Trick to finding A*A^T without wasting memory?

[mikeph]

Hi,

I have a rather large matrix A, and I need to find the eigenvalues of A*A^T. The problem is, when I type this in MATLAB I get a memory error. I am physically limited to 3GB so I'm now searching for mathematical tricks to find this without performing the full calculation.

I've successfully re-arranged similar things before which removed the memory error (I think as soon as MATLAB sees A' it creates a second matrix the exact same size as A, doubling memory use). This time I am a little stuck.

Any hints?

Thanks,
Mike

 

[caffenta]

Have you tried converting the matrix to single-precision floats? It will cut memory usage in half: A=single(A);

 

[D H]

Have you tried singular value decomposition? Matlab does that with the function svd(X). It also supplies an "economy size" singular value decomposition via svd(X,0).