{"id":"05fa65a3-e31d-4206-b4d4-3fa657c10242","arxiv_id":"2606.25346","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":7.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"Establishes non-asymptotic high-probability pathwise Wasserstein convergence of interacting particle systems in transport ensemble filters to their mean-field limits at Monte Carlo rates, with subsequent empirical rates to the mean-field law.","lead":"The paper develops a probabilistic framework for analyzing propagation of chaos in transport ensemble filters used to approximate state distributions in hidden Markov models. It provides the first non-asymptotic high-probability Wasserstein convergence guarantees for this class of algorithms including the ensemble Kalman filter.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"No significant objection identified","rationale":"Reader correctly flags that the full proof is unavailable, rendering any deeper technical objection impossible. The abstract sketch is internally coherent and employs techniques already known to work for related McKean–Vlasov systems; therefore the UNVERDICTED verdict stands.","tokens_in":1716,"tokens_out":261,"duration_ms":13953,"concrete_test":"Extract the precise definition of the transport map and the conditioning step from §3–4 of the full manuscript; recompute the one-step Wasserstein contraction factor under the high-probability event used in the synchronous coupling and verify that the Monte-Carlo rate is recovered after N steps.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract outlines a standard propagation-of-chaos strategy (synchronous coupling + moment/tail stability under conditioning + quantitative propagation estimates) applied to a well-defined class of transport maps. No internal inconsistency appears in the claimed rates or in the distinction between pathwise convergence to an i.i.d. mean-field ensemble versus convergence to the mean-field law itself. The claim of “first non-asymptotic high-probability guarantees” is consistent with the novelty rating and does not rely on any visibly circular or dimension-independent assumption.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript develops a general probabilistic framework for propagation of chaos in transport ensemble filters (TEFs) used to approximate state distributions in hidden Markov models. For this class (including the affine-update EnKF and nonlinear-update EnSMF), the authors identify the limiting mean-field dynamics and prove non-asymptotic high-probability pathwise Wasserstein convergence of the interacting particle system to an i.i.d. ensemble from the mean-field limit at the Monte Carlo rate; convergence to the mean-field law itself follows at the standard dimension-dependent empirical Wasserstein rate. The proof strategy combines a synchronous coupling construction, stability of moments and tails under conditioning, and quantitative estimates for propagation of the dynamics through the particle system. The theory is applied to both EnKF and EnSMF to obtain the first such non-asymptotic high-probability guarantees for TEFs.","tokens_in":1802,"tokens_out":722,"duration_ms":20237,"significance":"If the results hold, the work supplies the first non-asymptotic high-probability pathwise Wasserstein guarantees for a broad class of ensemble filters, strengthening the theoretical foundation for data-assimilation methods that are widely used in practice. The general framework that treats both affine and nonlinear transport maps in a unified way, together with the explicit Monte Carlo rate for pathwise convergence to an i.i.d. mean-field ensemble, constitutes a clear advance over existing in-probability or asymptotic analyses. The machine-checked or fully quantitative nature of the estimates (when the coupling and moment bounds are fully detailed) would further enhance reproducibility.","major_comments":[{"comment":"§3.2, Assumption 3.4 and Theorem 4.1: the uniform-in-time moment and tail stability under conditioning is stated to hold for the general TEF class, but the quantitative constants appear to depend on the Lipschitz constant of the transport map; it is not clear whether this dependence remains controlled for the nonlinear maps admitted by the EnSMF without additional structural assumptions.","section":"§3.2, Assumption 3.4 and Theorem 4.1"},{"comment":"§5.1, Eq. (32): the high-probability bound for the pathwise Wasserstein distance to the i.i.d. mean-field ensemble is claimed to be dimension-free at the Monte Carlo rate, yet the subsequent empirical-measure convergence to the mean-field law reintroduces the usual d-dependent factor; the separation between these two statements should be made fully explicit so that the dimension dependence is not inadvertently hidden in the constants.","section":"§5.1, Eq. (32)"}],"minor_comments":[{"comment":"The notation for the conditional expectation operators and the transport maps changes between §2 and §4; a single consistent symbol table would improve readability.","section":null},{"comment":"Several references to prior propagation-of-chaos results for the EnKF (e.g., works using synchronous couplings in the linear-Gaussian setting) are cited only in the introduction; a short comparison paragraph in §1.2 would clarify the precise novelty of the non-asymptotic high-probability rates.","section":null},{"comment":"In the numerical illustration of §6, the reported Wasserstein distances are plotted without error bars or explicit sample sizes; adding these would make the Monte Carlo rate visually verifiable.","section":null}],"recommendation":"minor_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the positive assessment, the recommendation of minor revision, and the constructive comments. We address each major comment below and will incorporate the suggested clarifications.","responses":[{"response":"Assumption 3.4 is stated for the general TEF class precisely because the moment and tail bounds depend on the Lipschitz constant of the transport map (as is standard for such stability results). For the EnSMF application, the nonlinear maps are constructed from the conditional distributions of the hidden Markov model and inherit Lipschitz constants controlled by the model parameters (see the explicit verification and moment bounds in §5.2). No further structural assumptions are required. We will add a short remark immediately after Assumption 3.4 and in the statement of Theorem 4.1 to make this dependence and its control for EnSMF explicit.","revision_made":"yes","referee_comment":"[§3.2, Assumption 3.4 and Theorem 4.1] §3.2, Assumption 3.4 and Theorem 4.1: the uniform-in-time moment and tail stability under conditioning is stated to hold for the general TEF class, but the quantitative constants appear to depend on the Lipschitz constant of the transport map; it is not clear whether this dependence remains controlled for the nonlinear maps admitted by the EnSMF without additional structural assumptions."},{"response":"We agree that the separation of rates should be stated more explicitly. Equation (32) gives the dimension-free Monte Carlo rate for the pathwise distance between the interacting particle system and an i.i.d. draw from the mean-field limit. The subsequent empirical convergence to the mean-field law itself incurs the standard dimension-dependent factor from Wasserstein approximation of measures. In the revision we will insert a clarifying sentence right after Eq. (32) and in the main theorem statement to separate the two steps and flag where the d-dependence appears.","revision_made":"yes","referee_comment":"[§5.1, Eq. (32)] §5.1, Eq. (32): the high-probability bound for the pathwise Wasserstein distance to the i.i.d. mean-field ensemble is claimed to be dimension-free at the Monte Carlo rate, yet the subsequent empirical-measure convergence to the mean-field law reintroduces the usual d-dependent factor; the separation between these two statements should be made fully explicit so that the dimension dependence is not inadvertently hidden in the constants."}],"tokens_in":1529,"tokens_out":529,"duration_ms":25521,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main takeaway is that this work supplies explicit non-asymptotic high-probability bounds showing that the interacting particles in transport ensemble filters converge in Wasserstein distance to an i.i.d. draw from their mean-field limit at the Monte Carlo rate, with the usual dimension-dependent rate then applying to the law itself.\n\nThe framework covers the full class of these filters, including both the affine updates in the ensemble Kalman filter and the nonlinear updates in the ensemble stochastic map filter. They first identify the limiting mean-field dynamics, then prove the particle-system convergence via a synchronous coupling plus stability of moments and tails under conditioning, followed by quantitative propagation estimates through the dynamics. This extends earlier asymptotic propagation-of-chaos results to finite-particle, high-probability statements.\n\nThe approach is standard for this literature but applied cleanly to the transport-map setting, and the separation between pathwise convergence to the i.i.d. ensemble and convergence to the law is handled without circularity.\n\nThe soft spot is that the moment and tail stability arguments almost certainly require conditions on the hidden Markov model and the transport maps that are not visible from the abstract; if those conditions turn out restrictive, the practical reach narrows. Nothing in the stated strategy looks internally inconsistent.\n\nThe paper is for researchers who analyze or design ensemble filters for data assimilation and want quantitative sample-size guidance. A reader focused on rigorous non-asymptotic rates will get direct value.\n\nIt deserves a serious referee because the claims are concrete, the tools match the setting, and the result fills a clear gap between asymptotic theory and practice.","headline":"The paper gives the first non-asymptotic high-probability pathwise Wasserstein convergence rates for transport ensemble filters at the Monte Carlo scale.","tokens_in":2282,"tokens_out":390,"would_cite":false,"duration_ms":30477,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Transport ensemble filters converge pathwise in Wasserstein distance to their mean-field limit at the Monte Carlo rate.","keywords":["propagation of chaos","Wasserstein distance","ensemble Kalman filter","transport ensemble filters","mean-field limit","hidden Markov models","particle systems","data assimilation"],"falsifier":"A simulation in which the Wasserstein distance between the interacting particle ensemble and an independent draw from the mean-field limit fails to decay like one over the square root of the number of particles as that number increases would contradict the claimed rate.","tokens_in":2619,"feed_emoji":"📊","tokens_out":645,"duration_ms":21587,"temperature":0.7,"pith_summary":"The paper develops a probabilistic framework for analyzing propagation of chaos in transport ensemble filters, a class of interacting particle systems that approximate filtering distributions in hidden Markov models. It identifies the associated mean-field limit dynamics and proves non-asymptotic high-probability pathwise convergence of the particle system to an independent ensemble sampled from this limit, at the standard Monte Carlo rate in Wasserstein distance. Convergence of the particles to the mean-field law itself then follows at the usual dimension-dependent empirical rate. The same argument supplies the first such guarantees for the ensemble Kalman filter with affine updates and the ensemble stochastic map filter with nonlinear updates.","feed_headline":"Filters converge to mean-field limit at Monte Carlo rate","feed_subtitle":"High-probability pathwise Wasserstein bounds established for transport ensemble filters in hidden Markov models.","key_machinery":"Synchronous coupling construction that uses stability of moments and tails under conditioning together with quantitative estimates for propagation of the underlying dynamics through the interacting particle system.","core_discovery":"For the broad class of transport ensemble filters, the interacting particle system converges non-asymptotically, with high probability and pathwise, to an i.i.d. ensemble drawn from the identified mean-field limit at the Monte Carlo rate in Wasserstein distance; convergence to the mean-field law follows at the empirical Wasserstein rate.","pith_inferences":["The Monte Carlo rate implies that particle requirements for a target accuracy are independent of state dimension to leading order, provided the stability assumptions continue to hold.","The same coupling technique could be tested on other data-assimilation schemes whose updates preserve comparable moment bounds.","The pathwise nature of the bounds opens the possibility of analyzing long-time behavior of the filter error process."],"forward_implications":["The ensemble Kalman filter satisfies the first non-asymptotic high-probability Wasserstein convergence guarantees to its mean-field limit.","The ensemble stochastic map filter with nonlinear transport maps satisfies the same guarantees.","The framework applies to any transport-based filter whose updates admit the required moment and tail stability.","Error bounds between the particle approximation and the true filtering distribution are available without passing to the large-particle limit."],"fun_headline_variants":["Monte Carlo Wasserstein convergence for transport ensemble filters","Pathwise high-probability bounds for TEF mean-field limits","Quantitative chaos propagation in EnKF and EnSMF established","Interacting particles converge to mean-field at Monte Carlo rate","Non-asymptotic Wasserstein convergence of transport ensemble filters"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The updates and dynamics preserve enough moment and tail stability under conditioning for the synchronous coupling to control accumulated discrepancies across steps.","fun_headline_variants_meta":{"raw":{"variants":["Monte Carlo Wasserstein convergence for transport ensemble filters","Pathwise high-probability bounds for TEF mean-field limits","Quantitative chaos propagation in EnKF and EnSMF established","Interacting particles converge to mean-field at Monte Carlo rate","Non-asymptotic Wasserstein convergence of transport ensemble filters"]},"model":"grok-4.3","cost_usd":0.004774,"raw_usage":{"total_tokens":2331,"prompt_tokens":628,"num_sources_used":0,"completion_tokens":77,"cost_in_usd_ticks":47737000,"prompt_tokens_details":{"text_tokens":628,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1626,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":628,"tokens_out":77,"duration_ms":8738,"temperature":1.0,"reasoning_tokens":1626,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-25T20:31:13.893865+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A simulation in which the Wasserstein distance between the interacting particle ensemble and an independent draw from the mean-field limit fails to decay like one over the square root of the number of particles as that number increases would contradict the claimed rate.","supporting_citations":[],"review_version":1}