REVIEW 3 major objections 4 minor 18 references
Limit theorems for a minimal random walk model
T0 review · 3 major / 4 minor · reviewed 2026-08-14 · deepseek-v4-flash
Pith's one-line read The superdiffusive minimal random walk, rescaled by $n^p/\Gamma(1+p)$, converges almost surely and in $L^m$ to a non-normal random variable with explicit moments.
desk verdict A genuinely useful limit-theory paper for the minimal random walk, with two fixable gaps: the alpha=-1 SLLN boundary case and an unproved non-normality assertion in Theorem 4. 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 proofs run through the martingale $M_n = (X_n - E[X_n])/a_n$, where $a_n = \prod_{j=1}^{n-1}(1+\alpha/j)$ and $a_n \sim n^{\alpha}/\Gamma(1+\alpha)$. The martingale differences satisfy the uniform bound $|D_j|\le 2/a_j$; Burkholder's inequality, together with convergence of $\sum_j 1/a_j^2$ (which happens exactly when $\alpha>1/2$), yields uniform $L^m$ bounds and hence almost sure and $L^m$ convergence. The exact moments of $X_n$ come from the conditional mean formula $E[\eta_{n+1}|\mathcal{F}_n] = q + \alpha X_n/n$, a recurrence, and a gamma-function summation identity; the moments of $M$ follow by expanding $(X_n/a_n - s)^k$.
What would settle it
Evaluate the moment formulas at $p=3/4$, $s=1$: compute $E(M^2)=2\Gamma(7/4)^2/\Gamma(5/2)-1$ and $E(M^4)$ from the paper's formula, then compare $E(M^4)/E(M^2)^2$ with $3$; a ratio different from $3$ confirms non-normality, while equality would refute the paper's central claim.
Extended reading notes
Core claim
The central claim is Theorem 4: when $q=0$ and $1/2<p<1$, $$\frac{X_n}{n^p/\Gamma(1+p)} - s \to M \quad \text{a.s. and in } L^m \text{ for every } m\ge 1,$$ where $M$ is a non-normal random variable with $E(M)=0$, $E(M^2)= \frac{2s\Gamma(1+p)^2}{\Gamma(1+2p)} - s^2$, and matching explicit third and fourth moments. The paper derives these moments from exact finite-$n$ formulas, and they violate the Gaussian moment relations, so the limit is not normal. The paper also proves that for $\alpha<1/2$ the centered walk, normalized by $\sqrt{n}$, converges to a normal law, that for $\alpha=1/2$ the correct normalizer is $\sqrt{n\log n}$ with a normal limit, that matching laws of the iterated logarithm hold, and that $X_n/n \to q/(1-\alpha)$ almost surely for the parameter range stated in Theorem 1.
Load-bearing premise
The argument needs the normalizer $a_n = \prod_{j=1}^{n-1}(1+\alpha/j)$ to be well defined and asymptotic to $n^{\alpha}/\Gamma(1+\alpha)$, together with the martingale-difference bound $|D_j|\le 2/a_j$; at $\alpha=-1$, which Theorem 1 includes, this sequence is undefined and the bound collapses.
Editorial extensions
If this is right
- For the parameter range stated in Theorem 1, the empirical frequency $X_n/n$ converges almost surely to the deterministic constant $q/(1-\alpha)$, so in the limit the walker's occupation density is known exactly.
- In the diffusive regime $\alpha<1/2$ with $q>0$, $(X_n - qn/(1-\alpha))/\sqrt{n}$ converges to a normal distribution with variance $q(1-p)/((1-\alpha)^2(1-2\alpha))$.
- On the marginally superdiffusive line $\alpha=1/2$, the correct scaling is $\sqrt{n\log n}$, and the limit is normal with variance $4q(1-p)$.
- In the strongly superdiffusive regime $q=0$, $1/2<p<1$, the rescaled walk converges to a non-normal $M$ whose mean is zero and whose variance and higher moments are given by closed formulas in terms of gamma functions.
- The same martingale argument extends to $q>0$ with $\alpha>1/2$, giving a non-normal limit in that superdiffusive region as well, as noted in Remark 2 of the paper.
Reading between the lines
- Because the limit $M$ is non-normal, single-trajectory observables such as time-averaged squared displacements in the superdiffusive phase should display sample-to-sample fluctuations with the cumulants computed here; this is a testable prediction for numerical simulations.
- The explicit first four moments suggest that the law of $M$ might be identified as a polynomial transform of a gamma-distributed random variable; inverting the full moment sequence would settle the exact distribution, which the paper does not do.
- The open case $q=0$, $p\le 1/2$ is not a routine gap: there the normalizer makes $\sum_j 1/a_j^2$ diverge, so the $L^2$-bounded martingale argument used here cannot apply and new ideas are needed.
- Only the structure $E[\eta_{n+1}|\mathcal{F}_n]=q+\alpha X_n/n$ and the bound $|D_j|\le 2/a_j$ are used, so any bounded-increment process with the same conditional mean should obey the same limit theorems.
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper studies the minimal random walk of Kumar, Harbola and Lindenberg, a one-dimensional non-Markovian process with unbounded memory, and proves limit theorems for it. Theorem 1 gives a strong law of large numbers for the whole parameter set alpha in [-1,1). Theorems 2 and 3 give a central limit theorem and a law of the iterated logarithm in the diffusive and marginally superdiffusive regimes. The main result, Theorem 4, treats the case q=0 and 1/2<p<1: the rescaled walk X_n/a_n - s is claimed to converge almost surely and in L^m to a non-normal random variable M whose first four moments are computed explicitly as functions of s and p. The proofs are based on a martingale difference decomposition, Burkholder's inequality, and Gamma-function asymptotics.
Significance. If the non-normality assertion in Theorem 4 is completed, the paper makes a useful contribution: it provides a rigorous non-Gaussian scaling limit with explicit moments for a simple unbounded-memory random walk in its superdiffusive regime, complementing recent mathematical work on elephant random walks. The martingale framework is appropriate, the moment recursions are carried out exactly, and the asymptotic formulas are parameter-free in the sense that p, q, and s are model inputs rather than fitted constants. Theorems 1-3 are standard in structure but fill a gap in the rigorous literature for this specific model. The main weakness is that the central claim that M is non-normal is asserted rather than proved.
major comments (3)
- [Section 4.6.2, proof of Theorem 4] The assertion that the limit M is non-normal is not established. After deriving the formulas for E(M^3) and E(M^4), the proof says to 'verify that they do not correspond to the third and fourth moment of a normally distributed random variable' but no verification is performed. A centered normal would satisfy E(M^3)=0 and E(M^4)=3(E(M^2))^2, and the third moment alone is inconclusive: for p=0.9, the equation E(M^3)=0 has a root s approximately 0.616 in (0,1). The fourth cumulant E(M^4)-3(E(M^2))^2 must be shown to be nonzero on the admissible parameter set, or at least at every point where the third moment vanishes. Since the non-Gaussian character of the limit is the advertised novelty of Theorem 4, this gap is load-bearing and needs to be closed.
- [Sections 4.1 and 4.3, Theorem 1] The proof of Theorem 1 relies on the normalizer a_n = product_{j=1}^{n-1}(1+alpha/j) and the bound |D_j| <= 2/a_j, but a_n is defined only for alpha > -1 and is zero or undefined at alpha = -1. Nevertheless Theorem 1 states the strong law for alpha in [-1,1), so the boundary case alpha = -1 is not covered by the proof. The same issue affects Theorems 2 and 3, whose hypotheses alpha <= 1/2 include alpha = -1 (for example p=0, q=1). The authors should either supply a separate argument for alpha = -1 or restrict the statements to alpha > -1.
- [Section 4.6, Theorem 4] The proof of the almost sure and L^m convergence uses the martingale L^m convergence theorem after showing sup_n E|M_n|^m < infinity for m>1. The case m=1 is not explicitly justified; it follows from the L^2 result by Holder's inequality, but the paper should state this since the theorem claims convergence in L^m for every m >= 1. This is a minor technical omission relative to the two issues above, but it should be fixed for completeness.
minor comments (4)
- [Theorem 4 statement] The displayed convergence 'Xn/npGamma(1+p)^{-1}-s' is ambiguous; it should be written as X_n Gamma(1+p)/n^p - s, or equivalently X_n/(n^p/Gamma(1+p)) - s, to make the normalizing constant clear.
- [Equation (11)] The expression E[eta_j^2 - 2p_j eta_j + p_j^2 | F_{j-1}] is slightly misleading because eta_j is not F_{j-1}-measurable; the equality is only valid after absorbing the difference into the o(1/a_j^2) term using Theorem 1. A short clarifying sentence would improve readability.
- [Lemma 2] The statement of Lemma 2 says 'b ot= a+1' but does not specify whether a and b can be such that the Gamma functions are undefined; in the applications here all arguments are nonnegative and the condition is satisfied, so this is only a presentation issue.
- [References] References [5] and [6] contain the typo 'BC.F. Coletti' instead of 'C.F. Coletti'; this should be corrected.
Circularity Check
No significant circularity: the paper's derivations are self-contained; the unverified non-normality assertion and the alpha = -1 boundary are proof gaps, not circular reductions.
full rationale
The main derivations are self-contained against standard martingale limit theory. The model parameters p, q, and s are inputs of the process definition (Section 2), not fitted to any target quantity. The scaling a_n = product_{j=1}^{n-1}(1 + alpha/j) and its asymptotic a_n ~ n^alpha/Gamma(1+alpha) are derived from the exact product and Stirling's formula (Eq. 5), and the moment formulas in Theorem 4 are obtained by exact recurrences plus Lemma 2, not by postulating the limit distribution. The martingale M_n = (X_n - E[X_n])/a_n is constructed from the model's conditional probabilities, and convergence is obtained from Burkholder's inequality and Lemma 1. No load-bearing premise is justified by a citation to the authors' own work: the prior papers [5, 6] are mentioned only in the introduction as related rigorous results on the elephant random walk and are not used in the proofs. The model definition from [12] is the object of study, not an imported conclusion. The paper's claim that the limit is non-normal is asserted rather than fully proved (the fourth-moment check is omitted), and the stated range alpha in [-1,1) for Theorem 1 includes a boundary case alpha = -1 where a_n is not defined and the bound |D_j| <= 2/a_j fails; however, these are correctness or rigor gaps, not instances of a derivation reducing to its own inputs. Accordingly, no circular step is identified and the circularity score is 0.
Assumptions & free parameters
assumptions (4)
- standard math Classical martingale limit theorems: Stout's strong law (Theorem 5), Hall-Heyde central limit theorem (Theorem 6), Stout's law of the iterated logarithm (Theorem 7), and Burkholder inequality (Theorem 8).
- standard math Gamma function asymptotic and summation identity (Lemma 2).
- domain assumption The model parameters satisfy 0 <= q <= 1, 0 <= p <= 1, and s in [0,1], with the process defined as in Section 2.
- standard math For alpha <= 1/2 and q > 0, the conditional success probability p_j' converges almost surely to q/(1-alpha) as j goes to infinity.
Cite this review
Pith. "Pith review of Limit theorems for a minimal random walk model." pith.science (2026). https://pith.science/paper/M6ANXBWX
@misc{pith2026190809199,
author = {Pith},
title = {Pith review of: Limit theorems for a minimal random walk model},
year = {2026},
howpublished = {\url{https://pith.science/paper/M6ANXBWX}},
note = {Machine review of arXiv:1908.09199}
}
abstract
We study the minimal random walk introduced by Kumar, Harbola and Lindenberg. It is a random process on $\{0, 1, \ldots \}$ with unbounded memory which exhibits subdiffusive, diffusive and superdiffusive regimes. We prove the law of large numbers for the whole parameter set. Then we prove the central limit theorem and the law of the iterated logarithm for the minimal random walk under diffusive and marginally superdiffusive behaviors. More interestingly, we establish a result for the minimal random walk when it possesses the three regimes; we show the convergence of its rescaled version to a non-normal random variable.
Reference graph
Works this paper leans on
-
[12]
Kumar, N., Harbola, U., Lindenberg, K., Memory-induced anomalous dynamics in a minimal random walk model. Phys. Rev. E 90, 022136 (2014)
work page 2014
-
[1]
Interacting elephant random walks
Arita, C.; Ragoucy, E. Interacting elephant random walks. Phys. Rev. E 98, 052118 (2018)
work page 2018
-
[2]
Elephant random walks and their connection to P´ olia-type urns
Baur, E.; Bertoin, J. Elephant random walks and their connection to P´ olia-type urns. Phys. Rev. E 94, 052134 (2016)
work page 2016
-
[3]
A martingale approach for the elephant random walk
Bercu, B. A martingale approach for the elephant random walk. J. Phys. A: Math. Theor. 51(1), 015201 (2018)
work page 2018
-
[4]
Memory-induced diffusive-superdiffusive transition: Ensemble and time-averaged observables Phys
Budini, A.A. Memory-induced diffusive-superdiffusive transition: Ensemble and time-averaged observables Phys. Rev. E 95, 052110 (2017)
work page 2017
-
[5]
Coletti, BC.F.; Gava, R.J.; Sch¨ utz, G. M. Central limit theorem for the elephant random walk. J. Math. Phys 58, 053303 (2017)
work page 2017
-
[6]
Coletti, BC.F.; Gava, R.J.; Sch¨ utz, G. M. A strong invariance principle for the elephant random walk. J. Stat. Mech. Theory Exp., 123207 (2017)
work page 2017
-
[7]
da Silva, M.A.A., Cressoni, J.C., Sch¨ utz, G.M., Viswanathan, G.M., Trimper, S., Non-Gaussian propaga- tor for elephant random walks. Phys. Rev. E 88, 022115 (2013)
work page 2013
Show all 18 references
-
[8]
Martingale Limit Theory and Its Application
Hall, P.; Heyde, C.C. Martingale Limit Theory and Its Application. Academic Press, New York (1980)
1980
-
[9]
Random walks exhibiting anomalous diffusion: elephants, urns and the limits of normality
Kearney, M.J., Martin, R.J. Random walks exhibiting anomalous diffusion: elephants, urns and the limits of normality. J. Stat. Mech., 013209 (2018)
2018
-
[10]
Dynamics of non-Markovian exclusion processes J
Khoromskaia, D.; Harris, R.J; Grosskinsky, S. Dynamics of non-Markovian exclusion processes J. Stat. Mech., P12013 (2014)
2014
-
[11]
Kumar, N., Harbola, U., Lindenberg, K., Memory-induced anomalous dynamics: Emergence of diffusion, subdiffusion, and superdiffusion from a single random walk model. Phys. Rev. E 82, 021101 (2010)
2010
-
[13]
Kumar, N., Harbola, U., Memory induced anomalous dynamics in a random walker with internal states. J. Stat. Mech., 103207 (2018)
2018
-
[14]
Random recursive trees and the elephant random walk
K¨ ursten, R. Random recursive trees and the elephant random walk. Phys. Rev. E 93, 032111 (2016)
2016
-
[15]
M.; Trimper, S
Sch¨ utz, G. M.; Trimper, S. Elephants can always remember: Exact long-range memory effects in a non- Markovian random walk. Phys. Rev. E 70, 045101 (2004)
2004
-
[16]
Stout, W. F. A martingale analogue of Kolmogorov’s law of the Iterated logarithm. Z. Wahrscheinlichkeit- stheorie 15, 279-290 (1970)
1970
-
[17]
Stout, W. F. Almost sure convergence, Probability and Mathematical Statistics, Academic Press, New York-London (1974)
1974
-
[18]
Probability with Martingales
Williams, D. Probability with Martingales. Cambridge University Press, Cambridge (1991). LIMIT THEOREMS FOR A MINIMAL RANDOM W ALK MODEL 13 UFABC - Centro de Matem´atica, Computac ¸˜ao e Cognic ¸˜ao. Avenida dos Estados, 5001, Santo Andr´e - S˜ao Paulo, Brasil e-mail: cristian...
1991
Reviewed August 14, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.