The Elephant in the Room: Fluid Dynamics in the age of Machine Learning

Prof. Matthew Juniper

Slide at 02:50

Nature 427(6972), 297 (2004)
essay turning points
A meeting with Enrico Fermi
How one intuitive physicist rescued a team from fruitless research
reach your calculated results, you had
Freeman Dyson
physical picture, and the forces are so strong that nothing converges. To
ne of the big turning points
reach your calculated results, you had
in my life was a meeting with
to introduce arbitrary cut-off proce-
Enrico Fermi in the spring of
dures that are not based either on solid
to introduce arbitrary cut-off proce-
1953. In a few minutes, Fermi politely
physics or on solid mathematics."
but ruthlessly demolished a programme
desperation asked whether
of research that my students and had
he was impressed by the agreement
been pursuing for several years. He
between our calculated numbers and his
dures that are not based either on solid
probably saved us from several more
measured numbers. He-replied,
years of fruitless wandering along a road
many parameters didyou
that was leading nowhere. I am eternally
for your calculations?' thought for
grateful to him for destroying our
moment about our off procedures
illusions and telling us the bitter truth.
and said, "Four." He said. remember
physics or on solid mathematics."
Fermi was one of the great physicists
my friend Johnny von Neumann
of our time, outstanding both as
say, with four parameters I can fit
theorist and as an experimenter. He
elephant, and with five I can make him
led the team that built the first nuclear
wiggle his trunk." With that, the conver-
reactor in Chicago in 1942. By 1953
sation was over. thanked Fermi for his
he was head of the team that built
time and trouble, and sadly took the next
the Chicago cyclotron, and was using
bus back to Ithaca to tell the bad news
it to explore the strong forces that
to students. Because it was important
"How
hold nuclei together. He made the
for students to have their names on
first accurate measurements of the
a published paper, we did not abandon
scattering of mesons by protons, an
calculations
immediately. We
experiment that gave the most direct
finished them and
evidence then available of the nature of
wrote long paper
the strong forces.
that was duly pub-
many arbitrary parameters did you use
At that time was a young professor
lished in the Physi-
of theoretical physics at Cornell Univer
cal Review with all
sity, responsible for directing the
Crossed paths: A discussion with Enrico
names on it.
research of a small army of graduate
Fermi (above) made Freeman Dyson
Then we dispersed
for your calculations?" I thought for a
students and postdocs. had put them
(right) change his career direction.
find other lines of
to work calculating meson-proton scat-
work. escaped to
tering, so that their theoretical calculations
package of our theoretical
Berkeley, California,
could be compared with Fermi's measure-
graphs to show to Fermi.
to start new career
ments. In 1948 and 1949 we had made
When I arrived in Fermi's
in ondensed-matter
moment about our cut-off procedures
similar calculations of atomic processes, using
office, handed the graphs to
physics.
the theory of quantum electrodynamics, and
Fermi, but he hardly glanced
Looking back after
found spectacular agreement between experi-
at them. He invited me to sit
fifty years, we can
ment and theory. Quantum electrodynamics
down, and asked me in a
clearly see that Fermi
and said, "Four." He said, "I remember
is the theory of electrons and photons
friendly way about the health
was right. The crucial
interacting through electromagnetic forces.
of my wife and our new-
discovery that made
Because the electromagnetic forces weak,
born baby son, now fifty
sense of the long
we could calculate the atomic processes
years old. Then he delivered
forces was the qua
precisely. By 1951, we had triumphantly
his verdict in a quiet, even voice. "There are
Mesonsa protonsal
finished the atomic calculations and were
two ways of doing calculations in theoretical
little bags of quarks. Before Murray Gell-
my friend Johnny von Neumann used to
looking for fresh fields to conquer. We
physics", he said. "One way, and this is the
Mann discovered quarks, no theory of the
decided to use the same techniques of calcu-
way prefer, is to have clear physical picture
strong forces could possibly have been
lation to explore the strong nuclear forces.
of the process that you are calculating. The
adequate. Fermi knew nothing about quarks,
We began by calculating meson-proton
other way is to have a precise and self-
and died before they were discovered But
say, with four parameters I can fit an
scattering, using a theory of the strong forces
consistent mathematical formalism. You
somehow he knew that something essential
known as pseudoscalar meson theory. By the
have neither." I was slightly stunned, but
was missing in the meson theories of
spring of 1953, after heroic efforts, we had
ventured to ask him why he did not consider
the 1950s. His physical intuition told him
plotted theoretical graphs of meson-proton
the pseudoscalar meson theory to be a self-
that the pseudoscalar meson theory could
scattering.We joyfully observed that our
consistent mathematical formalism. He
not be right. And so it was Fermi's intuition,
elephant, and with five I can make him
calculated numbers agreed pretty well with
replied, "Quantum electrodynamics is
and not any discrepancy between theory and
Fermi's measured numbers. So I made
good theory because the forces are weak,
experiment, that saved me and my students
an appointment to meet with Fermi and
and when the formalism is ambiguous we
from getting stuck in blind alley.
show him our results. Proudly, I rode the
have clear physical picture to guide us With
Institute for Advanced
Greyhound bus from Ithaca to Chicago with
the pseudoscalar meson theory there is no
Jersey
08540
USA.
wiggle his trunk."
2023-12-07 15:06:39
Publishing Group

References

1.
F. Dyson (2004) A meeting with Enrico Fermi. Nature

Share slide

Summary (AI generated)

In the article from Nature that I read, it discussed the pseudoscalar mison theory. However, it was ultimately proven to be rubbish. The theory could only fit the data by adjusting arbitrary parameters, which were not meaningful or significant. Therefore, if we interpret this to mean that a physical model should have fewer than four parameters, we are doing ourselves a disservice. This was never the intention.

In fact, in today's era of abundant data, I believe that we can utilize physics-based models with more than four parameters. The main point of this talk is to assert that when we have a large amount of data, specifically millions of data points, we can effectively train models in physics that require more than just four parameters.