The Dirichlet problem in \(C^*\)-algebras (Q1177885): Difference between revisions

Let \(A\) be a \(C^*\)-algebra, \(A^{**}\) its enveloping von Neumann algebra. The author uses the multiplier algebra \(M(A)\) for when \(A\) does not have a unit element. The article gives a non-commutative \(C^*\)-algebraic version of the classical solution to the Dirichlet problem, viz. \(\Delta u=0\) on a bounded domain \(D\) such that \(\displaystyle\lim_{a\in D,\ a\to b}u(a)=f(b)\) for each \(b\in\partial D\), mimicking the construction and hypotheses of \textit{E. B. Dynkin} [Markov processes. Berlin etc.: Springer-Verlag (1965; Zbl 0132.37901)] for the stochastic solution \(u(a)=\mathbb{E}_ a\{f[x(m_{\partial D})]\}\) where \(m_{\partial D}\) is the first passage time to \(\partial D\). The non-commutative Dirichlet structure is provided by a `quantum' semigroup \((\varphi_ t)\), \(t\geq 0\), whose infinitesimal generator takes the role of the Laplacian. The \(\varphi_ t\) are continuous completely positive contractions of \(A\) to itself which are Markov in the sense that for any state \(\omega\) of \(A\), \(\omega\cdot\varphi_ t\) remains a state of \(A\). The extension of \(\varphi_ t\) to \(A^{**}\) is denoted by \(\bar\varphi_ t\). The domain of definition is a closed two-sided ideal \(I\) and the boundary condition is represented by an element \(\alpha\) of \(A/I\). Denoting the annihilator of \(I\) in \(A\) by \(I'\), \(\partial I\) is defined to be \(A/(I+I')\). Solution comprises the construction of a `Dirichlet kernel', viz. a lifting \(\Lambda\) of \(\alpha\) to \(A\) which is harmonic on \(I\), i.e. \(\Delta_ I(\Lambda(\alpha))=0\); \(\Delta_ I\) is a localisation of \(\Delta\) defined by \(\Delta_ I(x)=\lim_{t\searrow0} \rho_ I\{t^{- 1}[\varphi_ t(x)-x]\}\), where \(\rho_ I\) denotes the restriction of \(A\), or \(M(A)\), to \(M(I)\) and the convergence is a non-commutative analogue of compact convergence. To construct \(\Delta\) the author first dilates the quantum dynamical system to obtain a `quantum Markov process' i.e. a system of von Neumann algebras \(M_ t\subset M\), \(t\geq 0\), \(M_ 0=A^{**}\), with covariant and filtering conditional expectations \((E_ t)\), and a semigroup \(\sigma_ t\) of normal endomorphisms of \(M\), such that \(\bar\varphi_ t(x)=E_ 0[\sigma_ t(x)]\) for \(x\in A^{**}\); \(M_ t\) and \(\sigma_ t(M)\) represent the `events' or observables, dependent on \([0,t]\) and \([t,+\infty)\) respectively. The kernel \(\Lambda\) is determined by the mapping of \(a\in A\) to \(E_ 0\{\sigma_ I(a)\}\in A^{**}\) where \(\sigma_ I(a)\) is the process stopped at \(\partial I\) by using an increasing family of projections taking the role of first exit time [cf. \textit{R. L. Hudson}, J. Funct. Anal. 34, 266--281 (1979; Zbl 0424.46048)]. The assumption of the locality hypothesis (cf. Dynkin, loc. cit.), here the existence of a certain approximate identity in \(I\), ensures that \(\Lambda\) is null on \(I\). To verify harmonicity and to be able to glue together the bits of \(\Lambda(A)\) and \((A/I)^{**}\) and \(I^{**}\) the author uses non-commutative analogues of the hypothesis that \((\varphi_ t)\) is a Feller semigroup, viz. \(\varphi_ t(A^{**})\in M(A)\) for all \(t>0\), and of the hypothesis of regularity of the boundary which ensures that \(I\) does not trap particles at its boundary (Dynkin, loc. cit.). An ergodicity hypothesis will ensure that \(\Lambda\) is Markovian. If \(I\) is `relatively compact' (i.e. \(A/I'\) has an identity element) then \(\Lambda\) is completely positive. Unfortunately the author does not give any examples, though he does mention that the Dirichlet problem might be solvable in \(C^*\)-algebras of foliations.