Thursday, November 17, 9:30 am in room MTH 3206, University of Maryland,
College Park
Semicirculant Preconditioners for First-order Systems of PDEs
Dr. Kurt Otto
Dept. of Computer Science, UMCP
kotto@cs.umd.edu
We consider solving time-dependent and time-independent systems
of first-order PDEs in 2D using CG-like iterative methods.
The systems are preconditioned using semicirculant preconditioners.
Analytical formulas for the eigenvalues are derived for scalar
model problems with correct boundary conditions and constant coefficients.
It is proved that, under certain conditions on the grid, the spectrum of
the time-independent problem asymptotically becomes two finite curve segments,
which are independent of the number of gridpoints. Using the information
on the spectra, the convergence of minimizing iterative methods like GMRES
is analyzed. Based on semicirculant preconditioners we also construct a memory
efficient direct solver for first-order systems of PDEs with proper boundary
conditions and constant coefficients in one space direction. This solver
could be used as a preconditioner for general problems. The convergence
properties are empirically verified both for the model problems,
and for more realistic problems, e.g., the Navier-Stokes equations
for a channel flow.
|