Quantum Neyman-Pearson test for Identifying Quantum Phases
Keio University Quantum Neyman-Pearson test
The quantum Neyman-Pearson test was used to create a new quantum phase classification approach. This strategy solves the disadvantages of machine learning and order parameter approaches, which sometimes require too much quantum data or prior information. By splitting subsystems and applying hypothesis testing, the authors avoid comprehensive state tomography's computational problems. Their numerical simulations show that this technique is very scalable and provides improved accuracy with fewer state copies. Using quantum measurements and standard post-processing, the research finds a better way to determine quantum phases.
The Many-Body Physics Challenge
Quantum phase classification is a fundamental many-body physics problem. Quantum phase transitions occur at absolute zero and are caused by external variables like pressure or magnetic fields, unlike classical systems where thermal fluctuations cause changes. Scientists have traditionally used order parameters, local observables that signal symmetry-breaking, or advanced quantum machine learning like Quantum Convolutional Neural Networks.
These established approaches are costly. QCNNs require many quantum state copies for training and reliable classification, while traditional order parameters may require extensive system expertise. As topological phases do not break local symmetries or have local order factors, they are defined by global properties like Chern numbers, making their definition unique.
An “Optimal” Solution
Akira Tanji, Hiroshi Yano, and Naoki Yamamoto used the quantum Neyman-Pearson test for statistical inference at Keio. This test is considered the best way to distinguish two quantum states since it optimizes the likelihood of a correct choice while allowing for errors.
Although powerful, the Neyman-Pearson test was deemed intractable for large systems. Complete state tomography tests demand double the computer resources for each qubit due to the “exponential growth” of Hilbert space.
To avoid the "curse of dimensionality," the researchers used partial tomography for partitioning. Separate groupings of qubits are examined instead of the complete quantum state. It then uses a majority vote to classify these subsystems, or reduced density matrices (RDMs), using the Neyman-Pearson test.
Outperforming AI
Computer simulations revealed shocking results. Novel method had lower classification error probabilities than QCNN in head-to-head comparisons.
In particular, the method was more resource-efficient. The QCNN needed less than a thousandth of the training data copies to reduce validation loss in a 15-qubit test. Existing quantum machine learning models employ expensive gradient-based variational learning, whereas the innovative method does not.
The researchers also showed that their method has a lower classical computational time complexity than the low-weight QCNN. The Keio method uses linear time with regard to system size, while the latter uses polynomial-log time.
Up to Quantum Limit Scaling
The researchers tested their method on the one-dimensional cluster-Ising model and the two-dimensional Toric code Hamiltonian. Ferromagnetic, antiferromagnetic, trivial, and symmetry-protected topological (SPT) phases were classified by the program.
Scalability was the method's most notable feature. The team verified high accuracy in systems up to 81 qubits, when full-state analysis is not possible. The researchers discovered that as system size increased, quantum phase transitions became “clearer in larger spin chains,” decreasing mistake probability.
The Future of Quantum Sensors
This innovation is crucial for experiments that combine quantum observations with traditional post-processing. The approach could be used on near-term quantum hardware (NISQ devices) for quantum communication and sensing by requiring fewer physical copies of a quantum state. “These findings underscore the promise of quantum hypothesis testing as a strong tool,” the scientists said, adding that it can overcome barren plateaus that hinder quantum machine learning.
In the future, the team plans to test the method on more complex phases like long-range entanglement or topological ordering. Now, scientists may analyze quantum alterations using a new, high-speed lens.