{"id":"5727a8d4-d1fb-4d0c-88f9-a8eeb18298b4","arxiv_id":"2607.17430","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For dual-scale memory diffusion, the sharp local threshold is q_c = N(ρ−1)(1+α∞)/2 with norm inflation below it, and global small-data existence/decay holds above ρ0 = max(1+2/[N(1+α0)], (1+α∞)/(1+α0)).","lead":"This mathematics paper finds the exact size threshold (in Lebesgue spaces) at which a nonlinear diffusion-with-memory equation switches from well-posed to instantly unstable, and it gives a companion global-existence threshold. The results apply to physical memory models such as Cole-Cole relaxation and Prabhakar multi-scale diffusion.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 3.1 Step 2 assumes phase convergence of ĝ(λ)λ^{α∞} to a positive real constant, but (H2) only bounds |ĝ|; without a phase-asymptotics hypothesis the coercivity of D(µ,η) — and hence the S^{-2}_{1,0} estimate and both thresholds — is unproved for the stated kernel class.","rationale":"The reader's weakest assumption isolates the same structural point I do: the uniform sectoriality/coercivity of D(µ,η) and the phase separation Φ = (1+α∞)φ_c are load-bearing for the whole L^q–L^p theory. I agree, and I emphasize that the precise unsupported step is in Lemma 3.1 Step 2, where the proof passes from the magnitude bounds in (H2) to a definite phase limit m(t,µ) → C₁ ∈ ℝ₊. That passage is not licensed by (2.2) and can fail for analytic kernels with slowly rotating phases. The proposed counterexample is admissible in the sense of (H1)–(H3) if the low-frequency end is cut off, and it would directly test whether the claimed S^{−2}_{1,0} bound is a theorem about the stated class or only about kernels with extra phase regularity. The Cole-Cole and Prabhakar examples satisfy the stronger condition, so the intended results are not false; they are simply conditional on an unstated strengthening of (H2). This justifies keeping the reader's CONDITIONAL verdict rather than upgrading to ACCEPT, but does not warrant REJECT because the core estimates are very likely correct for the canonical kernels. I also note, without changing the verdict, that Theorem 5.1's parameter interval may require m ≥ 1 to fit the stated L^m framework; this is a separate, patchable gap.","tokens_in":30845,"tokens_out":7388,"duration_ms":73913,"concrete_test":"For a model kernel ĝ_ε(λ) = λ^{−α∞} e^{iε log(λ/i)} (with a low-frequency cutoff), fix α∞ = 1/2, ε = 0.1, and φ_c satisfying (3.3). Along µ = re^{iφ_c}, compute the phase of m(t,µ) = t^{−α∞} ĝ_ε(µ/t) for t ∈ [10⁻⁶, 10⁻²] and r near the resonant scale |η|^{2/(1+α∞)} for η = 10², 10³. If Φ(t,µ) = φ_c − arg m(t,µ) crosses π or drifts by an amount comparable to ε log(1/t), the uniform coercivity bound |D| ≥ C(r+|η|² r^{−α∞}) fails as t→0⁺, demonstrating that Lemma 3.1 Step 2 is not justified under (H1)–(H2) as stated.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central local theory rests on Lemma 3.1, whose Step 2 needs a uniform lower bound for D(µ,η) = µ + |η|²µ^{−α∞}H(t,µ) along the deformed rays µ = re^{±iφ_c}. The proof sets m(t,µ) = µ^{−α∞}H(t,µ) = t^{−α∞}ĝ(µ/t) and asserts that 'hypothesis (H2) dictates' that m(t,µ) converges to C₁ ∈ ℝ₊ with phase ≈ −α∞φ_c, so that the phase separation Φ = (1+α∞)φ_c lies safely away from π and 2π. But (H2), as stated in (2.2), is only a two-sided magnitude bound: C₁|λ|^{−α∞} ≤ |ĝ(λ)| ≤ C₂|λ|^{−α∞}. It says nothing about arg ĝ(λ) or about convergence of ĝ(λ)λ^{α∞} to a positive real constant. A kernel such as ĝ(λ) = λ^{−α∞}e^{iε log(λ/i)} (with a low-frequency cutoff to satisfy (H3)) satisfies the magnitude bound but has no limiting argument; the phase of m(t,µ) then varies by ε log(1/t) as t→0⁺, so no fixed φ_c can make cos Φ ≥ −1+δ₀ uniformly. The coercivity inequality |D| ≥ C(r+|η|²r^{−α∞}), and therefore the Hörmander S^{−2}_{1,0} bound, is not established under the stated hypotheses. This is not an internal contradiction — the Cole-Cole and Prabhakar examples do have the required phase behaviour — but it means Theorems 4.1 and 4.2 are proved only for a narrower, unstated subclass of admissible kernels. The same concern was identified by the reader through (H1), but the precise gap is in (H2).","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the semilinear integro-differential Cauchy problem (1.1) with a dual-scale memory kernel. The main claims are: (i) local Hadamard well-posedness in L^q for q ≥ q_c := N(ρ−1)(1+α_∞)/2 (Theorem 4.1); (ii) instantaneous norm inflation, hence ill-posedness, for 1<q<q_c (Theorem 4.2); and (iii) global existence with algebraic decay for small initial data in L^{q_c}∩L^m when ρ > ρ_0 := max(ρ_F, (1+α_∞)/(1+α_0)) (Theorem 5.1 and Corollary 5.4). The linear tool is an L^q−L^p smoothing estimate for the resolvent S(t), obtained by deforming the Bromwich contour and interpreting the high-frequency symbol as a Hörmander S^{-2}_{1,0} multiplier.","tokens_in":31295,"tokens_out":17929,"duration_ms":177121,"significance":"If the main theorems are correct, they give an essentially complete critical Lebesgue theory for a broad class of super-diffusive memory equations, including the first sharp ill-posedness threshold below q_c and a Fujita-type exponent for the dual-scale model. The paper is ambitious and mostly self-contained, with detailed fixed-point arguments and a concrete application to Cole-Cole kernels. The thresholds are explicit and not fitted to data, and the examples are physically meaningful. However, the central linear estimate rests on a phase-coercivity step that is not justified by the stated hypotheses, and the ill-posedness theorem contains a logical gap in passing from Picard-iterate blow-up to discontinuity of the solution map.","major_comments":[{"comment":"The proof asserts that 'hypothesis (H2) dictates' that m(t,μ)=t^{-α_∞}ĝ(μ/t) converges as t→0+ to a positive real constant C_1, so that the phase of m is approximately −α_∞φ_c and the phase separation Φ=(1+α_∞)φ_c is uniformly bounded away from π and 2π. Hypothesis (2.2) is only a two-sided magnitude bound and contains no information on arg ĝ(λ) or on the limit of ĝ(λ)λ^{α_∞}. A kernel such as ĝ(λ)=λ^{-α_∞}e^{iε log(λ/i)} (with a low-frequency cutoff to satisfy (H3)) satisfies (2.2) but has phase varying with log(1/t), so no fixed φ_c can give the uniform bound cos Φ ≥ −1+δ_0. The coercivity |D(μ,η)| ≥ C(r+|η|^2 r^{-α_∞}), and hence the S^{-2}_{1,0} estimate, is therefore not established under the hypotheses as stated. This is load-bearing for Theorems 4.1, 4.2, and 5.1. Either add an explicit phase-asymptotics hypothesis, e.g. ĝ(λ)λ^{α_∞}→C_1∈(0,∞) uniformly in a sector, and verify it f","section":""},{"comment":"The verification of (H1) for the Cole-Cole and Prabhakar kernels only checks that the characteristic equation has no roots on the negative real axis. The hypothesis (H1) also requires a uniform sector aperture θ_∞ < πα_∞/(1+α_∞) and containment of all roots in |arg λ|≤π−θ(|λ|) with θ(|λ|)→θ_∞>0. For the Cole-Cole kernel, the large-frequency roots have argument tending to π/(1+α), exactly at the boundary of the region controlled by the aperture. A separate argument is needed to show that a uniform strict inequality θ_∞<πα_∞/(1+α_∞) holds; the absence of branch-cut zeros is not by itself sufficient. Without this, the canonical examples are not proved admissible, and the claim that the abstract framework applies to them is not yet justified.","section":""},{"comment":"The proof constructs data for which the first Picard iterate N_1(u_{0,k})(t_k) diverges in L^q norm. The theorem then concludes that the data-to-solution map fails to be uniformly continuous at the origin. The implication is not established. Continuity of the actual solution map does not imply the asserted uniform estimate ∥N_1(u_0)(t)∥_{L^q}≤C∥u_0∥_{L^q}^ρ, because N_1 uses the linear evolution S(s)u_0 inside the nonlinearity rather than the actual solution u(s). A standard CCT ill-posedness proof either constructs actual solutions with growing norm or proves a quantitative bound linking the first Picard iterate to the solution map. As written, the result rigorously proves norm inflation for the first Picard iterates only; the statement about the solution map is stronger than the argument supports.","section":""},{"comment":"The coercive lower bound |D(μ,η)|≥C(r+|η|^2 r^{-α_∞}) is asserted for all r along the deformed rays μ=re^{±iφ_c}. For r near 0 the argument |μ/t| is not large, so (H2) does not apply. In the Cole-Cole example, for small μ one has m(t,μ)=t^{-α_∞}ĝ(μ/t)→t^{-α_∞}γ^{-1}, so |m| is bounded and the term |η|^2 r^{-α_∞} in the lower bound is not present. The subsequent estimate ∫_0^1 r^{α_∞}dr/(C|η|^2) is therefore not justified as written. A separate treatment of the contour integral near the origin is needed; this might use (H3) or analyticity, but (H3) is not assumed in Lemma 3.1. This is another load-bearing gap in the derivation of the S^{-2}_{1,0} bound.","section":""}],"minor_comments":[{"comment":"The critical exponent q_c = N(ρ−1)(1+α_∞)/2 may be smaller than 1 for small N and ρ close to 1. The paper repeatedly states the theory for 1≤q<∞, so the critical case q=q_c would then be outside the Banach-space setting. Please clarify the range of parameters where q_c>1, or state the modifications for q<1.","section":""},{"comment":"The remark asserts that 'the multi-scale kernel dictates the structural crossover α_0<α_∞' and that m_c<q_c unconditionally. This is not true for Example 2 (multi-term fractional diffusion), where α_0=α>β=α_∞. The spectral-bridge condition ρ>(1+α_∞)/(1+α_0) is still meaningful when α_0>α_∞, but the stated inequality m_c<q_c and the accompanying 'tighter upper bound' discussion should be corrected or qualified.","section":""},{"comment":"The sentence 'the local contribution near the origin, controlled via (H3), is of order O(t^{-(1+α_0)}|ξ|^{-2})' is terse; the derivation of this specific rate and its uniformity over |ξ|≥1 should be spelled out, because it is used to justify the high-frequency branch estimate.","section":""},{"comment":"The term 'Hadamard well-posedness' is used to justify a uniform continuity assumption at the origin. In Theorem 4.1 the map is actually Lipschitz, so it would be cleaner to argue directly from the Lipschitz estimate and to state explicitly what regularity of the solution map is being assumed in the contradiction argument.","section":""},{"comment":"There are occasional notation inconsistencies, e.g. α_∞ vs α∞ and the use of both 'q_c' and 'm_c' before their formal definitions in Section 5. These do not affect the mathematics but should be harmonized.","section":""}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern about phase in (H2) is legitimate and central: the advertised theorem is proved for a class of kernels broader than the hypotheses actually control. I would ask the author to either add a phase-asymptotics condition to (H2) and verify it for all examples, or prove the coercivity of D(μ,η) from the root-location hypothesis alone. The ill-posedness statement also needs a rigorous link between first-Picard norm inflation and failure of the solution map. If these points are fixed, the paper could be a strong contribution; in its present form it is too conditional for acceptance."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a serious attempt to push the Weissler–Fujita machinery into dual-scale memory equations. The threshold formulas are attractive and the norm-inflation construction is real work. The main soft spot is a missing phase hypothesis in Lemma 3.1. The proof says hypothesis (H2) dictates that the modulation function m(t,µ) converges to a positive real constant, but (H2) only controls |ĝ|, not arg ĝ. A kernel like ĝ(λ) = λ^{-α∞} e^{i ε log(λ/i)} with a low-frequency cutoff satisfies the magnitude bound but has no limiting argument, so the phase separation Φ = (1+α∞)φ_c may not stay bounded away from π and 2π. As written, the coercivity of D(µ,η), the S^{-2}_{1,0} bound, and hence Theorems 4.1 and 4.2 are proved only for a narrower, unstated subclass. That is a real gap, not a manufactured one.\n\nWhat is genuinely new: the dual-scale treatment with distinct α∞ and α0, the critical Lebesgue threshold q_c, the spectral-bridge condition ρ0 = max(ρ_F, (1+α∞)/(1+α0)), and the Christ–Colliander–Tao style norm inflation below the threshold. The global existence/decay theorem is also structured sensibly, with the weight Θ(t) and the midpoint splitting doing the expected work. The citation pattern looks honest; the author's earlier single-scale papers are context, not load-bearing.\n\nOther soft spots are minor by comparison. The sectoriality hypothesis (H1) for the canonical examples is checked by noting that the imaginary part of the characteristic equation does not vanish on the negative real axis, which gives a sector but not the required uniform aperture θ∞ < πα∞/(1+α∞). And the Fujita-type exponent is only half-established: the paper shows divergence of the relevant integral for ρ ≤ ρ_F, but explicitly leaves the blow-up/nonexistence proof to future work. That is a limitation on the sharpness claims, not on the existence theorems.\n\nThe paper deserves a serious referee. The gap is likely fixable by adding a phase-asymptotics assumption to (H2) or deriving it from a strengthened (H1), and for the Cole-Cole and Prabhakar kernels the needed phase behavior is present. I would not desk-reject. But as is, I would not cite the general theorem without a fix; the results are conditional on an unstated subclass of admissible kernels. A good referee should spend time on Lemma 3.1 and ask the author to either prove the phase convergence under the stated hypotheses or make the stronger hypothesis explicit.","headline":"Serious dual-scale memory paper with a real proof gap: (H2) does not imply the phase convergence needed in Lemma 3.1, so the general thresholds are conditional, but the architecture is sound and the canonical examples likely survive.","tokens_in":31807,"tokens_out":4023,"would_cite":false,"duration_ms":40495,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35R09","35R11","45K05","35B53","35B33","35A01"],"pacs":[],"model":"deepseek-v4-flash","headline":"For a broad class of super-diffusive memory equations, local well-posedness in L^q holds exactly when q ≥ N(ρ−1)(1+α∞)/2, and fails below this threshold via instantaneous norm inflation.","keywords":["integro-differential equations","anomalous diffusion","super-diffusive transport","resolvent operators","pseudo-differential multipliers","critical Lebesgue spaces","instantaneous norm inflation","critical exponents"],"falsifier":"Find a kernel satisfying (H1)–(H3) whose characteristic denominator D(μ,η) violates the coercivity estimate |D(μ,η)| ≥ C(r + |η|^2 r^{-α∞}) along the branch-cut contour for small t, or exhibit an admissible kernel whose modulation m(t,μ) does not converge to a positive constant as t→0; then Lemma 3.1 fails and the threshold q_c is not valid. A simpler check: for the fractional-retardation kernel, compute the maximum of |D|^{-1} on the deformed contour numerically as t→0; if it grows without bound, the smoothing estimate and the norm-inflation argument collapse.","tokens_in":30679,"feed_emoji":"⚡","tokens_out":12216,"duration_ms":106501,"temperature":0.7,"pith_summary":"This paper studies a nonlinear heat-type equation in which the diffusion term is a time-convolution of the Laplacian with a memory kernel that behaves like a power law with two different exponents at short and long times (super-diffusive transport). It claims that local well-posedness in L^q is exactly limited by the critical index q_c = N(ρ−1)(1+α∞)/2, set by the short-time exponent: if q ≥ q_c, the initial-value problem is locally well-posed (with a smallness condition at q = q_c), and if 1 < q < q_c, the first Picard iterate of arbitrarily small data explodes in norm at arbitrarily small times, so the data-to-solution map is discontinuous at the origin. For global dynamics, the long-time exponent α_0 takes over: the paper identifies a critical global exponent ρ_0 = max(1+2/[N(1+α_0)], (1+α∞)/(1+α0)) such that small data in L^{q_c}∩L^m with m in (q_c/ρ, m_c) yield a unique global mild solution decaying algebraically. The thresholds are derived for a general class of admissible kernels, including pure fractional, multi-term, fractional-retardation, and multi-scale memory models. If correct, these formulas give the precise regularity boundary and global blow-up boundary for a broad family of anomalous transport equations.","feed_headline":"Below one critical exponent, super-diffusive solutions blow up","feed_subtitle":"Above it, small data give unique global solutions and algebraic decay; the paper proves both sides of the threshold.","key_machinery":"The argument rests on L^q–L^p smoothing bounds for the resolvent operator S(t) obtained by Laplace inversion and scaling of its Fourier symbol. The symbol lies in the pseudo-differential class S^{-2}_{1,0}, giving a gain of exactly two spatial derivatives and enforcing the dimensional restriction 1/q − 1/p < 2/N. The central controlling estimate is the phase separation Φ = (1+α_∞)φ_c in the scaled denominator D(μ,η) = μ + |η|^2 μ^{-α∞}H(t,μ); the structural condition θ_∞ < πα_∞/(1+α_∞) keeps this phase bounded away from π, yielding a uniform coercivity bound |D| ≥ C(r + |η|^2 r^{-α∞}) and hence the required derivative bounds. For ill-posedness, the same denominator controls the first Picard","core_discovery":"The core discovery is that the short-time memory exponent α_∞ sets the local regularity threshold q_c = N(ρ−1)(1+α∞)/2, while the long-time exponent α_0 sets the global blow-up threshold ρ_F = 1+2/[N(1+α_0)]. Local well-posedness holds for q ≥ q_c and fails for 1 < q < q_c, where the data-to-solution map is discontinuous at the origin via instantaneous norm inflation of the first Picard iterate. Global existence with algebraic decay holds for small data in L^{q_c}∩L^m when ρ > max(ρ_F, (1+α∞)/(1+α0)). The theory is developed for a general class of admissible dual-scale kernels and gives explicit thresholds for fractional, multi-term, fractional-retardation, and multi-scale memory models.","pith_inferences":["A concrete test of the threshold: in the fractional-retardation model with α_∞=1/3, α_0=0, N=3, and cubic source, the paper predicts q_c=4; direct numerical simulation of the Picard iterate at times t_k = τ* λ_k^{-2/(1+α∞)} should show the L^4-norm of the first iterate diverging as λ_k^{δ} with δ = 2(q_c−q)/(q(1+α∞)) for q just below 4.","The paper establishes global existence but not finite-time blow-up for ρ ≤ ρ_F; extending the classical critical-exponent blow-up argument to two-scale memory kernels would complete the global picture and is not addressed here.","Because the S^{-2} symbol is the only mechanism generating the thresholds, one may expect the same q_c formula for other power-like sources; verifying the persistence of the threshold for non-power nonlinearities would test the universality of the mechanism."],"forward_implications":["For every admissible kernel, the local well-posedness threshold is exactly q_c = N(ρ−1)(1+α_∞)/2: the problem is locally well-posed in L^q for q ≥ q_c (with a smallness condition at equality) and ill-posed for 1 < q < q_c.","The global threshold ρ_0 = max(1+2/[N(1+α_0)], (1+α∞)/(1+α0)) separates global existence from failure; if ρ ≤ ρ_0 the contraction framework collapses, and for ρ ≤ ρ_F the long-time integral of the nonlinear source diverges, pointing to finite-time blow-up.","When the global solution exists, it decays algebraically at the linear rate: limsup t^{β_{m,p}}‖u(t)‖_{L^p} ≤ C, so the memory tail governs the long-time profile.","The same abstract thresholds are computed explicitly for pure fractional, multi-term, fractional-retardation (with α_0 = 0), and multi-scale memory kernels, providing ready-to-use critical exponents in those models."],"fun_headline_variants":["Super-diffusive blow-up threshold pinned precisely","Critical exponent decides blow-up or decay in super-diffusive PDEs","Memory kernel sets blow-up threshold for super-diffusive transport","Instantaneous norm inflation marks ill-posedness barrier","Sharp threshold for solvability in super-diffusive equations"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The results stand or fall on the uniform coercivity of the scaled resolvent denominator along the deformed contours, specifically the phase-separation bound θ_∞ < πα_∞/(1+α_∞) that keeps the linear and memory terms from cancelling; if an admissible kernel violates this phase bound, the smoothing estimates and both thresholds collapse.","fun_headline_variants_meta":{"raw":{"variants":["Super-diffusive blow-up threshold pinned precisely","Critical exponent decides blow-up or decay in super-diffusive PDEs","Memory kernel sets blow-up threshold for super-diffusive transport","Instantaneous norm inflation marks ill-posedness barrier","Sharp threshold for solvability in super-diffusive equations"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001098,"raw_usage":{"total_tokens":4462,"prompt_tokens":827,"completion_tokens":3635,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":571,"completion_tokens_details":{"reasoning_tokens":3552}},"tokens_in":571,"tokens_out":3635,"duration_ms":22580,"temperature":1.0,"reasoning_tokens":3552,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T17:59:24.181659+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find a kernel satisfying (H1)–(H3) whose characteristic denominator D(μ,η) violates the coercivity estimate |D(μ,η)| ≥ C(r + |η|^2 r^{-α∞}) along the branch-cut contour for small t, or exhibit an admissible kernel whose modulation m(t,μ) does not converge to a positive constant as t→0; then Lemma 3.1 fails and the threshold q_c is not valid. A simpler check: for the fractional-retardation kernel, compute the maximum of |D|^{-1} on the deformed contour numerically as t→0; if it grows without bound, the smoothing estimate and the norm-inflation argument collapse.","supporting_citations":[],"review_version":1}