{"id":"47ecfe5a-cc74-449b-8b41-930e86cb7e25","arxiv_id":"1908.09055","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A Jordan-Kinderlehrer-Otto style variational scheme using L1 memory weights is shown to converge weakly in L1 to the unique weak solution of the time-fractional Fokker-Planck equation.","lead":"This paper builds a variational time-stepping scheme for a time-fractional Fokker-Planck equation, where each step solves a minimal-transport-cost problem, and proves the scheme converges to the PDE solution as the time step shrinks. It matters because a Wasserstein gradient-flow picture for this nonlocal-in-time diffusion was an open problem, and the scheme works as a practical algorithm.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Full-sequence convergence to a unique weak solution is unsupported: uniqueness for Definition 2.1 is imported from a merely formally equivalent reformulation, and Section 2.3 leaves well-posedness of (1.1) open.","rationale":"The reader's conditional verdict is supported. The scheme and its estimates are coherent: the Euler-Lagrange equation, L1 time-memory weights, energy and moment estimates, compactness, and discrete integration-by-parts all line up to show that any subsequential limit is a weak solution. I separate the two issues the reader flags. The right-sided L1 convergence is a genuine omission, since Theorem 3.1 proves only the left-sided error while the proof of Theorem 4.2 needs D_tau^alpha phi -> tD_T^alpha phi; however, this is a standard symmetric estimate and a short direct Taylor argument or the change of variables t -> T-t should repair it without changing the argument. The uniqueness issue is more load-bearing because it is the only step that converts subsequential convergence into the theorem's stated full convergence to a unique limit. Without uniqueness for Definition 2.1, the theorem as stated is not proven; with it, the conclusion follows. This does not require rejection: the paper's central construction is novel and the gap is addressable, so the conditional recommendation stands.","tokens_in":24158,"tokens_out":24641,"duration_ms":239219,"concrete_test":"Verify the equivalence of the weak-solution classes by an explicit calculation: take a weak solution rho of (2.7), apply the fractional integration-by-parts identity (2.4) to the term I^(1-alpha) d_t rho tested against a smooth phi with phi(T)=0, and check that it yields exactly (2.9); then repeat in the converse direction. If either direction produces an extra boundary term or requires additional regularity, the uniqueness imported from [10, Theorem 3.3] does not transfer to Definition 2.1 and Theorem 4.2 must be weakened to subsequential convergence.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing gap is in the statement of Theorem 4.2, not in the discrete construction. The proof of Theorem 4.2 shows that every subsequential L1 weak limit of the interpolants satisfies the weak formulation of Definition 2.1; the convergence of the whole sequence is then attributed to 'the unique weak solution'. But Section 2.3 states that no well-posedness result is available directly for (1.1), and the only cited existence/uniqueness result, [10, Theorem 3.3], is for the reformulation (2.7). Remark 2.1 derives the equivalence of (1.1) and (2.7) only formally, with no proof that the weak-solution classes coincide. Thus uniqueness in the sense of Definition 2.1 is not established. If that uniqueness fails, different subsequences could converge to different weak solutions and the asserted convergence rho_tau -> rho of the full sequence, and the word 'unique', have no basis. The right-sided L1 derivative limit used inside the proof is also not covered by Theorem 3.1, but it is a symmetric variant of the left-sided estimate and appears readily repairable; the uniqueness gap is the more consequential obstruction.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a Wasserstein gradient-flow (JKO-type) variational formulation for the time-fractional Fokker-Planck equation (1.1), where the Caputo time derivative is discretized with the L1 scheme. The discrete scheme, Scheme 4.1, replaces the immediate previous density in the classical JKO functional by a convex combination of all past densities, thereby encoding the memory of the fractional derivative. The main result, Theorem 4.2, asserts that the piecewise constant interpolants of the discrete minimizers converge weakly in L^1((0,T) x R^d) to the unique weak solution of (1.1) in the sense of Definition 2.1. The proof derives the Euler-Lagrange equation (5.17), establishes a priori estimates in Lemmas 5.3-5.5, uses a discrete integration-by-parts identity in Lemma 3.3, and passes to the limit in the discrete weak formulation. Numerical experiments in one and two dimensions illustrate the behavior of the scheme.","tokens_in":24314,"tokens_out":4592,"duration_ms":45748,"significance":"If the gaps identified below are repaired, this would be a valuable and novel contribution: it gives the first JKO-type variational formulation and convergent time discretization for a nonlocal-in-time Fokker-Planck equation, directly addressing a question raised by Kemppainen and Zacher. The proof strategy is sound and nearly all technical steps check out: the Euler-Lagrange equation, the a priori estimates, and the discrete integration by parts are genuinely derived and not fitted. The paper also ships a concrete computational algorithm based on entropic regularization, and the numerical section is reproducible in structure. The main theoretical claim is currently overstated because uniqueness of weak solutions in the sense of Definition 2.1 is not established within the paper, and one discrete limit used in the proof is not covered by the stated theorem. These are repairable, but they are load-bearing for the full convergence statement.","major_comments":[{"comment":"The assertion in Theorem 4.2 that the full sequence rho_tau converges to the unique weak solution of (1.1) in the sense of Definition 2.1 is not supported by the proof. The proof of Theorem 4.2 only shows that every subsequential L1 weak limit of the interpolants satisfies the weak formulation (2.9). Uniqueness in the class of Definition 2.1 is imported from [10, Theorem 3.3], but that theorem applies to the reformulation (2.7), and Remark 2.1 establishes equivalence of (1.1) and (2.7) only formally. Section 2.3 explicitly leaves a rigorous study of well-posedness of (1.1) to future work. If the weak-solution classes of (1.1) and (2.7) differ, different subsequences could converge to different weak solutions, and neither the word 'unique' nor the convergence of the full sequence follows. The theorem should either be weakened to subsequential convergence to a weak solution, or uniqueness for Definition 2.1 (or the equivalence of the two weak formulations) must be proved.","section":"Section 2.3 and Theorem 4.2"},{"comment":"The limit passage around Eq. (5.34) uses the convergence D_tau^alpha phi(t) -> t D_T^alpha phi(t) as tau -> 0, but Theorem 3.1 provides error estimates only for the left-sided L1 approximation dbar_tau^alpha phi and for the boundary term involving the weights b_n^(n). The right-sided discrete derivative D_tau^alpha defined in Eq. (3.12) is a symmetric analogue, and the required convergence is very likely true, but it is neither stated nor proved. Since the weak formulation (2.9) is obtained precisely from this term, the proof of Theorem 4.2 has a gap at a load-bearing step. Add an explicit error estimate or a convergence proof for the right-sided L1 approximation, along the same lines as the left-sided estimate in Theorem 3.1.","section":"Theorem 4.2 proof and Theorem 3.1"},{"comment":"The proof of the uniform second-moment bound invokes Corollary 5.1 with phi = |x|^2, but Corollary 5.1 is stated for phi in C_c^infty(R^d). The sentence 'This choice is justified by the finiteness of the second moment of each of the rho_n' is not a proof; a truncation/approximation argument is needed. This is a technical point, but it is load-bearing because Lemma 5.4 underpins Lemma 5.5 and the subsequent compactness argument in Theorem 4.2, and it should be addressed explicitly.","section":"Lemma 5.4"}],"minor_comments":[{"comment":"In the proof of Lemma 5.4, the text reads 'the assumptions F(rho_0)<infty and M_2(rho_0)<0'; the second condition should be M_2(rho_0)<infty.","section":"Lemma 5.4"},{"comment":"The admissible set A is defined as those u with the integral of u equal to the integral of u_n, but at step n the constraint should presumably reference a fixed mass, likely the initial mass or the previous iterate; the current definition appears to be a typo and should be corrected.","section":"Section 6.1, Eq. (6.38)"},{"comment":"Definition 2.1 uses test functions in C^infty([0,T] x R^d) with phi(T)=0, whereas the proof of Theorem 4.2 fixes phi in C_c^infty((-infty,T) x R^d) and does not explicitly justify that this class is sufficient to verify the weak formulation for all admissible test functions. A density or approximation statement should be added.","section":"Definition 2.1 and Theorem 4.2 proof"},{"comment":"There are several typos, including 'picewise-constant' before Eq. (4.15), 'duffusion' in the abstract, and the notation for the convex combination density is introduced as rho^{n-1} but sometimes typeset as rho^{n-1}; please unify the notation throughout Section 5.","section":"General presentation"}],"recommendation":"major_revision","confidential_remarks":"The central proof strategy is sound and the discrete machinery is well executed, but the main theorem as stated depends on a uniqueness result that the paper itself admits is not available for (1.1) in the sense of Definition 2.1. The authors should either prove uniqueness in this weak class, prove the equivalence of the weak formulations of (1.1) and (2.7), or restate the theorem as subsequential convergence. The right-sided L1 convergence gap is also directly repairable. I see no grounds for rejection, provided these load-bearing points are fixed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Read Duong-Jin on the Wasserstein gradient flow formulation of the time-fractional Fokker-Planck equation. The headline: the first JKO-type variational scheme for this equation is a genuine contribution, and the convergence proof is mostly sound, but the theorem as stated overreaches. Full-sequence convergence to 'the unique weak solution' is not actually proven, because uniqueness in their weak-solution sense is imported from Camilli-De Maio, who treat an equivalent reformulation, not (1.1) directly. The paper itself admits this in Section 2.3. So the honest statement is subsequential convergence to a weak solution, with full convergence contingent on establishing uniqueness.\n\nWhat's new and good: the scheme itself — L1 time discretization of the Caputo derivative encoded as a convex combination of all past densities in the Wasserstein term — is a natural and clever adaptation of JKO, and it directly answers the open question Kemppainen-Zacher flagged. The proof follows the classical route: Euler-Lagrange equation, a priori estimates on energy, entropy and second moment, discrete integration-by-parts, compactness, and passage to the limit. I checked the main steps; they line up. The limiting equation is exactly their weak formulation (2.9). The discrete integration-by-parts lemma (3.3) is a nice piece of work. Numerical experiments are illustrative, with caveats below.\n\nSoft spots, in decreasing order of seriousness. First, the uniqueness gap described above. The proof shows every subsequential limit satisfies the weak formulation; the word 'unique' needs a theorem the paper does not provide. This is load-bearing for the statement but not for the core construction — it is a repairable missing piece, not a fatal flaw. Second, inside the proof of Theorem 4.2, the limit D_tau^alpha phi -> tD_T^alpha phi is asserted without proof; Theorem 3.1 proves the left-sided estimate only. This looks symmetric and easily fixed, but as written it is a gap. Third, a minor sign slip in (5.33) — inconsequential. Fourth, the numerics use entropic regularization with gamma = 1/N and don't report spatial mesh sizes or Dykstra tolerances, so the observed rates are not fully reproducible. The rates also deteriorate as alpha increases, and the authors note the mechanism is elusive — fine to report, but it undercuts any claim of optimality.\n\nCitation pattern looks honest: Li-Liu [32] and Duong-Lu [14] are cited and distinguished; the paper does not overclaim novelty relative to them.\n\nWho this is for: researchers in Wasserstein gradient flows, fractional PDEs, and numerical analysis of nonlocal evolution equations. It deserves a serious referee, but the referee should push for a corrected theorem statement or a genuine uniqueness proof, and for the right-sided discrete estimate.\n\nRecommendation: send to peer review, with major revision required.","headline":"First JKO-type variational scheme for the time-fractional Fokker-Planck equation, with a mostly sound convergence proof whose stated full-sequence uniqueness result outruns the imported well-posedness.","tokens_in":24958,"tokens_out":3115,"would_cite":true,"duration_ms":30870,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35R11","49Q22","65M12","35Q84"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that a JKO-type variational scheme for the time-fractional Fokker-Planck equation converges weakly in $L^1$ to the unique weak solution as the time step tends to zero.","keywords":["Wasserstein gradient flow","time-fractional Fokker-Planck equation","JKO scheme","Caputo fractional derivative","L1 scheme","subdiffusion","convergence of time-discretization scheme","optimal transport"],"falsifier":"Concretely, exhibit two weak solutions of (1.1) in the sense of Definition 2.1 with the same initial datum, or verify directly whether the right-sided L1 approximant $D^\\alpha_\\tau \\phi$ converges to $tD^\\alpha_T \\phi$ as $\\tau\\to 0$; failure of either check would break the identification of the limit as the unique weak solution.","tokens_in":23850,"feed_emoji":"🧮","tokens_out":9650,"duration_ms":88253,"temperature":0.7,"pith_summary":"The paper addresses a question left open in the study of subdiffusion: whether the time-fractional Fokker-Planck equation, whose Caputo time derivative is nonlocal, admits a variational formulation in the Wasserstein space along the lines of the classical JKO scheme. It proposes a discrete scheme in which each step minimizes a Wasserstein-squared term plus the free energy, with the previous state replaced by a convex combination of all past densities whose weights come from the L1 approximation of the Caputo derivative. The main result is that, as the time step tends to zero, the piecewise-constant interpolants of the minimizers converge weakly in $L^1$ to a weak solution of the equation, identified with the unique weak solution imported from a related reformulation. This matters because it gives the first minimizing-movement (gradient-flow) characterisation and convergent time discretization for a nonlocal-in-time Fokker-Planck equation, and it reduces to the classical JKO scheme as the fractional order tends to one.","feed_headline":"Fractional Fokker-Planck equation gets a convergent JKO scheme","feed_subtitle":"Memory of the Caputo derivative is encoded in a convex combination of past densities; the limit is a weak solution.","key_machinery":"The machinery is the L1 scheme for the Caputo derivative together with its weights $b^{(n)}_i$, which turn the nonlocal fractional derivative into a weighted sum over past time levels and enter the scheme through the convex combination $\\bar{\\rho}^{n-1} = \\sum_{i=0}^{n-1}(-b^{(n)}_{n-i})\\rho^i$. A discrete integration-by-parts identity (Lemma 3.3) ties the left-sided L1 derivative used in the scheme to a right-sided L1 derivative acting on test functions, and Theorem 3.1 supplies the approximation errors that allow the limit passage. Around this, energy estimates on the free energy, second moments, and summed Wasserstein distances provide the compactness needed to extract a weakly convergent subsequence and identify the limit as a weak solution of (1.1).","core_discovery":"The central claim is Theorem 4.2: for any fixed final time $T>0$, under Assumption 4.1 on the confinement potential, the piecewise-constant interpolants $\\rho_\\tau$ of the minimizers of Scheme 4.1 converge, as $\\tau\\to 0$, weakly in $L^1((0,T)\\times\\mathbb{R}^d)$ to the unique weak solution $\\rho$ of the time-fractional Fokker-Planck equation (1.1) in the sense of Definition 2.1. The novelty is that the memory of the Caputo derivative is encoded in the discrete scheme through the convex combination $\\bar{\\rho}^{n-1}$ of all previously computed densities, with weights given by the L1 quadrature for the fractional derivative; the proof passes to the limit in the Euler-Lagrange equation using energy estimates, a discrete Gronwall inequality, and compactness in $L^1$. If the theorem is correct, it supplies the analogue, for fractional-in-time Fokker-Planck equations, of the classical JKO variational principle, and it recovers the classical JKO scheme as $\\alpha\\to 1$.","pith_inferences":["If the formal equivalence between (1.1) and (2.7) is made rigorous, the uniqueness assumption would be fully justified; a direct well-posedness proof for Definition 2.1 would also remove the main gap in the convergence theorem and let the result stand alone.","The proof's use of the right-sided L1 approximation, asserted but not proved, can be tested independently; establishing a symmetric error estimate would make the limit passage self-contained and likely transfer to other nonlocal-in-time gradient flows.","The same convex-combination memory mechanism could yield JKO-type schemes for other nonlocal-in-time equations, such as Volterra-type Fokker-Planck equations with more general memory kernels, where the L1 weights would be replaced by the corresponding quadrature.","A quantitative version of the theorem could be pursued by combining sharper L1 error estimates with regularity of the fractional Fokker-Planck solution; the empirical rates around 0.47 in Wasserstein distance provide a target for such a rate."],"forward_implications":["The time-fractional Fokker-Planck equation gains a variational (minimizing-movement) structure, so questions about equilibria, long-time behaviour, and stability can be studied through the discrete scheme rather than only through the PDE.","The scheme is a genuine fractional analogue of the JKO scheme and reduces to it when $\\alpha\\to 1$, giving a unified time-discretization family for subdiffusive and classical diffusion.","Because each step only costs one Wasserstein minimization plus evaluation of a convex combination of past densities, the memory of the fractional derivative is captured at essentially the computational price of the classical JKO scheme.","The numerical experiments indicate sublinear convergence in $L^1$, $L^2$, and Wasserstein distance, slower than the classical first-order rate, which suggests a separate convergence-rate theory is needed for the fractional case.","Convergence is in the weak $L^1$ sense, so further regularity of the limit would be needed to upgrade it to strong convergence or pointwise statements."],"supporting_citations":[{"why":"Supplies the JKO variational formulation, existence and uniqueness of minimizers, and the proof strategy of passing to the limit in the Euler-Lagrange equation.","marker":"[26]"},{"why":"Provides the L1 approximation of the Caputo derivative, its weights, and the error estimates used to build the scheme and to justify the limit passage.","marker":"[34]"},{"why":"Imports existence and uniqueness of weak solutions for the equivalent reformulation (2.7), which the paper uses to call the limit 'the unique weak solution'.","marker":"[10]"},{"why":"Gives the discrete fractional Gronwall inequality used to bound the second moments and energies of the approximating densities.","marker":"[33]"},{"why":"Supplies the Dunford-Pettis compactness criterion used to extract a weakly convergent subsequence in $L^1$.","marker":"[41]"},{"why":"Identifies the open problem of a JKO-type gradient flow for (1.1), motivating the paper and providing the long-time behaviour context.","marker":"[27]"}],"fun_headline_variants":["JKO scheme converges for time-fractional Fokker-Planck","Memory-laden gradient flow: convergent JKO for fractional FPE","Fractional Fokker-Planck tamed by convergent variational scheme","Caputo memory meets Wasserstein: a convergent JKO scheme","Gradient flow with memory: JKO converges for fractional FPE"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the weak solution of (1.1) in the sense of Definition 2.1 is unique, a fact imported from a formally equivalent reformulation (2.7) for which existence and uniqueness are known; if the two weak formulations do not coincide, the convergence theorem identifies the limit only up to that equivalence, and the assertion of convergence to 'the' unique solution has no independent basis.","fun_headline_variants_meta":{"raw":{"variants":["JKO scheme converges for time-fractional Fokker-Planck","Memory-laden gradient flow: convergent JKO for fractional FPE","Fractional Fokker-Planck tamed by convergent variational scheme","Caputo memory meets Wasserstein: a convergent JKO scheme","Gradient flow with memory: JKO converges for fractional FPE"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000784,"raw_usage":{"total_tokens":3435,"prompt_tokens":897,"completion_tokens":2538,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":513,"completion_tokens_details":{"reasoning_tokens":2445}},"tokens_in":513,"tokens_out":2538,"duration_ms":15901,"temperature":1.0,"reasoning_tokens":2445,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:26:05.455389+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Concretely, exhibit two weak solutions of (1.1) in the sense of Definition 2.1 with the same initial datum, or verify directly whether the right-sided L1 approximant $D^\\alpha_\\tau \\phi$ converges to $tD^\\alpha_T \\phi$ as $\\tau\\to 0$; failure of either check would break the identification of the limit as the unique weak solution.","supporting_citations":[{"cited_title":"Jordan, D","cited_arxiv_id":null,"evidence_quote":"Supplies the JKO variational formulation, existence and uniqueness of minimizers, and the proof strategy of passing to the limit in the Euler-Lagrange equation."},{"cited_title":"Lin and C","cited_arxiv_id":null,"evidence_quote":"Provides the L1 approximation of the Caputo derivative, its weights, and the error estimates used to build the scheme and to justify the limit passage."},{"cited_title":"Camilli and R","cited_arxiv_id":null,"evidence_quote":"Imports existence and uniqueness of weak solutions for the equivalent reformulation (2.7), which the paper uses to call the limit 'the unique weak solution'."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the discrete fractional Gronwall inequality used to bound the second moments and energies of the approximating densities."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the Dunford-Pettis compactness criterion used to extract a weakly convergent subsequence in $L^1$."},{"cited_title":"Kemppainen and R","cited_arxiv_id":null,"evidence_quote":"Identifies the open problem of a JKO-type gradient flow for (1.1), motivating the paper and providing the long-time behaviour context."}],"review_version":1}