Physics· Research · arXiv preprint in flight

Substrate Geometry

A research program that classifies shapes by what they guarantee under physical operation rather than by their symmetry. The working hypothesis: the oloid, a 3D shape discovered in 1929, is an optimal geometric primitive for rolling contact under stress.

How the Garden philosophy shows up here

This is the Gardener framework rendered into physical geometry. Instead of specifying a shape and checking if it works, the program generates 1,430 parametric variants and lets physics-based oracles tell you which geometries survive which regimes of stress. The shape isn't designed — it's selected by the environment. That's what a Gardener's proof looks like when the soil is Navier-Cauchy elasticity and the weather is fatigue cycling.

The thesis

Most geometric primitives used in engineering (cylinder, sphere, torus) were selected because they're easy to parameterize, easy to machine, and have closed-form contact solutions. None of those reasons is about physical performance under load. They're about computational and manufacturing convenience.

The oloid — a 1929 shape constructed from two perpendicular circles of equal radius with the shared tangent line — doesn't look like the kind of thing you'd pick for a bearing. But when you feed it through the oracle stack, it outperforms cylinders by 58-68× on shared physics dimensions: more even contact distribution, lower peak Hertz stress, lower thermal gradient at the contact line, slower wear progression.

The interesting claim isn't just "the oloid is good." It's that we've been picking engineering primitives for the wrong reasons and that a systematic shape-classification framework based on physical invariants would produce very different defaults.

The oracle stack

Seven physics-based oracles evaluate each candidate geometry independently. Contact distribution (how the load spreads). Hertz stress (peak contact pressure). Thermal (how heat flows at the interface under repeated contact). Wear (volume loss per cycle under the Archard model). Fatigue (cycles to crack initiation). Two more I'm keeping out of the public summary until the arXiv preprint lands.

Each oracle is a physics simulator in its own right, independently validated. The parametric search runs 1,430 variants through all seven and produces a two-tier invariant vector: four linear metrics that cluster tightly at 8×10⁻⁷, and three nonlinear fatigue metrics that diverge sharply. That divergence is the proof that the winning geometry is doing something that isn't reducible to the linear metrics — which is the interesting finding.

Current state

The LaTeX preprint is written and the validation runs are complete. The paper has been in outreach to Hellmuth Stachel (Professor Emeritus at TU Wien, one of the few living geometers who has published on the oloid), who provided the contact information of the grandson of the original figure who constructed the oloid in 1929. That lineage handoff is not something I expected to earn this year.

The preprint is waiting on a final pre-submission review before landing on arXiv. The outreach momentum matters more than the publication timeline — the real validation happens when someone with geometric intuition reads the paper and tells me what I missed.

What this substrate teaches

Physics is the purest test of the Gardener framework because the medium refuses to cooperate with Architect-style specification. You cannot tell an oloid to be optimal. You can only build the conditions under which optimality is detectable, then watch which shapes survive.

The lesson that transfers back to the other Deep Synthesis projects: every domain has a set of physical invariants you're either honoring or ignoring. The Gardener move is to make the invariants visible and let the medium arbitrate. The Architect move is to make the invariants conform to the spec.