{"id":"0a206ee6-79fe-4f66-86ab-c2fcf19ecd62","arxiv_id":"2412.10729","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A regularization inspired by Lefschetz thimbles stabilizes complex Langevin simulations in toy models, with a bias-correction step that restores the original expectation values.","lead":"This physics paper tests a fix for the complex Langevin method, a computational tool for quantum field theories with a 'sign problem' that ordinary Monte Carlo cannot handle. The authors add a carefully chosen regularization to force the dynamics onto a single Lefschetz thimble, then correct the bias to recover the original theory's results.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Bias-correction step Eqs. (12)-(13) is asserted but not demonstrated; it requires a second CL simulation for R alone and a ratio estimate, so accurate expectation values are not established in this manuscript.","rationale":"The reader's weakest assumption is the Salcedo conjecture invoked in Sec. 2.3. I agree that this is a limitation, but it is not the most load-bearing part of the central claim. The regularized models are explicitly checked against the correctness criterion of Eq. (4) in Figs. 1 and 3, and that criterion is a sufficient condition independent of the thimble conjecture; the single-thimble picture is a diagnostic heuristic. The genuinely unsupported step for the headline claim 'yields accurate expectation values after a bias correction' is the bias-correction procedure itself. Eq. (12) requires Q, which Eq. (13) obtains from CL expectation values of an auxiliary DSE observable. This requires a second CL simulation with respect to the regularization R alone, and the manuscript contains no evidence that this second simulation satisfies the correctness criterion. There is also a known statistical issue: Q is a ratio of noisy CL estimates, and the nonlinearity can introduce a bias that is not discussed. The paper explicitly defers all bias-corrected expectation values to companion [5], so the central numerical claim is not verifiable from this preprint alone. This supports the reader's CONDITIONAL verdict rather than changing it. The proposed test would settle the concern by comparing the CL-derived Q from Eq. (13) with the exact Q available in the one-dimensional cosine model, and by checking the R-only drift-magnitude criterion.","tokens_in":6981,"tokens_out":7647,"duration_ms":74050,"concrete_test":"Compute the exact Q = Z_R/Z_rho for the complex cosine model at beta=0.5, r=0.5 by direct numerical quadrature on [-pi,pi], where Z_rho = 2*pi*J_0(beta) and Z_R = integral of [r(x^2-pi^2)-e^{i beta}] dx = -4*pi^3*r/3 - 2*pi*e^{i beta}. Then run complex Langevin for the R-only weight, check that the drift-magnitude density p(u) decays exponentially, and extract Q from Eq. (13) using a Dyson-Schwinger observable O*. If the CL-derived Q disagrees with the exact Q beyond statistics, or if p(u) for R is power-law, the bias-correction step is invalid. Repeat the same comparison for the reduced SU(2) model at beta_2=(1+sqrt(3)i)/2 and r=-5.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Eqs. (12)-(13) of Sec. 2.4 are the only route from the regularized CL results back to the original theory, yet the manuscript contains no bias-corrected expectation values; all numerical checks are deferred to the companion paper [5]. The step is not a trivial identity: Q = Z_R/Z_rho requires two complex Langevin evaluations, <O*>_R and <O*_tilde>, and the paper provides no evidence that the R-only weight satisfies the correctness criterion, i.e., exponential decay of the drift-magnitude density p(u) of Eq. (4). For the cosine model, R(z)=r(z^2-pi^2)-e^{i beta} is not periodic and is rendered periodic by a non-holomorphic identification at x=+/-pi; the drift of R alone is not analyzed. Moreover, Q is a nonlinear ratio of stochastic estimates, so even unbiased CL estimators for the numerator and denominator leave a 1/N bias in Q; the paper does not quantify this. If <O*>_R is biased or the denominator in Eq. (13) is small, the corrected <O>_rho from Eq. (12) will not reproduce the original expectation values, and the headline claim of accurate expectation values after bias correction fails even though the regularized histograms are sharply localized.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This proceedings paper proposes a weight-regularization strategy inspired by Lefschetz thimbles to stabilize complex Langevin (CL) simulations. The authors add a regularization term R to the complex weight, design R so that the regularized theory has a single compact relevant thimble, and use a bias-correction identity (Eqs. (12)-(13)) to recover original observables. They present drift-magnitude densities p(u) from Fokker-Planck solutions for the complex cosine model and the reduced SU(2) Polyakov chain model, showing power-law decay for the original failing cases and exponential decay for the regularized cases, together with CL histograms. Actual corrected expectation values and several implementation details are deferred to the companion paper [5].","tokens_in":7198,"tokens_out":6153,"duration_ms":56656,"significance":"The proposal is potentially valuable: if the single-compact-thimble condition plus bias correction works in practice, it gives a concrete recipe for stabilizing CL in sign-problematic models and connects the thimble picture to CL correctness. The paper's strongest in-manuscript evidence is the independent Fokker-Planck check of the drift-magnitude criterion (Figs. 1 and 3), which goes beyond plotting histograms, and the algebraic derivation of the bias-correction identity in Eq. (12). However, as a standalone paper it does not yet establish the headline claim of accurate expectation values after bias correction, because no bias-corrected observables are reported and the R-only simulation needed in Eq. (13) is not checked against the correctness criterion.","major_comments":[{"comment":"The bias-correction identity is algebraically correct, but the paper never demonstrates the practical step. The only statements are that the numerical robustness is shown in [5] and that 'successful bias correction is discussed in detail in [5]'. Without a table or figure of corrected expectation values against exact or conventional results for at least one model, the conclusion that the approach 'yields accurate expectation values after a bias correction' is unsupported in this manuscript. Please include such results, or explicitly soften the claim to a recipe proposed and validated in the companion paper.","section":"Section 2.4 (Eqs. (12)-(13)) and Section 5"},{"comment":"The evaluation of Q requires a second CL simulation with the weight R alone. For the cosine model, R(z)=r(z^2-pi^2)-e^{i beta} is not periodic and is rendered periodic by a non-holomorphic identification at x=+/-pi; the paper does not analyze the thimble structure or the drift-magnitude density p(u) for the R-only process. If CL with R alone does not satisfy the correctness criterion of Eq. (4), then <O*>_R is biased and Eq. (13) cannot produce a valid Q. Please add a p(u) check, or equivalent evidence, for the R-only process for each model studied.","section":"Section 3 (Eqs. (13) and (15))"},{"comment":"Even if the CL estimators for the numerator and denominator in Eq. (13) are individually unbiased, Q is a ratio of stochastic estimates and therefore carries an O(1/N) bias; if the denominator <O*>_R - <O*>_tilde is small, the relative error is amplified. The manuscript gives no estimate of this bias or of the typical size of the denominator for the tested couplings. Please quantify this effect, for example by reporting Q with statistical errors and by checking stability of the corrected observables across independent subsamples.","section":"Section 2.4 (Eq. (13))"}],"minor_comments":[{"comment":"The abstract says the method solves the SU(N) Polyakov chain model, but the body only reports SU(2) results and refers to SU(3) in the companion paper; please adjust the wording to match the presented content.","section":"Abstract"},{"comment":"The caption says the original model shows power-law decay (green dashed line), while the text says the blue curve is the power-law decay; please make the color references consistent.","section":"Figure 1 caption and text"},{"comment":"The sentence 'The resulting non-holomorphic points at x=+/-pi do not affect the CL algorithm' is asserted without demonstration in this paper; if this is established in [5], please cite the specific result there, and otherwise add a short justification.","section":"Section 3, Eq. (15)"}],"recommendation":"major_revision","confidential_remarks":"This is a proceedings contribution that is explicitly dependent on the companion paper [5] for the central numerical validation. If the editors accept proceedings that summarize longer work, this dependence may be acceptable; otherwise the missing corrected observables are a genuine gap. I recommend major revision so the authors can either add the missing validation or restrict the claims accordingly."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Things you should know: this is a proceedings-length contribution that proposes a genuinely new way to stabilize complex Langevin. The trick is to add a regularization term to the weight that forces the Lefschetz-thimble structure down to a single compact thimble, run CL there, and then remove the bias using a Dyson-Schwinger identity to fix the partition-function ratio. The exposition is clear, and the idea is well motivated.\n\nWhat the paper actually delivers: for the complex cosine model and the SU(2) Polyakov chain, the authors solve the Fokker-Planck equation numerically and show that the drift-magnitude density p(u) decays exponentially for the regularized dynamics, while the unregularized model shows power-law decay. That is the right kind of evidence under the Nagata–Nishimura–Shimasaki criterion. The histograms in Figs. 2 and 4 are consistent with a single compact thimble controlling the process. The paper also honestly states that the underlying Salcedo conjecture is unproved and that the method has so far been tested on toy models.\n\nThe soft spots are real, and they sit where a stress-test lands. The central claim—that bias correction recovers accurate expectation values—is not demonstrated in this manuscript. Eqs. (12)-(13) give the reweighting identity, but Q is the ratio of two partition-function estimates from separate CL runs, and the paper gives no evidence that the R-only weight satisfies the correctness criterion. In the cosine model, the regularization is non-holomorphic at the boundaries, and the periodic continuation is imposed by hand; the authors say it does not affect CL, but that is deferred to the companion. None of this makes the method wrong, but it makes this version an announcement rather than the advertised result. The reader's conditional verdict is fair.\n\nI would send this to peer review: the idea deserves to be on the record, and the diagnostics are solid as far as they go. The referee should require that the bias-corrected numbers be clearly traceable to the companion paper and that the manuscript not overclaim. This is for lattice/CL people; the companion is the citable one. I would not cite the proceedings in my own work, but I would read the companion carefully.","headline":"A fresh thimble-inspired regularization idea for complex Langevin, with the correctness-criterion diagnostics in place but the bias correction—the load-bearing step—left to the companion paper.","tokens_in":7746,"tokens_out":3360,"would_cite":false,"duration_ms":29828,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proposes that complex Langevin converges correctly whenever a weight regularization forces a single compact Lefschetz thimble, and shows how to subtract the induced bias with a Dyson-Schwinger constraint.","keywords":["complex Langevin","sign problem","Lefschetz thimbles","weight regularization","bias correction","Polyakov chain model","cosine model","stochastic quantization"],"falsifier":"Find a model with exactly one relevant, compact thimble (verified by explicit thimble integration) whose complex Langevin drift-magnitude density decays exponentially, yet whose CL expectation values disagree with the exact thimble result; alternatively, run the regularized Polyakov or cosine model at a coupling where the bias correction should work and observe a discrepancy with exact results beyond numerical error.","tokens_in":6760,"feed_emoji":"🎲","tokens_out":6477,"duration_ms":51583,"temperature":0.7,"pith_summary":"Complex Langevin (CL) is a promising method for lattice systems with a sign problem, but it often converges to the wrong answer. This paper argues that the failure is controlled by the Lefschetz-thimble structure of the complexified action: when exactly one relevant, compact thimble is present, CL converges correctly, in line with an existing conjecture. The authors show how to engineer that condition by adding a carefully chosen weight regularization, and they give a systematic way to remove the bias this regularization introduces, using a Dyson-Schwinger constraint to compute the ratio of partition functions. They demonstrate the recipe on the complex cosine model and on the SU(N) Polyakov chain model at couplings where plain CL fails, restoring correct expectation values. A sympathetic reader would care because this turns the thimble picture, normally a diagnostic, into a practical stabilization tool for a wider class of sign-problem-plagued theories.","feed_headline":"Single-thimble weights make complex Langevin converge","feed_subtitle":"Add a weight term that forces one compact thimble; a Dyson-Schwinger constraint then removes the bias.","key_machinery":"The central object is the additive weight regularization $\\tilde{\\rho}(z)=\\rho(z)+R(z;r)$, which changes the effective action to $\\tilde S(z)=S(z)-\\ln[1+R(z;r)e^{S(z)}]$. The regularization is engineered so that, for large $|r|$, the relevant thimble connects to zeros of $\\tilde\\rho$ at the integration boundaries, leaving a single compact thimble (collecting point-symmetric or periodic copies). Two supporting mechanisms carry the argument: the drift-magnitude criterion $p(u;\\theta\\to\\infty)\\sim e^{-\\alpha u}$ for correct CL convergence, and the bias-correction equation (12) with $Q$ fixed by a Dyson-Schwinger constraint (13). The thimble flow equations define the relevant contours $J_\\sigma$ and anti-thimbles $K_\\sigma$ used to identify the structure.","core_discovery":"The central claim is that weight regularizations can be designed so that the regularized theory has a single relevant, compact Lefschetz thimble (up to model symmetries), and that under this condition complex Langevin converges to the correct result. The bias introduced by the regularization is not a dead end: an exact correction formula expresses the original expectation value as the regularized one plus a term proportional to the ratio of partition functions $Q=Z_R/Z_\\rho$, and $Q$ can be obtained from any observable with known zero expectation value via a Dyson-Schwinger equation. Applying this to the complex cosine model and the SU(2) Polyakov chain model—with extensions to SU(3) in the companion paper—the authors report that the drift-magnitude criterion is satisfied and expectation values agree with the exact thimble or known results after bias correction.","pith_inferences":["If the single-thimble conjecture is true in general, the single-compact-thimble condition is not merely sufficient but a systematic design target: any stabilization that enforces it will produce correct CL, and any failure of CL can be attributed to multiple or non-compact relevant thimbles.","The bias-correction formula suggests a general debiasing scheme for any controlled modification of the weight: as long as the modified weight is simulable by CL and an observable with a known zero expectation value exists, the original partition-function ratio can be measured and the bias removed.","A testable extension would be to apply the same regularization design to theories with continuous degrees of freedom in higher dimensions, where the compact-thimble condition is harder to visualize; success there would indicate that the method scales beyond the toy models.","The use of non-holomorphic regularization points at $x=\\pm\\pi$ (restored periodically) hints that mild non-holomorphicity in the drift may be harmless, which could be probed by constructing regularizations with different singularity placements."],"forward_implications":["For any model where a single relevant compact thimble can be enforced by a suitable additive term, complex Langevin can be made to converge correctly, with the original theory recovered by bias correction.","The drift-magnitude density $p(u;\\theta\\to\\infty)$ provides a practical, numerically checkable signal that the regularization has achieved the right thimble structure.","The same recipe extends from SU(2) to SU(3) Polyakov chains in the companion study, suggesting applicability to larger gauge groups.","Because the correction uses only observables with known zero expectation value from Dyson-Schwinger equations, the method is not limited to the specific models tested.","The paper anticipates that kernel transformations, which avoid bias correction entirely, are the natural next step toward gauge theories and real-time/finite-density applications."],"supporting_citations":[{"why":"Supplies the conjecture invoked in Section 2.3 that CL is unbiased when exactly one relevant compact thimble contributes, the load-bearing premise.","marker":"[4]"},{"why":"Provides the drift-magnitude decay criterion used to verify correct convergence of the regularized models.","marker":"[6]"},{"why":"Companion paper containing the full numerical results, bias-correction checks, and the SU(3) extension.","marker":"[5]"},{"why":"Establishes the connection between Lefschetz thimbles and complex Langevin dynamics that motivates the regularization design.","marker":"[3]"},{"why":"Gauge cooling is used to stabilize all CL simulations performed in this work.","marker":"[8]"}],"fun_headline_variants":["Thimble-inspired weights end complex Langevin failures","Single-thimble weights and DS correction stabilize CL","Lefschetz thimble regularization makes CL converge"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the conjecture, cited from the literature, that complex Langevin gives unbiased results whenever the theory has exactly one relevant, compact Lefschetz thimble (counting symmetric copies as one); if this conjecture is false or needs extra conditions, the regularized process could converge to a different theory and the bias correction would not recover the original expectation values.","fun_headline_variants_meta":{"raw":{"variants":["Thimble-inspired weights end complex Langevin failures","Single-thimble weights and DS correction stabilize CL","Lefschetz thimble regularization makes CL converge"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000741,"raw_usage":{"total_tokens":3262,"prompt_tokens":854,"completion_tokens":2408,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":470,"completion_tokens_details":{"reasoning_tokens":2357}},"tokens_in":470,"tokens_out":2408,"duration_ms":17462,"temperature":1.0,"reasoning_tokens":2357,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T15:40:11.298001+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find a model with exactly one relevant, compact thimble (verified by explicit thimble integration) whose complex Langevin drift-magnitude density decays exponentially, yet whose CL expectation values disagree with the exact thimble result; alternatively, run the regularized Polyakov or cosine model at a coupling where the bias correction should work and observe a discrepancy with exact results beyond numerical error.","supporting_citations":[],"review_version":1}