Generators and presentations of inverse subsemigroups of the monogenic free inverse semigroup

We will show that S is finitely generated. Finite presentability will then follow by Theorem Theorem.

By assumption, S=\langle T\rangle, where T:=S\setminus E(S) is the set of all non-idempotent elements of S. Note that T is always infinite.

For the rest of the proof of Theorem, we will fix a finite set T_1\subseteq T guaranteed by Proof Proof. We aim to show that there exists another finite set T_2 of non-idempotent elements of T, such that S = \langle T_1, T_2 \rangle. Recall a_{\text{min}}, b_{\text{min}}, p and the elements

from Lemma 3.2. Define

Let us consider an \mathscr{R}-class R_{a,b} of FI_1 which intersects S, and such that a+b>N. We know that R_{a,b} consists of all (-a,n,b) with -a\leq n\leq b, and that the second component of any triple in S is divisible by p. Therefore \{ (-a,qp,b)\colon -a\leq qp\leq b\}\subseteq R_{a,b}\cap S. We aim to show that the two sets are in fact equal, and that all these elements can be obtained just from \epsilon:=(-a,0,b), \alpha and \beta.

Following Lemma 3.2, write a and b as a = a_{\text{min}}+xp+r and b = b_{\text{min}}+yp+s where x,y \in \mathbb{N}_0, 0 \leq r, s <p. Then any element (-a, qp, b) \in R_{a,b}\cap S has -x \leq q \leq y. The following two claims show how to obtain the elements (-a,(q\pm 1)p,b) using u:=(-a,qp,b), \alpha and \beta.

We can now complete the proof of Theorem as follows. Recall that each \mathscr{D}-class in FI_1 is finite. Let T_{2} := \bigl(\bigcup_{i=1}^{N}D_i\bigr)\cap S, a finite set. Note that \alpha, \beta \in T_2 as a_{\text{min}}+b_{\alpha}, a_{\beta}+b_{\text{min}} \leq N. We claim that S = \langle T_1, T_2 \rangle. To see this note that E(S) \subseteq \langle T_1, T_2\rangle by Proof Proof. Now, consider an arbitrary R_{a,b} which intersects S. If a+b\leq N then R_{a,b}\cap S\subseteq D_{a+b} \subseteq T_2\subseteq \langle T_1,T_2\rangle. If a+b>N, then Claim ? gives R_{a,b}\cap S\subseteq \langle E(S),\alpha,\beta\rangle\subseteq \langle T_1,T_2\rangle. Hence S is finitely generated, and the proof of the theorem is complete.

Let \overline{S} be the inverse subsemigroup of S that is generated by all non-idempotent elements of S. Then S = \langle E(S), \overline{S} \rangle, and the corollary follows from Theorem.

Let x\in S and y\in \overline{S}, and we want to prove that xy,yx\in\overline{S}. We will prove that xy\in \overline{S}, and the claim for yx follows by symmetry. If x\in \overline{S} then xy\in\overline{S}; so let us suppose that x\in S\setminus\overline{S}, which certainly means that x is an idempotent. Now, if y is a non-idempotent element, then it is easy to see that xy is also non-idempotent, and hence xy\in\overline{S}. Finally, if y is an idempotent, then y=zz^{-1} for some non-idempotent element z by Claim Proof in the proof of Theorem, so that xy=(xz)z^{-1}\in \overline{S}.

It is clear that S is generated by a set in one to one correspondence with X_E\cup Y and that it satisfies the relations R \cup Q \cup P. Therefore, S is a homomorphic image of the inverse semigroup defined by the presentation (4.1). It remains to prove that for words u_1 and u_2 over X_E \cup Y, if the equality u_1 = u_2 holds in S, then it can be deduced from R \cup Q \cup P.

In order to prove this, we claim that for every word u\in (X_E\cup X_E^{-1}\cup Y\cup Y^{-1})^+ there exists a word \overline{u}\in (X_E\cup X_E^{-1})^+\cup (Y\cup Y^{-1})^+ such that u=\overline{u} is a consequence of R \cup Q \cup P. The assertion is non-obvious only if u contains letters from both X_E\cup X_E^{-1} and Y\cup Y^{-1}. In that case u contains a subword of the form x_e^\epsilon y^\delta or y^{-\delta} x_e^{-\epsilon} with e\in E, y\in Y, \epsilon,\delta\in\{-1,1\}. The element y^\delta y^{-\delta} represents an idempotent f from \overline{S}, and using Proposition 3.4 we have g:=ef\in E(S)\cap \overline{S}. Now we have the deduction

Taking inverses we see that also y^{-\delta}x^{-\epsilon} can be transformed into a word over Y using R \cup Q \cup P. Using this repeatedly, one can eliminate all the occurrences of letters from X_E\cup X_E^{-1} in u, and end up with a word \overline{u} solely over Y.

Now, returning to our relation u_1=u_2 that holds in S, consider \overline{u}_1, \overline{u}_2. If both \overline{u}_1 and \overline{u}_2 are words over X_E (resp. Y) then the relation \overline{u}_1=\overline{u}_2 is a consequence of R (resp. Q). Otherwise, we may assume without loss that \overline{u}_1\in (X_E\cup X_E^{-1})^+ and \overline{u}_2\in (Y\cup Y^{-1})^+. In this case they both represent an idempotent e\in E(S)\cap \overline{S}, and we have

In all the cases we have deduced \overline{u}_1=\overline{u}_2 as a consequence of R \cup Q \cup P, and then u_1=\overline{u}_1=\overline{u}_2=u_2 is a derivation of u_1=u_2, as required.

For a \in A and \epsilon=\pm 1 both y_{a}^\epsilon y_{a}^{-\epsilon} and x_{a^\epsilon a^{-\epsilon}} correspond to the idempotent a^\epsilon a^{-\epsilon}. Define:

By Theorem the inverse subsemigroup \overline{S} generated by all non-idempotent elements of S is finitely presented. Since A consists of non-idempotent elements, it can be extended to a finite generating set for \overline{S}, and then \overline{S} can be defined by a finite presentation over the resulting generating set. In other words, \overline{S} is defined by a finite presentation \operatorname{Inv} \langle Y \mid Q \rangle where Y_A \subseteq Y. We write Y_1 for the set Y\setminus Y_A. Each y_1 \in Y_1 corresponds to an element in S. This element can be written as a product of generators E(S)\cup A \cup A^{-1}, and so there is a word w_{y_1} over X_E\cup Y_A which corresponds to the element. Let

We claim that S is defined by

where T_3 := T_1 \cup T_2. Once we prove the claim, the proof of the corollary is easy to complete. Indeed, the generators in Y_1 are redundant via relations T_2. Thus we may eliminate those generators, which results in removing the relations T_2, and substituting w_{y_1} for y_1 in all other relations. But, in fact y_1 can only appear in relations Q, which after the substitution become a new finite set of relations Q'. Then, defining T:=Q' \cup T_1 completes the proof of the corollary.

We now turn to the proof of the claim. Recall the set P of relations in Theorem 4.1:

where each w_e is an arbitrary word over Y which represents the idempotent e. Thus, we may take w_e to have the form w_e\equiv w_1w_1^{-1}, w_1\in (Y\cup Y^{-1})^+. By Theorem 4.1, S is defined by

It is clear that S satisfies the relations in C \cup T_3, and so the set R\cup Q \cup P of relations implies C \cup T_3. We need to prove that the set R \cup Q\cup C\cup T_3 of relations implies P. To this end consider an arbitrary e \in E(S)\cap \overline{S}.

First, we use relations from T_2 to eliminate generators from Y_1 in w_1, yielding a word w_2\in (X_E\cup X_E^{-1}\cup Y_A\cup Y_A^{-1})^+, which must contain at least one occurrence of a letter from Y_A. Thus w_e=w_2 w_2^{-1} as a consequence of R \cup Q\cup C\cup T_3.

Now consider the last occurrence of a letter from Y_A in w_2; in other words suppose w_2 \equiv uy_a^\epsilon v, where y_a\in Y_A, \epsilon=\pm 1, and v\in (X_E\cup X_E^{-1})^*. If the word v is empty, denoting by g the idempotent represented by a^\epsilon a^{-\epsilon}, we have

When v is non-empty, it is equal to some x_f as a consequence of R; denoting by g the idempotent represented by a^\epsilon fa^{-\epsilon}, we have:

Denoting u x_g by w_3, in each case we have proved that w_e=w_2w_2^{-1}=w_3 w_3^{-1} as a consequence of R \cup Q\cup C\cup T_3, and the number of occurrences of letters from Y_A in w_3 is one less than in w_2. Continuing in this way we can successively eliminate all the occurrences of letters from Y_A, and therefore deduce w_e=x_e as a consequence of R \cup Q\cup C\cup T_3, which is the desired relation from P. This completes the proof of the claim, and hence of the corollary.

Aiming for a contradiction, suppose that S is finitely presented modulo E(S). Hence, S can be defined by a presentation

where X_E= \{x_e \colon e\in E(S)\} is a (necessarily infinite) set in one to one correspondence with E(S), \operatorname{Inv}\langle X_E\mid R\rangle is a presentation for E, and the sets Y and T are finite. For the purposes of this proof, we will take R to be the set of relations arising from the Cayley table of E, i.e. x_ex_f=x_{ef}, e,f\in E.

Let \overline{S} be the inverse subsemigroup of S generated by all non-idempotent elements of S. Using Theorem, we may without loss assume that the set Y corresponds to a finite set of non-idempotent elements in S, which generates \overline{S}. Furthermore, Theorem also gives that \overline{S} is defined by a finite presentation \operatorname{Inv} \langle Y \mid Q\rangle. Now Theorem 4.1 implies that S is defined by

where w_e\in (Y\cup Y^{-1})^+ represents e. Since both presentations (Proof) and (\ref{eq:new_no_fin_2}) define S, we can deduce T from Q\cup P. Since T is finite, only finitely many relations from Q\cup P are used in such a deduction. Hence there exists a finite subset P_1 of P such that S is defined by a presentation

Consider the element f:=e_1\dots e_k\in E(S). Then the relation x_{e_1}\dots x_{e_k} = x_f is a consequence of R. Also, we have x_f = w_f as a consequence of R\cup Q \cup P_1. Since w_f is a word over Y, there is a letter y \in Y and \epsilon=\pm 1 such that w_f \leq y^\epsilon y^{-\epsilon}. Recall that the semilattice of idempotents of \langle y \rangle consists of the elements of the form y^{-i}y^iy^jy^{-j} (see ?). In particular, there exist p,q\in\mathbb{N}_0 such that the idempotent g represented by y^{-p}y^py^qy^{-q} is strictly smaller than f. The relation y^{-p}y^py^qy^{-q}=x_g holds in S. In particular, the word x_g belongs to the language of the Schützenberger graph \operatorname{\textit{S} \, \Gamma}(y^{-p}y^py^{q}y^{-q}) with respect to X_E\cup Y.

On the other hand, \operatorname{\textit{S} \, \Gamma}(y^{-p}y^py^{q}y^{-q}) can be obtained by using Stephen’s procedure with respect to the presentation (?). We are going to reach a contradiction by showing inductively that at every stage of the procedure the following holds:

this will immediately imply that there will never be an edge labelled x_g. This is certainly the case at the start of the procedure, i.e. for the linear graph of y^{-p}y^py^{q}y^{-q} which contains no edges labelled by x_h for any h. Also, it is clear that a determination does not introduce new labels. Consider the possibility that an expansion of the graph that satisfies (?) yields a graph that does not.

Notice that the relations from Q contain no symbols from X_E, while those in P_1 contain only x_{e_1},\dots,x_{e_k}, and e_1,\dots,e_k\geq f. Hence, none of these relations can be used to introduce a label x_h with h<f. The remaining relations R have the form x_{h'} x_{h''}=x_h, where h,h',h''\in E and h=h'h''. Notice that h\geq f if and only if h',h''\geq f, and it follows that this relation does not introduce a label corresponding to an idempotent below f either. This completes the proof of the claim, and the theorem follows.

The equivalence of (i) and (ii) is an immediate consequence of Theorem Theorem. Also, (iii)\Rightarrow(i) is an easy consequence of the fact that \overline{S} is finitely generated (Theorem). For (i)\Rightarrow(iii) suppose that S is finitely generated, say by A. Recall that \overline{S} is an ideal of S (Proposition 3.4). It follows that S\setminus\overline{S}\subseteq \langle A\setminus\overline{S}\rangle. But A\setminus\overline{S} is a finite set of idempotents, and the subsemigroup generated by it is also finite.