Optimized continuous dynamical decoupling via differential geometry and machine learning

We introduce a strategy to develop optimally designed fields for continuous dynamical decoupling. Using our methodology, we obtain the optimal continuous field configuration to maximize the fidelity of a general one-qubit quantum gate. To achieve this, considering dephasing-noise perturbations, we employ an auxiliary qubit instead of the boson bath to implement a purification scheme, which results in unitary dynamics. Employing the sub-Riemannian geometry framework for the two-qubit unitary group, we derive and numerically solve the geodesic equations, obtaining the optimal time-dependent control Hamiltonian. Also, due to the extended time required to find solutions to the geodesic equations, we train a neural network on a subset of geodesic solutions, enabling us to promptly generate the time-dependent control Hamiltonian for any desired gate, which is crucial in circuit optimization.