# The smallest subobject $\sum{A_i}$ containing a family of subobjects {$A_i$}

In an Abelian category $\mathcal{A}$, let {$A_i$} be a family of subobjects of an object $A$. How to show that if $\mathcal{A}$ is cocomplete(i.e. the coproduct always exists in $\mathcal{A}$), then there is a smallest subobject $\sum{A_i}$ of $A$ containing all of $A_i$?

Surely this $\sum{A_i}$ cannot be the coproduct of {$A_i$}, but I have no clue what it should be.

-
push out of pull back of $A_i \to A$ – Alexander Thumm Aug 20 '11 at 9:31
@Alexander: That only works for a pair of subobjects. If you have three subobjects, the pullback may be trivial even when there are non-trivial pairwise intersections. – Zhen Lin Aug 20 '11 at 10:01
@Zhen: for three objects $A_1 + A_2 + A_3 = (A_1 + A_2) + A_3$ obviously – Alexander Thumm Aug 20 '11 at 10:38
You seem to have found the right generalization though. – Alexander Thumm Aug 20 '11 at 10:47

You are quite right that it can't be the coproduct, since that is in general not a subobject of $A$. Here are two ways of constructing the desired subobject:
1. As Pierre-Yves suggested in the comments, the easiest way is to take the image of the canonical map $\bigoplus_i A_i \to A$. This works in any cocomplete category with unique epi-mono factorisation.
2. Alternatively, the subobject $\sum A_i$ can be constructed by taking the colimit over the semilattice of the $A_i$ and their intersections. This construction can be carried out in any bicomplete category, but is not guaranteed to give a subobject of $A$ unless the category is nice enough.
Dear Zhen Li: I'll tell you what I'd have answered, and you (or someone else) will tell me why it's incorrect, if you're patient enough. The coproduct maps naturally to $A$, with some kernel. Thus, the quotient of the coproduct by the kernel can be viewed as a subobject. – Pierre-Yves Gaillard Aug 20 '11 at 9:45