From chemistry-request@ccl.net Mon Jul 29 08:56:05 1991
Date: Fri, 26 Jul 91 12:22:44 -0500
From: davem@hydra.convex.com (Dave Mullally)
To: CHEMISTRY@ccl.net
Subject: Re: MOPAC6.0 geometry optimization
Status: R


It turns out that the default, BFGS optimizer in MOPAC is 
extremely sensitive to the insignificant numerical differences 
in the wavefunction.  A change in the level of optimization, 
which changes the order of operations will typically effect 
the number of optimization cycles required for an optimization, 
and can make the optimization stop before it finds the true minimium.

I was able to avoid the problem by 2 methods:

1. Better wavefunction:

   The default criteria for SCF convergence in MOPAC 6.0 is not as
   tight as it was in MOPAC 5.0.  When I set the convergence criteria
   up with the PRECISE option, the optimization avoids getting stuck,
   and I was able to reproduce the MOPAC 5.0 energy.  (I also tried the
   NODIIS option without PRECISE, but it seemed to have no effect.)

   2. Stable optimization method:

   I suggest that you try the EF option.  This addition to MOPAC 6.0 
   has given us an optimization method which seems to be numerically
   stable.  I was able to reach the MOPAC 5.0 energy without having 
   to turn on the PRECISE keyword.

Dave Mullally
Convex Computer Corp.
Richardson, TX
USA