On localization for lattice Schrödinger operators involving Bernoulli variables (Q4664560)

From MaRDI portal

!

This is the item page for this Wikibase entity, intended for internal use and editing purposes. Please use the normal view instead:

scientific article; zbMATH DE number 2152973
Language Label Description Also known as
default for all languages
No label defined
    English
    On localization for lattice Schrödinger operators involving Bernoulli variables
    scientific article; zbMATH DE number 2152973

      Statements

      5 April 2005
      0 references
      Anderson-Bernoulli lattice Schrödinger operators
      0 references
      Anderson localization
      0 references
      pure point spectrum
      0 references
      integrated density of states
      0 references
      0 references
      On localization for lattice Schrödinger operators involving Bernoulli variables (English)
      0 references
      This paper investigates various problems related to the Anderson-Bernoulli lattice Schrödinger operators, namely discrete random operators of the form \[ H(\varepsilon) = \lambda \varepsilon_n \delta_{n,n'} + \Delta, \] where \(\lambda\) is a real parameter, \(\varepsilon = (\varepsilon_n)_{n \in {\mathbb Z}^d} \in \{1, -1\}^{{\mathbb Z}^d}\), \(\delta_{n,n'}\) is the Kronecker delta in \({\mathbb Z}^d\), and \(\Delta\) is the lattice Laplacian on \({\mathbb Z}^d\). The first (and main) part of the paper aims to develop a method of establishing localization results in dimension \(d > 1\). For \(d = 1\), \(H(\varepsilon)\) is known to exhibit Anderson localization (hence pure point spectrum) for all values \(\lambda \neq 0\), almost surely in \(\varepsilon = (\varepsilon_n)_{n \in \mathbb Z} \in \{1, -1\}^{\mathbb Z}\). There are two proofs known for this fact. The first is the Furstenberg-Lepage approach through the transfer matrix. The second is based on the supersymmetric formalism in the Bernoulli context. Neither of them could so far be extended to treat the Bernoulli model in \(d > 1\) (the transfer-matrix approach is 1D, naturally; the supersymmetric approach does apply but leads to substantial difficulties). Among the spectral properties, the easiest one to establish here is the presence of a point spectrum near the edge of the spectrum. This fact should be true in arbitrary dimension. The main difficulty of establishing a Wegner-type estimate, following the usual multiscale strategy in the case of Bernoulli variables, is that proving the required probabilistic statements on the Green's operator (resolvent) \(G(\varepsilon ; E) = (H(\varepsilon) - E)^{-1}\) (where \(E\) is the energy, i.e. the eigenvalue parameter) may not be achieved by varying a single variable (as in the case of a continuous distribution). In the present setting, `rare-event' bounds are obtained considering the dependence of eigenvalues on a large collection of variables \((\varepsilon_n)_{n \in \mathcal S} \). The key result used here is a new probabilistic estimate (Lemma 2.1) that is of independent interest. To carry out this type of analysis, one needs further information on sites \(n \in {\mathbb Z}^d\) where the eigenfunction \(\xi = (\xi_n)_{n \in {\mathbb Z}^d }\) is not `too small' (as is clear from the first order eigenvalue variation). While this issue is easily taken care of in \(d = 1\) (again using the transfer matrix), for \(d > 1\) our understanding of the set \[ \{n \in {\mathbb Z}^d : | n| \sim N \text{ and } (| \xi_n| > e^{-C N} \} \] for large \(N\), is rather poor. In order to avoid this issue, the author adopts in Theorem 3.1 and Section 4 a modification of the original Bernoulli model. For instance, the continuum version considered in Section 4 is given by (here \(\Delta\) denotes the standard Laplacian on \({\mathbb R}^d\).) \[ -\Delta + \lambda \sum_{n \in {\mathbb Z}^d} \varepsilon_n \phi(x - n) \] with \(\phi(x)\) exponentially decaying for \(| x| \to \infty\) (but not compactly supported as in the usual Bernoulli-setting). The last part of the paper deals with another unsolved issue on the IDS (Integrated density of states) \(N(E)\) of the one-dimensional \(H(\varepsilon)\). Given \(\lambda \neq 0\), \(N(E)\) is known to have only a limited smoothness in the energy \(E\) and this smoothness is expected to improve for \(\lambda \to 0\) (which is also heuristically reasonable). However, at this point, one does not even seem to be able to establish a Lipschitz property of \(N(E)\), taking \(\lambda\) small enough. In fact, the known arguments implying some Hölder-continuity produce an exponent \(\alpha(\lambda)\) which deteriorates \(\lambda \to 0\). In Section 5, the Figotin-Pastur formula and elementary martingale theory are used to obtain (Proposition 5.3) a uniform (and explicit) Hölder estimate in \(\lambda\) (keeping the energy \(E\) away from the edges \(\pm 2\), and zero). For the entire collection see [Zbl 1052.46001].
      0 references

      Identifiers

      0 references
      0 references
      0 references
      0 references
      0 references
      0 references
      0 references