Convergence Analysis\nand Predictions for Optimizing\nReciprocal Grids: A First-Principles and Machine Learning Study

When crystalline materials are investigated by performing\nfirst-principles\ndensity functional theory (DFT) calculations, the reciprocal grid\nshould be fine enough to obtain the converged total energy and electronic\nstructure. Herein, we performed a convergence test of the total energy\nfor the density of reciprocal points to determine fine enough reciprocal\ngrids for high-throughput calculations. Our results show that the\nnonlinearity of the band structures affects the convergence of the\ntotal energy, especially for materials with a finite band gap. We\nfurther investigate which physical properties make a finer reciprocal\ngrid necessary based on machine learning (ML) analysis. Our developed\nmodels using DFT-based features and elemental properties-based features\nexhibit <i>R</i><sup>2</sup> of 0.803 and 0.880, respectively.\nOur ML model quantitatively shows the importance of nonlinearity and\nband gaps in predicting errors in total energy calculations. Furthermore,\nour ML model using elemental features can be applied to estimate the\nappropriate reciprocal grid, facilitating high-throughput calculations.