Previous abstract Next abstract

Session 69 - The Solar Dynamo and Helioseismology.
Display session, Thursday, June 13
Tripp Commons,

[69.08] High Resolution Numerical Simulations of Global Solar Convection

M. Miesch, T. Clune, J. Toomre (JILA, Univ. of Colorado), G. Glatzmaier (Los Alamos Nat. Lab.)

We present a new computer code for simulating anelastic stellar convection in a rotating spherical shell which is based on an existing algorithm, but redesigned to take full advantage of the higher resolution possible on currently available parallel super-computing platforms. Similar previous studies have led to important insights into the dynamics of the solar convection zone, but are unable to reproduce several important features, in particular the latitudinal and radial angular velocity profile inferred from helioseismological inversion. The results from helioseismology imply a differential rotation which is constant on radial lines at mid latitudes in the convection zone, while numerical simulations generally exhibit profiles which tend to be constant on cylindrical surfaces aligned with the rotation axis. Spherical shell simulations by Glatzmaier and other results from convection in plane-parallel geometries suggest that the answer may lie in increasing the spatial resolution of the model. The relatively low resolutions of previous simulations only admit predominantly laminar flows, which are known to exhibit significantly different transport properties than turbulent flows, and which are therefore less applicable to the highly turbulent conditions in solar convection zone. To achieve the highest possible resolution, and therefore the most turbulent flows, on current (and near future) computational resources, our new implementation of Glatzmaier's earlier code is specifically suited to the hierarchical memories characteristic of MPPs, with careful consideration given to achieving both good serial performance as well as good scalability on this class of machines. The former is largely achieved through an extremely efficient implementation of the Legendre transform which constitutes the majority of the computational workload, while the latter is achieved primarily by a relatively complex load-balancing scheme. We discuss the implementation as well as the flow characteristics and transport properties of several simulations achieved on our initial target platform, the IBM SP-2 at the Cornell Theory Center, with spatial resolutions of up to 768 \times 1536 \times 129 (latitude, longitude, radial).

Program listing for Thursday