Papers · Preprint · v3

To Plot a Stone with Six Birds: Constructing and Auditing Emergent Geometry from Markov Dynamics

Distance and curvature from a Markov chain, with no coordinates assumed.

In plain words

Points are groups of microstates under a lens. Distance is the cost of the cheapest chain of macro transitions, each step costing the negative logarithm of its probability. Nothing else is assumed, and the paper proves in Lean that this always gives a valid distance.

A layer counts as coherent when its closure, persistence, route and distortion defects all vanish along a declared refinement. The paper builds coherent layers for three systems: block lenses on a grid walk, whose distances converge to taxicab distance on the unit square; recursive cells on a Sierpiński gasket walk, converging to a self-similar fractal; and reservoir chains on the sphere, converging to the round sphere, whose curvature a landmark transport estimator recovers.

Three walks lead through refinement at τ = 5 to three different geometries: a lazy grid walk to a taxicab square with an equal cost diamond, a gasket walk to a self-similar fractal of dimension log 3 / log 2, and reservoir chains to a round sphere whose curvature 1 is recovered.
Refining each walk and its supplied lens, with distance measured as the cost of the cheapest chain of transitions, yields a taxicab unit square, a self-similar fractal and a round unit sphere. At τ = 5 the four coherence defects vanish together along the refinement; the fractal has dimension log 3 / log 2, and a landmark transport estimator recovers the sphere's curvature of 1.

For the walk on the lattice, the cost of one long move converges to a quadratic form, so its square root is straight-line distance and Pythagoras becomes a statement about costs. The paper also marks the limits: spectrally learned lenses give valid distances but fail persistence, and a natural local estimator of curvature overestimates it by more than 20% even on exact spherical distances.

What it shows

  • A distance built from transition costs alone, always a valid metric, proved in Lean.
  • Coherent layers converging to the taxicab square, a Sierpiński gasket and the round sphere.
  • A Pythagorean cost law for long moves of the lattice walk.
  • Limits: learned lenses that fail persistence, and a curvature estimator that converges to the wrong value.

What it does not claim

Geometry here is a closure artifact of finite dynamics, not a claim about physical space.

Cite

Tsiokos, I. (2026). To Plot a Stone with Six Birds: Constructing and Auditing Emergent Geometry from Markov Dynamics. Zenodo. https://doi.org/10.5281/zenodo.23130783