Fig. 6: Application of the DeepH method to study moiré-twisted materials.
a, Workflow for studying twisted materials using DeepH, which uses the DFT results for small non-twisted structures as training data and makes predictions on twisted structures with arbitrary twist angle θ. b,c, Band structures computed by DFT and DeepH for TBGs (b: θ ≈ 3.48°, 1.08°) and TBBs (c: θ ≈ 9.43°, 7.34°). In c, the DFT bands for the magic angle θ ≈ 1.08° are adapted from ref. 49. d, Computation time to construct the DFT Hamiltonian matrices of TBGs and TBBs with varying system size by DFT self-consistent calculations versus by DeepH. For comparison, the calculations were all done by one compute node equipped with two AMD EPYC 7542 central processing units (CPUs), although DeepH works much more efficiently on graphics processor unit (GPU) nodes.
