Assume there are two directed systems $(A_i,f_{ij})$ and $(A'_i,f'_{ij})$ of groups over the directed set $(\mathbb{N},\le)$, such that for any $i,j$ there exists isomorphisms $\phi_{ij}:A_i\to A_i'$ and $\psi_{ij}:A_j\to A'_j$ such that $f'_{ij}\circ\phi_{ij}=\psi_{ij}\circ f_{ij}$. Is it true that $\varinjlim A_i$ is isomorphic to $\varinjlim A'_i$?

If it is true, is there some preferred isomorphism? What about direct limits in other categories and other directed sets?


This is false.

Example. Let $A_i = A'_i = \mathbf Z^{(\mathbf N)}$ be an infinitely generated free abelian group. For $i < j$, let $f_{ij} \colon A_i \to A_j$ be the map that kills the first $j$ coordinates, and $f'_{ij} \colon A'_i \to A'_j$ the map that kills the even coordinates up to $2j$: \begin{align} f_i((a_j)_j) &= (0,\ldots,0,a_{j+1},\ldots),\\ f'_i((a_j)_j) &= (a_1,0,a_3,0,\ldots,a_{2j-1},0,a_{2j+1},a_{2j+2},\ldots). \end{align} It is clear that the $A_i$ and $A'_i$ are pairwise isomorphic (even in finite subsets). However, the colimits are not isomorphic: $$A = \underset{\longrightarrow}{\operatorname{colim}} A_i = 0,$$ since an element $(a_1,\ldots,a_j,0,0,\ldots)$ in some $A_i$ disappears in $A_j$. On the other hand, $$A' = \underset{\longrightarrow}{\operatorname{colim}} A'_i = \mathbf Z^{(2\mathbf N-1)}$$ consists of all finite sequences $(a_1,\ldots) \in \mathbf Z^{(\mathbf N)}$ supported on odd coordinates only. Indeed, the even coordinates all die by the same argument, and the odd ones never change in the colimit.

  • $\begingroup$ Thanks a lot! If assume in addition that $A_i,A'_i$ are finitely generated abelian groups, will it be true? $\endgroup$ – ZZY May 29 at 18:33
  • $\begingroup$ I did try to think about this, but I was unable to come up with an example or a proof. The obvious proof is to extend to finite subsets of $\mathbf N$ first, and then to the limit. Both steps could go wrong (I think), and I see no reason to expect a positive answer. But please prove me wrong! $\endgroup$ – R. van Dobben de Bruyn May 29 at 18:52
  • 1
    $\begingroup$ Maybe I am missing something, but how is it obvious that these isomorphisms of $A_i$ and $A_i'$ commute with the $f_{ij}$ and $f_{ij}'$? OP does want the isomorphisms to be such that $f_{ij}' \circ \phi_{ij} = \psi_{ij} \circ f_{ij}$ (also to me it seems a bit odd that an isomorphism $A_i \to A_i'$ should also depend on $j$, so I am not sure if the way it is phrased now actually gives naturality). $\endgroup$ – Mark Kamsma May 30 at 16:58
  • 1
    $\begingroup$ @MarkKamsma: the dependence on $\phi$ and $\psi$ on both $i$ and $j$ is exactly the point of the question. If there were a single $\phi_i \colon A_i \stackrel\sim\longrightarrow A'_i$ commuting with all transition morphisms, these would trivially induce an isomorphism on the colimit. $\endgroup$ – R. van Dobben de Bruyn May 30 at 17:59
  • $\begingroup$ @R.vanDobbendeBruyn Oh right, now I see. I did indeed miss something, a slightly different construction for the isomorphisms than I first had in mind does work. $\endgroup$ – Mark Kamsma May 31 at 13:15

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.