I am seeking advice regarding the spectra I obtained by analyzing the first excited states of a squaraine dye in water. These spectra were generated using TDDFT calculations on 200 snapshots sampled every ~0.5 ns from a 300 ns trajectory (14,500 frames in total). The dye and nearby salt ions were modeled with the wb97x-d/def2-svpd functional and basis set, while water was represented as EFP fragments.
When the spectra are overlaid in groups of 100 snapshots, there is a noticeable population difference. Segmenting further into groups of 50 snapshots reveals a similar pattern.
Could these differences be attributed to microenvironmental effects (e.g., local interactions or solvent dynamics) rather than issues with convergence or equilibration? Any insights or suggestions on how to better interpret these results would be greatly appreciated.
Those are really the same thing, i.e., to say you have artifacts in the sampling is the same as saying the “micro-environmental effects” (a.k.a., heterogeneous broadening) is not sampled well enough. You can make the wiggles go away with a larger Gaussian (thus obtaining convergence in fewer snapshots), but at the expense of information loss if those wiggles are meaningful. Best to sample more snapshots until the answer stops changing.
Thank you for the clarification Dr. Herbert.
For my understanding, is this behavior expected for a relatively simple system like water? I have come across studies where authors report achieving convergence with fewer than 200 snapshots. Given that the overlayed spectra are derived from the convolution of the first half and the second half of the data, shouldn’t this represent an ensemble effect reflecting the chemistry of each segment?
I have also attached the spectra calculated using all 200 snapshots.
If I were to compute more snapshots and overlay the spectra in sets of 100 (as done here), should I expect the shape and position of the peaks to stabilize at some point? In other words, would the observed differences between sets of snapshots diminish ? Any additional thoughts or guidance on this would be greatly appreciated.
Should stabilize in the limit of infinity snapshots.
500 ps between snapshots is a quite large interval
