More information: personal opinions of Vicki Johnson

Scientists

My interest in N-body simulations began in 2002 with a small consulting contract to develop software to help undergraduates run N-body simulations accelerated by a GRAPE card. I have since continued to improve the NBodyLab software on my own demonstration web site as time permits. Since I am a software developer and not a scientist, I sought expert advice to choose and implement the core N-body simulation software. I received help from many scientists. Most helpful were Sverre Aarseth, Jun Makino, Peter Teuben and Bruce Elmegreen. Several weeks after we met, Piet Hut gave me a ticket to join him in Japan where he introduced me to Makino, for which I am grateful. Hut proposed undertaking several large projects, including co-authoring a book and starting a GRAPE-related company. I interacted with Hut for a few months but for me it was an unproductive and unpleasant experience.

Books

Scientific Simulations with Special-Purpose Computers--the GRAPE Systems
by Junichiro Makino and Makoto Taiji
John Wiley & Sons, 1998

This short book about the development of the GRAPE systems is technical but quite readable and exciting. It is a personal account of a remarkably focused effort, led by the young Japanese authors, to develop faster hardware to advance their particle simulation research in astrophysics (Makino) and proteins (Taiji). The design was applicable to gravitational N-body and molecular dynamics simulations. A variety of astrophysical studies using GRAPE's are presented in the book, including planetary formation, star clusters, galactic nuclei, galaxy interactions and galaxy formation. The authors (particularly Makino) have many times won international Gordon Bell Prizes for outstanding performance of their GRAPE supercomputer designs, competing against supercomputers costing tens and hundreds of millions of dollars (see their awards). The book is written so clearly, they make it look easy. I liked best the software sections, including descriptions of how the host controls the low-level GRAPE hardware/software, discussions of I/O interfaces and host ports, and the application software interfaces.

The authors are now busy developing next generation GRAPE systems, but I wish they would take time to write a second edition describing GRAPE developments from 1998-present. The papers submitted for the Gordon Bell competitions of 1999, 2000, 2001 and 2003 (links) tell part of the story of this evolving technology.

See also Makino's www.astrogrape.org

Gravitational N-Body Simulations: Tools and Algorithms (Cambridge Monographs on Mathematical Physics)
by Sverre Aarseth
Cambridge University Press, 2003

This comprehensive book is written by the best known developer of direct summation N-body codes. Aarseth states that the main purpose of the book is to provide algorithms for direct N-body simulations based on his personal experience. He describes the interplay of his own and others' research over the course of four decades. It contains many references to the GRAPE, which Aarseth refers to as "the simulator's dream machine". The first lesson I learned was that N-body simulations are much more than just the integrator. The book's organization is especially helpful for understanding practical software engineering aspects and N-body specific techniques, including sections on program organization, initial setup, accuracy and performance, algorithms for neighbor schemes and several regularization methods for two-body motion and compact subsystems. For readers who want to understand what types of scientific studies can be tackled with N-body simulations, there are chapters on simulations of star clusters, galaxies and planetary systems. The book offers good advice to newcomers: "it is much easier to take over one of the large codes" because it takes many years to master the combination of complicated methods required for accurate simulations. Aarseth's source codes are freely available. Aarseth's software and documentation, this book, and his professionalism led me to base the current version of NBodyLab on his NBODY4 software. [Follow-on note from Sverre Aarseth: Several Cambridge colleagues and students, as well as long-term visitors have made significant contributions to the NBODY4 package over many years.]

My wish list for future editions: include more graphs and illustrations, and an expanded version of the section on "Test Problems" for test cases and guidance on testing large simulations.

Aarseth, an avid mountain climber and wildlife enthusiast, dedicated his book "To the world's wild and magical places."

See also sverre.com, Aarseth's Cambridge web page, Aarseth's 1999 NBODY1 to NBODY6 survey paper, and the related 1998 Brouwer award.

The Gravitational Million-Body Problem
by Douglas Heggie, Piet Hut
Cambridge University Press, 2003

Although both books were published in 2003, Aarseth’s book above has over 40 pages with references to the GRAPE, and some 25 pages with references to the GRAPE-6. Only 12 pages in the Heggie & Hut book refer to the GRAPE, and it presents old GRAPE-3 (1993) and GRAPE-4 specifications, mentions a 2000 paper on the GRAPE-5 and says of the GRAPE-6 only that it is "coming". Yet in 2000, the GRAPE-6 was mature enough to win a top award in the Gordon Bell international supercomputer competition.

I expected extensive and up-to-date coverage of the GRAPE since co-author Hut has publicized his involvement in the GRAPE project on his own web and elsewhere. For example, in 2003 Hut was introduced in a talk at Columbia University as "the creator of ... the fastest parallel-processing computer in the world" and then Hut showed a picture of himself with a GRAPE card, talked about the GRAPE hardware and software "we have developed," and did not mention Jun Makino, as far as I could tell from the webcast.

In the preface Heggie & Hut mention difficulties of "attributing every advance to the originator", and "the question of who did what," even though they describe the field as "not a large one". These statements diminished my confidence in the book. If the attributions might not be correct, what about the rest?

Unlike the two books above, which focus on the personal accomplishments of the authors, the Heggie & Hut volume seems to be a survey book. I looked first for guidance on choosing simulation software. Appendix A gives pointers to codes. However, there are two easily-overlooked references at the beginning of Appendix A. The first one says: for a FORTRAN code see Binney & Tremaine 1987. But the code in Binney & Tremaine is an Aarseth code - why didn't Heggie & Hut credit Aarseth and/or instead point to Aarseth's current list of FORTRAN codes, which range from basic to advanced? (Binney and Tremaine generously acknowledged Aarseth.) A second reference in Appendix A says: "NBODYx A series of N-body codes" followed by a long ftp URL. But the URL is not identified as being Aarseth’s code, and the URL points to just one program, not a series of codes. When I read Appendix A, I did not pursue these two references, as they seemed obscure and their significance was not clear. Months later, after becoming familiar with Aarseth's work through his own book, I looked again at Appendix A and recognized Aarseth's association with these two references. Also, Hut and Heggie do not mention that Aarseth has N-body codes for GRAPE's, which have been many years in development.

In these cases and others, I found the Heggie & Hut style of presentation sometimes obscure, uneven and not helpful in my search for simulation software for NBodyLab. Each chapter is about ten pages with many graphics and sidebars, like a magazine, so the topics can be quickly skimmed, leaving the math and theory for later.

Acknowledgements

Vicki is grateful for assistance from computational astrophysicists, including Prof. Jun Makino, Dr. Sverre Aarseth, Dr. Peter Teuben, Dr. Bruce Elmegreen, and scientists at RIKEN. Makino has provided a GRAPE-6a for NBodyLab.org. An earlier version of NBodylab.org incorporated an MD-GRAPE2, loaned to Vicki by the Japanese research institute RIKEN.

Please use supercomputers for peaceful applications and oppose weaponization of space.

Vicki Johnson and Prof. Jun Makino at the Supercomputing 2003 conference. As a small protest against displays of warfare at these conferences, including soldiers in camouflage and military vehicles, Vicki brought a peace flag to SC2003. Many supercomputers are used in US weapons laboratories. If you are a student interested in computational science, please pursue peaceful lines of research, such as environmental modeling, biomedical research and simulations of natural phenomena.

Joint Vision 2020 outlines the US military vision for control of space. The planners state the world is becoming divided into the "haves" and "have-nots", which will create global conflicts; therefore, the US must weaponize space to maintain US superiority. Organizations such as the Global Network Against Weapons and Nuclear Power in Space oppose these views.