QAOA for Experiment Campaign Ordering

Ordering plasma shots and parameter sweeps to maximize information under machine-time limits is a combinatorial problem we test QAOA against.

Sequencing scarce machine time

Commissioning and characterizing the breeder and burner involves large parameter sweeps: negative-triangularity shape scans on the breeder, breeding-ratio configurations, plug-density operating points on the burner. Machine time is scarce and shots have setup dependencies (a configuration change costs conditioning time). Choosing an order that maximizes information gained per unit machine time is a sequencing optimization.

# Sequence pi over shots to maximize info minus transition cost
maximize  sum_i  I(shot_{pi(i)})  -  sum_i  t_setup( pi(i-1) -> pi(i) )
#   I(shot)      expected information gain (from twin/surrogate model)
#   t_setup      reconfiguration time between successive shots
# a prize-collecting sequencing / TSP-like structure -> NP-hard

Ising encoding of an ordering

An ordering is encoded with position-indexed binaries x_{s,p} = 1 if shot s occupies position p, with permutation constraints as penalties, the same recipe as the traveling-salesman Ising map, then handed to QAOA or a quantum annealer.

H = -sum_{s,p} I_s x_{s,p}
  + sum_{s,s',p} t_{s,s'} x_{s,p} x_{s',p+1}       # transition cost
  + P*( sum_s( sum_p x_{s,p} -1 )^2 + sum_p( sum_s x_{s,p} -1 )^2 )
#   permutation penalties: each shot once, each slot once

Where this connects and where it stops

• Information gain I_s is supplied by the classical digital twin and active-learning surrogates, not by the quantum step.
• The quantum solver, if used, only sequences; it never issues a machine command.
• Classical schedulers already run these campaigns; QAOA must beat them on Kronos instances to matter.

This is a clean, safety-isolated place to benchmark quantum optimization: a bad schedule wastes planning, not plasma. It remains research until a quantum method demonstrably outperforms the classical baseline. The active-learning loop that feeds it is described in the L0 surrogate pages; the honest overall stance is in the roadmap.