This paper was converted on www.awesomepapers.org from LaTeX by an anonymous user.
Want to know more? Visit the Converter page.

Closed inverse subsemigroups of graph inverse semigroups

Amal AlAli & N.D. Gilbert School of Mathematical and Computer Sciences
and the Maxwell Institute for the Mathematical Sciences,
Heriot-Watt University, Edinburgh EH14 4AS, U.K.
asa80@hw.ac.uk, N.D.Gilbert@hw.ac.uk
Abstract.

As part of his study of representations of the polycylic monoids, M.V. Lawson described all the closed inverse submonoids of a polycyclic monoid PnP_{n} and classified them up to conjugacy. We show that Lawson’s description can be extended to closed inverse subsemigroups of graph inverse semigroups. We then apply B. Schein’s theory of cosets in inverse semigroups to the closed inverse subsemigroups of graph inverse semigroups: we give necessary and sufficient conditions for a closed inverse subsemigroup of a graph inverse semigroup to have finite index, and determine the value of the index when it is finite.

AMS 2000 Mathematics subject classification: Primary 20M18
AMS 2000 Mathematics subject classification: Secondary 20M30, 05C25

1. Introduction

Graph inverse semigroups were introduced by Ash and Hall [3]: the construction, recalled in detail in section 2.2, associates to any directed graph Γ\Gamma an inverse semigroup 𝒮(Γ)\mathcal{S}(\Gamma) whose elements are pairs of directed paths in Γ\Gamma with the same initial vertex. If Γ\Gamma has a single vertex and nn edges with n>1n>1, then 𝒮(Γ)\mathcal{S}(\Gamma) is the polycyclic monoid PnP_{n} as defined by Nivat and Perrot [12]: if n=1n=1 then 𝒮(Γ)\mathcal{S}(\Gamma) is the bicyclic monoid BB with an adjoined zero. Ash and Hall give necessary and sufficient conditions on the structure of Γ\Gamma for 𝒮(Γ)\mathcal{S}(\Gamma) to be congruence-free, and they use graph inverse semigroups to study the realisation of finite posets as the posets of 𝒥{\mathcal{J}}–classes in finite semigroups. The structure of graph inverse semigroups as HNN extensions of inverse semigroups with zero was presented in [4, section 5]. For more recent work on the structure of graph inverse semigroups, we refer to [6, 10, 11]. Connections between graph inverse semigroups and graph CC^{*}–algebras have been fruitfully studied in [13].

As part of his study of representations of the polycylic monoids, Lawson [8] described all the closed inverse submonoids of a polycyclic monoid PnP_{n} and classified them up to conjugacy. We show in section 3 that Lawson’s description can be extended to closed inverse subsemigroups of graph inverse semigroups. As in Lawson’s study, there are three types: finite chains of idempotents, infinite chains of idempotents, and closed inverse subsemigroups of cycle type that are generated (as closed inverse subsemigroups) by a single non-idempotent element. In section 4 we apply Schein’s theory of cosets in inverse subgroups [14] to the closed inverse subsemigroups of graph inverse semigroups as classified in section 3: we give necessary and sufficient conditions for a closed inverse subsemigroup LL of 𝒮(Γ)\mathcal{S}(\Gamma) to have finite index, and determine the value of the index when it is finite.

2. Preliminaries

2.1. Cosets

Let SS be an inverse semigroup with semilattice of idempotents E(S)E(S). We recall that the natural partial order on SS is defined by

stthere existseE(S)such thats=et.s\leqslant t\Longleftrightarrow\text{there exists}\;e\in E(S)\;\text{such that}\;s=et\,.

A subset ASA\subseteq S is closed if, whenever aAa\in A and asa\leqslant s, then sAs\in A. The closure BB^{\uparrow} of a subset BSB\subseteq S is defined as

B={sS:sbfor somebB}.B^{\uparrow}=\{s\in S:s\geqslant b\;\text{for some}\;b\in B\}\,.

A subset LL of SS is full if E(S)LE(S)\subseteq L.

Let LL be a closed inverse subsemigroup of SS, and let tSt\in S with tt1Ltt^{-1}\in L. Then the subset

(Lt)={sS:there exists xL withsxt}(Lt)^{\uparrow}=\{s\in S:\text{there exists $x\in L$ with}\;s\geqslant xt\}

is a (right) coset of LL in SS. For the basic theory of such cosets we refer to [14]: the essential facts that we require are contained in the following result.

Proposition 2.1.

[14, Proposition 6.] Let LL be a closed inverse subsemigroup of SS.

  1. (a)

    Suppose that CC is a coset of LL. Then (CC1)=L(CC^{-1})^{\uparrow}=L.

  2. (b)

    If tCt\in C then tt1Ltt^{-1}\in L and C=(Lt)C=(Lt)^{\uparrow}. Hence two cosets of LL are either disjoint or they coincide.

  3. (c)

    Two elements a,bSa,b\in S belong to the same coset CC of LL if and only if ab1Lab^{-1}\in L.

We note that the cosets of LL partition SS if and only if LL is full in SS. The cardinality of the set of cosets of LL in SS is the index of LL in SS, denoted by [S:L][S:L].

The closed inverse submonoids of free inverse monoids were completely described by Margolis and Meakin in [9]. For other related work on inverse subsemigroups of finite index, see [2] and the first author’s PhD thesis [1].

2.2. Graph inverse semigroups

Let Γ\Gamma be a finite directed graph with vertex set V(Γ)V(\Gamma) and edge set E(Γ)E(\Gamma). Let 𝒫(Γ)\vec{{\mathscr{P}}}(\Gamma) be the path category of Γ\Gamma, with source and target maps 𝐝{\mathbf{d}} and 𝐫{\mathbf{r}}. We note that 𝒫(Γ)\vec{{\mathscr{P}}}(\Gamma) admits empty (or length zero) paths that consists of a single vertex. The graph inverse semigroup 𝒮(Γ)\mathcal{S}(\Gamma) of Γ\Gamma has underlying set

{(v,w):v,w𝒫(Γ),𝐝(v)=𝐝(w)}{0}\{(v,w):v,w\in\vec{{\mathscr{P}}}(\Gamma)\,,{\mathbf{d}}(v)={\mathbf{d}}(w)\}\cup\{0\}

equipped with the binary operation

(t,u)(v,w)={(t,pw)if u=pv in 𝒫(Γ),(pt,w)if v=pu in 𝒫(Γ),0otherwise.(t,u)(v,w)=\begin{cases}(t,pw)&\text{if $u=pv$ in $\vec{{\mathscr{P}}}(\Gamma)$},\\ (pt,w)&\text{if $v=pu$ in $\vec{{\mathscr{P}}}(\Gamma)$},\\ 0&\text{otherwise.}\end{cases}

This composition is illustrated in the following diagrams:

[Uncaptioned image]

The inverse of (v,w)(v,w) is given by (v,w)1=(w,v)(v,w)^{-1}=(w,v). The idempotents of 𝒮(G)\mathcal{S}(G) are the pairs (u,u)(u,u) and 0: if we identify E(𝒮(G))E(\mathcal{S}(G)) with 𝒫(G){0}\vec{{\mathscr{P}}}(G)\cup\{0\}, then 𝒫(G){0}\vec{{\mathscr{P}}}(G)\cup\{0\} becomes a semilattice with ordering given by

(2.1) uvif and only ifvis a suffix ofuu\leqslant v\quad\text{if and only if}\;v\;\text{is a suffix of}\;u

and composition (meet)

(2.2) uv={uif v is a suffix of u,vif u is a suffix of v,0otherwiseu\wedge v=\begin{cases}u&\text{if $v$ is a suffix of $u$},\\ v&\text{if $u$ is a suffix of $v$},\\ 0&\text{otherwise}\end{cases}

Hence ubu\wedge b is non-zero if and only if one of u,vu,v is a suffix of the other: in this case we say that u,vu,v are suffix comparable.

The natural partial order on non-zero elements of 𝒮(G)\mathcal{S}(G) is then given by (t,u)(v,w)(t,u)\leqslant(v,w) if and only if there exists a path pΓp\in\Gamma such that t=pvt=pv and u=pwu=pw: that is, we descend in the natural partial order from (v,w)(v,w) by prepending the same prefix to each of vv and ww., and ascend from (v,w)(v,w) by deleting an identical prefix from each of vv and ww. Recall that an inverse semigroup SS with zero 0S0\in S is said to be EE^{*}unitary, if whenever eE(S),e0e\in E(S),e\neq 0 and sSs\in S with ses\geqslant e then sE(S)s\in E(S). It is clear that graph inverse semigroups are EE^{*}–unitary. For further structural results about graph inverse semigroups, we refer to [6, 10]. Graph inverse semigroups as topological inverse semigroups have been recently studied in [11].

3. Closed inverse subsemigroups of graph inverse semigroups

Our first result generalizes – and closely follows – Lawson’s classification [8, Theorem 4.3] of closed inverse submonoids of the polycyclic monoids PnP_{n} to the closed inverse subsemigroups of graph inverse semigroups 𝒮(Γ)\mathcal{S}(\Gamma). Given Lawson’s insights, the generalization is largely routine, but it is perhaps slightly surprising that the classification extends from bouquets of circles (giving the polycyclic monoids as graph inverse semigroups) to arbitrary finite directed graphs, and so we have presented it in detail. Our notational conventions also differ slightly from those in [8].

Theorem 3.1.

In a graph inverse semigroup 𝒮(Γ)\mathcal{S}(\Gamma) there are three types of proper closed inverse subsemigroups LL:

  1. (a)

    Finite chain type: LL consists of a finite chain of idempotents.

  2. (b)

    Infinite chain type: LL consists of an infinite chain of idempotents.

  3. (c)

    Cycle type: LL has the form

    L=Lp,d={(vprd,vpsd):r,s0withva suffix ofp}{(q,q):qa suffix of d},L=L_{p,d}=\{(vp^{r}d,vp^{s}d):r,s\geqslant 0\;\text{with}\;v\;\text{a suffix of}\;p\}\cup\{(q,q):q\;\text{a suffix of }d\},

    where pp is a directed circuit in Γ\Gamma, dd is a directed path in Γ\Gamma starting at the initial point of pp, and where p,dp,d do not share a non-trivial prefix. In this case, LL is the smallest closed inverse subsemigroup of 𝒮(Γ)\mathcal{S}(\Gamma) containing (d,pd)(d,pd).

Proof.

It is easy to see that closed chains of idempotents are indeed closed inverse subsemigroups. For the cycle type, if L=Lp,dL=L_{p,d} then any two elements are suffix comparable, and we have

(q,q)(q,q)\displaystyle(q,q)(q^{\prime},q^{\prime}) =(qq,qq)for suffixes q,q of d,\displaystyle=(q\wedge q^{\prime},q\wedge q^{\prime})\;\text{for suffixes $q,q^{\prime}$ of $d$},
(q,q)(vprd,vpsd)\displaystyle(q,q)(vp^{r}d,vp^{s}d) =(vprd,vpsd)=(vprd,vpsd)(q,q).\displaystyle=(vp^{r}d,vp^{s}d)=(vp^{r}d,vp^{s}d)(q,q).

Now consider (vprd,vpsd)(wpjd,wpkd)(vp^{r}d,vp^{s}d)(wp^{j}d,wp^{k}d): write p=v0vp=v_{0}v and suppose that s<js<j. Then wpjd=wpjs1v0vpsdwp^{j}d=wp^{j-s-1}v_{0}vp^{s}d and so

(vprd,vpsd)(wpjd,wpkd)=(wpjs1v0vprd,wpkd)=(wpjs+rd,wpkd)L,(vp^{r}d,vp^{s}d)(wp^{j}d,wp^{k}d)=(wp^{j-s-1}v_{0}vp^{r}d,wp^{k}d)=(wp^{j-s+r}d,wp^{k}d)\in L\,,

and a similar calculation applies if s>js>j. If s=js=j and vv is a suffix of ww, say w=v1vw=v_{1}v, then

(vprd,vpsd)(wpsd,wpkd)\displaystyle(vp^{r}d,vp^{s}d)(wp^{s}d,wp^{k}d) =(vprd,vpsd)(v1vpsd,wpkd)\displaystyle=(vp^{r}d,vp^{s}d)(v_{1}vp^{s}d,wp^{k}d)
=(v1vprd,wpkd)\displaystyle=(v_{1}vp^{r}d,wp^{k}d)
=(wprd,wpkd)L\displaystyle=(wp^{r}d,wp^{k}d)\in L

and a similar calculation applies if s=js=j and ww is a suffix of vv. Hence LL is a subsemigroup of 𝒮(Γ)\mathcal{S}(\Gamma), and since the inverse of an element of LL is clearly also in LL, we deduce that LL is an inverse subsemigroup of 𝒮(Γ)\mathcal{S}(\Gamma). Since we ascend in the natural partial order in LL by deleting identical prefixes from the paths vprdvp^{r}d and vpsdvp^{s}d, or from a given suffix of dd, it is also clear that LL is closed.

If FF is a closed inverse subsemigroup of 𝒮(Γ)\mathcal{S}(\Gamma) and contains (d,pd)(d,pd), then for any m,n0m,n\geqslant 0 we have (pd,d)m(d,pd)n=(pmd,d)(d,pnd)=(pmd,pnd)F(pd,d)^{m}(d,pd)^{n}=(p^{m}d,d)(d,p^{n}d)=(p^{m}d,p^{n}d)\in F. Ascending in the natural partial order, we may obtain any element of Lp,dL_{p,d}, and so Lp,dFL_{p,d}\subseteq F.

Let LL be a closed inverse subsemigroup of 𝒮(Γ)\mathcal{S}(\Gamma). If ww and ww^{\prime} are paths occurring in elements of LL and are not suffix comparable, then the product of the idempotents (w,w)(w,w) and (w,w)(w^{\prime},w^{\prime}) in LL is equal to 0, and so 0L0\in L and by closure L=𝒮(Γ)L=\mathcal{S}(\Gamma). Hence if LL is proper, any two paths occurring in elements of LL are suffix comparable and hence have the same terminal vertex. By definition, if (u,v)𝒮(Γ)(u,v)\in\mathcal{S}(\Gamma) then u,vu,v have the same initial vertex: hence if (u,v)L(u,v)\in L then u,vu,v have the same initial and the same terminal vertex in Γ\Gamma. Suffix comparability then ensures that any proper closed inverse subsemigroup of 𝒮(Γ)\mathcal{S}(\Gamma) consisting entirely of idempotents is either a finite or an infinite chain. We note that in the second case, in order to obtain directed paths of arbitrary length, Γ\Gamma must contain a directed circuit.

We shall now describe those closed inverse subsemigroups of 𝒮(Γ)\mathcal{S}(\Gamma) which contain non-idempotent elements. Suppose that LE(L)L\neq E(L) is a closed inverse subsemigroup of 𝒮(Γ)\mathcal{S}(\Gamma). Then there exists (u,v)L,(u,v)\in L, with uvu\neq v, and we may assume that the path uu is shorter than the path vv. Hence uu is a suffix of vv and so v=puv=pu for some path pp. Since uu and vv have the same initial and terminal vertices, pp must be a directed circuit in Γ\Gamma. If pp and uu share a common prefix, with p=ap1p=ap_{1} and u=au1u=au_{1} then

(u1,p1u)(au1,ap1u)=(u,pu)(u_{1},p_{1}u)\geqslant(au_{1},ap_{1}u)=(u,pu)

and so by closure, (u1,p1u)L(u_{1},p_{1}u)\in L.

Amongst the non-idempotent elements (u,pu)L(u,pu)\in L, choose u=du=d to have smallest possible length, and then having chosen dd, choose pp to be a non-empty directed circuit of smallest possible length. Then d,pd,p do not share a non-trivial prefix. Now for any m0m\geqslant 0 we have (pmd,pmd)E(L)(p^{m}d,p^{m}d)\in E(L) and so, if (w1,w2)L(w_{1},w_{2})\in L, each wiw_{i} is a suffix of some directed path pmidp^{m_{i}}d. Since LL is closed, every suffix of dd is in LL and by minimality of |d||d|, every element of LL that contains a suffix of dd is an idempotent (q,q)(q,q). Hence if |wi|<|d||w_{i}|<|d| we have w1=w2w_{1}=w_{2}. So we may now assume that for i=1,2i=1,2 we have |wi||d||w_{i}|\geqslant|d|, and so w1=uprdw_{1}=up^{r}d, w2=vpsdw_{2}=vp^{s}d for some r,s0r,s\geqslant 0 and suffixes u,vu,v of pp. ∎

From the result of the previous Theorem, we may immediately conclude the following:

Corollary 3.2.

If the graph Γ\Gamma contains no directed circuit, then every proper closed inverse subsemigroup of 𝒮(Γ)\mathcal{S}(\Gamma) is a chain of idempotents.

Our next result, based on [8, Theorem 4.4] which treats the polycyclic monoids, classifies the closed inverse subsemigroups of a graph inverse semigroup up to conjugacy. We begin with the following definitions

Definition 3.3.

Let L=(u,u)L=(u,u)^{\uparrow} be a closed inverse subsemigroup of finite chain type in a graph inverse semigroup 𝒮(Γ)\mathcal{S}(\Gamma). We call the initial vertex of the directed path uu the root of LL.

Adapting ideas of [7, Section 1.3] from words to paths in Γ\Gamma:

Definition 3.4.

Two paths p,qp,q in Γ\Gamma are conjugate if there are paths u,vu,v in Γ\Gamma such that p=uvp=uv and q=vuq=vu. Equivalently, (see [7, Proposition 1.3.4]) there exists a path ww in Γ\Gamma such that wp=qwwp=qw. Conjugate paths must be directed circuits in Γ\Gamma, and conjugacy amounts to the selection of an alternative initial edge.

The following Lemma is due to Lawson and is extracted from the proof of [8, Theorem 4.4].

Lemma 3.5.

Let SS be an EE^{*}–unitary inverse semigroup. If HH and KK are conjugate closed inverse subsemigroups of SS with HSKH\neq S\neq K and HE(S)H\subseteq E(S) then KE(S)K\subseteq E(S). Moreover, if HH has a minimum idempotent, then so does KK.

Proof.

There exists sSs\in S with s1HsKandsKs1Hs^{-1}Hs\subseteq K\;\text{and}\;sKs^{-1}\subseteq H. Let kKk\in K: then k0k\neq 0 and sks1Hsks^{-1}\in H and so sks1E(S)sks^{-1}\in E(S). It follows that s1(sks1)s=(s1s)k(s1s)E(S)s^{-1}(sks^{-1})s=(s^{-1}s)k(s^{-1}s)\in E(S) and (s1s)k(s1s)k(s^{-1}s)k(s^{-1}s)\leqslant k. Since SS is EE^{*}–unitary, we deduce that kE(S)k\in E(S).

Now suppose that mHE(S)m\in H\subseteq E(S) is the minimum idempotent and that eKe\in K. Then mses1m\leqslant ses^{-1} and so

s1mss1ses1s=es1ses^{-1}ms\leqslant s^{-1}ses^{-1}s=es^{-1}s\leqslant e

and so s1mss^{-1}ms is a minimum idempotent in KK. ∎

Theorem 3.6.
  1. (a)

    Let LL be a closed inverse subsemigroup of 𝒮(Γ)\mathcal{S}(\Gamma) of finite chain type. Then all closed inverse subsemigroups conjugate to LL are of finite chain type. Two closed inverse subsemigroups L=(u,u)L=(u,u)^{\uparrow} and K=(v,v)K=(v,v)^{\uparrow} are conjugate in 𝒮(Γ)\mathcal{S}(\Gamma) if and only if they have the same root.

  2. (b)

    Let LL be a closed inverse subsemigroup of 𝒮(Γ)\mathcal{S}(\Gamma) of infinite chain type. Then all closed inverse subsemigroups conjugate to LL are also of infinite chain type. Two closed inverse subsemigroups of infinite chain type are conjugate if and only if there are idempotents (s,s)L(s,s)\in L and (t,t)K(t,t)\in K such that for all paths pp in Γ,\Gamma, we have that (ps,ps)L(ps,ps)\in L if and only if (pt,pt)K(pt,pt)\in K.

  3. (c)

    Let LL be a closed inverse subsemigroup of 𝒮(Γ)\mathcal{S}(\Gamma) of cycle type. The only closed inverse subsemigroups conjugate to LL are of cycle type. Moreover, Lp,dL_{p,d} is conjugate to Lq,kL_{q,k} if and only if pp and qq are conjugate directed circuits in Γ\Gamma.

Proof.

(a) It follows from Lemma 3.5 that if LL has finite chain type then so does every closed inverse subsemigroup conjugate to LL .

Suppose that LL and KK have the same root x0V(Γ)x_{0}\in V(\Gamma). Then (u,v)𝒮(Γ)(u,v)\in\mathcal{S}(\Gamma), and for any suffix ww of uu we have

(v,u)(w,w)(u,v)=(v,v)K.(v,u)(w,w)(u,v)=(v,v)\in K\,.

Similarly, for any suffix tt of vv, (u,v)(t,t)(v,u)=(u,u)L(u,v)(t,t)(v,u)=(u,u)\in L. Hence LL and KK are conjugate.

Conversely, suppose that LL and KK are conjugate, with conjugating element (p,q)𝒮(Γ)(p,q)\in\mathcal{S}(\Gamma), so that for any suffixes ww of uu and tt of vv we have

(q,p)(w,w)(p,q)Kand(p,q)(t,t)(q,p)L.(q,p)(w,w)(p,q)\in K\;\text{and}\;(p,q)(t,t)(q,p)\in L\,.

Then (q,p)(𝐫(u),𝐫(u))(p,q)K(q,p)({\mathbf{r}}(u),{\mathbf{r}}(u))(p,q)\in K, so that pp and 𝐫(u){\mathbf{r}}(u) are suffix-comparable: hence pp also ends at 𝐫(u){\mathbf{r}}(u), and (q,p)(𝐫(u),𝐫(u))(p,q)=(q,q)K(q,p)({\mathbf{r}}(u),{\mathbf{r}}(u))(p,q)=(q,q)\in K. Therefore qq is a suffix of vv. Similarly, pp is a suffix of uu.

Let v=v1qv=v_{1}q: then

(p,q)(v,v)(q,p)=(p,q)(v1q,v1q)(q,p)=(v1p,v1p)L(p,q)(v,v)(q,p)=(p,q)(v_{1}q,v_{1}q)(q,p)=(v_{1}p,v_{1}p)\in L

and so v1pv_{1}p is a suffix of uu. Let u=u0v1pu=u_{0}v_{1}p: then

(q,p)(u,u)(p,q)=(q,p)(u0v1p,u0v1p)(p,q)=(u0v1q,u0v1q)K(q,p)(u,u)(p,q)=(q,p)(u_{0}v_{1}p,u_{0}v_{1}p)(p,q)=(u_{0}v_{1}q,u_{0}v_{1}q)\in K

and so u0v1qu_{0}v_{1}q is a suffix of vv. But v=v1qv=v_{1}q and so u0u_{0} is a vertex (namely the root of LL), and u=v1pu=v_{1}p. Hence uu and vv have the same initial vertex, and so LL and KK have the same root.

(b) By Lemma 3.5 any closed inverse subsemigroup KK conjugate to LL must be of chain type, and by part (a) KK must be infinite. Suppose that (t,s)L(s,t)Kand(s,t)K(t,s)L(t,s)L(s,t)\subseteq K\;\text{and}\;(s,t)K(t,s)\subseteq L. Since 0K0\not\in K we have, for all (u,u)L(u,u)\in L, that ss is suffix comparable with uu and similarly for all (v,v)K(v,v)\in K, that tt is suffix comparable with vv. If we consider uu with |u||s||u|\geqslant|s| then ss must be a suffix of uu and by closure of LL we have (s,s)L(s,s)\in L. Similarly (t,t)K(t,t)\in K. Suppose that (ps,ps)L(ps,ps)\in L. Then (t,s)(ps,ps)(s,t)=(pt,pt)K(t,s)(ps,ps)(s,t)=(pt,pt)\in K and similarly if (pt,pt)K(pt,pt)\in K then (ps,ps)L(ps,ps)\in L.

Conversely, if ss and tt exist as in the Theorem and (w,w)L(w,w)\in L then ss is suffix comparable with ww.
If ww is a suffix of ss, with s=hws=hw, then

(t,s)(w,w)(s,t)=(t,hw)(w,w)(hw,t)=(t,t)K(t,s)(w,w)(s,t)=(t,hw)(w,w)(hw,t)=(t,t)\in K

and if ss is a suffix of ww with w=psw=ps then (ps,ps)L(ps,ps)\in L and so

(t,s)(w,w)(s,t)=(t,s)(ps,ps)(s,t)=(pt,pt)K.(t,s)(w,w)(s,t)=(t,s)(ps,ps)(s,t)=(pt,pt)\in K\,.

Similarly (s,t)K(t,s)L(s,t)K(t,s)\subseteq L, and LL and KK are conjugate.

(c) By parts (a) and (b), any closed inverse subsemigroup of 𝒮(Γ)\mathcal{S}(\Gamma) that is conjugate to Lp,dL_{p,d} must be of cycle type. Suppose that the closed inverse subsemigroups Lp,dL_{p,d} and Lq,kL_{q,k} are conjugate in 𝒮(Γ)\mathcal{S}(\Gamma), and so there exists (s,t)𝒮(Γ)(s,t)\in\mathcal{S}(\Gamma) such that

(3.1) (t,s)Lp,d(s,t)Lq,k\displaystyle(t,s)L_{p,d}(s,t)\subseteq L_{q,k}
(3.2) (s,t)Lq,k(t,s)Lp,d.\displaystyle(s,t)L_{q,k}(t,s)\subseteq L_{p,d}.

Since Lq,kL_{q,k} is closed and Lp,dL_{p,d} is the smallest closed inverse subsemigroup of 𝒮(Γ)\mathcal{S}(\Gamma) containing (d,pd)(d,pd), then (3.1) is equivalent to (t,s)(d,pd)(s,t)Lq,k(t,s)\,(d,pd)\,(s,t)\in L_{q,k}. Also, since 0Lq,k0\not\in L_{q,k} we must have ss\, suffix-comparable with uu and vv whenever (u,v)(u,v) is an element of Lp,dL_{p,d}\,. Hence (s,s)Lp,d(s,s)\in L_{p,d}\,, and similarly (t,t)Lq,k(t,t)\in L_{q,k}.

First suppose that s=upads=up^{a}d\, and t=vqbkt=vq^{b}k\, for some a,b0a,b\geqslant 0, where uu is a suffix of pp and vv is a suffix of qq. Write p=hup=hu: then

(t,s)(d,pd)(s,t)\displaystyle(t,s)\,(d,pd)\,(s,t) =(vqbk,upad)(d,pd)(upad,vqbk)\displaystyle=(vq^{b}k,up^{a}d)\,(d,pd)\,(up^{a}d,vq^{b}k)
=(vqbk,up)(u,vqbk)\displaystyle=(vq^{b}k,up)\,(u,vq^{b}k)
=(vqbk,uhvqbk)Lq,k.\displaystyle=(vq^{b}k,uhvq^{b}k)\in L_{q,k}.

It follows that uhvqbk=vqmkuhvq^{b}k=vq^{m}k\, for some m0m\geqslant 0. Comparing lengths of these directed paths, we see that m>bm>b, and then after cancellation we obtain uhv=vqmbuhv=vq^{m-b}. Hence uhuh is conjugate to some power of qq, and since uhuh is a conjugate of pp, we conclude that pp is conjugate to some power of qq.

Now suppose that ss is a suffix of dd and write d=csd=cs. With tt as before, we now obtain

(t,s)(d,pd)(s,t)=(vqbk,s)(cs,pcs)(s,vqbk)=(cvqbk,pcvqbk)Lq,k.(t,s)\,(d,pd)\,(s,t)=(vq^{b}k,s)\,(cs,pcs)\,(s,vq^{b}k)=(cvq^{b}k,pcvq^{b}k)\in L_{q,k}.

It follows that pcvqbk=cvqmkpcvq^{b}k=cvq^{m}k for some m0m\geqslant 0. Again m>bm>b and after cancellation we obtain pcv=cvqmbpcv=cvq^{m-b}. Here we see directly that pp is conjugate to a power of qq.

Now suppose that ss is a suffix of dd and write d=csd=cs, and that tt is a suffix of kk and write k=jtk=jt. We now obtain

(t,s)(d,pd)(s,t)=(t,s)(cs,pcs)(s,t)=(ct,pct)Lq,k.(t,s)\,(d,pd)\,(s,t)=(t,s)\,(cs,pcs)\,(s,t)=(ct,pct)\in L_{q,k}.

Since by assumption pp\, is not the empty path, we have ct=wqakct=wq^{a}k\, and pct=wqbkpct=wq^{b}k\, for some suffix ww of qq and some a,b0a,b\geqslant 0. Again comparing lengths, we see that b>ab>a, and then pct=pwqak=wqbkpct=pwq^{a}k=wq^{b}k. After cancellation we obtain pw=wqbapw=wq^{b-a} and again pp is conjugate to a power of qq.

Hence for each possibility of ss, we deduce from (3.1) that pp\, is conjugate to some power of qq\,. Using equation (3.2) we deduce similarly that qq is conjugate to a power of pp\,. Again comparing lengths, we conclude that pp and qq are conjugate.

Conversely, if p,qp,q are conjugate, suppose that p=uvp=uv and q=vuq=vu. Then it is esay to check that setting s=ks=k and t=vdt=vd furnishes a pair (s,t)(s,t) satisfying (3.1) and (3.2). ∎

Remark 3.7.

For the polycyclic monoids PnP_{n} (n2)(n\geqslant 2), we obtain the classification of closed inverse submonoids up to conjugacy given in [8, Theorem 4.4] by applying Theorem 3.6 to the graph Γ\Gamma with one vertex and nn loops labelled a1,,an.a_{1},\dots,a_{n}. For the case n=1n=1, with a single loop labelled aa, we obtain the graph inverse semigroup 𝒮(Γ)=B{0}\mathcal{S}(\Gamma)=B\cup\{0\}, where BB is the bicyclic monoid. A proper closed inverse subsemigroup LL of 𝒮(Γ)\mathcal{S}(\Gamma) cannot contain {0}\{0\} and so is a proper closed inverse subsemigroup of BB. If LE(B)L\subseteq E(B) then by Theorem 3.1, LL is either E(B)E(B) itself or is of finite chain type, and part (a) of Theorem 3.6, then shows that all closed inverse subsemigroup of finite chain type in BB are then conjugate. By Theorem 3.1, a closed inverse subsemigroup LL of BB of cycle type consists of elements of the form (qpr,qps)(qp^{r},qp^{s}) with r,s0r,s\geqslant 0, and where where p=amp=a^{m} for some m1m\geqslant 1 and q=akq=a^{k} for some kk with 0km10\leqslant k\leqslant m-1: that is, elements of the form (arm+k,asm+k)(a^{rm+k},a^{sm+k}). The subsemigroup LL is therefore isomorphic to the fundamental simple inverse ω\omega–semigroup BmB_{m}, discussed in [5, section 5.7].

4. The index of closed inverse subsemigroups in graph inverse semigroups

We first discuss the index of closed inverse subsemigroups of finite and infinite chain type in 𝒮(Γ)\mathcal{S}(\Gamma). For a fixed path ww in Γ\Gamma and a vertex vv of ww, we define Nv,wΓN^{\Gamma}_{v,w} to be the number of distinct directed paths in Γ\Gamma whose initial vertex is vv but whose first edge is not in ww. The empty path vv is one such path.

Theorem 4.1.
  1. (a)

    Let LL be a closed inverse subsemigroup of finite chain type in 𝒮(Γ)\mathcal{S}(\Gamma), with minimal element (w,w)(w,w). Then LL has infinite index in 𝒮(Γ)\mathcal{S}(\Gamma) if and only if there exists a non-empty directed circuit cc\, in Γ\Gamma and a (possibly empty) directed path gg from some vertex v0v_{0} of ww to a vertex of cc\, and with gg having no edge in common with cwc\cup w.

  2. (b)

    If L=(w,w)L=(w,w)^{\uparrow} and LL has finite index in 𝒮(Γ)\mathcal{S}(\Gamma) then

    [𝒮(Γ):L]=vV(w)Nv,wΓ.[\mathcal{S}(\Gamma):L]=\sum_{v\in V(w)}N^{\Gamma}_{v,w}\,.
  3. (c)

    Let LL be a closed inverse subsemigroup of infinite chain type in 𝒮(Γ)\mathcal{S}(\Gamma). Then LL has infinite index in 𝒮(Γ)\mathcal{S}(\Gamma).

Proof.

(a) Let LL have finite chain type. A coset representative of LL has the form (q,u)(q,u) where qq is some suffix of ww, and (q,u)(q,u) have the same initial vertex. If LL has infinite index, then there are infinitely many distinct choices for (q,u)(q,u) and since Γ\Gamma is finite, there must be a directed circuit in Γ\Gamma as described.

Conversely, suppose that g,cg,c\, exist. Let qq\, be the suffix of ww\, that has initial vertex v0v_{0}\,. Then qLq\in L and so for any k0k\geqslant 0\, the coset Ck=L(q,gck)C_{k}=L(q,gc^{k})^{\uparrow} exists. Now for k>l,k>l,\, we have if gg is non-empty, that

(q,gck)(q,gcl)1=(q,gck)(gcl,q)=0L(q,gc^{k})(q,gc^{l})^{-1}=(q,gc^{k})(gc^{l},q)=0\not\in L

and so by part (c) of Proposition 2.1, the cosets CkC_{k}\, and ClC_{l}\, are distinct. If gg is empty then we have

(q,ck)(q,cl)1=(q,ck)(cl,q)=(q,cklq)L(q,c^{k})\,(q,c^{l})^{-1}=(q,c^{k})\,(c^{l},q)=(q,c^{k-l}q)\not\in L

and again the cosets CkC_{k}\, and ClC_{l}\, are distinct.

(b) By part (a) there are no directed cycles accessible from any vertex of ww, and so Nv,wΓN^{\Gamma}_{v,w} is finite for each vertex vv of ww. A coset representative of LL has the form (s,t)(s,t) where ss is a suffix of ww. Suppose that two such elements, (s1,t1)(s_{1},t_{1}) and (s2,t2)(s_{2},t_{2}), represent the same coset. Then (s1,t1)(t2,s2)L(s_{1},t_{1})(t_{2},s_{2})\in L: in particular the product is non-zero and so t1,t2t_{1},t_{2} are suffix comparable. We may assume that t2=ht1t_{2}=ht_{1}: then (s1,t1)(ht1,s2)=(hs1,s2)(s_{1},t_{1})(ht_{1},s_{2})=(hs_{1},s_{2}) and this is in LL if and only if s2=hs1s_{2}=hs_{1}. Therefore L(s1,t1)=L(s2,t2)L(s_{1},t_{1})^{\uparrow}=L(s_{2},t_{2})^{\uparrow} if and only if (s2,t2)=(hs1,ht1)(s_{2},t_{2})=(hs_{1},ht_{1}), and so the distinct coset representatives are the pairs (s,t)(s,t) where ss is a suffix of ww, ss and tt have the same initial vertex, but do not share the same initial edge. It follows that the number of distinct cosets is vV(w)Nv,wΓ\sum_{v\in V(w)}N^{\Gamma}_{v,w}, and LL itself is represented by (𝐫(w),𝐫(w))({\mathbf{r}}(w),{\mathbf{r}}(w)).

(c) If LL has infinite chain type then the elements of LL comprise the idempotents determined by an infinite sequence of directed paths in Γ,\Gamma,\, each of which is a suffix of the other. Eventually then, we find a path cqcq where cc is a directed circuit, and (cq,cq)L(cq,cq)\in L. Then for each k0,k\geqslant 0,\, the element (q,ckq)(q,c^{k}q) represents a coset Ck=L(q,ckq).C_{k}=L(q,c^{k}q)^{\uparrow}\,.
Now for k>l,k>l,\,

(q,ckq)(q,clq)1=(q,ckq)(clq,q)=(q,cklq)L(q,c^{k}q)\,(q,c^{l}q)^{-1}=(q,c^{k}q)\,(c^{l}q,q)=(q,c^{k-l}q)\not\in L

and the cosets CkC_{k}\, and ClC_{l}\, are distinct. ∎

Example 4.2.

We illustrate the index computation in part (b) of Theorem 4.1 with Γ\Gamma equal to the finite chain with nn edges e1,,ene_{1},\dotsc,e_{n} and n+1n+1 vertices v0,v1,,vnv_{0},v_{1},\dotsc,v_{n}:

vn\textstyle{v_{n}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}en\scriptstyle{e_{n}}vn1\textstyle{v_{n-1}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}en1\scriptstyle{e_{n-1}}\textstyle{\dotsc\ignorespaces\ignorespaces\ignorespaces\ignorespaces}v1\textstyle{v_{1}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}e1\scriptstyle{e_{1}}v0\textstyle{v_{0}}

Here 𝒮(Γ)\mathcal{S}(\Gamma) is finite, and every closed inverse subsemigroup is of finite cycle type and has finite index. The number of paths in Γ\Gamma with initial vertex vjv_{j} is j+1j+1, and so

|𝒮(Γ)|=j=0n(j+1)2=j=1n+1j2=16(n+1)(n+2)(2n+3).|\mathcal{S}(\Gamma)|=\sum_{j=0}^{n}(j+1)^{2}=\sum_{j=1}^{n+1}j^{2}=\frac{1}{6}(n+1)(n+2)(2n+3)\,.

We let ww be the path ene1e_{n}\dotsm e_{1} and L=(w,w)L=(w,w)^{\uparrow}. Since ww has n+1n+1 suffixes, we have |L|=n+1|L|=n+1. An element (s,t)(s,t) lies in a coset of LL if and only if ss is a suffix of ww and 𝐝(s)=𝐝(t){\mathbf{d}}(s)={\mathbf{d}}(t): hence the total number of elements in all the cosets of LL is j=0n(j+1)=j=1n+1j=12(n+1)(n+2)\sum_{j=0}^{n}(j+1)=\sum_{j=1}^{n+1}j=\frac{1}{2}(n+1)(n+2).

Now Nvi,wΓ=1N^{\Gamma}_{v_{i},w}=1 since only the length zero path at viv_{i} is counted, and so [𝒮(Γ):L]=n+1[\mathcal{S}(\Gamma):L]=n+1.

Let qiq_{i} be the path eie1e_{i}\dotsc e_{1}, so that qn=wq_{n}=w, and set q0=v0q_{0}=v_{0}. The n+1n+1 cosets are then represented by the elements (qi,vi)(q_{i},v_{i}), 0in0\leqslant i\leqslant n, and

L(qi,vi)\displaystyle L(q_{i},v_{i})^{\uparrow} ={(qk,qk)(qi,vi):0kn}\displaystyle=\{(q_{k},q_{k})(q_{i},v_{i}):0\leqslant k\leqslant n\}^{\uparrow}
=({(qk,ekei+1:i<kn}{(qi,vi)})\displaystyle=(\{(q_{k},e_{k}\dotsm e_{i+1}:i<k\leqslant n\}\cup\{(q_{i},v_{i})\})^{\uparrow}
={(qn,enei+1),,(qi+1,ei+1),(qi,vi)}\displaystyle=\{(q_{n},e_{n}\dotsc e_{i+1}),\dotsc,(q_{i+1},e_{i+1}),(q_{i},v_{i})\}

and so |L(qi,vi)|=ni+1|L(q_{i},v_{i})^{\uparrow}|=n-i+1. Counting the total number of elements in all the cosets of LL we obtain

i=0n(ni+1)=j=1n+112(n+1)(n+2)\sum_{i=0}^{n}(n-i+1)=\sum_{j=1}^{n+1}\,\frac{1}{2}(n+1)(n+2)

as before.

We now discuss the closed inverse subsemigroups of cycle type.

Theorem 4.3.

A closed inverse subsemigroup Lp,dL_{p,d} of cycle type in 𝒮(Γ)\mathcal{S}(\Gamma), such that pp is a circuit with at least two distinct edges, has infinite index in 𝒮(Γ)\mathcal{S}(\Gamma).

Proof.

Write p=uvp=uv where each of u,vu,v is non-empty and one contains an edge not in the other. Let cc be the conjugate circuit vuvu. Then for k1,k\geqslant 1,\, the element (vd,ck)𝒮(Γ)(vd,c^{k})\in\mathcal{S}(\Gamma) and determines a coset Ck=Lp,d(vd,ck)C_{k}=L_{p,d}(vd,c^{k})^{\uparrow}. Then for k>l,k>l,\,

(vd,ck)(vd,cl)1=(vd,ck)(cl,vd)=(vd,cklvd)=(vd,upkld)Lp,d(vd,c^{k})\,(vd,c^{l})^{-1}=(vd,c^{k})\,(c^{l},vd)=(vd,c^{k-l}vd)=(vd,up^{k-l}d)\not\in L_{p,d}

and the cosets CkC_{k} and ClC_{l} are distinct. ∎

We now consider a graph Γ\Gamma containing an edge aa that is a directed circuit of length one, and a closed inverse subsemigroup Lam,dL_{a^{m},\,d} of cycle type.

Theorem 4.4.
  1. (a)

    A closed inverse subsemigroup L=Lam,dL=L_{a^{m},\,d} of 𝒮(Γ)\mathcal{S}(\Gamma) is of infinite index if there exists a directed cycle cc in Γ\Gamma that contains an edge ee with aea\neq e, and a (possibly empty) directed path gg from some vertex v0v_{0} of dd to a vertex of cc and with gg having no edge in common with cdc\cup d.

  2. (b)

    Let L=Lam,dL=L_{a^{m},\,d} where aa is a directed circuit in Γ\Gamma of length one, and there are no other directed circuits in Γ\Gamma attached to a vertex of dd. Then LL has finite index in 𝒮(Γ)\mathcal{S}(\Gamma), given by

    [𝒮(Γ):Lam,d]=(m1)N𝐝(a),aΓ+vV(d)Nv,dΓ{a}.[\mathcal{S}(\Gamma):L_{a^{m},d}]=(m-1)N^{\Gamma}_{{\mathbf{d}}(a),a}+\sum_{v\in V(d)}N^{\Gamma\setminus\{a\}}_{v,d}\,.
Proof.

(a) Suppose that c,gc,g exist and let qq be the suffix of dd with initial vertex v0v_{0}.

Let Ck=L(q,gck)C_{k}=L(q,gc^{k})^{\uparrow}\,. Then if gg is non-empty, for k>lk>l,

(q,gck)(q,gcl)1=(q,gck)(gcl,q)=0L(q,gc^{k})\,(q,gc^{l})^{-1}=(q,gc^{k})\,(gc^{l},q)=0\not\in L

and the cosets CkC_{k} and ClC_{l} are distinct. If gg is empty, then

(q,ck)(q,cl)1=(q,ck)(cl,q)=(q,cklq)L(q,c^{k})\,(q,c^{l})^{-1}=(q,c^{k})\,(c^{l},q)=(q,c^{k-l}q)\not\in L

and the cosets CkC_{k}\, and ClC_{l}\, are again distinct.

(b) We are now reduced to the case that the only directed circuits in Γ\Gamma that can be attached to a vertex of dd are powers of the loop aa. A coset representative of L=Lam,dL=L_{a^{m},d} has the form (ard,w)(a^{r}d,w) with r0r\geqslant 0, or (q,w)(q,w) where qq is a proper suffix of dd. Hence ww has the same initial vertex vv as dd or of some proper suffix of dd. We can only construct finitely many representatives of the form (q,w)(q,w). We do not need to consider paths of the form (d,akw)(d,a^{k}w) for any k0k\geqslant 0 since (d,akw)(w,d)=(d,akd)L(d,a^{k}w)(w,d)=(d,a^{k}d)\in L. The analysis in the proof of part (b) of Theorem 4.1 can then be repeated to show that the number of cosets obtained this way is vV(d)Nv,dΓ{a}\sum_{v\in V(d)}N^{\Gamma\setminus\{a\}}_{v,d}.

We now consider representatives of the form (ard,w)(a^{r}d,w) with r1r\geqslant 1. Here ww must have the form w=astw=a^{s}t for some s0s\geqslant 0 and some (possibly empty) directed path tt not containing the edge aa. If rs(modm)r\equiv s\pmod{m} then (ard,ast)(t,d)=(ard,asd)L(a^{r}d,a^{s}t)(t,d)=(a^{r}d,a^{s}d)\in L and so L(ard,ast)=L(d,t)=L(s,t1)L(a^{r}d,a^{s}t)^{\uparrow}=L(d,t)^{\uparrow}=L(s,t_{1})^{\uparrow} for some suffix ss of dd and path t1t_{1} with the same initial vertex as ss but not sharing the same first edge. Hence L(ard,ast)L(a^{r}d,a^{s}t)^{\uparrow} will be counted within the sum vV(d)Nv,dΓ{a}\sum_{v\in V(d)}N^{\Gamma\setminus\{a\}}_{v,d}. Now fix tt and consider the cosets L(ard,ast)L(a^{r}d,a^{s}t)^{\uparrow} with rs(modm)r\not\equiv s\pmod{m}. Now given L(ar1d,as1t)L(a^{r_{1}}d,a^{s_{1}}t)^{\uparrow} and L(ar2d,as2t)L(a^{r_{2}}d,a^{s_{2}}t)^{\uparrow} with s1s2s_{1}\geqslant s_{2}, we have

(ar1d,as1t)(ar2d,as2t)1=(ar1d,as1t)(as2t,ar2d)=(ar1d,as1s2+r2d)(a^{r_{1}}d,a^{s_{1}}t)(a^{r_{2}}d,a^{s_{2}}t)^{-1}=(a^{r_{1}}d,a^{s_{1}}t)(a^{s_{2}}t,a^{r_{2}}d)=(a^{r_{1}}d,a^{s_{1}-s_{2}+r_{2}}d)

and (ar1d,as1s2+r2d)L(a^{r_{1}}d,a^{s_{1}-s_{2}+r_{2}}d)\in L if and only if r2s2r1s1(modm)r_{2}-s_{2}\equiv r_{1}-s_{1}\pmod{m}. Hence for a fixed tt we can produce exactly m1m-1 distinct cosets of the form L(ard,ast)L(a^{r}d,a^{s}t)^{\uparrow}.

But for distinct paths t1t_{1} and t2t_{2}, as1t1a^{s_{1}}t_{1} cannot be suffix comparable with as2t2a^{s_{2}}t_{2} and so

(ar1d,as1t1)(ar2d,as2t2)1=0L(a^{r_{1}}d,a^{s_{1}}t_{1})(a^{r_{2}}d,a^{s_{2}}t_{2})^{-1}=0\neq L

and the cosets determined by distinct paths t1t_{1} and t2t_{2} are distinct. Hence each of the N𝐝(a),aΓN^{\Gamma}_{{\mathbf{d}}(a),a} paths tt starting at 𝐝(a){\mathbf{d}}(a), but not having aa as its initial edge, contributes m1m-1 cosets. ∎

Example 4.5.

As in Remark 3.7, we suppose that Γ\Gamma consists only of the vertex xx and a loop aa at xx so that the graph inverse semigroup 𝒮(Γ)\mathcal{S}(\Gamma) is the bicyclic monoid BB with a zero adjoined. From Theorem 4.1, the closed inverse submonoids of BB contained in E(B)E(B) have infinite index. Part (b) of Theorem 4.4 tells us that that the closed inverse submonoid Bm=Lam,xB_{m}=L_{a^{m},x} of BB has index mm.

Example 4.6.

Let Γ\Gamma be the following graph:

x\textstyle{x^{\prime}\ignorespaces\ignorespaces\ignorespaces\ignorespaces}h\scriptstyle{h}y\textstyle{y^{\prime}}x\textstyle{x\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}a\scriptstyle{a}e\scriptstyle{e}g\scriptstyle{g}y\textstyle{y\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces\ignorespaces}f\scriptstyle{f}k\scriptstyle{k}z\textstyle{z}

and let L=La2,efL=L_{a^{2},ef}. Then we have

Nz,efΓa=1,Ny,efΓa=2,Nx,efΓa=3,N^{\Gamma\setminus a}_{z,ef}=1\,,N^{\Gamma\setminus a}_{y,ef}=2\,,N^{\Gamma\setminus a}_{x,ef}=3\,,

counting the paths in the sets {z}\{z\} , {y,k}\{y,k\} and {x,g,gh}\{x,g,gh\} respectively, and Nx,aΓ=6N^{\Gamma}_{x,a}=6, counting the paths in the set {x,e,g,ef,ek,gh}\{x,e,g,ef,ek,gh\}. From part (b) of Theorem 4.4 we find that [𝒮(Γ),L]=12[\mathcal{S}(\Gamma),L]=12 and a complete set of coset representatives is

{(z,z),(f,y),(f,k),(ef,x),(ef,g),(ef,gh),\left\{(z,z),(f,y),(f,k),(ef,x),(ef,g),(ef,gh),\right.
(ef,a),(ef,ag),(ef,agh),(ef,ae),(ef,aek),(ef,aef)}.\left.(ef,a),(ef,ag),(ef,agh),(ef,ae),(ef,aek),(ef,aef)\right\}\,.

References

  • [1] A. AlAli, Cosets in Inverse Semigroups and Inverse Subsemigroups of Finite Index. PhD thesis, Heriot-Watt University, Edinburgh (2016).
  • [2] A. AlAli and N.D. Gilbert, Inverse subsemigroups of finite index in finitely generated inverse semigroups. Preprint (2016). arxiv.org/abs/1608.04254  .
  • [3] C.J. Ash and T.E. Hall, Inverse semigroups on graphs. Semigroup Forum 11 (1975) 140-145.
  • [4] E.R. Dombi and N.D. Gilbert, HNN extensions of inverse semigroups with zero. Math. Proc. Camb. Phil. Soc. 142 (2007) 25-39.
  • [5] J.M. Howie, Fundamentals of Semigroup Theory. London Math. Soc. Monographs, Oxford University Press (1995).
  • [6] D.G. Jones and M.V. Lawson, Graph inverse semigroups: their characterization and completion. J. Algebra 409 (2014), 444-473.
  • [7] M. Lothaire, Combinatorics on Words. Cambridge Univ.Press (1997).
  • [8] M.V. Lawson, Primitive partial permutation representations of the polycyclic monoids and branching function systems. Periodica Mathematica Hungarica 58(2), (2009) 189-207.
  • [9] S.W. Margolis and J.C. Meakin, Free inverse monoids and graph immersions. Internat. J. Algebra Comput. 3 (1993)
  • [10] Z. Mesyan and J.D. Mitchell, The structure of a graph inverse semigroup. Semigroup Forum 93 (2016) 111-130.
  • [11] Z. Mesyan, J.D. Mitchell, M. Morayne, and Y. Péresse, Topological graph inverse semigroups. Topology Appl. 208 (2016) 106-126.
  • [12] M. Nivat and J.F. Perrot, Une generalisation du monoide bicyclique. C. R. Acad. Sci. Paris Sér I Math. A 271 (1970) 824-827.
  • [13] A.L. Paterson, Graph inverse semigroups, groupoids, and their CC^{*}–algebras. J. Operator Theory 48 (2002) 645-662.
  • [14] B. Schein, Cosets in groups and semigroups. In Semigroups with applications, (eds J. M. Howie, W. D. Munn, and H. J. Weinert), World Scientific, Singapore (1992) 205-221.