Table 5 Mean absolute errors on ANI-1x subset

From: Materials Graph Library (MatGL), an open-source graph deep learning library for materials science and chemistry

Model

Energy (meV atom−1)

Force (eV Å−1)

M3GNet

4.565/4.592/3.746

0.092/0.093/0.085

M3GNet-TL

3.923/3.968/3.381

0.081/0.082/0.075

TensorNet

4.424/4.448/3.015

0.088/0.088/0.074

SO3Net

2.281/2.286/1.596

0.046/0.046/0.035

  1. The numbers are the calculated energy and force errors of M3GNet, TensorNet, and SO3Net compared to DFT. The “M3GNet-TL" indicates the transfer learning from the pre-trained M3GNet model on ANI-1xnr dataset. The numbers are listed in the order of training, validation, and test. The dataset was divided into training, validation, and test sets with a split ratio of 0.9, 0.05, and 0.05, respectively.
Back to article page