{"id":"30af8090-465e-4957-98b5-4a4f3d698764","arxiv_id":"2510.00737","paper_version":3,"verdict":"REJECT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"High-order (C^{k,1}) large-scale regularity is established for high-contrast stochastic elliptic homogenization, along with a non-iterative Caccioppoli inequality.","lead":"This paper claims a new large-scale regularity theorem for solutions of elliptic equations with random, high-contrast coefficients — where the contrast ratio may be infinite — and proves a new non-iterative Caccioppoli inequality. If correct, it would give high-order approximation by harmonic polynomials in a setting where only lower-order estimates were previously known.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The WLOG k=0 reduction in Assumption 1.1 and Step 2 of Theorem 4.1 is false for nonconstant antisymmetric coefficients: div(k∇u) contributes a nonzero first-order term, so the proof does not cover the advertised non-symmetric case.","rationale":"The reader identified the load-bearing weakness correctly: the proof reduces to k=0 by claiming the solution sets are unchanged, and this is false for nonconstant antisymmetric matrices. My own reading confirms the claim appears verbatim in Assumption 1.1 and is used in Step 2 of Theorem 4.1 as well as in Lemma 2.3. The explicit computation with periodic k shows the two operators differ by a first-order term. This is not merely a missing technical detail; it is the mechanism by which the proof transfers all regularity information from the homogenized symmetric equation to the original non-symmetric equation. The theorem may be salvageable through a more delicate argument treating ∇·k·∇u as a lower-order perturbation, but that argument is absent. The paper's novel Caccioppoli inequality may stand independently, but the main theorem as stated is unsupported. Therefore I do not change the reader's REJECT verdict.","tokens_in":32568,"tokens_out":6230,"duration_ms":62204,"concrete_test":"Analyze the explicit periodic example in dimension 2: a(x)=I+ε sin(2πx_1)J, with s=I and \\bar{a}=I (or \\bar{s}=I, \\bar{k}=0). Directly compute the solution spaces A_1 and \\bar{A}_1: the reduction in Step 2 predicts x_1 ∈ A_1, but -div(a∇x_1)=2πε cos(2πx_1)≠0, so x_1 is not a-solution. More decisively, solve the cell problem for linear-growth solutions with this k: the corrector equation contains the term ∇·k·∇(x_1+φ) and does not reduce to the k=0 equation. If the proof's reduction were valid, the corrector would be zero; it is not. This single computation settles that the WLOG k=0 step is invalid and that Theorem 4.1's proof does not cover nonconstant antisymmetric coefficients.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central proof of Theorem 4.1 depends on the reduction in Assumption 1.1 (p.3) and Step 2 (p.22) that \"we can universally assume without any loss of generality that k=0,\" justified by subtracting k from both a and \\bar{a} \"without altering the set of solutions.\" This is false for the nonconstant antisymmetric fields allowed by the paper. For smooth u, div(k∇u) = (∇·k)·∇u + k:D^2u, and while k:D^2u = 0 by antisymmetry, the first-order term (∇·k)·∇u is generically nonzero. Thus -div((s+k)∇u)=0 is not equivalent to -div(s∇u)=0. Concretely, in d=2 take s=I and k(x)=ε sin(2πx_1)J with J=[[0,1],[-1,0]]. Then -div((I+k)∇x_1) = -div(e_1 + ε sin(2πx_1)J e_1) = 2πε cos(2πx_1) ≠ 0, so the harmonic linear function x_1 is not a solution of the a-equation. The same false identification is repeated in Lemma 2.3 and in the notation discussion after (2.10). Because Steps 2.1–2.5 and the dimension count (4.4) are all carried out for the reduced symmetric equation, the advertised result for non-symmetric high-contrast coefficient fields is not established. The proof would need to control the additional first-order term ∇·k·∇u throughout the induction, which is not done.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper aims to prove a global high-order regularity theorem (Theorem 4.1) for high-contrast elliptic homogenization with possibly non-symmetric coefficient fields, alongside a non-iterative Caccioppoli inequality (Proposition 3.2). The main theorem asserts that, on large scales, any solution of polynomial growth is approximated by a homogenized harmonic polynomial with error O((X_H/3^n)^{θ/2}), and conversely, and that the dimensions of the solution spaces A_k and \\bar A_k agree. The proof follows the induction scheme of [AKM19]/[AK24], adapted to the high-contrast framework. The Caccioppoli inequality proof is presented as an independent contribution, and the paper is largely a reworking of the author's advisor's group framework [AK25].","tokens_in":33054,"tokens_out":4254,"duration_ms":35203,"significance":"If valid, the main theorem would be a significant extension of high-order regularity theory to high-contrast homogenization, with the dimension identity and large-scale polynomial approximation being natural and useful tools. The paper's non-iterative Caccioppoli inequality (Prop. 3.2) is a useful technical contribution in its own right, and the proof structure is careful and well-motivated. However, the advertised main result is not established because a central reduction (k=0 without loss of generality) is false for nonconstant antisymmetric parts. The proof therefore covers only the symmetric-coefficient case, despite the theorem being stated for non-symmetric matrices. This is a load-bearing gap that cannot be repaired by minor adjustments; the first-order term ∇·k·∇u would need to be controlled throughout the induction, which is not done.","major_comments":[{"comment":"The assertion that one can assume k=0 'without loss of generality' because subtracting k leaves the solution set unchanged is false for nonconstant antisymmetric k. For smooth u, ∇·(k∇u) = (∇·k)·∇u + k:D²u, and k:D²u=0, so the first-order term (∇·k)·∇u is generically nonzero. Example: d=2, s=I, k(x)=ε sin(2πx₁)[[0,1],[-1,0]]. Then -div((I+k)∇x₁)=2πε cos(2πx₁)≠0, so x₁ solves -Δu=0 but not -div((I+k)∇u)=0. Since Theorem 4.1 is stated for coefficient fields with non-symmetric matrices, the proof's reduction to a=s is invalid and the main theorem is not proven as stated.","section":"Assumption 1.1, p.3; Step 2 of Thm 4.1, p.22"},{"comment":"The same false identification is used in Lemma 2.3 ('We can also set k=0 as established in Chapter 1') and in the notation discussion after (2.10). Lemma 2.3 is applied in Steps 2.2–2.3 to estimate terms involving a∇w_j and a∇φ_j. Without an estimate for the additional first-order term coming from the antisymmetric part, the induction step from k−1 to k in Theorem 4.1 fails for non-symmetric a. The proof would need to propagate a bound on ∇·k·∇u through the harmonic approximation and the corrector argument, which is absent.","section":"Lemma 2.3 and discussion after (2.10)"},{"comment":"The appendix states that the change of variables u(x)=v(q₀^{-1}x) reduces (A.4) to the Laplace equation in Euclidean geometry. This is not generally true for non-symmetric a, and even for symmetric s it requires a precise choice of q₀ making q₀ᵀs q₀ a multiple of the identity. The formula defining q₀ in (2.5) is also garbled and not a definition. Since the spherical-harmonics estimates in Steps 3.2–3.4 and the dimension count in Step 6 rely on this reduction, the claimed properties of harmonic polynomials in the adapted geometry need a rigorous justification.","section":"Appendix A, p.35"}],"minor_comments":[{"comment":"The definition of q₀ is illegible: 'pq0qij := 3^{-k0} Q_{3^{k0}} |s^{-1}|^{1/2} (s^{1/2})_{ij}' leaves the objects Q and k₀ unexplained, and the expression is not a standard matrix definition.","section":"Eq. (2.5)"},{"comment":"The proof imports many parameters (Π, Θ, K_ΨS, m*, etc.) from [AK25, Cor. 4.3] without stating their definitions; this makes the verification of (2.3) hard to follow for a reader without the companion paper.","section":"Prop 2.1 proof"},{"comment":"The claim A₀ = Ā₀ is asserted without proof. A Liouville-type statement that all a-harmonic functions with sublinear growth are constant is not automatic for degenerate high-contrast coefficients and should be justified or cited precisely.","section":"Step 1 of Thm 4.1"},{"comment":"The notation \\bar a, \\bar s, and \\bar A is used in the display but the relationship to the homogenized matrix A in Assumption 1.1 is not stated clearly; the reader must infer that \\bar a = s and \\bar s = s.","section":"Notation in Thm 1.3 and (1.11)"}],"recommendation":"reject","confidential_remarks":"The paper's main theorem is not proven for the stated generality. The WLOG k=0 reduction is a genuine mathematical error, not a mere presentation issue. The paper could possibly be salvaged by restricting the main theorem to symmetric coefficient fields or by adding a substantial new argument controlling the first-order term ∇·k·∇u, but this would be a significant change of scope. The reliance on v2-specific lemmas from [AK25] is not itself disqualifying."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe thing to know: this paper announces the first high-order large-scale regularity theorem for high-contrast stochastic homogenization with non-symmetric coefficients, but the proof's central reduction — subtract the antisymmetric part k \"without altering the set of solutions\" — is false for nonconstant k. The stress-test note is right: for smooth u, div(k∇u) = (∇·k)·∇u, which is generically nonzero, so -div((s+k)∇u)=0 and -div(s∇u)=0 are different equations. The concrete d=2 example with s=I and k = ε sin(2πx_1) J kills the reduction: x_1 is a solution of the s-equation but not of the a-equation. The same step appears in Assumption 1.1, in the note after (2.10), and in Lemma 2.3, and the whole induction in Theorem 4.1 for non-symmetric fields is built on it. So Theorem 4.1 is not established in the advertised generality.\n\nThat said, the paper is not empty. The non-iterative Caccioppoli inequality in high contrast (Proposition 3.2) is new and looks like a real tool; it removes the energy norm from the right-hand side, which the iterative version in [AK25] does not do. The proof is independent of [AK25, Prop 2.5] and is a solid contribution in its own right, at least under the symmetric-coefficient interpretation. The high-order regularity argument for the symmetric case follows [AK24]/[AKM19] with serious modifications — adapted geometry, harmonic polynomials, excess decay — and if the k=0 issue were either removed by restating the theorem for a=s or fixed by controlling the first-order term, the proof may well be salvageable for the symmetric case.\n\nSoft spots in proportion: the δ parameter is dropped from the homogenization-error estimate without comment in the proof (the text admits it), and Lemma 3.1 cites \"Lemma 6.2 in v2\" of [AK25], which is not a stable reference. These are minor compared to the load-bearing false reduction. The paper also uses k both for the antisymmetric part and for the polynomial degree, which is confusing but not fatal.\n\nWho this is for: an expert in stochastic homogenization who wants the high-contrast machinery and a possible path to high-order regularity. The central theorem is currently unproven, so take the advertised non-symmetric statement with a grain of salt. It deserves a serious referee — the flaw is subtle and the symmetric-case core is potentially valuable — but as it stands, the title result is not supported.\n\nRecommendation: send to peer review, expect rejection or major revision, and ask the author to either prove the full non-symmetric case or restrict the main theorem to symmetric coefficients.","headline":"The paper has a genuine new Caccioppoli inequality and a plausible high-order regularity proof, but the advertised non-symmetric result rests on a false \"k=0 wlog\" reduction, so the main theorem as stated is unproven.","tokens_in":33446,"tokens_out":2697,"would_cite":false,"duration_ms":33472,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35B27","35J15","35B65"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that in high-contrast elliptic homogenization, every solution is approximated on all large scales by a homogenized polynomial of any fixed degree, with an explicit error controlled by the random minimal scale.","keywords":["high-contrast homogenization","stochastic homogenization","large-scale regularity","Caccioppoli inequality","high-order regularity","non-symmetric coefficients","harmonic polynomials","elliptic PDE"],"falsifier":"Take d=2 and a(x)=I+[[0,x_1],[-x_1,0]]. Then −∇·(a∇u)=0 is equivalent to −Δu+∂_2 u=0, whereas the symmetrized equation is −Δv=0; the solution sets differ, e.g. u(x)=e^{x_2} solves the first but is not harmonic. Checking whether the asserted polynomial approximation with rate (X_H/3^n)^{θ/2} holds for this coefficient field would settle whether the stated generality is real.","tokens_in":32498,"feed_emoji":"📐","tokens_out":6190,"duration_ms":54157,"temperature":0.7,"pith_summary":"The paper sets out to extend the classical large-scale regularity theory of stochastic homogenization — where coefficients are uniformly elliptic — to high-contrast fields whose ellipticity ratio is unbounded and whose coefficient matrices need not be symmetric. Its central claim is a global high-order regularity theorem: on every sufficiently large adapted cube, a solution lies within an explicit error O((X_H/3^n)^{θ/2}) of some k-th-order homogenized polynomial, and conversely every such polynomial has a nearby true solution. Along the way, it proves a non-iterative Caccioppoli inequality that bounds the energy of a solution on an inner cube directly by its L² norm on the outer cube, with no ellipticity ratio in the constant. If true, this gives the high-contrast setting the same quantitative polynomial structure that makes moderate-contrast homogenization useful, including an explicit dimension formula for the spaces of solutions with polynomial growth.","feed_headline":"High-contrast PDEs get polynomial regularity on large scales","feed_subtitle":"A new Caccioppoli estimate controls degenerate coefficients by homogenized polynomials, with explicit error.","key_machinery":"The load-bearing mechanism is the pair (coarse-grained matrices, adapted geometry). The coarse-grained matrix A(U) encodes the effective conductivity of a large block U, and Assumption 1.1 says that on all scales above the random threshold X_H these matrices are controlled by a fixed homogenized limit A up to a tolerance (X_H/3^n)^θ. Around this limit the paper constructs adapted cubes ♢_n = q_0((-1/2·3^n, 1/2·3^n)^d), with q_0 chosen from the homogenized symmetric part, so that the high-contrast equation becomes a Laplace-type equation after rescaling. The proof is carried by a new non-iterative Caccioppoli inequality (Proposition 3.2), which bounds the adapted energy on an inner cube by th","core_discovery":"On the paper's own terms, the discovery is that polynomial regularity survives degenerate, non-symmetric coefficients once the geometry is adapted to the homogenized limit. Theorem 4.1 asserts that, for every integer k, the space A_k of solutions with sub-(k+1)-growth is quantitatively isomorphic to the space of k-th-order homogenized harmonic polynomials: each side approximates the other with error controlled by (X_H/3^n)^{θ/2} in the relevant weighted norms. In particular, dim A_k equals the classical binomial dimension C(d+k-1,k)+C(d+k-2,k-1) of harmonic polynomials of degree at most k. The claim is not merely a qualitative Liouville theorem but a global, quantitative approximation valid","pith_inferences":["Editorial inference: the proof's reduction 'assume k=0' — subtracting the antisymmetric part of the coefficient field — is valid only for constant antisymmetric parts; for non-constant k, ∇·(k∇u) is generally a nonzero first-order term and changes the set of solutions. If so, Theorem 4.1 as stated is established only for symmetric coefficient fields.","Editorial inference: the adapted geometry suggests that the natural notion of regularity in high contrast is measured in the q_0 metric, not the Euclidean one; phrasing the C^{k,1} estimate entirely in q_0-balls would likely be the right framework for local and boundary versions.","Editorial inference: since the constants depend on X_H only through the threshold and its tail, one can read off almost-sure convergence rates; a natural test is to compare the theorem's prediction with explicit two-scale examples where X_H is large.","Editorial inference: a local, finite-scale version — stated by the author as future work — would follow by truncating the Haar-measure integrations at a stopping scale; the global theorem already implies such a local version at scales above that cutoff."],"forward_implications":["Quantitative large-scale regularity: on every scale 3^n ≥ X_H, solutions of the equation are within an explicit error of homogenized polynomials of any fixed degree k, with rate (X_H/3^n)^{θ/2}.","Dimension counting: the space of true solutions with polynomial growth of order ≤ k has dimension C(d+k-1,k)+C(d+k-2,k-1), so no polynomial orders are lost to degeneracy.","The non-iterative Caccioppoli estimate removes the energy norm from the right-hand side, so future high-contrast regularity proofs do not require infinite iteration over scales.","Both approximation directions hold, so the homogenized limit faithfully represents the full solution space at every polynomial order.","Because the approximation is global from the random scale upward, it supplies a missing ingredient for optimal quantitative estimates in high-contrast homogenization."],"fun_headline_variants":["High-contrast PDEs: polynomial regularity with explicit error","Global polynomial regularity for high-contrast homogenization","Caccioppoli inequality yields high-order regularity in high contrast","Sub-polynomial solutions match homogenized harmonic polynomials"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is the claim that one may assume k=0 without loss of generality — that subtracting the antisymmetric part of the coefficient matrix from both a and the homogenized matrix leaves the set of solutions unchanged; for a non-constant antisymmetric part, ∇·(k∇u) is generally a nonzero first-order term, so this is not true.","fun_headline_variants_meta":{"raw":{"variants":["High-contrast PDEs: polynomial regularity with explicit error","Global polynomial regularity for high-contrast homogenization","Caccioppoli inequality yields high-order regularity in high contrast","Sub-polynomial solutions match homogenized harmonic polynomials"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001179,"raw_usage":{"total_tokens":4617,"prompt_tokens":560,"completion_tokens":4057,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":304,"completion_tokens_details":{"reasoning_tokens":3991}},"tokens_in":304,"tokens_out":4057,"duration_ms":28625,"temperature":1.0,"reasoning_tokens":3991,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-04T13:23:59.783583+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take d=2 and a(x)=I+[[0,x_1],[-x_1,0]]. Then −∇·(a∇u)=0 is equivalent to −Δu+∂_2 u=0, whereas the symmetrized equation is −Δv=0; the solution sets differ, e.g. u(x)=e^{x_2} solves the first but is not harmonic. Checking whether the asserted polynomial approximation with rate (X_H/3^n)^{θ/2} holds for this coefficient field would settle whether the stated generality is real.","supporting_citations":[],"review_version":1}