{"id":"b3e49694-07ee-42c9-b51f-86b85af4bc37","arxiv_id":"2607.20051","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The Duval–Reiner majorization conjecture fails for every r≥5, while the r=2 inequality holds universally with an explicit equality classification.","lead":"A 2002 conjecture saying that the sums of the largest eigenvalues of a hypergraph's Laplacian are always bounded by degree data is false: the paper exhibits exact counterexamples at every index r≥5. It also proves the conjecture's statement for the second partial sum in all uniformities and classifies when equality holds.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified; the counterexample construction is internally consistent and verifiable.","rationale":"The reader's weakest assumption was the eigenvector identity (3), and I agree that this is the most load-bearing computational input in the counterexample construction. But the identity is not a hidden or circular step: it is an explicit finite integer-arithmetic claim that the paper supports by giving the full matrix and vector. I spot-checked the relevant rows and the arithmetic is consistent. The surrounding machinery - Lemma 3.4's block decomposition, Lemma 3.5's spectral reflection, Lemma 3.6's defect transfer, and the exact root bound for d* - all cohere. The only residual risk is a transcription error in the printed matrices, which is a verification issue rather than a substantive mathematical gap. I therefore do not see a reason to change the reader's ACCEPT verdict; a symbolic recomputation would raise confidence from MODERATE to HIGH.","tokens_in":19150,"tokens_out":13038,"duration_ms":122424,"concrete_test":"In a computer algebra system, enter the two 15x15 and 16x16 Gram matrices exactly as printed in Appendix A, compute det(xI - M_F15) and det(xI - M_F16) symbolically, and verify the factorizations in Lemma 3.1. Also compute M_F15 * y_eta - 7*y_eta with y_eta = (1,0,0,0,0,1,0,0,0,0,1,1,1,0,0)^T and confirm it is the zero vector. If these identities hold, the transfer construction and hence Theorem 1.2 stand.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The single most fragile point in the central claim is the exact eigenvector identity (3), M_F15 y_eta = 7 y_eta, together with the two characteristic-polynomial factorizations in Lemma 3.1. If (3) were wrong, ridge-whiskering would not preserve the defect and the universal counterexamples for r >= 7 would collapse. However, the appendix prints the full 15x15 matrix M_F15 and the vector y_eta, and the identity is directly checkable: the rows corresponding to facets containing {2,3} (indices 1,6,11,12,13) each give inner product 7 with y_eta, and all other rows give 0. The characteristic polynomials are also explicitly stated with enough matrix data for independent exact verification. The transfer identity Lemma 3.6 is derived cleanly from the spectral reflection M_F* = S(nI - M_F)S and the degree-counting identities, and the root-isolation proof of d* > 0 is exact, not numerical. I find no internal inconsistency or unsupported load-bearing assumption beyond the usual need to independently recheck the displayed arithmetic.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper disproves the majorization assertion of the Duval–Reiner conjecture by constructing explicit q-uniform families F with s_r(F) > D_r(F) for every r ≥ 5 (in some uniformity) and for every q ≥ 3 (at some index r ≥ 5). The construction uses two 3-uniform seeds F16 and F15, explicit characteristic polynomials, root isolation giving the positive defect d* = θ4+θ5−12, and two defect-preserving operations: ridge-whiskering (Lemma 3.4) and set-complement duality (Lemma 3.6). In parallel, the paper proves the universal inequality s_2(F) ≤ D_2(F) for all q-uniform families and classifies equality through core completion, a block-matrix trace bound, and Ky Fan variational arguments. The appendix provides the full 15×15 and 16×16 Gram matrices for direct verification of the characteristic polynomials.","tokens_in":19372,"tokens_out":3233,"duration_ms":28000,"significance":"If the counterexample construction is correct, it settles a long-standing conjecture in the negative, resolving the majorization question for all but indices 3 and 4. The paper also provides the first unrestricted-uniformity proof of the second partial-sum inequality, with a complete equality classification. A notable strength is the machine-checkable exactness: the counterexamples rely on explicit integer matrices, exact characteristic polynomial factorizations, an exact root-isolation proof of the tiny positive defect d*, and fully stated transfer lemmas. This is a significant contribution to spectral hypergraph theory and combinatorial Laplacian majorization. The r=2 theorem (Theorem 1.3) is substantive and appears correct, with the equality classification being a strong addition. The paper is honest about the quantifier structure and does not overclaim, and the remaining open cases s_3 and s_4 are clearly identified.","major_comments":[{"comment":"The counterexample claim for all r ≥ 5 depends on exact verification that the five largest eigenvalues of M_F15 at indices 8 and 9 are as stated. The list '7, 7−θ1, 6, 6, 7−θ2, 4, 4, 7−θ3' requires checking the relative ordering of the explicit eigenvalues; this follows from the root bounds θ1∈(0,1), θ2∈(2,3), θ3∈(4,5), θ4∈(5,6), θ5∈(6,7) and is correct, but the ordering verification is compressed. Since the entire counterexample family rests on this list, a short explicit display of the ordering would improve rigor.","section":"§3, Proposition 3.3 and root ordering"},{"comment":"The two seeds were 'identified through an exhaustive computer search on seven vertices,' but the search is not described and no search certificate is provided. The displayed matrices make the construction independently verifiable, so this is not a correctness issue, but it is a reproducibility gap: an interested reader cannot know whether the search was complete or how the seeds are special. A short paragraph describing the search space and the exact criterion would remove this gap.","section":"§3, search provenance of F15 and F16"},{"comment":"The equality case of s_2(F) = D_2(F) uses several compressed steps: the conclusion that t≥3 contradicts equality, the derivation that s_2(hat F) = 2m + p_1 only when t≤2, the identification of the two-dimensional maximizing subspace supported on F, and the passage from dim V_K ≥ 2 to condition (a). These steps are correct given Lemma 4.9 and the orthogonal decomposition in Lemma 4.6, but they are dense. Expanding the argument around the sentence 'the same orthogonal decomposition shows that all remaining eigenvalues ...' would materially improve the paper.","section":"§4, proof of equality classification in Theorem 1.3"},{"comment":"The transfer identity Δ_{f−r}(F*) = Δ_r(F) is stated for 1 ≤ r < f, and Theorem 1.2 uses it with r = 8 and r = 9 for F15 and F15(t). The argument is correct, including the degree-counting identity. The proof is clear and self-contained; no issue beyond the notation 'F*' which should be consistently typeset as F^\\star.","section":"§3, Lemma 3.6 transfer identity"}],"minor_comments":[{"comment":"The phrase 'while t ridge whiskers preserve its defect at r = 8' should read 'while t ridge-whisker facets preserve' or 'while t ridge whiskers preserve' with a hyphen. Also 'Both lie on {0, 1, ..., 6}' uses set notation without braces; prefer 'on the vertex set {0,1,...,6}'.","section":"Introduction, §1"},{"comment":"The vectors u_i and v_i are not normalized but are used as eigenvectors; the spectral decomposition formula is correct, but the word 'orthogonal eigenvectors' could be clarified as 'mutually orthogonal eigenvectors' since they are not unit vectors.","section":"§2, Lemma 2.5"},{"comment":"The sentence 'Bareiss fraction-free elimination ... entirely in Z[x]' is correct. In the appendix, 'the pivots are nonzero because the leading principal minors of xI−M_F are monic polynomials' is terse but valid; adding that the leading principal minors are monic of degree i would help.","section":"§3, Lemma 3.1"},{"comment":"The displayed matrices are legible, but the facet labels are only given in the surrounding text; placing the facet labels above the columns of the Gram matrices would ease direct verification.","section":"Appendix A"},{"comment":"The arXiv labels [13] and [19] are 'v1' versions that postdate the submission date of this paper. Since this is a future-dated timeline, this is only a minor housekeeping issue; the version numbers should be updated when available.","section":"References"}],"recommendation":"accept","confidential_remarks":"I found the counterexample construction and the r=2 theorem technically sound and verifiable from the displayed matrices. The exactness of the defect d* is a strong point: it is a root-location consequence, not a fitted parameter. The only reservations are the lack of detail on the exhaustive search that produced the seeds (reproducibility) and some compressed steps in the equality classification. Neither affects correctness. This paper is a clear fit for a top combinatorics journal."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper disproves the majorization assertion of Duval–Reiner for every r≥5 and every q≥3, and proves the r=2 inequality with full equality classification. That is a real result, not a tweak. The counterexample machinery is honest: two 3-uniform seeds, exact 15x15 and 16x16 Gram matrices in the appendix, characteristic polynomials obtained by exact integer arithmetic, and a ridge-whiskering/complement transfer lemma that propagates the defect. The key input, eigenvector identity (3) for the ridge {2,3}, is load-bearing but directly checkable from the printed matrix; I did not find a sign error. The positivity of d* is proved by root isolation, not numerics, and the paper is careful about quantifier order. No fitted parameters, no circular dependence on the conjecture.\n\nThe r=2 part is also well done. Core completion, Ky Fan, and a compression argument reduce to the complete-simplex boundary matrix; the equality classification is structural and plausible, with cases (i)-(iii) simple and case (iv) encoding the missing-core rank condition. This is the kind of theorem that will be cited.\n\nSoft spots, in order of size. First, the seed search itself is black-boxed: \"exhaustive computer search on seven vertices\" with no algorithm, search space, or code. That is not a flaw in the proof, since the appendix gives full verification, but it does limit the reader's ability to generalize or trust that no smaller seed was missed. Second, several displayed matrix identities and root intervals are asserted after \"exact multiplication\" or \"direct substitution\" without a reproducible script. Fine for a paper, but a referee with time on their hands should recheck them; I spot-checked the easy ones. Third, the introduction's claim that no complete majorization result was known in any unrestricted q≥3 is moderately strong, but it is consistent with the cited literature.\n\nOverall: the central claim is sound as far as I can tell, and the proof format—explicit matrices and exact polynomial identities—is the right level of verifiability. The paper deserves serious peer review. I would send it to a competent referee with instructions to check the appendix arithmetic and the transfer lemma, not to revisit the overall strategy.","headline":"A genuine disproof of the majorization assertion of Duval–Reiner, with explicit exact counterexamples and a solid universal r=2 inequality; deserves careful peer review.","tokens_in":19856,"tokens_out":2544,"would_cite":true,"duration_ms":23137,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05E45","05C65","05C50","15A42"],"pacs":[],"model":"deepseek-v4-flash","headline":"The sum of the r largest Laplacian eigenvalues can exceed the degree-based bound for every r≥5.","keywords":["simplicial Laplacian","spectral majorization","conjugate degree sequence","uniform hypergraphs","counterexamples","ridge-whiskering","set-complement duality","equality classification"],"falsifier":"Recompute, with exact integer arithmetic, the eigenvalues of the 16-facet seed and check that s_5(F16) > D_5(F16) (equivalently that the two largest roots of p_16 sum to more than 12), and independently verify the identity M_F15 y_η = 7y_η with ‖y_η‖²=5 for the ridge η={2,3}. A failure of either computation would invalidate the counterexample construction.","tokens_in":19018,"feed_emoji":"📐","tokens_out":6853,"duration_ms":61426,"temperature":0.7,"pith_summary":"The paper refutes a long-standing spectral majorization conjecture for higher-dimensional Laplacians. For every index r≥5 it constructs a q-uniform family of sets for which the sum of the r largest eigenvalues of the simplicial up-Laplacian strictly exceeds the r-th partial sum of the conjugate vertex-degree sequence, and it shows that every uniformity q≥3 admits such a violation at some r≥5. In the opposite direction, it proves the inequality is always true at r=2 and gives a complete classification of the families where equality holds. This leaves exactly two indices — r=3 and r=4 — as the only universal cases still open.","feed_headline":"Laplacian eigenvalue sums exceed degree bound for every r≥5","feed_subtitle":"The long-conjectured bound on higher Laplacian spectra fails from index 5 onward, leaving only r=3 and r=4 open.","key_machinery":"The counterexample construction is carried by two 3-uniform seed families, F15 and F16, whose Gram matrices and characteristic polynomials are computed exactly. The key engine is a ridge-whiskering operation: for a ridge η with signed incidence vector y_η satisfying M_F y_η = (d_F(η)+q−1) y_η, adjoining new facets of the form η∪{w_i} multiplies the characteristic polynomial by a known factor and preserves the spectral defect. Complementing the family in a larger ambient set transfers the defect at index r to index f−r. For the r=2 result, the paper completes the family to a full q-simplex on the core vertices and uses a variational maximum principle together with a trace bound for block matr","core_discovery":"The central claim is that the universal majorization bound s_r(F) ≤ D_r(F), which holds for graphs and was conjectured to hold for all q-uniform families, is false for r≥5. The proof produces, from two explicitly computed 3-uniform seed families on seven vertices, strict counterexamples at every index r≥5 via two operations: ridge-whiskering, which attaches fresh vertices along a distinguished ridge and shifts the spectrum, and set-complement duality, which transfers a spectral defect at index r to the complementary index f−r. The paper also establishes the second partial-sum inequality s_2(F) ≤ D_2(F) in all uniformities, with equality classified into four structural cases involving pendant","pith_inferences":["The existence of two complementary quantifier directions suggests checking whether r=3 and r=4 can also be broken by similar seed-transfer constructions; if so, the entire conjecture collapses at every index.","The margin of violation at r=5 is small (about 1.96×10^-4 in the seed example), hinting that quantitative refinements of the conjectured bound, perhaps with a correction depending on the family's structure, may still be true.","The rank condition on the missing-core matrix in the r=2 equality classification offers a general way to measure how close a family is to saturating the bound, potentially leading to stronger two-eigenvalue estimates.","The exact eigenvector identity that powers ridge-whiskering is delicate; exploring whether analogous identities hold for other seeds could either produce counterexamples at r=3 and r=4 or explain why those indices are special."],"forward_implications":["For every r≥5 the majorization inequality fails somewhere, so the original conjecture's main assertion is disproved outright.","The only universal inequalities that could still hold are at r=3 and r=4; these are now the sole open cases.","For every uniformity q≥3, some index r≥5 sees a violation, showing the failure is not confined to any single dimension.","The inequality at r=2 is true in all uniformities and equality is fully characterized by four explicit structural conditions.","Counterexamples can be chosen with empty intersection across all facets, so they cannot be dismissed as trivial cones over lower-dimensional examples."],"fun_headline_variants":["Simplicial Laplacian bound disproved for all r≥5","Duval-Reiner majorization fails from index 5 upward","Hypergraph Laplacian sum exceeds degree sum for some r≥5","s_2≤D_2 holds, but s_r≤D_r dies for r≥5"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The systematic counterexamples for all r≥7 and all q≥4 depend on a single exact eigenvector identity for one 15-facet seed family; if that identity fails — even though it is verified by exact arithmetic — the transfer construction no longer works.","fun_headline_variants_meta":{"raw":{"variants":["Simplicial Laplacian bound disproved for all r≥5","Duval-Reiner majorization fails from index 5 upward","Hypergraph Laplacian sum exceeds degree sum for some r≥5","s_2≤D_2 holds, but s_r≤D_r dies for r≥5"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00036,"raw_usage":{"total_tokens":1826,"prompt_tokens":829,"completion_tokens":997,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":573,"completion_tokens_details":{"reasoning_tokens":914}},"tokens_in":573,"tokens_out":997,"duration_ms":10033,"temperature":1.0,"reasoning_tokens":914,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T10:55:06.323126+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute, with exact integer arithmetic, the eigenvalues of the 16-facet seed and check that s_5(F16) > D_5(F16) (equivalently that the two largest roots of p_16 sum to more than 12), and independently verify the identity M_F15 y_η = 7y_η with ‖y_η‖²=5 for the ridge η={2,3}. A failure of either computation would invalidate the counterexample construction.","supporting_citations":[],"review_version":1}