🔬 condensed matter

Fragmented ETH: Prethermalization, Timescales, and Ensemble Inequivalence

This paper investigates how strong long-range interactions in finite quantum systems lead to Hilbert-space fragmentation and anomalously slow prethermalization, proposing a "fragmented eigenstate thermalization hypothesis" (fETH) to explain band-resolved thermalization and ensemble inequivalence without global ergodicity.

Original authors: C. L. Sriram, Soumya Kanti Pal, Lea F. Santos

Published 2026-07-20

📖 9 min read🧠 Deep dive

CC BY 4.0

Original authors: C. L. Sriram, Soumya Kanti Pal, Lea F. Santos

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 a universe where every particle is a dancer, and the music they follow is the fundamental laws of physics. For over a century, physicists have been trying to understand how these dancers, starting from a chaotic, frozen pose, eventually smooth out into a calm, predictable rhythm that we call "thermal equilibrium." This is the bridge between the wild, jittery world of quantum mechanics (where things are fuzzy and uncertain) and the steady, reliable world of thermodynamics (where a cup of coffee cools down at a predictable rate). The big question has always been: How does a pure quantum state, evolving on its own without any outside help, decide to act like a hot gas or a cold solid?

For a long time, the leading theory was that if the dancers are chaotic enough—like a mosh pit where everyone bumps into everyone else—they will eventually forget their starting positions and settle into a statistical average. This idea, known as the Eigenstate Thermalization Hypothesis (ETH), suggests that the system acts like a perfect mixer, blending all possible states together until everything looks the same. But what happens if the dancers aren't in a mosh pit? What if they are in a giant, perfectly symmetrical ballroom where they can only dance with specific partners, or if the music is so loud that they get stuck in a trance? This is the puzzle of "long-range interactions," where every particle feels the pull of every other particle, no matter how far away. It's a scenario that happens in real experiments with trapped ions and super-cooled atoms, and it breaks the usual rules of the dance floor.

In this new study, researchers C. L. Sriram, Soumya Kanti Pal, and Lea F. Santos dive into this chaotic ballroom to see what happens when the music is played at a "super long-range" volume. They discover that the system doesn't just mix; it shatters. Instead of one big, messy dance floor, the Hilbert space (the mathematical room where all possible dance moves live) gets fragmented into separate, isolated islands called "energy bands." The dancers can move freely within their own island, but they can't easily jump to another one. This leads to a strange, two-step process: the system first gets stuck in a "prethermal" trance, holding a pose for a very long time, before finally, very slowly, waking up and reaching true equilibrium.

The team finds that this "stuck" state isn't universal. Whether a specific observation (like the total spin of the dancers) gets stuck or relaxes immediately depends on a delicate dance between the type of observation and the starting pose of the system. They develop a mathematical toolkit to predict exactly how long this trance lasts and how high the "plateau" of the stuck state will be. Crucially, they show that even though the whole system is fragmented, chaos still reigns within each individual island. This allows them to propose a new version of the old thermalization rules, which they call "fragmented ETH" (fETH). This new rule changes how we compare systems of different sizes, requiring a special "selection rule" based on symmetry rather than just counting up by one. Finally, they reveal that this fragmentation causes a fundamental mismatch between two ways of calculating the system's temperature: the "microcanonical" way (looking at a specific energy island) and the "canonical" way (looking at the whole room). Because these two methods look at different parts of the dance floor, they give different answers, proving that in this fragmented world, the choice of how you measure the system actually changes the result, all without needing a traditional phase transition to explain it.

The Story of the Broken Dance Floor

The Setup: A Ballroom with a Twist
Imagine a massive ballroom filled with $L$ dancers (in this case, quantum spins). In a normal room, dancers only bump into their immediate neighbors. But in this experiment, the music is magical: every dancer feels a pull from every other dancer in the room, no matter how far apart they are. The strength of this pull depends on a number called $\alpha$ . If $\alpha$ is small (specifically, less than 1), the pull is "super long-range."

When the pull is infinitely strong and perfectly symmetrical (the "fully connected" limit where $\alpha = 0$ ), the ballroom has a secret structure. The dancers are organized into strict groups based on their total spin, creating distinct "energy bands." It's like the ballroom is divided into separate, invisible rooms. Dancers in one room can't easily talk to dancers in another.

The Problem: The Two-Step Stumble
The researchers asked: What happens if we nudge this perfect system just a tiny bit? We introduce a tiny imperfection (a small $\alpha > 0$ ) that breaks the perfect symmetry. In a normal system, a tiny nudge would just make the dancers mix faster. But here, something weird happens.

When the system is "quenched" (suddenly changed from a starting pose to the new music), it doesn't relax smoothly. Instead, it takes a two-step stumble:

The Trance (Prethermal Plateau): The system quickly settles into a "prethermal" state. It looks like it has reached equilibrium, but it's actually just stuck in a local groove. It holds this pose for a very long time.
The Awakening: Eventually, the tiny imperfections allow the dancers to slowly leak between the invisible rooms. Only then does the system finally reach true, global equilibrium.

The researchers found that the time it takes to escape this trance grows exponentially as the system gets closer to the perfect, fully connected limit. It's like trying to wake up from a deep sleep; the more perfect the sleep, the harder it is to wake up.

The Rules of the Trance
Not everything gets stuck in the trance. The paper reveals a surprising rule: it depends on what you are watching and how the dance started.

The Magnetization (The Total Spin): If you watch the total spin of the whole group, it relaxes immediately. It doesn't get stuck. Why? Because the total spin is a "symmetric" observer. It can't tell the difference between the dancers inside the same invisible room. It sees the whole room as one big blur, so it skips the trance entirely.
The Excitation Density (The Local Moves): If you watch a specific local move (like the first moment of excitation density), it does get stuck. This observer can see the differences between dancers in the same room.
The Starting Pose: Even for the local moves, the trance only happens if the starting pose of the dancers overlaps with the right "rooms." If you start in a pose that doesn't excite the right internal movements, you won't see the plateau.

The authors prove that the "Loschmidt echo" (a measure of how much the system remembers its starting pose) is the ultimate alarm clock. It always rings (leaves the trance) before any other observable does. This gives a strict lower bound on how long the trance lasts for any other measurement.

The New Map: Fragmented ETH
For a long time, physicists thought that if a system was fragmented, it couldn't thermalize at all. This paper overturns that idea. They show that while the whole ballroom is fragmented, the inside of each invisible room is actually chaotic and mixed up. The dancers within a single room are doing a mosh pit.

This leads to a new theory called Fragmented Eigenstate Thermalization Hypothesis (fETH).

Old Rule (ETH): To check if a system is thermal, you compare systems of size $L$ , $L+1$ , $L+2$ , etc.
New Rule (fETH): Because of the symmetry of the ballroom, you can't just compare any sizes. You have to follow a "selection rule." You can only compare systems where the size changes by multiples of 4 (e.g., $L$ , $L+4$ , $L+8$ ). If you compare $L=12$ and $L=13$ , you are comparing apples to oranges because the invisible rooms are arranged differently. But if you compare $L=12$ and $L=16$ , the rooms line up perfectly, and you can see the chaos smoothing out as the system gets bigger.

The Temperature Paradox
Finally, the paper tackles a classic problem in physics: "Ensemble Inequivalence." Usually, if you calculate the temperature of a system by fixing its energy (Microcanonical) or by fixing its temperature (Canonical), you get the same answer. But in this fragmented world, they don't.

The Microcanonical View: This looks at a single energy band (one invisible room). It sees the specific density of states inside that room.
The Canonical View: This looks at the whole ballroom, mixing all the rooms together.

Because the two methods are looking at different parts of the Hilbert space (one room vs. the whole building), they predict different temperatures for the same energy. The paper shows that this isn't because of a phase transition (like ice melting); it's purely because the spectrum is broken into bands. The "caloric curve" (temperature vs. energy) becomes multi-valued for the microcanonical view, while it stays smooth for the canonical view.

The Bottom Line
This work doesn't just say "long-range interactions are weird." It builds a complete, unified picture. It shows that approximate symmetry creates energy bands, which cause a two-stage relaxation process. It proves that chaos still exists within the bands, allowing for a new kind of thermalization (fETH). And it explains why different ways of measuring the system give different answers, all without needing a phase transition. The authors provide analytical formulas for the length of the trance and the height of the plateau, backed by simulations that confirm their predictions. They have turned a confusing mess of fragmented quantum states into a structured, predictable dance.

Technical Summary: Fragmented ETH: Prethermalization, Timescales, and Ensemble Inequivalence

Problem Statement
This work addresses the fundamental question of how finite quantum systems with strong long-range interactions approach thermal equilibrium. Specifically, it investigates systems where the interaction strength decays as a power law $J_{ij} \propto 1/|i-j|^\alpha$ with $\alpha < d$ (the super-long-range regime). In the fully connected limit ( $\alpha=0$ ), these systems possess full permutation symmetry, leading to a highly degenerate spectrum organized into energy bands. The central problem is understanding how the approach to equilibrium proceeds when global ergodicity is broken by this spectral fragmentation, yet the system is not strictly integrable. Previous work suggested that the Eigenstate Thermalization Hypothesis (ETH) might fail in such regimes, and the mechanisms governing the emergence of prethermal plateaus, their timescales, and the resulting statistical mechanics remained unclear.

Methodology
The authors analyze a one-dimensional spin-1/2 transverse-field Ising chain with super-long-range interactions, described by the Hamiltonian:
$\hat{H}(\alpha) = \sum_{i<j} \frac{J}{N^\alpha |i-j|^\alpha} \hat{\sigma}^x_i \hat{\sigma}^x_j + \frac{h}{L} \sum_{i=1}^L \hat{\sigma}^z_i$
where $N^\alpha$ is a Kac factor ensuring extensivity. The study focuses on the regime $0 < \alpha \ll 1$ , treating the system as a perturbation of the fully connected Lipkin-Meshkov-Glick (LMG) model ( $\alpha=0$ ).

The methodology combines:

Perturbative Theory: Using degenerate perturbation theory to derive analytical expressions for the survival probability and the dynamics of observables. This separates dynamics into fast interband dephasing and slow intraband evolution.
Numerical Simulations: Exact diagonalization of systems with sizes $L$ up to 16 (and scaling analysis up to $L=20$ ) to compute survival probabilities, entanglement entropy, and local observables.
Rigorous Bounds: Deriving a lower bound on prethermal lifetimes using the Loschmidt echo (survival probability) and trace-norm inequalities.
Finite-Size Scaling: Developing a new scaling framework for the Eigenstate Thermalization Hypothesis that respects the selection rules imposed by permutation and spin-inversion symmetries.

Key Contributions and Results

Fragmented Spectrum and Two-Stage Equilibration:
The authors demonstrate that for $0 < \alpha \ll 1$ , the many-body spectrum retains an approximate band structure inherited from the $\alpha=0$ limit. This leads to a two-stage equilibration process:
- Stage 1 (Fast): Rapid dephasing between different energy bands drives the system to a prethermal plateau.
- Stage 2 (Slow): Much slower dynamics within individual bands, governed by small energy splittings, eventually leads to thermal equilibrium.
  The lifetime of the prethermal plateau scales inversely with the perturbation strength, growing as $\alpha \to 0$ .
Conditions for Prethermalization:
Contrary to the survival probability, which always exhibits a prethermal plateau, the emergence of such plateaus for general observables is not universal. The authors derive precise criteria: a prethermal plateau appears only if (i) the observable has non-vanishing off-diagonal matrix elements between states that remain degenerate in the $\alpha=0$ limit, and (ii) the initial state has finite overlap with these specific states.
- Result: Collective observables invariant under full permutation symmetry (e.g., total magnetization $M^z$ ) do not exhibit prethermal plateaus because their off-diagonal matrix elements within degenerate bands vanish.
- Result: Non-symmetric observables (e.g., the first moment of excitation density $C_z$ ) exhibit plateaus only for specific initial states that populate bands with significant off-diagonal matrix elements.
Lower Bound on Prethermal Lifetimes:
The paper proves that the lifetime of the prethermal plateau for the Loschmidt echo provides a rigorous lower bound for the prethermal lifetime of any bounded observable. This implies that the survival probability is the earliest quantity to signal the breakdown of the prethermal regime, even though it may take the longest time to reach final equilibrium saturation.
Fragmented Eigenstate Thermalization Hypothesis (fETH):
The authors propose that while global ergodicity is broken, quantum chaos develops within individual energy bands. This supports a "band-resolved" formulation of ETH (fETH).
- Selection Rule: A critical methodological contribution is the identification of a symmetry-imposed selection rule for finite-size scaling. Due to spin-inversion symmetry, eigenstate expectation values can only be compared across system sizes $L, L+4, L+8, \dots$ . Standard consecutive scaling ( $L, L+2$ ) is invalid because it mixes different symmetry sectors.
- Validation: Numerical results confirm that within these symmetry-compatible sequences, eigenstate fluctuations decrease with system size, validating fETH.
Microscopic Origin of Ensemble Inequivalence:
The work identifies a microscopic mechanism for ensemble inequivalence in strongly long-range interacting systems, distinct from the conventional thermodynamic mechanism involving non-concave entropy or phase transitions.
- Mechanism: The microcanonical ensemble is confined to a single energy band (due to fETH), whereas the canonical ensemble samples states across multiple bands.
- Result: This mismatch causes the canonical and microcanonical caloric curves (inverse temperature vs. energy) to differ qualitatively. The microcanonical curve is multivalued (repeating structures for each band), while the canonical curve is smooth.

Significance
The paper establishes a unified framework connecting permutation symmetry, Hilbert-space fragmentation, prethermalization, and equilibrium statistical mechanics. It overturns the notion that Hilbert-space fragmentation precludes thermalization; instead, it reshapes the dynamical route to equilibrium and the statistical description of the equilibrium state.

The authors claim that their results apply to a broad class of Hamiltonians exhibiting Hilbert-space fragmentation due to approximate permutation symmetry. By providing analytical expressions for prethermal timescales and a rigorous lower bound on lifetimes, the work offers a predictive theory for when and how prethermal plateaus emerge. Furthermore, the development of the fETH framework and the associated selection rules for finite-size scaling provides a necessary correction to the analysis of thermalization in systems with strong long-range interactions, explaining ensemble inequivalence without invoking equilibrium phase transitions.

The Story of the Broken Dance Floor

More like this