Topos Theory — Elementary Toposes and Subobject Classifiers

An elementary topos is a cartesian closed category with finite limits that possesses a subobject classifier $Ω$ — a single object that internally represents the notion of "subset" or "truth value." Key result: every subobject of an object $X$ corresponds bijectively to a morphism $X \to Ω$ (its characteristic map), and $Ω$ carries the structure of an internal Heyting algebra, making the internal logic of a topos intuitionistic. Applies to: categorical logic, sheaf theory, algebraic geometry, set theory (forcing), and constructive mathematics. Who/when: Lawvere and Tierney, c. 1969–1970, building on Grothendieck's sheaf-theoretic toposes.

An elementary topos $E$ is a category that behaves enough like the category $Set$ to support a self-contained logic and mathematics, yet is defined by finitary categorical axioms: finite limits, exponentials, and a subobject classifier. The subobject classifier $Ω$ generalizes the two-element set ${true, false}$ ; in a topos the "truth values" form an internal Heyting algebra rather than a Boolean one, so the logic is generally intuitionistic. This matters because it unifies set theory, sheaf theory, and logic under one categorical roof: a model of (higher-order intuitionistic) logic is exactly a topos. The surprising fact is that from such spare axioms one recovers power objects, image factorizations, internal quantifiers, colimits, and a full higher-order type theory — all generated by the single classifying object $Ω$ .

The Core Insight

The defining move of topos theory is to internalize the membership relation. In $Set$ , a subset $S \subseteq X$ is the same data as its indicator function $χ_{S} : X \to {0, 1}$ . Topos theory takes this representability of subobjects as an axiom: there exists a universal object $Ω$ and a universal "element" $true : 1 \to Ω$ such that every monomorphism is a pullback of $true$ along a unique characteristic map. Once subobjects are represented by morphisms into $Ω$ , logical operations on subobjects become algebraic operations on $Ω$ , and the entire apparatus of logic becomes categorical.

Formal Statement

Subobject

For an object $X$ in a category $C$ , a subobject of $X$ is an equivalence class of monomorphisms $m : S ↣ X$ , where $m \sim m^{'}$ iff there is an isomorphism $φ$ with $m^{'} φ = m$ . The collection of subobjects is denoted $Sub (X)$ ; it is partially ordered by $[m] \leq [m^{'}]$ iff $m$ factors through $m^{'}$ .

Subobject Classifier

A subobject classifier in a category $C$ with finite limits is an object $Ω$ together with a monomorphism $true : 1 ↣ Ω$ (from the terminal object $1$ ) such that for every monomorphism $m : S ↣ X$ there is a unique morphism $χ_{m} : X \to Ω$ , the characteristic map, making the following square a pullback:

S ↓ m X! χ_{m} 1 ↓ true Ω

⟳The pullback condition says

S

is exactly the preimage

χ_{m}^{- 1} (true)

; uniqueness of

χ_{m}

encodes that subobjects are faithfully represented.

Elementary Topos

An elementary topos is a category $E$ that

has all finite limits (equivalently a terminal object and pullbacks),
is cartesian closed: for every object $X$ the functor $(-) \times X$ has a right adjoint $(-)^{X}$ (exponentials exist), and
has a subobject classifier $Ω$ .

The cartesian closure is expressed by the natural bijection

Hom_{E} (Z \times X, Y) ≅ Hom_{E} (Z, Y^{X}) .

⟳This is the currying isomorphism; it is natural in

Z

X

Y

and exhibits

(-)^{X} ⊣ (-) \times X

The functor $Sub : E^{op} \to Set$ sending $X \mapsto Sub (X)$ (with pullback as the action on morphisms) is then representable:

Sub (X) ≅ Hom_{E} (X, Ω), [m] ⟷ χ_{m} .

⟳By the universal property [1], pulling back

true

along

χ

recovers

m

, and distinct subobjects give distinct

χ

; representability is the Yoneda restatement of the classifier axiom.

Power Objects and Internal Hom into $\Omega$

In any elementary topos, every object $X$ has a power object $P (X) = Ω^{X}$ representing subobjects of $X \times (-)$ :

Sub (X \times Z) ≅ Hom_{E} (Z, Ω^{X}) .

There is a universal membership subobject $\in_{X} ↣ X \times Ω^{X}$ such that every relation arises by pullback along a unique classifying map $Z \to Ω^{X}$ .

⟳Combine [2] with

X \leftarrow Y

replaced by

Ω

, then apply [3]; the membership relation is

\in_{X} = (ev)^{- 1} (true)

where

ev : X \times Ω^{X} \to Ω

is evaluation.

$\Omega$ is an Internal Heyting Algebra

The object $Ω$ carries internal operations $\land, \lor, \Rightarrow: Ω \times Ω \to Ω$ and constants $true, false : 1 \to Ω$ making $(Sub (X), \leq)$ a Heyting algebra for every $X$ , naturally in $X$ . The Heyting implication satisfies the adjunction

a \land b \leq c ⟺ a \leq (b \Rightarrow c)

internally in each $Sub (X)$ .

⟳

\land

classifies the subobject

⟨ true, true ⟩ : 1 ↣ Ω \times Ω

;

\Rightarrow

classifies

{(p, q) : p \land q = p}

; the adjunction [5] is verified by transposing pullbacks.

Why It Holds / Derivation

Construction of $\wedge$ on $\Omega$ and the Heyting structure

We build the conjunction operation and show it induces meets on subobjects.

Step 1 — Construct $\land$ . Form the morphism

⟨ true, true ⟩ : 1 \to Ω \times Ω

. Because

1

is terminal and the composite

1 \to Ω \times Ω

is a monomorphism (it is split by either projection composed appropriately), it is a subobject of

Ω \times Ω

. By the classifier axiom [1] there is a unique map

\land = χ_{⟨ true, true ⟩} : Ω \times Ω \to Ω.

Step 2 — Meets of subobjects. Given

[m], [n] \in Sub (X)

with characteristic maps

χ_{m}, χ_{n} : X \to Ω

, form

⟨ χ_{m}, χ_{n} ⟩ : X \to Ω \times Ω

and compose with

\land

. The pullback of

true

along

\land \circ ⟨ χ_{m}, χ_{n} ⟩

is the simultaneous pullback of

true

along

χ_{m}

and along

χ_{n}

, i.e. the intersection

S \cap T

as a subobject of

X

. Hence

χ_{m \land n} = \land \circ ⟨ χ_{m}, χ_{n} ⟩, m \land n = m \cap n .

Step 3 — Order. Define

\leq

Ω

as the subobject

{(p, q) : p \land q = p} ↣ Ω \times Ω

, classified by some

\leq : Ω \times Ω \to Ω

. Under [3] this recovers the partial order on

Sub (X)

. Step 4 — Implication via the adjoint. The operation

(-) \land b : Sub (X) \to Sub (X)

preserves all meets and (using cartesian closure and that

Sub (X)

is a complete lattice — a consequence of finite limits plus

Ω^{X}

) has a right adjoint, the Heyting implication

b \Rightarrow (-)

, giving [5]. The internal implication

\Rightarrow: Ω \times Ω \to Ω

classifies the subobject

{(p, q) : p \leq q}

pulled into the diagonal. Step 5 — Distributivity and bottom.

false : 1 \to Ω

χ_{0 ↣ 1}

, the characteristic map of the initial object's unique mono into

1

. Distributivity of

\land

over

\lor

(where

\lor

is built from the image of

⟨ id, true ⟩ ⊔ ⟨ true, id ⟩

) follows from the adjunction in Step 4. Thus each

Sub (X)

is a Heyting algebra.

■

Why Heyting and not Boolean

The proof never produces a complemented lattice. In general $a \lor \neg a \neq = ⊤$ where $\neg a := (a \Rightarrow ⊥)$ . A topos is Boolean exactly when $Ω ≅ 1 + 1$ (equivalently the law of excluded middle holds internally). Generic toposes — e.g. sheaves on a non-discrete space — are not Boolean.

F. William Lawvere & Myles Tierney | 1937– / 1937–

Lawvere, following his 1963 thesis on functorial semantics, sought a purely categorical, elementary (first-order, finitary) axiomatization of "the category of sets" and of Grothendieck toposes. Working with Tierney around 1969–1970 at Dalhousie University, he isolated the subobject classifier as the crucial new ingredient, replacing the infinitary descent data of Grothendieck topologies with a single object $Ω$ and the operator $j : Ω \to Ω$ (a Lawvere–Tierney topology). This recast sheafification, forcing, and intuitionistic higher-order logic in a uniform categorical language, and turned topos theory into a foundation for mathematics.

Building Intuition

Think of $Ω$ as the object of truth values internal to the world $E$ . In $Set$ , $Ω = {0, 1}$ and $χ_{S}$ is the indicator function — asking "is $x \in S$ ?" returns a value in ${$ yes, no $}$ . In a sheaf topos $Sh (X)$ , the truth values are the open sets of $X$ : the "truth value of $x \in S$ " is the largest open set on which the property holds. So truth becomes local and stagewise, not global. This is why excluded middle fails: a property can hold on a dense open set without holding everywhere, so " $P$ or not $P$ " need not be globally true. Power objects $Ω^{X}$ are the internal "power sets," and exponentials $Y^{X}$ are internal function sets, so a topos genuinely is a self-contained universe of (constructive) sets.

The subobject classifier pullback square: a subobject as the preimage of $\mathsf{true}$

Worked Examples

The classifier in $\mathbf{Set}$

Given:

E = Set

, take

X = {1, 2, 3, 4}

and the subset

S = {2, 4}

with inclusion

m : S ↣ X

. Setup:

1 = {*}

, and we claim

Ω = {0, 1}

with

true : * \mapsto 1

. Steps:

Define $χ_{m} : X \to {0, 1}$ by $χ_{m} (x) = 1$ if $x \in S$ , else $0$ . Explicitly $χ_{m} (1) = 0, χ_{m} (2) = 1, χ_{m} (3) = 0, χ_{m} (4) = 1$ .
Compute the pullback of $true$ along $χ_{m}$ : it is ${x : χ_{m} (x) = 1} = {2, 4} = S$ . ✓
Uniqueness: any $χ^{'}$ with this pullback must send $S \to 1$ and $X ∖ S \to 0$ ; this is forced, so $χ^{'} = χ_{m}$ . ✓
Heyting structure: $\land$ is multiplication on ${0, 1}$ , $\lor = max$ , $\neg0 = 1, \neg1 = 0$ . Here $Ω = 1 + 1$ , so $Set$ is Boolean and excluded middle holds.

Ω_{Set} = {0, 1}, ∣ Sub (X) ∣ = 2^{∣ X ∣} = 2^{4} = 16

The classifier in the topos of $G$-sets and in presheaves

Given: Let

G = Z /2 = {e, g}

and let

E = Set^{G}

be the topos of left

G

-sets (functors

B G \to Set

). Question: What is

Ω

, and how many subobjects does the regular representation

G

(acting on itself) have? Steps:

A subobject of a $G$ -set $X$ is a $G$ -invariant subset. The classifier must record, for each point, "the set of $h \in G$ for which $h \cdot x$ lies in the subset" — i.e. truth values are subobjects of $G$ in the category, namely the set of subgroups-with-coset structure. For a general topos $C = Set^{C^{op}}$ , $Ω (c) = {sieves on c}$ .
For $B G$ (one object, morphisms $= G$ ), a sieve is a right-ideal $\subseteq G$ ; the right ideals of a group are $\emptyset$ and $G$ . So $Ω = {0, 1}$ as a set, but with $G$ acting trivially — $Ω ≅ 1 + 1$ and $Set^{G}$ is in fact Boolean (groups are groupoids).
Subobjects of the regular $G$ -set $X = G$ : invariant subsets are unions of orbits. $G$ acting on itself by left multiplication is transitive (one orbit), so $Sub (G) = {\emptyset, G}$ .

Ω_{Set^{G}} ≅ 1 + 1 (trivial action), ∣ Sub (G) ∣ = 2

Contrast (non-Boolean): in presheaves on the poset

∙ \to ∙

(the "walking arrow"),

Ω

has three global truth values

{0, \frac{1}{2}, 1}

— false, "true later", true — and the topos is not Boolean.

Common Misconceptions

$\Omega$ need not be $1+1$

A frequent error is to assume $Ω ≅ {true, false}$ in every topos. This holds only in Boolean toposes. In $Sh (X)$ , $Ω$ is the sheaf of open sets ( $Ω (U) = {$ open $V \subseteq U}$ ), which can have uncountably many sections. Treating $Ω$ as a two-element set destroys the intuitionistic content.

Monos vs. injections, epis vs. surjections

In a general topos monomorphisms and epimorphisms are defined by cancellation, not by underlying-set behavior. They do coincide with "pointwise injective/surjective" in $Set$ and many functor categories, but the classifier axiom is about monos, and one must verify a map is a mono categorically (left-cancellable) before forming $χ_{m}$ .

A cartesian closed category that is NOT a topos

The category $Cat$ of small categories is cartesian closed (the exponential $D^{C}$ is the functor category) and has finite limits. But $Cat$ has no subobject classifier: subcategories of a category $C$ cannot be uniformly classified by maps into a single object, because a "sub-thing" must specify both objects and arrows with closure conditions that are not representable. Hence $Cat$ is cartesian closed but not a topos — showing axioms (1),(2) do not imply (3).

A category with a "classifier-like" object failing uniqueness

In the category of topological spaces $Top$ , one might hope the Sierpiński space $S$ classifies open subobjects. Indeed open inclusions $U ↪ X$ correspond to continuous maps $X \to S$ . But not every monomorphism in $Top$ is an open inclusion (e.g. inclusion of a non-open subspace), and those are not classified. So $Top$ fails the classifier axiom for general monos: $S$ classifies only the open subobjects, not all subobjects. $Top$ is not a topos.

Connections

Elementary toposes sit at the crossroads of several fields. The Grothendieck toposes — categories of sheaves $Sh (C, J)$ on a site — are exactly the elementary toposes that are also cocomplete and have a generating set (Giraud's theorem), linking to Sheaf Theory and Grothendieck Topology. The internal Heyting algebra of $Ω$ ties the subject to Heyting Algebra and Intuitionistic Logic: the Mitchell–Bénabou language makes any topos a model of higher-order intuitionistic type theory, connecting to Categorical Logic and the Curry–Howard Correspondence. A Lawvere–Tierney topology $j : Ω \to Ω$ generalizes Grothendieck topologies and yields sheaf subtoposes, the categorical face of Forcing in set theory. Cartesian closure [2] relates $Ω^{X}$ to the Cartesian Closed Category notion underlying the Lambda Calculus.

Real-World Applications

Independence proofs in set theory. Cohen forcing is reconstructed as building a Boolean-valued or sheaf subtopos via a Lawvere–Tierney topology, giving categorical proofs of the independence of CH and AC (Tierney, Scott, Solovay tradition).
Synthetic differential geometry. Toposes containing infinitesimals (the Kock–Lawvere axiom) provide rigorous frameworks where the line $R$ has nilpotent elements, simplifying differential geometry and continuum physics.
Domain theory and denotational semantics. Toposes of sheaves and the effective topos model programming-language semantics; $Ω$ provides the type of propositions for verification.
Realizability and computability. The effective topos $E f f$ (Hyland) internalizes Kleene realizability, giving a topos whose internal logic captures computable mathematics; used in constructive analysis.
Algebraic geometry. The étale and Zariski toposes of a scheme classify cohomological data; subobject classifiers organize sheaves of sets used in descent and stack theory.
Type theory and proof assistants. Topos semantics underlies models of dependent type theory and homotopy type theory (via $\infty$ -toposes), informing systems like Coq/Lean conceptually.
Quantum logic and physics. The "topos approach" of Isham–Döring uses presheaf toposes over the poset of commutative subalgebras to give a contextual, intuitionistic logic for quantum systems.

Historical Development

1957–1963 — Grothendieck. In the Tôhoku paper and SGA, Grothendieck develops sheaves on sites; a topos is born as "a category of sheaves," the natural home for cohomology in algebraic geometry.
1963 — Lawvere's thesis. Functorial Semantics of Algebraic Theories recasts model theory categorically, seeding the program of axiomatizing mathematics in categorical terms.
1964 — Lawvere, ETCS. An Elementary Theory of the Category of Sets axiomatizes $Set$ categorically, a precursor to the topos axioms.
1969–1970 — Lawvere & Tierney. Identification of the subobject classifier $Ω$ and the operator $j$ ; the elementary (finitary, first-order) definition of a topos is given, divorcing it from Grothendieck's site data.
1971 — Lawvere–Tierney sheaves. Sheafification and forcing reconstructed via topologies $j : Ω \to Ω$ ; Cohen forcing seen topos-theoretically.
1972 — Freyd, Kock, Wraith, Mikkelsen. Proofs that elementary toposes have finite colimits and are exact; Mikkelsen shows power objects suffice.
1977 — Johnstone, Topos Theory. First comprehensive monograph systematizing the field.
1980s — Hyland, the effective topos. Realizability toposes connect topos theory to recursion theory and constructive mathematics.
1992 / 2002 — Mac Lane–Moerdijk; Johnstone's Elephant. Sheaves in Geometry and Logic and the two-volume Sketches of an Elephant become standard references, consolidating the subject.

Subtleties & Limits

When is a topos Boolean, and how "non-classical" can it be? The internal logic is intuitionistic; classifying exactly which categorical conditions force the law of excluded middle, the axiom of choice, or De Morgan's laws remains a delicate study (De Morgan toposes, Boolean toposes,

\neg\neg

-sheaves). Higher topos theory: the passage to

(\infty, 1)

-toposes (Lurie) reframes

Ω

as an object classifier and is central to homotopy type theory; the precise elementary axioms for an elementary $\infty$ -topos are still being worked out (Rasekh, Shulman). Independence and large cardinals: how topos-theoretic forcing relates to the full strength of class forcing and large-cardinal hypotheses is an active interface. Predicativity: whether key topos constructions can be carried out predicatively (without impredicative power objects) motivates predicative alternatives like Σ-frames and arithmetic universes, an ongoing program of Joyal and Maietti.

Topos Theory — Elementary Toposes and Subobject Classifiers

The Core Insight

Formal Statement

Subobject

Subobject Classifier

Elementary Topos

Power Objects and Internal Hom into $\Omega$

$\Omega$ is an Internal Heyting Algebra

Why It Holds / Derivation

Construction of $\wedge$ on $\Omega$ and the Heyting structure

Why Heyting and not Boolean

F. William Lawvere & Myles Tierney | 1937– / 1937–

Building Intuition

The subobject classifier pullback square: a subobject as the preimage of $\mathsf{true}$

Worked Examples

The classifier in $\mathbf{Set}$

The classifier in the topos of $G$-sets and in presheaves

Common Misconceptions

$\Omega$ need not be $1+1$

Monos vs. injections, epis vs. surjections

A cartesian closed category that is NOT a topos

A category with a "classifier-like" object failing uniqueness

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also

Before this

Where this leads