(Non-)coincidence of critical parameters for Poisson Zoos and Loop Soup Percolation on $\mathbb{Z^d}$, $d > 4$, and $\mathbb{T}_d$,$ d \ge 3$

Alexander Drewitz; Andreas Klippel; Christian M\"onch; Volker Betz

(Non-)coincidence of critical parameters for Poisson Zoos and Loop Soup Percolation on $\mathbb{Z^d}$, $d > 4$, and $\mathbb{T}_d$,$ d \ge 3$

T0 review · 0 major / 4 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read On Z^d with d≥5, the discrete random walk loop soup has its percolation threshold strictly above the metric-graph value 1/2.

desk verdict 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. read the letter →

arxiv 2608.11198 v1 pith:WHMWC7FW submitted 2026-08-11 math.PR

Volker Betz , Alexander Drewitz , Andreas Klippel , Christian Mönch This is my paper · ORCID

classification math.PR MSC 60K3582B4305C81

keywords PoissonzooloopsoupmetricgraphcablesystempercolationthresholdsusceptibilityBernoullienhancementregulartree

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

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.

What carries the argument

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.

What would settle it

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.

Watch

Extended reading notes

Core claim

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.

Load-bearing premise

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.

Editorial extensions

If this is right

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.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

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.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Dear colleague,

The 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.

What'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.

Soft 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.

The 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.

Who 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.

Bottom 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.

Referee Report

0 major / 4 minor

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.

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.

minor comments (4)

[Section 3, proof of (3.7)] 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 4.3, proof of Theorem 2.1(b)] 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 4.2, proof of Theorem 2.4] 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.
[Appendix A] 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.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the Z^d threshold separation rests on independent external inputs and an exact computation, and the tree theorems are proved by direct construction.

full rationale

The paper's central claims are not equivalent to their inputs by construction. Theorem 3.1(a) is derived from (i) the externally established cable-system threshold tilde alpha_c(Z^d)=1/2 (cited to [12,29]), and (ii) the one-animal mean formula M(alpha)=sum_{v != o}(1-(1-q(v)^2)^alpha) of [13, Prop. 5.1]. Neither input restates the conclusion: (i) gives only the left equality, while the strict inequality 1/2 < alpha_#(Z^d) follows from the paper's own exact estimate M(1/2)<1, obtained from S_d <= S_5 < 1.936. The bound S_5 is reproduced in Appendix A by exact rational arithmetic and an analytic Ball--Sterbenz tail bound, with the Python code included; it is not fitted or defined in terms of the target inequality. The M(alpha) formula is also used inside the paper in the tree proof through the two-point identity W_mu(u,v)=-log(1-F(u,v)F(v,u)), and the same normalization appears in [13]; the paper does not redefine q(v) to force M(1/2)<1. The tree results (Theorems 2.1, 2.4, 3.1(b)) are proved by a direct boundary-sprinkling exploration: Lemma 4.4 computes the expected boundary size, Lemma 4.5 constructs an i.i.d. Bernoulli bridge field from disjoint animal classes, and Proposition 4.6 couples the exploration to a Bienayme--Galton--Watson process. No parameter is fitted to the target critical value, and no prediction is a renamed fit. The self-citations [6,7,23] appear only as contextual examples of strict monotonicity results in Remark 2.3(b); they are not load-bearing. The only soft point is the imported normalization of the loop intensity alpha between the discrete soup, the cable-system fundamental loops, and [13, Prop. 5.1]; the paper states the identification of fundamental loops with the discrete soup (following [29]) and uses a consistent q(v)=G(o,v)/G(o,o). A normalization mismatch would be a correctness issue, not circularity, and no evidence of one is present. Therefore no circular step can be exhibited, and the paper is self-contained against external benchmarks for the purposes of this review.

Assumptions & free parameters 0 free parameters · 8 assumptions · 0 invented entities

The paper's central claims rest on standard probability results plus explicit structural assumptions (2.5), (2.6), and non-degeneracy. The Z^d comparison uses the independently established cable threshold 1/2 and formula (3.6) from [13]. No numerical parameters are fitted; all constants are either fixed by the model or explicit bounds.

assumptions (8)

domain assumption Metric graph loop soup percolation threshold is tilde α_c=1/2 for Z^d (d≥3) and T_d (d≥3).
Imported from [12,29]; used as the external benchmark in Theorem 3.1.
domain assumption One-animal mean formula M(α)=∑_{v≠o}(1-(1-q(v)^2)^α) with q(v)=G(o,v)/G(o,o) (from Proposition 5.1 of [13]).
Foundation of the Z^d strict-inequality proof in Theorem 3.1(a); not reproved.
domain assumption Finite-overlap condition C_*(μ)<∞ in (2.5).
Stated assumption for Theorems 2.1(a), 2.1(b) and 2.4; verified for the loop soup on T_d in the proof of Theorem 3.1(b).
domain assumption Exponential tail condition (2.6) on the animal-size distribution.
Assumption for the subcritical exponential tail statement, Theorem 2.1(b).
domain assumption Non-degeneracy: μ(A)>0 for some animal A with at least one edge.
Required for positive Bernoulli bridge parameters in Lemmas 4.5 and Theorem 2.4; degenerate zoos are handled separately.
standard math BK inequality for increasing events on disjoint animal-coordinate witnesses.
Used in the Aizenman-Newman differential inequality (Lemma 4.2); standard extension to independent Poisson variable coordinates.
standard math Simple random walk on Z^d is transient for d≥5 with finite Green function.
Used in Section 3 to define q(v), apply (3.6), and prove M(1/2)<1.
standard math Brownian loop measure and heat kernel on metric graphs (cable systems) have the stated properties, including the projection of fundamental loops.
Background for the cable system construction (3.1) and the Bernoulli enhancement extraction in Section 3.

how reviews work

0 comments

Cite this review

Pith. "Pith review of (Non-)coincidence of critical parameters for Poisson Zoos and Loop Soup Percolation on $\mathbb{Z^d}$, $d > 4$, and $\mathbb{T}_d$,$ d \ge 3$." pith.science (2026). https://pith.science/paper/WHMWC7FW

@misc{pith2026260811198,
  author       = {Pith},
  title        = {Pith review of: (Non-)coincidence of critical parameters for Poisson Zoos and Loop Soup Percolation on $\mathbbZ^d$, $d > 4$, and $\mathbbT_d$,$ d \ge 3$},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WHMWC7FW}},
  note         = {Machine review of arXiv:2608.11198}
}

abstract

In this article we investigate the (non)-coincidence of critical parameters for various related percolation problems. More precisely, for the random walk loop soup we show that on $\mathbb Z^d$, $d\ge 5$, the critical parameters for the percolation problems differ on the discrete graph and the respective metric graph. Moreover, on trees we deduce an analogous statement as well as the coincidence of the critical parameters for percolation and susceptibility for a more general class of percolation problems, the so-called Poisson zoo. Along the way we develop the useful notion of sensitivity to Bernoulli enhancements of such percolation problems with long range correlations, which builds on previously developed enhancement ideas.

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

12 extracted references · 9 canonical work pages

[28]

Domination by product measures

T. M. Liggett, R. H. Schonmann, and A. M. Stacey. “Domination by product measures”.Ann. Probab.25:1 (1997), pages 71–95.doi:10.1214/aop/1024404279(cited on page 5)

arXiv 1997
[29]

From loop clusters and random interlacement to the free field

T. Lupu. “From loop clusters and random interlacements to the free field”.Ann. Probab.44:3 (2016), pages 2117–2146.doi: 10.1214/15-AOP1019. arXiv:1402.0298 [math.PR] (cited on pages 1, 2, 7–10)

work page Pith review arXiv 2016
[30]

Loop percolation on discrete half-plane

T. Lupu. “Loop percolation on discrete half-plane”.Electron. Commun. Probab.21 (2016), Paper No. 30, 9.doi:10.1214/16-ECP4571. arXiv:1408.1045 [math.PR](cited on page 10)

work page Pith review arXiv 2016
[31]

Critical curve of loop percolation on the $d$-regular tree

L. Makowiec and A. Sapozhnikov.Critical curve of loop percolation on the𝑑-regular tree. 2026. arXiv:2606.27520 [math.PR](cited on page 6)

work page Pith review arXiv 2026
[32]

Martineau, R

S. Martineau, R. Poudevigne, and P. Rax.Stochastic domination and lifts of random variables in percolation theory. Preprint, arXiv:2504.02427 [math.PR] (2026). 2026 (cited on page 7)

work page arXiv 2026
[33]

Critical parameters for loop and Bernoulli percolation

P. Mühlbacher. “Critical parameters for loop and Bernoulli percolation”. English.ALEA, Lat. Am. J. Probab. Math. Stat.18:1 (2021), pages 289–308 (cited on page 7)

work page 2021
[34]

Nonamenable Poisson zoo

G. Pete and S. Rokob.Nonamenable Poisson zoo. Preprint, arXiv:2505.07145 [math.PR]. 2025. arXiv:2505.07145 [math.PR](cited on pages 1, 3)

work page Pith review arXiv 2025
[35]

Percolation of worms

B. Ráth and S. Rokob. “Percolation of worms”.Stochastic Processes Appl.152 (2022), pages 233– 288.doi:10.1016/j.spa.2022.06.018. arXiv:2107.03259 [math.PR](cited on page 3)

work page Pith review arXiv 2022

Show all 12 references

[36]

Phasetransitionandlevel-setpercolationfortheGaussian free field

P.-F.RodriguezandA. -S.Sznitman.“Phasetransitionandlevel-setpercolationfortheGaussian free field”.Comm. Math. Phys.320:2 (2013), pages 571–601.doi:10.1007/s00220-012-1649-y (cited on page 1)

2013 doi
[37]

Essential enhancements in Abelian networks: Continuity and uniform strict mono- tonicity

L. Taggi. “Essential enhancements in Abelian networks: Continuity and uniform strict mono- tonicity”.The Annals of Probability51:6 (2023), pages 2243–2264.doi:10.1214/23-AOP1647 (cited on page 7)

2023 doi
[38]

1+𝑜(1) asymptotics for loop percolation in five and higher dimensions

Q. Vogel. “1+𝑜(1) asymptotics for loop percolation in five and higher dimensions”.Int. Math. Res. Not. IMRN9 (2026), Paper No. rnag084, 18.doi:10.1093/imrn/rnag084(cited on pages 2, 10)

2026 doi
[39]

On clusters of Brownian loops in𝑑 dimensions

W. Werner. “On clusters of Brownian loops in𝑑 dimensions”. In:In and out of equilibrium 3. Celebrating Vladas Sidoravicius. Volume 77. Progr. Probab. Birkhäuser/Springer, Cham, 2021, pages 797–817.doi:10.1007/978-3-030-60754-8\_33(cited on pages 1, 2). 29

2021 doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.

[1] [28]

Domination by product measures

T. M. Liggett, R. H. Schonmann, and A. M. Stacey. “Domination by product measures”.Ann. Probab.25:1 (1997), pages 71–95.doi:10.1214/aop/1024404279(cited on page 5)

arXiv 1997

[2] [29]

From loop clusters and random interlacement to the free field

T. Lupu. “From loop clusters and random interlacements to the free field”.Ann. Probab.44:3 (2016), pages 2117–2146.doi: 10.1214/15-AOP1019. arXiv:1402.0298 [math.PR] (cited on pages 1, 2, 7–10)

work page Pith review arXiv 2016

[3] [30]

Loop percolation on discrete half-plane

T. Lupu. “Loop percolation on discrete half-plane”.Electron. Commun. Probab.21 (2016), Paper No. 30, 9.doi:10.1214/16-ECP4571. arXiv:1408.1045 [math.PR](cited on page 10)

work page Pith review arXiv 2016

[4] [31]

Critical curve of loop percolation on the $d$-regular tree

L. Makowiec and A. Sapozhnikov.Critical curve of loop percolation on the𝑑-regular tree. 2026. arXiv:2606.27520 [math.PR](cited on page 6)

work page Pith review arXiv 2026

[5] [32]

Martineau, R

S. Martineau, R. Poudevigne, and P. Rax.Stochastic domination and lifts of random variables in percolation theory. Preprint, arXiv:2504.02427 [math.PR] (2026). 2026 (cited on page 7)

work page arXiv 2026

[6] [33]

Critical parameters for loop and Bernoulli percolation

P. Mühlbacher. “Critical parameters for loop and Bernoulli percolation”. English.ALEA, Lat. Am. J. Probab. Math. Stat.18:1 (2021), pages 289–308 (cited on page 7)

work page 2021

[7] [34]

Nonamenable Poisson zoo

G. Pete and S. Rokob.Nonamenable Poisson zoo. Preprint, arXiv:2505.07145 [math.PR]. 2025. arXiv:2505.07145 [math.PR](cited on pages 1, 3)

work page Pith review arXiv 2025

[8] [35]

Percolation of worms

B. Ráth and S. Rokob. “Percolation of worms”.Stochastic Processes Appl.152 (2022), pages 233– 288.doi:10.1016/j.spa.2022.06.018. arXiv:2107.03259 [math.PR](cited on page 3)

work page Pith review arXiv 2022