{"id":"665d2a44-6aa8-477e-877c-cef81c4a5231","arxiv_id":"2607.23241","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Extends the Dacorogna–Moser transport-map construction to R^d for strictly asymptotically log-concave measures and derives optimality conditions for geodesics on the space of couplings.","lead":"This paper extends a classical method for building smooth transport maps between probability distributions from bounded regions to all of R^d, for a broad class of 'asymptotically log-concave' measures. It then uses these maps to describe shortest paths in the space of joint distributions with fixed marginals, settling an open question from a recent paper.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2.11 proof uses a demonstrably false integration-by-parts identity; the cancellation in Eq. (48) is invalid as written and must be repaired before the optimality conditions can be accepted.","rationale":"The reader's weakest-assumption concern focuses on the reflection-coupling estimates imported from [PW06]/[EZ19]. That is a legitimate concern about an external black box, but the proof gap I found is internal, concrete, and checkable: the integration-by-parts chain in (48) contains a false identity and a sign error. It affects Theorem 2.11, one of the paper's headline applications (rigorous optimality conditions and uniqueness of multipliers). The error is probably repairable—the cancellation in (48) can be recovered by applying the correct signs—so I do not recommend changing the conditional verdict to reject. But the theorem is not established as written, and the repair should be provided. Hence the reader's CONDITIONAL verdict remains appropriate; no change is needed.","tokens_in":26114,"tokens_out":28601,"duration_ms":259720,"concrete_test":"Take µ = γ on R and f(x) = x²/2 − 1/2. Compute ∫ Δ²f dµ = 0 and ∫ ∇log µ·∇f dµ = ∫ x² dγ = 1, directly contradicting the displayed equality in Eq. (48). Then re-derive (47) with correct integration-by-parts signs; if the cancellation holds, the flaw is repairable; if not, Theorem 2.11's conclusions (including uniqueness) are unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In the proof of Theorem 2.11, after Eq. (46), the paper derives the identities (47) via the displayed chain in (48). That chain contains a false equality: for f_t solving Δ_µ f_t = φ_t, it asserts ∫ Δ²f_t dµ = ∫ ∇log µ · ∇f_t dµ. This is not an identity. For µ = N(0,1) and f(x)=x²/2 − 1/2 (which satisfies ∫ f dµ = 0 and Δ_µ f = 1 − x²), the left side is ∫ f'''' dµ = 0 while the right side is ∫ x·x dµ = 1. There is also a sign error two lines earlier: ∫ ∇_x ρ_t · ∇Δf_t dxdy = −∫ Δ²f_t dµ, not +. If the signs are corrected, the two terms in (48) do cancel, so the identities (47) may be salvageable; but as written, the proof of Theorem 2.11 is invalid. Since Theorem 2.11 is the paper's main optimality-condition result, this is a load-bearing defect that must be fixed before the theorem can be considered established.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper proves an extension of the Dacorogna–Moser construction to the whole of R^d for strictly asymptotically log-concave reference measures with Lipschitz derivatives of the log-density (Theorem 1.2), based on global W^{n,∞} estimates for the elliptic operator Δ_μ = Δ + ∇log μ · ∇ obtained from parabolic semigroup estimates via reflection coupling (Proposition 2.1). It then applies this construction to the study of geodesics in the space of couplings Π(μ,ν): existence and Γ-convergence for an entropic regularization (Theorem 2.6), existence of Lagrange multipliers for the marginal constraints (Theorem 2.9), optimality conditions and uniqueness of the multipliers (Theorem 2.11), and weak-* convergence of the multipliers (Proposition 2.12). The paper also sketches a path-space interpretation of the regularized problem in Appendix A.","tokens_in":26399,"tokens_out":19352,"duration_ms":180565,"significance":"If the technical gaps are repaired, this would be a substantial contribution. Theorem 1.2 is a natural and nontrivial extension of the classical Dacorogna–Moser construction to unbounded domains, with quantitative global estimates, and the use of reflection-coupling semigroup estimates is well matched to the hypotheses. The application to geodesics in the coupling space addresses an open question from [CLP25], and the optimality/uniqueness results together with Γ-convergence and multiplier convergence provide a coherent variational framework. The paper also gives credit to several recent tools, including [CCE25] and [CL25]. However, several load-bearing computations in the proofs of Proposition 2.1 and Theorem 2.11 are currently invalid or incomplete, so the manuscript cannot be accepted without a substantive revision.","major_comments":[{"comment":"The displayed chain (48) is false as written. For μ=N(0,1) and f(x)=x²/2−1/2, one has Δ_μ f = 1−x², ∫ Δ²f dμ = ∫ f'''' dμ = 0, while ∫ ∇log μ · ∇f dμ = ∫ (−x)·x dμ = −1; hence the asserted equality ∫ Δ²f_t dμ = ∫ ∇log μ · ∇f_t dμ is not an identity. There is also a sign error two lines earlier: integration by parts gives ∫ ∇ρ · ∇Δf = −∫ ρ Δ²f, not +. The intended identities (47) are in fact true for ρ_t ∈ Π(μ,ν) because ∫ Δ_μ h dμ = 0 and the first marginal of ρ_t is μ, so the step is likely repairable, but the proof of Theorem 2.11 must be rewritten here.","section":"3.2, Theorem 2.11, Eq. (48)"},{"comment":"The expansion of A₀ uses pairings defined as ⟨∂t(ρv),(∇f,0)⟩ = ∫(∇f_t,0)·v_t dρ_t dt and ⟨div(ρv⊗v),(∇f,0)⟩ = ∫(∇²f_t v_t)·v_t dρ_t dt. This is not the first variation of A₀ along the perturbed curve ρ^δ_t = ((T^δ_t)^{-1},Id)_#ρ_t. The correct expansion contains −∫ ρ v·∂_t∇f dt and −∫ ρ (∇²f v)·v dt; the printed definitions have the wrong sign/content, and the subsequent 'substituting φ by −φ' does not repair the computation. This is a load-bearing step in the derivation of the optimality conditions.","section":"3.2, Theorem 2.11, Eq. (45)"},{"comment":"The proof claims ∥∂_I v_t∥∞ ≤ ∥∂_I φ∥∞ for the nonhomogeneous parabolic equation (26), whose right-hand side contains the source term −Σ ∂_J b · ∇∂_{I\\J} v. For such an equation the maximum principle gives an additional integral of the source; the displayed bound is false in general. Although the subsequent stochastic estimates may not require this exact bound, it is asserted as a justification for applying Itô's formula. Moreover, the integrals in the source term are bounded using the coupling probability p_s(x,y)=P(X_s≠Y_s), but the integrands involve Lipschitz functions of the processes; controlling these differences requires an estimate on E|X_s−Y_s| (or an equivalent coupling contraction bound), not merely on P(X_s≠Y_s). The manuscript should state and prove the needed coupling contraction estimate from [EZ19].","section":"3.1, Proposition 2.1, parabolic induction around Eq. (26)"},{"comment":"The line d/dt ∫ φ_t dμ = (1/2)∫ ∇·(μ φ_t) dx is wrong in two ways: it should involve μ∇φ_t rather than μφ_t, and the vanishing of the integral of a divergence over R^d requires spatial decay/boundary conditions that are not established. The conclusion ∫φ_t dμ = 0 is true and follows immediately from the fact that S_t is the symmetric Markov semigroup of ½Δ_μ in L²(μ); the proof should be rewritten accordingly. Without this step the pointwise definition of ψ = −∫₀^∞ φ_t dt and the mean-zero condition are not justified by the text.","section":"3.1, Proposition 2.1, elliptic part"}],"minor_comments":[{"comment":"The notation 'φ_t(x)∈L^1(dt⊗dμ(x))' is imprecise; it should be stated as a joint integrability condition for the function (t,x)↦φ_t(x).","section":"3.1, elliptic part"},{"comment":"Labeling an 'Informal' statement as a Proposition may confuse readers; consider calling it a Remark or stating precise hypotheses.","section":"Appendix A, Proposition A.1"},{"comment":"The reference [MACJC17] has a garbled author list; it should be M. Arnaudon, A.B. Cruzeiro, C. Léonard, and J.-C. Zambrini.","section":"References"},{"comment":"The construction of the velocity field v_t by pushforward under Lipschitz maps T,S should spell out the weak/a.e. differentiability interpretation if T,S are only Lipschitz; this is used later in the continuity equation.","section":"3.2, Theorem 2.6"},{"comment":"Several steps are described only as 'straightforward adaptation of [Bar20]' without checking the endpoint or compactness conditions specific to this setting. Please make these adaptations explicit, especially for the expansions leading to (45)–(49).","section":"3.2, Theorem 2.11"}],"recommendation":"major_revision","confidential_remarks":"The manuscript leans heavily on [CCE25, Theorem 1.4] and [CL25, Propositions 3.1/3.6] for essential infrastructure: the Lipschitz map from the Gaussian, the Poincaré inequality, and Gaussian-coupling geodesics. Both are arXiv preprints co-authored by the first author, and they are not included in the manuscript. This is not by itself a reason to reject, but it creates a verification risk; the editor may wish to ensure these works are publicly available in a form the referees can check. The paper's fit with math.AP is good, but the technical gaps in Sections 3.1 and 3.2 require a genuine revision before the claims can be accepted."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things to know. Theorem 1.2 is the first Dacorogna–Moser construction on all of R^d for strictly asymptotically log-concave measures, and the geodesic half answers the open problem from CLP25 with existence, Γ-convergence, and multiplier results. That is real progress. The second thing: the proof of Theorem 2.11 has a concrete false equality, and the parabolic/elliptic section has two unjustified analytic steps. None of this looks fatal, but the paper needs revision before the optimality conditions can be trusted as written.\n\nThe architecture is coherent. The idea of using reflection coupling to get global W^{n,∞} bounds for the elliptic and parabolic equations is exactly the right tool, and Theorem 1.2 is the missing infrastructure the literature has needed. The application to coupling-space geodesics is not window dressing: the existence result, the regularized problem, and the multiplier convergence genuinely extend what was known. The literature review is careful and positions the contribution accurately. I did not find any sign that the headline result is assumed in its own proof.\n\nNow the soft spots. In the proof of Theorem 2.11, Eq. (48) asserts ∫ Δ²f dµ = ∫ ∇logµ·∇f dµ. This is false: for µ = N(0,1), f(x)=x²/2−1/2, the left side is 0 and the right side is −1. There is also a sign error just above: ∫∇ρ·∇Δf = −∫Δ²f, not +. The identities (47) are probably salvageable—with the sign fixed, the sum becomes −∫Δ_µ(Δf)dµ, which vanishes under the same integration-by-parts assumptions—but as written the proof is invalid, and Theorem 2.11 is load-bearing.\n\nSection 3.1 also has two passages that need tightening. The sup-norm bound on ∂_I v_t from the nonhomogeneous parabolic equation is not justified as stated; the source term can contribute a (T−t) factor. And the step d/dt ∫φ_t dµ uses the divergence theorem without spelling out boundary decay. The second is a presentation gap rather than a deep problem, because Assumption (A.1) gives Gaussian-type tail decay. The first should be either proved or replaced. The paper leans heavily on [CCE25] and [CL25], both co-authored by the first author, for key infrastructure. That is not circular, but a referee should check those black boxes.\n\nWho is this for: people working on regular transport maps on R^d and on geodesics in Wasserstein submanifolds. My recommendation: send it to peer review. The core contribution is important and plausible, and the defects look repairable, but the revision must fix Eq. (48) and clean up the PDE estimates.","headline":"A genuinely useful R^d Dacorogna–Moser extension with a serious application to geodesics in coupling space, but the optimality proof contains a concrete false identity that must be repaired before the main theorems are accepted.","tokens_in":26904,"tokens_out":5481,"would_cite":true,"duration_ms":53945,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["49Q22","35B65","60J60"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper extends the Dacorogna–Moser transport-map construction to all of R^d for strictly asymptotically log-concave measures, and derives optimality conditions for geodesics in the space of couplings.","keywords":["transport maps","Dacorogna–Moser construction","asymptotically log-concave measures","reflection coupling","geodesics in the space of couplings","Lagrange multipliers","entropic regularization","Schrödinger bridge problem"],"falsifier":"Numerically compute the solution ψ of ∆µψ=φ on R^d for µ a Gaussian mixture and φ a sequence of compactly supported, mean-zero functions with growing support, and check whether ∥∇ψ∥_{W^{n,∞}} stays bounded by C∥φ∥_{W^{n,∞}} with C independent of the support; alternatively, estimate the reflection-coupling probability p_s(x,y) for large |x−y| and test the exponential decay with a positive rate required in the proof.","tokens_in":25988,"feed_emoji":"➡️","tokens_out":7692,"duration_ms":64491,"temperature":0.7,"pith_summary":"The paper proves that the classical Dacorogna–Moser construction of regular transport maps, previously confined to bounded domains, works on the whole space R^d for strictly asymptotically log-concave measures. The core technical advance is a global, uniform-in-time regularity estimate for the elliptic equation ∆ψ + ∇logµ·∇ψ = φ on R^d, obtained from parabolic semigroup estimates and reflection coupling. This yields an invertible transport map T with T#µ = (1+φ)µ and quantitative W^{n,∞} control on T−Id and T^{-1}−Id. The authors then use this construction to study geodesics in the space of probability measures on a product space with fixed marginals (couplings): they prove existence, derive optimality conditions with Lagrange multipliers, and introduce an entropic regularization whose minimizers converge as the regularization vanishes. Together these results answer an open question about the structure of geodesics in the space of couplings.","feed_headline":"Transport maps now built on whole R^d","feed_subtitle":"Strict log-concave tails get quantitative bounds, unlocking geodesic optimality conditions.","key_machinery":"Proposition 2.1, the parabolic–elliptic regularity estimate: the semigroup S_t generated by ½∆µ satisfies ∥∇S_tφ∥_{W^{n,∞}} ≤ c e^{−αt}∥φ∥ for t≥1 and ≤ c/√t ∥φ∥ for t≤1, and the unique solution ψ of ½∆µψ=φ, ∫ψ dµ=0 satisfies ∥∇ψ∥_{W^{n,∞}} ≤ C∥φ∥_{W^{n,∞}}. These estimates are proved by induction on the number of derivatives, using the Feynman–Kac formula, exponential contraction of the reflection-coupling probability, and standard small-time gradient estimates for diffusion semigroups. The strict asymptotic log-concavity of µ enters through the exponential contraction rate β>0 of the coupling.","core_discovery":"Under Assumption (A.1) — µ strictly asymptotically log-concave with ∇^j logµ Lipschitz for 1≤j≤n+1 — every mean-zero φ∈W^{n,∞} with φ≥−1+δ admits an invertible transport map T such that T#µ = (1+φ)µ, with T−Id and T^{-1}−Id in W^{n,∞} and ∥T−Id∥_{W^{n,∞}} + ∥T^{-1}−Id∥_{W^{n,∞}} ≤ C∥φ∥_{W^{n,∞}}. The map is generated by the flow of the vector field v_s = −∇ψ/(1+sφ), where ψ solves ∆µψ=φ. The proof of the required global regularity of ψ rests on short- and long-time smoothing bounds for the semigroup of ½∆µ, proved by Feynman–Kac representation and reflection-coupling contraction. The same bounds supply the compactness and competitor arguments for the geodesic problem.","pith_inferences":["A natural next step is to check whether the one-derivative loss between the bounded-domain and global theorems is intrinsic: if higher-order global uniform estimates are genuinely impossible, fractional regularity is likely the sharp endpoint.","The reflection-coupling estimates are probabilistic and depend only on tail contraction of the drift; the same strategy may extend the construction to non-Euclidean settings such as complete Riemannian manifolds with suitable curvature bounds.","The rigorous optimality conditions could support a gradient-flow theory in the space of couplings, potentially connecting the geodesic distance to projected Langevin dynamics and offering a quantitative handle on entropic regularization.","Numerical computation of ∥T−Id∥_{W^{n,∞}} for Gaussian mixtures (which are strictly asymptotically log-concave) could test the sharpness of the constants in the paper's estimates."],"forward_implications":["For any mean-zero φ in W^{n,∞} with φ≥−1+δ, a regular invertible transport map T with T#µ=(1+φ)µ exists on R^d with explicit W^{n,∞} control, extending sampling and functional-inequality tools beyond bounded domains.","Geodesics in the space Π(µ,ν) of couplings with fixed marginals exist whenever µ and ν are strictly asymptotically log-concave, a broader class than the previously treated strongly log-concave marginals.","The marginal constraints admit a unique Lagrange multiplier in the relevant dual space, making the formal optimality conditions (continuity equation plus momentum equation) rigorous.","The entropically regularized geodesic problem has a unique minimizer, Γ-converges to the original problem as ε→0, and its Lagrange multipliers converge weak-* to the unregularized one."],"fun_headline_variants":["Transport maps on R^d via log-concave tails","Extending Moser's transport to all of R^d","Global transport maps with log-concave tails","Optimality conditions for geodesics on couplings","Log-concave measures yield transport maps on R^d"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is that µ is strictly asymptotically log-concave with enough Lipschitz regularity so that the reflection-coupling contraction rate β in the semigroup estimates is strictly positive; if the drift ½∇logµ fails to be eventually contracting at large scales, the exponential decay in Proposition 2.1 fails and Theorem 1.2's transport map is not obtained.","fun_headline_variants_meta":{"raw":{"variants":["Transport maps on R^d via log-concave tails","Extending Moser's transport to all of R^d","Global transport maps with log-concave tails","Optimality conditions for geodesics on couplings","Log-concave measures yield transport maps on R^d"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000203,"raw_usage":{"total_tokens":1271,"prompt_tokens":844,"completion_tokens":427,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":588,"completion_tokens_details":{"reasoning_tokens":347}},"tokens_in":588,"tokens_out":427,"duration_ms":4259,"temperature":1.0,"reasoning_tokens":347,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T23:59:15.906462+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Numerically compute the solution ψ of ∆µψ=φ on R^d for µ a Gaussian mixture and φ a sequence of compactly supported, mean-zero functions with growing support, and check whether ∥∇ψ∥_{W^{n,∞}} stays bounded by C∥φ∥_{W^{n,∞}} with C independent of the support; alternatively, estimate the reflection-coupling probability p_s(x,y) for large |x−y| and test the exponential decay with a positive rate required in the proof.","supporting_citations":[],"review_version":1}