Thursday,
March 24, 2005, 12:15 p.m. GWC 604
Numerical approximation of nonsmooth solutions with spectral methods
Abstract
Spectral methods have been proved to be powerful computational methods
solving a wide range of physical and engineering problems.
As the spectral method is global, it provides the so-called spectral
convergence if the functions of interest are globally smooth. If they are
discontinuous, however, spectral accuracy is deteriorated and the overall
convergence is only O(1) in the maximum norm. Several reconstruction
methods have been developed to deal with this problem. These methods roughly
fall into two different theories: projection theory and direct-inverse theory.
Projection methods seek the reconstruction by means of projection of the
given spectral data into the new spectral space while the inverse methods
use the projection of the reconstruction onto the given spectral space
instead. In this talk, we will briefly explain these theories and discuss
the spectral filtering methods as one of the projection methods. Then we
introduce a new inverse method. The proposed method seeks a reconstruction
in a polynomial space by making the residue of the given spectral data and
the reconstruction orthogonal to the polynomial space. This method referred as
to the "inverse polynomial reconstruction method is unique, exact and
spectrally convergent. As the inverse method involves a matrix inversion
and it is known to be ill-posed, the maximum error grows exponentially
once the round-off errors become dominant. We show that there exist
polynomial basis sets which do not exhibit such exponential growth.
One such basis set is the Jacobi polynomials.
We also provide some numerical results including 1) the spectral simulation
of highly supersonic reactive cavity flows of scramjet engine and
2) the image reconstructions based on the wiggly Fourier images.
For further information please contact:
mittelmann@asu.edu