REVIEW 2 major objections 4 minor 19 references
Large deviations at the origin of random walk in random environment
T0 review · 2 major / 4 minor · reviewed 2026-08-12 · deepseek-v4-flash
Pith's one-line read The correct quenched rate for a random walk returning to the origin is realized by a walk that seeks out a rare periodic pocket whose environment emulates the boundary of the convex hull.
desk verdict A good idea with an honest frame, but the central lower-bound construction has a real gap at the tilted-drift hyperplane step. 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 load-bearing object is the dominant event $A^\varepsilon_{N,n,p}$: the walk moves in $a_N=\lfloor N/\log N\rfloor$ steps from 0 to the center of a ball of radius $\delta(\log N)^{1/d}$ where the environment is within $\varepsilon$ of a fixed strip-periodic environment $\omega_{n,p}$, remains in that ball for approximately $N$ steps, and then returns to 0. The construction of $\omega_{n,p}$ uses a strip-periodic tiling perpendicular to a rational vector: each strip is homogeneous with jump probabilities from the support, and the drifts of the tilted distributions are made perpendicular to that vector so that the projected one-dimensional walk is a martingale; Lemma 3.3 then controls the invariant measure and prescribes the fraction of time spent in each strip. Sion's minimax theorem supplies the saddle point that identifies the optimal tilted mixture, and a local limit theorem for random walks in periodic environments controls the probability of the loop inside the pocket.
What would settle it
Take a non-nestling or marginally nestling environment law in $d=2$ whose boundary minimizer $p^*$ has tilted drift directions that are not all perpendicular to any rational vector, for instance three support points whose post-tilting drifts span the plane. Compute $\inf_{\sigma\in K_P} I_\sigma(0)$ and check whether the approximating periodic environment required by Proposition 3.4 can be realized as a $(u;r_1,\dots,r_j)$-strip tiling. If no rational $u$ satisfies the perpendicularity condition while keeping the rate within $\varepsilon$, the paper's construction of the dominant event fails for that law.
Extended reading notes
Core claim
The central claim is Theorem 2.1. For a non-nestling or marginally nestling RWRE, the common value $I(0)=I_a(0)=I_q(0)$ equals $-\log(\sum_e \sqrt{p^*(e)p^*(-e)})$ for some $p^*$ in the boundary of $K_P$, the convex hull of the support of the one-site environment law. For every $\varepsilon>0$ there is a periodic environment $(n_\varepsilon,p_\varepsilon)$, realizable as alternating parallel strips of homogeneous environments from the support, whose rate at the origin is within $\varepsilon$ of $I(0)$. Moreover, for $P$-almost every environment, eventually in $N$ one can find a ball of radius $\delta(\log N)^{1/d}$ within distance $N/\log N$ where the environment is $\varepsilon$-close to that periodic environment, and the quenched probability that the walk goes to that ball, spends $O(N)$ time there, and returns to 0 has exponential rate between $-I(0)-\varepsilon$ and $-I(0)$. Thus the rare return event is staged in a spatially rare region that emulates the optimal boundary point of the convex hull.
Load-bearing premise
The construction assumes that the minimizer of the rate at the origin on the boundary of the convex hull can be approximated by at most $d$ jump distributions whose tilted average velocities all lie perpendicular to one common rational vector; if that hyperplane condition fails, the strip-periodic environment with the prescribed occupation times cannot be built and the lower bound does not follow.
Editorial extensions
If this is right
- The quenched rate at 0 is achieved by a spatially rare, approximately periodic pocket rather than by a single trap; the pocket's environment is built from jump probabilities on the boundary of the convex hull of the support.
- For every $\varepsilon>0$, a periodic environment whose site marginals come from the support of the environment law has rate at 0 within $\varepsilon$ of $I(0)$, so periodic environments are universal approximators for this endpoint rate.
- The probabilities of the travel to and from the pocket are negligible at exponential scale, so the exponential cost of the rare return is entirely determined by the rate of the periodic pocket.
- A homogenization statement for the viscous Hamilton-Jacobi equation with periodic potential is implicit, including the value of the homogenized Hamiltonian at 0, matching the rate-function identity.
Reading between the lines
- If the same mechanism extends to other boundary points, the boundary of $K_P$ and the saddle point of the minimax problem would play the role that the origin's minimizer plays here; that would give a variational description of quenched large deviations at nonzero velocities.
- The common-perpendicular rational vector in Lemma 3.3 is the main geometric constraint. Testing environment laws in $d\ge 3$ where the tilted drifts span more than a hyperplane would expose whether the strip construction is essential or an artifact of the proof.
- Because the annealed and quenched rates coincide at 0, the same periodic-pocket mechanism should also be the annealed picture: the environment law's rare fluctuations create the pocket, and the walk's behavior inside it is nearly deterministic. This suggests annealed large deviation events for RWRE can be studied through variational problems over environment laws concentrated at boundary points.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies nearest-neighbor random walk in an i.i.d. uniformly elliptic random environment on Z^d. Building on Varadhan's formula I_q(0)=I_a(0)=-inf_theta sup_{p in K_P} log sum_e e^{<theta,e>} p(e), it aims to identify a dominant mechanism for the atypical event {X_N=0} in the non-nestling and marginally nestling cases: the walk moves quickly to a ball of radius delta (log N)^{1/d} at distance O(N/log N) whose environment is epsilon-close to a periodic strip environment, spends time of order N there, and returns. The proof proceeds through a time-periodic random-walk approximation (Prop. 3.1), a strip-periodic environment lemma (Lemma 3.3), an approximation of the convex-hull variational problem by periodic environments whose jump probabilities lie in the support of the environment law (Prop. 3.4), and a Borel-Cantelli construction of good balls leading to Theorem 2.1(iii).
Significance. If completed, the paper would provide a valuable mechanistic description of the atypical event behind the quenched large deviation at the origin, complementing Zerner's trap picture for nestling walks and making precise the sense in which the walk seeks regions whose environment emulates the optimal boundary point of the convex hull of the support. Strengths of the manuscript include explicit and self-contained saddle-point computations (Lemma 3.2, Prop. 3.1), periodic environments constructed from the support rather than fitted to the target rate, and a concrete Borel-Cantelli construction of good balls. The main concern is that two load-bearing steps in the proof of Proposition 3.4 are not justified as written; both appear repairable, so the manuscript warrants a major revision.
major comments (2)
- [§3.2, proof of Proposition 3.4, Step 1] The construction of the rational vector u is not justified. The text uses the statement that the convex hull of D_S has dimension at most d-1 to infer a nonzero u perpendicular to all tilted drifts d(sigma*_i), and then asserts that u can be chosen rational. Neither inference follows. Equation (30), namely sum_i t*_i d(sigma*_i)=0, is a linear dependence of the tilted drifts; even when it yields a real u != 0, the dimension of the untitled drift set D_S is irrelevant because tilting by theta* can change the dimension of the drift set. Moreover, boundary attainment of inf_{K_P} I_sigma(0) does not by itself force the approximating set S to have drifts lying in a hyperplane. The rationality of u is asserted with 'by choosing the sigma_i appropriately' but no argument is given, and Lemma 3.3 is stated only for u in Q^d. Since the strip-periodic environment and the occupation-frequency identity (31) are the bridge to the lower bound, this is a load-bearing gap in Proposition 3.4 and hence in Theorem 2.1(ii)-(iii).
- [§3.2, proof of Proposition 3.4, Step 4] The passage from (33) to the point-probability lower bound is not justified. The inequality limsup_{n->infty} (1/n) log P_{0,omega_{n,p}}(X_n=0) >= limsup_{n->infty} (1/n) log P_{0,omega_{n,p}}(|X_n|<= n epsilon) - g(epsilon) is asserted from equicontinuity of the periodic-environment rate functions. An LDP alone controls probabilities of open or closed sets of velocities, not individual point probabilities; for a walk with nonzero asymptotic drift, I(0) need not be the infimum of I over the epsilon-ball, so P(|X_n|<= n epsilon) can be exponentially larger than P(X_n=0). A local limit theorem for the tilted periodic walk, of the kind used later in Lemma 4.1 via [T02], or an equivalent pointwise lower bound, is needed to control P(X_n=0). Without such an estimate the lower bound of Proposition 3.4 is not established as written.
minor comments (4)
- [§4.3, Lemma 4.2] In the proof of Lemma 4.2, the common value at zero-drift points is stated as d/2; with the normalization sum_i (p_i+q_i)=1, the value of sum_i sqrt(p_i q_i) at p_i=q_i=1/(2d) is 1/2, not d/2. The conclusion is unaffected once the factor is corrected.
- [§3.2, Lemma 3.3] The proof of Lemma 3.3 labels two consecutive subsections as 'Step 4', and the lower-bound half of the occupation-time estimate is only described verbally ('suitably bounding the sequence (tau_k) from below') rather than displayed. Since this lemma supplies the frequencies used later, the lower-bound computation should be written out.
- [§3.2, Lemma 3.3, Step 1] The sentence 'if we think of W as having periodic boundary conditions, then W has independent increments' is inaccurate: the projected chain W has transition probabilities that depend on the current strip index, so its increments are not independent. The subsequent martingale and optional-stopping arguments do not require independence, but the wording should be corrected.
- [Theorem 2.1(iii)] In the displayed chain of inequalities in Theorem 2.1(iii), the third limit is written as limsup_{n->infty} (1/N) log P_{0,omega}(X_N=0); the index should be N, not n.
Circularity Check
No significant circularity: the rate identity is derived from Varadhan's external variational formula plus explicit saddle-point and strip-periodic constructions; the flagged Proposition 3.4 issue is a proof gap, not a circular reduction.
full rationale
The derivation chain is self-contained rather than circular. The equality Ia(0)=Iq(0) and the variational formula are imported from Varadhan [V03] as external results, and they are not re-derived from the paper's own conclusions. The paper's positive contributions are explicit: the minimax saddle-point computation for time-periodic environments (Lemma 3.2 and Proposition 3.1), the strip-periodic occupation-frequency lemma (Lemma 3.3), the construction of a space-periodic environment whose rate approximates inf_{K_P} I_sigma(0) (Proposition 3.4), and the Borel-Cantelli/local-CLT argument in Section 4.2 showing that the quenched probability is realized by finding such environments in space. No parameter is fitted to the target rate: the frequencies t* and tilt theta* arise as a saddle point of the displayed log-moment generating functions, and the periodic environment is then constructed from those data, not chosen to match I(0). The self-citations [BMRS23], [BMRS23b], and [BDR11] appear only as contextual references in the introduction and are not load-bearing in any proof. Takenami [T02] and Varadhan [V03] are external, independently checkable ingredients. The only flagged difficulty is Proposition 3.4, Step 1: the passage from the linear relation sum_i t*_i d(sigma*_i)=0 (Eq. (30)) to the existence of a rational u in Q^d orthogonal to every tilted drift d(sigma*_i) is asserted with the phrase 'by choosing the sigma_i appropriately' and no proof. This is a missing justification for the applicability of Lemma 3.3 and therefore a correctness gap, but it is not a circular step: it does not assume the rate being proved or rename a fitted parameter as a prediction. Hence the circularity score is 0.
Assumptions & free parameters
assumptions (5)
- standard math Varadhan's quenched and averaged large deviation principles for uniformly elliptic i.i.d. environments, including the identity Ia(0) = Iq(0) and the variational formula.
- standard math Sion's minimax theorem provides a saddle point for Lambda over Delta_j times R^d.
- standard math Gartner-Ellis theorem gives the large deviation principle for time-periodic random walks.
- domain assumption Takenami's local limit theorem for random walks in periodic environments adapts to the nearest-neighbor setting.
- domain assumption Uniform ellipticity (kappa > 0) for the environment law.
Cite this review
Pith. "Pith review of Large deviations at the origin of random walk in random environment." pith.science (2026). https://pith.science/paper/IURZFLVA
@misc{pith2026241113875,
author = {Pith},
title = {Pith review of: Large deviations at the origin of random walk in random environment},
year = {2026},
howpublished = {\url{https://pith.science/paper/IURZFLVA}},
note = {Machine review of arXiv:2411.13875}
}
read the original abstract
We consider a random walk in an i.i.d. random environment on Zd and study properties of its large deviation rate function at the origin. It was proved by Comets, Gantert and Zeitouni in dimension d = 1 in 1999 and later by Varadhan in dimensions d >= 2 in 2003 that, for uniformly elliptic i.i.d. random environments, the quenched and the averaged large deviation rate functions coincide at the origin. Here we provide a description of an atypical event realizing the correct quenched large deviation rate in the nestling and marginally nestling setting: the random walk seeks regions of space where the environment emulates the element in the convex hull of the support of the law of the environment at a site which minimizes the rate function. Periodic environments play a natural role in this description.
Reference graph
Works this paper leans on
-
[1]
G. Barraquand and I. Corwin. Random walk in Beta-distributed random environment. Probab. Theory Relat. Fields , 167, 1057--1116 (2017)
work page 2017
- [2]
- [3]
- [4]
- [5]
- [6]
-
[7]
L. C. Evans. The perturbed test function method for viscosity solutions of nonlinear PDE. Proc. Roy. Soc. Edinburgh Sect. A , 119, 359--375 (1989)
work page 1989
-
[8]
A. Dembo and O. Zeitouni. Large deviations techniques and applications. Applications of Mathematics (New York), 38 Springer, New York (1998)
work page 1998
Show all 19 references
-
[9]
Drewitz and A.F
A. Drewitz and A.F. Ram\' rez. Selected topics in random walk in random environment. Springer Proc. Math. Stat., 69 Springer, New York, 23--83 (2014)
2014
-
[10]
Greven and F
A. Greven and F. den Hollander. Large deviations for a random walk in random environment. Ann. Probab. 22, 1381--1428 (1994)
1994
-
[11]
Lions, G
P.L. Lions, G. Papanicolaou and S.R.S. Varadhan. Homogenization of Hamilton-Jacobi equations. Preprint, 1987
1987
-
[12]
Rezakhanlou
F. Rezakhanlou. A prelude to the theory of random walks in random environments. Bull. Iranian Math. Soc. 37, 5--20 (2011)
2011
-
[13]
M. Sion. On general minimax theorems. Pacific J. Math. 8 (1): 171--176 (1958)
1958
-
[14]
Sznitman
A.-S. Sznitman. Slowdown estimates and central limit theorem for random walks in random environment. J. Eur. Math. Soc. , 2, 93--143 (2000)
2000
-
[15]
Sznitman
A.-S. Sznitman. On a class of transient random walks in random environment. Ann. Probab. , 29, 2, 724--765 (2001)
2001
-
[16]
Takenami
T. Takenami. Local limit theorem for random walk in periodic environment. Osaka J. Math. , 39, 867--895 (2002)
2002
-
[17]
Varadhan
S.R.S. Varadhan. Large deviations for random walks in random environment. Comm. Pure Applied Math. LVI, 1222--1245 (2003)
2003
-
[18]
A. Yilmaz. Equality of averaged and quenched large deviations for random walks in random environments in dimensions four and higher. Probab. Theory Relat. Fields (149)463-491 (2011)
2011
-
[19]
M. Zerner. Lyapounov exponents and quenched large deviations for multidimensional random walk in random environment. Ann. Probab. 26, 1446--1476 (1998)
1998
Reviewed August 12, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.