Analytic approximation of matrix functions in \(L^p\) (Q1028408)

The classical problem of analytic approximations of the functions on the unit circle \(T\) of the complex plane \(\mathbb C\) is: for a given function \(\varphi \in L^\infty\), to find the best \(H^\infty\) approximant to \(\varphi\), i.e., to find a bounded analytic \(\psi\) in the unit disk \(D\) such that \(\| \varphi - \psi \|_{L^\infty (T)}= \text{dist}_{L^\infty} (\varphi ,H^\infty )\). Under certain assumptions, the best approximation is unique. The above described approximation problem is closely related to Hankel operators on the Hardy classes \(H^2\). For a function \(\varphi \in L^\infty\), the Hankel operator \(H_\varphi : H^2 \to H^2_{-}\) (\(H^2_{-}\) is the orthogonal complement of \(H^2\) in \(L^2\)) is defined by \(H_\varphi f = P_- (\varphi f)\), where \(P_-\) is the orthogonal projection from \(L^2\) onto \(H^2_-\). Nehari proved that \(\| H_\varphi \|= \text{dist}_{L^\infty} (\varphi , H^\infty )\). This kind of problems has applications in control theory, but for the needs of control theory it is important to consider the case of matrix-valued functions. Let \(\Phi \in L^\infty (M_{m,n})\), that is, \(\Phi\) is a bounded function with values in the space \(M_{m,n}\) of \(m \times n\) matrices. The problem of best approximation is to find a bounded analytic matrix function \(Q\) of size \(m \times n\) such that \(\| \Phi - Q \|_{L^\infty (M_{m,n})}= \text{dist}_{L^\infty} (\Phi ,H^\infty (M_{m,n}))\), where \(H^\infty (M_{m,n})\) is the space of bounded analytic \(m \times n\) matrix functions. Here, for \(\Psi \in L^\infty (M_{m,n})\), we have set \(\| \Psi \| := \operatorname{ess}\sup_{z \in T} \| \Psi (z)\|_{M_{m,n}}\), where \(\| A \|_{M_{m,n}}\) denotes the operator norm of a matrix \(A \in M_{m,n}\) considered as an operator \(\mathbb C^n \to\mathbb C^m\). In the matrix case, we also have \(\| H_\Phi \|= \text{dist}_{L^\infty} (\Phi , H^\infty (M_{m,n}))\), where the Hankel operator \(H_\varphi : H^2(\mathbb C^n) \to H^2_-(\mathbb C^m)\) is defined as \(H_\Phi f = P_- (\Phi f)\). This time, \(P_-\) denotes the orthogonal projection \(L^2 (\mathbb C^m) \to H^2_-(\mathbb C^m)\). Unlike the scalar case, generically \(\Phi\) has infinitely many best approximants. In this paper, the authors consider the problem of approximation of matrix functions of the class \(L^p\) (\(2 \leq p < \infty\)) on \(T\) by matrix functions analytic in the unit disk in the norm of \(L^p\). For an \(m \times n\) matrix function \(\Phi\) in \(L^p\), they consider the Hankel operator \(H_\Phi : H^q(\mathbb C^n) \to H^2_-(\mathbb C^m)\), where \(q\) is such that \(1/p + 1/q = 1/2\). It turns out that the space of \(m \times n\) matrix functions \(\Phi \in L^p(M_{m,n}) \setminus H^p(M_{m,n})\) splits into two subclasses: the set of ``respectable'' matrix functions (those \(\Phi\) satisfying \(\| H_\varphi \|= \text{dist}_{L^p} (\Phi , H^\infty (M_{m,n}))\) and its complementary set of ``weird'' functions. For weird matrix functions, to obtain the distance formula, the authors consider Hankel operators defined on spaces of matrix functions. They also describe the set of \(p\)-badly approximable matrix functions (functions \(\Phi \in L^p(M_{m,n})\) such that \(\| \Phi - \Psi \|_{L^p(M_{m,n})} \geq \| \Phi \|_{L^p(M_{m,n})}\) for any \(\Psi \in H^p(M_{m,n})\)) in terms of special factorizations and give all best analytic approximants in the norm of \(L^p\). Moreover, the authors introduce the notion of \(p\)-superoptimal approximation of a matrix function \(\Phi \in L^p (M_{m,n})\) (as an extension of the concept of superoptimal approximant of a matrix function \(\Phi \in L^\infty (M_{m,n})\) introduced by \textit{N.\,J.\thinspace Young} [J.~Oper.\ Theory 15, 239--265 (1986; Zbl 0608.47020)]) and prove the uniqueness of the \(p\)-superoptimal approximant for each rational matrix function.