From: <HIDDEN>
Newsgroups: comp.soft-sys.matlab
Subject: Re: How to solve this non-convex quadratically constrained quadratic programming
Date: Sun, 22 Apr 2012 16:35:09 +0000 (UTC)
Organization: Xoran Technologies
Lines: 18
Message-ID: <jn1bvt$gdq$>
References: <3072521.689.1334707719706.JavaMail.geo-discussion-forums@ynmm9> <jmljp0$jbi$> <16329087.1512.1334757550839.JavaMail.geo-discussion-forums@ynmm10> <jmofmi$q63$> <jmpbbh$n13$> <jmrdm1$oam$>
Reply-To: <HIDDEN>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
X-Trace: 1335112509 16826 (22 Apr 2012 16:35:09 GMT)
NNTP-Posting-Date: Sun, 22 Apr 2012 16:35:09 +0000 (UTC)
X-Newsreader: MATLAB Central Newsreader 1440443
Xref: comp.soft-sys.matlab:765516

"Matt J" wrote in message <jmrdm1$oam$>...
> So, the optimal lambda must satisfy the above polynomial eigenvalue problem. But which solution is desired?  Rearranging the original Lagrange multiplier equation 
> lambda*x =H*x-g
> Premultiplying by x' on both sides and using x'*x=1 again leads to
> lambda =  x'*H*x-g'*x =objective
> So taking the minimum/maximum solution to the polynomial eigenvalue problem is equivalent to minimizing/maximizing the objective. 

I just realized that this part is wrong. The objective function is really

f(x) = 0.5*x'*H*x-g'*x 

So, it's not clear to me why we want to minimize over the admissible lambda. Oh well, I guess the Gander et al paper explains it.