{"id":"c260029b-1843-4103-9b0a-d2ccde5fd6c3","arxiv_id":"2607.16328","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"Spontaneous stochasticity is formalized as a measure-selection principle, and the paper proves that any probability measure on the nonunique inviscid solution set is realizable as the limiting law of some regularization.","lead":"A new theory paper gives spontaneous stochasticity — random-looking behavior in deterministic equations when smoothing is removed — a general mathematical definition: it is the selection of a probability law rather than a single state. Its main theorem shows that once an equation loses uniqueness, almost any random law can be manufactured by choosing a suitable smoothing, so the real science is in which smoothings and sampling rules are physical.","discovery_kind":"paradigm_shift","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Step 5's o(1) approximation is asserted without proof and is not a routine continuity argument; if it fails, co(E)⊂M and Corollary 3 collapse.","rationale":"The reader's weakest_assumption identifies Step 5's o(1) approximation as the load-bearing gap, and I agree. The central claim—that every probability measure on S0 is attainable—requires co(E) ⊂ M, which is exactly what Step 5 establishes. The asserted approximation is not a minor technicality: it is a statement about the stability of selected endpoints under the fast switching of the regularization field near a non-Lipschitz singularity, where small sup-norm perturbations can change the inviscid solution by O(1). The paper provides no proof, and the surrounding examples do not involve such switching. While Step 4's Lipschitz-membership issue is also a genuine proof gap, it is more plausibly repaired by standard smoothing; the o(1) step is a substantive dynamical claim that could actually be false. Thus the concern is load-bearing and not addressed. The verdict CONDITIONAL remains appropriate: the framework is coherent and the conclusion may hold, but the proof as written is incomplete. No change to the reader's verdict is needed.","tokens_in":48335,"tokens_out":9663,"duration_ms":97779,"concrete_test":"Take the paper's own one-dimensional system ˙x = √|x| with x0 < 0 and t > t⋆. Let fx and fy be the regularizations (AmbTϵ) from Section 6.1 with two distinct waiting times T_x ≠ T_y, so the selected endpoints are x ≠ y in S0. Construct fθ exactly as in Eq. (24). Numerically compute γ_θ(τ) = φ^τ_t(x0) for τ ranging over, say, 10^3 to 10^6 and compare with γ̂_θ(τ) = x â_θ(τ) + y(1−â_θ(τ)). If ∥γ_θ(τ) − γ̂_θ(τ)∥ does not tend to 0, the o(1) assertion fails. Since the regularized flow is explicit for (AmbTϵ), this check can also be done analytically by solving the mixed-field ODE on each interval where a_θ is nearly constant and tracking the accumulated endpoint error.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 1's proof hinges on Step 5, which claims that for the mixed field f_θ = a_θ f_x + (1−a_θ) f_y, the endpoint γ_θ(s) satisfies γ_θ(s) − γ̂_θ(s) = o(1), where γ̂_θ(s) is the convex combination of the x- and y-selecting endpoints weighted by the sign-limit â_θ(s). This is asserted without proof. It is not a routine continuity consequence: near non-Lipschitz singularities, flows are not continuous in sup-norm perturbations of the field (e.g., √x + η abandons the 'stay at 0' solution for arbitrarily small η). The coefficient a_θ(s) oscillates between near 1 and near 0 on a fast timescale, so the physical-time-t trajectory experiences a rapidly switching vector field. Whether the endpoint lands near x or y depends on the precise history of switches, especially if the trajectory passes through the singularity. The equality is a genuine geometric statement about how the two tube fields interact, and no argument is provided. If it fails for some nonunique inviscid system, then co(E) ⊂ M is not established, so M may be strictly smaller than M0, and Corollary 3—the advertised physical conclusion—collapses. The explicit examples in Section 6 do not exercise this switching construction, so they provide no corroboration.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a measure-theoretic definition of spontaneous stochasticity (SpSt) as a selection principle: for a regularized ill-posed inviscid problem and an ambient sampling measure, one studies the pushforward of the ambient measure under the regularized flow, with strong SpSt meaning convergence to a unique non-Dirac law. The central claim is an attainability theorem (Theorem 1, §4.1): if the inviscid problem is nonunique, then every probability measure supported on the compact inviscid endpoint set is attainable as the unique limiting law of some admissible regularization, so that M = M₀ = P(S₀), and Corollary 3 concludes that every non-Dirac law can be realized by a strong-SpSt regularization. The paper also derives a Dini-type necessary condition for nonuniqueness (§5), analyzes the relation to TBM and sensitivity-to-initial-data criteria (§3), and develops a semigroup/RG viewpoint (§§7–10) in which limiting statistics appear as statistical attractors. Explicit one-dimensional examples, including the √|x| family with an arcsine-type waiting-time law and an exit-time analysis for x^{1/3}, are worked out in detail, together with numerics in appendices.","tokens_in":48626,"tokens_out":18608,"duration_ms":188685,"significance":"If Theorem 1 and its proof are correct, the paper delivers a striking structural result: once nonuniqueness is present, essentially arbitrary inviscid statistics are realizable by suitable regularizations, shifting the physical content of SpSt onto the choice of regularization class and ambient measure. The paper is also valuable for its explicit, parameter-free computations in Section 6 and Appendix B, its careful comparison of SpSt, SpSt_turb, and TBM, and its semigroup/RG reformulation of limiting statistics as statistical attractors. The explicit examples and the detailed exit-time analysis in Section 9 are concrete and reproducible. However, the central attainability theorem has two load-bearing gaps in its proof — the admissibility of the tube-field construction and the unproved switching endpoint estimate — so the significance is conditional on those being repaired.","major_comments":[{"comment":"The field f(·,τ) = θ_τ F_τ + (1−θ_τ) f₀ is claimed to lie in V₀ as defined in Eq. (1). But θ_τ vanishes outside a tubular neighborhood of the selected inviscid trajectory, so f(·,τ) equals f₀ outside that neighborhood. For a non-Lipschitz f₀ — precisely the case in which nonuniqueness occurs — f(·,τ) is not globally Lipschitz and hence is not an admissible regularization in V₀. Even if the solution from x₀ is unique because it stays in the tube, the proof does not establish γ_x ∈ Γ, so E ⊂ M is not proven. The construction must be amended to use a global Lipschitz approximation of f₀ compatible with the tube, or the admissibility class must be revised with a proof that the regularized flow is single-valued for the relevant initial data.","section":"Section 4.1, Step 4 (tube-field construction)"},{"comment":"The pivotal assertion γ_θ(s) − γ̂_θ(s) = o(1) is stated without proof. It is not a routine continuity consequence: uniform convergence of f_θ(·,s) to f₀ does not imply convergence of finite-time flow maps when the limit field is non-Lipschitz (e.g., √x + η abandons the 'stay at zero' solution for arbitrarily small η). The coefficient a_θ(s) switches between the two tube fields on a fast scale, and the physical-time endpoint can depend on the entire switching history. Since this step is the only argument for co(E) ⊂ M, Theorem 1 and Corollary 3 collapse if it fails. The explicit examples in Section 6 do not exercise the switching construction, so they provide no corroboration. A proof, or a modified construction with quantitative estimates on the switching error, is required.","section":"Section 4.1, Step 5, Eq. (24)"}],"minor_comments":[{"comment":"The statement 'M₀ \\ E = M' contradicts the already-proved identity M = M₀; it should presumably read 'M₀ \\ E = M \\ E' or similar. Please correct this typographical-formulaic inconsistency.","section":"Corollary 3, Section 4.1"},{"comment":"Property 1 is asserted 'without proof'. As written, the implication TBM ⇒ liminf Var(ν_ε) ≥ L does not follow from Eq. (16), because the liminf of Var(κ_ε) at the endpoint does not control the Cesàro average of s ↦ Var(κ_s). Either add a hypothesis (e.g., continuity or a uniform lower bound on s ↦ Var(κ_s)) or downgrade the statement to a conjecture.","section":"Section 3.2.2, Property 1"},{"comment":"The notation T_ε is used both for the regularization parameter value and for the function s ↦ T_s entering the curve γ(s). This creates the appearance that (T_ε)#Leb^ε is a Dirac mass. Please make explicit that T_ε denotes the function T(s) and write T_s or T(·) in the pushforward statement.","section":"Section 6.2, Strong SpSt"},{"comment":"The Osgood argument should state explicitly that the modulus Ω is nondecreasing (or assume the appropriate Osgood condition on a nondecreasing function); otherwise the standard one-sided Osgood lemma does not directly apply to an arbitrary Ω with divergent integral.","section":"Appendix A.7, proof of Theorem 2"},{"comment":"Some displayed identities in Section 4.1 could benefit from numbering (e.g., the tube-field definition in Step 4 and the switching field in Step 5), since the current text refers to 'the previous construction' without precise equation numbers.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The paper is ambitious and, if the proof gaps are filled, would be a significant contribution. The main risk is the unproved Step 5 switching estimate in Theorem 1; I would recommend that the editor seek a referee with expertise in ODE continuation and singular perturbations to verify whether the construction can be repaired. The companion PDE claim [17] is cited but not proved here; that is fine, but the present paper's claims should not rely on it."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The main new thing here is a general measure-selection definition of spontaneous stochasticity, plus a maximality theorem: once the inviscid problem is nonunique, every probability measure on the endpoint set is realizable as the selected law of some regularization. The one-dimensional example is worked out to explicit densities, the RG/semigroup part gives a genuinely different angle, and the paper is honest about its own limitations. No data fitting, no hidden circularity. This is a serious contribution to a problem that has lacked a general definition.\n\nThe soft spots are in the proof of Theorem 1, and they are not cosmetic. The convex-combination step (Step 5) asserts gamma_theta(s) - hat_gamma_theta(s) = o(1) without proof. Near a non-Lipschitz singularity this is not a continuity statement; the trajectory experiences a rapidly switching vector field, and the endpoint could depend on the precise switching history. The stress-test note is right: this needs a real argument, and the explicit examples in Section 6 do not exercise the switching construction, so they do not corroborate it. Step 4 also has a V0/Lipschitz issue: the tube-constructed fields equal f0 outside the tube, and unless f0 is Lipschitz or the tube covers everything, membership in V0 is not automatic. Theorem 2's proof has a telescoping step and a one-sided Osgood lemma that are both asserted rather than shown.\n\nThese are plausibly repairable, but they are real gaps in the paper's central claim. If the switching step fails, M could be strictly smaller than M0, and Corollary 3 collapses. So the verdict is conditional, not a rejection. The framework itself is coherent, and the definitions are worth taking seriously even if the maximality theorem needs patching.\n\nFor a reading group: yes, this is exactly the kind of paper worth spending a session on. For citation: I would cite the definition and the example, but not Theorem 1 until the proof is fixed. A serious editor should send this to peer review; the referees should ask for a repaired Step 5 and a clean handling of V0 membership.","headline":"A serious and original framework for spontaneous stochasticity, with a central attainability theorem whose proof has a real, likely repairable gap in the switching-step.","tokens_in":49256,"tokens_out":2309,"would_cite":true,"duration_ms":26750,"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":"Spontaneous stochasticity is a measure-selection principle: whenever an inviscid problem is nonunique, every probability measure on the set of inviscid endpoint states can be realized as the selected law of a suitable regularization.","keywords":["spontaneous stochasticity","measure selection","inviscid limit","nonuniqueness","regularization","singular sets","statistical attractors","renormalization group"],"falsifier":"Take any nonunique inviscid ODE, for example x' = |x|^alpha with alpha in (0,1) at x = 0, and apply the Step-5 construction: choose two regularizations selecting endpoints x and y, build the mixed field with a_theta as in equation (24), compute gamma_theta(s) at the fixed observation time, and check whether gamma_theta(s) - (a_hat_theta(s) x + (1-a_hat_theta(s)) y) is o(1). If for some nonunique system this difference has a positive liminf along subsequences, then the endpoint does not track the convex combination and the equality M = M0 collapses; the falsifier would also produce a concrete n","tokens_in":1638,"feed_emoji":"🎲","tokens_out":1813,"duration_ms":51601,"temperature":0.7,"pith_summary":"This paper argues that spontaneous stochasticity — the emergence of non-Dirac probability statistics in a deterministic inviscid limit — is best understood as a measure-selection principle rather than a turbulence-specific anomaly. Its central attainability theorem shows that, in finite-dimensional systems, once the inviscid problem is nonunique, every probability measure supported on the set of inviscid endpoint states can be obtained as the unique limiting law of a carefully chosen regularization. The physical content of spontaneous stochasticity therefore shifts from the individual dynamical system to the choice of regularization class and ambient measure. The paper also ties nonuniqueness to singular sets detected by Dini-type directional growth, and shows that a semigroup or renormalization-group structure organizes the attainable statistics into statistical attractors.","feed_headline":"Any law can be a spontaneous-stochasticity limit","feed_subtitle":"Nonunique inviscid limits can be tuned to select any target probability measure; the physical content lies in the regularization class.","key_machinery":"The central object is the regularization curve gamma(epsilon) = phi^epsilon_t(x0), the state reached at fixed time t from fixed initial data under the regularized dynamics, together with its pushforward by the ambient measure Leb^epsilon — the normalized Lebesgue measure on [0, epsilon]. Spontaneous stochasticity is defined by whether these pushforwards converge to a non-Dirac measure (strong), have multiple subsequential limits (weak), or collapse to a Dirac mass despite a lack of selection (delta-LSP). The attainability theorem operates on the space Gamma of all regularization curves, and its two load-bearing constructions are: (1) a mixing function a_theta(s) = 1/2 + 1/2 tanh(s(sin 2pi s","core_discovery":"The core claim is Theorem 1: for finite-dimensional systems with continuous bounded inviscid fields and globally Lipschitz regularizations, the set M of all probability measures obtainable as subsequential weak limits of regularization-curve pushforwards is exactly M0 = P(S0), the full space of Borel probability measures on the compact connected set S0 of inviscid endpoint states. Moreover, for every mu in M0 there exists a one-parameter regularization curve gamma_mu whose only accumulation measure is mu. Corollary 3 draws the advertised consequence: whenever the inviscid problem (P0) is nonunique, every non-Dirac measure on S0 is realizable as the selected law of a strong-SpSt regularizatio","pith_inferences":["A direct but undeveloped consequence of Theorem 1 is that, in finite dimension, claims such as 'this system exhibits spontaneous stochasticity' are incomplete without specifying the regularization class and ambient measure; the theorem separates what is intrinsic to the inviscid equation (nonuniqueness and its singularities) from what is imposed by the regularization procedure.","One might test whether the full attainability result survives physically natural regularization classes — viscosity-like, numerical truncation, or stochastic noise — by characterizing the subset of M0 reachable within each class; the paper suggests this subset is where universality classes acquire physical meaning.","The construction behind Theorem 1 does not require the mixed regularization to be physically realizable, only admissible in the broad class V0; a skeptical reader could demand a version where the mixing field satisfies additional constraints such as gradient structure or divergence-freeness, which would likely shrink the attainable set."],"forward_implications":["Whenever the inviscid problem is nonunique, no single distinguished regularization is singled out by the mathematics: the full space of possible selected laws is attainable, so universality classes — regularizations sharing the same limiting statistics — become the meaningful object of study.","Turbulence-inspired criteria based on finite-time separation of nearby trajectories (the TBM criterion) are sufficient diagnostics for non-Dirac limiting laws but are not equivalent to spontaneous stochasticity; scale-averaging alone can produce non-Dirac statistics that TBM cannot detect.","Imposing a semigroup or renormalization-group structure on the regularization parameter constrains the attainable laws to invariant and ergodic measures of the RG dynamics, interpreted as statistical attractors; in one-dimensional gradient systems this recovers deterministic selection, so RG structure is a genuine restriction.","The same measure-selection definition extends beyond finite dimensions; in the authors' companion PDE work, a structured ambient measure — not the flat Leb^epsilon — is needed to obtain spontaneous stochasticity for a passive scalar."],"fun_headline_variants":["Tune regularization to pick any stochastic limit","Any probability law on inviscid states is selectable","Nonuniqueness lets you select any limiting law","Choose any law: it's a regularized limit"],"cache_read_input_tokens":50304,"weakest_assumption_plain":"The load-bearing premise is the convex-combination step of Theorem 1: near a non-Lipschitz singularity, a trajectory driven by the weighted field a_theta f_x + (1-a_theta) f_y is assumed to track the corresponding convex combination of the two selected endpoints, with an error vanishing as the regularization is removed — a genuine geometric tracking property that is not guaranteed by sup-norm closeness of the fields.","fun_headline_variants_meta":{"raw":{"variants":["Tune regularization to pick any stochastic limit","Any probability law on inviscid states is selectable","Nonuniqueness lets you select any limiting law","Choose any law: it's a regularized limit"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000394,"raw_usage":{"total_tokens":1951,"prompt_tokens":840,"completion_tokens":1111,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":584,"completion_tokens_details":{"reasoning_tokens":1049}},"tokens_in":584,"tokens_out":1111,"duration_ms":8823,"temperature":1.0,"reasoning_tokens":1049,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T03:18:35.121417+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take any nonunique inviscid ODE, for example x' = |x|^alpha with alpha in (0,1) at x = 0, and apply the Step-5 construction: choose two regularizations selecting endpoints x and y, build the mixed field with a_theta as in equation (24), compute gamma_theta(s) at the fixed observation time, and check whether gamma_theta(s) - (a_hat_theta(s) x + (1-a_hat_theta(s)) y) is o(1). If for some nonunique system this difference has a positive liminf along subsequences, then the endpoint does not track the convex combination and the equality M = M0 collapses; the falsifier would also produce a concrete n","supporting_citations":[],"review_version":1}