PM6

[Teodoro Laino]

Just for record:

The analytical derivatives for PM6 and other SE methods using d- 
orbitals have been implemented and are
available in the CVS.

Cheers,
Teo

On 30 Apr 2008, at 17:36, Teodoro Laino wrote:

> Dear All,
>
> good news:
> with the today commit, (finally !!), there's the possibility to run  
> PM6 semi-empirical calculation in
> CP2K (fully support for all elements) (at the moment only numerical  
> derivatives.. analytical will come soon).
>
> bad news:
> In the original version of PM6 there are some terms that make the  
> heat of formation a discontinuous function of
> the nuclei positions (are just correction terms, constants that  
> become zero for different topological configurations) that
> we didn't implement. Therefore for few situations there could be  
> discrepancies between the heat of formation of CP2K
> and the one of MOPAC (but the electronic energy should anyway be  
> the same (within numerical  errors (see below)).
> An example is H-CC-H, that has a constant correction term to the  
> heat of formation of the triple bond that becomes
> zero when the triple bond is broken (it's just a step function).
>
> Why have been these terms omitted?
> Simply because there's no trace of them in the PM6 paper.
>
> I'm sure that there could be other hidden corrections/parameters  
> not explicitly mentioned in the paper (classical correction
> like the one for correcting the sp2 pyramidalization, that we  
> didn't implement as well), so in case of big discrepancies
> compared with MOPAC2007 feel free to post here your input file and  
> a short description of the problem.
>
> What does it mean big discrepancies?
> These are the values for the heat of formation (kcalmol) of the  
> following systems (CP2K/MOPAC2007)
>
> HCl  (17.0356/17.0345)
> BrCl (2113.534/2113.493)
> TiO (202.714/202.633)
>
> If you have some discrepancy larger than the ones above, check  
> always the value of the electronic energy that
> must anyway be VERY similar (the only difference is in the way we  
> compute integrals.. so just kind of numerical
> errors, totally negligible).
> Have fun!!
>
> Teo
>