Q-VOLT: A Quantum–Classical Pipeline for Multi-Period Data-Center Siting on the IEEE 30-Bus System

Abstract

Q-VOLT is a hybrid quantum–classical pipeline for the multi-period data-center siting-and-scheduling problem on the IEEE 30-bus system with real California-ISO locational-marginal-price data and published hyperscale workload traces. I formulate the problem as a QUBO with per-job one-hot feasibility, generator-ramp coupling between consecutive timeslots, and a carbon-weighted operating cost. The pipeline solves this QUBO via a recursive QAOA-based reduction with three core ingredients: a multi-period formulation that respects DC-OPF ramp constraints, a feasibility-preserving cross-job RQAOA with always-on drain-guard, and a block-symmetric Ring-XY mixer with Dicke W_1 initialization that keeps QAOA dynamics inside the feasible subspace by construction. The pipeline is implemented via a patched build of qiskit-aer 0.17.2 with ARM SVE-512 vectorization on a Fujitsu PRIMEHPC FX700 cluster (A64FX CPU), with distributed statevector simulation up to 32 qubits; 10-seed multi-seed ablation evaluation at 30 and 32 qubits combines Fujitsu FX700 cluster runs (Seed 42) with NVIDIA H100 GPU-supplementary runs (Seeds 100-900). I additionally solve a 60-variable QUBO end-to-end with a ≤ 22-qubit quantum kernel via hybrid SDP-RQAOA reduction. Ablation studies at 30 and 32 qubits identify Dicke W_1 initialization as the dominant factor in feasibility preservation; the constraint-preserving mixer becomes feasibility-critical at 32q; CAISO LMP integration produces the largest measurable business-applicability deviation at each scale. Several auxiliary optimization techniques (warm-start, parameter learning, amplitude amplification) yield smaller but measurable improvements that begin to differentiate at n ≥ 30. Performance is instance-dependent and sensitive to subroutine choice; I frame Q-VOLT as a research testbed for feasibility-preserving hybrid quantum optimization on realistic grid problems, not a deployable production optimizer.