Stein's method on Wiener chaos (Q839413): Difference between revisions

The aim of the authors is to provide generalizations of the celebrated Berry-Esséen theorem, on the approximation of the Gaussian law, in the following three directions: -- first, to some approaching variables which are not necessarily normalized sums; -- second, to distances that can differ from the Kolmogorv one (used in the standard Berry-Esséen result); -- third, with the limiting Gaussian law replaced by a (centred) Gamma law. Their method relies mainly on Ch. Stein's central observation, that the distance between a stanard Gaussian variable \(Z\) and an arbitrary random variable \(Y\) is controlled by \(\sup_f \mathbb{E}[|b'\circ Y- Y\times f\circ Y|]\), where \(f\) seems some functional space (depending on the distance). The author apply this to variables \(Y\) belonging to the second chaos of a given Gaussian field, so that, for any functional \(F\) which is smooth (in the Malliavin sense), the above control is easily written: \[ \| 1+\langle DF, DL^{-1}F\rangle\|_2=\| 1-\textstyle{{1\over 2}}\langle DF,DF\rangle\|_2 \] (where \(L\) denotes the Arstein-Uhlenbeck generator, and \(D\) the Malliavin derivative). As a consequence, the following approximation result is deduced: if a sequence \((Z_n)\) of variables of the second Wiener chaos is such that \(\mathbb{E}(Z^2_n)\to 1\) and \(\mathbb{E}(Z^4_n)\to 3\), then their laws converge to \(N(0,1)\); moreover an explicit estimate for the total variation \(d(Z_n,Z)\) is supplied. A similar result is got, for \(Z\) a Gamma variable. The same method is also applied to the fractional Brownian motion.