§11 Short Exact Sequences and Semidirect Products
§11.1 Extensions—Short Exact Sequences
Definition 11.1 A short exact sequence of groups is a sequence consisting of two homomorphisms
satisfying the following conditions:
(1) G → H G\to H G → H is injective,
(2) H → K H\to K H → K is surjective, and
(3) the kernel of H → K H\to K H → K is equal to—not merely isomorphic to—the image of G → H G\to H G → H .
A short exact sequence is usually written as
1 → G → H → K → 1. 1\to G\to H\to K\to 1. 1 → G → H → K → 1.
Definition 11.2 We also say that H H H is an extension of K K K by G G G .
Question: What do the 1 1 1 's at the two ends mean?
The symbol 1 1 1 denotes the trivial group with one element.
The sequence above is “exact” because the image of each homomorphism is the kernel of the next homomorphism.
For example, the portion
means that the image of
is the kernel of
which means that
is injective.
For reasons that will become clearer later, short exact sequences are important because of the following idea:
we think of the group H H H as being built out of the groups G G G and K K K .
Example 11.1 We have the following short exact sequences:
(1)
Z / 2 Z → Z / 4 Z → Z / 2 Z , \mathbb{Z}/2\mathbb{Z}
\to
\mathbb{Z}/4\mathbb{Z}
\to
\mathbb{Z}/2\mathbb{Z}, Z /2 Z → Z /4 Z → Z /2 Z ,
where the first homomorphism sends
1 ↦ 2 ∈ Z / 4 Z , 1\mapsto2\in\mathbb{Z}/4\mathbb{Z}, 1 ↦ 2 ∈ Z /4 Z ,
and the second homomorphism sends
and
1 , 3 ↦ 0 ∈ Z / 2 Z . 1,3\mapsto0\in\mathbb{Z}/2\mathbb{Z}. 1 , 3 ↦ 0 ∈ Z /2 Z .
(2)
Z / 2 Z → Z / 2 Z × Z / 2 Z → Z / 2 Z . \mathbb{Z}/2\mathbb{Z}
\to
\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/2\mathbb{Z}
\to
\mathbb{Z}/2\mathbb{Z}. Z /2 Z → Z /2 Z × Z /2 Z → Z /2 Z .
We take
a ↦ ( a , 0 ) a\mapsto(a,0) a ↦ ( a , 0 )
and
( a , b ) ↦ b . (a,b)\mapsto b. ( a , b ) ↦ b .
Thus we may think of both
Z / 4 Z \mathbb{Z}/4\mathbb{Z} Z /4 Z
and the Klein four group as being built from two copies of
Z / 2 Z . \mathbb{Z}/2\mathbb{Z}. Z /2 Z .
But we see that different groups can be constructed from
Z / 2 Z \mathbb{Z}/2\mathbb{Z} Z /2 Z
in different ways.
We also have the following short exact sequences:
(3)
Z / 3 Z → S 3 → Z / 2 Z , \mathbb{Z}/3\mathbb{Z}
\to
S_3
\to
\mathbb{Z}/2\mathbb{Z}, Z /3 Z → S 3 → Z /2 Z ,
where
A 3 ≅ Z / 3 Z , A_3\cong\mathbb{Z}/3\mathbb{Z}, A 3 ≅ Z /3 Z ,
and this is the short exact sequence associated with the inclusion
A 3 → S 3 . A_3\to S_3. A 3 → S 3 .
and
(4)
Z / 3 Z → Z / 6 Z → Z / 2 Z . \mathbb{Z}/3\mathbb{Z}
\to
\mathbb{Z}/6\mathbb{Z}
\to
\mathbb{Z}/2\mathbb{Z}. Z /3 Z → Z /6 Z → Z /2 Z .
Thus we see that there are at least two different ways to construct a group of order 6 6 6 from
Z / 3 Z \mathbb{Z}/3\mathbb{Z} Z /3 Z
and
Z / 2 Z . \mathbb{Z}/2\mathbb{Z}. Z /2 Z .
Remark The examples above show that:
(1) an extension need not be a direct product;
(2) an extension
need not admit a map
such that
K → H → K = i d K ; K\to H\to K
=
\mathrm{id}_K; K → H → K = id K ;
(3) an extension of Abelian groups can be non-Abelian.
§11.2 Split Short Exact Sequences
Definition 11.3 If there exists a homomorphism
such that
K → H → K = i d K , K\to H\to K
=
\mathrm{id}_K, K → H → K = id K ,
we say that the short exact sequence
splits .
For readability, from now on we will write a short exact sequence
as
where L L L stands for “left” and R R R stands for “right”.
Since
is injective, from now on we will identify L L L with its image in H H H in order to simplify notation.
Example 11.2 Among the short exact sequences above:
(1)
Z / 2 Z → Z / 4 Z → Z / 2 Z , \mathbb{Z}/2\mathbb{Z}
\to
\mathbb{Z}/4\mathbb{Z}
\to
\mathbb{Z}/2\mathbb{Z}, Z /2 Z → Z /4 Z → Z /2 Z ,
(2)
Z / 2 Z → Z / 2 Z × Z / 2 Z → Z / 2 Z , \mathbb{Z}/2\mathbb{Z}
\to
\mathbb{Z}/2\mathbb{Z}\times\mathbb{Z}/2\mathbb{Z}
\to
\mathbb{Z}/2\mathbb{Z}, Z /2 Z → Z /2 Z × Z /2 Z → Z /2 Z ,
(3)
Z / 3 Z → S 3 → Z / 2 Z , \mathbb{Z}/3\mathbb{Z}
\to
S_3
\to
\mathbb{Z}/2\mathbb{Z}, Z /3 Z → S 3 → Z /2 Z ,
(4)
Z / 3 Z → Z / 6 Z → Z / 2 Z . \mathbb{Z}/3\mathbb{Z}
\to
\mathbb{Z}/6\mathbb{Z}
\to
\mathbb{Z}/2\mathbb{Z}. Z /3 Z → Z /6 Z → Z /2 Z .
Only (1) does not split.
Notice that, a priori, there is no reason to regard R R R as a subgroup of H H H .
For example, in
Z / 2 Z → Z / 4 Z → Z / 2 Z , \mathbb{Z}/2\mathbb{Z}
\to
\mathbb{Z}/4\mathbb{Z}
\to
\mathbb{Z}/2\mathbb{Z}, Z /2 Z → Z /4 Z → Z /2 Z ,
the second copy of
Z / 2 Z \mathbb{Z}/2\mathbb{Z} Z /2 Z
cannot naturally be “embedded” back into
Z / 4 Z \mathbb{Z}/4\mathbb{Z} Z /4 Z .
Proposition 11.1
Z / 2 Z → Z / 4 Z → Z / 2 Z \mathbb{Z}/2\mathbb{Z}
\to
\mathbb{Z}/4\mathbb{Z}
\to
\mathbb{Z}/2\mathbb{Z} Z /2 Z → Z /4 Z → Z /2 Z
does not split.
Proof:
Z / 2 Z \mathbb{Z}/2\mathbb{Z} Z /2 Z
contains only elements of order 1 1 1 and 2 2 2 .
Therefore, no homomorphism
j : Z / 2 Z → Z / 4 Z j:\mathbb{Z}/2\mathbb{Z}
\to
\mathbb{Z}/4\mathbb{Z} j : Z /2 Z → Z /4 Z
can have an image containing an element of order
⩾ 3 \geqslant3 ⩾ 3 .
Indeed, if
then we must also have
j ( g ) n = 1. j(g)^n=1. j ( g ) n = 1.
But in
Z / 4 Z \mathbb{Z}/4\mathbb{Z} Z /4 Z ,
both
[ 1 ] [1] [ 1 ]
and
[ 3 ] [3] [ 3 ]
have order 4 4 4 :
⟨ [ 1 ] ⟩ = { [ 1 ] , [ 2 ] , [ 3 ] , [ 0 ] } , ⟨ [ 3 ] ⟩ = { [ 3 ] , [ 6 ] = [ 2 ] , [ 5 ] = [ 1 ] , [ 0 ] } . \langle[1]\rangle
=
\{[1],[2],[3],[0]\},
\quad
\langle[3]\rangle
=
\{[3],[6]=[2],[5]=[1],[0]\}. ⟨[ 1 ]⟩ = {[ 1 ] , [ 2 ] , [ 3 ] , [ 0 ]} , ⟨[ 3 ]⟩ = {[ 3 ] , [ 6 ] = [ 2 ] , [ 5 ] = [ 1 ] , [ 0 ]} .
Therefore the image of any homomorphism
j : Z / 2 Z → Z / 4 Z j:\mathbb{Z}/2\mathbb{Z}
\to
\mathbb{Z}/4\mathbb{Z} j : Z /2 Z → Z /4 Z
must lie inside
{ [ 0 ] , [ 2 ] } ⊂ Z / 4 Z . \{[0],[2]\}
\subset
\mathbb{Z}/4\mathbb{Z}. {[ 0 ] , [ 2 ]} ⊂ Z /4 Z .
But this is the kernel of the map
Z / 4 Z → Z / 2 Z . \mathbb{Z}/4\mathbb{Z}
\to
\mathbb{Z}/2\mathbb{Z}. Z /4 Z → Z /2 Z .
Therefore no j j j can split the identity map on
R = Z / 2 Z . R=\mathbb{Z}/2\mathbb{Z}. R = Z /2 Z .
Here is a more dramatic example.
Example 11.3 The short exact sequence
Z → × n Z → Z / n Z \mathbb{Z}
\xrightarrow{\times n}
\mathbb{Z}
\to
\mathbb{Z}/n\mathbb{Z} Z × n Z → Z / n Z
does not split for any
n ≠ − 1 , 0 , 1. n\neq-1,0,1. n = − 1 , 0 , 1.
Any homomorphism from
Z / n Z \mathbb{Z}/n\mathbb{Z} Z / n Z
to
Z \mathbb{Z} Z
would have to send an element of order n n n to an element of finite order.
But Z \mathbb{Z} Z has no nonzero elements of finite order.
Therefore there is no injection
Z / n Z → Z . \mathbb{Z}/n\mathbb{Z}
\to
\mathbb{Z}. Z / n Z → Z .
Since this is our first attempt to understand short exact sequences, let us analyze the case in which R R R can be regarded as a subgroup of H H H .
If both L L L and R R R live inside H H H , then perhaps it is easier to accept the idea that H H H is “constructed” from L L L and R R R .
This leads us to the following definition.
Definition 11.4 If there exists a group homomorphism
such that the composite
R → j H → R R\xrightarrow{j}H\to R R j H → R
equals
then we say that the short exact sequence splits .
A choice of
is called a splitting .
Thus, if the short exact sequence is given by homomorphisms
ϕ : L → H , ψ : H → R , \phi:L\to H,
\qquad
\psi:H\to R, ϕ : L → H , ψ : H → R ,
then the definition means
ψ ∘ j = i d R . \psi\circ j
=
\mathrm{id}_R. ψ ∘ j = id R .
In particular, j j j is injective.
§11.3 Semidirect Products
In the example above, it is clear that
Z / n Z \mathbb{Z}/n\mathbb{Z} Z / n Z
cannot be regarded as a subgroup of
Z \mathbb{Z} Z .
So we arrive at a new idea.
We would like to be able to recognize semidirect products naturally, and we would also like to produce examples.
Let us analyze the situation.
As above, in a split short exact sequence, identify R R R with j ( R ) j(R) j ( R ) .
Then every element of R R R defines an action on H H H by conjugation:
h ↦ r h r − 1 . h\mapsto rhr^{-1}. h ↦ r h r − 1 .
Since L L L is normal,
r L r − 1 = L , rLr^{-1}=L, r L r − 1 = L ,
so this defines an action on L L L :
C r : l ↦ r l r − 1 . C_r:l\mapsto rlr^{-1}. C r : l ↦ r l r − 1 .
Moreover, this is a group isomorphism from L L L to itself.
We can prove that this defines a group homomorphism
R → A u t ( L ) , R\to\mathrm{Aut}(L), R → Aut ( L ) ,
given by
Equivalently, it is enough to prove
C r ∘ C r ′ = C r r ′ . C_r\circ C_{r^{\prime}}
=
C_{rr^{\prime}}. C r ∘ C r ′ = C r r ′ .
Indeed, for
h ∈ R h\in R h ∈ R ,
( C r ∘ C r ′ ) ( h ) = C r ( C r ′ ( h ) ) = C r ( r ′ h r ′ − 1 ) = r ( r ′ h r ′ − 1 ) r − 1 . (C_r\circ C_{r^{\prime}})(h)
=
C_r(C_{r^{\prime}}(h))
=
C_r(r^{\prime}hr^{\prime-1})
=
r(r^{\prime}hr^{\prime-1})r^{-1}. ( C r ∘ C r ′ ) ( h ) = C r ( C r ′ ( h )) = C r ( r ′ h r ′ − 1 ) = r ( r ′ h r ′ − 1 ) r − 1 .
Using associativity of the group operation,
( C r ∘ C r ′ ) ( h ) = ( r r ′ ) h ( r r ′ ) − 1 = C r r ′ ( h ) . (C_r\circ C_{r^{\prime}})(h)
=
(rr^{\prime})h(rr^{\prime})^{-1}
=
C_{rr^{\prime}}(h). ( C r ∘ C r ′ ) ( h ) = ( r r ′ ) h ( r r ′ ) − 1 = C r r ′ ( h ) .
This new map
R → A u t ( L ) R\to\mathrm{Aut}(L) R → Aut ( L )
is called the conjugation action of R R R on L L L .
Thus every splitting gives rise to a group homomorphism
R → A u t ( L ) . R\to\mathrm{Aut}(L). R → Aut ( L ) .
Here
A u t ( L ) \mathrm{Aut}(L) Aut ( L )
means the automorphism group of L L L as a group, not merely as a set.
Question: Fix two groups R R R and L L L .
A natural question is:
does every group homomorphism
R → A u t ( L ) R\to\mathrm{Aut}(L) R → Aut ( L )
produce a split exact sequence?
Another observation is that, given a splitting, both R R R and L L L become subgroups of H H H .
Moreover, their intersection contains only
Indeed, if a nonidentity element
existed, then the map
could not be injective.
Finally, since the orbits of the action of L L L cover all of H H H , we obtain
H = ⋃ r ∈ R L r . H
=
\bigcup_{r\in R}Lr. H = r ∈ R ⋃ L r .
That is,
Definition 11.5 Let L L L and R R R be subgroups of H H H . We define
L R = { g such that g = l r for some l ∈ L , r ∈ R } . LR
=
\{g~\text{such that}~g=lr
~\text{for some}~
l\in L,\ r\in R\}. L R = { g such that g = l r for some l ∈ L , r ∈ R } .
To prove that
when the short exact sequence
splits, we need the following lemma.
Lemma 11.2 Let
L → H → ψ R L\to H\xrightarrow{\psi}R L → H ψ R
be a short exact sequence.
Let
be the quotient homomorphism
Then there exists an isomorphism
such that
z ∘ q = ψ . z\circ q=\psi. z ∘ q = ψ .
That is, there exists z z z such that the diagram
commutes.
Proof: We are given a short exact sequence
L → H → ψ R , L\to H\xrightarrow{\psi}R, L → H ψ R ,
where
is surjective and
ker ( ψ ) = L . \ker(\psi)=L. ker ( ψ ) = L .
We want to prove that there exists an isomorphism
such that
z ∘ q = ψ , z\circ q=\psi, z ∘ q = ψ ,
where
is the quotient map.
Since ψ \psi ψ is surjective and
ker ( ψ ) = L , \ker(\psi)=L, ker ( ψ ) = L ,
the First Isomorphism Theorem gives
Define
by
z ( L h ) = ψ ( h ) . z(Lh)=\psi(h). z ( L h ) = ψ ( h ) .
This is well defined because if
L h 1 = L h 2 , Lh_1=Lh_2, L h 1 = L h 2 ,
then
h 1 h 2 − 1 ∈ L = ker ( ψ ) , h_1h_2^{-1}\in L=\ker(\psi), h 1 h 2 − 1 ∈ L = ker ( ψ ) ,
so
ψ ( h 1 ) = ψ ( h 2 ) . \psi(h_1)=\psi(h_2). ψ ( h 1 ) = ψ ( h 2 ) .
The map z z z is an isomorphism because:
z z z is a homomorphism since ψ \psi ψ is a homomorphism;
z z z is surjective since ψ \psi ψ is surjective;
z z z is injective because the kernel of z z z is precisely L L L , meaning that the only element mapped to the identity of R R R is L L L .
For every
h ∈ H h\in H h ∈ H ,
( z ∘ q ) ( h ) = z ( L h ) = ψ ( h ) , (z\circ q)(h)
=
z(Lh)
=
\psi(h), ( z ∘ q ) ( h ) = z ( L h ) = ψ ( h ) ,
so
z ∘ q = ψ . z\circ q=\psi. z ∘ q = ψ .
Therefore z z z is the required isomorphism making the diagram commute.
By the First Isomorphism Theorem, there exists an isomorphism
such that
z ∘ q = ψ , z\circ q=\psi, z ∘ q = ψ ,
which proves the lemma.
Once we have this lemma, we can prove the following corollary.
Corollary 11.3 If
is a splitting of
then
H = ⋃ r ∈ R L j ( r ) . H
=
\bigcup_{r\in R}Lj(r). H = r ∈ R ⋃ L j ( r ) .
Proof: By the definition of a splitting,
ψ ∘ j = i d R . \psi\circ j
=
\mathrm{id}_R. ψ ∘ j = id R .
On the other hand, by the lemma,
ψ = z ∘ q . \psi=z\circ q. ψ = z ∘ q .
Thus
z ∘ q ∘ j = i d R . z\circ q\circ j
=
\mathrm{id}_R. z ∘ q ∘ j = id R .
Since z z z is a group isomorphism, its inverse is a homomorphism, and therefore
q ∘ j = z − 1 . q\circ j
=
z^{-1}. q ∘ j = z − 1 .
Now let us interpret
The homomorphism q q q sends h h h to
Therefore,
sends r r r to the coset
L j ( r ) ∈ H / L . Lj(r)\in H/L. L j ( r ) ∈ H / L .
Since
z − 1 : R → H / L z^{-1}:R\to H/L z − 1 : R → H / L
is a bijection, for every coset
there exists a unique
r ∈ R r\in R r ∈ R
such that
L h = L j ( r ) . Lh=Lj(r). L h = L j ( r ) .
Since
⋃ H / L L h = H , \bigcup_{H/L}Lh
=
H, H / L ⋃ L h = H ,
this proves
⋃ r ∈ R L j ( r ) = H . \bigcup_{r\in R}Lj(r)
=
H. r ∈ R ⋃ L j ( r ) = H .
In the discussion above, we used j j j to identify
r ∈ R r\in R r ∈ R
with its image in H H H , so we write
⋃ r ∈ R L r = H . \bigcup_{r\in R}Lr
=
H. r ∈ R ⋃ L r = H .
Question: Fix
L , R ⊂ H . L,R\subset H. L , R ⊂ H .
If
L ∩ R = { 1 } , L\cap R=\{1\}, L ∩ R = { 1 } ,
L ⊂ H L\subset H L ⊂ H is normal, and
is H H H the semidirect product of L L L and R R R ?
This is exactly the right question, because the answer is yes.
Theorem 11.4 Fix a normal subgroup
and suppose
The following conditions are equivalent:
(1) A homomorphism
splits the short exact sequence
(2) An isomorphism
onto a subgroup
R ′ ⊂ H R^{\prime}\subset H R ′ ⊂ H
such that
R ′ ∩ L = { 1 } R^{\prime}\cap L=\{1\} R ′ ∩ L = { 1 }
and the map
L × R ′ → H L\times R^{\prime}\to H L × R ′ → H
is surjective.
(3) A group homomorphism
ϕ : R → A u t ( L ) . \phi:R\to\mathrm{Aut}(L). ϕ : R → Aut ( L ) .
Among these descriptions, we like the last one best because it makes no reference to the group H H H .
Once we construct a group homomorphism
ϕ : R → A u t ( L ) , \phi:R\to\mathrm{Aut}(L), ϕ : R → Aut ( L ) ,
we can construct a short exact sequence
Question: What is the group operation on H H H in terms of R R R and L L L ?
Proposition 11.5 Any homomorphism
ϕ : R → A u t ( L ) , r ↦ ϕ r \begin{aligned}
\phi:R&\to\mathrm{Aut}(L),\\
r&\mapsto\phi_r
\end{aligned} ϕ : R r → Aut ( L ) , ↦ ϕ r
defines a group H H H and a split short exact sequence
such that:
(1) the following defines a group structure on the set
H × H → H ( l 1 , r 1 ) ⋅ ( l 2 , r 2 ) : = ( l 1 ⋅ ϕ r 1 ( l 2 ) , r 1 r 2 ) . \begin{aligned}
H\times H&\to H\\
(l_1,r_1)\cdot(l_2,r_2)
&:=
(l_1\cdot\phi_{r_1}(l_2),r_1r_2).
\end{aligned} H × H ( l 1 , r 1 ) ⋅ ( l 2 , r 2 ) → H := ( l 1 ⋅ ϕ r 1 ( l 2 ) , r 1 r 2 ) .
The right-hand side is almost the usual group operation on
L × R L\times R L × R ,
except that before multiplying
l 1 l_1 l 1
and
l 2 l_2 l 2 ,
we “twist” l 2 l_2 l 2 into another element of L L L , namely
ϕ r 1 ( l 2 ) , \phi_{r_1}(l_2), ϕ r 1 ( l 2 ) ,
the value of r 1 r_1 r 1 under the homomorphism
ϕ : R → A u t ( L ) . \phi:R\to\mathrm{Aut}(L). ϕ : R → Aut ( L ) .
Moreover,
(2) the set
is a normal subgroup isomorphic to L L L ;
(3) the set
is a subgroup isomorphic to R R R .
Proof: Associativity:
( l 1 , r 1 ) ⋅ ( ( l 2 , r 2 ) ⋅ ( l 3 , r 3 ) ) = ( l 1 , r 1 ) ⋅ ( l 2 ⋅ ϕ r 2 ( l 3 ) , r 2 r 3 ) = ( l 1 ⋅ ϕ r 1 ( l 2 ⋅ ϕ r 2 ( l 3 ) ) , r 1 ( r 2 r 3 ) ) = ( l 1 ⋅ ϕ r 1 ( l 2 ) ⋅ ϕ r 1 ( ϕ r 2 ( l 3 ) ) , r 1 ( r 2 r 3 ) ) = ( l 1 ⋅ ϕ r 1 ( l 2 ) ⋅ ϕ r 1 r 2 ( l 3 ) , ( r 1 r 2 ) r 3 ) = ( l 1 ⋅ ϕ r 1 ( l 2 ) , r 1 r 2 ) ⋅ ( l 3 , r 3 ) = ( ( l 1 , r 1 ) ⋅ ( l 2 , r 2 ) ) ⋅ ( l 3 , r 3 ) . \begin{aligned}
(l_1,r_1)\cdot((l_2,r_2)\cdot(l_3,r_3))
&=
(l_1,r_1)\cdot
(l_2\cdot\phi_{r_2}(l_3),r_2r_3)\\
&=
(l_1\cdot
\phi_{r_1}(l_2\cdot\phi_{r_2}(l_3)),
r_1(r_2r_3))\\
&=
(l_1\cdot
\phi_{r_1}(l_2)\cdot
\phi_{r_1}(\phi_{r_2}(l_3)),
r_1(r_2r_3))\\
&=
(l_1\cdot
\phi_{r_1}(l_2)\cdot
\phi_{r_1r_2}(l_3),
(r_1r_2)r_3)\\
&=
(l_1\cdot\phi_{r_1}(l_2),r_1r_2)
\cdot(l_3,r_3)\\
&=
((l_1,r_1)\cdot(l_2,r_2))
\cdot(l_3,r_3).
\end{aligned} ( l 1 , r 1 ) ⋅ (( l 2 , r 2 ) ⋅ ( l 3 , r 3 )) = ( l 1 , r 1 ) ⋅ ( l 2 ⋅ ϕ r 2 ( l 3 ) , r 2 r 3 ) = ( l 1 ⋅ ϕ r 1 ( l 2 ⋅ ϕ r 2 ( l 3 )) , r 1 ( r 2 r 3 )) = ( l 1 ⋅ ϕ r 1 ( l 2 ) ⋅ ϕ r 1 ( ϕ r 2 ( l 3 )) , r 1 ( r 2 r 3 )) = ( l 1 ⋅ ϕ r 1 ( l 2 ) ⋅ ϕ r 1 r 2 ( l 3 ) , ( r 1 r 2 ) r 3 ) = ( l 1 ⋅ ϕ r 1 ( l 2 ) , r 1 r 2 ) ⋅ ( l 3 , r 3 ) = (( l 1 , r 1 ) ⋅ ( l 2 , r 2 )) ⋅ ( l 3 , r 3 ) .
The third equality holds because
is a homomorphism.
The fourth equality holds because
ϕ : R → A u t ( L ) \phi:R\to\mathrm{Aut}(L) ϕ : R → Aut ( L )
is a homomorphism.
Identity:
( 1 L , 1 R ) ⋅ ( l , r ) = ( 1 L ⋅ ϕ 1 R ( l ) , 1 R ⋅ r ) = ( 1 L ⋅ l , 1 R ⋅ r ) = ( l , r ) . \begin{aligned}
(1_L,1_R)\cdot(l,r)
&=
(1_L\cdot\phi_{1_R}(l),1_R\cdot r)\\
&=
(1_L\cdot l,1_R\cdot r)\\
&=
(l,r).
\end{aligned} ( 1 L , 1 R ) ⋅ ( l , r ) = ( 1 L ⋅ ϕ 1 R ( l ) , 1 R ⋅ r ) = ( 1 L ⋅ l , 1 R ⋅ r ) = ( l , r ) .
The second equality follows because ϕ \phi ϕ is a homomorphism, so
Inverse:
Proposition 11.6
( l , r ) − 1 = ( ϕ r − 1 ( l − 1 ) , r − 1 ) . (l,r)^{-1}
=
(\phi_{r^{-1}}(l^{-1}),r^{-1}). ( l , r ) − 1 = ( ϕ r − 1 ( l − 1 ) , r − 1 ) .
Proof:
( l , r ) ⋅ ( ϕ r − 1 ( l − 1 ) , r − 1 ) = ( l ⋅ ϕ r ( ϕ r − 1 ( l − 1 ) ) , r r − 1 ) = ( l ⋅ ϕ r r − 1 ( l − 1 ) , r r − 1 ) = ( l ⋅ l − 1 , r r − 1 ) = ( 1 L , 1 R ) . \begin{aligned}
(l,r)\cdot
(\phi_{r^{-1}}(l^{-1}),r^{-1})
&=
(l\cdot
\phi_r(\phi_{r^{-1}}(l^{-1})),
rr^{-1})\\
&=
(l\cdot
\phi_{rr^{-1}}(l^{-1}),
rr^{-1})\\
&=
(l\cdot l^{-1},
rr^{-1})\\
&=
(1_L,1_R).
\end{aligned} ( l , r ) ⋅ ( ϕ r − 1 ( l − 1 ) , r − 1 ) = ( l ⋅ ϕ r ( ϕ r − 1 ( l − 1 )) , r r − 1 ) = ( l ⋅ ϕ r r − 1 ( l − 1 ) , r r − 1 ) = ( l ⋅ l − 1 , r r − 1 ) = ( 1 L , 1 R ) .
The second equality follows because
ϕ : R → A u t ( L ) \phi:R\to\mathrm{Aut}(L) ϕ : R → Aut ( L )
is a homomorphism, so
ϕ r ∘ ϕ r ′ = ϕ r r ′ . \phi_r\circ\phi_{r^{\prime}}
=
\phi_{rr^{\prime}}. ϕ r ∘ ϕ r ′ = ϕ r r ′ .
The third equality follows because ϕ \phi ϕ is a group homomorphism, so
ϕ 1 = i d L . \phi_1
=
\mathrm{id}_L. ϕ 1 = id L .
Also,
( ϕ r − 1 ( l − 1 ) , r − 1 ) ⋅ ( l , r ) = ( ϕ r − 1 ( l − 1 ) ⋅ ϕ r − 1 ( l ) , r − 1 r ) = ( ϕ r − 1 ( l − 1 l ) , r − 1 r ) = ( ϕ r − 1 ( 1 L ) , 1 R ) = ( 1 L , 1 R ) . \begin{aligned}
(\phi_{r^{-1}}(l^{-1}),r^{-1})
\cdot(l,r)
&=
(\phi_{r^{-1}}(l^{-1})
\cdot
\phi_{r^{-1}}(l),
r^{-1}r)\\
&=
(\phi_{r^{-1}}(l^{-1}l),
r^{-1}r)\\
&=
(\phi_{r^{-1}}(1_L),
1_R)\\
&=
(1_L,1_R).
\end{aligned} ( ϕ r − 1 ( l − 1 ) , r − 1 ) ⋅ ( l , r ) = ( ϕ r − 1 ( l − 1 ) ⋅ ϕ r − 1 ( l ) , r − 1 r ) = ( ϕ r − 1 ( l − 1 l ) , r − 1 r ) = ( ϕ r − 1 ( 1 L ) , 1 R ) = ( 1 L , 1 R ) .
The second equality holds because
ϕ r − 1 : L → L \phi_{r^{-1}}:L\to L ϕ r − 1 : L → L
is a group homomorphism.
The final equality holds because
ϕ r − 1 \phi_{r^{-1}} ϕ r − 1
is a group homomorphism.
Therefore, this really does define a group.
Definition 11.6 We denote this group by
L ⋊ ϕ R . L\rtimes_{\phi}R. L ⋊ ϕ R .
When ϕ \phi ϕ is understood, we write
and call L ⋊ R L\rtimes R L ⋊ R a semidirect product of L L L and R R R .
Remark We say “a” semidirect product because different choices of ϕ \phi ϕ may produce different groups.
Although the word “and” is usually insensitive to order, L L L and R R R play very different roles here.
Remark Why do we use the symbol
Usually, when N N N is a normal subgroup of G G G , we write
N ◃ G . N\triangleleft G. N ◃ G .
The symbol
is a hybrid of
for “normal subgroup” and
for “product”.
Any diagram
gives rise to a map
R → A u t ( L ) . R\to\mathrm{Aut}(L). R → Aut ( L ) .
How?
By conjugation, since
L ⊂ H is normal , C h : L → L l ↦ h l h − 1 \begin{aligned}
L\subset H
&~\text{is normal},\\
C_h:L&\to L\\
l&\mapsto hlh^{-1}
\end{aligned} L ⊂ H C h : L l is normal , → L ↦ h l h − 1
is a group automorphism of L L L .
By essentially the same argument as in the exercise, we obtain a homomorphism
H → A u t ( L ) h ↦ C h . \begin{aligned}
H&\to\mathrm{Aut}(L)\\
h&\mapsto C_h.
\end{aligned} H h → Aut ( L ) ↦ C h .
The composite
is exactly the homomorphism ϕ \phi ϕ .
Question: How does R R R act on
L ⊂ L ⋊ R ? L\subset L\rtimes R? L ⊂ L ⋊ R ?
Proposition 11.7
( 1 L , r ) ⋅ ( l , 1 R ) ⋅ ( 1 L , r − 1 ) = ( ϕ r ( l ) , 1 R ) . (1_L,r)\cdot(l,1_R)\cdot(1_L,r^{-1})
=
(\phi_r(l),1_R). ( 1 L , r ) ⋅ ( l , 1 R ) ⋅ ( 1 L , r − 1 ) = ( ϕ r ( l ) , 1 R ) .
In other words, conjugation by r r r in
recovers ϕ r \phi_r ϕ r .
Proof:
( 1 L , r ) ⋅ ( l , 1 R ) ⋅ ( 1 L , r − 1 ) = ( 1 L ⋅ ϕ r ( l ) , r ⋅ 1 R ) ⋅ ( 1 L , r − 1 ) = ( ϕ r ( l ) ⋅ ϕ ( 1 L ) , r r − 1 ) = ( ϕ r ( l ⋅ 1 L ) , 1 R ) = ( ϕ r ( l ) , 1 R ) . \begin{aligned}
(1_L,r)\cdot(l,1_R)\cdot(1_L,r^{-1})
&=
(1_L\cdot\phi_r(l),r\cdot1_R)
\cdot(1_L,r^{-1})\\
&=
(\phi_r(l)\cdot\phi(1_L),rr^{-1})\\
&=
(\phi_r(l\cdot1_L),1_R)\\
&=
(\phi_r(l),1_R).
\end{aligned} ( 1 L , r ) ⋅ ( l , 1 R ) ⋅ ( 1 L , r − 1 ) = ( 1 L ⋅ ϕ r ( l ) , r ⋅ 1 R ) ⋅ ( 1 L , r − 1 ) = ( ϕ r ( l ) ⋅ ϕ ( 1 L ) , r r − 1 ) = ( ϕ r ( l ⋅ 1 L ) , 1 R ) = ( ϕ r ( l ) , 1 R ) .
We now have enough ingredients to prove Theorem 11.4.
The key ideas are:
is normal, then
H ↷ L H\curvearrowright L H ↷ L
by conjugation.
then R R R also acts by conjugation:
R → ϕ A u t ( L ) . R\xrightarrow{\phi}\mathrm{Aut}(L). R ϕ Aut ( L ) .
L ⋊ R L\rtimes R L ⋊ R is a group in which the conjugation action of R R R on L L L agrees with ϕ \phi ϕ .
Now let us prove
H ≅ L ⋊ R . H\cong L\rtimes R. H ≅ L ⋊ R .
To see why
L ⋊ R ≅ H , L\rtimes R\cong H, L ⋊ R ≅ H ,
we need a lemma.
Lemma 11.8 For every
there exist unique
and
such that
h = l ⋅ j ( r ) . h=l\cdot j(r). h = l ⋅ j ( r ) .
Proof: We know that the diagram
commutes.
That is,
z ∘ q = ψ . z\circ q=\psi. z ∘ q = ψ .
Given a splitting
H ← j R , H\xleftarrow{j}R, H j R ,
we have
z ∘ q ∘ j = ψ ∘ j = i d R . z\circ q\circ j
=
\psi\circ j
=
\mathrm{id}_R. z ∘ q ∘ j = ψ ∘ j = id R .
Since z z z is an isomorphism, it has an inverse
and therefore
q ∘ j = z − 1 . q\circ j
=
z^{-1}. q ∘ j = z − 1 .
The map z − 1 z^{-1} z − 1 is also a group isomorphism.
Since z − 1 z^{-1} z − 1 is a bijection, for every
h ∈ H h\in H h ∈ H
there exists a unique
r r r
such that
q ∘ j ( r ) = [ h ] ∈ H / L . q\circ j(r)
=
[h]
\in H/L. q ∘ j ( r ) = [ h ] ∈ H / L .
Hence there exists a unique
r ∈ R r\in R r ∈ R
such that
j ( r ) ∈ L h = [ h ] . j(r)\in Lh=[h]. j ( r ) ∈ L h = [ h ] .
Equivalently, there exists a unique
r ∈ R r\in R r ∈ R
such that
[ j ( r ) ] = [ h ] . [j(r)]=[h]. [ j ( r )] = [ h ] .
Thus there exists a unique
r ∈ R r\in R r ∈ R
such that
h = l ⋅ j ( r ) h=l\cdot j(r) h = l ⋅ j ( r )
for some
l ∈ L l\in L l ∈ L .
Of course, once h h h and j ( r ) j(r) j ( r ) are fixed, l l l is uniquely determined:
l = h ⋅ j ( r ) − 1 . l
=
h\cdot j(r)^{-1}. l = h ⋅ j ( r ) − 1 .
Therefore, for every
h ∈ H h\in H h ∈ H ,
there exist unique
l , r l,r l , r
such that
h = l ⋅ j ( r ) . h=l\cdot j(r). h = l ⋅ j ( r ) .
Now we can prove the following.
Theorem 11.9 Let
be a split short exact sequence, and let
ϕ : R → A u t ( L ) \phi:R\to\mathrm{Aut}(L) ϕ : R → Aut ( L )
be the induced action.
Then
H ≅ L ⋊ ϕ R . H\cong L\rtimes_{\phi}R. H ≅ L ⋊ ϕ R .
Proof: Consider the map
L ⋊ ϕ R → α H ( l , r ) ↦ l ⋅ j ( r ) . \begin{aligned}
L\rtimes_{\phi}R
&\xrightarrow{\alpha}
H\\
(l,r)
&\mapsto
l\cdot j(r).
\end{aligned} L ⋊ ϕ R ( l , r ) α H ↦ l ⋅ j ( r ) .
Then
( l 1 ⋅ ϕ r 1 ( l 2 ) , r 1 r 2 ) ↦ l 1 ϕ r 1 ( l 2 ) j ( r 1 ) j ( r 2 ) . (l_1\cdot\phi_{r_1}(l_2),r_1r_2)
\mapsto
l_1\phi_{r_1}(l_2)j(r_1)j(r_2). ( l 1 ⋅ ϕ r 1 ( l 2 ) , r 1 r 2 ) ↦ l 1 ϕ r 1 ( l 2 ) j ( r 1 ) j ( r 2 ) .
But by definition,
ϕ r 1 ( l 2 ) = j ( r 1 ) l 2 j ( r 1 ) − 1 . \phi_{r_1}(l_2)
=
j(r_1)l_2j(r_1)^{-1}. ϕ r 1 ( l 2 ) = j ( r 1 ) l 2 j ( r 1 ) − 1 .
Therefore,
α ( ( l 1 , r 1 ) ⋅ ( l 2 , r 2 ) ) = α ( ( l 1 ϕ r 1 ( l 2 ) , r 1 r 2 ) ) = l 1 ϕ r 1 ( l 2 ) j ( r 1 ) j ( r 2 ) = l 1 j ( r 1 ) l 2 j ( r 1 ) − 1 j ( r 1 ) j ( r 2 ) = l 1 j ( r 1 ) l 2 j ( r 2 ) = α ( ( l 1 , r 1 ) ) ⋅ α ( ( l 1 , r 2 ) ) . \begin{aligned}
\alpha((l_1,r_1)\cdot(l_2,r_2))
&=
\alpha((l_1\phi_{r_1}(l_2),r_1r_2))\\
&=
l_1\phi_{r_1}(l_2)j(r_1)j(r_2)\\
&=
l_1j(r_1)l_2j(r_1)^{-1}j(r_1)j(r_2)\\
&=
l_1j(r_1)l_2j(r_2)\\
&=
\alpha((l_1,r_1))
\cdot
\alpha((l_1,r_2)).
\end{aligned} α (( l 1 , r 1 ) ⋅ ( l 2 , r 2 )) = α (( l 1 ϕ r 1 ( l 2 ) , r 1 r 2 )) = l 1 ϕ r 1 ( l 2 ) j ( r 1 ) j ( r 2 ) = l 1 j ( r 1 ) l 2 j ( r 1 ) − 1 j ( r 1 ) j ( r 2 ) = l 1 j ( r 1 ) l 2 j ( r 2 ) = α (( l 1 , r 1 )) ⋅ α (( l 1 , r 2 )) .
Thus α \alpha α is a homomorphism.
By Lemma 11.8, for every
h ∈ H h\in H h ∈ H
there exist unique
l ∈ L l\in L l ∈ L
and
r ∈ R r\in R r ∈ R
such that
h = l ⋅ j ( r ) . h=l\cdot j(r). h = l ⋅ j ( r ) .
Therefore α \alpha α is a bijection.
This is enough to show that a split short exact sequence and a semidirect product contain the same amount of data.
we obtain
ϕ : R → A u t ( L ) \phi:R\to\mathrm{Aut}(L) ϕ : R → Aut ( L )
by conjugation.
H ← α L ⋊ R H\xleftarrow{\alpha}L\rtimes R H α L ⋊ R
is an isomorphism.
Hence we have a surjection
L ⋊ R → α H → ψ R . L\rtimes R
\xrightarrow{\alpha}
H
\xrightarrow{\psi}
R. L ⋊ R α H ψ R .
Since α \alpha α is an isomorphism,
ker ( ψ ∘ α ) = α − 1 ( ker ψ ) = α − 1 ( L ) = { ( l , 1 R ) } ⊂ L ⋊ R \begin{aligned}
\ker(\psi\circ\alpha)
&=
\alpha^{-1}(\ker\psi)\\
&=
\alpha^{-1}(L)\\
&=
\{(l,1_R)\}
\subset
L\rtimes R
\end{aligned} ker ( ψ ∘ α ) = α − 1 ( ker ψ ) = α − 1 ( L ) = {( l , 1 R )} ⊂ L ⋊ R
by Lemma 11.8.
Therefore we obtain a short exact sequence
L → L ⋊ R → ψ ∘ α R l ↦ ( l , 1 R ) \begin{aligned}
L&\to L\rtimes R
\xrightarrow{\psi\circ\alpha}
R\\
l&\mapsto(l,1_R)
\end{aligned} L l → L ⋊ R ψ ∘ α R ↦ ( l , 1 R )
with splitting
L ⋊ R ← R L\rtimes R\leftarrow R L ⋊ R ← R
( 1 L , r ) (1_{L},r) ( 1 L , r ) ↦ \mapsto ↦ r ~r r
The situation can be summarized by saying that the following diagram commutes:
That is, every sub-square that can be drawn is commutative.
Example 11.4 Recall that
S O n ( R ) ⊂ O n ( R ) SO_n(\mathbb{R})
\subset
O_n(\mathbb{R}) S O n ( R ) ⊂ O n ( R )
is a subgroup of index 2 2 2 .
By definition,
S O n ( R ) = ker ( O n ( R ) → det R × ) . SO_n(\mathbb{R})
=
\ker
\left(
O_n(\mathbb{R})
\xrightarrow{\det}
\mathbb{R}^{\times}
\right). S O n ( R ) = ker ( O n ( R ) d e t R × ) .
Every kernel is a normal subgroup, so
S O n ( R ) SO_n(\mathbb{R}) S O n ( R )
is normal.
Alternatively, one may use the fact that every subgroup of index 2 2 2 is normal.
Therefore we have a short exact sequence
Z / 2 Z \mathbb{Z}/2\mathbb{Z} Z /2 Z ≅ \cong ≅ 1 → S O n ( R ) → O n ( R ) → ± 1 → 1 1\to SO_{n}(\mathbb{R})\to O_{n}(\mathbb{R})\to{\pm 1}\to 1~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 1 → S O n ( R ) → O n ( R ) → ± 1 → 1 ⊂ \subset ⊂ R × \mathbb{R}^{\times} R ×
This sequence admits many different splittings.
For concreteness, take
For example,
O 2 ( R ) ← Z / 2 Z : j O_2(\mathbb{R})
\leftarrow
\mathbb{Z}/2\mathbb{Z}:j O 2 ( R ) ← Z /2 Z : j
( 1 0 0 1 ) \begin{pmatrix}
1&0\
0&1
\end{pmatrix} ( 1 0 0 1 ) ↦ \mapsto ↦ [ 0 ] [0] [ 0 ]
( 1 0 0 − 1 ) \begin{pmatrix}
1&0\
0&-1
\end{pmatrix} ( 1 0 0 − 1 ) ↦ \mapsto ↦ [ 1 ] [1] [ 1 ]
or
( − 1 0 0 1 ) \begin{pmatrix}
-1&0\
0&1
\end{pmatrix} ( − 1 0 0 1 ) ↦ \mapsto ↦ [ 1 ] [1] [ 1 ]
or
( 0 1 1 0 ) \begin{pmatrix}
0&1\
1&0
\end{pmatrix} ( 0 1 1 0 ) ↦ \mapsto ↦ [ 1 ] [1] [ 1 ]