[molpro-user] about uccsd(t)

Gerald Knizia knizia at theochem.uni-stuttgart.de
Tue Jan 1 16:34:14 GMT 2008

On Tuesday 01 January 2008 16:46, rachel.crespo at uam.es wrote:
>  tried to do some uccsd(t) and program shows the following error:
>  ? Error
>  ? No convergence
>  ? The problem occurs in uccsd
> I tried to change the number of iteractions but I can do it.

You can set the maximal number of iterations like
   {uccsd(t); maxit,137}
but most likely this is not your main problem here, as RHF-UCCSD(T) really is 
supposed to converge in at max 25 iterations except in the most pathological 
of cases or if something different is wrong.

Typical things to look out for:
  o Is the geometry specification correct? Molpro outputs bond lengths and 
bond angles in the beginning. You should check if they are reasonable. 
Especially look for Bohr/Angstrom conversion issues.
  o Is the state being optimized reasonable? (charge, spin, symmetry and so 
  o Are the chosen basis sets the ones you intended? E.g., if specifying basis 
sets manually, did you provide functions for each required angular momentum?

Other things to consider:
  o If using very large or very diffuse basis sets (e.g., AV5Z and larger), 
you might need to increase the integral accuracy in order to reach high 
convergence thresholds:

> In some cases the calculation stop with the same message but the energies
> seem to be converged.

In the iteration log two of the outputs are VAR(P) and VAR(S). These specify 
how the excitation amplitude vectors changed in the current iteration and are 
rather direct measures of the residual size, i.e., of how closely the current 
amplidues solve the uccsd equations they are supposed to solve. If these 
don't tend to zero, the equations are not solved and you cannot trust the 
energies obtained, even if they _seem_ converged.

That being said, the output you posted looks rather puzzling. I'd assume 
something like that could only happen if uccsd is jumping between two 
different fixed points in the iterations. Could you provide the input this 
originates from?
Gerald Knizia

More information about the Molpro-user mailing list