REVIEW 3 major objections 5 minor 23 references
Random walks on random walks: non-perturbative results in high dimensions
T0 review · 3 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash
Pith's one-line read This paper proves that in dimensions five and higher, a walker moving through a Poissonian swarm of independent particles travels at a deterministic linear speed, and in dimension nine and higher the walk converges to Brownian motion.
desk verdict The d≥9 FCLT is not established: the proved logarithmic mixing rate is too slow for the Σ√φ(2t)<∞ premise, and the theorem statement and proof disagree on the exponent; the SLLN/LDB for d≥5 look sound. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central object is the local environment process $(\xi_t)=(\omega_t(X_t), X_{t+1}-X_t)$, recording the environment bit and jump at the walker's current location, together with its uniform mixing coefficient $\varphi(t)$. The coefficient is the largest total-variation distance, over two arbitrary conditioned past observations, between the laws of the future tail of $(\xi_t)$. The key estimate, Theorem 3.5, bounds $\varphi(t)$ by $C(\log t)^{-d/2-2}$. The proof of that estimate rests on Theorem 4.2, a decomposition of the conditioned Poissonian environment: conditionally on any past observation, the field is, in law, the unconditioned field plus extra random-walk trajectories anchored to the observed path, with a stochastically dominated number of extra particles per anchor and heat-kernel decay bounds on their future influence. Strong ellipticity and diverging variance of the walk are then derived from the same decomposition.
What would settle it
One concrete check is to compute $S_d=\sum_{t\ge1}\sqrt{\varphi(2t)}$ under the paper's stated bound: since $\varphi(2t)\le C(\log t)^{-d/2-2}$, the terms are $C(\log t)^{-(d/2+2)/2}$, whose sum diverges for every $d$; this means the stated derivation of Theorem 2.2b requires a strictly stronger mixing estimate than Theorem 3.5 provides.
Extended reading notes
Core claim
The central discovery is that the random walk on random walks is diffusive in high dimensions at every positive density. The paper proves an SLLN and exponential large deviation bounds for $d\ge 5$ (Theorem 2.1), a weak annealed functional central limit theorem for $d\ge 5$ with a slowly varying normalization (Theorem 2.2a), and, for $d\ge 9$ with truly $d$-dimensional finite-range jump kernels, a full annealed functional central limit theorem with constant positive covariance matrix (Theorem 2.2b). The load-bearing estimate is the uniform mixing bound $\varphi(t)\le C(\log t)^{-d/2-2}$ of Theorem 3.5: conditioned on any two possible past observations of the environment along admissible walk trajectories, the laws of the future local environment differ by at most this amount in total variation. Combined with strong ellipticity and diverging variance, that bound feeds the general limit-law machinery of the companion paper [5] to produce the theorems. The paper also contrasts the results with low dimensions: for $d\le 2$, occupation-time tails of the environment decay slower than exponentially, so the exponential large deviation bound cannot hold there.
Load-bearing premise
The load-bearing premise is that the local environment process is uniformly mixing—given any two possible past observations, the future environment seen by the walker becomes close in total variation as the time gap grows—and, for the Brownian limit, that the decay is strong enough for $\sum_{t\ge1}\sqrt{\varphi(2t)}$ to converge; the paper proves the decay but does not show that this particular series converges from its stated logarithmic bound.
Editorial extensions
If this is right
- For $d\ge 5$ and any $\lambda>0$, the walker has a deterministic asymptotic velocity, so arbitrarily sparse moving environments do not induce sub-ballistic motion.
- Exponential large deviation bounds replace the stretched-exponential bounds of the earlier high-dimensional results wherever the models overlap.
- For $d\ge 9$ and truly $d$-dimensional kernels, the annealed walk converges to Brownian motion with positive-definite covariance, so the long-time spread grows as $\sqrt{n}$ in every direction.
- For $d\ge 5$, one-dimensional projections whose variance diverges satisfy a weak annealed functional central limit theorem, with a slowly varying normalization.
- The reduction to a uniform mixing estimate means any later improvement in $\varphi(t)$ automatically sharpens the limit theorems.
Reading between the lines
- Editorial inference (not a paper claim): the d≥9 Brownian conclusion is derived through the summability condition $\sum_{t\ge1}\sqrt{\varphi(2t)}<\infty$, but inserting the paper's own bound $\varphi(2t)\le C(\log t)^{-d/2-2}$ gives a divergent series; the theorem as stated therefore needs a sharper mixing estimate or a different argument.
- Editorial inference: the Section 6 decomposition of conditioned Poisson point processes is developed in a general setting and should transfer to other dynamic environments built from independent transient particles with finite-range motion.
- Editorial inference: the authors expect the logarithmic rate to be non-optimal; if a polynomial decay with exponent greater than 2 could be proved, the same machinery would give the Brownian limit at lower dimensions than the d≥9 threshold.
- Editorial inference: a concrete high-value test is to run the model at $d=4$ and $d=5$ with non-symmetric finite-range kernels and measure the velocity and variance; the paper predicts ballistic and weakly diffusive behaviour in those regimes.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This paper analyzes the random walk on random walks (RWRW) model on Z^d, in which the dynamic environment is a Poissonian field of independent finite-range simple random walks of density λ, and the walker X_t makes a jump with law α(0,·) or α(1,·) according as its current location is vacant or occupied. The main claims are: for d ≥ 5 and every λ > 0, a strong law of large numbers and exponential large deviation bounds for X_t/t (Theorem 2.1); for d ≥ 5, a weak annealed functional CLT with slowly varying normalization (Theorem 2.2a, first part); and, for d ≥ 9, a genuine annealed FCLT with positive definite covariance matrix for truly d-dimensional walks (Theorem 2.2b), together with a constant normalization in the second part of Theorem 2.2a. The proof strategy is to estimate a uniform mixing coefficient φ(t) for the local environment process via a new decomposition of the conditioned Poisson environment (Theorem 4.2, developed in Section 6), obtaining φ(t) ≤ C(log t)^{−d/2−2} (Theorem 3.5), and then to import limit theorems from the authors' companion paper [5, Proposition 3.4], whose hypotheses are φ(t) → 0 (plus strong ellipticity) for the SLLN/LDB and weak FCLT, and Σ√φ(2t) < ∞ for the Brownian scaling limit.
Significance. Conditional on the limit theorems being correct, the paper is a significant advance: it removes the small- or large-density restrictions of Blondel, Hilário, dos Santos, Sidoravicius and Teixeira (2019), extends the SLLN/LDB to all densities in d ≥ 5 with exponential rather than stretched-exponential deviations, and provides the first high-dimensional Brownian scaling result for this model at arbitrary density. The strength of the paper is that the mixing estimate is derived from first principles: the Poisson decomposition of Section 6 and the domination Theorem 4.2 are original tools, applied to classical heat-kernel bounds, with no fitting parameters or definitional circularity; the SLLN/LDB and the slowly varying waFCLT appear sound on my reading. The weakness is that the stress-test concern lands: the d ≥ 9 conclusions (Theorem 2.2b and the constant-σ part of Theorem 2.2a) are derived from a false summability implication, and the exponent in Theorem 3.5 is not the exponent proved in Lemma 5.2. The advertised Brownian scaling results therefore do not follow from the estimates supplied in the manuscript.
major comments (3)
- [Section 3, Proof of Theorem 2.2] The proof asserts that 'when d ≥ 9, by the bound in Theorem 3.5, it follows that Σ_{t≥1} √φ(2t) < ∞', and this summability is the only route used to reach Theorem 2.2b and the constant-σ part of Theorem 2.2a via Proposition 3.4. The implication is false: the estimate φ(2t) ≤ C(log(2t))^{−d/2−2} is compatible with, for example, φ(2t) = C(log(2t))^{−d/2−2}, for which √φ(2t) ≥ c (log t)^{−(d/4+1)}, and Σ_{t≥1} (log t)^{−p} diverges for every p > 0 (by Cauchy condensation, the condensed series is Σ_k 2^k (k log 2)^{−p} = ∞). Hence the bound in Theorem 3.5 does not verify the summability hypothesis of Proposition 3.4i–ii regardless of how the exponent is read, and Theorem 2.2b together with the h(n) = σ assertion of Theorem 2.2a are not established by the submitted argument. Theorem 2.1 and the slowly varying part of Theorem 2.2a are unaffected, since they require only φ(t) → 0, which both exponents supply. Repair requires either a genuinely faster decay of φ(t) (of order t^{−(2+ε)} would suffice for square-root summability) or a transfer principle from [5] that does not demand Σ√φ(2t) < ∞.
- [Section 3, Theorem 3.5, and Section 5, Lemma 5.2] The exponent in Theorem 3.5 does not match the exponent that the proof produces. Theorem 3.5 states φ(t) ≤ C(log t)^{−d/2−2}, while Lemma 5.2 states the bound Cn^{−d/2+2}, and the proof of Theorem 3.5 substitutes (T')^{−d/2+2} with T' = C log T, obtaining (log T)^{−d/2+2}; the two exponents differ by 4. As written, the stated theorem is strictly stronger than anything Lemma 5.2 and the proof of Theorem 3.5 supply, and Remark 3.1 compounds the confusion by describing Corollary 5.3 (exponent −d/2+2) as an improvement over Theorem 3.5. The authors should align the statement of Theorem 3.5, Lemma 5.2, and the proof. Note that this alignment does not fix the gap in Major Comment 1, since both candidate rates are logarithmic and neither implies Σ√φ(2t) < ∞.
- [Section 6.5, Theorem 6.6] The statement of the central decomposition theorem is not well-formed as printed. The display 'P^{(1)}_{I(1)}(·|C^{(1)}) is equal, in law, to Σ_y P^{(1)}_{I(1)}(y|C^{(1)}) Σ_{k=1}^{κ} L(Y^{(k)}|y)' mixes the weights P^{(1)}_{I(1)}(y|C^{(1)}) with an unnormalized sum of marginal laws L(Y^{(k)}|y), so the right-hand side is not readable as a probability measure; the fourth bullet, asserting independence of the Y^{(k)} given y, suggests that the intended object is the joint law of (Y^{(1)},...,Y^{(κ)}) given y, but the sum of marginals does not say this. The first bullet defines y = (y_1,...,y_K) with K ≤ κ, while the sums run to κ, and the third bullet introduces laws L(Y^{(k)}|y) for k > K that do not occur in the display; the stochastic domination sentence in the same theorem similarly writes Σ_{k=1}^{|y|} instead of κ. Since Theorem 6.6 is the basis of Theorem 4.2 and hence of the ellipticity and mixing estimates used throughout the paper, this statement needs a careful rewrite (mixture weights, joint law of the extra paths, and domination) before the proof of Theorem 4.2 can be checked.
minor comments (5)
- [Sections 3 and 4.2] The cross-references to the ellipticity and diverging-variance results are inconsistent with their statements: Section 3 ends with 'the proof of Proposition 3.3', but the result is stated as Lemma 3.3, and Section 4.2 contains 'Proof of Proposition 3.1' for a result stated as Lemma 3.1.
- [Section 3, Eq. (3.2)] In the proof of Lemma 3.3, Eq. (3.2) contains a typo, 'Pω(‖Xn − E(Xn)‖)2 > ǫ)': it should read Pω(‖Xn − E(Xn)‖₂ > ǫ); the notation W^{(j)}_{[1,i_j]} later in the same proof also uses inconsistent subscripts.
- [Section 5, proof of Lemma 5.2] In the proof of Lemma 5.2, the event {τ_m < ∞} is conditioned on Q[0,T] ≥ 1, but T is not defined at that point (the lemma's parameter is n); the closing sentence of the proof also refers to 'Proposition 5.2', which should be Lemma 5.2.
- [Section 4.2, Eqs. (4.6)–(4.8)] The products in (4.7)–(4.8) are indexed inconsistently: the factors run over s = 1,...,R−1 and s = R,...,|γ|, while the preceding lines index particles by i with anchor points z_i = −s; rewriting everything in terms of the anchors z_i would remove the ambiguity, and the symbol '∩' in the middle of (4.6) should be a multiplication sign.
- [Section 2, Theorems 2.1 and 2.2] The main theorems are stated for λ ∈ [0,∞), but the supporting lemmas (Lemma 3.1, Theorem 3.5, Proposition 3.4) assume λ > 0; the degenerate case λ = 0 reduces to a homogeneous random walk with kernel α(0,·) and is trivial, but the papers should say so explicitly.
Circularity Check
No circularity: the RWRW mixing estimate and limit theorems are derived from first principles; the d≥9 summability gap is a correctness issue, not a circular reduction.
full rationale
The paper's load-bearing input is the mixing estimate φ(t) ≤ C(log t)^{-(d/2+2)} (Theorem 3.5), and this is proved inside the paper from the Poisson construction of the environment (Theorem 4.2), a coupling argument (Lemma 5.2), and heat-kernel estimates for simple random walks (Lemma 4.3); no fitted parameter is involved and no conclusion is used to define an input. The final SLLN/LDB and FCLT statements are channeled through Propositions 3.2 and 3.4, which use the authors' companion paper [5]; although this is load-bearing self-citation, it is not circular, because the assumptions of [5] — strong ellipticity and φ-decay — are verified here independently, and [5]'s transfer theorem does not assume the RWRW conclusions. None of the seven circularity patterns is present: the model parameters are not fitted to the predicted quantities, no known result is renamed as a new object, and no uniqueness result or ansatz is imported by citation. A separate, non-circular correctness risk remains in the proof of Theorem 2.2: for d = 9, Theorem 3.5 gives φ(t) ≤ C(log t)^{-6.5}, which does not imply the condition Σ_t √φ(2t) < ∞ required by Proposition 3.4ii, since a series with terms (log t)^{-p} diverges for every p > 0; the asserted implication in the proof of Theorem 2.2 is therefore unsupported. Under the rubric, that is a correctness concern rather than a circular step.
Assumptions & free parameters
assumptions (3)
- standard math Heat kernel upper bounds for finite-range random walks on Z^d, d ≥ 3 (P(Y_t = y) ≤ c t^{-d/2})
- standard math Esseen's concentration function bound for sums of independent non-singular random vectors (Esseen 1968, Theorem 6.2)
- domain assumption Transfer principle of Bethuelsen and Völlering [5]: decay of φ(t) to 0 implies SLLN and LDBs, and the condition Σ√φ(2t)<∞ implies aFCLT
Cite this review
Pith. "Pith review of Random walks on random walks: non-perturbative results in high dimensions." pith.science (2026). https://pith.science/paper/FKZBQI5L
@misc{pith2026241113926,
author = {Pith},
title = {Pith review of: Random walks on random walks: non-perturbative results in high dimensions},
year = {2026},
howpublished = {\url{https://pith.science/paper/FKZBQI5L}},
note = {Machine review of arXiv:2411.13926}
}
abstract
Consider the dynamic environment governed by a Poissonian field of independent particles evolving as simple random walks on $\mathbb{Z}^d$. The random walk on random walks model refers to a particular stochastic process on $\mathbb{Z}^d$ whose evolution at time $t$ depends on the number of such particles at its location. We derive classical limit theorems for this instrumental model of a random walk in a dynamic random environment, applicable in sufficiently high dimensions. More precisely, for $d \geq 5$, we prove a strong law of large numbers and large deviation estimates. Further, for $d\geq 9$, we obtain a functional central limit theorem under the annealed law. These results are non-perturbative in the sense that they hold for any positive density of the Poissonian field. Under the aforementioned assumptions on the dimension they therefore improve on previous work on the model. Moreover, they stand in contrast to the anomalous behaviour predicted in low dimensions.
Reference graph
Works this paper leans on
-
[5]
S. A. Bethuelsen and F. V¨ ollering. Limit laws for random walk in a dynamic path-cone mixing random environment. ArXiv Preprint, 2024. https://arxiv.org/abs/2303.06756
work page Pith review arXiv 2024
-
[6]
O. Blondel, M. R. Hil´ ario, R. S. dos Santos, V. Sidoravicius and A. Teixeira. Random walk on random walks: higher dimensions. Electron. J. Probab., 24:Paper No. 80, 33, 2019
work page 2019
-
[1]
L. Avena and P. Thomann. Continuity and anomalous fluctuations in random walks in dynamic random environments: numerics, phase diagrams and conjectures. J. Stat. Phys. , 147(6):1041–1067, 2012
work page 2012
- [2]
- [3]
-
[4]
S. A. Bethuelsen and F. V¨ ollering. Absolute continuity and weak u ni- form mixing of random walk in dynamic random environment. Electron. J. Probab., 21:Paper No. 71, 32, 2016
work page 2016
-
[7]
O. Blondel, M. R. Hil´ ario, R. S. dos Santos, V. Sidoravicius, and Au - gusto Teixeira. Random walk on random walks: low densities. Ann. Appl. Probab., 30(4):1614–1641, 2020. 34
work page 2020
-
[8]
Sharp threshold for the ballisticity of the random walk on the exclusion process
G. Conchon–Kerjan, D. Kious, and P-F. Rodriguez. Sharp thre shold for the ballisticity of the random walk on the exclusion process. ArXiv preprint, 2024. https://arxiv.org/abs/2409.02096
work page Pith review arXiv 2024
Show all 23 references
-
[9]
J. T. Cox and D. Griffeath. Large deviations for Poisson systems of independent random walks. Z. Wahrsch. Verw. Gebiete , 66(4):543–558, 1984
1984
-
[10]
C. G. Esseen. On the concentration function of a sum of indepe ndent random variables. Z. Wahrscheinlichkeitstheorie und Verw. Gebiete , 9: 290–308, 1968
1968
-
[11]
Hil´ ario, F
M. Hil´ ario, F. den Hollander, V. Sidoravicius, R. ˜Soares dos Santos and A. Teixeira. Random walk on random walks. Electron. J. Probab. , 20: no. 95, 1–35, 2015
2015
-
[12]
M. ˜R. Hil´ ario, D. Kious, and A. Teixeira. Random walk on the simple symmetric exclusion process. Comm. Math. Phys. , 379(1):61–101, 2020
2020
-
[13]
den Hollander, H
F. den Hollander, H. Kesten, and V. Sidoravicius. Random walk in a high density dynamic random environment. Indagationes Mathematicae, 25(4):785 – 799, 2014
2014
-
[14]
Huveneers and F
F. Huveneers and F. Simenhaus. Random walk driven by simple exc lu- sion process. Electron. J. Probab. 20:, 1–42, 2015
2015
-
[15]
Huveneers and F
F. Huveneers and F. Simenhaus. Evolution of a passive particle in a one-dimensional diffusive environment. Electron. J. Probab. 28:, 1–31, 2023
2023
-
[16]
Huveneers
F. Huveneers. Response to a small external force and fluctu ations of a passive particle in a one-dimensional diffusive environment. Phys. Rev. E, 97:042116, Apr 2018
2018
-
[17]
Jara and O
M. Jara and O. Menezes. Symmetric exclusion as a random enviro nment: invariance principle. Ann. Probab., 48(6):3124–3149, 2020
2020
-
[18]
Kesten, M
H. Kesten, M. V. Kozlov, and F. Spitzer. A limit law for random walk in a random environment. Compositio Math. , 30:145–168, 1975
1975
-
[19]
G. F. Lawler and V. Limic. Random walk: a modern introduction , Cambridge University Press, Cambridge, 2010
2010
-
[20]
Peres, S
Y. Peres, S. Popov, and P. Sousi. On recurrence and transien ce of self-interacting random walks. Bulletin of the Brazilian Mathematical Society, New Series , 44(4):841–867, 2013. 35
2013
-
[21]
Salvi and F
M. Salvi and F. Simenhaus. Random walk on a perturbation of the infinitely-fast mixing interchange process. J. Stat. Phys. , 171(4):656– 678, 2018
2018
-
[22]
Ya. G. Sina ˘ ı. The limit behavior of a one-dimensional random walk in a random environment. Teor. Veroyatnost. i Primenen. , 27(2):247–258, 1982
1982
-
[23]
F. Solomon. Random walks in a random environment. Ann. Probability, 3:1–31, 1975. 36
1975
Reviewed August 12, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.