(657h) Extending the Reach of Particle-Based Simulations with Machine-Learned Models

Many of the most challenging problems in chemical engineering involve reactivity in complicated systems. First principles methods like density functional theory (DFT) are attractive for such studies but due to computational expense, are confined to problems of a few 100s atoms and 10s of ps. These limitations have motivated extensive work towards development of accurate, scalable, and reliable reactive force fields but these models generally rely on complicated functional forms requiring time consuming parameterization schemes. As a result, the physiochemical space over which parameters are available is limited.

In this work, we present the Chebyshev Interaction Model for Efficient Simulations (ChIMES), a new reactive force field and semi-automated 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 (and iteratively) generated by force matching to short DFT trajectories. ChIMES models are particularly well suited for studies that frequently rely on costly quantum simulations for elucidation and interpretation of experiments, and to date, have been successfully applied to problems including chemical reactivity under extreme conditions and surface chemistry, which can probe time and lengths that are many orders of magnitude greater that what can be achieved with standard DFT. Moreover, by overcoming this time and length scale gap, ChIMES enables direct comparisons to experiment for the first time, in many cases. Model details will be presented as will selected applications.

This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC52-07NA27344. LLNL-ABS-808063