Eigenstate Thermalization Hypothesis (ETH) for off-diagonal matrix elements in integrable spin chains

Using advanced Algebraic Bethe Ansatz techniques, this study demonstrates that off-diagonal matrix elements of local operators in the integrative isotropic spin-$1/2$ Heisenberg chain exhibit distinct exponential decay behaviors and follow Gumbel distributions depending on whether the eigenstates belong to the same or different thermodynamic macrostates.

The Gist

Imagine you have a giant, complex machine made of thousands of tiny, spinning tops (these are the "spins" in a quantum chain). You give this machine a sudden shake (a "quantum quench") and then let it run on its own.

The big question physicists ask is: How does this chaotic machine eventually settle down and look like a calm, hot cup of coffee? (This process is called "thermalization").

For most chaotic machines, we have a rulebook called the Eigenstate Thermalization Hypothesis (ETH). It says that if you look at the machine's internal "gears" (mathematical connections between different states), they should look like random noise that gets weaker and weaker as the machine gets bigger. Specifically, the "off-diagonal" gears (the ones connecting two different states) should vanish exponentially fast, like a whisper fading into silence.

But what happens if the machine isn't chaotic? What if it's "integrable"?
Integrable systems are like a perfectly tuned orchestra where every instrument follows a strict, predictable score. They have so many hidden rules (conserved quantities) that they shouldn't thermalize in the usual way. They should stay stuck in their initial state forever.

However, this paper investigates a specific type of integrable machine: the XXX Spin Chain (a line of quantum spins). The authors wanted to see if the "whispering" rule of ETH still applies here, or if the strict rules of the orchestra change the game entirely.

The Main Discovery: The "Gumbel" Surprise

The authors used a powerful mathematical tool called the Algebraic Bethe Ansatz (think of it as a super-advanced calculator that can solve these specific quantum puzzles without needing a supercomputer) to look at the "gears" (matrix elements) between different states.

Here is what they found, broken down simply:

1. The "Same Room" vs. "Different Rooms"

Imagine the machine has different "rooms" (macrostates) based on its energy and temperature.

• Same Room: If you compare two states inside the same room (e.g., two states that both look like a hot cup of coffee), the connection between them fades away exponentially as the machine gets bigger. This is similar to the standard ETH rule.
• Different Rooms: If you compare a state from a "hot room" to a state from a "frozen room," the connection vanishes much, much faster. It's like trying to whisper across a canyon; the signal dies out almost instantly.

2. The Shape of the Noise (The Big Twist)

This is the most exciting part. In standard chaotic systems, the "noise" of these connections follows a Gaussian distribution (the classic Bell Curve). If you plotted the strength of all the connections, you'd get a nice, symmetrical hill.

But in this integrable chain, the hill is lopsided!
The authors found that the connections follow a Gumbel distribution.

• Analogy: Imagine you are measuring the height of the tallest person in a crowd.
  • A Gaussian (Bell Curve) is like measuring the height of everyone in the crowd. Most people are average height; very few are giants or dwarfs.
  • A Gumbel distribution is like measuring the maximum height in many different crowds. It's skewed. You expect a few very tall outliers, and the "average" isn't the center of the action.

The paper shows that even though the system is "integrable" (predictable), the way these connections behave statistically is not Gaussian. It's "heavy-tailed," meaning there are rare, surprisingly strong connections that wouldn't exist in a chaotic system.

3. The "Bound States" (The Sticky Clumps)

In this quantum chain, particles can stick together to form "strings" or "bound states" (like a group of dancers holding hands and moving as one unit).

• Usually, calculating the math for these sticky clumps is a nightmare because the equations blow up with "singularities" (mathematical errors).
• The authors developed a clever way to smooth out these errors. They found that even with these sticky clumps present, the "Same Room" connections still fade exponentially, and the "Different Room" connections fade even faster. The sticky clumps don't break the rules; they just add a little extra flavor to the statistics.

Why Does This Matter?

1. It bridges two worlds: It shows that even in "predictable" integrable systems, there is a hidden layer of randomness that looks a lot like the chaotic ETH, but with a unique statistical fingerprint (the Gumbel distribution).
2. It explains the "Why": It proves that the difference between chaotic and integrable systems isn't just about whether they thermalize, but how the internal connections fluctuate.
3. Future Tech: Understanding how these quantum systems relax (or fail to relax) is crucial for building quantum computers. If a quantum computer is too "integrable," it might not thermalize correctly, leading to errors. If it's too chaotic, it loses information too fast. Knowing the exact "shape" of the noise helps engineers design better systems.

The Takeaway

Think of the quantum chain as a massive, complex dance floor.

• In a chaotic party, everyone moves randomly, and the chance of two specific people bumping into each other drops off smoothly (Gaussian).
• In this integrable dance, the dancers follow a strict choreography. You'd expect them to never mix. But the authors found that they do mix, just very rarely. And when they do mix, the pattern of those rare encounters is weird and lopsided (Gumbel), not the smooth bell curve we expected.

The paper essentially says: "Even in a perfectly ordered system, the chaos of the quantum world leaves a unique, non-Gaussian fingerprint."

Technical Summary

1. Problem Statement

The Eigenstate Thermalization Hypothesis (ETH) is the standard framework explaining how isolated quantum systems thermalize. For chaotic (non-integrable) systems, ETH posits that off-diagonal matrix elements of local operators between energy eigenstates decay exponentially with system size $L$ and follow a Gaussian distribution for their fluctuations.

However, integrable systems possess an extensive number of conserved charges, leading to non-thermal steady states (described by the Generalized Gibbs Ensemble). While the ETH scaling for diagonal matrix elements in integrable systems has been studied (showing $L^{-1}$ decay rather than exponential), the behavior of off-diagonal matrix elements in integrable lattice models remained an open question.

Previous work (Ref. [26]) investigated this in the Lieb-Liniger model (a 1D continuum field theory of bosons), finding:

• Off-diagonal elements between states in the same macrostate decay as $\exp(-c L \ln L - f L)$.
• Off-diagonal elements between states in different macrostates decay as $\exp(-d L^2)$.
• The statistics of these elements followed a Fréchet distribution.

The authors aim to determine if this scenario holds for integrable lattice models (specifically the spin-1/2 Heisenberg chain), which feature bound states (Bethe strings) absent in the repulsive Lieb-Liniger model. The presence of bound states significantly complicates the numerical computation of matrix elements due to singularities in the Algebraic Bethe Ansatz (ABA) formulas.

2. Methodology

The authors employ state-of-the-art Algebraic Bethe Ansatz (ABA) results to compute matrix elements of local operators between generic energy eigenstates.

• Model: The isotropic spin-1/2 Heisenberg chain (XXX chain) with Hamiltonian $H = \frac{1}{4} \sum \vec{\sigma}_i \cdot \vec{\sigma}_{i+1}$.
• Operators:
  • One-spin operators: $E^{(11)}_i = \frac{1}{2}(1 + \sigma^z_i)$.
  • Two-spin operators: $E^{(11)}_i E^{(22)}_{i+1}$ (acting on nearest neighbors).
• Thermodynamic Macrostates:
  • Same Macrostate: Eigenstates sampled from a thermal (Gibbs) ensemble at a fixed temperature $\beta$ and particle density $M/L$.
  • Different Macrostates: Eigenstates sampled from the zero-temperature (ground state) and infinite-temperature ensembles.
• Numerical Strategy:
  • Sampling: A Metropolis algorithm is used to sample eigenstates according to the Gibbs weight $e^{-\beta E}$.
  • Handling Strings: The authors utilize specific ABA formulas (Slavnov and Gaudin determinants) that have been regularized to handle Bethe strings (bound states).
    • For one-spin operators, the homogeneous limit and the limit of vanishing string deviations can be taken analytically, allowing the study of systems with strings up to $L \approx 240$.
    • For two-spin operators, the simultaneous limits are numerically challenging. Consequently, the authors restrict the two-spin analysis to eigenstates without bound states (no strings), reaching up to $L \approx 60$.
  • Arbitrary Precision: Due to the extreme smallness of off-diagonal elements ($\sim e^{-L}$), arbitrary-precision arithmetic is used to evaluate the ABA formulas.

3. Key Contributions

1. Extension to Lattice Models: This is the first study to verify ETH scaling for off-diagonal matrix elements in an integrable lattice model (XXX chain) containing bound states, moving beyond the continuum Lieb-Liniger model.
2. Resolution of Singularities: The paper successfully implements numerical strategies to regularize the fictitious singularities arising from the string hypothesis in the ABA formulas, enabling the calculation of matrix elements for large systems ($L \sim 240$).
3. Statistical Characterization: The authors provide a comprehensive statistical analysis of the distribution of matrix elements, contrasting integrable and chaotic behaviors.

4. Key Results

A. Scaling of Matrix Elements

The authors define $M_{ij} = \ln |\langle E_i | O | E_j \rangle|^2$.

• Same Macrostate (Thermal):
  • The off-diagonal matrix elements decay exponentially with system size $L$.
  • The scaling is found to be $|\langle E_i | O | E_j \rangle| \propto \exp(-c L \ln L - c_0 L)$.
  • This confirms the qualitative scenario found in the Lieb-Liniger model (Ref. [26]), indicating that the presence of bound states does not alter the fundamental decay law.
• Different Macrostates:
  • Matrix elements between eigenstates from different macrostates (e.g., ground state vs. infinite temperature) decay much faster.
  • The scaling follows $|\langle E_i | O | E_j \rangle| \propto \exp(-d L^2)$.

B. Probability Distribution Functions (PDFs)

A crucial finding concerns the statistical distribution of the matrix elements, which differs significantly from the chaotic ETH prediction.

• Distribution Type: The PDF of the logarithm of the matrix elements, $M_{ij}$, is well-described by a Gumbel distribution (Extreme Value Type I), rather than the Gaussian distribution predicted by standard ETH for chaotic systems.
  • $P(x) \propto e^{-(x-\mu)/\beta} e^{-e^{-(x-\mu)/\beta}}$.
• Contrast with Lieb-Liniger: This contrasts with Ref. [26], which reported a Fréchet distribution for the Lieb-Liniger model. The authors note that in the limit of large parameters, the Fréchet distribution can reduce to a Gumbel, but the specific parameters found here are distinct.
• Non-Gaussianity Evidence:
  • The authors compute the ratio $\Gamma = \overline{|O_{ij}|^2} / (\overline{|O_{ij}|})^2$.
  • For a zero-mean Gaussian distribution, $\Gamma = \pi/2$.
  • The numerical results show $\Gamma \sim 10^5$ (diverging with $L$), providing strong evidence that the matrix elements are non-Gaussian and consistent with the Gumbel statistics.

C. One-Spin vs. Two-Spin Operators

• One-spin operators: Results are robust up to $L=240$, including states with strings.
• Two-spin operators: Results are limited to $L \approx 60$ and states without strings. However, the data confirms the same scaling and Gumbel statistics as the one-spin case, suggesting universality across operator support size.

5. Significance and Implications

• Universality in Integrable Systems: The study suggests that the specific statistics of off-diagonal matrix elements (Gumbel distribution) and their scaling laws are robust features of integrable lattice systems, regardless of the presence of bound states.
• Distinction from Chaos: The results highlight a fundamental difference between integrable and chaotic systems: while both may exhibit exponential decay of off-diagonal elements (with logarithmic corrections in integrable cases), the fluctuation statistics are distinct (Gumbel vs. Gaussian). This implies that the "randomness" in integrable systems is governed by different statistical mechanics (potentially related to Free Probability Theory, as suggested in the discussion).
• Future Directions: The work opens the door to investigating the relationship between these statistics and Free Probability Theory (FPT) in integrable models. It also provides the necessary tools to reconstruct full-time dynamics of observables from matrix element statistics, extending previous work on the Lieb-Liniger model to lattice systems.

In summary, this paper establishes that for the integrable XXX chain, off-diagonal matrix elements decay exponentially (with logarithmic corrections) and follow Gumbel statistics, providing a refined understanding of thermalization (or lack thereof) in integrable lattice models.