{"id":"2579a374-862b-4f3e-be38-97862ca60cdc","arxiv_id":"2608.11198","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Loop soup percolation on Z^d, d≥5, has discrete threshold strictly above the cable system threshold 1/2; on trees, Poisson zoo percolation and susceptibility thresholds coincide, and the model is infinitesimally sensitive to Bernoulli enhancement.","lead":"This paper proves that on Z^d with d≥5, random walk loop soup percolation has different critical intensities on the discrete graph and on its metric graph counterpart, with the discrete threshold strictly above 1/2. It also introduces a sensitivity-to-enhancement framework for Poisson zoo percolation and proves threshold coincidences on regular trees.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: Theorem 3.1(a) is supported by an exact computation and standard identities; the imported α-normalization is the only soft point but no error surfaces.","rationale":"The reader's ACCEPT verdict is well supported. The central claim for Z^d depends on the one-animal mean being below 1 at intensity 1/2, and the paper establishes this through an exact Fourier/Bessel calculation plus a rigorous tail bound. The decisive numerical assertion S_5<1.936 is machine-checked with exact rational arithmetic and a reproducible script, which is independent support under the rules. The Brownian excursion extraction and the imported &tilde;α_c=1/2 are standard results; they are cited rather than proved, but the paper is not claiming them as new. The only genuine soft point is the reliance on the normalization in Eq. (3.6), exactly as the reader noted. I did not find an internal inconsistency, a missing step, or a circular argument. The limitation statements in the text (non-extendability to Z^3,Z^4, the ad hoc nature of the Z^d proof) are honest and do not affect the stated theorem. Therefore the verdict should remain ACCEPT, with the same moderate confidence; a quick re-derivation of (3.6) would raise confidence further but is not required for correctness.","tokens_in":22816,"tokens_out":50008,"duration_ms":421124,"concrete_test":"Independently re-derive Eq. (3.6) from the loop measure (2.1) via the two-point identity -log(1-G(o,v)G(v,o)/(G(o,o)G(v,v))) and confirm that the exponent is exactly α, not 2α or α/2; also check that the cable-system projection preserves the parameter α by comparing the total mass of fundamental loops in (3.1) with the discrete loop measure in (2.1). If both checks pass, the strict inequality 1/2<α_# follows directly.","verdict_should_be":"UNCHANGED","load_bearing_attack":"After reviewing the argument, no load-bearing flaw surfaces. The Z^d result rests on two imported inputs: the one-animal mean formula M(α)=∑(1-(1-q(v)^2)^α) from [13, Prop. 5.1] and the cable-system threshold &tilde;α_c(Z^d)=1/2 from [12,29]. Both are standard and internally consistent: the two-point identity underlying (3.6) is also used in the paper's own tree proof (proof of Theorem 3.1(b)), where W(u,v)=-log(1-F(u,v)F(v,u)) yields the same normalization. The numerical bound S_5<1.936 is exact and reproducible from the included Python code (Appendix A), and the domination of the cluster by a subcritical Bienaymé–Galton–Watson process is a standard exploration argument. If either imported input had a normalization shift, the strict inequality could fail, but there is no evidence of such a shift, and the margin is comfortable: S_d≤S_5<1.936 gives B_d<0.936, so even a moderate change in the α-exponent would not destroy M(1/2)<1.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies critical-parameter comparisons for long-range percolation models. In the Poisson-zoo framework, it proves on d-regular trees (d≥3) that the percolation threshold equals the susceptibility threshold under a finite-overlap condition, that subcritical clusters have exponential tails under a stronger exponential-tail condition, and that every non-degenerate zoo with finite overlap is infinitesimally sensitive to Bernoulli edge enhancement. It then applies these results to the random walk loop soup: on T_d it establishes infinitesimal sensitivity and hence strict separation of the discrete and cable-system thresholds; on Z^d with d≥5 it proves directly that the discrete susceptibility/percolation threshold lies strictly above the cable-system threshold 1/2. The Z^d proof combines the one-animal mean formula of Chang–Sapozhnikov with a Fourier computation bounding S_d ≤ S_5 and an exact rational computation plus explicit tail estimate showing S_5 < 1.936.","tokens_in":23024,"tokens_out":44430,"duration_ms":356483,"significance":"If correct, the paper answers the Cai–Ding question affirmatively for d≥5 and introduces a general enhancement-sensitivity notion for Poisson zoos, with strong tree theorems as a byproduct. The proofs are careful and largely self-contained: the finite-volume Aizenman–Newman inequality, the boundary-sprinkling exploration on trees, and the Green-function computation are all detailed, and the numerical bound is backed by exact rational arithmetic with reproducible code. The two main imported facts — the cable threshold of 1/2 from [12,29] and the one-animal mean formula from [13, Prop. 5.1] — are standard and are used consistently with the paper's own tree computation, which cross-checks the same normalization. The openness of the Z^3/Z^4 cases is explicitly acknowledged, and the paper does not overclaim beyond d≥5.","major_comments":[],"minor_comments":[{"comment":"The line 'Since 1−√(1−t)≤t due to (3.8)' is slightly confusing because (3.8) is introduced later for the continuity argument; the elementary inequality 1−√(1−t)≤t for t∈[0,1] is immediate and could be stated directly at that point.","section":"Section 3, proof of (3.7)"},{"comment":"The phrase 'As in the loop soup block argument' refers to a block argument that is not explicitly presented earlier in this paper; please clarify the reference or expand the explanation of the block exploration, since it is load-bearing for the exponential tail claim.","section":"Section 4.3, proof of Theorem 2.1(b)"},{"comment":"When stating that the Bernoulli enhancement is 'precisely a bridge field with parameter δ', it would be clearer to note explicitly that each undirected edge of the tree is identified with its unique orientation away from the root, so that the iid edge variables induce exactly the field on oriented edges required by Proposition 4.6.","section":"Section 4.2, proof of Theorem 2.4"},{"comment":"The tail bound is evaluated in floating point and compared against 0.0457 via a machine-precision assertion; although the margin is large and the comparison is safe, stating a fully rational or interval-arithmetic bound for the square-root terms would make the appendix completely exact.","section":"Appendix A"}],"recommendation":"minor_revision","confidential_remarks":"I found no load-bearing flaws. The only nontrivial external dependency is the normalization in the one-animal formula (3.6) from [13]; the paper cross-checks the same identity in its tree proof, and the numerical margin for the Z^d result is comfortable. The proof of Theorem 2.1(b) could benefit from a slightly more explicit block-exploration justification, but this is a presentation issue rather than a correctness concern."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe headline: this paper proves a real new result—on Z^d, d≥5, the random walk loop soup percolation threshold is strictly above the cable system's 1/2, for every such d. That had been known only for large d qualitatively; now it's quantitative and explicit. The tree part gives a general Poisson-zoo coincidence of percolation and susceptibility thresholds, and infinitesimal sensitivity to Bernoulli enhancement; the tree coincidence for the loop soup itself was already in Makowiec–Sapozhnikov, but the zoo version and exponential tails are new.\n\nWhat's done well: the proofs are careful. The finite-volume Aizenman–Newman inequality for zoos is a clean extension. The boundary-sprinkling exploration on trees is written out with enough detail that I could follow it. The Z^d computation reduces the threshold comparison to a Green function sum; the bound S_5<1.936 is backed by exact rational arithmetic in an appendix with code, plus a Ball–Sterbenz tail bound. That's reproducible evidence. The paper is honest about what it's not doing: the tree method doesn't port to Z^d, and the Z^d argument is ad hoc.\n\nSoft spots, in proportion: the Z^d theorem rests on two imported inputs—the cable threshold 1/2 from [12,29] and the one-animal mean formula from [13, Prop 5.1]. Both are standard, and the normalization is internally consistent, but if you don't trust those references the main inequality is unsupported. The construction of the independent Bernoulli enhancement in the cable system (the ℋ_{x,e} classes) is sketched; it's plausible but I'd want the excursion argument spelled out in the final version. These are minor-to-moderate referee requests, not flaws I can point to in the argument.\n\nThe half-plane example is a useful sanity check: it shows the cable enhancement need not lower the threshold, so the Z^d result is not an automatic monotonicity statement.\n\nWho this is for: loop soup and percolation people, especially those working on metric graph vs discrete comparisons and enhancement. It also gives a reusable framework for Poisson zoos on trees.\n\nBottom line: yes, this deserves a serious referee. The central claims look right, the numerical part is exact, and the imports are from solid sources. I'd engage with it.","headline":"Solid paper: proves strict threshold gap for loop soup on Z^d, d≥5, and gives a clean Poisson-zoo framework on trees; worth a serious referee.","tokens_in":23573,"tokens_out":2617,"would_cite":true,"duration_ms":20981,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60K35","82B43","05C81"],"pacs":[],"model":"deepseek-v4-flash","headline":"On Z^d with d≥5, the discrete random walk loop soup has its percolation threshold strictly above the metric-graph value 1/2.","keywords":["Poisson zoo","loop soup","metric graph","cable system","percolation threshold","susceptibility","Bernoulli enhancement","regular tree"],"falsifier":"Run a very high-precision simulation, or a rigorous finite-volume argument, for the discrete random walk loop soup on $Z^{5}$ at intensity $\\alpha$=1/2 and estimate the probability that the origin is in an infinite cluster: a strictly positive estimate would directly contradict alpha_c($Z^{5}$)>1/2, while zero supports it. As a cheaper check specific to the proof, recompute M(1/2) from formula (3.6) and verify whether the bound M(1/2)<1 is reproduced.","tokens_in":22603,"feed_emoji":"🌀","tokens_out":8305,"duration_ms":68039,"temperature":0.7,"pith_summary":"This paper asks whether replacing a graph by its cable system—a metric graph where each edge is a line segment—changes the percolation threshold of the random walk loop soup, and more generally of Poisson zoos. The main answer is that it does: on Z^d for d≥5, the discrete loop soup is still subcritical at intensity 1/2, while the cable-system soup has its critical point exactly at 1/2, so the discrete threshold is strictly larger. On every d-regular tree with d≥3, the paper proves the stronger statement that any positive independent Bernoulli sprinkle percolates at some subcritical intensity, and that the percolation and susceptibility thresholds coincide. A new notion, sensitivity to Bernoulli enhancement, is developed to express this, and the tree results are obtained through an exploration and boundary-sprinkling argument.","feed_headline":"Discrete loop soup percolates later than its cable version on Z^d","feed_subtitle":"For d≥5 the discrete soup's threshold sits strictly above the cable-system value 1/2.","key_machinery":"The Poisson zoo is the general model: each finite connected subgraph A (a 'graph animal') is independently present with probability 1-$e^{{-alpha mu(A)}}$, and edges are open if covered by a present animal; the loop soup is the special case where mu is the push-forward of the Markov loop measure under the trace map. The load-bearing machinery is the one-generation mean M($\\alpha$)=sum_{v≠o}(1-(1-q_G(v)^2)^$\\alpha$) with q_G(v)=G(o,v)/G(o,o), the finite-volume Aizenman–Newman differential inequality (d/dalpha) chi_hat_n($\\alpha$) ≤ C_*(mu) chi_hat_n($\\alpha$)^2 for the overlap constant C_*(mu)=sup_u sum_v W_mu(u,v), and, on trees, a boundary-sprinkling exploration in which the edge boundary of the root cluster is coupled to a Bienaymé–Galton–Watson process with offspring mean q b_mu^+($\\alpha$). The cable system contributes an independent Bernoulli enhancement built from non-fundamental metric loops, which is what makes the threshold comparison quantitative.","core_discovery":"The central result, Theorem 3.1(a), asserts that for G=Z^d, d≥5, the strict inequality 1/2 = tilde alpha_c(G) < alpha_#(G) ≤ alpha_c(G) holds: the discrete random walk loop soup's percolation threshold lies strictly above the cable-system's critical intensity 1/2, and this is detected already at the level of expected cluster size. For the d-regular tree, Theorem 2.1 shows alpha_c^mu(T_d)=alpha_#^mu(T_d) for every invariant Poisson zoo satisfying a finite-overlap condition, while Theorem 2.4 shows such a zoo is infinitesimally sensitive to Bernoulli enhancement; applied to the loop soup this yields the same threshold separation on trees. The proof on Z^d is a one-generation comparison: it computes the expected number of new vertices reached from the origin by a single occupied animal, shows via formula (3.6) that at $\\alpha$=1/2 this mean is <1, and then uses domination by a subcritical Bienaymé–Galton–Watson process to conclude finite susceptibility, hence alpha_#(Z^d)>1/2.","pith_inferences":["For Z^3 and Z^4 the finite-overlap condition fails (C_*(mu)=∞), and the paper notes the one-arm decay is slower than single-loop decay; a plausible next test is whether the strict threshold inequality still holds, perhaps with small loops playing a different role.","The sensitivity mechanism is not limited to loop soups: Theorem 2.4 applies to any non-degenerate Poisson zoo on a regular tree with finite overlap constant, so one could test the same 'any sprinkle percolates' phenomenon for Boolean models or worms on trees.","The exact S_5 computation suggests that the threshold gap on Z^d could be quantified further by sharpening return-probability bounds, and similar exact finite sums may give explicit lower bounds for alpha_#(Z^d) in all d≥5.","Interpreting the cable system as a dependent enhancement of the discrete soup, the paper's Bernoulli-bridge field construction shows that only non-fundamental loops that cross the middle of a cable are enough to create the enhancement; whether an analogous local crossing mechanism is available on other metric graphs remains open."],"forward_implications":["On Z^d for d≥5, the discrete random walk loop soup has no infinite cluster at intensity 1/2, whereas its cable-system version does; the gap between the two thresholds is at least the distance from 1/2 to alpha_#(Z^d).","On every d-regular tree, d≥3, the percolation threshold equals the susceptibility threshold for any invariant Poisson zoo with finite overlap constant, so a divergent expected cluster size is exactly the signal of percolation onset.","On such trees, every positive Bernoulli enhancement delta∈(0,1] creates percolation at some alpha strictly below the unenhanced critical value; hence the random walk loop soup on T_d is infinitesimally sensitive.","The cable-system enhancement strictly lowers the critical point on T_d, so the discrete and metric loop soup thresholds differ there as well.","Under the stronger exponential tail assumption, subcritical clusters on trees have uniformly exponential tails, so the phase transition is sharp in a quantitative sense."],"supporting_citations":[{"why":"Supplies the cable-system critical value 1/2 used as the left side of the strict inequality on Z^d.","marker":"[12]"},{"why":"Establishes that fundamental loops of the cable soup project to the discrete loop soup and that the cable system contains an independent Bernoulli enhancement; also gives critical non-percolation at 1/2.","marker":"[29]"},{"why":"Provides the one-animal mean formula M(alpha)=sum(1-(1-q(v)^2)^alpha) and the exploration-scheme idea used in both the Z^d and tree arguments.","marker":"[13]"},{"why":"Supplies the explicit return-probability bound used to control the tail in the exact S_5 calculation.","marker":"[3]"},{"why":"Gives the Markov loop measure normalization and the transience criterion behind formula (3.6).","marker":"[26]"}],"fun_headline_variants":["Loop soup on Z^d percolates above cable threshold for d≥5","Discrete loop soup threshold strictly above cable's on Z^d","For d≥5, loop soup percolation threshold > 1/2 on Z^d","Cable vs discrete loop soup: percolation gap on Z^d","Discrete loop soup needs higher intensity than cable on Z^d"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The strict inequality on Z^d rests on two imported facts — that the cable-system threshold is exactly 1/2 and that the one-animal mean M($\\alpha$) has the closed form (3.6) — and if either had a different normalization, the conclusion 1/2 < alpha_#(Z^d) would not follow from this argument.","fun_headline_variants_meta":{"raw":{"variants":["Loop soup on Z^d percolates above cable threshold for d≥5","Discrete loop soup threshold strictly above cable's on Z^d","For d≥5, loop soup percolation threshold > 1/2 on Z^d","Cable vs discrete loop soup: percolation gap on Z^d","Discrete loop soup needs higher intensity than cable on Z^d"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000803,"raw_usage":{"total_tokens":3525,"prompt_tokens":937,"completion_tokens":2588,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":553,"completion_tokens_details":{"reasoning_tokens":2485}},"tokens_in":553,"tokens_out":2588,"duration_ms":17035,"temperature":1.0,"reasoning_tokens":2485,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T14:14:53.379273+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a very high-precision simulation, or a rigorous finite-volume argument, for the discrete random walk loop soup on $Z^{5}$ at intensity $\\alpha$=1/2 and estimate the probability that the origin is in an infinite cluster: a strictly positive estimate would directly contradict alpha_c($Z^{5}$)>1/2, while zero supports it. As a cheaper check specific to the proof, recompute M(1/2) from formula (3.6) and verify whether the bound M(1/2)<1 is reproduced.","supporting_citations":[{"cited_title":"From loop clusters and random interlacement to the free field","cited_arxiv_id":"1402.0298","evidence_quote":"Establishes that fundamental loops of the cable soup project to the discrete loop soup and that the cable system contains an independent Bernoulli enhancement; also gives critical non-percolation at 1/2."}],"review_version":2}