The Calderón problem for Hilbert couples (Q1774807)

The aim of the present paper is to characterize all exact interpolation spaces \(\mathcal A\) and \(\mathcal B\) with respect to arbitrary Hilbert couples \(\overline {\mathcal H}\) and \(\overline {\mathcal K}\). It is well-known [cf. the author, ``Interpolation of Hilbert spaces'', (Ph.D. Thesis, Uppsala) (2001) or \textit{G. Sparr}, Stud. Math. 62, 229--271 (1978; Zbl 0393.46029)] that \(\mathcal A\) and \(\mathcal B\) are exact interpolation spaces provided that they are \(K\)-monotone. In general, these two properties are not the same. However, if \(\mathcal A\) and \(\mathcal B\) are exact interpolation spaces with respect to regular Hilbert couples \(\overline {\mathcal H}\) and \(\overline {\mathcal K}\), then they are \(\sqrt 2-K\)-monotone [cf. the author, loc. cit.]. The author proves that there is the following improvement of these results: Spaces \(\mathcal A\) and \(\mathcal B\) are exact interpolation spaces with respect to regular Hilbert couples \(\overline {\mathcal H}\) and \(\overline {\mathcal K}\) if and only if \(\mathcal A\) and \(\mathcal B\) are exact \(K\)-monotone. (In other words, regular Hilbert couples are Calderón couples). This result is a consequence of the following assertion: Let \(\overline {\mathcal H}\) and \(\overline {\mathcal K}\) be regular Hilbert couples. Let \(f \in \mathcal H_0 + \mathcal H_1\) and \(g \in \mathcal K_0 + \mathcal K_1\) be such that \(K (t, g; \overline {\mathcal K}) \leq \rho^{-1} K (t, f; \overline {\mathcal H}), t > 0\), for some \(\rho > 1\). Then there exists a~linear operator \(T\: \overline {\mathcal H} \to \overline {\mathcal K}\), \(\| T\| _{\overline {\mathcal H} \to \overline {\mathcal K}} \leq 1\), satisfying \(T f = g\). To prove this assertion, the author reduces the problem to the diagonal case when \(\overline {\mathcal H} = \overline {\mathcal K}\) and then to the finite-dimensional case. Another important step in the proof is the fact that the quasi-order defined by the Peetre \(K\)-functional coincides with the quasi-order defined by the \(K_2\)-functional. Since the \(K_2\)-functional of the regular Hilbert couple can be conveniently calculated, this enables the author to construct the operator \(T\) mentioned above. In the finite-dimensional case, the author makes use of the result to give a short proof of Donoghue's theorem on interpolation functions and a new proof of Löwner's theorem on monotone matrix functions.