Spectral decomposition for generalized domains of attraction (Q808513)

Let \(X_ 1,X_ 2,..\). be i.r. vectors in \({\mathbb{R}}^ k\) with a common distribution \(\mu\), and Y be a nondegenerate random vector in \({\mathbb{R}}^ k\) with a distribution \(\nu\). If there exist linear operators \(A_ n\) on \({\mathbb{R}}^ k\) and constants \(b_ n\in {\mathbb{R}}^ k\) such that \[ (*)\;A_ n(X_ 1+...+X_ n)-b_ n \Rightarrow Y, n\to \infty, \] then \(\mu\) is said to belong to the generalized domain of attraction of \(\nu\). All possible nondegenerate limit laws in (*) are called the operator- stable laws. Regular variation techniques [see author, Proc. Am. Math. Soc. 102, No.2, 341-348 (1988; Zbl 0648.26006)] are applied to investigate the asymptotic behaviour of the norming operators \(A_ n\). The spectral decomposition for an operator-stable law, related to the symmetries of the latter, is defined. Applications of these tools include a complete and concise description of moments behaviour, centering constants, convergence criteria, and tails behaviour for generalized domains of attraction.