|
In conversation with
John Guckenheimer
by Hinke Osinga
University of Bristol, UK
|
At the recent SIAM Conference on
Applied Dynamical Systems John Guckenheimer, chair of the Activity
Group on Dynamical Systems and former SIAM president, talked to Hinke
Osinga about the role of software in dynamical systems, the growing
demand for tools to investigate complex dynamical systems, and the roles of
mathematics and computer science departments in supporting the
development of numerical algorithms and software.
"From the beginning I was interested in real applications in
dynamical systems and I recognized that the computer is a really
valuable tool in doing that."
John's research has always been closely tied with computational
results. However, he feels that research on dynamical systems
theory involving algorithm development has often fallen into gaps
between disciplines. Mathematics, computer science and applications
disciplines have not fully embraced the challenges of providing the
best possible computational tools for the simulation and analysis of
complex systems. This was the theme of his SIAM Past Presidental
address (SIAM News, Volume 32, No. 8, October 1999).
Curriculum vitae of Professor John Guckenheimer
John got his undergraduate degree in 1966 from
Harvard and a PhD
from Berkeley in
1970. As with many mathematicians, there is an interesting story
behind how he decided to do a PhD in Berkeley. His plan was to join
the Peace Corps going to Nepal and so he went to Hawaii for their
training program. On his 21st birthday the last week of September he
was `deselected.' "I had also thought about doing a PhD and had
applied to three universities. However, I was on Hawaii and it was
already early afternoon. So, only the graduate office in Berkeley had
not yet gone home for the day, and they were happy to let me change my
mind and accept their offer of admission."
|
|
Before receiving his PhD form Berkeley,
John spent three months in Brasil and 1.5 years at Warwick. He spent
two years at the Institute
of Advanced Studies in Princeton, then one year at MIT before obtaining a
tenure track position in 1973 at the University of California at Santa Cruz. He stayed
there until 1985, by then a Full Professor in Mathematics, and moved
to Cornell
University. Cornell appealed to John for their really good
interdisciplinary interactions in applied mathematics. Furthermore,
Cornell was just starting its Theory Center and the Mathematical Sciences
Institute supported by the Army Research Office. He has been at
Cornell every since. |
A school picture from
2nd or 3rd grade. |
John's research interests are a combination of pure dynamical
systems theory, applications in (mainly) mathematical biology, and the
development of algorithms for the numerical exploration of the
associated models. In the first couple of years of his career he
worked on the problems proposed by Smale in his 1967 Manifesto for
Dynamical System Theory. He was further influenced by the research in
catastrophe theory of Christopher Zeeman (at Warwick) and René
Thom (whose book "Stabilité structurelle et
morphogénèse: esai d'une théorie
générale des modèles" was translated from French by
David Fowler while he was at Warwick). At Princeton he sat in on a
developmental biology seminar led by experimental biologist Malcolm
Steinberg. He also attended Gordon Conferences on mathematical biology
during this period and met George Oster at one of these. At the 1973
Gordon Conference, Jim Yorke talked about his now famous paper with Li,
Period 3 implies Chaos. When John moved to Santa
Cruz that year, he and George Oster started working together
on low-dimensional population models, influenced by the experiments on
blowflies done by Alexander J. Nicholson in Australia in the 1950s.
Oster also worked with Bob May on one-dimensional iterations as
population models; see, for example, Bifurcations and Dynamic Complexity in Simple Ecological
Models, The American Naturalist 110(974): 573-599,
1976.
|
David Fowler, Steve
Smale, John Guckenheimer, and Charles Pugh in Warwick,
1968 |
In those years, John was very much interested in one-dimensional
maps. He met Mitchell Feigenbaum at a Gordon conference in 1975 and
was blown away by his presentation. The work of Gollub, Swinney, and
Libchaber on chaotic fluid flows created an enormous amount of
excitement on the part of physicists in nonlinear dynamics that was
reflected in meetings sponsored by the New York Academy of Sciences in 1976 and 1978. He
spent the fall semester of his sabbatical in 1978-1979 at the
Courant Institute of
Mathematical Sciences where he met Ed Spiegel who got him involved
into fluid dynamics. "Courant really was an eye-opener for me to see
applied mathematics from a different perspective."
John Guckenheimer is best known for his book
Nonlinear
Oscillations, Dynamical Systems and Bifurcations of Vector Fields
with Phil Holmes. John met Phil for the first time in the summer of
1977 at a conference in Southampton, where he also met David Rand.
Phil and David had plans for writing a book in early 1980. David
visited Santa Cruz for a semester and attended an Institute of
Theoretical Physics program in Santa Barbara, but he decided not to
pursue the book project. John on the other hand "got tired of going
around giving survey lectures on dynamical systems". Both he and Phil
had lecture notes from courses they had taught, so they decided to
proceed. It took them about nine to twelve months to finish their
manuscript. They approached Springer-Verlag as a publisher for the
manuscript. Their editor, Walter Kauffman-Bühler, was willing to
take a risk in pricing the book at a lower cost than normal so that it
would be affordable by individuals. Walter said that books which sold
10,000 copies were mathematical best-sellers. "We crossed that
threshold quite a while ago." [Ed: Sales are now estimated at
approximately 20,000 copies.]
The influence of the computer on dynamical systems
John has always used the computer as a tool to investigate the
behavior of dynamical systems. While collaborating with Oster in 1974, he used the arpanet to access a mainframe at UCLA "doing things that today look ridiculously
simple". During a visit to Los Alamos in 1976, Alan Perelson, who
had been a postdoc of Aaron Katchalsky jointly with George Oster, helped
John to investigate the Lorenz system. These investigations led to his
subsequent work on the geometric structure of the Lorenz attractor. In
this period, physics students Doyne Farmer, Jim Crutchfield, Norm
Packard and Rob Shaw formed a group at Santa Cruz that used analog and
later digital computers to study nonlinear dynamics. John Guckenheimer
served as an informal advisor to this group. Thomas Bass recounts
their extracurricular efforts to predict in real time where a
roulette ball would stop in his book "The Eudaemonic Pie." John got his first
computer in 1984 (a SUN1 whose speed was 1 kiloflops at a price of $
30,000) and used it to run simulations of two-dimensional population
models and to perform normal-form calculations on fluid models.
|
|
|
In the California
Sierra, 1974. |
John with his son Matt in
1980 |
John started getting serious about developing a dynamical
systems tools package in 1989, after he had visited the Institute of
Advanced Studies for a special year on dynamical systems. His postdoc
Swan Kim had joined him in Princeton and had been working on a
set of bifurcation tools for studying iterations of the two-torus.
These efforts evolved into the computer package KAOS that was a
precursor to DsTool. MSI supported this research with money for a
SUN3. A year after Swan Kim finished KAOS, John obtained support from
NSF for three years to work on computational mathematics. He used it
to start the development of the dynamical systems toolbox
DsTool. His student Mark Myers was the lead person
in this effort. Mark brought his experience as an engineer at
JPL to his
graduate studies at Cornell. He was involved in the trajectory
planning for the Gallileo spacecraft, among other things. Mark's
experience in writing software was instrumental in making
DsTool a success. Graduate students
Patrick Worfolk and Rick Wicklin joined Mark in writing the initial
versions of DsTool. At the end of the grant, John sought funding to
continue the project, but was told that "writing software is a
commercial business." Not wanting to start a commercial venture, he
turned his efforts to research on algorithm development. For example,
he wrote the papers Computing Hopf Bifurcations I with Mark Myers and
Bernd Sturmfels, and
Defining Functions for Multiple Hopf Bifurcations with Willy
Govaerts and Alexander Khibnik.
Already in 1990 John's vision was that research in applied
mathematics should be coupled with good applications. He found good
examples for his research on computational methods for dynamical
systems in two different areas: the challenges of multiple
time scales in models for neuroscience, and the class of hybrid
systems in control theory problems. Software development remains
important in this research, but "I don't think that anybody has really
figured out what the role of software is in a research enterprise."
The evolution of open-source software has changed the landscape for
such work, but many research communities still have difficulty in
determining how to best support and recognize the development of
software for their communities.
The political side of things
In 1997-1998, when John was president of SIAM, he was very
concerned about the fact that in the US work on numerical analysis and
numerical algorithms was falling between the cracks, with neither
computer science nor mathematics wanting to take responsibility as
the primary disciplinary home for these research areas. He finds it
frustrating to see that this is still the case. "I find the
world of computational science and engineering a very confusing
landscape. There are different groups with very different
agendas. Selected areas such as numerical linear algebra and
optimization have developed definite niches. Some disciplines such as
astrophysics, atmospheric sciences, computational fluid dynamics,
computational chemistry and genomics have developed strong traditions
within CSE. Some groups are oriented towards building extremely large
computers and seem to measure work as much in terms of how many cycles
are used as in the scientific advances accomplished with these
computers."
Numerical analysis was a large component of computer science during
its formative years, but computer science as a discipline has evolved
away from this area. The discipline has become more focused on
information storage and retrieval, the internet and computers as
communication rather than computing devices. Data on PhDs in computer
science published by the Computer Research Association document how
few students are choosing to work in scientific computation or
numerical analysis. "Numerical analysis per se is something that seems
to be diminishing in computer science departments. Where does this
stuff get done?" The answer is "in between the cracks". Scattered
mathematics, applied mathematics and engineering departments provide
homes for research on numerical algorithms, but the unmet needs for
industry and science are enormous.
Industry would like to rely more heavily upon computation to
design and develop their products, but the algorithmic infrastructure
to support their efforts for the analysis of complex dynamical systems
simulations is inadequate. "We could be much better at this given the
hardware we have!" Without the proper acknowledgement and associated
support, relatively few talented people get drawn into software
development for dynamical systems. John thinks that the numerical
analysis of dynamical systems ought to be part of SIAM's continuing
efforts to strengthen computational science and engineering.
From a university perspective it is not clear what the most
effective strategies might be. Should one focus upon mathematics,
applied mathematics, computer science or other departments, or upon
interdisciplinary programs? Of course, things differ from country to
country, but in the US mathematics departments have little space for
this kind of research and teaching loads are typically higher than in
departments with a greater emphasis upon computational
science. "Developing software in a mathematics department is similar
to running a laboratory, but mathematics departments do not have the
resources to support laboratory research. Experimentalists typically
teach fewer formal classes than mathematicians." Computer science
departments are more entrepeneurial. "If the Federal Government would
make algorithm development a high priority as part of its initiatives
to develop computational science, then computer scientists would start
doing it again."
John can also envisage the start of new departments. He sees the
development of better algorithms and software for studying
increasingly complex systems as a bottleneck for the growth of
computational science and engineering. "I would like to see an
approach that brings more mathematics into these efforts. The ability
of mathematicians to distill the essence of these problems and pursue
novel approaches for their solution can make an enormous difference,
in the future as it has repeatedly in the past." And then, with an eye
on the research interests of his interviewer: "You would like to have a
toolkit for working with manifolds as computational objects. Is it
feasible? Sure it is feasible, but it takes a certain amount of
organization and it will not get recognized as fundamental research
until it is done and people see how great it is."
Snowbird Utah, May 2005.