Development of a Machine-Learning–Based Ionic-Force Correction Model for Quantum Molecular-Dynamic Simulations of Warm Dense Matter

Affordable calculations of accurate ionic forces are a key component of quantum molecular-dynamic (MD) simulations used to investigate warm-dense-matter systems. Within the Born–Oppenheimer approximation, Kohn–Sham (KS) density functional theory (DFT) offers such a balance for the calculation of the electronic contribution to the total ionic forces. Unfortunately, because of the computational cost of KS-DFT scaling as the cube of the system temperature, KS-DFT tends toward a prohibitive cost above 5 eV. Alternatively, orbital-free (OF) DFT is orders of magnitude faster, but its accuracy only converges with that of KS-DFT above ~15 eV. In this work we have developed a machine-learning–based model that uses ionic configurations to predict the difference in the ionic forces between KS and OF-DFT for a given ion. This allows OF ionic forces to be corrected in order to achieve KS accuracy with minimal additional cost. We will discuss the development of a new set of descriptors for the ionic configurations as well as provide training and testing results of the model for hydrogen at 1.0 g/cm³ between 3 and 15 eV. Furthermore, MD simulations are performed with the model and validated against reference KS results.