Re: [Help-glpk] big M

[Andrew Makhorin]

> numbers in GLPK are of type double. Hence the highest number that can
> be specified will be around 1.7E308 on an X86 platform.

On any platform where the IEEE 754 format is used to represent double's,
not only on X86 platform due to IEEE 754.

http://en.wikipedia.org/wiki/IEEE_754

> When using a big M approach, M should be choses as small as possible to
> avoid scaling problems. Hence a good choice of M is problem specific.

There are two different techniques, both called "big M". In linear
programming the "big M" technique may be used to find a feasible basic
solution; however, this technique is naive and using it may encounter
serious numerical difficulties on solving instances with more than 5
rows and columns. In mixed integer programming the "big M" technique is
used to model disjunctive conditions, like "x = 0 or x >= 1".