Alloy-Type Anderson-Bernoulli Model

• The alloy-type Anderson-Bernoulli model is defined as a random Schrödinger operator with Bernoulli-distributed disorder, applicable in both discrete and continuum settings.
• It exhibits unique spectral structures such as explicit two-band edge spectra and exactly computable integrated density of states under certain conditions.
• Researchers overcome challenges from the singular nature of Bernoulli disorder using methods like free-site multiscale analysis and quantitative unique continuation.

The alloy-type Anderson-Bernoulli model is a random Schrödinger operator in which the random field is generated by i.i.d. Bernoulli amplitudes coupled to a single-site profile. On the lattice, a basic discrete alloy-type form is

$$H_\omega=-\Delta+\lambda V_\omega,\qquad V_\omega(x)=\eta_x(\omega)=\sum_{i\in\mathbb Z^d}\omega_i\,u(x-i),$$

and the standard Anderson model is recovered when $u=\delta_0$. In the continuum, one works with alloy-type fields such as

$$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m)$$

or related bump-function variants. The subject includes local and non-local profiles, single-particle and multi-particle Hamiltonians, discrete and continuum settings, and several structured variants—periodic, hierarchical, matrix-valued, and long-range-hopping. What unifies these models is the Bernoulli single-site law: it is singular, so standard spectral averaging and density-based Wegner estimates usually fail, and the literature therefore develops alternative inputs such as transfer matrices, free-site multiscale analysis, quantitative unique continuation, or smoothing by non-local convolution (Tautenhahn et al., 2014, Ekanga, 2017, Ekanga, 2016).

1. Model classes and basic terminology

In the discrete alloy-type setting, the random potential is a convolution field built from i.i.d. couplings $(\omega_k)_{k\in\mathbb Z^d}$ and a summable single-site profile $u\in \ell^1(\mathbb Z^d)$. The potential values $V_\omega(x)$ are therefore generally correlated, and if $u$ changes sign then the model is non-monotone in the underlying couplings. These two features distinguish alloy-type models sharply from the standard i.i.d. site Anderson model and are already present before one specializes to Bernoulli disorder (Tautenhahn et al., 2014).

Continuum alloy-type models use the same architecture with translated bump functions. In one formulation,

$$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m),$$

while in another,

$$V(x;\omega)=\sum_{s\in\mathbb Z^d} V_s(\omega)\,\varphi_s(x-s).$$

Both are genuine alloy-type operators, with the randomness carried by i.i.d. amplitudes and the geometry carried by the single-site profile. Multi-particle versions replace the one-body field by the additive external potential

$$\mathbf V(\mathbf x,\omega)=\sum_{j=1}^N V(x_j,\omega),$$

and then add an interaction term $u=\delta_0$0 or $u=\delta_0$1 on configuration space (Ekanga, 2017, Monvel et al., 2010).

The Bernoulli specialization amounts to taking the common single-site law to be a two-point measure, for example $u=\delta_0$2 or $u=\delta_0$3. This includes the standard one-dimensional discrete model

$u=\delta_0$4

the matrix-valued quasi-one-dimensional continuum operator with Bernoulli channel amplitudes, and a variety of block, periodic, and long-range variants (Sánchez-Mendoza, 2021, Boumaza, 2010).

2. Ergodic spectral structure and almost-sure spectrum

For the standard one-dimensional discrete Anderson-Bernoulli operator with $u=\delta_0$5, the almost-sure spectrum is

$u=\delta_0$6

The regime $u=\delta_0$7 is the strong-disorder regime in which the two bands are separated, except possibly at one point when $u=\delta_0$8. This separated-band geometry underlies a large part of the explicit IDS theory for the one-dimensional Bernoulli model (Sánchez-Mendoza, 2020).

A structurally different spectral problem arises in the period-$u=\delta_0$9 “periodic Anderson-Bernoulli model,” where the even and odd sites use different Bernoulli laws: $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m)$$0 This is not the standard ergodic Anderson model over the one-step shift on $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m)$$1, but it becomes an ergodic family after enlarging the base dynamics to include the parity coordinate. Its natural transfer object is the set of four two-step matrices $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m)$$2, and the main theorem states that the almost sure spectrum consists of at most $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m)$$3 closed intervals. In the terminology of that paper, “periodic Anderson model” does not mean a deterministic periodic potential; it means a random Anderson-type model in which the distribution of the random potential is periodic in the site index (Wood, 2022).

Higher-dimensional local Bernoulli models retain the same ergodic almost-sure character but with different band geometry. In the two-dimensional large-disorder model with $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m)$$4,

$$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m)$$5

while the three-dimensional operator $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m)$$6 with $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m)$$7 has almost-sure spectrum $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m)$$8. In long-range-hopping models on $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m)$$9, after affine normalization one has

$(\omega_k)_{k\in\mathbb Z^d}$0

These formulas show that Bernoulli disorder often produces explicit two-band edge structure even when the kinetic term varies substantially (Li, 2020, Li et al., 2019, Liu et al., 3 Jul 2026).

3. Integrated density of states and spectral regularity

The integrated density of states is unusually explicit in several one-dimensional Bernoulli models. In the strong-disorder regime $(\omega_k)_{k\in\mathbb Z^d}$1, defining

$(\omega_k)_{k\in\mathbb Z^d}$2

one has the sharp bounds

$(\omega_k)_{k\in\mathbb Z^d}$3

together with the reflected upper-edge bounds near $(\omega_k)_{k\in\mathbb Z^d}$4. At the special energies $(\omega_k)_{k\in\mathbb Z^d}$5, $(\omega_k)_{k\in\mathbb Z^d}$6, the lower and upper bounds coincide, so the IDS is exactly computable and does not depend on $(\omega_k)_{k\in\mathbb Z^d}$7 (Sánchez-Mendoza, 2020).

This exactness extends from the integer sequence to a countable dense set of “rational energies.” If

$(\omega_k)_{k\in\mathbb Z^d}$8

then for each $(\omega_k)_{k\in\mathbb Z^d}$9 there is a critical $u\in \ell^1(\mathbb Z^d)$0 such that

$u\in \ell^1(\mathbb Z^d)$1

and

$u\in \ell^1(\mathbb Z^d)$2

Thus exact, disorder-independent IDS values occur on a dense set, but only above energy-dependent thresholds (Sánchez-Mendoza, 2021).

At small disorder, Bourgain proved a different kind of regularity for the one-dimensional Bernoulli model: under arithmetic assumptions on the coupling $u\in \ell^1(\mathbb Z^d)$3, the IDS $u\in \ell^1(\mathbb Z^d)$4 is $u\in \ell^1(\mathbb Z^d)$5-smooth on $u\in \ell^1(\mathbb Z^d)$6, with the proof proceeding through group expansion, the Furstenberg stationary measure, and smoothing of the Lyapunov exponent via Thouless’ formula. This is a noncommutative regularity mechanism rather than a density-based one (Bourgain, 2013).

Regularity can also appear because the alloy profile itself smooths the disorder. In the non-local two-particle lattice model with

$u\in \ell^1(\mathbb Z^d)$7

where $u\in \ell^1(\mathbb Z^d)$8 has staircase power-law decay, the cumulative single-site potential has characteristic function

$u\in \ell^1(\mathbb Z^d)$9

and hence a $V_\omega(x)$0 density. The same smoothing produces a finite-volume decomposition

$V_\omega(x)$1

with $V_\omega(x)$2 independent of the conditioned remainder and again having a $V_\omega(x)$3 density (Chulaevsky, 2017).

By contrast, discrete alloy-type models need not satisfy the conditional regularity hypotheses often used in abstract localization theory. In dimension $V_\omega(x)$4, the review of Tautenhahn and Veselić shows examples with

$V_\omega(x)$5

so the field is not uniformly $V_\omega(x)$6-Hölder continuous in the conditional sense. This obstruction is present already for regular single-site laws and therefore a fortiori for Bernoulli couplings (Tautenhahn et al., 2014).

4. Localization mechanisms and Wegner theory

The central technical difficulty in Anderson-Bernoulli problems is that Bernoulli disorder is singular. Standard Wegner proofs based on spectral averaging require absolute continuity of the single-site law and therefore fail in the Bernoulli case; this point is explicit in both the lattice and continuum multi-particle literature (Ekanga, 2016, Ekanga, 2017).

One important response is to import unusually strong one-dimensional single-particle inputs and lift them to interacting many-body systems. In the one-dimensional lattice $V_\omega(x)$7-particle model

$V_\omega(x)$8

the paper on “Wegner bounds for N-body interacting Bernoulli-Anderson models in one dimension” proves one-volume and two-volume Wegner bounds in the weak interaction regime and then invokes earlier multi-particle multiscale analysis to obtain spectral and strong dynamical localization. The key single-particle ingredient is the Carmona–Klein–Martinelli Bernoulli/singular Wegner estimate (Ekanga, 2016). The continuum analogue proves the same strategy for

$V_\omega(x)$9

using the one-dimensional continuum Bernoulli estimate of Damanik–Sims–Stolz and the second resolvent identity to pass from $u$0 to $u$1 (Ekanga, 2017).

Another response is free-site multiscale analysis plus quantitative unique continuation. In the two-dimensional large-disorder lattice model with $u$2, localization is proved outside small neighborhoods of finitely many exceptional energies. Those exceptional energies are Dirichlet eigenvalues of $u$3 restricted to connected finite subsets of a fixed finite box, and the method combines percolation geometry, a cutting procedure, a generalized Sperner lemma, and a discrete unique continuation theorem (Li, 2020). In three dimensions, localization near the lower spectral edge is proved for the Bernoulli model $u$4 by adapting the Bourgain–Kenig and Ding–Smart framework and supplying a new discrete unique continuation principle on $u$5 strong enough for the Wegner step (Li et al., 2019).

Long-range hopping requires a further modification because transfer matrices are unavailable and unique continuation becomes symbol-dependent. For $u$6 on $u$7, with $u$8 a long-range convolution operator whose Laurent symbol is rational, the paper of 2026 proves localization near the spectral edge and states that this is the first localization result for the long-range Anderson model with pure Bernoulli potentials. Its quantitative unique continuation principle comes from a finite-range recurrence obtained after rational-symbol renormalization, and the proof then uses a Bourgain–Kenig free-site multiscale scheme (Liu et al., 3 Jul 2026).

5. Structured and generalized variants

Several papers study Anderson-Bernoulli models in which the Bernoulli law is combined with additional deterministic structure. The period-$u$9 alternating model is the clearest example: randomness remains independent from site to site, but the even and odd coordinates use different Bernoulli distributions. Its block-transfer reduction to $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m),$$0 makes the spectral-support problem explicit and leads to the “at most four intervals” theorem (Wood, 2022).

A different structured variant is the quasi-one-dimensional matrix-valued continuum operator

$$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m),$$1

with $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m),$$2 a generic real symmetric interaction matrix. For almost every $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m),$$3, there is a finite critical set $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m),$$4 such that, on the explicit interval $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m),$$5, the model exhibits exponential localization and strong dynamical localization. The proof identifies the Fürstenberg group with $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m),$$6 away from finitely many energies by algebraic-geometric control of Lie-algebra generation (Boumaza, 2010).

Non-local alloy structure can itself regularize Bernoulli disorder. In the two-particle lattice model with staircase power-law profile $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m),$$7, the infinite-range single-site profile turns Bernoulli amplitudes into smooth cumulative potentials and yields frozen-bath Wegner estimates, eigenvalue comparison bounds, and low-energy strong dynamical localization (Chulaevsky, 2017). This is not the classical local Bernoulli model; it is a generalized alloy-type model in which non-locality is the essential smoothing mechanism.

Two long-range kinetic generalizations are also now available. One is the one-dimensional edge-localization theorem for pure Bernoulli disorder and rational-symbol hopping on $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m),$$8 (Liu et al., 3 Jul 2026). The other is the discrete alloy-type Bernoulli model on $$V(x,\omega)=\sum_{m\in\mathbb Z^d} q_m(\omega)\,f(x-m),$$9 with exponentially decaying Toeplitz hopping,

$$V(x;\omega)=\sum_{s\in\mathbb Z^d} V_s(\omega)\,\varphi_s(x-s).$$0

for which localization is proved near the upper spectral edge $$V(x;\omega)=\sum_{s\in\mathbb Z^d} V_s(\omega)\,\varphi_s(x-s).$$1. That work extends Bourgain’s 2004 Bernoulli analysis from the nearest-neighbor Laplacian to exponentially decaying long-range hopping by combining periodic approximants, Floquet–Bloch theory, a quantitative uncertainty principle, free sites, and Bourgain’s distributional inequality (Liu et al., 18 Aug 2025).

Finally, the hierarchical Anderson-Bernoulli model on $$V(x;\omega)=\sum_{s\in\mathbb Z^d} V_s(\omega)\,\varphi_s(x-s).$$2 adds a deterministic multi-scale background $$V(x;\omega)=\sum_{s\in\mathbb Z^d} V_s(\omega)\,\varphi_s(x-s).$$3 to the Bernoulli field. In this setting, Anderson localization is proved in $$V(x;\omega)=\sum_{s\in\mathbb Z^d} V_s(\omega)\,\varphi_s(x-s).$$4 for $$V(x;\omega)=\sum_{s\in\mathbb Z^d} V_s(\omega)\,\varphi_s(x-s).$$5, and also for $$V(x;\omega)=\sum_{s\in\mathbb Z^d} V_s(\omega)\,\varphi_s(x-s).$$6 under the conditions $$V(x;\omega)=\sum_{s\in\mathbb Z^d} V_s(\omega)\,\varphi_s(x-s).$$7 and $$V(x;\omega)=\sum_{s\in\mathbb Z^d} V_s(\omega)\,\varphi_s(x-s).$$8. This is a hierarchical analogue rather than the standard non-hierarchical Bernoulli model, but it shows that weak transversality plus Schur-complement renormalization and a site-mixed martingale can replace stronger unique continuation in high dimension (Liu et al., 21 Apr 2026).

6. Scope, limitations, and open directions

A recurring limitation is that many major alloy-type localization theorems are not Bernoulli theorems. The continuum multi-particle papers of Aizenman–Warzel’s general area represented here by Chulaevsky–Suhov-type results prove localization and dynamical localization for alloy-type random potentials, but they assume Hölder or log-Hölder continuity of the single-site distribution and therefore exclude Bernoulli randomness. The same is true for the weak-disorder, sign-indefinite lattice alloy model on $$V(x;\omega)=\sum_{s\in\mathbb Z^d} V_s(\omega)\,\varphi_s(x-s).$$9, whose Wegner estimate explicitly requires a bounded Hölder-continuous density and is stated not to work for distributions concentrated on finitely many points (Monvel et al., 2010, Chulaevsky et al., 2010, Cao et al., 2011).

Even within direct Bernoulli results, several problems remain model-specific. The period-$$\mathbf V(\mathbf x,\omega)=\sum_{j=1}^N V(x_j,\omega),$$0 alternating theory is explicit because there are only four block matrices; the corresponding paper states that the higher-period problem is open and even conjectures that for large enough period one may fail to obtain a finite union of intervals (Wood, 2022). The two-dimensional large-disorder lattice theorem excludes finitely many windows $$\mathbf V(\mathbf x,\omega)=\sum_{j=1}^N V(x_j,\omega),$$1, and the author remarks that localization inside those windows is plausible but not proved by the present method (Li, 2020). For long-range hopping on $$\mathbf V(\mathbf x,\omega)=\sum_{j=1}^N V(x_j,\omega),$$2, deterministic quantitative unique continuation fails in general; the rational-symbol class is singled out precisely because counterexamples exist for more general exponentially decaying hoppings, and the weaker super-exponential result is not strong enough for the Bernoulli localization proof carried out there (Liu et al., 3 Jul 2026).

The high-dimensional local problem remains especially delicate. The hierarchical paper states that the standard non-hierarchical discrete Anderson-Bernoulli model on $$\mathbf V(\mathbf x,\omega)=\sum_{j=1}^N V(x_j,\omega),$$3, $$\mathbf V(\mathbf x,\omega)=\sum_{j=1}^N V(x_j,\omega),$$4, remains open, even though the hierarchical analogue can now be treated (Liu et al., 21 Apr 2026). A plausible implication is that future progress will continue to depend on replacements for density-based Wegner estimates: sharper probabilistic unique continuation on $$\mathbf V(\mathbf x,\omega)=\sum_{j=1}^N V(x_j,\omega),$$5, new free-site combinatorics, or additional geometric input that plays the role of the deterministic barrier structure in hierarchical models.