{"id":"4959c209-44b0-415a-9e14-a5f821ba7086","arxiv_id":"2411.16078","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A second-order Crank-Nicolson finite element scheme is proved stable and O(tau-squared plus h-squared) accurate for a viscoelastic PDE with a variable-exponent Abel kernel, and numerical tests confirm the rates.","lead":"This paper studies a time-dependent 'Abel' kernel whose exponent changes over time, and claims it can smoothly transition from short-time quasi-exponential behavior to long-term power-law decay. It then builds and analyzes a second-order numerical scheme for a viscoelastic integro-differential equation using this kernel.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 4.3's O(tau^2) proof relies in (4.28) on a pointwise L^infinity bound for partial_t^2 Delta u that Theorem 3.5 does not supply; the gap is real but repairable by a Peano-kernel/Young-convolution argument.","rationale":"The reader's weakest assumption identifies the same load-bearing concern: equation (4.28) implicitly assumes a uniform-in-time pointwise bound on ||partial_t^2 Delta u||_{L^2} that Theorem 3.5 does not provide. I agree that this is the most serious gap in the proof of the central convergence claim. It is a genuine proof gap in the manuscript, not merely an aesthetic issue, because the displayed inequalities as written do not follow from the stated hypotheses. However, the gap is fixable: a Peano-kernel representation of the linear interpolation error converts the offending pointwise evaluations into integrals of partial_t^2 Delta u, and Young's convolution inequality with p > 1/alpha_* controls the singular kernel t^{alpha_* - 1} in L^{p'}. The numerical experiments in Table 5.1 show clean second-order rates and are consistent with the claimed theorem, which supports the conclusion that the scheme is second-order in practice, but the proof still needs the repair. Several supporting theorems from [40] are imported without proof and the phrase 'non-positivity' applied to a positive kernel is unexplained, but neither is as directly load-bearing as the (4.28) step. Since the concern is real yet repairable, and the reader already returned a conditional verdict, no change of verdict is needed.","tokens_in":18640,"tokens_out":15487,"duration_ms":144673,"concrete_test":"Independently re-derive (4.28) from Theorem 3.5 using the Peano-kernel form phi(s) - L1[phi](s) = integral G(s,theta) partial_t^2 phi(theta) dtheta, with |G(s,theta)| <= C tau, and apply Young's convolution inequality to tau sum_n ||R_2^n||. If the resulting exponent of tau is still 2 for some p > 1/alpha_*, the proof gap is expositional; if not, Theorem 4.3 requires an added L^infinity regularity assumption or the temporal order should be weakened to tau^{2 - 1/p}.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is Theorem 4.3, and the temporal part of the proof rests on the quadrature error bound (4.28). In that chain the authors write the linear-interpolation remainder as (1/2) partial_t^2 Delta u(xi_j)(s-t_{j-1})(s-t_j) with xi_j in (t_{j-1}, t_j), then bound the sum by Q tau^2 integral_0^{t_n} (t_n - s)^{alpha_* - 1} ds. This drop requires a uniform-in-time bound ||partial_t^2 Delta u(.,t)||_{L^2} <= Q on the spatial L^2 norm for all t in the interval. Theorem 3.5 only gives partial_t^2 Delta u in L^p(0,T; L^2) for finite p, which does not imply the needed L^infinity control. The same pointwise-evaluation pattern appears in the n=1 term of (4.27), although there the W^{3,p} embedding can supply the needed L^infinity bound for partial_t^2 u. Thus the O(tau^2) statement is not fully justified as written. The gap is structural rather than fatal: the interpolation error has a Peano-kernel representation, and after summing over n the convolution with t^{alpha_* - 1} can be controlled in L^{p'} for p > 1/alpha_*, recovering the same O(tau^2) from the available L^p regularity.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper introduces the variable-exponent Abel kernel k(t)=t^{α(t)-1}/Γ(α(t)) with α(0)=1 as a multiscale kernel, and applies it to the parabolic integro-differential equation (1.3)-(1.5) that models viscoelastic vibration. The authors prove well-posedness and regularity results (Theorems 3.2, 3.4, and 3.5), propose a Crank-Nicolson time discretization combined with linear-interpolation quadrature for the memory term and a piecewise-linear finite element spatial discretization (scheme (4.30)), and analyze the scheme with an exponentially weighted energy argument. The main theoretical claim is Theorem 4.3: under the regularity hypotheses of Theorem 3.5, the fully discrete solution satisfies ||u^m - U_h^m|| ≤ Q(τ² + h²). Numerical experiments in Section 5 confirm second-order temporal and spatial convergence and illustrate a crossover from short-time quasi-exponential to long-time power-law behavior.","tokens_in":18949,"tokens_out":20221,"duration_ms":166989,"significance":"The main theoretical contribution, if fully justified, is a rigorous second-order-in-time convergence proof for a PIDE with a variable-exponent Abel kernel, improving the first-order scheme of [40]. The exponentially weighted energy framework for the non-positive and non-monotone kernel is a useful extension of existing techniques, and the convergence tables in Table 5.1 provide clean numerical support for the claimed O(τ² + h²) accuracy. The paper is transparent about its reliance on [40] for several intermediate estimates, and the crossover dynamics in Fig. 5.2 is a nice qualitative illustration of the multiscale kernel. However, the proof of the temporal quadrature error contains a regularity gap that is load-bearing for Theorem 4.3; until that gap is repaired, the O(τ²) part of the central claim is not fully proven.","major_comments":[{"comment":"The bound (4.28) for the quadrature error R_2^n is not justified by the stated regularity assumptions. The derivation writes the linear-interpolation remainder of Δu as (1/2) ∂_t^2 Δu(ξ_j)(s-t_j)(s-t_{j-1}) with ξ_j∈(t_{j-1},t_j) and then bounds ||∂_t^2 Δu(ξ_j)|| by a generic constant Q. This requires a pointwise, uniform-in-time L²-space bound for ∂_t^2 Δu. Theorem 3.5, however, only yields ∂_t^2 u ∈ L^p(0,T; ˇH²), i.e. ∂_t^2 Δu ∈ L^p(0,T; L²) for finite p, which does not imply pointwise values of the second time derivative. The gap is repairable: replacing the pointwise evaluation by a Peano-kernel representation of the interpolation error and using Young's convolution inequality with the kernel t^{α_*-1} should recover O(τ²) from the available L^p regularity, but the argument must be written out.","section":"§4.2, Eq. (4.28)"},{"comment":"The proof of the spatial error estimate (4.34) is only sketched and omits a load-bearing estimate. In the error equation (4.32), the right-hand side contains δ_t^λ η^n, the time difference of the Ritz projection error. To obtain the h² term, one must estimate τ Σ_{n=1}^m ||δ_t^λ η^n||² (or the corresponding inner-product term) by Q h^4 ||∂_t u||_{L²(ˇH²)}², using the representation δ_t η^n = τ^{-1} ∫_{t_{n-1}}^{t_n} ∂_t η(s) ds. The sentence 'use ||η^m|| ≤ Q h²||u||_{L∞(H²)} and combine Theorem 4.2' does not supply this bound; only the pointwise-in-time Ritz error is bounded there, not its time difference. This is a standard and presumably correct estimate, but it needs to be displayed for the proof to be complete.","section":"§4.3, proof of Theorem 4.3"}],"minor_comments":[{"comment":"The word 'muitiscale' in the first bullet should be 'multiscale'.","section":"§1.1, bullet list"},{"comment":"The step from (4.22) to (4.26) is quite compressed. In particular, the 'selecting m*' argument should explicitly handle the case ||ρ^{m*}||=0 and should state which constants absorb the factors e^{-λt_n} and the discrete Gronwall step.","section":"§4.2, proof of Theorem 4.2"},{"comment":"Theorem 4.3 is stated as a stability result but also contains the error estimate (4.34); separating these two statements would make the logical structure clearer.","section":"§4.3, Theorem 4.3"},{"comment":"The paper relies on [40] for well-posedness, Lemma 3.3, and Theorem 3.4 without proof. Since these results are central to the regularity chain, a short paragraph at the beginning of Section 3 indicating exactly which statements are proved here and which are imported from [40] would improve readability.","section":"§3, Lemmas 3.3 and 3.4"},{"comment":"The classification of k_0(t)=e^{α'(0)t ln t} as 'quasi-exponential' is somewhat imprecise, since the exponent is proportional to t ln t rather than t; a brief clarification of the intended meaning would avoid confusion.","section":"§1.1, Eq. (1.2)"}],"recommendation":"major_revision","confidential_remarks":"The gap in Eq. (4.28) is real but appears repairable within the scope of the paper, and the numerical evidence is clean. The manuscript relies heavily on the authors' own [40] for well-posedness and intermediate regularity results, which is transparent but worth noting when considering novelty. I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a solid incremental advance over the authors' own first-order work [40]. It builds a Crank-Nicolson Galerkin scheme for a viscoelastic PIDE with a variable-exponent Abel kernel, proves stability and an O(τ²+h²) error estimate, and supports the rates with clean numerical tables. The genuinely new pieces are the high-order regularity estimates in Theorem 3.5 and the exponentially weighted energy argument in Section 4. The weighting is a natural way to handle the kernel's non-monotonicity, and the experiments show consistent second-order convergence in time and space.\n\nThere is one load-bearing gap a referee should catch. In the quadrature error bound (4.28), the authors evaluate ∂_t²∆u at an intermediate point ξ_j and then bound the sum by τ²∫(t_n−s)^{α*−1}ds. That step needs a uniform-in-time L² bound on ∂_t²∆u. Theorem 3.5 only gives ∂_t²∆u ∈ L^p(0,T;L²) for finite p, which does not imply L^∞ control. The gap is structural rather than fatal: a Peano-kernel representation followed by Young's convolution inequality with p > 1/α* should recover the same O(τ²) from the available L^p regularity. I'd ask the authors to supply that argument before publication.\n\nMinor issues: some supporting results are imported without proof from [40], which is transparent and acceptable for a follow-up, but the paper should state what is genuinely new. Calling the kernel 'non-positive' is confusing because it is positive everywhere; I assume they mean non-monotone or not of positive type. The crossover behavior itself already appears in [40]; the mechanical-vibration application is a demonstration, not a new concept.\n\nThe main proof gap is real but localized, and the numerical evidence plus the weighted energy framework justify sending this to a serious referee. If the (4.28) repair works as sketched, the central claim should hold. I'd cite it if I worked on variable-order PIDEs, and I'd bring it to a reading group as a useful example of a repairable regularity gap.","headline":"A solid second-order scheme for a variable-exponent Abel kernel PIDE, with a real but repairable regularity gap in the temporal error proof.","tokens_in":19490,"tokens_out":4137,"would_cite":true,"duration_ms":36178,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["45K05","65M12","65M60"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper establishes that the variable-exponent Abel kernel $k(t)=t^{\\alpha(t)-1}/\\Gamma(\\alpha(t))$ with $\\alpha(0)=1$ models a single relaxation process that is quasi-exponential at short times and power-law at long times, and proves a…","keywords":["variable-exponent Abel kernel","multiscale kernel","crossover dynamics","viscoelasticity","integro-differential equation","Crank-Nicolson method","Galerkin finite element","exponentially weighted energy"],"falsifier":"Take data satisfying Theorem 3.5 but with $\\partial_t^2\\Delta u$ having an $L^p$-integrable yet unbounded singularity at $t=0$ (for example behaving like $t^{-1/p}$), run the fully discrete scheme with decreasing $\\tau$ and fixed small $h$, and check whether $\\|u^m-U_h^m\\|/\\tau^2$ stays bounded. If it grows like $\\tau^{-\\varepsilon}$ or $\\log(1/\\tau)$, the regularity hypotheses in the theorem are insufficient for the stated second-order bound.","tokens_in":18429,"feed_emoji":"📉","tokens_out":8990,"duration_ms":78730,"temperature":0.7,"pith_summary":"The paper introduces the variable-exponent Abel kernel $k(t)=t^{\\alpha(t)-1}/\\Gamma(\\alpha(t))$ and argues that when $\\alpha(0)=1$ this single kernel produces a multiscale relaxation: quasi-exponential at short times and power-law decay at long times, with the crossover time adjustable through the exponent $\\alpha(t)$. It applies this kernel to a parabolic integro-differential equation for viscoelastic vibration and derives well-posedness and high-order regularity estimates for its solutions. The main numerical claim is a fully discrete Crank-Nicolson Galerkin scheme that is second order in time and space, $\\|u^m-U_h^m\\|\\le Q(\\tau^2+h^2)$, despite the kernel being neither positive nor monotone; the proof uses an exponentially weighted energy argument. The underlying aim is to turn a mathematically awkward variable-order memory kernel into a practical, provably accurate model for materials whose properties change under load.","feed_headline":"One kernel spans exponential and power-law decay","feed_subtitle":"A second-order Galerkin scheme now solves the resulting viscoelastic model with proven accuracy.","key_machinery":"The machinery that carries the argument is the linearly interpolated Crank-Nicolson discretization of the variable-exponent convolution $I^{(\\alpha(t))}\\phi(t)=\\int_0^t (t-s)^{\\alpha(t-s)-1}/\\Gamma(\\alpha(t-s))\\phi(s)\\,ds$, combined with an exponentially weighted energy framework. The paper rescales all discrete unknowns by $\\hat V^n=e^{-\\lambda t_n}V^n$ and uses the modified difference quotient $\\delta^\\lambda_t \\hat V^n=(\\hat V^n-e^{-\\lambda\\tau}\\hat V^{n-1})/\\tau$, then tests against the weighted average $\\Lambda_\\lambda(\\hat V^n)$; choosing $\\lambda$ large and $\\tau$ small makes the awkward non-positive, non-monotone quadrature coefficients harmless through estimates like $Q\\zeta(\\tau^{\\alpha_*}+\\lambda^{-\\alpha_*})\\le \\mu$. The regularity side uses Laplace-representation and semigroup smoothing estimates to control $\\partial_t^2\\Delta u$ and $\\partial_t^3 u$, which feed the interpolation remainder and Taylor terms of the error equation.","core_discovery":"On the paper's own terms, the central discovery is that the kernel in (1.1) is a multiscale object rather than a technical nuisance: with $\\alpha(0)=1$ it eliminates the initial singularity of constant-exponent Abel kernels and still reproduces their long-time power-law tail, so it can represent crossover dynamics that a single power law cannot. The paper then proves that the initial-boundary-value problem (1.3)-(1.5) with this kernel is well posed and that its solution enjoys enough regularity to support a second-order time discretization. The load-bearing theorem is Theorem 4.3: the fully discrete Crank-Nicolson Galerkin scheme (4.30) is stable and satisfies $\\|u^m-U_h^m\\|\\le Q(\\tau^2+h^2)$ under the regularity hypotheses of Theorem 3.5. Numerical experiments with $\\alpha(t)=1-\\frac45 t$ show second-order rates in both time and space and display the predicted crossover from the quasi-exponential short-time solution to the power-law long-time solution.","pith_inferences":["A natural extension not pursued in the paper is to derive an explicit relation between the crossover time and the decay rate of $\\alpha(t)$; such a relation could let one estimate the exponent dynamics from measured relaxation curves.","The proof uses only size and logarithmic-derivative bounds on $k$, so the same quadrature-weighting framework is likely to transfer to tempered or distributed-order Abel-type kernels.","A conservative reading of the error proof suggests the clean $\\tau^2$ rate rests on a uniform-in-time bound for $\\partial_t^2\\Delta u$, stronger than the $L^p$-in-time regularity proved in Theorem 3.5; constructing a borderline example would show whether the stated hypotheses are minimal.","For viscoelastic testing, the model implies a two-regime amplitude envelope in a single experiment, a falsifiable distinction from constant-exponent viscoelastic models."],"forward_implications":["The fully discrete scheme (4.30) is a rigorously second-order method for the variable-exponent viscoelastic PIDE, improving the first-order scheme of [40] and matching the convergence rates observed numerically.","The error bound separates temporal and spatial contributions, so time step and mesh size can be refined independently while retaining $O(\\tau^2+h^2)$ accuracy.","The exponentially weighted energy framework handles kernels that are neither positive nor monotone, so the proof strategy can be reused for other nonlocal memory terms with oscillatory coefficients.","With $\\alpha(0)=1$, the model predicts a finite initial response rate instead of the singularity seen for constant-exponent kernels, which is relevant to startup behavior in vibration problems.","The numerical experiments show a crossover from short-time quasi-exponential response to long-time power-law response, giving a concrete signature of the multiscale kernel in mechanical vibration."],"supporting_citations":[{"why":"Supplies the previous first-order method, resolvent-based regularity estimates, and the local-modification result that the present second-order scheme extends.","marker":"[40]"},{"why":"Provides the convolution quadrature and Laplace-transform representation used in the regularity proofs and in the treatment of the weakly singular kernel.","marker":"[23]"},{"why":"Supplies the contour-integral and semigroup smoothing estimates used repeatedly to bound solution derivatives.","marker":"[2]"},{"why":"Defines the spectral Sobolev spaces and the Ritz projection approximation property used in the spatial finite element error estimate.","marker":"[19]"},{"why":"Motivates the variable-order fractional framework and the choice of the variable-exponent kernel.","marker":"[35]"},{"why":"Supplies the definition of the variable-exponent integral operator used in the model equation.","marker":"[22]"}],"fun_headline_variants":["Multiscale kernel bridges exponential and power-law decay","Variable-exponent kernel models viscoelastic crossover","Second-order scheme for multiscale viscoelastic equations","Kernel tames singularity, keeps power-law tail","Crossover dynamics from one multiscale Abel kernel"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that, under the regularity stated in Theorem 3.5, the linear-interpolation remainder for the convolution term is pointwise $O(\\tau^2)$ on every time interval, which requires $\\partial_t^2\\Delta u$ to be uniformly bounded in $L^2$ at the mesh-dependent evaluation point; Theorem 3.5 delivers only an $L^p$-in-time bound for finite $p$, so if that uniformity fails the clean quadratic error estimate is not justified.","fun_headline_variants_meta":{"raw":{"variants":["Multiscale kernel bridges exponential and power-law decay","Variable-exponent kernel models viscoelastic crossover","Second-order scheme for multiscale viscoelastic equations","Kernel tames singularity, keeps power-law tail","Crossover dynamics from one multiscale Abel kernel"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00043,"raw_usage":{"total_tokens":2157,"prompt_tokens":864,"completion_tokens":1293,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":480,"completion_tokens_details":{"reasoning_tokens":1217}},"tokens_in":480,"tokens_out":1293,"duration_ms":8857,"temperature":1.0,"reasoning_tokens":1217,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T13:34:56.426618+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take data satisfying Theorem 3.5 but with $\\partial_t^2\\Delta u$ having an $L^p$-integrable yet unbounded singularity at $t=0$ (for example behaving like $t^{-1/p}$), run the fully discrete scheme with decreasing $\\tau$ and fixed small $h$, and check whether $\\|u^m-U_h^m\\|/\\tau^2$ stays bounded. If it grows like $\\tau^{-\\varepsilon}$ or $\\log(1/\\tau)$, the regularity hypotheses in the theorem are insufficient for the stated second-order bound.","supporting_citations":[{"cited_title":"Zheng, Y","cited_arxiv_id":null,"evidence_quote":"Supplies the previous first-order method, resolvent-based regularity estimates, and the local-modification result that the present second-order scheme extends."},{"cited_title":"Lubich, Convolution quadrature and discretized operational calcu lus","cited_arxiv_id":null,"evidence_quote":"Provides the convolution quadrature and Laplace-transform representation used in the regularity proofs and in the treatment of the weakly singular kernel."},{"cited_title":"Akrivis, B","cited_arxiv_id":null,"evidence_quote":"Supplies the contour-integral and semigroup smoothing estimates used repeatedly to bound solution derivatives."},{"cited_title":"Jin, Fractional differential equations-an approach via fractio nal derivatives , Appl","cited_arxiv_id":null,"evidence_quote":"Defines the spectral Sobolev spaces and the Ritz projection approximation property used in the spatial finite element error estimate."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Motivates the variable-order fractional framework and the choice of the variable-exponent kernel."},{"cited_title":"Lorenzo and T","cited_arxiv_id":null,"evidence_quote":"Supplies the definition of the variable-exponent integral operator used in the model equation."}],"review_version":1}