Invariance principle, multifractional Gaussian processes and long-range dependence (Q731682)

Let \(X= \{X_n(H), H\in(1/2,1), n\in\mathbb{N}\}\) be a centered Gaussian field and consider a continuous function \(h: \mathbb{R}\to[a, b]\subset(1/2,1)\). For every \(n\), \(N\in\mathbb{N}\) and \(t> 0\), put \(h^N_n= h(n/N)\) and \[ S^N_h(t)= \sum^{[Nt]}_{n=1} {X_n(h^N_n)\over N^{h^N_n}}. \] Furthermore, consider the following two assumptions: (i) for every \(M> 0\), the map \[ (j,k,H_1,H_2)\mapsto E[X_j(H_1)X_k(H_2)] \] is bounded on \(\{(j,k)\in\mathbb{N}^2,|j- k|\leq M\}\times [a,b]^2\), and (ii) there exists a continuous function \(R:[a, b]^2\to(0,\infty)\) such that \[ \lim_{j-k\to\infty}\;\sup_{(H_1,H_2)\in [a,b]^2}|(j- k)^{2-H_1- H_2} E[X_j(H_1) X_k(H_2)]- R(H_1, H_2)|= 0. \] For every \(t\), \(s\), \(H_1\), \(H_2\) define \[ {\mathcal R}(t,s; H_1, H_2)= R(H_1, H_2)I_{t\geq s}+ R(H_2, H_1)I_{t<s}. \] In this paper, the authors prove the fact that under assumptions (i) and (ii) the finite-dimensional distributions of \(S^N_h\) converge to those of a centered Gaussian process \(\widetilde S_h\) with covariance given for \(t,s\geq 0\) by \[ E[\widetilde S_h(t)\widetilde S_h(s)]= \int^t_0 d\theta \int^s_0 d\sigma{\mathcal R}(\theta, \sigma; h(\theta), h(\sigma))|\theta- \sigma|^{h(\theta)+ h(\sigma)- 2}. \] A representation and properties of the limit process are studied and some examples are shown.