Functional limit theorems for random quadratic forms (Q804076): Difference between revisions

Let \((X_ k)\) be a sequence of inependent random variables with E \(X_ k=0\), E \(X^ 2_ k=1\) for all \(n\geq 1\), and let \((a_{ij})\) be a symmetric double array of real numbers. Put \[ Q_ n=\sum^{n}_{i=1}\sum^{n}_{j=1}a_{ij}X_ iX_ j,\quad B_ n=var Q_ n,\quad t_{kn}=B_ k/B_ n,\quad k=0,1,2,...,n;\quad n\geq 1. \] Define \[ f_ n(t) = \begin{cases} B_ n^{-1/2}Q_ k, \text{ for } t=t_{kn},\;k=0,1,...,n,\\ \text{ linearly interpolated elsewhere, }\end{cases} \] \(0\leq t\leq 1\). Under some additional assumptions on \((X_ k)\) and \((a_{ij})\) it has been shown that \(f_ n\) converges weakly to Brownian motion. Define the process Q(t), \(t\geq 0\), by \[ Q(t) = \begin{cases} Q_ n \text{ for } t=B_ n, \\ \text{ linear interpolation elsewhere.} \end{cases} \] A strong invariance principle for Q(t) is established. An approach to the proof bases on Skorokhod embedding scheme for the martingale \(Q_ n\). As an application the compact law of the iterated logarithm like the Strassen one has been stated.