Every mathematical structure rests on the same primitive foundation: the **set**. [Groups](/page/Group) are sets with a binary operation; [topological](/page/Topology) spaces are sets equipped with a collection of designated subsets; measure spaces are sets with a $\sigma$-algebra and a measure. When a mathematician writes "$f: X \to Y$ is [continuous](/page/Continuity)," the objects $X$, $Y$, and $f$ are all defined in terms of sets. Yet the concept of a set is not as straightforward as it first appears. The naive intuition — "a set is any collection of objects" — leads to contradictions, and building a rigorous foundation requires care. This page develops the notion of a set from its naive origins through the paradoxes that forced an axiomatic treatment, defines the fundamental operations on sets, and establishes the first non-trivial results about their structure.

[motivation] ## Motivation ### The Naive Conception Georg Cantor, in his 1895 *Beiträge*, described a set as "a collection into a whole of definite, distinct objects of our perception or our thought." Under this conception, every property $P$ determines a set $\{x : P(x)\}$ — the collection of all objects satisfying $P$. This principle, called **unrestricted comprehension**, is powerful: it produces the natural numbers $\{1, 2, 3, \ldots\}$, the real line $\mathbb{R}$, the set of all continuous [functions](/page/Function) on $[0,1]$, and indeed the set of all sets. For most everyday mathematical purposes, this approach works perfectly well, and mathematicians routinely write set-builder notation $\{x : P(x)\}$ without incident. ### The Catastrophe: Self-Reference In 1901, Bertrand Russell observed that unrestricted comprehension allows a devastating self-reference. Consider the property $P(x) :\Leftrightarrow x \notin x$ — the property of not being a member of oneself. Under unrestricted comprehension, $R = \{x : x \notin x\}$ is a set. But is $R$ a member of itself? If $R \in R$, then by definition $R \notin R$; if $R \notin R$, then $R$ satisfies the defining property and so $R \in R$. Both cases lead to a contradiction. The naive conception of set, which seemed so natural, is inconsistent. This is not an isolated curiosity. The same self-referential mechanism produces other paradoxes: the Burali-Forti paradox (the "set of all ordinals" cannot be an ordinal), and Cantor's own paradox (the "set of all sets" leads to a contradiction with his theorem on power sets). The lesson is that collections which are "too large" or "too self-referential" cannot be sets. ### The Axiomatic Resolution The response to these paradoxes was to abandon the idea that every property defines a set and replace it with a carefully chosen list of axioms specifying exactly which set-forming operations are permitted. The most widely adopted system is **Zermelo-Fraenkel set theory with the Axiom of Choice** (ZFC). In ZFC, "set" is a primitive undefined term (just as "point" is in Euclidean geometry), and the axioms govern what one may do with sets: form pairs, take unions, separate elements by a property *within an existing set*, take power sets, and so on. Crucially, the Axiom Schema of Separation replaces unrestricted comprehension: given an existing set $A$ and a property $P$, one may form $\{x \in A : P(x)\}$, but not $\{x : P(x)\}$ in general. This seemingly small restriction blocks Russell's paradox while preserving all the set constructions that mathematics actually uses. [/motivation]

## The Axioms of ZFC

Working within ZFC, we do not define what a set *is* — the term is primitive. Instead, the axioms specify what sets exist and what operations produce new sets from old ones. The single primitive relation is **membership**: $x \in A$ means "$x$ is an element of $A$." All other notions (subset, union, function, cardinality) are defined from membership alone.

We state the axioms informally here; each can be made precise in the language of first-order logic with a single binary relation symbol $\in$.

The first axiom ensures that sets are determined entirely by their elements — two sets with the same members are the same set, regardless of how they were described or constructed.

[definition:Axiom Of Extensionality] **Axiom of Extensionality.** Two sets are equal if and only if they have exactly the same elements: \begin{align*} \forall A\, \forall B\, \bigl(A = B \iff \forall x\, (x \in A \iff x \in B)\bigr). \end{align*} [/definition]

Extensionality means that a set has no internal structure beyond its elements: $\{1, 2, 3\}$, $\{3, 1, 2\}$, and $\{1, 2, 2, 3\}$ are all the same set, because they have the same members. There is no notion of order or multiplicity.

The next axiom guarantees that the set-theoretic universe is not vacuous — at least one set exists.

[definition:Axiom Of Empty Set] **Axiom of the Empty Set.** There exists a set with no elements, denoted $\varnothing$: \begin{align*} \exists\, \varnothing\; \forall x\; (x \notin \varnothing). \end{align*} By extensionality, this set is unique. [/definition]

The existence of $\varnothing$ is the starting point from which all of ZFC's constructions proceed. From the empty set alone, the pairing and union axioms can build the finite ordinals $0 = \varnothing$, $1 = \{\varnothing\}$, $2 = \{\varnothing, \{\varnothing\}\}$, and so on.

The axiom that directly resolves Russell's paradox is separation. Unrestricted comprehension said: for any property $P$, the collection $\{x : P(x)\}$ is a set. Separation weakens this to: for any property $P$ and any *existing* set $A$, the collection $\{x \in A : P(x)\}$ is a set. The difference is that one can only carve out subsets of sets that already exist — one cannot conjure a set from a property alone.

[definition:Axiom Schema Of Separation] **Axiom Schema of Separation (Aussonderung).** For every set $A$ and every property $P(x)$ expressible in the language of set theory, there exists a set \begin{align*} B = \{x \in A : P(x)\}. \end{align*} [/definition]

With separation, the attempt to form $R = \{x : x \notin x\}$ fails: one can only form $\{x \in A : x \notin x\}$ for some pre-existing set $A$, which is a harmless subset of $A$ (and the contradiction shows only that $R_A \notin A$, not that $R_A$ doesn't exist).

### The Remaining Axioms

The other axioms of ZFC provide the tools for building new sets:

- **Pairing:** for any sets $a$ and $b$, the set $\{a, b\}$ exists. - **Union:** for any set $\mathcal{F}$ of sets, the union $\bigcup \mathcal{F} = \{x : \exists A \in \mathcal{F},\; x \in A\}$ exists. - **Power Set:** for any set $A$, the power set $\mathcal{P}(A) = \{B : B \subseteq A\}$ exists. - **Infinity:** there exists an infinite set (specifically, an inductive set containing $\varnothing$ and closed under the successor operation $x \mapsto x \cup \{x\}$). - **Replacement (Schema):** the image of a set under any definable function is again a set. - **Regularity (Foundation):** every nonempty set $A$ contains an element disjoint from $A$; equivalently, there is no infinite descending chain $\cdots \in a_2 \in a_1 \in a_0$. This rules out pathologies like $a \in a$. - **Choice:** for every family of nonempty sets, there exists a function selecting one element from each set.

Together, these axioms are strong enough to develop all of standard mathematics — number systems, functions, topological spaces, measure theory, algebra — while avoiding the paradoxes of naive set theory.

## Russell's Paradox

The following result, already discussed in the motivation, makes precise exactly why unrestricted comprehension is untenable. It is the simplest and most famous of the set-theoretic paradoxes, and its proof is a direct diagonalisation argument — the same technique that reappears in [Cantor's theorem](/theorems/621).