Replicator: Everything Copies Everything
Place a single live cell on an empty grid under the Replicator rule — B1357/S1357 — and watch what happens.
After 2 generations: 5 cells. After 4: a structured cluster. After 8: the cluster has begun producing copies. After 32: 8 distinct copies of the original single cell, arranged at the corners and edge midpoints of a large square. After 64: the pattern has the fractal structure of a Sierpinski triangle, with copies multiplying at every scale.
Now place your initials in binary on the grid. Same thing happens. Place a glider. Same thing. Place a random scattering of 10,000 cells. Same thing — 8 copies of that 10,000-cell scattering appear, each at a distance of 2ⁿ cells, each proceeding to make 8 copies of themselves.
Every pattern replicates. Not one special object, not an engineered machine — every single finite pattern that can be placed on the grid will eventually produce infinite copies of itself, arranged in a fractal lattice stretching across the infinite plane.
The rule was studied by Edward Fredkin, and it is sometimes called Fredkin’s rule. Understanding why it works requires understanding one of the most elegant pieces of mathematics in cellular automaton theory.
The Rule: B1357/S1357
The Replicator rule is:
- A dead cell is born if it has 1, 3, 5, or 7 live neighbors.
- A live cell survives if it has 1, 3, 5, or 7 live neighbors.
The pattern is immediate: cells are born or survive when their live neighbor count is odd. Equivalently: the rule fires when the parity (odd/even) of the live neighbor count is 1.
The birth and survival lists are identical: both are {1, 3, 5, 7}. This makes the rule symmetric between birth and death in a different way from Day & Night’s symmetry: in Replicator, alive and dead cells with the same neighbor count are treated identically. A live cell and a dead cell, each with 3 live neighbors, both survive/are born. A live cell and a dead cell, each with 2 live neighbors, both die/stay dead.
This odd-parity condition is the key to the universal replication property. It is not chosen arbitrarily — it is the specific condition that makes the rule equivalent to XOR arithmetic, which is what makes replication universal.
The Mathematics: Why Every Pattern Replicates
The proof of universal replication is surprisingly clean, and it repays careful attention.
Step 1: The XOR property.
Because the rule fires on odd neighbor counts — that is, when the sum of neighbor states is odd — the next state of any cell is determined by the XOR (parity) of its neighbors. XOR is an operation with a crucial linearity property: XOR(A + B) = XOR(A) XOR XOR(B), where we’re thinking of patterns as sets of live cells and XOR as symmetric difference (a cell is in A XOR B if it’s in exactly one of A or B).
This means: if you have two patterns A and B that don’t overlap, the n-th generation of (A union B) is equal to the XOR of the n-th generation of A alone and the n-th generation of B alone. The evolution of the combined pattern can be decomposed into the independent evolutions of its parts.
Step 2: A single live cell at generation 2ⁿ.
A single live cell at position (0, 0) at generation 0 produces, at generation 2ⁿ, exactly 8 live cells at positions (±2ⁿ, 0), (0, ±2ⁿ), (±2ⁿ, ±2ⁿ) — the corners and edge midpoints of a square of side length 2ⁿ⁺¹. When n is large, these 8 cells are arbitrarily far from each other and from the origin, meaning they form 8 isolated single cells, each capable of independently continuing to replicate.
Step 3: Any finite pattern replicates.
Any finite pattern is a finite union of single live cells (by definition — it’s just a set of cells). By the XOR superposition principle, the n-th generation of the whole pattern is the XOR of the n-th generations of each of its single cells. At generation 2ⁿ, each single cell has produced 8 copies of itself at positions ±2ⁿ along each axis. The XOR superposition means the whole pattern has produced 8 copies of itself at those same displacements — each copy is the superposition of the copies of each individual cell, which is the pattern.
When n is large enough that the copies don’t overlap (which happens once 2ⁿ is larger than the pattern’s diameter), the 8 copies are exact, non-interfering duplicates. Then each copy undergoes the same process, producing 8 copies each — and so on, forever.
The result is a fractal lattice: copies of the original pattern at every scale, arranged in a structure that resembles the Sierpinski triangle (or more precisely, the Sierpinski carpet, since we are in two dimensions with 8-fold replication).
This is not an approximation or a limiting behavior. It is exact, provable, and applies to literally any finite initial pattern, regardless of its size or structure.
What Replication Looks Like
The visual experience of watching the Replicator rule unfold is striking.
Start with a single cell. For the first few generations, it evolves like any other Life-like pattern — a small cluster of cells, updating locally, looking roughly symmetric. Then, around generation 4 to 8, you notice that the cluster seems to be trying to split — smaller clusters appearing at the edges. By generation 16 or 32, the fractal structure is unmistakable: the original cluster has become a growing lattice of copies, each copy separating from its neighbors as the replication distance 2ⁿ grows.
The copies are not all equally visible at any given moment, because during the copying process (the generations between 2ⁿ and 2ⁿ⁺¹), the copies interfere with each other, producing complex internal structures. The clean copies only fully emerge at the exact generations 2ⁿ, not in between.
Start with a more complex pattern — a recognizable shape — and the replication becomes more visually dramatic. At generation 2¹⁰ = 1024, your pattern has produced 8¹⁰ = 8 billion copies of itself, spread across the grid. At generation 2²⁰, the number of copies is astronomical.
The Paradox of Trivial Replication
Here is where the Replicator rule becomes philosophically interesting rather than just mathematically elegant.
Self-replication has a special status in the theory of computation and biology. John von Neumann spent years designing a self-replicating cellular automaton, because he believed (correctly) that proving replication was possible in a simple substrate would illuminate the mechanisms of biological replication. The design required 29 cell states, elaborate machinery, and the equivalent of a universal constructor — a machine that could build any object given a description. The point was not just to replicate, but to show that replication could be achieved through construction: reading a blueprint, interpreting it, and building the described object.
The Replicator rule makes von Neumann’s feat look like overkill. Every pattern replicates automatically, without any machinery, without any blueprint, without any universal constructor. Replication is not an achievement — it is a default. The rule doesn’t solve the problem of self-replication; it makes the problem disappear.
This should, on one reading, be deeply impressive: look how easy self-replication is! But it actually reveals something subtler: the Replicator rule’s replication is a completely different kind of thing from what von Neumann was studying.
Von Neumann’s self-replicating machine can replicate because it can build anything. Its replication is a special case of general construction. The replication is hard-won because the generality is hard-won. If you modified the machine to build only copies of itself — hardwiring the blueprint — you would lose the universality while keeping the replication. The replication would be cheaper; the machine would be less interesting.
The Replicator rule is that cheaper version, taken to an extreme. The “replication” is a pure XOR artifact — it doesn’t mean the pattern has any organizational sophistication. A random scattering of cells replicates just as faithfully as an intricate structure. The replication carries no information about the structure’s function or complexity. It is structural, not functional.
Comparison with HighLife’s Replicator
The contrast with HighLife’s replicator is clarifying.
HighLife’s replicator is one specific pattern that replicates in a rule that is not generally replicating. The replicator exploits the birth-at-6 condition of HighLife to copy itself — a mechanism that only works for that specific pattern (and patterns derived from it). Most HighLife patterns do not replicate. The replication is a property earned by one special configuration.
The Replicator rule’s replication is universal — it applies to every pattern, including trivial ones. It is a property of the rule, not of any particular configuration.
The HighLife replicator is harder to find, more interesting, and more revealing — precisely because it is rare. When a pattern replicates in a rule that doesn’t make replication easy, it tells you something about the pattern. It tells you that the pattern has a specific geometric self-consistency with the rule, that its structure is robust enough to propagate itself in a hostile environment. It tells you that the pattern has earned its replication.
When every pattern replicates by default, replication tells you nothing about the pattern. It tells you only about the rule.
This is the deep point: the Replicator rule trivializes self-replication, and in doing so, reveals that trivial self-replication is not the interesting phenomenon. The interesting phenomenon — the one von Neumann was studying, the one that connects to biological evolution, the one that enables open-ended complexity — requires that replication be difficult. Easy replication, universal replication, is mathematically beautiful but biologically and computationally empty.
Historical Context: Fredkin and Parity Rules
Edward Fredkin’s interest in the parity rule emerged from his broader research program on reversible computation and the physics of information. Fredkin was one of the pioneers of digital physics — the idea that the universe might be fundamentally computational, with physical reality arising from an underlying cellular automaton. He was interested in rules with specific mathematical properties: reversibility, conservation laws, time-reversal symmetry.
The parity rule satisfied several of these aesthetics. It was simple to state. It had elegant mathematical structure. Its replication property was provable by elementary argument. Terry Winograd proved in 1970 that Fredkin’s parity rule could be extended to N states as long as N is prime — a generalization that preserves the XOR structure. The rule became a standard example in the study of reversible and conservative cellular automata.
The name “Replicator” for the rule B1357/S1357 is somewhat informal — it’s used in the Life-like CA community (as catalogued in the LifeWiki) to distinguish this specific rule from Fredkin’s original parity-rule formulation (B1357/S02468, which is different). The B1357/S1357 rule shares the universal replication property because both birth and survival are triggered by odd counts, making the XOR superposition argument apply to both transitions.