Re: [Help-glpk] OPB output file anomaly

[Andrew Makhorin]

> Thanks for the explanation on MPS format.  I think I understand
> what you mean by doubling of rows due to objective function/free
> rows.  This does still leave me confused however.  The issue, from my
> perspective, is that I can do the following:

> 1.  Use glpsol with input files a.mod and a.dat (mathprog format) to
> get solution.
> 2.  I can also write out a.mps and a.cplex (and a.opb for that matter)
> using glpsol.
> 3.  Now if I use glpsol on a.mps or a.cplex I get a different
> answer from step #1, and that difference is the constant in the
> objective (which is now missing).

> So, more concretely:

> glpsol --math -m a.mod -d a.dat --wcpxlp a.cplex
> /* and subsequently */
> glpsol --cpxlp a.cplex

> provides different solutions.

> Something just doesn't seem right about that.  Am I still missing
> something fundamental here?

> I looked at the cplex file as you suggested and I see at the end of
> the objective function a comment line:

> \* constant term = -240720 *\

> And that, as I mentioned, is the difference between the solution in
> cplex input mode and the original mathprog mode.  Should the user just
> know to look for this comment in the cplex file to verify his/her
> solution?

> Any further clarification appreciated.  Thanks again for helping me
> understand this situation.

In principle, you may include the statement

   lpx_set_int_parm(LPX_K_MPSOBJ, 1);

immediately before a call to lpx_write_mps or lpx_write_freemps
(for example, in glplpx20.c). Then the objective function row will
always be written in output mps file, including the constant term.

Another, a better way to resolve the issue is to define a dummy
variable fixed at 1. For example:

   var one = 1;
   . . .
   minimize z: ... + 123.456 * one;

where 123.456 is a constant term.