$\DeclareMathOperator\SO{SO}$I posted this on MSE 10 days ago and it got 3 upvotes but no answers or comments, so I'm cross-posting to MO.

Background: The group of rotations $ \SO_3(\mathbb{R}) $ has a 24 element subgroup of integer points $ \SO_3(\mathbb{Z}) $ which is isomorphic to the symmetric group $ S_4 $.

Original Question: (answered almost immediately by LSpice with a great subgroup of $ \SO_5(\mathbb{R}) $)

Let $ G $ be a connected Lie group. For every injective group homomorphism $ \phi:\SO_3(\mathbb{Z}) \to G $ does there exist a unique extension $ \tilde{\phi} : \SO_3(\mathbb{R}) \to G $?

New Question:

Let $ G $ be a connected Lie group. Let $ \phi:\SO_3(\mathbb{Z}) \to G $ be a group homomorphism whose image is a maximal closed subgroup of $ G $. Does there exist a unique extension $ \tilde{\phi} : \SO_3(\mathbb{R}) \to G $?

Note that the only quotients of $ S_4 $ are $ S_4 $, cyclic 2 group trivial group and the six element dihedral group $ D_6 $. Of these only $ S_4 $ itself is a maximal closed subgroup of $ \SO_3(\mathbb{R}) $.

For $ G= \SO_3(\mathbb{R}) $ the answer is yes to both questions. Here is a proof:

First we show existence of an extension. Define $ \Gamma:=\SO_3(\mathbb{Z}) $ and $ \Gamma':=\phi(\SO_3(\mathbb{Z})) $. Isomorphic finite subgroups of $ \SO_3(\mathbb{R}) $ are conjugate. So there always exists some $ g \in G $ such that $ g \Gamma g^{-1}=\Gamma' $. So the map conjugation by $ g $ agrees with $ \phi $ up to some automorphism. From there we can conjugate by some $ x \in \Gamma $ to get an exact match (recall every automorphism of the symmetric group $ S_4 $ is inner). So conjugation by $ xg $ is an automorphism of $ G $ extending $ \phi $.

Now we show uniqueness of extensions. Any two extensions $ \phi_1 $ and $ \phi_2 $ of $ \phi $ are automorphisms of $ \SO_3(\mathbb{R}) $. Every automorphism of $ G $ is inner, so the extensions $ \phi_1$, $\phi_2 $ correspond to conjugation by $ g_1$, $g_2 $. Thus conjugation by $ g_2^{-1}g_1 $ fixes $ \Gamma $ pointwise. In other words $ g_2^{-1}g_1 $ centralizes $ \Gamma $. So in particular $ g_2^{-1}g_1 \in N_G(\Gamma) $. But $ \Gamma $ is self-normalizing $ N_G(\Gamma)=\Gamma $. So really $ g_2^{-1}g_1 \in \Gamma $. However $ \Gamma $ also has trivial center $ Z(\Gamma)=1 $. So we must have $ g_1=g_2 $ and thus uniqueness is proven.

EDIT: As Lspice points out the answer to the first question is no there does not always exist an extension and for the second question Lspice notes that if such an extension exists then $ G $ must be $ SO_3(\mathbb{R}) $.

First we show uniqueness. If $ \phi $ is trivial there is nothing to prove. If $ \phi $ is trivial and there exists an extension $ \tilde{\phi} $ then $ \tilde{\phi} $ must be injective since $ \SO_3(\mathbb{R}) $ is simple so since $ \Gamma':=\phi(\SO_3(\mathbb{Z})) $ is maximal closed and contained in $ \tilde{\phi}(\SO_3(\mathbb{R}))\cong \SO_3(\mathbb{R})$ we can conclude that $ G=\tilde{\phi}(\SO_3(\mathbb{R})) $.

Now we show existence. Suppose that $ \Gamma' $ is a maximal closed subgroup. Then $ G $ must be compact since $ \Gamma' $ is contained in a maximal compact. Also $ \Gamma' $ must contain the whole center otherwise it is not maximal closed (and $ G $ cannot be abelian because then it contains no maximal closed subgroups). So $ G $ must be semisimple since $ \Gamma' $ is finite. In fact I claim $ G $ must be simple. Otherwise we can pass to the universal cover and say $ G_1\times G_2 $ are two different semi simple factors with the lift of $ \Gamma' $ intersecting $ G_1 $ non trivially. Then the direct product of $ G_2 $ with the projection onto $ G_1 $ of lift of $ \Gamma' $ is a closed subgroup. So descending back to $ G $ we have that $ \Gamma' $ is not maximal closed, a contradiction. This is a general proof that a Lie group with a finite maximal closed subgroup must be simple and compact. For here we specialize to the case of $ S_4 $ subgroup. The only quotients are $ 1, C_2,D_6, S_4 $. It is fairly clear that of these only $ S_4 $ can be maximal closed in a Lie group since only an embedded $ S_4 $ can act irreducibly with respect to the adjoint representation of a compact simple Lie group. Moreover, the only compact simple Lie group for which this is true. Thus any connected Lie group containing a quotient of $ S_4 $ as a maximal closed subgroup must just be $ SO_3(\mathbb{R}) $.

  • 2
    $\begingroup$ The matrices of the form $w \oplus \det(w)^{-1}$ with $w$ a $4 \times 4$ permutation matrix form an $\operatorname S_4$ in $\operatorname{SL}_5(\mathbb R)$ that, I naïvely guess, does not sit inside an $\operatorname{SO}_3(\mathbb R)$ in $\operatorname{SL}_5(\mathbb R)$. $\endgroup$
    – LSpice
    Feb 24, 2022 at 3:18
  • 1
    $\begingroup$ Surely your revised question is impossible unless $G$ is already $\operatorname{SO}_3(\mathbb R)$? Indeed, $\tilde\phi(\operatorname{SO}_3(\mathbb R))$ is a compact, hence closed, subgroup of $G$ containing $\phi(\operatorname{SO}_3(\mathbb Z))$, properly since it is connected, hence must equal $G$; so that $G$ is a quotient of $\operatorname{SO}_3(\mathbb R)$—but $\operatorname{SO}_3(\mathbb R)$ is simple. $\endgroup$
    – LSpice
    Feb 24, 2022 at 5:02

1 Answer 1


Consider the group of matrices of the form $w \oplus \det(w)^{-1}$, with $w$ a $4\times4$ permutation matrix. This is an $\operatorname S_4$ inside $\operatorname{SL}_5(\mathbb R) \subseteq \operatorname{SL}_5(\mathbb C)$, and $\mathbb C^5$ decomposes as a sum of the trivial, the sign, and the reflection representation of $\operatorname S_4$. In particular, the space of fixed vectors for $\operatorname S_4$ in $\mathbb C^5$ is $1$-dimensional.

If this $\operatorname S_4$ were contained in an $\operatorname{SO}_3(\mathbb R)$ in $\operatorname{SL}_5(\mathbb R)$, or even in $\operatorname{SL}_5(\mathbb C)$, then the resulting (complex) $5$-dimensional representation of $\operatorname{SO}_3(\mathbb R)$ would be $0^5$, $2 + 0^2$, or $4$.

Since $\sum_{g \in \operatorname{SO}_3(\mathbb Z)} \chi_4(g) = 1(5) + 6(1) + 3(1) + 8(-1) + 6(-1)$ equals $0$, there are no fixed vectors for $\operatorname{SO}_3(\mathbb Z)$ in $4$. Clearly, there is at least a $2$-dimensional space of fixed vectors for $\operatorname{SO}_3(\mathbb Z)$ in $2 + 0^2$ and in $0^5$.

  • $\begingroup$ ya another more geometric way to see it is that you are basically splitting up $ S_4 $ between the two connected components of an $ O_4 $ inside $ SL_5 $ ( $ A_4 $ in $ SO_4 $ and the 12 odd parity elements of $ S_4 $ in the other component of $ O_4 $). So it is not possible for any connected group to contain such a copy of $ S_4 $ since it would intersect both connected components of $ O_4 $ nontrivially contradicting connectedness. In particular $ SO_3 $ is connected so it cannot contain your $ S_4 $ subgroup. $\endgroup$ Feb 24, 2022 at 4:21
  • $\begingroup$ Do you mind if I slightly change the question to perhaps make it a bit closer to what I'm really interested in? $\endgroup$ Feb 24, 2022 at 4:23
  • $\begingroup$ @IanGershonTeixeira, I'm not sure I understand your argument. It is certainly possible for a connected group to intersect multiple components of a disconnected group, so I don't think your argument is yet a proof. Or were you just discussing an intuition? $\endgroup$
    – LSpice
    Feb 24, 2022 at 4:23
  • $\begingroup$ @IanGershonTeixeira, in terms of changing the question, I think the usual etiquette after an answer has been given is to ask a new question. However, no, I do not mind, as long as you clearly indicate that the question was changed after this answer was given. $\endgroup$
    – LSpice
    Feb 24, 2022 at 4:24
  • $\begingroup$ Ya I was thinking along the lines of intuition. And I really appreciate that! I just updated the question and I made sure to acknowledge your answer and say why I changed the question. $\endgroup$ Feb 24, 2022 at 4:50

Your Answer

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

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