Operator algebras of idempotents (Q5933451)

A uniform algebra \(A\) of complex-valued functions on a compact Hausdorff space \(X\) can be represented isometrically as an algebra of operators on a (1-dimensional) Hilbert space by choosing a faithful representation of \(C(X)\) and restricting it to A. If \(K\) is an ideal in \(A\), then the quotient \(A/K\) is a Banach algebra. The author is using the fact, not obvious, that it can be represented isometrically as an operator algebra on a Hilbert space and that, moreover, matrices over \(A/K\) can be so represented as well. The result above originates from \textit{D. Sarason}'s work on the Nevanlinna-Pick problem for \(A\) the disc algebra [Trans. Am. Math. Soc. 127, 179-203 (1967; Zbl 0145.39303)]. Indeed, pick a finite set of points, the ``nodes'' on the disc, possibly with multiplicity, and let \(K\) denote the ideal of all functions that vanish on the nodes. Then the quotient \(A/K\) is a finite-dimensional Banach algebra. Using a quotient Hilbert space of the Hardy space \(H^2\), the quotient algebra can be realised as a subalgebra of the \(n \times n\) matrices over \(A\). Pick a finite subset \(x_{1},\dots,x_{k}\), in a compact Hausdorff space \(X\) and let \(K\) be the ideal of all functions on \(X\) that vanish on these nodes. The classical Nevanlinna-Pick interpolation problem, for \(X\) the unit disc, is to find a \(H^{\infty}\) function which takes prescribed values on the nodes. Pick's criterion for such a function to exist is that a certain matrix built from the data is positive-definite; hence the subsequent use of positive-definite \({k \times k}\) matrices \(P\) with non-vanishing entries. The author generalises the interpolation problem to \(M_{n}(A)\), with the sup norm, disregarding interpolating functions and being concerned only with interpolation sets, i.e., sets of k-tuples of \(n \times n\) matrices \({\mathcal{D}}^{n}(A,x_{1},\dots,x_{k}) = \{(F(x_{1}),\dots,F(x_{k}): F \in M_{n}(A), \|F\|\leq 1 \}\). One can think of \({\mathcal{D}}^{n}\) as representing the closed unit ball in \(M_{n}(A)\). The operator algebra \({\mathcal{A}}\) is the quotient of \(A\) by the ideal of functions vanishing at the nodes and the author identifies \(F \in M_{n}(A)\) with its canonical image in \(M_{n}({\mathcal{A}})\). Bounded operators on a Hilbert space \({\mathcal{H}}\) satisfying \(E_{i}E{j} = \delta_{ij}\) and \(E_{i}+.. +E_{k} = 1\) generate a \(k\)-idempotent operator algebra \({\mathcal{A}}\) on \({\mathcal{H}}\); the concrete operator system \({\mathcal{A}}^{*}{\mathcal{A}} = \text{span} \{ E_{i}^{*}E_{j}: i,j = 1,\dots,k \}\) is not invariant under completely isometric representations of \({\mathcal{A}}\). For a positive matrix \(P\) with nonzero entries, the author defines \({\mathcal{A}}(P)\) to be the \(\text{span} \{ E_{1},\dots,E_{k} \}\), where \(E_{i}={P^{1 \over 2} E_{ii} P^{-{1 \over 2}}}\) and the \(E_{ij}\) denote the canonical matrix units. The author's matricial Schur ideals \({\mathcal{S}}\), which the reviewer will call matricial semi-ideals, are sequences \(({\mathcal{S}}_{n})\) of sets in \(M_{k}(M_{n})^{+}\) such that if \(Q,R \in S_{n}\) then \(Q + R \in S_{n}\), and if \( B_{1},\ldots,B_{k}\) are \(m \times n\) matrices then \(\sum B_{i}Q_{ij}B_{j}^{*} \in S_{m}\). For a \(k\)-idempotent algebra \({\mathcal{A}}\), the definition of \({\mathcal{S}}_{n}({\mathcal{A}}^{*}{\mathcal{A}})\) is apparently meant to be the set of \((\Phi(E_{i}^{*}E_{j}))_{i,j = 1,\dots,k} \) where \(\Phi\) covers all completely positive mappings \({\mathcal{A}}^{*}{\mathcal{A}} \to M_{n}\). He proves that \({\mathcal{S}}({\mathcal{A}}^{*}{\mathcal{A}})\) is a matricial semi-ideal. The Schur terminology is used since \({\mathcal{S}}_{1}({\mathcal{A}}(P)^{*}{\mathcal{A}}(P))\), where \(P\) denotes a \(k \times k\) Pick matrix, is spanned by Schur products of scalar-entry matrices. He gives a very technical definition of the matricial semi-ideal determined by a \(k\)-idempotent algebra \({\mathcal{A}}\), denoted \({\mathcal{S}}({\mathcal{A}})\), seemingly unrelated to \({\mathcal{S}}_{n}({\mathcal{A}}^{*}{\mathcal{A}}))\), in terms of the \({\mathcal{D}}_{n}({\mathcal{A}})\). The definition also involves positivity of the tensor product of certain matrices in \(M_{k}(M_{n})\). There is already the definition of the interpolation set \({\mathcal{D}}^{n}\) in Section 1, but what is more likely meant is a closely related definition of \({\mathcal{D}}_{n}\) introduced in the proof of Theorem 4.2. Section 3 attempts, given a matricial semi-ideal \({\mathcal{S}}\), to construct from it a \(k\)-idempotent algebra which the author would denote by \({\mathcal{A}}({\mathcal{S}})\), and so to get some kind of duality between matricial Schur ideals and \(k\)-idempotent algebras. It is assumed that \({\mathcal{S}}\) is ``non-trivial'' and ``bounded''. The definition of boundedness ensures that the \(E_{i}\) which he constructs in \({\mathcal{A}}({\mathcal{S}})\) are bounded in the matrix norm. The definition of non-triviality is problematic, since in order to construct \({\mathcal{A}}({\mathcal{S}})\), the existence of \({\mathcal{A}}\) and its idempotents is assumed a priori. Nevertheless, it seems that the essential requisite, free of \({\mathcal{A}}\), for sufficiency for existence is that the invertible elements in \({\mathcal{S}}_{n}\) are dense in \({\mathcal{S}}_{n}\). The author then constructs the required \(k\)-idempotents \(E_{i}\) as \(\sum_{n}{\sum{\oplus} Q^{1 \over 2} (I_{n} \otimes E_{ii}) Q^{-{1 \over 2}}}\), where the second summation is over invertible matrices \(Q \in {\mathcal{S}}_{n}\). When \({\mathcal{A}}({\mathcal{S}})\) exists, then \({\mathcal{S}}({\mathcal{A}}({\mathcal{S}})^{*}{\mathcal{A}}({\mathcal{S}})) = {\mathcal{S}}\). He defines \({\mathcal{S}}\) to be affiliated with \({\mathcal{A}}\), if \({\mathcal{A}}({\mathcal{S}})\) and \({\mathcal{A}}\) are completely isometric. A \(k\)-idempotent algebra is algebraically isomorphic to \({\mathbb{C}}^{k}\) with pointwise multiplication and it can be determined by an assignment of an operator algebra matrix norm on \({\mathbb{C}}^{k}\). In Section 4, the author identifies the \(k\)-idempotent algebras with operator algebra norms on \({\mathbb{C}}^{k}\). The author's duality beteen matricial semi-ideals and \(k\)-idempotent algebras is meant to be analogous to the polar-bipolar duality for locally convex topological spaces. For sets of Banach-space norms on \({\mathbb{C}}^{k}\), the supremum and infimum are well-defined. The unit ball of the largest norm smaller than both operator-algebra norms is the polar of the intersection of the polars of the two balls. The smallest unit ball larger than two unit balls is simpler; it is the larger of the two. Analogously, given \(k\)-idempotent algebras \({\mathcal{A}}\) and \({\mathcal{B}}\) the author uses direct sums of the idempotents to construct a \(k\)-idempotent algebra, which he denotes by \(A \vee B\), which endows \({\mathbb{C}}^{k}\) with the larger of the two norms. He goes on to construct a dual \(k\)-idempotent algebra \(A \wedge B\) which will have the smallest ``unit ball'' such that \({\mathcal{D}}_{n}({\mathcal{A}} \wedge {\mathcal{B}}) \supset {\mathcal{D}}_{n}({\mathcal{A}}) \cup {\mathcal{D}}_{n}({\mathcal{B}})\), but relies on duality and the somewhat unclear Proposition 3.4 to get \({\mathcal{S}}_{n}({\mathcal{A}} \wedge {\mathcal{B}}) \subset {\mathcal{S}}_{n}({\mathcal{A}}) \cap {\mathcal{S}}_{n}({\mathcal{B}})\). Most of the remaining part of Section 4 consists of the construction of \({\mathcal{A}} \wedge {\mathcal{B}}\) (or a wedge product of a finite number of \(k\)-idempotent algebras) for which \({\mathcal{S}}_{n}({\mathcal{A}} \wedge {\mathcal{B}}) = {\mathcal{S}}_{n}({\mathcal{A}}) \cap {\mathcal{S}}_{n}({\mathcal{B}})\). In fact, \({\mathcal{A}} \wedge {\mathcal{B}}\) is the operator algebra generated by a pair of (completely contractive) images of \({\mathcal{A}}\) and \({\mathcal{B}}\) in the amalgamated free-product operator algebra \({\mathcal{A}} * {\mathcal{B}}\) described in \textit{D. P. Blecher} and \textit{V. I. Paulsen} [Proc. Am. Math. Soc. 112, 839-850 (1991; Zbl 0767.46037)]. The universal property for the free product is that, given a pair \(\pi_{i}\) of representations \({\mathcal{A}}\) and \({\mathcal{B}}\), respectively, on a Hilbert space \({\mathcal{H}}\), then there is a (completely contractive) operator algebra homomorphism \(\rho: {\mathcal{A}}*{\mathcal{B}} \to {\mathcal{L}}({\mathcal{H}})\) which restricts to the \(\pi_{i}\) on the component \(k\)-idempotent algebras (identifying elements of these with their images in the free product). The author then considers ways to factorize a matrix \(z\) over \({\mathcal{A}} * {\mathcal{B}}\) as a product of images of matrices over \({\mathcal{A}}\) and \({\mathcal{B}}\), respectively, made up of words of the form \(a_{1}b_{1} \cdots a_{m}b_{m}\), \(m\) arbitrary, with \(a_{k} \in {\mathcal{A}}, b_{k} \in {\mathcal{B}}\) and also allowing words beginning with a \(b\) rather than an \(a\) and/or ending with an \(a\) rather than with a \(b\). As in \textit{Blecher-Paulsen} [loc. cit.], the author considers all factorisations of matrices \(z\) with entries in the free product and defines a seminorm \(\|z\|\) to be the infimum of products of the norms of all the factors where the infimum is taken over all ways of factorising \(z\) into products of matrices over \({\mathcal{A}}\) and \({\mathcal{B}}\), following the scheme for the words of the free product. This infimum seminorm will equal the maximal operator seminorm. He defines \({\mathcal{A}} \wedge {\mathcal{B}}\) to be \(({\mathcal{A}} * {\mathcal{B}}) / {\mathcal{I}}\), where \({\mathcal{I}}\) denotes the ideal of elements of maximal operator seminorm \(0\). Using the \({\mathcal{D}}_{n}\), he shows that \({\mathcal{S}}_{n}({\mathcal{A}} \wedge {\mathcal{B}}) = {\mathcal{S}}_{n}({\mathcal{A}}) \cap {\mathcal{S}}_{n}({\mathcal{B}})\). The present paper is a continuation of \textit{V. I. Paulsen} [Integral Equations Oper. Theory 41, 38-62 (2001; Zbl 0997.47011)].