On Chromosome Crossover in Multimodal Adaptive Graph Evolution

Crossover operators have historically shown limited effectiveness in Cartesian Genetic Programming (CGP), with mutation-only approaches typically dominating the evolutionary process. In this paper, we explore a natural crossover mechanism enabled by Multimodal Adaptive Graph Evolution (MAGE), a multi-chromosome generalization of CGP that groups functions by return type and constrains graph mutation based on type coherence. MAGE's distinct architecture creates an opportunity for a chromosome-level crossover operator that was not previously feasible in traditional CGP. We implement and evaluate this crossover approach on selected problems from the Second Program Synthesis Benchmark Suite (PSB2), comparing it against a mutation-only strategy. Our experiments reveal that chromosome-level crossover offers mixed effectiveness in MAGE, with promising exceptions in problems like GCD and Twitter. This work continues the inquiry into effective crossover methods in graph-based genetic programming, revealing how effectiveness is problem dependent.