Developing computationally efficient potential models by genetic programming

The length and time scales of atomistic simulations are limited by the computational cost of the methods used to predict material properties. We have developed a machine learning algorithm based on genetic programming to discover computationally efficient and parsimonious potential models. Genetic programming is an evolutionary algorithm that can search the space of functional forms, facilitating the optimization of the computational efficiency without the need of selecting an expression a priori. Our approach was validated by rediscovering the Lennard Jones potential and the Sutton Chen embedded atom model from training data generated using these models. By using training data generated from density functional theory calculations, we found simple and fast potential models for elemental systems. We present our approach, the forms of the discovered models, and assessments of their transferability, accuracy and speed.