Take the 2-minute tour ×
Mathematics Stack Exchange is a question and answer site for people studying math at any level and professionals in related fields. It's 100% free, no registration required.

I'm reviewing for exams and came across this problem from an older exam:

Let $G$ and $H$ be groups and let $\phi,\psi : H\to \operatorname{Aut}(G)$ be homomorphisms from $H$ into the group of automorphisms of $G$. If there exists $f\in \operatorname{Aut}(H)$ and $a\in\operatorname{Aut}(G)$ such that $\psi(f(h))=a\phi(h)a^{-1}$ as automorphisms of $G$ for all $h\in H$, then $G\rtimes_\phi H \cong G\rtimes_\psi H$.

I've proved this (unless I've made a mistake) by explicitly defining a map $F:G\rtimes_\phi H \to G\rtimes_\psi H$ by $(g,h)\mapsto(a(g),f(h))$ and checking that it is an isomorphism. However, choosing the map mostly involved guess-and-checking. That left me feeling like I somehow didn't do it "properly" and that I might have missed something I was supposed to notice or understand. I realize this is somewhat subjective, but I'm a bit suspicious, because the way I went about it makes the problem just seem like a strange, tedious way to test that someone knows the definition of a semidirect product.

Is there some other way to go about this problem (maybe using some clever argument about conjugation by $a$ inside the automorphism group?) that somehow illuminates or suggests something about the structure going on, or should it otherwise be obvious in some way what the isomorphism is? I guess my question is: is there something deeper going on here?

share|improve this question
add comment

3 Answers

up vote 3 down vote accepted

I think that eventually you have to get down to defining the map explicitly and checking that it is an isomorphism.

But, conceptually: in general, an automorphism of $H$ will not induce an automorphism of $G\rtimes_{\phi} H$, because the action of $f(h)$ need not be related to the action of $h$ (of course, you get a different semidirect product, $G\rtimes_{\phi\circ f}H$, which may or may not be isomorphic). The automorphism $a$ provides the necessary "correction" by having $f(h)$ act on $a^{-1}(g)$ (instead of on $g$), so that the action of $f(h)$ on $a^{-1}(g)$ agrees with the original action of $h$ on $g$.

Intuitively: $H$ used to act on $G$ "in English", but after applying $f$ it is now trying to act "in Russian"; $a$ translates from Russian to English, so $(a^{-1}(g))^{f(h)}$ now makes sense, because $a^{-1}(g)$ is "in Russian", which $f(h)$ understands; then you translate the result back into English by applying $a$. It works exactly like a change-of-basis automorphism works for linear transformations. Interpreted that way, $(g,h)\mapsto (a(g),f(h))$ makes sense (that is, seems like the obvious thing to try), because we are simply performing the translations in both coordinates.

share|improve this answer
Thanks. The conceptual comments and the analogy both help a lot. –  matt May 9 '11 at 20:04
add comment

Both cases $G\rtimes_\phi H$ and $G\rtimes_\psi H$ are extensions of the group H by G. Extensions are classified by short exact sequences with in your case kernel G and image group H. The semi-direct products are exactly the so-called split extensions. To prove this the mentioned conjugation action appears naturally.

share|improve this answer
add comment

Denoting $c_a$ for conjugation with $a$, the assumption means that the following diagram commutes (how can I draw nicer diagrams?): $$\begin{array}{rcl} H\quad & \stackrel{f}{\longrightarrow} &\quad H \\ {\scriptstyle\phi}\downarrow\ \ & &\ \downarrow{\scriptstyle \psi}\\ \text{Aut}(H) & \stackrel{c_a}{\longrightarrow} & \text{Aut}(H) \end{array}$$

where the horizontal arrows are group isomorphisms. As Arturo explained, this is similar to a 'change of basis' of the actions $\phi$ and $\psi$.

share|improve this answer
There are better ways, but at least the horizontal arrows can be done with \stackrel. The rest, I sort of tried to align and size by hand. –  Arturo Magidin May 9 '11 at 20:00
add comment

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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