Dr. Raymond L. Orbach
Director
Office of Science
U.S. Department of Energy

RAND Graduate School
Santa Monica, CA
June 22, 2002

It is my great pleasure to congratulate the graduates of the RAND Graduate School on the successful completion of their doctorate. You join a distinguished group of policy analysts who are world leaders in their fields. Your own research encompasses health care quality measures in developing countries; family relationships and their measures; effectiveness of systems of care for at-risk children and their families; the transition of the Italian Army towards an All-Volunteer Force; storage and utilization of art collections in U.S. museums; hands-on science and student achievement; financial and structural consequences of “carve-out” health care services; patient experiences with healthcare in multi-cultural settings; and the consequences of class-size reduction policy in California. It is a remarkably board set of inquiries into the consequences of social policy. Your methods of inquiry are quantitative within a policy framework, something for which Rand is justifiably famous. You are to be congratulated on your achievements.

I have stressed the quantitative element in your studies. During my scientific career, computers were developed from the now “creaky” IMB 701, on which I did my thesis research in the wee hours of the morning, to the new ASCI machines which fill rooms the size of football fields, and use as much power and cooling as a small city. The speed of these latter, massively parallel machines, is now measured in “teraflops.” For the purpose of comparison, a “flop” is a floating point operation per second. Don’t worry about what a floating point means; just think of it as an algebraic operation on your hand held calculator. A “tera” means a trillion, or in our everyday language a million millions. So when I say 2 teraflops, that is a machine capable of 2 million million operations in one second, something that would take your hand held calculator 57,000 years to accomplish.

The reason I made you endure this bit of mathematics is that the teraflop era is upon us. The DOE civilian computer has a theoretical peak performance of 5 teraflops, and the so-called ASCI White computer at Lawrence Livermore has a peak performance of 12 teraflops. This would translate to 342,000 years on your hand held calculator, but now it takes just 1 second!

What does this mean for the U.S. scientific community, and for you? We are now in an era where computational simulation can inform our approach to science, and I believe the social science and humanities. We are now able to contemplate exploration of worlds never before accessible to mankind. We have used computers to solve sets of equations, physical laws too complicated to solve analytically. But now, we can simulate systems to discover physical laws for which there are no known predictive equations. This means that we will be able to model physical or social structures with hundreds of thousands, or maybe even millions, of “actors”, interacting with one another in a complex fashion. The speed of our new computational environment allows us to test different inter-actor (or inter-personal) relations to see what macroscopic behaviors can ensue. Thus, we may be able to use simulations to determine the nature of the fundamental “forces” or interactions between “actors.”

This approach to understanding complex systems is to be thought of in the same vein as experiment and analytic theory. In science of the 21st century, simulation and high-end computation are equal partners with theory and experiment.

Scientific leadership, the basis for our economic, physical, and intellectual prosperity depends on this triad, our being first in each component.

Alas, we have lost the lead in scientific computation. We are now second, and if we continue to dally, we will be third in this critical third leg of the triad.

Three months ago the Japanese announced operation of the earth simulator, a vast computer with a peak speed of 40 teraflops, 3 times the fastest U.S. machine’s theoretical peak performance. But there is more: the architecture of their machine was structured to solve a class of problems, climate change to begin with. Their sustained speed was close to 20 teraflops, or about 50% efficiency. The fastest U.S. computers exhibit an efficiency of approximately 10% because the United States has adopted a “one size fits all” philosophy. By some estimates, this leads to a lag of a factor of 40-50 for the National Center for Atmospheric Research’s climate model.

The full story is worse. Not only can climate change models run at these efficiencies, the Japanese have very recently shown similar capacities in fluid and plasma dynamics calculations.

Bluntly put, we are out of business in some critical areas of computational science. The earth simulator works on a grid 10 km on a side for climate models, while U.S. computers do no better than 100 km on a side. This means that U.S. simulations average over microclimates — mountains and coastal effects, river flow, cloud and storm systems, or hurricane storms. Averaging means that our models cannot credibly predict large scale fluctuations in climate change, critical for long-term drought and flood predictions.

What does it mean to lose scientific leadership, to be #2? There is a qualitative difference for large scale computational simulations as compared to conventional calculations – they do not travel. The great scientific discoveries will take place in Kanazawa, Japan, and not here.

What can we do about it? Should we do anything about it? I believe our country cannot afford to be second best, to be “good enough.” Our economy, our intellectual environment, literally our national security depends upon our scientific primacy. What very bright student will want to enter a field where we are #2? And what about Europe? They are already purchasing Japanese machines. Are we to be #3?

This does not only apply to climate change simulations. Also at risk is our leadership in nanoscience, accelerator design, astrophysics, combustion, materials science, and fusion energy research.

Biology will not be far behind: simulations can significantly aid our understanding of the systematics of cellular function. And the social sciences – Are systems approaches through simulations at hand?

The United States needs to face up to our dilemma, a construct of our own doing. We must move away from the pretense that a single computational architecture can work for all of our scientific needs.

The United States must embrace the concept of a diversity of computational architecture to address a variety of applications. We need to begin with the science, and scientists, starting with the problems which we wish to address. Bring to the table the computer scientists, the applied mathematicians, those who are good at algorithms, together with the chip makers and computer architects, those who can produce the computers devised for the scientific problem at hand.

The result will be machines enabling us to solve scientific and social problems of great importance. We shall be able to investigate systems of great complexity, and understand predictive laws for their behavior. We will free ourselves of the bounds of one-at-a-time processes, and learn the rules of collective behavior on a scale previously unknown. The opportunities are immense. We cannot afford to be second or third in this pursuit. We have the will and the capacity to fix this.

As a product of the Sputnik generation, I can personally attest to the vigor and vitality of the U.S. response. It is now incumbent upon us to repeat this dedication in this new era of computnik, to regain our scientific leadership and primacy.

You RGS graduates are first class. We need to provide you with a computational platform of comparable quality.

Graduates, I congratulate and salute you. Our hearts and very best wishes are with you. Good luck and God speed.