Multi-Scale Molecular Dynamics Simulations

TL;DR

This paper presents a multi-scale molecular dynamics simulation framework focusing on alkanes, introducing a Java-based system with novel methods to derive coarse-grained potentials from detailed atomistic data, enhancing simulation efficiency.

Contribution

It develops a multi-scale MD approach for alkanes, including a Java implementation and two methods for deriving coarse-grained potentials from atomistic simulations, with the minimisation method being most effective.

Findings

The minimisation method outperforms the inverse-Boltzmann approach.

The Java-based MD system incorporates global logical time for simulations.

Coarse-grained potentials are effectively derived from atomistic data.

Abstract

In molecular dynamics (MD), systems are molecules made up of atoms, and the aim is to determine their evolution over time. MD is based on a numerical resolution algorithm, whose role is to apply the forces generated by the various components, according to the equations of Newtonian physics. Molecular Dynamics is currently mainly used in materials science and molecular biology. In this document, we limit ourselves to {\it alkanes} which are non-cyclic carbon-hydrogenated chains. In the basic ``All-atom'' (AA) scale, all the atoms are directly simulated. In the ``United-atom'' (UA) scale, one considers grains that are composed of a carbon atom with the hydrogen atoms attached to it. Grains in the ``Coarse-grained'' (CG) scale are composed of two consecutive UA grains. In the multi-scale approach, one tries to use as much as possible the UA and CG scales which can be more efficiently simulated than the AA scale. In this document, we mainly put the focus on three topics. First, we describe an MD system, implemented in the Java programming language, according to the Synchronous Reactive Programming approach in which there exists a notion of a global logical time. This system is used to simulate molecules and also to build the potentials functions at the UA and CG scales. Second, two methods to derive UA and CG potentials from AA potentials are proposed and analysed. Basically, both methods rely on strong geometrical links with the AA scale. We use these links with AA to determine the forms and values of the UA and CG potentials. In the first method (called ``inverse-Boltzmann''), one considers data produced during several AA scale molecule simulations, and one processes these data using a statistical approach. In the second method (``minimisation method''), one applies a constrained-minimisation technique to AA molecules. The most satisfactory method clearly appears to be the minimisation-based one. The UA potentials we have determined have standard forms: they only differ from AA potentials by parameter values. On the opposite, CG potentials are non-standard functions. We show how to implement them with functions defined ``by cases''. Finally, we consider ``reconstructions'' which are means to dynamically change molecule scales during simulations. In particular, we consider automatic reconstructions based on the proximity of molecules.