[Lapack] [lapackers] bug in DGEHRD

[Julien Langou]

On Thu, 22 Apr 2010, James Demmel wrote:

> There is a permutation that makes this matrix 2x2 block triangular,
Mmm, I can not find this permutation ... I tried symamd and colamd and 
they can not either.

> > > Note that while
> > >    [V,D]=eig(A);
> > > does not work,
> > >    [Q,H]=hess(A);
> > >    [V,D]=eig(H);
> > >    V = Q*V;
> > > works ...
Definetley balancing ... I've just remembered that actually you can 
desactivate balancing in eig of matlab, all relative backward errors are 
back to 1e-16 with [V,D]=eig(A,'nobalance');.

Julien.