{"id":"34bbdb9d-8d88-459d-bb63-30730a638eac","arxiv_id":"2606.04357","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":7.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"Reverse diffusion from an OU process yields two new MCMC classes (pseudo-marginal and path-space) for sampling unnormalized densities without discretization or score estimation.","lead":"The paper develops Monte Carlo sampling methods that embed an unnormalized target density as the initial marginal of a finite-horizon diffusion and then use the reverse process to generate samples. The approach avoids discretization error and score estimation, relying only on Monte Carlo variability and change-of-measure techniques relative to an Ornstein-Uhlenbeck process.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Tractability of RN derivative for exact MH acceptance without score terms or discretization hinges on unshown cancellation in the Girsanov-type change of measure.","rationale":"The reader's weakest_assumption correctly isolates the single point where the no-discretization/no-score claim is least secured; without the explicit RN expression, the claim cannot be confirmed or refuted from the abstract alone, so the UNVERDICTED verdict stands.","tokens_in":1810,"tokens_out":369,"duration_ms":36065,"concrete_test":"Extract the explicit formula for the Radon-Nikodym derivative (the 'key theoretical result' referenced in the abstract) and substitute an unnormalized target (e.g., a 1D Gaussian mixture); verify whether the resulting expression depends only on the unnormalized density values, known OU transition densities, and path integrals that can be simulated exactly, with no residual score or normalizing-constant terms.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The strongest claim requires that the Radon-Nikodym derivative between the reverse diffusion transition kernel and the OU kernel be exactly evaluable from the unnormalized target density alone. Standard application of Girsanov's theorem to a diffusion with state-dependent drift (the reverse process) produces an integral involving the difference of drifts; the reverse drift contains the score of the marginal at each time, which is unavailable for an unnormalized target. The paper asserts this derivative remains tractable and free of score estimation, but the abstract provides no explicit expression or cancellation argument showing that score terms drop out or are replaced by quantities computable from the unnormalized density and OU transition densities alone. If such cancellation does not occur, the Metropolis-Hastings ratios in both algorithm classes require either score estimation or further approximation, undermining the central claim.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript proposes embedding an unnormalized target density as the time-0 marginal of a reverse diffusion over finite horizon T. It claims a tractable Radon-Nikodym derivative between the reverse transition kernel and an Ornstein-Uhlenbeck kernel that enables two Monte Carlo classes free of discretization error and score estimation: (i) a parallelizable pseudo-marginal MH scheme producing approximate i.i.d. samples, and (ii) a joint-path MCMC combining OU forward simulation with two Barker-type backward Metropolis updates. Monte Carlo variability is asserted to be the sole approximation source.","tokens_in":2007,"tokens_out":570,"duration_ms":33856,"significance":"If the claimed tractability of the change-of-measure holds without hidden score terms or further approximation, the work would provide a meaningful advance in exact (up to sampling error) methods for multimodal and dependent targets, offering both parallel i.i.d. sampling and an MCMC alternative to random-walk Metropolis that exploits the diffusion structure.","major_comments":[{"comment":"Abstract / central theoretical result: the manuscript asserts that the Radon-Nikodym derivative of the reverse diffusion transition w.r.t. the OU process is exactly evaluable from the unnormalized target alone, yet provides neither the explicit Girsanov integral expression nor the cancellation argument showing that score terms of the marginals drop out. This is load-bearing for both algorithm classes, as the Metropolis-Hastings acceptance probabilities in the pseudo-marginal and path-MCMC schemes cannot be implemented without score estimation or discretization unless the cancellation is demonstrated.","section":"Abstract / Theoretical Result"},{"comment":"§ on algorithm construction (pseudo-marginal MH and joint-path MCMC): the acceptance probabilities are stated to use the RN derivative directly via Barker Bernoulli factories, but without the explicit derivative or proof that it depends only on the unnormalized density and OU transition densities, the 'no score estimation' guarantee does not follow. The finite-horizon assumption that the target is recovered exactly at time 0 must also be verified to hold for the chosen diffusion.","section":"Algorithm descriptions"}],"minor_comments":[{"comment":"The abstract states that the methods 'perform well' on multimodal targets but supplies no quantitative comparison (effective sample size, mixing time, or wall-clock) against standard baselines such as random-walk Metropolis or HMC; this should be added for reproducibility.","section":"Abstract"},{"comment":"Notation for the diffusion coefficients, the choice of T, and the precise form of the OU process should be introduced with equation numbers in the main text rather than left implicit.","section":"Model setup"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for their careful reading and constructive feedback. We address each major comment below.","responses":[{"response":"We agree that the submitted manuscript does not contain the explicit Girsanov integral form of the Radon-Nikodym derivative or a self-contained cancellation argument for the score terms. The central theoretical result is stated to characterize this derivative, but the supporting derivation will be added as a new proposition with full proof in the revised manuscript. This will demonstrate that the derivative depends only on the unnormalized target density and the OU transition densities, enabling the claimed acceptance probabilities without score estimation.","revision_made":"yes","referee_comment":"[Abstract / Theoretical Result] Abstract / central theoretical result: the manuscript asserts that the Radon-Nikodym derivative of the reverse diffusion transition w.r.t. the OU process is exactly evaluable from the unnormalized target alone, yet provides neither the explicit Girsanov integral expression nor the cancellation argument showing that score terms of the marginals drop out. This is load-bearing for both algorithm classes, as the Metropolis-Hastings acceptance probabilities in the pseudo-marginal and path-MCMC schemes cannot be implemented without score estimation or discretization unless the cancellation is demonstrated."},{"response":"We will insert the explicit RN derivative expression into the algorithm sections of the revision to justify the Barker-factory acceptance probabilities. We will also add a dedicated paragraph verifying the finite-horizon construction: for the chosen diffusion and terminal distribution at time T, the marginal at time 0 recovers the target exactly (with Monte Carlo error as the only approximation).","revision_made":"yes","referee_comment":"[Algorithm descriptions] § on algorithm construction (pseudo-marginal MH and joint-path MCMC): the acceptance probabilities are stated to use the RN derivative directly via Barker Bernoulli factories, but without the explicit derivative or proof that it depends only on the unnormalized density and OU transition densities, the 'no score estimation' guarantee does not follow. The finite-horizon assumption that the target is recovered exactly at time 0 must also be verified to hold for the chosen diffusion."}],"tokens_in":1514,"tokens_out":420,"duration_ms":37319,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The new element is the explicit Radon-Nikodym derivative of the reverse diffusion transition kernel with respect to an OU process, which they turn into two algorithm families: a pseudo-marginal MH sequence that aims for approximate i.i.d. draws and a path-space MCMC that mixes forward OU simulation with two kinds of backward Metropolis updates using Barker factories.\n\nWhat works is the framing itself. Embedding the target as the exact marginal at time 0 of a finite-horizon diffusion and then using change-of-measure to remove the usual discretization and score errors is a clean conceptual move. The parallel structure of the first class and the conditional OU updates in the second are practical touches that address common bottlenecks in diffusion MCMC.\n\nThe soft spot is exactly where the stress-test note points: the central claim requires that the RN derivative be evaluable without score terms or further approximation. Standard Girsanov on a state-dependent reverse drift produces an integral over the drift difference; nothing in the abstract shows the cancellation that would leave only quantities computable from the unnormalized density and OU transitions. Without that step written out, the assertion that Monte Carlo variability is the sole error source remains unverified. The abstract also mentions good performance on multimodal targets but supplies no error analysis, implementation details, or numerical evidence.\n\nThe work is for researchers already working on diffusion-based or path-space samplers who are willing to check the missing derivation themselves. It deserves a serious referee because the representation is distinct from the cited reverse-diffusion literature and the algorithmic outline is concrete enough to review, even if the paper will need the full proof and experiments before it can be assessed properly.","headline":"The paper claims a Radon-Nikodym change-of-measure for reverse diffusion that yields exact MH steps without discretization or score estimation, but the abstract gives no derivation showing how the Girsanov integral stays tractable from the unnormalized density alone.","tokens_in":2494,"tokens_out":423,"would_cite":false,"duration_ms":27165,"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":"The Radon-Nikodym derivative of reverse diffusion transitions with respect to an Ornstein-Uhlenbeck process supplies a tractable change of measure that supports two classes of Monte Carlo algorithms free of discretization error and score es","keywords":["reverse diffusion","Monte Carlo sampling","Radon-Nikodym derivative","Ornstein-Uhlenbeck process","pseudo-marginal MCMC","diffusion path sampling","unnormalized densities","Barker acceptance"],"falsifier":"Generate samples from the proposed algorithms for a known target density whose normalizing constant is available, such as a standard bivariate Gaussian mixture, then compare the empirical distribution of the output against the known target; systematic deviation beyond Monte Carlo error would falsify the exact-recovery claim.","tokens_in":2742,"feed_emoji":"🔄","tokens_out":880,"duration_ms":38284,"temperature":0.7,"pith_summary":"The paper embeds an unnormalized target density exactly as the time-zero marginal of a diffusion process run forward over a finite horizon. It derives the Radon-Nikodym derivative between the reverse transition distribution and the corresponding Ornstein-Uhlenbeck transition, turning the reverse process into a change-of-measure problem whose weights remain computable. This representation underpins two sampler families that introduce no time-discretization bias and require no score-function approximation, so that Monte Carlo variability is the sole source of error. One family runs independent pseudo-marginal Metropolis-Hastings trajectories in parallel to produce approximate i.i.d. draws; the other targets the full diffusion path with an Ornstein-Uhlenbeck forward step and two backward Metropolis updates whose acceptance probabilities are realized exactly via Barker Bernoulli factories.","feed_headline":"Radon-Nikodym derivative gives discretization-free diffusion sampling","feed_subtitle":"Recovering the target as the exact initial marginal of a finite-horizon process removes both time-step error and score estimation from two n","key_machinery":"The Radon-Nikodym derivative of the reverse diffusion transition distribution with respect to the Ornstein-Uhlenbeck process, which supplies the change-of-measure weights used in the acceptance probabilities of both sampler classes.","core_discovery":"By characterizing the Radon-Nikodym derivative of the reverse diffusion transition distribution with respect to the Ornstein-Uhlenbeck process, the construction yields a tractable change-of-measure formulation. This formulation directly supports two Monte Carlo classes: a parallelizable sequence of pseudo-marginal Metropolis-Hastings algorithms that generate approximate independent samples, and a path-space MCMC algorithm that alternates Ornstein-Uhlenbeck forward simulation with two types of backward Metropolis steps implemented through Barker-type Bernoulli factories. Both classes recover the target marginal without time-discretization error or score estimation.","pith_inferences":["If the Radon-Nikodym construction extends to other reference processes whose transitions admit closed-form densities, the same change-of-measure strategy could be applied beyond the Ornstein-Uhlenbeck case.","The parallel structure of the first sampler class suggests straightforward distribution across independent compute nodes for very large sample sizes.","Because the horizon T is chosen so that the initial marginal matches the target exactly, any practical truncation of the diffusion at a finite but sufficiently large T would introduce a controllable approximation whose effect can be quantified by increasing T."],"forward_implications":["The first sampler class produces an approximate i.i.d. sample from the target and is fully parallelizable because each trajectory can be generated independently.","The second sampler class targets the joint law of the entire diffusion path on [0,T] by combining one Ornstein-Uhlenbeck forward update with two backward Metropolis updates.","Acceptance probabilities in the path-space algorithm are evaluated exactly using Barker-type Bernoulli factory constructions without additional approximation.","Numerical experiments indicate that both classes handle multimodal targets and complex dependence structures more effectively than random-walk Metropolis."],"fun_headline_variants":["Reverse diffusion Monte Carlo via Radon-Nikodym derivative","Discretization-free sampling from reverse diffusion transitions","Change of measure for exact Monte Carlo in diffusion processes","Parallelizable MCMC using reverse diffusion Radon-Nikodym"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"A finite horizon T exists such that the target density is recovered exactly as the marginal at time zero of the chosen diffusion, and the Radon-Nikodym derivative remains sufficiently tractable to evaluate Metropolis-Hastings acceptance probabilities without further approximation.","fun_headline_variants_meta":{"raw":{"variants":["Reverse diffusion Monte Carlo via Radon-Nikodym derivative","Discretization-free sampling from reverse diffusion transitions","Change of measure for exact Monte Carlo in diffusion processes","Parallelizable MCMC using reverse diffusion Radon-Nikodym"]},"model":"grok-4.3","cost_usd":0.005214,"raw_usage":{"total_tokens":2586,"prompt_tokens":786,"num_sources_used":0,"completion_tokens":62,"cost_in_usd_ticks":52137000,"prompt_tokens_details":{"text_tokens":786,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1738,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":786,"tokens_out":62,"duration_ms":23339,"temperature":1.0,"reasoning_tokens":1738,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-28T03:34:13.021133+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Generate samples from the proposed algorithms for a known target density whose normalizing constant is available, such as a standard bivariate Gaussian mixture, then compare the empirical distribution of the output against the known target; systematic deviation beyond Monte Carlo error would falsify the exact-recovery claim.","supporting_citations":[],"review_version":1}