[Help-glpk] [Fwd: Using glp_exact() after glp_simplex() has timed out]

[Andrew Makhorin]

-------- Forwarded Message --------
From: Brendan McKay <address@hidden>
To: address@hidden
Subject: Using glp_exact() after glp_simplex() has timed out
Date: Mon, 5 Jun 2017 16:50:18 +1000

Hi, I use the exact solver for my work in combinatorics and find it
very efficient for my problems (tens to hundreds of thousands of
variables but only a few dozen constraints).

GLPK 4.61 running on Ubuntu 14.04.5 LTS.

As recommended in the manual, I call glp_simplex() first and then call
glp_exact(), which usually gives a nice speedup.  It set an iteration
limit for glp_simplex() because sometimes it cycles.  However, if the
call to glp_simplex() returns due the number of iterations being exceeded
(return GLP_EITLIM), sometimes glp_exact() then runs forever.  I get a
never-ending sequence of outputs like this:

   26194:   infsum =                    144   (33)
   26406:   infsum =                    144   (33)
   26619:   infsum =                    144   (33)
   26831:   infsum =                    144   (33)
   27044:   infsum =                    144   (33)
   27257:   infsum =                    144   (33)
   27469:   infsum =                    144   (33)

I infer that glp_simplex() is leaving things in a state unsuitable for
starting glp_exact().  In such a case it would be enough to force
glp_exact() to begin with a blank slate rather than using the
left-overs from a timed-out glp_simplex().  But how can I achieve
that short of reconstructing the problem or keeping a copy?

Thanks, Brendan.