Open fredrik-johansson opened 1 month ago
Two related problem which I'm too lazy to create a new issue for right now: gr_mat_diagonalization
currently computes the R matrix by linear solving, then inverts R to obtain L. It is probably better (in terms of numerical stability or intermediate expression swell, where applicable) to do an independent linear solve for L. Also, the calculations can be simplified when all eigenvalues have multiplicity 1.
Currently
gr_mat_diagonalization
can be overridden by rings butgr_mat_eigenvalues
just has a default algorithm.