20.2 Computational Science

CP trained computational scientists ``accidentally'' by involving faculty, students, and staff in a research program whose success demanded interdisciplinary knowledge and work. Most of our students were at the Ph.D. level, although some undergraduates were involved through NSF REU (Research Experience for Undergraduates) and other research support. For instance, Felten made significant discoveries in new sorting algorithms (Section 12.7) while a physics undergraduate at Caltech. This work was awarded the prize for the best undergraduate research at Caltech during 1984. Felten is now in the Computer Science Ph.D. program at Washington University, Seattle.

We can ask the question of whether such interdisciplinary computational science can be incorporated in the academic curriculum as well as appearing in leading edge research projects? We can also ask if there is a role for a computational science at Ph.D., Masters, Undergraduate, and in K-12 precollege education.

We believe that computational science should be taught academically at all levels and not confined to research projects [Fox:92d]. We believe that there is an interdisciplinary core of knowledge for computational science. Further, this core contains fundamental issues and is far more than a programming course in Fortran, Lotus 1-2-3 or even more sophisticatedly, Express or MOVIE.

An education in computational science would include the basics of applied computer science, numerical analysis, and simulation. Computational scientists need a broader education than the typical physicist or computer scientist. Their training in basic computer science, and how to apply it, must be joined with an understanding of one or more application areas, such as physics and the computational approach to physics. Computational scientists will need a computer laboratory course so they become facile with the use of computers. These must be modern parallel supercomputers, and not just the personal computers or workstations now used for students in most universities. This broad education will only be possible if existing fields can teach their material more concisely. Considering a computational physicist, for example, the courses in applied computer science could substitute, for instance, advanced courses in quantum theory, or the parallel computer laboratory for an experimental physics lab. Thus, we could train a computational physicist with a reasonable knowledge of both physics and computation. Although the details of parallel computing are changing rapidly, the graduate of such an education will be able to track future changes. Computational science naturally links scientific fields to computer science. Here again, a specialization in computational science is an attractive option for computer scientists. An understanding of applications will allow computer scientists to develop better hardware and software. Computational scientists, whether in computer science or in an application field such as physics, will benefit directly from technology that improves the performance of computers by a factor of two or so each year. Their theoretical colleagues will not be assisted as well by technological improvements, so computational science can be expected to be a field of growing rewards and opportunities, as compared to traditional areas.

We believe that students educated in computational science will find it a rewarding and exciting experience, which should give them excellent job opportunities. Only a few universities offer such a degree, however, and often only at the Ph.D. level. Fledgling programs exist at Caltech, Cornell, Clemson, Denver, Illinois, Michigan, North Carolina, Princeton, Rice, Stanford, Syracuse, and U.C. Davis. The Caltech and Syracuse programs are both based in lessons from CP. These programs are diverse, and no national consensus as to the core knowledge of computational science has been developed. The NSF Supercomputer Centers at Cornell, Illinois, Pittsburgh, and San Diego have played a major role in enhancing the visibility and progress of computational science. However, these centers are set up outside of the academic framework of universities and do not contribute directly to developing computational science as an academic area. These centers, industry, the National laboratories, and indeed the federal government with its new high-performance computing and communication initiative, are all driving computational science forward. Academia is behind. Not only are there few computational science education programs, but few faculty who could teach such a curriculum. The poor job opportunities for computationalists in leading universities naturally discourages students entering the field and so again hinders the development of new educational programs. It will not be an easy issue to address, and probably only slow progress will be made as computational science gradually gains recognition in universities as a fundamentally exciting field. The inevitable dominance of parallel computing will help, as will the use of parallel computers in the NSF centers that have provided such a critical stimulus for computational science. Industry and the National laboratories already offer computational scientists excellent job opportunities, and the demand for such training will grow. Hopefully, this market pressure will lead to initiatives from within universities to hire, encourage, and promote new computational faculty, and educate students in computational science.

Consider the issues controlling the development of computational science in universities. As this field borrows and extends ideas from existing fields-computer science, biology , chemistry, physics, and so on-it will naturally face campus political hurdles as it challenges traditional and firmly held beliefs. These inevitable difficulties are exacerbated by administrative problems; many universities are facing a scenario of no growth, or even of declining funding and faculty size. This will mean that creation of new areas implies reductions in other areas. Computational science shares difficulties with other interdisciplinary areas, such as those associated with the growing interest in Planet Earth. The peer referee system used in the hiring and promoting of new faculty is perfect for ensuring high standards within the referees' domain of expertise. This tends to lead to very high-quality but isolated departments that find it hard to move into new areas. The same effect is seen in the peer review system used for the refereeing of scholarly papers and federal grants. Thus, universities find it hard to change, making it difficult for computational science to grow in academia. A key hurdle will be the development of some consensus in the community that computational science is, as we have asserted, fundamental and exciting. This needs to be quantified academically with the development of a core curriculum-a body of knowledge on which one can build computational science as an academic discipline.

There are two obvious approaches to filling the academic void identified as computational science. The boldest and simplest approach is to create an entirely new academic degree, ``Computational Science,'' administered by a new university department. This would give the field great visibility, and, once created, the independent department would be able to develop its educational program, research, and faculty hiring without direct interference from existing academic fields. Such a department would need strong support from the university administration to flourish, and even more so for its creation. This approach would not be easy to implement. There would be natural opposition from existing academic units for both good and not-so-good reasons. A telling critic could argue that a freestanding Computational Science program is premature; there is as yet no agreement on a core body of knowledge that could define this field. Students graduating from this program might find it hard to progress up the academic ladder at the vast majority of universities that do not have such a department.

These difficulties are avoided by the second strategy for computational science, which, rather than filling the void with a new department, would broaden the existing fields to ``meet in the middle.'' Students could graduate with traditional degrees and have a natural academic future. This is the approach taken by the existing university Computational Science and Engineering programs. For example, consider the two fields of chemistry and computer science. A computational scientist would graduate with either a Chemistry  or Computer Science degree. Later academic progress would be judged by the scientist's contributions to the corresponding base field. We have already argued that such an interdisciplinary education would allow the student to be a better chemist or a better computer scientist, respectively. Naturally, the chemistry graduate from the Computational Science program would not have received as complete an education in chemistry as is traditional for theoretical or experimental chemists. Some of the chemistry elective courses would have been replaced by computational science requirements. This change would need to be approved and evaluated by the Chemistry  faculty, who would also need to identify key chemistry requirements to be satisfied by computational scientists. New courses might include computational chemistry and those covering the basics of computer science, numerical analysis, and simulation. The latter set would be taught either by computer scientists or interdisciplinary Computational Science faculty. The education of a computational scientist within a Computer Science department could be handled similarly. This would have an emphasis on applied computer science, and a training in at least one application area.

In this scenario, a degree in Computational Chemistry  is equivalent to one in ``Chemistry  within the Computational Science program.'' On the computer science side, one could see degrees in ``Computer Science with a minor in Chemistry ,'' or a ``Ph.D. in Computer Science with a master's degree in Chemistry .'' At the academic level, we see an interdisciplinary program in computational science, but no separate department; faculty are appointed and students admitted to existing academic units. This approach to computational science allows us to develop and understand the core knowledge curricula in an evolutionary fashion. Implementing this more modest plan is certainly not easy, as one must modify the well-established degree requirements for the existing fields, such as chemistry and computer science. These modifications are easiest at the master's and especially at the Ph.D. level, and this is where most of the new programs have been established.

These seem to be very good reasons to establish undergraduate level Computational Science programs. We also need to create an awareness in the (K-12) educational system of the importance of computation, and the possibility of Computational Science degrees. In this way, more high school students may choose Computational Science educational programs and careers. Further, in K-12 one emphasizes a general education without the specialization normal in college. The breadth of computational science makes this very suitable for pre-college education. We also expect that high-technology environments-such as virtual reality front ends to an interactive fluid flow or other physical simulation on a parallel supercomputer-will prove to be a valuable teaching tool for today's Nintendo generation. Kids with a background in computational science will interact better with this modern computer environment and so learn more about traditional fields-for example, more about the physics of fluid flow in the sample simulation mentioned above.

Eventually, everybody will learn computational science-it will be part of any general education. When all students take two years at college of basic applied computer science-including but not at all limited to programming-then it will be natural to define computational science in all its flavors as an extension of these two years of base courses. Computation, like mathematics, chemistry, physics, and humanities, is essential in the education of tomorrow's scientists and engineers.