AMBER Archive (2002)

Subject: sander minimizations failing

From: John Bushnell (bushnell_at_chem.ucsb.edu)
Date: Tue Aug 13 2002 - 19:03:12 CDT

We are in the process of changing over from Amber6 to
Amber7 and are having troubles. Currently, the most
difficult problem to overcome is that we are seeing
LOTS of LINMIN errors when using sander to minimize
geometries between annealing steps. It works fine
for really simple molecules, but falls apart on more
complex systems. We have seen this for Amber7 compiled
using ifc on a P4 and ifc and g77 on an Athlon, all
of which seemed to pass the sander tests fine. We do
not have the library necessary to use nmode for
minimization. We seem to be able to (almost) get
things to work by converting the topology file using
new2oldparm and then using sander_classic from Amber6,
though that chokes on the raw coordinate file created
by saveamberparm, and is rather awkward. Also, we see
small but significant differences between the energy
of a particular structure given by sander_classic and
sander (Amber7). It appears that there have been some
significant changes in runmin.f since Amber6.

    Anyone have suggestions? Is the new minimizer
known to be buggy?

    Thanks for any help! - John