The Pulse of Invasion: How Immunity Loss Sculptures Epidemic Waves

The Architecture of Epidemic Waves

Imagine watching a disease spread across a landscape — not as a smooth, gradual advance, but as a jagged sequence of sudden eruptions separated by stretches of quiet, almost imperceptible change. That's what emerges from the mathematics in this paper: a new way of understanding how epidemics invade territory, built on the insight that the same mechanisms that make diseases recurring — the slow waning of immunity — also sculpt the shape of their spatial spread.

The finding, distilled to its essence: when a disease doesn't grant permanent protection, its geographic advance doesn't proceed as a uniform wavefront. Instead, it unfolds as an alternation between two modes — periods of gradual adaptation where the susceptible population slowly rebuilds, punctuated by rapid, explosive outbreaks that drive the infection forward. The rate at which immunity fades — the "slow" parameter in the model — turns out to be the master dial controlling both how long communities get to breathe between outbreaks and how far apart successive waves of infection sit from one another.

This isn't just a mathematical curiosity. It offers a framework for understanding why some diseases advance in distinct pulses rather than as a steady tide, and why the spacing between those pulses depends so intimately on the biology of immunity rather than the mechanics of transmission. It also opens a path toward predicting invasion speeds and understanding the interplay between temporal recurrence and spatial spread — two dimensions of epidemics that are often studied separately but, as this work shows, are deeply entangled.

The Science

To appreciate what the researchers did, it helps to start with a model that epidemiologists have used for nearly a century: the SIRS framework. The acronym captures the life cycle of infection. "S" stands for Susceptible — people who can catch the disease. "I" stands for Infected — those carrying and transmitting it. "R" stands for Recovered — people who've fought off the infection and temporarily gained protection. The final "S" is the twist that makes this model special: recovered individuals don't stay immune forever. They eventually lose their protection and return to the susceptible pool, setting the stage for future outbreaks. This makes SIRS the minimal model for diseases that don't confer lifelong immunity — think whooping cough, RSV, or seasonal coronaviruses.

The classical SIRS model, developed in its non-spatial form by Kermack and McKendrick in the 1920s, captures temporal dynamics: how the number of susceptibles, infected, and recovered changes over time in a well-mixed population. It answers questions like: Will an outbreak die out or persist? How many people will be infected in total? But it has a blind spot. It assumes everyone mixes uniformly with everyone else — no geography, no movement, no spatial structure.

Real epidemics don't work that way. When avian flu moves through poultry farms, it's not spreading through some homogeneous soup of birds. It's carried by infected individuals moving across the landscape, leaving trails of infection as they go. The geographic pattern of spread matters enormously for control strategies: where do you set up barriers? How do you time culling operations?

To bridge this gap, the researchers extended the SIRS model by adding a diffusion term — a mathematical way of describing how infected individuals move randomly through space, spreading the disease as they go. Specifically, they assumed diffusion acts only on the infected compartment. This is a deliberate simplification, but a reasonable one: in many outbreaks, it's the movement of infectious individuals that drives geographic advance, not the mobility of the susceptible or recovered population.

The resulting partial differential equation (PDE) system describes how the density of susceptibles and infected individuals changes over both time and space. But PDEs are notoriously difficult to solve directly. The researchers' key move was to notice that, under the right conditions, this PDE reduces to a system of ordinary differential equations (ODEs) — a massive simplification — by switching to a traveling wave coordinate.

Think of it this way: if an epidemic front moves at constant speed, you can change variables so that you're riding along with the front rather than standing still in the lab frame. In this moving reference frame, the PDE becomes an ODE — the spatial coordinate becomes the "time" variable along the orbit of the solution. This reduction is only valid for traveling fronts that move at constant speed, but the researchers' numerical simulations confirm that their model does produce exactly this kind of front.

Now comes the mathematical heart of the paper. The system exhibits a natural slow-fast structure because of a small parameter — the loss-of-immunity rate, denoted δ. In typical epidemiological settings, the infectious period is much shorter than the average duration of immunity. If you're sick for a week or two but immune for months or years, those processes operate on wildly different time scales. The mathematics of such separation — where a small parameter creates distinct fast and slow regimes — has been extensively developed in what geometricians call Geometric Singular Perturbation Theory (GSPT). The researchers deployed this entire machinery, analyzing the "layer problem" (the fast dynamics, what happens in a single outbreak), the "critical manifold" (the slow dynamics, what happens between outbreaks), and the transition between them.

What They Found

The analysis revealed a structure that is both elegant and, for the first time, systematically characterized in this epidemiological context.

The traveling front that connects a disease-free region ahead of the wave to an endemic region behind it is not a single smooth transition. Instead, it is a concatenation of distinct segments, each with its own character.

Near the endemic equilibrium — the state behind the front where the disease has permanently established itself — the dynamics are what the researchers call "local." They don't fit neatly into either the fast or slow category; they're the complex, three-dimensional behavior near a genuine equilibrium, where all three variables (susceptibles, infected, and the auxiliary diffusion variable) interact in intertwined ways.

As the orbit moves away from this equilibrium, it approaches a one-dimensional curve called the critical manifold. This is where the slow dynamics take over. On this manifold, the infected population and its spatial derivative are both zero — the system sits in a state of near-quiescence, with no active infection but with a reservoir of susceptible individuals slowly accumulating as immunity wanes. The susceptible population drifts gradually, controlled by the small δ parameter. This is the quiet phase between outbreaks, where the conditions for the next eruption are being quietly assembled.

Eventually, as susceptibles accumulate, the system reaches a tipping point — a threshold where the dynamics become unstable and the slow manifold can no longer contain the trajectory. This is the "exit" point. The orbit then executes a rapid, almost explosive transition governed by the layer problem: the fast dynamics take over, infection surges, and the epidemic "front" advances sharply.

This fast excursion carries the orbit to a region where infection is again low — but now at a different location in the susceptible-infected plane. The process then repeats: slow drift near the critical manifold, exit at a critical threshold, fast outbreak excursion, and so on. The result is a traveling front made of alternating slow and fast segments, like a heartbeat rhythm rather than a smooth pulse.

Spatial profiles of S and I showing sharp front formation

Label	Value
Disease-free	0
Infected (I)	0
Sharp transition	0

Figure 1a: Spatial profiles at fixed time. Red = susceptible, Black = infected. Parameter values: ℛ₀=3, δ=10⁻³, D=10.

The spatial profiles from numerical simulations show this beautifully. In the susceptible population (shown in red in Figure 1), the change is gradual — a slow rise from the low values behind the front to the full susceptible population ahead of it. In the infected population (in black), the change is abrupt — a sharp transition layer that defines the epidemic front itself. This coexistence of different spatial scales — slow variation in the susceptible background, rapid transitions in infection — is not an artifact of the numerics. It's a direct consequence of the slow-fast structure built into the model's parameters.

Composition of traveling front dynamics

Label	Value
Local dynamics near EE	20
Slow evolution	50
Fast excursion	30

Figure 5: Multiscale structure showing local (green), slow (blue), and fast (red) segments of the orbit.

The phase space trajectory in three dimensions makes the structure even clearer. The orbit winds through (S, I, J) space — where J is the spatial derivative of infection — following a path that hugs near the critical manifold for extended stretches, punctuated by fast loops that shoot out and back. The projection onto the (S, I) plane shows the same alternation: slow, nearly horizontal segments (where I changes little as S drifts) interrupted by sharp, curved excursions where infection spikes.

The researchers went further and derived a quantitative characterization of the transition between slow and fast regimes. This is where the "entry-exit mechanism" comes in. In singular perturbation theory, the transition from slow to fast dynamics is not instantaneous; there's a delay, a hysteresis-like effect where the system "waits" a certain time before leaving the slow manifold. The researchers derived an explicit integral condition that determines this exit time, showing it scales proportionally with the logarithm of the slow parameter δ. This gives a concrete prediction: the slower immunity is lost, the longer the system lingers in the quiet phase between outbreaks.

Exit time scaling with slow parameter δ

Label	Value
0	1
δ=10⁻³	0.7
δ=10⁻⁴	0.5
δ=10⁻⁵	0.35

Figure 6c: Ratio TE/Δx scaling as O(δ) as immunity loss rate decreases.

The exit time also translates directly into a spatial prediction: the distance the front travels during each slow segment. The researchers showed that the ratio between exit time and spatial extent follows an O(δ) scaling law, meaning that as immunity wanes more slowly, the spatial separation between successive outbreaks grows — but only in proportion to how slow the waning is.

Why This Changes Things

Epidemiologists have long known that diseases with temporary immunity tend to recur — the classic susceptible-recovered-susceptible cycle produces recurrent outbreaks even in closed populations. What's less well understood is how this temporal recurrence maps onto spatial spread.

This work provides a rigorous mathematical framework for that mapping. The key insight is that the slow loss of immunity doesn't just control the timing of recurrent outbreaks in a single location; it also controls the spatial structure of invasion. Each pulse of the temporal cycle corresponds to a segment of the traveling front. The quiet phase, where immunity wanes and susceptibles accumulate, corresponds to the slow stretch where the susceptible background changes gradually. The outbreak phase corresponds to the sharp transition layer that defines the epidemic front.

This has practical implications. If you're trying to predict the speed at which a disease will advance through a region, you need to know more than just the transmission parameters. You need to know the duration of immunity. Two diseases with identical transmissibility but different immunity durations will spread at different speeds — not because their infection dynamics differ, but because their slow-fast structure differs.

The finding also explains a qualitative pattern that has been observed but not systematically explained: the "pulsed" nature of some epidemic invasions, where surveillance data shows distinct waves rather than a smooth advancing front. This is exactly what you'd expect if the traveling front is built from alternating slow and fast segments. The pulses aren't anomalies; they're the signature of the underlying slow-fast architecture.

From a mathematical perspective, the paper contributes to a growing body of work applying geometric singular perturbation theory to epidemiological models. Previous work had analyzed the temporal dynamics of fast-slow SIRS models, but the spatial (traveling-wave) extension had remained largely unexplored. This paper shows that the GSPT framework — developed for abstract dynamical systems — maps naturally onto the epidemic context, and that the entry-exit mechanism, previously studied in mechanical and chemical systems, has a clear epidemiological interpretation.

The assumption that diffusion acts only on the infected compartment is a simplification, but it's a productive one. It keeps the analysis tractable while capturing the essential mechanism of spatial invasion. The paper's authors are careful to note that incorporating differential diffusion across compartments — infected individuals moving faster or slower than susceptibles, for instance — is a natural direction for future work. So is adding demographic dynamics (births and deaths), which would introduce additional parameters.

Another limitation is that the analysis focuses on the existence and structure of traveling fronts, not on calculating their exact speed. The speed c appears as a parameter in the equations, and the analysis shows that traveling fronts exist for a range of speeds, but the minimal or selected speed isn't derived from first principles. This is common in the traveling wave literature — existence is often established before uniqueness or selection — but it's a gap that future work could address.

What's Next

The framework opened by this paper suggests several promising directions.

First, the relationship between invasion speed and immunity duration deserves further exploration. The present work establishes the structural decomposition of the front; a natural next step would be to derive explicit formulas for how the speed depends on δ, ℛ₀, and other parameters. Such formulas would be directly useful for parameter estimation from surveillance data.

Second, the assumption of diffusion only on the infected compartment should be relaxed. Real populations don't move as a homogeneous group of infectives. Susceptible individuals move too — often in response to perceived risk, which introduces feedback between infection levels and mobility. Incorporating this feedback would make the model more realistic and, likely, more complicated. The geometric framework developed here provides a scaffold for that extension.

Third, the entry-exit mechanism derived in the paper could be tested against real epidemic data. For diseases with clear pulsed invasion patterns — some vector-borne diseases, for instance, or diseases with strong seasonal forcing — one could attempt to estimate the slow parameter from the spacing between waves and see whether the predictions match.

Finally, the connection to pattern formation deserves attention. The traveling fronts studied here are only one possible outcome of diffusive epidemic models. Under other conditions — different diffusion rates, additional nonlinearities — the same models can produce spatial patterns (Turing patterns, waves of infection that oscillate in time as well as space). The geometric framework could be extended to characterize these alternative regimes, building toward a complete picture of the spatial dynamics that simple epidemic models can support.

What this work ultimately offers is a new lens on an old problem. Epidemics have been modeled mathematically for a century, but the tools of modern dynamical systems theory — geometric singular perturbation theory, the entry-exit mechanism, the geometry of invariant manifolds — are still being brought to bear on them. The result is often a deeper understanding of structures that were glimpsed before but never quite resolved. In this case, the structure is the traveling front as a concatenation of slow and fast segments, and the resolution it offers is a quantitative link between the biology of immunity and the geometry of invasion. That link is new, and it's worth taking seriously.