# standard practice for large dense truncated svd computations?

What are the standard methods of computing the rank-k truncated SVD of large dense matrices? My literature search yields results only for large sparse matrices.

I assume for k small that you use a Krylov subspace method (this is what Matlab's svds does). But (empirically) how large can k get before these methods become impractical, and then what should one resort to?

-
Not sure why this had two downvotes... –  Yemon Choi Mar 15 '13 at 7:12
Would this question not be more appropriate for scicomp.stackexchange.com? –  guest Sep 11 '14 at 22:07

## 1 Answer

When dense matrix is so large that Krylov-based techniques become impractical, one have to rely on certain specific of a matrix to compute its low-rank approximation. Because different matrices have different properties, I doubt there are any standard approaches, which work as a silver bullet.

However, you can try to compute a low-rank approximation of your matrix by a cross interpolation: http://dx.doi.org/10.1007/s006070070031 http://www.researchgate.net/publication/251735015_How_to_find_a_good_submatrix/file/e0b49527254918b220.pdf

-