{"id":"f93ebe9a-6135-46ea-945e-33bc16d6b6b1","arxiv_id":"2512.02784","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The minimal number of linear measurements controlling ∇u in Korn's second inequality is exactly 2d−1, and in the first inequality it grows like 2d asymptotically.","lead":"This paper determines exactly how many scalar measurements of the gradient are needed for Korn-type inequalities: 2d−1 for the second version and asymptotically 2d for the first. The proof connects Korn inequalities to bilinear algebra and martingale theory, yielding a new dimension-optimal Korn-Hankel inequality.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proof of N′(d)=2d−1 applies the C-ellipticity theorem to the wrong subspace: it requires X^⊥ to avoid rank-one, but the text states X avoids rank-one; the Hankel construction supplies the missing condition, so the result is likely repairable but needs a revision.","rationale":"The reader’s verdict is CONDITIONAL, and I agree that the central claims are likely correct but that the paper needs revision. The reader identified the imported Kałamajska theorem as the load-bearing bridge. My stress-test narrows this to a more concrete internal problem: the upper-bound proof of N′(d)≤2d−1 states the ellipticity condition backwards. The operator is P_X(∇u), so C-ellipticity requires X^⊥ to avoid complex rank-one matrices; the proof says X avoids rank-one. This is not a cosmetic typo, because the rest of the proof and Remark 14 use the correct condition. However, the gap is repairable without changing the mathematical result: the Hankel subspace H(d) has dimension 2d−1 and its orthogonal complement avoids rank-one by a one-line polynomial argument. The paper already introduces H(d) and the Hankel Korn inequality, so the fix is natural. Thus the central claim is not undermined, but the proof as written is formally invalid at its most crucial step. The reader’s CONDITIONAL verdict remains appropriate; I would not reject the paper, but the revision must correct this proof and verify the C-ellipticity condition explicitly. I also credit the paper’s lower-bound argument, the algebraic characterizations, and the explicit Hankel construction, all of which are independent support for the main result.","tokens_in":24674,"tokens_out":18857,"duration_ms":174275,"concrete_test":"Rewrite the upper-bound proof of Section 4 with the correct hypothesis: take X=H(d), and verify the C-ellipticity condition used by [48, Theorem 4] by computing the principal symbol. For each ξ∈C^d\\{0}, the symbol of T_H at ξ is a↦P_H(a⊗ξ); show it is injective by noting that if P_H(a⊗ξ)=0, then the polynomial (Σ a_i z^i)(Σ ξ_j z^j) has all zero coefficients, hence a=0. Then confirm that the domain hypotheses of [48, Theorem 4] (star-shaped with respect to a ball, with extension to Lipschitz domains) are met for T_H. If [48, Theorem 4] indeed applies to this operator, the inequality ||u||_{H^1}≲||u||_{L^2}+||P_H(∇u)||_{L^2} follows and N′(d)=2d−1 is established. If, instead, the theorem as stated in the manuscript (X avoids rank-one) is used, no such X of dimension 2d−1 can be C-elliptic, so the proof must be corrected.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The exact value N′(d)=2d−1 is the central claim, and its upper bound rests on the proof in Section 4. The lower bound is sound: if dim_R X<2d−1, then X_C^⊥ has complex dimension >(d−1)^2 and, by the complex projective dimension bound, contains a nonzero rank-one matrix, yielding an infinite-dimensional space S={u:P_X∇u=0} on which H^1 and L^2 are equivalent, contradicting compactness. The upper-bound proof, however, is written with the wrong hypothesis. It says: “Let X be a subspace of M_d(C) such that X avoids all rank one matrices. Let T_X denote the differential operator T_X(u)=P_X(∇u).” It then invokes [48, Theorem 4] via condition (C). But for the operator P_X(∇u) to be C-elliptic, the required condition is that X^⊥ contains no nonzero complex rank-one matrix — equivalently, P_X(a⊗ξ)≠0 for all a≠0, ξ≠0. The paper itself states this correctly in Remark 14. The condition “X avoids rank-one” is not equivalent and does not imply C-ellipticity; for example, in d=2, X=span of a nonzero skew-symmetric matrix avoids rank-one, while X^⊥ contains many rank-one matrices and P_X is not elliptic. Thus, as written, the proof of N′(d)≤2d−1 is internally inconsistent. The repair is standard and almost explicit in the paper: take X=H(d), the Hankel subspace of real dimension 2d−1. Then H(d)^⊥ avoids rank-one, because if P_H(a⊗b)=0, all skew-diagonal sums of a_i b_j vanish, meaning the product of the two polynomials Σ a_i z^i and Σ b_j z^j is identically zero, forcing a=0 or b=0. So the central theorem is very likely correct, but the manuscript must either correct the quantifier in Section 4 to “X^⊥ avoids rank-one” or explicitly prove the upper bound by exhibiting H(d) and verifying the C-ellipticity condition before applying [48, Theorem 4]. This is the most load-bearing gap: the imported theorem is standard, but its application in the text is mis-specified.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies Chipot's minimal number of scalar measurements needed for Korn-type control of the gradient. It reformulates the problem in terms of subspaces of matrices avoiding rank-one matrices, proves the algebraic characterisations N(d)=g_R(d) and N'(d)=g_C(d), and derives the sharp values N'(d)=2d−1 and N(d)=2d(1−o(1)), together with a dimension-optimal Korn inequality based on the Hankel subspace H(d). The paper also develops a martingale–laminate technique to obtain sharp constants, treats the rectangular case, extensions to L^p, and a quantitative proof of Ornstein's non-inequality.","tokens_in":25149,"tokens_out":12764,"duration_ms":117882,"significance":"If fully correct, these results settle the growth rate of Chipot's minimal dimensions and, in particular, give the exact closed form for the second Korn inequality. The Hankel projection inequality is a genuinely stronger and dimension-optimal replacement for the classical Korn inequality. The martingale–laminate connection with Burkholder's sharp constants is a promising tool for coercivity questions. The paper is also honest about the algebraic characterisation being previously known (Remark 3) and the external C-ellipticity theorem is clearly identified. However, two load-bearing arguments are not correctly written as they stand: the upper bound for N'(d) uses the wrong rank-one hypothesis, and the sharpness proof of Theorem 2.2 omits the scaling needed to reach the claimed constant. These are repairable, but the revision must close them before the central claims are fully established.","major_comments":[{"comment":"The proof as written invokes [48] with the wrong hypothesis. It states: 'Let X be a subspace of M_d(C) such that X avoids all rank one matrices' and then claims that T_X(u)=P_X(∇u) satisfies condition (C). C-ellipticity of this operator requires that X_C^⊥ contains no nonzero complex rank-one matrix, equivalently P_X(a⊗ξ)≠0 for all nonzero a,ξ; this is exactly the condition stated in Remark 14. 'X avoids rank-one' is not equivalent and does not imply C-ellipticity. In fact H(d) itself contains rank-one matrices (e.g. the all-ones matrix), while H(d)^⊥ avoids rank-one by the polynomial-convolution argument. Thus the upper bound N'(d)≤2d−1 is not proved as written. The repair is straightforward: apply the argument to X=H(d) (or its complexification), verify H(d)_C^⊥∩R_1(C)=∅, and correct the wording. Because this step is load-bearing for the main exact value, it must be fixed explicitly in","section":"Section 4, Proof of N'(d)≤g_C(d)"},{"comment":"The proof of the sharpness direction is only a sketch and does not, as written, yield the claimed threshold c(p*−1). The martingale M_n=A f_n + B g_n with A∈X, B∈X^⊥ has rank-one increments only if rank(A±B)=1, whereas the theorem assumes rank(A±cB)=1. To use the stated hypothesis one must scale the A-component, e.g. M_n=(1/c)A f_n + B g_n, and then track the induced ratio ∥Q_X(M_n)∥/∥P_X(M_n)∥. Without this scaling the argument gives only the threshold p*−1. Since Theorem 2.2 underlies the sharp-constant claims and the quantitative Ornstein result, this gap needs to be closed in the revision.","section":"Section 2, Theorem 2.2"}],"minor_comments":[{"comment":"The notation m◦d is used without definition. The paper defines d◦d in Section 3.1; the Hopf–Stiefel function should be defined or a reference given.","section":"Section 5, Theorem 5.2"},{"comment":"The asterisked entries in Table 2 are intervals of best known bounds, but the caption does not say so. Please state explicitly that entries marked * are not exact values.","section":"Table 2"},{"comment":"In the cohomology computation, write the tensor product over Z_2 and use H^*(RP^{d-1};Z_2) to avoid ambiguity about coefficients.","section":"Theorem 3.2 proof"}],"recommendation":"major_revision","confidential_remarks":"The main theorems appear quite likely to be correct, and the flaws identified above are local and repairable; this is not a rejection. The self-citation [20] is to an unpublished manuscript, but since it is not needed for the present proofs, it does not affect the assessment. Please ensure the revision reconciles the statement in Section 4 with Remark 14 and completes the proof of Theorem 2.2."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: this is a real advance. The closed form N'(d)=2d−1 and the asymptotic N(d)~2d answer Chipot's growth question; the Hankel subspace gives a dimension-optimal Korn inequality; and the martingale-laminate machinery yields quantitative Ornstein-style counterexamples and sharp constants. The algebraic characterization (Theorem 1.2) is honestly credited as previously known, but its systematic exploitation is new. Anyone working in calculus of variations or PDE should look at this.\n\nThe paper does several things well. The lower bound N'(d)≥2d−1 is sound: if dim_R X<2d−1, the complexified complement contains a rank-one matrix, which gives an infinite-dimensional null space and contradicts compactness. The asymptotic N(d)~2d is built on established immersion bounds and is well sourced. The Hankel-polynomial interpretation and the constant estimate log(C(H(d))) ~ (2G/π)d are neat. The paper is also honest about limitations: Remark 14 states correctly that C-ellipticity requires X^⊥ to avoid rank-one matrices.\n\nNow the soft spot, and it is load-bearing. The stress-test note is right. In the Section 4 proof of N'(d)≤g_C(d), the text says: let X be a subspace of M_d(C) such that X avoids all rank-one matrices, and then invokes [48, Theorem 4] for T_X=P_X(∇u). But C-ellipticity for this operator requires the orthogonal complement X^⊥ to avoid rank-one matrices, not X itself. \"X avoids rank-one\" is neither necessary nor sufficient; for example, in d=2, the span of a nonzero skew-symmetric matrix avoids rank-one while its complement contains rank-one matrices and the projected operator is not elliptic. This is not a fatal flaw, because the intended subspace H(d) does satisfy the correct condition — if P_H(a⊗b)=0, the corresponding polynomial product vanishes identically, forcing a=0 or b=0 — so the theorem is very likely correct. But the proof as written is internally inconsistent and must either correct the quantifier to \"X^⊥ avoids rank-one\" or explicitly exhibit H(d) and verify C-ellipticity before applying [48, Theorem 4]. Since this upper bound is exactly what gives N'(d)=2d−1, the revision is substantive, not cosmetic.\n\nOther issues are minor. The unpublished self-citation [20] is attributed as the origin of Theorem 2.2 and should not carry that weight; Remark 5's pointer to [42, Theorem 3.21] looks like a mis-citation; and Theorem 2.2's sharpness part is only sketched. None of these affect the main theorems except the Section 4 issue above.\n\nBottom line: this paper deserves a serious referee. The central claims are probably correct, but the proof of the central upper bound needs a real correction. I would engage with it, and I would cite it once the revision is out.","headline":"Sharp results on Chipot's Korn-measurement problem, with a clean algebraic reformulation and a genuinely useful martingale/laminate toolkit — but the proof of the flagship upper bound N'(d)≤2d−1 is written with the wrong ellipticity hypothesis and needs a revision.","tokens_in":25728,"tokens_out":2993,"would_cite":true,"duration_ms":30219,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35A23","49J45"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that the second Korn inequality needs exactly 2d−1 scalar measurements of the gradient, and that the first Korn inequality needs asymptotically 2d measurements, resolving a question posed in 2021.","keywords":["Korn inequalities","rank-one convexity","quasiconvexity","laminates","martingales","Hankel matrices","minimal measurements","L1 non-inequality"],"falsifier":"Compute N'(3) directly: the paper predicts 5. A single four-functional family of linear measurements for which the H1-control inequality holds on a cube would refute the lower bound N'(d)≥2d−1; likewise, a domain that is not John where the Hankel inequality still holds would test the reach of the representation theorem.","tokens_in":24561,"feed_emoji":"🎲","tokens_out":6256,"duration_ms":60428,"temperature":0.7,"pith_summary":"This paper settles a question about Korn's inequalities: how many scalar measurements of the gradient are enough to recover a Korn-type estimate on the whole gradient? The author shows that for the second Korn inequality the answer is exactly 2d−1 in dimension d, and for the first Korn inequality the answer grows as 2d asymptotically. The proof converts the problem into an algebraic statement about subspaces of matrices that avoid rank-one matrices, and then uses a systematic correspondence between laminates and martingales to build sharp examples. A concrete payout is a dimension-optimal Korn inequality: the full H1 norm of a vector field is controlled by its L2 norm plus the L2 norm of the projection of its gradient onto the (2d−1)-dimensional space of Hankel matrices. A sympathetic reader would care because this shows that the gradient norm is far more redundant than the standard symmetric-part control suggests, and it opens a martingale-based toolbox for sharp constants in calculus of variations.","feed_headline":"2d−1 measurements are exactly enough to control gradients","feed_subtitle":"Only the Hankel part of the gradient is truly needed, so the gradient norm is far more redundant than symmetric-part control suggests.","key_machinery":"The load-bearing objects are, on the algebraic side, the rank-one-avoiding subspaces of matrix space, and on the analytic side, a systematic correspondence between finite-order laminates and dyadic martingales. A laminate is a probability measure on matrix space obtained by repeated rank-one splitting, equivalently a dyadic martingale whose increments have rank one. The paper shows that the rank-one convex envelope of functions of the form Cp∥PX∥p−∥QX∥p is nonnegative exactly when X⊥ contains no rank-one matrices, and it uses the sharp subordination inequality for Hilbert-space martingales to compute the optimal constant p*−1. The Hankel matrices H(d) then provide the explicit dimension-mini","core_discovery":"The central discovery is an exact algebraic characterisation of when a Korn-type inequality holds: a subspace X of d×d matrices gives the L2 control ∥∇u∥ ≲ ∥PX(∇u)∥ exactly when its orthogonal complement contains no non-zero rank-one matrix. This reformulates the minimal numbers as N(d)=d2−nR(d,1) and N'(d)=d2−nC(d,1), the codimensions of the largest real/complex subspaces avoiding rank-one matrices. From this, the paper derives the exact value N'(d)=2d−1, the asymptotic N(d)∼2d (with N(d)≤2d−2 and equality infinitely often), and the inequality N(d)=d only for d∈{1,2,4,8}. The explicit extremal subspace is the space of Hankel matrices—matrices constant on each skew-diagonal—whose dimension i","pith_inferences":["The algebraic characterisation suggests the hard part of N(d) for each d is a purely topological question about immersion dimensions of real projective spaces; a closed form for N(d) would follow from an unresolved equality between the algebraic minimum and the immersion dimension, which the paper leaves open.","The sharp constant conjecture in the paper—that the Lp Hankel-Korn constant factorises as the L2 constant times p*−1—would, if true, connect the problem to a long-standing conjecture about the quasiconvexity of certain rank-one convex envelopes; the martingale method already proves the rank-one-convex side.","The same martingale–laminate correspondence could be pushed to higher-order operators and to rectangular gradients, as the paper begins, potentially yielding dimension bounds for general linear operators beyond first order."],"forward_implications":["Second Korn inequality: exactly 2d−1 scalar linear measurements of ∇u suffice (and are necessary) to control the H1 norm on Lipschitz domains, for every d.","First Korn inequality: the minimal number N(d) satisfies N(d)∼2d, with N(d)≤2d−2 and equality for d=2n+1 infinitely often; N(d)=d only for d∈{1,2,4,8}.","The Hankel projection gives a concrete dimension-optimal Korn inequality: ∥u∥H1 ≲ ∥u∥L2 + ∥PH(∇u)∥L2.","The classical L1 non-inequality is quantified: any first-order homogeneous operator with a rank-2 matrix in its kernel fails L1 control, and the optimal constant in the Lp version grows at least like p*−1."],"fun_headline_variants":["Exact minimum: 2d−1 measurements for gradient control","Hankel matrices reveal sharp Korn bounds","Martingale-laminate method answers Chipot's question","Korn-type inequalities: best possible measurement count found","Algebraic test for Korn control: no rank-one in complement"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The exact value N'(d)=2d−1 depends on a pointwise representation theorem for operators whose symbol avoids complex rank-one matrices; if that representation fails (e.g., on domains that are not John domains), the dimension-optimal Hankel inequality for the second Korn inequality would collapse.","fun_headline_variants_meta":{"raw":{"variants":["Exact minimum: 2d−1 measurements for gradient control","Hankel matrices reveal sharp Korn bounds","Martingale-laminate method answers Chipot's question","Korn-type inequalities: best possible measurement count found","Algebraic test for Korn control: no rank-one in complement"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000333,"raw_usage":{"total_tokens":1806,"prompt_tokens":983,"completion_tokens":823,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":727,"completion_tokens_details":{"reasoning_tokens":743}},"tokens_in":727,"tokens_out":823,"duration_ms":8856,"temperature":1.0,"reasoning_tokens":743,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T18:55:00.733546+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute N'(3) directly: the paper predicts 5. A single four-functional family of linear measurements for which the H1-control inequality holds on a cube would refute the lower bound N'(d)≥2d−1; likewise, a domain that is not John where the Hankel inequality still holds would test the reach of the representation theorem.","supporting_citations":[],"review_version":1}