We present a new finite difference code for modeling three-dimensional thermal convection in a spherical shell using the "cubed sphere" method of Ronchi et al. [1]. The equation of motion is solved using a poloidal potential formulation of the equation of motion for an iso-viscous, infinite Prandtl number fluid on a finite difference grid and advective transport is implemented using the 2 nd -order MPDATA scheme of Smolarkiewicz [2]. Due to the high efficiency of multigrid acceleration, low memory requirements, and second-order accuracy of this model, we conclude that the cubed sphere method offers a great deal of potential for simulating complicated problems of fluid mechanics in spherical geometry.
IntroductionHighly viscous thermal convection in the rocky mantles of the solid planets is the primary process governing their thermal and mechanical evolution over long time scales. This process is driven primarily by the transfer of heat from the interior to the surface. Because thermal convection constitutes a non-linear problem, the main tool for studying finite amplitude convective motions is computer models. Due to the relative simplicity of the relevant equations in a parallel coordinate system, Cartesian domains are often used to obtain an approximation to convective motions in planetary mantles. This approximation has proven to be useful for many different problems, although it imposes a geometric symmetry between the upper and lower boundaries which does not exist in spherical geometry.In the first spherical shell models the most popular method for solving the relevant equations were spectral, i.e., using spherical harmonics as basis functions. This allows a great deal of simplification of the equations (Chandrasekhar [3]), however, computation of the Legendre transforms can be expensive. The method's primary limitation, however, is that the elegance and simplicity are destroyed when more complicated effects, such as laterally-varying viscosity, are included, and viscosity varies by many orders of magnitude in planetary bodies. In addition, spectral methods do not offer a straightforward implementation on parallel computers since the basis functions are not local.Recently, there has been an increase in the application of grid-based methods to mantle convection in spherical geometry. Methods such as finite elements and finite differences are local and can therefore more easily accommodate complex effects such as