I have a problem with the convergence of spectral matrix factorization: with a tolerance of 1e-8, all my data (H,S,Z) are NaN values. How could you explain that? thank you