Quantum Optics Enables Universal Computation with Cubic and Quartic Phases.

Researchers demonstrated universal quantum computation using continuous variables, achieving 96% success in generating cubic-phase states via reinforcement learning controlled quantum optical circuits. Number-resolving measurements constitute the sole non-Gaussian resource required, enabling direct quartic-phase gate generation without cubic gate decomposition.

The manipulation of quantum states represents a central challenge in the development of photonic quantum computers. Researchers are continually seeking methods to generate complex quantum states with high fidelity, essential for performing advanced quantum computations. A team led by Amanuel Anteneh, Léandre Brunel, Carlos González-Arciniegas, and Olivier Pfister, from the University of Virginia, detail a novel approach utilising deep reinforcement learning to prepare cubic- and quartic-phase gates – key components in continuous-variable quantum computing – within a photonic circuit. Their work, entitled ‘Deep reinforcement learning for near-deterministic preparation of cubic- and quartic-phase gates in photonic quantum computing’, demonstrates an average success rate of 96% in generating these states, relying solely on number-resolving measurements as a non-Gaussian resource.

Reinforcement Learning Facilitates Efficient Generation of Non-Classical Quantum States

Researchers have demonstrated the efficient generation of cubic and quartic phase states – essential resources for universal continuous-variable quantum computation – by employing deep reinforcement learning to control a quantum optical circuit. The study reports a 96% average success rate in generating cubic phase states, utilising number-resolving measurements as the sole non-Gaussian resource.

The investigation centres on manipulating photonic states to create highly entangled cluster states, fundamental to measurement-based quantum computation. This approach to quantum computation relies on creating a large, entangled multi-photon state, and then performing a series of measurements to drive the computation forward. A deep neural network, trained via reinforcement learning, actively optimises the parameters of the quantum optical circuit, directly addressing the challenge of generating the necessary entanglement for high-fidelity state preparation.

The system leverages number-resolving detectors – devices that precisely measure the number of photons in a given quantum state – to perform measurements on photons, driving the quantum computation forward. Researchers meticulously analyse the generated states using Wigner functions, a quasi-probability distribution that visualises quantum states in phase space, to verify their properties and quantify their fidelity – a measure of how closely the generated state matches the intended state. Furthermore, they assess the quarticity of the states, a metric indicating the degree of non-classicality and entanglement present. Results demonstrate that the choice of measurement settings significantly influences both fidelity and quarticity.

The study establishes that the same resources – number-resolving measurements and reinforcement learning-optimised control – enable the direct generation of a quartic-phase gate. This is significant because it eliminates the necessity for decomposing the quartic gate – a fundamental operation in quantum computation – into a series of cubic gates, a process that typically adds complexity and reduces efficiency. This innovative approach streamlines quantum circuit design and potentially improves computational efficiency.

The convergence of machine learning and quantum optics suggests a promising avenue for overcoming the challenges associated with building and controlling complex quantum systems. By allowing algorithms to learn optimal control strategies, researchers can potentially automate the process of quantum circuit design and optimisation. This acceleration of development promises to bring practical quantum technologies closer to realisation. The emphasis on number-resolving measurements as the sole non-Gaussian resource indicates a pathway toward building more resource-efficient quantum computers based on continuous variables. This research contributes to the growing body of evidence supporting the feasibility of continuous-variable quantum computation, and future work will likely focus on scaling these techniques to larger, more complex quantum circuits.