The central limit theorem for stochastic processes. II (Q1117547): Difference between revisions

[For part I see Ann. Probab. 15, 164-177 (1987; Zbl 0615.60008).] The measurability assumptions in two results of the reviewer and \textit{J. Zinn}, ibid. 12, 929-989 (1984; Zbl 0553.60037), are removed. One of them is that the central limit theorem for the empirical process \(n^{1/2}(P_ n-P)(f)\), \(f\in {\mathcal F}\), is equivalent to the central limit theorem for its Gaussian randomization \(n^{- 1/2}\sum^{n}_{i=1}g_ if(X_ i)\), \(f\in {\mathcal F}\), where \(\{g_ i\}\) are i.i.d. N(0,1) (or more generally satisfy \(0<\int^{\infty}_{0}P\{| g_ i| >t\}^{1/2}dt<\infty)\) independent of \(\{X_ i\}\), which are i.i.d. (P). The hypothesis here is that the sequence \((\{X_ i\},\{g_ i\})\) is Pr-perfect. The other result is, essentially, that if \({\mathcal F}\) is P-preGaussian (i.e. the \({\mathcal F}\)-indexed centered Gaussian process \(G_ P\) which has the covariance of \(\delta_{X_ 1}\) is sample-continuous for the \(L_ 2\) metric on \({\mathcal F})\) then the CLT for the empirical process holds in \(\ell^{\infty}({\mathcal F})\) iff for some \(\delta >0\) the random element \[ \sup_{E(G_ P(f)-G_ P(g))^ 2\leq \delta^ 2}| \sum^{n}_{i=1}\epsilon_ if(X_ i)/n^{1/2}| \] tends to zero in \(pr^*\), where \(\epsilon_ i\) are i.i.d. Rademacher-independent of \(\{X_ i\}\). The hypothesis here is that the sequence \(\{X_ i\}\) is Pr-perfect.