuer, C.V. [1966] On the equivalence of cancellative extensions of commutative
cancel/ative semigroups by groups, J. Austral. Math. Soc. 12 (1971) 187-192.
Heuer, C. V. & Miller, D. W. [ 1966] An extension problem for cance/lative semi-
groups, Trans. Amer. Math. Soc. 122 (1966) 499-515.
Hewitt, E. & Zuckerman, H.S. [1955] Finite dimensional convolution algebras,
Acta Math. 93 (1955) 67-119.
[1956] The l 1 -algebra of a commutative semigroup,
Trans. Amer. Math. Soc. 83 ( 1956) 70-97.
[ 1960] The multiplicative semigroup of integers modulo m,
Pacific J. Math. 10 (1960) 1291-1308.
Higgins, J.C. [1966] Finitely generated commutative Archimedean semigroups gener-
ated N-semigroups, Doct. Diss., Univ. California Davis, 1966.
[ 19691] A faithful canonical representation for finitely generated N-semigroups,
[1969S] Representing N-semigroups, Bull. Austral. Math. Soc. I (1969) 115-125.
Higgins, J.C. (cont'd) [1972] Subsemigroups of the additive positive integers,

Fibonacci Quart. 10 (1972) 225-230.
[ 1983] The varieties of commutative semigroups for which epis are onto,
Proc. Roy. Soc. Edinburgh Sect. A, 94 (1983) 1-7.
[ 1984] Semigroup epimorphisms, Proc. 1984 Marquette conf. on semigroups, 51-64.
Marquette Univ., Milwaukee Wis. 1985.
[ 1985] Epimorphisms, dominions, and semigroups,
Algebra Universalis 21 (1985) 225-233.
[1990] A short proof of Isbell's zigzag theorem, Pacific J. Math. 144 (1990) 47-50.
Higgins, J.C. & Tamura, T. [1973] Finitely generated N-semigroup and quotient
group, Proc. Japan Acad. 49 (1973) 323-327.
Hildebrant, J .A. [ 1984] The translational degree of a semigroup,
Hill, P. [ 1966] A solution to the nonvanishing character extension problem,
[1967] Characters of commutative semigroups, J. Algebra 5 (1967) 16-24.
Hindman, N. & Woan, W.J. [1992] Central sets in commutative semigroups and
partition regularity of systems of linear equations, Mathematika 40 (1993) 169-186.
Hinkle, C.V., Jr. [1972] Semigroups ofright quotients of a semigroup which is a
semilattice of groups, Semigroup Forum 5 (1972/73) 167-173.
[1974] Semigroups of right quotients of a semigroup which is a semilattice of groups,
J. Algebra 31(1974) 276-286.
Hippisley-Cox, S.D. [ 1992] Finite point compactifications of (N,+),
Hirabuki, S. & Matsuda, R. [1999] Note on the Krull-Akizuki theorem for semi-
groups, Bull. Yamagata Univ. Natur. Sci. 14 (1999) 155-159.
Hmelnitsky, I.L. [ 1985] On semigroups with the idealizer condition,
Hoa, L.T. [1991] A note on projective monomial surfaces,
Math. Nachr. 154 (1991) 183-188.
Hochster, M. [ 1972] Rings of invariants of tori, Cohen-Macaulay rings generated by
monomials, and polytopes, Ann. of Math. 96 (1972) 318-337.
Hoehnke, H.-J. [ 1965] Eine Charakterisierung des Radikals kommutativer Halbgrup-
pen, Arch. Math. 16 (1965) 412-413.
Holme, A. [1966] A general theory of structure spaces, Fund. Math. 68 (1966) 335-347.
Homma, M. [ 1996] The Weierstrass semigroup of a pair of points on a curve,
Arch. Math. (Basel) 67 (1996) 337-348.
Hora, R.B. & Kimura, N. [1972] Commutative semigroups of type (m,n,p,q). l
Monovariable cases, Proc. Japan Acad. 48 (1972) 555-560.
Hora, R.B. [1973] Onfactorizable simple and commutative semigroups,
Howie, J .M. [ 1968] Commutative semigroup amalgams,
J. Austral. Math. Soc. 8 (1968) 609-630.
[ 1976] An introduction to semigroup theory, Academic Press, London 1976.
REFERENCES. 393
[ 1995] Fundamentals of semigroup theory,

Clarendon Press, Oxford Univ. Press, New York 1995.
Howie, J .M. & Isbell, J .R. [ 1967] Epimorphisms and dominions, !1,
J. Algebra 6 (1967) 7-21.
Hsieh, S.C. [ 1982] A note on a saturated commutative semigroup,
Soochow J. Math. 8 (1982) 55-57.
Huang, I. C. [ 1995] A residue map and its applications to some one-dimensional rings,
Proc. Amer. Math. Soc. 123 (1995) no.8, 2369-2372.
Huang, W.C. [1996] Matrices which belong to an idempotent in a sandwich semigroup
of circulant Boolean matrices, Linear Algebra Appl. 249 (1996) 157-167.
Hule, H. [ 1976] Uber die Eindeutigkeit der Losungen algebraischer Gleichungssysteme,
J. Reine Angew. Math. 282 (1976) 157-161.
Hurwitz, A. [1893] Uber algebraischer Gebilde mit eindeutigen Transformationen in
sich, Math. Ann. 41 (1893) 403-442.
Huynh, D.T. [1986] A superexponentiallower bound for Grabner bases and Church-
Rosser commutative Thue systems, Inform. and Control 68 (1986) 196-206.
lgnat'eva, I. V. [ 1996] SH-approximatoin of semigroups by finite characters [Russian],
Modem algebra, No.I [Russian], 25-30. Rostov. Gos. Ped. Univ., Rostov-on-Don 1996.
Imaoka, T. [ 1976] Free products with amalgamation ofcommutative inverse semigroups,
J. Austral. Math. Soc. Ser. A 22 (1976) 246-251.
Inasaridze, H.N. [1964] Extensions ofsemigroups [Russian],
Soobsc. Akad. Nauk. Gruzin. SSR 33 (1964) 263-269.
[ 1965] Extensions of regular semigroups [Russian],
Soobsc. Akad. Nauk. Gruzin. SSR 39 (1965) 3-10.
[ 1967] Extensions of commutative inverse semigroups [Russian],
Sakharth, SSR Mecn. Akad. Moambe 46 (1967) 11-18.
Isbell, J .R. [ 1959] On the multiplicative semigroup of a commutative ring,
[ 1965] Epimorphisms and dominions,
Proc. Conf. Categorical Algebra (La Jolla 1965), 232-246, Springer, New York 1966.
[2000] On multiplicative generators of commutative rings,
Semigroup Forum 61 (2000) 315.
Iseki, K. [ 1957] Contributions to the theory of semigroups, VI,
Proc. Japan Acad. 33 (1957) 29-30.
lwani~, A. & Plonka, J. [1975] Linear independence in commutative semigroups,
Mat. Casopis Sloven. Akad. Vied 25 (1975) 333-338.
Jain, R.K. [1982] On Dedekind monoids, Semigroup Forum 24 (1982) 91-92.
[ 1983] Semigroups with primary ideals of prime power,
Acta Math. Hungar. 42 (1983) 165-169.
[1988] On Prilfer monoids, Ranchi Univ. Math. J. 19 (1988) 49-52.
[1988] Prufer monoids and Dedekind monoids, Ranchi Univ. Math. J. 20 (1989) 15-18.
Jarek, P. [1964] Commutative regular semigroups, Colloq. Math. 12 (1964) 195-208.
Jensen, B.A. & Miller, D.W. [1968] Commutative semigroups which are almost
finite, Pacific J. Math. 27 (1968) 533-538.
Jezek, J. & Kepka, T. [1975] Semigroup representations of commutative idempotent

abelian groupoids, Comment. Math. Univ. Carolinae 16 (1975) 487-500.
John, C.C., Jr. [1973] The construction offinite commutative semigroups,
Doct. Diss., Kansas State Univ. 1973.
[1974] A theorem onfree envelopes, Semigroup Forum 9 (1974175) 74-75.
Johnson, C.S., Jr. & McMorris, F.R. [1977] Commutative non-singular semi-
groups, Canad. Math. Bull. 20 (1977) 41-67.
Johnson, E. W. [ 1986] Primary factorization in semigroups,
Czechoslovak Math. J. 36 (Ill) (1986) 180-184.
Johnson, R.E. [1951] The extended centralizer of a ring over a module,
Jones, P. & Ligh, S. [1980] The multiplicative semigroup of integers modulo n,
Math. Japon. 25 (1980) 15-17.
Juan, L. [1993] Some results about symmetric semigroups,
Comm. Algebra 21 (1993) no.10, 3637-3645.
Jurgensen, H. [ 1990] Annotated tables of semigroups of orders 2 to 6,
Report 231, Dept. of Computer Sci., Univ. Western Ontario 1990.
Jurgensen, H., Migliorini, F., & Szep, J. [1991] Semigroups,
Akad. Kiado, Budapest 1991.
Jurgensen, H., Pallas, I., & Wick, P. [1976] Halbgruppenprogramme,
Bericht 1176, lnst. Informatik Prakt. Math., Univ. Kiel 1976.
Jurgensen, H. & Wick, P. [1973] Bestimmung der Unterhabgruppenverbande fur
zwei klassen endlicher Halbgruppen,
Computing (Arch. Elektron. Rechnen) 11 (1973) 337-351.
[1977] Die Halbgruppen der Ordnungen ~7, Semigroup Forum 14 (1977) 69-79.
Kaarli, K. & Marki, L. [1992] Affine complete commutative inverse semigroups,
Semigroups (Luino 1992), 144-151. World Sci. Publishing, River Edge NJ, 1993.
[ 1995] Affine complete commutative inverse semigroups, Algebra Colloq. 2 (1995) 51-78.
Kaarli, K., Marki, L., & Schmidt, E.T. [1985] Affine complete semilattices,
Monatsh. Math. 99 (1985) 297-309.
Kacman, S.I. [also spelled Katsman] [1979] Commutative semigroups with self-dual
lattice ofsubsemigroups, Semigroup Forum 18 (1979) 119-161.
Kainrath, F. & Lettl, G. [2000] Geometric notes on monoids,
Kanemitsu, M. & Bansho, S. [1999] On primary ideals of valuation semigroups,
Far East J. Math. Sci. I (1999) 27-32.
Kanemitsu, M., Okabe, A. & Matsuda, R. [1999] On polynomial extensions of
g-monoids, Algebras Groups Geom. 16 (1999) 269-276.
Kapur, D. & Narendran, P. [1983] The Knuth-Bendix completion procedure and
Thue systems, Fundations of software technology and theoretical computer science (Ban-
galore, 1983) 363-385, Tata Inst. Fund. Res., Bombay 1983.
[ 1985] The Knuth-Bendix completion procedure and Thue systems,
SIAM J. Comput. 14 (1985) no.4, 1052-1072.
Karasek, J. [ 1985] A remark concerning x -systems,
Arch. Math. (Bmo) 21 (1985) 177-180.
REFERENCES. 395
Karhumiiki, J. & Plandowski, W. [ 1996] On the size of independent systems of

equations in semigroups, Theoret. Comput. Sci. 168 (1996) 105-119.
Kato, T. [1979] On criteria ofg-hyperellipticity, Kodai Math. J. 2 (1979) 275-285.
Kaufman, R.P. [1966A] Maximal semicharacters,
[ 1966B] Interpolation of additive functionals, Studia Math. 27 (1966) 269-272.
[1967] Positive semicharacters, J. London Math. Soc. 42 (1967) 264-266.
Keimel, K. [ 1971] Baer extensions of rings and Stone extensions of semigroups,
Khan, N .M. [ 1982] Epimorphisms, dominions, and varieties of semigroups,
Khovansky, A.G. [1992] The Newton polytope, the Hilbert polynomial and sums of
finite sets, Functional Anal. Appl. 26 (1992) 276-281.
[ 1995] Sums of finite sets, orbits of commutative semigroups and Hilbert functions,
Functional Anal. Appl. 29 (1995) 102-112.
Kil'p, M. [1973] Commutative monoids all ofwhose principal ideals are projective,
Kim, A.C. & Neumann, B.H. [1991] Laws in a class ofgroupoids,
Discrete Math. 92 (1991) 145-158.
Kim, S.J. [1994] On the index of the Weierstrass semigroup of a pair of points on a
curve, Arch. Math. (Basel) 62 (1994) 73-82.
[ 1996] Semigroups which are not Weierstrass semigroups,
Bull. Korean Math. Soc. 33 (1996) 187-191.
Kimura, N. [1954] Maximal subgroups of a semigroup,
Kodai Math. Sem. Rep. 6 (1954) 85-88.
Kimura, N. & Tsai, Y.S. [1972] On power cancellative archimedean semigroups,
Proc. Japan Acad. 48 (1972) 553-554.
Kisielewicz, A. [1991] All pseudovarieties of commutative semigroups,
Semigroups with applications (Oberwolfach, 1991 ), 78-89; World Scientific, 1992.
[ 1994] Varieties of commutative semigroups, Trans. Amer. Math. Soc. 342 (1994) 275-306.
Kisielewicz, A. & Newrly, N. [1993] Polynomial densityofcommutativesemigroups,
Bull. Austral. Math. Soc. 48 (1993) 151-162.
Kist, J. [1963] Minimal prime ideals in commutative semigroups,
Proc. London Math. Soc. (3) 13 (1963) 31-50.
Kitsuneda, Y., Matsuda, R., & Sugatani, T. [1999] Note on /-dimensional
integrally closed Mori semigroups, Tamkang. J. Math. 30 (1999) 295-297.
Knebl, H. [1984] Ebene algebraische Kurven vom Typ p,q,
Knebusch, M. [1969] Grothendieck- und Wittringe von nichtausgearteten symmetri-
schen Bilinearformen, S.-B. Heidelberger Akad. Wiss. Math.-Natur. Kl. 1969/70, 93-157.
Kleitman, D.J., Rothschild, B.R., & Spencer, J.H. [1976] The number of
semigroups of order n, Proc. Amer. Math. Soc. 55 (1976) 227-232.
Kobayashi, Y. [ 1974] Irreducible bases of N-semigroups,
J. Math. Tokushima Univ. 8 (1974) 9-12.
Kobayashi, Y. (cont' d) [ 1977C] Conditions for commutative semigroups to have non-

trivial homomorphisms into nonnegative (positive) reals,
[ 1977E] On extending functionals on commutative semigroups,
SOrikaisekikenkyOsho K6ky0roku No.292 ( 1977) 90-97.
[ 1978] On a problem of commutative nil semigroups, Proceedings of the 2nd Seminar
on Semigroups, Tokyo Gakugei Univ., Tokyo, 1978, 25-28.
[1980] An extension theorem offunctionals on commutative semigroups,
Studia Math. 69 (1980) 87-90.
[1982] Cancellative and Archimedean ideal-extensions by commutative nil semigroups,
Kobayashi, Y. & Tamura, T. [1974] Quasi-order preserving homomorphisms of
commutative semigroups into the non-negative reals, Math. Japon. 22 (1977) 267-285.
Kogalovsky, S.R. [1961] The multiplicative semigroups of rings [Russian],
Kolibiarov_a, B. [1958] Ober kommutative periodische Halbgruppen,
Mat.-Fyz. Casopis Sloven. Akad. Vied 8 (1958) 127-135.
Komeda, J. [ 1982] On the existence of Weierstrass points with a certain semigroup
generated by 4 elements, Tsukuba J. Math. 6 (1982) 237-270.
[ 1983] On Weierstrass points whose first nongaps are four,
[ 1991] On primitive Schubert indices of genus g and weight g-1,
J. Math. Soc. Japan 43 (1991) 437-445.
[ 1992] On the existence of Weierstrass points whose first nongaps are five,
[ 1994] On the existence of Weierstrass gap sequences on curves of genus ~8,
[ 1997] Remarks on non- Weierstrass numerical semigroups,
Algebras and combinatorics (Hong Kong 1997), 313-319, Springer, Singapore 1999.
[1998] Non-Weierstrass numerical semigroups, Semigroup Forum 57 (1998) 157-185.
Kominek, Z. [ 1990] On approximately subadditive functions,
Demonstratio Math. 23 (1990) 155-160.
Kopytov, V.M. & Medvedev, N.Ya. [1994] The theory of lattice-ordered groups,
Kluwer Acad. Pub!., Dordrecht 1994.
[1996] Right ordered groups, Consultants Bureau, New York 1996.
Kowol, G. & Mitsch, H. [I 976] Polynomial functions over commutative semi-groups,
Kozlov, K.P. [1976] Semigroups with a minimal number of rigid cells in a variety of
commutative semigroups [Russian], Modem algebra no. 4 [Russian] 91-96,
Kranz, P. [1974] Sandwich and extension theorems on semigroups and lattices,
Comment. Math. Prace Mat. 18 (1974/75) 193-200.
[ 1979] Extension of additive jimctionals and semicharacters on commutative semigroups,
REFERENCES. 397
Krause, U. [ 1989] On monoids offinite real character,

Krivenko, V.M. [1985] Semigroups with normal monogenic subsemigroups [Russian],
Algebraic systems with one action and relation [Russian], 54-57.
Krob, D. [1987] MonoU/es et semi-anneaux complets,
Kruming, P.D. [ 1982] Homological and cohomological characteristics of generalized
periodic semigroup [Russian], Izv. Vyss. Ucebn. Zaved. Mat. 1982 no.5, 30-34.
Kublanovsky, S.I. [1983] Endomorphisms offinite semigroups [Russian], Associative
actions [Russian] 65-75, Leningrad. Gos. Ped. lnst., Leningrad 1983.
Kublanovsky, S.I. & Lesohin, M.M. [1976] Separability of semigroups with re-
spect to predicates [Russian], Algebraic Theory of Semigroups (Szeged, 1976) 309-333.
Colloq. Math. Soc. Janos Bolyai, North-Holland, Amsterdam, 1979.
Kulkarni, D.M. [1990] On the semigroup of an ordinary multiple point,
Kunz, E. [1966] Gruppenringe und Di.fferentiale, Math. Ann. 163 (1966) 346-350.
[ 1973] The value-semigroup of a one-dimensional Gorenstein ring,
[ 1987] Ober die Klassijikation numerischer Halbgrupper,
Univ. Regensburg, Fachbereit Math., Regensburg 1987.
Kunze, M. [1978] Die Halbgruppe der Grossenklassen eines geordneten lokalen Rings,
Semigroups (Oberwolfach, 1978), 19-25, Lecture Notes in Math. no.855,
Springer, Berlin 1981.
Kupcov, A.I. [also spelled Kuptsov] [1979E] A property of the semigroup of en-
domorphisms of commutative regular semigroups [Russian], Semigroup varieties and
semigroups ofendomorphisms [Russian] 80, Leningrad. Gos. Ped.Inst., Leningrad 1979.
[ 1979H] Some properties of the semigroup of homomorphisms of commutative regular
semigroups [Russian], Semigroup varieties and semigroups of endomorphisms [Russian]
67-79, Leningrad. Gos. Ped. lnst., Leningrad 1979.
[ 1980] Periodicity of a semigroup ofhomomorphisms of commutative regular semigroups
[Russian], Modem algebra [Russian] 89-98, Leningrad. Gos. Ped. lnst., Leningrad 1980.
[ 1981 B] The non-triviality of the semigroup of bilinear mappings of semigroups [Rus-
sian], Materials of VII Region conf. math. mech. [Russian], Tomsk 1981,35-36.
[1981H] Some properties of the functors hA and he in the category of commutative
regular semigroups [Russian], Modem algebra [Russian] 17-22,
Kupcov, A.I. & Lesohin, M.M. [ 1986] Quasiduality ofsemigroups [Russian], Semi-
groups with additional structures [Russian], 38-49. Leningrad. Univ., Leningrad 1986.
Kuroki, N. [1971] Note on congruences on the tensor product ofarchimedean commu-
tative semigroups, Comment. Math. Univ. St. Paul. 20 (1971 /72) 93-96.
[1973] Certain fundamental congruences on the tensor product of commutative inverse
semigroups, Czechoslovak Math. J. 23 (98) (1973) 252-256.
Lal, H. [1972P] A note on regular semigroups. II, Math. Education 6 (1972) A128.
[1972R] On Fuchs radical of a semigroup, Aligarh Bull. Math. 2 (1972) 25-32.
Lal, H. (coot' d) [1975] Commutative semi-primary semigroups,

Lal, R.N. [1967] Pseudogroup, Proc. Nat. Acad. Sci. India Sect. A 37 (1967) 32-36.
Lallement, G. [1971] On a theorem of Malcev,
Laplaza, M.L. [ 1964] Imbedding of cancel/alive semigroups in difference semigroups,
Rev. Mat. Hisp.-Amer. (4) 24 (1964) 103-112.
[1974] A note on commutative semigroups offinite length
[Spanish], Rev. Mat. Hisp.-Amer. (4) 34 (1974) 109-111.
Lau, G. [1979] Finite abelian semigroups represented into the power set offinite groups,
Lawson, J.D. [ 1996] The earliest semigroup paper?, Semigroup Forum 52 (1996) 55-60.
Leech, J. [ 1975] :H -coextensions of monoids,
Mem. Amer. Math. Soc. no.l57 (1975) 1-66.
Lenzi, D. [1980] The paracancel/ation law in commutative semigroups,
Acta Sci. Math. Acad. Sci. Hungar. 36 (1980) 65-69.
Lepetit, C. [ 1973] Decomposition en somme directe de monofdes fibres abeliens,
C. R. Acad. Sci. Paris Ser. A-8, 277 (1973) A877-A880.
Lesieur, L. [1955] Surles ideaux irreductibles d'un demi-groupe,
Rend. Sem. Mat. Univ. Padova 24 (1955) 29-36.
Lesohin, M.M. [also spelled Lesokhin] [ 1958] Certain properties of generalized
characters of semigroups, [Russian],
Leningrad. Gos. Ped. lnst. Ucen. Zap. 183 (1958) 277-286.
[ 1961] On the regularity of systems with external multiplication and the simplicity of
their components, [Russian], Leningrad. Gos. Ped. lost. Ucen. Zap. 218 (1961) 23-37.
[1963A] Hom_omorphisms into regular semigroups [Russian],
Sibirsk Mat. Z. 4 (1963) 1431-1432.
[1963M] Regularity of systems with exterior multiplication with regular first component
[Russian], Izv. Vyss. Ucebn. Zaved. Mat. 1963 no.4, 89-91.
[1963S] On the completeness of systems with exterior multiplication [Russian],
[1964G] Galois connections in systems with exterior multiplication [Russian],
[ 1964M] On the semigroup of a multiplicative commutative regular semigroup [Russian],
[ 1965] On the duality of complex characters and semi-characters of commutative semi-
groups [Russian], Mat. Sb. (N.S.) 66 (108) (1965) 378-383.
[1966A] Approximabi/ity of semigroups and separability of subsemigroups [Russian],
Interuniv. Sci. Sympos. General Algebra [Russian] 52-62, Tartu Gos. Univ., Tartu 1966.
[ 1966D] Duality of commutative semigroups and of sets with binary relations [Russian],
Herzen Lectures, XIX, Math. [Russian] 6-7, Leningrad. Gos. Ped. In st., Leningrad 1966.
[ 1966H] Additive semigroups ofhomomorphisms ofregular semigroups [Russian], Trudy
Naucn. Ob'ed. Prepodav. Fiz.-Mat. Fak. Ped. lnst. Dal'n. Vostok. 7 (1966) 54-67.
[1967A] The approximation ofsemigroups [Russian],
Leningrad. Gos. Ped. Inst. Ucen. Zap. 328 (1967) 147-171.
REFERENCES. 399
[1967D] Distributive operations on semigroups [Russian],

[ 1967E] Distributive operations on semigroups, II [Russian],
[1967S] Systems with exterior multiplication which are complete and quasi-complete
with respect to the first and second component [Russian],
[ 1968A] Finitary approximability of semigroups and finite separability of subsemigroups
[Russian], Leningrad. Gos. Ped. lnst. Ucen. Zap. 387 (1968) 134-158.
[ 1968C] Characters of commutative semigroups [Russian],
Symposium on Semigroup Theory and Appl. [Russian], Smolensk 1968, 17-27.
[1968D] Absolute duality of commutative semigroups [Russian],
[ 1969] Characters and bicharacters of semigroups [Russian], First All-Union Sympo-
sium on the Theory of Semigroups [Russian], 43-52. Ural. Gos. Univ., Sverdlovsk 1969.
[1970] Characters of commutative semigroups, I [Russian],
[1971A] Approximation ofsemigroups with respect to predicates [Russian],
[ 1971 B] Bicharacters of commutative semigroups [Russian],
Leningrad. Gos. Ped. Inst. Ucen. Zap. 464 (1971) no. 2, 4
[1971C] Characters of commutative semigroups, lJ [Russian],
[1971H] Homomorphisms of systems with exterior multiplication [Russian],
[ 1971 R] Reflexive semigroups [Russian], Theory of semigroups and its applications, No.2
[Russian], 36-40, 152. Izdat. Saratov. Univ., Saratov 1971.
[ 1972] Examination of semigroups by characters [Russian], Mini-Conference on Semi-
group Theory (Szeged, 1972) 38-40. J6szef Attila Univ., Szeged 1973.
[1974A] The finite approximation of semigroups [Russian], Theory of semigroups and
its applications, No.3 [Russian], 39-46, 152. Izdat. Saratov. Univ., Saratov 1974.
[ 19741] The finitary I-approximability of commutative semigroups [Russian], Modem
Algebra, No.I [Russian], 124-129. Leningrad. Gos. Ped. Inst., Leningrad 1974.
[1974~] The separability ofsubsemigroups by characters [Russian],
Mat. Casopis Sloven. Akad. Vied 24 (1974) 129-138.
[1975] The finitary approximability of semigroups with respect to predicates [Russian],
Modem Algebra, No.3 [Russian], 87-98. Leningrad. Gos. Ped. lnst., Leningrad 1975.
[ 1976] Finite approximability of semigroups of bilinear mappings [Russian], Modem
Algebra, No.5 [Russian], 88-92. Leningrad. Gos. Ped. Inst. im Gercena, Leningrad
1976.
[ 1977] The local finiteness of semigroups of bilinear complex functionals [Russian],
Modem Algebra, No.6 [Russian], 132-135. Leningrad. Gos. Ped. lnst., Leningrad 1977.
[ 1983] Separability ofsubsemigroups by complex characters [Russian], Algebraic actions
and orderings [Russian], 79-81. Leningrad. Gos. Ped. lnst., Leningrad 1983.
[ 1992] Duality ofregular commutative semigroups [Russian], Partitions and holomorphic
mappings of semigroups [Russian], 64-75. "Obrazovanie", St. Petersburg 1992.
Lesohin, M.M. & Golubov, E.G. [ 1966] Finitary approximability of commutative

semigroups [Russian], Ural. Gos. Univ. Mat. Zap. 5 (1966) tetrad' 3, 82-90.
Lesohin, M.M. & Kupcov, A.I. [1980] Duality and quasiduality of semigroups
[Russian], Modem algebra [Russian] 99-102, Leningrad. Gos. Ped. lnst., Leningrad 1980.
Lesohin, M.M. & Popyrin, A. V. [ 1987] Bicharacters of semigroups,
Lettl, G. [1988] Subsemigroups offinitely generated groups with divisor-theory,
Monatsh. Math. 106 (1988) 205-210.
Levin, R.G. [1968] On commutative nonpotent archimedean semigroups,
Pacific J. Math. 27 (1968) 365-371.
Levin, R.G. & Tamura, T. [1970] Notes on commutative power joined semigroups,
Pacific J. Math. 35 (1970) 673-679.
Li, S.Zh. [1986] On the tensor products in some subcategories of the category of
semigroups, J. Math. Res. Exposition (English ed.) 1986, no.2, 15-18.
[1987] The tensor product of cancellation semigroups [Chinese],
Yangzhou Shiyuan Xuebao Ziran Kexue Ban 1987, no.4, 33-37.
Li, S.Zh. & Liu, Zh.H. [1992] A note on injections of abelian groups [Chinese],
Qufi Shifan Daxue Xuebao Ziran Kexue Ban 18 (1992) no.4, 28-30.
Li, S.Zh. & Zhang, Y.F. [1993] Tensor products of separable semigroups [Chinese],
J. Math. (Wuhan) 13 (1993) 283-288.
Liberman, A.A. [ 1980] Lattice isomorphisms of cyclic semigroups [Russian],
Ural. Gos. Univ. Mat. Zap. 12, no.2, 100-111, ii (1980).
Libih, A.L. [1974] Local automorphisms of commutative monomorphic inverse semi-
groups [Russian], Studies in algebra, No.4 [Russian], 56-69, 125.
Izdat. Saratov. Univ., Saratov 1974.
Lidl, R. & Miiller, W.B. [1986] On commutative semigroups of polynomials with
respect to composition, Monatsh. Math. 102 (1986) 139-153.
Lipkovski, A. [ 1988] Newton polyhedra and irreducibility,
Math. Z. 199 (1988) 119-127.
Ljubenova, L. & Cvetkova, G. [ 1967] The subsemigroups of a certain free abelian
semigroup offinite rank [Bulgarian],
Godisnik Vyss. Tehn. Ucebn. Zaved. Mat. 4 ( no.1, 13-26.
Lomadze, D.D. [1992A] Approximation ofsemigroups with respect to Green equiva-
lences [Russian], Soobsc. Akad. Nauk. Gruzii 145 (1992) 265-268.
[1992B] SH-approximation of semigroups by finite commutative regular semigroups
[Russian], Partitions and holomorphic mappings of semigroups [Russian], 105-112.
"Obrazovanie", St. Petersburg 1992.
Lord, J .M. [ 1978] The structure of commutative Archimedean semigroups without idem-
potent, Doct. Diss., Univ. California Davis, 1978.
[1979] Isomorphism criterion and structure group description for iJt-semigroups,
Proc. Japan Acad. Ser. A Math. Sci. 55 (1979) 252-254.
Lorenzen, P. [1939] Abstrakte Begriindung der multiplikativen Idealtheorie,
Math. Z. 45 (1939) 533-553.
[1949] Uber halbgeordnete Gruppen, Math. Z. 52 (1949) 483-526.
[1952] Teilbarkeitstheorie in Bereichen, Math. Z. 55 (1952) 269-275.
REFERENCES. 401
Luedemann, J .K. [ 1976] The generalized translational hull of a semigroup,

Pacific J. Math. 67 (1976) 489-507.
Lugowski, H. [ 1966] Die Charackterisierung gewisser geordneter Halbmoduln mit Hilfe
der Erweiterungstheorie, Publ. Math. Debrecen 13 (1966) 237-248.
Lyapin, E.S. [1950S] Simple commutative associative systems [Russian],
Izv. Akad. Nauk SSSR Ser. Mat. 14 (1950) 275-282.
[1950R] Semisimple commutative associative systems [Russian],
Izv. Akad. Nauk SSSR Ser. Mat. 14 (1950) 367-380.
Lyapin, E.S. & Evseev, A.E. [1997) The theory of partial algebraic operations,
Kluwer Acad. Publ., Dordrecht 1997.
MacKenzie, R.E. [1954] Commutative semigroups, Duke Math. J. 21 (1954) 471-477.
Maclachlan, C. [1971] Weierstrass points on compact Riemann surfaces,
J. London Math. Soc. (2) 3 (1971) 722-724.
MacLane, S. [1963] Homology, Springer-Verlag, New York, 1963.
[ 1971] Categories for the working mathematician,
Springer-Verlag, New York, 1971. 2nd. Ed. 1998.
Malcev, A.I. [also spelled Mal'tsev] [1958] On homomorphisms onto finite groups
[Russian], Uch. Zap. Ivanov. Gos. Ped. lnst. 18 (1958) 49-60.
Mannepalli, V .L. [ 1976A] Multiplication semigroups,
Semigroup Forum II (1975/76) 310-327.
[19768) Multiplication semigroups. !1, Semigroup Forum 12 (1976) 145-164.
Mannepalli, V.L. & Satyanarayana, M. [1974] Monoids of the Dedekind type,
Semigroup Forum 9 (1974/75) 19-27.
Manukjanc, M.G. [ 1974) Construction of the semigroup of endomorphisms of certain
commutative regular semigroups, [Russian],
[ 1976A] The structure qf a semigroup of multiplications ofa class ofcommutative regular
semigroups [Russian], Theory of functions. Differential Equations and their applications,
no. I [Russian], 3-7, 220. Kalmyck. Gos. Univ., Elista, 1976.
[ 19768] Semigroups of multiplications of certain commutative semigroups [Russian],
Theory of functions. Differential Equations and their applications, no.! [Russian], 56-
72, 216. Kalmyck. Gos. Univ., Elista, 1976.
Marcus, D.A. [1974) Direct decompositions in free commutative monoids,
J. Combinatorial Theory Ser. A, 16 (1974) 286-312.
Markov, S.M. [ 1995) On the foundations of interval arithmetic, Scientific computing
and validated numerics (Wuppertal 1995), 307-313, Akademie Verlag, Berlin 1996.
Matsuda, R. [ 1997) The Krull-Akizuki theorem for semigroups,
Math. J. Ibaraki Univ. 29 (1997) 55-56.
[ 1998A] Some theorems for semigroups, Math. J. lbaraki Univ. 30 (1998) 1-7.
[ 19988] Note on overrings without a specified element,
Math. J. lbaraki Univ. 30 (1998) 9-14.
[1999] Note on Macaulay semigroups, Tsukuba J. Math. 23 (1999) 189-199.
Maury, G. [ 1958] Une caracterisation des demi-groupes noetheriens integralement clos,
C. R. Acad. Sci. Paris 247 (1958) 254-255.
Maury, G. (cont'd) [1959] Une caracterisation des demi-groupes noetheriens integrale-

ment clos, C. R. Acad. Sci. Paris 248 (1959) 3260-3261.
McAlister, D.B. [1968] Characters on commutative semigroups,
Quart. J. Math. Oxford Ser.(2) 19 (1968) 141-157.
McAlister, D.B. & O'Carroll, L. [1970] Finitely generated commutative semi-
groups, Glasgow Math. J. 11 (1970) 134-151.
[1970] Maximal homomorphic images of commutative semigroups,
Glasgow Math. J. 12 (1971) 12-17.
McMorris, F .R. [ 1971] The quotient semigroup of a semigroup that is a semilattice of
groups, Glasgow Math. J. 12 (1971) 18-23.
[1972] The maximal quotient semigroup, Semigroup Forum 4 (1972) 360-364.
McMorris, F .R. & Satyanarayana, M. [ 1972] Categorical semigroups,
McNeil, P.E. [1971] Group extensions ofnull semigroups,
Duke Math. J. 38 (1971) 491-497.
[ 1972] Finite commutative subdirectly irreducible semigroups,
Megyesi, L. & Pollak, G. [1968] Ober die Struktur der Hauptidealhalbgruppen, /,
[1977] Ober die Struktur der Hauptidealhalbgruppen, II,
Migliorini, F. & Szep, J. [ 1979] Equivalences, congruences, and decompositions in
semigroups, Riv. Mat. Univ. Parma (4) 5 (1979) 745-752.
[ 1981] On r -decompositions of semigroups, Recent developments in [ ... ] semigroups
(Oberwolfach, 1981), 385-403, Lecture Notes in Math. no.998, Springer, Berlin 1983.
Mitrovic, Z.M. [1976] Interval semigroups,
Univ. Beograd. Publ. Elektrotehn. Fak. Ser. Mat. Fiz. no. 544-576 (1976) 127-143.
[ 1977] Contribution to the theory of interval arithmetic, and its applications [Serbo-
Croatian], Univ. Beograd. Publ. Elektrotehn. Fak. Ser. Mat. Fiz. no. 601 (1977).
Mitsch, H. [ 1983] Semigroups and their lattice of congruences,
[1986] A natural order for semigroups, Proc. Amer. Math. Soc. 97 (1986) 384-388.
[1994] Semigroups and their natural order, Math. Slovaca 44 (1994) 445-462.
Mogiljanskaja, E.M. [ 1971] The semigroup of the subsemigroups of a commutative
semigroup [Russian], Leningrad. Gos. Ped. Inst. Ucen. Zap. 404 (1971) 250-256.
[ 1972] Definability of certain holoid semigroups by means of the semigroups of all their
subsets and subsemigroups [Russian],
Moh, T. T. & Heinzer, W. [ 1982] On the Liiroth semigroup and Weierstrass canonical
divisors, J. Algebra 77 (1982) 62-73.
Monzo, R.A.R. [1973] On categorical semigroups,
Moore, E.H. [1902] A definition of abstract groups,
Morel, A.C. [1979] Locally cyclic semigroups, Algebra Universalis 9 (1979) 391-394.
REFERENCES. 403
Moszner, Z. [ 1979] Sur I 'inegalite de translation,

Demonstratio Math. 11 (1978) 1095-1100 (1979).
Motzkin, T .S. & Selfridge, J .L. [1956] Semigroups of order five,
Bull. Amer. Math. Soc. 62 (1956) 14.
Murata, K. [1980] A note on arithmetics in semigroups,
Murphy, G.J. [1996] C*-algebras generated by commuting isometries,
Rocky Mountain J. Math. 26 (1996) 237-267.
Murty, A.S.R. [1983] Noetherian S-spaces and primary decompositions,
Mushtaq, Q. & Yusuf, S.M. [ 1987] On LA-semigroup defined by a commutative
regular semigroup, J. Natur. Sci. and Math. 27 (1987) 11-15.
[1988] On LA-semigroup defined by a commutative inverse semigroup,
Mat. Vesnik 40 (1988) 59-62.
Nagy, A. [200 1] Special classes of semigroups, Kluwer A cad. Pub!., Dordrecht 2001.
Nakajima, S. [1977] The endomorphism semigroup of a semigroup and its application,
SOrikaisekikenkyOsho KIJkyOroku No.292 ( 1977) 1-8.
Narendran, P., O'Dunlaing, C., & Rolletschek, H. [1985] Complexityofcertain
decision problems about congruential languages,
J. Comput. System Sci. 30 (1985) 343-358.
Narendran, P. & O'Dunlaing, C. [1989] Cancellativity infinitely presented semi-
groups, J. Symbolic Comput. 7 (1989) 457-472.
Nasirov, S.N. [1977] Representations ofsemigroups in Boolean algebras, Tashkent.
Gos. Univ. Sb. Naucn. Trudov No.623 Mat. Analiz i Geometrija (1980) 56-61, 94.
Nathanson, M.B. [2000] Growth of sumsets in abelian semigroups,
Neeb, K.-H. [1992] Toric varieties and algebraic monoids,
Sem. Sophus Lie 2 (1992) 159-187.
Nguyen, H.K. [1981] Schreiersche Erweiterungen von Halbgruppen,
Seminarberichte no. 34, Humboldt Univ., Sekt. Math., Berlin 1981.
Niculescu, C.P. [1995] A combinatorial property of abelian semigroups,
Rev. Roumaine Math. Pures Appl. 40 (1995) 669-671.
[ 1996] Birkhoff recurrence theorem and combinatorial properties of abelian semigroups,
Rev. Roumaine Math. Pures Appl. 41 (1996) 675-686.
Nikolaev, A.M. [1971] A tensor product ofsemigroups [Russian],
Leningrad. Gos. Ped. Inst. Ucen. Zap. 464 (1971) no.2, 122-127.
Niven, I. [ 1971] A characterization of complementing sets ofpairs of integers,
Duke Math. J. 38 (1971) 193-203.
No bauer, W. [ 1986] Some remarks on permutable chains ofpolynomials, Contributions
to general algebra 5 (Salzburg 1986), 247-256. Holder-Pichler-Tempsky, Vienna 1987.
f 1987] Vertauschbare Polynome: an den Grenzen der Koejjizientenvergleichsmethode,
Osterreich. Akad. Wiss. Math.-Natur. Kl. Sitzungsber. II, 196 (1987) 403-417.
Nordahl, T. [1973] Commutative semigroups whose proper subsemigroups are power
joined, Semigroup Forum 6 (1973) 35-41.
[1977] Abelian groups and N-semigroups, II, Proc. Japan Acad. 53 (1977) 8-10.
Novikov, B.V. [1990] Commutative semigroups with cancellation of dimension I [Rus-

sian], Mat. Zametki 48 (1990) 148-149.
Oda, T. [ 1988] Convex bodies and algebraic varieties,
Springer-Verlag, Berlin-New York, 1988.
Oknhiski, J. [1984] Semigroup algebras, Marcel Dekker, New York, 1991.
[ 1986] Commutative monoid rings with Krull dimension, Semigroups, theory and appli-
cations (Oberwolfach, 1986), 251-259, Lecture Notes in Math. no.1320,
Springer, Berlin-New York 1988.
Ovsyannikov, A.Ya. (also spelled Ovsjannikov] [1977A] Structural isomorphisms
of commutative semigroups with one defining relation [Russian],
[ 1977B] Lattice isomorphisms of commutative semigroups with a single defining relation
[Russian], Ural. Gos. Univ. Mat. Zap. 10, no.3 lssled. po Sovremen. Algebre, 138-172,
218 (1977).
[1981] Structural isomorphisms of commutative semigroups with one defining relation.
II [Russian], Ural. Gos. Univ. Mat. Zap. 12, no.3, 90-107, iii (1981).
Painter, P. [1966] Isomorphisms between certain subsemigroups of the positive integers
under multiplication, Amer. Math. Monthly 73 (1966) 62-64.
[ 1967] Isomorphisms between semigroups of integers,
Proc. Amer. Math. Soc. 18 ( 1967) 1100-1101.
Pales, z. [1989] A Stone-type theorem for abelian semigroups,
Arch. Math. (Basel) 52 (1989) 265-268.
Parizek, f!. [1957] Note on structure of multiplicative semigroup of residue classes,
Mat.-Fyz. Casopis Sloven. Akad. Vied 7 (1957) 183-185.
Parizek, B. & Schwarz, S. f1961] Semicharacters of the multiplicative semigroup of
integers modulo m, Mat.-Fyz. Casopis Sloven. Akad. Vied 11 (1961) 63-74.
Parker, E.T. [1954] On multiplicative semigroups of residue classes,
Patil, D.P. [ 1993] Minimal sets ofgenerators for the relation ideals of certain monomial
curves, Manuscripta Math. 80 (1993) 239-248.
Patil, D.P. & Singh, B. [ 1990] Generators for the derivation modules and the relation
ideals of certain curves, Manuscripta Math. 68 (1990) 327-335.
Paxia, G., Raciti, G., & Ragusa, A. [1992] On the Lilroth semigroup of curves
lying on a smooth quadric, Manuscripta Math. 75 (1992) 225-246.
Pedersen, F.D. & Sizer, W.S. [1978] Certain semigroups ofcomplexes of an abelian
group, Proceedings of the Conference on Semigroups in Honor of Alfred H. Clifford (New
Orleans, 1978), 215-221; Tulane Univ., New Orleans 1979.
Perel, W.M. [1957] Principal representations in commutative semigroups,
Peric, V. [1966] Zu denfastinvertierbaren Idea/en in kommutativen Ringen und Halb-
gruppen, Glasnik Mat. Ser. III 1 (21) (1966) 139-146.
Perrot, J .-F. [ 1965] Sur quelques families de parties des monoUJes abeliens libres,
C. R. Acad. Sci. Paris 261 (1965) 3008-3011.
Petrich, M. [ 1962] Semicharacters of the Cartesian product of two semigroups,
Pacific J. Math. 12 (1962) 679-683.
REFERENCES. 405
[ 1963C] Some theorems concerning the dual of a commutative semigroup,

Duke Math. J. 30 (1963) 25-32.
[1963F] The maximal semilattice decomposition of a semigroup,
[1964] On the structure of a class of commutative semigroups,
[1973] Introduction to semigroups, Charles E. Merrill, Columbus OH, 1973.
[1973E] L 'enveloppe de translations d'un demi-treillis de groupes,
Canad. J. Math. 25 (1973) 164-177.
[1977] Commutative semigroups, Rend. Sem. Mat. Fis. Milano 47 (1977) 61-66.
Pfister, G. & Steenbrink, J.H.M. [1992] Reduced Hilbert schemes for irreducible
curve singularities, J. Pure Appl. Algebra 77 (1992) 103-116.
Philip, J.M. [1974] A proof of Isbell's zigzag theorem, J. Algebra 32 (1974) 328-331.
Plappert, P. [1995] A sandwich theorem for monotone additive functions,
Plemmons, R.J. [ 1966] Cayley tables for all semigroups of order N~6,
Auburn Univ. 1966.
PondiHicek, B~ [ 1966] On the characters of semigroups whose idempotents form a
chain [Czech], Casopis Pest. Mat. 91 (1966) 4-7.
[ 1983] Atomicity of tolerance lattices of commutative semigroups,
Czechoslovak Math. J. 33 (I 08) (1983) 485-498.
[1985A] Modularity and distributivity of tolerance lattices of commutative separative
[ 1985B] Modularity and distributivity of tolerance lattices of commutative separative
lJ988A] Note on the congruence lattice of a commutative separative semigroup,
Casopis Pest. Mat. 113 (1988) 74-79.
[ 1988B] Commutative semigroups whose lattice of tolerances is Boolean,
[ 1988C] Principal tolerance trivial commutative semigroups,
[1988D] Direct decomposability of tolerances and congruences on semigroups,
Ponizovsky, I.S. [1962] A remark on commutative semigroups [Russian],
[1970] The mqtrix representations offinite commutative semigroups [Russian],
Sibirsk. Mat. Z. 11 (1970) 1098-1106, 1197-1198.
Popyrin, A.V. [1984] Properties of bilinear mappings of commutative semigroups
[Russian], Properties of semigroups [Russian], 122-130.
[1985] Weak duality of commutative semigroups [Russian], Algebraic systems with one
action and relation [Russian], 100-113. Leningrad. Gos. Ped. Inst., Leningrad 1985.
[1986] Separability of commutative semigroups by bicharacters [Russian], Semigroups
with additional structures [Russian], 101-108. Leningrad. Univ., Leningrad 1986.
Popyrin, A.V. (cont'd) [1988] Nondegenerate bilinear mappings of commutative semi-

groups [Russian], Homomorphisms [Russian], 111-115.
Porubsky, S. [1977] Commutative semi-primary x-semigroups,
Czechoslovak Math. J. 27 (102) (1977) 467-472. Correction, ibid. 28 (103) (1978) 505.
Poyatos, F. [1966] On irreducible compositions of abelian semigroups [Spanish], Proc.
Seventh Annual Meeting of Spanish Mathematicians (Valladolid 1966) [Spanish], 85-91.
Univ. Valladolid, Valladolid 1967.
[1967] On irreducible decompositions of commutative semigroups [Spanish],
Rev. Mat. Hisp.-Amer. (4) 27 (1967) 39-54.
[1968] Formulas of length for commutative semigroups [Spanish],
Rev. Mat. Hisp.-Amer. (4) 28 (1968) 95-127.
[ 1985] Archimedean decomposition of left S-semimodules and semirings,
Studia Sci. Math. Hungar. 20 (1985) 323-324.
[ 1988] Archimedean decomposition of commutative semigroups with operators,
Studia Sci. Math. Hungar. 23 (1988) 339-342.
Preston, G.B. [1975] Redei's characterization of congruences on finitely generated
free commutative semigroups, Acta Sci. Math. Acad. Sci. Hungar. 26 (1975) 337-342.
Pus, V. [1991] Combinatorial properties ofproducts of graphs,
[ 1992] On multiplicative bases in commutative semigroups,
European J. Combin. 13 (1992) 201-211.
Putcha, M. [ 1971] Commutative semigroups whose homomorphic images in groups are
[1973] Positive quasi-orders on semigroups, Duke Math. J. 40 (1973) 857-869.
[1975] Maximal cancellative subsemigroups and cancellative congruences,
[ 1976] Positive functions from S -indecomposable semigroups into partially ordered sets,
[ 1980] Cancellative subsemigroups of Archimedean semigroups,
Putcha, M. & Tamura, T. [ 1976] Homomorphisms of commutative cancel/ative
semigroups into nonnegative real numbers,
Putcha, M. & Weissglass, J. [1973] Aplications of semigroup algebras to ideal
extensions of semigroups, Semigroup Forum 6 (1973) 283-294.
Rachunek, J. [1987] Prime ideals in autometrized algebras,
Radelecki, S. [ 1986] On the properties of decompositions of commutative semigroups,
Mathematica (Cluij) 28 (51) (1986) 163-165.
Rankin, S.A., Reis, C.M., & Wang, M.L. [ 1976] Globally abelian transformation
semigroups, Semigroup Forum 13 (1976177) 221-228.
Ratschek, H. [ 1972] Teilbarkeitskriterien der Intervallarithmetik,
Ratschek, H. & Schroder, G. [ 1977] Representation of semigroups as systems of
compact convex sets, Proc. Amer. Math. Soc. 65 (1977) 24-28.
REFERENCES. 407
Redei, L. [ 1952] Die Verallgemeinerung der Schreierschen Erweiterungstheorie,

[ 1956] The theory offinitely generated commutative semigroups,
Akad. Kiado, Budapest 1956. English translation, Pergamon Press, Oxford, 1965.
Rees, D. [1940] On semi-groups, Proc. Cambridge Phil. Soc. 36 (1940) 387-400.
Repnitsky, V.B. [1994] On subsemigroup lattices without nontrivial identities,
Algebra Universalis 31 (1994) 256-265.
Repnitsky, V.B. & Katsman, S.I. [1988] Commutative semigroups the lattice of
whose subsemigroups satisfies a nontrivial identity [Russian],
Mat. Sb. (N.S.) 137 (179) (1988) 462-482, 575.
Ressel, P. & Ricker, W.J. [1998] Semigroup representations, positive definite func-
tions and abelian C* -algebras, Proc. Amer. Math. Soc. 126 (1998) 2949-2955.
Rim, D.S. & Vitulli, M.A. [1977] Weierstrass points and monomial curves,
J. Algebra 48 (1977) 454-4 76.
Rodseth, O.J. [1978] On a linear Diophantine problem of Frobenius,
Roiz, E.N. & Schein, B.M. [1978] Radicals ofsemigroups,
Rompke, J. [ 1975] Regular, commutative, maximal semigroups of quotients,
Canad. Math. Bull. 18 (1975) 99-104.
Rosales, J .C. [ 1995] Function minimum associated to a congruence on integral n-tuple
space, Semigroup Forum 51 (1995) 87-95.
[ 1996A] An algorithmic method to compute a minimal relation for any numerical semi-
group, Intemat. J. Algebra Comput. 6 (1996) 441-455.
[1996N] On numerical semigroups, Semigroup Forum 52 (1996) 307-318.
[ 1996S] On symmetric numerical semigroups, J. Algebra 182 (1996) 422-434.
[1997] On presentations ofsubsemigroups ofN.n, Semigroup Forum 55 (1997) 152-159.
Rosales, J.C. & Garcia-Garcia, J.I. [1999] Hereditarily finitely generated commu-
tative monoids, J. Algebra 221 (1999) 723-732.
Rosales, J.C. & Garcia-Sanchez, P.A. [1995] On complete intersection affine
semigroups, Comm. Algebra 23 (1995) no.14, 5395-5412.
[ 1998C] On Cohen-Macaulay and Gorenstein simplicial affine semigroups,
[ 1998N] On numerical semigroups with high embedding dimension,
J. Algebra 203 (1998) 567-578.
[ 1999] Finitely generated commutative semigroups,
Nova Science Publishers, Commack NY, 1991.
[1999F] Onfree affine semigroups, Semigroup Forum 58 (1999) 367-385.
[1999N] On normal affine semigroups, Linear Algebra Appl. 286 (1999) 175-186.
[ 1999P] Presentations for subsemigroups offinitely generated commutative semigroups,
Israel J. Math. I I 3 (1999) 269-283.
Rosales, J.C., Garcia-Sanchez, P.A. & Urbano-Blanco, J.M. [1998] On
Cohen-Macaulay subsemigroups of N2 , Comm. Algebra 26 (1998) no.8, 2543-2558.
408 COMMUTAT IVE SEMIGROUPS .
[1999] On presentations of commutative monoids,

lntemat. J. Algebra Comput. 9 (1999) 539-553.
Ross, K.A. [ 1959] A note on extending semicharacters on semigroups,
[1961] Extending characters on semigroups, Proc. Amer. Math. Soc. 12 (1961) 988-990.
Ruiz, J.M. [1985] Pythagorean real curve germs, J. Algebra 94 (1985) 126-144.
Sagastume Berra, A. E. [ 1957] Theory of ova,
Math. Notae 16 (1957/58) 43-59, 65-77.
Sankaran, N. [ 1961] On some permutable processes in semi-groups,
J. Madras Univ. Sect. B, 31 (1961) 97-107.
Sasaki, M. [1964] On semigroups of positive integer vectors,
Math. Japon. 9 (1964) 31-47.
[ 1965] On the isomorphism problem of certain semigroups constructed from indexed
groups, Proc. Japan Acad. 41 (1965) 763-766.
[ 1966] Commutative nonpotent archimedean semigroups with cancellation law, II,
Math. Japon. 11 (1966) 153-165.
[I 978] Commutative semigroups obtained by an abelian group and a generalized S-
function, Annual Rep. Fac. Ed. Iwate Univ. 38 (1978) no.3, 1-6.
[1982] Supplement to the paper: "Commutative semigroups obtained by an abelian
group and a generalized S-function",
Annual Rep. Fac. Ed. Iwate Univ. 42 (1982) no. I, 51-54.
[ 1983] fit -semigroups with an identity element,
Annual Rep. Fac. Ed. lwate Univ. 43 (1983) no.1, 57-68.
[ 1984] Non-Archimedean fit -semigroups without idempotents,
Annual Rep. Fac. Ed. Iwate Univ. 44 (1984) no.l, 1-12.
[ 1988] Isomorphism problem ofpower semigroups of integer semigroups,
Annual Rep. Fac. Ed. Iwate Univ. 48 (1988) no.l, 123-126.
Sasaki, M. & Tamura, T. [1971] Positive rational semigroups and commutative
power joined cancellative semigroups without idempotent,
Sasaki, M., Spake, R., & Tamura, T. [1986] Archimedean components of the
power semigroup of the group of rational numbers,
Proc. lOth Sympos. on Semigroups (Sakado 1986) 31-35, Josai Univ., Sakado 1987.
Sathaye, A. [1977] On planar curves, Amer. J. Math. 99 (1977) 1105-1135.
Sato, S. [Satoh, S.], Yama K., & Tokizawa, M. [1991] Semigroups of order 8,
Proc. 15th Sympos. on Semigroups, Shimane Univ. (1991) 7-16.
See also: Semigroup Forum 48 (1994) 1-23.
Satyanaray ana, M. [1971A] A class of commutative semigroups in which the idem-
patents are linearly ordered, Czechoslovak Math. J. 21 (96) (1971) 633-637.
[1971B] Commutative semigroups in which primary ideals are prime,
Math. Nachr. 48 (1971) 107-111.
[1972] Commutative primary semigroups,
[1973] On semigroups admitting ring structure. II, Semigroup Forum 6 (1973) 189-197.
[1975] On 0-radica/ of a semigroup, Math. Nachr. 66 (1975) 231-234.
REFERENCES. 409
[ 1977] On commutative semigroups which are unions of a finite number of principal

ideals, Czechoslovak Math. J. 27 (102) (1977) 61-68.
[ 1978] Structure and ideal theory of commutative semigroups, Czechoslovak Math. J. 28
(103) (1978) 171-180. Correction, ibid. 29 (104) (1979) 662-663.
Schein, B.M. [also spelled Sa1n] [1962] On subdirectly irreducible semigroups [Rus-
sian], Dokl. Akad. Nauk SSSR 144 (1962) 999-1002.
[1965] A class of commutative semigroups [Russian],
Publ. Math. Debrecen 12 (1965) 87-88.
[ 1966] Homomorphisms and subdirect decompositions of semigroups,
Pacific J. Math. 17 (1966) 529-547.
[1969] Commutative semigroups where congruences form a chain, Bull. Acad. Polon.
Sci. Ser. Sci. Math. Astron. Phys. 17 (1969) 523-527; Errata, ibid. 23 (1975) 1247-1248.
[ 1974] Injectives in certain classes of semigroups,
Semigroup Forum 9 (1974175) 159-171.
[1975] Commutative semigroups all of whose subsemigroups are densely embedded,
Acta Math. Acad. Sci. Hungar. 26 (1975) 327-330.
[1976] lrljectives commutative semigroups, Algebra Universalis 6 (1976) 395-397.
[1981C] Pseudosimple commutative semigroups, Monatsh. Math. 91 (1981) 77-78.
[19811] On two papers by B.M Schein, Semigroup Forum 23 (1981) 87-88.
Schmidt, J. [ 1975A] Quasi-decompositions, exact sequences, and triple sums of semi-
groups, I. General theory, Contributions to universal algebra (Szeged, 1975) 365-397.
See also Proc. Lattice Theory Conf., 180-206. Univ. Ulm, Ulm 1975.
[ 1975B] Quasi-decompositions, exact sequences, and triple sums of semigroups, II. Ap-
plications, Contributions to universal algebra (Szeged, 1975) 399-428.
Schreier, 0. [1926] Uber die Erweiterungen von Gruppen. /.,
Monatsch. Math. Phys. 34 (1926) 165-180.
Schiitzenberger, M.P. [ 1957] 'D -representation des demi-groupes,
C. R. Acad. Sci. Paris 244 (1957) 1994-1996.
Schwarz, S. [1943] On the theory ofsemigroups [Slovakian],
Sbomfk prac Prfrodovedekej Fak. Slovensk. Univ. Bratislava No.6 (1943).
[1953] Contribution to the theory of torsion semigroups [Russian],
[l954A] The theory of characters offinite commutative semigroups [Russian],
[1954B] Characters of finite commutative semigroups as class functions [Russian],
[ 1954C] On a Galois connection in the theory of characters of commutative semigroups
[Russian], Czechoslovak Math. J. 4 (1954) 291-295.
[ 1977] A counting theorem in the semigroup of Boolean circulant matrices,
Seitz, K. & Blickle, T. [1974] The structure of systems,
Report DM 1974-2, Dept. of Math., Karl Marx Univ. ofEconomics, Budapest 1974.
Seitz, K., Blickle, T., & Grega, B. [1975] On semigroups of type a with special
emphasis on certain applications in chemical engineering,
Report DM 1975-4, Dept. of Math., Karl Marx Univ. of Economics, Budapest 1975.
Selmer, E.S. [1977] On the linear Diophantine problem ofFrobenius,
Shershin, A. C. & Moore, J. T. [ 1965] Direct summands of abelian monoids,
Math. Notae 20 (1965) I 09-116.
Shevrin, L.N. [also spelled Sevrin] [1961] Semigroups with certain types ofsubsemi-
group lattices [Russian], Dokl. Akad. Nauk SSSR 138 (1961) 796-798.
[1963] Commutative semigroups offinite rank [Russian],
Uspehi Mat. Nauk 18 (1963) 201-204.
[ 1966] Eleme'!tary lattice properties of semigroups [Russian],
Sibirsk. Mat. Z. 7 (1966) 664-684.
Shevrin, L.N. ( ed.) [ 1989] The Sverdlovsk Tetrad: unsolved problems in the theory
of semigroups [Russian], Sverd1ovsk, 1989.
Shevrin, L.N. & Ovsyannikov, A.Ya. [1983] Semigroups and their subsemigroup
lattices [Russian], Semigroup Forum 27 (1983) 1-154.
[ 1990] Finitely assembled semigroups and ascending chain condition for subsemigroups,
Monash Conf. on Semigroup Theory (Melbourne, 1990) 269-284.
World Scientific Pub!., River Edge NJ, 1991.
Shirota, T. [ 1952] A generalization of a theorem of I. Kaplansky,
Osaka Math. J. 4 (1952) 121-132.
Shiryaev, V .M. [ 1985] Semigroups with semidistributive subsemigroup lattices [Rus-
sian], Semigroup Forum 27 (1983) 1-154.
Shleifer, F.G. [1981] Identities, quasi-identities, and algorithms [Russian],
Modem algebra [Russian] 128-143, Leningrad. Gos. Ped. Univ., Leningrad 1981.
Shoji, K. [ 1989] Commutative semigroups which are amalgamation bases,
Proc. 13th Sympos. on semigroups (Kyoto 1989) 4-8. Kyoto Sangyo Univ., Kyoto 1990.
Shortt, R.M. & Rao, K.P.S.B. [1989] The dual of a refinement algebra, General
topology and applications (Staten Island 1989) 335-367. Marcel Dekker, New York 1991.
Simmons, H. [ 1976] The word and torsion problems for commutative Thue systems,
Word problems, II (Conf. on Decision problems in algebra, Oxford 1976) 395-400, North-
Holland, Amsterdam-New York 1988.
Simon, J. [ 1979] The lattice of normal subsemigroups of an abelian semigroup [Czech],
Sb. Praci Ped. Fac. v Ostrave Ser. A 14 (1979) no.62, 3-8.
Simons, S. [1963] On the product of semigroups [Russian],
J. London Math. Soc. 38 (1963) 66-70.
Sitnikov, V.M. [1983] Semigroups with a Boolean lattice oftolerances [Russian],
Ural. Gos. Univ. Mat. Zap. 13, no.3, 146-158 (1983).
Skula, L. [1975] Extension of a partial x-operator to an x-operator [Czech],
Kniznice Odbom. Ved. Spisu Vysoke. Uceni Tech. v Bme B-56 (1975) 125-130.
[ 1976] On extensions of partial x -operators,
Spake, R. [ 1986A] Idempotent-free Archimedean components of the power semigroup
of the group of integers. I, Math. Japon. 31 (1986) 791-810.
REFERENCES. 411
[ 1986F] The semigroup of nonempty finite subsets of integers,

Intemat. J. Math. Math. Sci. 9 (1986) 605-616.
[ 1988Q] The semigroup of nonempty finite subsets of rationals,
Intemat. J. Math. Math. Sci. 11 (1988) 81-86.
[1988Z] Semilattice decomposition of the power semigroup of the group of integers.
Math. Japon. 33 (1988) 609-625.
Spake, R. & Hanlon, B. [1993] Archimedean components of 'Y(Z),
Math. Japon. 38 (1993) 55-57.
Speed, T.P. [1968] A note on commutative semigroups,
Stambolieva, G. C. [1972] Projectors in free commutative semigroups with a finite
number of generators [Bulgarian],
Godisnik Vyss. Tehn. Ucebn. Zaved. Mat. 8 (1972) no.4, 97-105.
Stenstrom, B. [1971] Flatness and localization over monoids,
Math. Nachr. 48 (1971) 315-334.
Storrer, H.H. [1976] An algebraic proof of Isbell's zigzag theorem,
Strecker, R. [ 1969] Veral/gemeine Schreiersche Halbgruppenerweiterungen,
Monatsb. Deutsch. Akad. Wiss. Berlin II (1969) 325-328.
Sylvester, J.J. [1884] Mathematical questions with their solutions,
Educational Times 41 (1884) 21.
Szasz, G. [1976] Unitary subsemigroups of commutative semigroups,
Szep, J. [ 1969] On the structure offinite semigroups,
[ 1970] On the structure offinite semigroups. II,
[1973] On the structure offinite semigroups. III,
Szimtenings, D. [1970] Primitivmengen in additiven abelschen Gruppen,
Taiclin, M.A. [1974] The isomorphism problem for commutative semigroups solved
affirmatively [Russian], Theory of models and its applications [Russian], 75-81.
Kazah. Gos. Univ., Alma-Ata, 1980.
[1980] The isomorphism problem for commutative semigroups, [Russian],
Mat. Sb. (N.S.) 93 (135) (1974) 102-128, 152.
Takahashi, M. [1984A] On semimodules, I, Kobe J. Math. I (1984) 67-97.
[l984B] On semimodules, II, Kobe J. Math. I (1984) 177-190.
[1985] On semimodules, Ill, Kobe J. Math. 2 (1985) 131-141.
[1987] Structures ofsemimodules, Kobe J. Math. 4 (1987) 79-101.
Tamura, T. [1953] Some remarks on semigroups and all types of orders 2,3,
J. Gakugei Tokushima Univ. 3 (1953) 1-11.
[1954] Notes on finite semigroups and determination of semigroups of order 4,
J. Gakugei Tokushima Univ. 5 (1954) 17-27.
Tamura, T. (cont'd) [1957] Commutative nonpotent archimedean semigroup with can-

celation law, J. Gakugei Tokushima Univ. 8 (1957) 5-11.
[1963] Minimal commutative divisible semigroups,
[ 1966] Notes on commutative Archimedean semigroups. L IL
Proc. Japan Acad. 42 (1966) 35-40, 545-548.
[1968] Construction of trees and commutative archimedean semigroups,
Math. Nachr. 36 (1968) 257-287.
[1969] Commutative semigroups whose lattice of congruences is a chain,
Bull. Soc. Math. France 97 (1969) 369-380.
[1970C] Commutative archimedean cancellative semigroups without idempotent,
Sem. Dubreil 1969/1970, Fasc. 2, Exp. DG5.
[1970F] Abelian groups and N-semigroups, Proc. Japan Acad. 46 (1970) 212-216.
[ 1970P] Finite union of commutative power joined semigroups,
[1971] On commutative exclusive semigroups, Semigroup Forum 2 (1971) 181-187.
[ 1973C] N-congruences on N-semigroups, J. Algebra 27 (1973) 11-30.
[ 1973S] Finitely generated steady N-semigroups,
[1974B] Basic study of N-semigroups and their homomorphisms,
[1974B] Irreducible N-semigroups, Math. Nachr. 63 (1974) 71-88.
[1975] Basic study of N-semigroups and their homomorphisms II,
[ 1976] Putcha 's problem on maximal cancellative subsemigroups,
Proc. Japan Acad. 52 (1976) 91-94.
[ 1977N] Commutative Archimedean cancellative idempotent-free semigroups with an
infinite cyclic group as structure group, J. Algebra 45 (1977) 343-390.
[ 1977S] Commutative subarchimedean semigroups,
SOrikaisekikenkyUsho Kokyfrroku No.292 (1977) 98-126.
[ 1977T] N-semigroups and their translation semigroups,
J. Austral. Math. Soc. Ser. A 24 (1977) 184-202.
[ 1978] Recent results and problems on commutative archimedean semigroups, Proceed-
ings of the 2nd Seminar on Semigroups, Tokyo Gakugei Univ., Tokyo, 1978, 69-79.
[ 1982N] Nil-orders of commutative nil-bounded semigroups,
[1982R] Commutative cancellative semigroups with nontrivial homomorphisms into non-
negative real numbers, J. Algebra 76 (1982) 25-41.
[ 1984C] On chains whose power semigroups are lattices,
[ 1984Z] On the power semigroup of the group of integers,
[1985] Archimedean components of the power semigroup of the infinite cyclic group,
Proc. 8th Sympos. on Semigroups, Shimane Univ. (1985) 22-25.
REFERENCES. 413
Tamura, T., Dehara, K., Iwata, T., Saito, H., & Tsukumo, K. [1960]
Semigroups of order 5, 6, 7, 8 whose greatest c-homomorphic image are unipotent
semigroups with groups, J. Gakugei Tokushima Univ. 11 (1960) 53-66.
Tamura, T. & Etterbeek, W. [1966] The lattice of congruences of locally cyclic
semigroups, Proc. Japan Acad. 42 (1966) 682-684.
Tamura, T. & Hamilton, H.B. [1971] Commutative semigroups with greatest group-
homomorphic image, Trans. Amer. Math. Soc. 173 (1972) 40 1-419; see also Proc. Japan
Acad. 47 (1971) 671-675.
[1972] Is minimal group congruence smallest? Semigroup Forum 4 (1972) 173-176.
Tamura, T. & Kimura, N. [1954] On decompositions of a commutative semigroup,
Kodai Math. Sem. Rep. 6 (1954) 109-112.
Tamura, T. & Sasaki, M. [1963] Semigroups of positive integer vectors,
Proc. Japan Acad. 39 (1963) 289-293.
Tamura, T. & Shafer, J. [1967] Power semigroups,
Math. Japon. 12 (1967) 25-32; errata, ibid. 29 (1984) 679.
Tanabe, T. & Matsuda, R. [1999] Note on Kaplansky's commutative rings,
Nihonkai Math. J. 10 (1999) 31-61.
Taylor, M. [1999] The generalized equation of bisymmetry: solutions based on can-
cellative abelian monoids, Aequationes Math. 57 (1999) 288-302.
Teissier, B. [1973] Appendix to Zariski [1973], Le prob/eme des modules pour les
branches planes, Ecole Polytechnique, Paris 1973.
Teissier, M. [ 1951] Sur les equivalences regulieres dans les demi-groupes,
C. R. Acad. Sci. Paris 232 (1951) 1987-1989.
Tetsuya, K., Hashimoto, T., Akazawa, T., Shibata, R., lnui, T., and
Tamura, T. [ 1955] All semigroups of order at most 5,
J. Gakugei Tokushima Univ. 6 (1955) 19-39 and Erratum.
Thanh, D. T. [ 1992] d-isomorphic semigroups of continuous JUnctions in locally com-
pact spaces, Acta Math. Hungar. 60 (1992) 103-105.
Thierrin, G. [1953] Quelques properietes des sous-groupoUles consistants d'un demi-
groupe abelien, C. R. Acad. Sci. Paris 236 (1953) 1837-1839.
[ 1954] Sur quelques properietes de certaines classes de demi-groupes,
C. R. Acad. Sci. Paris 239 (1954) 1335-1337.
[ 1956] Sur Ia structure des demi-groupes, PubI. Sci. Univ. Alger. Ser.A 3 ( 1956) 161-171.
Thoma, A. [ 1996] Affine semigroups and monomial varieties,
Comm. Algebra 24 (1996) no.7, 2463-2471.
Thron, R. [1981] Ober das Zutreffen einer Eigenschaft einer Menge auf die Elemente
der Potenzmenge, Beitrllge Algebra Geom. no.ll (1981) 41-50.
Tichy, R.F. [1979] Polynomial functions over monoids,
[ 1981] A remark on polynomial JUnctions over finite monoids,
Toader, Gh. [1990] A hierarchy of supermultiplicity of sequences in a semigroup,
Studia Univ. Babes-Bolyai Math. 35 (1990) no.4, 3-8.
Toea, A. [1977] Generalized monoids offractions,
Studia Univ. Babes-Bolyai Math. 22 (1977) no.l, 12-13.
Toffalori, C. [ 1980] Existentially closed Archimedean semigroups [Italian], Atti Acad.

Lincei Rend. Cl. Sci. Fis. Mat. Natur. (8) 67 (1979) 162-167 (1980).
Tokunaga, H. & Yoshihara, H. [ 1995] Degree of irrationality of abelian surfaces,J.
Algebra 97 (1994) 51-71.
Torres, F. [ 1994] Weierstrass points and double coverings of curves. With application:
symmetric numerical semigroups which cannot be realized as Weierstrass semigroups,
[ 1995] On certain N-sheeted coverings of curves and numerical semigroups which cannot
be realized as Weierstrass semigroups, Comm. Algebra 23 ( 1995) no.11, 4211-4228.
[1997] On 1-hyperelliptic numerical semigroups, Semigroup Forum 55 (1997) 364-379.
Trepetin, M.S. [1971] Semirings of endomorphisms of commutative nilpotent semi-
groups [Russian], Tartu Riikl. 01. Toimetised Vih. 277 (1971) 20-36; letter to the editors,
ibid. 300 (1972) 300.
[1974] The determinability of monogenic nilsemigroups by their semigroups of endo-
morphisms [Russian], Tartu Riikl. 01. Toimetised Vih. 336 (1974) 160-177.
Tretjakova, E.G. [1993] Commutative semigroups which are embeddable into epi-
Trnkova, V. [1975A] Representation ofsemigroups by products in a category, J. Algebra
34 (1975) 191-204. see also Comment. Math. Univ. Carolinae 16 (1975) 393-394.
[1975B] On a representation of commutative semigroups,
[ 1976] Isomorphisms of products and representations of commutative semigroups,
Algebraic theory of semigroups (Szeged, 1976), 657-683.
[ 1978] Cardinal multiplication of relational structures,
Algebraic methods in graph theory, Vol. I,II (Szeged, 1978), 763-792.
Colloq. Math. Soc. Janos Bo1yai, North-Holland, Amsterdam, 1971.
Trueman, D.C. [1979] Direct products of cyclic semigroups,
Semigroups, Academic Press, New York 1980, 103-110.
[ 1980] Finite direct products offinite cyclic semigroups and their characterization,
Proc. Roy. Soc. Edinburgh Sect. A 85 (1980) 337-351.
[1983] The lattice of congruences of direct products of cyclic semigroups and certain
other semigroups, Proc. Roy. Soc. Edinburgh Sect. A 95 (1983) 203-214.
[ 1985] Lattices of congruences on free finitely generated commutative semigroups and
direct products of cyclic monoids, Proc. Roy. Soc. Edinburgh Sect. A 100 (1985) 175-179.
Trung, N.V. & Hoa, L.T. [1986] Affine semigroups and Cohen-Macaulay rings
generated by monomials, Trans. Amer. Math. Soc. 298 (1986) 145-167.
Tully, E.J., Jr. [ 1966] A class of naturally partly ordered commutative archimedean
semigroups with maximal condition, Proc. Amer. Math. Soc. 17 (1966) 1133-1139.
[ 1969] Semigroups in which every ideal is a retract,
[1974] A class offinite commutative archimedean semigroups,
Math. Systems Theory 7 (1974) 289-293.
Turnwa1d, G. [1986] Permutable chains of polynomials of odd degree, Contributions
to general algebra 5 (Salzburg 1986), 407-413. Holder-Pichler-Tempsky, Vienna 1987.
REFERENCES. 415
Tutalar, H.I. [ I987] The Weierstrass nongap sequence beginning with 4, and a con-
struction of its algebraic function field, Doga Mat. 11 (1987) 135-I40.
Utumi, Y. [I956] On quotient rings, Osaka Math. J. 8 (1956) I-I8.
Vandiver, H.S. [I940] On the imbedding of one semi-group in another, with applica-
tions to semi-rings, Amer. J. Math. 62 (I940) 72-78.
Vinarek, J. [ I982] Representations of commutative semigroups by products of metric
0-dimensional spaces, Comment. Math. Univ. Carolinae 23 (1982) 7I5-726.
Waldi, R. [I972] Wertehalbgruppe und Singularitllt einer ebenen algebroiden Kurve,
Doct. Diss., Regensburg Univ., Regensburg I972.
[I980] Zur Konstruktion von Weierstrasspunkten mit vorgegebener Halbgruppe,
Wallace, A.D. [I953] A note on mobs, II, Anais Acad. Brasil Ci. cc 25 (1953) 335-336.
Wang, H.X. [1988] A note on endomorphism semirings ofsemimodules,
Kobe J. Math. 5 (1988) 155-160.
Ward, M. & Dilworth, R.P. [ 1939] The lattice theory of ova,
Ann of Math. 40 (1939) 7I2-730.
Warne, R.J. & Williams, L.K. [1961] Characters on inverse semigroups,
Warner, S. [1960] Mathematical induction in commutative semigroups,
Amer. Math. Monthly 67 (1960) 533-537.
Wegmann, H. [ 1966] Beitrage zur Zahlentheorie auffreie Halbgruppen. II.
Wehrung, F. [ 1996] Tensor products of structures with interpolation,
Pacific J. Math. 176 (1996) 267-285.
[19980] The dimension monoid of a lattice, Algebra Universalis 40 (1998) 247-4Il.
[I998E] Embedding simple commutative monoids into simple refinement monoids,
Semigroup Forum 56 (1998) I04-I29.
Wells, C. [1978] Extension theories for monoids, Semigroup Forum 16 (1978) 13-35.
Wiegandt, R. [1958] On complete semi-modules,
Acta Sci. Math. (Szeged) I9 (1958) 2I9-I23.
Woan, W.J. [I975] Minimal group congruences,
Semigroup Forum II (1975176) I78-I80.
Yamada, M. [I955] On the greatest semilattice decomposition of a semigroup,
Kodai Math. Sem. Rep. 7 (I955) 59-62.
[ I964] Construction of commutative z-semigroups,
Proc. Japan Acad. 40 (1964) 94-98.
[ I965] Construction offinite commutative semigroups,
Bull. Shimane Univ. Natur. Sci. no.I5 ( I965) 1-I1.
[ 1968] Commutative ideal extensions of null semigroups,
Mem. Fac. Lit. Sci. Shimane Univ. Natur. Sci. 1 (1968) no.1, 8-22.
Yamada, M. & Tamura, T. [1969] Note on finite commutative nil-semigroups,
Portugal. Math. 28 (1969) 189-203.
Yap, C.-K. [1991] A new lower bound construction for commutative Thue systems with
applications, J. Symbolic Comput. 12 (1991) 1-27.
Youssfi, E.H. [ 1998] Harmonic analysis on cone/ike bodies and holomorphic functions
on tube domains, J. Funct. Anal. 155 (1998) 381-435.
Zanardo, P. & Zannier, U. [ 197 5) The class semigroup of orders in number fields,
Math. Proc. Cambridge Phil. Soc. 115 (1994) 379-391.
Zariski, 0. [1973] Le probleme des modules pour les branches planes,
Ecole Polytechnique, Paris 1973.
Zelinka, B. [ 1975] Tolerance relations on semilattices,
Comment. Math. Univ. Carolinae 16 (1975) 333-338.
[ 1977] Tolerance relations on periodic commutative semigroups,
Zhang, Y.F., Li, S.Zh., & Wang, D.Sh. [1993] An abstract characterization of
Grothendieckfunctor, Northeast. Math. J. 9 (1993) 477-482.
Zheng, B.D. & Qi, Zh.K. [1997] Primary decompositions of commutative monoids
[Chinese], J. Harbin. lust. Tech. 29 (1997) no.6, 46-49.
AUTHOR INDEX
Abdali, S.K., xi Boguta, B.K., 195

Abel, N.H., xii Borisov, A.A., 191,201
Abrgan, 1., 188,192 Bouvier, A., 31
Abellanas, P., 46,90,194 Bouvier, C., 57
Adamek, J., 195 Brameret, M.P., 35,195
Adams, W.W., 328,332 Brauer, A., 40
Ahsan, J., 200 Bredihin, B.M., 188, 195
Akazawa, T., 187 Bredihin, D.A., 198
Albert, J.P., 199 Bresinsky, H., 43,44
Almqvist, G., 32 Briales, E., 43
Altmann, K., 56 Brillouet-Belluot, N., xi,201
Anderson, D.D., 189,193 Br(Z)ndsted, A., 52,53,54
Anderson, D.F., 43 Brown, D.R., 45,73,75
Anderson, J.A., 103,195,199 Brown, W.C., 43
Anderson, M., 45 Bruns, W., 56,188
Angermtlller, G., 44 Buchweitz, R.O., 42
Apery, R., 43,44 Budach, L., 141,143
Arendt, B.D., 73,86,193,197,203,211, Bulman-Fleming, S., 199,200
216,255 Burgess, D.A., 190
Aubert, K.E., 192, 193 Butzbach, P., 189
Aucoin, K.D., 192,195 Buzachi, K., 203
Aull, C.E., 189 Byeveld, S., 195
Byrd, R.D., 194
Backelin, J ., 41
Bansho, S., 192 Caillot, J .F., 190
Baransky, V.A., 191 Campillo, A., 43
Barr, M., 295,296,300 Carbonne, P., 41,44,310
Barucci, V., 41,43 Carlisle, W.H., 157
Batbedat, A., 201 Cavaliere, M.P., 43
Bazzoni, S., 189 Chacron, J., 25
Beck, J., 295,296,299,300 Chandran, V.R., 198
Berg, L., 189 Chapman, S., 57,67
Berthiaume, P., 32 Chen, Q.H., 192
Bertin, J., 41,44 Cho, J.R., 201
Biggs, R., 86 Chrislock, J.L., 158
Birjukov, A.P., 143 Ch'Uan, J.C., 188
Birkhoff, G., 95,97 Cibul'skite, D., 188
Bisgaard, T.M., 201 Ciric, A., xi,69
Blagojevic, D., 190 Cistov, A.L., 189
Blicke, T., 201 Clark, W.E., 190
Bogacheva, E.D., 196 Clarke, J.L., 203
Bogomolov, V.S., 200 Clifford, A.H., 1,16,37,39,57,58,62,69,
417
418 COMMUTATIVE 8EMIGROUPS
Clifford, A.H. (cont'd), 70,73,74,113 Feil, T., 45

Comfort, W.W., 103 Findlay, G.D., 32
Coppens, M., 42,43 Fischer, K.G., 43,65
Croisot, R., 70, I 04 Fontana, M., 41,43
Cudakov, N.G., 188 Forsythe, G.E., 187
Curtis, F., 41,43 Fountain, J.B., 32
Cvetkova, G., 57 Fraenkel, A.S., 188
Fraser, G .A., 199
D' Anna, M., 43 Freyd, P., 141
Darnel, M.R., 45 Frobenius, F.G., 25,40
de Bruijn, N.G., 188 Froberg, R., 41,43
Deddens, J.A., 41 Fuchs, L., 98
Dehara, K., 187 Fulp, R.O., 103,104,195,199
de Luca, A., 190
Delfino, D., 43 Galanova, J., 199
Delgado, F., 42,57 Galbiati, J .L., 190
Delorme, C., 43 Gale, D., 73,86
Dickinson, R.P., Jr., 86 Garcia-Garcia, J.l., 160-163
Dickson, L.E., 23 Garcia-Sanchez, P.A., 27,40,41 ,54,57, 144
Dilworth, R.P., 192 Gastinger, W., 40
Dobbertin, H., 189,190 Geroldinger, A., xi,57,62,193
Dobbs, D.E., 41,43 Gilmer, R., 62
Dorofeeva, M.P., 192 Gluskin, L.M., 39,76,192,193,198,200
Drbohlav, K., 141,146,197 Golubov, E.G., 100
Dress, A., 103 Gonchigdorzh, R., 189
Dubreil, P., 58,92, 194,197 Gonzales-Springberg, G., 57
Duske, J., 201 Goodearl, K.R., 45,190
Dyadchenko, G.G., 188 Gottlieb, C., 41,43
Gould, M., 195
Easdown, D., 32,195 Gould, V., 32
Ebanks, B.R., 127,201 Grabmeier, J., 103
Eberhart, C., 197 Greco, S., 43
Eigenthaler, G., 189 Green, J.A., 17
Eilenberg, S., 189,193 Grega, B., 201
Ego, M., 191 Grossman, P.A., 196
Emelicev, V.A., 143 Groza, Gh., 189
Ermolina, N.G., 196 Gubeladze, J., 56,188
Esterle, J., xi Gurican, J., 189
Etayo, J.J., 191
Etterbeek, W., 197 Haggkvist, R., 41,43
Evseev, A.E., 15,191 Halezov, V.A., 143
Ewald, G., 56 Hall, R.E., 39,86,160
Halter-Koch, F., 57,67
Faisant, A., 31 Hamilton, H.B., 34,45,74,83,86,197
Farrell, E.J., 189 Hancock, R., 200
AUTHOR INDEX 419
Hanlon, B., 194 Jensen, B.A., 40,191,197

Hansen, R. T., 188 Jezek, J., 201
Hartshorne, R., 189 John, C.C., Jr., 67,133,203,227
Hashimoto, T., 187 Johnson, C.S., Jr., 32,193
Head, T., 47,78,196,199 Johnson, E. W., 192,193
Heinzer, W., 43 Johnson, R.E., 32
Herzinger, K., 43 Jones, P., 188
Herzog, J., xii,43 Juan, L., 41
Heuer, C.V., 39,45,193 Jogensen, H., 1,2,70,187,191
Hewitt, E., 73,76,78,101,102,103,188
Higgins, J.C., 40,86,158,159,199,200 Kaarli, K., 198
Hildebrant, J.A., 193 Kacman [Katsman], S.l., 191
Hill, P., 103,104 Kainrath, F., 67
Hindman, N., 190 Kanemitsu, M., 192
Hinkle, C.V., Jr., 32 Kapur, D., 201
Hippisley-Cox, S.D., 188 Karasek, J., 192
Hirabuki, S., 192 Karhumaki, J., 189
Hmelnitsky, I.L., 192 Kato, T., 43
Hoa, L.T., 57 Kaufman, R.P., 194
Hochster, M., 54,57,62,67 Keimel, K., 199
Hoehnke, H.-J., 197 Kepka, T., 201
Holland, W.C., 190 Khan, M.F., 200
Holme, A., 192 Khan, N.M., 200
Homma, M., 42 Khovansky, A.G., 190,200
Hora, R.B., 190,192 Kil'p, M., 200
Howie, J.M., 1,199,200 Kim, A.C., 195
Hsieh, S.C., 200 Kim, S.J., 42
Huang, I.C., 43 Kimura, N., xii,19,70,75,83,190,199
Huang, W.C., 189 Kisielewicz, A., xiv,201
Hule, H., 192,196 Kist, J., 192
Hurwitz, A., 42 Kitsuneda, Y., 192
Huynh, D.T., 201 Knebl, H., 42
Knebusch, M., 189
lgnat' eva, I. V., I 00 Kleitman, D.J., 187
lmaoka, T., 200 Kobayashi, Y., 86,103,194,203
lnasaridze, H.N., 125,309 Kogalowsky, S.R., 189
Inui, T., 187 Kolibiarova, B., 115
Isbell, J.R., 189,199 Komeda, J., 42
lseki, K., 103 Kominek, Z., 194
Iskra, J.A., 195 Kopytov, V.M., 45
lwanik, A., 193 Koubek, V., 195
Iwata, T., 187 Kowol, G., 196
Kozlov, K.P., 195
Jain, R.K., 192,193 Kranz, P., 194
Jarek, P., 74 Krause, U., 57,67,194
420 COMMUTATIVE SEMIGROUPS
Krivenko, V.M., I92 Maclachlan, C., 42

Krob, D., I90 MacLane, S., 295,297,3I8
Kruming, P.D., 309 Malcev [Mal'tsev], A.I., 95,154,I57
Kubianovsky, S.l., I 00, I95 Mal'tsev, Yu.N., I89
Kulkarni, D.M., I89 Mannepalli, V.L., I92, I93
Kunz, E., 40,43, I28 Manukjanc, M.G., I95,I96
Kunze, M., I89 Marcus, D.A., I88
Kupcov [Kuptsov], A.l., I95, I96 Marijuan, C., 43
Kuroki, N ., I99 Marki, L., I98
Markov, S.M., xi,201
Lal, H., I92, I93 Matsuda, R., 52,192
Lal, R.N., 20I Maury, G .,31, 192
Lallement, G., I57 McAlister, D.B., 34,78,90,91,93,103,
Lambek, J., 32 I58,160
Laplaza, M.L., 4 7, I9I McDowell, K., I99,200
LaTorre, J.G., 46,73,75 McMorris, F.R., 32,193
Lau, G., I98 McNeil, P.E., 100, I93
Lawson, J.D., xii Medvedev, N.Ya., 45
Lee, W.S., I88 Megyesi, L., 193
Leech, J., xii,xiii,II5,I25 Migliorini, F., 2, 70
Lenzi, D., I88 Miller, D. W., 39,40,45,191, I93, I97
Lepetit, C., I88 Mitrovic, Z.M., 188
Lesieur, L., I92 Mitsch, H., 4, I96, I97
Lesohin [Lesokhin], M.M., 95,IOO,I03, Mogiljanskaja, E.M., 19l,I94
I04,I93,I95,I96 Moh, T. T., 43
Lett!, G., 57,67 Monzo, R.A.R., I93
Levin, R.G., 90,9I,92,I58 Moore, E.H., 25
Li, S.Zh., I99,200 Moore, J.T., I92
Liberman, A.A., I9I Morel, A.C., 191,192
Libih, A.L., I9I Morris, W., 56
Lid!, R., I89 Moszner, Z., 20 I
Ligh, S., I88 Motzkin, T.S., 187
Lipkovski, A., I89 Moller, W.B., 189
Liu, Zh.H., 200 Murata, K., I92
Ljubenova, L., 57 Murphy, G.J., xi
Lloyd, J.T., I94 Murty, A.S.R., 200
Lomadze, D.D., IOO Mushtaq, Q., 201
Lord, J.M., 73,86 Mustafin, T.G., 200
Lorenzen, P., 192
Loustaunau, P., 328,332 Nagy, A., xi,69
Luedemann, J.K., 32 Nakajima, S., I95
Lugowski, H., 125 Narendran, P., 201
Lyapin, E.S., I5,197 Nasirov, S.N., 195
Nathanson, M.B., 190
MacKenzie, R.E., I92 Neeb, K.-H., 57
AUTHOR INDEX 421
Neumann, B.H., 195 Porta, H., 188

Newrly, N., 201 Porubsky, S., 192,193
Nguyen, H.K., 120 Poyatos, F., 70,192,200
Niculescu, C.P., 190 Preston, G.B., xii,1,16,69,70,113,141
Niesi, G., 43 Pus, v., 190
Nikolaev, A.M., 199 Putcha, M., 73,193,194,196,198
Niven, I., 188
Nobauer, W., 189 Qi, Zh.K., 192
Nordahl, T., 45,83,84,192,197
Novikov, B.V., 309,310 Rachunek, 1., 192
Raciti, G., 43
O'Carroll, L., 34,78,91,158,160 Radelecki, S., 93
Oda, T., 57 Ragusa, A., 43
O'Dunlaing, C., 201 Rankin, S.A., 189
Oeljeklaus, E., 67 Rao, K.P.S.B., 190
Okabe, A., 192 Ratschek, H., 188,195
Okninski, J., 191 Redei, L., xi,xii,22,44,125,141,204,
Ovsyannikov, A.Ya., 160,190,191 206,208
Rees, D., 4,14
Painter, P., 188 Reis, C.M., 189
Pales, Z., 192 Repnitsky, V.B., 191
Palfy, P.P., 195 Ressel, P., xi, 195
Pallas, 1., 187 Restivo, A., 190
Parizek, B., 103,188 Ricker, W.J., xi,195
Parker, E. T., 198 Rim, D.S., 42
Patil, D.P., 41,43 Rodseth, O.J., 41
Pavlyucuk, A.K., 188 Ro'lz, E.N., 192
Paxia, G., 43 Rolletschek, H., 201
Pedersen, F.D., 194 Rompke, J., 32,76
Perel, W.M., 192 Rosales, J.C., 27,40,41,54,57,144,
Peric, V., 192 160-163,328,329,3 30
Perrot, J.-F., 189 Ross, K.A., 103
Petrich, M., 16,32,37,50,69,70,72,74,103, Rothschild, B.R., 187
158,160,193 Ruiz, J.M., 43
Pfister, G., 43
Philip, J.M., 199 Sagastume Berra, A.E., 187
Pis6n, P., 43 Saito, H., 187
Plandowski, W., 189 Salce, L., 189
Plappert, P., 194 Sankaran, N., 39
Plemmons, R.J., 187 Sasaki, M., 86,91,188,194
Plonka, J., 193 Sathaye, A., 41
Pollak, G., 193 Sato, S., 1,187
Pondelicek, B., 104,198 Satyanarayana, M., 143,189,192,193
Ponizovsky, I.S., xii,95, 107,194 Schein [Sa'ln], B.M., 39,98,100,102,104,
Popyrin, A.V., 100,196 106, 192, 197,200
Scherpenisse, C., 43 Szekely, G.J., 190

Schmidt, E.T., I98 Szep, J., 2,70
Schmidt, J., I20 Szimtenings, D., 194
Schreier, 0., I25
Schroder, G., I95 Taiclin, M.A., 144
Schotzenberger, M.-P., 20, I89, I93 Takahashi, M., 187,193,200
Schwarz, S., xii,I9,70,76,78,IOI,I03, Tamura, T., xii,25,34,45,48, 70, 73, 74,83,
I88,I89 84,86,91,92,93, 103,133,158,187,188,
Seitz, K., 20 I 190,193,194,197,201,203
Selfridge, J .L., I87 Tanabe, T., 192
Selmer, E.S., 4I,42 Taylor, M., xi,201
Shabir, M., 200 Teissier, B., 44
Shafer, J., I93,I94 Teissier, M., 104,212
Shapiro, J., 43,56 Tetsuya, K., 187
Shershin, A.C., I92 Thanh, D.T., I89
Shevrin [Sevrin], L.N., I60,166,I90,I9I Thierrin, G., 50,70,95
Shibata, R., I87 Thoma, A., 56
Shirota, T., I89 Thron, R., 201
Shiryaev, V.M., I9I Tichy, R.F., 196
Shleifer, F.G., I44 Toader, Gh., 201
Shoji, K., 200 Toea, A., 32,194
Shokuev, V.N., I88 Toffalori, C., 192
Shortt, R.M., I90 Tokizawa, M., 1,187
Simmons, H., 20 I Tokunaga, H., 43
Simon, J., I92 Torres, F., 42,43
Simons, S., 86 Trepetin, M.S., I95
Singh, B., 4I Tretjakova, E.G., 174
Sitnikov, V.M., I98 Tmkova, V., I95
Sizer, W.S., I94 Trueman, D.C., 197,I98
Skula, L., I92 Trung, N.V., 57
Spake, R., I94 Tsai, Y.S., 75,83
Speed, T.P., 20I Tsukumo, K., 187
Spencer, J.H., I87 Tully, E.J., Jr., 73,193,203
Stambolieva, G.C., 189 Tumwald, G., 189
Steenbrink, J.H.M., 43 Tutalar, H.l., 42
Stenstrom, B., 32,199
Stepp, J.W., 194 Urbano-Blanco, J.M., 57
Stickles, J., 189 Utumi, Y., 32
Stolarsky, K.B., 188
Storrer, H.H., 199 Vandiver, H.S., 29
Strecker, R., 125 Veronesi, M.L., 190
Stuth, C.J., 73,193,203,211,216,255 Vinarek, J., 195
Sugatani, T., 192 Vishnyakova, N.I., 203
Sylvester, J.J., 40 Vitulli, M.A., 42
Szasz, G., 197 Volkov, M.V., 82
AUTHOR INDEX 423
Waldi, R., 40, 42,43 Woan, W.J., 190,197

Wallace, A.D., 19 Woracek, H., 189
Wang, D.Sh., 199
Wang, H.X., 200 Yama, K., 1,187
Wang, M.L., 189 Yamada, M., xii,70,73,133,193,203
Ward, M., 192 Yap, C.-K., 201
Warne, R.J., 103 Yoshihara, H., 43
Warner, S., 201 Youssfi, E.H., xi
Wegmann, H., 188 Yusuf, S.M., 201
Wehrung, F., xi,189,190,199
Weissglass, J., 193 Zanardo, P., 189
Wells, C., 300,301,305 Zannier, U., 189
Wessels, U ., 56 Zariski, 0., 44
Wick, P., 1,187,191 Zelinka, B., 198
Wiegandt, R., 200 Zhang, Y.F., 192,199
Wiesenbauer, J., 189 Zheng, B.D., 192
Williams, L.K., 103 Zuckerman, H.S., 73,76,78,101,102,103,188
NOTATION
General conventions: MB 1 , Me 1 , MZ 1 , 333

A, B, Z: sets, semigroups
0 0 0 ,
P, 329
a, b,
0 0z: elements0 ,
pA, 205
A, 'B, Z: relations
p~, 205
0 0 0,
A,'B,ooo,Z: categories
PA,I70
E a : the f.. -class of a
Pe , Ponizovsky factor, I 08
a, {3, 0 w: mappings
0 0,
A, JR, Z: explicit semigroups

0 0 0 ,
Pe* , partial Ponizovsky factor, I 09
A, JR, Z: functors
0 0 0, PHom(S,T), 6
Q, 329
A-Z: q(a), 329
A( x) , archimedean component, 72 R a , strand group, 275
Ab, the category of abelian groups RA ( t) , strand group, 272
Bn, 298
R(r, s), relation, 348
BiHom(A,B,e), 196
sa, 278
en' 298,313 S a , strand, 244,275
D (a) , direction face
sa (t), 278
D( e), direction set
E A , extent cell SA (t), strand, 268
Ext(S,G), 13I,323 SBn, 319,324,325
Fx,20
sen' 3I9,324,325
St(H),I9
Gx,20
Ga, 275
szn' 3I9,324,325
Tn, 298,3II
G(S), 32
U, group of units, 45
H (S) , category, II8 U, identity class, 209
HA, 245,268 U ( s) , upper section, stabilizer,
Hn , cohomology group, 296,299 Y(S), 72
Hom(S,T), 6
zn' 298
I A ( t) , strand ideal, 272 Z(r,s,a), 350
im cp, 6
Im cp, 8 A-Z:
Irr (S), 256 A, 30I
JA, 245,268 'B, a semi lattice congruence
K, 368 e' a congruence; a category;
KA, 170 the category of CoSo
ker cp, 6 e*' Cogof. hull, 270
LA, I70 1J , J -congruence, 213
M,329 1J, direction set
425
426 NOTATION
£, extent cell family Other notation:

JC, Green's relation 17 s1 ' 4
X, 180,271 S, divisible hull, 48
/:..;, 253
S, real hull, 48
J\t(a, 159
Tn, 298
N, 71
(')' 148 e= e +s' category of objects over s
P, pure congruence, 104 TI, 296
P, Ponizovsky family, 170 W, 296,297
QA, 291 -+
Na , 301
8, separative congruence, 77
8, strand congruence, 240,268
iU, 296,297
'J, 46,83 V, 297
U, universal congruence, 12 "E, 296,297
U, "underlying" category En,i, 297
rj, 296,297
A-Z: c*, 228,235
A, 295,297 Sp, 174
JF, "free" functor s1e, quotient semigroup, 8
N, 2 S I I, Rees quotient, 15
N+ 2 SIH, where H s;;; G(S), 44
'
Ql+' 2 xI a, fraction, 29
IR+ 2 a+, a-, 205
'
§, circle group, 10 I
1U, "underlying" functor
c-l s' semigroup of fractions, 29
;2;' 4,20,205,208,232,355
V, 295,296
;2;9{, 17
Greek letters: ;2;N, 70
r(H), 19,182 ~. 328
]['' 119,185 1\, infimum, 2
a, boundary, 316 V, supremum, 205
a, divisor theory, 57 (X), 5
15' 296,298,314,333 (X I :R)' 26
m
E, 295,296,310
a -------+ b, 330
En, i, 296,311
'fl, 296
L, canonical injection
v, 296
1r, projection
7rn , 298,311
I;(K), 182
E, 185
INDEX
abelian group object, 297, 301-305 c.g.f.m. = c.g.f. monoid, 227

action of-, 297,306 c.m. = commutative monoid, 29
abelian group valued functor, 118 c.pc.r., 47
almost constant-, 138,318 c.s. = commutative semigroup, 29
almost null - , 131 cancellative
constant, 3 17 - congruence, 34,44,284,290
finite - , 119 - element, 30
surjecting - , 119 - semigroup, 30,36--67,82-86,160,
thin-, 118 162,309
action of abelian group object, 297,306 - semigroup with zero, 102
affine categorical ideal, 193
- completeness, 198 chain, 316
- semigroup, 56 character, 10 I, I 0 1-1 04
almost constant functor, 138,318 circle group, I 0 I
almost null functor, 131 ,366 class
antichain, 22,142 equivalence - , 7
aperiodic, 120 identity - , 209
approximable, 100 zero-, 210,290
finitely - , I 00 classifies extensions, 300,315
archimedean Clifford semigroup, 73,73-75, 101,116,121,
- cancellativity index, 147 195,198,309
- component, 72,116,23 5 closing element, 191
- decomposition, 72 coboundary, 298,314
- direction set, 260 minimal-, 333
- semigroup, 71,72,78-93,190 symmetric - , 319,324,325
complete-, 78,78-82,116,121,124 cochain, 298, 312-314
finite - , 78-82 consistent - , 339
finitely generated-, 158-162 minimal - , 333
associativity, 3,15 partial - , 340
symmetric-, 318-326,319,324
Beck cocycle, 298,314
-cohomology, 298, 295-300 minimal - , 333
-extension, 300, 305-309 symmetric - , 319,324,325
bicharacter, 195, 195-196 coextension, 120
bihomomorphism, 195, 195-196 group-, 121, 121-133
complete - , 196 9-!--, 120,115-117,122
regular - , 196 cofinite ideal, 210
bilinear mapping, 195 cohomology, 295-378
Birkhoff's Theorem, 97 - o f cyclic semigroup, 335
bounded semicharacter, 103 - o f one relator semigroup, 335
- o f partially free semigroups, 355
c.g.f. = complete group-free, 227 - of Volkov semigroup, 352
427
428 COMMUTATIVE 8EMIGROUPS
cohomology (cont' d) congruence (cont'd)

Beck - , 298,313,295- complete - , 265,275
Eilenberg-MacLane - , 309 conditions on -s, 197
triple - , 298, 295-300 direction set of-, 260
coideal, 218 finite - , 275,285
trace-, 290 finite nilmonoid - , 209
combinatorial, 120 group-, 197
commutative, 2 group-free, 233,250,227-258
- Clifford semigroup, 73 identity class of-, 209
-cohomology, 313, 295-378 irreducible-, 144
- group coextension, 121 J--, 211,219
- 9-C-coextension, 120 kernel function of-, 206,207,221
- inverse semigroup, 73 lattice of -s, 197
- regular semigroup, 73 nilmonoid - , 208, 208-226
-word, 310 power cancellative - , 46
commutativity, 3 proper-, 97
compatible well order, 328 pure-, 104
complementary face, 205 Redei group of-, 206,221
complete reduced nilmonoid - , 209
- bihomomorphism, 196 Rees - , 15,207
- congruence, 265,275 se~ilattice - , 70, 227-232
- group-free, see below separative - , 77
- integral closure, 62, 198 strand - , 240,244
- semigroup, 107,78,107-111,116,119, subcomplete - , 259-293,282,292
120,122,172,227-258 subelementary, 264,284,288-290
complete group-free, support of J - - , 216
- congruence, 233,250,227-258,275 trace - , 245,244-249,269,268-271
-hull, 270 universal - , 12
- semigroup, 45,120,232,232-258 zero class of - , 210
- strand group family, 285 consistent cochain, 339
complete semigroup (Krob ), 190 constant functor, 317
completion, 165,165-169 comer point, 213
monoid - , 168 cotriple, 296
sharp-, 165 cross section, 125
Completion Theorem, 153,173 cyclic semigroup, 5,6,24,335
component
- of decomposition, 96 decomposition, 70
archimedean - , 72,116,235 Ponizovsky - , 108-111
power joined - , 93 semilattice - , 70
conductor, 41 subdirect - , 96,95-111,144,145,166,
congruence, 8,7-13,197-198 167,179
- generated by relation, 13 defining
- on Clifford semigroup, 74 -basis, 371
- on free semigroup, 203-296,329 - vector, 345, 345-354
cancellative - , 34,44,284,290 degree lexicographic order, 328
INDEX 429
Dickson's Theorem, 22,142 exact

Diophantine semigroup, 59 - 9-C-coextension, 121
direct summand, 192 -sequence, 299,315
direct image - strand bases, 279
- of subsemigroup, 10 extended SchUtzenberger functor, 185,278
- of congruence, 10 extension, 120, 198
direct product, 95 -group, 131,323
direction face, 227,233,290 Beck - , 300, 305-309
direction set, 228,235,227-236,260, dense-, 198
259-261 ideal-, 15,15-17,37-39,193
archimedean - , 260 normal-, 198
finite - , 230 extent cells, 237,236--244,263,261-268
divisibility preorder, 17 extent cell family, 237,242,263,267,268,
divisible hull, 48 261-268
divisor theory, 57 sharp - , 263,264,275,285
dominion, 199
duality, 196 f.g. = finitely generated, 141
face, 50,204
Eilenberg-MacLane cohomology, 309 complementary - , 205
element direction - , 227,233,290
cancellative - , 30 factor, Ponizovsky, 108,246,256-258
closing - , 191 factor set, 126,319
idempotent - , 4 equivalent -s, 128
identity - , 3 normalized, 130
infinitely divisible - , 190 split-, 129,319
irreducible - , 208,232 filter, 218
minimal, I 05 finite
nilpotent - , 78 -congruence, 275,285
zero-, 4 - functor, 119
elementary semigroup, 109,112-114,117, - nilmonoid congruence, 209
121,124,135-137 - nilsemigroup, 105-107
homogeneous - , 136 - semigroup, 14,25,36,78-82,107-111,
orbits of - , 112 285,309
embedding, 9 finitely approximable, 100
empty finitely generated
-intersection, 12 - archimedean, 158-162
- product, 3,95 - nilsemigroup, I 05
-union, 12 - semigroup, 6,141-163,166,168,173,
endomorphisms, 195 190,284,351
epigroup, 79, 166, 174 hereditarily-, 161
index of-, 174 finitely presented, 143
epimorphisms, 199 finitely subdirectly irreducible, 98,134-137,
equivalent 154-157
- factor sets, 128 - abelian group, 99
- group coestensions, 125 - cancellative semigroup, 99
First Isomorphism Theorem, 8 group (cont'd)

fraction, 29 - coextension, see below
group of-s, 33 - congruence, 197
semigroup of-·-s, 29, 29-32 - extension, 131
free -free, 45,120,232-258
- abelian group, 20,205 - of fractions, 33
- commutative monoid, 23, 204-208 - of units, 18
- commutative semigroup, 20,20-23, - valued functor, 73
188,204-208,310,327-329 - with zero, 102
- commutative semigroup with zero, 24 circle-, 101
- semilattice, 204 extension-, 131,323
partially-, 255 Redei - , 44,206,221,346
fried chicken problem, 40 Schntzenberger - , 19,20,182-184,276
Frobenius number, 41 strand-, 273,271-278,284
function minimum, 329 structure - , 84
functor universal-, 32,32-37,74,80-81,
abelian group valued - , 118 216,346
almost constant - , 138,318 zero-, 366
almost null-, 131,366 group coextension, 121, 121-133
constant, 3 17 cross section of-, 125
finite - , 119 equivalent - , 125
group valued - , 73 factor set of-, 126
reduced, - , 366 split-, 129
semiconstant - , 369 group-free
strand group - , 277 - congruence, 233,250,227-258
strand monoid - , 277 -hull, 270
thin-, 118 - semigroup, 45,120,232,232-258
SchUtzenberger - , 119,117-120,121,
124,184-186,277 !Ji -coextension, 120,115-117,122
selective - , 362 exact-, 121
surjecting, 119 hereditarily finitely generated, 161
ho1oid, 120,201
gap, 42 homogeneous, 136,156
generated by ... subject to ... , 26, 25-28 homomorphic image, 9
generated by subset homomorphism 6, 6-12
ideal-, 13 - ofmonoids, 12
semigroup - , 5 - of semigroups, 6
subsemigroup - , 5 -Theorem, 8
generated by relation, 13 partial - , 15,80
generator, 5 pure-, 211
genus, 42 semigroup of -s, 6,195
global semigroup, 193 trimming - , 320
Green's preorder, 17 hull
Grobner basis, 331 complete group-free - , 270
group, I divisible - , 48
INDEX 431
hull (cont'd) integal closure, 62

real-, 48 complete - , 62
integrally closed semigroup, 62
ideal, 13, 192-193 intersection
- extension, see below empty-, 12
-generated by subset, 13 inverse image
- theory, 192 - of subsemigroup, 10
categorical - , 193 - of congruence, 11
cofinite - , 210 irreducible
conditions on -s, 193 - congruence, 144
lattice of -s, 193 - element, 208,232
nilmonoid - , 210 - N-semigroup, 86
order-, 218 - semigroup, 144, 154-157
prime-, 50,70 subdirectly - , 96
principal - , 14 weakly - , 134
semiprime - , 103 isomorphism, 6
ideal extension, 15,15-17,37-39,193 -problem, 27,144
- determined by a partial homomor- - Theorems, 8,1 0,11
phism, 17,39
- of nilsemigroup, 112 J -congruence, 211,219
-problem, 16
canonical homomorphism of-, 38 kernel, 14,109
retract - , 16,17 -function, 206,207,221,248,280
idempotent, 4 Kleitman-Rothschild-Spencer conjecture,
identity 187,203
-class, 209 Krull monoid, 57
-element, 3
image lattice, 189
- of homomorphism, 8 - of congruences, 197
- of mapping, 6 - of ideals, 193
direct - of subsemigroup, I 0 - of subsemigroups, 190-191
direct - of congruence, 10 length of element, 22
homomorphic - , 9 lexicographic order, 328
inverse - of subsemigroup, l 0 degree - , 328
inverse - of congruence, 11 linear admissible order, 329
independence of path, 344 Iin early independent subset, 193
index lower
- of conguence on N, 24 - section, 218
- of element, 24 - semilattice, 2
- o f epigroup, 174 Lilroth semigroup, 43
- o f semigroup, 147
induced Malcev's Theorem, 157
-congruence, 8,9,11,141 maximal embedding dimension, 41
infinitely divisible element, 190 minimal
injective semigroup, 200 - coboundary, 333
minimal (cont'd) object

- cochain, 333 -overS, 296
- cocycle, 333 abelian group-, 297, 30I-305
- element, I 05 one relator semigroup, 333-335
money-changing problem, 40 orbits, 112, I49
monoid, 3 monoid of-, 112,I49,290
-completion, I68 order
- of orbits, II2,290 -ideal, 218
free-, 23 degree lexicographic - , 328
nil-, 112 compatible well - , 328
Schtltzenberger - , I82,276 lexicographic - , 328
monomial mapping, I96 linear admissible - , 329
morphism prime-, 328
- o f abelian group objects, 297 overpath, 330, 327-360
- of group coextensions, 306 - method, 333, 327-360
- of objects over S, 296
multiplicity, 4I partial
- binary operation, I5
naturally partially ordered, I20 - cochain, 340
naturally totally ordered, 20 I -homomorphism, I5,80
negative part, 205 - Ponizovsky factor, I 09,238
nest of equivalence relations, 2I8 - semigroup, I5, 79-80
coherent - , 2I9 partially free semigroup, 255,254-258,
nilmonoid, 112 355-360,362
- congruence, 208,209,208-226 path, 330
-ideal, 210 independence of - , 344
unique factorization - , 210 period, 24
0-free - , 2I0,255 polynomial mapping, 196
nilpotent Ponizovsky decomposition, I 08-III
- element, 78 Ponizovsky factor, I08,238,246,256-258
- semigroup, I05,187 partial-, I09,238
nilsemigroup, 78, I 04-I 07,203,208-226 Ponizovsky family, I7I, I69-I86,262
- with enough minimal elements, I 05 induced by completion, I7I
finitely generated, 105 sharp-, I7I
ideal extension of-, 112 standard - , I73
subdirectly irreducible - , I 06,213 positive
weakly irreducible, 134 -part, 205
normal semigroup, 62 - relation, 346
normalized factor set, 130 power
N-semigroup, 82,82-86,90,92, I 58, I60, - of element, 3
I94,203 - cancellative, 46
irreducible - , 86 -joined, 88,90-93,158-I60
numerical semigroup, 40, 39-44 - - component, 93
free-, 4I - semigroup, 193
symmetric - , 4I preorder, 17
INDEX 433
preorder, 17 range of mapping, 6

divisibility - , 17 rank of semigroup, 46, 191
Green's - , 17 rational
presentation, 27,25-28,208,220,247,279 - semigroup, 90
standard - , 209,233 - subset, 193
prime reachable relation, 371
-ideal, 50,70 real hull, 48
-order, 328 realized relation, 348
principal ideal, 14 reduced
problem -cone, 48
fried chicken - , 40 -functor, 366
ideal extension-, 16 - nilmonoid congruence, 209
isomorphism-, 27,144 - semigroup, 45
semilattice composition - , 72 reduced rank, 46
money-changing - , 40 reductive semigroup, 39
word-, 27,143 Redei group, 44,206,221,346
product, 3 Redei's Theorem, 142,329
- of subsets, 5, 190 Rees congruence, 15,207
direct-, 95 Rees order, 4
empty - , 3,95 Rees quotient, 15
tensor - , 196, 199 regular bihomomorphism, 196
subdirect - , 96 1r-regular semigroup, 79
profinite semigroup, 100 refinement semigroup, 190
projection relation, 25
- from direct product, 95 - holds in semigroup, 25
-from subdirect product, 96 positive - , 346
- to face, 205 reachable, 371
- to Ponizovsky factor, 109 realized - , 348
- to quotient set, 7 vector - , 346
- to quotient semigroup, 8 verifiable - , 350
projective semigroup, 200 representations, 194
proper congruence, 97 residually finite, 100
pseudo-invertible semigroup, 79 retract ideal extension, 16,17
pure retraction, 16
- congruence, 104 Riesz interpolation property, 188, 190,199
-homomorphism, 211,255 root closed semigroup, 62
- subgroup, 60 root closure, 62
quasiorder, 17 s'Y, 174,173-179,180,183,185

quasi-universal free c.m., 56 saturated semigroup, 199
quasi-universal property, 56 Schutzenberger
quotient set, 7 - group, 19,20,182-184,276
quotient semigroup, 8 - monoid, 182,276
-monoid functor, 185,184-185,277
radical, 192,197 -functor, 119,117-120,121,124,
Schutzenberger functor (cont'd) semigroup (cont'd)

184-186,277 group-free-, 45,120,232,232-258
selective functor, 362 hereditarily finitely generated - , 161
semi character, 101, 101-104 injective - , 200
bounded-, 103 integrally closed - , 62
semiconstant functor, 369 inverse - , 73
semigroup, 1 irreducible-, 144, 144-147,154-157
- cohomology, 313, 295-378 Krull-, 57
- generated by . . . subject to ... , 26 LUroth - , 43
- o f fractions, 29, 29-32 N--, 82,82-86,90,92,158,160,194,203
- of homomorphisms, 6,195 naturally partially ordered-, 120
- presentation, 27 naturally totally ordered - , 201
affine-, 56 nil-, 78,104-107,203,208-226
aperiodic-, 120 nilpotent-, 105,187
approximable - , 100 normal-, 62
archimedean - , 71,72,78-93,190 numerical - , 40, 39-44
cancellative - , 30,36-{)7 ,82-86,160, one relator-, 333-335
162,309 partial-, 15, 79-80
cancellative - with zero, I 02 partially free - , 255,254-258,355-360
character of - , 101-104 power-, 193
Clifford-, 73,73-75,101,116,121,195, power joined-, 88,90-93,158-160
198,309 power cancellative - , 46
combinatorial-, 120 profinite - , 100
commutative - , 2 projective - , 200
complete-, 107,78,107-111,116,119, pseudo-invertible - , 79
complete- (Krob), 190 quotient - , 8
120,122,172 rank of-, 46,191
cyclic - , 6,24,335 radical of-, 192, 197
Diophantine - , 59 rational - , 90
divisible hull of-, 48 real hull of-, 48
elementary-, 109,112-114,117,121, reduced - , 45
124,135-137 reduced rank of-, 46
finite-, 14,25,36,78-82,105-111, reductive - , 39
285,309 refinement - , 190
finitely approximable - , 100 regular - , 73
finitely generated-, 6,141-163,166, ;r -regular - , 79
168,173,190,284,351 residually finite-, 100

finitely presented-, 143 root closed - , 62
finitely subdirectly irreducible - , 98, saturated - , 199
144-147,154-157 semicharacter of-, 101-104
face of-, 50 semi lattice of -s, 69
free-, 20,20-23,188,204-208,310, separable-, 100
327-329 separative-, 39,75, 75-78, 101-104
generator of - , 5 Stone-, 198
global - , 193 strongly noetherian, 160, 160-163
INDEX 435
semigroup (cont' d) strand (cont'd)

subcomplete - , 86,166,165-186 - group functor, 277
subdirectly irreducible - , 96,95-111, - monoid functor, 277
133-139,154-157 strongly noetherian, 160
subelementary - , 144,148-149, structure group, 84,90
155-156,173,177,181 S-set, 200
subgroup of-, 18,19 subcomplete
torsion free - , 46 - archimedean semigroup, 86, 86-90
Volkov - , 82,105,107,217,323,352,375 - congruences, 259-293,282,292
weakly irreducible - , 134, 156 - semigroup, 166, 165-186
Weierstrass - , 42 subdirect
semilattice, 2,4,197,198 - product, 96
- composition problem, 72 -decomposition, 96, 95-111,144,145,
- congruence, 70, 227-232 166,167,179
- decomposition, 70 subdirectly irreducible, 96,95-111,133-139,
- o f cancellative semigroups, 75, 75-78 154-157
- o f groups, 73, 73-75 - abelian group, 98
- of semigroups, 69 - cancellative semigroup, 98
lower-, 2 - nilsemigroup, I 06
subdirectly irreducible - , 96 finitely-, 98,134-137,144-147
universal - , 72, 116 subelementary congruence, 264,284,
semimodule, 200 288-290
semiprime ideal, 103 subelementary semigroup, 144,148-149,
separable semigroup, 100 155-156,173,177,181
separate homogeneous - , 156
- the elements, 96 subgroup, 18, 19
separative submonoid, 6
- congruence, 77 subsemigroup, 5
- semigroup, 39,75,75-78,101-104 - generated by subset, 5
sharp - o f N, 39-44
-extent cell family, 263,264,275,285 - o f z, 39
- Ponizovsky family, 171 conditions on -s, 191-192
- strand group family, 285 cyclic-, 5
short V -exact sequence, 299 lattice of -s, 190-191
split unitary - , 58
- coextension, 129 subset
- factor set, 129 linearly independent - , 193
standard rational - , 193
- Ponizovsky family, 173 sum, 3
- presentation, 209,233 support of J -congruence, 216
Stone semigroup, 198 surjecting functor, 119
strand, 244,268 symmetric
-base, 278,278-281,284,336-337 - coboundary, 319,324,325
- congruence, 240,244,268 - cochain, 318-326,319,324
-group, 273,271-278,284 - cocycle, 319,324,325
tensor product, 196,199 universal (cont' d)

Theorem -congruence, 12
Completion-, 153,173 -group, 32,32-37,74,80-81,216,346
Dickson's - , 22,142 - power cancellative semigroup, 46--47
First Isomorphism - , 8 - semilattice, 72, 116
Homomorphism - , 8 - semilattice of groups, 77
Isomorphism - , 8, 10,11 - separative semigroup, 78
Malcev's - , 157
Redei's - , 142 vector
Universal Coefficients - , 318 - relation, 346
thin functor, 118 defining - , 345, 345-354
Thue system, 201 verifiable relation, 350
tolerance relation, 198 via, 25
torsion free semigroup, 46 V -exact sequence, 299
trace, 245,244--249,250,269,270,268-271 Volkov semigroup, 82,105,107,217,323,
- coideal, 290 352,375
- congruence, 245,269
triple, 296 weakly irreducible, 134,156
- cohomology, 298, 295-300 Weierstrass semigroup, 42
type, 43 well order, 328,329
word
union -problem, 27,143
empty-, 12 commutative - , 310
unique factorization nilmonoid, 210
unit, 18 0-free nilmonoid, 210,255
group of -s, 18 zero
unitary subsemigroup, 58 - class, 210,290
universal -element, 4
- cancellative semigroup, 34 -group, 366
- c.pc.r. semigroup, 47
Advances in Mathematics
1. A. Nagy: Special Classes of Semigroups. 2001 ISBN 0-7923-6890-8

2. P.A. Grillet: Commutative Semigroups. 2001 ISBN 0-7923-7067-8
SPRINGER-SCIENCE+BUSINESS MEDIA, B.V.

Libro - (Advances in Mathematics 2) P. A. Grillet (Auth.) - Commutative Semigroups-Springer US (2001) PDF

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Libro - (Advances in Mathematics 2) P. A. Grillet (Auth.) - Commutative Semigroups-Springer US (2001) PDF

Загружено:

Авторское право:

Доступные форматы

Commutative Semigroups

J. Szep, Budapest University of Economics, Hungary

G. Erjaee, Shiraz University, Iran

W. Fouche, University of South Africa, South Africa

P. Grillet, Tulane University, U.S.A.

H.J. Hoehnke, Germany

F. Szidarovszky, University of Arizona, U.S.A.

P. Zecca, Universita di Firenze, Italy

SPRINGER-SCIENCE+BUSINESS MEDIA. B.V.

ISBN 978-1-4419-4857-1 ISBN 978-1-4757-3389-1 (eBook)

Printed on acid-free paper

All Rights Reserved

General structure theory

Chapter VI. Finitely generated semigroups 141

5. Strand bases 278

constitutes Chapters IX, X, and XI, contains constructions of subcomplete semi-

Slidell, Louisiana, March 2001.

This chapter contains basic first properties of commutative semigroups: idem-

1. A semigroup is a set S together with an associative binary operation on

The number of semigroups of order 9 has not been determined; it is known to

The number of commutative semigroups of order 9 is from Grillet [ 1996N].

multiplicative terms. (Constructing rings from semigroups is even more fruitful;

for every permutation a- of {1, 2, ... , n}.

hold for all x, y E S and m, n > 0 ( m, n ~ 0 if S has an identity element).

in case the latter exists. Hence the following construction:

Accordingly, commutative semigroups in which every element is idempotent

the product of one or more elements of X. A semigroup is finitely generated

2. HOMOMORPHISMS AND CONGRUENCES.

1. Let S and T be semigroups. A homomorphism of semigroups of S into

ker c.p = {(a, b) E S x S I c.p(a) = c.p(b)}.

Ea.(Eb.Ec) = Ea.Ebc = Ea(bc) = E(ab)c = Eab·Ec = (Ea.Eb).Ec

and the projection s

Thus ~ is a mapping ofT into U. Also ~ (cp(a)) = 1/J(a) for all a E S by

cp- 1 (T 1) = {xES I cp(x) E T}

If T' is a subsemigroup of T, then cp -l (T') is a union of e-classes and is

equivalently, the direct image of c ~ S x S under cp x cp : S x S ---+ T x T.

6. Similar results hold for monoids. When S and T are monoids, a

congruence e on S which contains ~; e is the intersection of all the congruences

Then a e b if and only if there exist n ~ 1 and s 1 , ... , sn E S such that a = s 1 ,

1. An ideal of a semigroup S is a subset I of S such that a E I implies

Proof. 8 1 X is the product in 8 1 , which is contained in S since either

semigroup E such that S is an ideal of E and Q is the Rees quotient Q = E / S.

a*b abE Q ifab # 0 in Q,

for all a, bE Q\ 0 and x, y E S, is a retract ideal extension of S by Q, and every

1. A preorder (also called quasiorder) is a binary relation which is reflexive

a ~:J-C b ~ a= tb for some t E S 1 ~ S 1a ~ S 1 b.

a JC b ~ a ~:J-C b and b ~:J-C a ~ S 1 a = S 1 b.

Proof. If a JC b, then a = tb and b = ua for some t, u E S 1 , ae = tbe and

Proof. Let hE H. Then thE H, h = uth for some u E 8 1 , and uH s;;; H,

Proposition 4.4. In a commutative monoid S, S\H1 is an ideal.

Proposition 4.6. For every 1-C-class H, r(H) = {gt I t E St (H)} is a

If a, b E H, then b = ta for some t E 8 1 , t E St (H) since 1-C is a

r(H) is the (left) SchOtzenberger group of H; it was discovered by Schiit-

5. FREE COMMUTATIVE SEMIGROUPS.

1. When a commutative semigroup S is generated by a subset X, every

l::xEX axx + l::xEX bxx = l::xEX (ax+ bx)x ·

'ExEX ax x ~ 'ExEX bx x if and only if ax ~ bx for all x E X

equivalently, Fx is the set of all linear combinations a = l:xEX ax x with

'P(l:xEX axx) = ITxEX f(xtx ·

<p (al xl + a2x2 + ... + anxn) = c.p(xltl c.p(x2t2 ... c.p(xn)an .

'P(l:xEX axx) = ITxEX c.p(xtx = ITxEX f(xtx

Conversely define a mapping t.p : Fx ~ S by

t.p(L:xEX axx) = flxEX f(xtx