{"id":"2771b3fc-3d7c-4f30-bcc7-f2877acda63d","arxiv_id":"2505.23572","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Linear programming upper bounds for sphere packing densities are proven for homogeneous spaces, including a hyperbolic bound conjectured by Cohn and Zhao.","lead":"This paper proves new upper limits on how densely equal-sized spheres can be packed in curved spaces such as hyperbolic space and Heisenberg groups. The method extends a famous technique from flat Euclidean space and settles a version of an open conjecture about hyperbolic sphere packings.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The LP inequality itself is sound for stationary point processes, but the paper's central claim about Bowen-Radin deterministic density hinges on Proposition 4.19 and Corollary 4.21, both deferred to the unpublished companion [30].","rationale":"The reader's weakest-assumption analysis identifies the same load-bearing bridge: the equality between deterministic Bowen-Radin density and stationary point-process intensity rests on Definition 4.15 and Proposition 4.19, with the latter's proof deferred to [30]. I find this concern accurate and central. The self-contained part of the argument is nevertheless credible: the estimate in Theorem 5.5 is valid because P is 2r-uniformly discrete, so all distinct center pairs lie outside the support where the witness function is nonpositive, and positive-definiteness of f supplies the normalization at the identity. Thus the probabilistic inequality is not the weak point. The genuine open question is whether every Bowen-Radin dense deterministic packing is the generic instance of a stationary point process with the same density. Since Theorem B and its hyperbolic special case Theorem A are advertised as bounds on △(r, G/K), the paper is conditional on an unpublished companion. The secondary issues—the equality typo in Theorem A and the restricted Schwartz witness class relative to the full Cohn-Zhao conjecture—do not change the verdict but should be corrected or qualified. Given the reader already marked the paper CONDITIONAL, my stress-test does not move the verdict.","tokens_in":28167,"tokens_out":10214,"duration_ms":109297,"concrete_test":"Provide the deferred proofs in [30]: (a) verify Definition 4.15 for (SO(n,1), SO(n)) by reducing F to a countable translation-invariant subset of continuous functions for which the Gorodnik-Nevo theorem yields a single conull set of invariantly generic point sets for every G-invariant probability measure; (b) prove Proposition 4.19 in the special cases of lattice orbits and regular model sets in H^n by computing D(Λ_r) and D_r(P0, x) explicitly and checking equality for every x. If the conull set cannot be chosen independently of h ∈ G and f ∈ F, then 'invariantly generic' in Definition 4.17 must be weakened or Corollary 4.21 fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem B is stated as an unconditional bound on the Bowen-Radin density △(r, G/K), but the proof reaches that conclusion through Corollary 4.21, which identifies △(r, X) with the probabilistic optimal density △prob(r, X) only if the invariant pointwise ergodic theorem of Definition 4.15 holds. That theorem is not proved here: its content is exactly that for every G-invariant probability measure on UD_{2r}(X) there is a conull set of invariantly generic point sets. Moreover, Proposition 4.19, which asserts that a generically measured deterministic packing has the same density as the associated stationary point process, is stated with proof deferred to [30]. If either statement fails, the deterministic conclusion of Theorem B is not established; the argument in Section 5.2 would still bound only △prob(r, X), as Remark 5.6 concedes. A further unaddressed technical point is that the family F in Definition 4.15 is uncountable (all Riemann integrable compactly supported functions on UD_{2r}(X)). Pointwise ergodic theorems provide one conull set per function, or per countable dense subset, and a single conull set for all f ∈ F and all h ∈ G requires a separability and translation-invariance argument that the manuscript does not supply. The equality sign in Theorem A is also an error; the intended statement is an upper bound.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a general framework for linear programming upper bounds on sphere packing densities in homogeneous spaces G/K arising from convenient Gelfand pairs (G,K,d,S(G,K)). The main result, Theorem B, asserts that for every witness function f in a Schwartz-like space with nonnegative spherical transform and appropriate sign condition, the Bowen-Radin optimal packing density satisfies \\Delta(r,G/K) \\le m_{G/K}(B(x0,r)) f(e)/\\hat f(1). The proof proceeds by associating to a 2r-uniformly discrete stationary point process its autocorrelation measures, applying the Plancherel-Godement theorem to pass to the spherical transform side, and then invoking a probabilistic formulation of packing density. A special case for hyperbolic spaces, Theorem A, is claimed to resolve a conjecture of Cohn and Zhao. The paper also gives explicit formula-level statements for Euclidean space, Heisenberg groups, and Riemannian symmetric spaces, and includes an appendix proving a spherical Bochner-Schwartz theorem for the Heisenberg group.","tokens_in":28490,"tokens_out":2358,"duration_ms":25332,"significance":"If the main theorem is established, it provides a common framework encompassing the Euclidean Cohn-Elkies bounds, hyperbolic packing bounds conjectured by Cohn and Zhao, and bounds for other homogeneous spaces, with a proof that does not rely on approximation by periodic packings. The use of positive-definite autocorrelation measures and the Plancherel-Godement theorem is elegant and, as far as Section 5.2 goes, internally coherent. The paper also ships a nontrivial self-contained appendix (Theorem A.1) for the Heisenberg case, and the explicit integral conditions in Examples 1.2 and 1.5 are directly usable by specialists. However, the deterministic content of the main theorems currently depends on results deferred to an in-preparation companion paper, and the stated equality in Theorem A is stronger than what the proof establishes; both issues must be resolved before the results can be fully credited.","major_comments":[{"comment":"The central bridge between deterministic Bowen-Radin density and probabilistic point-process density is Proposition 4.19, which asserts that every generically measured point set P0 has a well-defined density equal to the intensity-based density of any stationary point process with distribution \\mu_{P0}. The proof is deferred entirely to the companion paper [30], which is listed as 'In preparation'. Corollary 4.21, which identifies the deterministic optimal density \\Delta(r,X) with \\Delta_{prob}(r,X), depends directly on this proposition and on the invariant pointwise ergodic theorem. Since Theorem B is stated as an unconditional bound on \\Delta(r,G/K), the absence of a proof of Proposition 4.19 is a load-bearing gap. The manuscript should either include a full proof or explicitly state the theorem as conditional on the companion paper, with Remark 5.6 serving only as a fallback for the probabilistic bound.","section":"§4.2, Proposition 4.19 and Corollary 4.21"},{"comment":"The invariant pointwise ergodic theorem is asserted for the uncountable family F of all Riemann integrable compactly supported functions on UD_{2r}(X). A pointwise ergodic theorem typically yields a conull set of generic points for each fixed function, or for a countable separating family, and obtaining a single conull set simultaneously for all f in F and all h in G requires a separability and translation-invariance argument that is not supplied in the manuscript. Since the main theorem relies on this theorem to pass from generic point sets to stationary point processes, this is not a minor technicality. The author should either justify the existence of such a common conull set (for instance by proving that F admits a countable dense subfamily with the appropriate invariance properties for the specific groups in the examples) or replace F by a countable family for which the ergodic theorem is known.","section":"§4.2, Definition 4.15 and Example 4.16"},{"comment":"Theorem A is printed with an equality, \\Delta(r,\\mathbb{H}^n) = m_{\\mathbb{H}^n}(B(x0,r)) f(x0)/\\hat f(1). The proof in Section 5.2 (see Theorem 5.5) yields only an upper bound, and there is no matching lower bound or construction of packings attaining the right-hand side. The equality is therefore an error and should be replaced by '\\le'. This is a mathematical accuracy issue in the statement of the paper's headline result, not merely a typographical slip.","section":"Theorem A (Section 1.2)"},{"comment":"The proof correctly establishes the bound for the intensity i(\\Lambda) of a stationary point process and hence, by Proposition 4.10, for the probabilistic density D(\\Lambda). The final sentence 'And thus by Corollary 4.21 we obtain the result' silently imports the full weight of the invariant pointwise ergodic theorem and Proposition 4.19. Because those ingredients are unproved here, the logical status of the theorem's conclusion about \\Delta(r,X) should be flagged in the theorem statement itself, e.g. by stating the bound for \\Delta_{prob}(r,X) unconditionally and for \\Delta(r,X) conditional on the ergodic bridge.","section":"§5.2, Theorem 5.5, last sentence"}],"minor_comments":[{"comment":"Definition 4.17 defines generically measured point sets in UD_{2r}(X), while Proposition 4.19 states its conclusion for P0 \\in UD_r(X). The index r appears inconsistently; either the definition should use r-uniformly discrete sets or the proposition should use 2r. Please clarify.","section":"§4.2, Definition 4.17 vs Proposition 4.19"},{"comment":"There are several typographical slips: 'paair' in Section 1.5, 'Randon' in Lemma 2.5, 'Proposiiton' in Section 3.1, and 'bounded support' for functions in Proposition 3.7 that should be 'bounded' or 'compact support' consistently. These do not affect the mathematics but should be corrected.","section":"Throughout"},{"comment":"The remark states that Cohn and Zhao conjectured the bound with f continuous and integrable, but it does not comment on how the stronger Schwartz-like regularity assumption in Theorem A affects the comparison with their proposed class. A short remark on whether the results of [21] imply the bound for a wider class of f in the hyperbolic case would help the reader.","section":"§1.2, Remark 1.1(iii)"},{"comment":"The statement says the Abel transform Af is in W_r(a), but the proof only checks the sign condition and the Fourier nonnegativity. Please state explicitly which conditions of Definition 1.3 are verified and which are inherited from f \\in W_r(G/K), or add the missing verification.","section":"§5.3, Proposition 5.7"}],"recommendation":"major_revision","confidential_remarks":"The manuscript's central theorem is conditional on two unproved ingredients: Proposition 4.19, whose proof is deferred to a companion paper in preparation, and the uncountable-family invariant pointwise ergodic theorem in Definition 4.15. The equality sign in Theorem A is also plainly too strong. These are fixable in a revision provided the companion paper is available or the theorems are explicitly restated as conditional. However, the current dependence on an unpublished companion is significant for a journal publication, and the editor should ensure that the companion is either included or that the main theorems are reformulated so as not to rely on it."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a genuine step beyond Cohn–Elkies and Cohn–Zhao. The core inequality, bounding the intensity of a stationary 2r-uniformly discrete point process by f(e)/f-hat(1), looks sound, and the autocorrelation/Plancherel–Godement route is a real alternative to Poisson summation and periodic approximation. The Heisenberg formulas and the Bochner–Schwartz proof in Appendix A are new and useful, and Proposition 5.7 (Abel transform sends witnesses to Euclidean witnesses) is a nice observation.\n\nThe soft spots are concentrated where the paper turns probabilistic bounds into deterministic Bowen–Radin bounds. Proposition 4.19, the bridge from generically measured packings to stationary point processes with the same density, is deferred to the companion [30]. Corollary 4.21, which equates △(r,X) with △prob(r,X), depends on the invariant pointwise ergodic theorem in Definition 4.15, and that theorem is not proved here. The family F is uncountable, so standard ergodic theorems give one conull set per function; getting a single conull set for all Riemann integrable compactly supported functions and all h ∈ G needs a separability/translation-invariance argument that is not supplied. That is a genuine gap, not a nitpick. Also, Theorem A is printed with an equality sign; the proof gives an upper bound, and for nonperiodic packings equality would require a matching lower bound that is not there. The claim to have solved Cohn–Zhao should be softened: the witness class is smaller than the one they conjectured, so the conjecture as stated is not resolved, only a version with extra regularity.\n\nNone of this kills the paper's core. If the companion delivers Proposition 4.19 and the ergodic theorem is repaired, Theorem B stands as stated. As it is, the unconditional claim is conditional on unpublished work. The right move is to send it to a serious referee, with the explicit instruction that the deterministic bridge must be either proved or carved out, and the equality in Theorem A corrected to an inequality.\n\nI would bring it to reading group; there is enough here to discuss even with the caveat.","headline":"Unified LP packing bounds via autocorrelation measures are real, but the deterministic bridge rests on an unproved ergodic theorem in a companion paper.","tokens_in":28961,"tokens_out":1863,"would_cite":true,"duration_ms":18271,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["52C17","22E46","43A85","60G55"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper establishes a general linear programming upper bound on sphere-packing density in every 'convenient commutative space,' and derives the conjectured hyperbolic-space version as a special case.","keywords":["sphere packing","linear programming bounds","hyperbolic space","Gelfand pairs","homogeneous spaces","point processes","autocorrelation measures","Bowen-Radin density"],"falsifier":"Find a convenient commutative space and a witness function f for which the ratio m_{G/K}(B(x_0, r)) f(e)/bf(1) is strictly smaller than the density of an explicit generically measured packing; then the master inequality is false. Equivalently, exhibit a G-invariant probability measure on the space of 2r-uniformly discrete point sets with no conull set of invariantly generic point sets, which would break the link between deterministic and probabilistic packing density that the proof requires.","tokens_in":27981,"feed_emoji":"⚪","tokens_out":10814,"duration_ms":96745,"temperature":0.7,"pith_summary":"The paper proves that sphere-packing density in a broad class of homogeneous spaces is controlled by one linear programming inequality: for any convenient commutative space—a Gelfand pair (G, K) equipped with an invariant metric and a Schwartz-like function space—every admissible witness function f gives the bound △(r, G/K) ≤ m_{G/K}(B(x_0, r)) f(e)/bf(1). The main target is hyperbolic space, where this bound was conjectured but previously known only for periodic packings; the proof does not rely on approximating arbitrary packings by periodic ones. Instead, deterministic packings with well-defined density are viewed as generic instances of stationary random packings, so the bound becomes an intensity estimate i(Λ) ≤ f(e)/bf(1) for invariant point processes. If the theorems are correct, the same machinery yields explicit packing bounds in Riemannian symmetric spaces, Heisenberg groups, and any other space satisfying the needed ergodic hypothesis.","feed_headline":"Hyperbolic packing bound proved by random-packings route","feed_subtitle":"A commutative-space inequality covers Euclidean, hyperbolic, and Heisenberg packings without periodic approximation.","key_machinery":"The load-bearing object is the reduced autocorrelation measure η_Λ of a stationary 2r-uniformly discrete point process Λ, together with its spherical transform, the spherical diffraction. The argument uses two facts: η_Λ and η_Λ^+ are positive-definite bi-K-invariant measures—invariant under the compact subgroup K from both sides—so the Plancherel-Godement theorem gives them positive spherical transforms; and the identity η_Λ^+ = η_Λ + i(Λ)^2 m_G becomes the spectral identity bη_Λ^+ = bη_Λ + i(Λ)^2 δ_1. Feeding a witness function f with bf ≥ 0 into these measures yields i(Λ)^2 bf(1) ≤ bη_Λ^+(bf) = η_Λ^+(f) ≤ f(e) i(Λ), where the last step uses the sign condition f ≤ 0 beyond distance 2r. A Schwartz-like function space and a spherical Bochner-Schwartz theorem extend the argument from compactly supported functions to the witness class, and the invariant pointwise ergodic theorem transfers the intensity bound to deterministic Bowen-Radin density.","core_discovery":"On its own terms, the central claim is a master inequality: whenever (G, K, d, S(G, K)) is a convenient Gelfand pair, every witness function f that is non-positive outside radius 2r and has nonnegative spherical transform with bf(1) > 0 satisfies △(r, G/K) ≤ m_{G/K}(B(x_0, r)) f(e)/bf(1). For the hyperbolic pair (SO(n, 1), SO(n)), the spherical transform is written through Gauss hypergeometric functions, and the theorem yields the conjectured hyperbolic linear programming bound. The same inequality also recovers the classical Euclidean linear programming bound and supplies new explicit bounds for Heisenberg groups with the Cygan-Koranyi metric and for Riemannian symmetric spaces of noncompact and compact type. The proof works by replacing Poisson summation with the fact that the spherical diffraction of a stationary point process has an atom at the trivial character, so applying the positive transform of a witness function isolates the intensity term.","pith_inferences":["The spectral mechanism suggests the inequality should extend beyond Gelfand pairs to any homogeneous space where point-process autocorrelations are positive definite and the trivial character is a spectral atom; the Gelfand-pair assumption mainly makes the Plancherel side explicit.","The deferred bridge is the part to scrutinize: if the invariant pointwise ergodic theorem or the density-intensity equality (Proposition 4.19, proved in the companion paper) fails for some non-amenable group, Theorem B still bounds probabilistic packing density but not deterministic density.","The printed equality in the hyperbolic theorem appears to be a typographical slip; the surrounding argument and the general theorem establish an upper bound, and the equality should be read as that upper bound.","Because cut-and-project model sets are generically measured, the same route likely yields packing bounds for quasicrystal-like packings in hyperbolic and other symmetric spaces, an application the paper mentions only in passing."],"forward_implications":["If the master inequality holds, the hyperbolic linear programming conjecture is settled: for every n ≥ 2 and every admissible radial function h, △(r, H^n) ≤ C(r) h(0) / ∫_0^∞ h(t) sinh(t)^{n-1} dt.","The same bound applies to every irreducible Riemannian symmetric space of noncompact type, without needing to know whether optimal packings can be approximated by periodic ones.","Compact symmetric spaces, including spheres, inherit linear programming bounds for packings and codes as a special case.","Heisenberg groups with the Cygan-Koranyi metric acquire explicit linear programming bounds, showing the method reaches non-symmetric homogeneous spaces.","Since the proof never approximates general packings by periodic packings, it sidesteps the major open question of periodic approximation in high dimensions."],"supporting_citations":[{"why":"The Euclidean linear programming bound being generalized; supplies the witness-function framework.","marker":"[19]"},{"why":"The conjectured hyperbolic linear programming bound and its proof for periodic packings.","marker":"[21]"},{"why":"Introduces the deterministic density notion for hyperbolic packings that the paper generalizes.","marker":"[16]"},{"why":"Establishes density-intensity relations for generically measured packings, the basis for Proposition 4.19.","marker":"[17]"},{"why":"Constructs the autocorrelation measures and intensity formulas for stationary point processes used in the proof.","marker":"[8]"},{"why":"Provides the invariant pointwise ergodic theorem for semisimple Lie groups, a hypothesis of the main theorem.","marker":"[28]"},{"why":"Gives the Plancherel-Godement theorem that yields spherical transforms of positive-definite measures and distributions.","marker":"[27]"},{"why":"Supplies the spherical Bochner-Schwartz theorem needed to enlarge the witness class to Schwartz-like functions.","marker":"[5]"},{"why":"Companion paper to which the proof of the density-intensity bridge for generically measured packings is deferred.","marker":"[30]"}],"fun_headline_variants":["Hyperbolic packing bound proven via random-packings","LP bound for hyperbolic sphere packings proved","Commutative-space inequality nails hyperbolic packing","Cohn-Zhao hyperbolic packing conjecture resolved","Stationary point process yields packing bound"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything depends on the bridge assumption that a deterministic packing with a well-defined density gives a stationary random packing whose intensity equals that density, and that the pointwise ergodic theorem supplies enough such packings; the proof of this bridge is deferred to the companion paper.","fun_headline_variants_meta":{"raw":{"variants":["Hyperbolic packing bound proven via random-packings","LP bound for hyperbolic sphere packings proved","Commutative-space inequality nails hyperbolic packing","Cohn-Zhao hyperbolic packing conjecture resolved","Stationary point process yields packing bound"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000379,"raw_usage":{"total_tokens":1920,"prompt_tokens":759,"completion_tokens":1161,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":375,"completion_tokens_details":{"reasoning_tokens":1095}},"tokens_in":375,"tokens_out":1161,"duration_ms":9476,"temperature":1.0,"reasoning_tokens":1095,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T12:42:46.348177+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find a convenient commutative space and a witness function f for which the ratio m_{G/K}(B(x_0, r)) f(e)/bf(1) is strictly smaller than the density of an explicit generically measured packing; then the master inequality is false. Equivalently, exhibit a G-invariant probability measure on the space of 2r-uniformly discrete point sets with no conull set of invariantly generic point sets, which would break the link between deterministic and probabilistic packing density that the proof requires.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"The Euclidean linear programming bound being generalized; supplies the witness-function framework."},{"cited_title":"J.163 (2014), no","cited_arxiv_id":null,"evidence_quote":"The conjectured hyperbolic linear programming bound and its proof for periodic packings."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the deterministic density notion for hyperbolic packings that the paper generalizes."},{"cited_title":"Dedicata 104 (2004), 37–59","cited_arxiv_id":null,"evidence_quote":"Establishes density-intensity relations for generically measured packings, the basis for Proposition 4.19."},{"cited_title":"172, Princeton University Press, Princeton, NJ, 2010","cited_arxiv_id":null,"evidence_quote":"Provides the invariant pointwise ergodic theorem for semisimple Lie groups, a hypothesis of the main theorem."},{"cited_title":"Selberg, Séminaire Bourbaki, Vol","cited_arxiv_id":null,"evidence_quote":"Gives the Plancherel-Godement theorem that yields spherical transforms of positive-definite measures and distributions."},{"cited_title":"Barker,The spherical Bochner theorem on semisimple Lie groups, J","cited_arxiv_id":null,"evidence_quote":"Supplies the spherical Bochner-Schwartz theorem needed to enlarge the witness class to Schwartz-like functions."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Companion paper to which the proof of the density-intensity bridge for generically measured packings is deferred."}],"review_version":1}