Re: eig function project

[Ed Meyer]

On Mon, Apr 22, 2013 at 4:48 PM, Rafael Gonzalez <address@hidden> wrote:

> This way if the lapack folks
> don't fix *geev at least octave could pass the balancing counterexample given in
> the ML doc

A comment I forgot to include in the last mail, I'd say that the 'nobalance' option is in fact the workaround that Matlab came up with to avoid that feature from dgeev but it is still left to the user criteria. So, a Matlab like functionality is a must, but I was thinking that maybe including an 'auto' option that determines if balancing is required or not, could be a nice feature and a good GSoC project that would require a lot of understanding on when balancing is required.

Rafael González

Searching through old emails I came across one from one of the lapack
developers talking about the balance issue and how they were going to
try to come up with a better strategy. Unfortunately I lost contact with him
over the years so I don't know what ever became of that effort. I see that
he (Zhaojun Bai) is still teaching at UC Davis so I'll email him and see what
I can learn. Here is the old email:

> Message-Id: <address@hidden>
> Date: Fri, 7 Dec 2001 14:22:43 -0800 (PST)
> From: Zhaojun Bai <address@hidden>
> Reply-To: Zhaojun Bai <address@hidden>
> Subject: Re: xGGEVX bug
> To: address@hidden
> Cc: address@hidden
> MIME-Version: 1.0
> Content-Type: TEXT/plain; charset=us-ascii
> Content-MD5: rWD8Tjg2tbM33Qim12idJg==
> X-Mailer: dtmail 1.3.0 @(#)CDE Version 1.3.5 SunOS 5.7 sun4u sparc
> Content-Length: 2030
>
>
> Thank you for your report. It will be taken care in the next
> release of LAPACK.
>
> On the other hand, if I remember corrected, back to many years ago
> (5 years?),  you also helped us to deal with issues related to
> balancing, and matrix pencils with possible common null space.
> If this is correct, will you be interested to work with us
> to continue, we have a plan to deal with this issue, and more generally,
> on some cases of singular pencils. One method we are looking into is
> the Fix-Heiberger method and its improvement.
>
> I will be happy to further communicate with you on this.
>
> Sincerely, Zhaojun Bai


-- 
Ed Meyer