>>>>> Daniel Svozil writes: > I am trying to understand when CCSD is considered to be converged. > According to manual > (http://www.molpro.net/info/current//doc/manual/node322.html)
> "The convergence thresholds can be modified using > THRESH,ENERGY=thrden,COEFF=thrvar > Convergence is reached if the energy change is smaller than thrden > (default 1.d-6) and the square sum of the amplitude changes is > smaller than thrvar (default (1.d-10)." > What are amplitude changes? VAR(S) and VAR(P)? And what those > variables actually mean? Change in sibles and pairs energies? These are the sums of the squares of the amplitude changes between each pair of iterations [N and (N - 1)] for the singles and pairs, respectively. > There is also global directive GTHRESH, where COEFF key means > "Convergence threshold for coefficients in CI and CCSD (default > 1.d-4)" > (http://www.molpro.net/info/current//doc/manual/node62.html#command:gthresh), > but this is apparently different that what is said at CCSD THRESH > manual page. And moreover it seems to me, that the implicit value > of COEFF is 1.d-5. > How to print out CCSD coefficients to the output file? At present, you can use: ccsd print,cs=2,cp=2 to increase the print level, and this will output the singles (T1) and doubles (TIJ) amplitudes (and some other quantities) for each iteration. > I have very slowly converging dipole-bound anion and I would like > to relax convergence criteria, the energy change is negligible and > constant (1d-8), but the job doesn't stop here. You can control these thresholds in the way quoted from the manual above. So, for example: ccsd thresh,coeff=1d-3,energy=1d-5 will lead to THRVAR = 1.00D-06 and THRDEN = 1.00D-05. > Also, is it possible to restart CCSD calculation from the last > already done iteration (for both open (rsscd(t)) and closed shell > systems)? When I write int and wfu files and I use them in the next > run, only HF reference is restarted, but not CCSD. I don't think this is possible at present. -- Dr. Matt Hodges School of Chemistry University of Birmingham
