CCL:G: Gaussian03 Error: Density matrix is not changing but DIIS error - Suggested solutions

 Sent to CCL by: Eric Breynaert [eric.breynaert_._biw.kuleuven.be]
 Thanks for the suggestions. I'm giving an overview below.
 1/
  - SCF=qc will probably solve the problem, albeit at a cost
  - Change the SCF converger to either SD, Quadratic or Fermi
 2/
  - lower the symmetry of optimize with and optimize
    with the "nosymm" keyword
I solved the problem using a variation on the first suggestion. Normally the scf took less than 80 cycles to converge. So i used scf=(Maxconventionalcycles=100,xqc) which resulted in a good compromise between using scf=qc and optimisation speed. In the case of the DIIS error the scf always took more than 100 cycles before the error, so by adding scf=(Maxconventionalcycles=100,xqc) the scf switched to qc after 100 cycles in the standard DIIS mode. 
 Regards,
 Eric
 u wrote:
 
 Sent to CCL by: "Eric  Breynaert" [Eric.breynaert]=[biw.kuleuven.be]
 Hello all,
I'm trying to optimize a transition metal structure using DFT ub3lyp LANL2DZ basisset for the metal center , 6-31G(d) for all other elements o.a. O, H, C, S, ....
In most cases the optimisation succeeds. But sometimes g03 generates the following error after about 20 to 80 steps in the optimisation. 
  Density matrix is not changing but DIIS error= 2.02D-05 CofLast= 3.35D-01.
  The SCF is confused.
I've already btried different approaches to get arround this but nothing has succeeded. For example, recalculating force constants on the geometry obtained 1 or 2 steps before the error. Forcing g03 to make a new guess at every step of the optimisation using guess=(mix,always).
Has anyone encountered such a problem before, or do you know a solution to it ? 
 Thanks,
 Kind Regards,
Eric-- 
 Eric Breynaert
 Labo voor Colloïdchemie
 Kasteelpark Arenberg 23
 B-3001 Leuven
 Tel: +3216321457
 Fax: +3216321998
 Disclaimer: http://www.kuleuven.be/cwis/email_disclaimer.htm