Boost Efficiency for Stochastic Density Functional Theory with a Unified Strategy

Stochastic density functional theory (sDFT) can achieve linear scaling or even sub-linear scaling for calculating many ground state properties. However, a large number of stochastic orbitals are required to reduce stochastic noises, leading to significant computational costs. Therefore, developing noise reduction sDFT methods is necessary to increase the efficiency. We have proposed an unified approach that combines the overlapped fragmentation scheme and energy window scheme. This new approach can significantly reduce the stochastic noise even compared with the overlapped fragmentation scheme. The method was tested with a system of a g-center in bulk silicon.