It starts to move
Buoyancy acts on the density difference, not on a prescribed velocity. The rise speed readout is the drift of the bubble centroid, so it starts near zero and settles into a terminal value.
A gas bubble starts at rest inside a heavier liquid and rises under buoyancy. There is no interface to draw: the boundary between the phases emerges from the interaction between neighbouring densities.
Bubble size and rise speed are measured from the density field every 0.2 s; the aspect ratio is width ÷ height, so < 1 means the bubble has stretched into a vertical ellipse.
A single-phase solver has one density everywhere. To get two phases you add a force between neighbouring cells that depends on a pseudopotential, ψ(ρ). With the right weights and a negative interaction strength, uniform fluid becomes unstable and splits into a dense phase and a light one, separated by a diffuse interface a few cells thick.
That is the Shan–Chen model. It costs almost nothing: one extra loop to compute ψ, one to apply the interaction force, and the same collision and streaming as the single-phase solver. It also has the model's characteristic weakness — the interface is not thermodynamically consistent, so it produces the parasitic currents you can see as shimmer.
Buoyancy acts on the density difference, not on a prescribed velocity. The rise speed readout is the drift of the bubble centroid, so it starts near zero and settles into a terminal value.
The pressure field squeezes the rising bubble into a horizontal ellipse: the aspect ratio falls below 1. Raise gravity and the deformation gets stronger.
Switch to vorticity: a pair of counter-rotating vortices trails under the bubble. In a taller domain the column would start to wobble — that is the path to the zig-zag regime.
The domain is periodic and only 200×240 cells, so the bubble is never far from its own wake — it rises back into it. The density ratio is about 26:1 rather than the ~800:1 of water and air, which flattens the dynamics. And the interface is several cells wide by construction, so surface tension is not a tunable parameter here.
What the page is good for is the shape of the phenomenon: why a bubble deforms, how the wake forms, and what happens when you make it bigger, heavier or more viscous. That intuition is exactly what an interface-resolving solver would give you at a hundred times the cost.
Buoyancy pushes harder on the underside than on the top, so the pressure field squeezes the bubble into a flattened ellipse — the aspect ratio readout drops below 1 as it accelerates. At higher gravity it becomes an indented “skirt” shape, the classic regime map for rising bubbles.
Spurious currents. In pseudopotential models the interface carries a small parasitic velocity even at rest, because the discrete force is not exactly balanced by the pressure gradient. Raising viscosity damps them; they are the main visual artefact of this method.
Shan–Chen models lose stability as the density ratio grows, and 26:1 is roughly what this grid and relaxation time tolerate while staying clean. Water and air are about 800:1 — for that you need a different method (colour-gradient or free-energy models).
No. The cylinder page is single-phase: one density, no interface. Here an interaction force between neighbouring densities creates phase separation, which is why a bubble can exist at all.