Re: [Bug-glpk] Bug in GLPK (Maybe)

[Andrew Makhorin]

> Please see attached file.
> 
> 
> I have defined a mixed integer program in Lingo and solved it. Lingo
> reports that it finds an optimal solution. I have used Excel to verify
> that the solution is, in fact, feasible.
> I then used Lingo to export the model as an MPS file which GLPK can
> read and solve.
> 
> 
> When I ask GLPSOL to solve the model in the MPS file, GLPSOL tells me
> that there exists no feasible solution.
> I'm not sure if this bug is in GLPK or if Lingo has exported a bad MPS
> file, however I have been using the MPS export from Lingo quite a lot
> and haven't had any problems up until now!
> 
> 
> I am using GLPK Lab for Windows, running GLPSOL v4.43
> 

The standard says that "for integer variables, the default bounds on
columns are 0 and 1". The mps file you provided does not explicitly
specify that integer variables have no upper bound (INTORG/INTEND only
mark variables as integer), so all such variables are considered as
binary that leads to incorrect results.

For more details about the mps format please see:
http://www.cenapad.unicamp.br/parque/manuais/OSL/oslweb/features/featur11.htm
in particular, the topic "BOUNDS section".

I added a bound section into your mps file, in which case glpsol could
find an optimal solution (see patched.* in the zip attached).

I'd suggest you to specify some reasonable upper bounds for all integer
variables in your model. Thanks.

BugSomewhere.zip
Description: Zip archive