Paper detail

Parallel stochastic methods for PDE based grid generation

The efficient generation of meshes is an important step in the numerical solution of various problems in physics and engineering. We are interested in situations where global mesh quality and tight coupling to the physical solution is important. We consider elliptic PDE based mesh generation and present a method for the construction of adaptive meshes in two spatial dimensions using domain decomposition that is suitable for an implementation on parallel computing architectures. The method uses the stochastic representation of the exact solution of a linear mesh generator of Winslow type to find the points of the adaptive mesh along the subdomain interfaces. The meshes over the single subdomains can then be obtained completely independently of each other using the probabilistically computed solutions along the interfaces as boundary conditions for the linear mesh generator. Further to the previously acknowledged performance characteristics, we demonstrate how the stochastic domain decomposition approach is particularly suited to the problem of grid generation - generating quality meshes efficiently. In addition we show further improvements are possible using interpolation of the subdomain interfaces and smoothing of mesh candidates. An optimal placement strategy is introduced to automatically choose the number and placement of points along the interface using the mesh density function. Various examples of meshes constructed using this stochastic-deterministic domain decomposition technique are shown and compared to the respective single domain solutions using a representative mesh quality measure. A brief performance study is included to show the viability of the stochastic domain decomposition approach and to illustrate the effect of algorithmic choices on the solver's efficiency.