Maximality of positive operator-valued measures (Q2499023)

In this paper, \((\Omega, \Sigma)\) is a measurable space and \(ca_{+}(\Omega, \Sigma)\) is the set of all finite, positive, countably additive measures; for \( \mu \in ca_{+}(\Omega, \Sigma)\), \(ca_{+}(\Omega, \Sigma, \mu)= \{ \nu \in ca_{+}(\Omega, \Sigma): \nu \ll \mu \}\). \(\mathcal{H}\) is a separable Hilbert space, and \(L(\mathcal{H})\) is the Banach space of all linear bounded operators on \(\mathcal{H}\). A mapping \(M: \Sigma \to L(\mathcal{H})\) is a \(POV\) measure if \(P_{M}(\rho) \in ca_{+}(\Omega, \Sigma)\) for any density operator \(\rho\), where, for any \(\triangle \in \Sigma\), \(P_{M}(\rho)( \triangle)= \text{Trace}(\rho M(\triangle) )\); if \(M(\Omega)=I\), then it is called an \(OV\) probability measure; a spectral measure is an \(OV\) probability measure \(M\) for which \(M(\triangle)\) is an orthogonal projection operator for each \( \triangle \in \Sigma\). From the separability of \(\mathcal{H}\) it follows that there is a scalar probability measure \(\mu_{M}\) on \((\Omega, \Sigma)\), having the same sets of measure \(0\) as \(M\). If \(M\) is an \(OV\) probability measure, then \(P_{M}\) maps density operators into \(\text{Prob}(\Omega, \Sigma, \mu_{M})\). For a \(\triangle \in \Sigma\), \( \mathbb{I} (\triangle )\) will denote the operator of multiplication on \( L_{2}(\Omega, \Sigma, \mu_{M})\) by the indicator function of the set \(\triangle\). An \(L(\mathcal{H})\)-valued \(POV\) measure \(M\) on \((\Omega, \Sigma)\) is called non-degenerate if there exists a bounded linear operator \(V: \mathcal{H} \to L_{2}(\Omega, \Sigma, \mu_{M})\) such that \(M= V^{*} \mathbb{I}(\cdot) V\). For \(k=(1, 2)\), let \( M_{k}\) be an \(L(\mathcal{H})\)-valued \(POV\) measure on \((\Omega_{k}, \Sigma_{k})\) and \( \mu_{k}\) be a scalar-valued measure, having the same sets of measure \(0\) as \(M_{k}\). Then \(M_{2}\) is said to be dominating \(M_{1}\) (written as \(M_{2} \to M_{1}\)) if there is an affine function \(A: ca_{+}(\Omega_{2}, \Sigma_{2}, \mu_{2}) \to ca_{+}(\Omega_{1}, \Sigma_{1}, \mu_{1})\) such that \(P_{M_{1}}( \rho) =P_{M_{2}}(\rho) \) for every density operator \(\rho\) on \(\mathcal{H}\), and \(A(\text{Prob}(\Omega_{2}, \Sigma_{2}, \mu_{2})) \subset \text{Prob}(\Omega_{1}, \Sigma_{1}, \mu_{1})\). An \(L(\mathcal{H})\)-valued \(POV\) measure \(M\) on \((\Omega, \Sigma)\) is called maximal if, for any \(L(\mathcal{H})\)-valued \(POV\) measure \(N\) on some measurable space \((\Omega_{2}, \Sigma_{2})\) with \(N \to M\), we have \(M \to N\). In this paper, the author proves some representation theorems for \(POV\) measures \(M: \Sigma \to L(\mathcal{H})\). Some important results are: I. Let \(M\) be an \(L(\mathcal{H})\)-valued \(POV\) measure on \((\Omega, \Sigma)\). Then there is a complex Hilbert space \(\mathcal{K}\), an \(L(\mathcal{K})\)-valued spectral measure \(E\) on \((\Omega, \Sigma)\), with the same sets of measure \(0\) as \(M\), and a linear bounded operator \(V: \mathcal{H} \to \mathcal{K}\) such that \(M(\cdot)= V^{*} E(\cdot) V\). II. If \(M\) and \(N\) are two \(L(\mathcal{H})\)-valued \(POV\) measures on \((\Omega, \Sigma)\), with \(M\) non-degenerate and \( N(\triangle) \leq M(\triangle)\), for all \(\triangle \in \Sigma\), then there is a unique \(\phi \in L_{\infty}(\Omega, \Sigma, \mu_{M})\), \(0 \leq \phi \leq 1\), such that \( N(\triangle)= \int_{\triangle} \phi (x)\) \(M( dx)\), for all \(\triangle \in \Sigma\). III. If \(M_{1} \to M_{2}\) and \(M_{2}\) is non-degenerate, then \(M_{1}\) is also non-degenerate. IV. (The Main Result) (i) An \(L(\mathcal{H})\)-valued \(POV\) measure \(M\) on \((\Omega, \Sigma)\) is maximal if and only if there is a bounded linear operator \(V: \mathcal{H} \to L_{2}(\Omega, \Sigma, \mu_{M})\) such that \(M= V^{*} \mathbb{I}(\cdot) V\). (ii) Let \(\mathcal{K}\) be a complex Hilbert space, \(E\) an \(L(\mathcal{K})\)-valued spectral measure on \((\Omega, \Sigma)\), and \(V: \mathcal{H} \to \mathcal{K}\) a bounded linear operator. Assume that \(\mathcal{K} = \text{cl(span} \{ E(\triangle) V(h): \triangle \in \Sigma, \; h \in \mathcal{H} \})\). Then the \(POV\) measure \(M= V^{*} E(\cdot) V\) is maximal if and only if \(E\) generates a maximal commutative von Neumann algebra. Some other related results are also proved.