Why the timing matters
These arrows show the effective pull in the moving base’s frame. The fast wobble is exaggerated here. With suitable shaking, the restoring effect wins over gravity’s tendency to topple it.
An experiment in balance
Upside-down pendulums. One unexpected way to make instability stable.
Vertical shaking · on
Three free hinges. Nothing steering the lean.
Changing a setting restarts the same release, so comparisons stay fair.
Readouts · current experiment
Look a little closer
Gravity still pulls down. The timing of the movement changes what happens on average.
Near upright, the links sway back and forth. Staying near zero is different from being perfectly still.
These arrows show the effective pull in the moving base’s frame. The fast wobble is exaggerated here. With suitable shaking, the restoring effect wins over gravity’s tendency to topple it.
Three things to discover
Turn the drive off without resetting the chain. It starts falling from its current position. Turn it back on after a large fall and it may keep swinging: stability near upright does not mean recovery from every disturbance.
Try the three-link chain at 32 Hz and at 64 Hz. Then double its amplitude from 12 to 24 mm. Within this tested range, both too little and too much shaking can lose the upright balance.
One link and three links do not share the same stability window. The hinges allow collective patterns of motion. Acheson’s result for finitely many rigid links does not extend to a perfectly flexible rope.
For the curious
A planar chain of one, two or three massless rigid links, each 0.25 m long, with a 0.10 kg point mass at each end. Every hinge has viscous damping of 0.001 N·m·s. The base moves vertically on a prescribed path; it never measures or corrects the lean.
Amplitude means displacement from the centre: a 12 mm amplitude is a 24 mm peak-to-peak stroke. At 64 Hz this idealized drive reaches about 198 g of peak acceleration. The rendered apparatus illustrates the model; it is not a replica of a laboratory device. Rod thickness, decorative hardware and camera perspective do not change the equations. Air drag, elastic bending, collisions and actuator limits are omitted.
Switching the drive uses a smooth transition so the base does not jump. Reset returns to the same phase and initial state. A nudge applies a small horizontal impulse to the top mass. The chart shows angles relative to upright, wrapped to ±180°.
The slow-drive, large-amplitude and three-way comparison presets replay a three-second trial from identical initial conditions. This keeps their focus on the loss of upright balance, rather than prolonged post-collapse rotation in a model without collisions. Changing settings or using the drive/nudge controls returns to free exploration. Free trials pause shortly after a sustained fall, keeping attention on the initial loss of upright balance. This is an observation limit, not a threshold of physical validity. Upright trials have no time limit. At strong drive settings, detailed tumbling paths are sensitive to numerical precision. The phase diagram above is an explanatory drawing for a single pendulum, with the tiny fast wobble enlarged. It does not draw forces to scale. At real-time speed, sampled physical states form a brief motion trail; slow motion resolves the drive cycle.
For one pendulum, with angle measured from upright and vertical base displacement yb, the equation is Lθ̈ = (g + ÿb) sin θ, with the stated damping added. For linked pendulums the coupled mass matrix, centrifugal terms and hinge torques are integrated together. The full computational model is available as readable source.
The numerical checks compare the browser kernel against an independent high-accuracy integration, test timestep convergence, verify undriven energy and drive transitions, and reproduce stable and unstable releases for the settings used in the film. A finite simulation is not a proof of stability for all time or all starting conditions.
Nine scientific test groups passed, including eight independent trajectory comparisons. In the earlier release, the tested Mac’s three-chain browser comparison averaged 164.85 frames/s over ten seconds with no dropped simulation time. A phone-sized desktop viewport is not a physical-phone benchmark.
The simulation runs in your browser. Its code does not include account registration or analytics. Hosting requests and external links are explained in our Privacy notice. The animation pauses when this tab is hidden. Reduced-motion preferences start the experiment paused. All controls work from the keyboard; the readouts and explanations remain available independently of the 3D image. If 3D graphics are unavailable, a simpler two-dimensional view is used.
The integrator keeps its small physical timestep when a device is busy. Playback may slow instead of skipping through the motion. The simulation clock shows the actual time advanced. A real phone may perform differently from a phone-sized desktop browser.
The check measures animation timing and simulation cost for the current setup. Keep this tab visible.
Scientific starting points: D. J. Acheson, “A pendulum theorem” (1993); D. J. Acheson and T. Mullin, “Upside-down pendulums” (1993). The original theorem concerns linear stability under suitable drive conditions; the nonlinear model here explores particular finite disturbances.
Developed for THEORYMERIDIAN by Daniel Hashemi. Original procedural apparatus and explanatory graphics. Three.js is distributed under the MIT licence. Fraunces, Geist and Geist Mono are distributed under the SIL Open Font License. Third-party licences.
Companion to Why Shaking Stops It Falling | Upside-Down Pendulums and its three short films. Film links will be added when the releases are public.