First, suppose that $i:K \hookrightarrow M$ is an insertion in some concrete category $\mathcal{C}$, and that we have some morphism $f:T \to M$ such that $f(T) \subseteq K \subseteq M$. It seems to me evident (from the standard definitions of function, subset, insertion, etc.) that there is a morphism $g:T \to K$ such that $i \circ g = f$, the only difference between $f$ and $g$ being that $\mathrm{cod}(f) = M$ while $\mathrm{cod}(g) = K$. But how does one prove the existence of $g$ categorically? (I imagine that the key to this lies in casting the definitions of insertion and $f(T) \subseteq K$ in categorical language.) How would the statement of the theorem have to be changed so that it holds for an arbitrary category? (E.g. I imagine that $i:K \to M$ would be described merely as a monomorphism, rather than an insertion.)

Second, what does one call a function like $g$, which differs from another function $f$ in codomain only? Is there a general category-theoretical name for the relationship between morphisms $g$ and $f$ in this example?

(This reminds me of how one gets a function $h|_Y$ by restricting the function $h$ to some subset $Y$ of $\mathrm{dom}(h)$. Here, instead, I want to "restrict" the codomain.)


PS: I have been studying a bit of CT in my spare time for some weeks: IOW, I'm a rank noob! I know what (co)products are, and have a very fuzzy understanding of pullbacks/pushouts and (co)limits...

  • 1
    $\begingroup$ Would you mind defining insertion and $\mathrm{cod}$? $\endgroup$ – Rasmus Aug 20 '11 at 10:07
  • $\begingroup$ Given a set $X$ and a subset $Y$ of $X$, the insertion (aka inclusion, aka immersion) of $Y$ is the restriction $I_X|_Y:Y \to X$ of the identity map of $X$ to $Y$. $\mathrm{cod}(f)$ means "codomain of $f$". $\endgroup$ – kjo Aug 20 '11 at 10:25
  • $\begingroup$ This follows from the faithfulness of the forgetfull functor of your concrete category. $\endgroup$ – Rasmus Aug 20 '11 at 10:33
  • $\begingroup$ @Rasmus: how could this categorical proof work in the case where $\mathcal{C}$ is $\mathbf{Set}$ (so the forgetful functor doesn't enter into the picture)? BTW, I have a vague idea of what functors are, and what the forgetful functor is, thanks to Wikipedia, though I have not really reached functors yet in my independent study... $\endgroup$ – kjo Aug 20 '11 at 10:38
  • $\begingroup$ Well, if $\mathcal C$ is the category of sets, then $g$ is just $f$, but regarded as a function to $K$ instead of $M$, which is possible since $f(T)\subseteq K$. $\endgroup$ – Rasmus Aug 20 '11 at 10:40

The concept you need is the image of an arrow $f : T \to M$. It is defined by the obvious universal property: a monomorphism $\operatorname{im} f : H \rightarrowtail M$ is the image of $f$ just if there is an arrow $c : T \to H$ such that $f = (\operatorname{im} f) \circ c$ and for every morphism $g : T \to K$ and every monomorphism $h : K \to M$ such that $f = h \circ g$, there is a (necessarily unique) arrow $k : H \to K$ such that $g = k \circ c$ and $\operatorname{im} f = h \circ k$.

Exercise. Show that $\operatorname{im} f$ is unique up to unique isomorphism.

The trouble is that $\operatorname{im} f$ as defined above is not guaranteed to exist unless the category is nice enough. This happens, for example, when the category is a topos or an abelian category.

Once we have $\operatorname{im} f$, it's clear how we may obtain $g : T \to K$ given $i : K \rightarrowtail M$ and the fact that we have a monomorphism $m: H \rightarrowtail K$ (i.e. the witness of the inclusion $H \subseteq K$) — we just take the composite $m \circ c$.

A related concept is that of the regular coimage. This makes sense in any category with finite limits and colimits. First, we take the pullback of $f$ along $f$ to obtain the kernel pair $p_1, p_2 : S \to T$. (For example, in the category of sets, $S = \{ (t_1, t_2) \in T \times T : f(t_1) = f(t_2) \}$.) The regular coimage of $f$ is the coequaliser $e : T \to H$ of $p_1$ and $p_2$. Notice that since $f \circ p_1 = f \circ p_2$, there must be an arrow $m : H \to M$ such that $f = m \circ e$. In any regular category (e.g. a topos or an abelian category) this $m$ will also be the image in the sense defined above.

I don't know of a name for the phenomenon where two arrows differ ‘only in the codomain’. But it's easy enough to formulate this categorically: this happens when the (regular) coimage of the two arrows are isomorphic.

| cite | improve this answer | |
  • $\begingroup$ Thanks! I'm still working through the details of your answer. What is the exact definition witness (as in "the witness of the inclusion $H \subseteq K$) in CT? I've consulted the indices of several books, but none mention this term. $\endgroup$ – kjo Aug 20 '11 at 12:38
  • 1
    $\begingroup$ It's a term from logic, which I am using informally. A ‘witness’ is a specific instance of a thing which is asserted to exist. For example, when we say $H \subseteq K$ in the ETCS, we are asserting that there is a monomorphism $H \rightarrowtail K$; so a witness is a specific such monomorphism. (Actually, this is an abuse of notation, because we're only supposed to write $m \subseteq K$ where $m$ is a monomorphism into $K$!) Or a more elementary example: $2$ is a witness to the assertion that $6$ is composite, because $2$ is a factor of $6$. $\endgroup$ – Zhen Lin Aug 20 '11 at 12:59

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.