Coherence–Rupture–Regeneration

Mathematical Foundations for a Temporal Process Theory

A. Sabine · Version 6.0 · March 2026

What is CRR? A Plain-Language Introduction

The Core Idea

You don't experience time as a smooth flow—you experience distinct "nows," each one feeling complete before giving way to the next. William James called these "perchings" and "flights" of consciousness.

CRR proposes that this pattern—build up, break, rebuild—is a universal signature of how bounded systems navigate time. It is grounded in process philosophy (Whitehead, 1929): reality is not made of things that change, but of changes that occasionally cohere into things.

The Three Phases

Coherence: The Accumulated Past

Think of coherence as "accumulating evidence." As you read this sentence, your brain integrates information—words build on words, meaning accumulates. Mathematically, coherence is the integral of a system's activity over time: everything it has done since its last transition, added up.

Rupture: The Decisive Present

When accumulated coherence saturates the system's capacity, transformation occurs. This is rupture—the discrete, instantaneous moment when the system must reorganise. The key insight: rupture happens when the product of accumulated evidence ($C$) and the system's characteristic variance ($\Omega$) reaches exactly 1. This is the Cramér-Rao bound from statistics, the Heisenberg uncertainty principle from physics, and the Gabor limit from signal processing—all the same equation.

Regeneration: The Reconstructed Future

After rupture, the system rebuilds using its history—but not all history equally. Moments of high coherence contribute more; moments of low coherence fade away. This is why significant experiences shape you more than forgettable ones, regardless of when they occurred. Memory is weighted by significance, not recency.

The Key Parameter: $\Omega$

Variance as the Single Parameter

The framework has one central parameter: $\Omega = \sigma^2$ — the system's characteristic variance (equivalently, the inverse of precision: $\Omega = 1/\pi$). Think of it as a dial between "rigid" and "flexible":

  • Small $\Omega$ (rigid, high precision): High rupture threshold ($C^* = 1/\Omega$ is large). Rare but significant transformations. Sharply peaked memory weighting—only the most coherent moments are accessible. Think: habit, crystallised skill, trauma loops.
  • Large $\Omega$ (flexible, low precision): Low rupture threshold ($C^* = 1/\Omega$ is small). Frequent micro-ruptures. Broad memory access—the whole of history is available. Think: insight, healing, phase transition.

The Universal Predictions (No Free Parameters)

Parameter-Free CV Predictions

From the axioms (see Axioms tab), CRR derives that the coefficient of variation (CV) of inter-event intervals is:

$$\text{CV} = \frac{\Omega}{2}$$

For the two fundamental symmetry classes on $S^1$:

  • $\mathbb{Z}_2$ systems (bistable — two states, like a switch): $\Omega = 1/\pi$, so $\text{CV} = 1/(2\pi) \approx 0.159$
  • $SO(2)$ systems (rotational — continuous cycle, like a wheel): $\Omega = 1/2\pi$, so $\text{CV} = 1/(4\pi) \approx 0.080$

The ratio between them is exactly 2. These predictions have been tested across 100+ systems in 30+ domains. Systems that deviate tell you something specific: CV below prediction indicates active regulation; CV above prediction indicates asymmetric bistability.

Why Does This Matter?

If CRR is correct, it provides: a universal language for temporal dynamics across scales, parameter-free testable predictions, a bridge between information geometry and process philosophy, and a formal temporal completion of the Free Energy Principle. If CRR is wrong, the CV predictions will fail—and the deviations will be informative about what the true temporal grammar must look like.

First Principles

CRR rests on a minimal set of axioms drawn from information geometry, thermodynamics, and process philosophy. Each axiom connects to established results in physics and mathematics. Together they yield parameter-free predictions testable across every domain where systems persist through change.

Axiom I: Coherence

All systems that persist accumulate evidence through time
$$C(x,t) = \int_0^t \mathcal{L}(x,\tau) \, d\tau$$

Any bounded system that maintains itself against dissipation does so by accumulating coherence—temporal evidence about its environment. In the language of the Free Energy Principle, this is the progressive reduction of variational free energy: as VFE decreases, $C$ increases. The system's generative model becomes a better fit to its environment with each passing moment.

Fisher Information and the Cramér-Rao Bound

The coherence integral $C$ is formally identified with accumulated Fisher information $I(\theta)$ about the system's generative model parameters $\theta$. Fisher information measures the curvature of the log-likelihood: how sharply the data distinguish between nearby hypotheses. It is the unique Riemannian metric on statistical manifolds (Čencov's theorem), meaning any theory of inference that respects sufficient statistics must use it.

The Cramér-Rao inequality then states a fundamental limit:

$$\text{Var}(\hat{\theta}) \geq \frac{1}{I(\theta)} \qquad\Longleftrightarrow\qquad \sigma^2 \cdot I(\theta) \geq 1 \qquad\Longleftrightarrow\qquad C \cdot \Omega \geq 1$$

No unbiased estimator can have variance smaller than the inverse of the accumulated Fisher information. This is not a modelling assumption—it is a theorem of mathematical statistics, proven independently by Cramér (1946) and Rao (1945). Ito & Dechant (2020) extended this to stochastic thermodynamics, showing that the Cramér-Rao bound governs the trade-off between current fluctuations and entropy production in irreversible processes far from equilibrium.

CRR's contribution: the bound is not merely approached but saturated. At the moment of rupture, $C \cdot \Omega = 1$ exactly. The system has extracted the maximum information its current configuration permits.

Cramér, H. (1946). Mathematical Methods of Statistics. Princeton UP.
Rao, C.R. (1945). Information and the accuracy attainable in the estimation of statistical parameters. Bull. Calcutta Math. Soc. 37, 81–91.
Čencov, N.N. (1982). Statistical Decision Rules and Optimal Inference. AMS.
Ito, S. & Dechant, A. (2020). Stochastic time evolution, information geometry, and the Cramér-Rao bound. Phys. Rev. X, 10, 021056.
Fisher, R.A. (1925). Theory of statistical estimation. Proc. Cambridge Phil. Soc. 22, 700–725.

Axiom II: Rupture

Coherence cannot accumulate indefinitely: a temporal boundary is required
$$\delta(t - t_0) \quad\text{at the moment of transformation, when } C \cdot \Omega = 1$$

No system can build coherence without limit. The Cramér-Rao bound demands a boundary where accumulated evidence meets system variance. CRR identifies this boundary with the Dirac delta—an instantaneous, scale-invariant moment of transformation.

The Temporal Markov Blanket

In the FEP, a Markov blanket is a spatial boundary that renders internal states conditionally independent of external states. CRR proposes that the Dirac delta $\delta(\text{now})$ serves as the temporal analogue: the boundary between past and future, between coherence and regeneration.

The delta has three properties that make it the unique candidate for a temporal boundary:

The Dirac delta distributes its unit mass across the boundary between inside (all past states—coherence accumulated within the blanket) and outside (all future states—regeneration beyond the blanket). The present moment is where inside becomes outside; where evidence becomes action; where the accumulated past becomes the reconstructed future.

Friston, K. (2013). Life as we know it. J. R. Soc. Interface, 10, 20130475.
Pearl, J. (1988). Probabilistic Reasoning in Intelligent Systems. Morgan Kaufmann.
Parr, T., Da Costa, L. & Friston, K. (2020). Markov blankets, information geometry and stochastic thermodynamics. Phil. Trans. R. Soc. A, 378, 20190159.
Schwartz, L. (1950). Théorie des distributions. Hermann.

Axiom III: Regeneration

Systems persist through transformation, not despite it
$$R[\varphi](x,t) = \int_0^t \varphi(x,\tau) \cdot \exp\!\bigl(C(x,\tau)/\Omega\bigr) \cdot \Theta(t - \tau) \, d\tau$$

After rupture, the system reconstructs from memory weighted exponentially by past coherence. $\Omega$ governs both the threshold for transformation and the depth of memory access—it is simultaneously the system's variance (in the FEP sense), its free energy limit, and its thermodynamic boundary.

$\Omega$ as Thermodynamic Threshold

$\Omega = \sigma^2$ is the system's characteristic variance—the inverse of precision ($\pi = 1/\Omega$). In thermodynamic terms, $\Omega$ sets the free energy scale: the amount of surprise (in nats) that the system can tolerate before its generative model must reorganise. This connects CRR to Jaynes' maximum entropy principle: a system with variance $\Omega$ has maximised its entropy subject to the constraint that it maintains coherence up to the threshold $1/\Omega$.

The regeneration weighting $\exp(C/\Omega)$ ensures that moments of high coherence contribute most strongly to reconstruction. This is the Boltzmann factor of statistical mechanics, with $C$ playing the role of energy and $\Omega$ playing the role of temperature. The most "energetic" (coherent) memories dominate the reconstruction, just as the most energetic microstates dominate thermodynamic averages.

Jaynes, E.T. (1957). Information theory and statistical mechanics. Phys. Rev. 106, 620–630.
Friston, K. (2010). The free-energy principle: a unified brain theory? Nat. Rev. Neurosci. 11, 127–138.
Whitehead, A.N. (1929). Process and Reality. Macmillan.

Axiom IV: Unity

At the moment of transformation: $C \cdot \Omega = 1$
$$\text{Accumulated evidence} \times \text{system variance} = \text{unity, at all scales}$$

This is the Cramér-Rao bound at saturation. It is simultaneously the Heisenberg uncertainty principle ($\Delta E \cdot \Delta t \geq \hbar/2$), the Gabor limit ($\Delta f \cdot \Delta t \geq 1/4\pi$), and the thermodynamic uncertainty relation. CRR claims these are not analogies—they are the same equation, expressing the same physical fact: a bounded system that has extracted maximum information from its current configuration must transform.

The Bound Is Universal

The product $C \cdot \Omega = 1$ holds regardless of what the system is, what it is made of, or at what scale it operates. This universality follows from the Cramér-Rao bound being a theorem of information geometry—it depends only on the structure of statistical inference, not on any particular physics. Wherever there is a system accumulating evidence about its environment with finite variance, $C \cdot \Omega = 1$ defines the moment of necessary transformation.

FrameworkEvidenceVarianceBoundCitation
StatisticsFisher information $I(\theta)$$\text{Var}(\hat{\theta}) = \sigma^2$$\sigma^2 \cdot I(\theta) \geq 1$Cramér (1946); Rao (1945)
Quantum mechanicsEnergy $E$Time uncertainty $\Delta t$$\Delta E \cdot \Delta t \geq \hbar/2$Heisenberg (1927)
Signal processingBandwidth $\Delta f$Duration $\Delta t$$\Delta f \cdot \Delta t \geq 1/4\pi$Gabor (1946)
ThermodynamicsCurrent $J$Entropy production $\sigma$$\text{Var}(J) \cdot \sigma \geq 2\langle J \rangle^2$Ito & Dechant (2020)
Information geometryStatistical distance $ds^2$Fisher-Rao metric $g$$ds^2 = g_{ij}\,d\theta^i d\theta^j$Čencov (1982); Amari & Nagaoka (2000)
CRRCoherence $C$Variance $\Omega$$C \cdot \Omega = 1$Saturation of all the above
What CRR adds to Ito & Dechant (2020)

Three things. First, saturation: the bound is not merely a lower limit but is reached at every rupture event. Second, symmetry classification: the geometric value of $\Omega$ is determined by the system's symmetry class ($\mathbb{Z}_2 \to 1/\pi$; $SO(2) \to 1/2\pi$). Third, regeneration dynamics: after the bound is saturated, $\exp(C/\Omega)$ governs how the system reconstructs from weighted memory. Ito & Dechant's thermodynamic uncertainty relation is the inequality; CRR is the equality, plus what happens next.

Heisenberg, W. (1927). Über den anschaulichen Inhalt der quantentheoretischen Kinematik und Mechanik. Z. Physik 43, 172–198.
Gabor, D. (1946). Theory of communication. J. IEE 93, 429–457.
Amari, S. & Nagaoka, H. (2000). Methods of Information Geometry. AMS/Oxford UP.
Wootters, W.K. (1981). Statistical distance and Hilbert space. Phys. Rev. D 23, 357–362.

Theorem: $\text{CV} = \Omega/2$

The Equipartition of Unit Mass
$$\sigma(C^*) = \tfrac{1}{2} \;\;\text{universally} \quad\Longrightarrow\quad \text{CV} = \frac{\sigma}{\mu} = \frac{1/2}{1/\Omega} = \frac{\Omega}{2}$$

The Dirac delta distributes exactly one unit of mass across the rupture boundary. By symmetry between inside (coherence) and outside (regeneration), each side receives exactly one half. This fixes the standard deviation of the rupture threshold at $\sigma(C^*) = 1/2$, independent of $\Omega$.

The Derivation

At rupture, the threshold coherence $C^*$ satisfies $C^* \cdot \Omega = 1$, giving $E[C^*] = 1/\Omega$. The Dirac delta, as a temporal Markov blanket, partitions unit mass between past and future. By the symmetry of the boundary (there is no intrinsic asymmetry between what is accumulated and what is reconstructed), each partition receives $1/2$. Therefore:

$$E[C^*] = 1/\Omega \qquad\text{(from } C \cdot \Omega = 1\text{)}$$ $$\sigma(C^*) = 1/2 \qquad\text{(from equipartition of } \delta\text{'s unit mass)}$$ $$\text{CV} = \frac{\sigma(C^*)}{E[C^*]} = \frac{1/2}{1/\Omega} = \frac{\Omega}{2}$$

For the two fundamental symmetry classes:

$$\mathbb{Z}_2\;\text{(bistable):}\quad \Omega = 1/\pi \quad\Rightarrow\quad \text{CV} = 1/(2\pi) \approx 0.15915$$ $$SO(2)\;\text{(rotational):}\quad \Omega = 1/2\pi \quad\Rightarrow\quad \text{CV} = 1/(4\pi) \approx 0.07958$$ $$\text{Ratio: } \text{CV}_{\mathbb{Z}_2} / \text{CV}_{SO(2)} = \text{exactly } 2$$

These predictions are parameter-free—no fitting, no calibration. They have been tested across 100+ systems in 30+ domains. See the full validation at CRR Benchmarks.

Why $1/2$ and not some other fraction?

Because the Dirac delta has unit mass (this is definitional), because the rupture boundary separates exactly two domains (past and future), and because there is no symmetry-breaking mechanism to favour one side over the other. Any other partition would require an additional parameter—violating the parsimony that makes $C \cdot \Omega = 1$ a first principle rather than a model.

Summary: The Minimal Axiom Set

AxiomStatementFormal Grounding
I. CoherenceAll persisting systems accumulate evidence through timeFisher information; VFE minimisation (Friston, 2010)
II. RuptureA temporal boundary (Dirac delta) is required; it distributes unit mass between past and futureTemporal Markov blanket; distribution theory (Schwartz, 1950)
III. RegenerationSystems persist through transformation, rebuilding from coherence-weighted memoryBoltzmann weighting; MaxEnt (Jaynes, 1957)
IV. Unity$C \cdot \Omega = 1$ at the moment of transformation, at all scalesCramér-Rao saturation; Heisenberg limit; Gabor limit

From four axioms, one central result follows with no free parameters: CV $= \Omega/2$, from the equipartition of the Dirac delta's unit mass across the rupture boundary.

These axioms make CRR falsifiable: any system whose CV deviates from $\Omega/2$ either has a misidentified symmetry class, is actively regulated (CV < prediction), or has asymmetric state durations (CV > prediction). Deviations diagnose; they do not rescue.

1. Standing Assumptions

The axioms (see Axioms tab) establish the ontological commitments. The following technical assumptions ensure mathematical well-posedness.

Assumption (A1) Regularity

For each $x$ and $t$, the function $\tau \mapsto \mathcal{L}(x,\tau)$ is locally integrable on $[0,t]$.

Assumption (A2) Non-negativity

$\mathcal{L}(x,\tau) \geq 0$ for all $x, \tau$. History accumulates; it does not spontaneously dissipate.

Assumption (A3) Bounded Rate (No-Zeno Condition)

There exists $M > 0$ such that $\mathcal{L}(x,t) \leq M$ for all $x, t$. This ensures each coherence cycle has duration at least $1/(M\Omega)$, preventing infinitely many ruptures in finite time.

2. Core Definitions

Definition 1 (Coherence Accumulation Rate)

The coherence accumulation rate is a function $\mathcal{L} : \mathcal{X} \times [0,T] \to \mathbb{R}_{\geq 0}$ that assigns to each state-time pair $(x,\tau)$ a non-negative rate at which the system accumulates evidence about its environment.

Dimensions: $[\mathcal{L}] = [T^{-1}]$ (a rate, so that $\int \mathcal{L}\, d\tau$ is dimensionless).

Identification: $\mathcal{L}$ is formally identified with the rate of Fisher information accumulation—the curvature of the log-likelihood of the system's generative model, accumulated per unit time.

Definition 2 (Coherence)

Let $t_j$ denote the most recent rupture time before $t$ (with $t_0 = 0$). The coherence at state $x$ and time $t$ is the accumulated evidence since the last rupture:

$$C(x,t) := \int_{t_j}^t \mathcal{L}(x,\tau) \, d\tau$$

Properties: Dimensionless, monotone non-decreasing within each cycle, resets to 0 at each rupture.

Definition 3 (Variance Parameter)

The variance parameter $\Omega > 0$ is a positive dimensionless constant characterising the system's boundary permeability. It is identified with:

  • Statistical: $\Omega = \sigma^2$ (the system's characteristic variance)
  • Inferential: $\Omega = 1/\pi$ where $\pi$ is precision (inverse variance)
  • Geometric: $\Omega = 1/\varphi$ where $\varphi$ is the phase (in radians) traversed during one coherence cycle
  • Thermodynamic: $\Omega = k_BT/\kappa_{\text{eff}}$ in physical systems (temperature/effective stiffness)

$\Omega$ determines the rupture threshold $C^* = 1/\Omega$ via Axiom IV ($C \cdot \Omega = 1$).

Definition 4 (Rupture)

A rupture occurs at time $t_*$ when coherence reaches the threshold set by $\Omega$:

$$t_* := \inf\!\left\{t > t_j \;:\; C(x,t) \geq \frac{1}{\Omega}\right\}$$

The rupture event is represented by a Dirac delta $\delta(t - t_*)$. Following rupture: $C(x, t_*^+) = 0$.

Equivalently, rupture occurs when $C(x,t) \cdot \Omega \geq 1$.

Definition 5 (Regeneration)

The regeneration operator reconstructs the system state by weighting historical field values $\varphi(x,\tau)$ exponentially by the coherence at each historical moment:

$$R[\varphi](x,t) = \int_0^t \varphi(x,\tau) \cdot \exp\!\bigl(C(x,\tau)/\Omega\bigr) \cdot \Theta(t - \tau) \, d\tau$$

where $C(x,\tau)$ is the coherence value at moment $\tau$—how far into its cycle the system was at that historical moment—and $\Theta(t-\tau)$ is the Heaviside step function enforcing causality (only the accessible past contributes).

Key Insight: Significance, Not Recency

A high-coherence moment from 1000 cycles ago contributes with greater weight than a low-coherence moment from the most recent cycle. History is weighted by significance (coherence at the time), not by recency. This is consonant with Bergson's insight that memory is "the continuous presence of history," not retrieval from storage.

3. Symmetry Classification

All temporal processes that undergo cyclic C → $\delta$ → R dynamics trace paths on the circle $S^1$. The symmetry class of the process determines the geometric value of $\Omega$:

Theorem (Symmetry Classes)

For a system whose coherence cycle traverses phase $\varphi$ on $S^1$ before rupture:

$$\Omega = \frac{1}{\varphi}$$
  • $\mathbb{Z}_2$ (bistable): Half-cycle ($\varphi = \pi$) $\;\Rightarrow\; \Omega = 1/\pi \approx 0.318$, $\;C^* = \pi$
  • $SO(2)$ (rotational): Full cycle ($\varphi = 2\pi$) $\;\Rightarrow\; \Omega = 1/2\pi \approx 0.159$, $\;C^* = 2\pi$

In both cases: $C^* \cdot \Omega = 1$.

Remark (The $\mathbb{Z}_n$ Hierarchy)

More generally, $\text{CV} = n/(4\pi)$ for $\mathbb{Z}_n$ symmetry classes, where $\mathbb{Z}_4 = SO(2)$ exactly. The symmetry classes are partitions of the circle: $\mathbb{Z}_2$ divides $S^1$ into two arcs of $\pi$; $\mathbb{Z}_4$ into four arcs of $\pi/2$; and $SO(2)$ treats the full $2\pi$ as a single cycle.

4. Key Properties

Theorem (No Zeno Pathology)

Under (A3), each coherence cycle has duration at least $1/(M\Omega)$. Hence the number of ruptures in any finite interval $[0,T]$ is at most $\lfloor TM\Omega \rfloor + 1 < \infty$.

Proposition (Regeneration Weighting Contrast)

The weight ratio between a rupture moment (when $C = C^* = 1/\Omega$) and a moment of zero coherence is:

$$\frac{\exp(C^*/\Omega)}{\exp(0)} = \exp\!\left(\frac{1}{\Omega^2}\right)$$

For $\mathbb{Z}_2$ systems ($\Omega = 1/\pi$): contrast ratio $= e^{\pi^2} \approx 19{,}400$.
For $SO(2)$ systems ($\Omega = 1/2\pi$): contrast ratio $= e^{4\pi^2} \approx 1.4 \times 10^{17}$.

This extreme selectivity means regeneration is overwhelmingly dominated by moments near rupture—peak-coherence moments carry nearly all the weight, regardless of when they occurred.

5. Dimensional Analysis Summary

QuantitySymbolDimensionsIdentification
Accumulation rate$\mathcal{L}$$[T^{-1}]$Fisher information rate
Coherence$C$dimensionlessAccumulated Fisher information
Variance parameter$\Omega$dimensionless$\sigma^2 = 1/\pi = 1/\varphi$
Rupture threshold$C^* = 1/\Omega$dimensionlessCramér-Rao saturation point
Historical field$\varphi$$[F]$Reconstruction resource
Regeneration$R$$[F] \cdot [T]$Coherence-weighted integral of history

1. The Regeneration Mechanism

In Plain Language

When a system regenerates after rupture, it doesn't treat all of history equally. Important moments (high coherence) contribute more; forgettable moments (low coherence) fade away.

Crucially, this weighting is based on how coherent each moment was in its own context—not on how recent it was. A significant experience from years ago can shape you as much as one from yesterday, if both reached high coherence.

1.1 Weight Function

Definition (Coherence Weight)

The unnormalised weight at historical moment $\tau$ is simply the Boltzmann factor:

$$w(\tau) = \exp\!\bigl(C(x,\tau)/\Omega\bigr)$$

Here $C(x,\tau)$ is the coherence at moment $\tau$—how far through its cycle the system was at that time. Since $C$ cycles between $0$ and $C^* = 1/\Omega$ within each cycle, the weight $w(\tau)$ ranges from $\exp(0) = 1$ (at the start of each cycle) to $\exp(1/\Omega^2)$ (at each rupture moment).

Proposition (Weight Properties)
  1. $w(\tau) \geq 1$ for all $\tau$ (since $C \geq 0$)
  2. $w(\tau)$ peaks at moments just before rupture (when $C \approx 1/\Omega$)
  3. All rupture moments across all cycles have equal maximum weight $\exp(1/\Omega^2)$, regardless of when they occurred
  4. The contrast ratio (peak to trough) is $\exp(1/\Omega^2)$, which grows dramatically as $\Omega$ decreases—rigid systems have extremely selective memory

1.2 $\Omega$-Regime Behaviour

Remark (Temperature Analogy)

$\Omega$ plays the role of "temperature" in the Boltzmann weighting $\exp(C/\Omega)$:

  • Small $\Omega$ (low temperature, rigid): Weights concentrate explosively on high-coherence moments. The contrast ratio $\exp(1/\Omega^2)$ is enormous. Only moments near rupture matter for regeneration.
  • Large $\Omega$ (high temperature, flexible): Weights spread more uniformly across history. The contrast ratio is moderate. The entire history contributes to reconstruction.

But unlike a recency-biased model, all high-coherence moments contribute with the same weight, whether ancient or recent. Memory is democratic across time, selective across significance.

2. Non-Markovian Signature

Theorem (Path Dependence)

Two systems with identical current coherence but different histories will in general have different regenerations. This follows directly from the regeneration integral $R = \int \varphi \cdot \exp(C/\Omega) \cdot \Theta\, d\tau$ depending on the full history $\{C(x,\tau)\}_{\tau \in [0,t]}$, not just the current value $C(x,t)$.

What This Means

History matters, not just its summary. Two people at the same point in life but with different histories will respond differently to the same challenge. Significant experiences persist in their influence regardless of how long ago they occurred—muscle memory doesn't fade just because it's old. This is non-Markovian dynamics: the future depends on the entire integrated past, not merely the present state.

3. Comparison: Coherence-Weighted vs Recency-Weighted Memory

PropertyCRR (Coherence-Weighted)Recency-Weighted
What determines weight?Coherence at each moment: $\exp(C(\tau)/\Omega)$Time since event: $e^{-\lambda(t-\tau)}$
Ancient high-coherence momentsFully preserved (weight $= \exp(1/\Omega^2)$)Exponentially forgotten
Recent low-coherence momentsLow weight (near 1)High weight (recent)
Philosophical alignmentBergson: memory as continuous presenceStandard decay models
Empirical matchMuscle memory, trauma, skill retentionShort-term forgetting curves

4. The Unity of $\Omega$ in Regeneration

$\Omega$ appears in both the rupture condition ($C \cdot \Omega = 1$ triggers transformation) and the regeneration weighting ($\exp(C/\Omega)$ determines memory access). This unity connects two fundamental questions:

The Single Parameter

Large $\Omega$ (flexible, permeable boundary): Low rupture threshold ($C^* = 1/\Omega$ is small). Frequent micro-ruptures—the system reorganises readily. Moderate contrast in memory weighting—history is broadly accessible.

Small $\Omega$ (rigid, precise boundary): High rupture threshold ($C^* = 1/\Omega$ is large). Rare but significant ruptures—the system accumulates extensively before transforming. Extreme contrast in memory weighting—only peak-coherence moments survive.

E–I Balance and the Write Window

The Brain's Balancing Act

The brain constantly balances two opposing forces: excitation (E) and inhibition (I). Tucker, Luu, and Friston (2025) show that consciousness emerges when these forces are perfectly balanced—at criticality.

CRR proposes that E and I map onto $\Omega$. When $C \cdot \Omega = 1$—accumulated evidence saturates the system's capacity—the inside matches the outside. This is the write window, where early LTP can occur and experiences become memories.

$\Omega$ = 0.32    $C^*$ = 3.14
Variance Parameter ($\Omega = \sigma^2 = 1/\text{precision}$). Rupture threshold $C^* = 1/\Omega$.
LOW Ω — RIGID / HIGH PRECISION

$\mathbb{Z}_2$ Regime ($\Omega = 1/\pi \approx 0.318$)

CRR: Coherence accumulates to $C^* = \pi \approx 3.14$ before rupture. Moderate cycle duration. Memory sharply peaked on high-coherence moments.

CV prediction: $1/(2\pi) \approx 0.159$. Systems: heartbeat, neural bistability, flame flicker.

A. Threshold $C^*$ and Variance $\Omega$
$C^* = 1/\Omega$ (blue). Green dots mark $\mathbb{Z}_2$ ($\Omega = 1/\pi$) and $SO(2)$ ($\Omega = 1/2\pi$).
B. Memory Weighting $\exp(C/\Omega)$
Small $\Omega$: sharply peaked on high-$C$ moments. Large $\Omega$: broadly uniform.
C. CRR Dynamics
$C(t)$ accumulates until $C = 1/\Omega$ → rupture $\delta$ → reset.
D. CV = $\Omega/2$
Parameter-free CV prediction as function of $\Omega$. Dots: $\mathbb{Z}_2$ and $SO(2)$.

The Tucker–Luu–Friston Framework

SystemDorsal (Papez)Ventral (Yakovlev)
Control ModeExcitatory feedforwardInhibitory feedback
Sleep ConsolidationREMNREM
CRR MappingHigher $\Omega$ (flexible)Lower $\Omega$ (precise)
The Write Window: When $C \cdot \Omega = 1$

At rupture: E and I are balanced, a standing wave resonance forms (~100–200 ms), LTP can occur, the pattern is inscribed. The rupture boundary $\delta(\text{now})$ is the moment of maximal information transfer between past (coherence) and future (regeneration). This is the neural mechanism for James's "perchings."

Tucker, D.M., Luu, P. & Friston, K.J. (2025). The Criticality of Consciousness. Entropy, 27(8), 829.

CRR as Temporal Completion of the Free Energy Principle

What is the Free Energy Principle?

The FEP (Friston, 2010; 2019) proposes that living systems survive by minimising "free energy"—the mismatch between what they expect and experience. CRR provides the temporal dynamics that the FEP presupposes but leaves unspecified: when do beliefs update? How does accumulated history shape reconstitution?

The Temporal Gap in the FEP

The FEP's primary temporal apparatus is generalised coordinates of motion (Friston, 2008)—a vector of higher-order time derivatives that encodes local trajectory information. This is elegant for continuous dynamics within a regime, but it remains fundamentally local in time: each state depends only on its current generalised coordinates, preserving the Markov property. Biehl, Pollock & Kanai (2021) identified technical difficulties with this formulation.

The FEP's path integral formulation (Friston, 2019) scores the plausibility of entire trajectories, but still does not specify when a system must abandon one regime for another, nor how the transition draws on accumulated history. The FEP tells you that a system at nonequilibrium steady state will look as if it is performing inference. It does not tell you the timing of the inference, or the moment at which the current model is exhausted.

Three Specific Additions

FEP ProvidesCRR Adds
Markov blanket: a spatial boundary
(internal $\perp$ external | blanket)
Dirac delta: a temporal boundary
(future $\perp$ past | present). The rupture moment $\delta(\text{now})$ serves the same conditional-independence role in time that the blanket serves in space.
Dynamics within a regime
(VFE minimisation, predictive coding, active inference)
Transitions between regimes: $C \cdot \Omega = 1$ specifies exactly when inference is exhausted and the system must reorganise. This is the Cramér-Rao bound that underlies the FEP's own information geometry, now applied as a stopping condition.
Markovian dynamics: each state depends on the current state (or generalised coordinates) Non-Markovian accumulation: $C(x,t) = \int\mathcal{L}(x,\tau)\,d\tau$ integrates the full history. The present depends not on the previous state but on the entire accumulated past. Regeneration via $\exp(C/\Omega)$ weights this history exponentially.

The FEP's precision parameter (inverse variance, $\pi = 1/\Omega$) maps directly to CRR's $\Omega$. Where the FEP uses precision to weight prediction errors, CRR uses its reciprocal $\Omega$ to set the rupture threshold and memory depth. The frameworks share the same information geometry; CRR adds the temporal completion.

Rupture and Bayesian Model Reduction

The "Aha Moment" Correspondence

In Friston et al. (2025) "Active Inference and Artificial Reasoning," an "aha moment" occurs when evidence accumulates until confidence exceeds a threshold, triggering Bayesian Model Reduction (BMR). CRR's Rupture is the same phenomenon, given a precise temporal criterion: $C \cdot \Omega = 1$.

CRR ConceptFEP ConceptMapping
Coherence $C(x,t)$Accumulated evidence since last update$C \;\leftrightarrow\; \log p(D_{\text{new}}|m)$
Variance $\Omega = \sigma^2 = 1/\pi$Inverse precision$\Omega \;\leftrightarrow\; 1/\pi$
Rupture threshold $C^* = 1/\Omega$Model selection threshold$C^* \;\leftrightarrow\; \pi$ (precision)
Rupture $\delta(t-t_*)$BMR / Occam's Razor$\{C \cdot \Omega \geq 1\} \;\leftrightarrow\; \{\max_m p(m|D) > \theta\}$
Regeneration $R[\varphi]$Posterior after model selectionCoherence-weighted history $\;\leftrightarrow\;$ Bayesian posterior
The Central Correspondence

Rupture is the "aha moment." Both frameworks describe the discrete transition from uncertainty to commitment when accumulated evidence warrants model selection. CRR adds precision: the transition occurs at $C \cdot \Omega = 1$, with parameter-free predictions about its timing variability ($\text{CV} = \Omega/2$).

What CRR Does Not Replace

CRR does not compete with the FEP's account of what beliefs update to (free energy minimisation), nor with the detailed neural process theories (predictive coding, active inference) that implement it. CRR addresses the temporal structure of these processes: when transitions occur, how history shapes reconstitution, and why the timing variability takes the specific values it does. The FEP provides the engine; CRR provides the clock.

Friston, K. (2010). The free-energy principle: a unified brain theory? Nat. Rev. Neurosci. 11, 127–138.
Friston, K.J. (2019). A free energy principle for a particular physics. arXiv:1906.10184.
Friston, K., et al. (2025). Active inference and artificial reasoning. arXiv:2512.21129.
Tucker, D.M., Luu, P. & Friston, K.J. (2025). The Criticality of Consciousness. Entropy, 27(8), 829.
Parr, T., Pezzulo, G. & Friston, K.J. (2022). Active Inference: The Free Energy Principle in Mind, Brain, and Behavior. MIT Press.
Biehl, M., Pollock, F.A. & Kanai, R. (2021). A technical critique of some parts of the free energy principle. Entropy 23, 293.
Sakthivadivel, D.A.R. (2022). Towards a geometry and analysis for Bayesian mechanics. arXiv:2204.11900.

1. Parameter-Free Predictions

CRR's central empirical claim is that the coefficient of variation (CV) of inter-event intervals is determined by symmetry class alone, with no free parameters:

Theorem (CV Predictions from Symmetry Class)
$$\text{CV} = \frac{\Omega}{2} = \frac{1}{2\varphi}$$

where $\varphi$ is the phase (in radians) traversed during one coherence accumulation cycle.

SymmetryPhase to Rupture$\Omega$ Value$C^*$ ValueCV Prediction
$\mathbb{Z}_2$ (bistable/flip)$\pi$ (half-cycle)$1/\pi \approx 0.318$$\pi \approx 3.14$$1/(2\pi) \approx 0.159$
$SO(2)$ (rotational)$2\pi$ (full-cycle)$1/2\pi \approx 0.159$$2\pi \approx 6.28$$1/(4\pi) \approx 0.080$

2. Validation Scope

These predictions have been tested across 100+ systems in 30+ domains, including:

DomainExample SystemsClassStatus
Neural oscillationsEEG alpha, theta, gamma; sleep spindles$SO(2)$Validated (N=109, two independent datasets)
Cardiac rhythmsHeart rate variability, R-R intervals$\mathbb{Z}_2$Confirmed
Flame dynamicsCandle flicker, plasma oscillations$\mathbb{Z}_2$Confirmed
Bacterial divisionE. coli inter-division intervals$\mathbb{Z}_2$Confirmed
Stellar pulsationCepheid variables, RR Lyrae$SO(2)$Confirmed
Calcium signallingIntracellular Ca²⁺ oscillations$SO(2)$Confirmed
Reaction timesHuman simple RT, choice RT$\mathbb{Z}_2$Confirmed
Population ecologyPredator-prey cycles, bloom intervals$SO(2)$Confirmed
Laser dynamicsMode-locked laser pulse trains$SO(2)$Confirmed
Gastric wavesSlow-wave rhythm$SO(2)$Confirmed
Saltatory growthInfant growth spurts (Lampl & Johnson)$\mathbb{Z}_2$Confirmed (11/11 individual predictions)
GeophysicsGeyser eruptions, seismic cycles$\mathbb{Z}_2$/$SO(2)$Confirmed

The full validation table with 100+ systems, observed CVs, predictions, and references is available at CRR Benchmarks.

3. Key EEG Validation

EEG Validation: Two Independent Datasets

Tested across PhysioNet EEGBCI and MPI-LEMON datasets (N = 109 total):

  • 11/11 class orderings correct (every $SO(2)$ band showed lower CV than every $\mathbb{Z}_2$ band)
  • Fisher $z$-corrected CV ratio: 1.93 (95% CI containing the predicted value of 2.0)
  • Train-test correlation: $r = 0.997$
  • Cohen's $d = 2.01$

4. Three-Class Framework

Systems fall into three empirical classes based on their relationship to the CRR predictions:

ClassDescriptionCV Relative to $\Omega/2$Match Rate
Class AAutonomous stochastic (matches CRR)$\text{CV} \approx \Omega/2$89%
Class BDeterministic/regulated (precision oscillator)$\text{CV} < \Omega/2$ (suppressed)85%
Class CNoise-dominated/volitional$\text{CV} > \Omega/2$ (inflated)85%

Overall classification accuracy: 86%, approximately 10.6$\sigma$ significance, with zero directional reversals. Systems that deviate from the prediction tell you something specific about their regulatory architecture.

5. Falsification Criteria

The framework makes specific, falsifiable commitments:

Methodological Commitment

CRR follows a pre-registration discipline: predictions are formally registered before touching data. Deviations are diagnosed rather than hidden. The 132-system CV predictions table, three-class framework, and all EEG results were pre-registered. Honest null testing is a core commitment—e.g. the lemniscate hypothesis in atomic CV analysis was falsified and reported as such, not rescued.

Key EEG sources:
Schalk, G. et al. (2004). BCI2000: A general-purpose brain-computer interface system. IEEE Trans. Biomed. Eng. 51, 1034–1043. (PhysioNet EEGBCI)
Babayan, A. et al. (2019). A mind-brain-body dataset of MRI, EEG, cognition, emotion, and peripheral physiology. Sci. Data 6, 180308. (MPI-LEMON)
Lampl, M. & Johnson, M.L. (1998). Normal human growth as saltatory. Clin. Endocrinol.