Susceptibility for extremely low external fluctuations and critical behaviour of Greenberg-Hastings neuronal model

This paper demonstrates that in the Greenberg-Hastings neuronal model, the spontaneous activation probability acts as an external field that drives a dynamical phase transition, exhibiting clear finite-size scaling behavior and critical exponents as this probability approaches zero.

The Gist

The "Spark and the Forest Fire" Theory: Understanding Brain-Like Networks

Imagine you are looking at a massive, dark forest at night. Most of the trees are quiet and still. Occasionally, a tiny, random spark (like a lightning strike or a stray ember) might land on a tree. If that spark is strong enough, it might start a small fire. If the trees are close together and the conditions are just right, that small fire could turn into a massive, roaring wildfire that sweeps across the entire forest.

This paper studies a mathematical model of this exact phenomenon, but instead of trees and fire, it uses neurons and electrical activity. This is called the Greenberg-Hastings model.

Here is the breakdown of what the scientists discovered, using everyday language.

---

1. The "Background Noise" Problem (The Spontaneous Spark)

In a real brain, neurons don't just sit there in total silence; there is always a little bit of "background noise" or random activity. In the researchers' model, they call this $r_1$ (spontaneous activation).

For a long time, scientists noticed something weird: when they included this background noise, the mathematical "signals" of a major transition (like a brain moving from sleep to wakefulness) seemed to disappear. It was like trying to hear a whisper in a crowded stadium—the background noise was so loud it drowned out the important signals.

The Discovery: The researchers found that this background noise isn't just "clutter"—it actually acts like an external force (like a constant wind blowing through the forest). If you turn the noise down low enough, the "wildfire" (the massive brain activity) suddenly becomes visible again, and you can see the beautiful, mathematical patterns of how the system transitions from quiet to active.

2. The "Social Butterfly" vs. "The Lone Wolf" (Activation Mechanisms)

How does one neuron decide to "fire"? The paper identifies three ways this happens:

• The Lone Wolf (Spontaneous): A neuron fires all by itself, just because it felt like it (the random spark).
• The Social Butterfly (Single Activation): One neighbor neuron fires, and its signal is strong enough to pull the next one along.
• The Group Chat (Cooperative Activation): One neighbor isn't enough, but three or four neighbors firing at once creates a "group effort" that triggers the next neuron.

The Discovery: The researchers found that the "Group Chat" method is the real game-changer. If the network is loosely connected (like a few scattered trees), the "Lone Wolf" and "Social Butterfly" methods rule. But as the network gets more crowded and complex (like a dense jungle), the "Group Chat" becomes the dominant way the whole system catches fire. This "group effort" is what causes the transition to go from a slow, steady burn to a sudden, explosive burst.

3. The Mystery of the "Unknown Class" (The Mathematical Fingerprint)

In physics, every type of transition has a "fingerprint"—a specific set of numbers called critical exponents that tell you exactly what kind of process is happening. For example, water boiling has a different fingerprint than a forest fire.

The Discovery: When the scientists ran their simulations on complex, "small-world" networks (networks that mimic how human brains are wired), they found a fingerprint that doesn't match any known category. It’s like finding a new animal in the wild that has the wings of a bird but the scales of a fish. It doesn't fit the "standard" rules of physics they expected, suggesting that the way brain-like networks behave is even more unique and complex than we previously thought.

---

Summary: Why does this matter?

By understanding how tiny, random sparks turn into massive waves of activity, scientists can better understand:

• Brain States: How we move from sleep to being awake.
• Brain Disorders: How a tiny bit of "noise" might trigger a massive, uncontrolled wave of activity, like an epileptic seizure.
• Complexity: Why the brain is so much more than just a collection of individual cells, but a massive, interconnected "social network" of electricity.

Technical Summary

Technical Summary: Susceptibility for Extremely Low External Fluctuations and Critical Behaviour of Greenberg-Hastings Neuronal Model

1. Problem Statement

The study investigates the dynamical phase transitions of the Greenberg-Hastings (GH) neuronal model on Watts-Strogatz (small-world) networks. A long-standing puzzle in this model was the observation that the fluctuation susceptibility ($\chi$) associated with the average network activity ($f_a$) appeared to lack a size-dependent divergent peak, which is a hallmark of second-order phase transitions in statistical mechanics.

The authors aim to determine whether this lack of scaling is intrinsic to the model or caused by the presence of a small, constant spontaneous activation probability ($r_1$). They hypothesize that $r_1$ acts as an external field conjugated to the order parameter, thereby driving the system away from criticality and suppressing finite-size scaling.

2. Methodology

The researchers employed a combination of numerical simulations and analytical mean-field approximations:

• Model Setup: A three-state cellular automaton (quiescent, excited, refractory) on a Watts-Strogatz network with quenched exponential synaptic weights. The control parameter is the activation threshold $T$.
• Numerical Approaches for $r_1 = 0$: Since setting $r_1 = 0$ leads to an "absorbing state" (where all neurons become quiescent), standard simulations fail. To study the critical regime, they used two quasi-stationary methods:
  1. Reactivation Algorithm: Restarting the system with a fraction of excited neurons whenever it hits the absorbing state.
  2. Fixed Time Algorithm: Only considering network realizations that remain active for a predefined maximum time ($t_{max}$).
• Limit Analysis: They performed simulations with decreasing values of $r_1 \to 0$ to observe the emergence of scaling.
• Analytical Approach: A mean-field approximation was developed to model activation mechanisms (spontaneous, single, and cooperative) to compare with numerical data.
• Quantities Measured: Average activity ($f_a$), fluctuation susceptibility ($\chi$), and the first coefficient of the autocorrelation function ($AC(1)$) to estimate the critical threshold $T_c$.

3. Key Contributions and Results

• Identification of $r_1$ as an External Field: The study confirms that the "missing" scaling in previous literature was due to $r_1 > 0$. As $r_1 \to 0$, a clear finite-size scaling behavior emerges in the susceptibility. This proves that $r_1$ plays the role of an external field conjugated to the order parameter $f_a$.
• Discovery of a New Universality Class: By applying finite-size scaling to the $r_1 = 0$ case, the authors estimated critical exponents ($\beta/\nu d \approx 0.3$ and $\gamma'/\nu d \approx 0.25$). Crucially, these exponents do not match any known universality class (such as Directed Percolation), suggesting that the quenched disorder of the network or rare-region effects significantly alter the critical behavior.
• Characterization of Activation Mechanisms: The authors categorized activation into three types:
  1. Spontaneous: Driven by $r_1$.
  2. Single: One neighbor is strong enough to trigger activation ($W_{ij} > T$).
  3. Cooperative: The sum of multiple neighbors triggers activation.
     They found that for low connectivity ($\langle k \rangle$), single activation dominates. As $\langle k \rangle$ increases, the transition becomes increasingly dominated by cooperative mechanisms, eventually shifting the phase transition from continuous (second-order) to discontinuous (first-order).
• Validation of Mean-Field Theory: The mean-field approximation successfully predicted the critical threshold $T_c \approx \ln(k)/\lambda$ in the continuous regime, confirming that cooperative effects are negligible near the critical point for low-to-intermediate connectivity.

4. Significance

This work provides a fundamental clarification of the phase transition landscape in excitable media and neural models. By demonstrating how microscopic spontaneous activity can mask macroscopic critical scaling, the paper offers a cautionary framework for interpreting criticality in biological systems (like the human brain), where "noise" or spontaneous firing is always present. Furthermore, the identification of a potentially new universality class in small-world networks opens new avenues for studying how topological disorder and synaptic weight distributions influence the collective dynamics of neural networks.