Name
FULLCNV — Fully converge conjugate gradient at each SCF cycle.
Description
PSOLVE=CGDMS and
PSOLVE=QNDMS solve for the density matrix
using a congugate gradient (CG) procedure. Typically, a CG procedure iterates until all of its convergence
criteria are satisfied. For CGDMS and QNDMS, it is unnecessary to fully converge things at every single
SCF, since we only need the density to be converged at the final SCF step. This leads to the notion of
progressive convergence, where we limit the CG to only 4 steps. In general, progressive convergence will
greatly reduce the number of CG iterations without increasing (or only marginally increasing) the total number
of SCF iterations and so is the default behavior for these methods. FULLCNV forces the use of
full convergence instead of progressive convergence.
