A variance reduction method based on sensitivity derivatives (Q2495434)

A variance reduction method using sensitivity derivatives is developed to enhance the accuracy of Monte Carlo schemes for approximating the solutions of stochastic partial differential equations. Results of numerical experiments on Burgers' equation are given which indicate a significant decrease in the number of samples needed for the Monte Carlo schemes to obtain a given accuracy. Combination of the variance reduction method with a stratified sampling technique to obtain a further significant decrease in the number of samples needed is described, and its effectiveness is evidenced by numerical experiments.