Machine Learning Meets Quantum State Preparation. The Phase Diagram of Quantum Control

Abstract

The ability to prepare a physical system in a desired quantum state is central to many areas of physics such as nuclear magnetic resonance, cold atoms, and quantum computing. However, preparing a quantum state quickly and with high fidelity remains a formidable challenge. Here we tackle this problem by applying cutting edge Machine Learning (ML) techniques, including Reinforcement Learning, to find short, high-fidelity driving protocols from an initial to a target state in complex many-body quantum systems of interacting qubits. We show that the optimization problem undergoes a spin-glass like phase transition in the space of protocols as a function of the protocol duration, indicating that the optimal solution may be exponentially difficult to find. However, ML allows us to identify a simple, robust variational protocol, which yields nearly optimal fidelity even in the glassy phase. Our study highlights how ML offers new tools for understanding nonequilibrium physics.