During a MOZYME run, the steady accumulation of small errors (for example,
roundoff, and more important, the elimination of atoms from LMOs) will
result in the LMOs becoming non-orthogonal. These errors can be eliminated
almost completely by re-orthogonalizing the LMOs. If REORTH
is present, then the LMOs will be re-orthogonalized at the end of every
10th SCF calculation, starting with the first SCF.
A useful strategy would be to do geometry optimizations without using
REORTH, and allow the LMOs to slowly become more and more non-orthogonal.
Then, either at or near the end of the optimization, do a 1SCF with
REORTH to eliminate the accumulated errors.
The effect of REORTH is likely to be very small. If the value of THRESH
is increased, however, then the LMOs will become non-orthogonal faster, and
REORTH would then likely have a significant effect.