---
title: Exponential Dynamical Localisation in Expectation
url: https://www.emergentmind.com/topics/exponential-dynamic-localisation-in-expectation
type: topic
---

# Exponential Dynamical Localisation in Expectation

Searching arXiv for relevant papers on exponential dynamical localisation in expectation and closely related formulations.
tool call: arxiv_search({"query":"\"exponential dynamical localization\" expectation almost Mathieu operator", "max_results": 10, "sort_by": "relevance"})
tool call: arxiv_search({"query":"all:\"Exponential dynamical localization\" OR ti:\"Exponential dynamical localization\"", "max_results": 10, "sort_by": "relevance"})
Exponential dynamic localisation in expectation is a formulation of localisation for quantum dynamics in which the time-uniform transition amplitude between distant sites decays exponentially after averaging over an external parameter or random environment. In the arXiv literature supplied here, the phrase is used most explicitly for the supercritical almost Mathieu operator, where the average is taken over the phase \(\theta\), and the central conclusion is an exponential bound of the form
\[
\int_0^1 \sup_{t\in\mathbb R} \bigl|\langle \delta_k,e^{-itH_{\lambda,\alpha,\theta}}\delta_\ell\rangle\bigr|\,d\theta \le C e^{-\gamma |k-\ell|},
\]
for suitable \(\lambda\) and \(\alpha\) [1208.2674]. Closely related papers study adjacent but distinct notions—almost sure dynamical localisation, quenched localisation in probability, or deterministic zero-velocity transport bounds—making the qualifier “in expectation” mathematically decisive rather than terminological.

## 1. Formal statement of the notion

A canonical strong form of dynamical localisation is the exponentially decaying expectation bound
\[
\mathbb E\Bigl(\sup_t |\langle \delta_k,e^{-itH}\delta_\ell\rangle|\Bigr)\le Ce^{-\gamma|k-\ell|}.
\tag{1.1}
\]
In the quasi-periodic setting of the almost Mathieu operator, the expectation is not an i.i.d. disorder average but phase averaging over \(\theta\in[0,1]\). The paper "Exponential dynamical localization for the almost Mathieu operator" formulates the relevant eigenfunction-correlator decay through
\[
\gamma := \liminf_{k\to\infty} \left( -\frac{1}{|k|} \ln \mathbb E\Bigl(\sum_s |\varphi_s(0)|\,|\varphi_s(k)|\Bigr) \right),
\tag{1.2}
\]
where \(\{\varphi_s\}_s\) is a complete orthonormal basis of eigenfunctions and \(\mathbb E\) denotes integration over phase \(\theta\) [1208.2674].

The dynamical conclusion is then stated as Corollary 1.2:
\[
\int_0^1 \sup_{t\in\mathbb R} \bigl|\langle \delta_k,e^{-itH_{\lambda,\alpha,\theta}}\delta_\ell\rangle\bigr|\,d\theta \le C e^{-\gamma |k-\ell|}.
\tag{1.7}
\]
This is the precise sense in which the paper says the result “can be best called exponential dynamical localization in expectation” [1208.2674].

The same source also presents the abstract formulation in terms of bounded exponential moments of the position operator. A plausible implication is that the expectation formulation is designed to control both propagation amplitudes and weighted spatial moments of time-evolved states within a single exponential framework.

## 2. Operator-theoretic setting: the almost Mathieu regime

The operator under study is the one-frequency almost Mathieu operator on \(\ell^2(\mathbb Z)\),
\[
(H_{\lambda,\alpha,\theta}u)(n)=u(n+1)+u(n-1)+2\lambda \cos\bigl(2\pi(n\alpha+\theta)\bigr)\,u(n),
\tag{1.3}
\]
with coupling \(\lambda>0\), irrational frequency \(\alpha\), and phase \(\theta\) [1208.2674].

The localisation theorem is proved in the supercritical regime
\[
\lambda>1.
\]
The arithmetic hypothesis is expressed either as Diophantine frequency,
\[
\|q\alpha\|=\operatorname{dist}(q\alpha,\mathbb Z)\ge \frac{\kappa}{q^\tau},\qquad q\ge 1,
\tag{1.4}
\]
or, more generally, by requiring
\[
\beta(\alpha):=\limsup_{q\to\infty}\frac{-\ln \|q\alpha\|}{q}
\]
to be sufficiently small. The theorem is stated for \(\beta(\alpha)<\beta\) for some \(\beta>0\), so it includes all Diophantine \(\alpha\) and some weakly Liouville frequencies [1208.2674].

The phase parameter plays a dual role. Spectral localisation is available for almost every \(\theta\), but the dynamical statement is averaged over \(\theta\). This distinction is essential: the result is not uniform in \(\theta\), and it is not an almost-sure-in-\(\theta\) estimate with a \(\theta\)-independent constant. The expectation in this setting is therefore an ergodic phase average rather than a disorder expectation in the Anderson sense [1208.2674].

The paper also stresses that pure point spectrum with exponentially decaying eigenfunctions is a spectral statement, whereas dynamical localisation concerns the time-evolution operator \(e^{-itH}\). The latter is stronger: pure point spectrum alone does not automatically imply a uniform-in-time bound on propagation amplitudes [1208.2674].

## 3. Eigenfunction correlators and localisation centres

The proof mechanism proceeds through a reduction from dynamics to eigenfunction correlators. For a general family \(\{H_x\}_{x\in X}\) of self-adjoint operators on \(\ell^2(\mathbb Z^d)\), with pure point spectrum for \(\mu\)-a.e. \(x\), one chooses for each eigenfunction \(\varphi_{x;s}\) a localisation centre \(n_{x;s}\in\mathbb Z^d\) satisfying
\[
|\varphi_{x;s}(n_{x;s})| = \|\varphi_{x;s}\|_{\ell^\infty(\mathbb Z^d)}.
\tag{2.1}
\]
This centre decomposition is the structural device used to convert spectral localisation data into dynamical estimates [1208.2674].

The key abstract bound is
\[
\bigl|\langle \delta_k,e^{-itH_x}\delta_\ell\rangle\bigr| \le \sum_s |\varphi_{x;s}(k)|\,|\varphi_{x;s}(\ell)|
\]
and, after regrouping according to the centre \(n=n_{x;s}\),
\[
\bigl|\langle \delta_k,e^{-itH_x}\delta_\ell\rangle\bigr| \le \sum_n \left( \sum_{n_{x;s}=n} |\varphi_{x;s}(k)|^2 \right)^{1/2} \left( \sum_{n_{x;s}=n} |\varphi_{x;s}(\ell)|^2 \right)^{1/2}.
\]
After averaging in \(x\), the same structure persists with integrals over the parameter space [1208.2674].

In one dimension, the decisive hypothesis becomes an exponential bound on the averaged centre-grouped \(\ell^2\)-mass:
\[
\int \sum_{n_{x;s}=n} |\varphi_{x;s}(\ell)|^2\,d\mu(x) \le C e^{-2\gamma |n-\ell|}.
\tag{2.2}
\]
From this one obtains
\[
\int \bigl|\langle \delta_k,e^{-itH_x}\delta_\ell\rangle\bigr|\,d\mu(x) \le C\Bigl(\frac{1}{\gamma}+|k-\ell|\Bigr)e^{-\gamma |k-\ell|}.
\tag{2.3, \(d=1\)}
\]
Thus exponential decay of centre-grouped eigenfunction mass implies exponential decay of the averaged propagator. The conceptual point is that orthogonality is exploited within each centre sector rather than estimating the full eigenfunction expansion term-by-term [1208.2674].

## 4. Resonances, almost localisation, and phase averaging

The nontrivial model-specific input is an almost-localisation theorem for solutions of the almost Mathieu difference equation. Resonances are defined by the arithmetic condition that \(k\in\mathbb Z\) is \(\eta\)-resonant for \(\theta\) if
\[
\|2\theta-k\alpha\|\le e^{-\eta |k|}.
\tag{3.1}
\]
These resonant scales are the only locations where uniform exponential decay can fail [1208.2674].

Away from resonances, the invoked theorem states that for \(\lambda,C_0>1\), there exist \(\eta(\lambda)>0\) and \(\beta(\lambda,C_0)>0\) such that if \(\beta(\alpha)<\beta\), then there are \(C_1,\gamma>0\) with the following property: every solution \(u\) of
\[
h_{\lambda,\alpha,\theta}u = Eu
\]
satisfying
\[
u(0)=1,\qquad |u(n)|\le 1+|n|
\]
obeys
\[
|u(n)|\le C_1 e^{-\gamma |n|}
\tag{3.3}
\]
whenever \(n\) lies in the nonresonant region
\[
C_0(1+|k_j|)\le |n|\le C_0^{-1}|k_{j+1}|,
\]
where \(k_j\) are the \(\eta\)-resonances of \(\theta\) [1208.2674].

This almost-localisation estimate is converted into an averaged eigenfunction bound:
\[
\int_0^1 \left(\sum_{n_{\theta;s}=n} |\varphi_{\theta;s}(\ell)|^2\right)\,d\theta \le C_1^2 e^{-2\gamma |n-\ell|} + e^{-\frac{\eta}{C_0}|n-\ell|}.
\tag{3.4}
\]
The mechanism is twofold. For nonresonant phases, eigenfunctions centred at \(n\) decay exponentially at \(\ell\). For resonant phases, one uses only a trivial bound, but the measure of the exceptional phase set is exponentially small in \(|n-\ell|\). The expectation estimate is therefore created by phase averaging over a sparse resonant set rather than by a pointwise-in-\(\theta\) dynamical argument [1208.2674].

Once this averaged centre-grouped decay is inserted into the abstract reduction, one obtains the exponential propagator bound of Corollary 1.2. The same argument yields bounded exponential moments of the position operator, as stated in the abstract [1208.2674].

## 5. Distinction from neighbouring localisation concepts

The phrase “in expectation” can be confused with several nearby but inequivalent formulations. The distinction is explicit in the supplied literature.

For quasi-one-dimensional random operators on strips, the relevant result is an almost sure sharp exponential decay of the eigenfunction correlator,
\[
\mathbb P\left\{\limsup_{x\to \pm\infty}\frac1{|x|}\log Q_I(x,y)\le -\inf_{E\in I}\gamma_W(E)\right\}=1,
\tag{8}
\]
from which one gets almost sure dynamical localisation and exponential decay of the Fermi projection. The paper does not state an Aizenman-type expectation theorem; the probabilistic mode is almost sure, not in expectation [2110.00097].

For the parabolic Anderson model with Weibull potential, localisation is formulated for the normalised quenched solution \(u(t,z)/U(t)\). The central profile theorem states that, uniformly on a mesoscopic neighbourhood of the peak,
\[
\frac{ \log \left( \frac{u(t, z)}{U(t)}\right) }{ \frac{1}{\gamma}|z-Z_t^{(1, \rho)}|\log \log t }  \to -1
\qquad \text{in probability},
\]
and complete localisation means
\[
\frac{u(t, Z_t^{(1, \rho)})}{U(t)} \to 1
\qquad \text{in probability}.
\]
These are quenched high-probability statements about a random evolving mass profile, not annealed expectation bounds [1311.7634].

In the many-body setting, a different adjacent notion appears: deterministic zero-velocity Lieb–Robinson-type bounds such as
\[
\|A(t)-e^{itH_A^l}Ae^{-itH_A^l}\| \le c_{\mathrm{loc}} e^{-\mu l},
\tag{3}
\]
or low-energy commutator bounds of the form
\[
\bigl| \mathrm{tr}\!\left(\rho [A(t),B]\right)\bigr| \le \min(t,1)\, c_{\mathrm{mob}} e^{-\mu \dist}.
\tag{7}
\]
These assumptions imply exponential clustering of eigenvectors and, in one dimension, area laws and matrix-product-state approximability. However, they are deterministic statements for a fixed Hamiltonian and do not involve disorder averages \(\mathbb E[\cdot]\) [1409.1252].

A further distinct use of localisation occurs in theory-space Anderson models, where a disordered nearest-neighbour mass matrix yields exponentially localised mass eigenvectors,
\[
\left|v_j^{(k)}\right| \sim \left|v_{k_0}^{(k)}\right|\, e^{-\frac{|j-k_0|}{L_k}},
\]
and loop-generated deterministic non-local couplings may preserve, weaken, or enhance that static localisation. This is neither time-dependent dynamical localisation nor an expectation theorem [2501.05137].

The common misconception is therefore to identify every exponential localisation estimate with the same probabilistic notion. In the supplied corpus, “in expectation,” “almost sure,” “in probability,” and “deterministic” label materially different objects, norms, and averaging procedures.

## 6. Scope, limitations, and open directions

The expectation theorem for the almost Mathieu operator is proved for \(\lambda>1\) and for frequencies satisfying \(\beta(\alpha)<\beta\), in particular all Diophantine \(\alpha\) [1208.2674]. It does not treat all irrational frequencies, it does not cover the critical or subcritical regimes \(\lambda\le 1\), and it does not produce a pointwise-in-\(\theta\) exponential dynamical localisation statement with uniform constants. The decay rate \(\gamma\) is positive but non-explicit, because the underlying almost-localisation estimates are themselves nonquantitative [1208.2674].

The paper also isolates several natural unresolved questions. One is whether the limit in
\[
\gamma := \liminf_{k\to\infty} \left( -\frac{1}{|k|} \ln \mathbb E\Bigl(\sum_s |\varphi_s(0)|\,|\varphi_s(k)|\Bigr) \right)
\]
exists as an actual limit rather than only a \(\liminf\). Another is whether, in one dimension, \(\gamma\) coincides with the minimal Lyapunov exponent. A further question is whether analogous positive decay rates can be established for other quasi-periodic models in regimes of positive Lyapunov exponent [1208.2674].

Within the literature represented here, the significance of exponential dynamic localisation in expectation is therefore precise. It is a strong dynamical statement, stronger than mere pure point spectrum, formulated through exponentially decaying expected propagator amplitudes or eigenfunction correlators. In the almost Mathieu case the expectation is phase averaging; in other contexts one instead encounters almost-sure strip localisation, quenched localisation in probability for the parabolic Anderson model, deterministic zero-velocity transport suppression in many-body systems, or static localisation of mass eigenvectors in theory space [1208.2674]. This suggests that the phrase is best understood not as a generic synonym for localisation, but as a specific probabilistic mode of dynamical control.

Source: https://www.emergentmind.com/topics/exponential-dynamic-localisation-in-expectation