{"id":"e25b269e-eece-43a1-a3bb-54909d84c455","arxiv_id":"1908.09143","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A pseudospectral method computes the one-dimensional fractional Laplacian on R without domain truncation by deriving explicit series formulas for the operator applied to Fourier modes after mapping R to [0,π].","lead":"This paper develops a numerical method to compute the fractional Laplacian on the whole real line by mapping the line to a circle and computing the operator on each Fourier mode. The method avoids truncating the domain, and the authors demonstrate its accuracy on test functions and in simulations of Fisher's equation.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The CAS integrals look corroborated; the untested Fourier-extension step is the real gap, since a smooth bounded function with unequal limits can reduce the method to algebraic convergence.","rationale":"The reader's conditional verdict is appropriate, but the specific weakest assumption identified by the reader (the Mathematica evaluation of I1 and I2) is substantially mitigated by the independent Gaussian benchmark: Eq. (39) is a standard exact result, and the method's agreement with it for both even and odd extensions across 199 values of α would be very unlikely if formulas (18) were wrong. The genuinely load-bearing gap is the unexamined representation step: for a general bounded C² function, the Fourier series is taken over a 2π-periodic extension of u(s), and convergence depends on the extension's smoothness at the endpoints. The authors themselves emphasize that the extension choice matters (§3.2), but they provide no error analysis and no test on a function with unequal limits at ±∞ and non-flat endpoint derivatives. Such a function, e.g. arctan x, yields only algebraic Fourier decay, so the claimed accuracy for the full stated class is not established. This concern does not invalidate Theorem 2.2 or the numerical experiments as reported; it narrows the supportable scope of the abstract's claim. Hence the reader's CONDITIONAL verdict stands unchanged, but the condition should be on the representation/extension step rather than primarily on the CAS integrals.","tokens_in":19695,"tokens_out":37798,"duration_ms":371311,"concrete_test":"Take u(x)=arctan x with L=1. For α=0.5, 1, and 1.5, compute a reference (-Δ)^{α/2}u at the grid points by direct adaptive quadrature of the integral in (6) over R; for α=1 the exact value x/(1+x²) is available as a cross-check. Apply the proposed Mα with llim=500 for N=32, 64, 128, 256, 512, 1024 and record the maximum absolute error at the nodes. If the error decays only like N^{-1} and saturates well above machine precision, the representation step is the limiting factor and the paper's stated scope is too broad; if the error reaches machine precision, the extension concern is resolved.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 2.2's I1 and I2 are delegated to Mathematica, but the Gaussian benchmark (Table 2, Eq. 39) independently checks the resulting even- and odd-mode formulas over a sweep of α, so the CAS step is not the weakest point. The weakest point is the move from Theorem 2.2 to a method for 'a bounded regular function' in §2.3. The operator (13) needs u only on [0,π], but the pseudospectral coefficients (15)-(17) are those of an extension of u to [0,2π]; the accuracy of the truncated series is governed by the smoothness of that extension, not by the smoothness of u(x). For u(x)=arctan x, which is C∞ and bounded with unequal limits at ±∞, the mapped function u(s)=arctan(L cot s) satisfies u_s(π^-)=-1 while the even extension has u_s(π^+)=+1: a jump of 2 in the first s-derivative. Its Fourier coefficients therefore decay only as O(k^{-2}) and the L∞ truncation error is O(N^{-1}), not spectral. The paper tests only finite Fourier modes (u1, u2) and exp(-x²), whose mapped version is flat at both endpoints; no test or error bound covers C² functions with nonvanishing s-derivative at the endpoints. The abstract's blanket claim of accurate approximation for arbitrary bounded regular functions is therefore unsupported.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a pseudospectral method for computing the one-dimensional fractional Laplacian on R without truncating the domain. The authors map x to s via x = L cot(s), represent a function u(x) as a Fourier series in s, and derive in Theorem 2.2 explicit infinite-series formulas for (-Delta)^{alpha/2} e^{iks}. These formulas are used to build a differentiation matrix M_alpha acting on Fourier coefficients. Numerical tests compare the method against exact expressions for e^{i2s} and exp(-x^2), and the method is applied to Fisher-KPP fronts to recover the predicted exponential acceleration e^{t/alpha}. The paper claims that the method approximates the fractional Laplacian accurately and efficiently for bounded regular functions without truncation.","tokens_in":20005,"tokens_out":7537,"duration_ms":76290,"significance":"If Theorem 2.2 is correct, the operational-matrix construction is elegant and potentially useful: the matrix is reusable, the gamma-function factorization in (32)-(33) gives a stable implementation, and the benchmark results are quantitatively strong (errors near 1e-12 for e^{i2s}, 8e-12 for exp(-x^2), and Fisher slope estimates close to 1/alpha). The Gaussian test against the independent formula (39) and the Fisher front comparison with the Cabre-Roquejoffre theory provide genuine external validation. However, the paper's stated scope is broader than what is proved: the central derivation relies on unverified computer-algebra evaluations, no convergence or tail estimates are given for the series and matrix truncations, and the Fourier-extension step is not analyzed for general bounded C^infty functions with different limits at +/-infinity. With these gaps addressed, the paper would be a useful contribution to spectral methods for fractional operators.","major_comments":[{"comment":"The proof of the central formulas delegates the evaluation of I1 and I2 to Mathematica without an independent derivation. Since every subsequent formula and the matrix M_alpha rest on these evaluations, a complete derivation or a reproducible computer-algebra script is needed; the empirical benchmark against (39) is good evidence but does not replace a proof. In addition, the factorization d_kl = (1/pi) I1 I2 in Eq. (24) silently uses the pi-periodicity of |sin(s-eta)|^{alpha-1} and of e^{-i2ls} to shift the inner integration interval, which produces the factor e^{-i2l eta} that turns e^{ik eta} into e^{i(k-2l) eta}; this step should be stated explicitly.","section":"Theorem 2.2, Eqs. (24)-(30)"},{"comment":"The accuracy of representation (15) is governed by the smoothness of the even extension of u(s) to [0,2pi], not by the smoothness of u(x). For u(x) = arctan x with L = 1, u(s) = arctan(cot s) satisfies u_s(pi^-) = -1, while the even extension has u_s(pi^+) = +1, so the first derivative jumps and the Fourier coefficients decay only as O(k^{-2}); the L_infinity truncation error is then O(N^{-1}), not spectral. The tests in Section 3 use only finite Fourier modes and exp(-x^2), whose mapped version is flat at s = 0 and s = pi, so the abstract's claim for arbitrary bounded regular functions is unsupported. Please either prove an error estimate for the extension or explicitly restrict the class of functions considered.","section":"Section 2.3 and Section 3"},{"comment":"The assignment (-Delta)^{alpha/2}(e^{-iNs}) = 0 is introduced because \\hat u(-N) = \\hat u(N), and the truncation of the l-sum at l_lim is justified only by numerical stability. Both choices inject uncontrolled errors into M_alpha, and the abstract's phrase 'without using truncation' is misleading because the method truncates the Fourier series, the l-sum, and applies a Krasny filter. Please provide tail estimates for the l-sum and an error bound for the highest-mode treatment, or state these as heuristic parameters in a clearly delimited claim.","section":"Section 2.5, around Eq. (31)"}],"minor_comments":[{"comment":"In the second case of Eq. (19), the series over l includes l = 0, where sgn(l) is undefined; please specify that the l = 0 term is the separate constant -2/(k^2-4) and that the series runs over l != 0.","section":"Eq. (19)"},{"comment":"The bisection description contains a typo: 'u((x_j + u_{j+1})/2)' should read 'u((x_j + x_{j+1})/2)'.","section":"Section 4, paragraph after Eq. (41)"},{"comment":"The caption says L in {0.1, 0.2, ..., 1}, while the text says L in {0.1, 0.2, ..., 10}; one of these is wrong.","section":"Figure 2 caption"},{"comment":"Reference [28] is incomplete: it lacks volume, year, and page numbers.","section":"References"},{"comment":"The expression (-i sin(s) e^{is})^{1+alpha} requires a branch choice; please state the principal branch used in the numerical comparisons.","section":"Eq. (37)"},{"comment":"The phrase 'spacial shift' should be 'spatial shift'.","section":"Section 2.3"}],"recommendation":"major_revision","confidential_remarks":"The paper presents a promising method and strong numerical evidence, but the central theorem currently depends on unverified computer-algebra evaluations and the Fourier-extension issue for functions with unequal limits at +/-infinity is a real gap relative to the abstract's claim. I would encourage the editor to require the authors to either prove the CAS integrals, provide reproducible code, or restrict the claims; the Fisher front results are interesting but should be quantified more carefully (for example, by reporting sigma_0.5 for several alpha and comparing with 1/alpha with confidence intervals)."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. First, the explicit computation of (-Delta)^{alpha/2}e^{iks} in Theorem 2.2 is a real addition to the spectral toolbox for fractional operators on R; the x = L cot(s) mapping plus aliasing-aware truncation is a coherent way to avoid domain truncation. Second, the paper's claim about arbitrary bounded regular functions is broader than what is actually shown: the pseudospectral coefficients are those of a 2pi-periodic extension of u(L cot s), and the convergence rate is controlled by the smoothness of that extension, not just by the smoothness of u on R.\n\nThe paper does several things well. The operational matrix construction is thoughtful, especially the recursive gamma-function ratios and the l = l1 N + l2 aliasing decomposition. The numerical validation is genuinely strong: the e^{i2s} test reaches 5e-13 and the Gaussian test reaches 1e-12 against an independent exact formula, over a wide sweep of alpha. The Fisher front simulation matching the predicted exp(t/alpha) growth is also credible evidence that the method works in a nontrivial time-dependent setting.\n\nI am less worried than the reader's report about the Mathematica step. The proof of Theorem 2.2 does delegate the key integrals I1 and I2 to a CAS without derivation, and that should be fixed. But the Gaussian benchmark independently checks the resulting even- and odd-mode formulas across many alpha values, so the CAS result is empirically corroborated. That is not nothing.\n\nThe weakest point is the one the stress-test note identifies: the extension. For u(x) = arctan(x), the mapped function u(L cot s) has first-derivative jumps at s = 0 and s = pi after the standard even extension, so the Fourier coefficients decay only as O(k^{-2}) and the method reduces to algebraic, not spectral, accuracy. The paper tests only functions whose mapped versions are either periodic or flat at both endpoints, so the abstract's blanket claim for any bounded regular function is unsupported. There is also no convergence analysis, no code, and the ad hoc setting of the e^{-iNs} mode to zero deserves justification, though that is a minor concern in practice.\n\nThis paper deserves a serious referee. I would send it for review and ask for a revision: prove or at least derive the CAS integrals, test a function with unequal limits at plus and minus infinity, and either weaken the abstract's generality or explain the extension smoothness condition.","headline":"Genuinely new explicit formulas for the fractional Laplacian of Fourier modes, with strong numerical tests, but the spectral accuracy claim is only as good as the smoothness of the periodic extension of the mapped function, and the paper never tests a case where that extension is not flat at the endpoints.","tokens_in":20551,"tokens_out":4276,"would_cite":true,"duration_ms":47677,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65M70","35R11","65T50"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper establishes explicit Fourier series formulas for the fractional Laplacian of trigonometric modes on the whole real line, packaged as an operational matrix.","keywords":["fractional Laplacian","pseudospectral methods","rational Chebyshev functions","Fourier series","nonlocal Fisher equation","operational matrix","whole-line computation"],"falsifier":"Recompute the integrals $I_1$ and $I_2$ in (24) by an independent method, such as residue calculus or high-precision quadrature of the original singular integral (13) for a moderately odd $k$ and a few values of $\\alpha$, and compare with (18); a discrepancy at the level of machine precision in the odd-$k$ branch, where no closed form is available for comparison, would falsify the central formula.","tokens_in":19424,"feed_emoji":"🧮","tokens_out":6089,"duration_ms":53954,"temperature":0.7,"pith_summary":"The paper aims to compute $(-\\Delta)^{\\alpha/2}u$ numerically for smooth bounded functions on the whole real line, without truncating the domain to a finite interval. It maps $x\\in\\mathbb R$ to $s\\in[0,\\pi]$ via $x=L\\cot(s)$, expands the mapped function as a finite Fourier series, and reduces the whole task to the action of the fractional Laplacian on one mode $e^{iks}$. The central result is an explicit infinite-series formula for $(-\\Delta)^{\\alpha/2}e^{iks}$, encoded in an operational matrix. If the formula is correct, a user can approximate the fractional Laplacian for any regular function at spectral accuracy using only three choices: the mode count $N$, a truncation level $l_{\\rm lim}$, and the scale $L$. The authors test the method on explicit examples and use it to simulate Fisher's equation with fractional diffusion in the monostable case, recovering the theoretically predicted exponentially accelerating front speed.","feed_headline":"Fractional Laplacians on R computed without truncating the domain","feed_subtitle":"Mapping R to a circle turns the nonlocal operator into an explicit Fourier series formula, tested on Fisher fronts.","key_machinery":"The machinery is the composition of three steps. First, Lemma 2.1 rewrites the singular integral definition (1) as a Hilbert transform when $\\alpha=1$ and as a second-derivative integral kernel $\\frac{c_\\alpha}{\\alpha(1-\\alpha)}\\int u_{xx}(y)|x-y|^{\\alpha-1}\\,dy$ otherwise, requiring only bounded $C^2$ functions. Second, the algebraic map $x=L\\cot(s)$ turns $\\mathbb R$ into the finite interval $[0,\\pi]$ and Chebyshev or rational-Chebyshev series into Fourier series, so the operator acts on modes $e^{iks}$. Third, Theorem 2.2 supplies the explicit action on each mode, and Section 2.5 packages it as an operational matrix $M_\\alpha$ evaluated on the pseudospectral nodes $s_j=\\pi(2j+1)/(2N)$, with the infinite $l$-sum folded by aliasing into $l_1,l_2$ blocks and stabilized Gamma-function ratios computed recursively.","core_discovery":"The paper's central discovery is Theorem 2.2: for $\\alpha\\in(0,1)\\cup(1,2)$, the fractional Laplacian of the elementary trigonometric function $e^{iks}$ on $\\mathbb R$, after the mapping $x=L\\cot(s)$, equals an infinite series over $l\\in\\mathbb Z$ (equation (18), with separate even-$k$ and odd-$k$ branches), and for $\\alpha=1$ the same object is given by equation (19). The coefficients are ratios of Gamma functions with absolute-value arguments, multiplied by factors involving $\\cot(\\pi\\alpha/2)$ or a sign factor. This formula is the load-bearing component: once it is known, the fractional Laplacian of a Fourier-expanded function is obtained by applying the corresponding matrix $M_\\alpha$ to the Fourier coefficients, with no domain truncation anywhere.","pith_inferences":["If Theorem 2.2 holds, the same mode-by-mode strategy should extend to other kernels that are diagonalized by the cotangent map, such as Riesz derivatives or Hilbert-type operators, and even to $L^2(\\mathbb R)$ functions whose mapped versions have converging Fourier series.","Because the Hilbert transform is diagonalized by the sine-like basis functions mentioned in the paper, the odd-$k$ formula at $\\alpha=1$ could be checked directly against that basis, a computation the paper does not perform.","A testable practical extension would be an automatic rule for choosing $l_{\\rm lim}$ and $L$ as functions of $\\alpha$, $N$, and the decay of the target function; the paper gives heuristics but no selection criterion."],"forward_implications":["For any bounded regular function approximated by its $2N$-point Fourier interpolant (15), the fractional Laplacian at the nodes is approximated by a matrix-vector product; tests on $e^{i2s}$ and $e^{-x^2}$ show errors below $10^{-12}$ for large enough $N$ and $l_{\\rm lim}$.","The method avoids truncation error entirely because the map covers all of $\\mathbb R$; the only controlled parameters are $N$, $l_{\\rm lim}$, and the map scale $L$, with $L$ tunable by spectral interpolation without recomputing the matrix from scratch.","For even $k$ the formula reproduces known closed forms (e.g. (37) for $k=2$ and the Gaussian formula (39)); for odd $k$ it provides the first explicit series representation, with a special treatment at $\\alpha=1$.","The Fisher-KPP simulation confirms the theoretical prediction $c(t)\\sim e^{t/\\alpha}$ for monostable fractional fronts; the measured slope $\\sigma_{0.5}$ approaches $1/\\alpha$ as $\\alpha\\to2^-$, and for $\\alpha=0.5$ increases to $1.9865$ when $N$ and $L$ are enlarged."],"supporting_citations":[{"why":"Supplies the evaluations of the integrals $I_1$ and $I_2$ in (24) that Theorem 2.2 depends on.","marker":"[38]"},{"why":"Introduces the algebraic map and the rational Chebyshev functions that carry the domain transformation from $\\mathbb R$ to $[0,\\pi]$.","marker":"[29]"},{"why":"Previous pseudospectral treatment of a nonlocal operator on the whole real line that provides the mapping, the nodes, and the operational-matrix strategy adapted here.","marker":"[23]"},{"why":"Establishes that the cosine-like and sine-like basis functions are Hilbert-transform eigenfunctions, supporting the even-$k$ case of (19).","marker":"[32]"},{"why":"Supplies the exact Gaussian formula (39) used as a benchmark for the method.","marker":"[4]"},{"why":"Gives the theoretical prediction of exponentially accelerating front speeds in Fisher-KPP equations with fractional diffusion that the numerical experiments confirm.","marker":"[22]"}],"fun_headline_variants":["Fractional Laplacian on R: no truncation, exact Fourier series","Map R to circle: explicit fractional Laplacian, no truncation","Exact formula for fractional Laplacian via cotangent mapping","No truncation: fractional Laplacian on R via Fourier series","Cotangent trick: fractional Laplacian on R becomes a Fourier series"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method rests on the computer-algebra evaluation of the integrals $I_1$ and $I_2$ in (24), which the text reports without an independent derivation, together with an unstated periodicity of $|\\sin(s-\\eta)|^{\\alpha-1}$; if either ingredient is wrong, Theorem 2.2 and the matrix $M_\\alpha$ collapse.","fun_headline_variants_meta":{"raw":{"variants":["Fractional Laplacian on R: no truncation, exact Fourier series","Map R to circle: explicit fractional Laplacian, no truncation","Exact formula for fractional Laplacian via cotangent mapping","No truncation: fractional Laplacian on R via Fourier series","Cotangent trick: fractional Laplacian on R becomes a Fourier series"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000693,"raw_usage":{"total_tokens":3072,"prompt_tokens":815,"completion_tokens":2257,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":431,"completion_tokens_details":{"reasoning_tokens":2161}},"tokens_in":431,"tokens_out":2257,"duration_ms":13569,"temperature":1.0,"reasoning_tokens":2161,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:20:28.527665+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute the integrals $I_1$ and $I_2$ in (24) by an independent method, such as residue calculus or high-precision quadrature of the original singular integral (13) for a moderately odd $k$ and a few values of $\\alpha$, and compare with (18); a discrepancy at the level of machine precision in the odd-$k$ branch, where no closed form is available for comparison, would falsify the central formula.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the evaluations of the integrals $I_1$ and $I_2$ in (24) that Theorem 2.2 depends on."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Introduces the algebraic map and the rational Chebyshev functions that carry the domain transformation from $\\mathbb R$ to $[0,\\pi]$."},{"cited_title":"de la Hoz, C","cited_arxiv_id":null,"evidence_quote":"Previous pseudospectral treatment of a nonlocal operator on the whole real line that provides the mapping, the nodes, and the operational-matrix strategy adapted here."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes that the cosine-like and sine-like basis functions are Hilbert-transform eigenfunctions, supporting the even-$k$ case of (19)."},{"cited_title":"Pozrikidis, The Fractional Laplacian, CRC Press, Boca Raton, FL, 2016","cited_arxiv_id":null,"evidence_quote":"Supplies the exact Gaussian formula (39) used as a benchmark for the method."},{"cited_title":"Cabr´ e, J.-M","cited_arxiv_id":null,"evidence_quote":"Gives the theoretical prediction of exponentially accelerating front speeds in Fisher-KPP equations with fractional diffusion that the numerical experiments confirm."}],"review_version":1}