Some shift inequalities for Gaussian measures

From MaRDI portal





The authors consider inequalities for centered Gaussian measures \(\mu\) on a Hilbert space and translations \(h\) of the generating Hilbert space. Bounds are derived for a shifted Borel set \(E\) and their measure \(\mu (E+h)\). For instance a sharper bound than Anderson's inequality is found for shifts of symmetric convex sets \(E\). Inequalities of this type have important applications in statistics, where shift families are common statistical models, \(E\) is a critical region and \(\mu (E+h)\) is the power function. Section 4 includes further applications when \(\| h\|\) converges to infinity. The proofs are based on the Cameron-Martin density formula. It is my feeling that the inequalities of the theorem follow from the Neyman-Pearson lemma which can be applied to \(\mu\) and \(\mu*\varepsilon_{-\lambda h}\) if \(1_E\) is a test of the level \(\Phi (\theta)\) and \(\mu (E+ \lambda h)\) is the power function. In addition, this point of view shows that the inequality is sharp when \(E\) denotes the critical region of the Neyman-Pearson test. For the entire collection see [Zbl 0883.00024].











This page was built for publication: Some shift inequalities for Gaussian measures

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q4387146)