REVIEW 4 major objections 5 minor 11 references
Axis-Driven Random Walks on $\mathbb{Z}^2$ (transient cases)
T0 review · 4 major / 5 minor · reviewed 2026-08-12 · deepseek-v4-flash
Pith's one-line read A random walk on the positive quadrant of Z^2, pushed away from the origin only along the axes, is transient and superdiffusive for every repulsion strength alpha < 1/2, with explicit tail asymptotics.
desk verdict Genuinely new superdiffusive regime for a repulsive axis-driven walk, but the main theorem's right tail is wrong as printed and the key coupling lemma is only sketched; the core idea is plausible and worth refereeing after fixes. 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 machinery is the excursion decomposition: with $\eta_i$ the $i$-th time the walk hits an axis and $\rho_i$ the next return to the cone, write $Z_{\rho_i}=\sum_{j=1}^i(Z_{\rho_j}-Z_{\eta_j})+\sum_{j=1}^i(Z_{\eta_j}-Z_{\rho_{j-1}})$ and show the first, axis-run sum dominates. The paper obtains a law of large numbers for $Z_{\rho_i}$ from a one-step recurrence for the mean (Corollary 3.6) and a vanishing covariance bound (Proposition 3.8), then couples $Z$ to a one-sided walk $\tilde Z$ whose only $\alpha$-push is on one axis. That coupling makes the excursion count a renewal process with a $1/2$-stable subordinator limit, so the left and right tail constants come from stable-renewal theory and the arcsine law.
What would settle it
Simulate the walk for a fixed small $\alpha$, say $\alpha=0.1$, and for large $n$ estimate $\mathbb{P}(Z_n \le a\,n^{1/(2(1-\alpha))})$ at several small $a$; if the ratio of that probability to $(8/\sqrt{\pi})(a/c_1)^{(1-\alpha)/2}$ does not approach 1, or if the fraction of trajectories that visit both axes after time $n^{1-\epsilon}$ does not tend to 1, the central claim fails.
Extended reading notes
Core claim
On the paper's own terms, the discovery is a phase statement: once $\alpha>0$, the walk can no longer return to the origin, and its dominant coordinate obeys the two tail limits stated in Theorem 1. For any fixed small $a>0$, $$\lim_{n\to\infty} \mathbb{P}(Z_n \le a\,$n^{{1/(2(1-\alpha))}}$) = \frac{8}{\sqrt{\pi}}\left(\frac{a}{c_1}\right)^{(1-\$\alpha$)/2}, \qquad c_1=(2(1-\$\alpha$))^{1/(1-\$\alpha$)},$$ and $$\lim_{n\to\infty} \mathbb{P}(Z_n \ge $a^{{-1}}$\,$n^{{1/(2(1-\alpha))}}$) = G_{1/2}\big((c_1/a)^{2(1-\$\alpha$)}\big),$$ where $G_{1/2}$ is a $1/2$-stable distribution decaying faster than $e^{-a^{-1/2}}$ as $a\to0$. The mechanism is two-scale: the endpoint of the $i$-th axis excursion satisfies $Z_{\rho_i}/i^{1/(1-\alpha)}\to c_1$ in probability, while the number $N_n$ of axis excursions before time $n$ has the same $\sqrt{n}$ scale and $1/2$-stable fluctuations as for a simple random walk on the half-plane; combining those two scales yields the two tails.
Load-bearing premise
The whole calculation of the tail constants depends on Lemma 2.1's claim that after a short initial phase, the walk with probability tending to one never touches the other axis; the proof of that lemma is sketched rather than fully written out.
Editorial extensions
If this is right
- For every $0<\alpha<1/2$, no matter how small, the walk is transient: it escapes the origin with probability one in the dominant coordinate and the escape scale is $n^{1/(2(1-\alpha))}$, larger than the diffusive $\sqrt{n}$.
- The diffusion in the cone does not retain the particle; it sets the clock. Only about $\sqrt{n}$ axis excursions occur up to time $n$, each axis run reaches a distance of order (count)$^{1/(1-\alpha)}$, and the product of the two gives the superdiffusive normalization.
- The explicit constants give quantitative tails: for small thresholds, $\mathbb{P}(Z_n \le a\,n^{1/(2(1-\alpha))}) \sim (8/\sqrt{\pi})(a/c_1)^{(1-\alpha)/2}$, a prediction that simulations or numerical CDF estimates can check directly.
- The time spent on the axes up to time $n$ is negligible compared with $n$, so the walk inside the cones still looks diffusive; the superdiffusive scale is created entirely by the rare, long axis excursions.
Reading between the lines
- Editorial inference: the theorem is proved only for fixed $\alpha\in(0,1/2)$; letting $\alpha\to0$ and $n\to\infty$ simultaneously is not covered, so the crossover between this superdiffusive regime and ordinary recurrent diffusion at $\alpha=0$ remains an open question rather than a consequence of this paper.
- Editorial inference: the paper's closing remark that $\alpha\ge1$ gives a ballistic walk suggests a second unresolved window $1/2\le\alpha<1$; one plausible guess, not claimed here, is that a different fluctuation exponent or a phase transition separates it from the $\alpha<1/2$ regime.
- Editorial inference: if the coupling lemma is made fully rigorous and extended to the whole lattice $\mathbb{Z}^2$, where the paper's simulations suggest the walk eventually chooses a quadrant, the same two-scale argument would supply full-plane tail asymptotics; the paper explicitly does not claim this extension.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This paper studies an axis-driven nearest-neighbour random walk on the first quadrant of Z^2: in the interior of the quadrant it performs simple random walk, while on the coordinate axes it has a drift away from the origin with probability governed by a parameter α>0. The main theorem states that for 0<α<1/2 the dominant coordinate Z_n is of order n^{1/(2(1-α))}, with explicit left-tail constant c_2(a/c_1)^{(1-α)/2} and a right tail expressed through a 1/2-stable distribution. The proof combines a law of large numbers for the positions at successive returns to the axes, a study of the number of axis-excursions, and a coupling to a one-sided walk with a reflecting boundary.
Significance. If the results were fully established, they would provide a clean and interesting counterpart to the renewal-case paper [AD23]: an arbitrarily small repulsive force on the axes changes the walk from recurrent/diffusive to transient/superdiffusive, with explicit constants. The strength of the paper is its detailed moment analysis in Section 3 and the explicit computation of the constant c_1; the reduction of the fluctuation count to a 1/2-stable subordinator is a promising approach. However, the main theorem as stated contains an algebraic error in the right tail, the left tail appears inconsistent with the stable-subordinator scaling, and the coupling lemma connecting Z to the auxiliary walk is only sketched; these are central and require a revision.
major comments (4)
- [Theorem 1; §2.3, proof of Proposition 2.4] The right-tail formula is algebraically inconsistent with Corollary 2.3. Under the approximation X_n ≈ c_1 (N_n)^{1/(1-α)} used in the proof, the event {X_n ≥ a^{-1} n^{1/(2(1-α))}} corresponds to {N_n/n^{1/2} ≥ (a c_1)^{-(1-α)}}. Since Corollary 2.3 gives P(N_n/n^{1/2} ≥ u^{-1}) → G_{1/2}(u), the limit should be G_{1/2}((a c_1)^{1-α}), not G_{1/2}((c_1/a)^{2(1-α)}). The printed expression tends to 1 as a→0, whereas the correct tail should tend to 0; for α=0.1 and a=0.001 it also violates the union bound together with the left tail. This is a load-bearing error in the statement of the main theorem.
- [§2.1, Lemma 2.1] Lemma 2.1 is the only place where the original walk Z is coupled to the one-sided walk Z̃, and Proposition 2.2, Corollary 2.3, and Proposition 2.4 are all proved for Z̃. The proof of the lemma is a sketch: it asserts that N_n is of order n^{1/2}, that the walk reaches distance n^{(1-δ)/(2(1-α))} before time n, and that from such a distance it is impossible to reach the other axis, but none of these assertions is supplied with a quantitative probability bound that tends to 1. The sentence 'even if it requires adjusting ϵ according to δ' is not a proof. Since the theorem for Z inherits all tail constants from Z̃, this coupling must be proved rigorously or replaced by a direct argument.
- [§2.3, Lemma 2.6] The proof of Lemma 2.6 contains a product estimate that does not match the model: the displayed P_{(y,0)}(X_{ρ^{Z̃}} - x > ϵ n^{1/(2(1-α))}) = ∏_{m=y}^{x+ϵ n^{1/(2(1-α))}} (1 - m^{-α/2}) uses m^{-α/2}, whereas the horizontal-axis transition probabilities of Z̃ are 1 - (2m^α)^{-1}. In addition, the Gaussian-type bounds (7), (8), and (10) are asserted from a diffusion heuristic rather than proved for the discrete random walk. Because Lemma 2.6 controls the event C_{ϵ,n} that the last excursion before n does not affect the normalization, this is another load-bearing gap in Proposition 2.4.
- [§2.2, Corollary 2.3 and §2.3, Theorem 1] The left tail in Corollary 2.3 appears to have an exponent error. From Proposition 2.2, ρ^Z̃_i ≈ i^2 V_{1/2}(1) in law for a 1/2-stable subordinator V_{1/2}; hence P(N^Z̃_n ≤ n^{1/2}u) = P(ρ^Z̃_{⌊n^{1/2}u⌋} ≥ n) → P(u^2 V_{1/2}(1) ≥ 1) ∼ C u by the usual tail behaviour of V_{1/2}(1), not c_2 u^{1/2}. If this is not a typographical issue, the left-tail statements in Corollary 2.3 and Theorem 1, including the exponent (1-α)/2, need to be re-derived.
minor comments (5)
- [§2.1] The symbol Z is used for both the original walk and the auxiliary one-sided walk; this makes Lemma 2.1 and Proposition 2.4 unnecessarily confusing, and a distinct symbol such as Z̃ should be used throughout.
- [Figure 2] The caption reads 'Transition probabilities Transition probabilities of Z'; the duplicated phrase should be removed.
- [§1, around Eq. (4)] There is a typo in the displayed sum: 'P_{j≤i}(Z_{ρ_j} - Z_{ρ_j})' should presumably be 'P_{j≤i}(Z_{ρ_j} - Z_{η_j})' or the intended centered expression; please correct it.
- [§4] The section title 'Authorised the back-return on the axis' is ungrammatical; it should be something like 'Allowing back-return on the axis'.
- [Theorem 1 and Corollary 2.3] The phrase 'for any small u > 0' is informal; since the left-tail expression c_2 u^{1/2} exceeds 1 for u ≥ (1/c_2)^2, the intended range of validity should be stated explicitly.
Circularity Check
No significant circularity: the derivation is self-contained and the constants are derived, not fitted; the cited prior estimates are used as independent local-limit facts.
full rationale
I traced the derivation chain of Theorem 1 through Proposition 2.4, Corollary 2.3, Proposition 2.2, and Proposition 1.1, and then back to the moment estimates in Section 3. The constants c1 and c2 are computed from the transition probabilities of the walk and from the stable tail behavior of first-exit times, not fitted to the target tail probabilities. The local-limit estimates (18)-(20) are imported from the author's earlier work [AD23], and this is a real self-citation, but it is not circularity: these are parameter-free results about the simple random walk in a cone or half-plane, stated for starting points in the cone, and they do not assume the theorem being proved. They are load-bearing, but they are independent external support in the sense of the review rules. Lemma 2.1's coupling is sketched rather than fully rigorous, and it does use Proposition 1.1, but Proposition 1.1 is proved independently from moment estimates, so the coupling is not circular. The apparent algebraic mismatch in the right tail of Theorem 1 / Proposition 2.4 relative to Corollary 2.3 is a correctness concern, not a circularity concern: the conclusion is not identical to its input by construction, it is an erroneous substitution of the argument of G_{1/2}. I found no step where an input is defined in terms of the target, no fitted parameter is renamed as a prediction, and no load-bearing argument reduces to a self-citation chain. Hence the circularity score is 0.
Assumptions & free parameters
assumptions (3)
- domain assumption Local limit estimates for the simple random walk in the half-plane and quarter plane, stated as equations (18), (19), and (20) from [AD23].
- standard math Renewal and arcsine limit theorems for 1/2-stable subordinators, cited to Feller's book, Chapters XI.5, XIII.6, and XIV.3.
- domain assumption The simple random walk on the quarter plane is recurrent and crosses the axes infinitely often almost surely.
Cite this review
Pith. "Pith review of Axis-Driven Random Walks on $\mathbb{Z}^2$ (transient cases)." pith.science (2026). https://pith.science/paper/SSENBTKX
@misc{pith2026241114766,
author = {Pith},
title = {Pith review of: Axis-Driven Random Walks on $\mathbbZ^2$ (transient cases)},
year = {2026},
howpublished = {\url{https://pith.science/paper/SSENBTKX}},
note = {Machine review of arXiv:2411.14766}
}
abstract
Axis-driven random walks were introduced by P. Andreoletti and P. Debs [AD23] to provide a rough description of the behaviour of a particle trapped in a localized force field. In contrast to their work, we examine the scenario where a repulsive force (controlled by a parameter $\alpha$) is applied along the axes, with the hypothesis that the walk remains diffusive within the cones. This force gradually pushes the particle away from the origin whenever it encounters an axis. We prove that even with a minimal force (i.e., a small $\alpha$), the walk exhibits transient, superdiffusive behaviour, and we derive the left and right tails of its distribution.
Figures
Reference graph
Works this paper leans on
-
[1]
P. Andreoletti and P. Debs. Simple random walk on Z ^2 perturbed on the axis (renewal case). Electronic Journal of Probab. , pages 1--25, 2023
work page 2023
-
[2]
P. Andreoletti and P. Debs. The subdiffusive behavior of a recurrent axis-driven random walk. Preprint , 2024+
work page 2024
-
[3]
O. Blondel, M.R. Hilário, R.S. dos Santos, V. Sidoravicius, and A. Teixeira. Random walk on random walks: higher dimensions. Electron. J. Probab. , 24:1--33, 2019
work page 2019
-
[4]
Denis Denisov and Vitali Wachtel. Random walks in cones . The Annals of Probability , 43(3):992 -- 1044, 2015
work page 2015
-
[5]
W. Feller. An Introduction to Probability Theory, Vol. 2 . Wiley, New York, NY, third edition, 1968
work page 1968
-
[6]
G. Fayolle, R. Iasnogorodski, and V. Malyshev. Random Walks in the Quarter-Plane. Algebraic methods, boundary value problems, applications to queueing systems and analytic combinatorics. Springer, second edition, 2017
work page 2017
-
[7]
J. H. B. Kemperman. The oscillating random walk. Stochastic Processes and their Applications , 2(1):1--29, 1974
work page 1974
-
[8]
J. Kloas and W. Woess. Multidimensional random walk with reflections. Stochastic Process. Appl. , 129:336--354, 2019
work page 2019
Show all 11 references
-
[9]
Raschel and P
K. Raschel and P. Tarrago. Marches al\'eatoires dans un c\^one et fonctions discr\`etes harmoniques. Arxiv : https://arxiv.org/pdf/2211.02925 , 2022
2022 arXiv
-
[10]
T. D. Vo and M. Peigné. A functional limit theorem for lattice oscillating random walks. Latin American Journal of Probability and Mathematical Statistics , 20:1433--1457, 2023
2023
-
[11]
Zeitouni
O. Zeitouni. Lectures notes on random walks in random environment. St Flour Summer School , 2001
2001
Reviewed August 12, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.