I have a question on the following proof of the characterisation of normal subgroups of groups which are direct products of non-abelian simple groups:

Theorem: Let $G = G_1 \times \cdots \times G_n$, where $G_1, \ldots, G_n$ are non-abelian and simple, and let $N \unlhd G$. Then there exists some set $J \subseteq \{1,\ldots, n\}$ such that $$ N = \times_{j \in J} G_j \quad \mbox{ and } \quad G_k \cap N = 1 \mbox{ for } k \notin J. $$

The proof goes by induction on $n$. I just show the relevant part:

Suppose $k \in \{1,\ldots, n\}$ with $N\cap G_k \ne 1$. Then $G_k \le N$ because $G_k$ is simple., let $\overline G/G_k$, then $$ \overline G = \times_{i\ne k} G_i/G_k $$ with $G_i / G_k \cong G_i$ for $i \ne k$. By induction hypothesis there exists $J' \subseteq \{1, \ldots, n \}$ with $k \notin J'$ such that $$ \overline N = \times_{j\in J'} G_j/G_k \quad \mbox{ with } \quad G_i/G_k \cap \overline N = 1 \mbox{ for } i \notin j'. $$ But then for $J := J' \cup \{ k \}$ we have $$ N = \times_{j\in J} G_j \quad \mbox{ with } \quad N\cap G_i = 1 \mbox{ for } i \notin J. $$

The last step I do not understand, the logic is something like ''because of $$ G/N = G'/N $$ we can conclude $G = G'$'', but as I found out this logic does not hold in general, see the Extension problem. So whats going on here?

(remark on notation: do not know how to do the big $\times$...)

  • $\begingroup$ As for notation, I think most people write \prod, which gives $G = \prod_{i=1}^n G_i$. $\endgroup$ – André 3000 Mar 12 '15 at 20:36
  • $\begingroup$ Yes, but in the textbook they write it with the big $\times$, and $\prod_i G_i$ is simply $G_1G_2 \cdots G_n$, which need not be a direct product, just an ordinary product, so guess using your notation would be suboptimal. $\endgroup$ – StefanH Mar 12 '15 at 20:38
  • $\begingroup$ It's probably overkill for this, but as far as your Extension problem remark goes: Ayoub proved that if $G=H\times K$ is a direct product of finite groups, then every short exact sequence $0\rightarrow H\rightarrow G\rightarrow K\rightarrow 0$ splits as a direct product. The proof can be generalized to groups with both chain conditions without a whole lot of difficulty, as an aside, and it has also been established for profinite groups by Guralnick et al., and for Chernikov groups by authors I can't remember. $\endgroup$ – zibadawa timmy Mar 12 '15 at 20:38
  • 1
    $\begingroup$ If $N$ is a normal subgroup of a group $G$ and $N \le G_i \le G$ for $i=1,2$, then $G_1/N = G_2/N \Rightarrow G_1=G_2$. That follows immediately from the definition of quotient groups: $g \in G_1 \Rightarrow gN \in G_1/N \Rightarrow gN \Rightarrow G_2/N \Rightarrow g \in G_2$. $\endgroup$ – Derek Holt Mar 12 '15 at 20:40
  • $\begingroup$ Before applying your inductive argument, you have to prove that there exists a $k$ with $N \cap G_k \ne 1$. $\endgroup$ – Derek Holt Mar 12 '15 at 20:43

You want to use the following result:

If $G$ is any group, $K$ is a direct factor of $G$, and $H$ is a subgroup of $G$ containing $K$, then $K$ is a direct factor of $H$.

In your case, you would use this to conclude that $G_k$ is a direct factor of $N$, and the inductive hypothesis tells you that the remaining factors (if any) are some (possibly empty) collection of the remaining $G_i$.


Proof of the claimed result:

Suppose $G=K\times L$ and $K\subseteq H\subseteq G$. For any $h\in H$ there exists $a\in K,b\in L$ with $h=ab$. Since $K\subseteq H$ we have $a^{-1}h = b\in H$. Setting $T=\{b\in L \ : \ ab\in H \mbox{ for some } a\in K\}$, we may conclude that $T$ is a subgroup of $H$ (and $L$, in fact) such that $H=KT$, $K\cap T=\{1_G\}$, and $[K,T]=\{1_G\}$. It follows that $H$ is the direct product of $K$ and $T$.

Note that you may use slightly different notation depending on how you opt to write out direct products like this. Namely you may write $h=(a,b)$ instead, in which case we'd then say $(a^{-1},1)h = (1,b)\in H$, etc. This is just a technical distinction between whether we interpret $G$, at the level of sets, as the actual cartesian product of the sets $K$ and $L$, or if we simply use the $\times$ to denote the direct product conditions without necessarily implying a Cartesian product of the underlying sets. The latter is the more general situation, and is what I was using.

In your case, we would thus have that $N=G_k\times L$ for some normal subgroup $L$ of $\times_{i\neq k} G_i$. Apply the inductive hypothesis to $L$. In this way we actually avoid the quotients entirely, thus avoiding your concerns about the extension problem. Though, see my comment about a result of Ayoub which shows the problem doesn't exist for direct products of certain groups.

  • $\begingroup$ Thanks for your reply, but I still have several questions. First, strictly, by the isomorphism theorems we just could have $$ \overline G \cong \times_{i\ne k} G_i/G_k $$ not equality! And if I got you right, this "inverse" application of the isomorphism of quotient groups just works if the groups involved are direct products? Seems to be a quite deep result, used in my introductory textbook, and being still in the first few chapters... also I guess all we can get is $N \cong \times_{j\in J} G_i$, so not equality, and this is considerable weaker as a group could contain many isomorphic copies? $\endgroup$ – StefanH Mar 13 '15 at 12:03
  • $\begingroup$ @Stefan I've added proof details to my answer. In short, if the subgroup contains a direct factor then you can simply multiply out the terms from that factor and what's left is the commuting complement desired. $\endgroup$ – zibadawa timmy Mar 13 '15 at 20:46

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.