Some Remarks on the Cone of Completely Positive Maps between Von Neumann Algebras (Q4340950)

\(A\)-\(N\) bimodules, taken to be Hilbert spaces with left action by \(A\) and a normal right action by \(N\), are used to study \(CP(A,N)\), the space of completely positive maps with topology of pointwise weak* convergence, from a \(C^*\)-algebra \(A\) to a von Neumann algebra \(N\). Every \(\phi \in CP(A,N)\) defines an \(A\)-\(N\) bimodule \(H_{\phi}\). The bimodules generalise the usual representation theory of \(C^*\)-algebras as they correspond to certain representations of a \(C^*\)-completion of the algebraic tensor product of \(A\) and \(N\). \(L^{2}(N)\) denotes the standard Hilbert space for \(N\) with its structure as a \(N\)-\(N\) bimodule. A completely positive mapping \(\phi\) from \(A\) to \(N\), is said to be \(H\)-elementary if, for any \(x \in A\) it is a finite sum of mappings \(x \to T^{\ast}xT\), where the \(T\) are bounded linear maps \(L^{2}(N) \to H_{\phi}\) intertwining right actions of \(N\) (identified with left actions of \(N^{0}\), the opposite algebra of \(N\)). Let \(X_\phi\) denote the Hilbert \(N\)-module, assumed to be self-dual [see \textit{W. L. Paschke}, Trans. Am. Math. Soc. 182, 443-468 (1973; Zbl 0265.46056)] defined by \(\phi \in CP(A,N)\); the representation \(\pi_{\phi}\) of \(A\) as \(N\)-valued operators on \(X_{\phi}\) is defined by \(\pi_{\pi}(a \otimes n) = xa\otimes n\). When \(A\) is a von Neumann algebra \(M\) and \(\phi\) is normal, \((\pi,X)\) is called a \(M\)-\(N\) bimodule or a correspondence (in the sense of being a morphism, as defined by \textit{A. Connes} [`Non-commutative geometry', Academic Press (1994; Zbl 0818.46076)] and the Hilbert space is denoted by \(H(X)\) or \(H_{\phi}\); there are commuting normal representations \(\rho_{M}\) and \(\rho_{N^{0}}\) of \(M\) and \(N^0\) on \(H_\phi\). If both \(M\) and \(N\) have copies of a von Neumann algebra \(Z\) in their centres, \(\phi\in CP(M,N)\) is called weakly \(Z\)-nuclear if it is the limit of completely positive mappings factorising through matrix algebras over \(Z\), i.e., if \(\phi\) splits into \(\phi_{1}: A \to M_{n}(Z)\) and \(\phi_{2}: M_{n}(Z) \to N\). The main results concern density of cones of `elementary' mappings in \(CP(A,N)\) and seem to be the following: If \(Z\) is a von Neumann subalgebra of \({\mathcal {Z}}(M)\) then \(M\) is injective if and only if every \(Z\)-linear completely positive endomorphism is weakly \(Z\)-nuclear; for a correspondence with faithful \(\rho_{M}\) and \(\rho_{N^{0}}\), \(M\) or \(N\) injective and either \(Z\) equals either \( {\mathcal{Z}}(M)\) or \({\mathcal{Z}}(N)\), every weakly \(Z\)-nuclear element of \(CP(M,N)\) is the pointwise weak\(^{\ast}\) limit of a net of \(H\)-elementary mappings; if \(M\) is injective then every \({\mathcal{Z}}(M)\)-linear completely positive endomorphism is approximable in the weak* topology by \(H\)-elementary mappings, \(H = L^{2}(M)\); a \(\sigma\)-finite von Neumann algebra \(M\) is of type I if and only if every normal \({\mathcal{Z}}(M)\)-linear completely positive endomorphism has the form \(x \to \sum_{i \geq 1}{a_{i}^{\ast} x a_{i}}\) where the \(a_{i}\) in \(M\) and the series is weak* convergent.