关于指数分布无记忆性与概率密度函数（PDF）的疑问

Great question—this is a super common confusion when getting to grips with exponential distributions, so let's break it down step by step.

First, let's clear up the core mix-up: the constant occurrence rate you're referring to is the hazard rate (also called failure rate), not the probability density function (PDF). These are two distinct concepts, and that's where the apparent contradiction vanishes.

What's the hazard rate, exactly? #

The hazard rate λ for the exponential distribution represents the conditional probability rate of an event occurring at time x, given that the event hasn't happened before x. For the exponential distribution, this rate is constant:
h(x) = λ
This is directly tied to the "memoryless" property—if a transistor has survived 10 hours, its chance of failing in the next hour is the same as a brand new transistor's chance of failing in the first hour.

Why the PDF isn't a horizontal line #

The PDF f(x) = λe^(-λx) describes the absolute probability density of the event occurring at exactly time x. Here's why it decays exponentially even with a constant hazard rate:

• At time x=0, all your transistors are still functional. The density of failures here is high because there are many devices that could fail.
• As time increases, fewer and fewer transistors remain functional (this is captured by the survival function S(x) = e^(-λx), which gives the probability the event hasn't happened by time x).
• The PDF is actually the product of the hazard rate and the survival function: f(x) = h(x) * S(x). Since S(x) decays exponentially, even with a constant h(x), the PDF will follow an exponential decay curve.

A concrete example #

Let's say λ = 0.1 failures per hour (10% hourly failure rate for surviving transistors):

• At x=0, f(0) = 0.1 * e^0 = 0.1—high failure density because all transistors are intact.
• At x=10 hours, S(10) = e^(-1) ≈ 0.368, so f(10) = 0.1 * 0.368 ≈ 0.0368—lower failure density, because only ~37% of transistors are still around to fail. But the hazard rate is still 10%: of those 37%, 10% will fail in the next hour.

So to wrap up: there's no contradiction at all. The exponential distribution's constant hazard rate (your "constant occurrence rate") doesn't imply a flat PDF—it implies that the conditional chance of failure stays the same, while the absolute density of failures decreases over time as fewer subjects/items remain at risk.

内容的提问来源于stack exchange，提问作者The Pointer