The Two Kinds of Social Tipping Points

A single instigator can spark a revolution — or be silenced completely. Whether they succeed depends on a hidden mathematical threshold that governs collective behavior.

In a society on the edge, one person stepping forward can ignite a mass movement. But in another, the same act fizzles into silence. What determines whether a spark becomes a fire? A new analysis of Granovetter’s threshold model — a foundational framework for understanding social cascades — reveals that the fate of collective action hinges on a precise mathematical tipping point. For populations with a critical density of cautious individuals, the transition from inaction to mass mobilization is not gradual, but sudden: below a threshold parameter $α = 1$ , nearly everyone joins; above it, the movement dies with almost no one. And when the population includes a persistent fraction of highly resistant individuals, the shift becomes astonishingly smooth — so gradual that even in large societies, small changes in susceptibility can tip the balance. These findings aren’t just abstract math. They offer a rigorous lens for understanding everything from political uprisings and climate activism to the spread of misinformation and technological adoption.

The Science

Mark Granovetter’s 1978 threshold model is one of sociology’s most enduring ideas. It proposes that individuals decide to join a collective action — a protest, a strike, a viral trend — only when enough others have already done so. Each person has a personal threshold: the number of active participants required before they, too, will act. Some people join at the first sign of movement (low thresholds); others wait until nearly everyone is involved (high thresholds). The model captures how individual decisions, when aggregated, can produce explosive social change — or total stagnation.

While the model has been widely simulated and extended, exact analytical solutions have remained elusive, especially for finite populations where randomness plays a crucial role. José F. Fontanari’s new work (Fontanari, 2026) breaks this barrier by deriving an exact expression for the probability that a cascade stops at any given size $k$ in a population of $N$ people, starting from a single instigator.

The key innovation is modeling threshold heterogeneity using the Beta distribution, a flexible probability distribution bounded between 0 and 1. Each individual’s normalized threshold $x \in [0, 1]$ is drawn from $f (x; α, β)$ , where $α$ and $β$ are shape parameters:

f (x; α, β) = \frac{x ^{α - 1} ( 1 - x ) ^{β - 1}}{B ( α , β )}

Here, a small $α$ means many people have low thresholds — a volatile, easily mobilized society. A small $β$ indicates a significant number of high-threshold individuals — the cautious, skeptical types who resist joining until nearly everyone else has. This parameterization allows the model to span a wide range of social structures, from radicalized movements to deeply conservative societies.

Fontanari derives the exact probability $P_{N} (k)$ that a cascade halts with exactly $k$ active agents:

P_{N} (k) = (k - 1 N - 1) Q_{k} [1 - F (\frac{k}{N})]^{N - k}

where $Q_{k}$ is the probability that a group of $k$ individuals, including the instigator, can sustain a continuous chain of recruitment without stalling at any intermediate size. This recursive structure captures the full stochastic history of the cascade, making it possible to compute not just average outcomes, but the full distribution of possible cascade sizes.

The model assumes global interaction — every person can see how many others have joined — and synchronous updating, where all agents reassess their decision at each time step. While real-world networks are more complex, this simplification allows for exact solutions that serve as a benchmark for understanding the core mechanics of behavioral cascades.

What They Found

The most striking result is the existence of a sharp phase transition at $α = 1$ in the infinite-population limit. Below this point, a single instigator can trigger a global cascade involving nearly the entire population. Above it, the movement stalls almost immediately. But the nature of this transition depends critically on the value of $β$ , the parameter controlling the density of high-threshold individuals.

When $β = 1$ , corresponding to a power-law threshold distribution $F (x) = x^{α}$ , the transition is discontinuous — a first-order phase transition. The fraction of active individuals jumps abruptly from 0 to 1 as $α$ crosses 1. In finite populations, this manifests as a steep rise in activation probability over a narrow range of $α$ . The width of this critical region — the “scaling window” — shrinks as $N^{- 1/2}$ , meaning that in large societies, the transition becomes razor-sharp.

Critical Scaling for β=1

Width of the critical region for the power threshold distribution (β=1) as a function of population size N.

Critical Scaling for β=1
Label	Value
N = 8000	0.001
N = 16000	0.0005
N = 32000	0.00025
N = 64000	0.000125

This behavior is confirmed in Monte Carlo simulations (

Figure 1: Finite-size scaling of the cascade transition for the power function distribution.
(Left) The expected final macroscopic active fraction ρ∞\rho_{\infty} as a function of the susceptibility index α\alpha for system sizes N=8000,16000,32000N=8000,16000,32000, and 6400064000. The vertical dashed line marks the theoretical transition point at α=1\alpha=1.
(Right) Data collapse of the macroscopic fractions obtained by plotting the rescaled order parameter N1/2ρ∞N^{1/2}\rho_{\infty} against the finite-size scaling variable (α−1)N1/2(\alpha-1)N^{1/2}, confirming the critical exponent ν=2\nu=2. The vertical dashed line indicates the critical point (α−1)N1/2=0(\alpha-1)N^{1/2}=0. The cross symbol (×\times) at the coordinates (0,π/2)(0,\sqrt{\pi/2}) represents the exact analytical prediction for the finite-size scaling at α=1\alpha=1. In both panels, the yy-axis is displayed on a logarithmic scale to highlight the intersection points and the precise scaling behavior. All data points represent Monte Carlo ensemble averages over 10510^{5} independent threshold realizations. — Figure 1: Finite-size scaling of the cascade transition for the power function distribution. (Left) The expected final macroscopic active fraction ρ∞\rho_{\infty} as a function of the susceptibility index α\alpha for system sizes N=8000,16000,32000N=8000,16000,32000, and 6400064000. The vertical dashed line marks the theoretical transition point at α=1\alpha=1. (Right) Data collapse of the macroscopic fractions obtained by plotting the rescaled order parameter N1/2ρ∞N^{1/2}\rho_{\infty} against the finite-size scaling variable (α−1)N1/2(\alpha-1)N^{1/2}, confirming the critical exponent ν=2\nu=2. The vertical dashed line indicates the critical point (α−1)N1/2=0(\alpha-1)N^{1/2}=0. The cross symbol (×\times) at the coordinates (0,π/2)(0,\sqrt{\pi/2}) represents the exact analytical prediction for the finite-size scaling at α=1\alpha=1. In both panels, the yy-axis is displayed on a logarithmic scale to highlight the intersection points and the precise scaling behavior. All data points represent Monte Carlo ensemble averages over 10510^{5} independent threshold realizations. Source: José F. Fontanari

, left panel), where the expected final cascade size $ρ_{\infty}$ shifts from near-zero to near-one as $α$ increases through 1. The right panel shows a data collapse when plotting $N^{1/2} ρ_{\infty}$ against $(α - 1) N^{1/2}$ , confirming the predicted scaling exponent. At the critical point $α = 1$ , the finite-size scaling predicts $N^{1/2} ρ_{\infty} \to π /2 \approx 1.25$ , marked by the cross in

— a rare instance of an exact analytical prediction matching simulation data.

But when $β < 1$ , the story changes dramatically. Now, the population contains a persistent density of high-threshold individuals — those who will only act if nearly everyone else does. In this regime, the phase transition becomes continuous, and of infinite order: the active fraction grows smoothly from zero as $α$ decreases below 1. More remarkably, the finite-size scaling window contracts at a much slower rate — proportional to $(ln N)^{- 1}$ .

This means that in large but finite populations, the transition remains broad and gradual. Small changes in societal susceptibility can lead to large changes in mobilization, but there is no sudden jump. The system remains sensitive to perturbations even at scale.

Critical Scaling for β<1

Value of (α−1) at which ρ∞ reaches half-maximum for β=1/2, showing logarithmic scaling.

Critical Scaling for β<1
Label	Value
N = 8000	0.144
N = 16000	0.133
N = 32000	0.125
N = 64000	0.118
N = 128000	0.111

Simulations for $β = 1/2$ (

Figure 2: Cascade transition equilibrium for a Beta threshold distribution with shape parameter β=1/2\beta=1/2. (Left) The expected final macroscopic active fraction ρ∞\rho_{\infty} as a function of the control parameter α\alpha for system sizes N=8000,16000,32000N=8000,16000,32000, and 6400064000. (Right) A logarithmic magnification of the critical region, exposing the pronounced finite-size effects that are otherwise obscured on a linear scale. In both panels, the solid curves represent the infinite population limit obtained from the numerical solution of the fixed-point Equation (18). All data points are Monte Carlo ensemble averages computed over 10610^{6} independent threshold realizations. Source: José F. Fontanari

) show this smooth onset. The left panel displays $ρ_{\infty}$ rising continuously as $α$ decreases, with finite-size effects still visible even at $N = 128, 000$ . The right panel zooms in on the critical region, revealing how larger populations shift the curve but do not sharpen it abruptly.

When plotting $N ρ_{\infty}$ against $(α - 1) ln N$ , the data collapse onto a single curve (

), confirming the $(ln N)^{- 1}$ scaling. At $α = 1$ , the expected number of active agents approaches approximately 1.20, independent of $N$ — a signature of the infinite-order transition.

Discontinuous Phase Transition at α=1

Steady-state active fraction in the infinite-population limit for the power threshold distribution.

Discontinuous Phase Transition at α=1
Label	Value
α < 1	1
α > 1	0

Why This Changes Things

These findings reframe how we think about social change. Traditionally, tipping points are imagined as sudden, unpredictable shifts — like a straw breaking the camel’s back. But Fontanari’s work shows that the shape of the tipping point — whether it’s a cliff or a ramp — depends on the distribution of resistance within the population.

In societies where most people are willing to act early (low $\alpha$) and there are few ultra-cautious individuals (large $\beta$), change can be explosive. Think of the rapid spread of the #MeToo movement or the Arab Spring uprisings — once a critical mass was reached, participation surged almost overnight. These are $\beta = 1$-like regimes, where the transition is sharp and hard to reverse.

But in societies with many high-threshold individuals — those who need to see overwhelming consensus before acting — change is slower, more fragile, and more reversible. This might describe the adoption of plant-based diets, renewable energy, or new political ideologies in deeply polarized environments. Here, progress is incremental, and setbacks can unravel gains. The $(ln N)^{- 1}$ scaling means that even in large populations, small shifts in public sentiment can tip the balance — offering hope for change, but also vulnerability to backlash.

The $β$ parameter, in particular, emerges as a crucial but previously overlooked factor. Most models focus on the average threshold or the number of early adopters. But Fontanari shows that the tail of the distribution — the presence of highly resistant individuals — determines the very nature of the transition. A society can have the same average susceptibility but behave completely differently depending on whether it has a few diehard holdouts.

This has profound implications for activism and policy. In a $β < 1$ world, efforts should focus on reducing the number of high-threshold individuals — not just by persuasion, but by altering the social context so that joining feels less risky. Social proof, visibility, and institutional support can all lower perceived thresholds. In contrast, in a $β = 1$ world, the goal is to reach the critical $α$ threshold — a classic tipping point strategy.

The model also helps explain why some movements spread globally while others remain local. In a finite population, the probability of a large cascade depends on avoiding early stalling — a problem governed by $Q_{k}$ , the survival probability of the recruitment chain. The exact solution shows that even in volatile societies ($\alpha < 1$), a bad draw of thresholds in the first few steps can kill the cascade. This randomness is why identical movements can succeed in one country and fail in another.

What’s Next

Fontanari’s work is a theoretical breakthrough, but it raises urgent questions for real-world application. The model assumes global visibility — everyone knows how many people have joined. In reality, social influence is networked: we see only our friends, not the whole population. How do these phase transitions change on complex networks? Preliminary work suggests that clustering and community structure can suppress cascades, but an exact solution for networked thresholds remains out of reach.

Another open question is how thresholds evolve over time. The model treats them as fixed, but in reality, people update their beliefs based on experience. A protest that grows slowly might lower thresholds through momentum; one that stalls might increase them through discouragement. Incorporating adaptive thresholds could lead to richer dynamics, including oscillations or delayed tipping.

Perhaps most importantly, the model assumes homogeneous influence — each additional participant counts equally. But Granovetter himself emphasized the “strength of weak ties” and the outsized influence of close relationships. Extending the model to include weighted influence — where some people matter more than others — could bridge the gap between this exact solution and the messy reality of social movements.

Finally, while the Beta distribution is flexible, real threshold distributions are unknown. Survey data on willingness to act under different levels of participation could test whether $α$ and $β$ vary systematically across issues and cultures. Do climate activists have lower $α$ than vaccine skeptics? Do authoritarian regimes produce higher $β$ ?

The beauty of this work is that it turns vague ideas about “tipping points” and “social momentum” into precise, testable predictions. It shows that collective behavior isn’t just unpredictable chaos — it follows mathematical laws. And while no model captures every nuance, having an exact benchmark allows us to isolate which features matter most. In a world where social change feels both urgent and elusive, this is a rare source of clarity: we now know not just that tipping points exist, but what kind they are — and how to find them.