Let $G$ be a finite abelian group, and $g_1, \ldots, g_n \in G$ such that the cyclic groups that they generate are in direct sum $\langle g_1 \rangle \oplus \cdots \oplus \langle g_n \rangle$. Is it always possible to find elements $h_1, \ldots, h_n \in G$ and integers $a_1, \ldots, a_n$ such that the following three facts hold

  1. $g_i = a_i \cdot h_i$, for all $1 \leq i \leq n$,

  2. the cyclic subgroups generated by the $h_i$ are in direct sum, $H:=\langle h_1 \rangle \oplus \cdots \oplus \langle h_n \rangle$.

  3. $H$ is a direct summand of $G$?

(I asked this question on math.stackexchange, see https://math.stackexchange.com/questions/199928/smallest-pure-subgroup-containing-a-fixed-subgroup)

  • $\begingroup$ Do you really mean that the $g_i$'s generate a cyclic group $C$? if so, why can't you take all of the $h_i$ to be the same element, a generator of the largest cyclic group containing $C$. My first thought is that this solution would satisfy (1), (2) and (3)... But I could be wrong. $\endgroup$
    – Nick Gill
    Commented Sep 21, 2012 at 13:04
  • $\begingroup$ No, I have edited my question. "The cyclic groupS that they generate" insted of "The cyclic group that they generate" $\endgroup$
    – calc
    Commented Sep 21, 2012 at 13:17
  • 1
    $\begingroup$ It should be obvious. Your $H$ is the pure subgroup generated by $g_i$-s, and in finite groups, pure subgroups should be the same as direct summands. $\endgroup$
    – Bugs Bunny
    Commented Sep 21, 2012 at 13:21
  • $\begingroup$ @Bugs Bunny Can you give a reference for the "pure subgroup generated by"? Also, how do you find the generators $h_i$ each dividing $g_i$? $\endgroup$
    – calc
    Commented Sep 21, 2012 at 13:23
  • $\begingroup$ en.wikipedia.org/wiki/Pure_subgroup $\endgroup$
    – Bugs Bunny
    Commented Sep 21, 2012 at 13:24

2 Answers 2


I am not convinced that this is true, because the pure subgroup generated by the $g_i$ might not have the stipulated form as a direct sum of the $h_i$.

Let $G = {\mathbb Z}/16{\mathbb Z} \oplus {\mathbb Z}/4{\mathbb Z}$, $n=1$, and $g_1=(4,2)$. What could $h_1$ be?

We can prove that there is no such $h_1$ as follows. For any $g \in G$ of order 16, we have $8g = (8,0)$. Since $(8,0) \in \langle h_1 \rangle$, $\langle h_1 \rangle$ cannot be a direct summand of $G$ of order 4. So $\langle h_1 \rangle$ would have to be a direct summand of order 16. But then $4h_1 = (4,0)$ or $(12,0)$ and so $g_1 \not\in \langle h_1 \rangle$.

  • $\begingroup$ I think you mean $8g = (8,0)$... $\endgroup$ Commented Sep 21, 2012 at 17:02
  • $\begingroup$ Thanks but there is something I am missing. Why do you not start with $g_1=(2,1)$? $\endgroup$
    – calc
    Commented Sep 21, 2012 at 17:03
  • $\begingroup$ Anyway I am still not convinced about the pure subgroup generated by the $g_i$. What is it? You generate a subgroup with the $g_i$. Then you take the roots. Then? Is this a subgroup at all? Or do you have to still take the subgroup generated and continue? $\endgroup$
    – calc
    Commented Sep 21, 2012 at 18:41
  • $\begingroup$ @calc: A pure subgroup $S$ of $G$ is a subgroup such that $nG\cap S=nS$ for all $n$. You may not need to take roots; for example, in $G=\mathbf{Z}_4\times\mathbf{Z}_2$, $S=\langle (1,1)\rangle$ is pure: $2G\cap S = \{(0,0),(2,0)\} = 2S$, and that's the only one you need to worry about. Even though $2(1,0)\in S$, you do not need $(1,0)$ in $S$ for $S$ to be pure, you just need $2(1,0)$ to be twice something in $S$. $\endgroup$ Commented Sep 21, 2012 at 21:27
  • $\begingroup$ @Calc: But if you have a subgroup $S$ of $G$, and you let $H=\{g\in G\mid ng\in S, ng\neq 0\}$ (which I suspect is what you mean; otherwise, in a torsion group, adding "all roots" would mean adding the entire thing), then $H$ need not be a subgroup; e.g., with $G$ and $S$ as above, $(1,0)\in H$ (since $2(1,0)=(2,0)\in S$ is nonzero), and $(1,1)\in H$ (since it is in $S$), but $(1,1)-(0,1)\notin H$. $\endgroup$ Commented Sep 21, 2012 at 21:31

This old question just surfaced on MO for some reason, so I thought I'd add a comment to put Derek Holt's excellent answer in a wider context. Given a prime $p$, the problem of classifying pairs consisting of a finite abelian $p$-group of exponent $p^n$ together with a subgroup, up to isomorphism of pairs, is known to be of wild representation type for $n$ at least seven. See for example the paper of Ringel and Schmidmeier, "Submodule categories of wild representation type" (JPAA 2006). If what you're asking were true, it would imply a classification for such pairs. But more generally, trying to tame such pairs is destined to failure.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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