Sensitivity Analysis in Photodynamics: How Does the\nElectronic Structure Control <i>cis</i>-Stilbene Photodynamics?

The techniques of\ncomputational photodynamics are increasingly\nemployed to unravel reaction mechanisms and interpret experiments.\nHowever, misinterpretations in nonadiabatic dynamics caused by inaccurate\nunderlying potentials are often difficult to foresee. This work focuses\non revealing the systematic errors in the nonadiabatic simulations\ndue to the underlying potentials and suggests a thrifty approach to\nevaluate the sensitivity of the simulations to the potential. This\nissue is exemplified in the photochemistry of <i>cis</i>-stilbene, where similar experimental outcomes have been differently\ninterpreted based on the electronic structure methods supporting nonadiabatic\ndynamics. We examine the predictions of <i>cis</i>-stilbene\nphotochemistry using trajectory surface hopping methods coupled with\nvarious electronic structure methods (OM3-MRCISD, SA2-CASSCF, XMS-SA2-CASPT2,\nand XMS-SA3-CASPT2) and assess their ability to interpret experimental\nobservations. While the excited-state lifetimes and calculated photoelectron\nspectra show consistency with experiments, the reaction quantum yields\nvary significantly: either completely suppressing cyclization or isomerization.\nIntriguingly, analyzing stationary points on the potential energy\nsurface does not hint at any major discrepancy, making the electronic\nstructure methods seemingly reliable when treated separately. We show\nthat performing an ensemble of simulations with different potentials\nprovides an estimate of the electronic structure sensitivity. However,\nthis ensemble approach is costly. Thus, we propose running nonadiabatic\nsimulations with an external bias at a resource-efficient underlying\npotential (semiempirical or machine-learned) for the sensitivity analysis.\nWe demonstrate this approach using a semiempirical OM3-MRCISD method\nwith a harmonic bias toward cyclization.