<br><font size=2 face="sans-serif">Christian,</font>
<br>
<br><font size=2 face="sans-serif">Not much work has been done on either
algorithm recently, but some bugs have been fixed and performance seems
reasonable with some caveats.</font>
<br>
<br><font size=2 face="sans-serif">1. With the default Cholesky ordering
and factorization I would guess that a problem of the size you describe
would be fairly slow. You would need to download an alternative open
source ordering and factorization e.g. from University of Florida. With
the new build procedures it should be possible to make this easy, but I
am not an expert in automake etc. See ClpCholeskyUfl.?pp for where
to get code and what defines are needed. If your quadratic objective
is dense then it may still be fairly slow.</font>
<br>
<br><font size=2 face="sans-serif">2. The Simplex algorithm is just
primal. There is also a robust SLP (Sequential Linear Program) method
to obtain an approximate answer to any nonlinear objective function. Often
the best way to solve a problem is to run this first and then the Quadratic
Simplex which often takes zero iterations.</font>
<br>
<br><font size=2 face="sans-serif">3. Presolve with QP is not too
reliable.</font>
<br>
<br><font size=2 face="sans-serif">Having said all that I have just run
qgrow22 from maros test set - 440 rows and 946 columns using standalone
solver</font>
<br>
<br><font size=2 face="sans-serif">clp qgrow22.sif -presolve off -primals
=> QP simplex 7.8 seconds</font>
<br><font size=2 face="sans-serif">clp qgrow22.sif -presolve off -slp 20
-primals => QP simplex after SLP (does 2 QP iterations) 0.53 seconds</font>
<br><font size=2 face="sans-serif">clp qgrow22.sif -presolve off -barrier
=> QP barrier WITH bad ordering 0.26 seconds</font>
<br>
<br><font size=2 face="sans-serif">So see if you need to download a better
ordering.</font>
<br>
<br><font size=2 face="sans-serif">John Forrest</font>
<br>
<br>
<br>
<table width=100%>
<tr valign=top>
<td width=40%><font size=1 face="sans-serif"><b>Christian Kirches <christian.kirches@gmail.com></b>
</font>
<br><font size=1 face="sans-serif">Sent by: coin-lpsolver-bounces@list.coin-or.org</font>
<p><font size=1 face="sans-serif">09/16/2007 01:16 PM</font>
<td width=59%>
<table width=100%>
<tr valign=top>
<td>
<div align=right><font size=1 face="sans-serif">To</font></div>
<td><font size=1 face="sans-serif">coin-lpsolver@list.coin-or.org</font>
<tr valign=top>
<td>
<div align=right><font size=1 face="sans-serif">cc</font></div>
<td>
<tr valign=top>
<td>
<div align=right><font size=1 face="sans-serif">Subject</font></div>
<td><font size=1 face="sans-serif">[Coin-lpsolver] CLP: Question about
current state of QP solving capabilities.</font></table>
<br>
<table>
<tr valign=top>
<td>
<td></table>
<br></table>
<br>
<br>
<br><tt><font size=2>Dear COIN-OR developers,<br>
<br>
I have browsed through CLP's documentation, FAQ, and example programs,
<br>
and was left in doubt a bit about CLP's QP solving capabilities. On the
<br>
one hand, the FAQ entry clearly states that QP solving using the <br>
interior-point code is in an early state of development only, but <br>
nonetheless usually more efficient that the quadratic simplex code. On
<br>
the other hand, that FAQ entry dates back to 2004 and seemingly refers
<br>
to a 0.9x version wheres I find 1.5.0 being the most recent revision.<br>
<br>
I would be extremely grateful if you could elaborate on the current <br>
state of the two algorithms regarding the solution of convex / <br>
semidefinite QPs with a size of ~500 variables and ~1000 linear <br>
constraints ?<br>
<br>
Best regards,<br>
Christian Kirches<br>
<br>
-- <br>
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~<br>
Dipl.-Math. Christian Kirches<br>
<br>
Simulation and Optimization Group<br>
Interdisciplinary Center for Scientific Computing (IWR)<br>
Ruprecht-Karls-University of Heidelberg<br>
<br>
address: Im Neuenheimer Feld 368, D-69120 Heidelberg<br>
e-mail: christian.kirches@iwr.uni-heidelberg.de<br>
phone: +49 6221 54 8895 <br>
room: 414<br>
Private<br>
<br>
address: Stahlbuehlring 143, D-68526 Ladenburg<br>
e-mail: christian.kirches@gmail.com<br>
phone: +49 6203 922 681<br>
mobile: +49 176 21 72 37 22<br>
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~<br>
<br>
_______________________________________________<br>
Coin-lpsolver mailing list<br>
Coin-lpsolver@list.coin-or.org<br>
http://list.coin-or.org/mailman/listinfo/coin-lpsolver<br>
</font></tt>
<br>