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

Extending the Synchronous Fellow Traveler Property

Prohrak Kruengthomya Mahidol University, Faculty of Science, Department of Mathematics, Bangkok 10400, Thailand; e–mail: [email protected].    Dmitry Berdinsky Mahidol University, Faculty of Science, Department of Mathematics and Centre of Excellence in Mathematics, CHE, Bangkok 10400, Thailand; e–mail: [email protected].

Abstract

We introduce an extension of the fellow traveler property which allows fellow travelers to be at distance bounded from above by a function f(n)f(n) growing slower than any linear function. We study normal forms satisfying this extended fellow traveler property and certain geometric constraints that naturally generalize two fundamental properties of an automatic normal form – the regularity of its language and the bounded length difference property. We show examples of such normal forms and prove some non–existence theorems.

Keywords:     fellow traveler property, normal form, quasigeodesic, prefix–closed, Baumslag–Solitar group, wreath product

1 Introduction

The fellow traveler property is a cornerstone of the theory of automatic groups introduced by Thurston and others [8]. A normal form of a group satisfies the fellow traveler property if every two normal forms of group elements which are at distance one (with respect to some fixed set of generators) are kk–fellow travelers for some positive integer kk. The latter, informally speaking, means that when two fellow travelers are synchronously moving with the same speed along the paths labeled by such normal forms they are always at distance at most kk from each other. Each automatic group admits an automatic structure that includes a normal form which satisfies the fellow traveler property [8, § 2.3]. If a normal form of a group satisfies the fellow traveler property and its language is regular, then the group must be automatic.

The idea of extending automatic groups while retaining the fellow traveler property or its natural relaxations is not new. Requiring the fellow traveler property for a normal form but dropping the regularity condition for its language leads to the notions of a combing and a combable group. Bridson showed that there exist combable groups which are not automatic [4]111Note that the notion of a combable group used by Bridson is different from the one originally introduced by Epstein and Thurston [8, § 3.6].. A more general approach is to allow fellow travelers moving with different speeds. This leads to the notion of the asynchronous fellow traveler property. Thurston showed that the Baumslag–Solitar group BS(p,q)BS(p,q) for 1p<q1\leqslant p<q admits a normal form which satisfies the asynchronous fellow traveler and the language of this normal form is regular [8, § 7.4]. Bridson and Gilman showed that the fundamental group of a compact 33–manifold admits a normal form which satisfies the asynchronous fellow traveler property and its language is indexed [5].

In this paper we consider a relaxation of the fellow traveler property requiring fellow travelers to move with the same speed (synchronously) but allowing them to be at distance bounded from above by f(n)f(n), where f:+f:\mathbb{N}\rightarrow\mathbb{R}_{+} is a function growing slower than any linear function, e.g., nαn^{\alpha} for 0<α<10<\alpha<1 or log(n)\log(n), and nn is the distance that fellow travelers traversed starting from the origin, see Definition 5. In this case we say that a normal form satisfies the f(n)f(n)–fellow traveler property. First in the subsection 3.1 we notice that such normal forms can be trivially constructed. So then in the subsection 3.2 we introduce two types of geometric constraints each of which makes constructing such normal forms nontrivial. The first geometric constraint requires a normal form to be quasigeodesic. It is a relaxation of the bounded length difference property222Note the bounded length difference property is referred to as the comparable length property in [4]. Also, the bounded length difference property is incorporated in the notion of a combable group introduced by Epstein and Thurston. It is an open question whether there exists a combable group in the sense of Epstein and Thurston which is not automatic. [8, Lemma 2.3.9] which requires the geodesic length of a group element and the length of its normal form to be linearly comparable, see Definition 6. The second geometric constraint requires a normal form to be quasiregular. It is a relaxation of the regularity condition for the language of a normal form which requires that for each prefix uu of a normal form of a group element there exists a word vv of length at most c>0c>0 for which uvuv is a normal form of some group element, see Definition 7. Alternatively, quasiregularity can be considered as a relaxation of prefix–closedness; see also Definition 8 and Theorem 9.

The main results of the paper are as follows. In Theorems 14 and 15 we show that there exists no quasigeodesic normal form which satisfies the f(n)f(n)–fellow traveler property in a finitely presented group with the strongly–super–polynomial Dehn function and a non–finitely presented group, respectively; see Definition 12 for the notion of a strongly–super–polynomial function. In Theorem 16 we show relation between the notion of a Cayley distance function studied by Elder, Taback, Trakuldit and the second author [1, 2, 3] and quasigeodesic normal forms satisfying the f(n)f(n)–fellow traveler property. Namely, Theorem 16 shows that if a non–automatic group has a Cayley automatic representation for which the Cayley distance function h(n)h(n) grows slower than a linear function333The existence of such Cayley automatic representations is an open question., then this group admits a quasigeodesic normal form satisfying the h(n)h(n)–fellow traveler property. Theorem 17 shows that for 1p<q1\leqslant p<q the Baumslag–Solitar group BS(p,q)=a,t|tapt1=aqBS(p,q)=\langle a,t\,|\,ta^{p}t^{-1}=a^{q}\rangle admits a prefix–closed (so quasiregular) normal form satisfying the log(n)\log(n)–fellow traveler property. Theorem 20 shows that the wreath product 22\mathbb{Z}_{2}\wr\mathbb{Z}^{2} admits a prefix–closed normal form satisfying the n\sqrt{n}–fellow traveler property444Note that by Theorems 14 and 15 the groups BS(p,q)BS(p,q) for 1p<q1\leqslant p<q and 22\mathbb{Z}_{2}\wr\mathbb{Z}^{2} do not admit quasigeodesic normal forms satisfying the f(n)f(n)–fellow traveler property..

The rest of the paper is organized as follows. In Section 2 we recall the notions of a group, a normal form, an automatic group, the fellow traveler property and the relations \preceq and \ll for nondecreasing functions appeared in this paper. In Section 3 we introduce the f(n)f(n)–fellow traveler property, quasigeodesic and quasiregular normal forms. In Section 4 for quasigeodesic normal forms satisfying the f(n)f(n)–fellow traveler property we prove the non–existence theorems in groups with the strongly–super–polynomial Dehn function and non–finitely presented groups and show relation with the notion of a Cayley distance function. In Section 5 we show examples of quasiregular normal forms satisfying the f(n)f(n)–fellow traveler property. Section 6 concludes the paper.

2 Preliminaries

In this section we introduce necessary notations and recall some definitions.

Groups and normal forms. Let GG be a finitely generated infinite group and A={a1,,am}A=\{a_{1},\dots,a_{m}\}, where aiGa_{i}\in G for i=1,,mi=1,\dots,m, be a finite set generating GG: each element of GG can be written as a product of elements from AA and their inverses. We allow different ai,ajAa_{i},a_{j}\in A, iji\neq j, to be equal in GG and some elements in AA to be equal the identity eGe\in G. We denote by A1A^{-1} the set of formal inverses for elements in AA: A1={a11,,am1}A^{-1}=\{a_{1}^{-1},\dots,a_{m}^{-1}\} and by SS the union of AA and A1A^{-1}: S=AA1S=A\cup A^{-1}. We denote by dAd_{A} the word metric in GG relative to AA: for g1,g2Gg_{1},g_{2}\in G the distance dA(g1,g2)d_{A}(g_{1},g_{2}) is the length of a shortest word uSu\in S^{*} equal to g11g2g_{1}^{-1}g_{2} in GG. For a given gGg\in G we denote by dA(g)d_{A}(g) the distance between gg and the identity eGe\in G with respect to dAd_{A}: dA(g)=dA(e,g)d_{A}(g)=d_{A}(e,g). For a given w=s1sSw=s_{1}\dots s_{\ell}\in S^{*}, we denote by |w||w| the length of ww: |w|=|w|=\ell and by π(w)\pi(w) the group element s1sGs_{1}\dots s_{\ell}\in G, where π\pi refers to the canonical projection map π:SG\pi:S^{*}\rightarrow G.

A normal form of GG is a rule for assigning a word wSw\in S^{*} to a group element gGg\in G such that π(w)=g\pi(w)=g. The word ww is referred to as a normal form of gg. In this paper we always assume that a normal form is one–to–one: for each gGg\in G exactly one word wSw\in S^{*} is assigned. A normal form defines a language LSL\subseteq S^{*}. Similarly, a language LSL\subseteq S^{*} for which the restriction πL:LG\pi_{L}:L\rightarrow G is surjective and one–to–one defines a normal form of GG.

The fellow traveler property and automatic groups. Let wSw\in S^{*} be a word and t[0,+)t\in[0,+\infty) be a nonnegative integer. We define w(t)w(t) to be the prefix of ww of a length tt if t|w|t\leqslant|w| and ww if t>|w|t>|w|. Let LSL\subseteq S^{*} be a normal form of GG. We denote by s(n)s(n) a function s:[0,+)+s:[0,+\infty)\rightarrow\mathbb{R}_{+} defined as:

s(n)=max{dA(π(w1(t)),π(w2(t)))|tn,w1,w2L,aA,π(w1)a=π(w2)}.s(n)=\max\{d_{A}(\pi(w_{1}(t)),\pi(w_{2}(t)))\,|\,t\leqslant n,\,w_{1},w_{2}\in L,a\in A,\pi(w_{1})a=\pi(w_{2})\}. (1)

The function s(n)s(n) is the maximum distance between fellow travelers moving with the same speed along the paths in Γ(G,A)\Gamma(G,A) labeled by words w1w_{1} and w2w_{2} for which π(w1)a=π(w2)\pi(w_{1})a=\pi(w_{2}) for some aAa\in A.

Definition 1 (the fellow traveler property).

It is said that the normal form LSL\subseteq S^{*} satisfies the fellow traveler property if the function s(n)s(n) given by (1) is bounded from above by a constant.

Recall that the group GG is called automatic if it admits a normal form LSL\subseteq S^{*} for which the language LL is regular and for each aAa\in A the binary relation Ra={(u,v)|π(u)a=π(v)}R_{a}=\{(u,v)\,|\,\pi(u)a=\pi(v)\} is recognized by a two–tape synchronous automaton [8]. In this case the normal form LL satisfies the fellow traveler property. Equivalently, if GG admits a normal form LSL\subseteq S^{*} which satisfies the fellow traveler property and LL is regular, then GG must be automatic [8, Theorem 2.3.5]. In this paper we focus on groups which are not automatic (non–automatic groups).

The relations \preceq and \ll for nondecreasing functions. We denote by \mathbb{N} a set of all natural numbers which includes zero. For a given NN\in\mathbb{N} we denote by [N,+)[N,+\infty) the set [N,+)={n|nN}\left[N,+\infty\right)=\{n\in\mathbb{N}\,|\,n\geqslant N\}. Let \mathcal{F} be the set of all nondecreasing functions f:[N,+)+f:[N,+\infty)\rightarrow\mathbb{R}_{+}, where +={x0|x}\mathbb{R}_{+}=\{x\geqslant 0\,|\,x\in\mathbb{R}\}.

Definition 2 (\preceq relation).

For given f,hf,h\in\mathcal{F} we say that hfh\preceq f if there exist positive integers K,MK,M and a nonnegative NN such that h(n)Kf(Mn)h(n)\leqslant Kf(Mn) for all n[N,+)n\in\left[N,+\infty\right). We say that hfh\asymp f if hfh\preceq f and fhf\preceq h. If hfh\preceq f and hfh\not\asymp f, we say that hfh\prec f.

Definition 3 (\ll relation).

For given f,hf,h\in\mathcal{F} we say that hfh\ll f if there exists an unbounded function tt\in\mathcal{F} such that htfht\preceq f.

We denote by 𝔦:[0,+)+\mathfrak{i}:[0,+\infty)\rightarrow\mathbb{R}_{+} the identity function: 𝔦(n)=n\mathfrak{i}(n)=n. Note that f𝔦f\ll\mathfrak{i} is stronger than f𝔦f\prec\mathfrak{i}, that is, f𝔦f\ll\mathfrak{i} implies f𝔦f\prec\mathfrak{i}. Indeed, f𝔦f\ll\mathfrak{i} implies f𝔦f\preceq\mathfrak{i}. Now suppose that f𝔦f\ll\mathfrak{i} and 𝔦f\mathfrak{i}\preceq f. The inequality 𝔦f\mathfrak{i}\preceq f implies that there exist positive integers K,MK,M and a nonnegative integer NN such that nKf(Mn)n\leqslant Kf(Mn) for all n[N,+)n\in\left[N,+\infty\right). The inequality f𝔦f\ll\mathfrak{i} implies that there exist positive integers K,MK^{\prime},M^{\prime} and a nonnegative integer NN^{\prime} such that f(n)t(n)KMnf(n)t(n)\leqslant K^{\prime}M^{\prime}n for all n[N,+)n\in\left[N^{\prime},+\infty\right), where t(n)t(n) is some unbounded function. Therefore, f(n)KMt(n)nf(n)\leqslant\frac{K^{\prime}M^{\prime}}{t(n)}n for all n[N,+)n\in\left[N^{\prime},+\infty\right), which implies that Kf(Mn)KKMMt(Mn)nKf(Mn)\leqslant\frac{KK^{\prime}MM^{\prime}}{t(Mn)}n also for all n[N,+)n\in\left[N^{\prime},+\infty\right). Since t(n)t(n) is unbounded, we get a contradiction. Therefore, we have that 𝔦f\mathfrak{i}\not\preceq f, so 𝔦f\mathfrak{i}\prec f. The reverse (f𝔦f\prec\mathfrak{i} implies f𝔦f\ll\mathfrak{i}) in general is not true as it is shown in Example 4.

Example 4.

Let ni,i1n_{i},i\geqslant 1 be an infinite sequence defined recursively by the identities: n0=0n_{0}=0, n1=1n_{1}=1 and ni+1=ni22in_{i+1}=n_{i}2^{2i} for i1i\geqslant 1. Let f(n)f(n) be a function for which f(n)=2nif(n)=2n_{i} for nin<ni+1n_{i}\leqslant n<n_{i+1}. Clearly, f(n)𝔦f(n)\preceq\mathfrak{i}. Let us show that 𝔦f(n)\mathfrak{i}\not\preceq f(n). The inequality 𝔦f\mathfrak{i}\preceq f implies that there exist positive integers K,MK,M and NN such that nKf(Mn)n\leqslant Kf(Mn) for all n[N,+)n\in\left[N,+\infty\right). Therefore, if ni2iNn_{i}2^{i}\geqslant N, then ni2iKf(Mni2i)n_{i}2^{i}\leqslant Kf(Mn_{i}2^{i}). By the definition of f(n)f(n), if M<2iM<2^{i}, then f(Mni2i)=2nif(Mn_{i}2^{i})=2n_{i}. Therefore, if ni2iNn_{i}2^{i}\geqslant N and M<2iM<2^{i}, then ni2i2Knin_{i}2^{i}\leqslant 2Kn_{i}. The latter inequality is true only if 2i2K2^{i}\leqslant 2K and false, otherwise. Thus, 𝔦f\mathfrak{i}\not\preceq f. Therefore, f𝔦f\prec\mathfrak{i}. Let us show now that f≪̸𝔦f\not\ll\mathfrak{i}. The inequality f𝔦f\ll\mathfrak{i} implies that there exist positive integers K,MK,M and a nonnegative integer NN such that t(n)f(n)KMnt(n)f(n)\leqslant KMn for some unbounded tt\in\mathcal{F} and all n[N,+)n\in\left[N,+\infty\right). Since t(n)t(n) is unbounded, there exists i0i_{0} such that for all ii0i\geqslant i_{0}, t(ni)KMt(n_{i})\geqslant KM which implies that f(ni)=2ninif(n_{i})=2n_{i}\leqslant n_{i} for all ii0i\geqslant i_{0}. The latter is impossible for i>0i>0.

3 The f(n)f(n)–Fellow Traveler Property

We extend the fellow traveler property by allowing the distance between fellow travelers to be bounded from above by a nondecreasing function f(n)f(n). If f(n)f(n) is bounded from above by a constant, then one simply gets the fellow traveler property, see Definition 1. We will allow the function f(n)f(n) to be unbounded.

Definition 5 (the f(n)f(n)–fellow traveler property).

Let ff\in\mathcal{F} be a function for which f𝔦f\ll\mathfrak{i}. We say that a given normal form LSL\subseteq S^{*} of a group GG satisfies the f(n)f(n)–fellow traveller property if for the function s(n)s(n) given by (1) the inequality sfs\preceq f holds.

In Definition 5 the function f(n)f(n) is an upper bound for the distance between fellow travelers in coarse sense. It can be verified that Definition 5 does not depend on the choice of the set of generators AA. By the triangle inequality, for every normal form in GG we always have that s(n)2ns(n)\leqslant 2n for all nn\in\mathbb{N}. From Example 4 we see that there exists a function f𝔦f\prec\mathfrak{i} for which f(n)=2nf(n)=2n at infinitely many points. If f𝔦f\ll\mathfrak{i}, then, informally speaking, ff genuinely grows slower than 𝔦\mathfrak{i}. This explains the choice of the inequality f𝔦f\ll\mathfrak{i} over the inequality f𝔦f\prec\mathfrak{i} in Definition 5. For illustration of the f(n)f(n)–fellow traveler property see Fig. 1.

Refer to caption
Figure 1: The upper and lower curves show the paths labeled by the normal forms w1w_{1} and w2w_{2} of group elements gg and gaga, respectively. The pairs of dots and dashed curves connecting it show fellow travelers and shortest paths between them, respectively.

3.1 Normal forms satisfying the f(n)f(n)–fellow traveler property

We now show two ways of modifying a given normal form so the modified normal form satisfies the f(n)f(n)–fellow traveler property for some f𝔦f\ll\mathfrak{i}. Let us be given a normal form of a group GG defined by a language LSL\subseteq S^{*}.

First way. Let uSu\in S^{*} be a word defining a loop in Γ(G,S)\Gamma(G,S): π(u)=e\pi(u)=e. We define a language LSL^{\prime}\subseteq S^{*} as L={u2w|wL=|w|}L^{\prime}=\{u^{\ell^{2}}w\,|\,w\in L\wedge\ell=|w|\}. That is, for every wLw\in L we attach a prefix u2u^{\ell^{2}} to the word ww, where =|w|\ell=|w|. See Figure 2 for illustration. For a normal form given by a language LL^{\prime} the function s(n)s(n) is bounded from above by n\sqrt{n}: s(n)ns(n)\preceq\sqrt{n}. Indeed, let us consider two words w1Lw_{1}\in L and w2Lw_{2}\in L for which π(w1)a=π(w2)\pi(w_{1})a=\pi(w_{2}). Let |u|=c|u|=c, 1=|w1|\ell_{1}=|w_{1}| and 2=|w2|\ell_{2}=|w_{2}|. Without loss of generality we assume that 12\ell_{1}\leqslant\ell_{2}. We denote by nn a number of steps traversed by fellow travelers along the paths labeled by w1w_{1} and w2w_{2}. There are three different cases to consider:

  • If nc12n\leqslant c\ell_{1}^{2}, then the distance between fellow travelers is bounded from above by 2c2c.

  • If c12<nc22c\ell_{1}^{2}<n\leqslant c\ell_{2}^{2}, then the distance between fellow travelers is bounded from above by (1+c)(\ell_{1}+c), so it is strictly less than n/c+c\sqrt{n/c}+c.

  • If c22<nc22+2c\ell_{2}^{2}<n\leqslant c\ell_{2}^{2}+\ell_{2}, then the distance between fellow travelers is bounded from above by 1\ell_{1}+222\ell_{2}\leqslant 2\ell_{2}, so it is strictly less than 2n/c2\sqrt{n/c}.

From these three cases it can be seen that s(n)ns(n)\preceq\sqrt{n} for the normal form defined by the language LL^{\prime}.

Refer to caption
Figure 2: A curve on the left shows the path labeled by a normal form ww of a group element gg. A curve on the right shows the path labeled by the modified normal of gg: first it traverses the loop labeled by uu doing it 2\ell^{2} times and then it traverses the path labeled by ww.

Second way. Let w=s1smw=s_{1}\dots s_{m} be a nonempty word in the language LL. For each integer k{1,,m}k\in\{1,\dots,m\}, let us choose a loop ukSu_{k}\in S^{*} for which C1k|uk|C2kC_{1}\sqrt{k}\leqslant|u_{k}|\leqslant C_{2}\sqrt{k} for some fixed constants C1,C2>0C_{1},C_{2}>0. These loops uku_{k}, k=1,,mk=1,\dots,m can be chosen arbitrarily for each wLw\in L and k{1,,m}k\in\{1,\dots,m\}, where m=|w|m=|w|. Let w=s1u1s2u2smumw^{\prime}=s_{1}u_{1}s_{2}u_{2}\dots s_{m}u_{m}. If w=εw=\varepsilon, then we put w=εw^{\prime}=\varepsilon. We define a language L′′SL^{\prime\prime}\subseteq S^{*} as L′′={w|wL}L^{\prime\prime}=\{w^{\prime}\,|\,w\in L\}. See Figure 3 for illustration.

For a normal form given by a language L′′L^{\prime\prime} the function s(n)s(n) is bounded from above by n\sqrt{n}: s(n)ns(n)\preceq\sqrt{n}. Indeed, let us consider two words w1=s1sm1Lw_{1}=s_{1}\dots s_{m_{1}}\in L and w2=t1tm2Lw_{2}=t_{1}\dots t_{m_{2}}\in L for which π(w1)a=π(w2)\pi(w_{1})a=\pi(w_{2}). For the words w1w_{1} and w2w_{2}, let w1=s1u1sm1um1w_{1}^{\prime}=s_{1}u_{1}\dots s_{m_{1}}u_{m_{1}} and w2=t1v1tm2vm2w_{2}^{\prime}=t_{1}v_{1}\dots t_{m_{2}}v_{m_{2}}, respectively. Let nn be a number of steps traversed by fellow travelers along the paths labeled by w1w_{1} and w2w_{2}. We denote by k1k_{1} the integer for which w1(n)=s1u1sk1uk1q1w_{1}(n)=s_{1}u_{1}\dots s_{k_{1}}u_{k_{1}}q_{1}, where either k1=m1k_{1}=m_{1} and q1=εq_{1}=\varepsilon or q1q_{1} is a proper prefix of sk1+1uk1+1s_{k_{1}+1}u_{k_{1}+1}. Similarly, we denote by k2k_{2} the integer for which w2(n)=t1v1tk2vk2q2w_{2}(n)=t_{1}v_{1}\dots t_{k_{2}}v_{k_{2}}q_{2}, where either k2=m2k_{2}=m_{2} and q2=εq_{2}=\varepsilon or q2q_{2} is a proper prefix of tk2+1vk2+1t_{k_{2}+1}v_{k_{2}+1}. Now the distance between fellow travelers is bounded from above by k1+|q1|+k2+|q2|k1+(1+C2k1+1)+k2+(1+C2k2+1)k_{1}+|q_{1}|+k_{2}+|q_{2}|\leqslant k_{1}+(1+C_{2}\sqrt{k_{1}+1})+k_{2}+(1+C_{2}\sqrt{k_{2}+1}). On the other hand, we have that Ck132nCk_{1}^{\frac{3}{2}}\leqslant n and Ck232nCk_{2}^{\frac{3}{2}}\leqslant n for some constant C>0C>0. Therefore, s(n)n23s(n)\preceq n^{\frac{2}{3}} for the normal form defined by the language L′′L^{\prime\prime}.

Refer to caption
Figure 3: A curve on the left shows the path labeled by a normal form w=s1s2smw=s_{1}s_{2}\dots s_{m} of a group element gg. A curve on the right shows the path labeled by the modified normal form of gg: s1u1s2u2siuismums_{1}u_{1}s_{2}u_{2}\dots s_{i}u_{i}\dots s_{m}u_{m}.

3.2 Quasigeodesic and quasiregular normal forms

We introduce two kinds of geometric constraints for a normal form which originate from two basic properties of an automatic normal form: the bounded length difference property [8, Lemma 2.3.9] and the regularity of its language. These geometric constraints in general are an obstacle for trivial constructions of normal forms satisfying the f(n)f(n)–fellow traveler property shown in the subsection 3.1. In this paper normal forms satisfying these geometric constraints are referred to as quasigeodesic and quasiregular.

Quasigeodesic normal forms. Recall that the bounded length difference property for the normal form defined by a language LSL\subseteq S^{*} means that there exists a constant C>0C^{\prime}>0 such that for every w1,w2Lw_{1},w_{2}\in L for which π(w1)a=π(w2)\pi(w_{1})a=\pi(w_{2}) for some aAa\in A the inequality ||w1||w2||C||w_{1}|-|w_{2}||\leqslant C^{\prime} holds. Let C=max{C,|w0|}C=\max\{C^{\prime},|w_{0}|\}, where w0w_{0} is the normal form of the identity eGe\in G. Then the bounded length difference property implies that for every wLw\in L we have |w|C(dA(π(w))+1)|w|\leqslant C(d_{A}(\pi(w))+1). The latter is the notion of a quasigeodesic normal form introduced by Elder and Taback [7], see Definition 6. Note that for a normal form being quasigeodesic, in general, does not imply having the bounded length difference property.

Definition 6 (quasigeodesic normal form).

A normal form defined by a language LSL\subseteq S^{*} is said to be quasigeodesic if there is a constant C>0C>0 such that for every wLw\in L the inequality |w|C(dA(π(w))+1)|w|\leqslant C(d_{A}(\pi(w))+1) holds.

Note that if LSL\subseteq S^{*} is a quasigeodesic normal form in the sense of Definition 6, the paths labeled by elements of LL might not be quasigeodesics in the standard sense, i.e., not every subword uu^{\prime} of uLu\in L needs to satisfy the inequality |u|C(dA(π(u))+1)|u^{\prime}|\leqslant C(d_{A}(\pi(u^{\prime}))+1). We notice that normal forms constructed in the subsection 3.1 are not quasigeodesic. Indeed, for the first way we have that the normal form of a group element π(w)\pi(w) is w=u2ww^{\prime}=u^{\ell^{2}}w, where =|w|\ell=|w|. Since dA(π(w))d_{A}(\pi(w))\leqslant\ell and |w|=|u|2+2|w^{\prime}|=|u|\ell^{2}+\ell\geqslant\ell^{2}, the inequality |w|C(dA(π(w))+1)|w^{\prime}|\leqslant C(d_{A}(\pi(w))+1) cannot hold for all wLw\in L and some constant C>0C>0. For the second way we have that the normal form of a group element π(w)\pi(w) is w=s1u1s2u2smumw^{\prime}=s_{1}u_{1}s_{2}u_{2}\dots s_{m}u_{m}. Since dA(π(w))md_{A}(\pi(w))\leqslant m and w=m+|u1|++|um|C1m2m2w^{\prime}=m+|u_{1}|+\dots+|u_{m}|\geqslant C_{1}\frac{m}{2}\sqrt{\frac{m}{2}}, the inequality |w|C(dA(π(w))+1)|w^{\prime}|\leqslant C(d_{A}(\pi(w))+1) cannot not hold for all wLw\in L and some constant C>0C>0.

Quasiregular normal forms. In Definition 7 we introduce the notion of a quasiregular normal form. We consider it as a geometric version of the regularity of the language LSL\subseteq S^{*} defining a normal form. Indeed, if a language LSL\subseteq S^{*} is regular, then the normal form it defines is quasiregular. Note that a quasiregular normal form does not necessarily define a regular language LSL\subseteq S^{*}.

Definition 7 (quasiregular normal form).

We say that a normal form defined by a language LSL\subseteq S^{*} is quasiregular if there exists a constant c0c\geqslant 0 such that for each prefix uSu\in S^{*} of a word w=uvLw=uv\in L there is a word xSx\in S^{*} of length |x|c|x|\leqslant c for which uxLux\in L.

We notice that a normal form constructed using the first way in the subsection 3.1 is not quasiregular. Indeed, there are infinitely many words ukxLu^{k}x\in L, where uu is a fixed loop and |x|c|x|\leqslant c, that represent group elements for which the distances to eGe\in G are bounded by some constant from above. The number of such group elements is finite. Since we consider only one–to–one normal forms, the latter is impossible. Also, there does not seem to be any trivial construction of quasiregular normal forms using the second way in the subsection 3.1.

One can consider a more restricted version of the notion of a quasiregular normal forms by additionally requiring in Definition 7 that xx is a prefix of vv. This leads to the notion of a quasiprefix–closed normal form introduced in Definition 8.

Definition 8 (quasiprefix–closed normal form).

We say that a normal form defined by a language LSL\subseteq S^{*} is quasiprefix–closed if there exists a constant C0C\geqslant 0 such that for every prefix uSu\in S^{*} of a word w=uvLw=uv\in L there is a prefix xSx\in S^{*} of length |x|C|x|\leqslant C of the word vv for which uxLux\in L.

Prefix–closed normal forms are exactly quasiprefix–closed normal forms with the constant C=0C=0. A quasiprefix–closed normal form is quasiregular. The reverse in general is not true. However, Theorem 9 shows that if a quasiregular normal form satisfies the f(n)f(n)–fellow traveler property, then one can construct a quasiprefix–closed normal form which also satisfies the f(n)f(n)–fellow traveler property. So in the context of the f(n)f(n)–fellow traveler property the notions of a quasiregular normal form and a quasiprefix–closed normal form are equivalent.

Theorem 9.

Suppose LSL\subseteq S^{*} defines a quasiregular normal form for some constant c0c\geqslant 0 which satisfies the f(n)f(n)–fellow traveler property. Then one can construct a quasiprefix–closed normal form LSL^{\prime}\subseteq S^{*} for the constant C=4cC=4c which also satisfies the f(n)f(n)–fellow traveler property.

Proof.

First we note that if c=0c=0, then the normal form defined by the language LL is prefix–closed. So we further assume that c>0c>0. Now let uu be a prefix of length k(c+1)k(c+1) of a word w=uvLw=uv\in L for an integer k0k\geqslant 0. Since LL is a quasiregular normal form, there exists xSx\in S^{*} of length |x|c|x|\leqslant c for which uxLux\in L. Let ySy\in S^{*} be any word of length c|x|c-|x|. We define q(u)q(u) to be the word xyy1x1Sxyy^{-1}x^{-1}\in S^{*}, where x1x^{-1} and y1y^{-1} are the inverses of xx and yy, respectively. The word q(u)q(u) depends on uu only.

A word wLw\in L can be always written as a concatenation w=u1u2umtw=u_{1}u_{2}\dots u_{m}t of words u1,,umSu_{1},\dots,u_{m}\in S^{*} and tSt\in S^{*}, where |u1|==|um|=c+1|u_{1}|=\dots=|u_{m}|=c+1 and 0|t|<c+10\leqslant|t|<c+1. For each word wLw\in L we construct a word ww^{\prime} as follows:

w=u1q(u1)um1q(u1um1)umt.w^{\prime}=u_{1}q(u_{1})\dots u_{m-1}q(u_{1}\dots u_{m-1})u_{m}t. (2)

We define a language LL^{\prime} as L={w|wL}L^{\prime}=\{w^{\prime}\,|\,w\in L\}. Let us show that LL^{\prime} is a quasiprefix–closed normal form for the constant C=4cC=4c. We denote by xi,yiSx_{i},y_{i}\in S^{*} the words for which q(u1ui)=xiyiyi1xi1q(u_{1}\dots u_{i})=x_{i}y_{i}y_{i}^{-1}x_{i}^{-1}. Let uu^{\prime} be a prefix of ww^{\prime}. There are the following three possibilities.

  • The prefix uu^{\prime} is of the form u=u1q(u1)ui1q(u1ui1)piu^{\prime}=u_{1}q(u_{1})\dots u_{i-1}q(u_{1}\dots u_{i-1})p_{i}, where i=1,,m1i=1,\dots,m-1 and pip_{i} is a prefix of uiu_{i}. By the definition of xix_{i}, we have that u1ui1uixiLu_{1}\dots u_{i-1}u_{i}x_{i}\in L. Let u′′=u1q(u1)ui1q(u1ui1)uixiu^{\prime\prime}=u_{1}q(u_{1})\dots u_{i-1}q(u_{1}\dots u_{i-1})u_{i}x_{i}. It can be seen that u′′Lu^{\prime\prime}\in L^{\prime}, uu^{\prime} is a prefix of u′′u^{\prime\prime} and u′′u^{\prime\prime} is a prefix of ww^{\prime}. Let ui=piτiu_{i}=p_{i}\tau_{i}. Then u′′=uτixiu^{\prime\prime}=u^{\prime}\tau_{i}x_{i}, so |u′′||u|=|τi|+|xi|(c+1)+c4c|u^{\prime\prime}|-|u^{\prime}|=|\tau_{i}|+|x_{i}|\leqslant\left(c+1\right)+c\leqslant 4c.

  • The prefix uu^{\prime} is of the form u=u1q(u1)ui1q(u1ui1)uipiu^{\prime}=u_{1}q(u_{1})\dots u_{i-1}q(u_{1}\dots u_{i-1})u_{i}p_{i}, where i=1,,m1i=1,\dots,m-1 and pip_{i} is a prefix of q(u1ui)=xiyiyi1xi1q(u_{1}\dots u_{i})=x_{i}y_{i}y_{i}^{-1}x_{i}^{-1}. If pip_{i} is a prefix of xix_{i}, then we put u′′=u1q(u1)ui1q(u1ui1)uixiu^{\prime\prime}=u_{1}q(u_{1})\dots u_{i-1}q(u_{1}\dots u_{i-1})u_{i}x_{i}. If pip_{i} is not a prefix of xix_{i}, then we put u′′=u1q(u1)ui1q(u1ui1)uiq(u1ui)γiu^{\prime\prime}=u_{1}q(u_{1})\dots u_{i-1}q(u_{1}\dots u_{i-1})u_{i}q(u_{1}\dots u_{i})\gamma_{i}, where γi=ui+1xi+1\gamma_{i}=u_{i+1}x_{i+1} if i<m1i<m-1 and γi=umt\gamma_{i}=u_{m}t if i=m1i=m-1. It can be seen that u′′Lu^{\prime\prime}\in L^{\prime}, uu^{\prime} is a prefix of u′′u^{\prime\prime} and u′′u^{\prime\prime} is a prefix of ww^{\prime}. In the first case when pip_{i} is a prefix of xix_{i}, we have that |u′′||u||xi|c|u^{\prime\prime}|-|u^{\prime}|\leqslant|x_{i}|\leqslant c. In the second case when pip_{i} is not a prefix of xix_{i}, we have that |u′′||u|(2c1)+(c+1)+c=4c|u^{\prime\prime}|-|u^{\prime}|\leqslant(2c-1)+(c+1)+c=4c.

  • The prefix uu^{\prime} is of the form u=u1q(u1)um1q(u1um1)pmu^{\prime}=u_{1}q(u_{1})\dots u_{m-1}q(u_{1}\dots u_{m-1})p_{m}, where pmp_{m} is a prefix of umtu_{m}t. Let u′′=wu^{\prime\prime}=w^{\prime}. Then u′′Lu^{\prime\prime}\in L, uu^{\prime} is a prefix of u′′u^{\prime\prime} and u′′u^{\prime\prime} is a prefix of ww^{\prime}. We have that |u′′||u|(c+1)+c4c|u^{\prime\prime}|-|u^{\prime}|\leqslant(c+1)+c\leqslant 4c.

It is straightforward from (2) that if LL satisfies the f(n)f(n)–fellow traveler property, then LL^{\prime} also satisfies the f(n)f(n)–fellow traveler property. ∎

4 Quasigeodesic Normal Forms

This section discusses quasigeodesic normal forms in the context of the f(n)f(n)–fellow traveler property. In the subsection 4.1 we show that groups with the strongly–super–polynomial Dehn function and non–finitely presented groups do not admit quasigeodesic normal forms satisfying the f(n)f(n)–fellow traveler property. In the subsection 4.2 we show relation with the notion of a Cayley distance function.

4.1 Non–existence theorems

This subsection presents non–existence theorems for quasigeodesic normal forms satisfying the f(n)f(n)–fellow traveler property. First we consider finitely presented groups. We show that if the Dehn function of a group is strongly–super–polynomial (see Definition 12), then it does not admit a quasigeodesic normal form satisfying the f(n)f(n)–fellow traveler property (see Theorem 14). Then we consider non–finitely presented groups. We show that they do not admit quasigeodesic normal forms satisfying the f(n)f(n)–fellow traveler property (see Theorem 15).

Finitely presented groups. Let GG be a group. We assume first that GG is finitely presented. Let G=A|RG=\langle A\,|\,R\rangle be a finite presentation of GG. We denote by FAF_{A} a free group on AA. Let wFAw\in F_{A} be a reduced word for which π(w)=e\pi(w)=e in the group GG. Let us recall the definitions of a combinatorial area and the Dehn function.

Definition 10 (combinatorial area).

The area 𝒜(w)\mathcal{A}(w) with respect to the presentation A|R\langle A\,|\,R\rangle is the minimum NN for which w=i=1Nτi1ri±1τiw=\prod\limits_{i=1}^{N}\tau_{i}^{-1}r_{i}^{\pm 1}\tau_{i} in FAF_{A}, where riRr_{i}\in R and τiFA\tau_{i}\in F_{A}.

Definition 11 (Dehn function).

The Dehn function of GG with respect to the presentation A|R\langle A\,|\,R\rangle is the function δ:\delta:\mathbb{N}\rightarrow\mathbb{N} such that δ(n)=max{𝒜(w)|wFA,|w|n}\delta(n)=\max\{\mathcal{A}(w)\,|\,w\in F_{A},|w|\leqslant n\}.

Now we recall the notion of a strongly–super–polynomial function introduced in [1].

Definition 12 (strongly–super–polynomial function).

A non–zero function ff\in\mathcal{F} is said to be strongly–super–polynomial if n2ffn^{2}f\ll f.

Note that for every c>0c>0, one has that n2ffn^{2}f\ll f if and only if ncffn^{c}f\ll f [1]. Strongly–super–polynomial functions include, for example, exponential functions.

The following lemma is a key ingredient in the proof of Theorem 14.

Lemma 13.

If a group G=A|RG=\langle A\,|\,R\rangle admits a quasigeodesic normal form satisfying the f(n)f(n)–fellow traveler property, then there exist constants C,D>0C,D>0 and n00n_{0}\geqslant 0 such that for the Dehn function δ(n)\delta(n) of GG with respect to the presentation A|R\langle A\,|\,R\rangle the inequality δ(n)Dn2δ(Cf(Cn))\delta(n)\leqslant Dn^{2}\delta(Cf(Cn)) holds for all nn0n\geqslant n_{0}.

Proof.

Suppose that GG admits a quasigeodesic normal form satisfying the f(n)f(n)–fellow traveler property. Let LSL\subseteq S^{*} be a language defining such normal form in GG. Now let w=s1snSw=s_{1}\dots s_{n}\in S^{*} be a word for which π(w)=e\pi(w)=e in the group GG, where n=|w|n=|w|. For each i=1,,n1i=1,\dots,n-1, let uiLu_{i}\in L be the normal form of the group element gi=π(s1si)g_{i}=\pi(s_{1}\dots s_{i}). We first divide a loop ww into subloops: s1u11s_{1}u_{1}^{-1}, u1s2u21,,un2sn1un11u_{1}s_{2}u_{2}^{-1},\dots,u_{n-2}s_{n-1}u_{n-1}^{-1}, un1snu_{n-1}s_{n}.

For a given i{1,,n1}i\in\{1,\dots,n-1\} and j1j\geqslant 1 we denote by ui,jSu_{i,j}\in S the jjth symbol in the normal form uiSu_{i}\in S^{*}, if j|ui|j\leqslant|u_{i}|, and ui,j=eu_{i,j}=e, if j>|ui|j>|u_{i}|. In particular, a prefix of uiu_{i} of length j|ui|j\leqslant|u_{i}| is ui,1ui,ju_{i,1}\dots u_{i,j}. For a given i{1,,n2}i\in\{1,\dots,n-2\} and j1j\geqslant 1, let vi,jSv_{i,j}\in S^{*} be a shortest path connecting gi,j=π(ui,1ui,j)g_{i,j}=\pi(u_{i,1}\dots u_{i,j}) and gi+1,j=π(ui+1,1ui+1,j)g_{i+1,j}=\pi(u_{i+1,1}\dots u_{i+1,j}): gi,jvi,j=gi+1,jg_{i,j}\cdot v_{i,j}=g_{i+1,j}. For illustration see Figure 4.

Refer to caption
Figure 4: The outer cycle shows a loop w=s1sisi+1si+2snw=s_{1}\dots s_{i}s_{i+1}s_{i+2}\dots s_{n}. The directed edge labeled by si+1s_{i+1} leads from the group element gig_{i} to the group element gi+1g_{i+1}. The curves uiu_{i} and ui+1u_{i+1} show normal forms of the group elements gig_{i} and gi+1g_{i+1}, respectively. The curves vi,jv_{i,j} and vi,j1v_{i,j-1} show shortest paths between gi,j,gi+1,jg_{i,j},g_{i+1,j} and gi,j1,gi+1,j1g_{i,j-1},g_{i+1,j-1}, respectively. The directed edges labeled by ui,ju_{i,j} and ui+1,ju_{i+1,j} lead from gi,j1g_{i,j-1} to gi,jg_{i,j} and from gi+1,j1g_{i+1,j-1} to gi+1,jg_{i+1,j}, respectively.

For a given i{1,,n2}i\in\{1,\dots,n-2\}, let mi=max{|ui|,|ui+1|}1m_{i}=\max\{|u_{i}|,|u_{i+1}|\}-1. Now we divide each subloop uisi+1ui+11,i=1,,n2u_{i}s_{i+1}u_{i+1}^{-1},i=1,\dots,n-2 into smaller subloops wi,j,j=1,,miw_{i,j},j=1,\dots,m_{i}: if j=1j=1, then wi,1=ui,1vi,1ui+1,11w_{i,1}=u_{i,1}v_{i,1}u_{i+1,1}^{-1}, if 2jmi12\leqslant j\leqslant m_{i}-1, then wi,j=ui,jvi,jui+1,j1vi,j11w_{i,j}=u_{i,j}v_{i,j}u_{i+1,j}^{-1}v_{i,j-1}^{-1} and if j=mij=m_{i}, then wi,mi=ui,misi+1ui+1,mi1vi,mi1w_{i,m_{i}}=u_{i,m_{i}}s_{i+1}u_{i+1,m_{i}}^{-1}v_{i,m_{i}}^{-1}. Therefore, for the area 𝒜(w)\mathcal{A}(w) we have: 𝒜(w)𝒜(s1u11)+𝒜(un1sn)+i=1n2j=1mi𝒜(wi,j)\mathcal{A}(w)\leqslant\mathcal{A}(s_{1}u_{1}^{-1})+\mathcal{A}(u_{n-1}s_{n})+\sum\limits_{i=1}^{n-2}\sum\limits_{j=1}^{m_{i}}\mathcal{A}(w_{i,j}).

Let s(n)s(n) be the function defined by the equation (1) for the normal form LL and the set of generators AA. We have: |wi,j|s(j)+s(j1)+2|w_{i,j}|\leqslant s(j)+s(j-1)+2 for 1jmi11\leqslant j\leqslant m_{i}-1 and |wi,mi|3+s(j1)|w_{i,m_{i}}|\leqslant 3+s(j-1) for j=mij=m_{i}. Since the normal form LL is quasigeodesic, there exists a positive integer DD such that DnmiDn\geqslant m_{i} for all i=1,,n2i=1,\dots,n-2.

If s(n)s(n) is an unbounded function we may assume that s(D)max{|s1u11|,|un1sn|}s(D)\geqslant\max\{|s_{1}u_{1}^{-1}|,|u_{n-1}s_{n}|\}; in particular, s(D)1s(D)\geqslant 1. The total number of terms in the expression 𝒜(s1u11)+𝒜(un1sn)+i=1n2j=1mi𝒜(wi,j)\mathcal{A}(s_{1}u_{1}^{-1})+\mathcal{A}(u_{n-1}s_{n})+\sum\limits_{i=1}^{n-2}\sum\limits_{j=1}^{m_{i}}\mathcal{A}(w_{i,j}) is at most Dn2Dn^{2} and each term is bounded from above by 4s(Dn)4s(Dn). Therefore, 𝒜(w)Dn2δ(4s(Dn))\mathcal{A}(w)\leqslant Dn^{2}\delta(4s(Dn)) which implies that δ(n)Dn2δ(4s(Dn))\delta(n)\leqslant Dn^{2}\delta(4s(Dn)) for all nn. Since the normal form LL satisfies the f(n)f(n)–fellow traveler property, we have that sfs\preceq f which implies that there exists a constant C0>0C_{0}>0 and n00n_{0}\geqslant 0 such that s(n)C0f(C0n)s(n)\leqslant C_{0}f(C_{0}n) for all nn0n\geqslant n_{0}. Therefore, δ(n)Dn2δ(4s(Dn))Dn2δ(4C0f(C0Dn))\delta(n)\leqslant Dn^{2}\delta(4s(Dn))\leqslant Dn^{2}\delta(4C_{0}f(C_{0}Dn)). Let C=max{4C0,C0D}C=\max\{4C_{0},C_{0}D\}. Then we have that δ(n)Dn2δ(Cf(Cn))\delta(n)\leqslant Dn^{2}\delta(Cf(Cn)) for all nn0n\geqslant n_{0}.

If the function s(n)s(n) is bounded from above by a constant, then we immediately obtain that δ(n)=O(n2)\delta(n)=O(n^{2}). In particular, one can always get that δ(n)Dn2δ(Cf(Cn))\delta(n)\leqslant Dn^{2}\delta(Cf(Cn)) for all nn0n\geqslant n_{0} for some integer constants C,D>0C,D>0 and n00n_{0}\geqslant 0. ∎

Theorem 14.

If the Dehn function of a group G=A|RG=\langle A\,|\,R\rangle with respect to the presentation A|R\langle A\,|\,R\rangle is strongly–super–polynomial, then GG does not admit a quasigeodesic normal form that satisfies the f(n)f(n)–fellow traveler property.

Proof.

Let δ(n)\delta(n) be the Dehn function of a group GG with respect to a presentation A|R\langle A\,|\,R\rangle. Since the Dehn function δ(n)\delta(n) is strongly–super–polynomial, n2δ(n)δ(n)n^{2}\delta(n)\ll\delta(n). Therefore, there exist an unbounded function t(n)t(n)\in\mathcal{F} and constants K,M,N0K,M,N_{0} such that for all nN0n\geqslant N_{0}:

n2δ(n)t(n)Kδ(Mn).n^{2}\delta(n)t(n)\leqslant K\delta(Mn). (3)

We prove the theorem by contradiction. Assume that GG admits a quasigeodesic normal form satisfying the f(n)f(n)–fellow traveller property for some function f𝔦f\ll\mathfrak{i}. By Lemma 13 there exist integer constants C,D>0C,D>0 and N10N_{1}\geqslant 0 such that for all nN1n\geqslant N_{1}:

δ(n)Dn2δ(Cf(Cn)).\delta(n)\leqslant Dn^{2}\delta(Cf(Cn)). (4)

Let N2=max{N0,N1}N_{2}=\max\{N_{0},N_{1}\}. By the inequalities (3) and (4) we obtain that for all nN2n\geqslant N_{2}:

n2δ(n)t(n)Kδ(Mn)DKM2n2δ(Cf(CMn)).n^{2}\delta(n)t(n)\leqslant K\delta(Mn)\leqslant DKM^{2}n^{2}\delta(Cf(CMn)).

Therefore, δ(n)t(n)DKM2δ(Cf(CMn))\delta(n)t(n)\leqslant DKM^{2}\delta(Cf(CMn)) for all nN2n\geqslant N_{2}. Let N3=min{n| 2DKM2t(n)}N_{3}=\min\{n\,|\,2DKM^{2}\leqslant t(n)\} and N4=max{N2,N3}N_{4}=\max\{N_{2},N_{3}\}. Then, 2δ(n)δ(Cf(CMn))2\delta(n)\leqslant\delta(Cf(CMn)) for all nN4n\geqslant N_{4}.

Let N5=min{n|δ(n)1}N_{5}=\min\{n\,|\,\delta(n)\geqslant 1\} and N6=max{N4,N5}N_{6}=\max\{N_{4},N_{5}\}. Then, nCf(CMn)n\leqslant Cf(CMn) for all nN6n\geqslant N_{6} as if, otherwise, n>Cf(CMn)n>Cf(CMn) for some nN6n\geqslant N_{6}, then 2δ(n)δ(Cf(CMn))δ(n)2\delta(n)\leqslant\delta(Cf(CMn))\leqslant\delta(n) which leads to a contradiction with the inequality δ(n)1\delta(n)\geqslant 1. Since f𝔦f\ll\mathfrak{i}, there exists an unbounded function τ(n)\tau(n)\in\mathcal{F} and constants E,N7E,N_{7} such that f(CMn)τ(CMn)Enf(CMn)\tau(CMn)\leqslant En for all nN7n\geqslant N_{7}. Let N8=max{N6,N7}N_{8}=\max\{N_{6},N_{7}\}. From the inequalities nCf(CMn)n\leqslant Cf(CMn) and f(CMn)τ(CMn)Enf(CMn)\tau(CMn)\leqslant En we conclude that nτ(CMn)CEnn\tau(CMn)\leqslant CEn for all nN8n\geqslant N_{8}. As the function τ(n)\tau(n) is unbounded, we get a contradiction. ∎

Non–finitely presented groups. Now we assume that GG is a non–finitely presented group. In this case no additional assumptions are needed to show non–existence Theorem 15. Equivalently, Theorem 15 claims that if GG is a finitely generated group admitting a quasigeodesic normal form with the f(n)f(n)–fellow traveler property, then GG is finitely presented.

Theorem 15.

If GG is a non–finitely presented group, then GG does not admit a quasigeodesic normal form that satisfies the f(n)f(n)–fellow traveller property.

Proof.

First we notice that there exist infinitely many words wSw\in S^{*} for which π(w)=e\pi(w)=e such that for every decomposition of ww as the product w=j=1krj1ujrjw=\prod\limits_{j=1}^{k}r_{j}^{-1}u_{j}r_{j}, where rj,ujSr_{j},u_{j}\in S^{*} and π(uj)=e\pi(u_{j})=e for j=1,,kj=1,\dots,k, for some 1mk1\leqslant m\leqslant k the length of umu_{m} is greater than or equal to the length of ww: |um||w||u_{m}|\geqslant|w|. Indeed, if such infinitely many words did not exist, the group GG would be finitely presented. We denote the set of such words by WW.

We prove the theorem by contradiction. Assume that GG admits a quasigeodesic normal form satisfying the f(n)f(n)–fellow traveller property for some function f𝔦f\ll\mathfrak{i}. Let LSL\subseteq S^{*} be a language defining such normal form in GG. Let w=s1snw=s_{1}\dots s_{n} be a word from the set WW. For each i=1,,n1i=1,\dots,n-1, let uiLu_{i}\in L be the normal form of the group element gi=π(s1si)g_{i}=\pi(s_{1}\dots s_{i}). Similarly to Lemma 13, we divide a loop ww into subloops: s1u11s_{1}u_{1}^{-1}, u1s2u21u_{1}s_{2}u_{2}^{-1}, …, un2sn1un11u_{n-2}s_{n-1}u_{n-1}^{-1}, un1snu_{n-1}s_{n}. For a given i{1,,n1}i\in\{1,\dots,n-1\} and j1j\geqslant 1 we denote by ui,jSu_{i,j}\in S the jjth symbol in the normal form ujSu_{j}\in S^{*}, if j|ui|j\geqslant|u_{i}|, and ui,j=eu_{i,j}=e, if j>|ui|j>|u_{i}|. For a given i{1,,n2}i\in\{1,\dots,n-2\} and j1j\geqslant 1, we denote by vi,jSv_{i,j}\in S^{*} a shortest path connecting gi,j=π(ui,1ui,j)g_{i,j}=\pi(u_{i,1}\dots u_{i,j}) and gi+1,j=π(ui+1,1ui+1,j)g_{i+1,j}=\pi(u_{i+1,1}\dots u_{i+1,j}).

For a given i{1,,n2}i\in\{1,\dots,n-2\}, let i=max{|ui|,|ui+1|}1\ell_{i}=\max\{|u_{i}|,|u_{i+1}|\}-1. Similarly to the proof of Lemma 13, let us divide each subloop uisi+1ui+11u_{i}s_{i+1}u_{i+1}^{-1}, i=1,,n2i=1,\dots,n-2 into smaller subloops wi,j,j=1,,iw_{i,j},j=1,\dots,\ell_{i}: if j=1j=1, then wi,1=ui,1vi,1ui+1,11w_{i,1}=u_{i,1}v_{i,1}u_{i+1,1}^{-1}, if 2ji12\leqslant j\leqslant\ell_{i}-1, then wi,j=ui,jvi,jui+1,j1vi,j11w_{i,j}=u_{i,j}v_{i,j}u_{i+1,j}^{-1}v_{i,j-1}^{-1} and if j=ij=\ell_{i}, then wi,i=ui,isi+1ui+1,i1vi,i1w_{i,\ell_{i}}=u_{i,\ell_{i}}s_{i+1}u_{i+1,\ell_{i}}^{-1}v_{i,\ell_{i}}^{-1}.

Let s(n)s(n) be the function defined by the equation (1) for the normal form LL and the set of generators AA. Let =max{i|i=1,,n2}\ell=\max\{\ell_{i}\,|\,i=1,\dots,n-2\}. Then, |wi,j|4s()|w_{i,j}|\leqslant 4s(\ell) for all i=1,,n2i=1,\dots,n-2 and 1ji1\leqslant j\leqslant\ell_{i}, where we assume that \ell is big enough so s()1s(\ell)\geqslant 1. Since the normal form is quasigeodesic, there exists an integer constant C>0C>0 for which iCn\ell_{i}\leqslant Cn for all i=1,,n2i=1,\dots,n-2, so Cn\ell\leqslant Cn.

As sfs\preceq f for some f𝔦f\ll\mathfrak{i}, there exists an unbounded function tt\in\mathcal{F} for which s(n)t(n)𝔦s(n)t(n)\preceq\mathfrak{i}. In particular, s(n)18Cns(n)\leqslant\frac{1}{8C}n for all nNn\geqslant N for some NN. Therefore, if nNn\geqslant N, then |wi,j|n2|w_{i,j}|\leqslant\frac{n}{2} for all i=1,,n2i=1,\dots,n-2 and 1ji1\leqslant j\leqslant\ell_{i}. Furthermore, we may assume that nn is big enough so |s1u11|,|un1sn|n2|s_{1}u_{1}^{-1}|,|u_{n-1}s_{n}|\leqslant\frac{n}{2}. Since n2<n\frac{n}{2}<n and wWw\in W, we get a contradiction. ∎

4.2 Relation with a Cayley distance function

We find relation with the notion of a Cayley distance function studied in [1, 2, 3]. A Cayley distance function h:+h:\mathbb{N}\rightarrow\mathbb{R}_{+} is defined for an arbitrary bijection ψ:LG\psi:L\rightarrow G between a language LSL\subseteq S^{*} and a group GG by the following identity:

h(n)=max{dA(ψ(w),π(w))|wLn}ifLn,h(n)=\max\{d_{A}(\psi(w),\pi(w))\,|\,w\in L^{\leqslant n}\}\,\,\mathrm{if}\,\,L^{\leqslant n}\neq\varnothing,

where Ln={w||w|n}L^{\leqslant n}=\{w\,|\,|w|\leqslant n\} is the set of words in LL of length less than or equal to nn, and h(n)=0h(n)=0 if Ln=L^{\leqslant n}=\varnothing.

Cayley automatic groups were introduced by Kharlampovich, Khoussainov and Miasnikov [10]. We recall that a group GG is called Cayley automatic if there exists a bijection ψ:LG\psi:L\rightarrow G for which LL is a regular language and for each aAa\in A the relation Ra={(u1,u2)L×L|ψ(u1)a=ψ(u2)}R_{a}=\{(u_{1},u_{2})\in L\times L\,|\,\psi(u_{1})a=\psi(u_{2})\} is recognized by a two–tape synchronous automaton. The bijection ψ:LG\psi:L\rightarrow G is referred to as a Cayley automatic representation of GG. Cayley automatic groups extend automatic groups retaining exactly the same computational model but allowing an arbitrary bijection ψ:LG\psi:L\rightarrow G, not only a canonical mapping π:LG\pi:L\rightarrow G like in the notion of an automatic group.

In [3] it is asked if there exists a Cayley automatic representation of a non–automatic group GG such that for the Cayley distance function h:+h:\mathbb{N}\rightarrow\mathbb{R}_{+} the inequality h𝔦h\prec\mathfrak{i} holds. This problem can be slightly narrowed by requiring that h𝔦h\ll\mathfrak{i}. Theorem 16 shows that if such Cayley automatic representation exists, then GG admits a quasigeodesic normal form satisfying the h(n)h(n)–fellow traveler property.

Theorem 16.

If a non–automatic group GG has a Cayley automatic representation ψ:LG\psi:L\rightarrow G with the Cayley distance function h𝔦h\ll\mathfrak{i}, then there exists a quasigeodesic normal form LSL^{\prime}\subseteq S^{*} that satisfies the h(n)h(n)–fellow traveler property.

Proof.

First we describe how to construct a normal form LL^{\prime} from a given Cayley automatic representation ψ:LG\psi:L\rightarrow G. For a given uLu\in L let vSv\in S^{*} be a word corresponding to a shortest path between π(u)\pi(u) and ψ(u)\psi(u) in the Cayley graph Γ(G,A)\Gamma(G,A) for which ψ(u)=π(uv)\psi(u)=\pi(uv). We define the language LSL^{\prime}\subseteq S^{*} as L={uv|uL}L^{\prime}=\{uv\,|\,u\in L\}. Below we prove that LL^{\prime} defines a quasigeodesic normal form that satisfies the h(n)h(n)–fellow traveler property.

We now show that the normal form defined by LL^{\prime} is quasigeodesic. Let g=ψ(u)g=\psi(u) for some uLu\in L and w=uvLw=uv\in L^{\prime} be the normal form of gg. We need to show that |uv|CdA(g)+C|uv|\leqslant Cd_{A}(g)+C for some constant C>0C>0. Since ψ:LG\psi:L\rightarrow G is a Cayley automatic representation, there exists a constant C1>0C_{1}>0 for which |u|C1(dA(g)+1)|u|\leqslant C_{1}(d_{A}(g)+1). Let n=|u|n=|u|. Then |v|h(n)|v|\leqslant h(n). Therefore, |w|=|u|+|v|n+h(n)|w|=|u|+|v|\leqslant n+h(n). Since h𝔦h\ll\mathfrak{i}, then h(n)C2nh(n)\leqslant C_{2}n for some constant C2>0C_{2}>0 and all nN0n\geqslant N_{0}. Let C3=C2+1C_{3}=C_{2}+1. Then we have that |w|C3nC3C1(dA(g)+1)|w|\leqslant C_{3}n\leqslant C_{3}C_{1}(d_{A}(g)+1) for all nN0n\geqslant N_{0}. Therefore, there exists a constant C>0C>0 such that |w|C(dA(g)+1)|w|\leqslant C(d_{A}(g)+1) for all n0n\geqslant 0.

Refer to caption
Figure 5: The curves labeled u1u_{1} and u2u_{2} correspond to the words u1,u2Lu_{1},u_{2}\in L for which ψ(u1)=g1\psi(u_{1})=g_{1} and ψ(u2)=g2\psi(u_{2})=g_{2}. The dashed curves v1v_{1} and v2v_{2} show shortest paths between π(u1)\pi(u_{1}) and g1g_{1} and π(u2)\pi(u_{2}) and g2g_{2}, respectively. The directed edge labeled by aa leads from g1g_{1} to g2g_{2}.

Let us show that LL^{\prime} defines a normal form in GG that satisfies the h(n)h(n)–fellow traveler property. Let w1=u1v1Lw_{1}=u_{1}v_{1}\in L^{\prime} and w2=u2v2Lw_{2}=u_{2}v_{2}\in L^{\prime} be the normal forms of the group elements g1=π(w1)g_{1}=\pi(w_{1}) and g2=π(w2)g_{2}=\pi(w_{2}) for which g1a=g2g_{1}a=g_{2} for some aAa\in A. See Figure 5 for illustration. We denote by mm the minimum m=min{|u1|,|u2|}m=\min\{|u_{1}|,|u_{2}|\}. Let s(n)s(n) be the function defined by the equation (1) for the normal form LL^{\prime} and the set of generators AA. It can be shown that s(n)2h(n)+C0s(n)\leqslant 2h(n)+C_{0} for all nmn\leqslant m and some constant C0>0C_{0}>0; this is proved in details for all n0n\geqslant 0 in [2, Theorem 2.1]. If nmn\geqslant m, then by the triangle inequality dA(π(w1(n)),π(w2(n)))|v1|+|v2|+||u1||u2||+1d_{A}(\pi(w_{1}(n)),\pi(w_{2}(n)))\leqslant|v_{1}|+|v_{2}|+||u_{1}|-|u_{2}||+1. Since the representation ψ:LG\psi:L\rightarrow G is Cayley automatic, ||u1||u2||C2||u_{1}|-|u_{2}||\leqslant C_{2} for some constant C2C_{2}. Let C3=C2+1C_{3}=C_{2}+1. Therefore, for all nmn\geqslant m:

dA(π(w1(n)),π(w2(n)))|v1|+|v2|+C3h(|u1|)+h(|u2|)+C3h(m)+h(m+C2)+C32h(m+C2)+C32h(n+C2)+C3.\begin{split}d_{A}(\pi(w_{1}(n)),\pi(w_{2}(n)))\leqslant|v_{1}|+|v_{2}|+C_{3}\leqslant h(|u_{1}|)+h(|u_{2}|)+C_{3}\leqslant\\ h(m)+h(m+C_{2})+C_{3}\leqslant 2h(m+C_{2})+C_{3}\leqslant 2h(n+C_{2})+C_{3}.\end{split}

Let C4=max{C0,C3}C_{4}=\max\{C_{0},C_{3}\}. Then, for all n0n\geqslant 0, we have that dA(π(w1(n)),π(w2(n)))2h(n+C2)+C4d_{A}(\pi(w_{1}(n)),\pi(w_{2}(n)))\leqslant 2h(n+C_{2})+C_{4}. Let f(n)=2h(n+C2)+C4f(n)=2h(n+C_{2})+C_{4}. Since f(n)h(n)f(n)\preceq h(n), we get that s(n)h(n)s(n)\preceq h(n). Therefore, LL^{\prime} defines a normal form of GG satisfying the h(n)h(n)–fellow traveler property. ∎

5 Quasiregular Normal Forms

This section discusses quasiregular normal forms in the context of the f(n)f(n)–fellow traveler property. By Theorems 14 and 15 for groups with the strongly–super–polynomial Dehn function and non–finitely presented groups there exist no quasigeodesic normal forms satisfying the f(n)f(n)–fellow traveler property. In this section we show examples of quasiregular normal forms satisfying the f(n)f(n)–fellow traveler property for such groups.

5.1 Baumslag–Solitar groups

We consider a family of the Baumslag–Solitar groups BS(p,q)=a,t|tapt1=aqBS(p,q)=\langle a,t\,|\,ta^{p}t^{-1}=a^{q}\rangle for 1p<q1\leqslant p<q. Each group of this family has the exponential Dehn function, so by Theorem 14 it does not admit a quasigeodesic normal form satisfying the f(n)f(n)–fellow traveler property. In Theorem 17 we will show that each group of this family admits a quasiregular normal form satisfying the log(n)\log(n)–fellow traveler property.

Theorem 17.

Each group BS(p,q)BS(p,q) for 1p<q1\leqslant p<q admits a quasiregular normal form satisfying the log(n)\log(n)–fellow traveler property.

Proof.

Every group element gBS(p,q)g\in BS(p,q) for 1p<q1\leqslant p<q can be uniquely written as a freely reduced word over the alphabet {a±1,t±1}\{a^{\pm 1},t^{\pm 1}\} of the form:

w=wtεw1tε1ak,w=w_{\ell}t^{\varepsilon_{\ell}}\dots w_{1}t^{\varepsilon_{1}}a^{k}, (5)

where εi{+1,1}\varepsilon_{i}\in\{+1,-1\}, wi{ϵ,a,,ap1}w_{i}\in\{\epsilon,a,\dots,a^{p-1}\} if εi=1\varepsilon_{i}=-1, wi{ϵ,a,,aq1}w_{i}\in\{\epsilon,a,\dots,a^{q-1}\} if εi=+1\varepsilon_{i}=+1 and kk\in\mathbb{Z}. So the identity (5) defines a normal form in BS(p,q)BS(p,q). This normal form is prefix–closed, so it is quasiregular. Below we show that it satisfies the log(n)\log(n)–fellow traveler property.

Let A={a,t}A=\{a,t\} and dAd_{A} be the word metric in BS(p,q)BS(p,q) with respect to the generators aa and tt. Burillo and Elder [6] showed that there exist constants C1,C2,D1,D2>0C_{1},C_{2},D_{1},D_{2}>0 such that for all gBS(p,q)g\in BS(p,q):

C1(+log(|k|+1))D1dA(g)C2(+log(|k|+1))+D2.C_{1}(\ell+\log(|k|+1))-D_{1}\leqslant d_{A}(g)\leqslant C_{2}(\ell+\log(|k|+1))+D_{2}. (6)

We first consider a pair of group elements gg and gaga. Let waw_{a} be the normal form of gaga. Then wa=wtεw1tε1akw_{a}=w_{\ell}t^{\varepsilon_{\ell}}\dots w_{1}t^{\varepsilon_{1}}a^{k}. Clearly, dA(π(w(n)),π(wa(n)))1d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 1 for all n0n\geqslant 0.

Now we consider a pair of group elements gg and gtgt. We denote by wtw_{t} the normal form of gtgt. Let k=mq+rk=mq+r, where mm\in\mathbb{Z} and r{0,,q1}r\in\{0,\dots,q-1\}. We have three different cases:

  • Suppose r0r\neq 0. Then wt=wtεw1tε1artampw_{t}=w_{\ell}t^{\varepsilon_{\ell}}\dots w_{1}t^{\varepsilon_{1}}a^{r}ta^{mp}. Let u=wtεw1tε1u=w_{\ell}t^{\varepsilon_{\ell}}\dots w_{1}t^{\varepsilon_{1}}. If n|u|n\leqslant|u|, then dA(π(w(n)),π(wt(n)))=0d_{A}(\pi(w(n)),\pi(w_{t}(n)))=0. If |u|<n|u|+r+1|u|<n\leqslant|u|+r+1, then dA(π(w(n)),π(wt(n)))2r+2d_{A}(\pi(w(n)),\pi(w_{t}(n)))\leqslant 2r+2. If n>|u|+r+1n>|u|+r+1, then dA(π(w(n)),π(wt(n)))dA(artai)+dA(aj)d_{A}(\pi(w(n)),\pi(w_{t}(n)))\leqslant d_{A}(a^{r}ta^{i})+d_{A}(a^{j}), where i=min{|mp|,n(|u|+r+1)}sign(k)i=\min\{|mp|,n-(|u|+r+1)\}*\mathrm{sign}(k) and j=min{|k|,n|u|}sign(k)j=\min\{|k|,n-|u|\}*\mathrm{sign}(k). By (6), dA(artai)C2(1+log(|i|+1))+D2d_{A}(a^{r}ta^{i})\leqslant C_{2}(1+\log(|i|+1))+D_{2} and dA(aj)C2log(|j|+1)+D2d_{A}(a^{j})\leqslant C_{2}\log(|j|+1)+D_{2}. Therefore, dA(π(w(n)),π(wt(n)))2C2log(n+1)+C2+2D2d_{A}(\pi(w(n)),\pi(w_{t}(n)))\leqslant 2C_{2}\log(n+1)+C_{2}+2D_{2}.

  • Suppose r=0r=0 and either ε1=1\varepsilon_{1}=1 or =0\ell=0. Then wt=wtεw1tε1tampw_{t}=w_{\ell}t^{\varepsilon_{\ell}}\dots w_{1}t^{\varepsilon_{1}}ta^{mp}. Let u=wtεw1tε1u=w_{\ell}t^{\varepsilon_{\ell}}\dots w_{1}t^{\varepsilon_{1}}. If n|u|n\leqslant|u|, then dA(π(w(n)),π(wt(n)))=0d_{A}(\pi(w(n)),\pi(w_{t}(n)))=0. If n>|u|n>|u|, then dA(π(w(n)),π(wt(n)))dA(tai)+dA(aj)d_{A}(\pi(w(n)),\pi(w_{t}(n)))\leqslant d_{A}(ta^{i})+d_{A}(a^{j}), where i=min{|mp|,n|u|1}sign(k)i=\min\{|mp|,n-|u|-1\}*\mathrm{sign}(k) and j=min{|k|,n|u|}sign(k)j=\min\{|k|,n-|u|\}*\mathrm{sign}(k). By (6), dA(tai)C2(1+log(|i|+1))+D2d_{A}(ta^{i})\leqslant C_{2}(1+\log(|i|+1))+D_{2} and dA(aj)C2log(|j|+1)+D2d_{A}(a^{j})\leqslant C_{2}\log(|j|+1)+D_{2}. Therefore, dA(π(w(n)),π(wt(n)))2C2log(n+1)+C2+2D2d_{A}(\pi(w(n)),\pi(w_{t}(n)))\leqslant 2C_{2}\log(n+1)+C_{2}+2D_{2}.

  • Suppose r=0r=0, ε1=1\varepsilon_{1}=-1 and 1\ell\geqslant 1. Then wt=wtεw2tε2w1ampw_{t}=w_{\ell}t^{\varepsilon_{\ell}}\dots w_{2}t^{\varepsilon_{2}}w_{1}a^{mp}. Let v=wtεw2tε2w1v=w_{\ell}t^{\varepsilon_{\ell}}\dots w_{2}t^{\varepsilon_{2}}w_{1}. If n|v|n\leqslant|v|, then dA(π(w(n)),π(wt(n)))=0d_{A}(\pi(w(n)),\pi(w_{t}(n)))=0. If n>|v|n>|v|, then dA(π(w(n)),π(wt(n)))dA(ai)+dA(t1aj)d_{A}(\pi(w(n)),\pi(w_{t}(n)))\leqslant d_{A}(a^{i})+d_{A}(t^{-1}a^{j}), where i=min{|mp|,n|v|}sign(k)i=\min\{|mp|,n-|v|\}*\mathrm{sign}(k) and j=min{|k|,n|v|1}sign(k)j=\min\{|k|,n-|v|-1\}*\mathrm{sign}(k). By (6), dA(ai)C2log(|i|+1)+D2d_{A}(a^{i})\leqslant C_{2}\log(|i|+1)+D_{2} and dA(t1aj)C2(1+log(|j|+1))+D2d_{A}(t^{-1}a^{j})\leqslant C_{2}(1+\log(|j|+1))+D_{2}. Therefore, dA(π(w(n)),π(wt(n)))2C2log(n+1)+C2+2D2d_{A}(\pi(w(n)),\pi(w_{t}(n)))\leqslant 2C_{2}\log(n+1)+C_{2}+2D_{2}.

From these three cases we can see that s(n)log(n)s(n)\preceq\log(n). Therefore, the normal form given by the identity (5) satisfies the log(n)\log(n)–fellow traveler property. ∎

Remark 18.

We note that for the normal form in the proof of Theorem 17 the upper bound s(n)log(n)s(n)\preceq\log(n) is sharp. Indeed, let w=amq2w=a^{mq^{2}} for m>0m>0. Then wt=tampqw_{t}=ta^{mpq}. For n=mqn=mq we have that dA(π(w(n)),π(wt(n)))=dA(amq,tamq1)=dA(amqtamq1)=dA(tam(qp)1)d_{A}(\pi(w(n)),\pi(w_{t}(n)))=d_{A}(a^{mq},ta^{mq-1})=d_{A}(a^{-mq}ta^{mq-1})=d_{A}(ta^{m(q-p)-1}). By (6), dA(tam(qp)1)C1(1+log|m(qp)|)D1=C1(1log(qqp)+log(mq))D1d_{A}(ta^{m(q-p)-1})\geqslant C_{1}(1+\log|m(q-p)|)-D_{1}=C_{1}\left(1-\log\left(\frac{q}{q-p}\right)+\log(mq)\right)-D_{1}. Therefore, for n=mqn=mq we have that:

dA(π(w(n)),π(wt(n)))C1log(n)(D1+C1log(qqp)C1).d_{A}(\pi(w(n)),\pi(w_{t}(n)))\geqslant C_{1}\log(n)-\left(D_{1}+C_{1}\log\left(\frac{q}{q-p}\right)-C_{1}\right).

Therefore, by the triangle inequality we get that for all nn:

dA(π(w(n)),π(wt(n)))C1log(n)(D1+C1log(qqp)C1+2q).d_{A}(\pi(w(n)),\pi(w_{t}(n)))\geqslant C_{1}\log(n)-\left(D_{1}+C_{1}\log\left(\frac{q}{q-p}\right)-C_{1}+2q\right).

This implies that log(n)s(n)\log(n)\preceq s(n).

Remark 19.

Let us be given a normal form of BS(p,q)BS(p,q) for 1p<q1\leqslant p<q satisfying the f(n)f(n)–fellow traveler property. We denote by ma,ma1,mtm_{a},m_{a^{-1}},m_{t} and mt1m_{t^{-1}} the functions which send the normal form of a group element gBS(p,q)g\in BS(p,q) to the normal form of a group element ga,ga1,gtga,ga^{-1},gt and gt1gt^{-1}, respectively. One can notice that the functions ma,ma1,mtm_{a},m_{a^{-1}},m_{t} and mt1m_{t^{-1}} cannot be all computed in o(nlogn)o(n\log n) time on a one–tape Turing machine. Indeed, suppose each of the functions ma,ma1,mtm_{a},m_{a^{-1}},m_{t} and mt1m_{t^{-1}} is computed on a one–tape Turing machine in o(nlogn)o(n\log n) time. Hartmanis [9] and, independently, Trachtenbrot [12] showed that a language recognized on a one–tape Turing machine in o(nlogn)o(n\log n) time must be regular. This fact and the pumping lemma imply that if each of the functions ma,ma1,mtm_{a},m_{a^{-1}},m_{t} and mt1m_{t^{-1}} is computed on a one–tape Turing machine in o(nlogn)o(n\log n) time, then the normal form satisfies the bounded length difference property, so it is quasigeodesic; for the proof see [11, Theorem 4]. Thus, by Theorem 14, we arrive at a contradiction. However, it can be verified that for the normal form given by the identity (5) the functions ma,ma1,mtm_{a},m_{a^{-1}},m_{t} and mt1m_{t^{-1}} can be computed in O(n)O(n) time using a more powerful computational model – a two–tape Turing machine. Though this verification is not difficult we omit it as it is out of scope of this paper.

5.2 Wreath product 22\mathbb{Z}_{2}\wr\mathbb{Z}^{2}

We consider the wreath product 22=a,b,c|[ai1bj1cai1bj1,ai2bj2cai2bj2]=e,ab=ba,c2=e\mathbb{Z}_{2}\wr\mathbb{Z}^{2}=\langle a,b,c\,|\,[a^{i_{1}}b^{j_{1}}ca^{-i_{1}}b^{-j_{1}},a^{i_{2}}b^{j_{2}}ca^{-i_{2}}b^{-j_{2}}]=e,ab=ba,c^{2}=e\rangle. This group is non–finitely presented, so by Theorem 15 it does not admit a quasigeodesic normal form satisfying the f(n)f(n)–fellow traveler property. In Theorem 20 we will show that 22\mathbb{Z}_{2}\wr\mathbb{Z}^{2} admits a quasiregular normal form satisfying the n\sqrt{n}–fellow traveler property.

Every group element of 22\mathbb{Z}_{2}\wr\mathbb{Z}^{2} can be written as a pair (φ,z)(\varphi,z), where φ:22\varphi:\mathbb{Z}^{2}\rightarrow\mathbb{Z}_{2} is a function such that φ(ξ)\varphi(\xi) is not equal to the identity for at most finitely many ξ2\xi\in\mathbb{Z}^{2} and z2z\in\mathbb{Z}^{2}. For the group 2={(x,y)|x,y}\mathbb{Z}^{2}=\{(x,y)\,|\,x,y\in\mathbb{Z}\} we denote by aa and bb the generators a=(1,0)a=(1,0) and b=(0,1)b=(0,1). The group 2\mathbb{Z}^{2} is canonically embedded in 22\mathbb{Z}_{2}\wr\mathbb{Z}^{2} by mapping ξ2\xi\in\mathbb{Z}^{2} to (φe,ξ)(\varphi_{e},\xi), where φe:22\varphi_{e}:\mathbb{Z}^{2}\rightarrow\mathbb{Z}_{2} sends every element of 2\mathbb{Z}^{2} to the identity e2e\in\mathbb{Z}_{2}. We denote by cc the nontrivial element of 2\mathbb{Z}_{2}. The group 2\mathbb{Z}_{2} is canonically embedded in 22\mathbb{Z}_{2}\wr\mathbb{Z}^{2} by mapping cc to (φc,(0,0))(\varphi_{c},(0,0)), where φc:22\varphi_{c}:\mathbb{Z}^{2}\rightarrow\mathbb{Z}_{2} is a function such that φc(ξ)=e\varphi_{c}(\xi)=e if ξ(0,0)\xi\neq(0,0) and φc((0,0))=c\varphi_{c}((0,0))=c. We will identify a,ba,b and cc with the group elements (φe,a)(\varphi_{e},a), (φe,b)(\varphi_{e},b) and (φc,(0,0))(\varphi_{c},(0,0)) in 22\mathbb{Z}_{2}\wr\mathbb{Z}^{2}, respectively. The set A={a,b,c}A=\{a,b,c\} generates the group 22\mathbb{Z}_{2}\wr\mathbb{Z}^{2}.

Theorem 20.

The wreath product 22\mathbb{Z}_{2}\wr\mathbb{Z}^{2} admits a quasiregular normal form satisfying the n\sqrt{n}–fellow traveler property.

Proof.

Let Γ\Gamma be the infinite directed graph shown in Fig. 6 which is isomorphic to (;S)\left(\mathbb{N};\mathrm{S}\right), where S\mathrm{S} is the successor function S(n)=n+1\mathrm{S}(n)=n+1. The vertices of Γ\Gamma are identified with elements of 2\mathbb{Z}^{2}, each vertex in V(Γ){(0,0)}V(\Gamma)\setminus\{(0,0)\} has exactly one ingoing and one outgoing edges and the vertex (0,0)(0,0) has one outgoing edge and no ingoing edges. Let τ:2\tau:\mathbb{N}\rightarrow\mathbb{Z}^{2} be the mapping such that τ(0)=(0,0)\tau(0)=(0,0) and, for k>0k>0, τ(k)=(x,y)\tau(k)=(x,y) is the end vertex of a directed path in Γ\Gamma of length kk which starts in the vertex (0,0)(0,0).

Refer to caption
Figure 6: An infinite digraph Γ\Gamma and an element h22h\in\mathbb{Z}_{2}\wr\mathbb{Z}^{2}.

Now we define a normal form in the group 22\mathbb{Z}_{2}\wr\mathbb{Z}^{2} as follows. Let g=(φ,z)22g=(\varphi,z)\in\mathbb{Z}_{2}\wr\mathbb{Z}^{2}, where φ:22\varphi:\mathbb{Z}^{2}\rightarrow\mathbb{Z}_{2} and z2z\in\mathbb{Z}^{2}. Let u0=cu_{0}=c if φ(τ(0))=c\varphi(\tau(0))=c and u0=εu_{0}=\varepsilon if φ(τ(0))=e\varphi(\tau(0))=e. We denote by mm the maximum m=max{i|φ(τ(i))=c}m=\max\{i\,|\,\varphi(\tau(i))=c\}; if φ(ξ)=e\varphi(\xi)=e for all ξ\xi then we put m=0m=0. For a given integer i[1,m]i\in\left[1,m\right], let αi=a,b,a1\alpha_{i}=a,b,a^{-1} and b1b^{-1} if τ(i)τ(i1)\tau(i)-\tau(i-1) is equal to (1,0)(1,0), (0,1)(0,1), (1,0)(-1,0) and (0,1)(0,-1), respectively, and let βi=c\beta_{i}=c if φ(τ(i))=c\varphi(\tau(i))=c and βi=ε\beta_{i}=\varepsilon if φ(τ(i))=e\varphi(\tau(i))=e. Let ui=αiβiu_{i}=\alpha_{i}\beta_{i} for i[1,m]i\in\left[1,m\right] and uu be the concatenation u=u0u1umu=u_{0}u_{1}\dots u_{m}.

Let ll be the integer for which τ(l)=z\tau(l)=z. If l>ml>m, for a given i[1,lm]i\in\left[1,l-m\right] let vi=a,b,a1v_{i}=a,b,a^{-1} and b1b^{-1} if τ(m+i)τ(m+i1)\tau(m+i)-\tau(m+i-1) is equal to (1,0)(1,0), (0,1)(0,1), (1,0)(-1,0) and (0,1)(0,-1), respectively. If l<ml<m, for a given i[1,ml]i\in\left[1,m-l\right] let vi=a,b,a1v_{i}=a,b,a^{-1} and b1b^{-1} if τ(mi)τ(mi+1)\tau(m-i)-\tau(m-i+1) is equal to (1,0)(1,0), (0,1)(0,1), (1,0)(-1,0) and (0,1)(0,-1), respectively. If l=ml=m, let v=εv=\varepsilon. If lml\neq m, let v=v1vkv=v_{1}\dots v_{k}, where k=|lm|k=|l-m|.

Finally we define a normal form ww of the element g=(φ,z)g=(\varphi,z) as a concatenation of uu and vv: w=uvw=uv. Informally speaking, the normal form ww is obtained as follows. Imagine the lamplighter who moves along the graph Γ\Gamma starting from (0,0)(0,0) writing a,a1,ba,a^{-1},b and b1b^{-1} depending whether it moves right, left, up and down, respectively. The lamplighter writes cc if the lamp at the current position (x,y)2(x,y)\in\mathbb{Z}^{2} is lit: φ((x,y))=c\varphi((x,y))=c. After the lamplighter reaches the position f(τ(m))f(\tau(m)) it either moves further along Γ\Gamma, if l>ml>m, or goes back along Γ\Gamma, if l<ml<m, until it reaches the position τ(l)=z\tau(l)=z; while moving it writes a,a1,ba,a^{-1},b and b1b^{-1} depending whether it moves right, left, up and down, respectively. For illustration let us consider the group element h22h\in\mathbb{Z}_{2}\wr\mathbb{Z}^{2} shown in Fig. 6: a black square means that the lamp at the current position is lit, i.e., φ((x,y))=c\varphi((x,y))=c, a white square means that the lamp at the current position is unlit, i.e., φ((x,y))=e\varphi((x,y))=e, and a black circle shows the position of the lamplighter zz and that the lamp at this position is lit, i.e., φ(z)=c\varphi(z)=c. For this group element hh the word uu is as follows:

u=acba1a1b1cb1caaabbba1ca1a1a1b1b1cb1b1aacaaabcbbcbcbca1ca1a1a1a1a1cb1cb1b1b1b1cb1acacaaaaac,\begin{split}u=acba^{-1}a^{-1}b^{-1}cb^{-1}caaabbba^{-1}ca^{-1}a^{-1}a^{-1}b^{-1}b^{-1}cb^{-1}b^{-1}aacaaabcbbcbcbc\\ a^{-1}ca^{-1}a^{-1}a^{-1}a^{-1}a^{-1}cb^{-1}cb^{-1}b^{-1}b^{-1}b^{-1}cb^{-1}acacaaaaac,\end{split}

and the word vv is as follows:

v=a1a1a1a1a1a1a1bbbbbbaaaaaab1b1b1b1b1a1a1a1,v=a^{-1}a^{-1}a^{-1}a^{-1}a^{-1}a^{-1}a^{-1}bbbbbbaaaaaab^{-1}b^{-1}b^{-1}b^{-1}b^{-1}a^{-1}a^{-1}a^{-1},

and the normal form of hh is a concatenation of uu and vv. The described normal form of 22\mathbb{Z}_{2}\wr\mathbb{Z}^{2} is prefix–closed, so it is quasiregular. Let us show that it satisfies the n\sqrt{n}–fellow traveler property.

We start with a pair of group elements g=(φ,z)g=(\varphi,z) and ga=(φ,z)ga=(\varphi,z^{\prime}), where z=z+(1,0)z^{\prime}=z+(1,0). Let ll^{\prime} be the integer for which τ(l)=z\tau(l^{\prime})=z^{\prime}. Let wa=uavaw_{a}=u_{a}v_{a} be the normal form of gaga. We first notice that dA(π(w(n)),π(wa(n)))d_{A}(\pi(w(n)),\pi(w_{a}(n))) can be bounded from above by |ll||l-l^{\prime}| for every nn:

dA(π(w(n)),π(wa(n)))|ll|.d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant|l-l^{\prime}|. (7)

For a given p=(x,y)2p=(x,y)\in\mathbb{Z}^{2}, let r(p)=max{|x|,|y|}r(p)=\max\{|x|,|y|\} and k(p)k(p) be an integer for which τ(k(p))=p\tau(k(p))=p. One can notice the following lower and upper bounds for k(p)k(p):

(2r(p)1)21k(p)(2r(p)+1)21.(2r(p)-1)^{2}-1\leqslant k(p)\leqslant(2r(p)+1)^{2}-1. (8)

In particular, (8) implies that (2r(z)1)21l(2r(z)+1)21(2r(z)-1)^{2}-1\leqslant l\leqslant(2r(z)+1)^{2}-1 and (2r(z)1)21l(2r(z)+1)21(2r(z^{\prime})-1)^{2}-1\leqslant l^{\prime}\leqslant(2r(z^{\prime})+1)^{2}-1. Therefore, |ll|4(r(z)+r(z))|r(z)r(z)+1||l-l^{\prime}|\leqslant 4(r(z^{\prime})+r(z))|r(z^{\prime})-r(z)+1|. Since |r(z)r(z)|1|r(z^{\prime})-r(z)|\leqslant 1, we obtain that |ll|8(r(z)+r(z))|l-l^{\prime}|\leqslant 8(r(z)+r(z^{\prime})). Therefore, by (7) we obtain that for every nn:

dA(π(w(n)),π(wa(n)))16r(z)+8,dA(π(w(n)),π(wa(n)))16r(z)+8.d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 16r(z)+8,\,\,d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 16r(z^{\prime})+8. (9)

Now we notice that u=uau=u_{a}. Therefore, if n|u|n\leqslant|u|, then dA(π(w(n)),π(wa(n)))=0d_{A}(\pi(w(n)),\pi(w_{a}(n)))=0. For n>|u|n>|u| we consider the following three cases:

  • Suppose llml^{\prime}\geqslant l\geqslant m. If |u|<n|u|+(lm)|u|<n\leqslant|u|+(l-m), then dA(π(w(n)),π(wa(n)))=0d_{A}(\pi(w(n)),\pi(w_{a}(n)))=0. By (9) we have that dA(π(w(n)),π(wa(n)))16r(z)+8d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 16r(z)+8. Therefore, by (8), dA(π(w(n)),π(wa(n)))8(l+1+1)+8d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 8(\sqrt{l+1}+1)+8. If n>|u|+(lm)n>|u|+(l-m), then nln\geqslant l. Therefore, dA(π(w(n)),π(wa(n)))8n+1+16d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 8\sqrt{n+1}+16. Similarly, suppose llml\geqslant l^{\prime}\geqslant m. If |u|<n|u|+(lm)|u|<n\leqslant|u|+(l^{\prime}-m), then dA(π(w(n)),π(wa(n)))=0d_{A}(\pi(w(n)),\pi(w_{a}(n)))=0. By (9) we have that dA(π(w(n)),π(wa(n)))16r(z)+8d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 16r(z^{\prime})+8. Therefore, by (8), dA(π(w(n)),π(wa(n)))8(l+1+1)+8d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 8(\sqrt{l^{\prime}+1}+1)+8. If n>|u|+(lm)n>|u|+(l^{\prime}-m), then nln\geqslant l^{\prime}. Therefore, dA(π(w(n)),π(wa(n)))8n+1+16d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 8\sqrt{n+1}+16.

  • Suppose lmll^{\prime}\geqslant m\geqslant l. By (9) we have that dA(π(w(n)),π(wa(n)))16r(z)+8d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 16r(z)+8. Therefore, by (8), dA(π(w(n)),π(wa(n)))8(l+1+1)+8d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 8(\sqrt{l+1}+1)+8. If n>|u|n>|u|, then nln\geqslant l. Therefore, dA(π(w(n)),π(wa(n)))8n+1+16d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 8\sqrt{n+1}+16. Similarly, suppose lmll\geqslant m\geqslant l^{\prime}. By (9) we have that dA(π(w(n)),π(wa(n)))16r(z)+8d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 16r^{\prime}(z)+8. Therefore, by (8), dA(π(w(n)),π(wa(n)))8(l+1+1)+8d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 8(\sqrt{l^{\prime}+1}+1)+8. If n>|u|n>|u|, then nln\geqslant l^{\prime}. Therefore, dA(π(w(n)),π(wa(n)))8n+1+16d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 8\sqrt{n+1}+16.

  • Suppose mllm\geqslant l^{\prime}\geqslant l. If |u|<n|u|+(ml)|u|<n\leqslant|u|+(m-l^{\prime}), then dA(π(w(n)),π(wa(n)))=0d_{A}(\pi(w(n)),\pi(w_{a}(n)))=0. By (9) we have that dA(π(w(n)),π(wa(n)))16r(z)+8d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 16r^{\prime}(z)+8. Therefore, by (8), dA(π(w(n)),π(wa(n)))8(l+1+1)+8d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 8(\sqrt{l^{\prime}+1}+1)+8. If n>|u|+(ml)n>|u|+(m-l^{\prime}), then nln\geqslant l^{\prime}. Therefore, dA(π(w(n)),π(wa(n)))8n+1+16d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 8\sqrt{n+1}+16. Similarly, suppose mllm\geqslant l\geqslant l^{\prime}. If |u|<n|u|+(ml)|u|<n\leqslant|u|+(m-l), then dA(π(w(n)),π(wa(n)))=0d_{A}(\pi(w(n)),\pi(w_{a}(n)))=0. By (9) we have that dA(π(w(n)),π(wa(n)))16r(z)+8d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 16r(z)+8. Therefore, by (8), dA(π(w(n)),π(wa(n)))8(l+1+1)+8d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 8(\sqrt{l+1}+1)+8. If n>|u|+(ml)n>|u|+(m-l), nln\geqslant l. Therefore, dA(π(w(n)),π(wa(n)))8n+1+16d_{A}(\pi(w(n)),\pi(w_{a}(n)))\leqslant 8\sqrt{n+1}+16.

From these three cases we can see that dA(π(w(n)),π(wa(n)))nd_{A}(\pi(w(n)),\pi(w_{a}(n)))\preceq\sqrt{n}.

Now let us consider a pair of group elements g=(φ,z)g=(\varphi,z) and gc=(φ,z)gc=(\varphi^{\prime},z), where φ(γ)=φ(γ)\varphi(\gamma)=\varphi^{\prime}(\gamma) if γz\gamma\neq z and φ(γ)=φ(γ)c\varphi^{\prime}(\gamma)=\varphi(\gamma)c if γ=z\gamma=z. There are two different cases to consider: lml\geqslant m and l<ml<m. Let wc=ucvcw_{c}=u_{c}v_{c} be the normal form of gcgc. The case lml\geqslant m is straightforward: dA(π(w(n)),π(wc(n)))1d_{A}(\pi(w(n)),\pi(w_{c}(n)))\leqslant 1 for all nn.

Now let us consider the case l<ml<m. We denote by u~\widetilde{u} the following prefix of uu: u~=u0u1um1αm\widetilde{u}=u_{0}u_{1}\dots u_{m-1}\alpha_{m}. If n|u~|n\leqslant|\widetilde{u}|, then dA(π(w(n)),π(wc(n)))=0d_{A}(\pi(w(n)),\pi(w_{c}(n)))=0. Suppose now that n>|u~|n>|\widetilde{u}|. We assume that βm=ε\beta_{m}=\varepsilon. Let v~\widetilde{v} be the suffix of w(n)w(n) which follows u~\widetilde{u}: w(n)=u~v~w(n)=\widetilde{u}\widetilde{v}.

Let z0=τ(m)z_{0}=\tau(m) and z1=τ(k)z_{1}=\tau(k) be a position of the lamplighter for a group element π(w(n))\pi(w(n)). We denote by (x0,y0)(x_{0},y_{0}) and (x1,y1)(x_{1},y_{1}) the coordinates of z0z_{0} and z1z_{1}, respectively. There are two different cases: v~=αm+1βm+1βk1αk\widetilde{v}=\alpha_{m+1}\beta_{m+1}\dots\beta_{k-1}\alpha_{k} or v~=αm+1βm+1αkβk\widetilde{v}=\alpha_{m+1}\beta_{m+1}\dots\alpha_{k}\beta_{k}, where βk=c\beta_{k}=c. Let us analyze these two cases:

  • Suppose v~=αm+1βm+1αk\widetilde{v}=\alpha_{m+1}\beta_{m+1}\dots\alpha_{k}. Then wc(n)=u~cαm+1βm+1αk1βk1w_{c}(n)=\widetilde{u}c\alpha_{m+1}\beta_{m+1}\dots\alpha_{k-1}\beta_{k-1}, so we have that:

    π(wc(n))=π(w(n))ax0x1by0y1cax1x0by1y0αk1.\pi(w_{c}(n))=\pi(w(n))a^{x_{0}-x_{1}}b^{y_{0}-y_{1}}ca^{x_{1}-x_{0}}b^{y_{1}-y_{0}}\alpha_{k}^{-1}.

    Therefore, dA(π(w(n)),π(wc(n)))2|x1x0|+2|y1y0|+22(|x1|+|y1|+|x0|+|y0|+1)2(2r(z1)+2r(z0)+1)d_{A}(\pi(w(n)),\pi(w_{c}(n)))\leqslant 2|x_{1}-x_{0}|+2|y_{1}-y_{0}|+2\leqslant 2(|x_{1}|+|y_{1}|+|x_{0}|+|y_{0}|+1)\leqslant 2(2r(z_{1})+2r(z_{0})+1). By (8), 2r(z0)m+1+12r(z_{0})\leqslant\sqrt{m+1}+1 and 2r(z1)k+1+12r(z_{1})\leqslant\sqrt{k+1}+1. Therefore, dA(π(w(n)),π(wc(n)))2(k+1+1+m+1+1+1)d_{A}(\pi(w(n)),\pi(w_{c}(n)))\leqslant 2(\sqrt{k+1}+1+\sqrt{m+1}+1+1). Since nkmn\geqslant k\geqslant m, we have that dA(π(w(n)),π(wc(n)))4n+1+6d_{A}(\pi(w(n)),\pi(w_{c}(n)))\leqslant 4\sqrt{n+1}+6.

  • Suppose v~=αm+1βm+1αkc\widetilde{v}=\alpha_{m+1}\beta_{m+1}\dots\alpha_{k}c. Then wc(n)=u~cαm+1βm+1βk1αkw_{c}(n)=\widetilde{u}c\alpha_{m+1}\beta_{m+1}\dots\beta_{k-1}\alpha_{k}, so we have that:

    π(wc(n))=π(w(n))ax0x1by0y1cax1x0by1y0c.\pi(w_{c}(n))=\pi(w(n))a^{x_{0}-x_{1}}b^{y_{0}-y_{1}}ca^{x_{1}-x_{0}}b^{y_{1}-y_{0}}c.

    Therefore, dA(π(w(n)),π(wc(n)))2|x1x0|+2|y1y0|+2d_{A}(\pi(w(n)),\pi(w_{c}(n)))\leqslant 2|x_{1}-x_{0}|+2|y_{1}-y_{0}|+2. By exactly the same argument as in the previous case we have that dA(π(w(n)),π(wc(n)))4n+1+6d_{A}(\pi(w(n)),\pi(w_{c}(n)))\leqslant 4\sqrt{n+1}+6.

Therefore, if βm=ε\beta_{m}=\varepsilon, we have that dA(π(w(n)),π(wc(n)))nd_{A}(\pi(w(n)),\pi(w_{c}(n)))\preceq\sqrt{n}. If βm=c\beta_{m}=c, then swapping the role of w(n)w(n) and wc(n)w_{c}(n) in the argument above yields dA(π(w(n)),π(wc(n)))nd_{A}(\pi(w(n)),\pi(w_{c}(n)))\preceq\sqrt{n}. Thus we finally proved that s(n)ns(n)\preceq\sqrt{n}. ∎

Remark 21.

We note that for the normal form in the proof of Theorem 20 the upper bound s(n)ns(n)\preceq\sqrt{n} is sharp. A proof of this is as follows. For a given m0m\geqslant 0 let φm:22\varphi_{m}:\mathbb{Z}^{2}\rightarrow\mathbb{Z}_{2} be a function for which φm(τ(m))=c\varphi_{m}(\tau(m))=c and φm(p)=e\varphi_{m}(p)=e for pτ(m)p\neq\tau(m). We denote by gmg_{m} the group element gm=(φm,(0,0))g_{m}=(\varphi_{m},(0,0)). Let wmw_{m} and wmaw_{ma} be the normal forms of gmg_{m} and gmag_{m}a, respectively. Let n=m+1n=m+1 and zm=τ(m)z_{m}=\tau(m). Then π(wm(n))=(φm,zm)\pi(w_{m}(n))=(\varphi_{m},z_{m}) and π(wma(n))=(φ0,zm)\pi(w_{ma}(n))=(\varphi_{0},z_{m}). Let zm=(xm,ym)z_{m}=(x_{m},y_{m}). The distance dA(π(wm(n)),π(wma(n)))d_{A}(\pi(w_{m}(n)),\pi(w_{ma}(n))) is equal to 2(|xm|+|ym|+1)2(|x_{m}|+|y_{m}|+1). Indeed, in order to obtain gmag_{m}a from gmg_{m} the lamplighter moves from the position (xm,ym)(x_{m},y_{m}) to the position (0,0)(0,0) choosing a shortest route, switch on a lamp, moves back to the position (xm,ym)(x_{m},y_{m}) and switch off a lamp. In particular, dA(π(wm(n)),π(wma(n)))2r(zm)+1d_{A}(\pi(w_{m}(n)),\pi(w_{ma}(n)))\geqslant 2r(z_{m})+1. By (8), we have that (2r(zm)+1)2m+1(2r(z_{m})+1)^{2}\geqslant m+1. Therefore, dA(π(wm(n)),π(wma(n)))nd_{A}(\pi(w_{m}(n)),\pi(w_{ma}(n)))\geqslant\sqrt{n}. This implies that ns(n)\sqrt{n}\preceq s(n).

6 Discussion and Open Questions

Theorems 14 and 15 show that for a finitely presented group with the strongly–super–polynomial Dehn function or a non–finitely presented group there exists no quasigeodesic normal form satisfying the f(n)f(n)–fellow traveler property. The following question is apparent from these results.

  1. 1.

    Is there a quasigeodesic normal form satisfying the f(n)f(n)–fellow traveler property for some finitely presented non–automatic group with the Dehn function which is not strongly–super–polynomial? Some interesting candidates to consider this question include, for example, the Heisenberg group 3()\mathcal{H}_{3}(\mathbb{Z}) and the higher Heisenberg groups 2k+1()\mathcal{H}_{2k+1}(\mathbb{Z}), k>1k>1.

Theorems 17 and 20 show the existence of a quasiregular normal form satisfying the f(n)f(n)–fellow traveler property for Baumslag–Solitar groups BS(p,q)BS(p,q), 1p<q1\leqslant p<q, and the wreath product 22\mathbb{Z}_{2}\wr\mathbb{Z}^{2}. We leave the following question for future consideration.

  1. 2.

    Is there a quasiregular normal form satisfying the f(n)f(n)–fellow traveler property for the fundamental group of a torus bundle over a circle 2A\mathbb{Z}^{2}\rtimes_{A}\mathbb{Z}, where AGL(2,)A\in\mathrm{GL}(2,\mathbb{Z}) has two real eigenvalues not equal to ±1\pm 1? Recall that the latter guarantees that 2A\mathbb{Z}^{2}\rtimes_{A}\mathbb{Z} has at least exponential Dehn function, so no quasigeodesic normal form satisfying the f(n)f(n)–fellow traveler property exists in this case.

In addition to that there are other questions that might be worth considering. Is there a quasiregular normal form satisfying the f(n)f(n)–fellow traveler proper for BS(p,q)BS(p,q), 1p<q1\leqslant p<q, for some flog(n)f\ll\log(n)? Is there a quasiregular normal form satisfying the f(n)f(n)–fellow traveler proper for 22\mathbb{Z}_{2}\wr\mathbb{Z}^{2}, 1p<q1\leqslant p<q, for some fnf\ll\sqrt{n}? What are the other examples of groups which admit quasiregular normal forms satisfying the f(n)f(n)–fellow traveler property?

Acknowledgments

The authors thank the anonymous reviewer for useful comments.

References

  • [1] Berdinsky, D., Elder, M., Taback, J.: On the geometry of Cayley automatic groups. International Journal of Algebra and Computation 32, 383–409 (2022)
  • [2] Berdinsky, D., Trakuldit, P.: Towards quantitative classification of Cayley automatic groups. East–West J. of Mathematics 20(2), 107–124 (2018)
  • [3] Berdinsky, D., Trakuldit, P.: Measuring closeness between Cayley automatic groups and automatic groups. In: Klein, S., Martín-Vide, C., Shapira, D. (eds.) Language and Automata Theory and Applications, vol. 10792, pp. 245–257. Springer International Publishing (2018)
  • [4] Bridson, M.R.: Combings of groups and the grammar of reparameterization. Comment. Math. Helv. 78, 752–771 (2003)
  • [5] Bridson, M.R., Gilman, R.H.: Formal language theory and the geometry of 3–manifolds. Commentarii Mathematici Helvetici 71(1), 525–555 (1996)
  • [6] Burillo, J., Elder, M.: Metric properties of Baumslag–Solitar groups. International Journal of Algebra and Computation 25(5), 799–811 (2015)
  • [7] Elder, M., Taback, J.: CC–graph automatic groups. Journal of Algebra 413, 289–319 (2014)
  • [8] Epstein, D.B.A., Cannon, J.W., Holt, D.F., Levy, S.V.F., Paterson, M.S., Thurston, W.P.: Word Processing in Groups. Jones and Barlett Publishers. Boston, MA (1992)
  • [9] Hartmanis, J.: Computational complexity of one–tape Turing machine computations. Journal of the Association of Computing Machinery 15, 411–418 (1968)
  • [10] Kharlampovich, O., Khoussainov, B., Miasnikov, A.: From automatic structures to automatic groups. Groups, Geometry and Dynamics 8(1), 157–198 (2014)
  • [11] Kruengthomya, P., Berdinsky, D.: Cayley Linear–Time Computable Groups. journal of Groups, Complexity, Cryptology Volume 15, Issue 2, 1–22 (Apr 2024)
  • [12] Trachtenbrot, B.: Turing computations with logarithmic delay. Algebra i Logica 3 (1964), (In Russian) English translation in U. of California Computing Center, Tech. Rep. No. 5, Berkeley, Calif., 1966.