Spectral bounds for ergodic Jacobi operators
Abstract.
We consider ergodic Jacobi operators and obtain estimates on the Lebesgue measure and the distance between maximum and minimum points of the spectrum in terms of the Lyapunov exponent. Our proofs are based on results from logarithmic potential theory and their connections with spectral theory of Jacobi operators.
Key words and phrases:
ergodic Jacobi operators, spectral estimates, logarithmic capacity, Lyapunov exponent, density of states measure2020 Mathematics Subject Classification:
47B36, 47B80, 47E051. Introduction
The Jacobi operator acting on the Hilbert space is the self-adjoint operator associated with the infinite Jacobi matrix
given by
where , and for any . Discrete SchrΓΆdinger operators on the real line give a subclass, where for any . When the index set is one gets the two-sided Jacobi matrix, but for simplicity of the formulation we will stick with the index set .
If is a probability measure space and is an ergodic transformation, then one defines Jacobi parameters for and as
where and are measurable functions from to with positive, bounded and invertible, and bounded. Then defines an ergodic family of Jacobi operators with parameters .
Ergodic Jacobi, and specifically ergodic SchrΓΆdinger operators appear in solid state physics and quantum Hall effect, serving various models such as crystals, disordered systems or quasi-crystals, see [CFKS87, DF22] and references therein. Besides appearing in models from mathematical physics, the study of ergodic Jacobi and SchrΓΆdinger operators have been an active research area, combining ideas from dynamical systems, analysis, potential theory and topology. Moreover, Jacobi operators with stochastic parameters show unusual spectral behavior like dense point spectrum, singular continuous spectrum and Cantor spectrum. For example, the almost Mathieu operator is one of the well-studied ergodic Jacobi operators, where the Jacobi parameters are and with and . For any , the spectrum of the almost Mathieu operator is a Cantor set. This is known as the Ten Martini Problem and was solved by Avila and Jitomirskaya [AJ09]. For , the spectrum of the almost Mathieu operator is purely singular continuous, which was also proved by Jitomirskaya [J21].
The spectrum of is nonrandom as follows: For an ergodic family , there exists a set independent of such that for almost every with respect to the measure [CFKS87]. Throughout the paper we call this set the spectrum of . Same nonrandomness is valid for the absolutely continuous, the singular continuous and the pure point spectra [CFKS87].
In this paper we consider bounds on the Lebesgue measure and the distance between maximum and minimum points of the spectrum . Similar bounds were obtained for periodic Jacobi matrices [H24]. Our bounds depend on two fundamental object, the Lyapunov exponent and the density of states measure, both of which are introduced in Section 2. These bounds are obtained using tools from logarithmic potential theory, which are also introduced in Section 2.
The paper is organized as follows.
Section 2 includes preliminaries required for our results, namely some basics of logarithmic potential theory, and the definitions and some properties of the density of states measure and the Lyapunov exponent for ergodic Jacobi operators.
Section 3 includes proofs of the following results and some remarks on them.
-
β’
In Theorem 3.2, we consider an almost sure nonrandom upper bound for the Lebesgue measure of the spectrum in terms of the Lyapunov exponent and the off-diagonal parameters of an ergodic Jacobi operator.
-
β’
Theorem 3.5 shows an almost sure nonrandom lower bound on the difference between maximum and minimum energies in terms of the logarithmic energy of the density of states measure, or equivalently the Lyapunov exponent and the off-diagonal parameters of an ergodic Jacobi operator.
-
β’
In Theorem 3.9, we prove that in the case that the Lyapunov exponent is HΓΆlder continuous on the spectrum, the difference between maximum and minimum energies is almost surely bounded below by nonrandom numbers depending on the Lyapunov exponent and the off-diagonal parameters of an ergodic Jacobi operator and HΓΆlder continuity parameters of the Lyapunov exponent.
2. Preliminaries
We obtain our results using estimates on logarithmic capacity, so letβs recall some basics of logarithmic potential theory, which can be found e.g. in [R95].
Definition 2.1.
Let be a finite Borel measure, supported on a compact subset of the complex plane. Then logarithmic potential of is the function defined by
Definition 2.2.
Let be a finite Borel measure, supported on a compact subset of the complex plane. Its logarithmic energy is defined by
Definition 2.3.
Let be a compact subset of and be the set of Borel probability measures compactly supported inside . The measure is called the equilibrium measure for if
Now we are ready to define the logarithmic capacity.
Definition 2.4.
The logarithmic capacity of a subset of the complex plane is given by
In particular if is compact with equilibrium measure , then .
We will use following properties of the logarithmic capacity in our results:
Theorem 2.5.
([R95] Theorem 5.1.2(a), Corollary 5.2.4, Theorem 5.3.1, Theorem 5.3.2(c))
-
(a)
If , then .
-
(b)
If , then .
-
(c)
Let be a compact subset of , and let be a map satisfying
where , and and are positive constants. Then
(2.1) -
(d)
Let be a compact subset of . Then
(2.2) where denotes the Lebesgue measure of .
Next, we introduce two fundamental objects related with ergodic Jacobi operators: the density of states measures and the Lyapunov exponent. These two objects and more on the spectral theory of ergodic Jacobi operators can be found e.g. in [CFKS87, DF22, T00].
Definition 2.6.
Let be a family of ergodic Jacobi operators defined as in the Introduction section. The density of states measure is the measure defined by
(2.3) |
for bounded measurable , where
The density of states measure is a probability measure supported on the spectrum.
Given an ergodic family of Jacobi operators, the solutions of the eigenvalue problem
for and are represented by the transfer matrices as
where
(2.4) |
for , , and
(2.5) |
Following result introduces the Lyapunov exponent.
Theorem 2.7.
The Lyapunov exponent is subharmonic on the complex plane. It is harmonic and positive on . The famous Thouless formula relates the Lyapunov exponent and the density of states measure.
Theorem 2.8.
(Thouless Formula) ([S07], Theorem 7.1(f)) For every , we have
(2.6) |
where for .
Thouless Formula also relates spectral theory with potential theory as the right hand side of (2.6) is the negative of the logarithmic potential of the density of states measure.
3. Results, proofs and some remarks
In our results, the following lemma will be a main tool that represents the logarithmic capacity of the spectrum of an ergodic family of Jacobi operators in terms of the Lyapunov exponent and the equilibrium measure of the spectrum.
Lemma 3.1.
([S07] Equation (1.37)) Let be a family of ergodic Jacobi operators with parameters and be its Lyapunov exponent. Then for a.e. ,
(3.1) |
where and is the equilibrium measure of the spectrum.
Theorem 3.2.
Let be a family of ergodic Jacobi operators with parameters and be its Lyapunov exponent. Then for a.e. ,
(3.2) |
where and denotes the Lebesgue measure of the set .
Proof.
Remark 3.3.
The equality is obtained for the discrete Laplacian, i.e. , since and for .
Remark 3.4.
For some important classes of ergodic Jacobi operators, the Lyapunov exponent is zero on the spectrum, so inequality (3.2) becomes
Some of these classes are as follows:
-
(1)
subcritical or critical almost Mathieu operator: The almost Mathieu operator is one of the best-understood examples of ergodic SchrΓΆdinger operators, where the potential is with frequency and phase in and coupling constant . When (subcritical) or (critical), the Lyapunov exponent is zero everywhere on the spectrum.
-
(2)
periodic SchrΓΆdinger operators: By Floquet-Bloch theory the spectrum of a periodic SchrΓΆdinger operator has band-gap structure and purely absolutely continuous spectrum [Kuc16, T00]. Since the essential closure of the zero set of the Lyapunov exponent is the ac-spectrum [DF22], Lyapunov exponents for periodic Jacobi operators are zero on the spectrum.
-
(3)
analytic quasi-periodic SchrΓΆdinger operators without supercritical-regime: For analytic quasi-periodic SchrΓΆdinger operators, the potential is generated from the real valued analytic function defined on with phase and irrational frequency . Avila obtained spectral characterizations for analytic potentials using complexified Lyapunov exponents [Avi15]. He considers the Lyapunov exponent with
which is obtained by complexifying the phase using analyticity of the potential . We denote this complexified Lyapunov exponent by . Then the spectrum with an analytic potential decomposes into three mutually disjoint sets as follows:
-
β’
The energy is subcritical if is zero in a neighborhood of .
-
β’
The energy is critical if is zero at , but is not subcritical.
-
β’
The energy is supercritical if is positive at .
According to this classification of energies, if the supercritical regime is empty, then the Lyapunov exponent is zero everywhere on the spectrum.
-
β’
-
(4)
strictly ergodic subshift operators: This class of ergodic SchrΓΆdinger operators are given by strictly ergodic subshifts on finite alphabets [Dam07, S07]. It is expected that the majority of these operators have purely singular continuous spectrum as a Cantor set of zero Lebesgue measure and zero Lyapunov exponent on the spectrum [Dam07].
Theorem 3.5.
Let be a family of ergodic Jacobi operators with Jacobi parameters . Also let and , and denote the logarithmic energy of the measure . Then
(3.3) |
where is the density of states measure of . Equivalently we have
(3.4) |
where for a.e. and is the Lyapunov exponent of .
Proof.
Using Lemma 3.1 and the Thouless formula (Theorem 2.8) and recalling that denotes the logarithmic potential of a measure we get
(3.5) | ||||
(3.6) | ||||
(3.7) | ||||
(3.8) | ||||
(3.9) |
The inequality above follows from the definition of the equilibrium measure and the fact that density of states measure is a probability measure supported on .
Note that the logarithmic energy of the density of states measure is given in terms of the Lyapunov exponent by Thouless formula as
(3.10) |
so we can replace (3.9) by
On the other hand logarithmic capacity is monotone and the logarithmic capacity of an interval is the one fourth of the size of that interval (Theorem 2.5 (a),(b)), so we have
(3.11) |
and hence get the desired results. β
Remark 3.6.
The equality for (3.4) is obtained for the discrete Laplacian, i.e. , since and for .
Remark 3.7.
If the Lyapunov exponent is zero everywhere on the spectrum, specifically for the classes we discussed in Remark 3.4 we have
(3.12) |
Remark 3.8.
If the Lyapunov exponent is zero everywhere on , then we showed that
(3.13) |
Therefore if the spectrum is a continuum set in this case, then
(3.14) |
If the Lyapunov exponent is HΓΆlder continuous, then we obtain another lower estimate for the logarithmic capacity of the spectrum and hence for the difference between maximum and minimum energies in terms of the Lyapunov exponent and HΓΆlder continuity parameters.
Theorem 3.9.
Let be a family of ergodic Jacobi operators with parameters and be its Lyapunov exponent. Also let and . If is -HΓΆlder continuous on the spectrum for , i.e.
(3.15) |
for and , then
(3.16) |
where and denotes the Lebesgue measure of a set .
Proof.
Using items (a),(b),(c) of Theorem 2.5, our assumption and the estimate (3.11) we get
(3.17) |
On the other hand, the Lyapunov exponent is real-valued and we assumed it to be continuous on the spectrum, so is a compact subset of the real line. Therefore by item (d) of Theorem 2.5 we get
(3.18) |
and hence the desired result. β
Remark 3.10.
It is known that [DF22], so (3.16) is non-trivial if .
Remark 3.11.
If the image of the spectrum under the Lyapunov exponent is a continuum set, then the measure of is the difference between maximum and minimum values of over . Therefore inequality (3.16) becomes
(3.19) |
Moreover if the Lyapunov exponent is zero at some energy, e.g. if the ac-spectrum is non-empty, then from the non-negativity of the Lyapunov exponent, inequality (3.16) becomes
(3.20) |