Particle Life, Explained
After reading this you will understand the exact force law behind Particle Life, why an asymmetric rule table lets clusters swim under their own power, and how to read the difference between a frozen blob and a living creature.
What Particle Life is
Particle Life is a physics toy with almost no rules. Every particle belongs to one of a few species, drawn as a colour. A single square table, one row and one column per species, says how much each species is attracted to or repelled by every other. That table plus a short-range push, a decay of force with distance, and a friction term is the entire model. No particle knows what a cell or a worm is.
Run it and structure appears anyway. With three or four species and a random table you often see rings that orbit a core, chains that wriggle, and small clusters that crawl across the screen at a steady speed. Here is the hook: a cluster can move on its own only because the rules are unfair. If red is pulled toward blue with strength +0.6 while blue is pushed from red with strength -0.4, then red keeps chasing and blue keeps fleeing. The pair drifts. Nothing pushes back on the world, so momentum leaks in one direction and the cluster becomes a swimmer.
When to use it, and when not
Use Particle Life to build intuition about emergence: how local interaction rules with no global plan can produce shapes that look designed. It is an honest example of self-organisation because you can read every rule and still be surprised by the output.
Do not read it as biology. Real cells have membranes, metabolism, and genetics. Particle Life has none of these. The resemblance to bacterial swarms or chemically driven colloids is a resemblance of behaviour, not of mechanism. Treat any "creature" you see as a stable dynamical pattern, not an organism. If you want a model where selection actually acts on heritable traits, that is a different tool.
The pattern you see depends far more on the rule table than on the starting positions. Two runs from different random layouts but the same table usually settle into the same kinds of structures. That is the sign of a genuine attractor in the dynamics.
The force law
Consider two particles, one of species i and one of species j, separated by distance r. Let R be the interaction radius (particles beyond it feel nothing) and let \beta be the fraction of that radius taken up by the hard push. The rule table entry a_{ij} is the attraction strength of i toward j, a number in [-1, 1]. A common force profile is:
Read it piece by piece. For r \lt \beta R the force is negative (repulsive) and rises linearly from -1 at contact to 0 at r = \beta R: this is the short-range push that stops overlap. For \beta R \le r \lt R the force is a triangle scaled by a_{ij}, peaking halfway across the outer band and falling to zero at both ends. Beyond R there is no force at all. The sign of a_{ij} decides whether the mid-range interaction pulls (positive) or pushes (negative).
Each particle sums the force from every neighbour inside R, updates its velocity, and then loses a fixed fraction to friction every step:
Here f is the friction fraction (for example 0.1 means 10 percent of speed drains each step) and \Delta t is the timestep. Friction is what keeps the world from exploding: without it, energy from the asymmetric rules would build up without bound.
Why asymmetry makes creatures move
In real physics, Newton's third law forces a_{ij} = a_{ji}: if red pulls blue, blue pulls red back with equal strength, and the pair's centre of mass never accelerates on its own. Particle Life breaks this on purpose by letting the table be asymmetric, so a_{ij} \ne a_{ji}.
Work through the simplest case. Put one red at position 0 and one blue at position 1, both at rest, with a_{\text{red,blue}} = +0.6 (red is drawn to blue) and a_{\text{blue,red}} = -0.4 (blue flees red). Suppose at this separation the triangular profile gives a raw force of magnitude 0.5 before the table scaling. Red feels +0.6 \cdot 0.5 = +0.30 toward blue. Blue feels -0.4 \cdot 0.5 = -0.20, a push away from red. Blue sits ahead of red, so that push points the same way as red's pull. The net force on the pair is 0.30 + 0.20 = +0.50, which is (a_{\text{red,blue}} - a_{\text{blue,red}}) \cdot 0.5, pointing toward blue. The centre of mass accelerates. A stable red-blue clump becomes a self-propelled dot.
Flip the switch to symmetric rules and the model sets a_{\text{blue,red}} = a_{\text{red,blue}} = +0.6. Now both feel +0.30 toward each other, the net force is zero, and the pair only vibrates in place before friction stills it. That single change is the difference between a moving zoo and a field of static blobs.
If your world looks dead, check the diagonal of the table first. Slightly positive self-attraction (species like their own kind) gives particles something to cluster into. Then add off-diagonal asymmetry to make those clusters move.
A worked example with a cyclic table
Building a chaser from three species
Take three species (red, green, blue) with the numbers used above: interaction radius R = 80, push fraction \beta = 0.3 and friction f = 0.1. In the tool, set Species to 3 and pick the Chains rule set, which builds a table of the same shape with stronger numbers. Set a mildly cyclic table so each colour chases the next:
| toward red | toward green | toward blue | from |
|---|---|---|---|
| 0.4 | 0.6 | -0.3 | red |
| -0.3 | 0.4 | 0.6 | green |
| 0.6 | -0.3 | 0.4 | blue |
- The diagonal is
+0.4everywhere, so each colour weakly clumps with its own kind. Expect small single-colour cores. - Red is drawn to green (
+0.6) but green is pushed from red (-0.3). Red chases green. - Green is drawn to blue (
+0.6) but blue is pushed from green (-0.3). Green chases blue. - Blue is drawn to red (
+0.6) but red is pushed from blue (-0.3). Blue chases red. - The three-step chase cycle has no fixed point, so clumps circle and drift instead of settling. You get roving mixed-colour packs.
Check the net drive on a red-green pair as in the previous section. Take a mid-band raw force of 0.5. Red feels 0.6 \cdot 0.5 = 0.30 toward green; green feels -0.3 \cdot 0.5 = -0.15, a push away from red in the same direction. Net drive 0.30 + 0.15 = 0.45, which is (0.6 - (-0.3)) \cdot 0.5. That surplus, repeated every frame and topped up faster than friction removes it, is what pushes the pack along.
Explore the asymmetry yourself
Reading what you see
A few recurring outcomes tell you what the table is doing.
- Frozen blobs
- Round, static clusters that stop moving. The effective table is close to symmetric, so no net drive survives friction. Increase off-diagonal asymmetry to wake them up.
- Orbiting rings
- One species circles a core of another. This needs positive attraction inward and enough repulsion or asymmetry to keep the orbiter from falling in. Rings mean the system found a stable circulation rather than a fixed point.
- Chasers and worms
- Elongated moving structures. These are the signature of a chase cycle like the three-species table above. Their speed scales with the difference between the paired table entries.
- Uniform gas
- Particles spread evenly and never cluster. Attractions are too weak or friction too strong. Raise the diagonal or the mid-range strengths.
Speed gives you a rough number to check. If the mean drift of your chasers is near zero, the table is effectively symmetric no matter how the numbers look, because on average a_{ij} - a_{ji} cancels across pairs. A table where the upper triangle sums to +2.1 and the lower triangle to -1.8 has a net asymmetry of 0.3 per pair on average, and that surplus is what feeds visible motion.
Common mistakes
Turning friction to zero does not make the world more lively. It makes it explode. With no drain, the energy that asymmetry injects accumulates every frame and speeds run away until the simulation is noise. Keep f around 0.05 to 0.2 for stable, moving structures.
The other frequent error is confusing the two triangles of the table. The entry in row red, column green means "how red responds to green", not "how green responds to red". Swapping them silently makes your intended chaser flee instead. Read a row as "who this species is drawn toward or pushed from".
A subtler trap is the timestep. The integration is explicit, so a large \Delta t combined with strong forces can jump a particle straight through the short-range push and into overlap, where the linear repulsion is weakest. If clusters collapse to a point, shrink the timestep or the force strengths rather than adding more repulsion.
Finally, do not over-read a single lucky run. Because the model is chaotic, one seed may produce a striking creature that never returns. Mutate the table slightly and rerun a few times. If the structure survives small changes to the numbers, it is a real feature of that table. If it vanishes, it was a transient.
Related tools
Particle Life sits inside a family of models where local rules produce global order. For the classic three-rule version of collective motion, see the Boids Flocking Simulator. For a continuous cellular automaton that grows gliding lifeforms from a smooth kernel, try Lenia Continuous Life. Agent-based self-organisation appears again in the Ant Colony Pheromone Trails and the Slime Mould Networks, both of which build transport structure from simple deposition and decay. If you want selection acting on heritable traits rather than fixed rules, look at Evolving Creatures and the Genetic Drift Simulator. For synchronisation from local coupling, the Firefly Synchronization model shows the same emergence in the time domain.
Frequently asked questions
Why do the particles never overlap?
The force law has a hard repulsive band below r = \beta R that grows toward -1 as particles approach contact. That push always beats the mid-range attraction at short distance, so two particles keep a minimum separation set by \beta and the timestep.
Is Particle Life alive in any real sense?
No. There is no metabolism, no replication, and no heredity. What you see are stable patterns of a dynamical system. They resemble life because both are self-organising, but the mechanism is a fixed force table, not biology.
How many species should I use?
Two species can already chase and flee, but three to six give the richest behaviour because they allow chase cycles that never close, like the red to green to blue to red loop. Above six or seven the table has so many entries that random settings tend to average out toward a bland gas.
What does the symmetric-rules switch actually change?
It forces every pair of entries to be equal, setting a_{ij} = a_{ji}. That restores an effective Newton's third law, so the net force on any cluster's centre of mass becomes zero and self-propulsion disappears. The same world that swam before now freezes into static blobs.
Why does the world wrap around at the edges?
Wrapping removes walls, so no particle ever piles up against a boundary and distorts the statistics. A structure that leaves the right edge reappears on the left with the same velocity, which keeps the total behaviour uniform across the whole field.