Why three dimensionsThe hidden dial in the α formula — and the dimension that survives it

Every factor of the α formula has by now been derived from lattice physics. But one of those factors, quietly, carries a free parameter we have never turned: the dimensionality of space itself. The coordination number z in the formula is really d + 1, where d is the dimension of the lattice. Diamond gives z = 4 because d = 3. What if we had lived on a lattice of another dimension?

The d-dial

Slide the dimensional dimension of the lattice in the figure below. Only one integer value makes a physical electromagnetic coupling:

Sliding the dimension dial sweeps d from 2 to 5 continuously. The dot on the ruler shows the current 1/α; the green dashed line marks the physical value 137.036. Only d = 3 reaches it. At d = 2 the coupling is far too strong (1/α < 40); atoms would collapse under electromagnetic self-attraction. At d = 4 the coupling has halved in strength (1/α > 500); atoms barely bind. At d = 5 it is weaker by another factor of four. Dimensionality is not a free choice — it is selected by the requirement that the output be physical.

Notice how sensitive the output is. A one-unit change in d moves 1/α by roughly a factor of four. This is not a smooth slide: the whole formula has an exponential (R0 raised to z = d + 1), so integer dimensions give wildly different couplings. It is a knife-edge landscape with one viable point at d = 3. Click through the integer values:

What the dial does to matter

The 1/α ruler above is an abstract number. Here is that same dial's effect on real physics: three hydrogen-like atoms, each a fixed proton (red) with an orbiting electron (blue). The only knob is the electromagnetic coupling strength — exactly what α controls. At d = 3 the electrons hold stable, circular orbits. Slide to d = 2 and the attraction is nearly 4x too strong: the electrons spiral inward and slam into their protons. Slide to d = 4 and the attraction is 4x too weak: the electrons fly off in straight lines, the atoms evaporate. Only at d = 3 does matter exist in the form we know.

Three protons (fixed red disks) with electrons (blue dots with fading trails) orbiting them under Coulomb attraction proportional to α(d). Shared slider with the d-dial above — move either and both update. At d = 3, each electron holds a circular orbit indefinitely. At d < 3, the force exceeds what the initial tangential velocity can balance and electrons spiral in — eventually colliding with their protons (annihilation flash). At d > 3, the force is too weak to hold the orbital velocity; electrons fly off tangentially and escape. The simulation uses velocity-Verlet integration with soft-core protons (r ≥ 1.5 nominal units) to prevent numerical singularities, so a "crashed" electron rebounds chaotically inside the soft core rather than diverging. Press Reset to re-spawn electrons at the d = 3 Bohr radius with circular velocity.

This is what α — the whole formula we spent twelve chapters deriving — does to a universe. It sets the size of atoms (Bohr radius ∝ 1/α), the depth of their binding energy (∝ α²), and whether they exist at all. Change α by a factor of four, change d from 3, and hydrogen stops being hydrogen.

Why d = 2 and d = 4 fail

d = 2 — too strong to bind atoms
With a coupling of 1/α ≈ 35 (almost 4x the measured value), electrons would pull each other so hard that hydrogen's electron would spiral inward in a picosecond. Bohr radii would be 4x smaller, electron binding energies sixteen times larger, and chemistry \u2014 which lives in the narrow range where EM is strong enough to bind but weak enough to allow delicate valence structures \u2014 would not exist. Perturbation theory in α would also fail: the very expansion parameter that makes QED a well-defined theory would become order unity.
d = 4 — too weak to matter
At 1/α > 500, EM would be 4x weaker than in our universe. Atomic binding energies drop by 16x; thermal motion at room temperature would easily ionise hydrogen. Worse: 2D point-vortex annihilation (Chapter 7) goes away only because d ≥ 3; in d = 4 the vortex is a 2-dimensional object that suffers even stronger topological fluctuations. The BKT wall arguments that pin K at 2/π weaken. Whether electrons can even exist as stable excitations becomes an open question.

At d = 3 the combination clicks: EM is strong enough to bind atoms (α of order 0.01 keeps nuclei and electrons together), weak enough to treat perturbatively (allowing QED to work), and the lattice's 3D geometry is the unique d where 2D point-vortex annihilation (Chapter 7) can fail but 3D vortex-line topology persists (Chapter 8). Three constraints, one satisfying dimension.

Is this anthropic — or forced?

A careful claimWhether the observed dimensionality is anthropic (we live here because other dimensions are uninhabitable) or forced by the structure of the theory (no other dimension admits a self-consistent CLR attractor) is a question the coherence lattice does not yet answer. What it does show: the two options are not equivalent. In an anthropic framing, the dimension is a boundary condition imposed on a multiverse; in a forced framing, the dimension is a requirement of the equations. The d-dial figure shows what the formula says, nothing more.

A softer version of the forced argument can be made. The coherence lattice requires three features to work: (i) the CLR attractor (any d), (ii) a topologically stable vortex (needs d ≥ 3 because 2D point-vortices annihilate pairwise under the CLR, Chapter 7), and (iii) a physically viable α (narrowly around d = 3). The intersection of these three constraints is the single integer d = 3. Whether that counts as "forced" is a philosophical question; operationally, the theory picks it out uniquely.

Where this leaves us

The BKT formula gives 1/α = 137.032. The CODATA measurement is 137.035999. The gap is 29 ppm, or 3% of 1%, and it is not zero. That remaining difference is not a lattice effect; it is QED vacuum polarisation running αBKT (which lives at the matching scale) down to Q = 0 (the Thomson limit, where CODATA measurements sit). The next chapter performs that running — with a linked-cluster expansion over diamond subgraphs — and closes the gap to 1.5 parts per billion.