{"id":"d6db79ef-ee6b-489d-b7d9-eb03149c1c61","arxiv_id":"2611.17234","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":4,"one_line_summary":"Under spectral gap and return probability bounds, first-passage percolation on symmetric Markov kernels exhibits mean-field passage time, Gaussian hopcount, and geodesic approximated by an independent Markov chain.","lead":"This paper proves that for first-passage percolation on finite weighted graphs with symmetric Markov kernels, a uniform spectral gap and a bound on even-step return probabilities guarantee that the passage time, hopcount, and entire geodesic converge to the same limits as on the complete graph. The result covers many spatial structures including torus kernels, spread-out graphs, and finite groups, showing that the mean-field behavior is robust under spatial modulation.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek/deepseek-v4-flash","headline":"Assumption 1.2 (constant row sums) is used throughout the proofs but is not automatically satisfied for many natural spatial kernels; the paper's own Proposition 2.2 and Lemma A.1 provide normalization schemes, but these introduce a relative error that may accumulate in the path comparison and…","rationale":"The reader's weakest_assumption correctly identifies Assumption 1.2 (constant row sums) as the most restrictive condition. The paper provides normalization tools (Proposition 2.2, Lemma A.1) to handle near-constant row sums, but the error analysis in these tools is in operator norm and row-sum L^1, not in entrywise relative error. The path comparison (10) and the hopcount CLT rely on the labels in the CMJ trees evolving exactly according to PN, while the physical edge hazards are proportional to the original g_N(x,y). The paper's Theorem 1.6 and Corollary 2.3 claim universality for edge laws with a uniform relative hazard error o(1/log N), but this condition is applied to the normalized kernel PN, not to the original g_N. The step from near-constant row sums to a constant-row-sum PN introduces an entrywise error that may not be o(1/log N) uniformly, especially for kernels with spatially varying row sums that are not captured by the L^1 row-sum bound. This is a load-bearing concern because the path comparison is a central result and the paper explicitly discusses non-constant row sums as a direction for future work (citing [28]). The concrete test would settle whether the normalization suffices for the full path law or whether additional entrywise control is needed. The verdict should be CONDITIONAL: accept subject to verification that the normalization error is uniformly o(1/log N) entrywise, or that the path comparison holds under weaker entrywise control.","tokens_in":49409,"tokens_out":1157,"duration_ms":12070,"concrete_test":"Construct a family of symmetric kernels g_N on N vertices with row sums S_N(x) = 1 + delta_N(x) where max_x |delta_N(x)| = c / log N for a small constant c > 0, and with uniform spectral gap and return bound as in Assumption 1.3. Apply the normalization of Lemma A.1 to obtain a constant-row-sum kernel PN. Compute the entrywise relative error |PN(x,y) / (g_N(x,y)/S_N) - 1| for edges with non-negligible weight. If this error is not o(1/log N) uniformly over all edges, then simulate or analytically bound the total variation distance between the geodesic label distribution and the PN-Markov chain. If the distance does not tend to zero, the path comparison (10) fails for this family, showing that Assumption 1.2 cannot be relaxed to near-constant row sums without additional entrywise control.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim (Theorem 1.4) requires Assumption 1.2: the row sums S_N(x) = sum_{y != x} g_N(x,y) must be constant across vertices. The paper acknowledges this is a strong symmetry condition and provides Proposition 2.2 and Lemma A.1 to normalize kernels with small diagonal or near-constant row sums. However, the error introduced by these normalizations is controlled only in operator norm and entrywise, not in the quantities that enter the path comparison (10) and the hopcount CLT. Specifically, the reference kernel PN in (21) differs from the physical off-diagonal weights by a factor b_N(x)b_N(y) that is close to 1 but not exactly 1. The proof of Lemma 7.1 (path comparison) uses the identity (62) which relies on PN being exactly the transition matrix of the labels in the unpruned trees. If the physical edge weights are normalized by a different kernel (e.g., QN in Proposition 2.2), the labels in the CMJ trees evolve according to PN, but the actual edge hazards in the FPP model are proportional to QN(x,y) (or the original g_N(x,y)). The paper's Corollary 2.3 asserts that the conclusions of Theorem 1.6 hold with reference kernel PN and normalizer c^{1/theta} a_N, but the proof of Theorem 1.6 (Section 8) assumes that the dominating CMJ trees use PN as the label transition matrix and that the physical hazards are approximated by PN up to a relative error epsilon_N = o(1/log N). When the row sums are not constant, the normalization to constant row sums changes the off-diagonal entries by a factor that may vary across edges. The paper's Lemma A.1 controls this variation only in L^1 row sum, not entrywise. The path comparison (10) requires that the law of the geodesic labels be close in total variation to a PN-Markov chain. If the physical transition probabilities differ from PN by a spatially varying factor that is not uniformly o(1/log N) in the sup norm, the total variation distance may not vanish.","agreement_with_reader":"agree"},"referee_report":null,"author_rebuttal":null,"desk_editor":{"model":"deepseek/deepseek-v4-flash","letter":"This is a serious paper. It gives sufficient conditions—a uniform spectral gap and an O(1/N) even-step return bound—for a finite symmetric Markov kernel to produce the same mean-field passage-time, hopcount, and geodesic limits as the complete graph. The main result (Theorem 1.4) is new and the proof is rigorous. The two-source branching framework from the authors' earlier work is extended cleanly to finite kernels, and the spatial mixing estimates (Proposition 6.1) are the technical core. The applications section is extensive and honestly worked out.\n\nThe soft spot is Assumption 1.2: constant row sums. The paper acknowledges this is strong and provides normalization schemes (Proposition 2.2, Lemma A.1) that introduce a relative error controlled only in operator norm and row-sum, not entrywise. The stress-test note worries that this error may accumulate in the path comparison (10) and the hopcount CLT. On reading the paper, I think the concern is real but not fatal: the proof of Lemma 7.1 (path comparison) uses the identity (62) which relies on PN being exactly the transition matrix of the labels in the unpruned trees. If the physical edge weights are normalized by a different kernel (e.g., QN in Proposition 2.2), the labels in the CMJ trees evolve according to PN, but the actual edge hazards in the FPP model are proportional to QN(x,y) (or the original g_N(x,y)). The paper's Corollary 2.3 asserts that the conclusions of Theorem 1.6 hold with reference kernel PN and normalizer c^{1/θ} a_N, but the proof of Theorem 1.6 (Section 8) assumes that the dominating CMJ trees use PN as the label transition matrix and that the physical hazards are approximated by PN up to a relative error ε_N = o(1/log N). When the row sums are not constant, the normalization to constant row sums changes the off-diagonal entries by a factor that may vary across edges. The paper's Lemma A.1 controls this variation only in L^1 row sum, not entrywise. The path comparison (10) requires that the law of the geodesic labels be close in total variation to a PN-Markov chain. If the physical transition probabilities differ from PN by a spatially varying factor that is not uniformly o(1/log N) in the sup norm, the total variation distance may not vanish.\n\nThat said, the paper is mathematically honest: it states the assumption, gives examples where it holds, and provides a perturbation result for near-constant row sums. The central argument holds up under the stated conditions. The paper is for specialists in random metric spaces and first-passage percolation who want a unified spectral condition for mean-field behavior. It deserves a serious referee.\n\nRecommendation: send to peer review. The referee should check the path comparison step carefully, especially the transition from the normalized kernel to the physical edge weights in the non-constant row sum case. But the paper is solid and the contribution is real.","headline":"Spectral conditions for mean-field FPP: a clean, technically solid paper that does what it says, with one real soft spot (constant row sums) that the authors acknowledge but whose practical reach is narrower than the title suggests.","tokens_in":50485,"tokens_out":744,"duration_ms":9489},"pith_extraction":null,"created_at":"2026-11-13T08:52:34.850743+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}