(360aa) The Chebyshev Interaction Model for Efficient Simulations (ChIMES): Machine-Learned Interatomic Models for Quantum-Accurate Reactive Simulation

First principles methods such as density functional theory (DFT) are extremely powerful in their predictive power and ability to describe reactivity in atomistic simulations, however, high computational expense and poor scalability confines tractable problem space to time and length scales of a few hundred picoseconds and fewer than 10’s of nanometers, respectively. While semi-empirical methods like density functional theory tight-binding (DFTB) can improve accessible timescales while retaining much of the predictive power of DFT, application of DFTB is still largely confined to sub-10 nm length scales. Thus, extensive work has gone into the development of reactive molecular mechanics force fields. For the vast majority of these models, however, the complicated functional forms required to accurately describe bond formation and breaking leads to challenging and time-consuming parameterization schemes, which, in turn, limits the physiochemical space over which parameters are available.

In this work, we present the Chebyshev Interaction Model for Efficient Simulations (ChIMES), a new reactive force field and fitting framework that retains most of the accuracy of DFT while decreasing computational requirements by several orders of magnitude. ChIMES models are comprised of n-body atomic interactions constructed from linear combinations of Chebyshev polynomials, and are entirely linear in fitting coefficients. Thus, model parameters can be rapidly generated by force matching to short DFT trajectories. Model details will be presented as will application to a variety of systems.

This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344. Document release number LLNL-ABS-746816.