Comparison of Universal Graph Neural Network Force Fields: Lattice Constant Dependence of Energy#
Universal Graph Neural Network (GNN) force fields achieve higher versatility and accuracy than conventional force fields by utilizing neural networks. Many universities, research institutions, and companies have developed and released them. AdvanceSoft has modified LAMMPS to support various GNN force fields, making them available through Advance/NanoLabo.
In this case study, we will perform energy calculations for SrTiO3, a cubic perovskite crystal, as an example of comparing different GNN force fields. We will demonstrate their applicability to multi-component systems and verify whether they can accurately calculate energies even when the lattice constant deviates from the stable structure.
Creating Models with Different Lattice Constants#
Based on the SrTiO3 structure file (mp-5229) obtained from Materials Project, we created models by isotropically varying the lattice constant from -15% to +15%.
Results#
We calculated the energy of the models using each GNN force field. For comparison, we also performed DFT calculations using Quantum ESPRESSO and plotted the energy difference relative to the original structure.
The results show that although there are some deviations at ±15%, all GNN force fields can reproduce the DFT energy differences with sufficient accuracy. This type of analysis can be applied to predict the mechanical properties and structural transitions of materials.