[Coin-discuss] Probing problem

[Joerg Herbers]

I've got problems using CglProbing in preprocessing (via CglPreProcess) on an IP formulation with general integer variables.

In the formulation, a variable has lb=1 and ub=DBL_MAX. Since solval=1 in CglProbing.cpp (l. 2424), the "down" and "up" variables are set to 1.0 and 2.0, respectively (l. 2432-2435). Consequently, "movement" is set to 1.0-DBL_MAX (l. 2462). When the first alternative is tested in probing (l. 2497), colUpper is set to colUpper-movement=DBL_MAX+(1.0-DBL_MAX). This should give 1.0 - but it actually results in 0.0, leading to infeasibility and incorrect results of probing.

Is there a general recommendation of not using DBL_MAX as a marker for an open upper bound? Or could we make CglProbing "DBL_MAX-safe"?

Thanks in advance,
Joerg



-- 

Dr. Jörg Herbers
Airport Systems Division
INFORM GmbH, Pascalstraße 23, 52076 Aachen, Germany
Telefon: (+49) 2408 9456-24, Fax: (+49) 2408 9456-25
E-Mail: joerg.herbers at inform-ac.com, Web: http://www.inform-ac.com