TESTIMONY BY DR. MALVIN H. KALOS DIRECTOR, CORNELL THEORY CENTER TO THE SENATE COMMITTEE ON SCIENCE, TECHNOLOGY, AND SPACE HEARINGS ON S. 272, THE HIGH-PERFORMANCE COMPUTING ACT OF 1991 TUESDAY, MARCH 5, 1991 Mr. Chairman, it is a privilege to be invited to comment on the "High Performance Computing Act of 1991" in the company of such a distinguished group of representatives of government, industry, and academia. I am Malvin H. Kalos, Director of the Cornell Theory Center, and a professor of physics at Cornell University. The Theory Center is an interdisciplinary research unit of Cornell University, dedicated to the advancement and exploitation of high performance computing and networking for science, engineering, and industrial productivity. As you know, the Theory Center is one of the National Supercomputer Centers supported by the National Science Foundation. The Center also receives support from the State of New York, and from industry. My career spans 40 years of work with computers as a tool in physics and engineering. I have worked in universities, industry, and as a consultant to the Los Alamos, Livermore, and Oak Ridge national laboratories in research devoted to the application of high performance computing to further their missions. We are witnessing a profound transformation of our scientific and engineering cultures brought about by the advent and adoption of high-performance computing and communications as part of our technological society. The changes, some of which we see now, some of which we easily surmise, and some of which we can only guess at, have had and will continue to have wide-reaching benefits. Our economic well-being and the quality of our lives will be immeasurably improved. I salute the foresight and leadership of the authors and cosponsors of this Bill, and the Administration. Senator Gore, Congressmen Hollings and Brown, and the President all understand the deep and positive implications for our future. We are also grateful for the support of Congressmen Boehlert and McHugh whose backing of our efforts at Cornell and for the entire program has been very strong. The Director of the Office of Science and Technology Policy, Dr. Bromley, has done essential work in translating the ideas into effective policy. The Federal Coordinating Council for Science, Engineering, and Technology (FCCSET) has, for the first time, brought unity into the Federal approach to high-performance computing. This is a well designed, well integrated program that shows good balance between the need to exploit advancing supercomputing technology, the need for very high performance networking, and the need to bring these new tools to the widest possible community through research and education. I will begin with some historical and philosophical remarks about science, using the history of physics, which I know best. Science is not a dry collection of disconnected facts, however interesting. The essence of science is the dynamic network of interconnections between facts. For a scientist, making a connection never perceived before can be the highlight of a career; the more distant the connection, the more it is valued. Our aim is to connect all we know in a seamless web of understanding. Historically, the greatest contribution of the greatest scientists have been such connections: Newton's between the fall of an apple and the motion of the Moon and planets; Maxwell's between the phenomena of electricity, magnetism, and the propagation of light; Einstein's leap of understanding connecting quanta of light and the photoelectric effect. These connections must be, to the greatest extent possible, mathematical and quantitative, not merely verbal or qualitative. Making these connections in a quantitative way remains at the heart of pure science today, but it has become harder as we try to probe into more and more complex phenomena, phenomena that cannot be analyzed by the mathematical tools at our disposal. There are many important examples in science that shed light on this paradigm. Chemistry is one of our most important sciences, one that contributes enormously to our grasp of the physical world and one whose applications lie at the core of our understanding of materials we use, wear, and eat, and of our health. The fundamental understanding of chemistry lies in quantum mechanics and electricity, well understood since the 1930s. Yet the translation of that scientific understanding into quantitative knowledge about chemical materials and processes- - polymers, chemical catalysis, drugs both harmful and healing, is very far from complete. Quite properly, chemistry is still largely an experimental science. But the power of modern supercomputers is transforming the face of chemistry at every level. We are coming to understand how electrons cooperate to bind atoms into molecules, molecules into larger structures, and to elucidate their structural, dynamic, and biological effects. However, extraordinary numerical precision, which can only be attained by very powerful supercomputers, is required for this vital work. Many other areas of science involve this kind of systematic connection among different phenomena at different scales of length or energy, including biology and medicine, the physics of materials, and astrophysics. The role of computation in linking disparate scientific fields is not a contemporary development. The early evolution of modem computers was dominated in the 1940s and 1950s by John von Neumann, who was also a great mathematician. He designed computers so that the very difficult questions that underlie such scientific and engineering problems as fluid flow could be explored and understood. Only later was it recognized that computers were also important business tools. The essential role of computers in science and engineering were well appreciated by many groups in the United States, including the national laboratories, and their use contributed very much to the development of nuclear weapons, fusion technology, and the design of aircraft. The use of computers in academic science and engineering evolved more slowly, partly because of the failure of many to see the possibilities, partly because the policies of the Federal government at the time discouraged scientists from participating fully. My own career was impacted negatively by these policies. It was the leadership of a few scientists, notably Dr. Kenneth Wilson, who created the modern climate of respect for the accomplishments and possibilities of computational science in the future of our country. The constructive contributions of the Congress and the National Science Foundation in creating the National Supercomputer Centers are noteworthy. That creation was, in a profound sense, the mark of the entry by the mainstream of American research into the era of computational science at the heart of science and engineering. It is also important to note that computational science is now an essential tool in experimental science as it is currently practised. The most advanced scientific instruments, optical and radio telescopes, particle accelerators, and computers themselves are studied, designed, optimized, and verified with computer simulation. Data collection is usually automated with the help of computers, and the reduction to comprehensible data sets and pictures may involve enormous computations. Exchange of large data sets and the cooperative work in understanding them will require very large computations and very heavy use of future high capacity data networks. Finally, in many cases, even reduced data are incomprehensible except when studied in the light of complex theories that can be understood only by simulation. Now the entire scientific and engineering community of the country has the opportunity to exploit these new tools. Many researchers are. Important new scientific discoveries are being made. New ideas and connections are seen everywhere. More important, students and young scientists, who are always the very heart of any important scientific change, are involved. They are coming to understand the techniques, the promise, and the limitations of computational science. Their knowledge and its applications are the most important products of our efforts, and they will carry the message to the rest of our society and to the future. It is they who will have the most direct impact upon industry in the United States. The science made possible throughout the nation by the resources of the Theory Center spans all scales of length and energy from the galactic through the planetary through the earth's crust, the behavior of man-made structures, of materials at the microscopic level, to the physics of elementary particles. From another perspective, it spans the traditional disciplines of physics, chemistry, mathematics, biology, medicine, all fields of engineering, and agriculture and veterinary medicine. Although I describe research at or made possible by the Theory Center, the other National Centers, at San Diego, Champaign-Urbana, and at Pittsburgh, can easily list an equally impressive set of accomplishments in pure and multidisciplinary science. It is perhaps unfair to cite a few at the expense of so many others, but the work of Stuart Shapiro and Saul Teukolsky on fluids and fields in general relativity is outstanding and has been recognized by a significant prize, the Forefronts of Large-Scale Computation Award. Their research comprises both the development of mathematical and numerical methods for the exploration of astrophysical and cosmological phenomena and the use of these methods to develop quantitative understanding of the formation of black holes and the characteristics of gravitational radiation. John Dawson of UCLA uses the Theory Center resources to study the unexpected results of the Active Magnetic Particle Tracer Explorer experiments. In these, barium and lithium were injected into the earth's magnetosphere, creating, in effect, an artificial comet. The observations contradicted existing theories and simulations. Dawson and Ross Bollens constructed a hybrid theory and simulation that models the observed effect. Henry Krakauer of the College of William and Mary uses a modern "density functional" theory of electronic structure to examine the nature of the electron-phonon interaction, known to be responsible for low-temperature superconductivity. The aim is to determine its role in high- temperature superconductivity. Work like this is being carried out throughout the world and will require the fastest parallel supercomputers of the future. Having them available to American researchers, including those who are not at major research universities, gives them and American industry a competitive edge. The research of Harold Scheraga and his group at Cornell into the three-dimensional structure of proteins shows an equally broad range of activity: the investigation of the fundamental interactions of the amino acid units with each other and with solvent atoms, the basic computational techniques needed to find the optimal structure, and the biochemistry of proteins. This is research that is particularly well suited to highly parallel computing, and will require, in the long run, the full use of future teraflops machines. Understanding the properties of the earth's crust is the subject of the research of Larry Brown and the Consortium for Continental Reflection Profiling (COCORP). This national group uses the supercomputers to reduce, display, and interpret the huge data set that is gathered by seismic probing (to 30krn or more) of the continental crust. I cited earlier the fundamental importance of scientific computing in enabling the connections among different phenomena within scientific disciplines. Even more important is its role in permitting quantitative connections among different disciplines, that is, in supporting multidisciplinary research. Every one of the large problems that confront our society, and to whose solutions we expect science to contribute, is in some sense a multidisciplinary problem. For example, issues of the environment involve many sciences -- chemistry, physics, engineering, fluid flow, biology, and materials. Medicine is equally demanding in its call upon diverse science. As we have indicated, biochemistry and its relations to chemistry and physics plays a central role in medicine. But other areas are important as well. As part of my oral presentation, I will show a video of a supercomputing study of the uses of ultrasound in the treatment of eye tumors. The building of modem prosthetic devices uses many resources of computation, from the reduction of CAT scans to the computational optimization of the mechanical properties of the devices. Understanding blood flow in the heart requires a mastery of fluid dynamics of viscous media plus the knowledge of the elastic properties of the heart and its valves. Bringing the knowledge from these fields together to make quantitative predictions about the effects of some technological or regulatory proposal is a difficult undertaking, one that is utterly impossible without the use of computational modeling on high- performance computers. Computational modeling is the indispensable natural language of quantitative multidisciplinary research. An outstanding example of such work is that by Greg McRae of Carnegie Mellon University. He uses supercomputers and supercomputer-based visualization to explain from basic chemistry, fluid mechanics, meteorology, and engineering the scientific effect that underlie the development of air pollution in the Los Angeles Basin, and the probable effects of fuel changes and regulatory procedures. His results have been used to influence regulatory policy constructively. The Global Basins Research Network (GBRN), a consortium directed by Larry Cathles of the Geology Department of Cornell University and by Roger Anderson of Columbia University's Lamont-Dougherty Laboratory and which includes eight academic and 11 industrial partners, has as its goal the multidisciplinary understanding of the chemical, physical, and mechanical processes that occur in a sedimentary basin such as the one in the Gulf of Mexico below Louisiana. They have assembled a composite database of the observations of the basin and are using computational modeling to explain the data. But simply the collection and display in a coherent visual way has led to new and deeper understanding of the geology. The outcome of this understanding is very likely to improve oil recovery world-wide. I will also show a video clip of a visualization of the data set that was prepared jointly by the Theory Center and the GBRN. It is important to note that this research covers a wide range of partners, geographically dispersed, and the that the medium of information exchange is usually visual. High- performance networking is essential to the GBRN and to similar scientific enterprises. Another important development is the establishment at Cornell of the Xerox Design Research Institute, with the participation of the Theory Center, the Computer Science Department, and the School of Engineering. Directed by Gregory Zack of Xerox, and involving researchers from Xerox centers nationwide, the aim of the Institute, quite simply, is to improve Xerox's ability to bring better products more quickly to market. The techniques are those of computational and computer science. A vital aspect of the research is the development of methods whereby the geographically separate centers can effectively collaborate. Again, high-performance networking is key. As our reach extends, the necessary partners required to carry out important collaborative research will rarely be found at one institution or even in one part of the country. Essential experimental devices or data bases may exist anywhere. Rapid, concurrent access is essential, and at higher demands in bandwidth. The NREN is necessary for the full growth and exploitation of the scientific, technological, and educational implications of computational science. The GBRN and Xerox examples indicate how the greatest potential is for industrial use. The supercomputing community will soon find itself at a major crossroads -- where the increases in performance needed for the fulfillment of our scientific mandate will demand parallel architectures. To exploit these new machines, a major retooling of software and algorithms will have to take place. This is not a trivial undertaking, yet it must be started very soon if we are to make progress on the Grand Challenge problems in the mid-1990s. The High-Performance Computing and Communications program will offer us an essential opportunity to bridge the gap between today's high performance vector machines and tomorrow's highly parallel systems. I have emphasized how science and its application to societal problems are communal activities, activities that involve, more or less directly, the entire scientific community. Bringing to bear the transformation made possible by computational science in the most complete and positive way requires that its techniques and strategies be learned, used, and shared by the widest possible group of researchers and educators. That means advancing the art, acquiring the best and most powerful tools of hardware, software, and algorithms, and coupling the community in the tightest possible ways. The "High-Performance Computing Act of 1991" is a vital step in that direction.