Numerical precision problems

[DavideTisi]

Hi, I want to spend two words on a possible source of problems in the multivariate cepstral analysis. Using the multivariate methods, if you (wrongly) decorrelate from two linearly dependent currents, in some cases, it is possible that the code does not issue an error, and returns a result that seems reasonable, even though it is, of course, wrong. The reason is that if the two currents are linearly dependent theoretically, but not exactly numerically, the resulting determinant of the covariate matrix is small but not exactly zero.

[lorisercole]

Hi @DavideTisi! This is an old issue, but it may still be relevant in SporTran 1.0 since the core modules have not changed. I understand what happens, but it is quite hard to reproduce it. If you happen to have some data showing this problem, it would be very useful to draft a solution, or at least a check to be performed in such cases. Thanks