Semicontinuity and closed faces of \(C^*\)-algebras (Q1679447)

This interesting paper, written by one of the major contributors to investigations into the structure of \(C^*\)-algebras, is dedicated to the memory of another, Uffe Haagerup. Let \(A\) be a \(C^*\)-algebra, with self-adjoint part the JB-algebra \(A_{\text{sa}}\), having unit balls \(A_1\) and \(A_{\text{sa},1}\), let \(A^*\) and \(A^*_{\text{sa}}\) be their dual spaces with corresponding unit balls \(A^*_1\) and \(A^*_{\text{sa},1}\), and let \(A^{**}\) and \(A^{**}_{\text{sa}}\) be their bidual \(W^*\)-algebra and bidual JBW-algebra, respectively, with corresponding unit balls \(A^{**}_1\) and \(A^{**}_{\text{sa},1}\). Let \(A^*_{\text{sa},1, +}\) be the set of positive linear functionals on \(A\) of norm less than or equal to one, the subset \(S(A)\) of which, consisting of those elements of norm one, is the state space of \(A\) and the normal state space of the \(W^*\)-algebra \(A^{**}\). This paper, like many others on the properties of \(C^*\)-algebras, uses two fundamental theorems, the Gelfand representation for commutative \(C^*\)-algebras, that suggests results that may hold for non-commutative \(C^*\)-algebras, and the Kadison representation that shows that a \(C^*\)-algebra \(A\) can be identified with the complex Banach space of complex-valued weak\(^*\)-continuous bounded affine functions on the weak\(^*\)-compact convex set \(\text{conv}(S(A)\cup\{0\})\) that vanish at zero, whilst its bidual \(A^{**}\) can be identified with the complex Banach space of all bounded affine functions on \(\text{conv}(S(A)\cup\{0\})\) that vanish at zero, thereby connecting \(C^*\)-algebras with Choquet theory. A property that is much used in Choquet theory is that of lower-semicontinuity of bounded real affine functions on compact convex sets. \textit{C. A. Akemann} and \textit{G. K. Pedersen} [Duke Math. J. 40, 785--795 (1973; Zbl 0287.46073)] showed that three of the possible definitions of lower-semicontinuity of bounded real affine functions that vanish at zero on the compact convex set \(\text{conv}(S\cup\{0\})\) in the self-adjoint part of the dual \(A^*\) of the \(C^*\)-algebra \(A\) are equivalent. Since that date, their results have been used by many authors to investigate other properties of \(C^*\)-algebras. The state spaces of \(C^*\)-algebras and normal state spaces of \(W^*\)-algebras are sometimes used as models for those of statistical physical systems and it is usually possible to give `physical' interpretations of properties of the algebras. In this case, an element \(p\) of the complete orthomodular lattice \({\mathcal P}(A^{**})\) of self-adjoint idempotents (projections) in the \(W^*\)-algebra \(A^{**}\) is associated with a filter \(x\mapsto pxp\) on \(A^*\) the range of which represents a subsystem of the original system. Using the notation established above, let \(p\) be an element of \({\mathcal P}(A^{**})\) and let \[ \{p\}_{\prime}= \{x:x\in A^*_1,\, x(p)= 1\}. \] Then, \(\{p\}_{\prime}\) is a norm-closed face of the normal state space \(\{1\}_{\prime}\) of \(A^{**}\), where \(1\) denotes the identity in \(A^{**}\), and the mapping \(p\mapsto\{p\}_\prime\) is an order isomorphism from \({\mathcal P}(A^{**})\) onto the set \({\mathcal F}_n(S)\) of norm-closed faces of \(S\) ordered by set inclusion. A projection \(p\) is said to be compact if \(\{p\}_{\prime}\) is weak\(^*\)-closed and is said to be closed if the convex hull \(\text{conv}(\{p\}_{\prime}\cup\{0\})\) is weak\(^*\)-closed. The mapping \(p\mapsto\{p\}_{\prime}\) is an order isomorphism from the atomic lattice of compact projections in \(A^{**}\) onto that of the set \({\mathcal F}_{W^*} (S)\) of weak\(^*\)-closed faces of \(S\). Similarly, the mapping \(p\mapsto\text{conv}(\{p\}_\prime\cup\{0\})\) is an order isomorphism from the set of closed projections in \(A^{**}\) onto the set of weak\(^*\)-closed faces of \(\text{conv}(S\cup\{0\})\). When \(p\) is a fixed closed projection in the \(W^*\)-algebra \(A^{**}\), the JBW-algebra \(pA^{**}_{\text{sa}}p\) may be identified with the space of real-valued bounded affine functions on \(\text{conv}(\{p\}_\prime\cup\{0\})\) vanishing at zero and \(pA_{\text{sa}}p\) may be identified with the space of real-valued weak\(^*\)-continuous affine functions on \(\text{conv}(\{p\}_\prime\cup\{0\})\) vanishing at zero. There are then at least three possible definitions of the lower-semicontinuity of an element \(a\) of the JBW-algebra \(pA^{**}_{\text{sa}}p\). It can be seen that when \(p\) is closed, \(p\) lies in \(pA_{\text{sa}}p\) if and only if \(p\) is compact. It follows that if \(p\) is compact, then all three types of local lower-semicontinuity coincide, but when \(p\) is closed but not compact, at least two of the types differ. Using sophisticated techniques, the author reveals some of the local properties of the \(C^*\)-algebra \(A\) related to a closed projection \(p\), including useful tools for further investigations, such as interpolation theorems for the various types of local lower-semicontinuity, conditions under which for an element \(h\) of \(pA^{**}_{\text{sa}}p\), both \(h\) and \(-h\) share the same type of local lower-semicontinuity, and relations between such elements \(h\) and the local multiplier and quasi-multiplier Jordan algebras \(pM(A)_{\text{sa}}p\) and \(pQM(A)_{\text{sa}}p\). He himself goes on to use his results to give further characterizations of operator convexity and strong operator convexity of real-valued functions on real intervals. The paper contains several useful examples, used to refute possible conjectures, and concludes with some ideas for further research.