On the connection between Molchan-Golosov and Mandelbrot-van Ness representations of fractional Brownian motion (Q2426596)

The fractional Brownian motion with Hurst index \(H\in (0,1)\), or \(H\)-fBm, is the continuous centered Gaussian process \((B_t^H)_{t\in{\mathbb R}}\) with \(B_0^H=0\) almost surely and \[ \text{Cov}_{\mathbb P}(B_s^H,B_t^H)= \tfrac12 \left(| s| ^{2H}+| t| ^{2H}-| t-s| ^{2H}\right), \quad s,t\in\mathbb R. \] For \(H=1/2\), fractional Brownian motion is standard Brownian motion. The fractional Brownian motion is interesting from a theoretical and a practical points of view. Important tools when working with the fractional Brownian motion are the integral representations (the generalized Molchan-Golosov integral transform): for a fixed Hurst index \(K\in (0,1)\), on the one hand, there exists a \(K\)-fBm \((B_t^K)_{t\in\mathbb R}\) such that for all \(t\in [0,\infty)\) we have that \[ B_t^H=C(K,H)\int_0^t(t-s)^{H-K} \biggl(F(1-K-H,H-K,1+H-K,{{s-t}\over{s}} \biggr)\,dB_s^K, \quad \text{a.s.,} \] where the constant \(C(K,H)\) depends on only \(K\) and \(H\) [see \textit{C. Jost}, Stochastic Processes Appl. 116, No. 10, 1341--1357 (2006; Zbl 1102.60032)]. On the other hand, there exist a unique, up to modification, \(K\)-fBm \(({\widetilde{B}}_t^K)_{t\in{\mathbb R}}\) such that for all \(t\in{\mathbb R}\), it holds that (the generalized Mandelbrot-Van Ness integral transform): \[ B_t^H=C(K,H)\int_{\mathbb R} ((t-s)^{H-K}1_{(-\infty,t)}(s)-(-s)^{H-K}1_{(-\infty,0)}(s))\,d{\tilde B}_s^K, \quad {\text{a.s.,}} \] see \textit{V. Pipiras} and \textit{M. S. Taqqu} [J. Time Ser. Anal. 23, No. 4, 487--501 (2002; Zbl 1088.60035)]. In the paper a connection between the generalized Molchan-Golosov integral transform and the generalized Mandelbrot-Van Ness integral transform of fractional Brownian motion is proved.