Prediction theory in Hilbert Spaces: Operator-valued Szego Theory
Badr Missaoui, Nicholas H. Bingham
TL;DR
This work extends Szegő theory to operator-valued orthogonal polynomials on the unit circle in infinite-dimensional settings, enriching prediction theory for wide-sense stationary and nonstationary processes. It introduces an operator-valued OPUC framework via moments, Schur complements, and a Schur-structured Verblunsky recursion, then develops operator versions of Christoffel-Darboux kernels and Bernstein-Szegő approximations. The main result is an operator Szegő limit theorem relating Toeplitz-determinant regularizations to the product over Verblunsky coefficients, with a Wiener-algebra regularity condition ensuring the logarithmic determinant integral is well-defined. The findings generalize matrix-valued theory and provide a robust toolkit for analyzing spectral and predictive properties of large-scale or continuous-time operator-valued processes. ${$\displaystyle \lim_{n\to\infty} \mathrm{det}_{2}(T^{R}_{n})/\mathrm{det}_{2}(T^{R}_{n-1}) = \mathrm{det}\prod_{k=0}^{\infty} (I - \alpha_k \alpha_k^{\dagger}) = \exp\left( \frac{1}{2\pi} \int_{-\pi}^{\pi} \log \det \mu'(\theta) \, d\theta \right)$.}$
Abstract
In this paper, we extend some classical results of the Szego theory of orthogonal polynomials on the unit circle to the infinite-dimensional case, and we establish the corresponding Szego limit theorem.
