Re: [Bug-glpk] Possible GLPK - Sensitivity Report Bug

[Andrew Makhorin]

> I'm not sure whether if it's an already reported bug (or maybe it's
> not a bug either), I'm teaching GLPK as a part of my linear
> programming subject, as a part of our discussion we need to get the
> sensitivity reports, they show the right values for most parts but
> fail to show the right range values for certain things, specifically,

> * when there is slack/surplus within a constraint, it returns an
> activity range not including the initial value, so is misleading for
> inexperienced students as the range should actually be between the
> initial value and the current level of activity, objective value at
> break points also returns misleading values based on this, they should
> be equal to the initial optimal value,

> * for a nonbasic variable the objective value at break points should
> be equal to the initial optimal value, however it returns two other
> values, again misleading for inexperienced students,

> * and as a suggestion for improvement is it possible to have no
> activity ranges for columns, and no objective coefficient ranges for
> rows, again it can be quite misleading for inexperienced students,
> 

> I'm not sending any problem instances as these happen for all problems
> in general, but please let me know if you need an instance.

> If they actually are bugs or areas that needs to be improved I would
> love to help you as much as I can, please let me know how,
> 

Thank you for your interest in glpk.

The post-optimal analysis provided by the glpk api routines
glp_analyze_bound, glp_analyze_coef and glp_print_ranges is implemented
in the same way as in the mps/360 and mpsx/370 linear programming
packages and can be considered as a de-facto standard for such kind of
analysis.

Detailed description of the post-optimal analysis report printed by
glp_print_ranges and 'glpsol --ranges' can be found on pages 108-114 in
the reference manual included in the distribution (file doc/glpk.pdf).

If you think that some analysis components are computed incorrectly,
please provide corresponding lp instance and description of the bug.
Surely, misunderstanding of the analysis report by inexperienced users
cannot be considered as a bug.

If you would like to change the contents of the analysis report, you may
change/replace the api routine glp_print_ranges.