After Psicon-2020, I had a conversation with Prof. Konrad Patkowski regarding SAPT calculation using PSi4. He was kind to answer my questions. However, I thought (and as per his suggestion) to bring the summary of conversation to the forum hoping it would be beneficial to everyone doing SAPT calculations. I also have some follow up clarifications based on the replies. I request to help in clarifying them.
1.Q.) Usage of the keyword., ‘Freeze Core’ (freeze_core ) in the SAPT calculations (for both SAPT0 and SAPT-DFT) and how it will affect the result.
1.A.) Freeze core can be done by setting “freeze_core true”. It will have a minor effect on the dispersion and exchange-dispersion correction, and it will speed up the calculations a little (but not necessarily by a lot).
2.Q.) What could be the best functional to choose (Mo62X over B3LYP-D3 or any other better functional) for the SAPT0 and SAPT-DFT calculations
2.A) Strongly recommend the asymptotically corrected PBE0 functional: “SAPT_DFT_FUNCTIONAL PBE0”. There should be no -D3 because you include dispersion using SAPT, not using DFT. M06-2X is also likely a bad choice because it was fitted to reproduce some medium-range dispersion, which is not the job of the density functional in SAPT(DFT).
3.Q) The importance of the usage of the keyword "sapt_dft_grac_shift_a " , “sapt_dft_grac_shift_b” and “SAPT_DFT_MP2_DISP_ALG FISAPT”
3.A) 1. The keyword “SAPT_DFT_MP2_DISP_ALG FISAPT” is just an algorithm selection and it should only affect the performance, not the results. However, the values for “sapt_dft_grac_shift_a” and “sapt_dft_grac_shift_b” are very important and have to be adjusted for every different pair of molecules, as they control the amount of shift to ensure the asymptotic correctness of the density functional. Unfortunately, the documentation of these shifts has not yet made it into the manual, but you can read about them here (search for the GRAC string):
Follow up questions
Since it is suggested to use asymptotically corrected functional, how correct would it be to use the functional, CAM-B3LYP. The reason I mention this functional specifically, is because, as you already know, range separated and asymptotically corrected CAM-B3LYP is good (better than B3LYP, PBE0 which lack long range Hartree-Fock Exchange) for predicting charge transfer (CT) excitation energy and the system I am looking into is of CT nature. And this is why, I ask if it would be meaningful to use CAM-B3LYP?
Regarding the Gradient Regulated Asymptotic Correction scheme (GRAC), thank you very much for sharing the link to the page. I am learning of lot of things from the page, which otherwise I would not have been able to. I am really thankful to you for that. Is it ok to use the modulus of Kohn-Sham HOMO energy level, for instance , the HOMO of Zn-Porphyrin and PCBM optimized at B3LYP/6-31G(d) level of theory are -0.19 and -0.20 respectively.
However in order to be consistent with the functional, for carrying the geometry optimization (of both isolated and complex) and later on SAPT(DFT) (for which values for GRAC comes from the geometry optimization step) do you think it is advisable to stick to one and the same functional for these calculations and therefore as per your recommendation it should be the asymptotically corrected one , namely PBE0 or CAM-B3LYP (based on your opinion on CAM-B3LYP).
Kindly let me know if I am not clear or need any further information.