Path-space variational inference for non-equilibrium coarse-grained systems

JOURNAL OF COMPUTATIONAL PHYSICS

卷:314

Article

Harmandaris, Vagelis^3,4  Kalligiannaki, Evangelia³  Katsoulakis, Markos²  Plechac, Petr¹ 

[1] Univ Delaware, Dept Math Sci, Newark, DE 19716 USA

[2] Univ Massachusetts, Dept Math & Stat, Amherst, MA 01003 USA

[3] Univ Crete, Dept Math & Appl Math, Iraklion, Greece

[4] IACM FORTH, Fdn Res & Technol Hellas FORTH, IACM, GR-71110 Iraklion, Greece

关键词: Coarse graining;    Non-equilibrium;    Information metrics;    Machine learning;    Variational inference;    Stochastic optimization;    Time series;    Langevin dynamics;   

DOI  :  10.1016/j.jcp.2016.03.021

来源: Elsevier

【 摘 要 】

In this paper we discuss information-theoretic tools for obtaining optimized coarse-grained molecular models for both equilibrium and non-equilibrium molecular simulations. The latter are ubiquitous in physicochemical and biological applications, where they are typically associated with coupling mechanisms, multi-physics and/or boundary conditions. In general the non-equilibrium steady states are not known explicitly as they do not necessarily have a Gibbs structure. The presented approach can compare microscopic behavior of molecular systems to parametric and non-parametric coarse-grained models using the relative entropy between distributions on the path space and setting up a corresponding path-space variational inference problem. The methods can become entirely data-driven when the microscopic dynamics are replaced with corresponding correlated data in the form of time series. Furthermore, we present connections and generalizations of force matching methods in coarse-graining with path-space information methods. We demonstrate the enhanced transferability of information-based parameterizations to different observables, at a specific thermodynamic point, due to information inequalities. We discuss methodological connections between information-based coarse-graining of molecular systems and variational inference methods primarily developed in the machine learning community. However, we note that the work presented here addresses variational inference for correlated time series due to the focus on dynamics. The applicability of the proposed methods is demonstrated on high-dimensional stochastic processes given by overdamped and driven Langevin dynamics of interacting particles. (C) 2016 Elsevier Inc. All rights reserved.

【 授权许可】

Free