Renormalization-Group Analysis of the Many-Body Localization Transition in the Random-Field XXZ Chain

This paper argues that the Many-Body Localization transition in the random-field XXZ chain is governed by a two-parameter, Berezinskii-Kosterlitz-Thouless-like flow with a line of fixed points rather than the previously assumed one-parameter scaling and isolated Wilson-Fisher fixed point.

The Gist

The Big Picture: A Party That Never Ends (or Does It?)

Imagine a crowded dance party.

• The "Ergodic" State (Normal): Everyone is dancing, mixing, and swapping partners. Eventually, the energy spreads out evenly. If you look at any random group of people, they look like a mix of the whole crowd. This is how normal matter behaves; it "thermalizes" (reaches a comfortable, mixed state).
• The "Many-Body Localized" (MBL) State (The Freeze): Now, imagine the music stops, but the lights go out, and everyone is glued to their spot. No one moves, no one swaps partners, and the energy stays trapped where it started. The system never mixes. It remembers exactly how it started, forever.

For decades, physicists have been arguing about a specific type of quantum "party" (a chain of spinning atoms with random magnetic fields). They asked: If we add enough "disorder" (randomness) to the room, will the particles get stuck in the MBL state forever, or will they eventually break free and start dancing again?

This paper tries to answer that question by looking at the "traffic rules" of the universe, using a tool called the Renormalization Group (RG).

---

The Tool: The "Zoom Lens" (Renormalization Group)

To understand this, imagine you are looking at a forest through a camera lens.

1. Zoomed in: You see individual leaves and twigs (tiny quantum particles).
2. Zoomed out: You see the shape of the trees and the forest as a whole.

In physics, the Renormalization Group (RG) is like a camera that zooms out step-by-step. It asks: "If I look at this system from a slightly larger distance, does it look the same, or does it change?"

• If the system keeps changing as you zoom out, it's likely heading toward a "mixed" state (Ergodic).
• If the system stays frozen and looks the same no matter how much you zoom out, it's "Localized."

The authors use this "zoom lens" to trace the path (or trajectory) of the system as it gets bigger.

---

The Two Possible Stories

The authors looked at the data and found that the "traffic rules" (the math describing how the system changes) don't fit the old, simple story. Instead, they found two possible scenarios:

Scenario A: The "One-Parameter" Story (The Old Idea)

• The Metaphor: Imagine a ball rolling down a hill with a single, sharp peak in the middle.
• The Physics: If the disorder is weak, the ball rolls to the "mixed" side. If it's strong, it rolls to the "frozen" side. There is a clear, sharp line (a critical point) where the ball stops right at the peak.
• The Problem: When the authors tried to fit their data to this simple hill, the math didn't work. The "ball" didn't stop at a sharp peak; it behaved strangely.

Scenario B: The "Two-Parameter" Story (The New Idea)

• The Metaphor: Imagine a long, flat valley (a Line of Fixed Points) that ends at a cliff.
  • The Valley: This represents the Localized Phase. Once the system falls into this valley, it stays there. It doesn't need to roll down a steep hill; it just sits in a flat, stable groove.
  • The Cliff: This is the Critical Point. It's the very end of the valley.
  • The Other Side: If the system starts on the other side of the cliff, it rolls away into the "mixed" chaos.
• The Physics: The authors suggest that the transition isn't a sharp peak, but a long, flat road of "frozen" states that only ends at the transition point. This is similar to a famous physics concept called the BKT transition (named after physicists Berezinskii, Kosterlitz, and Thouless), which describes how vortices in a superfluid behave.

---

The "Newtonian Particle" Analogy

To make sense of the messy data, the authors did something clever. They turned the math into a story about a particle moving in a landscape.

• The Particle: Represents the state of the system (how "frozen" or "mixed" it is).
• The Landscape (Potential): A hilly terrain.
  • If the landscape is a confining valley (like a bowl), the particle is trapped. No matter how much disorder you add, it eventually bounces back and mixes. Result: No MBL phase exists.
  • If the landscape is non-confining (like a ramp that goes down forever), the particle can slide all the way to the bottom and stay there. Result: MBL exists.

By analyzing the data, the authors found the landscape looks like the ramp. The particle can slide down into a "frozen" state and stay there, suggesting that MBL is real in this system.

---

The "Noise" Problem (Why is this hard?)

Here is the catch: The authors are looking at very small systems (like a chain of 20 atoms).

• The Analogy: Imagine trying to hear a whisper in a hurricane.
• In the "frozen" state, the signal (the fact that particles are stuck) gets weaker and weaker as the system gets bigger. Eventually, it becomes so small that it looks like random static (noise).
• The authors had to do a massive amount of computer simulations (millions of samples) to prove that the "whisper" wasn't just random noise. They used a statistical test (Kolmogorov-Smirnov) to say, "Yes, this is a real signal, not just a fluke."

---

The Conclusion: What Did They Find?

1. The Old Model Failed: The system doesn't behave like a simple hill with a sharp peak. The "One-Parameter" scaling idea (the old way of thinking) doesn't fit the data.
2. The New Model Fits: The system behaves like a long, flat valley (a line of fixed points) that ends at a cliff. This is a "Two-Parameter" scaling.
3. The Verdict: Based on this new map, it looks like Many-Body Localization does exist in this specific chain of atoms. There is a critical point (around a disorder strength of $W \approx 4.4$) where the system switches from mixing to freezing.
4. The Caveat: Because the "frozen" signal is so weak and hard to measure, we need even bigger computers and more data to be 100% sure. But the current evidence strongly points to the "Two-Parameter" story being the correct one.

In a Nutshell

The authors took a complex quantum problem, mapped it onto a "landscape" where a particle rolls, and found that the terrain allows for a stable "frozen" state. This suggests that even when particles interact, they can get stuck in a permanent state of disorder, defying the usual laws of thermodynamics. It's a victory for the "frozen" party guests!

Technical Summary

1. Problem Statement

The paper addresses the long-standing debate regarding the existence and nature of the Many-Body Localization (MBL) transition in one-dimensional quantum systems, specifically the random-field Heisenberg XXZ spin chain.

• The Controversy: While perturbative theories (BAA) and early numerical studies suggested a stable MBL phase exists at strong disorder, recent non-perturbative arguments (e.g., thermal avalanches, many-body resonances) suggest that the localized phase might be unstable and that the system eventually thermalizes (ergodic) at any finite disorder strength in the thermodynamic limit.
• The Challenge: Standard numerical methods (Exact Diagonalization) are limited to small system sizes ($L \le 20$), making it difficult to distinguish between a true phase transition and a slow crossover or pre-thermal regime. Traditional one-parameter scaling hypotheses often yield inconsistent results (e.g., violating Harris bounds or requiring unphysical critical exponents) when applied to this data.
• The Goal: To reconstruct the Renormalization Group (RG) flow of the system from numerical data to determine if the MBL transition follows a standard one-parameter scaling (Wilson-Fisher type) or a more complex two-parameter scaling (Berezinskii–Kosterlitz–Thouless or BKT-like) scenario.

2. Methodology

The authors employ a rigorous RG analysis based on finite-size scaling of numerical data obtained via Exact Diagonalization (ED).

• Model: The random-field XXZ spin chain Hamiltonian:
  $$\hat{H} = J \sum_{i=1}^L \hat{S}_i \cdot \hat{S}_{i+1} + \sum_{i=1}^L h_i \hat{S}^z_i$$
  where $h_i$ are uniform random variables in $[-W, W]$. The disorder strength is $W$ (with $J=1$).
• Observable: The rescaled spectral gap ratio ($r$-parameter), denoted as $\phi$.
  $$\phi = \frac{r - r_P}{r_{WD} - r_P}$$
  where $r_P \approx 0.386$ (Poisson, localized) and $r_{WD} \approx 0.5307$ (Wigner-Dyson, ergodic). $\phi=0$ implies localization, $\phi=1$ implies ergodicity.
• Beta Function Construction: The authors define the beta function $\beta(\phi) = d \ln \phi / d \ln N$, where $N$ is the system volume (Hilbert space dimension). They reconstruct $\beta$ numerically from polynomial interpolations of $\phi(L)$ data.
• Theoretical Framework:
  • One-Parameter Scaling: Assumes $\beta$ depends only on $\phi$. This leads to a simple fixed point structure.
  • Two-Parameter Scaling: Inspired by recent work on the Anderson model on Random Regular Graphs (RRG) and infinite dimensions, the authors treat the RG flow as a dynamical system in phase space $(\alpha, \beta)$, where $\alpha = \ln \phi$.
  • Newtonian Analogy: The RG equations are mapped to the motion of a classical particle in a potential $V(\alpha)$.
    • $\ddot{\alpha} = \gamma(\alpha) = -V'(\alpha)$.
    • Confining Potential: All trajectories return to the ergodic fixed point ($\phi=1$) $\rightarrow$ No MBL transition.
    • Non-Confining Potential: Trajectories can escape to $\phi=0$ if energy is high enough $\rightarrow$ MBL Transition exists.

3. Key Contributions

1. Rejection of Simple One-Parameter Scaling: The authors demonstrate that attempting to collapse the XXZ data using standard one-parameter scaling (assuming a simple critical fixed point) leads to contradictions, such as a critical exponent $\nu$ violating the Harris bound ($\nu > 2/d$) and a non-linear scaling function at the critical point.
2. Mapping to BKT-like Flow: They propose that the MBL transition is better described by a two-parameter scaling hypothesis resembling a BKT transition. In this scenario, the localized phase corresponds to a line of fixed points rather than a single isolated critical point. The critical point is the endpoint of this line.
3. Dynamical System Formulation: By mapping the RG flow to a Newtonian particle in a potential, they provide a physical intuition for the transition. They identify the force law governing the flow as $\gamma(\alpha) \propto e^\alpha$ (corresponding to a potential $V(\alpha) \sim -e^\alpha$), which is non-confining.
4. Statistical Significance Analysis: They perform a Kolmogorov-Smirnov test to quantify the statistical noise in the data. They show that distinguishing the exponential decay of the MBL phase from statistical fluctuations requires an exponentially large number of samples ($N_{sig} \sim e^{0.6L}$), explaining why previous studies might have been inconclusive.

4. Key Results

• Ergodic Regime ($W < W_c$): The RG trajectories flow toward the ergodic fixed point ($\phi=1$) and eventually attach to a universal one-parameter scaling curve $\beta_0(\phi)$.
• Localized/Critical Regime ($W \ge W_c$):
  • The numerical data for large disorder fits the trajectory of a particle in a non-confining potential $V(\alpha) = -c e^\alpha$ (with $n \approx 1$).
  • This implies a power-law decay of the order parameter near the critical point: $\phi(L) \sim L^{-2}$.
  • Critical Disorder ($W_c$): By analyzing the energy of the trajectories (sign of the "energy" in the Newtonian analogy) and the zeros of the beta function, the authors estimate the critical disorder to be $W_c \approx 4.4 \pm 0.4$.
  • Universality: The scaling exponent $n=1$ and the power-law decay $\phi \sim L^{-2}$ match the findings for the Anderson model on Random Regular Graphs (RRG), suggesting the XXZ chain and the infinite-dimensional Anderson model belong to the same universality class.
• Two Scenarios: The analysis supports two possibilities:
  1. Transition Scenario: A true MBL phase exists for $W > W_c$, characterized by a line of fixed points and a BKT-like transition.
  2. No-Transition Scenario: If the potential were confining (which the data argues against), the system would eventually thermalize for any $W$. The data strongly favors the non-confining (transition) scenario.

5. Significance

• Resolution of Numerical Ambiguity: The paper provides a coherent framework that explains why previous one-parameter scaling attempts failed. It suggests that the "critical" behavior is not a simple fixed point but a flow along a line of fixed points, which is harder to detect with small system sizes.
• Methodological Advance: The approach of reconstructing the full beta function and mapping it to a dynamical system offers a robust tool for analyzing phase transitions in systems where standard scaling hypotheses fail or are ambiguous.
• Implications for MBL Stability: The results suggest that if an MBL transition exists, it is of the BKT type, characterized by a line of fixed points. This challenges the notion of a simple Wilson-Fisher critical point for MBL and aligns the phenomenology of 1D interacting systems with high-dimensional non-interacting models (RRG).
• Future Directions: The authors note that while their analysis is consistent with existing computing power, definitive confirmation requires even larger system sizes and massive statistics to resolve the exponential decay in the deep localized regime.

In summary, the paper argues that the Many-Body Localization transition in the random-field XXZ chain is best understood as a Berezinskii–Kosterlitz–Thouless-like transition involving a line of fixed points, rather than a standard second-order phase transition, providing a more consistent explanation for numerical data than previous models.