010 / CHAOS
Fermi acceleration
A wall swings back and forth and on average stays put, yet the balls bouncing off it speed up tens and hundreds of times. Between two walls the growth hits a ceiling, above a table with gravity it never stops.
Phase portrait
all hits of 400 ballsEach dot is a hit: horizontally the moment of the swing when it happened, vertically the ball speed after it. Between walls with a sine, solid lines and islands grow at the top: the ball cannot cross them, and that is the ceiling. Above the table the dots keep climbing.
Theory behind the experimentWhy a ball speeds up off a wall that on average stays put, where the ceiling comes from and why gravity removes itPlain explanation, equations and how the model works
In plain words
A ball bounces off a wall. If the wall is still, the ball leaves at the same speed. If the wall moves towards it, the ball leaves faster, like a tennis ball off a swinging racket. If the wall moves away, slower: that is how you cushion a ball by pulling your foot back.
The wall swings back and forth, so gains and losses alternate. But the ball meets the wall coming towards it more often than the receding one. Running in the rain, more drops hit your front than your back. Gains slightly outnumber losses, and the energy of the balls grows. It does not come from nowhere: the motor that swings the wall pays for it.
Enrico Fermi proposed this idea in 1949 to explain cosmic rays. Protons in space bounce off moving clouds of magnetised gas and slowly gather huge energies. Today the main acceleration is thought to happen at shock waves from supernova explosions.
In 1961 Stanislaw Ulam tested the idea on a computer with the simplest model: a ball between a fixed wall and a swinging one. Everyone expected endless acceleration, but found a ceiling. A fast ball hits the wall many times during one swing. It speeds up while the wall moves in and slows down again while it moves out, with almost no net gain.
Replace the second wall by gravity and the ceiling is gone. The faster the ball, the longer it flies, and the moment it meets the table becomes random. The energy grows like a random walk: up and down, but on average up, without limit.
Try it yourself
- Run "above a table" and "between walls" at ×100 and compare the average energy charts: in one case growth stops, in the other it does not.
- In "One hit" set the wall speed to 0.5 and the ball speed to 1. After the hit the ball flies at 2 and its energy is 4 times larger. Now make the wall speed −0.5.
- In the laboratory choose "between walls" with a sine, switch to ×100 and watch the phase portrait. At the bottom the dots are scattered at random, at the top solid lines and islands grow. The balls never cross them: that is the ceiling.
- Switch the wall to the triangle wave. The average energy grows for longer than with the sine, but unevenly: most balls stay slow and a few pull the average up. Look at the chart of each ball.
- Go "above a table" and switch to ×1000. The average energy grows more and more slowly but never stops, and the camera keeps zooming out.
- Lower gravity or the table amplitude. When the table accelerates downwards more weakly than gravity, the ball no longer lifts off on every swing and the acceleration changes.
- Press "New starts". The coloured balls behave differently each time, while the average over 400 repeats. That is chaos and the law of large numbers at once.
Physics and equations
The wall is infinitely heavy and the hit is elastic. In the frame of the wall the ball simply reverses its velocity. Going back to our frame, along the normal:
where is the wall velocity and the ball velocity before the hit. The energy change per unit mass is
It is positive when the wall moves towards the ball and negative when it recedes. The hit rate is proportional to the approach speed , so hits on an approaching wall are more frequent. Averaging over phases gives a mean gain of order per hit.
Between two walls, for smooth motion and a fast ball, there is an adiabatic invariant:
The ball makes many hits during one slow swing, and its speed follows the width of the chamber back and forth. Rigorously, on the plane "hit phase, speed" there are invariant curves at high speeds (KAM theorem). The chaotic region at the bottom is bounded by the first such curve, and that is the ceiling. For sufficiently smooth periodic wall motion the energy is proven to stay bounded.
The triangle wave has kinks, so the conditions of this theorem do not hold. In our runs the average energy with the triangle wave grows for longer than with the sine, thanks to a few balls that run far away. But the record among 400 balls barely grows after about 1000 swings, so these data cannot tell whether the triangle wave has a limit. For a smooth random wall a limit is expected too, only high up and after very long runs.
Above a table the flight time grows in proportion to the speed, and the phase of the next hit shifts more the faster the ball. At high speeds consecutive phases are nearly independent, and the speed performs a random walk with steps of order . The key parameter is
the ratio of the largest table acceleration to free fall. For the ball lifts off on every swing.
How the model works
Units are dimensionless: chamber width , swing period 1. The wall moves as
where is a sine, a triangle wave (the wall moves at constant speed and turns instantly) or a smooth random signal: a sum of 12 harmonics with random frequencies from 0.6 to 1.4 of the base one and the same mean square speed as the sine. As in the videos, , and gravity above the table is , so .
There is no time step: the program finds the exact moment of each hit. For smooth laws the distance to the wall is bounded from below by a parabola
where is the largest wall acceleration (plus gravity for the table). A step to the root of this parabola never jumps over the wall, so a ball cannot pass through it. For the triangle wave the wall moves uniformly on each piece and the hit time follows from a formula.
The 400-ball ensemble runs in the browser in double precision with the same algorithm as the Python version used for the videos. The site tests compare the first 40 hits of several balls for all six combinations of scene and law: the difference is at most . With a resting wall the energy of every ball is conserved.
The coloured balls and the wall "as in the video" match the videos exactly. The phase portrait shows a random sample of up to 250 thousand hits. The ceiling of the sine between walls shows best in the record, the largest energy among 400 balls: between swing 1000 and 4000 it changes by 1 %, while above a table it grows 2 to 2.3 times. The average energy with the sine keeps creeping up slowly for a long time while the balls fill the chaotic region below the ceiling.
Simplifications: the wall is infinitely heavy and does not feel the hits, the balls do not collide with each other, every hit is perfectly elastic.
Sources and further reading
- E. Fermi. On the origin of the cosmic radiation. Phys. Rev. 75, 1169 (1949). doi ↗
- S. M. Ulam. On some statistical properties of dynamical systems. Proc. Fourth Berkeley Symposium on Mathematical Statistics and Probability, vol. 3, p. 315 (1961).
- A. J. Lichtenberg, M. A. Lieberman. Regular and Chaotic Dynamics. Springer (1992), chapter 3: the Fermi–Ulam model and the bouncing ball.