{"id":"f18ef6ce-cce9-4db4-9533-7930b4575d58","arxiv_id":"2508.19814","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"On Comb(Z^2, f_gamma) with f_gamma(z) = floor(log^gamma ||z||_inf), two independent random walks collide finitely often a.s. if gamma > 1 and infinitely often if gamma <= 1; analogous thresholds are proved for fractal and percolation bases.","lead":"Two random walks on a comb graph over a planar base collide only finitely often when the teeth grow faster than log distance, and infinitely often when they grow at or below that rate. The paper pins down this threshold for Z^2, fractal, and percolation bases, settling a question of Barlow, Peres, and Sousi.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Borel-Cantelli partition in §4.4 omits height-0 base vertices and the top third of each tooth, so the finite-collision proof does not control all collisions; the gap is likely repairable with the paper's own heat-kernel bounds.","rationale":"The reader's weakest assumption identifies exactly the load-bearing gap: the claimed exhaustion of V by the regions \\tilde Q_{k,ℓ} is false because height-0 vertices and the top third of each tooth are omitted. This is not a minor typo; it directly affects the Borel-Cantelli step in the proof of the finite-collision half of Theorem 4.1, which is the paper's principal new claim. The same flaw appears in Section 3.1. I see no other concern that is more load-bearing. The gap is likely repairable rather than fatal: the omitted regions have small expected collision counts under the paper's heat-kernel estimates, so a corrected partition should restore the summability argument. Therefore the reader's CONDITIONAL verdict is appropriate, and no change is needed.","tokens_in":37562,"tokens_out":20153,"duration_ms":206520,"concrete_test":"Augment the partition in Section 4.4 by adding R^0_k = {(z,0): ||z||∞=k} and R^top_k = {(z,h): ||z||∞=k, h > 2^{j0+1}/3}, and recompute the probability of at least one collision in each added slice using the method of Lemma 4.11 with the paper's own bounds: for each slice, estimate E[Z] and the conditional lower bound E[Z | \\tilde Z>0] (constant for the bottom slice, proportional to the top-slice height for the top slice). If the resulting sum over k converges for γ>1, the finite-collision proof can be completed by partition augmentation; if it diverges, the conclusion is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In the proof of Theorem 4.1 (Section 4.4), the regions \\tilde Q_{k,ℓ} are defined as {(u,h): ||u||∞=k, ℓ/3 ≤ h ≤ 2ℓ/3} with ℓ ∈ L(k) = {2,4,...,2^{j0}}, where 2^{j0} ≥ log^γ(k). The union over ℓ covers heights only up to 2^{j0+1}/3, which is strictly less than log^γ(k) for many k (e.g., whenever log^γ(k) is just above a power of 2). Thus the top third of each tooth is omitted, and height h=0 is never included. The text's assertion that these sets exhaust V is false, so the Borel-Cantelli sum over P(\\tilde Z_{k,ℓ}>0) does not by itself bound collisions occurring in the omitted regions. The same issue affects Section 3.1 and hence Theorem 3.2. The gap is probably repairable: for the bottom slice h=0 and the top slice, the heat-kernel bounds in Proposition 4.2 / Lemmas 4.7–4.8 give expected collision counts per shell that, after dividing by the relevant region height for the conditional lower bound, yield probability bounds summable for γ>1. But as written, the proof is incomplete.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies collisions of two independent simple random walks on comb graphs Comb(\\tilde G, f) with a planar base. The main results are phase transitions for the number of collisions: for Comb(Z^2, f_γ) with f_γ(z)=⌊log^γ(||z||_∞∨1)⌋, infinitely many collisions occur for γ≤1 and only finitely many for γ>1, answering a question of Barlow, Peres, and Sousi; analogous transitions are proved for polynomial profiles on fractal-like bases and for logarithmic profiles over supercritical percolation clusters. The finite-collision proofs rest on new heat-kernel upper bounds, most notably Proposition 4.2 for Comb(Z^2, f_γ), combined with a Borel-Cantelli argument over space-time regions.","tokens_in":37902,"tokens_out":16338,"duration_ms":159678,"significance":"If correct, the paper resolves a natural open question and gives the first finite-collision examples over planar base graphs with logarithmic tooth profiles. The heat-kernel estimates in Section 4 are a substantial technical contribution and appear, for the most part, carefully derived. The paper is clearly written and makes good use of prior work (Barlow–Peres–Sousi, Barlow, Abe) without obvious circularity. However, the Borel-Cantelli partition used to conclude the finite-collision property does not cover the vertex set as claimed; this is a load-bearing gap, though it appears repairable with the paper's own estimates. A second, smaller gap concerns the application of Lemma 4.10 in Lemma 4.8 to vertices on teeth.","major_comments":[{"comment":"The assertion that the sets \\tilde Q_{k,ℓ} exhaust V is false. For \\tilde Q_{k,ℓ} = {(u,h): ||u||_∞=k, ℓ/3≤h≤2ℓ/3} with ℓ∈{2,4,...,2^{j0}} and 2^{j0}≥log^γ(k), the union over ℓ covers, in each tooth, only heights in [2/3, 2^{j0+1}/3]. Hence h=0 is never included, and for many k an interval near the top (and near the bottom) is omitted whenever log^γ(k) > 2^{j0+1}/3 or 2^{j0}/3 > 2/3. Consequently ∑ P_0(\\tilde Z_{k,ℓ}>0)<∞ does not rule out infinitely many collisions at base vertices or in the omitted tooth segments, and the appeal to [5, Cor. 2.3] is not justified. The same defect appears in §3.1, where \\tilde Q_{k,ℓ} = {(u,h): u∈A_k, ℓ/3≤h≤2ℓ/3} with 2^{j0}≥k^γ, so base vertices and positive fractions of each tooth are again uncovered. The proof must either enlarge the partition or add a separate summability argument for these regions; the paper's own bounds in Lemmas 3.7–3.8 and 4.7–4.","section":"§4.4 (and §3.1)"},{"comment":"In the proof of Lemma 4.8, the second term after the first display is bounded using 'Lemma 4.10 to the second term'. However, Lemma 4.10 is stated and proved only for vertices x=(z,0) on the base, whereas the second term involves P_y(X_{t-s}=x) with x=(z,h) for h≥0 and y on the boundary of B(0,k/2). No on-diagonal or off-diagonal upper bound for vertices at positive height on a tooth is proved (Lemma 4.4 covers only base vertices). This leaves the short-time bound of Proposition 4.2 without a complete proof as written. The gap is likely fixable by extending Lemma 4.4 to arbitrary (z,h) or by giving a separate off-diagonal estimate, but the current derivation is not self-contained.","section":"§4.3, Lemma 4.8"}],"minor_comments":[{"comment":"The definitions of \\tilde Q_{k,ℓ} in §4.4 and of \\tilde Q^μ_{k,ℓ} in §5 write '(u,ℓ) ∈ V' instead of '(u,h) ∈ V'; this is a typo that should be corrected.","section":"§4.4 and §5"},{"comment":"In the decomposition of the B^c term, the display reads '+ Ez[...]' where the first expectation is Ex; presumably this should be Ex throughout.","section":"Lemma 4.4"},{"comment":"The lower bound E[Z_{k,ℓ}|\\tilde Z_{k,ℓ}>0] ≥ cℓ is stated without the rounding details in the definition of the tooth heights; adding a sentence clarifying that ℓ/3 and 2ℓ/3 are interpreted up to integer parts would improve readability.","section":"§3.1, Lemma 3.9"}],"recommendation":"major_revision","confidential_remarks":"The paper's central results are very likely correct, and the heat-kernel estimates are impressive. The coverage gap in the Borel-Cantelli partition is, in my view, repairable with the bounds already present in the paper, but it is not a purely cosmetic issue: the exhaustion claim is explicitly false and the conclusion of the finite-collision property does not follow without an additional argument. The second issue in Lemma 4.8 is also local and repairable. I would encourage the editor to send the paper back for a careful revision rather than reject."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper resolves a question Barlow, Peres, and Sousi left open: on Comb(Z^2, f_gamma) with f_gamma(z) = floor(log^gamma(||z||_inf)), two random walks collide infinitely often iff gamma <= 1. The finite-collision side for gamma > 1 is new, and the proof is a serious piece of work. The heat-kernel estimates in Section 4 are the core, and they look carefully done. Extending the phase transition to pre-fractal bases and to supercritical percolation clusters adds real value; the percolation part is not just a routine adaptation.\n\nThat said, there is a real gap in the Borel-Cantelli argument. The sets Qtilde_{k,ell} used in Section 4.4 do not cover the whole graph: height-0 base vertices and the top third of each tooth are omitted, so the claim that they exhaust V is false. The same issue appears in Section 3.1 and affects Theorem 3.2 as written. The good news is that the gap looks repairable: the paper's own heat-kernel bounds should give summable expected collisions for the omitted regions, as the stress-test note suggests. But the authors need to either fix the coverage claim or add a separate control for the missing pieces. Right now the proof as written is incomplete, though the main conclusion is highly plausible.\n\nMinor point: the notation in Section 4.4 has a typo in the definition of Qtilde_{k,ell} (it writes (u, ell) instead of (u,h)), which should be caught in a revision.\n\nOverall: this deserves a serious referee. The contribution is important for the random-walk subfield, the technical core seems sound, and the gap is local and likely fixable. I would recommend sending it to peer review and asking the authors to address the partition coverage carefully.","headline":"Answers an explicit open question with real new results; the Borel-Cantelli partition gap is genuine but looks repairable.","tokens_in":38389,"tokens_out":1130,"would_cite":true,"duration_ms":14692,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60J10","05C81","60J35"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves the phase transition for collisions of two random walks on Z^2 combs: teeth growing like log^gamma produce infinitely many collisions exactly when gamma is at most 1.","keywords":["random walks","collisions","comb graphs","heat kernel","phase transition","Green kernel criterion","percolation","planar graphs"],"falsifier":"Check the coverage claim: in Section 4.4 (and Section 3.1), the sets Qtilde_{k,ell} with ell in {2, 4, ..., 2^{j0}} are said to exhaust the vertex set, but they do not contain (u,0) or tooth heights above 2 ell/3. One can directly test whether the omitted sets still satisfy summable collision probabilities using the paper's own heat-kernel bounds; if that sum diverges, the finite-collision claim would need a different proof.","tokens_in":37485,"feed_emoji":"🎲","tokens_out":4907,"duration_ms":53848,"temperature":0.7,"pith_summary":"The paper aims to pin down when two independent simple random walks on a comb graph—a planar base with vertical teeth whose length grows with distance from the origin—keep colliding or stop colliding. It proves a phase transition governed by how fast the teeth grow. On a base of Z^2 with teeth of length log^gamma(radius), walks started together collide infinitely often almost surely exactly when gamma is at most 1, and only finitely often when gamma is larger than 1; this answers a question left open by Barlow, Peres, and Sousi. The same transition is established for pre-fractal planar bases, at exponent beta minus alpha, and for typical supercritical Bernoulli percolation clusters in Z^2, again at gamma equals 1. The proof's workhorse is a sharp heat-kernel upper bound on the comb, obtained by concentrating on the time the walk spends in vertical excursions between horizontal steps.","feed_headline":"Log-growing teeth flip walk collisions to finite","feed_subtitle":"New heat-kernel bounds settle the Z^2 comb question at gamma = 1.","key_machinery":"The heat-kernel bound on the comb is the key object. The walk's horizontal steps are interleaved with vertical excursions whose lengths are governed by the tooth profile. Lemma 4.4 bounds the diagonal heat kernel by showing that, in the time window [t/2, t], the number of returns to the starting tooth is controlled by the horizontal component, a random walk on Z^2. A concentration argument on the number of horizontal steps (events A and B) plus an occupation-time estimate reduce the return count to a truncated Green kernel on Z^2, which is O(1). From this diagonal bound, off-diagonal estimates and a two-scale comparison (short time versus long time) yield the sharp on-diagonal-to-off-diagona","core_discovery":"The central claim is Theorem 4.1: on Comb(Z^2, f_gamma) with f_gamma(z) = floor(log^gamma(||z||_inf or 1)), two independent simple random walks started from the same vertex collide infinitely often almost surely if gamma is at most 1, and collide only finitely often almost surely if gamma is larger than 1. The infinite-collision side was known; the paper's new contribution is the finite side for gamma greater than 1, which requires proving the heat kernel from the origin to a vertex at sup-norm distance k is at most c/t for t at least k^2 log^gamma(k) and at most c/(k^2 log^gamma(k)) for smaller t. This tuned bound makes the expected number of collisions in concentric shells summable, so Bor","pith_inferences":["The concentration method around vertical excursions looks transferable to any recurrent planar base whose heat kernel is approximately 1/t on relevant scales; the paper does not claim this full generality.","The stated exhaustion of the vertex set by the regions Qtilde in Section 4.4 (and 3.1) omits height-zero base vertices and the top third of each tooth. The paper's own heat-kernel bounds appear to make the omitted regions summable, so the proof could be repaired by adding them, but the coverage claim needs correction or augmentation.","For the pre-fractal base, the critical exponent beta minus alpha is exactly the exponent appearing in the resistance scaling, suggesting that the collision threshold is governed by resistance growth rather than volume growth; this may be a general principle worth testing on other recurrent planar graphs."],"forward_implications":["For Comb(Z^2, f_gamma), two walks starting together meet infinitely often almost surely if gamma is at most 1, and only finitely many times if gamma is larger than 1.","The same gamma equals 1 phase transition holds when the base is a typical supercritical Bernoulli percolation cluster in Z^2 containing the origin.","For pre-fractal bases satisfying the stated volume and resistance conditions, the phase transition occurs at tooth-growth exponent gamma equal to beta minus alpha.","The Green-kernel criterion of Barlow, Peres, and Sousi is sharp at leading order for these planar bases: it predicts the exact transition exponent, and the heat-kernel analysis shows the crossover where the criterion stops being effective.","By the zero-one law, the infinite or finite collision property holds from every starting vertex once it holds from one vertex."],"supporting_citations":[{"why":"Supplies the Green-kernel criterion used to prove the infinite-collision side and poses the question about Z^2 combs that this paper answers.","marker":"[5]"},{"why":"First example of recurrent graphs with the finite collision property (Comb(Z,Z)), the phenomenon this paper extends.","marker":"[19]"},{"why":"Provides the volume and resistance conditions (VG),(RG) and the heat-kernel characterization used to handle pre-fractal bases.","marker":"[4]"},{"why":"Gives the quenched heat-kernel bounds on supercritical percolation clusters that Section 5 transfers to the comb.","marker":"[2]"},{"why":"Occupation-time estimate for planar random walk used to control the second moment of returns in Lemma 2.1.","marker":"[15]"},{"why":"Standard heat-kernel estimates and random-walk tools (including the reference for the planar heat kernel) invoked throughout the proofs.","marker":"[3]"},{"why":"Effective-resistance bounds in boxes for percolation clusters, used to get the Green-kernel lower bound and the finite-collision side for C_infinity.","marker":"[1]"}],"fun_headline_variants":["Comb walk collisions go finite when teeth grow faster than log","Log-growth teeth flip comb walk collisions from infinite to finite","Phase transition on Z^2 combs: finite collisions for gamma>1 log teeth","Barlow-Peres-Sousi question answered: finite collisions on log-gamma combs"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The Borel-Cantelli argument in Theorem 4.1 and Theorem 3.2 relies on the claim that the regions Qtilde_{k,ell} cover every vertex of the comb; in fact height-0 vertices and the top third of each tooth are omitted, so unless those omitted regions are separately controlled, the summed probabilities do not by themselves prove finitely many collisions.","fun_headline_variants_meta":{"raw":{"variants":["Comb walk collisions go finite when teeth grow faster than log","Log-growth teeth flip comb walk collisions from infinite to finite","Phase transition on Z^2 combs: finite collisions for gamma>1 log teeth","Barlow-Peres-Sousi question answered: finite collisions on log-gamma combs"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000971,"raw_usage":{"total_tokens":3981,"prompt_tokens":777,"completion_tokens":3204,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":521,"completion_tokens_details":{"reasoning_tokens":3124}},"tokens_in":521,"tokens_out":3204,"duration_ms":26905,"temperature":1.0,"reasoning_tokens":3124,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T15:27:45.900407+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Check the coverage claim: in Section 4.4 (and Section 3.1), the sets Qtilde_{k,ell} with ell in {2, 4, ..., 2^{j0}} are said to exhaust the vertex set, but they do not contain (u,0) or tooth heights above 2 ell/3. One can directly test whether the omitted sets still satisfy summable collision probabilities using the paper's own heat-kernel bounds; if that sum diverges, the finite-collision claim would need a different proof.","supporting_citations":[{"cited_title":"Barlow, Y","cited_arxiv_id":null,"evidence_quote":"Supplies the Green-kernel criterion used to prove the infinite-collision side and poses the question about Z^2 combs that this paper answers."},{"cited_title":"Krishnapur and Y","cited_arxiv_id":null,"evidence_quote":"First example of recurrent graphs with the finite collision property (Comb(Z,Z)), the phenomenon this paper extends."},{"cited_title":"Barlow, T","cited_arxiv_id":null,"evidence_quote":"Provides the volume and resistance conditions (VG),(RG) and the heat-kernel characterization used to handle pre-fractal bases."},{"cited_title":"Randomwalksonsupercriticalpercolationclusters","cited_arxiv_id":null,"evidence_quote":"Gives the quenched heat-kernel bounds on supercritical percolation clusters that Section 5 transfers to the comb."},{"cited_title":"Erdős and S","cited_arxiv_id":null,"evidence_quote":"Occupation-time estimate for planar random walk used to control the second moment of returns in Lemma 2.1."},{"cited_title":"Random walks and heat kernels on graphs, volume438of London Mathematical Society Lecture Note Series","cited_arxiv_id":null,"evidence_quote":"Standard heat-kernel estimates and random-walk tools (including the reference for the planar heat kernel) invoked throughout the proofs."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Effective-resistance bounds in boxes for percolation clusters, used to get the Green-kernel lower bound and the finite-collision side for C_infinity."}],"review_version":1}