Black Hole Excision with Multiple Grid Patches

Abstract

When using black hole excision to numerically evolve a black hole spacetime with no continuous symmetries, most 3+1 finite differencing codes use a Cartesian grid. It's difficult to do excision on such a grid, because the natural $r = \text{constant}$ excision surface must be approximated either by a very different shape such as a contained cube, or by an irregular and non-smooth "LEGO(tm) sphere" which may introduce numerical instabilities into the evolution. In this paper I describe an alternate scheme, which uses multiple $\{r \times (\text{angular coordinates}) \}$ grid patches, each patch using a different (nonsingular) choice of angular coordinates. This allows excision on a smooth $r = \text{constant}$ 2-sphere. I discuss the key design choices in such a multiple-patch scheme, including the choice of ghost-zone versus internal-boundary treatment of the interpatch boundaries, the number and shape of the patches, the details of how the ghost zones are "synchronized" by interpolation from neighboring patches, the tensor basis for the Einstein equations in each patch, and the handling of non-tensor field variables such as the BSSN $\tildeΓ^i$. I present sample numerical results from a prototype implementation of this scheme. This code simulates the time evolution of the (asymptotically flat) spacetime around a single (excised) black hole, using 4th-order finite differencing in space and time. Using Kerr initial data with $J/m^2 = 0.6$, I present evolutions to $t \gtsim 1500m$. The lifetime of these evolutions appears to be limited only by outer boundary instabilities, not by any excision instabilities or by any problems inherent to the multiple-patch scheme.