[Osi] How to define SOS constraint

[John Forrest]

Christian,

The main problem is the lack of an O in your SOS, which means the
branches have an element of randomness depending on your order.  The
main advantage of SOS is branching in the correct place using a set of
weights e.g. size of facility.  Also for long sets a set can be
evaluated in fewer branches but that again relies on there being an
order.

You could do N-way branching - each branch only allowing one variable.
This would be same as your big M method without variable/constraint
overhead and it would be more likely to make correct first choice.
Three way branching is a variant - first branch only allows variable
with largest value and other two branches fix that variable and half the
others to zero (choosing each half to have an equal selection of zero
and nonzero).

John Forrest