[Paper Review] Approximate Local Limit Theorems with Effective Rate and Application to Random Walks in Random Scenery

Rita Giuliano, Michel Weber|arXiv (Cornell University)|Dec 12, 2014

Stochastic processes and statistical mechanics26 references20 citations

TL;DR

This paper develops approximate local limit theorems with explicit, effective error bounds for sums of independent lattice-valued random variables using the Bernoulli part extraction method. It establishes a new local limit theorem with an effective remainder for random walks in random scenery, recovering classical results by Gnedenko and Gamkrelidze and providing a practical, quantitatively precise framework for applications in probability and number theory.

ABSTRACT

We show that the Bernoulli part extraction method can be used to obtain approximate forms of the local limit theorem for sums of independent lattice valued random variables, with effective error term, that is with explicit parameters and universal constants. We also show that our estimates allow to recover Gnedenko and Gamkrelidze local limit theorems. We further establish by this method a local limit theorem with effective remainder for random walks in random scenery.

Motivation & Objective

• To develop approximate local limit theorems with effective, explicit error terms for sums of independent lattice-valued random variables.
• To extend the Bernoulli part extraction method to provide quantitative estimates with universal constants and explicit parameters.
• To recover classical local limit theorems of Gnedenko and Gamkrelidze using this new method.
• To establish a local limit theorem with effective remainder for random walks in random scenery.
• To provide a practical, quantitatively precise alternative to characteristic function methods in local limit theory.

Proposed method

• Applies the Bernoulli part extraction method, originally developed by McDonald and rooted in Kolmogorov’s earlier ideas, to decompose lattice-valued random variables into a Bernoulli component and a remainder.
• Uses the probabilistic device of the Bernoulli part to transfer known results for Bernoulli sums to general lattice-valued sums.
• Derives an approximate local limit theorem of the form $\sqrt{\Sigma_n} \mathbb{P}(S_n = N) \approx \frac{D}{\sqrt{2\pi}} e^{-(N - M_n)^2 / (2\Sigma_n)} $ with an explicit error term.
• Establishes the effective remainder by controlling the error via moment conditions: $\int_{|x| \geq u} x^2 F(dx) = \mathcal{O}(u^{-2\alpha})$ for $0 < \alpha < 1/2$.
• Applies the method to weighted sums and random walks in random scenery by extending the representation to weighted Bernoulli walks.
• Uses conditional independence and the structure of the Bernoulli part to derive the distributional equivalence $\{S_m\} \stackrel{\mathcal{D}}{=} \{W_m + D M_m\}$, where $M_m$ is a weighted Bernoulli walk.

Experimental results

Research questions

• RQ1 Can the Bernoulli part extraction method yield approximate local limit theorems with explicit, effective error bounds for sums of independent lattice-valued random variables?
• RQ2 To what extent can the method recover the classical local limit theorems of Gnedenko and Gamkrelidze with explicit constants and parameters?
• RQ3 Can the method be extended to handle weighted sums and random walks in random scenery, where a common lattice span may not exist?
• RQ4 What moment conditions ensure an effective remainder term of order $\mathcal{O}(n^{-\alpha})$ with $0 < \alpha < 1/2$?
• RQ5 How does the method compare to characteristic function techniques in terms of quantitative precision and applicability?

Key findings

• The paper establishes an approximate local limit theorem with an effective remainder: $\sqrt{\Sigma_n} \mathbb{P}(S_n = N) = \frac{D}{\sqrt{2\pi}} e^{-(N - M_n)^2 / (2\Sigma_n)} + \mathcal{O}(n^{-\alpha})$ for $0 < \alpha < 1/2$, under the condition that the span $D$ is maximal and $\int_{|x| \geq u} x^2 F(dx) = \mathcal{O}(u^{-2\alpha})$.
• The method recovers the classical local limit theorems of Gnedenko and Gamkrelidze, including the case $\alpha = 1/2$ under the third moment condition $\mathbb{E}|X|^3 < \infty$.
• For random walks in random scenery, the paper derives a local limit theorem with an effective remainder using the same framework, extending the Bernoulli part method to this non-i.i.d. setting.
• The error term is explicit and universal, with constants independent of the underlying distribution beyond moment and span conditions.
• The method avoids reliance on characteristic functions and provides a more direct, probabilistically grounded alternative with quantifiable error control.
• The approach is shown to be robust in challenging models such as the probabilistic partition function model, where characteristic function methods fail due to large exponential moments.