QMaxCal: Path-Space Regularization for Open Quantum Control via Girsanov's Theorem

A new quantum-control technique, QMaxCal, applies Girsanov's theorem to reduce decoherence in open quantum systems, according to a preprint posted to arXiv on 18 June 2026 [1]. The method introduces two path-space regularizers that penalize the observable consequences of control on noise channels rather than the control amplitude itself [2]. The work targets a central obstacle in quantum computing: environmental noise that destroys quantum information. Open quantum systems under continuous monitoring produce classical measurement records whose drift reflects the noise experienced by the system. Because two evolutions sharing the same decoherence channels differ only in this drift, Girsanov's theorem provides a closed-form, differentiable estimator of the Kullback-Leibler divergence between their trajectory distributions [2]. The authors instantiate this estimator with two physically motivated reference measures, yielding two regularizers: the Wiener KL (KL_W), which is empirically more effective under certain noise-model conditions, and the drift-variance regularizer (R_DV), which works for all noise models [2]. Both regularizers are qualitatively distinct from existing penalties on control fluence or smoothness. They drive the system toward states where the effects of decoherence are minimal by penalizing the observable consequences of control on the decoherence channels rather than the control amplitude itself [2]. The researchers evaluated QMaxCal against unregularized gradient-based and reinforcement-learning baselines across single- and multi-qubit benchmarks and a multi-qubit chain calibrated to a published snapshot of the IBM Kingston processor [2]. On the calibrated IBM Kingston chain, the regularizers delivered gains of roughly 16% [2]. Across the broader test suite, infidelity was reduced by up to 50% [2]. The method also showed growing robustness to noise-model mismatch: gains increased from 17 percentage points at training noise to 27 percentage points under a 2.5x noise mismatch [2]. The preprint, hosted on the arXiv e-print repository — which as of November 2024 receives about 24,000 submissions per month — has not yet been peer-reviewed [6].