On Thursday, March 5, 2015 at 4:58:10 PM UTC-5, Sheehan Olver wrote:
>
>
> Hmm, maybe I’m posing the wrong problem then… I wanted a fast way to
> calculate the null space of a sparse matrix, where the basis spanning the
> null space is also sparse. And the dimension of the vector space is in the
> millions.
>
In that case, why not use:
min ||Lx||_1 subject to ||x||_1= 1 (or ||x||_\infty = 1)?
That can be transformed into an LP. (As I understand it, the only reason
for your norm constraint on x is to prevent the trivial x=0 solution, but
won't any norm work for that?)
