Advanced Software for the Calculation of
Thermochemistry, Kinetics, and Dynamics
Argonne PI: Al Wagner (CHM)
Argonne Collaborating Staff: Stephen Gray (CHM), Ron Shepard
(CHM), and Mike Minkoff
(MCS).
The Challenge: There are two challenges this project
addresses.
(1) The
computationally intensive core of computed rate constants is the Cumulative
Reaction Probability (CRP). Almost
all reaction rate calculations involve statistical approximations to the
CRP. The approximations have been tested on very narrow classes of
reactions (primarily those involving atoms reacting with diatomic molecules)
because of the computational difficulty of computing the CRP for reactions with
larger reactants. We are developing and parallelizing methods for
calculating the CRP so as to permit a much wider variety of tests of
approximate reaction rate theories.
(2) In developing new
methods for the CRP, efficient parallel eigensolvers and linear solvers are
needed. The same can also be said for
developing better parallel algorithms in other areas of chemical theory, in
particular, electronic structure calculations.
We have developed a new parallel iterative method for eigensolvers and
linear solvers called the Subspace
Projected Approximate Matrix (SPAM)
method. This subspace method allows the
physical insight of the application scientist to be mapped into the numerics
used to solve the application problem.
SPAM is very general and we are developing and testing its application
to chemistry problems such as those requiring CRP or electronic structure
calculations.
Argonne’s
Role: Both
projects above are built on the past efforts of the Chemical Dynamics Program
in the Chemistry Division and on a long term collaboration with computational
scientists in the Mathematics and Computer Science Division. In addition, both projects are connected to
the SciDAC TOPS project
that MCS is involved in.
The CRP project is based in part on the PETSc library of common
mathematical kernels
that will migrate into the TOPS library.
The SPAM project is being
carried out in part as a component of the TOPS development of eigensolvers.
At the SciDAC kickoff meeting of January 15-16, 2002, an overview (ppt) (pdf) of this programm was presented.
There are publications that pertinent to this
project and an ftp site. These can all
be accessed under either the CRP or the SPAM section of this web site.
A paper (pdf)
describing the dispersion fitted finite-difference method.
Related Presentations and Publications
M. Minkoff, R. Shepard, A. Wagner, “Application and the Extension
of the Subspace Projected Approximate Matrix (SPAM) Modification of the
Davidson Method,” Presentation
(ppt) (pdf) at the 2002 Copper Mountain Conference on
Iterative Methods, Copper Mountain, CO, March, 25-29, 2002.
Date: November 26, 2002