{"id":"68439a21-9cb4-4da8-825d-b64f465b335b","arxiv_id":"2412.01012","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"On a globally hyperbolic spacetime, any weak Kantorovich potential for the squared time-separation cost is locally semiconvex almost everywhere, implying optimal couplings are induced by transport maps.","lead":"This math paper proves that in curved spacetime, the best way to move mass between two probability clouds is governed by a well-behaved helper function, which yields an optimal transport map. It extends recent Riemannian regularity results to Lorentzian spacetimes for the squared time-separation cost, a step toward geometric applications of optimal transport in general relativity.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proposition 4.7, which underpins the full-measure open set Ω1 and the transport-map formula, is proven only in Minkowski spacetime and deferred to a theorem for a different cost; the transfer to general globally hyperbolic spacetimes for c2 is missing.","rationale":"The reader's weakest_assumption correctly identifies the incomplete proof of Proposition 4.7 as the main load-bearing gap. I traced the dependency chain: Theorem 6.5 (µ(Ω0)=1) uses Corollary 6.11, which uses Proposition 6.9, which invokes Proposition 4.7; Corollary 6.19 invokes Proposition 4.7 directly; and Corollary 6.2 uses Corollary 6.19 for the timelike subdifferential and the differentiability set. The paper's own proof of Proposition 4.7 is explicitly restricted to Minkowski spacetime and defers the general case to [17, Theorem 2.12], which concerns the cost (1.4). The cost c2 = c1^2 has different sublevel geometry near ∂J+, so the transfer is not a formality. I checked the other potential weak points identified in a second pass: the local semiconvexity obtained in the proof of Theorem 6.1 actually comes with a linear modulus (via Proposition 7.13 and Remark 7.14, together with [13, A15/A16]), so the appeal to a.e. differentiability in Corollary 6.2 is justified by the stronger property proved, not by the paper's weak definition alone. The twist condition, Theorem 5.1, and the construction of Ω1 appear internally consistent. No circularity, data, or authorship issues are present. The verdict should remain CONDITIONAL: the main theorem is plausible and well-structured, but Proposition 4.7 must be either fully proved for general globally hyperbolic spacetimes with c2 or replaced by an alternative argument establishing the timelike subdifferential property on a full-measure set.","tokens_in":46854,"tokens_out":18217,"duration_ms":163072,"concrete_test":"Write out a full proof of Proposition 4.7(i) for a general globally hyperbolic spacetime by repeating the Minkowski contradiction argument with points x_i = exp_{x0}((r i/m) u) for a spacelike unit vector u obtained as the projection of the lightlike direction, and verify that the estimate c2(x_{i+1},y_i)−c2(x_i,y_i) ≤ −C/√m holds uniformly using only (2.1) and dist(x0, supp ν)>0. If the estimate requires extra curvature or uniformity assumptions not stated in the paper, the theorem needs those assumptions or a different proof. Equivalently, check whether [17, Theorem 2.12] for the un-squared cost (1.4) actually implies the claim for c2 by showing optimality for c2 gives optimality for c1 on the relevant support; if not, the gap is real.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central regularity theorem (Theorem 6.1/1.3) and Corollary 6.2 rely on Proposition 4.7 in essential ways: Proposition 6.9 uses Proposition 4.7(ii) to conclude that the set of x0 with ∂c2φ(x0)∩I+(x0) nonempty has full µ-measure, and Corollary 6.19 uses Proposition 4.7 both to restrict to such x0 and to know that any y∈∂c2φ(x) is strictly timelike from x. Proposition 4.7 is stated for a general globally hyperbolic (M,g) and cost c2, but its proof says: 'We provide only a sketch of the proof for the case where (M,g)=(R^{1+n},⟨⟨·,·⟩⟩)' and then refers to Theorem 2.12 in [17], which is proved for the cost (1.4), not for c2. The forward reference to 'the more general Theorem 6.6' is also inaccurate: Theorem 6.6 is a local sup-bound statement, not a concentration-on-I+ statement. The Minkowski proof uses Lebesgue differentiation and the explicit affine structure of the light cone to obtain the uniform decrease c2(x_{i+1},y_i)-c2(x_i,y_i)≤−C/√m; in a curved spacetime one would need the same uniform estimate with constants independent of the chain, using the exponential map, the time function τ with (2.1), and the geometry of ∂J+. No such argument is supplied. Because the proof of Theorem 6.1 and the differentiability step in Corollary 6.2 both pass through this lemma, the main conclusion is conditional on this unproven transfer.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the Kantorovich optimal transport problem on a globally hyperbolic spacetime (M,g) for the Lorentzian cost c2(x,y) = (τ(y)-τ(x)-d(x,y))^2 on J+ and +∞ otherwise. It introduces π-solutions as weak Kantorovich potentials calibrated by a fixed optimal coupling, proves existence of π-solutions under strict timelikeness and causal compactness (Proposition 1.1/4.10), and establishes structural properties of c2-convex functions: local boundedness on the interior of their finiteness domain, countable n-rectifiability of the boundary of that domain, and a compactness condition for near-maximizers (Theorem 1.2/5.1). The main theorem (Theorem 1.3/6.1) asserts that, under absolute continuity of μ, disjoint supports, and existence of a π-solution, the potential is locally semiconvex on an open set Ω1 of full μ-measure. From this the paper derives that any optimal coupling admitting a π-solution is induced by a transport map satisfying ∂c2/∂x(x,T(x)) = -dφ(x), together with a uniqueness statement (Corollary 1.4/6.2). The central regularity theorem depends essentially on Proposition 4.7, which claims that optimal couplings are concentrated on the timelike region I+ and that for μ-a.e. x the c2-subdifferential contains a timelike point.","tokens_in":47187,"tokens_out":12021,"duration_ms":107770,"significance":"If fully supported, the paper would be a meaningful extension of Riemannian optimal-transport regularity theory to a Lorentzian cost: it proves local semiconvexity of weak potentials and a classical transport-map formula for a cost c2 that is not locally semiconcave on all of M×M but is locally semiconcave on I+. The manuscript contains substantial technical work: a Lagrangian formulation with an L2-exponential map, a detailed adaptation of Figalli-Gigli's proof of the structural properties of c2-convex functions, and a proof of local semiconcavity of c2 on I+ (Proposition 7.13). It is also transparent about the provenance of many arguments, which is a strength. However, the paper's central claim is conditional on Proposition 4.7, whose proof is only sketched in Minkowski space and whose transfer to general globally hyperbolic spacetimes for the cost c2 is not demonstrated. The main theorem and the transport-map conclusion lose their support unless that gap is closed.","major_comments":[{"comment":"Proposition 4.7 is stated for every globally hyperbolic spacetime and for the cost c2, but its proof is explicitly restricted to the Minkowski case and refers to [17, Theorem 2.12], which is proved for the cost (1.4), not for c2. Since c2 = c1^2 with c1(x,y) = τ(y)-τ(x)-d(x,y), c2-optimality does not automatically imply c1-optimality, so the citation does not transfer the result without further argument. This issue is load-bearing: Proposition 4.7(ii) is used in Proposition 6.9 to produce the full-measure set of points with a timelike element in the c2-subdifferential, and again in Corollary 6.19 and Corollary 6.2 to rule out lightlike pairs and to apply the twist condition on I+. The forward reference to 'the more general Theorem 6.6' is also inaccurate: Theorem 6.6 is a local strict-supremum bound, not a statement that optimal couplings are concentrated on I+. Please provide a complete proof of Proposition 4.7 for the cost c2 on general globally hyperbolic spacetimes, or a precise reduction that accounts for the quadratic transform.","section":"Section 4, Proposition 4.7"},{"comment":"Even in the Minkowski-space sketch, the argument assumes without proof that the set B and the partition sets {x ∈ B : |v_x - v| ≤ δ} are measurable, that the selected points x_i are Lebesgue points of one of these sets, and that inf_m φ(x_1^m) and sup_m φ(x_m^m) are finite independently of m. Since v_x is defined through an arbitrary choice of y ∈ (J+(x)\\I+(x)) ∩ supp(π) ∩ ∂c2φ, a measurable selection argument is needed before Lebesgue's differentiation theorem can be applied. If Proposition 4.7 is to remain a stated result, these measurability and selection issues need to be resolved in the full proof.","section":"Section 4, proof of Proposition 4.7(i)"},{"comment":"The sets D_i := {x ∈ B | (x, v_{x,k}) ∈ U_i for infinitely many k} are treated as closed sets and are used to extract a Lebesgue point, but no measurability or closedness proof is supplied. The vector v_{x,k} is selected from a maximizing geodesic that depends on x and on a sequence (y_{x,k}), so D_i need not be Borel as defined. Since the contradiction argument requires a measurable set of positive measure, a measurable selection for v_{x,k} (or an alternative construction) is necessary. This is a gap in the proof of Theorem 6.6, which is itself needed for the construction of Ω1 in Theorem 6.1.","section":"Section 6, proof of Theorem 6.6, Step 2"}],"minor_comments":[{"comment":"The text refers to 'Lemma 4.7', but the relevant statement is Proposition 4.7; please correct the cross-reference.","section":"Section 6, Corollary 6.19"},{"comment":"In the displayed equation, the expression '∂c2/∂x c2(x,y)' should read '∂c2/∂x(x,y)'.","section":"Section 6, proof of Corollary 6.2"},{"comment":"There are several typographical errors: 'coplete' should be 'complete' in the proof of Proposition 4.10; 'starightforward' should be 'straightforward' in the proof of Theorem 5.1; 'rectifibale' should be 'rectifiable' in Definition 7.22; and 'this set if of full μ-measure' should be 'this set is of full μ-measure' in Corollary 6.19.","section":"Throughout"},{"comment":"The paper repeatedly refers to 'the Lebesgue measure on M' (e.g., in Theorem 1.3, Proposition 4.7, and Section 6), but this measure is not defined. Please specify the reference measure, for example the measure induced by the fixed complete Riemannian metric h or by a volume form on M, and state the absolute-continuity condition relative to that measure.","section":"Sections 2 and 4"}],"recommendation":"major_revision","confidential_remarks":"The paper is technically rich and the main result is interesting, but the missing proof of Proposition 4.7 is load-bearing and is acknowledged by the author only as a Minkowski-space sketch with a reference to a result for a different cost. I would ask the editor to require either a complete proof of Proposition 4.7 for the cost c2 on general globally hyperbolic spacetimes or a precise reduction to existing results. The forward reference to 'the more general Theorem 6.6' in the proof of Proposition 4.7 appears to be a leftover from an earlier version and should be corrected; it currently points to a theorem with a different content. I do not see grounds for rejection, because the gap seems fixable within the paper's framework, but it is too central to accept as is."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The genuinely new result here is Theorem 1.3/6.1: local semiconvexity of pi-solutions for the c2 cost on a general globally hyperbolic spacetime. The transport-map corollaries are already known via other methods, and the author says so explicitly. That honesty is typical of the paper: it credits Figalli–Gigli for Theorem 1.2, Kell–Suhr for Proposition 1.1, and does not oversell what is inherited. The proof strategy is clear and the technical machinery—Lagrangian formulation, exponential map, local semiconcavity of c2 on I+, twist condition—is developed carefully and in detail. For a reader who works in Lorentzian optimal transport, the paper is useful and readable.\n\nThe soft spot is exactly what the stress-test note identifies: Proposition 4.7. It is stated for a general globally hyperbolic spacetime and the c2 cost, but the proof is only sketched for Minkowski space, and the forward reference to Theorem 6.6 does not cover it—that theorem is a local sup-bound statement, not a concentration-on-I+ result. The reference to [17, Theorem 2.12] is also not a direct transfer, because that theorem is proved for the cost (1.4), not for c2. This is not a cosmetic gap. Proposition 4.7 is used to ensure that the subdifferential of phi meets I+ on a set of full mu-measure, and that is what lets the proof invoke the twist condition and obtain the transport map formula in Corollary 6.2. The Minkowski proof uses Lebesgue differentiation and the affine structure of the light cone to get a uniform decrease estimate; in a curved spacetime the same estimate needs genuine geometry and is simply not supplied. So the central claim is conditional on a real, unproven transfer.\n\nBeyond that, I do not see other serious problems. The assumptions on the measures are natural, the paper does not fit parameters or hide circular reasoning, and the author is candid about which pieces are adaptations. The main theorem is plausible and the structure of the proof is sound up to this gap.\n\nThis paper deserves a serious referee, but the referee should insist on a complete proof of Proposition 4.7 in the general setting, or a clear statement of it as a separate theorem with full proof. If that can be supplied, the paper will be a solid contribution to Lorentzian optimal transport. If not, the main result is not yet established.","headline":"Local semiconvexity of pi-solutions for the squared time-separation cost is a genuine new result, but the proof leans on a central lemma that is only sketched in Minkowski and deferred; a serious referee should demand the general case.","tokens_in":47748,"tokens_out":2147,"would_cite":true,"duration_ms":23014,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["49Q22","49Q20","53C50"],"pacs":[],"model":"deepseek-v4-flash","headline":"On globally hyperbolic spacetimes, the squared time-separation cost yields locally semiconvex weak Kantorovich potentials, and hence optimal transport maps.","keywords":["optimal transport","Kantorovich potentials","semiconvexity","globally hyperbolic spacetime","Lorentzian cost","transport map","pi-solution"],"falsifier":"Find a globally hyperbolic spacetime and a pair of causally related probability measures with $\\mu$ absolutely continuous, disjoint supports, and an optimal coupling $\\pi$ that assigns positive mass to $J^+\\setminus I^+$, the lightlike boundary. Then Proposition 4.7(i) would be false in that setting, and the full-measure set $\\Omega_1$ of Theorem 6.1, together with the transport-map formula, could not exist. A concrete check would be to compute optimal couplings between measures with mass accumulating near a null cone in a nontrivial spacetime and test whether any mass is transported along lightlike geodesics.","tokens_in":46582,"feed_emoji":"🛰️","tokens_out":5039,"duration_ms":48412,"temperature":0.7,"pith_summary":"The paper studies optimal transport on a globally hyperbolic spacetime with the squared Lorentzian time-separation cost $c_2(x,y)=(\\tau(y)-\\tau(x)-d(x,y))^2$ on causally related pairs and $\\infty$ otherwise. Its central claim is that every weak Kantorovich potential, called a $\\pi$-solution, attached to an optimal coupling is locally semiconvex on an open set of full source-measure, provided the source measure is absolutely continuous and the two supports are disjoint. From this regularity the author recovers, in the classical way, the existence of an optimal transport map and the first-order formula $\\frac{\\partial c_2}{\\partial x}(x,T(x)) = -d_x\\varphi(x)$. The result matters because it extends the Riemannian squared-distance theory of optimal transport to Lorentzian geometry and explains why a well-behaved transport map still exists when the cost function is allowed to take the value infinity.","feed_headline":"Squared time-separation cost admits a spacetime transport map","feed_subtitle":"Weak Kantorovich potentials become locally semiconvex almost everywhere, giving a unique transport formula.","key_machinery":"The argument is carried by the cost function $c_2$, realized as the minimal action of the Lagrangian $L_2(v)=(d\\tau(v)-|v|_g)^2$ on future-directed causal tangent vectors. Its superlinear growth yields Theorem 1.2: a $c_2$-convex function $\\varphi$ is locally bounded on $\\Omega=\\operatorname{int}(D)$, the set $D\\setminus\\Omega$ is countably $n$-rectifiable, and $c_2$-subdifferentials over compact subsets stay relatively compact. The main theorem adds a measure-theoretic step: the set $\\Omega_0$ where the subdifferential is locally bounded away from the lightlike boundary $\\partial J^+$ has full $\\mu$-measure, and on a further full-measure subset $\\Omega_1$ the potential can be locally represented as a supremum over points with $d(x,y)\\ge\\delta$. Since $c_2$ is locally semiconcave on the timelike region $I^+$, a finite supremum of uniformly locally semiconvex functions is locally semiconvex; the twist condition, strict convexity of $L_2$ on timelike directions, then converts differentiability into the transport-map formula.","core_discovery":"On a globally hyperbolic spacetime $(M,g)$, take Borel probability measures $\\mu,\\nu$ that are causally related with $\\tau\\in L^2(\\mu)\\cap L^2(\\nu)$, disjoint supports, and $\\mu$ absolutely continuous with respect to the Lebesgue measure. The paper proves that for any optimal coupling $\\pi$ admitting a $\\pi$-solution $\\varphi$, there exists an open set $\\Omega_1\\subseteq\\Omega$ of full $\\mu$-measure on which $\\varphi$ is locally semiconvex. Consequently $\\pi$ is induced by a Borel transport map $T$ satisfying, for $\\mu$-almost every $x$, the equation $\\frac{\\partial c_2}{\\partial x}(x,T(x)) = -d_x\\varphi(x)$, with $T(x)$ uniquely determined at those points. If another optimal coupling exists, it cannot admit a $\\pi$-solution. This is the Lorentzian analogue of the Riemannian regularity theorem for squared-distance costs.","pith_inferences":["The author expects the absolute-continuity hypothesis can be weakened to '$\\mu$ gives zero mass to countably $n$-rectifiable sets'; if true, the regularity theorem would cover singular measures that avoid the rectifiable set $D\\setminus\\Omega$.","The proof singles out $c_2$ among Lorentzian costs: unlike the linear cost (1.4) or McCann's $q$-cost (1.5), this cost is superlinear and locally semiconcave only on $I^+$, suggesting that the same semiconvexity phenomenon may fail, or require substantially new ideas, for the other costs.","Because the only missing step for a fully general proof is the transfer of Proposition 4.7 from Minkowski space to general globally hyperbolic spacetimes for this cost, a direct proof of that concentration statement would complete the argument and is a natural intermediate goal to test."],"forward_implications":["If $\\mu,\\nu$ are strictly timelike, causally compact, with connected and disjoint supports, and $\\mu$ is absolutely continuous, then there is a unique optimal coupling and it is induced by a transport map (Corollary 1.5).","Every optimal coupling that admits a $\\pi$-solution is induced by a transport map $\\mu$-almost everywhere uniquely determined by $\\frac{\\partial c_2}{\\partial x}(x,T(x)) = -d_x\\varphi(x)$.","If a second optimal coupling exists alongside one with a $\\pi$-solution, the second coupling cannot admit a $\\pi$-solution, and neither can any convex combination of the two.","The set on which semiconvexity holds is open and of full $\\mu$-measure, so the weak Kantorovich potential is differentiable away from a $\\mu$-negligible set, exactly as in the Riemannian squared-distance case."],"supporting_citations":[{"why":"Supplies the Riemannian squared-distance theorem whose proof structure is adapted to the Lorentzian setting.","marker":"[14]"},{"why":"Provides the existence of $\\pi$-solutions for the cost (1.4) and the concentration result on $I^+$ that Proposition 4.7 extends to $c_2$.","marker":"[17]"},{"why":"Provides the semiconvexity and semiconcavity propositions used to pass from a local supremum representation to local semiconvexity.","marker":"[13]"},{"why":"Gives the standard differentiability theorem for locally semiconvex functions and general optimal transport background.","marker":"[29]"},{"why":"Establishes local semiconcavity of the negative Lorentzian distance on $I^+$, from which the semiconcavity of $c_2$ is derived.","marker":"[21]"},{"why":"Provides the Rockafellar construction, Kantorovich duality, and the fact that optimal couplings concentrate on $c_2$-monotone sets.","marker":"[2]"}],"fun_headline_variants":["Spacetime transport map from weak Kantorovich regularity","Weak potentials yield unique transport on spacetimes","Lorentzian regularity: transport maps from semiconvexity","Squared time-separation cost yields semiconvex potentials","Semiconvex weak potentials ensure spacetime optimal transport"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof depends on the claim that optimal couplings between such measures concentrate on the strictly timelike region $I^+$, but for this cost on a general globally hyperbolic spacetime that claim is only sketched and deferred to a reference using a different cost; if that transfer fails, the semiconvexity theorem and the transport-map conclusion lose their support.","fun_headline_variants_meta":{"raw":{"variants":["Spacetime transport map from weak Kantorovich regularity","Weak potentials yield unique transport on spacetimes","Lorentzian regularity: transport maps from semiconvexity","Squared time-separation cost yields semiconvex potentials","Semiconvex weak potentials ensure spacetime optimal transport"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000914,"raw_usage":{"total_tokens":3868,"prompt_tokens":833,"completion_tokens":3035,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":449,"completion_tokens_details":{"reasoning_tokens":2956}},"tokens_in":449,"tokens_out":3035,"duration_ms":19184,"temperature":1.0,"reasoning_tokens":2956,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T04:45:58.122534+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find a globally hyperbolic spacetime and a pair of causally related probability measures with $\\mu$ absolutely continuous, disjoint supports, and an optimal coupling $\\pi$ that assigns positive mass to $J^+\\setminus I^+$, the lightlike boundary. Then Proposition 4.7(i) would be false in that setting, and the full-measure set $\\Omega_1$ of Theorem 6.1, together with the transport-map formula, could not exist. A concrete check would be to compute optimal couplings between measures with mass accumulating near a null cone in a nontrivial spacetime and test whether any mass is transported along lightlike geodesics.","supporting_citations":[{"cited_title":"Local semiconvexity of Kantorovich potentials on non- compact manifolds.ESAIM Control Optim","cited_arxiv_id":null,"evidence_quote":"Supplies the Riemannian squared-distance theorem whose proof structure is adapted to the Lorentzian setting."},{"cited_title":"On the existence of dual solutions for Lorentzian cost func- tions.Ann","cited_arxiv_id":null,"evidence_quote":"Provides the existence of $\\pi$-solutions for the cost (1.4) and the concentration result on $I^+$ that Proposition 4.7 extends to $c_2$."},{"cited_title":"Optimal transportation on non-compact manifolds.Israel J","cited_arxiv_id":null,"evidence_quote":"Provides the semiconvexity and semiconcavity propositions used to pass from a local supremum representation to local semiconvexity."},{"cited_title":"Springer, Berlin, 2009 edition, 2008","cited_arxiv_id":null,"evidence_quote":"Gives the standard differentiability theorem for locally semiconvex functions and general optimal transport background."},{"cited_title":"Lectures in Mathematics ETH Z¨urich","cited_arxiv_id":null,"evidence_quote":"Provides the Rockafellar construction, Kantorovich duality, and the fact that optimal couplings concentrate on $c_2$-monotone sets."}],"review_version":1}