{"id":"6b720034-73ab-4fd2-b47c-f8bcebc0120f","arxiv_id":"2412.01190","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"A gradient-flow proof yields Wasserstein Jensen's inequality, barycenter well-posedness on RCD and extended metric measure spaces, and a new Barycenter-Curvature-Dimension condition with stability and geometric applications.","lead":"This mathematics paper proves new existence, uniqueness, and regularity results for Wasserstein barycenters, a natural way to average probability measures, on very general non-smooth and infinite-dimensional spaces. It also introduces a new curvature-dimension condition called BCD, defined through a Jensen inequality for entropy, and shows this condition is stable under measured Gromov-Hausdorff convergence.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The uniqueness and multi-marginal theorems for RCD(K,∞) rely on an optimal-map property cited only to a finite-dimensional reference; the core existence/Jensen theorem is not affected.","rationale":"Placeholder to ensure JSON validity.","tokens_in":39905,"tokens_out":18973,"duration_ms":171388,"concrete_test":"Check whether the Rajala–Sturm result [RS14] or another published theorem establishes existence and uniqueness of optimal transport maps from an absolutely continuous measure to an arbitrary measure on a strong CD(K,∞)/RCD(K,∞) space. If such a theorem exists, the gap reduces to a citation fix, and Theorems 5.9/5.16B/5.17B can be repaired by replacing [GRS16] with that reference. If no such theorem exists, ask the authors to provide a proof, or restrict all RCD(K,∞) uniqueness and multi-marginal statements to the case where the Monge property is known (e.g., RCD(K,N) or Hilbert/Gaussian spaces). A second useful check is to test the strict-convexity argument in Proposition 4.5 on an infinite-dimensional Gaussian Hilbert space, where Brenier maps are known to exist, to confirm the proof structure carries over.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's full advertised set of results includes uniqueness of barycenters and the multi-marginal Monge problem for RCD(K,∞) spaces. Theorems 5.9, 5.15B, 5.16B, and 5.17B all assume the weak Monge property: for any two absolutely continuous measures with finite W2 distance, a unique optimal transport map exists. The only supporting citation given in Proposition 5.15 is [GRS16], which treats finite-dimensional RCD(K,N) spaces; [CM17] also requires a finite dimension parameter (essentially non-branching MCP(K,N)). No proof or reference is supplied for infinite-dimensional RCD(K,∞). If this property is not known or fails, then the uniqueness theorem and every RCD(K,∞) multi-marginal existence/uniqueness/absolute-continuity result in Section 5.3 are unjustified as stated. This gap does not threaten Theorems 5.3 and 5.8: the EVI-to-Jensen argument and the tightness-based existence proof are self-contained and, under the paper's convention that reference measures are probability measures, appear sound. The central existence/Jensen claim therefore survives; the conditional verdict should concern only the broader theorem set.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper develops a theory of Wasserstein barycenters on non-compact, non-smooth, and extended metric measure spaces. Its central mechanism is to derive Jensen-type inequalities for the relative entropy from the existence of EVI_K (and EVI_K,N) gradient flows, then to use those inequalities as a priori estimates to obtain existence, absolute continuity, and uniqueness of barycenters without local compactness. The authors prove such Jensen inequalities for RCD(K,∞) and RCD(K,N) spaces, for abstract Wiener spaces, and for configuration spaces, and they introduce a new Barycenter-Curvature-Dimension (BCD) condition defined through barycentric Jensen inequalities. They prove stability of BCD(K,∞) under measured Gromov–Hausdorff convergence, prove existence of barycenters on BCD spaces under mild hypotheses, and derive applications including multi-marginal Brunn–Minkowski and functional Blaschke–Santaló type inequalities. The paper also treats the multi-marginal Monge problem with the barycenter cost for RCD spaces.","tokens_in":40146,"tokens_out":8737,"duration_ms":83073,"significance":"The EVI-to-Jensen argument is elegant and genuinely dimension-free; the existence and Jensen parts of the paper do not use fitted parameters and are essentially self-contained. The interpretation of Jensen's inequality as an a priori estimate is a valuable conceptual contribution, and the BCD condition together with its stability and geometric applications is natural and potentially influential. If all advertised results are correct, the paper unifies and extends several earlier barycenter results in Euclidean, Riemannian, Alexandrov, Wiener, and configuration-space settings. However, the advertised uniqueness and multi-marginal results for RCD(K,∞) rest on an infinite-dimensional weak Monge property that is not proved or correctly referenced; this gap affects a substantial part of the theorem set, even though the main existence/Jensen core survives.","major_comments":[{"comment":"The weak Monge property for RCD(K,∞) spaces is asserted without proof. The proof of Proposition 5.15 uses unique optimal transport maps between absolutely continuous measures in the RCD(K,∞) case and cites [CM17] and [GRS16], but both references treat finite-dimensional settings (essentially non-branching MCP(K,N) and RCD(K,N) with finite N). Theorems 5.16(B) and 5.17(B) inherit this gap, as does the claim in Theorem 5.9 that RCD(K,∞) spaces are important examples. Because Theorems 1.4, 1.10, and 1.11 advertise RCD(K,∞) uniqueness and multi-marginal results, this is load-bearing. Please either provide a proof or a valid reference for the weak Monge property in infinite-dimensional RCD(K,∞) spaces, or restrict the corresponding statements to finite-dimensional RCD(K,N) and state the infinite-dimensional cases as conditional assumptions.","section":"§5.3, Proposition 5.15(B), Theorems 5.16(B), 5.17(B), and Theorem 5.9"},{"comment":"The RCD(K,∞) parts of these proofs are only sketched as \"by induction and a similar truncation argument\". Even if the weak Monge property were available, the truncation argument would need to show that the reduced multi-marginal problem still has all marginals absolutely continuous, that the optimal plan remains optimal after truncation, and that the reduced marginals have finite entropy so that Proposition 5.15(B) applies. As written, this reduction is not demonstrated. The gap is not merely expository: it is used to remove the finite-entropy condition and to prove absolute continuity of the barycenter. Please supply the missing argument or state these results conditionally on the weak Monge property plus the truncation procedure.","section":"§5.3, proofs of Theorem 5.16(B) and Theorem 5.17(B)"}],"minor_comments":[{"comment":"There are typos, including \"curvature-dimesion\" in the abstract and \"As a by product\" in the introduction; these should be corrected.","section":"Abstract and general presentation"},{"comment":"The proposition states that strict convexity holds if one marginal is absolutely continuous, but the proof requires optimal transport maps from the absolutely continuous measure μ to both ν1 and ν2. The stated hypothesis is broader than the proof supports. The later application in Theorem 5.9 only needs maps between absolutely continuous measures, so restating the proposition in that narrower form would make it correct.","section":"§4.2, Proposition 4.5"},{"comment":"The introduction defines the multi-marginal cost with a factor 1/2, while Section 5.3 uses c(x1,...,xn)=inf_y Σ d²(x_i,y) without the factor. The factor does not affect barycenters but does affect the numerical equality between the transport cost and the barycenter functional; the convention should be fixed consistently.","section":"§5.3, cost function definition"},{"comment":"The text says the exponential volume growth condition holds \"for all x0 ∈ X\", but the definition used in Theorem 3.13 only assumes existence of some x0 and some c>0. Please make the base point and constant explicit in the proof.","section":"§6.3, Theorem 6.7, Step 3"},{"comment":"The paper defines the reference measure m as a probability measure, while standard RCD(K,∞) spaces are often stated for σ-finite reference measures. The tightness argument in Theorem 5.8(A) relies on a uniform entropy bound, which is used together with finite total mass of m; this normalization should be stated explicitly in the hypotheses of the theorem.","section":"§3.1, Definition 3.1 and Theorem 5.8(A)"}],"recommendation":"major_revision","confidential_remarks":"The central EVI/Jensen existence theorem and the BCD stability results appear sound and are likely publishable. The main risk is the RCD(K,∞) weak Monge gap: if the authors cannot supply a proof or reference, the paper can still be published by restricting the uniqueness and multi-marginal theorems to finite-dimensional RCD(K,N) and to the abstract Wiener-space cases where optimal maps are known. I would ask the editor to request that the authors either establish the infinite-dimensional weak Monge property or clearly modularize the claims so that the unsupported RCD(K,∞) statements do not appear as theorems without qualification."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The core idea here — deriving Wasserstein Jensen's inequality from EVI gradient flows rather than local compactness — is genuinely new and the proof is clean. The dimensional Jensen inequality (Theorem 5.11) appears new even on Euclidean space, and I don't see an obvious flaw in that argument. The BCD condition is a real addition to the Lott–Sturm–Villani toolbox, and the stability under mGH convergence is a nice result. For RCD(K,N) spaces, the multi-marginal existence and uniqueness theorems look solid.\n\nThe soft spot is the RCD(K,∞) case. The paper repeatedly asserts that RCD(K,∞) spaces satisfy the weak Monge property, citing [GRS16], but that reference is for finite-dimensional RCD(K,N) spaces. [CM17] also requires an essentially non-branching MCP(K,N) condition with finite N. As far as I know, the existence of a unique optimal map between arbitrary absolutely continuous probability measures with finite W2 distance is not established for general RCD(K,∞) spaces. This gap affects Theorems 5.9, 5.15, 5.16, and 5.17 as stated for RCD(K,∞). It does not affect Theorems 5.3 and 5.8, which are the heart of the paper and appear sound.\n\nThis is not a fatal flaw; it is a citation/proof gap in a set of applications. The authors should either prove the weak Monge property for RCD(K,∞) (or find a correct reference) or restrict those statements to spaces where it is known, e.g., by adding an explicit assumption. The central existence and Jensen results stand, and the BCD theory is unaffected.\n\nI'd send this to a serious referee. The paper is long but well-organized; the referee should focus on the RCD(K,∞) claims and verify the EVI_K,N proof. For a reading group, it's worth a session: the EVI-to-Jensen trick is a good concept, and the gap is instructive.","headline":"Strong paper with a genuine new proof strategy, but the RCD(K,∞) claims lean on an optimal-map property that is only cited to a finite-dimensional reference.","tokens_in":40688,"tokens_out":3588,"would_cite":true,"duration_ms":32097,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["53C23","51F99","49Q22"],"pacs":[],"model":"deepseek-v4-flash","headline":"Wasserstein barycenters of probability measures exist and stay regular on non-compact spaces with synthetic Ricci lower bounds, and the paper proves this through Jensen's inequality derived from EVI gradient flows, introducing a new…","keywords":["Wasserstein barycenter","metric measure space","curvature-dimension condition","Ricci curvature","optimal transport","gradient flow","Jensen's inequality","EVI"],"falsifier":"Exhibit two absolutely continuous probability measures at finite Wasserstein distance on an infinite-dimensional RCD(K,∞) space, such as an abstract Wiener space, that admit two distinct optimal transport plans; the uniqueness assertions of Theorems 5.9, 5.16(B), and 5.17(B) would then fail.","tokens_in":39700,"feed_emoji":"📐","tokens_out":8112,"duration_ms":65812,"temperature":0.7,"pith_summary":"This paper establishes that Wasserstein barycenters—the natural averaged probability measures in optimal transport—exist and are well behaved on non-compact, non-smooth spaces with synthetic lower Ricci bounds, and even on extended metric spaces such as abstract Wiener spaces, without any local compactness assumption. The route is to run the relative entropy along an EVI gradient flow in the Wasserstein space and integrate the flow inequality: this yields Jensen's inequality for barycenters, which then acts as an a priori estimate from which existence, uniqueness, and absolute continuity of barycenters follow. On spaces satisfying a new Barycenter-Curvature-Dimension condition, the same mechanism gives existence of barycenters, stability under measured Gromov–Hausdorff convergence, and geometric inequalities such as multi-marginal Brunn–Minkowski and functional Blaschke–Santalo type estimates. The result matters because it replaces compactness-based existence proofs with a dimension-free functional-analytic mechanism.","feed_headline":"Gradient flows guarantee Wasserstein barycenters without compactness","feed_subtitle":"Jensen's inequality from EVI gradient flows yields existence, uniqueness, and regularity on RCD and Wiener spaces.","key_machinery":"The load-bearing object is the $\\mathrm{EVI}_K$ gradient flow of the relative entropy in the Wasserstein space: a curve $(\\mu_t)$ satisfying $\\frac{1}{2}\\frac{d}{dt}W_2^2(\\mu_t,\\nu) + \\frac{K}{2}W_2^2(\\mu_t,\\nu) \\le \\mathrm{Ent}_m(\\nu)-\\mathrm{Ent}_m(\\mu_t)$. The paper's integral characterization of this flow inequality allows one to integrate it against the measure $\\Omega$; choosing the starting point to be the barycenter then yields Jensen's inequality directly. Jensen's inequality plays the role of an a priori estimate: it bounds the entropy of the barycenter, forces the barycenter to be absolutely continuous, and, with a weak Monge property, yields uniqueness. The same estimate through the $\\mathrm{EVI}_{K,N}$ version gives the dimension-dependent refinement, and the Barycenter-Curvature-Dimension condition abstracts this barycenter Jensen inequality into a new curvature-dimension condition.","core_discovery":"The paper's central claim is that, on an extended metric measure space where every measure with finite distance to the domain of the relative entropy starts an $\\mathrm{EVI}_K$ gradient flow of the entropy in the Wasserstein space—in particular on $\\mathrm{RCD}(K,\\infty)$ spaces and abstract Wiener spaces—any probability measure $\\Omega$ over the Wasserstein space with finite variance and finite expected entropy has a Wasserstein barycenter, every barycenter is absolutely continuous with respect to the reference measure, and Jensen's inequality $\\mathrm{Ent}_m(\\bar\\mu) \\le \\int \\mathrm{Ent}_m\\,d\\Omega - \\frac{K}{2}\\int W_2^2(\\bar\\mu,\\cdot)\\,d\\Omega$ holds. The same method, pushed through the finite-dimensional $\\mathrm{EVI}_{K,N}$ gradient flow, yields a dimension-dependent Jensen inequality that seems new even on $\\mathbb{R}^n$ and implies uniqueness and absolute continuity on $\\mathrm{RCD}(K,N)$ spaces. A second claim is that a new Barycenter-Curvature-Dimension condition, defined by requiring exactly this barycenter Jensen inequality, is stable under measured Gromov–Hausdorff convergence and suffices for existence of barycenters, multi-marginal optimal transport maps, and Brunn–Minkowski and Blaschke–Santalo type inequalities.","pith_inferences":["The paper leaves implicit that the same EVI-to-Jensen mechanism should give existence of barycenters in any barycenter space whose entropy functional has a contractive gradient flow, potentially far beyond Ricci bounds—for example on Poisson configuration spaces over manifolds with Ricci lower bound, where the EVI is already known.","The Barycenter-Curvature-Dimension condition is defined relative to the reference measure's entropy; a natural test is whether $\\mathrm{BCD}(K,\\infty)$ plus geodesicity actually forces the standard $\\mathrm{CD}(K,\\infty)$ condition, or whether a non-geodesic BCD space exists that has no Lott–Sturm–Villani structure, which would show the new condition is strictly broader.","The dimension-dependent Jensen inequality is new even on Euclidean space; applying it to indicator-type densities may yield quantitative finite-$N$ refinements of Brunn–Minkowski inequalities for Wasserstein barycenter sets.","A possible resolution of the weak Monge gap would be to prove the infinite-dimensional $\\mathrm{RCD}(K,\\infty)$ optimal transport map theorem by finite-dimensional approximation; until then, the uniqueness half of Theorems 5.9, 5.16(B), and 5.17(B) rests on an unproved premise."],"forward_implications":["On any $\\mathrm{RCD}(K,\\infty)$ space, every probability measure $\\Omega$ over the Wasserstein space with finite variance and finite expected entropy has a barycenter, and every barycenter is absolutely continuous with respect to the reference measure (Theorems 5.3 and 5.8).","Under the weak Monge property—valid in finite-dimensional RCD spaces—the barycenter is unique; the dimension-dependent Jensen inequality gives a refined uniqueness statement on $\\mathrm{RCD}(K,N)$ spaces whenever $\\Omega$ gives positive mass to finite-entropy measures (Theorem 5.9 and Corollary 5.13).","The multi-marginal optimal transport problem with cost $c(x_1,\\ldots,x_n)=\\inf_y \\sum_i \\tfrac12 d^2(x_i,y)$ has a unique Monge solution on RCD spaces, and finite-support Wasserstein barycenters of absolutely continuous measures are unique and absolutely continuous without a finite-entropy condition (Theorems 5.16 and 5.17).","The $\\mathrm{BCD}(K,\\infty)$ condition is closed under measured Gromov–Hausdorff convergence and, under exponential volume growth or a probability reference measure, guarantees existence of Wasserstein barycenters (Theorems 6.6 and 6.7).","$\\mathrm{BCD}(K,N)$ spaces satisfy a multi-marginal Brunn–Minkowski inequality, and $\\mathrm{BCD}(1,\\infty)$ spaces satisfy a functional Blaschke–Santalo type inequality (Propositions 6.8–6.10)."],"supporting_citations":[{"why":"Supplies the integral characterization of EVI gradient flows that converts the flow inequality into Jensen's inequality.","marker":"[DS08]"},{"why":"Characterizes RCD(K,∞) spaces by existence of EVI_K gradient flows of the relative entropy, making the main Jensen inequality applicable to them.","marker":"[AGS14]"},{"why":"Extends the EVI framework to extended metric measure spaces, covering abstract Wiener spaces and configuration spaces.","marker":"[AES16]"},{"why":"Is the prior existence theorem for Wasserstein barycenters under local compactness, the assumption this paper removes.","marker":"[LGL17]"},{"why":"Initiated the Euclidean Wasserstein barycenter theory and the link to multi-marginal optimal transport that the paper generalizes.","marker":"[AC11]"},{"why":"Cited for optimal transport maps on finite-dimensional RCD spaces, the basis for uniqueness and the weak Monge property.","marker":"[GRS16]"},{"why":"Provides optimal maps in essentially non-branching spaces, used for RCD(K,N) multi-marginal transport and barycenter regularity.","marker":"[CM17]"},{"why":"Supplies optimal transport maps and Wasserstein completeness on Wiener spaces, used for the abstract Wiener space example.","marker":"[FU04]"},{"why":"Equates RCD(K,N) with EVI_{K,N} gradient flows, used for the dimension-dependent Jensen inequality.","marker":"[EKS15]"}],"fun_headline_variants":["Wasserstein barycenters exist on RCD and Wiener spaces via EVI flows","New Barycenter-Curvature-Dimension condition is stable and yields inequalities","Jensen's inequality for barycenters holds without compactness","Dimension-dependent Jensen inequality new even on R^n","Geometric inequalities from barycenter curvature condition"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The uniqueness and absolute-continuity conclusions for infinite-dimensional RCD(K,∞) spaces rely on the weak Monge property—that any two absolutely continuous measures at finite Wasserstein distance admit a unique optimal transport map—yet the paper cites support for that property only in finite-dimensional RCD(K,N) spaces and gives no proof or reference for the infinite-dimensional case.","fun_headline_variants_meta":{"raw":{"variants":["Wasserstein barycenters exist on RCD and Wiener spaces via EVI flows","New Barycenter-Curvature-Dimension condition is stable and yields inequalities","Jensen's inequality for barycenters holds without compactness","Dimension-dependent Jensen inequality new even on R^n","Geometric inequalities from barycenter curvature condition"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000299,"raw_usage":{"total_tokens":1758,"prompt_tokens":1003,"completion_tokens":755,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":619,"completion_tokens_details":{"reasoning_tokens":667}},"tokens_in":619,"tokens_out":755,"duration_ms":7177,"temperature":1.0,"reasoning_tokens":667,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T04:38:14.171933+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Exhibit two absolutely continuous probability measures at finite Wasserstein distance on an infinite-dimensional RCD(K,∞) space, such as an abstract Wiener space, that admit two distinct optimal transport plans; the uniqueness assertions of Theorems 5.9, 5.16(B), and 5.17(B) would then fail.","supporting_citations":[],"review_version":1}