The definition of aug-cc-pVDZ is inconsistent with BSE, ORCA and Molpro

[Tibor Győri]

Dear all,

It seems to me that the basis set library that ships with NWchem is not consistent with the Basis Set Exchange or other programs like ORCA 5 and Molpro.

I tried a test calculation for the EtH + F· system, and was extremely surprised to find that the ROHF/AVDZ energy is lower than expected. Apparently NWchem, for some reason unknown to me, uses a different definition of the aug-cc-pVDZ basis set that contains a few extra basis functions, thus lowering the energy.

Pasting the appropriate output from BSE into the NWchem input brings the computed energy into agreement with other software.

Does anyone know why NWchem does not use the same definition of "aug-cc-pVDZ" as BSE and other programs? As I said, not even the number of basis functions matched. Is this intentional or a bug? If it is intentional, I feel like this is something that should be plastered all over the manual, like the different B3LYP definitions.

Best,

Tibor

[Edoardo Aprà]

Could post the two input files (and as the well the two associated output files) of the aug-cc-pVDZ problem you are reporting?

[Tibor Győri]

Sure thing.

Now that I actually diffed the outputs, I have a hunch that this is somehow related to spherical/cartesian basis set handling.

But it is still rather surprising.

[Edoardo Aprà]

The diff output on the input file is pretty  clear

$ diff nwchem_t.in     nwchem_t_lib.in | grep -v \#|grep -i basis
< BASIS "ao basis" SPHERICAL PRINT
> basis

If you believe that the documentation is not explicit enough about stating that cartesian functions are the default, please suggest how to improve it

https://nwchemgit.github.io/Basis.html#spherical-or-cartesian

A simple search of this group

https://groups.google.com/g/nwchem-forum/search?q=default%20cartesian

would show that this same issue has already been discussed a few times.  

[Tibor Győri]

Ah, thanks, I have missed that part of the manual, now it does make sense.

However, I think that this a case of "surprising deafults". The manual states that using cartesian with cc-pVnZ sets is not recommended. Yet NWchem defaults to using cartesian functions with cc basis sets.

Thus current default program behavior is to do things the manual recommends against. IMO NWchem should, at a minimum, print a big fat warning in the output file when using cartesian + cc sets.

The number of times this has been discussed supports that this is counter-intuitive behavior, a trap for new users to fall into. I will see if I can find a way to improve the manual or the program itself.