{"id":"a7a21499-13cb-4735-b698-deee8fb13bb9","arxiv_id":"2502.06500","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"The infinite-volume free energy for lattice spins has a gradient flow that coincides with the law of the Langevin dynamics, with uniqueness and exponential convergence under curvature and temperature conditions.","lead":"This paper constructs the gradient flow of the infinite-volume free energy for lattice spin systems and proves it coincides with the law of the interaction Langevin dynamics. It also proves exponential convergence to the unique minimizer when the spin manifold has positive Ricci curvature and the temperature is high enough.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"EVI constant in Theorem 3.7 uses ||Ψ||_L∞ where the proof requires ||∇²Ψ||_∞; Lemma 2.13 supplies only the latter, so the stated EVI and exponential-rate conclusions are not proven.","rationale":"The reader identified the same mismatch between the EVI constant in Eq. (3.12) and the displacement-convexity estimate in Lemma 2.13. My independent check confirms that the proof of Lemma 3.11 requires a lower bound on β∇²H_n, which can only come from ||∇²Ψ||_∞, not ||Ψ||_{L∞}. A concrete high-frequency example on the circle shows that ||Ψ||_{L∞} can be arbitrarily small while ||∇²Ψ||_∞ remains of order 1, so no hidden inequality repairs the stated constant. This is a genuine gap in the proof of Theorem 3.7, and because Corollaries 3.8 and 3.9 directly invoke that theorem, the paper as written is not fully sound. However, the main claim that the gradient-flow trajectories coincide with the Langevin dynamics depends on uniqueness, which would still hold if the EVI constant were corrected to κ − 2β||J||_{ℓ1}||∇²Ψ||_∞; only the positivity threshold and explicit rate in Corollary 3.9 would change. Thus the appropriate verdict remains CONDITIONAL, matching the reader's assessment, and no verdict adjustment is needed.","tokens_in":59045,"tokens_out":5215,"duration_ms":60214,"concrete_test":"Re-derive Step 2 of Lemma 3.11 using Lemma 2.13(2), replacing Eq. (3.12) by Kβ = κ − 2β||J||_{ℓ1}||∇²Ψ||_∞, and check that the rest of the EVI proof goes through. Separately, instantiate M = S¹, U = 0, κ = 0, J nearest-neighbour, β = 1, and Ψ_k(x,y) = k^{-2} cos(k(x−y)) for large k. Compute the least eigenvalue of β∇²H_n on Λ_n and compare it with the two candidate constants: the proof's required bound is −2β||J||_{ℓ1}, whereas Eq. (3.12) would claim only −2β||J||_{ℓ1} k^{-2}. The failure of the latter bound for large k demonstrates that the discrepancy is real and not a notational artifact.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Theorem 3.7 states Kβ := κ − 2β||J||_{ℓ1}||Ψ||_{L∞} in Eq. (3.12), and Lemma 3.11 applies it as the lower bound Ric_n + ∇²U_n + β∇²H_n ≥ Kβ before invoking [CEMS06, Prop. 4.2]. However, the only displacement-convexity estimate for the interaction term proven in the paper is Lemma 2.13(2): H_n is −2||J||_{ℓ1}||∇²Ψ||_∞-displacement convex. The L∞ norm of Ψ does not control the Hessian of Ψ: on M = S¹, take Ψ_k(x,y) = k^{-2} cos(k(x−y)); then ||Ψ_k||_{L∞} = k^{-2} while ||∇²Ψ_k||_{L∞} = 1. Thus the curvature condition used in the EVI proof is not a consequence of the stated assumptions unless the constant is corrected to ||∇²Ψ||_∞. Because this constant feeds into Corollary 3.8 (uniqueness of strong solutions) and Corollary 3.9 (exponential contraction and convergence rate), those statements as written rely on an unproven inequality. The identification of JKO limits with Langevin laws may survive after the correction, but Theorem 3.7 in its displayed form is unsupported.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a gradient-flow framework for the infinite-volume free energy of lattice spin systems with continuous spins on a compact Riemannian manifold, under a translation-invariant, symmetric, short-range (ℓ1-summable) interaction. The authors construct the flow through a JKO-type minimizing movement scheme that alternates solving a finite-volume variational problem and a stationarization step, and they construct the infinite-volume overdamped Langevin diffusion via a weighted ℓ2 embedding and finite-volume approximations. The central result is that both the gradient-flow trajectory and the law of the diffusion solve the same infinite-volume Fokker–Planck–Kolmogorov hierarchy in a dual sense; a regularity theorem upgrades weak/dual solutions to strong solutions, and an Evolution Variational Inequality is used to prove uniqueness and, under positive Bakry–Émery curvature and sufficiently high temperature, exponential convergence to the unique minimizer in specific Wasserstein distance and free energy.","tokens_in":59288,"tokens_out":5443,"duration_ms":48122,"significance":"If the results stand, this is a substantial contribution. It appears to be the first construction of the gradient flow of an infinite-volume free energy with genuinely interacting spins, and it provides a new route to uniqueness for infinite-volume Fokker–Planck–Kolmogorov equations, yielding convergence in a stronger metric (specific Wasserstein distance) than the dual-norm convergence available from earlier stochastic-analysis approaches. The paper is highly detailed and essentially self-contained, with careful finite-volume approximations, stationarization estimates, heat-kernel bounds, and honest discussion of limitations, including the obstacles to the non-compact extension in Section 6. The main conceptual architecture is sound; the issues found are local and, in my assessment, correctable.","major_comments":[{"comment":"The stated EVI constant Kβ := κ − 2β||J||_{ℓ1}||Ψ||_{L∞} is not supported by the proof. In the proof of Lemma 3.11 the authors invoke the curvature condition Ric_n + ∇²U_n + β∇²H_n ≥ Kβ with Kβ as in (3.12), but the only displacement-convexity estimate for H_n established in the paper is Lemma 2.13(2), which gives that H_n is −2||J||_{ℓ1}||∇²Ψ||_∞-displacement convex. The L∞ norm of Ψ does not control the Hessian norm of Ψ: on M = S¹, Ψ_k(x,y) = k^{-2}cos(k(x−y)) has ||Ψ_k||_{L∞}=k^{-2} while ||∇²Ψ_k||_{L∞}=1. Consequently, the condition needed to apply [CEMS06, Prop. 4.2] holds only with Kβ = κ − 2β||J||_{ℓ1}||∇²Ψ||_∞. As written, Theorem 3.7 and its consequences Corollary 3.8 and Corollary 3.9 rest on an unproven inequality.","section":"Theorem 3.7, Eq. (3.12) and Lemma 3.11"},{"comment":"There is an internal inconsistency in the high-temperature regime used for exponential convergence. Theorem 2.14 proves uniqueness of the minimizer under the condition β < (1/(2κ))(||J||_{ℓ1}||∇²Ψ||_∞)^{-1}, while Corollary 3.9 claims exponential convergence whenever Kβ > 0 with Kβ from (3.12), i.e. for β < κ/(2||J||_{ℓ1}||Ψ||_∞). Since ||∇²Ψ||_∞ can be much larger than ||Ψ||_∞, the corollary asserts convergence in a parameter regime for which the paper's own uniqueness result does not apply. Once the constant in Theorem 3.7 is corrected to use ||∇²Ψ||_∞, the threshold in Corollary 3.9 will match Theorem 2.14; as displayed, the statements are inconsistent.","section":"Corollary 3.9 versus Theorem 2.14"}],"minor_comments":[{"comment":"The sentence 'we divide by |Λ_n| in (3.11)' should refer to inequality (3.16), not to item (3.11) of Proposition 3.6.","section":"Proof of Theorem 3.7, Step 3"},{"comment":"There is a LaTeX artifact in the sentence introducing H(x): the string 'H(x) ??:=' should be replaced by the intended definition.","section":"Section 1.3"},{"comment":"The constant Kβ is referred to as 'as in (3.12)' before Theorem 3.7 has been stated; stating the corrected constant explicitly in Lemma 3.11 would improve readability and prevent the kind of mismatch identified in the major comments.","section":"Lemma 3.11"}],"recommendation":"major_revision","confidential_remarks":"The EVI constant issue appears to be a genuine but local error: the proof of Lemma 3.11 already contains the correct mechanism, and Lemma 2.13 gives the correct constant involving ||∇²Ψ||_∞. I recommend major revision rather than rejection, with the expectation that the authors can correct the statement of Theorem 3.7 and propagate the corrected constant through Corollaries 3.8 and 3.9, while re-checking any further places where the same mismatch may appear. The paper is otherwise impressive in scope and execution, and the main claims are likely to survive this correction."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Here is my read. The paper delivers the first rigorous JKO-style gradient flow for the infinite-volume free energy of continuous lattice spins (short-range, not necessarily finite-range), and it proves the flow agrees with the overdamped Langevin dynamics through the infinite-volume FPK hierarchy. That is a real advance. The JKO scheme with the stationarization/compatibilization step is new, the regularity theorem for weak/dual solutions is substantial, and the EVI route to uniqueness is well chosen. The paper is also honest about what it does not do (non-compact case, k-body interactions), and the literature review is fair; the reliance on earlier work is mostly inspirational, not load-bearing.\n\nThe soft spot flagged in the stress-test is real. Theorem 3.7 gives Kβ = κ − 2β‖J‖ℓ1‖Ψ‖L∞, but the proof of Lemma 3.11 needs the curvature bound Ric_n + ∇²U_n + β∇²H_n ≥ Kβ. The only displacement-convexity estimate supplied for H_n is Lemma 2.13(2): H_n is −2‖J‖ℓ1‖∇²Ψ‖∞-displacement convex. The L∞ norm of Ψ does not control its Hessian (the cosine example on S¹ is fine), so the displayed EVI, and with it Corollaries 3.8 and 3.9 as written, rest on an unproven inequality. This is not a dealbreaker: Theorem 2.14 already uses the correct constant with ‖∇²Ψ‖∞, and the proof structure repairs itself if Kβ is changed accordingly. But the authors need to make that correction; as submitted, the exponential rate in Corollary 3.9 is unsupported. I also note the proof of Lemma 3.11 says the inequality holds \"with Kβ as in (3.12)\" without giving the Hessian computation; a short argument should be added, not just a changed display.\n\nEverything else I checked is consistent with the stated architecture. I did not machine-check the estimates, but the bootstrap is detailed and the strategy is coherent. This is a paper for mathematical statisticians, statistical mechanics readers, and optimal-transport workers; it deserves a serious referee, and I expect acceptance after the repair.","headline":"A credible first construction of the infinite-volume gradient flow for interacting lattice spins, but Theorem 3.7 states the EVI constant with the wrong norm and needs a one-line repair before the exponential-rate results are proven.","tokens_in":59854,"tokens_out":2592,"would_cite":true,"duration_ms":26535,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K35","82B20","49Q22","60J60","58J65"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that, for infinite lattice systems of continuous spins on a compact manifold, the gradient flow of the infinite-volume free energy and the law of the spins under the overdamped Langevin dynamics coincide, because both…","keywords":["Wasserstein gradient flow","infinite-volume free energy","lattice spin systems","overdamped Langevin dynamics","Fokker-Planck-Kolmogorov equations","Evolution Variational Inequality","displacement convexity","specific Wasserstein distance"],"falsifier":"For a constant interaction potential $\\Psi \\equiv 1$ (so $\\nabla^2\\Psi \\equiv 0$), the interaction exerts no force and the flow reduces to independent diffusions with Bakry-Émery constant $\\kappa$, while the stated $K_\\beta$ would be $\\kappa - 2\\beta\\|J\\|_{\\ell^1}$; checking whether the EVI proof's curvature inequality $\\mathrm{Ric} + \\nabla^2 U + \\beta\\nabla^2 H \\ge K_\\beta\\, g$ holds in this example would settle whether the stated constant follows from the proof.","tokens_in":58786,"feed_emoji":"🧲","tokens_out":15273,"duration_ms":116699,"temperature":0.7,"pith_summary":"The central claim is that two a priori different descriptions of how a lattice spin system relaxes toward equilibrium — the Wasserstein gradient flow of the infinite-volume free energy, and the law of the spins under the infinite-volume overdamped Langevin dynamics — are in fact the same trajectory. The identification is achieved by showing that both objects satisfy the same hierarchy of coupled parabolic equations, the infinite-volume Fokker-Planck-Kolmogorov equations, that weak solutions of this hierarchy automatically have smooth finite-box densities, and that strong solutions are unique via an Evolution Variational Inequality. A second group of results controls the long-time behaviour: the free energy is non-increasing along the flow, and when the spin space has uniformly positive Ricci curvature and the temperature is high enough, the flow converges exponentially fast to the unique minimizer of the free energy in specific Wasserstein distance and in free energy. The relevance is that an interacting spin system in contact with a heat bath is thereby shown to be a steepest descent of its own free energy, a relation previously established only in finite volume.","feed_headline":"Infinite-volume free-energy flow equals Langevin dynamics","feed_subtitle":"Both trajectories solve the same infinite-volume Fokker-Planck equations, so they coincide.","key_machinery":"The argument is carried by the infinite-volume Fokker-Planck-Kolmogorov hierarchy — a family of coupled parabolic PDEs, one for each finite box $\\Lambda$, with drifts depending on conditional expectations of the full infinite-volume interaction — together with the specific Wasserstein distance and the free energy $F^\\beta$ built from specific relative entropy and interaction-energy density. Two analytic pillars support the main theorem: a regularity result stating that every weak solution of the hierarchy has $C^{1,2}$ local densities (proved by a bootstrap on Duhamel's principle with heat-kernel estimates on the compact manifold), and an Evolution Variational Inequality for $F^\\beta$ with respect to $W$ (proved by taking finite-volume EVIs to the limit), which yields uniqueness. The gradient flow is constructed by a discrete minimizing-movement scheme in the Wasserstein space with a stationarization step that converts finite-box minimizers into translation-invariant measures; the diffusion is constructed by embedding the configuration space into a weighted $\\ell^2$ Hilbert space in which the interaction gradient is Lipschitz, following classical infinite-volume SDE methods.","core_discovery":"On the space of translation-invariant spin measures, the paper defines the infinite-volume free energy $F^\\beta(P) = E(P) + \\beta H(P)$, with $E$ the specific relative entropy and $H$ the interaction-energy density, together with the specific Wasserstein distance $W(P,Q)=\\lim_n |\\Lambda_n|^{-1} W_n^2(P,Q)$. It constructs two evolutions: the gradient flow of $F^\\beta$ obtained as the limit of a discrete variational (JKO-type) scheme with a stationarization step, and the law of the infinite-volume overdamped Langevin dynamics obtained through a weighted Hilbert-space embedding. The paper then proves that both curves satisfy the same infinite-volume Fokker-Planck-Kolmogorov hierarchy in the dual sense; that weak solutions automatically regularize into strong solutions with smooth finite-box densities; and that strong solutions are unique because they satisfy an Evolution Variational Inequality with respect to $W$. Consequently, the trajectories of the gradient flow coincide with those obtained from the Langevin dynamics. Under a positive Bakry-Émery curvature bound on the spin space and for $\\beta$ small enough, the free energy has a unique minimizer and the flow converges to it exponentially in $W$ and in free energy.","pith_inferences":["Correcting the EVI constant to the Hessian-based value $\\kappa - 2\\beta\\|J\\|_{\\ell^1}\\|\\nabla^2\\Psi\\|_\\infty$ that the proof's estimates support would yield faster exponential rates for interactions with small Hessian; comparing the two constants on concrete models would test the sharpness of the paper's stated rate.","The stationarize-then-limit scheme should transfer to other infinite-volume variational problems lacking a product structure, such as random fields or point-process free energies, where similar coincidence theorems between variational flows and Markov dynamics could be derived.","The regularity bootstrap, which avoids classical potential-theoretic $L^p$–$L^q$ estimates, may apply to other nonlocal parabolic hierarchies (for instance mean-field Fokker-Planck systems with conditional drifts) where standard potential estimates are unavailable.","The equality of trajectories suggests a variational numerical scheme for spin dynamics: each minimizing-movement step is a finite-box optimal-transport problem and the stationarization step controls finite-size errors, offering a transport-based alternative to discretizing the stochastic differential equation."],"forward_implications":["The law of an infinite spin system under thermal agitation is exactly the Wasserstein steepest descent of its free energy, so variational tools such as displacement convexity and EVIs describe the physical relaxation process without finite-volume approximation.","Exponential convergence at high temperature holds simultaneously in specific Wasserstein distance and in free energy, a uniformity over all local observables that is stronger than earlier weak-dual convergence statements for the stochastic Heisenberg model.","Any weak solution of the infinite-volume Fokker-Planck-Kolmogorov equations is automatically a strong solution, so the distinction between the dual and strong formulations disappears for this system.","The variational scheme with stationarization provides an existence proof for infinite-volume Fokker-Planck equations that does not require constructing the underlying stochastic process first.","Trajectories started from different stationary initial measures contract in specific Wasserstein distance with rate $e^{K_\\beta t}$, giving quantitative stability of the dynamics with respect to initial data."],"supporting_citations":[{"why":"Introduces the variational (minimizing movement) scheme in Wasserstein space and the bootstrap regularity strategy that the paper adapts to infinite volume.","marker":"[JKO98]"},{"why":"Provides the displacement-convexity inequality on manifolds used to derive the EVI and the high-temperature uniqueness of the minimizer.","marker":"[CEMS06]"},{"why":"Constructs infinite-volume diffusion on the infinite torus by finite-box approximation and the martingale method; the model for Theorem 5.5.","marker":"[HS81]"},{"why":"Supplies the weighted Hilbert-space technique showing the interaction gradient is Lipschitz, the key device in the diffusion construction.","marker":"[LR85]"},{"why":"The Brenier-McCann optimal transport theorem on compact manifolds, used for Kantorovich potentials and Wasserstein geodesics in the variational and EVI arguments.","marker":"[McC01]"},{"why":"Shows the heat flow on a manifold is an EVI gradient flow with a constant tied to Ricci curvature, the template for the infinite-volume EVI and exponential convergence.","marker":"[Erb10]"},{"why":"Earlier exponential convergence to equilibrium for the stochastic Heisenberg model in a weak dual norm; the paper's stronger specific-Wasserstein statement generalizes it.","marker":"[Wic81]"},{"why":"Develops the specific-Wasserstein gradient-flow framework for stationary point processes, the closest prior construction that the spin-system version adapts.","marker":"[EHJM23]"},{"why":"Provides the classical definitions of specific relative entropy and interaction-energy density as limits over boxes, and the Gibbs variational principle used for comparison.","marker":"[FV18]"}],"fun_headline_variants":["Gradient flow matches Langevin dynamics in spin lattices","Infinite-volume spin free energy flow equals diffusion law","Spin systems: free energy flow and Langevin paths unify","High-temp spin systems: exponential free energy decay","Gradient flow and Langevin law share same Fokker-Planck"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the free energy satisfies an Evolution Variational Inequality with the stated constant $K_\\beta = \\kappa - 2\\beta\\|J\\|_{\\ell^1}\\|\\Psi\\|_{L^\\infty}$; the proof as written actually requires the displacement-convexity constant with $\\|\\nabla^2\\Psi\\|_\\infty$, so the stated EVI and the exponential rate rest on an inequality the paper does not establish.","fun_headline_variants_meta":{"raw":{"variants":["Gradient flow matches Langevin dynamics in spin lattices","Infinite-volume spin free energy flow equals diffusion law","Spin systems: free energy flow and Langevin paths unify","High-temp spin systems: exponential free energy decay","Gradient flow and Langevin law share same Fokker-Planck"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000665,"raw_usage":{"total_tokens":3082,"prompt_tokens":1039,"completion_tokens":2043,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":655,"completion_tokens_details":{"reasoning_tokens":1959}},"tokens_in":655,"tokens_out":2043,"duration_ms":12984,"temperature":1.0,"reasoning_tokens":1959,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T15:14:57.065273+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For a constant interaction potential $\\Psi \\equiv 1$ (so $\\nabla^2\\Psi \\equiv 0$), the interaction exerts no force and the flow reduces to independent diffusions with Bakry-Émery constant $\\kappa$, while the stated $K_\\beta$ would be $\\kappa - 2\\beta\\|J\\|_{\\ell^1}$; checking whether the EVI proof's curvature inequality $\\mathrm{Ric} + \\nabla^2 U + \\beta\\nabla^2 H \\ge K_\\beta\\, g$ holds in this example would settle whether the stated constant follows from the proof.","supporting_citations":[],"review_version":1}