A Monte Carlo method for Poisson's equation

Laplace operator, Helmholtz equation (reduced wave equation), Poisson equation (35J05) Monte Carlo methods (65C05) Applications to the sciences (65Z05)

Review text

A Monte Carlo method for estimating local solutions \(u(x_ 0)\) to the Dirichlet problem for Poisson's equation \(\nabla^ 2u(x)=-q(x)\), \(x\in \Omega\), \(u(x)=f(x)\), \(x\in \partial \Omega\) is developed and analyzed. Given a point \(x_ k\) in \(\Omega\), the method proceeds by sampling a point \(y_{k+1}\) in the largest ball \(B(x_ k)\) centered at \(x_ k\) and contained in \(\Omega\) according to the Green's function based at \(x_ k\) and sampling a point \(x_{k+1}\in \partial B(x_ k)\) uniformly. After a random number of iterations, say \(n^*\), \(x_{n^*}\) is within a pre-selected distance \(\delta\) of the boundary \(\partial \Omega\) and the iterations stop. Let \(x'\) be the nearest point of \(\partial \Omega\) to \(x_{n^*}\). The estimate of \(u(x_ 0)\) is taken as the sum \(z_ i=f(x')+\sum^{n^*-1}_{k=0}a(x_ k)q(y_{k+1})\). Here \(a(x_ k)\) is the normalization factor stemming from using Green's function as a distribution. The walk is repeated say N times and \(u(x_ 0)\approx S_ N=\frac{1}{N}\sum^{N}_{i=1}z_ i\). This procedure is called a modified ``walk on spheres. It is shown that the expectation \(E_{n^*}=O(| \ln \delta |)\) and the variance var \(S_ N=O(N^{-1})\) is achieved with O(N log N) computational operations. Some numerical experiments are conducted on a problem having a large temperature gradient near a boundary point.

Recommendations

Cites work

Cited in

(33)

This page was built for publication: A Monte Carlo method for Poisson's equation

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q753456)