{"id":"afc57b0a-e630-45b9-a5d1-47a9e2755c11","arxiv_id":"2607.05981","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Exponential mixing is proven for non-Markovian randomly forced dissipative systems, including the 3D primitive equations, under bounded mixing noises with conditional-distribution regularity.","lead":"This paper proves that a broad class of dissipative dynamical systems — from ordinary differential equations to the 3D primitive equations of atmosphere/ocean dynamics — converge to a unique statistical steady state exponentially fast when driven by bounded, mixing, non-Markov random noise. It extends the Kuksin–Shirikyan framework to non-stationary and continuous-path noises, which is what makes the PDE applications possible.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Section 5's infinite-dimensional proof is self-admittedly sketchy; the relaxation of [KS25]'s density hypothesis leaves the central Theorem 5.2 unverified as written.","rationale":"The reader's chosen weakest assumption was Hypothesis (SF) and its failure for unmodified H^1_loc noises, which is a legitimate applicability concern. However, the paper explicitly constructs modified processes satisfying (SF) and formulates Theorem 6.6/6.10 conditionally on Hypothesis (H), so that concern is partially mitigated. The more load-bearing issue, in my reading, is the self-acknowledged sketchiness of the infinite-dimensional proof in Section 5 and its reliance on non-restated results from [KS25] and [KS26]. This is mentioned in the reader's rationale but was not selected as the weakest assumption. Since the reader's overall verdict of CONDITIONAL already accounts for this sketchiness, my analysis does not move the verdict; it sharpens the reason for conditionality by pointing to the specific, as-yet-unverified transfer of the finite-dimensional coupling argument to infinite dimensions under a relaxed density hypothesis.","tokens_in":43816,"tokens_out":23615,"duration_ms":227030,"concrete_test":"Complete the proof of Lemma 5.6 and the 'rest repeats' paragraph in Section 5, importing only the published statements of [KS25, Prop. 6.1] and [B22, Prop. 3]. Verify explicitly that the squeezing estimate (4.4) holds with constants independent of the approximating subspace F_n and of τ, and that no step in the Doeblin coupling requires density of ∪F_n in E. If a step does require density, supply a substitute argument or the proof of Theorem 5.2 has a real gap.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central infinite-dimensional claim (Theorem 5.2) is supported by a proof sketch that explicitly says 'Our presentation below is sketchy, but missing details may be extracted from the work [KS25]' (Section 5). While the new Lemma 5.4 is proved, the subsequent key step Lemma 5.6 is deferred to '[KS25, Proposition 6.1]', and the paragraph 'The rest of the proof repeats Steps 3,4 in Section 4.1 and the arguments in Section 4.2, 4.3' is asserted without verifying that these carry over to infinite dimensions. This is not merely cosmetic: the paper expressly relaxes a hypothesis of [KS25] by not assuming that the union of the finite-dimensional subspaces F_n is dense in E (Remark before (ALC)). The referenced results in [KS25] and [KS26] are not restated, and it is not shown that their constants remain uniform under this relaxation. Since Theorem 5.2 is the foundation for the primitive-equations application (Theorem 6.10), an undetected gap here would leave the infinite-dimensional exponential-mixing claim unsupported. The finite-dimensional total-variation upgrade in Section 3.3 also relies on [KS26, Cor. 4.2] without proof, but the dual-Lipschitz part is not affected.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies dissipative discrete-time random dynamical systems u_k = S(u_{k-1}, η_k) driven by bounded random noises whose regular conditional distributions with respect to the past satisfy Lipschitz, recurrence and non-degeneracy conditions. It lifts the dynamics to a Markov process on the history space X = X × K and proves exponential mixing in total variation in finite dimensions (Theorem 3.3), exponential mixing in the dual-Lipschitz metric in infinite dimensions (Theorem 5.2), and existence/uniqueness of a two-sided weak solution. Applications are given to ODEs driven by random processes with H^1_loc trajectories and to the randomly forced primitive equations (Theorem 6.10). The paper explicitly extends the framework of Kuksin–Shirikyan [KS25] by allowing non-stationary noises and by not assuming that the union of the finite-dimensional subspaces F_n is dense in E.","tokens_in":44178,"tokens_out":7276,"duration_ms":82414,"significance":"If the results are complete, they constitute a valuable extension of [KS25] to non-stationary bounded mixing noises and to continuous-time noises with H^1_loc sample paths, and they provide an exponential-mixing statement for the randomly forced primitive equations, a system of current interest. The Kantorovich-functional Doeblin coupling argument is natural and the finite-dimensional proof is mostly explicit. However, the manuscript itself states in Section 5 that the infinite-dimensional proof is only a sketch, and several key estimates in both the finite- and infinite-dimensional parts are deferred to [KS25] and [KS26]. Thus the advertised infinite-dimensional theorem is not fully established within the present text.","major_comments":[{"comment":"Theorem 5.2 is a central advertised result and is the basis of the primitive-equations application (Theorem 6.10), but its proof is not contained in the manuscript. The text explicitly says 'Our presentation below is sketchy', the key Lemma 5.6 is deferred to '[KS25, Proposition 6.1]', and the final paragraph asserts that the rest repeats Steps 3,4 of Section 4.1 and Sections 4.2,4.3 without verifying that the constants N, q, p, l, κ remain uniform under the relaxed hypothesis (ALC), where ∪F_n is not assumed dense in E. This is load-bearing: an undetected gap would leave the infinite-dimensional exponential-mixing claim unsupported. Please supply a complete proof or state and prove the precise external result used, checking that its hypotheses cover the present relaxation.","section":"Section 5, Theorem 5.2 and the 'Sketch of the proof'"},{"comment":"The finite-dimensional total-variation conclusion (3.7) depends on two imported ingredients: the assertion that S_*(u,(ξ^⊥_F, Q_{kF})) has a Lipschitz density, and the estimate (3.20). Both are taken from '[KS26, Theorem 4.1 and Corollary 4.2]' without restating the results or their hypotheses. Since (3.7) is the main total-variation mixing statement, the reader cannot verify from the present text that all assumptions of [KS26] hold in this setting. Please restate these results, or give a self-contained proof.","section":"Section 3.3, derivation of (3.7)"},{"comment":"The proof of Lemma 5.5 says 'Since A(y)(F_∞) is dense in H', but the lemma's assumption is only that the closure of A(y)(F_∞) contains the finite-dimensional subspace G. The intended Dini-theorem argument can be made to work on the compact set Y × B_G(1) using the weaker assumption, but as written the proof is internally inconsistent. This is a correctable issue, but it is part of the already sketchy infinite-dimensional proof and should be fixed explicitly.","section":"Lemma 5.5, proof, Step 1"}],"minor_comments":[{"comment":"Typographical and grammatical issues: 'Under a linearised controllability assumptions', repeated '(η1)(η1)(η1)' labels, 'continuos', 'aslo', 'provied', and similar. These should be corrected.","section":"Abstract and Introduction"},{"comment":"Notation is confusing: 'H^1(R1_loc;H)' should presumably be H^1_loc(R;H), and the spaces H^1, H^1_0 used there are not defined consistently with Section 6.3. Please clarify.","section":"Section 6.5"},{"comment":"Hypothesis (H) is essentially the whole content of the applicability of the main theorems to continuous-time noises. The paper gives constructions showing that such processes exist, but it would help to state explicitly that Hypothesis (H) is an assumption and not a consequence of the preceding construction.","section":"Section 6.3, Hypothesis (H)"},{"comment":"The corollary is interesting and, if correct, gives a Dobrushin-style reconstruction of a process from its conditional laws. A brief comparison with the precise statement in [Dob70] would improve readability.","section":"Corollary 3.7"}],"recommendation":"major_revision","confidential_remarks":"The paper is honest about the sketchy nature of Section 5, and the finite-dimensional part is largely explicit. My main concern is that Theorem 5.2 is presented as a new theorem even though its proof is deferred to [KS25] and [KS26] without verifying the relaxed hypotheses. I would recommend asking the authors either to provide a complete proof of Theorem 5.2 or to restate, with verification, the exact results from [KS25] and [KS26] on which it relies. If the companion papers are published and the reductions are straightforward, the paper could be acceptable after these additions."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The finite-dimensional part of this paper is in good shape and worth engaging with seriously. The genuinely new material is there: non-stationary bounded mixing noises, a construction that handles H^1_loc sample paths by replacing the noise with a modified process η̂_k, and a weakening of the density hypothesis in the infinite-dimensional statement. The Markov-lifting argument and the Kantorovich-functional coupling are presented clearly, and the applications to ODEs with continuous bounded noise are plausible. The authors are honest that the finite-dimensional theorem overlaps with [KS25] under slightly stronger assumptions; the added value is in the non-stationary formulation and the modified noise.\n\nThe soft spot is exactly where the stress-test note points: Section 5. The text says outright that the presentation is sketchy and that missing details may be extracted from [KS25]. Lemma 5.6 is deferred to a proposition in [KS25], and the rest of the proof is asserted to repeat earlier steps without checking that uniformity of constants survives the relaxed hypothesis that ∪F_n need not be dense. That matters, because Theorem 5.2 is the foundation for the primitive-equations application in Theorem 6.10. This is not a fatal flaw in the underlying idea — the finite-dimensional proof is largely explicit, and the infinite-dimensional extension probably works with enough work — but as written the central infinite-dimensional claim is unverified. The reliance on results from [KS25] and [KS26] without restatement is also heavy; a referee should check whether those constants and estimates carry over unchanged.\n\nA minor but real caveat: Hypothesis (SF) fails for the natural H^1-continuous noise, and the paper fixes this by moving to the modified process η̂_k. That is legitimate, but it means the theorems apply only when such a modification exists and satisfies the assumptions. For the primitive equations, the verification is plausible but not fully detailed; a referee should push on this point too.\n\nWho gets value from this paper? People working on random dynamical systems and statistical hydrodynamics, especially those interested in non-Markovian bounded noise. The finite-dimensional result and the examples are a genuine contribution. But I would not cite the infinite-dimensional theorem in its current state without checking the deferred details, and I would not trust the primitive-equations claim until Section 5 is written out properly.\n\nRecommendation: send it to peer review, absolutely. A serious referee can separate the solid finite-dimensional core from the sketchy infinite-dimensional part. The paper deserves referee time and probably a major revision, not a desk reject.","headline":"Finite-dimensional mixing results are solid and worth reading; the infinite-dimensional theorem, which carries the primitive-equations application, is not fully supported as written and needs a careful referee.","tokens_in":44629,"tokens_out":1887,"would_cite":true,"duration_ms":23688,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37A25","37H30","35Q30","35Q56","37L40","35R60"],"pacs":[],"model":"deepseek-v4-flash","headline":"Dissipative systems driven by bounded mixing noise converge to a unique statistical state exponentially fast.","keywords":["randomly forced equation","mixing","Markov lifting","Kantorovich functional","non-Markov dynamics","dual-Lipschitz distance","total variation","primitive equations"],"falsifier":"For a concrete system meeting (SF),(RZ),(LCR),(GD) — for example S(u,η)=a u + η on a bounded interval with an i.i.d. noise smoothed by a Lipschitz moving average — numerically compute the total-variation or dual-Lipschitz distance between the laws of two solutions started at widely separated initial states as a function of time. The theorem predicts an exponential decay with a rate γ independent of the initial states; observing algebraic or non-uniform decay would refute the claim.","tokens_in":43706,"feed_emoji":"🌀","tokens_out":7450,"duration_ms":74078,"temperature":0.7,"pith_summary":"This paper proves that a broad class of dissipative random dynamical systems — including ODEs and the 3D primitive equations of atmospheric dynamics — driven by bounded random noises that mix in the past become exponentially mixing: the laws of any two solutions with different initial data converge to each other (and to a unique invariant statistical state) at an exponential rate. The key novelty is that the noise need not be stationary or Markovian; it only needs to have conditional distributions given its past that are Lipschitz in total variation, recurrent to zero, and non-degenerate in a finite-dimensional projection. The proof lifts the system to a Markov process on the infinite-dimensional history space and uses a Doeblin coupling with a Kantorovich functional to show exponential contraction. If correct, this gives a rigorous basis for statistical prediction in a wide range of randomly forced dissipative systems.","feed_headline":"Mixing noise forces exponential memory loss in dissipative systems","feed_subtitle":"New proof covers non-stationary and continuous-path noises, from ODEs to 3D primitive equations.","key_machinery":"The machinery is the Markovian lifting of the non-Markov system (1.1) to the product space X = X×K, where K is the space of past noise histories, via the transition probabilities P_k(U;·) = S_*(U, Q_k(ξ;·)), with Q_k the regular conditional distribution of the noise given its past. Hypothesis (SF) — that Q_k(ξ;·) is Lipschitz in ξ in total variation with respect to the weighted-history metric d(ξ,ξ') = Σ α^{|k|}∥ξ_k − ξ'_k∥ — makes the lifted process sufficiently regular for a Doeblin coupling. The coupling operators R_k, R'_k are built by combining a local 'stabilisation' step (using the Moore–Penrose right inverse of D_ηS to reduce the distance between two trajectories) with a maximal coup","core_discovery":"On the paper's own terms, the central claim is that under Hypotheses (SF), (RZ), (LCR), (GD) in finite dimensions — and (SF), (RZ), (DLP), (ALC), (GD) in infinite dimensions — system (1.1) is exponentially mixing: for initial states v,v' in X, ∥D(u_k(v)) − D(u_k(v'))∥_var ≤ C e^{−γk} in finite dimensions, and the dual-Lipschitz analogue in infinite dimensions. Moreover, there is a unique-in-law two-sided process {û_k} extending the dynamics in the sense D(û_k) = S_*(D(û_{k−1}, η̂_k)) with η̂ distributed as η. If the noise is stationary, {û_k} is the unique stationary process and its one-time marginals give the unique stationary measure. The proof works by lifting (1.1) to a Markov process on","pith_inferences":["The theorem's scope hinges on the (SF) condition, which fails for natural H^1_loc noises and forces the modification η̂_k = (ν_k, η^0_k); a systematic characterisation of noises admitting such a modification would reveal how broad the theory really is.","In infinite dimensions the authors only prove dual-Lipschitz mixing and explicitly doubt total-variation mixing; a natural test is to seek a concrete parabolic SPDE where TV mixing actually fails, to confirm the weaker metric is necessary.","Since the proof uses only the transition probabilities {Q_l} and not the process itself, the results can be read as a property of non-homogeneous Markov operators on the history space; this suggests potential applications to deterministic driving signals with suitable mixing properties.","The paper does not address mixing at continuous times beyond integer times for general systems; extending the rates to all real times with explicit constants would be a useful next step for the ODE application."],"forward_implications":["For any two initial states in the invariant set X, the total-variation distance between their laws decays exponentially in time (finite-dimensional case), so long-run statistics are uniquely determined and computable.","There exists a unique two-sided process û_k satisfying the system in law that attracts all trajectories; when the noise is stationary, this yields a unique stationary measure and exponential mixing of the system.","For randomly forced ODEs driven by bounded mixing noises with continuous paths, the laws of any two solutions converge in dual-Lipschitz distance at an exponential rate for all times t ≥ 0 (Corollary 6.9).","The 3D primitive equations of atmospheric dynamics, driven by bounded continuous mixing noise, are exponentially mixing in the dual-Lipschitz metric (Theorem 6.10).","Any family of transition probabilities on noise histories satisfying (SF) and (RZ) defines a unique random process that is feebly exponentially mixing — a Dobrushin-type reconstruction theorem (Corollary 3.7)."],"fun_headline_variants":["Bounded mixing noises force exponential mixing in dissipative systems","Exponential mixing proven for dissipative systems under bounded random noise","Noise-induced exponential memory loss in dissipative dynamics","From ODEs to primitive equations: exponential mixing under mixing noise","Mixing noise triggers exponential decay of correlations in dissipative systems"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The entire result rests on the assumption (SF) that the noise's conditional distribution given its past moves by no more than a Lipschitz multiple of the past's weighted total-variation distance; if this fails, the coupling argument breaks down, and it visibly fails for many natural continuous-path noises unless the noise is re-encoded as (ν_k, η^0_k).","fun_headline_variants_meta":{"raw":{"variants":["Bounded mixing noises force exponential mixing in dissipative systems","Exponential mixing proven for dissipative systems under bounded random noise","Noise-induced exponential memory loss in dissipative dynamics","From ODEs to primitive equations: exponential mixing under mixing noise","Mixing noise triggers exponential decay of correlations in dissipative systems"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000684,"raw_usage":{"total_tokens":2976,"prompt_tokens":815,"completion_tokens":2161,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":559,"completion_tokens_details":{"reasoning_tokens":2077}},"tokens_in":559,"tokens_out":2161,"duration_ms":15171,"temperature":1.0,"reasoning_tokens":2077,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T08:20:22.072073+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a concrete system meeting (SF),(RZ),(LCR),(GD) — for example S(u,η)=a u + η on a bounded interval with an i.i.d. noise smoothed by a Lipschitz moving average — numerically compute the total-variation or dual-Lipschitz distance between the laws of two solutions started at widely separated initial states as a function of time. The theorem predicts an exponential decay with a rate γ independent of the initial states; observing algebraic or non-uniform decay would refute the claim.","supporting_citations":[],"review_version":2}