The influence of the electronic structure method on intersystem crossing dynamics. The case of thioformaldehyde

The ability of different electronic structure methods to describe correctly intersystem crossing dynamics is evaluated, using thioformaldehyde as a test case. Mischievously, all methods considered—ranging from the multi-reference methods MRCISD, MS-CASPT2, or SA-CASSCF, to the single-reference methods ADC(2), CC2, and TDDFT in different flavours— provide the same state ordering and energies of the low-lying singlet and triplet electronic excited states within an acceptable error of 0.2–0.3 eV. However, the outcome of the non-adiabatic simulations after excitation to the lowest S1 (1nπ∗) state are dramatically different. While MS-CASPT2, ADC(2), BP86, and PBE do not transfer population to the triplet states within 500 fs—in consonance with experimental evidence—SA-CASSCF, B3LYP, and BHHLYP predict intersystem crossing yields between 3% and 21% within the same time. The different excited state dynamics can be rationalized by inspecting potential energy profiles along the C–S bond stretch mode and single-triplet energy gaps. It is found that already at a C–S bond length of 1.9Å, all the single-reference methods struggle to describe the correct asymptotic behavior of the potentials. Moreover, some methods, including SACASSCF, obtain incorrect 1nπ∗ − 3ππ∗ energy gaps, leading to compensation of errors (ADC(2), BP86, PBE), or wrong dynamics (SA-CASSCF, B3LYP, BHHLYP). Only the accurate MRCISD and MS-CASPT2 methods are able to describe the C–S bond correctly and thus able to deliver the correct potential energy surfaces and dynamics for the right reason. A correlation with the amount of Hartree-Fock exchange in the density functional and the easiness to access the 3ππ∗ state from the 1nπ∗ is able to explain the different behavior observed for GGA and hybrid functionals. It is thus illustrated that even in the case of a simple molecule, like CH2S, the sole assessment of vertical excitation energies as reliability predictors for non-adiabatic is inadequate. The reason is that ISC does not occur at the FC geometry, but rather at distorted geometries where the singlet-triplet gaps become small. Hence, a characterization of the potential energy surfaces beyond the Franck-Condon region is mandatory.