[Ipopt] Slow convergence for simple problem

[Acharjee Swagato]

Andreas,

Thanks for your suggestion. It worked well. I was also able to add more terms to the objective function and more complex constraints.

Swagato

________________________________
From: Andreas Waechter [mailto:awaechter.iems at gmail.com]
Sent: Wednesday, September 21, 2011 5:30 PM
To: Acharjee Swagato
Cc: ipopt at list.coin-or.org
Subject: Re: [Ipopt] Slow convergence for simple problem

Hi Swagato,

If you do not provide second derivatives to Ipopt, it uses a quasi-Newton approximation, which is not the same as computing the actual second derivatives, and can be much slower.  Since you have the Hessian (C) available, I strongly suggest you write the code to return it to Ipopt.

Andreas

On 09/20/2011 01:56 PM, Acharjee Swagato wrote:
Hi Andreas,

I checked the first derivatives using the derivative checker, there are no errors. I did not code the second derivative, so IPOPT is calculating it by itself. C is positive definite as I can solve the same problem using a quadratic optimizer.

Swagato

________________________________
From: ipopt-bounces at list.coin-or.org<mailto:ipopt-bounces at list.coin-or.org> [mailto:ipopt-bounces at list.coin-or.org] On Behalf Of Andreas Waechter
Sent: Tuesday, September 20, 2011 2:43 PM
To: ipopt at list.coin-or.org<mailto:ipopt at list.coin-or.org>
Subject: Re: [Ipopt] Slow convergence for simple problem

Hi Swagato,

Did you verify that your coded the first and second derivatives correctly, using the derivative checker?  Is C positive definite?  (If not, Ipopt might something take a large number of iterations getting out of a nonconvex region)

Andreas



On 09/20/2011 07:40 AM, Acharjee Swagato wrote:
Hello IPOPT team,

I am a new user of IPOPT. I have a newbie question. I am trying to solve a simple mean variance quadratic optimization problem of the form

min(X) - alpha*X + XCX'

I have some bounds on the variables and no other constraints. I am using the ma27 solver with no hessian approximation.

For a 2500 variable problem I do not see IPOPT converging for 10-15 minutes for 3000+ iterations using default tolerance. As a comparison I can solve the same problem using the quadprog library in R in much less time. I am wondering what I am doing wrong.

Thanks,
Swagato



*********************************************************************************** The information contained herein is not guaranteed. This is not a solicitation of any order to buy or sell. This communication, including any attachments, is for the exclusive use of the intended recipient(s) and/or the intended recipient's designees and may contain proprietary, confidential and/or privileged information. Any use, retention or dissemination by a person other than the intended recipient is strictly prohibited. If you are not the intended recipient or designee, please notify the sender immediately by return e-mail and delete/destroy all copies of this communication. Any opinions, conclusions or information herein not related to the official business of Natixis and any of its affiliates shall be understood as neither given nor endorsed by the company. Natixis and its affiliates reserve the right to review all incoming and outgoing e-mails.


_______________________________________________
Ipopt mailing list
Ipopt at list.coin-or.org<mailto:Ipopt at list.coin-or.org>
http://list.coin-or.org/mailman/listinfo/ipopt


*********************************************************************************** The information contained herein is not guaranteed. This is not a solicitation of any order to buy or sell. This communication, including any attachments, is for the exclusive use of the intended recipient(s) and/or the intended recipient's designees and may contain proprietary, confidential and/or privileged information. Any use, retention or dissemination by a person other than the intended recipient is strictly prohibited. If you are not the intended recipient or designee, please notify the sender immediately by return e-mail and delete/destroy all copies of this communication. Any opinions, conclusions or information herein not related to the official business of Natixis and any of its affiliates shall be understood as neither given nor endorsed by the company. Natixis and its affiliates reserve the right to review all incoming and outgoing e-mails.




***********************************************************************************


The information contained herein is not guaranteed. This is not a solicitation of any order to buy or sell. This communication, including any attachments, is for the exclusive use of the intended recipient(s) and/or the intended recipient's designees and may contain proprietary, confidential and/or privileged information. Any use, retention or dissemination by a person other than the intended recipient is strictly prohibited. If you are not the intended recipient or designee, please notify the sender immediately by return e-mail and delete/destroy all copies of this communication. 

Any opinions, conclusions or information herein not related to the official business of Natixis and any of its affiliates shall be understood as neither given nor endorsed by the company. Natixis and its affiliates reserve the right to review all incoming and outgoing e-mails.



-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20110929/1b845e07/attachment.html>