QAOAKit: A Toolkit for Reproducible Application and Verification of QAOA

Installation

Recommended: create an Anaconda environment

conda create -n qaoa python=3
conda activate qaoa

Note that current implementation requires significant amounts of RAM (~5GB) as it loads the entire dataset into memory. Linux and macOS are currently supported.

pip install QAOAKit
python -m QAOAKit.build_tables

Example

import networkx as nx
from qiskit.providers.aer import AerSimulator
from QAOAKit import opt_angles_for_graph, angles_to_qaoa_format
from QAOAKit.qaoa import get_maxcut_qaoa_circuit

# build graph
G = nx.star_graph(5)
# grab optimal angles
p = 3
angles = angles_to_qaoa_format(opt_angles_for_graph(G,p))
# build circuit
qc = get_maxcut_qaoa_circuit(G, angles['beta'], angles['gamma'])
qc.measure_all()
# run circuit
backend = AerSimulator()
print(backend.run(qc).result().get_counts())

Almost all counts you get should correspond to one of the two optimal MaxCut solutions for star graph: 000001 or 111110.

For graphs where no pre-optimized angles are available, the angles from "The fixed angle conjecture for QAOA on regular MaxCut graphs" (arXiv:2107.00677) will be returned.

Advanced usage

More advanced examples are available in examples folder:

• Using optimal parameters in state-of-the-art tensor network QAOA simulator QTensor: examples/qtensor_get_energy.py
• Transferring parameters to large unseen instances: unweighted (examples/Transferability_to_unseen_instances.ipynb) and weighted (examples/weighted_transfer.ipynb)
• Exploring the role of problem structure in QAOA performance examples/QAOA_symmetry_and_performance.ipynb
• Exploring the performance of QAOA on small graphs as a function of average degree: examples/performance.ipynb
• Running classical algorithms for MaxCut: examples/classical_algorithms.py
• Comparing QAOA with classical algorithms for MaxCut: examples/classical_vs_quantum.ipynb
• Clustering degenerate QAOA angles: examples/degenerate_optima_in_angle_space.py

Citation

Please cite the following paper when using QAOAKit:

@inproceedings{Shaydulin2021,
  doi = {10.1109/qcs54837.2021.00011},
  url = {https://doi.org/10.1109/qcs54837.2021.00011},
  year = {2021},
  month = nov,
  publisher = {{IEEE}},
  author = {Ruslan Shaydulin and Kunal Marwaha and Jonathan Wurtz and Phillip C. Lotshaw},
  title = {{QAOAKit}: A Toolkit for Reproducible Study,  Application,  and Verification of the {QAOA}},
  booktitle = {2021 {IEEE}/{ACM} Second International Workshop on Quantum Computing Software ({QCS})}
}

Consider citing relevant papers for the particular dataset you use as well.

Install from source

git clone https://github.com/QAOAKit/QAOAKit.git
cd QAOAKit
pip install -e .
python -m QAOAKit.build_tables
pytest

Nauty installation issues

If you have an issue like "Illegal Instruction (core dumped)", you may have to force pip to recompile Nauty binaries (pip install --no-binary pynauty pynauty), use conda (conda install -c conda-forge pynauty) or install Nauty separately: https://pallini.di.uniroma1.it/

Contributing

You should set up the linter to run before every commit.

pip install pre-commit
pre-commit install

Note that linter checks passing is a necessary condition for your contribution to be reviewed.

We are in the process of moving the codebase to numpy-style docstrings. See documentation here: https://numpydoc.readthedocs.io/en/latest/format.html