[Scip] soplex unable to parse LP file generated by zimpl

[George Nychis]

Hi Gerald and Stefan,

Thanks a bunch for the quick response!  You are correct, it is not a true
LP. The reason I tried to run it through a different solver is that scip is
not finding the optimal solution, one that I don't think is "hard" to
intellectual find, but may for some reason be hard to solve for?  It has
solved 10/10 tests I created for it (some of which are harder), but 1 of
these tests it does not find the optimal solution.

In my optimization, I am trying to find the optimal wireless frequency for
each network.  I give it the possible set of frequencies for each radio:
  set FR[R]  :=
    <1> { 2412 },
    <2> { 2412 },
    <3> { 2437 },
    <4> { 2437 },
    <5> { 2412, 2437 },
    <6> { 2412, 2437 };

By putting {2412,2437} for <5> and <6>, I allow for the optimization to
choose between those two frequencies for those two radios (that belong to
the same network -- they must end up on the same frequency).  When solving
through scip, it will choose 2412 for these radios which produces an
objective value of 2.40163428660051.

If I force the two radios to 2437 by removing 2412 from both's set (such
that there is no choice in the optimization problem), then the objective
value of 2.73735205292492 will be produced because the radios receive less
interference on 2437.

Intellectually, this is not hard for me to solve in my head.  Of course,
with 387 and 794 constraints, I know that what goes on in my head is not
what happens in the optimization.  But, the only "choices" in the
optimization are these frequencies and there are only 2.  So, I'm not sure
why it misses the optimal solution.

Does anyone have any suggestions when running through scip to try and find
the optimal solution?

I have included the ZIMPL file with a matching data.zpl that forces the
optimal solution.  You can change FR[R] in data.zpl to match what I pasted
in this e-mail and if you run it through scip it will find the non-optimal
solution.  Thanks for any help. This is my first attempt at mixed integer
programming, so this is all extremely new to me, and I appreciate all the
work you've all done the optimization suite.

- George




On Fri, Jul 26, 2013 at 4:34 AM, Stefan Vigerske
<stefan at math.hu-berlin.de>wrote:

> Hi,
>
> your model is not an LP, but a quadratic program. SCIP can handle this,
> but SoPlex cannot.
>
> Stefan
>
>
> On 07/26/2013 10:13 AM, George Nychis wrote:
>
>> I am trying to run an LP file through soplex that was generated by zimpl.
>>   However, when I try to I get:
>>
>> Loading LP file spectrum_optimization.lp
>> ELPFRD15 Syntax error in line 205
>> error while reading file "spectrum_optimization.lp"
>>
>> This points to the following lines:
>>
>>   airtime_eq_re at 19:
>>    - RadioAirtime#1 + GoodAirtime#1
>>    + [ + RadioAirtime#1 * RadioLossRate#1 ]  <--- line 205
>>   = 0
>>
>> I've attached the zmpl files and the lp output.  I am able to run this
>> through scip using both of these commands:
>> scip -f spectrum_optimization.zpl
>> scip -f spectrum_optimization.lp
>>
>> So, something seems to be wrong when parsing with soplex.   I'd greatly
>> appreciate any help.
>>
>> Thanks!
>> George
>>
>>
>>
>> ______________________________**_________________
>> Scip mailing list
>> Scip at zib.de
>> http://listserv.zib.de/**mailman/listinfo/scip<http://listserv.zib.de/mailman/listinfo/scip>
>>
>>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://listserv.zib.de/pipermail/scip/attachments/20130726/0d161c6f/attachment.html
-------------- next part --------------
A non-text attachment was scrubbed...
Name: forcesolve_data.zip
Type: application/zip
Size: 5777 bytes
Desc: not available
Url : http://listserv.zib.de/pipermail/scip/attachments/20130726/0d161c6f/forcesolve_data.zip