On Tue, Apr 17, 2012 at 20:17, Barry Smith <[email protected]> wrote:

The truth is likely that except for CG most of the methods were coded without regard to complex and are likely wrong but not necessarily wrong in a consistent way or even self-consistently wrong.

This fixes Alexander Grayver's reported problem with TFQMR. I still don't know if CG is correct, but it should be easy to check.