Formation energy prediction on Materials Project (test)
0.0188MAE (eV/atom)PotNet
Evaluation results
| Method | Links | |||||
|---|---|---|---|---|---|---|
| PotNet2023.06 | 0.0188 | — | — | — | — | |
| Matformer2023.06 | 0.021 | — | — | — | — | |
| Matformer2022.09 | 0.021 | — | — | — | — | |
| ALIGNNSource=Retrained, Number of layers=4 gen layers and 4 alignn layers, Learning rate=1e-3, Batch size=642022.09 | 0.022 | — | — | — | — | |
| ALIGNN2023.06 | 0.0221 | — | — | — | — | |
| MEGNetElements=89, Training structures=600002018.12 | 0.028 | — | — | — | — | |
| MEGNETSource=Original paper results2022.09 | 0.028 | — | — | — | — | |
| MEGNET2023.06 | 0.03 | — | — | — | — | |
| MEGNETSource=Retrained, Number of layers=3, Learning rate=1e-3, Batch size=128, Epochs=1000, Radius=4.02022.09 | 0.03 | — | — | — | — | |
| CGCNN2023.06 | 0.031 | — | — | — | — | |
| CGCNNSource=Retrained, Hidden dimensions=128, Batch size=256, Number of layers=3, Radius cutoff=8.0, Epochs=10002022.09 | 0.031 | — | — | — | — | |
| SchNet2023.06 | 0.033 | — | — | — | — | |
| GATGNN2023.06 | 0.033 | — | — | — | — | |
| SchNetSource=Retrained, Number of layers=6, Feature dimension=64, Learning rate=5e-4, Batch size=64, Epochs=500, Radius=12-th smallest distance2022.09 | 0.033 | — | — | — | — | |
| GATGNNSource=Retrained, Hidden dimensions=128, Batch size=256, Learning rate=5e-3, Epochs=500, Radius cutoff=4.02022.09 | 0.033 | — | — | — | — | |
| SchNetElements=89, Training structures=600002018.12 | 0.035 | — | — | — | — | |
| SchNetSource=Original paper results2022.09 | 0.035 | — | — | — | — | |
| CGCNNElements=87, Training structures=280462018.12 | 0.039 | — | — | — | — | |
| CGCNNSource=Original paper results2022.09 | 0.039 | — | — | — | — | |
| GATGNNSource=Original paper results2022.09 | 0.039 | — | — | — | — | |
| ALIGNN2022.09 | — | — | — | 49.94 | 71.1 | |
| CGCNN# of train data=28046, Unit=eV/atom2017.10 | — | 0.039 | — | — | — | |
| Ewald sum matrixTraining size (N)=3,0002017.12 | — | 0.49 | — | — | — | |
| ext. Coulomb matrixTraining size (N)=3,0002017.12 | — | 0.64 | — | — | — | |
| Matformer2022.09 | — | — | — | 55.86 | 75.02 | |
| SchNetTraining size (N)=3,000, Interaction blocks (T)=62017.12 | — | 0.127 | — | — | — | |
| SchNetTraining size (N)=60,000, Interaction blocks (T)=62017.12 | — | 0.035 | — | — | — | |
| sine matrixTraining size (N)=3,0002017.12 | — | 0.37 | — | — | — |