Optimal Spectral Inequalities for Landau Operators

• The paper establishes optimal spectral inequalities by providing explicit lower bounds on the L2-mass of spectral subspaces over thick sets with constants depending on energy, magnetic field, and geometry.
• It employs families of magnetic Bernstein inequalities and holomorphy arguments to extend classical Logvinenko–Sereda-type inequalities into the magnetic quantum setting, with demonstrated sharpness via Gaussian quasimodes.
• The results have practical implications for control theory and Anderson localization, ensuring null-controllability for the magnetic heat equation and enabling strong dynamical localization in continuum models.

Optimal spectral inequalities for Landau operators establish precise lower bounds on the $L^2$-mass of functions from spectral subspaces of finite energy when restricted to "thick" or relatively dense sets. These results generalize classical Logvinenko–Sereda-type inequalities to the magnetic quantum setting, providing explicit and asymptotically sharp constants dependent on energy, magnetic field strength, and geometric parameters of the sampling set. The principal technical tools are families of magnetic Bernstein inequalities, which control magnetic derivatives and facilitate holomorphy arguments crucial for the lower bounds. The sharpness of these inequalities is supported by extremal examples and tight Gaussian estimates, and they have significant implications for control theory, spectral theory, and Anderson localization.

1. The Landau Operator: Structure and Spectral Subspaces

In two dimensions, the Landau operator is the magnetic Laplacian with constant field $B>0$, commonly written in symmetric gauge as

$$H_B = (i\nabla - A)^2, \qquad A(x)=\tfrac{B}{2}(-x_2,\,x_1),$$

with spectral values $\sigma(H_B) = \{(2n+1)B\,|\,n=0,\,1,\,2,\dots\}$. The spectral projector onto energies up to $E$ is $P_E = 1_{(-\infty, E]}(H_B)$, so that the finite-energy spectral subspace is $\operatorname{Ran}1_{(-\infty, E]}(H_B)$ (Pfeiffer et al., 2023).

Higher-dimensional Landau operators adopt:

$H_B = (-i \nabla_x - A(x))^2,$

with $A(x) = B x$ for a constant skew-symmetric $B \in \mathbb{R}^{d \times d}$, and analogous spectral subspaces $B>0$0 are studied (Özcan et al., 5 Jan 2026).

2. Definition and Characterization of Thick Sets

A measurable set $B>0$1 is $B>0$2-thick if for given side-lengths $B>0$3 and $B>0$4, every axis-aligned rectangle $B>0$5 of side lengths $B>0$6 satisfies:

$B>0$7

These sets, also referred to as relatively dense, are precisely the geometric sampling sets for which optimal spectral inequalities are valid (Pfeiffer et al., 2023, Özcan et al., 5 Jan 2026). The parameter $B>0$8 (sampling ratio) enters only through universal constants in the inequalities.

3. Optimal Spectral Inequalities: Statements and Sharpness

The central result for $B>0$9 asserts that for $$H_B = (i\nabla - A)^2, \qquad A(x)=\tfrac{B}{2}(-x_2,\,x_1),$$0-thick $$H_B = (i\nabla - A)^2, \qquad A(x)=\tfrac{B}{2}(-x_2,\,x_1),$$1 and all $$H_B = (i\nabla - A)^2, \qquad A(x)=\tfrac{B}{2}(-x_2,\,x_1),$$2,

$$H_B = (i\nabla - A)^2, \qquad A(x)=\tfrac{B}{2}(-x_2,\,x_1),$$3

where $$H_B = (i\nabla - A)^2, \qquad A(x)=\tfrac{B}{2}(-x_2,\,x_1),$$4 and constants $$H_B = (i\nabla - A)^2, \qquad A(x)=\tfrac{B}{2}(-x_2,\,x_1),$$5 depend only on universal features (not $$H_B = (i\nabla - A)^2, \qquad A(x)=\tfrac{B}{2}(-x_2,\,x_1),$$6, $$H_B = (i\nabla - A)^2, \qquad A(x)=\tfrac{B}{2}(-x_2,\,x_1),$$7, or $$H_B = (i\nabla - A)^2, \qquad A(x)=\tfrac{B}{2}(-x_2,\,x_1),$$8) (Pfeiffer et al., 2023).

In dimensions $$H_B = (i\nabla - A)^2, \qquad A(x)=\tfrac{B}{2}(-x_2,\,x_1),$$9, for $\sigma(H_B) = \{(2n+1)B\,|\,n=0,\,1,\,2,\dots\}$0-thick $\sigma(H_B) = \{(2n+1)B\,|\,n=0,\,1,\,2,\dots\}$1,

$\sigma(H_B) = \{(2n+1)B\,|\,n=0,\,1,\,2,\dots\}$2

with $\sigma(H_B) = \{(2n+1)B\,|\,n=0,\,1,\,2,\dots\}$3 and the operator norm $\sigma(H_B) = \{(2n+1)B\,|\,n=0,\,1,\,2,\dots\}$4 (any norm equivalent to Frobenius) (Özcan et al., 5 Jan 2026).

Sharpness in $\sigma(H_B) = \{(2n+1)B\,|\,n=0,\,1,\,2,\dots\}$5 and the quadratic dependence on $\sigma(H_B) = \{(2n+1)B\,|\,n=0,\,1,\,2,\dots\}$6 and $\sigma(H_B) = \{(2n+1)B\,|\,n=0,\,1,\,2,\dots\}$7 is demonstrated by evaluating these bounds on explicit Landau-level Gaussian quasimodes and by considering limit regimes where box dimensions diverge.

4. Magnetic Bernstein Inequalities and Analyticity

Magnetic Bernstein inequalities provide uniform $\sigma(H_B) = \{(2n+1)B\,|\,n=0,\,1,\,2,\dots\}$8-bounds for all magnetic derivatives of functions in the spectral subspace:

$\sigma(H_B) = \{(2n+1)B\,|\,n=0,\,1,\,2,\dots\}$9

for $E$0 (Pfeiffer et al., 2023), and more generally for $E$1,

$E$2

with explicit control by energy and magnetic field (Özcan et al., 5 Jan 2026).

These bounds, together with factorial control on all derivatives, imply analyticity of $E$3 in a uniform complex polydisc about every real point, which is essential for extending classical Remez-type lower bounds to the magnetic setting.

5. Covariant Covering Arguments and Remez-Type Estimates

The proof technique adapts Kovrijkine's Logvinenko–Sereda methodology. The domain is covered by disjoint rectangles of side $E$4, partitioned into "good" and "bad" sets according to whether Bernstein and analyticity estimates hold locally. The $E$5-mass in bad rectangles is shown to be minor via dyadic summation and Chebyshev-type arguments.

On each good rectangle, a reduction to one-dimensional slices allows application of analytic Remez-type arguments: a holomorphic function's sup on an interval is quantitatively controlled by its mass on any positive measure subset. Summation over good rectangles yields the global inequality.

6. Links to Carlson–Landau and Lieb–Thirring Spectral Inequalities

Optimal spectral inequalities for Landau operators generalize and connect to the Carlson–Landau and Lieb–Thirring inequalities for periodic and magnetically perturbed Schrödinger operators (Ilyin et al., 2015). In periodic and cylinder geometries, the sharp constants depend on flux parameters:

$E$6

where $E$7 encodes magnetic flux dependence, and $E$8 is the classical phase-space constant. The minimal exponent $E$9 reflects the sharp semiclassical regimes, with best constants attained for large-field Landau potentials (Ilyin et al., 2015).

7. Applications: Null-Controllability and Anderson Localization

Spectral inequalities underpin major results in control theory: for the magnetic heat equation

$P_E = 1_{(-\infty, E]}(H_B)$0

null-controllability is achieved over any $P_E = 1_{(-\infty, E]}(H_B)$1-thick set $P_E = 1_{(-\infty, E]}(H_B)$2, with the control cost bounded sharply in terms of $P_E = 1_{(-\infty, E]}(H_B)$3 and $P_E = 1_{(-\infty, E]}(H_B)$4:

$P_E = 1_{(-\infty, E]}(H_B)$5

known to be optimal in the small-$P_E = 1_{(-\infty, E]}(H_B)$6 and large-$P_E = 1_{(-\infty, E]}(H_B)$7 regimes (Pfeiffer et al., 2023, Özcan et al., 5 Jan 2026).

For Anderson localization in continuum alloy-type models with Landau background, optimal Wegner estimates and scale-free unique continuation properties flow from these spectral inequalities. They enable rigorous proofs of strong dynamical localization near the spectral edge, with minimal assumptions on the support of the single-site potential (Pfeiffer et al., 2023).

Table 1: Core Inequalities and Domains

Inequality Type	Magnetic Domain	Best Constant Dependencies
Spectral (LS-type)	$P_E = 1_{(-\infty, E]}(H_B)$8, thick sets	$P_E = 1_{(-\infty, E]}(H_B)$9, $\operatorname{Ran}1_{(-\infty, E]}(H_B)$0, $\operatorname{Ran}1_{(-\infty, E]}(H_B)$1, $\operatorname{Ran}1_{(-\infty, E]}(H_B)$2
Carlson–Landau	$\operatorname{Ran}1_{(-\infty, E]}(H_B)$3, periodic	flux parameter $\operatorname{Ran}1_{(-\infty, E]}(H_B)$4
Lieb–Thirring	Cylinder, torus	$\operatorname{Ran}1_{(-\infty, E]}(H_B)$5, $\operatorname{Ran}1_{(-\infty, E]}(H_B)$6

These results collectively provide an optimal framework for quantitative spectral bounds in Landau-type operators, extending and synthesizing classical Euclidean and periodic analysis via sharp magnetic and analytic estimates (Pfeiffer et al., 2023, Özcan et al., 5 Jan 2026, Ilyin et al., 2015).