A heterogeneous central processing unit (CPU) and graphical processing unit (GPU) implementation for the variational optimization of the two-electron reduced density matrix (v2RDM) under three-particle N-representability conditions is presented. The 2RDM provides a polynomially scaling approximation to complete active space self-consistent field (CASSCF) theory. The enforcement of the partial three-body N-Representability constraint, T2, is benchmarked on the CPU and GPU algorithm using the first three molecules from the linear polyacene series. For v2RDM-based CASSCF calculations on anthracene with an active space of 14 electrons in 14 orbitals, a total speedup of 19.5 is observed when the algorithm is executed with a combination of an NVIDIA TITAN V GPU and an Intel Core i7-6850k CPU relative to the case when the algorithm is executed on the CPU alone. The enforcement of the full three-body positivity constraints is benchmarked against a system of H10. A total speedup of 4.2 is found when the algorithm is implemented with a GPU and CPU relative to a CPU alone.

GPU implementation of the three-body constraints in v2RDM-CASSCF

ACS Fall 2021

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.