[Fwd: identifying infeasible problem without using the LP presolver]

[Andrew Makhorin]

-------- Forwarded Message --------
From: Will Tipton <wtipton@janestreet.com>
To: help-glpk@gnu.org
Cc: John Rice <jrice@janestreet.com>
Subject: identifying infeasible problem without using the LP presolver
Date: Mon, 21 Mar 2022 11:15:44 -0400

> Hi help-glpk,
> 
> We're using GLPK's simplex algorithm to solve some medium-sized lp
> problems (hundreds of variables and constraints). We've found that the
> presolver produces bad solutions, and we suspect that it's due to
> numerical issues, since the constraints may be near-degenerate in many
> cases.
> 
> Running without the presolver usually works great. However, we can't
> rely on the results in this case, because the simplex algorithm
> returns an OK error code even when it fails to find a feasible
> solution.
> 
> So:
> How hard would it be to get errors in case there is no feasible
> solution when not using the presolver?
> Any tips on debugging presolver stability problems?
> Thanks in advance,
> Will