[SCIP] Questions on detecting unbounded
Yankai Cao
yankai.cao at gmail.com
Mon Feb 20 00:04:49 CET 2017
Hi, Gregor,
Thanks for your information. I have search the user parameters but cannot find the right parameter.
I am solving a number of (e.g. 1000) NLPs with different parameters in the model. Many cases are unbounded. The following is the output of SCIP for the smallest case:
node | left |LP iter|LP it/n| mem | dualbound | primalbound | gap
1 | 0 | 1 | - | 209k| -- |-1.000000e+05 | Inf
(node 1) LP relaxation is unbounded (LP 1)
1 | 2 | 4 | - | 210k| -- |-1.000000e+05 | Inf
* 2 | 1 | 6 | 5.0 | 211k| -- |-6.800000e+17 | Inf
100 | 99 | 6 | 0.1 | 241k| -- |-6.800000e+17 | Inf
…
node | left |LP iter|LP it/n| mem | dualbound | primalbound | gap
10575k| 9648k| 6 | 0.0 |2501M| -- |-1.000000e+18 | Inf
10575k| 9648k| 6 | 0.0 |2501M| -- |-1.000000e+18 | Inf
As you can see, at node 2, SCIP have already find a prime bound of -6.8e17, which is so small that I can view it as unbounded for my model. I want SCIP at node 2 to stop and return me this prime bound. However, SCIP is still running after exploring more than 10 million nodes. Setting a solution limit of 1 is not appropriate for my problem, because the primal bound in the node 1 is not large enough for me to declare that the problem is unbounded.
Regard,
Yankai
> On Feb 17, 2017, at 10:23 AM, Gregor Hendel <hendel at zib.de> wrote:
>
> Hi Yankai,
>
> please read the documentation of SCIP, especially search the limits in the user parameters
>
> http://scip.zib.de/doc/html/PARAMETERS.php
>
> to find something that matches your purpose. IMHO, there is nothing to speed up the detection of unboundedness. What you should rather do, is set up a proper lower bound for your model, and set a solution limit of 1 to stop if feasibility of this modified model was proven.
>
> Happy scipping,
> Gregor
>
>
>
> Am 17.02.2017 um 17:02 schrieb Yankai Cao:
>> Dear Benjamin,
>>
>>
>> Thanks so much for your reply. Is there any parameter I can set to help SCIP detect if the problem is unbound. For example, for my model, if SCIP find an upper bound <= ub (e.g. -1e10), I want SCIP to stop and declare the problem is unbound. Otherwise, it will takes SCIP much much more time to continue. Thanks.
>>
>>
>> Regards,
>> Yankai
>>
>>
>>> On Feb 17, 2017, at 7:19 AM, Benjamin Müller <benjamin.mueller at zib.de> wrote:
>>>
>>> Dear Yankai,
>>>
>>> you can use
>>>
>>> set/limits/objective
>>>
>>> in order to set an objective limit. SCIP will then only look for solutions that have a better value than the given objective limit. You can think of it as an artificial upper bound for your (minimization) problem.
>>>
>>> As far as I can see it, there is no parameter to set an initial dual bound in SCIP. There is SCIPupdateLocalDualbound in the API that could be called during the SCIP_STAGE_PROBLEM stage.
>>>
>>> Regards,
>>> Benjamin
>>>
>>>
>>> On 02/17/2017 07:24 AM, Yankai Cao wrote:
>>>> Hi, All,
>>>>
>>>> I am new to SCIP. I want to know how to set a parameter ub (e.g. -1e10) so that if SCIP find the upper bound is smaller than ub SCIP can stop and declare that the problem is unbounded?
>>>>
>>>> Also, how to provide known upper/lower bounds before b&b, which might help speed the solution time of SCIP?
>>>>
>>>> Thanks very much!
>>>>
>>>>
>>>> Regards,
>>>> Yankai
>>>>
>>>>
>>>> _______________________________________________
>>>> Scip mailing list
>>>> Scip at zib.de
>>>> http://listserv.zib.de/mailman/listinfo/scip
>>>>
>>> --
>>> ______________________________
>>> Benjamin Müller
>>> Zuse Institute Berlin
>>> Takustr. 7, 14195 Berlin
>>> benjamin.mueller at zib.de
>>> +49 30 841 85-195
>>
>> _______________________________________________
>> Scip mailing list
>> Scip at zib.de
>> http://listserv.zib.de/mailman/listinfo/scip
>
>
> _______________________________________________
> Scip mailing list
> Scip at zib.de
> http://listserv.zib.de/mailman/listinfo/scip
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listserv.zib.de/pipermail/scip/attachments/20170219/c4867c1f/attachment.html>
More information about the Scip
mailing list