Percolation on random 2-lifts

Paul Drouvill\'e

REVIEW 3 major objections 6 minor 2 references

Percolation on random 2-lifts

T0 review · 3 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Random double covers of non-tree transitive graphs strictly lower the percolation threshold.

desk verdict Genuinely new results and a natural model, but the strict-monotonicity proof has a load-bearing gap in the construction of the enhancement variables; likely fixable, so conditional rather than reject. read the letter →

arxiv 2506.01612 v1 pith:WKOGUT2B submitted 2025-06-02 math.PR math.CO

Paul Drouvillé This is my paper

classification math.PRmath.CO MSC 60K3505C8082B43

keywords random2-liftBernoullipercolationthresholdcriticalparametertransitivegraphexponentialdecayenhancedgaugetransformation

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

Percolation on a random 2-lift of a graph $G$ takes two copies of $G$ and crosses each pair of parallel edges with probability $q$. This paper proves that for every connected, transitive, infinite graph $G$ that is not a tree and has $p_c(G)<1$, the random lift $G_q$ has a strictly smaller percolation threshold than $G$, for every $q\in(0,1)$, almost surely. It also proves that the threshold $p_c(q)$ is continuous in $q$ on $(0,1)$ and that at $q=1/2$ the tail of the cluster of a vertex decays exponentially for every $p

What carries the argument

The load-bearing object is the random 2-lift $G_q=(V\times\{0,1\},E(\eta))$, with independent Bernoulli switching variables $\eta_e$; a pair of parallel edges is crossed exactly when $\eta_e=1$. The proofs work by couplings rather than by analyzing one realization's geometry: first, a coupling between percolation on $G_q$ and enhanced percolation on $G$, in which a fully open ball of radius $r$ can, with auxiliary probability $s$, annex its boundary sphere; second, Bernoulli variables $\beta_x$ that probabilistically certify that the two lifts of a vertex $x$ are connected within $D$ steps, replacing the deterministic bounded-fiber condition used in [MS19]; and third, at $q=1/2$, an exploration that reveals edges without always revealing both endpoints, together with a coupling of the 'remaining graph' after the origin's cluster is deleted. The $q=1/2$ step uses gauge transformations $\phi_S$ and the fact that the switching law is the Haar measure on $\mathbb{Z}/2\mathbb{Z}$, whose convolution with any explored geometry remains uniform; this is why the exponential-decay coupling is proved only at $q=1/2$.

What would settle it

Run two independent high-accuracy simulations of bond percolation on large finite toroidal 2-lifts of the square lattice at the same $q\in(0,1)$, and compare the two finite-size estimates of the quenched critical threshold; if they do not converge to a common value, the almost-sure constancy that Theorem 1.2 needs is false. A direct refutation of the strict inequality would be a single transitive non-tree graph with $p_c(G)<1$ and one $q\in(0,1)$ for which the limiting threshold satisfies $p_c(G_q)\ge p_c(G)$.

Watch

Extended reading notes

Core claim

The central claim is that the random choice of which edges cross between the two floors of a double cover lowers the percolation threshold strictly, in every nontrivial switching regime, and that the usual sharp phase transition persists in this non-transitive random setting. Theorem 1.2 states that if $G$ is connected, transitive, infinite, not a tree, and $p_c(G)<1$, then $p_c(G_q)<p_c(G)$ almost surely for every $q\in(0,1)$; the hypothesis that $G$ is not a tree is necessary, because a tree has two parallel copies with the same threshold. Theorem 1.1 says $q\mapsto p_c(q)$ is continuous on $(0,1)$ (in fact locally $1/2$-Hölder continuous), and Theorem 1.4 gives exponential decay of the cluster size at $q=1/2$, with a quenched version following for almost every realization of the random lift.

Load-bearing premise

The almost-sure conclusion in Theorem 1.2 rests on the unstated premise that the percolation threshold of the random 2-lift is the same for essentially every realization of the switching randomness, so that the average strict inequality proved in Section 4 can be applied to each typical random graph; if different realizations had different thresholds, strict inequality could hold on average but fail almost surely.

Editorial extensions

If this is right

For every non-tree transitive graph with $p_c(G)<1$, an arbitrarily small switching probability already makes percolation easier: $p_c(G_q)<p_c(G)$ for every $q\in(0,1)$.
Because $q=0$ gives two disjoint copies of $G$, continuity at $0$ means the threshold drops immediately as soon as switching is turned on.
At $q=1/2$, the subcritical regime is genuinely exponentially decaying: $\mathbb{P}(|C_o|\ge n)\le Ce^{-cn}$ for all $p<p_c(1/2)$, and the same holds for almost every realization of the random lift.
Continuity of $q\mapsto p_c(q)$ in $(0,1)$ is a locality-type result for a family of random, generally non-transitive graphs: the critical parameter varies continuously when the law of the random graph is deformed.

Reading between the lines

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

Editorial inference: the probabilistic twin-distance variables $\beta_x$ are the only place where the proof of Theorem 1.2 uses the special structure of a 2-lift; the same scheme should yield strict monotonicity for random $n$-lifts with i.i.d. permutation switching, provided the cycle-certificate lemmas generalize.
Editorial inference: the Borel–Cantelli argument that turns the annealed exponential bound into a quenched one is a general template: any uniform annealed exponential tail bound would transfer to almost every realization in the same way.
Editorial inference: the open question of exponential decay for $q\neq 1/2$ could be probed numerically by measuring cluster tails on large finite 2-lifts of the square lattice at $q=1/4$; the paper predicts the decay should hold, while Lemma 5.10 may fail.

Share X Bluesky LinkedIn Reddit HN

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

Bottom line: this is a genuinely interesting paper with a real gap in its central strict-monotonicity proof. The continuity theorem is clean, the exponential-decay section is ambitious, and the model is natural. But the proof of Proposition 4.1 does not actually show that the enhancement variables α_u are i.i.d. Bernoulli(s), and Theorem 1.2 leans on that. I'd send it to a serious referee, but I would not accept it as-is.

The three results are new. Continuity of p_c(q) via the coupling in Lemma 3.2 is a nice locality statement for a random graph family; the argument is self-contained and looks correct. Exponential decay at q=1/2 follows Vaneuville's ghost-field method, with a serious extra twist (Lemma 5.10) about the remaining graph; the coupling proof is sketched but the idea is plausible, and the q=1/2 restriction is honest. The strict monotonicity is the weak point. The proof imports [MS19]'s enhanced percolation but must replace deterministic bounded fibers with probabilistic control, which is a genuinely new difficulty.

The specific problem is in Step 2N+2, substep 6: the paper asserts that α_u is Bernoulli(s) with fixed s and that the α_u are independent. The definition sets α_u=1 exactly when a random number N of newly s-explored edges are open and β_x=1. N depends on the exploration history and on η, so the success probability is t·p^N, a mixture, not a fixed Bernoulli(s). Independence across overlapping Z-regions is also not established. Without α ~ B(s)^{⊗V(G)}, C∞ is not distributed as the augmented percolation cluster, and the inequality p_c(G_q) ≤ p_c(G,s) collapses. The reader's note about the 'a.s.' transfer is not the real issue—the 0-1 law plus the annealed inequality gives the quenched statement directly, no ergodicity needed. The real problem is the α construction.

That said, the gap is probably repairable. In [MS19] the analogous step uses a thinning argument: if the success probability is bounded below and the variables are locally dependent, you can extract a genuine Bernoulli(s) family stochastically dominated by the original. The paper needs to spell that out; it doesn't. Given that the rest of the strategy is sound and the missing argument is standard but nontrivial, I'd call this conditional rather than a reject. The paper is honest about what it does and doesn't prove—Question 5.14 is explicitly left open.

Who is this for? People working on percolation on graphs, especially locality and covering maps. It deserves a serious referee, and I'd be happy to see a revised version with the thinning argument filled in. For my own work I wouldn't cite Theorem 1.2 yet, though I'd consider citing the continuity result if I needed it.

Referee Report

3 major / 6 minor

Summary. The paper studies Bernoulli percolation on random 2-lifts G_q of a transitive graph G, obtained by switching each pair of parallel edges with probability q. It proves three results: the annealed critical parameter p_c(q) is continuous on (0,1) (Theorem 1.1); for connected non-tree transitive G with p_c(G) < 1, p_c(G_q) < p_c(G) almost surely (Theorem 1.2); and the cluster-size tail decays exponentially for p < p_c(1/2) (Theorem 1.4). The continuity proof is a coupling showing 1/2-Hölder continuity, the monotonicity proof adapts the augmented-percolation method of Martineau and Severo, and the exponential decay proof follows Vaneuville's ghost-field strategy with a new exploration framework for random graphs.

Significance. The continuity result is a clean and likely correct application of coupling and gives a new locality-type statement for random graph limits. The exponential decay section contains a novel exploration of random graph structure and an elegant use of the Haar-measure property of Bernoulli(1/2) switching variables to prove the key remaining-graph comparison; this is a valuable contribution in its own right. The strict monotonicity theorem, if established, would be a substantial extension of the Martineau–Severo result to covers without bounded fibers. However, the proof of Theorem 1.2 contains a gap in the construction of the enhanced percolation variables: the α_u variables are not shown to be i.i.d. Bernoulli(s), and their independence from the exploration is not proved. This gap is likely repairable by a thinning argument, but it is load-bearing for the main theorem as written.

major comments (3)

[4.2, Step 2N+2, substeps 4–6] The proof asserts in substep 6 that α_u is Bernoulli with parameter s and that the α_u's are independent, and the final paragraph of the proof uses this to conclude that C∞ has the law of augmented percolation with parameters (p,s). This does not follow from the construction: the success of substeps 3 and 4 requires all of a random number N of s-explored edges to be open, so the success probability is t·p^N (with N depending on p, η, and the exploration history). The marginal law of α_u is therefore a mixture rather than Bernoulli(s), and no independence across u is established. Without α ~ B(s)^{⊗V(G)} independent of ω, the comparison p_c(G_q) ≤ p_c(G,s) in the proof of Theorem 1.2 collapses. A thinning argument (for instance using the uniform bound M from property (D) and the lower bound p ≥ ε to choose s = t ε^M) is needed, but the manuscript does not supply it.
[4.2, substep 5] The variable β_x imported from Lemma 4.7 is a deterministic function of the switching configuration η (via the variables O_i), and the exploration process reveals information about η. The proof does not show that β_x is independent of the σ-field generated by the earlier exploration, nor that the events {β_x=1} for different vertices are conditionally independent given the history. These properties are required for the claimed joint law of the α_u variables and for the identification of C∞ as an augmented-percolation cluster.
[Theorem 1.2 and Section 4] The theorem is a quenched statement (the inequality holds a.s. for the random graph G_q), but the proof establishes the annealed inequality p_c(q) ≤ p_c(G,s) < p_c(G). The paper never explains why the quenched critical parameter p_c(G_q(η)) is almost surely equal to the annealed p_c(q). The equality follows from Ψ(p,q)=E[θ_η(p)] together with the Kolmogorov 0-1 law and monotonicity in p, but this argument should be written out because it is essential to the a.s. formulation.

minor comments (6)

[Proof of Theorem 1.4] In the proof of Theorem 1.4, the last display and the concluding sentence give C = 1/(1−2m_h(p′)), but the computation yields C = 1/(1−m_h(p′)); the bound is ψ_n(p) ≤ (1/(1−m_h(p′))) e^{-hn}.
[Proof of Lemma 5.10] In the proof of Lemma 5.10, the sentence 'Let e be a type (3) edge' refers to a type that has not been defined; from the previous paragraph this should read 'type (1)'.
[Lemma 5.7] In Lemma 5.7 and its proof, the event 'A ∩ Expl_k' is written without specifying that A is a subset of {0,1}^E × F × {0,1}^V while Expl_k is a subset of {0,1}^E × F; the text should clarify that the projection of A onto the first two coordinates is taken.
[Proof of Proposition 2.4] The proof of Proposition 2.4 invokes a 'factorization theorem' without naming it; this is an orbit–stabilizer count and should be stated as such.
[Introduction] The first sentence of the introduction misspells 'Hammersley' as 'Hammersely'.
[Remark 1.3] In Remark 1.3, the phrase 'G_q consists essentially of two copies of G in parallel' is true for trees but deserves a proof or a more precise statement: every 2-lift of a tree is isomorphic to two disjoint copies of G.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper's results are obtained by explicit couplings plus external theorems, with no fitted parameters or load-bearing self-citations.

full rationale

Walking the derivation chain: Theorem 1.1 (continuity) rests on Lemma 3.2, whose proof explicitly constructs a coupling from independent uniforms, X_e, Y_e, Z_e; the inequality p_c(q/(1-r)) >= (1-sqrt(r)) p_c(a) is derived in-text, not imported. Theorem 1.2 is assembled from two external ingredients: the paper's Proposition 4.1 (a concrete exploration coupling that uses the probabilistic twin-connection lemmas 4.3-4.7) and Proposition 4.2, quoted verbatim as "Proposition 4.2 from [MS19]". This is an external published theorem, not a self-citation; it supplies the strict inequality p_c(G,s) < p_c(G), so the claimed strictness is not manufactured by a fit or by definition. Theorem 1.4 follows Vaneuville's ghost-field method via Theorem 5.2 and Lemma 5.10, whose proof is an explicit coupling using the q=1/2 Haar measure. No parameter is fitted to the predicted quantity and no central claim is defined in terms of the conclusion. The paper's own flagged limitations — the acknowledgement that Paul Rax found a mistake in an earlier version of the continuity proof, and the statement that exponential decay is only proved at q=1/2 (Question 5.14) — are caveats about scope, not circularity. Two technical gaps are visible but non-circular: in Step 2N+2, substep 6 of Proposition 4.1, the assertion "therefore alpha_u is a Bernoulli variable with parameter s and the alpha_u variables are independent" is not justified by the preceding success events (the success probability is a mixture over histories and p), and the paper suppresses the standard thinning argument supplied in [MS19]; and the "a.s." in Theorem 1.2 is not explicitly derived, although it follows from the 0-1 law (if Psi(p,q)=1 then theta_eta(p)=1 a.s.) rather than requiring ergodicity. Neither defect is a circular reduction of the theorem to its inputs.

Assumptions & free parameters 0 free parameters · 6 assumptions · 0 invented entities

The central claims rest on standard percolation facts and on two imported results ([MS19] Proposition 4.2 and Vaneuville's method), plus an implicit ergodicity assumption. No free parameters are fitted; the proof constants are auxiliary. No invented entities are introduced.

assumptions (6)

domain assumption G is a non-empty simple undirected graph, connected and locally finite; most results assume G is transitive and infinite.
Stated in the introduction as the class of graphs studied; used in all theorems.
domain assumption For Theorem 1.2, G is not a tree and p_c(G) < 1.
Hypotheses of Theorem 1.2; Remark 1.3 shows trees are excluded because then p_c(G_q)=p_c(G).
standard math Kolmogorov 0-1 law makes the infinite-cluster event tail-measurable for the annealed measure, yielding a deterministic p_c(q).
Used in Section 1 to define p_c(q).
domain assumption Martineau-Severo Proposition 4.2: for transitive infinite G with p_c(G)<1, enhanced percolation has p_c(G,s) < p_c(G) for every r>=1 and s in (0,1].
Imported from [MS19] and used in the proof of Theorem 1.2.
domain assumption The i.i.d. switching variables are ergodic under Aut(G), so the quenched critical value is a.s. constant.
Implicitly needed for the a.s. statement in Theorem 1.2; not proven in the paper.
standard math Gauge transformations phi_S are graph isomorphisms of every realization of G_q.
Used in Sections 2 and 5; follows from the definition of the random 2-lift.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Percolation on random 2-lifts." pith.science (2026). https://pith.science/paper/WKOGUT2B

@misc{pith2026250601612,
  author       = {Pith},
  title        = {Pith review of: Percolation on random 2-lifts},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WKOGUT2B}},
  note         = {Machine review of arXiv:2506.01612}
}

abstract

Given a graph $G$, we consider a model for a random cover of $G$ by taking two parallel copies of $G$ and crossing every pair of parallel edges randomly with probability $q$ independently of each other. The resulting graph $G_q$, is a random $2$-lift of $G$ that may not be transitive but still probabilistically exhibit many properties of transitive graphs. Studying percolation in this context can help us test the reliability and robustness of our proofs methods in percolation theory. Our three main results on this model are the continuity of the critical parameter $p_c(G_q)$, for $q\in(0,1)$, the strict monotonicity $p_c(G_q)< p_c(G)$ and the exponential decay of the cluster size in the subcritical regime at $q=1/2$.

Figures

Figures reproduced from arXiv: 2506.01612 by the authors.

**Figure 2.** An example of a remaining graph Gfq with G = Z 2 seen from above. Since the graph in this figure is seen from above, note that an edge here represent a pair of twin edges in Gfq. Holes correspond to pairs of edges which were both explored and fully deleted in Gfq. Type (2) edges are represented in orange and type (1) edges in blue. Type (0) edges are drawn in black. Let us take a realization (ω, f) of a percolation … view at source ↗

**Figure 3.** The geometry of a type (1) edge (in green) and its lift depends on the [PITH_FULL_IMAGE:figures/full_fig_p027_3.png] view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

2 extracted references · 1 canonical work pages

[1]

The spectrum of random lifts

[AB87] Michael Aizenman and David Barsky. “Sharpness of the phase tran- sition in percolation models”. In: Communications in Mathematical Physics 108 (1987). [AG] Louigi Addario-Berry and Simon Griffiths. The spectrum of random lifts. arXiv: 1012.4097. [AG91] Michael Aizenman and Grimmett Geoffrey. “Strict monotonicity for critical points in percolation a...

work page Pith review arXiv 1987
[206]

Coincidence of critical points in percolation problems

[EH] Philip Easo and Tom Hutchcroft. The critical percolation probability is local. arXiv: 2310.10983. 29 [Men86] Mikhail Menshikov. “Coincidence of critical points in percolation problems”. In: Soviet Mathematics Doklady 33 (1986), pp. 856–859. [MS19] S´ ebastien Martineau and Franco Severo. “Strict monotonicity of per- colation thresholds under covering...

arXiv 1986

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

The spectrum of random lifts

[AB87] Michael Aizenman and David Barsky. “Sharpness of the phase tran- sition in percolation models”. In: Communications in Mathematical Physics 108 (1987). [AG] Louigi Addario-Berry and Simon Griffiths. The spectrum of random lifts. arXiv: 1012.4097. [AG91] Michael Aizenman and Grimmett Geoffrey. “Strict monotonicity for critical points in percolation a...

work page Pith review arXiv 1987

[2] [206]

Coincidence of critical points in percolation problems

[EH] Philip Easo and Tom Hutchcroft. The critical percolation probability is local. arXiv: 2310.10983. 29 [Men86] Mikhail Menshikov. “Coincidence of critical points in percolation problems”. In: Soviet Mathematics Doklady 33 (1986), pp. 856–859. [MS19] S´ ebastien Martineau and Franco Severo. “Strict monotonicity of per- colation thresholds under covering...

arXiv 1986