{"id":"73a4ef4f-6cef-41af-88c7-0415ee8c0e1c","arxiv_id":"2506.15333","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Any BV-in-time curve of probability measures with finite first moment and its singular flux admit a probabilistic representation by reparametrized Lipschitz curves, and by augmented BV curves with explicit jump transitions.","lead":"This paper proves that curves of probability measures with bounded variation in time, together with their possibly singular flux, can be represented as a superposition of trajectories solving an augmented characteristic system. The result extends the classical Ambrosio-Gigli-Savare superposition principle to the p=1 Wasserstein setting and describes what happens at jump times.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 4.7's reduction to the Euclidean norm is unproved; Lemmas 2.2(iii), A.1(2), and 2.7 fail for non-strictly-convex norms, so the main representation theorems are not yet established for arbitrary norms.","rationale":"The reader's weakest-assumption analysis identifies exactly the same point: the proof of Theorem 4.7 declares the Euclidean-norm reduction without proof, and the main theorems are stated for arbitrary norms. My stress-test confirms that this is not merely cosmetic: strict convexity is used at several concrete technical steps, and the key lemmas on which the projection identity depends are false for non-strictly-convex norms such as ℓ∞. The paper is otherwise careful, detailed, and honest about its limitations, and the gap is plausibly repairable by an approximation argument or by restricting the statement to strictly convex norms. Because the central claim as stated lacks a complete proof for the full claimed generality, the appropriate verdict is conditional acceptance rather than unconditional acceptance.","tokens_in":65963,"tokens_out":11880,"duration_ms":120914,"concrete_test":"Take R^2 with the ℓ∞ norm and attempt to re-derive the conclusion p♯ζ≺λ of Lemma 2.7 from assumptions (2.13)–(2.14) without invoking Lemma 2.2(iii). Concretely, inspect the step where (2.25) yields λj and λ̄j: check whether λj is a submeasure of λ using only the definition ζ≺θ, without strict convexity. If a sequence satisfying (2.13)–(2.14) can be produced for which p♯ζ is not a submeasure of λ, then Lemma 2.7 fails in that setting. If the step fails, verify whether an approximation of the ℓ∞ norm by strictly convex norms recovers Theorem 4.7's identities (4.32)–(4.34) in the limit; this would confirm the gap is repairable.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The proof of Theorem 4.7 begins with the declaration that 'it is not restrictive to assume that ... ||·|| is the Euclidean norm'. This reduction is load-bearing but unproved. All norm-dependent objects change when the norm is replaced: total variation |(µ,ν)|, the Wasserstein metric W1, the minimal flux selected by (1.13), and the submeasure relation used in Lemma 2.7. Strict convexity is used critically in Lemma 2.2(iii), in Lemma A.1(2), and in Lemma 2.7 to pass from (2.24)–(2.25) to the conclusion p♯ζ≺λ. For a non-strictly-convex norm such as ℓ∞ on R^2, Lemma 2.2(iii) is false: with θ=(1,0)δ, ζ=(1/2,1/2)δ, ζ_C=(1/2,−1/2)δ one has θ=ζ+ζ_C and |θ|∞=|ζ|∞+|ζ_C|∞, yet ζ is not a scalar multiple of θ. Since Theorems 5.1(2) and 6.5 rely on Theorem 4.7, the main results are not yet proved for the stated arbitrary-norm generality; either a norm-reduction argument must be supplied or the statements restricted to strictly convex norms.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies measure-valued solutions (µ,ν) to the continuity equation ∂tµ+divν=0 in [0,∞)×R^d, where µ is a curve of probability measures with finite first moment and ν is a Radon vector flux. It characterizes BV curves in (P1,W1) through such equations, selects minimal singular fluxes, constructs an auxiliary continuity equation in an augmented phase space driven by a bounded autonomous vector field, and derives two probabilistic representations: one by reparametrized Lipschitz curves (Theorem 5.1) and one by augmented BV curves with explicit absolutely continuous, Cantor, and jump parts (Theorem 6.5). The paper includes detailed proofs, several appendices, and examples illustrating the role of the minimality condition.","tokens_in":66190,"tokens_out":24186,"duration_ms":235323,"significance":"If correct, the paper provides a substantial and natural extension of the classical superposition principle to the BV and p=1 setting with singular fluxes. The minimal-flux selection and the augmented-phase-space method are original and likely to be useful for evolutionary PDEs, gradient flows, and optimal transport applications. The statements are precise, the proofs are detailed, and Section 7 carefully discusses the sharpness of the hypotheses through examples (e.g., Examples 7.2 and 7.3). The use of the classical superposition principle [11] is an external, standard tool and does not introduce circularity. The main reservations concern norm-dependence in the proof of Theorem 4.7 and a missing justification in the decomposition step of Proposition 5.2.","major_comments":[{"comment":"The proof begins with the declaration 'it is not restrictive to assume that ϱ=|(µ,ν)|, ||·|| is the Euclidean norm ..., and θ≡1'. The first and third reductions are legitimate, but the Euclidean-norm reduction is not justified by Lemma 4.6 alone, which only rescales the velocity field by a positive factor. The objects |(µ,ν)|, W1, the minimality condition (1.13), and the submeasure relation ≺ all depend on the chosen norm. Lemma 2.2(iii), which is used in the proof of Lemma 2.7 to pass from (2.24)–(2.25) to the conclusion λj≺λ, is false for non-strictly convex norms: taking the ℓ∞ norm on R^2 with θ=(1,0)δ, ζ=(1/2,1/2)δ and ζC=(1/2,-1/2)δ gives θ=ζ+ζC and |θ|∞=|ζ|∞+|ζC|∞, yet ζ is not a scalar multiple of θ. Since Lemma 2.7 is the step that identifies the projected limit as a submeasure of (µ,ν) in Theorem 4.7, the proof as written establishes the augmented representation only for strictly convex (in particular Euclidean) norms. The authors should either supply the missing norm-invariance argument (e.g., prove the Euclidean case and then apply Lemma 4.6 with θ = ||(τ_E,v_E)||_orig^{-1}, after observing that minimality and ≺ are invariant under equivalent norms) or restrict the statements of Theorems 5.1 and 6.5 to strictly convex norms.","section":"§4.2, Theorem 4.7 (proof)"},{"comment":"In the proof of (5.11), the statement 'Since νa≪µ and D0∩D+=∅, the second and third equalities follow' is not sufficient: a point (t,x) may be reached at D+-times by some curves and at D0-times by others, so the supports of e♯(x′ηL|D0) and e♯(t′ηL|D+) need not be disjoint merely because D0 and D+ are disjoint in the (s,y) domain. The missing argument is that the curves solve the autonomous system (5.7a), so that τ>0 on the former image and τ=0 on the latter, making the images disjoint up to negligible sets. As written, this step is incomplete, and it is used later in the proof of Theorem 5.3 and consequently in Theorem 6.5.","section":"§5.1, Proposition 5.2"}],"minor_comments":[{"comment":"The proof uses the same symbol λj both for the vector measure in (2.17) and for the positive limit in (2.20)–(2.25); this makes the argument very hard to follow and should be corrected with distinct notation.","section":"§2.2, Lemma 2.7"},{"comment":"The phrase 'in the sense of Definition 3.2. 2.3.' contains a stray '2.3.' and should be cleaned up.","section":"§3, Theorem 3.4(1)"},{"comment":"The formula has a duplicated parenthesis in v(t,u(t,r)))/τ(...); also, the equality with dνa/dµ should be interpreted on the set where τ>0, which deserves a brief remark.","section":"§6.3, Theorem 6.5, Eq. (6.27a)"},{"comment":"The proof invokes 'We can select the Euclidean norm' although the lemma is stated for a strictly convex norm; this is acceptable, but the assumption should be relaxed or the wording clarified, since Claims (1) and (3) do not require strict convexity.","section":"Appendix A, Lemma A.1(3)"}],"recommendation":"major_revision","confidential_remarks":"The main gap is the norm reduction in the proof of Theorem 4.7. It is substantive but appears fixable: the needed ingredients (Lemma 4.6, equivalence of norms on finite-dimensional spaces, and norm-invariance of the submeasure relation) are already close to the surface. If the authors prefer, restricting the main theorems to strictly convex norms would also remove the issue, at some loss of generality. The use of [11], co-authored by one of the present authors, is standard and not circular. I recommend major revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The short version: this is an important paper, and most of it is solid. It extends the superposition principle to p=1, BV-in-time flows, and singular fluxes using two genuinely new mechanisms: the augmented phase-space lifting and the augmented BV curves with explicit jump transitions. The minimal singular flux selection is a useful concept, and the examples (including non-minimal cases that still admit representations, and the failure of weak* stability) are honest and illuminating. The proofs are detailed, and the authors are upfront about limitations. People working in optimal transport, measure-valued PDEs, or gradient flows will want to read this.\n\nThe soft spot is real and load-bearing. Theorem 4.7, the essential bridge to both main representation results, begins its proof by declaring it is 'not restrictive' to take the Euclidean norm. That reduction is not justified, and the stress-test example shows why: Lemma 2.2(iii), which is used to pass to the submeasure relation in Lemma 2.7 and in the proof of Theorem 5.3, is simply false for non-strictly-convex norms. The counterexample with the ℓ∞ norm is concrete, and it means the total variation, the minimal flux, and the whole submeasure framework are not invariant under the claimed reduction. So the main theorems, as stated for arbitrary norms, are not yet proved. The likely fix is either to restrict the statements to strictly convex norms (covering ℓp for 1<p<∞) or to supply a real approximation argument. This is a gap but not a fatal flaw; the architecture of the paper is sound.\n\nI would send this to a serious referee. It deserves careful review, and the norm issue should be the first thing on the report. The reader's ACCEPT is reasonable in spirit, but underweights this point: a theorem stated for all norms with a proof that only treats Euclidean is not fully proven. Once the statements are corrected or the argument completed, this is a strong contribution. For now: revise, then publish.","headline":"A substantial and mostly rigorous extension of the superposition principle to BV curves and singular fluxes, but the main theorems are stated for arbitrary norms while the proofs only handle strictly convex norms; that gap needs fixing.","tokens_in":66745,"tokens_out":2566,"would_cite":true,"duration_ms":29453,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35F25","49Q22","28A33"],"pacs":[],"model":"deepseek-v4-flash","headline":"Bounded-variation measure flows still admit a superposition principle, provided the singular flux is made minimal.","keywords":["continuity equation","singular flux","BV curves","superposition principle","Wasserstein space","minimal flux","augmented phase space","measure-valued solutions"],"falsifier":"In R² with the ℓ∞ norm, run the construction of Theorem 4.7 on the minimal pair μ_t=(1−t)δ_{(0,0)}+tδ_{(1,1)}, ν=L¹|[0,1]⊗L¹|segment, and check whether the minimal flux is unique and whether the projection identity π♯σ=ν holds for every weak-∗ limit σ. If uniqueness or the identity fails, the Euclidean-norm reduction is load-bearing and the stated generality is false.","tokens_in":65726,"feed_emoji":"","tokens_out":7953,"duration_ms":81689,"temperature":0.7,"pith_summary":"Solving the continuity equation for probability measures with a singular flux—one not absolutely continuous in time-space—looks like it should break the classical superposition principle, which represents the solution as an average of characteristic curves. This paper establishes that the principle survives in the bounded-variation, p=1 setting: every solution pair (μ,ν) can, after replacing the singular part of ν by a minimal flux, be written as the projection of an auxiliary continuity equation in an augmented phase space driven by a bounded autonomous velocity field. From that auxiliary equation the authors obtain two probabilistic representations: one by reparametrized 1-Lipschitz curves solving the characteristic system ẏ=(τ(y),v(y)), and one by augmented BV curves that explicitly resolve the trajectory followed during each jump. The singular part of the flux is thereby translated into the Cantor and jump parts of the random BV curves, with jumps directed by the singular flux. A sympathetic reader would care because this converts a purely Eulerian evolution law with irregular flux into a Lagrangian particle picture, the missing tool for many measure-valued PDE applications.","feed_headline":"Singular flux still yields particle trajectories","feed_subtitle":"For BV measure flows, replacing a flux by its minimal part restores a Lagrangian picture of jumps.","key_machinery":"The argument is carried by three objects: the submeasure order ζ≺θ, meaning ζ=λθ for a Borel [0,1]-valued function λ, which selects minimal singular fluxes by minimizing |ζ| among submeasures with the same divergence; the augmented phase space I×R^d with artificial time s, in which the pair (μ,ν) is lifted to a solution σ of ∂sσ+∂tσ0+div σ=0 with a normalized autonomous velocity field (τ,v)=d(μ,ν)/d|(μ,ν)|; and the evaluation and projection maps that push η on Lipschitz curves back to (μ,ν). The bridge to BV curves is the class ABV of continuous maps u on the ordered parameter space I×[0,1] whose slices u(t,·) are constant except at jump times, where they are Lipschitz transition curves of constant speed; composing the reparametrization maps T and S converts arc-length Lipschitz trajectories into augmented BV curves without losing information. The crucial work of these objects is to carry the Eulerian singular flux information into Lagrangian jump transitions.","core_discovery":"The central discovery is that the pairing of a BV curve μ∈BV_loc(I;P1(R^d)) with a flux ν in ∂tμ+div ν=0 is not an obstruction to particle representations if ν is minimal. Theorem 5.1 asserts that for any P1-solution (μ,ν), after replacing ν by a minimal pair (μ,ν̄) with ν̄⊥≺ν⊥, there exists a probability measure η on reparametrized Lipschitz curves y=(t,x) with t(0)=0, t increasing to +∞, satisfying μ=e♯(t′η_L), ν̄=e♯(x′η_L), |(μ,ν̄)|=e♯(‖y′‖η_L), and η-a.e. curve solves y′(s)=(τ(y(s)),v(y(s))), y(0)=(0,x), where (τ,v) is the density of (μ,ν) with respect to |(μ,ν)|. Theorem 6.5 then gives a companion representation by a probability measure on augmented BV curves in which the absolutely continuous, Cantor, and jump parts obey explicit equations and the jump transitions follow the direction of ν⊥. Theorems 5.3 and 5.4 add fine structure: representing curves are injective under minimality, and when the sharp variation identity (3.7) holds, jump transitions are straight segments at constant speed.","pith_inferences":["The paper's examples show minimality is sufficient but not necessary; a natural next test is to identify the exact class of nonminimal pairs that still admit a representation, conjecturally those whose singular flux is reachable by flat-time trajectory segments.","The D+/D0 decomposition of trajectories suggests an estimation scheme for empirical data: given a time series of measures, infer the distribution of jump-start times and jump paths from the singular flux, then compare with η's marginals.","If the strict-convexity reduction fails for non-Euclidean norms, the augmented BV representation may still hold but with non-unique minimal fluxes; checking the ℓ∞ norm would settle whether uniqueness is essential or merely technical."],"forward_implications":["Every minimal P1-solution, including jump discontinuities, has a Lagrangian interpretation: η is concentrated on injective curves, so mass does not split and rejoin except as a measure-averaged superposition.","The absolutely continuous part of the flux is produced only along strictly increasing time segments (t′>0), while the singular part is produced on flat time segments (t′=0), giving a clean geometric split of νa and ν⊥.","When the sharp condition (3.7) holds, jumps are straight-line transitions at constant speed, so the BV metric variation matches the average Euclidean length of jump paths.","The BV representation gives direct formulas for the left and right limits μ−_t and μ+_t as the marginals of u(t,0) and u(t,1), allowing jump size to be read off from the path measure."],"supporting_citations":[{"why":"Provides the classical superposition principle for absolutely continuous Wasserstein curves and the gradient-flow theory that the BV result extends.","marker":"[11]"},{"why":"Establishes the metric superposition principle for BV curves in (P1,W1), the baseline the paper refines with flux information.","marker":"[1]"},{"why":"Supplies the characterization of absolutely continuous curves in Wasserstein spaces that motivates the augmented equation.","marker":"[34]"},{"why":"Decomposes solenoidal vector charges into elementary currents, the source of the submeasure and minimality idea.","marker":"[43]"},{"why":"Introduces the Young-measure technique linking continuity equations with low-regularity velocity fields to characteristic curves.","marker":"[5]"},{"why":"Connects the three superposition principles, for currents, continuity equations, and curves of measures, that the paper extends to the singular-flux setting.","marker":"[44]"},{"why":"Develops the reparametrization technique for rate-independent BV evolutions adapted here to build augmented BV curves.","marker":"[38]"}],"fun_headline_variants":["Minimal singular flux restores particle trajectories in BV flows","Superposition principle for BV curves with singular flux","BV measure flows: minimal flux yields Lagrangian representation","Particle paths survive singular flux via minimal choice"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The reduction to a strictly convex (Euclidean) norm is declared without a full proof, and minimality, uniqueness of minimal fluxes, and polar decomposition arguments depend on strict convexity; if the norm is not strictly convex, the equivalence steps in Theorem 4.7 can fail.","fun_headline_variants_meta":{"raw":{"variants":["Minimal singular flux restores particle trajectories in BV flows","Superposition principle for BV curves with singular flux","BV measure flows: minimal flux yields Lagrangian representation","Particle paths survive singular flux via minimal choice"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000218,"raw_usage":{"total_tokens":1580,"prompt_tokens":1229,"completion_tokens":351,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":845,"completion_tokens_details":{"reasoning_tokens":290}},"tokens_in":845,"tokens_out":351,"duration_ms":3576,"temperature":1.0,"reasoning_tokens":290,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T19:36:00.332499+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"In R² with the ℓ∞ norm, run the construction of Theorem 4.7 on the minimal pair μ_t=(1−t)δ_{(0,0)}+tδ_{(1,1)}, ν=L¹|[0,1]⊗L¹|segment, and check whether the minimal flux is unique and whether the projection identity π♯σ=ν holds for every weak-∗ limit σ. If uniqueness or the identity fails, the Euclidean-norm reduction is load-bearing and the stated generality is false.","supporting_citations":[{"cited_title":"Ambrosio, N","cited_arxiv_id":null,"evidence_quote":"Provides the classical superposition principle for absolutely continuous Wasserstein curves and the gradient-flow theory that the BV result extends."},{"cited_title":"Abedi, Z","cited_arxiv_id":null,"evidence_quote":"Establishes the metric superposition principle for BV curves in (P1,W1), the baseline the paper refines with flux information."},{"cited_title":"Lisini, Characterization of absolutely continuous curves in Wasserstein spaces , Calc","cited_arxiv_id":null,"evidence_quote":"Supplies the characterization of absolutely continuous curves in Wasserstein spaces that motivates the augmented equation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Decomposes solenoidal vector charges into elementary currents, the source of the submeasure and minimality idea."},{"cited_title":"Ambrosio, Transport equation and Cauchy problem for BV vector fields, Invent","cited_arxiv_id":null,"evidence_quote":"Introduces the Young-measure technique linking continuity equations with low-regularity velocity fields to characteristic curves."},{"cited_title":"Stepanov and D","cited_arxiv_id":null,"evidence_quote":"Connects the three superposition principles, for currents, continuity equations, and curves of measures, that the paper extends to the singular-flux setting."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Develops the reparametrization technique for rate-independent BV evolutions adapted here to build augmented BV curves."}],"review_version":2}