arXiv:1808.06371v2 [math.GR] 11 Sep 2019 n ugop n h oe rwhsre ihrsett n su set. any generating to of for respect choice with that any growt series Namely for growth relative functions the results. rational the series, those and growth to , conjugacy facts th any the ge surprising three , finite perhaps add of abelian choice is we ally the c it paper of different series, independent this are many growth In that of in results case few attention very the of are In lot a [14]. attracted groups have groups erated infiac fteerslscmsfo h atta hyhl fo hold [15 they in that group fact Heisenberg the integer set from generating the comes of t for results in and these groups genera [18]), of hyperbolic and (wit (a significance for [16], result proved series [6], corresponding was [5], Klarn growth 1) The (see of to complete work rational. equal the on also uniformly built that function is [23] growth) show Liardet standard to [3]. of others in rational and be [22] to Benson by shown eti uoai rus oee,ltl skonfrohrgener series. other growth ‘st for of known to types is other respect little or However, with Baumslag-S groups. series soluble automatic the growth certain groups, rational Coxeter example have for res to generators, Ma with shown rational another. been to are respect have series with transcendental growth and whose set generating groups exist there that AINLGOT NVRULYAEINGROUPS ABELIAN VIRTUALLY IN GROWTH RATIONAL h eae oin fgot ucin n rwhsre ffinite of series growth and functions growth of notions related The h tnadwihe rwhsre fayvrulyaeingopw group abelian virtually any of series growth weighted standard The 2010 ahmtc ujc Classification. Subject oeso n ugopof subgroup any of hieo nt egtdgnrtn e for set generating weighted finite of choice of classes gacy lyaeingroups. abelian ally Abstract. lyaeingroup abelian ally e od:Cnuaygot,rltv rwh oe growth coset growth, relative growth, Conjugacy words: Key S hsne o etecs ngnrl tl hwdi [28] in showed Stoll general. in case the be not need This . eso htaysbru fafiieygnrtdvirtu- generated finitely a of subgroup any that show We G rw ainly hs eut odrgrls fthe of regardless hold results These rationally. grows G rw ainlyrltv to relative rationally grows 1. G LXEVETTS ALEX Introduction rw ainly n httesto conju- of set the that and rationally, grows 1 06,2E5 05E15. 20E45, 20F65, G . G httesto right of set the that , yohrgroups other ny virtu- , ltrgroups, olitar eaigset. nerating n choice any r gop are bgroup, ett one to pect tn sets, ating eisof series h weight h e1980s he assof lasses tthere at virtu- a andard’ lisation ygen- ly .The ]. as er 2 ALEX EVETTS

Let G be a group with a finite generating set S, that generates G as a monoid. Let S∗ denote the set of all words over S (i.e. the free monoid on S). If we assign a positive integer weight, ω(s), to each element s ∈ S, we ∗ k can define the weight of any word s1s2 ··· sk ∈ S to be i=1 ω(si). In turn, we define the weight of an element g ∈ G, denoted ω(g), to be the minimal weight amongst all words that represent g. P We define the (standard) weighted growth function of G with respect to S and ω to be ω (1) σG,S(n) = #{g ∈ G | ω(g)= n}. The weighted growth series of a group G, with respect to a finite generating set S with weight function ω is defined as ∞ ω ω n (2) SG,S(z)= σG,S(n)z . n=0 X ω It is natural to ask under what conditions SG,S(z) can be expressed as a rational function, that is, when do there exist polynomials p, q with integer ω p(z) coefficients such that SG,S(z)= q(z) . The definition given via (1) and (2) is often referred to as strict or spherical growth. One can also consider cumula- tive growth, which instead counts elements with weight at most n. Note that the spherical growth series is rational if and only if the cumulative growth series is rational. In this paper we focus on virtually abelian groups, and study various gen- eralisations of the notion of growth series. In each case, the results will hold for any choice of weighted generating set. Let G be a group generated by a finite set S, with weight function ω, ω and consider a subgroup H. Let σH≤G,S (n) be the number of elements of H which have weight n with respect to the generators of G. We will call this the weighted growth function of H relative to G. The weighted growth series of H relative to G is then ∞ ω ω n (3) SH≤G,S(z)= σH≤G,S(n)z . n=0 X ω The asymptotic behaviour of the relative growth function σH≤G,S(n) has been studied in many papers (for example [11]). In the present paper we consider the much less studied formal power series. In Section 3 we prove the following. Theorem A (See Theorem 3.3). If G is virtually abelian, and H ≤ G is any subgroup, then the weighted growth series of H relative to G is rational with respect to any finite generating set S of G. One way to view standard growth is to consider the equivalence relation on S∗ where words are equivalent whenever they represent the same group RATIONAL GROWTH IN VIRTUALLY ABELIAN GROUPS 3 element. We then choose one minimal weight word from each equivalence class, and study the growth of the resulting language. Any other equivalence relation on S∗ thus yields a growth function (and corresponding series) in the same way. In Section 4, we study the growth series arising from the equivalence re- lation where words are considered to be equivalent whenever they represent elements of the same coset of a chosen subgroup H. This is known as coset growth. If the weight of every generator is 1, then the coset growth function counts the number of cosets intersecting the sphere of radius n in the Cayley graph (that do not intersect the sphere of radius n − 1). Coset growth has been studied for hyperbolic groups by Holt and others. See [21] for details. Here, we prove the following. Theorem B (See Theorem 4.1). If G is virtually abelian, and H ≤ G is any subgroup, then the weighted growth series of the set of right cosets H\G is rational with respect to any finite generating set S of G. In Section 5 we study the growth series arising from the equivalence re- lation where words are considered to be equivalent whenever they represent elements of the same conjugacy class. This is known as conjugacy growth, and has been studied for some time, for example in [10] and [4]. To the author’s knowledge, Babenko [2] was the first to introduce this form of growth. For a useful overview see Guba and Sapir’s paper [20]. Rivin studied the corre- sponding formal power series, and conjectured ([26], [27]) that a hyperbolic group has rational conjugacy growth series if and only if it is virtually cyclic. One direction of this was confirmed by Ciobanu, Hermiller, Holt, and Rees in [7], and the other by Antol´ınand Ciobanu in [1]. Recently, Gekhtman and Yang ([17]) have shown that the conjugacy growth series is transcendental for all (non-elementary) relatively hyperbolic groups, and a large class of acy- clindrically hyperbolic groups, with respect to any choice of finite generating set. Amongst other calculations, Mercier [24] has shown that the conjugacy growth series of the lamplighter group C2 ≀ Z is transcendental with respect to a certain choice of generating set. Section 5 of [7] contains calculations of conjugacy growth and related series in some virtually abelian groups, and [8] contains formulas for the conjugacy growth series of graph products. In the present paper we prove the following. Theorem C (See Theorem 5.1). If G is virtually abelian, then the weighted conjugacy growth series of G is rational with respect to any choice of finite generating set S. In light of the results mentioned above, we venture the conjecture that the conjugacy growth series of a finitely presented group that is not virtually abelian is transcendental, with respect to all finite generating sets. 4 ALEX EVETTS

Our key tool for all of the results will be the theory of polyhedral sets developed in [3]. These are defined to be subsets of Zr for some r> 0, made up of regions bounded by affine hyperplanes. Assigning a weight to each coordinate of Zr, and then extending this linearly to a weight function on the whole of Zr, we can calculate the growth series of any subset. It turns out that polyhedral sets have rational growth series (see Section 2.4 for definitions and results). Our strategy will always be to find a language (i.e. a subset of S∗) consisting of minimal representatives for the objects we wish to count, which is in one-to-one weight-preserving correspondence with a polyhedral set, and thus has rational growth series. The objects in question may be group elements, cosets, or conjugacy classes. In [3], it is shown that there exists a language of minimal representatives for the elements of a given virtually abelian group, which is in one-to-one weight-preserving correspondence with a finite collection of polyhedral sets. Thus the (standard) growth series is a rational function.

2. Preliminaries 2.1. Notation. Here we follow [3]. Let G be a finitely generated virtually abelian group. It is well known that such a group has a normal, finite-index subgroup isomorphic to Zn for some n. Throughout the paper, G will denote a finitely generated virtually abelian group with of finite index d, isomorphic to Zn. Choose a finite monoid generating set S. If w ∈ S∗, or W ⊆ S∗, we will write w, or W , to denote the element(s) of G represented by the given word(s). Write ǫ for the empty word. Let |w|S denote the length of w as a word in S∗. In this paper, N will contain zero, and we will write N+ = N \{0}. A function ω : S → N+ will be called a weight function. We extend this to ∗ ω : S → N+, so that ω(s1s2 ··· sl)= ω(s1)+ ω(s2)+ ··· + ω(sl) for any word s1s2 ··· sl. Define the weight of a group element as ω(g) = min{ω(w) | w ∈ S∗, w = g}. If ω(s) = 1 for all s ∈ S, this gives the usual notion of word length. n n Write X := S ∩ Z = {x1,...,xr} and Y := S \ (S ∩ Z ) = {y1,...,ys}, and call any word in Y ∗ a pattern. Definition 2.1. Let patt: S → Y be the map

ǫ if si ∈ X patt: si 7→ . (si if si ∈ Y This extends to a monoid patt: S∗ → Y ∗, which records those generators in a word which are not contained in Zn. We call patt(w) the pattern of w. RATIONAL GROWTH IN VIRTUALLY ABELIAN GROUPS 5

Definition 2.2. For a pattern π = yi1 yi2 ··· yik of length k, define the set of π-patterned words in S∗ as w w w π w1 wr r+1 w2r 2r+1 kr+1 wkr+r N W = {x1 ··· xr yi1 x1 ··· xr yi2 x1 ··· yik x1 ··· xr | wj ∈ }. Any group element represented by a word with pattern π can be represented by a word in W π (since powers of elements of X commute). For a fixed pattern π, W π is in one-to-one correspondence with non-negative integer vectors of length kr + r. We define m(π)= kr + r. When it is clear which pattern this refers to, we will just write m. Definition 2.3. Define a map ϕ: W π → Nm via

w1

w w w w2 w1 wr r+1 w2r 2r+1 kr+1 wkr+r   x1 ··· xr yi1 x1 ··· xr yi2 x1 ··· yik x1 ··· xr 7→ . , .   w   m which records the powers of the generators contained in Zn. Forw ∈W π, write ~w := ϕ(w), and for a subset V ⊆ W π, write −→ V := ϕ(V ) ∈ Nm. Note that ϕ is a bijection. Let w ∈ W π with the above form. Then the weight of w is given by k r k

(4) ω(w)= ω(xi)wjr+i + ω(yij ). j=0 i=1 j=1 X X X Let T ⊂ G be a choice of coset representatives for Zn\G. We can then T express an element of G uniquely as (a1,...,an) t for aj ∈ Z, t ∈ T . In order to pass from a patterned word to this standard form, we introduce some constants. T n Notation 2.4. For each xi ∈ X, let xi = (z1i,z2i,...,zni) ∈ Z , and define the n × r matrix

Z := [zji] = x x ··· xn , 1≤j≤n,1≤i≤r  1 2  which encodes the abelian part of the generating set.  We then have

w1 w1 w2 wr . Zn (5) x1 x2 ··· xr = Z  .  ∈ w  r   for any powers wi ∈ N. 6 ALEX EVETTS

Now we encode conjugation of elements of Zn via matrix multiplication. Let Zn −1 Zn ⊳ ei ∈ be the ith standard basis vector, and yk ∈ Y . Then ykeiyk ∈ G, −1 and we will let ykeiyk = (γ1i,k,γ2i,k,...,γni,k) for each 1 ≤ i ≤ n, yk ∈ Y . Define the n × n matrix

−1 T yke1yk −1 yke2y  k  Γk := [γji,k]1≤i,j≤n = . . .    y e y−1   k n k    Then

a1 a1 a2 a2   −1   (6) yk . yk =Γk . . .     an an         for any yk ∈ Y , ai ∈ Z. n So we can express powers of the xi generators as vectors in Z , and we can move powers of yk past such vectors by using the identity yk(a1,...,an) = T Γk(a1,...,an) yk. For the word

w1 w2 wr wr+1 w2r w2r+1 wkr+1 wkr+r w = x1 x2 ··· xr yi1 x1 ··· xr yi2 x1 ··· yik x1 ··· xr , we can use the identities (5) and (6) to move all the yk generators to the right, modifying the powers of xi as we go, without changing the element that is represented. Thus

w1 wr+1 wkr+1 (7) w = Z . +Γ Z . + ··· +Γ Γ ··· Γ Z . π   .  i1  .  i1 i2 ik  .   w w w   r  2r   m        To express this more compactly, we introduce further notation.  Definition 2.5. Consider the n × m matrix formed by placing the matrices

Z, Γi1 Z, Γi1 Γi2 Z... next to each other. The transposes of the rows of this π new matrix are m-dimensional vectors which we will call Ai as follows:

π T (A1 ) (Aπ)T  2  . := Z Γi1 Z Γi1 Γi2 Z ··· Γi1 Γi2 ··· Γik Z . .    π T    (An)    Definition 2.6. The word π itself represents some element of G, so we in- π troduce integers Bi so that we can write π in the standard form given by the RATIONAL GROWTH IN VIRTUALLY ABELIAN GROUPS 7 choice of coset representatives as π B1 π B2 π =  .  tπ, .   Bπ  n    for tπ ∈ T . Now we may rewrite equation (7) using scalar products as π π A1 · ~w B1 π π A2 · ~w B2 (8) w =  .  +  .  tπ.  . .      Aπ · ~w Bπ  n   n  π     So words in W represent the same element of G if and only if the scalar  π products of the corresponding vectors with the Ai s agree. π Zm π Z Remark 2.7. We emphasise that Ai ∈ , Bi ∈ , and tπ ∈ G are constant in the sense that they depend only on the pattern π. In particular, any two words with the same pattern represent elements of the same coset. π Nm Definition 2.8. For a pattern π = yi1 ··· yik of length k, let An+1 ∈ record the weights of the xi generators, ordered as follows: π T An+1 = (ω(x1),ω(x2),...,ω(xr),...,ω(x1),ω(x2),...,ω(xr)) . r weights, repeated k+1 times, giving an m-dimensional vector π Furthermore, let| Bn+1 record the weight{z of the word π, i.e. } k π Bn+1 = ω(yij ) j=1 X We can then express equation (4) more compactly using a scalar product: π π (9) ω(w)= An+1 · ~w + Bn+1. π π We now have a collection of vectors Ai and integers Bi which together put words in W π into the chosen standard form. 2.2. Structure constants for testing conjugacy. In what follows we in- troduce notation that we will need to prove Theorem 5.1. As above, we fix a transversal T for Zn\G. In a similar manner to above, we encode conjugation of an element

a1 . Zn ⊳  .  ∈ G a  n   8 ALEX EVETTS by some other element of G via multiplication by a matrix. Let t ∈ T , and ei n −1 n be the ith standard basis vector in Z . Then teit ∈ Z by normality. Let −1 T teit = (δ1i,t,δ2i,t,...,δni,t) , and write ∆t for the n×n matrix [δji,t]1≤i,j≤n −1 whose columns are teit . Then we have

a1 a1 . −1 . t  .  t = ∆t  .  a a  n  n     for any ai ∈ Z. Fix a pattern π. Recall the matrices Z, and Γj for each element yj ∈ Y . If

w1 w2 wr wr+1 wr+2 w2r wkr+1 wm π w = x1 x2 ··· xr yj1 x1 x2 ··· xr yi2 ··· yik x1 ··· xr ∈ W , −1 consider the group element twt for t ∈ T . With ∆t as defined above, equation (7) yields

w1 wr+1 −1 . . twt = ∆tZ  .  + ∆tΓj1 Z  .  + ···  w w  r  2r      w  kr+1 (10) + ∆ Γ ··· Γ Z . tπt−1. t j1 jk  .  wm      We make definitions analogous to 2.5 and 2.6 above.  Definition 2.9. For each pattern π, and t ∈ T , place the matrix products in π order and define m-dimensional vectors Ai,t as follows

π T A1,t π T  A2,t   := ∆ Z ∆ Γ Z ··· ∆ Γ ... Γ Z . . t t j1 t j1 jk  .    T    Aπ   n,t    Now we can express equation (10) in terms of scalar products:

π A1,t · ~w π A2,t · ~w twt−1 =  .  tπt−1. .   Aπ · ~w  n,t    Definition 2.10. We have tπt−1 = xs for some s ∈ T , x ∈ Zn. This x π π T depends only on π and t. Write x = B1,t,...,Bn,t .  RATIONAL GROWTH IN VIRTUALLY ABELIAN GROUPS 9

Thus we have π π A1,t · ~w B1,t Aπ · ~w Bπ − 2,t 2,t (11) twt 1 =  .  +  .  s.  . .      Aπ · ~w Bπ   n,t   n,t Now we have a collection of vectors that we can use to test whether two words   represent conjugate elements (provided we know that their coset representa- tives are conjugate). The following Lemma makes this precise. Lemma 2.11. Let v ∈ W π and w ∈ W µ for patterns π and µ, and let t ∈ T . Then v = twt−1 if and only if (1) π and tµt−1 are in the same Zn-coset, and π π µ µ (2) Ai · ~v + Bi = Ai,t · ~w + Bi,t for each 1 ≤ i ≤ n. Proof. From equation (8) we have

π π π T π π π T v = (A1 · ~w, A2 · ~w, . . ., An · ~w) + (B1 ,B2 ,...,Bn ) tπ. From equationn (11) we have o

−1 µ µ µ T µ µ µ T twt = A1,t · ~w, A2,t · ~w, . . ., An,t · ~w + B1,t,B2,t,...,Bn,t s where s is determinedn by the element tµt−1, and the result follows.  o  2.3. Reduction to finitely many patterns. In general, there are infinitely many patterns in Y ∗. It will be essential for the results on growth series that we have only finitely many patterns. Extend S to a new generating set

(12) S = {s1s2 ··· sk | si ∈ S, 1 ≤ k ≤ d}. For eachs ˜ ∈ S, assign the weight ω(˜s), which induces a weighted length e function ω on the words in S∗, and in turn on elements of G. This func- tion preserves thee weight of elements of G (although the length will change), meaning thee growth functionse are equal: ω n ωe n σG,S(n)z = σG,Se(n)z . n≥ n≥ X0 X0 Proposition 2.12 (Proposition 11.3 of [3]). Suppose g ∈ G. Then there exists a minimal weight word w ∈ S∗, with a pattern of length at most d (with respect to S∗), such that w = g. e This leads us to make the following definition. e Definition 2.13. For a virtually abelian group G generated by S, let Zn ∗ P := {π ∈ (S \ (S ∩ )) | |π|Se ≤ d}. Note that P is a finite set. e e 10 ALEX EVETTS

Then the previous proposition implies that any g ∈ G can be represented by a minimal weight word with a pattern in P .

2.4. Polyhedral Sets. The following definition and results follow Benson’s work [3]. However, the ideas appear in model theory as early as Presburger [25]. Results regarding rationality can be found in [12], and the ideas appear in the theory of Igusa local zeta functions (see [9]). These last are linked to subgroup growth [19], a different notion of growth in groups not considered here.

Definition 2.14. Let m ∈ N+, and let · denote the Euclidean scalar product. m Then for any u ∈ Z , a ∈ Z, b ∈ N+: (1) an elementary set is any subset of Zm of the form {z ∈ Zm | u·z = a}, {z ∈ Zm | u · z >a}, or {z ∈ Zm | u · z ≡ a mod b}, (2) a basic polyhedral set is any finite intersection of elementary sets, (3) a polyhedral set is any finite union of basic polyhedral sets. If P ⊂ Zm is polyhedral and additionally P ⊆ Nm, we call P a positive polyhedral set. We record some crucial facts about polyhedral sets. Proposition 2.15 (Proposition 13.1 of [3]). Polyhedral sets in Zm are closed under finite unions, finite intersections, and set complement.

′ Proposition 2.16 (Propositions 13.7 and 13.8 of [3]). Let E : Zm → Zm be an integral affine transformation (for some m,m′ > 0). That is, there is some ′ m′ × m matrix with integer entries and some q ∈ Zm such that E(p)= Ap + q ′ for p ∈ Zm. If P ⊆ Zm is a polyhedral set then E(P) ⊆ Zm is a polyhedral ′ set. If Q ⊆ Zm is a polyhedral set then the preimage E−1(Q) ⊆ Zm is a polyhedral set. We note that projection onto any subset of the coordinates of Zm is an integral affine transformation. Let P ⊆ Nm be a positive polyhedral set. Given some choice of weights Nm m (ω1,...,ωm) for the coordinates of , we assign the weight i=1 aiωi to the T vector (a ,...,am) ∈P. Define 1 P ω σP (n) = #{p ∈ P | ω(p)= n}, and the resulting weighted growth series ∞ ω ω n SP (z)= σP (n)z . n=0 X Proposition 2.17 (Proposition 14.1 of [3]). If P is a positive polyhedral set, ω the weighted growth series SP (z) is a rational function. RATIONAL GROWTH IN VIRTUALLY ABELIAN GROUPS 11

Benson’s result that virtually abelian groups have rational weighted growth series with respect to all generating sets follows from the following Theorem. Theorem 2.18 (Theorem 1.2 of [3]). Let G be a virtually abelian group, with any choice of finite weighted generating set S. For each pattern π ∈ P , with −→ P as in definition 2.13, there exists a set U π ⊂ S∗ such that U π ⊂ Nm(π) π is polyhedral, and the disjoint union π∈P U forms a language of unique minimal-weight representatives for the elements of eG. S π m(π) Since the xi generators in U correspond to the coordinates of N , and the contribution from the y generators is constant within U π, rational growth −→ k of U π (in the sense of Proposition 2.17) implies rational growth of U π with respect to S, and hence with respect to S. We will need the following Lemma concerning polyhedral sets. e Lemma 2.19. Let P be a polyhedral subset of Zm for some m ≥ 1. Suppose Zm k there exist polyhedral sets X1,...,Xk ⊂ such that P ⊆ i=1 Xi. Then there exist polyhedral sets Yi ⊆ Xi for each i such that Yi ∩ Yj = ∅ for i 6= j k S and P = i=1 Yi.

Proof. WeS induct on k. Consider k = 1, i.e. P ⊆ X1. Let Y1 = P ∩ X1, which is polyhedral as an intersection of polyhedral sets. In other words P = Y1. Now assume the statement is true for some k> 1. Let P be some polyhedral k+1 set, with polyhedral sets X1,...,Xk+1 such that P ⊆ i=1 Xi. Consider the polyhedral set k S Q := P ∩ Xi. i=1 [ k Now since Q⊆ i=1 Xi, the inductive hypothesis gives polyhedral sets Yi ⊆ Xi for 1 ≤ i ≤ k such that Yi ∩ Yj = ∅ for each i 6= j, 1 ≤ i, j ≤ k, and S k Q = Yi. i=1 [ Now let Yk+1 = P\Q, also a polyhedral set. We have Yk+1 ∩Q = ∅, and Yk+1 ⊆ Xk+1 (since the Xis cover P and Yk+1 does not intersect Q), and by definition k+1 P = Q ∪ Yk+1 = Yi. i=1 [ So the statement holds for k + 1. 

2.5. N-fold patterns. We develop a framework for dealing with N-tuples of patterned words for some finite N. 12 ALEX EVETTS

Definition 2.20. Let Q be any set of patterns (for a virtually abelian group with some choice of finite generating set). Let N be a positive integer. An N-fold pattern will be an N-tuple of patterns from Q. We will write π = N (π1, π2,...,πN ) ∈ Q . π Given an N-fold pattern π = (π1, π2,...,πN ), write W for the set of N-tuples of words with patterns given by π. More precisely, π W = w(1), w(2),...,w(N) w(i) ∈ W πi , 1 ≤ i ≤ N π n  o Let m( )= i m(πi). As above, we extract the powers of the xi generators with respect to each pattern and note that elements of W π are in one-to-one correspondenceP with vectors in Nm(π) via −−→ w(1) (1) (N) . w ,...,w 7→  .  . −−−→   w(N)     In the following Lemma, we show that if we have a polyhedral set of N- tuples in Nm(π), we can extract a minimal-weight element of each tuple, in a manner that preserves rational growth. Lemma 2.21. Let G be virtually abelian, generated by S, with weight func- π −→ tion ω. Fix some N-fold pattern π = (π1, π2,...,πN ). Let V ⊂ W . If V is a polyhedral set then there exists a language L ⊂ S∗ in one-to-one corre- spondence with V such that for each (v1,...,vN ) ∈ V , L contains exactly one element w ∈ {v1,...,vN }, with ω(w) ≤ ω(vi) for each 1 ≤ i ≤ N, and such that L has rational weighted growth series. Proof. Let

(1) (N) π (j) (k) Xj (π)= w ,...,w ∈ W ω w ≤ ω w , 1 ≤ k ≤ N n π      o be those elements in W where the jth component word has minimal weight. πk For each πk, define the m(π)-dimensional vector D as follows: 0 . k−1 . P m(πi) zeros  .   i=1 0  πk  πk  D := A  m(πk) rows  n+1   0      .  N m π  .   Pi=k+1 ( i) zeros    0      (1) (N) π Then for some N-tuple of words w ,...,w ∈ W , whose corresponding π −−→ Nm( ) πk πk (k) vector is ~z ∈ , we have D · ~z = An+1 · w . RATIONAL GROWTH IN VIRTUALLY ABELIAN GROUPS 13

For w ∈ W πj and w′ ∈ W πk , equation (9) implies that

′ πj πj πk ~′ πk ω(w) ≤ ω(w ) ⇔ An+1 · ~w + Bn+1 ≤ An+1 · w + Bn+1. Therefore we have −−−−→ π π π Nm( ) πj j πk πk Xj ( )= ~z ∈ | D · ~z + Bn+1 ≤ D · ~z + Bn+1 1≤k≤N k\=6 j n o π π Nm( ) πj πk πk j = ~z ∈ | (D − D ) · ~z ≤ Bn+1 − Bn+1 . 1≤k≤N k\=6 j n o

−−−−→ m(π) Thus each Xj (π) is a polyhedral subset of N (since it is a finite inter- section of elementary sets). Since every N-tuple in V must have at least one −−−−→ −→ N π minimal-weight element, we have V ⊆ j=1 Xj ( ). Lemma 2.19 then gives −−−→ −−−−→ us polyhedral sets Yj (π) ⊆ Xj (π) for each j which are pairwise disjoint, and −−−→ S −→ N π such that V = j=1 Yj ( ). π Let ρ : Nm( ) → Nm(πj ) be the projection onto just the coordinates cor- j S responding to the jth component of the N-tuple. Since images of polyhedral sets under projections are still polyhedral by Proposition 2.16, we have N −−−→ polyhedral sets of the form ρj (Yj (π)), each corresponding to a collection of words which are minimal-weight representatives for their N-tuple, and each growing rationally with respect to the generators of G (by Proposition 2.17). −→ −→ Since the Yj s are disjoint, and cover V , the union of these polyhedral sets corresponds to a language of unique, minimal-weight representatives for V , which grows rationally. 

3. Relative growth series Given a group G with finite generating set S, consider a subset H ⊂ G. One can construct the relative (weighted) growth series of H with respect to the generators of G by defining σ(n) to be the number of elements of H with weight n. We prove that for any subgroup of a virtually abelian group, the relative weighted growth series is rational (for any finite generating set and positive integer weight function). We do this by showing that we can use linear algebra to test whether or not a word in S∗ represents an element of H, and thus describe the language of words whose image in G is contained in H in terms of polyhedral sets. We then intersect with the language of minimal representatives for the elements of G to obtain a language of minimal representatives for the elements of H, which grows rationally. The following two Lemmas are elementary and are stated without proof. Lemma 3.1. Any subgroup of a virtually abelian group is virtually abelian. 14 ALEX EVETTS

In the next Lemma, we establish criteria for an element of a virtually abelian group to lie in a chosen subgroup.

Lemma 3.2. Let G be a finitely generated virtually abelian group, with nor- mal subgroup Zn of finite index d as usual. Let H < G, and choose a set n {h1,...,hc} ∈ H of coset representatives for H ∩ Z \H. Then this set can be n extended to a set {h1,...,hc,...,hd} of coset representatives for Z \G, and n an element xhi ∈ G, with x ∈ Z , is in H if and only if (1) 1 ≤ i ≤ c, and (2) x ∈ H ∩ Zn.

Theorem 3.3. Let G be a finitely generated virtually abelian group and let H be any subgroup of G. Then H has rational weighted growth series relative to any choice of generators of G (with any weight function).

Proof. The strategy of the proof is to use the criteria given in Lemma 3.2 to detect those elements of G which are contained in H, pattern by pattern, and show that they form polyhedral sets. We can then use the sets U π (from Theorem 2.18) to find minimal representatives for the elements of H. Fix a generating set S for G, and a weight function ω. We consider the expanded generating set S of Section 2.3, and the corresponding finite set of patterns P as in Definition 2.13. Since words with the same pattern represent elements of the same Zn-cosete (see Remark 2.7), only words whose pattern c Zn represents an element of i=1 hi can possibly be in H. Call the set of such patterns PH . F π Fix one of these patterns, π ∈ PH , and the resulting vectors Ai and integers π π Bi as in section 2. Consider a word w ∈ W . By design, we have an hi with n 1 ≤ i ≤ c so that w ∈ Z hi, so criterion (1) is satisfied. Now by criterion (2), n n w represents an element of H if and only if w ∈ (H ∩ Z )hi ⊂ Z hi. Since

π π A1 · ~w B1 π π A2 · ~w B2 w =  .  +  .  hi,  . .      Aπ · ~w Bπ  n   n      w ∈ H if and only if   π π A1 · ~w B1 π π A2 · ~w B2 (13)  .  +  .  ∈ H ∩ Zn. . .     Aπ · ~w Bπ  n   n      Now H ∩ Zn is a (free abelian) subgroup of Zn. Suppose it has dimension n f, and choose a basis {b1,..., bf }⊂ Z . Then w satisfies (13) if and only if RATIONAL GROWTH IN VIRTUALLY ABELIAN GROUPS 15 there exist integers a1,...,af so that π π A1 · ~w B1 π π A2 · ~w B2  .  +  .  = a1b1 + ··· + af bf . . .     Aπ · ~w Bπ  n   n  In other words,     π π b b b b (14) Ai · ~w + Bi = ei · (a1 1 + ··· + af f )= a1(ei · 1)+ ··· + af (e1 · f ) for each 1 ≤ i ≤ n. We will now express the set of all vectors satisfying (14) as a polyhedral set. For each 1 ≤ i ≤ n, define the (m + f)-dimensional vector

Aπ m rows  i   π  Ci = −ei · b1    .    .   f rows   −e · b    i f    π where the first m entries are the entries of the vector Ai , and the last f entries are −1 times the ith component of the basis vectors. Now a vector ~w ∈ Zm satisfies (14) for some i precisely when there is a vector ~v ∈ Zm+f , T π π with entries v1, v2,...,vm+f , such that ~w = (v1,...,vm) and Ai · ~w + Bi = ei · (vm+1b1 + ··· + vm+f bf ). We rewrite this last equation as

v1 ei · b1 vm+1 v2 ei · b2 vm+2 π π Ai ·  .  −  .  ·  .  = −Bi . . .       v  e · b  v   m  i f   m+f  i.e.       π π Ci · ~v = −Bi . m+f m Let pm : Z → Z denote projection onto the first m coordinates. We can then express the set of all ~w that satisfy (13) as the following n π Zm+f π π Nm X = pm ~v ∈ Ci · ~v = −Bi ∩ . i=1 ! \  Zm+f π π Note that {~v ∈ | Ci · ~v = −Bi } is an elementary set for each i, and therefore the intersection is a (basic) polyhedral set. Therefore Xπ is also a polyhedral set. It corresponds to all words in W π which represent elements of the subgroup H. We wish to find unique minimal-weight representatives for these elements. 16 ALEX EVETTS

Recall that the set U π ⊂ W π is a set of minimal-weight representatives for −→ the elements of W π. The intersection U π ∩ Xπ consists of a unique, minimal- weight vector representing each element of H that lies in W π (and is not represented by a word with another pattern, of smaller weight). Since this intersection is polyhedral, it grows rationally with respect to the generators of G. So we have finitely many rationally growing sets whose (disjoint) union corresponds to a language of unique, minimal-weight representatives for H. The relative growth series of H is the sum of the growth series of these in- dividual sets, so is itself rational. Note that we have not had to consider overlaps as the sets U π are already non-overlapping. 

Corollary 3.4. Finite unions of of virtually abelian groups have rational relative growth. Proof. First note that the growth series of a finite disjoint union of subsets of G is simply the sum of their individual growth series, and for subsets A ⊂ B ⊆ G, the growth series of B \ A is the difference of their individual growth series. We will induct on the number of subgroups in the finite union. For subgroups H1 and H2, let I = H1 ∩ H2. Then we can express their union as a disjoint union of three subsets,

H1 ∪ H2 = I ∪ (H1 \ I) ∪ (H2 \ I).

Since H1, H2, and I are all subgroups, they have rational relative growth, so each term in the above expression does also, and so H1 ∪ H2 has rational growth series. Now assume the union of k subgroups has rational growth. Consider a union of k + 1 subgroups:

k+1 k Hi = Hi ∪ Hk+1. i=1 i=1 [ [ k Let J = i=1 Hi ∩ Hk+1. Then we have a disjoint union   S k+1 k Hi = J ∪ Hi \ J ∪ (Hk+1 \ J) i=1 i=1 ! [ [ k+1 and so if J has rational growth then i=1 Hi has rational growth. But we can write J as the union of k subgroups: S k

J = (Hi ∩ Hk+1) i=1 [ so it has rational growth by the inductive hypothesis.  RATIONAL GROWTH IN VIRTUALLY ABELIAN GROUPS 17

4. Coset Growth series In this section we demonstrate the following: Theorem 4.1. Let G be a virtually abelian group with any choice of finite, weighted generating set S, and H any subgroup of G. Then the set of right cosets H\G has rational weighted growth series with respect to S. This is proved in two parts. Firstly, we generalise the main result of [3] to show that if a subgroup F is contained in a finite index abelian subgroup of G then there exists a language of minimal representatives for the cosets of F in G that grows rationally. For a general subgroup H, its intersection with the finite index abelian subgroup of G is abelian, and of finite index in H. We then consider the language of minimal representatives for the cosets of this intersection subgroup in G, and from these we choose a language of representatives for the cosets of H. We show that this language can be chosen so that it grows rationally. We prove the following special case of Theorem 4.1. Theorem 4.2. Let G be a virtually abelian group, with any choice of finite weighted generating set S, and with finite index normal subgroup Zn. If F is a subgroup of G contained in Zn, then the weighted growth of F \G with respect to S is rational. Theorem 1.2 of [3] is the case where F is the trivial group. The proof given here follows a similar structure to that in [3]. For each pattern π ∈ P , we establish an ordering on the words of W π which respects their weight, and use this to find a polyhedral set of minimal representatives for the cosets that intersect W π. We then show that the overlaps between these sets are also polyhedral, so can be removed while preserving rationality. Note that if F has finite index in G, F \G is finite and so the growth series is a polynomial, and so trivially rational. From now on we assume that F has infinite index in G, and hence [Zn : F ] is also infinite. We first establish a criterion for when two words represent elements of the π π same F -coset. We will again use the vectors A1 ,...,An (of dimension m(π)), µ µ π π µ µ A1 ,...,An (of dimension m(µ)) and integers B1 ,...,Bn , and B1 ...,Bn de- fined in section 2. Proposition 4.3. Let v ∈ W π, w ∈ W µ, for some patterns π,µ. Let F < Zn n be of rank f ≤ n, with basis {b1, b2,..., bf }⊂ Z . Then v and w are in the same coset of F in G if and only if (1) π and µ are in the same coset of Zn in G, and (2) there exist integers a1,...,af such that f π π µ µ A · ~v + B − (A · ~w + B )= ei · aj bj i i i i   j=1 X   18 ALEX EVETTS

for each 1 ≤ i ≤ n (where ei as usual denotes the ith standard basis vector in Zn).

Proof. Recall (see (8)) that the group elements represented by v and w are given by

π π π T π π π T v = (A1 · ~v, A2 · ~v,...,An · ~v) + (B1 ,B2 ,...,Bn ) tπ ∈ G and 

µ µ µ T µ µ µ T w = (A1 · ~w, A2 · ~w, . . ., An · ~w) + (B1 ,B2 ,...,Bn ) tµ ∈ G

n respectively, where tπ and tµ are the chosen representatives for the cosets Z π and Znµ. Now two words v, w represent elements of the same F -coset if and only if v(w)−1 ∈ H. This is equivalent to the existence of integer coefficients a1,...,af such that f −1 v(w) = aj bj . j=1 X Suppose that our words v and w do represent the same F -coset. Since n n Fg ⊂ Z g for any g ∈ G, v and w represent the same Z -coset, so tπ = tµ, which is precisely condition (1). We have

−1 π π π T π π π T v(w) = (A1 · ~v, A2 · ~v,...,an · ~v) + (B1 ,B2 ,...,Bn ) tπ −1 µ µ µ T µ µ µ T  · tµ − (A1 · ~w, A2 · ~w, . . . , An · ~w) − ( B1 ,B2 ,...,Bn) π π n µ µ π π µ µ T o = (A1 · ~v + B1 − A1 · ~w − B1 ,...,An · ~v + Bn − An · ~w − Bn ) . Thus

f π π µ µ π π µ µ T b Zn (A1 · ~v + B1 − A1 · ~w − B1 ,...,An · ~v + Bn − An · ~w − Bn ) = aj j ∈ , j=1 X which is equivalent to condition (2). Conversely, if v and w satisfy the conditions, then tπ = tµ and so

f −1 π π µ µ π π µ µ T b v(w) = (A1 ·~v+B1 −A1 · ~w−B1 ,...,An ·~v+Bn −An · ~w−Bn) = aj j j=1 X i.e. v(w)−1 ∈ F . 

We note the following special case, when π = µ. RATIONAL GROWTH IN VIRTUALLY ABELIAN GROUPS 19

π n Corollary 4.4. Let v, w ∈ W , F < Z with rank f and basis {b1, b2,..., bf }⊂ Zn. Then v and w are in the same F -coset if and only if there exist integers a1,...af so that f π b Ai · (~v − ~w)= ei · aj j j=1 X for each 1 ≤ i ≤ n.

In what follows we will need a version of Dickson’s Lemma.

Definition 4.5. Let I ⊆{1, 2,...,m} for some m, and define

m QI = {x ∈ Z | ei · x ≥ 0, if i ∈ I, ei · x ≤ 0, if i∈ / I},

m which is a closed orthant of Z . Define the ‘coordinate ordering’ on QI as follows: x ≤I y if and only if ei · x ≤ ei · y for i ∈ I and ei · x ≥ ei · y for i∈ / I.

Lemma 4.6 (Lemma A of [13]). Let X ⊆ QI for some I ⊆ {1, 2,...,m}. Then there exists a finite subset Y ⊂ X such that for all x ∈ X, there exists y ∈ Y with y ≤I x.

The result is more commonly stated for the case I = {1, 2,...,m} (and so m QI = N ) but this more general version follows directly from symmetry. We now build a polyhedral set of minimal weight coset representatives, for each pattern π. The following is a modification of the arguments in Section 6 of [3].

Proposition 4.7. Fix a pattern π ∈ P , and with F an infinite index subgroup of Zn ⊳ G, consider the set (F \G)π of right cosets which contain an element represented by a word in W π, i.e.

(F \G)π = {Fg ∈ F \G | Fg ∩ W π 6= ∅}.

π π Then there exists a set VF ⊂ W consisting of minimal-weight (amongst π π −→π W ) representatives for every coset in (F \G) , with the property that VF is a polyhedral set.

Proof. We will define an ordering on words in W π that is consistent with the weight ordering, and yields a unique minimal representative for each coset in (F \G)π. b b π π As above, fix a basis { 1,..., f } for F . Recall the vectors A1 ,...,An+1 ∈ Zm π π π π . Choose standard basis vectors An+2,...,AK so that the set {An+1,...,AK } spans Zm. Define an order on the words of W π as follows. We will write v ≤π w if and only if either v = w or there exist integers a1,...,af , and 20 ALEX EVETTS

0 ≤ i ≤ K − n such that f π b Ak · (~v − ~w)= ek · aj j , for 1 ≤ k ≤ n, j=1 π X Ak · (~v − ~w)=0, for n +1 ≤ k 0. We show that this is a partial order on W π. Firstly, note that the ordering is reflexive by definition. Next, we show transitivity. Suppose u,v,w ∈ W π with u ≤π v ≤π w. If u = v or v = w then clearly u ≤π w, so suppose u 6= v 6= w. ′ ′ ′ So we have integers a1,...,af ,a1 ...,af and 0 ≤ i,i ≤ K − n with

f π b Ak · (~u − ~v)= ek · aj j , for 1 ≤ k ≤ n, j=1 π X Ak · (~u − ~v)=0, for n +1 ≤ k 0 and f π ′ b Ak · (~v − ~w)= ek · aj j , for 1 ≤ k ≤ n, j=1 π X ′ Ak · (~v − ~w)=0, for n +1 ≤ k 0. Then for 0 ≤ k ≤ n we have f π π π ′ b Ak · (~u − ~w)= Ak · (~u − ~v)+ Ak · (~v − ~w)= ek · (aj + aj) j j=1 X π by linearity of the scalar product. Furthermore, Ak · (~u − ~w)=0 for n +1 ≤ ′ π ′ k< min(i,i ) and Ak · (~u − ~w) > 0 for k = min(i,i ), and thus u ≤π w, so ≤π is transitive. For antisymmetry, note that if v ≤π w and w ≤π v then either π π v = w or we must have Ak ·(~v − ~w)= Ak ·( ~w −~v) = 0 for each n+1 ≤ k ≤ K, π Zm and since the corresponding Ak s span , we must have ~v = ~w, i.e. v = w. Thus the order is a well-defined partial order (although not a total order). If we restrict ourselves to the words representing a single H-coset, this becomes a well-ordering. To see this, suppose we have an infinite descending chain of words in W π that represent the same coset:

w1 ≥π w2 ≥π ··· . π Nm Nm Since Ak ∈ for k>n and ~wi ∈ for all i, the sequences π π Ak · ~w1 ≥ Ak · ~w2 ≥ · · · RATIONAL GROWTH IN VIRTUALLY ABELIAN GROUPS 21 for each k>n consist of non-negative integers and so must stabilize, say after π π ik steps. Let imax = maxnn span , wimax = wimax +j and the sequence stabilizes. Note that two words representing the same coset can always be compared under ≤π, since Corollary 4.4 implies there are integers a1,...,af satisfying the definition. π Thus there is a unique ≤π-minimal word in W that represents each coset in (F \G)π. π π Note that if v ≤π w then An+1 · ~v ≤ An+1 · ~w and so ω(v) ≤ ω(w). Thus π the unique ≤π-minimal element in W that represents a given F -coset is also π π a minimal weight coset representative (amongst W ). Let VF denote the set π of all ≤π-minimal representatives in W , that is π π π (15) VF := {w ∈ W | If v ∈ W s.t. v ∈ F w then w ≤π v} . π To finish the proof, we need to show that VF corresponds to a polyhedral set in Zm. An element τ ∈ Zm will be called a translation (with respect to F , π) π if there exist integers a1,...,af and 0 ≤ i ≤ K − n such that Ak · τ = f b ek · j=1 aj j for 1 ≤ k ≤ n and π π π P An+1 · τ = ··· = An+i−1 · τ =0, An+i · τ > 0. Let T denote the set of all such translations. Suppose v, w ∈ W π. Then it is clear that w ∈ F v with w ≤π v if and only if there exists some τ ∈ T m with ~v = ~w + τ. The set T is contained in Z . Consider T ∩ QI for some I ⊆ {1,...,m}. By Lemma 4.6, there exists a finite set TI ⊂ T ∩ QI such that each element τ ∈ T ∩ QI has a bound τ0 ∈ TI such that τ0 ≤I τ. Let Tmin = I TI , the union of the minimal translations across all orthants. We now claim that S −→π Nm Nm VF = \ (τ + ). τ∈T [min It is not hard to see that this is a polyhedral set, which proves the Proposition. π Nm Nm To see the claim, first suppose that v ∈ VF but ~v/∈ \ τ∈Tmin (τ + ). m m So ~v ∈ τ + N for some τ ∈ Tmin, i.e. ~v = τ + ~w for some ~w ∈ N . This implies that there is some w ∈ W π which shares an F -cosetS with v such that w ≤π v. But this implies that v is not minimal, contradicting the assumption π that v ∈ VF . Nm Nm π Conversely, suppose that ~v ∈ \ τ∈Tmin (τ + ) and v∈ / VF . So π there exists some τ ∈ T and ~v0 ∈ V , with ~v = ~v0 + τ and v0 ∈ F v. In F S other words ~v − ~v0 is a translation. Choose I so that ~v − ~v0 ∈ QI , and then m m τ0 ∈ TI so that τ0 ≤ ~v − ~v0. We claim that ~v − τ0 ∈ N , i.e. ~v ∈ τ0 + N , contradicting our assumption. Indeed, for i ∈ I we have ei · τ0 ≤ ei · τ, so ei·(~v−τ0) ≥ ei·(~v−τ)= ei·~v0 ≥ 0, and for i∈ / I we have ei·(~v−τ0) ≥ ei·τ0 ≥ 0. m So for all i, ei · (~v − τ0) ≥ 0, and hence ~v − τ ∈ N as claimed.  22 ALEX EVETTS

Since Fg ⊂ Zng for any g ∈ G, any two words that represent the same F - coset must lie in the same Zn-coset. So we consider each Zn-coset separately. Section 2 tells us that for a given Zn-coset, say Znt for t ∈ T , there is a finite set of patterns Pt, over the extended generating set S, whose patterned sets contain representatives for all the elements of the coset (and no other −→π cosets). We take the corresponding polyhedral sets VF frome equation (15) for each π ∈ Pt, and combine them to find a language of representatives for the F -cosets within Znt. We may have pairs of words with different patterns that both represent the same coset, but we only wish to count the minimal −→π one. To prove Theorem 4.2, we show that these overlaps between the VF s are polyhedral, so can be removed without losing rationality.

π,µ Definition 4.8. Let π,µ ∈ Pt for some t ∈ T . Define the set R (resp. π,µ π µ R∗ ) consisting of all those elements of VF where there is an element of VF of strictly smaller (resp. equal) weight that represents the same coset:

π,µ π µ R :={v ∈ VF | ∃u ∈ VF , u ∈ F v, ω(u) <ω(v)} π,µ π µ R∗ :={v ∈ VF | ∃u ∈ VF , u ∈ F v, ω(u)= ω(v)}. We need to discard all of Rπ,µ for every pair π 6= µ, since we only want minimal words. If there exist two of more minimal weight representatives for the same coset with equal weight, we must choose exactly one and discard the rest. We make the following definition.

Definition 4.9. Pick a total order on the finite set Pt, denoted π1 < π2 < ··· . Let

πk πk πk,πi πk,πj U := V \ R ∪ R∗ . F F   i6 k j

2n+2+f Let 1j ∈ Z be the vector with a 1 at the jth entry and zeroes everywhere else. Define the vectors

0 .  .   2n + 2 zeroes 0    Ei = 1i − 1i n + −ei · b1  + +1    −ei · b2       .  f rows  .      −e · b   i f      for each 1 ≤ i ≤ n, and let En+1 = 1n+1 − 12n+2. Then define the polyhedral sets

n 2n+2+f 2n+2+f Φ := φ ∈ Z | φ · Ei =0 ∩ φ ∈ Z | φ · En+1 > 0 i=1 \   and

n 2n+2+f 2n+2+f Φ∗ := φ ∈ Z | φ · Ei =0 ∩ φ ∈ Z | φ · En+1 =0 . i=1 \   π −→π Zn+1 Let E : VF → denote the integral affine transformation

π π A1 · ~w + B1 π π A2 · ~w + B2 Eπ : ~w 7→  .  .   Aπ · ~w + Bπ   n+1 n+1   −→ for any pattern π (and write (Eπ)−1X for the preimage in V π of any X ⊆ ′ F n+1 ′ k k Z ). For any k > k, write pk : Z → Z for the projection onto the first k coordinates. We will show that

−−→π,µ π −1 π −→π µ −→µ (16) R = (E ) pn+1 E (VF ) ×E (VF ) ∩ p2n+2(Φ) , h  i which is a polyhedral set since projection is an affine transformation. Indeed, π,µ µ suppose that v ∈ R . So there exists u ∈ VF such that u ∈ F v and ω(u) <ω(v). By Corollary 4.4, there exist integers a1,...,af such that

f π π µ µ b Ai · ~v + Bi − (Ai · ~u + Bi )= ei · aj j j=1 X 24 ALEX EVETTS

π π µ µ for each 1 ≤ i ≤ n, and An+1 · ~v + Bn+1 > An+1 + Bn+1. In other words

π π A1 · ~v + B1 π π A2 · ~v + B2  .  .  π π  An+1 · ~v + Bn+1  µ µ   A1 · ~u + B1   µ µ   A · ~u + B   2 2  ∈ Φ,  .   .   µ µ  A · ~u + B   n+1 n+1  a   1   .   .     a   f    π µ and hence (E (~v), E (~u)) ∈ p2n+2(Φ), so ~v is contained in the right hand side of (16). Conversely, let ~v ∈ Nm(π) be contained in the right hand side of (16). Thus

π π −→π µ −→µ E (~v) ∈ pn+1 E (VF ) ×E (VH ) ∩ p2n+2(Φ) h  i so there exists z ∈ Zn+1 with −→ π z π −→π µ µ (E (~v), ) ∈ E (VF ) ×E (VF ) ∩ p2n+2(Φ).   µ z µ µ µ µ T That is, there exists u ∈ VF with = (A1 · ~u + B1 ,...,An+1 · ~u + Bn+1) π T and there exist integers a1,...,af such that (E (~v), z,a1,...,af ) ∈ Φ, and together this means that

π π A1 · ~v + B1 π π A2 · ~v + B2  .  .  π π  An+1 · ~v + Bn+1  µ µ   A1 · ~u + B1   µ µ   A · ~u + B   2 2  ∈ Φ.  .   .   µ µ  A · ~u + B   n+1 n+1  a   1   .   .     a   f    RATIONAL GROWTH IN VIRTUALLY ABELIAN GROUPS 25

From the definition of Φ, this implies that

f π π µ µ b Ai · ~v + Bi − (Ai · ~u + Bi )= ei · aj j j=1 X π π µ µ π for each 1 ≤ i ≤ n and An+1 · ~v + Bn+1 > An+1 · ~u + Bn+1. So v ∈ VF and µ π,µ −−→π,µ there exists u ∈ VF with u ∈ F v and ω(u) <ω(v), i.e. v ∈ R , and so R has the polyhedral form (16) as claimed. In an exactly analogous way,

−−→π,µ π −1 π −→π µ −→µ R∗ = (E ) pn+1 E (VF ) ×E (VF ) ∩ p2n+2(Φ∗) . h  i 

We now use the previous result to prove Theorem 4.1, that is, to show that for an arbitrary subgroup H ≤ G, the set of right cosets H\G has rational growth with respect to any choice of weighted generating set for G. The proof relies on the understanding of the structure of subgroups in Lemma 3.2 and the rationality of coset growth for free abelian subgroups in Theorem 4.2.

Proof of Theorem 4.1. First, we consider the coset structure of G. Let Zn be the maximal free abelian normal subgroup of G. Then we have H ∩ Zn ⊳ H, with finite index c ≤ d. Fix a choice of transversal {h1,...,hc} for the cosets (H ∩ Zn)\H. As in Lemma 3.2, we extend this to a transversal for Zn\G, write T = {t1,...,td}⊇{h1,...,hc}. Suppose that the free abelian group n n H ∩ Z has rank f, and fix a basis {b1,..., bf }⊂ Z . Consider a coset Hg ∈ H\G. Following the above discussion, we may decompose H as the finite union of its H ∩ Zn-cosets. We may also write T g = (g1,...,gn) t for some gi ∈ Z and t ∈ T . Therefore

c g1 n . Hg = (H ∩ Z )hj . t.    .  j=1 g [  n     Since H ∩ Zn is also a subgroup of G, contained in Zn, Theorem 4.2 provides a minimal weight representative for each coset of H ∩ Zn in G of the form n (H ∩ Z )hj g, and therefore a collection of c candidates for a minimal-weight representative for Hg, since the minimal-weight representative for Hg is one n of the c minimal-weight representatives for the cosets (H ∩ Z )hj g. We will express these candidates as c-tuples of words, (whose patterns together make c-fold patterns, see Section 2.5), and show that they correspond to polyhedral sets, from which rationality will follow. 26 ALEX EVETTS

c Fix a c-fold pattern π = (π1,...,πc) ∈ P (recalling the definition of P from Definition 2.13), and define

(17) π (1) (c) π1 πc V ( )= w ,...,w ∈ UH∩Zn ×···× UH∩Zn |

n  (j) n ∃g ∈ G s.t. w ∈ (H ∩ Z )hj g, 1 ≤ j ≤ c , o πk Zn where UH∩Zn is the set of minimal representatives for the cosets of H ∩ in G as defined in Definition 4.9, where H ∩Zn plays the role of F . Each element of V (π) consists of a c-tuple of candidates for a minimal weight representative of −−−→ the coset Hg. By Lemma 2.21, if V (π) is polyhedral then there is a language Lπ of minimal representatives for those cosets represented by V (π), which grows rationally. Every element of G has a minimal-weight representative with a pattern in P (by definition of P ), and so in particular every coset has a minimal-weight representative with pattern in P , and is therefore represented π in some V ( ). Thus the union π∈P c Lπ forms a language of minimal weight representatives for the set of cosets H\G. Since P c is finite, this union is a finite union of polyhedral sets,S and thus has rational growth. −−−→ We now show that V (π) is indeed a polyhedral set for each π ∈ P c, which will complete the proof. For some tuple in V (π), consider the element g as T in equation (17). This can be expressed as (g1,...,gn) t for some gi ∈ Z and t ∈ T . π π So we can decompose V ( ) as a finite union t∈T V ( ,t) where S π (1) (c) π1 πc V ( ,t)= w ,...,w ∈ UH∩Zn ×···× UH∩Zn |

n  n (j) n ∃g ∈ Z t s.t. w ∈ (H ∩ Z )hj g, 1 ≤ j ≤ c . o n T We wish to write an element of (H ∩ Z )hj (g1,...,gn) t in the standard form defined in section 2.1. Recall from section 2.2 that for any element T −1 T s ∈ T , we have a matrix ∆s so that s(a1,...,an) s = ∆s(a1,...,an) for any integers ai. For two coset representatives s,t ∈ T , their product st will n not necessarily be in T . So let xst ∈ Z and τst ∈ T be such that st = xstτst. Now suppose that

g1 g1 g2 g2 n γ ∈ (H ∩ Z )hj  .  t ⊆ H  .  t. . .     g  g   n  n     RATIONAL GROWTH IN VIRTUALLY ABELIAN GROUPS 27

(j) (j) So there exist integers λ1 ,...,λf , so that

g1 f g2 (j)b   γ = λk k hj . t ! . k=1   X g   n   g1 f g2  (j)b   = λk k + ∆hj . hj t . k=1   X g    n    g1 f g2  (j)b    = λk k + ∆hj . + xhj t τhj t, . k=1    X g     n      which is in the standard form. Thus for w(j) ∈ W πj , we have w(j) ∈ (H ∩ n T Z )hj g for some g = (g1,...,gn) t if and only if (j) Z (1) there exists λk ∈ for each 1 ≤ k ≤ f such that

π −−→ A j · w(j) πj 1 B1 g1 −−→ π πj (j) j f A · w  B2 g2 2   (j)b   + . = λ k + ∆hj . + xhj t, . . k .  .  . k=1 .  −−→  πj  X    πj  Bn  g  A · w(j)    n  n        and

(2) tπj = τhj t, πj πj T (where πj = (B1 ,...,Bn ) tπj as before). The first condition can be restated (j) Z as follows. There exists λk ∈ for each 1 ≤ k ≤ f such that

g1 f −−→ g2 πj (j) πj  (j)b    Ai · w + Bi = ei · λk k + ∆hj . + xhj t . k=1    X g     n      for each 1 ≤ i ≤ n and 1 ≤ j ≤ c. This can be re-written as

g1 f −−→ g2 πj (j) b   πj (18) Ai · w + λk(−ei · k) − ei · ∆hj . = −Bi + ei · xhj t. . k=1   X g   n   28 ALEX EVETTS

Then we may rewrite V (π,t) as follows.

π (1) (c) π1 πc T Zn V ( ,t)= w ,...,w ∈ UH∩Zn ×···× UH∩Zn ∃(g1,...,gn) ∈ , nj  λ( ) ∈ Zn, s.t. each w(j) satisfies (18) for each 1 ≤ i ≤ n k o π j Zm(π)+fc+n For a fixed and t ∈ T , we define vectors Li ∈ as follows:

0 . j−1 m(π ) zeroes  .   k=1 j  0  P  πj   A  m(πj ) rows  i    0      .  c  .   k=j+1 m(πj ) zeroes    0     P  0     j  .   Li :=  .   f(j − 1) zeroes    0     −e · b   i 1    .    .   f rows   −e · b    i f   0      .    .  f(c − j) zeroes     0      −1 ∆  n rows  i hj      where 1i∆hj is the matrix product of the row vector with 1 at the ith position and zeroes elsewhere, and ∆hj . Now, noting that

g1 g1 g2 g2      1i∆hj · . = ei · ∆hj . , . .      gn  gn           we see that a c-tuple of words w(1),...,w(c) satisfies (18) for some i precisely when there exists v ∈ Zm(π)+fc+n such that 

j πj Li · v = −Bi + ei · xhj t RATIONAL GROWTH IN VIRTUALLY ABELIAN GROUPS 29

−−→ −−→ (1) (c) m(π) and pm(π)(v)= w ,..., w ∈ Z . Therefore we have   c n −−−−→ m(π)+fc+n j πj V (π,t)= p π v ∈ Z L · v = −B + ei · xh t , m( )  i i j  j=1 i=1 n o \ \  π  which is a positive polyhedral subset of Zm( ). Now since finite unions of −−−→ −−−−→ π π polyhedral sets are polyhedral, V ( ) = t∈T V ( ,t) is polyhedral, which proves the Theorem.  S

5. Conjugacy Growth Series In this section we will prove the following. Theorem 5.1. Let G be a virtually abelian group, with finite generating set S, and weight function ω : S → N+. Then the weighted conjugacy growth series of G with respect to S is rational. In order to prove this Theorem, we show that the set of conjugacy classes of a virtually abelian group can be split into an infinite collection of finite classes, and an infinite collection of infinite classes. For the finite case, [3] gives us a way to find a minimal representative for each element of the conjugacy class, and express a full set of representatives using polyhedral sets. In the infinite case, we express each conjugacy class as a finite union of cosets of certain subgroups. Section 4 gives us a way to find a minimal representative word for a given coset, and to express a full set of such representatives using polyhedral sets. We thus find a finite set of candidates for a unique minimal representative for every conjugacy class (finite or infinite). This allows us to use Lemma 2.21 to extract a single such representative for each class, so that the polyhedral set description, and thus rational growth, is preserved. As above, we assume that G contains Zn as a normal subgroup, with [G: Zn] = d < ∞, and we let T be a choice of transversal for G/Zn such that 1G ∈ T . Furthermore, we fix an order on T :

T = {1,t2,t3,...,td}. First, we must understand the structure of conjugacy classes in virtu- ally abelian groups. Conjugacy classes have different structure depending n n on whether they are inside or outside the centralizer of Z , CG(Z ). Thus we consider these cases separately. Note that if one element of a coset Znt cen- n n n tralizes Z then t must centralize Z and hence the whole coset is in CG(Z ). n n n So both CG(Z ) and G \ CG(Z ) are unions of Z -cosets.

5.1. Conjugacy classes of elements inside the centralizer of Zn. 30 ALEX EVETTS

n Lemma 5.2. Let g ∈ CG(Z ). Then the conjugacy class of g has size at most d, and is given by −1 −1 −1 [g]= {tgt | t ∈ T } = {g,t2gt2 ...,tdgtd }. Proof. Let h ∈ G. Then hgh−1 = xtgt−1x−1 for some x ∈ Zn and t ∈ T . Since the centralizer of a normal subgroup is itself a normal subgroup, tgt−1 centralizes Zn, and so hgh−1 = tgt−1.  Recall the sets U π of minimal-length representatives for π-patterned words, introduced in Theorem 2.18. Each element of a conjugacy class has a unique minimal-weight representative, and this is contained in U π for some π ∈ P n (recall Definition 2.13). So by Lemma 5.2, each conjugacy class in CG(Z ) has at most d candidate words for a weight minimal representative. A d-tuple of candidates has a d-fold pattern, the d-dimensional vector where the entries are the patterns of the component words of the d-tuple. We will show that for each d-fold pattern in P d, the corresponding set of d-tuples of candidate representatives forms a polyhedral set. d Definition 5.3. Fix a d-fold pattern π = (π1, π2,...,πd) ∈ P where πj ∈ Zn −1 Zn CG( ) for each j, and tj π1tj ∈ πj for each 2 ≤ j ≤ d. Let

C(π) := w(1),...,w(d) ∈ U π1 ×···× U πd |

n  (j) (1) −1 w = tj w tj , for each 2 ≤ j ≤ d . Remark 5.4. Note that each tuple in C(π) corresponds to a conjugacyo class, and (by definition of the sets U πj ) the weight of a conjugacy class is realised by at least one of the words in the corresponding tuple. −−−→ Proposition 5.5. For each C(π), the set C(π) ⊂ Nm(π) is polyhedral.

(1) π1 (j) πj −1 Zn Proof. Consider w ∈ U and w ∈ U where tj π1tj ∈ πj . By (j) (1) −1 Lemma 2.11, w = tjw tj if and only if −−→ −−→ π1 (1) πj (j) πj π1 (19) Ai,tj · w − Ai · w = Bi − Bi,tj for each 1 ≤ i ≤ n. We express this using linear algebra. Define

π1 Ai,tj m(π1) rows 0   − . j 1 m π .  Pk=2 ( k) zeroes    j π  0   Fi ( )=  πj  −A  m(πj ) rows  i    0      .  d m π  .   Pk=j+1 ( k) zeroes    0        RATIONAL GROWTH IN VIRTUALLY ABELIAN GROUPS 31

−−→ −−→ T for each 1 ≤ i ≤ n, 2 ≤ j ≤ d. Then the vector w(1),..., w(d) satisfies (19) for some j if and only if   −−→ w(1) F j (π) · . = Bπj − Bπ1 i  .  i i,tj −−→ w(d)   for each 1 ≤ i ≤ n. Thus   d n −−−→ −−→ −−→ π π π π1 πd Zm( ) j π j π1 C( )= U ×···× U ∩ ~z ∈ Fi ( ) · ~z = Bi − Bi,tj .   j=2 i=1 n o \ \ This is therefore a polyhedral set.  5.2. Conjugacy classes of elements outside the centralizer of Zn. We express the conjugacy classes in terms of certain cosets, and use the sets π UF introduced in Definition 4.9 to find polyhedral sets of conjugacy class representatives. Definition 5.6. For any γ ∈ G, define the subgroup F (γ)= {[x, γ] | x ∈ Zn}. Note that this is indeed a subgroup of G, since if x, y ∈ Zn, we have [x, γ][y,γ]= xγx−1γ−1yγy−1γ−1 = xyγx−1γ−1γy−1γ−1 = [xy,γ]. Furthermore, since Zn is normal, [x, γ] ∈ Zn, and so F (γ) is a subgroup of Zn, and hence is free abelian. Remark 5.7. Let a ∈ Zn and t ∈ T . Then since Zn is normal, [x,at] = xatx−1t−1a−1 = xtx−1t−1aa−1 = [x, t]. So F (γ) depends only on the Zn- coset that γ is contained in. Thus if w ∈ W π then F (w)= F (π). If A and B are subsets of some group G, write BA for the conjugate of A by B, that is BA = {bab−1 | a ∈ A, b ∈ B}. n Lemma 5.8. If g ∈ G \ CG(Z ) then its conjugacy class is given by a union of finitely many cosets as follows [g]= [x,tgt−1] | x ∈ Zn tgt−1 = F (tgt−1)tgt−1. t∈T t∈T [  [ n n −1 Proof. Let g ∈ G \ CG(Z ), and suppose x ∈ Z . We have xgx = xgx−1g−1g = [x, g]g. Now the conjugacy class is given by Zn [g]= G{g} = {xtg(xt)−1 | x ∈ Zn,t ∈ T } = {tgt−1 | t ∈ T } Zn = {tgt−1} = x,tgt−1 | x ∈ Zn tgt−1. t∈T t∈T [ [   32 ALEX EVETTS



n Definition 5.9. Fix a d-fold pattern π = (π1,...,πd) where πj ∈ G\CG(Z ) −1 Zn for each j, and tj π1tj ∈ πj for 2 ≤ j ≤ d. Define

C′(π)= w(1),...,w(d) ∈ U π1 ×···× U πd F (π1) F (πd) n n  Z tj (j) (1) w ∈ w , 2 ≤ j ≤ d , n o  where U πi is as in Definition 4.9. That is, C′(π) is the set of d-tuples of F (πi) (j) words where each w is the unique minimal representative for its F (πj )-coset, (j) (1) n and w is conjugate to w via an element of Z tj . Proposition 5.10. Each element of C′(π) consists of a d-tuple of words representing elements of the same conjugacy class. Furthermore, the weight of the conjugacy class is realised by the word(s) of smallest weight in the d- tuple. Proof. Let w(1), w(2),...,w(d) ∈ C′(π). From the definition of C′(π), each (j) (1) w is conjugate to w , so each component word represents an element of the same conjugacy class. Now from Lemma 5.8, we see that each word rep- resents one of the finite number of cosets that make up the corresponding (j) conjugacy class. In fact, since each w is contained in UF (πj ), each com- ponent word is a minimal-weight representative for the coset. Therefore the minimal-weight representative(s) for the conjugacy class must be contained in {w(1), w(2),...,w(d)}.  −−−→ Proposition 5.11. The set C′(π) ∈ Nm(π) is a polyhedral set. Proof. We have n Z tj (1) (1) −1 (1) −1 (1) −1 w = F tj w tj tj w tj = F (πj )tj w tj .

n o  n  Z tj (j) (1) by Remark 5.7. Thus w ∈ w if and only if there exists y ∈ F (πj ) with n o (j) (1) −1 (20) w = ytj w tj . π ) ⊂ Zn. Let {bπj ,..., bπj } Write fj for the rank of the free abelian group F ( j 1 fj be a choice of basis for F (πj ). Then there exists y ∈ F (πj ) satisfying (20) if and only if there exist integers a1,...,afj with

fj (j) bπj (1) −1 w = ak k tj w tj . k X=1 RATIONAL GROWTH IN VIRTUALLY ABELIAN GROUPS 33

Expanding using identities (8) and (11) gives −−→ −−→ πj (j) π π1 (1) π1 A · w j A1,t · w B 1 B1 j 1,tj −−→ πj −−→ π πj (j) fj π1 (1) 1 A · w  B2  A · w  B2,tj  2   bπj 2,tj   + . tπj = ak + . tπj , . .  k  . .  .  .  k=1  .  .  π .  −−→  j  X  −−→  π1   πj (j) Bn   π1 (1) Bn,t  An · w      A · w   j       n,tj          or equivalently,

−−→ −−→ fj π1 (1) πj (j) bπj πj π1 (21) Ai,tj · w − Ai · w + ei · ak k = Bi − Bi,tj k X=1 for each 1 ≤ i ≤ n. We express this using linear algebra. For each 1 ≤ i ≤ n and 2 ≤ j ≤ d, consider the vectors

π1 Ai,tj m(π1) rows 0   − . j 1 m π .  Pk=2 ( k) zeroes    j π  0   Mi ( )=  πj  −A  m(πj ) rows  i    0      .  d m π  .   Pk=j+1 ( k) zeroes    0       and  0 − . Pj 1 f zeroes  .   k=2 k 0   bπj  ei · 1   πj   ei · b  j  2  N (π)=  .   fj rows i  .    .    πj   ei · b   fj   0       .  d f  .  Pk=j+1 k zeroes        0     d  Let f = j=1 fj, i.e. the sum of the ranks of the free abelian subgroups F (πj ), n Z tj j π (j) (1) and henceP the dimension of Ni ( ). Now by equation (21), w ∈ w n o 34 ALEX EVETTS precisely when there exist integers a1,a2,...af such that −−→ w(1) a −−→ 1 w(2) a2 M j (π) · + N j (π) ·   = Bπj − Bπ1 i . i . i i,tj  .  .  .    −−→ a  w(d)  f        for each 1 ≤ i ≤ n and 2 ≤ j ≤ d. We see that each of these identities defines an elementary set if we rewrite it as follows: | M j (π)   i   π | πj π ~z ∈ Nm( )+f · ~z = B − B 1  .   |  i i,tj       N j (π)   i   |            Taking the intersection of each elementary set, discarding the vector (a1,...,ar), and intersecting with the cartesian product of polyhedral sets UF (π1) ×···× π UF (πd), allows us to express the m( )-dimensional vectors corresponding to C′(π) as a polyhedral set: −−−→ ′ π C ( )= UF (π1) ×···× UF (πd) ∩  | M j(π) d n   i   m(π)+r | πj π1 p π ~z ∈ N · ~z = B − B  m( )   |  i i,tj  j=2 i=1     \ \  N j(π)   i   |            where, as before, pk denotes projection onto the first k coordinates.  We are now ready to prove Theorem 5.1.

Proof of Theorem 5.1. Each conjugacy class in G has a d-tuple of candidates for a weight-minimal representative (see Remark 5.4 and Proposition 5.10), given by an element of C(π) or C′(π) for an appropriate d-fold pattern π. The result will follow from the following claim: There exists a finite set of d-fold patterns, R, so that: (1) the candidate representatives for every conjugacy class in G (as d- tuples of words) have a d-fold pattern in R, and (2) no conjugacy class is represented by more than one d-fold pattern in R. RATIONAL GROWTH IN VIRTUALLY ABELIAN GROUPS 35

∗ For each π ∈ R, Lemma 2.21 yields a set Lπ ∈ S of unique, minimal-weight representatives for the tuples of C(π), or C′(π). It follows from the above claim that π∈R Lπ is a finite disjoint union ofe sets, forming a language of unique minimal-weight representatives for the conjugacy classes of G. Since S each Lπ has rational growth series, we conclude that G has rational conjugacy growth series. Now we prove the claim. If P is the finite set of patterns (with respect to S∗) providing minimal weight representatives for each element of G (as per Definition 2.13), consider the set of ordered d-tuples (π1, π2,...,πd) of −1 Zn elementse of P , with the condition that tiπ1ti ∈ πj . For any set of such d-tuples which are permutations of each other, choose only one (arbitrarily), and discard the others. Call the resulting reduced set of d-fold patterns R. This is clearly a finite set, and is sufficient to represent all d-tuples of elements of G. This proves part (1). To see part (2), note that the candidates are uniquely determined (either n the unique weight-minimal representatives for each element, in the CG(Z ) case, or the unique weight-minimal representatives for each coset component, n in the G \ CG(Z ) case). A tuple of candidates uniquely determines a d-fold pattern in R (since we have removed permutations). Thus the claim holds, and the theorem follows. 

Acknowledgments The author would primarily like to thank Laura Ciobanu for an immeasur- able amount of mathematical discussion and writing advice. Thanks are also due to Turbo Ho for a helpful discussion, and Jim Howie for useful comments on a draft of this paper.

References

[1] Y. Antol´ın and L. Ciobanu, Formal conjugacy growth in acylindrically hyperbolic groups, Int. Math. Res. Not. IMRN (2017), no. 1, 121–157. MR 3632100 [2] I. K. Babenko, Closed geodesics, asymptotic volume and the characteristics of growth of groups, Izv. Akad. Nauk SSSR Ser. Mat. 52 (1988), no. 4, 675–711, 895. MR 966980 [3] M. Benson, Growth series of finite extensions of Zn are rational, Invent. Math. 73 (1983), no. 2, 251–269. MR 714092 [4] E. Breuillard and Y. de Cornulier, On conjugacy growth for solvable groups, Illinois J. Math. 54 (2010), no. 1, 389–395. MR 2777001 [5] J. W. Cannon, The growth of the closed surface groups and compact hyperbolic coxeter groups, Circulated typescript, Univ. Wisconsin, 1980. [6] , The combinatorial structure of cocompact discrete hyperbolic groups, Geom. Dedicata 16 (1984), no. 2, 123–148. MR 758901 [7] L Ciobanu, S. Hermiller, D. Holt, and S. Rees, Conjugacy languages in groups, Israel J. Math. 211 (2016), no. 1, 311–347. MR 3474966 [8] L. Ciobanu, S. Hermiller, and V. Mercier, Formal conjugacy growth in graph products, In preparation 2019. 36 ALEX EVETTS

[9] Raf Cluckers, Julia Gordon, and Immanuel Halupczok, Integrability of oscillatory func- tions on local fields: transfer principles, Duke Math. J. 163 (2014), no. 8, 1549–1600. MR 3210968 [10] M. Coornaert and G. Knieper, Growth of conjugacy classes in Gromov hyperbolic groups, Geom. Funct. Anal. 12 (2002), no. 3, 464–478. MR 1924369 [11] T. C. Davis and A. Y. Olshanskii, Relative subgroup growth and subgroup distortion, Groups Geom. Dyn. 9 (2015), no. 1, 237–273. MR 3343353 [12] J. Denef, The rationality of the Poincar´eseries associated to the p-adic points on a variety, Invent. Math. 77 (1984), no. 1, 1–23. MR 751129 [13] L. E. Dickson, Finiteness of the Odd Perfect and Primitive Abundant Numbers with n Distinct Prime Factors, Amer. J. Math. 35 (1913), no. 4, 413–422. MR 1506194 [14] M. Duchin, Counting in groups: Fine asymptotic geometry, Notices of the AMS 63 (2016), no. 8, 871–874. [15] Moon Duchin and Michael Shapiro, The Heisenberg group is pan-rational, Adv. Math. 346 (2019), 219–263. MR 3908251 [16] D. B. A. Epstein, J. W. Cannon, D. F. Holt, S. V. F. Levy, M. S. Paterson, and W. P. Thurston, Word processing in groups, Jones and Bartlett Publishers, Boston, MA, 1992. MR 1161694 [17] I. Gekhtman and W. Yang, Counting conjugacy classes in groups with contracting elements, ArXiv e-prints, arXiv:1810.02969. [18] M. Gromov, Hyperbolic groups, Essays in , Math. Sci. Res. Inst. Publ., vol. 8, Springer, New York, 1987, pp. 75–263. MR 919829 [19] F. J. Grunewald, D. Segal, and G. C. Smith, Subgroups of finite index in nilpotent groups, Invent. Math. 93 (1988), no. 1, 185–223. MR 943928 [20] V. Guba and M. Sapir, On the conjugacy growth functions of groups, Illinois J. Math. 54 (2010), no. 1, 301–313. MR 2776997 [21] D. F. Holt, Automatic groups, subgroups and cosets, The Epstein birthday schrift, Geom. Topol. Monogr., vol. 1, Geom. Topol. Publ., Coventry, 1998, pp. 249–260. MR 1668284 [22] D. A. Klarner, Mathematical crystal growth. I, Discrete Appl. Math. 3 (1981), no. 1, 47–52. MR 604265 [23] F. Liardet, Croissance dans le groupes virtuellement ab´eliens, Ph.D. thesis, Universit´e De Gen`eve, 1997. [24] V. Mercier, Conjugacy growth series of some wreath products, ArXiv e-prints (2016), arXiv:1610.07868. [25] M. Presburger, Uber¨ die vollst¨andigkeit eines gewissen systems der arithmetik ganzer zahlen, in welchem die addition als einzige operation hervortritt, Comptes-rendus du I Congr`es des Math´ematiciens des Pays Slaves (1929), 92–101. [26] I. Rivin, Some properties of the conjugacy class growth function, Group theory, statis- tics, and cryptography, Contemp. Math., vol. 360, Amer. Math. Soc., Providence, RI, 2004, pp. 113–117. MR 2105439 [27] , Growth in free groups (and other stories)—twelve years later, Illinois J. Math. 54 (2010), no. 1, 327–370. MR 2776999 [28] M. Stoll, Rational and transcendental growth series for the higher Heisenberg groups, Invent. Math. 126 (1996), no. 1, 85–109. MR 1408557

Department of Mathematics, Heriot-Watt University, Edinburgh, Scotland, EH14 4AS E-mail address: [email protected]