Math AI HL: Poisson Distribution

The Poisson distribution is an indispensable tool for quantifying uncertainty in events that occur randomly and independently over fixed intervals of time or space. From modelling the number of calls to a customer service center in an hour to predicting the decay of radioactive particles, it provides a mathematical framework for understanding rare events—those where an event happens with a small probability in a given trial but many trials are possible. Mastering this distribution is crucial for IB Math AI HL students, as it connects probability theory to real-world phenomena in operations research, biology, and quality control, forming a key part of your toolkit for stochastic modelling.

The Poisson Process and Its Defining Conditions

A Poisson distribution models the number of times a specific event occurs within a fixed interval. This interval can be a unit of time (e.g., minutes, hours), length, area, or volume. The events described must arise from a Poisson process, which has three critical conditions for validity.

First, events must occur independently. The occurrence of one event does not make the next event more or less likely. Second, the average rate at which events occur, denoted by the Greek letter lambda $\lambda$, must be constant. This means $\lambda$ is proportional to the length of the interval. If you know $\lambda =5$ events per hour, then for a 2-hour interval, the average becomes $\lambda =10$. Third, two events cannot occur at exactly the same instant. When these conditions hold—independence, constant average rate, and singularity—the scenario can be modelled using the Poisson distribution.

Consider a busy pizza restaurant that receives an average of 15 delivery orders per hour. This is a classic candidate for Poisson modelling: orders (events) generally occur independently of one another, the average rate of 15 per hour is stable during peak times (constant $\lambda$), and it's improbable for two orders to be placed at the exact same microsecond.

The Poisson Probability Formula and Key Properties

If a random variable $X$ follows a Poisson distribution with mean rate $\lambda$, we write $X\sim \mathrm{Po}(\lambda )$. The probability that $X$ equals a specific non-negative integer value $k$ (where $k=0,1,2,...$) is given by the formula: $$P(X=k)=\frac{e^{-\lambda }{\lambda }^k}{k!}$$

Here, $e$ is Euler's number (approximately 2.71828), and $k!$ denotes "$k$ factorial". This formula directly calculates the probability of observing exactly $k$ events in the given interval.

A unique and powerful property of the Poisson distribution is that its mean and variance are equal. For $X\sim \mathrm{Po}(\lambda )$: $$E(X)=\lambda \text{and}\mathrm{Var}(X)=\lambda$$ This relationship simplifies analysis. If you know the average rate, you immediately know the variance, which measures the spread of the distribution. A higher $\lambda$ indicates a higher average count and also greater variability around that average.

Worked Example: A call center receives an average of 4 calls per minute. What is the probability they receive exactly 6 calls in a given minute?

Here, $\lambda =4$ and $k=6$. $$P(X=6)=\frac{e^{-4}\cdot 4^6}{6!}$$

First, calculate $4^6=4096$ and $6!=720$. $$P(X=6)=\frac{e^{-4}\cdot 4096}{720}$$ Using $e^{-4}\approx 0.0183156$: $$P(X=6)\approx \frac{0.0183156\cdot 4096}{720}\approx \frac{75.02}{720}\approx 0.1042$$ Thus, there is approximately a 10.4% chance of receiving exactly 6 calls in a minute.

Modelling Real-World Rare Events

The Poisson distribution excels at modelling rare events across diverse fields. In queuing theory, it models customer arrivals at a bank or website hits. In quality control and manufacturing, it models the number of defects found in a fixed length of fabric or on a circuit board. In natural occurrences, it can model the number of mutations in a strand of DNA per unit length or the number of stars found in a fixed volume of space. The "rareness" is contextual; it refers to the low probability of the event in any tiny sub-interval, even if the expected count $\lambda$ over the full interval is moderate.

A vital application is adjusting the interval. If a help desk gets an average of 3 calls per hour ($\lambda =3$ for a 1-hour interval), the distribution for a 15-minute period requires a new $\lambda$. Since 15 minutes is one-quarter of an hour, the average rate becomes $\lambda =3\times \frac{1}{4}=0.75$ calls per 15 minutes. You would then use $X\sim \mathrm{Po}(0.75)$ to find probabilities for that shorter interval.

Efficient Calculation Using Your GDC

While the formula is instructive, your Graphical Display Calculator (GDC) is essential for efficient problem-solving, especially for cumulative probabilities like $P(X\le k)$ or $P(X>k)$. You must be fluent with the Poisson functions in your GDC's statistics/distribution menu.

There are typically two key functions:

1. Poisson probability density (pdf): This calculates $P(X=k)$ for a given $\lambda$.
2. Poisson cumulative distribution (cdf): This calculates $P(X\le k)$ for a given $\lambda$.

GDC Example: Using the previous call center scenario ($\lambda =4$), find the probability of receiving at most 2 calls in a minute.

You need $P(X\le 2)=P(X=0)+P(X=1)+P(X=2)$. Instead of three separate calculations, use the Poisson cdf function. Input $\lambda =4$ and an upper bound of $k=2$. Your GDC will return $P(X\le 2)\approx 0.2381$.

To find the probability of receiving more than 5 calls, $P(X>5)$, use the complement rule: $P(X>5)=1-P(X\le 5)$. You would calculate $1-\text{cdf}(\lambda =4,k=5)$.

Common Pitfalls

1. Misapplying the Distribution: The most common error is using the Poisson distribution when its core conditions are violated. For example, modelling customer arrivals when they come in groups (violating singularity) or modelling traffic accidents at an intersection where one accident causes a backlog and reduces the immediate chance of another (violating independence) leads to inaccurate models. Always check for independence and a constant average rate before proceeding.

1. Confusing $\lambda$ with Probability: $\lambda$ is the average number of events, not a probability. It can be any positive real number, including values greater than 1. The related probability of an event occurring in a tiny interval is proportional to $\lambda$ but is not equal to it.

1. Incorrect Interval Adjustment: Forgetting to scale $\lambda$ proportionally when the interval size changes is a frequent computational mistake. If the given rate is 10 events per day, and the question asks about a 6-hour period, you must use $\lambda =10\times (6/24)=2.5$, not $\lambda =10$.

1. Overlooking the Complement Rule: Students often perform lengthy manual sums for probabilities like $P(X\ge 7)$. It is far more efficient to compute $1-P(X\le 6)$ using the GDC's cdf function. Recognizing when to use $P(X>k)=1-P(X\le k)$ saves time and reduces error.

Summary

• The Poisson distribution, $X\sim \mathrm{Po}(\lambda )$, models the count of rare, independent events occurring at a constant average rate $\lambda$ within a fixed interval of time, space, or volume.
• Its probability mass function is $P(X=k)=\frac{e^{-\lambda }{\lambda }^k}{k!}$, and it has the unique property that its mean and variance are both equal to $\lambda$.
• Valid application requires satisfying the conditions of a Poisson process: events are independent, the average rate $\lambda$ is constant, and two events cannot occur simultaneously.
• It has wide-ranging applications for modelling phenomena such as system arrivals, mechanical or biological defects, and natural random occurrences.
• Efficient calculation for both exact ($P(X=k)$) and cumulative ($P(X\le k)$) probabilities is performed using the pdf and cdf functions on your GDC, with careful attention to scaling $\lambda$ for different interval sizes.