Aha! I think I may have found a better way to do it using QR decomposition which I'm implementing anyway. And it works for general n by n matrices if I understand so this is good. Just a matter of seeing what will be faster now...

http://math.fullerton.edu/mathews/n2003/...odMod.html

i think using QR only will be slower. as it says on the page:

Quote:it takes many iterations and becomes time consuming.

but the good news is that it can be sped up and then it is probably the best method suiting your needs. one trick is to try and transform the original matrix to get as many zeros as possible. then the matrix mutiplications will be done faster.

Yeah, I'm thinking though for general case it is what I'll use. But for the small 3 by 3 thing, I'll just be doing it the normal way.