Mobility Edge for the Anderson Model on Random Regular Graphs

This paper proves that for the Anderson tight-binding model on random regular graphs with sufficiently large fixed degree and Gaussian disorder, the spectrum asymptotically consists of a finite delocalized interval flanked by two unbounded localized regions, thereby establishing the existence of a mobility edge by transferring spectral characterizations from the Bethe lattice limit to finite graphs.

The Gist

Imagine you are trying to walk through a crowded, chaotic city. Sometimes, the crowd is so disorganized and the obstacles so random that you get stuck in one spot, unable to move forward. Other times, the path is clear enough that you can wander freely across the whole city.

This paper is about figuring out exactly when you get stuck and when you can move freely in a very specific, mathematical version of this city.

Here is the breakdown of the story, the characters, and the big discovery, explained without the heavy math jargon.

The Setting: The "Disordered City"

In physics, this is called the Anderson Model.

• The City: A giant network of streets (a graph). In this paper, the city is a "Random Regular Graph." Imagine a city where every single house has exactly the same number of neighbors (say, 10), but the layout is completely random, like a tangled ball of yarn.
• The Walkers: Tiny particles (like electrons) trying to hop from house to house.
• The Disorder: Every house has a random "noise" or "obstacle" (like a bumpy floor or a loud noise) that tries to trap the walker.
• The Tug-of-War: There are two forces fighting:
  1. The Hop: The natural desire to move to the next house.
  2. The Trap: The random noise that wants to freeze the walker in place.

The Big Question: The "Mobility Edge"

For a long time, physicists knew that if the noise is very loud, everyone gets stuck (Localization). If the noise is very quiet, everyone runs free (Delocalization).

But they suspected there was a middle ground. They thought there might be a specific "line" in the city (called a Mobility Edge) where:

• Near the center of the city's energy spectrum, walkers can roam freely.
• Near the edges of the spectrum, walkers get trapped.

The big mystery was: Does this line actually exist in a finite, random city, or is it just a trick of infinite math?

The Strategy: The "Perfect Tree" vs. The "Real City"

To solve this, the authors used a clever trick involving two types of maps:

1. The Perfect Tree (The Bethe Lattice): Imagine a city that branches out perfectly like a tree, with no loops. It's a simplified, infinite version of the city. Mathematicians recently figured out exactly where the "Mobility Edge" is on this perfect tree.
2. The Real City (The Random Graph): This is the messy, finite city with loops and dead ends that we actually care about.

The Analogy:
Think of the Perfect Tree as a perfectly smooth, endless highway. You know exactly how fast you can drive on it.
The Real City is like a busy, winding neighborhood.
The authors' goal was to prove that if you zoom in close enough on the neighborhood, it looks exactly like the highway. Therefore, the rules that govern the highway should also govern the neighborhood.

The Discovery: The "Phase Diagram"

The authors proved that for a city with enough connections (a high "degree"), the map looks like this:

• The Center (Delocalized): In the middle of the energy spectrum, the particles are free agents. They spread out evenly across the whole city. They are not stuck in one house; they are everywhere at once.
• The Edges (Localized): As you move toward the very high or very low energy levels, the particles get trapped. They huddle in small clusters and refuse to move.
• The Edge (The Mobility Edge): There is a sharp, distinct line separating the "Free Zone" from the "Trapped Zone."

Why This Matters

This is a huge deal for a few reasons:

1. Math vs. Physics: Physicists have been guessing about this "Mobility Edge" for decades using simulations and intuition. This paper provides the first rigorous mathematical proof that this edge actually exists in a finite, random network. It turns a "hunch" into a "fact."
2. Quantum Computers: This relates to Many-Body Localization, a phenomenon that might help us build better quantum computers. If particles get stuck (localized), they might be able to store information without it getting scrambled by the environment. Understanding exactly when they get stuck is crucial for engineering these machines.
3. The "Finite" Problem: Many math proofs only work on infinite, perfect worlds. This paper bridges the gap, showing that the infinite rules actually hold true for finite, real-world-sized systems.

The "Secret Sauce" of the Proof

How did they do it?
They didn't just look at the whole city at once. They looked at the resolvent (a fancy math tool that measures how a particle reacts to a specific energy).

They proved that:

1. If the "reaction" on the Perfect Tree is weak, the particle is stuck in the Real City.
2. If the "reaction" on the Perfect Tree is strong, the particle is free in the Real City.

They had to overcome some tricky hurdles, like proving that the random loops in the city don't mess up the "tree-like" behavior too much. They used a technique called bootstrapping (using a small proof to build a bigger one) to show that the particles' behavior stabilizes as the city gets bigger.

The Bottom Line

This paper confirms that in a disordered world, there is a clear dividing line between chaos (where things move freely) and order (where things get stuck). It's like finding the exact temperature where water turns to ice: below that line, the flow stops; above it, the flow is free.

They didn't just say "it happens"; they drew the exact map of where it happens, proving that even in a messy, random world, there is a hidden, sharp structure.

Technical Summary

1. Problem Statement

The paper addresses the Anderson localization phenomenon in the context of the tight-binding model on random regular graphs (RRGs).

• The Model: The Hamiltonian is defined as $H = -tA + V$, where $A$ is the adjacency matrix of a random $d$-regular graph on $N$ vertices, $V$ is a diagonal matrix of independent Gaussian random variables (disorder), and $t$ is the inverse disorder strength (coupling constant).
• The Phenomenon: In disordered systems, wave functions can be either localized (exponentially decaying, particle trapped) or delocalized (extended, particle diffuses). The transition between these phases is marked by mobility edges.
• The Gap: While the behavior of the Anderson model on the Bethe lattice (infinite $d$-regular tree) is well-understood rigorously (including the existence of mobility edges for large $d$), the behavior on finite random regular graphs has been less clear. RRGs are the natural finite-size analogs of the Bethe lattice, but they contain long loops and finite-size effects that complicate the transfer of results from the infinite tree.
• Goal: To rigorously prove the existence of a mobility edge for the Anderson model on random regular graphs with fixed, sufficiently large degree $d$ and Gaussian disorder, showing that the spectrum consists of a central delocalized interval flanked by localized regions.

2. Methodology

The authors employ a strategy of transferring results from the limiting infinite model (Bethe lattice) to the finite model (RRG) through a series of rigorous probabilistic estimates.

A. The Limiting Model (Bethe Lattice)

The proof relies heavily on a recent characterization of the Bethe lattice spectrum by Aggarwal and Lopatto (2025). On the Bethe lattice, the existence of mobility edges is established via the behavior of the resolvent (Green's function) at the root, $R_{00}(z)$.

• Localization: Corresponds to $\lim_{\eta \to 0} \text{Im } R_{00}(E+i\eta) = 0$ in probability.
• Delocalization: Corresponds to $\lim_{\eta \to 0} \text{Im } R_{00}(E+i\eta) > c$ with positive probability.

B. Bridging Finite and Infinite Graphs

The core technical challenge is proving that the spectral properties of the finite RRG converge to those of the Bethe lattice. The authors establish four key pillars:

1. Local Tree-Likeness: RRGs are locally tree-like. The authors use the configuration model and Azuma-Hoeffding martingale concentration to show that the neighborhood of a vertex in an RRG is isomorphic to a tree with high probability as $N \to \infty$.
2. Resolvent Convergence: They prove that the diagonal resolvent entries of the RRG, $G_{kk}(z)$, converge in probability to the root resolvent of the Bethe lattice, $R_{00}(z)$. This is achieved using Combes-Thomas bounds (exponential decay of resolvent entries with distance) to show that the "boundary" of a large finite ball has negligible effect on the root.
3. Concentration of Resolvent Moments: To handle the randomness of the graph structure and the disorder, the authors prove concentration inequalities for fractional moments of the resolvent. They combine Gaussian concentration (for the potential $V$) with martingale arguments based on edge-switching (for the graph adjacency $A$).
4. Eigenvalue Density Bounds: They establish upper and lower bounds on the eigenvalue counting function (density of states) to ensure that the intervals of interest contain a macroscopic number of eigenvalues, preventing trivial concentration effects.

C. Criteria for (De)localization

The authors define observables based on the inverse participation ratio (IPR) and its generalizations. Specifically, they analyze $Q_I$, a quantity measuring the concentration of eigenvectors in an energy interval $I$.

• If $Q_I$ remains bounded as the interval shrinks, eigenvectors are delocalized.
• If $Q_I$ diverges, eigenvectors are localized.
  They prove that the behavior of $Q_I$ is directly determined by the limit of the imaginary part of the resolvent at the root.

3. Key Contributions and Results

Main Theorem (Theorem 1.3)

For a sufficiently large fixed degree $d$ and Gaussian disorder with variance 1, scaled by $t = g(d \ln d)^{-1}$, there exists a critical energy $M$ (the mobility edge) such that:

1. Localized Phase: For energies $|E| > M$, the system exhibits Anderson localization. The eigenvectors are exponentially localized, and the spectral measure is pure point.
2. Delocalized Phase: For energies $|E| < M$, the system exhibits delocalization. The eigenvectors are extended (not concentrated in small balls), and the spectral measure contains an absolutely continuous component.
3. Sharp Transition: The transition between these phases is sharp, defined by the condition $\rho(E) = 1/(4g)$, where $\rho$ is the Gaussian density.

Technical Innovations

• Uniform Moment Bounds: A significant technical contribution is Lemma 3.6, which proves a uniform bound on the second moment of the imaginary part of the resolvent in the delocalized regime. This is achieved via a bootstrap argument on the tail decay of the resolvent's imaginary part, moving from a Cauchy-type bound to a polynomial decay bound.
• Rigorous Transfer: The paper provides the first rigorous proof that the phase diagram of the Bethe lattice (specifically the existence of mobility edges) holds for finite random regular graphs, closing the gap between theoretical physics predictions and mathematical rigor for this specific model.

4. Significance

• Mathematical Physics: This work resolves a long-standing question regarding the validity of the Bethe lattice approximation for finite-dimensional disordered systems. It confirms that the "mean-field" behavior of the Bethe lattice persists in random regular graphs, which are the standard proxy for high-dimensional lattices.
• Many-Body Localization (MBL): The tight-binding model on RRGs is a crucial proxy for understanding Many-Body Localization in Fock space, where the effective graph is locally tree-like. Establishing the existence of mobility edges in this single-particle model provides a foundational step for understanding the stability of MBL phases.
• Methodological Advancement: The combination of Combes-Thomas estimates, configuration model coupling, and martingale concentration inequalities provides a robust toolkit for analyzing spectral properties of random operators on sparse graphs.

5. Conclusion

Liu and Lopatto successfully demonstrate that for random regular graphs with large fixed degree and Gaussian disorder, the spectrum undergoes a sharp transition from localized to delocalized states. By rigorously linking the finite graph to the Bethe lattice limit, they confirm the existence of mobility edges, validating the theoretical predictions of the Anderson model in this high-dimensional setting. The result is a complete phase diagram description for the model in the large-degree limit.