1D Kinetic Energy Density Functionals learned with Symbolic Regression

Abstract

Orbital-free density functional theory promises to deliver linear-scaling electronic structure calculations. This requires the knowledge of the non-interacting kinetic-energy density functional (KEDF), which should be accurate and must admit accurate functional derivatives, so that a minimization procedure can be designed. In this work, symbolic regression is explored as an alternative means to machine-learn the KEDF, which results into analytical expressions, whose functional derivatives are easy to compute. The so-determined semi-local functional forms are investigated as a function of the electron number, and we are able to track the transition from the von Weizs\"acker functional, exact for the one-electron case, to the Thomas-Fermi functional, exact in the homogeneous electron gas limit. A number of separate searches are performed, ranging from totally unconstrained to constrained in the form of an enhancement factor. This work highlights the complexity in constructing semi-local approximations of the KEDF and the potential of symbolic regression to advance the search.