{"id":"e04ba9b2-1c0e-43de-ac29-f49558206ec9","arxiv_id":"2607.25206","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For ε-periodic Fokker-Planck equations, the weighted-L2 minimizing-movement scheme commutes with homogenization and gives the effective equation, while the JKO scheme converges to a different limiting equation.","lead":"This paper proves that two time-discrete numerical schemes for a Fokker-Planck equation with rapidly oscillating coefficients behave differently in the homogenization limit: one preserves the correct effective equation, the other does not. It matters because it identifies which discretization is safe for multiscale gradient-flow simulations.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 1.8's JKO limit is conditional on an unproved W^{1,1} compactness assumption and on uniform strong convexity of Ahom; without these, the abstract's 'only weighted L2 recovers correct dynamics' is not established.","rationale":"The reader's weakest-assumption analysis correctly isolates the same load-bearing concern: Theorem 1.8 is the only route to the JKO limiting equation (1.23), and its proof depends on two unproved ingredients—W^{1,1} compactness and uniform strong convexity of Ahom. The paper itself flags the first in Remark 1.9, and the second is an additional assumption not guaranteed by the standing hypotheses. Without these, the abstract's claim that only the weighted-L2 scheme recovers the correct effective dynamics is not fully established. The weighted-L2 chain (Theorems 1.4 and 1.7) is supported by Γ-convergence and H1 estimates, but the JKO chain is conditional. This does not contradict the likely truth of the qualitative message; it does mean the conditional verdict is appropriate. I found no independent objection beyond the reader's, and I do not recommend changing the conditional verdict.","tokens_in":30406,"tokens_out":23241,"duration_ms":225296,"concrete_test":"Attempt to prove the missing compactness claim in Theorem 1.8: from the optimality condition in Theorem 4.14 and the a priori bound, show that the homogenized JKO trajectories are relatively compact in W^{1,1}, e.g. by establishing a uniform L1 equicontinuity estimate for the gradients, ∥∇µτ(·+h)−∇µτ(·)∥_{L1} → 0 uniformly in τ. Independently, check whether Ahom is uniformly strongly convex for a natural admissible A, such as a 2D piecewise-constant checkerboard conductivity; if Ahom has flat directions, Theorem 1.8 does not apply and the abstract's unconditional claim must be weakened.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The decisive step is in the proof of Theorem 1.8: to pass τ→0 inside the nonlinear coefficient (1/2∇Ahom)^{-1}, the authors use the Lipschitz bound (4.20), which requires both uniform strong convexity of Ahom and the assumed W^{1,1} convergence µτ→µ. Remark 1.9 explicitly concedes that establishing such compactness 'lies beyond the scope of this work.' A uniform W^{1,1} bound is not enough: W^{1,1} is not reflexive, and compactness in W^{1,1} needs an equicontinuity/tightness argument for the gradients, which is not supplied. Moreover, uniform strong convexity of Ahom is not a consequence of Assumption 1.1 for general d; homogenized Finsler metrics can develop flat directions, so (1/2∇Ahom)^{-1} may be multivalued and (4.20) fails. Thus the identification of the τ→0 limit with (1.23) is conditional. Because the abstract states unconditionally that 'only the scheme based on the weighted L2 metric recovers the correct effective dynamics,' the central comparative claim rests on this unproved hypothesis. The weighted-L2 half is well supported; the JKO half is not.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies homogenization limits of two time-discrete minimizing movement schemes for Fokker–Planck equations on the flat torus with rapidly oscillating diffusion A(x/ε). The first scheme uses a weighted L2 metric (backward Euler form); the second is an ε-JKO scheme with a Wasserstein-type distance induced by A. For fixed τ, Theorem 1.4 homogenizes the weighted L2 scheme to a scheme using the effective diffusion A*, while Theorem 1.5 homogenizes the ε-JKO scheme to a scheme using the Gromov–Hausdorff limiting cost WGH and the associated Finsler action Ahom. The paper then identifies τ→0 limits: Theorem 1.7 states the homogenized weighted L2 scheme converges to the effective Fokker–Planck equation (1.2); Theorem 1.8 states that, conditional on uniform strong convexity of Ahom and an unproved W^{1,1} compactness assumption, the homogenized JKO scheme converges to the quasilinear equation (1.23) with coefficient (1/2∇Ahom)^{-1}. The paper concludes that only the weighted L2 scheme is asymptotic-preserving.","tokens_in":30703,"tokens_out":8895,"duration_ms":80548,"significance":"The question addressed is timely and important: whether variational time discretizations preserve homogenization limits is directly relevant to asymptotic-preserving numerical schemes and to the theory of gradient flows in inhomogeneous media. The weighted-L2 half of the paper is clean: the Γ-convergence of the Dirichlet energy and the convergence of the weighted L2 distance are proved in detail, and the H1 compactness argument is sound. The homogenization of the ε-JKO scheme for fixed τ is also a worthwhile contribution, and the numerical experiment in 1D illustrates the claimed discrepancy. The paper makes good use of independent published results ([20], [9], [41]) rather than fitting parameters, and the conditional theorem is stated honestly. However, the central comparative claim — that only the weighted L2 scheme recovers the correct effective dynamics — rests on Theorem 1.8, whose hypotheses are not established in the paper. The abstract and Section 1.5 present this claim unconditionally, which is not supported by the proofs as written.","major_comments":[{"comment":"The load-bearing conclusion is conditional. Passing τ→0 inside (1/2∇Ahom)^{-1} in (4.25) requires both uniform strong convexity of Ahom and the assumed W^{1,1}(Td) convergence µτ→µ; the latter is explicitly left unproved in Remark 1.9 ('lies beyond the scope of this work'). A uniform W^{1,1} bound, which is all that §4.2.2 establishes, does not imply W^{1,1} compactness because W^{1,1} is not reflexive and no equicontinuity of gradients is supplied. Moreover, uniform strong convexity of Ahom is not a consequence of Assumption 1.1 in d>1; homogenized Finsler action functions can develop flat directions. Therefore the abstract's unconditional statement that 'only the scheme based on the weighted L2 metric recovers the correct effective dynamics' is not established. The theorem should either be proven under additional verifiable hypotheses, or the claims and abstract should be rephrased as","section":"Theorem 1.8 / Remark 1.9 / §4.2.2"},{"comment":"The proof of convergence of Wasserstein distances uses pointwise convergence of costs cε(x,y)→chom(x,y) and then asserts: 'Since Td×Td is compact, ε(δ) can be made independent of (x,y)'. This is not valid in general: pointwise convergence of continuous functions on a compact set does not imply uniform convergence. The lemma needs an equicontinuity/Dini-type argument for the family (cε) or an alternative stability theorem for optimal transport under Γ-convergent costs. Without this, the proof of Theorem 1.5 has a gap, although the statement is likely repairable.","section":"Lemma 4.5 / §4.1.2"},{"comment":"The weighted-L2 half of the paper's central claim depends on the τ→0 convergence of the backward Euler scheme. This is stated in Proposition 3.2 'without proof', and the proof of Theorem 1.7 is deferred to 'the same argument'. Since this is the half that is claimed to be correct and asymptotic-preserving, a proof or a precise reference with matching hypotheses should be included. The standard numerical-analysis references may not cover the exact weak formulation used here (with Vε and the H1 setting); a short proof would remove the asymmetry in rigor between the two halves.","section":"Theorem 1.7 / Proposition 3.2"}],"minor_comments":[{"comment":"Typo: 'weigthed L2 scheme' should be 'weighted L2 scheme'.","section":"Figure 1"},{"comment":"Remark 1.9 lists [26, 36, 14] as stronger convergence results for JKO schemes; these references are for standard Fokker–Planck or related equations and do not cover the homogenized Finsler metric considered here. Please clarify that the cited results do not imply the required W^{1,1} compactness for the present WGH metric.","section":"Remark 1.9"},{"comment":"In the displayed inequality before (4.16), the factor τ is written as 'τ × Cε(φ)/(2τ)' which simplifies to Cε/2; the final bound should make explicit that the constant depends on φ (and is O(τ) after using Lemma 4.9). The current display is dimensionally confusing.","section":"Lemma 4.10 / (4.16)"},{"comment":"Minor typos: 'Boreal sets' should be 'Borel sets'; 'Monge-Kantrovich' should be 'Monge–Kantorovich'; 'trajecotry' in Section 4.1; 'the prove' in Section 3.2 should be 'the proof'.","section":"General"}],"recommendation":"major_revision","confidential_remarks":"The paper contains a substantial and mostly well-executed analysis of the weighted-L2 scheme and of the fixed-τ homogenization of the ε-JKO scheme. The main problem is that the headline conclusion is stated more strongly than the theorems support. The authors should either prove the missing compactness/strong-convexity hypotheses (or at least verify them in the 1D case that is numerically shown) or explicitly frame the comparison as conditional on those hypotheses. If the conditional framing is adopted, the paper would be a solid contribution to the literature on variational schemes and homogenization; in its current form, the abstract overstates the result."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing to know: the paper has a solid half and a conditional half. The solid half is the weighted-L2 scheme: fixed-τ homogenization via Γ-convergence and H1 compactness, then τ→0 by backward-Euler convergence. That chain is in good shape. The conditional half is the JKO story: Theorem 1.8's limiting equation (1.23) rests on W1,1 compactness the authors explicitly say they don't prove, and on uniform strong convexity of Ahom that Assumption 1.1 doesn't guarantee. The abstract's \"only the weighted L2 recovers correct effective dynamics\" is stronger than what is proved.\n\nWhat's genuinely new: earlier work identified the continuous Wasserstein gradient-flow homogenization and the metric limit Wε→WGH. This paper asks what happens at the time-discrete level and shows the answer depends on the metric. The weighted-L2 chain is rigorous and clean, and the comparison between the two schemes is a real contribution. The 1D computation showing the JKO coefficient differs from A* is nicely concrete.\n\nSoft spots, in proportion: Lemma 4.5 jumps from pointwise cost convergence to uniform control of the integral; pointwise convergence on a compact set is not enough without an equicontinuity argument. Likely fixable, but as written it's a gap. The bigger issue is Theorem 1.8. A uniform W1,1 bound alone does not give compactness in W1,1, and the paper concedes in Remark 1.9 that the needed compactness is beyond its scope. Uniform strong convexity of Ahom is also an assumption, not a consequence of the stated hypotheses. So the advertised failure of JKO is established only conditionally. The qualitative message is credible and probably true—especially in 1D—but the abstract should not state it as an unconditional theorem.\n\nThe citation pattern looks honest: the targets come from independent published work, no parameters are fitted, and the paper doesn't disguise what it inherits from [20]. No code or data is shipped; the numerical figure is illustrative.\n\nWho is this for: applied analysts and numerical analysts working on asymptotic-preserving schemes for Fokker-Planck equations and on discrete gradient flows. It deserves a serious referee. I'd send it out, but with the expectation of major revision: either close the W1,1 compactness gap or rewrite the main claims so the conditional status is explicit. The weighted-L2 half can stand; the JKO half needs its caveats in the title or abstract.","headline":"The weighted-L2 half is solid and worth knowing; the JKO half is a credible but conditional story that the abstract overstates.","tokens_in":31204,"tokens_out":3435,"would_cite":true,"duration_ms":36705,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35A15","35B27","35Q84","37M15","46N40","49Q22"],"pacs":[],"model":"deepseek-v4-flash","headline":"For Fokker–Planck equations with rapidly oscillating diffusion, the choice of metric in a time-discrete gradient flow determines whether homogenization and the time-step limit commute: the weighted L2 metric recovers the effective dynamics","keywords":["homogenization","Fokker-Planck equation","gradient flows","minimizing movements","JKO scheme","weighted L2 metric","effective diffusion","Gamma-convergence"],"falsifier":"Run the 1D numerical example from the paper (A(x) = (1 + ½ sin 2πx)², ε = 10⁻⁵, small τ) and measure the effective diffusion coefficient of the homogenized JKO solutions as τ→0. If the spreading follows A* = (∫ A^{-1})^{-1} ≈ 0.65 rather than (∫ A^{-1/2})^{-2} ≈ 0.75, the central claim fails; conversely, reproducing ≈ 0.75 confirms it. A proof or counterexample of the W^{1,1} compactness of the homogenized JKO trajectories would directly settle the gap left open in the paper.","tokens_in":30234,"feed_emoji":"🔄","tokens_out":11262,"duration_ms":93759,"temperature":0.7,"pith_summary":"This paper asks whether natural time-discrete approximations of a Fokker–Planck equation with rapidly oscillating diffusion survive the homogenization limit ε→0. Two candidate schemes are compared: a minimizing-movement (JKO) scheme built on a transport-type metric, and a scheme built on a weighted L2 metric. The paper proves that the choice matters: homogenizing the weighted L2 scheme and then sending the time step τ to zero recovers the classical effective Fokker–Planck equation with coefficient A*, whereas the homogenized JKO scheme converges to a quasilinear equation with a different coefficient, equal in one dimension to (∫ A^{-1/2})^{-2} instead of A* = (∫ A^{-1})^{-1}. Therefore, among the two schemes studied, only the weighted L2 metric is asymptotic-preserving. If correct, this is a caution for the widespread use of JKO-type variational discretizations for multiscale gradient-flow problems, and it isolates the cause: the two limits ε→0 and τ→0 do not commute.","feed_headline":"Only a weighted L2 time step survives Fokker–Planck homogenization","feed_subtitle":"The standard JKO transport scheme converges to the wrong effective diffusion when ε→0 and τ→0.","key_machinery":"The load-bearing objects are the two metrics and their homogenized limits. For the weighted L2 scheme, the distance Kε(μ,ν) = (∫(μ−ν)²/πε dx)^{1/2} and the Dirichlet energy Qε homogenize to their counterparts with the classical elliptic-corrector coefficient A* = ∫(A + A∇χ). For the JKO scheme, the transport cost cε(x,y) = min∫⟨A(z/ε)^{-1}ż, ż⟩dt homogenizes (as a Lagrangian) to A_hom; the limiting equation's diffusion coefficient is then (1/2∇A_hom)^{-1}, obtained by Legendre duality from the homogenized action. The mismatch between A* and (1/2∇A_hom)^{-1} is the quantitative expression of the non-commutativity.","core_discovery":"For the Fokker–Planck equation ∂ₜμ − ∇·(A(x/ε)(∇μ + μ∇V)) = 0 on the flat torus with 1-periodic A, the paper proves (for fixed τ) Γ-convergence of both time-discrete schemes: the weighted L2 scheme homogenizes to a backward-Euler scheme for the effective equation with coefficient A* (Theorem 1.4), while the ε-JKO scheme homogenizes to a JKO scheme whose transport cost is generated by the homogenized action A_hom(ξ) = liminf_T (1/T)∫⟨A(tξ+φ)^{-1}(ξ+φ′), ξ+φ′⟩dt (Theorem 1.5). Sending τ→0, the homogenized weighted-L2 scheme gives the effective Fokker–Planck equation (1.2) (Theorem 1.7); the homogenized JKO scheme gives, under a W^{1,1} compactness assumption, the quasilinear equation (1.23) wi","pith_inferences":["A byproduct of the 1D formula is a cheap quantitative bench test: for any periodic A, a multiscale scheme that claims asymptotic preservation should reproduce A*, whereas a JKO-type transport scheme will produce (∫ A^{-1/2})^{-2}; for A(x) = (1 + ½ sin 2πx)² these two numbers already differ by roughly 15%.","The same mechanism — trajectory-level homogenization of the Lagrangian versus PDE-level homogenization — likely affects other Wasserstein gradient flows (e.g., aggregation–diffusion or porous-medium type) once the entropy or drift is modified, so the conclusion may extend well beyond the linear Fokker–Planck case studied here.","If the W^{1,1} compactness left open in Remark 1.9 fails for some initial data, the homogenized JKO scheme might have no well-defined τ→0 limit at all — an outcome that would strengthen rather than weaken the paper's warning that transport-step schemes need scrutiny."],"forward_implications":["The weighted L2 scheme is asymptotic-preserving: its homogenized limit at fixed τ is the backward-Euler discretization of the effective Fokker–Planck equation, and its τ→0 limit is exactly (1.2).","The homogenized JKO scheme is not asymptotic-preserving: sending τ→0 yields the quasilinear equation (1.23) with coefficient (1/2∇A_hom)^{-1}, which differs from A* in general.","In one dimension the mismatch is explicit: the JKO limit gives effective diffusivity (∫ A^{-1/2})^{-2} whereas the correct effective diffusivity is A* = (∫ A^{-1})^{-1}; the two limits ε→0 and τ→0 therefore do not commute for the JKO scheme.","The mechanism is identified as the blow-up, as ε→0, of the second derivative of the exponential map of the transport metric, which makes the JKO consistency constant diverge while the weighted L2 constant stays bounded."],"fun_headline_variants":["Weighted L2 metric wins in Fokker–Planck homogenization limit","JKO scheme fails: wrong effective diffusion after homogenization","Metric choice in time step alters homogenization of Fokker–Planck","Effective dynamics depend on time-discrete metric choice","Only weighted L2 recovers correct Fokker–Planck homogenization"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The conclusion that the homogenized JKO scheme converges to the quasilinear equation (1.23) rests on the unproved W^{1,1}(T^d) compactness of its time-discrete trajectories as τ→0 (explicitly left open in Remark 1.9) together with the assumed uniform strong convexity of the homogenized action A_hom; if either fails, the claimed limiting equation — and hence the claim that JKO fails to recover the effective dynamics — is not established.","fun_headline_variants_meta":{"raw":{"variants":["Weighted L2 metric wins in Fokker–Planck homogenization limit","JKO scheme fails: wrong effective diffusion after homogenization","Metric choice in time step alters homogenization of Fokker–Planck","Effective dynamics depend on time-discrete metric choice","Only weighted L2 recovers correct Fokker–Planck homogenization"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000643,"raw_usage":{"total_tokens":2774,"prompt_tokens":706,"completion_tokens":2068,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":450,"completion_tokens_details":{"reasoning_tokens":1976}},"tokens_in":450,"tokens_out":2068,"duration_ms":14558,"temperature":1.0,"reasoning_tokens":1976,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T03:05:41.840363+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the 1D numerical example from the paper (A(x) = (1 + ½ sin 2πx)², ε = 10⁻⁵, small τ) and measure the effective diffusion coefficient of the homogenized JKO solutions as τ→0. If the spreading follows A* = (∫ A^{-1})^{-1} ≈ 0.65 rather than (∫ A^{-1/2})^{-2} ≈ 0.75, the central claim fails; conversely, reproducing ≈ 0.75 confirms it. A proof or counterexample of the W^{1,1} compactness of the homogenized JKO trajectories would directly settle the gap left open in the paper.","supporting_citations":[],"review_version":1}