Quantum Nescimus: Improving the characterization of quantum systems from limited information

Abstract

We are currently approaching the point where quantum systems with 15 or more qubits will be controllable with high levels of coherence over long timescales. One of the fundamental problems that has been identified is that, as the number of qubits increases to these levels, there is currently no clear way to use efficiently the information that can be obtained from such a system to make diagnostic inferences and to enable improvements in the underlying quantum gates. Even with systems of only a few bits the exponential scaling in resources required by techniques such as quantum tomography or gate-set tomography will render these techniques impractical. Randomized benchmarking (RB) is a technique that will scale in a practical way with these increased system sizes. Although RB provides only a partial characterization of the quantum system, recent advances in the protocol and the interpretation of the results of such experiments confirm the information obtained as helpful in improving the control and verification of such processes. This thesis examines and extends the techniques of RB including practical analysis of systems affected by low frequency noise, extending techniques to allow the anisotropy of noise to be isolated, and showing how additional gates required for universal computation can be added to the protocol and thus benchmarked. Finally, it begins to explore the use of machine learning to aid in the ability to characterize, verify and validate noise in such systems, demonstrating by way of example how machine learning can be used to explore the edge between quantum non-locality and realism.