Molecular Simulations: Methods and Software Development
Overview
We develop open-source simulation software and new computational methods that make molecular simulation more accurate, reproducible, and accessible. However, molecular simulations are only as good as the methods and software behind them. We develop simulation methods and maintain open-source software so that the broader community can compute thermodynamic and transport properties reliably and reproducibly.
Approach
Our methodological work spans Monte Carlo, molecular dynamics, and machine-learning techniques, aimed at computing phase equilibria, free energies, and a broad range of thermophysical and transport properties. Two commitments run through this work. The first is reliability: many properties are easy to estimate badly, so we develop methods with built-in statistical assessment and document best practices the community can adopt. The second is openness: we release our software and methods so that others can reproduce our results and build on them. The tools and methods described below reflect both, and most exist because answering a specific scientific question required a capability that was missing. We develop new methods, contribute to community frameworks for reproducible simulation, and increasingly use machine learning both to improve the classical models simulation depends on and to predict properties directly.
Software
Cassandra
A comprehensive open-source atomistic Monte Carlo package that computes thermodynamic properties of fluids including vapor-liquid equilibrium using advanced sampling methods.
Explore Cassandra Monte Carlo Software
Read more about Cassandra about the Journal of Computational Chemistry
MoSDeF-Cassandra
A Python interface to Cassandra that uses MoSDeF.
PyLAT
A collection of tools to analyze LAMMPS output trajectories and compute properties using validated methods.
MoSDeF
The Molecular Simulation Design Framework: tools for setting up and running molecular simulations reliably.
Methods
Pseudo-Supercritical Path Sampling (PSCP)
A free energy approach for computing the melting points of solids and free energy differences between crystal polymorphs. The method has been implemented in LAMMPS.
View an article about PSCP in the Journal of Chemical Physics
Reliable Viscosity Calculation (Time Decomposition)
This is a reliable approach for computing the viscosity of a liquid using independent equilibrium molecular dynamics simulations coupled with automatic statistical assessment of the estimated value of the viscosity. J. Chem. Theory Comput. 2015, 11, 8, 3537–3546] Link: 10.1021/acs.jctc.5b00351
Best Practices for Computing Transport Properties
A collection of best practices for computing diffusivities and viscosities from molecular dynamics.
View an article from the Living Journal of Computational Molecular Science
OpenSPGen
An open source tool for computing sigma profiles, useful for property prediction and machine learning.
Force Field Parameterization Using Machine Learning
A collection of tools and methods that use Gaussian process surrogate models to rapidly optimize classical force fields.
View Force Field Parameterization Repositories.
Group Contribution and Gaussian Process Models for Property Prediction
A simple GP correction to a classic group contribution property prediction method that is fast and accurate for a range of properties.
Funding
This work has been supported by the National Science Foundation, Air Force Office of Scientific Research, and the Department of Energy.