1: \begin{abstract}

2: { We study the Langevin dynamics of a physical system with manifold structure $\mm\subset\bR^p$ based

3: on collected sample points   $\{\mx_i\}_{i=1}^n \subset \mm$ that probe the unknown manifold $\mm$. Through the diffusion map, we first learn the reaction coordinates $\{\my_i\}_{i=1}^n\subset \nn$ corresponding to $\{\mx_i\}_{i=1}^n$, where $\nn$ is a manifold diffeomorphic to $\mm$ and isometrically embedded in $\bR^\ell$ with $\ell \ll p$.  The induced Langevin dynamics on $\nn$ in terms of the reaction coordinates captures the slow time scale dynamics such as conformational changes in biochemical reactions.   To construct an efficient and stable approximation for the Langevin dynamics on $\nn$, we  leverage the corresponding Fokker-Planck equation on the manifold $\nn$ in terms of the reaction coordinates $\my$. We propose an implementable, unconditionally stable, data-driven finite volume  scheme for this Fokker-Planck equation,  which automatically incorporates the manifold structure of $\nn$.  Furthermore, we provide a weighted $L^2$ convergence analysis of the finite volume  scheme to the Fokker-Planck equation on $\nn$. The proposed finite volume  scheme leads to a Markov chain on $\{\my_i\}_{i=1}^n$ with an approximated transition probability and jump rate between the nearest neighbor points. After an unconditionally stable explicit time discretization, the data-driven finite volume scheme gives an approximated Markov process for the Langevin dynamics on $\nn$ and the approximated Markov process enjoys detailed balance, ergodicity, and other good properties.}

4: \end{abstract}

5: