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

Being Cayley automatic is closed under taking
wreath product with virtually cyclic groups

Dmitry Berdinskya,b aDepartment of Mathematics, Faculty of Science, Mahidol University, Bangkok, 10400, Thailand bCentre of Excellence in Mathematics, Commission on Higher Education, Bangkok, 10400, Thailand [email protected] Murray Elder School of Mathematical and Physical Sciences, University of Technology Sydney, Ultimo, NSW 2007, Australia [email protected]  and  Jennifer Taback Department of Mathematics, Bowdoin College, 8600 College Station, Brunswick, ME 04011, USA [email protected]
Abstract.

We extend work of the first author and Khoussainov to show that being Cayley automatic is closed under taking the restricted wreath product with a virtually infinite cyclic group. This adds to the list of known examples of Cayley automatic groups.

Key words and phrases:
Cayley automatic group; restricted wreath product; virtually infinite cyclic group; lamplighter group
2020 Mathematics Subject Classification:
20F65, 20E22, 68Q45
The second author is supported by Australian Research Council grant DP160100486
The third author acknowledges support from Simons Foundation grant 31736 to Bowdoin College.

1. Introduction

Cayley automatic groups, introduced by Kharlampovich, Khoussainov and Miasnikov in [7], generalise the class of automatic groups while retaining some key algorithmic properties. Namely, the word problem in a Cayley automatic group is decidable in quadratic time, and the first order theory for a (directed, labeled) Cayley graph of a Cayley automatic group is decidable. The family of Cayley automatic groups is larger than that of automatic groups, for example it includes all finitely generated nilpotent groups of nilpotency class two [7], the Baumslag-Solitar groups [1, 7], the higher rank lamplighter groups [3], and restricted wreath products of the form GG\wr\mathbb{Z} where GG is Cayley automatic [2].

Here we add to this list by extending [2] to restricted wreath products of the form GHG\wr H where GG is Cayley automatic and HH is virtually infinite cyclic. While this result is not surprising, the proof contains some subtleties which require care, and we believe is worth recording.

2. Automatic and Cayley automatic groups

We assume that the reader is familiar with the notions of regular languages, finite automata and multi-tape synchronous automata. For more details, we refer the reader to [6]. We say a language L(X)nL\subseteq(X^{*})^{n} is regular if it is accepted by a synchronous nn-tape automaton where nn\in\mathbb{N} and XX is a finite set, or alphabet.

For any group GG with finite symmetric generating set S=S1S=S^{-1}, let π:SG\pi\colon S^{*}\to G denote the canonical projection map. For wSw\in S^{*} let |w|S|w|_{S} denote the length of ww as a word in the free monoid SS^{*}, that is, |w|S|w|_{S} denotes the number of letters in the word ww.

Definition 2.1.

An automatic structure for a group GG is a pair (S,L)(S,L) where

  1. (1)

    SS is a finite symmetric generating set for GG;

  2. (2)

    LSL\subseteq S^{*} is a regular language;

  3. (3)

    π|L:LG\pi|_{L}\colon L\rightarrow G is a bijection;

  4. (4)

    for each aSa\in S the binary relation

    Ra={(u,v)L×Lπ(u)a=Gπ(v)}S×SR_{a}=\{(u,v)\in L\times L\mid\pi(u)a=_{G}\pi(v)\}\subseteq S^{*}\times S^{*}

    is regular, that is, recognised by a two-tape synchronous automaton.

A group is called automatic if it has an automatic structure with respect to some finite generating set.

It is a standard result, see, for example [6, Theorem 2.4.1], that if GG is automatic then GG has an automatic structure with respect to any finite generating set.

Cayley automatic groups were introduced in [7] with the motivation of allowing the language LL of normal forms representing group elements to be defined over an arbitrary alphabet Λ\Lambda rather than a generating set SS for GG.

Definition 2.2.

A Cayley automatic structure for a group GG is a 4-tuple (S,Λ,L,ψ)(S,\Lambda,L,\psi) where

  1. (1)

    SS is a finite symmetric generating set for GG;

  2. (2)

    Λ\Lambda is an alphabet and LΛL\subseteq\Lambda^{*} is a regular language;

  3. (3)

    ψ:LG\psi\colon L\rightarrow G is a bijection;

  4. (4)

    for each aSa\in S the binary relation

    Ra={(u,v)L×L|ψ(u)a=Gψ(v)}Λ×ΛR_{a}=\{(u,v)\in L\times L\,|\,\psi(u)a=_{G}\psi(v)\}\subseteq\Lambda^{*}\times\Lambda^{*}

    is regular, that is, recognised by a two-tape synchronous automaton.

A group is called Cayley automatic if it has a Cayley automatic structure (S,Λ,L,ψ)(S,\Lambda,L,\psi) with respect to some finite generating set SS.

As for automatic groups, if GG has a Cayley automatic structure (S,Λ,L,ψ)(S,\Lambda,L,\psi) and YY is another finite generating set for GG, then there exists a Cayley automatic structure (Y,ΛY,LY,ψY)(Y,\Lambda_{Y},L_{Y},\psi_{Y}) for GG. See [7, Theorem 6.9] for a proof of this fact.

3. Wreath products with virtually infinite cyclic groups

For two groups GG and HH, let G(H)G^{(H)} be the set of all functions γ:HG\gamma\colon H\rightarrow G with finite support, that is, such that γ(h)1G\gamma(h)\neq 1_{G} for at most finitely many hHh\in H. For a given γG(H)\gamma\in G^{(H)} and hHh\in H, we denote by γh\gamma^{h} the element of G(H)G^{(H)} for which γh(x)=γ(hx)\gamma^{h}(x)=\gamma(hx) for all xHx\in H. The restricted wreath product GHG\mathrel{\wr}H can be defined as the Cartesian product G(H)×HG^{(H)}\times H with the group multiplication given by the formula:

(γ,h)(γ,h)=(γ(γ)h1,hh).(\gamma,h)\cdot(\gamma^{\prime},h^{\prime})=\left(\gamma(\gamma^{\prime})^{\,h^{-1}},hh^{\prime}\right).

Equivalently, we can define GHG\wr H as

{(γ,h)hH,γkH(G)k where γ has finitely many nontrivial entries}\left\{(\gamma,h)\mid h\in H,\gamma\in\bigoplus_{k\in H}(G)_{k}\text{ where $\gamma$ has finitely many nontrivial entries}\right\}

with multiplication defined as above, where

(γ)h1=(kH(g)k)h1=kH(g)h1k.(\gamma^{\prime})^{\,h^{-1}}=\left(\bigoplus_{k\in H}(g)_{k}\right)^{\,h^{-1}}=\bigoplus_{k\in H}(g)_{h^{-1}k}.

Note that if GG is generated by S0GS_{0}\subseteq G and HH is generated by THT\subseteq H then GHG\wr H is generated by S0TS_{0}\cup T.

We prove the following theorem.

Theorem 3.1 (Wreath products with virtually infinite cyclic groups).

Let GG be a Cayley automatic group, and HH any virtually infinite cyclic group. Then GHG\wr H is Cayley automatic.

Proof.

Since GG is Cayley automatic, there exists a finite symmetric generating set S0S_{0} for GG, an alphabet Λ0\Lambda_{0}, a regular language L0Λ0L_{0}\subseteq\Lambda_{0}^{*}, a bijection ψ0:L0G\psi_{0}\colon L_{0}\to G, and a 2-tape automaton Ms\texttt{M}_{s} for each sS0s\in S_{0} with accepted language

L(Ms)={(u,v)L0×L0ψ0(v)=Gψ0(u)s}.L(\texttt{M}_{s})=\{(u,v)\in L_{0}\times L_{0}\mid\psi_{0}(v)=_{G}\psi_{0}(u)s\}.

Without loss of generality assume ψ0(ε)=1G\psi_{0}(\varepsilon)=1_{G}.

Let HH be a finite extension of its cyclic subgroup =t\mathbb{Z}=\langle t\rangle of index m+1m+1, and denote by tx0,tx1,,txm\langle t\rangle x_{0},\langle t\rangle x_{1},\dots,\langle t\rangle x_{m} the distinct right cosets of \mathbb{Z}, where x0=1Hx_{0}=1_{H}. Let

T={t,x1,,xm,t1,x11,,xm1};T=\{t,x_{1},\dots,x_{m},t^{-1},x_{1}^{-1},\dots,x_{m}^{-1}\};

then S=S0TS=S_{0}\cup T is a symmetric generating set for GHG\wr H. We identify a particular spanning tree 𝒮\mathcal{S} of the Cayley graph Γ(H,T)\Gamma(H,T) which consists of a “spine” corresponding to t\langle t\rangle, and at each vertex tkt^{k} there are mm “spokes” terminating at the mm vertices tkxjt^{k}x_{j} of HH, for kk\in\mathbb{Z} and 1jm1\leqslant j\leqslant m, as in Figure 1.

x1x_{1}x4x_{4}x2x_{2}x3x_{3}ttx1x_{1}x4x_{4}x2x_{2}x3x_{3}ttx1x_{1}x4x_{4}x2x_{2}x3x_{3}ttx2x_{2}x3x_{3}x1x_{1}x4x_{4}ttx2x_{2}x3x_{3}x1x_{1}x4x_{4}1H1_{H}
Figure 1. Part of a spanning tree 𝒮\mathcal{S} for Γ(H,T)\Gamma(H,T), where the index of =t\mathbb{Z}=\langle t\rangle in HH is 5.

As a concrete example, consider the infinite dihedral group

H=D=a,ba2,b2H=D_{\infty}=\langle a,b\mid a^{2},b^{2}\rangle

In this case we can take t=ab,x1=at=ab,x_{1}=a and 𝒮=Γ(D,T)\mathcal{S}=\Gamma(D_{\infty},T), as shown in Figure 2.

aattaattaattaattaa1D1_{D_{\infty}}tttttttta1a^{-1}a1a^{-1}a1a^{-1}a1a^{-1}a1a^{-1}
Figure 2. Part of the spanning tree 𝒮\mathcal{S} (shown in black) drawn inside Γ(D,T)\Gamma(D_{\infty},T), where T={t,a,t1,a1}T=\{t,a,t^{-1},a^{-1}\}. Note a=a1a=a^{-1} and ata=aaba=ba=t1ata=aaba=ba=t^{-1}.

We borrow some terminology from the lamplighter groups n\mathbb{Z}_{n}\wr\mathbb{Z} to describe elements of GHG\wr H. An element vGHv\in G\wr H can be thought of in two equivalent ways:

  1. (1)

    algebraically, as an element (γ,h)(\gamma,h) where γkH(G)k\gamma\in\bigoplus_{k\in H}(G)_{k} has finitely many nontrivial entries and hHh\in H.

  2. (2)

    geometrically, as a copy of 𝒮\mathcal{S} (or Γ(H,T)\Gamma(H,T)) where each vertex is marked by some element of GG, with all but finitely many vertices marked by 1G1_{G}, and the vertex hh of 𝒮\mathcal{S} is also marked with a pointer indicating the final position of the “lamplighter”. We refer to this marking as a configuration of 𝒮{\mathcal{S}}.

Write v=(γ,h)v=(\gamma,h) where γhiH(G)hi\gamma\in\bigoplus_{h_{i}\in H}(G)_{h_{i}} and hHh\in H. Since every element hh of HH can be written uniquely as h=tkxqh=t^{k}x_{q} for some kk\in\mathbb{Z} and 0qm0\leqslant q\leqslant m, the map ξ:H\xi:H\rightarrow\mathbb{Z} defined by ξ(tkxq)=k\xi(t^{k}x_{q})=k is well defined. Then the vertex corresponding to hHh\in H is an endpoint of a spoke attached to the vertex tξ(h)t^{\xi(h)}. For v=(γ,h)v=(\gamma,h) with γ\gamma as above, let

  1. (1)

    k=ξ(h)k_{*}=\xi(h),

  2. (2)

    k1=min{0,ξ(hi)(g)hiγ,(g)hi1G}k_{1}=\min\{0,\xi(h_{i})\mid(g)_{h_{i}}\in\gamma,(g)_{h_{i}}\neq 1_{G}\}, and

  3. (3)

    k2=max{0,ξ(hi)(g)hiγ,(g)hi1G}k_{2}=\max\{0,\xi(h_{i})\mid(g)_{h_{i}}\in\gamma,(g)_{h_{i}}\neq 1_{G}\}.

Additionally, let m1=min(k,k1)m_{1}=\min(k_{*},k_{1}) and m2=max(k,k2)m_{2}=\max(k_{*},k_{2}). Define the integer support of vv, denoted isupp(v)\mathrm{isupp}(v), to be the interval [m1,m2][m_{1},m_{2}]. The left endpoint of the integer support is the smallest kk so that either

  1. (1)

    vv has a nontrivial entry among the copies of Γ(G,S0)\Gamma(G,S_{0}) attached to the spine at the vertex tkxit^{k}x_{i} for some 0im0\leqslant i\leqslant m,

  2. (2)

    the final position of the lamplighter is tkxit^{k}x_{i} for some 0im0\leqslant i\leqslant m, or

  3. (3)

    all of kk_{*} and ξ(hi)\xi(h_{i}) are positive, that is, the lamplighter is never in a position along the spine with negative index, so m1=0m_{1}=0 denotes the starting position of the lamplighter.

The right endpoint of the integer support is defined analogously, where the 0 is included in the definition of k2k_{2} to account for the possibility that kk_{*} and all the ξ(hi)\xi(h_{i}) are negative.

To define our normal form, we mimic the standard “left-first” representation of elements of the lamplighter group n\mathbb{Z}_{n}\wr\mathbb{Z} (cf.  [4]). Given v=(γ,h)GHv=(\gamma,h)\in G\wr H, we describe a path traversed by the lamplighter from the vertex 1H1_{H} in 𝒮{\mathcal{S}} to its final vertex h𝒮h\in{\mathcal{S}}. If m1<0m_{1}<0, the lamplighter first moves left along the spine of 𝒮{\mathcal{S}} to the vertex labeled tm1t^{m_{1}}, and marks it with a possibly trivial element of GG. The lamplighter then visits tm1x1t^{m_{1}}x_{1} and marks it with a possibly trivial element of GG and returns to tm1t^{m_{1}}. This procedure is repeated for the vertices tm1x2,,tm1xmt^{m_{1}}x_{2},\cdots,t^{m_{1}}x_{m}. The lamplighter then proceeds to the vertex corresponding to tm1+1t^{m_{1}+1} and repeats the process of visiting the vertex at the end of each spoke in order and marking it with a possibly trivial element of GG. This continues until the lamplighter reaches the vertex corresponding to tm2t^{m_{2}}, where the process is repeated one last time. If m1=0m_{1}=0, the lamplighter begins the process of marking the vertices with possibly trivial elements of GG at 1H𝒮1_{H}\in{\mathcal{S}}, and then visits the spokes as described above, until it reaches the vertex labeled tm2t^{m_{2}} and marks the vertices tm2xjt^{m_{2}}x_{j} for 0jm0\leqslant j\leqslant m with possibly trivial elements of GG.

We refer to the subpath which starts at the vertex tm1t^{m_{1}} and ends at the vertex tm2t^{m_{2}} after having marked the vertices tixjt^{i}x_{j} for m1im2,0jmm_{1}\leqslant i\leqslant m_{2},0\leqslant j\leqslant m with possibly trivial elements of GG as the positive path, because when written as a word in the group generators, the exponents of tt are all positive. See Figure 3 for an example of a configuration with integer support [2,1][-2,1] where the positive path is marked.

x2x_{2}g1g_{1}x3x_{3}x1x_{1}x4x_{4}g2g_{2}ttx1x_{1}x4x_{4}x2x_{2}x3x_{3}x1x_{1}x4x_{4}x2x_{2}x3x_{3}ttx1x_{1}g4g_{4}g3g_{3}x4x_{4}x2x_{2}x3x_{3}ttx2x_{2}x3x_{3}x1x_{1}x4x_{4}tt1H1_{H}
Figure 3. The element t2x2g1x21x4g2x41t3g3x1g4x11t2x3t^{-2}x_{2}g_{1}x_{2}^{-1}x_{4}g_{2}x_{4}^{-1}t^{3}g_{3}x_{1}g_{4}x_{1}^{-1}t^{-2}x_{3} as a configuration of 𝒮\mathcal{S}. The integer support of this element is [2,1][-2,1] and the red arrow denotes the final position of the lamplighter. The positive path for this element is x1x11x2g1x21x3x31x4g2x41(tx1x11x2x21x3x31x4x41)2tg3x1g4x11x2x21x3x31x4x41x_{1}x_{1}^{-1}x_{2}g_{1}x_{2}^{-1}x_{3}x_{3}^{-1}x_{4}g_{2}x_{4}^{-1}(tx_{1}x_{1}^{-1}x_{2}x_{2}^{-1}x_{3}x_{3}^{-1}x_{4}x_{4}^{-1})^{2}tg_{3}x_{1}g_{4}x_{1}^{-1}x_{2}x_{2}^{-1}x_{3}x_{3}^{-1}x_{4}x_{4}^{-1} is shown in green.

Upon completing the positive path, one of two things will occur. It may be that the lamplighter is in its final position, and the path simply ends. If not, the lamplighter moves to its final position via a subpath of the form tkt^{k} or tkxqt^{k}x_{q} where k,k0k\in\mathbb{Z},k\leqslant 0. Note that since m2m_{2}, the right endpoint of isupp(v)\mathrm{isupp}(v), is the maximum of k2k_{2} and kk_{*}, the lamplighter will never be in a position along =t\mathbb{Z}=\langle t\rangle to the right of m2m_{2}, so the exponent kk is non-positive.

As the lamplighter travels along its positive path, we will wish to indicate two special positions: the first time the lamplighter is at the vertex corresponding to 1H=t01_{H}=t^{0}, and the first time the lamplighter is at the vertex which will be its final position. The integer support and the positive path are defined so that these are unique positions along the positive path.

The normal form for the Cayley automatic structure on GHG\wr H will be constructed in stages. We first define a normal form 𝒩0(Λ0T)\mathcal{N}_{0}\subseteq(\Lambda_{0}\cup T)^{*} for elements of GHG\wr H as follows. Given vGHv\in G\wr H with isupp(v)=[m1,m2]\mathrm{isupp}(v)=[m_{1},m_{2}], the above description allows us to uniquely represent vv as a word either of the form

(1) v=tnxq or v=tm1v1tv2tvstjxqv=t^{n}x_{q}\ \text{ or }\ v=t^{m_{1}}v_{1}tv_{2}\dots tv_{s}t^{j}x_{q}

where n,j,q,s,j0n,j,q,s\in\mathbb{Z},\ j\leqslant 0, s1, 0qms\geqslant 1,\ 0\leqslant q\leqslant m, m1+(s1)=m2m_{1}+(s-1)=m_{2} and

(2) vς=vς,0x1vς,1x11x2vς,2xm11xmvς,mxm1v_{\varsigma}=v_{\varsigma,0}x_{1}v_{\varsigma,1}x_{1}^{-1}x_{2}v_{\varsigma,2}\dots x_{m-1}^{-1}x_{m}v_{\varsigma,m}x_{m}^{-1}

with vς,tL0v_{\varsigma,t}\in L_{0}. If m1=km_{1}=k_{*} then we allow v1v_{1} to be trivial, otherwise v1v_{1} must be nontrivial. If m2=km_{2}=k_{*} we allow vςv_{\varsigma} to be trivial, otherwise vςv_{\varsigma} must be nontrivial. Each word vςv_{\varsigma} encodes a sequence of words (vς,0,,vς,m)L0m+1(v_{\varsigma,0},\dots,v_{\varsigma,m})\in L_{0}^{m+1} with ψ0(vς,0)\psi_{0}(v_{\varsigma,0}) labeling the vertex at position tm1+ς1t^{m_{1}+\varsigma-1} in 𝒮\mathcal{S} and ψ0(vς,i)\psi_{0}(v_{\varsigma,i}) labeling the end of the spoke at position tm1+ς1xit^{m_{1}+\varsigma-1}x_{i} for 1im1\leqslant i\leqslant m. Note that in Equation (1), the vςv_{\varsigma} are separated by instances of tt as the lamplighter moves along the positive path. Let 𝒩0(Λ0T){\mathcal{N}}_{0}\subseteq(\Lambda_{0}\cup T)^{*} denote the set of words of this form.

For example, the element in Figure 3 has 𝒩0\mathcal{N}_{0} normal form

t2x1x11x2v1,2x21x3x31x4v1,4x41tx1x11x2x21x3x31x4x41tx1x11x2x21x3x31x4x41tv4,0x1v4,1x11x2x21x3x31x4x41t2x3\begin{split}t^{-2}x_{1}x_{1}^{-1}x_{2}v_{1,2}x_{2}^{-1}x_{3}x_{3}^{-1}x_{4}v_{1,4}x_{4}^{-1}tx_{1}x_{1}^{-1}x_{2}x_{2}^{-1}x_{3}x_{3}^{-1}x_{4}x_{4}^{-1}tx_{1}x_{1}^{-1}x_{2}x_{2}^{-1}\\ x_{3}x_{3}^{-1}x_{4}x_{4}^{-1}tv_{4,0}x_{1}v_{4,1}x_{1}^{-1}x_{2}x_{2}^{-1}x_{3}x_{3}^{-1}x_{4}x_{4}^{-1}t^{-2}x_{3}\end{split}

where ψ0(v1,2)=g1,ψ0(v1,4)=g2\psi_{0}(v_{1,2})=g_{1},\psi_{0}(v_{1,4})=g_{2}, ψ0(v4,0)=g3\psi_{0}(v_{4,0})=g_{3}, ψ0(v4,1)=g4\psi_{0}(v_{4,1})=g_{4}, and in all other cases, vi,j=εv_{i,j}=\varepsilon where ψ0(ε)=1G\psi_{0}(\varepsilon)=1_{G}.

Next, we insert special symbols into the words in 𝒩0\mathcal{N}_{0} to obtain the intermediate language 𝒩1\mathcal{N}_{1}.

Let Λ1=Λ0T{B,C,B0,C}\Lambda_{1}=\Lambda_{0}\cup T\cup\{B,C,B_{0},C_{*}\} and Λ=Λ0{B,C,B0,C}\Lambda=\Lambda_{0}\cup\{B,C,B_{0},C_{*}\}. Let v=(γ,h)GHv=(\gamma,h)\in G\wr H be written in the form of Equation (1). Notice that all terms of the form vςv_{\varsigma} are part of the positive path. With vςv_{\varsigma} as in Equation (2), before each vς,jv_{\varsigma,j} we place the symbol CC, with one exception. If vς,jv_{\varsigma,j} is the label of the vertex hh of 𝒮{\mathcal{S}} which is the final position of the lamplighter, then precede vς,jv_{\varsigma,j} by the symbol CC_{*}. Before each term vςv_{\varsigma} we place the symbol BB, with one exception. If m1+ς1=0m_{1}+\varsigma-1=0 we place the symbol B0B_{0} in front of vςv_{\varsigma}, indicating the unique position along the positive path where the lamplighter is at the vertex 1H𝒮1_{H}\in{\mathcal{S}}.

Let 𝒩1Λ1{\mathcal{N}}_{1}\subseteq\Lambda_{1}^{*} denote the set of all words in 𝒩0{\mathcal{N}_{0}} where the symbols {B,C,B0,C}\{B,C,B_{0},C_{*}\} have been inserted as described. The word in 𝒩1\mathcal{N}_{1} for the element in Figure 3 is then

t2BCx1Cx11x2Cv1,2x21x3Cx31x4Cv1,4x41tBCx1Cx11x2Cx21x3Cx31x4Cx41tB0Cx1Cx11x2Cx21x3Cx31x4Cx41tBCv4,0x1Cv4,1x11x2Cx21x3Cx31x4Cx41t2x3.\begin{split}t^{-2}BCx_{1}Cx_{1}^{-1}x_{2}Cv_{1,2}x_{2}^{-1}x_{3}Cx_{3}^{-1}x_{4}Cv_{1,4}x_{4}^{-1}tBCx_{1}Cx_{1}^{-1}x_{2}Cx_{2}^{-1}x_{3}C_{*}x_{3}^{-1}\\ x_{4}Cx_{4}^{-1}tB_{0}Cx_{1}Cx_{1}^{-1}x_{2}Cx_{2}^{-1}x_{3}Cx_{3}^{-1}x_{4}Cx_{4}^{-1}tBCv_{4,0}x_{1}Cv_{4,1}x_{1}^{-1}x_{2}Cx_{2}^{-1}\\ x_{3}Cx_{3}^{-1}x_{4}Cx_{4}^{-1}t^{-2}x_{3}.\end{split}

To obtain the final normal form which will be the basis of the Cayley automatic structure for GHG\wr H, let 𝒩Λ{\mathcal{N}}\subseteq\Lambda^{*} denote the set of words in 𝒩1{\mathcal{N}}_{1} where all instances of the letters t,x1,,xm,t1,x11,,xm1t,x_{1},\dots,x_{m},t^{-1},x_{1}^{-1},\dots,x_{m}^{-1} from the set TT are removed. The word in 𝒩\mathcal{N} for the element in Figure 3 is then

BCCCv1,2CCv1,4BCCCCCB0CCCCCBCv4,0Cv4,1CCC.BCCCv_{1,2}CCv_{1,4}BCCCC_{*}CB_{0}CCCCCBCv_{4,0}Cv_{4,1}CCC.

Define the language

(3) L1={i=1p(βiΓi,0vi,0Γi,1vi,1Γi,2vi,2Γi,mvi,m)|vi,jL0,βi{B,B0},Γi,j{C,C},p1}.L_{1}=\left\{\prod_{i=1}^{p}\left(\beta_{i}\Gamma_{i,0}v_{i,0}\Gamma_{i,1}v_{i,1}\Gamma_{i,2}v_{i,2}\dots\Gamma_{i,m}v_{i,m}\right)\ \middle|\ \begin{array}[]{ll}v_{i,j}~{}\in L_{0},\\ \beta_{i}~{}\in\{B,B_{0}\},\\ \Gamma_{i,j}~{}\in\{C,C_{\ast}\},\\ p\geqslant 1\end{array}\right\}.

As L0L_{0} is a regular language, it follows that L1L_{1} is a regular language.

Recall that when vtkxqv\neq t^{k}x_{q}, if m1=km_{1}=k_{*} we allow v1v_{1} to be trivial, otherwise v1v_{1} must be nontrivial, and if m2=km_{2}=k_{*} we allow vsv_{s} to be trivial, otherwise vsv_{s} must be nontrivial. These conditions are easily verified by a finite state automaton inspecting, respectively, the first and last expressions in the product representing an element of L1L_{1} according to the following rules:

  1. (1)

    if CC_{*} occurs in the first factor in the product as expressed in Equation 3, then all vi,jv_{i,j} may be ε\varepsilon for 0jm0\leqslant j\leqslant m; if not, at least one vi,jv_{i,j} must be nontrivial.

  2. (2)

    if the CC_{*} occurs in the last factor in the product as expressed in Equation 3, then all vi,jv_{i,j} may be ε\varepsilon for 0jm0\leqslant j\leqslant m; if not, at least one vi,jv_{i,j} must be nontrivial.

Note that a finite state automaton can also easily verify that when all vςv_{\varsigma} are trivial, we have a normal form corresponding to tkxqt^{k}x_{q}. Let L2L1L_{2}\subseteq L_{1} be the set of all strings in L1L_{1} for which all three of these conditions are satsified. Since these conditions are easily checked by a finite state automaton, and L1L_{1} is a regular language, it follows that L2L_{2} is a regular language.

Finally, we verify that the string has only one occurrence each of BB_{*} and CC_{*}. Let

L=L2{pB0qCr,pCqB0rp,q,r(Λ{B0,C})}.L=L_{2}\cap\{pB_{0}qC_{\ast}r,pC_{\ast}qB_{0}r\mid p,q,r\in(\Lambda\setminus\{B_{0},C_{*}\})^{*}\}.

It follows that LL is regular and that L=𝒩L={\mathcal{N}}.

As a further example, note that if v=tnxqv=t^{n}x_{q}, the corresponding word in LL is as follows:

  1. (1)

    when n>0n>0, we have isupp(tnxq)=[0,n]\mathrm{isupp}(t^{n}x_{q})=[0,n] and the corresponding word is

    B0Cm+1(BCm+1)n1BCqCCmq;B_{0}C^{m+1}(BC^{m+1})^{n-1}BC^{q}C_{*}C^{m-q};
  2. (2)

    when n=0n=0, we have isupp(xq)=[0,0]\mathrm{isupp}(x_{q})=[0,0] and the corresponding word is

    B0CqCCmq;B_{0}C^{q}C_{*}C^{m-q};
  3. (3)

    when n<0n<0, we have isupp(tnxq)=[n,0]\mathrm{isupp}(t^{n}x_{q})=[n,0] and the corresponding word is

    BCqCCmq(BCm+1)n1B0Cm+1.BC^{q}C_{*}C^{m-q}(BC^{m+1})^{n-1}B_{0}C^{m+1}.

Given a word σL\sigma\in L, the symbols B0B_{0} and CC_{*} allow us to reconstruct the integer support of the corresponding element, as well as the final position of the lamplighter, that is, the coordinate hh. The words vi,jv_{i,j} correspond (via ψ0\psi_{0}) to elements of GG listed in a specified order. That is, we can deterministically reconstruct γhH(G)h\gamma\in\bigoplus_{h\in H}(G)_{h} and hHh\in H from σ\sigma. Formally, let ψ:LGH\psi\colon L\rightarrow G\wr H be the bijective map defined by

ψ(w)=ψ(k=1s(βkΓk,0uk,0Γk,1uk,1Γk,2uk,2Γk,muk,m))=tm1p1tp2tpstj\psi(w)=\psi\left(\prod_{k=1}^{s}\left(\beta_{k}\Gamma_{k,0}u_{k,0}\Gamma_{k,1}u_{k,1}\Gamma_{k,2}u_{k,2}\dots\Gamma_{k,m}u_{k,m}\right)\right)=t^{m_{1}}p_{1}tp_{2}\dots tp_{s}t^{j}

where

pi=ψ0(vi,0)x1ψ0(vi,1)x11x2ψ0(vi,2)xm11xmψ0(vi,m)xm1,p_{i}=\psi_{0}(v_{i,0})x_{1}\psi_{0}(v_{i,1})x_{1}^{-1}x_{2}\psi_{0}(v_{i,2})\dots x_{m-1}^{-1}x_{m}\psi_{0}(v_{i,m})x_{m}^{-1},

with ui,j=vi,ju_{i,j}=v_{i,j}, βk{B,B0}\beta_{k}\in\{B,B_{0}\}, Γk,j{C,C}\Gamma_{k,j}\in\{C,C_{*}\} and m1m_{1} calculated from the positions of B0B_{0} and CC_{*} as described above.

We claim that (S,Λ,L,ψ)(S,\Lambda,L,\psi) is a Cayley automatic structure for GHG\wr H. To prove this, we must show that for every generator sS=S0Ts\in S=S_{0}\cup T the set

Rs={(u,v)L×L|ψ(u)s=GHψ(v)}R_{s}=\{(u,v)\in L\times L|\psi(u)s=_{G\wr H}\psi(v)\}

is a regular language, that is, recognised by a 2-tape synchronous automaton. It suffices to do this for sS0{x1,,xm,t}s\in S_{0}\cup\{x_{1},\cdots,x_{m},t\}; see, for example, [5, Lemma 9].

First let sS0s\in S_{0} and suppose (u,v)Rs(u,v)\in R_{s}. Viewing ψ(u)\psi(u) as a configuration of 𝒮{\mathcal{S}} with finitely many vertices marked with elements of GG and a distinguished position for the lamplighter, we can easily see the effect of multiplication by ss on the normal form. Let tkxqt^{k}x_{q} denote the vertex of 𝒮{\mathcal{S}} which is the final position of the lamplighter in uu, marked by the element guGg_{u}\in G. Let ρuL0\rho_{u}\in L_{0} be such that ψ0(ρu)=gu\psi_{0}(\rho_{u})=g_{u}. To obtain the normal form word for ψ(u)s\psi(u)s we simply multiply ρu\rho_{u} by ss and verify that the multiplication is correct using the multiplier automaton Ms\texttt{M}_{s} given as part of the given Cayley automatic structure on GG. Therefore we need to accept pairs of strings (u,v)L×L(u,v)\in L\times L of the following form:

u=(Πi=1pβiΠj=0mCαi,j)Θu(Πi=p+2ςβiΠj=0mCαi,j)u=\left(\Pi_{i=1}^{p}\beta_{i}\Pi_{j=0}^{m}C\alpha_{i,j}\right)\Theta_{u}\left(\Pi_{i=p+2}^{\varsigma}\beta_{i}\Pi_{j=0}^{m}C\alpha_{i,j}\right)

and

v=(Πi=1pβiΠj=0mCαi,j)Θv(Πi=p+2ςβiΠj=0mCαi,j)v=\left(\Pi_{i=1}^{p}\beta_{i}\Pi_{j=0}^{m}C\alpha_{i,j}\right)\Theta_{v}\left(\Pi_{i=p+2}^{\varsigma}\beta_{i}\Pi_{j=0}^{m}C\alpha_{i,j}\right)

where βi{B,B0}\beta_{i}\in\{B,B_{0}\}, αi,jL0\alpha_{i,j}\in L_{0},

Θu=βp+1Cαp+1,0C𝜶𝒑+𝟏,𝒓Cαp+1,m\Theta_{u}=\beta_{p+1}C\alpha_{p+1,0}\cdots C_{*}\bm{\alpha_{p+1,r}}\cdots C\alpha_{p+1,m}

and

Θv=βp+1Cαp+1,0C𝜶𝒑+𝟏,𝒓Cαp+1,m\Theta_{v}=\beta_{p+1}C\alpha_{p+1,0}\cdots C_{*}\bm{\alpha^{\prime}_{p+1,r}}\cdots C\alpha_{p+1,m}

where (αp+1,r,αp+1,r)(\alpha_{p+1,r},\alpha^{\prime}_{p+1,r}) is accepted by the multiplier automaton Ms\texttt{M}_{s} given as part of the given Cayley automatic structure on GG. The bold highlighted symbols represent the only difference between the two words.

By [7] (see also [5, Lemma 8]) the language L0L_{0} is necessarily quasigeodesic. It follows that the difference between the lengths of αp+1,r\alpha_{p+1,r} and αp+1,r\alpha^{\prime}_{p+1,r} is uniformly bounded. As it is regular to check that two words are identical with a bounded shift, it follows that we can construct a 2-tape automaton which checks that the prefixes of uu and vv are identical, then calls Ms\texttt{M}_{s} to read (αp+1,r,αp+1,r)(\alpha_{p+1,r},\alpha^{\prime}_{p+1,r}), and finally checks that the suffixes of uu and vv are identical (with a bounded shift). Thus RsR_{s} is a regular language.

Next let xi{x1,,xm}x_{i}\in\{x_{1},\cdots,x_{m}\}, and suppose (u,v)Rxi(u,v)\in R_{x_{i}}. Writing ψ(u)\psi(u) as in Equation (1), we see that ψ(u)xi\psi(u)x_{i} ends in the letters xqxix_{q}x_{i}. The product xqxiHx_{q}x_{i}\in H is an element of some right coset txr\langle t\rangle x_{r}. That is, xqxi=tkxrx_{q}x_{i}=t^{k}x_{r} for some kk and rr. Viewing ψ(u)\psi(u) and ψ(v)\psi(v) as configurations in 𝒮{\mathcal{S}}, this means that the configurations are identical except for the final position of the lamplighter which is indicated by CC_{*} in the normal form. Note that as xqx_{q} and xix_{i} vary among the finite set of coset representatives, there are only a finite number of possible values of (k,r)(k,r) which arise.

The elements ψ(u)\psi(u) and ψ(v)\psi(v) may or may not have identical integer support. For example if ψ(u)=t10g1t20g2t5xq\psi(u)=t^{-10}g_{1}t^{20}g_{2}t^{-5}x_{q} and xqxi=t7xrx_{q}x_{i}=t^{-7}x_{r} then isupp(ψ(u))=isupp(ψ(v))\mathrm{isupp}(\psi(u))=\mathrm{isupp}(\psi(v)), whereas if ψ(u)=t10g1t20g2t5xq\psi(u)=t^{-10}g_{1}t^{20}g_{2}t^{-5}x_{q} and xqxi=t17xrx_{q}x_{i}=t^{-17}x_{r} then isupp(ψ(u))isupp(ψ(v))\mathrm{isupp}(\psi(u))\neq\mathrm{isupp}(\psi(v)).

If isupp(ψ(u))=isupp(ψ(v))\mathrm{isupp}(\psi(u))=\mathrm{isupp}(\psi(v)), then we simply need to check the two strings are identical except for the location of CC_{*}. If ψ(u)\psi(u) ends in xqx_{q} when written as in Equation (1), we have xqxi=tkxrx_{q}x_{i}=t^{k}x_{r}. Let π:Λ{C,C,B0}\pi\colon\Lambda^{*}\to\{C,C_{*},B_{0}\}^{*} be a homomorphism which is the identity on C,C,B0C,C_{*},B_{0} and sends all other letters to ε\varepsilon. Then π(u)\pi(u) and π(v)\pi(v) are identical strings except for the location of CC_{*} in each string. Observe the letter B0B_{0} is in the same position in each string since the integer supports of ψ(u)\psi(u) and ψ(v)\psi(v) are the same. Further observe that there exists an integer sq,is_{q,i} such that for every pair (u,v)Rxi(u,v)\in R_{x_{i}} which have the same integer support, if CC_{*} is the xxth letter of π(u)\pi(u) and the yyth letter of π(v)\pi(v), then xy=sq,ix-y=s_{q,i}.

Consider the language Xq,i{C,C,B0}×{C,C,B0}X_{q,i}\subseteq\{C,C_{*},B_{0}\}^{*}\times\{C,C_{*},B_{0}\}^{*} consisting of all pairs of strings, each of which contains exactly one CC_{*} letter and one B0B_{0} letter, where CC_{*} is the xxth letter of the first string and the yyth letter of the second string with xy=sq,ix-y=s_{q,i}, and B0B_{0} is in the same position in both strings. Since these conditions are regular to check, Xq,iX_{q,i} is a regular language.

Let

κ:Λ×Λ{C,C,B0}×{C,C,B0}\kappa\colon\Lambda^{*}\times\Lambda^{*}\to\{C,C_{*},B_{0}\}^{*}\times\{C,C_{*},B_{0}\}^{*}

be the map which in each coordinate is the identity on C,CC,C_{*} and B0B_{0} and sends all other letters to ε\varepsilon. Let YΛ×ΛY\subseteq\Lambda^{*}\times\Lambda^{*} be the language consisting of all pairs of strings such that for every positive integer zz the zzth letter of the first string and the second string is the same unless one of these letters is CC_{*} and the other is CC. The language YY is regular. Then the language

κ1(Xq,i)(L×L)Y\kappa^{-1}(X_{q,i})\cap\left(L\times L\right)\cap Y

is regular, and the union of these languages for 0qm+10\leqslant q\leqslant m+1 is exactly the subset of RxiR_{x_{i}} for which multiplication by xix_{i} does not change the integer support for the first entry.

Now consider all the possible ways that the integer support of ψ(u)\psi(u) can change upon multiplication by xix_{i}. Again assume ψ(u)\psi(u) ends in xqx_{q} when written as in Equation (1), and xqxi=tkxrx_{q}x_{i}=t^{k}x_{r}. We must consider the following cases.

  1. (1)

    ψ(u)=tnxq\psi(u)=t^{n}x_{q} and k0k\neq 0,

  2. (2)

    ψ(u)=tm1v1tv2tvstjxq\psi(u)=t^{m_{1}}v_{1}tv_{2}\dots tv_{s}t^{j}x_{q} with isupp(ψ(u))=[m1,m2],j0\mathrm{isupp}(\psi(u))=[m_{1},m_{2}],j\leqslant 0 and

    1. (a)

      k>jk>-j; in this case the integer support of vv extends further to the right of m2m_{2},

    2. (b)

      k<m1m2jk<m_{1}-m_{2}-j; in this case the integer support of vv extends further to the left of m1m_{1}.

Each of these cases can be handled in a manner similar to the above case, by considering the relative positions of CC_{*} and B0B_{0} in π(u),π(v)\pi(u),\pi(v). For the first case, if n>0,k>nn>0,k>-n then

u=B0Cm+1(BCm+1)n1BCqCCmqand v=B0Cm+1(BCm+1)n+k1BCrCCmr;\begin{split}u=B_{0}C^{m+1}(BC^{m+1})^{n-1}BC^{q}C_{*}C^{m-q}\\ \text{and }v=B_{0}C^{m+1}(BC^{m+1})^{n+k-1}BC^{r}C_{*}C^{m-r};\end{split}

if n>0,k<nn>0,k<-n then

u=B0Cm+1(BCm+1)n1BCqCCmqand v=BCrCCmr(BCm+1)kn1B0Cm+1\begin{split}u=B_{0}C^{m+1}(BC^{m+1})^{n-1}BC^{q}C_{*}C^{m-q}\\ \text{and }v=BC^{r}C_{*}C^{m-r}(BC^{m+1})^{-k-n-1}B_{0}C^{m+1}\end{split}

Analogous pairs of expressions can be worked out for n0n\leqslant 0; clearly all such pairs can be recognised by 2-tape automata since q,i,k,rq,i,k,r are fixed. We leave details of the remaining cases to the reader.

Finally, suppose (u,v)Rt(u,v)\in R_{t}. Writing ψ(u)\psi(u) as in Equation (1), we see that ψ(u)t\psi(u)t ends in the letters xqtx_{q}t. Once again, we can consider the case where the integer support of ψ(u)\psi(u) does not change, in which case we merely need to check the location of the CC_{*} letters in each word, and separately the case where the integer support of ψ(u)\psi(u) differs at one endpoint from the integer support of ψ(v)\psi(v). We follow the same reasoning as in the previous case of multiplication by xix_{i}; note that xqtx_{q}t is in some right coset of \mathbb{Z} in HH, so we can write xqt=tkxrx_{q}t=t^{k}x_{r}, for a possibly different coset representative xrx_{r}. We can therefore show that RtR_{t} is a regular language as well. The regular languages RsR_{s}, RxiR_{x_{i}} and RtR_{t} complete the construction of the Cayley automatic structure (S,Λ,L,ψ)(S,\Lambda,L,\psi) for GHG\wr H. ∎

References

  • [1] Dmitry Berdinsky and Bakhadyr Khoussainov. On automatic transitive graphs. In Developments in language theory, volume 8633 of Lecture Notes in Comput. Sci., pages 1–12. Springer, Cham, 2014.
  • [2] Dmitry Berdinsky and Bakhadyr Khoussainov. Cayley automatic representations of wreath products. International Journal of Foundations of Computer Sceince, 27(2):147–159, 2016.
  • [3] Sophie Bérubé, Tara Palnitkar, and Jennifer Taback. Higher rank lamplighter groups are graph automatic. J. Algebra, 496:315–343, 2018.
  • [4] Sean Cleary and Jennifer Taback. Dead end words in lamplighter groups and other wreath products. The Quarterly Journal of Mathematics, 56(2):165–178, 2005.
  • [5] Murray Elder and Jennifer Taback. 𝒞\mathcal{C}-graph automatic groups. J. Algebra, 413:289–319, 2014.
  • [6] David B. A. Epstein, James W. Cannon, Derek F. Holt, Silvio V. F. Levy, Michael S. Paterson, and William P. Thurston. Word Processing in Groups. Jones and Barlett Publishers. Boston, MA, 1992.
  • [7] Olga Kharlampovich, Bakhadyr Khoussainov, and Alexei Miasnikov. From automatic structures to automatic groups. Groups Geom. Dyn., 8(1):157–198, 2014.