M achine Learning-Driven Elasticity Prediction in Advanced Inorganic Materials via Convolutional Neural Networks

This dataset contains the elastic property data for 80,664 inorganic crystal structures, including predicted shear modulus and bulk modulus values. The process of generating the dataset is as follows: First, the shear modulus and bulk modulus data for 10,987 materials were collected from the Materials Project database using the Matbench v0.1 dataset, and two CGCNN (Crystal Graph Convolutional Neural Network) models were trained based on this data. These models successfully mapped the corresponding bulk and shear moduli from crystallographic information files (CIF) by extracting crystal structure information. To improve data quality, we applied several filtering criteria, removing materials with band gaps greater than 3.0 eV or less than 0.1 eV, and eliminating compounds containing radioactive elements. The models were then trained on the filtered dataset, and pre-trained CGCNN models were used to predict the elastic moduli for additional crystal structures. The final predicted dataset consists of two parts: one part from the Materials Project database, which contains 54,359 crystal structures (referred to as the Materials Project Elastic Dataset, MPED); the other part consists of 26,305 crystal structures discovered by Merchant et al. (referred to as the Nature Elastic Dataset, NED). Together, these two datasets form the comprehensive dataset of elastic properties for 80,664 inorganic crystal structures, providing essential reference data for material design and performance prediction.The data file contains the following information for each material: the Crystallographic Information File (CIF), Formula, Number of Atoms, Density (g cm⁻³), Volume (Å), Total Atomic Mass (amu), Shear Modulus (GPa), Bulk Modulus (GPa), Sound Velocity of Longitudinal Wave (m s⁻¹), Sound Velocity of Transverse Wave (m s⁻¹), Speed of Sound (m s⁻¹), Poisson's Ratio, and Debye Temperature (K).