Experimental semi-autonomous eigensolver using reinforcement learning

Area of Science:

• Quantum Computing
• Quantum Mechanics
• Artificial Intelligence

Background:

• Characterizing observables via Hermitian operators is essential in quantum mechanics.
• Eigensolvers are fundamental algorithms for quantum technologies.
• Current quantum devices require efficient algorithms for practical applications.

Purpose of the Study:

• To implement a semi-autonomous algorithm for approximating eigenvectors of Hermitian operators.
• To utilize single-shot measurements and a feedback loop for reduced resource demand.
• To explore the application of reinforcement learning in quantum eigensolving.

Main Methods:

• Implementation of a semi-autonomous eigensolver algorithm on an IBM quantum computer.
• Utilizing single-shot measurements and a classical feedback loop for pseudo-random adjustments.
• Framing the algorithm within the reinforcement learning paradigm.

Main Results:

• Achieved fidelities over 0.97 for single-qubit observable eigenvectors using ~200 measurements.
• Obtained fidelities over 0.91 for two-qubit observable eigenvectors using ~1500 measurements.
• Demonstrated low resource demand suitable for current quantum computing devices.

Conclusions:

• The developed algorithm efficiently approximates eigenvectors with high fidelity.
• This approach reduces the number of measurements required, making it practical for current quantum hardware.
• The work contributes to quantum devices capable of decision-making with partial information, supporting quantum artificial intelligence.