[Help-glpk] Re: mps format problems

[Andrew Makhorin]

Hi Serge,

> The .lp files is solved by glpsol and cplex qith exactly the same
> answer.  But the .mps file, even though it is solved by glpsol with
> the same solution, is rejected by Cplex, by Cbc (Coin-or) and by  
> Qsopt as INFEASIBLE.  So if everybody (except glpk) says the problem
> is infeasible, I think you will have to admit there is something  
> wrong in the mps file, no?

I am puzzled. As I noticed the primal infeasibility was detected on
solving lp relaxation, however, glpsol found nothing unusual on solving
lp relaxation as well as on solving mip:

------------------------------------------------------------------------
Problem:    ttbcube
Rows:       519
Columns:    551
Non-zeros:  1845
Status:     OPTIMAL
Objective:  R0000001 = 4 (MINimum)

Karush-Kuhn-Tucker optimality conditions:

KKT.PE: max.abs.err. = 8.88e-16 on row 2
        max.rel.err. = 8.88e-16 on row 6
        High quality

KKT.PB: max.abs.err. = 2.22e-16 on row 80
        max.rel.err. = 1.11e-16 on row 80
        High quality

KKT.DE: max.abs.err. = 0.00e+00 on column 0
        max.rel.err. = 0.00e+00 on column 0
        High quality

KKT.DB: max.abs.err. = 0.00e+00 on row 0
        max.rel.err. = 0.00e+00 on row 0
        High quality

End of output
------------------------------------------------------------------------
Problem:    ttbcube
Rows:       519
Columns:    551 (481 integer, 480 binary)
Non-zeros:  1845
Status:     INTEGER OPTIMAL
Objective:  R0000001 = 4 (MINimum) 4 (LP)

Integer feasibility conditions:

INT.PE: max.abs.err. = 2.66e-15 on row 59
        max.rel.err. = 1.78e-15 on row 4
        High quality

INT.PB: max.abs.err. = 1.33e-15 on row 420
        max.rel.err. = 9.99e-16 on column 128
        High quality

End of output
------------------------------------------------------------------------

KKT optimality conditions and integer feasibility conditions are checked
directly on the model data, so the solution is correct in both cases.

The most puzzling thing is that after converting the model from mps
format to cplex lp format the scip solver (available through the NEOS
server) found the optimal solution:

SCIP Status        : problem is solved [optimal solution found]
Solving Time (sec) : 22.69
Solving Nodes      : 926
Primal Bound       : +4.00000000000000e+00 (1 solutions)
Dual Bound         : +4.00000000000000e+00
Gap                : 0.00 %

Please try writing your instance in *free* mps format, i.e.

   glpsol --model ... --data ... --wfreemps foo.mps

and then solving it with cplex. If the infeasibility will disappear,
I will find the cause of the error.


Andrew Makhorin