Concentration inequalities and moment bounds for sample covariance operators (Q502859)

Let \(X, X_1, X_2,\ldots, X_n,\ldots\) be i.i.d. (independent and identically distributed) centered random variables in a separable Banach space with covariance operator \(\Sigma\). Also, let \(\hat{\Sigma}\) be the sample covariance operator based on \(X_1,X_2,\ldots,X_n\). Moreover, define \(r(\Sigma)=\frac{(\operatorname{E}\| X\|)^2}{\|\Sigma\|}\). It is shown that \(\operatorname{E}\|\hat{\Sigma}-\Sigma\|\|\Sigma \|\Big(\sqrt{}\frac{r(\Sigma)}{n}\vee\frac{r(\Sigma)}{n}\Big)\). Moreover, under an additional assumption that \(r(\Sigma)\lesssim n\), for some \(C>0\) and for all \(t\geq 1\), the concentration inequality \[ |\|\hat{\Sigma}-\Sigma \|-\operatorname{E}\| \hat{\Sigma-\Sigma \| }|\leq C \| \Sigma \| \Bigg(\sqrt{}\frac{t}{n}\vee \frac{t}{n} \Bigg) \] holds with probability at least \(1-e^{-t}\). Similarly, under the assumption that \(r(\Sigma)\gtrsim n\), the resulting concentration inequality \[ |\| \hat{\Sigma}-\Sigma \|-\operatorname{E}\| \hat{\Sigma}-\Sigma \||\leq C \| \Sigma \|\Bigg(\sqrt{}\frac{r(\Sigma)}{n}\sqrt{}\frac{t}{n}\vee \frac{t}{n}\Bigg) \] holds with the same probability \(1-e^{-t}\).