[Paper Review] On the Emerging Potential of Quantum Annealing Hardware for Combinatorial Optimization

Byron Tasseff, Tameem Albash+5 more|arXiv (Cornell University)|Oct 9, 2022

Quantum Computing Algorithms and Architecture22 citations

TL;DR

The paper benchmarks D-Wave's Advantage quantum annealer on hardware-native CBFM-P Ising instances and shows potential run-time benefits over classical heuristics for large-scale problems, though not a universal speedup.

ABSTRACT

Over the past decade, the usefulness of quantum annealing hardware for combinatorial optimization has been the subject of much debate. Thus far, experimental benchmarking studies have indicated that quantum annealing hardware does not provide an irrefutable performance gain over state-of-the-art optimization methods. However, as this hardware continues to evolve, each new iteration brings improved performance and warrants further benchmarking. To that end, this work conducts an optimization performance assessment of D-Wave Systems' most recent Advantage Performance Update computer, which can natively solve sparse unconstrained quadratic optimization problems with over 5,000 binary decision variables and 40,000 quadratic terms. We demonstrate that classes of contrived problems exist where this quantum annealer can provide run time benefits over a collection of established classical solution methods that represent the current state-of-the-art for benchmarking quantum annealing hardware. Although this work does not present strong evidence of an irrefutable performance benefit for this emerging optimization technology, it does exhibit encouraging progress, signaling the potential impacts on practical optimization tasks in the future.

Motivation & Objective

• Assess whether the latest D-Wave Advantage hardware can outperform classical optimization methods on hardware-native Ising problems.
• Characterize the problem classes and instance distributions that are challenging for classical solvers but tractable for QA hardware.
• Evaluate how run time scales with problem size and overheads when using QA hardware.
• Discuss limitations and future opportunities for quantum annealing in practical optimization tasks.

Proposed method

• Model Ising-based optimization as QUBO with variables in {-1,1}.
• Use hardware-native CBFM-P instances tailored to Pegasus topology on Advantage QPU with 537? - actually 5,387 variables and 25,324 couplings.
• Benchmark QA against classical algorithms: steepest coordinate descent, Gurobi IQP, simulated annealing, spin-vector Monte Carlo, parallel tempering with iso-energetic clustering moves.
• Fix annealing time at 62.5 microseconds and vary anneal-read cycles between 10 and 5,120 to control compute time.
• Apply spin reversal transforms via LEAP to mitigate control errors and report solution quality relative to best-known solutions.
• Analyze both ideal runtimes (no overhead) and wall-clock runtimes including cloud service overheads.

Experimental results

Research questions

• RQ1 Can the D-Wave Advantage hardware provide run-time benefits over classical methods for certain large-scale Ising instances?
• RQ2 How does problem size affect the competitiveness of quantum annealing versus advanced classical heuristics?
• RQ3 What are the limitations of current QA hardware in achieving near-global-optimal solutions on hardware-native instances?
• RQ4 What opportunities exist for future hardware and algorithmic improvements to amplify QA benefits?

Key findings

• QA hardware can produce high-quality solutions extremely quickly for large hardware-native instances.
• For the largest CBFM-P instances (up to 5,387 variables), QA solutions are within 0.2% of the best-known solution in short runtimes.
• In ideal conditions, QA demonstrated up to around 15x run-time improvement over best classical heuristics for matching solution quality on large problems; with cloud overheads, improvements remain substantial (~18x in wall-clock).
• Run-time benefits become evident starting around 1,000 variables and grow with problem size, with PT-ICM and SA being the strongest classical competitors.
• QA results are contingent on hardware-native instance hardness and do not guarantee universal speedups for real-world, sparse problems.