Renewal does not imply disappearance
In a generative account, persistence can mean that a relation is maintained across successive completions. It need not mean that a fixed piece of substance sits outside the process of renewal. The books use this perspective to connect renewal, memory, and observed continuity. [B04] [B17]
Our explicit model allows a precise but limited answer. Its total weight persists exactly. Its initial localized pattern does not. The local averaging rule spreads that pattern across the graph, approaching a uniform state. This separates a conservation law from the stability of a particular structure.
The distinction matters when discussing matter. A localized particle-like state would require a mechanism that maintains localization or supplies a meaningful moving pattern, along with observable properties such as dispersion and calibrated responses. The mere survival of a total quantity is insufficient to establish those properties.
Track conservation and concentration
Advance the central seed and compare two curves: total weight Σρᵢ and concentration C = Σρᵢ². Total weight stays at one; concentration falls from one toward 1/27. Change μ to vary the rate of spreading, then reset for a controlled comparison.
Explore · teaching model v1
Teal: generated values. Gold: highlighted readout or reference.
Enable JavaScript to explore this model. The figure, worked values, and equations explain the same construction.
At μ = 0 the update is the identity and every pattern remains unchanged. This is persistence by switching off dynamics. For positive μ in the allowed range, the connected graph mixes the central seed. A frozen pattern and a dynamically stable object are different ideas.
The stationary state and its modes
The uniform vector is stationary because every neighbour difference is zero. The lower bound on concentration follows from Cauchy–Schwarz: (Σρᵢ)² ≤ 27Σρᵢ². Equality requires all weights equal. Our initial central seed has concentration one.
The graph Laplacian is real symmetric, so its eigenmodes provide a basis. Each mode is multiplied by 1−μλ per update. For this three-by-three-by-three open graph, the one-dimensional path has eigenvalues 0, 1, and 3; the three-dimensional eigenvalues are sums of three such values. The largest is 9. For 0 < μ ≤ 1/6, every nonconstant mode has magnitude multiplier less than one, while the constant mode remains.
The negative multipliers possible near the upper end of the μ range describe sign alternation of a mode coefficient around equilibrium. They do not make the complete nonnegative state negative, because the full update is still a nonnegative averaging matrix. Mode coefficients and state weights are different quantities.
What a model of matter would add
A more ambitious reconstruction would need suitable modes, stable or sufficiently long-lived structures, interactions, and a mapping from mathematical parameters to measurable quantities. Phase, spin, dispersion, and mass cannot be inferred from the colour or size of a dot in this laboratory. The phase and quaternionic papers address additional mathematical structures under stated assumptions. [P01] [P03]
The existing quantum guide, matter section, and applications provide onward routes. This chapter supplies a useful negative result: the chosen diffusion rule conserves weight but does not support a stable localized particle. Finding that limitation is part of understanding the model.
Continue to predictions to turn the same rule into a calculation that can be checked independently of the illustration.

