{"id":"cf173cf8-b160-4db6-8bc4-92efc604db93","arxiv_id":"1908.09189","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The time-stepping discontinuous Galerkin method for fractional diffusion-wave problems is shown to achieve nearly optimal O(ln(T/τ)(sqrt(ln(1/h)) h^2 + τ)) accuracy even with nonsmooth data.","lead":"This paper proves convergence error estimates for a time-stepping discontinuous Galerkin method applied to a fractional diffusion-wave equation with rough data. The key results are nearly optimal rates in space and time even when the initial value or source term is nonsmooth, and the analysis is supported by four numerical experiments.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The advertised rates in Theorems 4.3 and 4.4 rest on the companion regularity Theorem 2.2, which is imported from the same authors' submitted paper [6] with only a 'trivial modification' of [6, Theorems 4.2]; until that transfer is verified, the error bounds are conditional.","rationale":"I read the paper in good faith and traced the argument for each advertised estimate. Theorems 4.1 and 4.2 are supported by the published semidiscrete results [5] together with the self-contained scalar DG analyses in Section 3. The proofs of Theorems 4.3 and 4.4 are internally coherent: the interpolation steps in Lemma 4.3, the Qτ/Pτ error bounds used in (69), and the ε-balancing in (68) all appear to work without introducing hidden powers of h or τ. The genuinely load-bearing point is Theorem 2.2, which is neither proved nor derived from a peer-reviewed source; the paper explicitly defers to an unpublished companion manuscript by the same authors with a 'trivial modification' claim. Since (9)-(11) are used at their endpoints and with constants that matter for the final rates, any failure or extra restriction there would directly void the central convergence results. This is a correctness risk, not merely a missing-reference issue, because the paper's novelty is precisely in extracting near-optimal nonsmooth-data rates from these regularity norms. The numerical experiments are encouraging but only verify rates relative to another discrete solution, so they cannot settle the regularity question. I therefore agree with the reader's conditional verdict and recommendation to require a full, self-contained proof or a published version of Theorem 2.2 before the main claims are accepted.","tokens_in":25517,"tokens_out":18732,"duration_ms":166609,"concrete_test":"Obtain arXiv:1901.02799 and transcribe the proof of [6, Theorem 4.2] into the notation of Section 2, checking each assertion (9), (10), (11) at the exact endpoints used here: γ=0 with β=α/(α+1) for Theorem 4.3, and γ=α+1/2 with β=0 and ε=(ln(1/h))^{-1} for Theorem 4.4. As a numerical cross-check, take Ω=(0,1), a single eigenmode φ_n, and f(t)=t^{α+1/2}φ_n (or f(t)=t^{α/2}φ_n for Theorem 4.3), then evaluate the explicit Mittag-Leffler representation of u to compute the ratios in (9)-(11) for n=1,...,10^4 and t∈(0,1). If any ratio grows like a negative power of λ_n or like (α+1/2-γ)^{-1}, the regularity transfer fails and the rates in Theorems 4.3-4.4 collapse.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 2.4 states the weak formulation and then asserts Theorem 2.2 with the sentence 'by a trivial modification of the proof of [6, Theorems 4.2]'. This theorem is the only source for the regularity norms entering the main error estimates: (9) supplies the 0H^{α+1/2}-to-0H^{α+3/2} and 0H^{1/2}(0,T;\\dot H^2) controls used in (68)-(69), (10) supplies the C([0,T];\\dot H^1) embedding needed in Theorem 4.3, and (11) supplies the near-\\dot H^2 spatial regularity that produces the h^{2(1-ε)} factor in Theorem 4.4. If the companion proof does not extend to the stated ranges — in particular to the endpoint γ=α+1/2 with the 1/√ε constant, to all 0≤β<∞, and to the negative-γ cases — then the right-hand sides of (45) and (46) are unsupported. The paper also leaves the bound (6) of Theorem 2.1 unproved, but that result is less central than Theorem 2.2. The numerical experiments in Section 5 are consistent with the claimed rates but do not isolate Theorem 2.2, since they compare discrete solutions only to each other and not to the true solution.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper analyzes the low-order time-stepping discontinuous Galerkin method for the fractional diffusion-wave equation u' - Δ D^{-α}_{0+}u = f, 0<α<1, with piecewise constant time and continuous piecewise linear space. The main results are four error estimates: Theorem 4.1 for nonsmooth initial data u0∈L²(Ω) and f=0; Theorem 4.2 for u0=0 and time-independent f∈L²(Ω); Theorem 4.3 for u0=0 and f∈L²(0,T; ˙H^{α/(α+1)}(Ω)), giving L∞(0,T;L²(Ω)) error O(h+√ln(1/h) τ^{1/2}); and Theorem 4.4 for u0=0 and f∈0H^{α+1/2}(0,T;L²(Ω)), giving the nearly optimal rate O(ln(T/τ)(√ln(1/h) h²+τ)). The proofs combine discrete Laplace transform arguments for two scalar fractional ODEs (Section 3), energy estimates and interpolation inequalities (Section 4), with the regularity of the weak solution imported from the companion paper [6]. Four numerical experiments are reported that are consistent with the claimed rates.","tokens_in":25842,"tokens_out":16486,"duration_ms":151876,"significance":"If the results are correct, this is a valuable contribution: it extends the convergence theory of the low-order DG method for fractional diffusion-wave problems to nonsmooth data, a regime that the earlier analyses in [10,14] did not cover. The Laplace-transform estimates in Section 3 are detailed and appear internally consistent, and the error estimates are stated with explicit structural constants rather than fitted parameters. The claimed rates are falsifiable and are tested numerically. The main weakness is that the central regularity Theorem 2.2 is not proved in the manuscript but imported from an unpublished companion paper by the same group, so the advertised error bounds are conditional on the validity of that transfer.","major_comments":[{"comment":"Theorem 2.2 is the load-bearing regularity input for Theorems 4.3 and 4.4, yet it is introduced with the sentence 'by a trivial modification of the proof of [6, Theorems 4.2]' and no proof is given. Specifically, estimate (9) with γ=0 provides the 0H¹(0,T; ˙H^{α/(α+1)}) and 0H^{-α}(0,T; ˙H^{2+α/(α+1)}) controls used in the proof of Theorem 4.3; estimate (9) with γ=α+1/2 provides the 0H^{α+3/2}(0,T;L²) and 0H^{1/2}(0,T; ˙H²) controls used in (69); estimate (11) with γ=α+1/2 and the 1/√ε constant is used in (68) to obtain the √ln(1/h) h² factor in Theorem 4.4; and estimate (10) provides the C([0,T]; ˙H¹) embedding used in Theorem 4.3. If [6] is not yet published, the manuscript is not self-contained on this point. The authors must either prove Theorem 2.2 in this paper or provide a complete, checkable derivation from [6, Theorems 4.2], including the endpoint γ=α+1/2 with the 1/√ε bound, the range γ≥(α-3)/4, and the range 0≤β<∞. Until then, the right-hand sides of (45) and (46) are unsupported.","section":"§2.4, Theorem 2.2"},{"comment":"The stability bound (6) in Theorem 2.1 is asserted with the sentence 'the proof of (6) is trivial and hence omitted.' Although this bound is apparently not used in the proofs of Theorems 4.1-4.4, it is still a stated theorem in a numerical analysis paper. The authors should either supply the proof, since it is claimed to be trivial, or downgrade the statement to a remark so that no unproved assertion remains in the main text. As it stands, this is a completeness gap, albeit not a load-bearing one for the central convergence claims.","section":"§2.3, Theorem 2.1(6)"}],"minor_comments":[{"comment":"Lemmas 2.1-2.4 are supported by the sentence 'For the proofs of the above lemmas, we refer the reader to [6, Section 3].' Since these lemmas are used throughout the paper and [6] is a companion preprint, it would improve self-containedness to include the proofs or to cite a published source for these standard fractional-calculus facts.","section":"§2.2"},{"comment":"The numerical experiments compare discrete solutions only against the finest discrete solution U^{11,16}, not against the exact solution. The word 'verify' in the section introduction is therefore too strong; the experiments are consistent with the predicted rates but do not test the constants or the precise dependence on the data norms. A sentence acknowledging this limitation would be appropriate.","section":"§5"},{"comment":"There are several typos and infelicities: 'thoerem' in the transposition-technique paragraph of §2.4, 'the this paper' in §2.2, 'combing' in the proofs of Theorems 3.2 and 3.3, and 'inerit' in the proof of Theorem 3.3. The reference list also has inconsistent formatting, e.g., page ranges for [7] and [8] use a hyphen in one place and an en-dash in another. These should be corrected.","section":"Throughout"},{"comment":"In the proof of Theorem 4.3, the step 'by Theorem 2.2 and Lemma 4.3' is quite compressed. It would help the reader if the authors explicitly identified which choices of the interpolation parameters (β,γ,r,s,θ) in Lemma 4.3 yield the required norm ‖u‖_{0H^{(1+ε+εα)/2}(0,T; ˙H^{1-ε})} and how Theorem 2.2 supplies the endpoint norms. This is not a correctness issue, but it would make the argument much easier to verify.","section":"§4.1"}],"recommendation":"major_revision","confidential_remarks":"The central concern is that Theorem 2.2, upon which the main rates depend, is imported from an unpublished companion paper by largely the same group. If the journal accepts submissions that rely on unpublished companion works, the authors should be required to include the proof or a fully detailed transfer argument. The current 'trivial modification' phrasing is not checkable. The rest of the technical content appears largely sound and the paper has genuine value, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nTwo things to know about this paper. It gives the first nonsmooth-data convergence rates for the piecewise-constant-in-time / piecewise-linear-in-space DG method for u' − Δ D^{−α} u = f. The main rates in Theorems 4.1–4.4 are genuinely new relative to the earlier work [10,14,5], including the O(ln(T/τ)(√ln(1/h) h² + τ)) bound for f ∈ 0H^{α+1/2}(0,T;L²). The technical core—discrete Laplace transform analysis of two fractional ODEs in Section 3, plus the interpolation arguments in Section 4—is carefully executed. I found no fitted parameters and no circular step; the ψ-function contour estimates are explicitly worked out.\n\nWhere are the soft spots? The stability bound (6) in Theorem 2.1 is asserted with “trivial and hence omitted”. That is a small gap, likely patchable with the paper's own tools, but a referee should ask to see it. The larger issue is Theorem 2.2, the regularity theorem that drives the final rates in Theorems 4.3 and 4.4. It is imported from the same authors' companion paper [6] via “a trivial modification of the proof of [6, Theorems 4.2]”, and no proof or statement of the needed endpoint cases appears here. The ranges used—particularly γ = α + 1/2 with the 1/√ε factor, all β ≥ 0, and the negative-γ cases—are exactly the ones the error bounds require. If that theorem does not extend as claimed, the right-hand sides of (45) and (46) are unsupported. This is a genuine gap, not a manufactured one. My sense is that the result is probably true, because the companion exists, but a standalone paper should not ask the reader to take that on faith.\n\nThe numerical experiments are consistent with the predicted rates, but they are self-referential: they compare discrete solutions to a finer discrete solution, never to the true solution, so they do not independently validate Theorem 2.2. That limits their probative value but does not contradict the theory.\n\nOverall: this paper deserves a serious referee. It is a real advance for nonsmooth-data analysis of fractional diffusion-wave discretizations, and the proof structure, apart from the import from [6], is sound. I would send it to review with the instruction to verify Theorem 2.2—either by requesting the companion proof or by asking the authors to include the transfer argument—and to fill the gap in (6). With those resolved, I would be comfortable citing it.","headline":"First nonsmooth-data convergence rates for the low-order time-stepping DG method for fractional diffusion-wave equations; the analysis is careful, but the key rates rest on a regularity theorem imported from a companion paper, so the result is conditional.","tokens_in":26385,"tokens_out":3331,"would_cite":false,"duration_ms":37482,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65M60","65M12","65M15","35R11"],"pacs":[],"model":"deepseek-v4-flash","headline":"A low-order time-stepping discontinuous Galerkin method is shown to converge at near-optimal rates for fractional diffusion-wave equations even when the data are nonsmooth.","keywords":["fractional diffusion-wave equation","discontinuous Galerkin method","time stepping","nonsmooth data","discrete Laplace transform","convergence analysis","error estimates","fractional ordinary equation"],"falsifier":"Take $u_0=0$, $f(t,x) = t^{\\alpha - 1/2}\\sin(\\pi x)$ on $\\Omega=(0,1)$, $T=1$, which lies in ${}_0H^{\\alpha+1/2}(0,T;L^2(\\Omega))$. Compute the scheme with $h$ small and $\\tau = 2^{-n}$ for $n=8,\\dots,14$; if $\\|u-U\\|_{L^\\infty(0,T;L^2)}/(\\tau \\ln(1/\\tau))$ grows without bound as $n$ increases while the source norm stays finite, the claimed $O(\\ln(T/\\tau)\\tau)$ rate fails.","tokens_in":25325,"feed_emoji":"📉","tokens_out":6289,"duration_ms":60408,"temperature":0.7,"pith_summary":"The paper analyzes the low-order time-stepping discontinuous Galerkin method for the fractional diffusion-wave equation $u' - \\Delta D^{-\\alpha}_{0+} u = f$, using piecewise constant functions in time and continuous piecewise linear functions in space. It establishes convergence with nearly optimal rates under nonsmooth data: for $f=0$ and $u_0 \\in L^2(\\Omega)$ the error at time $t_j$ is $O(h^2 t_j^{-\\alpha-1} + \\tau t_j^{-1})$, for $u_0=0$ and $f \\in L^2(0,T;\\dot H^{\\alpha/(\\alpha+1)}(\\Omega))$ it is $O(h + \\sqrt{\\ln(1/h)}\\,\\tau^{1/2})$, and for $u_0=0$ and $f \\in {}_0H^{\\alpha+1/2}(0,T;L^2(\\Omega))$ it is $O(\\ln(T/\\tau)(\\sqrt{\\ln(1/h)}\\,h^2 + \\tau))$. These are the first nonsmooth-data convergence results for this scheme, closing a gap left by earlier smooth-data analyses and showing that uniform time grids suffice for first-order temporal accuracy up to a logarithmic factor.","feed_headline":"Fractional wave DG method proven near-optimal for rough data","feed_subtitle":"The simplest low-order DG time step keeps near-optimal accuracy even for nonsmooth source terms and initial values.","key_machinery":"The proof is carried by the discrete Laplace transform of the numerical scheme together with an auxiliary function $$\\psi(z) = \\frac{e^z - 1}{\\Gamma(2+\\$\\alpha$)} \\sum_{k=1}^\\infty $k^{{1+\\alpha}}$ $e^{{-kz}}$,$$ which encodes the effect of the piecewise-constant time discretization on the fractional integral. The argument shows that $1 + \\mu\\psi(z)$ with $\\mu = \\lambda\\tau^{1+\\alpha}$ stays away from zero in a sector of the complex plane (Lemma 3.1) and satisfies pointwise lower bounds (Lemma 3.2), yielding $O(k^{-1})$ estimates for the scalar fractional ordinary equations to which the full problem is reduced (Theorems 3.1–3.3). Spatial and temporal interpolation estimates then transfer these scalar bounds to the full discrete solution.","core_discovery":"The central claim is that the low-order time-stepping DG scheme achieves nearly optimal convergence uniformly in time under nonsmooth data. The main estimate, stated as Theorem 4.4, is $$\\|u - U\\|_{L^\\infty(0,T;$L^{2}$(\\$\\Omega$))} \\lesssim \\ln(T/\\tau)\\bigl(\\sqrt{\\ln(1/h)}\\,$h^{2}$ + \\tau\\bigr)\\, \\|f\\|_{{}$_0H^{{\\alpha+1/2}}$(0,T;$L^{2}$(\\$\\Omega$))}$$ for $u_0=0$, and Theorem 4.1 gives the companion bound $\\|u(t_j) - U_j\\|_{L^2(\\Omega)} \\lesssim (h^2 t_j^{-\\alpha-1} + \\tau t_j^{-1})\\|u_0\\|_{L^2(\\Omega)}$. Because the solution of this fractional problem generally develops a singularity at $t=0$, the achievement is that no graded temporal meshes and no extra regularity of the initial value are required for nearly optimal rates.","pith_inferences":["The same discrete-Laplace-transform machinery likely extends to semilinear versions of the fractional diffusion-wave equation, provided a matching regularity theorem is available.","The logarithmic factor $\\ln(T/\\tau)$ in Theorem 4.4 may be an artifact of the duality argument and the summation of jumps; numerical experiments suggest that on data with mild smoothness the observed rate can be closer to $O(\\tau)$ than $O(\\tau \\ln(1/\\tau))$.","The sharp dependence of the auxiliary bound on $\\mu = \\lambda\\tau^{1+\\alpha}$ indicates that the stability constant may degrade as $\\alpha$ approaches 0 or 1; practitioners using extreme $\\alpha$ values might need to check whether the predicted rates are still visible at practical resolutions.","Because the load-bearing regularity estimates are imported from a companion paper with a 'trivial modification', a careful independent check of Theorem 2.2 for the full parameter range would be prudent before relying on the advertised rates."],"forward_implications":["Uniform time grids suffice: the temporal singularity at $t=0$ caused by nonsmooth data does not force graded meshes for the $O(\\tau)$ rate up to a logarithmic factor.","For a constant-in-time source $f(t)=v \\in L^2(\\Omega)$, the error is $O(t^{-\\alpha} h^2 + \\tau)$, so the spatial error has a $t^{-\\alpha}$ singularity while the temporal rate remains first order.","When $f \\in H^{\\alpha+1/2}$ with $f(0) \\ne 0$, combining Theorems 4.2 and 4.4 gives a bound containing both a $t^{-\\alpha}$ spatial term and a logarithmic temporal term (Remark 4.3).","The analyzed scheme coincides with the first discretization in the earlier work of McLean, Thomée, and Wahlbin [10], so the results close the nonsmooth-data gap for that method."],"supporting_citations":[{"why":"Supplies the weak-solution regularity estimates (Theorem 2.2) that all main error bounds rely on.","marker":"[6]"},{"why":"Provides the contour-integral representation of the solution and the nonsmooth-data error estimates for the spatially discrete problem used in Theorems 4.1 and 4.2.","marker":"[5]"},{"why":"Introduces the low-order time-stepping DG method and gives the stability bound in Theorem 2.1, the smooth-data baseline the paper extends.","marker":"[14]"},{"why":"Supplies the discrete Laplace transform technique and the integral representation of the auxiliary function used in Section 3.","marker":"[12]"},{"why":"Provides the time-stepping DG framework and the jump identity used as Lemma 4.4.","marker":"[18]"},{"why":"Supplies the interpolation-space tools behind Lemmas 4.1–4.3, including the K-method and the embedding estimates for fractional Sobolev spaces.","marker":"[17]"},{"why":"Gives the Hankel integral representation of the auxiliary function $\\psi$, used to continue it analytically and control its zeros.","marker":"[19]"},{"why":"Supplies the fractional-calculus composition and duality identities used throughout the proof.","marker":"[16]"}],"fun_headline_variants":["Low-order DG time stepping hits near-optimal rates for rough fractional waves","Simple DG scheme achieves near-optimal accuracy for rough data","No grading needed: low-order DG stays near-optimal for nonsmooth waves","DG with piecewise constants: near-optimal for rough fractional data","Robust near-optimality for nonsmooth fractional waves with simple DG"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof's rates rest on the weak-solution regularity bounds of Theorem 2.2, which the paper borrows from a companion analysis with only a 'trivial modification of the proof'; if those bounds fail, the error estimates have no foundation.","fun_headline_variants_meta":{"raw":{"variants":["Low-order DG time stepping hits near-optimal rates for rough fractional waves","Simple DG scheme achieves near-optimal accuracy for rough data","No grading needed: low-order DG stays near-optimal for nonsmooth waves","DG with piecewise constants: near-optimal for rough fractional data","Robust near-optimality for nonsmooth fractional waves with simple DG"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000931,"raw_usage":{"total_tokens":3948,"prompt_tokens":870,"completion_tokens":3078,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":486,"completion_tokens_details":{"reasoning_tokens":2985}},"tokens_in":486,"tokens_out":3078,"duration_ms":21395,"temperature":1.0,"reasoning_tokens":2985,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:19:58.970159+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take $u_0=0$, $f(t,x) = t^{\\alpha - 1/2}\\sin(\\pi x)$ on $\\Omega=(0,1)$, $T=1$, which lies in ${}_0H^{\\alpha+1/2}(0,T;L^2(\\Omega))$. Compute the scheme with $h$ small and $\\tau = 2^{-n}$ for $n=8,\\dots,14$; if $\\|u-U\\|_{L^\\infty(0,T;L^2)}/(\\tau \\ln(1/\\tau))$ grows without bound as $n$ increases while the source norm stays finite, the claimed $O(\\ln(T/\\tau)\\tau)$ rate fails.","supporting_citations":[{"cited_title":"Convergence analysis of a Petrov-Galerkin method for fractional wave problems with nonsmooth data","cited_arxiv_id":"1901.02799","evidence_quote":"Supplies the weak-solution regularity estimates (Theorem 2.2) that all main error bounds rely on."},{"cited_title":"Lubich, I","cited_arxiv_id":null,"evidence_quote":"Provides the contour-integral representation of the solution and the nonsmooth-data error estimates for the spatially discrete problem used in Theorems 4.1 and 4.2."},{"cited_title":"Mustapha and W","cited_arxiv_id":null,"evidence_quote":"Introduces the low-order time-stepping DG method and gives the stability bound in Theorem 2.1, the smooth-data baseline the paper extends."},{"cited_title":"McLean and K","cited_arxiv_id":null,"evidence_quote":"Supplies the discrete Laplace transform technique and the integral representation of the auxiliary function used in Section 3."},{"cited_title":"Thom´ ee","cited_arxiv_id":null,"evidence_quote":"Provides the time-stepping DG framework and the jump identity used as Lemma 4.4."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the interpolation-space tools behind Lemmas 4.1–4.3, including the K-method and the embedding estimates for fractional Sobolev spaces."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the Hankel integral representation of the auxiliary function $\\psi$, used to continue it analytically and control its zeros."},{"cited_title":"Samko, A","cited_arxiv_id":null,"evidence_quote":"Supplies the fractional-calculus composition and duality identities used throughout the proof."}],"review_version":1}