Strong-disorder expansion of the root-averaged density… — Plain-Language Explanation

Imagine you are standing in the middle of an infinite, perfectly symmetrical forest. Every tree in this forest has exactly the same number of neighbors (let's say $q+1$ ). This is the Bethe lattice, a mathematical shape that looks like a tree but goes on forever without any loops.

Now, imagine that every tree in this forest has a hidden, random "weight" attached to it. Some are heavy, some are light, and the weights are chosen randomly according to a specific rule. This is the Anderson model.

Physicists and mathematicians want to know: "If I send a wave of energy through this forest, how does it spread out? What does the 'density' of these energy waves look like?" This is called the Density of States.

Usually, calculating this is incredibly hard because the randomness of the weights makes the waves bounce around in chaotic, unpredictable ways. However, this paper focuses on a specific scenario: Strong Disorder. This means the random weights on the trees are so heavy and varied that they dominate the system. The "hopping" between trees (the connection) becomes a tiny, almost negligible perturbation compared to the massive weights.

Here is the simple breakdown of what the author, Masahiro Kaminaga, discovered:

1. The "Zoomed-In" View

Because the disorder is so strong, the author suggests we "zoom out" or rescale our view. Instead of looking at the raw energy numbers, we look at them relative to the strength of the disorder ( $\lambda$ ). It's like looking at a mountain range through a telescope; the individual rocks (the random weights) become the main feature, and the small paths between them (the tree connections) become secondary details.

2. The Magic of the "Tree" Shape

The forest isn't just any shape; it's a tree. In a tree, if you start at the root and walk a certain number of steps, you can only return to the start if you take an even number of steps. If you take an odd number of steps, you are guaranteed to be somewhere else.

The author uses this simple fact to prove something surprising: All the "odd-numbered" corrections to the energy density vanish.

Think of the calculation as a recipe. You have a main ingredient (the random weights).
You add "correction" ingredients to account for the tree connections.
The author proves that the 1st, 3rd, 5th, etc., correction ingredients are exactly zero. You only need to worry about the 2nd, 4th, 6th, etc.

3. The "Walk" Analogy

To figure out exactly what the energy density looks like, the author imagines a "random walker" moving through the forest.

The walker starts at the root, takes a few steps, and must return to the root.
The author calculates how many different ways the walker can do this and how often they visit specific trees.
Because the forest is a tree, these "walks" are very structured. They don't get stuck in loops (because there are no loops).
The final formula for the energy density is a sum of these specific walking patterns.

4. The Result: A Smooth, Predictable Curve

Even though the weights are random, the author proves that if you look at the "average" energy density over a specific range, it is smooth and predictable.

The Leading Term: The most important part of the answer is simply the distribution of the random weights themselves. If the weights are uniformly distributed (like a flat line), the energy density starts as a flat line.
The Corrections: The tree connections add small ripples to this line. The author provides a precise formula for these ripples.
- The first ripple (the 2nd order correction) depends on how many neighbors each tree has ( $q$ ) and the shape of the random weight distribution.
- The author explicitly calculates this first ripple for the case where the weights are uniformly distributed.

5. Why This Matters (According to the Paper)

Before this paper, we knew the energy density existed, but we didn't have a precise, step-by-step recipe to calculate it for strong disorder.

The paper provides a finite-order expansion. This means you can calculate the answer as accurately as you want by adding more terms to the recipe.
It proves that the answer is analytic, meaning it's a very smooth curve without any sharp breaks or jagged edges in the region they studied.
It connects the complex math of "random walks on trees" directly to the physical property of "how energy is distributed."

Summary Analogy

Imagine you are trying to predict the average height of a crowd of people standing on a bumpy, uneven floor (the random weights).

Old way: You try to measure every single person and every bump, which is impossible.
This paper's way: You realize the floor is so bumpy that the people's own heights matter most. The bumps between them (the tree connections) only cause tiny, specific adjustments.
The Discovery: Because the floor is shaped like a tree, the "wobbles" caused by the connections cancel out in a very specific way (the odd terms disappear). The author gives you a formula to calculate exactly how the floor's shape tweaks the average height, term by term.

In short, the paper takes a chaotic, random system and shows that under strong disorder, it behaves in a surprisingly orderly, calculable, and smooth way, thanks to the unique geometry of the tree-like forest.

1. Problem Statement

The paper investigates the Anderson model on the infinite $(q+1)$ -regular tree (Bethe lattice), defined by the Hamiltonian:
$H_{\lambda, \omega} = A + \lambda V_\omega$
where $A$ is the adjacency operator, $\lambda > 0$ is the disorder strength, and $V_\omega$ is a multiplication operator with independent, identically distributed (i.i.d.) random potentials $\{\omega_x\}$ having a common law $\mu$ .

The primary objective is to analyze the root-averaged density of states (DOS) in the strong-disorder regime ( $\lambda \to \infty$ ). Specifically, the paper focuses on the scaled energy window $E = \lambda \xi$ , where $\xi$ lies within the support of the single-site distribution $\mu$ .

Key Distinction: Unlike amenable graphs (e.g., $\mathbb{Z}^d$ ) where the DOS is defined via finite-volume eigenvalue counting limits, the Bethe lattice is non-amenable. The boundary volume is comparable to the bulk volume. Therefore, the author defines the DOS measure $N_\lambda$ strictly as the root-averaged spectral measure:
$N_\lambda(B) = \mathbb{E}[\langle \delta_0, E_{H_{\lambda, \omega}}(B) \delta_0 \rangle]$
where $\delta_0$ is the basis vector at the root. The goal is to prove the absolute continuity of this measure and derive an explicit asymptotic expansion of its density $n_\lambda(E)$ in powers of $\lambda^{-1}$ .

2. Methodology

The author employs a combination of random-walk expansion and complex analysis to overcome the challenge of controlling the resolvent near the real axis (where the DOS is defined).

A. Random-Walk Expansion

The resolvent $G_\lambda(0,0;z) = \langle \delta_0, (H_{\lambda, \omega} - z)^{-1} \delta_0 \rangle$ is expanded using the Neumann series in the region where $\text{Im}(z)$ is large. By treating the hopping term $A$ as a perturbation to the diagonal term $\lambda V_\omega - z$ , the resolvent is expressed as a sum over closed walks $\gamma$ on the tree:
$G_\lambda(0,0;z) = \sum_{n=0}^\infty (-1)^n \sum_{\gamma \in \Gamma_n(0,0)} \prod_{j=0}^n \frac{1}{\lambda \omega_{x_j} - z}$
where $\Gamma_n(0,0)$ is the set of closed walks of length $n$ starting and ending at the root. Taking the expectation $\mathbb{E}$ and scaling $z = \lambda \zeta$ leads to an expansion involving products of single-site Stieltjes transforms $s_k(\zeta) = \int \frac{d\mu(t)}{(t-\zeta)^k}$ .

B. Complex-Analytic Continuation

A major hurdle is that the standard Neumann series only converges for $\text{Im}(z) > q+1$ , which is far from the real axis where the DOS is needed.

Assumption: The single-site distribution $\mu$ has compact support and a density $\rho$ that is locally analytic on an interval $I^\sharp$ containing the target interval $I$ .
Technique: The author uses a Cauchy integral argument to show that the single-site Stieltjes transforms $s_k(\zeta)$ admit a holomorphic continuation from the upper half-plane to a complex neighborhood $\Omega_\delta(I)$ of the real interval $I$ .
Extension: Since the coefficients of the random-walk expansion are finite sums of products of these $s_k(\zeta)$ , the entire averaged resolvent $m_\lambda(\lambda \zeta)$ inherits this holomorphic continuation.

C. Tree Structure Exploitation

The specific geometry of the Bethe lattice is crucial:

Bipartite Nature: The tree is bipartite, meaning any closed walk starting and ending at the root must have even length. Consequently, all odd-order terms in the expansion vanish ( $M_n \equiv 0$ for odd $n$ ).
Occupation Profiles: Because the graph is a tree (no loops other than backtracking), the contribution of a closed walk is determined entirely by the occupation profile (how many times each vertex is visited) and the finite subtree visited. This allows the coefficients to be computed as finite combinatorial sums.

3. Key Contributions and Results

A. Main Theorem (Resolvent Expansion)

Under the local analyticity assumption on $\mu$ , the author proves that for sufficiently large $\lambda$ , the scaled averaged diagonal resolvent $m_\lambda(\lambda \zeta)$ admits a holomorphic continuation to a complex neighborhood of $I$ . It possesses a uniform asymptotic expansion:
$m_\lambda(\lambda \zeta) = \sum_{n=0}^N \lambda^{-n-1} M_n(\zeta) + R_{N,\lambda}(\zeta)$
where:

$M_n(\zeta)$ are holomorphic functions given by finite sums over closed walks of length $n$ .
The remainder $R_{N,\lambda}$ is bounded by $O(\lambda^{-N-2})$ .
Odd coefficients vanish: $M_n \equiv 0$ for all odd $n$ .

B. Density of States Expansion

By applying the Stieltjes inversion formula to the holomorphic continuation, the paper establishes that the root-averaged DOS measure is absolutely continuous on the scaled window $\lambda I$ . Its density $n_\lambda(E)$ has the expansion:
$n_\lambda(\lambda \xi) = \sum_{n=0}^N \lambda^{-n-1} a_n(\xi) + r_{N,\lambda}(\xi)$
where $a_n(\xi) = \pi^{-1} \text{Im} M_n(\xi)$ .

Specific Properties of Coefficients:

Leading Term ( $n=0$ ): $a_0(\xi) = \rho(\xi)$ , the density of the single-site distribution. This confirms the heuristic that in the strong-disorder limit, the DOS is dominated by the local potential.
Vanishing Odd Terms: $a_n(\xi) = 0$ for all odd $n$ .
Higher Order Terms: The coefficients $a_n$ for even $n \geq 2$ are finite sums determined by the occupation profiles of short closed walks on the tree. They depend explicitly on the branching number $q$ and the derivatives of the Stieltjes transforms of $\mu$ .

C. Explicit Calculation for Uniform Distribution

The paper explicitly computes the first non-zero correction term for the case where $\mu$ is the uniform distribution on $[-a, a]$ .

$a_0(\xi) = \frac{1}{2a}$
$a_2(\xi) = -\frac{q+1}{2a(a^2 - \xi^2)}$
The expansion becomes:
$n_\lambda(\lambda \xi) = \frac{1}{2a}\lambda^{-1} - \frac{q+1}{2a(a^2 - \xi^2)}\lambda^{-3} + O(\lambda^{-5})$
This result highlights that the correction term diverges as $\xi$ approaches the edges of the support ( $\pm a$ ), consistent with the requirement that the expansion holds strictly inside the support.

4. Significance

First Explicit Coefficient-Level Expansion: Prior to this work, strong-disorder expansions for the DOS on the Bethe lattice were either qualitative or lacked explicit coefficient formulas. This paper provides the first rigorous derivation of explicit, finite-order expansions where coefficients are combinatorial sums over tree walks.
Rigorous Justification of Heuristics: It mathematically validates the physical intuition that strong disorder leads to a DOS dominated by the single-site distribution, with hopping terms providing systematic, calculable corrections.
Methodological Advancement: The combination of random-walk expansions with complex-analytic continuation of Stieltjes transforms offers a robust framework for studying spectral properties of random operators on non-amenable graphs, bypassing the limitations of finite-volume approximations.
Structural Insight: The proof elegantly utilizes the tree structure (bipartiteness and lack of cycles) to simplify the expansion, showing that the "tree-ness" leads to the vanishing of odd-order terms and the computability of higher-order terms.

In summary, Kaminaga provides a rigorous mathematical foundation for the strong-disorder behavior of the Anderson model on the Bethe lattice, delivering a precise asymptotic series for the density of states that connects local disorder statistics with global spectral properties through the geometry of the underlying tree.

Strong-disorder expansion of the root-averaged density of states for the Anderson model on the Bethe lattice