Let $K$ be a cyclic group. Let $\phi,\psi: K\rightarrow Aut(H)$ be group homomorphisms such that there exists $\zeta\in Aut(H)$ satisfying $\phi(K)=\zeta \psi(K)\zeta^{-1}$. Then can we prove $H\rtimes_{\phi}K\simeq H\rtimes_{\psi}H$?

My idea is to use the following

Theorem: Let $K,H$ be groups. Let $\phi,\psi: K\rightarrow Aut(H)$ be group homomorphisms and $\zeta\in Aut(H)$, $\alpha\in Aut(K)$ and $\psi=\sigma_{\zeta}\circ\phi\circ\alpha$, where $\sigma_{\zeta}\circ \phi\circ\alpha(k)=\zeta(\phi(\alpha(k)))\zeta^{-1}$ for all $k\in K$. Then $H\rtimes_{\phi}K\simeq H\rtimes_{\psi}H$.

Let $K=\langle a\rangle$ generated by $a$. From the condition $\phi(K)=\zeta \psi(K)\zeta^{-1}$, we have $\phi(a)=\zeta \psi(a^n)\zeta^{-1}$ for some integer $n$. Then how to continue? Thanks.


If $$\zeta \varphi(a)\zeta^{-1} = \psi(a)^t$$ I believe the map $$ (h, a^i) \mapsto (h^{\zeta}, a^{it} ) $$ provides the required isomorphism from the semidirect product with respect to $\varphi$ to the one with respect to $\psi$. Here I am writing morphisms as exponents.

(Here I am only addressing the first question, with $K = \langle a \rangle $ cyclic.)

Addendum Having seen the second answer by Derek Holt, I now realise this is only a morphism, which should be an isomorphism only when $a \mapsto a^t$ is.

  • $\begingroup$ I think that every automorphism of a quotient of a finite cyclic group lifts to an automorphism of the whole group, so the result is true for finite cyclic groups $K$. $\endgroup$ – Derek Holt Jan 2 '14 at 14:26

The problem is that the action of $\zeta$ might not lift to the whole of $K$.

As an example, let $H=\langle z \mid z^7=1 \rangle = C_7$ and $K = \langle x,y \mid x^3=y^7=1,x^{-1}yx=y^2 \rangle$ be the nonabelian group of order $21$.

Define $\phi,\psi$ by $\phi(x)= (z \mapsto z^2), \phi(y)= 1$, and $\psi(x)= (z \mapsto z^4), \psi(y)= 1$.

So $\phi(K)=\psi(K)$ and we can take $\zeta=1$.

The two semidirect products are

$\langle x,y,z \mid x^3=y^7=z^7=1,yz=zy,x^{-1}yx=y^2,x^{-1}zx=z^2 \rangle$, and

$\langle x,y,z \mid x^3=y^7=z^7=1,yz=zy,x^{-1}yx=y^2,x^{-1}zx=z^4 \rangle$,

which are not isomorphic.

Added later: I realize that this is not a counterexample to the question asked, because $K$ is not cyclic. But I think the example is interesting, so I will not delete it.

  • $\begingroup$ It's nothing to do with opinions! The problem is that $K$ is not cyclic in my example! I got confused, because I would normally denote the normal subgroup by $K$ and the non-normal subgroup by $H$, and the OP has done the oppopsite. $\endgroup$ – Derek Holt Jan 2 '14 at 10:40

Second attempt. I believe this is false when $K$ is infinite cyclic, because automorphisms of finite quotients of $K$ do not usually lift to automorphisms of $K$. For example, the two groups

$\langle x,y \mid x^{11}=1, y^{-1}xy=x^4 \rangle$ and $\langle x,y \mid x^{11}=1, y^{-1}xy=x^5 \rangle$

(with $H = \langle x \rangle$ and $K = \langle y \rangle$) are not isomorphic, but the images of $\phi$ and $\psi$ are both equal to subgroup of ${\rm Aut}(H)$ of order $5$.

  • $\begingroup$ Thanks, and I still have some questions.(1)What do you mean by "finite quotient of $K$"? (2)Since $ζ\in Aut(H)$, what do you mean by "action of $ζ$ might not lift to the whole of $K$"? $\endgroup$ – Andrews Feb 15 at 16:26
  • $\begingroup$ A finite quotient of $K$ is a finite quotient group $Q := K/N$ for some normal subgroup $N$ of $K$. For an automorphism $\zeta$ of $Q$, it is possible that there is no automorphism $\theta$ of $K$ with $\theta(N) = N$ such that $\theta(g)N = \zeta(gN)$ for all $g \in K$. In that case we say that $\zeta$ does not lift to an automorphism of $K$. $\endgroup$ – Derek Holt Feb 15 at 17:43
  • $\begingroup$ So in your answer,$H=<x>\cong \Bbb Z_{11},K=<y>\cong \Bbb Z$, you mean treat $H$ as finite quotient of $K$ and $\phi_1(y)(x)=x^4, \phi_2(y)(x)=x^5$, $\phi_1,\phi_2$ are automorphism of $H$ but can't lift to $K$, is that right? And why that's related to the initial proposition $H⋊_ϕK≃H⋊_ψH$ may not correct? $\endgroup$ – Andrews Feb 16 at 2:22
  • $\begingroup$ No, the finite quotient of $K$ in question is a subgroup of ${\rm Aut}(H)$ of order $5$. This subgroup is the image of both $\phi$ and $\psi$. I left the proof of non-isomorphism of the two semidirect products as an exercise. $\endgroup$ – Derek Holt Feb 16 at 10:06
  • $\begingroup$ Thanks so much for your help these days. $H=\langle x\rangle\cong \Bbb Z_{11}, K=\langle y \rangle \cong \Bbb Z. \text{Aut}(H)=\Bbb Z_{11}^\times \cong \Bbb Z_{10}.$ $\phi_1(y)(x)=x^4, \phi_2(y)(x)=x^5, \phi_1(K)=\phi_2(K)=\{\bar 1, \bar3, \bar4, \bar5, \bar9\}\cong \Bbb Z_5$. ker$\phi_1 \cong$ ker$\phi_2=5\Bbb Z$. After some calculation I can derive that $H\rtimes_{\phi_1} K \not \cong H\rtimes_{\phi_2} K$, but I still don't know what is "the" automorphism of $\phi_1(K)=\phi_2(K)\cong \Bbb Z_5$ which can't lift to the whole $K \cong \Bbb Z$. Could you please point it out? Thanks. $\endgroup$ – Andrews Feb 16 at 15:21

I think, this is false. Take $H=Z^5$, $K$ cyclic of order 5 generated by $k$, $\phi(k)$ an element $c$ of order 5 and $\psi(k)=c^2$. I think, the corresponding semidirect products are not isomorphic since $c$ is not conjugate to $c^2$ in the group $GL(5,Z)$.

  • $\begingroup$ I don't think this works. The two semidirect products are isomorphic with the map inducing identity on $H$ and mapping $c$ to $c^2$. I think the result is true whenever $\phi$ and $\psi$ are both injective. $\endgroup$ – Derek Holt Jan 2 '14 at 9:25
  • $\begingroup$ @Derek Holt: What you defined is not a homomorphism. $\endgroup$ – Moishe Kohan Jan 2 '14 at 16:14
  • $\begingroup$ Sorry, it should be $c \mapsto c^3$ for an isomorphism $H\rtimes_\phi K \to H \rtimes_\psi K$. Using $\circ_\phi$ and $\circ_\psi$ for composition in the two semidirect products, we get $(h_1,k^i) \circ_\phi (h_2,k^j) = (h_1c^i(h_2),k^{i+j}) \mapsto (h_1c^i(h_2),k^{3(i+j)}) = (h_1,k^{3i})\circ_\psi (h_2,k^{3j})$. $\endgroup$ – Derek Holt Jan 2 '14 at 17:23

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.