Harmonic analysis on SL(2,R) and smoothness of the density of states in the one-dimensional Anderson model (Q1065462): Difference between revisions

Let \(h_{\omega}\) be a random operator in \(\ell^ 2({\mathbb{Z}})\) (one- dimensional Anderson model according to physical terminology) defined as \[ (1)\quad (h_{\omega}u)(n)=u(n+1)+u(n-1)+v_{\omega}(n)u(n),\quad n\in {\mathbb{Z}}, \] where \(v_{\omega}(n)\) are independent identically distributed random variables with distribution \(d\eta (v)=F(v)dv\). If \({\mathcal E}(n,m,\lambda)\) is the matrix of the resolution of identity of the selfadjoint operator \(h_{\omega}\), then the integrated density of states (ids) \(k(\lambda)\equiv E\{{\mathcal E}(0,0,\lambda)\}\) (\(E\{...\}\) denotes mathematical expectation). Another definition of ids is \(k(y)=\lim_{\ell \to \infty}\ell^{-1}S_ p{\mathcal E}_{\ell}(\lambda)\), where \({\mathcal E}_{\ell}(\lambda)\) is the resolution of identity of the restriction of \(h_{\omega}\) to \(\ell^ 2([0,\ell -1])\) with \(u(-1)=u(\ell)=0.\) Ids plays an important role in the Spectral analysis of random operators [see e.g. the reviewer, Spectral properties of random and almost periodic differential and finite-difference operators. Statistical physics and dynamical systems. Rigorous results, Pap. 2nd Colloq. Workshop, Köszeg/Hung. 1984, Prog. Phys. 10, 49-68 (1985); \textit{S. Kotani}, Proc. AMS Conf. Random matrices (1984)]. The function f(v) belongs to the space \(L^ P_{\alpha}\), \(1\leq p\leq \infty\), \(\alpha\geq 0\), iff there is a \(g(v)\in L^ P\) with the Fourier transform \(\hat g(\)k)\(=(1+k^ 2)^{\alpha /2}\hat f(k)\). The simple example of \(f\in L^ 1_{\alpha}\), \(0\leq \alpha <\), is the characteristic function of an interval. Theorem 1.1. If the density F(v) of the distribution of i.i.d. \(V_{\omega}(n)\) in (1) has compact support and belongs \(L^ 1_{\alpha}\) for some \(\alpha >0\), then ids k(\(\lambda)\) is a \(C^{\infty}\) function of \(\lambda\). The authors discuss also examples which show that the conditions of the theorem are close to be necessary. The proof is based on the known connection between k(\(\lambda)\) and the invariant measure of some Markov process on SL(2,\({\mathbb{R}})\) and on a subtle analysis of the increase of smoothness of convolution powers of the probability measures on SL(2,\({\mathbb{R}})\).