🔬 physics

Reducibility of higher-order to pairwise interactions: Social impact models on hypergraphs

This paper demonstrates that a general class of higher-order social impact models on hypergraphs can be exactly reduced to equivalent pairwise interaction models on weighted projected networks, enabling accurate macroscopic predictions of ordering dynamics for both linear and nonlinear voter models.

Original authors: Jaume Llabrés, Raúl Toral, Maxi San Miguel, Federico Vázquez

Published 2026-06-24

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Jaume Llabrés, Raúl Toral, Maxi San Miguel, Federico Vázquez

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

Imagine a crowded room where people are trying to decide between two opinions: "Team Red" or "Team Blue." Usually, we think of how people change their minds as a simple one-on-one conversation. If you talk to a friend who disagrees with you, you might flip your opinion. This is what scientists call a pairwise interaction.

But in real life, influence often happens in groups. You might be sitting at a table with five friends, and if four of them suddenly switch to Team Blue, you might feel pressured to switch too, even if you never had a private conversation with any single one of them. This is a higher-order interaction.

This paper tackles a big question: Can we simplify these complex group dynamics into simple one-on-one conversations without losing the truth?

The Big Discovery: The "Shadow Network"

The authors say yes, but with a twist. They proved that any model where people change their minds based on group pressure (on a "hypergraph," which is just a fancy map of groups) can be mathematically transformed into a model where people only talk to one person at a time (on a standard network).

However, to make this work, the "friendships" in this new, simplified network can't be equal. Some friendships must be weighted.

The Analogy of the "Weighted Friendship":
Imagine you are at a party.

In the real group scenario: You are influenced by a whole table of 5 people.
In the simplified scenario: You are only talking to one person, let's call him Bob.
The Twist: To make Bob's influence feel exactly like the influence of that whole table of 5, Bob's "voice" in your head has to be louder or more frequent. If the table had 5 people, maybe Bob counts as 5 conversations. If the table had 3 people, Bob counts as 3.

The paper shows exactly how to calculate these "weights" (how loud Bob's voice should be) so that the math of the group conversation is perfectly identical to the math of the one-on-one conversation.

Two Types of "Social Pressure"

The paper looks at two specific ways people change their minds, and the results are different for each:

1. The "Fair" Rule (Linear Voter Model)

Imagine a rule where your chance of changing your mind is directly proportional to how many people in the group disagree with you. If 50% of the group disagrees, you have a 50% chance of switching.

The Result: In this case, the "weights" of the friendships in our simplified network are static. They never change. They depend only on the structure of the groups (how many people are in the groups), not on who currently holds which opinion.
The Surprise: Because the weights are constant, the complex group behavior turns out to be exactly the same as if everyone were just randomly picking a friend to copy, regardless of the group sizes. The messy group dynamics collapse into a simple, standard "copy your neighbor" game. The specific details of the groups don't matter for the big picture; only the number of connections matters.

2. The "Intense" Rule (Nonlinear Voter Model)

Now, imagine a rule where the pressure is stronger. Maybe if 50% of the group disagrees, you don't just have a 50% chance of switching; maybe you have an 80% chance because the pressure feels overwhelming. This is a "nonlinear" rule.

The Result: Here, the "weights" of the friendships are dynamic. They change every second depending on who currently holds which opinion. If the group is mostly Red, the "Blue" friends in the simplified network get a louder voice to try to sway you.
The Good News: Even though the weights are constantly shifting and getting complicated, the authors found that if you just use a standard "copy your neighbor" model (ignoring the shifting weights), it still does a very good job of predicting the overall outcome. It's not a perfect 1-to-1 match like the first case, but it captures the main trends, especially in well-connected groups.

Why This Matters (According to the Paper)

The paper doesn't talk about curing diseases or predicting elections. It stays strictly in the realm of mathematical physics.

The main takeaway is that complexity can be reduced. You don't always need a super-computer to simulate a room full of people arguing in groups. You can often simulate a simpler network of one-on-one conversations, provided you adjust the "strength" of those connections correctly.

If the social pressure is "fair" (linear), the simplification is perfect and the connection strengths are fixed.
If the social pressure is "intense" (nonlinear), the simplification is an excellent approximation, even though the connection strengths are constantly shifting.

In short, the authors built a mathematical "translator" that turns complex group dynamics into a language of simple pairs, proving that sometimes, the whole is mathematically equivalent to the sum of its parts, as long as you know how to weigh those parts.

Technical Summary: Reducibility of Higher-Order to Pairwise Interactions in Social Impact Models on Hypergraphs

Problem Statement
The paper addresses the fundamental question of whether higher-order (HO) interactions, naturally encoded in hypergraphs and simplicial complexes, can be reduced to effective pairwise (PW) interactions on standard complex networks. While many biological and social systems involve polyadic interactions (groups of three or more entities), it remains unclear when a hypergraph representation is strictly necessary versus when HO effects can be captured by dyadic interactions. Specifically, the authors investigate whether node-update social impact models defined on hypergraphs can be exactly mapped to an equivalent model with pairwise interactions on a weighted projected network, preserving the microscopic transition probabilities.

Methodology
The authors develop a rigorous mapping framework for a general class of social impact models on hypergraphs.

Model Definition: They consider a population of $N$ nodes with binary states ( $s_i = \pm 1$ ). The dynamics follow a node-update process where a focal node $i$ selects an incident hyperedge $e$ and flips its state with a probability $f(\phi_e)$ , where $\phi_e$ is the fraction of nodes in $e$ (excluding $i$ ) holding the opposite state. The function $f$ represents the social impact.
Microscopic Mapping: The core methodological contribution is the derivation of a weighted projected network $G(V, L)$ $G (V, L)$ that is microscopically equivalent to the original hypergraph dynamics. The authors impose two conditions to determine the weights $\omega_{ij}$ $ω_{ij}$ of the projected network:
- Probability Conservation: The probability of selecting a neighbor within a specific hyperedge must match the probability of selecting that hyperedge in the HO dynamics.
- Flip Consistency: The probability of a node flipping its state by copying a neighbor in the opposite state must match the HO flipping probability.
  This yields a set of weights $\omega_{ij}$ that depend on the hypergraph structure and, generally, on the instantaneous configuration of node states.
Specific Cases: The framework is applied to "hypergraph-voter models" where the social impact function is a power law, $f(\phi) = \phi^q$ $f (ϕ) = ϕ^{q}$ . The authors analyze two regimes:
- Linear Case ( $q=1$ ): The hypergraph-linear voter model.
- Nonlinear Case ( $q \neq 1$ ): The hypergraph-nonlinear voter model.
Analysis: For homogeneous hypergraphs (fully connected, complete $d$ -hypergraphs, Erdős–Rényi, and $z$ -regular random hypergraphs), the authors derive macroscopic evolution equations. They employ a pair approximation (PA) for the linear case to analyze the dynamics on the projected network and compare analytical predictions with numerical simulations of the HO, reduced PW, and standard voter/nonlinear voter models.

Key Contributions and Results

Exact Microscopic Reducibility: The primary result is that any node-update social impact model on an arbitrary hypergraph can be exactly mapped to a pairwise interaction model on a weighted projected network. The mapping preserves the microscopic probabilities of state changes. The weights $\omega_{ij}$ encode the influence of higher-order structures; in general, these weights are state-dependent (time-dependent).
Linear Voter Model ( $q=1$ ):
- For the linear case, the derived weights on the projected network are static (state-independent). They depend only on topological properties (hyperdegrees and hyperedge orders).
- Using a pair approximation for uncorrelated weighted networks, the authors demonstrate that the macroscopic ordering dynamics (density of active links and fixation time) is independent of the specific heterogeneity of the weights.
- Consequently, the macroscopic behavior of the hypergraph-linear voter model is equivalent to that of the standard voter model on the unweighted projected network. The complex higher-order topology is effectively captured by the degree distribution of the projected network.
Nonlinear Voter Model ( $q \neq 1$ ):
- In the nonlinear regime, the weights of the projected network depend explicitly on the instantaneous system configuration, making them dynamic.
- Despite the complexity of the exact weights, numerical simulations show that the nonlinear voter model (NLVM) on the unweighted projected network (where weights are uniform) provides a good approximation of the macroscopic trends (specifically the plateau value of active link density) for well-connected hypergraphs.
- The NLVM on the projected network successfully reproduces the main ordering dynamics, suggesting it serves as an effective minimal model for empirical data analysis where microscopic weights are unknown.

Significance and Claims
The paper claims to establish a "microscopic reducibility" of higher-order interactions to pairwise ones. This finding implies that for node-update dynamics, the explicit representation of hyperedges is not strictly necessary to reproduce the stochastic evolution of the system; a weighted projected network suffices.

The authors emphasize that the linearity of the social impact function is the critical factor determining the nature of this reduction:

In the linear case, the reduction is robust and leads to a static, time-independent weighted network where macroscopic dynamics are insensitive to weight heterogeneity.
In the nonlinear case, while an exact microscopic mapping exists with dynamic weights, the system can still be effectively approximated by a simpler, unweighted nonlinear voter model on the projected network for macroscopic observables.

The work concludes that while higher-order interactions alter structural properties, their dynamical effects in these specific social impact models can be captured by effective pairwise interactions, provided the appropriate (potentially dynamic) weights are used or, in the nonlinear case, a suitable minimal model is selected. The paper notes that this exact mapping applies specifically to node-update dynamics; hyperedge-update dynamics (where all nodes in a group update simultaneously) cannot be mapped exactly to a pairwise system, though macroscopic equivalence may still exist in specific limits.

The Big Discovery: The "Shadow Network"

Two Types of "Social Pressure"

1. The "Fair" Rule (Linear Voter Model)

2. The "Intense" Rule (Nonlinear Voter Model)

Why This Matters (According to the Paper)

More like this