Skip to content

Model for energy_per_atom

  • Description: This is a benchmark to evaluate how accurately an AI model can predict the energy per atom using the JARVIS Three-Body Tight Binding dataset computed with Quantum Espresso (this data has been used to fit tight binding parameters for the entire periodic table). The dataset contains 65 different elements and their binary combinations. Here we use mean absolute error (MAE) to compare models with respect to DFT accuracy. External links: https://journals.aps.org/prmaterials/abstract/10.1103/PhysRevMaterials.7.044603, https://jarvis.nist.gov/jarvisqetb/, https://github.com/usnistgov/tb3py


Reference(s): https://github.com/aimat-lab/gcnn_keras, https://www.nature.com/articles/s41524-023-01012-9;https://hackingmaterials.lbl.gov/matminer, https://doi.org/10.1103/PhysRevMaterials.7.044603, https://www.nature.com/articles/s41524-021-00650-1

Model benchmarks

Model name Dataset MAE Team name Dataset size Date submitted Notes
kgcnn_schnetqe_tb16.3509kgcnn82957409-26-2023CSV, JSON, run.sh, Info
matminer_xgboostqe_tb2.2407UofT82957405-22-2023CSV, JSON, run.sh, Info
alignn_modelqe_tb0.7515ALIGNN82957401-14-2023CSV, JSON, run.sh, Info
matminer_rfqe_tb1.5049UofT82957405-22-2023CSV, JSON, run.sh, Info
kgcnn_coGNqe_tb0.0636kgcnn82957405-06-2023CSV, JSON, run.sh, Info