Wilson Factorization not converged

[jxrjxrjxr]

Hello, I'm currently using your code to infer the time-frequency domain Granger causality among two time series. I use the wavelet transform to get the spectral matrix as your code 'Example.ipynb' did, but the Wilson factorization process is not converged. However, if I only retain the real part of the spectral matrix, then the Wilson factorization is converged. So I wonder under which condition will the Wilson factorization be converged? Is there a criterion to judge that? Looking for your reply! Thank you.