The generalized minimum spanning tree problem: A parameterized complexity analysis of Bi-level optimisation

Bi-level optimisation problems have gained increasing interest in the field of combinatorial optimisation in recent years. With this paper, we start the runtime analysis of evolutionary algorithms for bi-level optimisation problems. We examine the NP-hard generalised minimum spanning tree problem and analyse the two approaches presented by Hu and Raidl [7] in the context of parameterised complexity (with respect to the number of clusters) that distinguish each other by the chosen representation of possible solutions. Our results show that a (1+1) EA working with the spanning nodes representation is not a fixed-parameter evolutionary algorithm for the problem, whereas the global structure representation enables to solve the problem in fixed-parameter time. Furthermore, we present hard instances for each approach and show that the two approaches are highly complementary by proving that they solve each other's hard instances very efficiently.