This lecture defines zero-probability events and discusses some counterintuitive aspects of their apparently simple definition, in particular the fact that a zero-probability event is not an event that never happens.

**There are common probabilistic settings in which zero-probability
events do happen all the time!**

Table of contents

As we said, the definition is very simple.

Definition
Let
be an event and denote its
probability by
.
We say that
is a **zero-probability event** if and only
if

Despite the simplicity of the definition, there are some features of zero-probability events that might seem paradoxical.

We illustrate these features with the following example.

Consider a probabilistic experiment whose set of possible outcomes, called sample space and denoted by , is the unit interval:

It is possible to assign probabilities in such a way that each sub-interval has probability equal to its length:

The proof that such an assignment of probabilities can be consistently performed is beyond the scope of this example, but you can find it in any elementary measure theory book (e.g. Williams - 1991).

As a direct consequence of this assignment, all the possible outcomes have zero probability:

Stated differently, **every possible outcome is a zero-probability
event**.

This might seem counterintuitive. In everyday language, a zero-probability event is an event that never happens.

However, this example illustrates that a zero-probability event can indeed happen.

Since the sample space provides an exhaustive description of the possible outcomes, one and only one of the sample points will be the realized outcome.

But we have just demonstrated that all the sample points are zero-probability events: as a consequence, the realized outcome can only be a zero-probability event.

Another apparently paradoxical aspect of this probability model is that the sample space can be obtained as the union of disjoint zero-probability events:where each is a zero-probability event and all events in the union are disjoint.

If we forgot that the additivity property of probability applies only to countable collections of subsets, we would mistakenly deduce thatand we would come to a contradiction: , when, by the properties of probability, it should be .

Of course, the fallacy in such an argument is that is not a countable set and, hence, the additivity property cannot be used.

The main lesson to be taken from this example is that a zero-probability event is not an event that never happens (also called an impossible event): in some probability models, where the sample space is not countable, zero-probability events do happen all the time!

**The reason for the apparent paradoxes is that the sample space
described above is uncountable.** It has
the
power of the continuum.

Sample spaces of this kind are very common in statistics. They implicitly arise every time that we define a continuous random variable.

But why do we define mathematical objects that have such counterintuitive properties?

If you are eager to read an answer to this question and you know what a random variable is, you can read our page on continuous random variables.

The notion of a zero-probability event plays a special role in probability theory and statistics because it underpins the important concepts of almost sure property and almost sure event.

Often, we want to prove that some property is almost always satisfied, or something happens almost always.

"Almost always" means that the property is satisfied for all sample points, except possibly for a negligible set of sample points.

The concept of zero-probability event is used to determine which sets are negligible: if a set is included in a zero-probability event, then it is negligible.

Definition
Let
be some property that a sample point
can either satisfy or not satisfy. Let
be the set of all sample points that satisfy the
propertyDenote
its complement (the set of all points not satisfying property
)
by
.
Property
is said to be **almost sure** if there exists a zero-probability
event
such that
.

An almost sure property is said to hold **almost
surely** (often abbreviated as **a.s.**). Sometimes, an
almost sure property is also said to hold **with probability
one** (abbreviated **w.p.1**).

Remember (see the lecture on probability) that some subsets of the sample space may not be considered events.

The above definition of almost sure property allows us to consider also sets that are not, strictly speaking, events.

However, in the case in which
is an event,
is called an **almost sure event** and we say that
happens almost surely.

Furthermore, since there exists an event such that and , we can apply the monotonicity of probability:which in turn implies . Finally, recalling the formula for the probability of a complement:

Thus, an almost sure event is an event that happens with probability .

Consider the sample space and the assignment of probabilities introduced in the previous example:

We want to prove that the eventis a zero-probability event.

Since the set of rational numbers is countable and is a subset of the set of rational numbers, is countable.

This implies that the elements of can be arranged into a sequence:

Furthermore, can be written as a countable union:

Applying the countable additivity property of probability, we obtainsince for every .

Therefore, is a zero-probability event.

This might seem surprising: in this probability model there are zero-probability events comprising infinitely many sample points!

It can also easily be proved that the eventis an almost sure event.

In fact,and by applying the formula for the probability of a complement, we get

Below you can find some exercises with explained solutions.

Let and be two events.

Let be a zero-probability event and

Compute .

Solution

is a zero-probability event, which means thatFurthermore, using the formula for the probability of a complement, we obtainSince , by monotonicity we obtainSince and probabilities cannot be greater than , this implies

Let and be two events.

Let be a zero-probability event and

Compute .

Solution

is a zero-probability event, which means thatFurthermore, using the formula for the probability of a complement, we obtainIt is also true thatSince , by monotonicity, we obtainSince and probabilities cannot be greater than , this implies Thus, putting pieces together, we get

Williams, D. (1991) Probability with martingales, Cambridge University Press.

Please cite as:

Taboga, Marco (2021). "Zero-probability events", Lectures on probability theory and mathematical statistics. Kindle Direct Publishing. Online appendix. https://www.statlect.com/fundamentals-of-probability/zero-probability-events.

The books

Most of the learning materials found on this website are now available in a traditional textbook format.

Featured pages

- Maximum likelihood
- Multinomial distribution
- Convergence in probability
- Normal distribution
- Chi-square distribution
- Beta function

Explore

Main sections

- Mathematical tools
- Fundamentals of probability
- Probability distributions
- Asymptotic theory
- Fundamentals of statistics
- Glossary

About

Glossary entries

- Loss function
- Null hypothesis
- Discrete random variable
- IID sequence
- Probability mass function
- Continuous random variable

Share

- To enhance your privacy,
- we removed the social buttons,
- but
**don't forget to share**.