Two issues regarding constraints in meta-optimization should be mentioned; they are as follows:

• Constraints can be made on an optimizer's control parameters in the same manner as for an optimization problem by implementing the EnforceConstraints() and Feasible() methods in the optimizer's class. This means the meta-optimizer will search for control parameters that are feasibly optimal, allowing you to search for control parameters that meet certain criteria; for example, they have certain relationships with each other, such as one parameter being smaller than the other, and so on. See the source code of the MOL optimizer for an example of this.
• Constraint satisfaction is ignored when determining how well an optimizer performs in making up the meta-fitness measure. This is an open research topic, but experiments suggest that an optimizer's control parameters should be meta-optimized for unconstrained problems. This will also yield good performance on constrained problems.