Resonance Proliferation Across Localization Transitions

The Big Picture: The "Frozen" vs. "Boiling" Quantum World

Imagine a quantum system (like a collection of interacting particles) as a giant, complex dance floor.

The "Frozen" State (Localization): In a perfectly frozen state, the dancers are stuck in their spots. They can wiggle a little, but they never swap places with anyone else. Information about where they started stays trapped in their local area. This is called Many-Body Localization (MBL).
The "Boiling" State (Thermalization): In a boiling state, everyone is dancing wildly, swapping partners, and mixing everything up until the whole floor looks the same. The system has "thermalized," meaning it has forgotten its starting point and reached equilibrium.

For a long time, physicists believed that if you made the "noise" (disorder) on the dance floor strong enough, the dancers would stay frozen forever, no matter how big the dance floor got. However, recent computer simulations have shown a confusing problem: as the dance floor gets bigger, the system seems to slowly start "thawing" and mixing, even when the noise is supposed to be strong enough to keep it frozen.

The Paper's Goal: The authors want to explain why this slow thawing happens. They argue it's caused by a chain reaction of "resonances."

The Core Concept: The "Resonance Chain Reaction"

Think of a resonance like two people on the dance floor who happen to have the exact same rhythm. Even if they are far apart, they can start swapping energy and moving together.

The Spark: At first, only a few pairs of dancers find each other's rhythm. They start a slow, rhythmic wobble (a resonance).
The Chain Reaction (Proliferation): Here is the tricky part. Once a pair starts wobbling, they change the rhythm of the people around them. This makes it easier for other pairs to find a matching rhythm.
The Avalanche: If this happens enough, you get a runaway effect. One pair wobbles, which helps two more pairs wobble, which helps four more, and so on. Eventually, the whole dance floor starts wobbling together, and the system "thaws" (thermalizes).

The paper asks: What determines whether the wobbles stay small and isolated, or explode into a full-blown chain reaction?

The Tool: The "Jacobi Algorithm" as a Detective

To study this, the authors use a mathematical tool called the Jacobi Algorithm. Imagine this as a very organized detective trying to solve the mystery of the dance floor.

The Job: The detective looks at the entire list of connections between every dancer.
The Method: The detective finds the strongest connection (the loudest wobble) and "silences" it by rotating the dancers into a new position. Then, they look for the next loudest connection and silence that one too.
The Clue: As the detective works, they keep a log of the size of the connections they silence.
- If the connections get smaller and smaller very quickly, the system is frozen (localized).
- If the connections stay large or start getting bigger again as the detective digs deeper, the system is boiling (thermalizing).

The authors developed a statistical method (called the Statistical Jacobi Approximation or SJA) to predict what this log of connections will look like without having to simulate the entire dance floor every time.

The Key Discovery: The "Thermostat" Exponent ( $\theta$ )

The authors found a single number, which they call $\theta$ (theta), that acts like a thermostat for the system. This number tells us how the "loudness" of the connections changes as the detective digs deeper.

$\theta$ is Positive (The Safe Zone): If $\theta$ stays positive, the connections get weaker and weaker. The chain reaction dies out. The system stays frozen. The dancers remain in their spots.
$\theta$ is Negative (The Danger Zone): If $\theta$ turns negative, the connections get stronger as you look deeper. The chain reaction takes off. The system melts into a boil.
The Tipping Point: The paper shows that there is a critical line. If the system starts with a positive $\theta$ but the "noise" is just right, the act of silencing the first few connections actually helps the next ones grow. $\theta$ flips from positive to negative, and the system crashes into thermalization.

What They Tested

The authors tested their theory on three different types of "dance floors":

Random Regular Graphs: A theoretical network where everyone is connected in a tree-like structure.
Levy-Rosenzweig-Porter Model: A random matrix model (a grid of numbers) with specific statistical properties.
Disordered Spin Chains: The standard model for real-world quantum materials (like a chain of magnets with random noise).

The Results:

In the first two models, their theory perfectly matched the computer simulations. They could predict exactly when the system would stay frozen and when it would melt.
In the third model (the real-world spin chain), they found the "slow drift" phenomenon. At intermediate levels of noise, the system starts out looking frozen ( $\theta$ is positive), but as the simulation digs deeper, $\theta$ flips negative. This explains why computer simulations see the system slowly thawing as it gets bigger: the "chain reaction" of resonances just needs more space (a bigger system) to get going.

The "Bounce" (Finite Size Effects)

The paper also explains a weird quirk in the computer data. When the system gets very close to melting, the numbers sometimes "bounce" back up, making it look like the system is freezing again. The authors explain this is an illusion caused by the system being too small. It's like trying to start a forest fire in a tiny pot; the fire starts to spread, but runs out of wood before it can really catch. In a truly infinite system, the fire would burn forever.

Summary

This paper provides a new mathematical "thermostat" ( $\theta$ ) to measure the stability of quantum systems. It explains that the slow melting of these systems isn't a glitch; it's a chain reaction of resonances. Just like a small spark can start a massive fire if the conditions are right, a few small quantum wobbles can trigger a cascade that eventually melts the entire system, explaining why larger systems seem less stable than smaller ones.

Technical Summary: Resonance Proliferation Across Localization Transitions

Problem Statement
Many-body localization (MBL) posits a robust phase of interacting quantum matter where generic initial states fail to thermalize. However, empirical numerical studies reveal a persistent issue: as system size increases, spectral probes of ergodicity (such as the $r$ -statistic) drift towards values characteristic of the ergodic phase, even at disorder strengths where localization is expected. This "finite-size drift" suggests a prethermal regime where thermalization occurs only at extremely late times or very large system sizes. While thermal avalanches (nucleated by rare weakly-disordered regions) are believed to destabilize the infinite-volume MBL phase, the slow drifts observed at numerically accessible sizes are plausibly attributed to the proliferation of many-body resonances. The paper addresses the lack of a satisfactory theoretical framework to describe how these resonances form, proliferate, and drive the crossover from localization to thermalization.

Methodology
The authors employ the Jacobi algorithm for diagonalizing Hamiltonians, viewing the iterative process as a probe of resonance formation. In this algorithm, the Hamiltonian is diagonalized via successive two-level rotations that decimate the largest off-diagonal matrix elements, denoted by $w$ . Large rotations (where the rotation angle $\eta \gtrsim \pi/4$ ) are identified as resonances.

To analyze the statistical properties of this process across disorder realizations, the paper utilizes the Statistical Jacobi Approximation (SJA). The SJA posits that the dynamics of local observables can be described by the statistical distribution of decimated matrix elements rather than tracking the exact sequence of rotations.

Sparse vs. Dense Regimes: The analysis distinguishes between a "sparse" regime (characteristic of localized systems where few large off-diagonal elements connect sites) and a "dense" regime (characteristic of thermalizing systems where matrix elements become uniformly distributed).
Resonance Exponent $\theta(w)$ : The core of the methodology is the definition of a scale-dependent exponent $\theta(w)$ that characterizes the power-law distribution of decimated elements, $\rho_{\text{dec}}(w) \propto w^{-2+\theta(w)}$ . The number of resonances above a scale $w$ , $n_{\text{res}}(w)$ , is derived from this distribution.
Flow Equations: The authors derive flow equations for $\theta(w)$ $θ (w)$ as the scale $w$ $w$ decreases (analogous to increasing flow time). Two derivations are presented:
1. A heuristic derivation (Sec. III) assuming a power-law ansatz for the matrix element distribution at all scales.
2. A systematic derivation (Sec. V) based on a functional flow equation for the distribution of matrix elements $\rho_H(h|n)$ , incorporating perturbations to account for matrix elements increasing under rotation (a necessary condition for resonance proliferation).

Key Contributions and Results

Flow Equation for Resonance Proliferation:
The paper derives a flow equation for $\theta(w)$ .
- In the localized phase, $\theta(w)$ flows to a stable, positive value ( $0 < \theta \le 1$ ), indicating that the number of resonances per state remains finite as $w \to 0$ . This phase is described by a line of fixed points.
- In the thermalizing (delocalized) phase, $\theta(w)$ becomes negative and diverges to $-\infty$ at a finite scale $w_c$ . This divergence signals the instability of the localized phase due to resonance proliferation.
- The transition between these phases is characterized by a separatrix flowing to $\theta(0)=0$ .
Universality Class and Critical Behavior:
- The heuristic flow equation predicts that the transition belongs to the Berezinskii–Kosterlitz–Thouless (BKT) universality class. This implies that the thermalization time $\tau$ diverges faster than any power law as the disorder strength approaches the critical value ( $\log \log \tau \sim \sqrt{\log(1/\Delta W)}$ ).
- The systematic functional flow equation (Sec. V) recovers the instability for $\theta < 0$ but predicts power-law scaling for the critical behavior, differing from the BKT prediction. The authors note that independent analyses of Hilbert space localization on Random Regular Graphs (RRG) favor the BKT scaling, suggesting the heuristic approach may capture the correct critical exponents despite its simplifications.
Numerical Validation:
The predicted flow of $\theta(w)$ is compared against exact Jacobi diagonalization numerics for three models:
- Levy-Rosenzweig-Porter (LRP) Random Matrix Ensemble: Shows clear agreement with the flow equation predictions, capturing the transition from localized ( $\theta > 0$ ) to delocalized ( $\theta < 0$ ) regimes.
- Anderson Model on Random Regular Graphs (RRG): Demonstrates qualitative agreement, including the "ramp" at large $w$ (due to initialization) and the subsequent flow towards negative $\theta$ in the delocalized phase.
- Disordered XXZ Spin Chain (Standard MBL Model): At intermediate disorder strengths, the numerics show $\theta(w)$ starting positive (sparse regime) but flowing to negative values as $w$ decreases. This captures the "slow thermalization" regime where resonance proliferation eventually drives delocalization, explaining the observed finite-size drifts.
Connection to Observables:
The paper establishes that the number of resonances $n_{\text{res}}(w)$ serves as a proxy for the Participation Ratio (PR) of eigenstates. Furthermore, the flow of $\theta(w)$ predicts the functional form of dynamical observables:
- In the prethermal regime (slowly varying $\theta$ ), autocorrelation functions and return probabilities exhibit stretched exponential decay with a slowly varying exponent.
- Near the divergence of $\theta$ , the decay crosses over to exponential behavior.

Significance and Claims
The paper claims to provide a significant step towards explaining the finite-size drifts observed in MBL simulations. By framing the problem in terms of resonance proliferation within the Hilbert space, the SJA offers a scale-dependent treatment of quantum dynamics analogous to Renormalization Group (RG) schemes, albeit without eliminating degrees of freedom.

The authors conjecture that the crossover in prethermal MBL is controlled by resonance proliferation, described by the SJA flow, before being cut off by other effects (like avalanches) or finite-size limits. They assert that the SJA successfully captures the physics of resonance-driven delocalization transitions in models like the LRP and RRG, and provides a qualitative description of the slow thermalization in real-space MBL models at intermediate disorder. The work suggests that the instability of the MBL phase at intermediate disorder is driven by the renormalization of the localization length via resonance formation, leading to a flow from a stable localized fixed point to a thermalizing instability.

The paper remains modest regarding the precise nature of the critical transition (BKT vs. power-law), acknowledging that the systematic functional derivation misses certain quantitative features (like the specific BKT exponents) found in the heuristic approach and independent RG analyses. It leaves the unification of resonance physics with rare-region avalanche effects as an open question for future work.