Improving pulse exploration in model predictive quantum control

Many typical quantum control problems can be solved by local methods for optimal control–relying on exploitation of the model information without fear of becoming trapped at solutions far from the global optimum. Nevertheless, if models of the experiment are inaccurate or the control tasks are sufficiently complex, additional global exploration might become necessary to avoid traps and find good solutions. In this work, we explore the exploitation / exploration tradeoff within model predictive quantum control (MPC). MPC is a local control scheme that utilizes experiment feedback online during the synthesis of the control pulse. With this modest request for data, MPC is able to design pulses which are robust to model uncertainty and system disturbances. In addition, sampling-based MPC algorithms offer a way to improve upon the exploration properties of traditional MPC by leveraging the predictions of a nominal system model or models. By combining model-based exploration with experiment feedback, sampling-based MPC is able to discover robust pulse sequences for hard problems in quantum optimal control.