- Infeasibility introduced by the error of the approximation, most noticeable when significant steps are made in the linearization.
- Infeasibility introduced by the activation of penalty breakers, where it was not otherwise possible to make a meaningful step in the linearization.
The infeasibility induced by the former diminishes as the solution converges, provided mild assumptions regarding the continuity of the functions describing the model are satisfied. The focus of any analysis of infeasibility in XSLP must therefore most often be on the penalty breakers (also called error vectors). For some problems, Xpress-SLP may terminate with a solution which is not sufficiently feasible for use in a desired application. The first controls to use to try to resolve such an issue are
-
XSLP_ECFTOL_A
-
The absolute linearization feasibility tolerance is compared for each constraint in the original, nonlinear problem to its violation by the current solution.
-
XSLP_ECFTOL_R
-
The relative linearization feasibility tolerance is compared for each constraint in the original, nonlinear problem to its violation by the current solution, relative to the maximum absolute value of the positive and negative contributions to the constraint.
