Session S17: Characterizing Quantum Computing Systems and Components II

Credibility of noisy intermediate-scale quantum computers

The value of quantum computers and simulators lie in solving efficiently and correctly problems that are hard classically. This is particularly relevant for simulation, sampling, and optimisation problems whose solutions cannot be verified efficiently classically, unlike integer factoring. I will present methods that can provide, for a given problem, an upper bound on the variation distance between an experimentally obtained output from a noisy quantum computer and the ideal output from a noiseless device. I will show how this can be achieved without the inherently infeasible method of simulating the quantum computation classically. I will highlight the vital interplay between empirical experimental observations and mathematical assumptions. I will conclude with some applications on judging the credibility of trusting noisy intermediate-scale quantum computers in simulating hard condensed-matter physics problems.
  

Logical Cooling for Noise Reduction in Analogue Quantum Simulation

Analogue quantum simulation is arguably the most promising near-term application of quantum computing. However, analogue simulators are susceptible to noise, and it is not known if realistic noise destroys their computational power. Here, we report our progress using a technique to remove errors in the computational basis of the system, without resorting to a full error correcting scheme, to both measure and increase an analogue quantum simulator's robustness to noise.   

Characterizing the Propagation of Gate Errors in Experimental Quantum Algorithms

Universal quantum computation requires a complete gate-set of single-qubit operations and a two-qubit entangling gate. As research progresses towards useful applications of NISQ-era quantum processors, infidelities in these operations limit the depth of algorithms accessible on near-term devices. While Clifford Randomized Benchmarking provides a rough metric for the fidelity of operations in a random quantum algorithm, this metric can fail to capture the buildup of coherent error in algorithms with significant internal symmetry, such as Trotterization. In this talk, we present recent measurements characterizing the propagation of gate errors in a Trotterized quantum algorithm performed using >99.9% fidelity single-qubit and 99.7% fidelity controlled-phase (CZ) gates between flux-tunable transmon qubits. We consider the role played by coherent and incoherent gate errors in deep quantum circuits, and we look at how their propagation depends on the choice of gate compilation.   

Time-Dependent Simulation of Superconducting Quantum Circuits in the Presence of Non-Markovian Noise

Understanding the practical impact of realistic noise on the performance of superconducting quantum circuits is vital to evaluating their prospects for potential applications and system architectures. This understanding is particularly limited at present for the more complex circuits involved in quantum annealing and reversible logic, which are subject to both damping and non-Markovian 1/f flux noise. In this presentation, we discuss a quantum trajectory-based dynamics simulation method for such circuits, and present progress towards its use in evaluating residual power dissipation in reversible logic circuits.   

Low Rank Density Matrix Evolution for Noisy Quantum Circuits

Quantum circuit simulators validate the accuracy of quantum computing hardware and facilitate the invention of quantum algorithms and the deployment of quantum solutions on real world problems. In this work, we present an algorithm for simulating noisy quantum circuits based on the fact that the effective dimensionality of a density matrix is low when the noise level is reasonable small. Under certain conditions on the noise level and circuit depth, we proof that the numerical rank of a density matrix grows only linearly with the number of qubits. This allow us to track the evolution of a compressed representation of a density matrix with exponentially less computational resource. We implement this algorithm in an in-house simulator, Quasar, showing that the low rank algorithm speeds up simulations more than two orders of magnitude against the standard full density matrix method, with a trade-off of small amount of error. We benchmark the performance of the algorithm with different noise channels, noise strength and circuit types. Finally, we implement instances of Grover’s search algorithm, showing that the low rank evolution benefits not only random circuit simulations, but also structured quantum algorithms.   

Fast estimation of sparse quantum noise

To achieve a scalable estimation of quantum noise we need to learn efficient and complete representations of that noise. We can do this by using descriptions that have clear and relevant physical assumptions baked in. Here I will present work on a scalable and complete protocol to learn a Pauli channel which only has s non-negligible Pauli error rates. So long as the number of error rates scales polynomially with the number of qubits, then this is an efficient protocol and requires only O(n s) experiments, linear in the number of qubits. The classical computational effort is also efficient in n. Learning these error rates is directly relevant to improving quantum error correction and we have already implemented this to efficiently learn all the errors on a 14-qubit quantum device.   

Random Circuit Metrics for Performance Assessment and Model Testing

Random circuit metrics, such as cross-entropy benchmarking, have recently emerged as a powerful set of tools to assess the performance of quantum processors. A natural question is to what extent these techniques can be used to learn noise characteristics of the processor. Here, we present results on extensions of random circuit metrics to testing error models. We demonstrate for small processors built from superconducting qubits that analysis of random circuit distributions is a viable method to compare candidate error models for device operation. Such models feed directly into the debugging cycle, and can be used to guide future operation towards optimal performance, or in the design of future devices.

This document does not contain technology or technical data controlled under either the U.S. International Traffic in Arms Regulations or the U.S. Export Administration Regulations.   

Benchmarking tools for NISQ systems

As the field marches towards quantum advantage with near-term quantum processors, it becomes imperative to characterize, verify, and validate performance. An outstanding scientific challenge in the community is a scalable set of metrics or experiments which can shed light on the usability of a device for near-term algorithms. Moreover, it becomes critical to explore techniques to extend the computational reach of noisy systems, be it through understanding underlying non-idealities, or more efficient circuit compilation. In this talk I will review the work we are doing at IBM to benchmark NISQ devices and I will discuss our recent results on quantum volume, large-scale entanglement and randomized benchmarking.


  

Model Refinement of Noisy Quantum Circuits Using Experimental Characterization

Current quantum processing units represent noisy intermediate scale quantum systems that tend to be poorly characterized. Accurate modeling of these devices can provide insight into the underlying noise as well as methods for mitigating errors. We present a test-driven methodology for quantifying QPU performance and characterizing NISQ behavior that offers an alternative to costly experimental characterizations using standard tomographic methods. We demonstrate modeling of noisy gate operations by fitting experimental characterization circuits using a series of bootstrapped numerical methods. We generate parameterized gate models that are composed easily to model noisy quantum circuits. We demonstrate the effectiveness of this modeling method for applications of GHZ state preparation and the Bernstein-Vazirani algorithm using a family of superconducting transmon QPUs. We quantify the accuracy of the generated models using the total variation distance between experimental observations and numerically simulated results. Our results show that model refinement from test-driven experimental characterization offers an accessible methodology for approximating performance of NISQ devices.   

Modeling leakage in superconducting quantum computers

Current quantum computing architectures are fragile and prone to various imperfections that limit computational capabilities. In order to better understand behavior of quantum hardware, we need to develop a theoretical framework that captures possibile sources of device errors.

Recent development in noise description allow us to characterize within statistical accuracy net effects of numerous unknown contributions. The most commonly used method for this is based on randomized Pauli channels (PC) approach, that can reflect circuit's fidelity.

In this talk, we discuss a potential extension to PC technique, that can in principle take into account leakage to non-computation basis. The method is based on Weyl matrices. We show the effect of projecting leakage subspace into a qubit space, and compare the method with PC. Finally, we briefly explain possibility of expanding this model to larger dimensions.   

Benchmarking noise with Quantum alternating operator ansatz (QAOA) circuits with a symmetry

QAOA as a quantum heuristic algorithm has attracted a lot of interest with its simple iterative structure. We propose adapting a symmetry-preserving QAOA circuit for benchmarking noise. Consider the system consisting of a number of subsets of qubits, we design the QAOA such that it preserves certain quantity of the subsets (e.g., a constant Hamming weight). When errors are present, the evolution will quickly escape from the preserved subspace. We study how this escaping rate scales with the iteration of the circuit and system size. Our analysis shows for a wide range of local noise channels the escaping rate can be exactly obtained with only the knowledge of the error model, thanks to the non-increasing reverse causal cone with QAOA level with respect to noise. We show that the analysis is stable against inhomogeneity in the subsystem, hence our scheme is feasible for benchmarking local noises in realistic situations.   

Entanglement in superconducting qubits and quantum foundations

Leggett has proposed an experimental program to probe the limits of quantum mechanics using superconducting qubits. With this in mind, we consider the number of particles entangled in certain superconducting qubit states. Implications for quantum foundations are discussed.   

Progress towards high-fidelity CZ gates in a tunable coupling architecture

We describe progress towards implementing high-fidelity CZ gates in the Sycamore tunable coupling architecture.