Berry-Esseen bounds in local limit theorems

Sergey Bobkov,Friedrich Götze
2024-07-30
Abstract:Berry-Esseen-type bounds are developed in the multidimensional local limit theorem in terms of the Lyapunov coefficients and maxima of involved densities.
Probability
What problem does this paper attempt to address?
This paper aims to solve the problem of normal approximation in the multi - dimensional local limit theorem, especially in the application of Berry - Esseen - type bounds. Specifically, the authors study the difference between the density function of the normalized form \(Z_n\) of the sum of independent random vectors and the standard normal density function, with the goal of quantifying this difference in a stronger sense, that is, pursuing the approximation of the density function in the sense of uniform convergence. ### Background of the Paper Consider \(n\) independent \(d\)-dimensional random vectors \(X_k\), each \(X_k\) has a mean of zero and a covariance matrix of \(\sigma_k^2I_d\) (where \(\sigma_k>0\)), and satisfies \(\sigma_1^2+\cdots+\sigma_n^2 = n\). Thus, the normalized sum \(Z_n=\frac{X_1+\cdots+X_n}{\sqrt{n}}\) has a zero mean and a unit covariance matrix \(I_d\). ### Main Problems The main problem of the paper is to quantify the difference between the density function \(p_n(x)\) of \(Z_n\) and the standard normal density function \(\phi(x)=\frac{1}{(2\pi)^{d/2}}e^{-|x|^2/2}\). Specifically, the authors focus on the following two main indicators: 1. **Maximum Density Difference**: \[ \Delta_n=\sup_{x\in\mathbb{R}^d}|p_n(x)-\phi(x)| \] 2. **Lyapunov Ratio**: \[ \beta_3=\sup_{|\theta| = 1}\left[\frac{1}{n}\sum_{k = 1}^n\mathbb{E}|\langle\theta,X_k\rangle|^3\right] \] \[ \beta_4=\sup_{|\theta| = 1}\left[\frac{1}{n}\sum_{k = 1}^n\mathbb{E}\langle\theta,X_k\rangle^4\right] \] ### Main Results The main results of the paper are to give upper bounds of \(\Delta_n\), which depend on \(\beta_3\) and \(\beta_4\) as well as some other parameters such as \(M\) and \(\sigma\). Specifically, the authors prove the following two theorems: 1. **Upper Bound in the General Case**: \[ \Delta_n\leq C\sigma^dM^2\frac{\beta_3}{\sqrt{n}} \] 2. **Upper Bound in the Case of Symmetric Distributions**: If \(\mathbb{E}\langle\theta,X_k\rangle^3 = 0\) for all \(\theta\in\mathbb{R}^d\) and \(k\leq n\) (especially when the distribution of \(X_k\) is symmetric), then: \[ \Delta_n\leq C\sigma^{2d}M^3\frac{\beta_4}{n} \] ### Methods To prove these results, the authors use the Fourier transform method and utilize the properties of characteristic functions. Specific steps include: 1. **Separation of Characteristic Functions**: Through the Statuljavičus theorem and its generalization, the authors derive the decay estimates of characteristic functions when far from the origin. 2. **Maximum of Convolution Densities**: Use the properties of convolution densities, especially the upper bounds of their maximum values, to control the behavior of density functions.