Inverse Semi Groups

Automatic S-acts and inverse semigroup presentations
Erzsebet Rita Dombi

Ph.D. Thesis
University of St Andrews
September, 2004
ii
To Zsolt
iv
Contents
Declaration vii
Abstract ix
Acknowledgements xi
1 Introduction 1
2 Preliminaries 5
2.1 Semigroups . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.2 Regular languages and nite state automata . . . . . . . . . . . . . . . . . . 11
2.3 Automatic groups and semigroups . . . . . . . . . . . . . . . . . . . . . . . 14
3 Automatic semigroup acts 19
3.1 Denitions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Sch utzenberger automatic regular semigroups . . . . . . . . . . . . . . . . . 25
3.3 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.4 Basic properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.5 Automatic versus Sch utzenberger automatic. . . . . . . . . . . . . . . . . . 47
3.6 Invariance under the change of generators. . . . . . . . . . . . . . . . . . . . 49
3.7 Fellow traveller property I. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
3.8 Fellow traveller property II. . . . . . . . . . . . . . . . . . . . . . . . . . . . 64
3.9 Fellow traveller property III. . . . . . . . . . . . . . . . . . . . . . . . . . . . 74
3.10 Finite presentability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
3.11 Equality and word problems . . . . . . . . . . . . . . . . . . . . . . . . . . . 78
3.12 Inverse free product of inverse semigroups. . . . . . . . . . . . . . . . . . . . 84
v
vi CONTENTS
4 Semidirect product 95
4.1 Presentations. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 95
4.2 Finite generation with respect to an action. . . . . . . . . . . . . . . . . . . 99
4.3 Finite generation. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 106
4.4 Finite presentability with respect to an action I. . . . . . . . . . . . . . . . 108
4.5 Examples I. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 119
4.6 Finite presentability with respect to an action II. . . . . . . . . . . . . . . . 129
4.7 Examples II. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 133
4.8 Matched and conjugate words . . . . . . . . . . . . . . . . . . . . . . . . . . 140
4.9 Finite presentability I. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143
4.10 Finite presentability II. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 150
4.11 Sch utzenberger automaticity . . . . . . . . . . . . . . . . . . . . . . . . . . 161
5 HNN extensions 167
5.1 Group HNN extensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 168
5.2 Yamamuras HNN extension . . . . . . . . . . . . . . . . . . . . . . . . . . . 170
5.3 Gilberts HNN extension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172
5.4 An alternative presentation for Gilberts HNN extension . . . . . . . . . . . 175
5.5 Finite generation of Gilberts HNN extension . . . . . . . . . . . . . . . . . 177
5.6 Finite presentability of Gilberts HNN extension . . . . . . . . . . . . . . . 179
5.7 A compelling consequence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
5.8 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
Declaration
I, Erzsebet Rita Dombi, hereby certify that this thesis, which is approximately 66000
words in length, has been written by me, that it is the record of work carried out by me
and that it has not been submitted in any previous application for a higher degree.
Signature. . . . . . . . . . . . . . . . . . . . . . . . . . . .Name Erzsebet Rita Dombi Date 03/09/04
I was admitted as a research student in September 2001 and as a candidate for the degree
of Ph.D.; the higher study for which this is a record was carried out in the University of
St Andrews between 2001 and 2004.
I hereby certify that Erzsebet Rita Dombi has fullled the conditions of the Resolutions
and Regulations appropriate for the degree of Ph.D. in the University of St Andrews and
that the candidate is qualied to submit this thesis in application for that degree.
Signature . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . Name Nik Ruskuc Date 03/09/04
In submitting this thesis to the University of St Andrews I understand that I am giving
permission for it to be made available for use in accordance with the regulations of the
University Library for the time being in force, subject to any copyright vested in the work
not being aected thereby. I also understand that the title and abstract will be published,
and that a copy of the work may be made and supplied to any bona de library or research
worker.
vii
viii DECLARATION
Abstract
To provide a general framework for the theory of automatic groups and semigroups, we
introduce the notion of an automatic semigroup act. This notion gives rise to a variety
of denitions for automaticity depending on the set chosen as a semigroup act. Namely,
we obtain the notions of automaticity, Sch utzenberger automaticity, R- and L-class auto-
maticity, etc. We discuss the basic properties of automatic semigroup acts. We show that
if S is a semigroup with local right identities, then automaticty of a semigroup act is inde-
pendent of the choice of both the generators of S and the generators of the semigroup act.
We also discuss the equality problem of automatic semigroup acts. To give a geometric
approach, we associate a directed labelled graph to each S-act and introduce the notion
of the fellow traveller property in the associated graph. We verify that if S is a regular
semigroup with nitely many idempotents, then Sch utzenberger automaticity is charac-
terized by the fellow traveller property of the Sch utzenberger graph. We also verify that
a Sch utzenberger automatic regular semigroup with nitely many idempotents is nitely
presented. We end Chapter 3 by proving that an inverse free product of Sch utzenberger
automatic inverse semigroups is Sch utzenberger automatic.
In Chapter 4, we rst introduce the notion of nite generation and nite presentability
with respect to a semigroup action. With the help of these concepts we give a necessary
and sucient condition for a semidirect product of a semilattice by a group to be nitely
generated and nitely presented as an inverse semigroup. We end Chapter 4 by giving a
necessary and sucient condition for the semidirect product of a semilattice by a group
to be Sch utzenberger automatic.
Chapter 5 is devoted to the study of HNN extensions of inverse semigroups from nite
generation and nite presentability point of view. Namely, we give necessary and sucient
conditions for nite presentability of Gilberts and Yamamuras HNN extension of inverse
semigroups. The majority of the results contained in Chapter 5 are the result of a joint
work with N.D. Gilbert and N. Ruskuc.
ix
x ABSTRACT
Acknowledgements
First of all I would like to thank my supervisor Professor Nik Ruskuc for all his help and
support during the last three years. He shared his great mathematical ideas, gave me
interesting problems to work on and always had time for discussing them. He carefully
read the drafts of this thesis and gave helpful advice on how I could improve my notation
system. He not only guided me in mathematics but shared his experience in how to be
an alien.
I wish to thank Dr Nick Gilbert for the inspiring discussions and for his encouragement.
I had an excellent time during collaboration with Nick Gilbert and Nik Ruskuc.
Next, I would like to thank Professor M aria Szendrei for all her support. She awoke
my interest in semigroup theory, gave hard and interesting problems and supported me in
coming to St Andrews.
Thanks goes to the Ph.D. students who made the department cheerful. I am more
than grateful to Alan Cain and Peter Gallagher for reading various parts of my thesis and
to Maja Waldhausen for all her help and support.
I wish to thank my husband Zsolt for all his patience during these years and his help
in creating the pictures in the thesis.
xi
xii ACKNOWLEDGEMENTS
Chapter 1
Introduction
In order to obtain certain properties of (innite) semigroups or groups quickly and e-
ciently with the help of computers, encoding them with nitely many pieces of information
is important. One way of encoding a semigroup or a group is to nd generators and rela-
tions between words on the generators that describe the semigroup or group in question.
This is the subject of the theory of semigroup and group presentations. Another way of
encoding is in terms of generators and nitely many nite state automata. This approach
led to the theory of automatic groups and semigroups.
The theory of automatic groups was introduced in the early 1980s by Epstein et al
[9]. The main aim was to introduce a large class of groups having solvable word problem.
Automatic groups have many pleasant properties. They are nitely presented, their auto-
maticity does not depend on the choice of the nite generating set, and most importantly
they are characterized by a geometric property, called the fellow traveller property of their
Cayley graphs.
That the (language theoretic) notion of an automatic group can be naturally extended
to semigroups was observed by Campbell et al [5] and the study of automatic semigroups
was initiated. Investigation began into whether properties of automatic groups carry
over in the case of automatic semigroups [5]. Automatic semigroups have solvable word
problem. On the other hand, it turned out that automatic semigroups behave more wildly
than automatic groups. Automatic semigroups are not necessarily nitely presented [5,
Example 4.4], automaticity of semigroups does depend on the choice of generators [5,
Example 4.5]. Moreover a geometric characterization of automatic semigroups with the
help of their Cayley graphs does not exists.
The response to the above results was twofold. On one hand alternative notions of
automaticity have been considered for semigroups [34], [19]. On the other hand semigroup
1
2 CHAPTER 1. INTRODUCTION
classes closely related to groups have been considered. For example, regarding automatic
completely simple semigroups, the group theoretic results generalize. That is, automatic
completely simple semigroups are nitely presented, automaticity does not depend on the
choice of generators and fellow traveller property of their Cayley graph characterizes them.
In [12] it is proved that automaticity of monoids does not depend on the choice of the
generating set.
The main purpose of Chapter 3 is to give a general framework for the theory of auto-
matic semigroups and groups and to extend the known results. In Section 1, we introduce
the notion of an automatic semigroup act or automatic S-act for short. Choosing the
S-act to be a set closely related to the semigroup S, we arrive at a wide variety of notions
of automaticity. If the S-act is the semigroup S, then we obtain the usual notion of au-
tomaticity. By choosing the S-act to be S/R or S/L, we introduce the notion of R-class
and L-class automaticity. With the assumption that S is a regular semigroup and the
S-act is an R-class of S we introduce the notion of Sch utzenberger automaticity. Com-
bining Sch utzenberger automaticity and R-class automaticity we introduce the notion of
a strongly Sch utzenberger automatic semigroup. In Section 3, we illustrate these notions
by giving examples.
In Sections 4 and 5, we discuss basic properties of automatic S-acts and compare the
notion of automaticity and Sch utzenberger automaticity of regular semigroups.
In Section 6, we show that if S is a semigroup with local right identities, then auto-
maticity of an S-act is independent of the choice of generators of S and also of the choice
of generators of the S-act. A consequence of this result is that automaticity of a regular
semigroup is independent of the choice of generators.
In Sections 7-9 we discuss a geometric approach. To each S-act we associate a directed
labelled graph and introduce the notion of the fellow traveller property in that graph. We
show that if an S-act is automatic then the fellow traveller property holds in the associ-
ated graph. In Section 8, we investigate the conditions under which the converse holds.
In particular, we verify that if S is a regular semigroup with nitely many idempotent
elements, then the fellow traveller property of a Sch utzenberger graph of an R-class R
does characterize Sch utzenberger automaticity of R. In Section 9, we give an example
of a semigroup S with innitely many idempotents, in which there is an R-class whose
Sch utzenberger graph possesses the fellow traveller property but is not Sch utzenberger
automatic.
As mentioned above, automatic semigroups and groups have solvable word problem.
In Section 10, we introduce the equality problem for S-acts, and show that the equality
3
problem is solvable in automatic S-acts.
Free products of automatic groups and semigroups are automatic. In Section 11,
we verify that inverse free product of Sch utzenberger automatic inverse semigroups is
Sch utzenberger automatic. The proof of this result is based on the description of the
Sch utzenberger graphs of the inverse free product of two inverse semigroups given by Jones
et al [25]. The underlying tool of the results in [25] is Stephens procedure for constructing
Sch utzenberger graphs with respect to an inverse semigroup presentation [35].
An extensive study of semigroup presentations began in the last decade. One of the
main areas of research is the study of presentations of semigroup constructions [33], [1],
[7]. Naturally, the main purpose is to give a necessary and sucient condition for the
semigroup construction to be nitely presented. In the last two chapters we investigate
two inverse semigroup constructions: one of them is the semidirect product of a semilattice
by a group; the other is an HNN extension (both in the sense of Gilbert and in the sense
of Yamamura) of an inverse semigroup.
We begin Chapter 4 by introducing the concept of nite generation and nite pre-
sentability with respect to a semigroup action. With the help of these concepts we give a
necessary and sucient condition for a semidirect product of a semilattice by a group to
be nitely generated and to be nitely presented as an inverse semigroup. We end Chapter
4 by giving a necessary and sucient condition for a semidirect product of a semilattice
by a group to be Sch utzenberger automatic.
The construction of an HNN extension was rst introduced when Higman, Neumann
and Neumann studied embeddability questions of groups. The construction was rst
considered for semigroups by Howie [20]. In recent years, two alternative notions for HNN
extensions of inverse semigroups have been given. One of these is due to Yamamura [36],
the other one to Gilbert [14]. We consider nite presentability of HNN extensions of
groups and inverse semigroups both in the sense of Gilbert and in the sense of Yamamura.
Considering Yamamuras HNN extension we see that the group theoretic result generalizes.
The main result of Chapter 5 is to give a necessary and sucient condition for Gilberts
HNN extension of an inverse semigroup to be nitely presented. Sections 3-8 of Chapter
5 contain results of a joint work with N.D.Gilbert and N.Ruskuc, and are accepted for
publication.
4 CHAPTER 1. INTRODUCTION
Chapter 2
Preliminaries
We begin by reviewing three main mathematical subjects under consideration in this
thesis. These are: semigroup theory, the theory of nite state automata and regular
languages and the theory of automatic groups and semigroups. To give a detailed survey
on these subjects is beyond the scope of the thesis. Our aim is to introduce the basic
denitions and results that are going to be used in the third chapter. For further details
in semigroup theory we refer the reader to [22], in the theory of nite automata and regular
languages [28],[21] and in automatic groups and semigroup [9] and [5]. At the beginning
of each subsequent chapter we give a further brief introduction containing the necessary
denitions and results to make the chapter self-contained.
2.1 Semigroups
We summarize some basic denitions and results from semigroup theory we will refer to
in the thesis.
Basic denitions
A non-empty set S together with an associative binary operation, usually called mul-
tiplication is called a semigroup. Let T be a non-empty subset of a semigroup S. We say
that T is a subsemigroup of S, if T is closed with respect to the multiplication. If S is a
semigroup and X is a non-empty subset of S, then the intersection of all subsemigroups
of S containing X is the subsemigroup X generated by X. If X = S, then we say that
S is generated by X.
Let X be a non-empty set. Let X
+
denote the set of all nite words (x
1
. . . x
n
), where
x
i
X for all 1 i n. The integer 1 n is the length |w| of the word (x
1
. . . x
n
). The
5
6 CHAPTER 2. PRELIMINARIES
set X
+
can be equipped with the following associative operation, called concatenation:
(x
1
. . . x
n
)(y
1
. . . y
m
) = (x
1
. . . x
n
y
1
. . . y
m
).
The semigroup obtained is called the free semigroup on X. We recall [33, proposition 1.1].
Proposition 2.1.1. Let X be a non-empty set and S be a semigroup. Then every mapping
f : X S extends uniquely to a homomorphism : X
+
S. If S is generated by X,
then is surjective.
Let S be a semigroup generated by X S. Let : X
+
S denote the homomorphism
extending the identity map : X S. For any two words u, v X
+
, we write u v, if u
and v are identical as words, and we write u = v, if u and v represent the same element
of S, that is, if u = v.
Let S be a semigroup. We say that 1 S is an identity element, if 1s = s1 = s holds
for all s S. It is easy to see that a semigroup can have at most one identity element. If
a semigroup S contains an identity element, then we call S a monoid. If S does not have
an identity element, then a monoid can be formed from S in the following way. We choose
a symbol 1 / S and dene 1s = s1 = s for all s S and dene 11 = 1. The monoid
obtained is denoted by S
1
.
Let S be a semigroup with at least two elements. We say that 0 S is a zero element,
if 0s = s0 = 0 holds for all s S. A semigroup S can have at most one zero element. If S
does not have a zero element, then a semigroup with zero element can be obtained in the
following way. We choose a symbol 0 / S and dene 0s = s0 = 0 for all s S and dene
00 = 0. The semigroup obtained is denoted by S
0
.
We say that e S is an idempotent, if ee = e. The set of idempotents of S will be
denoted by E(S).
Let S be a semigroup and R be a relation on S. We say that R is left (right) compatible
with the multiplication, if whenever s, t S are such that s R t, then for all a S,
as R at, (sa R ta) holds. We say that R is compatible with the multiplication, if whenever
s
1
, s
2
, t
1
, t
2
S are such that s
1
R t
1
and s
2
R t
2
hold, then s
1
s
2
R t
1
t
2
holds. A left
(right) compatible equivalence relation is called a left (right) congruence, and a compatible
equivalence relation is called a congruence on S.
Before we give examples for a left and a right congruence on a semigroup S, we
introduce the following notions. Let A be a non-empty subset of a semigroup S. We
say that A is a left (right) ideal in S, if for all s S and a A, sa A, (as A) holds.
If A is a left and a right ideal in S, then we say that A is an ideal in S. If A is an ideal
2.1. SEMIGROUPS 7
of S so that A S, then say that S is a proper ideal of S. We say that S is simple, if
it has no proper ideals. If a S, then the smallest left (right) ideal of S containing a is
S
1
a = {sa | s S
1
} (aS
1
= {as | s S
1
}), which we call the principal left (right) ideal
generated by a. The smallest ideal containing a is S
1
aS
1
= {sat | s, t S
1
}, which we call
the principal ideal generated by a.
Let S be a semigroup. We dene the following equivalence relations on S. For all
s, t S, we dene s L t, if and only if s and t generate the same principal left ideal, that
is, if and only if S
1
s = S
1
t. Similarly, for all s, t S, we dene s R t, if and only if s and
t generate the same principal right ideal, that is, if and only if sS
1
= tS
1
. Clearly, L is
a right congruence and R is a left congruence on S. It can be easily seen that L and R
commutes, and so D = L R = R L is the smallest equivalence containing L and R. We
denote by H the intersection of L and R. If H is trivial, then we call S a combinatorial
or aperiodic semigroup. We introduce our nal equivalence relation. For all s, t S, we
dene s J t, if and only if s and t generate the same principal ideal, that is, if and only
if S
1
sS
1
= S
1
tS
1
. We note that S is a simple semigroup if and only if J = S S. It
is immediate that H R, L J, and so D J holds as well. We call the relations
L, R, H, D, and J Greens relations. We denote the L-class [R, H, D, J-class] of an
element s S by L
s
[R
s
, H
s
, D
s
, J
s
]. Since D = L R = R L, the D-class of s S, can
be visualized in a convenient way by the egg-box picture:
L
s
R
s
s
The box represents a D-class, namely D
s
, the rows correspond to R-classes, the columns
to L-classes, and each cell represents an H-class.
Lemma 2.1.2. Let S be a semigroup and e E(S). Then er = r for all r R
e
and
le = l for all l L
e
.
Proof. See [22, Proposition 2.3.3].
With the help of left and right translations, a bijection can be given between L-, R-
and H-classes of a semigroup contained within a D-class.
Lemma 2.1.3 (Greens lemma). Let S be a semigroup and s, t S so that s L t.
Assume that s = at and t = bs. Then the mappings
a
: R
s
R
t
; x bx and
b
: R
t
R
s
; y ay
are mutually inverse L-class preserving bijections.
Proof. See [22, Lemma 2.2.2].
A subgroup of a semigroup S is a subsemigroup, which is a group. With the help of
Greens lemma the following propositions can be proved.
Proposition 2.1.4. The maximal subgroups of a semigroup S coincide with the H-classes
of S which contain an idempotent. Each subgroup of S is contained in an H-class of S.
Proof. See [22, Corollary 2.2.6].
Proposition 2.1.5. Let S be a semigroup, and s, t S such that s D t. Then st R
s
L
t
if and only if L
s
R
t
contains an idempotent.
We illustrate Proposition 2.1.5 with the help of the egg-box picture:
L
t
L
s
R
s
st s
R
t
t e
Regular semigroups
Throughout the thesis we shall be interested in regular semigroups. We rst generalize
the group theoretic notion of an inverse. Let S be a semigroup. We say that s S is
regular, if there exists x S so that s = sxs. We say that x is an inverse of s, if s = sxs
and x = xsx hold. It can be easily seen, that if s S is regular and s = sxs, then
s
= xsx is an inverse of s. Thus, we may deduce that s S is regular if and only if s

has an inverse. An element can have more than one inverse. The set of all inverses of an
element will be denoted by V (s). We call S a regular semigroup, if every element of S is
regular. Note that if s
V (s), then ss
, s
s E(S) and ss
R s L s
s. Thus in a regular
semigroup, every L- and R-class contains an idempotent. Also we have
Proposition 2.1.6. If s S is regular, then every element of D
s
is regular.
2.1. SEMIGROUPS 9
Let s S, where S is a regular semigroup. The locations of the idempotents in D
s
determines the location of inverses of s. Namely, we have
Proposition 2.1.7. Let S be regular semigroup and s S. Then
(1) For all s
V (s), we have s D s
.
(2) For all s
V (s), we have ss
R
s
L
s
and s
s L
s
R
s
.
(3) If e, f E(S) and e R s L f, then there exists s
V (s) such that ss
= e and
s
s = f.
(4) No H-class contains more than one inverse of s.
Proof. See [22, Theorem 2.3.4].
We illustrate the results of the above proposition with the help of the egg-box picture:
L
a
L
a
R
a
a aa
a
a
a a
L
f
L
e
R
e
a e
R
f
f a
The Greens relations in a regular semigroup can be described in the following way:
Proposition 2.1.8. Let S be a regular semigroup and s, t S. Then
(1) s L t, if and only if s
s = t
t for some s
V (s), t
V (t).
(2) s R t, if and only if ss
= tt
for some s
V (s), t
V (t).
(3) s H t, if and only if s
s = t
t and ss
= tt
for some s
V (s), t
V (t).
Let S be a regular semigroup and s, t S. To determine an inverse of st in terms of
inverses of s and t, we need to introduce the notion of a sandwich set.
Denition 2.1.9. Let S be a regular semigroup. The sandwich set of e, f E(S) is the
non-empty set
S(e, f) = {h E(S) | h = he = fh, ehf = ef}.
Proposition 2.1.10. Let S be a regular semigroup, s, t S. Let s
V (s), t
V (t) and
h S(s
s, tt
). Then t
hs
V (st).
With the help of Proposition 2.1.5, the following proposition can be veried.
Proposition 2.1.11. Let S be a regular semigroup, e E(S) and consider G = H
e
. Let
s, t R
e
such that s H t. Then Gs = Gt hold.
We give a brief summary on certain regular semigroup classes that will appear in the
thesis.
We say that S is a completely regular semigroup, if every H-class in S is a group. A
completely regular simple semigroup is called a completely simple semigroup. It can be
veried, that if S is a completely simple semigroup, then eSe is a group for all e E(S).
Completely simple semigroups play an important role in the description of the structure
of completely regular semigroups. Namely, if S is a completely regular semigroup, then
S/J is a semilattice, and each J-class is a completely simple subsemigroup of S. In other
words, each completely regular semigroup is a semilattice of completely simple semigroups.
The Rees-Sushkevich Theorem, which we now introduce, describes the structure of com-
pletely simple semigroups, and so a ner picture can be obtained from completely regular
semigroups.
Proposition 2.1.12 (Rees-Sushkevich). Let G be a group, and I non-empty sets.
Let P = (p
i
) be a I matrix with entries in G. Let S = I G , and dene a
multiplication on S by
(i, g, )(j, h, ) = (i, gp
j
h, ).
Then S is a completely simple semigroup. Conversely, every completely simple semigroup
is isomorphic to a semigroup constructed in this way.
The semigroup I G with the given multiplication is denoted by M[G, I, ; P] and
is called a Rees matrix semigroup over G.
In a regular semigroup an element can have more than one inverse. We say that S is an
inverse semigroup, if every element has exactly one inverse. The inverse of an element s is
denoted by s
1
. It can be easily seen that if S is an inverse semigroup and s, t S, then
(st)
1
= t
1
s
1
, (s
1
)
1
= s hold.
2.2. REGULAR LANGUAGES AND FINITE STATE AUTOMATA 11
Proposition 2.1.13. Let S be a semigroup. Then the following are equivalent.
(1) S is an inverse semigroup.
(2) S is regular and E(S) forms a semilattice.
(3) Every R- and L-class contains exactly one idempotent.
There is a natural partial order dened on inverse semigroups, which possesses pleasant
properties. Let S be an inverse semigroup and s, t S. We say that s t if and only
if s = et for some e E(S). It can be easily seen, that s t if and only if s
1
t
1
.
Furthermore, if s
1
t
1
and s
2
t
2
, then s
1
s
2
t
1
t
2
.
2.2 Regular languages and nite state automata
We will often refer to a nite non-empty set A as an alphabet. We may wish to adjoin an
element to A
+
, called the empty-word, to form the free monoid A
. The length of the

empty-word will be dened to be 0. A subset of L A
will be called a language over A.

If L and K are languages over A, then L K, L K, L \ K, LK = {uv | u L, v K},
L
= {} L LL LLL . . . and L
+
= L LL LLL . . . are also languages over A.
We often write L +K instead of L K.
A deterministic nite state automaton is a quintuple A = (, A, , p, T), where is a
nite non-empty set called the set of states, A is an alphabet, : A is a function,
called the transition function, p is called the initial state, T is the set of terminal
or nal states. We extend the transition function to a mapping
: A
as follows:
(q, ) q; (q, a) (q, a); (q, aw)
((q, a), w),

where q , a A, w A
. We say that a word w = a

1
. . . a
n
over A is recognized
by the automaton A if
(p, w) T. The language L(A) recognized (or accepted) by the

automaton A is the set of all elements w A
that are recognized by A.

A deterministic nite state automaton A = (, A, , p, T) can be visualized as a di-
rected labelled graph, whose vertices are elements of and there is an arrow from state q
to state r with label a, if (q, a) = r. If w = a
1
. . . a
n
L(A), then we say that the path
p
a
1
q
1
a
2
. . .
a
n1
q
n1
an
r
is successful.
A non-deterministic nite state automaton is a quintuple A = (, A, , p, T), where
is a nite non-empty set called the set of states, A is a nite non-empty set called the
alphabet, : (A {}) P(), where P() denotes the set of all subsets of , is a
function, called the transition function, p is called the initial state and T . The
elements of T are called terminal or nal states. To dene the language accepted by this
automaton, we need the following notion. Let w be a word over A {}. The value of
w is the word obtained from w by deleting all occurrences of in w. A path in A is a
sequence
q
0
a
1
q
1
a
2
. . .
a
n1
q
n1
an
q
n
,
where q
i
(q
i1
, a
i1
), (1 i n) and we say that a
1
. . . a
n
is the label of the path. We
say that w is accepted by A, if there exists a path from the initial state to a nal state
whose label is w or a word with the same value as w. The language L(A) recognized (or
accepted) by the automaton A is the set of all elements w A
that are recognized by A.

Similarly to deterministic nite state automata, non-deterministic nite state automata
can be visualized as directed labelled graphs.
Let A be an alphabet. A regular expression over A is a special type of word over
A {, (, ), +, ,
}. Namely, , , a, where a A are regular expressions. If r, r

1
, r
2
are
regular expressions, then so are (r
1
+ r
2
), (r
1
r
2
) and (r)
. Every regular expression

is formed in this way. Each regular expression r denes a language L(r). Namely, we
dene L() = , L() = {}, L(a) = {a}, where a A. We dene L(r
1
+ r
2
) = L(r
1
) +
L(r
2
), L(r
1
r
2
) = L(r
1
)L(r
2
) and L(r
) = (L(r))
, where r, r
1
, r
2
are regular expressions.
A language dened by a regular expression is called a regular language. The rst major
result in automata theory is due to Kleene, Rabin and Scott. For the proof, see for example
Theorem 4.1.2 and Theorem 5.2.1 in [28].
Theorem 2.2.1 (Kleene, Rabin, Scott). Let A be an alphabet and L be a language
over A. Then the following are equivalent:
(1) L is recognized by a deterministic nite state automaton.
(2) L is recognized by a non-deterministic nite state automaton.
(3) L is a regular language.
The following lemma provides a necessary condition for a language to be regular.
Lemma 2.2.2 (Pumping lemma). Let L be an innite regular language over the al-
phabet A. Then, there exists a positive integer N such that every word w L with length
2.2. REGULAR LANGUAGES AND FINITE STATE AUTOMATA 13
greater or equal to N can be factorized as w = xyz, where |xy| N, |y| 1 and xy
n
z L
for all n 0.
Next, we recall Proposition 2.2 in [5].
Proposition 2.2.3. Let X and Y be nite sets. Then the following hold:
(1) , X
+
and X
are regular languages.

(2) Any nite subset of X
is a regular language.
(3) If K X
and L Y
are regular languages, then K L, K L, K L, KL, K
and K
rev
= {x
1
. . . x
n
| x
n
. . . x
1
K} are regular languages.
(4) If K X
is a regular language and : X

+
Y
+
is a semigroup homomorphism,
then K is a regular language.
(5) If L Y
is a regular language and : X

+
Y
+
is a semigroup homomorphism,
then L
1
Before we nish this section, we introduce the following notion of automata theory.
Denition 2.2.4. Let A be an alphabet and L be a language over A. Let v A
. The
left quotient of L by v is the language
K = {w A
| vw L}.
Similarly, the right quotient of L by v is the language
N = {w A
| wv L}.
The following lemma will prove useful.
Lemma 2.2.5. Let L be a regular language over the alphabet A and let v A
. Then the
left (right) quotient of L by v is also a regular language.
2.3 Automatic groups and semigroups
Let X be a nite set, and let $ be a symbol not contained in X. Let X(2, $) = ((X $)
(X $)) \ {($, $)}. Dene
X
: X
X(2, $)
by
(a
1
. . . a
n
, b
1
. . . b
m
)
X
=
_
_
_
(a
1
, b
1
) . . . (a
n
, b
n
) if n = m,
(a
1
, b
1
) . . . (a
n
, b
n
)($, b
n+1
) . . . ($, b
m
) if n < m,
(a
1
, b
1
) . . . (a
m
, b
m
)(a
m+1
, $) . . . (a
n
, $) if n > m.
We recall [5, Proposition 2.2].
Proposition 2.3.1. Let X be a nite set.
(1) If K, L X
are regular languages, then (K L)

X
(2) If U (X
)
X
is a regular language, then
{u X
| (u, v)
X
U for some v X
}
is also a regular language.
We will also make use of [5, Proposition 2.4] and of [5, Proposition 2.3] :
Proposition 2.3.2. If L is a regular language over X, then {(w, w) | w L}
X
is a
regular language over X(2, $).
Proposition 2.3.3. Let L and K be regular languages over X(2, $). Then the language
W = {(u, w) X
| there exists v X
such that (u, v)

X
L, (v, w)
X
K}
X
is regular.
Let S be a semigroup. Assume that the set X S generates S and let : X
+
S
denote the homomorphism extending the identity map : X S. As introduced, for any
two words u, v X
+
, we write u v, if u and v are identical as words, and we write
u = v, if u and v represent the same element of S, that is, if u = v. We say that the
word problem is solvable in S, if an algorithm can be given with the help of which it can
be decided whether two words over X represent the same element of S.
Denition 2.3.4. Let S be a semigroup generated by a nite set X. Let L be a regular
language over X. We say that (X, L) is an automatic structure for S, if the following
conditions hold:
2.3. AUTOMATIC GROUPS AND SEMIGROUPS 15
(1) L = S;
(2) L
=
= {(u, v) L L | u = v}
X
is a regular language;
(2) L
x
= {(u, v) L L | ux = v}
X
is a regular language for all x X.
If S has an automatic structure (X, L), then we say that S is an automatic semigroup.
We say that (X, L) is an automatic structure with uniqueness for S, if whenever u, v L
are such that u = v, then u v holds.
Considering groups as semigroups, we obtain the notion of an automatic group.
Denition 2.3.5. Let G be a group generated a nite set X as a semigroup. Let L be
a regular language over X. We say that (X, L) is an automatic structure for G, if the
following conditions hold:
(1) L = G;
(2) L
=
= {(u, v) L L | u = v}
X
(2) L
x
= {(u, v) L L | ux = v}
X
If G has an automatic structure (X, L), then we say that G is an automatic group. We
say that (X, L) is an automatic structure with uniqueness for G, if whenever u, v L are
such that u = v, then u v holds.
We now recall [5, Corollary 5.5].
Proposition 2.3.6. If S is an automatic semigroup (group), then there exists an auto-
matic structure with uniqueness.
Automatic groups can be described with the help of a geometric property of their
Cayley graphs. First, we summarize some basic notions concerning the Cayley graph of a
group.
Let G be a group generated by a set X. We assume that X is closed under taking
inverses. The right Cayley graph =
X
(G) of G is a directed labelled graph, whose
vertices are elements of G, and there is an arrow from g to h with label x precisely when
h = gx. Clearly, there is an arrow from g to h with label x precisely when there is an
arrow from h to g with label x
1
. We dene a directed path between two vertices g and h
of to be a sequence of edges:
g = g
0
x
1
g
1
x
2
g
2
. . . g
n1
xn
g
n
= h
and say that the length of the path is n. Clearly is a connected graph, and there exists a
directed path between any two vertices of . Let g, h G. We dene the distance d
(g, h)
between g and h to be the length of the shortest path connecting g and h. Before we
dene the fellow traveller property, we introduce the following notation. If u x
1
. . . x
m
and t 1, then we let
u(t) =
_
x
1
. . . x
t
if t m,
x
1
. . . x
m
if t m.
Denition 2.3.7. Let G be a group generated by a nite set X. Let L be a regular
language over X and assume that L = G. We say that the Cayley graph =
X
(G)
possesses the fellow traveller property with respect to L, if there exists a constant k N,
such that whenever u, v L are such that d
(u, v) 1, then d
(u(t), v(t)) k for all

t 1.
The following result yields a powerful tool in proving results about automatic groups.
Proposition 2.3.8. Let G be a group. Then G is automatic if and only if the fellow
traveller property holds in the Cayley graph of G with respect to some regular language L.
We recall the following nice properties of automatic groups.
Proposition 2.3.9. Let G be an automatic group and assume that (X, L) forms an au-
tomatic structure for G. Then the following hold:
(1) If Y also generates G as a semigroup, then there exists a regular language K over
Y such that (Y, K) forms an automatic structure for G.
(2) The group G is nitely presented.
(3) The word problem is solvable in quadratic time.
Proof. See [9, Theorem 2.4.1, Theorem 2.3.12, Theorem 2.3.10].
Next, we discuss some of the basic properties of automatic semigroups. Let S be a
group generated by a set X. The (right) Cayley graph
X
(S) = of S is a directed
labelled graph, whose vertices are elements of S, and there is an arrow from s to t with
label x precisely when s = tx. Certain subgraphs, called Sch utzenberger graphs of the
Cayley graph of S will play an important role in the thesis. Let R be an R-class of S.
The Sch utzenberger graph of R is the Cayley graph of S restricted to R. In other words,
2.3. AUTOMATIC GROUPS AND SEMIGROUPS 17
the vertices of the Sch utzenberger graph of R are elements of R, and there is an arrow
from s to t, (s, t R) with label x, if s = tx.
The aim of giving a geometric characterization of automatic semigroups with the help
of the Cayley graph raises a few question. First of all the Cayley graph of a semigroup
is not necessarily connected. Even, if two vertices s and t are in the same connected
component, then there does not necessarily exists a directed path between them. These
considerations lead to the decision to ignore the direction of the edges in the Cayley graph
of S. A path between two vertices s and t of is a sequence of edges:
s = s
0
x
1
s
1
x
2
s
2
. . . s
n1
xn
s
n
= t
such that either (s
i
, x
i
, s
i+1
) or (s
i+1
, x
i
, s
i
) is an arrow in the Cayley graph of S, and say
that the length of the path is n. For any two vertices s and t we dene the distance d
(s, t)
between s and t to be the length of the shortest path connecting s and t and say that the
distance is innite if s and t belong to dierent components of .
Denition 2.3.10. Let S be a semigroup generated by a nite set X. Let be the
Cayley graph of S with respect to X. Let L be a regular language over X and assume
that L = S. We say that possesses the fellow traveller property with respect to L, if
there exists a constant k N, such that whenever u, v L are such that d
(u, v) 1,
then d
(u(t), v(t)) k for all t 1.

Proposition 3.12 of [5] tells us:
Proposition 2.3.11. Let S be an automatic semigroup. Then the fellow traveller property
holds in the Cayley graph of S with respect to some regular language L.
The converse of Proposition 2.3.11 does not necessarily hold. The fellow traveller prop-
erty holds in any semigroup with a zero element. To be more precise, if S is a semigroup
with a zero element, then the distance between any two vertices in the Cayley graph of S
is less then or equal to two. Thus, to demonstrate that the converse of Proposition 2.3.11
does not hold, one needs to take a semigroup with zero that is not automatic. Such a
semigroup can be formed by adjoining a zero element to a group that is not automatic.
Automatic semigroups behave more wildly than automatic groups. An example is
given in [5, Example 4.4] for a semigroup that is automatic but not nitely presented.
Another example [5, Example 4.5] shows that a semigroup can be automatic with respect
to one generating set, but not automatic with respect to another one. On the other hand,
if we consider monoids, then automaticity is independent of the choice of the generators
[12]. Also, if we consider semigroup classes that are somehow related to groups, such as
consider completely simple semigroups, then the group theoretic results generalize [4].
Proposition 2.3.12. Let S be an automatic semigroup generated by the set X. Then the
word problem is solvable in quadratic time.
Proof. See [5, Corollary 3.7].
Consider the alphabet X = {x
1
, x
2
, . . . , x
n1
, x
n
} and choose an ordering x
1
< x
2
<
. . . < x
n1
< x
n
. Let u, v X
+
. We dene u < v if and only if u is shorter then v, or they
have the same length, and u comes before v in the lexicographical order. This ordering is
called the shortlex order on X
+
.
Proposition 2.3.13. Let (X, L) be an automatic structure for a semigroup S. Let <
denote the shortlex order on X
+
. Then the language
K = {u | if (u, v) L
=
then u < v}
is regular.
Proof. See [5, Proposition 5.4].
The following proposition will also prove useful.
Proposition 2.3.14. Let (X, L) be an automatic structure for a semigroup S and let
w X
+
. Then L
w
= {(u, v) | u, v L, uw = v}
X
Proof. See [5, Proposition 3.2].
The notion of a padded product of regular languages is introduced in [10]:
Denition 2.3.15. Let X be an alphabet and let M, N be regular languages over (X
)
X
. The padded product of M and N is the language
M N = {(u
1
u
2
, v
1
v
2
) | (u
1
, v
1
)
X
M, (u
2
, v
2
)
X
N}
X
.
Theorem 3.3 of [10] gives a sucient condition on the padded product of regular languages
to be regular.
Proposition 2.3.16. Let X be an alphabet and let M, N be regular languages over (X
)
X
. If there exists a constant c such that for any two words u, v X
the following
property holds:
(u, v)
X
M then ||u| |v|| c,
then the padded product M N is a regular language.
Chapter 3
Automatic semigroup acts
3.1 Denitions
Finite state automata are easy to handle computationally. Therefore, some recent research
has aimed at encoding algebraic structures in terms of nite state automata. However,
there are dierent ways to dene what it means for an algebraic structure to be automatic.
Epstein et al. developed a theory of automatic groups, Campbell et al. realized that the
notion of an automatic group can be naturally extended to semigroups and investigated
to what extent the theory of automatic groups can be generalized to semigroups. It
turned out that automatic semigroups do not have as nice properties as automatic groups,
which led researchers to modify the denition of an automatic semigroup, see for example
[19],[34].
We mention that Khoussainov and Nerode represent another viewpoint of dening an
algebraic structure automatic [26]. Automatic groups (semigroups) in the sense of Epstein
et al. (Campbell et al.), viewed as unary structures where the unary operations are
multiplication by the generators on the right are automatic in the sense of Khoussainov
and Nerode.
Our aim in this chapter is to give a general framework for the theory of automatic
semigroups introduced by Cambell et al. and to extend the known results. To achieve
this goal we are going to dene automaticity of a very simple structure, called an S-
act. One of the reasons we are going to work with S-acts is that semigroups and groups
can be naturally considered as S-acts and S-acts proved to be a useful tool in studying
semigroups. On the other hand, certain sets for example an R- or an L-class can be
associated with a semigroup S, which can be viewed as a S-acts. This will allow us to
introduce other notions of automaticity, for example Sch utzenberger automaticity, R-class
automaticity.
19
20 CHAPTER 3. AUTOMATIC SEMIGROUP ACTS
Throughout this chapter S will denote a semigroup, X a generating set for S. We
assume that X S and denote by : X
+
S the homomorphism extending the identity
map : X S.
Let S = X be a semigroup and A be a non-empty set. A right action (left action) of
S on A is a function f : AS A; (a, s) a.s such that (a.s).t = a.(st) ((a.s).t = a.(ts))
holds for all s, t S, and we say that A is a right S-act (left S-act). We say that A is an
S-act, if it is either a left or a right S-act. Although it might seem a little unorthodox to
write left actions on the right, we will do so. The main reason for this is that we stick to
the rule in this chapter that a product s
1
s
2
. . . s
n
, (s
i
S, 1 i n) is read from left to
right. If we would write left actions of S on A on the left, then considering (s
1
s
2
. . . s
n
).a
would make us read s
1
s
2
. . . s
n
from right to left.
The subset A
0
= {a
1
, a
2
, . . . , a
n
, . . .} of A is said to generate the S-act A, if
_
a
j
A
0
a
j
.S = A.
If A
0
can be chosen to be a nite set, then we say that the S-act A is nitely generated. If
A
0
can be chosen to be a one element set, then we say that the S-act A is cyclic. Clearly,
if A
0
generates the S-act A, then it might happen that there exists s, t S so that s = t
and a
i
.s = a
j
.t for some a
i
, a
j
A
0
. To generalize the notion and basic properties (for
example uniqueness of an automatic structure) of an automatic group or semigroup, we
would like to use only as many elements of S as are necessary to generate the S-act A.
The following straightforward lemma is a crucial observation that will lead to the dention
of an automatic S-act.
Lemma 3.1.1. Let A be an S-act. Then A
0
generates A if and only if there exist subsets
T
1
, T . . . , T
n
, . . . of S such that
_
a
j
A
0
a
j
.T
j
=
_
a
j
A
0
{a
j
.t | t T
j
} = A.
If there exist regular languages L
1
, L
2
, . . . , L
n
, . . . over X such that L
j
= T
j
, then we
say that A is regularly generated by A
0
and {L
1
, L
2
, . . . , L
n
, . . .}, and use the notation
A =
_
a
j
A
0
a
j
.L
j
= {a
j
.l | l L
j
}.
Following semigroup and group theoretical conventions, we will also write a
i
.u = a
j
.v and
a
i
.(u x) = a
j
.v, (u, v) L
i
L
j
instead of a
i
.u = a
j
.v and a
i
.(u x) = a
j
.v, since
3.1. DEFINITIONS 21
the context will always make it clear over which alphabet the regular languages are being
taken, and what the homomorphism from the set of all words over that alphabet to S is.
Before we give the denition of an automatic S-act A, we give examples for right and
left S-acts.
Example 3.1.2. Let S be a semigroup and a right congruence on S. Dene f : S/S
S/; (s, t) (st). Since is a right congruence, f is a well dened function, and clearly
denes a right action of S on the factor set S/.
Example 3.1.3. Let S be a semigroup and a left congruence on S. Dene f : S/ S
S/; (s, t) (ts). Since is a left congruence, f is a well dened function, and clearly
denes a left action of S on the factor set S/.
Example 3.1.4. Any semigroup S can be considered as a right S-act, where the action is
right multiplication, that is f : SS S, (s, t) st. For convenience, when considering
a semigroup S as a right S-act, we write s t instead of s.t.
Example 3.1.5. Let S be a semigroup. Adjoin a zero element 0 to S and extend multi-
plication of S to S
0
= S {0} in the usual way; s0 = 0s = 0 for all s S and 0 0 = 0.
Let T be a subset of S with the condition that if
t (s
1
s
2
) T, where t T and s
1
, s
2
S, then t s
1
T. (3.1)
Let T
0
= T {0}. Then one can naturally consider T
0
as a right S
0
-act by dening
f : T
0
S
0
T
0
as:
(t, s)
_
t s if ts T,
0 otherwise.
To verify that (t s
1
) s
2
= t (s
1
s
2
) indeed holds for all s
1
, s
2
S, we consider the
following two cases:
(i) If t s
1
/ T, then t s
1
= 0 and it follows that (t s
1
) s
2
= 0 s
2
= 0 for all s
2
S.
On the other hand, by condition (3.1), t (s
1
s
2
) / T, hence t (s
1
s
2
) = 0, proving
that t (s
1
s
2
) = (t s
1
) s
2
.
(ii) If t s
1
T, then (t s
1
) s
2
= t (s
1
s
2
) is immediate.
Our nal example is a special case of the last example. This important subcase is the
basis of the denition of a Sch utzenberger automatic regular semigroup.
Example 3.1.6. Let S be a semigroup and adjoin a zero element 0 to S as in Example
3.1.5. Let R be a R-class of S. We verify that R fullls (3.1), hence a right action of S
0
on R
0
can be dened as in Example 3.1.5. Let r R, s
1
, s
2
S, such that r (s
1
s
2
) R.
Then r R rs
1
s
2
and by the denition of R, there exists t S such that r = rs
1
s
2
t, that
is r = rs
1
s
2
t. On the other hand rs
1
= r s
1
, verifying that r R rs
1
.
Next we introduce the notion of an automatic S-act A.
Denition 3.1.7. Let S be a semigroup generated by a nite set X. Let A be an S-act
and A
0
= {a
1
, a
2
, . . . , a
n
} be a nite subset of A. Let L
1
, L
2
, . . . , L
n
be regular languages
over X. We say that (X, L
1
, . . . , L
n
) forms an automatic structure for A with respect to
the generating set A
0
, if the following conditions hold:
(1) A is regularly generated by A
0
and {L
1
, . . . , L
n
}, that is:
n
_
j=1
a
j
.L
j
= A;
(2) L
(a
i
,a
j
)=
= {(u, v) L
i
L
j
| a
i
.u = a
j
.v}
X
is a regular language for all (a
i
, a
j
)
A
0
A
0
;
(3) L
(a
i
,a
j
)x
= {(u, v) L
i
L
j
| a
i
.(u x) = a
j
.v}
X
is a regular language for all
(a
i
, a
j
) A
0
A
0
and x X.
If an S-act A has an automatic structure (X, L
1
, . . . , L
n
) with respect to the generating set
A
0
, then we say that A is automatic with respect to the generating set A
0
. An automatic
structure (X, L
1
, . . . , L
n
) with respect to the generating set A
0
of an S-act A is said to be
with uniqueness, if for all a
i
, a
j
A
0
, a
i
.u = a
j
.v, ((u, v) L
i
L
j
) implies a
i
= a
j
and
u v.
The following two lemmas will prove useful. Both of the lemmas and their proofs are
based on [5, Proposition 3.1.].
Lemma 3.1.8. Let (X, L
1
, L
2
, . . . , L
n
) be an automatic structure for the S-act A with
respect to the generating set A
0
= {a
1
, a
2
, . . . , a
n
}. Let b A and x X. Then for each
i, (1 i n), the language {w L
i
| a
i
.(w x) = b} is regular.
3.1. DEFINITIONS 23
Proof. Since

n
j=1
a
j
.L
j
= A, there exists and hence we can x a language L
k
and u L
k
such that a
k
.u = b. Choose a language L
i
. It might happen that {w
L
i
| a
i
.(w x) = b} = , in which case it is a regular language by denition. Assume that
{w L
i
| a
i
.(w x) = b, x X} = and let v X
+
. Then (v, u)
X
L
(a
i
,a
k
)x
if and
only if v L
i
and a
i
.(v x) = a
k
.u. By Propositions 2.2.3 and 2.3.1, the language
K = {(w, u) | w L
i
, a
i
.(w x) = b}
X
= L
(a
i
,a
k
)x
{(v, u) | v X
+
}
X
is regular. Making use of Proposition 2.3.1, we obtain that
{w X
+
| (w, v)
X
K for some v X
+
} = {w X
+
| (w, u)
X
K}
= {w L
i
| a
i
.(w x) = b}
Similarly we can prove the following:
1
, L
2
, . . . , L
n
) be an automatic structure for the S-act A with
respect to the generating set A
0
= {a
1
, a
2
, . . . , a
n
} and let b A. Then for each i, (1
i n), the set {w L
i
| a
i
.w = b} is regular.
Now we claim that the denition of an automatic S-act is indeed a generalization of
the notion of an automatic semigroup. As in Example 3.1.4, when considering a semigroup
S as a right S-act, we assume that the action is right multiplication and we write s t
instead of s.t.
Proposition 3.1.10. A semigroup S generated by a nite set X is automatic if and only
if it is automatic as a right S-act with respect to the generating set X.
Proof. () Assume that (X, L) forms an automatic structure for S, where X =
{x
1
, . . . , x
n
}. Then L = S and the languages L
=
and L
x
(x X) are regular. By
Lemma 2.2.5, for each j, (1 j n), the language
L
j
= {w X
+
| x
j
w L}
is regular. We assume that none of the languages L
j
(1 j n) are empty. We clearly
have that
n
_
j=1
x
j
L
j
= L = S,
and hence S is regularly generated by X and L
1
, . . . , L
n
. We claim that (X, L
1
, . . . , L
n
)
is an automatic structure for the right S-act S with respect to the generating set X. By
denition,
K
(x
i
,x
j
)=
= {(u, v) L
i
L
j
| x
i
u = x
j
v}
X
= {(u, v)
X
X(2, $) | (x
i
, x
j
)(u, v)
X
L
=
},
and so it is a regular language by Lemma 2.2.5. Similarly,
K
(x
i
,x
j
)x
= {(u, v) L
i
L
j
| x
i
u x = x
j
v}
X
= {(u, v)
X
X(2, $) | (x
i
, x
j
)(u, v)
X
L
x
}
is a regular language by Lemma 2.2.5.
() Conversely, assume that (X, L
1
, . . . , L
n
) is an automatic structure for the right S-
act S with respect to the generating set X. Then
n
j=1
x
j
L
j
= S and so K =
n
j=1
x
j
L
j
is a regular language for which K = S. We show that (X, K) is an automatic structure
for S. By denition,
K
=
= {(u, v) | u, v K, u = v}
X
= {(u, v) | u x
i
k, v x
j
l, (k, l) L
i
L
j
, x
i
k = x
j
l}
X
=
_
1i,jn
(x
i
, x
j
)L
(x
i
,x
j
)=
,
which is a nite union of regular languages, and hence is regular. Similarly, for each x X,
the language
K
x
= {(u, v) | u, v K, u x = v}
X
= {(u, v) | u x
i
k, v x
j
l, (k, l) L
i
L
j
, x
i
kx = x
j
l}
X
=
_
1i,jn
(x
i
, x
j
)L
(x
i
,x
j
)x
is regular, since it is a nite union of regular languages.
For monoids the following proposition holds.
Proposition 3.1.11. A monoid M is automatic if and only if it is automatic as a right
M-act with respect to the generating set {1}.
Proof. That {1} is a generating set for the M-act M is immediate, since 1 M = M.
The lemma now follows from the denitions.
3.2. SCH
UTZENBERGER AUTOMATIC REGULAR SEMIGROUPS 25

3.2 Sch utzenberger automatic regular semigroups
It was proved in [8], that free inverse semigroups are not automatic. On the other hand,
free inverse semigroups have solvable word problem [22], [27]. These facts motivate us
to introduce a new denition of automaticity for regular semigroups, namely that of a
Sch utzenberger automatic regular semigroup. We formulate the new notion in the frame-
work of automatic S-acts, where the S-acts are R-classes of the semigroup. In the next
section, we show that free inverse semigroups are Sch utzenberger automatic.
Being Sch utzenberger automatic means that the regular semigroup satises a collection
of local properties. As indicated above these local properties are dened on the R-classes
of the regular semigroup. By introducing R-class automaticity in terms of left actions of
S, we connect the R-classes of S. Finally, combining the denition of Sch utzenberger au-
tomaticity and R-class automaticity, we introduce the notion of a strongly Sch utzenberger
automatic regular semigroup.
Throughout this chapter S will denote a semigroup, X a generating set for S. We
+
map : X S.
Before we give the denition of a Sch utzenberger automatic regular semigroup, we turn
our attention to the right S
0
-act R
0
given in Example 3.1.6.
Proposition 3.2.1. Let S be a regular semigroup generated by a nite set X, and let R
be an R-class of S. Then the following are equivalent:
(S1) There exists a regular language L over X
0
= X {0} such that (X
0
, L) forms
an automatic structure for the right S
0
-act R
0
with respect to some one element
generating set {s}.
(S2) There exists a regular language K over X and an element s R such that the
(i) s K = R;
(ii) K
=
= {(u, v) | u, v K, s u = s v}
X
(iii) K
x
= {(u, v) | u, v K, s (u x) = s v}
X
(S3) There exists a regular language N over X such that the following conditions hold:
(i) N = R;
(ii) N
=
= {(u, v) | u, v N, u = v}
X
(iii) N
x
= {(u, v) | u, v N, u x = v}
X
Proof. (S1) (S2) Assume that (X
0
, L) is an automatic structure for the right S
0
-
act R
0
with respect to the generating set {s}. Then s L = R
0
. By Lemma 3.1.9,
L
= {l L | s l = 0} is a regular language, hence K = L L
is also a regular language

and s K = R. It is also obvious that
K
=
= {(u, v) | u, v K, s u = s v}
X
= L
=
{(u, v) | u, v L
}
X
= L
=
(L
)
X
,
and so K
=
is a regular language. To show that K
x
is a regular language, note that
K
x
= L
x
(L
)
X
{(u, v) L L
| s (u x) = s v}
X
.
Let L
= {u L | s (u x) = 0}. By Lemma 3.1.8, L
is a regular language. It follows

that {(u, v) L L
| s (u x) = s v}
X
= (L
)
X
is also a regular language, and
hence we may deduce that K
x
(S2) (S3) Assume that there exists a regular language K over X such that con-
ditions (i) (iii) of (S2) hold. Let w be a word over X representing s. Then we have
R = s K = (wK). Dene N = wK. It is obvious that N is a regular language and
that N = R. Moreover
N
=
= {(u, v) | u, v N, u = v}
X
= {(u, v) | u wk
1
, v wk
2
, wk
1
= wk
2
}
X
= {(u, v) | u wk
1
, v wk
2
, s k
1
= s k
2
}
X
= (w, w)
X
K
=
,
hence is a regular language. Similarly we obtain that
N
x
= {(u, v) | u, v N, u x = v}
X
= (w, w)
X
K
x
(S3) (S1) Assume that there exists a regular language N over X such that condi-
tions (i) (iii) of (S3) hold. Adjoin a zero element 0 to S and extend multiplication in
the usual way. Dene a right action of S
0
on R
0
as before:
(r, s)
_
r s if rs R,
0 otherwise.
3.2. SCH
UTZENBERGER AUTOMATIC REGULAR SEMIGROUPS 27

Let e be an idempotent element of R. Such an element of R exists, since S is a regular
semigroup. Consider the regular language L = N {0}. Recall that e is a left identity in
its R-class. Then e L = e (N {0}) = R
0
. Since N = R, we obtain that
L
=
= {(u, v) | u, v L, e u = e v}
X
= {(u, v) | u, v N, u = v}
X
{(0, 0)}
= N
=
{(0, 0)}.
Similarly, if x X then
L
x
= {(u, v) | u, v L, e u x = e v}
X
= N
x
{(0, 0)},
and we have that
L
0
= {(u, v) | u, v L, e u 0 = e v}
X
= (0, 0)
which veries that (X
0
, L) is indeed an automatic structure for the right S
0
-act R
0
with
respect to the generating set {e}.
We note that in the proof of Proposition 3.2.1 the assumption that S is a regular
semigroup was only used to deduce that the R-class R contains an idempotent element.
Hence, Proposition 3.2.1 can be rephrased for all such R-classes of an arbitrary semigroup
that contains an idempotent. On the other hand, if we assume that S is a semigroup
in which every R-class contains an idempotent, then we have that S has to be a regular
semigroup.
We introduce now the denition of a Sch utzenberger automatic regular semigroup S.
Denition 3.2.2. Let S be a regular semigroup generated by a nite set X, and L be a
regular language over X. Let R be an R-class of S. We say that (X, L) is a Sch utzenberger
automatic structure for R, if the following conditions are satised:
(i) L = R;
(ii) L
=
= {(u, v) | u, v L, u = v}
X
(iii) L
x
= {(u, v) | u, v L, u x = v}
X
If an R-class R has a Sch utzenberger automatic structure (X, L), then we say that R is
Sch utzenberger automatic. A Sch utzenberger automatic structure (X, L) of an R-class R
is said to be with uniqueness, if u = v (u, v L) implies u v. Dene a regular semigroup
S to be Sch utzenberger automatic, if all of its R-classes are Sch utzenberger automatic.
Although Proposition 3.2.1 provides three dierent descriptions of a Sch utzenberger
automatic regular semigroup, we will usually work with (S2) and (S3). Clearly conditions
of (S3) seem to be the easiest to check. On the other hand, the advantage of (S2) is that
if k K, then for every prex

k of k, s

k R
s
, a property that is not necessarily true
when considering description (S3). That is if n N and n is a prex of n, then usually
n / R.
The next proposition tells us, that the notion of a Sch utzenberger automatic regular
semigroup is in fact a generalization of the notion of an automatic group.
Proposition 3.2.3. Let G be a group. Then the following are equivalent:
(1) G is automatic.
(2) The right G-act G is automatic.
(3) G is Sch utzenberger automatic.
Proof. (1) (2) See Proposition 3.1.10.
(1) (3) Follows from the fact that if G is a group then R = GG.
Sch utzenberger automaticity of a regular semigroup S is a collection of local properties.
To link the R-classes of S we dene R-class automaticity in terms of left S-actions on the
partially ordered set S/R.
First we need to dene a left action of a semigroup S on S/R. Let
f : S/RS S/R; (R
t
, s) R
st
.
Since R is a left congruence on S, f clearly denes a left S-act.
For our purposes, we will dene R-class automaticity of a semigroup S when S is a
monoid.
Denition 3.2.4. Let S be a monoid generated by a nite X. Let L be a regular language
over X. We say that (X, L) is an R-class automatic structure for S/R, if (X, L) is an
automatic structure for the left S-act S/R with respect to the generating set R
1
. That is,
(1) R
1
L = S/R;
(2) L
=
= {(u, v) | u, v L, R
1
.u = R
1
.v}
X
(3) L
x
= {(u, v) | u, v L, R
1
.(u x) = R
1
.v}
X
3.3. EXAMPLES 29
Combining the notions of Sch utzenberger automaticity and R-class automaticity we
introduce the notion of a strongly Sch utzenberger automatic regular semigroup.
Denition 3.2.5. We say that a regular semigroup is strongly Sch utzenberger automatic,
if it is Sch utzenberger automatic, and R-class automatic.
3.3 Examples
In this section we give several examples for automatic S-acts. Throughout this section S
will denote a semigroup, X a nite generating set for S. For the sake of simplicity we
+
map : X S.
Proposition 3.3.1. Let A be a nite S-act so that the action f : AS A is surjective.
Then A is automatic.
Proof. Let A = {a
1
, . . . , a
n
} and for each a
i
A, x an element t
i
S and an
element b
i
A such that a
i
= b
i
.t
i
. Since f is a surjective map, such elements exist.
Let T = {t
1
, . . . , t
n
}. For each t
i
T x a word w
i
over X for which w
i
= t
i
and let
L
i
= {w
i
}. Clearly A is regularly generated by L
1
, . . . , L
n
and {b
1
, . . . , b
n
}. On the other
hand the languages L
(b
i
,b
j
)=
and L
(b
i
,b
j
)x
(x X) are nite, and hence they are regular,
proving that A is indeed automatic.
Corollary 3.3.2. Let S be a monoid such that S/R is nite. Then S is R-class automatic.
Proof. The left action f : S/RS S/R is clearly surjective. Hence if S/R is nite,
then S/R is an automatic left S-action by Proposition 3.3.1, proving that S is indeed
R-class automatic.
Corollary 3.3.3. Let R be a nite R-class of a regular semigroup S. Then R is Sch utzen-
berger automatic.
Proof. If R is a nite R-class of S, then the right S
0
-act R
0
is nite, moreover the
right action f : R
0
S
0
R
0
is surjective. Hence by Proposition 3.3.1, the right S
0
-act
R
0
is automatic. Making use of Proposition 3.2.1 we obtain that R is Sch utzenberger
automatic.
We have the following immediate consequence of Corollary 3.3.3:
Corollary 3.3.4. Let S be a regular semigroup whose R-classes are nite. Then S is
Sch utzenberger automatic.
Combining Corollaries 3.3.2 and 3.3.3 we obtain:
Corollary 3.3.5. Let S be a nite regular semigroup. Then S is strongly Sch utzenberger
automatic.
Next, we give two examples for innite automatic S-acts.
Example 3.3.6. Let A be the double innite chain with an identity adjoined on top:
. . . < e
1
< e
0
< e
1
< . . . < 1.
Let G = g be the innite cyclic group. Let n, k Z and dene a right action of G on A
in the following way:
f : A G A; (e
n
, g
k
) e
n
.g
k
= e
n+k
; (1, g
k
) 1.g
k
= 1.
We claim that the right G-act A is automatic. (We note that the dened action can be
considered as a left action of G on A as well and it can be veried that the left G-act A
is automatic.)
Let X = {g, g
1
}, A
0
= {e
0
, 1}, L
0
= g
+
(g
1
)
+
{gg
1
} and L
1
= {g}. By
denition e
n
= e
0
.g
n
, n Z, thus A is regularly generated by A
0
and the regular languages
L
0
, L
1
. Clearly we have that the languages L
(e
0
,1)=
, L
(e
0
,1)x
, L
(1,e
0
)=
, L
(1,e
0
)x
(x X) are
empty, and hence they are regular. Moreover we have that
L
(1,1)=
= {(u, v) L
1
L
1
| 1.u = 1.v}
X
= (g, g);
L
(1,1)x
= {(u, v) L
1
L
1
| 1.(u x) = 1.v}
X
= (g, g);
L
(e
0
,e
0
)=
= {(u, v) L
0
L
0
| e
0
.u = e
0
.v}
X
= (g, g)
+
X
(g
1
, g
1
)
+
X
(gg
1
, gg
1
)
X
;
L
(e
0
,e
0
)g
= {(u, v) L
0
L
0
| e
0
.(u g) = e
0
v}
X
= (g, g)
+
X
($, g) (g
1
, g
1
)(g
1
, g
1
)
+
X
(g
1
, $)
(g
1
, g)($, g
1
) (g, g)(g
1
, $)
L
(e
0
,e
0
)
g
1
= {(u, v) L
0
L
0
| e
0
.(u g
1
) = e
0
v}
X
= (g
1
, g
1
)
+
X
($, g
1
) (g, g)(g, g)
+
X
(g, $)
(g, g)($, g
1
) (g, g
1
)(g
1
, $).
These languages are regular, and hence we obtain that (X, L
0
, L
1
) forms an automatic
structure with respect to the generating set A
0
for A.
3.3. EXAMPLES 31
Example 3.3.7. Let B be the innite antichain
. . . , e
1
, e
0
, e
1
, . . . .
Adjoin an identity 1 on top and a zero 0 on bottom, that is for each e
i
B, 0 < e
i
< 1
holds. Let A denote the lattice obtained. Let G = g be the innite cyclic group. Let
n, k Z and dene a right action of G on A in the following way:
f : A G A; (e
n
, g
k
) e
n+k
; (1, g
k
) 1; (0, g
k
) 0.
We claim that the right G-act A is automatic. (We note that the dened action can be
considered as a left action of G on A as well and it can be veried that the left G-act A
is automatic.)
Let X = {g, g
1
}, A
0
= {e
0
, 1, 0}, L
0
= g
+
(g
1
)
+
{gg
1
}, L
1
= L
2
= {g}.
By denition, e
n
= e
0
.g
n
, n Z, thus A is regularly generated by A
0
and the regular
languages L
0
, L
1
, L
2
. Let
(a, b) {(1, 0), (0, 1), (e
0
, 1), (1, e
0
), (e
0
, 0), (0, e
0
)}.
Clearly, the languages L
(a,b)=
, L
(a,b)x
, L
(a,b)=
, L
(a,b)x
(x X) are empty, and hence they
are regular. Moreover we have that
L
(1,1)=
= {(u, v) L
1
L
1
| 1.u = 1.v}
X
= (g, g);
L
(1,1)x
= {(u, v) L
1
L
1
| 1.(u x) = 1.v}
X
= (g, g);
for all x X. Similarly
L
(0,0)=
= {(u, v) L
2
L
2
| 0.u = 0.v}
X
= (g, g);
L
(0,0)x
= {(u, v) L
2
L
2
| 0.(u x) = 0.v}
X
= (g, g);
for all x X. Also we have that
L
(e
0
,e
0
)=
= {(u, v) L
0
L
0
| e
0
.u = e
0
.v}
X
= (g, g)
+
X
(g
1
, g
1
)
+
X
(gg
1
, gg
1
)
X
;
L
(e
0
,e
0
)g
= {(u, v) L
0
L
0
| e
0
.(u g) = e
0
v}
X
= (g, g)
+
X
($, g) (g
1
, g
1
)(g
1
, g
1
)
+
X
(g
1
, $)
(g
1
, g)($, g
1
) (g, g)(g
1
, $)
L
(e
0
,e
0
)
g
1
= {(u, v) L
0
L
0
| e
0
.(u g
1
) = e
0
v}
X
= (g
1
, g
1
)
+
X
($, g
1
) (g, g)(g, g)
+
X
(g, $)
(g, g)($, g
1
) (g, g
1
)(g
1
, $)
These languages are regular, and hence we obtain that (X, L
0
, L
1
, L
2
0
for A.
Example 3.3.8. The bicyclic monoid is strongly Sch utzenberger automatic.
To verify this, consider the bicyclic monoid given by the monoid presentation B =
p, q | pq = 1. We let X = {p, q} and denote by : X
+
S the homomorphism extend-
ing the identity map : X S. Each element s B can be written as s = q
i
p
j
, (i, j N),
moreover q
i
p
j
R q
n
p
m
if and only if i = n. To show that B is R-class automatic, consider
the regular language K = q
+
qp. Clearly R
1
.K = B/R. On the other hand
K
=
= {(u, v) K K | R
1
.u = R
1
.v}
X
= {(u, u) | u K}
X
,
K
p
= {(u, v) K K | R
1
.u p = R
1
.v}
X
= (q, q)
+
X
(q, $) (qp, qp)
X
,
K
q
= {(u, v) K K | R
1
.u q = R
1
.v}
X
= (q, q)
+
X
($, q) (qp, q)
X
,
and so (X, K) indeed forms an automatic structure for the left B-act B/R.
To show that B is Sch utzenberger automatic, consider an arbitrary R-class R of B.
Assume that q
i
p
j
R. As we have noted above, any element of R is of the form q
i
p
k
, k
N. Let L = q
i
p
. Clearly L = R and we have that

L
=
= {(u, v) L L | u = v}
X
= {(u, u) | u L}
X
,
L
p
= {(u, v) L L | u p = v}
X
= (q, q)
i
X
(p, p)
X
($, p),
L
q
= {(u, v) L L | u q = v}
X
= (q, q)
i
X
(p, p)
X
(p, $),
proving that (X, L) is a Sch utzenberger automatic structure for R. Since R was an arbi-
trary R-class of B, it follows that B is Sch utzenberger automatic.
Example 3.3.9. Polycyclic monoids are strongly Sch utzenberger automatic.
The polycyclic monoid on n generators (n 2) is given by the presentation
P
n
= p
1
, . . . , p
n
, p
1
1
, . . . , p
1
n
| p
i
p
1
i
= 1, p
i
p
1
j
= 0 for i = j.
Consider the generating set X = {0, 1, p
1
, . . . , p
n
, p
1
1
, . . . , p
1
n
} for P
n
. Denote by :
X
+
P
n
the homomorphism extending the identity map : X P
n
. To make notation
convenient, we let Y = {p
1
, . . . , p
n
} and Y
1
= {p
1
1
, . . . , p
1
n
}. Every non-zero element of
P
n
can be uniquely written as u
1
v, where u, v Y
+
. Moreover u
1
v R x
1
y, (u, v, x, y
3.3. EXAMPLES 33
Y
+
) if and only if u and x are identitcal words ([27, Chapter 9.3]). To show that P
n
is R-class automatic, we consider the regular language K = {0, 1} (Y
1
)
+
. Clearly
R
1
.K = P
n
/R. On the other hand, we have that
K
=
= {(u, v) K K | R
1
.u = R
1
.v}
X
= {(u, u) | u K}
X
= K
1
,
K
0
= {(u, v) K K | R
1
.u 0 = R
1
.v}
X
= (0, 0) (1, 0) ((Y
1
)
+
{0})
X
,
and hence they are regular languages. To show that K
p
i
and K
p
1
i
are regular languages
for all 1 i n, we make the following observations. By Lemma 2.2.5, the language
M
i
= {u (Y
1
)
+
| u p
1
i
w}
is regular, and so by Proposition 2.2.3, the language
N
i
= (Y
1
)
+
\ M
rev
i
is also regular. By Proposition 2.3.2, the language A = {(u, u)|u (Y
1
)
}
X
is regular.
Using the above introduced notation, we obtain that
K
p
i
= {(u, v) K K | R
1
.u p
i
= R
1
.v}
X
= (N
i
{0})
X
A(p
1
i
, $) (0, 0) (1, 1)
and
K
p
1
i
= {(u, v) K K | R
1
.u p
1
i
= R
1
.v}
X
= A($, p
1
i
) (0, 0) (1, p
1
i
)
are regular languages. Hence, we may deduce that P
n
is indeed R-class automatic.
Next, we show that P
n
is Sch utzenberger automatic. Clearly the R-class of 0 is
Sch utzenberger automatic. Consider an R-class R = R
0
of P
n
and assume that u
1
v R,
where u, v Y
+
. As we mentioned before, every element of R is of the form u
1
w, where
w Y
. Consider the regular language L = u

1
Y
. Clearly L = R, moreover
L
=
= {(v, w) L L | v = w}
X
= {(v, v) | v L}
X
= L
1
,
L
0
= {(v, w) L L | v 0 = w}
X
=
To prove that L
p
i
and L
p
1
i
are regular languages for all 1 i n, we make the following
observations. By Lemma 2.2.5, the language
M
j
= {v Y
+
| v wp
j
}
is regular, and so by Proposition 2.2.3, the language
N
i
=
n
_
j=1,i=j
M
j
is also regular. By Proposition 2.3.2, the language B = {(v, v)|v Y
}
X
is regular.
Using the above introduced notation, we obtain that
L
p
i
= {(v, w) L L | v p
i
= w}
X
= (u
1
, u
1
)
X
B($, p
i
)
L
p
1
i
= {(v, w) L L | v p
1
i
= w}
X
= (u
1
, u
1
)
X
B(p
i
, $) (u
1
N
i
{0})
X
,
proving that (X, L) is a Sch utzenberger automatic structure for R. It follows that B is
Proposition 3.3.10. Let X be a non-empty nite set. Then the following hold:
(1) The free band and the free semilattice on X are strongly Sch utzenberger automatic.
(2) The free completely simple semigroup on X is strongly Sch utzenberger automatic.
(3) The free inverse semigroup on X is Sch utzenberger automatic.
Proof. (1) It is known that if a band or a semilattice B is nitely generated then it is
nite. The assertion hence follows from Corollary 3.3.5.
(2) We rst describe the free completely simple semigroup on X. Fix z X and let
Y = {p
xy
| x, y X \ {z}}. Let Z = X Y . Consider the free group FG(Z) and denote
its identity element by 1. For all x X let p
zx
= p
xz
= 1, and consider the X X-matrix
P with (x, y) entry p
xy
. Then the free completely simple semigroup on X is isomorphic
to the Rees matrix semigroup CS
X
= M(X, FG(Z), X; P). Since X is nite, CS
X
has
nitely many R-classes, and hence is R-class automatic by Corollary 3.3.4. In Section 9,
we will verify that a completely simple semigroup is Sch utzenberger automatic if and only
if all of its maximal subgroups are automatic. Clearly the maximal subgroups of CS
X
are isomorphic to the free group FG(Z), which is known to be automatic, and hence the
assertion follows.
(3) It is known that the R-classes of the free inverse semigroup FIS(X) are nite, and
hence by Corollary 3.3.4, FIS(X) is indeed Sch utzenberger automatic.
3.4 Basic properties
One naturally expects automatic S-acts to bear certain properties: for example, having
automatic structures with uniqueness, invariance under the change of generators of S and
3.4. BASIC PROPERTIES 35
of the S-act A, etc. In this section we discuss the concept of an automatic structure
with uniqueness and the invariance of R-classes being Sch utzenberger automatic within
a D-class. Finally, in preparation for the fth section (discussion of automaticity being
invariant under the change of generators), we introduce three technical lemmas. Through-
out this section S will denote a semigroup, X a nite generating set for S. We assume
that X S and denote by : X
+
S the homomorphism extending the identity map
: X S. We begin this section with the following observation.
Lemma 3.4.1. Let A be an S-act regularly generated by L
1
, . . . , L
n
X
+
and A
0
. Assume
that L
(a
i
,a
j
)=
, a
i
, a
j
A
0
is a regular language. Then L
(a
j
,a
i
)=
is also a regular language.
Proof. Let A = (, X(2, $), , p, F) be a nite state automaton accepting the lan-
guage L
(a
i
,a
j
)=
. Consider the automaton B = (, X(2, $), , p, F), where (q, (x, y)) =
(q, (y, x)). Since (u, v) L
(a
i
,a
j
)=
if and only if (v, u) L
(a
j
,a
i
)=
, it is clear, that B is the
automaton accepting L
(a
j
,a
i
)=
.
Automatic structure with uniqueness
We prove in two steps that if A is an automatic S-act, then there exists an automatic
structure with uniqueness. In the rst step we show that if A is an automatic S-act,
then there exists an automatic structure (X, K
1
, . . . , K
m
) with respect to a generating set
B = {a
1
, . . . , a
m
}, such that a
i
.K
i
a
j
.K
j
= for all 1 i, j m, i = j. That
is, a
i
.u = a
j
.v, (u, v) K
i
K
j
implies a
i
= a
j
. In the second step following the
semigroup and group theoretical approach to the problem, by ordering the alphabet X
and introducing shortlex order on X
+
we construct regular languages L
1
, . . . , L
m
from
K
1
, . . . , K
m
, such that (X, L
1
, . . . , L
m
) will form an automatic structure with uniqueness
for A.
1
, . . . , L
n
) be an automatic structure with respect to the genera-
ting set A
0
= {a
1
, . . . , a
n
} for the S-act A. Then there exists a subset B = {b
1
, . . . , b
m
}
A
0
and regular languages K
1
, . . . , K
m
such that the following hold:
(i) b
i
.K
i
b
j
.K
j
= for all 1 i, j m, i = j.
(ii) (X, K
1
, . . . , K
m
) is an automatic structure for A with respect to the generating set
B.
Proof. For each i (1 i n 1), we let
L
i
=
n
_
j=i+1
{u L
i
| (u, v)
X
L
(a
i
,a
j
)=
for some v L
j
}.
Since L
(a
i
,a
j
)=
is a regular language for each j (i + 1 j n), the languages
{u L
i
| (u, v)
X
L
(a
i
,a
j
)=
for some v L
j
}
are regular by Proposition 2.3.1. Thus

L
i
is a regular language, since it is a nite union
of regular languages. Dene for each i (1 i n 1)
K
i
= L
i
L
i
and let K
n
= L
n
.
Note that we only kept those elements u of L
i
for which there exists no generator a
j
, i < j
of A and v L
j
such that a
i
.u = a
j
.v. By Proposition 2.2.3, K
1
, K
2
, . . . , K
n
are regular
languages. It might happen that K
i
= , in which case a
i
is a surplus generator, since then,
for all u L
i
, there exists v L
j
, i < j, such that a
i
.u = a
j
.v. Let B = {a
i
A
0
| K
i
= }.
That a
i
.K
i
a
j
.K
j
= , where a
i
, a
j
B, i = j, 1 i, j n follows from
the denition of the regular languages K
1
, . . . , K
m
. To give a more detailed reason: the
equation a
i
.u = a
j
.v, where (u, v) K
i
K
j
cannot hold, since if i < j, then by the
denition of K
i
, u should have been removed from L
i
, and similarly if j < i, then v should
have been removed from L
j
.
To prove that an automatic structure for A is obtained with respect to the generating
set B, we rst verify that we did not subtract necessary elements from L
1
, . . . , L
n
, that is
we show that
_
a
j
B
a
j
.K
j
= A.
Let b A. Since (X, L
1
, . . . , L
n
) is an automatic structure for A with respect to the
generating set A
0
, there exists a
i
A
0
and u L
i
such that a
i
.u = b. If u K
i
, then
we are nished. If u / K
i
, then there exists a
j
A
0
and v L
j
such that a
i
.u = a
j
.v and
i < j. Obviously we can suppose that j is the greatest such index, hence it follows by the
denition of K
j
that v K
j
, hence b = a
j
.v a
j
.K
j
.
It is immediate that for all (a
i
, a
j
) B B,
K
(a
i
,a
j
)=
= {(u, v) K
i
K
j
| a
i
.u = a
j
.v}
X
= L
(a
i
,a
j
)=
(K
i
K
j
)
X
,
hence is a regular language. Similarly we obtain that for all x X,
K
(a
i
,a
j
)x
= {(u, v) K
i
K
j
| a
i
.(u x) = a
j
.v}
X
= L
(a
i
,a
j
)x
(K
i
K
j
)
X
,
We have the following immediate consequence of Lemma 3.4.2.
Corollary 3.4.3. If A is an automatic S-act, then there exists an automatic structure
(X, K
1
, . . . , K
m
) with respect to some generating set B, with the help of which A can be
partitioned into m subsets A
1
, . . . , A
m
; that is to say that
m
_
j=1
A
j
= A and A
i
A
j
= for all 1 i, j m, i = j.
Proof. Let A be an automatic S-act. Making use of Lemma 3.4.2, we may assume
that (X, K
1
, . . . , K
m
) is an automatic structure with respect to a generating set B =
{b
1
, . . . , b
m
}, such that b
i
.K
i
b
j
.K
j
= for all 1 i, j m, i = j. Dene A
i
= b
i
.K
i
.
It is immediate that A
i
A
j
= for all 1 i, j m, i = j. Furthermore, since
m
j=1
b
j
.K
j
= A, we also have that

m
j=1
A
j
= A.
Before we turn to the second step, we recall the denition of the shortlex ordering.
Let X be an alphabet, and choose an ordering x
1
< x
2
< . . . < x
n
on X. The shortlex
ordering < on X
+
is dened as follows:
v < w v is shorter then w, or they have the same length
and v comes before w in the lexicographical order.
Proposition 3.4.4. If A is an automatic S-act, then A has an automatic structure with
uniqueness.
Proof. Let (X, K
1
, . . . , K
n
) be an automatic structure with respect to the generating
set A
0
= {a
1
, . . . , a
n
} for the S-act A. By Lemma 3.4.2, we may assume that a
i
.K
i

a
j
.K
j
= for all 1 i, j n, i = j. Hence, if a
i
.u = a
j
.v, u K
i
, v K
j
, then a
i
= a
j
and u, v K
i
. To construct an automatic structure (X, L
1
, . . . , L
n
) with uniqueness, we
need to reduce the number of elements of each regular language K
i
, (1 i n) in such
a way that for each a (a
i
.K
i
) we have exactly one w L
i
K
i
such that a
i
.w = a.
Dene
L
i
= {w K
i
| if (w, v) K
(a
i
,a
i
)=
then w < v in the shortlex order}.
By Proposition 2.3.13, L
i
is a regular language. Furthermore, we clearly have that
n
_
j=1
a
j
.L
j
= A.
If a
i
.u = a
j
.v, (u, v) L
i
L
j
, then as before a
i
= a
j
, u, v L
i
, and hence by the
denition of L
i
, u v. It follows that
L
(a
i
,a
j
)=
=
_
if i = j,
{(u, u) | u L
i
}
X
if i = j.
Thus L
(a
i
,a
j
)=
is a regular language for all (a
i
, a
j
) A
0
A
0
by Proposition 2.3.2. More-
over,
L
(a
i
,a
j
)x
= K
(a
i
,a
j
)x
(L
i
L
j
)
X
,
and hence is a regular language. We may deduce that (X, L
1
, . . . , L
n
) is indeed an auto-
matic structure with uniqueness with respect to the generating set A
0
for the S-act A.
An immediate consequence of Proposition 3.4.4:
Corollary 3.4.5. If S is an R-class automatic semigroup, then there exists an R-class
automatic structure for S/R with uniqueness.
Also we have:
Corollary 3.4.6. If R is a Sch utzenberger automatic R-class of a regular semigroup S,
then there exists a Sch utzenberger automatic structure for R with uniqueness.
Proof. Let R be a Sch utzenberger automatic R-class of a regular semigroup S. Then
by Proposition 3.2.1, the right S
0
-act R
0
is automatic with respect to some generating
set {s}, hence there exists an automatic structure (X, L) with uniqueness for the right
S
0
-act R
0
with respect to {s}. Following the construction of the regular languages K
and N in Proposition 3.2.1, we have that K = L L
, where L
= {l L | s l = 0}
and N = w(L L
), where w is a word over X representing s = es. Since (X, L) is an

automatic structure with uniqueness, and N
=
= (w, w)
X
K
=
, we obtain that
(u, v) N
=
(u, v) (w, w)
X
(L
=
(K K)
X
) u v,
hence (X, N) is a Sch utzenberger automatic structure with uniqueness for R.
Sch utzenberger automatic R-classes in a D-class.
We will focus on Sch utzenberger automatic R-classes of a regular semigroup S. To be
more precise, we show that the property of being Sch utzenberger automatic is invariant
within a D-class of a regular semigroup S. Recall that if s S, then R
s
denotes the
R-class of s.
Proposition 3.4.7. Let S be a regular semigroup and R be a Sch utzenberger automatic
R-class of S. Let s R, r S such that s D r. Then R
r
is Sch utzenberger automatic.
Proof. Suppose that (X, L) is a Sch utzenberger automatic structure for R. Then
L = R and the languages L
=
, L
x
(x X) are regular. On the other hand, since s D r,
there exists t S
1
such that s R t L r. In particular, we have that r = k t for some
k S
1
and it follows by Greens lemma, that we can give an L-class preserving bijection
: R
t
R
r
; x k x.
Let w be a word over X representing k, and let K = wL. Clearly K is a regular language.
We show that (X, K) is a Sch utzenberger automatic structure for R
r
. Since is a bijection
and L = R
t
, we obviously have that K = (wL) = k L = R
r
. Also,
if wl
1
= wl
2
then l
1
= l
2
holds, since if wl
1
= k l
1
= k l
2
= wl
2
, that is if (l
1
) = (l
2
), then l
1
= l
2
, since
is injective. Similarly,
if wl
1
x = wl
2
, (x X) then l
1
x = l
2
.
Thus we obtain that
K
=
= {(u, v) | u, v K, u = v}
X
= {(u, v) | u wl
1
, v wl
2
, (l
1
, l
2
L), wl
1
= wl
2
}
X
= (w, w)
X
L
=
is a regular language. Similarly, we obtain that
K
x
= {(u, v) | u, v K, u x = v}
X
= {(u, v) | u wl
1
, v wl
2
, (l
1
, l
2
L), wl
1
x = wl
2
}
X
= (w, w)
X
L
x
Technical lemmas
We are going to give two lemmas that will prove useful later on. First, we generalize
Proposition 3.2 of [5].
1
, . . . , L
n
) be an automatic structure with respect to the genera-
ting set A
0
for the S-act A. Let w X
+
and a
i
, a
k
A
0
. Then the language
L
(a
i
,a
k
)w
= {(u, v) L
i
L
k
| a
i
.uw = a
k
.v}
X
is regular.
Proof. Let w x
1
. . . x
t
and x a
i
, a
k
A
0
. Let L =
n
j=1
L
j
. Consider the set
K
1
=
n
_
j=1
L
(a
i
,a
j
)x
1
.
Since (X, L
1
, . . . , L
n
) is an automatic structure with respect to the generating set A
0
for
A, K
1
is a regular language. We note that
(u, v
1
)
X
K
1
(u, v
1
) L
i
L and a
i
.(u x
1
) = a
j
.v
1
,
where a
j
A
0
, v
1
L
j
.
Since K
1
is a regular language, the language
N
1
= {v
1
L | (u, v
1
)
X
K
1
for some u L
i
}
is regular by Proposition 2.3.1. Moreover the set N
1
determines the following index set
I
1
= {j | 1 j n there exists v L
j
N
1
}.
We next dene the languages K
2
, N
2
and the index set I
2
. Then we give a recursive
denition for K
l
, N
l
, I
l
(1 l < t). Let
K
2
=
_
mI
1
(
n
_
j=1
L
(am,a
j
)x
2
).
Since K
2
is a nite union of regular languages it is a regular language. We note that
(v
1
, v
2
)
X
K
2
(v
1
, v
2
) N
1
L and a
m
.(v
1
x
2
) = a
j
.v
2
where m I
1
, v
1
L
m
N
1
and a
j
A
0
, v
2
L
j
.
there exists u L
i
such that a
i
.(ux
1
x
2
) = a
j
.v
2
, where
a
j
A
0
, v
2
L
j
.
Since K
2
is a regular language, the set
N
2
= {v
2
L | (v
1
, v
2
)
X
K
2
for some v
1
N
1
}
forms a regular language by Proposition 2.3.1. We consider the following index set deter-
mined by N
2
:
I
2
j
N
2
}.
Giving the promised recursive denition for K
l
, N
l
, I
l
, (2 j < t), we let
K
l
=
_
mI
l1
(
n
_
j=1
L
(am,a
j
)x
l
).
Clearly K
l
is a regular language. We note that
(v
l1
, v
l
)
X
K
l
(v
l1
, v
l
) N
l1
L and a
m
.(v
l1
x
l
) = a
j
.v
l
where m I
l1
, v
l1
L
m
N
l1
and a
j
A
0
, v
l
L
j
.
there exists u L
i
such that a
i
.(ux
1
x
2
. . . x
l
) = a
j
.v
l
,
where a
j
A
0
, v
l
L
j
.
We have that the language
N
l
= {v
l
L | (v
l1
, v
l
) K
l
for some v
l1
N
l1
}
is regular. The index set determined by N
l
is
I
l
j
N
l
}.
Finally we consider the set
K
t
=
_
mI
t1
L
(am,a
k
)x
t
.
Clearly K
t
is a regular language, since it is a nite union of regular languages. We note
that
(v
t1
, v)
X
K
t
(v
t1
, v) N
t1
L
k
and a
m
.(v
t1
x
t
) = a
k
.v
where m I
t1
and v
t1
L
m
.
there exists u L
i
such that a
i
.ux
1
x
2
. . . x
t
= a
k
.v
and v L
k
.
Now the languages
K
x
1
x
2
= {(u, v
2
) L
i
N
2
| there exists v
1
N
1
such that
(u, v
1
) K
x
1
and (v
1
, v
2
) K
x
2
}
X
K
x
1
x
2
x
3
= {(u, v
3
) L
i
N
3
| there exists v
2
N
2
such that
(u, v
2
) K
x
1
x
2
and (v
2
, v
3
) K
x
3
}
X
.
.
.
K
x
1
...xt
= {(u, v) L
i
L
k
| there exists v
t1
N
t1
such that
(u, v
t1
) K
x
1
x
2
...x
t1
and (v
t1
, v) K
xt
}
X
are regular by Proposition 2.3.3. Clearly, if (u, v) K
x
1
...xt
, then (u, v) L
(a
i
,a
k
)w
. Let
(u, v) L
(a
i
,a
k
)w
. Since

n
j=1
a
j
.L
j
= A, we can construct the following sequence
a
i
.ux
1
= a
j
1
.v
1
, v
1
L
j
1
a
j
1
.v
1
x
2
= a
j
2
.v
2
, v
2
L
j
2
.
.
.
.
.
.
a
j
t1
.v
t1
x
t
= a
k
.v, v L
k
where (u, v
i
)
X
K
x
1
...x
i
(1 i t 1), and so (u, v) K
x
1
...xt
. Hence we may deduce
that K
x
1
...xt
= L
(a
i
,a
k
)w
, and so L
(a
i
,a
k
)w
is indeed a regular language.
Finally we introduce the following lemma:
1
, . . . , L
n
) be an automatic structure with uniqueness for the S-
act A, where the generating set is A
0
= {a
1
, . . . , a
n
}. Assume that a
n
= a
i
.u, u L
i
, n =
i. Let a
j
A
0
, j = n and x X. Then the languages
(a) L
(a
i
,a
j
)=
= {(v, w) | v uL
n
, w L
j
, a
i
.v = a
j
.w}
X
;
(b) L
(a
j
,a
i
)=
= {(w, v) | w L
j
, v uL
n
, a
j
.w = a
i
.v}
X
;
(c) L
(a
i
,a
j
)x
= {(v, w) | v uL
n
, w L
j
, a
i
.(v x) = a
j
.w}
X
;
(d) L
(a
j
,a
i
)x
= {(w, v) | w L
j
, v uL
n
, a
j
.(w x) = a
i
.v}
X
are regular.
Proof. (a) Let (X, L
1
, . . . , L
n
) be an automatic structure with uniqueness for the S-act
A, where the generating set is A
0
= {a
1
, . . . , a
n
}. Assume that a
n
= a
i
.u, u L
i
, n = i.
Then,
L
(a
i
,a
j
)=
= {(v, w) | v uL
n
, w L
j
, a
i
.v = a
j
.w}
X
= {(u v, w) | v L
n
, w L
j
, a
i
.u v = a
j
.w}
X
= {(u v, w) | v L
n
, w L
j
, a
n
. v = a
j
.w}
X
.
Hence if (v, w)
X
L
(a
i
,a
j
)=
, then v = u v for some v L
n
and ( v, w)
X
L
(an,a
j
)=
.
Since (X, L
1
, . . . , L
n
) is an automatic structure with uniqueness it follows that a
n
= a
j
and v w. Since by assumption a
j
= a
n
, we may deduce that L
(a
i
,a
j
)=
= , hence it is a
regular language.
(b) The proof is similar to that of part (a).
(c) First we note that
L
(a
i
,a
j
)x
= {(v, w) | v uL
n
, w L
j
, a
i
.(v x) = a
j
.w}
X
= {(u v, w) | v L
n
, w L
j
, a
i
.(u v x) = a
j
.w}
X
= {(u v, w) | v L
n
, w L
j
, a
n
.( v x) = a
j
.w}
X
.
Hence
(v, w) L
(a
i
,a
j
)x
v u v, v L
n
and ( v, w)
X
L
(an,a
j
)x
.
Let
A
1
= (
1
, X,
1
, p
0
, T
1
)
be a nite state automaton accepting the regular language uL
n
. We assume that |T
1
| = 1.
Let
A
2
= (
2
, X(2, $),
2
, q
0
, T
2
)
be a nite state automaton accepting the regular language L
(an,a
j
)x
. With the help of A
1
and A
2
we want to construct a nite state automaton A accepting L
(a
i
,a
j
)x
. Assume that
the length of the word u is m (a
i
.u = a
n
).
We make some remarks. If we input the word (x
1
. . . x
t
1
, y
1
. . . y
t
2
)
X
into A, then we
require A to check:
(P1) whether x
1
. . . x
t
1
is an element of uL
n
or not;
(P2) whether (x
m+1
. . . x
t
1
, y
1
. . . y
t
2
)
X
is an element of L
(an,a
j
)x
or not.
The rst requirement can be checked with the help of A
1
and the second one with the
help of A
2
. Note also that A
2
need to be set to work after reading through the rst m
(the length of u) letters of the input word. That is we have to construct an automaton in
such a way that it
recognizes that we have read the rst m letters;
stacks in the rst instance y
1
. . . y
m
X(2, $), so that after reading through the
rst m letters (x
1
. . . x
m
, y
1
. . . y
m
)
X
, we can set to work A
2
by inputting the letter
(x
m+1
, y
1
). At this stage, the stack has to change to y
2
. . . y
m+1
if y
m+1
= $ and
to y
2
. . . y
m
otherwise. Hence the stack always contains words of length less then or
equal to m.
empty its stack by the end of the procedure. In other words we want our automaton
to read all of (x
m+1
. . . x
t
1
, y
1
. . . y
t
2
)
X
with the help of A
2
. To be more precise,
we need to take into consideration that we may have the following situation occur:
The word (x
1
. . . x
t
1
, y
1
. . . y
t
2
)
X
is already read by the automaton A, but there is
still a word y
k
. . . y
t
2
of length less then or equal to m stacked, so that the word
($, y
k
) . . . ($, y
t
2
) is still waiting to beprocessed by A
2
. To be able to do that,
we need to lengthen the word (x
1
. . . x
t
1
, y
1
. . . y
t
2
)
X
by attaching the letter ($, $)
at most m times, to make sure that the empty word is stacked at the end of the
procedure.
Depending on how many times the letter ($, $) needs to be attached to the word
(x
1
, . . . x
t
1
, y
1
. . . y
t
2
)
X
so that property (P2) could be checked, we let
K
0
= {(v, w)
X
L
(a
i
,a
j
)x
| |v| m > |w|}.
Illustrating this case:
v
..
' E
m
w
..
$ $
w
..
That is K
0
contains words (v, w)
X
L
(a
i
,a
j
)
X
for which properties (P1) and (P2) can be
checked without attaching the letter ($, $).
For 1 j m1 we let
K
j
= {(v, w)
X
L
(a
i
,a
j
)x
| |v| m+j = |w|}.
Illustrating this case:
v
..
$ $ $ $
$ $
' E
m
' E
j
w
..
$
w
..
That is K
j
X
L
(a
i
,a
j
)
X
for which properties (P1) and (P2) can be
checked by attaching j times the letter ($, $).
We let
K
m
= {(v, w)
X
L
(a
i
,a
j
)x
| |v| m < |w|}.
v
..
$ $
$ $ $ $
' E
m
' E
m
w
..
w
..
That is K
m
X
L
(a
i
,a
j
)
X
for which properties (P1) and (P2)
can be checked by attaching at least m times the letter ($, $).
It is immediate that if (v, w)
X
L
(a
i
,a
j
)x
, then (v, w) K
j
for some 1 j m, since
the length of v = u v uL
n
has to be greater then m. Thus,
m
_
j=1
K
j
= L
(a
i
,a
j
)x
. (3.2)
We rst verify that the languages N
j
= K
j
($, $)
j
, (0 j m) are regular. For N
j
we let
L
j
= {0, 1, . . . , m};
I
j
= {w X
| |w| m};
J
j
= {0, 1, . . . , j}.
Attach a failure state FS to
1
and denote
1
= FS. We will explain why this step
was needed when we discuss how the transition function
j
of the below dened automaton
works. Consider the automaton B
j
= (Q
j
, X(2, $) ($, $),
j
, r
j
, F
j
) where
Q
j
= (
2
L
j
I
j
J
j
) FS;
r
j
= (p
0
, q
0
, 0, , 0);
F
j
= T
1
T
2
{m} {} {j},
where denotes the empty word. Before we dene the transition function
j
, we introduce
the following notations: for w = y
1
. . . y
t
I
j
we let
w =
_
y
2
. . . y
t
if w =
w if w =
and dene w $ = w.
Let (p, q, l, w, k) Q
j
, where w = y
1
. . . y
t
and let (x, y) X(2, $). Recall that A
1
is a
nite state automaton with exactly one nal state. Let
p
1
=
_
_
_
1
(p, x) if x = $
p if x = $, p = T
1
FS otherwise
and let
q
1
=
_
_
_
q if l < m
2
(q, (x, y
1
)) if w = , l = m
2
(q, (x, $)) if w = , x = $, l = m
The transition function
i
in given in the following tableau:
j
((p, q, l, w, k), (x, y))
(p
1
, q
1
, l + 1, wy, k) if y = $, l < m, k = 0; (R1)
(p
1
, q
1
, l + 1, w, k) if y = $, l < m, k = 0; (R2)
(p
1
, q
1
, l, wy, k) if x = $ or w = , l = m, k = 0 (R3)
(p
1
, q
1
, l, w, k + 1) if x = $, y = $, l = m, k < j, w = (R4)
FS if x = $, y = $, k > 0 (R5)
FS if x = $, y = $, k < j, w = (R6)
FS if (p, q, l, w, k) = FS (R7)
We now explain in more detail how the transition function works:
With rules (R1) and (R2), we
count how many letters of the input word we have read so far up to m;
ll up the stack with a word of length less then or equal to m;
begin to check with the help of A
1
, whether property (P1) holds for the input
word.
3.5. AUTOMATIC VERSUS SCH
UTZENBERGER AUTOMATIC. 47
With rule (R3), we
continue checking whether property (P1) holds for the input word;
begin checking with the help of A
2
, whether property (P2) holds for the input
word.
With rule (R4), we
continue checking whether property (P2) holds for the input word by emptying
the stack;
begin counting the number of ($, $) letters of the input word.
With rule (R5), we assure that
words are not accepted in which a letter ($, $) is followed by a letter (x, y), x =
$ or y = $.
With rule (R6), we assure that
the stack empties exactly when we read through j times the letter ($, $) con-
secutively.
With rules (R1)-(R3), with the denition of p
1
and with the help of the
assumption that A
1
is an automaton with one nal state we assure that
if (v, w)
X
is accepted by B
j
, then v uL
n
, that is v does not contain a letter
$ followed by a letter x = $.
With rule (R1)-(R4), we assure that
the input word satises property (P2).
We may now deduce that a word w is accepted by B
j
if and only if w N
j
, hence
for each 1 j m the language N
j
is regular. By Proposition 2.2.3, N
rev
j
= ($, $)
j
K
rev
j
is also a regular language. Making use of Lemma 2.2.5, we may deduce that K
rev
j
, and
hence K
j
is a regular language for each 1 j m. By (3.2) we have that L
(a
i
,a
j
)x
is a
nite union of regular languages, and hence is a regular language.
(d) The proof is similar to that of part (c). Basically we need to redene A
2
to be the
automaton accepting the regular language L
(a
j
,an)x
.
3.5 Automatic versus Sch utzenberger automatic.
Considering a regular semigroup S, the property of being Sch utzenberger automatic is a
collection of local properties namely it is required that all R-classes are Sch utzenberger
automatic meanwhile being automatic is a global property. In this section we discuss how
the two notions compare. Proposition 3.2.3 tells us that a group is automatic if and only if
it is Sch utzenberger automatic. We verify that under certain conditions automatic regular
semigroups are Sch utzenberger automatic. Moreover, we give an example that shows
that even strongly Sch utzenberger automatic regular semigroups might not be automatic.
Throughout this section S will denote a semigroup, X a nite generating set for S. We
+
map : X S.
Proposition 3.5.1. Let S be an automatic regular semigroup with automatic structure
(X, L). Assume that for every R-class R of S, there exists a regular language K L such
that K = R. Then S is Sch utzenberger automatic.
Proof. Let (X, L) be an automatic structure for S with uniqueness, s S, and consider
the R-class of s, R
s
. By assumption, there exists a regular language K L such that
K = R
s
. We see that (X, K) is a Sch utzenberger automatic structure for R
s
, since
K
=
= {(u, v) | u, v K, u = v}
X
= L
=
(K K)
X
and
K
x
= {(u, v) | u, v K, u x = v}
X
= L
x
(K K)
X
are regular languages which completes the proof.
Corollary 3.5.2. Let S be an automatic regular semigroup with nitely many R-classes.
Then S is Sch utzenberger automatic.
Proof. Let (X, L) be an automatic structure for S. Let S/R = {R
1
, R
2
, . . . , R
n
} and
choose idempotents e
1
, e
2
, . . . , e
n
so that e
i
R
i
. Let R be an R-class of S with idempotent
e. We show that the assumption of Proposition 3.5.1 is satised, that is we construct a
regular language K L such that K = R. Consider the automaton A = (, X, , p, F),
where
=
n
_
i=1
R
e
i
{p}; F = {R}; : (R
i
, x) R
xe
i
and : (p, x) R
x
.
Since R is a left congruence, is well dened. Note that if a word w x
1
x
2
. . . x
n
is
accepted by this automaton, that is if
p
x
1
R
x
1
x
2
R
(x
2
x
1
)
. . .
R
(x
n1
...x
1
)
xn
R
(xn...x
1
)
= R
is a successful path, then w
rev
R. Let M be the language accepted by this automaton
and let M
= M
rev
. By Proposition 2.2.3, K = M
L is a regular language, moreover we

have that K = R. The assertion follows by Proposition 3.5.1.
3.6. INVARIANCE UNDER THE CHANGE OF GENERATORS. 49
The following counterexample shows that the reverse statement does not hold even if S is
strongly Sch utzenberger automatic.
Example 3.5.3. Let S = [Y
2
; G
1
, G
2
, ], be a Cliord semigroup where the building
blocks are Y
2
, the two element chain, G
1
= a, b | the free group of rank 2, G
2
=
c, d, | c
2
= d
2
= 1 the free product of two cyclic groups of order two, and : G
1
G
2
is
the homomorphism dened by a = c, b = d. Then S is Sch utzenberger automatic, but
it is not automatic.
Indeed, since S has two R-classes, G
1
and G
2
we obtain by Corollary 3.3.2, that S is
R-class automatic. Moreover, the groups G
1
and G
2
are known to be automatic groups,
hence by Proposition 3.2.3 they are Sch utzenberger automatic and we may deduce that S
is strongly Sch utzenberger automatic. However, it is shown in [4] that S is not automatic.
3.6 Invariance under the change of generators.
The following problem is essential to consider when talking about automatic S-acts:
Main Problem: Let (X, L
1
, L
2
, . . . , L
n
) be an automatic structure with respect to the
generating set A
0
for the S-act A. Assume that the nite set Y also generates S and that
the nite set B also generates the S-act A. Do there exist regular languages K
1
, . . . , K
m
over Y such that (Y, K
1
, . . . , K
m
) forms an automatic structure with respect to the gen-
erating set B for the S-act A?
We can split the above problem into two subproblems:
Changing the generators of S. Let (X, L
1
, L
2
, . . . , L
n
) be an automatic structure
with respect to the generating set A
0
for the S-act A. Assume that the nite set Y also
generates S. Do there exist regular languages K
1
, . . . , K
n
1
, . . . , K
n
)
forms an automatic structure with respect to the same generating set A
0
for the S-act A?
Changing the generators of A. Let (X, L
1
, L
2
, . . . , L
n
) be an automatic structure
0
for the S-act A. Assume that the nite set B also
generates the S-act A. Do there exist regular languages K
1
, . . . , K
m
over X such that
(X, K
1
, . . . , K
m
) forms an automatic structure with respect to the generating set B for
the S-act A?
Giving a positive answer to both of the subproblems certainly gives a positive answer
to our main problem. We can rst change the generating set of S and then the generating
set of the S-act A.
It is known that automaticity of semigroups depends on the choice of the nite gen-
erating set. Hence with Proposition 3.1.10 in mind, we expect that in general we cannot
change the set of generators either of S or of A without losing the property of being au-
tomatic. We will verify in this section that if S is a semigroup with local right identities,
then an armative answer can be given to both of the subproblems, and hence to our
main problem.
As in the previous sections S will denote a semigroup, X a nite generating set for S.
We assume that X S and denote by : X
+
S the homomorphism extending the
identity map : X S. Also, S
0
= S {0} denotes the semigroup obtained by adjoining
a zero element 0 to S, and X
0
= X {0}.
We say that a semigroup S has local right identities, if for all s S there exists e E(S)
such that se = s. We note here that regular semigroups have local right identities.
Changing the generators of the semigroup
Proposition 3.6.1. Let S = X be a semigroup with local right identities and assume
that (X, L
1
, L
2
, . . . , L
n
) is an automatic structure with respect to the generating set A
0
for the S-act A. Assume that the nite set Y also generates S. Then there exist regular
languages K
1
, . . . , K
n
1
, . . . , K
n
) forms an automatic structure with
respect to the same generating set A
0
for the S-act A.
The proof of Proposition 3.6.1 follows from Lemmas 3.6.2, 3.6.4 and 3.6.5.
Lemma 3.6.2. Let S = X be a semigroup with local right identities and assume
that (X, L
1
, L
2
, . . . , L
n
) forms an automatic structure with respect to the generating set
A
0
= {a
1
, . . . , a
n
} for the S-act A. Assume that the nite set Y also generates S. Then
there exists a nite set Y
Y and regular languages K

1
, . . . , K
n
over Y
such that
(Y

, K
1
, . . . , K
n
) forms an automatic structure with respect to the same generating set
A
0
for the S-act A.
Proof. Let X = {x
1
, . . . , x
k
} and Y = {y
1
, . . . , y
m
}. Let : Y
+
S be the homo-
morphism extending the identity map on Y . Fix words v
1
, v
2
, . . . , v
k
over Y such that
x
i
= v
i
, (1 i k). Let Y
= Y {e
1
, . . . , e
m
}, where y
j
e
j
= y
j
for all 1 j m.
Clearly, since Y
Y , we have that Y
also generates S and that there exists a surjective

homomorphism : (Y
)
+
S such that |
Y
= . Making use of the fact that e
i
is a right
identity for y
i
, we have that for each word v
i
there exists f
i
{e
1
, . . . , e
m
} such that
(v
i
(f
i
)
t
) = v
i
, t N. Hence, by attaching the appropriate local right identities to the
words v
1
, . . . , v
k
, we obtain words w
1
, . . . , w
k
over Y
of equal length such that

x
i
= w
i
= v
i
= v
i
. (3.3)
Consider the map
: X (Y
)
+
, x
i
w
i
,
and extend it to a semigroup homomorphism : X
+
(Y
)
+
. In particular we have that
the diagram
X
+
(Y
)
+
S
d
d
d

is commutative, since
x = v
i
= v
i
= w
i
= (x), (3.4)
hence = . Let
K
i
= L
i
(1 i n).
By Proposition 2.2.3, K
i
is a regular language for all (1 i n). We verify that
(Y
, K
1
, K
2
, . . . , K
n
)
is an automatic structure with respect to the generating set A
0
for the S-act A.
Since (X, L
1
, L
2
, . . . , L
n
) is an automatic structure for the S-act A, we have that
n
j=1
a
j
.L
j
= A. To show that

n
j=1
a
j
.K
j
= A, let a A. Then there exists a
i
A
0
and u L
i
, such that a
i
.u = a. On the other hand u = (u) by (3.4) and also by
denition u K
i
. Thus we obtain that a
i
.(u) = a and we deduce that
n
_
j=1
a
j
.K
j
= A.
Let (a
i
, a
j
) A
0
A
0
. We show that the languages K
(a
i
,a
j
)=
and K
(a
i
,a
j
)y
, y Y
are
regular. Let : X(2, $)
+
Y
(2, $)
+
be the semigroup homomorphism induced by the
map
(x, y) (x, y)( )
Y
, (x, y) X X.
Since for each 1 i, j k the length of x
i
w
i
and x
j
w
j
is the same, we have that
for all (u
, v
) X
+
X
+
(u
, v
)
X
= (u
, v
)
Y
= (u
, v
)( )
Y
and we may deduce that

K
(a
i
,a
j
)=
= {(u, v) K
i
K
j
| a
i
.u = a
j
.v}
Y
= {(u
, v
) | (u
, v
) L
i
L
j
, a
i
.u
= a
j
.v
}
Y
= {(u
, v
) L
i
L
j
| a
i
.u
= a
j
.v
}( )
Y
= {(u
, v
) L
i
L
j
| a
i
.u
= a
j
.v
}
X
= L
(a
i
,a
j
)=
,
hence is a regular language by Proposition 2.2.3. Let y Y
and w X
+
such that
w y. Then,
K
(a
i
,a
j
)y
= {(u, v) K
i
K
j
| a
i
.(u y) = a
j
.v}
Y
= {(u
, v
) | (u
, v
) L
i
L
j
, a
i
.(u
y) = a
j
.v
}
Y
= {(u
, v
) | (u
, v
) L
i
L
j
, a
i
.(u
w) = a
j
.v
}
Y
= {(u
, v
) L
i
L
j
| a
i
.(u
w) = a
j
.v
}( )
Y
= {(u
, v
) L
i
L
j
| a
i
.(u
w) = a
j
.v
}
X
= L
(a
i
,a
j
)w
,
hence is regular by Lemma 3.4.8 and Proposition 2.2.3.
Denition 3.6.3. Let (X, L
1
, . . . L
n
) be an automatic structure for the S-act A. Let
L =
n
j=1
L
j
. We say that a letter x X is dispensable in L, if whenever a word w L
contains the letter x, the word w obtained from w by deleting all occurrences of x in w
represents the same element of S as w.
We note that in Lemma 3.6.2, all supplementary letters added to Y are dispensable in
K =
n
j=1
K
j
. To be more accurate: let w K. Then w = (x
i
1
. . . x
it
) = w
i
1
. . . w
it
for
some (x
i
1
. . . x
it
) L
i
. Recall that w
i
k
is obtained from v
i
k
by attaching an appropriate
local right identity (supplementary letter) to it, and also by (3.3) v
i
j
= w
i
j
. That is
deleting a supplementary letter from w means substituting certain w
i
j
s by v
i
j
, which does
not have any aect on the element represented by w.
Before we nish proving Proposition 3.6.1 we invoke the following technical lemma
which follows from the proofs of [12, Lemma 3.1] and [9, Theorem 2.4.1.].
Lemma 3.6.4. Let S be a semigroup generated by X = {x
1
, . . . , x
m
, e}. Assume that
Y = {x
1
, . . . , x
m
} also generates S and let e = x
i
1
. . . x
i
k
. Dene : X
+
Y
+
as follows:
For each word w, let w be the word obtained from w by substituting all kth occurrences
of e by the word x
i
1
. . . x
i
k
and delete all other occurrences. Then the following hold:
(1) If L is a regular language over X, then L is a regular language over Y .
(2) If J is a regular language over X(2, $), then
(J
l
)
r
= {(u, v) | (u, v)
X
J}
Y
is a regular language over Y (2, $).
Lemma 3.6.5. Let (X, K
1
, . . . , K
n
) be an automatic structure for the S-act A with respect
to the generating set A
0
= {a
1
, . . . , a
n
}, where X = {x
1
, . . . , x
m
, e}. Let K =

n
j=1
K
j
.
Assume that e is dispensable in K and that Y = {x
1
, . . . , x
m
} also generates S. Then
there exist regular languages L
1
, . . . , L
n
over Y such that (Y, L
1
, . . . , L
n
) is an automatic
structure for the S-act A.
Proof. Assume that e = y
1
. . . y
k
, y
i
Y (1 i k). Let : Y
+
S be the
homomorphism extending the identity map on : Y S. Dene the map : X
+
Y
+
in the following way. For each word w X
+
, replace all kth occurrences of e by y
1
. . . y
k
and delete all other occurrences of e in w. Let L
i
= K
i
, (1 i n). By Lemma 3.6.4(1),
L
i
is a regular language for each 1 i n. Let a
i
, a
j
A
0
. To show that (X, L
1
, . . . , L
n
)
forms an automatic structure with respect to the generating set A
0
for A, we need to
verify that the languages L
(a
i
,a
j
)=
, L
(a
i
,a
j
)y
(y Y ) are regular. First we note that since
e is dispensable in K we have by denition for all u
K that (u
) = u
, hence by
letting u u
we obtain
a
i
.u
= a
i
.u
= a
i
.(u
) = a
i
.u = a
i
.u.
Now we may deduce using Lemma 3.6.4 (2) that
L
(a
i
,a
j
)=
= {(u
, v
) L
i
L
j
| a
i
.u
= a
j
.v
}
Y
= {(u
, v
) | (u
, v
) = (u, v) for some (u, v) K

i
K
j
,
a
i
.u = a
j
.v}
Y
= {(u, v) L
i
L
j
| (u, v)
X
K
(a
i
,a
j
)=
}
Y
= (K
(a
i
,a
j
)=
l
)
r
is a regular language. Let y Y and w X
+
such that y = w. Then
L
(a
i
,a
j
)y
= {(u
, v
) L
i
L
j
| a
i
.(u
y) = a
j
.v
}
Y
= {(u
, v
) | (u
, v
) = (u, v) for some (u, v) K

i
K
j
,
a
i
.(u w) = a
j
.v}
Y
= {(u, v) L
i
L
j
| (u, v)
X
K
(a
i
,a
j
)w
}
Y
= (K
(a
i
,a
j
)w
l
)
r
is a regular language by Lemma 3.4.8 and Lemma 3.6.4(2).
Proof of Proposition 3.6.1. Assume that (X, L
1
, L
2
, . . . , L
n
) is an automatic structure
0
for the S-act A. Assume that the nite set Y also
generates S. Then, according to the proof of Lemma 3.6.2, there exists a nite set Y
Y
and regular languages N
1
, . . . , N
n
over Y
such that (Y

, N
1
, . . . , N
n
structure for A with generating set A
0
. Moreover, all letters added to Y are dispensable
in N =

n
j=1
N
j
. Making use of Lemma 3.6.5 we can remove the supplementary letters
added to Y one by one and obtain regular languages K
1
, . . . , K
n
such that (Y, K
1
, . . . , K
n
)
forms an automatic structure with respect to the generating set A
0
for the S-act A.
Corollary 3.6.6. Let S = X be a regular semigroup and R be an R-class of S. As-
sume that (X, L) is a Sch utzenberger automatic structure for R and that the nite set Y
also generates S. Then there exists a regular language K over Y such that (Y, K) is a
Sch utzenberger automatic structure for R.
Proof. Assume that (X, L) is a Sch utzenberger automatic structure for the R-class R
of the regular semigroup S. Then by Proposition 3.2.1 there exists a regular language L
1
over X
0
= X {0} such that (X
0
, L
1
) forms an automatic structure for the right S
0
-act
R
0
with respect to a one element generating set {s}. Since Y generates S, we clearly
have that Y {0} generates S
0
. Applying Proposition 3.6.1, we deduce that there exists
a regular language L
2
over Y
0
such that (Y
0
, L
2
) forms an automatic structure for R
0
with generating set {s}. Pulling back the argument for the R-class R with the help of
Proposition 3.2.1 we conclude that there exists a regular language K over Y such that
(Y, K) forms a Sch utzenberger automatic structure for R.
Changing the generators of the act.
Proposition 3.6.7. Let S be a semigroup with local right identities. Let (X, L
1
, . . . , L
n
) be
an automatic structure with respect to the generating set A
0
for the S-act A. Assume that
the nite set B also generates A. Then there exist regular languages K
1
, K
2
, . . . , K
m
over
X, such that (X, K
1
, . . . , K
m
) forms an automatic structure with respect to the generating
set B for the S-act A.
The proof of the proposition is based on Lemmas 3.6.8, 3.6.9.
Lemma 3.6.8. Let S be a semigroup with local right identities and (X, L
1
, . . . , L
n
) be an
automatic structure with respect to the generating set A
0
for the S-act A. Let a
n+1
A
and B = A
0
{a
n+1
}. Then there exist regular languages K
1
, . . . , K
n+1
over X such that
(X, K
1
, . . . , K
n+1
) forms an automatic structure with respect to the generating set B for
the S-act A.
Proof. By Proposition 3.4.4 we can assume that (X, L
1
, . . . , L
n
) is an automatic struc-
ture with uniqueness for the S-act A. Let a
n+1
A. Then there exists a unique generator
a
i
A
0
and a unique word u L
i
such that a
i
.u = a
n+1
. Let e be a local right identity
of S such that u e = u. Let w be a word over X representing e and L
n+1
= {w}. Note
that (uw) = u, and so
a
n+1
= a
i
.u = a
i
.(u w) = (a
i
.u) w = a
n+1
.w.
That is
a
n+1
.w = a
n+1
(3.5)
holds. We clearly have that

n+1
j=1
a
j
.L
j
= A. To show that (X, L
1
, . . . , L
n+1
) indeed
forms an automatic structure with respect to the generating set B we need to verify that
for all 1 j n the languages L
(a
j
,a
n+1
)=
, L
(a
j
,a
n+1
)x
and L
(a
n+1
,a
j
)x
are regular.
Since (X, L
1
, . . . , L
n
) is an automatic structure with uniqueness and since a
n+1
.w =
a
n+1
by (3.5), we have that
a
j
.v = a
n+1
.w a
j
.v = a
n+1
a
j
= a
i
and v u.
Thus L
(a
j
,a
n+1
)=
= for all 1 j n, i = j and L
(a
i
,a
n+1
)=
= (u, w)
X
, and hence they
are regular.
Next we consider the language L
(a
j
,a
n+1
)x
(1 j n). Let L
j
= {v L
j
| a
j
.(v x) =
a
n+1
}. By Lemma 3.1.8, L
j
is a regular language, and so
L
(a
j
,a
n+1
)x
= {(v, w) L
j
L
n+1
| a
j
.(v x) = a
n+1
.w}
X
= {(v, w) L
j
L
n+1
| a
j
.(v x) = a
n+1
}
X
= (L
j
L
n+1
)
X
= (L
j
{w})
X
is a regular language by Proposition 2.3.1.
Finally consider the language L
(a
n+1
,a
j
)x
. Let c = a
n+1
.(wx) and L
j
= {v L
j
| a
j
.v =
c}. By Lemma 3.1.9, L
j
is a regular language, and hence
L
(a
n+1
,a
j
)x
= {(w, v) L
n+1
L
j
| a
n+1
.(w x) = a
j
.v}
X
= {(w, v) L
n+1
L
j
| c = a
j
.v}
X
= (L
n+1
L
j
)
X
= ({w} L
j
)
X
is a regular language by Proposition 2.3.1, proving that (X, L
1
, L
2
, . . . , L
n+1
) is indeed an
automatic structure with respect to the generating set B for the S-act A.
Lemma 3.6.9. Let S be a semigroup with local right identities and let (X, L
1
, . . . , L
n
) be
0
= {a
1
. . . , a
n
} for the S-act
A. Assume that B = {a
1
, . . . , a
n1
} also generates the S-act A. Then there exist regular
languages K
1
, . . . , K
n1
over X such that (X, K
1
, . . . , K
n1
with respect to the generating set B for the S-act A.
Proof. Let (X, L
1
, . . . , L
n
) be an automatic structure with uniqueness with respect to
the generating set A
0
= {a
1
. . . , a
n
} for the S-act A. Assume that B = {a
1
, . . . , a
n1
}
also generates the S-act A. Then there exists a unique generator a
i
B and u L
i
such
that a
i
.u = a
n
. Dene
K
j
= L
j
for all (1 j n 1, i = j) and K
i
= L
i
uL
n
.
By Proposition 2.2.3, K
i
is a regular language, and we clearly have that
n1
_
j=1
a
j
.K
j
= A.
To prove that (X, K
1
, . . . , K
n1
) forms an automatic structure with respect to the gen-
erating set B, we need to verify that the languages K
(a
i
,a
j
)=
, K
(a
i
,a
j
)x
and K
(a
j
,a
i
)x
, x
X, (1 j n 1), are regular, since for all j = i, (1 j n 1) L
j
= K
j
.
Case 1. If a
j
= a
i
, then
K
(a
i
,a
j
)=
= {(v, w) K
i
K
j
| a
i
.v = a
j
.w}
X
= L
(a
i
,a
j
)=
{(v, w) uL
n
L
j
| a
i
.v = a
j
.w}
X
,
is a regular language by Lemma 3.4.9 and by the fact that (X, L
1
, . . . , L
n
) is an automatic
structure for A.
Case 2. If a
j
= a
i
, then
K
(a
i
,a
i
)=
= {(v, w) K
i
K
i
| a
i
.v = a
i
.w}
X
= L
(a
i
,a
j
)=
{(v, w) uL
n
L
i
| a
i
.v = a
i
.w}
X
{(w, v) L
i
uL
n
| a
i
.w = a
i
.v}
X
(u, u)
X
L
(an,an)=
= L
(a
i
,a
j
)=
(u, u)
X
L
(an,an)=
,
is a regular language by Lemma 3.4.9 and by the fact that (X, L
1
, . . . , L
n
) is an automatic
structure for A.
Case 3. If a
j
= a
i
, then
K
(a
i
,a
j
)x
= {(v, w) K
i
K
j
| a
i
.(v x) = a
j
.w}
X
= L
(a
i
,a
j
)x
{(v, w) uL
n
L
j
| a
i
.(v x) = a
j
.w}
X
and
K
(a
j
,a
i
)x
= {(w, v) K
j
K
i
| a
j
.(w x) = a
i
.v}
X
= L
(a
j
,a
i
)x
{(w, v) L
j
uL
n
| a
j
.(w x) = a
i
.v}
X
are regular languages by Lemma 3.4.9 and by the fact that (X, L
1
, . . . , L
n
) is an automatic
structure for A.
Case 4. If a
j
= a
i
, then
K
(a
i
,a
i
)x
= {(v, w) K
i
K
i
| a
i
.(v x) = a
j
.w}
X
= L
(a
i
,a
i
)x
{(v, w) uL
n
L
i
| a
i
.(v x) = a
i
.w}
X
{(w, v) L
i
uL
n
| a
i
.(w x) = a
i
.v}
X
(u, u)
X
L
(an,an)x
,
is a regular language by Lemma 3.4.9, Proposition 2.2.3 and the fact that (X, L
1
, . . . , L
n
)
is an automatic structure for A.
Proof of Proposition 3.6.7. Assume that (X, L
1
, . . . , L
n
0
= {a
1
. . . , a
n
} for the S-act A. Assume that
B = {b
1
, . . . , b
m
} also generates A. Then, by Lemma 3.6.8, there exist regular languages
N
1
, . . . , N
n+m
over X, such that (X, N
1
, . . . , N
n+m
respect to the generating set C = A
0
B for the S-act A. Since B also generates the
S-act A, by applying Lemma 3.6.9 we can subtract the generators of A
0
from C one by
one, so that we obtain regular languages K
1
, . . . , K
m
over X, such that (X, K
1
, . . . , K
m
)
forms an automatic structure with respect to the generating set B for the S-act A.
Proposition 3.6.7 has the following consequences, which provide a generalization of [9,
Theorem 1.1] and of [4, Proposition 1.4].
Corollary 3.6.10. Let S be a semigroup with local right identities. Assume that (X, L) is
an automatic structure for S and that the nite set Y also generates S. Then there exists
a regular language K over Y such that (Y, K) forms an automatic structure for S.
Proof. Let (X, L) be an automatic structure for the semigroup S, where S has local
right identities. Assume that X = {x
1
, . . . , x
n
}. Then by Proposition 3.1.10, S is an
automatic right S-act with generating set X. In particular we have that there exist regular
languages L
1
, . . . , L
n
over X such that (X, L
1
, . . . , L
n
respect to the generating set X for the right S-act S.
Assume that Y = {y
1
, . . . , y
m
} also generates the semigroup S. Then as shown in
Proposition 3.1.10, Y generates S as an S-act as well, hence by Proposition 3.6.7, there
exist regular languages K
1
, . . . , K
m
such that (Y, K
1
, . . . , K
m
with respect to the generating set Y for the S-act S. Making use of Proposition 3.1.10
repeatedly, we obtain that there exists a regular language K over Y such that (Y, K) forms
an automatic structure for the semigroup S.
Corollary 3.6.11. Let S be a regular semigroup. Assume that (X, L) is an automatic
structure for S and that the nite set Y also generates S. Then there exists a regular
language K over Y such that (Y, K) forms an automatic structure for S.
Proof. Since every regular semigroup has local right identities, our corollary immedi-
ately follows from Corollary 3.6.10.
We will make use of the following useful fact later on.
Proposition 3.6.12. Let S be a regular semigroup generated by a nite set X and R be an
R-class of S. Let s, t R. Assume that (X
0
, L) forms an automatic structure with respect
to the generating set {s} for the right S
0
-act R
0
. Then there exists a regular language K
over X
0
, such that (X
0
, K) forms an automatic structure with respect to the generating
set {t} for R
0
.
3.7. FELLOW TRAVELLER PROPERTY I. 59
Proof. Since (X
0
, L) forms an automatic structure with respect to the generating set
{s} for R
0
, we have that s L = R
0
and that the languages
L
=
= {(u, v) | u, v L, s u = s v}
X
L
x
= {(u, v) | u, v L, s ux = s v}
X
(x X
0
)
are regular. Since s R t, we also have that s = tt
s, t
V (t). Let w be a word over X

such that w = t
s and let K = wL. Note that s = t w. Clearly K is a regular language

over X
0
and t K = t (wL) = s L = R
0
. We verify that (X
0
, K) is an automatic
structure with respect to the generating set {t} for R
0
. Note that (k
1
, k
2
) K
=
if and only
if k
1
wl
1
and k
2
wl
2
for some l
1
, l
2
L, and t wl
1
= t wl
2
. Since t w = s, it follows
that twl
1
= twl
2
if and only if sl
1
= sl
2
if and only if (l
1
, l
2
) L
=
. Hence, we have that
K
=
= (w, w)
X
L
=
and so it is regular. Similarly we have that K
x
= (w, w)
X
L
x
, x X
0
proving that (X
0
, K) is an automatic structure with respect to the generating set {t} for
R
0
.
To summarize the main result of this section, we have:
Theorem 3.6.13. Let S = X be a semigroup with local right identities and assume that
(X, L
1
, L
2
, . . . , L
n
) forms an automatic structure with respect to the generating set A
0
for
the S-act A. Assume that the nite set Y also generates S and that the nite set B also
generates the S-act A. Then there exist regular languages K
1
, . . . , K
m
over Y such that
(Y, K
1
, . . . , K
m
) forms an automatic structure with respect to the generating set B for the
S-act A.
3.7 Fellow traveller property I.
In this section we rst associate a directed labelled graph
X
(A, S) to each S-act A and
introduce the notion of distance in
X
(A, S). With these tools, we give the denition of
the fellow traveller property and claim that the introduced notion is a generalization of the
fellow traveller property given for semigroups and groups. Finally we prove in this section
that if A is an automatic S-act, then
X
(A, S) possesses the fellow traveller property.
As before, S will denote a semigroup, X a nite generating set for S. We assume that
X S and we denote by : X
+
S the homomorphism extending the identity map
: X S. We denote by S
0
the semigroup obtained by adjoining a zero element 0 to S,
and we let X
0
= X {0}. If S is a group, then we will assume that X is closed under
taking inverses.
The associated graph
Intuitively we can think of an S-act A as a directed labelled graph
X
(A, S), in which
the vertices are elements of A, the labels are elements of X and there is an arrow from
a to b with label x precisely when a.x = b. We write the arrows of
X
(A, S) as ordered
triples (a, x, b) indicating that a.x = b. We let V(
X
(A, S)) denote the set of vertices and
A(
X
(A, S)) denote the set of arrows of
X
(A, S). Clearly
X
(A, S) is not necessarily a
connected graph.
We dene a path between two vertices a and b of =
X
(A, S) to be a sequence of
edges:
a = a
0
x
1
a
1
x
2
a
2
. . . a
n1
xn
a
n
= b
such that either (a
i
, x
i
, a
i+1
) A() or (a
i+1
, x
i
, a
i
) A(), (0 i n 1) and say
that the length of the path is n. For a, b V(), we dene the distance d
(a, b) between
a and b to be the length of the shortest path connecting a and b and say that the distance
is innite if a and b belong to dierent components of .
We give a list of S-acts A, when
X
(A, S) or a graph related to
X
(A, S) is a well-
known graph.
Example 3.7.1. If A is the right S-act S (see Example 3.1.4), then =
X
(A, S) is
the right Cayley graph =
X
(S) of S. Bearing in mind the denition of distance in
the Cayley graph of a semigroup introduced in the Preliminaries, we have for all a, b S
that d
(a, b) = d
(a, b). If S is a group then is a connected graph and (g, x, h) A()

if and only if (h, x
1
, g) A(). In other words, any two vertices are connected via a
directed path. To make it clear whether we consider S as a semigroup or a right S-act we
write Cayley graph
X
(S) of S and the graph
X
(S, S) respectively.
Example 3.7.2. If A is the right S
0
-act R
0
, where R is an R-class of S (see Example
3.1.6), then
X
(R
0
, S
0
) restricted to R is the Sch utzenberger graph S
X
(R) of R.
Fellow traveller property
Recall the following notation. If u x
1
. . . x
m
and t 1, then
u(t) =
_
x
1
. . . x
t
if t m,
x
1
. . . x
m
if t m.
Denition 3.7.3. Let A be an S-act regularly generated by L
1
, . . . , L
n
and the set A
0
=
{a
1
, . . . , a
n
}. The graph =
X
(A, S) is said to have the fellow traveller property with
respect to L
i
, L
j
and a
i
, a
j
, if there exists a constant k N such that whenever (u, v)
L
i
L
j
with d
(a
i
.u, a
j
.v) 1, then d
(a
i
.u(t), a
j
.v(t)) k for all t 1. We say that
X
(A, S) possesses the fellow traveller property with respect to L
1
, . . . , L
n
and A
0
, if it
possesses the fellow traveller property with respect to any two regular languages L
i
, L
j
and corresponding generators a
i
, a
j
.
We have seen that if S is a semigroup then
X
(S, S) is the Cayley graph of S. Now we
show that the fellow traveller property for S-acts is a generalization of the fellow traveller
property given for semigroups and groups.
Proposition 3.7.4. Let S be a semigroup generated by a nite set X = {x
1
, . . . , x
n
}.
Then the Cayley graph of S possesses the fellow traveller property with respect to some
regular language L if and only if
X
(S, S) possesses the fellow traveller property with
respect to some regular languages L
1
, . . . , L
n
and X.
Proof. () Assume that the Cayley graph =
X
(S) of S has the fellow traveller
property with respect to a language L. Then L = S and there exists a constant k N
such whenever d
(u, v) 1 with u, v L then d
(u(t), v(t)) k for all t 1. As in

Proposition 3.1.10, the languages L
i
= {u X
+
| x
i
u L} (1 i n) and X regularly
generates the S-act S. Let =
X
(S, S). Choose languages L
i
, L
j
and let (u, v) L
i
L
j
such that d
(x
i
u, x
j
v) 1. Bearing in mind that d
(a, b) = d
(a, b) for all a, b S,

(see Example 3.7.1) we have that d
(x
i
u, x
j
v) 1, and hence d
((x
i
u)(t), (x
j
v)(t)) k
holds for all t 1. In particular we have that d
(x
i
(u(t)), x
j
(v(t))) k for all t 1,
proving that the fellow traveller property holds in with respect to L
i
, L
j
and x
i
, x
j
.
Since L
i
, L
j
were arbitrarily chosen, we may deduce that
X
(S, S) possesses the fellow
traveller property.
() Assume that =
X
(S, S) possesses the fellow traveller property with respect to
some regular languages L
1
, . . . , L
n
and X. Then
n
j=1
x
j
L
j
= S and for any two chosen
languages L
i
, L
j
and corresponding generators x
i
, x
j
, there exists a constant k N such
that whenever d
(x
i
u, x
j
v) 1 with (u, v) L
i
L
j
then d
(x
i
(u(t)), x
j
(v(t))) k
for all t 1. As in Proposition 3.1.10, we let L =

n
j=1
x
j
L
j
. Then L = S. We
claim that the Cayley graph =
X
(S) has the fellow traveller property with respect
to L. Let N N be a constant such that for any two generators x
k
and x
m
that are
connected via a path in the distance d
(x
i
, x
j
) N. Let M = max(k, N). Assume
that d
(u, v) 1, (u, v L). Then u = x

i
u and v = x
j
v, where x
i
, x
j
X and
( u, v) L
i
L
j
. It follows that d
(x
i
u, x
j
v) 1 holds and we obtain that for all t 1,
d
(x
i
( u(t)), x
j
( v(t))) k M holds. Since d
(a, b) = d
(a, b) for all a, b S, we have

that for all t 1, d
(x
i
( u(t)), x
j
( v(t))) k M. To nish the proof we need to verify
that d
(x
i
, x
j
) M or equivalently that x
i
and x
j
are connected in = . But the latter
fact follows by our assumptions, since x
i
( u x) = x
j
v holds for some x X{}, where
denotes the empty word. That is, we have the following path in
x
i
u
x
i
u
x
x
i
u x = x
j
v x
j
v
We may deduce that the Cayley graph of S possesses the fellow traveller property with
respect to L.
Next we verify that if A is an automatic S-act, then
X
(A, S) possesses fellow traveller
property. We follow the group and semigroup theoretical proofs.
Proposition 3.7.5. Let S be a semigroup generated by a nite set X. Let A be an
automatic S-act. Then there exist regular languages L
1
, . . . , L
n
over X, such that
X
(A, S)
has the fellow traveller property with respect to L
1
, . . . , L
n
and A
0
.
Proof. Let (X, L
1
, . . . , L
n
) be an automatic structure for A with respect to the gener-
ating set A
0
= {a
1
, . . . , a
n
}. For each regular language L
(a
i
,a
j
)x
, x X {=} consider a
nite state automaton A
(a
i
,a
j
)x
accepting it and choose a constant N N, such that N is
greater then the number of states of any of the automata dened. Let =
X
(A, S).
Choose regular languages L
i
, L
j
and assume that d
(a
i
.u, a
j
.v) 1 holds, where
(u, v) L
i
L
j
. Without loss of generality we can assume that a
i
.(u y) = a
j
.v for
some y X {}, where denotes the empty word. Then (u, v)
X
is accepted by the
automaton A
(a
i
,a
j
)x
, where x = y if y X and x is the symbol = if y = . Start
reading the word (u, v)
X
in A
(a
i
,a
j
)x
and assume that after reading the rst t letters
(u(t), v(t))
X
, we are at state q. Let ( u, v)
X
be the shortest word over X(2, $) such that
reading ( u, v)
X
from state q, we arrive at a nal state of A
(a
i
,a
j
)x
. Clearly, the length
of ( u, v)
X
is less then N, since the number of states of the considered automaton is less
then N. Furthermore, since (u(t), v(t))
X
( u, v)
X
is accepted by A
(a
i
,a
j
)x
, we have the
following diagram in :
u
............................................................................................................................... .............................................................................................................................................. z
............................................................................................................................................................................................................................. ....................................................................
X
.................................................... .......................................................................
X
... ......... ........................ .......................................... x ....................................................................................... .............................................................. q
u(t)
a
j
x
v(t)
v
2N+1
a
i
c
T
That is,
d
(a
i
.u(t), a
j
.v(t)) | u| +| v| +|x| 2N + 1,
which proves that the fellow traveller property holds in with respect to L
i
, L
j
and a
i
, a
j
.
Since L
i
, L
j
were arbitrarily chosen, we may deduce that possesses the fellow traveller
property with respect to L
1
, . . . , L
n
and A
0
.
Corollary 3.7.6. Let S be an automatic semigroup. Then the Cayley graph of S possesses
the fellow traveller property.
Proof. Immediately follows from Proposition 3.1.10 and Propositions 3.7.4 and 3.7.5.
Corollary 3.7.7. Let S be a monoid and assume that A = S/R is an automatic left
S-act. Then
X
(A, S) has the fellow traveller property.
We will also make use of the following corollary.
Corollary 3.7.8. Let S be a semigroup generated by a nite set X and let A be an S-act.
Assume that (X, L
1
, . . . , L
n
) forms an automatic structure with respect to the generating
set A
0
= {a
1
, . . . , a
n
} for A. Let l N. Then there exists a constant k N, such that
whenever a
i
.uw = a
j
.v holds for some w X
+
, |w| l, where (u, v) L
i
L
j
, then
d
(a
i
.u(t), a
j
.v(t)) k holds in =
X
(A, S).
Proof. Let L = {w X
+
| |w| l}. Since for all a
i
, a
j
A
0
and w L the languages
L
(a
i
,a
j
)w
= {(u, v) | a
i
.uw = a
j
.v}
X
are regular by Lemma 3.4.8, nite state automata
A
(a
i
,a
j
)w
can be dened accepting these languages. We can choose a constant N N
that is greater than the number of states in any of the automata dened. The proof now
continues along the same lines as the proof of Proposition 3.7.5.
In preparation for the next section, we make the following observation. Let S be
a regular semigroup generated by a nite set X and let R be an R-class of S. Then,
for any regular language L over X and s R, for which s L = R
0
holds, the graph
=
X
(R
0
, S
0
) associated to the right S
0
-act R
0
possesses the fellow traveller property
with respect to L and {s}. Indeed, since we have for all s, t R
0
that d
(s u, s v) 2.
This fact ruins our hopes to connect Sch utzenberger automaticity of R and the fellow
traveller property of , and urges us to introduce the following denition.
Denition 3.7.9. Let S be a semigroup and R be an R-class of S. Let L be a regular
language over X and s R such that s L = R. The Sch utzenberger graph = S
X
(R)
is said to have the fellow traveller property with respect to L and {s}, if there exists a
constant k N such that in the Sch utzenberger graph of R whenever u, v L with
d
(s u, s v) 1, then d
(s u(t), s v(t)) k holds for all t 1.

Similarly to Proposition 3.7.5 and Corollary 3.7.8 the following proposition and corol-
lary can be proved:
Proposition 3.7.10. Let S be a regular semigroup generated by a nite set X and let
R be a Sch utzenberger automatic Rclass of S. Let s R. Then there exists a regular
language L over X, such that S
X
(R) has the fellow traveller property with respect to L
and {s}.
Corollary 3.7.11. Let S be a regular semigroup generated by a nite set X and R be an
R-class of S. Let s R and K be a regular language over X satisfying condition (S2)
of Proposition 3.2.1. Let l N. Then there exists a constant k N, such that whenever
s uw = s v for some w X
+
, |w| l, where u, v L, then d
(s u(t), s v(t)) k holds

in the Sch utzenberger graph =
X
(R).
3.8 Fellow traveller property II.
We have seen in the previous section that if A is an automatic S-act, then the fellow
traveller property holds in
X
(A, S). The question naturally arises whether the converse
holds or not:
Problem: Let A be an S-act. Is it true that if
X
(A, S) possesses the fellow traveller
property then A is automatic?
We know that groups can be characterized by the fellow traveller property, while semi-
groups usually cannot. Hence, with Propositions 3.1.10 and 3.7.4 in mind an armative
answer can be given if A is the right S-act S, where S is a group, but if A is the right
S-act S, where S is a semigroup then the answer is expected to be negative. As a reaction
to the result that fellow traveller property does not characterize automatic semigroups,
investigations continued in two directions:
(a) On one hand, semigroup classes closely related to groups came into the focus of
research; i.e. it was shown in [4] that completely simple semigroups can be charac-
terized by the fellow traveller property.
3.8. FELLOW TRAVELLER PROPERTY II. 65
(b) On the other hand, modication of the concept of automaticity has been considered
in [34]. With the introduction of the notion of prex-automatic monoids and the
notion of monoids of nite geometric type the following results are veried:
(i) Prex-automaticity is still a valid generalization of the group theoretical notion
of automaticity.
(ii) Prex-automatic monoids of nite geometric type are characterized by the fel-
low traveller property.
These considerations suggest that if we want to obtain a positive answer to our proposed
problem, then we need to have structural information and a niteness condition on the
S-act A.
The main purpose of this section is to prove
Theorem 3.8.1. Let S be a regular semigroup with nitely many idempotents. Assume
that each R-class possesses the fellow traveller property. Then S is Sch utzenberger auto-
matic.
The key to prove Theorem 3.8.1 is based on the following proposition.
Proposition 3.8.2. Let S be a regular semigroup with nitely many idempotents. Let R
be an arbitrary R-class of S and let e E(S) R. Then R is Sch utzenberger automatic
if and only if H
e
is automatic.
Proposition 3.8.2 is proved in three steps. The rst step is to nd a generating set for
H
e
in terms of generators of S. The second step is to prove that if the Sch utzenberger
graph of R possesses the fellow traveller property then the Cayley graph of H
e
possesses
the fellow traveller property. The nal step is to prove that that if H
e
is automatic, then
R is Sch utzenberger automatic.
First step. Let S be a regular semigroup generated by a set X, and let G be a
maximal subgroup of S. In particular, G is an H-class of S, and the identity element e of
G is an idempotent in S. Recall that e is a left identity in its R-class. Our aim is to nd
a generating set for G. In doing so, we will translate the results from [32] into our setting.
The following concept plays a central role. Consider the R-class R of e and choose
representatives e = r
0
, r
1
, r
2
, . . . of the H-classes contained in R. For each chosen
representative choose an inverse r
0
= e, r
1
, r
2
, . . . (r
i
V (r
i
)). Clearly er
i
= r
i
and
r
i
D r
j
. Illustrating what we have done:
R
e
r
0
= e r
1
r
2
r
3
r
3
r
1
r
2
For a word w over X and a representative r
i
, if r
i
w R e, then we let r
iw
denote the
representative of the H-class of r
i
w.
R
e
r
0
= e r
i
r
i
w
r
iw
We verify
Proposition 3.8.3. If X generates S, then
Y = {r
i
x(r
ix
)
e | r
i
x R e}
generates G as a semigroup.
Proof. First we show that Y G. Let r
i
x(r
ix
)
e Y . Then, r
ix
H r
i
x R e, and hence
R
(r
ix
)
L
r
i
x
contains an idempotent, namely (r
ix
)
r
ix
. It follows by Proposition 2.1.5,
that r
i
x(r
ix
)
L
(r
ix
)
R
r
i
x
R
e
. On the other hand, r
i
x(r
ix
)
H r
ix
(r
ix
)
R e, and
hence R
e
L
r
i
x(r
ix
)
contains an idempotent, namely r
ix
(r
ix
)
. Applying Proposition 2.1.5

one more time, we obtain that r
i
x(r
ix
)
e L
e
R
r
i
x
= G. Illustrating the argument:
R
e
r
0
= e r
i
x(r
ix
)
r
ix
(r
ix
)
r
i
x
r
ix
(r
ix
)
(r
ix
)
r
ix
By Proposition 2.1.5 we also have that for y
1
, y
2
Y , y
1
y
2
G holds. Let g G, and
assume that g = x
1
. . . x
n
, where x
i
X (1 i n). We verify that g can be written in
terms of generators in Y . Clearly g = eg = ex
1
. . . x
n
and hence r
0x
1
...xn
= r
0
= e. We
make the following further observations.
(1) Since e R ex
1
. . . x
n
, we have for all 1 i n that
e R ex
1
. . . x
i
. (3.6)
Moreover, since R is a left congruence, it follows that
h R hx
1
. . . x
i
for all h G, 1 i n. (3.7)
(2) For all 1 i < n,
r
0x
1
...x
i
x
i+1
R e. (3.8)
For, by denition and by (3.6), e R r
0
x
1
. . . x
i
H r
0x
1
...x
i
and hence it follows by
Proposition 2.1.11 that Gr
0
x
1
. . . x
i
= Gr
0x
1
...x
i
. So, there exists h G such that
hx
1
. . . x
i
= hr
0
x
1
. . . x
i
= er
0x
1
...x
i
= r
0x
1
...x
i
.
Multiplying each side on the right by x
i+1
, and making use of (3.7),we obtain that
r
0x
1
...x
i
x
i+1
R e indeed holds.
(3) Since r
0x
1
...x
i
x
i+1
H r
0x
1
...x
i+1
L ((r
0x
1
...x
i+1
)
r
0x
1
...x
i+1
), and since every idempotent
is a right identity in its L-class, we have that
r
0x
1
...x
i
x
i+1
= r
0x
1
...x
i
x
i+1
((r
0x
1
...x
i+1
)
r
0x
1
...x
i+1
). (3.9)
We are now ready to rewrite g in terms of generators of Y :
g = ex
1
x
2
. . . x
n
= r
0
x
1
x
2
. . . x
n
= (r
0
x
1
r
0x
1
r
0x
1
)x
2
. . . x
n
by (3.9),
= r
0
x
1
r
0x
1
(r
0x
1
x
2
r
0x
1
x
2
r
0x
1
x
2
) . . . x
n
by (3.9),
.
.
.
= r
0
x
1
r
0x
1
r
0x
1
x
2
r
0x
1
x
2
r
0x
1
x
2
. . . (r
0x
1
...x
n1
x
n
(r
0x
1
...xn
)
r
0x
1
...xn
).
Using the fact that e is a left identity in its R-class and that r
0x
1
...x
i
R e for all 1 i n
by (3.6), we obtain that
g = (r
0
x
1
r
0x
1
e)(r
0x
1
x
2
r
0x
1
x
2
e) . . . (r
0x
1
...x
n1
x
n
(r
0x
1
...xn
)
e).
To make notation convenient, we write [r

i
, x] instead of r
i
x(r
ix
)
e Y . Proposition
3.8.3 asserts that Y = {[r
i
, x] | r
i
x R e} generates G. By the proof of Proposition 3.8.3,
we have that for g G,
g = x
1
x
2
. . . x
n
= [r
0
, x
1
][r
0x
1
, x
2
] . . . [r
0x
1
...x
n1
, x
n
].
We observe here that the length of a word over X that represents an element of G does
not change when we rewrite it in terms of generators of Y .
Second step. We keep the notation of the rst step. We prove that if R = R
e
is a Sch utzenberger automatic R-class, and E(S) is nite, then G, our chosen maximal
subgroup, is automatic. We need the following lemma and its corollary.
Lemma 3.8.4. Let S be a regular semigroup generated by a nite set X and assume that
E(S) is nite. Let R be an R-class of S. There exists a constant c N such that whenever
s, t R are such that s = tx for some x X, then t w = s for some word w X
+
with
length less then or equal to c.
Proof. Since E(S) is nite, there exists a constant c
1
N such that each idempotent
e E(S) can be represented by a word of length less then or equal to c
1
. For each x X,
choose x
V (x). Since X is nite, there exists a contant c

2
N such that each chosen
element x
V (x) can be represented by a word of length less then or equal to c

2
. Let
c = 2c
1
+ c
2
. Assume that s R t and that sx = t. Then, by Proposition 2.1.8, ss
= tt
for some s
V (s) and t
V (t). Let h S(s
s, xx
). Then, by Proposition 2.1.10,

t
= x
hs
V (xs) = V (t), and so we obtain that

s = tt
s = tt
tt
s = tt
s = t x
hs
s.
By denition h E(S) and so we obtain that x
hs
s can be represented by a word of

length less then or equal to c.
An immediate corollary of Lemma 3.8.4:
Corollary 3.8.5. Let S be a regular semigroup generated by a nite set X and assume
that E(S) is nite. Let R be an R-class of S and denote by the Sch utzenberger graph
of R. Let k N. Then there exists a constant c N such that whenever s, t R are such
that d
(s, t) k, then s w = t for some word w X

+
with length less then or equal to c.
Proof. By denition, if d
(s, t) k, then there exists an undirected path

s = s
0
x
0
s
1
x
1
s
2
. . . s
n1
x
n1
s
n
= t,
such that n k. We want to verify that there exists a directed path with some bounded
length from s to t. Two cases can occur:
(i) If s
i
x
i
= s
i+1
, then our arrow point in the right direction.
(ii) If s
i+1
x
i
= s
i
, then by Lemma 3.8.4, there exists a constant c
1
and a word w X
+
,
whose length is less then or equal to c
1
such that s
i
w = s
i+1
. Hence arrows that
point in the wrong direction can be substituted by a sequence of at most c
1
arrows
pointing in the right direction.
We may now deduce that there is a directed path from s to t in the Sch utzenberger graph
of R, whose length is less then or equal to c = kc
1
.
Proposition 3.8.6. Let S be a regular semigroup generated by a nite set X and let G be
a maximal subgroup of S with identity element e. Assume that E(S) is nite and that the
Sch utzenberger graph of R
e
possesses the fellow traveller property. Then G is an automatic
group. In particular, if R
e
is Sch utzenberger automatic, then G is automatic.
Proof. For the sake of convenience, we assume that e X. We let : X
+
S
denote the homomorphism extending the identity map
1
: X S. Since R
e
possesses the
fellow traveller property, there exists a regular language L over X such that e L = R
e
.
By our assumptions E(S) is nite, and so R
e
contains nitely many H-classes. Choose
representatives e = r
0
, r
1
, . . . , r
n
of the H-classes contained in R
e
and also choose inverses
of the representatives e = r
0
, r
1
, . . . , r
n
(r
i
V (r
i
)). According to Proposition 3.8.3 and
using the notation introduced after Proposition 3.8.3, the nite set Y = {[r
i
, x] | x
X, r
i
x R e} generates G as a semigroup. Let : Y
+
G be the homomorphism
extending the map : Y G; [r
i
, x] r
i
xr
ix
e.
Next, we construct a regular language over Y that is mapped onto G. For this, let
L = {ew eL | (ew) G}.

It is immediate that

L = G. Consider the map
:

L Y
+
; x
1
x
2
. . . x
m
[r
0
, x
1
][r
0x
1
, x
2
] . . . [r
0x
1
...x
m1
, x
m
].
Let K =

L. Since

L = G, we have that K = G. We rst show that K is a regular lan-
guage over Y . Let A
1
= (
1
, X, , p, T) be a nite state automaton accepting the regular
language eL. Let / Y and Z = Y {}. Consider the automaton A = (, Y, , q, F),
where
= (
1
Y ) {q} FS, q = (p, ), F = T Y,
and the transition function is dened as follows:
(q, [r
j
, y]) = ((p, y), [r
j
, y]),
for ( p, [r
i
, x])
1
Y we dene
(( p, [r
i
, x]), [r
j
, y]) =
_
(( p, y), [r
j
, y]) if r
ix
= r
j
,
FS otherwise
and
(FS, [r
j
, y]) = FS.
By the denition of K, it is straightforward, that every word w K is accepted by A.
On the other hand, by the denition of , we have that every word accepted by A is an
element of K. We may now deduce that K is a regular language for which K = G.
We next prove that G is automatic by showing that the fellow traveller property holds
in the Cayley graph of G. The Sch utzenberger graph of R will be denoted by . We will
work with distances both in the Cayley graph of G and in the Sch utzenberger graph
of R. The notation will always make it clear in which graph the distance is understood.
First, we introduce the following upper bounds:
Since Y is nite, there exists c
1
N, such that for every generator [r
i
, x] Y there
exists a word w over X of length less then or equal to c
1
such that [r
i
, x] = w.
Since the set of representatives of the H-classes contained in R
e
is nite, there exists
c
2
N, such that r
i
e, (r
i
V (r
i
), 1 i n) can be represented by a word over X
of length less then or equal to c
2
. In particular e = r
0
can be represented by a word
with length less or equal to c
2
.
Let u, v K for which d
(u, v) 1 holds. We verify that there exists a constant c N

such that d
(u(t), v(t)) c holds for all t 1.

We may assume that u [r
i
, x] = v for some [r
i
, x] Y . Consider e u, e v

L for which
(e u) u and (e v) v. Let w be a word over X such that w = [r
i
, x]. We may
assume that |w| c
1
. Then e uw = e v, and hence d
(e u, e v) c
1
. Making use of
Corollary 3.7.11, we have that there exists k N such that d
(e u(t), e v(t)) k for all

t 1.
Let t 1. We verify that in the Sch utzenberger graph of R, the distance between the
vertices e u(t) and u(t) is less then c
2
. Assume that
u [r
0
, x
1
][r
0x
1
, x
2
] . . . [r
0x
1
...x
l1
, x
l
].
Then e u x
1
. . . x
l
and
u(t) = [r
0
, x
1
][r
0x
1
, x
2
] . . . [r
0x
1
...x
t1
, x
t
]
= r
0
x
1
r
0x
1
(e r
0x
1
)x
2
r
0x
1
x
2
e . . . (e r
0x
1
...x
t1
)x
t
r
0x
1
...xt
e
= (r
0
x
1
r
0x
1
r
0x
1
)x
2
r
0x
1
x
2
. . . r
0x
1
...x
t1
x
t
r
0x
1
...xt
e (since r
i
R e)
= r
0
(x
1
. . . x
t
)r
0x
1
...xt
e (by (3.9))
= e u(t) r
0x
1
...xt
e.
It follows that d
(u(t), e u(t)) c
2
. Similarly, d
(v(t), e v(t)) c
2
and so we have that
d
(u(t), v(t)) d
(u(t), e u(t)) +d
(e u(t), e v(t)) +d
(e v(t), v(t))
c
2
+k +c
2
.
Illustrating this, we have
[ri,xi]
c
2
c
2
.

_
_
.
.
.
.
_

.
.
.

_
.
.
.

w
.
.
.
.
.
.

By Corollary 3.8.5, there exists a constant c

3
N and a word w X
+
such that |w| c
3
and
u(t) (w) = u(t) (ew) = v(t).
It follows that (ew) G. Clearly (ew) can be represented by a word over X with length
less then or equal to 1 +c
3
. According to the remark after Proposition 3.8.3, rewriting a
word over X that represents an element of G does not change the length of the word and
so we have that (ew) can be represented by a word over Y with length less then or equal
to 1 + c
3
. We may now deduce that there exists a word w Y
+
, such that | w| 1 +c
3
,
and u(t) w = v(t) and so d
(u(t), v(t)) 1 +c
3
for all t 1.
Third step. We prove
Proposition 3.8.7. Let S be a regular semigroup nitely generated by X S and as-
sume that E(S) is nite. Let e E(S) and assume that H
e
is automatic. Then R
e
is
Proof. Assume that H
e
is generated by a nite set Y H
e
. Let : Y
+
H
e
denote
the homomorphism extending the identity map
1
: Y H
e
. Let K Y
+
be a regular
language such that (Y, K) form an automatic structure for H
e
with uniqueness. Since E(S)
is nite, D
e
contains nitely many L-classes. We choose representatives r
0
= e, r
1
, . . . , r
n
in each H-class contained in R
e
. For each representative r
i
, we choose an inverse r
i
V (r
i
).
Since S is generated by X, it is also generated by the nite set
Z = X Y {r
0
, . . . , r
n
}.
Let : Z
+
S denote the homomorphism extending the identity map : Z S. Let
L =
n
_
i=1
Kr
i
.
By Greens Lemma, L = R
e
. Moreover, since (Y, K) is an automatic structure with
uniqueness, for each s R
e
, there exists exactly one w L such that w = s, and so
L
=
= {(u, u)|u L}
Z
is a regular language. To show that (Z, L) is a Sch utzenberger
automatic structure for R
e
, we have to show that for each z Z, L
z
We rst x notation. Let
I = {(i, j) | 1 i, j n, r
i
z H r
j
}.
Observe that if (i, j) I, then r
i
zr
j
e H
e
. Indeed, since r
j
r
j
L
r
i
z
R
r
j
, we have
by Proposition 2.1.5, that r
i
zr
j
R
r
i
z
R
r
j
. Since r
j
r
j
L
r
i
zr
j
R
e
, we have r
i
zr
j
e
R
r
i
zr
j
L
e
= H
e
by Proposition 2.1.5. Thus, for each (i, j) I we can x a word w
i,j
Y
+
such that w
i,j
= (r
i
zr
j
e). For each (i, j) I, we let
L
i,j
= L
z
(Y
r
i
Y
r
j
)
Z
.
We rst verify that
L
z
=
_
(i,j)I
L
i,j
, (3.10)
then we show that L
i,j
is a regular language for each (i, j) I. It is immediate that
L
i,j
L
z
for each (i, j) I. To prove the reverse inclusion, we show the following stronger
statement:
( ur
i
, vr
j
)
Z
L
z
r
i
z H r
j
and ( u, v)
Y
K
w
i,j
. (3.11)
First, assume that ( ur
i
, vr
j
)
Z
L
z
. Then ur
i
z = vr
j
and it follows that r
i
z H r
j
. On
the other hand, multiplying each side on the right by r
j
e, we obtain that ur
i
zr
j
e = vr
j
r
j
e.
Since r
j
r
j
R e H v, we have vr
j
r
j
e = v and so ur
i
zr
j
e = v. It follows that ( u, v)
Z
K
w
i,j
.
For the converse, assume that ( u, v)
Z
K
w
i,j
and that r
i
z H r
j
. Then uw
i,j
= v,
and so ur
i
zr
j
e = v holds in S. Multiplying each side on the right by r
j
, we obtain that
ur
i
zr
j
er
j
= vr
j
. Since e R r
j
, we have er
j
= r
j
. By assumption r
i
z H r
j
L r
j
r
j
, and so
r
i
zr
j
r
j
= r
i
z. We may now deduce that ur
i
zr
j
er
j
= ur
i
z = vr
j
, and so ( ur
i
, vr
j
)
Z
L
z
.
Fix (i, j) I. We construct a nite state automaton accepting L
i,j
. To simplify
notation, we let w denote w
i,j
. Since (Y, K) is an automatic structure for H
e
, we have
by Proposition 2.3.14, that K
w
is a regular language. Let A
1
= (
1
, Y,
1
, p
1
, F
1
) be a
nite state automaton accepting K
w
. Consider the nite state automaton A
i,j
= (, Y
{r
i
, r
j
}, , p, F), where
= (
1
{, r
i
} {, r
j
}) {FS}; F = F
1
{r
i
} {r
j
}; p = (p
1
, , ).
The transition function is dened in the following way.
((q, , ), (y
1
, y
2
)) =
_
_
(
1
(q, (y
1
, y
2
)), , ) if y
1
, y
2
Y ,
(
1
(q, ($, y
2
)), r
i
, ) if y
1
= r
i
, y
2
Y ,
(
1
(q, (y
1
, $)), , r
j
) if y
1
Y, y
2
= r
j
,
(q, r
i
, r
j
) if y
1
= r
i
, y
2
= r
j
,
FS otherwise.
((q, r
i
, ), (y
1
, y
2
)) =
_
_
_
(
1
(q, ($, y
2
)), r
i
, ) if y
1
= $, y
2
Y ,
(q, r
i
, r
j
) if y
1
= $, y
2
= r
j
,
FS if y
1
= $.
Similarly,
((q, , r
j
), (y
1
, y
2
)) =
_
_
_
(
1
(q, (y
1
, $)), , r
j
) if y
1
Y, y
2
= $,
(q, r
i
, r
j
) if y
1
= r
i
, y
2
= $,
FS if y
2
= $,
and dene
((q, r
i
, r
j
), (y
1
, y
2
)) = FS.
Let u a
1
. . . a
k
and v b
1
. . . b
m
be words over Z and assume that (u, v)
Z
is accepted
by A
i,j
. Without loss of generality we may assume that k m. Since (u, v)
Z
is accepted
by A
i,j
, we have that the path
p
(a
1
,b
1
)
q
1
(a
2
,b
2
)
. . .
(a
k
,b
k
)
q
k
($,b
k+1
)
. . .
($,bm)
q
m
is successful, and so q
m
F. It follows that a
k
= r
i
and b
m
= r
j
and that the word
(a
1
. . . a
k1
, b
1
. . . b
m1
)
Y
K
w
. Keeping in mind (3.11), we indeed obtain that (u, v)
Z

L
i,j
.
Assume that (u, v)
Z
L
i,j
. Then u a
1
. . . a
k1
r
i
and v b
1
. . . b
m1
r
j
. Moreover,
by (3.11), the word (a
1
. . . a
k1
, b
1
. . . b
m1
)
Y
K
w
and so it is accepted by A
1
. From
the construction of A
i,j
, it follows htat (u, v)
Z
is indeed accepted by A
i,j
.
We may now deduce for each (i, j) I, the language L
i,j
is regular and so L
z
is a
regular language, since by (3.10), it is a nite union of regular languages. We have thus
proved that (Z, L) is a Sch utzenberger automatic structure for R
e
.
Combining Proposition 3.8.6 and 3.8.7, we obtain:
Corollary 3.8.8. Let S be a regular semigroup with nitely many idempotents. Let e
E(S). Then H
e
is automatic if and only if R
e
is Sch utzenberger automatic.
Corollary 3.8.9. Let S be a regular semigroup with nitely many idempotents. Let R be
an R-class of S and assume that the fellow traveller property holds in the Sch utzenberger
graph of R with respect to some regular language.Then R is Sch utzenberger automatic.
Proof. Let e R. If the fellow traveller property holds in R, then H
e
is automatic by
Proposition 3.8.6. If H
e
is automatic, then R
e
is Sch utzenberger automatic by Proposition
3.8.7.
3.9 Fellow traveller property III.
In the previous section, we veried that if S is a regular semigroup with nitely many
idempotents, then the fellow traveller property of S
R
implies that R is Sch utzenberger
automatic. In this section, we give an example which shows that Corollary 3.8.9 does not
hold without the niteness condition.
Let X = {x
0
, x
1
, x
2
, . . .} be an innite set, and consider the symmetric inverse monoid
I
X
. Let T = {
x
i
,x
j
: x
i
x
j
} . Clearly T is an inverse subsemigroup of I
X
, moreover
x
i
,x
j
R
xn,xm
x
i
= x
n
and
x
i
,x
j
L
xn,xm
x
j
= x
m
.
We may deduce that T is an aperiodic 0-bisimple inverse subsemigroup in I
X
. In particular,
T \ {} is a D-class of I
X
containing innitely many idempotents. Further on we denote
x
0
,x
0
by .
Next, we construct a nitely generated inverse subsemigroup S of I
X
containing T.
For this, consider the following three elements of I
X
:
: x
i

_
x
i+2
if i is even,
undened otherwise
: x
i

_
_
_
x
i+2
if i is odd,
x
1
if i = 0,
undened otherwise
3.9. FELLOW TRAVELLER PROPERTY III. 75
: x
i

_
x
i1
if i = 2n + 1, where n is not a prime,
undened otherwise
Let S be the inverse subsemigroup of I
X
generated by Y = T {
1
,
1
,
1
}.
Lemma 3.9.1. S is nitely generated by ,
1
,
1
and
1
.
Proof. We show, that any element of T can be written in terms of ,
1
,
1
,
1
. By
the denition of , , we have that
x
0
,x
2n
=
n
and
x
0
,x
2n+1
=
n+1
. (3.12)
On the other hand
x,y
=
x,x
0

x
0
,y
=
1
x
0
,x

x
0
,y
holds. It follows now by (3.12) that
S is generated by ,
1
,
1
and
1
as a semigroup.
For notational convenience, we introduce the alphabet
Z = {a, b, b
1
, c, c
1
, d, d
1
},
where a corresponds to , b
1
corresponds to
1
, c
1
corresponds to
1
and d
1
cor-
responds to
1
. Consider now the Sch utzenberger graph R of in the D-class T of
S:
x
0
,x
0
c

x
0
,x
1
c
x
0
,x
2
b
x
0
,x
3
d
_
c
x
0
,x
4
b
x
0
,x
5
c
x
0
,x
6
. . .
(Note that we omitted the inverse arrows). Let L = a b
. Our aim is to show that

S
R
possesses the fellow traveller property with respect to L and {}, and that R is not
Lemma 3.9.2. The Sch utzenberger graph S
R
possesses the fellow traveller property with
respect to L and {}.
Proof. All distances considered in the proof will be distances in the Sch utzenberger
graph = S
X
(R). Therefore we write d(s, t) instead of d
(s, t). Clearly L is a regular

language, and L = R. We verify that for all u, v L with d( u, v) 1, we have
that d( u(t), v(t)) 4. The following cases have to be dealt with:
(1) If u = v, then either u v a or u v b
n
or u v c
n
, n N. Thus,
d( u(t), v(t)) 4 clearly follows.
(2) The case, when u a = v is similar to the previous case.
(3) If u b = v, then either u a and v b, or u b
n
and v b
n+1
, n N. In
both cases, we clearly have that d( u(t), v(t)) 4.
(4) If u c = v, then either u a and v c, or u c
n
and v c
n+1
, n N. In
both cases, we clearly have that d( u(t), v(t)) 4.
(5) If u d = v, then u c
n+1
and v b
n
, where n is not a prime. To verify that
d( u(t), v(t)) 4 holds for all t 1, we consider the following cases:
(a) If t is even and t = 2, then
c
t
(c d) = c
t+1
d =
x
0
,x
2t+1
d =
x
0
,x
2t
= b
t
,
and hence d( u(t), v(t)) = 2 4.
(b) If t = 2, then
c
2
d = b,
(c) If t is odd and t is not a prime, then
c
t
(c d) = c
t+1
d =
x
0
,x
2t+1
d =
x
0
,x
2t
= b
t
,
(d) If t is an odd prime, then using the fact that t + 1 is even and making use of
(a), we obtain that
c
t
c
2
d = c
t+1
(c d) = b
t+1
,
and it follows that
c
t
(c
2
d b
1
) = b
t
.
That is, d( u(t), v(t)) = 4.
Lemma 3.9.3. The R-class R of is not Sch utzenberger automatic with respect to L and
.
3.10. FINITE PRESENTABILITY 77
Proof. One can easily see, that L
a
= L
=
= {(u, u) | u L}
Z
, that
L
b
= {(u, v) | u, v L, u b = v}
Z
= (a, b) (b, b)
+
($, b),
and that
L
c
= {(u, v) | u, v L, u c = v}
Z
= (a, c) (c, c)
+
($, c).
Hence, these languages are regular. Clearly, we have that (u, v) L
d
, if and only if
u = c
n+1
and v = b
n
, where n is not a prime. That is to say, that
L
d
= {(c
n+1
, b
n
) | n N, n is not prime}
Z
.
Using the pumping lemma one can easily verify that L
d
is not regular.
3.10 Finite presentability
Automatic groups are nitely presented. As far as automatic semigroups are concerned
an example is given in [5, Example 4.4] for an automatic semigroup that is not nitely
presented. On the other hand, as soon as we get closer to groups, for example consider
completely simple semigroups, we see that the group theoretical result generalizes [4], that
is being automatic does imply nite presentability.
Proposition 3.10.1. Let S be a Sch utzenberger automatic regular semigroup having
nitely many idempotents. Then S is nitely presented.
Proof. By Proposition 3.8.6, the maximal subgroups of S are automatic, and hence
they are nitely presented. According to [32, Theorem 4.1], a regular semigroup S with
nitely many idempotent elements is nitely presented if and only if all maximal subgroups
of S are nitely presented, and so we obtain the desired result.
The following proposition provides a generalization of [4, Corollary 1.2].
Proposition 3.10.2. Let S be an automatic regular semigroup with nitely many idem-
potents. Then S is nitely presented.
Proof. It follows by 3.5.2 that S is Sch utzenberger automatic. Making use of Propo-
sition 3.8.2, we have that the maximal subgroups of S are automatic, and hence they are
nitely presented. Making use of [32, Theorem 4.1] once again, we obtain our desired
result.
We end this section by proving:
Proposition 3.10.3. For a nitely generated completely simple semigroup S, the following
are equivalent:
(1) S is Sch utzenberger automatic;
(2) All maximal subgroups of S are automatic;
(3) S is automatic.
Proof. (1) (2) Finitely generated completely simple semigroups have nitely many
L- and R-classes. Hence, by Proposition 3.8.6 if a completely simple semigroup S is
Sch utzenberger automatic, then all of its maximal subgroups are automatic.
(2) (3) See [4, Theorem 1.1].
(3) (1) If S is automatic, then it is nitely generated, and hence it has nitely many
L- and R-classes. Making use of Corollary 3.5.2, we thus have that S is Sch utzenberger
automatic.
3.11 Equality and word problems
Let S be a semigroup generated by a nite set X. The word problem is said to be solvable
for S, if there exists an algorithm which decides whether or not given any two words
u, v X
+
represent the same element in S or not. Automatic groups and semigroups have
solvable word problem in quadratic time. In this section we will introduce the concept of
the equality problem for S-acts, and show that the equality problem is solvable for the
right S-act S if and only if the word problem is solvable for the semigroup S. Moreover
we consider the following two problems:
Problem 1. Is it true that automatic S-acts have solvable equality problem?
Problem 2. Is it true that (strongly) Sch utzenberger automatic regular semigroups have
solvable word problem?
Concerning Problem 1, we see that the semigroup theoretical approach to the word prob-
lem of automatic semigroups will carry over, that is, with slight modication of the proofs,
we will obtain an armative answer to Problem 1. As far as Problem 2 is concerned we will
give a partial answer. Throughout this section S will denote a semigroup, X a nite gen-
erating set for S. We assume that X S and denote by : X
+
S the homomorphism
extending the identity map : X S.
3.11. EQUALITY AND WORD PROBLEMS 79
Equality problem for S-acts. Let A be an S-act, generated by a nite set A
0
A.
If there exists an algorithm that decides whether or not for any (a
i
, a
j
) A
0
A
0
and for
any two given words u, v X
+
, a
i
.u = a
j
.v holds, then we say that the equality problem
is solvable for the S-act A.
Proposition 3.11.1. Let S be a semigroup generated by a nite set X. Then the word
problem is solvable for S if and only if the equality problem is solvable for the right S-act
S.
Proof. Recall that if S is generated by X, then the right S-act S is generated by X
(see Proposition 3.1.10). Assume that the word problem is solvable for S. Let x
i
, x
j

X, u, v X
+
and u
x
i
u, v
x
j
v. By our assumptions, there exists an algorithm
which decides whether or not x
i
u = u
= v
= x
j
v, proving that the equality problem
is solvable for the right S-act S.
For the converse, assume that the equality problem is solvable for the right S-act S.
Let u, v X
+
, and assume that u x
i
u
and that v x
j
v
. By our assumptions, there

exists an algorithm that decides whether or not u = x
i
u
= x
j
v
= v holds in S or not,
proving that the word problem is indeed solvable for S.
In giving a solution to Problem 1, we follow the semigroup theoretical approach.
Proposition 3.11.2. Let A be an automatic S-act. Let (X, L
1
, . . . , L
n
) be an automatic
0
for A. Then, there exists a constant N
such that for any a
i
A
0
and u L
i
the following hold for the elements a = a
i
.u or
a = a
i
.(u x) of A, where x X:
(i) There exists a
j
A
0
and v L
j
such that |v| |u| +N and a = a
j
.v.
(ii) If there exists a
j
A
0
and v L
j
, such that |v| > |u| +N and a = a
j
.v. Then there
exists innitely many w L
j
such that a = a
j
.w.
Proof. (i) For all (a
j
, a
k
) A
0
A
0
consider nite state automata A
(a
j
,a
k
)=
, A
(a
j
,a
k
)x
,
(x X) accepting the regular languages L
(a
j
,a
k
)=
, L
(a
j
,a
k
)x
respectively. Let N be greater
than the number of states in any of the automata dened.
Let a
i
A
0
and u L
i
. If a = a
i
.u, then (i) is straightforward. Assume that a =
a
i
.(u x), where x X. Since (X, L
1
, . . . , L
n
) is an automatic structure with respect to the
generating set A
0
, there exists a
j
A
0
and v L
j
such that a = a
j
.v. If |v| |u|+N, then
we are nished. Assume that |v| > |u| +N. Clearly (u, v)
X
is accepted by A = A
(a
i
,a
j
)x
.
Reading through all of u, we visit a state say q in A at least twice. Removing the subword
of v between successive visits to q, we get a shorter word v
1
, moreover (u, v
1
)
X
is still
accepted by A
(a
i
,a
j
)x
. Repeating this procedure as many times as necessary, we obtain a
word w, which satises that |w| |u| +N and a = a
j
.w.
(ii) Assume that there exists a
j
A
0
and v L
j
, such that |v| > |u| +N and a = a
j
.v.
In particular we have that (u, v)
X
is accepted by one of the automata dened; say by A.
After reading through all of u, we visit a state say q in A more then once. Inserting the
subword in between two successive visits to q in v in the appropriate place, we will get a
longer word v
1
so that (u, v
1
)
X
is still accepted by A. Repeating this process as many
times as we want, we get the desired result.
In the following proposition, we let S
1
denote the semigroup obtained by adjoining an
identity element to S.
Proposition 3.11.3. Let A be an automatic S-act. Then A is also an automatic S
1
-act.
Proof. Let (X, L
1
, . . . , L
n
set A
0
for A. Let Y = X {1} and extend the action of S on A to an action of S
1
on
A by dening a.1 = a for all a A. Dene K
i
= L
i
{1}. Clearly each K
i
is a regular
language over Y , and

n
j=1
a
j
.K
j
= A holds. To show that (Y, K
1
, . . . , K
n
) is indeed
0
for A, we claim that the
languages K
(a
i
,a
j
)=
, K
(a
i
,a
j
)y
, y Y are regular. If a
i
= a
j
, then
K
(a
i
,a
j
)=
= {(u, v) K
i
K
j
| a
i
.u = a
j
.v}
X
= L
(a
i
,a
j
)=
{(u, 1) | u L
i
, a
i
.u = a
j
}
X
{(1, v) | v L
j
, a
i
= a
j
.v}
X
,
and hence is regular by Lemma 3.1.9 and Proposition 2.3.1. If a
i
= a
j
, then
K
(a
i
,a
i
)=
= {(u, v) K
i
K
i
| a
i
.u = a
i
.v}
X
= L
(a
i
,a
i
)=
{(u, 1) | u L
i
, a
i
.u = a
i
}
X
{(1, v) | v L
i
, a
i
= a
i
.v}
X
{(1, 1)},
hence is regular. It is straightforward that for all (a
i
, a
j
) A
0
A
0
we have K
(a
i
,a
j
)
1
=
K
(a
i
,a
j
)=
. Let y X. Then
K
(a
i
,a
j
)y
= {(u, v) K
i
K
j
| a
i
.(u x) = a
j
.v}
X
= L
(a
i
,a
j
)y
{(u, 1) | u L
i
, a
i
.(u x) = a
j
}
X
{(1, v) | v L
j
, a
i
x = a
j
.v}
X
,
and hence is regular by Lemmas 3.1.8, 3.1.9 and Propositions 2.2.3 and 2.3.1.
Proposition 3.11.4. Let (X, L
1
, . . . , L
n
) be an automatic structure with respect to the
generating set A
0
for the S-act A. Let u X
+
, a
i
A
0
, and consider a = a
i
.u. Then we
can nd a generator a
j
A
0
and v L
j
such that a = a
j
.v in time proportional to |u|
2
.
Proof. By Proposition 3.11.3 we can consider A as an automatic S
1
-act. Let Y = X
{1} and consider the automatic structure (Y, K
1
, . . . , K
n
) with respect to the generating
set A
0
for A as dened in the proof of Proposition 3.11.3. For each (a
i
, a
j
) A
0
A
0
and x X, let A
(a
i
,a
j
)x
denote a nite state automaton recognizing the regular language
K
(a
i
,a
j
)x
. Let u x
1
. . . x
m
where x
i
X, (1 i m), and let a = a
i
.u. We will run the
following procedure:
Step 1. Consider the nite state automata A
(a
i
,a
1
)x
1
, . . . , A
(a
i
,an)x
1
in this order. We follow
a path in the automaton A
(a
i
,a
j
)x
1
, in which the rst component of the labels of the
edges are 1$$ . . .. We begin this procedure with the rst automaton, that is when
j = 1.
(a) If a nal state can be reached in A
(a
i
,a
j
)x
1
, then the second component of the
labels of the edges give a word
1
L
j
, so that a
i
.1x
1
= a
j
.
1
. Continue with
step 2.
(b) If a nal state cannot be reached, then we consider A
(a
i
,a
j+1
)x
1
and repeat (a)
with this automaton.
We note that since

n
j=1
a
j
.K
j
= A, we will nd a generator a
k
1
and
1
L
k
1
, so
that a
i
.1x
1
= a
k
1
.
1
.
Step 2. Consider the nite state automata A
(a
k
1
,a
1
)x
2
, . . . , A
(a
k
1
,an)x
2
in this order. We follow
a path in A
(a
k
1
,a
j
)x
1
, in which the rst component of the labels of the edges are
1
$$ . . .. We begin this procedure with the rst automaton, that is when j = 1.
(a
k
1
,a
j
)x
2
, then the second component of the
labels of the edges give a word
2
L
j
, so that a
k
1
.
1
x
2
= a
j
.
2
. Continue
the procedure with step 3.
(a
k
1
,a
j+1
)x
2
and repeat (a)
with this automaton.
Since

n
j=1
a
j
.K
j
= A, there exists a generator a
k
2
and a word
2
L
k
2
, so that
a
i
.1x
1
x
2
= a
k
1
.
1
x
2
= a
k
2
.
2
.
.
.
.
Step m. Consider the nite state automata A
(a
k
m1
,a
1
)xm
, . . . , A
(a
k
m1
,an)xm
in this order.
We follow a path in A
(a
k
m1
,a
j
)xm
, in which the rst component of the labels of the
edges are
m1
$$ . . .. We begin this procedure with the rst automaton, that is
when j = 1.
(a
k
m1
,a
j
)xm
, then the second component of
the labels of the edges give a word
m
L
j
, so that a
k
m1
.
m1
x
m
= a
j
.
m
.
(a
k
m1
,a
j+1
)x
2
and repeat
(a) with this automaton.
As before, we note that since

n
j=1
a
j
.K
j
= A, we will eventually nd a generator
a
k
A
0
and L
k
, so that
a
i
.1x
1
x
2
. . . x
m
= a
k
1
.
1
x
2
. . . x
m
= . . . a
k
m1
.
m1
x
m
= a
k
..
In the worst case, the time being taken to nd
j
is n |
j
|, since then we need to
input
i1
$$ . . . in n automata. By Proposition 3.11.2 there exists a constant N such that
|
i
| |
i1
| +N. So the time being taken to nd is O(
|u|
j=1
n(|1| +jN)) = O(|u|
2
).
Proposition 3.11.5. Let A be an automatic S-act, then the equality problem is solvable
for A in quadratic time.
Proof. Let (X, L
1
, . . . , L
n
set A
0
for the S-act A. Let u, v X
+
and let a
i
, a
j
A
0
. Let a = a
i
.u and b = a
j
.v.
Then, according to Proposition 3.11.4 we can nd generators a
k
, a
l
A
0
and words u
L
k
, v L
j
in quadratic time, so that a
i
.u = a
k
. u and a
j
.v = a
l
. v. With the help of
the automaton A
(a
k
,a
l
)=
accepting the regular language L
(a
k
,a
l
)=
we will obtain an answer
whether a
i
.u = a
j
.v in A.
An immediate consequence of Proposition 3.11.5 is:
Corollary 3.11.6. Let S = X be an R-class automatic monoid, and u, v X
. Then
we can decide in quadratic time whether u R v in S.
Also we have:
Corollary 3.11.7. Let S = X be a regular semigroup and R be a Sch utzenberger auto-
matic R-class of S. Let u, v X
+
. Then we can decide whether u, v R. Moreover if
u, v R, then we can decide whether u = v.
Proof. Using the notation of Proposition 3.2.1, let (X, N) be a Sch utzenberger auto-
matic structure for R, let L = N {0} and e E(R). According to Proposition 3.2.1,
(X
0
, L) is an automatic structure with respect to the generating set {e} for the right
S
0
-act R
0
. Let u X
+
. In the right S
0
-act R
0
we have that e.u = u if u R, and
e.u = 0 if u / R. By Proposition 3.11.5, we can decide in quadratic time whether or not
e.u = e 0 = 0.
Now let u, v X
+
. Decide rst whether or not e.u = e.0 and e.v = e.0. If e.u = e.0
and e v = e.0, then u = e u and v = e v are elements of R. By Proposition 3.11.5 again
we can decide whether or not u = e u = e v = v.
Corollary 3.11.8. Let S be a strongly Sch utzenberger automatic regular semigroup with
nitely many R-classes. Then the word problem is solvable for S in quadratic time.
Proof. Let S/R = {R
1
, . . . , R
n
} and let (X, N
i
) be a Sch utzenberger automatic struc-
ture for R
i
(1 i n). Let u, v X
+
. Since S is also R-class automatic we can decide
by Corollary 3.11.6 in quadratic time whether or not the elements represented by u and
v are R-related. Since S has nitely many R-classes, we can nd their R-class in nitely
many steps and then decide by Corollary 3.11.7 in quadratic time whether or not they
represent the same element in S.
To be more precise we decide with the following procedure whether or not u = v holds
in S. As in Corollary 3.11.7, let (X
0
, L
i
), where L
i
= N
i
{0} be an automatic structure
with respect to a one element generating set {e
i
}, where e
i
E(R
i
) for the right S
0
-act
R
0
i
. Let u, v X
+
. Since we have nitely many R-classes and hence S
0
-acts we can nd
in nitely many steps the idempotents e
i
, e
j
so that e
i
.u = e
i
.0, e
j
.v = e
j
.0. In other
words, we can nd in nitely many steps the Sch utzenberger automatic structure assigned
to the R-class of the element represented by u and the R-class of the element represented
by v. If the R-classes of the elements represented by u and v happens to be the same,
then as seen in Corollary 3.11.7 we can also decide whether or not u and v represent the
same element in quadratic time.
Whether the word problem is solvable for strongly Sch utzenberger automatic regular
semigroups with innitely many R-classes seems to be a more dicult question. We can
obviously decide whether or not the elements represented by the two given words belong
to the same R-class. If they happen to be in the same R-class, then a technique needs to
be developed with the help of which one can nd in nitely many steps the Sch utzenberger
automatic structure of the R-class of these elements, so that it can be decided whether
or not the two given words represent the same element. We end this section with the
following
Open question. Is it true that strongly Sch utzenberger automatic regular semigroups
have solvable word problem?
3.12 Inverse free product of inverse semigroups.
It is veried in [3] that the group free product of automatic groups is automatic, and in
[5] it is shown that the free product of automatic semigroups is automatic. In this section
we seek answer to the following problem:
Problem: Is the inverse free product of Sch utzenberger automatic inverse semigroups
Sch utzenberger automatic?
The results in [25] concerning the description of Sch utzenberger graphs of the inverse
free product of two inverse semigroups will serve as a basis of the answer of the above
problem. First we summarize some necessary notions and results needed for this section.
Throughout the section S
1
and S
2
denote two disjoint inverse semigroups. The inverse
free product S = S
1
Inv
S
2
is an inverse semigroup satisfying the following properties:
(i) There exists embeddings
1
: S
1
S,
2
: S
2
S;
(ii) If there exists an inverse semigroups T and homomorphisms
1
: S
1
T,
2
: S
2

T, then there exists a unique homomorphism : S T such that the following
diagram is commutative:
S
S
1
S
2
2
.
T
Assume that S
1
= InvX|P and S
2
= InvY |Q where X Y = . Then S
1

Inv
S
2
=
InvX Y |P Q.
The rst general result concerning the structure of the inverse free product of two
inverse semigroups was obtained by Jones in [24]. This structure theorem provides a set of
3.12. INVERSE FREE PRODUCT OF INVERSE SEMIGROUPS. 85
canonical forms for the elements of the inverse free product, with the help of which Greens
relations are described. A graph theoretical approach to describe the structure of the
inverse free product of two inverse semigroups is introduced in [25]. This approach employs
a technique introduced by Stephen [35] for obtaining Sch utzenberger graphs relative to
an inverse semigroup presentation. As our answer to our proposed problem relies on the
latter approach, we summarize the necessary notions and results of [25] regarding the
Sch utzenberger graphs of S
1
Inv
S
2
.
Let X be a nite set and let X
1
= {x
1
| x X} denote the set of formal inverses
of elements of X. Let be a directed labelled graph, where the labels of the edges are
elements of X X
1
. We say that is a word graph over X X
1
if it satises the
following two conditions:
(i) The graph is strongly connected, that is to say that for any two distinct vertices
v
1
, v
2
of , there exists a directed path from v
1
to v
2
.
(ii) If v
1
x
v
2
is an edge of , then v
2
x
1
v
1
is an edge of as well.
Let X and Y be disjoint nite sets, and let be a word graph over Z = XX
1
Y
Y
1
. The edges of can be coloured with two colours in such a way that the edges of
labelled with elements of XX
1
get one of the colours, and the edges of labelled with
elements of Y Y
1
get the other one. A subgraph of is said to be monochromatic, if all
of its edges have the same colour. We say that a path in is monochromatic, if all of its
edges have the same colour. A lobe is a maximal monochromatic subgraph of . A vertex
of is called an intersection vertex if it belongs to more then one lobe. A switchpoint
of a path p in is a vertex that is common to successive edges of dierent colour. The
switchpoint sequence of a path p in is the sequence of switchpoints p traverses, in order.
A path p in is called simple if there are no repeated vertices in p other then perhaps
its rst and last vertex, in which case p is called a simple cycle. A word graph over Z
is called cactoid, if the following two conditions hold:
(i) has nitely many lobes;
(ii) every simple cycle of is monochromatic.
The following important properties of cactoid graphs over Z is proved in Lemma 2.1. and
Corollary 2.2 of [25].
Proposition 3.12.1. Let be a cactoid graph over Z, and let v
1
, v
2
be two distinct vertices
of . Then the following hold:
(i) Any two simple paths from v
1
to v
2
traverse the same switchpoint sequence.
(ii) Distinct lobes of possess at most one common vertex.
Let be a cactoid graph. Fix a vertex v in , and let v
1
be a vertex distinct from v.
The switchpoint sequence of v
1
is the switchpoint sequence of a (in fact any) simple path
from v to v
1
. Denoting by l the length of the switchpoint sequence of an arbitrary vertex
v
1
of , a norm with respect to v is then dened for v
1
v
1
=
_
0 if v
1
= v,
1 +l otherwise
The following results are proved in [25, Lemma 2.3.] and [25, Corollary]:
Proposition 3.12.2. Let be a cactoid graph over Z and let v be a vertex of . Then
the following hold:
(i) Each lobe of possess a unique vertex called the root
of least norm with respect

to v.
(ii) For each lobe of ,
is either an intersection vertex of or is the vertex v.

(iii) For each vertex u =
of the lobe , u =
+ 1.
(iv) The chosen vertex v is the root of two distinct lobes if and only if it is an intersection
vertex in and it is the root of a unique lobe otherwise. Every other intersection
vertex of is the root of a unique lobe of .
One of the main results of [25] is Theorem 4.1.:
Theorem 3.12.3. Let S
1
be an inverse semigroup generated by a set X and S
2
be an
inverse semigroup generated by Y , where X Y = . Then the Sch utzenberger graphs of
S
1
Inv
S
2
are precisely the cactoid graphs over X X
1
Y Y
1
each of whose lobes is
isomorphic to a Sch utzenberger graph of S
1
or of S
2
.
We are ready to give an answer to our proposed problem.
Proposition 3.12.4. Let S
1
and S
2
be Sch utzenberger automatic inverse semigroups.
Then the inverse free product S = S
1

Inv
S
2
is also a Sch utzenberger automatic inverse
semigroup.
Step 1: notation. In the rst step, we set up the notation. Assume that S
1
is
generated by the nite set X, where X is closed under taking inverses. For the sake of
simplicity we assume that X S
1
. Let
1
: X
+
S
1
denote the semigroup homomor-
phism extending the identity map
1
: X S
1
. Assume that S
2
is generated by the nite
set Y S
2
, where Y is closed under taking inverses and X Y = . Let
2
: Y
+
S
2
denote the homomorphism extending the identity map
2
: Y S
2
. Let Z = X Y and
let : Z
+
S denote the homomorphism extending the identity map : Z S. Let
s S and consider the Sch utzenberger graph = S
s
(Z) of R
s
. By Theorem 3.12.3,
is a cactoid graph over Z, and so it has nitely many lobes each of which is isomorphic
to a Sch utzenberger graph of S
1
or of S
2
. Let
X
denote the set of lobes, whose edges
are labelled by elements of X and let
Y
denote the set of lobes, whose edges are labelled
by the elements of Y . Let
1
, . . . ,
n1
denote the intersection vertices of and let
0
be
the vertex corresponding to ss
1
. For the sake of convenience, we assume that
0
is not
an intersection vertex in . (Note that since we have n 1 intersection vertices, has n
lobes.) For each 1 i n 1 we x a simple path p
i
in from
0
to
i
. Denote by w
i
the word over Z that labels the path p
i
. Let W = {w
1
, . . . , w
n1
}. Clearly W is a regular
language over Z.
According to Proposition 3.12.2, each intersection vertex
i
is the unique root of one
of the lobes of . We let
i
denote the lobe whose root is
i
.
1

2
0

1

2
For all
i

X
we choose an R-class R
i
of S
1
, such that
i
is isomorphic to the
i
. The set of R-classes chosen in S
1
will be denoted by R
1
.
Similarly, for all
i

Y
we choose an R-class R
i
of S
2
, such that
i
is isomorphic to the
i
. The set of R-classes chosen in S
2
will be denoted by R
2
. To
simplify notation, we will actually think of
i
as the Sch utzenberger graph of the chosen
R-class R
i
. In particular, we will think of the vertices of
i
as elements of R
i
. Dene
W
X
= {w
i
W |
i

X
},
and let W
Y
= W \ W
X
. In other words, w
i
W
X
if and only if
0
w
i
=
i
is a root of a
lobe in
X
, and w
j
W
Y
if and only if
0
w
j
=
j
is a root of a lobe in
Y
.
By assumptions, S
1
and S
2
are Sch utzenberger automatic inverse semigroups. In
particular, for each R-class R
i
R
1
, we may assume by Propositions 3.6.12, 3.2.1 and
Proposition 3.4.4 that there exists a regular language K
i
over X such that the following
hold:
(1a) For each t V (
i
) there exists exactly one u K
i
with
i
u
1
= t and so
(K
i
)
=
= {(u, w) K
i
K
i
|
i
u =
i
w}
X
= {(u, u) | u K
i
}
X
is a regular
language.
(1b) (K
i
)
x
= {(u, w) K
i
K
i
|
i
(u x) =
i
w}
X
is a regular language for each
x X.
We let K
X
denote the set of regular languages obtained. Similarly, for each R-class
R
j
R
2
we may assume that there exists a regular language K
j
over Y such that the
following hold:
(2a) For each t R
j
there exists exactly one u K
j
with
j
u
2
= t and so (K
j
)
=
=
{(u, w) K
j
K
j
|
j
u =
j
w}
Y
= {(u, u) | u K
j
}
Y
(2b) (K
j
)
y
= {(u, w) K
j
K
j
|
j
(u y) =
j
w}
Y
is a regular language for each
y Y .
We let K
Y
denote the set of regular languages obtained. We let K = K
X
K
Y
=
{K
0
, K
1
, . . . , K
n1
}.
Step 2: the regular language. In the second step, we are going to construct a
regular language K with the help of the regular languages in K, so that
0
and K satisfy
that
0
K = R
s
.
Let
i
be an arbitrary lobe of with vertex set V (
i
) and root
i
. Let V
i
denote the
intersection vertices of belonging to
i
. Clearly V
i
is a nite set. Let

i
denote the
subgraph of
i
, whose vertex set is V (
i
) = (V (
i
) \ V
i
) {
i
}. Illustrating this:
0
1

2
Clearly
=

0
. . .

n1
and

j
= , for all 0 i, j n 1, i = j. (3.13)
For each v V
i
, consider the language
K
i,v
= {u K
i
|
i
u = v}. (3.14)
By Lemma 3.1.9, we have that K
i,v
is a regular language over X or over Y , depending on
whether
i
is an element of
X
or of
Y
. In fact, since K
i
satises (1a) or (2a), for each
v V
i
, K
i,v
has exactly one element. Let
K
i,V
i
=
_
vV
i
K
i,v
(3.15)
The language K
i,V
i
is regular, since it is a nite union of regular languages. Dene
K
i
= K
i
K
i,V
i
and L
i
=

K
i
{},
where denotes the empty word. Clearly

K
i
and L
i
are regular languages. Taking into
consideration that for all 1 i n 1, w
i
is path between
0
and
i
, the importance of
the languages dened can be illustrated in the following way in :
w
i
u
1
u
2
u
3
(u
1
, u
2
, u
3

K
i
)
That is to say that
0
(w
i
L
i
) = V (
i
). (3.16)
Moreover since K
i
is a language satisfying (1a) or (2a), we also have that for all v V (
i
)
there exists exactly one w L
i
so that
0
(w
i
w) = v holds. To have a complete picture,
we note that
0
(w
i
K
i,V
i
) = V
i
. (3.17)
Let
K = L
0
w
1
L
1
. . . w
n1
L
n1
,
where w
i
denotes the labels of the path we xed from
0
to
i
and L
i
is the regular
language over X or over Y so that
0
(w
i
L
i
) = V (
i
). It follows by (3.16) and (3.13)
that
0
K = R
s
.
Third step: the Sch utzenberger automatic structure. We verify that
0
and
K satisfy the last two conditions of (S2) of Proposition 3.2.1.
Clearly, by the denition of K and by the note after (3.16) we have that for all t R
s
,
there exists exactly one w K so that
0
w = t proving that
K
=
= {(u, w) |
0
u =
0
w}
Z
= {(u, u) | u K}
Z
.
Hence K
=
is a regular language by Proposition 2.3.2.
Next we claim that for all x X, the language
K
x
= {(u, w) K K |
0
(u x) =
0
w}
Z
is regular. It is straightforward that if (u, w) K
x
, then
0
(u x) and
0
w belong to the
same lobe say to
i
, where
i

X
. Moreover, since the last arrow is an element of
X and since by denition a lobe is a maximal monochromatic subgraph,
0
u and
0
w
have to be vertices of
i
. Depending on what kind of vertices (intersection vertex or not)
the elements
0
u and
0
w represent in the Sch utzenberger graph R
s
, the following four
cases can occur:
(1) Our rst case is when (u, w)
Z
is a word for which
0
u and
0
w are not intersection
vertices of . First we introduce the following notation. Let
(L
i
)
x
= (K
i
)
x
\ {(u, w) | u, w K
i,V
i
,
i
u x =
i
w}
Z
.
Since (K
i
)
x
is a regular language and since the latter set is nite, (L
i
)
x
is a regular
language.
We verify that words (u, w)
Z
K
x
that satisfy that
0
u,
0
vertices in a lobe
i

X
are exactly the words contained in the regular language
N
1
=
_
w
i
W
X
(w
i
, w
i
)
Z
(L
i
)
x
.
If (u, w) N
1
, then u w
i
u and w = w
i
w, where w
i
W
X
and ( u, w)
Z
(L
i
)
x
.
It follows that u, w / K
i,V
i
, in other words
i
u and
i
w are not intersection vertices
of . Moreover we have that
i
( u x) =
i
w holds. It follows, that
0
(u x) =
0
(w
i
u x) =
0
w
i
w =
0
w, proving that (u, w)
Z
K
x
.
For the converse, assume that (u, w)
Z
K
x
and that
0
u,
0
vertices in a lobe
i

X
. Then, as noted before
0
u and
0
w are vertices of
the same lobe
i

X
. It follows that u w
i
u, and w w
i
w, where u, w

K
i
. In
particular, we have that
i
( u x) = (
0
w
i
) u x =
0
w
i
w =
i
w.
In other words, ( u, w) (L
i
)
x
.
Since (L
i
)
x
is regular we have that (w
i
, w
i
)
Z
(L
i
)
x
is regular and N
1
is a nite union
of regular languages, proving that N
1
is regular.
(2) Our second case is when (u, w)
Z
K
x
is a word such that
0
u is not an intersection
vertex of and
0
w is an intersection vertices of . Again, we rst introduce the
notation we are going to use. For each
i

X
and for all intersection vertices
k
V
i
, we let w
i,k
K
i,V
i
K
i
so that
i
w
i,k
=
k
. Such an element exists, since
i
and
i
are isomorphic and since
i
K
i
1
= R
i
. We let
K
(i,x,k)
= {u K
i
|
i
u x =
k
} \ K
i,V
i
.
By Lemma 3.1.8 and Proposition 2.2.3, K
(i,x,k)
is a regular language. Let
M
1
= {(w
i
, w
k
) W
X
W
Y
|
0
w
k
V
i
}.
We will verify that words (u, w)
Z
K
x
that satisfy that
0
vertex of
i

X
and
0
w is an intersection vertex of
i
are exactly the words
contained in the language
N
2
=
_
(w
i
,w
k
)M
1
(w
i
, w
k
)
Z
(K
(i,x,k)
{$}
})
Z
,
where denotes the padded product.
If (u, w) N
2
, then w w
k
and u w
i
u, where u K
(i,x,k)
,
0
w
i
,
0
w
k
V
i
. In
particular we have that
i
u is not an intersection vertex and that
i
u x =
k
=
0
w
k
. Hence (
0
w
i
) u x =
0
w
k
holds, which veries that (u, w)
Z
K
x
.
For the converse assume that (u, w)
Z
K
x
such that
0
vertex of
i

X
and
0
w is an intersection vertex. Since
0
w is an intersection
vertex, we have that w w
k
, where
k
V
i
. Moreover, since
0
u is not an
intersection vertex in
i
, we have that u w
i
u, where u

K
i
. Also, since
(u, w)
Z
K
x
, we have that
k
=
0
w
k
=
0
w
=
0
u x
= (
0
w
i
) u x
=
i
u x,
verifying that u K
(i,x,k)
. Hence we may deduce that
(u, w)
Z
(w
i
, w
k
)
Z
(K
(i,x,k)
{$}
)
Z
N
2
.
Finally we verify that N
2
is a regular language over Z. By Proposition 2.2.3, the
language (K
(i,x,k)
{$}
)
Z
is regular. Since M = (w
i
, w
k
)
Z
is nite, it satises
the condition of Proposition 2.3.16 and hence we have that for all (w
i
, w
k
) M
1
the
language (w
i
, w
k
)
Z
(K
(i,x,k)
{$}
)
Z
is regular. We may now deduce that N
2
is
regular, since it is a nite union of regular languages.
(3) This case can be considered as the dual of the previous case. We consider words
(u, w)
Z
K
x
, where
0
u is an intersection vertex of and
0
w is not an
intersection vertices of . We introduce the following notations.
For each
i

X
and for all intersection vertices
k
V
i
, let w
i,k
K
i,V
i
K
i
so
that
i
w
i,k
=
k
. Such an element exists, since
i
and
i
are isomorphic and since
i
K
i
1
= R
i
. Let
K
(x,i,k)
= {u K
i
|
i
u =
k
x} \ K
i,V
i
.
By Lemma 3.1.9 and Proposition 2.2.3, K
(x,i,k)
is a regular language. Let
M
2
= {(w
k
, w
i
) W
Y
W
X
|
0
w
k
V
i
}.
We will verify that words (u, w)
Z
K
x
that satisfy that
0
u is an intersection
vertex of some
i

X
and
0
w is not an intersection vertex of
i
are exactly
the words contained in the language
N
3
=
_
(w
k
,w
i
)M
2
(w
k
, w
i
)
Z
({$}
K
(x,i,k)
)
Z
,
where denotes the padded product.
If (u, w) N
3
, then u w
k
and w w
i
w, where w K
(x,i,k)
,
0
w
i
,
0
w
k
V
i
.
In particular we have that
i
w is not an intersection vertex and that
0
w
k
x =
k
x =
i
w. Hence (
0
w
k
) x =
0
w
i
w holds, which veries that (u, w)
Z
K
x
.
For the converse assume that (u, w)
Z
K
x
such that
0
u is an intersection vertex
of
i

X
and
0
w is not an intersection vertex. Since
0
u is an intersection
vertex, we have that u w
k
, where
k
V
i
. Moreover, since
0
w is not an
intersection vertex in
i
, we have that w w
i
w, where w

K
i
. Also, since
(u, w)
Z
K
x
, we have that
k
x =
0
u
k
x =
0
w
= (
0
w
i
) w
=
i
w
verifying that w K
(x,i,k)
. Hence we may deduce that
(u, w)
Z
(w
k
, w
i
)
Z
({$}
K
(x,i,k)
)
Z
N
3
.
Finally we verify that N
3
is a regular language over Z. By Proposition 2.2.3, the
languages (K
(x,i,k)
{$}
)
Z
, where
0
w
k
V
i
are regular. Since M = (w
k
, w
i
)
Z
is nite it satises the condition of Proposition 2.3.16 and hence we have that for
all (w
k
, w
i
) M
2
the language (w
k
, w
i
)
Z
(K
(x,i,k)
{$}
)
Z
is regular. We may
now deduce that N
3
is regular, since it is a nite union of regular languages.
(4) Our nal case is when (u, w)
Z
K
x
is such that the vertices
0
u,
0
w are
intersection vertices of some
i

X
. In this case (u, w)
Z
K
x
if and only if the
word (u, w)
Z
is an element of the following language
N
4
= {(w
j
, w
k
) | w
j
, w
k
W,
0
w
j
x =
0
w
k
}
Z
.
Since W is nite, N
4
is nite, and hence is a regular language.
Now we may deduce that
K
x
= N
1
N
2
N
3
N
4
,
is a nite union of regular languages and hence is regular. That K
y
is a regular language
for all y Y can be proved in a similar way. Since R
s
was an arbitrary R-class of
S, we may deduce that every R-class of S is Sch utzenberger automatic, and hence S is
Chapter 4
Semidirect product
Semilattices and groups are some of the handiest examples of inverse semigroups. In
fact, every inverse semigroup can be described using semilattices and groups as building
blocks with a semidirect product construction. Namely, every inverse semigroup is an
idempotent separating homomorphic image of a subsemigroup of a semidirect product of
a semilattice by a group. In addition, a dual approach tells us that every inverse semigroup
is a subsemigroup in an idempotent separating homomorphic image of a semidirect product
of a semilattice by a group. In this chapter we give necessary and sucient conditions
for the construction of a semidirect product of a semilattice by a group to be nitely
generated, to be nitely presented and to be Sch utzenberger automatic.
4.1 Presentations.
Semigroup presentations.
We rst summarize the basic denitions and properties concerning semigroup presen-
tations.
Let X be a non-empty set. A semigroup presentation is an ordered pair X|P, where
P is a binary relation on the free semigroup X
+
. Let denote the congruence generated
by P on the free semigroup X
+
. The semigroup S = X
+
/ is said to be presented by the
generators X and relations P and we denote this by
S = X|P.
If X can be chosen to be a nite set, then we say that S is nitely generated, and if both X
and P can be chosen to be nite sets, then we say that S is nitely presented. Let w
1
, w
2
be words over X. We write w
1
w
2
, if w
1
and w
2
are identical as words and we write
95
96 CHAPTER 4. SEMIDIRECT PRODUCT
w
1
= w
2
, if w
1
and w
2
represent the same element of S. If w
1
= w
2
, then we also say that
S satises the relation w
1
= w
2
. We say that w
2
is obtained from w
1
by an application
of a relation of P, if w
1
u and w
2
v, where , X
and u = v or v = u is
a relation of P. We say that w
1
= w
2
is a consequence of relations in P, if there exists a
sequence of words
w
1

0
, . . . ,
m
w
2
,
where
j+1
is obtained from
j
by applying a relation of P.
We recall [33, Proposition 2.3.]:
Proposition 4.1.1. Let S be a semigroup generated by a set X and P be a binary relation
on X
+
. Then S = X|P if and only if the following two conditions hold:
(i) S satises all relations in P;
(ii) If w
1
, w
2
X
+
are such that w
1
= w
2
holds in S, then w
1
= w
2
is a consequence of
relations in P.
Let S be dened by the nite presentation X|P. One can obtain an alternative
semigroup presentation for S by applying the following elementary Tietze transformations:
(T1) adding a new generating symbol y and a new relation y = w, where
w X
+
;
(T2) if P possesses a relation of the form y = w, where w is a word over X
that does not contain the symbol y, then deleting the generating symbol
y and the relation y = w and replacing all occurrences of y by w in the
remaining relations;
(T3) adding a new relation w
1
= w
2
to P, in the case where w
1
= w
2
is a
consequence of relations in P;
(T4) deleting a relation w
1
= w
2
from P, in the case where w
1
= w
2
is a
consequence of relations in P {w
1
= w
2
}.
Proposition 2.5. of [33] tells us:
Proposition 4.1.2. Two (nite) semigroup presentations dene the same semigroup if
and only if one can be obtained from the other by applying (a nite sequence of ) elementary
Tietze transformations.
Inverse semigroup presentations.
In the rest of the thesis, we work frequently with inverse semigroup presentations.
Therefore we recall the basic denitions and properties concerning presentations of inverse
semigroups.
4.1. PRESENTATIONS. 97
We consider inverse semigroups as algebras of type (2,1), where the binary operation is
multiplication and the unary operation assigns to each element its unique (von Neumann)
inverse.
The class of inverse semigroups forms a variety and hence free inverse semigroups
exist. This fact enables us to dene inverse semigroup presentations. First, we recall the
description of the free inverse semigroup on a non-empty set X as a factor semigroup of
the free semigroup with involution on X.
Let X be a non-empty set and X
1
= {x
1
| x X}. Consider the free semigroup
F = (X X
1
)
+
and dene a unary operation on F in the following way: For each
y X X
1
, let
y
1
=
_
x
1
if y = x X
x if y = x
1
X
1
and dene
(y
1
. . . y
n
)
1
= y
1
n
. . . y
1
1
.
With the unary operation dened, (F, ,
1
) is the free semigroup with involution on X,
which we shall denote by FSI(X). Dene the following binary relation on FSI(X):
= {(uu
1
u, u) | u FSI(X)} {(uu
1
vv
1
, vv
1
uu
1
) | u, v FSI(X)}.
The congruence generated by is called the Wagner congruence which we denote by .
The factor-semigroup
FI(X) = FSI(X)/
is the free inverse semigroup on X. We will refer to the elements of as standard inverse
semigroup relations.
An inverse semigroup presentation is an ordered pair X|P, where P is a binary
relation on FSI(X). Let denote the congruence generated by P . The semigroup
S = FSI(X)/ is said to be presented as an inverse semigroup by the generators X and
relations P and we denote this by
S = InvX|P.
If X can be chosen to be a nite set, then we say that S is nitely generated as an inverse
semigroup, and if both X and P can be chosen to be nite, then we say that S is nitely
presented as an inverse semigroup. We note that if S is generated as an inverse semigroup
by X, then S is generated by XX
1
as a semigroup. In particular S is nitely generated
as a semigroup if and only if it is nitely generated as an inverse semigroup. Moreover, if S
is given by the inverse semigroup presentation InvX|P, then it is given by the following
semigroup presentation:
X X
1
|P .
Proposition 4.1.3. Let S be an inverse semigroup dened by the semigroup presentation
X|P. Then S is dened by the inverse semigroup presentation InvX|P.
Let S = InvX|P and let w
1
, w
2
be words over X X
1
. We write w
1
w
2
, if w
1
and w
2
are identical as words and we write w
1
= w
2
, if w
1
and w
2
represent the same
element of S. If w
1
= w
2
, then we also say that S satises the relation w
1
= w
2
. We say
that w
2
is obtained from w
1
by an application of a relation of P or of a standard inverse
semigroup relation, if w
1
u and w
2
v, where , (X X
1
)
and u = v or
v = u is a relation of P or a standard inverse semigroup relation. We say that w
1
= w
2
is
a consequence of relations in P and of standard inverse semigroup relations, if there exists
a sequence of words
w
1

0
, . . . ,
m
w
2
,
where
j+1
is obtained from
j
by applying a relation of P or a standard inverse semigroup
relation.
The following proposition is a modication of [33, Proposition 2.3.]:
Proposition 4.1.4. Let S be an inverse semigroup generated by a set X. Let P be a
relation on FSI(X) and let denote the set of standard inverse semigroup relations on
FSI(X). Then S = InvX|P if and only if the following two conditions hold:
(i) S satises all relations in P ;
(ii) If w
1
, w
2
(X X
1
)
+
are such that w
1
= w
2
holds in S, then w
1
= w
2
is a
consequence of relations in P .
Proof. (=) If S = InvX|P, then S = FSI(X)/ by denition, where is the
congruence generated by P and hence (i) and (ii) hold.
(=) Let F denote the free semigroup with involution on X and let : F S be
the unary homomorphism extending the identity map : X X. Let denote the
congruence generated by P . We show that S = InvX|P by verifying that ker = .
Since S satises all relations in P by (i), we clearly have that ker. Suppose now
that (w
1
, w
2
) ker. Then w
1
and w
2
represent the same element of S, and hence by (ii)
ker .
Let S = InvX|P. One can obtain an alternative inverse semigroup presentation for
S by applying the following elementary Tietze transformations:
4.2. FINITE GENERATION WITH RESPECT TO AN ACTION. 99
(T1) adding a new generating symbol y and a new relation y = w, where
w (X X
1
)
+
;
(T2) if P possesses a relation of the form y = w, where w is a word over
X X
1
that does not contain the symbols y or y
1
, then deleting the
generating symbol y and the relation y = w and replacing all occurrences
of y by w and y
1
by w
1
in the remaining relations;
(T3) adding a new relation w
1
= w
2
to P, in the case where w
1
= w
2
is a con-
sequence of relations in P and of standard inverse semigroup relations;
(T4) deleting a relation w
1
= w
2
from P, in the case where w
1
= w
2
is
a consequence of relations in P {w
1
= w
2
} and of standard inverse
semigroup relations.
Similarly to Proposition 4.1.2, we have:
Proposition 4.1.5. Two (nite) inverse semigroup presentations dene the same inverse
semigroup if and only if one can be obtained from the other by applying (a nite sequence
of ) elementary Tietze transformations.
4.2 Finite generation with respect to an action.
In the rst section of the third chapter, we introduced the notion of an S-act and among
others dened nite generation for S-acts. We also showed that every semigroup S can be
considered as an S-act, but in doing so we forget the structure which S possesses and only
keep the underlying set. Bearing in mind that an S-act A can have an algebraic structure
underlying it, we give a more sophisticated version of the denition of nite generation
and introduce the concept of nite presentability with respect to an action in Section 4.
Let (Y, ) be a semilattice and denote by the natural partial order on it. Recall that
x y if and only if x = x y. We say that y Y is a maximal element of Y if x Y,
y x always implies that x = y. We say that Y satises the maximum condition if it has
nitely many maximal elements, and for all x Y there exists a maximal element y Y
such that x y. The following lemma is immediate from the denitions.
Lemma 4.2.1. Let (Y, ) be a semilattice and let be an automorphism of (Y, ). Then
x y if and only if x y. In particular we have that y is a maximal element of Y if
and only if y is a maximal element of Y as well.
Next we recall the notion of a left action of a semigroup S on a semigroup T. Let
(T, ) and (S, ) be semigroups. We say that S acts on T by endomorphisms on the left, if
there exists a map f : T S T, (t, s)
s
t satisfying the conditions
(A1)
(s
1
s
2
)
t =
s
1
(
s
2
t);
(A2)
s
(t
1
t
2
) =
s
t
1
s
t
2
for all t, t
1
, t
2
T and s, s
1
, s
2
S. If S is a monoid, then we also require the following
condition to hold:
(A3)
1
t = t for all t T.
We say that S acts on T by automorphisms on the left, if besides conditions (A1), (A2)
(and (A3), if S is a monoid), the following conditions are satised:
(A4) for all t T and s S there exists

t T such that
s
t = t;
(A5) for all t
1
, t
2
T and s S,
s
t
1
=
s
t
2
implies that t
1
= t
2
.
The reason we deviated from our notation introduced in the third chapter, and write
s
t instead of t.s, is that we would like to keep the traditional notation used in semidirect
products, which will be given soon. We also mention that a right action of S on T by
endomorphisms (automorphisms) can be dened similarly.
Denition 4.2.2. Let (S, ) and (T, ) be semigroups. Assume that S acts on T on the
left by endomorphisms (automorphisms). We say that T is generated by T
0
T with
respect to the action of S, if T =
S
T
0
T
0
, where
S
T
0
= {
s
t | s S, t T
0
}. We say
that T is nitely generated with respect to the action of S if T
0
can be chosen to be a nite
subset of T.
Lemma 4.2.3. Let T be a nitely generated semigroup and assume that the semigroup S
acts on T on the left by endomorphisms. Then T is nitely generated with respect to the
action of S as well.
Proof. Clearly, if T is generated by a nite set Y , then T is nitely generated by Y
with respect to the action of S as well.
Finitely generated semilattices are nite. We give three examples for innite semilat-
tices that are nitely generated with respect to a group action dened on them. These
examples will be referred to throughout the fourth chapter to illustrate the results con-
cerning nite generation and nite presentability of a semidirect product of a semilattice
by a group. Our rst two examples have already been introduced in the third chapter,
these are Example 3.3.6 and Example 3.3.7.
Proposition 4.2.4. Let Y
be the double innite chain with an identity adjoined on top:

. . . < e
1
< e
0
< e
1
< . . . < 1.
Let G = g be the innite cyclic group. Let n, k Z and consider the following map:
f : Y
G Y
; (e
n
, g
k
)
g
k
e
n
= e
n+k
; (1, g
k
)
g
k
1 = 1.
With this action, G acts on Y
on the left by automorphisms, and Y
is generated by
Y
0
= {e
0
, 1} with respect to this action of G.
Proof: We rst show that conditions (A1)-(A5) hold.
(A1) Let y Y
and g
1
= g
n
, g
2
= g
m
, n, m Z. If y = 1, then clearly
g
1
g
2
1 = 1 =
g
1
1 =
g
1
(
g
2
1). If y = e
k
, for some k Z, then
g
1
g
2
y =
g
n+m
e
k
= e
k+n+m
=
g
n
e
k+m
=
g
n
(
g
m
e
k
) =
g
1
(
g
2
y),
proving that (A1) does indeed hold.
(A2) Let y
1
= e
n
, y
2
= e
m
and let h = g
k
, where n, m, k Z. Without loss of generality
we can assume that n m, and hence n +k m+k. Then
h
(y
1
y
2
) =
g
k
(e
n
e
m
) =
g
k
e
n
= e
n+k
= e
n+k
e
m+k
=
g
k
e
n
g
k
e
m
=
h
y
1
h
y
2
.
The case when y
1
and/or y
2
equals 1 can be veried similarly.
(A3) Clearly
1
y =
g
0
y = y for all y Y
.
(A4) Let e
n
Y
and g
m
G. By denition,
g
m
e
nm
= e
n
and also
g
m
1 = 1, verifying
that (A4) holds.
(A5) Let y
1
= e
n
, y
2
= e
m
and let h = g
k
G such that
h
y
1
=
h
y
2
. That is
e
n+k
=
g
k
e
n
=
g
k
e
m
= e
m+k
holds and hence we obtain that e
n
= e
m
, proving that (A5) is satised.
We may now deduce that G acts on Y
on the left by automorphisms. By (A4), for all

n Z, e
n
=
g
n
e
0
, and so the semilattice Y
is nitely generated with respect to the action

of G by 1 and e
0
.
Proposition 4.2.5. Let A be the innite antichain
. . . , e
1
, e
0
, e
1
, . . . .
i
A, 0 < e
i
< 1
holds. Let A
denote the (semi)lattice obtained. Let G = g be the innite cyclic group.

Let n, k Z and consider the following map:
f : A
G A
; (e
n
, g
k
)
g
k
e
n
= e
n+k
; (1, g
k
)
g
k
1 = 1; (0, g
k
)
g
k
0 = 0.
With this action, G acts on A
on the left by automorphisms, and A
is generated by
Y
0
= {e
0
, 1} with respect to this action of G.
Proof. We show that conditions (A1)-(A5) hold.
(A1) Clearly if y A
equals to 1 or to 0, then for all g

1
, g
2
G,
g
1
g
2
y =
g
1
(
g
2
y). Let
y = e
k
, g
1
= g
n
, g
2
= g
m
. Then
g
1
g
2
y =
g
n
g
m
e
k
=
g
n+m
e
k
= e
n+m+k
=
g
n
e
k+m
=
g
n
(
g
m
e
k
) =
g
1
(
g
2
y),
proving that (A1) indeed holds.
(A2) Let y
1
, y
2
A
and h = g
k
. If either y
1
= 0 or y
2
= 0 (or both), then clearly
h
(y
1
y
2
) =
h
y
1

h
y
2
. Similarly, if y
1
= 1 and y
2
A
arbitrary, then
h
(y
1

y
2
) =
h
y
2
= 1
h
y
2
=
h
y
1

h
y
2
. The only case remaining to consider is when
y
1
= e
n
, y
2
= e
m
for some n, m Z. If n = m, then we have
h
(y
1
y
2
) =
g
k
(e
n
e
m
) =
g
k
0 =
g
k
e
n
g
k
e
m
=
h
y
1
h
y
2
.
If n = m, then
h
(y
1
y
2
) =
g
k
(e
n
e
n
) =
g
k
e
n
=
g
k
e
n
g
k
e
n
=
h
y
1
h
y
2
.
verifying that (A2) holds.
(A3) By denition, it is clear, that for all y A
,
1
y =
g
0
y = y.
(A4) Let y = e
n
A and g
m
G. By denition,
g
m
e
nm
= e
n
. Also
g
m
1 = 1 and
g
m
0 = 0, verifying that (A4) holds.
(A5) Let y
1
, y
2
A
and let h = g
k
such that
h
y
1
=
h
y
2
. Then the following three cases
can occur:
(a) If
h
y
1
=
h
y
2
= 1, then y
1
= y
2
= 1.
(b) If
h
y
1
=
h
y
2
= 0, then y
1
= y
2
= 0.
(c) If
h
y
1
=
h
y
2
= e
m
, then
h
1
(
h
y
1
) =
h
1
(
h
y
2
) =
h
1
e
m
, from which it follows
that y
1
= y
2
=
h
1
e
m
.
Hence (A5) is satised.
We may deduce that G acts on A
on the left by automorphisms. Since by (A4) for all

n Z, e
n
=
g
n
e
0
, and since e
n
e
m
= 0 where n, m Z (n = m), we also obtain that the
semilattice A
is nitely generated with respect to the action of G by 1 and e

0
.
Before presenting our last example, we make a few basic observation about the free
semilattice generated by a set A. To be more precise, in our case A is going to be an
innite set. Let F be the free semilattice generated by A = {. . . , e
1
, e
0
, e
1
, . . .}. Since
F is the free semilattice, for every element f F, there exists a unique subset {f
1
, . . . , f
m
}
of A, such that
f = f
1
. . . f
m
(4.1)
and the following conditions hold:
(F1) f
i
= f
j
for all i = j, 1 i, j m.
(F2) If f = l
1
. . . l
n
, where l
i
= l
j
(i = j, 1 i, j n) then m = n and {f
1
, . . . , f
m
} =
{l
1
, . . . , l
n
}.
In the next proposition, if we write
f = f
1
. . . f
m
, (4.2)
then we assume that conditions (F1) and (F2) hold.
Proposition 4.2.6. Let F be the free semilattice generated by innitely many elements
A = {. . . , e
1
, e
0
, e
1
, . . .}. Adjoin an identity 1 on top that is for each e
i
A
0
, e
i
< 1
holds. Denote by F
the (semi)lattice obtained. Let G = g be the innite cyclic group.

Let n, k Z and consider the following map:
: F
G F
; (e
n
, g
k
)
g
k
e
n
= e
n+k
; (1, g
k
)
g
k
1 = 1,
where e
n
A. If e F so that e = f
1
f
2
. . . f
m
, then we dene
(e, g
k
)
g
k
e =
g
k
f
1
g
k
f
2
. . .
g
k
f
m
.
With this action, G acts on Y
by automorphisms. Furthermore Y
is generated by
Y
0
= {e
0
, 1} with respect to this action.
Proof. That f is a well-dened map follows, since (F1) and (F2) hold. We note that
for g
n
G, we have by denition that
g
n
(
g
k
f
1
g
k
f
2
. . .
g
k
f
m
) =
g
n
(
g
k
f
1
)
g
n
(
g
k
f
2
) . . .
g
n
(
g
k
f
m
)
=
g
n+k
f
1
g
n+k
f
2
. . .
g
n+k
f
m
.
We are now ready to show that conditions (A1)-(A5) hold.
(A1) Clearly if y equals to 1, then for all g
1
, g
2
G,
g
1
g
2
y =
g
1
(
g
2
y). Let y = f
1
. . . f
n
,
where f
1
, . . . , f
n
A and let g
1
= g
m
, g
2
= g
k
. Then on one hand
g
1
g
2
y =
g
m
g
k
(f
1
. . . f
n
) =
g
m+k
(f
1
. . . f
n
) =
g
m+k
f
1
. . .
g
m+k
f
n
.
On the other hand
g
1
(
g
2
y) =
g
m
(
g
k
(f
1
. . . f
n
)) =
g
m
(
g
k
f
1
. . .
g
k
f
n
) =
g
m+k
f
1
. . .
g
m+k
f
n
proving that (A1) indeed holds.
(A2) Let y
1
, y
2
F
and h = g
k
. If y
1
= 1 and y
2
F is arbitrary, then
h
(y
1
y
2
) =
h
y
2
=
h
1
h
y
2
=
h
y
1
h
y
2
. The only remaining case to consider is when y
1
, y
2
F.
Assume that y
1
= f
1
. . . f
m
and that y
2
= l
1
. . . l
n
, where f
i
, l
j
A. Then,
on one hand
h
(y
1
y
2
) =
g
k
(f
1
. . . f
m
l
1
. . . l
n
) =
g
k
f
1
. . .
g
k
f
m
g
k
l
1
. . .
g
k
l
n
.
On the other hand
h
y
1
h
y
2
=
g
k
(f
1
. . . f
m
)
g
k
(l
1
. . . l
n
) =
g
k
f
1
. . .
g
k
f
m
g
k
l
1
. . .
g
k
l
n
.
Thus (A2) holds.
(A3) By denition, it is clear that for all y F
,
1
y =
g
0
y = y.
(A4) Let e = f
1
. . . f
n
F and g
m
G. Let l =
g
m
e. By (A1) and (A3) we have that
g
m
l =
g
m
(
g
m
e) =
1
e = e. Also, if e = 1 then
g
m
1 = 1, verifying that (A4) holds.
(A5) Let y
1
, y
2
F
and let h = g
k
such that
h
y
1
=
h
y
2
. Then either y
1
= y
2
= 1 or
y
1
= y
2
= 0, or y
1
= f
1
. . . f
n
and y
2
= l
1
. . . l
m
. In the latter case we have
that
g
k
y
1
=
g
k
(f
1
. . . f
n
) =
g
k
(l
1
. . . l
m
) =
g
k
y
2
.
That is,
g
k
f
1
. . .
g
k
f
n
=
g
k
l
1
. . .
g
k
l
m
. Having in mind that (F1) and (F2) hold,
we have that
B
1
= {
g
k
f
1
, . . . ,
g
k
f
n
} = {
g
k
l
1
, . . . ,
g
k
l
m
} = B
2
,
where n = m. In particular, we have that for each
g
k
f
i
B
1
there exists exactly one
element
g
k
l
j
B
2
such that
g
k
f
i
=
g
k
l
j
and vica versa. It follows that
{f
1
, . . . , f
n
} = {l
1
, . . . , l
m
},
and hence y
1
= y
2
, verifying that (A5) is satised.
We may now deduce that G acts on F
on the left by automorphisms. By denition for

all n Z, e
n
=
g
n
e
0
. Moreover every element of F is a product of nitely many elements
of A, therefore we may deduce that F
is nitely generated with respect to the action of

G by 1 and e
0
.
Let (Y, ) be a semilattice and (G, ) be a group. Assume that G acts on Y on the
left by automorphisms. The semidirect product S = Y G of Y by G is the set Y G
equipped with the following multiplication
(e, g)(f, h) = (e
g
f, g h).
With this construction an inverse semigroup is obtained. We recall some basic properties
of the semidirect product S = Y G.
Proposition 4.2.7. Let (Y, ) be a semilattice and (G, ) be a group. Assume that G acts
on Y on the left by automorphisms and consider the semidirect product S = Y G. Then
we have that
(i) (e, g)
1
= (
g
1
e, g
1
);
(ii) (e, g) (f, h) if and only if e f in Y and g = h;
(iii) (e, g) E(S) if and only if g = 1;
(iv) (e, g) R (f, h) if and only if e = f.
(v) (e, g) L (f, h) if and only if
g
1
e =
h
1
f.
4.3 Finite generation.
In this section, we give a necessary and sucient condition for a semidirect product of a
semilattice by a group to be nitely generated. Throughout this section, if we say that
the group G acts on a semilattice Y , then it will be understood that G acts on Y on the
left by automorphisms.
Proposition 4.3.1. Let (Y, ) be a semilattice, and (G, ) be a group acting on Y . The
semidirect product S = Y G is nitely generated if and only if the following conditions
hold:
(i) G is nitely generated;
(ii) Y satises the maximum condition;
(iii) Y is nitely generated with respect to the action of G.
Proof. (=) Assume that S = Y G is nitely generated by the elements A =
{(e
1
, h
1
), . . . , (e
n
, h
n
)}, where A S. By Proposition 4.2.7, we obtain that A
1
=
{(
h
1
1
e
1
, h
1
1
), . . . , (
h
1
n
e
n
, h
1
n
)}. Let
X = {h
1
, h
2
, . . . , h
n
} and Y
0
= {e
1
, . . . , e
n
} {
g
1
e | (e, g) A}.
Let (e, g) S. Write
(e, g) = (f
1
, g
1
)(f
2
, g
2
) . . . (f
k
, g
k
)
where (f
j
, g
j
) A A
1
for all 1 j k. Then, on one hand
g = g
1
. . . g
k
,
where g
j
XX
1
verifying that G is generated by the nite set X. On the other hand,
e = f
1
g
1
f
2
. . .
g
1
...g
k1
f
k
,
where f
1
, . . . , f
k
Y
0
. It follows that for each e Y there exist f
1
, . . . , f
k
Y
0
, and
t
1
, . . . , t
k1
G such that
e = f
1
t
1
f
2
. . .
t
k1
f
k
, (4.3)
proving that Y =
G
Y
0
. We also obtain from (4.3), that e f
1
. That is to say that every
element of Y is less than or equal to an element of Y
0
with respect to the natural partial
order. In particular, we have that the maximal elements of Y are the maximal elements
of Y
0
, and so we deduce that Y satises the maximum condition.
4.3. FINITE GENERATION. 107
(=) For the converse, let G be a group acting on a semilattice Y , where G and Y
satisfy conditions (i) (iii). More precisely, assume that the nite set X generates G as
a group, and suppose that
G
Y
0
= Y . For the sake of simplicity we assume that X G
and Y
0
Y . Let Y
m
denote the set of maximal elements of Y and let Y
0
= Y
m
Y
0
. By
assumption, Y
m
is a nite set, and so Y
0
is a nite set as well. Moreover, we still have
that
G
Y
0
= Y . We claim that the set
A = {(e, 1): e Y
0
} {(e, h): e Y
m
, h X}
generates S as an inverse semigroup.
Let (e, g) Y
0
G. We verify that (e, g) can be written in terms of elements of AA
1
.
Let e Y
m
be such that e e and suppose that g = g
1
g
2
. . . g
k
where g
1
, g
2
, . . . , g
k

X X
1
. By Lemma 4.2.1, there exist f
1
, f
2
, . . . , f
k1
Y
m
such that
g
1
f
1
= e and
g
j
f
j
= f
j1
for all 2 j k 1. It follows that
g
1
...g
j
f
j
= e for all 1 j k 1, and we
obtain
(e, 1)( e, g
1
)(f
1
, g
2
) . . . (f
k1
, g
k
) = (e e
g
1
f
1
. . . (
g
1
g
2
...g
k1
f
k1
), g)
= (e e e . . . e
. .
k
, g) = (e e, g)
= (e, g),
verifying that any element of Y
0
G can be written in terms of elements of A A
1
.
Let e Y be arbitrary. Since
G
Y
0
= Y , there exist t
1
, . . . , t
k
G, and f
1
, . . . , f
k
Y
0
such that
e =
t
1
f
1
. . .
t
k
f
k
.
Since Y satises the maximum condition, there exists e Y
m
such that e e, and so
e = e e holds. Let g G and write g = t
1
t
2
. . . t
k
u where u G. Then
(e, g) = ( e, t
1
)(f
1
, t
1
1
t
2
) . . . (f
k1
, t
1
k1
t
k
)(f
k
, t
1
k
)( e, u).
By the above argument each component of the product can be written in terms of elements
of A A
1
, and thus S is nitely generated as an inverse semigroup by A.
We have the following useful corollary of the above proof.
Corollary 4.3.2. Let G be a group generated as a group by a nite set X and let 1
denote the identity element of G. Let Y be a semilattice satisfying the maximum condition
and let Y
m
denote the set of maximal elements of Y . Assume that G acts on Y on the
left and that the nite set Y
0
containing Y
m
generates Y with respect to the action of
G. Then the semidirect product S = Y G is generated as an inverse semigroup by
A = (Y
0
{1}) (Y
m
X).
We end this section by giving three examples for nitely generated semidirect products.
Example 4.3.3. Let Y
be the double innite chain with an identity element adjoined

on top and let G be the innite cyclic group. Dene the action of G on Y
as it was in
Proposition 4.2.4. Clearly G is nitely generated and Y
satises the maximum condition.

According to Proposition 4.2.4, Y
is nitely generated with respect to the action of G.

Hence, by Proposition 4.3.1, the semidirect product S = Y
G is nitely generated.
Example 4.3.4. Consider the semilattice A
introduced in Proposition 4.2.5 and let G

be the innite cyclic group. Dene the action of G on A
as it was in Proposition 4.2.5.

Clearly G is nitely generated and A
satises the maximum condition. According to

Proposition 4.2.5, A
is nitely generated with respect to the action of G. Hence, by

Proposition 4.3.1, the semidirect product S = A
Example 4.3.5. Consider the semilattice F
introduced in Proposition 4.2.6 and let G

be the innite cyclic group. Dene the action of G on F
as it was in Proposition 4.2.6.

Clearly G is nitely generated and F
satises the maximum condition. We have also

shown in Proposition 4.2.6 that F
is nitely generated with respect to the action of G.

Hence, by Proposition 4.3.1, the semidirect product S = F
4.4 Finite presentability with respect to an action I.
The main purpose of this section is to introduce the concept of nite presentability of a
semigroup with respect to a semigroup action. We investigate basic properties regarding
this new notion. Among other things, we prove that a nitely presented semigroup on
which a nitely generated semigroup acts on the left by endomorphisms is also nitely
presented with respect to the action of the nitely generated semigroup. We set the
concept introduced in this section into an inverse semigroup theoretic context in Section
6.
Throughout this section, if we say that a semigroup S acts on a semigroup T, then
it will be understood that S acts on T on the left by endomorphisms. Moreover, if
S = X|P, then we assume for the sake of convenience that X S. We let denote the
empty word.
4.4. FINITE PRESENTABILITY WITH RESPECT TO AN ACTION I. 109
Let X and Y be non-empty sets and let Z = X
. Consider the set

Z
Y = {
y | Z, y Y }.
In other words, the elements of
Z
Y consist of symbols of the form
x
1
...xn
y, where y
Y, x
i
X {}. Consider the map:
f : (
Z
Y )
+
Z (
Z
Y )
+
(
y, )
y, (y, ) y, (y
1
. . . y
n
, )
y
1
. . .
y
n
, (4.4)
where Z \ {}, Z and y
1
, . . . , y
n

Z
Y . Clearly, f satises (A1) and (A2) and so
Z acts on (
Z
Y )
+
. As before, we write the image of (y
1
. . . y
n
, ) under f as

(y
1
. . . y
n
).
That is,
(y
1
. . . y
n
)
y
1
. . .
y
n
.
In particular we obtain Y
Z
Y , since for all y Y,

y y. Throughout this section,
we consider (
Z
Y )
+
as a semigroup on which the semigroup Z acts, where the action is
determined by the map f, dened in (4.4).
Denition 4.4.1. Let X|P be a semigroup presentation, Z = X
and Y be a non-
empty set. A semigroup presentation with respect to the action of X|P is an ordered
pair Y |Q, where Q is a binary relation on (
Z
Y )
+
. Let
Q
A
= {
u
p =
v
q | (p = q) Q, (u = v) P, , Z}
{
u
w =
v
w | w (
Z
Y )
+
, (u = v) P, , Z}
{
p =

q | (p = q) Q, Z}.
We note that Q Q
A
. The semigroup dened by the semigroup presentation
Z
Y |Q
A
is
said to be presented with respect to the action of X|P by the generators Y and relations
Q and we denote this by
T = Act
X|P
Y |Q.
If Y and Q can be chosen to be nite sets, then we say that T is nitely presented with
respect to the action of X|P. If we x a presentation X|P for a semigroup S, then we
write
T = Act
S
Y |Q
instead of T = Act
X|P
Y |Q, and say that T is presented with respect to an action of S
by Y and Q.
Let T = Act
X|P
Y |Q, Z = X
and let w
1
, w
2
(
Z
Y )
+
. We say that w
2
is obtained
from w
1
by an application of a relation of Q
A
, if w
1
u and w
2
v, where
, (
Z
Y )
and u = v is a relation of Q
A
. We say that w
1
= w
2
is a consequence of
relations in Q
A
, if there exists a sequence of words
w
1

0
, . . . ,
m
w
2
such that
j+1
is obtained from
j
by applying a relation of Q
A
.
According to Proposition 4.1.1, we obtain:
Proposition 4.4.2. Let X|P be a semigroup presentation and Y be a non-empty set.
Let Z = X
and Q be a binary relation on (

Z
Y )
+
. Then T = Act
X|P
Y |Q if and only
if the following two conditions hold:
(i) T satises all relations in Q
A
;
(ii) If w
1
, w
2
(
Z
Y )
+
are such that w
1
= w
2
holds in T, then w
1
= w
2
is a consequence
of relations in Q
A
.
Every semigroup can be dened in terms of a semigroup presentation. The question
naturally arises whether a semigroup T on which a semigroup S acts can be dened in
terms of a presentation with respect to the action of S. We give an answer to this problem
in the next proposition.
Proposition 4.4.3. Let S and T be semigroups and assume that S acts on T. Then T
can be dened in terms of a presentation with respect to the action of S.
Proof. Fix a presentation X|P for S. Let Z = X
and let T be given by the

presentation Y |R. Since S acts on T, there exists a function
f : T S T; (t, s)
s
t
satisfying the following two conditions:
(A1)
(s
1
s
2
)
t =
s
1
(
s
2
t); and (A2)
s
(t
1
t
2
) =
s
t
1
s
t
2
for all t, t
1
, t
2
T and s, s
1
, s
2
S. For all y Y and Z, we x a word v
,y
over Y ,
so that
y = v
,y
holds. We note that because of conditions (A1) and (A2) and since f is well-dened, the
following hold: On one hand, if , Z are such that = holds in S, then for all y Y
v
,y
= v
,y
(4.5)
holds in T. More generally, if w y
1
. . . y
n
is a word over Y , then
v
,y
1
. . . v
,yn
= v
,y
1
. . . v
,yn
. (4.6)
On the other hand, for all y Y and , where , Z, if v
,y
y
1
. . . y
n
, then
v
,y
= v
,y
1
. . . v
,yn
(4.7)
holds in T. Let
Q = R {
y = v
,y
| Z, y Y }.
Consider the semigroup presentation
Z
Y |Q
A
, where
Q
A
= {
u
p =
v
q | (p = q) Q, (u = v) P, , Z}
{
u
w =
v
w | w (
Z
Y )
+
, (u = v) P, , Z}
{
p =

q | (p = q) Q, Z}.
We verify that T = Act
S
Y |Q, by showing that the semigroup presentation
Z
Y |Q
A
can be obtained from the semigroup presentation Y |R by applying elementary Tietze

transformations.
It is immediate that applying Tietze transformations of type (T1), the semigroup
presentation
Z
Y |Q is obtained from Y |R. We verify that
Z
Y |Q
A
can be obtained
from
Z
Y |Q by applying Tietze transformation of type (T3). In other words, we show
that every relation in Q
A
\ Q is a consequence of relations in Q.
First, consider a relation in R:
a
1
. . . a
n
= b
1
. . . b
m
.
Let , Z and (u = v) P. Let
1
u and
2
v.
It is immediate that the relation
1
=
2
holds in S. Bearing in mind that f is a well-dened
map and that (A2) holds, we obtain that
v
2
,a
1
. . . v
2
,an
= v
2
,b
1
. . . v
2
,bm
(4.8)
holds in T. In particular, we have that
1
(a
1
. . . a
n
)
1
a
1
. . .
1
a
n
= v
1
,a
1
. . . v
1
,an
applying relations in Q
= v
2
,a
1
. . . v
2
,an
by (4.5)
= v
2
,b
1
. . . v
2
,bm
by (4.8)
=

2
b
1
. . .
2
b
m
2
(b
1
. . . b
m
).
It can be similarly proved that for all Z,
(a
1
. . . a
n
) =

(b
1
. . . b
m
)
is a consequence of relations in Q.
Next, let w (
Z
Y )
+
and let
1
,
2
Z as dened before. Assume that w
1
y
1
. . .
n
y
n
, where
i
Z and

i
y
i
= y
i
, if
i
is the empty word. Let
i

1
i
and
i

2
i
, (1 i n). Since
1
=
2
holds in S, we also have that
i
=
i
holds
for all 1 i n. Making use of (4.5), we obtain that v
i
,y
i
= v
i
,y
i
holds in T for all
1 i n. Hence, we have that
1
w
1
y
1
. . .
1
n
y
n
1
y
1
. . .
n
y
n
= v
1
,y
1
. . . v
n,yn
= v
1
,y
1
. . . v
n,yn
by (4.5)
=

1
y
1
. . .
n
y
n
=

2
1
y
1
. . .
2
n
y
n
2
w.
Finally, consider a relation of the form

y = v
,y
in Q. Let
1
and
2
be the words over
X as dened before. Let
1
and
2
. Since
1
=
2
holds in S, we have that
= also holds in S. Assume that v
,y
y
1
. . . y
n
. Then,
y = v
,y
applying a relation in Q
= v
,y
by (4.5)
= v
2
,y
1
. . . v
2
,yn
by (4.7)
=

2
y
1
. . .
2
y
n
2
(y
1
. . . y
n
)
2
v
,y
.
It can be similarly proved that for all Z,
y) =

v
,y
is a consequence of relations in Q.
From the proof of Proposition 4.4.3 we obtain the following useful corollary:
Corollary 4.4.4. Let S = X|P, Z = X
and T = Y |R. Assume that S acts on T.

For each Z and y Y , let v
,y
be a word over Y representing

y. Let
Q = R {
y = v
,y
| Z, y Y }
and
Q
A
= {
u
p =
v
q | (p = q) Q, (u = v) P, , Z}
{
u
w =
v
w | w (
Z
Y )
+
, (u = v) P, , Z}
{
p =

q | (p = q) Q, Z}.
Then the following hold:
(1) Every relation in Q
A
(2) T = Act
S
Y |Q =
Z
Y |Q
A
=
Z
Y |Q.
Proposition 4.4.5. Let T be a nitely presented semigroup and assume that a nitely
generated semigroup S acts on T. Then T is also nitely presented with respect to the
action of S.
Proof. Fix a semigroup presentation X|P for S, where X is a nite set and let
Z = X
. Let T be dened by the semigroup presentation Y |R, where Y is a nite set

and R is a nite set of relations on Y
+
. For all y Y and Z we x a word v
,y
over
Y so that
y = v
,y
.
Let
Q = R {
y = v
,y
| Z, y Y },
and consider the semigroup presentation
Z
Y |Q
A
, where
Q
A
= {
u
p =
v
q | (p = q) Q, (u = v) P, , Z}
{
u
w =
v
w | w (
Z
Y )
+
, (u = v) P, , Z}
{
p =

q | (p = q) Q, Z}.
Making use of Corollary 4.4.4, we have that T =
Z
Y |Q
A
=
Z
Y |Q. Let
Q = R {
x
y = v
x,y
| x X, y Y }.
Since R is a nite set of relations and since X and Y are nite sets,

Q is a nite set of
relations. Let
Q
A
= {
u
p =
v
q | (p = q)

Q, (u = v) P, , Z}
{
u
w =
v
w | w (
Z
Y )
+
, (u = v) P, , Z}
{
p =

q | (p = q)

Q, Z}.
We claim that T is nitely presented with respect to the action of S, by verifying that
Z
Y |
Q
A
can be obtained from
Z
Y |Q by applying elementary Tietze transformations.
Clearly

Q
A
Q
A
. By Corollary 4.4.4, the elements of Q
A
\ Q and hence of

Q
A
\

Q are
consequences of relations in Q, and so we obtain
T =
Z
Y |Q

Q
A
.
Finally, we claim by induction on the length of words Z, that all relations of the form
y = v
,y
, where || 2 are consequences of relations in

Q
A
.
We recall that by (4.7), for all y Y and , where , Z, if v
,y
y
1
. . . y
n
,
then
v
,y
= v
,y
1
. . . v
,yn
. (4.9)
Let x
1
x
2
. Then
x
2
y = v
x
2
,y
is a relation in

Q, and so
x
1
x
2
y =
x
1
v
x
2
,y
is a relation in
Q
A
. Assume that v
x
2
,y
y
1
. . . y
n
. Then we obtain
x
1
x
2
y =
x
1
v
x
2
,y
applying a relation in

Q
A
x
1
y
1
. . .
x
1
y
n
= v
x
1
,y
1
. . . v
x
1
,yn
applying relations in

Q
= v
x
1
x
2
,y
by (4.9).
Next, assume that for all words over X with length less then m, where 2 m, the
relation

y = v
,y
is a consequence of relations in

Q
A
and let Z such that || = m.
Then = x for some x X and Z, where || = m1. Since
x
y = v
x,y
is a relation
in

Q we have that
x
y =

v
x,y
is an element of

Q
A
. Assume that v
x,y
y
1
. . . y
n
. Then,
we obtain
y
x
y =

v
x,y
applying a relation in

Q
A
(y
1
. . . y
n
)
y
1
. . .
y
n
= v
,y
1
. . . v
,yn
by the inductive hypothesis
= v
,y
by (4.9).
Hence, applying Tietze transformations of type (T4), we obtain
T =
Z
Y |
Q
A
= Act
S
Y |
Q,
proving that T is indeed nitely presented with respect to the action of S.
We introduce a sharper version of Corollary 4.4.4.
Corollary 4.4.6. Let S = X|P, Z = X
and T = Y |R. Assume that S acts on T.

For each Z and y Y , let v
,y
be a word over Y representing

y. Let
Q = R {
y = v
,y
| Z \ X, y Y }
and
Q = R {
x
y = v
x,y
| x X, y Y }.
Let
Q
A
= {
u
p =
v
q | (p = q) Q, (u = v) P, , Z}
{
u
w =
v
w | w (
Z
Y )
+
, (u = v) P, , Z}
{
p =

q | (p = q) Q, Z}.
Then the following hold:
(1) Every relation in

Q\ Q is a consequence of relations in Q
A
.
(2) Every relation in Q
A
(3) T = Act
S
Y |Q = Act
S
Y |
Q.
Finally, we prove:
Proposition 4.4.7. Let S = X|P, where X is a nite set and let Z = X
. Let T be a
semigroup on which S acts. Assume that the nite sets Y
1
and Y
2
generate the semigroup
T with respect to the action of S and that T = Act
S
Y
1
|Q, where Q is a nite set of
relations on (
Z
Y
1
)
+
. Then there exists a nite set of relations

Q on (
Z
Y
2
)
+
such that
T = Act
S
Y
2
|
Q.
Proof. Since Y
2
generates T with respect to the action of S, for each y Y
1
there exists
a word v
y
over
Z
Y
2
such that y and v
y
represent the same element of T. In particular, for
all Z, we have that

y and

v
y
represent the same element of T. The map
y v
y
,

y
v
y
can be extended to a homomorphism
: (
Z
Y
1
)
+
(
Z
Y
2
)
+
.
We note that for each w (
Z
Y
1
)
+
, w and w represent the same element of T. Consider
the word w
1
y
1
. . .
n
y
n
, where
i
Z and y
i
Y
1
for all 1 i n. Let Z and let
i

i
, where 1 i n. Then
(
w) (
1
y
1
. . .
n
y
n
)
=

1
v
y
1
. . .
n
v
yn
1
v
y
1
. . .
n
v
yn
)
=

(w),
and so for all w (
Z
Y
1
)
+
and Z,
(
w)
(w) (4.10)
holds. Similarly, since Y
1
generates T with respect to the action of S, for each y Y
2
there exists a word v
y
(
Z
Y
1
)
+
such that y and v
y
represent the same element of T. In
particular, for all Z, we have that

y and

v
y
represent the same element of T. The
map
y v
y
,

y
v
y
can be extended to a homomorphism
: (
Z
Y
2
)
+
(
Z
Y
1
)
+
,
and for each w (
Z
Y
2
)
+
, w and w represent the same element of T. Furthermore, for
all w (
Z
Y
2
)
+
and Z,
(
w)
(w). (4.11)
Since T = Act
S
Y
1
|Q, we have that T =
Z
Y
1
|Q
A
, where
Q
A
= {
u
p =
v
q | (p = q) Q, (u = v) P, , Z}
{
u
w =
v
w | w (
Z
Y
1
)
+
, (u = v) P, , Z}
{
p =

q | (p = q) Q, Z}.
Let
Q = {p = q | (p = q) Q} {y = y | y Y
2
}
and let
Q
A
= {
u
p =
v
q | (p = q)

Q, (u = v) P, , Z}
{
u
w =
v
w | w (
Z
Y
2
)
+
, (u = v) P, , Z}
{
p =

q | (p = q)

Q, Z}.
Our aim is to show that T = Act
S
Y
2
|

Q. By the denition of the homomorphisms
and and by (4.10), we have that T satises all relations in

Q and of

Q
A
as well.
Before we proceed and show that every relation w
1
= w
2
, (w
1
, w
2
(
Z
Y
2
)
+
) that holds
in T is a consequence of relations in

Q
A
, we verify that for all y Y
2
and Z
y = (
y)
holds in T. Keeping (4.10) and (4.11) in mind, we obtain for all y Y
2
and Z, that
(
y) ((
y)) (
(y))
(y) (4.12)
holds. Let y Y
2
and Z. Since y = y is a relation in

Q, we have that

y =

(y)
is a relation in

Q
A
. Having (4.12) in mind we have that

(y) (
y) and hence we
obtain that
y = (
y) (4.13)
holds in T.
Let w
1
, w
2
(
Z
Y
2
)
+
such that w
1
and w
2
represent the same element of T. We show
that w
1
= w
2

Q
A
. Since w
1
= w
2
, we have that w
1
= w
2
holds in T, and hence there exists a nite sequence of words over

Z
Y
1
:
w
1
u
1
, u
2
, . . . , u
m
w
2

such that u
i+1
is obtained from u
i
, (1 i m1) by applying a relation in Q
A
. Consider
now the following sequence of words over
Z
Y
2
:
w
1
u
1
, u
2
, . . . , u
m
w
2
.
If u
i+1
is obtained from u
i
, (1 i m 1) by applying a relation in Q, then u
i+1
is obtained from u
i
by applying a relation in

Q. Assume that u
i+1
is obtained from
u
i
, (1 i m1) by applying a relation in Q
A
\ Q. Then the following three cases can
occur:
(a) u
i+1
is obtained from u
i
by applying a relation of the form
u
p =
v
q,
where , Z, (p = q) Q and (u = v) P. Then u
i+1
is obtained from u
i
by
substituting the subword (
u
p) of u
i
by (
v
q). Keeping in mind (4.10), we have
that
(
u
p)
u
(p)
and that
(
v
q)
v
(q),
and so we may deduce that u
i+1
is obtained from u
i

Q
A
.
(b) u
i+1
is obtained from u
i
u
w =
v
w,
where , Z, w (
Z
Y
1
)
+
and (u = v) P. Then u
i+1
is obtained from u
i
by
substituting the subword (
u
w) of u
i
by (
v
w). By (4.10),
(
u
w)
u
(w)
and
(
v
w)
v
(w),
and hence we may deduce that u
i+1
is obtained from u
i
Q
A
.
(c) u
i+1
is obtained from u
i
p =

q,
4.5. EXAMPLES I. 119
where Z, (p = q) Q. Then u
i+1
is obtained from u
i
by substituting the
subword (
p) of u
i
by (
q). Since
(
p)
(p) and (
q)
(q),
we have that u
i+1
is obtained from u
i

Q
A
.
To nish our proof, we need to verify that w
1
= w
1
and w
2
= w
2
is a consequence
of relations in

Q
A
. Assume that w
1

1
y
1
. . .
n
y
n
, where
i
Z and y
i
Y
2
for all
1 i n. Keeping in mind (4.13) and that and are homomorphisms we obtain
w
1

1
y
1
. . .
n
y
n
= (
1
y
1
) . . . (
n
y
n
)
= (
1
y
1
. . .
n
y
n
)
w
1
.
Similarly w
2
= w
2
holds, and so w
1
= w
2

Q
A
.
4.5 Examples I.
The aim of this section is to illustrate through a sequence of examples the concept of
nite presentability with respect to a semigroup action. By Proposition 4.4.5, we know
that if a semigroup T is nitely presented and a nitely generated semigroup S acts on
it, then T is also nitely presented with respect to the action of S. To demonstrate
that T does not necessarily have to be nitely presented as a semigroup in order to be
nitely presented with respect to a semigroup action, we will consider innite semilattices.
The reason we choose innite semilattices is twofold. On one hand innite semilattices
are not nitely generated and hence they are not nitely presented. On the other hand
these examples serve as a preparation for considering semidirect products. We give an
example of an innite semilattice that is nitely presented as a semigroup with respect to
a semigroup action and we give two examples of semilattices that are not nitely presented
as a semigroup with respect to a semigroup action.

on top, and let G be the innite cyclic group. Dene the action of G on Y
as in Propo-
sition 4.2.4. Then Y
is nitely presented as a semigroup with respect to the action of

G.
Proof. Consider the semilattice Y
:
. . . < e
1
< e
0
< e
1
< . . . < 1.
Since Y
is innite, it is not nitely generated and hence it is not nitely presented. To

be more precise, if we let
R = {e
i
e
j
= e
j
e
i
= e
i
, e
i
1 = 1 e
i
= e
i
, 1 1 = 1 | i j, i, j Z},
then we have that Y
is dened by the semigroup presentation Y
|R. Let G = g be
the innite cyclic group. To make notation clear, we let X = {g, g
1
, 1} and let Z = X
.
Let
P = {gg
1
= g
1
g = 1, 1g = g, 1g
1
= g
1
, 11 = 1}.
Clearly G is dened by the semigroup presentation
G = X|P.
Dene the action of G on Y
as in Proposition 4.2.4. According to Proposition 4.2.4,

Y
is nitely generated by {e
0
, 1} with respect to the action of G. It follows that Y
is
generated by the nite set Y
0
= {e
0
, e
1
, 1} with respect to this action of G as well. Let
Q = {e
0
e
1
= e
1
e
0
= e
0
, e
0
e
0
= e
0
, 1 e
0
= e
0
1 = e
0
,
g
e
0
= e
1
,
1
e
0
= e
0
,
1
1 = 1,
g
1 = 1}
and
Q
A
= {
u
p =
v
q | (p = q) Q, (u = v) P, , Z}
{
u
w =
v
w | w (
Z
Y
0
)
+
, (u = v) P, , Z}
{
p =

q | (p = q) Q, Z}.
Let Y =
Z
Y
0
|Q
A
. We claim that Y
= Act
G
Y
0
|Q, by giving two homomorphism
: Y
Y and : Y Y
that are inverse to each other.

We rst show that the map e
n

g
n
e
0
, 1 1 induces a homomorphism : Y
Y
by showing that relations in R are mapped onto relations that hold in Y . First consider
a relation of the form e
n
e
m
= e
n
, where n m. Then
e
n
e
m
=
g
n
e
0
g
m
e
0
=
g
n
e
0
g
n
e
1
g
m
e
0
A
=
g
n
e
0
g
n+1
e
0
g
m
e
0
A
.
.
.
=
g
n
e
0
g
n+1
e
0
. . .
g
m1
e
0
g
m
e
0
A
=
g
n
e
0
g
n+1
e
0
. . .
g
m1
(e
0
e
1
) applying a relation in Q
A
=
g
n
e
0
g
n+1
e
0
. . .
g
m1
e
0
A
.
.
.
=
g
n
e
0
= e
n
,
and so (e
n
e
m
) = e
n
holds. It can be similarly proved that all relations of the form
e
m
e
n
= e
n
, where n m are mapped onto relations that hold in Y .
It is immediate by the denition of that 1 1 = 1 is mapped onto the relation
1 1 = 1.
Before we proceed, we rst claim that
g
k
1 = 1 holds in Y for all k 0 by induction.
Clearly
g
2
1 =
g
1 applying a relation in Q
A
= 1 applying a relation in Q.
Assume that
g
k
1 = 1 holds in Y for all k < m. Then 1 =
g
m1
1 holds. On the other hand
g
m1
(
g
1) =
g
m1
1 is a relation in Q
A
, and so we may deduce that
g
m
1 = 1 holds in Y .
We also note that
g
1
1
1 =
g
1
g
1,
g
1
g
1 =
1
1,
g
1
1
1 =
g
1
1 are relations in Q
A
, and
so
g
1
1 = 1
holds in Y . Furthermore, for all k 0,
g
k
1 = 1 holds.
Consider a relation of the form e
n
1 = e
n
. Then
e
n
1 =
g
n
e
0
1
=
g
n
e
0
g
n
1 applying relations in Q
A
=
g
n
e
0
applying a relations in Q
A
= e
n
,
and hence (e
n
1) = e
n
. It can be similarly proved that all relations of the form
1 e
n
= e
n
are mapped onto relations that hold in Y .
Before we dene a homomorphism from Y to Y
, we introduce the following notation.

Let Z, and assume that the group reduced word obtained from is g
n
. Then we let
|||| = n.
We verify that the map

e
0
e
||||
,

e
1
e
||||+1
,

1 1 induces a homomorphism
: Y Y
, by showing that relations in Q

A
are mapped onto relations that hold in Y
.
First, we consider relations in Q. We have the following seven cases:
(i) e
0
e
1
= e
0
e
1
= e
0
= e
0
= (e
0
e
1
). Similarly, (e
0
e
1
) = e
0
holds.
(ii) e
0
e
0
= e
0
e
0
= e
0
= e
0
= (e
0
e
0
).
(iii) (1 1) = 1.
(iv) (
g
e
0
) = e
1
= e
1
.
(v) (
1
e
0
) = e
0
= e
0
.
(vi) (
g
1) = 1 = (1).
(vii) (
1
1) = 1 = (1).
Thus, we may deduce that relations of Q are mapped onto relations that hold in Y
.
1
u and
2
v. Since (u = v) P, we
have that ||u|| = ||v||, and it follows that ||
1
|| = ||
2
||. Consider the relation e
0
e
1
= e
0
in Q. Then
(
1
e
0
) (
1
e
1
) = e
||
1
||
e
||
1
||+1
= e
||
1
||
= e
||
2
||
= (
2
e
0
) = (
1
(e
0
e
1
))
holds. Similarly, we have that
(
e
0
) (
e
1
) = e
||||
e
||||+1
= e
||||
= (
e
0
) = (
(e
0
e
1
))
holds. It can be proved similarly, that for any relation (p = q) Q,
(
1
p) = (
2
q) and (
p) = (
q)
hold. Finally, let
w
1
y
1
. . .
n
y
n
,
where
i
Z and y
i
Y
0
. Let
i

1
i
and
i

2
i
,
where 1 i n. Clearly, ||
i
|| = ||
i
|| for all 1 i n. It is straightforward that
(
1
w) = (
2
w) holds, if y
i
= 1 for all 1 i n. Assume that there exists y
i
= 1. Then,
applying the appropriate relations in Q
A
, we may assume that in w each y
i
= e
0
. Let
i
Z be such that ||
i
|| ||
j
|| for all 1 j m. Then, applying the appropriate
relations in Q
A
, we obtain that
1
(
1
y
1
. . .
n
y
n
) =

i
y
i
and that
2
(
1
y
1
. . .
n
y
n
) =

i
y
i
.
Taking into account that
(
i
y
i
) = e
||
i
||
= e
||
i
||
= (
i
y
i
)
holds, we obtain that (
1
w) = (
2
w) holds in Y
.
That and are inverse to each other follows from the following facts. On one hand
we have
(e
n
) = (
g
n
e
0
) = e
n
and 1 = 1.
On the other hand
(
e
0
) = (e
||||
) =
g
||||
e
0
=

e
0
,
(
e
1
) = (e
||||+1
) =
g
||||+1
e
0
=

(
g
e
0
) =

e
1
and
1 = 1.
Proposition 4.5.2. Let A
be the semilattice obtained by adjoining an identity and a

zero element to an innite antichain. Let G be the innite cyclic group. Dene the action
of G on A
as in Proposition 4.2.5. Then A
is not nitely presented as a semigroup

with respect to the action of G.
Proof. Let A be the innite antichain
. . . , e
1
, e
0
, e
1
, . . . .
Adjoin an identity 1 on top and a zero 0 on bottom. Then, for each e
i
A, 0 < e
i
< 1
holds. Let A
denote the (semi)lattice obtained. Since A
is an innite semilattice, it is
not nitely generated and hence it is not nitely presented. To be more precise, if we let
R = {e
i
e
j
= e
j
e
i
= 0, e
i
1 = 1 e
i
= e
i
, e
i
0 = 0 e
i
= 0,
e
i
e
i
= e
i
, 1 1 = 1, 0 0 = 0 | i, j Z, i = j},
then we have that A
is dened by the semigroup presentation A
|R. Let G = g be
1
, 1} and let Z = X
.
Let
P = {gg
1
= g
1
g = 1, 1g = g, 1g
1
= g
1
, 11 = 1}.
G = X|P.
Dene the action of G on A
as in Proposition 4.2.5. As we veried in Proposition 4.2.5,

A
is nitely generated by Y
0
= {e
0
, 1} with respect to the action of G.
Assume that A
is nitely presented with respect to the action of G. Then, according

to Proposition 4.4.7, there exists a nite set of relations Q on (
Z
Y
0
)
+
such that A
=
Act
G
Y
0
|Q. In other words, if we let
Q
A
= {
u
p =
v
q | (p = q) Q, (u = v) P, , Z}
{
u
w =
v
w | w (
Z
Y
0
)
+
, (u = v) P, , Z}
{
p =

q | (p = q) Q, Z}
then A
=
Z
Y
0
|Q
A
.
We make some observations. Every relation in Q is of the following form:
1
a
1
. . .
n
a
n
=

1
b
1
. . .
m
b
m
,
where a
i
, b
j
Y
0
and
i
,
j
Z for all 1 i n, 1 j m. Without loss of generality
we may assume that the following nite set of relations:
Q = {e
0
e
0
= e
0
, e
0
1 = 1 e
0
= e
0
, 1 1 = 1,
g
1 =
1
1 = 1}
is contained in Q. To make notation convenient, we let
Q
A
= {
u
p =
v
q | (p = q)

Q, (u = v) P, , Z}
{
u
w =
v
w | w (
Z
Y
0
)
+
, (u = v) P, , Z}
{
p =

q | (p = q)

Q, Z}.
As in the proof of the previous proposition, it can be veried that for all k Z, the relation
g
k
1 = 1
is a consequence of relations in Q
A
. Since for all k Z, the relations
g
k
e
0
g
k
e
0
=
g
k
e
0
and
g
k
e
0
g
k
1 =
g
k
1
g
k
e
0
=
g
k
e
0
are relations in Q
A
, we may assume that for all relations
1
a
1
. . .
n
a
n
=

1
b
1
. . .
m
b
m
,
in Q\

Q the following hold:
(a) a
i
= b
j
= e
0
for all 1 i n, 1 j m;
(b) there is no subword of p and of q of the form

e
0
e
0
;
(c)
i
,
j
are group reduced words for all 1 i n, 1 j m.
We also note that n = 1 if and only if m = 1, in which case the relation in Q \

Q reads
e
0
=

e
0
, and hence is redundant. Because of this fact, we assume that for all relations
1
a
1
. . .
n
a
n
=

1
b
1
. . .
m
b
m
in Q\

Q,
n, m 2
also holds. Let
I = { Z |

e
0
occurs in one of the relations in Q}.
By our assumptions, for every I, we have = g
k
for some k Z. Let i, j Z such
that g
i
, g
j
I, and if g
k
I, then j k i. Let d = i j.
Clearly, the relation
g
i+1
e
0
g
j1
e
0
=
g
j1
e
0
g
i+1
e
0
holds in A
. We claim that the above relation is not a consequence of the relations in

Q
A
. For this, assume that there exists a nite sequence of words
g
i+1
e
0
g
j1
e
0
u
0
, u
1
, . . . , u
n

g
j1
e
0
g
i+1
e
0
,
such that u
t+1
is obtained from u
t
, (0 t n 1) by applying a relation in Q
A
.
Let us take a closer look how the word u
1
can be obtained:.
(i) It is straightforward to see that we cannot apply any of the relations in Q to the
word u
0
, since
g
i+1
e
0
and
g
j1
e
0
do not occur in any of the relations in Q, and so to
obtain a dierent word from u
0
, we need to apply a relation in Q
A
\ Q.
(ii) It is also obvious that by applying a relation in

Q
A
, the order of the elements
g
i+1
e
0
and
g
j1
e
0
in the word p cannot be changed.
(iii) If a relation in Q
A
\

Q
A
can be applied, then it is either of the form
p =

q,
where (p = q) Q\

Q, and Z \ {}, where denotes the empty word, or of the
form
u
p =
v
q,
where (p = q) Q\

Q, and , Z. Considering both cases, it follows that there
exists a relation p = q in Q\

Q, such that p
g
l
e
0
g
k
e
0
, and a word g
m
Z, such
that
g
i+1
e
0
g
j1
e
0

g
m+l
e
0
g
m+k
e
0
.
It follows that l k = m+l mk = i + 1 j + 1 = d + 2, which contradicts our
choice of the number d. We may now deduce that relations in Q
A
\

Q
A
cannot be
applied to the word u
0
.
To summarize the above argument, we have that only relations in

Q
A
\

Q can be applied
to the word u
0
to obtain a dierent word u
1
, but applying relations in

Q
A
\

Q, the order
of the elements
g
i+1
e
0
and
g
j1
e
0
in the word p cannot be changed. Inductively, it can be
seen that for all 1 t n 1, u
t+1
is obtained from u
t
, (1 t n 1) by applying a
relation in

Q
A
\

Q, but doing so the order of the elements
g
i+1
e
0
and
g
j1
e
0
in the word u
t
cannot be changed. We may now deduce that the relation
g
i+1
e
0
g
j1
e
0
u
0
, u
1
, . . . , u
n

g
j1
e
0
g
i+1
e
0
,
is not a consequence of relation in Q
A
, and so A
is not nitely presented with respect

to the action of G.
Proposition 4.5.3. Let F
be the semilattice obtained by adjoining an identity element

to the free semilattice generated by innitely many elements. Let G be the innite cyclic
group. Dene the action of G on F
as in Proposition 4.2.6. Then F
is not nitely
presented as a semigroup with respect to the action of G.
Proof. Let F be the free semilattice generated by innitely many elements A =
{. . . , e
1
, e
0
, e
1
, . . .}. Adjoin an identity 1 on top. Then, for each e
i
A, e
i
< 1
holds. Denote by F
the (semi)lattice obtained. Clearly F
is not nitely presented,

since it is not nitely generated. To be more precise, if we let
R = {e
i
e
j
= e
j
e
i
, e
i
1 = 1 e
i
= e
i
, e
i
e
i
= e
i
, 1 1 = 1 | i, j Z},
then we have that F
is dened by the semigroup presentation A, 1|R. Let G = g be

1
, 1} and let Z = X
.
Let
P = {gg
1
= g
1
g = 1, 1g = g, 1g
1
= g
1
, 11 = 1}.
G = X|P.
Dene the action of G on F

F
0
= {e
0
, 1} with respect to the action of G.
Assume that F
is nitely presented with respect to the action of G. Then, according

to Proposition 4.4.7, there exists a nite set of relations Q on (
Z
Y
0
)
+
such that F
=
Act
G
Y
0
Q
A
= {
u
p =
v
q | (p = q) Q, (u = v) P, , Z}
{
u
w =
v
w | w (
Z
Y
0
)
+
, (u = v) P, , Z}
{
p =

q | (p = q) Q, Z},
then F
=
Z
Y
0
|Q
A
.
As in the previous proposition, we now make some observations. Every relation in Q
is of the following form:
1
a
1
. . .
n
a
n
=

1
b
1
. . .
m
b
m
,
where a
i
, b
j
Y
0
and
i
,
j
Z for all 1 i n, 1 j m. Without loss of generality
we may assume that the following nite set of relations:
Q = {e
0
e
0
= e
0
, e
0
1 = 1 e
0
= e
0
, 1 1 = 1,
g
1 =
1
1 = 1}
is contained in Q. To make notation convenient, we let
Q
A
= {
u
p =
v
q | (p = q)

Q, (u = v) P, , Z}
{
u
w =
v
w | w (
Z
Y
0
)
+
, (u = v) P, , Z}
{
p =

q | (p = q)

Q, Z}.
As in the previous propositions, it can be veried that
g
k
1 = 1
A
. Since for all k Z, the relations
g
k
e
0
g
k
e
0
=
g
k
e
0
and
g
k
e
0
g
k
1 =
g
k
1
g
k
e
0
=
g
k
e
0
are contained in Q
A
, we may assume that for all relations
p
1
a
1
. . .
n
a
n
=

1
b
1
. . .
m
b
m
q
in Q\

Q the following hold:
(a) a
i
= b
j
= e
0
for all 1 i n, 1 j m;
(b) there is no subword of p and of q of the form

e
0
e
0
;
(c)
i
,
j
are group reduced words for all 1 i n, 1 j m.
We also note that n = 1 if and only if m = 1, in which case the relation in Q \

Q reads
e
0
=

e
0
, and hence is redundant. Because of this fact, we assume that for all relations
1
a
1
. . .
n
a
n
=

1
b
1
. . .
m
b
m
in Q\

Q,
n, m 2
also holds. Let
I = { Z |

e
0
k
that g
i
, g
j
I, and if g
k
I, then j k i. Let d = i j.
Clearly, the relation
g
i+1
e
0
g
j1
e
0
=
g
j1
e
0
g
i+1
e
0
holds in F
. We claim that the above relation is not a consequence of the relations in

Q
A
. For this, assume that there exists a nite sequence of words
g
i+1
e
0
g
j1
e
0
u
0
, u
1
, . . . , u
n

g
j1
e
0
g
i+1
e
0
,
such that u
t+1
is obtained from u
t
, (0 t n 1) by applying a relation in Q
A
.
1
can be obtained:
4.6. FINITE PRESENTABILITY WITH RESPECT TO AN ACTION II. 129
(i) It is straightforward, that we cannot apply any of the relations in Q to the word u
0
,
since
g
i+1
e
0
and
g
j1
e
0
do not occur in any of the relations in Q, and so to obtain a
dierent word from u
0
, we need to apply a relation in Q
A
\ Q.
(ii) It is also obvious, that by applying a relation in

Q
A
, the order of the elements
g
i+1
e
0
and
g
j1
e
0
in the word p cannot be changed.
(iii) If a relation in Q
A
\

Q
A
can be applied, then it is either of the form
p =

q,
where (p = q) Q\

Q, and Z \ {}, where denotes the empty word, or of the
form
u
p =
v
q,
where (p = q) Q\

Q, and , Z. Considering both cases, it follows that there
exists a relation p = q in Q\

Q, such that p =
g
l
e
0
g
k
e
0
, and a word g
m
Z, such
that
g
i+1
e
0
g
j1
e
0

g
m+l
e
0
g
m+k
e
0
.
A
\

Q
A
cannot be
applied to the word u
0
.
To summarize the above argument, we have that only relations in

Q
A
\

Q can be applied
to the word u
0
to obtain a dierent word u
1
, but applying relations in

Q
A
\

Q, the order
of the elements
g
i+1
e
0
and
g
j1
e
0
in the word p cannot be changed. Inductively, it can be
seen that for all 1 t n 1, u
t+1
is obtained from u
t
, (1 t n 1) by applying a
relation in

Q
A
\

Q, but doing so the order of the elements
g
i+1
e
0
and
g
j1
e
0
in the word u
t
cannot be changed. We may now deduce that the relation
g
i+1
e
0
g
j1
e
0
u
0
, u
1
, . . . , u
n

g
j1
e
0
g
i+1
e
0
,
is not a consequence of relation in Q
A
, and so F
is not nitely presented with respect to

the action of G.
4.6 Finite presentability with respect to an action II.
In Section 4, we introduced the concept of a semigroup presentation with respect to an
action of a semigroup and discussed basic properties. In this section we introduce the
concept of an inverse semigroup presentation with respect to an action of a semigroup.
The new notion serves as a key point in giving a necessary and sucient condition for
nite presentability of a semidirect product of a semilattice by a group.
As in the previous sections, if we say that a semigroup S acts on a semigroup T,
then it will be understood that S acts on T on the left by endomorphisms. Moreover, if
S = X|P, then we assume for the sake of convenience that X S. We let denote the
empty word.
As in Section 4, we use the following notations. Let X and Y be non-empty sets, and
let Z = X
. Consider the set

Z
Y = {
y | Z, y Y }.
In other words, the elements of
Z
Y consist of symbols of the form
x
1
...xn
y, where y Y ,
x
i
X {}, where denotes the empty word. Let (
Z
Y )
1
denote the set of formal
inverses of elements of
Z
Y . Let
B =
Z
Y (
Z
Y )
1
.
Consider the following map:
f : B
+
Z B
+
;
(
y, )
y, ((
y)
1
, ) (
y)
1
, (y, ) y, (y
1
, ) y
1
, (4.14)
where Z \ {}, Z and y Y . For any word y
1
. . . y
n
over B, we dene
(y
1
. . . y
n
, )
y
1
. . .
y
n
.
Clearly, f satises (A1) and (A2) and so Z acts on B
+
on the left by endomorphisms. As
before, we write the image of (, y
1
. . . y
n
) under f as

(y
1
. . . y
n
). That is,
(y
1
. . . y
n
)
y
1
. . .
y
n
.
In particular, we have that Y Y
1
B. Throughout this section, we consider B
+
as a
semigroup on which the semigroup Z acts on the left by endomorphisms, where the action
is determined by the map f dened in (4.14).
Denition 4.6.1. Let X|P be a semigroup presentation, Y be a non-empty set and
Z = X
. Let B =
Z
Y (
Z
Y )
1
, where (
Z
Y )
1
is the set of formal inverses of elements
of elements
Z
Y . An inverse semigroup presentation with respect to the action of X|P is
4.6. FINITE PRESENTABILITY WITH RESPECT TO AN ACTION II. 131
an ordered pair Y |Q, where Q is a binary relation on B
+
. Let denote the standard
inverse semigroup relations on FSI(
Z
Y ), and let
Q
R
= {
u
p =
v
q | (p = q) Q , (u = v) P, , Z}
{
u
w =
v
w | w B
+
, (u = v) P, , Z}
{
p =

q | (p = q) Q , Z}.
We note that Q Q
R
. Let Q
A
= Q
R
\ . The semigroup dened by the inverse
semigroup presentation Inv
Z
Y |Q
A
is said to be presented as an inverse semigroup with
respect to the action of X|P by the generators Y and relations Q and we denote this by
T = InvAct
X|P
Y |Q.
If Y and Q can be chosen to be nite sets, then we say that T is nitely presented as an
inverse semigroup with respect to the action of X|P. If we x a presentation X|P for
a semigroup S, then we write
T = InvAct
S
Y |Q
instead of T = InvAct
X|P
Y |Q, and say that T is presented as an inverse semigroup
with respect to an action of S by Y and Q.
Let T = InvAct
X|P
Y |Q and Z = X
. Let B =
Z
Y (
Z
Y )
1
, where (
Z
Y )
1
is the
set of formal inverses of elements
Z
Y . Let w
1
, w
2
B
+
. We say that w
2
is obtained from
w
1
by an application of a relation of Q
R
if w
1
u and w
2
v, where , B
and u = v is a relation in Q
R
. We say that w
1
= w
2
R
,
if there exists a sequence of words
w
1

0
, . . . ,
m
w
2
such that
j+1
is obtained from
j
R
.
According to Proposition 4.1.4, we have:
Proposition 4.6.2. Let X|P be a semigroup presentation and Y be a non-empty set.
Let Z = X
, B =
Z
Y (
Z
Y )
1
and Q be a binary relation on B
+
. Then T =
InvAct
X|P
Y |Q if and only if the following two conditions hold:
(i) T satises all relations in Q
R
;
(ii) If w
1
, w
2
B
+
are such that w
1
= w
2
holds in T, then w
1
= w
2
is a consequence of
relations in Q
R
.
Next, we set Propositions 4.4.3, 4.4.5 and 4.4.7 into an inverse semigroup theoretic con-
text. Propositions 4.6.3, 4.6.4 and 4.6.5 can be veried similarly to Propositions 4.4.3, 4.4.5
and 4.4.7. To be more accurate, one needs to consider inverse semigroup presentations, Ti-
etze transformations regarding inverse semigroups and inverse semigroup homomorphisms
instead of semigroup presentations, Tietze transformations of semigroups and semigroup
homomorphisms.
Proposition 4.6.3. Let S be a semigroup and T be an inverse semigroup. Assume that
S acts on T. Then T can be dened in terms of an inverse semigroup presentation with
respect to the action of S.
Proposition 4.6.4. Let T be a nitely presented inverse semigroup and assume that S
is a nitely generated semigroup that acts on T. Then T is also nitely presented as an
inverse semigroup with respect to the action of S.
Proposition 4.6.5. Let S = X|P and let T be an inverse semigroup on which S acts.
Assume that the nite sets Y
1
and Y
2
generate T with respect to the action of S and that
T can be dened by a nite inverse semigroup presentation with respect to the action of
S in terms of Y
1
. Then T can be dened by a nite inverse semigroup presentation with
respect to the action of S in terms of Y
2
as well.
Finally, we prove the following useful proposition.
Proposition 4.6.6. Let S be a semigroup dened by the semigroup presentation X|P,
and let T be an inverse semigroup. Assume that T = Act
S
Y |Q. Then we have T =
InvAct
S
Y |Q. In particular, we have that if T is nitely presented as a semigroup with
respect to an action of S, then it is also nitely presented as an inverse semigroup with
respect to the same action of S.
Proof. Let Z = X
. If T = Act
S
Y |Q, then T is dened by the semigroup presentation
Z
Y |Q
A
, where
Q
A
= {
u
p =
v
q | (p = q) Q, (u = v) P, , Z}
{
u
w =
v
w | w (
Z
Y )
+
, (u = v) P, , Z}
{
p =

q | (p = q) Q, Z}.
4.7. EXAMPLES II. 133
Making use of Proposition 4.1.3, we have that T = Inv
Z
Y |Q
A
. Let denote the standard
inverse semigroup relations on FSI(
Z
Y ). To verify that T = InvAct
S
Y |Q, we need to
show that
A
= {
u
p =
v
q | (p = q) , (u = v) P, , Z}
{
p =

q | (p = q) , Z}.
A
. Since T is an inverse semigroup, the relations in
are consequences of relations in Q
A
, and hence the relations in
A
are also consequences
of relations in Q
A
.
4.7 Examples II.
Examples introduced in Section 5 involved inverse semigroups. To deepen the notion of an
inverse semigroup presentation with respect to a semigroup action, we continue working
on these examples. We investigate whether the semilattices Y
, A
and F
are nitely
presented as inverse semigroups with respect to the group actions dened on them.

on top, and let G be the innite cyclic group. Dene the action of G on Y
as in Propo-
sition 4.2.4. Then Y
is nitely presented as an inverse semigroup with respect to the

action of G.
Proof. According to Proposition 4.5.1, Y
is nitely presented as a semigroup with

respect to the action of G. Making use of Proposition 4.6.6, we may deduce that Y
is
nitely presented as an inverse semigroup with respect to the action of G as well.
Proposition 4.7.2. Let F
be the semilattice obtained by adjoining an identity element

to the free semilattice generated by innitely many elements. Let G be the innite cyclic
group. Dene the action of G on F
as in Proposition 4.2.6. Then F
is nitely presented
as an inverse semigroup with respect to the action of G.
Proof. Let F be the free semilattice generated by innitely many elements A =
{. . . , e
1
, e
0
, e
1
, . . .}. Adjoin an identity 1 on top. Then, for each e
i
A
0
, e
i
< 1
holds. Denote by F
the semilattice obtained. Clearly F
is not nitely presented as an

inverse semigroup, since it is not nitely generated. To be more precise, if we let
R = {e
i
e
i
= e
i
, e
i
1 = e
i
, 1 1 = 1 | i Z},
then we have that F
is dened by the inverse semigroup presentation InvA, 1|R. Let

G = g be the innite cyclic group. To make notation clear, we let X = {g, g
1
, 1} and
let Z = X
. Let
P = {gg
1
= g
1
g = 1, 1g = g, 1g
1
= g
1
, 11 = 1}.
G = X|P.
Dene the action of G on F
as in Proposition 4.2.6. As we veried in Proposition

4.2.6, F
0
= {e
0
, 1} with respect to the action of G. Let
B =
Z
Y
0
(
Z
Y
0
)
1
, and let
Q = {e
0
e
0
= e
0
, e
0
1 = e
0
, 1 1 = 1,
g
1 = 1,
1
e
0
= e
0
,
1
1 = 1}.
Let
Q
R
= {
u
p =
v
q | (p = q) Q , (u = v) P, , Z}
{
u
w =
v
w | w B
+
, (u = v) P, , Z}
{
p =

q | (p = q) Q , Z},
where denotes the standard inverse semigroup relations on FSI(
Z
Y
0
). Consider Y =
InvAct
G
Y
0
|Q. We claim that F
= InvAct
G
Y
0
|Q, by giving two inverse semigroup
homomorphism : F
Y and : Y F

We rst show that the map e
n

g
n
e
0
, 1 1 induces an inverse semigroup homomor-
phism : F
Y by showing that relations in R are mapped onto relations that hold in

Y . As in Proposition 4.5.3, we have that for all k Z, the relation
g
k
1 = 1
R
.
The following three cases have to be considered.
(i) e
i
e
i
=
g
i
e
0
g
i
e
0
=
g
i
e
0
= e
i
= (e
i
e
i
).
(ii) e
i
1 =
g
i
e
0
1 =
g
i
e
0
g
i
1 =
g
i
e
0
= e
0
= (e
0
1).
(iii) 1 1 = 1 1 = 1 = 1 = (1 1).
We may hence deduce that every relation in R is mapped onto a relation that holds in Y .
Before we dene a homomorphism from Y to F
, we introduce the following notation.

Let Z, and assume that the group reduced word obtained from is g
n
. Then we let
|||| = n.
We verify that the map
1
e
0
. . .
n
e
0
e
||
1
||
. . . e
||n||
,

1 1
induces a homomorphism : Y F
, by showing that relations in Q

R
are mapped onto
relations that hold in F
. It is immediate by the denition of that relations of Q are

mapped onto relations that hold in F
.
1
u and
2
v. Since (u = v) P, we
have that ||u|| = ||v||, and it follows that ||
1
|| = ||
2
||. Consider the relation e
0
e
0
= e
0
in Q. Then,
(
1
(e
0
e
0
)) = e
||
1
||
e
||
1
||
= e
||
1
||
= e
||
2
||
= (
2
e
0
)
holds. Moreover, we have that
(
(e
0
e
0
)) = e
||||
e
||||
= e
||||
= (
e
0
)
holds. It can be proved similarly, that for any relations (p = q) Q ,
(
1
p) = (
2
q) and (
p) = (
q)
holds.
That and are inverse to each other follows from the following observations. On one
hand
e
n
= (
g
n
e
0
) = e
n
and 1 = 1.
On the other hand
(
e
0
) = e
||||
=
g
||||
e
0
=

e
0
and 1 = 1.
Proposition 4.7.3. Let A
be the semilattice obtained by adjoining an identity and a

zero element to an innite antichain. Let G be the innite cyclic group. Dene the action
of G on A
as in Proposition 4.2.5. Then A
is not nitely presented as an inverse

semigroup with respect to the action of G.
Proof. Let A be the innite antichain
. . . , e
1
, e
0
, e
1
, . . . .
i
A, 0 < e
i
< 1
holds. Let A
denote the semilattice obtained. Clearly A
is not nitely presented as

an inverse semigroup. To be more precise, if we let
R = {e
i
e
j
= 0, e
i
1 = e
i
, e
i
0 = 0,
e
i
e
i
= e
i
, 1 1 = 1, 0 0 = 0 | i, j Z, i = j},
then we have that A
is dened by the inverse semigroup presentation InvA
|R. Let
G = g be the innite cyclic group. To make notation clear, we let X = {g, g
1
, 1} and
let Z = X
. Let
P = {gg
1
= g
1
g = 1, 1g = g, 1g
1
= g
1
, 11 = 1}.
G = X|P.
Dene the action of G on A

A
is nitely generated as a semigroup and hence as an inverse semigroup by Y

0
= {e
0
, 1}
with respect to the action of G.
Let B =
Z
Y
0
(
Z
Y
0
)
1
and let denote the standard inverse semigroup relations on
FSI(
Z
Y
0
). Assume that A
is nitely presented as an inverse semigroup with respect to

the action of G. Then, according to Proposition 4.6.5, there exists a nite set of relations
Q on B
+
such that A
= InvAct
G
Y
0
Q
R
= {
u
p =
v
q | (p = q) Q , (u = v) P, , Z}
{
u
w =
v
w | w B
+
, (u = v) P, , Z}
{
p =

q | (p = q) Q , Z}.
and Q
A
= Q
R
\ , then A
= Inv
Z
Y
0
|Q
A
. Without loss of generality we may assume
that the following set of relations
Q = {e
0
e
0
= e
0
, e
0
1 = e
0
, 1 1 = 1,
g
1 =
1
1 = 1}
is a subset of Q. For further use, we introduce the following notations. We let
Q
A
= {
u
p =
v
q | (p = q)

Q, (u = v) P, , Z}
{
p =

q | (p = q)

Q, Z}
and
A
= {
u
p =
v
q | (p = q) , (u = v) P, , Z}
{
p =

q | (p = q) , Z}.
It is immediate that for all Z,

e
0

e
0
=

e
0
is a relation in

Q
A
, and hence by
applying relations in we obtain that for all Z,
e
0
= (
e
0
)
1
.
In particular, every relation in is a consequence of relations in
P = {
a =

a,

a
b =

b
a | a, b Y
0
, , Z}.
We let
P
A
= {
u
p =
v
q | (p = q) P, (u = v) P, , Z}
{
p =

q | (p = q) P, Z}.
In a similar way as claimed in Proposition 4.5.3, we have that for any k Z
g
k
1 = 1
R
. Having these observations in mind, we assume that
for all relations
p a
1
. . . a
n
= b
1
. . . b
m
q
in Q\

Q, the following conditions hold:
(i) a
j
, b
i

Z
Y
0
for all 1 j n, 1 i m;
(ii) a
j
=

e
0
for all 1 j n, where is a group reduced word over Z;
(iii) b
i
=

e
0
for all 1 i m, where is a group reduced word over Z;
(iv) there is no subword of p and of q of the form

e
0
e
0
.
With these conditions, we also have that n = 1 if and only if m = 1, in which case our
relation p = q reads

e
0
=

e
0
, and hence is redundant. We thus assume, that for all
relations in Q\

Q:
p a
1
. . . a
n
= b
1
. . . b
m
q,
n, m 2 also holds.
We recall that for all Z, if = g
k
in G, then we let |||| = k. We need the
following notion, to verify that A
is not nitely presented as an inverse semigroup with

respect to the action of G. Consider a word
w
1
y
1
. . .
n
y
n
over
Z
Y , and assume
i
,
j
, (1 i, j n) are such that ||
i
|| ||
k
|| ||
j
|| for all
1 k n. Dene
d(w) = ||
j
|| ||
i
||,
and say that d(w) is the distance in w. Let
I = { Z |

e
0
k
that g
i
, g
j
I, and if g
k
I, then j k i. Let d = i j. It is straightforward that the
relation
p
g
i+1
e
0
g
j1
e
0
=
g
i+2
e
0
g
j2
e
0
q
holds in A
, and that
d(p) = d(q).
To show that A
= InvAct
G
Y
0
| Q does not hold, we need to verify that the above
relation is not a consequence of relations in (Q
R
\
A
) P
A
. Assume that there exists a
nite sequence of words
p u
0
, u
1
, . . . , u
m
q,
so that u
t+1
is obtained from u
t
applying a relation in (Q
R
\
A
) P
A
.
1
can be obtained from u
0
.
(a) It is immediate that we cannot apply any of the relation in Q, since
g
i+1
e
0
and
g
j1
e
0
do not occur in any of the relations in Q.
(b) Applying a relation in

Q
A
\

Q, the following words can be obtained:
(i) u
1

e
0
e
0
g
j1
e
0
, where = g
i+1
in G;
(ii) u
1

g
i+1
e
0
e
0
e
0
, where = g
j1
in G;
(iii) u
1

e
0
1
g
j1
e
0
, where = g
i+1
in G;
(iv) u
1

g
i+1
e
0
e
0
1, where = g
j1
in G;
We note that with the application of relations in

Q
A
\
Q, the distance in the obtained

word u
1
equals to the distance in u
0
.
(c) When applying a relation in P
A
, we may have the following situations occur:
(i) u
1

e
0
e
0
, where = g
i+1
and = g
j1
in G;
(ii) u
1

e
0
e
0
g
j1
e
0
, where = g
i+1
in G;
(iii) u
1

g
i+1
e
0
e
0
e
0
, where = g
j1
in G.
We note that in all three cases, d(u
1
) = d(u
0
).
(d) With the application of a relation in Q
R
\(Q

Q
A
P
A
R
A
), the we obtain a word
of the form
u
1

g
e
0
e
0
,
where = g
i+1
, = g
j1
in G. That relations in the set
Q
A
= {
u
p =
v
q | (p = q) Q\

Q, (u = v) P, , Z}
{
p =

q | (p = q) Q\

Q, Z \ {}}.
cannot be applied to u
0
follows from the following remarks. Assume that a relation
of Q
A
can be applied to u
0
. Then there exists a word g
m
Z and a relation relation
p = q in Q\

Q, such that p
g
l
e
0
g
k
e
0
and
g
i+1
e
0
g
j1
e
0

g
m+l
e
0
g
m+k
e
0
.
A
cannot be applied
to the word u
0
. To highlight the key point of the fourth case, we note that applying
a relation in Q
R
\ (Q

Q
A

A
P
A
) results a word u
1
so that d(u
1
) = d(u
0
).
Thus we may deduce, that u
1
is a word so that d(u
1
) = d(u
0
). We can make the same
observation, if we take a closer look how the word u
2
can be obtained from u
1
. With an
inductive argument we thus arrive to the conclusion, that
d(p) = d(u
0
) = d(u
1
) = . . . = d(u
m
) = d(r)
leading to a contradiction. We may hence deduce that A
is not nitely presented as an

inverse semigroup with respect to the action of G.
4.8 Matched and conjugate words
We introduce the concepts of matched and conjugate words inspired by the proof of Propo-
sition 4.3.1. The new notions and related notations will be frequently used in the next two
sections. As in the previous sections, if we say that a semigroup S acts on a semigroup T,
then it will be understood that S acts on T on the left by endomorphisms.
Matched words
We rst x the notation for this section. Let Y be a semilattice and G be a group acting
on Y . Assume that the semidirect product S = Y G is generated as an inverse semigroup
by the nite set A. For simplicity we suppose that A S, and we let B = A A
1
. Let
X = {g G | (e, g) A for some e Y }
and let
Y
0
= {e Y | (e, g) B for some g X X
1
}.
By Proposition 4.3.1 we may assume that Y
m
Y
0
, where Y
m
denotes the set of maximal
elements of Y . Let Z = X {1} and dene the following map:
n : Y
m
(Z Z
1
)
+
B
+
; (f
1
, g
1
g
2
. . . g
k
) (f
1
, g
1
)(f
2
, g
2
) . . . (f
k
, g
k
),
where
g
j
f
j+1
= f
j
, (1 j +1 k). By Lemma 4.2.1, f
j
Y
m
for all 1 j k. Moreover,
f
1
=
g
1
f
2
=
g
1
(
g
2
f
3
) = . . . =
g
1
(
g
2
(. . . (
g
k1
f
k
) . . .)) =
g
1
...g
k1
f
k
(4.15)
holds, and thus the word n(f
1
, g
1
g
2
. . . g
k
) represents the element (f
1
, g
1
g
2
. . . g
k
) in S. We
say that the word n(e, g
1
. . . g
k
) is matched to the word n(f, h
1
. . . h
m
), if e =
g
1
...g
k
f. We
introduce two lemmas regarding matched words.
Lemma 4.8.1. The word n(e, g
1
. . . g
k
1
. . . h
m
) if and only
if n(e, g
1
. . . g
k
)n(f, h
1
. . . h
m
) n(e, g
1
. . . g
k
h
1
. . . h
m
).
Proof. () Assume that n(e, g
1
. . . g
k
1
. . . h
m
). Then
we know that
e =
g
1
...g
k
f. (4.16)
On one hand, we have that
n(e, g
1
. . . g
k
)n(f, h
1
. . . h
m
) (e, g
1
) . . . (e
k
, g
k
)(f, h
1
) . . . (f
m
, h
m
),
4.8. MATCHED AND CONJUGATE WORDS 141
where
g
i
e
i+1
= e
i
, (1 i k 1, e
1
= e) and
h
j
f
j+1
= f
j
, (1 j m1, f
1
= f). On
the other hand, we have that
n(e, g
1
. . . g
k
h
1
. . . h
m
) (e, g
1
) . . . (e
k
, g
k
)(l
1
, h
1
) . . . (l
m
, h
m
),
where
g
i
e
i+1
= e
i
, (1 i k 1, e
1
= e),
g
k
l
1
= e
k
and
h
j
l
j+1
= l
j
, (1 j m 1).
Making use of (4.15) we obtain that
e =
g
1
...g
k
l
1
.
Bearing in mind (4.16), we thus obtain that
g
1
...g
k
l
1
=
g
1
...g
k
f
1
holds. It follows that l
j
= f
j
for all 1 j k verifying that
n(e, g
1
. . . g
k
)n(f, h
1
. . . h
m
) n(e, g
1
. . . g
k
h
1
. . . h
m
).
() Assume that
n(e, g
1
. . . g
k
)n(f, h
1
. . . h
m
) n(e, g
1
. . . g
k
h
1
. . . h
m
).
Hence, if
n(e, g
1
. . . g
k
) (e, g
1
) . . . (e
k
, g
k
)
n(f, h
1
. . . h
m
) (f, h
1
) . . . (f
m
, h
m
)
n(e, g
1
. . . g
k
h
1
. . . h
m
) (e, g
1
) . . . (e
k
, g
k
)(l
1
, h
1
) . . . (l
m
, h
m
),
then l
j
= f
j
, (1 j m, f
1
= f) and
e =
g
1
...g
k
l
1
=
g
1
...g
k
f
1
,
proving that the word n(e, g
1
. . . g
k
) is indeed matched to the word n(f, h
1
. . . h
m
).
The following lemma can be proved using similar techniques.
Lemma 4.8.2. Assume that n(e, g
1
. . . g
k
) is matched to n(f, h
1
. . . h
m
). Then
n(e, g
1
. . . g
k
)(f, g) n(e, g
1
. . . g
k
g)
holds for all g Z Z
1
.
The following lemma is immediate from the denitions. Before asserting the lemma,
we recall that by Lemma 4.2.1, if l Y
m
, then for any word g
1
. . . g
k
over Z Z
1
,
g
1
...g
k
l Y
m
.
Lemma 4.8.3. Consider the words u n(e, g
1
. . . g
k
) and v n( e, g
1
k
. . . g
1
1
), where
e =
g
1
...g
k
e. Then the following hold:
(i) u is matched to v;
(ii) v is matched to u;
(iii) uvu = v and vuv = v hold in S.
(iv) If w = n(f, g
1
k
. . . g
1
1
) such that uwu = u and wuw = w hold in S, then w v.
We note that considering u and v as elements of S, v = u
1
. Therefore, if u
n(e, g
1
. . . g
k
), then we let u
1
denote the word n( e, g
1
k
. . . g
1
1
), where e =
g
1
...g
k
e.
Conjugate words of idempotents
We keep the notation introduced so far. Before we introduce the notion of conjugate
words, we prove the following lemma.
Lemma 4.8.4. Let f Y and assume that
f =
g
1
...g
k
l,
where l Y
0
and g
1
, . . . , g
k
ZZ
1
. Let e Y
m
such that f e. Let u n(e, g
1
. . . g
k
).
Then the word
u(l, 1)u
1
represents the element (f, 1) in S.
Proof. Indeed, we have
u (l, 1) u
1
= (e, g
1
. . . g
k
) (l, 1) ( e, g
1
k
. . . g
1
1
)
= (e
g
1
...g
k
l
g
1
...g
k
e, 1)
= (e f e, 1)
= (f, 1).
4.9. FINITE PRESENTABILITY I. 143

Keeping the notation of Lemma 4.8.4, we say that
u (l, 1) u
1
is a conjugate word of the idempotent (
g
1
...g
k
l, 1). Clearly, any idempotent (
g
1
...g
k
l, 1) in
S can have several conjugate words representing it, depending on how many maximal
idempotents there are above it with respect to the natural partial order.
Notation. The set of all conjugate words of (
g
1
...g
k
l, 1) will be denoted by
c(
g
1
...g
k
l).
Clearly, if Y satises the maximum condition, then c(
g
1
...g
k
l) is a nite set for all l Y
0
and w g
1
. . . g
k
(Z Z
1
)
+
.
4.9 Finite presentability I.
The main result of this chapter is
Theorem 4.9.1. Let (Y, ) be a semilattice, and G be a group acting on Y on the left
by automorphisms. The semidirect product S = Y G is nitely presented as an inverse
semigroup if and only if the following conditions hold:
(i) G is nitely presented as a group;
(iii) Y is nitely presented as an inverse semigroup with respect to the action of G.
We prove direct implication in this section. As in the previous sections, if we say that
a group G acts on a semilattice Y , then it will be understood that G acts on Y on the left
by automorphisms. We rst x notation for this and the following section.
Notation
Let Y be a semilattice and G be a group acting on Y . Let S = Y G and let A be a
nite inverse semigroup generating set for S. We assume that A S. We let
and let
Y
0
= {e Y | (e, g) A A
1
for some g X X
1
}.
Because of Proposition 4.3.1 we assume that Y
m
, the nite set of maximal elements of Y ,
is contained in Y
0
. We let Z = XX
1
{1}. Since we are going to work with words over
three dierent alphabets, namely over A A
1
, X X
1
and over B =
Z
Y
0
(
Z
Y
0
)
1
,
we agree that words over A A
1
and X X
1
will be written in the usual way, i.e. we
write the letters consecutively. From the context it will be clear over which alphabet the
word being considered is taken. For the sake of legibility words over B will be written in
the following way:
a
1
. . . a
k
.
As in Section 6, we consider the semigroup B
+
as a semigroup on which Z acts, where the
action is dened by the rule (4.14) introduced in Section 6. As introduced in the previous
section, we dene
n : Y
m
(Z Z
1
)
+
B
+
; (f
1
, g
1
g
2
. . . g
k
) (f
1
, g
1
)(f
2
, g
2
) . . . (f
k
, g
k
),
where
g
j
f
j+1
= f
j
, (1 j + 1 k).
We also adopt the following notation. For a word w (e
1
, g
1
) . . . (e
k
, g
k
) (AA
1
)
+
,
we let
w e
1
g
1
e
2
. . .
g
1
...g
k1
e
k
and w g
1
. . . g
k
.
Note that the element of S represented by w is ( w, w). It is immediate that if w uv,
then
w u
b u
v.
We recall that FSI(A) denotes the free semigroup with involution on the set A. We rst
prove
Lemma 4.9.2. Let (Y, ) be a semilattice and G be a group acting on Y . Assume that
the semidirect product S = Y G is generated by A. Then for any w FSI(A),
ww
1
= w
holds in Y .
Proof. We prove the lemma by induction on the length of w. If |w| = 1, then w
(e, g) A A
1
and so (e, g)
1
= (
g
1
e, g
1
) by Proposition 4.2.7. To simplify notation,
let u ww
1
. Then
u e
g
(
g
1
e) = e e = e
verifying that u = w, when |w| = 1. Assume that
ww
1
= w
holds in Y for all w FSI(A), whose length is less then m, (m > 1). Let w FSI(A) so
that the length of w is m. Then w (e, g)w
1
, where w
1
is a word of length m 1 and
(e, g) A A
1
. To simplify notation, let u ww
1
and v w
1
w
1
1
. Then
u e
g
v
g v
(
g
1
e)
= e
g
w
1
g v
(
g
1
e) by the induction hypothesis
= e
g
w
1
e since g vg
1
= 1
= e
g
w
1
standard inverse semigroup relation
w
proving that u = w.
The necessary condition
We prove
Proposition 4.9.3. Let (Y, ) be a semilattice, and G be a group acting on Y . If the
semidirect product S = Y G is nitely presented as an inverse semigroup then the
(iii) Y is nitely presented as an inverse semigroup with respect to the action.
Proof. Assume that the semidirect product S = Y G is given by the presentation
S = InvA|P, where A S is a nite set and P is a nite set of relations. As introduced,
we let
and let
Y
0
= {e Y | (e, g) A A
1
for some g G}.
By Proposition 4.3.1 we may assume that Y
m
Y
0
. That condition (ii) holds, follows
by Proposition 4.3.1, and thus it remains to show that G is nitely presented as a group
and that Y is nitely presented with respect to the action as an inverse semigroup. Let
P = { r = p | (r = p) P}.
We show that
G = GrpX|
P.
We already know by Proposition 4.3.1 that G is generated by X. It is straightforward that
the relations of

P hold in G. We verify that any relation that holds in G is a consequence
of the relations in

P. For this, assume that the relation
g
1
g
2
. . . g
m
= h
1
h
2
. . . h
k
,
(g
i
, h
j
X X
1
) holds in G. Let f Y
m
. Then
n(f, g
1
g
2
. . . g
m
) = n(f, h
1
h
2
. . . h
k
),
holds in S. Thus, there exists a nite sequence of words
n(f, g
1
g
2
. . . g
m
) q
0
, q
1
, . . . , q
t
n(f, h
1
h
2
. . . h
k
)
where q
j+1
is obtained from q
j
by applying a relation in P or a standard inverse semigroup
relation. In particular we have that:
(i) If q
j+1
is obtained from q
j
by applying a relation in P, then q
j+1
is obtained from
q
j

P.
(ii) If q
j+1
is obtained from q
j
by applying a standard inverse semigroup relation, then
q
j+1
is obtained from q
j
by applying a sequence of standard group relations.
It follows that there exists a nite sequence of words
g
1
g
2
. . . g
m

0
,
1
, . . . ,
l
h
1
h
2
. . . h
k
,
such that
j+1
is obtained from
j

P or a standard group relation,
verifying that G = GrpX|
P.
We let Z = (X X
1
{1})
and
P =

P {xx
1
= x
1
x = 1, 1x = x, 1x
1
= x
1
, 11 = 1 | x X}.
Clearly G = X X
1
{1} | P.
Next, we claim that Y is nitely presented as an inverse semigroup with respect to
the action of G. We already know that Y satises the maximum condition and that Y is
nitely generated as a semigroup by Y
0
with respect to the action. We recall that (
Z
Y
0
)
1
denotes the set of formal inverses of elements of
Z
Y
0
. We let B =
Z
Y
0
(
Z
Y
0
)
1
. We
denote by the set of standard inverse semigroup relations on FSI(
Z
Y
0
). Let
Q = { r = p | (r = p) P} {e = f e | e, f Y
0
, e f}.
We show that
Y = InvAct
G
Y
0
|Q
by proving that Y = Inv
Z
Y
0
|Q
A
, where Q
A
= Q
R
\ and
Q
R
= Q {
u
p =
v
q | (p = q) Q , (u = v) P, , Z}
{
u
w =
v
w | w B
+
, (u = v) P, , Z}
{
p =

q | (p = q) Q , Z}.
It is clear that every relation of Q and hence of Q
R
holds in Y . Furthermore, for all Z,
the relation

e
e =

e is an element of Q
R
, and so it follows that
(
e)
1
=

e. (4.17)
Making use of this observation, we may assume that any relation that holds in Y is of the
form:
1
e
1
2
e
2
. . .
m
e
m
=

1
f
1
2
f
2
. . .
k
f
k
, (4.18)
where e
i
, f
j
Y
0
and
i
,
j
Z. Assume now that the above relation holds in Y . Let e
1
be a maximal element of Y , such that e
1
e
1
. Since Y satises the maximum condition,
such an element exists. In particular, the relation
1
e
1
1
e
1
=

1
( e
1
e
1
) =

1
e
1
,
is an element of Q
R
. Moreover,

1
e
1
Y
m
Y
0
by Lemma 4.2.1. Let e
0
=

1
e
1
. Then we
obtain
1
e
1
2
e
2
. . .
m
e
m
= e
0
1
e
1
2
e
2
. . .
m
e
m
.
With a similar argument, we also obtain that
1
f
1
2
f
2
. . .
k
f
k
= f
0
1
f
1
2
f
2
. . .
k
f
k
holds for some f
0
Y
m
Y
0
. We verify that (4.18) is a consequence of relations in Q
R
by
verifying that
e
0
1
e
1
2
e
2
. . .
m
e
m
= f
0
1
f
1
2
f
2
. . .
k
f
k
(4.19)
is a consequence of the relations in Q
R
in three steps.
Step 1. Consider the element (e
0
1
e
1
2
e
2
. . .
m
e
m
, 1) of S. Let
j

_
_
_
1
j

j+1
1 j m1
j+1
j = 0
1
j
j = m.
Dene words w
1,0
, . . . , w
1,m
over A A
1
w
1,j

_
n(e
j
,
j
) if e
j
Y
m
(e
j
, 1)n( e
j
,
j
) otherwise,
where e
j
Y
m
is such that e
j
e
j
. Clearly, the word
w
1
w
1,0
w
1,1
. . . w
1,m
represents (e
0

1
e
1

2
e
2
. . .
m
e
m
, 1). On the other hand, by the denition of the
words w
1,0
, . . . , w
1,m
we obtain that
w
1,j
=
_
e
j
. . . e
j
if e
j
Y
m
e
j
e
j
. . . e
j
otherwise,
and hence e
j
= w
1,j
R
. It follows that
e
0
1
e
1
2
e
2
. . .
m
e
m
= w
1
(4.20)
is also a consequence of the relations in Q
R
.
Step 2. We consider the element (f
0
1
f
1
2
f
2
. . .
k
f
k
, 1) of S and repeat the rst
step. We let
j

_
_
_
1
j

j+1
1 j k 1
j+1
j = 0
1
j
j = k.
Dene words w
2,0
, . . . , w
2,k
over A A
1
w
2,j

_
n(f
j
,
j
) if f
j
Y
m
(f
j
, 1)n(

f
j
,
j
) otherwise,
where

f
j
Y
m
is such that f
j

f
j
. The word
w
2
w
2,0
w
2,1
. . . w
2,k
represents (f
0

1
f
1

2
f
2
. . .
k
f
k
, 1). On the other hand, by the denition of the
words w
2,0
, . . . , w
2,k
we obtain that
w
2,j
=
_
f
j
. . . f
j
if f
j
Y
m
f
j

f
j
. . .

f
j
otherwise,
and hence f
j
= w
2,j
R
. It follows that
w
2
= f
0
1
f
1
2
f
2
. . .
k
f
k
(4.21)
is also a consequence of the relations in Q
R
.
Step 3. Since by (4.20)
e
0
1
e
1
2
e
2
. . .
m
e
m
= w
1
and by (4.21)
w
2
= f
0
1
f
1
2
f
2
. . .
k
f
k
hold, to show that
e
0
1
e
1
2
e
2
. . .
m
e
m
= f
0
1
f
1
2
f
2
. . .
k
f
k
R
, it is enough to verify that
w
1
= w
2
R
. Since
e
0
1
e
1
2
e
2
. . .
m
e
m
= f
0
1
f
1
2
f
2
. . .
k
f
k
holds in Y , the relation w
1
= w
2
holds in S, and hence there exists a nite sequence of
words
w
1
q
0
, q
1
, . . . , q
t
w
2
such that q
j+1
is obtained from q
j
by applying a relation in P or a standard inverse
semigroup relation. We consider the following three cases:
(1) If q
j+1
is obtained from q
j
by applying a relation in P, then we may write q
j

t
1
st
2
, q
j+1
t
1
zt
2
, where (s = z) P. It follows that ( s = z)

P and ( s = z)

P.
In particular

t
1
s =

t
1
z holds in G and we obtain that
q
j

t
1
t
1
s
c
t
1
s
t
2
=

t
1
t
1
z
c
t
1
s
t
2
=

t
1
t
1
z
c
t
1
z
t
2
q
j+1
verifying that q
j+1
is obtained from q
j
by applying relations in Q
R
.
(2) If q
j+1
is obtained from q
j
by applying a relation of the form ww
1
w = w, then we
may write q
j
t
1
ww
1
wt
2
, q
j+1
t
1
wt
2
. Let u ww
1
, and v ww
1
w. Clearly
v = w holds in G and we have that
q
j

t
1
t
1
u
c
t
1
u
w
c
t
1
v
t
2
=

t
1
t
1
w
c
t
1
u
w
c
t
1
v
t
2
since u = w by Lemma 4.9.2
=

t
1
t
1
w
b
t
1
w
c
t
1
v
t
2
since

t
1
u =
t
1
in G
=

t
1
t
1
w
c
t
1
v
t
2
by applying a relation in Q
R
=

t
1
t
1
w
d
t
1
w
t
2
since w = v in G
q
j+1
proving that q
j
= q
j+1
R
.
(3) If q
j+1
is obtained from q
j
by applying a relation of the form w
1
w
1
1
w
2
w
1
2
=
w
2
w
1
2
w
1
w
1
1
, then we may write
q
j
t
1
w
1
w
1
1
w
2
w
1
2
t
2
, q
j+1
t
1
w
2
w
1
2
w
1
w
1
1
t
2
.
Let u w
1
w
1
1
, v w
2
w
1
2
. Then
q
j

t
1
t
1
u
c
t
1
u
v
d
t
1
uv
t
2
=

t
1
t
1
u
t
1
v
t
1
t
2
since u = uv = 1 in G
=

t
1
t
1
v
t
1
u
t
1
t
2
standard inverse semigroup relation
=

t
1
t
1
v
c
t
1
v
u
d
t
1
vu
t
2
since v = vu = 1 in G
q
j+1
proving that q
j
= q
j+1
R
.
It follows that a nite sequence of words can be given w
1
u
0
, u
1
, . . . , u
l
w
2
such
that u
j+1
is obtained from u
j
R
and we may deduce that
Y = InvAct
G
Y
0
|Q.
4.10 Finite presentability II.
We keep the notation introduced at the beginning of Section 9. In addition, we recall that
for any element
g
1
...g
k
e Y
c(
g
1
...g
k
e)
4.10. FINITE PRESENTABILITY II. 151
denotes the set of conjugate words of the idempotent (
g
1
...g
k
e, 1) E(S). The main purpose
of this section is to prove
Proposition 4.10.1. Let (Y, ) be a semilattice, and G be a group acting on Y . Assume
that the following conditions hold:
(iii) Y is nitely presented as an inverse semigroup with respect to the action.
Then the semidirect product S = Y G is nitely presented as an inverse semigroup.
Since the proof of Proposition 4.10.1 is rather long, we rst summarize the main steps
of the proof.
1. We give an inverse semigroup generating set A for S = Y G and dene a set of
relations R on (A A
1
)
+
that hold in S.
2. In Proposition 4.10.2, we give a normal form for the elements of S. Proposition
4.10.2 is proved with the help of two technical lemmas: Lemma 4.10.3 and Lemma
4.10.4.
3. Making use of Proposition 4.10.2, we prove that any relation w
1
= w
2
that holds in
S is a consequence of relations in R.
Proof. Let G be a group acting on a semilattice Y and assume that G and Y satisfy
conditions (i) (iii). More precisely, let G = GrpX|P, where X is a nite set and P is a
nite set of relations. Let Y be a semilattice satisfying the maximum condition and let Y
m
denote the set of maximal elements of Y . Assume that Y = InvAct
G
Y
0
|Q, where Y
0
is
a nite subset of Y and Q is a nite set of relations. By Proposition 4.6.5 we may assume
that Y
m
Y
0
. Consider the semidirect product S = Y G. Making use of Corollary 4.3.2,
S is nitely generated as an inverse semigroup by
A = (Y
0
{1}) (Y
m
X).
By Proposition 4.2.7, we have
A
1
= (Y
0
{1}) (Y
m
X
1
).
Dene the following relations on (A A
1
)
+
:
(R1) (e, 1)(f, g) = (f, 1)(e, g), where e, f Y
m
;
(R2) (e, g) = (e, 1)(e, g), where e Y
m
;
(R3) (e, g) = (e, g)(f, 1), where f Y
0
,
g
f e Y
m
;
(R4) (e, g)(f, 1) = (

f, 1)(e, g)(f, 1), where f Y
0
, e,

f Y
m
,
g
f

f;
(R5) (e, 1)(f, g) = (e, 1)(f, g)( e, 1), where e, e, f Y
m
,
g
e = e;
(R6) (e, g)(f, h) = (e, 1)(

f, g)(f, h), where e, f,

f Y
m
,
g
f =

f;
(R7) (e, g)(f, h) = (e, g)(f, 1)( e, h), where e, e, f Y
m
,
g
e = e.
We denote by R the relations obtained.
For each relation g
1
. . . g
m
= h
1
. . . h
k
in P and e Y
m
consider the following relation
n(e, g
1
. . . g
m
) = n(e, h
1
. . . h
k
)
and denote the set of relations obtained by R
G
. Since Y
m
and P are nite sets, we have
that R
G
is a nite set of relations. For each (e, g) Y
m
X consider the relations
(e, g)(f, g
1
) = (e, 1) and (f, g
1
)(e, g) = (f, 1),
where f is the maximal element of Y for which
g
f = e. Denote the set of relations obtained
by R
M
.
For each relation
1
e
1
. . .
m
e
m
=

1
f
1
. . .
k
f
k
in Q we dene the following set of relations
{c
11
. . . c
1m
= c
21
. . . c
2k
| c
1j
c(
j
e
j
), c
2j
c(
j
f
j
)}.
Since Y satises the maximum condition, each of the sets c(
j
e
j
) and c(
j
f
j
) is nite and
hence
{c
11
. . . c
1m
= c
21
. . . c
2k
| c
1j
c(
j
e
j
), c
2j
c(
j
f
j
)}.
is a nite set. Denote the set of all relations obtained in this way by R
Y
.
For each e, f Y
0
such that e f consider the relation
(e, 1) = (f, 1)(e, 1)
and denote the set of all relations of this form by R
L
.
Let
R = R R
G
R
M
R
Y
R
L
.
Our aim is to show that S = InvA|R. Clearly R is a nite set of relations and all
relations in R hold in S. According to Proposition 4.1.4 we need to verify that any
relation w
1
= w
2
that holds in S is a consequence of relations in R and of standard
inverse semigroup relations. In order to prove this, we need the following proposition.
Proposition 4.10.2. Let w (e
1
, g
1
) . . . (e
k1
, g
k1
)(e
k
, 1) (AA
1
)
+
such that |w|
2. Then the relation
w = (e
1
, 1)c
2
. . . c
k1
v
where c
j
c(
g
1
...g
j
e
j+1
), v n(f, g
1
. . . g
k1
) for some f Y
m
and c
k1
v(e
k
, 1)v
1
, is
a consequence of relations in R.
We need the following two technical lemmas:
Lemma 4.10.3. Let w n(e, g
1
g
2
. . . g
k
)(f, 1), where f Y
0
. Then there exists l Y
m
such that the relation
w = (e, 1)n(l, g
1
g
2
. . . g
k
)
is a consequence of the relations in R. Moreover, if we let u n(l, g
1
g
2
. . . g
k
), then
u(f, 1)u
1
c(
g
1
g
2
...g
k
f).
Proof. Let

f Y
m
so that f

f. Let l
k
=
g
k
f. Let l
1
, l
2
, . . . , l
k1
Y
m
such that
l
j
=
g
j
l
j+1
holds. Such elements exist by Lemma 4.2.1. Since f

f, we also obtain that
g
1
...g
k
f l
1
. (4.22)
Assume that
n(e, g
1
. . . g
k
) (e, g
1
) . . . (e
k1
, g
k1
)(e
k
, g
k
).
Applying the appropriate relations in R we obtain
w (e, g
1
) . . . (e
k1
, g
k1
)((e
k
, g
k
)(f, 1))
= (e, g
1
) . . . (e
k1
, g
k1
)((l
k
, 1)(e
k
, g
k
)(f, 1)) by (R4)
= (e, g
1
) . . . (l
k1
, 1)(e
k1
, g
k1
)(l
k
, 1)(e
k
, g
k
)(f, 1) by (R4)
.
.
.
.
.
.
= (l
1
, 1)(e, g
1
)(l
2
, 1) . . . (l
k1
, 1)(e
k1
, g
k1
)(l
k
, 1)(e
k
, g
k
)(f, 1) by (R4)
= (e, 1)(l
1
, g
1
)(e
2
, 1) . . . (e
k1
, 1)(l
k1
, g
k1
)(e
k
, 1)((l
k
, g
k
)(f, 1)) by (R1)
= (e, 1)(l
1
, g
1
)(e
2
, 1) . . . ((e
k1
, 1)(l
k1
, g
k1
)(e
k
, 1))(l
k
, g
k
) by (R3)
= (e, 1)(l
1
, g
1
)(e
2
, 1) . . . (e
k1
, 1)(l
k1
, g
k1
)(l
k
, g
k
) by (R5)
.
.
.
.
.
.
= (e, 1)(l
1
, g
1
) . . . (l
k1
, g
k1
)(l
k
, g
k
) by (R5)
(e, 1)(n(l
1
, g
1
. . . g
k
)).
It is immediate from (4.22) that if u n(l
1
, g
1
g
2
. . . g
k
), then u(f, 1)u
1
c(
g
1
g
2
...g
k
f).
Lemma 4.10.4. Let w (f, 1)n(e, g
1
g
2
. . . g
k
), where f Y
m
. Then
w = (e, 1)n(f, g
1
g
2
. . . g
k
)
is a consequence of the relations in R.
Proof. Let f
1
= f and let f
2
, . . . , f
k
Y
m
such that
g
j
f
j+1
= f
j
holds. Assume that
n(e, g
1
g
2
. . . g
k
) (e, g
1
)(e
2
, g
2
) . . . (e
k1
, g
k1
)(e
k
, g
k
).
Then
w (f, 1)(e, g
1
)(e
2
, g
2
) . . . (e
k1
, g
k1
)(e
k
, g
k
)
= (e, 1)((f, g
1
)(e
2
, g
2
)) . . . (e
k1
, g
k1
)(e
k
, g
k
) by (R1)
= (e, 1)(f, g
1
)(e
2
, 1)(f
2
, g
2
) . . . (e
k1
, g
k1
)(e
k
, g
k
) by (R7)
.
.
.
.
.
.
= (e, 1)(f, g
1
)(e
2
, 1)(f
2
, g
2
) . . . ((e
k1
, 1)(f
k1
, g
k1
)(e
k
, 1))(f
k
, g
k
) by (R7)
= (e, 1)(f, g
1
)(e
2
, 1)(f
2
, g
2
) . . . ((e
k1
, 1)(f
k1
, g
k1
))(f
k
, g
k
) by (R5)
.
.
.
.
.
.
= (e, 1)(f, g
1
)(f
2
, g
2
) . . . (f
k1
, g
k1
)(f
k
, g
k
) by (R5)
(e, 1)n(f, g
1
. . . g
k
),
verifying the lemma.
Proof of Proposition 4.10.2. We proceed by induction on the length of w.
If |w| = 2, then w (f, g)(e, 1), where (f, g) A A
1
. We assume that (f, g)
Y
m
(X X
1
). The case when (f, g) A A
1
is such that g = 1 can be veried
similarly. Let e Y
m
such that
g
e e and let
g
1
e = l. By Lemma 4.2.1, l Y
m
.
Moreover,
w (f, g)(e, 1)
= (f, 1)((f, g)(e, 1)) by (R2)
= (f, 1)(( e, 1)(f, g))(e, 1) by (R4)
= (f, 1)(f, 1)( e, g)(e, 1) by (R1)
= (f, 1)( e, g)(e, 1) by R
L
= (f, 1)( e, g)((l, g
1
)( e, g)(e, 1)) standard inverse sg. relation
= (f, 1)( e, g)(e, 1)(l, g
1
)( e, g) standard inverse sg. relation
Clearly, ( e, g)(e, 1)(l, g
1
) c(
g
e), verifying that the proposition holds for all words of the
form (f, g)(e, 1).
Assume that the proposition is true for all words over AA
1
that satisfy the condition
stated in the proposition and whose length is less then m, where m 2. We consider the
following two cases:
Case 1. w (e
1
, g
1
) . . . (e
k1
, 1)(e
k
, 1);
Case 2. w (e
1
, g
1
) . . . (e
k1
, g
k1
)(e
k
, 1), where g
k1
= 1.
Case 1. The inductive hypothesis can be applied to the subword
(e
1
, g
1
) . . . (e
k1
, 1),
and hence we obtain that
w = (e
1
, 1)c
2
. . . c
k2
u(e
k
, 1),
where c
j
c(
g
1
...g
j
e
j+1
), u n(f, g
1
. . . g
k2
) for some f Y
m
and c
k2
u(e
k1
, 1)u
1
.
By Lemma 4.8.3, we know that u
1
is matched to u, and so applying a relation of type
(R3) we obtain that u
1
(f, 1) = u
1
. In particular, we have
c
k2
(f, 1) = c
k2
. (4.23)
On the other hand, by Lemma 4.10.3, there exists l Y
m
so that
n(f, g
1
. . . g
k2
)(e
k
, 1) = (f, 1)n(l, g
1
. . . g
k2
) (4.24)
holds, and if we let v denote the word n(l, g
1
. . . g
k2
), then
c
k1
v(e
k
, 1)v
1
c(
g
1
...g
k2
e
k
). (4.25)
Summarizing the above facts, we obtain that
w = (e
1
, 1)c
2
. . . c
k2
u(e
k
, 1)
= (e
1
, 1)c
2
. . . c
k2
u(e
k
, 1)(e
k
, 1)
= (e
1
, 1)c
2
. . . c
k2
(f, 1)v(e
k
, 1) by (4.24)
= (e
1
, 1)c
2
. . . c
k2
v(e
k
, 1) by (4.23)
= (e
1
, 1)c
2
. . . c
k2
vv
1
v(e
k
, 1) standard inverse sg. relation
= (e
1
, 1)c
2
. . . c
k2
v(e
k
, 1)v
1
v standard inverse sg. relation
= (e
1
, 1)c
2
. . . c
k2
c
k1
v
verifying that the proposition indeed holds in the rst case.
Case 2. If w (e
1
, g
1
) . . . (e
k1
, g
k1
)(e
k
, 1) where g
k1
= 1, then e
k1
Y
m
, since all
generators of the form (e, g), g = 1 have rst component in Y
m
. Applying (R2) we obtain
that
w (e
1
, g
1
) . . . (e
k1
, g
k1
)(e
k
, 1)
= (e
1
, g
1
) . . . (e
k1
, 1)(e
k1
, g
k1
)(e
k
, 1).
The inductive hypothesis can be applied to the subword
(e
1
, g
1
) . . . (e
k1
, 1),
and so we obtain
w = (e
1
, 1)c
2
. . . c
k2
u(e
k1
, g
k1
)(e
k
, 1),
where c
j
c(
g
1
...g
j
e
j+1
), u n(f, g
1
. . . g
k2
) for some f Y
m
and c
k2
u(e
k1
, 1)u
1
.
To unravel the information, we have that
u(e
k1
, 1)u
1
= (
g
1
...g
k2
e
k1
, 1),
and so
(f
g
1
...g
k2
e
k1
, 1) = (
g
1
...g
k2
e
k1
, 1) (4.26)
holds. Taking into account that e
k1
Y
m
, we thus obtain from (4.26) that f =
g
1
...g
k2
e
k1
. In particular, we have that u n(f, g
1
. . . g
k2
) is matched to (e
k1
, g
k1
),
and so by Lemma 4.8.1 we obtain
u(e
k1
, g
k1
) = n(f, g
1
. . . g
k2
g
k1
). (4.27)
By Lemma 4.8.3, we know that u
1
is matched to u, and so applying a relation of type
(R3), it follows that u
1
(f, 1) = u
1
. In particular, we have that
c
k2
(f, 1) = c
k2
. (4.28)
To summarize the above observations,
w = (e
1
, 1)c
2
. . . c
k2
u(e
k1
, g
k1
)(e
k
, 1)
= (e
1
, 1)c
2
. . . c
k2
n(f, g
1
. . . g
k2
g
k1
)(e
k
, 1).
We need one more observation to nish the argument. Making use of Lemma 4.10.3, we
have that there exists l Y
m
so that
n(f, g
1
. . . g
k2
g
k1
)(e
k
, 1) = (f, 1)n(l, g
1
. . . g
k2
g
k1
) (4.29)
holds and if we let v n(l, g
1
. . . g
k2
g
k1
), then
c
k1
v(e
k
, 1)v
1
c(
g
1
...g
k1
e
k
). (4.30)
Summarizing the above facts:
w = (e
1
, 1)c
2
. . . c
k2
n(f, g
1
. . . g
k2
g
k1
)(e
k
, 1)
= (e
1
, 1)c
2
. . . c
k2
n(f, g
1
. . . g
k2
g
k1
)(e
k
, 1)(e
k
, 1)
= (e
1
, 1)c
2
. . . c
k2
(f, 1)v(e
k
, 1) by (4.29)
= (e
1
, 1)c
2
. . . c
k2
v(e
k
, 1) by (4.28)
= (e
1
, 1)c
2
. . . c
k2
vv
1
v(e
k
, 1) standard inverse sg. relation
= (e
1
, 1)c
2
. . . c
k2
v(e
k
, 1)v
1
v standard inverse sg. relation
= (e
1
, 1)c
2
. . . c
k2
c
k1
v
proving Proposition 4.10.2.
We are now ready to prove that S = InvA|R. Recall that G = GrpX|P and
Y = InvAct
G
Y
0
|Q.
Proof of Proposition 4.10.1. Assume that the relation w
1
= w
2
holds in S. By applying a
relation of type (R3), we may assume that
w
1
(e
1
, g
1
) . . . (e
m
, 1) and that w
2
(f
1
, h
1
) . . . (f
k
, 1).
By Proposition 4.10.2, we have that
w
1
= (e
1
, 1)c
2
. . . c
m1
u, (4.31)
where c
j
c(
g
1
...g
j
e
j+1
), u n(e, g
1
. . . g
m1
) for some e Y
m
and c
m1
u(e
m
, 1)u
1
.
We write
(w
1
)
Y
(e
1
, 1)c
2
. . . c
m1
and recall the notation introduced in Section 9:
w
1
e
1
g
1
e
2
. . .
g
1
...g
m1
e
m
and w
1
g
1
. . . g
m1
.
Similarly we have that
w
2
= (f
1
, 1)d
2
. . . d
k1
v,
where d
j
c(
h
1
...h
j
f
j+1
), v n(f, h
1
. . . h
k1
) for some f Y
m
and d
k1
v(f
k
, 1)v
1
.
We write
(w
2
)
Y
(f
1
, 1)d
2
. . . d
k1
,
w
2
f
1
h
1
f
2
. . .
h
1
...h
k1
f
k
and w
2
h
1
. . . h
k1
.
We prove in four steps that w
1
= w
2
is a consequence of relations in R and of standard
inverse semigroup relations.
Step 1. Since w
1
= w
2
in S, we have that w
1
= w
2
holds in G, and hence there exists
a nite sequence of words
w
1

1
,
2
, . . . ,
t
w
2
such that
j+1
is obtained from
j
by applying a relation in P or a standard group relation.
It follows that there exists a nite sequence of words
n(e, g
1
. . . g
m1
)
1
, . . . ,
t
n(e, h
1
. . . h
k1
)
such that
j+1
is obtained from
j
by applying a relation in R
G
R
M
. In particular we
have a nite sequence of words
(w
1
)
Y
(n(e, g
1
. . . g
m1
)) (w
1
)
Y

1
, . . . , (w
1
)
Y

t
(w
1
)
Y
(n(e, h
1
. . . h
k1
)). (4.32)
Step 2. Since w
1
= w
2
in S, we have that w
1
= w
2
holds in Y , and hence there exists a
nite sequence of words
w
1

1
,
2
, . . .
q
w
2
such that
j+1
is obtained from
j
by applying a relation in Q
R
. It follows that there
exists a nite sequence
(w
1
)
Y

1
,
2
, . . . ,
q
(w
2
)
Y
such that
j+1
is obtained from
j
by applying a relation in R
Y
R
L
R
G
R
M
or a
standard inverse semigroup relation. In particular, we have a nite sequence of words
(w
1
)
Y
(n(e, h
1
. . . h
k1
))
1
(n(e, h
1
. . . h
k1
)), . . .
. . . ,
q
(n(e, h
1
. . . h
k1
)) (w
2
)
Y
(n(e, h
1
. . . h
k1
)).
Bearing in mind (4.32), we thus obtain that
(w
1
)
Y
(n(e, g
1
. . . g
m
)) = (w
2
)
Y
(n(e, h
1
. . . h
k1
)) (4.33)
is a consequence of relations in R and of standard inverse semigroup relations. In the
following two steps, we verify that
(w
2
)
Y
(n(e, h
1
. . . h
k1
)) = (w
2
)
Y
(n(f, h
1
. . . h
k1
))
is a consequence of relations in R, and thus verifying that w
1
= w
2
is a consequence of
relations in R.
Step 3. By (4.31), we have that
w
1
=
(w
1
)
Y
e
and so we obtain
w
1
= w
1
e.
On the other hand we know that w
1
= w
2
, and hence
w
2
= w
2
e. (4.34)
Step 4. We give a nite sequence of words
(w
2
)
Y
(n(e, h
1
. . . h
k1
))
0
, . . . ,
n
(w
2
)
Y
(n(f, h
1
. . . h
k1
))
where
j+1
is obtained from
j
, using relations in R. Recall that
(w
2
)
Y
(f
1
, 1)d
2
. . . d
k1
,
where d
k1
v(f
k
, 1)v
1
and v n(f, h
1
. . . h
k1
). Applying (R3), we obtain that
v
1
(f, 1) = v
1
and hence we get that
(w
2
)
Y
= (w
2
)
Y
(f, 1).
It follows that
(w
2
)
Y
(n(e, h
1
. . . h
k1
)) = (w
2
)
Y
(f, 1)(n(e, h
1
. . . h
k1
))
= (w
2
)
Y
(e, 1)(n(f, h
1
. . . h
k1
)) by Lemma 4.10.4.
Bearing in mind that w
2
= w
2
e by (4.34), there exists a nite sequence of words
(w
2
)
Y
(e, 1)
1
, . . . ,
l
= (w
2
)
Y
,
such that
j+1
is obtained from
j
using relations in R
Y
R
L
R
G
R
M
. It follows that
(w
2
)
Y
(e, 1)v
1
v, . . . ,
l
v = (w
2
)
Y
v,
verifying that w
1
= w
2
is indeed a consequence of relations in R and of inverse semigroup
relations.
We end this section by returning to our examples.
Example 4.10.5. Let Y

on top and let G be the innite cyclic group. Dene the action of G on Y
as it was in
Proposition 4.2.4. Clearly G is nitely presented and Y
satises the maximum condition.

According to Proposition 4.7.1, Y
is nitely presented as an inverse semigroup with

respect to the action of G. Hence, by Theorem 4.9.1, the semidirect product S = Y
G
is nitely presented as an inverse semigroup.
Example 4.10.6. Consider the semilattice F
introduced in Proposition 4.2.6 and let

G be the innite cyclic group. Dene the action of G on F
as it was in Proposition
4.2.6. Clearly G is nitely presented and F
satises the maximum condition. We have

also shown in Proposition 4.7.2 that F
is nitely presented as an inverse semigroup with

respect to the action of G. Hence, by Theorem 4.9.1, the semidirect product S = F
G
is nitely presented as an inverse semigroup.
Example 4.10.7. Consider the semilattice A
introduced in Proposition 4.2.5 and let

G be the innite cyclic group. Dene the action of G on A
as it was in Proposition
4.2.5. We veried in Proposition 4.7.3, that A
is not nitely presented as an inverse

semigroup with respect to the action of G. Hence, by Theorem 4.9.1, the semidirect
product S = A
G is not nitely presented.

4.11. SCH
UTZENBERGER AUTOMATICITY 161

4.11 Sch utzenberger automaticity
In this section, we give a necessary and sucient condition for a semidirect product of a
semilattice by a group to be Sch utzenberger automatic.
As in the previous sections, if we say that a group G acts on a semilattice Y , then
it will be understood that G acts on Y on the left by automorphisms. We recall the
notation introduced in Section 9: Let A be a semigroup generating set for the semidirect
product S = Y G of a semilattice by a group and assume that A S. For a word
w (e
1
, g
1
) . . . (e
m
, g
m
) A
+
, we let
w g
1
. . . g
m
.
Proposition 4.11.1. Let Y be a semilattice and G be a group acting on Y . Then the
semidirect product S = Y G is Sch utzenberger automatic if and only if the following
conditions hold:
(SA1) G is automatic;
(SA2) Y satises the maximum condition and is generated by a
nite set Y
0
with respect to the action of G;
(SA3) There exists an automatic structure (X, K) for G such that
for all e Y and f Y
0
the language
K
f
e
= {w K| e
w
f}
is regular.
Proof. (=) If S = Y G is Sch utzenberger automatic, then S is nitely generated.
It follows by Proposition 4.3.1, that G is generated by a nite set X, that Y satises the
maximum condition and is generated by a nite set Y
0
with respect to the action of G,
and hence (SA2) holds. Let Y
m
denote the set of maximal elements of Y . For the sake of
convenience we assume that X is closed under taking inverses, that 1 X G and that
Y
m
Y
0
Y . In particular we may assume that S is generated by the nite set
A = (Y
0
{1}) (Y
m
X).
Since X is closed under taking inverses, for each element (e, h) A, (e, h)
1
A holds
by Lemma 4.2.1 and Proposition 4.2.7. Thus A is a semigroup generating set for S. Let
: A
+
S denote the homomorphism extending the identity map
1
: A S and let
: X
+
G denote the homomorphism extending the identity map
2
: X G. Let
e Y and consider s = (e, 1) S. Let R denote the R-class R of s. By Proposition 4.2.7,
we have that
R = {(e, g)|g G}. (4.35)
Making use of Proposition 3.2.1, and of Corollaries 3.4.6 and 3.6.6 in Chapter 1, we
have that there exists a regular language L over A such that the following conditions hold:
(c1) s L = R;
(c2) L
=
= {(u, u)|u L}
A
and hence is a regular language;
(c3) L
a
= {(u, v)|u, v L, s u a = s v}
A
is a regular language for all a A.
We claim that G is automatic by verifying that the fellow traveller property holds in
the Cayley graph of G. Consider the map
: L X
+
; w w,
and let K =

L. It follows by (4.35) and (c1), that K = G.
The language K is regular, since if A = (, A, , p, F) is a nite state automaton
accepting L, then the automaton obtained from A by changing the labels (e, h) A to h
accepts K. Let w
1
and w
2
K such that
d
(w
1
, w
2
) 1.
Then w
1
x = w
2
for some x X. Consider the elements u
1
, u
2
of L for which u
1
w
1
and u
2
w
2
. Clearly, for all t 1
u
1
(t) w
1
(t) and

u
2
(t) w
2
(t). (4.36)
Let f Y
m
such that e
w
1
f =
b u
1
f. Since Y satises the maximum condition, such an
element of Y
m
exists by Lemma 4.2.1. Then s u
1
(f, x) = s u
2
, and hence
d
(s u
1
, s u
2
) 1.
Since R is Sch utzenberger automatic, the fellow traveller property holds in the Sch utzen-
berger graph of R by Proposition 3.7.10. Hence, there exists a constant k N such that
whenever u, v L with d
(s u, s v) 1, then d
(s u(t), s v(t)) k holds for all t 1.

It follows that for all t 1,
d
(s u
1
(t), s u
2
(t)) k.
Because of the observations made in Example 3.7.2 of Chapter 1, there exists a word
over A, whose length is less then or equal to k and
s u
1
(t) = s u
2
(t).
4.11. SCH

Bearing in mind (4.36) and that s = (e, 1), we obtain that w
1
(t) = w
2
(t), and hence
d
(w
1
(t), w
2
(t)) k,
proving that the fellow traveller property holds in the Cayley graph of G.
Let w
1
, w
2
K. Before we show that (SA3) holds, we verify that
w
1
= w
2
if and only if w
1
w
2
. (4.37)
Clearly, if w
1
w
2
, then w
1
= w
2
. Assume that w
1
= w
2
and let u
1
, u
2
L such that
u
1
w
1
and u
2
w
2
. In particular we have that s u
1
= (e, w
1
) and s u
2
= (e, w
2
) and
hence we may deduce that (u
1
, u
2
) L
=
. It follows by condition (c2) that u
1
u
2
and so
w
1
w
2
, verifying (4.37). In particular we have obtained that
w
1
w
2
if and only if u
1
u
2
. (4.38)
To show that (SA3) holds, let f Y
0
and consider the generator (f, 1) A. Our aim is
to show that
K
f
e
= {w K|e
w
f}
is a regular language. Note that
(u, v)
A
L
(f,1)
s u (f, 1) = s v
(e, u)(f, 1) = (e, v)
( u, v) K
=
, u K
f
e
.
Thus u v by (4.37) and hence bearing in mind (4.38) we may deduce that
(u, v) L
(f,1)
if and only if u v and u K
f
e
. (4.39)
By Proposition 2.3.1, we have that the language
L
f
e
= {u L | (u, v)
A
L
(f,1)
for some v A
}
is regular. It follows that the language
L
f
e
= { u K | u L
f
e
}
is regular as well. We nish our proof by showing that
L
f
e
= K
f
e
.
Let w

L
f
e
. Then w u for some u L
f
e
, and hence e
u
f holds, verifying that w K
f
e
.
Now let w K
f
e
. Then e
w
f. Let u L such that u w. Then by (4.39), (u, u) L
(f,1)
,
and so u L
f
e
, proving that w

L
f
e
.
(=) For the converse, assume that G is a group acting on a semilattice Y , where
G and Y satisfy conditions (SA1) (SA3). More precisely, assume that (X, K) is an
automatic structure with uniqueness for G, where X is closed under taking inverses. To
simplify notation, we assume that 1 X G. Let : X
+
Gdenote the homomorphism
extending the identity map
1
: X G. Suppose that Y is generated by the nite set
Y
0
with respect to the action of G. We assume that the set of maximal elements Y
m
is
contained in Y
0
and that for all e Y and f Y
0
, the language K
f
e
= {w K|e
w
f} is
regular. Let S = Y G. By Corollary 4.3.2, the nite set
A = (Y
0
{1}) (Y
m
X)
generates S as a semigroup. Let : A
+
S be the homomorphism extending the identity
map
2
: A S.
Let R be an arbitrary R-class of S and assume that (e, 1) R. To simplify notation,
denote (e, 1) by s. Choose e Y
m
such that e e. Since Y satises the maximum
condition, such a maximal element of Y exists. Dene
: K A
+
; w n( e, w)
where n is the function dened in Section 8.
Let L = K. We show that L is a regular language, by constructing a nite state
automaton accepting it. Let A
1
= (
1
, X, , p, T) be a nite state automaton accepting
K. Such an automaton exists, since K is a regular language. To simplify notation,
let B = (Y
m
X) and consider the automaton A = (, B, , (p, ( e, 1)), T B), where
= (
1
B) {FS} and
: ((r, (l, g)), (f, h))
_
((r, h), (f, h)) if
g
f = l,
FS otherwise.
By the denition of the function n, and since A
1
is an automaton accepting K, we have
that A is a nite state automaton accepting L.
By the denition of the function n and since K is an automatic structure with unique-
ness, we have that
n( e, w
1
) = n( e, w
2
) if and only if w
1
= w
2
. (4.40)
Since K = G and since R = {(e, g)|g G} by Proposition 4.2.7, s L = R holds.
4.11. SCH

Let u, v L and w
1
, w
2
K such that w
1
u and w
2
v. Then
(u, v)
A
L
=
s u = s v
(e, 1)n( e, w
1
) = (e, 1)n( e, w
2
)
(e, w
1
) = (e, w
2
)
(w
1
, w
2
)
X
K
=
.
It follows by (4.40) that (u, v)
A
L
=
if and only if u v, verifying that
L
=
= {(u, u) | u L}
A
.
Let (f, x) A and consider the language K
f
e
= {u K|e
u
f}. By our assumptions,
K
f
e
is a regular language. Let u, v L and w
1
, w
2
K such that w
1
u and w
2
v.
Then
(u, v)
A
L
(f,x)
s u (f, x) = s v
(e, 1)n( e, w
1
)(f, x) = (e, 1)n( e, w
2
)
(e, w
1
)(f, x) = (e, w
2
)
(w
1
, w
2
)
X
K
x
, w
1
K
f
e
.
It follows that
L
(f,x)
= K
x
1
X
( )
A
(K
f
e
K)( )
A
.
Since K, K
x
and K
f
e
are regular languages we have that K
x
1
X
( )
A
and (K
f
e
K)(
)
A
are regular languages as well. Thus we may deduce that L
(f,x)

Chapter 5
HNN extensions
In the 1950s, G. Higman, B. Neumann and H. Neumann investigated embeddability ques-
tions of groups. Among others, they considered the following problem. Let G be a group
and U, V be subgroups of G that are isomorphic via : U V . They asked whether
it is possible to embed G into a group H, so that H possesses an element t for which
t
1
ut = u holds for all u U. They showed that this is always possible; the construction
they gave for H is called an HNN extension of G.
The above problem was rst introduced for semigroups by Howie [20] in the following
settings. Let S be a semigroup, e E(S). Let U, V eSe be subsemigroups of S that are
isomorphic via : U V . Does there exists a semigroup H and t, t
H, where t
V (t)
so that S embeds in H, t
ut = u and tt
= t
t = e. Howie proved that if U and V are

unitary subsemigroups of eSe, then this is possible.
HNN extension of inverse semigroups was considered in recent years. There are two
alternative denitions for an HNN extension of an inverse semigroup. One is introduced
by Yamamura [36], the other by Gilbert [14]. In this chapter, we rst discuss group
HNN extensions from nite generation and nite presentability point of view. Regarding
nite generation, a sucient condition will be straightforward. Moreover, we shall see
that Brittons lemma is the key in giving a necessary and sucient condition for nite
presentability of an HNN extension of a group. Considering Yamamuras HNN extension,
we verify that the group theoretic results generalize. The role of Brittons lemma is played
by the strong HNN property in this context. Gilberts HNN extension is more delicate
and the J-preorder is the key notion in understanding nite generation and presentability.
Regarding nite presentability, a variation of the strong HNN property will help us in
giving a necessary and sucient condition. Results of sections 3-8 are the result of a joint
work [11] with N.Gilbert and N.Ruskuc. We give more detailed proofs for these results.
167
168 CHAPTER 5. HNN EXTENSIONS
5.1 Group HNN extensions
The construction of an HNN extension of a group arose while Higman, Neumann and Neu-
mann studied embeddability problems related to groups. The construction is employed in
solving algorithmic problems, e.g. undecidability of Markov properties of nitely presented
groups.
Denition 5.1.1. Let G = GrpX|P. Let U and V be subgroups of G that are isomor-
phic via : U V . Let t / X. The group presented by
H = GrpX, t |P, t
1
ut = u (u U)
is called the HNN extension of G associated with : U V . The element t is called the
stable letter of H.
In what follows, we assume that G = GrpX|P, that U and V are subgroups in G
that are isomorphic via : U V . To simplify notation, instead of
H = GrpX, t |P, t
1
ut = u (u U),
we write
H = GrpG, t |t
1
Ut = V .
The following theorem is due to Higman, Neumann and Neumann.
Theorem 5.1.2. The group G embeds in H = GrpG, t |t
1
Ut = V .
Proof. See [2, Chapter IV].
Clearly, if G is nitely generated, then the HNN extension H is also nitely gener-
ated. However, the converse is not necessarily true. To demonstrate this fact, we give an
example. Let C
2
= x be the cyclic group of order two and let
G = . . . C
2
C
2
C
2
. . . .
Clearly G is not nitely generated. In particular we have that G is generated by the
elements of the form
x
i
= (. . . , 1, 1, x, 1, 1, . . .),
where x appears exactly at the ith position. Denote the set of these elements by X. Let
P = {x
2
= 1
G
, xy = yx|x, y X}. Then G = GrpX|P. Let U = V = G. We consider
the following automorphism of G:
: G G; x
i
1
x
i
2
. . . x
in
x
i
1
+1
x
i
2
+1
. . . x
in+1
.
5.1. GROUP HNN EXTENSIONS 169
Consider the HNN extension of G associated with : G G:
H = GrpX, t |P, t
1
ut = u, u G.
Observe that for every i Z, we have x
i
= (t
1
)
i
x
0
t
i
. It follows that H is generated by
the nite set {x
0
, t} as a group.
Next, we are going to discuss when H is nitely presented. For this, we need the
following notion.
Denition 5.1.3. Consider the HNN extension H = GrpG, t |t
1
Ut = V . A pinch is a
word over G {t, t
1
} of the form t
1
ut, where u U or of the form tvt
1
, where v V .
Lemma 5.1.4 (Britton). Consider H = GrpG, t |t
1
Ut = V . Let
w g
0
t
1
g
1
t
2
. . . t
n
g
n
,
where g
j
G (0 j n) and
j
{1, 1} (1 i n). If w represents the identity
element of H, then it contains a pinch.
Proof. See [2, Chapter IV].
Proposition 5.1.5. Let G be dened by the nite group presentation X|P. Let U and
V be subgroups in G that are isomorphic via : U V . Then the HNN extension
H = GrpX, t |P, t
1
ut = u (u U)
is nitely presented if and only if U is nitely generated.
Proof. Assume that the HNN extension
H = GrpX, t |P, t
1
ut = u (u U)
is nitely presented. Then there exists a nite subset Y U such that H is isomorphic to
H
0
= GrpG, t | t
1
ut = u, (u Y ).
Let U
0
denote the subgroup of U generated by Y . In particular, we have that H
0
is the
HNN extension of G associated with : U
0
U
0
. By assumption U is not nitely
generated, and so there exists u U \ U
0
. Then t
1
ut(u)
1
= 1 holds in H and hence in
H
0
. Making use of Brittons lemma it follows that t
1
ut(u)
1
contains a pinch and so
u U
0
, leading to a contradiction.
For the converse, assume that U is nitely generated by Y . Then, it is immediate that
H = GrpX, t |P, t
1
yt = y (y Y ).

5.2 Yamamuras HNN extension
Yamamura generalized the notion of an HNN extension to inverse semigroups in such
a way that the construction possesses similar properties as in the case of groups. In
[36], it is shown that several important inverse semigroups, e.g. free inverse semigroups
and the bicyclic monoid arise as Yamamura HNN extensions of certain semilattices. The
Yamamura HNN extension is as powerful as in the case of groups for discussing algorithmic
problems. Among others, undecidability of Markov properties of nitely presented inverse
semigroups is proved in [36].
Denition 5.2.1. Let S = InvX|P be an inverse semigroup. Let e, f E(S). Let U
and V be inverse subsemigroups of S such that e U eSe and f V fSf. Assume
that U and V are isomorphic via : U V . The inverse semigroup S
dened by the
presentation
S
= InvX, t |P, tt
1
= e, t
1
t = f, t
1
ut = u (u U)
is called the Yamamura HNN extension of S associated with : U V . The element t is
called the stable letter of S
.
In what follows, we assume that S = InvX|P, that e, f E(S) and that e U eSe
and f V fSf are isomorphic subsemigroups in S via : U V . To simplify notation,
instead of
S
= InvX, t |P, tt
1
= e, t
1
t = f, t
1
ut = u (u U)
we write
S
= InvS, t |tt
1
= e, t
1
t = f, t
1
Ut = V .
The following theorem is proved by Yamamura in [36].
Theorem 5.2.2. The inverse semigroup S embeds in the Yamamura HNN extension S
=
InvS, t |tt
1
= e, t
1
t = f, t
1
Ut = V .
If S is nitely generated, then the HNN extension S
is also nitely generated. As in

the case of groups, the converse is not true. The same example as the one introduced in
the previous section demonstrates this fact.
The main purpose of this section is to discuss nite presentability of the Yamamura
HNN extension. For this, we need the following theorem proved by Yamamura [36, The-
orem 12].
5.2. YAMAMURAS HNN EXTENSION 171
Theorem 5.2.3. Let S be an inverse semigroup and consider the Yamamura HNN exten-
sion S
= InvS, t |tt
1
= e, t
1
t = f, t
1
Ut = V . Then t
1
St S = t
1
Ut = V.
The above theorem can be considered as a weaker version of Brittons lemma and is
called the strong HNN property by Yamamura. Whether Brittons lemma can be general-
ized for Yamamuras construction is still unknown. Yamamuras conjecture is that a fully
analogous result does not hold. We see that concerning nite presentability of Yamamuras
HNN extension, the group theoretic result generalizes.
Proposition 5.2.4. Let S be dened by the nite inverse semigroup presentation X|P.
Let e, f E(S) and let e U eSe and f V fSf be inverse subsemigroups that are
isomorphic via : U V . Then the Yamamura HNN extension
S
= InvX, t |P, tt
1
= e, t
1
t = f, t
1
ut = u (u U)
Proof. Assume that the Yamamura HNN extension
S
= InvX, t |P, tt
1
= e, t
1
t = f, t
1
ut = u (u U)
is nitely presented. Then there exists a nite subset Y U such that S
is isomorphic
to
S
0
= InvX, t |P, tt
1
= e, t
1
t = f, t
1
ut = u (u Y ).
Without loss of generality we may assume that e Y . Let U
0
be the inverse subsemigroup
of U generated by Y . Since e is the identity element of U
0
, we have that for all u U
0
, the
relation t
1
ut = u is a consequence of relations of the form t
1
yt = y, where y Y .
It follows that S
0
is the Yamamura HNN extension of S associated with : U
0
U
0
.
Making use of Theorem 5.2.3, we obtain that
t
1
St S = t
1
U
0
t = U
0
= U
0
. (5.1)
By assumption U is not nitely generated, and so there exists u U \ U
0
. Clearly
t
1
ut = u holds in S
, and so in S
0
. By (5.1), there exists u
0
U
0
such that t
1
ut = u
0
,
and hence u = u
0
. Since is an isomorphism it is injective, and so we obtain the
contradiction that u = u
0
. Thus U is indeed nitely generated.
For the converse, assume that U is generated by a nite set Y . Since e is the iden-
tity element of U, we obtain that the Yamamura HNN extension is dened by the nite
presentation
S
= InvX, t |P, tt
1
= e, t
1
t = f, t
1
ut = u (u Y ).

5.3 Gilberts HNN extension
Making use of the relationship between inductive groupoids and inverse semigroups, Gil-
bert introduced an alternative notion for an HNN extension of an inverse semigroup in
[14]. The relationship between the two notions of HNN extensions is discussed, moreover
maximal subgroups and the maximal group homomorphic image of the construction is
described [14]. In [13], Gilbert shows that free inverse semigroups, and the bicyclic monoid
arise as Gilberts HNN extension of certain semilattices. To be more accurate, the latter
case is set in a more general context. Namely, it is explained how Bruck-Reilly extensions
of monoids arise as Gilberts HNN extensions. The purpose of this section is to introduce
Gilberts notion of an HNN extension of an inverse semigroup, and to prove some useful
results.
First, we recall some denitions and results regarding inverse semigroups. We have
already mentioned, that the natural partial order has an important role in inverse
semigroup theory.
Denition 5.3.1. Let S be an inverse semigroup and s, t S. We dene
s t s = et
for some e E(S).
It is easy to see, that
s t s = ft for some f E(S)
s
1
t
1
.
Denition 5.3.2. Let S be an inverse semigroup and U be a subset of S. We say that U
is an order ideal of S, if whenever s u, where u U, then s U holds.
There is another relation which will be important in our considerations, namely the
J
preorder.
Denition 5.3.3. Let S be an inverse semigroup and s, t S. We dene
s
J
t s StS.
Note that s
J
t and t
J
s if and only if s J t. Moreover, if s t, then s
J
t. With
the help of the natural partial order and the D-relation the
J
preorder can be described
in the following way [27, Proposition 8]:
5.3. GILBERTS HNN EXTENSION 173
Lemma 5.3.4. Let S be an inverse semigroup and s, t S. Then s
J
t if and only if
there exists u S such that sDu t.
Denition 5.3.5. Let S be an inverse semigroup. A set F E(S) is said to J-dominate
S, if for every s S there exists f F such that s
J
f.
Notation. Let U be an inverse subsemigroup of S. Then for any F E(U) we let
U
F
= {u U | e, f F such that uu
1
e, u
1
u f};
this is easily seen to be an inverse subsemigroup of S.
Denition 5.3.6. Let S = InvX|P. Let U and V be inverse subsemigroups of S that
are order ideals. Assume that U and V are isomorphic via : U V . The inverse
semigroup S
U,
dened by the presentation
S
U,
= InvX {t
e
| e E(U)} |P, t
e
t
1
f
= ef, t
1
e
t
f
= (ef),
t
1
uu
1
ut
u
1
u
= u (u U)
is called the Gilbert HNN extension of S associated with : U V . The elements t
e
are
called the stable letters of S
U,
.
To make reference easier to certain relations, we x the following notations for this
section. Let S = InvX|P. We let U and V denote inverse subsemigroups of S that
are order ideals. We assume that U and V are isomorphic via : U V . For a subset
L E(U) we let
T
L
= {t
e
t
1
f
= ef, t
1
e
t
f
= (ef) | e, f L}
and for brevity we denote T
E(U)
by T. For a subset L E(U) and a subset K of U we let
W
L,K
= {t
1
e
at
f
= a | a K, e, f L, aa
1
e, a
1
a f}
and for brevity we denote W
E(U),U
by W. The following proposition will prove useful.
Proposition 5.3.7. (a) If e E(S) and f E(U), then et
f
= t
ef
. In particular, if
e f, then et
f
= t
e
.
(b) Let e, f E(U) and u U with uu
1
e and u
1
u f. Then the relation
t
1
e
ut
f
= u holds in S
U,
, and so
S
U,
= InvX {t
e
|e E(U)}|P, T, W.
(c) Fix a subset F E(U) and suppose that U
F
is generated by a subset Y U
F
. Then
for all u U
F
with uu
1
e and u
1
u f, e, f F, the relation t
1
e
ut
f
= u is a
consequence of the relations in P T
F
W
F,Y
.
Proof. (a) Let e E(S) and f E(U). We verify that et
f
= t
ef
is a consequence of
the dening relations of S
U,
and of standard inverse semigroup relations. Indeed,
et
f
= et
f
t
1
f
t
f
= eft
f
by applying a relation in T
= t
ef
t
1
ef
t
f
= t
ef
(ef) by applying a relation in T
= t
ef
(ef)t
1
ef
t
ef
by applying standard inv. sg. relations
= (ef)t
ef
since t
ef
(ef)t
1
ef
= ef
= t
ef
t
1
ef
t
ef
= t
ef
.
(b) Since uu
1
e, we have that euu
1
= uu
1
. Since u
1
u f, we have that u
1
uf =
u
1
u. Thus, with the previous part of the proposition in mind, we obtain that
t
1
e
ut
f
= t
1
e
uu
1
uu
1
ut
f
= t
1
uu
1
ut
u
1
u
= u.
It follows that all relations in W hold in S
U,
. On the other hand
{t
1
uu
1
ut
u
1
u
= u | u U} W,
and so we obtain that
S
U,
= InvX {t
e
| e E(U)}| P, T, W.
(c) We proceed by induction on the number of elements of Y required to express u U
as a product. The base case is dealt with by the relations with y Y . Let u = u
1
u
2
, with
u
1
, u
2
U
F
and assume that u
1
u
1
1
i = t
i
t
1
i
, u
1
1
u
1
g = t
g
t
1
g
and u
2
u
1
2
h =
t
h
t
1
h
, u
1
2
u
2
j = t
j
t
1
j
for some i, j, g, h F. Then u
1
= iu
1
g and u
2
= hu
2
j hold as a
consequence of P, and hence
t
1
e
ut
f
= t
1
e
u
1
u
2
t
f
= t
1
e
(iu
1
g) (hu
2
j)t
f
= t
1
e
t
i
(t
1
i
u
1
t
g
)(t
1
g
t
h
)(t
1
h
u
2
t
j
)t
1
j
t
f
= t
1
e
t
i
(u
1
) t
1
g
t
h
(u
2
) t
1
j
t
f
= (ei) (u
1
) (gh) (u
2
) (jf) = (e(iu
1
g)(hu
2
j)f)
= (eu
1
u
2
f) = u.
5.4. AN ALTERNATIVE PRESENTATION FOR GILBERTS HNN EXTENSION 175

Denition 5.3.8. Let S be an inverse semigroup. A product s
1
s
2
. . . s
n
of elements of S
is called a trace product if s
1
i
s
i
= s
i+1
s
1
i+1
for all 1 i n 1.
Lemma 5.3.9. Let S be an inverse semigroup and let s
1
, s
2
S. Then there exists
t
1
, t
2
S such that the product of s
1
s
2
equals to the trace product t
1
t
2
and t
1
s
1
, t
2
s
2
holds.
Proof. Consider the idempotent e = s
1
1
s
1
s
2
s
1
2
and let t
1
= s
1
e and t
2
= es
2
. Clearly
s
1
t
1
and s
2
t
2
. Moreover an easy calculation shows that t
1
1
t
1
= t
2
t
1
2
and that
t
1
t
2
= s
1
es
2
= s
1
s
2
.
By an easy induction the following lemma can be veried.
Lemma 5.3.10. Let S be an inverse semigroup. Then a product of elements s
1
s
2
. . . s
n
of S equals to a trace product t
1
t
2
. . . t
n
with t
i
s
i
for all 1 i n. Moreover, if
s = t
1
. . . t
n
is a trace product, then ss
1
= t
1
t
1
1
and s
1
s = t
1
n
t
n
.
In the HNN extension S
U,
, we shall consider products that are words on the alphabet
S {t
e
|e E(U)}: for such words, we can form trace products that preserve the structure
of the word. Combining Lemma 5.3.10 with Proposition 5.3.7(a) we obtain the following:
Lemma 5.3.11. An element h = s
1
t
1
f
1
s
2
t
2
f
2
s
n
t
n
fn
s
n+1
S
U,
, where
i
= 1 for
1 i n and s
i
S for 1 i n + 1 is equal to a trace product r
1
t
1
e
1
r
2
t
2
e
2
r
n
t
n
en
r
n+1
where e
i
f
i
for 1 i n and r
i
s
i
, r
i
S for 1 i n + 1.
Denition 5.3.12. Let S be an inverse semigroup. Let U and V be inverse subsemigroups
of S that are order ideals. Assume that U and V are isomorphic via : U V . Consider
the Gilbert HNN extension S
U,
. A pinch is a word over S {t
e
, t
1
e
| e E(U)} of the
form t
1
uu
1
ut
u
1
u
, where u U or of the form t
1
uu
1
(u)t
u
1
u
.
The following proposition will prove useful when proving Proposition 5.5.1.
Proposition 5.3.13. [18, 29] Let h = s
1
t
1
e
1
s
2
t
2
e
2
s
n
t
n
en
s
n+1
be a trace product in S
U,
,
with n > 0. If h is an idempotent then it contains a pinch.
5.4 An alternative presentation for Gilberts HNN exten-
sion
In preparation for discussing nite presentability of Gilberts HNN extension, we give an
alternative presentation for the HNN extension S
U,
. Throughout this section, we use
the notation introduced in the previous section.
Proposition 5.4.1. Let S = InvX|P be an inverse semigroup and let U, V be inverse
subsemigroups of S that are order ideals. Assume that U and V are isomorphic via :
U V . Let F be any J-dominant subset of U. Then
S
U,
= InvX {t
e
: e F}|P, T
F
, W
F,U
F
.
Proof. Let X
F
= X {t
e
: e F} and H
F
= InvX
F
|P, T
F
, W
F,U
F
. By Proposition
5.3.7, S
U,
= InvX {t
e
: e E(U)}|P, T, W. We show that S
U,
and H
F
are
isomorphic inverse semigroups by giving two inverse semigroup homomorphisms : H
F

S
U,
and : S
U,
H
F
Since T
F
T and W
F,U
F
W, we clearly have that the identity map on X
F
induces
an inverse semigroup homomorphism : H
F
S
U,
.
For each e E(U), choose and x u
e
U and e F such that
e = u
e
u
1
e
, u
1
e
u
e
e;
u
e
= e = e (e F).
We show that the mapping t
e
u
e
t
e
(u
e
)
1
and the identity map on X together induce
an inverse semigroup homomorphism : S
U,
H
F
. That is we show that maps
relations in S
U,
to relations that hold in H
F
.
Let e
1
, e
2
E(U). Clearly u
1
e
1
u
e
2
u
1
e
2
u
e
1
u
1
e
1
u
e
1
e
1
and u
1
e
2
u
e
1
u
1
e
1
u
e
2
u
1
e
2
u
e
2

e
2
, and we have
t
e
1
t
1
e
2
= u
e
1
t
e
1
(u
e
1
)
1
(u
e
2
)t
1
e
2
u
1
e
2
= u
e
1
t
e
1
(u
1
e
1
u
e
2
)t
1
e
2
u
1
e
2
= u
e
1
u
1
e
1
u
e
2
u
1
e
2
= e
1
e
2
= (e
1
e
2
)
using the relation t
e
1
(u
1
e
1
u
e
2
)t
1
e
2
= u
1
e
1
u
e
2
from W
F,U
F
and it follows that (t
e
1
t
1
e
2
) =
(e
1
e
2
) indeed holds in H
F
. Similarly,
t
1
e
1
t
e
2
= (u
e
1
)t
1
e
1
u
1
e
1
u
e
2
t
e
2
(u
e
2
)
1
= (u
e
1
)(u
1
e
1
u
e
2
)(u
e
2
)
1
= (u
e
1
u
1
e
1
u
e
2
u
1
e
2
) = (e
1
e
2
) = ((e
1
e
2
)).
and it follows that (t
1
e
1
t
e
2
) = ((e
1
e
2
)). These considerations show that all the relations
in T, once rewritten in terms of X
F
, follow from T
U
and W
F,U
F
in H
F
.
Now consider a relation t
1
e
1
ut
e
2
= u from W, where uu
1
e
1
and u
1
u e
2
, u
U, e
1
, e
2
E(U). Consider t
e
1
= u
e
1
t
e
1
(u
e
1
)
1
and t
e
2
= u
e
2
t
e
2
(u
e
2
)
1
. Note that
(u
1
e
1
uu
e
2
)(u
1
e
2
u
1
u
e
1
) = u
1
e
1
ue
2
u
1
u
e
1
= u
1
e
1
uu
1
u
e
1
= u
1
e
1
u
e
1
e
1
5.5. FINITE GENERATION OF GILBERTS HNN EXTENSION 177
and similarly (u
1
e
2
u
1
u
e
1
)(u
1
e
1
uu
e
2
) e
2
, hence u
1
e
1
uu
e
2
U
F
. In particular, we have
that t
1
e
1
u
1
e
1
uu
e
2
t
e
2
= (u
1
e
1
uu
e
2
) is a relation in W
F,U
F
and we obtain
t
1
e
1
ut
e
2
= (u
e
1
)t
1
e
1
u
1
e
1
uu
e
2
t
e
2
(u
e
2
)
1
= (u
e
1
)(u
1
e
1
uu
e
2
)(u
e
2
)
1
= (u
e
1
u
1
e
1
uu
e
2
u
1
e
2
)
= (e
1
ue
2
) = u = (u).
It clearly follows that (t
1
e
1
ut
e
2
) = (u) indeed holds in H
F
. This completes the check
that is a homomorphism.
Let us now check that and are inverse to each other. Indeed, for x X we clearly
have x = x and x = x. Likewise, for e F we have t
e
= t
e
. Now consider any
e E(U); we have
t
e
= (u
e
t
e
(u
e
)
1
) = u
e
t
e
(u
e
)
1
= u
e
u
1
e
t
e
= et
e
= t
e
,
using the relations t
1
e
u
e
t
e
= u
e
and et
e
= t
e
, which hold in S
U,
. This completes the
proof that and are mutually inverse.
5.5 Finite generation of Gilberts HNN extension
In this section, a necessary and sucient condition will be given for the Gilbert HNN
extension of inverse semigroups to be nitely generated.
As, in the previous section, we use the following notation. We let S denote an inverse
semigroup. We let U and V denote inverse subsemigroups of S that are order ideals. We
assume that U and V are isomorphic via : U V . For a subset L E(U) we let
T
L
= {t
e
t
1
f
= ef, t
1
e
t
f
= (ef) | e, f L}
and for brevity we denote T
E(U)
by T. For a subset L E(U) and a subset K of U we let
W
L,K
= {t
1
e
at
f
= a | a K, e, f L, aa
1
e, a
1
a f}
and for brevity we denote W
E(U),U
by W.
Proposition 5.5.1. Let S be a nitely generated inverse semigroup. Let U and V be
inverse subsemigroups of S that are order ideals. Assume that U and V are isomorphic
via : U V . Then the Gilbert HNN extension S
U,
is nitely generated if and only if
U is nitely Jdominated.
Proof. Suppose that S is nitely generated by X and that S
U,
is nitely generated.
Then there exists a nite subset F E(U), such that S
U,
is generated by X{t
f
|f F},
and hence by S {t
f
|f F}. Let e E(U) \ F. Then, in the Gilbert HNN extension
S
U,
, the stable letter t
e
is equal to some word
w s
1
t
1
f
1
s
2
t
2
f
2
s
n
t
n
fn
s
n+1
on the generating set S {t
f
|f F}, where
i
= 1 , (1 i n). By Lemma 5.3.11, t
e
may also be represented by a trace product
w
= r
1
t
1
e
1
r
2
t
2
e
2
r
n
t
n
en
r
n+1
where for each i , (1 i n), e
i
f
i
and r
i
s
i
for each i , (1 i n+1). By application
of Wrelations we may assume that w
contains no pinch. It is possible that w
is the
trace product t
e
, in which case e f for some f F, in particular e
J
f. Assume now
that w
is not identical to t
e
. Note that by Lemma 5.3.10 w
t
1
e
is a trace product and
by Proposition 5.3.13 the trace product w
t
1
e
contains a pinch. This must be the nal
segment t
n
en
r
n+1
t
1
e
, where r
n+1
= u for some u U and e = u
1
u. Hence eDuu
1
and
uu
1
f
n
F: that is, e
J
f
n
. Therefore, U is nitely Jdominated by the set F.
Conversely, suppose that U is nitely Jdominated by F. Given e E(U) suppose
that e
J
f with eDh and h f. Take u U with uu
1
= e and u
1
u = h. Then
t
1
e
= t
1
e
t
e
t
1
e
= t
1
e
uu
1
by applying relations in T
= t
1
e
uu
1
uu
1
= t
1
e
uhu
1
= t
1
e
ut
h
t
1
h
u
1
= ut
1
h
u
1
by applying relations in T,
and so t
e
= ut
h
(u)
1
holds. In particular we have that the following hold in S
U,
:
t
e
= ut
h
(u)
1
= ut
h
t
1
h
t
h
(u)
1
= uht
h
(u)
1
= uht
f
(u)
1
by Proposition 5.3.7
= ut
f
(u)
1
.
Thus we may deduce that t
e
is expressible as a word on the generators X {t
f
|f F},
and S
U,
5.6. FINITE PRESENTABILITY OF GILBERTS HNN EXTENSION 179
5.6 Finite presentability of Gilberts HNN extension
The main purpose of this section is to give a necessary and sucient condition for S
U,
to be nitely presented.
Proposition 5.6.1. Let S be a nitely presented inverse semigroup and let U, V be inverse
subsemigroups of S, that are order ideals. Assume that U and V are isomorphic via
: U V . Then the Gilbert HNN extension S
U,
is nitely presented if and only if U
is nitely Jdominated by a subset F E(U) such that the inverse subsemigroup U
F
is
nitely generated.
Proof. Assume that the inverse semigroup S is dened by the nite inverse semigroup
presentation X|P. Assume that U is Jdominated by a nite subset F E(U), and
that U
F
is generated by a nite set Y of words on X. As before, let X
F
= X{t
e
|e F}.
Then by Proposition 5.4.1,
S
U,
= InvX
F
|P, T
F
, W
F,U
F
.
On the other hand, since W
F,Y
W
F,U
F
and since all relations in W
F,U
F
are consequences
of relations in R T
F
W
F,Y
by Proposition 5.3.7 (c), we have that
InvX
F
|P, T
F
, W
F,U
F
= InvX
F
|P, T
F
, W
F,Y
.
Since F and Y are nite it is clear that the latter presentation is nite, hence S
U,
is
indeed nitely presented.
For the converse, suppose that S
U,
is a nitely presented inverse semigroup. In
particular, S
U,
is nitely generated and so by Proposition 5.5.1, there exists a nite
Jdominant subset F E(U) of U. Then, by Proposition 5.4.1,
S
U,
= InvX
F
|P, T
F
, W
F,U
F
,
where P and T
F
are nite sets of relations, but W
F,U
F
is possibly innite. Since S
U,
is
nitely presented by assumption, there exists a nite subset W
0
W
F,U
F
, such that
S
U,
= InvX
F
|P, T
F
, W
0
.
Consider the nite set
Y = {u U|e, f F such that (t
1
e
ut
f
= u) W
0
} F,
and let U
0
be the inverse subsemigroup of U generated by Y . Our aim is to show that
U
F
is nitely generated by showing that U
0
= U
F
. By denition of U
0
, we clearly have
U
0
U
F
. For the reverse inclusion we need the following variation of the strong HNN
embeddability.
Lemma 5.6.2. Let e, f F and U
e,f
= {u U
0
|uu
1
e, u
1
u f}. Then, in S
U,
,
we have
t
1
e
St
f
S = t
1
e
U
e,f
t
f
= U
e,f
.
Proof. Relations in W
F,U
F
imply immediately that t
1
e
U
e,f
t
f
= U
e,f
and that
t
1
e
U
e,f
t
f
t
1
e
St
f
S.
To prove the reverse inclusion we employ a technique used by Yamamura [36]. Let U
0
=
V
0
and S
be a copy of S with : S S
an isomorphism, let U
0
= U
0
. Since U
0
is
an inverse subsemigroup of S, the amalgam K = S
V
0
=U
0
S
can be considered. By a
theorem of Hall [16] (see also [30]), S and S
are canonically embedded in K, that is we

have embeddings
1
: S K,
2
: S
K. Moreover by the strong amalgamation property

of inverse semigroups [16] we have that S
1
S
2
= U
0
= V
0
. Let =
1
1

2
. Then it is
clear that : S
1
S
2
is an isomorphism; furthermore for all u U
0
, (u
1
) = (u)
2
=
(u)
1
. It follows by identifying S
1
with S, that |
U
0
= |
U
0
. Let K
1
be obtained from
K by adjoining an identity to it, and form the Yamamura HNN extension
K
= InvK
1
, t|tt
1
= 1 = t
1
t, t
1
st = s , (s S)
associating the subsemigroups (S
1
)
1
and (S
2
)
1
.
Next we show that the mapping t
e
et and the identity map on S together induce an
inverse semigroup homomorphism : S
U,
K
, that is we verify that maps relations

in S
U,
to relations that hold in K
. Let e, f F and u U
e,f
such that uu
1
e and
u
1
u f:
(t
e
)(t
1
f
) = (et)(ft)
1
= ett
1
f = ef = (ef),
(t
1
e
)(t
f
) = (et)
1
(ft) = t
1
eft = (ef) = (ef) = ((ef)),
(t
1
e
)(u)(t
f
) = (et)
1
u(ft) = t
1
euft = t
1
ut = u = u = (u).
Therefore relations in P T W are indeed mapped onto relations that hold in K
.
Let us now consider an arbitrary s t
1
e
St
f
S, and write s = t
1
e
zt
f
for some z S.
Then in K
we have
s = s = t
1
ezft t
1
eSft S.
5.7. A COMPELLING CONSEQUENCE 181
By the above mentioned strong amalgamation property, we have that
(et)
1
S(ft) S = t
1
eSft S = (eSf) S S S S
2
S
1
= U
0
= V
0
.
Therefore there exists u
0
U
0
such that the relation t
1
ezft = u
0
holds in K
. Let g, h
F be such that u
0
u
1
0
g and u
1
0
u
0
h, so that the relation u
0
= (gt)
1
u
0
(ht) = t
1
u
0
t
holds in K
. Now from t
1
ezft = t
1
u
0
t and tt
1
= t
1
t = 1 it follows that ezf = u
0
.
This, in turn, implies that u
0
u
1
0
e and u
1
0
u
0
f; in other words u
0
U
e,f
. Also,
notice that ezf = u
0
is a relation between elements of S, and hence it also holds in S
U,
.
So, returning to S
U,
, we have
s = t
1
e
zt
f
= t
1
e
ezft
f
= t
1
e
u
0
t
f
t
1
e
U
e,f
t
f
,
completing the proof.
We are now in the position to prove the remaining inclusion U
F
U
0
. Let u U
F
.
Then there exist e, f F such that uu
1
e and u
1
u f. It follows that t
1
e
ut
f
= u
holds in S
U,
. Applying Lemma 5.6.2, we have that there exists u
0
U
0
such that
u = u
0
, and so we obtain that u = u
0
U
0
, proving that U
F
= U
0
is indeed nitely
generated.
We conclude this section by formulating Proposition 5.6.1 in the special case when
E(U) satises the maximum condition. Recall that we say that E(U) satises the maxi-
mum condition if U has nitely many maximal idempotents F = {f
1
, f
2
, . . . , f
m
} and for
every idempotent e E(U) there exists f
i
F such that e f
i
. Note that in this case F
is also a nite Jdominant subset and U
F
= U holds, and so we may conclude:
Corollary 5.6.3. Let S be a nitely presented inverse semigroup and let U, V be inverse
subsemigroups of S that are order ideals. Assume that U and V are isomorphic via :
U V . Assume that E(U) satises the maximum condition. Then the HNN extension
S
U,
5.7 A compelling consequence
Proposition 5.6.1 and its proof have the following intriguing consequence. Assume that
S, U, V and are as in Proposition 5.6.1 and suppose that S
U,
is nitely presented.
Taking any nite J-dominant set F E(U), we have
S
U,
= InvX
F
|R, T
F
, W
0
,
where W
0
is a nite subset of W
F,U
F
. But then the converse part of the proof of Proposition
5.6.1 shows how this presentation yields a nite generating set for U
F
. In other words if
U
F
is nitely generated for some nite J-dominant subset F then U
G
is nitely generated
for every nite J-dominant subset G. In fact, we can prove this in full generality removing
the requirements that S be nitely presented, that U be an order ideal, and any mention
of V .
Proposition 5.7.1. Let S be an inverse semigroup and assume that the inverse sub-
semigroup U is nitely Jdominated. Then U
F
is nitely generated for some nite J
dominant set F if and only if U
G
is nitely generated for every nite Jdominant set
G.
The proof is divided into the following two lemmas.
Lemma 5.7.2. Let U
F
be nitely generated and g be an arbitrary idempotent of U. Then
U
F{g}
Proof. Assume that U
F
is generated by the nite set A. Let e F be such that
g
J
e. Then there exist k, l U, such that g = kel = kell
1
ek
1
= kll
1
ell
1
k
1
,
hence there exists v U such that g = vev
1
. From this we have that gv = ve = gve. We
claim that U
F{g}
is generated by A {gve}. Let s U
F{g}
. The following four cases
are to be considered.
(i) The case when s U
F
is straightforward.
(ii) Assume that ss
1
, s
1
s g. Then s = gsg = v(ev
1
sve)v
1
. Since ev
1
sve U
F
,
there exist a
1
, . . . , a
k
A A
1
such that ev
1
sve = a
1
. . . a
k
. From this we have
that s = ve a
1
. . . a
k
ev
1
= gve a
1
. . . a
k
(gve)
1
.
(iii) Assume that ss
1
g and s
1
s f for some f F. Then s = gsf = v(ev
1
sf).
Since ev
1
sf U
F
, there exist a
1
, . . . , a
k
A A
1
such that ev
1
sf = a
1
. . . a
k
.
From this we have that s = ve a
1
. . . a
k
= gve a
1
. . . a
k
.
(iv) The case when s
1
s g and ss
1
f for some f F can be proved similarly as
(iii).
Lemma 5.7.3. Assume that U
F
is nitely generated and that for some g F, G = F \{g}
is also Jdominant in U. Then U
G
5.8. EXAMPLES 183
Proof. Let U
F
be generated by the nite subset A which we may assume, without
loss of generality, is closed under taking inverses. There exists e F \ {g} for which
g
J
e, and hence g = vev
1
for some v U. Let A
denote the set of elements of A

for which aa
1
g and for which there is no other j F \ {g} such that aa
1
j; also
let A
= (A
)
1
. Notice that if a A
then a = ga, while if a A
, then a = ag. Let
B = A\ (A
). We show that U
G
is generated by
B =

B {a
i
ve: a
i
A\ A
} {ev
1
a
i
: a
i
A\ A
} {ev
1
a
i
ve: a
i
A}.
Let s U
G
U
F
. Then s = a
1
a
2
. . . a
k
for some a
1
, . . . , a
k
A. If a
i
A
, then substitute
a
i
by ga
i
; if a
i
A
, then substitute a
i
by a
i
g; and if a
i
A
, then substitute a
i
by
ga
i
g. Consider now all subwords w of the form a
i1
a
i
ga
i+1
g . . . ga
m1
ga
m
a
m+1
. These
subwords can be written in terms of B, since
w = a
i1
a
i
ve ev
1
a
i
ve ev
1
a
i+1
. . . ev
1
a
m1
ve ev
1
a
m
a
m+1
.
It follows that s can be written in terms of B, hence U
G
is generated by B.
5.8 Examples
Here we provide four examples of HNN extension of inverse semigroups.
Example 1. Let B be the bicyclic monoid, considered as the set N N with binary
operation (m, n)(p, q) = (m n + max(n, p), q p + max(n, p)). Now B is presented
as an inverse monoid by B = Inva|a = a
2
a
1
, where a = (0, 1). Let U = B and
V = {(m, n)|m, n > 0} with : U V the shift map (p, q) (p + 1, q + 1), so that
a = a
1
a
2
. It is clear that U = B is Jdominated by the identity element 1 = (0, 0), and
that U
{1}
= B is nitely generated. Hence the HNN extension B
B,
is nitely presented.
It is easy to check that
Inva, t|a = a
2
a
1
, tt
1
= aa
1
, t
1
t = a
1
a, t
1
at = a
1
a
2
gives a presentation.
Example 2. With B as before, suppose now that U = E(B) = {(m, m)|m N} with
: (m, m) (m + 1, m + 1). It is clear that U is nitely Jdominated by the identity
element 1 = (0, 0) of U, but U
{1}
= U is not nitely generated. Hence we may conclude
that the HNN extension B
U,
is nitely generated but not nitely presented.
Example 3. In this example we construct a nitely presented HNN extension S
U,
,
with U not nitely generated. Let S = Inva, e|aa
1
= 1, ae = 0, e
2
= e considered as a
semigroup with 0. It can be easily seen that
{a
i
a
j
, a
i
ea
j
|i, j N} {0}
is a set of normal forms for S. Consider the inverse subsemigroup
U = {a
i
ea
j
|i, j N} {0}
of S , together with the identity isomorphism : U U. We observe that U is isomorphic
to the innite aperiodic Brandt semigroup, and hence U is Jdominated by any of its
idempotents. Let F = {a
i
ea
i
}. Then U
F
= {a
i
ea
i
, 0} is nite, so certainly nitely
generated, and it follows that the HNN extension S
U,
is nitely presented. However, the
associated inverse subsemigroup U is not nitely generated.
Example 4. Our nal example shows that the maximum condition on E(U) is essential
in Corollary 5.6.3. We give an example of a nitely presented HNN extension of a nitely
presented inverse semigroup S, where the associated subsemigroups contain nitely many
maximal idempotents, but do not satisfy the maximum condition and are not nitely gen-
erated. Consider the semidirect product S = Y
G, introduced in Example 4.10.5. We

veried that S is nitely presented as an inverse semigroup. In particular, it is generated
by the set {(e
0
, 1), (1, g), (1, g
1
)} and if we correspond a to (1, g), b to (1, g
1
) and c to
(e
0
, 1), then we have that
S = Inva, b, c|ab = ba = 1, c
2
= c, cbc = bc, cac = ca.
Then it can be easily seen that
(i) {a
i
: i N} {b
i
: i N} {ab} {x
i
cy
j
: x, y {a, b} i, j N} is a set of normal
forms for S,
(ii) E(T) is an innite chain with an identity element adjoined on top:
. . . < b
2
ca
2
< bca < c < acb < a
2
cb
2
< . . . < 1,
(iii) the Hclass H
1
of 1 is isomorphic to the innite cyclic group,
(iv) K = T \H
1
is a bisimple aperiodic inverse subsemigroup of S, which is not nitely
generated.
It follows from (ii) and (iv) that, for any e E(S) \ {1}, the subsemigroup
U
e
= {u U : uu
1
, u
1
u e}
5.8. EXAMPLES 185
is nitely generated. To be more accurate, if e = (e
n
, 1), then U
e
is generated by the nite
set {(e
n
, 1), (e
n
, g), (
g
1
e
n
, g
1
)}.
Take two copies S
(1)
and S
(2)
of S and let T be their 0direct union. It is straightfor-
ward that T is nitely presented. Let f E(S
(2)
) and let U be the 0-direct union of K
with U
f
. Then U has one maximal idempotent f, and is not nitely generated since the
subsemigroup K contains an innite ascending chain of idempotents. Moreover since K
is bisimple, for any e E(K), the idempotents e and f are Jmaximal in U and
U
{e,f}
= {u U : uu
1
, u
1
u e or uu
1
, u
1
u f}
is isomorphic to the 0direct union of U
e
and U
f
, and hence is nitely generated. It follows
by Proposition 5.6.1, that the HNN extension H = S
U,
, where : U U is the identity
isomorphism, is nitely presented.
Bibliography
[1] I.M. Ara ujo, Presentation for semigroup constructions and related computational
methods, Ph.D. Thesis, University of St Andrews, 2000.
[2] G. Baumslag, Topics in Combinatorial Group Theory, Birkh auser Verlag, 1993.
[3] G. Baumslag, S.M. Gersten, M. Shapiro, H. Short, Automatic groups and amalgams,
J. Pure Appl. Algebra 76 (1991) 229316.
[4] C.M. Campbell, E.F. Robertson, N. Ruskuc, R.M. Thomas, Automatic completely
simple semigroups, Technical Report, University of Leicester, 2000.
[5] C.M. Campbell, E.F. Robertson, N. Ruskuc, R.M. Thomas, Automatic semigroups,
Theoretical Computer Science 250 (2001) 365391.
[6] C.M. Campbell, E.F. Robertson, N. Ruskuc, R.M. Thomas, Direct product of auto-
matic semigroups, J.Austral.Math.Soc. (Series A) 69 (2000) 1924.
[7] C. Carvalho, Presentations of semigroups and inverse semigroups, MSc Thesis, Uni-
versity of St Andrews, 2002.
[8] A.Cutting, A.Solomon, Remarks concerning nitely generated semigroups having reg-
ular sets of unique normal forms, J.Austral.Math.Soc. 70 (2001) 293309.
[9] D.B.A. Epstein, J.W. Cannon, D.F. Holt, S.V.F. Levy, M.S. Paterson, W.P. Thurston,
Word Processing in Groups, Jones and Barlett Publishers, 1992.
[10] L. Descalco, Automatic semigroups:Constructions and Subsemigroups, Ph.D. Thesis,
University of St Andrews (2002).
[11] E. Dombi, N.D. Gilbert, N. Ruskuc Finite presentability of HNN extensions of inverse
semigroups, Internat. J. Algebra Comput. (to appear).
187
188 BIBLIOGRAPHY
[12] A.J. Duncan, E.F. Robertson, N. Ruskuc, Automatic monoids and change of gener-
ators, Math.Proc.Camb.Phil.Soc. 127 (1999), 403409.
[13] N.D. Gilbert, Some constructions of inverse semigroups as HNN extensions, preprint.
[14] N.D. Gilbert, HNN extensions of inverse semigroups and groupoids, J. Algebra 272
(2004) 2745.
[15] S. Haataja, S.W. Margolis and J. Meakin, Bass-Serre theory for groupoids and the
structure of full regular semigroup amalgams, J. Algebra 183 (1996), 3854.
[16] T.E. Hall, Amalgamation of inverse and regular semigroups, Trans. Amer. Math. Soc.
246 (1978), 395406.
[17] P.J. Higgins, Notes on categories and groupoids, Van Nostrand Reinhold Math. Stud.
32, 1971.
[18] P.J. Higgins, The fundamental groupoid of a graph of groups, J. London Math. Soc.
(2) 13 (1976), 145149.
[19] M. Homann; Automatic semigroups, Ph.D. Thesis, University of Leicester, 2001.
[20] J.M. Howie; Embedding theorems for semigroups, Quart. J. Math (2) 14 (1963), 254
258.
[21] J.M. Howie; Automata and languages, Clarendon Press, Oxford, 1991.
[22] J.M. Howie, Fundamentals of semigroup theory, Clarendon Press, Oxford, 1995.
[23] T. Jajcayova, HNN extensions of inverse semigroups, Ph.D. Thesis, University of
NebraskaLincoln, 1997.
[24] P.R. Jones, Free products of inverse semigroups, Trans. Amer. Math. Soc. 282 (1984)
293317.
[25] P.R. Jones, S.W. Margolis, J. Meakin, J.B. Stephen, Free products of inverse semi-
groups II., Glasgow Math. J. 33 (1991) 373387.
[26] B.Khoussainov, A.Nerode, Automatic presentations of structures, Lecture Notes in
Comput.Sci., 813, Springer, Berlin 1995.
[27] M.V. Lawson, Inverse Semigroups, World Scientic, 1998.
BIBLIOGRAPHY 189
[28] M.V. Lawson, Finite automata, Chapman & Hall/CRC, 2004.
[29] E.J. Moore, Graphs of groups, word computations, and free crossed resolutions, Ph.D.
Thesis, University of Wales, Bangor, 2001.
[30] K.S. Nambooripad and F.J. Pastijn, Amalgamation of regular semigroups, Houston
J. Math. 15 (1989), 249254.
[31] M. Petrich, Inverse semigroups, John Wiley & Sons Publication, 1984.
[32] N. Ruskuc, Presentation for subgroups of monoids, J.Algebra 220 (1999), 365380.
[33] N. Ruskuc, Semigroup presentations, Ph.D. Thesis, University of St Andrews, 1995.
[34] P.V. Silva, B. Steinberg A geometric characterization of automatic monoids, preprint.
[35] J.B. Stephen, Presentations of inverse monoids, J. Pure Appl. Algebra 63 (1990),
81112.
[36] A. Yamamura, HNN extensions of inverse semigroups and applications, Internat. J.
Algebra Comput. 7 (1997), 605624.
[37] A. Yamamura, Locally full HNN extensions of inverse semigroups, J. Austral. Math.
Soc. 70 (2001), 235272.

Inverse Semi Groups

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Inverse Semi Groups

Uploaded by

Copyright:

Available Formats

Automatic S-acts and inverse semigroup presentations

Erzsebet Rita Dombi

= xsx is an inverse of s. Thus, we may deduce that s S is regular if and only if s

V (s) such that ss

. The length of the

will be called a language over A.

((q, a), w),

. We say that a word w = a

(p, w) T. The language L(A) recognized (or accepted) by the

that are recognized by A.

that are recognized by A.

}. Namely, , , a, where a A are regular expressions. If r, r

. Every regular expression

are regular languages.

are regular languages, then K L, K L, K L, KL, K

is a regular language and : X

is a regular language and : X

are regular languages, then (K L)

such that (u, v)

(u(t), v(t)) k for all

(u(t), v(t)) k for all t 1.

UTZENBERGER AUTOMATIC REGULAR SEMIGROUPS 25

= {l L | s l = 0} is a regular language, hence K = L L

is also a regular language

= {u L | s (u x) = 0}. By Lemma 3.1.8, L

is a regular language. It follows

UTZENBERGER AUTOMATIC REGULAR SEMIGROUPS 27

. Clearly L = R and we have that

. Consider the regular language L = u

), where w is a word over X representing s = es. Since (X, L) is an

L is a regular language, moreover we

Y and regular languages K

also generates S and that there exists a surjective

of equal length such that

and we may deduce that

) = (u, v) for some (u, v) K

) = (u, v) for some (u, v) K

V (t). Let w be a word over X

s and let K = wL. Note that s = t w. Clearly K is a regular language

(a, b). If S is a group then is a connected graph and (g, x, h) A()

(u, v) 1 with u, v L then d

(u(t), v(t)) k for all t 1. As in

(a, b) for all a, b S,

(u, v) 1, (u, v L). Then u = x

(a, b) for all a, b S, we have

(s u(t), s v(t)) k holds for all t 1.

(s u(t), s v(t)) k holds

. Applying Proposition 2.1.5

To make notation convenient, we write [r

V (x). Since X is nite, there exists a contant c

V (x) can be represented by a word of length less then or equal to c

V (t). Let h S(s

). Then, by Proposition 2.1.10,

V (xs) = V (t), and so we obtain that

s can be represented by a word of

(s, t) k, then s w = t for some word w X

(s, t) k, then there exists an undirected path

L = {ew eL | (ew) G}.

(u, v) 1 holds. We verify that there exists a constant c N

(u(t), v(t)) c holds for all t 1.

(e u(t), e v(t)) k for all

By Corollary 3.8.5, there exists a constant c

. Our aim is to show that

(s, t). Clearly L is a regular