When disorder traps a quantum particle—and when extended states survive

← All digests

Family 261 · Mathematical physics
Importance 5/5 · Provisional editorial rating
Sources consulted 2026-10-25

1. The question

Imagine a particle moving through an infinite crystal. In an ideal lattice, the local environment repeats, and wave interference can support states spread across the material. Now independently change the energy at every lattice site. Does arbitrarily weak disorder trap the particle, or can extended states survive?

The Anderson model turns this question into a random operator on an integer lattice. Family 261 contains two complementary claims for independent uniformly distributed site energies. One concerns two dimensions at every fixed positive disorder strength. The other concerns every fixed dimension at least three, at sufficiently weak positive disorder.

Their conclusions concern spectral type: the kind of energy decomposition supplied by the spectral theorem. This is closely related to transport, but it is not interchangeable with a quantitative statement that a wave packet diffuses at a specified rate. Keeping that distinction visible is essential to understanding both manuscripts.

2. Background and essential definitions

Let $\mathbb Z^d$ be the integer lattice in $d$ dimensions. A wave function $\psi$ assigns a complex amplitude to each site, with

$$\sum_{x\in\mathbb Z^d}|\psi(x)|^2<\infty.$$

These functions form the Hilbert space $\ell^2(\mathbb Z^d)$. The Hamiltonian combines hopping to nearest neighbors with a random diagonal potential:

$$(H_\lambda\psi)(x) =-\sum_{j=1}^d\bigl(\psi(x+e_j)+\psi(x-e_j)\bigr) +\lambda V_x\psi(x).$$

Here $e_j$ is the $j$-th coordinate vector, the variables $V_x$ are independent uniforms on $[-1,1]$, and $\lambda>0$ measures disorder. Changing the hopping sign gives a unitarily equivalent model on this lattice, so the different sign convention in the two-dimensional paper does not change its spectral claim.

The spectrum consists of energies at which the operator cannot be inverted continuously. In infinite dimension it is not just a list of eigenvalues. For a state $\psi$, the spectral theorem gives a measure $\mu_\psi$: its mass on an energy set describes how much of $\psi$ lies in that spectral range.

There are three basic types. A pure-point measure is concentrated on countably many atoms, corresponding to eigenvectors. An absolutely continuous measure has a density with respect to ordinary length on the energy axis. A singular continuous measure has no atoms but is concentrated on a set of length zero. Thus excluding eigenvectors alone does not establish absolute continuity.

Pure-point spectral type for the entire operator means that its eigenvectors form a complete orthonormal basis. It does not mean that every point in its spectrum is an eigenvalue. A countable dense collection of eigenvalues can have an interval as its closure.

By contrast, purely absolutely continuous spectrum on an interval means every state restricted to that energy interval has an absolutely continuous spectral measure. This is stronger than merely finding a nonzero absolutely continuous component there.

3. Development history and previous partial results

Anderson’s 1958 paper introduced random site energies as a mechanism suppressing transport. It provided the physical origin of the problem, rather than a modern proof of the particular almost-sure spectral assertions discussed here. The 1979 scaling theory of Abrahams, Anderson, Licciardello and Ramakrishnan sharpened the expected distinction between two and higher dimensions; M2 cites that work as physical motivation.

Rigorous localization developed through several methods. Fröhlich and Spencer’s 1983 multiscale analysis controlled Green functions and diffusion in strong-disorder or suitable extreme-energy regimes. The subsequent work of Fröhlich, Martinelli, Scoppola and Spencer established pure-point spectrum with exponentially decaying eigenfunctions in such regimes. The distinction matters: information at almost every fixed energy is not automatically a complete decomposition of a random spectral measure.

Aizenman and Molchanov’s fractional-moment method provided another powerful route to localization. Here one estimates an average of a fractional power of a resolvent entry, which is less sensitive to rare large values. These methods settled substantial regions of parameters without giving all-energy localization at arbitrarily weak disorder in two dimensions. M2 also distinguishes later spectral-edge localization for Bernoulli and broader bounded distributions from its own whole-spectrum uniform-law claim.

Delocalization was established earlier on trees, including Klein’s absolutely continuous spectrum results for the Bethe lattice. A tree has exponentially branching geometry; it is not a Euclidean lattice. Such theorems therefore do not automatically settle the problem on $\mathbb Z^3$.

Barry Simon’s Schrödinger Operators in the Twenty-First Century separated the higher-dimensional absolutely continuous energy-range question, the planar pure-point question, and diffusion. Both manuscripts use that separation. More recently, Black, Drogin and Hernández obtained quantitative weak-disorder diffusion and resolvent results for Gaussian potentials on disorder-dependent scales. Those finite-scale or weak-coupling statements have a different order of limits from a spectral-type theorem at fixed positive coupling.

4. What the new manuscripts claim

Two dimensions: a complete eigenbasis

M2 claims that for every fixed $\lambda>0$, the square-lattice operator almost surely has pure-point spectral type throughout its spectrum. In the centered uniform convention,

$$\sigma(H_\lambda)=[-4-\lambda,\,4+\lambda].$$

The eigenvalues are dense in this interval, and the eigenvectors form a complete orthonormal basis. An arbitrary nondegenerate uniform interval is covered after subtracting its midpoint as a constant energy shift.

The probability-one event is allowed to depend on $\lambda$. The theorem does not assert exponential decay of every eigenfunction, dynamical localization, or an estimate uniform in energy. Its asserted advance is the pure-point conclusion for every positive disorder, including arbitrarily small disorder, over the entire spectrum for this single-site law.

At least three dimensions: a nonempty continuous range

M1 claims that for each fixed $d\geq3$, there is a positive threshold $\lambda_d$. For every fixed $0<\lambda<\lambda_d$, the operator almost surely has purely absolutely continuous spectrum with nonzero weight on a specified open interval $I_d$. The interval does not shrink with $\lambda$.

For $d\geq4$, the paper chooses

$$I_d=-2d+\left(\frac1{200},\frac1{100}\right).$$

For $d=3$, it chooses

$$I_3=(-6+\varepsilon/2,-6+\varepsilon),$$

where $\varepsilon$ is a fixed sufficiently small positive constant. These intervals sit just above the lower edge of the free hopping band, not at the lower edge of the full disordered spectrum.

The conclusion excludes both point and singular continuous spectral mass on $I_d$. The thresholds may depend on dimension. The higher-dimensional threshold is specified through probability conditions for finite matrices, without an effective numerical bound; the three-dimensional threshold is existential.

M1 also combines its theorem with established edge localization. At the same fixed weak coupling, a separate interval near $-2d-\lambda$ has pure-point spectrum with exponentially decaying eigenfunctions. Thus the manuscript claims coexistence of a localized edge region and an absolutely continuous region. It does not locate a unique transition energy between them or classify everything in between.

5. Why the advance matters

Provisional importance: 5/5. The spectral-type questions addressed here are central problems in disordered quantum systems. The two claims would establish a particularly clear dimension-dependent contrast for the same elementary random potential law.

The planar result reaches arbitrarily weak disorder, where finite samples can look very different from the infinite-volume spectral conclusion. The higher-dimensional result establishes purity, rather than just a small extended component. It also gives an interval independent of the sufficiently small fixed coupling.

Their significance lies in the order of quantifiers. Fixing a nonzero disorder and then reaching the infinite-volume, real-energy boundary is stronger in a different direction from following transport for longer and longer times while simultaneously taking disorder to zero. Neither kind of theorem makes the other unnecessary; they describe different aspects of the model.

6. The main idea

Both papers work with resolvents, finite-volume matrices and controlled ways of revealing random potentials. The resolvent is

$$G(z)=(H-z)^{-1},\qquad \operatorname{Im}z>0.$$

Its entries measure the response at one site to a source at another. Approaching the real energy axis reveals spectral information, but near-resonances can make individual entries large. The technical task is to retain useful control as the artificial imaginary part disappears.

For M1, the starting operator has damping at selected sites. Each uniform random variable is represented by independent binary digits. Revealing digits gradually replaces artificial regularization with the actual random potential. A Schur complement eliminates completed sites while retaining their influence on the remaining matrix.

Three controls travel through this procedure: a self-consistency estimate, bounds preventing a resolvent row from concentrating at a few sites, and a spatial energy form derived from the Ward identity. Failed local updates leave a record; they cannot simply be retried as though already observed random values were fresh. The manuscript organizes pending sites and their later completion to preserve the true product distribution. The final finite-volume certificate keeps a positive spectral response on many sites. Spectral averaging then upgrades this response to pure absolute continuity.

M2 uses absorbing channels, called ports, and scattering matrices. Their squared amplitudes define an energy measuring differences between input values at different ports. Rare resonant cells and a hierarchy of progressively sparser ports produce comparisons under the original uniform law.

Two-dimensional geometry enters through logarithmic voltage profiles. A large annulus can separate inner and outer values at arbitrarily small capacity cost. The paper uses this feature to force a limiting affine scattering energy to vanish. It then obtains the resolvent summability needed for a rank-one spectral argument. Averaging over one independent site value transfers fixed-energy information to the random spectral measure, yielding the complete pure-point conclusion.

These are explanations of the stated mechanisms, not physical replacements of the random potential by a different distribution. The auxiliary ports, damping and conditioning belong to the analysis; the final claims concern the original Anderson operator.

7. Scope, limitations, and open questions

The two primary theorems use independent uniform site potentials and nearest-neighbor hopping. They do not automatically extend to Gaussian, Bernoulli, correlated or continuum disorder, nor to interacting particles.

The planar theorem does not claim exponential localization or uniform dynamical bounds. The higher-dimensional theorem does not give a full mobility-edge diagram, a diffusion coefficient, or effective usable disorder thresholds. Its probability-one event may depend on the fixed coupling, just as in the planar result. There is no asserted uniformity as dimension tends to infinity.

Natural further questions concern quantitative localization lengths, transport at fixed disorder, the spectral region between the specified intervals, and robustness under other potential laws. The manuscripts’ stated conclusions should therefore be read as sharp answers to their spectral-type questions, with these additional transport and classification questions kept separate.

8. Sources and further reading

Anderson’s original article and Black–Drogin–Hernández are linked in section 3. Other primary historical sources discussed by the manuscripts include Fröhlich–Spencer, Absence of diffusion in the Anderson tight binding model for large disorder or low energy, and Fröhlich–Martinelli–Scoppola–Spencer, Constructive proof of localization in the Anderson tight binding model. The spectral readout is related to Simon and Wolff’s rank-one perturbation and localization theorem. M1 and M2 give the precise versions and parameter regimes used in their comparisons with earlier work.

Source manuscripts

• M1. OpenAI, Absolutely Continuous Spectrum for Weak-Disorder Anderson Models in Dimensions at Least Three (September 23, 2026).
• M2. OpenAI, Pure-Point Spectrum for the Two-Dimensional Anderson Model at Every Positive Disorder (September 23, 2026).

Sources consulted on 25 October 2026 at repository revision 1581d4c. Download the manuscript citations.