Interactive · Field Notes · June 2026

The Buckyball Edge Automaton

A cellular automaton on the exact C60 truncated icosahedron — symbols on the faces, rules on the edges, twelve pentagons that curvature forces into being. Below the toy is an honest walk of where chasing it led: from a sphere that can only count, through added handles, to a homotopy-protected universal reversible computer — and the topological-computing ideas underneath it.

Field notes — what chasing this actually found

This page is also a record of an exploration, kept honest about what was discovery and what was rediscovery. The toy above was the starting point. The thread below is where it went.

The object

The buckyball is a truncated icosahedron: exactly 60 vertices, 90 edges, 32 faces — 12 pentagons and 20 hexagons — with full icosahedral symmetry. Because every vertex looks identical (two hexagons and a pentagon meet at each), the edges fall into only two orbits: 60 pentagon–hexagon edges and 30 hexagon–hexagon edges. That is why the automaton needs only two rules. The twelve pentagons are not a design choice. Euler's formula forces exactly twelve of them onto any sphere built this way: they are topological defects, the only sites where the geometry cannot be locally flat.

What the defects do

Run the automaton and the natural question is whether activity collects on those twelve pentagons. It mostly does not — a careful per-face measurement shows the dynamics mildly avoid them. What the defects actually set is global: every vertex on a C60 is identical, so every local plaquette carries the same holonomy, and the defects fix a global twist the sphere must close around. The honest version of the romantic claim ("curvature defects act as memory") is narrower and truer: the defects set a global constraint, not a local preference.

The mechanism is not new. Following the thread upward — defects, braiding, holonomy that computes — leads straight into established mathematics: topological defect braiding (Kitaev, Teo–Roy–Chen, the genon literature), and the connection between holonomy structure and computational power that Krohn–Rhodes theory has carried for sixty years. The genuinely defensible seam is small and specific: finite-order holonomy behaves like a finite-state machine, while faithful unbounded computation requires infinite-order holonomy. A bridge between known results, not a new continent — and worth saying plainly rather than overselling.

From a sphere to a computer, one handle at a time

The real thread was computational. Each step added exactly one structural ingredient, and each claim was executed in code rather than asserted — seven reproducible scripts by the end. The arc:

  1. Sphere → it counts, it cannot compute. The holonomy on a sphere is abelian — it can measure enclosed area but carries no order-dependence, so no real computation lives there.
  2. The sphere as an error-correcting code. The C60 face structure reads exactly as a [32, 9.5, 8] code over ℤ₄ (proven via Smith normal form), and the twelve pentagon defects own the failure modes. The geometry is a code.
  3. Torus control. On a torus the concentration of activity is geometric, 4.4× — establishing causation, not correlation, between curvature structure and dynamics.
  4. Genus-2 → it computes. Add handles and the holonomy goes non-abelian: braiding loops around the handles transforms data, order matters, and the result is protected by homotopy (contractible detours do nothing).
  5. Genus-2, full search → universal. A four-generator flat connection realizes the entire affine group Aff(2, ℤ₄) — all 1536 elements. Any invertible affine transform of the token is a braid word. A universal gate set for a concrete reversible model, encoded purely in how loops wind around handles.
  6. Genus-2 + junctions → Turing-complete. The one new primitive — state-dependent loop selection, where the token's value chooses which loop it traverses next, driven by holonomy triviality — supplies control flow. That turns the universal gate set into a programmable machine: a reversible counter machine, with multiplication verified in code.

The single move, repeated until it bit: add a handle, then don't artificially restrict the generators. The distance from a glowing buckyball image to a universal reversible processor is that move, six times.

What it means — topological and quantum computation

This is the classical skeleton of the architecture that makes topological quantum computing robust: computation carried by braiding and protected by homotopy, so local noise that doesn't change a loop's topology can't corrupt the result. That fault-tolerance-by-topology is exactly why the field (Kitaev's anyons, the genon literature) pursues it. What was built here is the classical, exact version of that skeleton over a small ring — proven to reach everything in its model — not a quantum device and not a claim of quantum speedup.

The honest ceilings, stated plainly. Three boundaries keep the claim defensible. It is classical — Turing-completeness is the bar for general computation and this clears it, but no quantum advantage is claimed. Completeness needs unboundedness — true Turing-completeness requires counters that grow without limit, which on a finite surface means unbounded winding; the architecture is complete, any physical instance is bounded (the same caveat every real tape carries). And the mechanism is not new: defect-holonomy braiding runs through Kitaev, Teo–Roy–Chen, and the genon work, and the link between holonomy structure and computational power is sixty years of Krohn–Rhodes theory. The one genuinely defensible seam is narrow and worth stating exactly: finite-order holonomy behaves like a finite-state machine; faithful unbounded computation requires infinite-order holonomy. A bridge between known results, not a new continent.

The shape that survived

Past the computer, one transferable idea remained: a plastic connection — a medium where transport rewrites the medium and the rewritten medium conditions later transport. Process and memory as one substance. Tested against neural networks it ties a fixed baseline on regular meshes while freely adapting, and on metric-free graphs with a curvature prior it recovers structure where the standard machinery has no metric to work from; a graph-neural-net holonomy classification, by contrast, was a clean null after three attempts. It earns its keep where there is a gap rather than a crowded baseline. That same plastic-connection shape turned out to be the missing formal mechanism behind one of this corpus's own papers, the Meteorology of Information Propagation — a curiosity worth following for anyone who wants to chase the abstract version. But the load-bearing result here is the computer.