{"id":"336a980c-0e3c-4f22-8d2e-f065a86d774e","arxiv_id":"2607.11695","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Rotated E8 yields a strict lattice packing density gain for the product of two 4-balls over D4×D4, and a one-parameter deformation of the densest 9-dimensional lattice lowers its spectral height.","lead":"This paper gives an exact counterexample in dimension 8 showing that packing densities of Cartesian products of convex bodies need not equal the product of the densities, and a numerical counterexample in dimension 9 showing that the densest lattice need not minimize spectral height of unit-volume flat tori. These settle long-open questions of Cassels, Zong, Sarnak and Chiu in the geometry of numbers.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection beyond the already-flagged floating-point gap for the height derivative.","rationale":"The paper's strongest geometric claim—the strict failure of the lattice, translative and unrestricted product formulae for B₄^{2}—is fully rigorous and elementary. The only provisional claim is the height counterexample for Λ₉. The reader correctly identified that this claim is supported only to floating-point accuracy and that the authors themselves decline interval certification. No deeper inconsistency, hidden assumption, or circularity is present. Consequently the CONDITIONAL verdict already records the precise remaining gap; no adjustment is warranted.","tokens_in":9725,"tokens_out":461,"duration_ms":4053,"concrete_test":"Re-run the height derivative and the finite-difference comparison of Section 3.4 with interval arithmetic (or mpmath/Arb at 50+ digits) and a cutoff large enough that the tail of both Epstein-type series is smaller than 10^{-8}; if the interval for dh/dε still lies entirely below −0.1 and the interval for h(G_{10^{-3}})-h(G_0) lies entirely below −10^{-5}, the numerical counterexample becomes rigorous and the remaining certification gap closes.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The reader's weakest_assumption already isolates the only soft spot that actually bears on a central claim: the sign of dh/dε|ε=0 and the inequality h(G_{10^{-3}})<h(G_0) rest on ordinary double-precision summation of the spectral zeta derivative over lattice/dual vectors with finite cut-offs (Sections 3.3–3.4, Remark 2). The lattice-product counterexample (Theorems 9–12, Corollary 11) is exact once the matrix H with H^{2}=5I and the root-projection identity rᵀHr=±2 are verified; both are elementary and machine-checkable. No further load-bearing gap appears: the Cohn–de Laat–Salmon upper bound is used only as a published numerical certificate, and the uniqueness of Λ₉ is taken from a cited preprint whose correctness is independent of the height calculation.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper gives two counterexamples in the geometry of numbers. For the product problem of Cassels (critical determinants) and Zong (lattice/translative packing densities), it exhibits a rotated copy of E8 that packs B_2^4 \times B_2^4 more densely than the product of the D4 packing densities, yielding the strict inequality δ_L(B\times B) ≥ (π^4/1024)(1+1/√5)^4 > π^4/256 = δ_L(B)^2 (and the analogous critical-determinant form); the same lower bound plus the Cohn–de Laat–Salmon upper bound on δ(B_2^4) also falsifies the unrestricted and translative product formulae. Independently, for the Sarnak/Chiu height-minimization conjecture on unit-volume flat tori, it constructs an explicit determinant-1 path G_ε through Gram matrices of the Korkine–Zolotarev lattice Λ_9 (the unique densest lattice in dimension 9) along which numerical evaluation of the spectral zeta derivative shows a negative derivative at ε=0 and a strict drop in height at ε=10^{-3}.","tokens_in":9962,"tokens_out":1304,"duration_ms":28554,"significance":"The results settle long-standing questions of Cassels and Zong in the negative already for four-dimensional balls, and supply the first evidence that densest lattices need not minimize height (already in dimension 9). The lattice-product counterexample is fully rigorous and elementary once the algebraic identities H^{2}=5I and rᵀHr=±2 on E8 roots are verified; the paper supplies machine-checkable ancillary scripts for the latter and an explicit rotation. The height calculation is reproducible via the same ancillaries and an explicit one-parameter family. These strengths (exact identities, open verification code, clean separation of the two independent constructions) make the work a solid contribution to packing and spectral geometry of lattices, even though the second claim remains numerical.","major_comments":[{"comment":"Sections 3.3–3.4 and Remark 2: the second central claim (that Λ_9 is not a local height minimizer, so the Sarnak/Chiu conjecture fails) rests entirely on ordinary double-precision summation of the spectral-zeta derivative over lattice and dual vectors with finite cut-offs. The reported values dh/dε|ε=0 ≈ -0.2035 and h(G_{10^{-3}})-h(G_0)≈-2.03\times10^{-4} are large enough that the sign is almost certainly correct, yet the paper explicitly declines interval-arithmetic certification. Because this is load-bearing for one of the two advertised counterexamples, either supply rigorous error bounds (interval arithmetic, explicit tail estimates, or multiprecision with a priori remainder control) or rephrase the result strictly as strong numerical evidence rather than a definitive counterexample.","section":"§3.3–3.4, Remark 2"},{"comment":"Theorem 15 and the paragraph preceding (11): uniqueness (up to isometry) of the densest unit-covolume lattice in dimension 9 is taken from the preprint [10]. The height calculation itself does not need uniqueness, but the logical step “the unique densest lattice is not a local height minimizer ⇒ the conjecture fails” does. If [10] remains unpublished, the claim should be weakened to “this particular densest lattice is not a local height minimizer” (still sufficient to cast serious doubt on the conjecture, but not a complete disproof).","section":"§3.2, Theorem 15"}],"minor_comments":[{"comment":"The numerical height evaluation (cut-offs, number of vectors retained, convergence checks) is only sketched; a short paragraph or table in the ancillary documentation listing the precise truncation radii used for the reported digits would improve reproducibility.","section":"§3.3"},{"comment":"In the unrestricted comparison of §2.5 the elementary bounds π>3.14 and √5<25/11 already suffice; the more precise Cohn table value can be relegated to a remark so that the argument is self-contained without external floating-point data.","section":"§2.5"},{"comment":"Typographical: “dimen-sions” (Acknowledgements), and the future date “July 13, 2026” on the title page should be checked against the arXiv stamp.","section":"Acknowledgements, title page"},{"comment":"The matrix M defining the perturbation A is presented without motivation; a one-sentence remark that it was chosen so that the first-order change in λ_1 of the dual is positive while that of the primal is negative would help the reader.","section":"§3.3"}],"recommendation":"major_revision","confidential_remarks":"Both counterexamples were discovered with substantial AI assistance (disclosed in the Acknowledgements). The first is clean and independent of that origin; the second inherits the usual reproducibility caveats of floating-point spectral sums. The journal may wish to decide whether a purely numerical disproof of a conjecture, even with open code, meets its standards for a “counterexample” or should be labelled “numerical evidence”. Fit for math.MG is excellent."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The paper settles Cassels’ critical-determinant question and Zong’s lattice/translative product questions with an exact construction: a suitable rotation of E8 packs B4×B4 at density (π4/1024)(1+1/√5)4, which beats the D4⊕D4 product value π4/256 by about 9.7 %. The same lower bound plus the published Cohn–de Laat–Salmon upper bound on δ(B4) also kills the unrestricted product formula. That part is elementary once the matrix H with H2=5I and the root-projection identity rTHr=±2 are checked; both are machine-verifiable and the ancillary scripts do it.\n\nThe second half is a one-parameter volume-preserving deformation of the Gram matrix of Λ9 (the unique densest unit-covolume lattice in dimension 9, per the recent Dutour Sikirić–van Woerden classification). Ordinary double-precision summation of the spectral-zeta derivative gives a clearly negative derivative at ε=0 and a strict drop at ε=10-3. The authors flag that they did not certify with interval arithmetic, so the height claim remains numerical evidence rather than a theorem.\n\nWhat is new is the explicit rotation that makes the product inequality strict and the concrete trace-zero path that moves height while slightly worsening packing density. Both start from classical lattices and standard functionals; there is no circularity. The writing is direct, the citations are appropriate, and the code is supplied.\n\nSoft spot is exactly the one already noted: the height numbers rest on floating-point cut-offs. Everything else is solid. This is for people who care about packing densities, critical determinants, or heights of flat tori. It deserves a serious referee; the product counterexample alone is publishable, and the height calculation is reproducible enough to be useful even if it stays numerical. I would send it out.","headline":"Exact rotated-E8 counterexample kills the Cassels–Zong product formulas; the height claim on Λ9 is only floating-point but cleanly presented.","tokens_in":10555,"tokens_out":491,"would_cite":true,"duration_ms":9019,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["11H06","52C17","11M41"],"pacs":[],"model":"grok-4.5","headline":"Two product and height optimization principles for lattices fail in dimensions 8 and 9.","keywords":["geometry of numbers","critical determinant","lattice packing","Cartesian product","flat tori","spectral height","E8 lattice","Korkine-Zolotareff lattice"],"falsifier":"Recompute the spectral height of the given one-parameter family of nine-dimensional Gram matrices with rigorous interval arithmetic or arbitrary-precision cut-offs; if the derivative at zero is non-negative or the height at ε=10^{-3} is not smaller, the second counterexample collapses.","tokens_in":10594,"feed_emoji":"□","tokens_out":732,"duration_ms":6586,"temperature":0.7,"pith_summary":"The paper shows that two long-standing optimization principles for lattices do not hold in every dimension. First, the densest packing of a product of two convex bodies need not equal the product of their densest packings: a carefully rotated copy of the E8 lattice packs two four-dimensional balls more densely than the product of the best four-dimensional lattice packings, and the same construction also beats the product of the best unrestricted four-dimensional sphere-packing densities. Second, among nine-dimensional flat tori of unit volume, the lattice that maximises the shortest nonzero vector is not even a local minimiser of the spectral height; an explicit one-parameter deformation lowers the height while preserving volume. The first counterexample is exact; the second is numerical but reproducible to ordinary floating-point precision. Together they demonstrate that product formulae and packing-height coincidences that hold in low dimensions can fail once the dimension reaches eight or nine.","feed_headline":"Lattice product formulas fail in dimensions 8 and 9","feed_subtitle":"Rotated E8 beats the product of two D4 packings; densest 9D lattice is not height-minimal","key_machinery":"A concrete orthogonal projection of E8 onto a pair of complementary four-planes defined by a symmetric integer matrix H with H^{2}=5I; after rotation the projected roots realise a block-norm minimum larger than that of D4⊕D4, producing the improved product packing density.","core_discovery":"For the Euclidean unit ball B in four dimensions the lattice packing density of the product body B\times B is at least (π^{4}/1024)(1+1/√5)^{4}, which is strictly larger than the product of the individual lattice packing densities π^{4}/256. The same lower bound, combined with a known rigorous upper bound on the unrestricted packing density of B, also yields a strict inequality for unrestricted congruent packings. Independently, an explicit determinant-preserving path through Gram matrices of the densest nine-dimensional lattice produces a negative derivative of height at the origin and a strictly smaller height value at a nearby parameter, showing that this densest lattice is not a local hei","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["Product packing equality fails for balls in dim 8","Rotated E8 beats D4 product packing density","Densest 9D lattice is not height-minimal","Cassels-Zong product formula fails in dim 8","Sarnak height conjecture fails for densest 9D lattice"],"cache_read_input_tokens":128,"weakest_assumption_plain":"The height calculation relies on ordinary floating-point summation of the spectral zeta derivative over finitely many lattice and dual vectors, without interval-arithmetic certification of the sign.","fun_headline_variants_meta":{"raw":{"variants":["Product packing equality fails for balls in dim 8","Rotated E8 beats D4 product packing density","Densest 9D lattice is not height-minimal","Cassels-Zong product formula fails in dim 8","Sarnak height conjecture fails for densest 9D lattice"]},"model":"grok-4.5","effort":"low","cost_usd":0.003428,"raw_usage":{"total_tokens":1122,"prompt_tokens":730,"num_sources_used":0,"completion_tokens":64,"cost_in_usd_ticks":34280000,"prompt_tokens_details":{"text_tokens":730,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":328,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":730,"tokens_out":64,"duration_ms":3195,"temperature":1.0,"reasoning_tokens":328,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-14T03:48:18.379666+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Recompute the spectral height of the given one-parameter family of nine-dimensional Gram matrices with rigorous interval arithmetic or arbitrary-precision cut-offs; if the derivative at zero is non-negative or the height at ε=10^{-3} is not smaller, the second counterexample collapses.","supporting_citations":[],"review_version":1}