Reinforcement Learning for Digital Quantum Simulation

Digital quantum simulation is a promising application for computers. Their free programmability provides the potential to simulate unitary evolution of any many-body Hamiltonian with bounded spectrum by discretizing time operator through sequence elementary gates, typically achieved using Trotterization. A fundamental challenge in this context originates from experimental imperfections involved which critically limits number attainable gates within reasonable accuracy and therefore achievable system sizes times. In work, we introduce reinforcement learning algorithm systematically build optimized circuits digital upon imposing strong constraint on allowed gates. With consistently obtain that reproduce physical observables as little three entangling long times large sizes. As concrete examples apply our formalism range Ising chain lattice Schwinger model. Our method makes larger scale possible scope current technology.