A sufficient condition for the CLT in the space of nuclear operators -- application to covariance of random functions (Q2497812)

Let \(H\) be an infinite-dimensional separable Hilbert space. It is proved the following central limit theorem (CLT) for a sequence \(T_1,T_2,\dots\) of i.i.d. centered trace class random operators from \(H\) to \(H\): if for some basis \((e_p)\) of \(H\) it holds \(\sum_p(E\| T_1(e_p)\|^2)^{1/2}<\infty\), then \(n^{-1/2}(T_1+\cdots+ T_n)\) converges weakly to a centered Gaussian trace class random operator with the same covariance operator as that of \(T_1\). This theorem is used to obtain a CLT for a sequence of empirical covariance operators \((\Gamma_n)\) defined by \(\Gamma_n(\cdots)= n^{-1}\sum^n_{k=1}\langle X_k,\cdot\rangle\,X_k\), where \(X_1,\dots, X_n\) is a sequence of i.i.d. centered random variables with values in \(H\) and \(\langle\cdot,\cdot\rangle\) is the inner product in \(H\). Further results as CLT for \(m\)-dependent trace class random operators as well as CLT for empirical cross-covariance operators are also provided. The study is motivated by some applications to statistics, particularly to inference on random curves.