Anderson Localization for the hierarchical Anderson-Bernoulli model on $\mathbb{Z}^d$

Abstract: In this paper, we prove Anderson localization for a hierarchical Anderson-Bernoulli model on lattice with arbitrary dimension, where the potential is characterized by a geometric hierarchical structure combined with fluctuations induced by independent and identically distributed (i.i.d.) Bernoulli random variables. Our method is also applicable to proving a probabilistic unique continuation result on $\mathbb{Z}^d$.

• The paper establishes Anderson localization for a hierarchical Anderson-Bernoulli model on Z^d, marking the first proof for singular Bernoulli potentials especially in high dimensions.
• It employs a refined multi-scale analysis combined with a novel site-mixed martingale method to overcome limitations of classical Wegner estimates and unique continuation.
• The results demonstrate exponential eigenfunction decay and uniform dynamical localization, with rigorous Wegner-type bounds ensuring strict spectral separation.

Anderson Localization in the Hierarchical Anderson-Bernoulli Model on $\mathbb{Z}^d$

Introduction and Context

This paper provides the first proof of Anderson localization for a hierarchical Anderson-Bernoulli model (ABM) on $\mathbb{Z}^d$ with Bernoulli-distributed i.i.d. random potentials for arbitrary $d \geq 1$. The study of localization phenomena for the random Schrödinger operator with Bernoulli potentials has been a longstanding problem, particularly in high dimensions $(d\geq 4)$ due to the failure of classical tools such as the Wegner estimate and unique continuation on discrete lattices.

The hierarchical model considered merges deterministic geometric barriers (with Newtonian scaling) and i.i.d. Bernoulli randomness. This structure creates a sharply contrasting scenario to the standard Anderson model, where the random potential is assumed to be absolutely continuous. The authors construct a sophisticated multi-scale analysis, circumventing the lack of absolute continuity and unique continuation by employing a weak (1D) transversality condition—combined with a novel site-mixed martingale argument—to establish exponential localization of eigenfunctions.

Main Results

Low Dimensions $(1\leq d \leq 3)$: For any coupling constant $0<\beta\leq 1$, Anderson localization for $H(\omega)$ is established at the bottom of the spectrum, i.e., for energies $E\in [0,h)$. This covers the expected regime of dynamical and spectral localization, even when the random potential is strictly Bernoulli.

High Dimensions $(d\geq 4)$: Given sufficiently large barrier height $h$ and barrier width parameter $\mathbb{Z}^d$0 (where $\mathbb{Z}^d$1 is the well density), Anderson localization for arbitrary $\mathbb{Z}^d$2 holds at the bottom of the spectrum. This is the first such result for singular Bernoulli potentials in $\mathbb{Z}^d$3, in either hierarchical or standard settings.

Dynamical Localization: In all dimensions and under respective conditions, the paper also proves uniform boundedness of the expected mean square displacement:

$\mathbb{Z}^d$4

almost surely in $\mathbb{Z}^d$5.

Technical Approach

Model Construction

The hierarchical potential consists of deterministic high-value barriers arranged on length scales $\mathbb{Z}^d$6 ($\mathbb{Z}^d$7), separating multiple potential wells at each scale (parameterized by $\mathbb{Z}^d$8). The random perturbation is added as $\mathbb{Z}^d$9, where each site is assigned $d \geq 1$0 or $d \geq 1$1 independently (Bernoulli variable).

Overcoming the Absence of Standard Wegner Estimates

The classical multi-scale analysis and Wegner estimate rely on absolute continuity, which fails for Bernoulli potentials. Instead, the authors exploit a cone property—a weak, one-dimensional form of transversality—along with a robust martingale framework. For $d \geq 1$2, tools like unique continuation (probabilistic and deterministic) and Sperner-type combinatorics are sufficient, while for $d \geq 1$3, only the cone property applies, necessitating a new probabilistic and structural approach.

Martingale Framework and Strict Spectral Monotonicity

A critical innovation is the site-mixed martingale construction in high dimensions. By iteratively controlling fluctuations at judiciously chosen sites—where a weak but nontrivial transversality is established along geometric chains ("cones")—the authors show that the local eigenvalue counting function forms a submartingale, which, by Azuma-Hoeffding's inequality, yields exponentially strong Wegner-type bounds for the hierarchical ABM.

This is accomplished via a refined multi-scale iteration: between Newtonian scales, the authors interpolate finer $d \geq 1$4-adic scales enabling precise analysis of eigenvalue transitions under perturbations at specific sites, ultimately showing that the number of eigenvalues near a fixed energy strictly decreases with high probability at each step, leading to spectral gaps scaling as fast as $d \geq 1$5.

Interpolation Between Scales and Schur/Orthogonality Techniques

In addition, the analysis makes clever use of Schur complements to relate eigenvalues at different scales and leverages approximate orthogonality of eigenfunctions to yet more tightly control multiplicities across scales.

Removal of $d \geq 1$6-Dependence

A technical challenge is ensuring the lower bound on the barrier height $d \geq 1$7 does not degenerate as $d \geq 1$8. The paper resolves this by expanding higher-order terms in the random walk-based Neumann series expansions, showing that sufficiently large but $d \geq 1$9-independent $(d\geq 4)$0 suffices for all $(d\geq 4)$1.

Generalized Localization and Spectral Separation

The authors' approach not only establishes localization but delivers full spectral separation for the hierarchical model: for large enough $(d\geq 4)$2, the spectra of different resonant blocks are separated by at least $(d\geq 4)$3 with high probability, even for Bernoulli-distributed randomness.

Notable Claims

• First localization result for i.i.d. Bernoulli potentials in $(d\geq 4)$4: This effectively closes a long-standing gap in the high-dimensional discrete Anderson-Bernoulli localization problem.
• Exponential decay rate for eigenfunctions: The method yields concrete, explicit bounds on the decay of eigenfunctions in terms of the geometrical parameters of the model.
• Strict monotonicity in eigenvalue counting: Through the interpolation and martingale schemes, the paper proves strong monotonicity properties on the dimension of spectral slices, even in regimes where unique continuation fails.
• Wegner bounds with polynomially decaying probability in scale: This is achieved despite the singularity of the Bernoulli potential.

Implications and Future Directions

Theoretical Implications: The techniques introduced have significant implications for the theory of random Schrödinger operators on lattices with singular potentials. The robust use of cone-based transversality and martingale arguments may point the way towards a full resolution of the Anderson localization conjecture in the standard (non-hierarchical) ABM for $(d\geq 4)$5.

Practical Implications: Instability of quantum tunneling under singular noise, as quantified here, carries direct relevance to transport properties in disordered quantum systems with discrete (non-continuous) random environments.

Extensions and Open Problems:

• Removal of $(d\geq 4)$6-dependence for the standard ABM: Can these methods be adapted to yield localization for the non-hierarchical lattice ABM, under weaker geometric constraints?
• Full-dimensional probabilistic unique continuation: The authors pose as an open problem whether the probabilistic unique continuation estimates derived (which currently depend on initial data) can be made uniform, enabling further advances in the theory.
• Applications to continuous models: While the orientation here is primarily discrete, the technical tools may inspire analogous arguments in the continuum setting.

Conclusion

This work achieves a major milestone in the mathematical understanding of Anderson localization in random media with singular disorder. By successfully adapting and extending multi-scale analysis beyond the limits of classical tools, the authors both solve the localization problem for the hierarchical ABM in all dimensions and introduce methods—particularly the site-mixed martingale and a fine-grained scale interpolation—that will likely have further impact on the theory of localization for models with minimal (Bernoulli) randomness.

Reference:

"Anderson Localization for the hierarchical Anderson-Bernoulli model on $(d\geq 4)$7" (2604.18989)