decomposition

[lichtkind]

Not only because panda doesnt work here i want to make a branch without AttrX::lazy and maybe the travis issue also resolve here , but before i want to finalize at least the decomposition section. i wan tto add QR (householder and givens) but due LU I reinvestigated crout and the only difference between jordan and crout LU is reall that crout seems to output (Lu dot D) aka L and Ru and the normal case is Lu and (D dot Ru) aka R. u stands here for unipotent. So there is potention for unification und to have one LU method since crout is only a differnt way to compute and write it doent doesnt change result.

next plan is also new-vandermonde and then finally somehing to iterate eigenvalues.

[lichtkind]

i asked my prof, the algorithm you implemented is wrong he says. we should doublecheck but if so, your mehtod would have to go :) since all the other cases are covered by what i wrote

[pierre-vigier]

I copied it from wikipedia
https://en.wikipedia.org/wiki/Crout_matrix_decomposition

On Fri, Jan 29, 2016 at 5:38 PM Herbert Breunung [email protected]
wrote:

i asked my prof, the algorithm you implemented is wrong he says. we should
doublecheck but if so, your mehtod would have to go :) since all the other
cases are covered by what i wrote

—
Reply to this email directly or view it on GitHub
#29 (comment)
.

[pierre-vigier]

Also, could we know where it is wrong? Or at least have one case where it does not work?

May i know why a version without AttrX::Lazy? By the way, Panda is workign again

[lichtkind]

panda never worked here, and today was exam but I will go Tuesday in library and check the details

[lichtkind]

well all i said is written there too: https://en.wikipedia.org/wiki/Crout_matrix_decomposition sounds like my prof is wrong i will write him too again.

[lichtkind]

i will check it in the library and then write here

[lichtkind]

i was now in library and will again because the 4 books i checked doesn't mention crout. so I wound why having him in the name if the result is the same anyway. I think it is better to let the user specify what he wants and to introduce an option if he wants his U matrix unipotent but L not.

btw unipotent would be candidate for another tester method.

[lichtkind]

i just let it open to remind me to check the workings of all decompositions

[lichtkind]

singular value decomposition will be important too

[lichtkind]

cholesky is proven to be right

[lichtkind]

SVD: singular value decompost and QR are missing