{"id":"3821ac24-1b57-4904-8aa0-420f41335ebd","arxiv_id":"2603.22431","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Korn's constant is shown to be ≤√3(p*−1) in every dimension, with lower bound p*−1, and conjectured to equal p*−1.","lead":"This paper proves a new, dimension-free bound for the constant in Korn's inequality, a core estimate in elasticity and analysis: the antisymmetric part of a gradient is controlled by the symmetric part with a constant at most √3(p*−1). The method borrows martingale techniques from harmonic analysis and links the optimal constant to a long-standing open problem in the calculus of variations.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proposition 4.1's proof omits centering of M and L^p convergence details; Theorem 1.1's upper bound depends on this.","rationale":"Reader verdict CONDITIONAL with medium risk; I agree the central claim is probably correct, but the proof as written has a precise technical gap. The most load-bearing link is Proposition 4.1, which underlies the dimension-free bound. The reader's weakest assumption (L^p convergence and differential subordination) points in the right direction, but the sharper issue is that the subordination theorem is invoked for a pair that does not satisfy X0=Y0=0; M0 is nonzero, and the proof must pass to the limit after centering. This is not an objection to the result, only to the submitted derivation. Because the missing steps are standard and likely fixable, the conditional verdict is correct. No ad hominem, no overstatement; the paper deserves revision, not rejection.","tokens_in":29060,"tokens_out":41117,"duration_ms":342186,"concrete_test":"Write out a complete proof of Proposition 4.1 for scalar f: (i) replace M by M−M_0 and verify Burkholder's hypotheses; (ii) use Itô's formula (or [9, Thm 3.9.1]) to identify lim_{T→∞} E(Z_T|Y_T=(x,0)) as the heat-truncated Riesz transform R_iR_j f and check convergence in L^p for all f∈L^p, with the truncations uniformly bounded. If the centered Burkholder step yields an extra term ∥P_T f∥_p that fails to vanish, or if the limiting operator differs from R_iR_j by a non-negligible multiplier, Proposition 4.1—and hence Theorem 1.1—is invalid. This is a single analytical verification of the missing details.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central upper bound in Theorem 1.1 is obtained by applying Burkholder's differential-subordination theorem to the matrix-valued heat martingales Z and M of §4.1. As written, the argument has a load-bearing gap: Theorem 3.2 requires X0=Y0=0 and d⟨X⟩≤d⟨Y⟩; here M^T_0 = P_T f(B_0), which is not zero for f∈L^p (unless f has zero mean). Thus the displayed inequality ∥Z_T∥≤(p*−1)∥M_T∥ is not justified. The fix is to compare Z to M−M_0, which yields an extra term ∥P_T f∥_p that disappears only after T→∞; the paper does not supply this centering. In addition, the L^p convergence of E(Z_T|Y_T=(x,0)) to R⊗R f for matrix-valued f is asserted as 'easy to check' without proof. Both steps are necessary: if the centered limit produces a residual T-dependent term, or if the convergence fails for general f∈L^p(M_d), the bound ∥A(u)∥≤√3(p*−1)∥E(u)∥ does not follow. The gaps appear repairable, but in the submitted form the proof of Proposition 4.1 is incomplete.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies optimal constants in Korn's inequality and their relation to Morrey's problem. The main claim (Theorem 1.1) is that for u∈Ẇ^{1,p}(R^d,R^d) one has ∥A(u)∥_{L^p}≤√3(p*−1)∥E(u)∥_{L^p}, hence C(p,d)≤√3(p*−1), together with C(p,d)≥p*−1; analogous statements are claimed for the full-gradient and trace-free versions, for Orlicz spaces, and for Muckenhoupt-weighted spaces. The upper bound is obtained by representing the matrix operator R⊗R through heat-extension martingales, applying Burkholder's differential-subordination theorem, and then using a linear-algebra lemma (Lemma 4.1) to improve from p*−1 to √3(p*−1). Lower bounds are obtained in dimension 2 by explicit radial witnesses and in general by a reduction to rank-one convexity importing results from the author's preprint [23]. The paper also draws a conceptual connection between the sharp constant and quasiconvexity of a Burkholder-type integrand.","tokens_in":29403,"tokens_out":20583,"duration_ms":218717,"significance":"If established, the main theorem would be the first dimension-free upper bound for the Korn constant with the optimal linear growth in p*−1, and it would give strong support to the natural conjecture that the sharp constant is p*−1. The approach via heat martingales is attractive and the explicit 2D witnesses in Proposition 4.4 and the linear-algebra lemma are convincing and useful. However, the manuscript as submitted is not ready: the key Burkholder step has an initial-value gap, the L^p convergence in Proposition 4.1 is asserted rather than proved, the trace-free bound rests on an unproved lemma, and the lower-bound half of Theorem 1.1 is essentially imported from another preprint of the author. These issues are local and probably repairable, but they are load-bearing for the central claims.","major_comments":[{"comment":"The application of Theorem 3.2 is not justified. Definition 2 and Theorem 3.2 require X_0=Y_0=0. For M_t^T=P_{T-t}f(B_t) with Brownian motion started from the Lebesgue measure, M_0^T=P_T f(B_0) is generically non-zero, while Z_0^T=0. Thus the pair (Z^T,M^T) is not differentially subordinate in the sense of the paper, and the displayed inequality ∥Z_T^T∥_{L^p}≤(p*−1)∥M_T^T∥_{L^p} does not follow from the stated theorem. A repair is possible — one can either state and prove the continuous-time Burkholder theorem with the initial condition |X_0|≤(p*−1)|Y_0|, or compare Z^T with M^T−M_0^T and control the extra term as T→∞ — but the manuscript supplies neither. The same defect is inherited by the weighted proof in §6.2.","section":"§4.1, Proposition 4.1"},{"comment":"The assertion 'As one easily checks, even in the matricial case we have convergence in L^p' is load-bearing. The operator (R⊗R)_T must be shown to converge to R⊗R in the strong operator topology on L^p(R^d,M_d(R)); for matrix-valued functions on an infinite measure space this is not completely routine, and the final bound is obtained by letting T→∞. A proof, at least for f∈C_c^∞ followed by a density argument, should be supplied.","section":"§4.1, Proposition 4.1"},{"comment":"The trace-free upper bound C_0(p,d)≤√3(p*−1) for d≥3 relies entirely on Lemma 4.2, but Lemma 4.2 is stated without proof; the text only says the proof is 'very similar' to that of Lemma 4.1. Since Lemma 4.2 is the only new ingredient for the trace-free case, its proof must be included. In addition, the p∈(1,2) case of Theorem 4.3 is only sketched ('using the convexity of t^{2/p}') with no displayed argument; this should be expanded.","section":"§4.3, Lemma 4.2"},{"comment":"The lower bounds C(p,d)≥p*−1 and C_0(p,d)≥p*−1 are quoted as a direct corollary of [23, Theorem 2.2] together with Lemma 5.2. Thus the paper does not contain a proof of half of Theorem 1.1; [23] is an unpublished preprint of the author. Please either include a self-contained derivation or state explicitly the imported theorem, its hypotheses, and its status. As written, a central part of the main theorem is not verifiable from the manuscript.","section":"§5, Theorem 5.1"},{"comment":"The abstract claims that in dimension 2 the paper obtains a bound sharp up to a factor of 1.158. No statement or proof of such a bound appears in the body; Theorem 1.1 gives the factor √3≈1.732. Please either add the advertised 2D theorem with a proof or reference, or correct the abstract to match the results actually proved.","section":"Abstract vs. body"}],"minor_comments":[{"comment":"The description of X_k as '(p(k−1)+1, 1/k) Gamma-distributed' is ambiguous between scale and rate parametrizations. Write the density or specify 'rate k' explicitly.","section":"§4.4, Eq. (25)–(28)"},{"comment":"The statement uses p and q as Φ's indices but defines them only later in Section 6.1. State p=i_Φ and q=s_Φ in the theorem statement.","section":"Theorem 1.2"},{"comment":"The notation ∇2(∞) and Δ2(∞) is used in Proposition 6.1 without definition; please define it or refer explicitly to the earlier Δ2/∇2 terminology.","section":"§6.1"},{"comment":"The notation in the calculation of N mixes ζ⊗ζ^t, (ζ⊗ζ^t)^†, and ζ⊗ζ^† in a way that is hard to follow. A short explanation of the convention would improve readability.","section":"§2, Proposition 2.2"},{"comment":"The scaling argument is sketched rather tersely. In particular, the use of f_d(rx_1,…,r x_d) and the need for f_d to be (or be approximated by) compactly supported functions should be made explicit, since f_d need not belong to L^p(R^d).","section":"§4.2, Proposition 4.3"}],"recommendation":"major_revision","confidential_remarks":"The technical gaps appear repairable, so I would not recommend rejection. However, I would ask the editor to ensure that the dependence on the companion preprint [23] is made explicit and that the paper is self-contained enough for the claims in Theorem 1.1. The abstract's 1.158 claim must be reconciled with the body before acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Main thing to know: the central result is real. A dimension-free upper bound C(p,d) ≤ √3(p*-1) for Korn's inequality is a genuine advance over Hytönen's dimension-dependent bound, and the method—matrix-valued heat martingales plus a linear-algebra trick—is the right kind of tool for this problem. Lemma 4.1 is a nice observation, and the Orlicz and weighted variants follow naturally from the same machinery. If the main theorem is correct, this is an important paper for elasticity, singular integrals, and the Beurling–Ahlfors program.\n\nThe soft spots are concentrated in the proof of Proposition 4.1. The paper applies Theorem 3.2 to the pair (Z^T, M^T), but that theorem requires both martingales to start at zero. For finite T, M^T_0 = P_T f(B_0) is not zero, so the inequality ∥Z_T∥ ≤ (p*-1)∥M_T∥ is not justified as stated. The fix is standard: center both martingales and let the extra term ∥P_T f∥_p vanish as T→∞. That is a one-line repair, but the paper does not supply it, and as written it is a genuine gap. The L^p convergence of the conditional expectation to R⊗R f for matrix-valued f is also asserted without proof; the scalar case is in the cited literature, so this is likely fine but still needs detail.\n\nThe lower bound for p < 2 is imported from the companion preprint [23], and for p≥2 it is proved in this paper via Prop 4.3 and 4.4. That is a dependency, not a fatal flaw, but it means part of Theorem 1.1 does not stand alone.\n\nOne more issue: the arXiv metadata abstract promises a 2D bound sharp up to a factor of 1.158, but the paper's own abstract and body never deliver that. It looks like a leftover from an earlier version and should be corrected.\n\nOverall, the main claim is probably right, the technique is interesting, and the gaps are localized and repairable. This deserves a serious referee; the likely outcome is major revision, not rejection.","headline":"The dimension-free Korn bound with factor √3 is a real new result and the proof outline is sound; the gaps are repairable, but the manuscript needs a few fixes before it can be trusted as written.","tokens_in":29895,"tokens_out":4450,"would_cite":true,"duration_ms":45311,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35A23","26D10","49J45","60G44","46E35","42B20"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves a dimension-free upper bound of √3(p∗−1) for the best constant in Korn's inequality, matching the conjectured sharp value up to a universal factor.","keywords":["Korn inequality","dimension-free bounds","sharp constants","martingale inequalities","Riesz transforms","quasiconvexity","rank-one convexity","Muckenhoupt weights"],"falsifier":"For a fixed p∈(1,∞) and d≥2, take a matrix-valued Schwartz function f and compare R⊗R(f) with the conditional expectation E[Z^T_T | Y^T_T = (x,0)] as T→∞; if the L^p difference fails to go to zero for some f, Proposition 4.1 and thus Theorem 1.1 collapse. Alternatively, test the inequality itself: any sequence u_n with ||A(u_n)||_{L^p} > √3(p∗−1)||E(u_n)||_{L^p} would falsify the theorem, while one with ratio strictly larger than p∗−1 would falsify the sharp-constant conjecture.","tokens_in":28962,"feed_emoji":"📐","tokens_out":12977,"duration_ms":118904,"temperature":0.7,"pith_summary":"Korn's inequality, which bounds the antisymmetric part of a vector field's gradient by its symmetric part, is a load-bearing tool in linearised elasticity and fluid mechanics. This paper establishes that the optimal constant C(p,d) is never larger than √3(p∗−1), with a matching lower bound p∗−1, where p∗ = max(p, p′). Because the upper bound does not depend on the ambient dimension d, coercivity estimates no longer degrade in high dimensions. The proof encodes the symmetric part of the gradient into the heat extension of a martingale and applies the classical differential-subordination inequality for martingales, with a tensor identity contributing the factor √3. It also yields dimension-free versions in Orlicz spaces and for Muckenhoupt weights, and it connects the sharp-constant question to the long-standing rank-one-convexity versus quasiconvexity problem.","feed_headline":"Dimension-free Korn bound lands within √3 of sharp","feed_subtitle":"The optimal constant is pinned between p*−1 and √3(p*−1) for every dimension, sharpening coercivity estimates.","key_machinery":"The space-time Brownian heat martingale M_t^T = P_{T−t} f(B_t) and its matrix-valued stochastic-integral analogue Z_t^T, built by integrating the heat kernel's spatial derivatives against the coordinates of the Brownian motion. For symmetric matrix inputs, the quadratic-variation comparison d⟨Z^T⟩ ≤ 3 d⟨M^T⟩ follows from the spectral identity on R^d⊗Sym(d): the antisymmetrisation operator I−σ satisfies |(I−σ)a|² ≤ 3|a|², with sharp constant 3. Plugging this comparison into the differential-subordination theorem and letting T→∞ transfers the martingale inequality into the dimension-free bound for the Riesz transform and hence for Korn's inequality.","core_discovery":"The central claim is that the Korn constant satisfies C(p,d) ≤ √3(p∗−1) and C(p,d) ≥ p∗−1 (trace-free version in d≥3 included); for generalised-radial fields the constant is exactly p∗−1, sharp in dimension two for p≥2. The proof begins with the identity A(u) = (R⊗R)E(u) − ((R⊗R)E(u))^t, reducing Korn's inequality to an L^p bound for the matrix-valued second-order Riesz transform. That operator is represented by a space-time Brownian martingale, against which the classical differential-subordination inequality yields p∗−1; a tensor lemma then shows that antisymmetrisation on symmetric-matrix inputs inflates the quadratic variation by at most 3, giving the √3. The lower bound p∗−1 is obtained","pith_inferences":["The martingale-representation route is likely to transfer to other elliptic homogeneous constant-coefficient operators whose symbol maps matrices into a fixed subspace; any such operator should admit a dimension-free L^p bound whenever a quadratic-variation lemma of the same type holds.","A concrete numerical experiment could test the sharp-constant conjecture: compute extremal ratios ||A(u)||_p/||E(u)||_p for large p in dimensions 3 and 4; if the supremum stays strictly below √3(p∗−1), the √3 is an artifact of the proof and the conjecture p∗−1 becomes more credible.","If the quasiconvexity statement fails, the construction would provide a new example of a rank-one convex but non-quasiconvex integrand, feeding directly into the open problem on the gap between the two notions in dimensions above two.","The dimension-free weighted estimate with heat weights suggests an analogous sharp weighted bound with the usual Muckenhoupt characteristic [w]_{A_p}, which would be the Korn analogue of the sharp weighted bound for classical singular integrals."],"forward_implications":["The Korn constant is bounded by √3(p∗−1) in every dimension, so coercivity estimates in elasticity and hydrodynamics no longer worsen as the dimension grows.","The constant is always at least p∗−1, and the gap to the upper bound is a universal factor √3; for generalised-radial fields the constant is exactly p∗−1, with sharpness on radial examples in two dimensions for p≥2.","The trace-free Korn inequality in d≥3 and the full-gradient variant satisfy the same dimension-free estimates, with the full-gradient constant bounded by √(3(p−1)²+1) for p≥2.","Korn's inequality holds in every Orlicz space in which it is possible (Φ satisfying the Δ₂ and ∇₂ conditions), with a dimension-free constant depending only on the Orlicz indices; the same holds in weighted L^p(w) for Muckenhoupt weights, dimension-free in the weight characteristic.","The sharp-constant conjecture is reduced to a single quasiconvexity question; if the extremal function is quasiconvex at 0, the Korn constant is p∗−1 and the Morrey-type gap between rank-one convexity and quasiconvexity is resolved for this family."],"fun_headline_variants":["Korn constant pinned to within √3 in all dimensions","Martingale proof settles Korn constant up to √3","Dimension-free Korn bound: √3 from sharp","For Korn, √3 is the universal factor","Korn's constant ties Morrey's problem, sharp to √3"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The upper-bound proof assumes, without proof, that the conditional expectations of the matrix-valued space-time Brownian martingale converge in L^p to the matrix Riesz operator R⊗R as the heat horizon tends to infinity; if that convergence fails, the dimension-free bound does not follow.","fun_headline_variants_meta":{"raw":{"variants":["Korn constant pinned to within √3 in all dimensions","Martingale proof settles Korn constant up to √3","Dimension-free Korn bound: √3 from sharp","For Korn, √3 is the universal factor","Korn's constant ties Morrey's problem, sharp to √3"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001687,"raw_usage":{"total_tokens":6508,"prompt_tokens":716,"completion_tokens":5792,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":460,"completion_tokens_details":{"reasoning_tokens":5713}},"tokens_in":460,"tokens_out":5792,"duration_ms":39611,"temperature":1.0,"reasoning_tokens":5713,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T05:42:05.238496+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a fixed p∈(1,∞) and d≥2, take a matrix-valued Schwartz function f and compare R⊗R(f) with the conditional expectation E[Z^T_T | Y^T_T = (x,0)] as T→∞; if the L^p difference fails to go to zero for some f, Proposition 4.1 and thus Theorem 1.1 collapse. Alternatively, test the inequality itself: any sequence u_n with ||A(u_n)||_{L^p} > √3(p∗−1)||E(u_n)||_{L^p} would falsify the theorem, while one with ratio strictly larger than p∗−1 would falsify the sharp-constant conjecture.","supporting_citations":[],"review_version":2}