On the numerical solution of a nonlinear stochastic Helmholtz equation with a multigrid preconditioner (Q1091104)

The paper reports some results on the numerical approximation for a nonlinear stochastic Helmholtz equation of the form \(\Delta u+k^ 2n^ 2(x,| u|)=0\), where the index of refraction n is a random function of x and a nonlinear function of \(| u|\). This equation, with suitable auxiliary conditions, is first linearized by Newton's method, and then discretized using a finite element discretization. The resulting linear system is solved iteratively using a preconditioned conjugate gradient method which incorporates a version of the multigrid method. A computer code is implemented, and a one-dimensional model problem is analyzed in order to study the numerical method; for this problem the authors found that the convergence for a multigrid preconditioner is independent of the mesh size for k fixed, roughly independent of the random coefficient, and to some extent independent of the nonlinearity. An unexpected bifurcation in the solution of the nonlinear stochastic model problem is also discovered. The applicability of this multigrid preconditioner method to two-dimensional problems is also discussed.