Computational Scenarios

Leo P. Kadanoff

Citation: Phys. Today 57(11), 10 (2004); doi: 10.1063/1.1839351

View online: http://dx.doi.org/10.1063/1.1839351
View Table of Contents: http://www.physicstoday.org/resource/1/PHTOAD/v57/i11
Published by the American Institute of Physics.

Additional resources for Physics Today

Homepage: http://www.physicstoday.org/
Information: http://www.physicstoday.org/about_us
Daily Edition: http://www.physicstoday.org/daily_edition
Computational Scenarios
Leo P. Kadanoff

S upernovae are among the most

dramatic events in nature. A star
undergoes a thermonuclear explosion
stars to predict the likelihood of such a
triggering event. The other is the ris-
ing blob of hotter and less dense mate-
interfaces. Thus, it is natural that
ASC has sponsored more than a dozen
experimental, theoretical, and simu-
and within a few seconds produces a rial started by the trigger and formed lational studies of the Rayleigh–Tay-
total energy flux comparable to that of by the Rayleigh–Taylor instability. lor instability.
an entire galaxy. To visualize the instability, imag- The results of the studies were sur-
Computer simulations are begin- ine a dense fluid placed above one of prising and disturbing. Contrary to
ning to catch the essential features of lesser density with the two separated expectations, variations in initial con-
these events. Recently, a large by an interface that is not quite flat. ditions, in computational methodol-
Chicago-led simulation showed an at- The pull of gravity will trigger a mo- ogy, or in computational precision
tractive new mechanism for one kind tion in which the two fluids will in- gave measurable variations in the
of supernova.1 Previous calculations terpenetrate one another to form a “universal” parameter, a. The Chicago
have often been done for spherically kind of mixing zone. The motion will group even found a kind of dynamical
symmetrical situations in which the be inhibited by the surface’s interfa- phase transition in its calculations
event is triggered by a heating event in cial tension and the fluids’ viscosities. within the computationally interest-
the center of the star. The new calcu- If these inhibiting effects can be neg- ing range. As the resolution of the
lations look at off-center triggering, lected, the motion will be basically calculation was improved, the finger-
which sends a very asymmetrical blob ballistic. A simple dimensional analy- like structure produced by the
of hot material rising up through the sis argument then gives an estimate Rayleigh–Taylor instability lost its
star. At the point where the bubble sur- of the height of the mixing zone. The left–right symmetry and became
faces, outer stellar layers are strongly height will grow in proportion to the quite asymmetrical. This change had
accelerated, fly over the star surfaces, square of the time, with a constant of a substantial effect on the mixing of
and then come together on the opposite proportionality called a. Arguments the fluids. Taken all together, these
side of the star. That focus then forms based upon the concepts of universal- different studies suggested that sev-
the center of the resulting explosion. ity, scale invariance, and the renor- eral different complex causes will re-
This entire sequence has been malization group back up the dimen- sult in a variation in the apparent
studied by a computer simulation that sional analysis. This approach put value of a by roughly a factor of two
gives many attractive results for com- forward by a group at Stony Brook within the parameter range studied.
parison with observational data. The University suggests that once the flow Even larger variations might be ex-
sequence of events might well explain gets well started, a might be a uni- pected if one tried to extrapolate to
previously considered anomalies in versal constant, independent of every- new situations. These variations pro-
the angular anisotropy of the events thing except perhaps the ratio of the duce a substantial impediment to ac-
and in the mass and velocity spectra densities of the two fluids.2 curate predictions.
Perhaps these outcomes should not
of the ejecta. If the mechanism sur- Stockpile stewardship have been entirely surprising. The
vives further testing, the results The studies mentioned here are part mathematics of the situation does in-
might help us understand one class of of the stockpile stewardship program dicate that in the limit of low viscos-
supernova events. of the US Department of Energy ity and surface tension, the problem of
Computer models are representa- weapons laboratories. The steward- Rayleigh–Taylor interface instability
tions of nature. Sometimes they can ship program uses extensive com- is ill-posed, which means that the re-
give faithful and accurate representa- puter simulations to maintain the re- sults are potentially very sensitive to
tions. Sometimes their accuracy is liability of our nuclear weapons. It small variations in initial data or com-
weak or unknown. Robert Batterman, also tests the simulations by using putational method. Thus, empirical
a philosopher of science, has argued university groups, in an effort called and mathematical evidence suggest
that the most accurate models do not ASC, for Advanced Simulation and that in this kind of instability prob-
necessarily give the most worthwhile Computing, to explore situations akin lem, predictions are likely to remain
results. His position is somewhat to ones that might be relevant for the subject to doubt no matter how much
counterintuitive. Nonetheless, I weapons stewardship. This program effort is invested in simulations of the
argue for it, using the example of this supports part of my research. Rayleigh–Taylor problem.
excellent and provocative supernova One important focus of ASC is the
simulation. test of computer programs that simu- Inaccurate and important
The simulated event starts with two late Rayleigh–Taylor instabilities. Despite these adverse results for a
imperfectly known processes. One is These instabilities play an important universal a, the Chicago ASC group
the off-center trigger. We simply do not role in many explosive events, includ- came up with two different applica-
know enough about fluctuations in ing supernovae. Gravity can produce tions of the same techniques—one to
the basic instability. In explosive sit- novae, and the other to the super-
Leo P. Kadanoff is a condensed mat- uations, the role of gravity is replaced novae calculation described above.
ter theorist at the University of Chicago. by that of the rapid acceleration of the The group pointed out that both re-

10 November 2004 Physics Today © 2004 American Institute of Physics, S-0031-9228-0411-210-6

Computer simulation shows a bubble rising through a model star.
The bubble was started in a slightly off-center position in a spher-
ically symmetrical environment. This completely asymmetrical
structure was a surprising outcome and has an important influ-
ence on the subsequent history of the star and its transition to su-
pernova behavior. Another simulation (see Plewa et al., reference
1) shows that after the bubble surfaces, material flies over the star
surface and dredges up material from the surface. Image was
constructed by the ASC center at the University of Chicago and
Argonne National Laboratory.

sults were important additions to this latter category as

knowledge of these astrophysical phe- well. I expect that
nomena. Fluid instabilities are signif- most astrophysical
icant ingredients in both calculations. calculations involv-
The Rayleigh–Taylor study, however, ing turbulent mixing
indicated that there may be difficul- will also remain
ties with numerically modeling these heuristic in character
instabilities. Thus the calculations for a long time to
were at the same time inaccurate and come, perhaps until
important! How can that be? we really understand
As I see it, there are two ways in turbulence.
which a simulation may be useful. To deny these cal-
First, the simulation may prove some- culations the honorable name of proof very short scales or very long ones.
thing. In such a case, the simulation is not to deny their value. Nothing can The new processes and structures are
will include carefully checked rou- be more valuable to science than to often completely unexpected. We need
tines that accurately model all the suggest how things might happen. methods to protect ourselves from
physical processes involved. These The Watson–Crick model of the dou- being overtaken by unexpected occur-
routines will have been validated and ble helix implied suggestions for how rences. Computer simulations of sim-
verified by a long series of investiga- biology might work. Those sugges- ple nonlinear systems can be one of
tions open to the usual processes of tions were of magnificent value. The our best tools for explorations that
broad scientific criticism and compar- suggestion of a new mechanism for uncover unexpected possibilities.
ison to a broad variety of observa- how a supernova might go off is of But there are also major dangers in
tional data. considerable value, even if the mech- simulations. Often simulations are di-
A second kind of simulation is the anism does not directly provide a rectly aimed at confirming our expec-
exploratory one that suggests new proof of its truth. tations, thereby throwing away the
mechanisms for complex physical possibility of finding anything new. In
processes. Such simulations are likely Making a star go boom addition, we simulators must be most
to contain components that represent How does the power of argumentation careful to distinguish between simula-
the best state of our knowledge, but provided by exploratory simulations tions as argument versus simulations
are not necessarily “true.” They are compare to that of rhetorical or order- as proof. There is a considerable risk
most effective when they use a sim- of-magnitude discussions? Since the of confounding the two approaches. It
plified model of the situation to isolate simulations must include everything is tempting to say that “supercom-
the essential elements of the new to make a star go boom, they provide puter simulations show . . .” when
processes. Their results are likely to an internal check of consistency and what is meant is more like “recent in-
be remembered in the words that de- completeness not available through vestigations have raised the possibil-
scribe the processes, rather than in words. On the other hand, some inter- ity that . . . .” In our writing, we all
any numerical computer output. mediate steps in the argument may are tempted to replace “it would
These exploratory calculations repre- have their weaknesses hidden in un- please us if . . .” by “we know
sent some sort of argument. Are they examined computer processes. Words that . . . ”. And if we scientists and en-
just a rhetorical device? Do they rep- may be better than computer output gineers join with all those around
resent an important improvement for showing up weak arguments. Com- us—in places high and low—who con-
over the traditional method for con- puter arguments often force us to rely found possibility with proof, and de-
ducting such arguments, via words upon the integrity and care of the in- sire with truth, who then will believe
supplemented by order-of-magnitude vestigators. So computers provide a us in anything we say?
estimates? useful but dangerous tool for the ex-
I’ll come back to these questions. ploration of complex systems. I had very helpful discussions with Alan
But first, I should point out that only There is an important technical Calder, Robert Rosner, Tomasz Plewa,
a few physically interesting calcula- reason for the increasing use of com- Steve Libby, Don Lamb, Robert Batter-
tions fall into the first category, that puters in exploring physical systems. man, and Guy DiMonte.
of proof. In a previous paper, I sug- In the past, analysis of point particles
gested that the computational studies used ordinary differential equations References
of neutrino production within the Sun to plot their position and momentum. 1. T. Plewa, A. C. Calder, D. G. Lamb,
may have achieved that status.3 How- More recent work has focused upon Astrophys. J. 612, L37 (2004); A. C.
fluids, plasmas, and even solids that Calder et al., http://www.arXiv.org/abs/
ever, studies of magnetic field gener- astro-ph/0405162.
ation by dynamos clearly fall into the are studied as continuum systems, 2. B. Cheng, J. Glimm, D. H. Sharp, Chaos
second category, that is, exploratory usually with the aid of partial differ- 12, 267 (2002).
calculations. Glass formation and ential equations. These systems and 3. L. P. Kadanoff, Comp. Sci. Eng. 6 (2),
fracture in solids remain mostly in equations can develop structures with 57 (2004). 䊏

http://www.physicstoday.org November 2004 Physics Today 11