A High-Resolution Cartesian Grid Method for the Approximation of Conservation Laws in Complex Geometries



Our efforts are directed towards the development of high-resolution Cartesian grid methods for the approximation of multidimensional systems of conservation laws in complex irregular geometries. A Cartesian grid approach is attractive, since away from the boundary it allows the use of standard high-resolution shock capturing methods that are more difficult to develop on unstructured (body fitted) grids. Furthermore, embedded boundary methods allow a more automated grid generation procedure around complex objects, which is important especially for three-dimensional problems.

The numerical challenge associated with a Cartesian grid embedded boundary approach is the so-called small cell problem. Near the embedded boundary the grid cells may be orders of magnitude smaller than regular Cartesian grid cells. Since standard explicit finite volume methods take the time step proportional to the size of a grid cell, this would typically require small time steps near an embedded boundary. We have developed a numerical method that overcomes the time step restriction, while seeking an accurate and conservative approximation near the embedded boundary as well as in the whole domain. Our approach to overcome the small cell problem is based on the so-called h-box method suggested in earlier work by Berger and LeVeque. The basic idea behind this method is to approximate numerical fluxes at the interface of a small cell based on values specified over regions of length h, where h depends on the size of a regular grid cell. In order to obtain a method that is stable for a CFL condition adequate for grid cells away from the boundary, the flux differences at cut cells should be of the size of the small irregular cell.

In order to gain a better understanding of these methods, we have studied the construction of high-resolution h-box methods for one-dimensional conservation laws on irregular grids. We  show that our method is:

 Numerical studies confirmed the same properties for the approximation of nonlinear systems, e.g. the Euler equations. This work is presented in: Recently we have extended these ideas to the two-dimensional case. In order to obtain the required cancellation property of numerical fluxes, we use a rotated grid method. We have developed a method that leads to a second order accurate approximation of boundary cells that may be orders of magnitude smaller than regular grid cells. Click  here for results showing the performance of our new embedded boundary method.