Simplicity of An
Simplicity of An
KEITH CONRAD
1. Introduction
A finite group is called simple when it is nontrivial and its only normal subgroups are
the trivial subgroup and the whole group.
For instance, a finite group of prime size is simple, since it in fact has no non-trivial
proper subgroups at all (normal or not). A finite abelian group G not of prime size, is not
simple: let p be a prime factor of |G|, so G contains a subgroup of order p, which is a normal
since G is abelian and is proper since |G| > p. Thus, the abelian finite simple groups are
the groups of prime size.
When n ≥ 3 the group Sn is not simple because it has a nontrivial normal subgroup An .
But the groups An are simple, provided n ≥ 5.
Theorem 1.1 (C. Jordan, 1875). For n ≥ 5, the group An is simple.
The restriction n ≥ 5 is optimal, since A4 is not simple: it has a normal subgroup of
size 4, namely {(1), (12)(34), (13)(24), (14)(23)}. The group A3 is simple, since it has size
3, and the groups A1 and A2 are trivial.
We will give five proofs of Theorem 1.1. Section 2 includes some preparatory material
and later sections give the proofs of Theorem 1.1. In the final section, we give a quick
application of the simplicity of alternating groups and give references for further proofs not
treated here.
2. Preliminaries
We give two lemmas about alternating groups An for n ≥ 5 and then two results on
symmetric groups Sn for n ≥ 5.
Lemma 2.1. For n ≥ 3, An is generated by 3-cycles. For n ≥ 5, An is generated by
permutations of type (2, 2).
Proof. That the 3-cycles generate An for n ≥ 3 has been seen earlier in the course. To show
permutations of type (2, 2) generate An for n ≥ 5, it suffices to write any 3-cycle (abc) in
terms of such permutations. Pick d, e 6∈ {a, b, c}. Then note
(abc) = (ab)(de)(de)(bc).
The 3-cycles in Sn are all conjugate in Sn , since permutations of the same cycle type in
Sn are conjugate. Are 3-cycles conjugate in An ? Not when n = 4: (123) and (132) are not
conjugate in A4 . But for n ≥ 5 we do have conjugacy in An .
Lemma 2.2. For n ≥ 5, any two 3-cycles in An are conjugate in An .
1
2 KEITH CONRAD
3. First proof
Our first proof of Theorem 1.1 is based on the one in [2, pp. 149–150].
We begin by showing A5 is simple.
Theorem 3.1. The group A5 is simple.
Proof. We want to show the only normal subgroups of A5 are {(1)} and A5 . This will be
done in two ways.
Our first method involves counting the sizes of the conjugacy classes. There are 5 conju-
gacy classes in A5 , with representatives and sizes as indicated in the following table.
Rep. (1) (12345) (21345) (12)(34) (123)
Size 1 12 12 15 20
If A5 has a normal subgroup N , then N is a union of conjugacy classes – including {(1)} –
whose total size divides 60. However, no sum of the above numbers which includes 1 is a
factor of 60 except for 1 and 60. Therefore N is trivial or A5 .
For the second proof, let N C A5 with |N | > 1. We will show N contains a 3-cycle. It
follows that N = An by Lemmas 2.1 and 2.2.
Pick σ ∈ N with σ 6= (1). The cycle structure of σ is (abc), (ab)(cd), or (abcde), where
different letters represent different numbers. Since we want to show N contains a 3-cycle,
we may suppose σ has the second or third cycle type. In the second case, N contains
((abe)(ab)(cd)(abe)−1 )(ab)(cd) = (be)(cd)(ab)(cd) = (aeb).
In the third case, N contains
((abc)(abcde)(abc)−1 )(abcde)−1 = (adebc)(aedcb) = (abd).
Therefore N contains a 3-cycle, so N = A5 .
4 KEITH CONRAD
Lemma 3.2. When n ≥ 5, any σ 6= (1) in An has a conjugate σ 0 6= σ such that σ(i) = σ 0 (i)
for some i.
For example, if σ = (12345) in A5 then σ 0 = (345)σ(345)−1 = (12453) has the same value
at i = 1 as σ does.
Proof. Let σ be a non-identity element of An . Let r be the longest length of a disjoint cycle
in σ. Relabelling, we may write
σ = (12 . . . r)π,
where (12 . . . r) and π are disjoint.
If r ≥ 3, let τ = (345) and σ 0 = τ στ −1 . Then σ(1) = 2, σ 0 (1) = 2, σ(2) = 3, and
σ (2) = 4. Thus σ 0 6= σ and both take the same value at 1.
0
4. Second proof
Our next proof is taken from [6, p. 108]. It does not use induction on n, but we do need
to know A6 is simple at the start.
Theorem 4.1. The group A6 is simple.
Proof. We follow the first method of proof of Theorem 3.1. Here is the table of conjugacy
classes in A6 .
Rep. (1) (123) (123)(456) (12)(34) (12345) (23456) (1234)(56)
Size 1 40 40 45 72 72 90
SIMPLICITY OF An 5
A tedious check shows no sum of these sizes, which includes 1, is a factor of 6!/2 except for
the sum of all the terms. Therefore the only non-trivial normal subgroup of A6 is A6 .
Now we prove the simplicity of An for larger n by reducing directly to the case of A6 .
Proof. Since A5 and A6 are known to be simple by Theorems 3.1 and 4.1, pick n ≥ 7 and
let N C An be a non-trivial subgroup. We will show N contains a 3-cycle.
Let σ be a non-identity element of N . It moves some number. By relabelling, we may
suppose σ(1) 6= 1. Let τ = (ijk), where i, j, k are not 1 and σ(1) ∈ {i, j, k}. Then
τ στ −1 (1) = τ (σ(1)) 6= σ(1), so τ στ −1 6= σ. Let ϕ = τ στ −1 σ −1 , so ϕ 6= (1). Writing
ϕ = (τ στ −1 )σ −1 ,
we see ϕ ∈ N . Now write
ϕ = τ (στ −1 σ −1 ),
Since τ −1 is a 3-cycle, στ −1 σ −1 is also a 3-cycle. Therefore ϕ is a product of two 3-cycles, so
ϕ moves at most 6 numbers in {1, 2, . . . , n}. Let H be the copy of A6 inside An corresponding
to the even permutations of those 6 numbers (possibly augmented to 6 arbitrarily if in fact
ϕ moves fewer numbers). Then N ∩ H is non-trivial (it contains ϕ) and it is a normal
subgroup of H. Since H ∼ = A6 , which is simple, N ∩ H = H. Thus H ⊂ N , so N contains
a 3-cycle.
5. Third proof
Our next proof is by induction, and uses conjugacy classes instead of Lemma 3.2. It is
based on [9, §2.3].
Lemma 5.1. If n ≥ 6 then every non-trivial conjugacy class in Sn and An has at least n
elements.
The lower bound n in Lemma 5.1 is actually quite weak as n grows. But it shows that
the size of each non-trivial conjugacy class in Sn and An grows with n.
Proof. For n ≥ 6, pick σ ∈ Sn with σ 6= (1). We want to look at the conjugacy class of σ in
Sn , and if σ ∈ An we also want to look at the conjugacy class of σ in An , and our goal in
both cases is to find at least n elements in the conjugacy class.
Case 1: The disjoint cycle decomposition of σ includes a cycle with length greater than
2. Without loss of generality, σ = (123 . . . ) . . . .
For 3 ≤ k ≤ n, fix a choice of ` 6∈ {1, 2, 3, k} (which is possible since n ≥ 5) and let
αk = (2k`) and βk = (3k`). Then αk σαk−1 has the effect 1 → 1 → 2 → k and βk σβk−1 has
the effect 1 → 1 → 2 → 2 and 2 → 2 → 3 → k. This tells us that the conjugates
α3 σα3−1 , . . . , αn σαn−1 , β3 σβ3−1 , . . . , βn σβn−1
are all different from each other: the conjugates by the α’s have different effects on 1, the
conjugates by the β’s have different effects on 2, and a conjugate by an α is not a conjugate
by a β since they have different effects on 1. Since these conjugates are different, the number
of conjugates of σ is at least 2(n − 2) > n. Because αk and βk are 3-cycles, if σ ∈ An then
these conjugates are in the An -conjugacy class of σ.
Case 2: The disjoint cycle decomposition of σ only has cycles with length 1 or 2. Therefore
without loss of generality σ is a transposition or a product of at least 2 disjoint transposi-
tions.
6 KEITH CONRAD
If σ is a transposition, then its Sn -conjugacy class is the set of all transpositions (ij)
2
where 1 ≤ i < j ≤ n, and the number of these permutations is n2 = n 2−n , which is greater
than n for n ≥ 6.
If σ is a product of at least 2 disjoint transpositions, then without loss of generality
σ = (12)(34) . . . , where the terms in . . . don’t involve 1, 2, 3, or 4.
For 5 ≤ k ≤ n, let αk = (12)(3k), βk = (13)(2k), and γk = (1k)(23). Then αk σαk−1 has
the effect
1 → 2 → 1 → 2, 2 → 1 → 2 → 1, k → 3 → 4 → 4,
βk σβk−1 has the effect
1 → 3 → 4 → 4, 3 → 1 → 2 → k, k → 2 → 1 → 3,
and γk σγk−1 has the effect
2 → 3 → 4 → 4, 3 → 2 → 1 → k, k → 1 → 2 → 3.
The conjugates of σ by the α’s are different from each other since they take different elements
to 4, the conjugates of σ by the β’s are different from each other since they take different
elements to 3, and the conjugates of σ by the γ’s are different from each other since they
take different elements to 3. Conjugates of σ by an α and a β are different since they send
1 to different places, conjugates of σ by an α and a γ are different since they send 2 to
different places, and conjugates of σ by a β and a γ are different since they send different
elements to 4 (1 for the β’s and 2 for the γ’s). In total the number of conjugates of σ we
have written down (which are all conjugates by 3-cycles, hence they are conjugates in An
if σ ∈ An ) is 3(n − 4), and 3(n − 4) ≥ n if n ≥ 6.
Proof. We argue by induction on n, the case n = 5 having already been settled by Theorem
3.1. Say n ≥ 6. Let N C An with N 6= {(1)}. Since N is normal and non-trivial, it contains
non-identity conjugacy classes in An . By Lemma 5.1, any non-identity conjugacy class in
An has size at least n when n ≥ 6. Therefore, by counting the trivial conjugacy class and
a non-trivial conjugacy class in N , we see |N | ≥ n + 1.
Using a wholly different argument, we now show that |N | ≤ n if N 6= An , which will be
a contradiction. Pick 1 ≤ i ≤ n. Let Hi ⊂ An be the subgroup fixing i, so Hi ∼ = An−1 . In
particular, Hi is a simple group by induction. Notice each Hi contains a 3-cycle.
The intersection N ∩ Hi is a normal subgroup of Hi , so simplicity of Hi implies N ∩ Hi
is either {(1)} or Hi . If N ∩ Hi = Hi for some i, then Hi ⊂ N . Since Hi contains a 3-cycle,
N does as well, so N = An by Lemmas 2.1 and 2.2. (This part resembles part of our first
proof of simplicity of An , but we will now use Lemma 5.1 instead of Lemma 3.2 to show
the possibility that N ∩ Hi = {(1)} for all i is absurd.)
Suppose N 6= An . Then, by the previous paragraph, N ∩ Hi = {(1)} for all i. Therefore
each σ 6= (1) in N acts on {1, 2, . . . , n} without fixed points (otherwise σ would be a non-
identity element in some N ∩ Hi ). That implies each σ 6= (1) in N is completely determined
by the value σ(1): if τ 6= (1) is in N and σ(1) = τ (1), then στ −1 ∈ N fixes 1, so στ −1 is
the identity, so σ = τ .
There are only n − 1 possible values for σ(1) ∈ {2, 3, . . . , n}, so N − {(1)} has size at
most n − 1, hence |N | ≤ n. We already saw from Lemma 5.1 that |N | ≥ n + 1, so we have
a contradiction.
SIMPLICITY OF An 7
6. Fourth proof
Our next proof, based on [3, p. 50], is very computational.
Proof. Let N C An be a non-trivial normal subgroup. We will show N contains a 3-cycle.
Pick σ ∈ N , σ 6= (1). Write
σ = π1 π2 · · · πk ,
where the πj ’s are disjoint cycles. In particular, they commute, so we can relabel them at
our convenience. Eliminate any 1-cycles from the product.
Case 1: Some πi has length at least 4. Relabelling, we can write
π1 = (12 · · · r)
with r ≥ 4. Let ϕ = (123). Then ϕσϕ−1 ∈ N and
ϕσϕ−1 = ϕπ1 ϕ−1 π2 · · · πk
= ϕπ1 ϕ−1 π1−1 σ
= (123)(123 · · · r)(132)(r · · · 21)σ
= (124)σ,
so (124) = ϕσϕ−1 σ −1 ∈ N .
Case 2: Each πi has length ≤ 3, and at least two have length 3 (so n ≥ 6). Without loss
of generality, π1 = (123) and π2 = (456). Let ϕ = (124). Then
ϕσϕ−1 = ϕπ1 π2 ϕ−1 π3 · · · πk
= ϕπ1 π2 ϕ−1 π2−1 π1−1 σ
= (124)(123)(456)(142)(465)(132)σ
= (12534)σ,
so ϕσϕ−1 σ −1 = (12534) ∈ N . Now run through Case 1 with this 5-cycle to find a 3-cycle
in N .
Case 3: Exactly one πi has length 3, and the rest have length ≤ 2. Without loss of
generality, π1 = (123) and the other πi ’s are 2-cycles. Then σ 2 = π12 is in N , and π12 = (132).
Case 4: All πi ’s are 2-cycles, so necessarily k > 1. Write π1 = (12) and π2 = (34). Let
ϕ = (123). Then
ϕσϕ−1 = ϕπ1 π2 ϕ−1 π3 · · · πk
= ϕπ1 π2 ϕ−1 π2−1 π1−1 σ
= (123)(12)(34)(132)(34)(12)σ
= (13)(24)σ,
so
ϕσϕ−1 σ −1 = (13)(24) ∈ N.
Let ψ = (135). Then
(13)(24)ψ(13)(24)ψ −1 = (13)(24)(135)(13)(24)(153)
= (13)(135)(13)(153)
= (135),
so N contains a 3-cycle.
8 KEITH CONRAD
7. Fifth proof
Our final proof is taken from [8, p. 295].
Let N C An with N not {(1)} or An . We will study N as a subgroup of Sn . By Lemma
2.4, N is not a normal subgroup of Sn . This means the normalizer of N inside Sn is a
proper subgroup, which contains An , so
(7.1) An = NSn (N ).
For any transposition τ in Sn , τ 6∈ NSn (N ) by (7.1), so τ N τ −1 6= N . Since N C An and
τ N τ −1 is a subgroup of An , the product set N · τ N τ −1 is a subgroup of An . We have the
chain of inclusions
N ∩ τ N τ −1 ⊂ N ⊂ N · τ N τ −1 ⊂ An ,
where the first and second are strict.
We will now show, for any transposition τ in Sn , that
(7.2) N ∩ τ N τ −1 C Sn , N · τ N τ −1 C Sn .
The proof of (7.2) is a bit tedious , so first let’s see why (7.2) leads to a contradiction.
It follows from (7.2) and Lemma 2.4 that
(7.3) N ∩ τ N τ −1 = {(1)}, N · τ N τ −1 = An
for any transposition τ in Sn . By (7.3), |An | = |N | · |τ N τ −1 | = |N |2 , so n! = 2|N |2 . This
tells us |N | must be even, so N has an element, say σ, of order 2. Then σ is a product of
disjoint 2-cycles. There is a transposition ρ in Sn which commutes with σ: just take for ρ
one of the transpositions in the disjoint cycle decomposition of σ. Then
σ = ρσρ−1 ∈ N ∩ ρN ρ−1 .
From (7.3), using ρ for the arbitrary τ there, N ∩ρN ρ−1 is trivial, so we have a contradiction.
(As another way of reaching a contradiction from the equation n! = 2|N |2 , we can use
Bertrand’s postulate – proved by Chebyshev – that there is always a prime strictly between
m and 2m for any m > 1. That means, taking m = n!/4, the ratio n!/2 can’t be a perfect
square.)
It remains to check the two conditions in (7.2). In both cases, we show the subgroups
are normalized by An and by τ , so the normalizer contains hAn , τ i = Sn .
First consider N ∩ τ N τ −1 . It is clearly normalized by τ . Now pick any π ∈ An . Then
πN π −1 = N since N C An , and
(7.4) π(τ N τ −1 )π −1 = τ (τ −1 πτ )N (τ −1 π −1 τ )τ −1 = τ N τ −1
since τ −1 πτ ∈ An . Therefore
π(N ∩ τ N τ −1 )π −1 = πN π −1 ∩ πτ N τ −1 π −1 = N ∩ τ N τ −1 ,
so An normalizes N ∩ τ N τ −1 .
Now we look at N · τ N τ −1 . Pick an element of this product, say
σ = σ1 τ σ2 τ −1 ,
where σ1 , σ2 ∈ N . Then, since N C An ,
τ στ −1 = τ σ1 τ σ2 τ −2 = τ σ1 τ σ2 ∈ τ N τ −1 · N = N · τ N τ −1 ,
which shows τ normalizes N · τ N τ −1 .
SIMPLICITY OF An 9
8. Concluding Remarks
The standard counterexample to the converse of Lagrange’s theorem is A4 : it has size 12
but no subgroup of index 2. For n ≥ 5, the groups An also have no subgroup of index 2,
since any index-2 subgroup of a group is normal and An is simple.
In fact, something stronger is true.
Corollary 8.1. For n ≥ 5, any proper subgroup of An has index at least n.
This is an analogue of Theorem 2.7.
Proof. Let H be a proper subgroup of An , with index m > 1. Consider the left multiplication
action of An on An /H. This gives a group homomorphism
ϕ : An → Sym(An /H) ∼ = Sm .
Let K be the kernel of ϕ, so K ⊂ H (why?) and K C An . By simplicity of An , K is trivial.
Therefore An injects into Sm , so (n!/2) | m!, which implies n ≤ m.
The lower bound of n is sharp since [An : An−1 ] = n. Corollary 8.1 is false for n = 4: A4
has a subgroup of index 3.
Remark 8.2. What the proof of Corollary 8.1 shows more generally is that if G is a finite
simple group and H is a subgroup with index m > 1, then there is an embedding of G into
Sm , so |G| | m!. With G fixed, this divisibility relation puts a lower bound on the index of
any proper subgroup of G.
A reader who wants to see more proofs that An is simple for n ≥ 5 can look at [4, pp. 247-
248] or [5, pp. 32–33] for another way of showing a non-trivial normal subgroup contains a
3-cycle, or at [1, §1.7] or [7, pp. 295–296] for a proof based on the theory of highly transitive
permutation groups.
References
[1] N. L. Biggs, A. T. White, “Permutation Groups and Combinatorial Structures,” Cambridge Univ. Press,
Cambridge, 1979.
[2] D. Dummit, R. Foote, “Abstract Algebra,” 3rd ed., J. Wiley, 2004.
[3] T. Hungerford, “Algebra,” Springer-Verlag, New York, 1980.
[4] N. Jacobson, “Basic Algebra I,” 2nd ed., W. H. Freeman, New York, 1985.
[5] S. Lang, “Algebra,” revised 3rd ed., Springer-Verlag, New York, 2002.
[6] J. Rotman, “Advanced Modern Algebra,” Prentice-Hall, Upper Saddle River, 2002.
[7] W. R. Scott, “Group Theory,” Dover, New York, 1987.
[8] M. Suzuki, “Group Theory I,” Springer-Verlag, Berlin, 1982.
[9] R. Wilson, “The Finite Simple Groups,” Springer-Verlag, New York, 2009,