Startseite Mathematik Mixed identities, hereditarily separated actions and oscillation
Artikel Öffentlich zugĂ€nglich

Mixed identities, hereditarily separated actions and oscillation

  • Aleksander Ivanov ORCID logo EMAIL logo und Roland Zarzycki
Veröffentlicht/Copyright: 5. Dezember 2024

Abstract

Given a topological đș-space, we consider equations with constants over đș. In particular, we formulate some very general conditions on words with constants w ⁹ ( y ̄ , g ̄ ) over đș which guarantee that the inequality w ⁹ ( y ̄ , g ̄ ) ≠ 1 has a solution in đș. These results are illustrated in some typical situations. In particular, standard actions of Thompson’s group đč and branch groups are considered.

1 Introduction

Let F n denote the free group of rank 𝑛. For w ∈ F n ∗ G , the equality w = 1 is called a mixed group identity over đș. When it is satisfied for all elements of đș, we call it a law with constants of đș. In this paper, we study mixed identities/laws with constants in groups of homeomorphisms of some natural topological/metric spaces.

Investigations of mixed identities were initiated by several Russian mathematicians in the 1970s and 80s; see [3, 2, 7, 20]. The topic was later developed in [8, 16, 19]. Very recently, groups without mixed identities (MIF groups) have become an attractive object of investigations under some geometric assumptions. In particular, Hull and Osin have observed in [10] that acylindrically hyperbolic groups with trivial finite radical are MIF. Papers [10, 11] contain many other interesting examples in connection with some dynamical and geometric properties studied in group theory. We also mention an observation from [10] that MIF can be reformulated in model-theoretic terms (and even in terms of algebraic geometry, as in [14, 15, 17]).

Investigation of mixed identities is fruitful in the context of permutation groups. For example, [10, Theorem 1.6] states that, in the class of highly transitive countable groups, property MIF is opposite to embeddability of Alt ⁹ ( N ) as a normal subgroup. It is shown in [6] that dense countable subgroups of automorphism/isometry groups of typical structures/spaces with some properties of universality are MIF. On the other hand, it is proved in [22, 18] that Thompson’s group đč is not MIF.

The latter fact together with Abert’s theorem that Thompson’s group đč does not satisfy any identity was the starting point of this paper. What mixed identities are not satisfied in Thompson’s group đč? Generalizing Abert’s property of separation we introduce hereditarily separating đș-spaces and prove that, under this condition, the group đș does not satisfy so-called oscillating identities with constants. Since the standard action of đč on [ 0 , 1 ] is hereditarily separating, we arrive at some partial answer to the above question. The same situation arises in the context of actions of branch groups on the boundary space. A short formulation of the main theorem is as follows (the exact formulation is given in Theorem 2.13 in Section 2 together with the corresponding definitions).

Theorem

Let đș be a group acting on a perfect Hausdorff space 𝒳 by homeomorphisms. Let w ⁹ ( x ̄ ) be a non-trivial reduced word from F n ∗ G such that w ⁹ ( x ̄ ) ∉ G . If đș hereditarily separates 𝒳 and w ⁹ ( x ̄ ) has a conjugate in F n ∗ G which is explicitly oscillating, then the inequality w ⁹ ( x ̄ ) ≠ 1 has a solution in đș.

Using this theorem, we will also see that explicit oscillation gives a reasonable substitute for MIF; see Section 2.3.

It is worth noting that the idea of oscillation that we introduce is straightforward: it is a condition on the family of supports of the constants of the word. In particular, our arguments are rather direct but technical. The examples given in the paper should justify this inconvenience: they show that the notion is really ubiquitous and unavoidable.

The major results of this paper appeared in some form in the PhD thesis of the second author (under supervision of the first) in [21, Section 2]. In [21, Sections 3 and 4], they are applied to investigations of đč-limit groups. These applications will form a separate article.

We view the present article as an attempt to generalize the property MIF in the context of đș-spaces. Theorem 2.13 gives a sufficient condition for the existence of a solution of a given inequality in a group having a hereditarily separating action. Techniques applied in the proof will be enriched in other parts of the paper in order to cover new cases under new circumstances. In particular, we develop the approach in Theorem 3.8, which deals with a finite set of inequalities and uses weaker assumptions. The case where đș is equipped with a topology is considered in Section 4.

The terminology and notation used in the paper are standard. In the final part of the introduction, we remind the reader of some facts about Thompson’s group đč. In the main body of the paper, it is assumed that the reader remembers them (in particular the statement below). We mainly follow [5].

Thompson’s groupđč is the group given by the following infinite group presentation:

⟹ x 0 , x 1 , x 2 , 
 ∣ ⁹ x j ⁹ x i = x i ⁹ x j + 1 , i ⁹ < j ⟩ .

In fact, đč is finitely presented,

F = ⟹ x 0 , x 1 ∣ [ x 0 ⁹ x 1 − 1 , x 0 − i ⁹ x 1 ⁹ x 0 i ] , i = 1 , 2 ⟩ .

We will use the following geometric interpretation of đč. Consider the set of all strictly increasing continuous piecewise-linear functions from the closed unit interval onto itself. The group đč is realized by the set of all such functions that are differentiable except at finitely many dyadic rational numbers and such that all slopes (derivatives) are integer powers of 2. The corresponding group operation is just composition. For further reference, it will be useful to give an explicit form for the generators x n , for n ≄ 0 , in terms of piecewise-linear functions,

x n ⁱ ( t ) = { t , t ∈ [ 0 , 2 n − 1 2 n ] , 1 2 ⁱ t + 2 n − 1 2 n + 1 , t ∈ [ 2 n − 1 2 n , 2 n + 1 − 1 2 n + 1 ] , t − 1 2 n + 2 , t ∈ [ 2 n + 1 − 1 2 n + 1 , 2 n + 2 − 1 2 n + 2 ] , 2 ⁱ t − 1 , t ∈ [ 2 n + 2 − 1 2 n + 2 , 1 ] .

For any dyadic subinterval [ a , b ] ⊂ [ 0 , 1 ] , let us consider the set of elements in đč which are trivial on its complement, and denote it by F [ a , b ] . We know that it forms a subgroup of đč, which is isomorphic to the whole group. Let us denote its standard infinite set of generators by x [ a , b ] , 0 , x [ a , b ] , 1 , x [ a , b ] , 2 , 
 , where, for n ≄ 0 , we have

x [ a , b ] , n ⁱ ( t ) = { t , t ∈ [ 0 , a + ( 2 n − 1 ) ⁱ ( b − a ) 2 n ] , 1 2 ⁱ t + 1 2 ⁱ ( a + 2 n − 1 2 n ) , t ∈ [ a + ( 2 n − 1 ) ⁱ ( b − a ) 2 n , a + ( 2 n + 1 − 1 ) ⁱ ( b − a ) 2 n + 1 ] , t − b − a 2 n + 2 , t ∈ [ a + ( 2 n + 1 − 1 ) ⁱ ( b − a ) 2 n + 1 , a + ( 2 n + 2 − 1 ) ⁱ ( b − a ) 2 n + 2 ] , 2 ⁱ t − b , t ∈ [ a + ( 2 n + 2 − 1 ) ⁱ ( b − a ) 2 n + 2 , b ] , t , t ∈ [ b , 1 ] .

Moreover, if Îč [ a , b ] denotes the natural isomorphism between đč and F [ a , b ] sending x n to x [ a , b ] , n for all n ≄ 0 , then for any f ∈ F by f [ a , b ] , we have the element Îč [ a , b ] ⁹ ( f ) ∈ F [ a , b ] < F .

We will repeatedly use the following fact (see [5, Lemma 4.2] and [12, Lemma 2.4]).

Fact

If

0 = x 0 < x 1 < x 2 < ⋯ < x n = 1 , 0 = y 0 < y 1 < y 2 < ⋯ < y n = 1

are partitions of [ 0 , 1 ] consisting of dyadic rational numbers, then there exists some f ∈ F such that f ⁱ ( x i ) = y i for i = 0 , 
 , n .

Furthermore, if x i − 1 = y i − 1 and x i = y i for some 𝑖 with 1 ≀ i ≀ n , then 𝑓 can be taken to be trivial on the interval [ x i − 1 , x i ] .

Many examples that occur in this paper can be easily exposed using the rectangle diagrams introduced by W. Thurston. The paper [21] contains the corresponding pictures.

2 Explicitly oscillating words and hereditarily separating actions

In this section, we study inequalities over groups of permutations which are similar to the separating actions introduced by Abert in [1].

Let đș be a group. Any inequality over đș can be considered as follows. Let w ⁹ ( y ̄ ) be a word over đș on 𝑡 variables y 1 , 
 , y t . We usually view it as a non-trivial element of F t ∗ G . In order to study the inequality w ⁹ ( y ̄ ) ≠ 1 , we will assume that w ⁹ ( y ̄ ) is reduced in F t ∗ G . If w ⁹ ( y ̄ ) ∉ F t , we usually assume that it is of the form

(2.1) w ⁹ ( y ̄ ) = u n ⁹ v n ⁹ u n − 1 ⁹ v n − 1 ⁹ 
 ⁹ u 1 ⁹ v 1 ,

where n ∈ N , u i ∈ F ⁹ ( y ̄ ) and v i ∈ G ∖ { 1 } for each i ≀ n . It is clear that any word with constants is conjugate to a word in this form.

Our basic concern is existence of solutions of the inequality w ⁹ ( y ̄ ) ≠ 1 in đș. The easiest case appears in the following definition.

Definition 2.1

If, in the form (2.1), v n ⋅ 
 ⋅ v 1 ≠ 1 , then we say that the word w ⁹ ( y ̄ ) has non-trivial product of constants (in supp ⁹ ( v n ⋅ 
 ⋅ v 1 ) ⊆ X ).

In this case, it is clear that the tuple of units 1 ̄ solves the inequality w ⁹ ( y ̄ ) ≠ 1 .

Remark 2.2

The subgroup of G ∗ F t consisting of words having trivial product of constants can be viewed as follows. The 𝑡-tuple of units 1 ̄ ∈ G t is the variety of the system of equations S triv : y i = 1 , 1 ≀ i ≀ t . Then, according to [14, Definition 2], the set of all words w ⁹ ( y ̄ ) ∈ F t ∗ G with w ⁹ ( 1 ̄ ) = 1 is the radical of the system S triv .

In order to study more general/interesting cases, we will concentrate on permutation groups. In this class of groups, we will introduce oscillating words. It turns out that they generalize words with non-trivial product of constants.

2.1 Explicitly oscillating words

Let đș be a permutation group on 𝑋. We distinguish some specific types of words over đș with respect to the action on 𝑋. Let w ⁹ ( y ̄ ) be in the form (2.1). Define

O w : = ⋂ i = 0 n − 1 v 0 − 1 v 1 − 1 
 v i − 1 ( supp ( v i + 1 ) ) ,

where v 0 = id . If w ⁹ ( y ̄ ) ∈ F t , then let O w : = X ∖ Fix ( G ) .

Definition 2.3

Let V ⊆ X . We say that the word w ⁹ ( y ̄ ) ∈ F t ∗ G is explicitly oscillating in 𝑉 if w ⁹ ( y ̄ ) is non-trivial and V ∩ O w ≠ ∅ . When V = X , we just say that w ⁹ ( y ̄ ) is explicitly oscillating.

Observe that any non-trivial initial/final segment of an explicitly oscillating word is also explicitly oscillating. Note also that if all v i are taken from the same cyclic subgroup of đș of prime (or infinite) order, then w ⁹ ( y ̄ ) is explicitly oscillating. It is also clear that, by conjugating an explicitly oscillating word w ⁹ ( y ̄ ) with n > 1 by an element of đș with support disjoint from ⋃ i = 1 n supp ⁹ ( v i ) , we obtain a word which is not explicitly oscillating.

In the situation where đș acts on a Hausdorff topological space 𝒳 by homeomorphisms, the set O w is open. In the main results of the paper, we usually assume that 𝒳 is perfect. This rules out the explicitly oscillating words of the following example.

Example 2.4

Consider the group of finitary permutations G = S fin ⁹ ( N ) . Suppose that w ⁹ ( y ̄ ) ∉ F t and w ⁹ ( y ̄ ) is in the form (2.1) with non-trivial v i . Then O w is finite.

Remark 2.5

Nevertheless, we will show in this paper how our results can be adapted to the discrete case; see for example Theorem 2.15.

The case of Thompson’s group đč is fundamental for us. It will serve as the main source of illustrations. They are usually based on some computations which are left to the reader. Details of these computations and the corresponding pictures are given in the PhD thesis of the second author; see [21].

Example 2.6

Consider Thompson’s group đč with its standard action on [ 0 , 1 ] . Let w 1 ⁹ ( y ) = y ⁹ x 1 ⁹ y − 1 ⁹ x 2 ⁹ y 2 ⁹ x 1 − 1 , where 𝑩 is a variable. In the notation from the definitions, we have v 1 = x 1 − 1 , v 2 = x 2 , v 3 = x 1 and hence

O w 1 = x 1 ⁹ x 2 − 1 ⁹ ( ( 1 2 , 1 ) ) ∩ x 1 ⁹ ( ( 3 4 , 1 ) ) ∩ ( 1 2 , 1 ) = ( 5 8 , 1 ) .

Thus we see that w 1 ⁹ ( y ) is explicitly oscillating (in ( 5 8 , 1 ) ).

Notation

We introduce the following notation. For an element v ∈ G , let v 1 = v and v 0 = 1 . For every w ⁹ ( y ̄ ) given in the form w ⁹ ( y ̄ ) = u n ⁹ v n ⁹ u n − 1 ⁹ v n − 1 ⁹ 
 ⁹ u 1 ⁹ v 1 as in (2.1) and for every set A ⊆ X , let

V w ⁹ ( A ) = { v j Δ j ⋅ 
 ⋅ v 1 Δ 1 ⁹ ( A ) ∣ ( Δ 1 , 
 , Δ j ) ∈ { 0 , 1 } j ,  1 ≀ j ≀ n } , V w − 1 ⁹ ( A ) = { v 1 Δ 1 ⋅ 
 ⋅ v j Δ j ⁹ ( A ) ∣ ( Δ 1 , 
 , Δ j ) ∈ { 0 , − 1 } j ,  1 ≀ j ≀ n } ,

and let

V w > 0 ⁹ ( A ) = { v j ⋅ 
 ⋅ v 1 ⁹ ( A ) ∣ 1 ≀ j ≀ n } .

When w ⁹ ( y ̄ ) ∈ F t , let V w > 0 ⁹ ( A ) = { A } .

2.2 Hereditarily separating actions by homeomorphisms

Definition 2.7

Let đș be a permutation group acting on an infinite set 𝑋.

  • We say that đș separates 𝑋 if, for every finite subset Y ⊂ X , the pointwise stabilizer stab G ⁹ ( Y ) does not stabilize any point outside 𝑌.

  • Assume that 𝒳 is a Hausdorff topological space, đș consists of homeomorphisms of 𝒳 and the set of fixed points Fix ⁹ ( G ) is finite. We say that đș hereditarily separates 𝒳 if, for every open and infinite subset Z ⊆ X and for every finite subset Y ⊂ Z , the subgroup stab G ⁹ ( ( X ∖ Z ) âˆȘ Y ) does not stabilize any point from Z ∖ ( Y âˆȘ Fix ⁹ ( G ) ) .

Remark 2.8

Separating actions were introduced in [1]. We mention the following observation from that paper. For every separating action of a group đș on 𝑋 and every finite Y ⊂ X , the orbits of the action of the pointwise stabilizer of 𝑌 in X ∖ Y are infinite. As a corollary, we see that, for a hereditarily separating action of đș on 𝒳 and an open and infinite subset Z ⊆ X , the action of the stabilizer stab G ⁹ ( X ∖ Z ) on Z ∖ Fix ⁹ ( G ) has only infinite orbits.

Example 2.9

Consider ℕ as a discrete space. The symmetric group Sym ⁹ ( N ) , the group of finitary permutations S fin ⁹ ( N ) and the alternating group Alt ⁹ ( N ) ≀ S fin ⁹ ( N ) then hereditarily separate ℕ.

Example 2.10

Observe that Thompson’s group đč is hereditarily separating with respect to its standard action on [ 0 , 1 ] . Indeed, suppose that 𝑍 is an open subset of [ 0 , 1 ] , Y : = { t 1 , t 2 , 
 , t s } ⊂ Z and t ∈ Z ∖ ( Y âˆȘ { 0 , 1 } ) . There is some non-trivial dyadic segment [ p , q ] ⊆ Z containing 𝑡 such that Y ∩ [ p , q ] = ∅ . Obviously, x [ p , q ] , 0 ∈ stab F ⁹ ( ( [ 0 , 1 ] ∖ Z ) âˆȘ Y ) ∖ stab F ⁹ ( { t } ) .

Example 2.11

The action of any weakly branch group on the boundary space of the corresponding infinite rooted tree is also hereditarily separating. To define this class, we will follow [4, 9]. We consider a group đș that isometrically acts on some rooted tree 𝑇. The vertices in 𝑇 that are at the same distance from the root are said to be at the same level. We say that the action of the group đș on 𝑇 is spherically transitive if đș acts transitively on each level of 𝑇. For any vertex t ∈ T , we define its rigid stabilizer to be the set of all elements from đș that stabilize T ∖ T t pointwise, where T t is the natural subtree hanging from 𝑡. A group đș is called weakly branch if it acts spherically transitively on some rooted tree 𝑇 so that the rigid stabilizer of every vertex is non-trivial. The boundary of a tree 𝑇, denoted by ∂ T , consists of the infinite branches starting at the root. There is a topology on ∂ T where the base of open sets is determined by subtrees T t . A weakly branch group obviously acts on the boundary ∂ T by homeomorphisms (in fact, by isometries with respect to the natural metric).

To see the statement of this example, we use an argument from [1]. Fix a weakly branch group đș and the corresponding tree 𝑇. Let X : = ∂ T . To see that đș hereditarily separates 𝒳, suppose that 𝑍 is an open subset of 𝒳,

Y : = { x 1 , x 2 , 
 , x ℓ } ⊂ Z

and x ∈ Z ∖ Y . Without loss of generality, assume that 𝑡 is the vertex of the infinite ray đ‘„ such that if any y ∈ X contains 𝑡, then y ∈ Z . Let 𝑘 be the level of 𝑡. Now choose a level k â€Č ≄ k such that the vertices in the infinite rays x 1 , 
 , x ℓ , x at the k â€Č -th level are all distinct. Let t 0 be the vertex of đ‘„ at level k â€Č . Let 𝑆 be the stabilizer of t 0 in đș and let 𝑅 be the rigid vertex stabilizer of t 0 in đș. Then 𝑆 acts spherically transitively on the infinite subtree T t 0 rooted at t 0 . Since 𝑅 is a non-trivial normal subgroup of 𝑆, we see that it cannot stabilize any infinite ray going through t 0 . In particular, there exists some g ∈ R such that g ⁹ ( x ) ≠ x . On the other hand, 𝑔 stabilizes every ray not going through t 0 . It follows that g ∈ stab G ⁹ ( ( X ∖ Z ) âˆȘ Y ) . This proves the statement of Example 2.11.

Before the formulation of Theorem 2.13, we introduce a notion which will be central in the proof. Let đș be a permutation group on 𝑋 and let w ⁹ ( y ̄ ) be a word over đș on 𝑡 variables y 1 , 
 , y t . Assume that w ⁹ ( y ̄ ) is reduced in F t ∗ G , w ⁹ ( y ̄ ) ∉ F t and w ⁹ ( y ̄ ) is written in the following form:

(2.2) w ⁹ ( y ̄ ) = u n , ℓ n ⁹ 
 ⁹ u n , 1 ⁹ v n ⁹ 
 ⁹ u 2 , ℓ 2 ⁹ 
 ⁹ u 2 , 1 ⁹ v 2 ⁹ u 1 , ℓ 1 ⁹ 
 ⁹ u 1 , 1 ⁹ v 1 ,

where u j , i j ∈ { y 1 ± 1 , 
 , y t ± 1 } , 1 ≀ i j ≀ ℓ j , and v j ∈ G ∖ { 1 } , 1 ≀ j ≀ n . Let

L j : = ∑ i = 1 j ℓ i , 1 ≀ j ≀ n .

From now on, we will say that L n is the length of w ⁹ ( y ̄ ) , which will be denoted by | w ⁹ ( y ̄ ) | . This terminology slightly disagrees with the usual one (where parameters are taken into account), but it will be helpful in our induction arguments.

For any 1 ≀ r ≀ L n , let

( w ) r ⁹ ( y ̄ ) = u d , s ⁹ 
 ⁹ u d , 1 ⁹ v d ⁹ 
 ⁹ u 2 , ℓ 2 ⁹ 
 ⁹ u 2 , 1 ⁹ v 2 ⁹ u 1 , ℓ 1 ⁹ 
 ⁹ u 1 , 1 ⁹ v 1

be the final segment of w ⁹ ( y ̄ ) of 𝑟 occurrences of letters from { y 1 ± 1 , 
 , y t ± 1 } , i.e. r = L d − 1 + s and 1 ≀ s ≀ ℓ d . In this case, we also define

[ w ] r ⁹ ( y ̄ ) = w ⁹ ( y ̄ ) ⁹ ( ( w ) r ⁹ ( y ̄ ) ) − 1

to be the corresponding initial segment.

Let g ̄ = ( g 1 , 
 , g t ) be a tuple from đș. By ( w ) r ⁹ ( g ̄ ) , we denote the value of ( w ) r ⁹ ( y ̄ ) in đș under the substitution y i = g i , 1 ≀ i ≀ t . To simplify notation, also let ( w ) 0 ⁹ ( y ̄ ) = 1 ∈ G .

Let p ∈ X . Define p r , g ̄ = ( w ) r ⁹ ( g ̄ ) ⁹ ( p ) for all 𝑟, 1 ≀ r ≀ L n .

Definition 2.12

We say that g ̄ is distinctive for w ⁹ ( y ̄ ) and 𝑝 if all the points

p = p 0 , g ̄ , v 1 ⁹ ( p 0 , g ̄ ) , 
 , p l 1 , g ̄ , v 2 ⁹ ( p l 1 , g ̄ ) , 
 , p L n , g ̄

are pairwise distinct.

This generalizes the corresponding definition of [1, p. 528] for words w ⁹ ( y ̄ ) ∈ F t . In the latter case, we just omit the members v j ⁹ ( p l j − 1 , g ̄ ) from the sequence. When Definition 2.12 holds, we obviously have w ⁹ ( g ̄ ) ⁹ ( p ) ≠ p , i.e. w ⁹ ( g ̄ ) ≠ 1 . The following statement is related to [1, Theorem 1.1].

Theorem 2.13

Let đș be a group acting on a perfect Hausdorff space 𝒳 by homeomorphisms. Let w ⁹ ( y ̄ ) be a non-trivial word over đș on 𝑡 variables, y 1 , 
 , y t , which is reduced in F t ∗ G and non-constant (i.e. w ⁹ ( y ̄ ) ∉ G ). If đș hereditarily separates 𝒳 and w ⁹ ( y ̄ ) has a conjugate in F t ∗ G which is explicitly oscillating, then the inequality w ⁹ ( y ̄ ) ≠ 1 has a solution in đș.

Moreover, assuming additionally that w ⁹ ( y ̄ ) ∈ F t or w ⁹ ( y ̄ ) is given in the form (2.1) and is explicitly oscillating, for every open set O â€Č ⊆ O w , there is a distinctive tuple g ̄ = ( g 1 , 
 , g t ) ∈ G t with respect to w ⁹ ( y ̄ ) and some p ∈ O â€Č such that, for all 𝑖 with 1 ≀ i ≀ t ,

  • supp ⁹ ( g i ) ⊆ ( ⋃ V w > 0 ⁹ ( O â€Č ) ) ∖ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) ,

  • each member of V w ⁹ ( O â€Č ) is g i -invariant.

Proof

Using conjugation (if necessary), we assume that if w ⁹ ( y ̄ ) ∉ F t , then w ⁹ ( y ̄ ) is an explicitly oscillating word written in the form (2.2). We preserve the notation given before the formulation. Note that, since 𝒳 is perfect, each open subset is infinite.

It is clear that we only need to prove the second part of the statement of the theorem. Fix O â€Č as in the formulation and any

p ∈ O â€Č ∖ ( ⋃ V w − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) ) .

The existence of such a 𝑝 follows from the fact that the set O â€Č ̄ ∖ O â€Č is nowhere dense in 𝒳 and the action of đș is continuous. The proof of the theorem is by induction. At the 𝑘-th step (where k ≀ L n ), we will show that

  • there is a tuple g ̄ = ( g 1 , 
 , g t ) ∈ G such that g ̄ is distinctive for ( w ) k and 𝑝.

  • In the condition above, we can choose g ̄ so that, for all 𝑖 with 1 ≀ i ≀ t ,

    supp ⁹ ( g i ) ⊆ ( ⋃ V w > 0 ⁹ ( O â€Č ) ) ∖ ( ⋃ V [ w ] k − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) )

    and each member of V w ⁹ ( O â€Č ) is g i -invariant.

We will use the following claim.

Claim M

For every r ≀ n , every ( Δ 1 ⁹ 
 ⁹ Δ r ) ∈ { 0 , 1 } r and every

q ∈ v r Δ r ⋅ 
 ⋅ v 1 Δ 1 ⁹ ( O â€Č ) ∖ ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ,

there is a neighborhood O ⊆ v r Δ r ⋅ 
 ⋅ v 1 Δ 1 ⁹ ( O â€Č ) of 𝑞 such that,

( † ) for all ⁹ V ∈ V w ⁹ ( O â€Č ) , O ∩ V ≠ ∅ âŸč O ⊆ V .

Indeed, let 𝑞 and O â€Č be as in the formulation of the claim. Let us enumerate the elements of V w ⁹ ( O â€Č ) by V m with m > 0 and inductively construct open sets O m corresponding to them. We start with O 0 = v r Δ r ⋅ 
 ⋅ v 1 Δ 1 ⁹ ( O â€Č ) . If, at the 𝑚-th step, q ∈ O m − 1 ∩ V m , then we define O m = O m − 1 ∩ V m . Now suppose that q ∉ O m − 1 ∩ V m . As q ∉ ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) , we find some open O m ⊆ O m − 1 ∖ V m containing 𝑞. After meeting all members of V w ⁹ ( O â€Č ) , we obtain the final 𝑂 which is non-empty and open. If w ⁹ ( y ̄ ) ∈ F t , then we define O = O â€Č , which finishes the proof of the claim.

If k = 1 , we may assume that ( w ) 1 is of the form y j ± 1 ⁹ v 1 for some 1 ≀ j ≀ t . When w ⁹ ( y ̄ ) ∈ F t , we replace v 1 by id and follow the argument below. Note that

v 1 ⁹ ( p ) ∉ ⋃ V [ w ] 1 − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) .

Applying the claim above, we find in v 1 ⁹ ( O â€Č ) a neighborhood 𝑂 of v 1 ⁹ ( p ) which satisfies ( † ). According to the assumptions above, when w ⁹ ( y ̄ ) ∉ F t , then we have p â€Č ≠ v 1 ⁹ ( p â€Č ) for all p â€Č ∈ O â€Č . Thus, for a non-trivial v 1 , the inequality p ≠ v 1 ⁹ ( p ) is satisfied.

Without loss of generality, suppose ( w ) 1 = y j ⁹ v 1 . Since đș hereditarily separates 𝒳, the orbit of v 1 ⁹ ( p ) with respect to

stab G ⁹ ( ( X ∖ O ) âˆȘ ⋃ V [ w ] 1 − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) )

is infinite. Thus we can choose

f ∈ stab G ⁹ ( ( X ∖ O ) âˆȘ ⋃ V [ w ] 1 − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) )

such that f ⁱ ( v 1 ⁱ ( p ) ) ∉ { p , v 1 ⁱ ( p ) } . Defining g j : = f and extending it by any ( t − 1 ) -tuple from

stab G ⁹ ( ( X ∖ O ) âˆȘ ⋃ V [ w ] 1 − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) ) ,

we obtain a 𝑡-tuple g ̄ distinctive for 𝑝 and ( w ) 1 .

It is easy to see that, for all i ≀ t ,

supp ⁹ ( g i ) ⊆ ( ⋃ V w > 0 ⁹ ( O â€Č ) ) ∖ ( ⋃ V [ w ] 1 − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) ) .

Now fix i ≀ t and V ∈ V w ⁹ ( O â€Č ) . If O ∩ V = ∅ , then obviously g i ⁹ ( V ) = V . On the other hand, if O ∩ V ≠ ∅ , then it follows from the construction of 𝑂 that O ⊆ V . Thus supp ⁹ ( g i ) is also a subset of 𝑉 and the condition g i ⁹ ( V ) = V is satisfied.

Let k > 1 . Assume that, for

( w ) k − 1 = u d , s ⁱ 
 ⁱ u d , 1 ⁱ v d ⁱ 
 ⁱ u 2 , ℓ 2 ⁱ 
 ⁱ u 2 , 1 ⁱ v 2 ⁱ u 1 , ℓ 1 ⁱ 
 ⁱ u 1 , 1 ⁱ v 1 ,

where k − 1 = L d − 1 + s , we can find a tuple g ̄ ∈ G such that the induction hypothesis is satisfied. According to the form of ( w ) k , we consider two cases.

Case 1:

( w ) k = u d , s + 1 ⁱ u d , s ⁱ 
 ⁱ u d , 1 ⁱ v d ⁱ 
 ⁱ u 2 , 1 ⁱ v 2 ⁱ u 1 , ℓ 1 ⁱ 
 ⁱ u 1 , 1 ⁱ v 1 ,

where k − 1 = L d − 1 + s , s ≄ 1 . If

p k , g ̄ ∉ { p i , g ̄ ∣ 0 ≀ i ≀ k − 1 } âˆȘ { v 1 ⁹ ( p 0 , g ̄ ) , 
 , v d ⁹ ( p L d − 1 , g ̄ ) } ,

then we have found an acceptable tuple g ̄ . Let us assume that p k , g ̄ = p m , g ̄ for some 0 ≀ m < k or p k , g ̄ = v m + 1 ⁹ ( p L m , g ̄ ) for some 0 ≀ m < d − 1 .

Let y j ± 1 be the first letter of ( w ) k . Replacing y j by y j − 1 and g j by g j − 1 if necessary, we may assume that u d , s + 1 = y j . Put

Y = { p i , g ̄ ∣ 0 ≀ i ≀ k − 2 } âˆȘ { v 1 ⁹ ( p 0 , g ̄ ) , 
 , v d ⁹ ( p L d − 1 , g ̄ ) }

for g ̄ chosen at the ( k − 1 ) -th step of induction. Since, by the induction hypothesis, p ∈ O â€Č and g s ± 1 ⁹ ( v r ⁹ 
 ⁹ v 1 ⁹ ( O â€Č ) ) = v r ⁹ 
 ⁹ v 1 ⁹ ( O â€Č ) for all r ≀ k and all s ≀ t , we see that p k − 1 , g ̄ ∈ v d ⁹ 
 ⁹ v 1 ⁹ ( O â€Č ) . We also know that

p k − 1 , g ̄ ∉ ⋃ V [ w ] k − 1 − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) .

As above, we choose a neighborhood O ⊆ v d ⁹ 
 ⁹ v 1 ⁹ ( O â€Č ) of the point

p k − 1 , g ̄ ∈ v d ⁹ 
 ⁹ v 1 ⁹ ( O â€Č )

satisfying condition ( † ). Since the action of đș is hereditarily separating, the orbit with respect to

stab G ⁹ ( ( X ∖ O ) âˆȘ ⋃ V [ w ] k − 1 − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) )

of p k − 1 , g ̄ is infinite. Let

Z = { g j − 1 ⁹ ( p i , g ̄ ) ∣ 0 ≀ i ≀ k − 1 } âˆȘ { g j − 1 ⁹ ( v i + 1 ⁹ ( p L i , g ̄ ) ) ∣ 0 ≀ i ≀ d − 1 } .

Since 𝑍 is finite, there exists some

f ∈ stab G ⁹ ( ( X ∖ O ) âˆȘ ⋃ V [ w ] k − 1 − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) âˆȘ Y )

taking p k − 1 , g ̄ outside 𝑍. Replacing g j by g j ⁹ f , we obtain a corrected tuple g ̄ . Since the element 𝑓 has been chosen from the stabilizer stab G ⁹ ( Y ) , the points

p 0 , g ̄ , v 1 ⁹ ( p 0 , g ̄ ) , p 1 , g ̄ , 
 , v d ⁹ ( p L d − 1 , g ̄ ) , p L d − 1 , g ̄ , 
 , p k − 1 , g ̄

are the same as before. On the other hand, p k , g ̄ is distinct from all the elements

p 0 , g ̄ , v 1 ⁹ ( p 0 , g ̄ ) , p 1 , g ̄ , 
 , v d ⁹ ( p L d − 1 , g ̄ ) , p L d − 1 , g ̄ , 
 , p k − 1 , g ̄ .

Since

f ∈ stab G ⁹ ( ( X ∖ O ) âˆȘ ⋃ V [ w ] k − 1 − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) âˆȘ Y ) ,

we see that

supp ⁹ ( f ) ⊆ ⋃ V w > 0 ⁹ ( O â€Č ) ∖ ( ⋃ V [ w ] k − 1 − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) ) .

This together with the induction hypothesis implies that

supp ⁹ ( g j ) ⊆ ⋃ V w > 0 ⁹ ( O â€Č ) ∖ ( ⋃ V [ w ] k − 1 − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) )

for the corrected g j .

Let V ∈ V w ⁹ ( O â€Č ) . If O ∩ V = ∅ , then by the choice of 𝑓 and induction, we have g j ⁹ ( V ) = V . If O ∩ V ≠ ∅ , then O ⊆ V . Thus, by the choice of 𝑓 and induction, 𝑓 and the original g j (defined at Step k − 1 ) stabilize 𝑉 setwise. Thus g j ⁹ ( V ) = V . On the other hand, all elements g s for s ≠ j have not been changed and thus automatically satisfy the required conditions. This finishes the proof of Case 1.

Case 2:

( w ) k = u d + 1 , 1 ⁱ v d + 1 ⁱ u d , s ⁱ 
 ⁱ u d , 1 ⁱ v d ⁱ 
 ⁱ u 2 , 1 ⁱ v 2 ⁱ u 1 , ℓ 1 ⁱ 
 ⁱ u 1 , 1 ⁱ v 1 ,

where k = L d + 1 . If

p k , g ̄ ∉ { p i , g ̄ ∣ 0 ≀ i ≀ k − 1 } âˆȘ { v 1 ⁹ ( p 0 , g ̄ ) , 
 , v d + 1 ⁹ ( p L d , g ̄ ) } , v d + 1 ⁹ ( p k − 1 , g ̄ ) ∉ { p i , g ̄ ∣ 0 ≀ i ≀ k − 1 } âˆȘ { v 1 ⁹ ( p 0 , g ̄ ) , 
 , v d ⁹ ( p L d − 1 , g ̄ ) } ,

then we have found an acceptable tuple g ̄ .

Assume the contrary. Suppose that y j ± 1 is the first letter of ( w ) k , and as in Case 1, we only consider the possibility u d + 1 , 1 = y j . Let u d , s = y j â€Č ± 1 . Then let

Y â€Č = { p i , g ̄ ∣ 0 ≀ i ≀ k − 1 } âˆȘ { v 1 ⁹ ( p 0 , g ̄ ) , 
 , v d ⁹ ( p L d − 1 , g ̄ ) } .

Assume that v d + 1 ⁹ ( p k − 1 , g ̄ ) ∈ Y â€Č . By the induction hypothesis,

p k − 1 , g ̄ ∈ v d ⁹ 
 ⁹ v 1 ⁹ ( O â€Č ) .

By Claim M, we find some neighborhood O ⊆ v d ⁹ 
 ⁹ v 1 ⁹ ( O â€Č ) of p k − 1 , g ̄ satisfying ( † ). Since the action of đș is hereditarily separating, the orbit of p k − 1 , g ̄ with respect to

stab G ⁹ ( ( X ∖ O ) âˆȘ ⋃ V [ w ] k − 1 − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) âˆȘ ( Y â€Č ∖ { p k − 1 , g ̄ } ) )

is infinite. We replace g j â€Č by some f â€Č ⁹ g j â€Č (or g j â€Č ⁹ f â€Č in the case u d , s = y j â€Č − 1 ), where

f â€Č ∈ stab G ⁹ ( ( X ∖ O ) âˆȘ ⋃ V [ w ] k − 1 − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) âˆȘ ( Y â€Č ∖ { p k − 1 , g ̄ } ) )

and f â€Č takes p k − 1 , g ̄ outside the finite set Y â€Č âˆȘ v d + 1 − 1 ⁹ ( Y â€Č ) . By the choice of O â€Č and 𝑂,

O ⊆ v d ⁹ 
 ⁹ v 1 ⁹ ( O â€Č ) ⊆ supp ⁹ ( v d + 1 ) ,

i.e. the corrected p k − 1 , g ̄ is not fixed by v d + 1 . Thus the corrected v d + 1 ⁹ ( p k − 1 , g ̄ ) surely omits the corrected Y â€Č .

Furthermore, by the choice of f â€Č , we still have

supp ⁹ ( g j â€Č ) ⊆ ⋃ V w > 0 ⁹ ( O â€Č ) ∖ ( ⋃ V [ w ] k − 1 − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) ) .

Similarly to Case 1, we see that the condition supp ⁹ ( f â€Č ) ⊆ O implies g j â€Č ± 1 ⁹ ( V ) = V for any V ∈ V w ⁹ ( O â€Č ) .

So it remains to consider the case when v d + 1 ⁹ ( p k − 1 , g ̄ ) ∉ Y â€Č , but

  • either p k , g ̄ = p j , g ̄ for some 0 ≀ j < k

  • or p k , g ̄ = v j + 1 ⁹ ( p L j , g ̄ ) for some 0 ≀ j ≀ d .

Let

Y = { p i , g ̄ ∣ 0 ≀ i ≀ k − 1 } âˆȘ { v 1 ⁹ ( p 0 , g ̄ ) , 
 , v d ⁹ ( p L d − 1 , g ̄ ) }

(i.e. redefine Y â€Č ) and

Z = { g j − 1 ⁹ ( p i , g ̄ ) ∣ 0 ≀ i ≀ k − 1 } âˆȘ { g j − 1 ⁹ ( v i + 1 ⁹ ( p L i , g ̄ ) ) ∣ 0 ≀ i ≀ d } .

Now observe that

v d + 1 ⁹ ( p k − 1 , g ̄ ) ∈ v d + 1 ⁹ 
 ⁹ v 1 ⁹ ( O â€Č ) ∖ ( ⋃ V [ w ] k − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) âˆȘ Y ) .

Choose the neighborhood O ⊆ v d + 1 ⁹ 
 ⁹ v 1 ⁹ ( O â€Č ) of the point v d + 1 ⁹ ( p k − 1 , g ̄ ) satisfying ( † ). Then there exists some

f ∈ stab G ⁹ ( ( X ∖ O ) âˆȘ ⋃ V [ w ] k − 1 ⁹ ( ⋃ V w ⁹ ( O â€Č ̄ ∖ O â€Č ) ) âˆȘ Y )

taking v d + 1 ⁹ ( p k − 1 , g ̄ ) outside 𝑍. Replacing g j by g j ⁹ f , we finish the proof exactly as in Case 1. ∎

Remark 2.14

Under the circumstances of Theorem 2.13, additionally assume that ( X , d ) is a metric space. Then the conclusion of Theorem 2.13 may be extended by the following statement: for any 𝜀, the tuple g ̄ can be chosen so that it additionally satisfies the inequality d ⁹ ( x , g i ⁹ ( x ) ) ≀ Δ for all i ≀ t and x ∈ X .

To see this, it is enough to control how g ̄ is corrected at each step of the inductive procedure in the proof of Theorem 2.13. If | w ⁹ ( y ̄ ) | = L n (as above), then at the 𝑘-th step of the induction, we choose the corresponding open set 𝑂 with the additional property that its diameter is less than Δ L n . Assume that the homeomorphism g i which is corrected at this step already satisfies d ⁹ ( x , g i ⁹ ( x ) ) ≀ k − 1 L n ⁹ Δ for all x ∈ X . Since the result of the correction, say g i â€Č , differs from g i by some f â€Č with support from 𝑂, we have d ⁹ ( x , g i â€Č ⁹ ( x ) ) ≀ k L n ⁹ Δ for all x ∈ X .

We also have a topology-free version of Theorem 2.13, which generalizes [1, Theorem 1.1]. For any A ⊆ X , we denote A 1 = A and A 0 = X ∖ A . Fix any w ⁹ ( y ̄ ) such that either w ⁹ ( y ̄ ) ∈ F t or w ⁹ ( y ̄ ) = u n ⁹ v n ⁹ 
 ⁹ u 1 ⁹ v 1 is in the form (2.1). Then, for every Δ ̄ = ( Δ 1 , 
 , Δ n ) ∈ { 0 , 1 } n , we denote by O w Δ ̄ the set

⋂ s = 1 n ( v s ⁹ 
 ⁹ v 1 ⁹ ( O w ) ) Δ s .

Theorem 2.15

Let đș act by permutations on some set 𝑋 and let w ⁹ ( y ̄ ) ∈ F t ∗ G be a reduced, non-constant word over đș on variables y 1 , 
 , y t . Assume that w ⁹ ( y ̄ ) ∈ F t or w ⁹ ( y ̄ ) = u n ⁹ v n ⁹ 
 ⁹ u 1 ⁹ v 1 is in the form (2.1) with O w ≠ ∅ . Assume also that, for every Δ ̄ ∈ { 0 , 1 } n ,

( ♱ ) if ⁹ O w Δ ̄ ≠ ∅ , then ⁹ stab G ⁹ ( X ∖ O w Δ ̄ ) ⁹ separates ⁹ O w Δ ̄ .

Then the inequality w ⁹ ( y ̄ ) ≠ 1 has a solution in đș.

Proof

If w ⁹ ( y ̄ ) ∈ F t , then O w = X and we simply apply [1, Theorem 1.1].

If w ⁹ ( y ̄ ) ∉ F t , then we follow the proof of Theorem 2.13 (keeping the corresponding notation). Note that, by ( ♱ ), the set O w Δ ̄ is infinite for O w Δ ̄ ≠ ∅ . For O â€Č = O w , we reformulate Claim M from the proof of Theorem 2.13 in the following form.

Claim M#

For every r ≀ n and every q ∈ v r ⁹ 
 ⁹ v 1 ⁹ ( O w ) , there is a unique tuple Δ ̄ ∈ { 0 , 1 } n such that q ∈ O w Δ ̄ ⊆ v r ⁹ 
 ⁹ v 1 ⁹ ( O w ) . The corresponding O w Δ ̄ satisfies the following:

( ‡ ) for all ⁹ V ∈ V w > 0 ⁹ ( O â€Č ) , O w Δ ̄ ∩ V ≠ ∅ âŸč O w Δ ̄ ⊆ V .

We prove Claim M# as follows. First observe that { O w Δ ̄ ∣ Δ ̄ ∈ { 0 , 1 } n } is a partition of 𝑋, and hence, for any q ∈ v r ⁹ 
 ⁹ v 1 ⁹ ( O w ) , there is a unique tuple Δ ̄ ∈ { 0 , 1 } n such that q ∈ O w Δ ̄ ⊆ v r ⁹ 
 ⁹ v 1 ⁹ ( O w ) . Now fix this tuple Δ ̄ , take any s ≀ n and suppose that O w Δ ̄ ∩ v s ⁹ 
 ⁹ v 1 ⁹ ( O w ) ≠ ∅ . It follows that Δ s = 1 in Δ ̄ . Thus we have O w Δ ̄ ⊆ v s ⁹ 
 ⁹ v 1 ⁹ ( O w ) .

We now apply the proof of Theorem 2.13. Let O â€Č = O w . At the 𝑘-th step of the induction, we show that

  • there is some p ∈ O â€Č and a tuple g ̄ = ( g 1 , 
 , g t ) ∈ G such that g ̄ is distinctive for 𝑝 and ( w ) k .

  • In the condition above, we can choose g ̄ so that, for all 𝑖, 1 ≀ i ≀ t ,

    supp ⁹ ( g i ) ⊆ V w > 0 ⁹ ( O â€Č ) and g i ⁹ ( v r ⁹ 
 ⁹ v 1 ⁹ ( O â€Č ) ) = v r ⁹ 
 ⁹ v 1 ⁹ ( O â€Č ) ⁹ for any ⁹ r ≀ n .

At Step 1, we fix any p ∈ O w . At Step k ≀ L n , by Claim M# , we obtain Δ ̄ ∈ { 0 , 1 } n such that p k − 1 , g ̄ ∈ O w Δ ̄ ⊆ v d ⁹ 
 ⁹ v 1 ⁹ ( O w ) and ( ‡ ‣ M# ) holds. Hence we may now apply ( ♱ ) and Remark 2.8 to see that, for any finite set 𝑌 not containing p k − 1 , g ̄ , the orbit of p k − 1 , g ̄ with respect to stab G ⁹ ( ( X ∖ O w Δ ̄ ) âˆȘ Y ) is infinite.

Furthermore, applying the proof of Theorem 2.13, we replace each occurrence of the neighborhood O ⊆ v d ⁹ 
 ⁹ v 1 ⁹ ( O â€Č ) of p k − 1 , g ̄ constructed there by the set O w Δ ̄ as in Claim M# . Consequently, in order to find the desired tuple g ̄ , we replace the usage of ( † ) from the proof of Theorem 2.13 by ( ‡ ‣ M# ). After these modifications, the proof of Theorem 2.13 works for Theorem 2.15. ∎

This theorem cannot be directly applied in the case of subgroups of S fin ⁱ ( N ) . In this case, condition ( ♱ ) does not hold for O w . In Section 3, we will show that this theorem can be applied to so-called oscillating identities.

2.3 Partial MIF

In this section, we give a curious application of Theorem 2.13. Let đș be a group and suppose that E ⊆ G . We denote by F t ∗ E the set of all reduced words from ( F t ∗ G ) ∖ G with constants from 𝐾. We remind the reader that F t is the free group F ⁹ [ y 1 , 
 , y t ] . Below, we denote by F ω the free group F ⁹ [ z 1 , 
 , z i , 
 ] .

Definition 2.16

Let đș be a group acting on a perfect Hausdorff space 𝒳 by homeomorphisms. We say that a subset E ⊆ G is e.o.-stable if, for every natural t > 0 , each reduced word from F t ∗ E is explicitly oscillating.

Example 2.17

Consider Thompson’s group đč and the subset F cf of all g ∈ F such that supp ⁹ ( g ) is cofinite in [ 0 , 1 ] . Then F cf is e.o.-stable. Indeed, when w ⁹ ( y ̄ ) is a non-trivial reduced word over F cf , O w is cofinite in [ 0 , 1 ] . Note that F cf ⋅ F cf = F .

When a group đș is a subgroup of some G ̂ , it is said that đș is existentially closed in G ̂ if, for every sequence of words

w 1 ⁹ ( z ̄ ) , 
 , w ℓ ⁹ ( z ̄ ) , w ℓ + 1 ⁹ ( z ̄ ) , 
 , w ℓ + k ⁹ ( z ̄ ) ∈ F ω ∗ G ,

if the system of equations and inequations

w 1 ⁹ ( z ̄ ) = 1 , 
 , w ℓ ⁹ ( z ̄ ) = 1 , w ℓ + 1 ⁹ ( z ̄ ) ≠ 1 , 
 , w ℓ + k ⁹ ( z ̄ ) ≠ 1

has a solution in G ̂ , then it has a solution in đș. It is easy to see that [10, Proposition 5.3] can be reformulated as follows.

A countable group đș is MIF if and only if đș is existentially closed in F t ∗ G for every natural 𝑡.

Moreover, the sufficiency is obvious. Thus the following statement is in close relation to it.

Proposition 2.18

Let đș be a group acting on a perfect Hausdorff space 𝒳 by homeomorphisms and let 𝑡 be a natural number. Assume that a subset E ⊆ G is e.o.-stable. Then, for every sequence of words

w 1 ⁹ ( z ̄ ) , 
 , w ℓ ⁹ ( z ̄ ) , w ℓ + 1 ⁹ ( z ̄ ) , 
 , w ℓ + k ⁹ ( z ̄ ) ∈ F ω ∗ G ,

if the system of equations and inequations

w 1 ⁹ ( z ̄ ) = 1 , 
 , w ℓ ⁹ ( z ̄ ) = 1 , w ℓ + 1 ⁹ ( z ̄ ) ≠ 1 , 
 , w ℓ + k ⁹ ( z ̄ ) ≠ 1

has a solution in F t ∗ G which, for every i ≀ k , takes w ℓ + i ⁹ ( z ̄ ) to some

w ℓ + i â€Č ⁹ ( y ̄ ) ∈ ( F t ∗ E ) ∖ { 1 } ,

then the system has a solution in đș.

Proof

Let | z ̄ | = s and let v 1 ⁹ ( y ̄ ) , 
 , v s ⁹ ( y ̄ ) be a solution of the system

w 1 ⁹ ( z ̄ ) = 1 , 
 , w ℓ ⁹ ( z ̄ ) = 1 , w ℓ + 1 ⁹ ( z ̄ ) ≠ 1 , 
 , w ℓ + k ⁹ ( z ̄ ) ≠ 1

in F t ∗ G such that, after the substitution z ̄ ← v ̄ into w ℓ + 1 ⁹ ( z ̄ ) , 
 , w ℓ + k ⁹ ( z ̄ ) , we obtain w ℓ + 1 â€Č ⁹ ( y ̄ ) , 
 , w ℓ + k â€Č ⁹ ( y ̄ ) from the formulation of the proposition. Let

u ⁹ ( y 0 , y ̄ ) = [ [ 
 ⁹ [ w ℓ + 1 â€Č ⁹ ( y ̄ ) , y 0 − 1 ⁹ w ℓ + 2 â€Č ⁹ ( y ̄ ) ⁹ y 0 ] , 
 ] , y 0 − k + 1 ⁹ w ℓ + k â€Č ⁹ ( y ̄ ) ⁹ y 0 k − 1 ] .

We view u ⁹ ( y 0 , y ̄ ) as an element of F ⁹ [ y 0 , y 1 , 
 , y t ] ∗ G . It is easy to see that the normal form of it is non-trivial and belongs to F t + 1 ∗ E , i.e. it is explicitly oscillating. Applying Theorem 2.13, we find g 0 , g 1 , 
 , g t ∈ G with u ⁹ ( g 0 , g ̄ ) ≠ 1 . Thus it follows that w ℓ + 1 â€Č ⁹ ( g ̄ ) ≠ 1 , 
 , w ℓ + k â€Č ⁹ ( g ̄ ) ≠ 1 . In particular, the elements v 1 ⁹ ( g ̄ ) , 
 , v s ⁹ ( g ̄ ) form a solution of the system w ℓ + 1 ⁹ ( z ̄ ) ≠ 1 , 
 , w ℓ + k ⁹ ( z ̄ ) ≠ 1 in đș. Since the tuple v 1 ⁹ ( y ̄ ) , 
 , v s ⁹ ( y ̄ ) is a solution of the system w 1 ⁹ ( z ̄ ) = 1 , 
 , w ℓ ⁹ ( z ̄ ) = 1 in F t ∗ G , the elements v 1 ⁹ ( g ̄ ) , 
 , v s ⁹ ( g ̄ ) form a solution of the system w 1 ⁹ ( z ̄ ) = 1 , 
 , w ℓ ⁹ ( z ̄ ) = 1 in đș. ∎

Using Example 2.17, we see that the statement of this proposition holds when đș is Thompson’s đč and E = F cf .

3 Inequalities in groups with hereditarily separating action

3.1 Oscillating words

In order to apply Theorem 2.13 to a broader class of words and a larger number of inequalities, we will introduce the notion of an oscillating word. Intuitively, it describes words that are explicitly oscillating after the transition from the region O w to some other place.

Suppose đș acts on some Hausdorff topological space 𝒳 by homeomorphisms and let w ⁹ ( y ̄ ) be a word over đș on 𝑡 variables such that w ⁹ ( y ̄ ) is reduced in F t ∗ G . If w ⁹ ( y ̄ ) ∉ F t , then we write it as follows:

w X ⁹ ( y ̄ ) = u X , n X ⁹ v X , n X ⁹ u X , n X − 1 ⁹ v X , n X − 1 ⁹ 
 ⁹ u X , 1 ⁹ v X , 1 ,

where n X ∈ N , u X , i depends only on variables and v X , i ∈ G ∖ { 1 } for 1 ≀ i ≀ n X . Let v X , 0 : = 1 .

Suppose that w X ⁹ ( y ̄ ) is not explicitly oscillating. We now describe a procedure which produces a family P os of open subsets of 𝒳 and a map which associates to each V ∈ P os a word w V ⁹ ( y ̄ ) which is explicitly oscillating in 𝑉. The case P os = ∅ is possible but not desirable. We call this procedure Transition .

In the description of it, we use the following notation. For an open A ⊆ X , let A 0 : = int ( X ∖ ( A âˆȘ Fix ( G ) ) ) and A 1 : = A ∖ Fix ( G ) . Below, we always assume that Fix ⁹ ( G ) is finite.

Transition

For each sequence Δ ̄ = ( Δ 1 , 
 , Δ n X ) ∈ { 0 , 1 } n X , we define the set

X Δ ̄ : = ⋂ i = 1 n X v X , 1 − 1 
 v X , i − 1 − 1 ( supp ( v X , i ) Δ i )

and consider the following family:

P 1 : = { X Δ ̄ ∣ Δ ̄ ∈ { 0 , 1 } n X } ∖ { ∅ } .

Since w X ⁹ ( y ̄ ) is not explicitly oscillating, X ( 1 , 
 , 1 ) = ∅ , i.e. not in P 1 . For each X Δ ̄ ∈ P 1 , we define a word w X Δ ̄ â€Č ⁹ ( y ̄ ) in the following way:

w X Δ ̄ â€Č ( y ̄ ) : = u X , n X v X , n X Δ n X u X , n X − 1 v X , n X − 1 Δ n X − 1 
 u X , 1 v X , 1 Δ 1 .

Then we reduce w X Δ ̄ â€Č ⁹ ( y ̄ ) in F t ∗ G . If, after reduction, the obtained word w â€Č ⁹ ( y ̄ ) is of the form 
 ⁹ v â€Č ⁹ u â€Č , where u â€Č contains only variables and v â€Č ∈ G ∖ { 1 } , then we conjugate w â€Č ⁹ ( y ̄ ) by ( u â€Č ) − 1 and denote the obtained word by w X Δ ̄ ⁹ ( y ̄ ) . Otherwise, we simply take w X Δ ̄ ( y ̄ ) : = w â€Č ( y ̄ ) . Let

W 1 = { w X Δ ̄ ⁹ ( y ̄ ) ∣ w X Δ ̄ ⁹ ( y ̄ ) ≠ 1 , X Δ ̄ ∈ P 1 } .

If there is some word w X Δ ̄ ⁹ ( y ̄ ) ∈ W 1 which is explicitly oscillating in X Δ ̄ , let

P os : = { X Δ ̄ ∈ P 1 ∣ w X Δ ( y ̄ ) is explicitly oscillating in X Δ ̄ } .

If there is no explicitly oscillating w X Δ ̄ ⁹ ( y ̄ ) ∈ W 1 , then for every X Δ ̄ ∈ P 1 , we repeat the process described above replacing 𝒳 by X Δ ̄ and w X ⁹ ( y ̄ ) by w X Δ ̄ ⁹ ( y ̄ ) . For every X Δ ̄ ∈ P 1 , we define the family P X Δ ̄ 2 and the corresponding set of words W X Δ ̄ 2 exactly as P 1 and W 1 were defined above. Now let

P 2 : = ⋃ { P X Δ ̄ 2 ∣ X Δ ̄ ∈ P 1 } .

Let W 2 be the set of all words w V ⁹ ( y ̄ ) for V ∈ P 2 defined as w X Δ ̄ ⁹ ( y ̄ ) above. If there is an explicitly oscillating word in W 2 , then we define

P os : = { V ∈ P 2 ∣ w V ( y ̄ ) is explicitly oscillating in V }

and finish the construction. If there is no explicitly oscillating word in W 2 , then we continue this procedure. If, for some k ∈ N , W k contains explicitly oscillating words or W k = ∅ , then the procedure terminates.

To unify our notation, we will denote W 0 : = { w X ( y ̄ ) } .

Lemma 3.1

Procedure Transition terminates after finitely many steps.

Proof

Suppose that, for some k > 1 , we are given a word w V ⁹ ( y ̄ ) ∈ W k − 1 that is not explicitly oscillating. Thus w V ⁹ ( y ̄ ) ∉ F t , i.e. w V ⁹ ( y ̄ ) contains some constants v V , i ∈ G , 1 ≀ i ≀ n V .

Let Δ ̄ ∈ { 0 , 1 } n V , V Δ ̄ ∈ P V k and w V Δ ̄ ⁹ ( y ̄ ) ∈ W k be obtained from w V ⁹ ( y ̄ ) as in the construction. Since w V ⁹ ( y ̄ ) was not explicitly oscillating,

U 1 , 
 , 1 = ⋂ i = 1 n V v V , 1 − 1 ⁱ 
 ⁱ v V , i − 1 − 1 ⁱ ( supp ⁱ ( v V , i ) 1 ) = ∅ .

Thus, for some 𝑖, 1 ≀ i ≀ n V , v V , i becomes id in the word w V Δ ̄ ⁹ ( y ̄ ) . Therefore, the number of constants in w V Δ ̄ ⁹ ( y ̄ ) is strictly smaller than the corresponding number in w V ⁹ ( y ̄ ) . We now see that, after finitely many steps, we find some 𝑘 such that either W k contains an explicitly oscillating word or the words w V ⁹ ( y ̄ ) are equal to 1 for all V ∈ P k . ∎

Definition 3.2

Under the notation above, the initial word w X ⁹ ( y ̄ ) is called oscillating if it is explicitly oscillating or procedure Transition terminates producing the set P os ≠ ∅ . If the initial word w X ⁹ ( y ̄ ) is not oscillating and procedure Transition terminates for some k ∈ N such that, for all V ∈ P k , w V ⁹ ( y ̄ ) = 1 , then the word w X ⁹ ( y ̄ ) is called rigid.

Assume that w X ⁹ ( y ̄ ) is oscillating but not explicitly oscillating. If V ∈ P os and w V ∉ F t (considered in F t ∗ G ), we view it in the form

w V ⁹ ( y ̄ ) = u V , n V ⁹ v V , n V ⁹ 
 ⁹ u V , 1 ⁹ v V , 1 ,

where n V ∈ N and, for each i ≀ n V , the subword u V , i depends only on variables and v V , i ∈ G ∖ { 1 } . Let v V , 0 = 1 .

Define

O ̂ w : = ⋃ V ∈ P os ( ( V ∖ Fix ( G ) ) ∩ ⋂ i = 0 n V − 1 v V , 0 − 1 v V , 1 − 1 
 v V , i − 1 ( supp ( v V , i + 1 ) ) ) .

In particular, for w V ⁹ ( y ̄ ) ∈ F t , we have n V = 0 and the contribution of w V ⁹ ( y ̄ ) to the above union is V ∖ Fix ⁹ ( G ) .

To unify notation in the case of explicitly oscillating w X ⁹ ( y ̄ ) , we put O ̂ w = O w .

Example 3.3

Consider Thompson’s group đč with its standard action on [ 0 , 1 ] . We now give several illustrations of the notions introduced above. There are graphic illustrations of these cases in [21].

(a) Let

w 2 ⁱ ( y ) = x [ 0 , 1 2 ] , 0 − 1 ⁱ y ⁱ x [ 1 2 , 1 ] , 1 − 1 ⁱ y − 1 ⁱ x [ 0 , 1 2 ] , 1 ⁱ y ⁱ x [ 0 , 1 2 ] , 2 − 1 .

The word w 2 ⁹ ( y ) is not explicitly oscillating. Indeed,

x [ 0 , 1 2 ] , 2 ⁹ x [ 0 , 1 2 ] , 1 − 1 ⁹ ( ( 1 2 , 1 ) ) ∩ ( 0 , 1 2 ) = ∅ .

We claim that it is oscillating. To see this we apply Transition . There are four constant segments in w 2 ⁹ ( y ) ,

v [ 0 , 1 ] , 1 = x [ 0 , 1 2 ] , 2 − 1 , v [ 0 , 1 ] , 2 = x [ 0 , 1 2 ] , 1 , v [ 0 , 1 ] , 3 = x [ 1 2 , 1 ] , 1 − 1 , v [ 0 , 1 ] , 4 = x [ 0 , 1 2 ] , 0 − 1 .

It is clear that

supp ⁹ ( v [ 0 , 1 ] , 1 ) 1 = ( 3 8 , 1 2 ) and supp ⁹ ( v [ 0 , 1 ] , 1 ) 0 = ( 0 , 3 8 ) âˆȘ ( 1 2 , 1 ) , supp ⁹ ( v [ 0 , 1 ] , 2 ) 1 = ( 1 4 , 1 2 ) and supp ⁹ ( v [ 0 , 1 ] , 2 ) 0 = ( 0 , 1 4 ) âˆȘ ( 1 2 , 1 ) , supp ⁹ ( v [ 0 , 1 ] , 3 ) 1 = ( 3 4 , 1 ) and supp ⁹ ( v [ 0 , 1 ] , 3 ) 0 = ( 0 , 3 4 ) , supp ⁹ ( v [ 0 , 1 ] , 4 ) 1 = ( 0 , 1 2 ) and supp ⁹ ( v [ 0 , 1 ] , 4 ) 0 = ( 1 2 , 1 ) .

Thus the family P 1 for w 2 ⁹ ( y ) is { ( 0 , 1 4 ) , ( 1 4 , 3 8 ) , ( 3 8 , 1 2 ) , ( 1 2 , 3 4 ) , ( 3 4 , 1 ) } . Hence we obtain five reduced words

( w 2 ) ( 0 , 1 4 ) â€Č ⁹ ( y ) = x [ 0 , 1 2 ] , 0 − 1 ⁹ y , ( w 2 ) ( 1 4 , 3 8 ) â€Č ⁹ ( y ) = x [ 0 , 1 2 ] , 0 − 1 ⁹ x [ 0 , 1 2 ] , 1 ⁹ y , ( w 2 ) ( 3 8 , 1 2 ) â€Č ⁹ ( y ) = x [ 0 , 1 2 ] , 0 − 1 ⁹ x [ 0 , 1 2 ] , 1 ⁹ y ⁹ x [ 0 , 1 2 ] , 2 − 1 , ( w 2 ) ( 1 2 , 3 4 ) â€Č ⁹ ( y ) = y , ( w 2 ) ( 3 4 , 1 ) â€Č ⁹ ( y ) = y ⁹ x [ 1 2 , 1 ] , 1 − 1 .

The corresponding words

( w 2 ) ( 0 , 1 4 ) ⁹ ( y ) , ( w 2 ) ( 1 4 , 3 8 ) ⁹ ( y ) , ( w 2 ) ( 1 2 , 3 4 ) ⁹ ( y ) and ( w 2 ) ( 3 4 , 1 ) ⁹ ( y )

are explicitly oscillating. Note that ( w 2 ) ( 3 8 , 1 2 ) ⁹ ( y ) is non-trivial and not explicitly oscillating in ( 3 8 , 1 2 ) . Indeed,

supp ⁹ ( x [ 0 , 1 2 ] , 0 − 1 ⁹ x [ 0 , 1 2 ] , 1 ) ∩ supp ⁹ ( x [ 0 , 1 2 ] , 2 − 1 ) = ∅ .

This finishes the procedure and we see that w 2 ⁹ ( y ) is oscillating, where

O ̂ w 2 = ( 0 , 3 8 ) âˆȘ ( 1 2 , 1 ) .

(b) The following word is explicitly oscillating and has trivial product of constants:

w 3 ⁱ ( y ) = y ⁱ x 1 ⁱ y − 1 ⁱ x 1 − 1 .

Indeed,

O w 3 = x 1 ⁹ ( ( 1 2 , 1 ) ) ∩ ( 1 2 , 1 ) = ( 1 2 , 1 ) ,

and x 1 ⁱ x 1 − 1 = id .

On the other hand, the word

w 4 ⁱ ( y ) = y ⁱ x 1 ⁱ y − 1 ⁱ x [ 0 , 1 2 ] , 0 ⁱ y 2 ⁱ x 1 − 1

is not explicitly oscillating, because

O w 4 ⁹ ( y ) = x 1 ⁹ x [ 0 , 1 2 ] , 0 − 1 ⁹ ( ( 1 2 , 1 ) ) ∩ x 1 ⁹ ( ( 0 , 1 2 ) ) ∩ ( 1 2 , 1 ) = ∅ .

Since x 1 ⁱ x [ 0 , 1 2 ] , 0 ⁱ x 1 − 1 = x [ 0 , 1 2 ] , 0 and supp ⁱ ( x [ 0 , 1 2 ] , 0 ) = ( 0 , 1 2 ) , the word w 4 ⁱ ( y ) has non-trivial product of constants. We will see in Proposition 3.7 that this guarantees the w 4 ⁱ ( y ) is oscillating. The corresponding pictures can be found in [21].

Example 3.4

We start with a word, denoted by w 5 ⁹ ( y ) , which is rigid under the standard action of Thompson’s group đč. Using this, we construct w 6 ⁹ ( y 1 , y 2 ) such that Transition needs two steps in order to show that it is oscillating.

(a) Let

w 5 ⁱ ( y ) = y − 1 ⁱ x 1 ⁱ y ⁱ x [ 0 , 1 2 ] , 0 ⁱ y − 1 ⁱ x 1 − 1 ⁱ y ⁱ x [ 0 , 1 2 ] , 0 − 1 .

This word has four constant segments

v [ 0 , 1 ] , 1 = x [ 0 , 1 2 ] , 0 − 1 , v [ 0 , 1 ] , 2 = x 1 − 1 , v [ 0 , 1 ] , 3 = x [ 0 , 1 2 ] , 0 , v [ 0 , 1 ] , 4 = x 1 ,

which define the following supports:

supp ⁹ ( v [ 0 , 1 ] , 1 ) 1 = ( 0 , 1 2 ) and supp ⁹ ( v [ 0 , 1 ] , 1 ) 0 = ( 1 2 , 1 ) , supp ⁹ ( v [ 0 , 1 ] , 2 ) 1 = ( 1 2 , 1 ) and supp ⁹ ( v [ 0 , 1 ] , 2 ) 0 = ( 0 , 1 2 ) , supp ⁹ ( v [ 0 , 1 ] , 3 ) 1 = ( 0 , 1 2 ) and supp ⁹ ( v [ 0 , 1 ] , 3 ) 0 = ( 1 2 , 1 ) , supp ⁹ ( v [ 0 , 1 ] , 4 ) 1 = ( 1 2 , 1 ) and supp ⁹ ( v [ 0 , 1 ] , 4 ) 0 = ( 0 , 1 2 ) .

The family P 1 for w 5 ⁹ ( y ) is { ( 0 , 1 2 ) , ( 1 2 , 1 ) } . Thus Transition gives two words

( w 5 ) ( 0 , 1 2 ) â€Č ⁹ ( y ) = y − 1 ⁹ y ⁹ x [ 0 , 1 2 ] , 0 ⁹ y − 1 ⁹ y ⁹ x [ 0 , 1 2 ] , 0 − 1 , ( w 5 ) ( 1 2 , 1 ) â€Č ⁹ ( y ) = y − 1 ⁹ x 1 ⁹ y ⁹ y − 1 ⁹ x 1 − 1 ⁹ y .

In fact, both of them are equal to id . Hence W 1 is empty and therefore w 5 ⁹ ( y ) is rigid.

(b) In this example, we show that Transition sometimes needs several steps. Let 𝑣 and v â€Č be elements of đč with supp ⁹ ( v ) = [ 0 , 1 ] = supp ⁹ ( v â€Č ) . When [ a , b ] is a dyadic subinterval of [ 0 , 1 ] , let v [ a , b ] and v [ a , b ] â€Č be the corresponding elements defined on [ a , b ] . Let

v [ 0 , 1 ] , 1 = v [ 0 , 1 4 ] â€Č ⁹ x [ 1 4 , 1 2 ] , 0 ⁹ x [ 1 2 , 3 4 ] , 0 ⁹ v [ 3 4 , 1 ] , v [ 0 , 1 ] , 6 = v [ 0 , 1 4 ] â€Č ⁹ x [ 1 4 , 1 2 ] , 0 − 1 ⁹ x [ 1 2 , 3 4 ] , 0 − 1 ⁹ v [ 3 4 , 1 ] , v [ 0 , 1 ] , 7 = x [ 0 , 1 4 ] , 0 ⁹ v [ 1 4 , 1 2 ] â€Č ⁹ v [ 1 2 , 3 4 ] ⁹ x [ 3 4 , 1 ] , 0 , v [ 0 , 1 ] , 12 = x [ 0 , 1 4 ] , 0 − 1 ⁹ v [ 1 4 , 1 2 ] â€Č ⁹ v [ 1 2 , 3 4 ] ⁹ x [ 3 4 , 1 ] , 0 − 1 .

Note that these elements of đč have the same support: ( 0 , 1 ) ∖ { 1 4 , 1 2 , 3 4 } . Thus, computing supp 0 of them, we obtain ∅ in each case. In order to define the remaining constants, we use 1 2 -versions of the word w 5 defined in (a). Let

v [ 0 , 1 ] , 2 = v [ 0 , 1 ] , 8 = x [ 0 , 1 4 ] , 0 − 1 ⁱ x [ 1 2 , 3 4 ] , 0 − 1 , v [ 0 , 1 ] , 3 = v [ 0 , 1 ] , 9 = x [ 0 , 1 2 ] , 1 − 1 ⁱ x [ 1 2 , 1 ] , 1 − 1 , v [ 0 , 1 ] , 4 = v [ 0 , 1 ] , 10 = x [ 0 , 1 4 ] , 0 ⁱ x [ 1 2 , 3 4 ] , 0 , v [ 0 , 1 ] , 5 = v [ 0 , 1 ] , 11 = x [ 0 , 1 2 ] , 1 ⁱ x [ 1 2 , 1 ] , 1 .

Let

w 6 = v 12 ⁱ y 2 − 1 ⁱ y 1 − 1 ⁱ v 11 ⁱ y 1 ⁱ v 10 ⁱ y 1 − 1 ⁱ v 9 ⁱ y 1 ⁱ v 8 ⁱ y 2 ⁱ v 7 ⁱ y 1 ⁱ y 2 ⋅ v 6 ⁱ y 2 − 1 ⁱ y 1 − 1 ⁱ v 5 ⁱ y 1 ⁱ v 4 ⁱ y 1 − 1 ⁱ v 3 ⁱ y 1 ⁱ v 2 ⁱ y 2 ⁱ v 1 .

We omit [ 0 , 1 ] -indexes. Note that, between v 12 ⁹ y 2 − 1 and y 2 ⁹ v 7 , we have [ 0 , 1 2 ] - and [ 1 2 , 1 ] -versions of w 5 placed in parallel. The same word appears between v 6 ⁹ y 2 − 1 and y 2 ⁹ v 1 . The family of open subsets of [ 0 , 1 ] computed by Transition at Step 1 is as follows: P 1 = { ( 0 , 1 4 ) âˆȘ ( 1 2 , 3 4 ) , ( 1 4 , 1 2 ) âˆȘ ( 3 4 , 1 ) } . Applying the analysis of case (a), we obtain that the subword of w 6 between v 12 ⁹ y 2 − 1 and y 2 ⁹ v 7 is trivial for each element of P 1 . The same argument shows that the subword of w 6 between v 6 ⁹ y 2 − 1 and y 2 ⁹ v 1 is trivial too. After cancellation of any subwords y 2 − 1 ⁹ y 2 that appear, we have that w 6 becomes v 12 ⁹ v 7 ⁹ y 1 ⁹ y 2 ⁹ v 6 ⁹ v 1 on each element of P 1 . Note that supp ⁹ ( v 12 ⁹ v 7 ) = ( 1 4 , 3 4 ) ∖ { 1 2 } and supp ⁹ ( v 6 ⁹ v 1 ) = ( 0 , 1 4 ) âˆȘ ( 3 4 , 1 ) . In particular, the word v 12 ⁹ v 7 ⁹ y 1 ⁹ y 2 ⁹ v 6 ⁹ v 1 is not explicitly oscillating and the open sets mentioned in the previous sentence form P 2 . It is easy to see that

( w 6 ) supp ⁱ ( v 12 ⁱ v 7 ) = y 1 − 1 ⁱ y 2 − 1 ⁱ v 12 ⁱ v 7 and ( w 6 ) supp ⁱ ( v 6 ⁱ v 1 ) = y 1 ⁱ y 2 ⁱ v 6 ⁱ v 1 .

These words are explicitly oscillating.

The following lemma exhibits a correspondence between the existence of solutions of the inequalities w X ⁹ ( y ̄ ) ≠ 1 and w V ⁹ ( y ̄ ) ≠ 1 , where w V ⁹ ( y ̄ ) is derived from w X ⁹ ( y ̄ ) by Transition . We use the notation of this section and Transition .

Lemma 3.5

Suppose that k ≄ 1 , U ∈ P k , p ∈ U and g ̄ = ( g 1 , 
 , g t ) ∈ G . Assume w U ⁹ ( g ̄ ) ⁹ ( p ) ≠ p , where w U ⁹ ( y ̄ ) ∈ W k corresponds to 𝑈. If, for every 𝑖, 1 ≀ i ≀ t , and every ( Δ 1 , 
 , Δ j ) ∈ { 0 , 1 } j , 1 ≀ j ≀ n X , the element g i stabilizes v X , j Δ 1 ⁹ 
 ⁹ v X , 1 Δ j ⁹ ( U ) setwise, then w X ⁹ ( g ̄ ) ≠ 1 .

Proof

Fix some g ̄ ∈ G satisfying the conditions of the lemma. Suppose r ≀ k , U ⊆ V ⊆ V â€Č , V ∈ P V â€Č r and let w V â€Č ⁹ ( y ̄ ) be a word from W r − 1 which is not explicitly oscillating. Consider the word w V ⁹ ( y ̄ ) ∈ W r , which is obtained from w V â€Č ⁹ ( y ̄ ) by the appropriate reductions and conjugation. We will show that if, for some p â€Č ∈ U , w V ⁹ ( g ̄ ) ⁹ ( p â€Č ) ≠ p â€Č , then we have w V â€Č ⁹ ( g ̄ ) ⁹ ( p â€Čâ€Č ) ≠ p â€Čâ€Č , where p â€Čâ€Č ∈ U is obtained from p â€Č by (possibly) several applications of g i , 1 ≀ i ≀ t . This proves the lemma by induction starting with the case r = k and w U ⁹ ( y ̄ ) ∈ W k , where p â€Č : = p ∈ U as in the formulation. For ease of notation, we will put p â€Č = p below. Let

w V â€Č ( y ̄ ) : = u V â€Č , n V â€Č v V â€Č , n V â€Č u V â€Č , n V â€Č − 1 v V â€Č , n V â€Č − 1 
 u V â€Č , 1 v V â€Č , 1 ,

where n V â€Č ∈ N ∖ { 0 } , u V â€Č , i depends only on variables and we have v V â€Č , i ∈ G ∖ { 1 } for i ≀ n V â€Č . Assume that w V ⁹ ( y ̄ ) is obtained by reductions and conjugation from the word

w V Δ ̄ â€Č â€Č ⁹ ( y ̄ ) = u V â€Č , n V â€Č ⁹ v V â€Č , n V â€Č Δ n V â€Č ⁹ u V â€Č , n V â€Č − 1 ⁹ v V â€Č , n V â€Č − 1 Δ n V â€Č − 1 ⁹ 
 ⁹ u V â€Č , 1 ⁹ v V â€Č , 1 Δ 1

for some Δ ̄ ∈ { 0 , 1 } n V â€Č . Note that, from the description of Transition and the assumption that every element g i stabilizes every v X , j Δ 1 ⁹ 
 ⁹ v X , 1 Δ j ⁹ ( U ) setwise, we have that, for every j ≀ n V â€Č , elements of g ̄ stabilize v V â€Č , j ⁹ 
 ⁹ v V â€Č , 1 ⁹ ( U ) setwise. At this step of induction, we additionally assume that, in order to produce w V ⁹ ( y ̄ ) from w V â€Č ⁹ ( y ̄ ) , conjugation is not used. Otherwise, we replace 𝑝 by an appropriate u â€Č ⁹ ( g ̄ ) − 1 ⁹ ( p ) which is still in 𝑈.

Let us compute w V â€Č ⁹ ( g ̄ ) ⁹ ( p ) . To simplify notation, for each j ≀ n V â€Č , denote by p j â€Č the point u V â€Č , j − 1 ⁹ ( g ̄ ) ⁹ v V â€Č , j − 1 ⁹ 
 ⁹ u V â€Č , 1 ⁹ ( g ̄ ) ⁹ v V â€Č , 1 ⁹ ( p ) and by p j the point

u V â€Č , j − 1 ⁹ ( g ̄ ) ⁹ v V â€Č , j − 1 Δ j − 1 ⁹ 
 ⁹ u V â€Č , 1 ⁹ ( g ̄ ) ⁹ v V â€Č , 1 Δ 1 ⁹ ( p ) .

We put p 1 â€Č = p 1 = p and prove by induction that p j = p j â€Č .

Claim

Assume j ≀ n V â€Č and, for each i ≀ j , the points p i and p i â€Č coincide. Then

( † ) u V â€Č , j ⁹ ( g ̄ ) ⁹ v V â€Č , j ⁹ ( p j â€Č ) = u V â€Č , j ⁹ ( g ̄ ) ⁹ v V â€Č , j Δ j ⁹ ( p j ) .

Note that, in the case j = n V â€Č , the claim implies the following in/equality:

w V â€Č ⁹ ( g ̄ ) ⁹ ( p ) = u V â€Č , n V â€Č ⁹ ( g ̄ ) ⁹ v V â€Č , n V â€Č ⁹ ( p n V â€Č â€Č ) = w V ⁹ ( g ̄ ) ⁹ ( p ) ≠ p .

This will finish the proof of the lemma.

Proof of the claim. If Δ j = 1 , then u V â€Č , j ⁹ ( g ̄ ) ⁹ v V â€Č , j = u V â€Č , j ⁹ ( g ̄ ) ⁹ v V â€Č , j Δ j and we are done. Now assume that Δ j = 0 . We claim that p j â€Č ∈ X ∖ supp ⁹ ( v V â€Č , j ) . Indeed, since

V = V Δ ̄ â€Č = ⋂ i = 1 n V â€Č v V â€Č , 1 − 1 ⁹ 
 ⁹ v V â€Č , i − 1 − 1 ⁹ ( supp ⁹ ( v V â€Č , i ) Δ i ) ,

for each ℓ ≀ n V â€Č − 1 , either

v V â€Č , ℓ ⁹ 
 ⁹ v V â€Č , 1 ⁹ ( V ) ⊆ supp ⁹ ( v V â€Č , ℓ + 1 ) ⁹ or v V â€Č , ℓ ⁹ 
 ⁹ v V â€Č , 1 ⁹ ( V ) ⊆ X ∖ supp ⁹ ( v V â€Č , ℓ + 1 ) .

Since, for every i ≀ t and ℓ ≀ n V â€Č , the element g i stabilizes v V â€Č , ℓ ⁹ 
 ⁹ v V â€Č , 1 ⁹ ( U ) , we see that

u V â€Č , j − 1 ⁹ ( g ̄ ) ⁹ v V â€Č , j − 1 ⁹ 
 ⁹ u V â€Č , 1 ⁹ ( g ̄ ) ⁹ v V â€Č , 1 ⁹ ( U ) ⊆ ( X ∖ supp ⁹ ( v V â€Č , j ) ) .

Since p ∈ U , we have p j â€Č ∈ X ∖ supp ⁹ ( v V â€Č , j ) and

u V â€Č , j ⁹ ( g ̄ ) ⁹ v V â€Č , j ⁹ ( p j â€Č ) = u V â€Č , j ⁹ ( g ̄ ) ⁹ ( p j â€Č ) = u V â€Č , j ⁹ ( g ̄ ) ⁹ v V â€Č , j Δ j ⁹ ( p j ) . ∎

Example 3.6

Consider any G ≀ S fin ⁹ ( N ) with respect to the action on X = N . Assume that the word

w ⁹ ( y ̄ ) = u n ⁹ v n ⁹ u k − 1 ⁹ v k − 1 ⁹ 
 ⁹ u 1 ⁹ v 1

is given in the form (2.1) and the corresponding word u n ⁹ u n − 1 ⁹ 
 ⁹ u 1 (after reduction of all v i ) is non-trivial. Then w ⁹ ( y ̄ ) is oscillating. Indeed, applying the first step of Transition , we see that the region

X 0 ̄ : = ⋂ i = 1 n X v X , 1 − 1 
 v X , i − 1 − 1 ( supp ( v X , i ) 0 )

belongs to P os . As we already know, S fin ⁹ ( N ) is hereditarily separating on ℕ. By Abert’s theorem from [1], the inequality u k ⁹ u k − 1 ⁹ 
 ⁹ u 1 ≠ 1 has a solution in S fin ⁹ ( N ) . By Lemma 3.5, we obtain a solution of w ⁹ ( y ̄ ) ≠ 1 . This observation gives some kind of extension of Theorem 2.15.

The following proposition shows that oscillation unifies explicit oscillation with non-triviality of product of constants.

Proposition 3.7

Assume that đș acts on a Hausdorff topological space 𝒳 by homeomorphisms. Let w ⁹ ( y ̄ ) ∈ F t ∗ G be a word in the form (2.1). If w ⁹ ( y ̄ ) has non-trivial product of constants, then w ⁹ ( y ̄ ) is oscillating.

Proof

Assume that w ⁹ ( y ̄ ) is not explicitly oscillating and is in the form as in the beginning of the section,

w X ⁹ ( y ̄ ) = u X , n X ⁹ v X , n X ⁹ u X , n X − 1 ⁹ v X , n X − 1 ⁹ 
 ⁹ u X , 1 ⁹ v X , 1 ,

and supp ⁱ ( v X , n X ⁱ v X , n X − 1 ⁱ 
 ⁱ v X , 1 ) is a non-empty subset of 𝒳. Let 𝑝 belong to this support. Let

p j = v X , j ⁹ 
 ⁹ v X , 1 ⁹ ( p ) , 1 ≀ j ≀ n X .

We may assume that, when the point p j belongs to supp ⁹ ( v X , j ) ̄ , it already belongs to supp ⁹ ( v X , j ) . Indeed, if

p j ∈ supp ⁹ ( v X , j ) ̄ ∖ supp ⁹ ( v X , j ) ,

then we replace 𝑝 by a sufficiently close p â€Č ∈ supp ⁹ ( v X , n X ⁹ v X , n X − 1 ⁹ 
 ⁹ v X , 1 ) so that the corresponding p j â€Č belongs to supp ⁹ ( v X , j ) . Using continuity of all v X , i , we arrange that, for each i ≠ j ,

p i ∈ supp ⁹ ( v X , i ) âŸč p i â€Č ∈ supp ⁹ ( v X , i ) , p i ∉ supp ⁹ ( v X , i ) ̄ âŸč p i â€Č ∉ supp ⁹ ( v X , i ) ̄ .

Repeating such replacements enough times, we eventually obtain the required property.

For each j ≀ n X , we define Δ j to be 1 if p j ≠ p j − 1 and 0 otherwise. In particular, we see that, when Δ j = 0 , the point p j does not belong to supp ⁹ ( v X , j ) ̄ . Then it is easy to see that p ∈ X Δ ̄ , where

X Δ ̄ : = ⋂ i = 1 n X v X , 1 − 1 
 v X , i − 1 − 1 ( supp ( v X , i ) Δ i ) .

Let w X Δ ̄ ⁹ ( y ̄ ) be the corresponding word obtained by Transition . It has fewer constants than w ⁹ ( y ̄ ) has, by the assumption that the latter is not explicitly oscillating. Furthermore, w X Δ ̄ ⁹ ( 1 ̄ ) ⁹ ( p ) = w X ⁹ ( 1 ̄ ) ⁹ ( p ) and 𝑝 obviously belongs to the support of the product of constants of w X Δ ̄ ⁹ ( y ̄ ) . We now apply the same procedure to w X Δ ̄ ⁹ ( y ̄ ) and iterate it until we obtain an explicitly oscillating word. The necessity of the latter output follows from the fact that, at each step, we obtain a shorter word with non-trivial product of constants. ∎

3.2 Solving a system of inequalities in the case of oscillating words

In this section, we present a theorem which gives a sufficient condition for a system of inequalities over đș to have a solution in đș. In the formulation, we use Definition 3.2 and the notation given after it. In particular, recall that if w X ⁹ ( y ̄ ) is oscillating but not explicitly oscillating, then

O ̂ w : = ⋃ V ∈ P os ( ( V ∖ Fix ( G ) ) ∩ ⋂ i = 0 n V − 1 v V , 0 − 1 v V , 1 − 1 
 v V , i − 1 ( supp ( v V , i + 1 ) ) ) ,

where, for each V ∈ P os , the word w V ⁹ ( y ̄ ) is viewed in the form

u V , n V ⁱ v V , n V ⁱ 
 ⁱ u V , 1 ⁱ v V , 1

with v V , i ∈ G ∖ { 1 } and u V , i depending only on variables, i ≀ n V ( v V , 0 = 1 ). Also, recall that, in the case of explicitly oscillating w X ⁹ ( y ̄ ) , we put O ̂ w = O w .

Theorem 3.8

Let đș act on a perfect metric space ( X , ρ ) by homeomorphisms. Let { w 1 ⁹ ( y ̄ ) , w 2 ⁹ ( y ̄ ) , 
 , w m ⁹ ( y ̄ ) } be a set of reduced and non-constant words from F t ∗ G on 𝑡 variables y 1 , 
 , y t . If đș hereditarily separates 𝒳 and every w j ⁹ ( y ̄ ) , for j ≀ m , is oscillating, then the set of inequalities

w 1 ⁹ ( y ̄ ) ≠ 1 , w 2 ⁹ ( y ̄ ) ≠ 1 , 
 , w m ⁹ ( y ̄ ) ≠ 1

has a solution in đș.

Moreover, for any collection { O j } such that O j is an open subset of the set O ̂ w j , j ≀ m , there is a solution ( g 1 , 
 , g t ) of this set of inequalities such that supp ⁹ ( g i ) ⊆ ⋃ j = 1 m ( ⋃ V w j ⁹ ( O j ) ) for 1 ≀ i ≀ t .

Proof

Without loss of generality, assume that, when w j ⁹ ( y ̄ ) ∉ F t ,

w j ⁹ ( y ̄ ) = u j , n j ⁹ v j , n j ⁹ 
 ⁹ u j , 1 ⁹ v j , 1 for ⁹ 1 ≀ j ≀ m ⁹ with ⁹ v i ∈ G ∖ { 1 } .

Fix the collection { O j } from the statement of the theorem. For every j ≀ m , choose a point o j ∈ O j so that V w j ⁹ ( { o j } ) ∩ V w j â€Č ⁹ ( { o j â€Č } ) = ∅ when j , j â€Č ∈ { 1 , 
 , m } and j ≠ j â€Č . Using continuity of all v j , s , for every j ≀ m , we choose some open ball B j ⊆ O j such that the following conditions are satisfied:

  • o j ∈ B j for every j ≀ m and

  • ( ⋃ V w j ⁹ ( B j ) ) ∩ ( ⋃ V w j â€Č ⁹ ( B j â€Č ) ) = ∅ when j , j â€Č ∈ { 1 , 
 , m } and j ≠ j â€Č .

Now we construct a sequence ( g ̄ 0 , g ̄ 1 , 
 , g ̄ m ) , where g ̄ i = ( g i , 1 , 
 , g i , t ) , such that, for every 𝑖 with 1 ≀ i ≀ m , the following conditions are satisfied:
  • supp ⁹ ( g i ̄ ) ⊆ ⋃ j = 1 i V w j ⁹ ( B j ) ,

  • for all j , ℓ with 1 ≀ j ≀ t , 1 ≀ ℓ ≀ m and ℓ ≠ i , the restriction g i , j ↟ B ℓ coincides with g i − 1 , j ↟ B ℓ ,

  • g ̄ i is a solution of the set of inequalities w 1 ⁹ ( y ̄ ) ≠ 1 , 
 , w i ⁹ ( y ̄ ) ≠ 1 .

Fix g ̄ 0 : = ( 1 , 
 , 1 ) ∈ G t . Suppose that, after k − 1 steps, the tuple

g ̄ k − 1 = ( g k − 1 , 1 , 
 , g k − 1 , t )

is defined, k ≀ m . At the 𝑘-th step, we will modify the action of elements of this tuple on the ball B k so that g ̄ k satisfies w k ( g ̄ k ) ↟ B k ≠ id ↟ B k (i.e. g ̄ k is a solution of the inequality w k ⁹ ( y ̄ ) ≠ 1 ). For the 𝑘-th word w k ⁹ ( y ̄ ) , we consider two cases.

Case 1: w k ⁹ ( y ̄ ) is explicitly oscillating. Let us apply Theorem 2.13 to the word w k ⁹ ( y ̄ ) and the set B k ⊆ O w k . We obtain some solution f ̄ = ( f 1 , 
 , f t ) ∈ G of the inequality w k ⁹ ( y ̄ ) ≠ 1 such that supp ⁹ ( f i ) ⊆ V w k > 0 ⁹ ( B k ) , 1 ≀ i ≀ t . Now, for every i ≀ t , we define g k , i : = f i g k − 1 , i . Thus g ̄ k is also a solution of w k ⁹ ( y ̄ ) ≠ 1 . Since

⋃ i = 1 t supp ⁱ ( f i ) ⊆ V w k > 0 ⁱ ( B k ) and ⋃ i = 1 t supp ⁱ ( g k − 1 , i ) ⊆ ( X ∖ V w k ⁱ ( B k ) ) ,

the tuple ( g k , 1 , 
 , g k , t ) still is a solution of w j ⁹ ( y ̄ ) ≠ 1 for 1 ≀ j ≀ k − 1 .

Case 2: w k ⁹ ( y ̄ ) is oscillating but is not explicitly oscillating. Applying Transition , we then obtain non-empty P os corresponding to w k ⁹ ( y ̄ ) and B k . Thus there is some U ∈ P os , U ⊆ B k , and a word w U ⁹ ( y ̄ ) , which is derived from w k ⁹ ( y ̄ ) by cancellations of constants and reductions, which is explicitly oscillating on 𝑈. By Theorem 2.13, there is some f ̄ such that, for every i ≀ t , supp ⁹ ( f i ) ⊆ V w k ⁹ ( B k ) , for every j ≀ n k , f i stabilizes v j ⁹ 
 ⁹ v 1 ⁹ ( U ) setwise and w U ⁹ ( f ̄ ) ⁹ ( p ) ≠ p for some point p ∈ U . Thus, by Lemma 3.5, we have w k ⁹ ( f ̄ ) ≠ 1 .

Now, similarly to above, for every i ≀ t , we define g k , i : = f i g k − 1 , i . This gives a solution of all the inequalities w j ⁹ ( y ̄ ) ≠ 1 for j ≀ k .

Thus, after 𝑚 steps of the algorithm, we obtain a tuple g ̄ m , which is the solution of the system w 1 ⁹ ( y ̄ ) ≠ 1 , 
 , w m ⁹ ( y ̄ ) ≠ 1 . Moreover, for any 1 ≀ i ≀ t , supp ⁹ ( g i ) ⊆ ⋃ j = 1 m V w j ⁹ ( O j ) . ∎

Remark 3.9

The conclusion of Theorem 3.8 can be extended by the following additional statement: for any 𝜀, the solution g ̄ can be chosen so that it additionally satisfies the inequality d ⁹ ( x , g i ⁹ ( x ) ) ≀ Δ for all i ≀ t and x ∈ X .

To see this, it is enough to add the following argument at each step of the inductive procedure of the proof of Theorem 3.8. Using Remark 2.14 at each step of the induction, we choose the corresponding solution f ̄ of w k ⁹ ( y ̄ ) ≠ 1 with the additional property that d ⁹ ( f i ⁹ ( x ) , x ) ≀ Δ for all i ≀ t and x ∈ X . Then one easily sees by inspection of the proof that the choice of the family { B j ∣ 1 ≀ j ≀ m } guarantees that d ⁹ ( g i ⁹ ( x ) , x ) ≀ Δ for all j ≀ t and x ∈ X .

Remark 3.10

One might wonder if, in order to solve the system of inequalities given in Theorem 3.8, it is possible to apply the trick of commutators used in the proof of Proposition 2.18. We should mention here that the situation of Proposition 2.18 is very special, where the commutator of oscillating words is again an oscillating word. This is not true in general and cannot be applied in Theorem 3.8.

4 Actions of topological groups

4.1 Polish đș-spaces

A Polish space (group) is a separable, completely metrizable topological space (group). The corresponding metric extends to tuples by

d ⁱ ( ( x 1 , 
 , x m ) , ( y 1 , 
 , y m ) ) = max ⁱ ( d ⁱ ( x 1 , y 1 ) , 
 , d ⁱ ( x m , y m ) ) .

Let ( X , d ) be a Polish space and Iso ⁹ ( X ) the corresponding isometry group endowed with the pointwise convergence topology. Then Iso ⁹ ( X ) is a Polish group. A compatible left-invariant metric can be obtained as follows. Fix a countable dense set S = { s i ∣ i ∈ { 1 , 2 , 
 } } ⊆ X . Define, for two isometries đ›Œ and đ›œ of 𝒳,

ρ S ⁹ ( α , ÎČ ) = ∑ i = 1 ∞ 2 − i ⁹ min ⁹ ( 1 , d ⁹ ( α ⁹ ( s i ) , ÎČ âą ( s i ) ) ) .

Let đș be a closed subgroups of Iso ⁹ ( X ) . We fix a dense countable set ΄ ⊂ G and distinguish a base of G t consisting of all sets of the following form. Let s ̄ 1 , 
 , s ̄ t be a sequence of tuples from 𝑆, q ∈ Q and let h 1 , 
 , h t be a sequence from ΄. Define

N q ( s ̄ 1 , h 1 , 
 , s ̄ t , h t ) = { ( g 1 , 
 , g t ) ∣ d ( g i ( s ̄ i ) , h i ( s ̄ i ) < q ,  1 ≀ i ≀ t } .

The family of all N q ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t ) forms a base of the topology of G t . This material can be found in any textbook on descriptive set theory, for example [13].

The following theorem is related to Theorem 3.8 and Remark 3.9.

Theorem 4.1

Let đș be a closed subgroup of the isometry group Iso ⁹ ( X ) of a perfect Polish space 𝒳. Assume that the action of đș is hereditarily separating on 𝒳. Then, for any oscillating word w ⁹ ( y ̄ ) from F t ∗ G on 𝑡 variables, y 1 , 
 , y t , the set { g ̄ ∣ w ⁹ ( g ̄ ) ≠ 1 } is dense in G t .

Proof

Let us fix an inequality w ⁹ ( y ̄ ) ≠ 1 over đș and assume that w ⁹ ( y ̄ ) is explicitly oscillating. If w ⁹ ( y ̄ ) ∉ F t , we may assume that it is in the form (2.2). Let N q ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t ) be a basic open set defined before the theorem. Let 𝑝 be an element of 𝒳 which does not occur in s ̄ 1 , 
 , s ̄ t . Define

U q ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t , p , w ⁹ ( y ̄ ) )

to be the set of tuples from N q ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t ) which are distinctive for w ⁹ ( y ̄ ) and 𝑝. It is easy to see that U q is open in N q . For example, one can repeat the argument given in [1, p. 530] (extended by application of homeomorphisms v j ).

Assume that N q ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t ) ≠ ∅ . We will prove that if

p ∈ O w ∖ ⋃ V w − 1 ⁹ ( ⋃ V w ⁹ ( O ̄ w ∖ O w ) ) ,

then the set U q ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t , p , w ⁹ ( y ̄ ) ) is not empty. Since this can be applied to every 𝑞 and every tuple s ̄ 1 â€Č , h 1 â€Č , 
 , s ̄ t â€Č , h t â€Č such that s ̄ i ⊆ s ̄ i â€Č and h i â€Č agrees with h i on s ̄ i , 1 ≀ i ≀ t , we would obtain that U q ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t , p , w ⁹ ( y ̄ ) ) is dense in N q ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t ) .

The case | w ⁹ ( y ̄ ) | = 0 is degenerate; we have N q = U q . We now apply some ideas from the proof of Theorem 2.13. At the 𝑘-th step of the induction, we will show that

  • there is a tuple

    ( g 1 , 
 , g t ) ∈ U q ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t , p , ( w ) k ⁹ ( y ̄ ) ) .

  • In the condition above, we can choose g ̄ so that, for all 𝑖 with 1 ≀ i ≀ t ,

    supp ⁹ ( g i ) ⊆ ( ⋃ V w > 0 ⁹ ( O w ) ) ∖ ( ⋃ V [ w ] k − 1 ⁹ ( ⋃ V w ⁹ ( O w ̄ ∖ O w ) ) )

    and each member of V w ⁹ ( O w ) is g i -invariant.

We start by fixing some ( g 1 ∘ , g 2 ∘ , 
 , g t ∘ ) ∈ N q ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t ) and q â€Č < q such that d ⁹ ( g i ∘ ⁹ ( s ̄ i ) , h i ⁹ ( s ̄ i ) ) < q â€Č for all 𝑖 with 1 ≀ i ≀ t . We will assume w ⁹ ( y ̄ ) ∉ F t . When w ⁹ ( y ̄ ) ∈ F t , the argument below works. In fact, after removing the corresponding v i , it becomes easier. In particular, we assume that ( w ) 1 ⁹ ( y ̄ ) is of the form y j ± 1 ⁹ v 1 for some 1 ≀ j ≀ t . When w ⁹ ( y ̄ ) ∉ F t , then p â€Č ≠ v 1 ⁹ ( p â€Č ) for all p â€Č ∈ O w . Thus, according to the assumptions on 𝑝, for a non-trivial v 1 , the inequality p ≠ v 1 ⁹ ( p ) is satisfied.

Without loss of generality, suppose ( w ) 1 ⁹ ( y ̄ ) = y 1 ⁹ v 1 . If g 1 ∘ ⁹ v 1 ⁹ ( p ) ∉ { p , v 1 ⁹ ( p ) } , then

( g 1 ∘ , g 2 ∘ , 
 , g t ∘ ) ∈ U q ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t , p , ( w ) 1 ⁹ ( y ̄ ) ) .

In the contrary case, using the argument of Step k = 1 of the proof of Theorem 2.13 and Remark 2.14, we can find

( g 1 â€Č , g 2 â€Č , 
 , g t â€Č ) ∈ U q − q â€Č ⁹ ( s ̄ 1 , id , 
 , s ̄ t , id , p , ( w ) 1 ⁹ ( y ̄ ) )

such that

( w ) 1 ⁹ ( g ̄ â€Č ) ⁹ ( p ) ∉ { p , v 1 ⁹ ( p ) , ( g 1 ∘ ) − 1 ⁹ ( p ) , ( g 1 ∘ ) − 1 ⁹ ( v 1 ⁹ ( p ) ) } .

Then

( g 1 ∘ ⁹ g 1 â€Č , g 2 ∘ ⁹ g 2 â€Č , 
 , g t ∘ ⁹ g t â€Č ) ∈ U q ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t , p , ( w ) 1 ⁹ ( y ̄ ) ) ,

i.e. we may take g i = g i ∘ ⁹ g i â€Č , 1 ≀ i ≀ n . It is worth noting that, instead of q − q â€Č , one could take arbitrarily small q â€Čâ€Č . In particular, ( g 1 ∘ ⁹ g 1 â€Č , g 2 ∘ ⁹ g 2 â€Č , 
 , g t ∘ ⁹ g t â€Č ) can be chosen arbitrarily close to ( g 1 ∘ , g 2 ∘ , 
 , g t ∘ ) .

If | w ⁹ ( y ̄ ) | > 2 , then for the second step, we additionally need that

g 1 ⁹ ( v 1 ⁹ ( p ) ) ∉ s ̄ 1 âˆȘ ⋯ âˆȘ s ̄ t .

This can be arranged at the first step by the stronger demand that

g 1 â€Č ⁹ ( v 1 ⁹ ( p ) ) ∉ { p , v 1 ⁹ ( p ) , ( g 1 ∘ ) − 1 ⁹ ( p ) , ( g 1 ∘ ) − 1 ⁹ ( v 1 ⁹ ( p ) ) } âˆȘ ( g 1 ∘ ) − 1 ⁹ ( s ̄ 1 ) âˆȘ ⋯ âˆȘ ( g 1 ∘ ) − 1 ⁹ ( s ̄ t ) .

In fact, our argument shows that the set

{ ( g 1 , g 2 , 
 , g t ) ∈ U q ( s ̄ 1 , h 1 , 
 , s ̄ t , h t , p , ( w ) 1 ( y ̄ ) ) | ( w ) 1 ( g ̄ ) ( p ) ) ∉ { p , v 1 ( p ) } âˆȘ s ̄ 1 âˆȘ ⋯ âˆȘ s ̄ t }

is dense in N q ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t ) .

For a fixed g ̄ from this part of U q ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t , p , ( w ) 1 ⁹ ( y ̄ ) ) , take q 1 < q with g ̄ ∈ U q 1 ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t , p , ( w ) 1 ⁹ ( y ̄ ) ) . At Step 2, we look for an element of

U q − q 1 ⁹ ( v 1 ⁹ ( p ) ⁹ s ̄ 1 , g 1 , 
 , s ̄ t , g t , p , ( w ) 2 ⁹ ( y ̄ ) ) ⊂ N q ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t ) .

Skipping further considerations of this step, we jump to Step 𝑘.

After Step k − 1 , we have some q k − 1 < q and some tuple g ̄ in

U q k − 1 ⁹ ( s ̄ 1 â€Č , h 1 â€Č , 
 , s ̄ t â€Č , h t â€Č , p , ( w ) k − 1 ⁹ ( y ̄ ) ) ⊆ N q ⁹ ( s ̄ 1 , h 1 , 
 , s ̄ t , h t ) ,

where s ̄ i ⊆ s ̄ i â€Č , i ≀ t and each h i â€Č agrees with h i on s ̄ i . We look for a tuple from

U q − q k − 1 ⁹ ( s ̄ 1 â€Č , h 1 â€Čâ€Č , 
 , s ̄ t â€Č , h t â€Čâ€Č , p , ( w ) k ⁹ ( y ̄ ) )

where each h i â€Čâ€Č agrees with g i on s ̄ i â€Č . Here we again have two cases.

Case 1:

( w ) k ⁹ ( y ̄ ) = u d , s + 1 ⁹ u d , s ⁹ 
 ⁹ u d , 1 ⁹ v d ⁹ 
 ⁹ u 2 , 1 ⁹ v 2 ⁹ u 1 , ℓ 1 ⁹ 
 ⁹ u 1 , 1 ⁹ v 1 ,

where k − 1 = L d − 1 + s , s ≄ 1 . Our argument is a slight modification of the corresponding part in the proof of Theorem 2.13. We preserve the notation of that proof. In particular, for a fixed g ̄ as above, we take the corresponding p i , g ̄ , i ≀ k .

If

p k , g ̄ ∉ { p i , g ̄ ∣ 0 ≀ i ≀ k − 1 } âˆȘ { v 1 ⁹ ( p 0 , g ̄ ) , 
 , v d ⁹ ( p L d − 1 , g ̄ ) } ,

then we have found an acceptable tuple g ̄ . Let us assume that p k , g ̄ = p m , g ̄ for some 0 ≀ m < k or p k , g ̄ = v m + 1 ⁹ ( p L m , g ̄ ) for some 0 ≀ m < d − 1 .

As in the proof of Theorem 2.13, we may assume that u d , s + 1 = y j and again put

Y : = { p i , g ̄ ∣ 0 ≀ i ≀ k − 2 } âˆȘ { v 1 ( p 0 , g ̄ ) , 
 , v d ( p L d − 1 , g ̄ ) } .

The neighborhood O ⊆ v d ⁹ 
 ⁹ v 1 ⁹ ( O w ) of the point p k − 1 , g ̄ ∈ v d ⁹ 
 ⁹ v 1 ⁹ ( O w ) is chosen as before. We can take it sufficiently small so that, when we define

f ∈ stab G ⁹ ( ( X ∖ O ) âˆȘ ⋃ V [ w ] k − 1 − 1 ⁹ ( ⋃ V w ⁹ ( O w ̄ ∖ O w ) ) âˆȘ Y ) ,

we have that max 1 ≀ i ≀ t ⁹ { d ⁹ ( s ̄ i â€Č , f ⁹ ( s ̄ i â€Č ) ) } ≀ q − q k − 1 . Since we want

p k , g ̄ ∉ Y âˆȘ s ̄ 1 âˆȘ ⋯ âˆȘ s ̄ t ,

we arrange 𝑓 taking p k − 1 , g ̄ outside 𝑍, where

Z := { g j − 1 ⁹ ( p i , g ̄ ) ∣ 0 ≀ i ≀ k − 1 } âˆȘ { g j − 1 ⁹ ( v i + 1 ⁹ ( p L i , g ̄ ) ) ∣ 0 ≀ i ≀ d − 1 } âˆȘ ( g j ) − 1 ⁹ ( s ̄ 1 â€Č ) âˆȘ ⋯ âˆȘ ( g j ) − 1 ⁹ ( s ̄ t â€Č ) .

We apply hereditary separation at this point. Replacing g j by g j ⁹ f , we obtain a corrected tuple g ̄ . By the choice of q k − 1 and 𝑂, this tuple represents

U q − q k − 1 ⁹ ( s ̄ 1 â€Č , h 1 â€Čâ€Č , 
 , s ̄ t â€Č , h t â€Čâ€Č , p , ( w ) k ⁹ ( y ̄ ) ) ⊂ U q ⁹ ( s ̄ 1 â€Č , h 1 â€Č , 
 , s ̄ t â€Č , h t â€Č , p , ( w ) k ⁹ ( y ̄ ) ) .

We finish the proof of Case 1 as in Theorem 2.13.

Case 2:

( w ) k = u d + 1 , 1 ⁱ v d + 1 ⁱ u d , s ⁱ 
 ⁱ u d , 1 ⁱ v d ⁱ 
 ⁱ u 2 , 1 ⁱ v 2 ⁱ u 1 , ℓ 1 ⁱ 
 ⁱ u 1 , 1 ⁱ v 1 ,

where k = L d + 1 . This corresponds to Case 2 of the proof of Theorem 2.13 and it can be treated in the same fashion. The rest of the argument for explicitly oscillating w ⁹ ( y ̄ ) is clear.

Let us consider the case when w ⁹ ( y ̄ ) is oscillating but not explicitly oscillating. Then, applying Transition , we obtain non-empty P os corresponding to w ⁹ ( y ̄ ) . Thus there is some word w V ⁹ ( y ̄ ) derived from w ⁹ ( y ̄ ) by cancellation of constants and reduction, which is explicitly oscillating. Then, fixing N q ⁹ ( x ̄ 1 , h 1 , 
 , x ̄ t , h t ) , we repeat the proof above replacing w ⁹ ( y ̄ ) by w V ⁹ ( y ̄ ) . Take a point p ∈ V according to the requirements of that proof. Then we see that U q ⁹ ( x ̄ 1 , h 1 , 
 , x ̄ t , h t , p , w V ⁹ ( y ̄ ) ) is dense in N q ⁹ ( x ̄ 1 , h 1 , 
 , x ̄ t , h t ) . In particular, so is the set

{ ( g 1 , g 2 , 
 , g t ) ∈ N q ∣ w V ⁹ ( g ̄ ) ≠ 1 } .

Thus, by Lemma 3.5, we have that { ( g 1 , g 2 , 
 , g t ) ∈ N q ∣ w ⁹ ( g ̄ ) ≠ 1 } is dense in N q ⁹ ( x ̄ 1 , h 1 , 
 , x ̄ t , h t ) . ∎

4.2 Topological actions of locally compact groups

A locally compact group đș carries a left đș-invariant Haar measure 𝜇. An action of đș on a set 𝑋 is called topological if each point stabilizer stab G ⁹ ( x ) , x ∈ X , is closed and of measure 0. An action is called strongly topological if, for any finite Y ⊆ X , each point stabilizer stab G ⁹ ( Y âˆȘ { x } ) , x ∈ X ∖ Y , is closed in stab G ⁹ ( Y ) and has measure 0 under the Haar measure for stab G ⁹ ( Y ) . Under the assumption that all point-stabilizers are non-trivial, this condition strengthens separating actions.

In this section, we concentrate on strongly topological actions of locally compact groups. We will assume that 𝑋 is a Polish space, say ( X , d ) , and đș acts on 𝒳 isometrically and strongly topologically. We preserve the notation of Section 4.1.

If đ» is a closed subgroup of đș, it has a left Haar measure 𝜆. If đŸ is a closed subgroup of đ» of zero measure (and of infinite index), then the space H / K inherits a natural measure λ ̄ which is equivariant, i.e. đ»-translates of λ ̄ -negligible sets are λ ̄ -negligible. As đŸ is a closed subgroup of đ», it also has a left Haar measure, say ÎŒ K .

The following is [1, Lemma 3.1].

Lemma 4.2

Let H 0 be a 𝜆-measurable subset of đ» such that, for almost all đŸ-cosets D ⊆ H , we have ÎŒ K ⁹ ( D ∩ H 0 ) = 0 . Then λ ⁹ ( H 0 ) = 0 .

We will apply this lemma to left cosets of đ» instead of đ». These cosets are considered under an obvious extension of the measure of đ». Then H 0 from the formulation will be a subset of some N q ⁹ ( s , id ) , s ∈ S . Typically, đŸ from the lemma arises as the stabilizer of some p ∈ X . Under our assumptions, ÎŒ ⁹ ( stab G ⁹ ( p ) ) = 0 .

The following theorem is a version of [1, Theorem 1.5]. It looks slightly technical, but in Corollary 4.4, we give a short and natural formulation.

Theorem 4.3

Let đș be a locally compact topological group acting strongly topologically on a Polish space ( X , d ) by isometries. Assume that, for every finite Y ⊂ X , the stabilizer stab G ⁹ ( Y ) is not trivial. Let s ∈ S , q ∈ Q .

  • Let w ⁹ ( y ̄ ) be an explicitly oscillating word such that O w contains B ( | w ⁹ ( y ̄ ) | + 1 ) ⁹ q ⁹ ( s ) and each B ℓ ⁹ q ⁹ ( s ) with 1 ≀ ℓ ≀ | w ⁹ ( y ̄ ) | + 1 is infinite and invariant with respect to each constant v i of the word w ⁹ ( y ̄ ) .

  • Let Îł ̄ be the random 𝑡-tuple in N q ⁹ ( s , id ) .

Then, almost surely, Îł ̄ satisfies the mixed inequality w ⁹ ( y ̄ ) ≠ 1 .

Proof

We adapt the proof of [1, Theorem 1.5]. Let x ̄ 1 , 
 , x ̄ t be a sequence of tuples from 𝒳 and let h 1 , 
 , h t be a sequence from N q ⁹ ( s , id ) . Define

A ⁹ ( x ̄ 1 , h 1 , 
 , x ̄ t , h t ) = { ( g 1 , 
 , g t ) ∣ g i ⁹ ( x ̄ i ) = h i ⁹ ( x ̄ i ) ,  1 ≀ i ≀ t } .

We view this set as follows. Let G 0 be the direct product of stabilizers

stab G ⁹ ( x ̄ 1 ) × ⋯ × stab G ⁹ ( x ̄ t ) .

Then A ⁹ ( x ̄ 1 , h 1 , 
 , x ̄ t , h t ) is a right coset of G 0 . It inherits the natural topology and measure from G 0 .

In order to introduce the next notation used in the proof, let w ⁹ ( y ̄ ) be an explicitly oscillating word and ℓ a natural number. If w ⁹ ( y ̄ ) ∉ F t , we may assume that it is in the form (2.2). Now let 𝑝 be an element of B ℓ ⁹ q ⁹ ( s ) such that neither 𝑝 nor v 1 ⁹ ( p ) occurs in x ̄ 1 , 
 , x ̄ t . Write U ⁹ ( x ̄ 1 , h 1 , 
 , x ̄ t , h t , p , w ⁹ ( y ̄ ) ) for the set of tuples from A ⁹ ( x ̄ 1 , h 1 , 
 , x ̄ t , h t ) which are distinctive for w ⁹ ( y ̄ ) and 𝑝 and the elements

p = p 0 , g ̄ , v 1 ⁹ ( p 0 , g ̄ ) , 
 , p l 1 , g ̄ , v 2 ⁹ ( p l 1 , g ̄ ) , 
 , p L n , g ̄

do not occur in x ̄ 1 , 
 , x ̄ t . Then 𝑈 is open in 𝐮. This situation is similar to that which appeared in the proof of Theorem 4.1. Again, in order to show openness, we can repeat the argument given in [1, p. 530] (extended by application of homeomorphisms v j ).

Applying induction on k = | w ⁹ ( y ̄ ) | , we will prove that if B ( k + ℓ ) ⁹ q ⁹ ( s ) ⊂ O w , every B m ⁹ q ⁹ ( s ) with 1 ≀ m ≀ k + ℓ + 1 is infinite and invariant with respect to each constant v i of the word w ⁹ ( y ̄ ) ,

x ̄ 1 âˆȘ ⋯ âˆȘ x ̄ t ⊂ B ℓ ⁹ q ⁹ ( s ) and { p , v 1 ⁹ ( p ) } ⊂ B ℓ ⁹ q ⁹ ( s ) ∖ ( x ̄ 1 âˆȘ ⋯ âˆȘ x ̄ t ) ,

then the set U ⁹ ( x ̄ 1 , h 1 , 
 , x ̄ t , h t , p , w ⁹ ( y ̄ ) ) is almost surely in

( N q ⁹ ( s , id ) ) t ∩ A ⁹ ( x ̄ 1 , h 1 , 
 , x ̄ t , h t ) ,

i.e. ÎŒ ⁹ ( ( N q ⁹ ( s , id ) t ∩ A ) ∖ U ) = 0 .

The case | w ⁹ ( y ̄ ) | = 0 is degenerate and obvious: A = U . When w ⁹ ( y ̄ ) ∈ F t , the argument coincides with the corresponding one from the proof of [1, Theorem 1.5]. Basically, this is a simplified version of the argument below for w ⁹ ( y ̄ ) ∉ F t . We will also apply some ideas from the proof of Theorem 2.13.

Consider Step k > 0 . We assume that the statement

ÎŒ ⁹ ( ( ( N q ⁹ ( s , id ) ) t ∩ A ) ∖ U ) = 0

holds for all words of length k â€Č < k , all tuples of parameters x ̄ 1 â€Č , h 1 â€Č , 
 , x ̄ t â€Č , h t â€Č with x ̄ 1 â€Č , 
 , x ̄ t â€Č ∈ B ℓ ⁹ q ⁹ ( s ) and all 𝑝 taken as it was described above. We will assume that ( w ) 1 is of the form y i ± 1 ⁹ v 1 for some 1 ≀ i ≀ t . Our argument also works in the case v 1 = 1 (which is now allowed). Note that, in the case v 1 ≠ 1 , according to the assumptions on 𝑝, we have p ≠ v 1 ⁹ ( p ) and v 1 ⁹ ( p ) ∈ B ℓ ⁹ q ⁹ ( s ) .

Without loss of generality, suppose ( w ) 1 = y i ⁹ v 1 . The set

{ ( g 1 , g 2 , 
 , g t ) ∈ A | g i ⁹ ( v 1 ⁹ ( p ) ) ∈ x ̄ 1 âˆȘ h 1 ⁹ ( x ̄ 1 ) âˆȘ ⋯ âˆȘ x ̄ t âˆȘ h t ⁹ ( x ̄ t ) âˆȘ { p , v 1 ⁹ ( p ) } }

is a finite union of cosets of the direct product

S v 1 ⁹ ( p ) = stab G ⁹ ( x ̄ 1 ) × ⋯ × stab G ⁹ ( { v 1 ⁹ ( p ) } âˆȘ x ̄ i ) × ⋯ × stab G ⁹ ( x ̄ t ) .

Since đș acts strongly topologically on 𝒳, we deduce that, for almost all

( g 1 , g 2 , 
 , g t ) ∈ ( N q ⁹ ( s , id ) ) t ∩ A ,

we have

g i ⁹ ( v 1 ⁹ ( p ) ) ∉ x ̄ 1 âˆȘ h 1 ⁹ ( x ̄ 1 ) âˆȘ ⋯ âˆȘ x ̄ t âˆȘ h t ⁹ ( x ̄ t ) âˆȘ { p , v 1 ⁹ ( p ) } .

For such a tuple and for a fixed p â€Č of the form g i ( ( v 1 ( p ) ) , let h i â€Č be any element of N q ⁹ ( s , id ) mapping { v 1 ⁹ ( p ) } âˆȘ x ̄ i to { p â€Č } âˆȘ h ⁹ ( x ̄ i ) . Note that p â€Č ∈ B ( ℓ + 1 ) ⁹ q ⁹ ( s ) .

Consider

U ⁹ ( x ̄ 1 , h 1 , 
 , { v 1 ⁹ ( p ) } âˆȘ x ̄ i , h i â€Č , 
 , x ̄ t , h t , p â€Č , [ w ] 1 ⁹ ( y ̄ ) )

in ( N q ⁹ ( s , id ) ) t ∩ A ⁹ ( x ̄ 1 , h 1 , 
 , { v 1 ⁹ ( p ) } âˆȘ x ̄ i , h i â€Č , 
 , x ̄ t , h t ) . Since

B ( ℓ + 1 + | [ w ] 1 | ) ⁱ q ⁱ ( s ) ⊆ O [ w ] 1

(by the inductive assumptions and assumptions on constants in w ⁹ ( y ̄ ) ) and

p â€Č ∈ B ( ℓ + 1 ) ⁹ q ⁹ ( s ) ∖ ( { p , v 1 ⁹ ( p ) } âˆȘ x ̄ 1 âˆȘ ⋯ âˆȘ x ̄ t ) ,

applying induction, we see that

U ⁹ ( x ̄ 1 , h 1 , 
 , { v 1 ⁹ ( p ) } âˆȘ x ̄ i , h i â€Č , 
 , x ̄ t , h t , p â€Č , [ w ] 1 ⁹ ( y ̄ ) )

is almost surely in ( N q ⁹ ( s , id ) ) t ∩ A ⁹ ( x ̄ 1 , h 1 , 
 , { v 1 ⁹ ( p ) } âˆȘ x ̄ i , h i â€Č , 
 , x ̄ t , h t ) . In particular,

U ⁹ ( x ̄ 1 , h 1 , 
 , x ̄ i , h i , 
 , x ̄ t , h t , p , w ⁹ ( y ̄ ) )

is almost surely in ( N q ⁹ ( s , id ) ) t ∩ A ⁹ ( x ̄ 1 , h 1 , 
 , { v 1 ⁹ ( p ) } âˆȘ x ̄ i , h i â€Č , 
 , x ̄ t , h t ) . Using Lemma 4.2 in the situation when K = stab G ⁹ ( x ̄ 1 , 
 , { v 1 ⁹ ( p ) } âˆȘ x ̄ i , 
 , x ̄ t ) , we conclude that

U ⁹ ( x ̄ 1 , h 1 , 
 , x ̄ t , h t , p , w ⁹ ( y ̄ ) )

is almost surely in ( N q ⁹ ( s , id ) ) t ∩ A ⁹ ( x ̄ 1 , h 1 , 
 , x ̄ t , h t ) .

Let us fix a mixed inequality w ⁹ ( y ̄ ) ≠ 1 and assume that w ⁹ ( y ̄ ) is explicitly oscillating and satisfies the conditions of the formulation of the theorem. Take p ∈ B q ⁹ ( s ) . Apply the claim proved by induction to A = G t and U ⁹ ( p , w ⁹ ( y ̄ ) ) where ℓ = 1 . We obtain that ÎŒ ⁹ ( ( N q ⁹ ( s , id ) ) t ∖ U ) = 0 . ∎

The following statement is proved by the same proof as in Theorem 4.3. One only has to replace balls B ℓ ⁹ q ⁹ ( s ) by 𝒳 and N q ⁹ ( s , id ) by đș.

Corollary 4.4

Let đș be a locally compact topological group acting strongly topologically on a Polish space ( X , d ) by isometries. Assume that, for every finite Y ⊂ X , the stabilizer stab G ⁹ ( Y ) is not trivial. Let Îł ̄ be a random 𝑡-tuple in đș and w ⁹ ( y ̄ ) an explicitly oscillating word such that O w = X . Then, almost surely, Îł ̄ satisfies the mixed inequality w ⁹ ( y ̄ ) ≠ 1 .

Corollary 4.5

Let đș be a compact topological group acting topologically on a Polish space ( X , d ) by isometries. Assume that the action is separating. Let Îł ̄ be a random 𝑡-tuple from đș and w ⁹ ( y ̄ ) an explicitly oscillating word such that O w = X . Then, almost surely, Îł ̄ satisfies the mixed inequality w ⁹ ( y ̄ ) ≠ 1 .

Proof

To apply Theorem 4.3, we only have to verify that the action of đș on 𝒳 is strongly topological. This verification coincides with the argument of [1, Theorem 1.3]. ∎

Corollary 4.6

Let đș be a profinite weakly branch group. Let đș act on a rooted tree 𝑇 spherically transitively such that the rigid vertex stabilizers are non-trivial. Let 𝒳 be the boundary of 𝑇. Let Îł ̄ be a random 𝑡-tuple from đș and w ⁹ ( y ̄ ) an explicitly oscillating word such that O w = X . Then, almost surely, Îł ̄ satisfies the mixed inequality w ⁹ ( y ̄ ) ≠ 1 .

Proof

As đș is closed in the profinite topology of Aut ⁹ ( T ) , each stabilizer stab G ⁹ ( x ) , x ∈ X , is closed. By Example 2.11, đș hereditarily separates 𝒳. By Corollary 4.5, we have the conclusion of the corollary. ∎

4.3 Topological actions of automorphism groups

Here we will consider the situation of Section 4.2 in the context of [6]. A word over a group đș with a single variable is of the following form:

(4.1) w ⁱ ( y ) = h k ⁱ y m k ⋅ 
 ⋅ h 1 ⁱ y m 1 , where ⁱ h 1 , 
 , h k ∈ G ⁱ and ⁱ m 1 , 
 , m k ∈ Z ∖ { 0 } .

We admit the possibility that h k = 1 . According to [10, Remark 5.1], in order to show that đș is MIF, it suffices to prove that there are no laws of the form w ⁹ ( y ) = 1 , where w ⁹ ( y ) is as above. Section 6 of [6] gives a method of analysis of such words in the case of the automorphism group of some standard continuous structures. In fact, the authors detect a certain (model-theoretic) property of these structures which guarantees that any w ⁹ ( y ) of the form (4.1) is not a law. Adapting this property to the general situation of đș-spaces, we arrive at the following definition.

Definition 4.7

Assume that đș acts on an infinite set 𝑋 by permutations. We say that ( G , X ) is discerning if, for every finite A ⊂ X , every non-trivial stab G ⁹ ( A ) -orbit 𝑂 and every non-trivial g ∈ G , the intersection supp ⁹ ( g ) ∩ O is not empty.

Lemma 6.13 of [6] states the following.

Assume that ( G , X ) is discerning and g 1 , g 2 ∈ G . Then, for every finite A ⊂ X and every a ∈ X ∖ A , there is some g ∈ g 2 ⋅ stab G ⁹ ( A ) such that { g ⁹ ( a ) , g 1 ⁹ g ⁹ ( a ) } ⊂ supp ⁹ ( g 1 ) ∖ ( A âˆȘ { a } ) .

In [6], this property is applied to constants h i of the word (4.1). These h i are viewed as g 1 in the formulation.

The assumption that đș is a locally compact group with a strongly topological action on a space 𝒳 already implies a statement of this kind: for every finite A ⊂ X and every a ∈ X ∖ A , all elements 𝑔 of the coset g 2 ⋅ stab G ⁹ ( A ) almost surely have the property that { g ⁹ ( a ) , g 1 ⁹ g ⁹ ( a ) } ⊂ X ∖ ( A âˆȘ { a } ) . Indeed, since stab G ⁹ ( A âˆȘ { a } ) is of measure 0 in stab G ⁹ ( A ) , we see that so is the set

{ g ∣ g ⁹ ( a ) ∈ ( A âˆȘ { a } ) ⁹ or ⁹ g ⁹ ( a ) ∈ g 1 − 1 ⁹ ( A âˆȘ { a } ) } .

This suggests the following definition, a measurable version of the statement of [6, Lemma 6.13].

Definition 4.8

Assume that a locally compact group đș has a strongly topological action on a topological space 𝒳 and g 1 ∈ G . We say that g 1 is measure discerning (m-discerning) if, for every g 2 ∈ G , every finite A ⊂ X and every a ∈ X ∖ A , almost surely, all elements 𝑔 of the coset g 2 ⋅ stab G ⁹ ( A ) have the property that { g ⁹ ( a ) , g 1 ⁹ g ⁹ ( a ) } ⊂ supp ⁹ ( g 1 ) ∖ ( A âˆȘ { a } ) .

Note that, when a ∉ g 2 − 1 ⁹ ( A âˆȘ Fix ⁹ ( g 1 ) ) , then existence of 𝑔 as in the definition also follows from the assumption of hereditary separation.

The following proposition is a measurable version of [6, Theorem 6.10].

Proposition 4.9

Assume that a locally compact group đș has a strongly topological action on a Hausdorff topological space 𝒳. Then, for every non-trivial word w ⁹ ( y ) whose constants are m-discerning, a random element g ∈ G almost surely satisfies the inequality w ⁹ ( y ) ≠ 1 .

Proof

We adapt the proof of [6, Theorem 6.10]. Assume that w ⁱ ( y ) is in the form (4.1). Since the action is strongly topological, for a random g ∈ G , almost surely, the elements a , g ⁱ ( a ) , 
 , g m 1 − 1 ⁱ ( a ) are pairwise distinct (where, say, m 1 > 0 ). Applying m-discerning (for g 1 = h 1 and g 2 = 1 ) together with Fubini’s theorem, we see that, almost surely, the elements

a , g ⁱ ( a ) , 
 , g m 1 − 1 ⁱ ( a ) , g m 1 ⁱ ( a ) , h 1 ⁱ ( g m 1 ⁱ ( a ) )

are pairwise distinct.

Repeating the argument k − 1 more times, we obtain that, almost surely, the elements

a , g ⁱ ( a ) , 
 , g m 1 − 1 ⁱ ( a ) , g m 1 ⁱ ( a ) , h 1 ⁱ ( g m 1 ⁱ ( a ) ) , g ⁱ ( h 1 ⁱ ( g m 1 ⁱ ( a ) ) ) , 
 , h k ⁱ ( 
 ⁱ ( g m 2 ⁱ ( h 1 ⁱ ( g m 1 ⁱ ( a ) ) ) ) ⁱ 
 )

are pairwise distinct. We see that, almost surely, a random element of đș satisfies the inequality w ⁹ ( y ) ≠ 1 . ∎

When đș acts on 𝒳 strongly topologically and transitively, the space 𝒳 can be considered under the measure inherited from đș (after identification of 𝒳 with G / stab G ⁹ ( p ) ). Then note that, for each m-discerning h ∈ G , the set Fix ⁹ ( h ) is of measure 0. In particular, in typical situations where 𝒳 is not discrete and the conditions of Proposition 4.9 hold, there are non-m-discerning elements in đș. For example, this is the case of [1, Corollary 1.6] which concerns the automorphism group of a 𝑑-regular tree ( d > 2 ) acting on the boundary of the tree. On the other hand, the following general statement holds.

Assume that a locally compact group đș has a strongly topological action on a Hausdorff topological space 𝒳. If every non-trivial element of đș is m-discerning, then the group đș is MIF.

Indeed, by [10, Remark 5.1], a group is MIF exactly when it does not have laws with constants depending on a single variable. It remains to apply Proposition 4.9.

The discussion above suggests that the condition that all non-trivial elements of đș are m-discerning is very restrictive. The authors do not know any interesting example of this kind (in particular, when 𝒳 is not discrete and the stabilizers of finite subsets are non-trivial).

Acknowledgements

The authors are grateful to the referee for corrections and for helpful and stimulating remarks.

  1. Communicated by: Rachel Skipper

References

[1] M. AbĂ©rt, Group laws and free subgroups in topological groups, Bull. Lond. Math. Soc. 37 (2005), no. 4, 525–534. 10.1112/S002460930500425XSuche in Google Scholar

[2] V. S. Anashin, Mixed identities and mixed varieties of groups (in Russian), Mat. Sb. (N. S.) 129(171) (1986), no. 2, 163–174, 303. Suche in Google Scholar

[3] V. S. Anaơin, Mixed identities in groups (in Russian), Mat. Zametki 24 (1978), no. 1, 19–30, 141. Suche in Google Scholar

[4] L. Bartholdi, R. I. Grigorchuk and Z. Ć uniáž±, Branch groups, Handbook of Algebra. Vol. 3, Elsevier/North-Holland, Amsterdam (2003), 989–1112. 10.1016/S1570-7954(03)80078-5Suche in Google Scholar

[5] J. W. Cannon, W. J. Floyd and W. R. Parry, Introductory notes on Richard Thompson’s groups, Enseign. Math. (2) 42 (1996), no. 3–4, 215–256. Suche in Google Scholar

[6] M. Etedadialiabadi, S. Gao, F. Le MaĂźtre and J. Melleray, Dense locally finite subgroups of automorphism groups of ultraextensive spaces, Adv. Math. 391 (2021), Paper No. 107966. 10.1016/j.aim.2021.107966Suche in Google Scholar

[7] I. Z. Golubchik and A. V. MikhalĂ«v, Generalized group identities in classical groups (in Russian), Zap. Nauchn. Sem. Leningrad. Otdel. Mat. Inst. Steklov. (LOMI) 114 (1982), 96–119, 219. Suche in Google Scholar

[8] N. L. Gordeev, Freedom in conjugacy classes of simple algebraic groups and identities with constants (in Russian), Algebra i Analiz 9 (1997), no. 4, 63–78. Suche in Google Scholar

[9] R. I. Grigorchuk, Just infinite branch groups, New Horizons in Pro-𝑝 Groups, Progr. Math. 184, BirkhĂ€user, Boston (2000), 121–179. 10.1007/978-1-4612-1380-2_4Suche in Google Scholar

[10] M. Hull and D. Osin, Transitivity degrees of countable groups and acylindrical hyperbolicity, Israel J. Math. 216 (2016), no. 1, 307–353. 10.1007/s11856-016-1411-9Suche in Google Scholar

[11] B. Jacobson, A mixed identity-free elementary amenable group, Comm. Algebra 49 (2021), no. 1, 235–241. 10.1080/00927872.2020.1797073Suche in Google Scholar

[12] M. Kassabov and F. Matucci, The simultaneous conjugacy problem in groups of piecewise linear functions, Groups Geom. Dyn. 6 (2012), no. 2, 279–315. 10.4171/ggd/158Suche in Google Scholar

[13] A. S. Kechris, Classical Descriptive Set Theory, Grad. Texts in Math. 156, Springer, New York, 2012. Suche in Google Scholar

[14] O. Kharlampovich and A. Myasnikov, Irreducible affine varieties over a free group. I. Irreducibility of quadratic equations and Nullstellensatz, J. Algebra 200 (1998), no. 2, 472–516. 10.1006/jabr.1997.7183Suche in Google Scholar

[15] O. Kharlampovich and A. Myasnikov, Irreducible affine varieties over a free group. II. Systems in triangular quasi-quadratic form and description of residually free groups, J. Algebra 200 (1998), no. 2, 517–570. 10.1006/jabr.1997.7184Suche in Google Scholar

[16] V. V. Nesterov and A. V. Stepanov, An identity with constants in a Chevalley group of type F 4 , Algebra i Analiz 21 (2009), no. 5, 196–202. 10.1090/S1061-0022-2010-01119-1Suche in Google Scholar

[17] Z. Sela, Diophantine geometry over groups. I. Makanin–Razborov diagrams, Publ. Math. Inst. Hautes Études Sci. 93 (2001), 31–105. 10.1007/s10240-001-8188-ySuche in Google Scholar

[18] P. SƂanina and R. Zarzycki, Laws with constants in Thompson’s group đč, J. Group Theory 22 (2019), no. 5, 783–793. 10.1515/jgth-2018-0134Suche in Google Scholar

[19] A. Stepanov, Free product subgroups between Chevalley groups G ⁱ ( Ω , F ) and G ⁱ ( Ω , F ⁱ [ t ] ) , J. Algebra 324 (2010), no. 7, 1549–1557. Suche in Google Scholar

[20] G. M. Tomanov, Generalized group identities in linear groups, Mat. Sb. (N. S.) 123(165) (1984), no. 1, 35–49. Suche in Google Scholar

[21] R. Zarzycki, Limit groups with respect to Thompson’s group đč and other finitely generated groups, Ph.D. Thesis, WrocƂaw, 2009. Suche in Google Scholar

[22] R. Zarzycki, Limits of Thompson’s group đč, Combinatorial and Geometric Group Theory, Trends Math., BirkhĂ€user/Springer, Basel (2010), 307–315. 10.1007/978-3-7643-9911-5_14Suche in Google Scholar

Received: 2023-02-19
Revised: 2024-10-23
Published Online: 2024-12-05
Published in Print: 2025-05-01

© 2024 Walter de Gruyter GmbH, Berlin/Boston

Heruntergeladen am 30.12.2025 von https://www.degruyterbrill.com/document/doi/10.1515/jgth-2023-0028/html
Button zum nach oben scrollen