Distinguished varieties (Q2495343): Difference between revisions

The authors consider a special class of bordered (algebraic) varieties that are contained in the bidisk \(\mathbb D^2\) in \(\mathbb C^2\). A non-empty set \(V\subset\mathbb C^2\) is a distinguished variety if there is a polynomial \(p \in\mathbb C[z,w]\) such that \(V =\{(z,w)\in \mathbb D^2:p(z,w)=0\}\) and such that \({\overline V}\cap\partial (\mathbb D^2) ={\overline V}\cap(\partial \mathbb D)^2\). This condition means that the variety exists in the bidisk through the distinguished boundary of the bidisk, the torus. We use \(\partial V\) to denote the set given by this condition: topologically, it is the boundary of \(V\) within \(Z_p\), the zero set of \(p\), rather than in all of \(\mathbb C^2\). It is always assumed that \(p\) is chosen to be minimal, i.e., so that no irreducible component of \(Z_p\) is disjoint from \(\mathbb D^2\) and so that \(p\) has no repeated irreducible factors. Why should one single out distinguished varieties from other bordered varieties? One of the most important results in operator theory is T.~Ando's inequality: if \(T_1\) and \(T_2\) are commuting operators, and both of them are of norm 1 or less, then for any polynomial \(p\) in two variables the inequality \(\| p(T_1,T_2)\|\leqslant\| p\|_{\mathbb D^2}\) holds. The first main result of the paper is that if \(T_1\) and \(T_2\) are matrices then the Ando's inequality can be improved to \(\| p(T_1,T_2)\|\leqslant\| p\|_V,\) where \(V\) is some distinguished variety depending on \(T_1\) and \(T_2\). Indeed, in the proof of this inequality, the authors construct co-isometric extensions of the matrices that naturally live on this distinguished variety. So when studying bivariable matrix theory, rather than operator theory, one is led inexorably to study distinguished varieties. Conversely, the authors show that all distinguished varieties can be represented as \(\{(z, w)\in\mathbb D^2 : {\text{det}}(\Psi(z) -wI) = 0\}\) for some analytic matrix-valued function \(\Psi\) on the disk that is unitary on \(\partial\mathbb D\). This shows that the study of distinguished varieties leads back to operator theory. Consider the natural notion of isomorphism of two distinguished varieties. A function \(\Phi\) is holomorphic on a set \(V \subset\mathbb C^2\) if, at every point \(\lambda \in V\), there is a non-empty ball \(B(\lambda,\varepsilon)\) centered at \(\lambda\) and an analytic mapping of two variables defined on \(B(\lambda,\varepsilon)\) that agrees with \(\Phi\) on \(B(\lambda,\varepsilon)\cap V\). Two distinguished varieties \(V_1\) and \(V_2\) are isomorphic if there is a function \(\Phi\) that is holomorphic on \(V_1\) and continuous on \(\overline V_1\) such that \(\Phi\) is a bijection from \(\overline V_1\) onto \(\overline V_2\) and such that \(\Phi^{-1}\) is holomorphic on \(V_2\). By the maximum modulus principle, \(\Phi\) must map the boundary of \(V_1\) onto the boundary of \(V_2\). It follows that \(\Phi = (\phi_1,\phi_2)\) is a pair of inner functions, i.e., a pair of holomorphic scalar-valued functions that each have modulus one on \(\partial V_1\). So studying isomorphism classes of distinguished varieties is closely connected to the rich structure of inner functions. The authors show that, under fairly general conditions, a pair of inner functions \((\phi_1,\phi_2)\) on a set \(X\) must map \(X\) into a distinguished variety. Another reason to study distinguished varieties comes from considering the Pick problem on the bidisk. This is the problem of deciding, given points \(\lambda_1,\dots, \lambda_N \in\mathbb D^2\), and values \(w_1, \dots,w_N \in\mathbb C\), whether there is a function in \(H^\infty(\mathbb D^2)\), the bounded analytic functions on \(\mathbb D^2\) that interpolates the data and is of norm at most 1. The problem is called extremal if there is an interpolating function of norm exactly 1, but not less. If an extremal Pick problem is given, the solution may or may not be unique. The second main result of the paper is that there is always a distinguished variety on which the solution is unique. One can think then of the Pick problem as having two parts: (a) Solve the problem on the distinguished variety where the solution is unique. (b) Parametrize all the extensions of the solution to the whole bidisk. The authors give a formula for problem (a). The extension problem (b) is non-trivial: unless the distinguished variety is isomorphic to a disk, there will always be some functions that cannot be extended to the whole bidisk without increasing the norm. Obviously a function arising from a Pick problem will be extendable, but what distinguishes such functions remains mysterious. If one starts with an inner function on \(V\) and wants to extend this to a rational inner function on \(\mathbb D^2\), there may be more than one extension. However, there is a restriction on the degree: if the variety is of rank \((n_1,n_2)\), i.e. there are generically \(n_1\) sheets above every first coordinate and \(n_2\) above every second coordinate, then any regular rational inner extension of degree \((d_1,d_2)\) must have \(d_1n_1 + d_2n_2\) equal to the number of zeroes that the original function had on \(V\). An admissible kernel \(K\) on a set \(\{\lambda_1, \dots, \lambda_N\} \subset\mathbb D^2\) is a positive definite \(N\times N\)-matrix such that \([(1-\lambda_i^r{\overline \lambda}_j^r)K_{ij}]\geqslant0\), \( r=1,2\). It is known that studying all the admissible kernels on a set is essential to understanding the Pick problem. We say that an \(N\)-point extremal Pick problem is minimal if none of the \((N-1)\)-point subproblems is extremal. The authors prove that if the uniqueness variety is not the whole bidisk, then it at least contains a distinguished variety running through the nodes: Let \(N\geqslant2\), and let \(\lambda_1,\dots, \lambda_N\) and \(w_1,\dots,w_N\) be the data for a minimal extremal Pick problem on the bidisk. The uniqueness variety \(\mathcal U\) contains a distinguished variety \(V\) that contains each of the nodes.