A Physics-Informed Bayesian Approach for RIS Optimization in Complex Wireless Channels

We introduce a Bayesian optimization framework for sum rate maximization in RIS-assisted MU-MIMO systems. The decision variables include discrete RIS phase shifts and continuous array precoding vectors. This yields a non-convex, combinatorial problem that challenges conventional optimization methods. This paper proposes a Bayesian surrogate model with physics-informed bases and geometric embeddings to navigate the large configuration space. The work greatly reduces computational complexity and pilot overhead while delivering fast convergence towards near-optimal performance.