Opinion: Computers and the New Face of Science
Working on my Master's thesis in the early '70s, it was just an exhilarating experience to run my Fortran code on one of the fastest machines in Europe, a Univac 1108, at 1 MFlop/s.
Fifteen years later, when I started at CSC-Scientific Computing Ltd.-on the arrival of the Cray X-MP4, newspapers ran big headlines of the astounding machine approaching the 1 GFlop/s barrier. Another fifteen years later grid technology makes the 1 TFlop/s capacity commonplace, and the top-end performances close on 100 TFlop/s.
The consequences of Moore's Law for raw computing power keep amazing even the most cynical professional. To put it in concrete terms for an actual computational science project: a calculation that took in 1980 one full year to complete, can now be done in under five seconds.
How does this translate our progress in science? For example, has it led to an exponential increase in our understanding and knowledge in the physical sciences, where computers have been widely applied? There is obviously no simple answer to this question, and the profound analysis of the impact of computing remains to be done. Two developments are nevertheless easy to spot.
First, the number of scientists and engineers who routinely use large-scale computing has grown enormously, but not perhaps exponentially. Computational science and engineering, for example in the physical science realm, has established itself as a major undertaking. One might even worry that it could become too successful and start to divert resources and talent away from experimental and empirical activities, the bedrock the scientific enterprise.
Secondly, the complexity of the problems that can now be meaningfully tackled with the help of computing has increased dramatically. In my own field, materials physics and nanoscience, it is easy to list spectacular examples.
Twenty years ago, it was a year's project to calculate from first principles the properties of a simple substance such as perfect, crystalline silicon at zero temperature. Now, in a similar time, one can predicatively simulate, again from first principles, the behavior of defected, polycrystalline silicon under the attack of different reactive chemical etchants at high temperatures and pressures.
Some of the key words in today's efforts are nonlinearity, non-equilibrium, multi-scale, multi-science. However, these words also imply that the problems are tough, and get exponentially tougher as we increase the number of degrees of freedom. Even with Moore's Law, progress in understanding is sometimes painstakingly slow. Computational scientists certainly have their work cut out for them. Of course, the challenges are not limited to physical sciences. Systems biology, earth system sciences, linguistics and network studies are just some examples of hard but important problems.
The above arguments are, however, somewhat simplistic. They see science as a passive beneficiary of Moore's Law and computer industry.
It is my firm belief that in fact science is increasingly about the collection, organization and transformation of information, not just computing and number-crunching. Scientific research is gradually turning from its reductionist traditions to more integrative, "systems" approaches. In many areas, vast amounts of data are being generated, collected, stored and shared. Modelling and simulation, the traditional realm of computational sciences, are being integrated to data curation and mining, and also to data generation in experimental and observational activities. Information technology will weld together all sorts of pieces of information and data, utilizing archives and search engines, and transcending geographical borders.
High-performance computing continues to play an important role. New scientific disciplines, separately and in multidisciplinary complexes, will adopt the computational paradigm. The topend machines are crucial for the cutting-edge complex problems. They are also the source for inspiration. But it is the interwoven, hierarchical cyber-infrastructure as a whole that will carry the day. Computing technology and computer science, hardware, software, algorithms and theory, will be the great enablers. Without too much hype, it is fair to say that applied computer science will have a similar role as mathematics has had from the seventeenth century, providing the crucial apparatus for scientific discovery.
- Risto Nieminen
This essay originally appeared in CSCnews, a magazine from CSC, Finland's state-owned center for computational science located in Keilaniemi, Finland, near Helsinki. Reprinted with permission.