{"id":"e18f6a1a-b166-40cf-a3fc-b30ff78507b7","arxiv_id":"1909.00439","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A virtually torsion-free hierarchically hyperbolic group either has uniform exponential growth or its Cayley graph is quasi-isometric to Z times a space.","lead":"This paper proves that virtually torsion-free hierarchically hyperbolic groups either have uniform exponential growth or are quasi-isometric to a product of the integers with another space. The result settles uniform exponential growth for several previously open classes, including certain CAT(0) cubical groups of dimension three or more, and unifies many known proofs.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proposition 4.2's non-G-invariant case applies a pairwise-orthogonality cardinality bound to X^N.B without establishing pairwise orthogonality; the needed case split is missing.","rationale":"The central claim is a dichotomy: either uniform exponential growth or quasi-isometric to Z x E. The proof hangs on Proposition 4.2, which splits all generating sets into a ping-pong case and an invariant-domains case. The reader's weakest_assumption points to the external uniform translation-length bound (Lemma 2.23), and that bound is certainly load-bearing for uniformity. However, Lemma 2.23 is a cited prior result; absent evidence it is false, reliance on it is normal practice. The Proposition 4.2 gap is internal: the argument that a non-G-invariant B forces |X^N.B| >= N+1 only works if the sets in X^N.B are pairwise orthogonal, and the proof does not establish this. The same paragraph later implicitly handles the non-orthogonal subcase, so the intended fix is small, but as written the dichotomy is incomplete. There is also a mismatch between the stated X^N and the proved X^{2N+1}, which propagates into the length bound M in Theorem 4.1: the free-semigroup generators from Case 1 have word length bounded by (2N+1)k1, not k1, so the M chosen in the proof is too small unless enlarged. These are concrete, checkable issues. Because they are repairable and do not appear to require a new idea, I keep the reader's CONDITIONAL verdict rather than moving to REJECT. I partially agree with the reader: the Proposition 4.2 gap was mentioned in the rationale, but the weakest_assumption field highlighted Lemma 2.23 instead.","tokens_in":23660,"tokens_out":24259,"duration_ms":221680,"concrete_test":"Re-derive Proposition 4.2 with an explicit case split: (i) if X^N.B has a non-orthogonal pair, verify the final conjugacy argument yields s,t in X^{2N+1} with Big(s),Big(t) non-orthogonal; (ii) if X^N.B is pairwise orthogonal, verify the counting contradiction applies. Then recompute the word-length bounds in Propositions 4.3 and 4.4 for the free-semigroup generators, e.g., |s^{k1}| <= (2N+1)k1, and check whether the M defined in Theorem 4.1 can be enlarged to cover them; if the case split fails or the length bounds cannot be made independent of X, Theorem 1.1 is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Proposition 4.2's proof of the key dichotomy has a gap in the non-G-invariant branch. It asserts that if B = X^N.B is not G-invariant then |X^N.B| >= N+1, contradicting 'at most N pairwise orthogonal elements.' This only follows if X^k.B (or at least X^N.B) is pairwise orthogonal. But B = the union over s in X of Big(s) is a union over generators; different Big(s) need not be orthogonal, and images under different words need not be orthogonal. The proof never establishes pairwise orthogonality of X^N.B. If X^N.B contains non-orthogonal V1,V2, the contradiction is invalid; however, then the final paragraph of the proof can produce s,t of word length at most 2N+1 with non-orthogonal big-set domains, giving Item 1. The written proof omits this case split, so as written the dichotomy is not proven. The statement/proof mismatch (Item 1 says X^N; proof yields X^{2N+1}) compounds this. Theorem 4.1's M must also absorb the (2N+1) factor for the ping-pong powers; as written M >= k1 is insufficient for elements s^{k1} with |s| <= 2N+1. These are repairable, but the central dichotomy and uniformity constants are not rigorously established as written.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies uniform exponential growth for virtually torsion-free hierarchically hyperbolic groups (HHGs). The main theorem (Theorem 1.1) states that every such group either has uniform exponential growth or its Cayley graph is quasi-isometric to a product of the form Z×E. The proof strategy is a dichotomy: if short words act loxodromically on two non-orthogonal domains, ping-pong arguments produce a uniformly short free subsemigroup; otherwise the paper shows that a certain set of domains is an invariant pairwise-orthogonal collection, and a structural analysis yields either uniform exponential growth, virtual abelianness, or a product decomposition. The paper derives several corollaries, including uniform exponential growth for acylindrically hyperbolic HHGs, for groups with Morse or quasi-convex subgroups, and a quantitative Tits alternative under hierarchical acylindricity.","tokens_in":23863,"tokens_out":11878,"duration_ms":110937,"significance":"If correct, the main theorem is a significant contribution: it provides a new unified proof of uniform exponential growth for several classes of non-positively curved groups, including the first proof for certain CAT(0) cubical groups of dimension at least three, and it gives a quasi-isometric restriction on any virtually torsion-free HHG that fails to have uniform exponential growth. The structural theorem for the non-uniform-growth case (Theorem 1.10) is a strong, falsifiable statement, and the use of consistent tuples to obtain product decompositions is a useful contribution. The paper also gives credit to the relevant background machinery and does not appear to be circular; however, one load-bearing self-citation, Lemma 2.23 from the authors' earlier paper [AB18], supplies the uniform translation-length lower bound that makes the ping-pong constants independent of the generating set.","major_comments":[{"comment":"In the branch where B is not G-invariant, the proof asserts that |X^N.B| >= N+1 contradicts the fact that there are at most N pairwise orthogonal elements. This contradiction is only valid if X^N.B is pairwise orthogonal, but the proof does not establish this: B is a union of Big(s) over generators, and images of different domains under different words need not be orthogonal. The argument can be repaired by splitting into subcases: if X^N.B contains two non-orthogonal domains V1,V2, then conjugating the corresponding generators by words of length at most N gives elements of length at most 2N+1 with non-orthogonal big-set domains, which is exactly the desired conclusion; otherwise X^N.B is pairwise orthogonal and the cardinality contradiction applies. As written, the dichotomy is not proven. In addition, the statement of Item 1 says elements lie in X^N, while the proof actually produces elements in X^{2N+1}; the later text in Section 4.1 uses the latter bound.","section":"§4, Proposition 4.2"},{"comment":"The constant M is chosen as max{k1, 2n0+k2, k3+2, 3(k4+2)(N+1)!}, but this does not control the X-lengths of the elements that actually generate the free semigroup. In Case 1 the generators are powers of s and t with |s|,|t| <= 2N+1, so their X-lengths can be as large as (2N+1)k1, not merely k1. In the nested case, the word t^{n0} s^{k2} t^{-n0} has X-length at most 2N n0 + N k2, not 2n0 + k2. The proof of Proposition 4.4 also says to replace 2N+1 by 2n0+1, which ignores the factor N carried by s. The theorem would still hold after enlarging M by a factor depending only on N and the hierarchy constant, but as written the asserted uniform word-length bound in Theorem 4.1 does not follow from the displayed choice of M.","section":"§4, Proof of Theorem 4.1"}],"minor_comments":[{"comment":"In the ping-pong estimate, the text writes d_V(ρ_U^V, t^{k(2N+1)!} x) >= τ0 |k|, but the exponent is k(2N+1)! and the lower bound should involve τ0 k(2N+1)!; the displayed estimate appears to omit the factor (2N+1)!.","section":"§4, Proposition 4.3"},{"comment":"The phrase 'replacing 2N+1 with 2n0+1' is inaccurate because s has X-length at most N; the correct replacement should involve max(N, 2n0+N).","section":"§4, Proof of Proposition 4.4"},{"comment":"The proof of Corollary 1.4 asserts that if G is quasi-isometric to a product with unbounded factors, then an infinite quasi-convex subgroup is either coarsely dense in G or has bounded diameter. This dichotomy is not justified in the text; please add a proof or a reference.","section":"Corollary 1.4"},{"comment":"There are several minor typographical issues: 'By By Proposition 2.27' in the proof of Theorem 4.1, 'CAT(0) cubical groups' in the abstract where 'CAT(0) cubical spaces' is meant, and in Example 1.7 the statement that a group 'is isometric to the product of two trees' should presumably refer to its Cayley graph.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The gap in Proposition 4.2 is the main substantive obstruction, and it appears repairable by adding the missing case split. The constants issue in Theorem 4.1 is mechanical once the word-length factors are absorbed. I recommend requesting a revision that fixes these points; if the authors do so, the paper should be suitable for publication. The self-citation of [AB18, Lemma 1.8] is load-bearing for uniformity, so the editor may wish to verify independently that the constant τ0 is indeed independent of the generating set."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a genuinely strong paper. The main theorem—virtually torsion-free HHGs either have uniform exponential growth or are quasi-isometric to Z×E—is new and consequential, and the corollaries are real: acylindrically hyperbolic HHGs have uniform exponential growth, certain CAT(0) cubical groups in dimension at least 3 get their first proof, and there is a quantitative Tits alternative. The proof strategy is coherent: use short loxodromic words and ping-pong; if none exist, force an invariant orthogonal collection of quasi-line domains and decompose. The exposition is careful and the external machinery is used appropriately. I think the central theorem is correct.\n\nThe soft spots are real but repairable. Proposition 4.2 has a logical gap in the non-G-invariant case. The inequality |X^N.B| ≥ N+1 only contradicts the bound on pairwise orthogonal sets if X^N.B is pairwise orthogonal; otherwise the same assumption already gives the non-orthogonal pair the proposition needs. The written argument doesn't split those cases, so as written the dichotomy isn't proved. The fix is one paragraph: if X^N.B contains two non-orthogonal elements, we are done; if not, it is pairwise orthogonal and the cardinality contradiction goes through. There is also a statement/proof mismatch: Item 1 says s,t ∈ X^N, but the construction only gives length at most 2N+1, and later sections silently use X^{2N+1}. On top of that, Theorem 4.1's M should absorb the (2N+1) factor when taking powers of those words; defining M ≥ k1 is not enough if |s| can be 2N+1. These are fixable, but together they make the current version conditional rather than final.\n\nOne thing to check in refereeing: the uniform translation-length lower bound τ0 from AB18 is load-bearing. It is not circular, but uniformity across generating sets is what makes the ping-pong exponents independent of the generating set, so the referee should verify that the citation says what the paper needs.\n\nThis paper is for geometric group theorists working on hierarchical hyperbolicity, growth, or CAT(0) cube complexes. It deserves serious refereeing. My vote: accept after minor-to-moderate revision, with Proposition 4.2 checked closely.","headline":"A strong structural theorem for hierarchically hyperbolic groups with a repairable gap in Proposition 4.2 and a constants bookkeeping issue; worth serious refereeing.","tokens_in":24441,"tokens_out":5568,"would_cite":true,"duration_ms":117502,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["20F65","20F67","20F69"],"pacs":[],"model":"deepseek-v4-flash","headline":"Virtually torsion-free hierarchically hyperbolic groups either have uniform exponential growth or are quasi-isometric to $\\mathbb{Z}\\times E$; if the dichotomy holds, acylindrically hyperbolic HHGs grow uniformly.","keywords":["hierarchically hyperbolic groups","uniform exponential growth","acylindrically hyperbolic groups","CAT(0) cubical groups","quantitative Tits alternative","big set","translation length","quasi-isometric product"],"falsifier":"Exhibit a virtually torsion-free HHG and a sequence of infinite-order elements $g_n$ with a domain $U_n\\in\\operatorname{Big}(g_n)$ such that the stable translation length $\\tau_{U_n}(g_n)$ tends to zero; that would directly refute Lemma 2.23. Alternatively, find a sequence of generating sets for one HHG with no two elements of length bounded by any fixed $M$ generating a free semigroup, while the Cayley graph is not quasi-isometric to $\\mathbb{Z}\\times E$.","tokens_in":23402,"feed_emoji":"📈","tokens_out":6665,"duration_ms":64027,"temperature":0.7,"pith_summary":"The paper proves a dichotomy for virtually torsion-free hierarchically hyperbolic groups (HHGs), a broad class that includes hyperbolic groups, mapping class groups, and many CAT(0) cubical groups. Every such group either has uniform exponential growth—word counts grow exponentially at a rate bounded away from zero for every finite generating set—or its Cayley graph is quasi-isometric to $\\mathbb{Z}\\times E$ for some space $E$. If the dichotomy is right, every virtually torsion-free acylindrically hyperbolic HHG has uniform exponential growth, and certain CAT(0) cubical groups of dimension at least three receive their first proof of it. The paper also records a structural description of the exceptional non-uniform case and a quantitative Tits alternative under extra hypotheses.","feed_headline":"HHGs grow uniformly unless they split off a Z-factor","feed_subtitle":"Hierarchically hyperbolic groups with virtual torsion-freeness either grow uniformly or quasi-isometrically factor out the integers.","key_machinery":"The proof runs on the hierarchical structure itself: the group is encoded by an index set of domains, each carrying a hyperbolic coordinate space $C_U$, with projection maps satisfying consistency axioms. The load-bearing objects are the big set $\\operatorname{Big}(g)$, the set of domains on which a group element acts with unbounded orbit, and a uniform lower bound $\\tau_0>0$ on the stable translation length of every infinite-order element on each domain in its big set (Lemma 2.23). With that bound, the ping-pong exponents and word-length bounds become independent of the generating set. The dichotomy is decided by whether two short words have big sets meeting non-orthogonal domains: if so, ping-pong on the projected hyperbolic spaces produces a uniformly short free semigroup; if not, the domains in the generating set's big sets form a $G$-invariant family of pairwise orthogonal domains, forcing a quasi-isometric product decomposition $\\mathbb{Z}^{|B|}\\times E$.","core_discovery":"The central claim is Theorem 1.1: let $(G,S)$ be a virtually torsion-free hierarchically hyperbolic group. Then either $G$ has uniform exponential growth, or there is a space $E$ such that the Cayley graph of $G$ is quasi-isometric to $\\mathbb{Z}\\times E$. The proof in fact establishes a trichotomy: for every generating set either two words of uniformly bounded length generate a free semigroup, or $G$ is virtually abelian, or $G$ is quasi-isometric to $\\mathbb{Z}^{|B|}\\times E$ with $B$ a $G$-invariant collection of pairwise orthogonal domains whose associated hyperbolic spaces are uniformly quasi-lines. From this, Corollary 1.3 follows: virtually torsion-free HHGs that are acylindrically hyperbolic have uniform exponential growth. The authors also show that when the top-level hyperbolic space $C_S$ is non-elementary, the free semigroup can be upgraded to a genuine free subgroup, giving a quantitative Tits alternative.","pith_inferences":["The dichotomy suggests that any counterexample to Gromov's question inside the HHG class would have to be virtually abelian or quasi-isometric to $\\mathbb{Z}\\times E$; ruling out exponential non-uniform growth among such products would close Question 1.9 affirmatively.","Because the constants depend only on the hierarchy constants and $\\tau_0$, the theorem yields explicit, computable growth bounds for any HHG whose hierarchy data are known; one could turn the proof into an algorithm that, given a presentation and hierarchy constants, outputs the uniform word-length bound.","If all CAT(0) cubical groups are HHGs, as conjectured, the result would give uniform exponential growth for every virtually torsion-free cubical group that is not quasi-isometric to $\\mathbb{Z}\\times E$, extending the first-proof status beyond dimension three."],"forward_implications":["Every virtually torsion-free HHG whose Cayley graph is not quasi-isometric to a nontrivial product has uniform exponential growth.","Virtually torsion-free acylindrically hyperbolic HHGs have uniform exponential growth; this covers non-elementary hyperbolic groups, non-exceptional mapping class groups, and many CAT(0) cubical groups.","An HHG with an asymptotic cone containing a cut-point, or with an unbounded Morse quasi-geodesic, has uniform exponential growth.","If the top-level hyperbolic space $C_S$ is non-elementary, and $G$ is not quasi-isometric to $\\mathbb{Z}\\times E$, then every generating set contains two elements of uniformly bounded length generating a free subgroup.","A virtually torsion-free HHG without uniform exponential growth has a very restricted shape: a $G$-invariant set of pairwise orthogonal domains whose hyperbolic spaces are uniformly quasi-lines, with all other unbounded domains orthogonal to them."],"supporting_citations":[{"why":"Supplies the uniform translation-length lower bound $\\tau_0$ used to keep ping-pong exponents independent of the generating set.","marker":"[AB18, Lemma 1.8]"},{"why":"Provides the definition of hierarchically hyperbolic spaces and groups, the distance formula, and the product decomposition used in the quasi-isometric conclusion.","marker":"[BHS19]"},{"why":"Gives the ping-pong criterion that turns independent loxodromic isometries with lower-bounded translation length into a free semigroup.","marker":"[BF18, Proposition 11.1]"},{"why":"Classifies automorphisms of HHGs, including the big set and the dichotomy elliptic versus axial, used to locate short loxodromic elements.","marker":"[DHS17]"},{"why":"Transfers uniform exponential growth between a group and a finite-index subgroup, letting the proof reduce to torsion-free groups.","marker":"[SW92, Lemma 3.4]"},{"why":"Provides the acylindrical-action tool that upgrades free semigroups to genuine free subgroups in the quantitative Tits alternative.","marker":"[Fuj15]"},{"why":"Supplies the acylindricity and endpoint-stabilizer facts used to produce independent loxodromic isometries in the top-level hyperbolic space.","marker":"[DGO16]"}],"fun_headline_variants":["HHG growth: uniform unless Z-factor splits off","Uniform exponential growth for HHGs, barring Z-factor split","HHGs grow uniformly or quasi-isometrically split Z","Virtually torsion-free HHGs: uniform growth or Z-factor","No Z-factor means uniform growth for HHGs"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof assumes a fixed positive lower bound $\\tau_0$ on the stable translation length of any infinite-order element on any domain in its big set, with $\\tau_0$ independent of the generating set; if some element's translation length on a big domain can be arbitrarily small, the short-word constants collapse and uniformity is lost.","fun_headline_variants_meta":{"raw":{"variants":["HHG growth: uniform unless Z-factor splits off","Uniform exponential growth for HHGs, barring Z-factor split","HHGs grow uniformly or quasi-isometrically split Z","Virtually torsion-free HHGs: uniform growth or Z-factor","No Z-factor means uniform growth for HHGs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000946,"raw_usage":{"total_tokens":4005,"prompt_tokens":877,"completion_tokens":3128,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":493,"completion_tokens_details":{"reasoning_tokens":3047}},"tokens_in":493,"tokens_out":3128,"duration_ms":20211,"temperature":1.0,"reasoning_tokens":3047,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T05:56:41.932053+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Exhibit a virtually torsion-free HHG and a sequence of infinite-order elements $g_n$ with a domain $U_n\\in\\operatorname{Big}(g_n)$ such that the stable translation length $\\tau_{U_n}(g_n)$ tends to zero; that would directly refute Lemma 2.23. Alternatively, find a sequence of generating sets for one HHG with no two elements of length bounded by any fixed $M$ generating a free semigroup, while the Cayley graph is not quasi-isometric to $\\mathbb{Z}\\times E$.","supporting_citations":[],"review_version":1}