Re: [Help-glpk] infeasible optimal solution

[Andrew Makhorin]

> The result shows there exists optimal solution, consequently the
> optimal solution is assumed to be feasible.
> However, there are small violations to the feasibility, such as
> that net value of Eth_ex (-3.614e-016), Form_ex (-1.917e-015),
> Lac_ex (-1.593e-015) and dr_gly (-2.002e-015), which should be >= 0.
> Although they are extremely small discrepancies, it is not
> acceptable in my application, absolutely feasible must be
> guaranteed. 

> Is there any method can be adopted to circumvent my problem? 
> Any ideas will be greatly appreciated.

If the solution status is optimal or primal feasible and the reported
value of a (basic) variable violates its lower/upper bound, you can
safely replace it by the corresponding bound, for example:

   x = lpx_get_col_prim(lp, j);
   if (x < 0.0) x = 0.0;

(assuming that the lower bound of variable j is 0).

Another way is enabling the round option:

   lpx_set_int_parm(lp, LPX_K_ROUND, 1);

before calling lpx_get_col_prim. For details see the reference manual.