Weighted Sub-fractional Brownian Motion: Covariance Structure, Path Properties and Euler Approximation

Abstract

Weighted sub-fractional Brownian motion was introduced as a centered Gaussian process with covariance $$Q_{a,b}(s,t)=\frac{1}{1-b}{\int }_0^{s\wedge t}{u}^a[(s-u{)}^b+(t-u{)}^b-(s+t-2u{)}^b]du.$$ This kernel was proved to be positive definite for $a>-1$ and $b\in [0,2]\setminus 1$, and also for $a>-1$, $-1<b<0$, and $a+b+1\ge 0$. In this paper, we replace the power weight $u^a$ by a measurable non-negative function $f:{\mathbb{R}}_{+}\to {\mathbb{R}}_{+}$ satisfying $${\int }_0^{t}f(u)(t-u{)}^{b}du<\infty ,t>0,$$ and consider the kernel $$R_{f,b}(s,t)=\frac{1}{1-b}{\int }_0^{s\wedge t}f(u)[(s-u{)}^b+(t-u{)}^b-(s+t-2u{)}^b]du.$$ We prove that, for $b>-1$, $b\ne 1$, the kernel $R_{f,b}$ is positive definite for every such function $f$ if and only if $b\in [0,1)\cup (1,2]$. For every $b\in (-1,0)\cup (2,\infty )$, we construct a non-negative function $f$ satisfying the integrability condition for which $R_{f,b}$ is not positive definite. The case $b=1$ is obtained as a logarithmic limit. For the associated Gaussian process, we study Hölder regularity, total and quadratic variation, non-stationarity, and long-range dependence. For $b\in (0,1)\cup (1,2]$, we also define differential equations driven by this process and establish pathwise convergence of the corresponding Euler approximation, together with strong $L^p$-rates under the additional regularity assumptions stated below. The logarithmic boundary $b=1$ is included in the covariance theory but is not part of the pathwise-equation and Euler results.

61 pages