The presence of low-lying minimum energy conical intersections (MECIs) is thought to be one of the primary mechanisms responsible for nonradiative decay, and multireference quantum chemical methods are often used to locate and characterize these phenomena. The complete active space self consistent field (CASSCF) method is widely considered to be the standard for describing such systems. While CASSCF gives a qualitatively accurate description of the electronic properties in these cases, the calculation is computationally expensive, and the presence of multiple solutions to the CASSCF equations can give rise to nonconvergence of the electronic wavefunction and discontinuities in the potential energy surface. Certain "two-step" methods such as the improved virtual orbital (IVO) (Potts, D. M. et. al. J. Chem. Phys. 2001, 114, 2592-2600) and configuration interaction singles natural orbitals (CISNO) (Shu, Y. and Levine, B. G. submitted for publication) methods, where the orbital and configuration interaction (CI) vector optimization steps are uncoupled, perform favorably in terms of wavefunction convergence, accuracy, and overall computational cost. We have implemented a vectorized CASCI algorithm using graphical processing units, and observe significant speedups when compared to highly optimized CPU code running on modern hardware when computing moderately sized active spaces (106-108 configurations). This improvement is in addition to the ~100x speedup the TeraChem SCF code offers over competitive CPU based packages (Molpro), and has allowed us to study the local energy landscape in the region of MECIs of molecules and clusters on the nanoscale.

Graphical processing unit acceleration of "two-step'' complete active space configuration interaction (CASCI) methods

249th National Meeting (2015)

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