{"id":"45deb5fc-1013-44b4-be2b-20008fe0637d","arxiv_id":"1908.01136","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A generalized convolution quadrature theory with a shift parameter theta is developed for fractional operators, unifying many existing shifted numerical schemes.","lead":"This paper builds a unified theory for approximating fractional integrals and derivatives at shifted grid points, generalizing Lubich's convolution quadrature with a shift parameter. It also proposes new quadrature formulas, analyzes their stability, and applies them to a time-fractional diffusion equation.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Superconvergence conditions Eqs. (2.38) and (2.45) fail the paper's own consistency test; the order-3 claim for Eq. (2.37) is not satisfied by the stated theta.","rationale":"The reader's weakest_assumption pointed to the unproved identities (2.4)-(2.5). Those identities are actually checkable directly from the convolution structure and homogeneity under scaling, so that particular concern is weaker than stated. However, the reader's secondary observation about Eqs. (2.38) and (2.45) is correct and is the main load-bearing issue for the paper's advertised superconvergence novelty. A direct Taylor expansion of the paper's own consistency condition shows the theta in (2.38) does not make Eq. (2.37) order 3; the two-order-3 values of theta satisfy theta=(1-alpha ± sqrt(1-alpha/3))/2. This error affects Example 2 and Example 4, and any downstream claim relying on those specific superconvergence points, but it does not by itself invalidate Theorem 2.8, whose proof structure follows Lubich once (2.4)-(2.5) are granted. The paper needs revision of these examples and re-examination of the claimed new superconvergence formulas, so the existing CONDITIONAL verdict remains appropriate.","tokens_in":18851,"tokens_out":31570,"duration_ms":276246,"concrete_test":"Set alpha=1 and take theta=1-1/2-sqrt(2/3)/2 as in (2.38). Expand F(h)=h e^{theta h}(1-e^{-h})^{-1}[1-(1/2+theta)(1-e^{-h})] to order h^3 with a computer algebra system. The h and h^2 coefficients are 0 and approximately 0.0791, so F(h)=1+0.0791h^2+O(h^3), giving consistency order 2, not 3. Repeat with theta=(1-alpha+sqrt(1-alpha/3))/2; for alpha=1 this is sqrt(2/3)/2≈0.408, and the h^2 coefficient vanishes, so order 3 is obtained. Finally, test m=1 in (2.45) against the same expansion to confirm that the claimed reduction to (2.38) carries the same error.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The equivalence theorem 2.8 is probably salvageable: identities (2.4)-(2.5) do follow by writing Omega_h f as a discrete convolution evaluated at x+theta h, followed by scaling, so the reader's weakest assumption is not the main defect. The real defect is in the flagship superconvergence examples. For omega(xi)=(1-xi)^(-alpha)[1-(alpha/2+theta)(1-xi)] in Eq. (2.37), consistency of order 3 requires the coefficient of h^2 in F(h)=h^alpha e^{theta h}(1-e^{-h})^(-alpha)[1-(alpha/2+theta)(1-e^{-h})] to vanish. A direct Taylor expansion gives F(h)=1+[alpha(3alpha-1)/24 - (alpha/2+theta)alpha/2 + (alpha/2+theta)/2 - theta^2/2]h^2+O(h^3). Setting this to zero yields theta=(1-alpha ± sqrt(1-alpha/3))/2. For alpha=1, the value from (2.38) is theta=1-1/2-(1/2)sqrt(2/3)≈0.09175, and the h^2 coefficient is 1/12 - theta^2/2≈0.0791≠0, so the scheme is only order 2, not order 3. Since (2.45) reduces to (2.38) when m=1, the defect propagates to Example 4. The central framework and Theorem 2.8 are not directly refuted, but an advertised novelty, the new superconvergence points, is unsupported.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a 'shifted convolution quadrature' (SCQ) framework for approximating the Riemann-Liouville fractional integral I^α at shifted nodes x_{n-θ}. The main theoretical result, Theorem 2.8, states that under a condition-ω, a SCQ is convergent of order p if and only if it is stable and consistent of order p, with consistency defined by h^α e^{θh}ω(e^{-h}) = 1 + O(h^p). The authors derive a transformation theorem from classical CQ generating functions, introduce shift-generalized Newton-Gregory formulas, and present several examples (shifted Grünwald, WSGL, generalized BDF2-θ) including claimed new third-order superconvergence schemes. Section 3 studies stability regions as a function of θ, and Section 4 applies a second-order shift-generalized Newton-Gregory formula to a time-fractional diffusion equation with finite elements, deriving stability and error estimates and reporting numerical tests.","tokens_in":19187,"tokens_out":22817,"duration_ms":196054,"significance":"If Theorem 2.8 and the examples are correct, the framework would unify many existing shifted schemes under one theory and explain superconvergence as a choice of θ. The paper is explicit and constructive: it gives concrete generating functions, stability-region descriptions, and numerical verification, which are strengths. However, the advertised superconvergence points in Examples 2 and 4 are algebraically wrong as stated, and the central equivalence theorem relies on unproved identities and omitted lemma details. The general framework is likely salvageable, but the current version overclaims.","major_comments":[{"comment":"The proof of the main equivalence theorem is not self-contained. The identities (2.4) and (2.5) are load-bearing: they are used to pass from consistency to convergence in the usual Lubich argument, but the text only says their correctness 'can be directly checked' and gives no derivation. Lemma 2.5 and Lemma 2.6, which are the shifted analogues of Lubich's lemmas, are both stated with proofs omitted ('the rest argument ... is omitted here' and 'almost the same as Lemma 3.2 in [1]'). Please give complete proofs of (2.4)-(2.5) and of the two lemmas in the shifted setting, or explicitly identify which steps of Lubich's proof fail to transfer.","section":"Section 2, Theorem 2.8 and Eqs. (2.4)-(2.5)"},{"comment":"The advertised order-3 superconvergence point is incorrect. For ω(ξ)=(1-ξ)^{-α}[1-(α/2+θ)(1-ξ)], a direct expansion gives h^α e^{θh}ω(e^{-h}) = 1 + [α(5-3α)/24 + θ(1-α)/2 - θ^2/2]h^2 + O(h^3). The h^2 coefficient vanishes for θ = ((1-α) ± sqrt(1-α/3))/2, not for θ = 1 - α/2 - (1/2)sqrt(1-α/3) as stated in (2.38). For α=0, (2.38) gives θ=1/2, but the expansion is 1+h^2/8+O(h^4), so only order 2; for α=1, (2.38) gives θ≈0.0918, while the required values are ±1/sqrt(6)≈±0.408. Thus the claim that the scheme (2.37) is consistently of order 3 at the stated θ is false.","section":"Section 2, Example 2, Eqs. (2.37)-(2.38)"},{"comment":"The order-3 claim for the two-term WSGL-type generating function inherits the algebraic defect of Example 2: for m=1, (2.45) reduces to (2.38), which is not the correct root of the consistency condition computed above. The statement that the coefficients in (2.44) 'can be easily checked' to give stability and consistency of order 3 is not accompanied by the check, and the special case m=1 shows the stated formula does not have the claimed order. Please correct the condition or withdraw the superconvergence claim, and include the actual consistency expansion for general m.","section":"Section 2, Example 4, Eqs. (2.43)-(2.45)"}],"minor_comments":[{"comment":"The two displayed formulas are both marked with convergence orders, but the subsequent discussion of superconvergence switches to the two-term formula (2.37), which is not the same as the second formula in (2.36); please clarify what is being claimed.","section":"Section 2, Eq. (2.36)"},{"comment":"The sentence 'One can check that this choice of θ satisfies the condition-ω' should be replaced by an explicit verification, especially because the corrected consistency condition has two roots and the stability and condition-ω requirements may select only one of them.","section":"Section 2, Eq. (2.38)"},{"comment":"The proof is omitted with a reference to Lubich's Theorem 2.1; since the dependence of the stability region on θ is a stated contribution, include at least a brief derivation of the two boundary sets in (3.12).","section":"Section 3, Theorem 3.4"},{"comment":"The proof asserts that Θ_n is increasing and that the limit exists, but gives no argument for either statement; please complete the proof or state the missing estimates so that the bound (4.13) follows.","section":"Section 4, Lemma 4.2"},{"comment":"In the rows for α=0.5, θ=0.1 without the starting part, the reported rates jump from 2.4993 to -0.9761 as τ is refined; the text should comment on this non-monotone behavior, since it may otherwise cast doubt on the convergence claim.","section":"Section 4, Table 5"}],"recommendation":"major_revision","confidential_remarks":"The superconvergence error is contained in Examples 2 and 4, which are presented as new contributions; however, Theorem 2.8 and the general SCQ framework do not depend on those examples. The paper is likely salvageable, but the authors need to correct or remove the erroneous claims and supply the missing proofs. I would not recommend rejection based on the current evidence."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The useful part of this paper is the framework: defining shifted convolution quadrature with a parameter theta and proving an equivalence between stability, consistency, and convergence in Theorem 2.8, in the style of Lubich. That is a real generalization and should be of interest to anyone working on numerical fractional calculus. Theorem 2.11, which converts a CQ generating function into an SCQ one, is a workable tool, and the extension of the correction technique to shifted nodes is worth having. The PDE application in Section 4, using a second-order shift-generalized Newton-Gregory scheme with a discrete fractional Gronwall inequality, is competently done and the numerical tests support it.\n\nThe problem is with the advertised superconvergence examples. Eq. (2.38) claims the generating function (2.37) is order 3 when theta = 1 - alpha/2 - (1/2)sqrt(1 - alpha/3). Direct Taylor expansion against the paper's own consistency condition (2.7) shows that for alpha=1, the h^2 coefficient is about 0.0791, not zero, so the scheme is only order 2. Since (2.45) reduces to (2.38) when m=1, the same defect propagates. That means the claimed new superconvergence points are unsupported, and the phrase 'one can easily check' is hiding a real bug.\n\nOther soft spots are less serious. The proofs of Lemmas 2.5 and 2.6 are skipped with 'same as Lubich,' and the identities (2.4)-(2.5) are asserted as directly checkable. They are in fact checkable, which is good, but the paper should show the calculation rather than leave it to the reader. If it turns out that the commutativity fails for any shifted scheme, Theorem 2.8 collapses; I don't think it does, but that's exactly the spot to be careful about.\n\nSo: the central framework is probably correct and useful, but the novelty claim about superconvergence is not. This paper deserves a serious referee, but it needs a major revision that either fixes those Taylor expansions or removes the examples. I would not cite it in its current form.","headline":"A useful shifted-CQ framework undercut by erroneous superconvergence claims that need correcting before publication.","tokens_in":19735,"tokens_out":2388,"would_cite":false,"duration_ms":21913,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["26A33","65D30","65L20","65M12","65M60"],"pacs":[],"model":"deepseek-v4-flash","headline":"A shifted convolution quadrature converges to order p exactly when it is stable and consistent to order p, provided the generating function satisfies a mild zero-free condition.","keywords":["shifted convolution quadrature","Riemann-Liouville fractional calculus","generating functions","stability regions","superconvergence","fractional Gronwall inequality","time-fractional diffusion equation","numerical convergence theory"],"falsifier":"Compute both sides of the homogeneity identity $(E_h^{\\alpha}t^{\\beta-1})(x)=x^{\\alpha+\\beta-1}(E_{h/x}^{\\alpha}t^{\\beta-1})(1)$ for, say, $\\alpha=1/2$, $\\theta=1/4$, $\\beta=3/2$, using the second-order shift-generalized Newton-Gregory weights; a single value of $x/h$ where the two sides differ would show the proof of Theorem 2.8 does not transfer. The commutativity identity $E_h^{\\alpha}(f*g)=(E_h^{\\alpha}f)*g$ can be tested similarly on $f(t)=t^{\\beta-1}$ and $g(t)=t^{\\gamma-1}$ for small $\\beta,\\gamma$.","tokens_in":18646,"feed_emoji":"🧮","tokens_out":9991,"duration_ms":87141,"temperature":0.7,"pith_summary":"Shifted convolution quadrature approximates the Riemann-Liouville fractional integral $I^\\alpha f$ at a shifted node $x_{n-\\theta}$ by a weighted history sum plus starting corrections. This paper's central aim is to give such shifted schemes the same kind of complete theory that classical convolution quadrature has: a scheme is convergent of order $p$ if and only if it is stable and consistent of order $p$, with consistency read from the generating function through $h^\\alpha e^{\\theta h}\\omega(e^{-h})=1+O(h^p)$. If true, this unification matters because it brings many existing second-order schemes that evaluate fractional integrals at half-integer or otherwise shifted nodes under one framework, so results on starting weights, correction, superconvergence, and stability regions transfer to all of them at once. The paper also uses the theory to design new generating functions and to analyze a second-order scheme for the time-fractional diffusion equation.","feed_headline":"Stable plus consistent decides shifted fractional convergence","feed_subtitle":"The theory covers second-order shifted rules and preserves correction and stability tools for fractional PDEs.","key_machinery":"The central object is the generating function $\\omega(\\xi)=\\sum_{j=0}^{\\infty}\\omega_j\\xi^j$ for the convolution weights; all three properties in the main theorem are conditions on it. Stability is a bound on the coefficients, consistency is the behavior of $h^\\alpha e^{\\theta h}\\omega(e^{-h})$ as $h\\to0$, and Condition-$\\omega$ asks that $\\omega$ be analytic and zero-free in the open unit disk, with boundary exponents no larger than $\\Re\\alpha$. The argument is carried by two identities stated in the paper without proof: the convolution-error identity $E_h^{\\alpha}(f*g)=(E_h^{\\alpha}f)*g$ and the homogeneity identity $(E_h^{\\alpha}t^{\\beta-1})(x)=x^{\\alpha+\\beta-1}(E_{h/x}^{\\alpha}t^{\\beta-1})(1)$. For constructing schemes, the key expansion is $\\sum_{i=0}^{\\infty}\\gamma_i(1-\\xi)^i=\\xi^{\\theta}(-\\ln\\xi/(1-\\xi))^{-\\alpha}$, whose truncated versions define the shift-generalized Newton-Gregory formulas.","core_discovery":"On the paper's own terms, the central discovery is Theorem 2.8: a shifted convolution quadrature whose convolution weights are generated by $\\omega(\\xi)$ and satisfy Condition-$\\omega$ converges to $I^\\alpha$ at order $p$ exactly when it is stable and consistent of order $p$. Stability is the coefficient bound $\\omega_n=O(n^{\\alpha-1})$; consistency is the shifted condition $h^\\alpha e^{\\theta h}\\omega(e^{-h})=1+O(h^p)$; convergence is tested on the power functions $t^{\\beta-1}$ for $\\beta$ outside $\\{0,-1,-2,\\dots\\}$. The proof works by generalizing the two classical consistency lemmas to the shifted setting, and it leans on two algebraic properties of the convolution error, equations (2.4) and (2.5), which let the error commute with convolution and rescale with $h$. Alongside the equivalence, the paper proves that any classical convolution-quadrature generating function can be shifted to order $p$ by multiplying by a truncated series for $\\xi^{\\theta}$, and it introduces the shift-generalized Newton-Gregory family whose coefficients are fixed by $\\xi^{\\theta}(-\\ln\\xi/(1-\\xi))^{-\\alpha}$.","pith_inferences":["The two unproved identities (2.4)-(2.5) are the natural place to probe the theory's boundary: if they fail for some nonzero $\\theta$, the equivalence theorem would still hold for many practical schemes, but its proof would need a different route; a direct symbolic check on specific power functions would map where that boundary lies.","The consistency condition suggests a systematic superconvergence search: since the leading error is read off from $h^\\alpha e^{\\theta h}\\omega(e^{-h})$, one can tune $\\theta$ to cancel one error term at a time, potentially producing shifted schemes of arbitrarily high order from low-order base rules without adding more coefficients.","The strong dependence of stability regions on $\\theta$ opens an optimization problem the paper touches only indirectly: among all SCQ schemes of a fixed order, which shift parameter and generating-function structure maximize the sector of stability while keeping the convolution weights nonnegative enough for Grönwall-type arguments?"],"forward_implications":["Every classical convolution-quadrature generating function, including fractional BDF-$p$, the fractional trapezoidal rule, and the fractional BN-$\\theta$ method, can be converted into a shifted generating function of the same order by multiplying with a truncated expansion of $\\xi^{\\theta}$ (Theorem 2.11).","The starting-weight correction technique from classical convolution quadrature survives the shift: for functions with the mild boundary singularity $x^{\\beta-1}g(x)$, suitable starting weights restore the full order $p$ uniformly on intervals bounded away from zero (Theorem 2.9).","The shift parameter controls absolute stability: for the same base rule, different $\\theta$ values can give $A(\\pi/2)$-stability, conditional stability, or no stable real interval at all, so the theory turns $\\theta$ into a design parameter.","The second-order shift-generalized Newton-Gregory scheme applied to the time-fractional diffusion equation satisfies the discrete fractional Grönwall inequality for a range of $\\theta$, yielding stability and an optimal $(\\tau^2 + h^{r+1})$ error estimate.","The superconvergence seen in existing shifted schemes, such as taking $\\theta=-\\alpha/2$, is explained as a special choice of $\\theta$ that raises the consistency order of a generating function that would otherwise be lower order."],"supporting_citations":[{"why":"supplies the classical convolution quadrature theory, its equivalence theorem, and the generating-function framework that the shifted theory generalizes.","marker":"[1]"},{"why":"gives the classical equivalence result for linear multistep methods that Theorem 2.8 extends to fractional shifted schemes.","marker":"[3]"},{"why":"provides the second- and third-order shifted approximations whose generating functions the SCQ framework reproduces and extends.","marker":"[4]"},{"why":"documents superconvergence in finite-difference schemes for subdiffusion, the phenomenon the paper reinterprets through shifted consistency.","marker":"[5]"},{"why":"defines a class of second-order weighted and shifted fractional difference operators that appear here as stable and consistent SCQs.","marker":"[6]"},{"why":"supplies the discrete fractional Grönwall inequality used to prove stability and the $\\tau^2+h^{r+1}$ error estimate for the time-fractional diffusion equation.","marker":"[13]"},{"why":"introduces the shifted fractional difference formula that the SCQ framework covers as a first-order shifted rule.","marker":"[16]"},{"why":"gives explicit higher-order fractional derivative formulas, including the BDF2-$\\theta$ style generating function analyzed as an SCQ.","marker":"[18]"}],"fun_headline_variants":["Shifted quadrature: stable and consistent means convergent","Two conditions decide shifted fractional convergence","The shift rule: stability plus consistency yields convergence","Convergence check: stable and consistent shifted weights"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing assumption is that the shifted convolution error obeys the two identities (2.4)-(2.5), commuting with convolution and rescaling with $h$ exactly as the classical error does, with the paper only saying these are directly checked and giving no proof.","fun_headline_variants_meta":{"raw":{"variants":["Shifted quadrature: stable and consistent means convergent","Two conditions decide shifted fractional convergence","The shift rule: stability plus consistency yields convergence","Convergence check: stable and consistent shifted weights"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000517,"raw_usage":{"total_tokens":2503,"prompt_tokens":938,"completion_tokens":1565,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":554,"completion_tokens_details":{"reasoning_tokens":1507}},"tokens_in":554,"tokens_out":1565,"duration_ms":12083,"temperature":1.0,"reasoning_tokens":1507,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:23:21.835459+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute both sides of the homogeneity identity $(E_h^{\\alpha}t^{\\beta-1})(x)=x^{\\alpha+\\beta-1}(E_{h/x}^{\\alpha}t^{\\beta-1})(1)$ for, say, $\\alpha=1/2$, $\\theta=1/4$, $\\beta=3/2$, using the second-order shift-generalized Newton-Gregory weights; a single value of $x/h$ where the two sides differ would show the proof of Theorem 2.8 does not transfer. The commutativity identity $E_h^{\\alpha}(f*g)=(E_h^{\\alpha}f)*g$ can be tested similarly on $f(t)=t^{\\beta-1}$ and $g(t)=t^{\\gamma-1}$ for small $\\beta,\\gamma$.","supporting_citations":[{"cited_title":"Lubich, Discretized fractional calculus, SIAM J","cited_arxiv_id":null,"evidence_quote":"supplies the classical convolution quadrature theory, its equivalence theorem, and the generating-function framework that the shifted theory generalizes."},{"cited_title":"Dahlquist, Convergence and stability in the numerica l integration of ordinary differential equations, Math","cited_arxiv_id":null,"evidence_quote":"gives the classical equivalence result for linear multistep methods that Theorem 2.8 extends to fractional shifted schemes."},{"cited_title":"Dimitrov, Numerical approximations for fractional d ifferential equations, Journal of Fractional Cal- culus and Applications, 2015, 5(3S)(22): 1-45","cited_arxiv_id":null,"evidence_quote":"provides the second- and third-order shifted approximations whose generating functions the SCQ framework reproduces and extends."},{"cited_title":"Gao, H.W","cited_arxiv_id":null,"evidence_quote":"documents superconvergence in finite-difference schemes for subdiffusion, the phenomenon the paper reinterprets through shifted consistency."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"defines a class of second-order weighted and shifted fractional difference operators that appear here as stable and consistent SCQs."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the discrete fractional Grönwall inequality used to prove stability and the $\\tau^2+h^{r+1}$ error estimate for the time-fractional diffusion equation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"introduces the shifted fractional difference formula that the SCQ framework covers as a first-order shifted rule."},{"cited_title":"Gunarathna, H.M","cited_arxiv_id":null,"evidence_quote":"gives explicit higher-order fractional derivative formulas, including the BDF2-$\\theta$ style generating function analyzed as an SCQ."}],"review_version":1}