Fig. 3: The prediction of HamGNN on several periodic solids that are not present in the training sets. | npj Computational Materials

Fig. 3: The prediction of HamGNN on several periodic solids that are not present in the training sets.

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

Fig. 3: The prediction of HamGNN on several periodic solids that are not present in the training sets.

a–d Crystal structures of pentadiamond, Moiré twisted bilayer graphene (TBG), Si (MP-1199894), and SiO2 (MP-1257168). e–h Comparison of the Hamiltonian matrix elements predicted by HamGNN and those calculated by OpenMX for pentadiamond, TBG, Si (MP-1199894), and SiO2 (MP-1257168). i–l Comparison of the energy bands predicted by HamGNN (solid line) and those (dashed line) calculated by OpenMX for pentadiamond, TBG, Si (MP-1199894), and SiO2 (MP-1257168).

Back to article page