Self-Organized Quasi-Criticality (SOqC)

• Self-organized quasi-criticality is a dynamical regime where systems hover near a phase transition due to local adaptation and non-conservative dynamics.
• Mathematical models use coupled slow-fast dynamics with adaptive feedback, producing finite power-law avalanche statistics and broad control parameter distributions.
• SOqC challenges strict criticality by explaining scale-invariant behavior in complex systems like neural networks, atmospheric phenomena, and dissipative models.

Self-organized quasi-criticality (SOqC) describes dynamical regimes in which complex, driven systems with local adaptation and/or non-conservative dynamics spontaneously approach and maintain a broad region near a critical point of a phase transition, but do not achieve true criticality in the thermodynamic limit. Systems exhibiting SOqC display approximate scale invariance and power-law statistics (avalanches, correlations) across finite ranges, but key signatures required for full criticality—such as a singularly tuned control parameter, infinite correlation length, or vanishing variance of the control parameter—are absent. SOqC bridges the gap between ideal self-organized criticality (SOC) and realistic, dissipative systems across physics, neuroscience, atmospheric science, and beyond.

1. Conceptual Foundations and Distinction from SOC

Classical SOC is characterized by self-tuning to a non-equilibrium continuous phase transition, achieved through local conservation and infinite separation of driving and relaxation timescales—the archetype being the Bak–Tang–Wiesenfeld (BTW) sandpile. True SOC yields singular tuning of a global control parameter (e.g., mean energy, branching ratio), vanishing variance in the thermodynamic limit, and signatures such as power-law event size distributions with exponents governed by universality classes (typically conserved directed percolation, C-DP) (Buendía et al., 2020, Costa et al., 2014, Tian et al., 2023).

SOqC emerges when core SOC requirements are relaxed, notably:

• Non-conservative dynamics: Conservation (of energy, mass, stress, or “activity”) is broken in the bulk by dissipation (Buendía et al., 2020, Tian et al., 2023).
• Finite timescale separation: The ratio of driving to relaxation is large but finite, so the system cannot perfectly segregate slow input from fast dissipation (Kinouchi et al., 2018).
• Adaptive feedback: The control parameter (e.g., threshold, synaptic strength, gain) is itself a dynamical variable tuned by local homeostatic feedbacks rather than strict conservation (Kinouchi et al., 2020, Menesse et al., 2021, Kinouchi et al., 2018).

In SOqC, the system hovers around the transition point with a finite-width distribution of control parameters and large, but non-diverging, susceptibility and correlation lengths (Palmieri et al., 2018, Tian et al., 2023). The stationary distribution of the control parameter remains broad (not delta-peaked), and the system’s temporal evolution displays quasi-stationary or slowly drifting behaviors (Manna, 2023, Buendía et al., 2020).

2. Mathematical Formulation and Feedback Mechanisms

The mathematical structure of SOqC is generically that of coupled slow-fast dynamical systems, often incorporating nonconservative Langevin or difference equations alongside adaptive feedback:

$$\partial_t \rho(\mathbf{x}, t) = \left[ a + \omega E(\mathbf{x}, t) \right] \rho - b \rho^2 + D \nabla^2 \rho + \sigma \sqrt{\rho} \eta(\mathbf{x}, t)$$

$$\partial_t E(\mathbf{x}, t) = \nabla^2 \rho - \epsilon \rho + f(t)$$

where:

• $\rho(\mathbf{x}, t)$ is the local activity (density of active sites, neurons, etc.),
• $E(\mathbf{x}, t)$ is the local “energy” or control parameter,
• $\epsilon > 0$ is the bulk dissipation rate,
• $f(t)$ is possibly a slow or stochastic driving term (Buendía et al., 2020, Tian et al., 2023).

Integration over avalanches or the system’s history introduces non-Markovian memory terms in the dynamics for $\rho$, breaking strict SOC and yielding dynamical percolation universality rather than C-DP (Tian et al., 2023).

In adaptive networks, dynamical rules for the control parameter $E$ (e.g., synaptic weights $W_{ij}$, neuronal gains $\Gamma_i$, or thresholds $\theta_i$) typically follow

$$E[t+1] = E[t] + \frac{1}{\tau}(A - E[t]) - u \rho[t]$$

with relaxation (homeostasis: $A$), utilization/depression ($u$), and time constant $\tau$ (Kinouchi et al., 2020, Menesse et al., 2021, Kinouchi et al., 2018). The corresponding order parameter $\rho$ (network activity, firing rate, etc.) interacts nonlinearly with $E$; the coupled map admits a weakly stable focus near threshold (a quasi–critical attractor) (Kinouchi et al., 2018).

3. Statistical and Scaling Laws in SOqC

SOqC systems exhibit:

• Power-law avalanche statistics for sizes $S$ and durations $T$, $P(S) \sim S^{-\tau}$, $P(T) \sim T^{-\alpha}$, over a finite range (Tian et al., 2023, Kinouchi et al., 2020, Manna, 2023, Spineanu et al., 2014). Typical exponents are $\tau \approx 3/2$, $\alpha \approx 2$ in mean field, but actual values may depend on the underlying model and parameter regime.
• Finite cutoffs for avalanche sizes (and correlation lengths), governed by system size, the width of the quasi-critical region, and non-universal model parameters (Palmieri et al., 2018, Tian et al., 2023).
• Fat-tailed distributions of instantaneous correlation lengths $P(\xi) \sim \xi^{-\lambda_c}$, with $2 < \lambda_c \le 3$; mean $\langle \xi \rangle$ is finite, but the variance may diverge for $\lambda_c \le 3$ (Palmieri et al., 2018).
• Broken finite-size scaling: As the system hovers over a range of near-critical parameter values, data collapse is only approximate, and the upper cutoff in scaling laws does not scale cleanly with system size (Tian et al., 2023, Palmieri et al., 2018).

Table: Key Features Contrast

Feature	SOC	SOqC
Conservation	Exact (bulk)	Broken (bulk dissipation)
Time-scale separation	Infinite	Large but finite
Control parameter PDF	$\delta(E-E_c)$	Broad, fat tailed
Critical exponents	Universal (C-DP)	Approximate, parameter-dependent
Correlation length	Diverges ($L \to \infty$)	Large but finite
Susceptibility	Diverges	Broad peak, finite

(Buendía et al., 2020, Costa et al., 2014, Tian et al., 2023, Palmieri et al., 2018)

4. Canonical Models and Application Domains

Sandpiles & Lattice Automata: SOqC arises when open boundaries or strict conservation are removed, as in Manna’s infinite-lattice sandpile with only local injection and no dissipation. The bulk self-organizes to a quasi-steady state with scale-free avalanches, but true stationarity is only asymptotically approached, and the range of scaling is constrained by the system size and domain growth (Manna, 2023).

Bak–Sneppen and Punctuated Equilibrium: Mapping atmospheric convection to the Bak–Sneppen model with $K=2$ demonstrates how slow forcing and threshold-triggered local responses yield correlated, near-marginal configurations with scale-free cluster distributions. The master equation for updating “barriers” recovers observed scaling exponents for convective event sizes and durations ($\approx 1.3$–$1.4$), but the system never undergoes a true phase transition; the quasi-equilibrium “hovers” near but not at the instability (Spineanu et al., 2014).

Neuronal Networks: Adaptive networks with dynamic synapses, gains, or thresholds (e.g., LHG model, integrate-and-fire with threshold adaptation) implement local feedbacks that stabilize around the critical region but do not eliminate finite excursions and stochastic oscillations (“quasicycles”). The result is a weakly stable focus in the $(\rho, E)$ phase space, with avalanche and interspike interval statistics reflecting SOqC rather than strict SOC (Kinouchi et al., 2020, Kinouchi et al., 2018, Menesse et al., 2021).

Forest-Fire and Dissipative Models: The Drossel-Schwabe forest-fire model, as analyzed by Juhász et al., displays “weak criticality”—event size and correlation length distributions approximate power laws, but the average correlation length $\langle \xi \rangle$ remains finite and the critical region is a broad peak, not a singularity (Palmieri et al., 2018, Buendía et al., 2020).

5. Functional and Physical Implications

SOqC resolves the apparent contradiction between scale-invariant avalanches and the absence of singular critical points or diverging correlation lengths in real, dissipative, or driven adaptive systems (Tian et al., 2023). In neural models and possibly biological brains, SOqC provides a robust regime where critical-like processing advantages (e.g., maximal dynamic range, flexibility) persist without pathological runaway, and without requiring fine-tuned external control.

Atmospheric convection and precipitation, earthquake faults, and even some epidemic models exhibit behaviors compatible with SOqC rather than pure SOC: bursts or avalanches with broad (but finite) scaling, broad residence-time distributions around “critical” values, large fluctuations, but no infinite correlation or susceptibility (Spineanu et al., 2014, Palmieri et al., 2018).

A critical implication is that many empirical systems previously claimed to be SOC are, upon closer scrutiny, more accurately described by SOqC—particularly when conservation laws are violated or adaptation is mediated by slow, local feedbacks (Buendía et al., 2020, Kinouchi et al., 2020).

6. Analytical, Numerical, and Experimental Diagnostics

Rigorous identification of SOqC relies on a battery of statistical tests:

• Maximum-likelihood exponent estimation for event size and duration distributions, with bootstrapping and cutoff determination (Tian et al., 2023).
• Finite-size scaling collapse: Test for consistent scaling of cutoff with system size, examining deviations that signal quasi-criticality (Manna, 2023, Tian et al., 2023).
• Temporal analysis: Power spectra and autocorrelations of control parameter time-series reveal slow, sawtooth stochastic oscillations indicative of a weakly stable focus (Kinouchi et al., 2018, Kinouchi et al., 2020).
• Distribution of instantaneous correlation lengths: Extract $P(\xi)$ from snapshots, test for fat tails and the limiting value of the exponent $\lambda_c$ (Palmieri et al., 2018).
• Comparative analysis: Cross-referring SOC diagnostics (e.g., vanishing variance of the control parameter, strict data collapse) versus SOqC features (finite variance, skewed scaling) (Costa et al., 2014).

7. Open Questions and Research Directions

Several key open problems persist:

• Universality: The precise boundaries between SOC and SOqC universality classes as parameters (dissipation, adaptation time constants) are varied (Buendía et al., 2020, Kinouchi et al., 2018).
• Role of Adaptation: How complex, multiscale or metaplastic homeostatic rules adjust the breadth and location of the quasi-critical regime (Kinouchi et al., 2020, Menesse et al., 2021).
• Experimental Validation: Direct detection of SOqC-specific phenomena (fat-tailed control parameter distributions, stochastic focus oscillations) in cortical recordings, climate time series, or real sandpiles (Kinouchi et al., 2020, Tian et al., 2023).
• Generalization: Applicability to other dynamical systems—e.g., epidemic, voter, or synchronization models—with varying forms of adaptation or feedback (Menesse et al., 2021).
• Theoretical Extension: Extension to first-order transitions (self-organized bistability), Hopf (oscillatory criticality), and networks with spatial heterogeneity (Buendía et al., 2020, Kinouchi et al., 2020).

A plausible implication is that SOqC forms an umbrella for a wide class of “dirty critical” phenomena in natural and engineered systems—expanding classical SOC theory to include realistic adaptive and dissipative settings.

---

References

• (Spineanu et al., 2014): Regimes of self-organized criticality in the atmospheric convection
• (Manna, 2023): Nonstationary but quasisteady states in Self-organized Criticality
• (Costa et al., 2014): Can dynamical synapses produce true self-organized criticality?
• (Kinouchi et al., 2018): Stochastic oscillations produce dragon king avalanches in self-organized quasi-critical systems
• (Palmieri et al., 2018): The Emergence of Weak Criticality in SOC systems
• (Tian et al., 2023): Theoretical foundations of studying criticality in the brain
• (Kinouchi et al., 2020): Mechanisms of self-organized quasicriticality in neuronal networks models
• (Menesse et al., 2021): Homeostatic Criticality in Neuronal Networks
• (Buendía et al., 2020): Feedback mechanisms for self-organization to the edge of a phase transition