🔬 physics

Loops, not groups: Long cycles are responsible for discontinuous phase transitions in higher-order network contagions

This paper demonstrates that in higher-order network contagions, discontinuous phase transitions arise specifically from long cycles that enable feedback between distinct transmission chains, rather than from local group effects alone.

Original authors: Leah A. Keating, Laurent Hébert-Dufresne

Published 2026-06-30

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Leah A. Keating, Laurent Hébert-Dufresne

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

The Big Question: Why do some ideas spread in a "jump"?

Imagine you are trying to start a fire.

Simple Contagion: If you light a match, it catches a dry leaf, which catches a twig, which catches a branch. The fire grows slowly and steadily. This is like a standard virus or a simple rumor; one person telling another is usually enough.
Complex Contagion: Sometimes, an idea or behavior doesn't spread until a person hears it from multiple sources. For example, you might ignore a new diet trend if one friend suggests it, but if three friends suggest it, you suddenly join in. This is "complex contagion."

Scientists have long believed that the reason these "complex" trends cause sudden, explosive jumps (discontinuous phase transitions) is because people are in groups (like triangles of friends). They thought, "If you are in a group, hearing the same thing twice from different people in that group makes you more likely to act."

This paper says: "Actually, that's not the whole story. It's not the group itself; it's the long loops."

The Setting: The "Triangle" Network

The researchers studied a specific type of social network made of two things:

Individual connections: Just two people talking (a line).
Triangles: Three people all connected to each other (a group).

They used a mathematical model to simulate how a "contagion" (like a trend or virus) spreads through this web.

The Experiment: Separating "Groups" from "Loops"

The authors realized that in a triangle, there are two ways a person can hear a message twice:

The Group Effect: Person A tells Person B, and Person B tells Person C. Person C hears it from B, but B got it from A. It's one long chain of gossip inside the group.
The Loop Effect: Person A tells Person C directly. At the same time, Person B (who got the message from A) tells Person C. Now, Person C hears the message from two separate paths that started at the same source but took different routes. This creates a "loop."

The researchers ran simulations to see which of these two scenarios caused the "explosive jump" in adoption.

The Surprising Results

They tested three scenarios:

1. Groups with No Loops (The "Chain" Scenario)

Setup: People get more likely to adopt if they hear it twice within their triangle, but only if it comes from a single chain of transmission (A $\to$ B $\to$ C).
Result: The spread grew slowly and smoothly. Even with "group pressure," the transition was gentle.
Analogy: Imagine a line of people passing a bucket of water. Even if the last person needs two buckets to put out a fire, if the water comes from one single line, the fire goes out gradually.

2. Loops with No Groups (The "Cross-Connection" Scenario)

Setup: People get more likely to adopt if they hear it from two separate transmission chains that happen to meet at the same person (A $\to$ B $\to$ C and A $\to$ C).
Result: Boom. The system jumped suddenly from "nothing happening" to "everything happening." This is the discontinuous phase transition.
Analogy: Imagine two different fire hoses aiming at the same spot. If both hit at once, the pressure builds up instantly, causing a sudden, massive change.

3. Groups AND Loops (The Real World)

Setup: Both effects are present.
Result: The sudden jump still happened. But when the researchers turned off the "loop" effect (keeping only the "group" effect), the jump disappeared.

The Conclusion: It's the Detour, Not the Group

The paper's main claim is that long cycles (loops) are the secret ingredient for sudden, explosive changes in how things spread.

Groups alone (hearing the same thing twice from a single chain of friends) make things spread faster, but they don't cause the sudden "jump."
Loops (hearing the same thing from two independent paths that cross) are what cause the system to snap from "no spread" to "total spread."

Why Does This Matter?

For a long time, scientists thought that simply having "groups" (like cliques or triangles) was enough to explain why some social movements explode overnight. This paper corrects that view. It suggests that the structure of the network—specifically, whether there are long, winding paths that allow a message to arrive at a person from two different directions simultaneously—is what actually triggers the explosion.

If you want to predict when a trend will go viral or when a rumor will take over, you shouldn't just look at how many groups people are in. You need to look at how many loops exist in the network that allow information to circle back and reinforce itself from different angles.

Technical Summary: Loops, not groups: Long cycles are responsible for discontinuous phase transitions in higher-order network contagions

Problem Statement
Recent studies on dynamics in higher-order networks have frequently claimed that the structural presence of groups (e.g., triangles or cliques) is the primary driver of discontinuous phase transitions in complex contagion processes. However, standard mean-field analyses often fail to distinguish between the specific network pathways utilized by dynamical processes. Specifically, these models struggle to disentangle the effects of "group effects" (multiple exposures stemming from a single transmission chain within a group) from "loop effects" (multiple exposures arising from separate transmission chains intersecting at a group). This paper addresses the need to identify which structural mechanism—groups or long cycles—is truly responsible for the defining feature of complex contagions: the discontinuous phase transition in the size of global cascades.

Methodology
The authors employ a self-consistent branching process approach on Newman-Miller networks, a class of networks composed of independent edges and triangles (3-cliques) with a joint degree distribution. The study extends percolation theory to complex contagions by tracking individual exposures to distinguish between different higher-order effects.

Network Model: The network is parametrized by a joint probability distribution $p_{s,t}$ of edge-degree ( $s$ ) and triangle-degree ( $t$ ). The authors utilize a doubly-Poisson distribution for analytical tractability.
Simple Contagion Baseline: The authors first establish a baseline using bond percolation where transmission probability $T$ is constant. They derive self-consistent equations for the probability $u$ (edge not leading to the giant component) and $v$ (triangle not leading to the giant component) to calculate the giant component size $S$ and existence probability $R$ .
Complex Contagion Extension: The model is extended to complex contagion where the transmission probability increases upon a second exposure. Crucially, the authors introduce two distinct parameters to separate mechanisms:
- $T_\Delta$ : The increased transmission probability when a second active node in a triangle was activated within the same triangle (group effect).
- $T_\curvearrowright$ : The increased transmission probability when a second active node in a triangle was activated via a separate external transmission chain (loop effect).
Directional Percolation: To accurately calculate the size of the giant component in complex contagion, the authors distinguish between "outward" percolation (calculating the probability of a giant component existing) and "inward" percolation (calculating the size of the component). They define $u_{in}$ and $v_{in}$ to track the probability of a node not being connected to the giant component when transmission flows from neighbors to the node.
Simulation Validation: Theoretical predictions are validated against Monte Carlo simulations on networks with up to 2,000,000 nodes, analyzing the coefficient of variation (CoV) and finite-size scaling of the critical point $T_{peak}$ .

Key Contributions

Decoupling Mechanisms: The primary methodological contribution is the derivation of a self-consistent solution (Eq. 14) that explicitly separates the impact of intra-group transmission chains from inter-group loops. This allows for the isolation of the variable $T_\curvearrowright$ (loop effect) from $T_\Delta$ (group effect).
Directional Analysis: The paper provides a framework for calculating the size of the giant component in complex contagion by considering the directionality of percolation, a necessity arising from the breaking of symmetry between component size and existence probability in complex contagion models.
Identification of the Critical Mechanism: The study identifies that the discontinuous phase transition is not an inherent property of groups alone, but specifically requires the presence of long cycles where separate transmission chains intersect.

Results
The authors compare three scenarios using the parameters $\alpha_\Delta$ (scaling for $T_\Delta$ ) and $\alpha_\curvearrowright$ (scaling for $T_\curvearrowright$ ):

Groups and Loops ( $\alpha_\Delta = 10, \alpha_\curvearrowright = 10$ ): Both mechanisms are active. The system exhibits a discontinuous phase transition in the size of the giant component ( $S$ ), characterized by a jump in size and a hysteresis loop (stable and unstable equilibria).
Groups, No Loops ( $\alpha_\Delta = 10, \alpha_\curvearrowright = 1$ ): Only intra-group reinforcement is active. The system exhibits a continuous phase transition. While the size and probability of the giant component increase, the nature of the transition remains continuous, similar to simple contagion.
Loops, No Groups ( $\alpha_\Delta = 1, \alpha_\curvearrowright = 10$ ): Only inter-chain loop reinforcement is active. The system exhibits a discontinuous phase transition in the size of the giant component, matching the behavior of traditional complex contagion models.

Furthermore, finite-size scaling analysis of the Coefficient of Variation (CoV) confirms these findings. Cases with $\alpha_\curvearrowright = 1$ show CoV increasing with system size (characteristic of continuous transitions), while cases with $\alpha_\curvearrowright = 10$ show CoV saturation and a critical point $T_{peak}$ that scales with $1/\log(N)$ , supporting the critical role of cycle length.

Significance and Claims
The paper argues that the common claim—that the mere existence of groups causes discontinuous phase transitions in higher-order networks—is an oversimplification. The authors conclude that long cycles (specifically, the intersection of separate transmission chains) are the essential mechanism for discontinuous phase transitions in complex contagions. Group effects alone, without the presence of these loops, produce continuous phase transitions.

This finding refines the understanding of higher-order network dynamics, suggesting that features of hypergraphs which hinder or promote cycles (such as the overlap between hyperedges) will correspondingly hinder or promote discontinuous transitions. The work provides a more nuanced mathematical toolbox for distinguishing between structural motifs in spreading processes, moving beyond the assumption that group dynamics are solely responsible for abrupt phase changes.