Quantum Optimization Benchmarks

This repository accompanies the accepted paper published in Quantum Science and Technology:

From Circuits to Hardware: Benchmarking Standard and Qubit-Efficient Quantum Optimization on Real Hardware

Monit Sharma and Hoong Chuin Lau (2026)

The earlier pre-publication study is preserved on the arxiv branch: A Comparative Study of Quantum Optimization Techniques for Solving Combinatorial Optimization Benchmark Problems.

The project is built around a hardware-aware benchmark of gate-based quantum optimization on four NP-hard 0-1 combinatorial problem classes:

• Multi-Dimensional Knapsack Problem (MDKP)
• Maximum Independent Set (MIS)
• Quadratic Assignment Problem (QAP)
• Market Share Problem (MSP)

Rather than reporting only simulator-side objective values, the benchmark compares methods under realistic execution constraints, including logical qubit requirements, transpiled circuit growth, two-qubit gate counts, backend eligibility, and execution robustness on real hardware.

What The Paper Studies

The paper compares three broad families of quantum optimization methods:

• Variational energy-minimization methods: VQE, CVaR-VQE
• QAOA-style methods: QAOA, MA-QAOA, WS-QAOA
• Qubit-efficient methods: PCE, QRAO

All methods are evaluated through a shared problem-to-QUBO workflow and then studied in both simulator and hardware settings. The goal is not to claim quantum advantage, but to establish a reproducible empirical baseline for how these methods behave once compilation, routing, noise, and recovery are treated as first-class parts of the experiment.

Main Benchmark Message

The paper’s central findings are reflected directly in this repository:

• No single method family dominates across all four benchmark problems.
• Lower logical qubit count improves executability, but does not by itself guarantee better recovered hardware solution quality.
• Simulator conclusions do not always survive hardware execution.
• Compilation overhead, transpiled depth, and two-qubit gate growth materially affect practical outcomes.

In other words, this repository is meant to support full-stack benchmarking, not just logical-circuit comparison.

Repository Structure

The repo has two complementary layers.

1. Original benchmark datasets and notebook workflows

• Market_Share
• Maximum_Independent_Set
• Multi_Dimension_Knapsack
• Quadratic_Assignment_Problem

These folders preserve the benchmark sources, problem-specific notebooks, and historical artifacts used during development.

2. Reproducible research pipeline

The packaged workflow lives under research_benchmark/README.md. It provides:

• a common qobench CLI for listing instances and running model/QUBO experiments
• a hardware-facing runner for VQE, CVaR-VQE, QAOA, MA-QAOA, WS-QAOA, PCE, and QRAO
• a simulator-only runner
• a smoke-test runner across all supported methods
• structured output directories for results, logs, checkpoints, and paper artifacts

Benchmark Scope In This Repo

The repository includes the benchmark ingredients described in the paper:

• problem instances across MDKP, MIS, QAP, and MSP
• standard and qubit-efficient quantum optimization implementations
• simulator and hardware execution workflows
• preserved logs, summaries, plots, and result tables

Important artifact locations:

• research_benchmark/research_benchmark/results_hardware/
• research_benchmark/research_benchmark/results_simulator/
• research_benchmark/all_logs/
• research_benchmark/checkpoints/

Those directories are intentionally versioned so readers can inspect the paper-facing outputs directly.

Quickstart

Use Python 3.12 for the pinned environment (including CPLEX). Dependency audits and update policy are documented in Dependency security.

From the repository root:

python3.12 -m venv .venv
.venv/bin/python -m pip install --upgrade pip
.venv/bin/python -m pip install -r requirements.txt
.venv/bin/python -m pip install -e research_benchmark

You can also use the repo helpers:

make venv
make install
make test

Common Commands

List benchmark instances:

make list-instances PROBLEM=mis LIST_LIMIT=5

Run one classical/QUBO benchmark instance:

make run PROBLEM=mis INSTANCE=Maximum_Independent_Set/mis_benchmark_instances/1tc.8.txt

Run one simulator benchmark:

make simulate PROBLEM=mkp METHOD=qaoa INSTANCE=Multi_Dimension_Knapsack/MKP_Instances/sac94/hp/hp1.dat

Run the short all-method smoke suite:

make smoke PROBLEM=mis

Installed console scripts:

qobench --help
qobench-hardware --help
qobench-smoke --help

Documentation Guide

Start with these files:

Path	Purpose
research_benchmark/README.md	Main runnable research pipeline.
docs/PROJECT_LAYOUT.md	Repository map.
docs/REPRODUCIBILITY.md	Environment and reproduction notes.
research_benchmark/research_benchmark/README.md	Checked-in paper artifacts.
research_benchmark/all_logs/README.md	Archived raw logs.
research_benchmark/checkpoints/README.md	Resume metadata for long runs.

Citation

If you use this repository, please cite the accepted paper:

@article{Sharma_2026,
  title={From circuits to hardware: benchmarking standard and qubit-efficient quantum optimization on real hardware},
  volume={11},
  ISSN={2058-9565},
  url={http://dx.doi.org/10.1088/2058-9565/ae94a4},
  DOI={10.1088/2058-9565/ae94a4},
  number={4},
  journal={Quantum Science and Technology},
  publisher={IOP Publishing},
  author={Sharma, Monit and Lau, Hoong Chuin},
  year={2026},
  month={Sept},
  pages={045029}
}

License

MIT. See LICENSE.