Researchers developed, tested modeling programs at the Ohio Supercomputer Center; awarded 15M compute hours on Oak Ridge National Laboratory’s newly installed “Jaguar”

With complex computer modeling programs fine-tuned at the Ohio Supercomputer Center, a team guided by researchers from OSC and Louisiana State University will be among the first to road-test the world’s second-fastest supercomputer.

The interdisciplinary team of researchers, led by Mark Jarrell, Ph.D., a physicist at LSU, and Karen Tomko, Ph.D., a senior systems developer/engineer at OSC, has been awarded 15 million compute hours on the “Jaguar,” Oak Ridge National Laboratory’s newly upgraded Cray XT supercomputer. With its improvements, the Jaguar is one of only two computers world-wide that compute on the petascale level, with a peak performance of 1.6 petaflops.

The team’s work on modeling strongly correlated materials was one of 20 projects selected for ORNL’s “Petascale Early Science Allocation Program.” They have been developing their computing codes over the past year on OSC’s IBM Cluster 1350, located in Columbus on the west campus of The Ohio State University.  The OSC cluster was ranked among the top 60 supercomputers in the world when it was installed in early 2008.

“It’s an honor to be selected as one of the first groups to have access to Jaguar,” Tomko said. “Our time at OSC has given the team a chance to identify numerical instabilities in our algorithms and to find solutions. Additionally, it has enabled us to tune the code for execution on multi-core systems while using as many as 2,048 CPU cores. Our runs on Jaguar will use CPU cores that total an order of magnitude more.”

A total of 500 million compute hours have been awarded to 20 projects for this special, six-month period. To give the numbers context, 1 million processor-core hours could be spent, in theory, by using 1,024 processor-cores for about 1,000 hours, or 40 days. Or, the allotted time could be spent by using 10,000 processor-cores for a period of 100 hours, or less than five days.

The goals of this early phase, according to ORNL’s web site, are to deliver important, high-impact science results and advancements; harden the system for production; and embrace a broad user community capable of and prepared for using the system.

Jarrell, Tomko, and the team of applied mathematicians, computational physicists and computer scientists from the University of Cincinnati, Oak Ridge National Laboratory and the University of California-Davis will definitely meet those goals with their work to develop a massively parallel, multi-scale method to study strongly correlated materials, a wide class of materials with unusual electronic and magnetic properties.  Ultimately, this project will lead to better materials that companies can use to develop better technology for areas as diverse as energy production, electronic devices, medical science and national security.

Researchers have long wanted to study highly intricate, strongly correlated materials at their three different levels: short length scales, single atomic particles such as one electron; long length scales, a group of atoms; and intermediate length scales, a collection of 10 or so atoms. But before petascale computing, it was mathematically overwhelming to evaluate materials at intermediate length scales. Because of a variety of factors, the equations that looked at the interactions of the middle lengths quickly scaled off the charts.

“Our approach, from the treatment of the correlations at intermediate length scales to the integration of the three length scales, is completely new,” Jarrell said. “Our goal is to develop computational methods that separate a material’s correlations by length scale. These algorithms will enable physicists, for the first time, to accurately study the complex interactions of strongly correlated materials.”

The “Next Generation Multi-scale Quantum Simulations Software for Strongly Correlated Materials” project is funded by the U.S. Department of Energy, Scientific Discovery through Advanced Computing.