Magnetism Simulation Receives 1998 Gordon Bell Award
A team of scientists from Oak Ridge National Laboratory's Computer Science & Mathematics,
Computational Physics & Engineering, and Metals & Ceramics Divisions working with the National
Energy Research Scientific Computing Center (NERSC) at Lawrence Berkeley National Laboratory
received the 1998 Gordon Bell Award for their parallel computer simulation of metallic magnetism.
The ORNL-led team, which also included collaborators at the Pittsburgh Supercomputing Center
and the University of Bristol (UK), performed a 1,024-atom first-principles simulation of metallic
magnetism in iron which ran at 657 Gigaflops (billions of calculations per second) on a 1024-processor Cray/SGI T3E supercomputer.
On Sunday, November 8, the code ran at 1.02 Teraflops (trillions of calculations per second)
on a larger system at Cray -- the first real science application to achieve a Teraflop!
Funded as one of the U.S. Department of Energy's Grand Challenges, the group developed the computer code to provide a better microscopic understanding of metallic magnetism, which has applications
in fields ranging from computer data storage to power generation and utilization.
"One of the goals of this project is to address critical materials problems on the micro-structural
scale to better understand the properties of real materials," said Malcolm Stocks, a materials scientist at Oak Ridge and leader of the project.
The work was performed by Malcolm Stocks, William A. Shelton, Donald M.C. Nicholshon, Xiaoguang Zhang,
Xindon Wang, and Balazs Ujfalussy, Oak Ridge National Laboratory; Andrew Canning, NERSC, Lawrence Berkeley Laboratory; Yang Wang, Pittsburgh Supercomputing Center; and B.L. Gyorffy, H.H. Willis Physics Laboratory, UK.
ORNL researchers combined CLM (constrained local moment--which maintains the orientational configuration) with the LSMS method and validated it on a bcc Fe system (where results at certain phases can be confirmed by experiment), achieving 657 Gflops in the process.
In initial testing, we found that the constraining fields converged rapidly (a few tens of iterations), while convergence of the charge and magnetization densities required 800 iterations. Multiple scattering processes outside a local interaction zone (LIZ) centered on each atom are ignored.
512-atom CLM state of prototypical paramagnetic bcc Fe. The top frame shows the SCF magnetic moments, the bottom frame shows the corresponding transverse constraining fields. Atom positions are denoted by spheres, magnetic moments by arrows and constraining fields by cones. The magnitudes are color coded as indicated. These calculations are for LIZ27 and were performed on the T3E900 LC512 at NERSC.
