TRANSITIVE GROUP ACTIONS

KEITH CONRAD

1. Introduction
Every action of a group on a set decomposes the set into disjoint orbits. The group
acts on each of the orbits and an orbit does not have sub-orbits because unequal orbits are
disjoint, so the decomposition of a set into orbits could be considered as a “factorization” of
the set into “irreducible” pieces for the group action. Our focus here is on these irreducible
parts, namely group actions with a single orbit.
Definition 1.1. A action of a group on a set is called transitive when the set is nonempty
and there is exactly one orbit.
Example 1.2. For n ≥ 1, the usual action of Sn on {1, 2, . . . , n} is transitive since there is
a permutation sending 1 to every other number. The orbit of 1 is {1, 2, . . . , n}.
Example 1.3. For n ≥ 3, the usual action of An on {1, 2, . . . , n} is transitive since the
3-cycles (12n), (13n), . . . , (1 n − 1 n), (1n2) send 1 to every other number, so the orbit of 1
is {1, 2, . . . , n}. Notice A2 does not act transitively on {1, 2}, since A2 is trivial.
Example 1.4. For n ≥ 3, the usual action of Dn on the vertices of a regular n-gon is
transitive since each vertex of the n-gon can be carried to all the other vertices by the
rotations in Dn .
In Section 2 we give some further examples (and non-examples) of transitive actions.
Section 3 gives a few general properties of transitive actions. Doubly transitive actions are
the subject of Section 4 and they are applied in Section 5 to prove simplicity of most of the
groups PSL2 (F ). Highly transitive actions are used in Section 6 to prove most alternating
groups are simple. In Section 7 we look at equivalence relations preserved by a group
action, which leads to a concept lying between transitivity and double transitivity, called
primitivity. For further reading, see [1] and [3].

2. More examples
Example 2.1. Let the group Rn act on itself by translations: for v ∈ Rn , Tv : Rn → Rn
by Tv (w) = w + v. Since v = Tv (0), every vector is in the orbit of 0, so this action is
transitive. Concretely, this just means you can move to each point in Rn from 0 by a
suitable translation.
Example 2.2. The usual action of GL2 (R) on R2 is not transitive, since 0 is in its own
2
a

the action of GL2 (R) on R − {0} is transitive.
orbit. However,
1

To see why, pick a non-zero
vector v = b . We will find an A ∈ GL2 (R) such that A 0 = v, which means every v 6= 0
is in the GL2 (R)-orbit of 10 . If a 6= 0, let A = ( ab 01 ). If b 6= 0, let A = ( ab 10 ). These

matrices are invertible in each case, and they send 10 to ab = v.

1
2 KEITH CONRAD

Example 2.3. The action of SL2 (R) on R2 − {0} is also transitive. Indeed, in the previous
example replace the choice of ( ab 01 ) with ( ab 1/a
0
) when a 6= 0 and ( ab 10 ) with ( ab −1/b ) when
1

a

a

0
b 6= 0. We’ve found a matrix with determinant 1 that sends 0 to b when b 6= 0.
Example 2.4. The action of the orthogonal group O2 (R) on R2 − {0} is not transitive.
Its orbits are the circles centered at (0, 0). If we pick a particular circle centered at (0, 0),
such as the unit circle, then the action of O2 (R) on that circle is transitive.
Now we turn to examples (and non-examples) of transitive actions using abstract groups.
Example 2.5. Let a group G act on itself by left multiplication. Since g = g · e, every
element is in the orbit of e, so there is one orbit. When G = Rn , this is exactly Example
2.1.
Example 2.6. Let G be a group with a subgroup H. The action of G by left multiplication
on the coset space G/H has one orbit, since gH = g · H: the orbit of H ∈ G/H is the whole
coset space. Example 2.5 is the special case where H is the trivial subgroup.
Example 2.7. The action of G on itself by conjugation is not transitive if |G| > 1. Indeed,
the orbits of the conjugation action are the conjugacy classes of G and {e} is its own
conjugacy class. What about the conjugation action of G on G − {e}? This is not transitive
if G is finite and |G| > 2 since finite groups with at least three elements have at least three
conjugacy classes.1 (There are infinite groups where the non-identity elements form a single
conjugacy class.)
Example 2.8. The conjugation action of G on its subgroups of a fixed size may or may not
be transitive. If the size is a maximal prime power dividing |G| then the action is transitive
(conjugacy of p-Sylow subgroups), but otherwise it need not be. For instance, there could
be a normal subgroup of some size and other subgroups of the same size.

3. Properties of transitive actions

Our first theorem about transitive actions is an equivalence between its definition and
two other descriptions.
Theorem 3.1. Let G act on a nonempty set X. The following conditions are equivalent:
(1) the action of G on X is transitive (there is one G-orbit in X),
(2) for some x ∈ X, every element of X has the form gx for some g ∈ G,
(3) for all x ∈ X, every element of X has the form gx for some g ∈ G.
Proof. (1) ⇒ (2) If the action is transitive then X is a G-orbit, say the orbit of x. All
elements of this orbit are gx for some g ∈ G.
(2) ⇒ (3) All elements in an orbit have the same orbit, so if X is the G-orbit of some
point in X then X is the G-orbit of every point in X.
(2) or (3) ⇒ (1): from either (2) or (3), all elements of X lie in the same G-orbit, so the
action of G on X is transitive. 
A convenient way of describing (3) is: for each x and y in G, y = gx for some g ∈ G.
This puts the initial element x and target element y on the same footing (since x = g −1 y).
Theorem 3.2. If a finite group G acts transitively on X then |X| | |G|.
1See Example B.2 in https://kconrad.math.uconn.edu/blurbs/grouptheory/conjclass.pdf.
 TRANSITIVE GROUP ACTIONS 3

Proof. Pick x ∈ X. Since the G-orbit of x is X, the set X is finite and the orbit-stabilizer
formula tells us |X| = [G : Stabx ], so |X| | |G|. 
Example 3.3. Let p be prime. If G is a subgroup of Sp and its natural action on {1, 2, . . . , p}
is transitive then p | |G| by Theorem 3.2, so G contains an element of order p by Cauchy’s
theorem. The only elements of order p in Sp are p-cycles, so every subgroup of Sp whose
natural action on {1, 2, . . . , p} is transitive contains a p-cycle.
Theorem 3.4. Suppose G acts on two finite sets X and Y and there is a function f : X → Y
that respects the G-actions: f (gx) = gf (x) for all g ∈ G and x ∈ X. If the action is
transitive on Y then |Y | | |X|.
Proof. The set X can be decomposed into disjoint subsets according to the f -values of the
elements. That is,
[
X= f −1 (y),
y∈Y

and the sets f −1 (y)are disjoint. We will show |f −1 (y)| = |f −1 (y 0 )| for y and y 0 in Y , and
therefore |X| equals |Y | times the common size of the inverse images.
For y and y 0 in Y , write y 0 = g0 y. Then we get maps f −1 (y) 7→ f −1 (y 0 ) and f −1 (y 0 ) 7→
f (y) by x 7→ g0 x and x0 7→ g0−1 x0 . These maps are inverses of each other, so |f −1 (y)| =
−1

|f −1 (y 0 )|. 
As an exercise, show Theorem 3.2 is a special case of Theorem 3.4.
Here is a cute application of Theorem 3.4 to counting Sylow subgroups. For a group
G, np (G) denotes the size of Sylp (G). If H ⊂ G and N C G, then np (H) ≤ np (G) and
np (G/N ) ≤ np (G). Might these inequalities really be divisibilities? Not always. There are
several copies of A4 in A5 , and n3 (A4 ) = 4 while n3 (A5 ) = 10. However, if we stick to
normal subgroups only, then we do get a divisibility relation on the Sylow counts.
Corollary 3.5. When G is a finite group and N C G, np (N ) | np (G) and np (G/N ) | np (G).
Proof. Let X = Sylp (G) and Y = Sylp (N ). The group G acts on both X and Y by
conjugation. By the Sylow theorems, the action of N on Y is transitive, so the action of G
on Y is transitive. (Also the action of G on X is transitive, but we won’t need this.) There
is a natural map f : X → Y given by f (P ) = P ∩ N . (That P ∩ N is a p-Sylow subgroup of
N is the first part of Theorem A.1.) Since g(P ∩ N )g −1 = gP g −1 ∩ gN g −1 = gP g −1 ∩ N ,
f respects the conjugation action of G on X and Y . Now use Theorem 3.4 to see |Y | | |X|,
so np (N ) | np (G).
To show np (G/N ) | np (G), use X = Sylp (G) again but now let Y = Sylp (G/N ). Once
again G acts on both X and Y by conjugation, and the action on Y (and on X) is transitive.
Let f : X → Y by f (P ) = P N/N (that this is p-Sylow subgroup of G/N is the second part
of Theorem A.1) and check f respects the action of G on the two sets. 
Our next theorem about transitive actions is fundamental. It says that Example 2.6 is
actually the most general example: every transitive action can be viewed as a left mul-
tiplication action on cosets of a subgroup, even though it may not appear that way at
first.
Theorem 3.6. A transitive action of a group G is equivalent to an action of G by left
multiplication on some coset space G/H.
4 KEITH CONRAD

Proof. Let G act transitively on X. We want to show this action is the same as the left
multiplication action of G on some coset space G/H.
Pick x0 ∈ X. Every element of X has the form gx0 for some g ∈ G. Consider the map
G → X by g 7→ gx0 . This is onto by transitivity. Let H = Stabx0 . For g and g 0 in G,
gx0 = g 0 x0 ⇐⇒ g −1 g 0 ∈ H ⇐⇒ gH = g 0 H.
This shows there is a bijection G/H → X by gH 7→ gx0 . (Concretely, since (gh)x0 = gx0
for all h ∈ H, it makes sense to associate the whole coset gH with the point gx0 .) This
bijection between G/H and X respects the G-actions on both sides. To see this, pick x ∈ X
and g ∈ G, and set y = gx. What coset in G/H corresponds to x? Writing x = g0 x0 ,
the coset corresponding to x is g0 H. Similarly, since y = g(g0 x0 ) = (gg0 )x0 , the coset
corresponding to y is (gg0 )H = gg0 H. Thus, when x ↔ g0 H, we see that gx ↔ gg0 H, so
the G-actions on X and G/H correspond to each other by the bijection between them. 
To summarize the above proof, when G acts transitively on X fix an x0 ∈ X and let
H = Stabx0 . Then G/H is in bijection with X by letting gH correspond to gx0 , and this
bijection identifies left multiplication of G on G/H with the action of G on X.
Example 3.7. Let’s look at a transitive action that does not appear to be a coset action
2

on R − {0} by
at first, and understand why it really is. We consider the action of GL2 (R)
1
matrix-vector multiplication. We saw in Example 2.2 that the orbit of 0 takes us through
the whole space.
Following the idea in the proof of Theorem 3.6, we are going to show this action of
GL2 (R) on R2 − {0} is the same as the action of GL2 (R) on a certain left coset space
GL2 (R)/H.
Define H = Stab(1) = {A ∈ GL2 (R) : A 10 = 10 }. Carrying out the matrix-vector



0
multiplication explicitly, the stabilizing condition A 10 = 10 means the first column of A

is 10 , so


  
1 x
H= : y 6= 0 .
0 y
∼ 1x y x −1
(Note H = Aff(R) by ( 0 y ) 7→ ( ) .)
0 1
Theorem 3.6 tells us the coset space GL2 (R)/H looks like R2 − {0}. Let’s try to under-
stand how this works. For a typical matrix ( ac db ) in GL2 (R), which matrices belong to the
left coset ( ac db )H? For all ( 10 xy ) in H (x and y vary, except y 6= 0),
    
a b 1 x a ax + by
= .
c d 0 y c cx + dy
The second column in the matrix on the right is the vector ( ac db ) xy . As xy runs through

all vectors with y 6= 0, ax+by a

cx+dy runs through all vectors except the scalar multiples of c ,
so symbolically    
a b a ∗
H= ,
c d c ∗
where the ∗ means “put anything here, keeping the matrix invertible.” (A
matrix ( ac ∗∗ )
is invertible if and only if the second column is not a scalar multiple of ac .) Thus, the
essential data in a left coset of H is just the common first column of all the matrices of the
coset. We can see now how cosets in GL2 (R)/H naturally correspond to non-zero vectors
of R2 : associate to a left coset the common first column of the matrices in that coset.
 TRANSITIVE GROUP ACTIONS 5

Now we check that the GL2 (R)-action on non-zero vectors in R2 matches the left
multiplication action of GL2 (R) on GL2 (R)/H under our “first column” correspondence
GL2 (R)/H → R2 − {0}.
Pick a left coset {( αβ ∗∗ ) ∈ GL2 (R)} of H, where α and β are fixed. Let v = αβ . For


α∗
all ( ac db ) in GL2 (R), we have ( ac db )v = aα+bβ


cα+dβ . On the coset side, write a typical ( β ∗ ) in
αγ
explicit form, say as ( β δ ). Then
    
a b α γ aα + bβ aγ + bδ
= .
c d β δ cα + dβ cγ + dδ
Passing back to the cosets by ignoring the second columns, we have
    
a b α ∗ aα + bβ ∗
= .
c d β ∗ cα + dβ ∗
This matches the way ( ac db ) multiplies the vector αβ , so the actions of GL2 (R) on R2 − {0}

by matrix-vector multiplication and on GL2 (R)/H by left multiplication agree.

It is left to the reader to examine how R2 − {0} can be viewed as a left coset space of
SL2 (R) using Example 2.3.
Example 3.8. Consider the usual action of Sn on {1, 2, . . . , n}, which is transitive. We
will interpret this as a left coset action in the spirit of Theorem 3.6. The stabilizer of the
last number n is naturally identified with Sn−1 ⊂ Sn . All elements in a left coset σSn−1
in Sn have a common value at n since Sn−1 fixes n. That is, for all π ∈ Sn−1 we have
(σπ)(n) = σ(n). Conversely, if σ1 (n) = σ2 (n) then σ1−1 σ2 (n) = n, so σ1−1 σ2 ∈ Sn−1 . Thus
σ1 Sn−1 = σ2 Sn−1 . Associating each left coset of Sn−1 in Sn to the common value of its
members at n (a number in {1, 2, . . . , n}) converts the multiplication action of Sn on the
left coset space Sn /Sn−1 into the usual action of Sn on {1, 2, . . . , n}.
Theorem 3.9. Let G act transitively on X. If N C G then all orbits of N on X have the
same cardinality.
Proof. Pick two points x and y in X. By transitivity of G, we can write y = gx for some
g ∈ G. A bijection between the orbits N x and N y = N gx = gN x is t 7→ gt. (This bijection
depends on the choice of g such that y = gx.) 
Example 3.10. Let G = D4 act on the four vertices of a square in the natural way and let
N = Z(G) = {1, r2 }. The N -orbits on the four vertices are two orbits of length 2, namely
pairs of opposite vertices. On the other hand, the subgroup {1, s} is not normal, where s is
reflection across a diagonal of the square, and the orbits of the vertices under this subgroup
do not all have equal size: there are two orbits of length 1 and one orbit of length 2.
Example 3.11. Let p be prime. If G is a subgroup of Sp and its natural action on
{1, 2, . . . , p} is transitive then the action of a nontrivial normal subgroup N CG on {1, 2, . . . , p}
is also transitive! Indeed, the N -orbits on {1, 2, . . . , p} all have the same size by Theorem
3.9, say m, so m > 1 since some N -orbit is not a point (because N is nontrivial in Sp ). Then
counting {1, 2, . . . , p} by counting the orders of different N -orbits, we get p = m·|{N -orbits}|
with m > 1, so m = p, which means there is only one N -orbit.
By Example 3.3, N contains a p-cycle.
Theorem 3.12 (Frattini). If G acts on X and H is a subgroup of G, then the following
are equivalent:
6 KEITH CONRAD

(1) H acts transitively on X,

(2) G acts transitively on X and G = H Stabx for some x ∈ X.
When this occurs, G = H Stabx for every x ∈ X.
Proof. Suppose H acts transitively on X. Then obviously G acts transitively on X. Fix
an x ∈ X. For g ∈ G, gx = hx for some h ∈ H, so h−1 g ∈ Stabx . Thus g ∈ H Stabx , so
G = H Stabx . Conversely, if G acts transitively on X and G = H Stabx for some x ∈ X
then X = Gx = H Stabx x = Hx, so H acts transitively on X. 
Example 3.13. Let G be a finite group and N C G. Since G acts by conjugation on N , it
acts on Sylp (N ) (for each prime p). The conjugation action of N on Sylp (N ) is transitive
by the Sylow theorems, so for each P ∈ Sylp (N ) we have G = N StabP = N NG (P ).
Corollary 3.14. Let G act on X. If H is a subgroup of G such that H acts transitively
on X and Stabx ⊂ H for some x ∈ X then H = G.
Proof. In Theorem 3.12, use for x the point such that Stabx ⊂ H. Then Theorem 3.12
becomes G = H Stabx = H. 
For a group G, consider the natural action of the group Aut(G) on G: an automorphism
ϕ sends g ∈ G to ϕ(g) ∈ G. The identity element in G is a fixed point: ϕ(e) = e for all
ϕ ∈ Aut(G). Could the action of Aut(G) on G − {e} be transitive? That is, could all pairs
of non-identity elements of a group be linked by an automorphism of the group? One case
where this happens is G = (Z/(p))n : Aut(G) equals GLn (Z/(p)) and the transitivity of
GLn (Z/(p)) on (Z/(p))n follows from the fact that each nonzero vector can be extended to
a basis. We now show this is the only situation where Aut(G) acts transitively on G − {e}
and G is finite.
Theorem 3.15. Let G be finite with |G| > 1. If Aut(G) acts transitively on G − {e} then
G∼
= (Z/(p))n for some prime p.
Proof. Elements that are linked by an automorphism have the same order, so the hypothesis
in the theorem implies all non-identity elements of G have the same order. Let p be a prime
factor of |G|. Cauchy’s theorem gives us an element with order p, so all non-identity elements
have order p. Thus |G| is a power of p. Since G is a non-trivial p-group, it has a non-trivial
center. Elements that are linked by an automorphism are both in or both not in the center,
so by the hypothesis of the theorem every non-identity element of G is in the center. Thus
G is abelian. Since each non-zero element has order p, we can view G as a vector space over
Z/(p), necessarily finite-dimensional since G is finite. Picking a basis shows G ∼ = (Z/(p))n
for some n. 

4. Doubly transitive group actions

Some group actions don’t just take every element to every other element, but can do so
in pairs. Of course, we have to assume our set has at least two elements.
Definition 4.1. An action of a group G on a set X, with |X| ≥ 2, is called doubly transitive
when, for all ordered pairs of distinct elements (x, x0 ) and (y, y 0 ) in X, there is a g ∈ G such
that y = gx and y 0 = gx0 .
The distinctness of elements means x 6= x0 and y 6= y 0 . We say g takes the pair (x, x0 ) to
the pair (y, y 0 ).
 TRANSITIVE GROUP ACTIONS 7

Example 4.2. When |X| = 2, every non-trivial action of G on X is doubly transitive.

Writing X = {x1 , x2 }, the only ordered pairs of distinct elements are (x1 , x2 ) and (x2 , x1 ).
The identity in G sends each pair to itself, and an element of G that acts non-trivially on
X must send the pair (x1 , x2 ) to (x2 , x1 ) and vice versa.
Example 4.3. Let F be a field and let Aff(F ) act on F by ( a0 1b )·x = ax+b. (Such an action
on F by matrices is exactly the effect of linear polynomials on F under composition.) This
is doubly transitive: for ordered pairs (x, x0 ) and (y, y 0 ) of distinct elements in F , finding
a ∈ F × and b ∈ F such
that ax + b = y and ax0 + b = y 0 amounts to solving for a and b
a y
in the equation ( xx0 11 ) b = y0 , which can be done since the matrix is invertible. We have

a 6= 0 since y 6= y 0 .
Remark 4.4. The possibility x = y (or y 0 ) or x0 = y 0 (or y) in Definition 4.1 is allowed.
For instance, if |X| ≥ 3 and x, y, and z are different elements of X then (x, y) and (x, z) are
ordered pairs of distinct elements in X, so a doubly transitive action admits a g ∈ G such
that gx = x and gy = z. In particular, every doubly transitive action is transitive (this is
obvious when |X| = 2).
Example 4.5. For n ≥ 2, Sn acts doubly transitively on {1, 2, . . . , n}. When n = 2 this
follows from Example 4.2. When n ≥ 3 it is easy to see that the action of Sn can take the
ordered pair (1, 2) to every other ordered pair of distinct numbers from 1 to n.
Example 4.6. For n ≥ 4, An acts doubly transitively on {1, 2, . . . , n}. Indeed, if two
ordered pairs of distinct numbers have no elements in common then we might as well write
the ordered pairs as (1, 2) and (3, 4). Then the even permutation (13)(24) takes the first
pair to the second: 1 goes to 3 and 2 goes to 4. (Don’t confuse the notation for ordered
pairs with the notation for transpositions!) If the ordered pairs have one element in common
then they might as well be (1, 2) and (1, 3). Then the 3-cycle (234) sends the first pair to
the second. However, A3 does not act doubly transitively on {1, 2, 3}: there is no σ ∈ A3
taking the pair (1, 2) to the pair (2, 1). Similarly, A2 does not act doubly transitively on
{1, 2} since A2 is trivial.
Example 4.7. The action of Dn on the n vertices of a regular n-gon is not doubly transitive
for n ≥ 4 (it is for n = 3). For instance, an nonidentity element of Dn that fixes a vertex v
is not a rotation and thus has to be the reflection across the line through v and the center
of the n-gon. That reflection acts on the ertices other than v with orbits of size 2, which is
less than n − 1 when n ≥ 4.
Example 4.8. Although GL2 (R) acts transitively on R2 − {0} (Example 2.2), it does not
act doubly transitively. The reason has to do with linear dependence. For A ∈ GL2 (R)
and v ∈ R2 − {0}, A takes the pair (v, −v) to the pair (Av, −Av), which are negatives of
each other. In particular, given linearly independent vectors v1 and v2 , it is impossible for
a matrix in GL2 (R) to take a pair (v, −v) to the pair (v1 , v2 ).
Theorem 3.9 tells us the orbits of a normal subgroup of a group acting transitively share
the same cardinality. We can say more about orbits of a normal subgroup when the action
of the original group is doubly transitive.
Theorem 4.9. Suppose G acts doubly transitively on a set X. Each normal subgroup N CG
acts on X either trivially or transitively.
8 KEITH CONRAD

Proof. Suppose N does not act trivially: nx 6= x for some x ∈ X and n 6= 1 in N . Pick
y and y 0 in X with y 6= y 0 . There is g ∈ G such that y = gx and y 0 = g(nx). Then
y 0 = (gng −1 )(gx) = (gng −1 )(y) and gng −1 ∈ N , so N acts transitively on X. 
Example 4.10. The action of A4 on {1, 2, 3, 4} is doubly transitive and the normal sub-
group {(1), (12)(34), (13)(24), (14)(23)} C A4 acts transitively on {1, 2, 3, 4}.
Example 4.11. Let Aff(F ) act on F by ( a0 1b )x = ax + b. This is doubly transitive and the
normal subgroup N = {( 10 1b ) : b ∈ F } acts transitively (by translations) on F .
Example 4.12. We noted in Example 4.7 that the action of D4 on the 4 vertices of a square
is not doubly transitive. Consistent with Theorem 4.9, recall from Example 3.10 that the
normal subgroup {1, r2 } of D4 acts on the vertices neither trivially nor transitively.
In Remark 4.4, we observed that a doubly transitive action on a set X with |X| ≥ 3
has to be able to fix an arbitrary element x while sending each element 6= x to every other
element 6= x. This reflects a transitivity of the action of the stabilizer subgroup of x on the
set X − {x}. (For each g ∈ Stabx and y ∈ X − {x}, y 6= x =⇒ gy 6= gx = x, so Stabx
acts on X − {x}.) In fact, this transitivity of Stabx on X − {x} is equivalent to double
transitivity of G on X:
Theorem 4.13. Let G act on X with |X| ≥ 3. The action is doubly transitive if and only
if, for each x ∈ X, the group Stabx acts transitively on X − {x}.
The if direction is false if |X| = 2 and G acts trivially on X: the action is not doubly
transitive but Stabx is transitive on the one element set X − {x}.
Proof. If G acts doubly transitively on X and x ∈ X then Stabx acts transitively on X −{x}
by Remark 4.4.
To prove the converse, assume for each x ∈ X that the action of Stabx on X − {x} is
transitive. We consider two ordered pairs (x1 , x2 ) and (y1 , y2 ) in X × X, with x1 6= x2 and
y1 6= y2 . Our goal is to find an element of G taking the first pair to the second.
Usually we can do this in two steps. Use elements of Stabx1 and Staby2 with the successive
effects
(x1 , x2 ) 7→ (x1 , y2 ) 7→ (y1 , y2 ).
The only time this recipe doesn’t work is when x1 = y2 (why?).
If x1 = y2 , choose some z 6= x1 , y1 in X. (There is such a z since |X| ≥ 3.) Now use
elements of Stabx1 , Stabz , and Staby1 to obtain
(x1 , x2 ) 7→ (x1 , z) 7→ (y1 , z) 7→ (y1 , y2 ). 
Example 4.14. We can use Theorem 4.13 to give alternate explanations of Examples 4.5
and 4.6. Taking G = Sn for n ≥ 2, the stabilizer of a point in {1, 2, . . . , n} acts on the
complement of that point just like Sn−1 in its natural action. Since we already know the
natural action of Sn−1 is transitive, Theorem 4.13 tells us Sn acts doubly transitively for
n ≥ 3 (and it is obvious for n = 2). A similar argument shows An acts doubly transitively
for n ≥ 4. (We need n − 1 ≥ 3 in order for An−1 to act transitively on {1, 2, . . . , n − 1}, by
Example 1.3.) Note A3 acts transitively but not doubly transitively on {1, 2, 3}.
Corollary 4.15. If a finite group acts doubly transitively on a set then the group has even
size.
 TRANSITIVE GROUP ACTIONS 9

Proof. Let G act doubly transitively on X, with |X| = n ≥ 2. Pick x1 ∈ X and x2 ∈

X − {x1 }. Set H = Stabx1 , so [G : H] = n (Theorem 3.2). Since n | |G|, we may assume
n ≥ 3. Then H acts transitively on X − {x1 } by Theorem 4.13. Set K = H ∩ Stabx2 , so
[H : K] = |X − {x1 }| = n − 1. Thus
|G| = [G : H][H : K]|K| = n(n − 1)|K|.
Either n or n − 1 is even, so 2 | |G|. 

Note it is the group in Corollary 4.15 that has to have even size, not the set. For instance,
Aff(F ) acts doubly transitively on F for every field F and for finite F | Aff(F )| = |F |(|F |−1)
has even size but F can have odd size.
Theorem 4.13 leads to two other characterizations of double transitivity.
Corollary 4.16. Let G act on X with |X| ≥ 2. Fix x0 ∈ X. The action is doubly transitive
if and only if it is transitive and Stabx0 acts transitively on X − {x0 }.
Proof. The result is clear when |X| = 2 (Example 4.2), so take |X| ≥ 3.
The “only if” direction follows from Theorem 4.13. Conversely, assume the “if” hypoth-
esis holds. We will show for each y ∈ X that Staby acts transitively on X − {y}. Then G
acts doubly transitively by Theorem 4.13.
Write y = gx0 . Then Staby = g Stabx0 g −1 . For z1 , z2 6= y we have g −1 z1 and g −1 z2
not equal to g −1 y = x0 . Then by hypothesis, some h ∈ Stabx0 satisfies hg −1 z1 = g −1 z2 ,
so ghg −1 z1 = z2 . Since ghg −1 ∈ Stabgx0 = Staby , the group Staby acts transitively on
X − {y}. 
Corollary 4.17. Let G act on X with |X| ≥ 2 and let H be the stabilizer subgroup of a
point in X. Then the action of G on X is doubly transitive if and only if it is transitive
and
G = H ∪ HgH
for some g 6∈ H, in which case this is true for every g 6∈ H.
Proof. If |X| = 2 then G acts doubly transitively if and only if G acts transitively. When the
action is transitive H has index 2. Subgroups of index 2 are normal, so the decomposition
in the theorem holds because HgH = gH; we get the decomposition of G into two (left)
H-cosets.
Now let |X| ≥ 3. Let x be a point having H as its stabilizer. Pick g 6∈ H, so gx 6= x. If
the action is doubly transitive then it is transitive and by Theorem 4.13 X − {x} = Hgx.
For all g 0 ∈ G, if g 0 6∈ H then g 0 x = hgx for some h ∈ H, so g 0 = hge h ∈ H. Thus
h for some e
G = H ∪ HgH. This union is disjoint since H fixes x and no element of HgH fixes x.
Conversely, if G acts transitively and we have a decomposition G = H ∪ HgH for some
g 6∈ H then the union is disjoint and H sends gx to all of X − {x} (because X = Gx).
Therefore G acts doubly transitively by Corollary 4.16. 
Remark 4.18. There is another way (besides Theorem 4.13 and Corollaries 4.16 and 4.17)
to characterize double transitivity. When G acts on X it also acts on X × X in a natural
way, by g · (x, y) = (gx, gy). This is easily checked to be a group action. Since gx = gy
if and only if x = y, G acts separately on the diagonal ∆ = {(x, x) : x ∈ X} and on its
complement X × X − ∆ = {(x, y) : x 6= y}. The action of G on X is doubly transitive if
and only if G acts transitively on X × X − ∆.
10 KEITH CONRAD

Note Theorem 4.13, Corollaries 4.16 and 4.17, and Remark 4.18 give characterizations of
doubly transitive actions. The following theorem gives only a necessary (not a sufficient)
condition for double transitivity.
Theorem 4.19. If G acts doubly transitively on X then the stabilizer subgroup of each
point in X is a maximal subgroup of G.
A maximal subgroup is a proper subgroup contained in no other proper subgroup.
Proof. Pick x0 ∈ X and let H = Stabx0 . Assume K is a subgroup strictly containing H. By
Corollary 4.17, G = H ∪ HgH for each g 6∈ H. Pick g ∈ K − H. Then G = H ∪ HgH ⊂ K,
so K = G. 
Example 4.20. According to Example 4.5 and Theorem 4.19, Sn−1 is a maximal subgroup
of Sn (with index n) for n ≥ 2.
The converse of Theorem 4.19 is false: by Corollary 4.15 a finite group of odd size has
no doubly transitive actions, but it does have actions where all the stabilizer subgroups are
maximal subgroups (consider left multiplication of G on G/H with H a maximal subgroup
of G).
Let F be a field. The action of GL2 (F ) on F 2 − { 00 } is not doubly transitive since


linearly dependent vectors can’t be sent to linearly independent vectors by a matrix. Since
linearly dependent vectors in F 2 lie along the same line through the origin, consider the
action of GL2 (F ) on the one-dimensional subspaces of F 2 : A ∈ GL2 (F ) sends the line
L = F v to the line A(L) = F (Av). (Equivalently, we are letting GL2 (F ) act on P1 (F ),
the projective line over F .) Not only does this action of GL2 (F ) turn out to be doubly
transitive, but the restriction of this action to SL2 (F ) is doubly transitive.
Theorem 4.21. For every field F , the action of SL2 (F ) on the one-dimensional subspaces
of F 2 is doubly transitive. In particular, the action of GL2 (F ) is also doubly transitive.
Proof. The action of SL2 (F ) on F 2 − { 00 } is transitive (Example 2.3 for F = R), so its

action on the one-dimensional subspaces of F 2 is also transitive. Thus, to show the action of
SL2 (F ) on the one-dimensional subspaces of F 2 is doubly transitive we will
follow Corollary
4.16 and show the stabilizer subgroup of the one-dimensional subspace F 10 acts transitively
on the other one-dimensional subspaces.
The stabilizer subgroup of F 10 in SL2 (F ) is


    
1 1
StabF (1) = A ∈ SL2 (F ) : A ∈F
0 0 0
  
a b
= ∈ SL2 (F )
0 d
  
a b ×
(4.1) = : a ∈ F ,b ∈ F .
0 1/a
Pick one-dimensional subspaces F v and F w with neither equal to F 10 . That means the


lines F v and F w each contain a vector with a non-zero second coordinate, so each
of these
a
vector with second coordinate 1, say F v = F 1 and F w = F y1 . Since
x


lines contains
x y
( 10 y−x 1 y−x


1 ) 1 = 1 , ( 0 1 ) sends F v to F w. Thus SL2 (F ) acts doubly transitively. 
One can formulate the idea of a triply-transitive action, and more generally a k-fold
transitive action for an integer k ≥ 1: for all ordered k-tuples (x1 , . . . , xk ) and (y1 , . . . , yk )
 TRANSITIVE GROUP ACTIONS 11

of distinct elements in the set, some element of the group sends xi to yi for all i. For
instance, Sn is n-fold transitive on {1, 2, . . . , n} for all n and An is (n − 2)-fold transitive
on {1, 2, . . . , n} for n ≥ 3. An action that is k-fold transitive is `-fold transitive for ` < k,
so a triply transitive action is doubly transitive and transitive.
Example 4.22. Let F be a field. The action of GL2 (F ) on the one-dimensional subspaces
of F 2 is triply transitive.
It is enough to show the particular one-dimensional subspaces F 10 , F 01 , and F 11

can be sent by some matrix in GL2 (F ) to every other triple of distinct one-dimensional
subspaces F u, F v, F w (in this order). We need to find an A ∈ GL2 (F ) such that A 10 ∈ F u,

A 01 ∈ F v, and A 11 ∈ F w. Since F u 6= F v, u and v are linearly independent and thus are

a basis of F 2 . Write w = αu + βv with α, β ∈ F . We have α, β 6= 0 since F w is not F u or

F v. Let A be the 2 × 2
matrix (αu
βv), which is invertible since the columns are linearly
1 0 1


independent. Then A 0 = αu, A 1 = βv, and A 1 = w.
Example 4.23. Let F be a field. The action of SL2 (F ) on the one-dimensional subspaces
2
of F 2 is triply transitive if and only if F × = F × .
Generalizing Corollary 4.16, a group G acting transitively on a set X acts triply transi-
tively if and only if Stabx ∩ Staby acts transitively on X − {x, y} for some distinct pair x
and y in X. Taking G = SL2 (F ) acting on the one-dimensional subspaces of F 2 ,
  
a 0 ×
StabF (1) ∩ StabF (0) = :a∈F .
0 1 0 1/a
For all t 6= 0 or 1 in F , there is an a such that ( a0 1/a
0 a
) 11 ∈ F 1t if and only if 1/a ∈ F 1t ,

i.e., t = a2 .
The condition F × = (F × )2 holds when F is algebraically closed (e.g., F = C). When F
is finite, F × = (F × )2 if and only if F has characteristic 2.
Example 4.24. Let F be a field. The action of Aff(F ) on F is doubly transitive (Example
4.3) but it is not triply transitive when |F | ≥ 3: for t 6= 0, 1 in F there is no ( a0 1b ) in Aff(F )
that sends 0 to 0, 1 to 1, and t to t + 1 since the first two conditions force the matrix
to be ( 10 01 ), which sends t to t. More generally, for x 6= y in F the two-point stabilizer
Stabx,y := Stabx ∩ Staby in Aff(F ) is trivial. This action is also faithful; for instance, a
matrix in Aff(F ) is determined by where it sends 0 and 1.
Conversely, every faithful doubly transitive action on a finite set with trivial two-point
stabilizers is equivalent to the natural action of the affine group of a finite near field [3,
§7.6]. The set of all finite faithful doubly transitive actions (with no assumptions on the
two-point stabilizers) is described in [3, §7.7].
Remark 4.25. Examples of faithful k-fold transitive actions with k ≥ 4 other than the
natural actions of Sn (n ≥ 4) and An (n ≥ 6) are rare: the only others are related to four of
the five Mathieu groups (we do not define them here, but will meet them again in Appendix
6). Two are 4-fold transitive and two are 5-fold transitive.
Here is the doubly transitive refinement of Theorem 3.15.
Theorem 4.26. Let G be finite with |G| > 1. If Aut(G) acts doubly transitively on G − {e}
then G ∼= (Z/(2))n or G ∼= Z/(3).
Proof. By Theorem 3.15, G ∼ = (Z/(p))n . If n ≥ 2, pick linearly independent v and w in G. If
p > 2 then v, −v, and w are distinct elements of G and there is no A ∈ Aut(G) ∼= GLn (Z/(p))
12 KEITH CONRAD

such that Av = v while A(−v) = w. So if Aut(G) acts doubly transitively on G then p = 2

or n = 1. If n = 1 then an automorphism of G ∼ = Z/(p) that fixes a non-zero element fixes
all elements, so by double transitivity there can be at most 2 non-zero elements. Thus if
n = 1 we have p − 1 ≤ 2, so p ≤ 3. 
The converse is true. To see that GLn (Z/(2)) acts doubly transitively on (Z/(2))n − {0},
pick x 6= x0 and y 6= y 0 in (Z/(2))n − {0}. Then, letting F = Z/(2),
G∼ = F n = F x + F x0 + U = F y + F y 0 + V
for suitable subspaces U and V . Let A : F n → F n be a linear map such that Ax = y,
Ax0 = y 0 , and A identifies a basis of U with a basis of V . Then A ∈ GLn (Z/(2)), so the
action of Aut(G) on G − {e} is doubly transitive. The group Aut(Z/(3)) = (Z/(3))× acts
doubly transitively on Z/(3) − {0} since it is a non-trivial action on a set of size 2.
Corollary 4.27. Let G be finite with |G| > 1. If Aut(G) acts triply transitively on G − {e}
then G ∼
= (Z/(2))2 .
Proof. Since |G − {e}| ≥ 3, by Theorem 4.26 we have G = ∼ (Z/(2))n for some n ≥ 2.
Let v and w be linearly independent in (Z/(2))n . Then by triple transitivity Stabv ∩ Stabw
acts transitively on the remaining non-zero vectors. Since a linear map fixing v and w also
fixes v + w, the whole group is just {0, v, w, v + w}. Therefore n = 2. 

5. Simplicity of PSL2 (F )
The goal of this section is to use the doubly transitive action of SL2 (F ) on the one-
dimensional subspaces of F 2 to prove the simplicity of most groups PSL2 (F ) from the
following criterion of Iwasawa.
Theorem 5.1 (Iwasawa). Let G be a group that acts doubly transitively on a set X. Suppose
for some x ∈ X that Stabx has an abelian normal subgroup whose conjugate subgroups
generate G. If [G, G] = G then G/K is a simple group, where K is the kernel of the action
of G on X.
The kernel of an action is the kernel of the homomorphism G → Sym(X); it’s those g
that act like the identity permutation on X. An action is faithful if and only if it has a
trivial kernel.
Proof. To show G/K is simple we will show the only normal subgroups of G lying between
K and G are K and G. Let K ⊂ N ⊂ G with N C G. Let H = Stabx , so H is a maximal
subgroup of G (Theorem 4.19). Since N H is a subgroup of G containing H, either N H = H
or N H = G. By Theorem 4.9, N acts trivially or transitively on X, so N ⊂ K or N H = G
(check!). If N ⊂ K then N = K by hypothesis.
Now suppose N H = G. Let U be the abelian normal subgroup of H in the hypothesis:
its conjugate subgroups generate G. Since U C H, N U C N H = G. Then for g ∈ G,
gU g −1 ⊂ g(N U )g −1 = N U , which shows N U contains all the conjugate subgroups of U .
By hypothesis it follows that N U = G.
Thus G/N = (N U )/N ∼ = U/(N ∩ U ) is abelian, so [G, G] ⊂ N . Since G = [G, G] by
hypothesis, we have N = G. 
Example 5.2. We can use Theorem 5.1 to show A5 is a simple group. Its natural action
on {1, 2, 3, 4, 5} is doubly transitive and faithful. Let x = 5, so Stabx ∼
= A4 , which has the
 TRANSITIVE GROUP ACTIONS 13

abelian normal subgroup

{(1), (12)(34), (13)(24), (14)(23)}.
The A5 -conjugates of this subgroup generate A5 since the (2,2)-cycles in A5 are all conjugate
and generate A5 . It remains to show [A5 , A5 ] = A5 . The commutator subgroup [A5 , A5 ]
contains every (2,2)-cycle:
(abc)(abd)(abc)−1 (abd)−1 = (ab)(cd).
Therefore [A5 , A5 ] = A5 , so A5 is simple.
We will apply Theorem 5.1 to G = SL2 (F ) acting doubly transitively on the set of one-
dimensional subspaces of F 2 . What is the kernel K of this action? That is, which matrices
A ∈ SL2 (F ) carry each one-dimensional subspace of F 2 back to itself? If ( ac db ) preserves
the lines F 10 and F 01 then c = 0 and b = 0, so ( ac db ) = ( a0 d0 ). The determinant is 1, so

d = 1/a. If ( a0 1/a0
) preserves the line F 11 then a = 1/a, so a = ±1. This means the matrix

( ac db ) equals ±( 10 01 ), which is the center of SL2 (F ). Thus G/K = PSL2 (F ).

Let x = F 10 . Its stabilizer subgroup in SL2 (F ) is given in (4.1). This subgroup has an


abelian normal subgroup
    
1 ∗ 1 λ
U= = :λ∈F .
0 1 0 1
Note U ∼ = F by ( 1 λ ) 7→ λ. For example, ( 1 λ )−1 = ( 1 −λ ).
0 1 0 1 0 1
Theorem 5.3. The subgroup U and its conjugates generate SL2 (F ). More precisely, every
element of SL2 (F ) is the product of at most 3 elements from U and its conjugates.
We will arrive at a proof of Theorem 5.3 after a definition and a lemma.
Definition 5.4. Each matrix in SL2 (F ) that is conjugate to a matrix in U is called a
transvection.
If F = R, then U consists of horizontal shears: each vector is moved to another vector
on the same horizontal line and the x-axis is fixed pointwise. Intuitively, a transvection is
a shear transformation in some (not necessarily horizontal) direction.
Example 5.5. Since ( 01 −1 1λ 0 −1 −1 = ( 1 0 ), the matrices ( 1 0 ) are all transvections.
0 )( 0 1 )( 1 0 ) −λ 1 ∗1
A general transvection looks like
−1 
α2 λ
   
α β 1 λ α β 1 − αγλ
= ,
γ δ 0 1 γ δ −γ 2 λ 1 + αγλ
where αδ − βγ = 1. Notice there is no dependence on β or δ on the right side.
Lemma 5.6. For each
nonzero vector v in F 2 , there is a transvection or a product of two
transvections taking 10 to v.
Proof. The basic idea is this: if v is not on the line F 10 then a shear in some direction will

take 10 to v. If v is on the line F 10 , then we can move 10 off that line by some shear and




then move it back onto that line to land on v by another shear. This is where the one or
two transvections

in the lemma come from.
Write v = xy . We look for ( αγ βδ ) in SL2 (F ) and λ ∈ F such that
α2 λ
      
1 − αγλ 1 1 − αγλ ? x
= = .
−γ 2 λ 1 + αγλ 0 −γ 2 λ y
14 KEITH CONRAD

We want 1 − αγλ = x, −γ 2 λ = y, and αδ − βγ = 1. Use λ = −y, γ = 1 and α = (x − 1)/y;

we need y 6= 0. In this case, we can get αδ − βγ = 1 with β = −1 and δ = 0. The
−1 1 −y (x−1)/y −1 −1 x −(x−1)2 /y 1 x
transvection ( (x−1)/y



1 0
)( 0 1 )( 1 0
) = ( y 2−x
) sends 0 to y = v.
What if y = 0? Then x 6= 0. The transvection t1 = ( 11 01 ) sends v = 0 to xx , whose
x



second coordinate is
non-zero. Therefore,
byxthe

previous paragraph there is a transvection
t2 sending 10 to xx , so t−1 1


1 t 2 sends 0 to 0 = v. 

Now we prove Theorem 5.3.

Proof. Pick M ∈ SL2 (F ) and set v = M 10 . By Lemma 5.6 there is a product h of 1 or

2 transvections such that h 10 = v. Then (h−1 M ) 10 = 10 , so h−1 M = ( 10 ∗∗ ). Since the

determinant of h−1 M is 1, h−1 M = ( 10 µ1 ) for some µ. This last matrix is a transvection, so

M = h( 10 µ1 ) is a product of at most 3 transvections. 

Remark 5.7. Here is a second proof of Theorem 5.3. Given a matrix ( ac db ) in SL2 (F ) we
can write it as a product of three matrices of type ( 10 ∗1 ) or ( 1∗ 01 ) when b or c is non-zero. If
b 6= 0 then
     
a b 1 0 1 b 1 0
= .
c d (d − 1)/b 1 0 1 (a − 1)/b 1
If c 6= 0 then
     
a b 1 (a − 1)/c 1 0 1 (d − 1)/c
= .
c d 0 1 c 1 0 1

If b = 0 and c = 0 then the matrix is ( a0 1/a

0
). We can conjugate this to a matrix with
a a−1/a
non-zero upper-right entry as ( 10 11 )( a0 1/a
0
)( 10 11 )−1 = ( 0 1/a
), provided a2 6= 1. Then the
a a−1/a
explicit calculations above express ( 0 1/a
) as a product of three matrices of type ( 10 ∗1 ) or
( 1∗ 10 ) so we can conjugate back to express ( a0 1/a 0
) as a product of three transvections. The
remaining cases are ( 10 01 ) and ( −1 0
0 −1 ). The identity matrix is ( 10 11 )( 01 −1
1 ) and its negative
1 0 −1 4 1 4 −1 4 2 −1 1 1 2 −1 −1
is ( −1 1 )( −1 3 )( 0 1 ), where ( −1 3 ) = ( 1 0 )( 0 1 )( 1 0 ) is a transvection.
Allowing four transvections, we can write all the diagonal matrices in SL2 (F ) explicitly
in terms of matrices of type ( 10 1∗ ) and ( 1∗ 10 ):
      
a 0 1 0 1 1 1 0 1 −1/a
= .
0 1/a (1 − a)/a 1 0 1 a−1 1 0 1
So far F can be an arbitrary field. Now we reach a result where we need |F | ≥ 4.
Theorem 5.8. If |F | ≥ 4 then [SL2 (F ), SL2 (F )] = SL2 (F ).
Proof. We compute an explicit commutator:
−1  −1 
1 b(a2 − 1)
   
a 0 1 b a 0 1 b
= .
0 1/a 0 1 0 1/a 0 1 0 1
Since |F | ≥ 4, there is an a 6= 0, 1, or −1 in F , so a2 6= 1. Using this value of a and
letting b run over F shows [SL2 (F ), SL2 (F )] contains U . Since the commutator subgroup
is normal, it contains every subgroup conjugate to U , so [SL2 (F ), SL2 (F )] = SL2 (F ) by
Theorem 5.3. 
 TRANSITIVE GROUP ACTIONS 15

Theorem 5.8 is false when |F | = 2 or 3: SL2 (Z/(2)) = GL2 (Z/(2)) is isomorphic to S3

and [S3 , S3 ] = A3 . In SL2 (Z/(3)) the 2-Sylow subgroup is normal with index 3, so the
quotient by it is abelian. Therefore the commutator subgroup of SL2 (Z/(3)) lies inside the
2-Sylow subgroup (in fact, the commutator subgroup is the 2-Sylow subgroup).
Corollary 5.9. If |F | ≥ 4 then the group PSL2 (F ) is simple.
Proof. The action of SL2 (F ) on the one-dimensional subspaces of F 2 satisfies the hypotheses
of Iwasawa’s theorem with K equal to the center of SL2 (F ). 
Since PSL2 (Z/(2)) ∼= S3 and PSL2 (Z/(3)) ∼ = A4 are not simple, the constraint on |F | in
Corollary 5.9 is necessary.
By similar arguments, if n ≥ 3 then PSLn (F ) is a simple group for every field F . This is
proved by studying the action of SLn (F ) on the one-dimensional subspaces of F n (i.e., on
the projective space Pn−1 (F )). The restriction |F | ≥ 4 from the n = 2 case does not arise
when n ≥ 3 since [SLn (F ), SLn (F )] = SLn (F ) for every F . Once n ≥ 3 there is enough
room to move around even when |F | < 4.
By comparison to PSL2 (F ), the groups PGL2 (F ) are never simple, although it is instruc-
tive to run through the above proof with GL2 (F ) in place of SL2 (F ) to see where things
go wrong with the application of Iwasawa’s theorem. They start off well, since the action
of GL2 (F ) on the one-dimensional subspaces of F 2 is doubly transitive (even triply transi-
tive). But we run into a problem with the GL2 -analogue of Theorem 5.8: the commutator
subgroup of GL2 (F ) is not GL2 (F ) but is SL2 (F ) for |F | ≥ 4. (All commutators in GL2 (F )
have determinant 1; now use Theorem 5.8.) The commutator subgroup of GL2 (F ) is a
proper subgroup of GL2 (F ) even when |F | < 4.

6. Simplicity of An
Exploiting the highly transitive action of An−1 on {1, 2, . . . , n − 1}, we will prove An is
simple for n ≥ 5.
A group action is called regular when the action is equivalent to the left multiplication
action of the group on itself.
Lemma 6.1. If a group G admits a faithful doubly transitive action on a set X and Stabx
is a simple group for some x ∈ X then every non-trivial proper normal subgroup of G acts
regularly on X.
Proof. Set Hx = Stabx for each x ∈ X. Then all the Hx ’s are conjugate (so isomorphic) to
each other and therefore are simple groups by hypothesis. Since G acts doubly transitively,
Hx is a maximal subgroup of G (Theorem 4.19).
Assume there is a normal subgroup N not equal to {e} or G. Since N is non-trivial
and the action of G on X is faithful, N does not act trivially on X. Therefore N acts
transitively on X (Theorem 4.9). Pick x ∈ X. Since N C G, we have N ∩ Hx C Hx . Since
Hx is a simple group, N ∩ Hx = {e} or N ∩ Hx = Hx . Let’s eliminate the second possibility.
If N ∩ Hx = Hx for some x then Hx ⊂ N , so N = Hx or N = G because Hx is a maximal
subgroup of G. As N 6= G we get N = Hx , but then N does not act transitively. This is
a contradiction, so N ∩ Hx = {e} for each x. Hence N acts transitively on X with trivial
stabilizers, so N acts regularly on X. 
Theorem 6.2. Let G be a group with a faithful triply transitive action on a finite set X
and assume Stabx is a simple group for some x ∈ X. Then G is simple or |X| is a power
of 2 or is 3. If the action is 4-fold transitive then G is simple.
16 KEITH CONRAD

Before proving Theorem 6.2 we note the primary consequence.

Corollary 6.3. For n ≥ 5, An is a simple group.
Proof. We argue by induction on n. The case n = 5 was handled in Example 5.2 using
Iwasawa’s simplicity criterion (Theorem 5.1).
Now we take n ≥ 6 and assume An−1 is simple. In Theorem 6.2 let G = An in its natural
action on X = {1, 2, . . . , n}. The action is faithful and is (n − 2)-fold transitive, so 4-fold
transitive since n ≥ 6. Thus An is a simple group. 
Theorem 6.2 is also applicable to four of the five Mathieu groups. The five Mathieu
groups are denoted M11 , M12 , M22 , M23 , and M24 . We will not define these groups, but
each Mn has a highly (i.e., at least triply) transitive faithful action on a set of size n. The
groups M11 and M23 act 4-fold transitively, M12 and M24 act 5-fold transitively, and M22
acts triply transitively. The simplicity of M11 is proved in [2]. The action of M22 has point
stabilizer isomorphic to PSL3 (F4 ), which is simple, so M22 is simple by Theorem 6.2. For
n = 12, 23, and 24, the point stabilizer of Mn is Mn−1 . None of these n’s is a power of 2 or
is 3, so Theorem 6.2 implies simplicity of M12 , M23 , and M24 from simplicity of M11 and
M22 .
Remark 6.4. The only finite simple groups admitting a 4-fold transitive action are the
Mathieu groups except M22 and the alternating groups An for n ≥ 6. Therefore Theorem
6.2 has no applications to proving groups are simple beyond the ones we have already made.
Now we prove Theorem 6.2.
Proof. Let n = |X| ≥ 3. Assume G is not simple and let N be a non-trivial proper normal
subgroup. From Lemma 6.1, N acts regularly on X, so |N | = n. We are going to show
N ∼= (Z/(2))m for some m or N ∼ = Z/(3), so n = 2m or n = 3. Then we we will get a
contradiction if G acts 4-fold transitively, so G is simple in that case.
Fix a point x0 ∈ X. Let
H = Stabx0 ⊂ G.
Then |G| = |H| · n = |H||N |. The subgroup H acts on X − {x0 }. We can also make H
act on N by conjugation. Conjugations on N fix the identity, so we will think about the
conjugation action of H on the set N − {e}. These two actions of H, on X − {x0 } in the
natural way and on N − {e} by conjugation, are equivalent to each other. Let’s see why.
First, we can set up a bijection between X − {x0 } and N − {e}. Since N acts regularly on
X, for each x ∈ X (even x = x0 ) there is a unique g ∈ N such that gx0 = x. Set ϕ(x) = g.
Then ϕ(x0 ) = e, so ϕ : X − {x0 } → N − {e} is a bijection. For h ∈ H, how are the effects
of h on x ∈ X − {x0 } and on ϕ(x) ∈ N − {e} related? Its effect on ϕ(x) is by conjugation:
h · ϕ(x) = hϕ(x)h−1 .
This element of N − {e} sends x0 to
(hϕ(x)h−1 )(x0 ) = h(ϕ(x)(h−1 (x0 ))) = h(ϕ(x)(x0 )) = h(x),
so ϕ(h(x)) = h · ϕ(x) by the definition of ϕ (check!). Thus ϕ respects the H-actions on
X − {x0 } and on N − {e}. It shows how the actions of H on these two sets are equivalent.
Since G acts on X triply transitively, H acts on X −{x0 } doubly transitively and therefore
its action on N − {e} is doubly transitive. The (conjugation) action of H on N is a
homomorphism H → Aut(N ), so Aut(N ) acts doubly transitively on N − {e}. By Theorem
4.26, N ∼= (Z/(2))m or N ∼ = Z/(3). Thus |X| = |N | = 2m or 3.
 TRANSITIVE GROUP ACTIONS 17

Assume G acts 4-fold transitively on X. Then the action of H on X − {x0 } is 3-fold

transitive, so the conjugation action of H on N − {e} is triply transitive. Therefore Aut(N )
acts triply transitively on N − {e}, so N ∼ = (Z/(2))2 by Corollary 4.27. Since H is simple
and acts non-trivially on N , the homomorphism H → Aut(N ) ∼ = GL2 (Z/(2)) ∼ = S3 is
∼ ∼
injective. Therefore H = Z/(2) or H = Z/(3), so |G| = |H| · 4 = 8 or 12. Since H is a
maximal subgroup of G, the possibility |H| = 2 (and |G| = 8) can’t occur. Thus |H| = 3
and |G| = 12. There are 5 groups of size 12, up to isomorphism. In all of them except A4 ,
the subgroup of size 3 is a normal subgroup. A normal subgroup of size 3 and an element of
order 2 generate a subgroup of size 6. Thus, since H is maximal in G, we have G ∼ = A4 . But
a calculation of the subgroups of size 3 in A4 shows the action of G on G/H is equivalent
to the natural action of A4 , which is not 4-fold transitive. We have a contradiction. 

7. G-equivalence relations and primitivity

Some useful properties of doubly transitive actions, such as Theorem 4.9, are true for a
broader class of group actions that are called primitive actions. The relation looks like this:
doubly transitive actions ⊂ primitive actions ⊂ transitive actions.
We will approach the definition of primitivity (Definition 7.12) as an outgrowth of the
consideration of equivalence relations preserved by a group action.
Let G act transitively on X. For a (nonempty) subset Y ⊂ X the subsets gY (as g runs
over G) share the same cardinality and cover X:
[
X= gY.
g∈G

For instance, if Y is a one-point set then the fact that the union is X is exactly the condition
of transitivity of the action. If |Y | > 1, the different gY ’s may or may not partially overlap.
(When we speak of “different” gY ’s we mean different subsets, not just different g’s.)
Example 7.1. Let GL2 (R) act on R2 − {0} by matrix-vector multiplication. If Y is a
one-dimensional subspace without the origin then gY is also a one-dimensional subspace
without the origin and the different gY ’s do not overlap.
Example 7.2. Let Dn for n ≥ 3 act on the vertices of a regular n-gon and label the vertices
1, 2, . . . , n in counterclockwise order (so 1 and 2 are adjacent vertices, for instance). If we
take Y = {1, 2} then rY = {2, 3}, so Y ∩ rY = {2}. The different sets gY cover the vertex
set as g runs over Dn , but there are some proper nonempty overlaps between them.
For composite n there are subsets Y of the vertex set of a regular n-gon such that the
different gY ’s for g ∈ Dn don’t partially overlap: d equally spaced vertices of a regular
n-gon forming regular d-gons where d | n and 1 < d < n . See Figure 1.
For prime n ≥ 3, the only proper subsets Y of the vertex set such that the gY ’s don’t
overlap are individual vertices: if the gY ’s don’t overlap then the size of the vertex set is
|Y | times the number of different gY ’s, so |Y | | n. Therefore |Y | = 1.
To say, back in the general setting of a transitive action, that different gY ’s do not
overlap is the same as saying they form a partition of X. A partition of a set is the
same as equivalence classes for an equivalence relation on the set, so when the gY ’s form a
partition of X they provide us with an equivalence relation on X that is preserved by G:
x ∼ x0 =⇒ gx ∼ gx0 for all g ∈ G, so in fact x ∼ x0 ⇐⇒ gx ∼ gx0 for all g ∈ G. Conversely,
each equivalence relation on X that is preserved by G will have equivalence classes that
18 KEITH CONRAD

2
3 2 3 2

3 1 4 1 4 1

5 6 5 6
4

Figure 1. Equally spaced vertices as equivalence relations for n = 4, 6.

partition X and if Y is one of the equivalence classes then the rest are gY as g varies since
G acts transitively on X. For example, vertices of regular d-gons in the vertices of a regular
n-gon for a fixed divisor d of n form a basic example of a partition of the vertices that is
preserved by the action of Dn .
Definition 7.3. When G acts on X, a G-equivalence relation on X is an equivalence relation
satisfying x ∼ x0 ⇒ gx ∼ gx0 for all g ∈ G and x, x0 ∈ X.
If |X| > 1 then there are always at least two G-equivalence relations on X: the equivalence
relation whose equivalence classes are individual points of X and the equivalence relation
having all of X as a single equivalence class. Can there be others?
Lemma 7.4. Let G act transitively on X and Y ⊂ X be a non-empty subset. The following
conditions are equivalent:
(1) for all g1 and g2 in G, the subsets g1 Y and g2 Y are either equal or are disjoint,
(2) for each g ∈ G, the subset gY either equals Y or is disjoint from Y .
Proof. The second condition is clearly a special case of the first. Since g1 Y ∩ g2 Y =
g2 (g2−1 g1 Y ∩ Y ), the first condition follows from the second. 
Theorem 7.5. Let G act transitively on X and let H be the stabilizer subgroup of a point
in X. The G-equivalence relations on X are in bijection with the intermediate subgroups
H ⊂ K ⊂ G. Moreover, in the equivalence relation corresponding to K each equivalence
class has size [K : H].
Proof. Say H = Staby , for some y ∈ X. Suppose there is a G-equivalence relation on X.
Let Y ⊂ X be the equivalence class containing y. Every equivalence class has the form gY
for some g ∈ G, so all equivalence classes have the same size. What is it?
Set
K = StabY = {g ∈ G : gY = Y }.
Since the gY ’s partition X, we can also say
(7.1) K = {g ∈ G : gY ∩ Y 6= ∅}.
(That is, as soon as gY and Y overlap they coincide because equivalence classes partition
X.) So gy ∈ Y if and only if g ∈ K. In particular, H ⊂ K: if h ∈ H then hy = y ∈ Y ,
so h ∈ K. Since every element of X has the form gy for some g ∈ G, Y = Ky. Thus
|Y | = |Ky| = [K : H].
 TRANSITIVE GROUP ACTIONS 19

Given a G-equivalence relation on X, (7.1) gives us a subgroup between H and G. Con-

versely, suppose K is a subgroup between H and G. Define Y = Ky. The sets gY (g ∈ G)
partition X. To see this, it suffices by Lemma 7.4 to check that gY ∩ Y 6= ∅ =⇒ gY = Y .
Suppose g(ky) = k 0 y, so k 0−1 gky = y. Then k 0−1 gk ∈ H, so g ∈ k 0 Hk −1 ⊂ K, so
gY = gKy = Ky = Y . We have produced a partition {gY : g ∈ G} of X where Y is
the equivalence class of y and K = {g : gy ∈ Y }.
This work provides a bijection between the subgroups between H and G and the G-
equivalence relations on X, where the equivalence relation ∼ leads to the intermediate
subgroup {g ∈ G : gy ∼ y} and the intermediate subgroup K leads to the equivalence
relation gy ∼ g 0 y ⇐⇒ gK = g 0 K. 
Letting X = G/H, G-equivalence relations on X are the same as fibers of the natural
maps G/H → G/K where H ⊂ K ⊂ G.
Example 7.6. If G = Dn acts in the usual way on the vertices of a regular n-gon, H = {1, s}
for a reflection s across a line through a vertex, and d | n then the equivalence relation of
vertices lying on a regular d-gon inside the n-gon corresponds to the subgroup K = Dd
containing H. Note [K : H] = 2d/2 = d is the size of each equivalence class.
If we apply Theorem 7.5 to the action of G on itself by left multiplication, whose stabilizer
subgroups are trivial, then it tells us that the G-equivalence relations on G that are preserved
by the left multiplication action are precisely the left coset decompositions of G by different
subgroups of G. Each left coset decomposition comes from a particular subgroup, and that
is how the subgroups of G match up with the G-equivalence relations on G.
Corollary 7.7. When a finite cyclic group G acts transitively on a set X, for each divisor
d of |X| there is one G-equivalence relation on X whose equivalence classes have size d.
Proof. Let H be the stabilizer of a point of X, so |X| = [G : H] and d | [G : H]. Since G is
cyclic, there is exactly one subgroup K between H and G such that [K : H] = d. 
Remark 7.8. In the literature, the equivalence classes in a G-equivalence relation on X
are called blocks and the totality of equivalence classes for a given equivalence relation is
called a block system. If we view X as G/H, a block system is simply the left cosets of a
subgroup K lying between H and G; a block is one of those cosets.
As a nice application of G-equivalence relations to group theory, we will prove an exten-
sion of part of Sylow’s third theorem due to Weisner [4]. Our argument is based on some
notes of H. Lenstra (which in turn are based on an argument of H. Wielandt.)
Theorem 7.9 (Weisner). Let G be a finite group and p be a prime dividing |G|. For every
p-subgroup H ⊂ G, the number of intermediate p-subgroups H ⊂ K ⊂ G with a fixed size is
≡ 1 mod p.
The case of Theorem 7.9 where H is trivial and the subgroups have maximal p-power
size is part of Sylow’s third theorem. The case when H is trivial and the fixed size is an
arbitrary p-power dividing |G| is due to Frobenius (1895).
We will derive Theorem 7.9 from the following result about group actions.
Theorem 7.10. Let G be a finite group acting transitively on a set X. Assume the stabilizer
subgroup of a point of X is a p-subgroup of G, where p is a prime. If pm | |X| then the
number of G-equivalence relations on X whose equivalence classes each have size pm is
≡ 1 mod p.
20 KEITH CONRAD

To deduce Theorem 7.9 from Theorem 7.10, apply Theorem 7.5 to the usual action of G
on G/H: it shows the G-equivalence relations on G/H are in bijection with the subgroups
between H and G, with the size of an equivalence class in an equivalence relation being
equal to the index of H in the subgroup corresponding to that equivalence relation. Since
H is a p-subgroup of G, an intermediate subgroup H ⊂ K ⊂ G is a p-subgroup if and only
if [K : H] is a p-power.
Theorems 7.9 and 7.10 are equivalent to each other. They are saying the same thing in
different languages.
Now we prove Theorem 7.10.

Proof. Write |X| = rpm . Fix a point m

x ∈ X. Set T = {Y ⊂ X : |Y | = p }, on which G
rpm
acts from the left. Note |T | = pm .
For Y ∈ T , the sets gY as g varies cover X (G acts transitively on X) and each has size
pm , so
|X| ≤ |{gY : g ∈ G}| · |Y |
with equality precisely when {gY : g ∈ G} is a partition of X. Writing out the sizes of X
and Y explicitly, this becomes
(7.2) r ≤ |{gY : g ∈ G}|
with equality if and only if {gY : g ∈ G} is a partition of X.
Set
KY = {g ∈ G : gY = Y }.
This is the stabilizer subgroup of Y for the action of G on T .
Choose y ∈ Y and let Hy = Staby = {g : gy = y}. By hypothesis this is a p-subgroup of
G. Write
Ye = {g ∈ G : gy ∈ Y }.
This is a subset of G. (There is no reason to expect it is closed under inversion, so it need
not be a subgroup of G.) Easily Hy ⊂ Ye and KY ⊂ Ye . For g and g 0 in Ye , gy = g 0 y if and
only if gHy = g 0 Hy . Since Hy ⊂ Ye , Ye is a union of as many left Hy -cosets as there are
elements of Y , so
(7.3) |Ye | = |Y ||Hy |,
which is a power of p.
If k ∈ KY and g ∈ Ye then kgy ∈ kY = Y , so kg ∈ Ye . Thus KY acts by left multiplication
on Ye . Since Ye is a subset of G, its KY -orbits are right KY -cosets, so |KY | | |Ye |. Therefore
by (7.3) KY is a p-subgroup of G.
Since {gY : g ∈ G} is a G-orbit in T and the stabilizer of Y is KY ,
|G| [G : Hy ]|Hy | rpm |Hy |
|{gY : g ∈ G}| = [G : KY ] = = = = rpi(Y )
|KY | |KY | |KY |
for some integer i(Y ). By (7.2) pi(Y ) ≥ 1, so i(Y ) ≥ 0. (If p does not divide r then we could
say i(Y ) ≥ 0 because rpi(Y ) ∈ Z.)
Now we apply the orbit-stabilizer formula to the G-action on T . Fix an element y ∈ X.
Each G-orbit in T is a collection of sets {gY : g ∈ G}; it covers X so we can choose Y to
contain y. (At least one of the sets gY contains y and we can relabel one of those as Y by
 TRANSITIVE GROUP ACTIONS 21

the transitivity of G on X.) Let Y1 , . . . , Yd be representatives for the different G-orbits in

T . Then
X d
|T | = |{gYj : g ∈ G}|,
j=1
so
 m Xd
rp
= rpi(Yj ) .
pm
j=1
rpm rpm −1



Since pm =r pm −1 ,
 m  Xd
rp − 1
(7.4) = pi(Yj ) ≡ |{j : i(Yj ) = 0}| mod p.
pm − 1
j=1

When does a Y ∈ T have i(Y ) = 0? From the definition of i(Y ), its vanishing is equivalent
to |{gY : g ∈ G}| = r, which is equivalent to the sets {gY : g ∈ G} forming a partition of
X. (See (7.2) and the surrounding text.) This partition contains equivalence classes for a
G-equivalence relation on X where the classes have size pm . Therefore
(7.5) |{j : i(Yj ) = 0}| = |{G-equiv. relns. with classes of size pm }|.
Notice the binomial coefficient on the left side of (7.4) is determined entirely by r and pm ,
not by the finer group structure of G. So far, G has been an arbitrary group admitting a
transitive action on some set with size rpm and having p-subgroups as its point-stabilizers.
A particular example of this is the group G = Z/(pm r) acting on itself from the left (trivial
stabilizers). With this choice, Corollary 7.7 implies (7.5) equals 1, so the left side of (7.4)
is 1 modulo p. Therefore by (7.4), (7.5) is 1 mod p for each G fitting the hypotheses of the
theorem. 
Remark 7.11. The trick at the end of the proof with the switch to the cyclic group can be
avoided. The left side of (7.4) can be calculated directly modulo p using a result of Lucas:
if a = a0 + a1 p + · · · + a
n pn and

b
= b0 +
n

b1 p + · · · + bn p where 0 ≤ ai , bi ≤ p − 1 for i < n
a a0 a1 an
and an , bn ≥ 0, then b ≡ b0 b1 · · · bn mod p. In particular, since

rpm − 1 = (p − 1) + (p − 1)p + · · · + (p − 1)pm−1 + (r − 1)pm ,

pm − 1 = (p − 1) + (p − 1)p + · · · + (p − 1)pm−1 ,
m −1
Qm−1 p−1
r−1

the congruence of Lucas shows rp
pm −1 ≡ i=0 p−1 · 0 ≡ 1 mod p.

Definition 7.12. Let G act transitively on X with |X| > 1. The action is called primitive
if there are no G-equivalence relations on X other than the two equivalence relations of
individual points and the whole set.
If an action of G on X is not transitive then the action has a G-equivalence relation on
X other than the two trivial relations: its orbits! (Well, the orbits are a trivial equivalence
relation when the action is trivial, but in that case every partition of X is a G-equivalence
relation). Therefore it is natural to include transitivity in the definition of a primitive
action: only transitive actions could have no non-trivial G-equivalence relations. Let’s look
at some examples and nonexamples.
22 KEITH CONRAD

Example 7.13. For n ≥ 2, the action of GLn (R) on Rn −{0} is transitive but not primitive:
a non-trivial GLn (R)-equivalence relation on Rn − {0} that respects the group action is
v ∼ v 0 when v 0 = cv for some c ∈ R× . This is saying the one-dimensional subspaces of Rn
with the origin excluded are equivalence classes in Rn − {0} for a relation preserved by the
usual action of GLn (R).
Example 7.14. When n ≥ 3 is a composite number, the action of Dn on the vertices of a
regular n-gon is transitive but not primitive: if d | n and 1 < d < n then the vertices of the
regular d-gons inside the n-gon are the equivalence classes for a non-trivial Dn -equivalence
relation on the vertex set.
Example 7.15. If |X| is prime then every transitive action of G on X is primitive because
a subset Y ⊂ X such that the gY ’s partition X has size dividing a prime, so |Y | is 1 or |X|.
In particular, when p is an odd prime the natural action of Dp is primitive.
To show a transitive action on a set of non-prime size is primitive it is useful to reformulate
the condition. Then more examples of primitive actions will easily follow.
Theorem 7.16. Let G act transitively on X. The following conditions are equivalent:
(1) the action is primitive,
(2) for some x ∈ X, Stabx is a maximal subgroup of G,
(3) for every x ∈ X, Stabx is a maximal subgroup of G.
Proof. Since G acts transitively, stabilizer subgroups of different points in X are conjugate,
so (2) and (3) are equivalent to each other. Properties (1) and (2) are equivalent by Theorem
7.5. 
Concretely, a primitive G-action is equivalent to the left multiplication action of G on
G/H where H is a maximal subgroup of G.
Example 7.17. In Example 7.13 we saw that for n ≥ 2, the natural transitive action of
GLn (R) on Rn − {0} is not primitive. The stabilizer subgroup of e1 = (1, 0, . . . , 0) (viewed
as a column vector) is the group of invertible n × n matrices
 
1 ∗ ··· ∗
 0 ∗ ··· ∗ 
(7.6)
 
 .. .. . . .. 
 . . . . 
0 ∗ ··· ∗
which is not a maximal subgroup in GLn (R) since it is strictly contained in the subgroup of
matrices that are like (7.6) but have a nonzero entry in the upper left. That is the subgroup
of invertible n × n matrices that preserve the subspace Re1 rather than fix each element of
it, which is what matrices in (7.6) do.
Example 7.18. We already saw that the natural action of Dn is not primitive when n is
composite (Example 7.14). In terms of maximal subgroups, the stabilizer subgroup of the
vertex 1 is Stab1 = {1, s} = hsi, and if n is composite this subgroup is not maximal: letting
d | n and 1 < d < n we have hsi ⊂ hrn/d , si ⊂ hr, si. When n = p is an odd prime, the
natural action of Dp is primitive since the vertex set has prime size (or the subgroup hsi
has index p in Dp so it is maximal).
Corollary 7.19. A doubly transitive group action is primitive.
 TRANSITIVE GROUP ACTIONS 23

Proof. Use Theorems 4.19 and 7.16. 

Example 7.20. The natural actions of Sn (n ≥ 3) and An (n ≥ 4) on {1, 2, . . . , n} are
primitive, as is the action of Aff(F ) on F and the actions of GL2 (F ) and SL2 (F ) on the
one-dimensional subspaces of F 2 .
By Corollary 7.19, doubly transitive actions are special cases of primitive actions. Prim-
itive actions are transitive, so primitivity lies between transitivity and double transitivity.
The inclusions are strict, e.g., the natural action of Dn on a regular n-gon (n ≥ 3) is transi-
tive but not primitive for composite n and is primitive but not doubly transitive for prime
n. The natural action of A3 is primitive (Example 7.15) but not doubly transitive. Table 1
summarizes the situation.

Type Space
transitive G/H
primitive G/H, H maximal
doubly transitive G/H, G = H ∪ HgH
Table 1. Levels of Transitivity

In a sense, transitive and primitive G-actions are almost “dual” concepts. If G acts on
the sets S and T , let a G-map from S to T be a function f : S → T that respects the actions:
f (gs) = gf (s) for all g ∈ G and s ∈ S. Then the action of G on a set X is transitive if
and only if every G-map to X is surjective, while the action on X is primitive if and only
if the action is non-trivial and every (nonconstant) G-map out of X is injective. (We met
G-maps in a concrete setting in Corollary 3.5.)
Theorem 4.9 and Theorem 5.1 (Iwasawa’s theorem) generalize from doubly transitive
actions to primitive actions, as follows.
Theorem 7.21. Suppose G acts primitively on a set X. Every normal subgroup N C G
acts on X either trivially or transitively.
Proof. Let H = Stabx be the stabilizer of a point in X. Then H ⊂ N H ⊂ G. Since H
is maximal, N H = H or N H = G. If N H = H then N ⊂ H, so N ⊂ gHg −1 for all g.
Since the stabilizer subgroup of each point in X is conjugate to H, we conclude that N acts
trivially on X. Now suppose N H = G. Then
X = Gx = N Hx = N x,
so N acts transitively on X. 
Remark 7.22. Using Theorem 7.21 in place of Theorem 4.9, Lemma 6.1 remains true if
the doubly transitive hypothesis is replaced with primitivity, as the reader can check.
Theorem 7.23 (Iwasawa). Suppose G acts primitively on a set X and, for some x ∈ X,
Stabx has an abelian normal subgroup whose conjugate subgroups generate G. If [G, G] = G
then G/K is a simple group, where K is the kernel of the action of G on X.
Proof. The proof is just like the proof of Theorem 5.1; replace the reference to Theorem 4.9
with Theorem 7.21. 
There are groups whose simplicity is proved using a primitive action that is not doubly
transitive, e.g., simplicity of most projective symplectic groups.
24 KEITH CONRAD

Appendix A. Sylow subgroups under intersection and quotient

We prove here a theorem about Sylow subgroups needed in the proof of Corollary 3.5.
Theorem A.1. Let G be a finite group and N be a normal subgroup.
(1) If P is a p-Sylow subgroup of G then P ∩ N is a p-Sylow subgroup of N .
(2) P N/N is a p-Sylow subgroup of G/N .
Proof. (1): The group P ∩ N is a p-group since it is contained in P . To show it is a p-Sylow
subgroup of N we will show the index [N : P ∩ N ] is not divisible by p. The set P N is a
subgroup of G since N C G and |P N | = |P ||N |/|P ∩ N |, so [N : P ∩ N ] = |N |/|P ∩ N | =
|P N |/|P | = [P N : P ], which is a factor of [G : P ] and thus is not divisible by p. (Here is an
alternate proof. By the Sylow theorems P ∩ N is contained in a p-Sylow subgroup of N , say
K. Then K, being a p-subgroup of G, is contained in a conjugate of P : K ⊂ gP g −1 . Thus
g −1 Kg ⊂ P . Also g −1 Kg ⊂ g −1 N g = N , so g −1 Kg ⊂ P ∩ N ⊂ K. Since |K| = |g −1 Kg|,
we get |P ∩ N | = |K|, so P ∩ N = K is a p-Sylow subgroup of N .)
(2): First observe that P N/N is a p-group (either because every element has p-power
order or because P N/N ∼ = P/(P ∩ N )). Using the inclusions
G ⊃ P N ⊃ N, G ⊃ P N ⊃ P,
the first one shows [G/N : P N/N ] = [G : P N ] and the second one shows [G : P N ] ≡
6
0 mod p. Therefore P N/N is a p-subgroup of G/N whose index is not divisible by p, so
P N/N is a p-Sylow subgroup of G/N . 

References
[1] N. Biggs and A. T. White, “Permutation Groups and Combinatorial Structures,” Cambridge Univ. Press,
Cambridge, 1979.
[2] R. Chapman, An Elementary Proof of the Simplicity of the Mathieu Groups M11 and M23 , Amer. Math.
Monthly 102 (1995), 544–545.
[3] J. Dixon and B. Mortimer, “Permutation Groups,” Springer-Verlag, New York, 1996.
[4] L. Weisner, Some properties of prime-power groups, Trans. Amer. Math. Soc. 38 (1935), 485–492.