Re: [Help-glpk] Modelling Advice Request - Project Tasks

[Andrew Makhorin]

> Below is my version of a GMPL implementation that uses the
> transitivity constraints.  This works pretty well short of using row
> generation.  I used the xij+xji = 1 relation to reduce the number of
> variables and constraints.  Would the GMPL/GLPK preprocessor be just
> as efficient as doing this in the modeling step?

It must be, however, currently this feature is disabled. I implemented
preprocessing of row doubletons (ax + by = c) to eliminate one of the
variables, but sometimes this causes excessive round-off errors.
Probably it would be reasonable to distinguish between elimination of
binary complements and a more general lp/mip context.