A general LIL for \(B\)-valued geometrically weighted series under dependent assumption (Q663081): Difference between revisions

Let \(\{X, X_n, n\geq 0\}\) be a sequence of identically distributed \(\psi\)-mixing dependent random variables with values in a type 2 Banach space \(B\) with topological dual \(B^*\). The series \(\xi(\beta)=\sum^\infty\limits_{n=0}\beta^nX_n\) is considered if \(\sum^\infty\limits_{n=1}\psi(n)<\infty, \quad \sum^{\infty}\limits_{n=1}\operatorname P\{\| X\| \geq c_n\}<\infty, \quad c_n>0\), \(\frac{1}{c_n}\sum^n_{k=0} X_k\rightarrow 0\) in probability, where \(\frac{c_n}{\sqrt{n}}\uparrow\infty\) and, for all \(\varepsilon > 0\), exists \(m_{\varepsilon}>0\) such that \(\frac{c_n}{c_m}\leq(1+\varepsilon)\frac{n}{m}\), \(n\geq m\geq m_{\varepsilon}\). The main statement of the paper is Theorem 1.1. It states that \[ \lim\sup\limits_{\beta \nearrow 1} c^{-1}_{(1-\beta^2)^{-1}}\Big\| \sum^{\infty}\limits_{n=0}\beta^{n}X_n\Big\| =\alpha_1 \] with probability one, where \[ \alpha_1:=\sup\Big\{\alpha\geq 0:\sum_{n=1}^\infty\frac{1}{n}\exp\Big(-\frac{\alpha^2 c^2_n}{2n\sup\limits_{f\in B^*_1}\operatorname E f^2(X I(\| X\| \leq c_n))}\Big)=\infty\Big\}<\infty \] Taking some special norming sequence \(c_n\) (for example \(c_n=\sqrt{nh(n)}\) where \(h(\cdot)\) is a slowly varying function) one can get the law of iterated logarithm when \(\operatorname Ef^2(x)\) is infinite.