In January 2020, epidemiologists were already modelling COVID-19 spread using differential equations developed over a century earlier. The SIR model — first formulated by Kermack and McKendrick in 1927 — had been sitting in the mathematical literature for nearly a hundred years when it became the most urgently relevant piece of applied mathematics on earth. How did a handful of equations predict millions of outcomes — and where did they fall short?
From SIR to SEIR: Building the Model
The SIR model divides a population into three compartments that individuals flow through in sequence: Susceptible (not yet infected), Infected (currently infectious), and Recovered (immune). Three coupled ordinary differential equations govern the flow:
Where N is total population, β is the transmission rate (contacts per day × probability of transmission per contact), and γ is the recovery rate (1/infectious period). The model conserves population: S + I + R = N at all times.
SEIR extends the model by adding an Exposed compartment — people who have been infected but are not yet infectious. This captures the incubation period, which for COVID-19 averaged around five days. The flow becomes S → E → I → R, governed by a fourth parameter σ (the rate at which exposed individuals become infectious). The SEIR equations:
The power of this model is extraordinary given its simplicity. With just three or four parameters — all measurable from early case data — you can predict peak timing, total number affected, and the effect of interventions. The equations can be solved numerically in milliseconds even for populations of millions.
R0: The Most Important Number in Epidemiology
From the SIR equations, a critical quantity emerges naturally: the basic reproduction number R0 = β/γ. This represents the average number of secondary infections produced by a single infected individual in a fully susceptible population. R0 is the single most important parameter in epidemic dynamics:
- R0 < 1: Each infected person infects fewer than one other person on average. The epidemic dies out.
- R0 = 1: The epidemic is at the tipping point — stable, neither growing nor shrinking.
- R0 > 1: Each infected person infects more than one other. The epidemic grows exponentially until susceptibles run out.
From R0, the herd immunity threshold follows directly: H = 1 − 1/R0. This is the fraction of the population that must be immune (through infection or vaccination) before R0 drops below 1 and the epidemic begins to decline. Comparative R0 values across diseases:
For COVID-19 with R0 ≈ 2.5–3, the herd immunity threshold sits at roughly 60–67%. For measles with R0 ≈ 12–18, the threshold is 92–94% — which is why measles vaccination coverage in schools must be near-universal to prevent outbreaks. The math is unforgiving.
Flattening the Curve: How Interventions Work
In the SEIR framework, public health interventions have precise mathematical meanings. Reducing transmission — through masks, physical distancing, ventilation, school closures — directly reduces β. Better treatment reduces the infectious period and increases γ. Vaccination moves individuals directly from S to R without passing through E and I at all.
"Flattening the curve" is the graphical consequence of reducing β: the peak of the I(t) curve is lowered and spread over a longer time period, while the total area under the curve (total infections) is reduced. The goal is to keep peak I below healthcare system capacity — the horizontal line representing available ICU beds, ventilators, and staff.
Non-pharmaceutical interventions and rebound: Lockdowns reduce β dramatically, suppressing the epidemic curve. But they do not increase immunity. When restrictions lift and β returns to baseline, a susceptible population faces a second wave — the epidemic essentially resumes where it left off. This was the central dilemma of COVID-19 policy: suppress transmission at the cost of immunity accumulation.
The models sometimes failed to match reality. The SEIR assumption of homogeneous mixing — that any individual is equally likely to contact any other individual — breaks down completely when superspreader events drive transmission. A single choir rehearsal or nightclub can generate hundreds of cases; a randomly selected individual generates near-zero. Network epidemiology, which maps disease spread along actual social contact graphs, captures this heterogeneity far better.
Beyond SEIR: Modern Epidemic Modelling
SEIR is the workhorse, but modern epidemic models have grown far more sophisticated. SEIRD adds a Deaths compartment. SEIQR models quarantine. SEIRS models waning immunity, where recovered individuals return to the susceptible pool — crucial for COVID variants where immunity decays over months. Age-stratified models apply different β values between age groups, capturing the radically different COVID mortality rates between 20-year-olds and 80-year-olds.
Agent-based models abandon differential equations entirely, simulating individual people with personal characteristics — age, occupation, household, contact network. Each agent follows probabilistic rules for exposure, infection, and recovery. These models can capture superspreader dynamics, spatial heterogeneity, and behavioural responses (people reducing contacts when they feel ill) that compartmental models miss. The cost is computational: running a million-agent simulation requires serious hardware and careful validation.
The most consequential SEIR-family model of the pandemic was Imperial College London's Report 9, published by Neil Ferguson and colleagues on 16 March 2020. It estimated 510,000 UK deaths under an unmitigated epidemic and influenced the British government's pivot to lockdown policy within days. Whatever its limitations — and critics were thorough in identifying them — it demonstrated that mathematical epidemiology had moved from academic curiosity to policy-critical infrastructure.