Sample Spaces and Events

Suppose you toss a coin three times and ask: what is the chance of getting more heads than tails? Before any number can be attached to this question, we must first pin down some things with total precision. We must define what the possible outcomes even are. We must also define what it means to "get more heads than tails" as a mathematical object. This is the quiet, foundational work that makes all of probability theory possible. It translates a vague experiment into a precise universe of outcomes. It also translates a vague event description into a subset of that universe. This translation matters because probability is, at its core, a theory of measurement on sets. Once an experiment is encoded as a sample space, events are encoded as subsets of it. Then the entire machinery of set theory becomes available. This includes unions, intersections, complements, and De Morgan's laws. These tools let us reason about combinations of events: "heads on the first toss or fewer than two heads total," "not both A and B," "at least one of three events." Without this encoding, such compound statements would be linguistic puzzles. With it, they become algebra. The big picture is this: probability theory sits on top of two ingredients. These are a set of outcomes and a rule for assigning numbers (probabilities). This rule assigns numbers to certain subsets of that set. This article is entirely about the first ingredient and its bookkeeping. It covers outcomes, sample spaces, events, and the algebra of events. It deliberately postpones the question of how numbers get assigned. That question belongs to The Axioms of Probability. Here, the goal is to make the set-theoretic scaffolding airtight. Every later theorem about probability inherits any looseness left at this stage. By the end of this article you will be able to do several things. You will construct sample spaces for finite, countable, and continuous experiments. You will express complex verbal event descriptions as precise set expressions. You will manipulate events correctly using union, intersection, complement, and set difference. You will recognize mutually exclusive and exhaustive events. You will also prove and apply De Morgan's laws for events. These laws are indispensable in every computation that follows in Combinatorial Probability.

Key takeaways
  • A sample space Ω\Omega is the set of all possible outcomes of an experiment; each outcome ω∈Ω\omega \in \Omega is mutually exclusive with every other.
  • An event is any subset A⊆ΩA \subseteq \Omega; it "occurs" exactly when the realized outcome ω\omega lies in AA.
  • Set operations translate verbal logic directly: "or" is union, "and" is intersection, "not" is complement.
  • Two events are mutually exclusive (disjoint) if A∩B=∅A \cap B = \varnothing; they cannot both occur on the same trial.
  • De Morgan's laws — (A∪B)c=Ac∩Bc(A\cup B)^c = A^c \cap B^c and (A∩B)c=Ac∪Bc(A\cap B)^c = A^c \cup B^c — let you rewrite "not (this or that)" and "not (this and that)" correctly, avoiding a classic reasoning error.
  • Key idea: encoding an experiment as a set Ω\Omega and events as subsets of Ω\Omega turns probability into set algebra, the load-bearing structure for everything that follows.

Where this lives today

The sample-space/event formalism is the universal starting point of every quantitative field touching uncertainty. In statistics, hypothesis tests and confidence intervals are built by specifying the sample space of possible data and defining rejection regions as events. In statistical mechanics (physics), the "microstates" of a system form a sample space, and macroscopic observables correspond to events (unions of microstates with a shared property), directly paralleling the coin/spinner constructions here. In computer science, randomized algorithms and average-case analysis model all possible random choices (coin flips, hash collisions) as a sample space, with "algorithm fails" or "collision occurs" as events whose probability is bounded. In cryptography, security proofs model an adversary's random coins and the key-generation randomness as part of a sample space, and security is phrased as a bound on the probability of the "adversary wins" event. In actuarial science and finance, the space of possible future economic scenarios is the sample space, and claims, defaults, or losses are modeled as events, forming the basis for risk pricing built later on Random Variables defined on that same space. In quantum information, event algebras generalize to projections on Hilbert spaces, but the classical Boolean event algebra developed here is the base case that motivates the generalization.

Ask the tutor: How does the choice of sample space (ordered vs. unordered outcomes) affect probability calculations in a sampling-without-replacement problem?

DefinitionSample Space, Outcome, and Event

Let an experiment be any procedure with an uncertain result that can nonetheless be described by listing its possible results in advance. The sample space Ω\Omega of an experiment is the set of all possible outcomes of the experiment, subject to two requirements: 1. Exhaustiveness: every possible result of the experiment corresponds to exactly one element of Ω\Omega. 2. Mutual exclusivity of outcomes: the elements of Ω\Omega are pairwise distinct — on any single performance of the experiment, exactly one ω∈Ω\omega \in \Omega is realized. An element ω∈Ω\omega \in \Omega is called an outcome (or sample point, or elementary event). An event is any subset A⊆ΩA \subseteq \Omega (for uncountable Ω\Omega, one restricts to a suitable class of "measurable" subsets, but for finite or countable Ω\Omega every subset qualifies). The event AA is said to occur on a given performance of the experiment, with realized outcome ω\omega, if and only if ω∈A\omega \in A. Two special events are always available: the certain event Ω\Omega itself (which always occurs) and the impossible event ∅\varnothing (which never occurs).

Intuition

Think of Ω\Omega as a complete, non-overlapping partition of "everything that could happen" into indivisible atoms, the outcomes. An event is simply a region of this space — a collection of atoms grouped together because they share some property of interest. The event "the die shows an even number" is not itself an outcome; it is the set {2,4,6}\{2,4,6\}, a union of three outcomes. This shift — from asking "what happened?" (an outcome) to asking "did property PP happen?" (an event, a set) — is exactly what lets us combine properties using the Boolean connectives and, or, not, which correspond precisely to ∩\cap, ∪\cup, and complementation. Once events are sets, compound statements translate mechanically. "AA or BB occurs" means the realized outcome lies in A∪BA \cup B. "AA and BB both occur" means it lies in A∩BA \cap B. "AA does not occur" means it lies in Ac=Ω∖AA^c = \Omega \setminus A. Venn diagrams make these translations visually obvious, but the power of the framework is that it also works flawlessly for sample spaces too large or too abstract to draw, such as the space of all infinite sequences of coin flips.

Figure
Rolling a fair die: the sample space Ω={1,2,3,4,6}\Omega=\{1,2,3,4,6\}... wait, corrected below
Ω = {1, 2, 3, 4, 5, 6}, one die roll 1 2 3 4 5 6 A = "even outcome" = {2, 4, 6} Green circles are the outcomes composing the event A ⊆ Ω.

The mathematician behind it — placeholder note

(Section continues below in required order.)

TheoremDe Morgan's Laws for Events

Let Ω\Omega be a sample space and let A,B⊆ΩA, B \subseteq \Omega be events. Then:

(A∪B)c=Ac∩Bc,(A \cup B)^c = A^c \cap B^c,
(A∩B)c=Ac∪Bc,(A \cap B)^c = A^c \cup B^c,

where Xc:=Ω∖XX^c := \Omega \setminus X denotes the complement of XX relative to Ω\Omega. More generally, for any index set II (possibly infinite) and events {Ai}i∈I\{A_i\}_{i \in I},

(⋃i∈IAi)c=⋂i∈IAic,(⋂i∈IAi)c=⋃i∈IAic.\left(\bigcup_{i \in I} A_i\right)^c = \bigcap_{i \in I} A_i^c, \qquad \left(\bigcap_{i \in I} A_i\right)^c = \bigcup_{i \in I} A_i^c.
Proof

We prove the general infinite-index version; the two-event case is the special case I={1,2}I = \{1,2\}.

Technique: double inclusion (mutual subset containment) via the definitions of complement, union, and intersection as sets of outcomes satisfying quantified conditions. This is a purely set-theoretic argument and uses no probability axiom, so it cannot be circular with respect to anything in The Axioms of Probability.

Part 1: (⋃i∈IAi)c=⋂i∈IAic\left(\bigcup_{i \in I} A_i\right)^c = \bigcap_{i \in I} A_i^c. (⊆\subseteq) Let ω∈(⋃i∈IAi)c\omega \in \left(\bigcup_{i\in I} A_i\right)^c. By definition of complement, ω∈Ω\omega \in \Omega and ω∉⋃i∈IAi\omega \notin \bigcup_{i \in I} A_i. By definition of union, "ω∈⋃iAi\omega \in \bigcup_i A_i" means "there exists i∈Ii \in I with ω∈Ai\omega \in A_i." Negating this existential statement gives: "for all i∈Ii \in I, ω∉Ai\omega \notin A_i." For each such ii, since ω∈Ω\omega \in \Omega and ω∉Ai\omega \notin A_i, we have ω∈Aic\omega \in A_i^c. As this holds for every i∈Ii \in I, by definition of intersection ω∈⋂i∈IAic\omega \in \bigcap_{i \in I} A_i^c. (⊇\supseteq) Let ω∈⋂i∈IAic\omega \in \bigcap_{i \in I} A_i^c. By definition of intersection, ω∈Aic\omega \in A_i^c for every i∈Ii \in I, i.e. ω∈Ω\omega \in \Omega and ω∉Ai\omega \notin A_i for every i∈Ii \in I. Hence the statement "there exists i∈Ii \in I with ω∈Ai\omega \in A_i" is false, i.e. ω∉⋃i∈IAi\omega \notin \bigcup_{i \in I} A_i. Combined with ω∈Ω\omega \in \Omega, this gives ω∈(⋃i∈IAi)c\omega \in \left(\bigcup_{i \in I} A_i\right)^c. Both inclusions hold, so the two sets are equal, proving Part 1.

Part 2: (⋂i∈IAi)c=⋃i∈IAic\left(\bigcap_{i \in I} A_i\right)^c = \bigcup_{i \in I} A_i^c. Apply Part 1 to the events {Aic}i∈I\{A_i^c\}_{i \in I} in place of {Ai}i∈I\{A_i\}_{i\in I}, using the involution property (Xc)c=X(X^c)^c = X for any X⊆ΩX \subseteq \Omega (immediate from the definition of relative complement, since ω∈(Xc)c  ⟺  ω∈Ω,ω∉Xc  ⟺  ω∈Ω,¬(ω∈Ω∧ω∉X)  ⟺  ω∈X\omega \in (X^c)^c \iff \omega\in\Omega, \omega \notin X^c \iff \omega \in \Omega, \neg(\omega\in\Omega \wedge \omega \notin X) \iff \omega \in X, using ω∈Ω\omega \in \Omega throughout). Part 1 with AiA_i replaced by AicA_i^c gives

(⋃i∈IAic)c=⋂i∈I(Aic)c=⋂i∈IAi.\left(\bigcup_{i\in I} A_i^c\right)^c = \bigcap_{i \in I} (A_i^c)^c = \bigcap_{i \in I} A_i.

Taking complements of both sides (and using involution again on the left) yields

⋃i∈IAic=(⋂i∈IAi)c,\bigcup_{i \in I} A_i^c = \left(\bigcap_{i \in I} A_i\right)^c,

which is Part 2. ■\blacksquare

∎
CorollaryFinite two-event case, restated

For events A,B⊆ΩA, B \subseteq \Omega: "AA or BB does not occur" is equivalent to "AA does not occur and BB does not occur," and "not both AA and BB occur" is equivalent to "AA does not occur or BB does not occur." This is immediate from the Theorem with I={1,2}I=\{1,2\}.

Worked examples

Example 1 — A biased spinner with three unequal regions, plus a fair coin. An experiment consists of two stages: (i) spin a pointer on a disk divided into three unequal regions labeled RR (red), GG (green), BB (blue), and record which region it stops in; (ii) independently flip a fair coin and record Heads (HH) or Tails (TT). We are asked to build the sample space and express three events as subsets. Step 1: Construct Ω\Omega. Each performance of the experiment produces an ordered pair (spinner result, coin result). Since the spinner has 3 outcomes and the coin has 2, and every combination is a genuinely distinct, exhaustive, mutually exclusive possibility, the sample space is the Cartesian product

Ω={R,G,B}×{H,T}={RH,RG-typo-none,RT,GH,GT,BH,BT}.\Omega = \{R,G,B\} \times \{H,T\} = \{RH, RG\text{-typo-none}, RT, GH, GT, BH, BT\}.

Listing carefully (six elements, 3×2=63\times2=6):

Ω={RH,RT,GH,GT,BH,BT}.\Omega = \{RH, RT, GH, GT, BH, BT\}.

Each of these six symbols is one outcome — an atom of the sample space. Step 2: Verify the sample space axioms. Exhaustiveness holds: every possible joint result is one of these six pairs, since the spinner must land in exactly one of R,G,BR,G,B and the coin must show exactly one of H,TH,T. Mutual exclusivity holds: no performance of the experiment can simultaneously realize two different pairs, e.g. it cannot be both RHRH and GTGT at once. Step 3: Translate three events into subsets.

  • AA = "the spinner lands on red." As a subset of Ω\Omega: A={RH,RT}A = \{RH, RT\}, since these are precisely the outcomes whose first coordinate is RR.
  • BB = "the coin shows heads." Then B={RH,GH,BH}B = \{RH, GH, BH\}.
  • CC = "the spinner does not land on blue, or the coin shows tails." Parsing carefully: "spinner not blue" is the event {R,G}×{H,T}={RH,RT,GH,GT}\{R,G\}\times\{H,T\} = \{RH,RT,GH,GT\}; "coin shows tails" is {RT,GT,BT}\{RT, GT, BT\}; the word "or" means union. So
C={RH,RT,GH,GT}∪{RT,GT,BT}={RH,RT,GH,GT,BT}.C = \{RH,RT,GH,GT\} \cup \{RT,GT,BT\} = \{RH, RT, GH, GT, BT\}.

Note BH∉CBH \notin C: it is the only outcome excluded, since it is the only one where the spinner is blue and the coin is not tails — check: is BH∈CBH \in C? Spinner is blue (fails "not blue"), coin is heads (fails "tails"), so both disjuncts fail, hence BH∉CBH \notin C, confirming ∣C∣=5|C| = 5. Step 4: Compute A∩BA \cap B and interpret. A∩B={RH,RT}∩{RH,GH,BH}={RH}A \cap B = \{RH,RT\} \cap \{RH,GH,BH\} = \{RH\}. In words: "spinner red and coin heads" happens only via the single outcome RHRH, which matches intuition directly. Step 5: Compute (A∪B)c(A\cup B)^c via De Morgan and verify by direct listing. First A∪B={RH,RT}∪{RH,GH,BH}={RH,RT,GH,BH}A \cup B = \{RH,RT\} \cup \{RH,GH,BH\} = \{RH, RT, GH, BH\}, so (A∪B)c=Ω∖{RH,RT,GH,BH}={GT,BT}(A\cup B)^c = \Omega \setminus \{RH,RT,GH,BH\} = \{GT, BT\}. Now via De Morgan's Laws (proved above): Ac=Ω∖{RH,RT}={GH,GT,BH,BT}A^c = \Omega\setminus\{RH,RT\} = \{GH,GT,BH,BT\}, and Bc=Ω∖{RH,GH,BH}={RT,GT,BT}B^c = \Omega \setminus \{RH,GH,BH\} = \{RT,GT,BT\}. Then

Ac∩Bc={GH,GT,BH,BT}∩{RT,GT,BT}={GT,BT}.A^c \cap B^c = \{GH,GT,BH,BT\} \cap \{RT,GT,BT\} = \{GT, BT\}.

This matches (A∪B)c={GT,BT}(A\cup B)^c = \{GT,BT\} exactly, confirming the theorem numerically in this instance.

Example 2 — Sampling without replacement from an urn, with a subtle event description. An urn contains 4 balls labeled 1,2,3,41,2,3,4; balls 1,21,2 are red and balls 3,43,4 are blue. Two balls are drawn without replacement and without regard to order (i.e., we record only the unordered pair drawn, not which was drawn first). Step 1: Construct Ω\Omega. Since order is not recorded, an outcome is an unordered pair of distinct labels from {1,2,3,4}\{1,2,3,4\}. There are (42)=6\binom{4}{2}=6 such pairs:

Ω={{1,2}, {1,3}, {1,4}, {2,3}, {2,4}, {3,4}}.\Omega = \big\{\{1,2\},\ \{1,3\},\ \{1,4\},\ \{2,3\},\ \{2,4\},\ \{3,4\}\big\}.

Each is genuinely one outcome (mutual exclusivity: the drawn pair cannot simultaneously be {1,2}\{1,2\} and {3,4}\{3,4\}; exhaustiveness: every possible draw is one of these six pairs). Step 2: Define events.

  • DD = "both balls drawn are the same color." Checking each pair against colors (1,21,2 red; 3,43,4 blue): {1,2}\{1,2\} same color (red-red) ✓; {1,3}\{1,3\} mixed ✗; {1,4}\{1,4\} mixed ✗; {2,3}\{2,3\} mixed ✗; {2,4}\{2,4\} mixed ✗; {3,4}\{3,4\} same color (blue-blue) ✓. So
D={{1,2},{3,4}}.D = \big\{\{1,2\}, \{3,4\}\big\}.
  • EE = "ball 1 is drawn." The pairs containing label 1 are {1,2},{1,3},{1,4}\{1,2\}, \{1,3\}, \{1,4\}, so E={{1,2},{1,3},{1,4}}E = \{\{1,2\},\{1,3\},\{1,4\}\}.
  • FF = "at least one red ball is drawn." A pair fails to contain a red ball only if both its elements are blue, i.e. only for {3,4}\{3,4\}. So F=Ω∖{{3,4}}={{1,2},{1,3},{1,4},{2,3},{2,4}}F = \Omega \setminus \{\{3,4\}\} = \{\{1,2\},\{1,3\},\{1,4\},\{2,3\},\{2,4\}\}, a set of size 5.

Step 3: A compound event and mutual exclusivity check. Consider D∩ED \cap E: pairs that are same-color and contain ball 1. From D={{1,2},{3,4}}D = \{\{1,2\},\{3,4\}\}, only {1,2}\{1,2\} contains ball 1 (since {3,4}\{3,4\} does not). So D∩E={{1,2}}D \cap E = \{\{1,2\}\}, a single outcome — both balls red, and specifically ball 1 is one of them. Step 4: Are DD and FcF^c mutually exclusive? Fc={{3,4}}F^c = \{\{3,4\}\} (complement of FF found in Step 2, size 6−5=16-5=1). Is D∩Fc=∅D \cap F^c = \varnothing? We have D={{1,2},{3,4}}D = \{\{1,2\},\{3,4\}\} and Fc={{3,4}}F^c = \{\{3,4\}\}, so D∩Fc={{3,4}}≠∅D \cap F^c = \{\{3,4\}\} \neq \varnothing. They are not mutually exclusive — indeed {3,4}\{3,4\} (no red ball drawn) is simultaneously an instance of "same color" and "no red ball," which makes sense since two blue balls are automatically same-colored. Step 5: Apply De Morgan explicitly. Compute (D∪F)c(D \cup F)^c. First, D∪F={{1,2},{3,4}}∪{{1,2},{1,3},{1,4},{2,3},{2,4}}=ΩD \cup F = \{\{1,2\},\{3,4\}\} \cup \{\{1,2\},\{1,3\},\{1,4\},\{2,3\},\{2,4\}\} = \Omega (every pair is listed once we include {3,4}\{3,4\} from DD and the rest from FF). So (D∪F)c=∅(D\cup F)^c = \varnothing. Checking via De Morgan: Dc=Ω∖D={{1,3},{1,4},{2,3},{2,4}}D^c = \Omega\setminus D = \{\{1,3\},\{1,4\},\{2,3\},\{2,4\}\} and Fc={{3,4}}F^c = \{\{3,4\}\}; then Dc∩Fc={{1,3},{1,4},{2,3},{2,4}}∩{{3,4}}=∅D^c \cap F^c = \{\{1,3\},\{1,4\},\{2,3\},\{2,4\}\} \cap \{\{3,4\}\} = \varnothing, matching exactly.

CounterexampleWhy "exhaustive and mutually exclusive outcomes" cannot be dropped

Drop the mutual exclusivity of outcomes requirement in the definition of a sample space, and let outcomes be allowed to overlap as sets. Consider rolling a single fair die, but suppose someone (incorrectly) proposes the outcome list

Ω′={ "even","odd","greater than 4" }\Omega' = \{\,\text{"even"}, \text{"odd"}, \text{"greater than 4"}\,\}

as the sample space, intending these three labels as the "outcomes." This fails badly. The labels "even" and "greater than 4" are not mutually exclusive: rolling a 6 makes both true simultaneously, so it is impossible to say the experiment "resulted in" a single, unambiguous element of Ω′\Omega'. Rolling a 6 does not correspond to exactly one point of Ω′\Omega'; it corresponds to two ("even" and "greater than 4" both hold), while rolling a 5 corresponds to a different two ("odd" and "greater than 4"). Consequently the entire event/subset formalism collapses: the event A={"even"}A=\{\text{"even"}\} was supposed to represent "the die shows an even number," but since "even" and "greater than 4" can occur together, treating them as distinct atomic outcomes double-counts the roll of 66 across two different "elementary events." Any later probability assignment (say, via counting outcomes) built on Ω′\Omega' would be internally inconsistent — for instance, naive equally-likely counting would assign probability 1/31/3 to each of the three labels, summing to 11, even though a single roll of 6 makes two of the three labels simultaneously true, which is incoherent for a partition-based model. The correct sample space is Ω={1,2,3,4,5,6}\Omega = \{1,2,3,4,5,6\}, in which every outcome is a single face value, genuinely mutually exclusive from every other, and the properties "even," "odd," "greater than 4" are then correctly recovered as the events (subsets) {2,4,6}\{2,4,6\}, {1,3,5}\{1,3,5\}, {5,6}\{5,6\} respectively — sets that are allowed to overlap, because events, unlike outcomes, need not be disjoint. This is exactly why the definition insists mutual exclusivity is a property of outcomes, not of events.

Scope, edge cases & common traps

SituationCorrect treatmentCommon trap
Continuous experiment (e.g. a dart hitting [0,1][0,1])Ω=[0,1]\Omega=[0,1]; only a suitable class of subsets (Borel sets) are treated as eventsAssuming every subset is automatically an event; in continuous spaces this can fail to admit a consistent probability (non-measurable sets)
Sequential/infinite experiments (e.g. flip a coin forever)Ω={H,T}N\Omega = \{H,T\}^{\mathbb N}, the set of infinite sequencesTrying to list Ω\Omega by enumeration; it is uncountable, so outcomes must be described structurally, not listed
Compound experimentsΩ\Omega = Cartesian product of component sample spaces (Example 1)Forgetting that order/labeling conventions (ordered vs. unordered draws) change ∣Ω∣|\Omega| and must match the physical description
"At least one of A,BA,B"A∪BA \cup BConfusing with A∩BA \cap B, or with exclusive-or
"Neither AA nor BB"(A∪B)c=Ac∩Bc(A\cup B)^c = A^c \cap B^cWriting Ac∪BcA^c \cup B^c instead — the classic De Morgan sign error
Mutually exclusive vs. independentMutual exclusivity is a set relation (A∩B=∅A\cap B=\varnothing); independence is a probabilistic relation defined laterTreating disjointness and independence as synonyms — in fact disjoint events with positive probability are always dependent

The mathematician behind it

The formal identification of "event" with "subset of a sample space," and the accompanying algebra of unions, intersections, and complements, crystallized through the joint development of set theory (Georg Cantor, late 19th century) and modern axiomatic probability. While the qualitative idea of combining chances via logical connectives dates to Pascal, Fermat, and Huygens in the 1650s–1660s, the precise set-theoretic scaffolding used throughout this article is due to the 20th-century measure-theoretic reformulation of probability.

Origins · 1 minAndrey Nikolaevich Kolmogorov and how this was discovered
AN
Andrey Nikolaevich Kolmogorov
1903–1987 · Soviet mathematician, Moscow School of Probability

Kolmogorov's 1933 monograph Grundbegriffe der Wahrscheinlichkeitsrechnung ("Foundations of the Theory of Probability") gave probability its modern axiomatic footing by explicitly building it atop measure theory: a sample space Ω\Omega, a σ\sigma-algebra of events, and a probability measure satisfying countable additivity. Before Kolmogorov, "probability" lacked a universally agreed formal object; his framework — sample spaces and events as sets, exactly as presented here — became the unquestioned foundation for all subsequent probability theory, including The Axioms of Probability built directly on top of it.

History
- 1654 — Pascal and Fermat correspond on the "problem of points," implicitly enumerating equally likely cases (proto-sample-spaces).
- 1657 — Huygens publishes De Ratiociniis in Ludo Aleae, the first printed treatise on probability, reasoning via enumerated outcomes.
- 1895–1897 — Cantor's foundational papers establish rigorous set theory, supplying the language of subsets and operations later used for events.
- 1900 — Hilbert's sixth problem calls for the axiomatization of probability alongside physics, motivating a formal foundation.
- 1933 — Kolmogorov's Grundbegriffe fixes the modern sample-space/event/measure framework used verbatim today.
What exactly distinguishes an "outcome" from an "event"?

An outcome is a single, indivisible element ω\omega of the sample space Ω\Omega — the most granular possible description of what happened, such as "the die showed a 4." An event is any subset A⊆ΩA \subseteq \Omega, a collection of outcomes grouped by a shared property, such as "the die showed an even number," which is the subset {2,4,6}\{2,4,6\}. Every outcome ω\omega determines a special one-element event {ω}\{\omega\}, sometimes called an elementary event, but not every event is an outcome — most events (like "even") bundle several outcomes together. The distinction matters because probabilities are formally assigned to events (subsets), and the probability of an outcome is recovered as the probability of its singleton event.

Why must outcomes be mutually exclusive and exhaustive, but events need not be?

Outcomes model "what actually happens" on one performance of the experiment, and exactly one thing happens — hence outcomes must partition all possibilities without overlap (mutual exclusivity) while covering every possibility (exhaustiveness). Events, by contrast, are descriptive labels for properties of the outcome, and a single outcome can simultaneously satisfy several properties — e.g. rolling a 6 satisfies both "even" and "greater than 4." If events had to be disjoint, we could never express such natural overlapping descriptions, and compound statements built from "and"/"or" would be impossible to represent as sets, defeating the entire purpose of the framework, as the counterexample above demonstrates.

Can two different sample spaces describe the same experiment?

Yes, and this is common and legitimate. Rolling two dice can be modeled with the ordered sample space {1,…,6}2\{1,\dots,6\}^2 (36 outcomes, tracking which die is which) or, if the dice are physically indistinguishable and only the unordered result matters, with an unordered sample space of 21 outcomes. Both are valid provided the exhaustiveness and mutual-exclusivity requirements hold for the chosen description; the choice affects how outcomes are counted but, when probabilities are assigned consistently with the physical randomness, does not change the probability of any well-posed event. Choosing the sample space that matches what is physically observable and distinguishable is essential to avoid double-counting or under-counting later in Combinatorial Probability.

What happens when the sample space is infinite or continuous?

When Ω\Omega is countably infinite (e.g. "number of coin flips until first head," Ω={1,2,3,… }\Omega = \{1,2,3,\dots\}) the same subset-based definition of event applies without change: any subset is an event. When Ω\Omega is uncountable (e.g. a dart landing anywhere in [0,1][0,1]), it turns out that not every subset can be consistently assigned a probability; one restricts attention to a well-behaved collection of subsets called a σ\sigma-algebra (typically the Borel sets). This refinement is developed fully in The Axioms of Probability; for finite and countable spaces, which dominate introductory examples, this subtlety never arises and every subset is automatically an event.

Is ∅\varnothing really an event, and can it ever "occur"?

Yes, ∅\varnothing is formally an event because it is a subset of Ω\Omega (the empty subset). By definition, an event AA occurs when the realized outcome ω\omega satisfies ω∈A\omega \in A; since no ω\omega can belong to ∅\varnothing, the event ∅\varnothing never occurs, on any performance of the experiment. It represents "the impossible event" and plays the role of the additive identity once probabilities are introduced (its probability will always be 00), while Ω\Omega itself is the "certain event," always occurring since every outcome trivially belongs to Ω\Omega.

How do "mutually exclusive" and "exhaustive" apply to events, as opposed to outcomes?

A collection of events A1,A2,…A_1, A_2, \dots is called mutually exclusive if Ai∩Aj=∅A_i \cap A_j = \varnothing for every i≠ji \neq j, meaning no two of them can occur simultaneously — this generalizes the built-in mutual exclusivity of single outcomes to arbitrary events. The collection is called exhaustive if ⋃iAi=Ω\bigcup_i A_i = \Omega, meaning some event in the collection must occur. When a collection of events is both mutually exclusive and exhaustive, it is called a partition of Ω\Omega; the set of all singleton outcomes {{ω}:ω∈Ω}\{\{\omega\}: \omega \in \Omega\} is the finest possible partition, but coarser partitions (like "even" vs. "odd" for a die) are equally legitimate and often more useful.

Why does this article avoid assigning any actual probabilities?

Because sample spaces and events are purely set-theoretic objects — they specify what can happen and how to describe combinations of possibilities, with no reference yet to how likely anything is. Probabilities are a separate ingredient, a function from events to numbers in [0,1][0,1] satisfying specific axioms (non-negativity, normalization P(Ω)=1P(\Omega)=1, and countable additivity over disjoint events). Building that function rigorously, and deriving its consequences, is the subject of The Axioms of Probability; keeping the two stages strictly separate is what makes each stage's logic transparent and prevents subtle circular reasoning about "equally likely outcomes" before the axioms justify it.

Where this leads