Machine learning for qutrit-based quantum computing and simulation with Rydberg atoms

• Applicants: Professorin Dr. Monika Aidelsburger; Professorin Dr. Annabelle Bohrdt
• Subject Area: Optics, Quantum Optics and Physics of Atoms, Molecules and Plasmas ; Theoretical Condensed Matter Physics
• Term: since 2026
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 550495627

The project applies machine-learning techniques to optimize single- and two-qutrit gates in neutral atom tweezer arrays. We employ Bayesian optimization and reinforcement learning, starting with numerical simulations that incorporate experimentally benchmarked decoherence and loss channels for Ytterbium atoms in optical arrays. Based on these results, we will target experimental implementations to demonstrate increased fidelities and enhanced robustness of various gate operations. While single-atom operations can be well controlled in this way, optimizing two-atom interactions for digital and analog simulation will require more sophisticated control protocols. We will include both Rydberg-blockade and resonant dipole-dipole interactions to optimize control of the qutrit subspace for state preparation and readout. Specific protocols will be developed for applications such as the t-j model, relevant for studying unconventional superconductivity. Our control strategies will also support a broader range of applications involving qutrit subspaces.

• DFG Programme: Research Units
• Subproject of: FOR 5919: Machine learning for complex quantum states
• Co-Investigators: Dr. Marin Bukov; Professor Dr. Martin Gärttner