Numerical Linear Algebra (NUI Galway 2013/2014)
Here is the sparsity pattern for a matrix arising from the
numerical solution to a partial differential equation using a sparse
grid finite element method (joint work with my Ph.D. student,
Stephen Russell).
The image on the left is based on using a "natural" lexicographic
ordering for the nodes in the mesh. If this matrix is factored using
a standard Cholskey method, then many new non-zero entries in
creased in the factors.
One can reduce the this number by first permuting the rows and
columns of the matrix. The second image from the left is permuted
using Matlab's implementation of the approximate minimum degree
permutation. The third uses the column approximate minimum
degree algorithm of Larimore and Davis. The fourth uses
METIS. The
sparsity patterns of the corresponding factors are shown below.