Optimal Hamiltonian recognition of unknown quantum dynamics

Chengkai Zhu, Shuyu He, Yu-Ao Chen, Lei Zhang, Xin WangNpj 2026 · DOI

Abstract

Identifying unknown Hamiltonians from their quantum dynamics is a pivotal challenge in quantum technologies. In this paper, we introduce Hamiltonian recognition, a framework that bridges quantum hypothesis testing and quantum metrology, aiming to identify the Hamiltonian governing quantum dynamics from a known set of Hamiltonians. To identify $H$ for an unknown qubit quantum evolution $\exp(-iH\theta)$ with unknown $\theta$, from two or three orthogonal Hamiltonians, we develop a quantum algorithm for coherent function simulation, built on two quantum signal processing (QSP) structures. It can simultaneously realize a target polynomial based on measurement results regardless of the chosen signal unitary for the QSP. Utilizing semidefinite optimization and group representation theory, we prove that our methods achieve the optimal average success probability, taken over possible Hamiltonians $H$ and parameters $\theta$, decays as $O(1/k)$ with $k$ queries of the unknown unitary transformation. Furthermore, we demonstrate the validity of our protocol on a superconducting quantum processor. We also investigate a physically motivated recognition task for Heisenberg Hamiltonians, providing numerical evidence for effective multi-qubit quantum system recognition. This work presents an efficient method to recognize Hamiltonians from limited queries of the dynamics, opening new avenues in composite channel discrimination and quantum metrology.

Introduction

In quantum information science, characterizing the underlying Hamiltonian $H$ of a system is essential for building reliable quantum devices. Traditional approaches typically fall into two categories: quantum metrology, where the Hamiltonian is known and a parameter $\theta$ (such as time or field strength) is estimated, and Hamiltonian learning, which aims to reconstruct the entire Hamiltonian from scratch.

This work introduces an intermediate regime termed Hamiltonian recognition. In this framework, the goal is to distinguish between a discrete set of candidate Hamiltonians (e.g., $H \in \{X, Z\}$) when the evolution parameter $\theta$ is unknown. This constitutes a composite hypothesis testing problem, as each candidate Hamiltonian generates a continuous set of possible unitary transformations $U = e^{-iH\theta}$. The challenge is to develop a protocol that can distinguish between these sets of unitaries with maximum accuracy using a limited number of queries $k$.

Methods: Quantum Signal Processing for Recognition

The core of the proposed recognition protocol is Quantum Signal Processing (QSP). QSP provides a powerful framework for transforming an unknown unitary into a desired polynomial function of its eigenvalues. By carefully designing a sequence of signal-processing rotations, the authors construct a quantum circuit that amplifies the signature of the target Hamiltonian while remaining invariant to the unknown parameter $\theta$.

Quantum circuit for Hamiltonian recognition using QSP, showing the sequential application of signal unitaries and processing gates.

For binary recognition, the protocol uses a sequential strategy where the unknown unitary is interleaved with fixed $R_z$ or $R_x$ rotations. For more complex ternary recognition (choosing between three candidates), the researchers developed a "coherent function simulation" method. This involves a multi-qubit circuit that executes different QSP sequences in superposition, allowing for the simultaneous testing of multiple hypotheses. The optimality of these protocols was rigorously established using semidefinite programming (SDP) and group representation theory, showing that the QSP-based approach reaches the fundamental theoretical limits of success probability.

Key Results and Scaling

A primary finding of this research is the optimal scaling of the recognition error. The average success probability for binary recognition of orthogonal Hamiltonians is shown to be $(k + \max\{p, 1-p\}) / (k+1)$, where $p$ is the prior probability. This implies that the error in identifying the correct Hamiltonian decays as $O(1/k)$ with the number of queries $k$.

Performance of the Hamiltonian recognition protocol across different query numbers $k$ and parameter values $\theta$.

Experimental validation on a 13-qubit superconducting processor confirmed these theoretical predictions. Despite the presence of hardware noise, the observed error decay followed the $1/k$ trend across a wide range of sampled $\theta$ values. Furthermore, numerical simulations for multi-qubit Heisenberg Hamiltonians using variational quantum circuits demonstrated that the framework can be extended to complex many-body systems, maintaining high success probabilities even in higher-dimensional Hilbert spaces.

Conclusion

Hamiltonian recognition provides a robust and efficient bridge between quantum metrology and hypothesis testing. By leveraging the flexibility of QSP, this work establishes a fundamental limit on how accurately we can identify quantum dynamics from limited queries. The $1/k$ scaling of the error highlights the power of coherent signal processing in quantum sensing. Future research will focus on extending these analytical results to non-orthogonal Hamiltonians and investigating the impact of specific noise models on the protocol's performance in large-scale quantum systems.

Summary updated

Abstract

Identifying unknown Hamiltonians from their quantum dynamics is a pivotal challenge in quantum technologies. In this paper, we introduce Hamiltonian recognition, a framework that bridges quantum hypothesis testing and quantum metrology, aiming to identify the Hamiltonian governing quantum dynamics from a known set of Hamiltonians. To identify $H$ for an unknown qubit quantum evolution $\exp(-iHθ)$ with unknown $θ$, from two or three orthogonal Hamiltonians, we develop a quantum algorithm for coherent function simulation, built on two quantum signal processing (QSP) structures. It can simultaneously realize a target polynomial based on measurement results regardless of the chosen signal unitary for the QSP. Utilizing semidefinite optimization and group representation theory, we prove that our methods achieve the optimal average success probability, taken over possible Hamiltonians $H$ and parameters $θ$, decays as $O(1/k)$ with $k$ queries of the unknown unitary transformation. Furthermore, we demonstrate the validity of our protocol on a superconducting quantum processor. We also investigate a physically motivated recognition task for Heisenberg Hamiltonians, providing numerical evidence for effective multi-qubit quantum system recognition. This work presents an efficient method to recognize Hamiltonians from limited queries of the dynamics, opening new avenues in composite channel discrimination and quantum metrology.
BibTeX
@inproceedings{zhu2026optimal,
  author = {Chengkai Zhu and Shuyu He and Yu-Ao Chen and Lei Zhang and Xin Wang},
  title = {{Optimal Hamiltonian recognition of unknown quantum dynamics}},
  booktitle = {Npj},
  year = {2026},
  doi = {10.1038/s41534-026-01182-6},
}
Figure