[molpro-user] Integral thresholds for CCSD(T)

[Gerald Knizia]

On Wednesday 12 March 2008 11:51, you wrote:
> I'm having trouble converging the CCSD in a CCSD(T) calculation. I
> managed to converge the CCSD doing a direct calculation, and
> tightening the thresholds using:
>
> {ccsd
> direct, THR_DTRAF=1.D-12, THR_DKEXT=1.D-12
> start,4000.2
> save,4000.2}

A way to accomplish this in the non-direct CCSD is to set:

   gthresh,twoint=1.e-15
   gthresh,oneint=1.e-15

The default for twoint (1e-12) is sometimes too high when tightly packed 
molecules with high symmetric are treated with aug-cc-pV5Z and up, and then 
CCSD will refuse to converge. The above statements tend to fix this.

> The VAR(P) doesn't go below ~0.5D-08. So the question is, is there
> anything to tune to try to get the CCSD converged? 

The above should usually work. If not, specifying something like

   ccsd(t),shifts=0.2,shiftp=0.2,thrdis=1.0;diis,1,1,15,1

(i.e., increasing the DIIS dimension) might help. If molecules converge well 
in smaller bases, level shifts of 0.0 might lead to faster convergence.

> If not, is the (T) evaluation expected to give reliable results if I just
> loosen the  convergence threshold of the CCSD part?

This is hard to predict. Total energies are probably reasonable, but you need 
to be very careful with basis set extrapolation procedures.
-- 
Gerald Knizia