Skip to content
Prev 206553 / 398503 Next

optimization problem

Dear Hans,

I agree with your comments.  My intuition was that the quadratic form would be better behaved  than the radical form (less nonlinear!?).  So, I was "hoping" to see a change in behavior when the cost function was altered from a radical (i.e. sqrt) form to quadratic, but I was still surprised to actually see it.  I don't have a good explanation for this.  

I came up with the idea of solving Klaus' problem as an LSAP problem.  I did not know that the LSAP approach to this and similar problems has already been considered by Nick Higham.  I asked Nick last night about this problem thinking that he might know of more direct solutions to this problem (e.g. some kind of SVD or related factorizations). He said that I should take a look at the PhD thesis of one of his students.

Take a look at Section 3.5 of the PhD thesis

   Parallel Solution of SVD-Related Problems, With Applications

at

http://www.maths.manchester.ac.uk/~higham/misc/past-students.php


This thesis proposed algorithms for the current problem and different versions of it under the heading of "Procrustes-type" problems.  I have to read this and get a better handle on it.  However, I would not be able to get to this for another two weeks.  If you have any insights from this, in the meanwhile, do share with us.

Best regards,
Ravi.

____________________________________________________________________

Ravi Varadhan, Ph.D.
Assistant Professor,
Division of Geriatric Medicine and Gerontology
School of Medicine
Johns Hopkins University

Ph. (410) 502-2619
email: rvaradhan at jhmi.edu


----- Original Message -----
From: "Hans W. Borchers" <hwborchers at googlemail.com>
Date: Sunday, January 17, 2010 3:54 am
Subject: Re: [R] optimization problem
To: r-help at stat.math.ethz.ch