How a binary field emergesWhat the Coherence Theorem looks like when you let go of a whole network

Chapter 4 gave us a rule for one bond and an inequality for one network: I ≥ 0. Now we want to see it. We will run the CLR on a grid, from random initial conditions, and watch the theorem play out. Every bond is deciding, locally. The whole is climbing, globally. Nothing is directed from above.

Two bonds sharing a node

Before a whole grid, pause at the simplest case where bonds can influence each other: three oscillators in a chain, two bonds. Call them A–B and B–C. Bond AB sees the alignment of A and B; bond BC sees the alignment of B and C. Both run the CLR. Neither one can see the other directly. But they share node B — so whatever phase B settles into, both bonds feel.

Three oscillators, two living bonds. Each bond's K is evolving under the CLR, seeing only its own two endpoints. But the shared middle oscillator couples them implicitly. Set ωB near ωA — bond AB thrives while BC may struggle. Set all three nearly equal — both bonds come alive. Spread the ω's apart — B gets pulled in two directions, both bonds weaken. K_AB = 0.00, K_BC = 0.00

The two bonds make independent local decisions but their outcomes are entangled. What B does reflects the tug-of-war between A and C. This is how a whole lattice's dynamics composes out of atomic bond-level rules.

A small ring

Scale up. Six oscillators in a ring, six bonds, every one living. Each bond sees only its two endpoints. But every oscillator sits between two bonds, so every decision ripples.

Six oscillators on a ring. Each bond's K is a small live number; the whole ring is running the CLR. Watch what happens as K's evolve: some bonds strengthen to K*, some die to zero, and the ring settles into a specific pattern of live bonds. Reset repeatedly to see the basin structure — same dynamics, different final topologies depending on initial conditions. alive = 0, dead = 0

Notice: the ring settles on a pattern. Not the same pattern every time — resets give different outcomes. But every outcome is binary. No bond lingers in the middle. Each one chooses alive or dead, and the choices compose into a coherent-enough structure for the ring's frequencies to lock.

The aggregate form of I ≥ 0

Chapter 4 proved the Coherence Theorem globally: I(t) = dC/dt ≥ 0. But there is a more concrete way to see why it must hold, which becomes important now that many bonds are evolving at once. The global flux decomposes exactly into per-bond fluxes:

I(t)  =  ∑(i,j) ∈ E   Iij(t)  ,  with   Iij(t)  ≥  0 for each bond

Click any symbol to see what it means.

In words The global intelligence flux I(t) is the sum of every single bond's own contribution. Each bond's contribution Iij(t) is non-negative by the construction of the CLR — every bond is performing gradient ascent on its piece of coherence capital. When you sum non-negative numbers, the sum is non-negative. That's the theorem at the level of locality: no bond cheats, so the whole can't lose. This is why a global ascent is enforced by purely local rules. Each bond, acting on information only about its own endpoints, pushes the same global quantity upward.

Two attractors, everywhere

Recall the death threshold from Chapter 4: a bond survives only if cos(Δθ) > 4/r. Below that threshold the potential has no interior well; the only stable K is zero. Above it, there's a single well at K = K*. Every bond is solving the same simple equation at equilibrium:

R0(K*)  ·  cos(Δθ)  =  2K* / r

Click any symbol to see what it means.

In words At equilibrium, each bond balances two forces. On the left: the pull toward alignment — the Bessel ratio R0(K*) scaled by the observed alignment cos(Δθ). On the right: the regularization cost of maintaining a strong coupling, 2K*/r. When alignment is high enough, the equation has a positive solution K* > 0 (alive). When alignment is too low, the only solution is K* = 0 (dead). There is no continuum of intermediate states. Every single bond in a network, after the CLR has finished running, sits at one of these two attractors.

When a whole network of bonds runs the CLR, each bond finds one of these two attractors for itself. The result, aggregated over thousands of bonds, is a binary field: a distribution of K values with two sharp peaks and almost nothing in between.

The full simulation

Here is a proper grid — 14 by 14 oscillators, each connected to its four neighbors, every bond living. Start from random initial K's and random phases. Let go. Watch the field polarize.

A note on the controlsω spread sets how much the oscillators' natural frequencies disagree — the wider the spread, the harder it is for bonds to lock. r is the regularization parameter inside the CLR, the signal-to-noise ratio that sets the death threshold cos(Δθ) > 4/r. In the full theory on the diamond lattice with vortex topology, r is not a free parameter — it is determined self-consistently (≈ 5.9) by the requirement that bulk couplings equal 16/π². In this standalone grid we expose it as a slider so you can feel how it shapes the bifurcation. Phases / bonds / both toggles what you see: phase-colored cells, the K-field as line thickness, or both at once.

Top: the grid — in phases mode, cells colored by each oscillator's phase; in bonds mode, cells dim and every bond drawn as a line whose opacity is proportional to K (dead bonds fade, alive bonds glow); in both mode, combined. Bottom: the live K histogram (watch it crystallize bimodal) and the coherence-capital / intelligence-flux traces. Reset repeatedly to see the basin structure. I_phase = 0.00, C = 0.000

The field is the memory

What you are looking at — the particular pattern of alive bonds at the end of a run — is the network's learned connectivity. It was not imposed. It was produced by the dynamics, from initial conditions that did not know what pattern was going to emerge. The CLR found a structure consistent with the current natural frequencies and locked it in. If you hit reset and run again, a different pattern forms. If you wait long enough, each pattern stays stable — the binary nature of the attractor provides its own inertia.

A network that has found a stable K-field has, in a very literal sense, formed a memory of the frequencies and the topology it was given. Its pattern of alive bonds holds that memory even when the oscillators' motion is interrupted.

This is already remarkable. But a memory is only as useful as what you can do with it. In the next chapter we ask: what if we drive the network with a specific pattern? What if we shape its natural frequencies externally, or apply a stimulus? Does the CLR learn a specific pattern and retain it when the stimulus is removed? Does it recognize the pattern when shown it again? The answer, as you might by now suspect, is yes — and the analogy to both Chladni plates and neural memory gets very tight, very fast.