Hi, i have a problem where i have to calculate the SVD of typically thousands of small matrices A independently. Each matrix A has 8 rows x 9 columns, maybe i extend it by one row to get 9x9. In my problem, I need the singular vector corresponding to the smallest singular value of A.

Are there any libraries available to do such kind of stuff (many threads, each does the SVD of a small matrix) ? If not, what approach could be good ? It should be relatively easy to implement, even if it’s not the fastest out there - maybe something similar as in http://www.miislita.com/information-retrie…3-full-svd.html ? Power method may be useul, i’m not sure.

Alternatively, is there a way to re-formulate the problem mathematically so that the ‘many SVDS of small matrices’ can be re-cast into 'one SVD of a big matrix ? Advantage would be i could solve it with CULA or so. Either way, i think it wouldn’t make sense as the computational complexity would rise, i suppose.

Thx for any help, Hannes