Table 1 Comparison of the MAEs of the Hamiltonian matrices predicted by HamGNN, DeepH, and PhiSNet for the molecule structures from molecular dynamics trajectories.

From: Transferable equivariant graph neural networks for the Hamiltonians of molecules and solids

Dataset

PhiSNet (meV)

DeepH (meV)

HamGNN (meV)

Water

0.808

1.048

0.760

Ethanol

0.288

0.601

0.195

Malondialdehyde

0.340

0.547

0.270

Uracil

0.487

0.470

0.264

  1. The lowest MAE among the considered models for each subsystem is displayed in bold.
Back to article page