Using svds(A,k)


(Deepayan) #1

Hi,

I’m relatively new to the CVX software. I’m wondering why we cannot use svds(A,k) from MATLAB to compute the sum of the k largest singular values in CVX (and instead need to use the lambda_sum_largest)

Also, how does the math of the sum of the k largest singular values follow from this : lambda_sum_largest([zeros(n),X;X’,zeros(n)],k)

Thanks a lot.


(Mark L. Stone) #2

You can’t just do whatever you want with CVX; you have to follow its rules.

As for your 2nd paragraph, see 19. in p. 148 of “Lectures on Modern Convex Optimizatio”, Aharon Ben-tal and Arkadi Nemirovski https://www2.isye.gatech.edu/~nemirovs/Lect_ModConvOpt.pdf .

Perhaps you’ve read Computing or bounding a modified Ky Fan norm and Represent Schatten-p norm in CVX and https://github.com/nathanieljohnston/QETLAB/blob/master/kpNorm.m ?