Sparse selection of bases in neural-network potential for crystalline and liquid Si

TL;DR

This paper introduces a systematic method using forward stepwise regression to develop a neural-network interatomic potential for silicon, reducing the number of basis functions while maintaining high accuracy for crystalline and liquid states.

Contribution

The paper presents a novel application of stepwise regression to efficiently select bases in neural-network potentials, improving interpretability and reducing complexity.

Findings

Accurate potential with fewer bases achieved

Good agreement with ab-initio results for crystalline properties

Reliable dynamic property predictions for liquid Si

Abstract

The neural-network interatomic potential for crystalline and liquid Si has been developed using the forward stepwise regression technique to reduce the number of bases with keeping the accuracy of the potential. This approach of making the neural-network potential enables us to construct the accurate interatomic potentials with less and important bases selected systematically and less heuristically. The evaluation of bulk crystalline properties, and dynamic properties of liquid Si show good agreements between the neural-network potential and ab-initio results.