Two-Dimensional CN Material Structure Prediction Based on Machine Learning

Recently, inspired by the widespread application of machine learning in balancing accuracy and cost, this study aims to reduce the cost of atomic simulation by using machine learning algorithms to build generalized linear regression and artificial neural network models to train the energy, force, and stress of two-dimensional (2D) Carbon-nitrogen (CN) material. The predicted results are compared with those based on density functional theory (DFT), and the results show that both the network and linear regression models are effective in predicting energy, and the neural network has better performance in predicting force. Compared with quantum mechanical simulations, the cost of these two methods are reduced by several orders of magnitude. The results demonstrate that ML methods can quickly and accurately predict the energy and force of 2D CN materials, which improves the problem of high cost in traditional trial-and-errorr methods and DFT calculations. We believe that our research provides a feasible method for materials calculation and design, and this machine learning method can be widely used in materials science, providing an efficient, accurate, and economical tool for material design and optimization.