[Coin-discuss] osicplex strange behavior, writeMps bugs.

[Matthew Saltzman]

On Thu, 23 Jan 2003 Vivian.DeSmedt at aisystems.be wrote:

> Dear Osi,
>
>
> This mail to report some problems.
>
> first:
>
> In attachemnent you will find a very small test suite.
>
> (See attached file: main.zip)
>
> It pass well for Cpl and Glpk but the last problem fail for Cplex.
> If you look at the differences between problem 3 and problem 4 you will
> notice that I add just an upper bound to the variable wich shouldn't modify
> the answer.
>
> This look like a problem in the cplex osi interface.

I will take a look at this soon.

> second:
>
> It seems to me that there is a bug in the writeMps methode of Cpl and Glpk.
> Both forget to take into account the type of problems (minization or
> maximization) I guess both suppose minimization.
>
> To prove that I generate problem for minimization and maximization and
> nothing change.
>
> This is after all my test just a bug report.
> Tell me what you thing about it.

As I understand it, there is no convention associated with the objective
function in an MPS file.  The objective row is just another row of the
matrix that happens not to have bounds on the value of the slack variable.
It's the user's responsibility to provide separately the identity of the
row that is the objective if there are several such rows (though the
default is commonly to take the first one) and whether the objective is to
be minimized or maximized (the default is solver-dependent, but in OSI it
is minimization).

> Vivian.

-- 
		Matthew Saltzman

Clemson University Math Sciences
mjs at clemson.edu
http://www.math.clemson.edu/~mjs