Structure vs Exactness in Loop Coefficients

By John Gavel

In working through higher-order loop corrections in the framework, I encountered a situation that is more interesting than the result itself. This issue between structure and operation keeps pulling be back and forth.

Here we have two distinct candidate expressions emerged for the four-loop coefficient $\delta_4$. Both match numerical constraints to within experimental precision (~$10^{-9}$), and both reduce residual error down to ~$10^{-13}$ scale. At that level, empirical distinction is no longer meaningful.

What remains is not a question of correctness, but of structure.

Candidate 1: Combinatorial Closure

The first candidate is:

$$ \delta_4 = -\frac{F + K - 1}{2K} = -\frac{31}{24} $$

This expression is built directly from icosahedral structural quantities:

• $F = 20$ (faces)
• $K = 12$ (coordination number)
• $K - 1 = 11$ (self-exclusion constraint)
• $2K = 24$ (bidirectional coordination shell)

Its strength is precision and simplicity. It collapses multiple structural counts into a single rational ratio with extremely low deviation from the observed coefficient.

However, its limitation is interpretability across loop order. It does not naturally extend from $\delta_2$ or $\delta_3$ in a generative way. It is structurally valid, but locally assembled rather than globally enforced.

Candidate 2: Loop-Progression Law

The second candidate is:

$$ \delta_4 = -\left(\frac{4}{\pi_3} + \frac{\eta_{sub}}{\pi}\right) $$

This expression continues a visible progression:

• $\delta_2 = -4K$
• $\delta_3 = -(\pi + 4/\pi_2)$
• $\delta_4 = -(4/\pi_3 + \eta_{sub}/\pi)$

The key feature is the emergence of the $4/\pi_d$ sequence, where $\pi_d$ reflects the closure dimension of each loop level:

• $4/\pi_2 = 4/3$
• $4/\pi_3 = 1$

The term $\eta_{sub}/\pi$ introduces substrate efficiency filtering into holonomy scaling.

Unlike the first candidate, this formulation is explicitly generative. It describes a rule that produces $\delta_n$ rather than a standalone fit for $\delta_4$.

Its weakness is that it is slightly less numerically sharp in isolation.

The Actual Problem

At first glance, this appears to be a precision dispute. It is not. This is a model identity problem under degeneracy:

When multiple expressions converge to the same empirical value, the deciding criterion becomes whether the structure generalizes or merely compresses. The funny thing is I'd prefer both happening at the same time. 

Resolution Criterion

The comparison reduces to two axes:

• Candidate 1: maximizes local compression (exact rational closure from geometry)
• Candidate 2: maximizes generative continuity (law across loop hierarchy)

Neither is falsified. Both sit within numerical indistinguishability.

So the decision shifts from “which is correct?” to:

Which structure survives extension to $\delta_5$ and beyond without reparameterization?

Conclusion

In regimes where empirical resolution saturates, theory selection becomes architectural rather than numerical. Yet this is such a pull on my intuition that I figured I'd share it and see what others think or at least show others my thinking. To me:

One expression explains the value.

The other explains the sequence.

So, could they eventually be the same mechanism?