[Ipopt] What shall I do when "Not enough memory" is encountered

[Jonathan Hogg]

Weifeng,

If methods to reduce the memory required by MA27 are insufficient, we 
have an experimental interface for our out-of-core solver HSL_MA77 that 
uses disk rather than memory to store the factors, but will be 
considerably slower than using an in-core solver.

If this is of interest please contact me directly and I can give you 
more detailed instructions (as an academic you will qualify for the 
academic licence to our more recent codes).

When I get time (hopefully after Easter) I'll pull together various bug 
fixes and performance improvements before making the interface more 
widely available via the Ipopt svn tree.

If the memory issues are due to poor scaling, using MA57 rather than 
MA27 as the linear solver may help as the current Ipopt default is to 
use a matching based scaling internal to the linear solver.

Jonathan Hogg
STFC

PS If you have any problems that are large enough to tax MA27/MA57 
inside IPOPT, we would love to get access to them (under an NDA if 
necessary) so we can improve our linear solvers.

Andreas Waechter wrote:
> Hi Weifeng,
> 
> Please send emails regarding Ipopt to the Ipopt mailing list.
> 
> As for your question, you might want to play with the options
> 
> # ma27_liw_init_factor:
> # ma27_la_init_factor:
> # ma27_meminc_factor:
> 
> Also, choosing smaller pivot tolerances might help (the default is rather 
> small already though).  Furthermore, making sure that your problem is 
> well-scaled and well-posed will probably reduce the fill-in in the linear 
> solver.
> 
> Regards,
> 
> Andreas
> 
> On Thu, 1 Apr 2010, Weifeng Chen wrote:
> 
>> Hi Andreas,
>>
>> I am a doctoral student of Department of Control Science and Engineering of
>> Zhejiang University.
>>
>> Recently, I used IPOPT (Version 3.8.1) compiled with MS VS 2008 to solve a
>> large-scale optimization problem which was establish with Aspen Plus.
>> Unfortunately, IPOPT exited with a message, "Not enough memory". The linear
>> solver utilized was MA27 and the second derivative information was
>> approximated by means of BFGS. How shall I handle this problem? Looking
>> forward to your reply! Thank you very much!
>>
>> Information on the optimization problem
>> Number of variables:
>> 106375
>> Number of equality constraints:                                    106374
>> Number of non-zero elements in constraint Jacobian:     1492803
>>
>> The computation was carried out on the machine SONY VGN-SR58 running Windows
>> XP and with Intel(R) Core (TM) 2 Duo CPU P8800 and 4.0G RAM memory.
>>
>> Best regards,
>>
>> Weifeng Chen
>> Department of Control Science and Engineering
>> Zhejiang University, Hangzhou 310027
>> People's Republic of China
>>
> 
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/ipopt