Reinforcement Learning for Digital Quantum Simulation.

Area of Science:

• Quantum Information Science
• Computational Physics
• Machine Learning

Background:

• Digital quantum simulation aims to model complex many-body systems using quantum computers.
• Experimental limitations, such as noise and gate fidelity, restrict the scale and duration of achievable simulations.
• Optimizing quantum circuits is crucial for overcoming these limitations in noisy intermediate-scale quantum (NISQ) devices.

Purpose of the Study:

• To develop a reinforcement learning algorithm for systematically constructing optimized quantum circuits for digital quantum simulation.
• To address the challenge of limited quantum gate counts due to experimental imperfections.
• To enable accurate simulations of larger quantum systems and longer evolution times on NISQ hardware.

Main Methods:

• A reinforcement learning algorithm was designed to generate optimized quantum circuits under strict gate count constraints.
• The algorithm was applied to construct circuits for simulating the unitary evolution of many-body Hamiltonians.
• The performance of the generated circuits was evaluated by reproducing physical observables.

Main Results:

• The reinforcement learning approach consistently produced optimized quantum circuits capable of accurate simulations.
• Remarkably, circuits with as few as three entangling gates were sufficient for simulating large systems (up to 16 qubits) and long times.
• The method was successfully applied to simulate a long-range Ising chain and the lattice Schwinger model.

Conclusions:

• Reinforcement learning offers a powerful strategy for engineering efficient quantum circuits for digital quantum simulation.
• The developed method significantly enhances the capabilities of NISQ devices by pushing the boundaries of achievable system sizes and simulation times.
• This work demonstrates a pathway to more scalable and accurate quantum simulations leveraging current experimental technology.