Coupling Brownian loop soups and random walk loop soups at all polynomial scales

Wei Qian

REVIEW 3 major objections 4 minor 3 cited by

This paper constructs a coupling between Brownian loop soups and random walk loop soups that works for every polynomial scale of loop length, in every dimension, for both discrete- and continuous-time random walks, with error probability th

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-03 12:27 UTC pith:AVD6PUPX

load-bearing objection The main theorem is false as stated: the Brownian and random-walk selection conditions pick different block indices for every d≠2 (and for continuous-time d=2), so the claimed one-to-one correspondence cannot hold; the underlying a_n construction is a good idea that might be salvageable. the 3 major comments →

arxiv 2601.02992 v3 pith:AVD6PUPX submitted 2026-01-06 math.PR

Coupling Brownian loop soups and random walk loop soups at all polynomial scales

Wei Qian This is my paper

classification math.PR MSC 60J6560G5560J27

keywords Brownian loop souprandom walk loop soupcouplingKMT couplingmesoscopic loopspolynomial scalesPoisson point processcontinuous-time random walk

verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

The paper claims to remove the long-standing restriction θ>2/3 in dimension 2 and θ>2d/(d+4) in higher dimensions for the coupling between Brownian and random walk loop soups. The new coupling works for all θ in (0,2), meaning it controls loops at all polynomial scales, including the previously inaccessible mesoscopic loops. The central new idea is a carefully chosen increasing sequence a_n ≈ 2n/d that aligns the block boundaries of the two loop soups, so that each block of random walk loops has exactly the same total mass as the corresponding block of Brownian loops. The result would matter because mesoscopic loops are believed to influence scaling limits of connectivity and multiplicative chaos, and no previous coupling controlled them simultaneously over a macroscopic region.

Core claim

The main theorem asserts that for every d≥1, θ∈(0,2), and a>0, there exists a coupling of the continuous-time random walk loop soup and the Brownian loop soup such that, outside an event of probability at most cλr^d N^{-a}, there is a one-to-one correspondence between random walk loops with rescaled time larger than N^{θ-2} and Brownian loops with χ_N(t_γ) larger than N^{θ-2}, with time difference O(N^{-2}) and spatial difference O(N^{-1} log N) (or a weaker bound in d≥3 for discrete time). The proof constructs the Brownian loop soup from the random walk loop soup by using a sequence a_n defined by a recurrence that matches the loop mass in each block, together with a KMT-type bridge couplin

What carries the argument

The key object is the increasing sequence {a_n} defined by the recurrence a_1^{-d/2} - a_n^{-d/2} = (2π)^{d/2}(d/2)Q_d(n) (and its discrete-time analogue), which satisfies a_n = 2n/d + O(1). It is used to define the time-slicing function χ_N, and it guarantees that the total mass of continuous-time random walk loops of length in [2n,2n+2] equals the total mass of Brownian loops of length in [a_n,a_{n+1}], enabling a block-by-block Poisson coupling. This sequence is what replaces the ad-hoc thresholds in earlier couplings and allows the removal of the θ lower bound.

Load-bearing premise

The proof assumes that the two threshold conditions in the theorem select exactly the same collection of block indices; this alignment is exact only when the constant in a_n ≈ 2n/d matches the random-walk time rescaling (d=2), so for d≠2 the two sets may not have equal sizes.

What would settle it

For d=1 and d=3, compute the expected number of continuous-time random walk loops with rescaled time > N^{θ-2} and of Brownian loops with χ_N(t_γ) > N^{θ-2}; their ratio converges to a constant different from 1, which would contradict the existence of a bijection on an event of probability tending to 1.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

If correct, the coupling controls mesoscopic loops of diameter as small as N^{-1+ε} for any ε>0, which were previously inaccessible.
The error probability can be made N^{-a} for any a>0 (with a trade-off in the spatial bound for d≥3), much stronger than earlier polynomial bounds.
The result applies in all dimensions d≥1 and both discrete- and continuous-time random walks, with a KMT bridge coupling obtained for the continuous-time case.
It provides a route to quantitative convergence rates for random walk loop soup cluster boundaries to CLE, since global control of mesoscopic loops is needed.
The block-matching construction is simple enough that it may generalize to other discrete/continuous process pairs whose loop measures can be matched by reparametrization.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The threshold alignment in the theorem statement appears to rely on a_n ≈ 2n/d matching the random-walk time rescaling; this holds exactly only for d=2, so for other dimensions the two sets in the theorem may have different cardinalities and the claimed bijection may need the thresholds adjusted, for example by replacing N^{θ-2} with (2/d)N^{θ-2} on the random-walk side.
If such an adjustment is made, the same block-matching proof would likely go through, suggesting the main mathematical contribution is the a_n-sequence construction rather than the exact threshold values in the theorem.
The new KMT coupling for continuous-time random walk bridges in all dimensions could be applied to other problems requiring strong approximations of random walk paths, independent of the loop soup context.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

The main theorem is false as stated: the Brownian and random-walk selection conditions pick different block indices for every d≠2 (and for continuous-time d=2), so the claimed one-to-one correspondence cannot hold; the underlying a_n construction is a good idea that might be salvageable.

read the letter

You should know up front that the central theorem of this paper is not correct. The construction pairs Brownian and random-walk loops block by block using a sequence a_n ~ 2n/d, but the selection conditions in Theorems 1.2 and 1.3 do not align on the two sides except in the discrete-time d=2 case. The Brownian side uses chi_N(t)>N^{θ-2}, which selects exactly blocks n>N^θ. The random-walk side uses t_{\tildeγ}>N^{θ-2}, which for a loop in block n means n>(d/2)N^θ. For d>2 the Brownian set has extra blocks (N^θ,(d/2)N^θ); for d<2 the random-walk set has those extras. The expected number of unpaired loops is ~ λr^d N^{d(1-θ/2)}, unbounded for every θ<2, so no bijection can hold on an event of probability tending to one. Even in continuous time with d=2, a one-block boundary mismatch appears when N^θ is not an integer, and the expected extra loops ~ N^{2-2θ} already violate the claimed arbitrary-a error bound for small θ. Section 3.3 only controls bad events for N^θ

That said, the paper is not careless. The observation that only the leading term of the loop-mass Taylor expansion is needed, and the definition of a_n via the recurrence (3.2) to make block masses match exactly, are genuinely new and would be useful if the targeting were fixed. The KMT coupling between continuous-time random walk bridges and Brownian bridges (Lemma 2.5) is a clean addition, and the exposition is clear. The citation to [1] as claiming sharpness is honest, and the author correctly identifies that the prior barrier may not be inherent.

The appropriate audience is researchers working on loop soup couplings and scaling limits. The idea may be salvageable by reformulating the theorem with a block-index condition on both sides (or by adding the unpaired loops in a controlled way). As written, the claim is false. I would not accept this version, but I would send it to peer review rather than desk reject: the construction is novel enough that a referee should document the mismatch and give the author a chance to fix it. If the fix works, the result would be a real advance.

Referee Report

3 major / 4 minor

Summary. The paper proposes a coupling between Brownian loop soups in R^d and random-walk loop soups on Z^d (both continuous- and discrete-time), claimed to hold for all d≥1 and for all polynomial scales, i.e. for every θ∈(0,2). The construction refines the Lawler–Trujillo Ferreras / Sapozhnikov–Shiraishi framework by introducing a block sequence {a_n} chosen so that the total mass of random-walk loops in block n equals the Brownian-loop mass in the corresponding time interval. The author then uses KMT-type bridge couplings to control the spatial and temporal errors for paired loops. The main theorems (Theorems 1.2 and 1.3) assert a one-to-one correspondence between random-walk loops with rescaled time > N^{θ−2} and Brownian loops with χ_N(t)>N^{θ−2}, with failure probability at most cλr^d N^{-a} for arbitrary a>0.

Significance. If the main theorems were correct, this would be a substantial advance: it would remove the long-standing restriction θ>2/3 in d=2 and θ>2d/(d+4) in d≥3, and would provide couplings at all polynomial scales with very small failure probability. The construction is elegant: the recurrence defining {a_n} makes the block masses match exactly by construction, and the proof transparently identifies the KMT estimates needed. Those are genuine strengths. However, the central claim is undermined by a mismatch between the two threshold conditions used in the theorem, as detailed below. The paper also contains useful intermediate results and a clear exposition of the continuous-time case.

major comments (3)

[Theorems 1.2–1.3, Eq. (1.13), Lemma 3.1] The two threshold conditions in the theorem select different sets of block indices. For a Brownian loop in block n, χ_N(tγ)=n/N^2 by (1.13), so the condition χ_N(tγ)>N^{θ−2} is equivalent to n>N^θ. For a continuous-time random-walk loop in block n, the rescaled time is t_{\tildeγ}=\tilde T/(dN^2) with \tilde T∈[2n,2n+2), so t_{\tildeγ}>N^{θ−2} is equivalent, up to boundary effects, to n>(d/2)N^θ. Lemma 3.1 gives a_n=2n/d+O(1), so the coupling pairs block n with block n. Thus for d≠2 the selected block indices differ on an interval of length ~|d/2−1|N^θ. For d>2 the Brownian set contains extra blocks n∈(N^θ, dN^θ/2); for d<2 the random-walk set contains extra blocks n∈(dN^θ/2,N^θ). The expected number of such unpaired loops is of order λr^d N^{d(1−θ/2)}, which is unbounded for every θ<2. Hence the two sets cannot have a one-to-one correspondence on an event of probability tending to 1, le
[Section 3.3, event A; Section 3.4] The proof does not address the discrepancy interval. In Section 3.3, the event A is defined only over N^θ<n<N^k, and the subsequent estimates (3.9)–(3.11) control only that range; (3.7) controls n≥N^k. The mismatch interval between (d/2)N^θ and N^θ is never treated. Consequently, even if one accepted the bridge estimates, the sentence on p. 11 that 'on the event Ac ... the coupling satisfies the conditions of Theorem 1.2' is unjustified: the event Ac says nothing about the extra loops in the discrepancy interval, which are precisely the loops that break the claimed bijection. The same gap is inherited by the discrete-time proof in Section 3.4.
[Theorem 1.2, continuous-time d=2] The mismatch is not only a factor-d/2 artefact. For d=2 in continuous time, the block-index thresholds agree asymptotically, but there is still a one-block mismatch at the boundary. For block n=N^θ, the random-walk condition t_{\tildeγ}>N^{θ−2} is satisfied with probability 1 because \tilde T∈[2N^θ,2N^θ+2), while the Brownian condition χ_N(tγ)>N^{θ−2} is not satisfied because χ_N=N^{θ−2}. The expected number of unmatched random-walk loops in this boundary block is of order λr^2N^{2−2θ}, which is unbounded for θ<1. Thus even in the dimension where the original Lawler–Trujillo Ferreras coupling works, the strengthened theorem fails for the range θ∈(0,1).

minor comments (4)

[Eq. (3.4)] In the displayed sum, the index in Q_d(n) should be Q_d(k); as written, the summand does not depend on the summation variable.
[Section 3.2] The variables Y(n,z;m) are described as 'complex random variables' but then are said to be uniform on a d-dimensional hypercube. For d≠2 they should be called random vectors, not complex random variables.
[Section 2.3] In the definition of μcrw, the transition kernel is written p_t(x,y) but the integral is over z only; presumably this should be p_t(z,z) (or the integral should be over x,y with the kernel evaluated appropriately).
[Theorem 1.3, Eq. (1.18)] For d≥3, the spatial error bound N^{(a−d)/(2d)} log N is not vanishing when a≥d. The theorem states 'for every a>0', which may mislead readers; Remark 1.4 does acknowledge this, but the main statement could be phrased more carefully.

Circularity Check

0 steps flagged

No circularity: the coupling is constructed by deliberately matching block masses, and the substantive closeness estimates come from external KMT-type bridge couplings.

full rationale

The paper's construction in Section 3.2 defines the sequence {a_n} by (3.2) so that (3.3) holds, i.e. Q_d(n) equals the Brownian loop mass of the interval [a_n,a_{n+1}]. This deliberately makes the Poisson counts for each random-walk loop block and each Brownian loop block equal, and this is the standard mechanism of a coupling proof: the two loop soups are built on the same Poisson processes. The one-to-one correspondence at the block level is therefore by construction, not a derived prediction. The nontrivial content of Theorems 1.2 and 1.3 is the spatial closeness of paired bridges and the small failure probability, which are proved using external KMT-type bridge couplings (Theorem 2.4, Lemmas 2.2, 2.3, 2.5) and Poisson tail estimates. These inputs are independent of the theorem's conclusion and are not supplied by a self-citation: the bridge-coupling results are due to Lawler–Trujillo Ferreras, Sapozhnikov–Shiraishi, Dimitrov–Wu, and Komlós–Major–Tusnády, with no overlap with the present author. The possible block-index mismatch between the Brownian threshold chi_N(t)>N^{theta-2} and the random-walk threshold t_{tilde gamma}>N^{theta-2} for d != 2 is a substantive correctness question about whether the constructed bijection covers exactly the claimed sets, but it is not a circularity: the theorem's conclusion is not assumed in the construction; rather the construction may fail to satisfy the stated theorem for those parameters. Remark 1.4 even flags a limitation on the error probability for d>=3. Thus no circular step is present.

Axiom & Free-Parameter Ledger

0 free parameters · 4 axioms · 0 invented entities

No new fitted constants or invented physical entities appear. The sequence {a_n} is a construction, not a parameter fitted to data. The main external inputs are KMT couplings and loop-mass expansions from prior literature.

axioms (4)

standard math KMT bridge couplings: Theorem 2.1 and Lemma 2.2 from [6], Lemma 2.3 from [10], Theorem 2.4 and Lemma 2.5 from [3].
The whole proof is built on these external strong approximations of random-walk bridges by Brownian bridges.
standard math Asymptotic expansions for random-walk and continuous-time loop masses: (1.7), (1.8), and (1.11) with (2.4).
The definition of a_n in (3.2) and Lemma 3.1 rely on these expansions from [2], [5], and [6].
domain assumption Brownian loop soup scaling invariance and the Poissonian structure of the loop soups.
These are standard facts used to recognize the constructed collection Aλ as a Brownian loop soup.
standard math Coordinate decomposition of the continuous-time random walk bridge into independent one-dimensional bridges.
Used to derive Lemma 2.5 from Theorem 2.4.

pith-pipeline@v1.3.0-alltime-deepseek · 12072 in / 37688 out tokens · 328206 ms · 2026-08-03T12:27:38.785838+00:00 · methodology

0 comments

read the original abstract

Lawler and Trujillo Ferreras constructed a well-known coupling between the Brownian loop soups on $\mathbb{R}^2$ and the (discrete-time) random walk loop soups on $\mathbb{Z}^2$ (one rescales the random walk loops by $1/N$, their time parametrizations by $1/(2N^2)$, and lets $N\to \infty$), which led to numerous applications. It nevertheless only holds for loops with time length at least $N^{\theta-2}$ for $\theta \in(2/3,2)$. In particular, there is no control on mesoscopic loops with time length less than $N^{-4/3}$ (i.e. roughly diameter less than $N^{-2/3}$). This coupling was subsequently extended by Sapozhnikov and Shiraishi to $\mathbb{Z}^d$ with $d\ge 3$, for loops with time length at least $N^{\theta-2}$, for $\theta \in(2d/(d+4),2)$. In this paper, we find a simple way to remove the restriction $\theta>2d/(d+4)$, so that such a coupling works for all $\theta\in (0,2)$, i.e. for loops at all polynomial scales. We establish couplings for both discrete-time and continuous-time random walk loop soups on $\mathbb{Z}^d$, for $d\ge 1$. As an intermediate step, we also establish a KMT coupling between the continuous-time random walk bridge on $\mathbb{Z}^d$ and the Brownian bridge on $\mathbb{R}^d$.

discussion (0)

Forward citations

Cited by 3 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

The Brownian loop-catcher
math.PR 2026-07 conditional novelty 9.0

Brownian loop-catchers are the unique random subsets of a Brownian excursion that, when decorated by an independent loop soup of intensity -c/2, recover the excursion; they interpolate between LERW and Brownian trace.
Multi-window trace connectivity in subcritical planar Brownian loop soups
math.PR 2026-07 accept novelty 7.0

For q≥3 separated shrinking disks, the probability that one loop-soup cluster meets all q disks factors as the product of one-arm probabilities up to logarithmic exponents.
Multi-window trace connectivity in subcritical planar Brownian loop soups
math.PR 2026-07 unverdicted novelty 6.0

In subcritical planar Brownian loop soups, the probability that one trace cluster meets q ≥ 3 separated shrinking discs factorizes as the q-fold product of one-arm probabilities, up to arbitrary loss in the logarithmi...

Reference graph

Works this paper leans on

12 extracted references · cited by 2 Pith papers

[1]

Aïdékon, N

E. Aïdékon, N. Berestycki, A. Jego, and T. Lupu. Multiplicative chaos of the Brownian loop soup.Proc. Lond. Math. Soc. (3), 126(4):1254–1393, 2023

2023
[2]

Ball and J

K. Ball and J. Sterbenz. Explicit bounds for the return probability of simple random walks. J. Theoret. Probab., 18(2):317–326, 2005

2005
[3]

Dimitrov and X

E. Dimitrov and X. Wu. KMT coupling for random walk bridges.Probab. Theory Related Fields, 179(3-4):649–732, 2021

2021
[4]

Komlós, P

J. Komlós, P. Major, and G. Tusnády. An approximation of partial sums of independentRV’s and the sampleDF. I.Z. Wahrscheinlichkeitstheorie und Verw. Gebiete, 32:111–131, 1975

1975
[5]

G. F. Lawler and V. Limic.Random walk: a modern introduction, volume 123 ofCambridge Studies in Advanced Mathematics. Cambridge University Press, Cambridge, 2010

2010
[6]

G. F. Lawler and J. A. Trujillo Ferreras. Random walk loop soup.Trans. Amer. Math. Soc., 359(2):767–787, 2007

2007
[7]

G. F. Lawler and W. Werner. The Brownian loop soup.Probab. Theory Related Fields, 128(4):565–588, 2004

2004
[8]

Le Jan.Markov paths, loops and fields, volume 2026 ofLecture Notes in Mathematics

Y. Le Jan.Markov paths, loops and fields, volume 2026 ofLecture Notes in Mathematics. Springer, Heidelberg, 2011. Lectures from the 38th Probability Summer School held in Saint- Flour, 2008, École d’Été de Probabilités de Saint-Flour

2026
[9]

T. Lupu. Convergence of the two-dimensional random walk loop-soup clusters to CLE.J. Eur. Math. Soc. (JEMS), 21(4):1201–1227, 2019

2019
[10]

Sapozhnikov and D

A. Sapozhnikov and D. Shiraishi. On Brownian motion, simple paths, and loops.Probab. Theory Related Fields, 172(3-4):615–662, 2018

2018
[11]

Sheffield and W

S. Sheffield and W. Werner. Conformal loop ensembles: the Markovian characterization and the loop-soup construction.Ann. of Math. (2), 176(3):1827–1917, 2012

1917
[12]

van de Brug, F

T. van de Brug, F. Camia, and M. Lis. Random walk loop soups and conformal loop ensembles. Probab. Theory Related Fields, 166(1-2):553–584, 2016. 14

2016

[1] [1]

Aïdékon, N

E. Aïdékon, N. Berestycki, A. Jego, and T. Lupu. Multiplicative chaos of the Brownian loop soup.Proc. Lond. Math. Soc. (3), 126(4):1254–1393, 2023

2023

[2] [2]

Ball and J

K. Ball and J. Sterbenz. Explicit bounds for the return probability of simple random walks. J. Theoret. Probab., 18(2):317–326, 2005

2005

[3] [3]

Dimitrov and X

E. Dimitrov and X. Wu. KMT coupling for random walk bridges.Probab. Theory Related Fields, 179(3-4):649–732, 2021

2021

[4] [4]

Komlós, P

J. Komlós, P. Major, and G. Tusnády. An approximation of partial sums of independentRV’s and the sampleDF. I.Z. Wahrscheinlichkeitstheorie und Verw. Gebiete, 32:111–131, 1975

1975

[5] [5]

G. F. Lawler and V. Limic.Random walk: a modern introduction, volume 123 ofCambridge Studies in Advanced Mathematics. Cambridge University Press, Cambridge, 2010

2010

[6] [6]

G. F. Lawler and J. A. Trujillo Ferreras. Random walk loop soup.Trans. Amer. Math. Soc., 359(2):767–787, 2007

2007

[7] [7]

G. F. Lawler and W. Werner. The Brownian loop soup.Probab. Theory Related Fields, 128(4):565–588, 2004

2004

[8] [8]

Le Jan.Markov paths, loops and fields, volume 2026 ofLecture Notes in Mathematics

Y. Le Jan.Markov paths, loops and fields, volume 2026 ofLecture Notes in Mathematics. Springer, Heidelberg, 2011. Lectures from the 38th Probability Summer School held in Saint- Flour, 2008, École d’Été de Probabilités de Saint-Flour

2026

[9] [9]

T. Lupu. Convergence of the two-dimensional random walk loop-soup clusters to CLE.J. Eur. Math. Soc. (JEMS), 21(4):1201–1227, 2019

2019

[10] [10]

Sapozhnikov and D

A. Sapozhnikov and D. Shiraishi. On Brownian motion, simple paths, and loops.Probab. Theory Related Fields, 172(3-4):615–662, 2018

2018

[11] [11]

Sheffield and W

S. Sheffield and W. Werner. Conformal loop ensembles: the Markovian characterization and the loop-soup construction.Ann. of Math. (2), 176(3):1827–1917, 2012

1917

[12] [12]

van de Brug, F

T. van de Brug, F. Camia, and M. Lis. Random walk loop soups and conformal loop ensembles. Probab. Theory Related Fields, 166(1-2):553–584, 2016. 14

2016