{"id":"c9472c51-b72f-4c41-935f-24b8b6ba2450","arxiv_id":"2411.12910","paper_version":2,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Vanishing diffusivity uniquely selects the solution of the advection equation for divergence-free BV vector fields singular only at the initial time, including Depauw's non-uniqueness example.","lead":"For divergence-free vector fields that are rough, of bounded variation on every time interval away from the starting time, this paper proves that adding a small amount of diffusion and letting it vanish selects a unique solution of the advection equation, even when the diffusion-free equation admits infinitely many solutions. The result covers Depauw's benchmark example and also rules out anomalous dissipation for the selected limit.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The load-bearing step is the one-line time-reversal use of Ambrosio in Theorem 3.3, which goes beyond its global-BV hypothesis; the missing epsilon-localization closes it, so no substantive objection survives.","rationale":"I read the full text and reconstructed the proof around the two delicate places. The backward uniqueness in Theorem 3.3 is indeed stated with an unjustified direct appeal to Theorem 2.1: the hypothesis of Theorem 2.1 is global BV integrability, while the paper's class is only L^1_loc((0,T];BV). This is not merely cosmetic, because Lemma 3.8 and the duality characterization in Section 4 depend on θ_χ being the unique limit of the backward diffusive solutions. However, the reader's proposed repair is valid: fixing ε>0 makes b ∈ L^1([ε,T];BV), so after time reversal the field is in L^1((0,T-ε);BV) and Ambrosio's theorem applies with zero initial datum; letting ε→0 gives uniqueness on (0,T]. I checked the remaining steps: Theorem 2.4's parabolic well-posedness for b ∈ L^2 is standard and the commutator argument is sound; Lemma 3.8's Arzelà-Ascoli argument supplies C([0,T];w-L^2) convergence; Lemma 3.9's mollified duality formula follows from the characteristic representation; and the no-anomalous-dissipation argument in Section 4, despite the OCR-ambiguous notation for the restarted solution in (4.7)-(4.10), is correct once that distinction is restored. The manuscript would be improved by spelling out the epsilon-localization and by explicitly justifying the energy equality used in (4.6), but neither point threatens the central theorem. I therefore find no load-bearing objection.","tokens_in":21034,"tokens_out":27707,"duration_ms":278390,"concrete_test":"Write out the epsilon-localization for Theorem 3.3: for ε>0 set u(s)=θ(T-s), note that u solves ∂_s u + div((-b(T-s))u)=0 on (0,T-ε) with u(0)=0, and that -b(T-·) ∈ L^1((0,T-ε);BV); apply Theorem 2.1 to conclude u=0 on (0,T-ε), then let ε→0. If this derivation cannot be completed for some b in the class, the backward well-posedness used in Lemma 3.8 and Section 4 is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof that the backward problem (BW) is well-posed (Theorem 3.3, Step 2) invokes Ambrosio's Theorem 2.1 after time reversal in one sentence, even though the vector field is only assumed to lie in L^1_loc((0,T];BV). The global hypothesis of Theorem 2.1 is b ∈ L^1((0,T);BV), and the reversed field -b(T-·) generally has non-integrable BV norm near the reversed initial time. This is the most load-bearing step: Lemma 3.8 and the duality identification in Section 4 collapse if the backward limit θ_χ is not unique. The gap is, however, closed by an epsilon-localization argument: for every ε>0, b ∈ L^1([ε,T];BV), so the reversed field lies in L^1((0,T-ε);BV); applying Theorem 2.1 on (0,T-ε) with zero initial datum forces the reversed solution to vanish there. Since ε is arbitrary, the homogeneous backward solution vanishes on (0,T], proving uniqueness. Thus the central claim is not endangered, although the manuscript would benefit from writing this argument out. The other fragile-looking points, including the parabolic well-posedness for b ∈ L^2 and the energy identities in Section 4, check out after restoring the distinction between the original and restarted diffusive solutions in (4.7)-(4.10).","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the advection equation ∂tρ + div(bρ) = 0 on the torus for divergence-free vector fields b in L1_loc((0,T];BV(Td;Rd)) ∩ L2((0,T)×Td;Rd). Theorem 1.4 claims that for every bounded initial datum there is a unique vanishing diffusivity solution, that the weak-* limit of solutions along mollified fields b*wδ is this same solution for every standard mollifier, and that the diffusive family satisfies limsup_{ν→0} ν∫|∇ρν|2 = 0, i.e., no anomalous dissipation. The proof proceeds by proving uniqueness of a backward advection problem (Theorem 3.3), establishing duality identities for both the diffusive and the mollified problems (Lemma 3.7 and Lemma 3.9), and then using these identities to characterize any forward vanishing diffusivity limit. The final part of the paper adapts an energy argument to rule out anomalous dissipation. The class of fields includes Depauw's example, for which the undiffused equation admits infinitely many bounded weak solutions.","tokens_in":21079,"tokens_out":15930,"duration_ms":152479,"significance":"If correct, the result is significant: it identifies parabolic regularization as a selection principle in a rough-vector-field regime where the inviscid equation is highly nonunique, and it also yields a quantitative statement on the absence of anomalous dissipation. The backward-duality strategy is natural and elegant, and the paper is largely self-contained: Theorem 2.4, the parabolic well-posedness result, is proved in full rather than only cited. The claims are concrete and falsifiable, and the class of admissible fields is clearly delineated. However, as written the proof contains two load-bearing gaps: the uniqueness step for the backward problem applies the global BV well-posedness theorem in a setting where only local BV integrability is available, and the no-anomalous-dissipation argument uses identical notation for three different diffusive solutions. Both issues are repairable and I do not see a substantive threat to the main conclusion, but they need to be fixed in the text.","major_comments":[{"comment":"The uniqueness step for the backward problem rests on an unstated epsilon-localization. The proof says 'Switching to the time variable t~=T-t, we can refer to Theorem 2.1 to conclude v=0', but Theorem 2.1 requires the vector field to belong to L1((0,T);BV), whereas the hypothesis is only b∈L1_loc((0,T];BV)∩L2; after time reversal the field -b(T-·) need not be in L1((0,T);BV) because the BV norm may blow up near the reversed initial time. Since Lemma 3.8 and the duality identification in Section 4 would collapse if the backward solution θχ were not unique, this step is load-bearing. It is repairable: for each ε>0, b∈L1([ε,T];BV), so -b(T-·)∈L1((0,T-ε);BV); applying Theorem 2.1 on (0,T-ε) with zero initial datum forces the reversed solution to vanish there, and letting ε↓0 gives uniqueness on (0,T]. Please add this argument.","section":"Theorem 3.3, Step 2 (uniqueness of (BW))"},{"comment":"The no-anomalous-dissipation argument uses the symbol ρν_n for three different objects. The displayed inequality after Eq. (4.6) reads 2ν_n∫_δ^T |∇ρν_n|2 ≤ 2ν_n∫_δ^T |∇ρν_n|2 + 2ν_n∫_δ^T |∇(ρν_n-ρν_n)|2, and the subsequent text refers to 'the unique solution to (ν-PDE) on [δ,1] with initial datum ρ(δ,·)' while keeping the same notation. As printed, the inequality is a tautology and the estimates (4.7)-(4.10) cannot be parsed. The intended proof is reconstructible by writing, for example, ρ̄ν_n for the restarted solution on [δ,T] with initial datum ρ(δ,·), and then applying the energy estimates to ρν_n, ρ̄ν_n, and ρν_n-ρ̄ν_n. This correction is essential for the reader to follow the proof of claim (ii).","section":"Section 4, proof of no anomalous dissipation (after Eq. (4.6))"}],"minor_comments":[{"comment":"In both lemmas, the phrase 'by Lemma 3.6, [0,T]∋t↦∫θχφdx is continuous' should refer to Theorem 3.3, not Lemma 3.6; Lemma 3.6 concerns the diffusive backward problem, while the continuity of θχ is part of Theorem 3.3.","section":"Lemmas 3.8 and 3.9"},{"comment":"The sentence 'we can pass into the limit δ↓0 in the weak formulation of (ν−BW)' should refer to the nondiffusive weak formulation (BW); the same slip appears in Lemma 3.9, where the weak formulation of (3.2) is cited instead of that of (δ-BW).","section":"Theorem 3.3, Step 1"},{"comment":"The mollified equation is written as ∂tvδ+νΔvδ+div(bδvδ)=rδ with rδ:=div(bvδ-(bv)*wδ); the first term on the left is inconsistent with this definition of rδ and should read div(bvδ).","section":"Lemma 3.6, Step 2"},{"comment":"The phrase 'Let ρ be a vanishing viscosity solution' should read 'vanishing diffusivity solution', matching the terminology used in Definition 1.2 and Proposition 2.5.","section":"Section 4, first paragraph"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a solid paper and the main theorem is new. For divergence-free b in L^1_loc((0,T];BV) ∩ L^2, the vanishing viscosity limit selects a unique bounded weak solution of the advection equation, and the selected solution is the same one obtained by mollifying b. Depauw's vector field falls in this class, so the theorem answers a real question in the selection program. It also proves no anomalous dissipation.\n\nThe architecture is good. They prove backward uniqueness for the non-diffusive problem, get convergence of backward diffusive solutions in C([0,T];w-L^2), and use the duality formula to identify every forward accumulation point. That is a clean, honest strategy, and the paper credits Pitcho and De Rosa-Park appropriately. The parabolic well-posedness for L^2 fields is re-proved in Section 2, which is useful and self-contained.\n\nThe soft spots are real but limited. The one-line time-reversal invocation of Ambrosio in Theorem 3.3 Step 2 goes beyond the stated global-BV hypothesis. You need to apply Ambrosio on [ε,T] and vary ε. That works because the terminal datum is zero, but the paper should have written it out. Without that, Lemma 3.8 and the duality identification have a hole. The no-anomalous-dissipation argument also uses the same symbol for the original and restarted parabolic solutions in (4.7)-(4.10), which cost me real effort to untangle. Neither issue is fatal; both are exposition.\n\nThe citation pattern is fine. [21] covers mollification selection only, [1] the global BV class, and the counterexamples in [10] and [16] are in different settings. The paper correctly positions itself.\n\nWho should read it: anyone working on non-uniqueness and selection for transport equations, or on anomalous dissipation for passive scalars. It is a genuine step forward, not a marginal remark. I would send it to a serious referee—ideally one who has worked with BV vector fields and duality arguments—and would expect the main theorem to survive with the epsilon-localization made explicit.","headline":"A genuinely new vanishing-diffusivity selection theorem, with a proof that is sound after a short epsilon-localization repair; worth refereeing seriously.","tokens_in":21845,"tokens_out":5094,"would_cite":true,"duration_ms":44498,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35A02","35D30","35Q49","34A12"],"pacs":[],"model":"deepseek-v4-flash","headline":"Vanishing diffusion selects a unique solution of the advection equation for rough divergence-free fields, including the example with infinitely many weak solutions.","keywords":["vanishing diffusivity","advection equation","transport equation","BV vector fields","rough divergence-free fields","selection principle","anomalous dissipation","duality method"],"falsifier":"Exhibit two distinct bounded weak solutions of the backward problem (BW) for some divergence-free field in the stated class and some smooth $\\chi$; the duality characterization in Section 4 would then collapse. A direct numerical check on the classical non-uniqueness example is also decisive: if two sequences $\\nu_i\\to 0$ produce different weak-star limits of the forward diffusive solutions, Theorem 1.4 is false.","tokens_in":20613,"feed_emoji":"🌊","tokens_out":9826,"duration_ms":91631,"temperature":0.7,"pith_summary":"The paper asks whether adding a small diffusion term to a linear advection equation singles out one weak solution when the velocity field is too rough for uniqueness. It answers yes for divergence-free fields that are of bounded variation away from the initial time and square-integrable in space-time: for every bounded initial datum, the parabolic solutions converge to one well-defined limit as the diffusivity tends to zero, independent of how the limit is taken. This covers the classical example of a divergence-free field with infinitely many distinct bounded advection solutions. The paper also proves that the selected limit does not dissipate energy anomalously and coincides with the limit obtained by first smoothing the velocity field.","feed_headline":"Vanishing diffusion picks one advection solution for rough fields","feed_subtitle":"Even for the classic field with infinitely many bounded weak solutions, the diffusive limit is unique.","key_machinery":"The machine is the backward problem. For each smooth test function $\\chi$, the paper solves the backward transport equation $\\partial_t\\theta_\\chi+\\operatorname{div}(b\\theta_\\chi)+\\chi=0$ with zero data at time $T$, and shows that the backward parabolic solutions $\\theta_\\chi^\\nu$ converge to $\\theta_\\chi$ in $C([0,T];w-L^2(\\mathbb{T}^d))$ as $\\nu\\to 0$. The duality identity $\\iint\\rho^\\nu\\chi=\\int\\rho_{\\mathrm{in}}\\theta_\\chi^\\nu(0)\\,dx$ transfers this backward uniqueness to the forward family $\\rho^\\nu$, so every weak-star limit $\\rho$ must satisfy the same pairing $\\iint\\rho\\chi=\\int\\rho_{\\mathrm{in}}\\theta_\\chi(0)\\,dx$. Since $\\chi$ was arbitrary, the forward limit is uniquely characterized. The no-anomalous-dissipation conclusion then follows from the parabolic energy balance $\\nu\\int|\\nabla\\rho^\\nu|^2\\le \\tfrac12(\\|\\rho_{\\mathrm{in}}\\|_{L^2}^2-\\|\\rho^\\nu(T)\\|_{L^2}^2)$, combined with the $L^2$ conservation of the selected limit on intervals away from time zero. Here BV means bounded variation: the spatial distributional derivative of the velocity field is a finite measure.","core_discovery":"The central claim is Theorem 1.4. Let $b$ be a divergence-free vector field in $L^1_{\\mathrm{loc}}((0,T];BV(\\mathbb{T}^d;\\mathbb{R}^d))\\cap L^2((0,T)\\times\\mathbb{T}^d;\\mathbb{R}^d)$ and let $\\rho_{\\mathrm{in}}\\in L^\\infty(\\mathbb{T}^d)$. Then the unique bounded solutions $\\rho^\\nu$ of the advection--diffusion equation with diffusivity $\\nu$ have a weak-star limit as $\\nu\\to 0$, and that limit is independent of the vanishing sequence: there is a unique vanishing diffusivity solution. Moreover, the same unique limit is obtained by solving the undiffused equation along the mollified fields $b*w_\\delta$ for any standard mollifier $w$, and $\\limsup_{\\nu\\to 0}\\nu\\int_0^T\\int_{\\mathbb{T}^d}|\\nabla\\rho^\\nu|^2\\,dx\\,dt=0$. The proof characterizes the limit through the duality formula $\\iint\\rho\\chi=\\int\\rho_{\\mathrm{in}}\\theta_\\chi(0)\\,dx$, where $\\theta_\\chi$ is the unique bounded solution of the backward advection equation with source $\\chi$ and zero terminal datum; uniqueness of $\\theta_\\chi$ forces all forward approximate solutions to collapse onto one object.","pith_inferences":["The same duality route suggests that any forward approximation scheme whose adjoint converges to the unique backward solution $\\theta_\\chi$ would select the same limit; the paper establishes this for diffusion and for mollification, but the mechanism is general.","Repairing the time-zero regularity gap in the backward uniqueness step by applying the classical theorem on each compact interval $[\\varepsilon,T]$ would make the proof robust and likely allow an even stronger initial singularity, since the backward source is supported away from $t=0$.","An immediate testable extension is the continuous-dependence question left open in the paper: if the backward solution map $\\chi\\mapsto \\theta_\\chi(0)$ is stable, then the vanishing diffusivity solution should depend continuously on initial data in the weak-$L^2$ topology.","A numerical experiment on the classical non-uniqueness example, computing the diffusive limit for several $\\nu$ and comparing with the distinct non-unique weak solutions, would directly exhibit which of the infinitely many solutions parabolic selection chooses."],"forward_implications":["For this class of fields, the parabolic regularization limit is a well-defined functional of the initial datum and the velocity field, independent of the sequence $\\nu_i\\to 0$.","The infinitely many weak solutions of the undiffused equation in the classical non-uniqueness example are not selected by parabolic regularization; the selected solution is singled out by the backward adjoint problem.","Mollifying the velocity field and adding diffusion are equivalent selection mechanisms in this class, so smoothing the field does not change the limiting solution.","The selected solution conserves its $L^2$ norm on every time interval away from the initial singularity and has zero anomalous dissipation, so no energy is lost to unresolved small scales in the limit."],"supporting_citations":[{"why":"Supplies the well-posedness theorem for the undiffused transport equation used to prove uniqueness of the backward problem.","marker":"[1]"},{"why":"Constructs the divergence-free BV vector field with infinitely many distinct bounded advection solutions, the central example covered by Theorem 1.4.","marker":"[13]"},{"why":"Establishes well-posedness of the parabolic advection-diffusion equation for $L^2$ velocity fields, giving the unique diffusive solutions whose limits define vanishing diffusivity solutions.","marker":"[5]"},{"why":"Proposes the backward-uniqueness and duality strategy for selection by regularization, which the paper adapts from mollification to vanishing diffusivity.","marker":"[21]"},{"why":"Supplies the energy argument used to prove the no-anomalous-dissipation part of Theorem 1.4.","marker":"[12]"},{"why":"Earlier treatment of weak and vanishing-viscosity solutions of advection-diffusion equations with rough coefficients that the present paper builds on.","marker":"[4]"}],"fun_headline_variants":["Unique diffusive limit for rough advection fields","Vanishing diffusion selects one advection solution","Rough fields: diffusion picks a single solution","Depauw field case: unique diffusive limit exists"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole identification rests on the backward advection problem having a unique bounded weak solution; the uniqueness step cites a global well-posedness theorem although the field is only locally BV near time zero, a gap that appears repairable but is not written out in the paper.","fun_headline_variants_meta":{"raw":{"variants":["Unique diffusive limit for rough advection fields","Vanishing diffusion selects one advection solution","Rough fields: diffusion picks a single solution","Depauw field case: unique diffusive limit exists"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000257,"raw_usage":{"total_tokens":1554,"prompt_tokens":898,"completion_tokens":656,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":514,"completion_tokens_details":{"reasoning_tokens":596}},"tokens_in":514,"tokens_out":656,"duration_ms":6594,"temperature":1.0,"reasoning_tokens":596,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T17:08:42.066757+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Exhibit two distinct bounded weak solutions of the backward problem (BW) for some divergence-free field in the stated class and some smooth $\\chi$; the duality characterization in Section 4 would then collapse. A direct numerical check on the classical non-uniqueness example is also decisive: if two sequences $\\nu_i\\to 0$ produce different weak-star limits of the forward diffusive solutions, Theorem 1.4 is false.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the well-posedness theorem for the undiffused transport equation used to prove uniqueness of the backward problem."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Constructs the divergence-free BV vector field with infinitely many distinct bounded advection solutions, the central example covered by Theorem 1.4."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes well-posedness of the parabolic advection-diffusion equation for $L^2$ velocity fields, giving the unique diffusive solutions whose limits define vanishing diffusivity solutions."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Proposes the backward-uniqueness and duality strategy for selection by regularization, which the paper adapts from mollification to vanishing diffusivity."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Earlier treatment of weak and vanishing-viscosity solutions of advection-diffusion equations with rough coefficients that the present paper builds on."}],"review_version":1}