In going from AmberTools version 1.2 to 1.3, we changed the default quantum
chemistry program from mopac6 to our Amber's "sqm" program. Here are a
few comments about this change:
1. Why did we do this? We were never able to find a mopac variant that would
compile and run cleanly with all common fortran compilers. Our Mopac6 code
required GNU compilers (either g77 or gfortran), and it was a pain to have to
use different compilers for different parts of Amber. More importantly, we
could never really get the same am1-bcc charges on different machines, since
Mopac6 (at least in our hands) would never truly converge the geometry
optimization.
[As an aside, we also wanted to distribute "sqm" for purposes other than
just computing am1-bcc charges; sqm implements many other semi-empirical
and DFTB Hamiltonians, and will be the basis for for QM and QM/MM
developments in Amber as we move forward. For an overview of technical
differences between mopac our our codes, see R.C. Walker, M.F. Crowley
and D.A. Case. The implementation of a fast and accurate QM/MM potential
method in Amber. J. Comput. Chem. 28, 1019-1031 (2008). ]
2. Do am1-bcc jobs take too long? In an effort to get truly reproducible
am1-bcc charges across machines, operating systems and compilers, we set
geometry convergence defaults very tight in AmberTools1.3. This may be a
problem if you are doing lots of molecules, or if you are used to much faster
(but less precise) results from mopac. You can use the "-ek" keyword in
antechamber to change the precision requested, or you could change the
defaults (so you don't need the -ek keyword) by editing line 38 in
amber11/src/antechamber/common.c, and recompiling. One set of parameters
that has been suggested by Romain Wolf is this:
grms_tol=0.05, tight_p_conv=1, scfconv=1.d-8
3. Do am1-bcc jobs completely fail? We have found some cases where the new
codes fail at the scf convergence stage, and are working on improving our scf
convergence code. You should apply bugfix.1 for AmberTools1.3, which can be
found at the Amber web site. This doesn't fix the problem, just gives a
better error message and quits sooner. If you find such a failure, especially
one that worked with AmberTools 1.2, please consider posting your structure
to the mailing list -- that will help us test proposed fixes.
...thanks...dave case
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Jan 07 2010 - 07:30:04 PST
