Effectiveness of quantum annealing for continuous-variable optimization

TL;DR

This paper evaluates quantum annealing's ability to optimize continuous-variable functions, comparing hardware performance with classical algorithms and demonstrating potential advantages under ideal conditions.

Contribution

It introduces a method to apply quantum annealing to continuous problems via domain-wall encoding and benchmarks hardware against classical algorithms, highlighting the importance of coherence.

Findings

D-Wave 2000Q matches classical algorithms in limited time domains

TEBD simulation shows coherent quantum annealing outperforms other methods

Quantum annealing has potential to outperform classical algorithms with reduced noise

Abstract

The application of quantum annealing to the optimization of continuous-variable functions is a relatively unexplored area of research. We test the performance of quantum annealing applied to a one-dimensional continuous-variable function with a rugged energy landscape. After domain-wall encoding to map a continuous variable to discrete Ising variables, we first benchmark the performance of the real hardware, the D-Wave 2000Q, against several state-of-the-art classical optimization algorithms designed for continuous-variable problems to find that the D-Wave 2000Q matches the classical algorithms in a limited domain of computation time. Beyond this domain, classical global optimization algorithms outperform the quantum device. Next, we examine several optimization algorithms that are applicable to the Ising formulation of the problem, such as the TEBD (time-evolving block decimation) to simulate ideal coherent quantum annealing, simulated annealing, simulated quantum annealing, and spin-vector Monte Carlo. The data show that TEBD's coherent quantum annealing achieves far better results than the other approaches, demonstrating the effectiveness of coherent tunneling. From these two types of benchmarks, we conclude that the hardware realization of quantum annealing has the potential to significantly outperform the best classical algorithms if thermal noise and other imperfections are sufficiently suppressed and the device operates coherently, as demonstrated in recent short-time quantum simulations.