Life from a Table of Numbers
A few kinds of beads in a glass tank. Each kind has one number for each kind, attract or repel, and nothing else. A symmetric table lets them line up and stop. A lopsided one grows things that walk, chase, split and merge.
Scroll
One Number
The tank holds a single kind of bead. Between any two beads there is one thing only: a push or a pull along the line joining them, felt within a radius rMax and not at all beyond it.
The shape of the force is fixed. Closer than 0.3 rMax, two beads always push apart, whatever their kind. Further out comes a tent-shaped pull whose height is the number a. A positive a attracts; a negative one repels.
At every step each bead adds up its neighbours’ pushes and pulls, changes its velocity a little by the total, and moves a little. Its speed halves every 0.04 seconds, as if in a thick liquid. Here a = +0.6: the beads gather into round drops and come to rest.
A Table
More kinds, more numbers. With three kinds of bead, each kind has a number for each kind, itself included: nine in all, in three rows and three columns. Row i, column j says how kind i treats kind j.
The knobs on the console are this table. The cup at the head of a row holds the kind that feels; the cup at the head of a column, the kind that is felt. A pointer turned right pulls, turned left pushes; the lamp under it glows warm for a pull and red for a push.
This table is symmetric: red towards teal is the same number as teal towards red. The beads sort themselves into layers, one wrapped round another like liquids that will not mix. Once sorted, they stay still.
The Third Law
Newton’s third law: the forces two bodies exert on each other are equal and opposite. With a symmetric table every pair’s push and pull cancel, the beads’ forces on one another add up to zero, and they cannot change the swarm’s total momentum; only friction can, and friction only drags it towards zero. Such a system has an energy; friction drains it bit by bit until everything stops.
Nothing makes the table symmetric. The bat-handle switch at the top left of the console flips between two tables: the original, and its symmetric part, each pair of numbers averaged. The meter reads what is left when all the beads’ forces on one another are added up, as a share of the sum of their sizes; its scale is logarithmic, each major division ten times the last.
Every few seconds the four kinds in the tank change tables. With the symmetric one they fall still within seconds and the meter reads zero. Back on the original, the pushes no longer cancel; the beads move again, and keep moving.
Chase and Flight
The simplest lopsided table needs only two kinds. Red towards teal is +0.6, teal towards red −0.6: red is drawn to teal, teal is pushed away by red. Red chases, teal flees. A few red beads nestle into a crescent of teal; the teal moves off, the red follow, and the little clump runs away together like a comet with a tail.
Nobody fitted them with motors. They run only because the two beads push each other with different strengths: the total force is not zero, so they speed up until friction holds them back. The total momentum in the readout is no longer zero.
Membrane and Core
Five kinds, a random table, nothing else changed. The tank fills with small units: one kind gathered in the middle, another drawn round it in a thin ring, like a core and a membrane.
They shuffle about and shoulder each other aside; knocked apart, they gather again, while loose beads keep streaming between them. The rule has no cells and no membranes in it; those names are ours. All that can be written down is the five-by-five table.
The Specimen Rack
One rule; change the table and something else grows. Each of the four watch glasses in the rack holds a structure grown from one table, at the same scale as the tank.
The tank brings them out in turn: comets running with a tail, a unit with a membrane, worms drawn out long and writhing, and rings within rings. The names are ours; there is no agreed classification of these things.
Any Table at All
Set a table at random and, more often than not, something moves. The figure below is our own run: a hundred random tables for four kinds, each started from the same scatter of 1,500 beads and run for 20 seconds, with the mean speed taken over the last second. Of the hundred, 98 were still going faster than 2 mm a second.
The same hundred tables, each reduced to its symmetric part and run again: not one went faster than 2 mm a second, and all were winding down. Whatever keeps moving has a lopsided table.
Lopsided in Nature
Between two real particles, action and reaction are always equal. But when the interaction passes through something else, such as a flowing plasma, a flowing liquid, or a chemical that one releases and the other takes up, what each particle feels can differ.
A 2015 study worked out such mixtures and tested them on two-dimensional bilayers of charged microparticles in a plasma: the two kinds settle at different temperatures. In 2020 chemists watched one kind of oil droplet chase another through a surfactant solution while the other fled, driven by a net flow of oil from one droplet to the other.
A 2021 paper brought lopsided interactions into the theory of phase transitions: with them, an ordered state can keep moving for ever, two groups chasing each other round in a circle, for instance. In the tank three kinds chase head to tail, the first the second, the second the third, the third the first. That is not the paper’s model, only the same kind of lopsidedness.
Where It Came From
In the 1987 “boids”, each bird looks only at a few neighbours, matching their heading, closing in and keeping clear, and the flock forms by itself. Particle life is simpler still: no heading, no field of view, only pushes and pulls.
Handing many particles a lopsided table of attractions comes from a program of the 2010s called Clusters, which its author says he first put online around 2016. Many versions followed under the name particle life; the shape of the force and the constants here are taken from one published version.
The Bench
One table, up to six kinds of bead. Every number can be set; beads can be brushed into the tank; a table can be saved as a line of characters and read back.
Try thisGive two species opposite numbers for each other, then switch on “symmetric table only”: does anything still run? What does the total momentum do?
What the model leaves out
The tank is a square of side 2, 52 cm across on the model. Its opposite edges are joined: a bead leaving on the right comes back on the left, as is usual in particle-life programs, so the walls never push and the total momentum changes only through the beads’ forces on one another and friction. rMax = 0.15, which is 3.9 cm on the model; the beads are drawn 9 mm across only so they can be seen: in the rule a bead is a point with no size, and the spacing comes from the short-range push. Friction half-life 0.04 and force factor 10 are those of the published code; the time step is 0.01, half of its 0.02. At 0.02 a dense clump under a symmetric table never comes to rest: each step is too long, overshoots the steep push up close and hands the energy back, so a symmetric table would seem to keep moving for ever. The published code’s default rMax and world size also differ from ours. Each step runs on your graphics card and finds neighbours through a grid of cells, giving the same sums as checking every pair on the CPU (checked in the tests and on the page). Each chapter’s state was computed on the CPU beforehand and saved; on arriving at a chapter it is placed in the tank, and from then on everything is computed live. The materials, lamps and meter are display.
A common misreading
“Particle life breaks the laws of physics” is not quite right. The beads in the rule are not isolated bodies: a lopsided table amounts to every bead quietly borrowing momentum from a medium around it. Real non-reciprocal interactions work the same way; there is always a third party, a flowing gas or liquid or a chemical fuel, to settle the account. A driven system does not break the conservation of momentum; it only looks that way if you count the particles alone.
Sources
- Jeffrey Ventrella, Clusters (program and notes)
- Jeffrey Ventrella, an introduction to Clusters (updated 2025)
- Tom Mohr, particle-life-cli (the force law and constants; MIT licence)
- Tom Mohr, particle-life-app (GPL-3.0)
- Hunar Ahmad, particle-life (MIT licence)
- Craig W. Reynolds (1987), Flocks, herds and schools: A distributed behavioral model, Computer Graphics 21(4): 25–34
- A. V. Ivlev et al. (2015), Statistical Mechanics where Newton’s Third Law is Broken, Physical Review X 5, 011035
- C. H. Meredith et al. (2020), Predator–prey interactions between droplets driven by non-reciprocal oil exchange, Nature Chemistry 12: 1136–1142
- M. Fruchart, R. Hanai, P. B. Littlewood, V. Vitelli (2021), Non-reciprocal phase transitions, Nature 592: 363–369
- S. Saha, J. Agudo-Canalejo, R. Golestanian (2020), Scalar Active Mixtures: The Non-Reciprocal Cahn-Hilliard Model, Physical Review X 10, 041009
- R. Soto, R. Golestanian (2014), Self-Assembly of Catalytically Active Colloidal Molecules, Physical Review Letters 112, 068301
- A. Kumar et al. (2024), Automating the Search for Artificial Life with Foundation Models, arXiv:2412.17799
Data
The force law and its constants (β = 0.3, friction half-life 0.04, force factor 10) follow the published particle-life code by Tom Mohr (github.com/tom-mohr/particle-life-cli, MIT licence); we halve its time step of 0.02 to 0.01. Our tank, grid and every table on the page are our own. Particle-life systems with an asymmetric attraction table go back to Jeffrey Ventrella’s Clusters (ventrella.com/Clusters). The names of the structures in the specimen rack are ours; the rule has no notion of cells or worms.
The text, pictures and models on this page are licensed CC BY-NC 4.0: free to use and adapt for teaching and other non-commercial purposes, with credit to world.lapla.si. Third-party material listed above keeps its own licence.

