Rate of convergence of some self-attracting diffusions (Q2485751)

A self-attracting diffusion \((X_t)_t\) is by definition the solution of a stochastic integral equation (SIE) of the following form: \(X_t = B_t + \int_0^t\int_0^s f(X_s-X_u)\,du\,ds\), where \(B_t\) is one-dimensional Brownian motion and \(f\) is a decreasing function. If \(f\) is measurable and locally bounded, the SIE has a unique weak solution [see \textit{S. Herrmann} and \textit{B. Roynette}, Math. Ann. 325, 81--96 (2003; Zbl 1010.60033)]. The authors consider the asymptotic behaviour of the sample paths, in particular the (rate of) convergence as \(t\to\infty\). The main result of the paper is as follows: Let \(f\) be odd, \(C^1\) decreasing and such that \(|f(x)-f(y)|\geq C_\gamma |x-y|^\gamma\) for small \(|x-y|\) and suitable constants \(\gamma, C_\gamma\). Then, for all \(\mu < (1+\gamma)^{-1}\) it is shown that \(\lim_{t\to\infty} (t/\ln t)^\mu \sup_{s\geq t}|X_s-X_t| = 0 \) almost surely. The proof is based on a comparison result where the convergence (rate) of \(X_t\) is compared with a certain Markov process with non-homogeneous drift. In a first step it is shown that \(X_t\) converges towards the mean process. If this convergence is fast enough, then it is possible to get the above mentioned a.s.\ convergence. The paper concludes with a simple proof of the rate of convergence in the so-called linear case where \(f(x)=-ax\).