I'm studying a model in the field of complex systems regarding the epidemic spreading. The model is the susceptible-infected model, i.e., there is a population of N subjects and each of them can either be infected or susceptible so that: $S(t)+I(t)=N$ and $N$ is the total population. The model is discrete in the sense that infections happen in discrete time-step. The model defines a rate $\beta$ as $\textit{the average number of people each infectious person spreads the disease to each day}$ and writes the differential equation:

$$\frac{dI}{dt}=\beta \frac{S(t)}{N}I(t)$$

I don't understand this equation, as far as I am concerned, I would write an equation like this:

$$I(t+1)=I(t)+\beta I(t)$$ which means that the number of infected people at time $t+1$ is equal to the number of people who were infected at time $t$ plus the average number of susceptible people who are infected during a timestep (recover from the disease is not included in the model), but my formula would lead to $\frac{dI}{dt}=\beta I(t)$ so I don't understand why in this model they also include $\frac{S(t)}{N}$. As $\beta$ is defined, isn't the number of susceptible people, in a certain way, already included in that rate? Why do I have to multiply by $S(t)$? Being the definition of $\beta$ I think that $\beta I(t)$ already gives the average number of people getting infected in a timestep, where am I wrong? How should a person approach to this kind of problems if he wants to write a differential equation?


2 Answers 2


You almost nailed it. Notice that $$I(t+1)-I(t) = \frac {dI}{dt}$$

Now, why they use $\beta \alpha I(t)$ instead of simply $\beta I$,- the reason is simple. Limiting reason is that infection each day can't spread to more people than there are uninfected/susceptible ones. Besides as pool of infected persons grows,- chances grows that typical infected person will make contact with already infected person, so $\beta$ must go down with time, due to these pair of reasons. This is accumulated in a $\alpha = S/N$ coefficient which decreases available pool of persons $\alpha I$ for infection in each step.


To get impression how $I(t)$ chart looks like, you can try to integrate such expression :

$$ \int \frac {dI}{I} = \int \beta \frac SN dt \tag 1$$

Knowing that $S=N-I(t)$, and assuming a-priory that $I(t)$ "should go" like $e^{\beta t}$ if boundaries would be ignored, we can re-write (1) as :

$$ \int \frac {dI}{I} = \int \beta \left( 1- \frac {e^{\beta t}}{N} \right) dt \tag 2$$

Integrating both sides and solving for $I(t)$ we get :

$$ I\left(t\right) = I_0 \exp\left(\beta t-\frac{\exp\left(\beta t\right)}{N}\right) \tag 3$$

which looks like

enter image description here

It may not be really like that, but I hope you get the reasons of model behind.

  • 1
    $\begingroup$ Thank you for your answer, so the problem is the DE I propose has an exponential growth solution which is incompatible with a finite number of people; to take into account this last fact we can multiply by the coefficient $\alpha$. But still I can't figure out why we multiply exactly for $S/N$. $\endgroup$
    – Salmon
    Commented May 24 at 13:19
  • 1
    $\begingroup$ Exactly. For the choosing $\alpha$ to be specifically S/N ratio, I think because it cancels exponential growth of infected population nicely on small S numbers. $\endgroup$ Commented May 24 at 16:08

Let me first note that SI model is a simplified version of a more general SIR model (in fact, there is a whole zoo of related models.)

The number of persons that get infected at time step $t$ is proportional to the number of susceptible persons $S(t)$ and the number of infected persons $I(t)$. Indeed, the probability that an infected person meets a susceptible person and transmits infection is the probability that one person is infected, $I(t)/N$ times the probability that the other person is susceptible $S(t)/N$. So the change is $$ I(t+1) - I(t) \propto \frac{I(t)}{N}\frac{S(t)}{N} $$

$\beta$ is just a proportionality coefficient that assures correct rate, so that the results agree with the experimental data (e.g., in a densely populated city people run into each other more frequently than in countryside, which results in higher $\beta$. Or it may account for seasonal conditions, e.g., in winter people spend more time indoors and transmission of infection is more likely. It also accounts for the infectivity of the disease.)


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.