On decoupling of functions of normal vectors. II (Q382392): Difference between revisions

In an earlier paper [Math. Notes 92, No. 3, 362--368 (2012); translation from Mat. Zametki 92, No. 3, 401--409 (2012; Zbl 1269.60021)], the authors proved the following result: Suppose \(Y_i\), \(i=1,\dots,d\), are standard normal random variables. Let \(c^{-}\) and \(c^{+}\) be the smallest and the largest eigenvalues of the correlation matrix of the random vector \(\bar Y= (Y_1,\dots,Y_d)\). Then \[ c^{-}\sum_{i=1}^d||\phi_i(Y_i)||_2^2\leq ||\sum_{i=1}^d\phi_i(Y_i)||_2^2\leq c^{+}\sum_{i=1}^d||\phi_i(Y_i)||_2^2 \] for all measurable functions \(\phi_i\) satisfying \(\operatorname{E}[\phi_i(Y_i)]=0.\) Moreover, the constants \(c^{-}\) and \(c^{+}\) are the best possible. This result was extended two correlated standard normal random vectors with the best constants in the paper cited earlier. The authors prove a similar result for \(N\) standard normal random vectors in this paper. However, the constants obtained in this case are not the best in the sense discussed above.