Re: high-precision eigenvalue solver


Hans Mittelmann wrote:

> if you need the eigenvalue closest to zero try inverse iteration. You
> need a linear system solver for your matrix then.

I'm not sure, but I got the impression that what he needed was not
a small eigenvalue, but the small difference of two non-small
eigenvalues.
It's why I suggested reformulation.

Reformulation might be necessary anyway.
If the matrix entry errors are around 10**-16
and he needs an eigenvalue around 10**-16
he is out of luck unless the errors are correlated.
Correlated errors suggest the entries in the matrix
should be formulas instead of the results of formulas.
I wonder how big the matrix is.

.