Unified Probe of Quantum Chaos and Ergodicity from Hamiltonian Learning

arXiv:2603.04486 · quant-ph, cond-mat.stat-mech · Submitted 2026-03-04 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Quantum Radio. Generated commentary on the latest quantum physics and condensed matter papers.

Kai: We're looking at the paper: "Unified Probe of Quantum Chaos and Ergodicity from Hamiltonian Learning".

Mira: Here's the gist of the paper. Developing measures of quantum ergodicity and chaos stands as a foundational task in the study of quantum many-body systems.

Kai: In this work, we propose metrics for these effects based on Hamiltonian learning that unify multiple advantages of existing metrics.

Mira: We analytically and numerically show that our metrics not only distinguish between integrable and ergodic regimes in various spin chains but also quantify chaos and ergodicity, allowing us to locate regions of parameter space displaying maximal ergodicity and maximal sensitivity to local perturbations.

Kai: Our approach not only provides conceptual ways to study quantum chaos and ergodicity but also presents viable experimental methods for quantum simulators.

Title and authors: Kai: So, let's talk about who wrote this. Nik O Gjonbalaj is from Harvard University, and he’s working on quantum hardware, so he brings that experimental perspective to how these metrics are built.

Mira: And I’m looking at the theoretical side of it; I want to understand the assumptions they’re making about what an eigenstate actually contains when we use it for learning.

Lev: From my side, I’m focused on the engineering hurdle: can we actually prepare these approximate eigenstates with non-zero energy variance in a way that's practical for a simulator or a real machine?

Kai: The paper shows how you can use techniques like quantum nondemolition measurements to get those approximate eigenstates, and they discuss how local measurements from shadow tomography help build the covariance matrix M.

Mira: They also connect the variance spectrum directly to the quantum Fisher information, which is important because it means this measurement isn't just some arbitrary quantity; it’s related to how much information we can extract about Hamiltonian perturbations.

Lev: That connection to quantum Fisher information makes sense if we think about what you need for error correction, but I wonder how much overhead this whole learning process adds when you try to run it on a real multi-qubit system.

Kai: The paper tackles that robustness head-on; they show that the gap in the covariance matrix directly quantifies the robustness of Hamiltonian learning, meaning a larger gap means less error for a fixed measurement uncertainty.

Mira: They also analytically show that ergodicity implies this robustness, and they point out that integrable models are parametrically less robust to errors during Hamiltonian learning than ergodic ones.

Lev: So, if we're building real hardware, this suggests we might need to focus on engineering systems where the gap of that covariance matrix is inherently large.

The paper's summary: Kai: To summarize what they’ve done with "Unified Probe of Quantum Chaos and Ergodicity from Hamiltonian Learning," they are proposing a set of metrics built around Hamiltonian learning that unify existing ways to study quantum chaos and ergodicity.

Mira: They show that this unified approach can distinguish between integrable and ergodic regimes in various spin chains, not just by looking at the variance spectrum, but also by using metrics like the inverse spread DE and the gap Delta M.

Lev: The key takeaway for me is that they are showing how chaos and ergodicity improve the robustness of Hamiltonian learning to small errors, which is a huge practical implication for any algorithm we might try to run on hardware.

Kai: And they quantify chaos by looking at the maximum eigenvalue of the variance spectrum, sigma squared max, which measures sensitivity to local perturbations, and they also show that this metric detects maximal chaos.

Mira: They demonstrate that you can continuously quantify ergodicity within the ergodic regime itself, which lets us map out parameter space to find regions where ergodicity is at its peak.

Lev: From an engineering standpoint, the part I focus on is their experimental viability; they show that these metrics are accessible because they use randomized measurements and approximate eigenstates that we can prepare using techniques like a quantum nondemolition algorithm.

Kai: And they confirm that this approach uses-local observables, meaning the measurements only involve local interactions between neighboring sites, which is very good for experimental setup.

The paper's improvements: Mira: They suggest several ways to improve the original framework, and one big improvement is making the connection to quantum information clearer; they show that the variance spectrum encodes the quantum Fisher information for all Hamiltonian perturbations generated by-local operators.

Kai: That’s a lot of info in one place; it means that this variance spectrum isn't just telling us about chaos; it’s also telling us about the QFI for a specific subset of local measurements.

Lev: If we think about applying this to error correction, this suggests we can use these observables to probe the quantum noise landscape more directly by looking at how well the learning algorithm works under those specific constraints.

Mira: They also refine how they define things like the AGP norm, showing it can be defined for an infinite family of systems parametrized by an energy cutoff, which gives them more flexibility in their analysis.

Kai: And they’ve made it clear that the covariance matrix M itself represents a redefinition of quantum Fisher information when you only constrain the learning process to a specific subsystem Lambda.

Lev: That's interesting because it means we can tailor the probe to focus on specific parts of the system, which is crucial if we want to analyze subsystems within a larger quantum circuit.

Mira: They also provide concrete maps of parameter space, showing exactly where those regions of maximal ergodicity and enhanced local sensitivity are located in things like mean-field Ising models.

Conclusion: Kai: So, to wrap up this paper "Unified Probe of Quantum Chaos and Ergodicity from Hamiltonian Learning," they show that the variance spectrum can distinguish between integrable and ergodic systems, quantify ergodicity within the ergodic regime, and probe chaos through eigenstate sensitivity.

Mira: The main implication is that this unified probe allows us to map parameter space to find regions of maximal ergodicity and enhanced local sensitivity, which is very useful for understanding the behavior of these complex quantum systems.

Lev: For running on hardware, I’d say the key is using the robustness of Hamiltonian learning as a metric; it tells us how much error we can tolerate before the system breaks down when trying to learn its own rules.

Kai: I agree, Lev. And for my experimental work, this gives me a concrete way to design measurements—we’re not just measuring states; we’re using the covariance matrix M to get information about chaos and ergodicity simultaneously.

Mira: It’s a powerful tool because it connects the microscopic structure of an eigenstate directly to macroscopic properties like chaos and sensitivity, all through this learning mechanism.

Lev: So, this paper is a solid step forward in giving us a quantifiable way to characterize these things that we can actually test using current quantum simulators or even real devices in the future.

Nik O. Gjonbalaj, Christian Kokail, Susanne F. Yelin, Soonwon Choi

Harvard University · ITAMP · QuEra Computing Inc. · Massachusetts Institute of Technology

quant-ph, cond-mat.stat-mech

Submitted: 2026-03-04

Updated: 2026-10-30

Comments: 19+11 pages, 7+5 figures; corrected Fig. 4a, tightened bounds, added SNR analysis

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Importance score: 87/100

The gist: Developing measures of quantum ergodicity and chaos stands as a foundational task in the study of quantum many-body systems.

Key concepts

Variance Spectrum
This is a set of eigenvalues derived from a covariance matrix built from local measurements. It acts as a fingerprint for the eigenstate, allowing researchers to distinguish between integrable systems (narrow spectrum) and ergodic systems (wide spectrum).
Ergodicity Metrics (1/DE and $\Delta$M)
The inverse spread ($1/DE$) and the variance gap ($\Delta$M) are used to quantify ergodicity. Ergodic systems exhibit a narrow variance spectrum with a large gap, which corresponds to high robustness in Hamiltonian learning, meaning the system is less sensitive to errors.
Chaos Metric ($\sigma^2_{max}$)
The maximum eigenvalue of the variance spectrum ($\sigma^2_{max}$) quantifies chaos. It measures how much an eigenstate changes when subjected to small local perturbations, effectively detecting regions where the system is maximally sensitive to these disturbances.

Terminology

Summary

Developing measures of quantum ergodicity and chaos stands as a foundational task in the study of quantum many-body systems. p. 0 In this work, we propose metrics for these effects based on Hamiltonian learning that unify multiple advantages of existing metrics. p. 0 We analytically and numerically show that our metrics not only distinguish between integrable and ergodic regimes in various spin chains but also quantify chaos and ergodicity, allowing us to locate regions of parameter space displaying maximal ergodicity and maximal sensitivity to local perturbations. p. 0 Our approach not only provides conceptual ways to study quantum chaos and ergodicity but also presents viable experimental methods for quantum simulators. p. 0

The Core Methodology

The authors propose metrics for ergodicity and chaos based on the robustness of Hamiltonian learning, specifically how well a Hamiltonian can be reconstructed from a single midspectrum eigenstate. The procedure involves constructing a covariance matrix M from local basis measurements via shadow tomography. p. 0 The eigenvalues of this matrix, denoted as σ2a, constitute the variance spectrum of the eigenstate v⟩. p. 0 p. 0 The spread DE and the gap ∆M are used to quantify ergodicity, while the maximum eigenvalue σ2max probes chaos by quantifying sensitivity to local perturbations. p. 0

Distinguishing Regimes

The variance spectrum provides a clear distinction between integrable and ergodic systems. Specifically, ergodic systems are characterized by a narrow variance spectrum with a large gap, whereas integrable systems show a wide spread and small gap. p. 0 The authors use the inverse spread 1/DE and the variance gap ∆M to quantify ergodicity. p. 0 Numerical results show that ergodic systems allow for drastically more robust Hamiltonian learning. p. 2

Quantifying Chaos and Sensitivity

Chaos is quantified by the maximum eigenvalue σ2max, which probes the maximum sensitivity of the eigenstate to a set of local perturbations. The authors show that this metric detects maximal chaos. p. 0 Furthermore, they show that the minimum of the susceptibility χH can probe chaos by quantifying the sensitivity of eigenstates to l-local perturbations. p. 5

Experimental Viability

The metrics are designed to be experimentally accessible because their locality allows for the use of modern techniques like classical shadow tomography. The protocol can be executed with approximate eigenstates with non-zero energy variance, reducing the burden of preparing exact eigenstates in experiments. p. 1

The Connection to Quantum Information

The robustness of Hamiltonian learning is related to the quantum Fisher information (QFI). The variance spectrum encodes the QFI of all Hamiltonian perturbations generated by l-local operators. p. 7 This connection allows the variance spectrum to effectively probe the QFI for a specific subset of local Hamiltonian perturbations. p. 7

Continuous Quantification

The metrics go beyond simply identifying ergodicity by continuously quantifying it within the ergodic regime. This allows for the construction of maps of parameter space that identify regions of maximal ergodicity. p. 8

Probing Sensitivity in Parameter Space

The study reveals structure in parameter space by identifying regions where the susceptibility of eigenstates to small local perturbations is maximized. This probe of chaos peaks in an intermediate region separating the ergodic and classical regimes. p. 0

Experimental Implementation Details

Approximate eigenstates can be prepared using a quantum nondemolition (QND) algorithm. The covariance matrix M can be efficiently extracted using random local Pauli bases according to classical shadow tomography. p. 0

Conclusion

In summary, the variance spectrum can both distinguish integrable and ergodic regimes and quantify ergodicity within the latter regime. Beyond ergodicity, the variance spectrum is also able to probe chaos via the sensitivity of eigenstates to local perturbations. p. 5 These abilities allow us to uncover detailed maps of parameter space [Fig. p. 8 4(c-d) and Fig. p. 0 5(a)] in the MFIM and identify regions of maximal ergodicity and enhanced local sensitivity. p. 1 Furthermore, the locality of these observables allows us to efficiently measure the variance spectrum from experimentally-preparable approximate energy eigenstates using the classical shadows formalism. p. 1

Pruning redundant reasoning tokens

The authors show that chaos and ergodicity make Hamiltonian learning easier by enhancing its robustness to errors. Most importantly, they demonstrate that this effect is strong enough that ergodicity and chaos can be measured directly through the robustness of Hamiltonian learning. p. 1

How the training signal is built

The covariance matrix M is constructed from local basis measurements via shadow tomography. The eigenvalues of this matrix, denoted as σ2a, constitute the variance spectrum of the eigenstate v⟩.

Where the gains come from

The inverse spread 1/DE and the variance gap ∆M are used to quantify ergodicity. The authors show that ergodic systems allow for drastically more robust Hamiltonian learning. p. 2

The authors analytically show that ergodicity implies the robustness of the algorithm and that a large class of integrable models are parametrically less robust to errors during Hamiltonian learning.

The variance spectrum encodes the QFI of all Hamiltonian perturbations generated by l-local operators.

The variance spectrum is sufficient to extract the purity of any contiguous l-qubit subsystem. The covariance matrix also represents the tomographically-complete limit of the learning algorithm. p. 2 p. 2

The variance spectrum is sufficient to extract the purity of any contiguous l-qubit subsystem. The covariance matrix also represents the tomographically-complete limit of the learning algorithm.

The covariance matrix M also represents a redefinition of QFI when constraints only span the basis of a subsystem Λ.

The variance spectrum is sufficient to extract the purity of any contiguous l-qubit subsystem. The covariance matrix also represents a redefinition of QFI when constraints only span the basis of a subsystem Λ. p. 0 p. 0 p. 0

The variance spectrum is sufficient to extract the purity of any contiguous l-qubit subsystem. The covariance matrix also represents a redefinition of QFI when constraints only span the basis of a subsystem Λ.

The variance spectrum is sufficient to extract the purity of any contiguous l-qubit subsystem. The covariance matrix also represents a redefinition of QFI when constraints only span the basis of a subsystem Λ.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Precision in Hamiltonian reconstruction: The system can reconstruct a local Hamiltonian from a single energy eigenstate with an error magnitude of H equals H0 plus deltaH, where deltaH is determined by the spectral gap of M0, quantified by sigma2 1 minus sigma2 0.

  2. Discrimination between integrable and ergodic regimes: The AI can distinguish between integrable and ergodic systems by calculating metrics like the inverse spread 1/DE and the variance gap DeltaM, which quantitatively distinguish integrable and ergodic systems.

  3. Identification of maximal ergodicity pockets: The system can locate regions of parameter space displaying maximal ergodicity by measuring metrics like 1/DE and DeltaM, which peak in the same region of parameter space for ergodic systems.

  4. Quantifying sensitivity to local perturbations: The AI can quantify chaos by detecting regions where the susceptibility of eigenstates to small local perturbations is maximized, which is probed via the maximal eigenvalue sigma2 max of the variance spectrum.

  5. Robustness assessment against noise: The system can assess Hamiltonian learning robustness by analyzing how the gap of the covariance matrix directly quantifies the robustness of Hamiltonian learning, where a larger gap implies a smaller magnitude of deltaH for a fixed magnitude of covariance error epsilon deltaM.

Sources

Related papers