Possible Duplicate:
How to reorder variables to produce a banded matrix of minimum bandwidth?
What is the time and space complexity of direct sparse solvers (e.g., UMFPACK, SUPERLU, PARDISO, etc.) for finite element matrices in 2D and 3D? How does the complexity splits into the creation of the inverse and the forward/backward substitution? Can anybody make some notes on this topic and/or post here some references?