[Help-glpk] Re: mps format problems

[Andrew Makhorin]

> I am experimenting with various solvers and techniques and I have
> relied on glpsol to translate my MathProg models to .lp and .mps  
> files but I am now convinced that the .mps files are not equivalent
> to the MathProg (or even the .lp) files.

I did find that that caused the error. Thank you for your report.

In your mps file there is column C0000151 (corresponding to variable
maxnbs in your model) which is marked as integer, however, it is
missing in the BOUNDS section. By default glpk assigns its lower/upper
bounds to 0 and +inf, resp. On the other hand the mps format
specification (IBM OSL) states that any integer column is considered
as binary (i.e. with upper bound 1) until its upper bound is specified
explicitly in the BOUNDS section (to make it unbounded the user should
specifiy its upper bound as 1e30). To my opinion this is a quirk,
because many mip solvers do not like unbounded integer variables. Of
course, I will introduce an appropriate option in corresponding api
routines and glpsol.

To avoid the error you need to specify both lower and upper bounds
for every integer variable in the model.


Andrew Makhorin