Numerical Evaluation of Functionals based on Variance Minimisation

Abstract

AbstractNumerically evaluating the effect of a functional on a function is a very common task in scientific computing. The definite integral of a function over a domain is an example, differentiating a function in a certain point into a certain direction is another one.We developed a generic method to compute the effect of a functional using a linear approximation formula. The method is designed to generate the nodes and weights needed to approximate different functionals using a single set of tools: it regards the target function as a stochastic field and uses a user–defined covariance function for this field to minimise the error made by the approximation formula.The resulting formulas are optimal in an average case sense: all possible realisations of this stochastic field are taken into account while computing the solution. This results in nodes and weights that evaluate the target functional applied to any realisation with a minimised average error. The space of all realisations of such a stochastic field can be of infinite dimension whereas classical approaches often only consider a finite dimensional space of functions. (© 2008 WILEY‐VCH Verlag GmbH & Co. KGaA, Weinheim)