Logarithmic regularity of spectral measures on infinite graphs

This paper establishes that the expected spectral measures of self-adjoint operators on infinite unimodular weighted graphs satisfy a logarithmic Hölder regularity estimate under natural geometric conditions, extending the classical Craig–Simon theorem beyond Euclidean lattices to diverse settings including group algebras, random operators, and quasi-transitive graphs.

The Gist

Imagine you are trying to understand the "sound" of a massive, infinite instrument. In mathematics, this instrument is an infinite graph (a network of dots and lines that goes on forever), and the "sound" is its spectrum.

The spectrum tells you what frequencies (or energy levels) the system can vibrate at. Usually, these vibrations come in two flavors:

1. Discrete notes: Like a piano key, where the sound is a sharp, distinct spike.
2. Continuous noise: Like a violin bow sliding across a string, where the sound is a smooth smear of frequencies.

This paper, written by Charles Bordenave, asks a specific question: How "smooth" is the noise? If you look at a tiny slice of the spectrum (a very small interval of frequencies), how much "sound" (probability) is packed into that slice?

The author proves that for a wide class of these infinite networks, the sound is incredibly smooth. It doesn't just avoid sharp spikes; it avoids them so thoroughly that the amount of sound in a tiny interval shrinks very slowly as the interval gets smaller. Specifically, the paper proves a "logarithmic regularity" rule.

The Core Metaphor: The Infinite Hotel and the Elevator

To understand how the proof works, imagine an infinite hotel where every room is a point on the graph. The "operator" is a rule that tells you how to move from one room to another (like a random walk or a wave traveling through the network).

The author uses a clever trick called "Monotone Labelling" (which he improved from previous work). Think of this as assigning a floor number to every room in the hotel.

1. The Elevator Trick: The author finds a special "elevator" (a mathematical map to the integers) that lets you order the rooms. You can say, "Room A is on floor 10, Room B is on floor 11."
2. The "Prodigy" Rooms: In this ordering, some rooms are special "Prodigy" rooms. A room is a Prodigy if it has a neighbor on a lower floor, and all its other neighbors are on even lower floors.
3. The Logic: If you try to create a sharp, distinct "note" (an atom in the spectrum) that is trapped in a small area, the math shows that the wave function (the vibration) would have to grow impossibly fast as it moves up the floors. Because the "elevator" forces a specific structure on the connections, the wave gets "squeezed" out. It can't stay sharp; it has to spread out.

The author strengthens this idea by showing that even if the hotel has complex, random decorations (random weights on the connections), as long as the building has a certain "directional" structure (called indicability, meaning you can map the infinite network onto a simple line of integers), the sound remains smooth.

What Did They Actually Prove?

The paper establishes three main results, moving from simple to complex:

1. Group Algebras (The Pure Math Case):
   If your infinite graph is built from a specific type of group (a mathematical structure with a "direction" you can follow, like a free group or a surface group), the spectrum has no sharp spikes. The amount of "sound" in a tiny interval $I$ is bounded by a formula involving the natural log of the interval's size.

   • Analogy: No matter how small a slice of the frequency spectrum you take, you will never find a single, isolated note. It's always a smear.

2. Random Operators (The "Anderson" Model):
   The author extends this to graphs where the connections are random (like the famous "Anderson model" in physics, which models electrons in a disordered material). Even if the material is messy and random, as long as the underlying grid has that "directional" structure, the spectrum remains smooth.

   • Analogy: Imagine a forest where the trees are randomly placed. Usually, you might expect chaotic, jagged patterns. But if the forest is planted on a grid that has a "slope," the chaos smooths out. The "density of states" (how many energy levels exist) follows the same logarithmic rule.

3. Quasi-Transitive Graphs (The Complex Case):
   Finally, the paper handles graphs that look the same from a distance but might have different "local" structures (like a crystal with a repeating pattern that has a few different types of atoms). The author shows that you can break these complex graphs down into smaller, manageable blocks and apply the same logic.

   • Analogy: Think of a tiled floor where the pattern repeats, but some tiles are slightly different colors. You can still predict the overall "sound" of the floor by looking at how the tiles connect in the repeating pattern.

The "So What?" (According to the Paper)

The paper explicitly states that these results:

• Extend the Craig-Simon Theorem: This is a famous old result that only worked for grids in standard space (like $\mathbb{Z}^d$). This paper proves it works for much more complex, infinite shapes.
• Apply to Specific Groups: It works for groups like "Artin groups," "braid groups," and "surface groups."
• Handle Randomness: It works for "Anderson-type models" (disordered systems) and "anisotropic percolation" (randomly broken connections), provided the randomness doesn't break the underlying directional structure.

Crucially, the paper does not claim:

• That this solves problems in quantum computing or medical imaging.
• That it predicts the behavior of real-world materials in a lab.
• That it works for every possible infinite graph (it requires a specific geometric condition called "unimodularity" and "indicability").

Summary in One Sentence

By using a clever "floor-numbering" system to organize infinite networks, the author proves that for a vast class of these networks, the energy levels are so smoothly distributed that they cannot form sharp, isolated spikes, a result that holds true even when the network is random or complex.

Technical Summary

Technical Summary: Logarithmic Regularity of Spectral Measures on Infinite Graphs

Problem Statement
This paper investigates the regularity of spectral measures associated with self-adjoint, locally defined operators on infinite weighted graphs. The study operates within the framework of unimodular random graphs, a setting that encompasses:

1. Operators in the group algebra $\mathbb{C}[\Gamma]$ of a finitely generated group $\Gamma$.
2. Random operators whose distributions are quasi-invariant under a group action (e.g., Anderson-type models).
3. Benjamini–Schramm limits of sequences of operators on finite graphs.

The central objective is to establish quantitative bounds on the expected spectral measure $\mu$ (or the density of states) for intervals $I$ in the spectrum. Specifically, the paper seeks to prove a logarithmic Hölder regularity estimate of the form:
$$\mu(I) = O\left(\frac{1}{\ln(C/|I|)}\right)$$
where $|I|$ is the length of the interval. This result addresses the absence of atoms (pure point spectrum) and the continuity of the spectral measure, extending previous work that focused primarily on the absence of atoms for singletons.

Methodology
The core methodological innovation is a strengthened version of the monotone labelling method, originally introduced in a joint work with Sen and Virág [8]. The approach proceeds as follows:

1. Geometric Decomposition via Homomorphisms: The authors utilize the existence of a surjective homomorphism $\phi: \Gamma \to \mathbb{Z}$ (or more generally, an indicable structure) to decompose the graph vertices (or blocks of vertices) based on the values of $\phi(x)$.
2. Classification of Vertices: Given a labelling $\eta$ derived from $\phi$, vertices are classified into three categories:
   • Prodigy: Vertices with a unique neighbor of strictly lower label, where all other neighbors of that neighbor have even lower labels.
   • Level: Vertices where all neighbors have labels less than or equal to their own, but they are not prodigies.
   • Bad: Vertices that do not fit the above categories.
3. Iterative Eigenvalue Control: The proof analyzes the eigenvalue equation $(P - \lambda)f = 0$ iteratively along the decomposition. It demonstrates that the value of an eigenfunction $f$ at a "prodigy" vertex is determined by its values at lower-level vertices.
4. Von Neumann Algebra Framework: To extend this finite-dimensional intuition to infinite graphs, the authors embed the operators into a von Neumann algebra of finite type associated with the unimodular random graph. The existence of a faithful, normal tracial state $\tau$ (the expectation of the spectral measure) allows the use of von Neumann dimension to control the spectral mass.
5. Block Labelling: A significant technical extension involves "monotone block labelling," where the decomposition is applied to partitions of the vertex set (blocks) rather than individual vertices. This allows the method to handle operators with matrix-valued weights (quasi-transitive graphs) and more complex dependencies.

Key Contributions and Results

• Theorem 1 (Group Algebras): For an element $p = p^* \in \mathbb{C}[\Gamma]$ with support $S$, if $(\Gamma, S)$ is $(a, k)$-indicable (meaning there exists a homomorphism $\phi$ where $\phi(a) = k$ is strictly maximal among $\phi(S)$), the spectral measure $\mu_p$ satisfies the logarithmic regularity bound. This implies $\mu_p$ has no atoms. This refines previous results by allowing non-amenable groups and generalizing from singletons to intervals.
• Theorem 2 (Invariant Random Operators): The result extends to right-invariant random operators (e.g., Anderson tight-binding models and anisotropic percolation). If the operator norm is bounded and the weight associated with the "maximal" generator $a$ is bounded away from zero, the expected spectral measure satisfies the same logarithmic bound. Notably, the bound is uniform over the distribution of other weights, provided the critical weight condition holds.
• Theorem 3 (Quasi-Transitive Graphs): The framework is generalized to operators on $C^V \otimes \ell^2(\Gamma)$ (quasi-transitive graphs). By using block labelling and inductive arguments on the partition of the vertex set $V$, the authors control the regularity of the expected spectral measure, reducing the problem to sub-blocks where the operator is invertible or simpler.
• Theorem 6 (General Indicable Groups): For any indicable group and any finite symmetric generating set $S$, there exists a self-adjoint operator $p$ with support $S$ satisfying the logarithmic regularity condition. This is achieved by constructing $p$ such that one generator dominates the others in magnitude, ensuring the invertibility condition required by the monotone labelling method.

Significance and Claims
The paper claims to extend the classical Craig–Simon theorem (which establishes logarithmic regularity for Anderson models on $\mathbb{Z}^d$) to a much broader class of non-abelian and non-amenable settings, including:

• Operators in group algebras of indicable groups (e.g., free groups, Artin groups, braid groups, surface groups).
• Anisotropic percolation models where specific edge probabilities are set to 1.
• Quasi-transitive operators with random weights.

The authors emphasize that their results provide a uniform bound on the logarithmic regularity that does not depend on the specific distribution of the non-critical weights (e.g., in the Anderson model, the potential $V_x$ need not have a density or be independent, only that its distribution is compact).

The paper acknowledges that the logarithmic rate $1/\ln(1/|I|)$ is likely optimal for this class of assumptions, citing the Dyson singularity on $\mathbb{Z}$ and the work of Kotowski and Virág on the lamplighter group $\mathbb{Z}_2 \wr \mathbb{Z}$, where the spectral measure behaves as $C/(\ln \epsilon)^2$.

Limitations and Future Directions
The authors note that their method relies heavily on the existence of a homomorphism to $\mathbb{Z}$ (indicability). They identify a gap in establishing absolutely continuous spectrum (as opposed to just the absence of atoms) for these general settings. Furthermore, extending these tools to periodic operators on Riemannian manifolds is identified as a challenge due to the corresponding von Neumann algebras not being of finite type. The paper positions itself as a step toward understanding spectral connectedness and regularity in complex group-theoretic and geometric settings, building upon the foundational work of Lück, Atiyah, and others on $L^2$-invariants.