{"id":"58238e26-8bd0-403a-8a93-9d0e4d436803","arxiv_id":"1908.08441","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For any bounded domain, the behavior of optimal eigenvalues over unions of its scaled copies encodes whether Pólya's conjecture holds, yielding a trichotomy and a packing-density lower bound.","lead":"This math paper studies the sequence of best k-th eigenvalues you can get by gluing together scaled copies of a fixed shape, and shows the pattern of these best values reveals whether a famous conjecture about vibrating membranes (Pólya's conjecture) holds for that shape. It also finds a new lower bound for these eigenvalues in terms of how densely the shape can pack into space.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 2.6's one-line perturbation argument is the load-bearing step for both main theorems and needs a rigorous proof.","rationale":"Read in good faith: the paper's main theorems are plausible and the surrounding infrastructure (existence Lemmas 2.1–2.2, sub/superadditivity, Fekete lemma, packing lower bound) is coherent. The flagged concern is not a demonstrated counterexample but an under-justified structural lemma on which the trichotomy and Theorem 1.4 both depend. I traced Lemma 2.6 through Sections 3 and 4: Proposition 3.5 uses the additive identity; Theorem 1.4's proof uses it repeatedly and also needs the subset version; Algorithm 5.2 is exactly the recursive decomposition. The 'one also sees' sentence is the place where a hidden assumption would enter. I do not find an internal inconsistency or a known counterexample; a rigorous re-derivation would settle it. Hence the reader's CONDITIONAL verdict is appropriate and no verdict change is needed.","tokens_in":26086,"tokens_out":27284,"duration_ms":268345,"concrete_test":"Prove Lemma 2.6 rigorously by a two-step perturbation: (i) for a component whose j_q-th eigenvalue is strictly below λ_k^*, shrink it by factor (1−ε) and expand every other component by the volume-preserving factor, then use continuity of each component's j_q-th eigenvalue to show the k-th eigenvalue of the union is strictly below λ_k^* for all sufficiently small ε; (ii) for any subcollection S of components with total rank j', show the normalized subcollection realises λ_{j'}^* by replacing S with a scaled λ_{j'}^*-optimizer and distributing the freed volume to all components. If step (i) or (ii) fails for some generator (e.g., disk or 1:5 rectangle), compute the resulting improved competitor numerically via Algorithm 5.2 and check whether λ_k decreases.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim rests on Lemma 2.6: every optimizer Ω_k^* splits into scaled optimizers Ω_{j_q}^*, with λ_k^{*d/2} = Σ λ_{j_q}^{*d/2}. The proof's key assertion—'One also sees that the largest eigenvalue smaller or equal to λ_k^*(R) of each component has to be equal to λ_k^*(R)'—is not demonstrated. If some component had its relevant eigenvalue strictly below λ_k^*, the intended shrink-that-component/expand-the-others variation needs a quantitative check: the kth eigenvalue must drop, which requires that expanding the other components strictly lowers every eigenvalue that was at λ_k^* and that the shrunk component's selected eigenvalue remains below λ_k^*; this is plausible but not automatic when eigenvalues coincide or when components have different ranks. The same variational principle is silently used later when Theorem 1.4 bounds ν(Ξ_k): the proof needs that every subcollection of components of an optimizer is itself an optimizer for the sum of its ranks (so λ_{j'}^{*d/2} = Σ_{q in subset} λ_{j_q}^{*d/2}). Lemma 2.6 as stated only gives this for individual components, not for arbitrary subsets. The trichotomy, Theorem 1.4, and the recursive numerical algorithm all inherit this gap. The gap appears fixable by a minimal-volume replacement argument, but as written it is the least secure link.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies, for a fixed bounded domain Omega, the family R(Omega) of all disjoint unions of scaled and isometric copies of Omega with total volume at most 1 (for Dirichlet) or at least 1 (for Neumann). For each k it analyzes the extremal values lambda_k^*(R) = inf lambda_k and mu_k^*(R) = sup mu_k over this family. The main results are: existence of extremizers and sub/superadditivity of the extremal eigenvalue sequences (Section 2); a Wolf-Keller type decomposition of every extremizer into scaled extremizers of smaller rank (Lemmas 2.6 and 2.7); a trichotomy (Theorem 1.3) relating whether the generator Omega is an extremizer infinitely often to the validity of Polya's conjecture in R(Omega); an equivalence, under two-term Weyl asymptotics, between the strong Polya conjecture and convergence of the largest scaling coefficient to 1 (Theorem 1.4); a lower bound for the Dirichlet limit in terms of a newly defined asymptotic packing density (Theorem 1.8); and numerical experiments for the disk, the square, and a 1:5 rectangle (Section 5).","tokens_in":26405,"tokens_out":20579,"duration_ms":190256,"significance":"The paper offers a genuinely new structural framework for Polya's conjecture: rather than seeking universal bounds over all domains, it shows that the validity of the conjecture for a fixed generator is encoded in the asymptotic behavior of optimizers in R(Omega). If the technical decomposition lemmas are made fully rigorous, the trichotomy and the packing-density bounds form a valuable contribution that recovers and strengthens results of Polya and Urakawa. The paper is also honest that the numerical experiments do not prove the conjecture, and the recursive algorithm based on Lemma 2.6 is a useful byproduct. A notable strength is that the results contain no fitted parameters; the constants come from Weyl's law, Kroger's bound, and the geometric definition of packing density.","major_comments":[{"comment":"The proof hinges on the unproved variational assertion that the largest eigenvalue smaller or equal to lambda_k^*(R) of each connected component must equal lambda_k^*(R). The one-sentence perturbation argument is not a complete proof: a quantitative check is needed to show that shrinking a component whose relevant eigenvalue is strictly below lambda_k^*(R) and expanding the others strictly lowers the k-th eigenvalue, including when eigenvalues coincide or when components have different ranks. This lemma is subsequently used in Theorem 1.3, Theorem 3.4, Proposition 3.5, Theorem 1.4, and Algorithm 5.2, so the gap is load-bearing and should be closed with a fully expanded argument.","section":"Section 2, Lemma 2.6 (pp. 13-15)"},{"comment":"The proofs require a stronger decomposition property than Lemma 2.6 states: not only is each connected component of an optimizer an optimizer for its own rank, but any subcollection of components is an optimizer for the sum of the ranks. For instance, the equality lambda_{j'}(Xi_k)^{d/2} = lambda_{j'}^*(R)^{d/2} in the proof of Theorem 1.4 (1) implies (2), and the statement in Lemma 3.8 that Omega_k^{*(n_j)} realises lambda_{n_j k}^*(R), both use exactly this stronger property. Please state and prove the stronger decomposition property or give a separate argument for it.","section":"Section 3, Theorem 1.4 (pp. 20-22) and Lemma 3.8 (p. 24)"},{"comment":"In the decomposition j_k = n_k j + r with 0 <= r < j, the case r = 0 is allowed, but the construction (17) then involves lambda_r(Omega), which is undefined for the Dirichlet Laplacian because the numbering starts at k = 1. When j divides j_k, the first term in (17) should be omitted or handled separately. As written, the proof is incomplete along any subsequence for which r = 0.","section":"Section 3, Theorem 1.4, proof of (3) implies (1) (p. 22)"}],"minor_comments":[{"comment":"The proof has inconsistent notation: the cardinality bound should be n^d rather than n (since |nV| = n^d for |V| = 1), and the scaling map should be n_i^{-1/d} R^i V rather than n_i^{-1/d} n V, with density rho_i = n_i / R^{id}. As written, the construction of the asymptotic packing is not clear.","section":"Section 4, Proposition 4.2 (p. 27)"},{"comment":"The sentence \"Since Omega satisfies the two-term Weyl law (1)\" should refer to the fundamental domain V, consistent with the theorem statement; as written it is a typo.","section":"Section 4, Theorem 1.11 proof (p. 29)"},{"comment":"The justification of the last inequality in display (8) is confusingly worded: the point is that a_p >= b_{k-p}, so both terms are bounded by a_p; the phrase involving 'max{a_p, b_{k-p}}' is not the needed argument.","section":"Section 3, Lemma 3.2 proof (p. 17)"},{"comment":"The pseudocode never sets ranks[k] to {k} when no improvement is found, because the branch 'if minrank == k' is inside the 'then' block where minrank has just been set to j. The initialisation of ranks[k] should be moved before the loop or the no-improvement case handled after the loop.","section":"Section 5, Algorithm 5.2 (p. 31)"},{"comment":"The text says 'In all four cases' although three generators are studied (disk, square, and 1:5 rectangle); also 'logarithimc' and 'converhing' are typos.","section":"Section 1.5 (pp. 7-8)"},{"comment":"The sentence 'Summing up these identities for j from 1 to p-1' uses the wrong index (it should be q), and there is a stray comma in 'Omega_{n,q}'.","section":"Section 2, Lemma 2.6 proof (p. 14)"},{"comment":"The statement that 'f_0,...,f_{k-1}, phi in H^1(V) generates a k-dimensional subspace' counts k+1 functions; the dimension should be k+1, or the variational argument should be phrased directly via the orthogonal complement of the first k eigenfunctions.","section":"Section 2, Lemma 2.7 proof (p. 16)"},{"comment":"The manuscript contains numerous typographical errors ('satisy', 'adressed', 'reunion', 'strenghtening', 'a forth bullet point') that should be corrected in a final version.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"I recommend major revision. The central idea is sound and the results would be a solid contribution to the spectral optimization literature, but the proof of Lemma 2.6 and the strengthened subcollection property used in Theorem 1.4 and Lemma 3.8 need to be written out rigorously. The edge case r=0 in the proof of Theorem 1.4 is also easily fixable. I see no concerns about novelty, citation practice, or fit with the journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nRead this one. The core idea is good: restrict the Colbois–El Soufi machinery to families R(Ω) generated by one domain, where extremizers exist, and read Pólya's conjecture off optimizer behavior. The trichotomy (Thm 1.3), the r_{1,k}->1 equivalence under two-term Weyl (Thm 1.4), and the packing-density lower bound (Thm 1.8) are genuinely new and the paper earns its keep. It also recovers Pólya's tiling result and improves Urakawa's bound in a natural way. No fitted parameters, no circularity.\n\nThe soft spots are real but localized. Lemma 2.6 is the load-bearing wall: it claims every optimizer splits into scaled optimizers for smaller ranks with an additive eigenvalue identity. The proof's key step is one sentence—'the largest eigenvalue smaller or equal to λ*_k of each component has to be equal to λ*_k'—with no quantitative check. The stress-test reader is right that this needs a rigorous minimal-volume replacement argument. It also matters that Theorem 1.4 later needs the same property for arbitrary subcollections of components, not just individual components; Lemma 2.6 as stated doesn't give that. I think the property is true, but it is not proved, and both main theorems rest on it.\n\nThere are smaller presentation gaps: Theorem 1.4's proof writes jk = nk j + r with 0 ≤ r < j, then uses λ_r(Ω); when r = 0 that eigenvalue is undefined. That is an edge case, not a fatal flaw. The computational section has no shipped code and the figures are light, but the numerics are not the load-bearing part.\n\nWould I bring this to reading group? Maybe; the idea is worth knowing. Would I cite it? Only after the Lemma 2.6 gap is closed; right now the proofs are not dependable enough to cite as theorems. But this is exactly the kind of paper that should go to peer review rather than desk rejection: important, structurally original, probably correct, and in need of a careful revision. Send it out; ask the referee to pin down Lemma 2.6.","headline":"Genuinely new structural results on Pólya's conjecture, but the main theorems rest on one underproved variational lemma that a referee should pin down before publication.","tokens_in":26913,"tokens_out":2896,"would_cite":false,"duration_ms":28933,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35P15","35P20","49R05"],"pacs":[],"model":"deepseek-v4-flash","headline":"Scaled copies of a domain decide Pólya's conjecture.","keywords":["Pólya conjecture","Laplace eigenvalues","spectral optimization","Dirichlet eigenvalues","Neumann eigenvalues","packing density","two-term Weyl law","Wolf-Keller decomposition"],"falsifier":"For a bounded domain $\\Omega$ whose generator is optimal infinitely often, check the $k$-th Dirichlet eigenvalue of a volume-one union of its scaled copies: if any such eigenvalue is smaller than $(2\\pi)^2(\\omega_d|\\Omega|)^{-2/d}k^{2/d}$, then Pólya's conjecture fails in $R(\\Omega)$ despite the generator recursing infinitely often, which contradicts Theorem 1.3.","tokens_in":25923,"feed_emoji":"🧩","tokens_out":9381,"duration_ms":78791,"temperature":0.7,"pith_summary":"This paper studies what happens when a single bounded domain $\\Omega$ is used as the only building block: the family $R(\\Omega)$ of all disjoint unions of scaled copies of $\\Omega$ with fixed total volume. The authors prove that the sequence of optimal $k$-th Dirichlet (or Neumann) eigenvalues within $R(\\Omega)$ carries enough information to decide whether Pólya's conjecture holds for every domain in that family. A trichotomy (Theorem 1.3) classifies all possibilities: either the generator $\\Omega$ itself is an optimal union infinitely often, in which case Pólya's conjecture holds; or it is optimal only finitely often and the optimal value hits the Weyl constant infinitely often (conjecture holds); or it is optimal only finitely often and the optimal value attains its infimum infinitely often, which is the case where the conjecture fails. Under the two-term Weyl law, the strong form of Pólya's conjecture becomes equivalent to the largest scaled copy in the optimal union converging to the full domain size. This yields new proofs of Pólya's tiling theorem and a strengthening of the packing-density bound of Urakawa.","feed_headline":"Scaled copies of a domain decide Pólya's conjecture","feed_subtitle":"A trichotomy links the geometry of optimal eigenvalue unions to the conjecture's truth.","key_machinery":"The load-bearing object is the Wolf-Keller decomposition (Lemma 2.6): every optimiser $\\Omega^*_k$ in $R(\\Omega)$ splits into scaled copies of optimisers for smaller ranks, $\\Omega^*_k = \\bigsqcup_q \\alpha_q \\Omega^*_{j_q}$, with $\\lambda^*_k(R)^{d/2} = \\sum_q \\lambda^*_{j_q}(R)^{d/2}$. This additive identity is what makes the sequence $\\lambda^*_k^{d/2}$ subadditive, justifies the recursive dynamic-programming search over partitions used in the numerical section, and underlies the propagation arguments that connect recurrence of the generator to the infimum $L$. The second mechanism is the packing-density estimate of Section 4, which turns the fact that an optimal union can be geometrically embedded into another domain into a lower bound on $L$ in terms of the asymptotic packing density $\\rho_\\Omega$.","core_discovery":"The central discovery is that the geometric behaviour of the optimisers encodes the validity of Pólya's conjecture. For a fixed generator $\\Omega$, let $\\lambda^*_k(R)$ be the best possible $k$-th Dirichlet eigenvalue among volume-one unions of scaled copies of $\\Omega$. The sequence $\\lambda^*_k(R)^{d/2}$ is subadditive, so by Fekete's lemma it converges to an infimum $L$; Pólya's conjecture in $R(\\Omega)$ holds exactly when $L$ equals the Weyl constant $(2\\pi)^d/\\omega_d$. The trichotomy sharpens this: if the generator itself recurs as an optimiser infinitely often, the conjecture holds; if it recurs only finitely often, the conjecture holds precisely when the optimal value equals the Weyl constant infinitely often, and fails precisely when the optimal value attains its infimum infinitely often. In the presence of the two-term Weyl law, the strong Pólya conjecture is equivalent to the largest scale factor $r_{1,k}$ of the optimiser tending to $1$ as $k$ grows, so the geometry of the optimal union at high frequency is a complete proxy for the conjecture.","pith_inferences":["If the observed log-density of ranks where the generator is an optimiser converges to a constant greater than $0.8$ for all tested domains, the trichotomy suggests this constant is a new spectral invariant of $\\Omega$ that may distinguish domains for which Pólya's conjecture is open (like the disk) from tiling domains (like the square), and this can be tested on other shapes such as ellipses or tr","The equivalence of strong Pólya with $r_{1,k} \\to 1$ turns a spectral inequality into a computable geometric condition; checking whether $r_{1,k}$ approaches $1$ for a candidate domain at very high $k$ provides a numerical test that could flag a counterexample long before a proof is found.","The packing density $\\rho_\\Omega$ could be refined to a spectral packing density that measures how efficiently copies of $\\Omega$ can be embedded while preserving eigenvalue ordering, which might connect to questions in discrete geometry about densest packings of non-convex tiles."],"forward_implications":["If $\\Omega$ tiles $\\mathbb{R}^d$, Pólya's conjecture holds for every union of its scaled copies (Corollary 1.9).","If $\\Omega$ simply tiles $\\mathbb{R}^d$ and its fundamental domain obeys the two-term Weyl law, then $\\Omega$ satisfies the strong Pólya conjecture and is itself an optimiser infinitely often (Theorem 1.11).","The universal lower bound $\\inf_k \\lambda^*_k(R)^{d/2}/k \\geq \\rho_\\Omega (2\\pi)^d/\\omega_d$ holds for every domain, strengthening Urakawa's lattice-packing bound in terms of a more flexible packing density (Theorem 1.8).","Under the two-term Weyl law, the strong Pólya conjecture in $R(\\Omega)$ is equivalent to the largest scaled-copy coefficient $r_{1,k}$ converging to $1$ along every subsequence (Theorem 1.4).","For the disk, square, and $1:5$ rectangle, numerical optimisers up to rank $66{,}000$ have at most five connected components, consistent with the conjecture (Section 5)."],"supporting_citations":[{"why":"Establishes subadditivity and superadditivity for extremal eigenvalues over all bounded domains, the framework this paper restricts to unions of scaled copies.","marker":"[14]"},{"why":"Supplies the decomposition of optimal eigenvalue domains into scaled copies of lower-rank optimisers, adapted here as Lemma 2.6 and used by the numerical algorithm.","marker":"[34]"},{"why":"Provides the two-term Weyl asymptotics that drive the equivalence in Theorem 1.4.","marker":"[31]"},{"why":"Original proof of Pólya's inequalities for tiling domains, recovered here as Corollary 1.9.","marker":"[30]"},{"why":"Introduces lattice packing density lower bounds for Dirichlet eigenvalues, strengthened here by the asymptotic packing density.","marker":"[32]"},{"why":"Gives the Neumann eigenvalue upper bound used to establish linear boundedness and the superadditive limit.","marker":"[24]"},{"why":"Proves the Neumann version of the Wolf-Keller decomposition, adapted as Lemma 2.7.","marker":"[29]"}],"fun_headline_variants":["Optimal scaled-domain unions encode Pólya's conjecture","Pólya's conjecture read from optimal domain unions","Trichotomy of optimizer behavior ties to Pólya's conjecture","How scaled copies of a domain probe Pólya's conjecture","Eigenvalue optimizers: a proxy for Pólya's conjecture"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole structure leans on the claim that in an optimal union every component must carry an eigenvalue exactly equal to the union's optimal value, a variational step the paper states briefly rather than proves in full.","fun_headline_variants_meta":{"raw":{"variants":["Optimal scaled-domain unions encode Pólya's conjecture","Pólya's conjecture read from optimal domain unions","Trichotomy of optimizer behavior ties to Pólya's conjecture","How scaled copies of a domain probe Pólya's conjecture","Eigenvalue optimizers: a proxy for Pólya's conjecture"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000927,"raw_usage":{"total_tokens":3972,"prompt_tokens":949,"completion_tokens":3023,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":565,"completion_tokens_details":{"reasoning_tokens":2937}},"tokens_in":565,"tokens_out":3023,"duration_ms":20546,"temperature":1.0,"reasoning_tokens":2937,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:41:36.849457+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a bounded domain $\\Omega$ whose generator is optimal infinitely often, check the $k$-th Dirichlet eigenvalue of a volume-one union of its scaled copies: if any such eigenvalue is smaller than $(2\\pi)^2(\\omega_d|\\Omega|)^{-2/d}k^{2/d}$, then Pólya's conjecture fails in $R(\\Omega)$ despite the generator recursing infinitely often, which contradicts Theorem 1.3.","supporting_citations":[{"cited_title":"Colbois and A","cited_arxiv_id":null,"evidence_quote":"Establishes subadditivity and superadditivity for extremal eigenvalues over all bounded domains, the framework this paper restricts to unions of scaled copies."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the decomposition of optimal eigenvalue domains into scaled copies of lower-rank optimisers, adapted here as Lemma 2.6 and used by the numerical algorithm."},{"cited_title":"Safarov & D","cited_arxiv_id":null,"evidence_quote":"Provides the two-term Weyl asymptotics that drive the equivalence in Theorem 1.4."},{"cited_title":"1961 On the eigenvalues of vibrating membranes,Proc","cited_arxiv_id":null,"evidence_quote":"Original proof of Pólya's inequalities for tiling domains, recovered here as Corollary 1.9."},{"cited_title":"Urakawa, Lower bounds for the eigenvalues of the fixed vibrating membrane problems,Tôhoku Math","cited_arxiv_id":null,"evidence_quote":"Introduces lattice packing density lower bounds for Dirichlet eigenvalues, strengthened here by the asymptotic packing density."},{"cited_title":"Kröger, Upper bounds for the Neumann eigenvalues on a bounded domain in Euclidean space,J","cited_arxiv_id":null,"evidence_quote":"Gives the Neumann eigenvalue upper bound used to establish linear boundedness and the superadditive limit."},{"cited_title":"Poliquin and G","cited_arxiv_id":null,"evidence_quote":"Proves the Neumann version of the Wolf-Keller decomposition, adapted as Lemma 2.7."}],"review_version":1}