For any problem with optimal solution k, a superset of that problem formed by relaxing the rule set will have an optimal solution at least as efficient as k.

It comes up in AI a few times.