betaFeedback

WorksMathematics

An example of a non-sofic hyperbolic group

Abstract

We construct a torsion-free, two-dimensional, word-hyperbolic group which is not sofic. Our construction builds on OpenAI's construction of a non-residually finite hyperbolic group, see OpenAI 252.

Introduction

For a positive integer nn, let Sym⁡(n)\operatorname{Sym}(n) denote the symmetric group on {1,…,n}\{1,\ldots,n\}. Its normalized Hamming metric and the fixed-point fraction of a permutation are given by

dH(σ,ρ)=∣{a:σ(a)≠ρ(a)}∣n,fix⁡(σ)=∣{a:σ(a)=a}∣n.d_H(\sigma,\rho)=\frac{\lvert\{a:\sigma(a)\ne\rho(a)\}\rvert}{n},\qquad\operatorname{fix}(\sigma)=\frac{\lvert\{a:\sigma(a)=a\}\rvert}{n}.

We write id for the identity permutation. A group GG is sofic if, for every finite subset F⊂GF\subset G containing 11 and every η>0\eta>0, there are a positive integer nn and a map ϕ:F→Sym⁡(n)\phi:F\to\operatorname{Sym}(n), sending 11 to the identity permutation, such that

dH(ϕ(a)ϕ(b),ϕ(ab))<η(a,b,ab∈F),fix⁡(ϕ(a))<η(a∈F∖{1}).d_H(\phi(a)\phi(b),\phi(ab))<\eta\quad(a,b,ab\in F),\qquad\operatorname{fix}(\phi(a))<\eta\quad(a\in F\setminus\{1\}).

Thus multiplication of elements in FF is preserved on most points of the finite set, while each specified nonidentity element fixes only a small proportion of the points. The notion was introduced by Gromov [8]; see also Weiss [19]. All amenable groups and all residually finite groups are sofic; see [16]. The question whether all groups are sofic was open for a long time.

The first proof of the existence of a non-sofic group was given by OpenAI [15]. Kun and the second author [11] constructed non-sofic generalized wreath products and doubles of residually finite groups. Chapman and Peled [4] gave a conditional construction of a non-sofic hyperbolic group using random simplicial complexes. The aim of this paper is to construct a torsion-free word-hyperbolic group which is not sofic. We build on OpenAI’s construction of a non-residually finite torsion-free hyperbolic group [14]. Recall that a finitely generated group is word-hyperbolic if geodesic triangles in a Cayley graph for a finite generating set are uniformly thin: each side lies within a fixed distance of the other two sides, see [3] for details.

We obtain the group as the fundamental group of a finite complex. By a triangle complex we mean a two-dimensional cell complex obtained from a graph by attaching triangles, each side being mapped homeomorphically onto an edge. Several edges may have the same endpoints. In our construction, each closed face is embedded. We assign a length to every edge and give the faces the metrics of triangles in the hyperbolic plane of curvature −1-1, with the prescribed side lengths. The complex then carries the induced path metric.

A geodesic metric space is CAT⁡(−1)\operatorname{CAT}(-1) if distances between points on the sides of a geodesic triangle are at most the distances between the corresponding points of a triangle in the hyperbolic plane with the same side lengths. Corresponding points have the same distances from the endpoints of the corresponding sides. The metric is locally CAT⁡(−1)\operatorname{CAT}(-1) if each point has a neighbourhood which, with its intrinsic path metric, is CAT⁡(−1)\operatorname{CAT}(-1). We refer to [3] for these notions and their basic properties. Our main result is the following.

Theorem A. There exists a finite connected triangle complex KK with a locally CAT⁡(−1)\operatorname{CAT}(-1) metric such that G=π1(K)G=\pi_1(K) is torsion-free, word-hyperbolic, and non-sofic.

2020 Mathematics Subject Classification. Primary 20F67; Secondary 20E26, 20F65, 05C38, 15A03.

Key words and phrases. Sofic groups, hyperbolic groups, negatively curved complexes, matrix rank, finite-field interpolation, graphs of large girth.

Put q=9⋅1022+1q = 9 \cdot10^{22} + 1. The presentation of GG in § 3 has

1012(q−1)q4+6⋅1017q5−2<4⋅1013210^{12}(q-1)q^{4} + 6 \cdot10^{17}q^{5} - 2 < 4 \cdot10^{132}

generators,

1012(q−1)q5−107(q−1)5<6⋅1014910^{12}(q-1)q^{5} - 10^{7}(q-1)^{5} < 6 \cdot10^{149}

relations.

The group GG in Theorem A admits no proper cocompact action by cubical automorphisms on a CAT(0) cube complex. Indeed, by Agol’s theorem [1 Theorem 1.1] and the work of Haglund and Wise [9 Theorem 4.4], a word-hyperbolic group admitting such an action is residually finite and therefore sofic.

Corollary B. There exists a nontrivial torsion-free word-hyperbolic group which has no nontrivial sofic quotient.

Proof. If every nontrivial torsion-free word-hyperbolic group had a nontrivial sofic quotient, the theorem of Kapovich and Wise [10 Theorem 3.4], applied to the class of sofic groups, would imply that every torsion-free word-hyperbolic group is residually sofic and hence sofic by [7]. This contradicts Theorem A. □

The construction starts with a family of affine lines in Fq5\mathbb{F}_q^{5}, where Fq\mathbb{F}_q is the field with qq elements and qq is prime. Its incidence graph has one vertex for each point and each line, and an edge whenever the point lies on the line. This is the graph D(5,q)D(5,q) of Lazebnik and Ustimenko [12]. Its girth, the length of its shortest simple cycle, is at least ten [13 Proposition 2.1(iii)]. In § 2, we select a family of lines with distinct distinguished points, called marks, and delete the incidence with its own mark for each line in a specified subfamily.

The retained incidences determine a triangle complex with three vertices. Each retained line gives an array of edges between two of these vertices. Choosing two rows and two columns gives a closed four-edge path, traversed with alternating orientations.

In § 3, we combine polynomial interpolation with the rank estimate of Alon–Hall–Knauer–Pinchasi–Yuster [2 Theorem 6]. We prove that every assignment of permutations to the oriented edges has either a triangular boundary word moving a fixed positive fraction of points or one of the four-edge words fixing such a fraction. The bound is independent of the degree of the permutations.

In § 4, we use the girth bound to construct a locally CAT(−1-1) metric and prove that the four-edge paths are non-nullhomotopic. Soficity would allow both fractions in the preceding paragraph to be arbitrarily small. § 5 derives the contradiction.

Acknowledgments. This paper is based on an AI-generated construction obtained by Harjas Sandhu on October 7, 2026, which was in turn based on the OpenAI breakthrough [14]. The second author worked through the proof to make it readable and to reduce unnecessary overhead. OpenAI’s GPT Astra helped draft this document. We simplified various constructions and related them to previous work. The aim of this note is to give a reasonably self-contained presentation of the result. An independent AI-assisted proof was given by Douglas–Mghirbi [5].

From the construction in [14] we use the marked lines, triangular relations associated to Cartesian products, interpolation identity, and comparison of real and finite-field ranks. We prove these ingredients and combine them with the rank theorem and the incidence relations of D(5,q)D(5,q) to obtain the inequality for approximate permutation assignments. We thank Ian Agol for pointing out the relevance of Dvir’s work [6] to the crucial interpolation argument, see Lemma 2.1. We view this document as a commentary on OpenAI 252 [14].

Marked lines from D(5,q)D(5,q)

We fix

r=106,h=3⋅1011,q=h2+1=9⋅1022+1,s=q−110.r = 10^{6}, \qquad h = 3 \cdot10^{11}, \qquad q = h^{2} + 1 = 9 \cdot10^{22} + 1, \qquad s = \frac{q-1}{10}.

The integer qq is prime.1 We consider two copies of Fq5\mathbb{F}_q^{5}, points and labels. We write points as (t,y,z,w,v)(t,y,z,w,v) and labels as [a,b,c,d,e][a,b,c,d,e]. The incidence relation between points and labels is defined by

b−y=at,c−z=bt,d−w=ay,e−v=az.(1)b-y=at,\qquad c-z=bt,\qquad d-w=ay,\qquad e-v=az. \tag*{(1)}

These equations define a bipartite graph D(5,q)D(5,q) as in [13 Section 2]. By [13 Proposition 2.1], this graph is qq-regular, has q5q^{5} vertices in each part, and has girth at least ten. These properties also follow from [12 Proposition 2.1 and Theorem 3.3].

For a fixed label, the incident points form the full affine line

ℓ[a,b,c,d,e]={(t,b−at,c−bt,d−ab+a2t,e−ac+abt):t∈Fq}.(2)\ell_{[a,b,c,d,e]}=\{(t,b-at,c-bt,d-ab+a^{2}t,e-ac+abt):t\in\mathbb{F}_{q}\}. \tag*{(2)}

Different labels determine different lines. Indeed, the normalized direction (1,−a,−b,a2,ab)(1,-a,-b,a^{2},ab) determines a,ba,b, and the point with first coordinate zero then determines c,d,ec,d,e.

Consider the set of incidence edges satisfying a=ta=t. Given a point (t,y,z,w,v)(t,y,z,w,v), the condition a=ta=t and the incidence equations uniquely determine a label by

a=t,b=y+t2,c=z+bt,d=w+ty,e=v+tz.a=t,\qquad b=y+t^{2},\qquad c=z+bt,\qquad d=w+ty,\qquad e=v+tz.

Thus each point is contained in exactly one of these incidence edges. Conversely, for a fixed label [a,b,c,d,e][a,b,c,d,e], the parametrization (2) gives exactly one incident point with first coordinate t=at=a. Thus each label is also contained in exactly one of these incidence edges. These incidence edges therefore form a perfect matching. Since each matched pair satisfies (1), the point matched to a label lies on the corresponding affine line. We henceforth index each line by its matched point, writing ℓi\ell_i for the line marked at i∈Fq5i\in\mathbb{F}_{q}^{5}.

Put S={0,…,s−1}⊂FqS=\{0,\ldots,s-1\}\subset\mathbb{F}_{q}, using the standard integer representatives, and set

I=S5,J=(Fq∖{−1})×Fq4.I=S^{5},\qquad J=(\mathbb{F}_{q}\setminus\{-1\})\times\mathbb{F}_{q}^{4}.

We retain all lines ℓi\ell_i with i∈Ji\in J and distinguish those with i∈Ii\in I. For each distinguished line, we delete only its incidence with its own mark. All incidences of the other retained lines remain. Thus, for x∈Fq5x\in\mathbb{F}_{q}^{5}, the remaining incidences at xx are indexed by

J(x)={i∈I:x∈ℓi∖{i}}∪{i∈J∖I:x∈ℓi}.(3)J(x)=\{i\in I:x\in\ell_i\setminus\{i\}\}\cup\{i\in J\setminus I:x\in\ell_i\}. \tag*{(3)}

The bipartite graph Γ\Gamma on Fq5⊔J\mathbb{F}_{q}^{5}\sqcup J, whose two parts represent points and retained lines, has an edge from xx to ii precisely when i∈J(x)i\in J(x). It is a subgraph of D(5,q)D(5,q) and therefore has girth at least ten.

For a fixed point and a prescribed first label coordinate aa, (1) uniquely determines b,c,d,eb,c,d,e. The first coordinate of the matched point is aa, so the retained lines are precisely those with a≠−1a\ne-1. Thus q−1q-1 retained lines pass through each point before deletion. A point loses one incidence if it belongs to II, and none otherwise. Consequently,

∣J(x)∣={q−2=(h−1)(h+1),x∈I,q−1=h⋅h,x∉I.|J(x)|= \begin{cases} q-2=(h-1)(h+1), & x\in I,\\ q-1=h\cdot h, & x\notin I. \end{cases}

For each point xx, put (sx,tx)=(h−1,h+1)(s_x,t_x)=(h-1,h+1) if x∈Ix\in I and (sx,tx)=(h,h)(s_x,t_x)=(h,h) otherwise. All these factors are at least rr. Choose a bijection

J(x)⟶(Z/sxZ)×(Z/txZ),i⟼(αx(i),βx(i)).(4)J(x)\longrightarrow(\mathbb{Z}/s_x\mathbb{Z})\times(\mathbb{Z}/t_x\mathbb{Z}),\qquad i\longmapsto(\alpha_x(i),\beta_x(i)). \tag*{(4)}

The sets J(x)J(x) may overlap, but their bijections are chosen separately for each point.

The following interpolation argument is adapted from [14 Section 2.5, equation (2.13)] to the grid S5S^{5}. It uses low-degree polynomials on affine lines, as in Dvir’s proof of the finite-field Kakeya conjecture [6], applied here to the entries of rank-one matrices.

Lemma 2.1. There are matrices Hx∈MI(Fq)H_x\in M_I(\mathbb{F}_q), indexed by x∈Fq5x\in\mathbb{F}_{q}^{5}, such that

rank⁡Hx≤1,∑x:i∈J(x)Hx={−Eii,i∈I,0,i∈J∖I.(5)\operatorname{rank}H_x\le1,\qquad \sum_{x:i\in J(x)}H_x= \begin{cases} -E_{ii}, & i\in I,\\ 0, & i\in J\setminus I. \end{cases} \tag*{(5)}

Proof. For c∈Sc\in S, let pcp_c be the Lagrange basis polynomial associated with the node cc in SS, given by

pc(T)=∏d∈S∖{c}T−dc−d.p_c(T)=\prod_{d\in S\setminus\{c\}}\frac{T-d}{c-d}.

It is the unique polynomial of degree at most s−1s-1 satisfying

pc(d)={1,d=c,0,d∈S∖{c}.p_c(d)= \begin{cases} 1, & d=c,\\ 0, & d\in S\setminus\{c\}. \end{cases}

For i=(ci1,…,ci5)∈Ii=(c_{i1},\ldots,c_{i5})\in I and X=(X1,…,X5)X=(X_1,\ldots,X_5), set

fi(X)=∏j=15pcij(Xj),f=(fi)i∈I,Hx=f(x)f(x)T.f_i(X)=\prod_{j=1}^{5}p_{c_{ij}}(X_j),\qquad f=(f_i)_{i\in I},\qquad H_x=f(x)f(x)^{\mathsf T}.

We have rank⁡Hx≤1\operatorname{rank}H_x\le1. For i,i′∈Ii,i'\in I, fi(i′)f_i(i') equals 1 if i=i′i=i' and 0 otherwise, so Hi=EiiH_i=E_{ii} for i∈Ii\in I. Each fif_i has total degree at most 5(s−1)5(s-1). It follows that every entry of ffTff^{\mathsf T}, restricted to an affine line, is a polynomial of degree at most 10(s−1)=q−11<q−110(s-1)=q-11<q-1.

It is elementary to see that the sum of the values of a polynomial of degree less than q−1q-1 over Fq\mathbb{F}_q is zero. Applying this observation entrywise gives

∑x∈ℓiHx=0(i∈J).\sum_{x\in\ell_i}H_x=0\qquad(i\in J).

For i∈Ii\in I, deleting the term at its mark gives −Hi=−Eii-H_i=-E_{ii}. For i∈J∖Ii\in J\setminus I, the full line is retained, so its sum remains zero. This proves (5). □

A finite permutation obstruction

Our next aim is to define a connected triangle complex whose fundamental group will have all the required properties. We use the construction in § 2 and write [r]={0,…,r−1}[r]=\{0,\ldots,r-1\}. We follow the construction in [14 Section 2.3]. Consider three vertices O,V,WO,V,W and define oriented edges

Ai,u,v:V⟶W,Lx,u,j:O⟶V,Rx,v,k:O⟶W.A_{i,u,v}:V\longrightarrow W,\qquad L_{x,u,j}:O\longrightarrow V,\qquad R_{x,v,k}:O\longrightarrow W.

Here i∈Ji\in J, u,v∈[r]u,v\in[r], x∈Fq5x\in\mathbb{F}_q^5, j∈Z/sxZj\in\mathbb{Z}/s_x\mathbb{Z}, and k∈Z/txZk\in\mathbb{Z}/t_x\mathbb{Z}. For each point xx, i∈J(x)i\in J(x), and u,v∈[r]u,v\in[r], attach a triangle with boundary, read in path order,

Lx,u,αx(i)+vAi,u,vRx,v,βx(i)+u−1.(6)L_{x,u,\alpha_x(i)+v}A_{i,u,v}R_{x,v,\beta_x(i)+u}^{-1}. \tag*{(6)}

Denote the resulting finite triangle cell complex by KK. There are LL- and RR-edges, and hence KK is connected.

Let G=π1(K,O)G=\pi_1(K,O). Choose a point x0∈Fq5x_0\in\mathbb{F}_q^5. The edges cV:=Lx0,0,0c_V:=L_{x_0,0,0} and cW:=Rx0,0,0c_W:=R_{x_0,0,0} form a spanning tree of the graph of vertices and edges of KK. Let cOc_O be the constant path at OO. For an oriented edge e:A→Be:A\to B, define

ge=[cAecB−1]∈G.(7)g_e=[c_Aec_B^{-1}]\in G. \tag*{(7)}

Denote the elements associated with Ai,u,vA_{i,u,v}, Lx,u,jL_{x,u,j}, and Rx,v,kR_{x,v,k} by a(i,u,v)a(i,u,v), l(x,u,j)l(x,u,j), and r(x,v,k)r(x,v,k), respectively. We multiply loops in path order. Collapsing the spanning tree gives a finite presentation of GG with generators a(i,u,v),l(x,u,j),r(x,v,k)a(i,u,v),l(x,u,j),r(x,v,k), with the index ranges specified above, and relations

l(x,u,αx(i)+v)a(i,u,v)r(x,v,βx(i)+u)−1=1(8)l(x,u,\alpha_x(i)+v)a(i,u,v)r(x,v,\beta_x(i)+u)^{-1}=1 \tag*{(8)}

for every point xx, i∈J(x)i\in J(x), and u,v∈[r]u,v\in[r], together with

l(x0,0,0)=1,r(x0,0,0)=1.l(x_0,0,0)=1,\qquad r(x_0,0,0)=1.

Our next aim is to understand how this presentation interacts with the assignment of permutations to the generators. We identify a permutation σ∈Sym⁡(n)\sigma\in\operatorname{Sym}(n) with the matrix PσP_\sigma defined by Pσea=eσ(a)P_\sigma e_a=e_{\sigma(a)}, where eae_a is the aa-th standard coordinate vector. Thus inverse matrices are transposes, and tr⁡(Pσ)=nfix⁡(σ)\operatorname{tr}(P_\sigma)=n\operatorname{fix}(\sigma). Assign arbitrary permutations of degree nn to the oriented edges of KK. Denote the permutations assigned to Ai,u,vA_{i,u,v}, Lx,u,jL_{x,u,j}, and Rx,v,kR_{x,v,k} by A(i,u,v)A(i,u,v), L(x,u,j)L(x,u,j), and R(x,v,k)R(x,v,k), respectively. Reversing an edge replaces its permutation by its inverse. For a triangular face indexed by τ=(x,i,u,v)\tau=(x,i,u,v), define

δx,i,u,v=dH(L(x,u,αx(i)+v)A(i,u,v)R(x,v,βx(i)+u)−1,id⁡).(9)\delta_{x,i,u,v}=d_H\bigl(L(x,u,\alpha_x(i)+v)A(i,u,v)R(x,v,\beta_x(i)+u)^{-1},\operatorname{id}\bigr). \tag*{(9)}

This is exactly the defect of the corresponding relation being satisfied in the permutation group. For i∈Ii\in I, u≠u′u\ne u' and v≠v′v\ne v', define the rectangular product

Qi(u,u′,v,v′)=A(i,u,v)A(i,u′,v)−1A(i,u′,v′)A(i,u,v′)−1.(10)Q_i(u,u',v,v')=A(i,u,v)A(i,u',v)^{-1}A(i,u',v')A(i,u,v')^{-1}. \tag*{(10)}

Let λi\lambda_i be the average of fix⁡(Qi(u,u′,v,v′))\operatorname{fix}(Q_i(u,u',v,v')) over all r2(r−1)2r^2(r-1)^2 ordered tuples (u,u′,v,v′)(u,u',v,v') with u≠u′u\ne u' and v≠v′v\ne v'.

The key technical result is the following theorem. It gives a lower bound on the sum of defects of the defining relations together with a measure of non-triviality of the rectangular words wi,u,u′,v,v′:=w_{i,u,u',v,v'}:= a(i,u,v)a(i,u′,v)−1a(i,u′,v′)a(i,u,v′)−1a(i,u,v)a(i,u',v)^{-1}a(i,u',v')a(i,u,v')^{-1}. We will later show that all these rectangular words are non-trivial in GG.

Theorem 3.1. Every such assignment, in every degree n≥1n \ge1, satisfies

q55≤∑τδτ+r28∑i∈Iλi.(11)\frac{q^5}{5} \le\sum_{\tau} \delta_{\tau} + \frac{r^2}{8}\sum_{i\in I}\lambda_i. \tag*{(11)}

The proof of this theorem occupies the remainder of this section and needs various lemmas. The following result appears in [14 Lemma 2.4]. It is also a consequence of Gabber’s bound on torsion in cokernels; see [18 Lemma 1] and [17 Lemma 3.1]. We include a direct proof.

Lemma 3.2. If an integer matrix MM has row norms at most r\sqrt{r} and q>rq > r is prime, then

rank⁡FqM≥12rank⁡RM.\operatorname{rank}_{\mathbb{F}_q} M \ge\frac{1}{2}\operatorname{rank}_{\mathbb{R}} M.

Proof. Set d=rank⁡RMd = \operatorname{rank}_{\mathbb{R}} M and choose a nonsingular d×dd \times d minor BB if d>0d > 0. By Hadamard’s inequality, we have 1≤∣det⁡B∣≤rd/21 \le\lvert\det B\rvert\le r^{d/2}. Let d′=rank⁡FqBd' = \operatorname{rank}_{\mathbb{F}_q} B and choose d′d' rows forming a basis modulo qq. Subtract integer lifts of their linear combinations from the remaining rows. These row operations preserve the determinant and make d−d′d-d' rows divisible by qq. Hence qd−d′q^{d-d'} divides det⁡B\det B, so

qd−d′≤rd/2.q^{d-d'} \le r^{d/2}.

Since q>rq > r, this implies d′≥d/2d' \ge d/2. If d=0d = 0, there is nothing to prove. □

Lemma 3.3. For i∈Ji \in J, let Yi=(A(i,u,v))u,v∈[r]Y_i = (A(i,u,v))_{u,v\in[r]}, an rn×rnrn \times rn integer block matrix. For every i∈Ii \in I,

rank⁡FqYi≥r2n2(2r−1+(r−1)2λi)≥n(r4−r2λi8).(12)\operatorname{rank}_{\mathbb{F}_q} Y_i \ge\frac{r^2n}{2(2r-1+(r-1)^2\lambda_i)} \ge n\left(\frac{r}{4}-\frac{r^2\lambda_i}{8}\right). \tag*{(12)}

Proof. We apply [2 Theorem 6]. For a nonzero 0-1 matrix AA, this theorem gives the following estimate in terms of the bipartite graph whose vertices are the rows and columns and whose edges correspond to entries equal to 1:

rank⁡RA≥e2∑xdx2−e+4c4.(13)\operatorname{rank}_{\mathbb{R}} A \ge\frac{e^2}{\sum_x d_x^2-e+4c_4}. \tag*{(13)}

Here ee is the number of edges, dxd_x the vertex degrees, and c4c_4 the number of unoriented four-cycles. For A=YiA=Y_i, each of the 2rn2rn vertices has degree rr and e=r2ne=r^2n. Every four-cycle uses two different row blocks and two different column blocks: two rows of one permutation block cannot share a column, and the analogous assertion holds for columns. For each ordered choice of these four blocks, the cycles are counted by the fixed points of (10). Each unoriented cycle is counted four times, giving 4c4=nr2(r−1)2λi4c_4=nr^2(r-1)^2\lambda_i. By (13), we get

rank⁡RYi≥r2n2r−1+(r−1)2λi.\operatorname{rank}_{\mathbb{R}}Y_i \ge\frac{r^2n}{2r-1+(r-1)^2\lambda_i}.

Each row of YiY_i has exactly rr entries equal to 1, so Lemma 3.2 proves the first bound in (12). For the second inequality, use (1+t)−1≥1−t(1+t)^{-1}\ge1-t for t≥0t\ge0 to obtain

r22(2r−1+(r−1)2λ)≥r4(1+rλ/2)≥r4−r2λ8(λ≥0).\frac{r^2}{2(2r-1+(r-1)^2\lambda)} \ge\frac{r}{4(1+r\lambda/2)} \ge\frac{r}{4}-\frac{r^2\lambda}{8}\qquad(\lambda\ge0).

The following lemma extends the exact factorization and rank bounds in [14 equations (2.11)–(2.12)] to permutation assignments with triangular defects.

Lemma 3.4. Over Fq\mathbb{F}_q, let Zx=∑i∈J(x)YiZ_x=\sum_{i\in J(x)}Y_i, an rn×rnrn\times rn matrix. For every x∈Fq5x\in\mathbb{F}_q^5,

rank⁡Zx≤n+n∑i∈J(x)u,v∈[r]δx,i,u,v.(14)\operatorname{rank}Z_x\le n+n\sum_{\substack{i\in J(x)\\u,v\in[r]}}\delta_{x,i,u,v}. \tag*{(14)}

Proof. Permutation matrices P,QP,Q differing on kk points have at most kk nonzero columns in P−QP-Q. Thus rank⁡(P−Q)≤ndH(P,Q)\operatorname{rank}(P-Q)\le n d_H(P,Q) over every field. Bi-invariance gives

rank⁡(A(i,u,v)−L(x,u,αx(i)+v)−1R(x,v,βx(i)+u))≤nδx,i,u,v.\operatorname{rank}\left(A(i,u,v)-L(x,u,\alpha_x(i)+v)^{-1}R(x,v,\beta_x(i)+u)\right)\le n\delta_{x,i,u,v}.

Fix a point xx. In ZxZ_x, replace each permutation matrix by the corresponding product in the preceding display. For fixed u,vu,v, the pairs (αx(i)+v,βx(i)+u)(\alpha_x(i)+v,\beta_x(i)+u), with i∈J(x)i\in J(x), traverse the whole Cartesian product once. The resulting matrix is

\begin{pmatrix} \sum_j L(x,0,j)^{-1}\\ \vdots\\ \sum_j L(x,r-1,j)^{-1} \end{pmatrix} \begin{pmatrix} \sum_k R(x,0,k)&\cdots&\sum_k R(x,r-1,k) \end{pmatrix}. \tag*{(15)}\end{pmatrix}

Both factors have rank at most nn, so their product has rank at most nn. Adding the difference in one matrix block increases the rank by at most the rank of that difference. Subadditivity of rank now gives (14).

Proof of Theorem 3.1. Recall that Hx∈MI(Fq)H_x\in M_I(\mathbb{F}_q) and Yi,Zx∈Mrn(Fq)Y_i,Z_x\in M_{rn}(\mathbb{F}_q). We use the identifications

MI(Fq)⊗FqMrn(Fq)≅MI(Mrn(Fq))≅Ms5rn(Fq).M_I(\mathbb{F}_q)\otimes_{\mathbb{F}_q}M_{rn}(\mathbb{F}_q)\cong M_I\left(M_{rn}(\mathbb{F}_q)\right)\cong M_{s^5rn}(\mathbb{F}_q).

Thus the matrices have s5×s5s^5\times s^5 blocks, each of size rn×rnrn\times rn. The symbol ⊗\otimes denotes the Kronecker product: the (j,k)(j,k) block of Hx⊗ZxH_x\otimes Z_x is (Hx)jkZx(H_x)_{jk}Z_x. We write ⨁i∈IYi\bigoplus_{i\in I}Y_i for the block diagonal matrix with diagonal blocks YiY_i and zero off-diagonal blocks. From (5), we obtain the tensor identity of [14 equation (2.14)]:

−⨁i∈IYi=∑x∈Fq5Hx⊗Zx.(16)-\bigoplus_{i\in I}Y_i=\sum_{x\in\mathbb{F}_q^5}H_x\otimes Z_x. \tag*{(16)}

Indeed, expanding ZxZ_x and changing the order of summation makes the coefficient of YiY_i equal to −Eii-E_{ii} for i∈Ii\in I, and zero for i∈J∖Ii\in J\setminus I. The rank on the left of (16) is the sum of the individual ranks. Since rank⁡Hx≤1\operatorname{rank}H_x\le1 by Lemma 2.1, each summand on the right satisfies rank⁡(Hx⊗Zx)=rank⁡Hxrank⁡Zx≤rank⁡Zx\operatorname{rank}(H_x\otimes Z_x)=\operatorname{rank}H_x\operatorname{rank}Z_x\le\operatorname{rank}Z_x. We apply Lemma 3.4 at every point xx. Each triangular face τ=(x,i,u,v)\tau=(x,i,u,v) has a unique associated point, so its defect is counted exactly once. Rank subadditivity therefore gives

∑i∈Irank⁡FqYi≤∑x∈Fq5rank⁡FqZx≤∑x∈Fq5(n+n∑i∈J(x)u,v∈[r]δx,i,u,v)=n(q5+∑τδτ).\sum_{i\in I}\operatorname{rank}_{\mathbb{F}_q}Y_i\le\sum_{x\in\mathbb{F}_q^5}\operatorname{rank}_{\mathbb{F}_q}Z_x\le\sum_{x\in\mathbb{F}_q^5}\left(n+n\sum_{\substack{i\in J(x)\\u,v\in[r]}}\delta_{x,i,u,v}\right)=n\left(q^5+\sum_\tau\delta_\tau\right).

Using the second bound of Lemma 3.3 and cancelling the positive real number nn, we obtain

rs54−q5≤∑τδτ+r28∑i∈Iλi.\frac{rs^5}{4}-q^5\le\sum_\tau\delta_\tau+\frac{r^2}{8}\sum_{i\in I}\lambda_i.

Since q<1023q<10^{23} and rs5=59049⋅10111>5⋅10115rs^5=59049\cdot10^{111}>5\cdot10^{115}, we have rs5/4−q5>q5/4>q5/5rs^5/4-q^5>q^5/4>q^5/5. This proves (11).

The negatively curved complex

We equip every face of the complex KK from § 3 with the hyperbolic triangle of curvature −1-1 with angles

θO=π/2,θV=θW=π/5.(17)\theta_O=\pi/2,\qquad\theta_V=\theta_W=\pi/5. \tag*{(17)}

The angle sum is 9π/10<π9\pi/10<\pi, so this triangle exists. Assign to all LL- and RR-edges the equal side lengths opposite θW\theta_W and θV\theta_V, respectively, and to all AA-edges the side length opposite θO\theta_O. This includes edges contained in no face. All gluings are therefore isometric.

An edge germ at a vertex is an incident edge together with that endpoint. The angular link has one vertex for each edge germ and one edge for each face corner. The length of a link edge is the angle at the corresponding corner. A path in a graph is reduced if it has no immediate edge reversal. A reduced closed path is cyclically reduced if the last edge is not the inverse of the first edge. We set the distance between directions in different link components equal to infinity.

We adapt the link argument of [14 Lemma 2.3] to the present girth bound and angles.

Lemma 4.1. Every nonempty cyclically reduced closed path in any vertex link has length at least 2π2\pi. At either VV or WW, two different AA-directions with the same index ii have distance at least 2π2\pi, or belong to different components.

Proof. At OO, the link is simple and bipartite: a corner between Lx,u,jL_{x,u,j} and Rx,v,kR_{x,v,k} determines i∈J(x)i \in J(x) uniquely by αx(i)=j−v\alpha_x(i)=j-v and βx(i)=k−u\beta_x(i)=k-u. Every nonempty cyclically reduced closed path therefore has at least four edges and length at least 4θO=2π4\theta_O=2\pi.

We map the link at VV to Γ\Gamma by sending the direction Ai,u,vA_{i,u,v} to the line vertex ii and the direction Lx,u,jL_{x,u,j} to the point xx. Each corner maps to the corresponding incidence edge. We claim that this is an immersion, that is, the map is injective on incident edges at every vertex. Indeed, at Ai,u,vA_{i,u,v}, a fixed point zz determines at most one corner, namely that of the face indexed by (z,i,u,v)(z,i,u,v) if i∈J(z)i \in J(z). At Lx,u,jL_{x,u,j}, a fixed index ii determines v=j−αx(i)v=j-\alpha_x(i) modulo sxs_x. There is at most one such v∈[r]v \in[r], since sx≥rs_x \ge r. At WW, the same argument uses u=k−βx(i)u=k-\beta_x(i) modulo txt_x and tx≥rt_x \ge r.

Under these immersions, both a nonempty cyclically reduced closed link path and a reduced path joining distinct AA-directions with the same index ii map to nonempty closed walks in Γ\Gamma with no immediate reversal at an internal vertex. Every such walk contains a simple cycle and hence has at least ten edges. Since link edges at VV and WW have length π/5\pi/5, both assertions follow. □

Proposition 4.2. The universal cover K~\widetilde{K} is a proper CAT(−1-1) space. The group G=π1(K,O)G=\pi_1(K,O) is torsion-free and word-hyperbolic, and KK is a finite two-dimensional classifying space for GG.

Proof. We first subdivide the complex so that cells meet in single faces. Subdivide each edge at its midpoint and each face by joining an interior point to its vertices and edge midpoints by geodesics. Since the closed cells embed and each edge and face has a distinct new vertex, the subdivision is a finite simplicial complex with the same intrinsic metric. The angular link at a new face-interior vertex is a circle of length 2π2\pi. At an edge-interior vertex it consists of length-π\pi paths between the two edge directions.

The original angular links have the same metric after subdivision. By Lemma 4.1, their injective circuits have length at least 2π2\pi, and all circuits in the new links have length at least 2π2\pi. It follows from the metric graph criterion [3 Lemma II.5.6] and the link condition [3 Theorem II.5.2] that KK is locally CAT(−1-1).

The lifted complex is locally finite and has finitely many shapes. Its length metric is complete by [3 Theorem I.7.13]. By local compactness and [3 Proposition I.3.7], it is proper and geodesic. The Cartan–Hadamard theorem [3 Theorem II.4.1] now implies that the universal cover K~\widetilde{K} is globally CAT(−1-1). Deck transformations act freely, properly and cocompactly by isometries. A finite group of isometries of a complete CAT(0) space fixes a point [3 Corollary II.2.8]. Since the deck action is free, no nonidentity deck transformation has finite order. The space is hyperbolic [3 Chapter III.H], so the Švarc–Milnor lemma [3 Proposition I.8.19] implies that GG is word-hyperbolic. Since K~\widetilde{K} is contractible, KK is a classifying space. □

It remains to show that the rectangular words studied before are all non-trivial in GG. This is the content of the following lemma. We adapt the argument following [14 equation (2.6)] to the CAT(−1-1) metric constructed above.

Lemma 4.3. For i∈Ii \in I, u≠u′u \ne u' and v≠v′v \ne v', the element wi;u,u′,v,v′w_{i;u,u',v,v'} is nonidentity in GG.

Proof. The edge path p=Ai,u,vAi,u′,v′−1Ai,u′,v′Ai,u,v−1p=A_{i,u,v}A^{-1}_{i,u',v'}A_{i,u',v'}A^{-1}_{i,u,v} is a loop based at VV, and wi;u,u′,v,v′w_{i;u,u',v,v'} is the class of the loop cVpcV−1c_Vpc_V^{-1} based at OO. At every turn of pp, including the closing turn, the two AA-directions are distinct and have the same index ii. By Lemma 4.1, their link distance is greater than π\pi. It follows that pp is a positive-length closed local geodesic, by the link description of angles [3 Chapter I.7]. If this loop were nullhomotopic, it would lift to a positive-length closed local geodesic in K~\widetilde{K}. Every local geodesic in a CAT(−1-1) space is a geodesic [3 Proposition II.1.4]. This is impossible for a positive-length closed path. □

Proof of the main result

Let HH be a sofic group and let n>0n>0. Given finitely many words in a finite alphabet of elements of HH and their inverses, choose a finite set F⊂HF \subset H containing the alphabet, its inverses, all word prefixes, and 1. Let ϕ:F→Sym⁡(n)\phi:F\to\operatorname{Sym}(n) satisfy the inequalities in the definition of soficity from the introduction. For a word of length ℓ≥1\ell\ge1 with kk inverse letters, evaluate it using the assigned permutations of the positive letters. Its distance from the image under ϕ\phi of the corresponding group element is at most (ℓ−1+k)η(\ell-1+k)\eta. Indeed, each inverse substitution changes the evaluated permutation by less than η\eta, since ϕ(a)ϕ(a−1)\phi(a)\phi(a^{-1}) is within η\eta of id⁡\operatorname{id}. Each successive prefix multiplication introduces an error of less than η\eta. The asserted bound follows from bi-invariance and the triangle inequality.

Proof of Theorem A. The geometric assertions follow from Proposition 4.2. By Lemma 4.3, all rectangular elements are nonidentity. Assume that GG is sofic. Apply the preceding estimate in GG to all triangular and rectangular words, with the elements geg_e as the alphabet. Assign ϕ(ge)\phi(g_e) to each edge symbol. Each triangular word has length three, one inverse letter, and group value 11. Hence δτ<3η\delta_\tau<3\eta. Each rectangular word has length four and two inverse letters, so its evaluated permutation is within 5η5\eta of ϕ(w)\phi(w). Since w≠1w\ne1, the latter permutation has fixed-point fraction less than η\eta. Thus the evaluated rectangle has fixed-point fraction less than 6η6\eta, and λi<6η\lambda_i<6\eta. These estimates also hold when some alphabet elements equal 11.

Each point xx indexes r2∣J(x)∣r^2|J(x)| triangular faces. By the degree count in § 2, there are r2((q−1)q5−s5)r^2((q-1)q^5-s^5) triangle indices in total. The data are fixed independently of nn and η\eta. Applying (11) from Theorem 3.1 to the permutations just constructed, and using the s5s^5 distinguished lines, gives

q55≤∑τδτ+r28∑i∈Iλi≤3r2((q−1)q5−s5)η+r28⋅6s5η≤3r2q5η.\frac{q^5}{5}\le\sum_\tau\delta_\tau+\frac{r^2}{8}\sum_{i\in I}\lambda_i\le3r^2\bigl((q-1)q^5-s^5\bigr)\eta+\frac{r^2}{8}\cdot6s^5\eta\le3r^2q^5\eta.

Choosing 0<η<5⋅10−370<\eta<5\cdot10^{-37} gives a contradiction to our choices for qq and rr. □\square

References

Harjas Sandhu, The University of Queensland, St Lucia, QLD 4072, Australia

Email address: [email protected]

Andreas Thom, Faculty of Mathematics, TU Dresden, 01062 Dresden, Germany

Email address: [email protected]

References

  1. [1]I. Agol, The virtual Haken conjecture, with an appendix by I. Agol, D. Groves, and J. Manning, Doc. Math. 18 (2013), 1045–1087.arxiv.org/abs/1204.2810
  2. [2]N. Alon, H. T. Hall, C. Knauer, R. Pinchasi, and R. Yuster, On graphs and algebraic graphs that do not contain cycles of length 4, J. Graph Theory 68 (2011), no. 2, 91–102.DOI
  3. [3]M. R. Bridson and A. Haefliger, Metric Spaces of Non-Positive Curvature, Grundlehren der mathematischen Wissenschaften, vol. 319, Springer-Verlag, Berlin, 1999.DOI
  4. [4]M. Chapman and Y. Peled, Cocycle stability in permutations of random simplicial complexes, preprint, 2025, arXiv:2509.21566v2.arxiv.org/abs/2509.21566
  5. [5]S. Douglas and N. Mghirbi, Non-soficity of OpenAI’s torsion-free hyperbolic group, preprint, 2026, https://github.com.com/Asiape/hyperbolic-group-not-sofic/blob/main/paper/paper.pdf.DOI
  6. [6]Z. Dvir, On the size of Kakeya sets in finite fields, J. Amer. Math. Soc. 22 (2009), no. 4, 1093–1097.DOI
  7. [7]G. Elek and E. Szabó, On sofic groups, J. Group Theory 9 (2006), no. 2, 161–171.
  8. [8]M. Gromov, Endomorphisms of symbolic algebraic varieties, J. Eur. Math. Soc. 1 (1999), no. 2, 109–197.DOI
  9. [9]F. Haglund and D. T. Wise, Special cube complexes, Geom. Funct. Anal. 17 (2008), no. 5, 1551–1620.DOI
  10. [10]I. Kapovich and D. T. Wise, The equivalence of some residual properties of word-hyperbolic groups, J. Algebra 223 (2000), no. 2, 562–583.DOI
  11. [11]G. Kun and A. Thom, Nonsofic wreath products of residually finite groups, preprint, 2026, arXiv:2608.06222v3.arxiv.org/abs/2608.06222
  12. [12]F. Lazebnik and V. A. Ustimenko, Explicit construction of graphs with an arbitrary large girth and of large size, Discrete Appl. Math. 60 (1995), nos. 1–3, 275–284.DOI
  13. [13]F. Lazebnik, V. A. Ustimenko, and A. J. Woldar, A new series of dense graphs of high girth, Bull. Amer. Math. Soc. (N.S.) 32 (1995), no. 1, 73–79.arxiv.org/abs/math/9501231
  14. [14]OpenAI, A torsion-free hyperbolic group that is not residually finite, family 252, manuscript dated 23 September 2026, https://github.com/openai/math/blob/main/preprints/a-torsion-free-hyperbolic-group-that-is-not-residually-finite-September-23-2026/paper.pdf.hub.valency.io/works/W_38ukwq7p
  15. [15]OpenAI, Nonsofic groups exist, Chapter 3 of Ten Advances in Mathematics and Theoretical Computer Science, 2026, pp. 78–95, updated 6 August 2026, https://cdn.openai.com/pdf/ten-proofs-oai.pdf.
  16. [16]V. G. Pestov, Hyperlinear and sofic groups: a brief guide, Bull. Symbolic Logic 14 (2008), no. 4, 449–480.arxiv.org/abs/0804.3968
  17. [17]R. Sauer, Volume and homology growth of aspherical manifolds, Geom. Topol. 20 (2016), no. 2, 1035–1059.DOI
  18. [18]C. Soulé, Perfect forms and the Vandiver conjecture, J. Reine Angew. Math. 517 (1999), 209–221.
  19. [19]B. Weiss, Sofic groups and dynamical systems, Sankhyā Ser. A 62 (2000), no. 3, 350–359.

Paper details

vlnc:W_c3cz2cty
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 7

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

AI-generated construction, OpenAI’s GPT Astra, AI-assisted proof

View statement
Looking for related papers…
Contents