On recurrent properties of Fisher-Wright's diffusion on \((0,1)\) with mutation (Q2239789)

This short paper proves a refined exponential recurrent bound for a one-dimensional Fisher-Wright diffusion process living on the interval \((0,1)\) subject to mutations. This bound implies an exponential rate of convergence towards the invariant measure. More precisely, consider the following one-dimensional stochastic differential equation \[ dX_t =\left[\#1\right]{a(1-X_t) -bX_t}dt + \varepsilon \sqrt{X_t\left(\#1\right){1-X_t}}dW_t \] with parameters \(a,b,\varepsilon >0\). The starting point \(X_0=x\in (0,1)\) is deterministic and \((W_t)\) is a one dimensional standard Brownian motion (under \({\mathbb P}_x\)). Such equations were introduced for the study of population genetics independently by Wright and Fisher and remain a topical area of investigations until now. In the context of the paper, the model is subject to mutations and the parameters \(a\) and \(b\) stand for the mutation rates (for an account on such models see [\textit{L. Chen} and \textit{D. W. Stroock}, SIAM J. Math. Anal. 42, No. 2, 539--567 (2010; Zbl 1221.35013); \textit{C. L. Epstein} and \textit{R. Mazzeo}, SIAM J. Math. Anal. 42, No. 2, 568--608 (2010; Zbl 1221.35063)] in the authors' reference list) and \(\varepsilon\) as a selection parameter. Well-known classical results on stochastic differential equations ensure that there is a pathwise unique strong solution for this equation and that this solution is a strong Markov process. Since \((0,1)\) is topologically equivalent to \(\mathbb R\), one may apply Feller's test to show that this solution remains in the open interval \((0,1)\) for all times whenever the condition \(\min(a,b)>\varepsilon^2/2\) is satisfied (Feller's condition). Although one may wish to apply general results ensuring the existence of an invariant measure with an exponential rate of convergence for general one-dimensional models (see for, e.g., [\textit{L. H. Duc} et al., Stochastic Processes Appl. 128, No. 10, 3253--3272 (2018; Zbl 1434.60138)]), the strategy is quite different here and the main result of the paper gives a deeper insight on the recurrent properties of the Fisher-Wright diffusion process with a more detailed explicit result for the recurrence properties of the solution. More precisely, for any \(\alpha\in (0,1)\) let \(\tau_\alpha = \inf\left(\#1\right){t\geq 0~:~X_t\in [\alpha, 1-\alpha]}\) stand for the time where the solution enters \([\alpha, 1-\alpha]\). Then, assuming that Feller's condition is satisfied, the main result of the paper asserts that for any constant \(c>0\), there exists a point \(\alpha\in (0,1/2)\) and \(m>0\) such that \[ \mathbb E_x \mathrm{e}^{c\tau_\alpha} \leq C(m)\,c\,\alpha^{m+1}\left(\#1\right){(1-x)^{-m} + x^{-m}} + 1 \] with \(C(m) = \frac{2}{\min(a,b)m - \varepsilon^2m(m+1)/2}\). In a nutshell, the methodology of proof is to draw clever consequences of the application of Itô's formula to the function \(v(t,x) = x^{-m}{\exp}(ct)\). The stated inequality ensures that there are exponential moments for the returns from very small neighborhood of \(0\) or \(1\) to a compact set in \((0,1)\) and moreover that these exponential moments are quantified by an explicit bound derived from the data. As a by product of their main result, the authors recover the exponential rate of convergence of the distribution of \(X_t\) towards the invariant measure w.r.t the total variation metric (the computations are not written in the text but are announced for a forthcoming research study concerning more general diffusions). Note also that the paper contains an independent and short proof that the solution remains \({\mathbb P}_x\)-a.s. in \((0,1)\) whenever Feller's condition is satisfied which is inspired by \textit{I. I. Gikhhman} [``A short remark on Feller's square root condition'', Preprint, \url{https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1756450}]. \textbf{NB}: all number references are given by the authors' reference list.