2. Towers and exemplary towers: review of Collin Bleak’s results

In the proof of Theorem 1.2, as a crucial tool, we use the result of C. Bleak on the existence of arbitrarily high towers in a non-solvable subgroup of $PL_o(I)$ , [Reference Bleak7]. First, we would like to introduce the following notions essentially borrowed from [Reference Bleak7], with a slightly different terminology.

Definition 2.2. We will say that a strict tower

\begin{align*}T = [(f_1, \ldots, f_n);\; (a_1,b_1), \ldots, (a_n, b_n)]\end{align*}

is suitable if for any nonzero integer $p$ and for all $1\leq i \lt j \leq n$ , the following condition holds: $ f_i^p([a_j,b_j])\cap \displaystyle \mathop{\cup }_{i+1\leq k\leq n} supp (f_k) = \emptyset \ (1)$ .

To explain the existence of exemplary towers, we would like to make a digression into some of the results of C. Bleak. For the rest of this section, let $\Gamma \leq PL_o(I)$ . The following notions and notations are all borrowed directly from [Reference Bleak7].

We will call the convex hull of a point in $I$ under the action of $\Gamma$ an orbital of $\Gamma$ , if this convex hull contains more than one point. We note that the orbitals are open intervals. If $g\in \Gamma$ , we will refer to an orbital of the group $\langle g \rangle$ as an orbital of $g$ . If an open interval $A$ is an orbital of $g$ , then the pair $(A,g)$ will be called a signed orbital of $G$ . $g$ will be called the signature of the signed orbital $(A,g)$ .

Given a set $Y$ of signed orbitals of $G$ , the symbol $S_Y$ will refer to the set of signatures of the signed orbitals in $Y$ . Similarly, the symbol $O_Y$ will refer to the set of orbitals of the signed orbitals of $Y$ . We note that the set of signed orbitals of $PL_o(I)$ is a partially ordered set under the lexicographic order induced from the partial order on subsets of $I$ (induced by inclusion) in the first coordinate, and the left total order of the elements of $PL_o(I)$ in the second coordinate.

A tower $T$ of $G$ is a set of signed orbitals which satisfies the following two criteria.

1. 1. $T$ is a chain in the partial order on the signed orbitals of $G$ .

2. 2. For any $A\in O_T$ , $T$ has exactly one element of the form $(A,g)$ .

Given a tower $T$ of $G$ , if $(A,g), (B,h)\in T$ then one of $A\subseteq B$ and $B\subseteq A$ holds, with equality occurring only if $g = h$ as well. Therefore, one can visualize the tower as a stack of nested levels that are always getting wider as one goes “up” the stack.

The cardinality of the set $O_T$ will be called the height of the tower $T$ . Besides the cardinality, we also want to make use of the order structure of towers which allows to define the following more sensitive notions: if there is an order-preserving injection from ${\mathbb{N}}$ to $T$ , then we will say $T$ is tall, and if there is an order-preserving injection from $-{\mathbb{N}}$ to $T$ , then we will say $T$ is deep. If $T$ is both deep and tall, then we will say $T$ is a bi-infinite tower; in the latter case, there will be an order-preserving injection from ${\mathbb{Z}}$ to $T$ .

A major result of [Reference Bleak7] is the following beautiful geometric characterization of solvable subgroups of $PL_o(I)$ .

If $G$ admits two signed orbitals $(A,g)$ and $(B,h)$ so that $A = (a_1,a_2)$ and $B = (b_1, b_2)$ , with $a_1 \lt b_1 \lt a_2 \lt b_2$ then we will say that $G$ admits a transition chain of length two. One can similarly define transition chains of any (finite) length, but we will have no need for that generality here.

If $A = (a,b)$ is an orbital of the group $G$ , and $G$ has an element $g$ which has an orbital $B = (c,d)$ so that either $c = a$ or $d = b$ , then we say that $g$ has an orbital that shares an end with $A$ .

Given an orbital $A$ of $H$ , we say that $h$ realizes an end of $A$ if some orbital of $h$ lies entirely in $A$ and shares an end with $A$ . If $g$ and $h$ are elements of $PL_o(I)$ and there is an interval $B = (a,b)\subset I$ so that both $g$ and $h$ have $B$ as an orbital, then we will say that $g$ and $h$ share the orbital $B$ .

We will say that an orbital $A$ of a group $H\leq PL_o(I)$ is imbalanced if some element $h\in H$ realizes one end of $A$ , but not the other, and we will say $A$ is balanced if whenever an element $h\in H$ realizes one end of $A$ , then $h$ also realizes the other end of $A$ (note that $h$ might do this with two distinct orbitals). A subgroup $H\leq PL_o(I)$ will be called balanced if given any subgroup $G\leq H$ , and any orbital $A$ of $G$ , every element of $G$ which realizes one end of $A$ also realizes the other end of $A$ . In the case where $H$ has a subgroup $G$ with an imbalanced orbital, then we will say that $H$ is imbalanced.

We say a tower $T$ is an exemplary tower if the following two additional properties hold:

1. 1. Whenever $(A,g), (B,h)\in T$ then $(A,g)\leq (B,h)$ implies the orbitals of $g$ are disjoint from both ends of the orbital $B$ .

2. 2. Whenever $(A, g), (B, h)\in T$ then $(A,g)\leq (B,h)$ implies no orbital of $g$ in $B$ shares an end with $B$ .

C. Bleak proves several technical results which indicate the plethora of exemplary towers in $PL_o(I)$ . The following two lemmas are stated in [Reference Bleak7] as Lemmas 1.4 and 2.12, respectively.

Thus, the existence of a transition chain of length two implies the existence of infinite towers, and the absence of the transition chain of length two implies that all towers are nice, that is, exemplary. It turns out the absence of a transition chain of length two also implies certain nice properties of the group itself. The following result is stated as Remark 4.9 in [Reference Bleak7].

We also need the following two technical results from [Reference Bleak7].

Combining the above results we can now claim the following lemma.

Proof. If $\Gamma$ does not admit a transition chain of length two, then the claim follows from Theorem 2.4 and Lemma 2.6. If $\Gamma$ is not balanced then the claim follows from Corollary 2.10. Thus, we can assume that $\Gamma$ is balanced and admits a transition chain of length two. Then by Lemma 2.5, $\Gamma$ admits an infinite tower. Then by Lemma 2.8, it admits an exemplary bi-infinite tower.

3. Proof of Theorem 1.2

For any natural number $r$ , let $G_r = (\ldots ((\mathbb{Z}\wr \mathbb{Z})\wr \mathbb{Z})\wr \ldots \mathbb{Z})\wr \mathbb{Z}$ where the iterated wreath product is taken $r$ times. The group $G_r$ can be defined inductively as $G_0 = 1, G_{i+1} = G_i\wr \mathbb{Z} \;:\!=\; \mathbb{Z}\ltimes \displaystyle \mathop{\oplus }_{n\in \mathbb{Z}}G_i, 0\leq i\leq r-1$ . In the wreath product $G_i\wr \mathbb{Z}$ , the standard generator of the acting group $\mathbb{Z}$ will be denoted by $g_{r-i}$ . (in [Reference Bleak7], the group $G_r$ is denoted by $W_r$ ).

The following lemma will be useful; the idea of its proof is essentially borrowed from the proof of Lemma 2.3 in [Reference Akhmedov1].

Proof. Since a free group on two generators contains a free group on $k$ generators for any $k\geq 3$ , it is sufficient to prove the claim for $k=2$ . We will do this by induction; more precisely, it suffices to prove the following claim: if $A, B$ are nontrivial groups and $A$ satisfies no law in two variables of length less than $n\geq 4$ , then the wreath product $B\wr A = A\ltimes \displaystyle \mathop{\oplus }_{i\in A}B$ satisfies no law in two variables of length less than $n+1$ .

Indeed, let $w_1, w_2\in A$ with no relation of length less than $n$ . An element of $\displaystyle \mathop{\oplus }_{i\in A}B$ can be written as $(x_g)_{g\in A}$ where all but finitely many “coordinates” are 1. Let $b$ be a non-identity element of $B$ , and $t = (x_g)_{g\in A}$ be the element of $\displaystyle \mathop{\oplus }_{i\in A}B$ where $x_g = b$ for $g=1$ and $x_g = 1$ otherwise.

Then there is no relation of length less than $n+1$ among the elements $tw_1$ and $w_2$ . Indeed, let $W(x,y)$ be a nontrivial reduced word of length $k \lt n+1$ in the alphabet $\{x^{\pm 1}, y^{\pm 1}\}$ such that $W(tw_1,w_2) = 1\in B\wr A$ . For every $1\leq i\leq k$ , let also $W_i$ be the prefix of $W$ of length $i$ . Then $W(w_1,w_2) = W_k(w_1,w_2) = 1\in A$ , moreover, for at least one $i\in \{1,\ldots, k\}$ there exists $j\in \{1,\ldots, k\}\backslash \{i\}$ such that $W_i(w_1,w_2) = W_j(w_1, w_2)$ . Indeed, we have

\begin{align*}W(tw_1, w_2) = \left ( \displaystyle \mathop {\Pi }_{i=1}^s(W_{n_i}t^{\epsilon _i}W_{n_i}^{-1}) \right ) W(w_1,w_2)\end{align*}

where $1\leq s\leq k, \epsilon _i\in \{-1,1\}, 1\leq i\leq s$ and $0\leq n_1 \lt n_2 \lt \dots \lt n_s\leq k$ (here, we define $W_0 = 1$ ).

Now, we have $W_{n_i}t^{\epsilon _i}W_{n_i}^{-1}\in \displaystyle \mathop{\bigoplus }_{\alpha \in A}B$ for all $1\leq i\leq s$ . Moreover, by the re-indexing action of $A$ on $\displaystyle \mathop{\bigoplus }_{\alpha \in A}B$ , for each of the terms $W_{n_i}t^{\epsilon _i}W_{n_i}^{-1}$ we have only one nontrivial (i.e. non-identity) coordinate consisting of $b$ or $b^{-1}$ depending if $\epsilon _i = 1$ or $\epsilon _i = -1$ respectively. Hence, to have a necessary cancelation, we need to have two distinct $p,q\in \{1,\dots, s\}$ such that $W_{n_p} = W_{n_q}$ (and $\epsilon _p = -\epsilon _q$ ). Thus, we can take $i = n_p, j = n_q$ .

Thus, we established that for at least one $i\in \{1,\ldots, k\}$ there exists $j\in \{1,\ldots, k\}\backslash \{i\}$ such that $W_i(w_1,w_2) = W_j(w_1, w_2)$ . But this implies a relation of length less than $n$ among $w_1$ and $w_2$ .

Let us note that in the proof of Lemma 3.1, we are using a careful re-examination of the word $W(tw_1, w_2)$ . We view the latter as a word in the alphabet $\{t^{\pm 1}, w_1^{\pm 1}, w_2^{\pm 1}\}$ . There will be no consecutive occurrences of $t^{\pm 1}$ ; as we read through from left to right, we write $W(tw_1, w_2)$ as a product of conjugates of $t^{\pm 1}$ . The residue at the very right will be the word $W(w_1, w_2)$ . For example, if $W(x,y) = [x,y] = xyx^{-1}y^{-1}$ , then we will have

\begin{align*}W(tw_1,w_2) = [tw_1,w_2] = tw_1w_2w_1^{-1}t^{-1}w_2^{-1} \end{align*}

\begin{align*}= (1t1)(w_1w_2w_1^{-1}t^{-1}w_1w_2^{-1}w_1^{-1})[w_1,w_2].\end{align*}

We also would like to observe the following simple lemma.

Now, let $\Gamma$ be a finitely generated non-solvable subgroup of $PL_o(I)$ , $s = d(\Gamma )$ (i.e. $s$ is the minimal cardinality for a generating subset of $\Gamma$ ). Without loss of generality, we may assume that $\Gamma$ is irreducible, that is, it has no global fixed point in the interval $(0,1)$ .

For every finite generating set $X$ of $\Gamma$ , we will fix the left-invariant Cayley metric on $\Gamma$ with respect to $X$ , and let $B_k(1;\; X)$ denote the ball of radius $k$ centered at identity element $1\in \Gamma$ , for all $k\geq 1$ . We also let

\begin{align*}S_k(\Gamma, X, c) = \{\gamma \in B_k(1;\; X) \ | \ \gamma '(0) = c\}, \ S_k(\Gamma, X) = \{\gamma \in B_k(1;\; X) \ | \ \gamma '(0) = 1\},\end{align*}

\begin{align*}C_k(\Gamma, X) = \{c\in \mathbb {R}_{+} \ | \ S_k(\Gamma, X, c) \neq \emptyset \}.\end{align*}

Fix a positive integer $m$ , and let $q = 2m^2$ . By Lemma 3.1, there exists $r\in \mathbb{N}$ such that in the group $G_r$ there exist $s$ elements $w_1, w_2, \ldots, w_s$ such that there is no relation of length less than $q$ among $w_1, \ldots, w_s$ . Let $g_1, \ldots, g_r$ be the standard generators of $G_r$ and let $w_i = W_i(g_1, \ldots, g_r)$ , $1\leq i\leq s$ , where $W_i$ is a reduced word in the free group of rank $r$ formally generated by the letters $g_1, \ldots, g_r$ .

Since $\Gamma$ is non-solvable, the commutator subgroup $[\Gamma, \Gamma ]$ is non-solvable. Then by Remark 2.3 and Lemma 2.11, there exists an ordered $(r+1)$ -tuple $(f_1, \ldots, f_r, f_{r+1})$ of elements of $[\Gamma, \Gamma ]$ which form a suitable tower of height $r+1$ . Let $0 \lt d \lt D \lt 1$ such that $\displaystyle \mathop{\cup }_{1\leq i\leq r+1}supp(f_i)\subset (d,D)$ .

Then, we can find $\epsilon _0 \gt 0$ , a finite generating set $S$ of $\Gamma$ of cardinality $s$ , and a suitable tower $(h_1, \ldots, h_{r}, h_{r+1})$ of elements of $\Gamma$ with bases $(a_i, b_i)\subset (0, \epsilon _0), 1\leq i\leq r+1$ and $\Omega = \displaystyle \mathop{\cup }_{1\leq i\leq r+1}supp(h_i)$ such that $S\cap S^{-1} = \emptyset$ and the following conditions hold:

1. (i) for all $\beta \in B_{2m}(1;\; S), \beta (\Omega )\subset (0, \epsilon _0)$ ;

2. (ii) for any two distinct $c_1, c_2 \in \mathbb{R}_{+}$ , and for all $\beta _1\in S_{2m}(\Gamma, S, c_1), \beta _2\in S_{2m}(\Gamma, S, c_2)$ , $\beta _1(\Omega )\cap \beta _2(\Omega ) = \emptyset$ ;

3. (iii) for all $c\in \mathbb{R}_{+}, \beta _1, \beta _2\in S_{2m}(\Gamma, S, c)$ , and for all $x\in \Omega$ , $\beta _1(x) = \beta _2(x)$ (so, in particular, for all $\beta \in S_{2m}(\Gamma, S)$ and for all $x\in \Omega, \ \beta (x) = x$ ).

Indeed, let $X_0 = \{\alpha _1, \dots, \alpha _s\}$ be any generating set of $\Gamma$ of cardinality $s$ . Without loss of generality, we may assume that

\begin{align*}1\leq \alpha^{\prime} _1(0) \leq \dots \leq \alpha^{\prime}_s(0) \ \mathrm {and} \ \alpha^{\prime}_s(0) \gt 1.\end{align*}

Let $\delta \gt 0$ such that $\alpha _s$ has no singularity in $(0,\delta )$ . By irreducibility, there exists $\phi \in \Gamma$ such that $\phi (D) \lt \delta$ . For $n\geq 1$ and $1\leq i\leq r+1$ , we let $f_i^{(n)} = \psi _n^{-1} f_i\psi _n$ , where $\psi _n = \phi ^{-1}\alpha _s^n$ . Let also $(d_n, D_n) = \psi _n^{-1}((d,D)), n\geq 1$ . Notice that the interval $(d_n, D_n)$ (in particular, its subset $ \mathop{\cup }_{1\leq i\leq r+1}supp(f_i^{(n)})$ ) converges to zero as $n\to \infty$ ; moreover, there exists a positive integer $p$ such that for all $k\geq p$ , we have $\frac{D_k}{d_k} = \frac{D_p}{d_p}$ .

We can choose a finite generating set $S = \{\gamma _1, \dots, \gamma _s\}$ such that the inequality $\min \{c\in C_{2m}(\Gamma, S) \ | \ c \gt 1\} \gt \frac{D_p}{d_p}$ holds. To see this, first, we let $X_1 = \{\beta _1, \dots, \beta _s\}$ where $\beta _s = \alpha _s$ , and $\beta _i = \alpha _i\alpha _{s}^{n_i}, 1\leq i\leq s-1$ for some $n_{s-1}\leq \dots \leq n_2\leq n_1$ such that $1 \lt \beta^{\prime}_s(0) \leq \beta^{\prime}_{s-1}(0) \leq \dots \leq \beta _1'(0)$ and $\beta _{1}'(0) \gt \frac{D_p}{d_p}$ . After this, we modify the generating set $X_1$ further by letting $S = \{\beta _1, \beta _1^{\lambda }\beta _2, \beta _1^{\lambda ^2}\beta _{3}, \dots, \beta _1^{\lambda ^{s-1}}\beta _{s}\}$ where $\lambda = 4m+1$ .

Then, for sufficiently big $n$ , we can take $h_i = f_i^{(n)}, 1\leq i\leq r+1$ to satisfy the claims (i)–(iii).

Now, let $A_m$ be a minimal subset of $B_m(1;S)$ such that $A_m\cap S_m(\Gamma, S, c) \neq \emptyset$ for every $c\in C_m(\Gamma, S)$ ; and for all $k\in \{1, \ldots, s\}$ , let

\begin{align*}\omega _k = \displaystyle \prod _{\gamma \in A_m}(\gamma v_k\gamma ^{-1}), \ \eta _k = \displaystyle \prod _{\gamma \in A_m}(\gamma u_k\gamma ^{-1})\end{align*}

where $v_k = W_k(h_1, \ldots, h_r), u_k = v_k^m$ and for the products (the formulas for $\omega _k$ and $\eta _k$ ) we choose any linear order on the set $A_m$ .

Notice that because of conditions (i)–(iii), for any two $\gamma ', \gamma ''\in B_m(1)$ , we have $[\gamma 'v_k(\gamma ')^{-1}, \gamma ''v_k(\gamma '')^{-1}] = 1$ and $[\gamma 'u_k(\gamma ')^{-1}, \gamma ''u_k(\gamma '')^{-1}] = 1$ , so the order in the products $\displaystyle \prod _{\gamma \in A_m}(\gamma v_k\gamma ^{-1})$ and $\displaystyle \prod _{\gamma \in A_m}(\gamma u_k\gamma ^{-1})$ does not matter.

Now, let $S^{(m)} = \{\eta _1\gamma _1\eta _2, \eta _2\gamma _2 \eta _3, \ldots, \eta _s\gamma _s\eta _1, \omega _1, \ldots, \omega _s\}$ . Then, since $\omega _k^m = \eta _k$ for all $1\leq k\leq s$ , the set $S^{(m)}$ generates $\Gamma$ , and there is no relation of length less than $m$ among the elements of $S^{(m)}$ .

Indeed, let $R = R(\eta _1\gamma _1\eta _2, \eta _2\gamma _2 \eta _3, \ldots, \eta _s\gamma _s\eta _1, \omega _1, \ldots, \omega _s)$ denote such a relation. Then, we can write

\begin{align*}R(\eta _1\gamma _1\eta _2, \eta _2\gamma _2 \eta _3, \ldots, \eta _s\gamma _s\eta _1, \omega _1, \ldots, \omega _s) = R_0(\theta _1R_1\theta _1^{-1})(\theta _2R_2\theta _2^{-1})\ldots (\theta _nR_n\theta _n^{-1})R_{n+1}\end{align*}

where $n\leq m$ , $\theta _i\in B_m(1)$ for all $1\leq i\leq n$ , and $R_j = R_j(\eta _1, \ldots, \eta _s, \omega _1, \ldots, \omega _s)$ is a reduced word of length at most $m$ for all $0\leq j\leq n+1$ ; moreover, $R_1, \ldots, R_n$ are nontrivial.

Notice that for all $g\in B_m(1;\; S)$ , the shift $gB_m(1;\; S)$ still contains $1\in \Gamma$ . Then we obtain a nontrivial relation $V(v_1, \ldots, v_s)$ among $v_1, \ldots, v_s$ of length at most $2m^2$ .Footnote ² We can write $V(v_1, \ldots, v_s) = W(h_1, \ldots, h_r) = 1$ where $W$ is a nontrivial reduced word. Notice that $V(v_1, \ldots, v_s) = W(h_1, \ldots, h_r)$ represents a map in $PL_o(I)$ , while $V(w_1,\ldots, w_s)$ represents a word in $G_r$ which by our choices does not represent an identity element in $G_r$ . Then, by Lemma 3.2, for some word $U = U(h_1, \ldots, h_r)$ and for all $x\in U((a_{r+1}, b_{r+1}))$ we have $W(x)\notin U((a_{r+1}, b_{r+1}))$ , thus $W(x)\neq x$ . Since $m$ is arbitrary, we conclude that $girth(\Gamma ) = \infty$ .

The main idea of the proof of Theorem 1.2 is described below. For any given $q, k$ , in a suitable tower of sufficiently big height $r$ , formed by PL-maps $\phi _1, \ldots, \phi _r, \phi _{r+1}$ , one can find words $w_1, \ldots, w_k$ in the alphabet $\{\phi _1^{\pm 1}, \ldots, \phi _r^{\pm 1}\}$ such that the corresponding elements (let us denote them by $\overline{w_1}, \ldots, \overline{w_k}$ ) do not have a relation of length less than $q$ . This is because, upon the action on the innermost base of the tower (an orbital of $\phi _{r+1}$ ), a suitable tower behaves as if the maps generate a genuine copy of $G_r$ , for elements in the ball of certain radius. The problem is how to find a finite generating set without a short relation among the generators, not just among some elements. For this, we pick up a tower with sufficient height such that the PL homeomorphisms forming this tower are supported in a very small interval. This interval can be made arbitrarily small; therefore, using irreducibility of $\Gamma$ , one can push this support (interval) close enough to the end 0 of the interval $[0,1]$ such that the new support $\Omega$ satisfies conditions (i)–(iii), namely, (i) any PL map $\beta$ from the ball $B_m(1)$ still keeps $\Omega$ inside an interval $(0,\epsilon _0)$ ; (ii) the image of $\Omega$ by elements of $B_m(1)$ with different slopes at 0 are disjoint; and (iii) the images of $\Omega$ by elements of $B_m(1)$ with the same slope at 0 are the same. Then we pick up a generating set which involves the elements $\overline{w_1}, \ldots, \overline{w_k}$ . By properties (i)–(iii), we again obtain a short relation among $\overline{w_1}, \ldots, \overline{w_k}$ thus a contradiction.

4. Girth of subgroups with hyperbolic-like elements

In this section, we will prove Theorem 1.5.

Let $d(\Gamma ) = s$ , and $m$ be a natural number. We will find $s+2$ generators of $\Gamma$ such that there is no relation of length $m$ or less in $\Gamma$ in these generators. (since $m$ is arbitrary, this proves that $girth(\Gamma ) = \infty$ ).

Let $S = \{X_1, \ldots, X_s\}$ be a finite generating set of $\Gamma$ , and $S^{\ast }$ be the symmetrization of $S$ , that is $S^{\ast } = \{X_1, \ldots, X_s, X_1^{-1}, \ldots, X_s^{-1}\}$ . Let also $p_0\in (0,1)$ (one could take $p_0 = \frac{1}{2}$ ). We can find a natural number $N \gt 4m$ and a sequence $0 = c_0 \lt c_1 \lt c_2 \lt \ldots \lt c_{2N} \lt c_{2N+1} = 1$ such that the following three conditions are satisfied.

1. (i) $p_0\in (c_{N}, c_{N+1})$ ;

2. (ii) for all $X\in S^{\ast }$ and $p\in \{p_0, c_1, \dots, c_{2N}\}, X(p)\notin \{c_1, \ldots, c_{2N}\}\backslash \{p\}$ ;

3. (iii) $W(X_1, \ldots, X_s, Y_N)(p_0)\subset (c_1, c_{2N})$ for all reduced words $W$ of the form $W_0Y_N^{n_1}W_1Y_N^{n_2}\ldots W_{k-1} Y_N^{n_k}W_k$ where $Y_N$ is any orientation-preserving homeomorphism satisfying the conditions $Y_N(c_i) = c_{i+4}, 1\leq i\leq 2N-4$ , $n_1, \ldots, n_k\in \{-1, 1\}$ , and $W_i$ is a reduced word in the alphabet $\{X_1^{\pm 1}, \ldots, X_s^{\pm 1}\}$ of length $L_i, 0\leq i\leq k$ where $\displaystyle \sum _{i=0}^kL_i \leq 2m$ .

Notice that condition (iii) implies that

\begin{align*}W(X_1, \dots, X_s, Y_n)(p_0) \in (c_{1+4d}, c_{2N-4d})\end{align*}

where $d = 2m - \displaystyle \mathop{\sum }_{i=0}^{k}L_i$ .

Let $\delta = \displaystyle \min _{0\leq i\leq 2N} |c_{i+1}-c_i|$ and $\epsilon \lt \frac{1}{8}\min \{\delta, |p_0-c_N|, |p_0-c_{N+1}|\}$ be such that $X(p-\epsilon, p+\epsilon )\cap (q-\epsilon, q+\epsilon ) = \emptyset$ for all $X\in S^{\ast }$ and distinct $p, q\in \{p_0, c_1, \dots, c_{2N}\}$ . Then we can find a natural number $M \gt 2m$ and elements $\gamma, \theta \in \Gamma$ such that:

1. (iv) $Fix(\gamma ) = \{0, a_1, a_2, \ldots, a_N, 1\}, Fix(\theta ) = \{0, b_1, b_2, \ldots, b_N, 1\}$ and for all $1\leq i\leq N$ , the inequalities $|a_i-c_{2i-1}| \lt \epsilon$ and $|b_i-c_{2i}| \lt \epsilon$ hold;

2. (v) for all $n\geq M$ , we have $\gamma ^{\pm n}(U_{\gamma })\subset V_{\gamma }$ where

   \begin{align*}U_{\gamma } = \displaystyle \mathop {\sqcup } _{0\leq i\leq N}(a_i + \epsilon, a_{i+1} - \epsilon ), V_{\gamma } = \displaystyle \mathop {\sqcup } _{0\leq i\leq N+1}(a_i - \epsilon, a_i + \epsilon );\end{align*}

3. (vi) for all $n\geq M$ , we have $\theta ^{\pm n}(U_{\theta })\subset V_{\theta }$ where

   \begin{align*}U_{\theta } = \displaystyle \mathop {\sqcup } _{0\leq i\leq N}(b_i + \epsilon, b_{i+1} - \epsilon ), V_{\theta } = \displaystyle \mathop {\sqcup } _{0\leq i\leq N+1}(b_i - \epsilon, b_i + \epsilon ).\end{align*}

It is straightforward to make all these arrangements. (Let us also clarify that we define $a_0 = b_0 = 0$ and $a_{N+1} = b_{N+1} = 1$ .)

Notice that $p_0\in U_{\gamma } \cup U_{\theta }$ and $p_0\notin V_{\gamma } \cup V_{\theta }$ . Now, let $r \geq 2M$ , and $S' = \{\gamma ^r, \theta ^r, \gamma ^{r^2}X_1\theta ^{r^2}, \ldots, \gamma ^{sr^2}X_s\theta ^{sr^2} \}$ .

Then if $W_0$ is a nontrivial reduced word in these generators of length at most $m$ , and if $W'$ is any suffix of $W_0$ in the alphabet $S'$ , then because of (i)–(vi), we have $W'(p_0)\in (c_1, c_N)$ .

Indeed, $W'$ can be written as:

\begin{align*}W' = \alpha _1X_{j_1}^{\epsilon _1}\alpha _2X_{j_2}^{\epsilon _2}\dots \alpha _kX_{j_k}^{\epsilon _k}\alpha _{k+1}\end{align*}

where for all $1\leq i\leq k$ , $j_i\in \{1,\dots,s\}, \epsilon _i\in \{-1,1\}$ and $\alpha _i, 1\leq i\leq k+1$ belong to the set:

\begin{align*}\{\gamma ^n \ : \ |n| \gt M \}\cup \{\theta ^n \ : \ |n| \gt M \}\cup \{\gamma ^m\theta ^n \ : \ |m|, |n| \gt M \}\cup \{\theta ^m\gamma ^n \ : \ |m|, |n| \gt M \}.\end{align*}

From the re-writing $W'$ as a suffix of $W_0$ , we obtain that $k\leq m$ . On the other hand, for all $1\leq i\leq k, \epsilon \in \{-1,1\}$ and $x\in (c_4, c_{N-4})$ we have the inequality $Y_N^{-1}(x) \lt \alpha _i^{\epsilon }(x) \lt Y_N(x)$ . Then letting $W_i = X_{n_i}^{\epsilon _i}, 1\leq i\leq k$ , we obtain that $|W_i|= 1, 1\leq i\leq k$ in the alphabet $S^{\ast }$ , hence $ \displaystyle \mathop{\sum }_{i=1}^{k}|W_i| = k \leq m \lt \frac{N}{4}$ .

By condition (iii), for every suffix $W''$ of $W'$ , we have $W''(p_0)\in (c_1, c_{2N})$ , in fact, $W''(p_0)\in (c_{1+4d}, c_{2N-4d})$ where $d = 2m-L$ and $L$ is the number of occurrences of $X_i^{\pm 1}, 1\leq i\leq s$ in $W''$ . So, we have $W'(p_0)\in (c_{1+4d}, c_{2N-4d})$ with $d = 2m-k$ .

Then, inductively (ping-pong argument), $W_0(p_0)\in V_{\gamma }\cup V_{\theta }$ ; more precisely, $W_0(p_0)\in V_{\gamma }$ if $W_0$ starts with $\gamma ^{\pm 1}$ and $W_0(p_0)\in V_{\theta }$ if $W_0$ starts with $\theta ^{\pm 1}$ . Then, $W_0(p_0)\neq p_0$ , hence $W_0\neq 1$ .

The proof of Theorem 1.5 uses the well-known “ping-pong idea” where one designs a certain “ping-pong table” which could involve two or more sets (in our case $U_{\gamma }$ and $U_{\theta }$ ), and (in one of the versions) the point (in our case $p_0$ ) taken outside of these sets arrives to one of these sets and then either jumps from one to another never returning back to its original position or in a slightly more subtle version as in our case, even if it leaves $U_{\gamma }\cup U_{\theta }$ (in our case this might happen by the action of $X\in S^{\ast }$ ), then immediately returns back to it. In condition (i) of our proof, $p_0$ is taken in the middle interval just for convenience. We are following the orbit of it by the consecutive suffixes of $W_0$ . Notice that by our choice of $r$ , there will not be any occurrence of $X_i^{\pm 1}X_j^{\pm 1}$ in $W_0$ ; moreover, the exponents of both $\gamma$ and $\theta$ are sufficiently big. By condition (ii) and the choice of $\epsilon$ , our point stays away from $Fix (\gamma )\cup Fix(\theta )$ whenever $X\in S^{\ast }$ is applied. It may get close to $Fix (\gamma )$ , then hits $U_{\gamma }$ and is immediately taken away from it by $X\in S^{\ast }$ or by sufficiently big powers of $\theta$ . Similar effect happens when we get close to $Fix(\theta )$ . Condition (iii) guarantees that we stay in the field of action where arrangements are suitable.

We would like to give a precise definition of a hyperbolic-like element.

The proof of the following theorem utilizes the main idea of the proof of Theorem 1.5.

Proof. Without loss of generality, we may assume that $1\notin S$ . Since $X$ is Hausdorff, the attractive and repelling points $p_a, p_r$ in condition (iii) are unique. By condition (ii), we also have $p_a, p_r\notin (S^{\star }\backslash \{1\})\Omega _k, 0\leq k\leq m-1$ . Then we can claim that there exists a natural number $M$ such that for all $0\leq k\leq m-1$ , there exist unique and distinct $p_a^{(k)}, p_r^{(k)}\in \Omega _{k+1}$ and disjoint open neighborhoods $U_{p_a^{(k)}}, U_{p_r^{(k)}}$ of $p_a^{(k)}$ and $p_r^{(k)}$ , respectively, such that for all $n \gt M$ and for all $x\in (S^{\star }\backslash \{1\})\Omega _k$ , $\gamma ^{n}(x)\in U_{p_a^{(k)}}, \gamma ^{-n}(x)\in U_{p_r^{(k)}}$ ; moreover, $(U_{p_a^{(k)}}\sqcup U_{p_r^{(k)}})\cap (S^{\star }\backslash \{1\})\Omega _k = \emptyset$ . Then for sufficiently big $n$ , there is no relation of length less than $m-2$ among the elements of the generating set $\{\gamma, \gamma ^n\gamma _1\gamma ^n, \gamma ^{2n}\gamma _2\gamma ^{2n}, \dots, \gamma ^{sn}\gamma _s\gamma ^{sn}\}$ where $S = \{\gamma _1, \dots, \gamma _s\}$ .

Indeed, such a relation would be of the form:

\begin{align*}W = \gamma ^{n_1}\delta _1\gamma ^{n_2}\delta _2\dots \gamma ^{n_r}\delta _r\gamma ^{n_{r+1}}\end{align*}

where $r\leq m-2$ , $|n_i| \geq n - m, 1\leq i\leq r+1$ , and $\delta _i\in S^{\star }\backslash \{1\}, 1\leq i\leq r$ . Let $y\in (S^{\star }\backslash \{1\})z$ . Then for sufficiently big $n$ , we have $W(y)\in U_{p_a^{(r+2)}}\sqcup U_{p_r^{(r+2)}}$ . Hence, $W(y)\neq y$ . Since $m$ is arbitrary, we conclude that $girth(\Gamma ) = \infty$ .

We will borrow the following definition from [Reference Akhmedov3].

While proving Theorem 1.2, we indeed proved a bit more, namely, for any non-solvable finitely generated group $\Gamma$ of $PL_o(I)$ , we proved that $girth _{2d}(\Gamma ) = \infty$ where $d = d(\Gamma )$ . Also, in the proof of Theorem 1.5, we indeed proved that $girth _{d+2}(\Gamma ) = \infty$ . With a slightly different argument, one can improve this result showing that $girth _{d+1}(\Gamma ) = \infty$ ; and in Theorem 4.4, one can prove that $girth _{|S|+1}(\Gamma ) = \infty$ .