<br><font size=2 face="sans-serif">Dear Andreas and Tobias,</font>
<br><font size=2 face="sans-serif">thanks for the fast reply.</font>
<br>
<br><font size=2 face="sans-serif">Let me first give you the background to our problem: </font>
<br><font size=2 face="sans-serif">Im running a pilot project where the intention is to optimize the performance and production of a large pulp and paper mill in Sweden. We have thus developed a large set of process models (in Modelica/Dymola), together with a database and an user-interface. We also have a MATLAB based tool for automatically translating the process models to FORTRAN, including first order derivatives. The discretization is done using the trapetzoidal method. We also have an advanced interface for transfering information from the database (which is connected to the millwide information system) to the FORTRAN-based program. This framework would be very difficult to replace or even change and is not considerable.</font>
<br>
<br><font size=2 face="sans-serif">Currently we are solving our optimization problem with SNOPT. As you know, SNOPT allows for the user to separate the NLP into a linear and non-linear part, respectively. This we have implemented as well. However, besides that, SNOPT is claimed not to care of the structure of the Jacobian. We have thus ordered the variables and equations of the discretized DAE system in a variable- and equation-based order, ie. x=[x1(1),x1(2),..,x1(N),x2(1),...xn(N)]. This means that there is not block strucure at all in our Jacobian matrix.</font>
<br>
<br><font size=2 face="sans-serif">However, it is easy for us to change the variable and equation order if it is expected to make any difference and if it is convinient to make changes to the file constr.F. As I mentioned in the previous mail, it would be easy to generate a block-structure of the Jacobian as the one in prof. Bieglers presentation.</font>
<br><font size=2 face="sans-serif">I read abot the option "</font><font size=2 face="Courier New">ISELBAS</font><font size=2 face="sans-serif">" in the IPOPT.README file, but it is not clear to me how the information on the block-structure is specified. Is there some other document descibing the algorithms in constr.F??</font>
<br>
<br><font size=2 face="sans-serif">Best Regards</font>
<br><font size=2 face="sans-serif">Jens</font>
<br>
<br>
<br><font size=1 color=white face="Arial">Message from coin-ipopt-request@www-124.ibm.com@www-124.ibm.com received on 09/06/2003 06:01 PM</font>
<table width=100%>
<tr valign=top>
<td>
<td><font size=1 face="sans-serif">09/06/2003 06:01 PM</font>
<br>
<br>
<td>
<td><font size=1 face="sans-serif"><b>coin-ipopt-request@www-124.ibm.com@www-124.ibm.com</b></font>
<br>
<br><font size=1 color=red face="sans-serif">Please respond to coin-ipopt@www-124.ibm.com</font>
<br><font size=1 color=red face="sans-serif">Sent by coin-ipopt-admin@www-124.ibm.com</font></table>
<br>
<table width=100%>
<tr valign=top>
<td>
<div align=center>
<br></div>
<td><font size=1 face="sans-serif"> To: coin-ipopt@www-124.ibm.com</font>
<br><font size=1 face="sans-serif"> cc: (bcc: Jens Pettersson/SECRC/ABB)</font>
<br><font size=1 face="sans-serif"> Subject: Coin-ipopt digest, Vol 1 #26 - 1 msg</font></table>
<br>
<br>
<br><font size=2 face="Courier New">Send Coin-ipopt mailing list submissions to<br>
coin-ipopt@www-124.ibm.com<br>
<br>
To subscribe or unsubscribe via the World Wide Web, visit<br>
http://www-124.ibm.com/developerworks/oss/mailman/listinfo/coin-ipopt<br>
or, via email, send a message with subject or body 'help' to<br>
coin-ipopt-request@www-124.ibm.com<br>
<br>
You can reach the person managing the list at<br>
coin-ipopt-admin@www-124.ibm.com<br>
<br>
When replying, please edit your Subject line so it is more specific<br>
than "Re: Contents of Coin-ipopt digest..."<br>
<br>
<br>
Today's Topics:<br>
<br>
1. Re: IPOPT and optimal control problem (Andreas Waechter)<br>
<br>
--__--__--<br>
<br>
Message: 1<br>
Date: Fri, 5 Sep 2003 17:56:55 -0400 (EDT)<br>
From: Andreas Waechter <andreasw@watson.ibm.com><br>
To: jens.pettersson@se.abb.com<br>
cc: coin-ipopt mailing list <coin-ipopt@www-126.southbury.usf.ibm.com>,<br>
Larry Biegler <lb01@andrew.cmu.edu><br>
Subject: [Coin-ipopt] Re: IPOPT and optimal control problem<br>
<br>
Dear Jens,<br>
<br>
Many thanks for your message and your interest in Ipopt.<br>
<br>
(I'm sending the message to the Ipopt mailing list, since this is the<br>
channel for questions about the package. If possible, please use this<br>
for your future message.)<br>
<br>
You wrote:<br>
<br>
> I have a few questions regarding IPOPT and the use of it as a solver for<br>
> optimal control problem. In your thesis, as well as in the enclosed<br>
> presentations given by Larry Biegler on p. 16 (Decomposition of<br>
> Large-scale NLP) , it is mentioned that the performance can be increased<br>
> by specifying some variables as independent and the others as dependent.<br>
> This is shown to increase the speed significantly. How can this be<br>
> implemented in the current version of IPOPT?? I have seen in the<br>
> DYNOPT.README file that the file constr.f does this, but for systems of<br>
> DAE solved by orthogonal collection.<br>
><br>
> Can it be done in some other way, for example, we can easily partition our<br>
> optimal control problem according to the figure on p16, but how is that<br>
> information provided to IPOPT??<br>
<br>
The story with the dependent and independent variables is that this allows<br>
one to compute the search directions for the nonlinear optimization<br>
problem (which for an optimal control problem usually includes a<br>
discretized version of the DAE system) in two steps (by solving two<br>
linear systems) instead of one bigger one. We call the first approach the<br>
"reduced space approach" and the other one the "full-space approach".<br>
You can find details on the approached in Section 3.2 in my thesis.<br>
<br>
Which approach is more efficient depends on the particular properties of<br>
the problem that you are solving, in particular on the number of degrees<br>
of freedom (i.e. n-m, where n is the number of variables and m is the<br>
number of equality constraints, assuming that according to the problem<br>
statements that Ipopt accepts, all inequality constraints except for<br>
bounds have been reformulated into equality constraint and slacks).<br>
<br>
The reduced space approach can be more efficient than the full space<br>
approach if the number of degrees of freedom is small (e.g. in the optimal<br>
control problems that we considered), and if one can solve the linear<br>
system involving the basis matrix "C" (where C is defined in Eqn. (2.27)<br>
in my thesis) efficiently. For the dynamic optimization decomposition<br>
(which is implemented in the IPOPT package as DYNOPT), we exploit the<br>
almost block diagonal structure of the basis matrix C - for this we<br>
replace the "general-purpose" version of the file constr.f by an<br>
appropriate one.<br>
<br>
The variable partition for the reduced space approach can also be helpful<br>
if no second derivatives are available, and one wants to use a<br>
reduced-space quasi-Newton method in order to approximate the missing<br>
information (IPOPT options IQUASI=-5,-4,-3,-2,-1,1,2,3,4,5) instead of a<br>
limited memory BFGS approximation (IQUASI=-6,6) for the full space option.<br>
<br>
Note that the input to the DYNOPT version is essentially only the DAE<br>
system, and the DYNOPT package does the discretization for you (as you<br>
said using orthogonal collocation). IPOPT itself is only the nonlinear<br>
optimizer, i.e. if you want to solve an optimal control problem with Ipopt<br>
not using orthogonal collocation, you will have to either pass the large<br>
discretized ODE system as constraints directly to Ipopt (via eval_c and<br>
eval_a), or you will have to add something to the DYNOPT part of the<br>
package.<br>
<br>
I'm not sure which option you are interested in. In order to answer your<br>
question on how to pass the variable parition information to Ipopt let me<br>
assume that you do not want to use the discretization implemented in<br>
DYNOPT and instead provide your discretized version of the DAE system to<br>
IPOPT as constraints. In that case, the parameter ISELBAS (see the<br>
README.IPOPT file) determines how the dependent and independent variables<br>
are to be chosen. In particular, you can list the numbers of the<br>
(in)dependent variables in a file, or choose the first variables as the<br>
(in)dependent ones.<br>
<br>
I hope this helps. Please let me know if this not clear, or if I didn't<br>
understand your question.<br>
<br>
Best regards,<br>
<br>
Andreas<br>
<br>
<br>
<br>
--__--__--<br>
<br>
_______________________________________________<br>
Coin-ipopt mailing list<br>
Coin-ipopt@www-124.ibm.com<br>
http://www-124.ibm.com/developerworks/oss/mailman/listinfo/coin-ipopt<br>
<br>
<br>
End of Coin-ipopt Digest<br>
</font>
<br>
<br>