= Two-event probability in a stochastic SIR model
Consider a <stochastic SIR model> initially containing $N\geq1$ <susceptible individuals> and one infectious individual. Use infection rate $\beta SI$ and <recovery rate> $\gamma I$, with no births or loss of <immunity>. Let $T_2$ be the time of the second event. The event $T_2\leq t$ means at least two events by time $t$.
The first event must be an <infection>: a first recovery leaves no infectious individuals and stops the process. The first-infection time has unconditioned <probability density function> $\beta N e^{-a_0s}$, where $a_0=\beta N+\gamma$. After that infection, $S=N-1$, $I=2$ and the total rate is $a_1=2\beta(N-1)+2\gamma$. Conditioning on its time gives
$$
\Pr(T_2\leq t)=\int_0^t\beta N e^{-a_0s}\left[1-e^{-a_1(t-s)}\right]ds.
$$
The bracket is the <probability> of another event within the remaining interval, obtained from an <exponential distribution>. Consequently $\Pr(T_2\leq t)=\beta N a_1t^2/2+O(t^3)$ at short times.
This is not the <probability> of exactly two events by $t$: later events are permitted in the integral. Counting exactly two would additionally require survival without a third event, with the post-second-event rate depending on whether that second event was an <infection> or a recovery. Frequency-dependent infection rates require replacing $\beta SI$ by the model's population-normalized rate throughout.
Back to article page