{"id":"a24a9ada-0313-4cab-a625-e429b37cafc0","arxiv_id":"2412.16902","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For the logarithmic Schrödinger equation with H2 solutions and L∞ potentials, the EWI-FS method converges in L2 at O(τ|lnτ|² + h²|lnh|) under a CFL-type step size restriction.","lead":"This paper proves a near-optimal error bound for a standard numerical scheme, the exponential wave integrator Fourier spectral method, applied to the logarithmic Schrödinger equation. It establishes first-order temporal and second-order spatial convergence under minimal H2 regularity, subject to a time step restriction that the numerics show is necessary.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The advertised 'L∞-potential' applicability of Theorem 1 rests on an H2 well-posedness claim that the cited references do not support for LogSE with V∈L∞.","rationale":"The numerical analysis itself appears coherent in the 1D/2D setting: the local truncation error, the H2-conditional stability estimate, and the induction close under the stated CFL-type restriction in the written proof for d≤2; the 3D extension would require replacing the H^{7/4} Sobolev embedding by an H^s embedding with s>3/2, which is a patchable presentation issue. The main defect is the mismatch between the advertised guarantee of H2 solutions for LogSE with L∞ potential and the cited support, which the paper itself acknowledges covers only V=0 for the LogSE and only power-type nonlinearities for V∈L∞. This directly affects the central claim that the result 'can be directly applied to the LogSE with low regularity L∞-potential,' because without an H2 well-posedness theorem for that class, the assumption (2.15) may be vacuous for some admissible V. The numerical square-well example provides encouraging evidence for at least one discontinuous potential, but it does not establish the general well-posedness assertion. This is the same weakest assumption identified by the reader, and the CONDITIONAL verdict appropriately asks the authors to correct the overstatement or supply the missing reference.","tokens_in":17818,"tokens_out":33524,"duration_ms":285243,"concrete_test":"Inspect the proofs of Hayashi-Ozawa (2024, Theorem 1.2) and Bao et al. (2019b, Theorem 2.2) for LogSE with V=0 and Kato (1987) for NLSE with V∈L∞; determine whether the H2 well-posedness argument extends verbatim when V∈L∞ is added to the LogSE. If yes, cite the extension and the concern is resolved; if no, revise the abstract and Theorem 1 to state the error estimate as conditional on (2.15) without claiming the guarantee for general L∞ potentials.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Section 2.2, the paper states assumption (2.15) as 'known H2 well-posedness' and cites Carles-Gallagher 2018, Bao et al. 2019b, Kato 1987, and Hayashi-Ozawa 2024. The immediately following text admits those results are for LogSE with V≡0 and for NLSE with L∞ potential, not for LogSE with general V∈L∞. The abstract nevertheless claims the H2 assumption is 'theoretically guaranteed' for the present equation. If the LogSE with a merely L∞ potential can lack H2 solutions, or if the existing well-posedness proofs require V∈H¹ (e.g., for integration by parts), then the theorem's advertised application to arbitrary L∞ potentials is not established. The square-well potential test in Section 4.1 shows a particular discontinuous V with apparent H^{2.5} regularity, but one example cannot supply the missing general guarantee.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript analyzes a first-order exponential wave integrator with Fourier spectral spatial discretization (EWI-FS) for the periodic logarithmic Schrödinger equation with an L∞ potential. The main result, Theorem 1, proves an L2 error bound of order O(τ|ln τ|² + h²|ln h|) under an H2 regularity assumption on the exact solution, an L∞ potential, and a CFL-type step-size restriction τ|ln τ| ≲ h²/|ln h|; an H1 error bound of order O(√τ|ln τ| + h|ln h|) is also stated. The proof combines a local truncation error estimate, an H2-conditional L2-stability estimate obtained by the energy method, and a mathematical induction using inverse inequalities. Numerical experiments confirm the convergence orders and the necessity of the step-size restriction, and the method is applied to soliton collisions in 1D and vortex dipole dynamics in 2D.","tokens_in":18013,"tokens_out":16448,"duration_ms":135534,"significance":"If the proof is completed, the result is a significant advance: it would give the first nearly optimal a priori error estimate for the LogSE under H2 regularity, with a general L∞ potential, improving on existing FDTD and time-splitting analyses. The identification of a CFL-type restriction caused by the logarithmic singularity rather than by the potential is an interesting and plausible new phenomenon. The numerical section is substantial and supports the theoretical rates. The proof has two genuinely load-bearing gaps that are, in my view, repairable within the scope of the manuscript: the well-posedness citation does not cover the claimed L∞-potential setting, and the induction step uses an unjustified constant dependence from the stability proposition.","major_comments":[{"comment":"The assumption (2.15) is introduced as 'known H2 well-posedness' of the LogSE (1.1), and the abstract and conclusion state that the H2-solution assumption is 'theoretically guaranteed'. However, the text immediately following (2.15) concedes that the cited results of Carles–Gallagher 2018, Bao et al. 2019b, and Hayashi–Ozawa 2024 are for the LogSE with V ≡ 0, while Kato 1987 is for the NLSE with L∞ potential. No cited theorem covers the LogSE with a general V ∈ L∞. Since the advertised advantage over the existing literature is precisely applicability to L∞ potentials, this gap is load-bearing. The authors should either prove H2 well-posedness for (1.1) with V ∈ L∞, supply a reference that does cover this case, or explicitly reformulate Theorem 1 as conditional on (2.15) and remove the 'theoretically guaranteed' wording from the abstract and conclusion.","section":"Section 2.2, assumption (2.15)"},{"comment":"Proposition 3 states a stability bound of the form e^{Cs τ}‖v0 − w0‖L2 + C(M0)τ²|ln τ|(1 + M1) + C(M2)τ²|ln τ|, where M2 = ‖w0‖H2. In the induction in Theorem 1, Proposition 3 is applied with w0 = ψn, so M2 = ‖ψn‖H2, which the induction controls only by C1|ln h|. In passing to (3.33), the term C(M2)τ²|ln τ| is replaced by C(M0)τ²|ln τ|‖ψn‖H2. This replacement does not follow from the statement of Proposition 3: the notation C(α) denotes a generic constant depending on α, and no inequality such as C(M2) ≤ C(M0)M2 is established. Since ‖ψn‖H2 grows logarithmically as h → 0, an unspecified dependence on M2 could invalidate the induction. The proof needs to make the dependence on M2 explicit, for instance by using the L2 bound on ψn together with inverse inequalities and the CFL condition to control the R2 term in Proposition 3, or by stating Proposition 3 with a polynomial dependence on M2.","section":"Section 3.3–3.4, Proposition 3 and Eq. (3.33)"}],"minor_comments":[{"comment":"The induction proves the L2 bound with a factor τ|ln τ||ln h|, while Theorem 1 is stated with τ|ln τ|². The conversion is not shown and should be stated: the CFL condition implies τ ≤ h²/|ln h| ≤ h² for h < e⁻¹, hence |ln h| ≤ |ln τ|/2, so τ|ln τ||ln h| ≤ (1/2)τ|ln τ|².","section":"Section 3.4, Eqs. (3.34)–(3.35) and Theorem 1"},{"comment":"Remark 2 asserts improved error bounds and a relaxed step-size restriction under ψ ∈ C([0,T];H^m_per), m > 2, but no proof is given. If this statement is intended as a theorem, a proof or at least a detailed proof sketch should be included; otherwise it should be labelled as a conjecture or a claim to be established elsewhere.","section":"Remark 2"},{"comment":"In Eq. (3.25), the chain ∥v(t)∥L∞ ≲ ∥v(t)∥_{H^{7/4}} ≤ ∥v0∥_{H^{7/4}} + C t^{1/8} C(∥v0∥L∞) ≤ C(M0) should state explicitly that the last inequality uses the Sobolev embedding H^{7/4} ↪ L∞ and the assumption ∥v0∥_{H^{7/4}} ≤ M0. The current notation is understandable but slightly compressed.","section":"Section 3.3, proof of Proposition 3"},{"comment":"Minor typographical issues: the keyword 'voterx dipole' should presumably be 'vortex dipole'; 'Cauchy-Schwartz' in Section 3.3 should be 'Cauchy-Schwarz'; and expressions such as 'H2.5-regularity' would benefit from a hyphen or explicit definition.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the scope of IMA JNA and the main theorem is likely correct, but two issues are load-bearing and need repair: the well-posedness citation for L∞ potentials does not support the abstract's guarantee, and the induction step relies on an unproved constant dependence in Proposition 3. Both appear fixable without changing the overall approach. I do not see circular reasoning: the imported technical tools are published results, not the target theorem. I would be willing to review a revised version."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here's the short version: this is a genuinely useful paper for the LogSE numerics literature, and the main theorem looks right. The EWI-FS scheme itself is not new—it's borrowed from the authors' earlier NLSE work—but the analysis that handles the log nonlinearity is. They prove an O(τ|lnτ|² + h²|lnh|) L² error bound under H² solution regularity and L∞ potential, which improves on the H⁴ requirement for FDTD and the half-order rate for time-splitting. They also identify a CFL-type restriction τ|lnτ| ≲ h²/|lnh| that is genuinely needed: the numerics show clear order reduction when it's violated. That's a real finding, not an artifact.\n\nThe proof is the main event. The H²-conditional L²-stability estimate using the Cazenave-Haraux identity (3.4) is the right tool, and the induction with inverse inequalities closes cleanly. The numerical section is thorough and convincing, including the square-well potential case where the solution appears to have H^2.5 regularity. I have no substantive doubt about the theorem itself.\n\nThe soft spots are real but not fatal. The biggest one is the well-posedness claim. The abstract and conclusion state that H²-solution is 'theoretically guaranteed', but the cited references (Carles-Gallagher, Bao et al., Kato, Hayashi-Ozawa) cover LogSE without potential or NLSE with L∞ potential—not LogSE with a general L∞ potential. The authors actually acknowledge this in Section 2.2, then still use the strong wording. That's an overstatement. The theorem itself is conditional on an H² solution existing, so it stands, but the advertised applicability to arbitrary L∞ potentials is not yet established. The square-well test is one example, not a proof. The fix is easy: either prove the well-posedness (maybe a short perturbative argument from the V≡0 case) or change the language to 'assuming H² well-posedness'. A referee should ask for this.\n\nMinor issues: Remark 2 is stated without proof; acceptable for an auxiliary result but a sketch would help. The conversion from τ|lnτ||lnh| to τ|lnτ|² in the final bound is implicit; it follows from the CFL condition since τ ≤ h for small h, but it should be spelled out. No code is provided for the numerics; not required for this journal, but a reproducibility note would be nice.\n\nBottom line: send it to a serious referee. The analysis is solid, the CFL restriction is interesting and well-demonstrated, and the paper advances the LogSE numerics. With the well-posedness wording fixed, I'd be happy to cite it.","headline":"Solid new error analysis for EWI-FS on LogSE with a real CFL restriction, but the abstract oversells the H2 well-posedness guarantee.","tokens_in":18529,"tokens_out":4525,"would_cite":true,"duration_ms":37537,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["65M15","65M70","35Q55"],"pacs":[],"model":"deepseek-v4-flash","headline":"For the logarithmic Schrödinger equation, the exponential wave integrator Fourier spectral method is proved to converge with nearly optimal L² error under only H² regularity of the solution and an L∞ potential.","keywords":["logarithmic Schrödinger equation","exponential wave integrator","Fourier spectral method","error estimate","low regularity potential","H2 solution","CFL condition","vortex dipole dynamics"],"falsifier":"Rerun the one-dimensional convergence test with initial datum $\\psi_0(x)=x|x|^{0.51}e^{-x^2/2}$, $V\\equiv 0$, and $\\tau=h^2/8$ on $(-16,16)$ up to $T=1$; Theorem 1 predicts an $L^2$ error of order $h^2|\\ln h|$, so if the measured rate in $h$ is worse than quadratic up to logarithms while $\\tau|\\ln\\tau| \\le h^2/|\\ln h|$ holds, the central bound is wrong. Conversely, fixing $h$ and taking $\\tau|\\ln\\tau| \\gg h^2/|\\ln h|$ should produce clear order reduction; if no such reduction appears, the claimed necessity of the CFL condition fails.","tokens_in":17618,"feed_emoji":"⚛️","tokens_out":9213,"duration_ms":81221,"temperature":0.7,"pith_summary":"This paper proves that the exponential wave integrator Fourier spectral method solves the logarithmic Schrödinger equation with L² error of order $O(\\tau|\\ln\\tau|^2 + h^2|\\ln h|)$ under the mildest regularity currently available: an $H^2$ solution and a merely bounded $L^\\infty$ potential. This is the first near-optimal convergence result for the LogSE at this regularity level, improving on earlier time-splitting analyses that reached only half-order in time and on finite-difference analyses that demanded $H^4$-regularity. The proof works directly with the singular logarithm, overcoming the non-Lipschitz nonlinearity through an energy-based stability estimate and an induction that keeps the numerical $H^2$ norm under control. A CFL-type condition $\\tau|\\ln\\tau| \\lesssim h^2/|\\ln h|$ is proved necessary for stability and is confirmed by numerical experiments. If correct, the result makes accurate LogSE simulations feasible for low-regularity potentials, including disorder models.","feed_headline":"LogSE solver gets near-optimal error bound under H2 data","feed_subtitle":"Only H2 solutions and bounded potentials needed; a CFL condition ties time step to mesh squared.","key_machinery":"The central object is the exponential wave integrator Fourier spectral method: a first-order exponential integrator in time built from Duhamel's formula with the nonlinearity frozen at each time node, combined with Fourier spectral spatial discretization on a periodic domain. Its analysis rests on two devices: (i) an $H^2$-conditional $L^2$-stability estimate, proved with the energy method and the imaginary-part inequality for the logarithmic nonlinearity, which bounds the nonlinear contribution by a constant times the $L^2$ difference instead of by an unbounded Lipschitz constant; and (ii) a mathematical induction using inverse inequalities on the trigonometric space that controls the numerical $H^2$ norm by $|\\ln h|$ while the CFL-type condition keeps the induction closed. A regularized logarithmic nonlinearity $g_\\varepsilon$ and a difference estimate via $\\varepsilon$ supply the local truncation error bounds.","core_discovery":"The central claim is that the EWI-FS scheme is stable and convergent under the minimal $H^2$ well-posedness assumption $\\psi \\in C([0,T];H^2_{\\mathrm{per}}) \\cap C^1([0,T];L^2)$, with nearly optimal $L^2$ error $O(\\tau|\\ln\\tau|^2 + h^2|\\ln h|)$ and $H^1$ error $O(\\sqrt{\\tau}|\\ln\\tau| + h|\\ln h|)$. The log nonlinearity is not Lipschitz near zero; this is handled by a new $H^2$-conditional $L^2$-stability estimate proved with the energy method and the algebraic inequality $|\\operatorname{Im}[(g(z_1)-g(z_2))(z_1-z_2)]| \\le 2|z_1-z_2|^2$, which prevents the singularity from entering the stability constant exponentially. Mathematical induction with discrete Gronwall inequalities and inverse inequalities controls $\\|\\psi^n\\|_{H^2} \\lesssim |\\ln h|$, and the CFL restriction compensates for logarithmic factors in the truncation error. With higher $H^m$ regularity the bound improves to $O(\\tau|\\ln\\tau| + h^m|\\ln h|)$ under a relaxed step restriction. Numerical experiments validate the convergence orders and show that violating the CFL condition causes order reduction.","pith_inferences":["If the CFL-type restriction is intrinsic to the logarithmic singularity rather than to the low-regularity potential, unregularized first-order schemes such as Lie–Trotter splitting should exhibit a similar $\\tau \\lesssim h^2$ threshold; the stability technique developed here gives a template for proving such restrictions.","The numerical observation that odd data with $\\psi'_0(0)\\neq 0$ produce roughly $H^{3.5-}$ solutions suggests that $H^2$ regularity is close to the sharp threshold for the LogSE, so removing the logarithmic factors in the error estimate may require new well-posedness theory rather than sharper analysis.","Disorder models with genuinely white-noise potentials fall outside the $L^\\infty$ assumption, so extending this convergence theory to rougher potentials would require an $L^2$-based control of the projected nonlinearity that the present proof does not supply.","The necessity of the CFL condition even for single-Gausson dynamics implies that high-accuracy LogSE simulations must balance mesh size and time step globally, not only near zeros of the solution."],"forward_implications":["First-order temporal and second-order spatial $L^2$ convergence is achieved with $H^2$ solutions and $L^\\infty$ potentials, removing the $H^4$ requirement imposed by earlier finite-difference analyses of the LogSE.","The CFL-type restriction $\\tau|\\ln\\tau| \\lesssim h^2/|\\ln h|$ is a genuine feature of the logarithmic singularity: violating it causes measurable order reduction, so practical simulations must couple the time step to the mesh size.","The $H^1$ error bound $O(\\sqrt{\\tau}|\\ln\\tau| + h|\\ln h|)$ provides a quantified derivative error under the same minimal regularity assumption.","For solutions with $H^m$ regularity for $m>2$, the error improves to $O(\\tau|\\ln\\tau| + h^m|\\ln h|)$ under the slightly relaxed condition $\\tau|\\ln\\tau| \\lesssim h^2$, so smoother data recover higher-order spatial accuracy.","The method extends to Neumann and Dirichlet boundary conditions and supports physically motivated simulations such as Gausson collisions under disorder potentials and vortex dipole dynamics in two dimensions."],"supporting_citations":[{"why":"Establishes the algebraic imaginary-part inequality for the logarithmic nonlinearity that powers the H2-conditional L2-stability estimate.","marker":"Cazenave and Haraux, 1980"},{"why":"Proves H2 well-posedness of the LogSE and supplies regularized-log estimates used in the local truncation error analysis.","marker":"Bao et al., 2019b"},{"why":"Provides the recent H2 well-posedness result for the LogSE cited as the theoretical guarantee for the solution regularity assumption.","marker":"Hayashi and Ozawa, 2024"},{"why":"Introduces the EWI-FS method and its optimal error bounds for the NLSE with L∞ potential, the baseline technique this paper adapts.","marker":"Bao and Wang, 2024a"},{"why":"Gives companion optimal EWI estimates for low-regularity potentials and nonlinearities whose CFL-free status contrasts with the LogSE's restriction.","marker":"Bao and Wang, 2024b"},{"why":"Proves the Lipschitz-type estimates for the regularized logarithmic nonlinearity used to bound the truncation errors.","marker":"Zhang and Wang, 2024"},{"why":"Provides the earlier half-order time-splitting error bound for the LogSE under H2 regularity that this paper improves.","marker":"Bao et al., 2019a"},{"why":"Supplies the inverse inequalities on trigonometric spaces used in the induction to convert L2 error bounds into H7/4 and H2 control of the numerical solution.","marker":"Shen et al., 2011"}],"fun_headline_variants":["Near-optimal error bound for log Schrodinger solver","LogSE method achieves near-optimal error with minimal H2 data","Stable EWI-FS scheme for log nonlinearity: optimal error","H2 sufficiency: log Schrodinger solver hits near-optimal bound","CFL restriction yields near-optimal LogSE error estimates"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof assumes the exact solution has $H^2$ spatial regularity with an $L^2$ time derivative, and because the cited well-posedness results do not explicitly cover the LogSE with a general $L^\\infty$ potential, that regularity is the load-bearing premise.","fun_headline_variants_meta":{"raw":{"variants":["Near-optimal error bound for log Schrodinger solver","LogSE method achieves near-optimal error with minimal H2 data","Stable EWI-FS scheme for log nonlinearity: optimal error","H2 sufficiency: log Schrodinger solver hits near-optimal bound","CFL restriction yields near-optimal LogSE error estimates"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000261,"raw_usage":{"total_tokens":1700,"prompt_tokens":1159,"completion_tokens":541,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":775,"completion_tokens_details":{"reasoning_tokens":453}},"tokens_in":775,"tokens_out":541,"duration_ms":4847,"temperature":1.0,"reasoning_tokens":453,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T06:00:16.397503+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Rerun the one-dimensional convergence test with initial datum $\\psi_0(x)=x|x|^{0.51}e^{-x^2/2}$, $V\\equiv 0$, and $\\tau=h^2/8$ on $(-16,16)$ up to $T=1$; Theorem 1 predicts an $L^2$ error of order $h^2|\\ln h|$, so if the measured rate in $h$ is worse than quadratic up to logarithms while $\\tau|\\ln\\tau| \\le h^2/|\\ln h|$ holds, the central bound is wrong. Conversely, fixing $h$ and taking $\\tau|\\ln\\tau| \\gg h^2/|\\ln h|$ should produce clear order reduction; if no such reduction appears, the claimed necessity of the CFL condition fails.","supporting_citations":[],"review_version":1}