Metastable behaviors of diffusion processes with small parameter (Q1915544)

We consider scaling limits of the diffusion processes \((x^\varepsilon (t), P_x)\) on an orientable \(\sigma\)-compact \(d\)-dimensional Riemannian manifold \(\mathcal M\) generated by \({\mathcal L}^\varepsilon = (\varepsilon^2/2) \Delta - (1/2)\text{grad } U\) with a small parameter \(\varepsilon > 0\). Here \(U\) is assumed to be \(U \in C^\infty\), \(U \geq 0\) and \(\lim_{|x|\to \infty} U(x) = + \infty\), and to have the property that, if we denote the connected components of \(N := \{x \in \mathbb{R}^d; U(x) = 0\}\) by \(N_1, \dots, N_l\), i.e., \(N = \bigcup^l_{i = 1} N_i\), there exists \(V_0 > 0\) such that for all \(1 \leq i,j \leq l\), \(i \neq j\), \(V_0 = \min_\varphi \max_{t \in [0,1]} U(\varphi (t))\), where the minimum is taken over all \(\varphi \in C([0,1], {\mathcal M})\) such that \(\varphi (0) \in N_i\) and \(\varphi (1) \in N_j\). It is shown that the finite-dimensional distributions of the scaled process \(\{x^\varepsilon (\varepsilon^\mu e^{V_0/\varepsilon^2} t)\}_{t \geq 0}\) converge as \(\varepsilon \downarrow 0\) to those of a Markov jump process \(X(t)\) living on \(N\) for an appropriate choice of \(\mu\). To do this, we prepare the asymptotic behaviors of probabilities or expectations concerning exit time and exit position which is sharper than those via the Freidlin-Wentzell theory. There the asymptotics of the principal eigenvalue and eigenfunction for the Dirichlet boundary value problem is studied.