1 Introduction
An algebra is finitely based if the identities it satisfies are finitely axiomatisable; otherwise, it is nonfinitely based. The celebrated theorem of Oates and Powell [Reference Oates and Powell20], published in 1964, states that all finite groups are finitely based. In the decade that followed, finite members from other classes of algebras such as lattices [Reference McKenzie19], associative rings [Reference Kruse7, Reference L’vov8] and Lie rings [Reference Bahturin and Ol’šanskiĭ2] were also shown to be finitely based. However, this is not true in general. In the 1960s, Perkins [Reference Perkins21] published the first examples of nonfinitely based finite semigroups, one of which is the well-known Brandt monoid
$B_2^1$
of order six. The discovery of this example focused attention upon the finite basis problem for small semigroups. In particular, is there a nonfinitely based semigroup of order less than six? After several decades of cumulative work, a complete solution has been found for all semigroups of order up to six: every semigroup of order five or less is finitely based [Reference Lee9, Reference Trahtman and Lyapin22] and there are only four nonfinitely based semigroups of order six (including
$B_2^1$
) up to isomorphism [Reference Lee and Li16–Reference Lee and Zhang18].
This paper is concerned with involution semigroups, that is, unary semigroups
$(S,{}^*)$
that satisfy the identities
$$ \begin{align} (x^*)^* \approx x \quad \text{and} \quad (xy)^* \approx y^*x^*; \end{align} $$
the unary operation
$^*$
is an involution of S, and S is the semigroup reduct of
$(S,{}^*)$
. Common examples of involution semigroups include groups
$(G,{}^{-1})$
under inversion
$^{-1}$
and multiplicative matrix semigroups
$(M_n,{}^T)$
over any field under the usual matrix transposition
$^T$
.
With respect to the finite basis problem, involution semigroups have not been considered as much as semigroups, perhaps due to the supposition that a finite involution semigroup
$(S,{}^*)$
and its reduct S are simultaneously finitely based; but this has been refuted by recent examples [Reference Jackson, Volkov, Blass, Dershowitz and Reisig6, Reference Lee10, Reference Lee12]. An interesting example is the monoid
$A_0^1$
, obtained by adjoining a unit element to the semigroup
$$ \begin{align*} A_0=\langle e,f \,|\, e^2=e,\, f^2=f,\, ef=0\rangle \end{align*} $$
of order four, with involution
$^*$
given by the transposition
$e \leftrightarrow f$
on
$A_0^1$
, that is,
$^*$
interchanges e and f but fixes every other element. It is long known that the monoid
$A_0^1$
of order five is finitely based [Reference Edmunds4], but recently, Gao et al. [Reference Gao, Zhang and Luo5] have shown that the involution monoid
$(A_0^1, {}^*)$
is nonfinitely based. This result is surprising given that every semigroup of order up to five is finitely based [Reference Lee9, Reference Trahtman and Lyapin22]. As in the case for semigroups, it is natural to ask if there exists a nonfinitely based involution semigroup of order less than five [Reference Gao, Zhang and Luo5, Question 1.3]. The objective of the present article is to provide an answer to this question for involution monoids.
Theorem 1.1. Up to isomorphism, the involution monoid
$(A_0^1,{}^*)$
of order five is the unique smallest nonfinitely based algebra in the class of all involution monoids.
Notation and background information are first given in Section 2. An outline of the proof of Theorem 1.1 is then given in Section 3, while the finer details of the proof are deferred to Sections 4–6.
Question 1.2. Is there a nonfinitely based involution semigroup of order five or less?
Since every involution semigroup of order up to three is finitely based [Reference Lee14], any example that positively answers Question 1.2 is of order four or five. If the answer to Question 1.2 is negative, then
$(A_0^1,{}^*)$
is also the unique smallest nonfinitely based involution semigroup. Refer to the monograph of Lee [Reference Lee15] for more information on the finite basis problem for involution semigroups.
2 Preliminaries
Most of the notation and background material of this article are given in this section. Refer to the monograph of Burris and Sankappanavar [Reference Burris and Sankappanavar3] for more information.
2.1 Words
Let
$\mathcal {A}$
be a countably infinite alphabet that excludes the symbol
$1$
, and let
$\mathcal {A}^* =\{x^* \,|\, x \in \mathcal {A}\}$
be a disjoint copy of
$\mathcal {A}$
. Elements of
$\mathcal {A} \cup \mathcal {A}^*$
are called variables. The free involution semigroup over
$\mathcal {A}$
is the free semigroup
$F_{\mathsf {inv}}(\mathcal {A}) = ({\mathcal {A}} \cup {\mathcal {A}}^{*})^+$
with unary operation
$^*$
given by
$(x^*)^* = x$
for all
$x \in \mathcal {A}$
and
$$ \begin{align*} (x_1x_2 \cdots x_n)^* = x_n^* x_{n-1}^* \cdots x_1^* \end{align*} $$
for all
$x_1,x_2,\ldots ,x_n \in \mathcal {A} \cup \mathcal {A}^*$
. The free involution monoid over
$\mathcal {A}$
is
$F_{\mathsf {inv}}^1(\mathcal {A}) = F_{\mathsf {inv}}(\mathcal {A}) \cup \{ 1\}$
, where
$1$
is the empty word with
$1^* = 1$
. Elements of
$F_{\mathsf {inv}}^1(\mathcal {A})$
are called words and elements of
$\mathcal {A}^+ \cup \{1\}$
are called plain words. A word
$\mathbf {u}$
is a factor of a word
$\mathbf {v}$
if
$\mathbf {p}\mathbf {u}\mathbf {q} = \mathbf {v}$
for some
$\mathbf {p}, \mathbf {q} \in F_{\mathsf {inv}}^1(\mathcal {A})$
.
The plain projection of a word
$\mathbf {u} \in F_{\mathsf {inv}}(\mathcal {A})$
, denoted by
$\overline {\mathbf {u}}$
, is the plain word obtained from
$\mathbf {u}$
by removing all occurrences of the symbol
$^*$
. The content of a word
$\mathbf {u}$
, denoted by
$\operatorname {\mathsf {con}}(\mathbf {u})$
, is the set of variables occurring in
$\mathbf {u}$
; the number of times that a variable x occurs in
$\mathbf {u}$
is denoted by
$\operatorname {\mathsf {occ}}(x,\mathbf {u})$
. A variable
$x \in \mathcal {A} \cup \mathcal {A}^*$
is simple in
$\mathbf {u}$
if
$\operatorname {\mathsf {occ}}(\overline {x},\overline {\mathbf {u}})=1$
; otherwise, it is nonsimple. A word
$\mathbf {u}$
is simple if every variable in
$\mathbf {u}$
is simple in
$\mathbf {u}$
. Let
$\operatorname {\mathsf {sim}}(\mathbf {u})$
denote the set of simple variables occurring in
$\mathbf {u}$
and
$\operatorname {\mathsf {non}}(\mathbf {u})$
denote the set of nonsimple variables occurring in
$\mathbf {u}$
.
For any
$\mathbf {u} \in F_{\mathsf {inv}}^1(\mathcal {A})$
and
$x_1, x_2, \ldots , x_n \in \mathcal {A}$
, let
$\mathbf {u}[x_1, x_2, \ldots , x_n]$
denote the word obtained from
$\mathbf {u}$
by retaining the variables
$x_1, x_1^*, x_2, x_2^*, \ldots , x_n, x_n^*$
. In particular,
$\mathbf {u}[\operatorname {\mathsf {sim}}(\mathbf {u})]$
is obtained from
$\mathbf {u}$
by retaining its simple variables.
Example 2.1. If
$\mathbf {u} =x^*xy^*x^2yz^*yx^*$
with
$x,y,z\in \mathcal {A}$
, then
-
•
$\overline {\mathbf {u}}=x^2yx^2yzyx$
; -
•
$\operatorname {\mathsf {con}} (\mathbf {u})=\{x, x^*, y, y^*, z^*\}$
; -
•
$\operatorname {\mathsf {occ}} (x, \mathbf {u}) = 3$
,
$\operatorname {\mathsf {occ}} (x^*,\mathbf {u}) =2$
,
$\operatorname {\mathsf {occ}} (y, \mathbf {u}) = 2$
,
$\operatorname {\mathsf {occ}} (z^*,\mathbf {u}) = \operatorname {\mathsf {occ}} (y^*,\mathbf {u})=1$
; -
•
$\operatorname {\mathsf {occ}} (x, \overline {\mathbf {u}}) = 5$
,
$\operatorname {\mathsf {occ}}(y,\overline {\mathbf {u}}) =3$
,
$\operatorname {\mathsf {occ}} (z,\overline {\mathbf {u}})=1$
; -
•
$\operatorname {\mathsf {sim}}(\mathbf {u})= \{z^*\}$
,
$\operatorname {\mathsf {non}}(\mathbf {u})= \{x, x^*, y, y^*\}$
; -
•
$\mathbf {u}[x]=x^*x^3x^*$
,
$\mathbf {u}[x,y]=x^*xy^*x^2y^2x^*$
,
$\mathbf {u}[y,z]=y^*yz^*y$
.
2.2 Identities
An identity is an expression
$\mathbf {u}\approx \mathbf {v}$
formed by words
$\mathbf {u},\mathbf {v} \in F_{\mathsf {inv}}^1(\mathcal {A})$
. An involution semigroup
$(S,{}^*)$
satisfies an identity
$\mathbf {u} \approx \mathbf {v}$
if, for any substitution
$\varphi : \mathcal {A} \to S$
, the elements
$\mathbf {u}\varphi $
and
$\mathbf {v}\varphi $
of S coincide; in this case,
$\mathbf {s} \approx \mathbf {t}$
is also said to be an identity of
$(S,{}^*)$
.
An involution monoid that satisfies an identity
$\mathbf {u} \approx \mathbf {v}$
also satisfies the identity
$\mathbf {u}[x_1, x_2,\ldots ,x_n] \approx \mathbf {v}[x_1, x_2,\ldots ,x_n]$
for any
$x_1, x_2,\ldots ,x_n \in \mathcal {A}$
, since assigning the unit element
$1$
to a variable x in an identity is effectively the same as removing all occurrences of x and
$x^*$
.
For any involution semigroup
$(S,{}^*)$
, a set
$\Sigma $
of identities of
$(S,{}^*)$
is an identity basis for
$(S,{}^*)$
if every identity of
$(S,{}^*)$
can be deduced from
$\Sigma $
. An involution semigroup is finitely based if it has some finite identity basis; otherwise, it is nonfinitely based.
2.3 Periodic commutative involution semigroups
Perkins [Reference Perkins21] proved that every commutative semigroup is finitely based. In this subsection, a similar result is established for involution semigroups.
Proposition 2.2. Every periodic commutative involution semigroup is finitely based.
Recall that a semigroup S is periodic if it satisfies the identity
$x^i \approx x^{i+j}$
for some
$i,j \geq 1$
. If
$m \ge 1$
is the least such that S satisfies
$x^m \approx x^{m+j}$
for some
$j \geq 1$
, and k is the least such that S satisfies
$x^m \approx x^{m+k}$
, then S is
$(m,k)$
-periodic. An involution semigroup
$(S,{}^*)$
is
$(m,k)$
-periodic if S is
$(m,k)$
-periodic. An identity
$\mathbf {u} \approx \mathbf {v}$
of an
$(m,k)$
-periodic involution semigroup
$(S,{}^*)$
is reduced if the words
$\mathbf {u}$
and
$\mathbf {v}$
belong to the set
$$ \begin{align*} \{ x_1^{e_1}x_2^{e_2}\cdots x_n^{e_n} \mid 0 \leq e_1, e_2, \ldots, e_n<m+k \} \end{align*} $$
for some distinct variables
$x_1, x_2, \ldots , x_n \in \mathcal {A} \cup \mathcal {A}^*$
.
Let
$\mathbf {u}\approx \mathbf {v}$
be a reduced identity of an
$(m,k)$
-periodic involution semigroup
$(S,{}^*)$
. For any integers
$p,q,s,t$
such that
$0 \leq p, q, s, t < m+k$
, a nonempty set
$U_{(p,q,s,t)} =\{x_1,x_2,\ldots , x_n \}$
of variables from
$\mathsf {con}(\overline {\mathbf {u}\mathbf {v}})$
is called the
$(p, q, s, t)$
-block of
$\mathbf {u}\approx \mathbf {v}$
if
$U_{(p,q,s,t)}$
is the maximal subset of
$\mathsf {con}(\overline {\mathbf {u}\mathbf {v}})$
such that for each variable
$x_i$
in
$U_{(p,q,s,t)}$
,
$$ \begin{align*} \operatorname{\mathsf{occ}}(x_i,\mathbf{u}) = p, \quad \operatorname{\mathsf{occ}}(x_i^*,\mathbf{u})=q, \quad \operatorname{\mathsf{occ}}(x_i,\mathbf{v}) = s \quad \text{and} \quad \operatorname{\mathsf{occ}}(x_i^*,\mathbf{v}) = t. \end{align*} $$
Note that
$(p, q, s, t)\neq (0,0,0,0)$
because
$U_{(p,q,s,t)}\neq \emptyset $
. The length of
$U_{(p,q,s,t)}$
is denoted by
$|U_{(p,q,s,t)}|$
. For instance, if
$\mathbf {u} \approx \mathbf {v}$
is an identity with reduced words
$$ \begin{align*} \mathbf{u}=x_1^2x_2^3x_3^3x_4^3x_5^2x_1^*(x_2^*)^2(x_3^*)^2(x_4^*)^2x_5^*x_6^* \quad\text{and}\quad\mathbf{v}=x_1^6x_5^6(x_2^*)^2(x_3^*)^2(x_4^*)^2, \end{align*} $$
then
$\{x_1 , x_5\}$
,
$\{x_2, x_3, x_4\}$
and
$\{x_6\}$
are the
$(2, 1, 6, 0)$
-block, the
$(3, 2, 0, 2)$
-block and the
$(0, 1, 0, 0)$
-block of
$\mathbf {u}\approx \mathbf {v}$
, respectively.
For any reduced identity
$\mathbf {u}\approx \mathbf {v}$
, since each component of a quadruple
$(p, q, s, t)$
is from
$\{ 0,1,2,\ldots , m+k-1 \}$
and
$(p, q , s, t)\neq (0,0,0,0)$
, the number of possible quadruples is
$r=(m+k)^4 - 1$
. Encode these quadruples so that we can refer to the i-block
$U_i$
of
$\mathbf {u}\approx \mathbf {v}$
, where
$1\leq i \leq r$
, instead of the
$(p, q, s, t)$
-block
$U_{(p,q,s,t)}$
of
$\mathbf {u}\approx \mathbf {v}$
. An r-dimensional vector
$\overrightarrow {\mathbf {l}} \in \mathbb {N}^r$
(where
$\mathbb {N} = \{ 0,1,2,\ldots \}$
) is called the length vector of blocks for a reduced identity
$\mathbf {u}\approx \mathbf {v}$
if the ith component
$$ \begin{align*} \overrightarrow{\mathbf{l}}(i) &= \begin{cases} |U_i| & \text{if the } i\text{-block of } \mathbf{u}\approx \mathbf{v} \text{ exists}, \\ 0 & \text{otherwise}. \end{cases} \end{align*} $$
It is routine to check that every reduced identity
$\mathbf {u}\approx \mathbf {v}$
of an
$(m,k)$
-periodic commutative involution semigroup can be uniquely determined by some r-dimensional vector. Let
$\overrightarrow {\mathbf {l}_1}$
and
$\overrightarrow {\mathbf {l}_2}$
be r-dimensional length vectors corresponding to reduced identities
$\mathbf {u}_1 \approx \mathbf {v}_1$
and
$\mathbf {u}_2 \approx \mathbf {v}_2$
, respectively. Define a partial order
$\preceq $
on
$\mathbb {N}^r$
such that
$\overrightarrow {\mathbf {l}_1} \preceq \overrightarrow {\mathbf {l}_2}$
if
$\overrightarrow {\mathbf {l}_1}(i) \leq \overrightarrow {\mathbf {l}_2}(i)$
for all
$i \in \{ 1,2,\ldots , r\}$
. Similar to the argument given by Perkins [Reference Perkins21, Section 4] for the case of semigroups, we can deduce a ‘long’ identity from a ‘short’ identity using some appropriate substitution, that is, if
$\overrightarrow {\mathbf {l}_1} \preceq \overrightarrow {\mathbf {l}_2}$
, then
$\mathbf {u}_1 \approx \mathbf {v}_1$
implies
$\mathbf {u}_2\approx \mathbf {v}_2$
.
Proof of Proposition 2.2.
Let
$(S,{}^*)$
be any periodic commutative involution semigroup, say
$(S,{}^*)$
is
$(m,k)$
-periodic. Suppose that
$(S,{}^*)$
is nonfinitely based. Then there exists an infinite set
$$ \begin{align*} \Sigma=\{ \mathbf{u}_1\approx \mathbf{v}_1, \ \mathbf{u}_2\approx \mathbf{v}_2, \ \mathbf{u}_3\approx \mathbf{v}_3, \ldots \} \end{align*} $$
of identities of
$(S,{}^*)$
such that for each
$i \geq 1$
, the first i identities
$$ \begin{align*}\Sigma_i = \{ \mathbf{u}_1\approx \mathbf{v}_1, \, \mathbf{u}_2\approx \mathbf{v}_2, \, \ldots, \, \mathbf{u}_i\approx \mathbf{v}_i \} \end{align*} $$
do not imply the
$(i+1)$
st identity
$\mathbf {u}_{i+1}\approx \mathbf {v}_{i+1}$
. Since
$(S,{}^*)$
is commutative, each identity
$\mathbf {u}_i\approx \mathbf {v}_i \in \Sigma $
can be converted into reduced form and is thus uniquely associated with some length vector
$\overrightarrow {\mathbf {l}_i} \in \mathbb {N}^r$
. Since
$\Sigma _i \nvdash \mathbf {u}_{i+1}\approx \mathbf {v}_{i+1}$
for each
$i \geq 1$
, the length vectors
$\overrightarrow {\mathbf {l}_1},\overrightarrow {\mathbf {l}_2},\overrightarrow {\mathbf {l}_3}, \ldots $
corresponding to
$\mathbf {u}_1\approx \mathbf {v}_1, \mathbf {u}_2\approx \mathbf {v}_2,\mathbf {u}_3\approx \mathbf {v}_3,\ldots $
are distinct.
The set
$\{\overrightarrow {\mathbf {l}_1},\overrightarrow {\mathbf {l}_2},\overrightarrow {\mathbf {l}_3}, \ldots \}$
of infinitely many pairwise distinct r-dimensional vectors must contain two vectors
$\overrightarrow {\mathbf {l}_k}$
and
$\overrightarrow {\mathbf {l}_{\ell }}$
with
$k< \ell $
such that
$\overrightarrow {\mathbf {l}_k}\preceq \overrightarrow {\mathbf {l}_{\ell }}$
. Indeed, this can be proved by induction on the dimension r. If
$r=1$
, the conclusion holds obviously. Suppose that infinitely many pairwise distinct
$(r-1)$
-dimensional vectors contain two
$\preceq $
-related vectors. Then we can show that it also holds for the r-dimensional vectors. Let
$L_1=\{\overrightarrow {\mathbf {l}_1}, \overrightarrow {\mathbf {l}_2}, \overrightarrow {\mathbf {l}_3}, \ldots \}$
. We can find an infinite set
$L_2 = \{\overrightarrow {\mathbf {l}_{p_1}}, \overrightarrow {\mathbf {l}_{p_2}}, \overrightarrow {\mathbf {l}_{p_3}}, \ldots \} \subseteq L_1$
for some
$p_1< p_2< p_3 < \cdots $
such that for at least one component, say the ith component with
$1\leq i\leq r$
, one has
$\overrightarrow {\mathbf {l}_{p_1}}(i) \leq \overrightarrow {\mathbf {l}_{p_2}}(i) \leq \overrightarrow {\mathbf {l}_{p_3}}(i) \leq \cdots $
. By the inductive hypothesis, we can find two distinct vectors
$\overrightarrow {\mathbf {l}_k}, \overrightarrow {\mathbf {l}_{\ell }} \in L_2$
such that
$\overrightarrow {\mathbf {l}_k}\preceq \overrightarrow {\mathbf {l}_{\ell }}$
. Therefore, there exists some j such that
$\Sigma _j \vdash \mathbf {u}_{j+1}\approx \mathbf {v}_{j+1}$
, which is a contradiction.
3 Proof of Theorem 1.1
Since a finite involution monoid is finitely based if it is either commutative (Proposition 2.2) or of order at most three [Reference Lee14], it suffices to consider those that are noncommutative and of order four or five. It is routine to check with a computer that every involution monoid of order four is commutative and so is finitely based, and there are only six noncommutative involution monoids of order five:
$$ \begin{align*} & \begin{array}{c | c c c c c} M_1 & 1 & 2 & 3 & 4 & 5 \\ \hline 1 & 1 & 1 & 1 & 1 & 1 \\ 2 & 1 & 1 & 1 & 1 & 2 \\ 3 & 1 & 1 & 1 & 1 & 3 \\ 4 & 1 & 1 & 2 & 1 & 4 \\ 5 & 1 & 2 & 3 & 4 & 5 \\ \hline x^* & 1 & 2 & 4 & 3 & 5 \end{array} && \begin{array}{c | c c c c c} M_2 & 1 & 2 & 3 & 4 & 5 \\ \hline 1 & 1 & 1 & 1 & 1 & 1 \\ 2 & 1 & 1 & 1 & 1 & 2 \\ 3 & 1 & 1 & 1 & 3 & 3 \\ 4 & 1 & 2 & 1 & 4 & 4 \\ 5 & 1 & 2 & 3 & 4 & 5 \\ \hline x^* & 1 & 3 & 2 & 4 & 5 \end{array} && \begin{array}{c | c c c c c} M_3 & 1 & 2 & 3 & 4 & 5 \\ \hline 1 & 1 & 1 & 1 & 1 & 1 \\ 2 & 1 & 1 & 1 & 1 & 2 \\ 3 & 1 & 1 & 2 & 1 & 3 \\ 4 & 1 & 1 & 2 & 2 & 4 \\ 5 & 1 & 2 & 3 & 4 & 5 \\ \hline x^* & 1 & 2 & 4 & 3 & 5 \end{array} \\[0.05in] & \begin{array}{c | c c c c c} M_4 & 1 & 2 & 3 & 4 & 5 \\ \hline 1 & 1 & 1 & 1 & 1 & 1 \\ 2 & 1 & 1 & 1 & 2 & 2 \\ 3 & 1 & 2 & 3 & 1 & 3 \\ 4 & 1 & 1 & 1 & 4 & 4 \\ 5 & 1 & 2 & 3 & 4 & 5 \\ \hline x^* & 1 & 2 & 4 & 3 & 5 \end{array} && \begin{array}{c | c c c c c} M_5 & 1 & 2 & 3 & 4 & 5 \\ \hline 1 & 1 & 1 & 1 & 1 & 1 \\ 2 & 1 & 1 & 1 & 2 & 2 \\ 3 & 1 & 2 & 3 & 2 & 3 \\ 4 & 1 & 1 & 1 & 4 & 4 \\ 5 & 1 & 2 & 3 & 4 & 5 \\ \hline x^* & 1 & 2 & 4 & 3 & 5 \end{array} && \begin{array}{c | c c c c c} M_6 & 1 & 2 & 3 & 4 & 5 \\ \hline 1 & 1 & 1 & 1 & 4 & 4 \\ 2 & 1 & 2 & 3 & 4 & 5 \\ 3 & 3 & 3 & 3 & 5 & 5 \\ 4 & 1 & 4 & 1 & 4 & 4 \\ 5 & 3 & 5 & 3 & 5 & 5 \\ \hline x^* & 1 & 2 & 4 & 3 & 5 \end{array} \end{align*} $$
The involution monoid
$(M_1,{}^*)$
, which appears in [Reference Lee13] as
$\langle \mathsf {Rq}\{xx^*\},{}^*\rangle $
, is finitely based; in particular, its identities are axiomatised by (1.1) and
$$ \begin{align*} x^3 \approx x^2, \quad xyx \approx x^2y, \quad xyx \approx yx^2, \quad xyx^* \approx xx^*y, \quad xyx^* \approx yxx^*, \quad (x^*)^2 \approx x^2. \end{align*} $$
The involution monoids
$(M_2,{}^*)$
,
$(M_3,{}^*)$
and
$(M_4,{}^*)$
are shown to be finitely based in Sections 4, 5 and 6, respectively.
The involution monoid
$(M_5,{}^*)$
is isomorphic to
$(A_0^1,{}^*)$
and so is nonfinitely based [Reference Gao, Zhang and Luo5], while it follows from Adair [Reference Adair1] that the identities of
$(M_6,{}^*)$
are axiomatised by (1.1) and
$$ \begin{align*} x^2 \approx x, \quad xx^*x \approx x, \quad xx^*yxy \approx xy, \quad xyxy^*y \approx xy. \end{align*} $$
4 The involution monoid
$(M_2,{}^*)$
If
$x, x^* \in \operatorname {\mathsf {con}} (\mathbf {u})$
for some
$x \in \mathcal {A}$
, then
$\{x, x^*\}$
is a mixed pair of
$\mathbf {u}$
. A word is mixed if it has some mixed pair. A word without mixed pairs is bipartite.
Lemma 4.1 (Lee [Reference Lee11, Lemma 9]).
Let
$\mathbf {u}$
and
$\mathbf {v}$
be any bipartite words such that
$\operatorname {\mathsf {con}}(\mathbf {u})=\operatorname {\mathsf {con}}(\mathbf {v})$
. Then an involution semigroup satisfies
$\mathbf {u} \approx \mathbf {v}$
if and only if it satisfies
$\overline {\mathbf {u}}\approx \overline {\mathbf {v}}$
.
Lemma 4.2. Let
$(M,{}^*)$
be any involution monoid that satisfies the identities
$$ \begin{align} x^*yx \approx xyx, \quad xyx^* \approx xyx, \quad x^*yx^* \approx xyx. \end{align} $$
Suppose that M is finitely based. Then
$(M,{}^*)$
is also finitely based.
Proof. There exists some set
$\Sigma $
of identities of
$(M,{}^*)$
such that
$\{$
(1.1), (4.1)
$\}$
$ \cup\ \Sigma $
is an identity basis for
$(M,{}^*)$
. In view of the identities (4.1), the identities in
$\Sigma $
can be assumed to be formed by words whose nonsimple variables are all plain; note that these words are bipartite. If
$\operatorname {\mathsf {con}}(\mathbf {u}) \neq \operatorname {\mathsf {con}}(\mathbf {v})$
for some
$\mathbf {u} \approx \mathbf {v} \in \Sigma $
, then
$(M,{}^*)$
satisfies either
$x^{a} \approx x^*$
or
$x^{b} \approx 1$
for some
$a, b\geq 1$
. Note that
$x\approx (x^*)^{a}\overset {({\scriptstyle 4.1})}\approx x^{a} \approx x^*$
if
$a \geq 2$
and
${x\approx x^{b+1}\overset {({\scriptstyle 4.1})}\approx x^{b}x^* \approx x^*}$
. Hence,
$(M,{}^*)$
satisfies the identity
$x \approx x^*$
, and so
$(M,{}^*)$
is finitely based since M is finitely based. If
$\operatorname {\mathsf {con}}(\mathbf {u}) = \operatorname {\mathsf {con}}(\mathbf {v})$
for all
$\mathbf {u} \approx \mathbf {v} \in \Sigma $
, then by Lemma 4.1, the identities in
$\Sigma $
can be chosen to be plain. In other words,
$\Sigma $
is a set of identities of the monoid M. By assumption, there exists a finite identity basis
$\Sigma _0$
for M, so that
$\Sigma _0$
implies
$\Sigma $
. Therefore,
$\{$
(1.1), (4.1)
$\}$
$ \cup\ \Sigma _0$
implies
$\{$
(1.1), (4.1)
$\}$
$\cup\ \Sigma $
and so is a finite identity basis for
$(M,{}^*)$
.
Corollary 4.3. The involution monoid
$(M_2,{}^*)$
is finitely based.
Proof. It is routine to check that the involution monoid
$(M_2,{}^*)$
satisfies the identities (4.1). Since
$M_2$
is finitely based [Reference Edmunds4], the result holds by Lemma 4.2.
5 The involution monoid
$(M_3,{}^*)$
Proposition 5.1. The identities (1.1) and
$$ \begin{align*} \begin{aligned} &\quad\qquad\quad x^4 \approx x^3, \quad xyx \approx x^2y, \quad xyx \approx yx^2, \\ &xyx^* \approx xx^*y, \quad xyx^* \approx yxx^*, \quad (x^*)^3 \approx x^3, \quad (x^*)^2 \approx x^2 \end{aligned} \end{align*} $$
constitute an identity basis for
$(M_3,{}^*)$
.
For any variables
$x_1,x_2, \ldots , x_n \in \mathcal {A}$
in strict alphabetical order, define
-
•
$x_1^{e_1} x_2^{e_2} \cdots x_n^{e_n}$
, where
$e_1,e_2,\ldots ,e_n \in \{ 2,3\}$
, to be a plain restricted word; -
•
$\mathbf {x}_1 \mathbf {x}_2 \cdots \mathbf {x}_n$
, where
$\mathbf {x}_i \in \{ x_ix_i^*, x_i^*x_i \}$
, to be a mixed restricted word.
It is easy to show that the identities in Proposition 5.1 can be used to convert every word in
$F_{\mathsf {inv}}^1(\mathcal {A})$
into some word of the form
$\mathbf {p}\mathbf {m}\mathbf {s}$
, where
$\mathbf {p} \in \mathcal {A}^+ \cup \{ 1 \}$
is a plain restricted word,
$\mathbf {m} \in F_{\mathsf {inv}}^1(\mathcal {A})$
is a mixed restricted word, and
$\mathbf {s} \in F_{\mathsf {inv}}^1(\mathcal {A})$
is a simple word such that the sets
$\operatorname {\mathsf {con}}(\mathbf {p})$
,
$\operatorname {\mathsf {con}}(\overline {\mathbf {m}})$
and
$\operatorname {\mathsf {con}}(\overline {\mathbf {s}})$
are pairwise disjoint; in this section, such a word
$\mathbf {p}\mathbf {m}\mathbf {s}$
is said to be in canonical form.
It is routine to check that
$(M_3,{}^*)$
satisfies the identities in Proposition 5.1. Let
${\mathbf {u}_1 \approx \mathbf {u}_2}$
be any identity of
$(M_3,{}^*)$
, where
$\mathbf {u}_i = \mathbf {p}_i\mathbf {m}_i\mathbf {s}_i$
is in canonical form for each
$i \in \{1,2\}$
. In the remainder of this section, it is shown that
$\mathbf {u}_1 = \mathbf {u}_2$
. This completes the proof of Proposition 5.1.
Lemma 5.2.
$\mathbf {p}_1 = \mathbf {p}_2$
.
Proof. Suppose that
$\mathbf {p}_1 \neq \mathbf {p}_2$
. Then there are two cases.
Case 1:
$\operatorname {\mathsf {con}}(\mathbf {p}_1) = \operatorname {\mathsf {con}}(\mathbf {p}_2)$
. Then
$\operatorname {\mathsf {occ}}(x,\mathbf {p}_1) \neq \operatorname {\mathsf {occ}}(x,\mathbf {p}_2)$
for some
$x \in \mathcal {A}$
, so that
$\{\mathbf {p}_1[x], \mathbf {p}_2[x]\} = \{ x^2, x^3\}$
by the definition of plain restricted words. It follows that
$\mathbf {u}_1[x] \approx \mathbf {u}_2[x]$
is the identity
$x^3 \approx x^2$
; but this identity is not satisfied by
$(M_3,{}^*)$
, giving a contradiction.
Case 2:
$\operatorname {\mathsf {con}}(\mathbf {p}_1) \neq \operatorname {\mathsf {con}}(\mathbf {p}_2)$
. Generality is not lost by assuming the existence of some
$x \in \operatorname {\mathsf {con}}(\mathbf {p}_1) \backslash \operatorname {\mathsf {con}}(\mathbf {p}_2)$
. If
$x \in \operatorname {\mathsf {con}}(\mathbf {m}_2)$
, then
$\mathbf {u}_1[x] \approx \mathbf {u}_2[x]$
is either
$x^2 \approx xx^*$
,
$x^2 \approx x^*x$
,
$x^3 \approx xx^*$
or
$x^3 \approx x^*x$
. If
$x \in \operatorname {\mathsf {con}}(\mathbf {s}_2)$
, then
$\mathbf {u}_1[x] \approx \mathbf {u}_2[x]$
is either
$x^2 \approx x$
,
$x^2 \approx x^*$
,
$x^3 \approx x$
or
$x^3 \approx x^*$
. If
$x \notin \operatorname {\mathsf {con}}(\mathbf {m}_2\mathbf {s}_2)$
, then
$\mathbf {u}_1[x] \approx \mathbf {u}_2[x]$
is either
$x^2 \approx 1$
or
$x^3 \approx 1$
. However, these ten identities are not satisfied by
$(M_3,{}^*)$
, giving a contradiction.
Lemma 5.3.
$\mathbf {m}_1 = \mathbf {m}_2$
.
Proof. Suppose that
$\mathbf {m}_1 \neq \mathbf {m}_2$
. Then there are two cases.
Case 1:
$\operatorname {\mathsf {con}}(\mathbf {m}_1) = \operatorname {\mathsf {con}}(\mathbf {m}_2) = \{ x_1,x_1^*, x_2,x_2^*, \ldots ,x_n,x_n^*\}$
for some variables
$x_1, x_2,\ldots, $
$x_n \in \mathcal {A}$
in strict alphabetical order. Then by the definition of mixed restricted words,
$\mathbf {m}_1 = \mathbf {x}_1 \mathbf {x}_2 \cdots \mathbf {x}_n$
and
$\mathbf {m}_2 = \mathbf {x}_1' \mathbf {x}_2' \cdots \mathbf {x}_n'$
, where
$\mathbf {x}_i, \mathbf {x}_i' \in \{ x_ix_i^*, x_i^*x_i \}$
for all i. The assumption
$\mathbf {m}_1 \neq \mathbf {m}_2$
implies that
$\mathbf {x}_j \neq \mathbf {x}_j'$
for some j. Therefore,
$\mathbf {u}_1[x_j] \approx \mathbf {u}_2[x_j]$
is the identity
$x_jx_j^* \approx x_j^*x_j$
; but this identity is not satisfied by
$(M_3,{}^*)$
, giving a contradiction.
Case 2:
$\operatorname {\mathsf {con}}(\mathbf {m}_1) \neq \operatorname {\mathsf {con}}(\mathbf {m}_2)$
. Generality is not lost by assuming the existence of some
$x \in \mathcal {A}$
such that
$x,x^* \in \operatorname {\mathsf {con}}(\mathbf {m}_1) \backslash \operatorname {\mathsf {con}}(\mathbf {m}_2)$
. If
$x \in \operatorname {\mathsf {con}}(\mathbf {p}_2)$
, then
$x \in \operatorname {\mathsf {con}}(\mathbf {p}_1)$
by Lemma 5.2, whence
$\operatorname {\mathsf {con}}(\mathbf {p}_1)$
and
$\operatorname {\mathsf {con}}(\overline {\mathbf {m}_1})$
are not disjoint, contradicting the choice of
$\mathbf {p}_1$
and
$\mathbf {m}_1$
. Hence,
$x \notin \operatorname {\mathsf {con}}(\mathbf {p}_2)$
. Clearly,
$x^* \notin \operatorname {\mathsf {con}}(\mathbf {p}_2)$
because the word
$\mathbf {p}_2$
is plain. Therefore, the remaining possibilities are
$x \in \operatorname {\mathsf {con}}(\overline {\mathbf {s}_2})$
and
$x \notin \operatorname {\mathsf {con}}(\overline {\mathbf {s}_2})$
. If
$x \in \operatorname {\mathsf {con}}(\overline {\mathbf {s}_2})$
, then
$\mathbf {u}_1[x] \approx \mathbf {u}_2[x]$
is either
$xx^* \approx x$
,
$xx^* \approx x^*$
,
$x^*x \approx x$
or
$x^*x \approx x^*$
. If
$x \notin \operatorname {\mathsf {con}}(\overline {\mathbf {s}_2})$
, then
$\mathbf {u}_1[x] \approx \mathbf {u}_2[x]$
is either
$xx^* \approx 1$
or
$x^*x \approx 1$
. However, these six identities are not satisfied by
$(M_3,{}^*)$
, giving a contradiction.
Lemma 5.4.
$\mathbf {s}_1 = \mathbf {s}_2$
.
Proof. Recall that
$(M_3,{}^*)$
satisfies
$\mathbf {p}_1\mathbf {m}_1\mathbf {s}_1 \approx \mathbf {p}_2\mathbf {m}_2\mathbf {s}_2$
, where for each
$i \in \{1,2\}$
, the sets
$\operatorname {\mathsf {con}}(\mathbf {p}_i)$
,
$\operatorname {\mathsf {con}}(\overline {\mathbf {m}}_i)$
and
$\operatorname {\mathsf {con}}(\overline {\mathbf {s}}_i)$
are pairwise disjoint. Since
$\mathbf {p}_1 = \mathbf {p}_2$
and
$\mathbf {m}_1 = \mathbf {m}_2$
by Lemmas 5.2 and 5.3, it follows that
$(M_3,{}^*)$
satisfies
$\mathbf {s}_1 \approx \mathbf {s}_2$
. It is then easy to show that
$\mathbf {s}_1 = \mathbf {s}_2$
.
6 A finite basis for
$(M_4,{}^*)$
Proposition 6.1. The identities (1.1) and
$$ \begin{align} xyxzx & \approx xyzx, \end{align} $$
$$ \begin{align} x^2y^2 & \approx y^2x^2, \end{align} $$
$$ \begin{align} xx^*y & \approx yxx^*, \end{align} $$
$$ \begin{align} x^*yxzx \approx x^*yxz&, \quad xzxyx^* \approx zxyx^* \end{align} $$
$$ \begin{align} xyx^*zx &\approx yzxx^*, \end{align} $$
$$ \begin{align} xx^* &\approx x^*x, \end{align} $$
$$ \begin{align} x^{\circledast_1} hxyky^{\circledast_2} &\approx x^{\circledast_1} hyxky^{\circledast_2}, \end{align} $$
$$ \begin{align} x^{\circledast_1} hy^{\circledast_2} kxy &\approx x^{\circledast_1} hy^{\circledast_2} kyx, \end{align} $$
$$ \begin{align} xyhx^{\circledast_1} ky^{\circledast_2} &\approx yxhx^{\circledast_1} ky^{\circledast_2}, \end{align} $$
where
$\circledast _1, \circledast _2 \in \{1,*\}$
, constitute an identity basis for
$(M_4,{}^*)$
.
Some basic results are given in Section 6.1. A canonical form for words forming identities of
$(M_4,{}^*)$
is given in Section 6.2. Results established in these two subsections are then used to prove Proposition 6.1 in Section 6.3.
Remark 6.2. The identities (6.1a)–(6.1d) actually imply the latter identities (6.1e)–(6.1i) and so constitute an identity basis for
$(M_4,{}^*)$
. However, as we will see shortly, the identities (6.1e)–(6.1i) are crucial to the proof of Proposition 6.1.
6.1 Basic results
Remark 6.3. It is routine to check that the involution monoid
$(M_4,{}^*)$
satisfies the identities (6.1) but not any of the identities
$$ \begin{align} \begin{aligned} xyx & \approx x^2y, \quad xyx^* \approx xx^*y, \quad xyx^* \approx x^*xy, \quad xyx^* \approx x^*yx, \\ xyx & \approx yx^2, \quad xyx^* \approx yxx^*, \quad xyx^* \approx yx^*x, \quad x^2y \approx yx^2. \end{aligned} \end{align} $$
A word
$\mathbf {u} \in F_{\mathsf {inv}}^1(\mathcal {A})$
is
$2$
-limited if for any
$x \in \mathcal {A}$
, the total number of times x and
$x^*$
occur in
$\mathbf {u}$
is at most two, that is,
$\operatorname {\mathsf {occ}}(x,\mathbf {u}) + \operatorname {\mathsf {occ}}(x^*,\mathbf {u}) \leq 2$
. An identity is
$2$
-limited if it is formed by a pair of 2-limited words.
Lemma 6.4. Given any word
$\mathbf {u} \in F_{\mathsf {inv}}^1(\mathcal {A})$
, there exists some
${\mathrm 2}$
-limited word
$\mathbf {u}'$
such that
$\operatorname {\mathsf {sim}}(\mathbf {u}) = \operatorname {\mathsf {sim}}(\mathbf {u}')$
,
$\operatorname {\mathsf {non}}(\mathbf {u}) = \operatorname {\mathsf {non}}(\mathbf {u}')$
and (6.1)
$\vdash \mathbf {u} \approx \mathbf {u}'$
.
Proof. It is easy to see that the identities
$\{$
(6.1a), (6.1d), (6.1e)
$\}$
can be used to convert any word
$\mathbf {u}$
into some
${\mathrm 2}$
-limited word
$\mathbf {u}'$
satisfying
$\operatorname {\mathsf {sim}}(\mathbf {u}) = \operatorname {\mathsf {sim}}(\mathbf {u}')$
and
${\operatorname {\mathsf {non}}(\mathbf {u}) = \operatorname {\mathsf {non}}(\mathbf {u}')}$
.
Define a relation
$\sim $
on
$F_{\mathsf {inv}}^1(\mathcal {A})$
by
$\mathbf {u} \sim \mathbf {v}$
if
$\mathbf {u}$
and
$\mathbf {v}$
are the same word up to arrangement of their variables. Equivalently,
$\mathbf {u} \sim \mathbf {v}$
if and only if
$xy \approx yx \vdash \mathbf {u} \approx \mathbf {v}$
.
Lemma 6.5. Let
$\mathbf {u} \approx \mathbf {v}$
be any
${\mathrm 2}$
-limited identity of
$(M_4,{}^*)$
. Then
-
(i)
$\operatorname {\mathsf {sim}}(\mathbf {u}) = \operatorname {\mathsf {sim}}(\mathbf {v})$
and
$\operatorname {\mathsf {non}}(\mathbf {u}) = \operatorname {\mathsf {non}}(\mathbf {v})$
; -
(ii)
$\mathbf {u} \sim \mathbf {v}$
; -
(iii)
$\mathbf {u}[\operatorname {\mathsf {sim}}(\mathbf {u})] = \mathbf {v}[\operatorname {\mathsf {sim}}(\mathbf {v})]$
.
Proof. (i) First, suppose that
$x \in \operatorname {\mathsf {con}}(\mathbf {u}) \backslash \operatorname {\mathsf {con}}(\mathbf {v})$
. Generality is not lost by assuming that
$x \in \mathcal {A}$
. Let
$\varphi : \mathcal {A} \to M_4$
denote the substitution that maps x to
$3$
and every other variable to
$5$
. Then
$$ \begin{align*} \mathbf{u}\varphi = \begin{cases} 1 & \text{if } x^* \in \operatorname{\mathsf{con}}(\mathbf{u}), \\ 3 & \text{otherwise}; \end{cases} \quad \mathbf{v}\varphi = \begin{cases} 4 & \text{if } x^* \in \operatorname{\mathsf{con}}(\mathbf{v}), \\ 5 & \text{otherwise}. \end{cases} \end{align*} $$
Therefore, the contradiction
$\mathbf {u}\varphi \neq \mathbf {v}\varphi $
is obtained. Hence, the variable x does not exist, so that
$\operatorname {\mathsf {con}}(\mathbf {u}) = \operatorname {\mathsf {con}}(\mathbf {v})$
.
Now suppose that
$x \in \operatorname {\mathsf {sim}}(\mathbf {u}) \backslash \operatorname {\mathsf {sim}}(\mathbf {v})$
. Since
$x \in \operatorname {\mathsf {con}}(\mathbf {u}) = \operatorname {\mathsf {con}}(\mathbf {v})$
, we have
$x \in \operatorname {\mathsf {non}}(\mathbf {v})$
. Let
$\psi : \mathcal {A} \to M_4$
denote the substitution that maps x to
$2$
and every other variable to
$5$
. Then
$\mathbf {u} \psi = 2$
and
$\mathbf {v} \psi = 1$
, resulting in the contradiction
$\mathbf {u}\psi \neq \mathbf {v}\psi $
. Therefore, the variable x does not exist, so that
$\operatorname {\mathsf {sim}}(\mathbf {u}) = \operatorname {\mathsf {sim}}(\mathbf {v})$
; this, together with
$\operatorname {\mathsf {con}}(\mathbf {u}) = \operatorname {\mathsf {con}}(\mathbf {v})$
, implies that
$\operatorname {\mathsf {non}}(\mathbf {u}) = \operatorname {\mathsf {non}}(\mathbf {v})$
.
(ii) This is an easy consequence of part (i) because
$\mathbf {u}$
and
$\mathbf {v}$
are 2-limited words.
(iii) Suppose that
$\mathbf {u}[\operatorname {\mathsf {sim}}(\mathbf {u})] \neq \mathbf {v}[\operatorname {\mathsf {sim}}(\mathbf {v})]$
. Then there exist
$x, y \in \operatorname {\mathsf {sim}} (\mathbf {u}) = \operatorname {\mathsf {sim}} (\mathbf {v})$
such that x precedes y in
$\mathbf {u}$
but y precedes x in
$\mathbf {v}$
. Hence,
$\mathbf {u} [x,y] \approx \mathbf {v} [x,y]$
is the identity
$xy \approx yx$
, which implies that
$(M_4,{}^*)$
is commutative, a contradiction.
Lemma 6.6. Let
$\mathbf {u} \approx \mathbf {v}$
be any
${\mathrm 2}$
-limited identity of
$(M_4,{}^*)$
.
-
(i) Suppose that either
$\operatorname {\mathsf {sim}}(\mathbf {u}) = \emptyset $
or
$\operatorname {\mathsf {sim}}(\mathbf {v}) = \emptyset $
. Then (6.1)
$\vdash \mathbf {u} \approx \mathbf {v}$
. -
(ii) Suppose that either
$\operatorname {\mathsf {non}}(\mathbf {u}) = \emptyset $
or
$\operatorname {\mathsf {non}}(\mathbf {v}) = \emptyset $
. Then
$\mathbf {u} = \mathbf {v}$
.
Proof. (i) By Lemma 6.5(i), we have
$\operatorname {\mathsf {sim}}(\mathbf {u}) = \operatorname {\mathsf {sim}}(\mathbf {v}) = \emptyset $
and
$\operatorname {\mathsf {non}}(\mathbf {u}) = \operatorname {\mathsf {non}}(\mathbf {v})$
, so that both
$\mathbf {u}$
and
$\mathbf {v}$
consist entirely of nonsimple variables. Since
$\mathbf {u} \sim \mathbf {v}$
by Lemma 6.5(ii), the identities (6.1g)–(6.1i) can be used to convert
$\mathbf {u}$
into
$\mathbf {v}$
.
(ii) By Lemma 6.5(i), we have
$\operatorname {\mathsf {sim}}(\mathbf {u}) = \operatorname {\mathsf {sim}}(\mathbf {v})$
and
$\operatorname {\mathsf {non}}(\mathbf {u}) = \operatorname {\mathsf {non}}(\mathbf {v}) = \emptyset $
, so that both
$\mathbf {u}$
and
$\mathbf {v}$
are simple words. Therefore,
$\mathbf {u} = \mathbf {u}[\operatorname {\mathsf {sim}}(\mathbf {u})] = \mathbf {v}[\operatorname {\mathsf {sim}}(\mathbf {v})] = \mathbf {v}$
by Lemma 6.5(iii).
6.2 Canonical form
Any alphabetical order
$\prec $
on
$\mathcal {A}$
can be extended to a total order
$\prec $
on
$\mathcal {A} \cup \mathcal {A}^*$
in the following manner:
$x \prec x^*$
for all
$x \in \mathcal {A}$
and for all
$x,y \in \mathcal {A} \cup \mathcal {A}^*$
, define
$x \prec y$
if
$\overline {x} \prec \overline {y}$
. An ordered word is a word of the form
$$ \begin{align*} x_1^{e_1} x_2^{e_2} \cdots x_n^{e_n}, \end{align*} $$
where
$x_1,x_2, \ldots ,x_n \in \mathcal {A} \cup \mathcal {A}^*$
with
$x_1 \prec x_2 \prec \cdots \prec x_n$
and
$e_1,e_2,\ldots ,e_n \geq 1$
.
In this section, a 2-limited word
$\mathbf {u}$
with
$\operatorname {\mathsf {sim}}(\mathbf {u}) \neq \emptyset $
and
$\operatorname {\mathsf {non}}(\mathbf {u}) \neq \emptyset $
is said to be in canonical form if
$$ \begin{align} \mathbf{u} = \mathbf{u}_0 \prod_{i=1}^m (\mathbf{s}_i\mathbf{u}_i) \end{align} $$
for some
$m \geq 1$
, where
-
(CF1)
$\mathbf {u}_0, \mathbf {s}_1, \mathbf {u}_m \in F_{\mathsf {inv}}^1(\mathcal {A})$
and
$\mathbf {s}_2, \mathbf {s}_3, \ldots , \mathbf {s}_m, \mathbf {u}_1, \mathbf {u}_2, \ldots , \mathbf {u}_{m-1} \in F_{\mathsf {inv}}(\mathcal {A})$
; -
(CF2)
$\mathbf {u}[\operatorname {\mathsf {sim}}(\mathbf {u})] = \mathbf {s}_1 \mathbf {s}_2 \cdots \mathbf {s}_m$
; -
(CF3)
$\mathbf {u}_0 = x_1x_1^* x_2x_2^* \cdots x_rx_r^*$
for some
$x_1,x_2,\ldots ,x_r \in \mathcal {A}$
with
$x_1 \prec x_2 \prec \cdots \prec x_r$
and
$r \geq 0$
; -
(CF4)
$\mathbf {u}_1, \mathbf {u}_2, \ldots , \mathbf {u}_{m-1} \in \operatorname {\mathsf {non}}(\mathbf {u})^+$
and
$\mathbf {u}_m \in \operatorname {\mathsf {non}}(\mathbf {u})^+ \cup \{1\}$
are bipartite ordered words.
Lemma 6.7. Let
$\mathbf {u}$
be any
${\mathrm 2}$
-limited word such that
$\operatorname {\mathsf {sim}}(\mathbf {u}) \neq \emptyset $
and
$\operatorname {\mathsf {non}}(\mathbf {u}) \neq \emptyset $
. Then the identities (6.1) can be used to convert
$\mathbf {u}$
into a word in canonical form.
Proof. Write
$\mathbf {u} = \prod _{i=1}^m (\mathbf {s}_i\mathbf {u}_i)$
, where
$\mathbf {s}_1 \in F_{\mathsf {inv}}^1(\mathcal {A})$
and
$\mathbf {s}_2, \mathbf {s}_3, \ldots , \mathbf {s}_m \in F_{\mathsf {inv}}(\mathcal {A})$
are maximal factors of
$\mathbf {u}$
formed by simple variables, and
$\mathbf {u}_1, \mathbf {u}_2, \ldots , \mathbf {u}_{m-1} \in F_{\mathsf {inv}}(\mathcal {A})$
and
$\mathbf {u}_m \in F_{\mathsf {inv}}^1(\mathcal {A})$
are maximal factors of
$\mathbf {u}$
formed by nonsimple variables.
Suppose that some
$\mathbf {u}_i$
contains a mixed pair
$\{ x,x^* \}$
. Then apply the identities (6.1f)–(6.1i) to group x and
$x^*$
together as some factor
$xx^*$
of
$\mathbf {u}_i$
, and apply the identity (6.1c) to move
$xx^*$
to the left of
$\mathbf {s}_1$
.
The procedure in the previous paragraph can be repeated on every mixed pair of every
$\mathbf {u}_i$
, so that every
$\mathbf {u}_i$
no longer has a mixed pair and so is bipartite. The factors of the form
$xx^*$
that are collected on the left of
$\mathbf {s}_1$
can be rearranged by the identity (6.1c) to form the prefix
$\mathbf {u}_0$
satisfying (CF3). Note that since
$\mathbf {u}$
is 2-limited, the prefix
$\mathbf {u}_0$
does not share any variable with the rest of the word.
Therefore,
$\mathbf {u} = \mathbf {u}_0 \prod _{i=1}^m (\mathbf {s}_i\mathbf {u}_i')$
, where each
$\mathbf {u}_i'$
is a bipartite word obtained from
$\mathbf {u}_i$
by removing all its mixed pairs. If
$\mathbf {u}_i'$
is empty for some
$i < m$
, then
$\mathbf {s}_i$
and
$\mathbf {s}_{i+1}$
can be combined into a single maximal factor of
$\mathbf {u}$
formed by simple variables:
$$ \begin{align*} \mathbf{u} = \mathbf{u}_0 \cdots \mathbf{s}_i \mathbf{u}_i' \cdot \mathbf{s}_{i+1} \mathbf{u}_{i+1}' \cdots = \mathbf{u}_0 \cdots (\mathbf{s}_i \cdot \mathbf{s}_{i+1}) \mathbf{u}_{i+1}' \cdots. \end{align*} $$
The resulting word is of the form (6.3) satisfying (CF1). Now apply the identities (6.1g)–(6.1i) to rearrange each
$\mathbf {u}_i$
(
$1 \leq i \leq m$
) into an ordered word, so that (CF4) is satisfied. It is clear that (CF2) is also satisfied since no simple variable has been introduced or removed, and the order of appearance of the simple variables has not been changed.
6.3 Proof of Proposition 6.1
It suffices to show that any identity
$\mathbf {u} \approx \mathbf {v}$
of
$(M_4,{}^*)$
is deducible from the identities (6.1). By Lemmas 6.4 and 6.5, we may further assume that
-
(a)
$\mathbf {u}$
and
$\mathbf {v}$
are 2-limited; -
(b)
$\mathbf {u} \sim \mathbf {v}$
; -
(c)
$\mathbf {u}[\operatorname {\mathsf {sim}}(\mathbf {u})] = \mathbf {v}[\operatorname {\mathsf {sim}}(\mathbf {v})]$
.
If either
$\operatorname {\mathsf {sim}}(\mathbf {u}) = \operatorname {\mathsf {sim}}(\mathbf {u}) = \emptyset $
or
$\operatorname {\mathsf {non}}(\mathbf {u}) = \operatorname {\mathsf {non}}(\mathbf {v}) = \emptyset $
, then (6.1)
$\vdash \mathbf {u} \approx \mathbf {v}$
by Lemma 6.6. Therefore, it remains to address the case when
$\operatorname {\mathsf {sim}}(\mathbf {u}) = \operatorname {\mathsf {sim}}(\mathbf {v}) \neq \emptyset $
and
$\operatorname {\mathsf {non}}(\mathbf {u}) = \operatorname {\mathsf {non}}(\mathbf {v}) \neq \emptyset $
, whence by Lemma 6.7, the words
$\mathbf {u}$
and
$\mathbf {v}$
can be assumed to be in canonical form, say
$$ \begin{align*} \mathbf{u} = \mathbf{u}_0 \prod_{i=1}^m (\mathbf{s}_i\mathbf{u}_i) \quad \text{and} \quad \mathbf{v} = \mathbf{v}_0 \prod_{i=1}^n (\mathbf{t}_i\mathbf{v}_i). \end{align*} $$
It follows from (a) and (CF3) that
-
(d)
$\operatorname {\mathsf {con}}(\mathbf {u}_0) \cap \operatorname {\mathsf {con}}(\mathbf {s}_i\mathbf {u}_i) = \operatorname {\mathsf {con}}(\mathbf {v}_0) \cap \operatorname {\mathsf {con}}(\mathbf {t}_i\mathbf {v}_i) = \emptyset $
for all
$i \geq 1$
.
The results in the remainder of this subsection verify that
$\mathbf {u} = \mathbf {v}$
. The proof of Proposition 6.1 is therefore complete.
Lemma 6.8.
$m = n$
and
$\mathbf {s}_i = \mathbf {t}_i$
for all i.
Proof. Suppose that
$y_1,y_2 \in \operatorname {\mathsf {sim}}(\mathbf {u}) = \operatorname {\mathsf {sim}}(\mathbf {v})$
are such that
$y_1y_2$
is a factor of
$\mathbf {u}$
but not of
$\mathbf {v}$
. Then
$y_1y_2$
is a factor of some
$\mathbf {s}_j$
but is not a factor of any
$\mathbf {t}_1,\mathbf {t}_2,\ldots ,\mathbf {t}_n$
. However, since
$\mathbf {s}_1 \mathbf {s}_2 \cdots \mathbf {s}_m = \mathbf {t}_1 \mathbf {t}_2 \cdots \mathbf {t}_n$
by (c), the word
$y_1y_2$
is a factor of
$\mathbf {t}_1 \mathbf {t}_2 \cdots \mathbf {t}_n$
. It follows that for some j, the last variable of
$\mathbf {t}_j$
is
$y_1$
and the first variable of
$\mathbf {t}_{j+1}$
is
$y_2$
; in other words,
$y_1 \mathbf {v}_j y_2$
is a factor of
$\mathbf {v}$
. By (CF4), the factor
$\mathbf {v}_j$
contains some nonsimple variable of
$\mathbf {v}$
, say
$x^\circledast $
with
$\circledast \in \{1,*\}$
. Then by (a) and (CF4),
$$ \begin{align*} \mathbf{u}[x,y_1,y_2] & \in \{ x^{\circledast_1} x^{\circledast_2} y_1y_2, \, x^{\circledast_1} y_1y_2 x^{\circledast_2}, \, y_1y_2 x^{\circledast_1} x^{\circledast_2}\} \quad\mbox{and} \\ \mathbf{v}[x,y_1,y_2] & \in \{ x^{\circledast_3} y_1 x^{\circledast_4} y_2, \, y_1(x^{\circledast_3})^2 y_2, \, y_1 x^{\circledast_3} y_2 x^{\circledast_4} \} \end{align*} $$
for some
$\circledast _1,\circledast _2,\circledast _3,\circledast _4 \in \{ 1,*\}$
. Now (b) implies that
$\mathbf {u}[x,y_1,y_2] \sim \mathbf {v}[x,y_1,y_2]$
, whence it is routine to check that for any
$\mathbf {u}[x,y_1,y_2] \approx \mathbf {v}[x,y_1,y_2]$
, there exists an appropriate
$i \in \{1,2\}$
such that
$\mathbf {u}[x,y_i] \approx \mathbf {v}[x,y_i]$
is one of the following identities:
$$ \begin{align*} x^2 y_i \approx y_i x^2, \quad (x^*)^2 y_i \approx y_i (x^*)^2, \quad x^{\circledast_1} x^{\circledast_2} y_i \approx x^{\circledast_3} y_i x^{\circledast_4}, \quad y_i x^{\circledast_1} x^{\circledast_2} \approx x^{\circledast_3} y_i x^{\circledast_4}, \end{align*} $$
where
$\circledast _1,\circledast _2,\circledast _3,\circledast _4 \in \{ 1,*\}$
are such that
$\{\circledast _1,\circledast _2\} = \{ \circledast _3,\circledast _4 \}$
. However, by Remark 6.3, none of these identities is satisfied by
$(M_4,{}^*)$
, so we have a contradiction.
Therefore, for any
$y_1,y_2 \in \operatorname {\mathsf {sim}}(\mathbf {u}) = \operatorname {\mathsf {sim}}(\mathbf {v})$
, the word
$y_1y_2$
is a factor of
$\mathbf {u}$
if and only if it is a factor of
$\mathbf {v}$
. The present lemma thus follows from (c).
Lemma 6.9.
$\mathbf {u}_0 = \mathbf {v}_0$
.
Proof. Suppose that
$\operatorname {\mathsf {con}}(\mathbf {u}_0) \neq \operatorname {\mathsf {con}}(\mathbf {v}_0)$
, say
$x,x^* \in \operatorname {\mathsf {con}}(\mathbf {u}_0) \backslash \operatorname {\mathsf {con}}(\mathbf {v}_0)$
. Then since
$x,x^* \in \operatorname {\mathsf {non}}(\mathbf {v})$
by (b), the variables
$x,x^*$
occur in the factors
$\mathbf {v}_1, \mathbf {v}_2, \ldots , \mathbf {v}_n$
. However, these factors are bipartite by (CF4), so the variables
$x,x^*$
cannot occur in the same
$\mathbf {v}_i$
, whence their occurrence in
$\mathbf {v}$
must sandwich some simple variable y. Then
$\mathbf {u}[x,y] = xx^*y$
and
$\mathbf {v}[x,y] \in \{ xyx^*, x^*yx \}$
. It follows that
$(M_4,{}^*)$
satisfies an identity from (6.2), which is impossible by Remark 6.3. Therefore,
$\operatorname {\mathsf {con}}(\mathbf {u}_0) = \operatorname {\mathsf {con}}(\mathbf {v}_0)$
, whence
$\mathbf {u}_0 = \mathbf {v}_0$
by (CF3).
Lemma 6.10.
$\mathbf {u}_m = \mathbf {v}_m$
.
Proof. Suppose that
$\operatorname {\mathsf {con}}(\mathbf {u}_m) \neq \operatorname {\mathsf {con}}(\mathbf {v}_m)$
, say
$x \in \operatorname {\mathsf {con}}(\mathbf {u}_m) \backslash \operatorname {\mathsf {con}}(\mathbf {v}_m)$
. Generality is not lost by assuming that
$x \in \mathcal {A}$
. It follows from (a), (b), (d) and (CF4) that there are three cases. (In each case, let y be any simple variable in
$\mathbf {s}_m$
.)
Case 1:
$x^\circledast \in \operatorname {\mathsf {con}}(\mathbf {u}_i)$
for some
$i \in \{ 1,2,\ldots , m-1\}$
with
$\circledast \in \{ 1,* \}$
and
$x^* \notin \operatorname {\mathsf {con}}(\mathbf {v}_m)$
. Then
$$ \begin{align*} \mathbf{u} & = \mathbf{u}_0 \cdot \mathbf{s}_1 \underbrace{\mathbf{u}_1 \cdot \mathbf{s}_2 \mathbf{u}_2 \cdots \mathbf{s}_{m-1} \mathbf{u}_{m-1}}_{x^\circledast} \cdot \, \mathbf{s}_m \underbrace{\mathbf{u}_m}_x, \\ \mathbf{v} & = \mathbf{v}_0 \cdot \mathbf{s}_1 \underbrace{\mathbf{v}_1 \cdot \mathbf{s}_2 \mathbf{v}_2 \cdots \mathbf{s}_{m-1} \mathbf{v}_{m-1}}_{x \text{ and } x^\circledast} \cdot \, \mathbf{s}_m \mathbf{v}_m. \end{align*} $$
Hence,
$\mathbf {u}[x,y] \approx \mathbf {v}[x,y]$
is either
$x^\circledast yx \approx xx^\circledast y$
or
$x^\circledast yx \approx x^\circledast xy$
, which contradicts Remark 6.3.
Case 2:
$x^* \in \operatorname {\mathsf {con}}(\mathbf {u}_i)$
for some
$i \in \{ 1,2,\ldots , m-1\}$
and
$x^* \in \operatorname {\mathsf {con}}(\mathbf {v}_m)$
. Then
$$ \begin{align*} \mathbf{u} & = \mathbf{u}_0 \cdot \mathbf{s}_1 \underbrace{\mathbf{u}_1 \cdot \mathbf{s}_2 \mathbf{u}_2 \cdots \mathbf{s}_{m-1} \mathbf{u}_{m-1}}_{x^*} \cdot \, \mathbf{s}_m \underbrace{\mathbf{u}_m}_x, \\ \mathbf{v} & = \mathbf{v}_0 \cdot \mathbf{s}_1 \underbrace{\mathbf{v}_1 \cdot \mathbf{s}_2 \mathbf{v}_2 \cdots \mathbf{s}_{m-1} \mathbf{v}_{m-1}}_x \cdot \, \mathbf{s}_m \underbrace{\mathbf{v}_m}_{x^*}. \end{align*} $$
Hence,
$\mathbf {u}[x,y] \approx \mathbf {v}[x,y]$
is
$x^* yx \approx xyx^*$
, which contradicts Remark 6.3.
Case 3:
$\operatorname {\mathsf {occ}}(x,\mathbf {u}_m)=2$
. Then
$$ \begin{align*} \mathbf{u} & = \mathbf{u}_0 \cdot \mathbf{s}_1 \mathbf{u}_1 \cdot \mathbf{s}_2 \mathbf{u}_2 \cdots \mathbf{s}_{m-1} \mathbf{u}_{m-1} \cdot \mathbf{s}_m \underbrace{\mathbf{u}_m}_{x^2}, \\ \mathbf{v} & = \mathbf{v}_0 \cdot \mathbf{s}_1 \underbrace{\mathbf{v}_1 \cdot \mathbf{s}_2 \mathbf{v}_2 \cdots \mathbf{s}_{m-1} \mathbf{v}_{m-1}}_{\text{two occurrences of } x} \cdot \, \mathbf{s}_m \mathbf{v}_m. \end{align*} $$
Hence,
$\mathbf {u}[x,y] \approx \mathbf {v}[x,y]$
is
$yx^2 \approx x^2y$
, which contradicts Remark 6.3.
Since all three cases are impossible, we must have
$\operatorname {\mathsf {con}}(\mathbf {u}_m) = \operatorname {\mathsf {con}}(\mathbf {v}_m)$
.
Now suppose that
$\mathbf {u}_m \neq \mathbf {v}_m$
. Then by (CF4), there exists some
$x \in \operatorname {\mathsf {non}}(\mathbf {u}) = \operatorname {\mathsf {non}}(\mathbf {v})$
such that
$\operatorname {\mathsf {occ}}(x,\mathbf {u}_m) \neq \operatorname {\mathsf {occ}}(x,\mathbf {v}_m)$
. Generality is not lost by assuming that
${\operatorname {\mathsf {occ}}(x,\mathbf {u}_m) =2}$
and
$\operatorname {\mathsf {occ}}(x,\mathbf {v}_m)=1$
with
$x \in \mathcal {A}$
. Then
$$ \begin{align*} \mathbf{u} & = \mathbf{u}_0 \cdot \mathbf{s}_1 \mathbf{u}_1 \cdots \mathbf{s}_{m-1} \mathbf{u}_{m-1} \cdot \mathbf{s}_m \underbrace{\mathbf{u}_m}_{x^2}, \\ \mathbf{v} & = \mathbf{v}_0 \cdot \mathbf{s}_1 \underbrace{\mathbf{v}_1 \cdots \mathbf{s}_{m-1} \mathbf{v}_{m-1}}_x \cdot \, \mathbf{s}_m \underbrace{\mathbf{v}_m}_x. \end{align*} $$
Let y be any simple variable in
$\mathbf {s}_m$
. Then,
$\mathbf {u}[x,y] \approx \mathbf {v}[x,y]$
is
$yx^2 \approx xyx$
, which contradicts Remark 6.3. Consequently,
$\mathbf {u}_m = \mathbf {v}_m$
.
Lemma 6.11.
$\mathbf {u}_i = \mathbf {v}_i$
for all
$i = 1,2, \ldots , m-1$
.
Proof. Suppose that
$\ell \in \{1,2,\ldots ,m-1\}$
is the least index such that
$\mathbf {u}_\ell \neq \mathbf {v}_\ell $
. Then
$\mathbf {u}_i = \mathbf {v}_i$
for all
$i < \ell $
. First, suppose that
$\operatorname {\mathsf {con}}(\mathbf {u}_\ell ) \neq \operatorname {\mathsf {con}}(\mathbf {v}_\ell )$
. Then generality is not lost by assuming the existence of some plain variable
$x \in \operatorname {\mathsf {con}}(\mathbf {u}_\ell ) \backslash \operatorname {\mathsf {con}}(\mathbf {v}_\ell )$
. It follows from (a), (b), (d) and (CF4) that there are four cases. (In each case, let y be any simple variable in
$\mathbf {s}_{\ell +1}$
.)
Case 1:
$x^\circledast \in \operatorname {\mathsf {con}}(\mathbf {u}_i)$
for some
$i \in \{ 1,2,\ldots , \ell -1\}$
with
$\circledast \in \{ 1,* \}$
and
$x^* \notin \operatorname {\mathsf {con}}(\mathbf {v}_\ell )$
. Then
$$ \begin{align*} \mathbf{u} & = \mathbf{u}_0 \cdot \mathbf{s}_1 \underbrace{\mathbf{u}_1 \cdots \mathbf{s}_{\ell-1} \mathbf{u}_{\ell-1}}_{x^\circledast} \cdot \, \mathbf{s}_\ell \underbrace{\mathbf{u}_\ell}_x \cdot \, \mathbf{s}_{\ell+1} \mathbf{u}_{\ell+1} \cdots \mathbf{s}_m \mathbf{u}_m, \\ \mathbf{v} & = \mathbf{u}_0 \cdot \mathbf{s}_1 \underbrace{\mathbf{u}_1 \cdots \mathbf{s}_{\ell-1} \mathbf{u}_{\ell-1}}_{x^\circledast} \cdot \, \mathbf{s}_\ell \mathbf{v}_\ell \cdot \mathbf{s}_{\ell+1} \underbrace{\mathbf{v}_{\ell+1} \cdots \mathbf{s}_m \mathbf{v}_m}_x. \end{align*} $$
Hence,
$\mathbf {u}[x,y] \approx \mathbf {v}[x,y]$
is
$x^\circledast xy \approx x^\circledast yx$
, which contradicts Remark 6.3.
Case 2:
$\operatorname {\mathsf {occ}}(x,\mathbf {u}_\ell )=2$
. Then
$$ \begin{align*} \mathbf{u} & = \mathbf{u}_0 \cdot \mathbf{s}_1 \mathbf{u}_1 \cdots \mathbf{s}_{\ell-1} \mathbf{u}_{\ell-1} \cdot \mathbf{s}_\ell \underbrace{\mathbf{u}_\ell}_{x^2} \cdot \, \mathbf{s}_{\ell+1} \mathbf{u}_{\ell+1} \cdots \mathbf{s}_m \mathbf{u}_m, \\ \mathbf{v} & = \mathbf{u}_0 \cdot \mathbf{s}_1 \mathbf{u}_1 \cdots \mathbf{s}_{\ell-1} \mathbf{u}_{\ell-1} \cdot \mathbf{s}_\ell \mathbf{v}_\ell \cdot \mathbf{s}_{\ell+1} \underbrace{\mathbf{v}_{\ell+1} \cdots \mathbf{s}_m \mathbf{v}_m}_{\text{two occurrences of } x}. \end{align*} $$
Hence,
$\mathbf {u}[x,y] \approx \mathbf {v}[x,y]$
is
$x^2y \approx yx^2$
, which contradicts Remark 6.3.
Case 3:
$x^\circledast \in \operatorname {\mathsf {con}}(\mathbf {u}_i)$
for some
$i \in \{\ell +1, \ell +2, \ldots , m\}$
with
$\circledast \in \{ 1,* \}$
and
$x^* \notin \operatorname {\mathsf {con}}(\mathbf {v}_\ell )$
. Then
$$ \begin{align*} \mathbf{u} & = \mathbf{u}_0 \cdot \mathbf{s}_1 \mathbf{u}_1 \cdots \mathbf{s}_{\ell-1} \mathbf{u}_{\ell-1} \cdot \mathbf{s}_\ell \underbrace{\mathbf{u}_\ell}_x \cdot \, \mathbf{s}_{\ell+1} \, \underbrace{\mathbf{u}_{\ell+1} \cdots \mathbf{s}_m \mathbf{u}_m}_{x^\circledast}, \\ \mathbf{v} & = \mathbf{u}_0 \cdot \mathbf{s}_1 \mathbf{u}_1 \cdots \mathbf{s}_{\ell-1} \mathbf{u}_{\ell-1} \cdot \mathbf{s}_\ell \mathbf{v}_\ell \cdot \mathbf{s}_{\ell+1} \, \underbrace{\mathbf{v}_{\ell+1} \cdots \mathbf{s}_m \mathbf{v}_m}_{x \text{ and } x^\circledast}. \end{align*} $$
Hence,
$\mathbf {u}[x,y] \approx \mathbf {v}[x,y]$
is either
$xyx^\circledast \approx yxx^\circledast $
or
$xyx^\circledast \approx yx^\circledast x$
, which contradicts Remark 6.3.
Case 4:
$x^* \in \operatorname {\mathsf {con}}(\mathbf {u}_i)$
for some
$i \in \{\ell +1, \ell +2, \ldots , m\}$
and
$x^* \in \operatorname {\mathsf {con}}(\mathbf {v}_\ell )$
. Then
$$ \begin{align*} \mathbf{u} & = \mathbf{u}_0 \cdot \mathbf{s}_1 \mathbf{u}_1 \cdots \mathbf{s}_{\ell-1} \mathbf{u}_{\ell-1} \cdot \mathbf{s}_\ell \underbrace{\mathbf{u}_\ell}_x \cdot \, \mathbf{s}_{\ell+1} \, \underbrace{\mathbf{u}_{\ell+1} \cdots \mathbf{s}_m \mathbf{u}_m}_{x^*}, \\ \mathbf{v} & = \mathbf{u}_0 \cdot \mathbf{s}_1 \mathbf{u}_1 \cdots \mathbf{s}_{\ell-1} \mathbf{u}_{\ell-1} \cdot \mathbf{s}_\ell \underbrace{\mathbf{v}_\ell}_{x^*} \cdot \, \mathbf{s}_{\ell+1} \, \underbrace{\mathbf{v}_{\ell+1} \cdots \mathbf{s}_m \mathbf{v}_m}_x. \end{align*} $$
Hence,
$\mathbf {u}[x,y] \approx \mathbf {v}[x,y]$
is
$xyx^* \approx x^*yx$
, which contradicts Remark 6.3.
Since all four cases are impossible, we must have
$\operatorname {\mathsf {con}}(\mathbf {u}_\ell ) = \operatorname {\mathsf {con}}(\mathbf {v}_\ell )$
. Then by (CF4), there exists some
$x \in \operatorname {\mathsf {non}}(\mathbf {u}) = \operatorname {\mathsf {non}}(\mathbf {v})$
such that
$\operatorname {\mathsf {occ}}(x,\mathbf {u}_\ell ) \neq \operatorname {\mathsf {occ}}(x,\mathbf {v}_\ell )$
. Generality is not lost by assuming
$\operatorname {\mathsf {occ}}(x,\mathbf {u}_\ell ) = 2$
and
$\operatorname {\mathsf {occ}}(x,\mathbf {v}_\ell )=1$
with
$x \in \mathcal {A}$
. Then
$$ \begin{align*} \mathbf{u} & = \mathbf{u}_0 \cdot \mathbf{s}_1 \mathbf{u}_1 \cdots \mathbf{s}_{\ell-1} \mathbf{u}_{\ell-1} \cdot \mathbf{s}_\ell \underbrace{\mathbf{u}_\ell}_{x^2} \cdot \, \mathbf{s}_{\ell+1} \mathbf{u}_{\ell+1} \cdots \mathbf{s}_m \mathbf{u}_m, \\ \mathbf{v} & = \mathbf{u}_0 \cdot \mathbf{s}_1 \mathbf{u}_1 \cdots \mathbf{s}_{\ell-1} \mathbf{u}_{\ell-1} \cdot \mathbf{s}_\ell \underbrace{\mathbf{v}_\ell}_x \cdot \, \mathbf{s}_{\ell+1} \, \underbrace{\mathbf{v}_{\ell+1} \cdots \mathbf{s}_m \mathbf{v}_m}_x. \end{align*} $$
Let y be any simple variable in
$\mathbf {s}_{\ell +1}$
. Then,
$\mathbf {u}[x,y] \approx \mathbf {v}[x,y]$
is
$x^2y \approx xyx$
, which contradicts Remark 6.3. Consequently, the index
$\ell $
does not exist and the present lemma is established.
Acknowledgements
The authors are very grateful to the anonymous referees whose careful reading and helpful suggestions led to the improvement of this paper. They also thank Edmond W. H. Lee for pointing out that every involution monoid of order four is commutative and for his help in checking and revising this paper.
