dhw@iti.org (David H. West) (05/10/91)
I saw (but can't find) a report that someone has discovered a way to hybridize Monte Carlo and Gaussian Numerical Integration: you specify a tolerance e and a number of dimensions k, and the method computes an integer N and a set of N points (and presumably weights) in the unit k-dimensional hypercube such that in some probabilistic sense function values at those points allow the integral of the function to be estimated with precision better than e for "almost all" functions, AND the behavior of N with k is much better than the e^-k that Monte Carlo gives. Can anyone point me to this work? Please email. thanks, -David West dhw@iti.org