Академический Документы
Профессиональный Документы
Культура Документы
Advances in Mathematics
VOLUME2
Series Editor:
Advisory Board:
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.
Preface . . . . . . . . . . . . . . • • . . XI
vii
viii CONTENTS.
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
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.
ELEMENTARY PROPERTIES.
1. FIRST RESULTS.
1
2 I. ELEMENTARY PROPERTIES.
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:
8 I. ELEMENTARY PROPERTIES.
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) =
2. HOMOMORPHISMS AND CONGRUENCES. 9
<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;.
10 I. ELEMENTARY PROPERTIES.
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 };
2. HOMOMORPHISMS AND CONGRUENCES. 11
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
3. IDEALS.
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
16 I. ELEMENTARY PROPERTIES.
4. DIVISIBILITY.
(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
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.
Fx={aEGxla>O};
5. FREE COMMUTATIVE SEMIGROUPS. 21
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:
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,
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
24 I. ELEMENTARY PROPERTIES.
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.
Then r.p o t = f.
CANCELLATIVE SEMIGROUPS.
1. SEMIGROUPS OF FRACTIONS.
29
30 II. CANCELLATIVE SEMIGROUPS.
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
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
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
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)
34 II. CANCELLATIVE SEMIGROUPS.
r
s ~ s1e
~lY
G *-x G(Sie)
for some y,t E S. Hence axyt = xt and xla = xytlt = 1(xy). Thus 1 is
surjective.
3. CANCELLATIVE SEMIGROUPS.
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.
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
42 II. CANCELLATIVE SEMIGROUPS.
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
44 II. CANCELLATIVE SEMIGROUPS.
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 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.
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
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
equivalently, xi = 'PI (xk+I, ... , xn), ... , xk = 'Pk(xk+I, ... , xn), where
6. FACES.
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.
52 II. CANCELLATIVE SEMIGROUPS.
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
54 II. CANCELLATIVE SEMIGROUPS.
7. FREE EMBEDDING.
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
56 II. CANCELLATIVE SEMIGROUPS.
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.
(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.
60 II. CANCELLATIVE SEMIGROUPS.
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) .
(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).
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.
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,
ek = oa 1k 1\ ... 1\ oark E F
has the following properties. The k-th coordinate of ek is
SEMILATTICE DECOMPOSITIONS.
1. GENERAL RESULTS.
69
70 Ill. 8EMILATTICE DECOMPOSITIONS.
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~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
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.
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,
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
/
X---+ T
1~
K,
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 ).
(a, b I a 2 b4 = a 5 b2 ) = ( a, b I b2 = a3 )
( 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.
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
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,
'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);
SUBDIRECT DECOMPOSITIONS.
1. SUBDIRECT PRODUCTS.
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
2. SEPARATIVE SEMIGROUPS.
by Fulp & Hill [ 1965] and Pondelicek [ 1966] for semi characters. More general
forms of duality were investigated by Lesohin [ l966D], [ l968D].
3. NILSEMIGROUPS.
N = ( a, b I a 3 = a 2 b = a b2 = b3 = 0 ) .
Therefore the 'Y-classes of N are: {a, b}, { a 2 , ab, b2 } , and { 0} . We see that
N /'Y is cyclic. D
Example 3.4. Let N be the commutative nilsemigroup
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.
106 IV. SUBDIRECT DECOMPOSITIONS.
4. PONIZOVSKY DECOMPOSITIONS.
(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.
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-
4. PONIZOVSKY DECOMPOSITIONS 109
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.
(2) for every xES there exists a greatest e E E(S 1 ) such that 1re(x) #- 0
in~·
4. PONIZOVSKY DECOMPOSITIONS 111
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
(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.
112 IV. SUBDIRECT DECOMPOSITIONS.
5. ELEMENTARY SEMIGROUPS.
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.
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.
1. DIVIDING BY JC.
115
116 V. GROUP COEXTENSIONS.
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'){.
2. SCHfrTZENBERGER FUNCTORS.
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.
118 V. GROUP COEXTENSIONS.
St(A) ~ St{B)
gAl lgB
r(A) 7a r(B)
commutes; then
(ga) s = (r~g)(as)
3. COEXTENSIONS.
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
(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.
122 V. GROUP COEXTENSIONS.
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
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
'
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,
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
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
(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:
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)
((g,a)(h,b))(k,c)
(g,a)((h,b)(k,c))
(g bc s a,bc hac sab,c kab ' a (be)) ·
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)
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)
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
4. GROUP COEXTENSIONS 131
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
132 V. GROUP CO EXTENSIONS.
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. 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
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.
5. 8UBDIRECTLY IRREDUCIBLE 8EMIGROUPS 135
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
136 V. GROUP COEXTENSIONS.
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;
138 V. GROUP CO EXTENSIONS.
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.
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
2. SUBDIRECT DECOMPOSITIONS.
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.
3. SUBELEMENTARY SEMIGROUPS.
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
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.
150 VI. FINITELY GENERATED SEMIGROUPS.
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
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
4. THE COMPLETION THEOREM. 153
(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;
5. IRREDUCIBLE SEMIGROUPS.
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
All these semigroups are finite; in the last case S is elementary with finitely
many finite orbits. Thus:
6. ARCHIMEDEAN SEMIGROUPS.
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
160 VI. FINITELY GENERATED SEMIGROUPS.
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.
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.
162 VI. FINITELY GENERATED SEMIGROUPS.
SUBCOMPLETE SEMIGROUPS.
1. COMPLETIONS.
165
166 VII. SUBCOMPLETE SEMIGROUPS.
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}
168 VII. SUBCOMPLETE SEMIGROUPS.
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.
<;;;;
2. PONIZOVSKY FAMILIES.
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.
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.
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
'ljJ (ajk) a k -1 .
Thus 'ljJ = r.p. D
4. PROPERTIES.
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:
Example 4.5. Let S = { xn I n ~ 0} be the free c.m. with one generator
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.
St(K) ~ St(H)
aKl l~H
r(K) o;:+ r(H)
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:
~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)
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.
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
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.
[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-
194 VIII. OTHER RESULTS.
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]).
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].
3. HOMOMORPHISMS AND CONGRUENCES. 197
4. OTHER TOPICS.
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.
203
204 IX. NILSEMIGROUPS
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:
(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
b, c -=/= a; thus a is not irreducible. This proves the first part of the statement.
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.
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.
.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 ).
.k
Example 3.6
Example 3.7
4. NESTS.
N = ( x,y I x 3 = x 2 y = xy 2 = y4 = 0, x 2 = xy = y 3 ).
.k
.m
Example 4.3
(2) et ~ 'Dt;
(3) if k,l,m,n E K and k -l = m- n, then, for all t E H,
5. EXAMPLES.
.k
Example 5.1
222 IX. NILSEMIGROUPS
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
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:
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
~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:
~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(
GROUP-FREE SEMIGROUPS.
1. SEMILATTICE CONGRUENCES.
227
228 X. GROUP-FREE SEMIGROUPS.
(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
230 X. GROUP-FREE SEMIGROUPS.
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.
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' · · ·} ·
232 X. GROUP-FREE SEMIGROUPS.
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.
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).
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}),
236 X. GROUP-FREE SEMIGROUPS.
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.
·I ·I·
·I· I·
Example A: extent cells Example B: extent cells
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
Proposition 3.3. Let e be the c.gj congruence on F induced by a surjective
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:
Proposition 3.4. Let e be the c.gj congruence on F induced by a surjective
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
3. EXTENT CELLS. 239
' '
·I ·I·
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'.
(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'.
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
242 X. GROUP-FREE SEMIGROUPS.
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.
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,
244 X. GROUP-FREE SEMIGROUPS.
En, and E0 .
4. TRACE CONGRUENCES.
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:
246 X. GROUP-FREE SEMIGROUPS.
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
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
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).
248 X. GROUP-FREE SEMIGROUPS.
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
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
(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)
5. MAIN RESULT.
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
Example 5.4
(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
254 X. GROUP-FREE SEMIGROUPS.
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.
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
256 X. GROUP-FREE SEMIGROUPS.
(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"
(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 ( ~) .
SUBCOMPLETE SEMIGROUPS.
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.
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};
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
262 XI. SUBCOMPLETE SEMIGROUPS.
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-
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
2. EXTENT CELLS. 267
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).
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.
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
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
272 XI. SUBCOMPLETE SEMIGROUPS.
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
Example A Example C
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
Proposition 4.4. Let e be the congruence induced on F by a surjective
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.
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
4. STRAND GROUPS. 277
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.
5. STRAND BASES.
c E f*(g) if and only if c E (EA :g+) n (EA :g-) njA(g) for some A E 'D,
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.
(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
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
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},
Example C
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,
288 XI. SUBCOMPLETE SEMIGROUPS.
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.
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);
290 XI. SUBCOMPLETE SEMIGROUPS.
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
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
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.
(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
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
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
0---+ C 1(-
T,G-) ---+ ... ---+ C n(-
T,G-) -----'8n-'--)- C n+I(-
T,G-) ---+ ...
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
P~G
P2l f
G -----+ S
p
-+-
Hom-e(Na ,H) ~ Ha
where a = 1rc and b = 1rd, and 9ed = g~ + gd; similarly hed = h~ + hd. Hence
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
3. BECK EXTENSIONS.
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:
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.
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
for all [x 1, ... ,xml E Tn+ 1 , by Lemma 4.1; when i < n, E~,i = V E~- 1 , i and
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)
sends v : T n -+ G to
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
(8)
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
'Pa(t,z) = 'Y'Trt ,z ut
for all (t, z) E X a. Then
for all (t,z) E Xa, where t = [x1, ... ,xml E Tn and xj = X1···Xj-1 xj+1
···Xm.
<fJa(t,z) = l1rt,z ut = ut
~
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.
318 XII. CoMMUTATIVE SEMIGROUP CoHOMOLOGY.
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 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.
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; 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).
320 XII. COMMUTATIVE SEMIGROUP COHOMOLOGY.
(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.
(Z")
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
Therefore ~u E S Z 2 ( S, G) .
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
' '
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
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.
holds in Gxyz for all x,y,z E S. We note that A(x,y,z) is trivial if x = z and
324 XII. COMMUTATIVE SEMIGROUP COHOMOLOGY.
1. OVERPATHS.
327
328 XIII. THE OVERPATH METHOD.
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
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, 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).
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.
. 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
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
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)
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 ·
u "' r 9 1r(r-x)
wxEX, x~r X X
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
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.
by (S+) in Lemma XI.6.l. Adding these equalities yields sa- sb- a+ b- sm1-
···-skER
m a· D
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
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).
338 XIII. THE OVERPATH METHOD.
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'
' '
'"'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
(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.
za = "'
~xEX, x:S;a ax 9xa-x E G1ra ·
E G1ra,
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.
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
ua· p+q· c
' '
=
' '
ua·p·b
,'
+ ub·q·c ·
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
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
4. DEFINING VECTORS.
a-b
We write this equality as
a- b L:mEM Pm v(m),
348 XIII. THE OVERPATH METHOD.
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}.
350 XIII. THE 0VERPATH METHOD.
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;
4. DEFINING VECTORS. 351
'\" 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
S = ( c,d I c3 = c2 d = cd2 = d 4 = 0, c2 = cd = d 3 );
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
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
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
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,
( **) 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
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 ·
( 8v ) X "'""' 7rm(y)
= vx + L.JyEX, tx,yoiO tx,y 'Y7rm(x) vy .
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.
(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)?
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.
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 ).
with pkv(mi) ::::J 0 in G since pkmi ::::J pkq(mi) for all mi E Me.
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
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
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.
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
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).
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
3. NILMONOIDS.
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
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.
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
372 XIV. SEMIGROUPS WITH ZERO CoHOMOLOGY.
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
5
" ,,
ql q3
0
:~
. . . 1
• • • • m2
0 4
Example 3.5
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
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.
378 XIV. SEMIGROUPS WITH ZERO CoHOMOLOGY.
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.
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.
379
380 COMMUTATIVE SEMIGROUPS.
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.!,
Amer. Math. Soc., Providence RI, 1961.
[ 1967] The algebraic theory of semigroups, val. II,
Amer. Math. Soc., Providence RI, 1967.
Comfort, W. W. [ 1959] The isolated points in the dual of a commutative semigroup,
Proc. Amer. Math. Soc. 11 (1960) 227-233.
Comfort, W.W. & Hill, P. [1966] On extending nonvanishing semicharacters,
Proc. Amer. Math. Soc. 17 (1966) 936-941.
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
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,
J. Reine Angew. Math. 293 (1977) 1-17.
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,
Czechoslovak Math. J. 26 (101) (1976) 477-505.
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
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,
Semigroup Forum 18 (1979) 371-380.
[ 1981] A remark on polynomial JUnctions over finite monoids,
Semigroup Forum 22 (1981) 391-392.
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.
414 COMMUTATIVE SEMIGROUPS.
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,
Manuscripta Math. 30 (1980) 257-278.
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,
Czechoslovak Math. J. 11 (1961) 150-155.
Warner, S. [1960] Mathematical induction in commutative semigroups,
Amer. Math. Monthly 67 (1960) 533-537.
Wegmann, H. [ 1966] Beitrage zur Zahlentheorie auffreie Halbgruppen. II.
J. Reine Angew. Math. 221 (1966) 150-159.
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.
416 COMMUTATIVE SEMIGROUPS.
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,
Czechoslovak Math. J. 27 (102) (1977) 167-169.
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
417
418 COMMUTATIVE 8EMIGROUPS
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,
425
426 NOTATION
'fl, 296
L, canonical injection
v, 296
1r, projection
7rn , 298,311
I;(K), 182
E, 185
INDEX
427
428 COMMUTATIVE 8EMIGROUPS