Lately, I have run into the fact that there are some functionals in Gaussian for which 3rd derivatives are not supported. This means the following jobs are unavailable with them:
*never experienced myself, but these methods definitely deal with the 3rd derivatives
No blame for Gaussian, Inc. — everything is in the Reference. But it is at the end of description of DFT Methods in the printed Reference, so can be overlooked (I did so); on the web page, it is at the "Availability" tab (that's probably more convenient).
Affected functionals for G09 are:
For G16:
Of course, all higher-rung functionals constructed from these are also affected, e.g., TPSSh in G09. That means TPSS in G16 does have 3rd derivatives — another reason to buy the new version!
When a job dies due to such an unavailability, one of the following messages is printed multiple times (probably it is the number of second derivatives which would be differentiated again):
Invalid value of MaxDer in TPSSx
No func 3rd derivs with HSE (interestingly, I got it while using LC-wPBE).
Mentioning HSE is interesting because it is HSE03 exchange which is used for the short-range in LC-wPBE and not the PBE one, both in original paper by Vydrov and Scuseria of 2006 and as implemented in Gaussian. This is no nearly obvious, but it's how it is. Some renowned authors (for example, Takao Tsuneda in his year 2014 monograph about DFT; see p. 129). still erroneously assume it to be PBE.
- Polar=DCSHG (hyperpolarizabilities)
- SCRF=ExternalIteration (state-specific solvation)
- Freq=Raman (Raman intensities) *
- Freq=Anharmonic (anharmonic corrections) *
- maybe there are more, but I have never run into them…
*never experienced myself, but these methods definitely deal with the 3rd derivatives
No blame for Gaussian, Inc. — everything is in the Reference. But it is at the end of description of DFT Methods in the printed Reference, so can be overlooked (I did so); on the web page, it is at the "Availability" tab (that's probably more convenient).
Affected functionals for G09 are:
| Exchange (x): | Correlation (c): | Hybrid: | |
|---|---|---|---|
| Gill96, P, BRx, PKZB, TPSS, wPBEh, PBEh | PKZB, TPSS | HSE1PBE, HSE2PBE |
For G16:
| Exchange (x): | Correlation (c): | Hybrid: | |
|---|---|---|---|
| G96, P86, PKZB, wPBEh, PBEh | PKZB | OHSE1PBE, OHSE2PBE |
Of course, all higher-rung functionals constructed from these are also affected, e.g., TPSSh in G09. That means TPSS in G16 does have 3rd derivatives — another reason to buy the new version!
When a job dies due to such an unavailability, one of the following messages is printed multiple times (probably it is the number of second derivatives which would be differentiated again):
Invalid value of MaxDer in TPSSx
No func 3rd derivs with HSE (interestingly, I got it while using LC-wPBE).
Mentioning HSE is interesting because it is HSE03 exchange which is used for the short-range in LC-wPBE and not the PBE one, both in original paper by Vydrov and Scuseria of 2006 and as implemented in Gaussian. This is no nearly obvious, but it's how it is. Some renowned authors (for example, Takao Tsuneda in his year 2014 monograph about DFT; see p. 129). still erroneously assume it to be PBE.
Thanks a lot for clarifying this!
ReplyDeleteThank you this is really helpful. I ran into this issue while optimizing the energy of an excited state with LC-wPBE. Were you able to proceed or did you have to switch methods?
ReplyDeleteNo, I had to switch methods. If it is not implemented, then it is not implemented :-/ Apparently the same is situation with Gaussian 16, as only for TPSS and BRx the feature was implemented in that release. We can wait until next minor release, though.
DeleteThank You all for your "thank you's" ! :-)
Actually it is a very good question! It seems that in Gaussian LC-wPBE is constructed not from the PBE (with ω=0.4) but from HSE (called wPBEh by Gaussian – see my DF catalogue), for which the third-order derivatives are not available. So maybe LC-wPBE can actually be recovered, at least in its year 2006 version, as for it T. Tsuneda states in his book on DFT that it is indeed constructed from just the PBE. I will ask Gaussian people and update the post! Thank You for Your reaction!
DeleteHi Igors, did you ever hear back from the "Gaussian people"? :P
DeleteYes, sorry, I did it but postponed updating the post for too long. Turns out they are right, and definition of LC-wPBE is far from obvious. It does use PBE correlation, though.
Delete