Scalable Methods for Quantum Error Characterization

Quantum computing offers fundamentally new ways of information processing. Current quantum computers are reaching a size where they could be used to study real-world problems. However, they are still plagued by undesired interactions between different quantum bits (crosstalk) and between quantum bits and the environment (decoherence). These effects limit the size of quantum algorithms that can be successfully executed. To improve the computational capabilities of current quantum computers, crosstalk errors and decoherence of a quantum computer need to be characterized and mitigation strategies need to be developed. With a growing number of qubits, these tasks become very costly. In this thesis, we develop more flexible and efficient methods to construct a so-called sparse Pauli-Lindblad error model of a superconducting quantum processor, proposed by IBM in [1]. It characterizes the device noise using a sparse and low-weight set of Pauli error rates, assumed to appear on locally connected qubits, neighboring the qubits with gates applied. In this thesis we show that this restriction creates an inconsistency between describing gate errors as acting before or after the application of an ideal gate, since conjugation by the gate can map a low-weight error process onto a higher-weight one that falls outside the original model. We resolve this inconsistency with an extended model derived from a physical description of the noise acting during the gate, showing that it still scales linearly with the number of qubits. Since the standard characterization technique cannot resolve the individual rates of some of these processes, we further develop a modified characterization protocol, along with a scalable measurement scheme. Validating both approaches in a simulated environment with known ground-truth parameters, we show that our protocol is more accurate than the original one and is the only one able to characterize the additional error processes introduced by the extended model.