{"id":"23479419-816a-4aea-9131-cd8d13a1c30c","arxiv_id":"1909.02910","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":1,"one_line_summary":"A new path integral Monte Carlo variant, stochastic direct estimators, computes large equilibrium isotope effects with zero thermodynamic integration error and improved statistical convergence.","lead":"Path integral calculations of isotope effects are slow and carry several types of error. This paper combines an existing free-energy perturbation approach with a stochastic change of atomic mass to remove one error source and speed up calculations of large isotope effects, including all deuterated forms of methane in a single simulation.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"SDE's single-simulation estimator assumes λ mixing; Appendix D admits low-T degradation and no bound is given, leaving low-T/large-J bias unexcluded.","rationale":"The paper presents a careful derivation of the stochastic direct estimator and validates it on a harmonic model, methane, and methanium, where five methods agree. The estimator algebra in the appendices is sound, and the empirical agreement among TI, STI, DE, SDE, and ODE is strong evidence that the method is correctly implemented in the tested regimes. The only load-bearing soft spot is the λ-mixing and ergodicity of the extended ensemble used in SDE, which the authors themselves flag in Appendix D and in the reference to Supplementary Section II. Because the central single-simulation claim depends on the trajectory visiting all discrete λj values sufficiently often, and because no mixing guarantee is provided, the general claim for low-temperature and large-J applications is conditional. This matches the reader's weakest-assumption analysis, and the CONDITIONAL verdict remains appropriate: the demonstrated results are credible and the limitation is disclosed, but a mixing test would substantially strengthen the paper.","tokens_in":22368,"tokens_out":8380,"duration_ms":599416,"concrete_test":"In the 8-dimensional harmonic model of Sec. IIIB, run SDE at βℏω0 = 32 and 64 with J = 16, 32, 64, 128, and compare ln(IE) with the exact finite-P analytical value; also measure the integrated autocorrelation time of the λ label and the number of visits to each λj. If the deviation from the exact value exceeds the statistical error for large J or low T, or if the autocorrelation time grows with J faster than the simulation can compensate, the λ-mixing assumption fails and the central claim needs an explicit ergodicity test. Reproducing the slow-λ model of Supplementary Section II would confirm the failure mode.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim—that one SDE run yields unbiased estimates of all factors in Eq. (17)—requires the stochastic λ chain to visit every reference mass λj often enough for each conditional average to converge. This is a mixing assumption, not a variance assumption: if the chain stays near a subset of λ values, the per-factor averages are biased, and the product in Eq. (17) is wrong even with infinite sampling of the visited region. The paper provides no bound on λ mixing and contains an explicit admission of the risk: Appendix D states that the smallest λ step is limited by the discretization J and therefore the procedure becomes less efficient at lower temperatures, because the conditional distribution used in the simple λ-move becomes sharply peaked for large P. The authors also point to a model in Section II of the Supplementary Material where λ exploration is slow. The numerical validation covers the harmonic model and methane/methanium at T ≥ 200 K; the largest isotope effects are reported at 200 K, which is not a stringent test of the low-T regime, and no systematic study of J scaling is shown. Hence the single-simulation advantage is established only in regimes where λ mixing happens to be fast; the general claim for larger isotope effects at lower temperature is conditional on an unproven ergodicity property.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes \"stochastic direct estimators\" (SDE), a path integral Monte Carlo method that combines stepwise direct estimators with the authors' earlier stochastic change of mass, so that a single simulation evaluates all factors in the factored isotope-effect expression (Eq. (17)) and therefore all sequential deuteration isotope effects (e.g., CH4-xDx/CH4 for x=1..4) with zero thermodynamic integration error. The required λ- and mass-scaled moves are described in Appendix D, optimal discretization and reference-mass choices are analyzed in Appendix C, and convergence/divergence properties of the original direct estimators are analyzed in Appendices A and B. The method is validated on an eight-dimensional harmonic model, where it reproduces exact finite-P results, and on full-dimensional methane and methanium, where five independent path integral methods agree. The paper also provides a procedure for estimating the path integral discretization error of isotope effects (Appendix E).","tokens_in":22501,"tokens_out":7005,"duration_ms":71507,"significance":"The paper is careful and mostly convincing: it contains nontrivial analytical derivations, validates the new estimator against exact harmonic results, cross-checks five methods on two molecular systems, reports statistical errors via block averaging, and honestly identifies the central limitation in Appendix D. If the method's single-simulation advantage holds in the intended low-temperature regime, it is practically valuable because it removes thermodynamic integration error, returns all sequential isotope effects from one run, and improves convergence when computational resources are limited. However, the central advantage depends on rapid exploration of the λ dimension, and the evidence provided does not yet establish that dependence in the regime where the largest isotope effects occur.","major_comments":[{"comment":"The unbiasedness of Eq. (17) in a single SDE run requires the stochastic λ chain to visit every reference value λ_j sufficiently often so that each conditional average ⟨Z⟩^{(λ_j)} converges. This is a mixing requirement, not merely a variance requirement. Appendix D explicitly states that the simple λ-move 'becomes less efficient at lower temperatures' because the smallest λ step is limited by J, and Section II of the Supplementary Material reports a model in which λ exploration is slow. No quantitative diagnostic (acceptance rates, transition counts between λ values, or a lower-temperature benchmark) is provided. Since the numerical tests cover only T ≥ 200 K, the paper does not yet establish the method's central advantage in the low-temperature/large-isotope-effect regime. Please add such diagnostics or explicitly restrict the claim to the temperature range in which λ mixing is verified.","section":"Section IIF, Eq. (17), Appendix D"},{"comment":"The abstract and Section IIF attribute a statistical-error reduction to the stochastic λ moves via a sample-reshuffling argument, but the RMSE panels (Figs. 2(c) and 4(c)) show SDE and DE errors that are statistically indistinguishable at most temperatures. The clear, demonstrated advantage of SDE is single-run ergodicity and the ability to output all sequential isotope effects from one simulation, as illustrated in Figs. 3 and 5. Please either demonstrate a regime in which the variance is actually reduced or revise the wording so that the claimed benefit is stated as single-run ergodicity rather than decreased statistical error.","section":"Section IIF and Figs. 2-5"}],"minor_comments":[{"comment":"The table header 'ln(IE) (CD+5/CD+5)' should read 'CD+5/CH+5'; as printed, the ratio is between identical species.","section":"Table II (caption and header)"},{"comment":"The abstract cites Cheng and Ceriotti, J. Chem. Phys. 141, 244112 (2015), while the reference list gives 2014; the two should be reconciled.","section":"Abstract vs. Reference 3"},{"comment":"The phrase 'Z^{0,1}_{sc} is bound' should be 'is bounded', and likewise for 'bound observable'.","section":"Appendix B"},{"comment":"The phrase 'the same number of different Monte Carlo steps' should probably read 'the same number of Monte Carlo steps'.","section":"Section III.A"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a solid incremental contribution that builds heavily on the authors' own Ref. 16. The main risk is overstatement of the SDE advantage in the low-temperature regime; if the authors add mixing diagnostics or temper the claims, I would support publication. I do not see evidence of circularity: the external validation against exact harmonic results and the agreement among five methods are convincing for the tested cases."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is a competent methods paper, and the central new idea holds up: combining stepwise direct estimators with the authors' stochastic change of mass gives a single-simulation route to equilibrium isotope effects, with zero thermodynamic integration error and statistical errors that match or beat separate stepwise simulations. The validation is careful: exact harmonic benchmarks, agreement among five methods on methane and methanium, block-averaged statistical errors, and even a heuristic explanation of why the original direct estimators work better than expected. The analytical convergence results in Appendices A and B and near-optimal discretization rule in Appendix C are real new content. The W2 estimator for discretization error is a nice side product.\n\nThe main soft spot is exactly the one the stress-test note names. The single-simulation advantage assumes the stochastic lambda chain explores all reference masses fast enough. The paper has no proven bound on this mixing, and Appendix D openly says the smallest lambda step is limited by J, making the move less efficient at low temperature because the conditional distribution becomes sharply peaked for large P. The authors also point to a supplementary model where lambda exploration is slow. So the general low-temperature claim is conditional on an unproven ergodicity property. That said, for the tested systems at 200 K it appears to converge, and the authors are refreshingly honest about the limitation.\n\nTwo other reservations. No code, input files, or data are provided, so reproducibility is limited. And the discretization error is estimated (and shown to be a few per mille), but the headline values are not corrected; the authors note both interpretations, which is fair, but the reader must decide which value to quote. I'll note that the reader's soundness score of 7 is about right. The statistical-error-reduction claim is demonstrated numerically rather than derived, and the mixing issue is a genuine open question. But neither undermines the central method for the regime tested.\n\nThis paper deserves peer review, not a desk reject. It is a solid contribution for anyone doing path integral isotope effect calculations, and the W2 estimator will be cited even if the SDE mixing question never gets fully resolved. Send it out.","headline":"Solid methods paper; SDE is a real and useful combination, but the low-temperature lambda-mixing caveat keeps it conditional rather than definitive.","tokens_in":23102,"tokens_out":2843,"would_cite":true,"duration_ms":191438,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"Combining stepwise direct estimators with a stochastic change of mass lets a single path integral Monte Carlo simulation evaluate large equilibrium isotope effects with zero thermodynamic integration error and lower statistical error than…","keywords":["equilibrium isotope effects","path integral Monte Carlo","direct estimators","stochastic change of mass","thermodynamic integration","free energy perturbation","methane deuteration","methanium"],"falsifier":"Run a low-temperature (e.g. 100 K) path integral Monte Carlo simulation for a floppy molecule with a small number of mass intervals, record the sequence of accepted mass values, and measure the integrated autocorrelation time of the mass index; if this autocorrelation time exceeds the autocorrelation time of the configurational coordinates, then increasing the run length will reduce the statistical error of the stochastic direct estimator more slowly than separate stepwise runs of the same total cost, contradicting the claimed single-run advantage.","tokens_in":22085,"feed_emoji":"⚛️","tokens_out":15760,"duration_ms":130622,"temperature":0.7,"pith_summary":"Equilibrium isotope effects—ratios of quantum partition functions of isotopically substituted molecules—are normally computed with path integral simulations that must control three sources of error: discretization error from a finite Trotter number, statistical error from Monte Carlo sampling, and thermodynamic integration error from numerically integrating over a mass-change parameter. This paper argues that the integration error can be made exactly zero by combining two previously separate strategies: direct estimators, which evaluate a partition-function ratio by free-energy perturbation, and a stochastic change of mass, in which a single Monte Carlo trajectory visits many intermediate masses. The resulting 'stochastic direct estimators' factor a large isotope effect into a product of small factors and evaluate every factor in the same simulation, so one run yields not just the end-point ratio but the whole sequence of deuteration isotope effects. In tests on an exactly solvable harmonic model, methane, and methanium, the method matches thermodynamic integration and the stepwise direct estimators, and in short runs it reaches a converged result with fewer total potential-energy evaluations than methods requiring many separate simulations.","feed_headline":"Stochastic mass jumps drop isotope-effect errors in one run","feed_subtitle":"A single Monte Carlo trajectory samples all intermediate masses, yielding every deuteration ratio at once.","key_machinery":"The central object is the mass-scaled direct estimator $Z_{\\mathrm{sc}}^{\\lambda',\\lambda''}$ of Eq. (14), whose coordinates are scaled as $r^{(s)}_{\\lambda',\\lambda'',i}=r^{(C)}_{i}+\\sqrt{m_i(\\lambda')/m_i(\\lambda'')}(r^{(s)}_{i}-r^{(C)}_{i})$, so that a partition-function ratio becomes a Boltzmann-factor ratio of potential energies. The load-bearing identity is the stepwise factorization $Q_P(1)/Q_P(0)=\\prod_{j=1}^{J}\\langle Z_{\\mathrm{sc}}^{\\lambda_j,\\lambda_j}\\rangle^{(\\lambda_j)}/\\langle Z_{\\mathrm{sc}}^{\\lambda_j,\\lambda_{j-1}}\\rangle^{(\\lambda_j)}$, which splits a large isotope effect into $J$ small factors, each evaluated with the sampling weight of a convenient reference mass $\\bar{\\lambda}_j$. The stochastic engine is a Monte Carlo move that changes the mass parameter $\\lambda$ among the discrete values $\\{\\lambda_j\\}$ using acceptance probabilities shaped by a piecewise-linear umbrella biasing potential; it is the same move that earlier removed the integration error from thermodynamic integration, now restricted to discrete jumps. The combination means every factor in the product is gathered from one trajectory, and the umbrella potential is adjusted so that the trajectory visits the intermediate masses roughly in proportion to the factors' statistical weights.","core_discovery":"On the paper's own terms, the central discovery is that the free-energy-perturbation approach to isotope effects can be made practical for large mass changes by evaluating its stepwise factors inside a single path integral Monte Carlo simulation that stochastically changes the molecular mass among a discrete set of intermediate values. The estimator is built on the mass-scaled direct estimator $Z_{\\mathrm{sc}}^{\\lambda',\\lambda''}$, which expresses a small partition-function ratio as an exponential of potential-energy differences at coordinates scaled by the square root of the mass ratio. When these small ratios are multiplied along the mass path, the product has zero integration error by construction, and sharing a single trajectory among all factors reduces the total computational cost and improves convergence in the nonergodic short-time regime. Numerically, the method reproduces exact finite-Trotter harmonic isotope effects, agrees with thermodynamic integration, stochastic thermodynamic integration, stepwise direct estimators, and the original direct estimators on $\\mathrm{CD}_{4}/\\mathrm{CH}_{4}$ and $\\mathrm{CD}_{5}^{+}/\\mathrm{CH}_{5}^{+}$, and extracts all sequential isotope effects $\\mathrm{CH}_{4-x}\\mathrm{D}_{x}/\\mathrm{CH}_{4}$ and $\\mathrm{CH}_{5-x}\\mathrm{D}^{+}_{x}/\\mathrm{CH}^{+}_{5}$ from one run. The paper also finds that the original non-stepwise mass-scaled direct estimator, previously expected to fail for larger isotope effects, remains accurate far beyond that range, and derives sufficient conditions for its convergence and for its divergence.","pith_inferences":["If the discrete $\\lambda$-walk mixes more slowly than the configurational space, the single-run advantage disappears; measuring the $\\lambda$ autocorrelation time as a routine diagnostic would tell practitioners when SDE should be replaced by separate stepwise runs.","The discretization-error estimator $W_2$ of Appendix E could be combined with higher-order factorizations of the Boltzmann operator to estimate and subtract the $O(P^{-n})$ error from one run, effectively approaching the quantum limit without a sequence of Trotter numbers.","Because the mass-scaled coordinate transformation is a nonlocal mapping of the kind used in targeted free-energy perturbation, the same stochastic-$\\lambda$ trick could be transferred to other targeted perturbations, provided the mapping can be evaluated on demand for discrete intermediate states.","The surprising robustness of the original direct estimators hints that in typical deuterium substitution the dangerous divergence condition $m_i(1)>2m_i(0)$ is rarely approached, so simpler estimators may remain adequate for moderate isotope effects; the safe regime is bounded by the ratio of the heavier to lighter mass."],"forward_implications":["A single stochastic run converges to a given statistical error with fewer total potential-energy evaluations than stepwise direct estimators, which must equilibrate $J$ separate simulations; the advantage is largest in the short-run, nonergodic regime.","All sequential isotope effects of the form $\\mathrm{CH}_{4-x}\\mathrm{D}_{x}/\\mathrm{CH}_{4}$ and $\\mathrm{CH}_{5-x}\\mathrm{D}^{+}_{x}/\\mathrm{CH}^{+}_{5}$ become available from the same simulation, not only the end-point $\\mathrm{CD}_{4}/\\mathrm{CH}_{4}$ or $\\mathrm{CD}_{5}^{+}/\\mathrm{CH}_{5}^{+}$ ratio.","Stochastic direct estimators have zero thermodynamic integration error by construction, while deterministic thermodynamic integration retains a discretization error that vanishes only as the number of mass intervals tends to infinity.","In the large-$J$ limit the statistical error of stochastic direct estimators approaches that of stochastic thermodynamic integration, and the deviations are already small at moderate $J$.","The original mass-scaled direct estimators are applicable over a much wider range of isotope effects than previously assumed, with rigorous sufficient conditions for convergence derived in an appendix."],"supporting_citations":[{"why":"Introduces the mass-scaled direct estimator $Z_{\\mathrm{sc}}$; the new stochastic direct estimators build directly on this estimator.","marker":"[3]"},{"why":"The authors' earlier stochastic change of mass with umbrella biasing; supplies the $\\lambda$-move acceptance procedure and the zero-integration-error proof that SDE inherits.","marker":"[16]"},{"why":"Proposes free-energy-perturbation direct estimators for isotope effects; the conceptual starting point of the stepwise factorization in Eq. (17).","marker":"[15]"},{"why":"Shows that large isotope effects can be factored into a product of smaller ones; the stepwise idea the paper generalizes and optimizes.","marker":"[30]"},{"why":"Establishes thermodynamic integration with respect to mass, the baseline method against which all five methods are compared.","marker":"[11]"},{"why":"Provides the inverse-square-root mass interpolation of Eq. (7), used for all $\\lambda$-dependent mass values.","marker":"[1]"},{"why":"Supplies exact analytical isotope effects for harmonic systems at finite Trotter number, used to validate all five methods.","marker":"[43]"},{"why":"Provides the potential energy surface used in the methanium $\\mathrm{CH}_5^+$ simulations.","marker":"[47]"}],"fun_headline_variants":["One Monte Carlo run gets all isotope effects","Mass-jump trick cuts isotope-effect cost","Zero integration error for isotope ratios","All deuteration ratios from a single trajectory"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The single-simulation advantage assumes that the random walk over intermediate masses mixes quickly enough at the temperature of interest, even though the smallest mass jump is fixed by the chosen discretization and the paper notes that this becomes less efficient at lower temperatures.","fun_headline_variants_meta":{"raw":{"variants":["One Monte Carlo run gets all isotope effects","Mass-jump trick cuts isotope-effect cost","Zero integration error for isotope ratios","All deuteration ratios from a single trajectory"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000251,"raw_usage":{"total_tokens":1648,"prompt_tokens":1129,"completion_tokens":519,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":745,"completion_tokens_details":{"reasoning_tokens":466}},"tokens_in":745,"tokens_out":519,"duration_ms":5783,"temperature":1.0,"reasoning_tokens":466,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:03:35.391752+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a low-temperature (e.g. 100 K) path integral Monte Carlo simulation for a floppy molecule with a small number of mass intervals, record the sequence of accepted mass values, and measure the integrated autocorrelation time of the mass index; if this autocorrelation time exceeds the autocorrelation time of the configurational coordinates, then increasing the run length will reduce the statistical error of the stochastic direct estimator more slowly than separate stepwise runs of the same total cost, contradicting the claimed single-run advantage.","supporting_citations":[{"cited_title":"Karandashev \\ and\\ author J","cited_arxiv_id":null,"evidence_quote":"The authors' earlier stochastic change of mass with umbrella biasing; supplies the $\\lambda$-move acceptance procedure and the zero-integration-error proof that SDE inherits."},{"cited_title":"P\\' e rez \\ and\\ author O","cited_arxiv_id":null,"evidence_quote":"Proposes free-energy-perturbation direct estimators for isotope effects; the conceptual starting point of the stepwise factorization in Eq. (17)."},{"cited_title":"Van\\' i c ek , author W","cited_arxiv_id":null,"evidence_quote":"Establishes thermodynamic integration with respect to mass, the baseline method against which all five methods are compared."},{"cited_title":"Ceriotti \\ and\\ author T","cited_arxiv_id":null,"evidence_quote":"Provides the inverse-square-root mass interpolation of Eq. (7), used for all $\\lambda$-dependent mass values."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies exact analytical isotope effects for harmonic systems at finite Trotter number, used to validate all five methods."},{"cited_title":"Jin , author B","cited_arxiv_id":null,"evidence_quote":"Provides the potential energy surface used in the methanium $\\mathrm{CH}_5^+$ simulations."}],"review_version":1}