Is a probability space the same as a probability distribution?

Not exactly. A distribution can describe a random variable derived from an underlying probability space.

Must the sample space be finite?

No. It can be countably infinite, continuous, or a more complex collection such as sequences.

Why include an event collection?

It specifies the sets to which the probability measure applies and supports consistent set operations, especially in infinite settings.

AI Glossary / Statistics and probability / Intermediate

Probability space

Definition

A probability space is the mathematical structure used to describe uncertain outcomes and assign probabilities consistently. It consists of three parts: a sample space containing possible outcomes, a collection of events called a sigma-algebra, and a probability measure assigning numbers to those events. Usually written (Omega, F, P), it provides the foundation on which random variables, distributions, and expectations are defined.

What it means

The sample space answers what could happen. The event collection answers which sets of outcomes the model treats as measurable events. The probability measure answers how likely those events are under the model.

For a small finite example, every subset of the sample space can be an event, making the construction straightforward. With continuous outcomes or infinite sequences, the event collection becomes more important. Separating these three ingredients prevents a list of possible outcomes from being mistaken for a complete probability specification and clarifies how different random quantities relate to the same underlying experiment.

How it works

Define outcomes at the level needed for the question. Choose an event collection that contains the whole sample space and is closed under complements and countable unions. These closure properties allow consistent reasoning about combinations of events.

Assign a probability measure with nonnegative values, total probability one, and countable additivity for disjoint events. In a finite space, assigning nonnegative masses summing to one to individual outcomes determines probabilities of all subsets by addition.

Define random variables as measurable functions of outcomes. Several variables on the same space can share information or be dependent. Their distributions describe the probabilities induced by those functions. Independence is an additional relationship between events or variables; it is not a consequence of writing them on the same space. A mathematically valid construction also remains a model whose relevance to real observations must be assessed.

Three practical examples

A two-stage service workflow models whether each stage succeeds and derives the total number of successful stages.
A simulator defines a probability space for random arrivals and computes waiting times as functions of the simulated outcomes.
An AI evaluation model treats sampled examples and randomized algorithm choices as sources of uncertainty within a specified experiment.

Think of it this way

Think of a probability space as an organized catalog. It contains possible complete scenarios, a rulebook for grouping scenarios into events, and a consistent allocation of probability to those groups. A random variable then reads a particular measurement from each cataloged scenario.

Worked example

Consider an original model for two checks, A and B, each returning zero or one. The sample space contains four ordered outcomes: (0,0), (0,1), (1,0), and (1,1). Let the event collection contain all subsets, and assign masses 0.50, 0.20, 0.10, and 0.20 respectively.

The masses are nonnegative and sum to one. The event A = 1 consists of (1,0) and (1,1), so its probability is 0.10 + 0.20 = 0.30. Similarly, P(B = 1) = 0.40, while P(A = 1 and B = 1) = 0.20.

The checks are not independent because 0.20 differs from 0.30 × 0.40 = 0.12. Given B = 1, the probability that A = 1 is 0.20/0.40 = 0.50.

Define X = A + B. Its probabilities at zero, one, and two are 0.50, 0.30, and 0.20. The expected count is 0 × 0.50 + 1 × 0.30 + 2 × 0.20 = 0.70. This distribution is derived from the underlying space rather than specified independently of it.

Illustrative case study

Fictional case study: Cedar Simulation models a workflow with several automated checks. Separate spreadsheets contain individual success probabilities, but no one has specified whether checks share a common source of failure.

The team develops a joint scenario model before calculating workflow reliability. It documents which events are represented and how each reported metric is derived from them. This fictional case illustrates the value of a coherent probability foundation. It does not claim that enumerating scenarios alone identifies their correct real-world probabilities.

Common misconceptions and mistakes

The sample space is only one part of a probability space. Listing possible outcomes without assigning probabilities does not finish the model. Likewise, equal treatment in a list does not imply that outcomes are equally likely.

Do not confuse an outcome with an event: an event is a set of outcomes, which may contain one, several, or infinitely many elements. A probability-zero event in a continuous model need not be logically impossible; individual exact values can have zero probability while intervals have positive probability. Finally, variables defined on the same probability space are not automatically independent or identical.

Business use

Most business users do not need measure-theoretic notation to work with AI probabilities. The underlying questions are still practical: which scenarios are possible, which events matter, and how are their probabilities connected?

These questions become important when combining forecasts, simulating workflows, or estimating joint failures. Maintain a common definition of outcomes and time horizons across teams. Otherwise, apparently compatible probabilities may refer to different populations or experiments and cannot be combined coherently. Use a level of mathematical detail proportionate to the complexity of the application.

Check your understanding

Four possible outcomes are listed, but their probabilities are unspecified. Does this determine a unique probability space?

Answer: No. The outcome list alone does not specify the probability measure or, formally, the event collection. Many probability assignments could be consistent with that same list.

Related terms

Sample space
Event
Random variable
Independence
sigma-algebra
probability measure
joint distribution
expectation
measurable function

Learn next

Begin with finite sample spaces and probabilities of unions, intersections, and complements. Define several random variables on one space and derive their distributions. Then study countable additivity and measurable events to understand why continuous probability requires more structure than a finite table.

Sources

The Aalto Dictionary of Machine Learning defines probability spaces and their sample space, event collection, and probability measure: https://aaltodictionaryofml.github.io/ADictML_Main.pdf

The two-check model, derived distribution, calculations, and fictional case are original. The finite example uses all subsets as events; more general spaces require an appropriate sigma-algebra.

Previous: Probability simplex Next: Probability theory