Bug in aug-cc-pcv[t,q,5,6]z basis sets?

1

First, please, let me send you a big, big THANK YOU for making such excellent quantum chemistry software available open source to the general public!

Unfortunately, it seems, I hit a bug:

I computed the HF energies for B+ (4 electrons) with an
aug-cc-pcv[t,q,5,6]z basis.

Here are the HF energies from Psi4 (left) and from a well-known other program (right):

aug-cc-pcvtz -24.23687687337318 -24.236876244574

aug-cc-pcvqz -23.75349206862587 -24.237476720808

aug-cc-pcv5z -23.75374913354539 -24.237547009679

aug-cc-pcv6z -23.75377754514268 -24.237569639624

Here is the same for C2+:

aug-cc-pcvtz -36.40748210426819 -36.407480789809

aug-cc-pcvqz -35.72868177282092 -36.408289781221

aug-cc-pcv5z -35.72921212773497 -36.408419504400

aug-cc-pcv6z -35.72926641187144 -36.408484020785

While the aug-cc-pcvtz values agree, the others differ.
I have no doubt that the other program’s values are OK as those program’s aug-cc-pcv6z CCSD(T) value for B+ agrees well with nearly accurate
literature data.

Nevertheless, again many thanks for this excellent software!

Best regards

1 Like
2

Yes, we do have a problem with single atom runs converging to the wrong roots in SCF. You can fix this problem in three ways. Use a SAD guess with fractional occupation numbers:

set guess sad
set sad_frac_occ true

Use a GWH guess:

set guess gwh

Use a guess basis:

set basis_guess cc-pvdz

At this point we are still trying to figure out which consistently gives the correct answer. There are other examples where the above strategies break down. Other programs do this by using a precomputed density (effectively a weaker and cheaper basis_guess); however, this breaks down if the molecule is in a strange state which is often difficult to detect. I will say that the current default CORE guess for a single atom is certainly not optimal. More feedback is always welcome so that we can build up a library of these corner cases!

Glad you find Psi4 useful!

3

Thanks for the quick reply!

I tried “set basis_guess cc-pvdz” and it converges to the proper SCF root. So, Psi4 works fine even for atoms.

Anyway, Quantum Chemistry is about molecules, not about atoms and the basis sets or optimized to describe bonds. I was just starting to get the program running at all in my environment.

Excellent work! Thanks again!

Ludger