12 Oct
2018
12 Oct
'18
9:42 p.m.
On 09.10.2018 14:59, Vitaly Magerya wrote:
It appears that I've messed up the formatting and left the useless "matrix cmp" variable in the patch. Sorry about that. Here are the fixed versions.
Thank you, Vitaly! Maybe it is just impossible to avoid unnormal numbers without calling .normal() in these matrix algorithms. Your recent patches make me wonder if the heuristics in matrix::determinant() surrounding variable normal_flag are really a good idea at all. In any case, Laplace elimination does return an unnormal 1 on matrix{{a/(a-b), 1}, {b/(a-b), 1}}, and this is only remedied by the very strange line 795 in matrix.cpp. What do you think? All my best, -richy.