🔬 physics

An exact N-strain epidemic model using bond percolation

This paper presents an analytically exact N-strain epidemic model using bond percolation on Erdos-Renyi and scale-free random networks, introducing competitive and collaborative sequential branching processes to investigate emergent graph structures and disease spreading dynamics via generating functions.

Original authors: Peter Mann, V. Anne Smith, John B. O. Mitchell, Simon Dobson

Published 2026-07-20

📖 8 min read🧠 Deep dive

CC BY 4.0

Original authors: Peter Mann, V. Anne Smith, John B. O. Mitchell, Simon Dobson

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 bustling city where everyone is connected by invisible threads, forming a giant, tangled web. In the world of science, this web is called a network, and the threads are the connections between people, computers, or even cells. Sometimes, something spreads through this web—like a rumor, a computer virus, or a disease. Scientists use a clever trick called bond percolation to study how this spreading works. Think of it like a game of "connect the dots" where you only get to draw a line between two dots if you roll a lucky number. If you roll high enough often enough, the dots suddenly link up into one massive, city-spanning cluster. If you roll low, you just get tiny, isolated islands of dots. This simple game helps us understand when a small spark becomes a massive fire.

Now, imagine that instead of just one fire, we have a whole series of them, one after another. Maybe it's a flu season that changes every year, or a new strain of a virus that shows up after the last one fades. The big question is: How does the first fire change the city for the second one? Does it leave the city in ruins, making it hard for the next fire to spread? Or does it somehow clear the way? This is the puzzle that a team of researchers at the University of St Andrews set out to solve. They wanted to see what happens when you run this "connect the dots" game not just once, but an arbitrary number of times, and how the shape of the city changes with every new round.

The Great Game of "Who Gets Infected?"

The paper introduces two very different ways these "seasonal diseases" can play out, which the authors call Competitive and Collaborative branching processes. To understand them, let's imagine a massive party where the guests are the nodes (people) and the handshakes are the edges (connections).

The Competitive Branching Process: The "Cross-Immunity" Game

In this scenario, imagine a group of partygoers gets infected by "Flu A." Once they recover, they become immune to Flu A, but more importantly, they are immune to everything else that comes next. They are effectively removed from the game; they can't catch the next bug, and they can't pass it on.

The researchers found that when a new strain (Flu B) arrives, it can only spread among the people who didn't catch Flu A. This leaves a smaller, more broken-up group of people. The authors call this the Residual Graph (RG). It's like the party has been cleared out, leaving only the people who stayed in the corners.

Here is the twist: As more and more strains arrive, the "party" gets smaller and more fragmented. The authors discovered that for a new strain to cause a big outbreak, it needs to be more contagious than the one before it. Why? Because the easy-to-reach people (the ones with lots of friends) were likely already taken by the first few strains. The remaining people are the ones with fewer connections, making it harder for the virus to jump from person to person.

In their simulations, they showed that if the first strain is too strong, it might actually "kill" the party for everyone else. It breaks the network so badly that even if a super-contagious strain arrives later, it can't find enough people to connect with to start a big outbreak. It's like if the first fire burns down all the bridges; the second fire, no matter how hot, can't cross the river.

The Collaborative Branching Process: The "Coinfection" Game

Now, let's flip the script. In this version, imagine that to catch "Flu B," you must have already caught "Flu A." You can't get the new bug unless you are already infected by the old one. This is like a video game where you have to beat Level 1 to unlock Level 2.

This creates a very different kind of shrinking. Instead of removing people, the virus is forced to travel only through the people who have already been infected by the previous rounds. The authors call this the Giant Connected Component (GCC). It's like the virus is trapped in a shrinking bubble.

The researchers found that in this game, the bubble gets smaller and smaller with every new strain. The network of people who have caught all the previous strains becomes a tiny, fractured island. Even if the new strain is super contagious, it can't spread far because there simply aren't enough people left in the "bubble" to pass it to. The authors showed that the outbreak size for each new strain is always smaller than the one before it. It's a game of diminishing returns where the virus eventually runs out of fuel.

What Did They Actually Find?

The team used a powerful mathematical tool called generating functions (think of it as a super-advanced calculator that can handle infinite possibilities at once) to predict exactly how big these outbreaks would be. They didn't just guess; they ran thousands of computer simulations on networks that looked like random social webs (called Erdős-Rényi graphs) and compared their math to the simulation results. The math and the simulations matched perfectly.

Here are the key takeaways from their work:

The "Harder to Reach" Rule: In the competitive game, every new strain faces a higher barrier to entry. If the first strain infects 50% of the network, the second strain needs to be significantly more contagious just to infect the same number of people. The authors proved that the "critical threshold" (the minimum contagiousness needed to start a big outbreak) goes up with every single generation.
The "Total Infection" Surprise: You might think that if you have five different diseases, they would eventually infect everyone. But the authors found something counter-intuitive. If the first disease is too strong, it might actually reduce the total number of people infected in the long run. This happens because the first disease burns out the "highly connected" people (the social butterflies), leaving behind a network of isolated people that the later diseases can't reach. The total number of infected people can actually dip before rising again.
The "Coexistence" Limit: There is a sweet spot where multiple strains can coexist. If the first strain is too weak, the second one takes over easily. If it's too strong, it destroys the network for everyone else. But there's a middle ground where the network is just fractured enough to let the next strain in, but not so broken that it stops the game entirely. The authors mapped out exactly where this line is.
The "Shrinking Bubble" in Collaboration: In the coinfection game, the authors showed that the network of eligible hosts shrinks exponentially. The more strains you add, the smaller the group of people who have caught all of them becomes. Eventually, the network becomes so fractured that the virus can't spread at all, no matter how contagious it is.

Why Should We Care?

While this paper is about math and abstract networks, the story it tells is very real. It helps us understand how diseases evolve. If a virus mutates and becomes a new strain, does it need to become super-contagious to survive? The answer depends on how the previous strains changed the population.

The authors suggest that in a world of seasonal diseases, we shouldn't just look at how contagious a new virus is. We also need to look at how the previous viruses changed the social fabric. Did the last flu season leave us with a population that is hard to reach? Or did it leave us with a population that is easy to infect?

The paper doesn't claim to have a cure for the flu, nor does it predict the next pandemic. Instead, it gives us a new way to look at the invisible web of connections that holds our society together. It shows us that the history of an outbreak matters just as much as the virus itself. Whether the next wave of disease will be a massive wave or a tiny ripple might depend entirely on how the waves before it reshaped the shore.

In the end, the authors have built a precise map of how these "seasonal" diseases interact with the structure of our world. They showed that in the game of infection, the board changes with every move, and the rules for winning get harder with every round. Whether the game is a competition for hosts or a collaboration of infections, the result is the same: the network is a living, breathing thing that remembers every step of the journey.

Technical Summary: An Exact N-strain Epidemic Model using Bond Percolation

Problem Statement
This paper addresses the need to extend the established equivalence between bond percolation and the Susceptible-Infected-Removed (SIR) epidemic process from single-strain or two-strain models to an arbitrary, finite number ( $N$ ) of sequential strains. While previous work has examined two-strain scenarios involving either perfect cross-immunity (where the second strain spreads on the residual graph of the first) or perfect coinfection (where the second strain spreads on the giant connected component of the first), this study generalizes these concepts to $N$ sequential generations. The authors aim to analytically determine the emergent topological structures, expected outbreak sizes, and critical thresholds for these sequential processes on random networks, specifically Erdős-Rényi and scale-free graphs.

Methodology
The authors employ the method of generating functions, pioneered by Newman, Strogatz, and Watts, to derive exact analytical solutions for the system. The core of the methodology involves defining two distinct sequential branching processes:

Competitive Branching Process (Cross-Immunity Model): In this scenario, each subsequent strain ( $i$ ) performs bond percolation on the Residual Graph (RG) left by the previous $i-1$ strains. Nodes infected by previous strains are removed from the network for subsequent generations. The authors utilize the cavity method to track the probability that a neighbor is not part of the current Giant Connected Component (GCC), accounting for the history of previous percolations. They derive a hierarchy of self-consistent equations to solve for the probability $u_i$ that a neighbor is not infected by the $i$ -th strain, given it was not infected by previous strains.
Collaborative Branching Process (Coinfection Model): In this scenario, each subsequent strain performs bond percolation exclusively on the GCC formed by the previous strains. Infection by the current strain requires prior infection by all previous strains. This process creates a complex state space where the number of possible neighbor infection histories grows exponentially ( $2^{i-1}$ for the $i$ -th generation). The authors model this using a perfect binary tree structure of infection histories, deriving recursive expressions for the probability of specific infection histories and the prior probabilities of nodes occupying those states.

The analysis focuses on calculating the size of the GCC ( $A_i$ ) for each generation, the critical transmissibility ( $T_{i,c}$ ) required to sustain an epidemic, and the coexistence threshold ( $T_{i,x}$ ) which defines the maximum transmissibility of a previous strain that still allows the RG (or GCC) to support a future epidemic.

Key Contributions and Results

Analytical Framework for N-Strains: The paper provides a generalizable analytical framework using generating functions to solve for $N$ sequential strains, moving beyond the specific $N=2$ cases studied in prior literature.
Competitive Process Dynamics:
- Outbreak Size: The authors derive exact expressions for the outbreak size of the $i$ -th strain ( $A_i$ ) as a function of the transmissibilities of all preceding strains.
- Increasing Epidemic Thresholds: A key finding is that the critical transmissibility ( $T_{i,c}$ ) required for a strain to form a GCC increases with each generation ( $T_{i,c} \geq T_{i-1,c}$ ). This is because high-degree nodes are preferentially removed into the GCCs of earlier strains, leaving the RG populated by lower-degree nodes that are harder to infect.
- Coexistence Threshold: The study extends the definition of the coexistence threshold ( $T_{i,x}$ ) to $N$ strains. It is shown that as the number of generations increases, the gap between the epidemic threshold and the coexistence threshold widens, meaning the phase space where strains can coexist expands.
- Total Infected Fraction: Counter-intuitively, the total fraction of the network infected is not monotonic with the transmissibility of the first strain. Low transmissibility in early strains can lead to a local minimum in total infections because it removes high-degree nodes without forming a large GCC, thereby isolating the nodes they connected to and preventing subsequent strains from reaching them.
Collaborative Process Dynamics:
- Exponential State Complexity: The authors demonstrate that the number of distinct infection histories required to describe the system doubles with each generation ( $2^{i-1}$ new states).
- Fracturing Topology: In the collaborative model, the available network for each subsequent strain is the GCC of the previous one. The results show that the average degree of the remaining substructure decreases with each generation, and the degree distribution becomes increasingly skewed toward lower degrees. Consequently, the outbreak size ( $A_i$ ) monotonically decreases with the generation index, eventually leading to the burnout of the epidemic due to the fractured topology of the substrate.
Validation: The analytical results are validated against Monte Carlo simulations on Erdős-Rényi networks ( $V=35,000$ ), showing excellent agreement for both outbreak sizes and critical thresholds.

Significance and Claims
The paper claims to provide an "exact" analytical method for understanding the emergent structures of random networks undergoing sequential bond percolation. The authors assert that their model offers a rigorous interpretation of seasonal $N$ -strain disease spreading, where the interaction between strains (either competitive or collaborative) fundamentally alters the network's topology and the potential for future outbreaks.

Specifically, the work highlights that:

Network Structure Evolution: Sequential percolation processes do not merely reduce the network size; they qualitatively change the degree distribution and connectivity of the remaining nodes, making subsequent epidemics increasingly difficult to sustain (in the competitive model) or limiting the spread due to topological fracturing (in the collaborative model).
Critical Thresholds: The epidemic threshold is not a static property of the network but a dynamic function of the history of previous infections.
Modeling Utility: The competitive model serves as a tool to study the structure of the Residual Graph, while the collaborative model provides insight into the structure of the Giant Connected Component under repeated constraints.

The authors maintain a modest scope, focusing on the theoretical derivation and validation of these properties on random graphs, without proposing specific real-world disease interventions or future experimental proposals beyond the mathematical framework presented.

The Great Game of "Who Gets Infected?"

The Competitive Branching Process: The "Cross-Immunity" Game

The Collaborative Branching Process: The "Coinfection" Game

What Did They Actually Find?

Why Should We Care?

More like this