Machine Learning-Driven Modeling and Characterization of High-Fidelity Quantum Control in Rydberg Atom Systems

An optical tweezers-trapped array of Rydberg atoms provides an intriguing platform for realising quantum computing. Recent proposals and experimental demonstrations have shown high-fidelity gates within Rydberg atom systems. Despite these significant strides, achieving fault-tolerant computation demands further enhancements in gate fidelity. Various error sources, including phase and intensity fluctuations in lasers, alignment errors, optical potential variations, finite lifetime of Rydberg atoms, and significant atomic motion during gate operations, must be considered to boost fidelity.

To address this problem we have created a digital twin of the device in question, implementing a very comprehensive model of the system. We employ automatic differentiation, ensemble averaging, and adjoint equation-based approaches. These techniques enable us to develop protocols with state-of-the-art fidelities that are robust against the aforementioned sources of noise. Additionally, we characterise these noise sources to better understand their impact on fidelity and offer a comprehensive error budget for the optimised protocols, providing valuable guidance for their experimental implementation.

Furthermore, the versatile toolbox we have developed can readily be applied to analogous challenges in other qubit modalities, such as trapped ions.