Using other sparse solvers with ordering (AMD, typically), we find that the ordering process has a big affect on performance, but takes some time.

For our application, the matrix pattern does not change much between iterations. We've found that we can re-use the permutation vector, avoiding ordering at each iteration, and achieve noticeable increases in speed. Whenever the matrix size changes, or after 'N' iterations, we re-order and continue with a new permutation vector.

Can this be done in Eigen?

thanks in advance

Kevin

Sure, if the pattern does not change, you can call factorize() instead of compute() for the other runs, see the second example there: http://eigen.tuxfamily.org/dox-devel/gr ... tml#title4

thanks, again.

Kevin