Quantum sensing could measure level leakage with boosted accuracy using AI pulses

Hefei 230026, China Research Institute of Quantum Technology, The Hong Kong Polytechnic University, Hong Kong Level leakage is a dominant error source in quantum state control and requires accurate characterisation for large-scale quantum computing. Quantum sensing offers a viable route for high-precision estimation of level leakage, but its sensitivity can be limited by non-unitary leakage dynamics and interference between the free Hamiltonian and signal generator. A reinforcement learning-based quantum sensing algorithm addresses these challenges for Hermitian three-level and effective dissipative two-level systems.

The algorithm adopts an experimentally feasible pulse parameter optimisation strategy. Research objectives include developing a method for accurately characterising level leakage in quantum systems relevant to large-scale quantum computation, specifically improving the precision of level leakage estimation using quantum sensing techniques. The approach designs a reinforcement learning (RL) based algorithm to optimise quantum sensing protocols. This work focuses on both Hermitian three-level and effective dissipative two-level systems.

An experimentally feasible pulse parameter optimisation strategy forms the core of the RL implementation. This application of reinforcement learning addresses limitations imposed by non-unitary leakage dynamics and Hamiltonian interference within quantum sensing, enhancing sensitivity in estimating level leakage for targeted quantum systems.

Deep Reinforcement Learning Nears Quantum Limits in Leakage Estimation

A classical Fisher Information (CFI) reached close to of the corresponding Quantum Fisher Information (QFI), representing almost tenfold greater precision than previously possible levels in level leakage estimation. Attaining this threshold unlocks accurate characterisation vital for building scalable and reliable quantum computers, something unattainable with earlier methods limited by non-unitary dynamics and signal interference. Deep reinforcement learning discovered experimentally viable pulse sequences to enhance sensing performance across both Hermitian three-level systems and effective dissipative two-level models, utilising joint phase and detuning modulation as key optimisation parameters.

Employing joint phase and detuning modulation adjusts the timing and energy levels of pulses applied to qubits; it serves as a key parameter within an experimental pulse sequence design strategy applicable to both Hermitian three-level systems and effective dissipative two-level models simulating real quantum hardware imperfections. Detailed simulations showed that this algorithm enhanced measurement sensitivity across these model systems, bringing CFI values close to their theoretical upper limits defined by QFI calculations using a finite difference approximation for estimating parameter sensitivities.

Accurately pinning down level leakage, the unwanted escape of a qubit’s state, is vital to building larger, more stable quantum computers, offering a promising new set of tools in that effort through precise characterisation of errors within simple and slightly more complex systems.

Despite acknowledging the substantial computational resources demanded by reinforcement learning approaches which may not immediately scale to the largest contemplated quantum processors, this development establishes a methodology for detailed error characterisation; it provides a means of precisely identifying level leakage in systems beyond those easily assessed by existing techniques. Even small inaccuracies accumulate as qubit numbers grow, making enhanced precision over current methods important when estimating these errors.

By discovering optimal pulse parameters automatically through experimentation, a reinforcement learning algorithm moves beyond traditional control sequences reliant on manual design, characterising level leakage (the loss of qubit state) in both Hermitian three-level systems and effective dissipative two-level models. Achieving measurement precision approaching fundamental limits defined by Quantum Fisher Information demonstrates that machine learning can refine quantum sensing protocols for improved accuracy, essential as quantum computers scale up in complexity.

This research demonstrated a method to precisely characterize level leakage, the unwanted escape of a qubit’s state, using a reinforcement learning algorithm applied to both Hermitian three-level and effective dissipative two-level systems. The algorithm automatically discovers pulse parameters which enhance the sensitivity of quantum sensors, bringing measurement performance close to theoretical limits established by calculations using Quantum Fisher Information. This refined characterization is important because even small inaccuracies accumulate as quantum processors grow more complex. The authors note this approach demands substantial computational resources but establishes a methodology for detailed error assessment beyond current techniques.

👉 More information
🗞 High-precision measurement of level leakage via intelligent coherent control
✍️ Ting Wang, Shi-Qiang Qiao, Xue-Ke Song, Liu Ye and Dong Wang (Anhui University); Hang Xu (Shanghai Jiao Tong University); Haipeng Cao (University of Science and Technology of China); Zhao-Feng Su (The Hong Kong Polytechnic University)
🧠 ArXiv: https://arxiv.org/abs/2611.15031

Stay current

See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.

Avatar of Muhammad Rohail T.

Latest Posts by Muhammad Rohail T.: