[molpro-user] Convergence of Triples in UCCSD(T)-F12 calculation

[Gerald Knizia]

On Thursday 19 November 2009 23:12, Ralf Tonner wrote:
> I performed a series of UCCSD(T)-F12 calculations with the VDZ-F12 basis
> set and Molpro2009.1, Patchlevel 20 (binary).
>
> Several calculations converged well except for one radical anion.
>
> After the (converged) UCCSD part I get the error message:
>    > ?CONVERGENCE NOT REACHED AFTER MAX. ITERATIONS
>    > ir12=                   257
>    > ir12=                   257
>    >
>    > ?ERROR: NO CONVERGENCE. TRIPLES WILL NOT BE DONE. THIS CAN BE AVOIDED
>    > USING THE NOCHECK OPTION.
>
> (please find output-file attached).
>
> Is there a way to tweak the convergence in this part of the calculation?
> Since the calculation takes about a week to get to this point I would
> appreciate any hints.

It is the CCSD itself which is not converging (it gives up after the maximum 
number of iterations, it's just that this error message might be better 
placed somewhat before). (T) triples are not iterative, they should work 
whenever there is a CCSD wave function.

About why the CCSD does not converge I cannot really tell you much. As you are 
dealing with an anion, you would probably be better off with a fully 
augmented basis set (like AVDZ). But if that causes the RHF to need so many 
iterations or the T1/D1 diagnostics to be so high is unsure (normally both 
would be indications of a multireference case; your molecule might be one due 
to its near-symmetry. But maybe RHF also just found a state of wrong 
spin/symmetry somehow).
-- 
Gerald Knizia