Re: [eigen] SVD with singular matrices

[ Thread Index | Date Index | More Archives ]

2010/9/30 Cyrille Berger <cberger@xxxxxxxxxxx>:
> Why not call the second one FastSVD (when fixed) ? Then people will have an
> incitive to look at the documentation before using it.

Half of common matrix decompositions are unsafe/inaccurate in the same
way. For example, HouseholderQR is exactly in the same way; same for
Tridiagonalization (hence selfadjointeigensolver) and Hessenberg
decomposition (hence general eigensolver).

This is not Eigen-specific, it's the algorithms themselves. The SVD
bug we're discussing in this thread is Eigen-specific, but that's a
different thing.

Thus, as sad as it is to say, most matrix algorithms are unsafe.
Safeness is the exception (like Full-pivoting or two-sided Jacobi
algorithms), not the rule...


> --
> Cyrille Berger

Mail converted by MHonArc 2.6.19+