{"id":"9e93e0e5-9c84-487f-8c90-d1b7a8d4fca3","arxiv_id":"2506.08494","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A general FB semigroup theorem produces multi-function sharp versions of hypercontractivity, Hausdorff-Young, log-Sobolev, and noisy Borell inequalities with necessary and sufficient Gaussian covariance conditions.","lead":"This paper proves multi-function versions of major inequalities in Gaussian analysis, covering Hausdorff-Young, hypercontractivity, log-Sobolev, and noise stability. It gives a single general theorem with necessary and sufficient conditions in terms of covariance matrices.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 11 is false as printed: the §4 proof of log-Sobolev from Theorem 10 uses L instead of −L and garbles the φ′ factor, so the advertised FB corollary is not established.","rationale":"The reader's conditional verdict is reasonable, and my read does not move it. The FB core, especially Proposition 2 and Theorem 13, appears structurally sound: the local-to-global step applies Jensen to the convex function F''/F' y² exactly as stated, and the real forward case can be obtained from Proposition 1 by the implicit but unstated replacement F → −F, which flips both the convexity hypothesis and the local sign. Thus the reader's named weakest assumption, while under-documented, is not a fatal flaw. The serious defect is in Section 4: the Mehler derivative is computed with the wrong sign, and a φ′ factor is lost and then reinserted inconsistently. As a result Theorem 11 is false as printed, as the exponential test function shows. Since Theorem 11 is one of the paper's advertised main corollaries, the manuscript cannot be accepted as is, but the defect is localized and repairable (use −L and the correct constant p²λmin/[2(pλmin−1)]), so rejection is too strong. Hence the reader's conditional verdict stands unchanged.","tokens_in":34914,"tokens_out":39040,"duration_ms":461774,"concrete_test":"Set n=1, k=1, λmin=1, p=2, and f(x)=e^{εx} in Theorem 11. Compute L f/f = ε² − εx, so the printed RHS is 2·E[e^{2εx}(ε²−εx)] = −2ε²e^{2ε²}, while the entropy LHS is E[e^{2εx}·2εx] − e^{2ε²} log(e^{2ε²}) = 2ε²e^{2ε²}. The asserted inequality 2ε²e^{2ε²} ≤ −2ε²e^{2ε²} fails for every ε≠0. Replacing L by −L in the RHS gives +2ε²e^{2ε²}, matching the LHS to leading order and aligning with Gross's log-Sobolev inequality; this single check settles whether the sign and constant in Theorem 11 are a typo or a substantive error.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 4's derivation of Theorem 11 is internally inconsistent. The paper defines L = Δ − x·∇, so on Hermite polynomials L H_β = −|β| H_β and T_z = z^{−L}; hence ∂_r T_{φ(r)} = (φ′/φ)(−L)T_{φ(r)}, not (φ′/φ)L T_{φ(r)} as written. The proof also defines M g_φ = (1/φ) g_φ Σ L(T_φ f_j)/(T_φ f_j), dropping the φ′ factor, and later reinserts −r² φ′/φ E∏...Σ L(...)/... . The sign and magnitude errors propagate to the statement: the printed RHS is p²λmin/(2√(pλmin−1)) E F Σ Lf_j/f_j, whereas the corrected calculation gives −L and constant p²λmin/[2(pλmin−1)]. For n=1, k=1, λmin=1, p=2, f(x)=e^{εx}, the entropy LHS is +2ε²e^{2ε²} while the printed RHS is −2ε²e^{2ε²}, so the inequality fails at order ε². This does not falsify Theorems 13–14 by itself, but it means the paper's advertised log-Sobolev application, and the claim that all listed corollaries follow from the FB framework, are not established as printed. The separate sign gap between Proposition 1 (concavity/reverse) and Theorem 14 (convexity/forward) is repairable by applying Proposition 1 to −F, but this reduction is never stated.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a general \"FB\" framework for sharp multifunction Gaussian inequalities. Theorem 13 states that, under convexity of (t,y) -> F''(t)/F'(t) y^2, the inequality E F(B(|T_{z1} f1(ξ1)|,...,|T_{zn} fn(ξn)|)) <= F(E B(|f1(ξ1)|,...,|fn(ξn)|)) holds for all polynomials if and only if a local matrix condition (2.27) holds; Theorem 14 gives the analogous forward/reverse statement for real noise operators under convexity of F''/|F'| y^2. From these, the paper derives n-function complex and real hypercontractivity, n-function Hausdorff-Young inequalities, a correlated log-Sobolev inequality, reverse Hölder estimates for Gaussian chaoses, a noisy Gaussian-Jensen inequality, noisy Borell, and a covariance-based characterization related to Brascamp-Lieb. The main proofs are semigroup interpolation arguments: local conditions are obtained by second-order Taylor expansions, and sufficiency is shown by a Beckner-Janson type flow with a Jensen step. The exposition is detailed and self-contained, but Section 4 contains a sign error that makes Theorem 11 false as printed, and the reduction between the concavity/reverse setting of Proposition 1 and the convexity/forward setting of Theorem 14 is not stated.","tokens_in":35245,"tokens_out":15563,"duration_ms":176658,"significance":"If Theorems 13 and 14 are correct, the paper gives a substantial unifying framework: the covariance conditions are derived rather than fitted, and the corollaries connect several classical sharp inequalities with multifunction analogues. The semigroup proof strategy is original and the scope of applications is broad. However, the advertised log-Sobolev corollary is not established as printed, and the main real theorem relies on an unstated sign reduction. These issues are load-bearing for the paper's claims, although they appear repairable within the manuscript's approach.","major_comments":[{"comment":"The derivative identity for the Mehler flow has the wrong sign. Since L = Δ − x·∇ satisfies L H_β = −|β| H_β and T_z = z^{−L}, one must have ∂_r T_{φ(r)} = (φ′/φ)(−L) T_{φ(r)}, not (φ′/φ) L T_{φ(r)}. This sign error propagates into the definition of M g_{φ(r)} and into the final inequality (2.24). The printed statement is false: for n=1, k_1=1, λ_min=1, p=2 and f(x)=e^{εx}, the left-hand side of (2.24) is 2ε^2 e^{2ε^2}, while the printed right-hand side is −2ε^2 e^{2ε^2}. A corrected calculation gives Ent(f^p) ≤ p^2 λ_min/[2(pλ_min−1)] E(∏ f_j^p) Σ (−L f_j)/f_j, which for n=1, p=2 recovers the classical Gross inequality. As printed, the correlated log-Sobolev inequality and the claim that the listed corollaries follow from the FB framework are not established. This error should be fixed and the constants re-checked.","section":"Section 4, proof of Theorem 11, Eq. (4.8) and the line following it"},{"comment":"The bridge between Proposition 1 and Theorem 14 is not stated. Proposition 1 assumes that (t,y) ↦ F''(t)/|F'(t)| y^2 is concave and proves the reverse inequality (3.4) with the local condition (3.5) ≤ 0 and C(s) nonincreasing. Theorem 14 instead assumes convexity of the same quantity and concludes the forward inequality (2.28) with (2.29) ≥ 0. The transfer by applying Proposition 1 to −F is valid, but it is never written down, and the sign conventions in the Jensen step of Proposition 1 depend on F' through the |F'| in the denominator. Without this reduction, the proof of Theorem 14 is incomplete as presented.","section":"Section 3, Proposition 1 versus Theorem 14"},{"comment":"The text says twice 'for the implication (2.26) implies (2.27)' with different differentiability assumptions. The second occurrence should read '(2.27) implies (2.26)'. This is a typo, but it should be corrected because the two directions of an if-and-only-if theorem are being discussed.","section":"Section 2.4, list item after Theorem 13"}],"minor_comments":[{"comment":"There are several typos and misspellings: 'Mahler transform' should be 'Mehler transform', 'Seciton' in Section 4, 'moroever' in Theorem 11, 'measuralef' in the proof of Theorem 16, 'fameous' and 'haflspaces' in Section 2.6, and 'Borrel' in Theorem E.","section":"Throughout"},{"comment":"The matrix condition (2.9) contains denominators s_j^2; the statement should explicitly assume s_j ≠ 0 or discuss the limiting case separately.","section":"Theorem 2 and related conditions"},{"comment":"After the change of variables, the notation switches between (x,y), (u,v), and (ξ1,ξ2) without a consistent renaming; this makes the displayed computation harder to follow.","section":"Theorem 7 proof"},{"comment":"Even after correcting the sign, the text should define the Ornstein-Uhlenbeck generator and the Mehler transform convention once more near the proof, since the identity T_z = z^{−L} is central to the computation.","section":"Section 4, proof of Theorem 11"}],"recommendation":"major_revision","confidential_remarks":"The core FB theorems appear to be a genuine contribution, and the defects I found are localized. The sign error in Section 4 makes Theorem 11 false as printed, but the corrected calculation is close to the classical log-Sobolev inequality, so this should be fixable. I recommend major revision rather than rejection, provided the authors carefully re-derive the whole of Section 4 and explicitly state the reduction used in Theorem 14."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a serious paper with a real central framework, but as it stands the advertised log-Sobolev application is wrong as printed and the paper needs repair before acceptance. The FB theorem (Theorems 13–14) is a substantial extension: it gives n-function complex hypercontractivity with derived covariance conditions, and recovers Beckner, Bonami–Nelson, Borell, Paouris–Chen–Dafnis, and Neeman as special cases. The noisy Gaussian–Jensen inequality and the noisy Borell corollary are genuinely new. The proofs are mostly self-contained semigroup arguments; the conditions are derived, not fitted. That is real work and it deserves credit.\n\nThe soft spot is Theorem 11, and the stress-test is correct. With L = Δ − x·∇, the Mehler transform is T_z = z^{−L}, so ∂_r T_φ(r) = (φ′/φ)(−L)T_φ(r), not (φ′/φ)L as written. The proof also drops the −log E g^r term when differentiating ψ(r). The printed inequality is false: for n = 1, p = 2, λmin = 1, f(x) = e^{εx}, the entropy LHS is +2ε²e^{2ε²} while the printed RHS is −2ε²e^{2ε²}. This is not a minor typo in a constant; the statement itself fails. So the log-Sobolev corollary, and the claim that all listed applications follow from the framework, are not established as printed. Theorems 13–14 may still be fine—the error is in Section 4, not in the main FB proof—but the paper must fix this before it can be trusted.\n\nThere is also a smaller sign/convexity gap: Proposition 1 assumes concavity of (t,y) ↦ F″(t)/|F′(t)| y² and proves a reverse inequality, while Theorem 14 assumes convexity and proves a forward inequality; the bridge via −F is never stated. That is repairable, but it should be explicit.\n\nIf the authors correct Theorem 11 (or remove it and the dependent claims), fix the derivative calculation, and reconcile the convexity signs, the core results look credible. This paper deserves a serious referee, not a desk reject, and I would send it back with a clear request for revision.","headline":"A substantial FB framework with derived covariance conditions, but Theorem 11 as printed is false and the log-Sobolev application needs repair.","tokens_in":35813,"tokens_out":4101,"would_cite":false,"duration_ms":44406,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["42B20","42B35","47A30","42A38"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that a covariance matrix condition is equivalent to a whole family of sharp multifunction Gaussian inequalities, with the Beckner–Janson flow as the proof engine.","keywords":["Hausdorff–Young inequality","hypercontractivity","Beckner–Janson flow","multifunction inequalities","Gaussian correlation","Brascamp–Lieb inequality","log-Sobolev inequality","noise stability"],"falsifier":"Compute $C'(0)$ directly for $F(t)=-t$, $n=2$, $B(c_1,c_2)=c_1c_2$, $r=(0,0)$, and covariance $\\rho>0$: the local condition (2.29) reduces to the scalar $\\rho\\ge 0$, and the global inequality is the true statement $E[f_1(\\xi_1)f_2(\\xi_2)]\\ge Ef_1(\\xi_1)Ef_2(\\xi_2)$. A discriminating case is $F(t)=t^\\alpha$ with $0<\\alpha<1$, where the convexity assumption of Theorem 14 fails but Proposition 1's reverse inequality should coincide exactly with the known reverse hypercontractivity direction; if the directions disagree, the sign bridge is broken.","tokens_in":34697,"feed_emoji":"📐","tokens_out":10399,"duration_ms":117244,"temperature":0.7,"pith_summary":"The paper claims that a large family of sharp inequalities in Gauss space—multifunction complex and real hypercontractivity, the Hausdorff–Young inequality, log-Sobolev, Gaussian–Jensen, and Borell noise stability—are all governed by one abstract equivalence. For smooth functions $F$ and $B$, the inequality $E F(B(|T_{z_1} f_1(\\xi_1)|,\\ldots,|T_{z_n} f_n(\\xi_n)|)) \\le F(E B(|f_1(\\xi_1)|,\\ldots,|f_n(\\xi_n)|))$ holds for all polynomials if and only if a local matrix quadratic form (2.27) is nonnegative, provided the curvature map $(t,y)\\mapsto \\frac{F''(t)}{F'(t)}y^2$ is convex. The real-parameter version (Theorem 14) extends the equivalence to all measurable functions and to reverse inequalities, with convexity of $\\frac{F''(t)}{|F'(t)|}y^2$ and the local condition (2.29). Choosing $F(t)=t^\\alpha$ and $B(t)=t_1^{p_1}\\cdots t_n^{p_n}$ reduces the condition to two-sided covariance bounds whose eigenvalues give sharp constants, so each classical single-function inequality is recovered as the $n=1$ case. A sympathetic reader should care because the paper's payoff is structural: one flow, one matrix condition, and many sharp inequalities follow.","feed_headline":"One covariance check unlocks multifunction inequalities","feed_subtitle":"A single flow-based FB theorem turns hypercontractivity, log-Sobolev, and noisy Borell into one semidefinite condition.","key_machinery":"The machinery is the Beckner–Janson flow, an interpolation $C(s)$ built by applying two heat flows, $Q^u_s$ and $Q^x_{1-s}$, to functions $g_p(u,x,s)=\\int f_p(A_pu+r_pA_px+\\sqrt{1-r_p^2}y)\\,d\\gamma_{1-s}(y)$ in the real case, so that $C(0)$ is the left side of the target inequality and $C(1)$ is the right side. The proof differentiates $C(s)$, uses the heat equation to express $C'(s)$ as a sum of local terms, and applies Jensen's inequality, which is exactly where the convexity or concavity assumption on $(t,y)\\mapsto \\frac{F''(t)}{F'(t)}y^2$ enters to move the inner heat flow outside the quadratic term. The remaining block-matrix quadratic form collapses to the local condition (2.29) or (2.27), so monotonicity of the flow is equivalent to that matrix being semidefinite in the required direction. This reduces a global functional inequality to a pointwise matrix check.","core_discovery":"The central claim is that the if-and-only-if route from local matrix condition to global inequality is valid for the $(F,B)$ pair, not just for power functions. In the complex case, Theorem 13 states that for $|z_j|\\le 1$, $F'>0$, $B_m>0$, and convex $(t,y)\\mapsto \\frac{F''(t)}{F'(t)}y^2$, inequality (2.26) holds for every polynomial $f_j$ exactly when the quadratic form (2.27) is nonnegative at every $c>0$ and every complex $w_p$. The real case Theorem 14 weakens the hypotheses, requiring no polynomial growth condition, no positivity of $B_m$, and allowing measurable test functions, and it gives both forward and reverse inequalities, with the direction tied to the sign of $F'$ and to convexity of $\\frac{F''(t)}{|F'(t)|}y^2$. From these the paper derives Theorem 1 (n-function complex hypercontractivity) and Theorem 9 (n-function forward and reverse real hypercontractivity) as corollaries, and from those the sharp multifunction Hausdorff–Young inequality, the correlated log-Sobolev inequality, moment comparisons for Gaussian chaoses, and the noisy Borell theorem.","pith_inferences":["Beyond the paper: because the local condition is a semidefinite matrix inequality, the sharp constants for given covariance matrices and exponents could be computed automatically, turning the theorems into a feasibility test rather than a case-by-case analysis.","Beyond the paper: the framework leaves open entropy-type outer functions such as $F(t)=t\\log t$; if the convexity condition holds for such $F$, the same flow would produce correlated entropy or log-Sobolev inequalities for products of many functions beyond Theorem 11.","Beyond the paper: the noisy Gaussian–Jensen theorem suggests that Brascamp–Lieb inequalities with additional noise operators could be derived by choosing $B$ to encode pairwise interactions, a direction the paper only touches through the covariance characterization."],"forward_implications":["Taking $n=1$ recovers Beckner's complex hypercontractivity and hence the sharp Hausdorff–Young inequality, Bonami–Nelson real hypercontractivity, reverse hypercontractivity, and Gross's log-Sobolev inequality as corollaries of one theorem.","The covariance condition in Theorem 9 gives an if-and-only-if test: for any correlated Gaussian family and any exponents $p_j$, forward or reverse hypercontractivity is decided by a matrix inequality, with the sharp noise parameter read off from $\\lambda_{\\min}$ and $\\lambda_{\\max}$.","Theorem 6 supplies a sharp multifunction Hausdorff–Young inequality with explicit Gaussian extremizers and best constant $(p\\lambda_{\\min})^{\\sum k_j/2}$.","Theorem 11 yields a correlated, best-constant log-Sobolev inequality for products $\\prod_j f_j^{p_j}(\\xi_j)$ with constant $p^2\\lambda_{\\min}/(2\\sqrt{p\\lambda_{\\min}-1})$, interpolating Gross's inequality when the Gaussian coordinates are independent.","Theorem 16 generalizes Borell's noise stability to two different noise levels $r_1,r_2$ applied to the two sets, with the sharp bound expressed by the same function $M$ and equality on parallel halfspaces."],"supporting_citations":[{"why":"Supplies the $n=1$ complex hypercontractivity and sharp Hausdorff–Young inequality that the multifunction theorem must recover as a base case.","marker":"[11]"},{"why":"Introduces the continuous Beckner–Janson flow, the monotonicity engine used to prove the FB theorem.","marker":"[37]"},{"why":"Shows standard heat-flow monotonicity fails for Hausdorff–Young, which is why the subtler Beckner–Janson flow is needed.","marker":"[4]"},{"why":"Gives the n-function covariance characterization of Hölder and Brascamp–Lieb inequalities that Theorem 9's condition reduces to when the noise parameters vanish.","marker":"[22]"},{"why":"Supplies reverse real hypercontractivity, the template for the reverse direction of Theorem 14.","marker":"[15]"},{"why":"Is the single-pair noise stability theorem that Theorem 16 extends to two different noise levels.","marker":"[16]"},{"why":"Provides the multiversion noise stability bound that Theorem 15 recovers when all noise parameters are zero.","marker":"[48]"},{"why":"Establishes the $n=1$ $(P,Q)$-hypercontractivity case that Theorem 13 recovers and generalizes.","marker":"[32]"},{"why":"Identifies Gaussian kernels as the only maximizers, supporting the sharpness claims such as the constant in Theorem 6.","marker":"[44]"}],"fun_headline_variants":["One covariance check, many inequalities","Sharp Hausdorff-Young for many functions","A flow-based theorem unifies hypercontractivity and log-Sobolev","Multiversion Hausdorff-Young via semidefinite condition","Covariance check unifies hypercontractivity and beyond"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof assumes the sign convention that convexity of $(t,y)\\mapsto \\frac{F''(t)}{F'(t)}y^2$ makes the interpolation flow monotone in the direction needed for the forward inequality, with the reverse case handled by an unstated reduction of $F$ to $-F$; if that sign is wrong, the local matrix condition does not force the global inequality.","fun_headline_variants_meta":{"raw":{"variants":["One covariance check, many inequalities","Sharp Hausdorff-Young for many functions","A flow-based theorem unifies hypercontractivity and log-Sobolev","Multiversion Hausdorff-Young via semidefinite condition","Covariance check unifies hypercontractivity and beyond"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000936,"raw_usage":{"total_tokens":4038,"prompt_tokens":1015,"completion_tokens":3023,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":631,"completion_tokens_details":{"reasoning_tokens":2943}},"tokens_in":631,"tokens_out":3023,"duration_ms":23796,"temperature":1.0,"reasoning_tokens":2943,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T05:13:31.408355+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute $C'(0)$ directly for $F(t)=-t$, $n=2$, $B(c_1,c_2)=c_1c_2$, $r=(0,0)$, and covariance $\\rho>0$: the local condition (2.29) reduces to the scalar $\\rho\\ge 0$, and the global inequality is the true statement $E[f_1(\\xi_1)f_2(\\xi_2)]\\ge Ef_1(\\xi_1)Ef_2(\\xi_2)$. A discriminating case is $F(t)=t^\\alpha$ with $0<\\alpha<1$, where the convexity assumption of Theorem 14 fails but Proposition 1's reverse inequality should coincide exactly with the known reverse hypercontractivity direction; if the directions disagree, the sign bridge is broken.","supporting_citations":[{"cited_title":"Ivanisvili, A","cited_arxiv_id":null,"evidence_quote":"Introduces the continuous Beckner–Janson flow, the monotonicity engine used to prove the FB theorem."},{"cited_title":"older and reverse H\\","cited_arxiv_id":null,"evidence_quote":"Gives the n-function covariance characterization of Hölder and Brascamp–Lieb inequalities that Theorem 9's condition reduces to when the noise parameters vanish."},{"cited_title":"Neeman , A multidimensional version of noise stability , Electron.\\ Commun.\\ Probab.\\ 19 (2014), no","cited_arxiv_id":null,"evidence_quote":"Provides the multiversion noise stability bound that Theorem 15 recovers when all noise parameters are zero."},{"cited_title":"Hariya , A unification of the hypercontractivity and its exponential variant of the Ornstein--Uhlenbeck semigroup , J.\\ Funct.\\ Anal.\\ 275 (2018), no","cited_arxiv_id":null,"evidence_quote":"Establishes the $n=1$ $(P,Q)$-hypercontractivity case that Theorem 13 recovers and generalizes."},{"cited_title":"Lieb , Gaussian kernels have only Gaussian maximizers , Invent.\\ Math.\\ 102 (1990), no","cited_arxiv_id":null,"evidence_quote":"Identifies Gaussian kernels as the only maximizers, supporting the sharpness claims such as the constant in Theorem 6."}],"review_version":1}