Re-examining the quantum volume test: Ideal distributions, compiler optimizations, confidence intervals, and scalable resource estimations

Abstract

The quantum volume test is a full-system benchmark for quantum computers that is sensitive to qubit number, fidelity, connectivity, and other quantities believed to be important in building useful devices. The test was designed to produce a single-number measure of a quantum computer's general capability, but a complete understanding of its limitations and operational meaning is still missing. We explore the quantum volume test to better understand its design aspects, sensitivity to errors, passing criteria, and what passing implies about a quantum computer. We elucidate some transient behaviors the test exhibits for small qubit number including the ideal measurement output distributions and the efficacy of common compiler optimizations. We then present an efficient algorithm for estimating the expected heavy output probability under different error models and compiler optimization options, which predicts performance goals for future systems. Additionally, we explore the original confidence interval construction and show that it underachieves the desired coverage level for single shot experiments and overachieves for more typical number of shots. We propose a new confidence interval construction that reaches the specified coverage for typical number of shots and is more efficient in the number of circuits needed to pass the test. We demonstrate these savings with a $QV=2^{10}$ experimental dataset collected from Quantinuum System Model H1-1. Finally, we discuss what the quantum volume test implies about a quantum computer's practical or operational abilities especially in terms of quantum error correction.