Long-memory Hermite functional processes in functional time series spectral asymptotics

Abstract

Functional Non-Central Limit Theorems (NCLTs) are derived, in the functional spectral domain (see RuizMedina, 2022), for square integrable transformations, mapping an H_{1}-valued Gaussian stationary functional sequence into a nonlinear H_{2}-valued sequence, where H_{1} and H_{2} are separable Hilbert spaces. Linear and nonlinear operator-based scalings are considered, leading to different limit LRD Hermite functional processes. Under bounded linear operator scalings, an infinite-dimensional version of Theorem 5.3.1 in Pipiras and Taqqu (2017) is obtained, involving a functional reduction theorem, extending Theorem 5.3.3 in Pipiras and Taqqu (2017). Alternatively, when an operator-valued exponent based nonlinear scaling is applied, we pay attention to the LRD dominant set at each active output direction, in the leading chaos. Both scaling scenarios allow the derivation of NCLTs under similar settings of conditions.