The question of which quantum entangled states exhibit “steering,” a form of quantum correlation, remains a fundamental question in quantum information theory, and researchers are now presenting a new approach to tackling the problem. Yanning Jia, Fenzhuo Guo, and Mengyan Li from Beijing University of Posts and Telecommunications, along with Haifeng Dong and Fei Gao from Beihang University, will develop a framework to determine whether a particular entangled state can be described by a “local hidden state” model, effectively testing its maneuverability. Their method uses machine learning to efficiently sample measurements and tune parameters to build optimal models, and the team demonstrates its effectiveness by accurately evaluating the operability of various quantum states. In addition to confirming existing analytical results for specific states, this study also suggests that carefully selected measurements may reveal maneuverability that may remain hidden, providing a major advance in understanding and exploiting this subtle quantum phenomenon.
Researchers propose a machine learning-based framework that employs batch sampling of measurements and gradient-based optimization to build an optimal local hidden state (LHS) model. This could enable efficient testing of quantum steering criteria and accelerate advances in quantum communications and computation.
Proof of quantum nonstability using LHV model
Scientists have developed a machine learning framework to determine the controllability of entangled states, an important step in advancing quantum information science. This study addresses the challenge of verifying whether certain entangled states can be described by a local hidden state (LHS) model that is not manipulable. This new approach utilizes batch sampling of measurements and gradient-based optimization to build an optimal LHS model and effectively navigate the complex measurement space. The method involves building an LHV model that mimics the behavior of the quantum state, reparameterizing the parameters to ensure a physically meaningful representation, and defining a loss function based on the trace distance between the LHV model's predictions and the quantum state.
Gradient descent optimization iteratively adjusts the model's parameters to minimize the loss function, and by demonstrating that the LHV model perfectly reproduces the quantum state, convergence to zero proves that the state is inoperable. The experiments focused on a two-qubit Werner state and a two-qubit isotropic state to verify the performance of the method. For the Werner state, the team achieved results that saturate the known analytical visibility limits under three Pauli measurements, an arbitrary projection measurement (PVM), and an arbitrary positive operator value measurement (POVM). This means that the model accurately predicts the maneuverability limits of these conditions across a wide range of measurement types. For the isotropic state, this study was successful in matching the analytical limits established when using any PVM.
This breakthrough does more than simply match existing analytical results. Scientists have investigated the operability of isotropic states under arbitrary POVMs, but the exact analytical limits are currently unknown. Measurements confirm that the critical visibility of steerability is lower using POVM compared to PVM, suggesting that POVM can more effectively reveal the steerable nature of these states. This innovative approach provides a powerful tool to overcome the limitations of previous numerical methods, characterize quantum stability, and advance quantum information processing.
Stability verification using machine learning optimization
This work presents a new machine learning framework for determining the controllability of quantum states, an important aspect in understanding entanglement. Scientists have developed a method to efficiently sample measurements, build optimal local hidden state models by employing gradient-based optimization techniques, and effectively test whether a particular entangled state can be explained by classical hidden variables. The team successfully applied this approach to the analysis of the two-qubit Werner state and the two-qubit isotropic state, achieving analytical limits for stability in different measurement types. Results show that the framework accurately identifies steerable states, matches known analysis limits for projective measurements, and extends the analysis to more general positive operator value measurements. Importantly, our results suggest that employing these more general measurements can reveal the stability of potentially overlooked states, indicating potential benefits in characterizing quantum entanglement. This framework employs an iterative process of sampling measurements, expressing response functions, and constructing hidden states to minimize the trace distance between the LHS and the quantum ensemble to determine stability.
Machine learning reveals stability of quantum states
Scientists have developed a method to efficiently sample measurements, build optimal local hidden state models by employing gradient-based optimization techniques, and effectively test whether a particular entangled state can be explained by classical hidden variables. The team successfully applied this approach to the analysis of the two-qubit Werner state and the two-qubit isotropic state, achieving analytical limits for stability in different measurement types. Results show that the framework accurately identifies steerable states, matches known analysis limits for projective measurements, and extends the analysis to more general positive operator value measurements. This finding suggests that employing these more general measurements can reveal the stability of states that might otherwise be overlooked, demonstrating potential benefits in characterizing quantum entanglement.
👉 More information
🗞 A general framework for building local hidden state models for determining stability
🧠ArXiv: https://arxiv.org/abs/2512.21848
