🔢 mathematics

Emergent Prethermal Symmetries for Scalable Hamiltonian Learning

This paper introduces a scalable Hamiltonian learning protocol for geometrically local many-body systems that achieves Heisenberg-limited precision using only static single-qubit fields to generate emergent prethermal symmetries, thereby eliminating the need for trusted multi-qubit operations or rapid control pulses.

Original authors: Myeongjin Shin, Junseo Lee, Iman Marvian, Yu Tong

Published 2026-09-24

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Myeongjin Shin, Junseo Lee, Iman Marvian, Yu Tong

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

In the quantum world, the rules of interaction are written in a mathematical blueprint called a Hamiltonian. This blueprint dictates how the tiny particles within a system push and pull on one another, determining everything from how a material conducts electricity to how a quantum computer processes information. To understand a quantum system, scientists must first read this blueprint. However, reading it is notoriously difficult. As a quantum system evolves over time, its internal parts become so deeply entangled that local information—the clues needed to decipher the blueprint—fades away into a chaotic blur. This process, known as thermalization, is like a cup of hot coffee cooling down until it matches the room temperature; the specific history of how it got hot is lost. To learn the blueprint before this information disappears, scientists have traditionally relied on rapid, complex control pulses or trusted multi-particle operations. But these methods hit a wall: if the system is unknown, you cannot trust the complex controls needed to measure it, and if the controls are too slow, the information vanishes before you can read it.

A team of researchers has now found a way to bypass this dilemma, proving that it is possible to learn the blueprint of a complex quantum system with extreme precision using only simple, static magnetic fields applied to individual particles. Their work demonstrates that by carefully tuning these static fields, they can create a temporary "pause" in the system's natural tendency to thermalize. This pause, which they call a prethermal state, acts as a protective window where the system's internal dynamics remain coherent and informative for a much longer time than expected. Within this window, the researchers can extract the precise strengths of the interactions between particles, achieving a level of accuracy that was previously thought to require much more sophisticated and risky control methods.

The core of their discovery lies in how they manipulate the system's energy landscape. Instead of trying to freeze the entire system or force it into a specific state, they apply a strong, steady field to individual qubits, the basic units of quantum information. This field is strong enough to impose a temporary order, creating what the researchers describe as an emergent symmetry. In this state, the system behaves as if it has a hidden rule that prevents it from scrambling its information, even though it is still evolving. Crucially, the strength of this field does not need to increase as the system gets larger; it only needs to grow very slowly relative to the desired precision. This means the method scales efficiently, working just as well for a system of a thousand particles as it does for a dozen, without requiring exponentially more powerful equipment.

To read the blueprint, the researchers designed an experiment that resembles a classic technique called Ramsey interferometry, but adapted for a many-body environment. They prepare the system in a simple starting state and let it evolve under the influence of the unknown interactions and their static control fields. By measuring specific properties of individual particles at the end of this evolution, they can detect a signal that oscillates at a frequency directly tied to the interaction strengths they wish to learn. Because the prethermal symmetry keeps the system from thermalizing, this signal remains clear and distinct for a long time, allowing the researchers to measure the frequency with high precision. The longer they can measure, the more accurate their reading of the blueprint becomes, a relationship known as the Heisenberg limit, which represents the ultimate precision allowed by the laws of physics.

A significant challenge in such experiments is that measuring one part of a system often disturbs the rest, and doing this for many particles simultaneously is usually impossible without the signals interfering with each other. The researchers solved this by using a clever spatial strategy. They identified specific "defect" sites within the system where they would apply their measurements and control fields. By ensuring these sites were far enough apart from one another, they could run experiments on all of them at the same time. The distance between the sites was large enough that the influence of one measurement did not bleed into the others, allowing the team to gather data from the entire system in parallel. This parallelization drastically reduced the total time required to learn the full blueprint, making the process efficient enough to be practical for large-scale quantum devices.

The researchers then took the raw frequency data from these experiments and fed it into a classical computer algorithm. This algorithm worked like a reverse-engineering tool, comparing the measured frequencies against a mathematical model to deduce the exact values of the interaction strengths. The team proved that this mathematical problem was stable, meaning that small errors in the measurements would not lead to wild guesses about the interactions. Instead, the algorithm could reliably converge on the correct answer, even in the presence of experimental noise. They demonstrated that this entire process—from preparing the system to calculating the final result—could be done with a total evolution time that scales efficiently with the size of the system and the desired precision.

This work fundamentally changes the requirements for learning quantum systems. It shows that the expensive, high-speed, and complex control operations previously thought necessary are not actually required. Instead, simple static fields, which are easier to calibrate and less prone to error, are sufficient to unlock the secrets of complex quantum matter. The researchers established that by leveraging the physics of prethermalization, one can delay the loss of information long enough to extract it with the highest possible precision. This opens the door to characterizing and verifying large quantum processors and simulators without needing to know their inner workings beforehand, a crucial step toward building reliable quantum technologies. The findings suggest that the path to understanding complex quantum systems does not require fighting against thermalization with brute force, but rather working with the system's natural dynamics to create a temporary, stable window for observation.

Technical Summary: Emergent Prethermal Symmetries for Scalable Hamiltonian Learning

1. Problem Setup and Motivation

The paper addresses the problem of learning the coefficients of a geometrically local many-body Hamiltonian $H(\lambda) = \sum_{a} \lambda_a P_a$ from real-time evolution data. The goal is to estimate the parameter vector $\lambda$ to an accuracy $\epsilon$ with failure probability $\delta$ .

A central challenge in this domain is the "Heisenberg limit," where the total evolution time required to achieve precision $\epsilon$ scales as $O(1/\epsilon)$ , which is information-theoretically optimal. However, achieving this limit typically requires complex quantum control. Existing protocols often rely on:

Trusted many-qubit operations: Which are difficult to implement when the Hamiltonian itself is unknown.
Rapidly pulsed control: Where the control interval $\tau$ must scale as $O(\sqrt{\epsilon})$ . This creates a "precision floor" because real gates have finite duration; once $\tau$ falls below the gate time, the protocol fails.

The authors identify a tension: Heisenberg-limited learning requires control to prevent thermalization (which erases local information), yet such control is often unavailable when the system is not yet well-characterized. The paper asks if Heisenberg-limited learning is possible using only static single-qubit fields whose strength grows only polylogarithmically with $1/\epsilon$ and is independent of the system size $n$ .

2. Methodology

The proposed protocol utilizes emergent prethermal symmetries to delay thermalization without requiring fast or trusted multi-qubit control. The methodology proceeds in three main stages:

A. Static Fields and Prethermal Symmetry
The protocol applies random static single-qubit fields to the system. Specifically, for a chosen "defect" site $k$ , a field of strength $\nu$ is applied to a local neighborhood $B = B_R(k)$ , with the field strength halved at the defect site itself and zero outside $B$ .

Mechanism: This setup creates an approximate conservation law (a "charge") associated with the field. Using the ADHH (Abanin-De Roeck-Ho-Huveneers) recursion, the authors construct a quasi-local unitary transformation that maps the physical Hamiltonian to an effective Hamiltonian.
Result: In this effective frame, the system possesses an emergent symmetry that commutes with the effective Hamiltonian up to a quasi-exponentially small remainder. This suppresses thermalization for timescales of order $O(1/\epsilon)$ , provided the field strength $\nu$ is sufficiently large (specifically $\nu \sim \text{polylog}(1/\epsilon)$ ).

B. Local Spectroscopy and Frequency Estimation
Within this prethermal window, the system exhibits coherent dynamics. The protocol prepares a product state (a superposition of local field eigenstates at the defect) and measures single-qubit Pauli observables.

Signal: The measurement yields a Ramsey-like signal $S(t)$ that oscillates at a frequency determined by an effective transition energy $E(\lambda)$ .
Robust Estimation: Using robust frequency estimation algorithms (a variant of robust phase estimation), the protocol extracts these transition energies with Heisenberg-limited scaling ( $O(1/\epsilon)$ time) despite the presence of static noise and finite field strength.

C. Local Approximation and Classical Inversion
A critical technical contribution is showing that the global effective transition energy can be approximated locally.

Locality: Through a Taylor expansion of the ADHH recursion in a complex parameter $z$ , the authors prove that the transition energy at a defect depends primarily on Hamiltonian terms within a local patch of radius $R \sim O(\log(1/\epsilon))$ . Contributions from distant terms decay exponentially.
Optimization: The learning problem is formulated as a least-squares optimization: minimizing the discrepancy between measured transition energies and a forward model $F(\mu)$ computed on local patches.
Stability: The authors prove that the Jacobian of this forward map is well-conditioned in the strong-field limit. They construct a fixed-preconditioner gradient descent algorithm that converges exponentially to the true coefficients $\lambda$ , provided the field strength $\nu$ exceeds a threshold independent of $n$ .

D. Parallelization
To achieve scalability, the protocol runs experiments on multiple "defects" simultaneously.

Separation: Defects are chosen to be spatially separated by a distance $L \sim O(\log(1/\epsilon))$ .
Decoupling: Using the connected-word expansion of the ADHH recursion, the authors show that interactions coupling two separated defects only appear at high orders (degree $P \sim L/r_0$ ). By truncating the effective Hamiltonian at this order, the dynamics at each defect are shown to be effectively independent, allowing parallel data collection without significant cross-talk.

3. Key Contributions

Scalable Heisenberg-Limited Learning: The paper presents the first protocol that achieves Heisenberg-limited scaling ( $O(1/\epsilon)$ total evolution time) for geometrically local many-body Hamiltonians using only static single-qubit control.
System-Size Independence: Unlike previous methods where control parameters (like field strength or pulse intervals) might scale with system size $n$ , the required field strength $\nu$ and the number of recursion steps are independent of $n$ . They scale only polylogarithmically with $1/\epsilon$ .
Emergent Prethermal Symmetries as a Resource: The work establishes emergent prethermal symmetries not just as a physical phenomenon, but as a practical resource for quantum learning, enabling long-time coherent dynamics without complex control.
Rigorous Analytic Continuation: The authors extend the ADHH recursion to non-Hermitian interactions via analytic continuation. This allows for rigorous bounds on Taylor truncation errors and derivative estimates, which are crucial for proving the stability of the classical inverse problem.
Parallelization Strategy: The protocol demonstrates how to parallelize learning across $n$ sites using $O(\log^d(1/\epsilon))$ batches, reducing the total evolution time dependence on $n$ from linear to logarithmic.

4. Main Results

The main theorem (Theorem 1.1 and Theorem 7.8) states that for a $d$ -dimensional lattice of $n$ qubits with geometrically local interactions:

Accuracy: The protocol produces an estimate $\hat{\lambda}$ such that $\|\hat{\lambda} - \lambda\|_\infty \leq \epsilon$ with probability at least $1-\delta$ .
Total Evolution Time: The total time the system evolves under the unknown Hamiltonian is:
$T_{\text{total}} = O\left( \frac{\log^d(1/\epsilon) \log^2(n/\delta)}{\epsilon} \right)$
This saturates the Heisenberg limit up to polylogarithmic factors.
Field Strength: The required static field strength is:
$\nu \geq O\left( \max\left\{ \log^d(1/\epsilon), \log(1/\epsilon)[1 + \log \log(1/\epsilon)]^3 \right\} \right)$
This strength is independent of $n$ .
Resources: The protocol uses product state preparation, single-qubit measurements, and non-adaptive experiments. The classical post-processing time is polynomial in $n$ , $1/\epsilon$ , and $\log(1/\delta)$ .

5. Significance and Claims

The authors claim that this work resolves the tension between the need for control to achieve Heisenberg scaling and the lack of trusted control in unknown systems. By leveraging prethermalization, they show that neither trusted many-qubit operations nor rapidly pulsed control are necessary for scalable Hamiltonian learning.

The paper emphasizes that their results are established through rigorous mathematical proofs involving:

Analytic continuation of the ADHH recursion.
Local-norm estimates and Lieb-Robinson bounds.
Stability analysis of the classical inverse map.

The work opens a route to characterizing many-body systems where fast or trusted control is unavailable, suggesting that prethermal symmetries can serve as a general resource for scalable quantum learning and certification. The authors note that while the protocol is robust to certain State Preparation and Measurement (SPAM) errors, it relies on the ability to apply static fields and perform single-qubit measurements.