Green's function Monte Carlo algorithms for elliptic problems. (Q1418588)

The authors consider three versions of the walk on balls Monte Carlo method for the direct computation of linear functionals of solutions to elliptic boundary value problems. The algorithms are constructed using Green's function analysis to define the conditions under which the integral transformation kernel of the integral representation for the boundary-value problem is non-negative. The kernels are defined in terms of radially symmetric density functions where one of the three derived variants is classical. Numerical experiments show that all methods under consideration need essentially the same number of Monte Carlo steps and show the same convergence behavior. Nevertheless, the algorithm based on a constant density function shows higher efficiency than the other two (including the classical one) because it has the lowest computational complexity.