Skip to content
Tutorials Center
My learning

Topics

Lesson 20 of 36 · Poisson Processes

Interarrival and Arrival Times

See why waiting times between Poisson events are exponential, why the n-th arrival is Erlang, and how memorylessness works, all checked by simulation.

The previous lesson counted events in a window. There is a second, equally useful view: look at the times at which the events occur. Write T₁, T₂, T₃, … for the arrival times and X₁ = T₁, X₂ = T₂ - T₁, … for the gaps between them, called interarrival times.

For a Poisson process with rate λ, the gaps have a beautifully simple law. This lesson explains it and uses it to simulate a Poisson process quickly and exactly.

Gaps are exponential

The first event happens after time t exactly when there are zero events in (0, t]. The count is Poisson(λt), so

Text
P(X₁ > t) = P(N(t) = 0) = e^(-λt)

That is the survival function of an exponential distribution with rate λ. Its density is λ e^(-λt) for t ≥ 0, its mean is 1/λ and its variance is 1/λ².

Because increments are independent and stationary, after each event the process "starts afresh". So X₁, X₂, … are independent and identically distributed Exponential(λ). This is the other half of the characterization:

A counting process whose interarrival times are independent Exponential(λ) random variables is a Poisson process with rate λ.

If calls arrive at 3 per hour, the average gap is 1/3 hour = 20 minutes.

Arrival times are Erlang

The n-th arrival time is a sum: Tₙ = X₁ + … + Xₙ. A sum of n independent exponentials has an Erlang (Gamma with integer shape) distribution:

Text
mean(Tₙ) = n/λ        variance(Tₙ) = n/λ²

The two views are linked by one equation: Tₙ ≤ t exactly when N(t) ≥ n. Waiting for the n-th event to arrive by time t is the same as having at least n events by t.

Memorylessness

The exponential distribution has a famous property:

Text
P(X > s + t | X > s) = P(X > t)

If you have already waited s minutes for the next event, the remaining wait has the same distribution as a fresh wait. The process has no memory of how long ago the last event was. This is why "the bus is due any minute now" is nonsense for a true Poisson stream: the expected remaining wait never shrinks.

Memorylessness belongs to the exponential (and its discrete cousin, the geometric). Real waiting times for scheduled services are not memoryless. Do not use a Poisson model for a timetable.

Check it by simulation

Python Exponential gaps and memorylessness

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The two probabilities agree up to sampling noise, and both are close to e^(-λt).

Generating a Poisson process from gaps

This gives a fast, exact simulator: keep adding exponential gaps until you pass the end of the observation window.

Python Build a Poisson process from exponential gaps

Press Run (or Ctrl + Enter) to see the result here.

Program input · what you'd type while the program runs

Type the answers before you press Run, one per line. Each time the program reads a line (with input()), it gets the next line from this box.

The counts come out Poisson with mean and variance near λT = 6, even though we never used the Poisson formula. That is the equivalence of the two views in action.

Given the count, arrivals are uniform

One more fact is handy. If you are told that exactly n events happened in [0, T], the unordered arrival times are n independent uniform points on [0, T]. So "the events are scattered completely at random" has a precise meaning. A practical use: if 5 calls arrived in an hour, the chance that the first one came in the first 10 minutes is 1 - (5/6)^5 ≈ 0.60.

flowchart LR
  S["Start"] -->|"X1 ~ Exp(λ)"| E1["Event 1 (T1)"]
  E1 -->|"X2 ~ Exp(λ)"| E2["Event 2 (T2)"]
  E2 -->|"X3 ~ Exp(λ)"| E3["Event 3 (T3)"]

Independent exponential gaps join the arrival times.

Check your understanding

Quick check · 4 questions

of 4 answered You scored / 4 · perfect!

  1. 1. Events arrive as a Poisson process with rate 5 per minute. What is the mean gap between events?

    The mean of an exponential gap is 1/λ = 1/5 = 0.2 minutes.
  2. 2. You have waited 10 minutes for the next Poisson event. What is the distribution of the remaining wait?

    Memorylessness: the remaining wait is again Exponential(λ).
  3. 3. What is the distribution of the third arrival time T3?

    T3 is the sum of three independent Exponential(λ) gaps, which is Erlang(3, λ).
  4. 4. Tₙ ≤ t is the same event as which statement about the counting process?

    The n-th event has happened by time t exactly when at least n events are counted by t.

Summary

  • Interarrival times of a Poisson process are independent Exponential(λ), with mean 1/λ.
  • The n-th arrival time Tₙ is Erlang: mean n/λ, variance n/λ².
  • The exponential is memoryless, so the past never shortens the wait.
  • Adding exponential gaps until you pass T simulates a Poisson process exactly.
  • Given N(T) = n, the arrival times are uniform on [0, T].

Next: Superposition and Thinning