Duality and singular continuous spectrum in the almost Mathieu equation (Q1384842)

The authors consider an almost Mathieu operator \(h\) defined on \(\ell^2(\mathbb{Z})\), where \(h_{\lambda,\alpha,\theta}u(n) \underline\Delta u(n+1)+ u(n- 1)+\lambda\cos(\pi\alpha n+\theta)u(n)\) and study properties of the spectrum, in particular spectral properties at the self-dual point \(\lambda= 2\). Much has been proved recently about a dual relation of the spectra for a given \(\lambda\) and its dual \(4/\lambda\). The duality is explained by considering a solution \(\{a_n\}\) of the almost Mathieu equation: \[ a_{n+ 1}+ a_{n-1}+\lambda\cos(\pi\alpha n+ \theta)a_n= E a_n, \] with \(a_n\in \ell^1\). Define \(\varphi(x)= e^{i(\pi \alpha n+\theta) x}\). Then for any \(\eta\) the sequence \(u(n)= \varphi(n+ \eta/(\alpha \pi))\) obeys the equation: \[ u(n+ 1)+ u(n- 1)+ (4/\lambda)\cos (\pi\alpha n+ \eta)u(n)= (2E/\lambda) u(n). \] A naive conjecture linking the spectra at \(\lambda\) and \(4/\lambda\) was shown to be incorrect by Last. The authors prove more precise results: A point spectrum for \(\lambda\) implies an a.c. spectrum for \(4/\lambda\); an a.c. spectrum for \(\lambda\) implies a point or s.c. spectrum for \(4/\lambda\); an s.c. spectrum for \(\lambda\) implies a point or an a.c. (or s.c.) spectrum for \(4/\lambda\). (It is not known if there is any s.c. spectrum for \(\lambda<2\).) Next, the authors consider the measurability of eigenvalues. Their results imply for example that for irrational \(\alpha\) there is no a.c. spectrum if \(\lambda= 2\), and for \(\lambda<2\) the operator \(h_{4/\lambda,\alpha, \theta}\) has no a.c. spectrum and \(h_{\lambda,\alpha,\theta}\) has no point spectrum. Before reading this article, it is useful to read the expository article of \textit{Y. Last} [XI-th Int. Congr. Math. Phys., Paris, 1994, International Press, Cambridge/Mass, 366-372 (1995)].