Structured Quantum Optimal Control under Bandwidth and Smoothness Constraints-An Inexact Proximal-ADMM Approach for Low-Complexity Pulse Synthesis

TL;DR

This paper introduces an inexact Proximal-ADMM method for quantum pulse synthesis that incorporates bandwidth, smoothness, and sparsity constraints, aiming to explore low-complexity control frontiers rather than achieving absolute fidelity.

Contribution

It presents a novel structured-control optimization framework combining multiple constraints within a single loop, benchmarking against standard methods and analyzing its role in low-complexity frontier exploration.

Findings

Method stabilizes a low-complexity fidelity frontier.

Structured fidelities reach 0.6672 for qutrits and 0.6342 for two-qubits.

Lower complexity solutions compared to unconstrained quasi-Newton methods.

Abstract

Quantum optimal control is often judged by nominal fidelity alone, even though realistic pulse-design studies must also account for bandwidth, amplitude, and smoothness constraints. I study this structured-control regime with an inexact Proximal-ADMM framework that combines gate-infidelity minimization with $L_1$ sparsity, total-variation regularization, explicit band-limit projection, and box constraints in a single loop. The method is benchmarked against GRAPE, standard Krotov, and L-BFGS-B on a single-qubit $X$ gate, a leakage-prone qutrit task, and a two-qubit entangling gate. Across ten random seeds, Pareto scans, ablations, filtered-baseline fairness checks, significance analysis with false-discovery-rate correction, and robustness tests, the method is not a universal winner in either nominal fidelity or wall-clock cost. Its value is instead to expose and stabilize a low-complexity frontier of the fidelity-complexity landscape. After retuning the PADMM budgets and warm-start lengths, the qutrit and two-qubit structured fidelities rise to 0.6672 +- 0.0001 and 0.6342 +- 0.0003, respectively, while preserving markedly lower complexity than unconstrained quasi-Newton solutions. These values remain well below deployment-grade gate thresholds, so the contribution should still be read as a numerical framework for constrained pulse synthesis rather than as a finished route to immediately deployable high-fidelity gates. Training-time robust optimization yields only task-dependent gains, with the clearest effect appearing in qutrit drift robustness and amounting to a small absolute improvement. The results therefore position PADMM as a constraint-native framework for low-complexity frontier exploration, not as a replacement for unconstrained high-fidelity solvers.