I was running a geometry optimization at MCSCF level followed by a frequency calculation. The numerical Hessian calculation took too long
for the queue and now I have to restart it.
The problem is that after NSERCH=0 in the PUNCH file I can only find
the occupied MCSCF orbitals, virtual orbitals are missing
(ie the $VEC field contains only the first 80 occupied orbitals,
not all the 348 that I was expecting).
I have restarted the job by using the optimized geometry using as
a guess only the occupied natural orbitals found in the PUNCH file,
however, the MCSCF procedure converged to a slightly different
energy (10^-5 hartree above the original energy).
My first question is: how can I make MCSCF punch all the files
for each geometry optimization step?
Second, I am not sure the energy is accurate enough to restart the
frequency calculation. Indeed the error of 10^-5 hartree is of the
same order of the energy differences of the numerical differentiation
scheme. Anyone has any suggestion on how to
improve the restart of the frequency calculation?