The sphere: Kuramoto with room to moveSection 1 of the Lohe module — the same coupling move, one dimension at a time, until the circle becomes a hypersphere

This module is the summit of the curriculum: the machinery that connects the firefly physics of primer 00 to what a transformer is actually doing, token by token, layer by layer. This first section makes one move only — taking the oscillator off its circle and giving it a sphere — and makes sure you can see that nothing essential changed and one essential thing appeared. It assumes phase and coupling (00) and the dot product (01).

1. From the circle to the sphere

Kuramoto's oscillator — the one you dragged and heard in the earlier primers — is a point on a circle, pulled toward its neighbors. The circle was never the important part. The important part was the move: look at your neighbors, feel how far you are from agreement, and slide a little toward them — without ever leaving the surface you live on.

In 2009 Max Lohe asked what happens if the state is not an angle but a unit vector — an arrow of length one, living on the surface of a sphere. In two dimensions the sphere is the circle and you get Kuramoto back exactly. In three dimensions it is the globe. In five hundred dimensions it is a hypersphere no one can picture — and the update rule does not care. It is the same in every dimension:

Move toward the average of your neighbors; then re-project back onto the sphere.

The same sentence, in symbols (skippable) xi ← normalize( xi + ε Σj Kij ( xj − (xi·xj) xi ) ).  Here xi is oscillator i's unit vector, Kij is how strongly j pulls on i, and ε is a small step. The odd-looking subtraction is doing one honest job: xj − (xi·xj)xi is the part of the pull toward xj that lies along the sphere's surface — the dot product from primer 01 measures how much of xj points "straight out" at i's position, and that part gets removed so the motion stays on the surface. Then normalize cleans up the rounding. That is the whole model.

Everything you learned on the circle survives the trip. Weak coupling: everyone wanders by their own drift. Strong coupling: the population locks. The transition is sharp. But the sphere has something the circle barely had: room. On a circle, agreeing means one thing. On a sphere there are many directions to agree in — so a population can split into several internally-locked clusters that sit at different places on the surface. Structured synchrony, the thing the whole library keeps returning to, becomes geometrically visible.

2. See the generalization: one knob, two worlds

Don't take the paragraph above on faith — watch it. Below, the same experiment runs twice at once: on the left, twenty-four Kuramoto oscillators on the circle you know from primer 00; on the right, twenty-four Lohe oscillators on the sphere. One coupling knob drives both. Slide it from zero and watch the two worlds cross their thresholds together — same physics, one dimension apart. The readout under each is its order parameter R, the length of the population's average arrow: near 0 scattered, near 1 locked.

The circle (S¹, Kuramoto) and the sphere (S², Lohe) under one coupling knob. Low K: both populations wander by their private drifts. Slide up: both lock, sharply. The generalization is not a new theory — it is the same sentence spoken in a bigger room. Everything from here up to a transformer's hypersphere is this picture with more dimensions you cannot see and do not need. R₁ = —   R₂ = —

3. The sphere, live: three wirings

Forty oscillators on the sphere, each with its own private drift (its ω, the sphere version of a natural frequency). Slide K and switch the wiring:

Lohe oscillators on S². The readout tracks the order parameter R = |mean vector| (1 = one tight cluster, near 0 = spread out or balanced clusters) and a cluster count. Try: uniform at K ≈ 2 (collapse to one), two communities (two separate locked clusters — R can be low while order is high, which is why R alone is a blunt instrument), agreement-chosen (several clusters condense from geometry alone and persist). Sound on: a soft chord whose loudness is R — a labeled mapping, agreement made audible. R = —

4. What the wiring can be

Notice what the third regime above did: nobody drew that graph. The wiring was decided, instant by instant, by the states themselves — who agrees with whom right now chooses who talks to whom right now. That idea has a name, effective coupling, and it deserves more than a button: it is the subject of the next section, where you can watch the coupling matrix itself as a live object, and feel the difference between a network whose wiring lives and one whose wiring is frozen. After that, section 3 cashes the whole module out: a token moving through a transformer as a path on this very sphere.

What's real here The simulations in this section are illustrative Lohe dynamics on the ordinary 3-dimensional sphere — chosen because you can see it. Real models live on spheres of hundreds to thousands of dimensions, where two random directions are almost always nearly perpendicular (concentration of measure) — which is precisely what makes the high-dimensional versions of these dynamics tractable at all. The claims connecting any of this to transformers are argued and measured in the essays, and this module's last section walks you to them.