Quantum Architecture Search

OptCRLQAS with ReaPER+ achieved the lowest energy error in the reported 5-qubit Heisenberg, 6-qubit BeH2, and 8-qubit H2O comparisons against non-RL baselines. For 12-qubit H2O with OptCRLQAS, fixed ReaPER achieved the lowest minimum energy error among listed replay variants. The QAS reward scheme rewarded reaching the energy threshold, penalized episode failure, and otherwise used normalized cost improvement. The QAS scaling claims are limited because experiments only reached tasks up to 12 qubits. Quantum architecture search targeted parameterized circuits that minimize molecular or spin Hamiltonian energy.