A family of log-correlated Gaussian processes

TL;DR

This paper introduces a new family of log-correlated Gaussian processes indexed by metric spaces, derived as limits of bi-fractional Brownian motions, with stochastic integral representations and connections to aggregated models.

Contribution

It defines a novel class of log-correlated Gaussian processes for metric spaces, including their stochastic integral representations and scaling limit properties.

Findings

Processes arise as limits of bi-fractional Brownian motions

Stochastic integral representations are provided for measure definite kernels

Processes are scaling limits of certain aggregated models

Abstract

A family of log-correlated Gaussian processes indexed by metric spaces is introduced, when the metric is conditionally negative definite. These processes arise as the limit of bi-fractional Brownian motions indexed by $(H,K)$ scaled by $K^{-1/2}$ as $K\downarrow 0$ with $H\in(0,1/2]$ fixed. When the metric is in addition a measure definite kernel, stochastic-integral representations of the generalized processes when evaluated at a test function are provided. The introduced processes are also shown to be the scaling limits of certain aggregated models.