From law of the iterated logarithm to Zolotarev distance for supercritical branching processes in random environment

Yinna Ye
2024-06-21
Abstract:Consider $(Z_n)_{n\geq0}$ a supercritical branching process in an independent and identically distributed environment. Based on some recent development in martingale limit theory, we established law of the iterated logarithm, strong law of large numbers, invariance principle and optimal convergence rate in the central limit theorem under Zolotarev and Wasserstein distances of order $p\in(0,2]$ for the process $(\log Z_n)_{n\geq0}$.
Probability
What problem does this paper attempt to address?
The problems that this paper attempts to solve are several limit theorems of supercritical branching processes in a random environment, especially the convergence rate problems under the Law of the Iterated Logarithm (LIL), the Strong Law of Large Numbers (LLN), the Invariance Principle, and the Central Limit Theorem (CLT). Specifically, the author focuses on the specific forms and properties of these limit theorems under the Zolotarev distance and the Wasserstein distance. ### Background and Motivation 1. **Research on Branching Processes in a Random Environment**: - Branching processes in a random environment (BPRE) are an important probability model, widely used in fields such as biology, physics, and sociology. - Since the 1970s, the asymptotic behavior of BPRE has been widely studied, but many specific limit theorems still lack in - depth exploration. 2. **Deficiencies in Existing Research**: - Although there have been many studies on BPRE, the limit theorems and convergence rate problems under certain specific distance measures (such as the Zolotarev distance and the Wasserstein distance) have not been fully resolved. ### Main Contributions 1. **New Limit Theorems**: - The author established the convergence rates of the Law of the Iterated Logarithm (LIL), the Strong Law of Large Numbers (LLN), the Invariance Principle, and the Central Limit Theorem (CLT) under the Zolotarev distance and the Wasserstein distance. - These results fill the gaps in the existing literature and provide a more comprehensive theoretical framework. 2. **Methods and Tools**: - The author utilized the latest developments in martingale limit theory, especially the progress in the Invariance Principle and the Law of the Iterated Logarithm, to derive these results. - The properties of the Zolotarev distance and the Wasserstein distance were used, and these distances are widely applied in fields such as machine learning, computer vision, and microbiome research. ### Specific Problems - **Law of the Iterated Logarithm (LIL)**: - The author proved that under certain conditions, \(\limsup_{n \to \infty} \frac{\log Z_n - n\mu}{\sqrt{n \log \log n}} = +\sqrt{2\sigma}\) and \(\liminf_{n \to \infty} \frac{\log Z_n - n\mu}{\sqrt{n \log \log n}} = -\sqrt{2\sigma}\) hold almost surely. - **Strong Law of Large Numbers (LLN)**: - As a direct corollary of LIL, the author proved that \(\lim_{n \to \infty} \frac{\log Z_n}{n} = \mu\) holds almost surely. - **Invariance Principle**: - The author proved that there exists a common probability space such that \(\log Z_n - n\mu - B(n\sigma^2) = o(\sqrt{n \log \log n})\) holds almost surely, where \(B(t)\) is a standard Brownian motion. - **Convergence Rate of the Central Limit Theorem (CLT)**: - The author gave the convergence rates under the Zolotarev distance and the Wasserstein distance, specifically \(\zeta_r\left(\frac{\log Z_n - n\mu}{\sqrt{n\sigma}}, N\right) \leq \frac{C}{n^{\delta/2}}\) and \(W_r\left(\frac{\log Z_n - n\mu}{\sqrt{n\sigma}}, N\right) \leq \frac{C}{n^{\delta/2}}\) for \(r \in [\delta, 1]\)