Rates of convergence for increments of Brownian motion (Q1106540)

This paper deals with rates of convergence for certain moving averages of Brownian motions. Let \(\mu\) be a mean-zero Gaussian measure on a suitable Banach space B and \(\{W(t)\}^ a \mu\)-Brownian motion in B. Let \({\mathcal K}\) be the unit ball of the Hilbert space generating the \(\mu\)-Wiener measure on \(C_ B[0,1]\). Put \(Lx=\max (1,\log x)\) and the random set \[ J_ n=\{(W(k+(\cdot)a(k))-W(k))/a^{1/2}(k):k=0,1,...,n\}\quad and\quad \epsilon_ n=\epsilon /A(n)^{1/2}, \] where a:[0,\(\infty)\to [1,\infty)\), \(A(x)=L(x/a(x))+LLx\), and \(\epsilon >0\). The first main result is that for a certain class of functions a(x), with probability 1, \(J_ n/(2A(n))^{1/2}\) is eventually contained in the \(\epsilon_ n- neighborhood\) of \({\mathcal K}\). While an ``inner result'' is also obtained: with probability 1, \({\mathcal K}\) is eventually contained in the \(\epsilon_ n\)-neighborhood of \(J_ n/(2A(n))^{1/2}.\) The second main result is concerning the random set \[ E_ n=\{W(k+(\cdot)a(k))-W(k):k=0,1,2,...,n\}. \] Under suitable conditions on a(n) and for \(\epsilon_ n\) suitably chosen in terms of a(n) and the lower bounds for the logarithm of the lower tail distribution of \(\mu_ w\), it is shown that for fixed \(\delta >0\), with probability 1, \(E_ n/b(n)\) is eventually contained in the \(\epsilon\)-neighborhood of \((1+\delta){\mathcal K}\), where \(b(n)=(2a(n)A(n))^{1/n}\). If \(\delta\in (0,1)\), and \(\{\) a(n)\(\}\) satisfies \(n/a(n)\geq (Ln)^ c\) with \(c>1/\delta (1-\delta)\), then with probability 1, \((1-\delta){\mathcal K}\) is eventually contained in the \(\epsilon_ n\)-neighborhood of \(E_ n/b(n).\) An improvement of a result of \textit{E. Bolthausen} [Ann. Probab. 6, 668- 672 (1978; Zbl 0391.60036)] dealing with Strassen's law of the iterated logarithm is also obtained.