🔬 physics

Partisan voter model on complex networks: Dynamics of local ordering

This paper investigates the dynamics of local ordering in the partisan voter model on complex networks, demonstrating that while partisan bias redistributes active links without altering their total density on uncorrelated networks, preference-dependent structural correlations qualitatively modify the stationary state by creating distinct regimes of local ordering.

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

Published 2026-06-04

📖 4 min read☕ Coffee break read

CC BY 4.0

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

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 large town square where everyone holds one of two signs: a Red Flag or a Blue Flag. This is a classic model of how opinions spread, known as the "Voter Model." Usually, people just look at their neighbors and copy whoever they see most often. If they see more Reds, they might switch to Red. It's a game of pure imitation.

But in real life, people aren't blank slates. They have fixed preferences. Maybe you were born into a family that loves Red, or you just really want to be Red, even if your neighbors are Blue. This paper studies what happens when people have these stubborn internal biases while trying to fit in with their neighbors.

The researchers, Jaume Llabrés, Maxi San Miguel, and Raúl Toral, looked at how this plays out in two different types of "towns" (networks).

1. The Random Town (Uncorrelated Networks)

First, they imagined a town where everyone is connected to random neighbors, like a giant, messy web of friendships where who you know has nothing to do with your political preference.

The Setup: Everyone has a fixed "bias" (let's say, a 20% or 50% chance they prefer to be Red).
The Discovery: Even though people are biased, the total amount of arguing in the town stays exactly the same as if they had no bias at all.
- The Analogy: Imagine a dance floor. If everyone is just copying the person next to them, there's a certain amount of "mismatched dancing" (people holding different flags). Adding a personal preference doesn't stop the dancing or make the floor more chaotic overall. It just changes who is dancing with whom.
The Shift: While the total noise stays the same, the type of noise changes.
- If you are biased toward Red, you are more likely to be a "satisfied Red" (holding a Red flag and liking it).
- The paper found that as the bias gets stronger, the town becomes full of "satisfied" people (Reds who like being Red, Blues who like being Blue) and fewer "confused" people (Reds who are holding Blue flags because they copied a neighbor).
- Key Takeaway: In a random crowd, your personal bias doesn't change the overall level of disagreement; it just rearranges who is disagreeing with whom.

2. The Clumpy Town (Preference-Based Networks)

Next, they looked at a more realistic scenario: Homophily. This is the "birds of a feather flock together" rule. In this town, people are more likely to make friends with others who share their preference, not just their current opinion.

The Setup: People with a Red preference are more likely to connect to other Red-preferrers. People with a Blue preference connect to Blue-preferrers.
The Discovery: Here, the rules change completely. The structure of the town itself changes the outcome.
- The "Heterophilic" Zone (Opposites Attract): If people with opposite preferences are forced to hang out together (like a forced integration), and they also have strong internal biases, the town becomes a chaotic mess. Everyone is arguing constantly because they are surrounded by people they disagree with, but they refuse to change their minds. The "active links" (arguments) skyrocket.
- The "Homophilic" Zone (Like Attracts Like): If people mostly hang out with their own kind, the town splits into quiet, segregated neighborhoods. The Red-preferrers form a quiet Red village, and the Blue-preferrers form a quiet Blue village. Inside these villages, everyone agrees. The total amount of arguing drops to almost zero.
Key Takeaway: When your social circle is built on your biases, those biases can either supercharge the chaos (if you are forced to mix with opposites) or silence the noise (if you only hang out with your own kind).

The Big Picture

The paper uses a clever mathematical tool called a "pair approximation" (think of it as a way to count not just who is holding a flag, but who is holding a flag next to whom) to prove these points.

They found that you cannot just look at the "total amount of disagreement" in a society. You have to look at the composition of that disagreement:

Are the people arguing with each other because they have different preferences?
Are they arguing because they are "unsatisfied" (holding a flag they don't like)?

The Final Lesson:
In a random world, your personal stubbornness just reshuffles the deck; the game remains the same. But in a world where we choose our friends based on our preferences, that stubbornness can either tear the community apart into constant conflict or build walls that create peaceful, isolated bubbles. The paper shows that how we connect is just as important as what we believe when it comes to how a society settles down.

Technical Summary: Partisan Voter Model on Complex Networks: Dynamics of Local Ordering

Problem Statement
The standard voter model (VM) is a foundational framework in the statistical physics of social dynamics, describing how local interactions lead to global consensus. However, the standard VM assumes dynamical neutrality, where agents have no intrinsic preference for either state. This limitation restricts its applicability to social and political contexts where individual responses are biased by persistent predispositions or partisan identities. The "partisan voter model" (PVM) addresses this by assigning each agent a fixed binary preference that biases state updates toward alignment with that preference. While previous work established that this bias breaks the conservation of magnetization and induces stable coexistence in the mean-field limit, the interplay between this dynamical bias and complex network structures remains to be fully characterized. Specifically, it is unclear how fixed preferences interact with sparse, heterogeneous connectivity and structural correlations (such as homophily) to determine local ordering.

Methodology
The authors investigate the PVM on complex networks using a combination of analytical pair approximation and numerical Monte Carlo simulations. The study is divided into two main scenarios:

Uncorrelated Random Networks: The authors analyze the PVM on Erdős–Rényi networks where preferences bias the update rule but do not influence network topology. To move beyond mean-field theory, they develop a detailed pair approximation. Unlike the standard VM, where a single density of active links (links connecting nodes in different states) suffices, the PVM requires decomposing the total density of active links ( $\rho$ ) into four distinct classes based on the preferences and satisfaction levels of the connected agents:
- $\rho_+$ : Links between agents with preference $p=+1$ .
- $\rho_-$ : Links between agents with preference $p=-1$ .
- $\rho_s$ : Links between "satisfied" agents (state aligned with preference).
- $\rho_u$ : Links between "unsatisfied" agents (state unaligned with preference).
  The authors derive a closed system of differential equations (Eqs. 20a–20f) governing the time evolution of these densities and global variables (imbalance $\Delta$ and satisfaction density $\Sigma$ ).
Preference-Dependent Networks: The authors extend the analysis to networks generated via a Barabási–Albert homophily model. In this setting, the same fixed preferences that bias the dynamics also determine the network structure. A parameter $h$ controls the tendency to form links with agents of the same preference ( $h > 1/2$ , homophilic) or opposite preference ( $h < 1/2$ , heterophilic). For this case, where structural correlations are strong, the analytical pair approximation is insufficient, and the authors rely on numerical simulations to characterize the stationary state.

Key Contributions and Results

Decomposition of Local Order: The primary analytical contribution is the derivation of a pair approximation that resolves the densities of the four classes of active links. The authors demonstrate that in the PVM, active links are not dynamically equivalent; their evolution depends critically on the preferences of the connected nodes.
Dynamics on Uncorrelated Networks:
- The analytical predictions show excellent agreement with Monte Carlo simulations on Erdős–Rényi networks.
- A key finding is that while the partisan bias ( $\epsilon$ ) redistributes the active links among the four classes (increasing $\rho_s$ and decreasing others as $\epsilon$ increases), the total stationary density of active links ( $\rho_{st}$ ) remains unchanged and identical to that of the standard voter model on the same network.
- The total density is given by $\rho_{st} = \xi = \frac{\mu-2}{2(\mu-1)}$ , where $\mu$ is the average degree. The bias merely reorganizes local disorder without altering the global level of activity.
- The detailed pair approximation is shown to be superior to a reduced scheme (which tracks only total $\rho$ ) in capturing transient dynamics, particularly the non-monotonic relaxation observed at high bias strengths.
Dynamics on Preference-Dependent Networks:
- When network structure is coupled to preferences, the total stationary density of active links becomes dependent on both the bias strength $\epsilon$ and the homophily parameter $h$ .
- The authors identify three distinct regimes in the $(h, \epsilon)$ $(h, ϵ)$ parameter space:
  1. Heterophilic/High Bias: High disorder with large active link densities due to competition between structural heterophily and strong individual bias.
  2. Intermediate Regime: Moderate local ordering with clusters of intermediate size.
  3. Homophilic/High Bias: Strong local ordering and low active link densities, leading to fragmented, segregated communities where internal consensus is easily reached.
- In this regime, increasing $\epsilon$ has opposite effects depending on the structural context: it enhances activity in heterophilic networks but suppresses it in homophilic networks.

Significance and Claims
The paper claims to clarify the distinct roles of dynamical partisan bias and structural assortativity in opinion dynamics. The authors argue that in binary-state dynamics with quenched agent types, the total density of active links is insufficient to characterize local order; a finer description resolving the classes of active links is necessary.

The work provides an analytical framework to study partisan opinion dynamics beyond mean-field theory, demonstrating that:

On uncorrelated networks, dynamical bias acts as a redistributor of local disorder without changing the total amount of disorder.
On correlated networks, the interplay between dynamical bias and structural homophily qualitatively modifies the stationary state, allowing network organization to either enhance or suppress local activity.
The separation of these mechanisms is crucial for understanding how fixed preferences interact with social connectivity to produce collective behaviors, offering a more nuanced view than models that treat bias and structure as a single entangled effect.

1. The Random Town (Uncorrelated Networks)

2. The Clumpy Town (Preference-Based Networks)

The Big Picture

More like this