1.2.? Poisson

---

title: "Poisson Distribution"

Poisson Distribution

A machine turns its gears.

Not rhythmically. Not predictably. But events happen.

A click. Another. Silence. Then three in a row.

This is not pure chaos.

It is randomness with structure.

And this is where the Poisson distribution begins.

---

What does Poisson measure?

The Poisson distribution answers a very particular kind of question:

How many times does something happen in a fixed interval?

That interval may be:

• a stretch of time
• a region of space
• a segment of material
• or any bounded domain in which events may occur

Examples:

• How many emails arrive in one hour?
• How many cars pass a checkpoint in ten minutes?
• How many typos appear in one page?
• How many mutations occur in a strand of DNA?
• How many stars are found in a small patch of sky?

This is already a clue:

Poisson is not about measuring how much. It is about counting how many events occur.

---

A different kind of randomness

Some random variables measure values:

• height
• weight
• temperature
• speed

Poisson does not.

Poisson counts occurrences.

The random variable is something like:

“number of arrivals” “number of failures” “number of calls” “number of impacts”

So the possible values are:

$$ 0,;1,;2,;3,;\dots $$

This makes Poisson a discrete distribution.

---

The central idea

The Poisson distribution appears when events:

1. occur independently
2. happen with a stable average rate
3. are counted over a fixed interval
4. are individually rare, even if many opportunities exist

This combination matters.

It means the world is not following a clock, but neither is it arbitrary in the wildest sense. There is a stable background tendency, and on top of that tendency, individual events appear unpredictably.

That is the Poisson world.

---

The parameter: $\lambda$

Poisson is governed by a single parameter:

$$ \lambda $$

This lambda represents the expected number of events in the interval.

If on average 3 customers arrive every 10 minutes, then:

$$ \lambda = 3 $$

If on average 12 radioactive emissions are detected per second, then:

$$ \lambda = 12 $$

If on average 0.4 printing errors appear per page, then:

$$ \lambda = 0.4 $$

So $\lambda$ is not the number that must happen. It is the number the process tends to orbit around.

---

The formula

If a random variable $X$ follows a Poisson distribution with parameter $\lambda$, then the probability of observing exactly $k$ events is:

$$ P(X = k) = \frac{\lambda^k e^{-\lambda}}{k!} $$

where:

• $k = 0, 1, 2, 3, \dots$
• $\lambda > 0$

This formula gives the probability of seeing exactly $k$ events in the chosen interval.

---

Reading the formula like a human

A formula should not be treated as sacred decoration. It should be read.

Let us slow it down:

$$ P(X = k) = \frac{\lambda^k e^{-\lambda}}{k!} $$

$\lambda^k$

This reflects the pressure toward producing $k$ events. Larger $\lambda$ makes larger counts more plausible.

$k!$

This compensates for overcounting arrangements. When several events happen, their ordering inside the count does not matter.

$e^{-\lambda}$

This is the damping factor. It prevents the probabilities from exploding and ensures the whole distribution sums to 1.

So the formula is not a random artifact. It balances:

• the tendency for events to occur
• the combinatorial correction
• the normalization that keeps probability coherent

---

The deep origin: rare events

This is the part that turns memorization into understanding.

Poisson is what emerges when you have:

• many opportunities
• each with tiny probability
• but with a stable expected total

That should sound familiar.

It is the logic of the binomial distribution pushed into a special limit.

If:

$$ X \sim \text{Binomial}(n, p) $$

then the expected number of successes is:

$$ n p $$

Now imagine:

• $n$ becomes very large
• $p$ becomes very small
• but the product stays constant

$$ n p \to \lambda $$

Then the binomial model approaches the Poisson model:

$$ \text{Binomial}(n,p);\longrightarrow;\text{Poisson}(\lambda) $$

This is one of the most important conceptual facts about Poisson.

It is not just “a counting formula.” It is the natural shape of rare events aggregated over many chances.

---

Visual intuition

You can imagine time split into many tiny slices.

In each tiny slice:

• either nothing happens
• or one event happens

If the slices are tiny enough, multiple events in the same slice become negligible.

Most slices contain nothing. A few contain one event. Across the full interval, those tiny possibilities accumulate.

That accumulation is Poisson.

flowchart LR
    A[Many tiny opportunities] --> B[Each event is rare]
    B --> C[Average rate stays stable]
    C --> D[Count total events]
    D --> E[Poisson distribution]

---

A timeline picture

Here is the feeling of a Poisson process in time:

timeline
    title Random arrivals in one interval
    00:00 : no event
    00:08 : event
    00:21 : no event
    00:37 : event
    00:41 : event
    00:58 : no event

Notice what matters:

• not regular spacing
• not a deterministic rhythm
• but a count over the interval

Poisson does not care exactly when each event occurs inside the interval. It cares about how many occurred.

---

Example

Suppose a help desk receives, on average, 4 requests per hour.

Then:

$$ \lambda = 4 $$

What is the probability of receiving exactly 2 requests in one hour?

Using the formula:

$$ P(X=2)=\frac{4^2 e^{-4}}{2!} $$

$$ P(X=2)=\frac{16e^{-4}}{2}=8e^{-4} $$

Numerically,

$$ P(X=2)\approx 0.1465 $$

So there is about a 14.65% chance of receiving exactly two requests in that hour.

---

Shape of the distribution

When $\lambda$ is small, the distribution is heavily concentrated near 0.

When $\lambda$ grows, the center moves rightward and the distribution becomes more spread out.

flowchart TD
    A[Small λ] --> B[Most mass near 0]
    C[Moderate λ] --> D[Mass spreads over several counts]
    E[Large λ] --> F[Distribution becomes more bell-like]

This gives a useful intuition:

• small rate → events are scarce
• large rate → events are more common
• larger $\lambda$ also means larger variability in absolute terms

---

Mean and variance

A remarkable property of the Poisson distribution is this:

$$ \mathbb{E}[X] = \lambda $$

and also

$$ \mathrm{Var}(X) = \lambda $$

So in a Poisson model:

• the mean equals $\lambda$
• the variance also equals $\lambda$

This is elegant, and also diagnostically useful.

If observed count data has variance much larger than the mean, a plain Poisson model may be too simple.

---

Why this matters

Poisson is one of the first places where probability feels like a real language of the world.

It tells us that even when individual events are unpredictable, aggregate behavior may still be mathematically intelligible.

That is a deep idea.

A single arrival is uncertain. But the pattern of arrivals is not beyond thought.

Poisson is one answer to a larger question:

How can disorder still produce a stable form?

---

When Poisson is a good model

Poisson tends to be appropriate when:

• you count events in a fixed interval
• events are approximately independent
• the average rate is roughly constant
• simultaneous events are negligible
• the count is not naturally capped by a small fixed number of trials

Examples include:

• arrivals
• accidents
• defects
• emissions
• failures
• calls
• mutations

---

When Poisson is not a good model

Poisson can fail when:

• events affect one another
• the rate changes over time
• counts cluster in bursts
• data contains too many zeros
• variance is much larger than the mean

In those cases, we may need richer models.

This matters because technical correctness is not just about applying a formula. It is about recognizing whether reality actually matches the assumptions.

---

Poisson vs Binomial

These two distributions are close cousins, but they are not the same thing.

Binomial

Ask:

How many successes occur in a fixed number of trials?

• fixed number of attempts
• each trial has success/failure
• parameterized by $n$ and $p$

Poisson

Ask:

How many events occur in a fixed interval?

• no explicit fixed number of trials
• events arise over time or space
• parameterized by $\lambda$

A useful way to think about it:

• Binomial counts success across a known set of chances
• Poisson counts events when the chances are too many or too small to track individually

flowchart LR
    A[Binomial] --> B[Fixed number of trials]
    B --> C[Success / failure]
    C --> D[np = expected count]

    D --> E[If n is large and p is small]
    E --> F[Poisson approximation]

    G[Poisson] --> H[Fixed interval]
    H --> I[Count of events]

---

A more intuitive summary

Poisson is the mathematics of scattered happenings.

Not one large event. Not a measured magnitude. But many small possible occurrences, each uncertain, together producing a stable statistical shape.

It is the distribution of:

• arrivals without schedule
• defects without intention
• emissions without rhythm
• coincidences under a constant background tendency

In this sense, Poisson is one of probability’s most revealing ideas:

randomness can have architecture

---

Final compression

A Poisson random variable counts how many times an event occurs in a fixed interval when:

• events are independent
• the average rate is constant
• individual events are rare
• we care about the total count

Its formula is:

$$ P(X = k) = \frac{\lambda^k e^{-\lambda}}{k!} $$

Its mean and variance are both:

$$ \lambda $$

And its conceptual origin is the limit of many tiny chances accumulating into a finite expected count.

That is not just a formula.

That is a worldview.

---

Optional: quick Python check

import math

λ = 4
k = 2

probability = (λ**k) * math.exp(-λ) / math.factorial(k)
probability

You should obtain approximately:

0.1465

---

Exercise prompts

1. A website receives on average 6 visits per minute. What is the probability of receiving exactly 8 visits in one minute?

2. A book has on average 0.3 typos per page. What is the probability that a page has no typos?

3. Explain, in words, why Poisson can be understood as a limit of the binomial distribution.

4. Give one real process that is plausibly Poisson and one that is not. Justify both.