8. The Proper Optics: Why the Logarithmic Scale Is Structural
Geometry is not true; it is convenient. — Henri Poincaré
From phenomenon to necessity
In the previous chapter, an experimental fact was established with clarity:
the same deterministic operator $M$, constructed over the same arithmetic region, exhibits radically different spectral statistics when observed under different scales.
Linear sampling revealed a statistical regime compatible with local decorrelation, whereas logarithmic sampling led systematically to the emergence of universal correlations.
This contrast was not presented as a paradox, but as an optical effect: two lenses, two possible readings of the same object.
In this chapter, the metaphor is set aside and the inevitable question is addressed:
why is the logarithmic lens not merely effective, but structurally necessary?
Scale as part of the object
Up to this point, scale was treated as an external parameter — a choice made by the observer. That perspective is now insufficient.
In the context of prime arithmetic, scale is not neutral. It is embedded in the very structure of the object under observation.
Since Gauss, it has been known that the average density of primes around $x$ decreases as
\[\frac{1}{\ln(x)}.\]This means that the number line is not homogeneous from the standpoint of prime structure. Regions farther from unity are not merely “larger”: they are structurally more sparse.
Consequently, a linear ruler does not measure this space adequately. It compresses dense regions and dilutes sparse ones, distorting any attempt at a global reading.
The logarithmic scale, by contrast, acts as a natural change of coordinates: it reparametrises the axis in such a way that the average variation in prime density becomes approximately uniform.
The role of $\Delta_\pi(x)$ under reparametrisation
The operator $M$ does not depend directly on $\pi(x)$, but on its oscillatory combination:
\[\Delta_\pi(x) = \pi(x) - 2\, \pi(x/2).\]This function captures precisely the local deviation from the expected average behaviour.
The crucial point is that:
- under a restricted linear parametrisation, $\Delta_\pi(x)$ varies slowly;
- under a broad logarithmic parametrisation, $\Delta_\pi(x)$ traverses regimes with fluctuations across multiple scales.
Thus, the difference observed in the spectrum of $M$ does not originate in the operator, but in the way the operator’s argument explores the arithmetic structure.
The linear lens reduces the effective variability of the signal. The logarithmic lens restores its diversity of phases.
Effective complexity and phase mixing
The matrix
\[M_{ij} = \cos\, (\Delta_\pi(x_i)\, \ln(x_j)) + \cos\, (\Delta_\pi(x_j)\, \ln(x_i))\]is, by construction, symmetric and deterministic.
However, its degree of effective complexity depends on the diversity of phases present in the terms $\Delta_\pi(x_i)\, \ln(x_j)$.
- In narrow linear windows, these phases are highly correlated.
- In broad logarithmic windows, a mixing of phases occurs, comparable to that observed in deterministic chaotic systems.
What the spectrum “recognises” is not randomness, but richness of interference.
Universality does not emerge because the system is random, but because it is sufficiently complex when observed at the appropriate scale.
The optical experiment
Notebook 08 (08_the_discovery_lens.ipynb) 
introduces neither new operators nor new statistics.
It performs only one conceptual operation:
the operator is held fixed, while the manner in which the number line is traversed is varied systematically.
In doing so, it demonstrates that:
- Poisson statistics are not a defect of the operator;
- GOE statistics are not an artefact of normalisation;
- both are legitimate readings, corresponding to distinct geometric regimes.
Scale acts as a regime selector.
What this chapter does not claim
It is important to state explicitly the limits of what has been established:
- it is not claimed that the logarithmic scale “creates” universality;
- it is not claimed that every arithmetic function would produce the same effect;
- it is not claimed that GOE statistics are the only possible outcome.
What is claimed is more restricted and more robust:
when arithmetic structure is observed at its natural scale, the complexity required for the emergence of universality is present.
Point of Rest
By the end of this chapter, scale ceases to be a technical parameter and becomes part of the structure of the problem.
The logarithmic lens is identified as geometrically natural for the observation of the arithmetic signal.
The contrast between Poisson and GOE statistics is no longer interpreted as a change of object, but recognised as a change of regime.
In all cases, the operator $M$ remains deterministic and unaltered.
In the next chapter, this understanding will be taken further. We will investigate which features of the operator are essential to the observed universality and which can be modified without destroying it, separating structure from contingency.
| $\gets$ Previous Chapter | Contents | Next Chapter $\to$ |