Frequently asked questions

Not in any normal sense. There are no players, no moves and no way to win. It is a 'zero-player game': you choose the starting configuration and then you are a spectator. Conway called it a game because it came out of the same period of thinking about combinatorial games, and because 'game' invited people in where 'two-dimensional two-state cellular automaton with a Moore neighbourhood' would not.

Conway spent roughly eighteen months testing rules on a Go board looking for one that satisfied three conditions: no small pattern should obviously grow forever, some patterns should nonetheless appear to, and small patterns should be able to run a long time before their fate becomes clear. Most rules fail all three - they either die within a few generations or fill the plane with static. B3/S23 sits exactly on the boundary.

Yes, and this is a proved result rather than a metaphor. Gliders act as signals; guns produce them on demand; eaters and blocks act as gates. Logic gates were demonstrated in the early 1970s, a sketch of universality was published in Winning Ways in 1982, and Paul Rendell built an explicit Turing machine inside Life in 2000, extended in 2010 with an unbounded tape. Anything your computer can compute, a sufficiently large Life pattern can compute - just very, very slowly.

Yes. In 2006 Brice Due completed the OTCA metapixel, a 2048x2048 Life pattern that behaves exactly like a single Life cell, including a configurable rule. Tile the plane with metapixels and you are running Life inside Life, at a cost of 35,328 generations per meta-generation. The construction is fully recursive, so you can go as deep as you have patience and memory for.

No. Guns, puffers and breeders grow forever, and there is no general algorithm that can tell you in advance whether an arbitrary pattern will settle. That is the halting problem, and because Life is Turing complete it applies here directly. For most patterns the only way to find out what happens is to run it - what Wolfram calls computational irreducibility.

Because determinism and predictability are different things. Life is completely deterministic - the next generation is a function of this one, with no randomness anywhere. But for most patterns there is no shortcut: the fastest possible way to know the state at generation ten million is to compute all ten million generations. Determinism guarantees that the future is fixed. It guarantees nothing about your ability to see it in advance.

A configuration with no possible predecessor: you can draw it, but it can never arise from any earlier generation. Their existence in Life follows from the Moore-Myhill theorem of 1962, and the first explicit example was found in 1971. The smallest known ones are now surprisingly compact.

One cell per generation, called c by analogy with the speed of light. Information cannot travel faster because a cell's next state depends only on its immediate neighbours. A glider manages c/4 diagonally; the classic lightweight, middleweight and heavyweight spaceships manage c/2 orthogonally. Slower speeds are extremely varied - the copperhead, found in 2016, travels at c/10.

Constantly. Distributed soup-searching projects have processed trillions of random starting configurations and still turn up objects nobody has catalogued. The copperhead (2016) was a genuinely new spaceship speed class after forty-five years. Sir Robin (2018), the first elementary knightship, was found with SAT solvers. New oscillator periods are still being closed off. The space is effectively infinite and the tools keep improving.

Because the alternative is worse. Life is defined on an infinite plane and no computer has one, so every implementation compromises. A bounded grid mutilates anything that reaches the edge - a glider hitting a wall usually collapses into a block. Wrapping the edges into a torus at least keeps everything alive and moving. You can switch it off in the World panel if you want hard edges.

Three things. Cell states live in flat typed arrays with a reusable write buffer, so stepping allocates nothing. Neighbour indices are precomputed into lookup tables when the grid is created, removing modulo arithmetic from the inner loop. And each step only visits the bounding box of the live population plus a one-cell margin, so a glider in a 640x400 world costs about the same as a glider in a 40x40 one.

Run Length Encoded is the plain-text interchange format Life software has used since about 1980. 'b' is a dead cell, 'o' is a live one, '$' ends a row, '!' ends the pattern, and a number repeats the following symbol - so '3o' means three live cells. Everything on this site is stored in it, which means anything here can be pasted into Golly or any other Life program and vice versa.

Yes. The classic patterns are decades-old community knowledge and are not owned by anyone. Every pattern page has a download button that produces a standard .rle file with attribution in the comment headers.

Directly, less than the name suggests - it is not a model of any specific biological system. Indirectly, a great deal. Von Neumann's automaton derived the logical structure of self-reproduction before DNA's structure was known. Cellular automata are used seriously to model excitable media such as cardiac tissue, epidemics, forest fires, traffic flow, crystal growth and tumour development. And the pigmentation of the Conus textile sea snail's shell really is produced by a process closely resembling elementary automaton Rule 30.

He was ambivalent to the point of irritation. He acknowledged that it had introduced more people to mathematics than anything else he did, but he considered his work on surreal numbers, sporadic groups and combinatorial game theory far more important, and disliked being introduced as 'the Game of Life man'. He made a certain peace with it late in life. There is a good account of the tension in Siobhan Roberts's biography, Genius At Play.