Re: [Bug-glpk] GLPK 4.46 fails to find solution with enlight13.mps

[Andrew Makhorin]

> I tried to solve the problem “enlight13” of the MIPLIB 2010 set
> ( http://miplib.zib.de/ ). The outcome is below. The issue is that an
> integral solution indeed exists and other solves can apparently find
> it. Using cuts doesn’t seem to help the situation.
> 
> // OUTPUT --------------------
> 
>  
> 
> GLPSOL: GLPK LP/MIP Solver, v4.46
> 
> Parameter(s) specified in the command line:
> 
> enlight13.mps

> [...]

> +5506999: mip =     not found yet >=     tree is empty        (0;
> 297291)
> 
> PROBLEM HAS NO INTEGER FEASIBLE SOLUTION
> 
> Time used:   532.4 secs
> 
> Memory used: 35.7 Mb (37408391 bytes)
> 
> 

Thank you for your report.

Glpsol reports an incorrect result, because the input mps file is badly
formed. Namely, there are some lines in the BOUNDS section:

 LI bnd               y#2#2             0
 LI bnd               y#2#3             0
 LI bnd               y#2#4             0
etc.

which set lower bounds to 0 while upper bounds for corresponding
variables remain unset. Since these variables are marked as integer,
their upper bounds are set to 1, i.e. all such variables are considered
as binary rather than general integer variables having no upper bound.
You may fix this error by changing LI to PL in the mps file. For more
details about the mps format please see the glpk reference manual
included in the distribution. Note also that most mip instances from
miplib 2010 are hard for glpk and solving them to optimlity may take too
long time.


Andrew Makhorin