{"id":"b9c069d7-112e-4ba3-863e-9c21c466a222","arxiv_id":"2608.10241","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For any isotropic convex body in R^n, the subgaussian body has bounded volume ratio against the centroid body, sharp mean width O(√log n), and an orthonormal basis with subgaussian constants O(√log n).","lead":"New bounds show that the body encoding subgaussian directions of a convex high-dimensional body has the same volume as its variance ellipsoid, up to a constant independent of dimension. This sharpens the recently solved Milman problem and gives full orthonormal bases of directions with small subgaussian constants for every isotropic convex body.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The volume-ratio theorem's dimension-free constant rests on Lemma 2.2, whose e^{-C_1 p} Gaussian-mass bound for A_p is imported from [20, Prop. 2.4] and inherits unverified small-ball/negative-moment estimates; any dimension-dependent constant there would degrade Theorem 1.1 to only vrad ≤ C√n.","rationale":"The reader correctly identified Lemma 2.2 and its dependence on Bizeul's small-ball estimate and the negative-moment equivalence as the weakest assumption. My reading confirms that this is the load-bearing point: the entire volume-ratio argument hangs on the product of Gaussian masses ∏ e^{-C_1 2^k} = e^{-C n}, which requires an absolute constant in Lemma 2.2. The paper does not prove Lemma 2.2, and the cited sources [5], [6], and [12, Thm 6.9] are preprints or a self-authored survey, so the central theorem is conditional on external verification. No internal inconsistency, circularity, or data manipulation was found; the remaining arguments of the paper are coherent and the constants check out. Because the claimed dimension-free form of Theorems 1.1, 1.3 and 1.2 would collapse if Lemma 2.2's constant were dimension-dependent, acceptance should be conditional on an independent verification of Lemma 2.2/Proposition 2.1 with absolute constants.","tokens_in":18363,"tokens_out":47156,"duration_ms":455785,"concrete_test":"Re-derive Lemma 2.2 from Proposition 2.1 without invoking [20, Prop. 2.4]: starting from w_{-p}(Z_p) ≈ √p L_K and the small-ball estimate (2.1), compute γ_n(C0√(np)L_K Z_p(K)^∘) for p up to 2c_0 n and verify that the resulting lower bound is e^{-C_1 p} with C_1 absolute. Test the passage on the isotropic cross-polytope (or product-exponential measures), where Z_n and A_n can be computed explicitly; if for p=n the exponent in γ_n(A_n) is c n ln n rather than C n, then Lemma 2.2 fails in the range p ∼ n and the Gaussian-correlation product in Theorem 1.1 does not yield the required e^{-C n} volume lower bound.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The proof of Theorem 1.1 reduces to the lower bound γ_n(A) ≥ e^{-C_2 n} for the intersection A = ∩_{k=0}^{k0} A_{2^k}. By Gaussian correlation, γ_n(A) ≥ ∏ γ_n(A_{2^k}), and each factor is supplied by Lemma 2.2 as e^{-C_1 2^k}. Lemma 2.2 is not proved in the paper; it is attributed to Letwin–Mikulincer [20, Prop. 2.4], which in turn relies on Proposition 2.1 (w_{-p}(Z_p) ≈ √p L_K). Proposition 2.1 depends on Bizeul's small-ball estimate (2.1) and on the negative-moment equivalence I_{-(n-1)}(µ) ≈ I_2(µ), cited to [11] and to the self-authored survey [12, Thm 6.9]. The constant C_1 must be absolute, and the range must include p up to c_0 n. If either source carries a dimension-dependent constant—for example, an n^β factor in the small-ball exponent or an equivalence ratio 1+o(1) that degrades with n—then the product over dyadic p gives only γ_n(A) ≥ e^{-ω(n)}. The subsequent volume step gives vol(A) ≥ (2π)^{n/2} e^{-ω(n)}, and Blaschke–Santaló then yields vrad(Ψ2(K)) ≤ C√n (or worse) instead of an absolute bound. That would destroy Theorems 1.1, 1.3 and 1.2 in their dimension-free form. Since the paper supplies no derivation of Lemma 2.2 and the cited sources are very recent preprints, this is the single most load-bearing unverified point in the central claim.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the subgaussian body Ψ2(K) of a centered convex body K⊂R^n, whose support function is the ψ2-norm of linear functionals on K. Its central result, Theorem 1.1, asserts that the volume ratio (vol_n(Ψ2(K))/vol_n(Z2(K)))^{1/n} is bounded by an absolute constant, where Z2(K) is the L2-centroid body. The proof passes to an isotropic affine image, represents Ψ2(K) as a polytope-type convex hull of dyadic L_p-centroid bodies, takes polars, and uses the Gaussian correlation inequality together with a lower bound γ_n(A_p)≥e^{-C_1 p} for the sets A_p=C_0√(np)L_K Z_p(K)° (Lemma 2.2, quoted from Letwin–Mikulincer [20]) to obtain a Gaussian-mass lower bound for the intersection. From this the authors derive volume estimates, prove projection bounds for isotropic bodies, construct orthonormal bases with subgaussian constants C√(n/(n−k+1)), prove sharp mean-width bounds, and give an explicit uniformly subgaussian basis for unconditional isotropic bodies via Bobkov–Nazarov and a Fourier basis construction.","tokens_in":18764,"tokens_out":14759,"duration_ms":137290,"significance":"If correct, the results are substantial: Theorem 1.1 upgrades the existence of a single subgaussian direction, recently proved by Letwin and Mikulincer, to a dimension-free bound on the whole subgaussian body relative to Z2(K), and it yields clean consequences for projections, mean width, orthonormal bases, and the ℓ-position of Ψ2(K). The proof architecture is transparent: the affine-invariance reduction, the dyadic centroid-body approximation, and the Gaussian-correlation step are elegant, and the unconditional case gives an explicit basis with optimal behavior. The paper is also honest about its dependencies and does not appear circular: it uses [20]'s A_p sets as a tool rather than assuming the final subgaussian-basis conclusion. The main caveat, which is load-bearing, is that Lemma 2.2 is imported from a very recent preprint and is not proved in the manuscript.","major_comments":[{"comment":"Theorem 1.1 rests entirely on the lower bound γ_n(A_p)≥e^{-C_1 p} for 1≤p≤2c_0n, stated as Lemma 2.2. This lemma is not proved in the paper; it is quoted from [20, Prop. 2.4], which in turn depends on Proposition 2.1, Bizeul's small-ball estimate (2.1) from [5], and the negative-moment equivalence I_{-(n-1)}(µ)≈I_2(µ) cited to [11] and [12, Thm 6.9]. Since all of these are either very recent preprints or a self-cited survey, and since a dimension-dependent constant in any one of them would replace the product over dyadic p by e^{-ω(n)} and reduce Theorem 1.1 to vrad(Ψ2(K))≤C√n (also degrading Theorems 1.2 and 1.3), this is a load-bearing point. I ask the authors to include a complete proof of Lemma 2.2, or at least a precise derivation showing that the hypotheses of [20, Prop. 2.4] hold with absolute constants over the full range p≤2c_0n.","section":"§2, Lemma 2.2 and §3.1"},{"comment":"The advertised sharp mean-width estimate w(Ψ2(K))≤C√ln(en) depends on Bizeul's optimal M-estimate [6], a preprint, through Lemma 3.4, and also on the 'almost isotropic' comparison fν(0)^{1/m}≈Lν. Please state exactly which theorem from [6] is used and provide a derivation of the comparison, so the reader can verify that all constants are absolute. If [6] is not yet available in final form, an appendix containing the needed argument would resolve this dependency.","section":"§3.3, Theorem 3.6 and Lemma 3.4"}],"minor_comments":[{"comment":"The sentence 'Using the change of variables t=2s√(2ln(en))' reuses the symbol t for a new variable after t was already used as the threshold in Theorem 3.9; please rename one of the variables for clarity.","section":"§3.4, Proof of Proposition 3.11"},{"comment":"There is a typo in the phrase 'known facts abour Ball's bodies'; it should read 'about'.","section":"§3.3, Lemma 3.4"},{"comment":"The constant c_5 appears in equation (3.5) without being introduced; since the paper uses generic constants elsewhere, please clarify or use standard notation.","section":"§3.3, Eq. (3.5)"},{"comment":"In the Sudakov inequality statement, the constant is written 'where c>0' but the inequality as stated needs an absolute constant independent of n and K; please make this explicit.","section":"§4.1"},{"comment":"In the construction of the Fourier basis, the indexing for n even and n odd is correct, but the sentence 'If n is even, define in addition v_{n-1}=u_{n/2}' could be made clearer by explicitly stating that u_{n/2} is the real vector (1,-1,1,-1,...)/√n, which is already done in the following line.","section":"§3.5, Lemma 3.15"}],"recommendation":"major_revision","confidential_remarks":"The central theorem relies on Lemma 2.2, which is quoted from a very recent preprint [20] and ultimately from other preprints [5,6]. I would advise the editor to ask the authors to make this dependency verifiable by providing a self-contained proof of Lemma 2.2 or a detailed verification with explicit absolute constants. This is a verifiability issue, not a suspicion of circularity; the paper's own reasoning appears sound if the imported lemma holds."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's my take. This is a good, honest paper. The main result — volume ratio of the subgaussian body with respect to the L2-centroid body is bounded by an absolute constant — is a genuine improvement over the earlier C√log n bound, and the proof is short and transparent. I also like the projection estimate (n/m)^{1/2} and the mean-width bound O(√log n), which are sharp for the cross-polytope. The orthonormal basis results are new, and the explicit Fourier basis for unconditional bodies is a nice touch.\n\nThe paper reads well. The reduction to the isotropic case and the polar-intersection step are clean, and the use of Gaussian correlation is elegant. I did not find any internal gap in the arguments that are actually written.\n\nThe soft spot is exactly what the stress-test flags: Lemma 2.2 is the engine of Theorem 1.1, and it is not proved here. It is imported from Letwin–Mikulincer, which in turn rests on Bizeul's small-ball estimate and a negative-moment equivalence. Those inputs are from very recent preprints. If any of them carries a hidden dimension-dependent constant, the dyadic product would give e^{-ω(n)} mass, and Theorem 1.1 would collapse to a √n bound. The paper is transparent about the dependence, and the sources are credible, but the load-bearing constant C_1 in Lemma 2.2 deserves explicit scrutiny. I'd want a referee to verify that [20, Prop. 2.4] indeed yields an absolute C_1 for all p ≤ c_0 n.\n\nThis is not a flaw in the paper per se; it is the normal business of building on fast-moving work. Still, the authors could make the paper more self-contained by adding a short proof or a precise statement of Lemma 2.2 with the constants spelled out.\n\nBottom line: this deserves a serious referee. It is a strong contribution to a central topic, with new theorems and honest citations. I would send it to review, and I would cite the volume-ratio theorem in my own work.","headline":"A clean upgrade of subgaussian-body geometry from polylog to dimension-free bounds, but the central lemma is imported from a very recent preprint.","tokens_in":19390,"tokens_out":4167,"would_cite":true,"duration_ms":38501,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["52A40","46B06","52A23","60D05"],"pacs":[],"model":"deepseek-v4-flash","headline":"For every centered convex body, the subgaussian body \\(\\Psi_2(K)\\) has volume ratio bounded by an absolute constant with respect to the \\(L_2\\)-centroid body \\(Z_2(K)\\), a global strengthening of Milman's subgaussian-direction problem.","keywords":["hyperplane conjecture","log-concave measures","isotropic convex bodies","subgaussian directions","volume distribution in high dimensions","centroid bodies","psi-2 norm","Gaussian correlation inequality"],"falsifier":"Take a sequence of isotropic convex bodies, for example normalized cubes and cross-polytopes in dimensions \\(n=10,\\dots,1000\\), and compute or rigorously estimate \\(\\left(\\mathrm{vol}(\\Psi_2(K))/\\mathrm{vol}(Z_2(K))\\right)^{1/n}\\); if this ratio is unbounded in \\(n\\), Theorem 1.1 is false. On the input side, exhibit an isotropic log-concave measure and a point \\(y\\) with \\(\\mu\\{|x-y|_2\\le \\sqrt{\\varepsilon n}\\}>\\$varepsilon^{{c_0 n}}$\\) for some fixed small \\(\\varepsilon\\), which would falsify the small-ball inequality (2.1) that Lemma 2.2 depends on.","tokens_in":18142,"feed_emoji":"📐","tokens_out":8662,"duration_ms":80500,"temperature":0.7,"pith_summary":"Every centered convex body \\(K\\subset \\mathbb{R}^n\\) has a body \\(\\Psi_2(K)\\) that records the \\(\\psi_2\\)-norm of each linear functional, and the paper proves that its volume ratio against the \\(L_2\\)-centroid body \\(Z_2(K)\\) is bounded by an absolute constant independent of \\(n\\) and of \\(K\\). This global statement goes beyond the existence of one subgaussian direction, which was recently solved, and controls the whole collection of directions at once. In the isotropic case the same mechanism yields sharp estimates for the mean width and for the volume radii of orthogonal projections, and produces orthonormal bases whose vectors have small subgaussian constants. The engine is a representation of \\(\\Psi_2(K)\\) as a convex hull of rescaled centroid bodies, combined with Gaussian-measure lower bounds for their polars and the Gaussian correlation inequality.","feed_headline":"Every convex body has a bounded subgaussian body","feed_subtitle":"Its volume ratio to the L2-centroid body is an absolute constant, for all dimensions.","key_machinery":"The load-bearing object is the family of \\(L_p\\)-centroid bodies \\(Z_p(K)\\) together with the representation \\(\\Psi_2(K)\\approx \\mathrm{conv}\\{Z_p(K)/\\sqrt{p}:1\\le p\\le n\\}\\). Because \\(p\\mapsto \\|\\langle\\cdot,\\xi\\rangle\\|_p/\\sqrt{p}\\) stabilizes once \\(p\\approx n\\), only \\(p\\le n\\) matter, and centroid comparison \\(Z_{2p}\\approx Z_p\\) reduces \\(p\\) to dyadic powers. The second engine is Lemma 2.2: for \\(A_p=C_0\\sqrt{np}\\,L_K\\, Z_p(K)^\\circ\\), one has \\(\\gamma_n(A_p)\\ge $e^{{-C_1 p}}$\\), which follows from an optimal small-ball estimate for isotropic log-concave measures and negative-moment equivalence. The Gaussian correlation inequality combines these lower bounds across dyadic \\(p\\), giving \\(\\gamma_n(\\cap_k A_{2^k})\\ge $e^{{-C_2 n}}$\\), and Blaschke–Santalo converts this into the volume-ratio estimate. This exact mechanism turns the existence of one subgaussian direction into control of the whole body.","core_discovery":"The central discovery is Theorem 1.1: for every centered convex body \\(K\\subset \\mathbb{R}^n\\) normalized to volume one, \\(\\left(\\mathrm{vol}_n(\\Psi_2(K))/\\mathrm{vol}_n(Z_2(K))\\right)^{1/n}\\le C\\) with \\(C\\) an absolute constant. Since \\(\\Psi_2(K)\\) also contains a constant multiple of \\(Z_2(K)\\), this bounds the volume radius of the subgaussian body by an absolute constant. The proof reduces to the isotropic case by affine invariance, cuts \\(\\Psi_2(K)\\) into dyadic pieces via \\(\\Psi_2(K)\\approx \\mathrm{conv}\\{Z_{2^k}(K)/\\sqrt{2^k}\\}\\), and uses the Gaussian correlation inequality on the polar sets \\(A_{2^k}=C_0\\sqrt{n2^k}\\,L_K\\, Z_{2^k}(K)^\\circ\\). The same mechanism supplies an orthonormal basis with decaying subgaussian constants (Theorem 1.2), projection radius bounds (Theorem 1.3), mean width \\(\\le C\\sqrt{\\ln(en)}\\) (Theorem 1.4), a basis with all constants \\(\\le C\\sqrt{\\ln(en)}\\) (Theorem 1.5), and an explicit uniform basis in the unconditional case.","pith_inferences":["The dyadic Gaussian-correlation argument is insensitive to the particular shape of \\(K\\) beyond the small-ball estimate, so the same volume-ratio bound should transfer to every centered log-concave measure; Theorem 3.2 already takes a step in this direction and could likely be pushed further to control \\(\\Psi_2\\) of marginals and products.","The explicit Fourier-basis construction for unconditional bodies suggests a testable extension: for unconditional \\(K\\), the optimal subgaussian basis may be chosen from a single universal orthonormal system independent of \\(K\\), which could be explored numerically for \\(\\ell_p\\)-balls and Orlicz balls.","The projection bound is sharp for the normalized cross-polytope, so if a matching lower bound held beyond that example it would identify bodies with large subgaussian projections as being \\(\\ell_1\\)-like; the paper only establishes sharpness in that one class.","The proof uses the boundedness of the isotropic constant as an input, but the volume-ratio mechanism itself is driven by small-ball behavior; isolating how much of the dependence on the isotropic constant is removable would clarify whether the dimension-free constant can be made completely explicit."],"forward_implications":["Milman's problem follows again as a corollary: a volume-ratio bound of \\(C\\) forces a direction \\(v\\) with \\(h_{\\Psi_2(K)}(v)\\le C' h_{Z_2(K)}(v)\\), i.e. a uniformly subgaussian direction.","For any isotropic \\(K\\) and any \\(m\\)-dimensional subspace \\(F\\), \\(\\mathrm{vrad}(P_F(\\Psi_2(K)))\\le C\\sqrt{n/m}\\); in particular \\(\\Psi_2(K)\\subseteq C\\sqrt{n}\\,B_2^n\\), so all \\(\\psi_2\\)-norms of linear functionals are at most \\(O(\\sqrt{n})\\).","One can construct an orthonormal basis with \\(\\|\\langle\\cdot,v_k\\rangle\\|_{\\psi_2}\\le C\\sqrt{n/(n-k+1)}\\), meaning most vectors are uniformly subgaussian and only the last few directions deteriorate.","A random orthonormal basis is, with high probability, a complete basis of subgaussian directions with constants \\(\\le C\\sqrt{\\ln(en)}\\); for unconditional isotropic bodies an explicit Fourier-type basis achieves a uniform constant \\(C\\).","The mean width bound \\(w(\\Psi_2(K))\\le C\\sqrt{\\ln(en)}\\), combined with \\(M(\\Psi_2(K))\\le C\\), places the subgaussian body in the \\(\\ell\\)-position up to a logarithmic factor."],"supporting_citations":[{"why":"Introduced the \\(A_p\\) polar sets and the Gaussian-intersection strategy that the proofs of The main theorems inherit.","marker":"[20]"},{"why":"Supplies the optimal small-ball estimate (2.1) that yields the negative-moment equivalence behind Proposition 2.1 and Lemma 2.2.","marker":"[5]"},{"why":"Gives the optimal mean-width bound \\(w(K)\\le C\\sqrt{n\\ln(en)}\\) used for Theorem 1.4 and Lemma 3.4.","marker":"[6]"},{"why":"Proves the Gaussian correlation inequality used to multiply the \\(\\gamma_n(A_{2^k})\\) lower bounds.","marker":"[27]"},{"why":"Provides the projection identity \\(P_F(Z_p(\\mu))=Z_p(\\pi_F(\\mu))\\) and the centroid-body framework used in Theorem 1.3.","marker":"[24]"},{"why":"Establishes the \\(\\psi_2\\) upper bound for isotropic unconditional bodies used to obtain the explicit uniform basis in Corollary 3.16.","marker":"[7]"},{"why":"Provides the previous \\(\\sqrt{\\ln(en)}\\) volume-ratio bound that Theorem 1.1 strengthens, along with the representation (2.5).","marker":"[14]"},{"why":"Supplies the small-ball and negative-moment equivalence results used in the derivation of Proposition 2.1.","marker":"[11]"}],"fun_headline_variants":["Subgaussian body volume ratio to L2-centroid body is absolute constant","Bounded subgaussian body: volume ratio to centroid body is constant","Subgaussian body volume ratio: absolute constant, all dimensions","Isotropic convex bodies: subgaussian body has bounded volume ratio","Subgaussian body bounds: volume ratio to centroid body is absolute constant"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The argument collapses if the optimal small-ball estimate for isotropic log-concave measures fails to hold with a dimension-free exponent; any dimension dependence in the bound on the mass of small Euclidean balls around points would make the Gaussian-correlation step lose its dimension-free character.","fun_headline_variants_meta":{"raw":{"variants":["Subgaussian body volume ratio to L2-centroid body is absolute constant","Bounded subgaussian body: volume ratio to centroid body is constant","Subgaussian body volume ratio: absolute constant, all dimensions","Isotropic convex bodies: subgaussian body has bounded volume ratio","Subgaussian body bounds: volume ratio to centroid body is absolute constant"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000636,"raw_usage":{"total_tokens":2932,"prompt_tokens":948,"completion_tokens":1984,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":564,"completion_tokens_details":{"reasoning_tokens":1890}},"tokens_in":564,"tokens_out":1984,"duration_ms":14432,"temperature":1.0,"reasoning_tokens":1890,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T04:11:19.614675+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a sequence of isotropic convex bodies, for example normalized cubes and cross-polytopes in dimensions \\(n=10,\\dots,1000\\), and compute or rigorously estimate \\(\\left(\\mathrm{vol}(\\Psi_2(K))/\\mathrm{vol}(Z_2(K))\\right)^{1/n}\\); if this ratio is unbounded in \\(n\\), Theorem 1.1 is false. On the input side, exhibit an isotropic log-concave measure and a point \\(y\\) with \\(\\mu\\{|x-y|_2\\le \\sqrt{\\varepsilon n}\\}>\\$varepsilon^{{c_0 n}}$\\) for some fixed small \\(\\varepsilon\\), which would falsify the small-ball inequality (2.1) that Lemma 2.2 depends on.","supporting_citations":[{"cited_title":"Dimension-free Gaussian tail estimates for linear functionals on convex bodies","cited_arxiv_id":"2605.10939","evidence_quote":"Introduced the \\(A_p\\) polar sets and the Gaussian-intersection strategy that the proofs of The main theorems inherit."},{"cited_title":"Royen,A simple proof of the Gaussian correlation conjecture extended to some multivariate gamma distri- butions, Far East J","cited_arxiv_id":null,"evidence_quote":"Proves the Gaussian correlation inequality used to multiply the \\(\\gamma_n(A_{2^k})\\) lower bounds."},{"cited_title":"Paouris,Concentration of mass in convex bodies, Geom","cited_arxiv_id":null,"evidence_quote":"Provides the projection identity \\(P_F(Z_p(\\mu))=Z_p(\\pi_F(\\mu))\\) and the centroid-body framework used in Theorem 1.3."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes the \\(\\psi_2\\) upper bound for isotropic unconditional bodies used to obtain the explicit uniform basis in Corollary 3.16."},{"cited_title":"Giannopoulos, G","cited_arxiv_id":null,"evidence_quote":"Provides the previous \\(\\sqrt{\\ln(en)}\\) volume-ratio bound that Theorem 1.1 strengthens, along with the representation (2.5)."},{"cited_title":"Dafnis and G","cited_arxiv_id":null,"evidence_quote":"Supplies the small-ball and negative-moment equivalence results used in the derivation of Proposition 2.1."}],"review_version":1}