[molpro-user] ScO bond length and basis sets

[Kirk Peterson]

Dear Emine,

note that with older versions of Molpro the DK integrals were a bit numerically unstable for the larger core-valence DK basis sets. Perhaps this is the cause of your problems with 5Z ?  The newest Molpro version is better in this regard.  The problems arise due to linear dependencies in the fully uncontracted basis set, which is used for an RI in the DK integral evaluation. The solution is to slightly modify the basis set - looking for cases where the extra tight s,p,d correlating functions might overlap too strongly with the underlying HF primitive exponents and just replacing the former by uncontracting the latter.

best regards,

-Kirk

On May 16, 2014, at 11:06 AM, Kucukbenli Emine <emine.kucukbenli at epfl.ch> wrote:

> Dear Kirk, 
> Many thanks for your reply, much appreciated!
> The results make a lot more sense like this.
> 
> Reporting here just for future reference:
> 
> Now the 3s and 3p not-correlated results for bond length looks like
> {uccsd(t); occ, 9,3,3,0; closed, 8,3,3,0; core, 6,2,2,0;}
> 
> Tz:   1.696  #cc-pwcvTz-DK/aug-cc-pwcvTz-DK 
> Qz:   1.697  #cc-pwcvQz-DK/aug-cc-pwcvQz-DK
> 5z:    not-converged   #cc-pwcv5z-DK/aug-cc-pwcv5z-DK
> 
> while the correlated results with the same basis set is
> Tz: 1.673  
> Qz: 1.668
> 5z :  not-converged
> 
> The Qz bond length is remarkably spot on with the experimental one: 1.668 Angstrom
> I am trying to get the 5z to converge without much success up to now.
> But already things are looking better,
> 
> Thank you so much again!
> Emine Kucukbenli,
> Institute of Materials, EPFL, Switzerland
> 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.molpro.net/pipermail/molpro-user/attachments/20140516/1a0d2b75/attachment.html>