🧬 biology

Stochastic SIS Model with Dynamic Node Activity: Epidemic Threshold and Stationary States

This paper extends the classical stochastic SIS model to networks with dynamic node activity, demonstrating that while activity rates do not alter the epidemic threshold or stationary infection levels, they significantly slow down transient dynamics by allowing temporarily disabled infected nodes to act as a reservoir that sustains the infection.

Original authors: Alexey V. Nikiforov, Alexey A. Magazev

Published 2026-10-29

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Alexey V. Nikiforov, Alexey A. Magazev

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

In the digital world, networks are the invisible scaffolding that holds our modern life together. From the computers in an office building to the devices in a smart home, these systems rely on constant connections to share information. However, just as a virus can spread through a crowd of people, malicious software can travel through these digital connections, jumping from one device to another. Scientists have long used mathematical models to predict how such infections behave, treating computers like people who can be either healthy or sick. A classic model, known as SIS, assumes that once a computer is infected, it can recover and become healthy again, only to potentially catch the virus later. This framework helps researchers understand when an infection will die out on its own and when it will become a permanent, lingering problem. But these traditional models often assume that every computer is always online and ready to interact, a condition that rarely matches reality. In the real world, devices go offline for updates, sleep mode, or simply because they are turned off, creating a dynamic where the network's ability to spread information changes constantly.

A team of researchers from Russia set out to explore what happens when this constant connectivity is broken. They developed a new way to simulate how a computer virus spreads when the devices themselves are not always available to participate. In their model, every computer has two distinct states: it is either "enabled," meaning it is online and ready to communicate, or "disabled," meaning it is offline and silent. Crucially, a disabled computer cannot catch a virus, nor can it pass one on, but it also cannot recover if it is already infected. It simply waits in a state of suspended animation until it comes back online. The researchers wanted to know if this constant switching between being online and offline would change the fundamental rules of how an epidemic spreads, specifically whether it would raise or lower the threshold needed for an infection to take hold and become permanent.

To investigate this, the team built a computer simulation involving a network of one thousand nodes, representing individual devices. They created a virtual environment where these devices were connected in a random pattern, similar to how computers might be linked in a typical office network. They then introduced a virus and watched how it moved through the system over time. The simulation allowed them to adjust the "activity rate," which determined how frequently devices switched between being online and offline. They tested scenarios where the virus was weak and likely to die out, as well as scenarios where the virus was strong and likely to become a permanent fixture in the network. By running thousands of these simulations, they could observe the average behavior of the infection and compare their findings against their mathematical predictions.

The results revealed a surprising stability in the core rules of the epidemic. The researchers found that the critical threshold for an infection to become a permanent problem did not change, regardless of how often the devices went offline. Whether the computers were constantly switching states or staying put, the point at which the virus became unstoppable remained exactly the same as in the classic models. If the virus was strong enough to spread in a static network, it would also spread in this dynamic one. The activity of the nodes did not make the network more or less vulnerable to the initial takeover; the fundamental math of the outbreak remained unchanged.

However, while the final outcome was the same, the journey to get there was dramatically different. The researchers discovered that the speed at which the infection spread or died out was heavily influenced by how often the devices went offline. In a situation where the virus was weak and destined to die out, the infection disappeared much more slowly when the devices were frequently switching between online and offline states. This counterintuitive result occurs because a disabled infected computer acts as a hidden reservoir. When an infected device goes offline, it cannot recover, so it holds onto the virus while waiting to come back online. This delays the recovery process, allowing the virus to persist longer than it would if every device were always active and able to heal itself.

The same slowing effect was observed when the virus was strong and destined to become permanent. In these cases, the infection took much longer to reach its final, stable level when the activity rate was high. The constant switching created a bottleneck; infected devices spent so much time offline that they could not transmit the virus to their neighbors as quickly as they would in a static network. It was as if the infection was trying to run a race, but the runners were frequently forced to stop and wait, slowing the entire group's progress. Despite this delay, the infection eventually reached the same high level of prevalence as it would have in a network where everyone was always online. The final number of infected devices depended only on the strength of the virus and the network's structure, not on how often the devices went to sleep.

These findings suggest that while the dynamic nature of modern networks does not make them fundamentally more or less susceptible to an outbreak, it does significantly alter the timeline of the event. For those trying to manage or contain a digital infection, the activity of the devices is a crucial factor. A network where devices frequently go offline might appear to be slowing down an outbreak, but it is actually just delaying the inevitable, keeping the virus alive in a dormant state for longer periods. The study provides a clearer picture of how real-world constraints, like devices turning off, interact with the spread of digital threats, showing that the rules of the game remain the same, even if the pace of play changes.

Technical Summary: Stochastic SIS Model with Dynamic Node Activity

Problem Statement
Mathematical modeling of spreading processes, such as computer virus propagation, typically relies on compartmental models like the Susceptible-Infected-Susceptible (SIS) framework. While classical models assume static network topologies or continuous node availability, real-world systems often exhibit dynamic node activity where nodes (e.g., workstations or mobile devices) periodically switch between active (enabled) and inactive (disabled) states. In such scenarios, the underlying network topology remains fixed, but only enabled nodes participate in transmission. The central problem addressed in this paper is to determine how this dynamic node availability affects the epidemic threshold and the stationary states of infection spreading, specifically investigating whether the classical SIS threshold is preserved and how the activity rate influences transient dynamics.

Methodology
The authors propose a stochastic SIS model defined on a connected undirected graph $G = (V, L)$ with $N$ nodes. Each node possesses two independent binary attributes:

Activity State: Enabled ( $E$ ) or Disabled ( $D$ ).
Epidemiological State: Susceptible ( $S$ ) or Infected ( $I$ ).

This results in four possible node states: $(D,S), (D,I), (E,S), (E,I)$ . The dynamics are governed by discrete-time probabilistic transitions:

Activity Transitions: Nodes switch between $D$ and $E$ with probabilities $\lambda_+$ and $\lambda_-$ , independent of their infection status.
Epidemiological Transitions:
- An enabled susceptible node becomes infected with probability $\mu_a \approx \beta k_a i(t)$ , where $k_a$ is the node degree, $\beta$ is the transmission probability, and $i(t)$ is the fraction of infected nodes among the enabled population.
- An enabled infected node recovers with probability $\delta$ .
- Disabled nodes retain their epidemiological state; they cannot infect others nor recover while disabled.

The study employs two complementary approaches:

Mean-Field Approximation: A continuum analogue is derived by assuming statistical equivalence among nodes and describing the system via fractions of nodes in each state ( $s_D, i_D, s_E, i_E$ ). This yields a system of ordinary differential equations to analyze stationary states and stability.
Stochastic Simulations: Direct simulations are performed on Erdős–Rényi random graphs ( $N=1000$ ) to verify analytical predictions and quantify finite-size effects. The simulations use synchronous updates and ensemble averaging over multiple graph realizations and epidemic runs.

Key Contributions

Model Formulation: The paper introduces a stochastic SIS model that explicitly decouples node activity from epidemiological status, allowing for a rigorous analysis of systems where transmission is contingent on node availability.
Analytical Derivation: The authors derive the mean-field equations and identify the stationary states, proving the existence of a transcritical bifurcation at the epidemic threshold.
Threshold Preservation: A primary theoretical finding is that the epidemic threshold, defined by the basic reproduction number $R_0 = \beta \langle k \rangle / \delta$ , remains identical to the classical SIS result and is independent of the activity parameters ( $\lambda_+, \lambda_-$ ).
Transient Dynamics Analysis: The paper quantifies the counterintuitive effect of the activity rate on the speed of convergence. It demonstrates that increasing the activity switching rate ( $\lambda$ ) slows down both the decay of infection in the subcritical regime ( $R_0 < 1$ ) and the convergence to the endemic state in the supercritical regime ( $R_0 > 1$ ).

Results

Stationary States:
- Infection-Free State: Exists and is stable when $R_0 < 1$ .
- Endemic State: Exists and is stable when $R_0 > 1$ . The stationary fraction of infected nodes among the enabled population is $1 - 1/R_0$ .
- Independence: The stationary level of infection depends solely on $R_0$ and is independent of the activity rate $\lambda$ in the symmetric case ( $\lambda_+ = \lambda_- = \lambda$ ).
Transient Dynamics (The "Reservoir" Effect):
- Subcritical Regime ( $R_0 < 1$ ): Increasing $\lambda$ slows the exponential decay of the infection. This is explained by the fact that infected nodes in the disabled state cannot recover; they act as a "reservoir" that sustains the infection longer than in a static network.
- Supercritical Regime ( $R_0 > 1$ ): Increasing $\lambda$ slows the convergence to the stationary plateau. High activity rates cause infected nodes to spend significant time in the disabled state, where they cannot transmit the virus, leading to an initial dip in infection levels before the system stabilizes.
Simulation vs. Theory: Stochastic simulations on Erdős–Rényi graphs confirm the analytical predictions regarding the threshold and the independence of the stationary state from $\lambda$ . However, simulations show a systematic deviation (approx. 2–4% lower infection levels) from the mean-field prediction in the supercritical regime, attributed to the mean-field averaging of the infection probability over the degree distribution and finite-size effects.

Significance
The paper claims that its primary significance lies in extending the classical mean-field SIS model to the case of dynamic node availability. It establishes that while the epidemic threshold is robust and unaffected by dynamic node activity, the temporal dynamics are significantly altered. Specifically, the finding that higher activity rates can paradoxically prolong the persistence of an infection (by preventing recovery in the disabled state) offers a nuanced understanding of spreading processes in systems with intermittent connectivity, such as networks of devices that periodically go offline. The work provides a theoretical basis for understanding how the "reservoir" effect of disabled nodes influences the time scales of epidemic control and convergence.

Technical Summary: Stochastic SIS Model with Dynamic Node Activity

More like this