{"id":"6ed414c5-c304-4668-8946-94c8f422852f","arxiv_id":"2412.15026","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The authors characterize matrix weights for which all multilinear Calderón-Zygmund operators are bounded, showing equivalence with a multilinear Muckenhoupt condition and with boundedness of a tensor-product maximal operator.","lead":"This mathematics paper proves the full set of conditions on matrix-valued weights under which multilinear singular integral operators are bounded on weighted spaces, unifying multilinear and matrix weight theories. A generalist might care because it settles a natural open problem in harmonic analysis and introduces convex-geometric tools with potential use beyond this setting.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Definition 6.12's non-degeneracy condition forces the kernel to be constant; the Riesz example does not satisfy it, leaving the necessity direction of Theorem A unproved.","rationale":"The reader identified Proposition 2.4, the reducing-operator construction for quasinorms, as the weakest assumption. That proposition appears essentially correct: the John ellipsoid and Carath\\'eodory argument give the stated dimension-dependent approximation, and the application to averaging operators is coherent. The genuine load-bearing problem is elsewhere. The necessity direction of Theorem A is proved by exhibiting one directionally non-degenerate multilinear CZ operator, namely a multilinear Riesz transform, and applying Theorem 6.13. Definition 6.12(b) is quantifier-heavy: C and Q' are fixed, and then for every \\alpha\\in(0,1) a bounded error kernel S must exist. Because the identity must hold for all vector-valued f supported in Q, it is a pointwise kernel identity. The bound |S|\\le|Q|^{-m} then forces the normalized kernel |Q|^m C K to lie in the interval [1,(1+\\alpha)/(1-\\alpha)] for every \\alpha simultaneously. Sending \\alpha to zero forces the kernel to be constant, which is impossible for the Riesz kernel. The example's proof of |S|\\le|Q|^{-m} only uses K\\ge C^{-1}|Q|^{-m} and ignores the necessary upper bound; for the chosen Q' and small \\alpha the inequality is simply false. Since Theorem 6.13 and hence (i)\\Rightarrow(iii) depend on this example, the central claim is not yet fully supported. A repaired non-degeneracy condition, for instance requiring the identity only for one sufficiently large \\alpha chosen after fixing the constants in Theorem 6.16, or allowing Q' to depend on \\alpha in a way that makes the kernel variation small, could likely salvage the argument, but the current text does not contain it. The appropriate verdict is therefore conditional on a corrected non-degeneracy proof, not unconditional acceptance.","tokens_in":62309,"tokens_out":44093,"duration_ms":378643,"concrete_test":"Test Example 6.19 as written: take m=1, d=2, Q=[0,1]^2, Q'=[2,3]\\times[0,1], and the Riesz kernel K(x,y)=(x_1-y_1)/|x-y|^3. Compute m_0=\\inf_{(x,y)\\in Q'\\times Q}K(x,y) and M=\\sup_{(x,y)\\in Q'\\times Q}K(x,y). If M/m_0>3, then no scalar C satisfies the necessary interval condition 1\\le |Q|^m C K(x,y)\\le 3 for \\alpha=1/2, so property (b) fails for this Q'. More generally, verify that Definition 6.12(b) implies |Q|^m C K\\equiv1 by letting \\alpha\\to0; since no nonconstant kernel satisfies this, the non-degeneracy definition must be weakened or the proof of Theorem 6.13 must be revised.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The necessity implication (i)\\Rightarrow(iii) of Theorem A is proved via Theorem 6.13, which applies only to 'directionally non-degenerate' operators. Definition 6.12(b) requires: for a fixed C and Q', for every \\alpha\\in(0,1) there is S with |S|\\le|Q|^{-m} such that for all f supported in Q, \\otimes_j\\langle f_j\\rangle_Q = (1-\\alpha)C\\,\\tilde T(f)(x) + \\alpha\\int S(x,y)\\otimes_j f_j(y_j)\\,dy on Q'. Since x\\notin\\operatorname{supp}f, the kernel representation applies, and the identity forces the pointwise equation (1-\\alpha)C K(x,y)+\\alpha S(x,y)=|Q|^{-m}. Hence 1\\le |Q|^m C K(x,y)\\le (1+\\alpha)/(1-\\alpha) for a.e. (x,y). Letting \\alpha\\to0 forces |Q|^m C K to be identically 1 on Q'\\times Q^m. No nonconstant Calder\\'on-Zygmund kernel can satisfy this. For the Riesz kernel in Example 6.19 with Q'=Q+2m\\ell(Q)e_1, K is strictly positive on Q'\\times Q^m but not constant; for example when m=1, d=2, Q=[0,1]^2, Q'=[2,3]\\times[0,1], the ratio \\sup K/\\inf K is about 10.5, so for \\alpha=1/2 the required interval [1,3] cannot contain the normalized kernel values. The displayed verification of |S|\\le|Q|^{-m} checks only a lower bound and incorrectly drops the absolute value and upper bound. Thus the advertised directional non-degeneracy of the Riesz transform is not established, and the proof of (i)\\Rightarrow(iii) has a genuine gap.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a theory of multilinear matrix weights in the tensor product of finite-dimensional Hilbert spaces and claims a full Hunt–Muckenhoupt–Wheeden-type characterization: for exponents p_j ∈ (1,∞] with 1/p = Σ 1/p_j, the boundedness of all multilinear Calderón–Zygmund operators on the matrix-weighted spaces is equivalent to boundedness of a tensor-product convex-body maximal operator and to a multilinear matrix Muckenhoupt condition A_⃗p defined by uniform boundedness of averaging operators. The main technical contributions are reducing operators for quasinorms via the John ellipsoid and Carathéodory theorems, a Roudenko-type characterization of the A_⃗p class, reverse Hölder/Fujii–Wilson estimates, strong-type bounds for tensor-product convex-body maximal operators, sparse convex-body domination for multilinear Calderón–Zygmund operators, and a new notion of directional nondegeneracy used to prove the necessity direction of Theorem A. The paper also states quantitative bounds for sparse and maximal operators in terms of the multilinear characteristic.","tokens_in":62652,"tokens_out":9670,"duration_ms":78241,"significance":"If the main theorem is correct, this is a substantial advance: it extends the scalar multilinear Muckenhoupt theory and the linear matrix-weight theory into a genuinely multilinear matrix-weight setting, including the quasi-Banach range p<1, and it introduces useful machinery (quasinorm reducing operators, convex-body sparse domination, directional nondegeneracy) that is likely to be reused. The paper is detailed and self-contained in many places: the tensor algebra in Section 3, the reduction of the maximal operator bounds in Section 5, and the sparse domination in Section 6 are carefully developed. The quantitative bounds recover known sharp linear results. However, the proof of the necessity direction of the main theorem rests on a nondegeneracy condition that, as stated, appears to be unsatisfiable by any nonconstant Calderón–Zygmund kernel; this is a load-bearing gap that prevents the paper from establishing its central characterization in its current form.","major_comments":[{"comment":"The directional non-degeneracy condition in Definition 6.12(b) forces the kernel to be constant on Q' × Q^m, so it is empty for nonconstant Calderón–Zygmund kernels. Indeed, for f_j supported in Q and x ∈ Q', the kernel representation of Remark 6.5 applies, and the identity stated in (b) becomes, for all f_j, ∫_{Q^m} [(1−α)C K(x,y) + α S(x,y) − |Q|^{-m}] ⊗_j f_j(y_j) dy = 0. Since the f_j are arbitrary, the scalar factor vanishes a.e.; the bound |S(x,y)| ≤ |Q|^{-m} then gives |Q|^m C K(x,y) ∈ [1, (1+α)/(1−α)] for every α ∈ (0,1). Letting α → 0 forces C K(x,y) = |Q|^{-m} for a.e. (x,y) ∈ Q' × Q^m. No nonconstant Calderón–Zygmund kernel can satisfy this. Consequently Theorem 6.13 has no non-vacuous hypotheses, and the proof of implication (i)⇒(iii) of Theorem A given in Section 7 is not valid.","section":"Section 6.4, Definition 6.12(b)"},{"comment":"The verification that the multilinear Riesz transform is directionally non-degenerate is invalid for two reasons. The displayed computation proves only |Q|^{-m} − (1−α)C_{m,d}K(x,y) ≤ α|Q|^{-m}, which is a one-sided bound; since the left-hand side can be negative, this does not imply |S(x,y)| ≤ |Q|^{-m}. Moreover, the missing two-sided bound cannot hold: for m = 1, d = 2, Q = [0,1]^2 and Q' = [2,3] × [0,1], the normalized kernel C_{1,2}K takes values from 3/10 to √10, while for α = 1/2 the required interval is [1,3]. Thus the Riesz transform is not shown to satisfy Definition 6.12(b), and the only advertised example of a directionally non-degenerate operator does not work.","section":"Example 6.19"}],"minor_comments":[{"comment":"There is a typo in 'noncommutavity'; it should read 'noncommutativity'.","section":"Abstract and Section 1"},{"comment":"The notation for coordinates of y_j is inconsistent: the text writes y_j = (y^1_j, ..., y^m_j), but since y_j ∈ R^d the last index should run to d, not m.","section":"Example 6.19"},{"comment":"The product of directional Banach function spaces is defined only for matrix-weighted spaces in Definition 2.2, but Proposition 3.1 and elsewhere use the notation X_1 × ... × X_m as if it were a general construction; a clarifying remark could prevent confusion.","section":"Definition 2.2 and Proposition 3.1"},{"comment":"The reference [Ler24] is listed as 'Published online (early view)' without complete bibliographic data; please provide the full citation.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The paper is substantial and the positive direction of Theorem A is developed in convincing detail, but the necessity direction has a genuine gap: the directional nondegeneracy condition in Definition 6.12(b) is too rigid and forces the kernel to be constant, and the Riesz-transform example does not satisfy it. This blocks acceptance as is. The error is localized to Section 6.4, and a revision that relaxes condition (b) to a suitable inequality or otherwise repairs the proof of (i)⇒(iii) could make the central claim valid. The reader's report appears overly optimistic in rating soundness 8; the skeptic's objection is correct."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"I read the paper carefully after seeing your note, and I think the stress-test concern lands. The necessity direction of Theorem A, (i)⇒(iii), is proved through Theorem 6.13, which requires a directionally non-degenerate operator. Definition 6.12(b) is very strong. For each α it gives an identity whose kernel is (1−α)CK + αS equal to |Q|−m on Q′×Q^m. Since |S|≤|Q|−m pointwise, this forces (1−α)CK to lie between (1−α)|Q|−m and (1+α)|Q|−m. Because the identity must hold for every α, letting α→0 pins CK to be exactly |Q|−m a.e. on Q′×Q^m. So K has to be constant. No nonconstant Calderón–Zygmund kernel can satisfy that. The paper's Riesz example is the only illustration, and its verification of |S|≤|Q|−m is wrong: it computes α|S| = |Q|−m − (1−α)CK and drops the absolute value, assuming the expression is nonnegative. It need not be. Concretely, with Q=[0,1]^2, Q′=[2,3]×[0,1], the normalized Riesz kernel takes values spanning a ratio of about 10, so for α=1/2 the required interval [1,3] cannot contain them. The example fails.\n\nThat is a load-bearing flaw, not a cosmetic typo. As stated, Theorem A is not proved. However, the rest of the paper is genuinely good. The reducing operators for quasinorms via John ellipsoid plus Carathéodory look correct, and the characterization of averaging operators built on them is a solid piece of work. The convex body sparse domination for multilinear CZ operators with Dini kernels appears new and carefully done, and the quantitative bounds in Theorem B, while not sharp, are a real contribution. The sufficiency direction (iii)⇒(i) and the maximal operator results stand on their own. The paper gives proper credit to the prior literature and the structure is coherent up to the nondegeneracy section.\n\nWho gets value from this? Harmonic analysts working on matrix weights and multilinear theory. The sparse domination and reducing-operator machinery are worth studying even if the main theorem needs repair. I do not share the reader's ACCEPT in the current form. I would send it to peer review, but with a clear demand: fix the nondegeneracy condition—maybe allow a constant factor or an approximate identity that degrades gracefully—or find another route to necessity. The sufficiency half could be publishable separately. As it stands, the claim of a full characterization is not supported.","headline":"The necessity direction of Theorem A has a genuine gap: the directional nondegeneracy condition forces the kernel to be constant, and the Riesz example's verification drops the absolute value; the sufficiency side is substantial.","tokens_in":63215,"tokens_out":6082,"would_cite":false,"duration_ms":49421,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["42B20","46E25","46E30"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper establishes a full multilinear matrix weighted analogue of the Hunt–Muckenhoupt–Wheeden theorem: for any tuple of exponents, a tuple of matrix weights is admissible for all multilinear Calderón–Zygmund operators exactly when a…","keywords":["multilinear singular integrals","matrix weights","Muckenhoupt weights","convex body domination","tensor products","Hardy–Littlewood maximal operator","Fujii–Wilson conditions","quasinorms"],"falsifier":"Test Theorem A on a two-factor example: choose explicit noncommuting $2\\times2$ matrix weights and compute, over all cubes, the norms of the averaging operators on the tensor-product weighted space and the norms of the multilinear Riesz transform from Example 6.19; a single tuple for which one of these quantities is finite while the other is infinite would disprove the claimed equivalence.","tokens_in":62071,"feed_emoji":"📐","tokens_out":8245,"duration_ms":73147,"temperature":0.7,"pith_summary":"The paper aims to settle, in full generality, which tuples of matrix weights are admissible for multilinear singular integrals. Its central result, Theorem A, says that for any exponents $1<p_j\\le\\infty$ with $\\frac1p=\\sum_j\\frac1{p_j}$, an $m$-linear Calderón–Zygmund operator acting on tensor products of finite-dimensional Hilbert spaces is bounded on the corresponding matrix-weighted Lebesgue spaces for every such operator if and only if the weights satisfy one uniform condition on cube averages—the multilinear matrix Muckenhoupt condition $A_{\\vec p}$—and this is also equivalent to boundedness of a tensor-product convex-body maximal operator. This unifies the scalar multilinear weight theory and the linear matrix weight theory, and it extends the characterization to the quasi-Banach range $p<1$. The quantitative forms show that the same single class controls sparse convex-body operators, maximal operators, and Calderón–Zygmund operators with explicit bounds.","feed_headline":"A complete Muckenhoupt criterion for multilinear matrix weights","feed_subtitle":"One uniform averaging condition decides when every multilinear Calderón–Zygmund operator and the maximal operator are bounded.","key_machinery":"The paper's load-bearing machinery is a new construction of reducing operators for quasinorms. Given a lower-semicontinuous quasinorm $\\rho$ on an $n$-dimensional Hilbert space, Proposition 2.4 produces a positive self-adjoint operator $A$ with $K^{-2n}\\rho(u)\\le \\|Au\\|\\le n^{1/2}\\rho(u)$, combining the John ellipsoid theorem with Carathéodory's theorem on convex hulls. This turns a multilinear quasinorm such as $\\rho_{W,Q,p}(u)=(\\fint_Q |W(x)u|^p\\,dx)^{1/p}$, where $p$ may be below $1$, into a comparable Euclidean norm. These reducing operators convert boundedness of averaging operators into a single operator-norm condition, which is the definition of $A_{\\vec p}$, and they feed the quantitative bounds through Fujii–Wilson characteristics. The other central objects are the tensor-product convex-body maximal operator $M^K$ and the convex-body sparse operator $A^K_S$, whose pointwise domination of $m$-linear Calderón–Zygmund operators reduces weighted singular-integral bounds to estimates for sparse convex-body sums.","core_discovery":"The paper establishes a full multilinear matrix weighted analogue of the classical scalar Muckenhoupt theory. For $1<p_j\\le\\infty$, $\\frac1p=\\sum_j\\frac1{p_j}$, and matrix weights $W_j$ on finite-dimensional Hilbert spaces $H_j$, the tensor-product weight $W=\\bigotimes_j W_j$ defines weighted spaces on $\\bigotimes_j H_j$. The authors prove that the following are equivalent: every $m$-linear Calderón–Zygmund operator $T$, extended componentwise through the tensor product, is bounded from $\\prod_j L^{p_j}_{W_j}(\\mathbb{R}^d;H_j)$ into $L^p_W$; the tensor-product convex-body maximal operator $M^K$ is bounded; and the tuple belongs to the multilinear matrix Muckenhoupt class $A_{\\vec p}$, meaning all cube averaging operators $1_Q\\bigotimes_j \\int_Q f_j\\,dx$ are uniformly bounded. The equivalence is quantitative: sparse convex-body operator and maximal operator bounds are controlled by the $A_{\\vec p}$ characteristic and scalar Fujii–Wilson conditions, and the lower bound uses a new notion of directional nondegeneracy, illustrated by the multilinear Riesz transform.","pith_inferences":["The quasinorm reducing-operator lemma is a transferable tool: any weighted quasi-Banach function space with a tensor-product structure should admit the same averaging-operator characterization, so the argument may extend to $p<1$ linear matrix weights or to other quasi-Banach lattices.","Because a single concrete operator (a multilinear Riesz transform) already forces the full class, one may conjecture that future characterisations can be stated with one test operator instead of 'all Calderón–Zygmund operators'.","The gap between the sharp scalar exponent for the multilinear maximal operator and the matrix exponent obtained here suggests that any improvement will need a genuinely noncommutative reverse Hölder estimate, not a reduction to scalar weights."],"forward_implications":["A tuple of matrix weights can be certified admissible by checking only the uniform boundedness of tensor-product averaging operators over all cubes.","The tensor-product convex-body maximal operator is bounded exactly on the same class, so the maximal and singular-integral theories share one weight condition.","Quantitative bounds for sparse convex-body operators and Calderón–Zygmund operators are controlled by the $A_{\\vec p}$ characteristic and Fujii–Wilson constants, recovering the sharp linear matrix bounds when $m=1$.","Every $m$-linear Calderón–Zygmund operator with a Dini-smooth kernel is pointwise dominated by a sparse convex-body operator, so weighted estimates reduce to estimating these sparse sums.","Directionally nondegenerate operators, including the multilinear Riesz transform, are strong enough to force the $A_{\\vec p}$ condition, giving a concrete lower-bound test."],"supporting_citations":[{"why":"Supplies the scalar multilinear Muckenhoupt class and maximal operator that the matrix characterization extends.","marker":"[LOP+09]"},{"why":"Introduces convex body domination and the sparse convex-body operator for matrix-weighted linear Calderón–Zygmund operators, which the paper adapts to the multilinear setting.","marker":"[NPTV17]"},{"why":"Defines the convex-set valued maximal operator and characterizes matrix $A_p$ weights by its boundedness, serving as the template for the tensor-product maximal operator.","marker":"[BC23]"},{"why":"Introduces the Christ–Goldberg maximal operator for matrix weights and its scalar characterization used in the quantitative bounds.","marker":"[CG01]"},{"why":"Proves matrix $A_p$ characterizations via maximal operators and nondegenerate Hilbert transform arguments, the basis for the directional nondegeneracy lower bound.","marker":"[Gol03]"},{"why":"Provides the sparse domination scheme for bilinear Calderón–Zygmund operators with Dini-smooth kernels that is generalized to $m$-linear operators.","marker":"[DHL18]"},{"why":"Gives the sharp reverse Hölder inequality for $A_\\infty$ weights used to estimate Fujii–Wilson characteristics in the quantitative bounds.","marker":"[HPR12]"},{"why":"Supplies the sharp bound for the Christ–Goldberg maximal operator that the linear case of the paper recovers.","marker":"[IM19]"}],"fun_headline_variants":["Matrix Muckenhoupt equivalence for multilinear operators","Multilinear matrix weights: one cube condition suffices","Sharp matrix Muckenhoupt for multilinear singular integrals","Tensor weights: the multilinear Muckenhoupt criterion","Averaging cubes characterize multilinear matrix boundedness"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole argument rests on the lemma that any lower-semicontinuous quasinorm on a finite-dimensional Hilbert space—a length function whose triangle inequality holds only up to a fixed constant—is comparable to the norm induced by a positive matrix, with a comparability constant depending only on dimension.","fun_headline_variants_meta":{"raw":{"variants":["Matrix Muckenhoupt equivalence for multilinear operators","Multilinear matrix weights: one cube condition suffices","Sharp matrix Muckenhoupt for multilinear singular integrals","Tensor weights: the multilinear Muckenhoupt criterion","Averaging cubes characterize multilinear matrix boundedness"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00133,"raw_usage":{"total_tokens":5438,"prompt_tokens":1000,"completion_tokens":4438,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":616,"completion_tokens_details":{"reasoning_tokens":4357}},"tokens_in":616,"tokens_out":4438,"duration_ms":24054,"temperature":1.0,"reasoning_tokens":4357,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T11:42:14.790456+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Test Theorem A on a two-factor example: choose explicit noncommuting $2\\times2$ matrix weights and compute, over all cubes, the norms of the averaging operators on the tensor-product weighted space and the norms of the multilinear Riesz transform from Example 6.19; a single tuple for which one of these quantities is finite while the other is infinite would disprove the claimed equivalence.","supporting_citations":[],"review_version":1}