Re: [Help-glpk] information on presolve results

[Ali Baharev]

Thank you for your response.

One reason is debugging. I generate the LP problems by linearizing
nonlinear problems. A redundant nonlinear constraint may or may not be
nonredundant in the LP problem after linearization. Identifying
redundant rows / columns could help me debugging the nonlinear part.
Another reason is numerical stability and performance issues.

What do you advise?

Thank you,

Ali

On Jan 10, 2008 3:35 PM, Andrew Makhorin <address@hidden> wrote:
> > Is there a way to get information from the presolve routines why
> > certain rows / columns can be eliminated from the model?
>
> > My LP problems only have continuous variables.
>
> The lp presolver (see the module glplpp) keeps all information about
> applied transformations in an internal stack, however, this information
> is not intended to be used out of the lp presolver. Besides, it would be
> problematic to analyze it adequately due to its derivative nature. For
> example, some column may be eliminated, because other column in the same
> row was fixed by its reduced costs obtained from other row that made the
> former column an implied slack variable; i.e. there may not be a clear
> reason why this or that column has been removed.
>
> For which purpose do you need to obtain such information?
>
>