Challenges and Opportunities for Progress in CSEJune 15, 2009
Shown in Miami at the 2009 SIAM Conference on Computational Science and Engineering are SIAG/CSE officers (from left) Kirk Jordan (vice chair), Tammy Kolda (chair), Carol Woodward (secretary/treasurer), and Uli Rüde (program director). Two of them---Jordan and Woodward---chaired the organizing committee for the conference, and all helped with coverage for SIAM News, including the accompanying article on Tinsley Oden’s invited talk; a collection of articles on education-related conference sessions can be found in the May issue. Conference photos by Susan Whitehouse.
It's a rare talk at a SIAM conference that begins with a discussion of the philosophy of science in general, and Karl Popper's principle of falsification in particular.
J. Tinsley Oden, in an invited talk that set the scene for much that would follow at this year's SIAM Conference on Computational Science and Engineering, made the point that simulation-based engineering science, and the closely related CSE, are new areas of science that fit within Popper's general framework for the creation of scientific knowledge, with some differences. The scientific method, Oden pointed out, builds on observations, typically from experiment, to form theories that are tested by further observations. A hypothesis, by the popperian principle of falsification, becomes a legitimate scientific theory only if it can be refuted by observational evidence.
Clearly, Oden considers CSE an emerging area, "an important discipline in its own right," with far-reaching implications for all of science and technology: "CSE is going to permeate every area of science and technology, enriching applied math in the process."
Simulation-based engineering science (SBES) and CSE, used almost interchangeably by Oden (and in this article), make possible the discovery of things that cannot be observed directly and allow the testing of multiple hypotheses. (The two reports mentioned at the end of this article offer a good perspective on SBES.)
Today, Oden said, "epochal changes are occurring in science and engineering"---a reference in part to rapidly advancing computational capabilities---computing at the petascale now and at the exascale in the not-too-distant future---and in part to emerging "super algorithms" that enable performance improvements greater than those provided by computing power alone. These advances are making possible serious computational modeling and simulation of complex real-world systems. Mathematical models were long thought to provide at most qualitative information, Oden said. That has changed significantly in the last decade: People have begun to expect quantitative results, and in particular predictions whose uncertainty is quantifiable.
Validation and verification (V&V) and uncertainty quantification (UQ) were major themes both of Oden's talk and of the SIAM conference (which included a well-attended four-hour tutorial on techniques for UQ). Here again, Popper came to the fore. Parameters in models have errors, from uncertainties in measurements, among many sources, and variation. Just as a model (or a theory) cannot be proved, but only refuted, we cannot prove that a code (with specific parameters) provides the "correct" answer; rather, we must calibrate the models, by comparing model output with observed measurements, relying also on an a posteriori probability distribution function that characterizes the uncertainty in the answers.
Threaded throughout Oden's talk was a discussion of barriers to and opportunities for progress in SBES. We need a major investment in algorithms that will allow researchers to take advantage of steady advances in hardware, he said, mentioning one striking consequence of the ongoing evolution of hardware: People are developing algorithms for machines that don't yet exist. And as a practical matter, modifications in codes themselves must keep pace with the advances in hardware and algorithms.
A further challenge lies in the multiscale nature of complex systems. As Oden pointed out, our entire knowledge of physical systems is partitioned into scales, with each discipline tending to have its own scale intervals (in space and time); SBES needs to transcend scales.
Oden made a case for the effectiveness of team-based interdisciplinary work in CSE. For example, he considers climate modeling an area that is ripe for CSE, and especially for V&V: "You'd think that V&V would be well established in the field, but many of the issues of V&V have not found their way into climate modeling."
Finally, despite the emergence of CSE as an important area of study, Oden is cautious about proposals to create separate academic departments of CSE. "In what sense," he asks, "will CSE be a discipline in its own right---as opposed to, say, computational chemistry?" He identifies two camps: a user camp that views CSE as a provider of tools (albeit tools that will change), and a camp that focuses on the mathematical/computational methods and defines CSE as an area of scientific discovery.
Invited speaker Tinsley Oden, introduced in Miami by Carol Woodward, is director the Institute for Computational Engineering and Sciences (ICES) at the University of Texas, Austin. University structures that allow CSE to flourish were a much-discussed topic at the conference, and ICES offers one extremely successful model.
Within ICES is the longstanding CAM (Computational and Applied Mathematics) program, recently renamed CSEM (Computational Science, Engineering, and Mathematics) to reflect the increasingly interdisciplinary nature of the program, whose faculty hold tenure in 17 departments. CSEM students (there are currently 57) receive a heavy dose of graduate-level mathematics, computer science, and computing. Each student works with a mentor, and writes a proposal for research in an application area (e.g., quantum mechanics, acoustics, flow in porous media). CSEM awards about five doctorates a year. Most graduates get multiple job offers; about half go to academia, a quarter to government labs, and some of the rest to industry.
The program provides a unique atmosphere for students, who, in turn, make key contributions to the program's success: "Students are the glue that makes everything work," Oden said in a phone conversation with SIAM News. "I envy them. They really thrive in that environment.""
CSE needs a free flow of students and faculty," he says. "To treat it as an isolated academic department with the usual departmental barriers could destroy it. CSE has to be done by people working in interdisciplinary teams---that's how the important problems are solved."---JMC
Two Key Reports on SBES
Tinsley Oden chaired the National Science Foundation's Blue Ribbon Panel on Simulation-Based Engineering Science that produced the 2006 report "Revolutionizing Engineering Science through Simulation." The report focused on the future of SBES, defined as the application of computational models to the study and prediction of physical events or the behavior of engineered systems.
The report can be found at http://www.nsf.gov/pubs/reports/sbes_final_report.pdf.
SIAM was also well represented on a later blue ribbon panel convened to perform a comparative assessment of opportunities and challenges in SBES worldwide. The report of that panel, whose members included Linda Petzold and George Karniadakis, is titled "An International Assessment of Research and Development in Simulation-based Engineering and Science."
Recently released by the World Technology Evaluation Center in Baltimore, Mary-land, the report is available at http://www.wtec.org/sbes/.