Loop clusters on complete graphs

Yves Le Jan

REVIEW 3 major objections 6 minor 35 references

Loop clusters on complete graphs

T0 review · 3 major / 6 minor · reviewed 2026-08-16 · deepseek-v4-flash

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

Pith's one-line read The number of fixed-size loop clusters on a large complete graph converges to a mixed Poisson law with an explicit Gamma-driven mixing variable.

desk verdict A useful short paper on loop cluster asymptotics for complete graphs, but Proposition 5 is missing a factor (α/d)^k for d≥2 and needs a local correction. read the letter →

arxiv 2504.16976 v2 pith:FK25W5XL submitted 2025-04-23 math.PR

Yves Le Jan This is my paper · ORCID

classification math.PR MSC 60C0560J2760G60

keywords MarkovloopsloopclusterscompletegraphmixedPoissondistributionGammacumulantsrandompartitionslarge

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 studies the random partition of the complete graph $K_n$ generated by a Poissonian ensemble of Markov loops with intensity parameter $\alpha$, unit conductances, and constant killing $\kappa$. It seeks the asymptotic census of small clusters: for each fixed cluster size $d\ge 2$, the number $|I_d|$ of clusters occupying exactly $d$ vertices converges in distribution to a mixed Poisson law, with the mixing variable an explicit function $H_\alpha=\exp(-dZ_\alpha/\kappa)$ of a Gamma$(\alpha,1)$ variable; for $\alpha=1$ this gives a closed-form integral density. The same analysis shows the proportion of isolated vertices converges to $R_\alpha=\exp(-Z_\alpha/\kappa)$, and that with probability tending to 1 the partition contains clusters visiting more than $n^{1-\epsilon}$ vertices. The result gives a complete asymptotic description of the small-cluster side of the loop-cluster partition on the complete graph, and it identifies the dominant mechanism: asymptotically, a fixed-size cluster is created by a single $d$-gon loop rather than by an accumulation of smaller loops.

What carries the argument

The engine is the Green-matrix determinant formula for loop clusters: for any partition $\pi$, $P(C_\alpha \succeq \pi)=(\prod_i \det G_{B_i}/\det G)^\alpha$. On the complete graph this becomes a product of moments of $Y^{(\alpha)}=\exp(Z_\alpha/(n+\kappa))$, because the determinant of the Green matrix on a $d$-set is $1/((n-d+\kappa)(n+\kappa)^{d-1})$. The cluster probabilities are then rewritten using cumulants $c_d^{(\alpha)}$, whose asymptotic $\alpha(d-1)!\,n^{-d}$ is obtained in Lemma 2 by sandwiching the connectedness probability of an isolated $d$-set: below by the probability that exactly one $d$-gon appears, above by that event plus bounds on loops of total size exceeding $d$. This single-$d$-gon dominance mechanism converts the exact but unwieldy inclusion-exclusion formula into a Poisson-mixture limit.

What would settle it

Check the arithmetic connecting Propositions 3 and 5: Proposition 3 gives $\lim_n \mathbb{E}[(|I_d|)_k] = \alpha^k d^{-k}(\kappa/(kd+\kappa))^\alpha$, whereas the mixed-Poisson law in Proposition 5 has factorial moments $\mathbb{E}[H_\alpha^k]=(\kappa/(kd+\kappa))^\alpha$; for $d\ge 2$ these differ by $d^{-k}$, so a direct verification of which formula matches a simulation of $K_n$ for large $n$ would settle the exact limiting statement.

Watch

Extended reading notes

Core claim

The central discovery is an exact asymptotic law for the small-cluster census. For fixed $\alpha>0$ and fixed $d\ge 2$, let $I_d$ be the set of $C_\alpha$-clusters of cardinality $d$ in $K_n$ with unit conductances and killing $\kappa$. Proposition 5 asserts that as $n\to\infty$, $|I_d|$ converges in distribution to a mixture of Poisson distributions: $P(|I_d|=k)\to \mathbb{E}[H_\alpha^k e^{-H_\alpha}/k!]$, where $H_\alpha=\exp(-dZ_\alpha/\kappa)$ and $Z_\alpha$ is Gamma$(\alpha,1)$. In the case $\alpha=1$, this becomes $\int_0^1 \frac{x^k}{k!}e^{-x}\frac{\kappa}{d}x^{\kappa/d-1}\,dx$. The proof establishes that the asymptotic factorial moments are $\alpha^k d^{-k}(\kappa/(kd+\kappa))^\alpha$, and shows that the probability an isolated $d$-set is connected is asymptotically equal to the probability that it contains exactly one $d$-gon loop and no other loops. Alongside this, Proposition 6 shows that for every $\epsilon>0$, clusters of size larger than $n^{1-\epsilon}$ exist with probability tending to 1; whether clusters of linear size $cn$ exist is left open.

Load-bearing premise

Everything rests on treating the loops inside each candidate cluster as independent of the loops touching the remaining vertices and on the complete graph's self-similarity under replacing a $d$-set by a full graph with killing $n-d+\kappa$; if that step is not valid, the factored product of cluster probabilities behind Proposition 5 collapses.

Editorial extensions

If this is right

For fixed $\kappa$ and $\alpha$, all factorial moments of $|I_d|$ have explicit limits, so moment methods determine the full limiting distribution without further inclusion-exclusion.
For $\alpha=1$, the limiting distribution has a one-line integral formula, allowing direct computation of quantities such as the probability of seeing no clusters of size $d$.
Asymptotically, a fixed-size cluster is caused by a single $d$-cycle; larger loop configurations contribute only lower-order corrections.
With probability tending to 1, the loop-cluster partition has macroscopic clusters spanning more than $n^{1-\epsilon}$ vertices for every $\epsilon>0$, while clusters of linear size remain an open question.

Reading between the lines

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

The same determinant-cumulant route should apply to other dense graph families with high symmetry; the mixing variable would still be Gamma-driven, with the killing parameter replaced by an effective spectral parameter.
The 'one cycle dominates' lemma suggests a general principle for loop soups on graphs of bounded local cycle density: the leading contribution to a small cluster is a single simple cycle, which could be tested numerically on tori or sparse random graphs.
The open question of linear-size clusters could be approached by refining Proposition 6's second-moment calculation: if the visited-set size of the longest loop concentrates near its length, the existence of $cn$-clusters would follow for small $c$; conversely, a superpolylogarithmic upper tail would rule them out.

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

Short version: the paper does a clean job of specializing the loop cluster framework to complete graphs, but the headline limit theorem (Prop 5) is wrong for d≥2. The stress-test note is correct. From Prop 3 and Lemma 2, the k-th factorial moment of |I_d| converges to (α/d)^k (κ/(kd+κ))^α. The mixed Poisson law stated in Prop 5 uses H_α = exp(-dZ_α/κ), whose factorial moments are (κ/(kd+κ))^α. So for d≥2 the claimed distribution has the wrong moments. The fix is local: replace H_α by (α/d)H_α, or equivalently state the limit as a mixed Poisson with that mixing variable. For d=1, Lemma 2's asymptotic c_d ∼ α(d-1)!n^{-d} fails (the first cumulant tends to 1, not α/n), so the isolated-vertex result in Prop 4 is fine, but the paper should say Lemma 2 is for d≥2.

What is good: the derivation of Prop 3 via Green determinants and the factorization of loop sets is elegant; the reduction of connectivity to isolated d-gons in Lemma 2's proof is a nice argument; Prop 6 gives a simple existence proof for large clusters, modulo a trivial typo (n^ε/ε should be n^{-ε}/ε or 1/ε). The self-citation is legitimate—Prop 1 and 2 are the author's own framework, and the new results are derived from them rather than assumed.

Soft spots: the missing factor is load-bearing because Prop 5 is the main advertised result. The step from factorial moment convergence to distributional convergence is also not justified; given the moments grow like (α/d)^k, you need a bit more than "consequently." That said, the method is sound and the correction is contained. The d=1 exception should be stated explicitly.

Who this is for: people working on Markov loop soups and random partitions on dense graphs. They will find Prop 3 and Lemma 2 useful as a reference model, but they should not quote Prop 5 as is.

Recommendation: send it to a referee, but with a note that the central limit statement needs to be corrected. If the author fixes the mixing variable and cleans up the moment-to-distribution step, this becomes a solid short paper.

Referee Report

3 major / 6 minor

Summary. The paper studies the random partition C_alpha of the complete graph K_n induced by a Poissonian ensemble of Markov loops of intensity alpha, with unit conductances and constant killing kappa. Building on the author's earlier determinant formulas for the probability that C_alpha is thinner than a given partition, the paper derives exact factorial moments for the number |I_d| of clusters of a fixed finite size d (Prop. 3), an asymptotic evaluation of the relevant cumulants (Lemma 2), and a claimed limit theorem: |I_d| converges in distribution to a mixed Poisson law with mixing variable H_alpha = exp(-d Z_alpha / kappa), where Z_alpha is Gamma(alpha,1) (Prop. 5). The paper also proves that, for any epsilon>0, clusters of size larger than n^{1-epsilon} exist with probability tending to 1 (Prop. 6), and discusses the relation to the Erdos-Renyi tree census.

Significance. The paper's approach is elegant and uses powerful, well-established tools: Green-function determinants, moment/cumulant relations, and the author's prior loop-cluster formalism. The exact factorial moment formula in Prop. 3 and the sharp asymptotics of the cumulants in Lemma 2 are valuable and appear correct. If the distributional result were established, it would give a complete explicit small-cluster census for loop clusters on complete graphs, complementing classical random graph results. The paper is concise and mostly rigorous, with explicit determinant and moment computations. However, as detailed below, the central limit statement contains a factor error that must be corrected.

major comments (3)

[§4, Proposition 6] Proposition 5 is inconsistent with the factorial moments derived in Proposition 3 and Lemma 2. From Proposition 3 and Lemma 2, E[|I_d|(|I_d|-1)...(|I_d|-k+1)] converges to (alpha/d)^k (kappa/(kd+kappa))^alpha. The mixed Poisson law with mixing variable H_alpha = exp(-d Z_alpha / kappa) has factorial moments E[H_alpha^k] = (kappa/(kd+kappa))^alpha. The two sequences differ by a factor (alpha/d)^k for every k>=1 and every d>=2, and by alpha^k for d=1. For instance, for d=2 and alpha=1, the derived first factorial moment tends to (1/2) kappa/(kappa+2), whereas the claimed limit has mean kappa/(kappa+2). Hence Proposition 5 does not follow from the preceding asymptotics. The mixing variable should presumably be H_alpha = (alpha/d) exp(-d Z_alpha / kappa), and the displayed alpha=1 density must be modified accordingly.
[§4, Proposition 6] The Markov bound in the proof of Proposition 6 is typeset as n^epsilon/epsilon, which is greater than 1 for large n; with this bound the claimed lower bound (1 - n^{-alpha epsilon/2})(1 - n^epsilon/epsilon) is negative and cannot tend to 1. The intended estimate must be of order n^{-epsilon}/epsilon (or 1/(epsilon n^epsilon)); with that correction the argument appears to work. This is a localized but necessary correction.
[§3, Proposition 5] The passage from convergence of factorial moments to convergence in distribution is not justified. Even after correcting the mixing variable, the paper should either cite a standard moment-convergence theorem for nonnegative integer-valued random variables whose factorial moments converge to those of a mixed Poisson law with bounded mixing variable, or provide a short proof. As written, the distributional conclusion in Proposition 5 is an assertion rather than a consequence of the moment computations.

minor comments (6)

[Abstract] The word 'emsembles' should be 'ensembles'.
[§4, proof of Proposition 6] The word 'demoting' should be 'denoting'.
[§3, note after Lemma 2] The reference to 'lemma 3' should be to 'Lemma 2'.
[§3, Proposition 3] The notation c_d^alpha is not defined at first use; it refers to the d-th cumulant of Y^(alpha) defined before formula (4), but the superscript should be made consistent (for example, c_d^(alpha)).
[§3, Proposition 5] The displayed formula for the alpha=1 case appears to be missing an equality sign and, once the factor error in the mixing variable is fixed, needs to be recomputed; the current display does not follow from the stated mixing variable.
[Remark 3.2] The sentence about expansions in powers of x = 1/(n+1) and the triangle A087903 would benefit from a brief explanation or reference.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity; the loop-cluster limit laws are derived from prior independent loop-measure identities and not from the target claims.

full rationale

The paper's derivation chain is not circular in the sense defined by the review rules. Proposition 3 is obtained from Proposition 1 and Proposition 2, which are quoted from the author's earlier published work ([4], [6]) as general loop-cluster results. Those results are external tools: they do not contain the target limit law for |I_d|, and their assumptions do not include the conclusion of Proposition 5. The subsequent asymptotic analysis uses Lemma 2, which is proved by direct estimates of the probability that an isolated set is connected, not by assuming the Poisson-mixture limit. The self-similarity of the complete graph and the spatial independence of Poisson loop ensembles are structural properties used in the proof, not renamings of the desired conclusion. Lemma 2's proof invokes formula 4.18 from the author's own book [6] to bound a moment; that cited identity is a parameter-free result about loop measures and Green functions with stated assumptions that do not include the target statement, so it counts as independent support rather than circular self-citation. The reader's note about a possible factorial-moment mismatch between Proposition 3 plus Lemma 2 and the mixing variable in Proposition 5, if correct, would be a mathematical consistency or correctness issue, not a circularity issue: the claimed limit is not equivalent by construction to the preceding asymptotic moments, it is in tension with them. Under the hard rules of this circularity pass, correctness gaps and internal inconsistencies do not constitute circularity unless a prediction reduces to a fitted input or a self-citation chain. No such reduction is exhibited. The self-citations are frequent but load-bearing only as external theorems, and the central novelty, the small-cluster census in Proposition 5, is not assumed in those cited inputs.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The paper has no fitted parameters: α and κ are inputs of the loop soup model. It relies on the standard Markov loop cluster framework and several published theorems, all non-circular relative to the new claims. No new entities are introduced.

assumptions (5)

domain assumption Loop cluster partition Cα defined via Poisson process of Markov loops on a finite graph (intensity αν), as in [6] section 2.1 and 5.1.
The paper's object of study is this loop soup; the definition and its properties are taken from prior work without proof.
domain assumption Proposition 1: P(Cα ≥ π) = (∏_i det(G_{B_i})/det(G))^α for any partition π, from [4].
Used directly in Section 2 to compute thinning probabilities on K_n.
domain assumption Proposition 2: inclusion-exclusion formula for P(Cα=π), from [4].
Used to obtain formula (4) for P(Cα={X}) and eventually the factorial moments.
domain assumption Formula 4.18 in [6] for the exponential moment of the total loop length in a subset D.
Invoked in Lemma 2 to bound the probability of large total loop size.
domain assumption Chang's theorem 1-2 in [1]: under the normalized loop measure on K_n, log |ℓ|/log n converges to Uniform[0,1].
Used in Section 4 to prove the existence of loops and clusters of size > n^{1-ε}.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Loop clusters on complete graphs." pith.science (2026). https://pith.science/paper/FK25W5XL

@misc{pith2026250416976,
  author       = {Pith},
  title        = {Pith review of: Loop clusters on complete graphs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/FK25W5XL}},
  note         = {Machine review of arXiv:2504.16976}
}

read the original abstract

We investigate random partitions of complete graphs defined by Poissonian emsembles of Markov loops

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 32 canonical work pages

[1]

Graphs and digraphs

Mehdi Behzad, Gary Chartrand, Linda Lesniak-Foster. Graphs and digraphs. PWS Publishers, Boston, Mass., 1979

work page 1979
[2]

Non-backtracking loop soups and statistical mechanics on spin networks

Federico Camia, Marcin Lis. Non-backtracking loop soups and statistical mechanics on spin networks. Ann Institut Poincar\'e 18, 403–433 (2017)

work page 2017
[3]

Markov loops in discrete spaces

Yinshan Chang, Yves Le Jan. Markov loops in discrete spaces. arXiv:1402.1064. Probability and Statistical Physics in St. Petersburg. Proceedings of Symposia in Pure Mathematics 91 (2016)

work page Pith review arXiv 2016
[4]

Yinshan Chang Loop cluster on the discrete circle arXiv:1311.7583

work page Pith review arXiv
[5]

Jacobian Tori Associated with a Finite Graph and Its Abelian Covering Graphs

Motoko Kotani, Toshikazu Sunada. Jacobian Tori Associated with a Finite Graph and Its Abelian Covering Graphs. Advances in Applied Mathematics 24, pp 89–110. (2000)

work page 2000
[6]

A convergence result on the lengths of Markovian loops

Yinshan Chang. A convergence result on the lengths of Markovian loops. Electron. Commun. Probab. 20. no. 73, 1–11. (2015)

work page 2015
[7]

The Brownian loop soup

Gregory Lawler, Wendelin Werner. The Brownian loop soup. PTRF 128 565-588 (2004)

work page 2004
[8]

Lando and A.K

S.K. Lando and A.K. Zvonkin. Graphs on Surfaces and their Applications. Springer. (2004)

work page 2004

Show all 35 references

[9]

Markov loops, determinants and Gaussian fields arXiv:math/0612112

Yves Le Jan. Markov loops, determinants and Gaussian fields arXiv:math/0612112

arXiv
[10]

Advanced Combinatorics

Louis Comtet. Advanced Combinatorics. 2d edition. Reidel Publishing Company. Dordrecht. (1974)

1974
[11]

On the evolution of random graphs, Publ

Paul Erdos, Alfred Renyi. On the evolution of random graphs, Publ. Math. Inst. Hung. Acad. Sci., 5, 17-61 (1960)

1960
[12]

Markovian loop clusters on graphs

Yves Le Jan, Sophie Lemaire. Markovian loop clusters on graphs. arXiv:1211.0300. Illinois Journal of Mathematics Volume 57, 525–558 (2013)

2013 arXiv
[13]

Markov paths, loops and fields

Yves Le Jan. Markov paths, loops and fields. \' E cole d'\' E t\' e de Probabilit\' e s de Saint-Flour XXXVIII - 2008. Lecture Notes in Mathematics 2026. Springer-Verlag, Berlin-Heidelberg (2011)

2011
[14]

Random walks and physical fields

Yves Le Jan. Random walks and physical fields. Springer (2024)

2024
[15]

Markov loops, free field and Eulerian networks

Yves Le Jan. Markov loops, free field and Eulerian networks. arXiv:1405.2879. J. Math. Soc. Japan

arXiv
[16]

Markov loops, coverings and fields

Yves Le Jan. Markov loops, coverings and fields. Ann. Fac. Sci. Toulouse Math. (6) 26 (2017), no. 2, 401–416

2017
[17]

Markov loops topology

Yves Le Jan. Markov loops topology. arXiv:1707.05106

arXiv
[18]

Poisson ensembles of loops of one-dimensional diffusions

Titus Lupu. Poisson ensembles of loops of one-dimensional diffusions. arXiv:1302.3773

arXiv
[19]

From loop clusters and random interlacement to the free field

Titus Lupu. From loop clusters and random interlacement to the free field. arXiv:1402.0298. . Ann. Probab. 44 (2016), no. 3, 2117–2146

2016 arXiv
[20]

A note on Ising random currents, Ising-FK, loop-soups and the Gaussian free field

Titus Lupu, Wendelin Werner. A note on Ising random currents, Ising-FK, loop-soups and the Gaussian free field. Electron. Commun. Probab. 21 (2016), Paper No. 13, 7 pp

2016
[21]

Loop percolation on discrete half-plane

Titus Lupu. Loop percolation on discrete half-plane. arXiv:1408.1045

arXiv
[22]

Massey, \ Algebraic Topology: An Introduction \ Springer (1967)

William S. Massey, \ Algebraic Topology: An Introduction \ Springer (1967)

1967
[23]

Gauge theories as a problem of constructive quantum field theory and statistical mechanics

Erhard Seiler. Gauge theories as a problem of constructive quantum field theory and statistical mechanics. Lecture Notes in Physics
[24]

A Model for Positively Correlated Count Variables

Jesper Møller, Ege Rubak. A Model for Positively Correlated Count Variables. International Statistical Review (2010), 78, 1, 65–80

2010
[25]

Ray-Knight Theorem: a short proof

Christophe Sabot, Pierre Tarrès. Ray-Knight Theorem: a short proof. arXiv:1311.6622

arXiv
[26]

Arbres, amalgames, SL_ 2 Asterisque 46

Jean-Pierre Serre. Arbres, amalgames, SL_ 2 Asterisque 46
[27]

Richard P. Stanley. Enumerative Combinatorics, Vol. 2. Cambridge University Press
[28]

Symanzik, Euclidean quantum field theory

K. Symanzik, Euclidean quantum field theory. Scuola
[29]

Graphs on Surfaces

Bojan Mohar, Carsten Thomassen. Graphs on Surfaces. Johns Hopkins Studies in the Mathematical Sciences. (2001)

2001
[30]

Combinatorial Stochastic Processes

Jim Pitman. Combinatorial Stochastic Processes. 32th St Flour Summer School. Lecture Notes in Math.1875 Springer Berlin (2006)

2006
[31]

On Tree Census and the Giant component in Sparse Random Graphs

Boris Pittel. On Tree Census and the Giant component in Sparse Random Graphs. Random Structures and Algorithms, Vol 1, No 3 (1990)

1990
[32]

Random Graphs and Complex Networks

Remco Van der Hofstad. Random Graphs and Complex Networks
[33]

On the spatial Markov property of soups of unoriented and oriented loops

Wendelin Werner. On the spatial Markov property of soups of unoriented and oriented loops. arXiv:1508.03696. S\' e minaire de de Probabilit\' e s

arXiv
[34]

Appendix in Graphs on Surfaces and their Applications, by S.K

Don Zagier. Appendix in Graphs on Surfaces and their Applications, by S.K. Lando and A.K. Zvonkin. Springer. (2004)

2004
[35]

Generating random spanning trees more quickly than the cover time

David Bruce Wilson. Generating random spanning trees more quickly than the cover time

Pith tools

Reviewed August 16, 2026 · model on record in the stance chip above.

[1] [1]

Graphs and digraphs

Mehdi Behzad, Gary Chartrand, Linda Lesniak-Foster. Graphs and digraphs. PWS Publishers, Boston, Mass., 1979

work page 1979

[2] [2]

Non-backtracking loop soups and statistical mechanics on spin networks

Federico Camia, Marcin Lis. Non-backtracking loop soups and statistical mechanics on spin networks. Ann Institut Poincar\'e 18, 403–433 (2017)

work page 2017

[3] [3]

Markov loops in discrete spaces

Yinshan Chang, Yves Le Jan. Markov loops in discrete spaces. arXiv:1402.1064. Probability and Statistical Physics in St. Petersburg. Proceedings of Symposia in Pure Mathematics 91 (2016)

work page Pith review arXiv 2016

[4] [4]

Yinshan Chang Loop cluster on the discrete circle arXiv:1311.7583

work page Pith review arXiv

[5] [5]

Jacobian Tori Associated with a Finite Graph and Its Abelian Covering Graphs

Motoko Kotani, Toshikazu Sunada. Jacobian Tori Associated with a Finite Graph and Its Abelian Covering Graphs. Advances in Applied Mathematics 24, pp 89–110. (2000)

work page 2000

[6] [6]

A convergence result on the lengths of Markovian loops

Yinshan Chang. A convergence result on the lengths of Markovian loops. Electron. Commun. Probab. 20. no. 73, 1–11. (2015)

work page 2015

[7] [7]

The Brownian loop soup

Gregory Lawler, Wendelin Werner. The Brownian loop soup. PTRF 128 565-588 (2004)

work page 2004

[8] [8]

Lando and A.K

S.K. Lando and A.K. Zvonkin. Graphs on Surfaces and their Applications. Springer. (2004)

work page 2004