fig1

Figure 1. The workflow for calculating κ_L using PINK begins with the input of CIF files representing crystal structures. Starting with these CIF files, the framework utilizes CGCNN to predict the bulk and shear modulus, while also extracting crystal information such as volume, number of atoms, and density. These parameters are subsequently used to calculate both longitudinal and transverse sound velocities, which are essential for determining the Grüneisen parameter and the average speed of sound. All of these parameters are incorporated into Equation (2), which includes the Grüneisen parameter (γ), volume (V), temperature (T), and other variables necessary for predicting κ_L. PINK: Physical-informed kappa; CIF: crystallographic information file; CGCNN: crystal graph convolutional neural network.