Improving Monte Carlo simulations by Dirichlet forms (Q2565527)

The author considers the framework of an error structure, i.e. a probability space endowed with a local Dirichlet form having the square field operator \(\Gamma\) and the associated \(L^2\)-generator \(A\). The aim of the paper is to show that, if one can simulate in the error structure framework the random variable \(X\) together with \(\Gamma[X]\) and \(A[X]\), then the Monte Carlo simulations of the expectations and densities of \(X\) can be accelerated significantly. Examples supporting these facts are exposed for error structure frameworks developed on Wiener space, Poisson space, and Monte Carlo space. For the particular situation when \(X\) is real-valued and \(X\) can be simulated together with \(A[X]\), \(\Gamma[X]\), and \(\Gamma[X, 1/X]\), the paper provides an explicit formula for the density of \(X\) that can be simulated at the speed of the law of large numbers.