The probability of small Gaussian ellipsoids and associated conditional moments (Q2365733): Difference between revisions

Let \(\{x_ i\}^ \infty_{i=1}\) be a sequence of i.i.d. Gaussian random variables and \(z=\sum^ \infty_{i=1}x^ 2_ i/a^ 2_ i\), where \(\{a_ i\}^ \infty_{i=1}\) is a sequence of given numbers satisfying \(\sum^ \infty_{i=1}1/a^ 2_ i<\infty\). Theorem 1. Let \(a_ i\) be an increasing sequence such that \(a_ i\sim i^{\beta/2}\) as \(i\to\infty\) for some \(\beta>1\). Assume that \(\sum^ \infty_{i=1}i^{1/2}b^ 2_ i<\infty\). Then \(E\left(\exp\left(\sum^ \infty_{i=1}b_ ix_ i\right)| z\leq\varepsilon\right) @>>\varepsilon\to 0> 1\). Moreover, when \(\beta=2\), instead of \(\sum^ \infty_{i=1}i^{1/2}b^ 2_ i<\infty\) it is enough to assume that \(\sum^ \infty_{i=1}\ln(i)b^ 2_ i<\infty\). The authors modify the proof of Cramer's theorem to get implicit upper and lower bounds of \(P(z\leq\varepsilon)\) in order to prove Theorem 1. Theorem 1 is applied to compute an \(L^ 2\) type Onsager-Machlup functional for diffusions.