An Iterative Dynamic Algorithm for Active Space Selection using Self-Healing Diffusion Monte Carlo

Multideterminant Diffusion Monte Carlo (DMC) displays improved accuracy compared to single determinant DMC. Methods, such as Self-Healing Diffusion Monte Carlo (SHDMC), iteratively improve a multideterminant trial wavefunction. Methods such as, CISD(T), CIPSI, or CAS-SCF are often needed to construct multideterminant trial wavefunctions. These methods are often computationally impractical for use in large solids. Therefore, developing alternative algorithms for active space selection, that can be easily applied to large periodic systems, is timely. For this goal, we have developed a dynamic determinant active space selection algorithm that is very well suited for use within SHDMC algorithm. This is a determinant "branching" algorithm that enables us to find important higher order excited determinants using determinant coefficient information from SHDMC. With each iteration we identify which determinants are important with progressively improving accuracy. To test this SHDMC based branching algorithm, we applied it to the ground state energy of a small unit cell of graphene and verified the quality of our results via comparison with results obtained from CIPSI. Application of this SHDMC based branching algorithm to monolayer MoS2 is currently in progress.