I am interested in understanding how surrogates are effectively integrated into evolutionary algorithms (EAs). Specifically, I would like guidance on how to handle the approximation function when using static surrogates (no building of the surrogate model is needed). I'm familiar with multi-objective optimization and evolutionary algorithms, but I have never used surrogates before. I wonder whether the actual fitness function needs to be called periodically (e.g., every x generations) or if there are specific approaches to address this. Additionally, I would appreciate any relevant research papers or references on this topic.

Thank you for your insights!