Spectral properties of not necessarily self-adjoint linear differential operators (Q1813500)

The author studies, by functional analytic techniques, a class of higher- order ordinary differential operators which, although somewhat special in form, reflect the general structure of large classes of higher differential operators. The basic question is this: given a higher-order linear ordinary differential operator \(L\) on \([1, \infty)\), how does one calculate the number of solutions to \(Lf= \lambda f\) which lie in \(L_ 2[1, \infty)\), for a complex number \(\lambda\), and how does one calculate the essential spectrum of \(A- \lambda I\), where \(A\) is a differential operator, between the minimal and maximal operators, associated with the expression \(L\)? The author answers these questions completely for a class of expressions which correspond to polynomial coefficient expressions, with certain restrictions put on the powers and leading coefficients of the polynomials. One interesting aspect of the paper is that one might expect that the study of this sort of problem is best done by asymptotic formulas for the solutions. Indeed, there have been many successes obtained by this approach. However, the results of this paper go well beyond any such results known to the reviewer, and thus indicate the power also of functional analytic methods to decide questions, such as those above, which are of a functional analytic nature. Without looking into too much detail about the large number of results of this paper, we give a sketch of the results. Define a ``simple expression'' \(M_ 0\) to be an expression of type \(M_ 0= \sum^ r_{\sigma= 0} a_ \sigma t^{\alpha(\sigma)} y^{(\rho(\sigma))}\), where \(y^{(m)}\) denotes the \(m\)th derivative of \(y\), and where \(r\) is a positive integer, \(\rho(\sigma)\) is a nonnegative integer with \(0= \rho(0)< \rho(1)< \rho(r)= n\), and \(\alpha(\sigma)\in\mathbb{R}\), with \(\alpha(0)= 0\), \(\alpha(1)\leq \rho(1)\), and \[ 1\geq {\alpha(\sigma)- \alpha(\sigma- 1)\over \rho(\sigma)- \rho(\sigma- 1)}\geq {\alpha(\sigma+ 1)- \alpha(\sigma)\over \rho(\sigma+ 1)- \rho(\sigma)} \] for \(\sigma= 1,\dots, r- 1\). Let \(\sigma(1)<\cdots< \sigma(s- 1)\) denote those indices \(\sigma\) \((\sigma= 1,\dots, r- 1)\) for which the strong inequality holds above. Suppose that \(\text{sgn}((- 1)^{\rho(\sigma)/2}a_ \sigma)=\text{const}\) for all \(\sigma\geq \sigma(1)\) such that \(\rho(\sigma)\) is even, and \(\text{sgn}((- 1)^{\rho(\sigma+ 1)/2} a_ \sigma)= \text{const}\) for all \(\sigma\geq \sigma(1)\) such that \(\rho(\sigma)\) is odd. Let \(L= M_ 0+ M\), where the hypotheses on the coefficients of \(M\) are such that relative compactness of the operators associated with \(M\) compared with those associated with \(M_ 0\) may be proved; part of the theory is the quite nontrivial calculation of just what coefficients may be allowed for \(M\). Then the questions of the first paragraph are answered for \(L\), in terms of direct computations from the coefficients of \(M_ 0\). Many examples are discussed in the paper, one of which is a sixth-order example, due to the reviewer, of an expression \(L= - D^ 3 t^ \gamma D^ 3+ \beta t^{\gamma- 6}\), with \(\gamma> 6\) and \(\beta> 0\), which has more than three \(L_ 2\) solutions, settling a problem of W. N. Everitt. However, the exact number of solutions was not calculated by the reviewer; this was done later by Paris and Wood, who used powerful asymptotic techniques. This paper answers the questions of the first paragraph completely for expressions \(L\) of the above type, obtaining by a different method the results of Paris and Wood. In conclusion, this paper is a valuable contribution to the literature, using techniques which are not the standard ones, and which in the hands of the author have become very powerful.