SODFT calculation on small fullerene with aug-cc-pvtz fails to converge

[Researcher Joe]

Hello, community!

I'm doing an SODFT calculation on a small C34 fullerene with aug-cc-pvtz basis. In the first few scf cycles the energy quickly drops to unrealistically low values and keeps dropping (I performed longer simulations, it never recovers from those unrealistically low values). For this project I cannot use a basis set without diffuse functions and I need to do spin-orbit calculation. I know that the problem is probably due to the diffuse functions, I know that in standard DFT calculations the solution would be along the lines of basis set projection, etc., but I think basis set projection is not implemented in SODFT. 

Any suggestions on how to converge my calculation?

Input and output files attached.

Thank you!

[Miro]

Hello Joe !

I would try first quick (scalar relativistic) geometry optimization of your system. After that in obtained geometry I would try DFT in big basis, and if it works, I would proceed to SODFT.

Or if you are sure your geometry is OK, try not SODFT, but only SOHF.

Best, Miro

[Researcher Joe]

Hi Miro, 

Thank you for your advice but it doesn't help. The problem is not related to geometry. I can optimize geometry in SODFT with cc-pvtz basis or I can optimize geometry in DFT with aug-cc-pvtz. The problem arises only when I use this optimized structure to do SODFT calculation with aug-cc-pvtz, i.e., it is clearly related to the addition of diffuse functions while doing SODFT energy calculation. Also, I do not know what did you mean by mentioning SOHF, but if you mean doing SODFT calculation using HF exchange, then this is also something I tried and the result is same as when I use xc b3lyp, i.e., SODFT calculation converges with cc-pvtz basis but fails to converge with aug-cc-pvtz basis.

Any other ideas, please?

Thank you!

[niri....@gmail.com]

Hi Joe,

The reason it fails for aug-cc-pvtz with SODFT is because of linear dependencies. The SODFT module does not have the same handling as the scalar DFT code yet. It is on our todo list.

Best,

-Niri

[Researcher Joe]

Hi Niri,

Thank you for the explanation. So, what'd you suggest? Eliminate linear dependencies by hand (there are so many of them... is there an efficient way to do it by hand that you can point me to)? Or give up on doing this SODFT calculation with aug-cc-pvtz in NWChem? I like the program, so would like to stick with it if possible.