Computational Irreducibility
The universe runs on a simple rule — yet to know what it does, there's no shortcut but to let it run, step by relentless step.
Here’s a puzzle that sits underneath the whole project. Suppose you knew the exact rule for the universe — the single rewriting rule, written on an index card. Could you then predict what it does a billion steps from now, without the bother of actually running it?
For some systems, yes. If a planet orbits on an ellipse, you can jump straight to where it’ll be next Tuesday with a formula; you don’t simulate every instant in between. That’s a shortcut, and most of physics is the art of finding them.
The claim of computational irreducibility is that, in general, no such shortcut exists. To find out what the system does after steps, you have to effectively perform all steps. The system is its own fastest simulator. This idea isn’t new to the Physics Project — Wolfram carried it over from his earlier A New Kind of Science, where it grew out of watching staggeringly simple rules produce behaviour no formula could anticipate. [setup]
You can’t read the answer off the rule
The previous note, Causal Invariance, showed that the order of rewrites doesn’t change the final causal structure. That might tempt you to think the outcome is therefore “knowable in advance.” It isn’t — knowing that a destination exists is not the same as knowing what it is without travelling there.
Wolfram’s clearest demonstration uses the simplest computer imaginable: a single row of cells, each black or white, each updated from only itself and its two neighbours. The entire law is an 8-bit number. Below, rule 30 starts from one black cell — before you press anything, try to predict what the bottom row looks like.
chaotic — used as a randomness source.Try this: guess first, then press Play. Rule 30 dissolves into a pattern with no visible order — so irregular it has been used as a randomness source — yet it is completely deterministic. Flip to rule 90 (a clean fractal) or rule 110 (proven to be a universal computer) to feel how wildly the same tiny setup can behave. Your only honest route to the bottom row was to compute every row above it.
The same is true of the hypergraph rules that actually build space in this model. Here each edge becomes a little triangle with a new node — try once more to call the generation-5 shape before stepping:
{x, y} → {x, y}, {y, z}, {z, x} — each edge becomes a triangle with a new nodeNotice what just happened. The rules are trivial. You hold them in full. And yet your only reliable route to the later structure was to let them unfold — to do the computation. There was no peeking ahead.
Why a lawful universe still surprises us
This is Wolfram’s proposed answer to an old tension: how can a universe be both completely rule-based and genuinely unpredictable in detail? The resolution is that determinism and predictability are different things. The rule fixes the future entirely — but extracting that future requires as much computational work as the universe itself is doing, so nothing inside the universe can race ahead of it. [setup]
From that single idea, several familiar features start to look less mysterious:
- Apparent randomness. A sequence produced by an irreducible process can pass every test for randomness while being fully determined. What we call “random” may often be deterministic-but-irreducible — there’s simply no compressing description shorter than the process itself. [setup]
- An arrow of time. Because there’s no shortcut to the end state, the system has to be run forward, step after step, to get there. An observer who is themselves computationally bounded can only coarse-grain what they see — and from that, Wolfram argues, the Second Law of thermodynamics (the one-way slide toward disorder) emerges. [derived-in-model]
A skeptic should hear the seams here: that apparent randomness comes from irreducibility is an interpretive stance carried from A New Kind of Science, and the Second Law story is a model-internal derivation, not a settled theorem about our universe. The grounded version is below.
The precise version (and why physics is still possible)
Computational irreducibility, in Wolfram’s words, names outcomes that are “obtainable only by running the system step by step” — there is no general procedure that reaches step in substantially fewer than operations. [setup] It’s a property of the process, not a limit of our cleverness.
This raises an obvious worry: if the universe is irreducible, how is any physics possible? Wolfram’s answer is that irreducibility is the generic case, but it leaves room for pockets of computational reducibility — special slices where regular, predictable behaviour does emerge and a shortcut does exist. He goes further and proposes that the great reducible pockets we’ve found already have names: general relativity, quantum mechanics, and statistical mechanics are each cast as an island of predictability inside an otherwise irreducible world. [conjecture]
So the picture is not “everything is unpredictable.” It’s “almost everything is irreducible, and science lives in the rare reducible pockets” — which is also why a final theory could be simple yet leave most of what happens forever requiring simulation to know. [conjecture]
Grounded in Wolfram’s 2020 announcement and the 2021 one-year update; the lineage to A New Kind of Science is noted in the book-length treatment.
The next note, The Principle of Computational Equivalence, asks the natural follow-up: if simple rules are this powerful, how powerful — and is our universe’s rule special, or just one capable computer among countless equals?
Sources for this page: 2020 announcement · 2021 one-year update · A Project to Find the Fundamental Theory of Physics