Nevanlinna-Pick meromorphic interpolation: The degenerate case and minimal norm solutions (Q1018149)

Let \(\mathcal S\) stand for the Schur class of analytic functions mapping the unit disk \(\mathbb D\) into its closure \(\overline{\mathbb D}\) and let \(\mathcal B_k\) be the set of finite Blaschke products of degree \(k\). Denote by \(\mathcal S_k\) the generalized Schur class of meromorphic functions \(f(z)\) having the Kreǐn-Langer representation, \(f(z)=s(z)/b(z)\), where \(s \in \mathcal S\) and \(b\in\mathcal B_k\) do not have common zeros. Via nontangential boundary limits, \(S_k\)-functions can be identified with the functions from the unit ball of \(L^\infty(\mathbb T)\) which admit a meromorphic continuation inside the unit disk with total pole multiplicity equal \(k\). Also \(\mathcal S_k\)-functions can be characterized as meromorphic functions \(f\) on \(\mathbb D\) with the associated kernel \(\frac{1-f(z)\overline{f(\zeta)}}{1-z\overline\zeta}\) having \(k\) negative squares on \(\rho(f)\), the domain of analyticity of \(f\). Equivalently, all Schwarz-Pick matrices (which are clearly Hermitian) \[ P_n(f;z_1,\dots,z_n):=\left[\frac{1-f(z_i)\overline{f(z_j)}}{1-z_i{\overline z}_j}\right]_{i,j=1}^n,\quad z_1,\dots,z_n\in\rho(f), \] have at most \(k\) negative eigenvalues (counted with multiplicities), and at least one such matrix has exactly \(k\) negative eigenvalues. Denote by \(\pi(X)\), \(\nu(X)\) and \(\delta(X)\) respectively the numbers of positive, negative and zero eigenvalues, counted with multiplicities, of a Hermitian matrix \(X\). The author is mostly concerned about the Nevanlinna-Pick interpolation problem: {\(\mathbf N\mathbf P_k\):} Given \(n\) distinct points \(z_1,\dots,z_n\in\mathbb D\), complex numbers \(f_1,\dots,f_n\) and an integer \(k\geqslant 0\), find all functions \(f\in\mathcal S_k\) (if exist) which are analytic at \(z_1,\dots, z_n\) and satisfy \[ f(z_i) =f_i,\quad \text{for}\quad i =1,\dots,n.\tag{1} \] As in the classical case \(k = 0\), necessary and sufficient conditions for the problem \(\mathbf N\mathbf P_k\) to have a solution can be given in terms of the Pick matrix \[ P:=\left[\frac{1-f_i{\overline f}_j}{1-z_i{\overline z}_j}\right]_{i,j=1}^n\tag{2} \] of the problem. These conditions are contained in Theorem 1 below. Recall that a Hermitian matrix \(P\) of rank \(d\) is said to be saturated if every \(d\times d\) principal submatrix of \(P\) is invertible. Theorem 1. Let \(P\) be the Pick matrix of the problem \(\mathbf N\mathbf P_k\). Then: (1) The problem has infinitely many solutions if and only if \(k\geqslant\nu(P) +\delta(P)\). (2) The problem has a unique solution if and only if \(k =\nu(P)\), \(\delta(P) > 0\), and \(P\) is saturated. (3) Otherwise, the problem has no solutions. The next theorem gives a simple test to verify whether or not a matrix \(P\) is saturated. Theorem 2. Let \(P\in\mathbb C^{n\times n}\) be of the form (2) and let \(d:=\text{rank}\,P <n\). Then \(P\) is saturated if and only if at least one \((d + 1)\times (d + 1)\) principal submatrix \(\widetilde P\) of \(P\) with \(\text{rank}\,\widetilde P = d\) is saturated. Finally, the author discusses the existence of solutions of the problem \(\mathbf N\mathbf P_k\) of the minimal possible \(L^\infty\)-norm. Let \(H_k^\infty\) be the set of all functions \(f\) having the Kreǐn-Langer representation, where \(s \in H^\infty\) and \(b \in \mathcal B_k\) may have common zeros. From this definition it follows that \(S_k = (H_k^\infty \setminus H_{k-1}^\infty)\cap\mathcal BL^\infty\), where \(\mathcal BL^\infty\) denotes the unit ball of \(L^\infty(\mathbb T)\). Let \(\mathbf S:=\{g: g(z_i)=f_i\quad \text{for}\quad i =1,\dots,n\}\) be the set of all functions \(g\) satisfying the interpolation conditions (1), let \[ P^{(\lambda)}:=\left[\frac{\lambda^2-f_i{\overline f}_j}{1-z_i{\overline z}_j}\right]_{i,j=1}^n \] and let \(\lambda_0\geqslant \lambda_1\geqslant \cdots\geqslant \lambda_m > 0\) be all positive solutions of the equation \(\text{det}\, P^{(\lambda)}\!=\!0\). Then \[ \mu_k:=\inf_{g\in{\mathbf S}\cap H_k^\infty}\| g\|_\infty= \inf_{g\in{\mathbf S}\cap (H_k^\infty\setminus H_{k-1}^\infty)}\| g\|_\infty= \begin{cases} \lambda_k&\text{if \(k\leqslant m\)},\\ 0 &\text{if \(k> m\)}. \end{cases} \] The contribution here is a reasonably simple criterion for the existence of a function \(g_{k,\min} \in{\mathbf S}\cap H_k^\infty\) for \(k\leqslant m\) such that \(\| g_{k,\min}\|_\infty =\mu_k\). The case \(k >m\) is simple: the function \(g\equiv 0\) (the only function in \(H_k^\infty\) with \(\| g\| =\mu_k = 0\)) belongs to \(\mathbf S\) if and only if all interpolation conditions are homogeneous: \(f_1 =\dots = f_n = 0\) (in which case the equation \(\text{det}\, P^{(\lambda)}= 0\) has no positive solutions). Theorem 3. Let \(0\leqslant k\leqslant m\). There exists a (unique) function \(g\in\mathbf S\cap H_k^\infty\) with \(\| g\|_\infty =\mu_k =\lambda_k\) if and only if the matrix \(P^{(\lambda)}\) is saturated. This extremal function belongs to \(H_k^\infty\setminus H_{k-1}^\infty\) if and only if either \(k = 0\) or \(\lambda_k <\lambda_{k-1}\).