Challenge 03 — 10–85+ atoms · beat the published benchmark curve

[kateebonner]

The same encoding at benchmark scale: 10–85+ qubits, where analog quantum optimization meets its published limits. arXiv:2511.22967 reports MIS solution quality for adiabatic and QAOA schedules on random unit-disk graphs degrading with qubit number. Reproduce an instance at matched size and shots, and beat the published value with a better schedule.

REFERENCE. arXiv:2511.22967 benchmarks MIS solution quality for the quantum adiabatic algorithm and QAOA on random unit-disk graphs, on neutral-atom QPUs from ~10 to 85+ atoms, and reports solution quality degrading with system size.

TASK. Reproduce the paper's instances at one or more system sizes and optimize the full pulse schedule — waveform shapes, sweep profile, timing, and (where the instance permits) register geometry — to raise solution quality above the published curve, scored on the paper's own metric at matched instance size and shot count. The device envelope caps hardware at 80 atoms: instances above 80 are simulation-only, so run your hardware validation at a matched size of 80 atoms or fewer.

BASELINE. The published curve itself. As a starting schedule, the starter kit ships the Challenge 02 ramp stretched to $T = 6\ 000$ ns (252 ns rise/fall, 5 496 ns hold) with the same $\Omega_b$, $\delta_0$, $\delta_f$; it will not beat the paper — it is scaffolding, not a strategy.

SCORING METRIC. Report the paper's solution-quality metric; where a self-contained number is needed, use the approximation ratio

$$ \mathcal{R} = \langle |S_{\mathrm{meas}}| \rangle / \alpha(G) $$

over valid independent sets $S_{\mathrm{meas}}$, at matched shots.

SUCCESS. Solution quality above the published value at the same system size.

BONUS. Formulate a portfolio-optimization QUBO (background: arXiv:2201.02773), map it to an MIS instance embeddable as a unit-disk register, and run it through the same pipeline.

---

Submission (Discord team channel by 16:30)

• Waveforms + register (Pulser sequence, JSON, or code that generates them)
• Simulated score (paper's metric / $\mathcal{R}$) with the published value you beat, matched instance size and shots
• Pasqal Cloud job IDs for your hardware validation runs
• Three sentences: what you changed relative to the baseline, and why it worked

Device limits: see the device envelope — and verify Device.specs at runtime.