The Operating Speed Difference Model of Atlas Platform Based on Deep Learning

Abstract

Orbit-free density functional theory is expected to become an ideal theoretical method for large-scale quantum mechanical simulation because it can well balance computational accuracy and speed in large-scale system simulation. With the development of kinetic energy density functionals and localized pseudopotentials, orbit-free density functional theory is also expected to be applied to large-scale simulations. However, further development of efficient and stable corresponding software is required for the widespread application of orbit-free density functional theory. This paper develops an orbit-free density functional theory package ATLAS based on the real-space finite difference method. It uses the numerical stability of its direct energy minimization algorithm, through the simulation test of some periodic systems of Mg, Al, and Al3Mg, and the comparison with the calculation results of PROFESS software package based on orbital density functional theory and CASTEP software package based on KS density functional theory. It is shown that this technology can achieve accurate and stable numerical results in large-scale calculation simulation, which can provide a new choice for scarce calculation simulation software based on orbital density functional theory, and is expected to promote the broader application of orbital density functional theory in the large-scale calculation simulation field.