Basic HTML version of Foils prepared 8 November 1995

Foil 35 Banded Matrix Computational Complexity

From CPS615 Module on Iterative PDE Solvers CPS615 Basic Simulation Track for Computational Science -- Fall Semester 95. by Geoffrey C. Fox


One can show that in a general sparse matrix, it is straightforward to preserver the "end" zeros
000...000X000...000XXX000...000X000...000 shows structure of a row of A in 2D Laplace example
The first and last zeros can be explicitly taken into account by setting range of for loops properly
The inside zeros "fill" i.e. become nonzero during process of Gaussian elimination
Thus if bandwidth B i.e.
000...000X000...000XXX000...000X000...000
Then computational complexity is not N3 as for a full (no zeros accounted) but NB2



© on Tue Oct 28 1997