Anderson localisation in infinite and finite resonator crystals with long-range interactions

arXiv:2611.03286 · math-ph, math.MP · Submitted 2026-11-02 · Read on arXiv

Listen

Radio episode about this paper

Transcript

Introduction to the show: ident: Mathematical Physics. The day in mathematical physics: rigorous models of physical systems.

Kai: We're looking at the paper: "Anderson localisation in infinite and finite resonator crystals with long-range interactions".

Mira: Here's the gist of the paper. The paper establishes Anderson localization for random capacitance operators in periodic geometries of dimensions d=2, 3, with one resonator per cell, where inverse material factors are independent and uniformly distributed.

Kai: The analysis considers the full capacitance interaction, whose coefficients decay exponentially with lattice separation.

Mira: Part I focuses on the infinite operator, its symbol, and localization properties.

Kai: Under coordinate reflection symmetry, spectral and dynamical localization hold near the upper spectral edge for every fixed positive disorder strength W.

Title and authors: Kai: The paper "Anderson localisation in infinite and finite resonator crystals with long-range interactions" by Ammari, Fu, Jing, Qiu, and Uhlmann is setting up Anderson localization for random capacitance operators in these periodic systems. It focuses on how the long-range nature of the capacitance affects that localization.

Mira: What's interesting about this is that they aren't just looking at infinite systems; they are also quantifying the concentration of individual modes for finite capacitance matrices, which is crucial when we move to real hardware.

Lev: When you talk about finite capacitance matrices, you’re talking about truncated operators, and understanding how those behave relative to the infinite ones is where the real computational overhead lies for error correction.

Kai: They establish that these two matrix sequences share the same limiting density of states, which suggests that even if we truncate our system, the big picture of how many states there are remains consistent with the infinite case.

Mira: And they do surface-order trace comparisons, which is a way to check if the finite clusters truly converge to something meaningful when you take their size to infinity.

Lev: For running on hardware, that convergence implies that we can use estimates from truncated matrices to predict how modes behave in the full system, which simplifies the complexity of checking boundary conditions.

Kai: So, they are basically saying that even though we have finite clusters, the spectral convergence is strong enough to transfer localization properties from those truncated matrices directly onto individual modes of the finite capacitance matrix.

The paper's summary: Mira: To summarize what this paper does, it proves Anderson localization for random capacitance operators in two and three dimensions with one resonator per cell, where the inverse material factors are independently distributed. They keep the full capacitance interaction whose coefficients decay exponentially with lattice separation.

Kai: The main result is that under coordinate reflection symmetry, spectral and dynamical localization hold near the upper spectral edge for every fixed positive disorder strength W. This means you get localization near the top of your spectrum regardless of how much randomness you put in, as long as it's not too strong.

Lev: Localization near the upper edge is good because it keeps the spectrum somewhat predictable, but I wonder how strong this localization actually is when we consider real-world noise levels.

Mira: They also get an explicit localization window of energy width proportional to W at weak disorder, and they give you a specific interval for small disorder called IW a which is written as one times M over one plus alpha W times M over one plus alpha W where alpha is some constant.

Kai: That explicit window size proportional to W gives us a concrete energy range to check in simulations or experiments, which is much better than just saying localization happens somewhere in the vicinity of the edge.

Lev: But what about when we look at finite clusters? The summary mentions they establish estimates for truncated matrices and eigenvalue spacing, which deals with how localized the individual modes are within those finite samples.

Mira: They show that for any fixed W greater than zero, there exist constants A W zero such that with probability at least one minus C W every normalized eigenvector psi j corresponding to an eigenvalue E j in IW has a center n j in lambda for which its value at n is less than or equal to N p plus five over two e minus nu W alpha n minus n j over two where nu is some constant.

Kai: That estimate, with the probability one minus C W, gives us a way to control how localized the wavefunction itself is in the finite sample, tying it directly back to the disorder strength W.

The paper's improvements: Lev: The paper suggests a few key improvements. First, they use that variational boundary comparison and material-variable eigenvalue-spacing estimate to transfer localization from truncated matrices to individual modes of the finite capacitance matrix.

Mira: That transfer is important because it allows them to quantify concentration estimates for finite clusters, specifically showing that in a bounded sample of cell size epsilon, all except for O epsilon minus d log(one over epsilon) modes in the localization interval concentrate within radii O epsilon log(one over epsilon) with high probability.

Kai: That epsilon dependence is interesting; it tells us how much the localization region shrinks as we look at smaller and smaller clusters, and that scaling is logarithmic, which is relatively slow decay.

Lev: That logarithmic scaling of the concentration radius suggests that for practical hardware realization, you might be able to get a decent spatial localization with reasonably small cluster sizes.

Mira: They also derive a material-variable eigenvalue-spacing estimate, which they use in conjunction with a variational boundary comparison to transfer this localization from truncated matrices to individual modes of the finite capacitance matrix.

Kai: And this is connected to their correlator estimate, which is written as E Q t pi lambda (x, y; IW) less than or equal to A W e minus nu W alpha x minus y where nu is some constant.

Lev: That final result, with that correlator bound proportional to e to the negative of W, is what really matters for experimentalists because it shows exponential decay in space between points x and y within the localization interval.

Conclusion: Kai: So to wrap up, this paper "Anderson localisation in infinite and finite resonator crystals with long-range interactions" establishes Anderson localization for random capacitance operators, showing spectral and dynamical localization near the upper edge under fixed positive disorder strength.

Mira: And they give us explicit windows for localization, like IW a which is written as one times M over one plus alpha W times M over one plus alpha W where alpha is some constant, which helps define exactly where we expect to find those localized states.

Lev: What this means for the hardware side is that if you have long-range interactions and some disorder, you can predict a specific energy region where modes will be localized with high probability.

Kai: And they prove that the absence of absolutely continuous spectrum follows from part one, and the common density of states is shown to be zero to one times M over zero to one times M almost surely by proposition two point two.

Mira: So the final result for Anderson localization with long-range interactions is given by theorem six point one which states that the correlator bound E Q t pi lambda (x, y; IW) less than or equal to A W e minus nu W alpha x minus y where IW is written as one times M over c minus delta W to one times M.

Lev: The paper also shows that almost surely the spectrum of H pi in IW is pure point, and the coordinate vectors are total, meaning the spectrum in IW is pure point with simple eigenvalues.

Kai: That’s a strong conclusion because it means that for these specific energy intervals, you have a discrete set of states instead of a continuous band.

Mira: It also gives us that the frequency distribution is independent of epsilon, which proves that the absence of absolutely continuous spectrum follows from part one.

Habib Ammari, Xin Fu, Wenjia Jing, Jiayu Qiu, Alexander Uhlmann

ETH Zurich · Hong Kong Institute for Advanced Study

math-ph, math.MP

Submitted: 2026-11-02

Updated: 2026-11-02

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 100/100

The gist: The paper establishes Anderson localization for random capacitance operators in periodic geometries of dimensions d=2, 3, with one resonator per cell…

Key concepts

Anderson Localization
This phenomenon describes how quantum particles become trapped in a disordered material, preventing them from propagating freely. In this context, it means that the energy states (eigenvalues) of the capacitance operator are localized in space, meaning their corresponding wavefunctions decay exponentially away from certain points.
Long-Range Interactions
The capacitance between resonators decays exponentially with distance. This long-range nature is crucial because it allows for a unified analysis across different system sizes and helps establish spectral properties that are independent of the cluster size, leading to universal localization results.
Localization Window (IW)
This is a specific energy range where all or most eigenstates are localized. The paper derives explicit windows, such as IW_a = (Ea(W), cM), whose width is directly proportional to the disorder strength W. This window defines the region of energy where localization is guaranteed.
Correlator Bound
This mathematical estimate quantifies how quickly the system's response at two different spatial points ($x$ and $y$) decays. The result, EQ t_varpi,Lambda(x, y; IW) <= A_W e^{- u W |x-y|}, shows that this correlation decays exponentially with distance $|x-y|$, a hallmark of localization.

Terminology

Summary

The paper establishes Anderson localization for random capacitance operators in periodic geometries of dimensions d=2, 3, with one resonator per cell, where inverse material factors are independent and uniformly distributed. p. 0

Where the paper says it

We establish Anderson localisation for random capacitance operators and quantify concentration of individual modes of finite capacitance matrices for clusters of subwavelength high-contrast resonators. The underlying geometry is fully periodic in dimensions d = 2, 3, with one resonator per cell, and the inverse material factors are independent and uniformly distributed.

The analysis considers the full capacitance interaction, whose coefficients decay exponentially with lattice separation. p. 0
Where the paper says it

The underlying geometry is fully periodic in dimensions d = 2, 3, with one resonator per cell, and the inverse material factors are independent and uniformly distributed. We retain the full capacitance interaction, whose coefficients decay exponentially with lattice separation.

Part I focuses on the infinite operator, its symbol, and localization properties. p. 3

Where the paper says it

The paper is organised into three parts. Part I studies the infinite operator, its symbol, and localisation.

Under coordinate reflection symmetry, spectral and dynamical localization hold near the upper spectral edge for every fixed positive disorder strength W. p. 0
Where the paper says it

We retain the full capacitance interaction, whose coefficients decay exponentially with lattice separation. Under coordinate reflection symmetry, spectral and dynamical localisation hold near the upper spectral edge for every fixed positive disorder strength.

For weak disorder W, an explicit localization window of energy width proportional to W is obtained: IW = (cM - delta W, cM), where delta W proportional to W. p. 0
Where the paper says it

Under coordinate reflection symmetry, spectral and dynamical localisation hold near the upper spectral edge for every fixed positive disorder strength. At weak disorder W, we obtain an explicit localisation window of energy width proportional to W; prescribed positive intervals are localised at sufficiently large disorder.

Furthermore, for sufficiently small disorder W, a specific interval IW a = cM / (1 + aW), cM is explicitly given. p. 3
Where the paper says it

For each fixed 0 < a < 1/2 and sufficiently small W, Theorem 7. 1 gives the explicit interval IW,a = cM 1 + aW , cM .

Part II addresses the finite-cluster problem by establishing estimates for truncated matrices and eigenvalue spacing. p. 3

Where the paper says it

Part I studies the infinite operator, its symbol, and localisation. Part II establishes the estimates for truncated matrices and eigenvalue spacing needed for the finite-cluster problem.

The two matrix sequences share the same limiting density of states, and surface-order trace comparisons are performed. p. 3
Where the paper says it

The asymptotic-equivalence method of [7] extends to this diagonal weighting. The two matrix sequences have the same limiting density of states; we also obtain a surface-order trace comparison and convergence of their spectra.

For any fixed W > 0, there exist constants A W 0 such that with probability at least 1 - CW, every normalized eigenvector psi j corresponding to an eigenvalue E j in IW has a center n j in lambda for which psi j(n) at most N p + 5/2 e-nu W n-n j/2. p. 3
Where the paper says it

More precisely, let Ω ⊂ R d be a bounded Lipschitz sample containing Nε ≍ ε −d cells. For every fixed p > 0, with probability at least 1 − CW,pN −p ε , all but OW,p(ε 1−d log(1/ε)) modes in the localised interval concentrate in balls of radius OW,p(ε log(1/ε)), with squared tail at most N −p ε ; see Theorem 15.

This result, combined with a variational boundary comparison and a material-variable eigenvalue-spacing estimate, transfers localization from truncated matrices to individual modes of the finite capacitance matrix. p. 0
Where the paper says it

At weak disorder W, we obtain an explicit localisation window of energy width proportional to W; prescribed positive intervals are localised at sufficiently large disorder. A variational boundary comparison and a material-variable eigenvalue-spacing estimate transfer localisation from truncated matrices to individual modes of the finite capacitance matrix.

In a bounded sample of cell size epsilon, all except for O(epsilon 1-d (1/epsilon)) modes in the localization interval concentrate within radii O(epsilon (1/epsilon)) with high probability, and these estimates are uniform for small contrast. p. 0
Where the paper says it

In a bounded sample of cell size ε, with probability tending to one, all except for O(ε 1−d log(1/ε)) modes in the localisation interval concentrate within radii O(ε log(1/ε)). The estimates are uniform for the contrast-dependent capacitance family at small contrast.

Part III treats finite capacitance matrices, spectral convergence, and numerical illustrations. p. 3

Where the paper says it

Part II establishes the estimates for truncated matrices and eigenvalue spacing needed for the finite-cluster problem. Part III treats finite capacitance matrices, spectral convergence, and numerical illustrations.

A correlator estimate is derived: E Q t delta,, lambda(x, y; IW) at most A W e-nu W x-y for every fixed W > 0 and on positive intervals satisfying condition (10. p. 3
Where the paper says it

4 give the same conclusion on prescribed positive intervals under an explicit large-disorder condition. For each fixed 0 < a < 1/2 and sufficiently small W, Theorem 7.

7). Almost surely, the spectrum of H in IW is pure point, its eigenvalues are simple, and every corresponding eigenfunction satisfies psi(n) at most C, psi e-nun-y/2 for every 0 < nu < nu W and any y with psi(y) not equal to 0. p. 1
Where the paper says it

Recent OpenAI preprints [23, 24] report pure-point spectrum at every positive disorder in two dimensions and an absolutely continuous spectral interval at weak disorder in dimensions at least three, for nearest-neighbour models with independent uniform site potentials. Several recent works on resonator crystals provide the starting point of the present analysis that gives probabilistic localisation estimates for the infinite random operator and simultaneous concentration estimates for individual modes of the finite matrix corresponding to a finite cluster of resonators.

The explicit edge window is IW a = (E a(W), cM), with the energy width delta a(W) = cM - E a(W) = acMW / (1 + aW), and the frequency interval given by omega M sqrt 1 + aW < omega < omega M. p. 0
Where the paper says it

Under coordinate reflection symmetry, spectral and dynamical localisation hold near the upper spectral edge for every fixed positive disorder strength. At weak disorder W, we obtain an explicit localisation window of energy width proportional to W; prescribed positive intervals are localised at sufficiently large disorder.

Localization outside an exceptional family is established in Theorem 15. p. 1

Where the paper says it

For Anderson models with short-range hopping, the mathematical theory of localisation is based on multiscale analysis [18, 30] and fractional-moment estimates [2, 3]; see also [28]. Band-edge localisation and associated Wegner and density of states estimates for classes of random acoustic and electromagnetic operators were established in [17, 15].

1, which states that for every fixed W > 0 and on positive intervals satisfying (10. p. 0
Where the paper says it

Under coordinate reflection symmetry, spectral and dynamical localisation hold near the upper spectral edge for every fixed positive disorder strength. At weak disorder W, we obtain an explicit localisation window of energy width proportional to W; prescribed positive intervals are localised at sufficiently large disorder.

7), there is an excluded set B delta,, lambda,p(I) of eigenvalue indices in I such that every normalized eigenvector psi f j with E f j in I and j B delta,, lambda,p(I) has a center n j in lambda satisfying x n-n j > R p (2+N) psi f j(n) squared at most N - p. The final result is that the correlator bound E Q t, lambda(x, y; IW) at most A W e-nu W x-y holds for every fixed W > 0 and on positive intervals satisfying (10. p. 6
Where the paper says it

The correlator also bounds e −itGε,δ,ϖ 1I (Hϖ), by taking f(E) = e −itωε,δ(E) on I in (2. Thus the same spatial dynamical-localisation bound holds for the reduced positive-frequency evolution.

7).

The common density of states is shown to be [0, cM] almost surely by Proposition 2. p. 6

Where the paper says it

Proposition 2. For every fixed W > 0, σ(Hϖ) = [0, cM] almost surely.

2. The frequency distribution is the push-forward of the reduced energy distribution under E sqrt E, which is independent of epsilon. p. 6
Where the paper says it

ANDERSON LOCALISATION WITH LONG-RANGE INTERACTIONS 7 Notation Meaning α ∈ Y ∗ , αM Quasimomentum and upper-edge corner c(α) = Cbα scalar capacitance symbol C, cr, cM Lattice operator, convolution coefficients, upper edge γ0, A0 Decay constants: |cr| ≤ A0e −γ0|r|1 ϖ, ξn, W Configuration, independent variables, disorder width mn, Mϖ, Bϖ Inverse material factors, their diagonal, its inverse Hϖ, Ht,f ϖ,Λ Self-adjoint infinite and finite operators ω, δ, E Angular frequency, contrast, reduced spectral energy ωM Reduced edge frequency vr √ δcM/ε ∆W , IW Fixed disorder-dependent edge width and interval νW Positive decay rate in the localisation bound µW = 1 + W/2, λ = W/µW Mean inverse material factor and centred disorder Heϖ = µW Hϖ Centred operator in Section 8 s = µW E Centred energy in the interior and mobility-edge analysis Ea(W), IW,a Quantitative threshold and edge interval ε, Iε Cell size and indices in a bounded sample Table 1. Principal notation.

This proves that the absence of absolutely continuous spectrum follows from Part I. p. 2
Where the paper says it

ANDERSON LOCALISATION WITH LONG-RANGE INTERACTIONS 3 prove absence of absolutely continuous spectrum for one-dimensional capacitance operators with multiplicative disorder and critical long-range interactions. That result extends the one-dimensional long-range approach of [19] and motivates the present spectral question.

The final result for Anderson localization with long-range interactions is given by Theorem 6. p. 2
Where the paper says it

ANDERSON LOCALISATION WITH LONG-RANGE INTERACTIONS 3 prove absence of absolutely continuous spectrum for one-dimensional capacitance operators with multiplicative disorder and critical long-range interactions. That result extends the one-dimensional long-range approach of [19] and motivates the present spectral question.

1: E Q t, lambda(x, y; IW) at most A W e-nu W x-y, where IW = (cM - delta W, cM), uniformly in all finite lambda and x, y in lambda. p. 0
Where the paper says it

We establish Anderson localisation for random capacitance operators and quantify concentration of individual modes of finite capacitance matrices for clusters of subwavelength high-contrast resonators. The underlying geometry is fully periodic in dimensions d = 2, 3, with one resonator per cell, and the inverse material factors are independent and uniformly distributed.

Almost surely, the spectrum of H in IW is pure point. p. 1
Where the paper says it

Band-edge localisation and associated Wegner and density of states estimates for classes of random acoustic and electromagnetic operators were established in [17, 15]. Recent OpenAI preprints [23, 24] report pure-point spectrum at every positive disorder in two dimensions and an absolutely continuous spectral interval at weak disorder in dimensions at least three, for nearest-neighbour models with independent uniform site potentials.

The coordinate vectors are total, so the spectrum in IW is pure point. p. 1
Where the paper says it

Band-edge localisation and associated Wegner and density of states estimates for classes of random acoustic and electromagnetic operators were established in [17, 15]. Recent OpenAI preprints [23, 24] report pure-point spectrum at every positive disorder in two dimensions and an absolutely continuous spectral interval at weak disorder in dimensions at least three, for nearest-neighbour models with independent uniform site potentials.

Choosing f(E) = e-itE proves (6. p. 2
Where the paper says it

For every fixed W > 0, Theorem 6. 1 proves spectral and dynamical localisation

3).

The quantitative theorem for a fixed interval I = (E 0, E 1) with 0 W*(I;, mu), then E in I, eta>0 X x in lambda e mux-y/2 E delta x, (H, lambda-E-i eta)-1 delta y < infinity, with one bound independent of lambda and y. p. 6

Where the paper says it

ANDERSON LOCALISATION WITH LONG-RANGE INTERACTIONS 7 Notation Meaning α ∈ Y ∗ , αM Quasimomentum and upper-edge corner c(α) = Cbα scalar capacitance symbol C, cr, cM Lattice operator, convolution coefficients, upper edge γ0, A0 Decay constants: |cr| ≤ A0e −γ0|r|1 ϖ, ξn, W Configuration, independent variables, disorder width mn, Mϖ, Bϖ Inverse material factors, their diagonal, its inverse Hϖ, Ht,f ϖ,Λ Self-adjoint infinite and finite operators ω, δ, E Angular frequency, contrast, reduced spectral energy ωM Reduced edge frequency vr √ δcM/ε ∆W , IW Fixed disorder-dependent edge width and interval νW Positive decay rate in the localisation bound µW = 1 + W/2, λ = W/µW Mean inverse material factor and centred disorder Heϖ = µW Hϖ Centred operator in Section 8 s = µW E Centred energy in the interior and mobility-edge analysis Ea(W), IW,a Quantitative threshold and edge interval ε, Iε Cell size and indices in a bounded sample Table 1. Principal notation.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Precise localization of individual modes in finite clusters can be achieved by using a variational boundary comparison and a material-variable eigenvalue-spacing estimate transfer localisation from truncated matrices to individual modes of the finite capacitance matrix. This allows for quantifying concentration estimates, such as all except for O(epsilon 1-d log(1/epsilon)) modes in the localisation interval concentrate within radii O(epsilon log(1/epsilon)).

  2. The system can perform localization analysis on finite clusters by utilizing the finite capacitance matrix and proving a common limiting density of states for the truncated and finite matrices, which aids in understanding spectral convergence.

  3. Anderson localization estimates can be extended to the material-disorder operator by applying a material-variable averaging argument, allowing for the transfer of real-energy congruence to yield bounds like (7.25) EAϖ,Λ(E)−1(x, y)ϑ ≤ BW e−νW x−y1.

  4. The AI can determine the almost-sure spectrum of the random operator is exactly [0, cM] by proving that "for every fixed W > 0, sigma(Hϖ) = [0, cM] almost surely," which implies pure point spectrum with simple eigenvalues in this interval.

  5. The system can identify an explicit localisation window for spectral components proportional to the disorder strength by deriving (7.3) IW,a = (Ea(W), cM), ∆a(W) = cM − Ea(W) = acMW 1 + aW, which is sufficient for localization under specific conditions.

  6. The AI can rigorously bound the number of boundary-affected modes by proving that at most (12.5) SΛ,W W2 of these eigenvectors can have squared mass in SΛ,W greater than W2, providing a quantitative control over exceptional modes in finite samples.

  7. The system can establish spectral and dynamical localization by combining spectral averaging in one material variable with independent resampling, which leads to the correlator bound (6.1) EQ tϖ,Λ(x, y; IW) ≤ AW e−νW x−y1.

Abstract

We establish Anderson localisation for random capacitance operators and quantify concentration of individual modes of finite capacitance matrices for clusters of subwavelength high-contrast resonators. The underlying geometry is fully periodic in dimensions d=2,3, with one resonator per cell, and the inverse material factors are independent and uniformly distributed. We retain the full capacitance interaction, whose coefficients decay exponentially with lattice separation. Under coordinate reflection symmetry, spectral and dynamical localisation hold near the upper spectral edge for every fixed positive disorder strength. At weak disorder W, we obtain an explicit localisation window of energy width proportional to W; prescribed positive intervals are localised at sufficiently large disorder. A variational boundary comparison and a material-variable eigenvalue-spacing estimate transfer localisation from truncated matrices to individual modes of the finite capacitance matrix. In a bounded sample of cell size epsilon, with probability tending to one, all except for O(epsilon 1-d (1/epsilon)) modes in the localisation interval concentrate within radii O(epsilon (1/epsilon)). The estimates are uniform for the contrast-dependent capacitance family at small contrast. We also prove a common limiting density of states for the truncated and finite matrices and illustrate concentration numerically.

Sources

Related papers