From: "Bruno Luong" <b.luong@fogale.findmycountry>
Newsgroups: comp.soft-sys.matlab
Subject: Re: Finding the nearest matrix with real eigenvalues
Date: Fri, 3 Sep 2010 11:12:07 +0000 (UTC)
Organization: FOGALE nanotech
Lines: 17
Message-ID: <i5ql67$sp0$>
References: <i5j67v$4s4$> <i5jicf$29u$> <i5jn7k$pdd$> <i5jvad$c2b$> <i5p6tb$f0b$> <i5p8sn$mid$> <i5qi10$6oo$>
Reply-To: "Bruno Luong" <b.luong@fogale.findmycountry>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
X-Trace: 1283512327 29472 (3 Sep 2010 11:12:07 GMT)
NNTP-Posting-Date: Fri, 3 Sep 2010 11:12:07 +0000 (UTC)
X-Newsreader: MATLAB Central Newsreader 390839
Xref: comp.soft-sys.matlab:667402

"Matt J " <mattjacREMOVE@THISieee.spam> wrote in message <i5qi10$6oo$>...
> "Bruno Luong" <b.luong@fogale.findmycountry> wrote in message <i5p8sn$mid$>...

> Bruno, although there are "floating" things in this norm, i.e., the scaling of the eigenvalues, the solution is invariant to them. 

But that's even more disturbing. There is a family of "norms" that depends on scaling and the input, whereas the solution does not depend on the scaling. This solution cannot be *characterized* as the minimum of a "clean" cost function. The minimum of the norm is a property. 

I'll caricature here, but I could also tell:

B = V'*real(D)*V, solution proposed by Roger, minimizes the cost function 

norm(Ap-B, 'fro') where Ap:=V'*real(D)*V, and [V D]=eig(A).

That claims of course does not have any intrinsic value and it's certainly true. In a way you are doing more or less like this, you define a norm that is floating.