Up to this point, the lattice has been sorting itself out locally. Bonds choose alive or dead, PLMs assemble, the K-field freezes. Now something strange happens. If you run the CLR from random initial conditions on a plain, featureless grid, topological defects appear. Vortices — points around which the phase winds by a full turn — nucleate spontaneously out of the noise. Nobody placed them. Nothing about the lattice geometry demands them. They form because the dynamics had no other way to reconcile the phases.
This is the hinge of the whole derivation. Everything before this chapter is about how couplings organize themselves. Everything after is about what the presence of vortices forces the CLR to do. The fine structure constant lives in that forcing.
Let's start where topology starts: with a loop.
There are two circles involved, and keeping them straight is the whole game. The first is the loop you walk — a path through space, closing back on itself. The second is the phase circle — the space of possible angles any one oscillator could be pointing in. Every point along the loop has an oscillator sitting on it, and that oscillator's phase is a point on the phase circle. So walking the loop traces a second walk, over on the phase circle.
The winding number answers a simple question: as you walk once around the loop, how many times does the phase circle complete its own lap? If it wobbles and returns to where it started, the answer is zero. If it ticks off one full revolution of the phase circle by the time you finish the loop, the answer is +1. Two revolutions, +2. One backwards, −1. The number is always an integer, because coming back to your starting point on the loop means coming back to the same spot on the phase circle — and you can only get there by completing a whole number of laps.
Click any symbol to see what it means.
Easier to see than to read. Below, the loop on the left has twelve oscillators spaced around it. Each one's colour is its phase — the hue directly encodes its angle on the phase circle, so red means 0, yellow is a quarter turn ahead, green is half a turn, and so on, coming back to red after a full lap. On the right sits the phase circle itself. A slow walker travels around the loop; a pointer on the phase circle shows what phase the walker is seeing right now. Watch how many times the pointer laps its circle while the walker does one lap of its own.
A note on time A fair question at this point: these are oscillators, so shouldn't their phases be turning? Yes. What you've been seeing is the system in its co-rotating frame — a view where we subtract the shared rotation rate and show only the relative phases. Hit Let them oscillate on the figure above: the oscillators on the left now tick forward in the lab frame, and you'll see every cell's colour cycle continuously through the spectrum. That's the oscillation — oscillators oscillating. The phase circle on the right, though, keeps showing the topology: the pointer tracks the spatial winding only, unaffected by the common rotation. After every complete lap of the walker, the pointer has completed exactly n full turns, even while all twelve cells have been cycling around the colour wheel many times over. That is the precise sense in which the winding is a topological invariant: the oscillators can do whatever they like in time, and the integer n sits there unmoved.
Winding is quantized. You cannot smoothly dial it from 0 to 1, or anywhere in between. It is protected: any continuous deformation of the phases leaves it alone. Noise cannot rub it out. Only a discontinuous event — some bond's phase difference crossing past half a turn — can flip it. That protection is what gives vortices their astonishing robustness. Once a vortex has formed somewhere on a lattice, you cannot get rid of it by gentle jostling; you have to shove phases past the cut, which costs real energy.
Now promote the loop to a 2D grid. A vortex is a configuration of phases on the plane where every loop surrounding a particular point has non-zero winding. The point is the vortex core. The phase there is undefined — there is no consistent angle you can assign the central site without breaking continuity with its neighbours. Around the core, the phases arrange themselves into a pinwheel: walk any path encircling the core and the phase turns through a full revolution by the time you return.
Below is a 16×16 grid with a +1 vortex seeded at its centre and then relaxed so the phases smooth out into a clean pinwheel. Each cell's colour is its phase, same mapping as before. The small burgundy mark in the middle is the vortex core — the one spot on the grid where no consistent phase can be assigned. On top of the grid sits a moveable rectangular loop (the burgundy outline). Drag it anywhere. The number above it is the winding computed by summing phase differences around its perimeter.
A note on geometry The real system we're building toward is the 3D diamond lattice, whose 2D slices are honeycomb. We're using a square grid here only because winding is easier to read off a rectangle and square lattices are what most readers have intuition for. The topology of what a vortex is, and of how winding is counted around a loop, is identical on any 2D lattice. The argument for why diamond and not any of the other possibilities is the whole story of Chapter 8.
The buttons offer two other configurations worth walking through. A −1 antivortex is a vortex of opposite sign: walk a loop around it and the phase winds the other way, through one full turn backwards. The rainbow still pinwheels around the core, only now it cycles the opposite direction. Same topological object, opposite charge.
The vortex / antivortex pair is where the physics starts showing its hand. On a finite region with uniform boundary conditions, the total winding of the whole region is fixed at zero. Sum the winding of every plaquette on the grid and interior contributions cancel in pairs; what survives is just the winding around the outer boundary, which we've set to zero. So any vortex that shows up must come with an antivortex as a partner. You cannot make one without the other. They are born together, and when they eventually meet they annihilate together — the +1 and the −1 collide, their windings cancel, and the stored topological energy is released as a burst of phase waves.
Click the pair preset and try it with the loop. Enclose just the + core — winding reads +1. Move the loop over to enclose just the − core — −1. Then grow the loop until it encloses both cores at once — it reads 0, exactly. From outside the pair, the two defects cancel each other. There is no topology there at all.
This pairing law is going to matter in a moment. When vortices nucleate out of random noise, they do not appear one at a time. They emerge as balanced pairs — always one +1 with one −1, always conserving the total. The question the CLR is about to face is not can a vortex exist, but once a pair has nucleated, will the Shannon channel keep the pair alive, or will it destroy it.
We are about to do something that feels almost magical the first time you see it. Start the grid with completely random phases and a small, random initial coupling on every bond — no structure at all. Run the Kuramoto equation for the phases and the Shannon part of the CLR for the couplings, in lockstep. Without any guidance or seeding, vortices form. They are not put there. They emerge because the phases, trying to align locally with their neighbours, find themselves in a pattern whose contradictions can only be resolved by winding.
And then, a few thousand steps later, something else happens. The same CLR that let the vortices form now starts to kill them. You can watch it happen.
Before the figure, the picture to hold in your head. A vortex is a ring of phases winding by 2π around a central point. Near that central point, the phases cannot agree: neighbouring oscillators have large phase differences, and the Shannon channel looks at those mismatched bonds and rules them unfit. It kills them. So Shannon, acting alone, carves out a dead hole at each vortex core — a small region where the graph has no live bonds at all. And that hole wants to spread: its edges are still where the phase field is stressed. Left unchecked, the damage radiates outward and eats away more and more of the lattice.
The Fiedler channel's job is to stop the spread. It looks at the graph's algebraic connectivity — the Laplacian's second eigenvector v2 — which lights up wherever the graph is about to sever. Then it pushes coupling back into the bonds at that boundary, keeping them barely alive. The result is a protective ring of weak bonds around the would-be hole: enough to keep the graph connected, but not enough to flatten the vortex's phase winding (which is what we want to preserve for the BKT physics).
We run it twice, side by side, same initial conditions. The left grid uses the Shannon channel alone. The right grid uses the full CLR — Shannon plus Fiedler.
The Fiedler vector v2 is the smooth function on the graph that minimises the weighted gradient subject to orthogonality with the constant mode — the direction along which the graph is easiest to cut. On a uniform lattice it is a smooth wave; near a would-be bottleneck it concentrates dramatically on the cut. The CLR reads that eigenvector and uses the squared gradient (vi − vj)2 as a per-bond structural score: bonds at bottlenecks score high, bulk bonds score near-zero, and after mean-subtraction the total push is exactly zero. Everything it adds somewhere, it subtracts somewhere else.
Two things have to happen before the α derivation can fire, and we have just seen that both are genuine requirements, not assumptions.
First, the lattice must be at least three-dimensional. Our 2D demo produced the right vortex dynamics for the first thirty thousand steps — Fiedler held the cores alive, Shannon tried to kill them, the contest resolved exactly as the paper claims. And then the vortex pair drifted into each other and annihilated. Point defects in 2D always eventually find their partners; the topological constraint dissolves in finite time. In 3D, vortices are lines (the topology is π1(S¹) = ℤ on circles transverse to a line rather than on spheres around a point). Lines cannot collide pointwise; they have to collapse along their whole length, which costs energy scaling with length. That is why the paper needs a 3D lattice — and we'll see in Chapter 8 that the five independent filters it must satisfy select diamond uniquely.
Second, the CLR needs a ceiling. The CLR wants K as high as possible. The vortex we saved cannot survive above a specific critical coupling called KBKT = 2/π: above it, the vortex unbinds from its partner and the topology is gone; below it, stable. So the CLR climbs and the vortex caps, and the equilibrium sits exactly at the wall. That meeting point is where α lives. Chapter 9 derives the value.