For some cases Nested Dissection (provided in METIS lib) does give better results (lower # non-zeros in LU decomp) (for example the link above), in general tho, colamd provides has better results.
Potentially adding both of colamd from SS and symamd from SS.
Honestly, the renumbering is such a smaller cost in overall cfd sims that running each kind of renumbering alg to find the one with the least fill-in makes sense
Notes: prelim testing via matlab funcs show that colamd is better than symamd even on sym mats. Example tested on: https://www.mathworks.com/help/matlab/math/graphical-representation-of-sparse-matrices.html
For some cases Nested Dissection (provided in METIS lib) does give better results (lower # non-zeros in LU decomp) (for example the link above), in general tho, colamd provides has better results.
Potentially adding both of colamd from SS and symamd from SS.
Honestly, the renumbering is such a smaller cost in overall cfd sims that running each kind of renumbering alg to find the one with the least fill-in makes sense