SALMON 2.3: Implementation of divide-and-conquer ground-state initialization for large-scale real-time TDDFT
In large-scale real-time time-dependent density functional theory (TDDFT), preparing the ground-state electronic structure can be more expensive than the subsequent time propagation, limiting simulations of nonequilibrium electron dynamics in realistic systems containing thousands of atoms. This bottleneck is especially important for disordered materials, liquids, nanostructures, and heterogeneous condensed-matter systems, where nonlinear and strong-field phenomena such as high-harmonic generation and light-induced phase transitions require explicit real-time treatment. SALMON is an open-source first-principles code for light-matter interaction simulations based on real-time TDDFT on real-space grids, supporting massively parallel calculations with MPI combined with OpenMP or GPU acceleration. In SALMON 2.3, we implement a divide-and-conquer density functional theory (DC-DFT) scheme and combine it with a postprocessing method that reconstructs spatially extended Kohn-Sham orbitals of the entire system. These reconstructed global orbitals are used directly as initial states for the standard real-time, real-space TDDFT module of SALMON. The resulting workflow connects efficient DC-DFT ground-state preparation to conventional real-time TDDFT. The DC-DFT self-consistent-field procedure exhibits linear scaling with system size, addressing a major bottleneck in large-scale electron-dynamics simulations while retaining the robustness and broad applicability of SALMON's established time-propagation scheme. We describe the computational procedure, parallelization strategy, and input/output design. Weak-scaling tests using Si supercells on Fugaku confirm the linear-scaling behavior. Accuracy tests for a 512-atom amorphous Si system and a bulk H2O liquid system containing 4134 atoms demonstrate that the workflow enables practical large-scale real-time TDDFT simulations.