Time advantage in QAOA MaxCut over classical solutions

Directory structure

The general approach is to have data an presentation layers separate à-la MVC pattern.

• data/*.nc - data files
• data/generators/ - scripts that generate the data
• plots/*.ipynb - scripts that generate figures
• plots/pdf/ - pdf output from figure generators

Nice tools used

You'll need these to fully understand the code

• QTensor - tensor network simulator with focus on MaxCut
• Cartesian Explorer - a handy tool to map multi-dimensional data
• xarray - used to store datasets with their coordinates in netcdf format
• Gurobi and gurobipy - classical optimization of combinatorial problems.

Gurobi installation

See instructions on the official site, the general steps are the following:

1. Register at the website gurobi.com
2. Unpack an archive
3. Run licensing code (included in data/generators/*)

Reproducibility

All the figures in the paper are generated by notebooks in this repository.

• Figure 3: plots/Longrange correlations [figure 3].ipynb
• Figure 4: plots/Time bounds [figure4].ipynb
• Figures 5 and 6: plots/Quantum match gurobi time [figure5, 6].ipynb
• Figure 7: plots/Match quality multi-shot QAOA[figure 7].ipynb
• Figure 8: plots/Cost variance vs N [figure8].ipynb

data in data/xarray_0.19.0 were obtained using an older version of xarray.It was converted to json which was then converted to json and compressed usng script data/convert/convert_xarray.py. Some of them failed to load and may need to be converted using older version of xarray.

---

Note: if you don't see figures in notebooks, select "File -> Trust Notebook"