Approximation of linear functionals on a Banach space with a Gaussian measure (Q1082039): Difference between revisions

The paper is a study of linear functionals on a Banach space illustrated by the integration example \(Sf=\int^{1}_{0}f(t)dt.\) The problem considered is to find an optimal algorithm with respect to 3 criteria (i) it has the minimal average error, i.e. \(\int_{F_ 1}E(Sf-\phi)\mu df\) is a minimum where E denotes the error in Sf using the algorithm \(\phi\) and \(\mu\) is a probability measure on the \(\sigma\)-field \(B(F_ 1)\) of Borel sets from the Banach space \(F_ 1\); (ii) it has maximal probability of not exceeding a given error bound so that given \(\epsilon >0\) \[ \Pr ob \{E(Sf-\phi)\leq \epsilon \}=_{u} \Pr ob \{E(Sf-u)\leq \epsilon \}; \] and (iii) if \(\phi_ n\) denotes the algorithm that depends on the cardinality of the information (such as the number of integration points) then \(| Sf-\phi_ n|\) has the best rate of convergence as \(n\to \infty\) for almost all functions. Thus optimality is defined in the average, probabilistic and asymptotic senses. It is shown that the \(\mu\)-spline algorithms are optimal in all senses and that adaptive information does not improve the algorithms. Applications are considered and the results are derived using a Wiener measure for the space \(C^ r[0,1]\).