sign in sign up
<<
Home , Computational physics, Quality (physics)

COMPUTATIONAL the field of massively parallel . However, one should not be too critical as there are efforts under way to identify new Overcoming the Isolation paradigms. Consider, for instance, the field of computational . -gas Berni Alder, Professor Emeritus at the University of California, Berkeley, who pioneered methods have not been successful in simu­ the use of methods in hydrodynamics, took the opportunity as a keynote lations as the largest computers can only speaker at PC’94 to reflect on why , in spite of its successes, handle 106 for 10-8 seconds. Monte remains somewhat isolated in academic circles. Carlo methods have been shown to be very effective, but only at low Reynolds numbers. Although there remains the problem of materials research. The situation arises be­ Using a hybrid approach that combines the handling the interaction between particles in cause an undergraduate curriculum in com­ of particle scattering at boundaries simulating, for example, Schrödinger’s equa­ putational physics has not been defined, and with continuum hydrodynamic methods away tion, Monte Carlo methods represent a great because there is some confusion as to from the interface, it is becoming possible to success for computational physics. However, what a computational represents. tackle more relevant high-Reynolds number other major challenges have not yet been Claudio Rebbi, in summarizing below how problems. The crucial Issue is to match the tackled successfully. They Include three- a curriculum can be developed, goes a Navier-Stoke’s formalism with particle-like dimensional turbulence where there has long way to explaining what computational by ensuring that conservation been little progress for 60 years as distance physics tries to achieve. However, the com­ laws are satisfied. The first results for a mag­ and time-scales are very large, and pattern munity itself is partly responsible as it has not netic read-write head floating just above a recognition where the hopes for massively focussed sufficiently on theoretical aspects rapidly spinning magnetic recording disc parallel computation have not materialized. and the development of , but ten­ show that one can realistically simulate the There will be little progress in pattern recog­ ded instead to be absorbed by advances in development of fluid-flow Instabilities. If nition until much more is built into computer technology. Hence, in spite of con­ Rebbi’s Ideas are widely adopted it should be computing nodes, thus mimicking neural net­ siderable discussion and development, few possible to develop other paradigms, and works. The folding of DNA molecules and qualitatively new algorithms have emerged in envisage further successes. why it takes place so quickly are poorly understood; a new paradigm is needed as Predicting three-dimensional protein struc­ random folding clearly does not work. People tures. Attempts to reliably predict the struc­ are groping for ideas, although it seems that tures of proteins are in their infancy. One the presence of water is crucial in helping approach is based on the idea that native pro­ guide the DNA molecules through metastable tein structures are in the state with the lowest free energy. The interaction between the pro­ intermediate states (see figure). tein and its water solvent is known to play a What is perhaps sad is the fact that com­ dominant role in the stability of the protein. putational are often not accepted This interaction can be introduced by assu­ as equal partners in the academic world in ming that it is equal to the sum over all atoms of the atomic solvation parameter times the acces­ spite of their considerable achievements sible surface area. Braun and Mumenthaler reported at PC '94 that adding this simplified semi- which include, for example, calculating the empirical energy surface term to the standard conformational energy term and carrying out a mass of the proton from first principles (the minimization using Monte Carlo simulations gave folded structures with the correct three-dimen­ challenge now Is to calculate the mass of the sional topology. This is illustrated in the figure for a small test protein having three hydrogen atom). They have tended to seek packed helices. One of the low-energy simulated structures (the structure on the left) closely homes outside mainstream physics in areas resembles the structure on the right determined by NMR. So its seems that efforts to predict such as mechanical engineering and applied protein structures are moving in the right direction.

The majority of applications of computa­ Teaching Computational Physics tional physics involve numerical computa­ tions. The discipline must thus be taught by Claudio Rebbi of Boston University, in arguing that the teaching of computational using a fair amount of . But physics must become an part of , offers some general prin­ there is much more to computational physics ciples as a guide. than the study of algorithms and numerical methods. Computational physics is the art of formulating and solving physics problems on the computer. It is this art which we must COURSE PROJECTS IN COMPUTATIONAL PHYSICS convey to students. To accomplish this we Calculate the thermodynamic properties of crystals. must keep the teaching of computational Solve an equation for the Internal energy of a crystal based on a quantized oscillator physics anchored to the treatment of actual and the Debye approximation for the frequency of the normal modes of vibration. This physics problems. These can provide the introduces numerical integration. Going beyond the Debye approximation to use a more background and motivation for the introduc­ realistic dispersion formula demonstrates that the computational as opposed to the ana­ tion of algorithms. The consideration of a lytical perspective simplifies the calculation. concrete problem may, moreover, give the opportunity to illustrate the difference bet­ Calculate the trajectories of N-bodies subjected to mutual gravitational interaction. ween an analytically minded approach and A prototype project for the general class of molecular dynamic simulations. Introduces a computational approach, and how by various algorithms for solving ordinary differential equations with initial value data. changing one’s perspective one can fre­ Leads to a discussion of stability and the need to employ more sophisticated tech­ quently achieve a simpler and more efficient niques, such as the adaptive time step. With careful choices of initial data one can solution. Finally, as code is developed for explore chaotic behaviour. Also Introduces advanced computing technology (e.g., imple­ a teaching project, the students can refine mentation on massively parallel computers). programming skills and be exposed to ad­ Calculate the energy of a superconducting vortex in the of Ginzburg, Landau and vanced computing technology. Abrikosov. A Matrix Minimize a function of the several variables that parameterize the field configuration. We therefore essentially have a framework Introduces fundamental algorithms (e.g., method of conjugate gradients). for teaching computational physics consisting Calculate the energy levels of a bound state of a heavy quark and antiquark with a of a matrix of physics problems, numerical potential that has a Coulombic component and a linear component. methods and computing technology. Such a Introduces boundary value ordinary differential equations. Goes beyond learning the framework provides an excellent educational algorithms to checking the predictions of the model against experimental data. platform, while also reflecting the integration of knowledge that plays a fundamental role

158 Europhys. News 25 (1994) in progressing algorithmic research. For we often find that developing more efficient ways to carry out some computation rests on 1997 IUPAP Computational Physics Conference understanding the physics of the phenome­ The International Union of Pure and (IUPAP), in reappraising its role, non under study. In my own field of research, decided at its General Assembly last year in Japan to explore the possibility of forming a one of the most demanding computations Special Commission for Computational Physics. A working group was set up which will consists in the calculation of quark propaga­ shortly present a proposal to IUPAP’s Executive Committee for the new Commission (the tors in the background of a non-uniform Executive has the authority to establish the Commission). gauge field, which must be carried out many It is proposed that the Commission should adopt a broad times to simulate quantum fluctuations. In perspective by promoting “the exchange of Information mathematical terms, the calculation consists and views among members of the International community in solving a very large system of linear equa­ of physicists in the area of of computational studies of pro­ tions (with a number of variables that can blems related to physics including: reach into the millions), with a matrix of coef­ - numerical and symbolic models and algorithms ficients which is very sparse, but non-uniform for simulating physical systems; and with poor- condition numbers. - computational control and data processing Ideally, one would like to apply multigrid in experiments; methods or other acceleration techniques. - computing environments; What one sees is that the ability to implement - the physical basis of computing machinery.’’ David Andersen from Cali­ such methods is closely coupled to under­ The new Commission also sees itself as the body that fornia who chairs IUPAP’s standing the physical properties of the lowest sanctions international conferences in the field of computa­ computational physics work­ eigenstates of the quark field, so algorithmic tional physics. The working group has therefore agreed ing groups. progress and progress in unravelling the with representatives of the various regional physical socie­ physics go hand in hand. Issues of physics ties that a IUPAP-sponsored conference should be held every three years, with the joint and computing are not separate, but should EPS-APS conference alternating during the intervening years. Consequently, the next be seen as complementary. joint EPS-APS conference (Physics Computing ’95) will be held next In Pittsburg, followed Putting into Practice by PC'96 in Cracow with the IUPAP event in 1997 in Beijing. There is an emerging consensus that computational physics should be taught at the undergraduate level. This is a difficult SUPERCOMPUTING CENTRES challenge. An undergraduate course on ad­ vanced computing in physics based on the matrix approach was given for the first time A Variety of Approaches last year at Boston University as part of Supercomputing facilities in the various east and central Europe are evolving at a coordinated set of new undergraduate different rates in the various countries, with Poland adopting the most adventuresome courses on supported by attitude. the US National Science Foundation. Some of the projects used in the course are sum­ POLAND acquired a 4-processor Cray Y-MP EL this marised in the insert. Poland’s main supercomputing centre is year. The KBN recently signed an agreement My experience in teaching the course was the Academic Computing Centre based in with Cray for 6.5 M$US to be spent over 2 that I assumed too much about the students’ Cracow’s Institute of . It years to upgrade supercomputing resources. background preparation. This is a real diffi­ was established some 20 years ago and now culty because teaching computational phy­ serves 11 institutes in Cracow, but following a CZECH REPUBLIC sics must proceed concurrently with the rest recent policy change it will see its user com­ The Institute of Physics of the Czech of the curriculum. One way to help solve the munity grow to include other institutes. Most Republic’s Academy of Sciences acquired a problem is to use the tools whose utilization investments have been funded by the State Cray Y-MP EL in 1992. Owing to financial we wish to teach as the basis for making Committee for Scientific Research (KBN) constraints, it has proved difficult to develop many mathematical concepts accessible. For which runs the country’s research grants a large user base (users are concentrated in instance, the notion of a Laplacian operator scheme (see EN, April 1994). The most re­ physics, with most coming from universities). can be introduced using image manipulation cent investment is an agreement signed a The Institute recently formed a consortium to procedures. The idea is to start with pixels few months ago to purchase one of Convex’s seek support from the Czech government’s that form an image and blur the Image by latest generation of massively parallel super­ Foundation for the Development of Science replacing their numerical values with the computers (a 16-processor machine in the to build up a National Centre for Supercom­ averages of their neighbours. Exemplar series that offers scalable parallel puting that would be in a position enlarge its A more serious — and I believe general — processing with from 1 to 128 central pro­ client base and support by working for indus­ problem is to make room for new courses cessing units). The Centre already has three trial companies. These could include US in an already overcrowded curriculum. Ex­ Convex machines, the first of which — and firms that have been contracted by the go­ posing students in general courses to com­ the first Convex in central and eastern vernment to upgrade Soviet-designed nu­ putational projects is not a solution because Europe — was acquired about four years clear reactors. The immediate need is to there is much more to the methodology of ago. Some six other Convex computers are increase the number of workstations and computational physics than seeing the re­ distributed around the country (in Torun, improve software resources. However, a pro­ sults of its applications. However, by cleverly Warsaw University, the Oil and Gas Institute ject along these lines was not approved this integrating the power provided by modern in Cracow, the Institute of ) computational tools in the overall curriculum and two are in industry. A deskside Y-MP EL Cray . we may be able to help students acquire Warsaw’s Centre for Scientific Supercom­ more easily all of the many notions that are puting based at the Institute of Applied Ma­ indispensable for a physicist. thematics, Computer Science and of the Polish Academy of Sciences is equip­ This text is taken from an invited talk given ped with a so-called “baby” Cray (an 8-pro- by Claudio Rebbi at the 6th Joint EPS-APS In­ cessor machine in the Cray Y-MP EL series ternational Conference on Physics Computing with up to 32 CPUs). Warsaw University also (Lugano; 22-26 August 1994) and published in the proceedings. has a computing centre based on an IBM 3090 which was delivered in 1990 to form PC’96 : 7th JOINT EPS-APS an European Academic Research Network INTERNATIONAL CONFERENCE (EARN) node — the first in east and central Europe — as part of IBM’s Central European ON PHYSICS COMPUTING Academic Initiative. There is a computer Pittsburg, PA, USA June 1996 centre attached to the Polish Academy of Contact: D. Barnes, LANL Sciences Institute of Biology in Poznan which

Europhys. News 25 (1994) 159