AMSOL and MOPAC
Netters,
The message attached below describes a chimeric MOPAC version modified by
inclusion of the subroutines from AMSOL which calculate aqueous free energies
of solvation. AMSOL is Cramer and Truhlar's modification of AMPAC 2.1.
I must point out that this simple transfer of subroutines is doomed
because of the way in which the two programs calculate derivatives with
respect to nuclear motion, i.e. geometry optimization. AMSOL requires use of
the keyword 'DERINU'. It is unfortunate that this keyword 'appears' to be a
request for derivatives calculated numerically. In fact, the default
optimization techniques for both AMPAC and MOPAC calculate derivatives
numerically. Moreover, the gradients are calculated as the sum of all
pairwise contributions from every unique two-atom pair where one of the atoms
HAS BEEN MOVED. Without getting bogged down in the details, this is a pleasantly
fast procedure because the energy calculations involve rather small matrices --
at most 8 by 8. However, the theory underpinning AMSOL does not allow this
decomposition to be made -- ALL pairwise interactions are affected by the
movement of a single atom, specifically the Coulomb radii used in the
calculation of G(ENP) are changed purely as a result of the new geometry.
Moreover, the movement of any one atom may cause a significant change in both
the density and Fock matrices as a result of charge redistribution, and the
above approach does not permit the Coulomb radii to relax accordingly. It is
thus deficient in two significant respects. To arrive at true stationary
points, gradients must be calculated using the full Fock and density matrices.
This is what 'DERINU' accomplishes. From my perusal of MOPAC 6.0 such an option
is not incorporated into that code (indeed, the AMPAC code makes it clear that
this is regarded as somewhat archaic, but with solvation effects turned on it
can not be avoided).
Thus, structures optimized by the modified MOPAC are NOT true stationary
points, although they may be quite close depending on the molecule in
question. In general, however, molecules which are either highly polarizable
or incorporate atoms with large partial charges will not be treated well by
the pairwise-summation approach.
We are not advocating AMPAC over MOPAC (or vice versa) in any way. It
would not be terribly difficult to modify the MOPAC subroutine DCART.f to
permit the equivalent gradient calculations to be performed. However, the
latest version of MOPAC does not come so equipped, and that is one of the
chief reasons we chose to create AMSOL as opposed to "MOSOL".
Chris Cramer
Date: Mon, 10 Feb 92 17:05:33 GMT
From: James P. Schmidt <jims &$at$& duce.medicine.rochester.edu>
To: CHEMISTRY &$at$& ccl.net
Subject: amsol
Date: Mon, 10 Feb 92 16:58:11 EST
From: kpc23%CAS.BITNET &$at$& OHSTVMA.ACS.OHIO-STATE.EDU (Kevin P. Cross Ext.
3192 Room 2209B)
Subject: Useful Numeric Property Data
To: CHEMISTRY &$at$& ccl.net.bitnet
Status: R
Dear netters:
Chemical Abstracts Service is interested in
the types of numeric data you would like to see
associated with structures and how you would use
such data. This could be data for molecular modeling,
computational chemistry, etc. In particular, we
are interested in useful types of data that could
be calculated for large numbers of structures
that users might find beneficial when searching
in conjuction with structures.
We welcome any and all suggestions!
Sincerely,
~
Kevin P. Cross, Ph.D.
Research
Chemical Abstracts Service
2540 Olentangy River Road
P.O. Box 3012
Columbus, OH 43210
~
kpc23 &$at$& cas.org Internet
kpc23 &$at$& cas Bitnet