On the uniqueness of inverse spectral problems associated with incomplete spectral data (Q1754615): Difference between revisions

Consider the equation \[ -y'' + q(x) y = \lambda y \] and the boundary conditions \[ y'(0) = h y(0)\text{ and }y'(a) = f(\lambda) y(a) \] for some \(a \in (0,1]\). In the case of constant \(f\), a result of \textit{V. A. Marchenko} [Dokl. Akad. Nauk SSSR, n. Ser. 72, 457--460 (1950; Zbl 0040.34301)] states that the eigenvalues \(\{ \lambda_i \}_{i=0}^{\infty}\) and appropriately defined norming constants \(\{ \alpha_i \}_{i=0}^{\infty}\) uniquely determine all the coefficients of this boundary value problem, and all this data are actually needed. Subsequent research by various authors shows that the same situation occurs when one or both of the boundary conditions depend rationally on the eigenvalue parameter. A closely related question of which part of the above spectral data can be replaced by the partial knowledge of the coefficients is still an active area of current research. In particular, if \(f\) is a given constant, the identity \[ f = \frac{1}{\alpha_0} - \frac{1}{\pi} + \sum_{i=1}^{\infty} \left( \frac{1}{\alpha_i} - \frac{2}{\pi} \right), \] which is an easy consequence of the Gelfand-Levitan-Marchenko equation (cf., e.g., [\textit{B. M. Levitan}, Inverse Sturm-Liouville problems. Utrecht: VNU Science Press (1987; Zbl 0749.34001), Section 2.10]), shows that one of the norming constants can be omitted, but all eigenvalues must be specified. When \(f\) is a bounded (respectively, unbounded) rational Herglotz--Nevanlinna function with \(d\) poles, one can similarly deduce that \(2d+1\) (respectively, \(2d+2\)) elements (i.e., as many as the number of coefficients of \(f\)) can be omitted from the spectral data, since in this case we have \(2d\) (respectively, \(2d+1\)) more identities between the elements of the spectral data and the coefficients of \(f\) (about half of these identities can be found in the reviewer's paper [``On two-spectra inverse problems'', Preprint, \url{arXiv:1803.02567}] and the remaining ones can be obtained from Parseval's formula by following the methods of his older paper [Proc. Inst. Math. Mech., Natl. Acad. Sci. Azerb. 25, 35--40 (2006; Zbl 1220.34041)]). The main theorem of the paper under review tries to generalize these results to meromorphic Herglotz-Nevanlinna functions with positive poles \(\nu_i\) satisfying \[ \sum_{i \geq 0} \frac{1}{\nu_i^\rho} < \infty \] for some \(\rho \in (0,1)\) and show that the amount of spectral data that can be omitted depends on the growth properties of the poles of \(f\) (the details are somewhat technical to state here). The authors' motivation for considering this class of functions comes from the fact that it includes in particular \(m\)-functions of problems on \([a,1]\) (after a suitable shift in the spectral parameter). The proof of the main theorem is based on the method of \textit{F. Gesztesy} and \textit{B. Simon} [Trans. Am. Math. Soc. 352, No. 6, 2765--2787 (2000; Zbl 0948.34060)]. Reviewer's remarks: Unfortunately, the poor English, the large number of misprints and other inaccuracies in this paper make it very difficult to read. For example, one can only guess what the authors mean by (3.16) if the left-hand side is always an integer, but the right-hand side is not; or when they write ``\(n_S(t) = 0\) for \(t \in [0,1)\)''. The main theorem is certainly not true as stated in the case when \(f\) is bounded and has a finite number of poles, with \(S_0 = S = \sigma(L) \setminus \{ \lambda_0 \}\) and \(A(t) \equiv 0 \equiv B(t)\) being an obvious counterexample (see the above discussion). On the other hand, if the authors consider only unbounded rational Herglotz--Nevanlinna functions, as suggested by the sentence containing (2.3), then why does (2.1) contain two cases?