Machine learning for optimizing quantum control

 

Controlling the quantum state of a system using external parameters (classical, time dependent terms in the Hamiltonian) is important both for applications and fundamental experiments. In many cases, there is decoherence from the environment which limits the time available to reach a desired state, the parameters of the system is not known with sufficient accuracy, or the control operations are not perfect. It can therefore be important to optimize the control operations to meet certain requirements or constraints. One approach to this is to use machine learning methods to find robust optimal control operations. We have recently applied one such method, Deep Q Learning (DQL), to a simple example problem (the splitting of the wavefunction of a particle in  a box). The details of the problem and the method we applied are available in the paper: Vegard B. Sørdal and  Joakim Bergli, Deep reinforcement learning for robust quantum optimization, arXiv:1904.04712. In this work, we only demonstrated that the DQL method can be applied to this type of problems, but we did not evaluate its performance in any detailed manner.