no. 17 · Wednesday 18 November 2026
A Dobrushin-Major-Taqqu Theorem for Hilbert Space-Valued Random Variables
Abstract
Let $\{X_t\}_{ t \in \mathbb{Z}}$ be a stationary Gaussian process taking values in a separable Hilbert space $\mathbb H_1$, with long-range dependence defined in the spectral domain through a memory operator $A:\mathbb H_1 \to \mathbb H_1$. For square-integrable, nonlinear operators $G:\mathbb H_1\to\mathbb H_2$, taking values in another separable Hilbert space $\mathbb H_2$, we establish a non-central limit theorem for the centered partial-sum process of $\{G[X_t]\}_{t \in \mathbb{Z}}$, whenever $G[X_t]$ inherits the long-range dependence behavior from its Gaussian input process. We identify an operator normalization that accounts for both the underlying memory structure, characterized by the memory operator $A$, and the nonlinear transformation $G$. The limiting Hilbert space-valued process is, in general, non-Gaussian and can be expressed as multiple Wiener integrals involving only finitely many Wiener chaos orders along fixed directions in $\mathbb{H}_2$. The proof introduces a Hilbert space-valued generalization of the reduction principle, identifying the dominant Hermite components and showing that the remainder is negligible under our operator normalization. We employ our results to prove non-central limit theorems for nonlinear transformations of functional FARIMA processes and sojourn functionals of temporally dependent spherical Gaussian fields.
- fields
- Probability