{"id":"5dfbd2bc-b0f9-4e8b-ad15-c7c9970f08e8","arxiv_id":"2611.17066","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For stationary random measures in dimension at least 3, the asymptotic boundary transport cost, asymmetric transport cost, and mean transportation cost are all equal.","lead":"This paper proves that three different ways of measuring the cost of matching random points to a uniform background become equal in the limit of large systems. The result resolves a question from earlier work and shows that boundary effects and asymmetric constraints do not affect the asymptotic transport cost.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"No significant objection identified. The proof of Theorem 2.2 holds up under line-by-line checking; the only soft spot is the optimizer-existence step from [18,21] (flagged by the reader), which is cited within its stated hypotheses and is confirmed to cover the Poisson-to-Lebesgue application of…","rationale":"The reader's ACCEPT verdict with HIGH confidence and low correctness risk is justified. I verified the central claims step by step. Proposition 3.1 (upper bound) is sound: Lemma 3.3's monotone-convergence argument is valid because Cost_p(q) < ∞ implies ∫_{Λ_1×R^d}|x-y|^p dq_ω < ∞ almost surely, and the projection id_{Λ_n} being 1-Lipschitz onto the convex cube Λ_n gives cost control. Lemma 3.4's exponent bookkeeping is consistent: n^{d-1+α}·n^{-d+ε} = n^{ε+α-1} and ε - pα are both negative for α = 1/(p+1) and ε < p/(p+1). The later use of Lemma 3.4 does not suffer from a dimension problem: although the boundary slice argument would naively suggest a factor m·n^{d-1}, the slices (Λ_m + z) ∩ ∂Λ_n partition the boundary, so Σ_z E[λ_n((Λ_m+z)∩∂Λ_n)] ≈ E[λ_n(∂Λ_n)] and hence E[r_n(Λ_m)] ≤ n^{-d}E[λ_n(∂Λ_n)] = o(n^{-ε}). Proposition 3.6's marginal verification is sound: the second marginal follows from a truncation argument with lim_R R^{-p} · limsup n^{-d}E[W^p_{a,Λ_n}] = 0, and the first marginal uses (3.7) plus the mass transport principle on the unimodular group Z^d to upgrade ≤ η to equality, using E[η(∂Λ_1)] = E[ξ(∂Λ_1)] = 0 from the equal-intensity hypothesis. Section 4 is correct: the transport map on Q_n has Jacobian κ^{1/d}, giving density κ/κ = 1, i.e. the second marginal is L restricted to Λ_n; the remaining boundary mass is invisible to Adm_b because ∂Λ_n has zero Lebesgue measure; and the (4.3) decay uses Chebyshev-Bernstein on the Poisson variable η(Λ_n) with d ≥ 3. The asymmetric case is handled by the gluing estimate followed by a chain of inequalities that closes using Theorem 2.2 for the equivariant pair (η, L). The only genuine external dependency is the existence of the Cost_p optimizer from [18,21], which the reader correctly identifies as the weakest assumption; it is used within its stated hypotheses and is exactly the setting of [21] for the Poisson-Lebesgue constant c_∞(p,d), so I do not regard it as a defect. The minor presentational gap in Proposition 3.2's Portmanteau invocation is closed by a standard truncation whose required tail bound I verified. No change to the verdict is needed.","tokens_in":16285,"tokens_out":34064,"duration_ms":445934,"concrete_test":"Re-derive the closing inequality of Proposition 3.2 without invoking the Portmanteau theorem directly: truncate the cost as min(|x-y|^p, M)·1_{Λ_1}(x)·1_{Λ_R}(y), pass the vague limit along the subsequence realizing the liminf, then let M and R tend to infinity, showing limsup_n n^{-d}E[∫_{Λ_1 × Λ_R^c} |x-y|^p dΠ_n(x,y)] → 0 as R → ∞ via the estimate E[Π_n(Λ_1 × Λ_R^c)] ≤ (R-2)^{-p} · n^{-d}E[W^p_{a,Λ_n}(η,ξ)] and the finiteness of the limsup in (3.1). If that tail bound failed, Cost_p(Π_∞) ≤ liminf n^{-d}E[W^p_{a,Λ_n}] would need repair; my check shows it holds.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I checked the main estimates and found no genuine defect. The single most load-bearing step is Section 3.1's 'Let q be such an optimizer' (existence of a minimizer of Cost_p from [18,21], requiring a compact standard Borel space with a measurable flow, sentence [b61], plus the finiteness assumption Cost_p < infinity of [b74]). Proposition 3.1 and Lemma 3.3 rest entirely on this optimizer; if it failed, the upper bound for the boundary transport would break. This is, however, a cited published framework used within its stated hypotheses, and the specific application to a Poisson point process and Lebesgue measure (Theorem 1.2) is exactly the case covered by [21]. I also verified the delicate steps that a careful reader would worry about: Lemma 3.4's boundary-mass bound o(n^{d-epsilon}) with epsilon < p/(p+1); the L^1-vanishing of r_n(Λ_m), which follows from E[r_n(Λ_m)] ≤ n^{-d}E[λ_n(∂Λ_n)] = o(n^{-epsilon}); the marginal construction in Proposition 3.6, including the closing mass-transport-principle step; and Section 4's map T(x) = κ^{1/d}x on Q_n, whose Jacobian κ^{1/d} makes the pushforward of κL restricted to Q_n equal to L restricted to Λ_n (a scaling calculation gives density 1, so the second marginal is exactly L, not a scaled copy). The (4.3) decay via the event Θ_n and Bernstein needs α in (1/2, 1-1/d), i.e. d ≥ 3, matching Proposition 4.1's hypothesis. The one text-level gap is the closing line of Proposition 3.2: the cost |x-y|^p on Λ_1 × R^d is unbounded and non-compactly supported, so 'lower semicontinuity and the Portmanteau theorem' strictly requires a truncation plus a uniform-in-n tail bound; that bound follows from E[Π_n(Λ_1 × Λ_R^c)] ≤ (R-2)^{-p} · n^{-d}E[W^p_{a,Λ_n}(η,ξ)] together with (3.1), so the gap is presentational, not load-bearing.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek/deepseek-v4-flash","summary":"The paper establishes equality between several asymptotic limits of the transportation cost between stationary random measures with different boundary conditions, specifically the boundary transport and the asymmetric transport, and relates them to the mean transportation cost Cost_p defined via equivariant couplings. The main result (Theorem 2.2) shows that for any two jointly equivariant random measures with equal intensity, the limits of n^{-d} E[Wa^p_{Λ_n}(η,ξ)] and n^{-d} E[Wb^p_{Λ_n}(η,ξ)] exist and equal Cost_p(η,ξ). This implies Theorem 1.2: for d≥3, p≥1, the constants c(p,d), c_∞(p,d), and c_a(p,d) coincide. The proof uses a comparison with the optimizer of the mean transportation cost, constructing competitors for the boundary transport and extracting an equivariant coupling from the asymmetric transport via tightness and a subsequence argument. The paper also handles the semidiscrete matching case (Poisson to Lebesgue) by showing that the normalization by κ_n does not affect the limits.","tokens_in":16932,"tokens_out":7484,"duration_ms":363300,"significance":"If the result holds, it resolves a question raised in [15] about the equality of constants appearing in thermodynamic limits of the random optimal matching problem with different boundary conditions. The approach is general, not relying on PDE techniques, and applies to arbitrary jointly equivariant random measures, not just Poisson processes. The paper provides a clean connection between local transportation distances and the global mean transportation cost framework of Huesmann–Sturm. The result is significant for the theory of optimal transport for random measures and for Euclidean combinatorial optimization problems.","major_comments":[{"comment":"The proof of Lemma 3.3 uses the estimate E[∫_{Λ_n\\Λ_{n-√n}×Λ_n^c} |x-y|^p dq] ≤ (n^d - (n-√n)^d) E[∫_{Λ_1×R^d} |x-y|^p dq]. This relies on the equivariance of q and the fact that the cost is integrated over Λ_1×R^d. However, the region Λ_n\\Λ_{n-√n} is a thin shell of width √n, and the estimate seems to assume that the cost on each unit cell is bounded by the cost on Λ_1×R^d. But the cost |x-y|^p is not bounded, and the integral over Λ_1×R^d may be infinite if the coupling has heavy tails. The argument uses that the expectation of the cost on Λ_1×R^d is finite (since Cost_p(η,ξ)<∞), but the inequality (n^d - (n-√n)^d) E[∫_{Λ_1×R^d} ...] is valid only if the integrand is nonnegative and the measure is translation-invariant. The step appears to apply the equivariance to each unit cell in the shell, but the shell is not a union of full unit cubes; the boundary cells are cut. A more careful covering argument is needed to justify that the total cost on the shell is bounded by (surface area) times the cost on a reference cell. The authors should provide a rigorous estimate using a partition of the shell into cubes of side length 1, handling the partial cubes at the boundary. This is a load-bearing step because Lemma 3.3 is essential for Proposition 3.1.","section":"Section 3.1, Lemma 3.3","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The proof of Lemma 3.4 uses the estimate E[qn(∂Λ_n × Λ_{n-n^α})] ≤ n^{-pα} E[∫_{Λ_n×Λ_{n-n^α}} |x-y|^p dq_n]. This inequality follows from Markov's inequality: for any point (x,y) with y in Λ_{n-n^α}, the distance |x-y| is at least n^α? Actually, if x is in ∂Λ_n (the boundary layer) and y is in Λ_{n-n^α}, then the distance is at least n^α? The boundary ∂Λ_n is defined as the set of points in Λ_n that are within distance 1 of the boundary? The paper defines ∂Λ_n as the boundary of the cube, but in the context of the asymmetric transport, the optimizer q_n is a measure on Λ_n×Λ_n, and ∂Λ_n is the boundary of Λ_n (the set of points that are not in the interior). The distance between a point on the boundary and a point in Λ_{n-n^α} can be as small as 0 if the boundary point is near the corner? Actually, Λ_{n-n^α} is a smaller cube concentric with Λ_n, so the distance from any point in Λ_n\\Λ_{n-n^α} to any point in Λ_{n-n^α} is at least n^α? The minimal distance between the inner cube and the outer shell is n^α? Let's check: Λ_n = (-n/2, n/2)^d, Λ_{n-n^α} = (-(n-n^α)/2, (n-n^α)/2)^d. The distance between a point on the boundary of Λ_n (say at x=(n/2,0,...)) and a point in Λ_{n-n^α} (say at y=((n-n^α)/2,0,...)) is n^α/2. So yes, the distance is at least n^α/2. So the inequality is valid up to a constant factor. However, the term E[qn(∂Λ_n × (Λ_n\\Λ_{n-n^α}))] is bounded by (n^d - (n-n^α)^d) which is of order n^{d-1+α} (since the volume of the shell is ~ d n^{d-1} n^α). The paper writes (n^d - (n-n^α)^d) without justification that this bounds the mass of q_n on that region. The bound E[qn(∂Λ_n × (Λ_n\\Λ_{n-n^α}))] ≤ E[η(Λ_n\\Λ_{n-n^α})] = n^d - (n-n^α)^d, because q_n's first marginal is λ_n ≤ η, and λ_n(Λ_n\\Λ_{n-n^α}) ≤ η(Λ_n\\Λ_{n-n^α}). This is valid. So the estimate is fine. The comment is not a major issue; I withdraw this.","section":"Section 3.2, Lemma 3.4","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The proof that Π_∞ has marginals η and ξ uses the mass transport principle (Lemma 2.5 of [18]) to conclude that the inequality (3.8) together with equality of total mass implies equality. However, the mass transport principle is stated for Z^d-stationary random measures on R^d. The paper applies it to the function f(z1,z2) = E[Π_∞(Λ_{1,z1} × Λ_{1,z2})]. The argument requires that Π_∞ is Z^d-equivariant, which is established. But the mass transport principle also requires that the random measure is Z^d-stationary and that the measure on the diagonal is zero almost surely. The paper notes that Π_∞(∂Λ_1 × R^d) ≤ η(∂Λ_1)=0 a.s. and similarly for the other marginal, so the diagonal condition holds. However, the mass transport principle as stated in [18, Lemma 2.5] is for a function f: Z^d × Z^d → R that is translation-invariant and satisfies certain summability conditions. The paper uses it to deduce that ∑_z f(0,z) = ∑_z f(z,0). This is valid if f is nonnegative and translation-invariant. The paper defines f(z1,z2) = E[Π_∞(Λ_{1,z1} × Λ_{1,z2})]. Translation invariance follows from Z^d-equivariance of Π_∞. The summability condition is satisfied because the total mass is finite. So this step is correct. No major issue.","section":"Section 3.2, Proposition 3.6","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The proof of (4.3) uses the event Θ_n = {|η(Λ_n) - n^d| ≤ n^{αd}} with α in (1/2, 1-1/d). The condition d≥3 is needed to ensure that 1-1/d > 1/2, so that such α exists. The paper then claims that on Θ_n, n^{-d} Wb(κ_n L, L) ≲ n^p n^{pd(α-1)}. The derivation: from the estimate Wb(κ_n L, L) ≲ n^{d+p} (1-κ_n^{-1})^p (1+κ_n) for κ_n>1, and on Θ_n, |κ_n - 1| ≤ n^{αd - d} = n^{d(α-1)}. Then (1-κ_n^{-1}) ≈ (κ_n-1)/κ_n ≈ n^{d(α-1)}. So the bound becomes n^{d+p} n^{pd(α-1)} = n^{d+p + pd(α-1)} = n^{d + p(1 + d(α-1))}. But the paper writes n^p n^{pd(α-1)}? Actually, the text says: \"n^{-d}Wb(κ_nL,L) ≲ n^p n^{pd(α-1)}\". Let's check: n^{-d} times n^{d+p} n^{pd(α-1)} = n^p n^{pd(α-1)}. Yes, that matches. Then they say \"choosing α such that dα - d < -1 this term becomes of lower order.\" But the exponent in the bound is p + pd(α-1) = p(1 + d(α-1)) = p(dα - d + 1). For this to go to 0 as n→∞, we need dα - d + 1 < 0, i.e., dα - d < -1, which is equivalent to α < 1 - 1/d. Combined with α > 1/2, we need 1/2 < α < 1 - 1/d, which requires d ≥ 3. So the condition d≥3 is necessary. The paper states this correctly. However, the bound on the complement event uses n^p E[|κ_n - 1| 1_{Θ_n^c}] and then applies Cauchy-Schwarz and Bernstein inequality. The Bernstein inequality for a Poisson random variable gives P(|η(Λ_n) - n^d| ≥ t) ≤ 2 exp(-t^2/(2(n^d + t))). With t = n^{αd}, we get P(Θ_n^c) ≤ 2 exp(-c n^{2αd - d}) = 2 exp(-c n^{d(2α-1)}). Since α > 1/2, this decays exponentially. Then E[|κ_n - 1| 1_{Θ_n^c}] ≤ sqrt(E[|κ_n-1|^2]) sqrt(P(Θ_n^c)) ≤ n^{-d/2} exp(-c n^{d(2α-1)}/2). So the term is negligible. The paper's estimate is correct. The only subtlety is that the bound for Wb(κ_n L, L) on Θ_n uses the inequality (1-κ_n^{-1})^p ≤ (κ_n-1)^p (since κ_n≥1 on Θ_n? Actually, on Θ_n, κ_n may be less than 1 as well; the paper treats κ_n>1 and κ_n<1 separately. For κ_n<1, the bound is n^{d+p} (1-κ_n)^p. On Θ_n, |κ_n-1| ≤ n^{d(α-1)}, so the bound is similar. So the estimate holds. The proof is sound.","section":"Section 4, Proposition 4.1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper constructs a competitor for Wa_{Λ_n}(η,L) by gluing an optimizer of Wa_{Λ_n}(η,κ_n L) and an optimizer of Wb_{Λ_n}(κ_n L, L). The gluing lemma for boundary transport is cited from [13]. However, the gluing lemma in [13] is for the boundary transport distance Wb, which is defined for measures with possibly different masses. The paper uses it to glue π_1 (a measure on Λ_n×Λ_n with marginals λ≤η and κ_n L) and γ (a measure on Λ_n×Λ_n with marginals κ_n L and L, but γ is an optimizer of Wb_{Λ_n}(κ_n L, L) and may have mass on ∂Λ_n×Λ_n and Λ_n×∂Λ_n). The paper then defines π = glue(π_1, γ) and then projects to get a competitor for Wa. The construction seems plausible, but the paper should provide more details on how the gluing is performed and why the resulting measure is admissible for Wa. In particular, the gluing lemma for boundary transport typically requires that the two measures have the same total mass on the domain, but here π_1 has second marginal κ_n L (restricted to Λ_n) and γ has first marginal κ_n L (restricted to Λ_n). The gluing should produce a measure on Λ_n×Λ_n×Λ_n with the appropriate marginals. Then the projection onto the first and third coordinates gives a measure on Λ_n×Λ_n whose first marginal is λ (≤ η) and second marginal is L (since γ's second marginal is L). However, the paper adds γ⌞(∂Λ_n×Λ_n) separately to account for mass that was transported to the boundary in γ. This seems to handle the boundary mass correctly. But the argument is sketchy; a more rigorous justification would be welcome. Since this is a minor point in the proof of Proposition 4.1, and the main idea is clear, I do not consider it a major issue.","section":"Section 4, asymmetric case","authors_acknowledge":null,"authors_acknowledge_p":null}],"minor_comments":[{"comment":"The abstract states 'We establish equality between several asymptotic limits of the transportation cost between stationary random measures, with respect to the Euclidean distance cost raised to the p-th power, with different boundary conditions.' This is accurate but could be more specific about the main result (Theorem 2.2).","section":"Abstract","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The statement of Theorem 1.1 uses notation λ^a for the absolutely continuous part of λ, but this is not defined in the text. It is standard but should be defined for completeness.","section":"Section 1, Theorem 1.1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The definition of Wa_{Λ}^p(μ,ν) in (2.2) is given as inf_{λ≤μ} W_p^Λ(λ, ν⌞Λ). The paper notes that λ is allowed to charge mass to the boundary ∂Λ, but then W_p^Λ(λ, ν⌞Λ) is defined only if λ(Λ)=ν(Λ). The paper should clarify that λ is a measure on Λ (or on R^d) such that λ≤μ on Λ and λ(Λ)=ν(Λ). The text says 'λ is allowed to charge mass to the boundary ∂Λ, as for the r.h.s. to be well-defined λ(Λ)=ν(Λ).' This is clear enough.","section":"Section 2.1, definition of Wa","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The definition of c_∞(p,d) in (2.5) uses η as a 1-homogeneous Poisson point process and ξ=L. The paper should explicitly state that η is a Poisson point process with intensity 1.","section":"Section 2.2, line after (2.4)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The proof of Lemma 3.3 uses the estimate (n^d - (n-√n)^d) which is of order n^{d-1/2}. The paper writes 'which is of order n^{d-1/2}'. This is correct, but the constant should be checked: (n^d - (n-√n)^d) = n^d (1 - (1-1/√n)^d) ≈ n^d * d/√n = d n^{d-1/2}. So the order is n^{d-1/2}.","section":"Section 3.1, Lemma 3.3","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The proof uses the notation A(n) ≲ B(n) to mean there exists a universal constant C>0 such that A(n) ≤ C B(n) for all n. This is fine.","section":"Section 3.2, Lemma 3.4","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper defines Q_n as a measure on R^d × R^d × Ω. The construction is standard but could be clarified: Q_n(A×B×Ω') = ∫_{Ω'} Π_n^ω(A×B) dP(ω).","section":"Section 3.2, after (3.3)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The proof of tightness uses the estimate E[Π_n(R^d × Λ_m)] ≤ m^d (n+2)^d / n^d? The paper writes '≤ m^d (n+2)^d' but then divides by n^d? Actually, the text: '≤ m^d n+2 n^d ≲ m^d'. This seems to be a typo: it should be '≤ m^d ( (n+2)/n )^d ≲ m^d'. The expression 'n+2 n^d' is ambiguous. Please rewrite.","section":"Section 3.2, Lemma 3.5","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"In the estimate for the second marginal, the paper writes '≲ m^d ∥φ∥_∞ lim_{R→∞} R^{-p} limsup ...'. The exponent -p seems to come from the assumption that |x-y|^p ≥ R^p for x in Λ_R^c and y in Λ_m. But the coupling Π_n may have mass where x is far and y is close, and the cost is large. The argument uses that ∫_{Λ_R^c × Λ_m} |x-y|^p dΠ_n ≤ R^{-p} ∫_{R^d × Λ_m} |x-y|^p dΠ_n? Actually, they use the inequality: for x in Λ_R^c, |x-y| ≥ R - m/2? The paper uses R^{-p} as a bound for 1/|x-y|^p? Let's check: they write '≲ m^d ∥φ∥_∞ lim_{R→∞} R^{-p} limsup_{n→∞} E[∫_{R^d × Λ_1} |x-y|^p dΠ_n(x,y)]'. This seems to assume that for x in Λ_R^c, |x-y| ≥ R (up to a constant), so that 1 ≤ |x-y|^p / R^p. Then the integral over Λ_R^c × Λ_m is bounded by R^{-p} times the integral of |x-y|^p over R^d × Λ_m. But the integral over R^d × Λ_m is not necessarily finite? However, the limsup is finite because it equals limsup n^{-d} E[Wa^p_{Λ_n}(η,ξ)] which is assumed finite. So the argument is valid. The paper should make this step clearer.","section":"Section 3.2, proof of Proposition 3.6","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"In the estimate for Wb_{Λ_n}(κ_n L, L) when κ_n>1, the paper writes '≲ n^{d+p} (1-κ_n^{-1})^p (1+κ_n)'. Then they simplify to 'n^{d+p} (1-κ_n^{-1})^p (1+κ_n)'. Later they write '≤ n^{d+p} (1-κ_n^{-1})^p (1+κ_n)'. The constant factors are omitted, which is fine for asymptotic statements. However, the final bound 'n^{d+p} (1-κ_n^{-1})^p (1+κ_n)' seems to have an extra factor (1+κ_n) that is not present in the earlier derivation. The derivation: from the estimate Wb ≤ κ_n (|1-κ_n^{1/d}|^p ∫_{Q_n} |x|^p dx + ∫_{Λ_n\\Q_n} dist(x,∂Λ_n)^p dx). Then they bound ∫_{Q_n} |x|^p dx ≲ (n/2 κ_n^{-1/d})^{d+p} = n^{d+p} κ_n^{-1-p/d}. And ∫_{Λ_n\\Q_n} dist^p ≲ n^{d-1} (n/2 (1-κ_n^{-1/d}))^{p+1} = n^{d+p} (1-κ_n^{-1/d})^{p+1}. Then they combine to get Wb ≲ κ_n ( n^{d+p} κ_n^{-1-p/d} |1-κ_n^{1/d}|^p + n^{d+p} (1-κ_n^{-1/d})^{p+1} ). Then they simplify using |1-κ_n^{1/d}| ≈ |1-κ_n^{-1/d}|? Actually, they write '= n^{d+p} (1-κ_n^{-1/d})^p [1 + κ_n (1-κ_n^{-1/d})]'. This seems to assume that |1-κ_n^{1/d}|^p = κ_n^{-p/d} |1-κ_n^{-1/d}|^p? Let's check: 1-κ_n^{1/d} = κ_n^{1/d} (κ_n^{-1/d} - 1) = -κ_n^{1/d} (1-κ_n^{-1/d}). So |1-κ_n^{1/d}|^p = κ_n^{p/d} |1-κ_n^{-1/d}|^p. Then the first term becomes κ_n * n^{d+p} κ_n^{-1-p/d} * κ_n^{p/d} |1-κ_n^{-1/d}|^p = n^{d+p} |1-κ_n^{-1/d}|^p. The second term is κ_n * n^{d+p} (1-κ_n^{-1/d})^{p+1} = n^{d+p} κ_n (1-κ_n^{-1/d})^{p+1}. So the sum is n^{d+p} (1-κ_n^{-1/d})^p (1 + κ_n (1-κ_n^{-1/d})). Then they bound (1 + κ_n (1-κ_n^{-1/d})) ≤ (1+κ_n) because (1-κ_n^{-1/d}) ≤ 1. So the final bound is n^{d+p} (1-κ_n^{-1/d})^p (1+κ_n). Then they further bound (1-κ_n^{-1/d})^p ≤ (1-κ_n^{-1})^p because 1-κ_n^{-1/d} ≤ 1-κ_n^{-1} for κ_n>1? Actually, for κ_n>1, κ_n^{-1/d} > κ_n^{-1}, so 1-κ_n^{-1/d} < 1-κ_n^{-1}. So (1-κ_n^{-1/d})^p ≤ (1-κ_n^{-1})^p. So the bound becomes n^{d+p} (1-κ_n^{-1})^p (1+κ_n). This is what the paper writes. The derivation is correct but the steps are compressed. It would be helpful to include more intermediate steps for clarity.","section":"Section 4, Proposition 4.1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper writes 'where the last inequality is due to another application of (4.2), Proposition 3.1 and the fact that c_∞(p,d)<∞ for d≥3 can be found in [21].' The reference [21] is Huesmann–Sturm 2013. This is acceptable.","section":"Section 4, after (4.4)","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper constructs a competitor for Wa by gluing and then uses Young's inequality. The estimate is: Wa_{Λ_n}(η,L) ≤ (1+ε) Wa_{Λ_n}(η,κ_n L) + C ε^{p-1} Wb_{Λ_n}(κ_n L, L). This is analogous to (4.1). The paper should justify the constant C ε^{p-1} (or C/ε^{p-1}) from Young's inequality. Typically, Young's inequality gives ab ≤ ε a^p + C_ε b^q with 1/p+1/q=1. Here they use |x-y|^p ≤ (1+ε) |x-y|^p + C_ε |y-z|^p. The constant C_ε is of order ε^{-(p-1)}? Actually, the standard inequality is (a+b)^p ≤ (1+ε) a^p + C_p ε^{-(p-1)} b^p for p≥1. So the constant should be C ε^{-(p-1)} rather than C ε^{p-1}. The paper writes C ε^{p-1}, which would be small when ε is small, but the inequality requires a large constant when ε is small. This appears to be a typo. In (4.1) and (4.2), the paper writes C/ε^{p-1}? Actually, the text says 'C ε^{p-1}' in both (4.1) and (4.2). Let's check: 'Wb_{Λ_n}(η,L) ≤ (1+ε) Wb_{Λ_n}(η,κ_n L) + C ε^{p-1} Wb_{Λ_n}(κ_n L,L).' This is likely a mistake: the constant should be C ε^{-(p-1)}. The same issue appears in the asymmetric case. The authors should correct this. However, the subsequent argument sends ε→0, and if the constant is ε^{p-1}, then the term vanishes as ε→0, which is even better. But the inequality (a+b)^p ≤ (1+ε) a^p + C_p ε^{-(p-1)} b^p is standard; using ε^{p-1} would give a weaker bound (since ε^{p-1} is small, the term is smaller, but the inequality may not hold). The authors should verify the correct constant. This is a minor issue because the argument only requires that the term goes to 0 as ε→0, which holds for both forms if the constant is finite. However, if the inequality is wrong, the proof might be invalid. I recommend the authors check and correct the constant.","section":"Section 4, asymmetric case","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper writes 'This estimate is sufficient to deduce the claim. Indeed, take the limit in n and then send ε to 0 in the first step, we have by (2.3) in the third estimate, Proposition 4.1 in the first equality and Theorem 2.2 in the last two equalities'. The chain of inequalities is a bit confusing. It would be clearer to write: liminf n^{-d} E[Wa(η,L)] ≤ liminf n^{-d} E[Wa(η,κ_n L)] ≤ limsup n^{-d} E[Wa(η,κ_n L)] ≤ limsup n^{-d} E[Wb(η,κ_n L)] = limsup n^{-d} E[Wb(η,L)] = c_∞(p,d) = liminf n^{-d} E[Wa(η,L)], hence all limits exist and equal c_∞. This is what the paper does, but the presentation is a bit messy.","section":"Section 4, asymmetric case","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper uses the notation Λ_n for the cube (-n/2, n/2)^d. In Lemma 3.4, the notation ∂Λ_n is used for the boundary layer? Actually, the paper defines ∂Λ_n as the boundary of Λ_n, but in the context of the asymmetric transport, the optimizer q_n is a measure on Λ_n × Λ_n, and the mass on ∂Λ_n × Λ_n refers to mass that is transported from the boundary? The paper should clarify that ∂Λ_n denotes the set of points in Λ_n that are within distance 1 of the boundary? In the definition of Wa, the measure λ is allowed to charge mass to the boundary ∂Λ, but the coupling is on Λ×Λ. The paper's Lemma 3.4 uses q_n(∂Λ_n × Λ_n) where ∂Λ_n is the boundary of Λ_n. But if ∂Λ_n is the topological boundary, it has Lebesgue measure zero, and the Poisson process almost surely has no points there. So q_n(∂Λ_n × Λ_n) would be zero almost surely. The paper must be using a thickened boundary, e.g., the set of points within distance 1 of the boundary. The text says '∂Λ_n' but likely means the 'boundary layer' of width 1. This should be clarified. In the proof, they estimate E[qn(∂Λ_n × (Λ_n\\Λ_{n-n^α}))] ≤ (n^d - (n-n^α)^d), which is the volume of the shell, suggesting that ∂Λ_n is a set of positive measure. So it is indeed a thickened boundary. The paper should define this explicitly.","section":"General","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper writes 'For α = 1/(p+1), we can estimate ...'. The choice α = 1/(p+1) is used to balance the exponents. The estimate uses n^{-pα} times the cost. This is fine.","section":"Section 3.2, Lemma 3.4","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper writes 'Due to tightness, the sequence {Q_n}_n is vaguely compact and thus admits accumulation points.' This is correct because the space of Radon measures on a locally compact separable metric space is vaguely compact for bounded sets. The paper assumes Ω is compact, so R^d × R^d × Ω is locally compact. The tightness condition sup_n Q_n(K) < ∞ for all compact K ensures vague compactness. The paper verifies this for sets of the form A×Ω with A compact in R^d×R^d. This is sufficient because any compact set in R^d×R^d×Ω is contained in a product of a compact set in R^d×R^d and Ω (since Ω is compact). So the argument is valid.","section":"Section 3.2, after Lemma 3.5","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper uses the mass transport principle [18, Lemma 2.5] to conclude that ∑_z f(0,z) = ∑_z f(z,0). The reference [18] is Huesmann (2016). The lemma is stated for a function f: Z^d × Z^d → R that is translation-invariant and satisfies ∑_z f(0,z) < ∞. The paper defines f(z1,z2) = E[Π_∞(Λ_{1,z1} × Λ_{1,z2})]. Translation invariance follows from Z^d-equivariance. The summability condition holds because ∑_z f(0,z) = E[Π_∞(Λ_1 × R^d)] = E[η(Λ_1)] = 1 < ∞. So the application is valid.","section":"Section 3.2, proof of Proposition 3.6","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper writes 'The main obstruction to arguing analogously in the asymmetric case is the fact that Wa_{Λ_n} does not define a distance and it is therefore unclear how to deduce the analogue of (4.1) and (4.2).' The paper then provides a different argument using gluing. This is a strength of the paper.","section":"Section 4, Proposition 4.1","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"The paper notes that under suitable concentration assumptions, the proof generalizes to equivariant measures η, ξ. This is a remark, not a claim.","section":"Section 4, last paragraph","authors_acknowledge":null,"authors_acknowledge_p":null},{"comment":"Reference [15] is listed as 'M. Goldman, M. Huesmann, and D. Trevisan, Asymptotics for Random Quadratic Transportation Costs. arxiv preprint arXiv:2409.08612, (2026).' This is a preprint; the paper should note if it has been accepted or published.","section":"References","authors_acknowledge":null,"authors_acknowledge_p":null}],"recommendation":"minor_revision"},"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"You should know two things: the main result is correct as far as I can check, and the only genuinely soft spot is presentational, not mathematical. The paper proves c(p,d)=c_infinity(p,d)=c_a(p,d) for d>=3 and all p>=1, answering the p=2 question from [15] and extending it to all powers and to the asymmetric constant. The vehicle is Theorem 2.2: for jointly equivariant random measures of equal intensity, the boundary and asymmetric local limits both exist and equal Cost_p. That is a clean general statement, and the proof reaches it without PDE regularity, by comparing against the infinite-volume optimizer from [18,21].\n\nThe proof is careful. Lemma 3.3's o(n^d) boundary-leakage estimate, Lemma 3.4's o(n^{d-epsilon}) boundary-mass bound, and the marginal construction in Proposition 3.6 all check out. Section 4's normalization step, replacing Lebesgue intensity by kappa_n L, is where I expected a hidden bug, but the affine map T(x)=kappa^{1/d}x on Q_n has the right Jacobian, and the Bernstein argument correctly forces d>=3, matching Proposition 4.1. The authors also state plainly [b31] that in full generality the four constants cannot agree, as they do not for a stationarised lattice, so the scope is honest.\n\nSoft spots, in proportion. The existence of the optimizer q for Cost_p is load-bearing and imported from [18,21]; it needs a compact standard Borel space with a measurable flow and finite Cost_p. That is a published framework used within its stated hypotheses, and the Poisson-to-Lebesgue application is exactly the case [21] covers, so I do not count it as a defect. A referee should still confirm that Theorem 2.2's assumptions are stated so readers know what is being assumed. The one text-level gap is the closing line of Proposition 3.2: lower semicontinuity plus Portmanteau is invoked for the unbounded cost |x-y|^p on Lambda_1 x R^d. Strictly, one needs a truncation and a uniform-in-n tail bound; the tail bound follows from (3.1), so the fix is routine. That is all I found.\n\nThe paper resolves an open question in a small but active area. It is not a paradigm shift, but it is a genuinely useful result with sound proofs and an honest citation pattern. I would send it to a serious referee for a probability or mathematical physics journal, and I would cite it if I worked on random matching.","headline":"The main theorem holds up: the equality of the three matching constants is proved cleanly, with only a routine presentational gap in Proposition 3.2.","tokens_in":17531,"tokens_out":3180,"duration_ms":23566},"pith_extraction":null,"created_at":"2026-11-13T10:35:34.412172+00:00","model_set":{"reader":"deepseek/deepseek-v4-flash"},"pipeline_version":"pith-pipeline@v1.4.1-daily-deepseek-or-medium","input_kind":null,"input_chars":null,"input_total_chars":null,"passes_missing":null,"published_standing":null,"published_confidence":null,"objection_audit":null,"run_record":null,"basis":{"kind":"not_recorded","source":null,"chars_given":null,"chars_total":null},"receipt":null,"falsifier":null,"supporting_citations":[],"lane_accuracy":{"status":"not yet measured","lane":{"model":"deepseek/deepseek-v4-flash","input_kind":null},"url":"https://pith.science/accuracy"},"review_version":1}