A LIL for independent non-identically distributed random variables in Banach space and its applications (Q931493): Difference between revisions

Let \((B,\| \cdot\| )\) be a separable Banach space and \(X_1,X_2,\ldots\) be independent (in general not identically distributed) \(B\)-valued mean zero random variables. Assume that \((c_n)_{n\in \mathbb{N}}\) is nondecreasing sequence of numbers and, for any \(\rho >1\), there exists a sequence of integers \((n_k)_ {k\in \mathbb{N}}\) such that \((n_{k+1} - n_k)_{k\in \mathbb{N}}\) is eventually nondecreasing in \(k\). Suppose also that \( 1< \liminf_{k\to \infty}c_{n_{k+1}}/c_{n_k} \leq \limsup_{k\to \infty}c_{n_{k+1}}/c_{n_k} \leq \rho. \) Set \[ \alpha_0(\rho):=\sup\left\{\alpha: \sum_{k=1}^{\infty} \frac{1}{n_{k+1} - n_k}\sum_{n=n_k+1}^{n_{k+1}}\exp\left(-\frac{\alpha^2c_n^2}{2H_n}\right) = \infty\right\} \] where \(H_n = \sup_{\| f\| \leq 1} \sum_{i=1}^n {\mathbf E}f^2(X_i)I\{\| X_i\| \leq c_i\}\). If \(S_n /c_n \to 0\) in probability then under specified conditions involving truncated random variables \(X_n\), \(n\in \mathbb{N}\), one has \(\alpha_0(\rho) \leq \limsup_{n\to \infty} \| S_n\| /c_n \leq \rho^4 \alpha_0(\rho)\) a.s. Corollary 1 provides the conditions which guarantee the relation \(\limsup_{n\to \infty} \| S_n\| /c_n = \alpha_0\) a.s. whenever \[ \alpha_0:= \sup\left\{\alpha: \sum_{n=1}^{\infty} \frac{1}{n} \exp\left(-\frac{\alpha^2 c_n^2}{2H_n}\right) = \infty\right\} < \infty. \] The last result is applied to prove the LIL for the empirical covariance of Hilbertian autoregressive processes.