{"id":"cfd6c403-8fa2-4d14-ada7-326b391d7b1b","arxiv_id":"1908.04440","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A rough isometry invariant, the quasi-hyperbolicity constant, is introduced and computed exactly for Lp spaces and for the α-snowflake of the real line.","lead":"This paper defines a new number, the quasi-hyperbolicity constant, that measures how far a metric space is from being Gromov hyperbolic. It computes this number exactly for Lp spaces and for snowflaked Euclidean lines, giving metric geometers a compact large-scale invariant.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified; the Lp calculation is sound and the snowflake-line optimization, though dense, appears correct.","rationale":"I read the paper in good faith and focused on the strongest claims: the rough-isometry invariant C as a numerical invariant in [1,2], the exact Lp values, and the snowflake-line calculation. The Banach-space arguments are direct and coherent. The lower bound via the James constant and the upper bound via roundness fit together cleanly; the cited external results are used distinctly and without circularity. The Lp classification argument correctly reduces the lower bound to ℓp^2 and the upper bound to the stated roundness constants. For Theorem 6.6, the reduction to the compact domain D and the curve D0 is geometrically natural, and Lemma 6.13's Lagrange-multiplier condition is plausible. I checked the main algebra of the monotonicity argument and found no fatal flaw. The only concrete defects are minor typographical slips in Lemma 6.10, where t+s=1 should read t+s=0 in two places; these do not affect the mathematical conclusion once corrected. Therefore the reader's ACCEPT verdict stands. The optimization in Lemma 6.13 remains the least machine-checked step, so I recommend a symbolic re-derivation and numerical cross-check as a low-cost verification, but I do not regard it as a load-bearing objection.","tokens_in":22121,"tokens_out":48218,"duration_ms":454286,"concrete_test":"Re-derive equation (6.14) from the tangency equations for F and G in Lemma 6.13 using a computer algebra system for symbolic α, and independently verify that t ↦ (1−t^{1−α})/(1−t)^{1−α} is decreasing on (0,1); then run a global numerical maximization of Δ over the compact domain D for α = 1/4, 1/2, and 3/4 and compare with m^α from Theorem 6.6. A mismatch would indicate an optimization error; agreement would close the identified gap.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claims survive scrutiny. The Lp computation is well supported: Proposition 5.2 gives the sharp ℓp^2 value, the lower bound C(B) ≥ J(B) together with Gao–Lau and Komuro–Saito–Tanaka supplies √2 and the Hilbert rigidity, and the roundness upper bound in Theorem 5.11 yields max{2^{1/p}, 2^{1−1/p}} after applying the cited roundness values for Lp. The remaining delicate step is the exact snowflake-line value, Theorem 6.6. The reduction in Lemma 6.10 is valid in intent; the text contains small slips (the boundary of D1 is t+s=0, not t+s=1, and F=1 there), but correcting these preserves the argument. Lemma 6.13's Lagrange calculation and the monotonicity of t ↦ (1−t^{1−α})/(1−t)^{1−α} check out in outline. I did not find a load-bearing error.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces the quasi-hyperbolicity constant C(X,d) of a metric space, defined as the infimal µ for which a (µ,δ)-four-point inequality holds for some δ ≥ 0, together with a restricted version C0(X,d) with δ = 0. The authors prove basic properties: bounded spaces have C = 0, unbounded spaces have C ∈ [1,2], C is a rough isometry invariant, C = C0 for four-point scalable spaces, and unbounded Gromov hyperbolic spaces have C = 1. They show C0 ≤ √2 for Ptolemaic 2-round spaces and hence for subspaces of CAT(0)-spaces, and C ≥ √2 for every Banach space of dimension at least two, with a Hilbert-space rigidity statement in dimension at least three. For Lp-spaces of dimension at least two they obtain the exact value C(Lp) = max{2^{1/p}, 2^{1−1/p}}. For snowflaked metrics they prove the general bound C0(X,d^α) ≤ 2^α and compute the exact value for the α-snowflake of the Euclidean line: C(R, d_E^α) = m^α, where m ≥ 1 solves (m−1)^α + (m+1)^α = 2. The paper also contains examples clarifying the difference between C and C0, the behavior under quasi-isometry, and a lower bound C0 ≥ √2 for Riemannian manifolds of dimension greater than one.","tokens_in":22176,"tokens_out":25235,"duration_ms":235163,"significance":"If the results hold, the paper introduces a useful numerical rough-isometry invariant that quantifies deviation from Gromov hyperbolicity and takes values in [1,2] for unbounded spaces. The Lp calculation is a highlight: it combines external sharp results on the James constant and roundness with explicit four-point configurations, and it is free of fitted parameters. The exact snowflake-line value in Theorem 6.6 is a nontrivial optimization result and is likely to be of independent interest. The paper also carefully documents the failure of quasi-isometry invariance for non-intrinsic spaces and gives clean examples separating C from C0. Overall, the central claims are well supported by explicit and checkable arguments.","major_comments":[{"comment":"The proof of Lemma 6.10 states that the boundary of D1 is D0 ∪ {(t,s) ∈ D : t+s = 1} and that F(t,s) = 1 if and only if t+s = 1. This is incorrect: on D, F takes the value 1 on the segment t+s = 0 (for example F(0,0) = 1), while along t+s = 1 the values of F are generally larger than 1. Since the proof uses the claim F = 1 on this boundary component to discard it in favor of D0, the written proof is not valid as it stands. The intended argument is readily repaired by replacing t+s = 1 with t+s = 0; with that correction, the reduction of the maximum to D0 goes through. The correction should be made explicitly because this lemma is load-bearing for the exact value in Theorem 6.6.","section":"Lemma 6.10 and Theorem 6.6"}],"minor_comments":[{"comment":"In the proof of Proposition 2.2(i), the text reads \"xy + zw ≤ (xz + yw) + (xw + zw)\" and later repeats \"xw + zw\"; both occurrences should read \"xw + yz\".","section":"Proposition 2.2(i)"},{"comment":"The proof of Lemma 6.13 asserts without justification that Fs(a,b) − Gs(a,b) ≠ 0 before applying the implicit function theorem. This is a necessary condition for the argument; please add a short verification or a reference to a calculation establishing that the difference does not vanish on D0.","section":"Lemma 6.13"},{"comment":"After treating the case a < b in detail, the proof states that the expression in (6.14) is positive if a > b without showing the corresponding variable transformation. A brief indication of the symmetric argument would improve readability and completeness.","section":"Lemma 6.13"},{"comment":"There is a typo: \"Quesition 3.7\" should be \"Question 3.7\".","section":"Question 3.7"}],"recommendation":"minor_revision","confidential_remarks":"The paper's main claims appear sound. The flaws I found are local: an incorrect boundary description in Lemma 6.10 that is easily corrected, a small unproved step in Lemma 6.13, and a few typos. Once the authors make these corrections, I expect the paper to be suitable for publication. The Lp computation and the snowflake-line optimization are the strongest parts of the work."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know: this paper introduces a quasi-hyperbolicity constant C(X,d), a rough isometry invariant that quantifies deviation from Gromov hyperbolicity. The main results are exact: for Lp spaces it equals max{2^{1/p}, 2^{1−1/p}}, for the α-snowflake of the Euclidean line it is m^α with (m−1)^α+(m+1)^α=2, and for CAT(0) spaces the restricted constant is ≤√2. It also gives counterexamples showing C=1 does not imply hyperbolicity outside the roughly geodesic setting. These are genuinely new and useful calculations, not just a repackaging of known inequalities.\n\nThe paper is honest and well-structured. It uses James constants and roundness carefully, and it clearly labels the places where numerical evidence suggests but does not prove an exact value (e.g., the d_2^α snowflake conjecture and the X_m family). The citation pattern is good: it leans on external theorems (Gao–Lau, Komuro–Saito–Tanaka, Bridson, Enflo) and does not hide behind self-citations.\n\nThe main soft spot is Lemma 6.13, the long optimization argument behind the snowflake-line theorem. It is intricate, with several pages of sign analysis, and it is not machine-checked. There are also small slips in Lemma 6.10: the boundary of D1 is t+s=0 (where F=1), not t+s=1, and the text's claim that F=1 iff t+s=1 is wrong. The stress-test note and my own reading suggest these are correctable typos, and the argument can be patched without changing the conclusion, but a referee should ask the authors to rewrite that section more cleanly. Beyond that, the proofs are direct and checkable; the Lp calculation is particularly clean.\n\nThis is a solid paper for metric geometry and Banach space theory. It does not resolve a long-open question, but it introduces a natural invariant and computes it in enough cases to be a useful reference. I would send it to peer review and expect acceptance after minor revisions.\n\nRecommendation: accept with revisions; worth citing.","headline":"A clean new invariant with exact Lp and snowflake-line computations; the proofs are direct except for one intricate optimization that looks right.","tokens_in":22794,"tokens_out":2689,"would_cite":true,"duration_ms":27295,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["51K05","46B20","51F99","51M10"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper introduces the quasi-hyperbolicity constant and shows that it is a rough isometry invariant in [1,2], equal to 1 on unbounded Gromov hyperbolic spaces, and computable exactly for L_p spaces and snowflaked lines.","keywords":["quasi-hyperbolicity constant","four-point inequality","Gromov hyperbolic","rough isometry invariant","James constant","roundness","L_p spaces","snowflake metric"],"falsifier":"For any fixed $\\alpha\\in(0,1)$, numerically maximize $\\Delta(x,y,z,w)=(|x-y|^\\alpha+|z-w|^\\alpha)/\\max\\{|x-z|^\\alpha+|y-w|^\\alpha,|x-w|^\\alpha+|y-z|^\\alpha\\}$ over four points of $\\mathbb{R}$; if the maximum exceeds $m^\\alpha$ with $(m-1)^\\alpha+(m+1)^\\alpha=2$, the theorem is false. Equivalently, evaluate the left side of equation (6.14) on a dense grid of $(a,b)$ with $-1<a<b<1$ and $a+b>0$; any zero would break the sign analysis behind Lemma 6.13.","tokens_in":21822,"feed_emoji":"📐","tokens_out":14297,"duration_ms":110889,"temperature":0.7,"pith_summary":"The paper introduces the quasi-hyperbolicity constant $C(X,d)$, the smallest multiplier $\\mu$ in a four-point inequality that may carry an additive error $\\delta$, and proposes it as a rough-isometry invariant measuring how far a metric space deviates from Gromov hyperbolicity. For unbounded spaces $C$ lies in $[1,2]$, and it equals $1$ for every unbounded Gromov hyperbolic space. The paper's exact calculations are $C(L_p)=\\max\\{2^{1/p},2^{1-1/p}\\}$ for any separable $L_p$ space of dimension at least two, $C(B)\\ge\\sqrt{2}$ for every Banach space of dimension at least two, and $C(\\mathbb{R},d_E^\\alpha)=m^\\alpha$ for the $\\alpha$-snowflake of the Euclidean line, where $m$ solves $(m-1)^\\alpha+(m+1)^\\alpha=2$. These results give a computable large-scale invariant that distinguishes Euclidean, Banach, and snowflaked geometries.","feed_headline":"A new constant measures deviation from hyperbolic geometry","feed_subtitle":"It is a rough-isometry invariant, equals 1 on hyperbolic spaces, and is computed exactly for L_p and snowflaked lines.","key_machinery":"The load-bearing object is the $(\\mu,\\delta)$-four-point inequality $xy+zw\\le\\mu\\max\\{xz+yw,xw+yz\\}+2\\delta$ together with the ratio $\\Delta(x,y,z,w)=(xy+zw)/\\max\\{xz+yw,xw+yz\\}$. For Banach spaces the proof uses the James constant $J(B)$ as a lower bound and the roundness invariant $r(B)$ as an upper bound via $C(B)\\le 2^{1/r(B)}$. For the snowflaked line, the argument reduces $C(\\mathbb{R},d_E^\\alpha)$ to a two-variable maximization of two rational functions $F$ and $G$ over a compact domain, and Lemmas 6.10 and 6.13 show the maximum is attained on the diagonal $t=s$, converting the problem into the scalar equation $(m-1)^\\alpha+(m+1)^\\alpha=2$.","core_discovery":"The central claim is that the infimum over $\\mu$ in the $(\\mu,\\delta)$-four-point inequality defines a meaningful numerical invariant of a metric space, and that this invariant can be computed exactly for broad classes of spaces. In the paper's terms: for any separable measure space with $\\dim L_p\\ge 2$, $C(L_p(\\Omega,\\Sigma,\\mu))=\\max\\{2^{1/p},2^{1-1/p}\\}$; for any Banach space of dimension at least two, $C(B)\\ge\\sqrt{2}$; for any CAT(0) space, $C_0\\le\\sqrt{2}$; and for the $\\alpha$-snowflake of the real line, $C(\\mathbb{R},d_E^\\alpha)=m^\\alpha$ with $(m-1)^\\alpha+(m+1)^\\alpha=2$. The constant is a rough isometry invariant, takes values in $[1,2]$ on unbounded spaces, and equals $1$ for unbounded Gromov hyperbolic spaces; a proper CAT(0) space with $C=1$ is necessarily Gromov hyperbolic.","pith_inferences":["The combination of the James-constant lower bound and roundness upper bound suggests a numerical scale on which Hilbert spaces are the roundest Banach spaces; whether $C(B)=\\sqrt{2}$ characterizes Hilbert spaces in all dimensions is a natural open test.","The diagonal-maximization step for the snowflake line is a transferable strategy: any one-dimensional metric with a scaling action and a two-variable ratio may reduce to a scalar equation, so the same method could give exact constants for snowflakes of other normed lines.","The conjecture $C(\\mathbb{R}^n,d_2^\\alpha)=2^{\\alpha/2}$ for $n\\ge 2$ is directly testable by the same $\\Delta$-ratio numerics; the line case's deviation from $2^{\\alpha/2}$ shows the constant genuinely depends on dimension, not only on the snowflake exponent.","Because Proposition 2.9 identifies $C=C_0$ for snowflaked normed spaces, the restricted constant $C_0$ is the right computational target for such spaces, and differences between $C_0$ and $C$ measure small-scale versus large-scale geometry."],"forward_implications":["Unbounded Gromov hyperbolic spaces have $C=1$, so within a rough isometry class the value $1$ detects large-scale hyperbolicity; for proper CAT(0) spaces the converse also holds.","Every Banach space of dimension at least two has $C(B)\\ge\\sqrt{2}$, and if the dimension is at least three then equality $C(B)=\\sqrt{2}$ forces $B$ to be a Hilbert space.","For separable $L_p$ spaces, $C(L_p)=\\max\\{2^{1/p},2^{1-1/p}\\}$, so the constant is minimized at $p=2$ with value $\\sqrt{2}$ and approaches $2$ as $p\\to 1$ or $p\\to\\infty$.","For any metric space and any $0<\\alpha\\le 1$, $C_0(X,d^\\alpha)\\le 2^\\alpha$, and this upper bound is sharp for $(\\mathbb{R}^n,d_\\infty^\\alpha)$ with $n\\ge 2$.","Snowflaking the real line strictly raises the quasi-hyperbolicity constant above $1$ for every $0<\\alpha<1$; in particular $C(\\mathbb{R},d_E^{1/2})=\\sqrt{5}/2$."],"supporting_citations":[{"why":"Introduces Gromov hyperbolic spaces, the class whose deviation the new constant measures.","marker":"[Gro87]"},{"why":"Supplies the equivalence between $\\delta$-hyperbolicity and the four-point inequality that the new definition generalizes.","marker":"[V¨ ai05]"},{"why":"Gives the James-constant lower bound $J(B)\\ge\\sqrt{2}$ for non-trivial Banach spaces.","marker":"[GL90]"},{"why":"Shows that dimension at least three with $J(B)=\\sqrt{2}$ forces a Hilbert space, completing Theorem 5.8.","marker":"[KST16]"},{"why":"Introduces roundness and supplies $r(L_p)=p$ for $1\\le p\\le 2$, yielding the upper bound in Corollary 5.12.","marker":"[Enf69]"},{"why":"Supplies $r(L_p)=1/(1-1/p)$ for $2\\le p\\le\\infty$, completing the $L_p$ upper bound.","marker":"[LTW97]"},{"why":"Provides the classification of separable $L_p$ spaces used to reduce the lower bound to the $\\ell^2_p$ subspace.","marker":"[JL01]"},{"why":"Provides the Flat Plane Theorem used to show a proper CAT(0) space with $C=1$ is Gromov hyperbolic.","marker":"[Bri95]"},{"why":"Gives the CAT(0) four-point subembedding used to prove Ptolemaic and 2-round properties, giving $C_0\\le\\sqrt{2}$.","marker":"[BH99]"}],"fun_headline_variants":["Quasi-hyperbolicity: a new rough-isometry invariant","How non-hyperbolic is a space? A new constant answers","Exact deviation from hyperbolicity: L_p and snowflake cases","Rough isometry invariant: quasi-hyperbolicity constant in [1,2]","New metric constant measures hyperbolicity deviation"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The exact snowflake-line value rests on Lemma 6.13, which asserts—by sign analysis of equation (6.14) and monotonicity of $t\\mapsto (1-t^{1-\\alpha})/(1-t)^{1-\\alpha}$—that the maximum on the curve $F=G$ occurs at $t=s$; if that optimization step hides an error, the formula $C(\\mathbb{R},d_E^\\alpha)=m^\\alpha$ is unproved.","fun_headline_variants_meta":{"raw":{"variants":["Quasi-hyperbolicity: a new rough-isometry invariant","How non-hyperbolic is a space? A new constant answers","Exact deviation from hyperbolicity: L_p and snowflake cases","Rough isometry invariant: quasi-hyperbolicity constant in [1,2]","New metric constant measures hyperbolicity deviation"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000725,"raw_usage":{"total_tokens":3260,"prompt_tokens":968,"completion_tokens":2292,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":584,"completion_tokens_details":{"reasoning_tokens":2200}},"tokens_in":584,"tokens_out":2292,"duration_ms":17400,"temperature":1.0,"reasoning_tokens":2200,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:43:54.877365+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For any fixed $\\alpha\\in(0,1)$, numerically maximize $\\Delta(x,y,z,w)=(|x-y|^\\alpha+|z-w|^\\alpha)/\\max\\{|x-z|^\\alpha+|y-w|^\\alpha,|x-w|^\\alpha+|y-z|^\\alpha\\}$ over four points of $\\mathbb{R}$; if the maximum exceeds $m^\\alpha$ with $(m-1)^\\alpha+(m+1)^\\alpha=2$, the theorem is false. Equivalently, evaluate the left side of equation (6.14) on a dense grid of $(a,b)$ with $-1<a<b<1$ and $a+b>0$; any zero would break the sign analysis behind Lemma 6.13.","supporting_citations":[],"review_version":1}