Methoden der Zwischenprobleme zweiter Art bei Schranken für Eigenwerte rationaler Eigenwertaufgaben. (Methods of intermediate problems of the second kind for eigenvalues of rational eigenvalue problems) (Q917874)

Consider the eigenvalue problem \[ (1)\quad If-\lambda Af-\lambda^ 2H_ 0f+\sum^{n}_{k=1}(\lambda^ 2/(\lambda -a_ k))H_ kf=M(\lambda)f=0, \] where (I): A is a compact, \(H_ 0,...,H_ n\) are finite dimensional operators on a complex Hilbert space \({\mathcal H}\) and \(a_ 0,...,a_ n\) are different nonzero complex numbers, or (II): the conditions (I) hold with A, \(H_ 0,...,H_ n\) selfadjoint, \(a_ 0,...,a_ n\in {\mathbb{R}}\), \(H_ 0\), \(\frac{1}{a_ 1}H_ 1\),...,\(\frac{1}{a_ n}H_ n\) positive and \({\mathcal H}\) separable. Assuming (I), one calls a number \(\lambda_ 0\in {\mathbb{C}}\) an eigenvalue of \(M(\lambda)\) if \(M(\lambda)^{-1}\) has a pole at \(\lambda_ 0\). One can associate with \(M(\lambda)\) an operator \(\hat M\equiv {\mathcal L}({\mathcal H}^{n+2})\) such that the eigenvalues and (generalized) eigenvectors of \(M(\lambda)\) are directly related to those of \(I-\lambda\hat M;\) here \[ \hat M=\left[\begin{matrix} A_ 0 & H_ 0 & {1\over a_ 1}H_ 1 & \ldots & {1\over a_ n}H_ n \\ P_ 0 & 0 & 0 & & 0 \\ P_ 1 & 0 & {1\over a_ 1}P_ 1 \\ \vdots & & & \ddots \\ P_ n & 0 & 0 & \ldots & {1\over a_ n}P_ n \end{matrix}\right], \] where \(P_ k\) denotes the orthogonal projection along the kernel \(N(H_ k)\). The paper derives a Weinstein-Aronszajn formula for \(M(\lambda)\). Next, assuming (II) to hold one considers the space \({\mathcal Z}={\mathcal H}^{n+2}/N(\hat K)\), \(\hat K=diag(I,H_ 0,\frac{1}{a_ 1}H_ 1,...,\frac{1}{a_ n}H_ n),\) selfadjoint and positive, and the inner product \(([f],[g])=(\hat Kf,g).\) In the completion \(\bar {\mathcal Z}\) of \({\mathcal Z}\) with respect to this inner product one considers the induced operator \(\overline{[\hat M]}\) which is selfadjoint and compact; the eigenvalues of M(\(\lambda\)) coincide with those of \([I]-\lambda\overline{[\hat M]}\). Applying the Rayleigh-Ritz procedure to \(\overline{[\hat M]}\) one obtains upper bounds for the positive eigenvalues of \(M(\lambda)\), i.e., lower bounds for the positive eigenvalues of the operator \(\overline{[\hat M]}\). Next, upper bounds for the eigenvalues of \(\overline{[\hat M]}\) are found by means of the so- called method of intermediate problems. The procedure is quite technical, and the results are less satisfactory than those for the lower bounds. Two examples, taken from eigenvalue problems for ordinary differential equations where the eigenvalue parameter occurs polynomially in the boundary conditions, are used for explaining the methods. The final section of the paper deals with the special case where, assuming (II), the (nonnegative) eigenvalues and an orthonormal system of eigenvectors of A are known.