I've found a general categorical construction which I'm not familiar with.

Suppose that we have the square shown, with categories $A$, $B_i$, $C$ and functors $F_i$ and $G_i$ such that the diagram commutes ($G_0 F_0 = G_1 F_1$), and that $A$ is discrete in the sense that it has only the identity morphisms.

$$ \begin{array}{rcccl} A & {} & \xrightarrow{F_0} & {} & B_0 \\ {} & {} & {} & {} & {} \\ {\scriptstyle F_1}\downarrow & {} & {} & {} & \downarrow{\scriptstyle G_0} \\ {} & {} & {} & {} & {} \\ B_1 & {} & \xrightarrow{G_1} & {} & C \\ \end{array} $$

Then define $A^*$ as follows: the objects of $A^*$ are the same as the objects of $A$; a morphism $m \in A^*(X, Y)$ is a pair $(m_0, m_1)$ such that $m_i \in B_i(F_i(X), F_i(Y))$ and $G_0(m_0) = G_1(m_1)$. I believe that this makes $A^*$ a category with obvious functors $A \to A^*$, $A^* \to B_i$ making the whole diagram commute.

$$ \begin{array}{rcccl} A & {} & \xrightarrow{F_0} & {} & B_0 \\ {} & \searrow & {} & \nearrow & {} \\ {\scriptstyle F_1}\downarrow & {} & A^* & {} & \downarrow{\scriptstyle G_0} \\ {} & \swarrow & {} & {} & {} \\ B_1 & {} & \xrightarrow{G_1} & {} & C \\ \end{array} $$

For example, suppose $A$ is the discrete category $\operatorname{Top}_{\operatorname{dis}}$ of topological spaces with identity morphisms, $B_0$ is the usual category $\operatorname{Top}$, $B_1$ is the category $\operatorname{Set}_{\operatorname{inj}}$ of sets with injective maps between them, and $C$ is the usual category $\operatorname{Set}$. The functors are all the obvious inclusions/forgettings. Then $A^*$ is the category of topological spaces with injective continuous maps between them. It's a kind of hybrid between $\operatorname{Top}$ and $\operatorname{Set}_{\operatorname{inj}}$.

$$ \begin{array}{rcccl} \operatorname{Top}_{\operatorname{dis}} & {} & \xrightarrow{} & {} & \operatorname{Top} \\ {} & \searrow & {} & \nearrow & {} \\ {}\downarrow & {} & A^* & {} & \downarrow{} \\ {} & \swarrow & {} & {} & {} \\ \operatorname{Set}_{\operatorname{inj}} & {} & \xrightarrow{} & {} & \operatorname{Set} \\ \end{array} $$


  • Does this all work?
  • Is this a known construction? In particular, does it have some nice universal property?
  • 2
    $\begingroup$ In your specific example, $A^*$ is the same as the pullback in the usual sense. $\endgroup$ – Zhen Lin Nov 18 '14 at 19:52

Here is a better construction, namely the pullback, which makes no mention of $A$ (which is doing no work in your example): the objects are pairs $(c_0, c_1)$ where $c_i \in \text{Ob}(B_i)$ such that $G_0(c_0) = G_1(c_1)$. The morphisms $(c_0, c_1) \to (d_0, d_1)$ are pairs $(f_0, f_1)$ where $f_i \in \text{Hom}(c_i, d_i)$ such that $G_0(f_0) = G_1(f_1)$ (note that to state this condition we need the corresponding condition on objects first).

In your example, once you've picked a topological space $c_0$ then $c_1$ is constrained to be the underlying set of $c_0$ which is why you didn't see the interesting behavior on objects. But in general this construction behaves interestingly on objects as well.

However, it has one substantial drawback: it is not invariant under equivalence of (diagrams of) categories. That is, it really depends on (diagrams of) categories up to isomorphism rather than equivalence, which is generally a bad sign in category theory. There is a better-behaved construction which replaces the equality $G_0(c_0) = G_1(c_1)$ with an isomorphism called the homotopy pullback. There is also a lax version which replaces the equality with a morphism, not necessarily an isomorphism, which I guess should be called the lax 2-pullback or something but which is more conventionally called the comma category.

  • $\begingroup$ The lax pullback is the lax limit of the obvious diagram. It is not equivalent to the comma category. On the other hand, the pseudopullback (= pseudolimit of the obvious diagram) is equivalent to the iso-comma category. There is no substantial difference between the pseudopullback (also called "homotopy pullback", but I prefer to avoid that phrase in this context) and the ordinary pullback if one of the functors is an isofibration, which is the case in the OP's specific example. $\endgroup$ – Zhen Lin Nov 19 '14 at 9:17
  • $\begingroup$ Thank you, this pullback might be what I need. However, it's not clear to me how to define $A^*$ using a pullback: a pullback of the $G_i$ loses the information in the $F_i$, and a pullback of the $G_i F_i$ is discrete. It seems likely that a combination of a pullback and something else will give me the original construction, but we don't have that yet. $\endgroup$ – Hew Wolff Nov 20 '14 at 15:58
  • $\begingroup$ @Hew: is there any other example you're interested in? Again, in your only example, $A$ is doing no work and the $F_i$ are not important. $\endgroup$ – Qiaochu Yuan Nov 21 '14 at 4:10
  • $\begingroup$ There is, but it's a little complicated. :-) I'll try to find a more interesting example. $\endgroup$ – Hew Wolff Nov 21 '14 at 22:43

$A^*$ is the functor pullback (thank you to responders above) combined with the "inverse image of a functor" described in my other question here. We can get the pullback $P$, then its universal property gives us a functor $F: A \to P$ with a commutative diagram. Then in the notation of that other question, we can factor $F$ into $A \to F^{-1}(P) \to P$, and take $A^* = F^{-1}(P)$.

I'm still not sure whether this is useful. I need to think some more about the principle of equivalence invariance which responders have mentioned.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.