Coupled but distant

Gady Kozma; Itai Benjamini

REVIEW 4 major objections 4 minor 9 references

Coupled but distant

T0 review · 4 major / 4 minor · reviewed 2026-08-11 · deepseek-v4-flash

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

Pith's one-line read Two simple random walks on the four-dimensional integer lattice can be coupled so that, with positive probability, their traces never meet.

desk verdict New d=4 non-intersection coupling via Hall's theorem, but Lemma 8's coupling measure is not properly defined as written, and the 2^T normalization is wrong. read the letter →

arxiv 2412.16600 v1 pith:BNBQFFE7 submitted 2024-12-21 math.PR

Itai Benjamini , Gady Kozma This is my paper · ORCID

classification math.PR MSC 60G5060J1005D15

keywords randomwalkcouplingintersectionoftracesZ^4Hall'smarriagetheoremhittablepathsmomentmethodtransientgraphs

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

The paper proves that in $Z^{4}$, two simple random walks started at neighbouring sites can be coupled so that their paths never visit a common vertex, and this happens with positive probability. This settles the borderline case of a question that is trivial in dimension at least 5, impossible in dimension 2, and still open in dimension 3. The proof is non-constructive: it uses Hall's marriage theorem to show that, at every scale, enough short paths of the two walkers can be paired so that paired paths are disjoint and end far apart, and then these matchings are combined over growing balls. If the argument is correct, it shows that dimension 4 belongs to the side where such a coupling exists.

What carries the argument

The load-bearing object is a bipartite matching between two sets A1 and A2 of length-T paths, each set consisting of paths that satisfy a good event and have a small chance of being hit by the other walker. Edges join paths whose traces are disjoint and whose first hitting points on the outer sphere are more than m / log m apart. Hall's condition is verified by Lemma 6, which bounds the probability that a random walk is hittable with probability at least 1 − ε by C ε / $log^{{1/4}}$ n. The coupling measure is then obtained from the matching φ by chopping both paths at the time they first reach the outer sphere. The proof of Lemma 6 itself rests on moment estimates for the sum of intersections of one walk with many independent walks, together with 'good time' estimates that guarantee a typical walk has many periods where future intersections are abundant.

What would settle it

Recompute the normalization in Lemma 8 explicitly: for Ai sets of nearest-neighbour paths of length T in $Z^{4}$, count |Ai|, define the measure by summing over the matching after chopping, and check whether the resulting total mass is at most 1 and whether the marginal on the first coordinate equals the original stopped random walk law. If the total mass exceeds 1, or if the marginal weights a stopped path by the number of its length-T extensions, then the Hall-matching argument as written does not yield a coupling and the proof fails at this point.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is the theorem: in $Z^{4}$, for two simple random walks R1 and R2 started at neighbouring sites, there exists a coupling with P(R1 ∩ R2 = ∅) > 0. The construction is inductive. The walkers are coupled while running from the boundary of a ball of radius $2^{{n^2}}$ to a ball of radius $2^{{(n+1)^2}}$; at each step a Hall-marriage matching pairs paths that do not intersect and whose endpoints are separated by at least m / log m, where m is the outer radius. The probabilistic input is that a typical random walk path is 'unhittable': with high probability, a second independent walk meets it with probability not close to 1, and the event that it is hittable has probability at most about ε / $log^{{1/4}}$ n. The proof obtains this from moment bounds on the number of intersections with many independent walks, following classical intersection estimates for four-dimensional random walks.

Load-bearing premise

The construction's validity depends on the step where the matching from Hall's theorem is converted into a true coupling; the proof asserts this without checking the marginals, and its path counting appears to use 2^T as the number of length-T walks, while in $Z^{4}$ there are 8^T such walks, so this normalization step is not yet established.

Editorial extensions

If this is right

If the theorem is correct, then in Z^4 two random walkers started at neighbouring sites can be coupled so that the event of disjoint traces has positive probability.
The proof yields a quantitative hittability estimate: the probability that a random walk starting on the boundary of a large ball is hittable by an independent walk with probability at least 1−ε is at most K ε / log^{1/4} n.
The cases left open by the paper are dimension 3 and the question of whether a Markovian coupling exists.
The paper states no application for the theorem.

Reading between the lines

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

The normalization in Lemma 8 appears to use 2^T as the number of length-T nearest-neighbour paths, while in Z^4 there are 8^T such paths; if the count is merely a typo and the true count is used, the marginals of the chopped measure still need to be checked separately.
A stronger statement may be within reach: the same quantitative hittability bounds could allow a coupling where the probability of intersection is made arbitrarily small, not merely positive.
If the marginals of the Hall-marriage measure cannot be verified, the theorem's proof would need a different construction, leaving the existence of such a coupling in Z^4 unresolved.
The proof's structure suggests a possible numerical test in Z^3: if the analogue of Lemma 6 fails there, that would explain why the three-dimensional case is hard.

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

Quick take: the theorem is new and the Hall-marriage strategy is genuinely different, but Lemma 8 as written does not produce a verified coupling. The counting uses 2^T where Z^4 has 8^T length-T paths, and the measure defined at the end of the lemma has no marginals check. These are fixable, but they are load-bearing: the induction starts with that lemma.

What's good: the result is the first positive construction in d=4, and the paper says plainly that d≥5 is trivial, d=2 impossible, d=3 open. The moment estimates in Lemmas 2, 5, and 6 are detailed and mostly follow Lawler's method; they look plausible. Lemma 9's use of good times is in the same spirit. The paper is honest about having no application and asks sensible open questions. The non-explicit use of Hall's marriage theorem is a nice idea. The citation to Kozma's Lemma A.11 is a published result, so not a circularity problem.

Soft spots: Lemma 8 is the one place where an actual coupling measure is constructed, and it has real issues. First, the normalization: A'_i is described as 'paths of length T' on Z^4, so the natural count is 8^T, not 2^T. The proof asserts |A_i| ≥ 2^T(1 - Cµ) from Lemma 6, but Lemma 6 is a probability bound and the conversion to this cardinality count is never explained. If 2^T is a typo for 8^T, the rest of the cardinality arguments have to be rescaled, and the measure defined at the end still isn't a probability measure as written: it puts mass |A1|/2^T on the matched pairs and then adds an arbitrary coupling on the complements, with no check that the total mass is 1 or that the marginals are the stopped SRW laws. Without that check, the inductive step has no verified starting measure. The chopped paths introduce many-to-one multiplicities that are not controlled. These are fixable in principle—define the measure on full length-T paths with 8^{-T} weights, then the marginals work out—but the current text doesn't do it. The direction of the inequality in the definition of A_i also looks off at first glance, though the later argument may be salvageable.

For whom: this is for the random-walk coupling crowd and anyone interested in intersection properties of SRW. It's a solid contribution to that niche, not a field-reshaping one.

Recommendation: send it to a serious referee. The theorem deserves to be in the literature if Lemma 8 can be repaired. But the referee should be asked to check the normalization and marginals carefully; as written, the proof doesn't close.

Referee Report

4 major / 4 minor

Summary. The paper claims that in Z^4 one can couple two simple random walks, started at neighboring sites, so that their traces are disjoint with positive probability. The proof uses Hall's marriage theorem: Lemma 6 shows that the set of paths that are 'hittable' by an independent walk with probability close to 1 is small, and Lemma 8 turns this into a coupling on one annulus via a Hall matching between path sets. An induction over growing scales then yields the theorem. The moment estimates in Lemmas 1-7 and Lemma 9 follow Lawler's methods and are presented in detail; the final step, however, rests on Lemma 8, whose proof as written has several load-bearing gaps and errors, and on a final induction estimate that appears arithmetically incorrect.

Significance. If the theorem is correct, it is a striking and non-obvious result: it exhibits a coupling of two transient random walks in the critical dimension 4 that keep apart with positive probability, complementing the known impossibility in d=2 and triviality in d≥5. The strategy of proving existence via Hall's marriage theorem and moment estimates on intersection counts is original and potentially useful. The paper is honest about having no application and about the open three-dimensional case. The detailed estimates in Lemmas 1-7 and Lemma 9 appear plausible and are a valuable contribution in themselves. However, the central coupling construction in Lemma 8 is not rigorously established as written, and the induction in the proof of the theorem contains an arithmetic error; both seem fixable but require substantial revision.

major comments (4)

[Lemma 8, paragraph starting 'Let A\'i be the set of paths...'] The counting normalization is inconsistent with the state space. The proof uses 2^T as the number of length-T paths in Z^4, but a nearest-neighbor path of length T on Z^4 has 8^T possibilities. Consequently, the statements |A_i| ≥ 2^T(1−Cµ), |∂{γ}| > µ2^T, and ε = |B|2^{−T} are not probabilities under the uniform measure on path space. Lemma 6 controls probabilities with respect to the uniform measure on all length-T paths, i.e., relative to 8^T. This is not a cosmetic issue: for example, if P(E_i) ≈ log^2 n/n^2, then |A'_i| ≈ 8^T log^2 n/n^2, so the claimed lower bound |A_i| ≥ 2^T(1−Cµ) is not a consequence of Lemma 6 and is in fact numerically incompatible with the smallness of P(E_i). The proof must be renormalized with 8^T wherever counts are compared with probabilities.
[Lemma 8, definition of A_i] The definition of A_i appears to be inverted. The text defines A_i = A'_i \ {γ : P(R_{3−i}∩γ ≠ ∅) < 1−3µ}, which keeps the highly hittable paths, and then claims 'by definition, every γ ∈ A1 satisfies that P(R2∩γ ≠ ∅) > 3µ'. Later it also claims 'any γ ∈ A2 satisfies that P(γ∩R1 = ∅) > 3µ'; these two statements are inconsistent with the displayed definition. The intended definition must be the complement, A_i = A'_i \ {γ : P(R_{3−i}∩γ ≠ ∅) ≥ 1−3µ}, so that every retained path has avoidance probability at least 3µ. Under the displayed definition, the derivation of the lower bound |∂{γ}| > µ2^T does not follow, because a highly hittable path has few avoiding partners, not many.
[Lemma 8, final paragraph ('The coupling is derived from φ...')] The measure µ(γ,δ) is not shown to have the correct marginals. The text asserts that a coupling is a measure and defines µ+ν, but it never verifies that the first marginal of this measure equals the law of R1 stopped at ∂B(m), nor that the second marginal equals that of R2. A verification would need to use the fact that the events E_i depend only on the path up to the first hitting time of ∂B(m), so that the sets A_i are extension-closed, and that a stopped path of length ℓ has exactly 8^{T−ℓ} extensions to length T. The role of the 'arbitrary coupling' ν on A^c_1 × φ(A1)^c is also left unspecified; to obtain correct marginals one must take ν to be the independent product of the stopped-walk laws restricted to those complementary sets. As written, the construction does not establish that the joint distribution is a coupling of the two random walks.
[Proof of the theorem, final induction estimate] The per-step success probability is miscomputed. The text states that lemma 8 gives success probability 1 − C(P(E1)+P(E2)+1/log 2n^2) = 1 − O(log^2 n/n^2). But 1/log(2n^2) ≈ 1/(2 log n), which dominates the stated P(E_i) ≈ log^2 n/n^2, so the per-step failure probability is of order 1/log n, not O(log^2 n/n^2). Since ∑_{n≥n1} 1/log n diverges, the lower bound p_{n+1} ≥ p_n(1 − C/log n) does not keep p_n bounded away from zero; the induction as written fails. The proof can likely be repaired by choosing steps at doubly exponentially growing scales, but this is not what the manuscript does and the displayed equality is false.

minor comments (4)

[Lemma 8] The normalization 2^T appears in several places ('|A_i| ≥ 2^T(1−Cµ)', '|∂{γ}| > µ2^T', 'ε = |B|2^{−T}') and must be made consistent with the chosen path-counting convention; as written these expressions do not agree with each other or with the uniform measure on paths.
[Lemma 8, last paragraph] The phrase 'ν is an arbitrary coupling of A^c_1 with φ(A1)^c (say i.i.d.)' is too vague; to make the marginal check work, ν must be the independent product of the relevant stopped-walk laws, and this should be stated explicitly.
[Throughout] There are numerous OCR-style typos and duplicated words, e.g., 'In this caseIn this case', 'SinceSince', 'the effectthe effect', and 'ww' for 'x_v − x_w'. These should be cleaned up in revision.
[Lemma 5] The notation 'R0[σ0, σ0 + n]' appears with σ0 undefined; presumably it should be σ_i or the time of the first intersection. Please clarify.

Circularity Check

0 steps flagged · score 2.0 of 10

No circular derivation: the coupling is produced by Hall's theorem from independent random-walk estimates, not from its own conclusion.

full rationale

The paper's theorem is an existence statement, and its proof is a constructive (non-explicit) Hall-marriage argument, not a consequence of the claimed result. The estimates feeding Lemma 8—Lemmas 1 through 6 and Lemma 9—are derived from standard Green-function estimates (Lawler's book, an external source), elementary random-walk lemmas, and moment calculations. The only author-self-citation occurs in Appendix Lemma A.11, where the authors cite Kozma's published Acta Mathematica paper for a classic exit-probability estimate used inside Lemma 9. That cited result is a standalone published proof, is not a restatement of the target theorem, and does not by itself force the coupling to exist; the main derivation has substantial independent content. Possible issues in Lemma 8 concerning the counting normalization or verification of marginals would be proof gaps or correctness concerns, not circularity: the lemma does not define its measure in terms of the conclusion, and no fitted parameter is renamed as a prediction. There is no equation in the paper that is equivalent by construction to the theorem, and no uniqueness claim imported from the authors' prior work is used to rule out alternatives. The score of 2 reflects only the minor self-citation in Lemma A.11, which is not load-bearing for the central claim.

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

The proof introduces no new entities or fitted constants. It relies on standard random walk theory from Lawler's book, the local central limit theorem, Harnack's inequality, and one self-cited published theorem for a ball-exit bound. The constants C, c, K, M, and lambda are existential and can be chosen once the estimates are proven; they are not data-fitted.

assumptions (6)

standard math Green's function asymptotic G(0,x) ≈ |x|^{-2} in Z^4
Invoked in Lemma 1 and Lemma A.1 via Lawler [5, Thm 1.5.4]; underpins the main intersection estimates.
standard math Local central limit theorem for simple random walk on Z^4
Used in Lemmas A.2, A.4, A.7 and in the proof of Lemma 9 to estimate occupation probabilities.
standard math Harnack inequality for positive harmonic functions on Z^4
Used in Lemma A.10, cited to Lawler-Limic [6, Thm 6.3.9], to bound the exit point distribution.
standard math Hall's marriage theorem
The core combinatorial tool in Lemma 8 to produce a matching between path sets; standard.
domain assumption Exit-ball-before-half-space bound: P(R exits B(x,k) before B(n)) ≤ C(n-|x|)/k
Lemma A.11, cited to the second author's own paper [4]; load-bearing in Lemma 9 to show random walks are good with high probability.
standard math Lawler's intersection estimates for random walks (theorems 4.3.3 and 4.4.1 in [5])
Used in Lemma A.8 and in the initial step of the theorem proof to get positive probability that independent walks do not intersect.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Coupled but distant." pith.science (2026). https://pith.science/paper/BNBQFFE7

@misc{pith2026241216600,
  author       = {Pith},
  title        = {Pith review of: Coupled but distant},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/BNBQFFE7}},
  note         = {Machine review of arXiv:2412.16600}
}

read the original abstract

We construct a coupling of two random walks in 4 dimensions so that their traces do not intersect with positive probability.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

9 extracted references · 9 canonical work pages

[1]

Holroyd, James Martin, David B

Omer Angel, Alexander E. Holroyd, James Martin, David B. Wilson and Peter Wink- ler, Avoidance coupling. Electron. Commun. Probab. 18 (2013), no. 58, 13 pp

work page 2013
[2]

Itai Benjamini, Ori Gurel-Gurevich and Oded Schramm,Cutpoints and resistance of random walk paths. Ann. Probab. 39:3 (2011), 1122–1136

work page 2011
[3]

Fifth edition

Rick Durrett, Probability—theory and examples. Fifth edition. Cambridge Series in Statistical and Probabilistic Mathematics, 49. Cambridge University Press, Cam- bridge, 2019

work page 2019
[4]

Acta Math

Gady Kozma,The scaling limit of loop-erased random walk in three dimensions. Acta Math. 199 (2007), no. 1, 29–152

work page 2007
[5]

Lawler,Intersections of random walks

Gregory F. Lawler,Intersections of random walks. Probability and its Applications. Birkhäuser Boston, Inc., Boston, MA, 1991

work page 1991
[6]

Lawler and Vlada Limic,Random walk: a modern introduction

Gregory F. Lawler and Vlada Limic,Random walk: a modern introduction. Cambridge Studies in Advanced Mathematics, 123. Cambridge University Press, Cambridge, 2010

work page 2010
[7]

Coupling from the past

David A. Levin and Yuval Peres,Markov chains and mixing times. Second edition. With contributions by Elizabeth L. Wilmer. With a chapter on "Coupling from the past” by James G. Propp and David B. Wilson. American Mathematical Society, Providence, RI, 2017

work page 2017
[8]

Corrected reprint of the 1992 ori- ginal

Torgny Lindvall,Lectures on the coupling method. Corrected reprint of the 1992 ori- ginal. Dover Publications, Inc., Mineola, NY, 2002

work page 1992

Show all 9 references

[9]

various authors,Hall’s marriage theorem, Wikipedia, retrieved 6/4/2024

2024

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [1]

Holroyd, James Martin, David B

Omer Angel, Alexander E. Holroyd, James Martin, David B. Wilson and Peter Wink- ler, Avoidance coupling. Electron. Commun. Probab. 18 (2013), no. 58, 13 pp

work page 2013

[2] [2]

Itai Benjamini, Ori Gurel-Gurevich and Oded Schramm,Cutpoints and resistance of random walk paths. Ann. Probab. 39:3 (2011), 1122–1136

work page 2011

[3] [3]

Fifth edition

Rick Durrett, Probability—theory and examples. Fifth edition. Cambridge Series in Statistical and Probabilistic Mathematics, 49. Cambridge University Press, Cam- bridge, 2019

work page 2019

[4] [4]

Acta Math

Gady Kozma,The scaling limit of loop-erased random walk in three dimensions. Acta Math. 199 (2007), no. 1, 29–152

work page 2007

[5] [5]

Lawler,Intersections of random walks

Gregory F. Lawler,Intersections of random walks. Probability and its Applications. Birkhäuser Boston, Inc., Boston, MA, 1991

work page 1991

[6] [6]

Lawler and Vlada Limic,Random walk: a modern introduction

Gregory F. Lawler and Vlada Limic,Random walk: a modern introduction. Cambridge Studies in Advanced Mathematics, 123. Cambridge University Press, Cambridge, 2010

work page 2010

[7] [7]

Coupling from the past

David A. Levin and Yuval Peres,Markov chains and mixing times. Second edition. With contributions by Elizabeth L. Wilmer. With a chapter on "Coupling from the past” by James G. Propp and David B. Wilson. American Mathematical Society, Providence, RI, 2017

work page 2017

[8] [8]

Corrected reprint of the 1992 ori- ginal

Torgny Lindvall,Lectures on the coupling method. Corrected reprint of the 1992 ori- ginal. Dover Publications, Inc., Mineola, NY, 2002

work page 1992