Limits and Colimits as Universal Cones

Intuition Beginner

Many constructions in mathematics answer the same kind of question: given a pattern of objects and arrows, find the single best object that sits over the whole pattern and feeds into every piece of it consistently. The pair of two sets has a product, two functions out of a common source can be combined, a system of nested groups has one object capturing all of them at once. Each is the same move dressed differently.

A diagram is just a chosen pattern: some objects, with some arrows between them, drawn inside a category. A cone over that pattern is one extra object on top, with one arrow running down to each object in the pattern, arranged so that following an arrow inside the pattern never changes where you end up. The cone is consistent with the pattern's own internal arrows.

There are usually many cones over a pattern. The limit is the best one: a cone so efficient that every other cone factors through it in exactly one way. Best here means most economical, not biggest. The limit packages all the information the pattern forces, and nothing extra.

Flip every arrow and you get the mirror idea. A cone underneath the pattern, receiving one arrow from each piece, is a cocone; the most efficient cocone is the colimit. Limit and colimit are the same notion seen from two sides.

Visual Beginner

Picture a diagram as a few dots on a table connected by some strings (the arrows of the pattern). A cone hangs one apex dot above the table, with a straight string from the apex down to each table dot. The cone is allowed only if sliding down to a dot and then walking along a table string lands you at the same place as sliding straight down to the second dot.

feature	a cone over a pattern	the limit (best cone)
has	an apex with one leg to each object	an apex with one leg to each object
obeys	every triangle to the pattern's arrows commutes	the same commuting condition
compares to others	many cones are possible	every other cone factors through it once
flipping arrows gives	a cocone (apex below)	the colimit

The picture shows why "best" is a comparison, not a size. Among all apexes that hang consistently over the pattern, one is universal: every competitor sends a single string up to it, and that string explains the competitor entirely. That uniqueness is the whole content of the word limit.

Worked example Beginner

Work in the category of sets, where objects are sets and arrows are functions. Take the smallest interesting pattern: two separate sets $A=\{a_1,a_2\}$ and $B=\{b_1,b_2,b_3\}$ with no arrows between them.

Step 1. A cone over this pattern is an apex set $C$ with one function $C\to A$ and one function $C\to B$. Since the pattern has no internal arrows, there is no commuting condition to check; any pair of functions out of $C$ counts as a cone.

Step 2. Guess the best apex. Try the set of ordered pairs $A\times B=\{(a_1,b_1),(a_1,b_2),(a_1,b_3),(a_2,b_1),(a_2,b_2),(a_2,b_3)\}$, which has $2\times 3=6$ elements. Its two legs are "read off the first slot" and "read off the second slot".

Step 3. Check it is best. Suppose some other apex $C$ has legs $f:C\to A$ and $g:C\to B$. Define one function $C\to A\times B$ by sending each element $c$ to the pair $(f(c),g(c))$. Reading off the first slot recovers $f$, the second recovers $g$, and no other function to $A\times B$ does this. So every cone factors through $A\times B$ in exactly one way.

What this tells us: the product of two sets is the limit of the two-object pattern, and "best cone" picks it out without ever mentioning ordered pairs in advance. The same recipe with arrows reversed gives the disjoint union $A+B$, with five elements, as the colimit.

Check your understanding Beginner

Formal definition Intermediate+

Fix a category $\mathcal{C}$ and a small category $\mathcal{J}$, the index (or shape) category. A diagram of shape $\mathcal{J}$ in $\mathcal{C}$ is a functor $F:\mathcal{J}\to \mathcal{C}$ ^{[Mac Lane 1998]}. The objects $F(j)$ are the vertices of the diagram and the morphisms $F(u):F(j)\to F(k)$, for $u:j\to k$ in $\mathcal{J}$, are its internal arrows. Writing ${\Delta }_C:\mathcal{J}\to \mathcal{C}$ for the constant diagram at an object $C$ (every vertex $C$, every arrow ${\mathrm{id}}_C$) packages the apex.

Definition (cone). A cone over $F$ with apex $C$ is a natural transformation $\lambda :{\Delta }_C\Rightarrow F$. Concretely it is a family of legs ${\lambda }_j:C\to F(j)$, one per object $j\in \mathcal{J}$, such that for every morphism $u:j\to k$ the triangle commutes: $F(u)\circ {\lambda }_j={\lambda }_k$. A morphism of cones $(C,\lambda )\to (C^{\prime },{\lambda }^{\prime })$ is a morphism $h:C\to C^{\prime }$ in $\mathcal{C}$ with ${\lambda }_j^{\prime }\circ h={\lambda }_j$ for all $j$. Cones over $F$ and their morphisms form a category $\mathrm{Cone}(F)$.

Definition (limit). A limit of $F$ is a terminal object of $\mathrm{Cone}(F)$: a cone $(L,\pi )$ such that every cone $(C,\lambda )$ admits a unique morphism of cones $⟨\lambda ⟩:C\to L$, that is, a unique $h$ with ${\pi }_j\circ h={\lambda }_j$ for all $j$. The apex $L$ is written $\lim F$ and the legs ${\pi }_j$ are the projections or limit cone. The universal property is the existence-and-uniqueness clause; it determines $(L,\pi )$ up to a unique isomorphism of cones.

Dually, a cocone under $F$ with apex $C$ is a natural transformation $F\Rightarrow {\Delta }_C$: legs ${\iota }_j:F(j)\to C$ with ${\iota }_k\circ F(u)={\iota }_j$. A colimit of $F$ is an initial object of the category of cocones — a cocone $(L,\iota )$ through which every cocone factors uniquely. Equivalently, a colimit of $F$ in $\mathcal{C}$ is a limit of $F^{\mathrm{op}}:{\mathcal{J}}^{\mathrm{op}}\to {\mathcal{C}}^{\mathrm{op}}$, read back in $\mathcal{C}$; this is the duality of 41.01.01 applied to the cone construction.

The standard finite shapes instantiate the definition. The empty category gives the empty diagram, whose limit is the terminal object $1$ (a cone with no legs is an apex, and terminality of the empty-diagram cone is exactly terminality in $\mathcal{C}$) and whose colimit is the initial object $0$. The discrete two-object shape gives the product $A\times B$ with projections, dual to the coproduct $A+B$. The parallel-pair shape $\bullet \rightrightarrows \bullet$ gives the equalizer of $f,g:A\to B$, the universal $e:E\to A$ with $f\circ e=g\circ e$, dual to the coequalizer. The cospan $\bullet \to \bullet \leftarrow \bullet$ gives the pullback (fibre product) $A{\times }_{C}B$, dual to the pushout over a span.

Definition ((co)completeness). A category is complete if every diagram indexed by a small category has a limit, and finitely complete if every diagram indexed by a finite category has a limit. Cocomplete and finitely cocomplete are the dual conditions. $\mathbf{Set}$, $\mathbf{Grp}$, $\mathbf{Ab}$, $\mathbf{Top}$, ${\mathbf{Vect}}_k$, and $R\text{-}\mathbf{Mod}$ are complete and cocomplete.

Counterexamples to common slips Intermediate+

• A cone is not just any family of legs. The commuting condition $F(u)\circ {\lambda }_j={\lambda }_k$ is essential. Dropping it gives a cone over the discrete diagram on the same objects — a product — which generally differs from the limit of the diagram with its arrows. For a parallel pair $f,g$ the equalizer is a proper subobject of the product factor exactly because the condition $f\circ \lambda =g\circ \lambda$ bites.
• Limits need not be subobjects of products in a general category. In $\mathbf{Set}$ every limit embeds into a product (Theorem below), but this realisation uses that $\mathbf{Set}$ has products and equalizers; it is a theorem, not the definition.
• The product in $\mathbf{Grp}$ is the direct product, but the coproduct is the free product, not the direct sum. Coproducts in non-abelian categories rarely look like disjoint unions of underlying sets; the universal property, not a set-level guess, fixes them. In $\mathbf{Ab}$ the finite coproduct does coincide with the product (the biproduct), a feature special to additive categories.

Key theorem with proof Intermediate+

The signature result is that the universal property defining a limit is exactly terminality in the cone category, and that this characterisation forces uniqueness up to canonical isomorphism.

Theorem (limits are terminal cones; uniqueness up to unique iso). Let $F:\mathcal{J}\to \mathcal{C}$ be a diagram. A cone $(L,\pi )$ over $F$ is a limit of $F$ if and only if it is a terminal object of $\mathrm{Cone}(F)$. Any two limits of $F$ are connected by a unique isomorphism of cones; in particular their apexes are uniquely isomorphic in $\mathcal{C}$ ^{[Mac Lane 1998]}.

Proof. By definition a morphism of cones $(C,\lambda )\to (L,\pi )$ is a morphism $h:C\to L$ in $\mathcal{C}$ with ${\pi }_j\circ h={\lambda }_j$ for every $j$. So a cone $(L,\pi )$ is terminal in $\mathrm{Cone}(F)$ precisely when for every cone $(C,\lambda )$ there is exactly one such $h$. That clause is verbatim the universal property in the definition of a limit. The two notions therefore coincide; no further argument is needed for the equivalence.

For uniqueness, suppose $(L,\pi )$ and $(L^{\prime },{\pi }^{\prime })$ are both terminal in $\mathrm{Cone}(F)$. Terminality of $(L^{\prime },{\pi }^{\prime })$ produces a unique cone morphism $h:L\to L^{\prime }$; terminality of $(L,\pi )$ produces a unique cone morphism $h^{\prime }:L^{\prime }\to L$. The composite $h^{\prime }\circ h:L\to L$ is a cone morphism $(L,\pi )\to (L,\pi )$, and so is ${\mathrm{id}}_L$; terminality of $(L,\pi )$ allows exactly one such self-morphism, forcing $h^{\prime }\circ h={\mathrm{id}}_L$. Symmetrically $h\circ h^{\prime }={\mathrm{id}}_{L^{\prime }}$. Hence $h$ is an isomorphism of cones, and it is the only cone morphism $L\to L^{\prime }$, so the isomorphism is unique. Applying the forgetful map $\mathrm{Cone}(F)\to \mathcal{C}$ that returns the apex, $h$ is an isomorphism $L\cong L^{\prime }$ in $\mathcal{C}$ compatible with the projections. $\square$

Bridge. This terminal-cone reformulation builds toward the entire calculus of the chapter and appears again in 41.02.02, where the construction of every limit from products and equalizers is carried out by realising the cone category's terminal object inside a single equalizer of two maps between products. The foundational reason the argument is so short is that a universal property is a (co)representability statement, so two solutions are compared by their own defining maps; this is exactly the uniqueness pattern already met for initial and terminal objects in 41.01.01. The construction is dual to the colimit case: a colimit is an initial cocone, and the same proof read in ${\mathcal{C}}^{\mathrm{op}}$ gives uniqueness of colimits, so one argument discharges both. Putting these together, the bridge is that "best cone" and "terminal object of the cone category" are two names for one piece of data, which is what later lets limits be manipulated as objects — pulled back, preserved by functors, and computed pointwise in functor categories.

Exercises Intermediate+

Advanced results Master

Theorem (a category has all small limits iff it has all small products and all equalizers). If $\mathcal{C}$ has all products indexed by sets and all equalizers of parallel pairs, then $\mathcal{C}$ is complete; dually, products and equalizers being replaced by coproducts and coequalizers gives cocompleteness. The construction realises $\lim F$, for $F:\mathcal{J}\to \mathcal{C}$ with $\mathcal{J}$ small, as the equalizer of two maps ${\prod }_{j\in \mathrm{Ob}\mathcal{J}}F(j)\rightrightarrows {\prod }_{u:j\to k}F(k)$, one assembling the projections and one applying the diagram's arrows. The full statement, with the two maps written out and the cone correspondence verified, is the subject of 41.02.02; here the result anchors the examples below.

Theorem (concrete (co)limits across the corpus). In $\mathbf{Set}$ the limit of $F$ is the set of compatible families $\{(x_j)\mid x_j\in F(j),F(u)(x_j)=x_k\}\subseteq {\prod }_{j}F(j)$, with projections as legs; products are Cartesian products, equalizers are solution subsets, pullbacks are fibre products $\{(a,b)\mid f(a)=g(b)\}$. In $\mathbf{Grp}$, $\mathbf{Ab}$, $\mathbf{Top}$, and $R\text{-}\mathbf{Mod}$ limits are computed on underlying sets and then carry the induced structure, because the forgetful functors to $\mathbf{Set}$ preserve and reflect limits. The kernel of a module map is the equalizer with the zero map; the inverse limit $\underleftarrow{\lim }{M}_i$ of an inverse system of modules — the construction underlying $p$-adic integers and completions — is the limit over a cofiltered poset, and the direct limit $\underrightarrow{\lim }{M}_i$ is the dual colimit over a filtered poset. The fibre product of schemes is the pullback in the category of schemes, gluing affine pieces by their universal property.

Theorem (limits in functor categories are computed pointwise). Let $\mathcal{C}$ be complete and $\mathcal{D}$ a small category. Then the functor category $[\mathcal{D},\mathcal{C}]=\mathbf{Fun}(\mathcal{D},\mathcal{C})$ is complete, and for a diagram $G:\mathcal{J}\to [\mathcal{D},\mathcal{C}]$ the limit is the functor $d\mapsto {\lim }_{j}G(j)(d)$, with the limit cone evaluated objectwise. Evaluation functors ${\mathrm{ev}}_d:[\mathcal{D},\mathcal{C}]\to \mathcal{C}$ jointly create these limits. The dual statement computes colimits pointwise. This is why limits of presheaves, sheaves of abelian groups, and diagrams of representations are taken value by value.

Synthesis. Putting these together, the universal cone is a single organising idea that the rest of the subject specialises in every direction. The foundational reason a limit is unique and well-behaved is that it is a terminal object of $\mathrm{Cone}(F)$, so it inherits the rigidity of all universal constructions; this is exactly the rigidity that lets one prove the products-plus-equalizers theorem once and read off completeness for $\mathbf{Set}$, $\mathbf{Grp}$, $\mathbf{Top}$, and module categories at a stroke. The construction is dual throughout: colimit is initial cocone, so every theorem here generalises by reversing arrows in $\mathcal{C}$, and the inverse limit of $p$-adic completions is the formal mirror of the direct limit of localisations. The bridge to the next chapter is that the limit assignment is functorial and right adjoint to the diagonal — developed in 41.03.01 — which is why pointwise computation in functor categories works and why right adjoints preserve limits, the central insight that makes limits portable across the entire categorical landscape and connects them to the representability machinery of 41.04.01.

Full proof set Master

Proposition 1 (a right adjoint statement: limits as a representing object). For $F:\mathcal{J}\to \mathcal{C}$ with $\mathcal{C}$ locally small, an object $L$ with a cone $\pi$ is a limit of $F$ if and only if there is a natural isomorphism ${\mathrm{Hom}}_{\mathcal{C}}(C,L)\cong \mathrm{Cone}(C,F)$ in the variable $C$, where $\mathrm{Cone}(C,F)$ is the set of cones over $F$ with apex $C$.

Proof. A cone over $F$ with apex $C$ is a family ${\lambda }_j:C\to F(j)$ with $F(u)\circ {\lambda }_j={\lambda }_k$; precomposition with $h:C^{\prime }\to C$ sends it to $({\lambda }_j\circ h)$, making $C\mapsto \mathrm{Cone}(C,F)$ a functor ${\mathcal{C}}^{\mathrm{op}}\to \mathbf{Set}$. Given a cone $(L,\pi )$, define ${\Phi }_C:\mathrm{Hom}(C,L)\to \mathrm{Cone}(C,F)$ by $h\mapsto ({\pi }_j\circ h)$; naturality in $C$ is immediate from associativity. The map ${\Phi }_C$ is a bijection for every $C$ precisely when every cone with apex $C$ factors as $\pi \circ h$ for a unique $h$ — the universal property. So $L$ is a limit iff $\Phi$ is a natural isomorphism, i.e. $L$ represents the cone functor. $\square$

Proposition 2 (the limit of an inverse system is the compatible-family submodule). Let $(M_i,{\phi }_{ij})$ be an inverse system of $R$-modules over a directed poset $I$ (so ${\phi }_{ij}:M_j\to M_i$ for $i\le j$, compatible). Then $\underleftarrow{\lim }{M}_i=\{(x_i)\in {\prod }_i{M}_i\mid {\phi }_{ij}(x_j)=x_i\text{for all}i\le j\}$, an $R$-submodule of the product, with the restricted projections as limit cone.

Proof. The displayed set $L$ is closed under addition and scalar action because each ${\phi }_{ij}$ is $R$-linear, hence a submodule of ${\prod }_i{M}_i$. Its projections ${\pi }_i:L\to M_i$ satisfy ${\phi }_{ij}\circ {\pi }_j={\pi }_i$ by the defining relation, so $(L,\pi )$ is a cone. Given any cone $(N,\lambda )$ with ${\phi }_{ij}\circ {\lambda }_j={\lambda }_i$, the map $h:N\to {\prod }_i{M}_i$, $n\mapsto ({\lambda }_i(n))$, lands in $L$ because ${\phi }_{ij}({\lambda }_j(n))={\lambda }_i(n)$, and it is the unique module map with ${\pi }_i\circ h={\lambda }_i$ since its components are forced. So $(L,\pi )$ is terminal in $\mathrm{Cone}$, i.e. the limit. $\square$

Proposition 3 (pointwise limits in a functor category). Let $\mathcal{C}$ be complete, $\mathcal{D}$ small, and $G:\mathcal{J}\to [\mathcal{D},\mathcal{C}]$ a diagram with $\mathcal{J}$ small. Define $L:\mathcal{D}\to \mathcal{C}$ on objects by $L(d)={\lim }_{j}G(j)(d)$. Then $L$ extends to a functor and is the limit of $G$, with limit cone given objectwise.

Proof. For a morphism $\delta :d\to d^{\prime }$ in $\mathcal{D}$, the maps $G(j)(\delta ):G(j)(d)\to G(j)(d^{\prime })$ form a morphism of the $j$-indexed diagrams, so by the universal property of $L(d^{\prime })$ there is a unique $L(\delta ):L(d)\to L(d^{\prime })$ commuting with the projections; functoriality of $L$ follows from uniqueness applied to ${\mathrm{id}}_d$ and to composites. The objectwise limit cones ${\pi }_j^d:L(d)\to G(j)(d)$ assemble into natural transformations ${\pi }_j:L\Rightarrow G(j)$ because each $G(j)$ is a functor and the projections were chosen compatibly. Given a cone $(K,\lambda )$ in $[\mathcal{D},\mathcal{C}]$, evaluating at each $d$ gives a cone over $G(-)(d)$, hence a unique $h_d:K(d)\to L(d)$; naturality of $h$ follows because both $L(\delta )\circ h_d$ and $h_{d^{\prime }}\circ K(\delta )$ are the unique mediator at $d^{\prime }$. So $h:K\Rightarrow L$ is the unique cone morphism, and $L=\lim G$ computed pointwise. $\square$

Proposition 4 (a complete category with products and equalizers; the cone equalizer). If $\mathcal{C}$ has all small products and all equalizers, then for any small $F:\mathcal{J}\to \mathcal{C}$ the limit exists and is the equalizer $E$ of the pair $s,t:{\prod }_{j}F(j)\rightrightarrows {\prod }_{(u:j\to k)}F(k)$, where $s$ has component $F(u)\circ {\mathrm{pr}}_j$ and $t$ has component ${\mathrm{pr}}_k$.

Proof. An element-free reading: a map $C\to {\prod }_{j}F(j)$ is a family $({\lambda }_j)$, and $s,t$ are arranged so that $s\circ ⟨\lambda ⟩=t\circ ⟨\lambda ⟩$ holds iff $F(u)\circ {\lambda }_j={\lambda }_k$ for every $u:j\to k$, which is exactly the cone condition. The equalizer $e:E\to {\prod }_{j}F(j)$ is the universal map with $s\circ e=t\circ e$, so $E$ carries a universal cone ${\pi }_j={\mathrm{pr}}_j\circ e$: any cone $(C,\lambda )$ gives $⟨\lambda ⟩:C\to {\prod }_{j}F(j)$ equalizing $s,t$, hence a unique factorisation through $E$, and conversely every map to $E$ yields a cone. Thus $E=\lim F$. The detailed verification and the dual coequalizer construction are completed in 41.02.02. $\square$

Connections Master

• Limits from products and equalizers 41.02.02. The Advanced-results theorem that completeness reduces to products plus equalizers is proved in full in 41.02.02, where the two maps between products are written explicitly and the cone-equalizer correspondence of Proposition 4 is verified in both the limit and colimit directions. That unit owns the reduction theorem; the present unit owns the definition of the universal cone it relies on, so the two are a definition-then-construction pair.

• Limits as adjoints to the diagonal 41.03.01. The limit assignment $\lim :[\mathcal{J},\mathcal{C}]\to \mathcal{C}$ is right adjoint to the constant-diagram functor $\Delta :\mathcal{C}\to [\mathcal{J},\mathcal{C}]$, and the colimit is the left adjoint; the cone condition of this unit is precisely the hom-set bijection of that adjunction. 41.03.01 develops this, from which "right adjoints preserve limits" and the pointwise computation proved here both follow as structural consequences.

• Representability and the Yoneda perspective 41.04.01. Proposition 1 recasts a limit as a representing object for the cone functor $C\mapsto \mathrm{Cone}(C,F)$, exactly the representability viewpoint developed in 41.04.01; the Yoneda lemma there turns "the limit represents the cone functor" into the statement that the limit is determined by its functor of points, tying the universal-cone definition to the embedding of a category into its presheaves.

• Inverse and direct limits in algebra 01.02.09. The inverse limit of modules (Proposition 2) and the direct limit (Exercise 6) are the special cofiltered and filtered cases used throughout the algebra corpus surveyed in 01.02.09 — completions, $p$-adic integers, and localisations — so this unit supplies the categorical universal property that those concrete constructions instantiate, and 01.02.09 supplies the running algebraic examples.

Historical & philosophical context Master

The universal-property formulation of limits grew directly out of the functorial outlook introduced by Eilenberg and Mac Lane in 1945. Inverse and direct limits of groups and modules were in use earlier — Pontryagin and Steenrod employed them in topology, and the systematic treatment in Eilenberg and Steenrod's Foundations of Algebraic Topology (1952) ^{[Eilenberg-Steenrod 1952]} codified inverse and direct systems over directed sets — but these predate the general cone definition. The recognition that products, equalizers, pullbacks, and the older inverse limits are all instances of a single universal construction over an arbitrary index category is due to the categorical school of the 1950s and was given its standard form by Mac Lane in Categories for the Working Mathematician (1971; 2nd ed. 1998) ^{[Mac Lane 1998]}, which fixed the terminology of cones, limits, and the reduction of limits to products and equalizers.

Grothendieck's reworking of homological algebra and algebraic geometry made (co)limits load-bearing: the AB axioms of his 1957 Tôhoku paper classify abelian categories by which limits and colimits exist and how they interact with exactness, and the fibre product of schemes is the pullback that lets geometry be done relative to a base. Kan's introduction of adjoint functors in 1958 then showed that limits and colimits are themselves adjoints to the diagonal, which is the form in which they enter the modern theory and the reason the subject treats existence of limits as a representability question rather than an ad hoc construction.

Bibliography Master

@book{MacLane1998limits,
  author    = {Mac Lane, Saunders},
  title     = {Categories for the Working Mathematician},
  edition   = {2},
  publisher = {Springer},
  series    = {Graduate Texts in Mathematics 5},
  year      = {1998}
}

@book{Riehl2016limits,
  author    = {Riehl, Emily},
  title     = {Category Theory in Context},
  publisher = {Dover},
  year      = {2016}
}

@book{Borceux1994,
  author    = {Borceux, Francis},
  title     = {Handbook of Categorical Algebra 1: Basic Category Theory},
  publisher = {Cambridge University Press},
  series    = {Encyclopedia of Mathematics and its Applications 50},
  year      = {1994}
}

@book{EilenbergSteenrod1952,
  author    = {Eilenberg, Samuel and Steenrod, Norman},
  title     = {Foundations of Algebraic Topology},
  publisher = {Princeton University Press},
  year      = {1952}
}

@article{Grothendieck1957,
  author  = {Grothendieck, Alexander},
  title   = {Sur quelques points d'alg\`ebre homologique},
  journal = {T\^ohoku Mathematical Journal},
  volume  = {9},
  year    = {1957},
  pages   = {119--221}
}

@article{Kan1958,
  author  = {Kan, Daniel M.},
  title   = {Adjoint functors},
  journal = {Transactions of the American Mathematical Society},
  volume  = {87},
  year    = {1958},
  pages   = {294--329}
}