Paperlane

no. 12 · Tuesday 10 November 2026

Deterministic Principal Ideals of Dependent Random Operators

Guangqian Zhao

math.PR · v4 · 30 pp

Abstract

We classify the algebraic principal ideals of alternating quadratic random operators under dimension-free two-sided curvature bounds. For the centered evaluation of a Hilbert two-form, an explicit covariance compensator determines the principal ideal after every fixed compact perturbation. For fixed spectral choices, the compensator is a compact linear function of the full covariance with explicit finite-dimensional approximation bounds; its class modulo the independent Gaussian ideal is intrinsic. For fixed spatial weights of rank at least two and a trace-class kernel, all attainable principal ideals form an exact order interval. Every intermediate ideal is realized by a Gaussian covariance arbitrarily close to the identity while each row retains its standard Gaussian law. For these spatial weights, trace class is also the exact kernel domain for raw evaluation under every law in the curvature class. The proof uses a conditional-residual criterion for linear operators, polarization, and finite-rank approximation. Applications include Wiener area and dependent matrices with doubly monotone variance weights. A logarithmic example separates principal ideals with identical membership in every Schatten class.

comments
30 pages
fields
Probability
pdf ↗ page arXiv:2611.03495 ↗