English

Cram\'er-type Moderate Deviation for Quadratic Forms with a Fast Rate

Probability 2021-11-02 v1

Abstract

Let X1,,XnX_1,\dots, X_n be independent and identically distributed random vectors in Rd\mathbb{R}^d. Suppose EX1=0\mathbb{E} X_1=0, Cov(X1)=Id\mathrm{Cov}(X_1)=I_d, where IdI_d is the d×dd\times d identity matrix. Suppose further that there exist positive constants t0t_0 and c0c_0 such that Eet0X1c0<\mathbb{E} e^{t_0|X_1|}\leq c_0<\infty, where |\cdot| denotes the Euclidean norm. Let W=1ni=1nXiW=\frac{1}{\sqrt{n}}\sum_{i=1}^n X_i and let ZZ be a dd-dimensional standard normal random vector. Let QQ be a d×dd\times d symmetric positive definite matrix whose largest eigenvalue is 1. We prove that for 0xεn1/60\leq x\leq \varepsilon n^{1/6}, \begin{equation*} \left| \frac{\mathbb{P}(|Q^{1/2}W|>x)}{\mathbb{P}(|Q^{1/2}Z|>x)}-1 \right|\leq C \left( \frac{1+x^5}{\det{(Q^{1/2})}n}+\frac{x^6}{n}\right) \quad \text{for}\ d\geq 5 \end{equation*} and \begin{equation*} \left| \frac{\mathbb{P}(|Q^{1/2}W|>x)}{\mathbb{P}(|Q^{1/2}Z|>x)}-1 \right|\leq C \left( \frac{1+x^3}{\det{(Q^{1/2})}n^{\frac{d}{d+1}}}+\frac{x^6}{n}\right) \quad \text{for}\ 1\leq d\leq 4, \end{equation*} where ε\varepsilon and CC are positive constants depending only on d,t0d, t_0, and c0c_0. This is a first extension of Cram\'er-type moderate deviation to the multivariate setting with a faster convergence rate than 1/n1/\sqrt{n}. The range of x=o(n1/6)x=o(n^{1/6}) for the relative error to vanish and the dimension requirement d5d\geq 5 for the 1/n1/n rate are both optimal. We prove our result using a new change of measure, a two-term Edgeworth expansion for the changed measure, and cancellation by symmetry for terms of the order 1/n1/\sqrt{n}.

Keywords

Cite

@article{arxiv.2111.00679,
  title  = {Cram\'er-type Moderate Deviation for Quadratic Forms with a Fast Rate},
  author = {Xiao Fang and Song-Hao Liu and Qi-Man Shao},
  journal= {arXiv preprint arXiv:2111.00679},
  year   = {2021}
}

Comments

33 pages

R2 v1 2026-06-24T07:20:14.382Z