[CP2K:3177] Re: error in bond dissociation energy of gas phase molecule

[Fawzi Mohamed]

On 6-apr-11, at 23:09, Jun wrote:

> Many thanks for the comments, Fawzi.
>
>> That is exactly what I wanted to comment on:
>> 1) GTH is a pseudopetential, and also has scalar relativistic
>> correction (should be small in this case, but you never know)
>
> I guess relativistic correction in O pseudopotential might be too
> small to lower the binding energy by 0.5 eV. Possibly, something else
> in the pseudopotential.
>
>> 2) basiset for GTH pseudopotential are different from those on all
>> electrons, they should be more or less comparable, but there is no
>> guarantee, what can be compared is the basset limit in both cases
>
> We used the same basis set as used in Gaussian (6-31g*).  The basis
> set convergence was checked in Gaussian, and the calculated binding
> energy is the same as in literature.

In the inputs you gave you use TZV2P-GTH as basis set which is *not*  
6-31g*, using a pseudopotential the wavefunction is different from the  
all electron wavefunction, so you cannot use the same basis for the  
core part.
Aside from this 6-31g* can't even be properly represented with 280 Ry  
cutoff.
To compare you need to go to the basis set limit in *both* gaussian  
and cp2k (all electron calculation will probably have a larger BSSE),  
6-31g* is not a particularly good basis set.
You cannot use the same basis set in Gaussian and cp2k (unless you do  
all-electron GAPW).
If really the basiset limit of GPW-GTH is much different from the one  
of gaussian, then yes probably it is an effect due to the  
pseudopotential.

Please note that gaussian used to call B3LYP a functional that does  
not correspond to the B3LYP functional published (they used a  
different VWN uniform density parameterization).
I heard that the latest version might have fixed this, but I am not  
sure. (still I don't expect this difference to produce such large  
energy changes)

>> 3) a cutoff of 280 for OH is a bit on the low side, especially if you
>> don't do any smoothing of the xc part, I would increase that
>
> We did check this. Increasing the cutoff to 400 doesn't make a
> difference.
> In the all electron calculation, the cutoff seems even lower than 280
> (directly taken from sample input without any change). We should check
> this. But not sure this could be the cause of the increase of 0.8 eV
> in binding energy. I am more concerned on the other settings for GAPW
> calculations I don't understand.

using the GTH pseudo both GAP and GAPW should give the same results  
(allelectron is different).
Please note that
1) sample (regtest) files are built to execute quickly, not to give  
accurate results
2) convergence wrt. cutoff is *not* variational

Fawzi