Next generation supercomputers like Yellowstone favor mesh-based models like CAM-SE (HOMME) that can scale without loss of performance up to 100,000 processor-cores or more. However, from an optimization standpoint, the best way to mesh the globe remains an open question. The regular longitude-latitude (RLL) grid and the cubed-sphere are familiar choices, but many other options exist. Despite its simplicity, the RLL grid suffers from meridian convergence at the poles leading to: numerical singularities, CFL stability restrictions, and the need for expensive polar-filtering, collectively know as "the pole problem." In this talk, we examine the promising Yin-Yang mesh, a composite mesh combining two overlapping RLL grids that avoids the pole problem while retaining the orthogonality and quasi-regularity of the regular longitude-latitude grid. We demonstrate its performance on several transport benchmarks on the sphere, in a discontinuous Galerkin (DG) environment, and show that the Yin-Yang + DG combination has the potential to be very competitive for atmospheric modeling.
◘Mesa Laboratory – Main Seminar Room ◘
Wednesday April 25, 2012