Electronic structure calculations are the computational bottleneck in all nonadiabatic dynamics simulations, exhibiting both a high prefactor and unfavorable scaling behavior. In particular, dynamics calculations that require multireference methods have been limited to small molecular systems. We alleviate this problem by implementing graphical processing unit (GPU) accelerated state-averaged complete active space self-consistent field (SA-CASSCF). Our algorithm is ~10-100x faster than the nearest competitive code and exhibits effective scaling of O(N2). To demonstrate the utility of this development to nonadiabatic dynamics simulations, we have applied GPU accelerated SA-CASSCF coupled to ab initio multiple spawning (AIMS) to unravel the photodynamics of provitamin D3, the largest simulation of its kind. We also discuss parametric methods that efficiently build dynamic electron correlation into the SA-CASSCF method, alleviating the single greatest shortcoming of the method.

GPU-accelerated state-averaged complete active space self-consistent field and derivative methods enable accurate, large-scale nonadiabatic dynamics simulations

253rd National Meeting (2017)

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