Probability Spaces, Events, and Probability Axioms

The measure-theoretic language for reasoning about random phenomena

Statistics

Probability

Sample spaces, events, sigma-algebras, and the axioms that make probability calculus consistent.

Author

Jiahong Sun

Published

September 6, 2026

Modified

September 6, 2026

1 Topic at a Glance

One-sentence summary: A probability space specifies which outcomes are possible, which collections of outcomes can be assigned probabilities, and the rules those probabilities obey.

Item	Description
Domain	Probability & Randomness
Main question	How can uncertainty be described without contradiction?
Objects	Outcomes, events, and probability measures
Typical use	Sampling models, stochastic processes, and statistical data-generating models
Prerequisites	Basic set notation

2 The three ingredients

A probability space is a triple $(\Omega,\mathcal F,P)$.

$\Omega$ is the sample space, the set of elementary outcomes.
$\mathcal F$ is a sigma-algebra of events: the subsets of $\Omega$ to which probabilities may be assigned.
$P$ is a probability measure on $(\Omega,\mathcal F)$.

For a finite experiment it is usual, and harmless, to take $\mathcal F=2^\Omega$. For an experiment whose result is a real number, not every subset need be measurable; the Borel sigma-algebra generated by intervals is the standard choice. This technical restriction is what permits a probability theory compatible with limits and integration.

2.1 Events and set operations

An event $A\in\mathcal F$ occurs when the realized outcome $\omega$ belongs to $A$. Set operations encode ordinary logic:

Logic	Event operation
not $A$	complement $A^c$
$A$ and $B$	intersection $A\cap B$
$A$ or $B$	union $A\cup B$
exactly one of $A,B$	symmetric difference $A\triangle B$

Closure of $\mathcal F$ under complements and countable unions means that these logical combinations remain legitimate events. Closure under countable operations is essential for statements such as “the sequence eventually stays in a set.”

3 Kolmogorov’s axioms

For every $A\in\mathcal F$, $P$ satisfies Kolmogorov’s axioms:

Non-negativity: $P(A)\geq0$.
Normalization: $P(\Omega)=1$.
Countable additivity: for pairwise disjoint $A_1,A_2,\ldots$,

\[ P\!\left(\bigcup_{i=1}^{\infty}A_i\right)=\sum_{i=1}^{\infty}P(A_i). \]

The third axiom does more than the finite addition rule: it constrains probabilities of limiting events. It is the bridge from a finite coin-toss model to probability distributions and convergence results.

3.1 Consequences worth deriving

Because $\Omega=A\mathbin{\dot\cup}A^c$,

\[ 1=P(\Omega)=P(A)+P(A^c),\qquad P(A^c)=1-P(A). \]

If $A\subseteq B$, then $B=A\mathbin{\dot\cup}(B\setminus A)$, so $P(A)\leq P(B)$. For arbitrary $A,B$, split their union into disjoint pieces:

\[ P(A\cup B)=P(A)+P(B)-P(A\cap B). \]

In particular, disjoint events have $P(A\cap B)=0$, but events of probability zero are not necessarily impossible. A continuously distributed measurement can equal one exact value with probability zero while still producing some exact value. “Almost surely” means “except on a probability-zero event,” not “logically always.”

4 A sequencing example

Suppose one read is selected uniformly from a finite library of $N$ reads. Let $A$ be “the read maps to gene $g$” and $B$ be “the read has a base-quality score above 30 at a chosen position.” The elementary outcome is the selected read, and $P(A)=|A|/N$.

If low-quality reads are disproportionately unmapped, $A$ and $B$ overlap less than expected under independence. The inclusion–exclusion identity prevents double-counting reads satisfying both criteria. The independent experimental unit here is a sampled read for this sampling calculation; it is not a biological replicate for inference about expression across organisms.

5 Assumptions and common pitfalls

Pitfall	Why it fails	Better practice
Treating every subset of an infinite space as an event	A coherent probability measure cannot in general be assigned to all subsets	State the sigma-algebra or use standard Borel events
Equating disjointness with independence	Disjoint non-null events cannot occur together, whereas independent events may	Check $P(A\cap B)=P(A)P(B)$ for independence
Reading $P(A)=0$ as impossibility	Continuous models assign zero probability to individual values	Use “almost surely” precisely
Ignoring the sampling mechanism	Probabilities refer to a specified random experiment	Define the unit, sampling frame, and source of randomness

6 Connections to the Knowledge Map

Counting rules and combinatorial probability uses finite sample spaces to calculate probabilities.
Probability distributions collects common probability models built on these axioms.
Statistics overview places this foundation within the wider knowledge map.

7 Take-home Messages

A probability model is $(\Omega,\mathcal F,P)$, not merely a list of plausible outcomes.
Events are measurable sets; set algebra supplies the logic of probability statements.
Countable additivity is what makes probability compatible with limits.
A probability-zero event can occur; it is only negligible under the model.

8 References

Kolmogorov, A. N. Foundations of the Theory of Probability. Chelsea, 1950.
Billingsley, P. Probability and Measure, 3rd ed. Wiley, 2012.

--- title: "Probability Spaces, Events, and Probability Axioms" subtitle: "The measure-theoretic language for reasoning about random phenomena" description: "Sample spaces, events, sigma-algebras, and the axioms that make probability calculus consistent." author: "Jiahong Sun" date: "2026-09-06" date-modified: "2026-09-06" categories: [Statistics, Probability] toc: true number-sections: true draft: false --- # Topic at a Glance > **One-sentence summary:** A probability space specifies which outcomes are possible, which collections of outcomes can be assigned probabilities, and the rules those probabilities obey. | Item | Description | |---|---| | Domain | Probability & Randomness | | Main question | How can uncertainty be described without contradiction? | | Objects | Outcomes, events, and probability measures | | Typical use | Sampling models, stochastic processes, and statistical data-generating models | | Prerequisites | Basic set notation | # The three ingredients A probability space is a triple $(\Omega,\mathcal F,P)$. - $\Omega$ is the **sample space**, the set of elementary outcomes. - $\mathcal F$ is a **sigma-algebra** of events: the subsets of $\Omega$ to which probabilities may be assigned. - $P$ is a probability measure on $(\Omega,\mathcal F)$. For a finite experiment it is usual, and harmless, to take $\mathcal F=2^\Omega$. For an experiment whose result is a real number, not every subset need be measurable; the Borel sigma-algebra generated by intervals is the standard choice. This technical restriction is what permits a probability theory compatible with limits and integration. ## Events and set operations An **event** $A\in\mathcal F$ occurs when the realized outcome $\omega$ belongs to $A$. Set operations encode ordinary logic: | Logic | Event operation | |---|---| | not $A$ | complement $A^c$ | | $A$ and $B$ | intersection $A\cap B$ | | $A$ or $B$ | union $A\cup B$ | | exactly one of $A,B$ | symmetric difference $A\triangle B$ | Closure of $\mathcal F$ under complements and countable unions means that these logical combinations remain legitimate events. Closure under countable operations is essential for statements such as “the sequence eventually stays in a set.” # Kolmogorov's axioms For every $A\in\mathcal F$, $P$ satisfies Kolmogorov's axioms: 1. **Non-negativity:** $P(A)\geq0$. 2. **Normalization:** $P(\Omega)=1$. 3. **Countable additivity:** for pairwise disjoint $A_1,A_2,\ldots$, $$ P\!\left(\bigcup_{i=1}^{\infty}A_i\right)=\sum_{i=1}^{\infty}P(A_i). $$ The third axiom does more than the finite addition rule: it constrains probabilities of limiting events. It is the bridge from a finite coin-toss model to probability distributions and convergence results. ## Consequences worth deriving Because $\Omega=A\mathbin{\dot\cup}A^c$, $$ 1=P(\Omega)=P(A)+P(A^c),\qquad P(A^c)=1-P(A). $$ If $A\subseteq B$, then $B=A\mathbin{\dot\cup}(B\setminus A)$, so $P(A)\leq P(B)$. For arbitrary $A,B$, split their union into disjoint pieces: $$ P(A\cup B)=P(A)+P(B)-P(A\cap B). $$ In particular, disjoint events have $P(A\cap B)=0$, but events of probability zero are not necessarily impossible. A continuously distributed measurement can equal one exact value with probability zero while still producing some exact value. “Almost surely” means “except on a probability-zero event,” not “logically always.” # A sequencing example Suppose one read is selected uniformly from a finite library of $N$ reads. Let $A$ be “the read maps to gene $g$” and $B$ be “the read has a base-quality score above 30 at a chosen position.” The elementary outcome is the selected read, and $P(A)=|A|/N$. If low-quality reads are disproportionately unmapped, $A$ and $B$ overlap less than expected under independence. The inclusion--exclusion identity prevents double-counting reads satisfying both criteria. The independent experimental unit here is a sampled read for this sampling calculation; it is *not* a biological replicate for inference about expression across organisms. # Assumptions and common pitfalls | Pitfall | Why it fails | Better practice | |---|---|---| | Treating every subset of an infinite space as an event | A coherent probability measure cannot in general be assigned to all subsets | State the sigma-algebra or use standard Borel events | | Equating disjointness with independence | Disjoint non-null events cannot occur together, whereas independent events may | Check $P(A\cap B)=P(A)P(B)$ for independence | | Reading $P(A)=0$ as impossibility | Continuous models assign zero probability to individual values | Use “almost surely” precisely | | Ignoring the sampling mechanism | Probabilities refer to a specified random experiment | Define the unit, sampling frame, and source of randomness | # Connections to the Knowledge Map - [Counting rules and combinatorial probability](counting-rules-and-combinatorial-probability.qmd) uses finite sample spaces to calculate probabilities. - [Probability distributions](distributions.qmd) collects common probability models built on these axioms. - [Statistics overview](../index.qmd) places this foundation within the wider knowledge map. # Take-home Messages 1. A probability model is $(\Omega,\mathcal F,P)$, not merely a list of plausible outcomes. 2. Events are measurable sets; set algebra supplies the logic of probability statements. 3. Countable additivity is what makes probability compatible with limits. 4. A probability-zero event can occur; it is only negligible under the model. # References - Kolmogorov, A. N. *Foundations of the Theory of Probability*. Chelsea, 1950. - Billingsley, P. *Probability and Measure*, 3rd ed. Wiley, 2012.

Logic	Event operation
not \(A\)	complement \(A^c\)
\(A\) and \(B\)	intersection \(A\cap B\)
\(A\) or \(B\)	union \(A\cup B\)
exactly one of \(A,B\)	symmetric difference \(A\triangle B\)