⚛️ quantum physics

Optimal Hamiltonian recognition of unknown quantum dynamics

This paper introduces an optimal quantum algorithm for recognizing unknown Hamiltonians from a known set by leveraging quantum signal processing and semidefinite optimization to achieve the best possible average success probability with limited queries, a method validated on superconducting quantum processors and extended to multi-qubit systems.

Original authors: Chengkai Zhu, Shuyu He, Yu-Ao Chen, Lei Zhang, Xin Wang

Published 2026-02-26

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Chengkai Zhu, Shuyu He, Yu-Ao Chen, Lei Zhang, Xin Wang

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine you are a detective trying to solve a mystery, but you can't see the culprit. You only have a "black box" that performs a specific magic trick every time you ask it to. Your goal is to figure out which magic trick the box is doing, but there's a catch: you don't know when the trick happens or how long it lasts.

This is the core problem of Hamiltonian Recognition, a new method developed by researchers at the Hong Kong University of Science and Technology (Guangzhou).

Here is a simple breakdown of their breakthrough, using everyday analogies.

1. The Mystery: The "Black Box" of Physics

In the quantum world, everything evolves according to a rule called a Hamiltonian. Think of a Hamiltonian as the "recipe" or the "script" that tells a quantum particle how to move and change over time.

Usually, scientists try to figure out the entire recipe by watching the particle for a long time (this is called "learning"). But sometimes, you don't need the whole recipe. You just need to know: "Is the particle following Recipe A or Recipe B?"

Recipe A (Hamiltonian X): The particle spins like a top on a table.
Recipe B (Hamiltonian Z): The particle spins like a coin flipping in the air.

The problem is that you don't know how long the particle has been spinning (the time parameter $\theta$ ). It could be a split second or a full minute. You just know it's one of the two recipes.

2. The Old Way vs. The New Way

The Old Way (Quantum Tomography):
Imagine trying to guess a song by listening to a single, random 5-second clip. If you don't know the tempo, it's incredibly hard. You might need to listen to thousands of clips to be sure. This is slow and inefficient.

The New Way (Hamiltonian Recognition):
The researchers created a "super-listener" algorithm. Instead of just listening passively, they actively interact with the black box in a very specific, rhythmic pattern.

They use a technique called Quantum Signal Processing (QSP).

The Analogy: Imagine you are trying to identify a specific drumbeat in a noisy room. Instead of just standing there, you start tapping your own stick in a complex, mathematical rhythm that matches the drum. If the drum is playing the "X" beat, your tapping creates a perfect harmony. If it's the "Z" beat, your tapping creates a specific, predictable dissonance.
By tapping (querying) the black box $k$ times, the algorithm amplifies the signal of the correct recipe until it's impossible to miss.

3. The Magic Trick: "Perfect Discrimination"

The most mind-blowing part of this paper is what they discovered about entanglement (a spooky connection between particles).

Usually, in quantum physics, to solve hard problems, you need to use entangled particles (like having two detectives who can instantly talk to each other across the room). It's expensive and hard to do.

The Discovery:
The researchers proved that for this specific task, you don't need entanglement at all.

They showed that even if you use just a single particle and ask the black box questions one by one (sequentially), you can perfectly distinguish between two completely different "recipes" (like X and Z) after a finite number of tries.
It's like being able to tell the difference between a violin and a flute just by tapping your foot in a specific rhythm, without needing a second instrument to help you.

4. How Good Is It? (The Speed)

The researchers proved their method is mathematically optimal.

If you ask the black box $k$ times, your chance of making a mistake drops by a factor of $1/k$ .
The Metaphor: If you ask once, you might be wrong 50% of the time. If you ask 10 times, you are wrong only 10% of the time. If you ask 100 times, you are wrong only 1% of the time. You can't do better than this; it's the fastest possible speed allowed by the laws of physics.

5. Real-World Testing

The team didn't just do this on paper. They took their algorithm to a real quantum computer (a superconducting processor made by Tencent).

They fed the computer unknown "recipes" (rotations of X and Z).
The computer successfully identified the recipe with high accuracy, confirming that the math works in the real, noisy world.

6. Why Does This Matter?

This isn't just about guessing a recipe. It's a fundamental building block for the future of quantum technology:

Better Sensors: It helps us build sensors that can detect tiny changes in magnetic fields or gravity with extreme precision.
Faster Diagnostics: If a quantum computer is acting weird, this method can quickly tell you what is wrong without needing to fully map out the entire system.
Efficiency: It proves we can get the most information out of the fewest number of queries, saving time and energy.

Summary

Think of this paper as the invention of a super-efficient "tuning fork" for the quantum world. Instead of needing a massive orchestra (entanglement) to identify a sound, this new method uses a single, perfectly timed tap to instantly know exactly which song the universe is playing. It bridges the gap between guessing a hypothesis and measuring a value, making quantum technology faster, cheaper, and more reliable.

1. Problem Definition

The paper addresses the challenge of Hamiltonian recognition, a task situated at the intersection of quantum hypothesis testing and quantum metrology.

Goal: Identify an unknown Hamiltonian $H$ governing a quantum evolution $U_H(\theta) = \exp(-iH\theta)$ from a known finite set of candidate Hamiltonians $\{H_0, H_1, \dots, H_{m-1}\}$ .
Constraints:
- The parameter $\theta$ (evolution time) is unknown and can vary arbitrarily within a range $[0, t]$ .
- The protocol must be time-independent (single-shot measurement) and cannot rely on controlling $\theta$ .
- The objective is to maximize the average success probability over all possible Hamiltonians in the set and all possible values of $\theta$ .
Distinction: Unlike standard Hamiltonian learning (which estimates parameters of a known model) or certification, this task is a "composite hypothesis testing" problem where the hypothesis space includes continuous parameters ( $\theta$ ).

2. Methodology

The authors develop a framework combining Quantum Signal Processing (QSP), Semidefinite Programming (SDP), and Group Representation Theory.

A. Theoretical Framework

Tester Formalism: The problem is modeled using quantum testers (generalized POVMs for channels). The success probability is formulated as an optimization over $k$ -slot testers (parallel, sequential, or general strategies with indefinite causal order).
Performance Operator: The average Choi operator over the unknown parameter $\theta$ is defined as $\Omega^{(k)}_H(t) = \frac{1}{t}\int_0^t |U_H(\theta)\rangle\rangle\langle\langle U_H(\theta)|^{\otimes k} d\theta$ .
Optimality Proof: The authors use SDP duality to derive upper bounds on the success probability. By constructing feasible dual solutions using group representation theory, they prove that their proposed protocols achieve these bounds, establishing global optimality.

B. Quantum Algorithms

The core of the solution relies on Quantum Signal Processing (QSP) to construct specific polynomial functions of the unknown parameter $\theta$ .

Binary Recognition (Two Orthogonal Hamiltonians):
- Scenario: Distinguishing between two orthogonal single-qubit Hamiltonians (e.g., $X$ and $Z$ ).
- Protocol: A sequential QSP circuit applies a sequence of rotation gates $e^{iZ\phi_j}$ interleaved with the unknown unitary $e^{-iH\theta}$ .
- Mechanism: The QSP phase sequence $\vec{\phi}$ $ϕ$ is tuned such that the resulting measurement probability approximates a specific polynomial $f_k(\theta)$ $f_{k} (θ)$ .
  - If $H=Z$ , the error is zero (perfect discrimination).
  - If $H=X$ , the error probability follows a polynomial that averages to $O(1/k)$ .
- Result: The protocol achieves an average success probability of $\frac{k + \max(p, 1-p)}{k+1}$ , where $p$ is the prior probability.
Ternary Recognition (Three Orthogonal Hamiltonians):
- Scenario: Distinguishing between three mutually orthogonal Hamiltonians (e.g., $X, Y, Z$ ).
- Protocol: A coherent quantum algorithm utilizing two QSP circuits simultaneously.
- Mechanism: The circuit inputs a state $|000\rangle$ and applies the unknown unitary interleaved with rotations for both $\{X, Z\}$ and $\{Y, Z\}$ recognition tasks. A specific unitary $U(\phi_x, \phi_y)$ and a Toffoli gate are used to amplify the success probability for $X$ and $Y$ cases while maintaining zero error for $Z$ .
- Result: For odd $k$ , the optimal average success probability is $\frac{k + \max(p_0, p_1, p_2)}{k+1}$ .
General and Multi-Qubit Extensions:
- Non-Orthogonal Axes: Numerical SDP solutions show that the $\{X, Z\}$ protocol remains effective even when axes are not perfectly orthogonal, though optimality requires specific adaptations.
- Multi-Qubit Systems: A variational quantum algorithm is proposed for distinguishing Heisenberg-type Hamiltonians in 2- and 3-qubit systems, demonstrating that success probability scales with query count $k$ .

3. Key Contributions

Optimal Scaling: Proved that the average error for recognizing orthogonal Hamiltonians decays as $O(1/k)$ with $k$ queries. This scaling is proven to be globally optimal for any strategy (parallel, sequential, or indefinite causal order).
Entanglement Not Required: Demonstrated that for the optimal recognition of orthogonal single-qubit Hamiltonians, entanglement provides no advantage. A sequential protocol using single-qubit gates suffices to reach the theoretical limit.
Perfect Discrimination of Sets: Revealed a counter-intuitive result (Corollary 2): For any $k$ , there exist two disjoint sets of $k$ unitary operations (where elements within sets are not necessarily orthogonal) that can be perfectly distinguished using only $k$ adaptive queries without entanglement.
Experimental Validation: Successfully implemented the binary recognition protocol on a Tencent Surface-13 superconducting quantum processor, observing the predicted error decay and confirming the protocol's robustness against realistic noise.

4. Key Results

Theorem 1 (Binary): For orthogonal $H_0, H_1$ , the optimal success probability is $Suc = \frac{k + \max(p, 1-p)}{k+1}$ .
Theorem 3 (Ternary): For mutually orthogonal $H_0, H_1, H_2$ , the optimal success probability is $Suc = \frac{k + \max(p_0, p_1, p_2)}{k+1}$ (for odd $k$ ).
Error Decay: The average error scales as $1/(k+1)$ . The variance of the success probability also vanishes rapidly as $k$ increases.
Experimental Data: On the superconducting processor, increasing the number of queries $k$ from 1 to higher values significantly reduced misclassification rates between $X$ and $Z$ evolutions, matching theoretical predictions.

5. Significance

Bridging Fields: The work unifies concepts from quantum metrology (parameter estimation limits) and composite hypothesis testing (channel discrimination).
Efficiency: It provides an efficient method to identify Hamiltonians from limited dynamic queries, which is crucial for quantum device verification and characterizing unknown quantum systems.
Foundational Insight: It challenges the assumption that entanglement is always necessary for optimal quantum discrimination, showing that for specific composite tasks, sequential single-qubit strategies are sufficient and optimal.
Future Directions: The paper opens avenues for extending these QSP-based recognition methods to many-body systems and non-orthogonal Hamiltonian sets, potentially leading to new quantum algorithms for system identification.