Re: [Help-glpk] A question on row generation

[Andrew Makhorin]

>>  but
>> each with a different 'branching constraint' (that is, the constraint
>> added to define the branches, for example, if in the parent node, the
>> optimal solution of an integer variable x* = 2.5 is chosen to branch
>> on, you add x<=2 and x>=3 to define branch A and B respectively).
> 
> Branching on variable needs adding no rows; this just changes current
> bounds of the branching variable. It is possible to branch on constraint
> (for example, on disjunctive or GUB constraint), however, currently this
> feature is not used in glpk.

To branch on constraint you may add a free row to the current
subproblem prior to branching (or use some existing row) and then
specify lower and upper bounds of that row for each branch in the same
way as for branching on variable. Then each child subproblem will
inherit that row with corresponding bounds specified.