A bound on the rate of convergence in the central limit theorem for renewal processes under second moment conditions

Gesine Reinert, C. Yang

TL;DR

This paper provides a new, explicit bound on how quickly a renewal process converges to a normal distribution, even when we only know the second moment (variance) of the inter-arrival times. This is important for practical applications where observations are limited to finite time intervals and precise error bounds are needed.

What Problem Does This Paper Address?

The Central Limit Theorem (CLT) is a fundamental result in probability, stating that sums of independent random variables tend towards a normal distribution. Renewal processes, which model events occurring at random intervals, are a key application of the CLT. However, in real-world scenarios, we often only observe these processes over a finite time. This means we need to know how fast the process approaches its normal limit, not just that it eventually does.

Existing results for renewal processes often require stronger assumptions, like the existence of a third moment for the inter-arrival times, to provide bounds on this convergence rate. This paper addresses the gap by providing an explicit, non-uniform bound on the rate of convergence for renewal processes when only the second moment (variance) of the inter-arrival times is known. This is a more practical assumption in many applications.

Methods

The core methodology employed is Stein's method, a powerful technique for bounding the distance between probability distributions. Specifically, the authors adapt existing Stein's method machinery for sums of independent random variables to the context of renewal processes.

Key aspects of their approach include:

  • Non-uniform bounds: Unlike uniform bounds that hold for all values, non-uniform bounds provide tighter estimates for specific ranges of the variable.
  • Explicit constants: The paper focuses on deriving explicit, albeit sometimes large, constants in their bounds. This is crucial for practical applications where the exact magnitude of the error matters.
  • Decomposition: The problem is broken down into bounding different terms arising from the application of Stein's method.
  • Tail bounds: Standard results on the tails of the normal distribution are used to bound specific components of the error.
  • Berry-Esseen theorem: As an intermediate step, they derive explicit bounds for a non-central Berry-Esseen theorem under second moment conditions, which is a valuable result in itself.

Datasets

Not discussed in the paper.

Key Insights and Ideas

  • Second moment is sufficient for explicit bounds: The paper demonstrates that it's possible to derive explicit, non-uniform bounds on the rate of convergence for renewal processes using only the second moment (variance) of inter-arrival times. This relaxes a common requirement in previous work.
  • Stein's method is adaptable: The research showcases the flexibility of Stein's method in tackling complex probabilistic structures like renewal processes, even when moving beyond standard assumptions.
  • Practical implications of explicit bounds: The emphasis on explicit constants, even if large, is important for practitioners who need to quantify error margins in applications involving renewal processes.
  • Intermediate result of independent interest: The derived explicit bounds for a non-central Berry-Essén theorem under second moment conditions are presented as a valuable standalone contribution.

Key Results

The main result is Theorem 3.1, which provides an explicit non-uniform bound on the rate of convergence for the renewal CLT. The bound is of the order $t^{-1/2}$, where $t$ is the time parameter of the renewal process. The paper also provides explicit constants for this bound, though they are noted to be large and not optimized, making the bound less informative for small values of $n$ (number of renewals).

Limitations and Future Work

  • Large constants: The authors acknowledge that the explicit constants in their main bound (Theorem 3.1) are large and not optimized. This limits the informativeness of the bound for small values of $n$. Future work could focus on optimizing these constants.
  • Deterioration near expectation: The bound deteriorates for $t$ close to the expected value $n\mu$. This is a common characteristic of non-uniform bounds and suggests that different approaches might be needed for very precise estimates around the mean.
  • Focus on non-uniform bounds: While providing explicit non-uniform bounds is a strength, future work could explore uniform bounds under similar second-moment conditions.

Summary updated

Abstract

A famous result in renewal theory is the central limit theorem for renewal processes. Since, in applications, usually only observations from a finite time interval are available, a bound on the Kolmogorov distance to the normal distribution is desirable. We provide an explicit non-uniform bound for the renewal central limit theorem based on Stein’s method and track the explicit values of the constants. For this bound the inter-arrival time distribution is required to have only a second moment. As an intermediate result of independent interest we obtain explicit bounds in a non-central Berry–Esseen theorem under second moment conditions.
BibTeX
@article{reinert2020bound,
  author = {Gesine Reinert and C. Yang},
  title = {{A bound on the rate of convergence in the central limit theorem for renewal processes under second moment conditions}},
  journal = {J. Appl. Probab.},
  year = {2020},
  doi = {10.1017/jpr.2019.101},
}
Figure