From borchers@nmt.edu Wed May 24 19:24:42 2006
Date: Wed, 24 May 2006 16:53:16 -0600
From: Brian Borchers <borchers@nmt.edu>
To: hwolkowi@uwaterloo.ca
Subject: hwolkowi@uwaterloo.ca


I've finally had a chance to sit down and go over the revised version of
"Solving Large-Scale Semidefinite Programs in Parallel"  Unfortunately,
the author did very little to address the concerns expressed in my
original report, and so I cannot recommend the paper for publication.

In particular, 
 
1. I asked the author to provide DIMACS error measures for the solutions 
produced by his code.  This would be extremely easy to do, but the author
has not done it.

2. I asked the author for computational results on problems that are too
large to be easily solved by primal-dual codes on a desktop PC.  The author
has not expanded the test set.

3. I commented that the parallel efficiencies were very poor in some cases
and that the author has not discussed the parallel efficiency of the code.
The author has not addressed this point.  
 
4. The author claims in the abstract that his code "delivers faster
solution times than previously possible" without providing evidence to
support this claim.  The claim remains in the revised paper and the
author's only response is that the claim "isn't entirely unreasonable."
Research would be so much easier if we could simply make any claim that
wasn't entirely unreasonable!