Session: COMP Poster Session:

Computational models can provide insight into photochemistry, where light interacts with molecules to drive chemical reactions.
Multireference theories based on configuration interaction (CI) accurately describe the electronic structure of photochemical reactions.
This work describes two improvements to the CI methodology that enable simulation of complex photochemical processes.

Photochemical reactions with a single reaction coordinate can be accurately simulated with small orbital active spaces.
More complicated reactions require correlation of additional electrons.
The seminal graphics processing unit (GPU) implementation of a CI algorithm enabled routine use of wave functions with more than 150 million basis functions - a 10x improvement over contemporary alternatives.
In the present work I describe technical and algorithmic advances that further extend GPU CI by >10x, enabling time-dependent simulation of molecules with multiple reaction coordinates, or where many electrons need to be correlated.

In contrast to static studies, time-dependent simulation explores nonequilibrium regions of conformational space.
Degeneracies between electronic states can hinder convergence in Slater determinant (SD) basis CI calculations.
Despite nealy 4 decades of work trying to circumvent the poor convergence of the SD basis, technical reasons make SDs preferred over the more robust configuration state function (CSF) basis.
To take advantage of both computational efficiency of the SD basis and robustness of the CSF basis, we have developed a GPU CSF/SD transformation that enables a hybrid-basis CI program.
Taken together, GPU CI algorithms, which enable more than two orders-of-magnitude larger orbital spaces (measured in SDs) than conventional CI programs, and hybrid CSF/SD basis diagonalization methods, provide an efficient way to simulate photochemistry using time-dependent first-principles methods.

Myoglobin HF-CAS-(16,16)-CI

Todd Martinez

Appearances