Information-Theoretic Measures¶

Four families of measurable quantities bear on the signature the Four-Model Theory predicts: neuronal avalanche exponents, detrended fluctuation analysis, and branching parameters as convergent criticality signatures, and Lempel-Ziv complexity as a measure of how rich the dynamics are.

The theory's requirement is free compute — Wolfram Class-4 (universal-computation) capability actually deployed for open-ended self-modeling — and criticality is the dynamical signature this leaves in the substrate. That signature is stated qualitatively; to test it empirically, it must be translated into measurable quantities. The neuroscience of criticality has developed precisely these tools, consolidated in the ConCrit framework (Algom & Shriki, 2026) and the ~140-dataset meta-analysis of Hengen and Shew (2025). In the biological case the theory commits to the convergence of three signatures -- branching ratio, DFA exponent, and avalanche statistics -- with the DFA exponent as its primary operational measure. Lempel-Ziv complexity is a fourth, related family: it indexes the complexity (differentiation) of the dynamics rather than criticality itself.

Lempel-Ziv Complexity (LZc)¶

Lempel-Ziv complexity measures the algorithmic complexity of a signal — roughly, how many distinct patterns it contains. Applied to neural signals (EEG, MEG), LZc quantifies the informational richness of brain dynamics.

LZc is not itself a criticality measure. It measures the complexity dimension of the dynamics -- how rich the computation is -- which the theory separates from extent, how much of the substrate is recruited into the critical process. It is useful as a complexity proxy that co-varies with criticality across states, not as a test of criticality.

Empirical findings consistently show that LZc tracks consciousness level: - High LZc: Normal waking, psychedelic states (at or slightly past criticality, short of runaway) - Intermediate LZc: REM sleep (near criticality) - Low LZc: Deep NREM, propofol at surgical depth (subcritical) - Low LZc despite high co-activation: Generalized seizure -- hypersynchronous, low-differentiation activity that has left Class 4 even though most of cortex is active

Schartner et al. (2017) found increased spontaneous MEG signal diversity under psychoactive doses of ketamine, LSD, and psilocybin. Ketamine's rising entropy fits the distinction the Four-Model Theory draws: propofol pushes the substrate subcritical, and at surgical depth reportable experience is absent, while ketamine does not push it subcritical but disrupts its inputs, leaving consciousness present but disconnected.

Neuronal Avalanche Exponents¶

A neuronal avalanche is a cascade of neural activity triggered by a single event and propagating through the network. At criticality, the distribution of avalanche sizes follows a power law: many small avalanches, fewer medium ones, very few large ones, with no characteristic scale.

The critical exponent (typically near -3/2 for size distribution and -2 for duration distribution) is the signature of self-organized criticality (Beggs & Plenz, 2003). Deviations from these exponents indicate departure from criticality: - Steeper exponents (subcritical): Activity dies out too quickly — avalanches are too small - Shallower exponents (supercritical): Activity propagates too freely — avalanches grow uncontrollably - Power-law with critical exponents: The system is at the edge — information propagates across the network without either dying out or exploding

This provides a quantitative test of whether a system shows the criticality signature the theory predicts, with two caveats. Non-critical processes can also produce apparent power laws (Touboul & Destexhe, 2017), which is why the theory commits to the convergence of several signatures rather than any one. And the markers need not come from a critical point at all: an extended phase of long-range order produces avalanche power laws and long-range correlations with no critical point, and such a phase still counts as a Class 4 regime (Sipling, Zhang & Di Ventra, 2026).

Detrended Fluctuation Analysis (DFA)¶

DFA measures long-range temporal correlations in a signal. At criticality, neural dynamics exhibit temporal correlations that extend across multiple timescales — a single perturbation influences dynamics seconds to minutes later, producing a DFA exponent α in the range 0.6–0.9 (between uncorrelated noise at 0.5 and non-stationary drift above 1.0). The theory designates α as its primary operational measure: it is computable from standard EEG/MEG without spike-level resolution and avoids the distributional ambiguities of power-law fitting in finite data.

This measure captures one aspect of Class 4 dynamics: temporal depth. A system at criticality does not simply respond to the current input — it integrates information across time, maintaining a "memory" of past states that influences future dynamics. This temporal integration is precisely what the theory requires for sustained self-simulation: the explicit models must maintain coherent content across time, not merely react to moment-by-moment input.

Branching Parameter (sigma)¶

The branching parameter measures the average number of descendant activations triggered by a single neural activation. At criticality, sigma = 1: each activation triggers, on average, exactly one subsequent activation. Activity neither dies out (sigma < 1, subcritical) nor explodes (sigma > 1, supercritical).

Priesemann et al. (2013, 2014) found that the waking brain operates slightly below criticality (sigma ≈ 0.98) — a small safety margin that prevents seizure-like runaway activation while retaining enough dynamic range and long-range correlation for coherent self-simulation. This "slightly subcritical" finding is consistent with the theory: the brain operates near the edge of chaos, not necessarily at it, balancing computational power against stability.

Figure¶

graph LR
    subgraph "Four Measures of Criticality"
        LZ["Lempel-Ziv<br/>Complexity<br/><em>Algorithmic<br/>complexity of signal</em>"]
        AV["Avalanche<br/>Exponents<br/><em>Power-law<br/>size distribution</em>"]
        DFA["DFA<br/>Exponent<br/><em>Long-range<br/>temporal correlations</em>"]
        BR["Branching<br/>Parameter σ<br/><em>Activation<br/>propagation ratio</em>"]
    end

    LZ --> CRIT["CRITICALITY<br/>ASSESSMENT<br/>Class 4?"]
    AV --> CRIT
    DFA --> CRIT
    BR --> CRIT

    CRIT --> CON{"Free compute<br/>at work?"}
    CON -->|"All measures<br/>in critical range"| YES["Yes:<br/>Class 4 signature"]
    CON -->|"Measures indicate<br/>sub/supercritical"| NO["No:<br/>signature absent"]

    style CRIT fill:#e74c3c,stroke:#333,color:#fff
    style YES fill:#2ecc71,stroke:#333,color:#000
    style NO fill:#95a5a6,stroke:#333

Four complementary measures bear on a single question: is the system operating in the Class 4 regime? Avalanche exponents capture spatial propagation, DFA captures temporal depth, and the branching parameter captures activation dynamics -- the three criticality signatures whose convergence the theory commits to. Lempel-Ziv complexity adds the richness of the dynamics as a complexity proxy. Together, they provide a quantitative operationalization of the signature the theory predicts — the fingerprint of free compute spent on self-modeling.

Key Takeaway¶

The criticality signature is measurable. Three convergent signatures (avalanche exponents, DFA, branching parameter), with Lempel-Ziv complexity as a complexity proxy, quantify the Class 4 regime in neural tissue, enabling empirical testing of the theory's physical foundation: free compute at work.