5
$\begingroup$

I wondered whether there is an analogue of the Krull-Schmidt theorem for real Lie-groups. More precisely, what conditions do you have to impose on a connected finite dimensional Lie group $G$ so that for each direct product decomposition $G=G_1 \times ... \times G_s$ the isomorphism classes of $G_1,...,G_s$ are uniquely determined up to reordering?

I'm mainly interested in compact Lie groups, maybe this restriction simplifies the matter.

$\endgroup$
3
  • 2
    $\begingroup$ I guess you mean connected? Because otherwise, all discrete groups are real Lie groups... $\endgroup$
    – YCor
    Jan 27, 2015 at 22:04
  • 1
    $\begingroup$ @YCor: Of course. $\endgroup$
    – Dominik
    Jan 28, 2015 at 15:06
  • 1
    $\begingroup$ Then each direct summand is a connected Lie subgroup, and the chains of connected closed subgroups have bounded length (bounded by the dimension). Is this enough to make the proof of Krull-Schmidt work? $\endgroup$
    – YCor
    Jan 28, 2015 at 15:47

2 Answers 2

1
$\begingroup$

I see that you are restricting to (connected) compact Lie groups. This means that $G$ is a product of a semi-simple group and a central torus. The semi-simple piece decomposes uniquely (up to isomorphism) into a product of normal subgroups and the torus decomposes $T^n=S^1 \times \ldots \times S^1$. Thus

\begin{equation} G=G_{ss_1}\times\ldots \times G_{ss_k}\times S^1 \times \ldots S^1 \end{equation}

Where the first $k$ terms are semi-simple. If you demand that the factors are indecomposable and that they are Lie subgroups, ie no irrational winding on the torus, then this decomposition is unique. I hope this is what you wanted.

I have also been thinking about the non-compact case. I'm going to sketch my thoughts so we can get some discussion going, but there will be mistakes and it is unfinished. If G is semi-simple or abelian we get the same decomposition as before. Suppose $G$ is not semi-simple or solvable. Levi decomposition does give us something, but the Levi factor $G_{ss}$ is not normal in general, and it might even be non-closed which would probably stop us right here. On the Lie algebra level, we can split off the kernel $\mathfrak{g}_{ss_0}=\text{Ker}(\varphi)$ of the $\mathfrak{g}_{ss}$-representation $\varphi:\mathfrak{g}_{ss}\rightarrow \text{End}(\mathfrak{r})$, where $\mathfrak{r}$ is the solvable radical of $\mathfrak{g}$. $\mathfrak{g}_{ss_0}$ decomposes uniquely as it is semi-simple.

I don't know what to do with the complement to $\mathfrak{g}_{ss_0}$, and describing the unique decomposition directly seems hard. The approach I was considering was to use that $\mathfrak{g}_{ss}$-representations decompose into irreducibles, that an ideal of the radical is necessarily a submodule, and possibly split up $\mathfrak{g}_{ss}$ further by taking the kernels of sub-representations. In the end this didn't seem to help much.

$\endgroup$
5
  • 1
    $\begingroup$ It is true that any connected compact Lie group decomposes uniquely into a product of a semisimple group and a central torus. However, this product does not have to be direct. For example, $SU(n)$ is not a direct product of a semisimple group and a torus. $\endgroup$ Feb 1, 2015 at 21:15
  • 1
    $\begingroup$ @MikhailBorovoi I would call $SU(n)$ semi-simple. It does not need to decompose further. There is no torus in this case. $\endgroup$ Feb 2, 2015 at 8:38
  • 1
    $\begingroup$ Excuse me, I meant that $U(n)$ is not a direct product of a semisimple group and a torus. $\endgroup$ Feb 3, 2015 at 9:54
  • 1
    $\begingroup$ @MikhailBorovoi I see the problem now. $U(n)$ is only covered by $SU(n)\times S^1$, not equal to. If I find the time I will edit the answer to account for this situation. Thank you. $\endgroup$ Feb 5, 2015 at 13:53
  • 2
    $\begingroup$ Even if there is no centre, a semisimple group need not be the direct product of its normal, almost simple subgroups. Consider, for example, the quotient $G$ of $\operatorname{SU}_2 \times \operatorname{SU}_2$ by the diagonally embedded central $\mu_2$. The relevant subgroups of this are both $\operatorname{SU}_2$, but the quotient of $G$ by one of the $\operatorname{SU}_2$s is the adjoint group $\operatorname{SU}_2/\mu_2$, not $\operatorname{SU}_2$. $\endgroup$
    – LSpice
    Mar 7, 2023 at 0:06
1
$\begingroup$

For ease of writing, I'll say a compact connected Lie group $G$ has the Krull–Schmidt property if, up to order, it has a unique decomposition as a direct product where each factor itself cannot be decomposed any further.

One relatively nice sufficient condition for $G$ to have the Krull–Schmidt property given as follows.

Proposition: Suppose $G$ is a compact connected Lie group and that the center of $G$ is at most one dimensional. Then $G$ has the Krull–Schmidt property.

Sketch: The group $G$ has a cover of the form $G_0\times G_1\times \dotsb \times G_n$ where $G_0 = (S^1)^a$, $a\in \{0,1\}$ and every other $G_i$ is non-abelian and simple (in the sense that any proper normal subgroup is finite). Call the covering map $\pi$.

If $N\trianglelefteq G$ is a closed normal subgroup, then it is easy to see that $N$ is of the form $\pi(N_0\times N_1\times \dotsb \times N_n)$ where $N_0 = \{1\}$ or $(S^1)^a$, and each $N_i$ is either trivial or equal to $G_i$. (See, e.g., this MSE question. This doesn't use the fact that $a\leq 1$.)

Now, suppose we have a decomposition $G = K_1\times \dotsb\times K_k$ with each $K_i$ indecomposable. As each factor is normal, it follows from the previous paragraph that each $\pi(G_i)$ is contained in some $K_j$. (Note that if $a > 1$, in general $\pi(G_0)$ need not be contained in a single $K_j$.)

Since each $K_i$ intersects $K_j$ only in the identity, it follows that if $\pi(G_i)$ and $\pi(G_j)$ intersect non-trivially (i.e., in more than just the identity), they must like in the same $K_\ell$ factor.

Now, create an equivalence relation on $\{G_0,\dotsc, G_n\}$ as the transitive closure of the symmetric, reflexive relation where $G_i\sim G_j$ if $\pi(G_i)\cap\pi(G_j)$ contains more than the identity. This equivalence relation, of course, partitions $\{G_0,\dotsc, G_n\}$. For any set in the partition, all the $G_i$ in that set must be contained in the same $K_j$. If some $K_j$ contains elements from two different sets in the partition, then that $K_j$ can be decomposed further, giving a contradiction.

Thus, we find that the unique decomposition $G= K_1\times \dotsb\times K_k$ with each $K_i$ indecomposable is obtained from the above partition. Since the partition is uniqueley defined by the group, $G$ must have the Krull–Schmidt property $\square$.

On the other hand, if you allow $G$ to have a center of dimension $2$, it may fail to have the Krull–Schmidt property.

$\DeclareMathOperator\SU{SU}$Proposition: Suppose $\mu$ denotes a fixed non-trivial $3$rd root of unity. Then the group $G:= (S^1\times S^1\times \SU(2)\times \SU(3))/ \langle (1,-1,-I_2,I_3), (1,\mu, I_2, \mu I_3)\rangle$, where $I_k$ denotes the $k\times k$ identity mtarix, does not have the Krull–Schmidt property.

Sketch: Note that the first $S^1$ factor does not participate in the quotienting. As such, $G$ naturally has a decomposition as $G = S^1 \times (S^1\times \SU(2)\times \SU(3)/ \langle (-1,-I_2,I_3), (\mu, I_2, \mu I_3)\rangle)$. Regardless of whether the second factor further decomposes (it doesn't) it's obvious that this decomposition has an $S^1$ factor. We will now find another decomposition of $G$ which has no $S^1$ factor.

To that end, we consider the subgroups $U,V\subseteq S^1\times S^1$ where $$U = \{(u^2,u):u\in S^1\} \text{ and } V = \{(v^3,v):v\in S^1\}.$$

It is easy to verify that $U\cap V = \{(1,1)\}$ and that $U\cdot V = S^1\times S^1$. That is, every $(z,zw)\in S^1\times S^1$ has a unique representation in the form $uv$ with $u\in U$ and $v\in V$.

So, we can replace the $S^1\times S^1$ factor with $U\times V$. The old point $(1,-1) \in S^1\times S^1$ has the form $(-1,1)$ in $U\times V$ and the old point $(1,\mu)\in S^1\times S^1$ has the form $(1,\mu)\in U\times V$.

Thus, we see $$G \cong (U\times V\times \SU(2)\times \SU(3))/ \langle (-1,1,-I_2,I_3), (1,-\mu, I_2, \mu I_3)\rangle.$$ In this new description, the point $(-1,1,-I_2, I_3)$ only affects $U\times \{1\}\times \SU(2) \times \{I_3\}$, and has quotient $U(2)$, while the point $(1,-\mu, I_2, \mu I_3)$ only affects $\{1\}\times V\times \{I_2\}\times \SU(3)$, and has quotient $U(3)$. Thus, we have shown that $G \cong U(2)\times U(3)$.

To complete the proof, we need only note that neither $U(2)$ and $U(3)$ can split off an $S^1$ in a decomposition. It is easy to convince yourself the only possible decompositions look like $S^1\times H$ where $H$ is finitely covered by $\SU(2)$ or $\SU(3)$. But if $H$ is simply connected, then $Z(S^1\times H)\cong S^1\times \mathbb{Z}/k\mathbb{Z}$ is not isomorphic to $Z(U(k))\cong S^1$ (for $k=2,3$) since the former group has too many elements of order $k$. If $H$ is not simply connected, then $\pi_1(U(k))\cong\mathbb{Z}$ is not isomorphic to $\pi_1(S^1\times H) \cong \mathbb{Z}\times \mathbb{Z}/k\mathbb{Z}$. $\square$

$\endgroup$

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.