Re: [Bug-glpk] Problematic model

[Andrew Makhorin]

> the attached MILP model exits with:

> *  5400:   objval =   6.264244529e+00   infeas =   0.000000000e+00 (0)
> *  5450:   objval =   5.968440326e+00   infeas =   0.000000000e+00 (0)
> OPTIMAL SOLUTION FOUND
> Integer optimization begins...
> +  5450: mip =     not found yet >=              -inf        (1; 0)
+  5481: >>>>>>   6.036226238e+00 >=   5.968440326e+00   1.1% (32; 0)
> GLPK internal error: x <= ub; file glpios03.c, line 265
> Abort

> Maybe it is a scaling problem, but just wanted to let you know in case it 
> has triggered a bug. I have a very similar model which runs without any 
> problems.

Thank you for your bug report.

This bug was already reported earlier; please see:
http://lists.gnu.org/archive/html/bug-glpk/2007-12/msg00004.html
http://lists.gnu.org/archive/html/bug-glpk/2007-12/msg00005.html

In your instance there are free (unbounded) integer variables p[]
and q[] whose values in optimal solution seem to be very large,
from about 10000 to 60000. This makes serious troubles for the mip
solver, because the underlying lp solver cannot provide sufficient
_absolute_ accuracy on computing values of these variables. On the
other hand, you might declare these variables as continuous and then
round-off their values in optimal solution to nearest integers. Due
to their large values the round-off error must not be significant.

p[1]         B        6435.25  ->  6435
p[2]         B        12824.6  -> 12825
p[3]         B        19063.2  -> 19063  etc.
p[4]         B        25118.7
p[5]         B        30932.6
p[6]         B          36449
p[7]         B        41614.7
p[8]         B        46380.1
p[9]         B        50699.1
p[10]        B        54530.3
p[11]        B        57836.7
p[12]        B        60586.5
p[13]        B        62753.2
p[14]        B        64315.9
p[15]        B        65259.5