[Ipopt] Memory limitations in Matlab-interface IPOPT when using MA27

[Andreas Waechter]

MC19 is used to computed scaling factors for the matrix.  By default, it 
only kicks in if the linear system solutions are not accurate.  But by 
setting 'linear_scaling_on_demand" to "no", you will make sure the 
scaling happens from the beginning.  See 
http://www.coin-or.org/Ipopt/documentation/node150.html

Andreas Waechter

Associate Professor
Department of Industrial Engineering and Management Sciences
McCormick School of Engineering
Northwestern University
Evanston, IL 60208
USA
http://users.iems.northwestern.edu/~andreasw/

On 08/16/2012 05:40 AM, Jonathan Hogg wrote:
> At a guess its because MA57 will be using scaling by default and MA27 
> will not.
>
> For most problems scaling makes little real difference (except that 
> MA57 runs more slowly), but for others it
> significantly reduces (or eliminates completely) need for pivoting and 
> the related growth in memory requirements.
>
> You can probably test this by disabling scaling in MA57 and seeing if 
> you get similar errors.
>
> Jonathan.
>
> On 16/08/12 08:39, Lemoine, Roberto wrote:
>> Dear IPOPT developers/users:
>>
>> I have a question. I am using the Matlab-interfaced version of IPOPT 
>> v.3.10.0 (run with Matlab R2011b in a computer with Windows 7), 
>> successfully compiled as described in the Interface website:
>>
>> bandersnatch.dyndns.info/~djr/ipopt_compile_steps.pdf
>>
>> My optimization problem is relatively small (1000 degrees of freedom 
>> or less). However, IPOPT solves the problem successfully only if I 
>> use the linear solver MA57 (I have METIS also installed). However, 
>> when I use MA27, IPOPT fails to start iterating, and shows the 
>> failure message:
>>
>> EXIT: Not enough memory
>>
>> This does not happen with MA57 (even MUMPS makes it to start 
>> iterating, but fails after some iterations due to problems in the 
>> restoration phase). I have played around by increasing the parameters 
>> ma27_liw_init_factor, ma27_la_init_factor and ma27_meminc_factor, and 
>> the problem remains.
>>
>> Does anybody have an idea why MA27 could be having this problems, 
>> while MA57 not? Is there a difference in the memory management of the 
>> solvers? Thanks a lot in advance.
>>
>> Best regards,
>>
>>
>> Roberto Lemoine-Nava
>>
>> Lehrstuhl für Systemdynamik und Prozessführung
>> Fakultät Bio- und Chemieingenieurwesen
>> Technische Universität Dortmund
>> Emil-Figge-Str. 70
>> 44221 Dortmund
>>
>> Tel.: +49 (0) 231/755-5171
>>
>>
>>
>> _______________________________________________
>> Ipopt mailing list
>> Ipopt at list.coin-or.org
>> http://list.coin-or.org/mailman/listinfo/ipopt
>