WAVES · EXPERIMENT 003

Chladni figures

Change the frequency and watch sand form patterns on a vibrating plate.

Plate in perspective470 Hz
The turquoise actuator moves the centre point of the plate. Illustrative perspective: vibrations are slowed and exaggerated; grain hops are schematic. Sand positions and vibration intensity come from the current model.
Loading and preparing plate modes…470 Hz
Loading and preparing plate modes…
Gold: sand · turquoise: excitationCompute: 0.0 ms

Nearest natural frequency (among 80 modes): —

Logarithmic sweep: 100–3000 Hz in 20 seconds. Stop to let a pattern emerge.

20,000 is intended for powerful devices; start with 2,500 on a phone.

Try a resonance

Mode frequencies coupled to the selected excitation. Give the sand time to settle.

Theory behind the experimentWhy sound gathers sand into patterns and how the plate’s shape shapes its resonancesPlain explanation, equations and how the model works

In plain words

Make a thin plate vibrate and different parts of it move differently. Some go up and down a lot, others barely move. Grains on the busy parts get tossed and pushed around, while in the quiet places they stay put. That’s how a pattern slowly forms.

The lines where the plate is almost still are called nodal lines. The figure shows how the plate itself vibrates, not the shape of sound in the air. Sound and pattern share the same frequency, but they’re different things.

A plate has its own natural shapes of vibration, each with its own frequency. Drive the plate near one of those frequencies, in a way that can actually excite that shape, and the response jumps. That’s resonance. Changing the frequency moves you from one pattern to another, and between resonances you see a mix of several shapes.

The pattern depends on the plate’s shape, how it’s held, and where it’s pushed. Push at a point that barely moves in some shape, and that shape hardly gets excited. So the same frequency can give different patterns depending on how the plate is driven.

Try it yourself

  1. Press one of the resonance buttons, then nudge the frequency by hand. Compare how sharp the pattern stays.
  2. Switch between square, circle and triangle. Each shape has its own set of vibrations and frequencies, so the same hertz give different resonances.
  3. In the advanced settings switch to “Off-centre force” and move the force point around. Keep in mind this also changes how the plate is held, so the difference isn’t only about where it’s pushed.

Physics and equations

Bending of a thin elastic plate is described by Kirchhoff-Love theory. Without damping the equation is

ρh ∂2w∂t2+D ∇4w=p(x,y,t)\rho h\,\frac{\partial^2 w}{\partial t^2}+D\,\nabla^4 w=p(x,y,t)
D=Eh312 (1−ν2)D=\frac{E h^3}{12\,(1-\nu^2)}

Here ww is the transverse (out-of-plane) displacement in metres, ρ\rho the material density in kg/m³, hh the thickness, EE Young’s modulus in pascals, ν\nu Poisson’s ratio, DD the bending stiffness in N·m, and pp the external load per unit area. The operator ∇4\nabla^4 captures how the plate’s curvature changes from point to point. This is the equation for bending a stiff plate; a stretched membrane like a drum follows a different one.

Complicated motion is easiest to split into natural shapes:

w(x,y,t)=∑nqn(t) ϕn(x,y)w(x,y,t)=\sum_n q_n(t)\,\phi_n(x,y)

The function ϕn\phi_n is a shape, and qnq_n says how strongly that shape takes part in the motion at the moment. It’s convenient to make ϕn\phi_n dimensionless so that qnq_n is in metres. In moving-centre mode the prescribed motion of the support is handled separately, and the sum describes the plate’s deformation relative to it.

Under harmonic driving, the complex amplitude of a single shape is

q^n(ω)=Fn(ω)ωn2−ω2+2iζ ωnω\hat q_n(\omega)=\frac{F_n(\omega)}{\omega_n^2-\omega^2+2i\zeta\,\omega_n\omega}

ω=2πf\omega=2\pi f is the driving frequency in radians per second, ωn\omega_n the shape’s natural frequency, and ζ\zeta a dimensionless damping ratio. FnF_n says how strongly the drive excites that particular shape (already divided by its modal mass, so it isn’t just a force in newtons). The imaginary unit ii lets one number carry both amplitude and phase. Responses of different shapes add up with their phases, so at one point they can reinforce or cancel each other.

As the frequency approaches a natural one, the denominator gets small and the response grows. Damping keeps it finite and sets the width of the resonance. But a strong response only appears if the drive actually couples to that shape, meaning FnF_n isn’t too small.

How the model works

The plate is aluminium: Young’s modulus 69 GPa, density 2700 kg/m³, Poisson’s ratio 0.33, thickness 1.5 mm. The 30 cm size is the side of the square, the diameter of the circle or the side of the equilateral triangle. The outer edges are free.

In moving-centre mode the centre’s vertical motion is prescribed and the slopes there are free. In force mode the centre is fixed and clamped against tilting, and the force acts at the chosen point. These are two different ways of holding the plate.

The natural shapes are precomputed with the Rayleigh-Ritz method, and the page adds up 80 of them with their amplitudes and phases. The displayed field is normalised: good for comparing patterns, but it doesn’t give absolute amplitude or energy in physical units. The field is proportional to the squared vibration amplitude, while the size of a grain’s random step is proportional to the amplitude itself.

The sand is a qualitative random model. Grains jump harder where the plate vibrates more and nearly freeze in quiet spots, so that’s where they pile up. There’s no built-in pull toward nodal lines. Collisions, friction and real grain bounces aren’t simulated, and the speed at which the pattern forms isn’t physically calibrated. In the perspective view the vibrations are slowed down and exaggerated for clarity.

More on accuracy

The frequency goes from 10 to 5000 Hz. For the square and circle the calculation was compared with a finer one (polynomial orders 15 and 17) only up to 1500 Hz: natural frequencies shifted by up to 1.5% with a moving centre and 5.3% with a clamped one. This checks how much the result depends on the choice of basis, not how well it matches a real plate. Convergence for the triangle isn’t established. Accuracy above 1500 Hz and agreement with a real experiment haven’t been checked.

Sources and further reading

MIT OpenCourseWare: Bending Response of Plates ↗: a lecture on plate bending theory. It doesn’t cover sand motion; it’s about plates.

Experiment video

Chladni figures

Watch the short video, then try changing the model’s parameters yourself.

Video in Russian

Watch on YouTube