[Ipopt] Ipopt Stochasticity

[Tony Kelman]

Non-repeatability is usually due to parallel BLAS, or occasionally a memory allocation bug (several of which may have been fixed since 3.9.1).

For Anil, does this still happen if you set the linear_solver option to ‘mumps’? On Mac, I think my most recent mex files were using the Apple vecLib BLAS libraries, which I believe are parallel and based on MKL. There might be a way to force vecLib to run single-threaded, but I’m not sure how. You could also build mex files using the netlib BLAS (I think I have a few of those around, though IIRC your problems would occasionally run into exception-handling issues on Mac with those mex files).

For Jochen, what version of BLAS were you using? How are you calling Ipopt, through which interface or via a standalone application, etc?

Try setting output_file and a high value for file_print_level to see exactly where two runs start diverging, and verify that all the initial data is exactly the same like you think it should be.

-Tony


From: Steimel, Jochen 
Sent: Friday, September 13, 2013 7:40 AM
To: ipopt at list.coin-or.org 
Subject: Re: [Ipopt] Ipopt Stochasticity

Dear all,

 

i have encountered a similar problem as Mr. Rao. Even if I call IPOPT in direct succession with no changes in options or starting points the result might vary, ranging from SOLVE_SUCCEEDED to RESTORATION_FAILED. Does anybody have an idea what might cause this behavior? I am using IPOPT 3.9.1 with linear solver MA57 on a Windows Machine with a Intel Core 2-Duo processor. I only encounter this problem in difficult optimizations. If the problem is trivial, it always converges.

 

Regards

Jochen

 

 

Von: ipopt-bounces at list.coin-or.org [mailto:ipopt-bounces at list.coin-or.org] Im Auftrag von Anil Rao
Gesendet: Freitag, 13. September 2013 14:39
An: ipopt at list.coin-or.org
Betreff: [Ipopt] Ipopt Stochasticity

 

All

I have a problem that I have set up using the latest precompiled Matlab mex files (those that Tony Kelman graciously provided). I am having an issue that Ipopt runs a different number of iterations each time the problem is solved. I change nothing between calls to Ipopt, I simply solve the exact same problem repeatedly with the same initial guess. Does anyone have a reason why I may be having the aforementioned issue of non-repeatability?

Anil Rao





-- 
Anil V. Rao
Associate Professor
Department of Mechanical and Aerospace Engineering
University of Florida
Gainesville, FL 32611-6250
Tel:  (352) 672-1529
E-mail:  anilvrao at gmail.com



--------------------------------------------------------------------------------
_______________________________________________
Ipopt mailing list
Ipopt at list.coin-or.org
http://list.coin-or.org/mailman/listinfo/ipopt
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20130913/28f090a4/attachment.html>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: ~WRD000.jpg
Type: image/jpeg
Size: 823 bytes
Desc: not available
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20130913/28f090a4/attachment.jpg>