5.3 The Exponential Distribution

The Exponential Distribution

The exponential distribution models the time you wait until some event happens: a light bulb burning out, the next customer walking in, or a machine breaking down. It's one of the most common continuous distributions you'll encounter, and its connection to the Poisson distribution makes it especially useful for real-world problems involving event rates.

Exponential Distribution Probability Calculations

The exponential distribution works well whenever you're measuring how long until something occurs, assuming events happen at a roughly constant rate.

The probability density function (PDF) is:

$f(x) = \lambda e^{-\lambda x}$ for $x \geq 0$

• $\lambda$ (lambda) is the rate parameter, meaning the average number of events per unit time
• $e$ is the mathematical constant ≈ 2.71828

The cumulative distribution function (CDF) gives you the probability that the event occurs before time $x$:

$F(x) = 1 - e^{-\lambda x}$ for $x \geq 0$

To find the probability that an event occurs in a specific time interval $[a, b]$, subtract the CDF values:

$P(a \leq X \leq b) = F(b) - F(a) = e^{-\lambda a} - e^{-\lambda b}$

The mean (expected waiting time) is $\frac{1}{\lambda}$, and the variance is $\frac{1}{\lambda^2}$. Notice these are both determined entirely by $\lambda$. If a store averages 3 customers per hour ($\lambda = 3$), the expected time between arrivals is $\frac{1}{3}$ of an hour, or 20 minutes.

Quick example: Suppose light bulbs fail at a rate of $\lambda = 0.1$ per month. What's the probability a bulb lasts less than 6 months?

$F(6) = 1 - e^{-0.1 \times 6} = 1 - e^{-0.6} \approx 1 - 0.5488 = 0.4512$

So there's about a 45% chance the bulb fails within 6 months.

Memoryless Property Applications

The memoryless property is what makes the exponential distribution unique among continuous distributions. It means the probability of waiting a certain additional amount of time doesn't depend on how long you've already waited.

Mathematically:

$P(X > s + t \mid X > s) = P(X > t)$ for non-negative $s$ and $t$

Here's what that looks like in practice: say a car battery has an exponentially distributed lifespan. If it's already lasted 3 years, the probability it lasts at least 2 more years is the same as the probability a brand-new battery lasts at least 2 years. The battery doesn't "remember" that it's been running for 3 years.

This property makes the exponential distribution a good fit when the event rate stays constant over time, regardless of age or history. Some real-world examples:

• Call centers: The probability of a call arriving in the next 5 minutes is the same whether the last call came 2 minutes ago or 30 minutes ago.
• Device failures: For components that fail randomly (not from wear and tear), the chance of failure in the next hour stays constant.

Keep in mind that the memoryless property is an assumption. It works well for random failures but poorly for things that wear out gradually, like brake pads or tires.

Poisson vs. Exponential Distributions

These two distributions are two sides of the same coin. They both describe a Poisson process (events occurring randomly at a constant average rate), but they answer different questions.

• Poisson distribution (discrete): How many events occur in a fixed time period?
• Exponential distribution (continuous): How long between consecutive events?

If events occur at rate $\lambda$ per unit time in a Poisson process:

• The number of events in a fixed time $t$ follows a Poisson distribution with parameter $\lambda t$
• The time between consecutive events follows an exponential distribution with rate $\lambda$

Converting between the two is straightforward:

1. If you know the average number of events per unit time ($\lambda$), the average time between events is $\frac{1}{\lambda}$
2. If you know the average time between events ($\mu$), the average event rate is $\frac{1}{\mu}$

For example, if earthquakes strike a region at an average rate of 2 per year (Poisson with $\lambda = 2$), then the time between earthquakes follows an exponential distribution with a mean of $\frac{1}{2}$ year, or 6 months.