[Pw_forum] comparing gaussian03 with pwscf

[O. Baris Malcioglu]

I completely agree with you, but I think what I have in mind is much 
more trivial than that. When calculating a system in Gaussian using AE 
wavefunctions you are to find a basis set that describes all atoms in 
the structure simultaneously, unless you are using something like an 
ONIOM model, which has added complications. For example, when using 
PWSCF, you can use  La.pbe-nsp-van.UPF 
<http://www.pwscf.org/pseudo/1.3/UPF/La.pbe-nsp-van.UPF> for the 
Lanthanide and say C.pbe-van_ak.UPF 
<http://www.pwscf.org/pseudo/1.3/UPF/C.pbe-van_ak.UPF>  for the carbon, 
and since the valance set will consist of  (different combination of) 
planewaves on -both- sides, the both PPs are interoperable. On the 
contrary, using something like 6-31G* for the Carbon and LanL2DZ for the 
Lanthanide is not possible straightforwardly in a Gaussian AE 
calculation since the very definition of valance basis functions are 
different on these sets. Looking at the corresponding manual at 
http://www.gaussian.com/g_ur/m_basis_sets.htm you can see that for a 
basis set that describes both La and C simultaneously, the choices are 
quite limited, and coarse if used straightforwardly. Thus, I think there 
might be a chance that the charge difference  is arising from having 
forced to use too coarse basis sets.

Well, I hope another view on the matter helps.

O. Baris Malcioglu
SISSA CM sector,
Trieste,  Italy



Axel Kohlmeyer wrote:
> i beg to differ. there may be basis set completenes issues when
> using a too small basis set, but why should you need to tailor
> the carbon basis set for lanthanoids? the whole point of having
> defined basis sets is, that those _are_ balanced. so that should
> at most be a minor effect.
>
> i would rather expect the differences arising from much
> more obvious issues. i.e.,
>
> a) the fact that you have to project your plane wave wavefunction
> on an atomic basis set and that this projection cannot be
> complete, since you don't have a complete basis set that
> you project on. rather you use a minimal basis built from the
> atomic pseudowavefunctions. you can easily lose 1-2 electrons here.
>
> b) as paolo already noted, the differences between different ways
> to perform the population analysis. e.g. mulliken vs. loewdin vs.
> nbo vs. ...
>
> if you have access to the CPMD (www.cpmd.org) code you can easily see
> how large those effects can be, as the code allows to project on
> either atomic pseudo wavefunction or a minimal slater basis (or any
> gaussian basis you define in the input) as well as it supports mulliken
> and loewdin like analysis. comparing the dependence of the thusly
> calculated partial charges from plane waves on the basis set and
> method is quite educational... (not to mention the
> impact of pseudopotentials, plane wave cutoff etc...).
>
> we are frequently tempted to put too much trust into the many
> digits of numbers that we calculate. their absolute accuracy
> can be _much_ less.
>
> cheers,
> axel.