{"id":"7d0d267e-7eec-4cae-af56-6ce8ebc9726f","arxiv_id":"2607.14991","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper's claimed exact Morrey admissibility threshold via resolution data is invalid: for f=x^2+y^2 it gives 2/3, while the promised RLCT equality gives 1/2.","lead":"This paper claims that the range of exponents for which the gradient of a singular interaction kernel is integrable is governed by algebraic-geometric data from resolution of singularities. The characterization is not correct as stated: a simple example already contradicts the claimed equality with the real log-canonical threshold.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Log-resolutions do not monomialize ∇f; for f=x^2+y^4 the formula gives 2/3 instead of the true threshold 1/2.","rationale":"The reader's weakest_assumption correctly identifies the non-monomiality of ∇f as the point where the proof of Theorem 2.11(i) breaks. The paper explicitly assumes in Section 2.1 and in the proof that ∇f∘π = w∏ y_i^{M_i} with a single M_i; this is not a property of log-resolutions. The example f=x^2+y^4 is decisive because it satisfies all hypotheses (real-analytic, isolated zero, K=1/f with two-sided comparability), and a direct computation gives MTI=1/2 while the standard log-resolution data give 2/3. The reader's additional algebra point about part (iii) is also valid—with M_j=ν_j−1 the denominator is ν_j+1, not 2ν_j—so the claimed sharpness for Newton non-degenerate germs is false even when part (i)'s formula is applied correctly; for f=x^2+y^2, p*=2/3 while rlct0/2=1/2. Both issues are independent and load-bearing; the monomialization failure is the more fundamental one. No amount of fixing part (iii) can save part (i) without a correct treatment of the variable order of |∇f| along the exceptional divisor.","tokens_in":17128,"tokens_out":41517,"duration_ms":316262,"concrete_test":"For f=x^2+y^4, compute directly ∫_{B_ε} |∇(1/f)|^p dx dy. Use the weighted blow-up x=s^2, y=sv: f=s^4(1+v^4), |∇f|=2s^2√(1+4s^2v^6), dxdy=2s^2 ds dv, so the integral behaves like ∫ s^{2-6p} ds, which converges iff p<1/2. Compare with Theorem 2.11(i) applied to the standard blow-up (ν=2, a=1, M=1), which yields p*=2/3. The mismatch settles the concern.","verdict_should_be":"REJECT","load_bearing_attack":"Theorem 2.11(i) is proved by pulling back ∇K under a log-resolution and assuming ∇f∘π = w(y)∏ y_i^{M_i} with a constant order M_i on each exceptional divisor. This is the step where the claimed iff characterization lives. A log-resolution makes f and the Jacobian monomial, but it does not monomialize the vector-valued gradient. For f=x^2+y^4, the standard blow-up at the origin (x=u, y=uv) gives f=u^2(1+u^2v^4), |J|=|u|, and |∇f|=2u√(1+4u^2v^6). The vanishing order of |∇f| along E={u=0} is 1 for generic v, but the region v~u^{-1/2} contributes an additional u^{-1/2} in the v-integral, so the naive Fubini reduction ∫ u^{1-3p} du is invalid; the true integrability threshold of ∇(1/f) is p<1/2, not the 2/3 obtained from (a+1)/(2ν−M)=(1+1)/(4−1). The same non-monomiality invalidates part (i) generally; the formula is not even resolution-invariant (the weighted blow-up with weights (2,1) gives a=2, ν=4, M=2 and p*=1/2). Thus the central characterization rests on a false premise.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies local L^p integrability (and hence Morrey admissibility) of ∇K for singular kernels of the form K = 1/f, where f is a real-analytic function with an isolated zero. Under a two-sided comparability assumption |∇K| ≍ |∇f|/|f|^2, it claims a complete geometric characterization: ∇K ∈ L^p_loc iff p < p* := min_i (a_i+1)/(2ν_i − M_i), where (ν_i, M_i, a_i) are asserted to be derived from a log-resolution of {f=0}. It further claims that the real log-canonical threshold gives a lower bound p := (1/2) rlct0(f) ≤ p*, with equality for Newton non-degenerate germs. The proof is based on pulling back ∇K via a log-resolution and reducing to one-dimensional monomial integrals.","tokens_in":17520,"tokens_out":11594,"duration_ms":98431,"significance":"If correct, the result would give a birational-invariant formula for the sharp Morrey admissibility threshold of a broad class of interaction kernels, with a computable Newton-polyhedron criterion. The paper also connects the threshold to the real log-canonical threshold and to the authors' program on volume asymptotics and persistent topology. However, the central derivation and the sharpness theorem are not sound; the main theorem fails in elementary examples, and the advertised Newton-polyhedron criterion is contradicted by radial examples already treated in the paper. The significance is therefore contingent on a proof that the manuscript does not provide.","major_comments":[{"comment":"The proof assumes that a log-resolution simultaneously monomializes the gradient: ∇f∘π = w(y)∏ y_i^{M_i} with a single order M_i on each exceptional component. A log-resolution only monomializes f and the Jacobian; it does not monomialize the vector-valued gradient. For f = x^2+y^4, the standard blow-up x=u, y=uv gives f=u^2(1+u^2v^4), |∇f|=2u√(1+4u^2v^6), |J|=|u|. The pulled-back integrand is u^{1−3p}(1+4u^2v^6)^{p/2}(1+u^2v^4)^{−2p}, whose dependence on v is not a pure power; the region v ≍ u^{−1/2} contributes an extra u^{−1/2}. The claimed Fubini reduction to ∫ u^{1−3p} du (giving p* = 2/3) is therefore invalid; the actual threshold is p < 1/2. Thus the iff characterization in (i) rests on a false premise.","section":"§2, 'Local integrability via log resolutions' and Theorem 2.11(i)"},{"comment":"There is an algebra error: substituting M_j = ν_j − 1 into 2ν_j − M_j gives ν_j + 1, not 2ν_j. The equality (a_j+1)/(2ν_j−M_j) = (a_j+1)/(2ν_j) is false. Consequently the claimed characterization of equality p* = p, and in particular the assertion that p* = p for every Newton non-degenerate germ, is unsupported. For f = x^2+y^2, a log-resolution gives ν=2, a=1, M=1 (so M=ν−1), but the formula with the correct denominator gives p* = (1+1)/(4−1) = 2/3, while p = (1/2)·(2/2) = 1/2. Hence the sharpness theorem is false as stated.","section":"Theorem 2.11(iii), proof step 3"},{"comment":"The examples compute thresholds using the exponent −p(2ν) + a, i.e. implicitly setting M=0, in contradiction with the definition M_i ≥ ν_i − 1. For f = x^2+y^3, the paper reports ν=6, a=4 and p < 5/12 from ∫ |y|^{−12p+4} dy. But with M ≥ ν−1 = 5, the same monomial scheme would give a different exponent, and the claimed sharpness p* = 5/12 does not follow. The radial counterexample f = x^2+y^2 already shows that the Newton-polyhedron criterion of §2.3 (p* = 1/(2d_NP)) is wrong: it gives 1/2 instead of the actual threshold 2/3.","section":"Example 2.3 and §3 computations"},{"comment":"The step-by-step criterion asserts p* = 1/(2 d_NP(f)) for Newton non-degenerate germs, citing Theorem 2.11(iii). Since (iii) is false, the criterion is not established. For f = x^2+y^2 (Newton non-degenerate), d_NP=1 and the criterion gives p*=1/2, contradicting the direct computation p*=2/3 for ∇(1/f) in R^2. The advertised resolution-independent characterisation of the maximal integrability of ∇K therefore fails in the simplest isotropic case.","section":"§2.3, computational criterion, steps (1)–(4)"}],"minor_comments":[{"comment":"The symbols ν_i and N_i are introduced as identical, but the text uses both in formulae without consistently distinguishing them. This is confusing, especially in the statement of Theorem 2.11 where ν_i appears but the notation summary lists N_i as the integral notation.","section":"Notation summary, §2"},{"comment":"The normalization of the real log-canonical threshold is nonstandard: the usual definition is sup{c>0: |f|^{-c} is locally integrable}, but the remark also identifies rlct0(f) = min_i (a_i+1)/ν_i after claiming a factor of two convention. The relationship with the divisorial data should be stated precisely; as written, the same symbol is used for two different normalizations.","section":"Remark 2.5"},{"comment":"The table for f=x^m+y^n lists p* = (m+n)/(2mn). For m=n=2 this gives 1/2, but the actual threshold for ∇(1/(x^2+y^2)) in R^2 is 2/3. The table therefore propagates the error in Theorem 2.11(iii).","section":"Example 3.1 table and figures"},{"comment":"The paper cites several arXiv preprints by the authors [15,16] as part of the same program. While self-citation is not improper, the refereed status of these preprints is unclear; the manuscript should indicate which results are independent of these preprints.","section":"References and self-citations"}],"recommendation":"reject","confidential_remarks":"The manuscript's central theorem is not supported. The gradient-monomialization assumption is false, and the algebra in Theorem 2.11(iii) is wrong. The examples in Section 3 contradict the main theorem when computed correctly. These are load-bearing errors that cannot be repaired within the manuscript's scope; a full revision would need a fundamentally different method for computing the L^p threshold of ∇(1/f), or a restriction to a class where a genuine simultaneous monomialization of f and ∇f exists."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should know before reading: the headline claim is false as written. The iff characterization in Theorem 2.11 depends on the assumption that a log-resolution of (f=0) also monomializes the vector field ∇f, and that assumption is not available. For f=x^2+y^4, the standard blow-up gives M=1 and p*=2/3, while the weighted blow-up gives M=2 and p*=1/2; the formula is not resolution-independent. And Theorem 2.11(iii) contains a simple algebra error: with M_j=ν_j−1 the denominator is ν_j+1, not 2ν_j. So the claimed equality with (1/2)rlct0 for Newton non-degenerate germs is wrong even for f=x^2+y^2, where p*=2/3 and (1/2)rlct0=1/2.\n\nThat said, the paper is not empty. The question is good: Morrey admissibility of anisotropic kernels is exactly the kind of thing that should be governed by divisorial data, and the idea of using the RLCT as a systematic lower bound is natural and likely correct. The two-sided comparability hypothesis is honestly stated, and the Newton-polyhedron examples are clean and easy to check. The connection to volume asymptotics, singular learning theory, and persistent topology gives useful context; the self-citations are numerous but not fraudulent.\n\nThe soft spots are the load-bearing ones. The proof's reduction to one-dimensional monomial integrals requires |∇f∘π|≈∏|y_i|^{M_i}, with a single order per exceptional component. A log-resolution only gives monomial forms for f and the Jacobian. In general the vanishing order of |∇f| varies along a component, as x^2+y^4 shows, and the naive Fubini step collapses. There is also a more subtle issue: the examples in Section 3 only keep exceptional divisors and ignore strict-transform components, so the claimed computations are incomplete even in the cases where the formula might be repairable. The lower bound p≤p* seems solid; the exact formula and the sharpness claim do not.\n\nWho is this for? Anyone thinking about singular interaction kernels and birational invariants. The paper deserves a serious referee if it is revised to either prove the monomialization step (which I doubt) or downgrade the main theorem to a conjecture plus a clean sufficient condition. As is, I would not cite it.","headline":"The main theorem rests on a false monomialization assumption and an algebra slip, but the underlying connection between RLCT and Morrey admissibility is worth taking seriously.","tokens_in":17999,"tokens_out":6035,"would_cite":false,"duration_ms":53947,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35Q92","35A01","14B05","35B65","35B40","32S45"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims that the exact Morrey admissibility threshold for singular interaction kernels is a birational invariant determined by a log-resolution, and that it equals half the real log-canonical threshold for Newton non-degenerate si","keywords":["aggregation equations","Morrey spaces","real log-canonical threshold","resolution of singularities","singular interaction kernels","Newton polyhedron","anisotropic singularities","admissibility threshold"],"falsifier":"For f(x,y)=x^2+y^4 (Newton non-degenerate), the formula gives p* = 3/8. Direct integration in the sector near the y-axis, writing x = rψ, y = r, yields |∇(1/f)|^p ≍ r^{-3p} (ψ^2+4r^4)^{p/2} (ψ^2+r^2)^{-2p}; scaling ψ = rt makes the integral over r behave like ∫_0^1 r^{2-6p} dr, which converges for every p < 1/2. Hence ∇(1/f) ∈ L^{0.4}_loc for p = 0.4, contradicting p* = 0.375 if the theorem's iff is meant to hold.","tokens_in":16983,"feed_emoji":"📐","tokens_out":17883,"duration_ms":143302,"temperature":0.7,"pith_summary":"The paper studies viscous aggregation equations and asks which singular interaction kernels admit local well-posedness in Morrey spaces. The velocity field involves ∇K convolved with the density, so the critical question is the local L^p-integrability of ∇K near the kernel's singularity. For kernels comparable to 1/f with f a real-analytic function with an isolated zero, the paper claims a complete geometric answer: ∇K ∈ L^p_loc exactly when p lies below a threshold p* encoded in a log-resolution of the zero set. The real log-canonical threshold gives a computable lower bound for p*, and the bound is exact for Newton non-degenerate singularities. If correct, admissibility becomes a birational invariant computable from Newton data rather than a case-by-case analytic estimate.","feed_headline":"Resolution data fix the exact Morrey threshold","feed_subtitle":"For singular kernels of type 1/f, the admissible p-range of ∇K is a resolution-theoretic minimum, computable from Newton geometry.","key_machinery":"The load-bearing object is the log-resolution π: X̃→R^n of the zero set of f, together with its divisorial data (ν_i, M_i, a_i): the vanishing order of f along an exceptional component E_i, the vanishing order of |∇f| along E_i, and the Jacobian discrepancy exponent of π. The mechanism is the monomial reduction: in adapted coordinates, f∘π, ∇f∘π, and |Jπ| are assumed to be monomials, so |∇(1/f)|^p |Jπ| ≍ ∏ |y_i|^{-p(2ν_i−M_i)+a_i}, and Fubini reduces L^p integrability to the one-dimensional conditions −p(2ν_i−M_i)+a_i > −1. The minimum of the ratios (a_i+1)/(2ν_i−M_i) is the Morrey Threshold Index.","core_discovery":"Under the two-sided comparability |∇K| ≍ |∇f|/|f|^2 near an isolated zero of a real-analytic f, the paper proves that ∇K ∈ L^p_loc if and only if p < p* = min_i (a_i+1)/(2ν_i−M_i), where ν_i, M_i, a_i are the vanishing orders of f and of |∇f| and the discrepancy exponent along each exceptional component of a log-resolution of (f=0). Thus the Morrey Threshold Index of K is p*. The real log-canonical threshold rlct0(f) = min_i (a_i+1)/ν_i gives the lower bound p* ≥ (1/2)rlct0(f), with equality whenever a divisor attaining rlct0(f) has M_i = ν_i−1, in particular for Newton non-degenerate germs.","pith_inferences":["The proof of Theorem 2.11 assumes ∇f∘π is a single monomial w ∏ y_i^{M_i} per exceptional component; standard log-resolutions do not guarantee this. For f(x,y)=x^2+y^4, the vanishing order of |∇f| varies along the divisor, so the claimed iff may hold only as a sufficient condition, and the exact threshold could require a refined invariant.","A direct computation for x^2+y^4 suggests the true L^p threshold for ∇(1/f) is larger than the paper's formula gives; if confirmed, the equality p* = (1/2)rlct0(f) for Newton non-degenerate germs would need modification.","The Newton-polyhedron pipeline could be turned into a screening rule: compute d_NP(f) and check whether |∇f| has constant vanishing order on the face realizing the Newton distance; when it does not, the admissible p-range is likely wider than 1/(2d_NP), so the RLCT bound is conservative.","The same divisorial machinery could be applied to kernels of the form log|f| or to singular sets of positive dimension, where stratified valuations would replace a single divisor; the paper leaves that extension open."],"forward_implications":["The admissible range of Morrey exponents for singular kernels K=1/f is fixed by a log-resolution of the defining function, so anisotropic kernels such as (x^2+y^3)^{-λ} are handled by the same criterion as isotropic ones.","For Newton non-degenerate germs, p* equals 1/(2d_NP(f)), where d_NP(f) is the Newton distance, making the threshold directly computable from the Newton polyhedron.","The RLCT bound p < (1/2)rlct0(f) is sufficient and is exactly sharp precisely when ∇f vanishes to the minimal order ν_i−1 on a divisor attaining the RLCT; otherwise the true threshold is larger.","Kernels with radial cancellation (like the Newtonian |x|^{2-d}) can lie outside the hypothesis and be integrable beyond the resolution-theoretic threshold, indicating the criterion captures worst-case anisotropic concentration."],"fun_headline_variants":["Resolution data fix exact Morrey exponent threshold","Morrey p-range: geometric min from singularities","Newton geometry yields exact Morrey admissibility bound","Analytic zero sets dictate Morrey well-posedness"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that a log-resolution of (f=0) simultaneously turns the gradient into a single monomial w ∏ y_i^{M_i} along each exceptional component; a log-resolution only monomializes f and the Jacobian, so for germs like x^2+y^4, where the vanishing order of |∇f| varies along the divisor, the reduction to one-dimensional integrals and the claimed iff do not follow.","fun_headline_variants_meta":{"raw":{"variants":["Resolution data fix exact Morrey exponent threshold","Morrey p-range: geometric min from singularities","Newton geometry yields exact Morrey admissibility bound","Analytic zero sets dictate Morrey well-posedness"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000357,"raw_usage":{"total_tokens":1823,"prompt_tokens":848,"completion_tokens":975,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":592,"completion_tokens_details":{"reasoning_tokens":914}},"tokens_in":592,"tokens_out":975,"duration_ms":9482,"temperature":1.0,"reasoning_tokens":914,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T00:32:28.640878+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For f(x,y)=x^2+y^4 (Newton non-degenerate), the formula gives p* = 3/8. Direct integration in the sector near the y-axis, writing x = rψ, y = r, yields |∇(1/f)|^p ≍ r^{-3p} (ψ^2+4r^4)^{p/2} (ψ^2+r^2)^{-2p}; scaling ψ = rt makes the integral over r behave like ∫_0^1 r^{2-6p} dr, which converges for every p < 1/2. Hence ∇(1/f) ∈ L^{0.4}_loc for p = 0.4, contradicting p* = 0.375 if the theorem's iff is meant to hold.","supporting_citations":[],"review_version":1}