{"id":"2ee3ceaa-01e6-47ec-93c4-8ce9ad457064","arxiv_id":"2601.01159","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A Weinberg-regularized iterative solver for the coupled-channel Lippmann-Schwinger equation computes single-column S matrices with quadratic scaling and closed channels included.","lead":"A new solver for quantum scattering called WISE computes one column of the scattering matrix at quadratic cost instead of the usual cubic cost, by iterating the Lippmann-Schwinger equation and removing the divergent Weinberg eigenvalues that make naive iterations blow up. It reproduces benchmark He+CO and CO+N2 results and could make state-to-state scattering calculations for large molecular systems tractable.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'guaranteed convergence' claim rests on an unverified spectral assumption: no proof or residual check that the Arnoldi step captures every eigenvalue with |eta|>=1 and that K is diagonalizable on that subspace.","rationale":"The reader's weakest_assumption identified the same issue: the regularization proof assumes a complete eigenbasis and that Arnoldi finds all divergent eigenvalues. I agree this is the most load-bearing concern. The paper's headline contribution is 'guaranteed convergence regardless of strength or anisotropy'; every downstream step (Born series for regularized source, reconstruction, S-matrix extraction) depends on rho(K_R)<1. That inequality is not guaranteed by the construction as written unless the spectral decomposition is exact and complete. The paper provides numerical evidence that the method works for He+CO and CO+N2, which is real support, but it does not provide the missing guarantee. The quadrature issue is secondary: it affects the 'exact quantum results' wording but can be fixed by adaptive grids; the spectral issue, if real, invalidates the method's core promise. Therefore I recommend keeping the CONDITIONAL verdict. The authors should add a residual check, a theorem on the spectral decomposition of the discretized LS kernel, or at least a numerical certification that rho(K_R)<1 for the reported systems.","tokens_in":19275,"tokens_out":9603,"duration_ms":259748,"concrete_test":"Compute the N=343 CO-N2 kernel as in Methods, run the Arnoldi eigensolver twice with different starting vectors and a tight residual tolerance (e.g., 1e-10), and record the set of Ritz values with |eta|>=1. Then form the matrix-free operator K_R = K - K_D and estimate rho(K_R) by power iteration or a second Arnoldi pass on K_R. If the two runs give different n_D or if any eigenvalue of K_R has magnitude >=1 (or is within numerical error of 1), the divergent subspace was not completely captured and the 'guaranteed convergence' claim fails. Also compute max ||K v_k - eta_k v_k|| and left-right biorthogonality residuals for the divergent basis; large residuals indicate the regularization used in Eqs. (16)-(21) is not exact.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Equation (16) defines K_D from the right/left eigenvectors of K with |eta_k|>=1, and the text after Eq. (16) asserts rho(K_R)<1 'by construction'. This is exactly the load-bearing step. For a non-symmetric matrix, a separable sum of eigenvectors removes the selected eigenvalues only if K is diagonalizable on that subspace. If the subspace contains a defective eigenvalue (Jordan block) or if Arnoldi misses an eigenvalue just outside the unit circle, K_R retains spectral radius >=1 and the Born iteration used for u0^(R) and v_k^(R) in Eqs. (17)-(18) is not guaranteed to converge; Eq. (19) then need not represent the solution of Eq. (4). The paper gives no theorem for diagonalizability of the discretized LS kernel, no completeness check for the ARPACK run, and no residual or rho(K_R) verification. The Methods text says the Arnoldi method 'allows us to selectively converge' the divergent eigenvalues, but no shift/invert strategy or error tolerance is specified. This is not a claim that the method fails on the tested systems; the He+CO benchmark is encouraging. It is a claim that the central guarantee is asserted rather than established. The grid-size discrepancy in Fig. 3(b) is a separate accuracy limitation; it does not bear on the iterative convergence guarantee.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript introduces the Weinberg-regularized Iterative Series Expansion (WISE) algorithm for quantum coupled-channel scattering. The method recasts the coupled-channel equations as a multichannel Lippmann-Schwinger equation for a single column of the S-matrix, discretizes the radial coordinate, and solves the resulting linear system by Born iteration. To cure divergence of the Born series, the kernel is decomposed into a divergent part K_D built from Weinberg eigenvalues with |η|≥1 and a regularized part K_R claimed to have spectral radius <1; the divergent part is treated by a small linear system. Closed channels are included through a ratio-propagation construction of the Green's function. The algorithm is validated against standard CC results for He+CO and CO+N2 collisions, and timing benchmarks are reported for channel counts up to 343.","tokens_in":19582,"tokens_out":5684,"duration_ms":57215,"significance":"If the central claims hold, this is a substantial methodological advance: quadratic scaling for a single S-matrix column, explicit inclusion of closed channels, and removal of the Born-series divergence would extend rigorous CC calculations to systems currently considered intractable. The He+CO benchmark is strong and clean: it reproduces independent CC reference results, including Feshbach resonances, without fitted parameters. The CO+N2 application to a molecule-molecule system with strong anisotropy is also a meaningful stress test. The matrix-free implementation and the stable ratio-propagation construction of the Green's function are well-designed engineering. However, the paper's main guarantee is asserted rather than proved, and several numerical details are under-specified; the work is promising but needs revision before the central claims are fully supported.","major_comments":[{"comment":"The statement after Eq. (16) that ρ(K_R)<1 'by construction' is the load-bearing step for the guaranteed-convergence claim. The construction using right and left eigenvectors of K is valid only if the discretized kernel is diagonalizable on the subspace spanned by all eigenvalues with |η|≥1 and if the Arnoldi/ARPACK run returns every such eigenvalue. For a non-symmetric matrix, a defective Jordan block or a missed eigenvalue leaves ρ(K_R)≥1; then the Born series for u0^(R) and v_k^(R) in Eqs. (17)-(18) need not converge, and Eq. (19) need not represent the solution of Eq. (4). Please provide a theorem with explicit assumptions, or add numerical verification: residuals ||K v_k - η_k v_k|| and ||K† w_k - η_k* w_k||, a completeness check for the selected divergent set, and an estimate of ρ(K_R).","section":"Spectral Regularization and Guaranteed Convergence, Eq. (16)"},{"comment":"The abstract and text describe the CO+N2 results as 'exact quantum results,' but Fig. 3(b) shows visible residual differences between WISE and CC. The Methods state that Δr=0.1 'ensures strict convergence (<1%) of benchmark CC calculations' and then attribute the WISE residuals to the uniform radial discretization. This conflates two sources of error: the trapezoid quadrature in Eq. (10) is not identical to the propagator used in the benchmark, and the same grid step can give different quadrature errors in the two formulations. Please quantify the residuals and report a convergence study in Δr, or an adaptive quadrature, before using the word 'exact'.","section":"Application, Fig. 3(b), and Methods (grid)"},{"comment":"The fixed-point iteration is stopped when the L2 norm of the update ||u^(n)-u^(n-1)||_2 falls below 0.1. This is an absolute tolerance, not a relative one; for wavefunction components of order unity it is a very loose stopping criterion. Since the paper claims numerically exact agreement with CC, please justify this tolerance or perform a convergence test with ε decreased by factors of 10 and report the effect on the S-matrix elements.","section":"Methods, Numerical Implementation of the Iterative Algorithm"},{"comment":"The quadratic-scaling claim is supported by timing data only up to N=343 and by visual fits in Fig. 4. The total time includes an Arnoldi eigensolver whose iteration count is not bounded as N grows, and the Outlook estimates n_D ∝ N (up to 10^5), in which case a dense solve of the n_D×n_D M-matrix would be O(n_D^3). To support 'quadratic scaling guaranteed,' please give a complexity argument for the eigensolver and regularization steps, or report scaling of each stage with N over a wider range.","section":"Scaling and Summary/Outlook"}],"minor_comments":[{"comment":"The abstract as supplied mentions a 'contour-integral construction' that does not appear anywhere in the body; the convergence guarantee is instead based on spectral regularization. Please reconcile the abstract with the actual method.","section":"Abstract"},{"comment":"The statement that K_R v_k ≈0 should be an equality if K_D is constructed from the complete divergent eigenspace; the approximate equality signals an unquantified Arnoldi error. Please explain or remove the approximation.","section":"Summary of the Algorithm"},{"comment":"The M-matrix in Eq. (21) is described as sparse, but the implementation uses the dense LAPACK driver ZGESV. Clarify when a sparse solver would be needed and how the sparse structure would be exploited.","section":"Methods, Numerical Implementation"},{"comment":"Please report ARPACK restart parameters, the number of eigenvalues requested, and residual tolerances so that the calculation is reproducible and the claimed convergence of the divergent subspace can be assessed.","section":"Methods, ARPACK"}],"recommendation":"major_revision","confidential_remarks":"The paper fits the journal's scope and presents a genuinely promising method. My main concern is rigor of the central guarantee: the spectral regularization proof is asserted rather than established, and the numerical implementation has loose tolerances and under-specified Arnoldi parameters. I would not reject the paper; these issues are fixable with additional analysis, residual checks, and convergence tests. If the authors can supply those, the contribution would be significant."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"WISE is a genuine step forward for single-column CC calculations: it marries Thomas's iterative Lippmann-Schwinger idea with a stable way to include closed channels and a spectral regularization that appears to fix the divergence problem. The He+CO benchmark, Feshbach resonances included, is convincing, and the scaling plot shows a clean quadratic trend.\n\nWhat is actually new: the ratio-propagation construction for the closed-channel Green's function, which avoids exponentially growing intermediates, and the use of Weinberg eigenvalues to split the kernel into a divergent part (treated exactly) and a regularized remainder (treated by Born iteration). The demonstration that adding one closed channel pushes eigenvalues outside the unit circle and makes the naive Born series diverge is a clean and useful diagnostic. The debt to Thomas and to Weinberg is acknowledged; the new content is the closed-channel treatment and the regularization scheme.\n\nThe soft spots are real but addressable. The big one is the word 'guaranteed.' Equation (16) defines K_D from the eigenvectors with |η|≥1, and the text asserts ρ(K_R)<1 'by construction.' That is only guaranteed if the Arnoldi step found every such eigenvalue and if K is diagonalizable on that subspace. A non-symmetric discretized kernel can have defective eigenvalues, and an implicitly restarted Arnoldi run can miss eigenvalues without a residual check. The paper supplies neither a theorem nor a numerical verification of ρ(K_R)<1. The He+CO results are encouraging, but they do not establish the general claim, and the Introduction's 'regardless of the strength or anisotropy of the interaction potential' overstates what is shown.\n\nSecond, the CO+N2 validation: Fig. 3(b) shows visible residual grid-size differences while the abstract says 'exact quantum results.' The methods text honestly attributes these to the uniform grid at Δr=0.1 a0, but the abstract and the body's 'excellent agreement' should be reconciled, and a converged or adaptive grid should be used for the headline figure. Also, the Born iteration convergence tolerance ε=0.1 is loose; the sensitivity to that choice is worth showing.\n\nThird, the abstract mentions a 'contour-integral construction' that never appears in the paper. Either implement it or drop the phrase. Finally, the scaling evidence stops at N=343, with no error bars, and the N=1e5 extrapolation is speculative; fine for a methods paper if labeled as such.\n\nBottom line: the core idea is credible, the benchmarks are meaningful, and the paper deserves a serious referee. I would send it to review, but I would want the spectral guarantee qualified or verified, the grid convergence cleaned up, and the abstract/text inconsistencies fixed before publication. Recommendation: engage with it. It is a real methods contribution, not a toy.","headline":"A credible O(N^2) single-column coupled-channel method with genuinely useful closed-channel handling and a convincing He+CO benchmark; the 'guaranteed convergence' label currently outruns the spectral verification in the paper.","tokens_in":20115,"tokens_out":2988,"would_cite":true,"duration_ms":28636,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":["34.50.-s","34.50.Ez"],"model":"deepseek-v4-flash","headline":"This paper claims that a regularized Born series solves coupled-channel quantum scattering in O(N²) operations with guaranteed convergence.","keywords":["coupled-channel scattering","Lippmann-Schwinger equation","Born series","Weinberg eigenvalues","spectral regularization","quadratic scaling","closed channels","Feshbach resonances"],"falsifier":"For a system with a closed channel, compute the full spectrum of the regularized kernel K_R after regularization: if any eigenvalue has magnitude at least one, the claimed guaranteed convergence fails. On the numerical side, rerun the CO+N₂ benchmark with a halved radial step, e.g., Δr = 0.05 a₀, and check that the WISE cross-sections move onto the reference coupled-channel curve; if they do not, the discretized kernel, not the regularization, is controlling the error.","tokens_in":19100,"feed_emoji":"⚛️","tokens_out":4587,"duration_ms":48522,"temperature":0.7,"pith_summary":"The paper aims to break the cubic-scaling barrier of standard coupled-channel scattering calculations by solving the Lippmann-Schwinger equation for a single column of the S-matrix instead of propagating a full wavefunction matrix. Its WISE algorithm identifies the Weinberg eigenvalues of the scattering kernel that lie outside the unit circle, subtracts their subspace, and iterates the remaining regularized kernel, whose spectral radius is below one. This makes closed channels and Feshbach resonances tractable and reduces the cost from O(N³) to O(N²) per calculation. The authors test the method on He+CO and CO+N₂ collisions, reporting agreement with benchmark coupled-channel results and a measured quadratic scaling of total CPU time.","feed_headline":"Scattering calculations drop from cubic to quadratic cost","feed_subtitle":"A regularized Born series removes divergent eigenvalues, opening coupled-channel quantum dynamics to far more channels.","key_machinery":"The load-bearing object is the spectrum of the discretized multichannel Lippmann-Schwinger kernel K = G U_off, and in particular its Weinberg eigenvalues η_k, defined as eigenvalues of that kernel. The WISE decomposition K = K_R + K_D splits off eigenvectors with |η_k| ≥ 1 into a separable divergent operator K_D, leaving a regularized remainder K_R whose spectral radius is less than one. The full solution is then the convergent Born series for K_R plus a finite correction involving the divergent eigenvectors, with coefficients fixed by a small linear system. A stable ratio-propagation method constructs the Green's function without explicit evaluation of exponentially growing closed-channel w","core_discovery":"The paper's central claim is that the multichannel Lippmann-Schchinger equation for one column of the S-matrix can be solved in O(N²) operations with guaranteed convergence, regardless of interaction strength or anisotropy. The divergence of the naive Born series is traced to the presence of Weinberg eigenvalues with magnitude at least one. The WISE algorithm splits the kernel into a divergent part built from those eigenvectors and a regularized remainder with spectral radius strictly below one, solves the divergent part exactly through a small linear system, and iterates the convergent series for the rest. Closed channels are incorporated through a stable ratio-propagation construction of t","pith_inferences":["A cheap, auditable robustness check is left implicit: after forming K_R, one could directly verify that its spectral radius is below one by testing K_R v_k ≈ 0 for each removed eigenvector; adding this residual check would turn the guarantee from an assumption into a numerical fact.","Because the benchmarks show the eigenvalue search dominating total cost, future gains for very large channel counts likely depend on improving the eigensolver step more than on the Born iterations themselves.","The residual differences visible in the CO+N₂ results could plausibly be removed by adaptive radial grids at no extra asymptotic cost, which the paper notes but does not implement.","The Weinberg spectrum could be repurposed as a principled channel-selection tool: channels that push eigenvalues outside the unit circle are precisely the ones responsible for strong coupling, so they indicate which closed channels matter most."],"forward_implications":["State-to-state scattering for molecules prepared in a specific incoming quantum state can be computed in O(N²) time, a quadratic reduction from the O(N³) cost of standard propagation methods.","Closed channels can be included robustly, so Feshbach resonances and threshold effects are captured rather than lost in open-channel-only approximations.","Highly anisotropic molecule-molecule systems such as CO+N₂ become feasible for systematic energy-resolved calculations, including the scans needed for atmospheric line-shape modeling.","The regularization supplies a formal convergence guarantee for the iterative series, removing the need for ad-hoc acceleration schemes in a broad class of coupled-channel problems.","Weinberg eigenvalue spectra provide a quantitative diagnostic of how strongly perturbative an interaction potential is, which can guide basis-set and method choices."],"fun_headline_variants":["Quadratic scaling with guaranteed convergence for quantum scattering","WISE method: quadratic cost for coupled-channel quantum dynamics","Regularized Born series fixes divergence, cuts scattering cost","Quantum coupled-channel calculations now run at quadratic cost","Closed channels handled, cost drops to quadratic in scattering"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The guarantee stands only if the eigensolver finds every Weinberg eigenvalue with magnitude at least one and the kernel is diagonalizable on that divergent subspace; if an eigenvalue is missed or defective, the regularized kernel may still have spectral radius at least one and the series can diverge, and the numerical agreement additionally assumes the uniform radial grid is fine enough to represent the Green's operator faithfully.","fun_headline_variants_meta":{"raw":{"variants":["Quadratic scaling with guaranteed convergence for quantum scattering","WISE method: quadratic cost for coupled-channel quantum dynamics","Regularized Born series fixes divergence, cuts scattering cost","Quantum coupled-channel calculations now run at quadratic cost","Closed channels handled, cost drops to quadratic in scattering"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000206,"raw_usage":{"total_tokens":1221,"prompt_tokens":720,"completion_tokens":501,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":464,"completion_tokens_details":{"reasoning_tokens":426}},"tokens_in":464,"tokens_out":501,"duration_ms":5628,"temperature":1.0,"reasoning_tokens":426,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T12:53:40.417459+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a system with a closed channel, compute the full spectrum of the regularized kernel K_R after regularization: if any eigenvalue has magnitude at least one, the claimed guaranteed convergence fails. On the numerical side, rerun the CO+N₂ benchmark with a halved radial step, e.g., Δr = 0.05 a₀, and check that the WISE cross-sections move onto the reference coupled-channel curve; if they do not, the discretized kernel, not the regularization, is controlling the error.","supporting_citations":[],"review_version":1}