New methods for prediction of elastic constants based on density functional theory combined with machine learning

Abstract

Elastic constants play critical roles in researching mechanical properties, but they are usually difficult to be measured. While density functional theory (DFT) calculations provide a reliable method to meet this challenge, the results contain inherent errors caused by various approximations. The data-driven approach of machine learning also laid a foundation for predicting material properties. In order to increase the accuracy of theoretical calculations results, in this paper we investigate using machine learning methods to both correct the elastic constants by DFT calculation, and to directly predict elastic constants. The single-hidden layer feedforward neural network trained by back propagation algorithm (SLFN), general regression neural network (GRNN) and support vector machine for regression (SVR) techniques are employed to build regression models to correct the elastic constants by DFT calculation for metal or metallic binary alloys. We also build regression models to predict the elastic constants of metallic binary alloys with cubic crystal system rather than using DFT calculations. It has been demonstrated that the elastic constants corrected by regression models has higher accuracy than those calculated by DFT, and the elastic constants of binary alloys directly predicted by model using the outperformed SLFN technique is prospective.