Re: [eigen] Computing left nullspace of A when rows > cols

[Keir Mierle <mierle@xxxxxxxxx>]

Just to second a request for this: Many applications use the SVD to
form a basis for the nullspace of A, where A is 'wide'; this is the
same here as the left nullspace of a 'tall' matrix.

In libmv, I solve this problem by padding wide matrices with zeros on
the bottom until they are square, then pass to the SVD. Suboptimal,
but it works well enough for now.

I will see about the SVD later this week.

Keir

On Fri, Jan 30, 2009 at 8:34 AM, W. James MacLean
<james.maclean@xxxxxxxxxxx> wrote:
> Hi,
>
> I would like to use Eigen to compute the left nullspace of a matrix A where
> A is n by m, and n > m. Assuming we denote the decomposition as A = U S
> V^*, then (currently) the U returned by Eigen is n by m, and the columns
> corresponding to the left nullspace are missing.
>
> Is it possible to either (1) compute these easily using some other facility
> already in Eigen, or (2) modify Eigen so it outputs the full n by n U
> matrix?
>
> I know that, generally, I could do this by transposing A and computing the
> right nullspace, but then cols > rows and Eigen doesn't seem to support
> this for SVD at present.
>
> Cheers,
>
> James
>
> PS- I *really really* like what I've seen of Eigen so far - way cool!
>
>
>
>
>