📝 SummarXiv

Abstract

Computational materials science has evolved toward materials informatics, where large datasets of complex, multispecies compounds are generated and evaluated using density functional theory (DFT). Materials genome projects mine these datasets for candidates with breakthrough properties, but existing databases remain limited to compounds with relatively small unit cells due to computational cost. Exascale computers now provide the power to simulate larger and more chemically realistic systems, but fully realizing this potential requires DFT codes that can efficiently scale to thousands of processors. Our real-space multigrid (RMG) DFT code's grid-decomposition approach scales nearly linearly with the number of GPUs, even for simulations exceeding thousands of atoms. This scalability makes RMG a compelling tool for high-throughput DFT studies of materials that would otherwise be bottlenecked in other codes (for example, by global Fast Fourier Transforms in plane-wave DFT). In this work, we present pyRMG, a Python package designed to streamline the setup and execution of RMG DFT calculations. Built on the pymatgen and ASE Python packages, pyRMG automates input generation and convergence checking, and integrates with modern job schedulers (e.g., Flux) on leadership-class platforms such as Frontier and Perlmutter. We demonstrate pyRMG for a high-throughput study of interfacial strain and twist-angle effects in lattice-matched, 2D Bi$_2$Se$_3$/NbSe$_2$ heterostructures, which form large, anisotropic supercells. Our results link strain and twist angle to material informatics properties, including stability and band gap, and show that pyRMG can initialize and converge challenging RMG-based workflows with limited user intervention.