Applied Math/Computing Programs at DOE: Focus Is on Next-generation CapabilitiesDecember 13, 2001
On the agenda at the recent meeting of SIAM's Committee on Science Policy were two visitors from the U.S. Department of Energy: Dan Hitchcock and Charles Romine. Between them, they did much to clarify the intricacies of the important relationship between DOE's Office of Science and the SIAM community.
Hitchcock is the senior technical adviser for advanced scientific computing in DOE's Office of Advanced Scientific Computing Research (ASCR---confusingly, to some ears, pronounced "OSCAR"). He is also acting program manager of the Applied Mathematics Research Program. From the perspective of his two jobs, he provided an insightful look at near- and long-term research interests of DOE, and some of the programs created to meet those needs.
SciDAC (Scientific Discovery through Advanced Computing) is one such program. As the program manager in ASCR's Mathematical, Information, and Computational Sciences (MICS) Division who is responsible for SciDAC, Romine spoke to SIAM News in August, shortly after the announcement of the first round of awards.
The SciDAC awards---51 in all, for a total of $57 million in FY 2001---fund research teams working to produce scientific simulation codes that take advantage of the capabilities of terascale computers to address ever more complex problems of interest to DOE. Romine lists the most important application areas, which arise across the Office of Science, as computational climate dynamics, high-energy and nuclear physics, fusion energy, and basic energy science, with computational chemistry.
The program has multidisciplinary teams called ISICs---Integrated Software Infrastructure Centers---in both applied mathematics and computer science. The ISICs have ties to the applications, but are focused mainly on development---and deployment---of the technology. The key issue, Hitchcock says, is that "the market for most high-performance software is too small to be commercial." The ISICs are working to make robust, usable software available to scientists.
"Computer scientists are the 'glue' between the applications and the advanced technology," Hitchcock says. "We need people who are 'bilingual' if we are going to progress on some of the important large problems."
The ISICs (four in computer science, three in applied mathematics) have from 12 to 25 members, about half of them in universities. In all, SciDAC participants represent more than 50 colleges, universities, and companies, and 13 DOE labs.
Some "outstanding ideas" proposed during the initial round of SciDAC "were not competitive under present constraints," says Romine, who hopes that the SciDAC program can be expanded. The applications in the current ISICs for the most part require PDEs, he points out. As a result, "some fundamental algorithms are not covered under the existing ISICs." With a flat budget for this year (FY 2002), however, new SciDAC proposals are not being sought. Attention is focused, rather, on the base program---"things you need to do," Hitchcock says, "to make sure you have the software you need" for the machines and systems that will be available in five or ten years. Thinking ahead to 2006, Hitchcock names algorithmic complexity as a key issue to be considered: "If algorithms are too complex, you don't gain much with each teraflop, even if things scale perfectly."
Briefly describing the base program, Hitchcock refers to continued investments in computational fluid dynamics, optimization, linear algebra, differential equations, and mesh generation. DOE has a long-standing interest in mathematical biology, he points out, citing the proposed (FY 2003) Genomes to Life initiative; given the current climate, the emphasis could well be on how certain microbes work. Other ongoing interests are predictability (with the possibility of a SIAM-conducted workshop in the coming year) and large data sets (with proposals now welcome in the area of feature extraction, an area with both mathematical and computer science elements).
Overall, Hitchcock says, ASCR is interested in the next generation of tools. It is also interested in future generations of scientists, hoping to introduce this year a young investigator program at universities (funded at about $2 million). And, of course, the Computational Science Graduate Fellowship program so energetically supported by the late Fred Howes is doing its part to produce the computational scientists who will be critical to the solution of most large scientific problems in the future. The CSGF program, which celebrated its tenth anniversary last year, has supported more than 150 students at approximately 50 universities. Fellows must be in full-time PhD programs at U.S. universities, where their program of study must include coursework in mathematics, computer science, and a scientific or engineering discipline. Information about the program, and applications for FY 2002-03 fellowships, can be found at www.krellinst.org/CSGF/CSGF.html. The application deadline is January 16, 2002.
As to the success of the CSGF program, Romine advises anyone interested to attend one of the annual conferences at which third-year fellows present their research: "It's the highest signal-to-noise ratio you'll ever hear."
Funding for applied mathematics in the Department of Energy's Office of Science.