What are your favorite concrete examples of limits or colimits that you would compute during lunch?

(The title was initially "What are your favorite concrete examples that you would compute on the table during lunch to convince a working mathematician that the notions of limits and colimits are not as dreadful as they might appear?" but it was too long.)

This is a pretty basic question which may be more appropriate for another website. However, the examples I am looking for should appeal to a working mathematician. That is why I am asking here.

Tomorrow I shall have lunch with a mathematician whom notions of limits and colimits make nervous in general. He feels that calculating small examples may help him overcome his fear. More precisely, what he is asking for are examples of various small diagrams (including fancy ones, for instance with loops) in familiar categories (topological spaces, abelian groups, &c.) whose limit or colimit we could calculate together over lunch, in the hope that he would get a better understanding of what are limits and colimits when they are taken over diagrams other than those giving pullbacks or pushouts (of which he already has a feeling).

Do you know of some particular instances of diagrams, in categories familiar to the working mathematician, whose calculation of the limit or colimit seems particularly illuminating? Or at least which could help a nervous mathematician overcome his fear of general limits and colimits?

I could come up with ad hoc examples, but perhaps there is better than that?

EDIT: Sorry for the tardy edit. I went to sleep after asking the question and just woke up thinking "I should have made it Community Wiki, added a big-list tag and provided more details". Thanks for the answers so far. (By the way, I myself am unsure as to what extent this question is appropriate for MO, but if the three votes to close could be shortly explained I would nevertheless appreciate it.)

The examples should appeal to a mathematician working in geometry and topology. For some reasons he really would like to make concrete computations. It seems that he has been faced with the following situation (to which I myself have never been faced; that is why I am asking here, in the hope that someone else already has been in the same situation as his): he is given a fancy (not the most usual) small diagram in Top or Ab or whatnot and wants to compute the limit or the colimit. Somehow he is afraid of this. I feel that my question is somewhat too broad and unprecise, but this is the best I have come up with given what I was asked myself.

• ${{1 + 1 = 2}}$. – Tom Leinster Jan 17 '14 at 16:37
• Seriously: it could be worth mentioning that products and coproducts generalize multiplication and addition of natural numbers. – Tom Leinster Jan 17 '14 at 16:38
• What's your friend's background? – Qiaochu Yuan Jan 17 '14 at 18:15
• Thanks to all who have answered so far. I have edited the post accordingly, giving some more details at the end. (Background: geometry and topology.) – Jonathan Chiche Jan 18 '14 at 0:10
• The three votes to close are as "too broad," but this seems like a good question to me. – Douglas Zare Jan 18 '14 at 0:51

I'm not sure what would work for this individual, but I'd be tempted to turn this around, Jeopardy! style. That is, instead of being presented with a diagram and trying to compute its limit/colimit, take some construction and devise a diagram which naturally expresses the construction as a limit or colimit.

So for example, this might be too easy, but consider the construction $X/A$ where $A$ is a subspace of a topological space $X$. Is this naturally a limit or colimit? Well, it's a colimit, but of what? Again, this may be too easy since your friend is comfortable around pushouts. For extra credit: what is the sensible meaning of $X/\emptyset$?

Or, take the graph of a function like $y = x^2$. Can this be thought of as a limit or colimit? This time it's a limit, namely the equal-izer of two functions from $\mathbb{R}^2$ to $\mathbb{R}$. (There's a more general lesson to be learned here, that limits are generally loci of suitable equations.)

How about the localization $\mathbb{Z}[1/p]$ where we invert a prime? Perhaps a little harder, do the same for the localization $\mathbb{Z}_p$. Or (would this be too familiar?) how would you express the $p$-adics as a limit?

Or, come up with the condition that a presheaf over a space is a sheaf. This might be either too familiar or too abstract, however. It might be best to take more concrete examples like the ones above. These are all off the top of my head, though, and somewhat untested by me personally.

• As an undergraduate reading "An elementary theory of the category of sets" I was really struck by how naturally the concept of a graph arises as the limit of the diagram. The formal definition of a function in ZFC is really the graph of the function. But just dualizing everything in sight gives the "cograph" of a function. Namely, if $f:A \to B$ is a function, its cograph is the partition of $A \sqcup B$ into sets of the form ${b}\cup f^{-1}[{b}]$, which gives a perfectly good alternative definition of a function in ZFC. – Steven Gubkin Jan 18 '14 at 3:15
• Somehow this alternate definition, while the same formal object, is really the "right way" to think about a function when it is respecting equivalence classes. I think this definition of function would make the 1st isomorphism theorem somewhat more obvious for instance - simply because it emphasizes that any function partitions its domains into equivalence classes indexed by the codomain. – Steven Gubkin Jan 18 '14 at 3:18
• @StevenGubkin Yes, when we draw pictures of functions between e.g. finite sets, with bags of dots and arrows indicating the assignments, we are really drawing the cographs. For many purposes this is the right or more illuminating picture to draw. – Todd Trimble Jan 18 '14 at 3:32
• In the spirit of Todd's "equal-izer" comment, the two standard ways of specifying a manifold are, in essence, as a limit or colimit. E.g. if you describe a sphere as the solution-set of $x^2+y^2+z^2 = 1$, you're describing it as the equalizer of two maps $\mathbb{R}^3 \to \mathbb{R}$ (and $\mathbb{R}^3$ is a product, another kind of limit). On the other hand, if you describe it as a gluing-together of patches, say two copies of the disk $D$ overlapping in a cylinder $C$, then you're describing it as the coequalizer of two maps $C\to D+D$ (and $D+D$ is a coproduct, another type of colimit). – Tom Leinster Jan 18 '14 at 16:49

It seems that he has been faced with the following situation (to which I myself have never been faced; that is why I am asking here, in the hope that someone else already has been in the same situation as his): he is given a fancy (not the most usual) small diagram in Top or Ab or whatnot and wants to compute the limit or the colimit. Somehow he is afraid of this.

Then it seems that what your friend really needs is to learn methods for computing limits or colimits! Here are some facts which should help with your friend's difficulties in various familiar categories.

• In any category, to compute arbitrary small limits and colimits it suffices to compute products and equalizers resp. coproducts and coequalizers. This is a good exercise; for a solution see, for example, this blog post.
• In $\text{Set}$, the product of a family of sets is its Cartesian product in the usual sense, and the coproduct of a family of sets is its disjoint union in the usual sense.

• In $\text{Set}$, the equalizer of a pair of functions $f, g : X \to Y$ is $\{ x \in X | f(x) = g(x) \}$, and the coequalizer is the quotient of $Y$ by the equivalence relation generated by setting $f(x) \sim g(x)$ for all $x$.

• Many categories $C$ are equipped with a forgetful functor $U : C \to \text{Set}$ with a left adjoint (the "free object" functor). It follows that $U$ preserves limits, so when limits exist in $C$ they are computed as limits in $\text{Set}$ on underlying sets. For example, this is true of

• Topological spaces, graphs
• Groups, abelian groups, rings, various other algebraic categories
• Dually, if $C$ is equipped with a forgetful functor $U$ that has a right adjoint, then $U$ preserves colimits, so when colimits exist in $C$ they must be computed as colimits in $\text{Set}$ on underlying sets. For example, this is true of

• Topological spaces, graphs
• If $C$ is a reflective subcategory of a category $D$, let $U : C \to D$ denote the inclusion and let $F : D \to C$ denote its left adjoint. By point 4, limits in $C$ can be computed in $D$. But it is in addition true that colimits in $C$ can be computed by first computing the colimit in $D$ and then applying $F$. For example, this is true of

• The inclusion of abelian groups into groups (the left adjoint is abelianization)
• The inclusion of compact Hausdorff spaces into spaces (the left adjoint is Stone-Cech compactification)
• The inclusion of sheaves into presheaves (the left adjoint is sheafification)
• To an analyst, compute the colimit along $K\subseteq\mathbb R^n$ of the spaces $C^\infty_K(\mathbb R^n)$ of smooth functions on $\mathbb R^n$ with support contained in $K$. You can build othr spaces which will ring a bell in this way, like $L^1_{\mathrm{loc}}$, and friends.

• For an algebrist, construct the localization of a commutative ring as a colimit.

• For the complex analyst, construct the space of germs of analitic functions as a colimit.

• For a more general audience, convince them that directed limits and colimits are just unions and intersections of sorts. For example, show them that a ring/module is the colimit of its finitely generated subobjects, that the cantor set is a colimit, etc.

• Do the kernel and the cokernel, of course. Generalize to equalizers and coequalizers to show that you get useful notions in more general categories which serve similar purposes.

My facourite example is that the integers $\mathbb Z$ may be obtained from a pushout diagram

$$\begin{matrix} \{0,1\} & \to & \{ 0\} \\ \downarrow && \downarrow \\ \mathcal I & \to & \mathbb Z \end{matrix}$$ in the category of groupoids where $\mathcal I$ is the groupoid with two objects $0,1$ and non identity arrows $\iota:0 \to 1, \iota^{-1}:1 \to 0$ This pushout also nicely models the way the circle $S^1$ is obtained from the unit interval $[0,1]$ by identifying $0$ and $1$. Thus we see one value of the notion of colimit, of which pushout is a special case, namely to compare concepts in different categories. Also, we know most things internally about the groupoid $\mathcal I$, whereas the integers are infinite.

• I was thinking about this example too, except mine was the coequalizer in $\mathbf{Cat}$ of the two object inclusions of the terminal category $\mathbf{1}$ into the arrow category $\mathbf{2}$. The coequalizer is the monoid of natural numbers made into a 1-object category. This example appears in Lawvere's Category of Categories paper, and in Mathematics Made Difficult. – Todd Trimble Jan 18 '14 at 21:53
• @Todd: That example is a good one, and can also be phrased as a pushout in the category $Cat$ of small categories, noting that $Cat \to Sets$ is a bifibration of categories. Thanks also for the references. – Ronnie Brown Jan 20 '14 at 18:00
• @Todd: I mention that the groupoid example of the integers was in essence in my 1967 paper on van Kampen's theorem and was in the 1968 book on Topology. – Ronnie Brown Jan 21 '14 at 15:02

I think that producing beautiful examples strongly depends on the background of your friend. Nevertheless, the following is the best I can find.

Let $\cal C$ be small; compute the following limits and colimits

1. $\varprojlim {\cal C}(x,-)$
2. $\varprojlim {\cal C}(-,x)$
3. $\varinjlim {\cal C}(x,-)$
4. $\varinjlim {\cal C}(-,x)$

where $\mathcal C(-,x)\colon {\cal C}^\text{op}\to \bf Set$, $\mathcal C(x,-)\colon {\cal C}\to \bf Set$ are representable hom-functor, and $x$ is any fixed object.

Fundamental example: The free topological group G over a convergent sequence, in the sense of Graev.

This is the finest topological group with countably many generators, so that the generators converge to the identity. The basic behavior of colimits of Hausdorff compacta plays a crucial role in understanding this construction, sketched as follows.

Appetizer: Fix a group G and show in general that the collection of topologies on G (under which G is a topological group) forms a complete lattice under inclusion, ( key lemma: the union of any collection of topological group topologies on G is a subbasis for a topological group) ( intersections of such topologies behave much more opaquely but save that for dessert).

Main course: Let G be the free group on countably many generators $x_1$, $x_2$,...

Summon the finest topological group topology $\tau$ on G such that $x_1, x_2$,... $\rightarrow$ 1.

The appetizers ensure $\tau$ exists, but is $(G,\tau)$ sequential? Is it metrizable? Is it normal?

The answers are yes, no, and yes, but this is entirely nonobvious. However, much of the mystery is absorbed by basic theory of colimits of Hausdorff compacta.

Notably, $(G,\tau)$ is the direct limit of countably many nested sequential Hausdorff compacta, and hence normal and sequential. Moreover (by an ultimately straightforward miracle), $(G,\tau)$ X $(G,\tau)$ is a sequential space. The utility of the latter fact is that to check continuity of group multiplication, it suffices to check that group multiplication is continuous over convergent sequences.

( To justify the use of `miracle' the plausible claim that the product of two sequential spaces is sequential, is generally false. (Recall a sequential space is a topological quotient of a metric space, or (nonobviously) equivalently, a space whose closed sets are precisely those which are closed under convergent sequences)).

Dessert : Despite the apparent naturality of the construction at hand, we are surrounded by categorical pathology. Notably, fixing the countable free group G and considering the (complete) lattice1 of topological group topologies on G, and the larger (complete) lattice2 of all topologies on G, inclusion of lattice1 into lattice2 is NOT a homomorphism. (The pathology remains even if we restrict the conversation to sequential topologies).

Antacid: The root pathology is created by the general failure of colimits to commute with products. However, fixing a category and making modest adjustments to standard defintions (such as canonically refining the standard product of sequential spaces to preserve the sequential property) can create useful structures and powerful mathematics.

• Colimit over a discrete category (sum)
• limit over a discrete category (product)

Another example: colimit and limit in a category which is a partially ordered set (inf and sup).

The Arens-Michael envelope is an example of a natural colimit in Functional analysis. Intuitively it is an operation (a functor) that turns each topological algebra $A$ into its nearest from the outside holomorphic algebra $\text{Env}A$.

As an illustration, if $A$ is the algebra ${\mathcal P}(M)$ of polynomials on an affine algebraic manifold $M$, then its Arens-Micael envelope $\text{Env}A$ is exactly the algebra ${\mathcal O}(M)$ of holomorphic functions on $M$: $$\text{Env}{\mathcal P}(M)={\mathcal O}(M).$$
The non-commutative case is especially interesting, since it generates non-trivial computations in non-commutative geometry. For example, the "polynomial version" of the so-called $az+b$-group is turned into its "holomorphic version": $$\text{Env}{\mathcal P}_q({\mathbb C}^\times\ltimes{\mathbb C})={\mathcal O}_q({\mathbb C}^\times\ltimes{\mathbb C}),$$
for $|q|=1$, but the formula becomes different in the case of $|q|\ne 1$: $$\text{Env}{\mathcal P}_q({\mathbb C}^\times\ltimes{\mathbb C})={\mathcal O}({\mathbb C}^\times)\underset{\delta^q}{\overset{z}{\odot}}{\mathcal P}^\star({\mathbb C})$$
(see details in my paper).

Similarly, there are natural functors of taking

Finding the measurable envelope $\equiv$ the space of ergodic components (with respect to a given measure) of the tail equivalence relation on an infinite product space is equivalent to finding the corresponding projective limit in an appropriate category. However, this problem would usually require a quite hefty lunch.