군론

Group Extensions

Group extensions as short exact sequences and semidirect products

Posted Updated

In this post we discuss extensions of groups.

Group Extensions

Definition 1 For two groups \(G, F\), an extension of \(G\) by \(F\), \(\mathcal{E}\), means the following pair satisfying the condition \(\ker p=\im i\):

\[\mathcal{E}: F\overset{i}{\hookrightarrow}E\overset{p}{\twoheadrightarrow}G\]

The intuition for this is the following pair

\[\mathcal{E}_0: F \rightarrow F\oplus G \rightarrow G\]

which we call the trivial extension. However, in the above situation, while the relation

\[G\cong E/\ker p=E/\im i\]

holds in general by the first isomorphism theorem, one must be careful because this does not imply that

\[E\cong (E/i(F))\oplus i(F)\]

holds. In any case, from this computation we see that for \(E\) to be an extension of \(G\) by \(F\), it is equivalent that \(E\) has a suitable normal subgroup \(F'\) such that \(F'\) is isomorphic to \(F\) and \(E/F'\) is isomorphic to \(G\).

Now, for fixed \(G\) and \(F\), the collection of all extensions of \(G\) by \(F\) forms a category. Morphisms here are given as follows.

Definition 2 For two extensions \(\mathcal{E}_1: F \rightarrow E_1 \rightarrow G\) and \(\mathcal{E}_2:F \rightarrow E_2 \rightarrow G\), a morphism from \(\mathcal{E}_1\) to \(\mathcal{E}_2\) means \(u:E_1 \rightarrow E_2\) making the following diagram

commute.

Then if \(u:E_1 \rightarrow E_2\) is an isomorphism as a group homomorphism, the inverse of \(u\), \(u^{-1}: E_2 \rightarrow E_1\), also satisfies the condition of Definition 2, and therefore we can verify that \(u\) is an isomorphism as a morphism between extensions.

Extending the previous definition, we define the following.

Definition 3 If an extension of \(G\) by \(F\), \(\mathcal{E}:F \rightarrow E \rightarrow G\), is isomorphic to the extension

\[\mathcal{E}_0:F \rightarrow F\oplus G \rightarrow G\]

then we call it a trivial extension.

Then the following holds.

Proposition 4 For an extension \(\mathcal{E}:F \rightarrow E \rightarrow G\), the following are all equivalent.

  1. \(\mathcal{E}\) is a trivial extension.
  2. A retraction \(r: E \rightarrow F\) exists.
  3. A section \(s: G \rightarrow E\) exists such that \(s(G)\) can be chosen to lie in the centralizer of \(i(F)\).

Of course, here retraction and section mean group homomorphisms, not merely functions. ([Set Theory] §Retraction and Section, ⁋Definition 2)

Proof of Proposition 4

First, assume the first condition and consider the following diagram:

Then from this, we may define the retraction \(r:E \rightarrow F\) as \(\pr_1\circ u\), and \(s:G \rightarrow E\) as \(u^{-1}\circ\iota_2\).

Conversely, assume that the second condition holds. Then \((r,p): E \rightarrow F\oplus G\) is an isomorphism between the given extension and \(F \rightarrow F\oplus G \rightarrow G\). Similarly, assume the third condition. Then, since \(s(G)\) is contained in the centralizer of \(i(F)\), from \(F\oplus G\), via the weak direct product of \(F\) and \(G\), we can construct a morphism to \(E\).

If \(i(F)\) is contained in the center of \(E\), \(Z(E)\), then in the third condition we may ignore the relationship between \(s(G)\) and \(i(F)\). ([Algebraic Structures] §Group Action, ⁋Definition 12)

Definition 5 An extension \(\mathcal{E}:F \rightarrow E \rightarrow G\) is called a central extension if the image of \(F\) in \(E\) is contained in the center of \(E\).

Semidirect Products of Groups

On the other hand, the reason extensions other than the trivial extension exist is that, as seen above, for a group \(G\) and any of its normal subgroups \(N\), the formula

\[G\cong (G/N)\oplus N\]

does not always hold. However, the converse does not always hold either.

Definition 6 Let two groups \(N,H\) and a group homomorphism \(\tau:H \rightarrow \Aut(N)\) be given. Then the semi-direct product of \(N\) and \(H\) with respect to \(\tau\), \(N\rtimes_\tau H\), is the group given by the set \(N\times H\) with the following operation:

\[(x_1,y_1)(x_2,y_2)=(x_1\tau(y_1)(x_2), y_1y_2)\]

Then one can show that \(N\rtimes_\tau H\) has a group structure under the above operation; in this case the identity element of \(N\rtimes_\tau H\) is \((e_N, e_H)\) and the inverse of \((x,y)\) is \((\tau(y^{-1})(x^{-1}), y^{-1})\). Moreover, the following holds.

Proposition 7 Define two maps \(i: N \rightarrow N\rtimes_\tau H\) and \(p: N\rtimes_\tau H\rightarrow H\) by the formulas

\[i(x)=(x, e_H),\qquad p(x,y)=y\]

Then these maps are group homomorphisms, and

\[\mathcal{E}_\tau: N \overset{i}{\rightarrow} N\rtimes_\tau H\overset{p}{\rightarrow} H\]

obtained from them is an extension of \(H\) by \(N\). Moreover, if we define a map \(s: H \rightarrow N\rtimes_\tau H\) by the formula

\[s(y)=(e_N, y)\]

then \(s\) is a section of \(p\). In particular, if \(\tau\) is a trivial homomorphism, then since \(s(H)\) is contained in the centralizer of \(i(N)\), by Proposition 4, \(\mathcal{E}_\tau\) is a trivial extension.

The proof of this is a straightforward computation.

Now suppose the \(N,H\) examined above are subgroups of a specific group \(G\). If \(N\) is a normal subgroup of \(G\), then for each \(h\in H\), the inner automorphism \(\rho_h\) defined by it is an automorphism of \(N\), and thus \(\rho: H \rightarrow \Aut(N)\) is defined. ([Algebraic Structures] §Group Action, ⁋Definition 10) Then from the above proposition we obtain the following.

Corollary 8 Let a group \(G\), and in \(G\) a normal subgroup \(N\), and in \(G\) a subgroup \(H\) be given. If \(N\cap H=\{e_G\}\) and \(NH=G\) hold, then the group homomorphism defined by the formula

\[N\rtimes_\rho H \rightarrow G;\qquad (x,y)\mapsto xy\]

is an isomorphism.

Proof

It suffices to construct an inverse. To do this, for an arbitrary element \(g\) of \(G\), we must find \(x\in N\) and \(y\in H\) satisfying \(g=xy\); such \(x, y\) exist by the condition \(NH=G\), and they are unique by the condition $N\cap H={e_G}. The rest is a straightforward computation.

In this case we say that \(G\) is the (internal) semi-direct product of \(N\) and \(H\). The difference between the external semi-direct product and the internal semi-direct product is merely where we started from, and is not important.


References

[Bou] Bourbaki, N. Algebra I. Elements of Mathematics. Springer. 1998.


Comments

  • Public repository: a deleted comment stays in the Git history.
  • The password and the email are encrypted and kept in Cloudflare KV alone.

No comments yet.

Leave a comment