NCSA NEWS

'Easy to talk about, almost impossible to do'

Story posted December 12, 2006

As director of the University of Texas at Austin's Institute for Computational Engineering and Sciences, J. Tinsley Oden leads an interdisciplinary research center for faculty and graduate students. Institute members focus on computational studies of geoscience, materials science, acoustics, and a host of other fields. This breadth made Oden a natural choice to chair the National Science Foundation's Blue Ribbon Panel on Simulation-Based Engineering Science (SBES). The panel released its findings in February 2006. Oden talked to Access' J. William Bell about the future of SBES and what its future practitioners will need.

This report comes on the heels of a number of other, relatively similar reports. What findings stand out here?

Simulation has really opened the horizons of research -- enabling new scientific discoveries and the design of new engineered systems -- to a point that was unimaginable even a few years ago. We're reaching a point in our history in which the capabilities of modern large-scale computing systems, the development of new algorithms, and interest in a list of new medical, manufacturing, and imaging technologies have made simulation indispensable in advancing science, in preserving the competitiveness of the United States, and maintaining the security and health of its citizens. We very strongly believe that this is a discipline that will impact every aspect of human existence, and it needs to be embraced by the major academic institutions, funding agencies, and other entities that are stewards of science and technology in this country.

So simulation-based engineering science is altering and expanding traditional approaches.

You know, it's interesting to me how the philosophy of science and how we try to learn about the behavior of the physical universe have blended in with many aspects of simulation and computation. Traditionally, humankind made hypotheses about the way nature behaved. Theory stands as long as no contradictions are observed. Conversely, we also made observations and then tried to develop or validate theories. Now, we can expand the realm of theoretical science through computer simulation, and we can augment and expand observation through simulation as well. What simulation-based engineering science is about is that the very tools that enable the expansion of theory and observation can be used for prediction, and prediction is the essence of engineering. This is the idea behind simulation-based engineering.

Traditional education in science and engineering presented a largely qualitative view of nature, with quantitative analysis reserved for only simple systems describable by meager calculations. New computer simulation has transformed science and engineering into very quantitative disciplines that provide amazing tools for predictions -- for virtual looks into the future at the way things work that obey scientific principles. Our educational systems are struggling to keep up with this transformation.

The report mentions several high-end possibilities that arise from rendering the qualitative as quantitative -- digital cities that simulate the operation of a whole city as a single system and predictive medicine, for example. Those are clearly very interdisciplinary. How do you capture and use that interdisciplinarity to its fullest?

This is a major issue. For the most part our current educational system, the organizational structures found in most universities, is not set up to handle the kind of interdisciplinary research necessary to really exploit the potential of these technological advances in simulation and computational science. Many universities are trying to find ways to accommodate interdisciplinary research. But there are many factors confronting the structure of the public university today that discourage interdisciplinary work.

And so what sort of changes need to be made?

I can point to the current budget structures in the universities I'm aware of. They are designed to compartmentalize knowledge into disciplines that made sense 20 or 30 years ago. Now, the way we support students and the way we fund academic departments provide little incentive or opportunity for faculty and students to cross the barriers from one intellectual discipline to another. In fact, there's a great deal of resistance to that. We have to find a way to break down these barriers, and it could be that redefining the structure of colleges will be necessary. It could mean developing new schools in computational science, as is being considered in a few places, or integrating simulation into much of the curriculum. It could also represent restructuring the reward systems in ways that encourage interdisciplinary work.

Would this not only change education but really, conceivably, change the way those graduate students, who will some day be running those departments and colleges, think about these issues in general?

Today's students are the glue that will hold together traditionally disjoint disciplines. They're in the position to take courses across several departments, to work as teams. They can be expected to provide the exchange between one discipline and another. [The faster] universities can find ways to give the next generation credit for these interactions, the faster the exchange happens.

There is also the tendency for students to be content with the limited knowledge they obtain in formal courses and standard degree programs. I often tell my students, "Look, you were supposed to have gotten that undergraduate degree in approximately four years. But once you graduate you're expected to learn much faster because you have the basic tools to teach yourself. There's nothing that you can't pick up in one or two years or even in a semester. And taking courses is eventually the most inefficient way for learning that there is. You need to reach a threshold in your education at which you have no qualms about diving into a subject and learning what it would take to use it or to make a contribution to it." This mindset about learning and research is essential for interdisciplinary research and for education in simulation-based engineering science.

Let's talk about the tools those students will need. Exploiting that potential requires a couple of things that NCSA is particularly interested in: powerful computers and the environments that synthesize that infrastructure and make it more useable. What sort of insights in selecting and using those computers and environments arise from looking at it from an SBES perspective?

Let's begin with software. Over the last decade there's been a strong tendency to develop incremental improvements to large, specialized codes. So-called legacy codes are very prominent in industry and in government laboratories, and because of their size and complexity, any improvements made are generally cosmetic and incremental at best. This can represent a serious detriment to advances in computational science. We believe that there are new algorithms, approaches, and hardware on the horizon that can only be fully exploited if we're not afraid to make a major investment in new software. In many cases, we need to retire some of these simulation codes that have been developed over the last 10 or 15 years and start anew.

Then, there is the enormous opportunity provided by the dramatic increases in speed and performance of computers and the added performance made possible by new algorithms. It will take some time to calibrate our own perceptions of computing to use these new capabilities effectively...

I personally feel that it's time to bite the bullet and introduce these new methodologies into the next generation of codes. There's a natural reluctance to change them because the people who use them are comfortable with them, know how to supply data to them, know what all the tricks are to get them to run, and so forth. But, sooner or later, if we're really going to exploit these opportunities to lift simulation to the next level; we're going to have to pay the price to do these major changes in a lot of our software.

The report certainly acknowledges that by talking about long-term support for high-risk research. When taking that approach, how do you walk the line between high-risk and good stewardship of the money that goes into it?

Well that's the $64,000 question. It's difficult to do. One argument is to look at the history of basic research in the U.S. Over and over again, the support of high-risk basic research has paid huge dividends. The same will be true with research on simulation science and technology. With regard to the sizable investment in software development, one must cope with the fact that the codes developed in universities come and go and often die, and students who helped write them come and go and move on. Rarely does one find the resources and the continuity in a university environment to put together really large-scale codes that do engineering and science problems. There are exceptions. But by and large, you have one or two individuals who take it on as their life's work to keep the code moving and to continue to add new features. Perhaps we can develop a research infrastructure to make this process manageable and maintainable.

Let's talk about the concept of "tyranny of scale" that's mentioned throughout the report. Why don't you describe that for us briefly?

The entire body of knowledge of the universe is partitioned into scales...Nowadays, the applications of interest that are arising in medicine, biology, material science, and manufacturing involve events where multiple time scales or spatial scales are encountered. One example is in nanomanufacutring. Today, semiconductor devices are manufactured that are 50 nanometers across. The manufacturing process may take place over seconds or a minute, but the behavior of these molecular systems may take place over a nanosecond. So there's a huge disparity between the timeframe of interest to a scientific application and the timeframe that can be covered by today's simulation tools. We want to know about an event over many spatial or temporal scales, but our current methodologies might be valid over one scale or perhaps two. This is easy to talk about, but almost impossible to do. Techniques for bridging scales are under scrutiny across many disciplines. These are problems that are virtually impossible to handle any other way than by simulation.