4

Does the category $\mathbf{FreeAb}$ of free abelian groups have sequential colimits?

A sequential colimit is a colimit of the countable shape

$\bullet \to \bullet \to \bullet \to \cdots,$

i.e. indexed by $(\mathbb{N},\leq)$. This is a special case of a filtered colimit. A few weeks ago Jeremy Rickard showed in MSE/5025660 that $\mathbf{FreeAb}$ does not have filtered colimits, and although the filtered colimit there (the finitely generated subgroups of $\mathbb{Z}^{\mathbb{N}}$) was countable, it is clearly not sequential. I am not sure if it can be made sequential somehow, but I doubt it, since I think we would need the cardinal $\aleph_1$ to index such a sequence, not $\aleph_0$, which I need.

Notice that, as in the linked question, I am not asking if $\mathbf{FreeAb}$ is closed under sequential colimits taken in $\mathbf{Ab}$. This is a stronger property, and it fails miserably: the colimit of $\mathbb{Z} \xrightarrow{2} \mathbb{Z} \xrightarrow{2} \mathbb{Z} \xrightarrow{2} \cdots$ in $\mathbf{Ab}$ is not free, it is $\mathbb{Z}[1/2]$. But the colimit of that sequence in $\mathbf{FreeAb}$ does exist and is zero.

Of course, some other sequential colimits also exist and are even the same as in $\mathbf{Ab}$, such as $\mathbb{Z} \hookrightarrow \mathbb{Z}^2 \hookrightarrow \mathbb{Z}^3 \hookrightarrow \cdots$ with transition maps $x \mapsto (\dotsc x,0)$, whose colimit is the free abelian group $\mathbb{Z}^{\oplus \mathbb{N}}$. Another interesting example is the sequence of maps $\mathbb{Z}^n \to \mathbb{Z}^{n+1}$, $a \mapsto (a,a)$, but I don't know how the colimit looks like here, concretely.

The question is equivalent to: given a sequential colimit of free abelian groups taken in $\mathbf{Ab}$, does it have a "liberation"? More generally, we might ask which abelian groups admit a "liberation" at all, except for the free abelian groups of course. By what I wrote before, $\mathbb{Z}[1/2]$ admits a "liberation", namely $0$.

  • @downvoter: what can I improve? – Martin Brandenburg Feb 15 '25 at 15:54
  • Maybe I misunderstand the question, but what you call "sequential colimit" seems to be what is sometimes also called "colimit of a chain". More precisely, those colimits whose diagram is an ordinal. For those it is well known that their existence is equivalent to the existence of directed/filtered colimits. – Mark Kamsma Feb 16 '25 at 09:30
  • 1
    No, for a sequential colimit the index set is $\mathbb{N}$. I will clarify this. (But it is of course also written in the nLab page that I linked to.) – Martin Brandenburg Feb 16 '25 at 09:31
  • 1
    Maybe worth noting : the category of torsion-free abelian groups is a reflective subcategory of abelian groups, so it has all colimits (even if they may differ from the colimits in $\mathbf{Ab}$), and it coincides with free abelian groups in the finitely generated case. In particular, all finitely generated abelian groups admit a "liberation" : just quotient out the torsion. But I does not help much for examples like $\mathbb{Q}$, which is torsion-free but not free. – Arnaud D. Feb 17 '25 at 09:08

0 Answers0