(9e) Enhancing Lammps for Big Data Applications

LAMMPS is a popular open-source software package for large-scale molecular dynamics (MD) simulations. A single massively parallel MD simulation can require tens to thousands of cores and hours of wall clock time. Historically, LAMMPS was not designed to interface with large databases on the fly. However, certain simulation methods, such as reactive MD and Monte Carlo routines, can take advantage of recent advances in cheminformatics and databases prepared for use by machine learning. In this work, we detail major changes to the LAMMPS core source code that prepare it to dynamically interact with chemical databases. First, we introduce type labels to allow atom types and their higher-order interactions (bonds, angles, dihedrals, and impropers) to be represented as alphanumeric strings, e.g., the standard atom type strings for a given force field. Second, we eliminate a long-standing restriction that the exact number of types must be specified when creating the simulation box. Finally, we augment LAMMPS molecule templates to include types. In this way, molecule templates become a stand-alone description of a molecular topology that can be added to running simulations on the fly. The power of this feature is demonstrated via the REACTER protocol, a method for modeling chemical reactions in classical MD simulations. We introduce a comprehensive library of reactions compatible with classical force fields and relevant to organic and polymer chemistry, which transforms REACTER into a plug-and-play reactive MD tool. A library for REACTER allows a particular reaction to be specified by name or queried via a substructure search, and the reaction database is referenced to activate that mechanism in the simulation. In particular, scanning through complex configurations of various polymer chemistries is made possible. This automated generation of atomic configurations for polymeric and composite materials is a particularly challenging bottleneck in the polymer informatics field.