Quantum annealing (QA) is a global search-heuristic designed to solve NP-hard optimization problems. Currently, D-Wave Systems is the only commercial QA company, boasting a line of chips that have over 2000 working qubits. Unfortunately, their API lacks many useful features for exploring the actual physics and efficacy of QA as an optimization tool. Further, the typical QA routine is prone to errors for a variety of reasons: noise, problem mis-specification, minor-embedding errors, and using incorrect annealing parameters. This work attempts to alleviate both of these issues with a single Python package: dwaveutils. With it, we explore two case studies: simulations of transversefield Ising Hamiltonians and forward-reverse error-mitigation annealing. The first shows the utility that dwaveutils brings when it comes to submitting and post-processing problems on D-Wave. The second shows the power of having a numeric annealing solver that allows for exploration of modified annealing routines. With it, we've uncovered several potentially fruitful investigations that could reduce errors on modern devices with no cost in ancilla qubits and allow for a new feature: graph introspection. Overall, our package allows for easy exploration and improvements of modern quantum annealing.
We consider a range of unconventional modifications to quantum annealing (QA), applied to an artificial trial problem with continuously tunable difficulty. In this problem, inspired by ``transverse field chaos'' in larger systems, classical and quantum methods are steered toward a false local minimum. To go from this local minimum to the global minimum, all
N spins must flip, making this problem exponentially difficult to solve. We numerically study this problem by using a variety of new methods from the literature: inhomogeneous driving, adding transverse couplers, and other types of coherent oscillations in the transverse field terms (collectively known as RFQA, which stands for Random Field Quantum Annealing or Radio Frequency Quantum Annealing). We show that all of these methods improve the scaling of the time to solution (relative to the standard uniform sweep evolution) in at least some regimes. Comparison of these methods could help identify promising paths towards a demonstrable quantum speedup over classical algorithms in solving some realistic problems with near-term quantum annealing hardware.
Quantum annealing, a quantum computing paradigm, offers potential for accelerating computationally intensive problems like constraint satisfaction. This paper explores the integration of quantum annealing with dynamic relaxation to tackle complex constraint satisfaction problems (CSPs) with high-dimensional constraints. We propose a novel hybrid approach where quantum annealing is used to efficiently explore the solution space, followed by dynamic relaxation to refine the solution and improve its quality. The core mechanism centers around leveraging the quantum annealing's ability to find near-optimal solutions, coupled with the dynamic relaxation's capability to iteratively improve the solution's precision. This research aims to address the limitations of traditional CSP solvers and provides a promising avenue for enhancing performance in challenging scenarios.