PM6 - validated

[Teodoro Laino]

Dear All,

this mail is just to let know all people interested in the  
semiempirical (SE) method PM6 (or people that may want to give it a  
try) that
the method has been EXTENSIVELY validated compared to MOPAC2007.
The number of molecules tested at the moment is 2021 and the whole  
test requires some comments:

(1) The error on the electronic energy and core-core energy is always  
smaller than 0.1%.

(2) For 38 molecules the electronic energy is between 0.01% and 0.1%  
and for all the rest of the set it is much smaller than 0.01%  
(reflecting
most of the times in errors smaller than 1 kcal/mol).
I didn't check carefully the 38 molecules (out of 2021) but at first  
sight the difference in slightly larger errors is due
to the convergence of the SCF to different solutions for CP2K and  
MOPAC (In quite few cases it was a terrible nightmare to have CP2K and
MOPAC converged at the same value).

(3) In a very limited number of cases the errors for the core-core  
energy is around 0.1-0.2% clearly indicating a missing term in the  
treatment
of the core-core for those specific kinds of nuclei. Nothing to  
really worry about since at the moment the problem is localized only
for Fe-Si atoms (Si-Fe pair interaction.. while Si and Fe with other  
atoms are ok).
So.. Keep in mind that for Fe-Si looks like there is a kind of  
missing term in the energy compared to the information published on
the paper J.MOL.MOD. 13:1173-1213..

(4) Heat of formation: this is the worst part of the game and the  
most fragile.. the amount of errors in the heat of formation is quite
large. Why that? Personally I think that  heat of formations have  
been tuned in order to  reproduce the experimental data in a
quite wild way.
Most of the time have been added functions (not even published) non  
continuous, depending on the topology or atomic kinds,
just to get back the correct heat of formation. Maybe I'm wrong but,  
for the purpose for which PM6 could be useful to the CP2K community,
I don't think that heat of formations deserve so much attention.
Anyway, please, keep in mind that heat of formation printed by cp2k  
could be totally unreliable.
This should not be seen as limitation, since if you use the SE for MD  
or GEO_OPT or METADYN or whatever dynamical
methods you're always interested in differences of total energy..  
anyway.. it's important to know that..


Attached the whole set of data generated by this comparison (with  
electronic energies, core-core energies, and heat of formations:  
CP2K, MOPAC,
%Error) for your info and fun ;-).

Ciao,
Teo
-------------- next part --------------
A non-text attachment was scrubbed...
Name: out.12-05-2008
Type: application/octet-stream
Size: 327809 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20080512/0d385dbe/attachment.obj>
-------------- next part --------------