Slow places collect traffic
Imagine a corridor with a thick carpet halfway along it. People keep trying to walk, but they move more slowly over the carpet. Even with no desire to gather there, the slow half becomes busier: people spend more time in it.
A swimmer that slows in a region behaves similarly. In a simple steady state, the number arriving each second must equal the number leaving. A region traversed at half the speed can contain twice the density. This example has a speed pattern imposed from outside; the interesting step comes when the swimmers themselves create it.
Now let the crowd make itself slow
Suppose a swimmer moves more slowly whenever it senses more neighbours. A patch that happens to become denser now also becomes slower. New arrivals linger, the patch becomes denser still, and a small fluctuation may grow. This is the central feedback behind motility-induced phase separation, usually shortened to MIPS.
Self-propelled repulsive particles can also slow because they block one another. Two particles pushing head-on remain in contact until they turn enough to escape. A density-dependent speed model describes one coarse version of that slowdown. Its particles need not be literal hard disks; the local density is measured over a chosen neighbourhood.
The two descriptions answer related questions at different scales. To predict a particular colloidal material, the collision law, particle shape and hydrodynamics matter. To isolate the feedback, it is useful to prescribe how speed falls with density and test what follows.
Worked example
An ordinary attraction gives a different test
Prepare a crowd in three ways: passive wandering, self-propelled motion that slows in dense regions, and passive motion with an attractive force. All three can show clumps in a small box. Now switch off propulsion while keeping the attraction unchanged.
The attractive particles still pull on their neighbours. The motility mechanism loses its drive when swimming stops; with positional Brownian diffusion left on, its crowd disperses. This intervention distinguishes causes more clearly than comparing two attractive-looking snapshots.
There is a second useful intervention. Keep swimming on but remove the dependence of speed on density. Freely crossing particles now have no reason to linger in a crowd. Continued motion by itself is insufficient for this particular clustering mechanism.
A cluster is a beginning, not a phase diagram
A few hundred particles sometimes form one large patch because the box is small. Changing the seed may change its lifetime; changing the neighbourhood radius may change its apparent density. A bright group of dots is a useful observation, but macroscopic phase coexistence calls for more work.
Measure density distributions over time, compare several box sizes at fixed mean density, and repeat with independent seeds. A local-density histogram groups sampled neighbourhoods by how crowded they are. The browser takes one sample per particle; equal-area grid cells give a different weighting. Two populations suggest dense and dilute regions. Whether those populations survive increasing size and longer observation is the next question. The method chapter turns the feedback into a stability calculation.