50
90

Asymptotics and Concentration Bounds for Bilinear Forms of Spectral Projectors of Sample Covariance

Abstract

Let X,X1,,XnX,X_1,\dots, X_n be i.i.d. Gaussian random variables with zero mean and covariance operator Σ=E(XX)\Sigma={\mathbb E}(X\otimes X) taking values in a separable Hilbert space H.{\mathbb H}. Let {\bf r}(\Sigma):=\frac{{\rm tr}(\Sigma)}{\|\Sigma\|_{\infty}} be the effective rank of Σ,\Sigma, tr(Σ){\rm tr}(\Sigma) being the trace of Σ\Sigma and Σ\|\Sigma\|_{\infty} being its operator norm. Let \hat \Sigma_n:=n^{-1}\sum_{j=1}^n (X_j\otimes X_j) be the sample (empirical) covariance operator based on (X1,,Xn).(X_1,\dots, X_n). The paper deals with a problem of estimation of spectral projectors of the covariance operator Σ\Sigma by their empirical counterparts, the spectral projectors of Σ^n\hat \Sigma_n (empirical spectral projectors). The focus is on the problems where both the sample size nn and the effective rank r(Σ){\bf r}(\Sigma) are large. This framework includes and generalizes well known high-dimensional spiked covariance models. Given a spectral projector PrP_r corresponding to an eigenvalue μr\mu_r of covariance operator Σ\Sigma and its empirical counterpart P^r,\hat P_r, we derive sharp concentration bounds for bilinear forms of empirical spectral projector P^r\hat P_r in terms of sample size nn and effective dimension r(Σ).{\bf r}(\Sigma). Building upon these concentration bounds, we prove the asymptotic normality of bilinear forms of random operators P^rEP^r\hat P_r -{\mathbb E}\hat P_r under the assumptions that nn\to \infty and r(Σ)=o(n).{\bf r}(\Sigma)=o(n). In a special case of eigenvalues of multiplicity one, these results are rephrased as concentration bounds and asymptotic normality for linear forms of empirical eigenvectors. Other results include bounds on the bias EP^rPr{\mathbb E}\hat P_r-P_r and a method of bias reduction as well as a discussion of possible applications to statistical inference in high-dimensional principal component analysis.

View on arXiv
Comments on this paper