Exploring Unstructured Poisson Solvers for FDS

Many presently employed Poisson solvers which offer a sufficient amount of computational efficiency and robustness are limited to the use of regular geometries and corresponding grid structures. This restriction can affect the precise representation of complex fire scenarios involving complex bodies and flow obstructions. Usually, their parallel application in the context of multi-core architectures contributes to a further impairment. The current solver for the Poisson equation in FDS is based on the use of local FFT methods on the single meshes of the underlying domain decomposition. This approach has proven to be computationally efficient and accurate in a multitude of cases. But due to its restriction to rectilinear meshes and its purely local character, there are two possible drawbacks, namely the presence of velocity errors based on penetrations into immersed obstacles as well as possibly large velocity errors along mesh interfaces. In order to face this challenge, several alternative Poisson solvers of direct and iterative type are examined which basically apply global solution strategies spanning over the whole domain decomposition. Furthermore, their ability to deal with unstructured grids along with the exact setting of boundary conditions on internal obstacle surfaces shall be analyzed. The paper and its associated talk are intended to give some insights into the current state of development and to compare the pros and cons of the different Poisson approaches with respect to their efficiency and accuracy.