Skip to content
Tutorials Center
My learning

Topics

Lesson 21 of 36 · Poisson Processes

Superposition and Thinning of Poisson Processes

Merge independent Poisson streams, split one stream by coin flips, and use competing exponentials. Both rules are proved by intuition and checked in code.

Poisson processes are easy to combine and easy to take apart. Two independent streams of events can be merged into one, and one stream can be split into several by labelling events at random. In both cases the result is again Poisson. These two rules, superposition and thinning, make many queueing and reliability calculations short.

Superposition: merging streams

Suppose emails arrive at rate λ₁ = 2 per hour from colleagues and at rate λ₂ = 3 per hour from newsletters, independently. Put both streams in one inbox. The merged stream is a Poisson process with rate

Text
λ = λ₁ + λ₂ = 5 per hour

Why? The count in a window is the sum of two independent Poisson counts, and a sum of independent Poissons is Poisson with the sum of the means. The independence and stationarity of increments carry over.

For k streams with rates λ₁, …, λₖ, the merged rate is λ₁ + … + λₖ.

Which stream does the next event come from?

The next event of the merged stream is the earlier of two independent exponential waits, Exp(λ₁) and Exp(λ₂). Two facts about this race:

  • The time to the first event is Exp(λ₁ + λ₂).
  • The first event comes from stream 1 with probability λ₁ / (λ₁ + λ₂), independent of when it arrives.

With rates 2 and 3, the chance that the next email is from a colleague is 2/5 = 0.4.

Thinning: splitting a stream

Now go the other way. Take a Poisson process with rate λ. Give every event a label "A" with probability p and "B" with probability 1 - p, independently for each event. Then:

  • The A events form a Poisson process with rate pλ.
  • The B events form a Poisson process with rate (1 - p)λ.
  • The two processes are independent.

The independence is surprising: knowing that many A events occurred tells you nothing about B events, even though they come from the same source. This happens only because the total count is Poisson. (If the total number of events were fixed instead, A and B counts would be negatively correlated.)

Example: a shop sees 30 customers per hour, and each customer buys with probability 0.2 independently. Buyers arrive as a Poisson process with rate 6 per hour, non-buyers with rate 24 per hour, independent of each other.

Simulating both rules

Python Superposition and the race of two streams

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 merged count has mean and variance near 50, the minimum has mean near 0.2, and the stream-1 chance is near 0.4.

Python Thinning gives independent Poisson streams

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.

Each part is Poisson (mean equals variance), and the correlation is near 0 only for the Poisson total. The fixed-total case shows why the Poisson property matters.

Putting the rules to work

Example: a server and two types of jobs. Jobs arrive at rate 10 per minute. A job is "heavy" with probability 0.3. Heavy jobs form a Poisson process with rate 3 per minute. The probability of no heavy job in 30 seconds is e^(-3 × 0.5) = e^(-1.5) ≈ 0.223.

Example: competing failures. A machine fails from cause X at rate 0.01 per day and from cause Y at rate 0.04 per day. Time to the first failure is exponential with rate 0.05, mean 20 days. The failure is of type Y with probability 0.04/0.05 = 0.8.

flowchart LR
  S1["Stream 1: rate λ1"] --> M["Merged: rate λ1 + λ2"]
  S2["Stream 2: rate λ2"] --> M
  P["One stream: rate λ"] --> A["A events: rate pλ"]
  P --> B["B events: rate (1-p)λ"]

Merging adds rates; thinning splits them in proportion.

When a problem mentions "each event independently is of type A with probability p", think thinning. When it mentions "two independent sources", think superposition. Then everything reduces to rates.

Check your understanding

Quick check · 4 questions

of 4 answered You scored / 4 · perfect!

  1. 1. Two independent Poisson streams have rates 1.5 and 2.5 per minute. What is the rate of the merged stream?

    Rates add under superposition: 1.5 + 2.5 = 4.
  2. 2. In the merged stream of rates 1.5 and 2.5, what is the probability the next event is from the first stream?

    λ1/(λ1+λ2) = 1.5/4 = 0.375.
  3. 3. A Poisson process with rate 20 per hour is thinned by keeping each event with probability 0.25. What is the rate of the kept events?

    Kept events form a Poisson process with rate pλ = 0.25 × 20 = 5.
  4. 4. After thinning a Poisson process, how are the kept and discarded streams related?

    Thinning a Poisson process produces independent Poisson processes.

Summary

  • Superposition: independent Poisson streams merge into a Poisson stream with the summed rate.
  • In a race of exponentials, the winner is stream i with probability λᵢ / Σλ, and the first-event time is Exp(Σλ).
  • Thinning: independent labelling splits a Poisson stream into independent Poisson streams with rates pλ and (1-p)λ.
  • Independence after thinning is special to the Poisson total.

Next: Nonhomogeneous and Compound Poisson Processes