Running Quantum Computers in Discovery Mode

Abstract

Using a 36-qubit quantum processor, we demonstrate that, by operating in conjunction with a classical machine learning agent, quantum computers can discover instances of interesting quantum many-body dynamics. The central object in this new mode of use of a quantum device is an "interest function" defined for a given circuit (family) instance that can be evaluated on a quantum computer. The circuit is adapted by the learning agent to maximize interest. We illustrate this approach using two examples and show that, within a sufficiently general circuit family, two simple interest functions based on (i) binary classifiability of evolved states and (ii) spectral properties of the unitary circuit, are maximized by discrete time crystals (DTCs) and dual-unitary circuits, respectively. For the classifiability-based interest function, we implement the protocol on a superconducting quantum processor and find that it indeed discovers DTCs with high probability. For the dual-unitaries, our simulations of the dynamics suggest that an interest-function optimization would have set us close to a discovery of such unitaries. Our results using quantum devices and accompanying simulations suggest that learning agents with access to quantum-computing resources can almost autonomously discover new phenomena in many-body quantum dynamics, and establish the design of good interest functions optimizable in hybrid devices as a paradigm for quantum many-body physics.

10 + 8 pages, 4+6 figures