{"id":"0229dd96-06d7-4666-b7dc-0d0dea840057","arxiv_id":"2608.13052","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For any measurable fundamental domain K of a lattice Lambda, the flat torus spectral gap obeys lambda_SG(T_Lambda) >= pi^2/(3||Cov_K||_op), with sharp constants and an equivalence under a sectional tiling condition.","lead":"This note proves sharp inequalities linking the covariance of any fundamental domain of a lattice to the spectral gap and Cheeger constant of the associated flat torus. It also shows that under a sectional tiling condition the bound becomes an equivalence, and that isotropic Voronoi cells give dimension-free lower bounds.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified.","rationale":"The reader's CONDITIONAL verdict is appropriate: the main theorems are proven and correct, but the note contains an unverified remark and an imprecise abstract phrase. I do not see a load-bearing flaw in the central claim itself. The reader's weakest_assumption (the sectional tiling hypothesis in Theorem 1.5) is a real restriction on the upper-bound side, but it is stated precisely and the theorem is proved as stated. My only slight disagreement is with the abstract's wording 'equivalence up to numerical constants,' which should read 'up to constants depending on delta' unless delta is assumed bounded below. This is a wording fix, not a mathematical error, so the verdict remains unchanged.","tokens_in":8067,"tokens_out":28707,"duration_ms":280550,"concrete_test":"Recompute the directional variance in Lemma 1.1 for a non-axis-aligned fundamental domain in Z^2, e.g. the parallelogram generated by (1,0) and (N,1), and verify that <Cov_K e_2,e_2> = 1/12 and <Cov_K e_1,e_1> >= 1/12; if either fails, the universal bound is unsound.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central lower-bound claim, Lemma 1.1 and Eq. (1.2), is correct: for any measurable fundamental domain K, the character x -> <xi,x> mod 1 pushes the uniform distribution on K forward to the uniform distribution on R/Z, so the one-dimensional variance in every nonzero dual direction is at least 1/12. The isoperimetric lower bound (Theorem 1.2) also checks out: the BV translation estimate and the averaging identity yield Per(A) >= 2v(1-v)/R1(K), and the Cheeger--Buser relations convert this into the stated spectral consequences. Theorem 1.5 is internally consistent: the sectional tiling hypothesis gives f_Y(0)=|xi|, and Hensley's one-dimensional log-concave bound supplies the upper estimate, with the delta-dependence explicitly stated. The only issues I find are presentation-level. Remark 4.2 contains an unverified, ChatGPT-attributed claim with an undefined symbol p, and the abstract's phrase 'equivalence up to numerical constants' is stronger than the delta-dependent upper bound 1/(2 delta) in Theorem 1.5; for the standard cube delta=1/n, so the ratio of upper to lower bound grows with dimension. Neither issue affects the validity of the central lower bounds.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This note studies the flat torus T_Λ=R^n/Λ and relates spectral and isoperimetric parameters to the covariance structure of a fundamental domain K. Lemma 1.1 shows that for every measurable fundamental domain K with finite second moment and every nonzero dual vector ξ, ⟨Cov_K ξ,ξ⟩ ≥ 1/12; consequently λ_SG(T_Λ)=4π²λ_1(Λ*)² ≥ π²/(3‖Cov_K‖_op). Theorem 1.2 (after Hadwiger) proves I_{T_Λ}(v) ≥ 2v(1−v)/R_1(K), hence D_Che(T_Λ) ≥ 1/R_1(K). These bounds yield Corollary 1.3: for unimodular lattices with isotropic Voronoi cell, λ_1(Λ*), λ_SG(T_Λ), and D_Che(T_Λ) are bounded below by a universal constant, using the Klartag–Lehec resolution of the slicing problem. Corollary 1.4 shows that no dimension-free converse of (0.1) holds for general Voronoi cells, and Example 4.1 gives a family of isotropic Voronoi cells with unbounded spectral gap. Theorem 1.5 proves a converse under a sectional tiling hypothesis, giving ‖Cov_K‖_op λ_1(Λ*)² ≤ 1/(2δ). The paper explicitly marks Remark 4.2 as an unverified ChatGPT-attributed claim.","tokens_in":8261,"tokens_out":25056,"duration_ms":231444,"significance":"The main lower-bound argument is elementary, elegant, and checkable: Lemma 1.1 uses the uniform mod-1 distribution and gives the sharp parameter-free constant 1/12, and Theorem 1.2 is proved by a complete BV translation estimate and averaging argument. The connection between the KLS conjecture and flat tori via the Klartag–Lehec slicing theorem is a clean and appealing observation. The central lower-bound theorems and the conditional upper bound in Theorem 1.5 are sound. The paper is also transparent about provenance and limitations. I found no load-bearing error in the main proofs; the issues are local notational and presentational, plus an unverified remark that should be removed or rigorously supported.","major_comments":[],"minor_comments":[{"comment":"The notational convention \"put n = 2m\" is inconsistent with the displayed equality λ_1(BW_m)^2 = 2^{(m−1)/2} = √(n/2): for n = 2m, √(n/2) = √m, which is not equal to 2^{(m−1)/2} for m ≥ 3 (e.g., m=3 gives 2 versus √3). The standard Barnes–Wall lattice BW_m has dimension 2^m, and with n = 2^m the displayed equality is correct. Please correct \"n=2m\" to \"n=2^m\" (and similarly the phrase \"representation on R^{2m}\"), or define the lattice family precisely with a reference.","section":"Example 4.1, §4"},{"comment":"This remark contains an unverified claim attributed to ChatGPT, uses an undefined symbol p, and does not specify how p depends on n. The assertion that this is \"essentially the maximal growth rate\" is not proved in the text. Since the remark is not used in any proof and Example 4.1 already supplies the necessary counterexample, I recommend deleting it or replacing it with a rigorously verified statement or a precise citation.","section":"Remark 4.2, §4"},{"comment":"The abstract's phrase \"equivalence (up to numerical constants)\" is stronger than what Theorem 1.5 establishes: the upper bound carries the factor 1/(2δ), and when δ is not a universal constant (e.g., δ=1/n for the standard cube) the multiplicative gap between the upper and lower bounds grows with dimension. Please rephrase as \"up to constants depending on δ\" or explicitly assume δ is a universal constant.","section":"Abstract and §1.5"},{"comment":"The text ascribes to [19] the rate c log(1+n)^{-1/2}, but the title of [19] states O(log^{1/4} n). Please reconcile the quoted rate with the cited source.","section":"§1.3, Eq. (1.6) and Ref. [19]"}],"recommendation":"minor_revision","confidential_remarks":"The AI Declaration and the unverified ChatGPT-attributed Remark 4.2 are unusual for this venue. Independent of mathematical correctness, the editor may wish to require that ChatGPT attributions be removed, that the unverified claim be either verified or deleted, and that the Barnes–Wall example be corrected notationally before publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a clean, honest note that gets the main things right. Lemma 1.1 is a genuinely nice observation—the uniform mod-1 argument giving the sharp 1/12 bound is airtight, and the spectral gap consequence is immediate. The isoperimetric half is also in good shape: the BV proof of Hadwiger's bound extends it to all lattices, and the application to the flat-torus KLS analogue via Klartag–Lehec is legitimate, not forced. Theorem 1.5's sectional tiling criterion is new and the proof is checkable; the explicit delta-dependence is a plus.\n\nSoft spots, in proportion. Remark 4.2 is a real problem: it is ChatGPT-attributed, explicitly unverified, and contains an undefined symbol p. It should be proved, removed, or replaced with a citation. The abstract's phrase \"equivalence up to numerical constants\" overstates Theorem 1.5, since the upper bound is 1/(2δ) and δ can be as small as 1/n for the standard cube, so the ratio grows with dimension. That is a presentation-level overreach, not a flaw in the lower bound, but it should be rephrased. The AI Declaration is unusually transparent, but it leaves reproducibility questions only for Remark 4.2 and the counterexample discussion—the main theorems are proven in the text.\n\nI checked the Barnes–Wall example and the Autissier–Magazinov/Siegel counterexample: both look correct. Corollary 1.4 is fine. No load-bearing flaw anywhere. The reader's strongest claim holds up on reading.\n\nThis paper is for lattice theorists, spectral geometers, and people in high-dimensional probability who want sharp, explicitly stated constants. It deserves a serious referee and, after a minor revision that cleans up Remark 4.2 and the abstract, it should be accepted.","headline":"A short, honest note that gets the main inequalities right: the 1/12 variance bound is sharp and clean, the isoperimetric extension is solid, and only the unverified ChatGPT remark and an overstrong abstract phrase need attention.","tokens_in":8843,"tokens_out":1419,"would_cite":true,"duration_ms":15163,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["58J50","35P15","52C07"],"pacs":[],"model":"deepseek-v4-flash","headline":"A tile's variance floor controls the spectral gap and isoperimetric constant of any flat torus.","keywords":["flat torus","spectral gap","Cheeger constant","covariance of fundamental domain","Voronoi cell","KLS conjecture","slicing problem","lattice geometry"],"falsifier":"Compute $\\langle\\operatorname{Cov}_K\\xi,\\xi\\rangle$ for any measurable fundamental domain $K$ of $\\mathbb{Z}^n$ and a coordinate direction $\\xi$; if the value drops below $1/12$, Lemma 1.1 fails, while the paper predicts this never happens and that the unit cube attains equality. For Theorem 1.5, search for a convex fundamental domain satisfying the sectional tiling hypothesis for which $\\|\\operatorname{Cov}_K\\|_{\\mathrm{op}}\\lambda_1(\\Lambda^*)^2$ exceeds $1/(2\\delta)$; a numerical search over low-dimensional lattice polytopes would settle whether the stated constants are sharp.","tokens_in":7827,"feed_emoji":"📐","tokens_out":15757,"duration_ms":137094,"temperature":0.7,"pith_summary":"This paper shows that the geometry of a flat torus is controlled from below by the covariance of any measurable tile out of which the torus is built. For every lattice fundamental domain $K$ and every nonzero dual vector $\\xi$, the variance of $\\langle x,\\xi\\rangle$ over $K$ is at least $1/12$; consequently the spectral gap satisfies $\\lambda_{\\mathrm{SG}}(\\mathbb{T}_\\Lambda)=4\\pi^2\\lambda_1(\\Lambda^*)^2\\ge \\pi^2/(3\\|\\operatorname{Cov}_K\\|_{\\mathrm{op}})$, and the isoperimetric (Cheeger) constant obeys $D_{\\mathrm{Che}}(\\mathbb{T}_\\Lambda)\\ge 1/R_1(K)$. When the tile is convex and its covariance is proportional to the identity (isotropic), the recent resolution of the slicing problem yields universal positive lower bounds on the spectral gap, the Cheeger constant, and the shortest nonzero dual vector, independent of dimension -- a flat-torus analogue of the KLS conjecture. The paper also shows that the reverse inequality fails in general, even when the tile is the central Voronoi cell (the set of points no farther from a given lattice point than from any other) and that cell is isotropic, and that a structural sectional tiling condition restores equivalence up to constants.","feed_headline":"1/12: the variance floor behind flat-torus spectral gaps","feed_subtitle":"For any lattice tile, the shortest dual vector and Cheeger constant obey sharp dimension-free lower bounds.","key_machinery":"The load-bearing object is the covariance matrix $\\operatorname{Cov}_K$ of a measurable tile $K$ of the lattice, together with its directional variances and operator norm. The spectral bound rests on the fact that every nontrivial character of the flat torus pushes Haar measure forward to the uniform measure on $\\mathbb{R}/\\mathbb{Z}$, giving each dual direction a variance floor of $1/12$. The isoperimetric bound uses a directional translation estimate for sets of finite perimeter: the measure of the symmetric difference of a set and its translate by $z$ is bounded by the integral of $|\\langle z,\\nu_A\\rangle|$ over the reduced boundary; averaging this over a fundamental domain and applying Fubini's theorem yields the isoperimetric lower bound in terms of $R_1(K)$, the maximal expected absolute projection of a uniform point of $K$ onto a unit vector. The upper-bound half of Theorem 1.5 uses the slice density of a convex body along a unit direction $\\nu$; this density is log-concave by Brunn-Minkowski, the sectional tiling hypothesis identifies its value at zero with $|\\xi|$ for the primitive dual vector $\\xi$ normal to the tiling hyperplane, and the one-dimensional log-concave estimate $f(0)\\le 1/(\\sqrt{2}\\sigma)$ forces the variance $\\sigma^2=\\langle\\operatorname{Cov}_K\\nu,\\nu\\rangle$ to be at most $1/(2|\\xi|^2)$. Finally, the dimension-free lower bounds pass through the recently proved bound on isotropic constants of convex bodies.","core_discovery":"The paper's central discovery is a sharp variance estimate for dual directions. Let $\\Lambda\\subset\\mathbb{R}^n$ be a full-rank lattice, $K$ a measurable fundamental domain of finite second moment, and $\\xi\\in\\Lambda^*\\setminus\\{0\\}$. The normalized Lebesgue measure on $K$ pushes forward under the character $x\\mapsto \\langle x,\\xi\\rangle \\bmod 1$ to Haar measure on $\\mathbb{R}/\\mathbb{Z}$, so the variance $\\langle\\operatorname{Cov}_K\\xi,\\xi\\rangle$ is at least the variance of the uniform distribution on $[-1/2,1/2]$, namely $1/12$. Since $\\lambda_{\\mathrm{SG}}(\\mathbb{T}_\\Lambda)=4\\pi^2\\lambda_1(\\Lambda^*)^2$, this yields $\\lambda_{\\mathrm{SG}}(\\mathbb{T}_\\Lambda)\\ge \\pi^2/(3\\|\\operatorname{Cov}_K\\|_{\\mathrm{op}})$. The same tiling structure, through a directional translation estimate for sets of finite perimeter, gives $D_{\\mathrm{Che}}(\\mathbb{T}_\\Lambda)\\ge 1/R_1(K)$. Combining these with the resolved slicing bound on isotropic constants gives a universal $c>0$ such that every unimodular lattice with an isotropic convex fundamental domain (in particular an isotropic Voronoi cell) satisfies $D_{\\mathrm{Che}}(\\mathbb{T}_\\Lambda),\\lambda_{\\mathrm{SG}}(\\mathbb{T}_\\Lambda),\\lambda_1(\\Lambda^*)\\ge c$. There exist lattices with isotropic Voronoi cells for which no dimension-free upper bound holds; under the sectional tiling hypothesis, however, $\\|\\operatorname{Cov}_K\\|_{\\mathrm{op}}\\lambda_1(\\Lambda^*)^2$ is confined to the interval $[1/12,1/(2\\delta)]$.","pith_inferences":["Editorial inference: the variance inequality behaves like an uncertainty principle for lattices -- no measurable tile can place less than $1/12$ variance on any dual direction; a natural test is whether an analogous floor holds for tilings by non-convex tiles in other periodic geometries.","Editorial inference: the sectional tiling condition is exactly the missing ingredient that upgrades a one-sided covariance bound to an equivalence; it would be worth investigating how common such tiling sections are among natural families of fundamental domains such as zonotopes and alcoved polytopes.","Editorial inference: the universal constant $c$ in Corollary 1.3 is not made explicit; tracking the constants through the slicing bound and the log-concave density estimates would produce a concrete numerical value, which the paper does not provide.","Editorial inference: the existence of isotropic Voronoi cells with spectral gap growing like $\\sqrt{n}$ suggests an extremal question -- maximize $\\lambda_{\\mathrm{SG}}(\\mathbb{T}_\\Lambda)$ among unimodular lattices with isotropic Voronoi cell; the Hermite-constant bound places this maximum between order $\\sqrt{n}$ and order $n$."],"forward_implications":["From the covariance of any measurable tile one obtains a dimension-free lower bound on the spectral gap and Cheeger constant of the corresponding flat torus, so bounding a tile's covariance certifies a spectral gap.","For every unimodular lattice whose Voronoi cell is isotropic, the torus satisfies the KLS-type conclusion: $D_{\\mathrm{Che}}(\\mathbb{T}_\\Lambda)$, $\\lambda_{\\mathrm{SG}}(\\mathbb{T}_\\Lambda)$, and $\\lambda_1(\\Lambda^*)$ are all bounded below by a universal constant independent of dimension.","Isotropy alone does not give a matching upper bound: there are lattices with isotropic Voronoi cells whose torus spectral gap grows like a positive power of the dimension.","Under the sectional tiling hypothesis, $\\|\\operatorname{Cov}_K\\|_{\\mathrm{op}}\\lambda_1(\\Lambda^*)^2$ lies between $1/12$ and $1/(2\\delta)$, so the torus spectral gap and the Neumann spectral gap of the convex tile are comparable up to a $\\sqrt{\\log(1+n)}$ factor.","For convex tiles satisfying the sectional tiling condition, covariance bounds and spectral bounds are interchangeable up to constants."],"supporting_citations":[{"why":"identifies the eigenfunctions of the flat-torus Laplacian and hence the formula $\\lambda_{\\mathrm{SG}}=4\\pi^2\\lambda_1(\\Lambda^*)^2$.","marker":"[10]"},{"why":"supplies the original isoperimetric bound for lattice-periodic sets in terms of $R_1(K)$, which becomes Theorem 1.2.","marker":"[13]"},{"why":"proves the dimension-free upper bound on isotropic constants of convex bodies that makes the KLS-type lower bounds universal.","marker":"[18]"},{"why":"gives the lower-bound half of the perimeter-spectral comparison used in (1.3).","marker":"[7]"},{"why":"gives the upper-bound half of the perimeter-spectral comparison under non-negative Ricci curvature used in (1.3).","marker":"[9]"},{"why":"yields the one-dimensional log-concave density estimate at zero used in the sectional-tiling upper bound.","marker":"[14]"},{"why":"supplies the mean-value estimate on covering radius needed to construct counterexamples to a dimension-free converse.","marker":"[27]"},{"why":"provides the sharp lower bound on the second moment of a lattice Voronoi cell used in those counterexamples.","marker":"[20]"}],"fun_headline_variants":["Sharp variance 1/12 bounds flat-torus spectra","Flat tori: universal spectral gap from 1/12 variance","Variance floor yields dimension-free torus bounds","Isotropic lattices: KLS holds for flat tori"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"All dimension-free converse statements rest on the sectional tiling hypothesis: a convex fundamental domain with barycenter at the origin must have a linear hyperplane section that tiles that hyperplane by lattice translations and whose normal-direction variance is a fixed fraction of the maximal variance; without this hypothesis, the paper shows the matching upper bound can fail.","fun_headline_variants_meta":{"raw":{"variants":["Sharp variance 1/12 bounds flat-torus spectra","Flat tori: universal spectral gap from 1/12 variance","Variance floor yields dimension-free torus bounds","Isotropic lattices: KLS holds for flat tori"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000342,"raw_usage":{"total_tokens":2090,"prompt_tokens":1361,"completion_tokens":729,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":977,"completion_tokens_details":{"reasoning_tokens":661}},"tokens_in":977,"tokens_out":729,"duration_ms":7187,"temperature":1.0,"reasoning_tokens":661,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T17:48:31.832964+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute $\\langle\\operatorname{Cov}_K\\xi,\\xi\\rangle$ for any measurable fundamental domain $K$ of $\\mathbb{Z}^n$ and a coordinate direction $\\xi$; if the value drops below $1/12$, Lemma 1.1 fails, while the paper predicts this never happens and that the unit cube attains equality. For Theorem 1.5, search for a convex fundamental domain satisfying the sectional tiling hypothesis for which $\\|\\operatorname{Cov}_K\\|_{\\mathrm{op}}\\lambda_1(\\Lambda^*)^2$ exceeds $1/(2\\delta)$; a numerical search over low-dimensional lattice polytopes would settle whether the stated constants are sharp.","supporting_citations":[{"cited_title":"Fuglede, Commuting self-adjoint partial differential operators and a group theoretic problem,J","cited_arxiv_id":null,"evidence_quote":"identifies the eigenfunctions of the flat-torus Laplacian and hence the formula $\\lambda_{\\mathrm{SG}}=4\\pi^2\\lambda_1(\\Lambda^*)^2$."},{"cited_title":"Hadwiger,Gitterperiodische Punktmengen und Isoperimetrie, Monatsh","cited_arxiv_id":null,"evidence_quote":"supplies the original isoperimetric bound for lattice-periodic sets in terms of $R_1(K)$, which becomes Theorem 1.2."},{"cited_title":"Klartag and J","cited_arxiv_id":null,"evidence_quote":"proves the dimension-free upper bound on isotropic constants of convex bodies that makes the KLS-type lower bounds universal."},{"cited_title":"Cheeger,A lower bound for the smallest eigenvalue of the Laplacian, in: Problems in Analysis (Papers dedicated to Salomon Bochner), Princeton Univ","cited_arxiv_id":null,"evidence_quote":"gives the lower-bound half of the perimeter-spectral comparison used in (1.3)."},{"cited_title":"De Ponti and A","cited_arxiv_id":null,"evidence_quote":"gives the upper-bound half of the perimeter-spectral comparison under non-negative Ricci curvature used in (1.3)."},{"cited_title":"Hensley,Slicing convex bodies—bounds for slice area in terms of the body’s covariance, Proc","cited_arxiv_id":null,"evidence_quote":"yields the one-dimensional log-concave density estimate at zero used in the sectional-tiling upper bound."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the mean-value estimate on covering radius needed to construct counterexamples to a dimension-free converse."},{"cited_title":"Magazinov,A proof of a conjecture by Haviv, Lyubashevsky and Regev on the second moment of a lattice Voronoi cell, Adv","cited_arxiv_id":null,"evidence_quote":"provides the sharp lower bound on the second moment of a lattice Voronoi cell used in those counterexamples."}],"review_version":1}