CHAPTER 1

Groups

1.1. Definitions and Easy Facts

Definition 1.1. A group is a nonempty set G equipped with a binary opera-
tion · satisfying the following axioms.
(i) Associativity : (a · b) · c = a · (b · c) for all a, b, c 2 G.
(ii) Existence of identity : There exists an element e 2 G, called the identity
of G, such that e · a = a · e = a for all a 2 G.
(iii) Existence of inverse : For each a 2 G, there exists an element a 1 2 G,
called the inverse of a, such that a · a 1 = a 1 · a = e.
If a · b = b · a for all a, b 2 G, G is called abelian. (It is more appropriate to treat
abelian groups as Z-modules; see §2.6.)
In an abstract group (G, ·), we write a · b as ab for a, b 2 G. The cardinality
|G| is called the order of G
Fact. Let G be a group.
(i) The identity e of G and the inverse of a 2 G are unique.
(ii) (Cancellation) For a, b, c 2 G, ab = ac , b = c; ba = ca , b = c; ab = e
if and only if a and b are inverses of each other.
(iii) For a1 , . . . , an 2 G, parentheses are not needed to define the product
a1 · · · an because of the associativity. Moreover, (a1 · · · an ) 1 = an 1 · · · a1 1 .
Proof. (i) Let e0 be another identity of G. Then e0 = e0 e = e. Let b be
another inverse of a. Then b = be = b(aa 1 ) = (ba)a 1 = ea 1 = a 1 . ⇤
Examples. Here are a few examples of groups; certain terms involved are
either defined later in the book or borrowed from other areas of mathematics.
Abelian groups : (Z, +), (Q, +), (R, +), (C, +); the multiplicative group (F ⇥ , ·)
of a field F , where F ⇥ = F \ {0}, e.g., (Q⇥ , ·), (R⇥ , ·), (C⇥ , ·), etc. Let n 2 Z+ and
Zn = {x + nZ : x 2 Z}, where x + nZ = {x + nk : k 2 Z} is the congruence class of
x modulo n. For x + nZ, y + nZ 2 Zn , define (x + nZ) + (y + nZ) = x + y + nZ and
(x + nZ) · (y + nZ) = xy + nZ, which are well-defined. Then (Zn , +) is an abelian
group with |Zn | = n. Let Z⇥ n = {x + nZ 2 Zn : gcd(x, n) = 1}. Then (Zn , ·) is an
⇥

abelian group with |Zn | = (n), where (n) = |{0  x  n 1 : gcd(x, n) = 1}|
⇥

is the Euler -function. An element x + nZ in Zn or Z⇥ n is written as x when the

abuse of notation is not harmful.
Automorphism groups : These are groups of bijections which preserve certain
structures; e.g., SX = the group of all permutations of a set X; GL(V ) = the group
of all invertible linear transformations of a vector space V ; the group of isometries of
a metric space; groups of automorphisms of groups, rings, modules, fields, graphs,
etc.
1
2 1. GROUPS

Groups as invariants : These are groups that are used to detect intrinsic differ-
ences in mathematical structures; e.g., the fundamental group and the homology
groups of a topological space; the ideal class group in number theory; various char-
acteristic subgroups of a group.
Laws of exponents. Let G be a group. For a 2 G and m, n 2 Z, am an =
am+n , (am )n = amn . (Note: a0 := e; a n := (a 1 )n for n 2 Z+ .)
Additive notation (usually for abelian groups). We write a + b for ab, 0 for
e, a for a 1 and na for an .
Cartesian product. If G and H are groups, then G ⇥ H is a group with its
operation defined componentwise.
Homomorphism. Let G and H be groups. A map f : G ! H is called a
homomorphism if f (ab) = f (a)f (b) for all a, b 2 G. It follows that f (eG ) = eH ,
where eG and eH are the identities of G and H, respectively, and f (a 1 ) = f (a) 1
for a 2 G. The kernel of f is ker f = f 1 (eH ) = {a 2 G : f (a) = eH }. A
homomorphism f is one-to-one if and only if ker f = {eG }. If f : G ! H is a
bijective homomorphism, it is called an isomorphism, in which case f 1 : H ! G
is also an isomorphism. A one-to-one homomorphism f : G ! H is also called
an embedding of G in H and is denoted as f : G ,! H. Two groups G and H
are called isomorphic, denoted as G ⇠ = H, if there is an isomorphism between
them. Isomorphic groups have the same structure. Let Aut(G) be the set of all
automorphisms of G (isomorphisms from G to G). The group (Aut(G), ) is called
the automorphism group of G.
Example. The general linear group of degree n over a field F , denoted by
GL(n, F ), is the multiplicative group of all n ⇥ n invertible matrices over F .
The determinant, det : GL(n, F ) ! F ⇥ , A 7! det A, is a homomorphism, and
SL(n, F ) := ker(det) is the special linear group of degree n over F .
Example. Let G the set of all Möbius transformations of C [ {1}, i.e., func-
tions of the form (az + b)/(cz + d), where ad bc 6= 0. Then (G, ) is a group and
the map
: GL(2, C) ! G
h a b i az + b
7 !
c d cz + d
is a homomorphism with ker = {aI2 : a 2 C⇥ }.
Subgroups. Let (G, ·) be a group. A subset H ⇢ G is called a subgroup of G,
denoted as H < G, if (H, ·) is a group. If H < G, the identity of H is the same
as the identity of G; this follows from eH eH = eH . For ; =
6 H ⇢ G, H < G if and
only if H is closed under multiplication and inversion. The intersection of a family
of subgroups is a subgroup.
Proposition 1.2. Let G be a group and ; = 6 H ⇢ G.
(i) H < G if and only if ab 1 2 H for all a, b 2 H.
(ii) If |H| < 1, then H < G if and only if H is closed under multiplication.
Proof. (ii) (() We only have to show that H contains e and is closed under
inversion. Let a 2 H. The map x 7! ax from H to H is one-to-one hence is onto.
 1.1. DEFINITIONS AND EASY FACTS 3

Thus there exists x 2 H such that ax = a and hence e = x 2 H. Also, there exists
y 2 H such that ay = e, so a 1 = y 2 H. ⇤
For X ⇢ G, the intersection
\
hXi := H
H:X⇢H<G

is the smallest subgroup of G containing X; it is called the subgroup of G generated

by X. We have
hXi = {ae11 · · · aenn : n 0, a1 , . . . , an 2 X, e1 , . . . , en 2 Z}.
A group G is said to be finitely generated if G = hXi for some X ⇢ G with |X| < 1.
If G = hai, G is called a cyclic group.
Example. Let F be a field and X ⇢ GL(n, F ) be the set of all elementary
matrices. Then hXi = GL(n, F ).
Cosets. Let H < G be groups and a 2 G. The left coset of H in G with
representative a is aH = {ah : h 2 H}; the right coset with representative a is
Ha = {ha : h 2 H}. Let G/H = {aH : a 2 G} and H\G = {Ha : a 2 G}. (Here
H\G does not mean set difference.) One has aH = bH if and only if b 1 a 2 H;
Ha = Hb if and only if ab 1 2 H. The cosets in G/H (H\G) form a partition of
G.
Proposition 1.3. Let H < G.
(i) |G/H| = |H\G|.
(ii) |G| = |G/H| |H|.
Proof. (i) For X ⇢ G, let X 1 = {x 1 : x 2 X}. The map ↵ : G/H ! H\G,
aH 7! (aH) 1 = Ha 1 is a bijection.
(ii) The map f : G ! G/H, a 7! aH is onto. For each aH 2 G/H, f 1 (aH) =
aH, so |f 1 (aH)| = |aH| = |H|. Hence |G| = |G/H| |H|. ⇤
In Proposition 1.3, |G/H| is called the index of H in G and is denoted
F by [G : H].
If K F
< H < G, then [G : K] F = [G : H][H : K]. (Assume that G = i2I ai H and
H = j2J bj K. Then G = i2I,j2J ai bj K.)

Corollary 1.4 (Lagrange’s theorem). If H < G and |G| < 1, then |H| |G|.
For a 2 G, the order of a is
(
min{n 2 Z+ : an = e} if there is n 2 Z+ such that an = e,
o(a) = |hai| =
1 otherwise.

Fact. If |G| < 1 and a 2 G, then o(a) |G| and a|G| = e.

Example (Euler’s theorem). For n 2 Z+ and x 2 Z with gcd(x, n) = 1, we
have x (n) ⌘ 1 (mod n).
⇥
Proof. In Z⇥
n , we have x
(n)
= x|Zn | = 1. ⇤

Fact. Let H and K be subgroups of G. Then |H||K| = |HK||H \ K|, where

HK = {hk : h 2 H, k 2 K}. If H and K are finite, then |HK| = |H||K|/|H \ K|.
4 1. GROUPS

Proof. Define f : H ⇥ K ! HK, (h, k) 7! hk. For hk 2 HK, f 1 (hk) =

{(ha, a 1 k) : a 2 H \ K} and hence |f 1 (hk)| = |H \ K|. Therefore |H ⇥ K| =
|HK||H \ K|. ⇤

1.2. Normal Subgroups and Quotient Groups

Definition 1.5. Let H < G be groups. H is called a normal subgroup of G,
denoted as H C G, if aH = Ha for all a 2 G. If H C G, then
(G/H) ⇥ (G/H) ! G/H
(aH, bH) 7 ! abH
is a well-defined binary operation on G/H which makes G/H a group. (If aH = a1 H
and bH = b1 H, then abH = ab1 H = aHb1 = a1 Hb1 = a1 b1 H.) The group G/H is
called the quotient group or the factor group of G by H.
The canonical homomorphism. Let H C G. Then
⇡: G ! G/H
a 7 ! aH
is an onto homomorphism with ker ⇡ = H.
Fact. Let H < G. Then H C G , aHa 1
= H for all a 2 G , aHa 1
⇢H
for all a 2 G. (Note: For a given a 2 G, aHa 1
⇢ H does not imply aHa 1
= H;
see Exercise ??.)
Fact. Let f : G ! H be a homomorphism and let K < H. Then f (G) < H and
f 1 (K) < G. If K C H, then f 1 (K) C G. If f is onto, then K C H , f 1 (K) C
G. ((() For all x 2 G, xf 1 (K)x 1 ⇢ f 1 (K), so K f (xf 1 (K)x 1 ) =
f (x)Kf (x) , where f (x) covers H as x runs through G.)
1

Fact. Let G be a group.

(i) H C G if and only if H = ker f for some homomorphism f : G ! K.
(ii) The center of G is Z(G) = {a 2 G : ax = xa for all x 2 G}. We have
Z(G) C G. Moreover, H C G for all H < Z(G).
(iii) If [G : H] = 2, H C G.

Proof. (i) ()) We have H = ker ⇡, where ⇡ : G ! G/H is the canonical

homomorphism.
(iii) Let a 2 G. If a 2 H, then aH = H = Ha. If a 2
/ H, then aH = G\H = Ha
since G = H t aH = H t Ha. ⇤

Fact. Let G be a group. For H, K ⇢ G, define HK = {hk : h 2 H, k 2 K}.

(i) If H < G and K < G, then HK < G if and only if HK = KH.
(ii) If H < G and K C G, then HK < G.
(iii) If H C G and K C G, then HK C G.

Proof. (i) ()) We have HK = (HK) 1

=K 1
H 1
= KH.
(() Since HK(HK) 1 = HKK 1
H 1
= HKH = HHK = HK, we have
HK < G by Proposition 1.2 (i). ⇤
 1.2. NORMAL SUBGROUPS AND QUOTIENT GROUPS 5

Theorem 1.6 (The correspondence theorem). Let N C G and let A = {K <

G : N ⇢ K}, B = the set of all subgroups of G/N . Then the map
f: A ! B
K 7 ! K/N
is a bijection. Moreover, K/N C G/N if and only if K C G.
Proof. Let ⇡ : G ! G/N be the canonical homomorphism. Define g : B ! A,
! ⇡ 1 ( ). We show that f g = idB and g f = idA .
For each 2 B, we have (f g)( ) = ⇡ 1 ( )/N = ⇡(⇡ 1 ( )) = . (⇡(⇡ 1 ( )) =
since ⇡ is onto.)
For each K 2 A, we have (g f )(K) = g(K/N ) = ⇡ 1 (K/N ). Note that
x 2 ⇡ 1 (K/N ) , ⇡(x) = xN 2 K/N , x 2 K. Hence ⇡ 1 (K/N ) = K.
Since ⇡ : G ! G/N is an onto homomorphism, for K 2 A, K/N C G/N ,
⇡ 1 (K/N ) = K C G. ⇤
Universal mapping property of the quotient group. Let N C G and
let f : G ! H be a homomorphism such that ker f N . Then there is a unique
homomorphism f¯ : G/N ! H such that the following diagram commutes.
f
................................................
G H
... .......
... ........
.. ......
......
⇡ ...... .
......
.
..
.
......... ...... f¯
......
. ......

G/N
Proof. Define f¯(aN ) = f (a) for all aN 2 G/N . ⇤
Three isomorphism theorems.
The first isomorphism theorem. Let f : G ! H be a homomorphism.
Then G/ ker f ⇠
= f (G).
Proof. By the universal mapping property, f : G ! H induces a homomor-
phism f¯ : G/ ker f ! f (G). Clearly, f¯ is onto and ker f¯ = {ker f }. Hence f¯ is an
isomorphism. ⇤
The second isomorphism theorem. Let H < G and N C G. Then
HN/N ⇠
= H/(N \ H).
Proof. The map f : H ! HN/N , h 7! hN is an onto homomorphism with
ker f = N \ H. ⇤
The third isomorphism theorem. Let N C G, H C G and N ⇢ H. Then
(G/N ) (H/N ) ⇠
= G/H.
Proof. The map f : G/N ! G/H, aN 7! aH is a well-defined onto homo-
morphism with ker f = H/N . ⇤
Example. For n 2 Z+ , we have Zn = Z/nZ. If G is a cyclic group, then
G⇠
= Z or Zn .
Proof. Let G = hai. Then f : Z ! G, m ! 7 am is an onto homomorphism.
⇠
If ker f = {0}, then G = Z. If ker f 6= {0}, let n be the smallest positive integer in
ker f . Then ker f = nZ, so G ⇠
= Z/nZ = Zn . ⇤
6 1. GROUPS

If gcd(m, n) = 1, then Zm ⇥ Zn ⇠
= Zmn . The map Zmn ! Zm ⇥ Zn , x + mnZ 7!
(x + mZ, x + nZ) is a homomorphism, which is one-to-one and hence is onto.
Fact. If H, K C G, HK = G, and H \ K = {e}, then G = ⇠ H ⇥ K.

Proof. For all h 2 H, k 2 K, we have hkh 1 k 1 2 H\K, whence hkh 1 k 1 =

e, i.e., hk = kh. Define f : H ⇥ K ! HK, (h, k) 7! hk. Then f is an isomor-
phism. ⇤
Conjugation. For a 2 G, the conjugation by a is the map
a : G ! G
x 7 ! axa 1 ,
which is an automorphism of G. A subgroup H < G is normal if and only if
a (H) = H for all a 2 G; H is called a characteristic subgroup of G if f (H) = H
for all f 2 Aut(G). A characteristic subgroup is normal; the converse is false, e.g.,
Z2 ⇥ {0} is a normal subgroup but not a characteristic subgroup of Z2 ⇥ Z2 .
The map f : G ! Aut(G), a 7! a is a homomorphism with ker f = Z(G); the
image f (G) = { a : a 2 G} is called the inner automorphism group of G and is
denoted by Inn(G). We have Inn(G) ⇠ = G/Z(G). For ↵ 2 Aut(G), ↵ a ↵ 1 = ↵(a) ,
hence Inn(G) C Aut(G).
Fact. For any group G with |G| > 2, |Aut(G)| > 1.
Proof. If G is nonabelian, then |Inn(G)| = |G/Z(G)| > 1. Assume that G is
abelian. If a2 6= e for some a 2 G, then x 7! x 1 is a nonidentity automorphism of
G. If a2 = e for all a 2 G, then G is a vector space over Z2 of dimension > 1. In
this case, Aut(G) = GL(G). Let E be a basis of G and choose distinct ✏1 , ✏2 2 E.
There exists f 2 GL(G) such that f (✏1 ) = ✏2 . ⇤
The commutator subgroup and abelianization. An element of G of the
form [x, y] := x 1 y 1 xy (x, y 2 G) is called a commutator of G. The commutator
subgroup of G is G0 = h{[x, y] : x, y 2 G}i.
Proposition 1.7.
(i) G0 is a characteristic subgroup of G and G/G0 is abelian.
(ii) Let H < G. Then H C G with G/H abelian if and only if H G0 .
Proof. (i) The set {[x, y] : x, y 2 G} is invariant under any automorphism of
G, hence G0 C G. For x, y 2 G, since x 1 y 1 xy 2 G0 , we have xyG0 = yxG0 , i.e.,
(xG0 )(yG0 ) = (yG0 )(xG0 ), whence G/G0 is abelian.
(ii) ()) For x, y 2 G, since G/H is abelian, (xH)(yH) = (yH)(xH), so
1
x y 1 xy 2 H.
(() Since G/G0 is abelian, H/G0 C G/G0 , whence H C G. Moreover, G/H ⇠
=
(G/G0 )/(H/G0 ), which is abelian. ⇤
By Proposition 1.7 (ii), G0 is the smallest normal subgroup H of G such that
G/H is abelian; G/G0 is called the abelianization of G.
Normal subgroups at top and bottom. Let H < G. If H ⇢ Z(G) or
H G0 , then H C G.
Fact. If H < Z(G) is such that G/H is cyclic, then G is abelian.
 1.2. NORMAL SUBGROUPS AND QUOTIENT GROUPS 7

S
Proof. Let G/H = haHi, where a 2 G. Then G = n2Z an H. For any x, y 2
G, we have x = an1 h1 and y = an2 h2 , where n1 , n2 2 Z and h1 , h2 2 H ⇢ Z(G).
Then xy = an1 h1 an2 h2 = an1 +n2 h1 h2 = an2 +n1 h2 h1 = yx. ⇤

If G is abelian, every subgroup of G is normal. The converse is false, as shown

by the next example.
Example (The quaternion group). The quaternion group is Q8 = {±1, ±i, ±j, ±k}
equipped with a multiplication defined by the rules i2 = j 2 = k 2 = 1, ij = k,
jk = i, ki = j, ik = j, kj = i, ji = k; see Figure 1.1. One can show ⇥ 0that
⇤
Q8 is indeed a group
⇥ ⇤ by realizing it as a subgroup of GL(2, C). Let A = 10 ,
1

B = [ 0i 0i ], C = 0i 0i 2 GL(2, C), and G = {±I, ±A, ±B, ±C}. Note that A, B, C

obey the same rules as i, j, k; in particular, G is closed under multiplication and
hence G < GL(2, C). The map f : G ! Q8 , ±I 7! ±1, ±A 7! ±i, ±B 7! ±j,
±C 7! ±k is a bijection such that f (xy) = f (x)f (y) for all x, y 2 G. Hence Q8 is
a group and f : G ! Q8 is an isomorphism. It is easy to see that Z(Q8 ) = {±1}
and that every subgroup of Q8 is normal. It is also easy to see that Z(Q8 ) = Q08 .

......
......
i ............
..... ........
.......
i .......
.....
.... .... .... .....
... ...
... ij = k ... ...
... ik = j
.... ... .... ...
.. ... .. .
......... . jk = i ..... ......... kj = i
. .. . .
j .. k ki = j j .... k ji = k
....... ..... ........... ...
............................... .............................

Figure 1.1. Multiplication in Q8

Hamiltonian groups. A nonabelian group G is called Hamiltonian if all its

subgroups are normal. An elementary abelian p-group is a vector space over Zp for
some prime p.
Theorem 1.8 (Dedekind and Baer). A group G is Hamiltonian if and only if
G⇠ = Q8 ⇥ A ⇥ B, where A is an elementary abelian 2-group and B is an abelian
group whose elements all have odd order.
Lemma 1.9. Let G be a group and let x, y 2 G be such that [x, y] commutes
with x and y. Then
(i) [xi , y j ] = [x, y]ij , i, j 2 Z;
i
(ii) (xy)i = xi y i [y, x](2) , i 2 N.
Proof. (i) For i 2 Z, we have
[x, y][xi , y] = (x 1
y 1
xy)x i y 1 i
x y = x i (x 1
y 1
xy)y 1 i
x y = [xi+1 , y].
Thus [x, y]i = [xi , y] by induction on i. In the same way, for j 2 Z, we have
[xi , y j ][xi , y] = [xi , y j+1 ], and induction j shows that [xi , y j ] = [xi , y]j = [x, y]ij .
(ii) It takes 2i transpositions yx 7! xy to transform the product (xy)i to xi y i ,
each of which requires a multiplication by [x, y]. Hence
i
(xy)i [x, y](2) = xi y i .
Note that [x, y] 1
= [y, x]. ⇤
8 1. GROUPS

Proof of Theorem 1.8. The identity element and the identity subgroup of
a group both will be denoted by 1.
(() Assume that G ⇠= Q8 ⇥ A ⇥ B, where A, B are as stated, and let H < G.
By Exercise ??, H = H1 ⇥ H2 , where H1 < Q8 ⇥ A and H2 < B. It suffices to
show that H1 C Q8 ⇥ A. If H1 contains an element of order 4, then H1 {h2 : h 2
H1 } = Z(Q8 ) ⇥ 1 = (Q8 ⇥ A)0 , so H1 C Q8 ⇥ A. If H1 contains no element of order
4, then H1 ⇢ Z(Q8 ) ⇥ A = Z(Q8 ⇥ A), which also implies that H1 C Q8 ⇥ A.
()) 1 Let x, y 2 G be such that c := [x, y] 6= 1. Since hxi, hyi C G, we
have c 2 hxi \ hyi, i.e., xi = c = y j for some i, j 2 Z \ {0}. By Lemma 1.9 (i),
ci = [xi , y] = [c, y] = 1, hence o(x), o(y) < 1. Assume that x, y are chosen such
that [x, y] 6= 1 and o(x) + o(y) is minimal.
2 Let p be a prime factor of o(x). The minimality of o(x) + o(y) implies
1 = [xp , y] = cp , i.e., o(c) = p. Since p is determined by c, it is the only prime
factor of o(x). Hence o(x) is a power of p, and the same is true for o(y). Write
r s
x↵p = c = y p , ↵, 2 Z, r, s 2 N, p - ↵, . Then o(x) = pr+1 and o(y) = ps+1 .
Let ↵0 , 0 2 Z be such that ↵0 ↵ ⌘ 1 ⌘ 0 (mod p). Then
0 r 0
p ↵0 ↵pr 0 0 0 0
ps 0 s
x =x = c↵ = y↵ = y↵ p ,
0 0 0 0 0 0 0 0
where c↵ = [x , y ↵ ]. Replacing x, y, c by x , y ↵ , c↵ , respectively, we may
r s
assume that xp = c = y p . Note that r, s > 0 since otherwise [x, y] = 1.
r s
3 In 2 , assume that r s. Since x p y does not commute with x, by the
r s r s
minimality of o(x) + o(y), o(x p y) o(y) = ps+1 . By Lemma 1.9 (i), [x p , y]
commutes with x and y. Then by Lemma 1.9 (ii) and (i),
ps s ps
pr
pr
]( 2 ) = [y, x] ( ) = cpr (ps
s s
pr ps pr s
1 6= (x y)p = x y [y, x 2 1)/2
.
Thus p - pr (ps 1)/2 and hence p = 2 and r = 1. Consequently, o(x) = 4,
x2 = y 2 , yxy 1 = x 1 . Let Q = hx, yi. It follows from Theorem 1.29 that Q is a
homomorphic image of Q8 . Since Q is nonabelian we have Q ⇠ = Q8 .
4 We claim that G = QC, where C = {g 2 G : gu = ug for all u 2 Q}. In
a b
fact, for g 2 G we have gxg 1 = x±1 = x( 1) , gyg 1 = y ±1 = y ( 1) , a, b 2 {0, 1}.
Then y x g commutes with x and y, i.e. y x g 2 C.
a b a b

5 We claim that C has no element of order 4. If, to the contrary, o(g) = 4

for some g 2 C, then g 2/ Q. (Otherwise, g 2 Q \ C = Z(Q), which is impossible.)
Then o(gx) = 2 or 4. Since [gx, y] 6= 1, we have ygxy 1 = (gx) 1 , i.e. g = g 1 ,
which is a contradiction.
6 We claim that C is abelian. (Otherwise, C would be a Hamiltonian group
without element of order 4, which contradicts 3 .) For each g 2 C, gx does not
commute with y. By 1 , o(gx) < 1 hence o(g) < 1. Since C is an abelian group
whose elements are of finite order 6= 4, we can write C = AB, where A < C consists
of elements of order 1 or 2 and B < C consists of elements of odd order. Then
G = QC = QAB ⇠ = QA ⇥ B. Since Q \ A < A are vector spaces over Z2 , Q \ A has
a complement, i.e., A = (Q \ A)A0 for some A0 < A with 1 = (Q \ A) \ A0 = Q \ A0 .
Thus QA = QA0 ⇠ = Q ⇥ A0 . In conclusion, G ⇠ = Q ⇥ A0 ⇥ B. ⇤
 1.3. THE SYMMETRIC GROUP 9

1.3. The Symmetric Group

The symmetric group of a set X, denoted by SX , is the group of all permutations
of X. If |X| = n, we may assume that X = {1, . . . , n} and we write SX = Sn ; we
have |Sn | = n!.
Cayley’s theorem. Every group G is isomorphic to a subgroup of SG , i.e., G
is embedded in SG . If |G| = n, G is embedded in Sn .
Proof. Define
f: G ! SG f (a) : G ! G
where
a 7 ! f (a), x 7 ! ax.
Then f is a homomorphism with ker f = {e}. ⇤
Notation. A permutation 2 Sn is denoted by
!
1 ··· n
.
(1) · · · (n)
For distinct i1 , . . . , ik 2 {1, . . . , n}, (i1 , . . . , ik ) 2 Sn is the permutation which maps
i1 to i2 , i2 to i3 , ..., ik 1 to ik , ik to i1 , and leaves other elements fixed; it is called
a k-cycle. A transposition is a 2-cycle; a 1-cycle is the identity of Sn .
Fact.
(i) Disjoint cycles commute with each other.
(ii) (i1 , i2 , . . . , ik ) = (i2 , . . . , ik , i1 ) = · · · .
(iii) (i1 , i2 , . . . , ik ) 1 = (ik , . . . , i2 , i1 ).
(iv) (i1 , . . . , ik ) 1
= ( (i1 ), . . . , (ik )) for 2 Sn .
Theorem 1.10. Every 2 Sn is a product of disjoint cycles and the cycles
are uniquely determined by .
Proof. Existence. We use induction on n. The integers i (1), i 2 N, cannot
be all distinct; there exist 0  i1 < i2 such that i1 (1) = i2 (1), i.e., i2 i1 (1) = 1.
Now let i be the smallest positive integer such that i (1) = 1. Then ⌧ :=
(1, (1), . . . , i 1 (1)) 1 fixes 1, (1), . . . , i 1 (1), hence ⌧ 2 S{1,...,n}\{ (1),..., i 1 (1)} .
By the induction hypothesis, ⌧ is a product of disjoint cycles involving {1, . . . , n} \
{ (1), . . . , i 1 (1)}. Therefore = (1, (1), . . . , i 1 (1))⌧ is a product of disjoint
cycles.
Uniqueness. We also use induction on n. Assume that = (1, i2 , . . . , ik )↵ =
(1, j2 , . . . , jl ) , where ↵ ( , respectively) is a product of disjoint cycles involving
{1, . . . , n} \ {1, i2 , . . . , ik } ({1, . . . , n} \ {1, j2 , . . . , jl }). We claim that k = l. If to
the contrary k < l, then 1 = (1, i2 , . . . , ik )k (1) = k (1) = (1, j2 , . . . , jl )k (1) = jk+1 ,
which is a contradiction. For 2  s  k, is = s 1 (1) = js , hence (1, i2 , . . . , ik ) =
(1, j2 , . . . , jl ) and ↵ = . By the induction hypothesis, the disjoint cycles in ↵ and
are the same. ⇤
Fact. If the disjoint cycles in 2 Sn are of lengths l1 , . . . , lk , then o( ) =
lcm(l1 , . . . , lk ).
Fact. Sn is generated by each of the following subsets.
(i) {(1, 2), (1, 3), . . . , (1, n)}.
(ii) {(1, 2), (2, 3), . . . , (n 1, n)}.
10 1. GROUPS

(iii) {(1, 2), (1, 2, . . . , n)}.

Proof. (i) Note that (i1 , . . . , ik ) = (i1 , i2 )(i2 , i3 ) · · · (ik 1 , ik ), where (i, j) =
(1, i)(1, j)(1, i) if 1, i, j are distinct. Then the conclusion follows from Theorem 1.10.
(ii) We generate the subset in (i) by the identity (1, j) = (j 1, j)(1, j 1)(j
1, j), 2  j  n, and induction on j.
(iii) We generate the subset in (ii) by the identity (i, i + 1) = (1, 2, . . . , n)(i
1, i)(1, 2, . . . , n) 1 and induction on i. ⇤
The sign of a permutation. Let Z[x1 , . . . , xn ] denote the set of polynomials
in x1 , . . . , xn with coefficients in Z. For f (x1 , . . . , xn ) 2 Z[x1 , . . . , xn ] and 2 Sn ,
define (f ) = f (x (1) , . . . , x (n) ). If f, g 2 Z[x1 , . . . , xn ], a, b 2 Z, and , ⌧ 2 Sn ,
then (af + bg) = a (f ) + b (g), (f g) = (f ) (g), and (⌧ (f )) = ( ⌧ )(f ). For
2 Sn , we have
⇣ Y ⌘ Y Y
(xi xj ) = (x (i) x (j) ) = sign( ) (xi xj ),
1i<jn 1i<jn 1i<jn

where sign( ) 2 {±1}. The map sign : Sn ! ({±1}, ·) is a homomorphism (onto

when n 2). To see this claim, consider , ⌧ 2 Sn and compute
Y ⇣ Y ⌘ ⇣ ⇣ Y ⌘⌘
sign( ⌧ ) (xi xj ) = ( ⌧ ) (xi xj ) = ⌧ (xi xj )
1i<jn 1i<jn 1i<jn
⇣ Y ⌘ Y
= sign(⌧ ) (xi xj ) = sign(⌧ )sign( ) (xi xj ).
1i<jn 1i<jn

Hence sign( ⌧ ) = sign( )sign(⌧ ). Clearly, sign(1, 2) = 1; hence sign(i, j) = 1

since (i, j) is conjugate to (1, 2). The kernel of sign, denoted by An , is the alternating
group of degree n.
Fact.
(i) Each 2 Sn is a product of either an even number of transpositions (not
necessarily disjoint) or an odd number of transpositions, but not both;
is called even or odd accordingly. An is the set of even permutations.
(ii) |An | = n!/2 for n 2.
(iii) sign(i1 , . . . , ik ) = ( 1)k 1 .
Proof. (iii) Note that (i1 , . . . , ik ) = (i1 , i2 )(i2 , i3 ) · · · (ik 1 , ik ). ⇤
Proposition 1.11. An is generated by 3-cycles.
Proof. We only have to show that (i, j)(k, l) (i 6= j, k 6= l) is generated
by 3-cycles. If {i, j} = {k, l}, then (i, j)(k, l) = id. If |{i, j} \ {k, l}| = 1, we
may assume that {i, j} = {1, 2}, {k, l} = {1, 3}. Then (1, 2)(1, 3) = (3, 2, 1). If
{i, j} \ {k, l} = ;, then (i, j)(k, l) = ((i, j)(j, k))((j, k)(k, l)) where |{i, j} \ {j, k}| =
1 = |{j, k} \ {k, l}|. ⇤
Let n 3. The dihedral group Dn is the subgroup of Sn generated by ↵ =
(1, . . . , n) and = (2, n)(3, n 1) · · · (dn/2e, bn/2c + 2).
Facts about Dn .
(i) o(↵) = n, o( ) = 2, ↵ 1 = ↵ 1 .
(ii) |Dn | = 2n, Dn = {↵i j : 0  i  n 1, 0  j  1}.
 1.3. THE SYMMETRIC GROUP 11

(iii) (
{id, ↵n/2 } if n is even,
Z(Dn ) =
{id} if n is odd.

Proof. (i) ↵ 1
= (. . . , n 1, n, 1, 2, 3, . . . ) 1
= (. . . , 3, 2, 1, n, n 1, . . . ) =
↵ 1
.
(ii) By (i), G := {↵i j : 0  i  n 1, 0  j  1} is closed under multiplication,
whence Dn = G. We claim that the elements ↵i j , 0  i  n 1, 0  j  1, are all
distinct. Assume that ↵i1 j1 = ↵i2 j2 for some 0  i1 , i2  n 1, 0  j1 , j2  1.
Then ↵i1 i2 = j2 j1 , hence 1 = j2 j1 (1) = ↵i1 i2 (1) ⌘ 1+i1 i2 (mod n). Thus
i1 = i2 , and hence j1 = j2 . ⇤
Let Xn be a regular n-gon in R2 with vertices labeled by 1, 2, . . . , n (Figure 1.2).
A symmetry of Xn is a rigid motion in R3 that maps Xn to itself. A symmetry of
Xn can be treated as a permutation of the vertices of Xn .
..
.
...........
1
..
...
........... ........
........
.....
...... .
. ........
.. ........
........ .. ........
7 ................ .. ........ 2
...
.. .. ...
... ...
... ..... ...
... ...
... .. ...
... .. ...
... ...
... ..... ...
...
... ...
... .. ...
6 .... .. ... 3
...
... .
. . . .. .
... .
.. .. .
... . .
... ..
... .. ...
... .. ...
... ...
...
... ..... ....
.
... ...
............................................................
.
5 ... 4
...
.

Figure 1.2. Regular heptagon

Fact. Dn is the group of symmetries of Xn .

Proof. Let 2 Sn be a symmetry of Xn with (1) = i. Then ↵ i is a
symmetry of Xn which fixes 1. Thus ↵ i is either the identity or the reflection ,
i.e. ↵ i = j , j = 0 or 1. Therefore = ↵i j 2 Dn . ⇤
Definition 1.12. A group G 6= {e} is called simple if the only normal sub-
groups of G are {e} and G.
Fact. An abelian group G is simple if and only if G ⇠
= Zp where p is a prime.
Simplicity of An (n 5). If n 5, An is simple.
Proof. 1 We claim that all 3-cycles in An are conjugate in An . Let i1 , . . . , i5 2
{1, . . . , n} be distinct. Let
! !
1 2 3 4 5 ··· 1 2 3 4 5 ···
↵= or
i1 i2 i3 i4 i5 · · · i1 i2 i3 i5 i4 · · ·
be such that ↵ 2 An . Then ↵(1, 2, 3)↵ 1
= (i1 , i2 , i3 ).
2 Let {id} 6= N C An . To show that N = An , it suffices to show that N
contains a 3-cycle. (By 1 and the normality of N , N contains all 3-cycles. By
12 1. GROUPS

Proposition 1.11, N = An .) Let id 6= 2 N be written as a product of disjoint

cycles.
Case 1. Assume that has a cycle of length r 4, say, = (1, 2, . . . , r)↵. Let
= (1, 2, 3) 2 An . Then N 3 1
( 1
) = (r, . . . , 2, 1)(2, 3, 1, 4, . . . , r) = (1, 3, r).
Case 2. Assume that has two 3-cycles, say, = (1, 2, 3)(4, 5, 6)↵. Let
= (1, 2, 4) 2 An . Then N 3 1
( 1
) = (3, 2, 1)(6, 5, 4)(2, 4, 3)(1, 5, 6) =
(1, 4, 2, 6, 3), and we are in Case 1.
Case 3. Assume that one cycle of has length 3 and all others have length
 2. Then 2 is a 3-cycle.
Case 4. Assume that all cycles of have length  2, say, = (1, 2)(3, 4)↵. Let
= (1, 2, 3) 2 An . Then N 3 ( 1
) = (1, 2)(3, 4)(2, 3)(1, 4) = (1, 3)(2, 4) =: ⌧ .
Let ✏ = (1, 3, 5). Then N 3 ⌧ (✏⌧ ✏ ) = (1, 3)(2, 4)(3, 5)(2, 4) = (1, 3, 5).
1
⇤

Remark. A3 is also simple, but A4 is not. We have V := {id, (1, 2)(3, 4), (1, 3)(2, 4),
(1, 4)(2, 3)} C S4 since V is closed under multiplication and is invariant under con-
jugation in S4
Fact 1.13. The normal subgroups of Sn are {id}, An and Sn , and, in addition,
V for n = 4.
Proof. The claim is easily verified for n  4. For n 5, assume that H C Sn
and H 6 An . Then H \ An C An and H \ An 6= An , hence |H \ An | = 1. Thus
|H| = |HAn |/|An |  |Sn |/|An | = 2. The normality of H in Sn forces H = {id}. ⇤

1.4. Group Actions

Let X be a set and be a permutation of X. When applying to x 2 X, we
can write to the left or the right of x, i.e., (x) or (x) . In the first notation, the
rule of composition is (⌧ )(x) = ⌧ ( (x)), and the symmetric group on X is denoted
by SX ; in the second notation, the rule of composition is (x)(⌧ ) = ((x)⌧ ) , and
the symmetric group on X is denoted by X S. The map 7! 1
is an isomorphism
between SX and X S.
Definition 1.14. Let G be a group and X be a set. A left (right) action of G
on X is a homomorphism : G ! SX (X S). Usually, for g 2 G, the image (g)
is still denoted by g. So, for a left (right) action, gx = (g)(x) (xg = (x) (g)).
Equivalently, a left action on G on X is a map
G⇥X ! X
(g, x) 7 ! gx
such that ex = x and g1 (g2 x) = (g1 g2 )x for all x 2 X and g1 , g2 2 G. A right
action of g on X is a map X ⇥ G ! X with similar properties. In this book, an
action is treated as a left action if the side is not mentioned.
Example. Let V be a vector space over a field F , and let HomF (V, F ) be the
dual space of V (the vector space of linear maps from V to F ). Then
GL(V ) ⇥ V ! V
(f, x) 7 ! f (x)
 1.4. GROUP ACTIONS 13

is a left action of GL(V ) on V , and

HomF (V, F ) ⇥ GL(V ) ! HomF (V, F )
(l, f ) 7 ! l f
is a right action of GL(V ) on HomF (V, F ).
Definition 1.15. Let G act on X. For x 2 X, [x] := {gx : g 2 G} is the
G-orbit of x and Gx := {g 2 G : gx = x} < G is the stabilizer of x. The G-orbits
form a partition of X. If X has only one G-orbit, we say that G acts transitively
on X.
Fact. Let G act on X.
(i) For x 2 X, [G : Gx ] = |[x]|.
(ii) For g 2 G and x 2 X, Ggx = gGx g 1
.
Proof. (i) Define
f: G/Gx ! [x]
gGx 7 ! gx.
Then f is a well-defined bijection. ⇤
The class equation. Let G act on G by conjugation: G ⇥ G ! G, (a, x) 7!
axa 1 . In this case, the orbit [x] = {axa 1 : a 2 G} is the conjugacy class of x
in G, and the stabilizer Gx = {a 2 G : axa 1 = x} =: CG (x) is the centralizer
of x in G. Note that |[x]| = 1 if and only if x 2 Z(G). Let {xi : i 2 I} be a set
of representatives of the conjugacy classes of G not contained in Z(G). Then [x],
x 2 Z(G), and [xi ], i 2 I, enumerate all the conjugacy classes of G. Hence
X X X
(1.1) |G| = |[x]| + |[xi ]| = |Z(G)| + [G : CG (xi )].
x2Z(G) i2I i2I

Equation (1.1) is called the class equation of G.

Normalizer. Let G be a group and let S(G) be the set of all subgroups of
G. Let G act on S(G) by conjugation: G ⇥ S(G) ! S(G), (a, H) 7! aHa 1 . For
H < G, the stabilizer GH = {a 2 G : aHa 1 = H} =: NG (H) is the normalizer of
H in G. The normalizer NG (H) is the largest subgroup K of G such that H C K.
The index [G : NG (H)] is the number of conjugates of H in G.
p-group actions. Let p be a prime. A group G is called a p-group if for every
a 2 G, o(a) is a power of p. A finite group G is a p-group if and only if |G| is a
power of p; this follows from Sylow’s theorem (§1.5).
Proposition 1.16. Let G be a finite p-group acting on a finite set X, and let
X0 = {x 2 X : gx = x for all g 2 G}. Then |X| ⌘ |X0 | (mod p).
Proof. For x 2 X, [x] = {x} if and only if x 2 X0 . Enumerate the G-orbits
of X as [x1 ], . . . , [xn ], where {x1 , . . . , xk } = X0 . Then for each k < i  n, |[xi ]| is
a power of p and |[xi ]| > 1. Hence
n
X
|X| = |[xi ]| = k + |[xk+1 ]| + · · · + |[xn ]| ⌘ k (mod p).
i=1
⇤
14 1. GROUPS

Table 1.1. Examples of Proposition 1.16

G X action X0 conclusion in particular

N,
G conjugation N \ Z(G) |N \ Z(G)| ⌘ 0 (mod p) |N \ Z(G)| > 1
{e} 6= N C G
left [NG (H) : H] ⌘ [G : H] if p | [G : H],
H G/H, H < G NG (H)/H
multiplication (mod p) NG (H) 6= H

Proposition 1.17 (Facts about finite p-groups). Let p be a prime and G be a

nontrivial finite p-group.
(i) Z(G) 6= {e}.
(ii) If H < G with [G : H] = p, then H C G.
(iii) If |G| = p2 , then G is abelian.
Proof. (i) and (ii) are special cases of Table 1.1.
(iii) We have |Z(G)| = p or p2 , i.e., |G/Z(G)| = p or 1. Thus G/Z(G) is cyclic,
and hence G is abelian. ⇤

Recall that the notation H ,! G means that H is embedded in G, i.e., there is

a one-to-one homomorphism from H to G.
Proposition 1.18. Let G be a group. If there exists H < G such that [G :
H] = n < 1, then there exists N C G such that N ⇢ H and [G : N ] | n!.
Proof. Let G act on G/H: (a, gH) 7! agH. Let N be the kernel of this
action. Then N C G, N ⇢ H, and G/N ,! SG/H = Sn . ⇤

Example 1.19. Let G be a simple group with |G| > 2. If there exists H < G
such that [G : H] = n > 1, then G ,! An .
Proof. In the proof of Proposition 1.18, N = {e} and G = G/N ,! SG/H =
Sn . We may assume that G ⇢ Sn . We claim that G ⇢ An . If G 6⇢ An , then
[G : G \ An ] = 2, and hence G \ An C G, which is a contradiction. ⇤

Theorem 1.20 (Burnside’s lemma). Let G be a finite group acting on a finite

set X. Then the number of G-orbits in X is
1 X
n= Fix(g).
|G|
g2G

where Fix(g) = |{x 2 X : gx = x}|.

Proof. Let X = {(g, x)P : g 2 G, x 2 X, gx = x}. Counting (g, x) 2 X in the
order of g, x, we have |X | = g2G Fix(g). Counting (g, x) 2 X in the order of x, g,
P
we also have |X | = x2X |Gx |. Let [x1 ], . . . , [xn ] be the G-orbits in X. Then
X n X
X Xn X Xn X Xn
|G| |G|
|Gx | = |Gx | = = = |G| = n|G|.
i=1 x2[xi ] i=1
|[x]| i=1
|[xi ]| i=1
x2X x2[xi ] x2[xi ]
P
Therefore n|G| = g2G Fix(g). ⇤
 1.5. SYLOW’S THEOREM 15

1.5. Sylow’s Theorem

Theorem 1.21 (Sylow). Let G be a finite group with |G| = pe m, where p is a
prime, e > 0, and p - m.
(i) Every p-subgroup (including {e}) of G is contained in a subgroup of order
pe . (A subgroup of G of order pe is called a Sylow p-subgroup.)
(ii) All Sylow p-subgroups are conjugate in G.
(iii) Let np denote the number of Sylow p-subgroups of G. Then np | m and
np ⌘ 1 (mod p).
e
Proof. 1 Let X = {X ⇢ G : |X| = pe }. Then |X | = ppem ⌘ m 6⌘ 0
(mod p). Let G act on X : G ⇥ X ! X , (g, X) 7! gX. There exists X 2 X such
that |[X]| 6⌘ 0 (mod p), i.e., |G|/|GX | 6⌘ 0 (mod p), so pe |GX |. Choose x0 2 X.
Then |GX | = |GX x0 |  |GX X| = |X| = pe . Hence P := GX is a Sylow p-subgroup
of G.
2 Let P = {gP g 1 : g 2 G}, and let P act on P by conjugation. Then {P }
is the only orbit with one element. (If P1 2 P is such that P1 6= P and {P1 } is
an orbit, then P P1 is a p-subgroup of G with |P P1 | > pe , which is a contraction.)
Therefore |P| ⌘ 1 (mod p) by Proposition 1.16. Since |P| = [G : NG (P )] |G|, we
have |P| m.
3 Let P1 be any p-subgroup of G, and let P1 act on P by conjugation. Since
|P| ⌘ 1 (mod p), there exists P2 2 P such that {P2 } is a P1 -orbit. Then P1 ⇢ P2
since otherwise P1 P2 would be a p-subgroup of G with |P1 P2 | > pe . This also shows
that P is the set of all Sylow p-subgroups of G, i.e., np = |P|. ⇤

Proposition 1.22. Let P be a Sylow p-subgroup of a finite group G. Then

NG (NG (P )) = NG (P ).
Proof. Let x 2 NG (NG (P )). We have xP x 1 ⇢ xNG (P )x 1 = NG (P ). Since
both xP x 1 and P are Sylow p-subgroups of NG (P ), by Sylow’s theorem, there
exists y 2 NG (P ) such that xP x 1 = yP y 1 = P , hence x 2 NG (P ). ⇤

Proposition 1.23. Let H C G, |H| < 1, and let P be a Sylow p-subgroup of

H. Then G = HNG (P ).
Proof. Let x 2 G. Then P and xP x 1 are conjugate in H, i.e., xP x 1
=
yP y 1 for some y 2 H. Thus y 1 x 2 NG (P ) and x = y(y 1 x) 2 HNG (P ). ⇤

The method in the above two proofs is called the Frattini argument.
Semidirect product. Let N and H be groups and let ✓ : H ! Aut(N ) be a
homomorphism. Then H acts on N as follows: for h 2 H and n 2 N , hn = ✓(h)(n),
i.e., each h 2 H acts on H as an automorphism ✓(h) of N . The semidirect product
N o✓ H is a group (N ⇥ H, ·), where
(n1 , h1 ) · (n2 , h2 ) = (n1 h1 n2 , h1 h2 ).
The identity of N o✓ H is (eN , eH ); the inverse of (n, h) 2 N o✓ H is (h 1 n 1 , h 1 ).
Let G = N o✓ H, N 0 = N ⇥ {eH } and H 0 = {eN } ⇥ H. Then N 0 ⇠ = N , H0 ⇠ = H,
N C G, H < G, N \ H = {e}, and G = N H . We write N o✓ H as N o H
0 0 0 0 0 0

when the meaning of ✓ is implied by the context.

16 1. GROUPS

On the other hand, if G is a group such that G = N H, where N C G, H < G,

N \ H = {e}, we have a homomorphism ✓ : H ! Aut(N ), h 7! h( )h 1 and an
isomorphism N o✓ H ! G, (n, h) 7! nh.
Let q be a prime and Z⇥ q = (Zq \ {0}, ·). The map Aut(Zq ) ! Zq ,
⇥
7! (1)
is an isomorphism. For the proof of Proposition 1.24 (ii), we will use the fact that
Z⇥q is cyclic (Proposition 3.43).

Proposition 1.24. Let G be a group of order pq, where p < q are primes.
(i) If p - q 1, then G ⇠
= Zpq .
(ii) If p | q 1, then G ⇠
= Zpq or G ⇠
= Zq o✓ Zp , where ✓ : Zp ! Aut(Zq ) is
any nontrivial homomorphism.
Proof. Let P be a Sylow p-subgroup and Q a Sylow q-subgroup of G. By
Sylow’s theorem, Q C G.
(i) Since p - q 1, we also have P C G. Thus G = P Q ⇠
= P ⇥Q ⇠
= Zp ⇥Zq ⇠
= Zpq .
(ii) We have G ⇠= QoP ⇠ = Zq o✓ Zp for some homomorphism ✓ : Zp ! Aut(Zq ).
If ✓ is the trivial homomorphism, then Zq o✓ Zp = Zq ⇥ Zp ⇠ = Zpq . It remans
to show that if ✓1 , ✓2 : Zp ! Aut(Zq ) are two nontrivial homomorphisms, then
Z q o ✓1 Z p ⇠
= Zq o✓2 Zp . Since both ✓1 (Zp ) and ✓2 (Zp ) are subgroups of Aut(Zq ) of
order p and since Aut(Zq ) ⇠ = Z⇥q is cyclic, we have ✓1 (Zp ) = ✓2 (Zp ). Let ↵ : Zp ! Zp
be the isomorphism such that ✓1 = ✓2 ↵.

✓ 1
Zp .....................................
⇠ ✓1 (Zp )
...
= ...
... ...
... ...
↵.........
...
.
.........
id
...
.
.. ..
✓2
Zp .....................................
⇠ ✓2 (Zp )
=
Then it is easy to check that
Z q o ✓1 Z p ! Z q o ✓2 Z p
(x, y) 7 ! (x, ↵(y))
is an isomorphism. ⇤
Example 1.25. If |G| = pqr, where p < q < r are primes, then the Sylow
r-subgroup of G is normal.
Proof. Let Q, R < G be such that |Q| = q and |R| = r. Assume to the
contrary that R 6C G. Then nr = pq, i.e., NG (R) = R. If Q C G, then R C QR,
and hence NG (R) QR, which is a contradiction. If Q 6C G, we have nq = r or pr.
Then G has pq(r 1) elements of order r and at least r(q 1) elements of order q.
Note that pq(r 1) + r(q 1) > pqr, which is a contradiction. ⇤
Example. If G is a simple group of order 60, then G ⇠
= A5 .
Proof. 1 Let P be the set of all Sylow 5-subgroups of G. Then |P| = 6.
Let G act on P by conjugation. Then G ,! SP = S6 , so we assume that G < S6 .
In fact, G ⇢ A6 , since otherwise, [G : G \ A6 ] = 2 and G \ A6 C G, which is a
contradiction.
2 Let X = (A6 /G) \ {G} and note that T |X | = 5.1 Let G act on X by left
multiplication. The kernel of this action is 2A6 G , which is not equal to G.
 1.6. FINITELY GENERATED ABELIAN GROUPS 17

(Otherwise, G C A6 , which contradicts the simplicity of A6 .) Thus G ,! SX = S5

and by the same argument as in 1 , G ,! A5 . ⇤

Proposition 1.26. A group G of order pn q, where p and q are distinct primes

and n is a positive integer, is not simple.
Proof. Assume to the contrary that G is simple. Let Sylp (G) denote the set
of Sylow p-subgroups of G. Then |Sylp (G)| = q. Let P1 , P2 2 Sylp (G) be such that
|P1 \ P2 | = max{|P \ P 0 | : P, P 0 2 Sylp (G), P 6= P 0 },
and let A = P1 \ P2 .
We claim that NG (A) is not a p-group. Assume the contrary. Then there exists
P 2 Sylp (G) such that NG (A) ⇢ P . Thus A ⇢ NP1 (A) ⇢ P1 \ P . Since A ( P1 ,
we have A ( NP1 (A) (Table 1.1). By the maximality of |P1 \ P2 |, we must have
P = P1 . In the same way, P = P2 , which is a contradiction.
Therefore, NG (A) contains a Sylow q-subgroup Q of G. For all a 2 Q, A =
aAa 1 ⇢ aP1 a 1 . Since G = QP1 , Q acts transitively on Sylp (G) by conjugation.
Thus \
A⇢ P.
P 2Sylp (G)
T T
Since P 2Sylp (G) P C G, we have P 2Sylp (G) P = {e}. Thus we have proved that
P1 \ P2 = {e} for all P1 , P2 2 Sylp (G) with P1 6= P2 . Then G has q(pn 1)
elements of order pi , i > 0. The remaining pn q q(pn 1) = q elements make up
at most one Sylow q-subgroup. Thus the Sylow q-subgroup of G is normal, which
is a contradiction. ⇤

1.6. Finitely Generated Abelian Groups

The direct sum of a family of abelian groups Ai (i 2 I) is
M
Ai = (ai )i2I : ai 2 Ai for all i 2 I and ai = 0 for all but finitely many i ,
i2I
Q
which is a subgroup
L of the direct
Q product i2I Ai = {(ai )i2I : ai 2 Ai for all i 2 I}.
When |I| < 1, i2I Ai = i2I Ai .
Theorem 1.27 (Structure of finitely generated abelian groups). Every finitely
generated abelian group G is isomorphic to
Zpe11 ··· Zpess Zr ,
where p1 , . . . , ps are primes (not necessarily distinct), ei > 0, r 0. The numbers
pe11 , . . . , pess (the elementary divisors of G) and r (the rank of G) are uniquely
determined by G.
Proof. This is a special case of Theorem 2.57, the structure theorem of finitely
generated modules over a PID. ⇤

Remark 1.28.
(i) The number of nonisomorphic abelian groups of order pe11 · · · pess is
P (e1 ) · · · P (es ), where P (e) is the number of partitions of e.
18 1. GROUPS

(ii) Assume that A, B, C are finitely generated abelian groups. Then A A ⇠ =

B B implies A ⇠ = B; A C ⇠ = B C implies A ⇠ = B. (Use the
structureL
theorem.) These are false if A, B, C are not finitely generated.
Let A = n2N Z and B = Z. Then A A ⇠ = B A but A ⇠ 6 B. Corner’s
=
theorem (Theorem 2.66) shows that there is an abelian group A such that
A A A⇠ = A but A A ⇠ 6 A. Let B = A A. Then B B ⇠
= =A A
but B =⇠
6 A.

1.7. Free Groups and Presentations

The free group on X. Let X be a set and let
W = xe11 · · · xess : s 0, xi 2 X, ei 2 {±1} .
Here, xe11 is a formal product and is called a word of length s in X. Two
· · · xess
words u, v 2 W are called equivalent, denoted as u ⇠ v, if v can be obtained from u
by deleting and inserting strings of the form xx 1 and x 1 x, x 2 X. Note that ⇠
is an equivalence relation on W; the equivalence class of u 2 W is denoted by [u].
Let FX = W/ ⇠ be the set of ⇠ equivalence classes of W, and for [u], [v] 2 W/ ⇠,
define [u] · [v] = [uv], where uv is the concatenation of u and v. Then (FX , ·) is a
group, which is called the free group on X; the rank of FX is rank FX = |X|.
Let G be any group and f : X ! G be any map. Then there is a unique
homomorphism f¯ : FX ! G such that the diagram
FX
......
........ ......
...... ¯
... ...... f
◆ ..... ......
......
......
..... ........
. ......

...............................................
X f
G

commutes, where ◆(x) = [x], x 2 X. Every group is a homomorphic image of a free

group.
A word w 2 W is called reduced if it contains no strings xx 1 or x 1 x, x 2 X.
Fact. Every element of W/ ⇠ contains a unique reduced word.
Proof. Uniqueness. Let R be the set of all reduced words in W. Define
f: X ! SR
x 7 ! f (x),
where
f (x) : R ! R
(
xxe11 · · · xess if xe11 =
6 x 1
,
xe11 · · · xess 7 !
xe22 · · · xess if xe11 = x 1
.
To see that f (x) is indeed a permutation of R, note that it has an inverse
1
f (x) : R ! R
(
x 1 xe11 · · · xess if xe11 =
6 x,
xe11 · · · xess 7 !
xe22 · · · xess e1
if x1 = x.
There exists homomorphism f¯ : FX ! SR such that f¯([x]) = f (x) for all x 2 X.
Assume that xe11 · · · xess , y1f1 · · · ytft 2 R are such that [xe11 · · · xess ] = [y1f1 · · · ytft ].
 1.7. FREE GROUPS AND PRESENTATIONS 19

Then xe11 · · · xess = f (x1 )e1 · · · f (xs )es (1) = f¯([xe11 · · · xess ])(1) = f¯([y1f1 · · · ytft ])(1) =
f (y1 )f1 · · · f (yt )ft (1) = y1f1 · · · ytft . ⇤
By the above fact, ◆ : X ! FX is one-to-one, and hence X can be regarded as
a subset of FX .
Presentation. Let X be a set and R a set of words in X. Define
hX | Ri = FX /NR ,
where NR is the normal closure of {[r] : r 2 R} in FX , i.e., the smallest normal
subgroup of FX containing {[r] : r 2 R}. The group hX | Ri is called a presentation;
the elements in X are called generators and the elements in R are called relations.
If G ⇠
= hX | Ri, hX | Ri is called a presentation of G and we usually write G =
hX | Ri.
Theorem 1.29. Let G be a group generated by X 0 ⇢ G. Let hX | Ri be a
presentation such that there is an onto map X ! X 0 , x 7! x0 having the property
that for every r 2 R, the corresponding product in G is identity. Then there is a
unique onto homomorphism hX | Ri ! G such that [x]NR 7! x0 for all x 2 X.
Proof. The map f : X ! G, x 7! x0 induces a homomorphism f¯ : FX ! G
such that [x] 7! x0 . Since G = hX 0 i, f¯ is onto. By assumption, {[r] : r 2 R} ⇢ ker f¯,
whence NR ⇢ ker f¯. Thus f¯ induces an onto homomorphism f˜ from FX /NR = hX |
Ri to G such that [x]NR 7! x0 . The uniqueness of f˜ is obvious since hX | Ri is
generated by {[x]NR : x 2 X}. ⇤
In a presentation hX | Ri = FX /NR , an element [u]NR , where u is word in X,
is usually denoted by u. A relation uv 1 , where u, v are words in X, is denoted as
u = v.
Fact. FX1 ⇠= FX2 if and only if |X1 | = |X2 |.
Proof. ()) For i = 1, 2, let Ni be the normal closure of {a2 : a 2 FXi } in FXi .
From FX1 ⇠ = FX2 , we have FX1 /N1 ⇠ = FX2 /N2 . Since u2 = 1 for all u 2 FXi /Ni , it is
easy to see that FXi /Ni is a vector space over Z2 . We claim that dimZ2 (FXi /Ni ) =
|Xi |. (Then |X1 | = dimZ2 (FX1 /N1 ) = dimZ2 (FX2 /N2 ) = |X2 |.) Let V be a vector
space with basis X1 . The map X1 ! V , x 7! x induces a homomorphism f : FX1 !
V . Clearly, N1 ⇢ ker f ; hence f induces a homomorphism f¯ : FX1 /N1 ! V . On
the other hand, since X1 is a basis of V , there is a homomorphism g : V ! FX1 /N1
such that g(x) = xN1 for all x 2 X1 . Clearly, f¯ and g are inverses of each other,
so FX1 /N1 ⇠= V , whence dimZ2 (FX1 /N1 ) = dimZ2 V = |X1 |. ⇤
Fact. Let X = A t B, and let H be the subgroup of FX generated by A and
N be the normal closure of B in FX . Then FX = HN and H \ N = {1}. In
particular, F/N ⇠
=H⇠= FA .
Proof. We only have to show that H \ N = {1}. Define
: X ! F
(A
x if x 2 A,
x 7 !
1 if x 2 B.
Then induces a homomorphism ¯ : FX ! FA such that N ⇢ ker ¯ and ¯|H : H !
FA is an isomorphism. If w 2 H \ N , then ¯|H (w) = ¯(w) = 1, hence w = 1. ⇤
20 1. GROUPS

Examples.
(i) The dihedral group Dn ⇠
= h↵, | ↵n = 2
= 1, ↵ 1
=↵ 1
i.
(ii) The infinite dihedral group D1 = h↵, | = 1, 2
↵ 1
=↵ 1
i.
(iii) The quaternion group Q8 = ⇠ hx, y | x = y , yxy
2 2 1
=x 1
i. (Note: In the
presentation, the relations imply that x4 = 1.)
Proof. Q8 = hi, ji, where i, j satisfy the relations i2 = j 2 , jij 1
=i 1
.
By Theorem 1.29, there is an onto homomorphism
(1.2) G := hx, y | x2 = y 2 , yxy 1
=x 1
i ! Q8
such that x 7! i, y 7! j. Every element in G can be written as xi y j ,
0  i  3, 0  j  1, whence |G|  8. Hence the homomorphism in (1.2)
is an isomorphism. ⇤
(iv) The generalized quaternion group Q4n = hx, y | xn = y 2 , yxy 1 = x 1 i.
The relations in the presentation imply that x2n = 1. Let ⇠ = e⇡i/n , A =
⇥⇠ ⇤ ⇥ ⇤
⇠ 1
, B = 1 1 . There is an isomorphism Q4n ! hA, Bi ⇢ GL(2, C)
such that x 7! A, y 7! B. We have |Q4n | = 4n.
(v) The braid group on n strands is
Bn = h 1, . . . , n 1 | i i+1 i = i+1 i i+1 , 1  i  n 2; i j = j i, |i j| > 1i.
Each generator i is an n-strand braid with a single crossing. The product
of two braids is obtained by connecting their corresponding strands. (See
Figure 1.3.)
.... .. ... ... .... ..
.... ...... ... ... .... ......
..... ..... ... .. ..... .....
..... ... .....
1 ...
.. .. . ... ..
. .. . 2
1 2 i i+1 n 1 2 i i+1 n ..... ..... ... ... ...... ......
. .
....... ........ ........ .......... .......
. .
. ....... ......... .......... .......... .......
.... .... .... .... .... .... .... .... .... .... .... .... .. .. . ... . ..
... ... ... ... ... ... ... ... ... ... ... ... ... ..... ......
......... ..... ....... ...
........
..
...
..
...
..
...
..... ......
...
..
..
...
..
...
..
...
..
... ....
..
..... ......
...
..
..
...
..
... 2 ...
... ..
.
.... ....
.......... = .... .... ...........
.
...
... 1
... ... ... ... ... ... ........ ... ... .. ..
... ... ........ ... ... ... ... . ... ... ....... .......... .......... .......... ....... ....... .......... .......... .......... .......
... ... .............. ... ... ... ... .. ...........
. ... ... .... .. ... ... .... ..
... ... .... ...
... ... ... ... .... ...
... ... ..... ...... ..... ......
.. .. ....... ... ... .......
... ... .... ... ... ... ... ... .... ... ... ... 1 .... ........ .... ... .... ........ 2
... ... ... ... ... ... ... ... ... ... ... ... .. . .
. .. .. ... .. ... ..
... ... .... .... ... ... ... ... .... .... ... ... ... ... .... .... ... ...
. . . . . . . .
1 =
i i 1 2 1 2 1 2

Figure 1.3. The braid group

Theorem 1.30 (The Nielson-Schreier theorem). Let F be a free group on X

and G < F . Then G is a free group. If [F : G] = m < 1, then rank G =
m|X| + 1 m. where |X| may be infinite.
Proof. For A ⇢ F , define A 1
= {a 1
: a 2 A}.
1 Let R be the set of all right cosets of G in F . For each C 2 R, let |C|
denote the minimum length of reduced words in C. We choose a representative
tC of length |C| for each C 2 R inductively. First choose tG = 1. Let D 2 R be
such that tC have been chosen for all C 2 R with |C| < |D|. Choose uy 2 D,
reduced, such that y 2 X [ X 1 and uy is of length |D|. Then |Gu| = |D| 1,
hence tGu has been chosen, and D = Guy = GtGu y. We choose tD = tGu y. Let
T = {tC : C 2 R}. Then T is a system of representatives of R with the property
 1.7. FREE GROUPS AND PRESENTATIONS 21

that if t 2 T is reduced, its initial segments also belong to T ; we say that T has
the Schreier property.
2 For each w 2 F , let w̄ 2 T be such that Gw = Gw̄, and define (w) =
ww̄ 1
2 G. Let Y = { (tx) : t 2 T, x 2 X, tx 2
/ T }. For all w, u 2 F , we have
1
(1.3) (w) (w̄u) = ww̄ 1
w̄uw̄u = wuwu 1
= (wu).
Letting u = x 2 X in (1.3) gives (wx) = (w) (w̄x), where (w̄x) = 1 if w̄x 2 T
and (w̄x) 2 Y if w̄x 2 / T . Hence (wx)hY i = (w)hY i. This implies that
(w) 2 hY i for all w 2 F . In particular, for g 2 G, g = (g) 2 hY i. Therefore
G = hY i.
For t 2 T and w 2 F , by (1.3), (tw) (tww 1 ) = (t) = 1, i.e., (tw) 1 =
(tww 1 ). Note that each t 2 T can be written as t = t1 w, where t1 = tw 1 2 T .
It follows that Y 1 = { (tx 1 ) : t 2 T, x 2 X, tx 1 2
/ T }.
3 Assume that t1 , t2 2 T and y1 , y2 2 X [ X 1 are such that ti yi 2 / T . Let
si = ti yi . We claim that in reduced form, s1 1 and s1 1 t2 do not start with y1 1 .
If, to the contrary, s1 1 = y1 1 u in reduced form, then s1 = u 1 y1 in reduced form.
Since T has the Schreier property, u 1 2 T . Since Gt1 y1 = Gs1 = Gu 1 y1 , we
have Gt1 = Gu 1 , whence t1 = u 1 . Then t1 y1 = s1 2 T , which is a contradiction.
If, to the contrary, s1 1 t2 = y1 1 v in reduced form, then t2 = s1 y1 1 v. Since s1
(in reduced form) does not end with y1 and v does not start with y1 , s1 y1 1 (in
reduced form) is an initial segment of t2 (in reduced form). Thus s1 y1 1 2 T . Since
Gs1 y1 1 = Gt1 y1 y1 1 = Gt1 , we have s1 y1 1 = t1 . Again t1 y1 = s1 2 T , which is a
contradiction.
We also claim that in reduced form, t2 and s1 1 t2 do not end with y2 1 . If,
to the contrary, t2 = uy2 1 in reduced form, then u 2 T . Thus t2 y2 = u 2 T ,
which is a contradiction. If, to the contrary, s1 1 t2 = vy2 1 in reduced form, then
s1 = t2 y2 v 1 . Since t2 (in reduced form) does not end with y2 1 and v 1 does not
start with y2 1 , t2 y2 (in reduced form) is an initial segment of s1 (in reduced form).
Thus t2 y2 2 T , which is a contradiction.
4 Let t1 , . . . , tn 2 T and y1 , . . . , yn 2 X [ X 1 be such that ti yi 2
/ T and
1
(ti , yi ) 6= (si+1 , yi+1 ), 1  i < n, where si = ti yi (1  i  n). We have
(t1 y1 ) (t2 y2 ) · · · (tn yn ) = t1 y1 s1 1 t2 y2 s2 1 · · · tn yn sn 1 .
By 3 , the reduced form of the above product is
⇤ y1 ⇤ y2 ⇤ · · · ⇤ yn ⇤ .
In particular,
(1.4) (t1 y1 ) (t2 y2 ) · · · (tn yn ) 6= 1.

5 Assume that (t1 , y1 ) 6= (t2 , y2 ), where ti 2 T , yi 2 X [ X 1 and ti yi 2 / T.

Let si = ti yi . Note that (t1 y1 ) 1 = (s1 y1 1 ), where (s1 , y1 1 ) 6= (s2 , y2 1 ). By
(1.4), (s1 y1 1 ) (t2 y2 ) 6= 1, i.e., (t1 y1 ) 6= (t2 y2 ). Therefore, the map {(t, y) 2
T ⇥ (X [ X 1 ) : ty 2 / T } ! Y [ Y 1 , (t, y) 7! (ty) is a bijection. Moreover, (1.4)
implies that G = hY i is a free group on Y .
6 We have
(1.5) rank G = |Y | = |(T ⇥ X) \ C|,
22 1. GROUPS

where C = {(t, x) 2 T ⇥ X : tx 2 T }. The map

C ! ( \ {1}
T
t if t ends with x 1 ,
(t, x) 7 !
tx if t does not end with x 1

is a bijection with inverse

T \ {1} ! C
(
(t, x) if t ends with x 1 (x 2 X),
t 7 !
(tx 1 , x) if t ends with x 2 X.
Hence |C| = |T | 1. When |T | < 1, by (1.5), rank G = |T ||X| + 1 |T |. ⇤
The Burnside problems.
The general Burnside problem. If G is a finitely generated group and all
elements of G are of finite order, is G necessarily finite?
The Burnside problem. The free Burnside group of rank m and exponent n,
denoted by B(m, n), is the presentation hx1 , . . . , xm | xn = 1 for all words x in x1 ,
. . . , xm i. For which integers m, n > 0, is B(m, n) finite?
The restricted Burnside problem. Let M be the intersection of all normal
subgroups of B(m, n) of finite index, and let B0 (m, n) = B(m, n)/M . Is B0 (m, n)
finite for all m, n > 0?
The general Burnside problem was answered negatively in 1964 by Golod and
Shafarevich [11]. For the Burnside problem, B(m, n) is known to be finite for some
pairs (m, n) and infinite for many other pairs (m, n) [16]; the complete answer to the
problem is not known. The restricted Burnside problem was answered affirmatively
by Zelmanov in 1992 [32, 33].

1.8. Nonabelian Groups of Order  30

Order pq, p | q 1. Orders 6, 10, 14, 21, 22, 26 are covered by Proposi-
tion 1.24.
Order 8. G ⇠
= D4 or Q8 .
Proof. Clearly, G has no element of order 8. However, G has an element of
order 4. (Otherwise, x2 = e for all x 2 G, whence G is abelian.) Let a 2 G be
such that o(a) = 4, and choose b 2 G \ hai, so G = ha, bi. Since |G/hai| = 2,
we have b2 2 hai, hence b2 = a2 or e since o(b2 ) = 1 or 2. Since hai C G, we
have bab 1 2 hai, hence bab 1 = a 1 . If b2 = a2 , then G ⇠
= Q8 ; if b2 = e, then
⇠
G = D4 . ⇤
Order 12. G ⇠ = A4 or D6 or Z3 o↵ Z4 , where ↵ : Z4 ! Aut(Z3 ) is the
homomorphism defined by ↵(1)(1) = 1.
Proof. Let P be a Sylow 3-subgroup of G and Q be a Sylow 2-subgroup
of G. Then G acts on G/P by left multiplication which gives a homomorphism
f : G ! SG/P with ker f ⇢ P . If ker f = {e}, then G ,! S4 , so G ⇠ = A4 . If
ker f = P , then P C G, hence G = P o Q, where Q ⇠ = Z4 or Z2 ⇥ Z2 . If Q ⇠
= Z4 ,
then G ⇠= Z3 o↵ Z4 . If Q ⇠
= Z2 ⇥ Z2 , we show that G ⇠
= D6 . Write P = hai. Then
there exists b 2 Q \ {e} such that ab = ba. (Otherwise, the map Q ! {a, a 1 },
 1.8. NONABELIAN GROUPS OF ORDER  30 23

Table 1.2. Nonabelain groups of order  30

|G| G #
6 S3 1
8 D 4 , Q8 2
10 D5 1
12 A4 , D 6 , Z 3 o ↵ Z 4 3
14 D7 1
D8 , Q16 , D4 ⇥ Z2 , Q8 ⇥ Z2 , Z8 o 1 Z2 , Z8 o 2 Z2 ,
16 9
(Z4 ⇥ Z2 ) o 3 Z2 , (Z4 ⇥ Z2 ) o 4 Z2 , Z4 o 5 Z4
18 D9 , S3 ⇥ Z3 , (Z3 ⇥ Z3 ) o Z2 3
20 D10 , Z5 o 1 Z4 , Z5 o 2 Z4 3
21 Z7 o Z 3 1
22 D11 1
D4 ⇥ Z3 , Q8 ⇥ Z3 , Z3 o✏1 Z8 , Z3 o✏2 (Z4 ⇥ Z2 ),
24 Z3 o✏3 (Z4 ⇥ Z2 ), Z2 ⇥ D6 , Z3 o✏4 (Z2 ⇥ Z2 ⇥ Z2 ), 12
Q 8 o ✏ 5 Z 3 , Z 3 o ✏ 6 Q 8 , Z 3 o ✏7 D 4 , Z 3 o ✏8 D 4 , S 4
26 D13 1
27 Z9 o⇣1 Z3 , (Z3 ⇥ Z3 ) o⇣2 Z3 2
28 D14 , Z7 o⌘ Z4 2
30 D15 , D5 ⇥ Z3 , S3 ⇥ Z5 3

b 7! bab 1 , would be one-to-one.) Thus o(ab) = 6. Choose c 2 Q \ habi. It is clear

that G = habi o hci ⇠
= Z6 o Z2 ⇠= D6 . ⇤
⇠
=
There is an isomorphism hx, y | x3 = y 4 = 1, yxy 1 = x 1 i ! Z3 o↵ Z4
given by x 7! (1, 0), y 7! (0, 1). Let a = ((1, 2, 3), 2), b = ((1, 2), 1) 2 S3 ⇥ Z4 and
T = ha, bi < S3 ⇥ Z4 . Then |T | = 12 and T = ha, b | a3 = b2 , bab 1 = a 1 i. There
⇠
=
is also an isomorphism hx, y | x3 = y 4 = 1, yxy 1 = x 1 i ! T given by x 7! a2 ,
y 7! b.
Order 16. Table 1.3. (See [3, §118].)
Order 18. G ⇠
= D9 or S3 ⇥Z3 or (Z3 ⇥Z3 )o Z2 , where : Z2 ! Aut(Z3 ⇥Z3 )
is the homomorphism defined by (1)(x, y) = (x, y).
Proof. We have G = P o Q, where P is the Sylow 3-subgroup of G and
Q is a Sylow 2-subgroup of G. If P ⇠ = Z9 , we have G ⇠ = D9 . If P ⇠
= Z3 ⇥ Z3 ,
then G ⇠= (Z3 ⇥ Z3 ) o✓ Z2 . If ✓(1) fixes a nonidentity element of Z3 ⇥ Z3 , then
G⇠= Z3 ⇥ (Z3 o Z2 ) ⇠
= Z3 ⇥ S3 ; otherwise, G ⇠= (Z3 ⇥ Z3 ) o Z2 . ⇤

⇠ D10 or Z5 o Z4 or Z5 o Z4 , where
Order 20. G = 1, 2 : Z4 ! Aut(Z5 )
1 2
are the homomorphisms defined by 1 (1)(1) = 2, 2 (1)(1) = 1.
Proof. Let hai C G be the Sylow 5-subgroup and P be a Sylow 2-subgroup.
If P = hbihci, where o(b) = o(c) = 2, we may assume bab 1 = a. (If bab 1 = a 1
24 1. GROUPS

Table 1.3. Nonabelian groups of order 16

group presentation
D8 hx, y | x = y 2 = 1, yxy 1 = x 1 i
8

Q16 hx, y | x4 = y 2 , yxy 1 = x 1 i

hx, y, z | x4 = y 2 = z 2 = 1, yxy 1 = x 1
,
D 4 ⇥ Z2
xz = zx, yz = zyi
hx, y, z | x2 = y 2 , yxy 1 = x 1 ,
Q8 ⇥ Z2
z 2 = 1, xz = zx, yz = zyi
Z8 o 1 Z2 ,
hx, y | x8 = y 2 = 1, yxy 1
= x5 i
1 : Z 2 ! Aut(Z 8 ), 1 (1)(1) = 5
Z8 o 2 Z2
hx, y | x8 = y 2 = 1, yxy 1
= x3 i
2 : Z2 ! Aut(Z8 ), 2 (1)(1) = 3
(Z4 ⇥ Z2 ) o 3 Z2 ,
! Aut(Z4 ⇥ Z2 ),
3 : Z2 ( hx, y, z | x4 = y 2 = z 2 = 1, xy = yx,
(1, 0) 7! (1, 0) zxz 1 = x, zyz 1 = x2 yi
3 (1) :
(0, 1) 7! (2, 1)
(Z4 ⇥ Z2 ) o 4 Z2 ,
4 ! Aut(Z4 ⇥ Z2 ),
: Z2 ( hx, y, z | x4 = y 2 = z 2 = 1, xy = yx,
(1, 0) 7! (1, 1) zxz 1 = xy, zyz 1 = yi
4 (1) :
(0, 1) 7! (0, 1)
Z4 o 5 Z4
hx, y | x4 = y 4 = 1, yxy 1
=x 1
i
5 : Z4 ! Aut(Z4 ), 5 (1)(1) = 1

and cac 1 = a 1 , then (bc)a(bc) 1 = a.) Then G = habihci, where Z10 ⇠ = habi C
G. Thus G ⇠ = Z10 o Z2 ⇠ = D10 . If P = hbi ⇠ = Z4 , then G ⇠ = Z5 o Z4 , where
: Z4 ! Aut(Z5 ) is a homomorphism such that (1)(1) 6= 1. If (1)(1) = 2 or 3,
G⇠ = Z5 o 1 Z4 ; if (1)(1) = 1, G ⇠= Z5 o 2 Z4 . To see that Z5 o 1 Z4 ⇠
6 Z5 o 2 Z4 ,
=
note that Z(Z5 o 1 Z4 ) = {(0, 0)} but Z(Z5 o 2 Z4 ) = h(0, 2)i. ⇤

Order 24. Table 1.4. (See [3, §126].)

Order 27. (Exercise ??) G =⇠ Z9 o⇣ Z3 or (Z3 ⇥ Z3 ) o⇣ Z3 , where ⇣1 : Z3 !

1 2
Aut(Z9 ) is the homomorphism defined by ⇣1 (1)(1) = 4, and ⇣2 : Z3 ! Aut(Z3 ⇥ Z3 )
is the homomorphism defined by
(
(1, 0) 7! (1, 0),
⇣2 (1) :
(0, 1) 7! (1, 1).

Order 28. G ⇠ = D14 or Z7 o⌘ Z4 , where ⌘ : Z4 ! Aut(Z7 ) is the homomorphism

defined by ⌘(1)(1) = 1.
 1.8. NONABELIAN GROUPS OF ORDER  30 25

Table 1.4. Nonabelian groups of order 24

group presentation
D4 ⇥ Z3
Q8 ⇥ Z3
Z 3 o ✏1 Z 8 ,
ha, b | a3 = b8 = 1, bab 1
=a 1
i
✏1 : Z8 ! Aut(Z3 ), ✏1 (1)(1) = 1
Z3 o✏2 (Z4 ⇥ Z2 ),
ha, b, c | a3 = b4 = c2 = 1, bc = cb,
✏2 : Z4 ⇥ Z2 ! Aut(Z3 ),
bab 1 = a, cac 1 = a 1 i
✏2 (1, 0)(1) = 1, ✏2 (0, 1)(1) = 1
Z3 o✏3 (Z4 ⇥ Z2 ),
ha, b, c | a3 = b4 = c2 = 1, bc = cb,
✏3 : Z4 ⇥ Z2 ! Aut(Z3 ),
bab 1 = a 1 , cac 1 = ai
✏3 (1, 0)(1) = 1, ✏3 (0, 1)(1) = 1
Z2 ⇥ D6
Z3 o✏4 (Z2 ⇥ Z2 ⇥ Z2 )
✏4 : Z2 ⇥ Z2 ⇥ Z2 ! Aut(Z3 ), ha, b, c, d | a3 = b2 = c2 = d2 = 1,
✏4 (1, 0, 0)(1) = 1, bc = cb, cd = dc, db = bd,
✏4 (0, 1, 0)(1) = 1, bab = a 1 , cac 1 = a, dad 1 = ai
1

✏4 (0, 0, 1)(1) = 1
Q 8 o ✏5 Z 3 ,
✏5 : Z3 ! ( Aut(Q8 ), ha, b, c | a2 = b2 , bab 1 = a 1 ,
i 7! j c3 = 1, cac 1 = b, cbc 1 = abi
✏5 (1) :
j 7! k
Z 3 o ✏6 Q 8 ,
ha, b, c | a3 = 1, b2 = c2 , cbc 1
=b 1
,
✏6 : Q8 ! Aut(Z3 ),
bab 1 = a, cac 1 = a 1
i
✏6 (i)(1) = 1, ✏6 (j)(1) = 1
Z 3 o ✏7 D 4 ,
ha, b, c | a3 = b4 = c2 = 1, cbc 1 = b 1
,
✏7 : D4 ! Aut(Z3 ),
bab 1 = a, cac 1 = a 1 i
✏7 (↵)(1) = 1, ✏7 ( )(1) = 1
Z 3 o ✏8 D 4 ,
ha, b, c | a3 = b4 = c2 = 1, cbc 1 = b 1
,
✏8 : D4 ! Aut(Z3 ),
bab 1 = a 1 , cac 1 = ai
✏8 (↵)(1) = 1, ✏8 ( )(1) = 1
S4 ha, b | a4 = b3 = (ab)2 = 1i

Proof. Let P be the Sylow 7-subgroup of G and Q be a Sylow 2-subgroup of

G. Then P ⇠
= Z7 and G ⇠= P o Q. If Q ⇠ = Z4 , G ⇠
= Z7 o⌘ Z4 . If Q ⇠
= Z2 ⇥ Z2 ,
⇠
G = D14 . ⇤

Order 30. G ⇠
= D15 or D5 ⇥ Z3 or S3 ⇥ Z5 .
Proof. Let P be a Sylow 5-subgroup of G and Q be a Sylow 3-subgroup of
G. By Example 1.25, P C G, thus P Q < G, whence P Q ⇠ = Z5 ⇥ Z3 . Hence
26 1. GROUPS

G⇠ = (Z5 ⇥ Z3 ) o✓ Z2 , where ✓ : Z2 ! Aut(Z5 ⇥ Z3 ) is the homomorphism defined

by
( ( (
(1, 0) 7! ( 1, 0), (1, 0) 7! (1, 0), (1, 0) 7! ( 1, 0),
✓(1) : or or
(0, 1) 7! (0, 1), (0, 1) 7! (0, 1), (0, 1) 7! (0, 1).
In these three cases, G ⇠
= D5 ⇥ Z3 , S3 ⇥ Z5 and D15 , respectively. ⇤

1.9. Group Extensions

i p
A sequence of groups and homomorphisms 1 ! K ! G ! Q ! 1 is called
a short exact sequence if i is one-to-one, p is onto, and im i = ker p; in this case,
G is called an extension of K by Q. If there is a homomorphism j : Q ! G
such that p j = idQ , the short exact sequence (or the extension) is called split.
p
i
Given K, G, Q, there exists a split extension 1 ! K ! G Q ! 1 if and only if
j
G⇠
= K o Q.
i p
The Q-module induced by an extension. Let 0 ! K ! G ! Q ! 1 be
an extension, where K is abelian and is written additively. Then Q acts on K as
automorphsims as follows: Let j : Q ! G be any lifting, i.e., p j = idQ ; the action
of Q on K is defined by
Q⇥K ! K
1 1
(x, a) 7 ! i (j(x)i(a)j(x) ).
Note that in the above, j(x)i(a)j(x) 1 is independent of the choice of j since K is
abelian. This action of Q on K makes K a (left) Q-module (§2.6). To recap, an
i p
extension 0 ! K ! G ! Q ! 1 with abelian K induces a Q-module K.
i p
Given a Q-module K, an extension 0 ! K ! G ! Q ! 1 is said to realize the
Q-module K if the Q-module K coincides with the one induced by the extension.
Let K be a Q-module and f : Q ⇥ Q ! K be a function. Define a binary
operation · on K ⇥ Q by:
(a, x) · (b, y) = a + xb + f (x, y), xy .
Then (K ⇥ Q, ·) is a group if and only if
(1.6) xf (y, z) f (xy, z) + f (x, yz) f (x, y) = 0 for all x, y, z 2 Q.
The function f is called a 2-cocycle if (1.6) holds. The group (K ⇥ Q, ·) is denoted
by G(K, Q, f ) if f is a 2-cocycle.
Remark 1.31.
(i) Equation (1.6) is equivalent to the associativity of · in K ⇥ Q.
(ii) Equation (1.6) implies that f (1, y) = f (1, 1) and f (x, 1) = xf (1, 1) for all
x, y 2 Q. Moreover, (1.6) implies the existence of identity and the inverse
in G(K, Q, f ): eG(K,Q,f ) = ( f (1, 1), 1), (a, x) 1 = ( x 1 a f (1, 1)
f (x 1 , x), x 1 ).
(iii) If f is a 2-cocycle, then there exists an exact sequence
i p
(1.7) 0 ! K ! G(K, Q, f ) ! Q ! 1
 1.9. GROUP EXTENSIONS 27

where
i: K ! G(K, Q, f )
(1.8)
a 7 ! (a f (1, 1), 1),

p: G(K, Q, f ) ! Q
(1.9)
(a, x) 7 ! x.
Moreover, extension (1.7) realizes the Q-module K. When f = 0, the
extension 1.7 is split with the homomorphism j : Q ! G(K, Q, 0), x 7!
(0, x).
i p
Let 0 ! K ! G ! Q ! 1 be an extension with abelian K which makes K a
Q-module, and let j : Q ! G be a lifting. Then : K ⇥ Q ! G, (a, x) = i(a)j(x),
is a bijection. Let · be the operation in K ⇥ G such that (K ⇥ Q, ·) is a group and
: (K ⇥ Q, ·) ! G is an isomorphism. Then for (a, x), (b, y) 2 K ⇥ Q,
1 1
((a, x) · (b, y)) = i(a)j(x)i(b)j(y) = i(a)j(x)i(b)j(x) j(x)j(y)j(xy) j(xy)
1 ⇤
= i(a)i(xb)j(x)j(y)j(xy) j(xy) = i(a + xb + j (x, y))j(xy),
where
(1.10) j ⇤ (x, y) = i 1
(j(x)j(y)j(xy) 1
).
Hence
(a, x) · (b, y) = (a + xb + j ⇤ (x, y), xy).
By Remark 1.31 (i), j ⇤ is a 2-cocycle and (K ⇥ Q, ·) = G(K, Q, j ⇤ ). Moreover, the
diagram
i0 p0
0 ! K
? ! G(K,?Q, j ⇤ ) ! Q
? ! 1
(1.11) ? ? ?
yid y yid
i p
0 ! K ! G ! Q ! 1
commutes, where i0 and p0 are defined by (1.8) and (1.9). If j 0 : Q ! G is another
lifting, then j 0 (x)j(x) 1 = i h(x) for some function h : Q ! K. Moreover,
⇤
(1.12) j 0 (x, y) j ⇤ (x, y) = xh(y) h(xy) + h(x).

The second cohomology group. Let K be a Q-module. Let Z 2 (Q, K) be

the abelian group of all 2-cocycles f : Q ⇥ Q ! K. A function f : Q ⇥ Q ! K
is called a 2-coboundary if f (x, y) = xh(y) h(xy) + h(x) for some h : Q ! K.
Let B 2 (Q, K) be the set of all 2-coboundaries. Then B 2 (Q, K) < Z 2 (Q, K). The
group H 2 (Q, K) = Z 2 (Q, K)/B 2 (Q, K) is called the second cohomology group of Q
with coefficients in K.
Equivalence of extensions. Let K be a Q-module and let
i p
(1.13) 0 !K !G !Q !1
i0 p0
(1.14) 0 ! K ! G0 ! Q ! 1
28 1. GROUPS

be two extensions realizing the Q-module K. The two extensions are called equiv-
alent if there exists an isomorphism : G ! G0 such that the following diagram
commutes.
i p
0 ! K ? ! G? ! Q ? ! 1
? ? ?
yid y yid
i0 p0
0 ! K ! G0 ! Q ! 1

Theorem 1.32 (The meaning of H 2 (Q, K)). Let K be a Q-module, and let
E(Q, K) be the set of all equivalence classes of extensions of K by Q realizing the
Q-module K. There is a bijection ↵ : H 2 (Q, K) ! E(Q, K), where ↵(0) is the class
of split extensions.
Proof. For brevity, an extension 0 ! K ! G ! Q ! 1 realizing the Q-
module K is simply denoted by G and its equivalence class is denoted by [G].
Define ↵ : H 2 (Q, K) ! E(Q, K), f + B 2 (Q, K) 7! [G(K, Q, f )]. Note that ↵ is
well-defined. If f1 , f2 2 Z 2 (Q, K) are such that
f1 (x, y) f2 (x, y) = xh(y) h(xy) + h(x), x, y 2 Q,
for some h : Q ! K, then
: G(K, Q, f1 ) ! G(K, Q, f2 )
(a, x) 7 ! (a + h(x), x)
is an isomorphism and the diagram
i1 p1
0 ! K
? ! G(K,?Q, f1 ) ! Q
? ! 1
? ? ?
yid y yid
i2 p2
0 ! K ! G(K, Q, f2 ) ! Q ! 1
commutes, where i1 , p1 , i2 , p2 are defined by (1.8) and (1.9).
Next, define : E(Q, K) ! H 2 (Q, K), [G] 7! j ⇤ + B 2 (Q, K), where j : Q ! G
is a lifting and j ⇤ 2 Z 2 (Q, K) is given by (1.10). We claim that is well-defined.
Let [G] = [G0 ] 2 E(Q, K) and let j : Q ! G and j 0 : Q ! G0 be liftings. Since
[G] = [G0 ], we have a commutative diagram
i p
0 ! K
? ! G? ! Q
? ! 1
? ? ?
yid =y
⇠ yid
i0 p0
0 ! K ! G0 ! Q ! 1
⇤
Since j : Q ! G0 is also a lifting, by (1.12), ( j)⇤ ⌘ j 0 (mod B 2 (Q, K)). It
remains to show that ( j) = j . For x, y 2 Q, we have
⇤ ⇤

1
( j)⇤ (x, y) = i0 (( j)(x)( j)(y)( j)(xy) 1
)
0 1
=i ( (j(x)j(y)j(xy) 1
)) = i 1
(j(x)j(y)j(xy) 1
) = j ⇤ (x, y).
For f 2 Z 2 (Q, K), define a lifting j : Q ! G(K, Q, f ), j(x) = (0, x). Then it is
easy to check that j ⇤ = f . Hence
↵(f + B 2 (Q, K)) = ([G(K, Q, f )]) = j ⇤ + B 2 (Q, K) = f + B 2 (Q, K),
 1.10. NORMAL AND SUBNORMAL SERIES 29

i.e., ↵ is the identity map. For [G] 2 E(Q, K), let j : Q ! G be a lifting. Then
↵ ([G]) = ↵(j ⇤ + B 2 (Q, K)) = [G(K, Q, j ⇤ )] = [G],
where the last equality follows from (1.11). Hence ↵ is also the identity map.
i
Therefore ↵ is a bijection. We have ↵(0) = [G(K, Q, 0)], where 0 ! K !
p
G(K, Q, 0) ! Q ! 1 is a split extension since the lifting Q ! G(K, Q, 0), x 7! (0, x)
is a homomorphism. ⇤
Corollary 1.33. Let K be a Q-module. If H 2 (Q, K) = 0, then every extension
of K by Q realizing the Q-module K is a semidirect product K o Q.
Proposition 1.34. Let K be a Q-module where |K| = m, |Q| = n and
gcd(m, n) = 1. Then H 2 (Q, K) = 0.
Proof. Let f 2 Z 2 (Q, K). Then
xf (y, z) f (xy, z) + f (x, yz) f (x, y) = 0 for all x, y, z 2 Q.
Summing over z 2 Q gives
xh(y) h(xy) + h(x) = nf (x, y),
P
where h(x) = z2Q f (x, z). Therefore nf 2 B 2 (Q, K). Since gcd(m, n) = 1, we
have f 2 B 2 (Q, K). ⇤
Complement. Two subgroups H, K of G are called complements of each other
if G = HK and H \ K = {0}.
Corollary 1.35. If |G| = mn, where gcd(m, n) = 1, and G has an abelian
normal subgroup K of order m, then K has a complement.
Proof. By Proposition 1.34, H 2 (G/K, K) = 0, hence G ⇠
= K o (G/K). ⇤
Theorem 1.36 (Schur-Zassenhaus). If |G| = mn, where gcd(m, n) = 1, and
G has a normal subgroup K of order m, then K has a complement.
Proof. We use induction on m. If K has a nontrivial subgroup N such that
N C G, by the induction hypothesis, K/N has a complement H/N in G/N . Since
N C H, |H| = |N |n, gcd(|N |, n) = 1 and |N | < m, by the induction hypothesis
again, H has a subgroup of order n.
Therefore, we may assume that K is a minimal normal subgroup of G. Let
P 6= 1 be a Sylow subgroup of K. Since G = K NG (P ) (Proposition 1.23),
G/K = K NG (P )/K ⇠ = NG (P )/K \ NG (P ) = NG (P )/NK (P ). If NG (P ) 6= G,
then |NK (P )| < |K| = m. The induction hypothesis applied to NK (P ) C NG (P )
implies that NG (P ) has a subgroup of order n.
If NG (P ) = G, i.e., P C G, then 1 6= Z(P ) C G. Since Z(P ) ⇢ K, the
minimality of K implies that K = Z(P ), which is abelian. Then Corollary 1.35
applies. ⇤

1.10. Normal and Subnormal Series

Definition 1.37. Let G be a group. A sequence
(1.15) G = G0 B G1 B · · · B Gm = {e}
is called a subnormal series of G; Gi /Gi+1 , 0  i  m 1, are the factor groups
of the series, and the length of the series is |{0  i  m 1 : Gi 6= Gi+1 }|. If
30 1. GROUPS

Gi C G for all 1  i  m, (1.15) is called a normal series. If Gi /Gi+1 is simple for

all 0  i  m 1, (1.15) is called a composition series. If Gi /Gi+1 is abelian for
all 0  i  m 1, (1.15) is called a solvable series.
A (sub)normal series
(1.16) G = H0 B H1 B · · · B Hn = {e}
is called a refinement of (1.15) if {G0 , . . . , Gm } ⇢ {H0 , . . . , Hn }. Two subnormal
series S : G = G0 B · · · B Gm = {e} and T : G = H0 B · · · B Hn = {e} are called
equivalent if there is a bijection between the nontrivial factors of S and those of T
such that the corresponding factors are isomorphic.
Theorem 1.38 (The Schreier refinement theorem). Any two (sub)normal se-
ries of G have equivalent refinements.
Lemma 1.39 (Zassenhaus). Let A⇤ , A, B ⇤ , B be subgroups of G such that A⇤ C
A and B ⇤ C B. Then A⇤ (A \ B ⇤ ) C A⇤ (A \ B), B ⇤ (A⇤ \ B) C B ⇤ (A \ B) and
A⇤ (A \ B) ⇠ B ⇤ (A \ B)
= ⇤ ⇤ .
A⇤ (A \ B ⇤ ) B (A \ B)

A B
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
. .

A⇤ (A \ B) B ⇤ (A \ B)
......... ......
......... ........
......... ........
......... .........
.........
......... .
...
..........
.
..
......... .........
..... ........

A\B
A⇤ (A \ B ⇤ ) ...
... B ⇤ (A⇤ \ B)
...
... ......... . ...
......... ... ........
... ......... ... ........ ...
... ......... ......... ...
... ......... ...
... .
...
..........
. ...
... ......... ....... ...
......... ... .........
.. ..... . ......... ..

A⇤ (A \ B ⇤ )(A⇤ \ B) B⇤
... .. ......... ...
........ .........
... ........ ......... ...
... ......... ......... ...
... .
...
..........
. ......... ...
... ...... ......... ...
........ .........
.. ......... ..... ..

A⇤ \ B A \ B⇤
Proof. Let D = (A⇤ \ B)(A \ B ⇤ ). We claim that D C A \ B. Since A⇤ C A,
we have A⇤ \ B C A \ B; since B ⇤ C B, we have A \ B ⇤ C A \ B. Hence
D = (A⇤ \ B)(A \ B ⇤ ) C A \ B.
It suffices to show
A⇤ (A \ B) ⇠ A \ B
(1.17) = .
A⇤ (A \ B ⇤ ) D
⇠ (A \ B)/D.) Define
(By symmetry, we have B ⇤ (A \ B)/B ⇤ (A⇤ \ B) =
A\B
: A⇤ (A \ B) !
D
ax 7 ! Dx, a 2 A⇤ , x 2 A \ B.
1 We claim that is well defined. If ax = a0 x0 , where a, a0 2 A⇤ and x, x0 2
A \ B, then x0 x 1 = (a0 ) 1 a 2 A⇤ \ (A \ B) = A⇤ \ B ⇢ D, and hence Dx = Dx0 .
 1.10. NORMAL AND SUBNORMAL SERIES 31

2 Obviously, is onto.
3 We show that is a homomorphism. For ax, a0 x0 2 A⇤ (A \ B), we have
(axa0 x0 ) = (axa0 x 1 xx0 ) = Dxx0 = (ax) (a0 x0 ) since xa0 x 1 2 A⇤ .
4 We have ker = A⇤ D = A⇤ (A⇤ \ B)(A \ B ⇤ ) = A⇤ (A \ B ⇤ ). Now (1.17)
follows from the first isomorphism theorem. ⇤
Proof of Theorem 1.38. Let S : G = G0 B · · · B Gm = {e} and T : G =
H0 B · · · B Hn = {e} be two (sub)normal series. Let
Gij = Gi+1 (Gi \ Hj ), Hij = Hj+1 (Gi \ Hj ), 0  i  m, 0  j  n,
where Gm+1 = {e} and Hn+1 = {e}. Then
Gi 1,n = Gi0 = Gi , 1im 1,
Hm,j 1 = H0j = Hj , 1jn 1,
Gm 1,n 1 = Hm 1,n 1 = Gm 1 \ Hn 1.

We have a refinement S 0 for S

G = G00 B G01 B ··· B G0,n 1 B
G10 B G11 B ··· B G1,n 1 B
S0 : .. .. ..
. . .
Gm 1,0 B Gm 1,1 B ··· B Gm 1,n 1 B {e}
and a refinement T for T
0

G
q
H00 H01 ··· H0,n 1
5 5 5
H10 H11 ··· H1,n 1
T :0 5 5 5
.. .. ..
. . .
5 5 5
Hm 1,0 Hm 1,1 ··· Hm 1,n 1
5 5 5
{e}
In fact, it follows from the Zassenhaus lemma that Gi,j+1 C Gij and Hi+1,j C Hij .
Moreover, if S and T are normal series, then Gij C G and Hij C G for all i, j.
We claim that S 0 and T 0 are equivalent. It suffices to show that Gij /Gi,j+1 ⇠ =
Hij /Hi+1,j for all 0  i  m 1, 0  j  n 1. By the Zassenhaus lemma,
Gi+1 (Gi \ Hj ) ⇠ Hj+1 (Gi \ Hj )
Gij /Gi,j+1 = = = Hij /Hi+1,j .
Gi+1 (Gi \ Hj+1 ) Hj+1 (Gi+1 \ Hj )
⇤
Fact. Every finite group has a composition series.
Theorem 1.40 (Jordan-Hölder). Any two composition series of a group G are
equivalent.
32 1. GROUPS

Proof. Let S and T be two composition series of G. By Theorem 1.38, S and

T have refinements S 0 and T 0 , respectively, such that S 0 and T 0 are equivalent.
However, S and T do not allow proper refinements, hence S 0 = S and T 0 = T . ⇤
Examples of composition series. Sn B An B {id}, where n 5. S4 B
A4 B V B h(1, 2)(3, 4)i B {id}.
Derived subgroups and solvable groups. Let G be a group. Define G(0) =
G and G(i+1) = (G(i) )0 (the commutator subgroup of G(i) ) for i 0; G(i) is called
the ith derived subgroup of G. If G(n) = {e} for some n 2 Z+ , G is called solvable.
Example. We have
8
>
< An if n 5,
Sn0 = An , 0
An = V if n = 4,
>
:
{id} if n  3.
Proof. Since Sn /An is abelian, Sn0 ⇢ An . Since (i, j, k) = (i, j)(i, k)(i, j) 1
(i, k) 1 2 Sn0 for all distinct i, j, k 2 {1, . . . , n}, we have An ⇢ Sn0 .
If n 5, {id} = 6 A0n C An and An is simple, hence A0n = An . For n = 4, A4 /V
is abelian, hence A04 ⇢ V . Since A04 C S4 and A04 6= {id}, we have A04 = V . ⇤
Proposition 1.41. G is solvable if and only if G has a solvable series.
Proof. ()) G = G(0) B G(1) B · · · B G(n) = {e} is a solvable series.
(() Let G = G0 B G1 B · · · B Gn = {e} be a solvable series. Then Gi G0i 1
(n)
for all 1  i  n. Thus {e} = Gn G0n 1 G00n 2 · · · G0 = G(n) . ⇤
Fact. Let N C G. Then G is solvable if and only if N and G/N are both
solvable.
Proof. ()) Assume that G(n) = {e}. Then N (n) = {e} and (G/N )(n) = {N }.
(() There exist m, n 2 Z+ such that (G/N )(m) = {N } (i.e., G(m) ⇢ N )
and N (n) = {e}. Then G(m+n) = (G(m) )(n) ⇢ N (n) = {e}. (Alternatively, the
numerator of a solvable series of G/N together with a solvable series of N form a
solvable series of G.) ⇤
Example. Finite p-groups are solvable. Sn (n 5) is not solvable.
Theorem 1.42 (The Burnside p-q theorem). If |G| = pa q b , where p, q are
distinct primes and a, b 2 N, then G is solvable.
The standard proof of Theorem 1.42 requires representation theory; see [7, §34].
Theorem 1.43 (The Feit-Thompson theorem). Every finite group of odd order
is solvable.
The original proof of Theorem 1.43 is over 250 pages [10].
For groups H, K < G, define [H, K] = h{[h, k] : h 2 H, k 2 K}i, where
[h, k] = h 1 k 1 hk. If H C G and H ⇢ K ⇢ G, then K/H ⇢ Z(G/H) if and only
if [K, G] ⇢ H.
Central series and nilpotent groups. A normal series
{e} = G0 C G1 C · · · C Gn = G
 1.10. NORMAL AND SUBNORMAL SERIES 33

such that Gi+1 /Gi ⇢ Z(G/Gi ) is called a central series of G. If G has a central
series, G is called nilpotent.
Given a group G, to build a central series {e} = G0 C G1 C · · · of G from
bottom up greedily, one would choose Gi such that Gi /Gi 1 = Z(G/Gi 1 ). On the
other hand, to build a central series G = H0 B H1 B · · · from top down greedily,
one would choose Hi = [Hi 1 , G]. Let
(
Z0 (G) = {e},
Zi (G) be such that Z(G/Zi 1 (G)) = Zi (G)/Zi 1 (G), i > 0,
(
0 (G) = G,

i (G) = [ i 1 (G), G], i > 0.

The sequences of normal subgroups of G
{e} = Z0 (G) C Z1 (G) C · · ·
and
G= 0 (G) B 1 (G) B ···
are called the upper and lower central series of G, respectively. The upper (lower)
central series is a central series of G if and only if Zn (G) = G ( n (G) = {e}) for
some integer n 0.
Proposition 1.44. Let G be a group. The following statements are equivalent.
(i) G is nilpotent.
(ii) Zs (G) = G for some s 0.
(iii) t (G) = {e} for some t 0.
Proof. (i) ) (ii) and (iii). Let
{e} = G0 C G1 C · · · C Gn = G
be a central series. We use induction on i to show that Gi ⇢ Zi (G) for all i 0. As-
sume that Gi 1 ⇢ Zi 1 (G) for some i > 0. Then [Gi , G] ⇢ Gi 1 ⇢ Zi 1 (G), which
implies that Gi /Zi 1 (G) ⇢ Z(G/Zi 1 (G)) = Zi (G)/Zi 1 (G), i.e., Gi ⇢ Zi (G).
Therefore Zn (G) = G. Similarly, we have i (G) ⇢ Gn i for all 0  i  n, hence
n (G) = {e}.

(ii) or (iii) ) (i). A finite upper or lower central series is a central series. ⇤

Fact. Nilpotent groups are solvable.

Proof. A central series is a solvable series. (Alternatively, G(i) ⇢ i (G).) ⇤

Example. Let 3  n = 2t m, 2 - m and consider the dihedral group

Dn = h↵, | ↵n = 2
= 1, ↵ 1
=↵ 1
i.
It is easy to see that Dn0 = ha2 i and Dn00 = 1, whence Dn is solvable. Now we
compute the upper and lower central series of Dn .
First assume that m 3. We claim that
( i
h↵n/2 i if 0  i  t,
(1.18) Zi (Dn ) =
h↵m i if i > t.
34 1. GROUPS

i
To see (1.18), use induction on i. Assume that i < t and Zi (Dn ) = h↵n/2 i. Then
Dn /Zi (Dn ) = h↵ ¯ , ¯i ⇠
= Dn/2i , where ↵ ¯ = ↵Zi (Dn ), ¯ = Zi (Dn ), o(¯↵) = n/2i ,
i+1 i+1
o( ¯) = 2, ¯↵¯¯ 1 = ↵ ¯ 1 . Hence, Z(Dn /Zi (Dn )) = h↵ ¯ n/2 i = h↵n/2 i/Zi (Dn ),
so Zi+1 (Dn ) = h↵ n/2i+1
i. Since Dn /Zt (Dn ) = Dn /h↵m i ⇠= Dm , Z(Dn /Zt (Dn )) =
Zt (Dn )/Zt (Dn ), so Zt+1 (Dn ) = Zt (Dn ). It follows that Zi (Dn ) = Zt (Dn ) for all
i > t.
Now assume that m = 1, i.e., n = 2t , t > 1. We claim that
( i
h↵n/2 i if 0  i < t,
Zi (Dn ) =
Dn if i t.

For i < t, the proof is the same as above. Since Dn /Zt 1 (Dn ) = Dn /h↵2 i ⇠ = Z2 ⇥Z2 ,
which is abelian, we have Zt (Dn ) = Dn .
i
For the lower central series, we claim that i (Dn ) = h↵2 i for i 1. We use
induction on i. Since ↵2 = 1
↵ 1 ↵ 2 [Dn , Dn ], we have h↵2 i ⇢ 1 (Dn ). On
the other hand, h↵2 i C Dn and |Dn /h↵2 i| = 2 or 4. Thus Dn /h↵2 i is abelian,
2i
i.e., h↵2 i 1 (Dn ). Hence 1 (Dn ) = h↵ i. Assume that i (Dn ) = h↵ i for some
2
i+1 i i i+1
i > 0. Then ↵2 = 1
↵ 2 ↵2 2 [ i (Dn ), Dn ], hence h↵2 i ⇢ i+1 (Dn ).
2i+1 2i i+1 i+1 i+1
Clearly, in Dn /h↵ i, ↵ h↵2 i commutes with ↵h↵2 i and h↵2 i. Hence
i i+1 i+1 i i+1
h↵2 i/h↵2 i ⇢ Z(Dn /h↵2 i), so i+1 (Dn ) = [h↵2 i, Dn ] ⇢ h↵2 i. Therefore
2i+1
i+1 (Dn ) = h↵ i.
From the above, Dn is nilpotent if and only if n is a power of 2.

The upper and lower central series of a nilpotent group may differ. Clearly,
Zi (G ⇥ H) = Zi (G) ⇥ Zi (H) and i (G ⇥ H) = i (G) ⇥ i (H). Let G be any
nontrivial abelian group and H be a nonabelian group of order p3 . Then

Z0 (G ⇥ H) = {(eG , eH )}, Z1 (G ⇥ H) = G ⇥ H 0 , Z2 (G ⇥ H) = G ⇥ H,
0
2 (G ⇥ H) = {(eG , eH )}, 1 (G ⇥ H) = {eG } ⇥ H , 0 (G ⇥ H) = G ⇥ H.

Proposition 1.45. If G is nilpotent and H G, then H 6= NG (H).

Proof. Let {e} = G0 C G1 C · · · C Gn = G be a central series. Choose i such

that Gi ⇢ H but Gi+1 6⇢ H. Since Gi+1 /Gi ⇢ Z(G/Gi ), Gi+1 /Gi ⇢ NG/Gi (H/Gi ),
i.e., Gi+1 ⇢ NG (H). ⇤

Fact. Finite p-groups are nilpotent.

Proof. Let |G| = pk . If Zi (G) 6= G, then Zi+1 (G)/Zi (G) = Z(G/Zi (G)) is
nontrivial, i.e., Zi (G) ( Zi+1 (G). Thus Zn (G) = G for some n 0. ⇤

Proposition 1.46. Let G be a finite group. Then G is nilpotent if and only if

G is isomorphic to the direct product of its Sylow subgroups.

Proof. ()) We only have to show that every Sylow subgroup P of G is normal.
By Proposition 1.22, NG (NG (P )) = NG (P ). If NG (P ) 6= G, by Proposition 1.45,
NG (NG (P )) 6= NG (P ), which is a contradiction. Therefore NG (P ) = G, i.e., P C G.

(() Note that H ⇥K is nilpotent if and only if H and K are both nilpotent. ⇤
 1.11. EXAMPLES OF AUTOMORPHISM GROUPS 35

1.11. Examples of Automorphism Groups

Automorphism groups of Sn and An .
Lemma 1.47. If f 2 Aut(An ) maps a 3-cycle to a 3-cycle, then there exists
↵ 2 Sn such that f ( ) = ↵ ↵ 1 for all 2 An .
Proof. When n  3, the claim is obviously true. Now assume that n 4.
First note that f maps all 3-cycles to 3-cycles. When n = 4, every 3-cycle is
conjugate to (1, 2, 3) or (1, 2, 3) 1 in A4 . When n 5, all 3-cycles are conjugate in
An .
Let f (1, 2, 3) = (a1 , a2 , a3 ). Since (a1 , a2 , a3 )f (1, 2, 4) = f (1, 2, 3)(1, 2, 4)
= f (1, 3)(2, 4) has order 2, after a cyclic shift of (a1 , a2 , a3 ), we have f (1, 2, 4) =
(a1 , a2 , a4 ). For i 5, since (a1 , a2 , a3 )f (1, 2, i) = f (1, 3)(2, i) has order
2, we must have f (1, 2, i) = (a1 , a2 , ai ) or (a2 , a3 , c) or (a3 , a1 , d), where ai 2 /
{a1 , . . . , a4 } and c, d 2
/ {a1 , a2 , a3 }. In the last two cases, o f (1, 2, 4) f (1, 2, i) 6=
2, which is a contradiction. Hence we must have f (1, 2, i) = (a1 , a2 , ai ). Therefore
we have proved that
1
f (1, 2, i) = (a1 , a2 , ai ) = ↵(1, 2, i)↵ , i 3,
where ↵ 2 Sn is such that ↵(i) = ai , 1  i  n. Since An is generated by (1, 2, i),
3  i  n, we have f ( ) = ↵ ↵ 1 for all 2 An . ⇤

Lemma 1.48. Let G < Sn be such that [Sn : G] = n. Then there exists
✓ 2 Aut(Sn ) such that ✓(G) = { 2 Sn : (1) = 1} ⇠
= Sn 1 .
Proof. Let : Sn ! SSn /G be the action of Sn on Sn /G by left multiplication.
Then ker ⇢ G. Since the only normal subgroup of Sn with order dividing (n 1)!
is {id} (Fact 1.13), we have ker = {id}, i.e., is an isomorphism. Write Sn /G =
{G1 , . . . , Gn }, where G1 = G. Define f : {1, . . . , n} ! Sn /G, i 7! Gi , and define

✓: Sn ! Sn
1
7 ! f ( ) f.

Then ✓ 2 Aut(Sn ). For 2 G, ✓( )(1) = f 1

( ) f (1) = f 1
( ) (G) =
f 1 (G) = 1. ⇤

Theorem 1.49.
(i) For n 4, r : Aut(Sn ) ! Aut(An ), f !
7 f |An is an isomorphism.
(ii) Assume that n =
6 6. Then
(
⇠ 1 if n  2,
Aut(Sn ) = Inn(Sn ) =
Sn if n 3, n 6= 6,
8
>
<1 if n  2,
⇠
Aut(An ) = Z2 if n = 3,
>
:
Sn if n 4, n 6= 6.
(iii) [Aut(S6 ) : Inn(S6 )] = 2. More precisely, Aut(S6 ) ⇠
= S6 oZ2 but Aut(S6 ) ⇠
6 =
S6 ⇥ Z 2 .
36 1. GROUPS

Proof. The proof is based on Janusz and Rotman [19]. First note that r
maps Aut(Sn ) to Aut(An ) since An is the only subgroup of Sn of order |An |.
1 We claim that for n 4, r is one-to-one, whence we may write Aut(Sn ) ⇢
Aut(An ).
Let f 2 Aut(Sn ) be such that f |An = idAn . To prove that f = id, it suffices
to show that f (1, 2) = (1, 2). For each 2 An , we have f ((1, 2)) f ((1, 2)) 1 =
f ((1, 2) (1, 2) 1 ) = (1, 2) (1, 2) 1 . Thus (1, 2) 1 f ((1, 2)) 2 Z(An ) = {id}, i.e.,
f ((1, 2)) = (1, 2).
2 For n 4, n 6= 6, we claim that Aut(Sn ) = Inn(Sn ) and r : Aut(Sn ) !
Aut(An ) is an isomorphism.
Let g 2 Aut(An ) and let 2 An be a 3-cycle. Then g( ) is a product of k
disjoint 3-cycles. If k > 1, we have
1 1
(1.19) 3(n 3)! = |CAn ( )| = |CAn g( ) | = 3k k!(n 3k)!.
2 2
The integer solutions of (1.19) are (n, k) = (n, 1) and (n, k) = (6, 2), which con-
tradicts the assumptions. Therefore we have k = 1, and by Lemma 1.47, g = r(f )
for some f 2 Inn(Sn ). Hence r Inn(S ) : Inn(Sn ) ! Aut(An ) is onto. By 1 ,
n
Aut(Sn ) = Inn(Sn ) and r : Aut(Sn ) ! Aut(An ) is an isomorphism.
3 The remaining claims in (ii) about Aut(Sn ) and Aut(An ) for n  3 are
obvious. For example, to see that Aut(S3 ) = Inn(S3 ), note that for each f 2
Aut(S3 ), f (1, 2) is a 2-cycle and f (1, 2, 3) is a 3-cycle. Hence |Aut(S3 )| 
3 · 2 = |Inn(S3 )|.
4 We claim that [Aut(A6 ) : Inn(S6 )]  2. For any f 2 Aut(A6 ) \ Inn(S6 ), by
Lemma 1.47, f maps every 3-cycle to a product of two disjoint 3-cycles. Note that
in A6 , the number of 3-cycles equals the number of products of two disjoint 3-cycles.
Thus f maps every product of two disjoint 3-cycles to a 3-cycle. Therefore, for any
f, g 2 Aut(A6 ) \ Inn(S6 ), f g maps 3-cycles to 3-cycles, and hence f g 2 Inn(S6 )
by Lemma 1.47. This proves the claim.
5 We claim that Aut(S6 ) 6= Inn(S6 ). Let P be the set of the 6 Sylow 5-
subgroups of S5 , and let : S5 ! SP (= S6 ) be the action of S5 on P by conjuga-
tion. Since S5 acts transitively on P, 6 |S5 / ker |. Since A5 is only one nontrivial
normal subgroup of S5 and 6 - |S5 /A5 |, we have ker = {id}. Thus (S5 ) is a tran-
sitive subgroup of SP of index 6. Therefore, we have proved that S6 has a transitive
subgroup G of index 6. By Lemma 1.48, there exists f 2 Aut(S6 ) such that f (G)
fixes 1 and hence is not a transitive subgroup of S6 . It follows that f 2
/ Inn(S6 ).
6 By 4 and 5 , we have
Inn(S6 ) ( Aut(S6 ) ⇢ Aut(A6 )
and [Aut(A6 ) : Inn(S6 )]  2. Thus Aut(S6 ) = Aut(A6 ), [Aut(S6 ) : Inn(S6 )] = 2,
and (i) holds for n = 6.
7 From 5 , we know that there exists g 2 Aut(S6 ) \ Inn(S6 ) such that g(S5 )
is a transitive subgroup of S6 . Such a g can be determined as follows: For each
2 S6 , let 2 Inn(S6 ) denote ( ) 1 . Replacing g by g for a suitable
2 S6 , we may assume that g(⇢) = ⇢, where ⇢ = (1, 2, 3, 4, 5). By Remark 1.50
(i), g((5, 6)) = (⇤, ⇤)(⇤, ⇤)(⇤, ⇤). Replacing g by i⇢ g for a suitable 0  i  4,
 1.11. EXAMPLES OF AUTOMORPHISM GROUPS 37

we may assume that g((5, 6)) = (a, b)(c, d)(5, 6), where (a, b)(c, d) = (1, 2)(3, 4) or
(1, 3)(2, 4) or (1, 4)(2, 3). Hence
g((i, 6)) = g(⇢i (5, 6)⇢ i ) = ⇢i g((5, 6))⇢ i
= (a + i, b + i)(c + i, d + i)(i, 6),
where 1  i  5 and the sums are taken modulo 5. If (a, b)(c, d) = (1, 2)(3, 4), then
g((3, 5, 6)) = g((5, 6)(3, 6)) = g((5, 6))g((3, 6))
= (1, 2)(3, 4)(5, 6)(4, 5)(1, 2)(3, 6) = (3, 5)(4, 6),

which is impossible. If (a, b)(c, d) = (1, 3)(2, 4), then

g((4, 5, 6)) = g((5, 6)(4, 6)) = g((5, 6))g((4, 6))
= (1, 3)(2, 4)(5, 6)(5, 2)(1, 3)(4, 6) = (2, 6)(4, 5),

which is also impossible. Thus we must have (a, b)(c, d) = (1, 4)(2, 3), so
(1.20) g((i, 6)) = (1 + i, 4 + i)(2 + i, 3 + i)(i, 6), 1  i  5.
Note that g is completely determined by (1.20) since S6 is generated by (i, 6),
1  i  5. (In fact, since g(⇢) = ⇢, g is completely determined by g((5, 6)).) We
claim that g 2 = id. For this claim, it suffices to show that g 2 ((5, 6)) = (5, 6). We
have
g 2 ((5, 6)) = g((1, 4)(2, 3)(5, 6)) = g((1, 4))g((2, 3))g((5, 6)),
where, by (1.20),
g((1, 4)) = g((4, 6))g((1, 6))g((4, 6)) = (1, 3)(5, 6)(2, 4),
g((2, 3)) = g((3, 6))g((2, 6))g((3, 6)) = (6, 5)(2, 1)(4, 3),
g((5, 6)) = (1, 4)(2, 3)(5, 6).

From the above, one quickly checks that g 2 ((5, 6)) = (5, 6). Therefore Aut(S6 ) =
Inn(S6 )hgi ⇠
= S6 o Z 2 .
To prove that Aut(S6 ) 6⇠ = S6 ⇥ Z2 , we show that Z(Aut(S6 )) = {id}. Assume
to the contrary that there exists id 6= f 2 Z(Aut(S6 )). Then f 2 / Inn(S6 ) since
Z(Inn(S6 )) ⇠= Z(S 6 ) is trivial. By Remark 1.50 (i), f ((1, 2)) = (⇤, ⇤)(⇤, ⇤)(⇤, ⇤).
Choose 2 S6 such that fixes (1, 2) but not f ((1, 2)). Then ( f )((1, 2)) 6=
(f )((1, 2)), whence f 2 / Z(Aut(S6 )), which is a contradiction. ⇤

Remark 1.50.
(i) Let f 2 Aut(S6 ) \ Inn(S6 ). By step 4 in the above proof, f maps every
3-cycle to a product of two disjoint 3-cycles, and vice versa. It follows
that f maps every transposition to a product of three disjoint transposi-
tions and, by counting, vice versa. (Assume to the contrary that f ((1, 2))
is not a product of three disjoint transpositions. Since f (A6 ) = A6 ,
f ((1, 2)) is a transposition, and by conjugation, so is f ((1, 3)). Then
f ((1, 2, 3)) = f ((1, 3)(1, 2)) = f ((1, 3))f ((1, 2)), which cannot be a prod-
uct of two disjoint 3-cycles.)
(ii) By Lemma 1.48 and Theorem 1.49, if n 6= 6, every subgroup G < Sn with
|G| = (n 1)! must fix one of 1, . . . , n. This is false for n = 6 (step 5 of
the above proof).
38 1. GROUPS

The Automorphism group of GL(n, F ). Let F be a field and n 2. For

each P 2 GL(n, F ), 2 Aut(F ) and homomorphism : GL(n, F ) ! F ⇥ , define
gP, , : GL(n, F ) ! GL(n, F )
A 7 ! (A)P A P 1 ,
where A is the matrix obtained by applying to A entrywise. Then gP, , 2
Aut GL(n, F ) and all such automorphisms form a subgroup G < Aut GL(n, F ) .
Define
⌧ : GL(n, F ) ! GL(n, F )
A 7 ! (A 1 )T .
/ G unless n = 2; see Exercise ??. The full
Then ⌧ 2 Aut GL(n, F ) , but ⌧ 2
automorphism group Aut GL(n, F ) is generated by G and ⌧ ; see [8].
 CHAPTER 2

Rings and Modules

2.1. Rings, Basic Definitions

Definition 2.1. A ring is a nonempty set R equipped with two operations +
and · such that
(i) (R, +) is an abelian group;
(ii) (ab)c = a(bc) for all a, b, c 2 R;
(iii) a(b + c) = ab + ac and (a + b)c = ac + bc for all a, b, c 2 R.
If ab = ba for all a, b 2 R, R is called a commutative ring. If there exists an element
1R 2 R such that 1R a = a1R = a for all a 2 R, 1R is called the identity of R.
Subring. Let (R, +, ·) be a ring. A subset S ⇢ R is called a subring of R if
(S, +, ·) is a ring. The center of a ring R is Z(R) = {a 2 R : ax = xa for all x 2 R},
which is a subring of R.
Cartesian product. If R and S are two rings, then R ⇥ S is a ring with
operations defined componentwise.
Homomorphism. Let R and S be rings. A map f : R ! S is called a
homomorphism if f (a + b) = f (a) + f (b) and f (ab) = f (a)f (b) for all a, b 2 R. An
isomorphism is a bijective homomorphism.
In general, a ring may not have identity, e.g., 2Z. If S is a subring of R, any
of the following could happen: (i) R has identity but S does not (R = Z, S = 2Z);
(ii) S has identity but R does not (R = Z ⇥ 2Z, S = Z ⇥ {0}); (iii) R and S both
have identity but 1R 6= 1S (R = Z ⇥ Z, S = Z ⇥ {0}). If R and S are two rings
with identity, a homomorphism f : R ! S does not necessarily map 1R to 1S .
Declaration. In this book, unless specified otherwise, it is assumed that a ring
has identity; if S is a subring of R, then 1S = 1R ; a homomorphism maps identity
to identity.
Basic properties of rings. Let R be a ring.
(i) 0R a = a0R = 0R for a 2 R, where 0R is the identity of (R, +).
(ii) (na)b = a(nb) = n(ab), m(na) = (mn)a for a, b 2 R, m, n 2 Z.
(iii)
⇣Xn ⌘⇣X
m ⌘ X n Xm
ai bj = ai bj , ai , bj 2 R.
i=1 j=1 i=1 j=1
(iv) Assume that a1 , . . . , as 2 R are pairwise commutative. Then
X ✓ ◆
n
(a1 + · · · + as ) =n
ai11 · · · aiss ,
i +···+i =n
i 1 , . . . , i s
1 s

n
where i1 ,...,is = n!/(i1 ! · · · is !) 2 Z is the multinomial coefficient.
39
40 2. RINGS AND MODULES

The multiplicative group. An element a 2 R is called a unit or said to be

invertible if there exists b 2 R such that ab = ba = 1R ; b is called the (multiplicative)
inverse of a and is denoted by a 1 . The set of all units of R is denoted by R⇥ ;
(R⇥ , ·) is the multiplicative group of R.
Types of rings. An integral domain is a commutative ring R with 1R 6= 0
and with no zero divisors. (If a, b 2 R \ {0} are such that ab = 0, both a and b are
called zero divisors of R.) A division ring (skew field ) is a ring R with 1R 6= 0 and
R⇥ = R \ {0}. A field is a commutative division ring.
Examples. Commutative rings: Z; Zn (for a + nZ, b + nZ 2 Zn , define (a +
nZ)(b + nZ) = ab + nZ); C[0, 1] = the ring of continuous functions from [0, 1] to R.
Fields: Q, R, C, Zp (p prime).
Integral domains (not fields): Z; the polynomial ring over an integral domain.
Noncommutative rings: Mn (R) = the ring of n ⇥ n matrices over a ring R.
Subring generated by a subset. Let R be a ring and let X ⇢ R. Let
XP ⇤
= {x1 · · · xn : n 0, xi 2 X}. The smallest subring of R containing X is
n
{ i=1 ⇡i : n 0, ⇡i 2 X ⇤ }, which is called the subring of R generated by X. If S
is a subring of R such that all s 2 S commute with all x 2 X, then the smallest
subring of R containing S [ X is
nXn o
S[X] := ai ⇡i : n 0, ai 2 S, ⇡i 2 X ⇤ ,
i=1
p
which p
is called the subring of R generated by X over S. For example, Z[ 2] =
{a + b 2 : a, b 2 Z} ⇢ R.
Endomorphism ring. Let A be an abelian group, and let End(A) be the set
of all endomorphisms of A, i.e., homomorphisms from A to A. The endomorphism
ring of A is (End(A), +, ).
Fact. Every ring R is isomorphic to a subring of End((R, +)).
Proof. Define
f: R ! End((R, +))
r 7 ! f (r),
where
f (r) :
(R, +) ! (R, +)
x 7 ! rx.
Then f is a one-to-one ring homomorphism. ⇤
Example (The real quaternions, a division ring which is not a field). A real
quaternion is a formal sum a1 + a2 i + a3 j + a4 k, where a1 , . . . , a4 2 R. Let
H = {a1 + a2 i + a3 j + a4 k : a1 , . . . , a4 2 R},
and in H, define the addition componentwise and define the multiplication by
the distributive laws and the rules i2 = j 2 = k 2 = 1, ij = k, jk = i, ki = j,
ik = j, kj = i, ji = k. Then H becomes a ring. The only axiom that needs
checking is the associativity of the multiplication. P Let X = {1, i,Pj, k} and note
that the multiplication is associative in X. For a = x2X ax x, b = y2X by y, c =
P P P
z2X cz z 2 H, we have (ab)c = x,y,z2X ax by cz (xy)z = x,y,z2X ax by cz x(yz) =
a(bc). For z = a1 + a2 i + a3 j + a4 k, define z̄ = a1 a2 i a3 j a4 k; we have
 2.1. RINGS, BASIC DEFINITIONS 41

z z̄ = a21 + a22 + a23 + a24 2 R. For 0 6= z 2 H, z 1

= (1/z z̄)z̄, whence H is a division
ring.
Group rings. Let G be a group (written multiplicatively) and R P be a ring.
The group ring of G over R, denoted by R[G], is the set of formal sums g2G rg g,
where rg 2 R and rg 6= 0 for only finitely many g 2 G. The operations in R[G] are
defined by
X X X
rg g + sg g = (rg + sg )g,
g2G g2G g2G
⇣X ⌘⇣ X ⌘ X⇣ X ⌘
rh h sk k = rh sk g.
h2G k2G g2G h,k2G
hk=g
P
If X ⇢ G is closed under multiplication and e 2 X, then R[X] = { g2X rg g 2
R[G]} is a subring of R[G].
Characteristic. The characteristic of a ring R, denoted by char R, is the
smallest n 2 Z+ such that na = 0 for all a 2 R. If no such n exists, then char R = 0.
Let o(1R ) denote the additive order of 1R . Then char R = o(1R ) if o(1R ) < 1, and
char R = 0 if o(1R ) = 1. For example, char Zn = n, char Q = 0.
Fact. If D is an integral domain, then char D is either 0 or a prime.
Ideals. Let R be a ring. A subset I ⇢ R is called a left (right ) ideal of R if I
is a subgroup of (R, +) and ax 2 I (xa 2 I) for all a 2 R and x 2 I. An ideal is a
two-sided ideal.
For X ⇢ R, the ideal of R generated by X, denoted by hXi or (X), is
nX n o
ai xi bi : n 0, ai , bi 2 R, xi 2 X ;
i=1

it is the smallest ideal containing X. An ideal generated by one element is called a

principal ideal.
Sum and product of ideals. Let I, J be left (right) ideals of R. Define
I + J = {a + b : a 2 I, b 2 J};
I + J is the smallest left (right) ideal of R containing I [ J.
If I and J are ideals of R, define
nX n o
IJ = ai bi : n 0, ai 2 I, bi 2 J ;
i=1

IJ is an ideal of R and IJ ⇢ I \ J.
The quotient ring. Let I be an ideal of R. Then R/I is an abelian group. For
a + I, b + I 2 R/I, define (a + I)(b + I) = ab + I. The multiplication is well-defined
and (R/I, +, ·) is a ring, which is called the quotient ring of R by I. The map
⇡: R ! R/I
r 7 ! r+I
is an onto ring homomorphism, called the canonical homomorphism. If a, b 2 R are
such that a b 2 I, we write a ⌘ b (mod I).
42 2. RINGS AND MODULES

The kernel of a ring homomorphism f : R ! S is ker f = f 1 (0S ). A subset

I ⇢ R is an ideal of R if and only if I = ker f for some homomorphism f : R ! S.
Proposition 2.2 (Universal mapping property). Let f : R ! S be a homo-
morphism of rings and let I be an ideal of R such that I ⇢ ker f . Then there exists
a unique homomorphism f¯ : R/I ! S such that the following diagram commutes.

f
...............................................
R S
... .........
... .......
.. ......
⇡ ...... ..
..
.......
....
........ ...... f¯
......
. ......

R/I
Isomorphism theorems.
(i) Let f : R ! S be a homomorphism of rings. Then R/ ker f ⇠ = f (R).
(ii) If S is a subring of R and I is an ideal of R, then S + I is a subring of R
and (S + I)/I ⇠ = S/S \ I.
(iii) Let I ⇢ J be ideals of R. Then (R/I)/(J/I) ⇠ = R/J.
The correspondence theorem. Let I be an ideal of R. Let A be the set of
all ideals of R containing I and B be the set of all ideals of R/I. Then the map
A ! B, J 7! J/I, is a bijection. Let S be the set of all subrings of R containing
I and T be the set of all subrings of R/I. Then the map S ! T , S 7! S/I, is a
bijection.
m-adic topology. Let R be a ring and m be an ideal of R. For each x 2 R,
{x + mn : n 2 N} forms a neighborhood base of x. The topology on R defined by
this neighborhood base is called the m-adic topology. The following mappings are
continuous in the m-adic topology.
(i) R ⇥ R ! R, (x, y) 7! x + y;
(ii) R ! R, x 7! x;
(iii) R ⇥ R ! R, (x, y) 7! xy.
Proof. (i) (x + mn ) + (y + mn ) ⇢ x + y + mn .
(ii) (x + mn ) ⇢ x + mn .
(iii) (x + mn )(y + mn ) ⇢ xy + mn . ⇤
A ring R endowed with a topology such that the mappings (i) – (iii) are contin-
uous is called a topological ring. Thus R with the m-adic topology is a topological
ring. The ideal mn is both open and closed. In fact
S for every x 2n m , x + m ⇢ m
n n n
,
hence m is open. On the other hand, R \ m = x2R\mn (x + m ) is open, so m is
n n n
T1
closed. The m-adic topology is Hausdorff if and only if n=0 mn = {0}. The m-adic
topology is discrete if and only if m is nilpotent, i.e., mn = 0 for some n > 0.

2.2. Prime Ideals and Maximal Ideals

Definition 2.3. An ideal P of a ring R is called a prime ideal if (i) P 6= R,
and (ii) if A, B are ideals of R such that AB ⇢ P , then A ⇢ P or B ⇢ P .
An ideal M of R is called maximal if M 6= R and there is no ideal strictly
between M and R. Maximal left (right) ideals are defined in the same way.
Proposition 2.4. Let P be an ideal of R such that P 6= R.
 2.2. PRIME IDEALS AND MAXIMAL IDEALS 43

(i) If for a, b 2 R, ab 2 P implies that a 2 P or b 2 P , then P is prime.

(ii) If R is commutative, the converse of (i) is true.
Proof. (i) Suppose that A and B are ideals such that AB ⇢ P and A 6⇢ P .
Choose a 2 A \ P . Then for all b 2 B, ab 2 AB ⇢ P , whence b 2 P . Hence B ⇢ P .
(ii) Assume that ab 2 P . Then (a)(b) = (ab) ⇢ P , whence (a) ⇢ P or
(b) ⇢ P . ⇤

If R is not commutative, the converse of (i) is false. Let R = M2 (F ), where

F is any field. The only ideals of R are 0 and R, so 0 is a prime ideal of R. Let
A = [ 1 0 ], B = [ 0 1 ] 2 R. Then AB = 0 but A 6= 0, B 6= 0.
Proposition 2.5. Let R be a ring and I 6= R be a (left) ideal of R. Then I is
contained in a maximal (left) ideal of R.
Proof. Consider all (left) ideals J such that I ⇢ J 63 1 and use Zorn’s lemma.
⇤

Theorem 2.6. Let R be a commutative ring and I be an ideal of R.

(i) I is prime if and only if R/I is an integral domain.
(ii) I is maximal if and only if R/I is a field.
(iii) If I is maximal, then I is prime.
Proof. (i) ()) We only have to show that R/I has no zero divisors. Assume
that a + I, b + I 2 R/I are such that (a + I)(b + I) = I. Then ab 2 I, whence a 2 I
or b 2 I, i.e., a + I = I or b + I = I.
(() Let a, b 2 R be such that ab 2 I. Then (a + I)(b + I) = 0, hence a + I = I
or b + I = I, i.e., a 2 I or b 2 I.
(ii) ()) Let a 2 R \ I. By the maximality of I, (a) + I = R. Hence there exists
b 2 R and x 2 I such that ab + x = 1. Then (a + I)(b + I) = 1 + I, whence (a + I)
is invertible in R/I.
(() The only ideals of the field R/I are I/I and R/I. Thus there are no ideals
of R strictly between I and R. ⇤

Fact. If I is an ideal of a ring R such that R/I is a division ring, then I is a

maximal ideal. The converse is false: Let F be a field. Then 0 is a maximal ideal
of M2 (F ) but M2 (F ) is not a division ring.
Proposition 2.7. Let I1 , . . . , In be ideals of R such that I1 + · · · + In = R and
Ii Ij = {0} for all i 6= j. Write 1 = e1 + · · · + en , where ei 2 Ii . Then we have the
following conclusions.
(i)
(
ei if i = j,
(2.1) ei ej =
0 if i =
6 j.

(ii) Ii is a ring with identity ei . (It follows that e1 , . . . , en are unique.) More-
over, e1 , . . . , en are in the center of R and Ii = Rei .
(iii) R ⇠= I1 ⇥ · · · ⇥ In .
44 2. RINGS AND MODULES

Proof. (i) If i 6= j, then ei ej 2 Ii Ij = {0}, i.e., ei ej = 0. It follows that

ei = ei (e1 + · · · + en ) = e2i .
(ii) Let x 2 Ii . For each j 6= i, xej 2 Ii Ij = {0}, hence x = x(e1 +· · ·+en ) = xei .
In the same way, ei x = x.
Since ei is the identity of Ii and ei x = 0 = xei for all x 2 Ij , j 6= i, we see that
ei is in the center of R. Since Rei ⇢ Ii = Ii ei ⇢ Rei , we have Ii = Rei .
(iii) The map f : R ! I1 ⇥ · · · ⇥ In , a 7! (ae1 , . . . , aen ) is an isomorphism; its
inverse is g : I1 ⇥ · · · ⇥ In ! R, (x1 , . . . , xn ) 7! x1 + · · · + xn . ⇤
An element e in a ring is called an idempotent if e2 = e. Elements e1 , . . . , en
satisfying (2.1) are called orthogonal idempotents.
Theorem 2.8 (The Chinese remainder theorem). Let I1 , . . . , In be ideals of a
ring R such that Ii + Ij = R for all i 6= j. Then the map
f: R ! (R/I1 ) ⇥ · · · ⇥ (R/In )
a 7 ! (a + I1 , . . . , a + In )
is an onto homomorphism with ker f = I1 \ · · · \ In ; that is, for ai 2 R, 1  i  n,
there exists a 2 R, unique modulo I1 \ · · · \ In , such that a ⌘ ai (mod Ii ) for all
1  i  n.
Proof. We only have to show that f is onto. It suffices to show that there
exists a 2 R such that
(
1 (mod I1 ),
a⌘
0 (mod Ii ), 2  i  n.
For each 2  i  n, since I1 +Ii = R, there exists ai 2 I1 such that ai ⌘ 1 (mod Ii ).
Then a = (1 a2 ) · · · (1 an ) has the desired property. ⇤
Corollary 2.9. Let m1 , . . . , mn 2 Z+ be such that gcd(mi , mj ) = 1, i 6=
j, and let a1 , . . . , an 2 Z be arbitrary. Then there exists x 2 Z, unique modulo
lcm(m1 , . . . , mn ), such that x ⌘ ai (mod mi ) for all 1  i  n.
Example. Let X be a compact topological space and C(X, R) be the ring of
all continuous functions from X to R. For each a 2 X, let Ma = {f 2 C(X, R) :
f (a) = 0}. Then Ma , a 2 X, are all the maximal ideals of C(X, R).
Proof. Since C(X, R)/Ma ⇠ = R is a field, Ma is maximal.
Let M be a maximal ideal of C(X, R). Assume to the contrary that M 6= Ma
for all a 2 X. Then for each a 2 X, there exists fa 2 M such that fa (a) 6= 0.
Therefore fa (x)2 > 0 for all x in an open neighborhood Ua of a. Let Ua1 , . . . , Uan be
a finite cover of X. Then fa21 +· · ·+fa2n 2 M is invertible, and hence M = C(X, R),
which is a contradiction. ⇤

2.3. Factorization in Commutative Rings; UFD, PID and ED

Let R be a commutative ring. For a, b 2 R we say that a divides b, denoted as
a | b, if b = ax for some x 2 R. If a | b and b | a, then a and b are called associates,
denoted as a ⇠ b. If R is an integral domain, then a ⇠ b if and only if a = bu
for some u 2 R⇥ . An element a 2 R \ (R⇥ [ {0}) is called irreducible if a = bc
(b, c 2 R) implies that b or c is a unit; a 2 R \ (R⇥ [ {0}) is called prime if a | bc
(b, c 2 R) implies that a | b or a | c.
 2.3. FACTORIZATION IN COMMUTATIVE RINGS; UFD, PID AND ED 45

Definition 2.10 (PID). An integral domain R is called a principal ideal domain

(PID) if every ideal of R is principal.
Definition 2.11 (UFD). An integral domain R is called a unique factorization
domain (UFD) if the following two conditions are satisfied.
(i) For each a 2 R \ (R⇥ [ {0}), a = c1 · · · cn for some irreducible c1 , . . . , cn 2
R.
(ii) If c1 · · · cn = d1 · · · dm , where ci , dj 2 R are irreducible, then n = m and
after relabeling the indices suitably, ci ⇠ di , 1  i  n.
Definition 2.12 (ED). An integral domain R is called a Euclidean domain
(ED) if there is a function @ : R \ {0} ! N, called the Euclidean function, such that
(i) @(a)  @(ab) for all a, b 2 R \ {0};
(ii) for any a 2 R and 0 6= b 2 R, there exist q, r 2 R such that a = qb + r,
where r = 0 or @(r) < @(b).
Remark.
(i) If @ satisfies (i) and (ii) of Definition 2.12, so does @ min{@(x) : x 2
R \ {0}}. Thus, we may assume that 0 is in the image of @.
(ii) Let R be an ED and 0 6= x 2 R. Then x 2 R⇥ if and only if @(x) =
min @(y) : y 2 R \ {0} .
Proposition 2.13. Let R be a commutative ring.
(i) An element p 2 R is prime if and only if (p) is a nonzero prime ideal of
R.
(ii) If a 2 R is irreducible, then (a) is maximal in {(b) : b 2 R \ (R⇥ [ {0})}.
The converse is true if R is an integral domain.
(iii) If R is an integral domain, then every prime p 2 R is irreducible.
(iv) If R is a UFD, then p is a prime if and only if p is irreducible.
Proof. (i) ()) Assume that a, b 2 R are such that ab 2 (p). Then p | ab,
whence p | a or p | b, i.e., a 2 (p) or b 2 (p). Hence (p) is a prime ideal of R.
(() Let a, b 2 R be such that p | ab, i.e., ab 2 (p). Then a 2 (p) or b 2 (p),
i.e., p | a or p | b.
(ii) Let P = {(b) : b 2 R \ (R⇥ [ {0})}. Assume that a 2 R is irreducible and
that (a) ⇢ (b), where b 2 R \ (R⇥ [ {0}). Then a = bc for some c 2 R. Since a is
irreducible and b 2/ R⇥ , we have c 2 R⇥ , whence (a) = (b). Thus (a) is maximal
in P. Now assume that R is an integral domain and (a) is maximal in P. Clearly,
a2 / R⇥ [ {0}. Assume that a = bc for some b 2 R, c 2 R \ R⇥ . Then (a) ⇢ (c). By
the maximality of (a), we have (a) = (c), i.e., a ⇠ c. Since R is an integral domain,
it follows that b 2 R⇥ . Therefore a is irreducible.
(iii) Suppose that p = ab, a, b 2 R. Since p | ab and p is prime, we may assume
that p | a, i.e., a = pu for some u 2 R. Then p = pub, whence ub = 1, i.e., b is a
unit.
(iv) (() Assume that p | ab, a, b 2 R \ {0}. Then pq = ab for some q 2 R. By
the uniqueness of factorization, p appears in the factorization of a or b, i.e., p | a or
p | b. ⇤
46 2. RINGS AND MODULES

If R is not a UFD, pan irreducible element is not necessarily p a prime. For

example, consider 2 2 Z[ 5]. If 2 = xy for some x, y 2 Z[ 5], then 4 = |2|2 =
|x|2 |y|2 . It follows that one of |x|2 and |y|2 , say p
|x|2 , is 1, hence
p x is a unit. Therefore
p
2 is irreducible.
p However, since 2 | 6 = (1 + 5)(1 5) and 2 - (1 + 5),
2 - (1 5), 2 is not a prime.
Fact. ED ) PID ) UFD.
Proof. ED ) PID. Let R be an ED and I 6= {0} be an ideal of R. Let a 2 I
be such that @(a) is the smallest. Then I = (a).
PID ) UFD. Let R be a PID.
Existence of factorization. Let a 2 R \ (R⇥ [ {0}). Assume to the contrary
that a is not a product of finitely many irreducibles. Since a is not irreducible,
a = a1 a01 , where a1 , a01 2 R \ (R⇥ [ {0}) and, without loss of generality, a1 is not
a product of finitely many irreducibles.
S1 Write a1 = a2 a02 , ... In this
S1 way we have
(a) ( (a1 ) ( (a2 ) ( · · · . Since i=1 (ai ) is an ideal of R, we have i=1 (ai ) = (b)
for some b 2 R. Then b 2 (ai ) for some i, whence (ai+1 ) ⇢ (b) ⇢ (ai ), which is a
contradiction.
Uniqueness of factorization. First show that every irreducible element a of R
is a prime. (By Proposition 2.13 (ii), (a) is a maximal ideal, hence (a) is a prime
ideal and a is a prime.) Then use induction on the number of irreducible factors in
the factorization. ⇤
Examples of EDs. Z, the polynomial ring in one variable over a field, and
([25, §5.4])
p
Z[ d], d = 2, 1, 2, 3, 6, 7, 11, 19,
p
Z[(1 + d)/2], d = 11, 7, 3, 5, 13, 17, 21, 29, 33, 37, 41, 57, 73.

Example (UFD 6) PID). The polynomial ring Z[x] is a UFD (Theorem 2.27).
In Z[x], (2, x) is not a principal ideal.
p
Example 2.14 (PID 6) ED). Let ↵ = (1 + 19)/2. We claim that Z[↵] is
a PID but not an ED.
Proof. 1 Z[↵] is not an ED.
The units of Z[↵] are ±1. In fact, it is easy to see that u 2 Z[↵] is a unit if and
only if |u|2 = 1. Assume to the contrary that Z[↵] is an ED with the Euclidean
function @. We may assume that 0 2 im @. Let ✏ 2 Z[↵] be such that @(✏) is the
smallest in Z+ . Then ✏ 2
/ Z[↵]⇥ . We have
2 = q✏ + r, q 2 Z[↵], r = 0, ±1,
so q✏ = 1, 2, 3. Thus |✏| 2
1 , 2 , 3 , whence |✏|2 = 1, 2, 4, 3, 9. Also,
2 2 2

↵ = q 1 ✏ + r1 , q1 2 Z[↵], r1 = 0, ±1,
p
so q1 ✏ 2 19/2 + (1/2){±1, 3}. Thus |✏|2 (19 + 12 )/4 or (19 + 32 )/4, i.e., |✏|2 5
or 7. Then |✏|2 = 1, which is impossible since ✏ 2
/ Z[↵]⇥ .
2 We claim that for every z 2 C, there exists q 2 Z[↵] such that either
|z q| < 1 or |z q/2| < 1/2.
Let z = x + yi. We may assume that z belongs to the closed parallelogram
0, 1, ↵+1, ↵ (since there is a p 2 Z[↵] such that z +p belongs to that parallelogram);
 2.3. FACTORIZATION IN COMMUTATIVE RINGS; UFD, PID AND ED 47

see Figure 2.1. We want to show that z has distance < 1 from one of the dots or
has distance < 1/2 from one of the circles. For this purpose, we may assume
that z belongs to
p the closed triangle 0, 1/2, ↵. Assume that |z ↵/2|
p 1/2,pthen
(x 1/4)2 p +(y p19/4)2 1/4. p Since |xp 1/4|  1/4, we have |y 19/4| 3/4,
i.e., y  ( 19 3)/4 or y ( 19 + 3)/4. In the first case, |z 0| < 1; in the
second case, |z ↵| < 1.
...........................................
......... .......
....... ......
.......... .....
..... .....
....
...... ...
.... ...
...
..
. ...
... ...
.... ...
...
↵ ↵+1 ... ↵
... .... ...
....... • .........................................................• . • ...
.. ........ .. .. ... .......
... ......... ... ... ... ..... ....
... .......... ... ... ... . ..
. .
.
... .... .... .... ... .... ... .. ....
. ..
.
...
... .. ... ... ... .. ... ... ... ...
... ... ... ... ... ... ... ... .. ...
... ... .. .. ... ... ... .. .. ...
... .. ..... .....
. .
. ..
. ..... .......................................... ...
....
........... . ... .....
... ... ... ... .... ... ...... ... ... ...... .....
... ... ... ... ... ... ... ............ .... .... ..... ............
...↵/2.... .... ..... ... ... ... ............
... .. .............
.
... ... ... ... .
. ..
. ..... ↵/2 ... ........................ ....
... .. ... ... .... ... ... . .. ..
... ... .... ... .. .............................. .... ...
... ... ....
. ............... .. ............. ....
... .... .... ........ ..... ..
... ... ...
... ... ..
... .. .. . . .........
. .. .
. ............
.
. ..... ..
... ... ... ... .. .. ..... ...... .... ... ..........
... ... ... ... ... ..... ............. ................... .......
... ... .... ...
... .... ... ... .. ... ... ... ... ...
... .. ... ... ... .. ..
. ..
. ... ...
... ... ... ... ... ... ... ... .... ...
... ... ... ... ... ... .... .... ... ...
....... ...
...... ... ......... ... ... ..... ...
....... ... ........ .... .
.
............................... . ...
• . ....
...
...
.. . . .
.. . .
. . .. . . . .. ............. . . . . . . . . . . . ...• ................................................ ... • .
. .
... ...
0 1/2 1 ... 0 ..
.
... ...
...
... ...
... ...
... ..
... ....
.
..... .
..... .....
...... .....
....... .....
......... .......
............................................

Figure 2.1. Example 2.14

3 Z[↵] is a PID.
Let I 6= {0} be an ideal of Z[↵]. Let 0 6= 2 I be such that | |2 is the smallest.
We claim that I = ( ). For any 2 I, by 2 , there exists q 2 Z[↵] such that
| / q| < 1 or | / q/2| < 1/2. If | / q| < 1, then | q | < | |, whence
q = 0, so 2 ( ). Now assume that | / q/2| < 1/2. Then |2 q | < | |,
whence = (q/2) . We claim that q/2 2 Z[↵]. Assume the contrary. Then
q = a + b↵, where a, b 2 Z and at least one of a, b is odd.
(i) Assume that a is odd and b is even. Then (q + 1)/2 2 Z[↵]. Thus /2 =
(q + 1)/2 2 I and 0 < | /2| < | |, which is a contradiction.
(ii) Assume that a is even and b is odd. We have
q↵
¯ = a↵
¯ + 5b = (a + 5b) a↵ = a0 + b0 ↵ =: q 0 ,
where a0 is odd, b0 is even and (q 0 /2) 2 I. This is case (i).
(iii) Assume that a, b are both odd. We have
q↵
¯ = (a + 5b) a↵ = a0 + b0 ↵ =: q 0 ,
where a0 is even, b0 is odd and (q 0 /2) 2 I. This is case (ii). ⇤
Gauss integers. The ring of Gauss integers, Z[i] = {a + bi : a, b 2 Z}, is an
ED with @(↵) = |↵|2 . We have Z[i]⇥ = {↵ 2 Z[i] : |↵|2 = 1} = {±1, ±i}.
Proof. Let ↵, 2 Z[i] with 6= 0. There exists q 2 Z[i] such that |↵/ q| <
1, hence |↵ q| < | |. ⇤
48 2. RINGS AND MODULES

Primes in Z[i]. Let ↵ 2 Z[i] be neither 0 nor a unit. Then ↵ is a prime, i.e.,
irreducible, if and only if
(i) ↵ ⇠ p for some prime p 2 Z with p ⌘ 1 (mod 4) or
(ii) |↵|2 is a prime in Z.
Proof. (() Assume (i). Suppose to the contrary that p is not a prime in Z[i].
Then p = , where , 2 Z[i], | |2 > 1, | |2 > 1. Since p2 = | |2 | |2 in Z, it
follows that p = | |2 , whence p 6⌘ 1 (mod 4), which is a contradiction.
Assume (ii). If ↵ = , where , 2 Z[i], then |↵|2 = | |2 | |2 in Z, whence
| | = 1 or | | = 1, i.e., or is a unit.
2 2

()) We have |↵|2 = p1 · · · pn , where p1 , . . . , pn are primes in Z. Since ↵ | ↵↵

¯=
p1 · · · pn and ↵ is prime, ↵ | pk =: p for some k. Then |↵|2 p2 in Z, hence |↵|2 = p
or p2 . If |↵|2 = p, we have (ii). Now assume that |↵|2 = p2 . Since ↵ | p, we have
p = u↵ for some u 2 Z[i]. Then |u|2 = 1, i.e., u is a unit. It remains to show
that p ⌘ 1 (mod 4). If p = 2 or p ⌘ 1 (mod 4), by Lemma 2.15, p = a2 + b2 for
some a, b 2 Z. Then ↵ = u 1 p = u 1 (a + bi)(a bi) is not irreducible, which is a
contradiction. ⇤
Lemma 2.15. Let p be an odd prime integer. Then the following are equivalent.
(i) p ⌘ 1 (mod 4).
(ii) 1 is a square in Zp .
(iii) p = a2 + b2 for some a, b 2 Z.
Proof. (i) ) (ii). Since 4 | p 1 = |Z⇥
p | and Zp is cyclic (Proposition 3.43),
⇥

there exists x 2 Zp with o(x) = 4, hence 1 = x .

⇥ 2

(ii) ) (iii). We claim that p is not irreducible in Z[i]. Assume the contrary. By
(ii), there exists x 2 Z such that p | x2 +1 = (x+i)(x i), hence p | x+i or p | x i,
say p | x + i. Then x + i = p(a + bi) for some a, b 2 Z, whence 1 = pb, which is
impossible. Therefore, p = ↵ , where ↵, 2 Z[i] are nonunits. Thus p2 = |↵|2 | |2
in Z, and hence p = |↵|2 (= | |2 ). ⇤
Theorem 2.16 (Sum of two squares). Let x 2 Z+ have factorization x =
· · · pemm q1f1 · · · qnfn , where p1 , . . . , pm , q1 , . . . , qn are distinct primes with pj ⌘ 1
pe11
(mod 4) and qk = 2 or qk ⌘ 1 (mod 4). Then x = a2 + b2 for some a, b 2 Z if and
only if e1 , . . . , em are all even.
Proof. (() We have qk = |↵k |2 for some ↵k 2 Z[i]. Hence x = |↵|2 , where
e /2 e /2
↵ = p11 · · · pmm ↵1f1 · · · ↵nfn 2 Z[i].
()) We have x = ↵↵ ¯ for some ↵ 2 Z[i]. Assume to the contrary that ej is
odd for some j and write ej = 2l + 1. Since pj is a prime of Z[i] and p2l+1
j ¯,
| ↵↵
ej 1
we have pl+1
j ¯ , say pl+1
| ↵ or ↵ j | ↵. Then pj x = |↵/pl+1
j | 2 Z, which is a
2

contradiction. ⇤
gcd and lcm. Let R be a commutative ring and X ⇢ R. An element d 2 R is
called a greatest common divisor of X, denoted by gcd(X), if
(i) d | x for all x 2 X and
(ii) if c 2 R is such that c | x for all x 2 X, then c | d.
An element m 2 R is called a least common multiple of X, denoted by lcm(X), if
(i0 ) x | m for all x 2 X and
 2.4. FRACTIONS AND LOCALIZATION 49

(ii0 ) if c 2 R is such that x | c for all x 2 X, then m | c.

p
A setpmay not p have a gcd or lcm. For p example, in Z[ 3], let a = 2 · 2 =
(1 + 3)(1 + 3) and b = 2(1 + 3). One can show that gcd(a, b) and
lcm(a, b) do not exist. If a set has a gcd (lcm), all its T gcds (lcms) are associates. If
R is a PID, then hgcd(X)i = hXi and hlcm(X)i = x2X hxi.
Assume that R is a UFD. Primes of R which are associates will be treated
as being the same. Let P be the set of all distinct primes in R. Then for each
x 2 R \ {0},
Y
x⇠ p⌫p (x) ,
p2P

where ⌫p (x) 2 N and ⌫p (x) > 0 for

Q only finitely many p 2 P. We define ⌫p (0) = 1
for all p 2 P. Moreover, define p2P pep = 0 if ep = 1 for some p 2 P or ep > 0
for infinitely many p 2 P. Then
Y Y
gcd(X) ⇠ pinf{⌫p (x):x2X} , lcm(X) ⇠ psup{⌫p (x):x2X} .
p2P p2P

2.4. Fractions and Localization

The ring of fractions. Let R be a commutative ring. A multiplicative set
of R is a subset S such that 1 2 S, 0 2 / S and S is closed under multiplication.
For (r, s), (r0 , s0 ) 2 R ⇥ S, define (r, s) ⇠ (r0 , s0 ) if there exists s1 2 S such that
s1 (rs0 r0 s) = 0. The relation ⇠ is an equivalence relation on R⇥S; the equivalence
class of (r, s) in R ⇥ S is denoted by r/s. Let S 1 R = (R ⇥ S)/ ⇠ = {r/s : r 2
R, s 2 S}. For r/s, r0 /s0 2 S 1 R, define
r r0 rr0 r r0 rs0 + sr0
· 0 = 0, + 0 = .
s s ss s s ss0
The operations are well-defined and (S 1 R, +, ·) is a commutative ring, which is
called the ring of fractions of R by S. If R is an integral domain, then so is S 1 R.
If R is a integral domain and S = R \ {0}, then S 1 R is a field, which is called the
fractional field of R or the quotient field of R (not to be confused with the quotient
ring by an ideal).
Examples. The fractional field of Z is Q. The fractional field of the polynomial
ring over a field F is the field of rational functions over F .

Proposition 2.17. Let R be a commutative ring and S be a multiplicative set

of R.
(i) The map
1
S : R ! S R
r 7 ! r/1
is a homomorphism. For every s 2 S, S (s) is a unit of S 1 R.
(ii) S is one-to-one if and only if S contains no zero divisors.

Proposition 2.18 (Universal mapping property). Let R be a commutative

ring and S be a multiplicative set of R. Let T be another commutative ring and
50 2. RINGS AND MODULES

f : R ! T be a homomorphism such that f (S) ⇢ T ⇥ . Then there is a unique

homomorphism f¯ : S 1 R ! T such that the following diagram commutes.
f
...............................................
R T
... .
.........
... ...... .
... ......
. ......
S .... .
.......
.... f¯
..
.
........ ......
. ......

S 1R
Proof. Existence. Define f¯ : S 1 R ! T , r/s 7! f (r)f (s) 1
.
Uniqueness. Assume that g : S 1 R ! T is another homomorphism such that
g S = f . Then for each r 2 R and s 2 S, g(r/s)f (s) = g(r/s)g( S (s)) =
g((r/s)(s/1)) = g( S (r)) = f (r), hence g(r/s) = f (r)f (s) 1 . ⇤

Local rings. A local ring is a ring R with a unique maximal left ideal M ; this
definition is left-right symmetric by the following proposition.
Proposition 2.19. Let R be a ring.
(i) If R is local, the unique maximal left ideal of R is R \ R⇥ .
(ii) R is local if and only if R \ R⇥ is closed under addition.
Proof. (i) Let M be the unique maximal left ideal of R. Then M = {x 2 R :
x has no left inverse}. (x 2 R has no left inverse , Rx 6= R , Rx is contained in
a maximal left ideal of R , Rx ⇢ M .) Clearly, M ⇢ R \ R⇥ . If, to the contrary,
there exists a 2 R \ R⇥ such that a 2 / M , then a has a left inverse b 2 R but b
does not have a left inverse, i.e., b 2 M . Then 1 ab 2 / M , hence 1 ab has a left
inverse. By Exercise ??, 1 ba = 0 has a left inverse, which is a contradiction.
(ii) (() Clearly, R \ R⇥ is an ideal of R. Let M be any maximal left ideal of
R. Since M ⇢ R \ R⇥ , we have M = R \ R⇥ , which is unique. Hence R is local. ⇤

If R is commutative, the proof of Proposition 2.19 (i) is simpler: Every x 2

R \ R⇥ is contained in a maximal ideal of R, whence x 2 M (the unique maximal
ideal of R). Therefore R \ R⇥ ⇢ M .
If R is a commutative local ring with the unique maximal ideal M , R/M is
called the residue field of R. For example, let p be a prime and n > 0. Then Zpn
is a local ring with maximal ideal pZpn and residue field Zpn /pZpn ⇠
= Zp .
Localization. Let R be a commutative ring and P be a prime ideal of R.
Then S = R \ P is a multiplicative set of R. Moreover, S 1 R is a local ring with
maximal ideal S 1 P := {r/s : r 2 P, s 2 S}. If r/s 2 (S 1 R) \ (S 1 P ), where
r 2 R and s 2 S, then r 2 R \ P = S, so r/s is invertible in S 1 R. The ring S 1 R
is called the localization of R at P and is denoted by RP . For example, let p 2 Z
be a prime. Then Z(p) = {a/b : a, b 2 Z, p - b}.

2.5. Polynomial Rings

Polynomial ring in one indeterminate. Let R be a ring. A polynomial in
x (the indeterminate) with coefficients in R is a formal sum
f = a0 + a1 x + · · · + an xn , n 2 N, ai 2 R.
 2.5. POLYNOMIAL RINGS 51

The degree of f , denoted by deg f , is max{i : ai 6= 0} if ai 6= 0 for some i, and

is 1 if ai = 0 for all i. Let R[x] denote the set of all polynomials in x with
coefficients in R and define + and · in R[x] as follows:
n
X n
X n
X
ai xi + bi x i = (ai + bi )xi ,
i=0 i=0 i=0
⇣X
n ⌘⇣X
m ⌘ X⇣
n+m X ⌘
ai xi bj x j = a i bj x k .
i=0 j=0 k=0 i+j=k

Then (R[x], +, ·) is a ring, which is called the polynomial ring in x over R.

Polynomial ring in a set of indeterminates. Let R be a ring and let X
be a set of symbols (indeterminates). Let A be the set of all functions ↵ : X ! N
such that ↵(x) = 0 for almost all (all but finitely many) x 2 X. A polynomial in
X with coefficients in R is a formal sum
X
f= a↵ X ↵ ,
↵2A

where a↵ = 0 for almost all ↵ 2 A. Let R[X] be the set of all polynomials in X
with coefficients in R and define + and · in R[X] as follows:
X X X
a↵ X ↵ + b↵ X ↵ = (a↵ + b↵ )X ↵ ,
↵2A ↵2A ↵2A
⇣X ⌘⇣ X ⌘ X⇣ X ⌘
↵
a↵ X b X = a↵ b X .
↵2A 2A 2A ↵+ =

Then (R[X], +, ·) is a ring, which is called the polynomial ring in X over R. For
↵ 2 A, let supp ↵ = {x 2 X : ↵(x) > 0}, which is finite. If supp ↵ = {x1 , . . . , xn },
↵(x ) ↵(x )
we write X ↵ = x1 1 · · · xn n . For each f 2 R[X], there exist x1 , . . . , xn 2 X
such that f 2 R[x1 , . . . , xn ].
Let F be the free abelian group on X, written multiplicatively. (A free abelian
group on X is a free Z-module on X; see §2.6.) Let
X = {xd11 · · · xdnn : n 0, xi 2 X, di 2 Z+ } ⇢ F,
which is closed under multiplication. The subring R[X ] of the group ring R[F ] is
precisely the polynomial ring R[X].

Proposition 2.20 (Universal mapping property). Let R[X] be the polynomial

ring in X over R. Let S be another ring and f : R ! S be a homomorphism. Let
: X ! S be a function such that every element in (X) commutes with every
element in (X) [ f (R). Then there exists a unique homomorphism f¯ : R[X] ! S
such that the following diagram commutes.
f
...............................................
R S
.. ........... ...
.... .. .
..... ........
... ..... ....
... ¯ ..
...... ...
...
...
f.........
...
.
... ..... ...
.... ...
.. .....
......... ........ ...
. ... ..
...
R[X] ...........................................
X
52 2. RINGS AND MODULES

Proof. Define f¯ : R[X] ! S by

X X
ad1 ,...,dn xd11 · · · xdnn 7! f (ad1 ,...,dn ) (x1 )d1 · · · (xn )dn .
d1 ,...,dn d1 ,...,dn

⇤
Proposition 2.21. If X and Y are disjoint sets of indeterminates, then (R[X])[Y ] ⇠
=
R[X [ Y ].
Proof. By Proposition 2.20, there exist homomorphisms g : (R[X])[Y ] !
R[X [Y ] and h : R[X [Y ] ! (R[X])[Y ] such that the following diagrams commute.

⇢ ⇢
R[X] .............................................................. R[X [ Y ] R (R[X])[Y ]
..............................................................

... ... ... ... ... ...

... ........ ....... ... ........ .......
..... .....
... ..... ..... ... ..... ....
...
..
......
. .. ...
..
......
. ...
... .... ... ... ..
.....
... ..... ... .....
..
..... ... ..... ...
...
.. g.............. ...
...
.. h............. ...
\ ..... .
.....
..
. ... [
....
\ ..... .
.....
. ... [
....
... ..... ... .....
... ..... .. ... ..... ..
... ..... ... ... ..... ...
... ..
......
. ... ... ..
......
. ...
.. ..
... ..... ... ... ..... ...
....... ..... ... ....... ..... ...
..... .....
. ..... .. . ..... ..

(R[X])[Y ].............................................................. Y R[X [ Y ] .............................................................. X [ Y

The map h g : (R[X])[Y ] ! (R[X])[Y ] is identity on R [ X [ Y . Since (R[X])[Y ]

is generated by X [ Y over R, we have h g = id. In the same way, g h = id. ⇤
Proposition 2.22 (The division algorithm). Let R be a ring and let f, g 2
R[x] be such that the leading coefficient of g is a unit. Then there exist unique
q, r, q 0 , r0 2 R[x] such that
f = qg + r and f = gq 0 + r0 ,
where deg r < deg g and deg r0 < deg g.
Fact. If F is a field, then F [x] is an ED with Euclidean function @(f ) = deg f .
P
Let R be a commutative ring, f = d1 ,...,dn ad1 ,...,dn xd11 · · · xdnn 2 R[x1 , . . . , xn ]
P d1
and (c1 , . . . , cn ) 2 Rn . We write f (c1 , . . . , cn ) = d1 ,...,dn ad1 ,...,dn c1 · · · cn . If
dn

f (c1 , . . . , cn ) = 0, (c1 , . . . , cn ) is called a root or zero of f .

Fact.
(i) Let R be a commutative ring, f 2 R[x] and c 2 R. Then f (c) = 0 if and
only if x c | f .
(ii) If D is an integral domain and 0 6= f 2 D[x] with deg f = n, then f has
at most n distinct roots in D.
Derivative. Let R be a commutative ring and let f = a0 + · · · + an xn 2 R[x].
The derivative of f is f 0 = a1 + 2a2 x + · · · + nan xn 1 . All applicable differentiation
rules hold.
The multiplicity of a root. Let R be a commutative ring, 0 6= f 2 R[x]
and c 2 R. Then f can be uniquely written as f = (x c)m g, where m 2 N and
g 2 R[x], g(c) 6= 0; the uniqueness of the expression follows from the fact that x c
is not a zero divisor. The integer m is called the multiplicity of the root c of f ; c is
a multiple root of f (i.e., with multiplicity m > 1) if and only if f (c) = f 0 (c) = 0.
 2.5. POLYNOMIAL RINGS 53

The Hasse derivative. Let R be a commutative ring and let f (x) = a0 +

a1 x + · · · + an xn 2 R[x]. For c 2 R, we have
Xn Xn Xi ✓ ◆
i i k
(2.2) f (x) = ai (x c + c) =i
ai c (x c)k
i=0 i=0
k
k=0
Xn ⇣X n ✓ ◆ ⌘ Xn
i
= ai ci k (x c)k = (@ (k) f )(c)(x c)k ,
k
k=0 i=k k=0

where
n ✓ ◆
X i
@ (k) f = ai xi k
2 R[x].
k
i=k

The polynomial @ (k) f is called the kth order Hasse derivative of f ; we have f (k) =
k! @ (k) f .
Properties of the Hasse derivative. Let f, g 2 R[x] and a, b 2 R.
(i) @ (k) (af + bg) (k)
P = a@ (i) f + b@ (k) g.
(ii) @ (f g) = i+j=k (@ f )(@ (j) g).
(k)

(iii) @ (k) f (x + a) = (@ (k) f )(x + a).

(iv) c 2 R is a root of f of multiplicity m if and only if (@ (0) f )(c) = · · · =
(@ (m 1) f )(c) = 0.
Proof. For (i) – (iii), use (2.2) and treat c as an indeterminate. ⇤
Definition 2.23 (Content). Let D be a UFD and let 0 6= f = a0 +· · ·+an xn 2
D[x]. The content of f is C(f ) = gcd(a0 , . . . , an ). If C(f ) ⇠ 1, f is called primitive.
Lemma 2.24 (Gauss). Let D be a UFD and let f, g 2 D[x] be primitive. Then
f g is primitive.
Proof. Assume to the contrary that there is an irreducible p 2 D such that
p | C(f g). Let : D[x] ! (D/(p))[x] be the homomorphism induced by the
canonical homomorphism D ! D/(p). Then 0 = (f g) = (f ) (g), where (f ) 6=
0, (g) 6= 0. However, since D/(p) is an integral domain, so is (D/(p))[x], and we
have a contradiction. ⇤
Corollary 2.25. Let D be a UFD and f, g 2 D[x] be nonzero. Then C(f g) ⇠
C(f )C(g).
Proposition 2.26. Let D be a UFD and F be its fractional field. For f 2
D[x] \ D, f is irreducible in D[x] if and only if f is primitive and is irreducible in
F [x].
Proof. ()) Clearly, f is primitive. Assume to the contrary that f is not
irreducible in F [x], i.e., f = gh, g, h 2 F [x], deg g > 0, deg h > 0. Choose
a, b 2 D \ {0} such that ag, bh 2 D[x]. Then abf = (ag)(bh) 2 D[x], hence
ab = C(abf ) = C(ag)C(bh). Now f = (ag)(bh)/ab = (ag/C(ag)) · (bh/C(bh)),
where ag/C(ag), bh/C(bh) 2 D[x] have degree > 0, which is a contradiction.
(() Assume to the contrary that f = gh, where g, h 2 D[x] are nonzero and
nonunits. Since f is irreducible in F [x], one of g and h has degree 0. Then f is not
primitive, which is a contradiction. ⇤
54 2. RINGS AND MODULES

Theorem 2.27. Let D be a UFD. Then D[x] is also a UFD. The irreducible
elements of D[x] are precisely the irreducible elements of D and the primitive poly-
nomials in D[x] which are irreducible in F [x], where F is the fractional field of
D.
Proof. The second claim follows from Proposition 2.26. It remains to show
that D[x] is a UFD.
Existence of factorization. Let f 2 D[x] be nonzero and non-invertible. Since
F [x] is a UFD, f = f1 · · · fn , where fi 2 F [x] is irreducible. Choose 0 6= ai 2 D
such that ai fi 2 D[x] and write ai fi = ci gi , where ci 2 D and gi 2 D[x] is primitive
and is irreducible in F [x]. Then
a1 · · · an f = (a1 f1 ) · · · (an fn ) = (c1 g1 ) · · · (cn gn ) = c1 · · · cn g1 · · · gn .
Comparing the contents of both sides, we have c1 , · · · cn /a1 · · · an 2 D. Now
c1 , · · · cn
f= g1 · · · gn ,
a1 · · · an
where c1 , · · · cn /a1 · · · an is a product of irreducibles in D.
Uniqueness of factorization. Suppose that
(2.3) a 1 · · · a m f 1 · · · f n = b1 · · · b s g 1 · · · g t ,
where a1 , . . . , am , b1 , . . . , bs 2 D are irreducible and f1 , . . . , fn , g1 , . . . , gt 2 D[x] are
irreducible of degree > 0. Comparing the contents of the two sides of (2.3) gives
a1 · · · am ⇠ b1 · · · bs . Hence m = s and, after relabeling indices, ai ⇠ bi .
In F [x], f1 · · · fn ⇠ g1 · · · gt . Thus n = t and, after relabeling indices, fj ⇠ gj
in F [x]. Therefore fj = (u/v)gj for some u, v 2 D \ {0}, i.e., vfj = ugj . Then
v = C(ufj ) ⇠ C(ugj ) = u in D, hence fj ⇠ gj in D[x]. ⇤
Corollary 2.28. If D is a UFD and X is a set of indeterminates, then D[X]
is a UFD.
Proof. If |X| < 1, the conclusion follows from Theorem 2.27. In general,
note that for each f 2 D[X], there exists X 0 ⇢ X with |X 0 | < 1 such that
f 2 D[X 0 ]. Moreover, f 2 D[X 0 ] is irreducible if and only if f is irreducible in
D[X]. ⇤
Eisenstein’s criterion. Let D be a UFD with fractional field F and let f =
a0 + · · · + an xn 2 D[x], n > 0. If there is an irreducible element p 2 D such that
p - an , p | ai for 0  i  n 1 and p2 - a0 , then f is irreducible in F [x].
Proof. Assume to the contrary that f = gh, where g, h 2 F [x], deg g > 0,
deg h > 0. Then there exist g1 , h1 2 D[x] such that f = g1 h1 and g1 ⇠ g and
h1 ⇠ h in F [x]; see the proof of Proposition 2.26. Let : D[x] ! (D/(p))[x] be
the homomorphism induced by the canonical homomorphism D ! D/(p). Then
(an )xn = (g1 ) (h1 ). Since D/(p) is an integral domain, we have (g1 ) = ↵xk ,
(h1 ) = xl , ↵, 2 D/(p). Since k  deg g1 , l  deg h1 , and k + l = n =
deg g1 + deg h1 , we have k = deg g1 and l = deg h1 , hence k, l > 0. Then p | g1 (0)
and p | h1 (0), whence p2 | g1 (0)h1 (0) = a0 , which is a contradiction. ⇤
Example. Let p be a prime. Then p (x) = 1 + x + · · · + xp 1 2 Q[x] is
irreducible.
Pp Applying Eisenstein’s criterion to p (x + 1) = ((x + 1)p 1)/x =
p i 1
i=1 i x yields the assertion.
 2.6. MODULES, DEFINITIONS AND BASIC FACTS 55

2.6. Modules, Definitions and Basic Facts

Definition 2.29. Let R be a ring (not required to have identity). A left R-
module (a left module over R) is an abelian group (A, +) equipped with a scalar
multiplication R ⇥ A ! A, (r, a) 7! ra such that for r, s 2 R and a, b 2 A,
(i) r(a + b) = ra + rb;
(ii) (r + s)a = ra + sa;
(iii) r(sa) = (rs)a.
A right R-module is an abelian group (A, +) equipped with a scalar multiplication
A ⇥ R ! A. (a, r) 7! ar such that the analogies of (i) – (iii) hold. A left (right)
R-module is sometimes denoted by R A (AR ). If R has identity and
(iv) 1R a = a for all a 2 A,
A is called a unitary left R-module.
Declaration. Unless specified otherwise, all modules are assumed to be uni-
tary. A module is assumed to be left if the side is not specified.
Examples of modules. Abelian groups are Z-modules. A vector space over
a field F is an F -module. Left (right) ideals of a ring R are left (right) R-modules.
Let Mm⇥n (R) be the set of all m ⇥ n matrices over a ring R. Then it is a left
Mm (R)-module and a right Mn (R)-module where the scalar multiplications are
matrix multiplications.
Let V be a vector space over a field F and let ↵ : V ! V be a linear trans-
formation. For each f 2 F [x] and v 2 V , define f v = f (↵)v. Then V is an
F [x]-module.
Let A be an abelian group. For each a 2 A and f 2 End(A), define f a = f (a).
Then A is an End(A)-module. Let G be a group acting on A as automorphisms.
Then A becomes
P a Z[G]-module with the scalarP multiplication defined as follows:
for r = g2G ↵g g 2 Z[G] and a 2 A, ra = g2G ↵g ga. We say that A is a
G-module.
Example. Let f : R ! S be a ring homomorphism and let A be an S-module.
For r 2 R and a 2 A, define ra = f (r)a. Then A becomes an R-module.
Homomorphism. Let A, B be R-modules. A function f : A ! B is called a
homomorphism, or an R-map, if f (a + b) = f (a) + f (b) and f (ra) = rf (a) for all
a, b 2 A and r 2 R. An isomorphism is a bijective homomorphism.
Submodule. Let A be an R-module and B ⇢ A. B is called a submodule of
A if B with the inherited operations is an R-module. A nonzero module is called
simple or irreducible if it does not have any proper submodule. The submodules of
R R are precisely left ideals of R.
If X ⇢ A, the smallest submodule of A containing X, called the submodule
generated by X, is
nX n o
hXi = (X) := ri xi : n 2 N, ri 2 R, xi 2 X .
i=1

Quotient module. Let A be an R-module and B be a submodule of A.

Let A/B be the quotient abelian group. For a + B 2 A/B and r 2 R, define
r(a + B) = ra + B. Then A/B is an R-module, called the quotient module of A by
B.
56 2. RINGS AND MODULES

Isomorphism Theorems.
The first isomorphism theorem. Let f : A ! B be a homomorphism of
R-modules. Then
f¯ : A/ ker f ! im f
a + ker f 7 ! f (a)
is an isomorphism.
The second isomorphism theorem. Let A, B be submodules of an R-
module. Then (A + B)/B ⇠
= A/(A \ B).
The third isomorphism theorem. Let C ⇢ B ⇢ A be R-modules. Then
(A/C)/(B/C) ⇠
= A/B.
Direct product and external direct sum. Let {AQ i : i 2 I} be a family of
R-modules. The direct product of {A Qi : i 2 I}, denoted by i2I Ai , is the cartesian
product of Ai , i 2 I.
Q Elements in i2I iA are of the form (a i )i2I , where ai 2 Ai .
The direct product i2I Ai is an R-module with addition and scalar multiplication
defined componentwise. The external direct sum of {Ai : i 2 I} is
M(ex) n Y o
Ai = (ai )i2I 2 Ai : ai = 0 for almost all i ,
i2I i2I
Q L(ex) Q
which is a submodule of i2I Ai . If |I| < 1, then i2I Ai = i2I Ai .
Sum and internal direct sum. If {Ai : i 2 I} is a family of submodules of
an R-modules A, the submodule
⌦ [ ↵ nX o
Ai = ai : ai 2 Ai , ai = 0 for almost all i
i2I i2I
P P
is called the sum of {Ai : i 2 I} and is denoted by i2I Ai . If Ai \ j2I\{i} Aj =
P
{0} for all i 2 I, then i2I Ai is called an internal direct sum and is denoted by
L(in)
i2I Ai . There is an isomorphism
M(ex) M(in)
Ai ! Ai
i2I i2I
X
(ai )i2I 7 ! ai .
i2I
L(ex) L(in) L
When distinction is unecessary, we write both and as .
Bimodule. Let R and S be rings. An (R, S)-bimodule is an abelian group
A which is a left R-module and a right S-module such that (ra)s = r(as) for all
a 2 A, r 2 R, s 2 S. An (R, S)-bimodule A is denoted by R AS . Homomorphisms
between (R, S)-bimodules are called (R, S)-maps.
Hom. For R-modules R A, R B, let HomR (R A, R B) be the abelian group of all
R-maps from A to B. Let S be anther ring.
(i) If R AS is a bimodule, then HomR (R AS , R B) is a left S-module. For
f 2 HomR (R AS , R B), s 2 S and a 2 A, we define (sf )(a) = f (as).
(ii) If R BS is a bimodule, then HomR (R A, R BS ) is a right S-module. For
f 2 HomR (R A, R BS ), s 2 S and a 2 A, we define (f s)(a) = (f (a))s.
 2.6. MODULES, DEFINITIONS AND BASIC FACTS 57

Free modules. Let A be an R-module. A subset X 2 A is called linearly

independent if r1 x1 + · · · + rn xn = 0 (ri 2 R, x1 , . . . , xn 2 X distinct) implies that
r1 = · · · = rn = 0. The set X is called a basis of A if X is linearly independent and
hXi = A. If A has a basis X, A is called a free module (on X); in this case,
M(in) M(ex)
A= Rx ⇠= R.
x2X x2X

If all bases of A have the same cardinality, this common cardinality is denoted by
rank A. If A is free with a basis X and B is another R-module, then every function
f : X ! B can be uniquely extended to an R-map f¯ : A ! B. Every R-module is
a quotient of a free R-module.
Theorem 2.30. Let D be a division ring. Then every D-module V is free.
Any two bases of V have the same cardinality.
Proof. A maximal linearly independent subset of V , which exists by Zorn’s
lemma, is a basis.
Let X and Y be two bases of V . If |X| = 1 or |Y | = 1, we have |X| = |Y |
by the next lemma. Now assume that X = {x1 , . . . , xn } and Y = {y1 , . . . , ym }.
Assume to the contrary that n > m. We have
2 3 2 3 2 3 2 3
x1 y1 y1 x1
6 . 7 6 7 6 7 6 7
6 . 7 = A 6 .. 7 and 6 .. 7 = B 6 .. 7
4 . 5 4 . 5 4 . 5 4 . 5
xn ym ym xn
for some matrices A 2 Mn⇥m (D) and B 2 Mm⇥n (D). It follows that AB = In .
There exists an invertible C 2 Mn⇥n (D) such that
" #
⇤
CA = .
0 ··· 0
Thus (0, . . . , 0, 1)C = (0, . . . , 0, 1)CAB = 0, which is a contradiction. ⇤
Lemma 2.31. Let R be a ring and F be a free R-module with an infinite basis
X. Then every basis of F has the same cardinality as X.
Proof. Let Y be another basis of F . We claim that |Y | = 1. Otherwise, since
each y 2 Y is a linear combination of finitely many x 2 X, F would be generated
by a finite subset X1 of X. However, any x 2 X \ X1 is not a linear combination
of elements in X1 , which is a contradiction.
For each x 2 X, there exists a finite subset {y1 , . . . , yn } ⇢ Y such S that x =
r1 y1 + · · · + rn yn , ri 2 R. DefineSf (x) = {y1 , . . . , yn }. We claim that x2X f (x) =
Y . Since X is spanned by Y1 := x2X f (x) ⇢ Y , Y is spanned by Y1 , hence Y1 = Y .
Now,
[
|Y | = f (x)  |X|@0 = |X|.
x2X
By symmetry, |X|  |Y |. Hence |X| = |Y |. ⇤
Let D be a division ring. A D-module V is called a vector space over D; its
dimension is dimD V = |X|, where X is any basis of V .
Fact.
58 2. RINGS AND MODULES

(i) If W ⇢ V are vector spaces over D, then dim V = dim W + dim(V /W ).

(ii) (The dimension formula) If V and W are subspaces of some vector space
over D, then
dim V + dim W = dim(V + W ) + dim(V \ W ).
Proof. (i) Let X be a basis of W and extend X to a basis X t Y of V . Then
y + W , y 2 Y , are all distinct and form a basis of V /W . Hence dim V /W = |Y |.
(ii) Define a D-map
f: V ⇥W !
V +W
(v, w) 7 !
v + w.
Then f is onto and ker f = {(v, v) : v 2 V \ W } ⇠
= V \ W . Hence
dim V + dim W = dim(V ⇥ W ) = dim(im f ) + dim(ker f )
= dim(V + W ) + dim(V \ W ).
⇤

The invariant dimension property. A ring R is said to have the invariant

dimension property (IDP) if for every free R-module F , any two bases of F have
the same cardinality. Division rings (Theorem 2.30) and commutative rings (the
next proposition) have IDP. The rank of a free module F over a ring with IDP,
denoted by rank F , is the cardinality of any basis of F .
Proposition 2.32. A commutative ring R has IDP.
Proof. We may assume that 1R 6= 0. Let F be a free R-module and let X be
a basis of F . Let I be a maximal ideal of R. Then R/I is a field. Let IF be the
submodule of F generated by {ax : a L 2 I, x 2 F }. Then F/IF
L is a vector space
over R/I. We may assume that F = L x2X R. Then
L IF = x2X I and we have
L
an R/I-module isomorphism F/IF = ( x2X R)/( x2X I) ⇠ = x2X (R/I). Thus
|X| = dimR/I (F/IF ), which is independent of X. ⇤
Example (A ring without IDP). Let R be a ring and R A, R B be R-modules.
Let E = EndR (B). Then HomR (A, B) is a left E-module with the scalar multi-
plication defined by composition. If R A1 and R A2 are two R-modules, we have an
E-module isomorphism
HomR (A1 A2 , B) ⇠ = HomR (A1 , B) HomR (A2 , B).
In fact, let ⇡i : A1 A2 ! Ai be the projection and ◆i : Ai ! A1 A2 be the
embedding. Then
↵: HomR (A1 A2 , B) ! HomR (A1 , B) HomR (A2 , B)
f 7 ! (f ◆1 , f ◆2 )
and
: HomR (A1 , B) HomR (A2 , B) ! HomR (A1 A2 , B)
(f1 , f2 ) 7 ! f 1 ⇡1 + f 2 ⇡2
are E-maps which are inverses of each other.
Let R B be such that R B ⇠= R B R B. (For example, let B be a direct sum of
infinitely many copies of R.) Then we have an E-module isomorphism
E = HomR (B, B) ⇠ = HomR (B B, B) ⇠ = HomR (B, B) HomR (B, B) = E E.
 2.7. PROJECTIVE AND INJECTIVE MODULES 59

It follows that E ⇠
= E n for every integer n > 0.
Q1
Example (A direct product that is not free). We show that i=1 Z is not a
free Z-modules. Let
n 1
Y o
A = (a1 , a2 , . . . ) 2 Z : for every k > 0, 2k | ai for almost all i .
i=1
Q1
We claim that A is not free. (Then by Theorem 2.36, i=1 Z is not free.) Clearly,
|A| 2@0 > @0 . Assume to the contrary Lthat A is free. Then rank A > @0 . Every
1
coset of 2A in A contains an element in i=1 Z, hence A/2A is countable. Thus
dimZ2 (A/2A)  @0 . However, rank A = dimZ2 (A/2A), which is a contradiction.

2.7. Projective and Injective Modules

Exact sequence. A sequence of R-modules and R-maps
fi 1 fi fi+1
· · · ! Ai 1 ! Ai ! Ai+1 ! · · ·
f g
is called exact if im fi 1 = ker fi for all i. An exact sequence 0 ! A ! B ! C ! 0
f g
is called a short exact sequence. Two short exact sequences 0 ! A ! B ! C ! 0
0 0
f g
and 0 ! A0 ! B 0 ! C 0 ! 0 are called isomorphic if there exist isomorphisms
↵, , such that the diagram
f g
0 ! A
? ! B
? ! C
? ! 0
? ? ?
↵y y y
f0 g0
0 ! A0 ! B0 ! C0 ! 0
commutes.
f g
Proposition 2.33. Let 0 ! A ! B ! C ! 0 be a short exact sequence of
R-modules. Then the following statements are equivalent.
(i) There is an R-map h : C ! B such that g h = idC .
(ii) There is an R-map k : B ! A such that k f = idA .
f g ◆ ⇡
(iii) 0 ! A ! B ! C ! 0 is isomorphic to 0 ! A !
1
A C !2 C ! 0.
f g
If (i) – (iii) are satisfied, the short exact sequence 0 ! A ! B ! C ! 0 is said to
be split.
Proof. (i) ) (iii). The diagram
f g
............................................... ............................................... ............................................... ...............................................
0 A B .......... .... .... .... .... .... C 0
.. ..
h ..
....... ....... .......
..... ..... .....
.. .. ..
idA ..... ...
...
... id
.... C
..... ... ..
.. ... ...
... ... ...
◆1 ⇡2
0 ................................................ A ...........................................A C ........................................... C ................................................ 0
commutes, where
: A C ! B
(a, c) 7 ! f (a) + h(c).
By the five lemma (the next lemma), is an isomorphism.
60 2. RINGS AND MODULES

(ii) ) (iii). The diagram

f g
............................................... ................................................ ............................................... ...............................................
0 A ........... .... .... .... .... .... B C 0
... k ... ...
... ... ...
... ... ...
... ... ...
. ... ...
idA ...... ...
...
... idC
...
... ... ...
..
......... ......... .........
. . .
◆1 ⇡2
0 ............................................... A ...........................................
A C ........................................... C ...............................................
0
commutes, where
: B ! A C
b 7 ! k(b), g(b) .
By the five lemma, is an isomorphism.
(iii) ) (i) and (ii).
◆ 1 ⇡2
.......................................... .....
0 ................................................
A ............. .... .... .... ....
⇡1
A C ............................◆........................................ C ................................................
0
2
... ... ...
... ... ...
... ... ...
... ... ...
. ... ...
↵ ...... ...
...
...
...
... ... ...
..
......... ......... .........
. . .
f g
............................................... ............................................... ............................................... ...............................................
0 A .......... .... .... .... .... .... B .......... .... .... .... .... .... C 0
k h

In the above diagram, let k = ↵ ⇡1 1

and h = ◆2 1
. ⇤

Lemma 2.34 (The five lemma). Let

f1 f2 f3 f4
A?1 ! A?2 ! A?3 ! A?4 ! A?5
? ? ? ? ?
↵1 y ↵2 y ↵3 y ↵4 y ↵5 y
g1 g2 g3 g4
B1 ! B2 ! B3 ! B4 ! B5
be a commutative diagram of R-modules with exact rows.
(i) If ↵1 is surjective and ↵2 , ↵4 are injective, then ↵3 is injective.
(ii) If ↵5 is injective and ↵2 , ↵4 are surjective, then ↵3 is surjective.
(Note: surjective = onto, injective = one-to-one.)
Proof. (i) Let a3 2 ker ↵3 . Then ↵4 f3 (a3 ) = g3 ↵3 (a3 ) = 0. Since ↵4 is
injective, f3 (a3 ) = 0, whence a3 = f2 (a2 ) for some a2 2 A2 . Let b2 = ↵2 (a2 ). Then
g2 (b2 ) = ↵3 (a3 ) = 0, so b2 = g1 (b1 ) for some b1 2 B1 . Let a1 2 A1 be such that
↵1 (a1 ) = b1 . Then ↵2 (a2 f1 (a1 )) = ↵2 (a2 ) ↵2 f1 (a1 ) = b2 g1 ↵1 (a1 ) = b2 b2 =
0, whence a2 = f1 (a1 ). Therefore, a3 = f2 (a2 ) = 0.
(ii) Let b3 2 B3 . Then g3 (b3 ) = ↵4 (a4 ) for some a4 2 A4 . Since ↵5 f4 (a4 ) =
g4 ↵4 (a4 ) = g4 g3 (b3 ) = 0, we have f4 (a4 ) = 0, so a4 = f3 (a3 ) for some a3 2 A3 .
Since g3 (b3 ↵3 (a3 )) = ↵4 (a4 ) g3 ↵3 (a3 ) = ↵4 (a4 ) ↵4 f3 (a3 ) = ↵4 (a4 ) ↵4 (a4 ) =
0, b3 ↵3 (a3 ) = g2 (b2 ) for some b2 2 B2 . Let a2 2 A2 be such that b2 = ↵2 (a2 ).
Then ↵3 (a3 + f2 (a2 )) = ↵3 (a3 ) + ↵3 f2 (a2 ) = ↵3 (a3 ) + g2 ↵2 (a2 ) = ↵3 (a3 ) + g2 (b2 ) =
b3 . ⇤

Projective modules. An R-module P is called projective if for every surjective

R-map p : A ! B and R-map f : P ! B, there exists an R-map g : P ! A such
 2.7. PROJECTIVE AND INJECTIVE MODULES 61

that the diagram

P
.... ...
.. ...
g .... .. ...
...
.... ... f
.... ..
.... ........
....... .
p
.................................... ....................................
A B 0
commutes. Free modules are projective.

Theorem 2.35 (Characterizations of projective modules). Let P be an R-

module. The following statements are equivalent.
(i) P is projective.
i p
(ii) Every short exact sequence 0 ! A ! B ! P ! 0 is split.
(iii) There exists an R-module K such that K P is free.

Proof. (i) ) (ii).

P
.... ...
...
g .... ...
.... ...
.... ... id
.... ..
.... .......
........ ..
i p
.................................... .................................... .................................... ....................................
0 A B P 0
(ii) ) (iii). There is a surjection (surjective R-map) p : F ! P , where F is a
p
free R-module. Since the short exact sequence 0 ! ker p ,! F ! P ! 0 is split,
we have F ⇠ = ker p P .

(iii) ) (i). Consider the following diagram.

F =K P
... ...........
..
.. .... ....
... ⇡ .... .... ◆
.. ... ..
.. ........ ....
. ..
...
g1 ..
..
P
... .... ....
.. g... ...
. ....
. ....
... .... ... f
... .... ..
................... ........
.
p
.................................... ....................................
A B 0
Since F is projective, there exists an R-map g1 : F ! A such that pg1 = f ⇡. Let
g = g1 ◆. Then pg = pg1 ◆ = f ⇡◆ = f . ⇤
L
Fact. Let Ai (i 2 I) be R-modules. Then i2I Ai is a projective R-module if
and only if each Ai is a projective R-module.

Proof. This follows from Theorem 2.35 (iii). ⇤

Pullback. Let

A
?
?
(2.4) yf
B ! C
g
62 2. RINGS AND MODULES

be a diagram of R-modules. Define D = {(a, b) 2 A ⇥ B : f (a) = g(b)}, ↵ : D ! A,

(a, b) 7! a, and : D ! B, (a, b) 7! b. Then
↵
D
? ! A
?
? ?
y yf
B ! C
g

is a commutative diagram of R-modules. (D, ↵, ) is called the pullback of (2.4). If

g is onto, then ↵ is onto. In fact, for a 2 A, there exists b 2 B such that f (a) = g(b).
Then (a, b) 2 D and a = ↵(a, b). The map (↵, ) : D ! A ⇥ B is one-to-one. The
pullback (D, ↵, ) has the following universal mapping property: If
↵1
D?1 ! A
?
? ?
1y yf
B ! C
g

is another commutative diagram, then there exists a unique R-map : D1 ! D

such that the following diagram commutes.
D. 1..................
... ..... .........
... ...... .........↵1
... ..... .........
... ..... .........
..... .........
... ..... .........
... . . ......... .
... ......... ...........
...
.................................
...
...
...
D ↵ A
... ... ...
.. ... ...
1 .... ... ...
... ... ... f
...
... .. ..
... ...... ........
... ... .
........
.
................................
B g C
In Theorem 2.35, (ii) ) (i) can also be proved using a pullback:
..
... ↵
...............................................
0 ...........................................
ker ↵ .............................................. D .......... .... .... .... .... .... P ...............................................
0
... .... ...
... ...
... .... ...
... .... ...
... .... ...
... .... ... f
...
... .... ...
.... ..
....... .... ........
... ........ ...
p
............................................... ...............................................
A B 0
Note that ↵ is onto since p is onto.
Example. Let R = Z6 . The R-module Z3 is projective (Z3 Z2 ⇠
= R) but not
free.
Theorem 2.36. Let F be a free module over a PID R and A be a submodule
of F . Then A is free with rank A  rank F .
Proof. Let X be a basis of F . Let
Y = {(f, Z, Y ) : Z ⇢ Y ⇢ X, f : Z ! A\hY i one-to-one, f (Z) is a basis of A\hY i}.
For (f1 , Z1 , Y1 ), (f2 , Z2 , Y2 ) 2 Y, define (f1 , Z1 , Y1 ) (f2 , Z2 , Y2 ) if Y1 ⇢ Y2 , Z1 ⇢
Z2 and f2 |Z1 = f1 . Then (Y, ) is a nonempty poset in which every chain has
an upper bound. By Zorn’s lemma, (Y, ) has a maximal element (f0 , Z0 , Y0 ). It
suffices to show Y0 = X.
Suppose to the contrary that Y0 6= X. Choose x0 2 X \ Y0 and let
I = {r 2 R : rx0 + y 2 A for some y 2 hY0 i}.
 2.7. PROJECTIVE AND INJECTIVE MODULES 63

Then I is an ideal of R, hence I = hsi for some s 2 R. If s = 0, then A\hY0 [{x0 }i =

A\hY0 i, whence (f0 , Z0 , Y0 [{x0 }) 2 Y and (f0 , Z0 , Y0 [{x0 }) ⌫ (f0 , Z0 , Y0 ), which
is a contradiction. Thus s 6= 0. Let u 2 A be such that u = sx0 +y for some y 2 hY0 i.
We claim that
(2.5) A \ hY0 [ {x0 }i = A \ hY0 i hui.
First we show that A \ hY0 [ {x0 }i = A \ hY0 i + hui. If w 2 A \ hY0 [ {x0 }i, then
w = tx0 + z for some z 2 hY0 i and t 2 R with s | t. Now we have w (t/s)u 2
A \ hY0 i, hence w 2 A \ hY0 i + hui. Next we show that hY0 i \ hui = {0}. If
au = y 0 for some a 2 R and y 0 2 hY0 i, then a(sx0 + y) = y 0 , so a = 0. Thus
A \ hY0 i + hui = A \ hY0 i hui, and claim (2.5) is proved. Now f0 (Z0 ) [ {u}
is a basis of A \ hY0 [ {x0 }i. Extend f0 : Z0 ! A \ hY0 i to g : Z0 [ {x0 } !
A \ hY0 [ {x0 }i by setting g(x0 ) = u. Then (g, Z0 [ {x0 }, Y0 [ {x0 }) 2 Y and
(g, Z0 [ {x0 }, Y0 [ {x0 }) ⌫ (f0 , Z0 , Y0 ), which is a contradiction. ⇤
Remark. If rank F < 1, Theorem 2.36 can be proved by induction on rank F ;
the argument is similar to the above proof but Zorn’s lemma is not needed (Exer-
cise ??).
Theorem 2.37. Every projective module over a PID is free.
Proof. Let P be a projective module over a PID R. By Theorem 2.35 (iii),
P is a submodule of a free R-module. By Theorem 2.36, P is free. ⇤
Theorem 2.38 ([2, 24, 29]). Let k be a field. Then every projective module
over k[x1 , . . . , xn ] is free.
In Theorem 2.38, the case where the projective module is not finitely generated
was proved by Bass [2]; the case where the projective module is finitely generated
is known as Serre’s conjecture and the Quillen-Suslin theorem. See [22, Ch. III]
for some elementary proofs of Serre’s conjecture.
Projective modules over a local ring.
Theorem 2.39 (Kaplansky [20]). Every projective module over a local ring
(not necessarily commutative) is free.
Lemma 2.40. If A is a direct sum of countably generated R-modules and B is
a direct summand of A, then B is a direct sum of countably generated R-modules.
L
Proof. Let A = i2I Ai , where
P each Ai is countably generated. Let A =
B C. For each J ⇢ I, put AJ = i2J Ai . Let
X = (J, L) : J ⇢ I, AJ = AJ \ B + AJ \ C, L is a family of countably
L
generated submodules of B such that AJ \ B = L2L L .
For (J1 , L1 ), (J2 , L2 ) 2 X , define (J1 , L1 ) (J2 , L2 ) if J1 ⇢ J2 and L1 ⇢ L2 . Then
(X , ) is a nonempty poset in which S everySchain has an upper bound. In fact, if
(Jj , Lj ) is a chain in (X , ), then ( j Jj , j Lj ) 2 X . By Zorn’s lemma, (X , )
has a maximal element (J0 , L0 ).
We claim that J0 = I. (Then the conclusion of the lemma follows.) Assume to
the contrary that there exists i1 2 I \ J0 . Let J1 = {i1 } and AJ1 = hx11 , x12 , . . . i.
Write x1j = x01j + x001j , where x01j 2 B, x001j 2 C. Each x01j (x001j ) is contained
S1
in AJ for some finite J ⇢ I, hence j=1 {x01j , x001j } ⇢ AJ2 for some countable
64 2. RINGS AND MODULES

J2 ⇢ I. Write AJ2 = hx21 , x22 , . . . i, x2j = x02j + x002j , x02j 2 B, x002j 2 C. Then
S1
j=1 {x2j , x2j } ⇢ AJ3 for some countable J3 ⇢ I. In general, we have Ji ⇢ I with
0 00

|Ji |  @0 such that

AJi ⇢ AJi+1 \ B + AJi+1 \ C.
S1
Let J = i=0 Ji . Then
⇤

AJ ⇤ ⇢ AJ ⇤ \ B + AJ ⇤ \ C.
Since AJ0 \ B is a direct summand of AJ0 and AJ0 is a direct summand of A,
AJ0 \ B is a direct summand of A. Hence AJ0 \ B is a direct summand of AJ ⇤ \ B
(Exercise ??). Since AJ ⇤ = AJ ⇤ \ B AJ ⇤ \ C and AJ0 = AJ0 \ B AJ0 \ C, we
have
AJ ⇤ ⇠ AJ ⇤ \ B AJ ⇤ \ C
= .
AJ 0 AJ 0 \ B AJ 0 \ C
Thus (AJ ⇤ \ B)/(AJ0 \ B) is a homomorphic image of AJ ⇤ /AJ0 . Since AJ ⇤ is
countably generated, so is (AJ ⇤ \ B)/(AJ0 \ B). We have
AJ ⇤ \ B = (AJ0 \ B) L,
where L ⇠
= (AJ ⇤ \ B)/(AJ0 \ B) is countably generated. Thus (J ⇤ , L0 [ {L}) 2 X ,
which contradicts the maximality of (J0 , L0 ). ⇤
Proof of Theorem 2.39. Let R be a local ring with maximal ideal m (=
R \ R⇥ ). Let P be a projective module over R.
1 We claim that every x 2 P is contained in a free direct summand of P . There
exists an R-module Q such that F := P Q is free. Let U be a basis of F . Assume
x 6= 0 and write x = a1 u1 + · · · + an un , ai 2 R, u1 , . . . , un 2 U distinct. Assume
that U is chosen such that n is as small as possible. Then for each 1  i  n,
(2.6) ai 2
/ a1 R + · · · + ai 1R + ai+1 R + · · · + an R.
(If an = a1 b1 + · · · + an 1 bn 1 for some b1 , . . . , bn 1 2 R, then x = a1 (u1 + b1 un ) +
· · · + an 1 (un 1 + bn 1 un ). Note that {u1 + b1 un , . . . , un 1 + bn 1 un , un } [ U 0 is
a basis of F , where U 0 = U \ {u1 , . . . , un }. This contradicts the minimality of n.)
Write ui = yi + zi , yi 2 P , zi 2 Q. Then
(2.7) a 1 u 1 + · · · + a n u n = a 1 y1 + · · · + a n yn .
Write
2 3 2 3
y1 u1
6.7 6 . 7
(2.8) 6 . 7⌘C6 . 7 (mod hU 0 i),
4.5 4 . 5
yn un
where C 2 Mn (R). By (2.7) and (2.8), we have
[a1 , . . . , an ] = [a1 , . . . , an ]C,
i.e., [a1 , . . . , an ](I C) = 0. By (2.6), all entries of I C are in m. Since R is
local, C is invertible in Mn (R). Thus by (2.8), {y1 , . . . , yn } [ U 0 is a basis of F .
Let Y = hy1 , . . . , yn i. Then x 2 Y and Y is free and is a direct summand of F . By
Exercise ??, Y is a direct summand of P .
2 Since P is a direct summand of a free R-module, by Lemma 2.40, P is a
direct sum of countably generated R-modules. Thus we may assume that P is
countably generated.
 2.7. PROJECTIVE AND INJECTIVE MODULES 65

Let P = hx1 , x2 , . . . i. By 1 , P = F1 P1 , where F1 is free and x1 2 F1 .

Write x2 = x02 + x002 , where x02 2 F1 , x002 2 P1 . By 1 again, P1 = F2 P2 , where
F2 is free and x002 2 F2 . Write x3 = x03 + x003 , x03 2 F1 F2 , x003 2 P2 , ... Then
P = F1 F 2 · · · . ⇤
Injective modules. An R-module E is called injective if for every injective
R-map i : A ! B and every R-map f : A ! E, there exists an R-map g : B ! E
such that the diagram
i
0 .................................... A .................................... B
... ....
...
... ....
....
f ...... .
... g
.. ....
....... ..
. .........

E
commutes.
Q
Fact. Let {Ei : i 2 I} be a family of R-modules. Then i2I Ei is injective if
and only if Ei is injective for all i 2 I.
Proof. ())
....................................
◆
....................................
0 A B
... .... ...
...
... gi.... .... .....
. .
f ...... .... ....
.. .... .
...
....... .......
.. ..
... ...
...
..
E i
.
... g gi = ⇡i g.
...
... ... ..
.......... ....
.
.. .
◆i ...... ......⇡i.....
... .. ...
..
............ .........
Q . .. ...

j2I Ej
(()
.....................................
◆
.....................................
0 A B
... .... ....
...
... .... ...
f ...... h... .... .....
. ..
.. .... ...
....... .... ...
Q ... ....... ...
...
j2I j...... hi E .. h(b) = (hi (b))i2I , b 2 B.
... ..
... ...
... ...
⇡i ...... ......
... ...
....... ....... .
. ..

Ei
⇤
Pushout. Let
f
A
? ! B
(2.9) ?
gy

C
be a diagram of R-modules. Let S = (f (a), g(a)) : a 2 A ⇢ B C, D =
(B C)/S, ↵ : B ! D, b 7! (b, 0) + S, and : C ! D, c 7! (0, c) + S. Then
f
A
? ! B
?
? ?
gy y↵
C ! D
66 2. RINGS AND MODULES

is a commutative diagram of R-modules. (↵, , D) is called the pushout of (2.9). If

f is one-to-one, then so is . If c 2 ker , then (0, c) 2 S, i.e., (0, c) = (f (a), g(a))
for some a 2 A. Thus f (a) = 0, hence a = 0 and c = g(a) = 0. The map
↵⇡1 + ⇡2 : B ⇥ C ! D is onto. The pushout (↵, , D) has the following universal
mapping property: If
f
A
? ! B
?
? ?
gy y↵1
C ! D1
1

is another commutative diagram, then there exists a unique R-map : D ! D1

such that the following diagram commutes.
f
.................................
A B ...
.
... ... .....
... ... ...
. ... ...
g ...... ↵ .. .
. ...
...
. . ...
......... ......... ...
. . ... ↵1
...
...
................................ ...
C ........
D .....
.....
...
...
........ ..... ...
........ ..... ...
........ .
........ .....
........
........ ..... ......
........ . .
........ ........
.
1 ........
........
.............
.D1
Proposition 2.41 (Characterizations of injective modules). Let E be an R-
module. The following statements are equivalent.
(i) E is injective.
i p
(ii) Every short exact sequence 0 ! E ! A ! B ! 0 is split.
(iii) If E is a submodule of A, then A = E B for some submodule B of A.
Proof. (i) ) (ii).
i p
..................................... ..................................... ..................................... .....................................
0 E A B 0
... ....
...
... ....
. ....
id ...... .... g
. ....
....... .......
.
... ...

E
(ii) ) (i). Use a pushout
...............................................
i
...............................................
0 A B
... .... ...
... ...
... .... ...
... .... ...
. .... ...
f ...... .... ... ↵
... .... ...
.. .... ..
....... .... ........
. ....... .

0 ...............................................
E
.......... .... .... .... .... ....
................................................ D ........................................ coker ........................................
0

where coker = D/im . Note that since i is one-to-one, so is .

(ii) ) (iii). 0 ! E ,! A ! A/E ! 0 is split.
(iii) ) (ii). Obvious. ⇤
Remark. Theorem 2.45 also provides a quick proof of (iii) ) (i).
Theorem 2.42 (Baer’s criterion). An R-module E is injective if and only if
for any left ideal L of R and R-map ↵ : L ! E, ↵ can be extended to an R-map
: R ! E.
 2.7. PROJECTIVE AND INJECTIVE MODULES 67

Proof. (() Consider a diagram

i
0 ! A
? ! B
?
fy

E
We may assume that A ⇢ B and i is the inclusion. Let
S = {(C, h) : A ⇢ R C ⇢ B, h : C ! E is an R-map, h|A = f }.
For (C1 , h1 ), (C2 , h2 ) 2 S, define (C1 , h1 ) (C2 , h2 ) if C1 ⇢ C2 and h2 |C1 = h1 .
Then (S, ) is a nonempty poset in which every chain has an upper bound. By
Zorn’s lemma, (S, ) has a maximal element of (C0 , h0 ). It remains to show that
C0 = B.
Assume to the contrary that there exists b 2 B \C0 . Let L = {r 2 R : rb 2 C0 }.
Then L is a left ideal of R, and ↵ : L ! E, r 7! h0 (rb) is an R-map. Extend ↵ to
an R-map : R ! E. Define
h1 : C0 + Rb ! E
c + rb 7 ! h0 (c) + (r).
We claim that h1 is a well-defined R-map. If c + rb = c0 + r0 b, where c, c0 2 C0
and r, r0 2 R, then (r r0 )b = c0 c 2 C0 . Thus h0 (c0 c) = h0 ((r r0 )b) =
↵(r r0 ) = (r r0 ), and hence h1 is well-defined. Note that (C0 + Rb, h1 ) 2 S
and (C0 + Rb, h1 ) ⌫ (C0 , h0 ), which is a contradiction. ⇤
Divisible modules. Let R be an integral domain. An R-module D is called
divisible if for each y 2 D and 0 6= r 2 R, there exists x 2 D such that rx = y. D
is divisible if and only if rD = D for all 0 6= r 2 R.
Fact. Let D, Di (i 2 I) be modules over an integral domain R.
L
(i) Di is divisible for all i 2 I if and only if i2I Di is divisible.
(ii) If D is divisible and E ⇢ D is a submodule, then D/E is divisible.
(iii) If D is injective, then D is divisible.
Proof. (iii) Let y 2 D and 0 6= r 2 R. Consider the diagram
....
0 ....................................
Rr .................................... R
... ....
... ....
...
....
f ...... .... g
.. ....
........ ....
. .......

D
where f (r) = y. Then rg(1) = g(r) = f (r) = y. ⇤
Remark. A divisible module over an integral domain is not necessarily injec-
tive (Exercise ??).
Proposition 2.43. Let D be a module over a PID R. Then D is injective if
and only if D is divisible.
Proof. (() Let I 6= 0 be an ideal of R and f : I ! D be an R-map. We
have I = hai for some 0 6= a 2 R. Since D is divisible, there exists x 2 D such that
ax = f (a). Define g : R ! D, r 7! rx. Then g is an R-map and g|I = f . By Baer’s
criterion, D is injective. ⇤
68 2. RINGS AND MODULES

Proposition 2.44. Every abelian group A can be embedded in a divisible

abelian group.
L L L
Proof. A ⇠= ( i2I Z)/K ,! ( i2I Q)/K, where ( i2I Q)/K is divisible.
⇤

Theorem 2.45. Every R-module A can be embedded in an injective R-module.

Proof. By Proposition 2.44, there exists a Z-module embedding f : A ! B,
where B is a divisible abelian group. Then we have R-module embeddings
f¯
A ! HomZ (Z RR , Z A) ! HomZ (Z RR , Z B),
where for a 2 A and ↵ 2 HomZ (Z RR , Z A),
(a) : R ! A f¯(↵) : R ! B
r 7 ! ra, r 7 ! f (↵(r)).
By the next lemma, HomZ (Z RR , Z B) is an injective R-modules. ⇤

Lemma 2.46. Let R be a ring and B be a divisible abelian group. Then

HomZ (Z RR , Z B) is an injective R-module.
Proof. Let L be a left ideal of R and f : L ! HomZ (R, B) an R-map. Let
g: L ! B
x 7 ! f (x)(1R ).
Then g is a Z-map, so it extends to a Z-map ḡ : R ! B since B is an injective
Z-module. For each r 2 R, define
h(r) : R ! B
y 7 ! ḡ(yr).
Then h(r) 2 HomZ (R, B), h : R ! HomZ (R, B) is an R-map and h|L = f . By
Baer’s criterion, HomZ (R, B) is an injective R-module. ⇤

2.8. Chain Conditions

Let R A be an R-module. Two finite descending (or ascending) sequences of
submodules
A = A0 A1 ··· An = {0}
A= A00 A01 ··· A0m = {0}
are called equivalent if there is a bijection between {Ai 1 /Ai : 1  i  n, Ai 1 )
Ai } and {A0j 1 /A0j : 1  j  m, A0j 1 ) A0j } such that the corresponding factors
are isomorphic. A descending sequence A = A0 A1 · · · An = {0} is called a
composition series of A if Ai 1 /Ai is simple for all 1  i  n.
Theorem 2.47 (Schreier). Any two finite descending (or ascending) sequences
of submodules of a module R A have equivalent refinements.
Theorem 2.48 (Jordan-Hölder). Any two composition series of a module RA
are equivalent.
 2.8. CHAIN CONDITIONS 69

The proofs of Theorems 2.47 and 2.48 are the same as the proofs in the group
case; see Theorems 1.38 and 1.40.
In general, an infinite sequence (of any objects) A1 , A2 , . . . is said to stabilize
if there exists n such that An = An+1 = · · · .
ACC and DCC. An R-module A is said to have the ascending chain condition
(ACC) if every ascending sequence of submodules of A stabilizes. A is said to have
the descending chain condition (DCC) if every descending sequence of submodules
of A stabilizes.
Examples. Z as a Z-module has ACC but no DCC. Let p be a prime and let
Z(p1 ) be the subgroup of Q/Z defined by
na o
Z(p1 ) = + Z 2 Q/Z : a, b 2 Z, b = pi for some i 0 .
b
Then every proper subgroup of Z(p1 ) is generated by 1/pi + Z for some i 0.
Since D1 E D1 E
0= 0
+Z ( 1
+ Z ( ··· ,
p p
Z(p1 ) as a Z-module has DCC but no ACC.
Proposition 2.49. Let A be an R-module.
(i) A has ACC , every nonempty family of submodules of A contains a
maximal element , every submodule of A is finitely generated.
(ii) A has DCC , every nonempty family of submodules of A contains a
minimal element.
Proof. We only prove that if every submodule of A is finitely generated, then
A has ACC. The other claims in the proposition are obviously true.
S1 Let A0 ⇢ A1 ⇢ · · · be an ascending S sequence of submodules of A. Then
1
A
i=0 i = (a 1 , . . . , a )
Sk1 for some a 1 , . . . , a k 2 i=0 Ai . Choose n such that a0 , . . . , ak
2 An . Then An = i=0 Ai , hence An = An+1 = · · · . ⇤
Proposition 2.50. A module R A has a composition series if and only if A has
both ACC and DCC.
Proof. ()) Assume that A has a composition series with n+1 terms. Assume
to the contrary that A does not have ACC or DCC. Then there is a squence of
submodules of A:
A = A0 ) A1 ) · · · ) An+1 = {0}.
Any refinement of this sequence has at least n + 1 nonzero factors hence cannot be
equivalent to the composition series of A. This is a contradiction to Theorem 2.47.
(() We construct a composition series A = A0 A1 · · · as follows. Let
A0 = A. If A0 6= 0, since A has ACC, among all proper submodules of A0 , there
is a maximal one, say, A1 . Clearly, A0 /A1 is simple. By induction, there are
submodules A0 A1 A2 · · · such that Ai /Ai+1 is simple for all i and Ai+1 is
defined whenever Ai 6= 0. Since A has DCC, the above descending sequence must
stop at An , i.e., An = 0. Now, A = A0 A1 ··· An = 0 is a composition
series of A. ⇤
Definition 2.51. A ring R is called left (right) noetherian if the module R R
(RR ) has ACC; R is called left (right) artinian if the module R R (RR ) has DCC;
R is called noetherian (artinian) if it is both left and right noetherian (artinian).
70 2. RINGS AND MODULES

The Hopkins-Levitzki theorem (Theorem 4.26). A left (right) artinian ring

is left (right) noetherian.
A proof is given in §4.3.
Theorem 2.52 (Hilbert basis theorem). If R is a left (right) noetherian ring,
then so is R[x1 , . . . , xn ].
Proof. We only have to show that R[x] is left noetherian. Assume to the
contrary that there exists a left ideal I of R[x] which is not finitely generated. Let
f0 2 I \ {0} be of the smallest degree. Then I 6= (f0 ). Let f1 2 I \ (f0 ) be of the
smallest degree. In general, let fn+1 2 I \ (f0 , . . . , fn ) be of the smallest degree.
Let dn = deg fn . Then d0  d1  · · · . Let an be the leading coefficient of fn
and consider the ascending sequence (a0 ) ⇢ (a0 , a1 ) ⇢ · · · of R R. Since R is left
noetherian, there exists m such that (a0 , . . . , am ) = (a0 , . . . , am , am+1 ), hence
am+1 = r0 a0 + · · · + rm am , ri 2 R.
Let
m
X
f = fm+1 ri fi (x)xdm+1 di
.
i=0
Then f 2 I \ (f0 , . . . , fm ) and deg f < dm+1 , which is a contradiction. ⇤
i p
Proposition 2.53. Let 0 ! A ! B ! C ! 0 be an exact sequence of R-
modules. Then B has ACC (DCC) if and only if both A and C have ACC (DCC).
Proof. B has ACC ) A and C have ACC.
Let A1 ⇢ A2 ⇢ · · · be an ascending sequence of submodules of A. Then i(A1 ) ⇢
i(A2 ) ⇢ · · · is an ascending sequence of submodules of B, hence it stabilizes. Thus
the sequence A1 ⇢ A2 ⇢ · · · also stabilizes.
Let C1 ⇢ C2 ⇢ · · · be an ascending sequence of submodules of C. Then
p 1 (C1 ) ⇢ p 1 (C2 ) ⇢ · · · is an ascending sequence of submodules of B, so it
stabilizes. Since Ci = p(p 1 (Ci )), the sequence C1 ⇢ C2 ⇢ · · · also stabilizes.
A and C have ACC ) B has ACC.
Let B1 ⇢ B2 ⇢ · · · be an ascending sequence of submodules of B. Then there
exists n > 0 such that for all k > 0, p(Bn ) = p(Bn+k ) and i 1 (Bn ) = i 1 (Bn+k ).
We have a commutative diagram
i p
............................................... 1 ............................................... Bn ............................................... p(Bn ) ...............................................
0 i (Bn ) 0
... ..... ...
... ..... .. ...
... ... ...
... ... ...
... ... ...
... id ... ... id
... ... ...
... ... ...
....... ........ .......
... . ...
i p
0 ............................................... i 1(Bn+k) ...............................................Bn+k............................................... p(Bn+k ) ...............................................
0
with exact rows. By the five lemma, Bk = Bn+k . ⇤
Proposition 2.54. Let R be a left noetherian (artinian) ring. Then every
finitely generated R-module A has ACC (DCC).
Proof. We have A ⇠ = Rn /K for some n 2 N and R K ⇢ Rn . Since R has ACC,
by Proposition 2.53, R and Rn /K have ACC.
n
⇤
i p
Proposition 2.55. Let 0 ! A ! B ! C ! 0 be an exact sequence of R-
modules.
 2.9. FINITELY GENERATED MODULES OVER A PID 71

(i) Assume that A = hXi and C = hY i for some X ⇢ A and Y ⇢ C. Choose

Z ⇢ B such that p(Z) = Y . Then B = hi(X) [ Zi. In particular, if A
and C are finitely generated, then B is finitely generated.
(ii) If R is left noetherian, then B is finitely generated if and only if both A
and C are finitely generated.
Proof. (ii) ()) By Proposition 2.56 (i), A is finitely generated. ⇤

Proposition 2.56. Let R be a left noetherian ring and M be a finitely gener-

ated R-module.
(i) Every submodule of M is finitely generated.
(ii) If R is a PID and M is generated by n elements, then every submodules
of M can be generated by  n elements.
Proof. (i) Let M = hx1 , . . . , xn i and let S be a submodule of M . Use induc-
tion on n.
If n = 1, M = hx1 i ⇠
= R/I for some left ideal I of R. Then S ⇠ = J/I for some
left ideal J of R with J I. Since R is left noetherian, J is fnitely generated, and
so is J/I.
Assume that n > 1. Let M1 = hx1 , . . . , xn 1 i, and consider the exact sequence
0 ! S \ M1 ! S ! S/(S \ M1 ) ! 0.
Since S \M1 ⇢ M1 , by the induction hypothesis, S \M1 is finitely generated. Since
S/(S \ M1 ) ⇠
= (S + M1 )/M1 ⇢ M/M1 = hxn + M1 i, S/(S \ M1 ) is also finitely
generated. Thus S is finitely generated.
(ii) Same as the proof of (i): For n = 1, the conclusion holds since R is a PID.
For n > 1, by the induction hypothesis, S \M1 is generated by  n 1 elements and
S/(S \ M1 ) is cyclic. By Proposition 2.55 (i), S is generated by  n elements. ⇤

2.9. Finitely Generated Modules over a PID

Let R be a ring and A be an R-module. For X ⇢ A, define ann(X) = {r 2 R :
⇠ R/ann(x); x is called a torsion
rx = 0 for all x 2 X}. For x 2 A, we have Rx =
element if ann(x) 6= 0.
Modules over a quotient ring. Let R be a ring and I be an ideal of R. If A is
an R/I-module, then A is automatically an R-module with the scalar multiplication
ra := (r + I)a, r 2 R, a 2 A; submodules of R A are precisely submodules of R/I A.
If A is a left R-module such that I ⇢ ann(A), then A is also an R/I-module with
the scalar multiplication (r + I)a := ra, r 2 R, a 2 A.
Theorem 2.57 (Structure of finitely generated modules over a PID). Let A
be a finitely generated module over a PID R. Then
(2.10) A = Rz1 ··· Rzs ,
where
(2.11) R 6= ann(z1 ) ··· ann(zs ).
Moreover, ann(z1 ), · · · , ann(zs ) are uniquely determined by (2.10) and (2.11). (Note
that Rzi ⇠
= R/ann(zi ).)
72 2. RINGS AND MODULES

In the above theorem, assume that ann(zi ) = (di ), 1  i  s, dt 6= 0, dt+1 =

· · · = ds = 0. Write
di = pe1i1 · · · pekik , 1  i  t,
where p1 , . . . , pk 2 R are distinct irreducibles and eij 2 N. Then
h M i
e
(2.12) A⇠ = R/(d1 ) · · · R/(dt ) Rs t ⇠ = R/(pj ij ) Rs t .
1it
1jk

The integer s t is called the rank of A; d1 , . . . , dt are called the invariant factors
e
of A; pj ij with eij > 0 are called the elementary divisors of A. The list of invariant
factors determines the list of elementary divisors and vice versa. Two finitely gen-
erated modules over a PID are isomorphic if and only if they have the same rank
and the same invariant factors (elementary divisors).
Proof of Theorem 2.57. Existence of the decomposition of (2.12).
1 Let Ator = {a 2 A : ua = 0 for some 0 6= u 2 R}. Then A/Ator is torsion-
free. By the next lemma, A/Ator ⇠
= Rr for some r 2 N. Thus the exact sequence
0 ! Ator ,! A ! A/Ator ! 0 is split, hence
A⇠= Ator (A/Ator ).

2 For each irreducible p 2 R, let

A(p) = {a 2 A : pn a = 0 for some n > 0}.
Then M
Ator = A(p),
p
where the sum is over finitely many irreducibles p 2 R.
3 Assume that pn A(p) = 0 but pn 1 A(p) 6= 0. Let a 2 A(p) be such that
pn 1
a 6= 0. Then Ra ⇠ = R/(pn ) (as R-modules and as R/(pn )-modules). Using
Baer’s criterion, it is easy to see that R/(pn ) is an injective R/(pn )-module (Exer-
cise ??). Since Ra is an injective submodule of A(p) (as R/(pn )-modules), we have
A(p) = Ra B for some R/(pn )-submodule B of A(p). (Note that B is automati-
cally an R-module.) Apply the same argument to B and continue on. Since A(p)
is finitely generated, it has ACC (Proposition 2.54). So eventually,
A(p) ⇠= R/(pn1 ) · · · R/(pnk ).

Uniqueness of the decomposition of (2.12). Let

hM⇣ ⌘i
A = Rr R/(pn(p,1) ) · · · R/(pn(p,ip ) ) ,
p

where p runs through all irreducibles of R, n(p, i), 1  i  ip , are positive integers,
and ip = 0 for almost all p. Then r = rank(A/Ator ). For every irreducible p 2 R
and every n 1,
dimR/(p) pn 1
A/pn A = r + {1  i  ip : n(p, i) n} .
Therefore, dimR/(p) (pn 1 A/pn A) dimR/(p) (pn A/pn+1 A) gives the number of times
pn appears in the list of elementary divisors of A. ⇤
Lemma 2.58. Let R be a PID. If A is a finitely generated torsion-free R-
module, then A is free.
 2.9. FINITELY GENERATED MODULES OVER A PID 73

Proof. Assume that A = hx1 , . . . , xn i. Let {y1 , . . . , ym } be a maximal linearly

independent subset of {x1 , . . . , xn }. Then for every 1  i  n, there exists 0 6= ai 2
R such that ai xi 2 hy1 , . . . , ym i. Let a = a1 · · · an . Then aA ⇢ hy1 , . . . , ym i ⇠
= Rm ,
hence aA is free. Since A is torsion-free, aA = A. ⇠ ⇤
There is another proof pf Theorem 2.57 using the Smith normal form of a matrix
over a PID; the invariant factors of the module A in the theorem are precisely the
invariant factors of an associated matrix over R.
Another proof of Theorem 2.57. Existence of decomposition (2.10).
Since A is finitely generated, we may assume that A = Rn /K, where K is a
submodule of Rn . Since R is a PID, by Proposition 2.56, K is finitely generated.
(In fact, by Theorem 2.36, K is free of rank m  n.) Let K = (f1 , . . . , fm ) and
write 2 3 2 3
f1 e1
6 . 7 6 7
6 . 7 = C 6 .. 7 ,
4 . 5 4.5
fm en
where e1 , . . . , en is the standard basis of Rn and C 2 Mm⇥n (R). There exist
P 2 GL(m, R) and Q 2 GL(n, R) such that
2 3
d1
6 .. 7
6 . 0 7
6
P CQ = 6 7,
7
4 dr 5
0 0
where d1 | d2 | · · · | dr 6= 0. (The above matrix is the Smith normal form of C and
d1 , . . . , dr are the invariant factors of C; see [18, §3.7].) Assume that d1 = · · · =
da = 1 and da+1 2 / R⇥ . Let
2 3 2 3 2 3 2 3
f1 f10 e1 e01
6 . 7 6 . 7 6 7
1 6 .. 7
6 7
P6 . 7 6 . 7 6 .. 7
4 . 5 = 4 . 5 and Q 4 . 5 = 4 . 5 .
0
fm fm en e0n
Then 2 3
2 3 d1 2 3
f10
6 7 e1
0
6 . 7 6 ..
6 . 7=6 . 07 6 .. 7
7 6
7
4 . 5 6 74 . 5,
0
4 dr 5 0
fm en
0 0
and K = (f10 , . . . , fm
0
) = (d1 e01 , . . . , dr e0r ). Since Rn = Re01 · · · Re0n and K =
Rd1 e1 · · · Rdn en , where di is defined to be 0 if i > r, we have
0 0

A = Rn /K = ⇠ Re0 /Rd1 e0 · · · Re0 /Rdn e0

1 1 n n
⇠
= R/(d1 ) ··· R/(dn )
= R/(da+1 ) ··· R/(dn ).
Let wi = 1 + (di ) 2 R/(di ), a + 1  i  n. Then R/(di ) = Rwi , ann(wi ) = (di ) and
A⇠= Rwa+1 · · · Rwn .
74 2. RINGS AND MODULES

Uniqueness of ann(z1 ), . . . , ann(zs ). Assume that

A = Rz1 ··· Rzs = Rw1 ··· Rwt ,
where R 6= ann(z1 ) · · · ann(zs ) and R 6= ann(w1 ) · · · ann(wt ). We show
that s = t and ann(zi ) = ann(wi ).
Without loss of generality, assume that s t. Let (w10 , . . . , ws0 ) = (0, . . . , 0, w1 , . . . , wt ).
Then
(2.13) A = Rz1 ··· Rzs = Rw10 ··· Rws0 ,
where ann(z1 ) · · · ann(zs ) and ann(w10 ) · · · ann(ws0 ). It suffices to show
that ann(zi ) = ann(wi0 ) for all 1  i  s.
First, ann(zs ) = ann A = ann(ws0 ). Let 1  i < s and let ann(zi ) = (di ). By
(2.13),
Rdi zi+1 · · · Rdi zs = di A Rdi wi0 · · · Rdi ws0 .
Hence 2 3 2 3
wi0 zi+1
6 . 7 6 . 7
di 6 . 7 6 . 7 C 2 M(s i+1)⇥(s i) (R).
4 . 5 = di C 4 . 5 ,
ws0 zs
There exists P 2 GL(s i + 1, R) such that P C = [ 0 ···⇤ 0 ], hence
2 3
2 3 2 3 ⇤
0
wi zi+1 6.7
6 . 7 6 . 7 6 .. 7
di P 6 . 7 6 . 7 6 7
4 . 5 = di P C 4 . 5 = 6 7 .
4⇤5
ws0 zs
0
⇤
Write P = [ pi ··· ps ]. Then
2 3
wi0
6 . 7
di [pi , . . . , ps ] 6 . 7
4 . 5 = 0.
ws0
It follows that di pj wj0 = 0, i  j  s, since Rwi0 · · · Rws0 is a direct sum. Thus
di pj 2 ann(wj0 ) ⇢ ann(wi0 ), i  j  s. Since P is invertible, gcd(pi , . . . , ps ) = 1,
whence di 2 ann(wi0 ). Thus ann(zi ) = (di ) ⇢ ann(wi0 ). By symmetry, ann(wi0 ) ⇢
ann(zi ). ⇤

Example. Let
2 3
3 1 1 0 1
6 2 0 0 2 47
6 7
C=6 7
4 3 1 11 0 15
2 0 0 2 4
and A = Z /{xC : x 2 Z }. The Smith normal
5 4
form of C is
2 3
1
6 2 7
6 7
6 7,
4 12 5
0 0
 2.9. FINITELY GENERATED MODULES OVER A PID 75

whence A ⇠
= Z2 Z12 Z2 . The elementary divisors of A are 2, 22 , 3 and rank A = 2.

The rational canonical form of a linear transformation. Let V be an

n-dimensional vector space over a field F and let T 2 EndF (V ). For f 2 F [x] and
v 2 V , define f v = f (T )v, then V becomes an F [x]-module. By Theorem 2.57,
V = F [x]z1 ··· F [x]zs ,
wherePann(zi ) = di are such that d1 | d2 | · · · | ds 6= 0 and deg d1 > 0. Let
ni
di = j=0 aij xj , ai,ni = 1. Then zi , xzi , . . . , xni 1 zi is an F -basis of F [x]zi , and
2 3 2 3
zi zi
6 7 6 7
6 xzi 7 6 xzi 7
T66 .. 7
7 = M (d )
i 6
6 . 7 7,
4 . 5 4 .. 5
x n i 1 zi x n i 1 zi
where 2 3
0 1
6 7
6 0 1 7
6 7
6 · · 7
M (di ) = 6
6
7
7
6 · · 7
6 7
4 0 1 5
ai,0 · · · · ai,ni 1
Ss
is the companion matrix of di . Now i=1 {x zi : 0  j  ni 1} is an F -basis of
j

V , and 2 3 2 3
z1 z1
6 .. 7 6 .. 7
6 . 7 6 . 7
6 7 2 366 7
6 n1 1 7 7
6x z1 7 M (d1 ) 6 x n 1 1 z1 7
6 7 6 7 6 7
6 .. 7 6 .. 76 .. 7
T6 . 7=4 . 566 . 7.
6 7 7
6 z 7 M (ds ) 6 zs 7
6
6 s 7 7
6 .. 7 6 .. 7
6 7 6 7
4 . 5 4 . 5
x n s 1 zs x ns 1 zs
The invariant factors d1 , . . . , ds of the F [x]-module V can be computed from
the matrix of T . Assume that T is defined by
2 3 2 3
✏1 ✏1
6.7 6.7
T 4 . 5 = A 4 .. 7
6 . 7 6
5,
✏n ✏n
where ✏1 , . . . , ✏n is a basis of V and A 2 Mn (F ). Define
: F [x]n ! V 2 3
✏1
6.7
(f1 , . . . , fn ) 7 ! (f1 , . . . , fn ) 6 .7
4 . 5.
✏n
76 2. RINGS AND MODULES

Then is an onto F [x]-map. Given (f1 , . . . , fn ) 2 F [x]n , by the division algorithm,

(f1 , . . . , fn ) = y(xI A) + (a1 , . . . , an ) for some y 2 F [x]n and (a1 , . . . , an ) 2 F n .
Then
2 3 2 3 2 3
✏1 ✏1 ✏1
6.7 6.7 6.7
(f1 , . . . , fn ) 4 .. 5 = y(xI A) + (a1 , . . . , an ) 4 .. 5 = (a1 , . . . , an ) 4 .. 7
6 7 6 7 6
5.
✏n ✏n ✏n

Hence (f1 , . . . , fn ) 2 ker if and only if (a1 , . . . , an ) = 0, so ker = {y(xI A) :

y 2 F [x]n }. Thus we have an F [x]-module isomorphism

V ⇠
= F [x]n / ker = F [x]n /{y(xI A) : y 2 F [x]n }.

By the second proof of Theorem 2.57, the invariant factors d1 , . . . , ds of the F [x]-
module V are precisely the nonunit invariant factors of xI A 2 Mn⇥n (F [x]).

2.10. Tensor and Hom

Tensor. Let R be a ring and let AR and R B be R-modules. Let F be the free
Z-module generated by A ⇥ B and let K be the Z-submodule of F generated by
the following elements, where a, a0 2 A, b, b0 2 B, and r 2 R:
8
> 0 0
<(a + a , b) (a, b) (a , b),
(2.14) (a, b + b0 ) (a, b) (a, b0 ),
>
:
(ar, b) (a, rb).

The quotient Z-module F/K is called the tensor product of A and B over R and
is denoted by A ⌦R B; for a 2 A and b 2 B, the element P(a, b) + K 2 A ⌦R B is
denoted by a ⌦ b. Each element of A ⌦R B is a finite sum i ai ⌦ bi , ai 2 A, bi 2 B.
If S, R, T are rings and S AR and R BT are (S, R)- and (R, T )-binomdules, re-
spectively, then A⌦R B is an (S, T )-bimodule with the scalar multiplications defined
by
⇣X ⌘ X
s a i ⌦ bi = sai ⌦ bi , s 2 S,
i i
⇣X ⌘ X
a i ⌦ bi t = ai ⌦ bi t, t 2 T.
i i

In the above, the left multiplication by s is well-defined since it is the map A⌦R B !
A ⌦R B induced by the map A ⇥ B ! A ⌦R B, (a, b) 7! sa ⌦ b. In the same way,
the right multiplication by t is well-defined.
Let S AR , R BT , S CT be bimodules. A mapping f : A ⇥ B ! C is called an
(S, R, T )-map if
(i) for each b 2 B, f (·, b) : S A ! S C is an S-map;
(ii) for each a 2 B, f (a, ·) : BT ! CT is a T -map;
(iii) f (ar, b) = f (a, rb) for all a 2 A, b 2 B, r 2 R.

Theorem 2.59 (Universal mapping property). Let f : S AR ⇥ R BT ! S CT be

an (S, R, T )-map. Then there is a unique (S, T )-map f¯ : S A ⌦R BT ! S CT such
 2.10. TENSOR AND HOM 77

that the diagram

f
S AR ⇥ R BT .......................................
S CT
... ...
... ........
..... .
... .....
.. .
......
.
i ...... .
....
..... ¯
. ..... f
.......... .....
.....
.. .....

SA ⌦ R BT

commutes, where i : A ⇥ B ! A ⌦R B is defined by i(a, b) = a ⌦ b.

Proof. Existence of f¯. Let F be the free Z-module generated by A ⇥ B
and let K be the Z-submodule of F generated by the elements in (2.14). First,
f induces a Z-map f˜ : F ! C. Clearly, ker f˜ K, hence f˜ induces a Z-map
f¯ : F/K = A ⌦R B ! C such that the above diagram commutes. Conditions (i)
and (ii) imply that f¯ is an (S, T )-map.
Uniqueness of f¯. Since A ⌦R B is generated by i(A ⇥ B), f¯ is unique. ⇤

Proposition 2.60. Let f : S AR ! S A0R be an (S, R)-map and g : R BT ! R BT0

be an (R, T )-map. Then there is a unique (S, T )-map A ⌦R B ! A0 ⌦R B 0 , denoted
by f ⌦ g, such that a ⌦ b 7! f (a) ⌦ g(b).
Proof. Existence of f ⌦ g. Define h : A ⇥ B ! A0 ⌦R B 0 , h(a, b) = f (a) ⌦ g(b),
which is an (S, R, T )-map. Hence h induces unique (S, T )-map h̄ : A ⌦R B !
A0 ⌦R B 0 such that the diagram
h
A⇥B ..........................................
A0 ⌦ R B 0
... ..........
... .......
... ......
.. ......
......
i ...... .
.....
.
..
.
. ...... h̄
.......... ......
......
.. ......

A ⌦R B

commutes. Clearly, h̄(a ⌦ b) = f (a) ⌦ g(b).

Uniqueness of f ⌦ g. If l : A ⌦R B ! A0 ⌦R B 0 is another (S, T )-map such that
l(a ⌦ b) = f (a) ⌦ g(b), then l i = h. By the uniqueness of h̄, we have l = h̄. ⇤

Properties of tensor. Let S, R, T, U be rings and let S AR , S (Ai )R , i 2 I,

R BT , R (Bj )T ,
j 2 J, and T CU be bimodules. We have the following isomorphisms
of bimodules:
⇣M ⌘ M⇣ ⌘
(2.15) ⇠
S (Ai ) ⌦R BT = S (Ai ) ⌦R BT ,
i2I i2I
⇣M ⌘ M
(2.16) SA ⌦R (Bj )T ⇠
= SA ⌦R (Bj )T ,
j2J j2J
(2.17) (S A ⌦R B) ⌦T CU ⇠
= S A ⌦R (B ⌦T CU ).

Proof. We only prove (2.15) and (2.17).

L
L 1 Proof of (2.15). Let ⇡k : i2I Ai ! Ak be0 the L projection and ◆k : Ak !
i2I A i be the embedding, and similarly define ⇡ k : i2I (Ai ⌦R B) ! Ak ⌦R B
78 2. RINGS AND MODULES

L
and ◆0k : Ak ⌦R B ! i2I (Ai ⌦R B). Let
X ⇣M ⌘ M
f= ◆0i (⇡i ⌦ idB ) : Ai ⌦ R B ! (Ai ⌦R B),
i2I i2I i2I

and ⇣M ⌘
X M
g= (◆i ⌦ idB ) ⇡i0 : (Ai ⌦R B) ! Ai ⌦R B.
i2I i2I i2I

Clearly, f and g are both (S, T )-maps. We claim that they are inverses of each
other.
Let a 2 Ak and b 2 B. Then
⇣X ⌘
(f g)(◆0k (a ⌦ b)) = f (◆i ⌦ idB ) ⇡i0 ◆0k (a ⌦ b) = f ((◆k ⌦ idB )(a ⌦ b))
i2I
⇣X ⌘
= f (◆k (a) ⌦ b) = ◆0i (⇡i ⌦ idB ) (◆k (a) ⌦ b)
i2I
X
= ◆0i (⇡i ◆k (a) ⌦ b) = ◆0k (a ⌦ b).
i2I
L
Since i2I (Ai ⌦R B) is generated by ◆k (a ⌦ b), k 2 I, a 2 AkL
0
, b 2 B, we have
f g = id. Similarly, (g f )(◆k (a) ⌦ b) = ◆k (a) ⌦ b. Since ( i2I Ai ) ⌦R B is
generated by ◆k (a) ⌦ b, k 2 I, a 2 Ak , b 2 B, we have g f = id.
2 Proof of (2.17). For each c 2 C, define

fc : A⇥B ! A ⌦R (B ⌦T C)
(a, b) 7 ! a ⌦ (b ⌦ c).

Clearly, fc is an (S, R, Z)-map, and hence it induces an (S, Z)-map f¯c : A ⌦R B !

A ⌦R (B ⌦T C) such that f¯c (a ⌦ b) = a ⌦ (b ⌦ c). Now define

g: (A ⌦R B) ⇥ C ! A ⌦R (B ⌦T C)
⇣X ⌘ ⇣X ⌘ X
a i ⌦ bi , c 7 ! f¯c a i ⌦ bi = ai ⌦ (bi ⌦ c).
i i i

Clearly, g is an (S, T, U )-map, and hence it induces an (S, U )-map ↵ : (A ⌦R B) ⌦T

C ! A⌦R (B ⌦T C) such that ↵((a⌦b)⌦c) = a⌦(b⌦c) for all a 2 A, b 2 B, c 2 C.
In the same way, there is an (S, U )-map : A ⌦R (B ⌦T C) ! (A ⌦R B) ⌦T C such
that (a ⌦ (b ⌦ c)) = (a ⌦ b) ⌦ c for all a 2 A, b 2 B, c 2 C. Then ↵ and are
inverses of each other. ⇤

Algebra. Let K be a commutative ring. An algebra over K, or a K-algebra,

is a ring A such that (A, +) is a left K-module and r(ab) = (ra)b = a(rb) for all
r 2 K and a, b 2 A. Since K is commutative, (A, +) is actually a (K, K)-bimodule.
Every ring A is a Z-algebra and a Z(A)-algebra, where Z(A) is the center of A. A
K-algebra homomorphism is a ring homomorphism which is also a K-map.
Theorem 2.61. Let A and B be algebras over a commutative ring K. Then
K A ⌦ K B has a K-algebra structure satisfying

(a ⌦ b)(a0 ⌦ b0 ) = (aa0 ) ⌦ (bb0 ), a, a0 2 A, b, b0 2 B.

 2.10. TENSOR AND HOM 79

Proof. 1 For (a, b) 2 A ⇥ B, define

f(a,b) : A⇥B ! K A ⌦ K BZ
(x, y) 7 ! (ax) ⌦ (by),
which is a (K, K, Z)-map. Hence f(a,b) induces a (K, Z)-map f¯(a,b) : K A ⌦K BZ !
¯
K A ⌦K BZ such that f(a,b) (x ⌦ y) = (ax) ⌦ (by) for all x 2 A, y 2 B.

2 For any v 2 K A ⌦K BZ , define

gv : A⇥B ! KA ⌦ K BZ
(a, b) 7 ! f¯(a,b) (v),
which is a (K, K, Z)-map. It induces a (K, Z)-map ḡv : KA ⌦ K BZ ! KA ⌦ K BZ
such that ḡv (a ⌦ b) = f¯(a,b) (v) for all a 2 A, b 2 B.
3 For u, v 2 K A ⌦K BZ , define
uv = ḡv (u).
We claim that with this multiplication, KA ⌦K BZ is a K-algebra. First note that
0 0 0 0
(2.18) (a ⌦ b)(a ⌦ b ) = (aa ) ⌦ (bb ) for all a, a0 2 A, b, b0 2 B.
For u1 , u2 , v 2 K A ⌦K BZ , we have
(2.19) (u1 + u2 )v = ḡv (u1 + u2 ) = ḡv (u1 ) + ḡv (u2 ) = u1 v + u2 v.
For a 2 A, b 2 B, we have
(a ⌦ b)(u1 + u2 ) = ḡu1 +u2 (a ⌦ b) = f¯(a,b) (u1 + u2 ) = f¯(a,b) (u1 ) + f¯(a,b) (u2 )
= ḡu1 (a ⌦ b) + ḡu2 (a ⌦ b) = (a ⌦ b)u1 + (a ⌦ b)u2 .
Hence
(2.20) v(u1 + u2 ) = vu1 + vu2 .
The associativity of the multiplication in K A ⌦K BZ follows from (2.18) – (2.20).
For r 2 K, a, a0 2 A and b, b0 2 B, we have
r((a ⌦ b)(a0 ⌦ b0 )) = r((aa0 ) ⌦ (bb0 )) = (raa0 ) ⌦ (bb0 )
= ((ra) ⌦ b)(a0 ⌦ b0 ) = (r(a ⌦ b))(a0 ⌦ b0 ),
and in the same way, r((a ⌦ b)(a0 ⌦ b0 )) = (a ⌦ b)(r(a0 ⌦ b0 )). Hence r(uv) = (ru)v =
u(rv) for all u, v 2 K A ⌦K BZ . ⇤
Hom. For (S, R)-bimodules S AR and S BR , let Hom(S,R) (A, B) denote the
abelian group of all (S, R)-maps from A to B. Hom(R,Z) and Hom(Z,R) are denoted
by Hom(R, · ) and Hom( · ,R) , respectively.
Fact. Let S, R, T be rings.
(i) For bimodules R AS and R BT , Hom(R, · ) (A, B) is an (S, T )-bimodule with
the scalar multiplications defined by
(sf t)(a) = f (as)t, s 2 S, t 2 T, f 2 Hom(R, · ) (A, B), a 2 A.
(ii) For bimodules S AR and T BR , Hom( · ,R) (A, B) is a (T, S)-bimodule with
the scalar multiplications defined by
(tf s)(a) = tf (sa), s 2 S, t 2 T, f 2 Hom( · ,R) (A, B), a 2 A.
80 2. RINGS AND MODULES

Hom(↵, ).
(i) For ↵ 2 Hom(R,S) (A0 , A) and 2 Hom(R,T ) (B, B 0 ), we have an (S, T )-
map
Hom(↵, ) : Hom(R, · ) (A, B) ! Hom(R, · ) (A0 , B 0 )
f 7 ! f ↵.
(ii) For ↵ 2 Hom(S,R) (A0 , A) and 2 Hom(T,R) (B, B 0 ), we have an (T, S)-
map
Hom(↵, ) : Hom(·,R) (A, B) ! Hom( · ,R) (A0 , B 0 )
f 7 ! f ↵.
f
..........................................
A. B.
.. ...
......... ...
... ...
↵ ...... ...
...
..
.... ..........
... ..

A0 ..........................................
f ↵
B0

Fact. Let S, R, T be rings.

(i) For (R, S)-bimodules R AS , R (Ai )S , i 2 I, and (R, T )-bimodules R BT ,
R (Bj )T , j 2 J, we have (S, T )-bimodule isomorphisms
⇣M ⌘ Y
(2.21) Hom(R, · ) Ai , B ⇠= Hom(R, · ) (Ai , B),
i2I i2I
⇣ Y ⌘ Y
(2.22) Hom(R, · ) A, Bj ⇠
= Hom(R, · ) (A, Bj ).
j2J j2J

(ii) For (S, R)-bimodules S AR , S (Ai )R , i 2 I, and (T, R)-bimodules T BR ,

T (Bj )R , j 2 J, we have (T, S)-bimodule isomorphisms
⇣M ⌘ Y
(2.23) Hom( · ,R) Ai , B ⇠
= Hom( · ,R) (Ai , B),
i2I i2I
⇣ Y ⌘ Y
(2.24) Hom( · ,R) A, Bj ⇠
= Hom( · ,R) (A, Bj ).
j2J j2J

Proof. We only prove (2.21) and (2.22).

L
L 1 Proof of (2.21). Let ◆k : Ak ! i2I Ai be the embedding and ⇡k :
A
i2I i ! A k be the projection. Define
⇣M ⌘ Y
↵ : Hom(R, · ) Ai , B ! Hom(R, · ) (Ai , B)
i2I i2I
f 7 ! ◆i )i2I , (f
Y ⇣M ⌘
: Hom(R, · ) (Ai , B) ! Hom(R, · ) Ai , B
i2I i2I
X
(fi )i2I 7 ! f i ⇡i .
i2I
L P
(In the above, for each x 2 i2I Ai , i2I fi ⇡i (x) is a finite sum.) Then ↵ and
are (S, T )-maps and are inverses of each other.
 2.10. TENSOR AND HOM 81

Q
2 Proof of (2.22). Each (fj )j2J 2 j2J Hom(R, · ) (A, Bj ) can be treated as a
Q
map (fj )j2J : A ! j2J Bj , a 7! (fj (a))j2J . Clearly,
Y ⇣ Y ⌘
Hom(R, · ) (A, Bj ) ! Hom(R, · ) A, Bj
j2J j2J
(fj )j2J 7 ! (fj )j2J
is an (S, T )-isomorphism. ⇤

Remark. The opposite ring of a ring (R, +, ·) is Rop = (R, +, ⇤), where a ⇤ b =
ba for all a, b 2 R. An (S, R)-bimodule is precisely an (Rop , S op )-bimodule. This
is the reason for the left-right symmetry in the statements concerning tensor and
hom. For example, (2.16) can be derived from (2.15) as follows:
⇣M ⌘ ⇣M ⌘
S A ⌦R (Bj )T ⇠ = T op (Bj ) ⌦Rop AS op (Exercise ??)
j2J j2J
M⇣ ⌘
⇠
= T op (Bj ) ⌦Rop AS op ((2.15))
j2J
M⇣ ⌘
⇠
= SA ⌦R (Bj )T (Exercise ??).
j2J

Equation (2.23) can be derived form (2.21) as follows:

⇣M ⌘ ⇣M ⌘
Hom( · ,R) Ai , B = Hom(Rop , · ) Ai , B
i2I i2I
Y
⇠
= Hom(Rop , · ) (Ai , B) ((2.21))
i2I
Y
= Hom( · ,R) (Ai , B).
i2I

Proposition 2.62 (Right exactness of tensor). Let R be a ring.

f g
(i) Let R B ! R C ! R D ! 0 be an exact sequence of left R-modules and
let AR be a right R-module. Then
idA ⌦f idA ⌦g
A ⌦R B ! A ⌦R C ! A ⌦R D ! 0
is exact.
f g
(ii) Let BR ! CR ! DR ! 0 be an exact sequence of right R-modules
and let R A be a left R-module. Then
f ⌦idA g⌦idA
B ⌦R A ! C ⌦R A ! D ⌦R A ! 0
is exact.
Proof. We only prove (i). Since g f = 0, we have (idA ⌦ g) (idA ⌦ f ) =
idA ⌦ 0 = 0. Since g is onto, it is clear that idA ⌦ g is onto. It remains to show
that ker(idA ⌦ g) ⇢ im(idA ⌦ f ). Define

↵: A⇥D ! (A ⌦R C)/im(idA ⌦ f )
(a, d) 7 ! a ⌦ c + im(idA ⌦ f ), c 2 g 1
(d).
82 2. RINGS AND MODULES

We claim that ↵ is well-defined. If c1 , c2 2 g 1 (d), then c1 c2 2 ker g = im f

and hence a ⌦ c1 a ⌦ c2 = a ⌦ (c1 c2 ) 2 im(idA ⌦ f ). Clearly, ↵ is a (Z, R, Z)-
map, so it induces a Z-map ↵ ¯ : A ⌦R D ! (A ⌦R C)/im(idA ⌦ f ) such that
¯ (a ⌦ d) = a ⌦ c + im(idA ⌦ f ), where g(c) = d, i.e., such that the following diagram
↵
commutes, where ⇡ is the canonical homomorphism.
idA ⌦g.....................
.....
A ⌦R D
....... ...
....... ...
.......
....... ...
...
...
A ⌦R C ... ↵
...
¯
....... ..
.......
....... ........
....... ...
....
⇡ ......................
(A ⌦R C)/im(idA ⌦ f )
Now it is clear that ker(idA ⌦ g) ⇢ im(idA ⌦ f ). ⇤

Proposition 2.63 (Left exactness of Hom). Let R be a ring.

f g
(i) Let 0 ! R B ! R C ! R D be an exact sequence of left R-modules and
let R A be a left R-module. Then
Hom(idA ,f ) Hom(idA ,g)
0 ! Hom(R, · ) (A, B) ! Hom(R, · ) (A, C) ! Hom(R, · ) (A, D)
is exact.
f g
(i0 ) Let 0 ! BR ! CR ! DR be an exact sequence of right R-modules
and let AR be a right R-module. Then
Hom(idA ,f ) Hom(idA ,g)
0 ! Hom( · ,R) (A, B) ! Hom( · ,R) (A, C) ! Hom( · ,R) (A, D)
is exact.
f g
(ii) Let R B ! R C ! R D ! 0 be an exact sequence of left R-modules and
let R A be a left R-module. Then
Hom(g,idA ) Hom(f,idA )
0 ! Hom(R, · ) (D, A) ! Hom(R, · ) (C, A) ! Hom(R, · ) (B, A)
is exact.
f g
(ii0 ) Let BR ! CR ! DR ! 0 be an exact sequence of right R-modules
and let AR be a right R-module. Then
Hom(g,idA ) Hom(f,idA )
0 ! Hom( · ,R) (D, A) ! Hom( · ,R) (C, A) ! Hom( · ,R) (B, A)
is exact.
Proof. We only prove (i) and (ii).
(i) Clearly, Hom(idA , g) Hom(idA , f ) = Hom(idA , g f ) = Hom(idA , 0) = 0. It
remains to show that ker(Hom(idA , f )) = 0 and ker(Hom(idA , g)) ⇢ im(Hom(idA , f )).
Let ↵ 2 ker(Hom(idA , f )). Then f ↵ = 0. Since f is one-to-one, we have
↵ = 0. Let 2 ker(Hom(idA , g)). Then g = 0, whence im ⇢ ker g = im f .
Let f1 : im f ! B be the inverse of f and let = f1 2 Hom(R, · ) (A, B). Then
=f = Hom(idA , f )( ) 2 im(Hom(idA , f )).

. .
A
...... ...
...... ...
......
.
.......... ...
...
.
...... ...
..
......... ...
....
. .
. . .... .........
......... f
........................................................
B ........................................................ im f
f1
 2.10. TENSOR AND HOM 83

(ii) We have Hom(f, idA ) Hom(g, idA ) = Hom(g f, idA ) = Hom(0, idA ) = 0. It
remains to show that ker(Hom(g, idA )) = 0 and ker(Hom(f, idA )) ⇢ im(Hom(g, idA )).
Let ↵ 2 ker(Hom(g, idA )). Then ↵ g = 0. Since g is onto, we have ↵ = 0. Let
2 ker(Hom(f, idA )). Then f = 0, i.e., 0 = (im f ) = (ker g). Thus induces
an R-map ¯ : C/ ker g ! A such that the lower triangle in the following diagram
commutes. Moreover, g induces an R-isomorphism ḡ : C/ ker g ! D such that the
upper triangle in the diagram commutes.
g
C. ............................................................. D
. .
... ..... . .
... ..... ........ ....
... ..... ... ...
... ..... ... ....
..... . ..
... ..... .
.....⇡ . .
...
... .....
..... ḡ ...... ...... ḡ 1
... ..... ... ....
... .....
.. ..... .. ..
..... ... ....
.......... ......... ... ........
.. . .... .. ..

A ..................................................
¯
C/ ker g

Let = ¯ ḡ 1 2 Hom(R, · ) (D, A). Then = g = Hom(g, idA )( ) 2

im(Hom(g, idA )). ⇤
Proposition 2.64. Let S, R, T be rings and S AR , R BT , S CT be binomules.
(i) (Adjointness of A ⌦R and Hom(S, · ) (A, )) There is a Z-isomorphism
Hom(S,T ) (A ⌦R B, C) ⇠
= Hom(R,T ) (B, Hom(S, · ) (A, C)).
(ii) (Adjointness of ⌦R B and Hom( · ,T ) (B, )) There is a Z-isomorphism
⇠ Hom(S,R) (A, Hom( · ,T ) (B, C)).
Hom(S,T ) (A ⌦R B, C) =
Proof. (i) Define
↵: Hom(S,T ) (A ⌦R B, C) ! Hom(R,T ) (B, Hom(S, · ) (A, C))
f 7 ! ↵(f ),
where
↵(f ) : B ! Hom(S, · ) (A, C)
b 7 ! f ( · ⌦ b).
Given g 2 Hom(R,T ) (B, Hom(S, · ) (A, C)), define
ḡ : A⇥B ! C
(a, b) 7 ! g(b)(a),
which is an (S, R, T )-map. Hence ḡ induces an (S, T )-map g̃ : A ⌦R B ! C such
that g̃(a ⌦ b) = g(b)(a). Define
: Hom(R,T ) (B, Hom(S, · ) (A, C)) ! Hom(S,T ) (A ⌦R B, C)
g 7 ! g̃.
It is easy to see that ↵ and are both Z-maps and ↵ = id and ↵ = id.
(ii) Define
: Hom(S,T ) (A ⌦R B, C) ! Hom(S,R) (A, Hom( · ,T ) (B, C))
f 7 ! (f ),
where
(f ) : A ! Hom( · ,T ) (B, C)
a 7 ! f (a ⌦ · ).
84 2. RINGS AND MODULES

Given g 2 Hom(S,R) (A, Hom( · ,T ) (B, C)), define

ḡ : A⇥B ! C
(a, b) 7 ! g(a)(b),
which is an (S, R, T )-map. Hence ḡ induces an (S, T )-map g̃ : A ⌦R B ! C such
that g̃(a ⌦ b) = g(a)(b). Define
: Hom(S,R) (A, Hom( · ,T ) (B, C)) ! Hom(S,T ) (A ⌦R B), C)
g 7 ! g̃.
Then and are both Z-modules and = id and = id. ⇤

Module of fractions. Let R be a commutative ring and S ⇢ R be a multi-

plicative set. Let A be an R-module. For (a, s), (a0 , s0 ) 2 A ⇥ S, define (a, s) ⇠
(a0 , s0 ) if and only if s1 (s0 a sa0 ) = 0 for some s1 2 S. Then ⇠ is an equiva-
lent relation on A ⇥ S; the ⇠ equivalence class of (a, s) is denoted by a/s. Let
S 1 A = (A ⇥ S)/ ⇠ and for a/s, a0 /s0 2 S 1 A and r/t 2 S 1 R, define
a a0 s0 a + sa0 ra ra
+ 0 = , = .
s s ss0 ts ts
These operations are well-defined and under these operations S 1 A becomes an
S 1 R -module, called the module of fractions of A by S. The map
: A ! S 1A
a 7 ! a/1
is an R-map, which is one-to-one if and only if sa 6= 0 for all s 2 S and a 2 A \ {0}.
Moreover, given any R-map f : A ! B, where A is an R-module and B is an
S 1 R -module, there is a unique S 1 R -map f¯ : S 1 A ! B such that the following
diagram commutes.
f
..................................
A B
... ..............
...
... .....
.....
...
. ......
.
......... ... ¯
..
..... f
.. .....

1
S A
Fact 2.65. In the above notation, S 1
A⇠
=S 1
R ⌦R A as S 1
R -modules.
Proof. The R-map f : A ! S 1 R ⌦R A, a 7! (1/1) ⌦ a induces an S 1 R -
map f¯ : S 1 A ! S 1 R ⌦R A such that f¯(a/1) = (1/1) ⌦ a. On the other hand,
the (S 1 R, R, Z)-map S 1 R ⇥ A ! S 1 A, (r/s, a) 7! ra/s, induces an S 1 R -map
g : S 1 R ⌦R A ! S 1 A such that g((r/s)⌦a) = ra/s. Clearly, f¯ and g are inverses
of each other. ⇤

2.11. Corner’s Theorem

Free ring. Let X be a set and X be the set of words in X:
X = {x1 · · · xn : n 0, x1 , . . . , xn 2 X}.
(Note: The words here are slightly different from those
P in free groups where the
exponents are ±1.) Let ZhXi be the set of formal sums x2X ax x, where ax 2 Z for
 2.11. CORNER’S THEOREM 85

P P
all x P
2 X and ax 6= 0 for only finitely many x. For A = x2X ax x, A0 = x2X a0x x,
B = y2X by y 2 ZhXi, define
X
A + A0 = (ax + a0x )x,
x2X
X⇣ X ⌘
AB = ax by z,
z2X xy=z

where xy 2 X is the concatenation of x and y. Then ZhXi becomes a ring, which

is called the free ring on X. Every function f : X ! R, where R is a ring, can be
uniquely extended to a homomorphism f¯ : ZhXi ! R. For S ⇢ ZhXi, the quotient
ring ZhXi/hSi, where hSi is the ideal of ZhXi generated by S, is the ring generated
by X subject to the relations s = 0, s 2 S.
The following theorem was used in Remark 1.28 (ii) to illustrate a strange
behavior of certain abelian groups that are not finitely generated. An abelian
group is called reduced if its only divisible subgroup is {0}.
Theorem 2.66 (Corner [6]). Let r 2 Z+ . There exists a countable reduced
torsion-free abelian group G such that for m, n 2 Z+ , Gm ⇠
= Gn if and only if
m ⌘ n (mod r).
The proof of Theorem 2.66 included below relies on another theorem (Theo-
rem 2.67) by Corner whose proof can be found in [5].
Theorem 2.67 (Corner [5]). Let A be a ring such that (A, +) is countable,
reduced, and torsion-free. Then A ⇠
= End(G) for some countable, reduced, torsion-
free abelian group G.
Lemma 2.68. Let G be an abelian group and let !1 , !2 two idempotents in
End(G). Then !1 G ⇠ = !2 G if and only if there exist x, y 2 End(G) such that
!1 = xy and !2 = yx.
Proof. (() Define y ⇤ : !1 G ! !2 G, !1 a 7! y!1 a and x⇤ : !2 G ! !1 G,
!2 a 7! x!2 a. Then x⇤ and y ⇤ are inverses of each other.
()) Let ↵ : !1 G ! !2 G and : !2 G ! !1 G be inverse isomorphisms of each
other. Then y := ↵!1 and x := !2 have the desired property. ⇤
Proof of Theorem 2.66. Let A be the ring generated by xi , yi , 0  i  r,
subject to the relations
( r
1 if i = j, X
(2.25) yi x j = xi yi = 1.
0 if i =
6 j, i=0

1 We claim that (A, +) is free abelian. Let F be the free ring with generators
xi , yi , 0  i  r. Consider the following subsets of F :
(2.26)
A = {xi1 · · · xim yjn · · · yj1 , n, m 0, 0  ik , jl  r, (im , jn ) 6= (r, r)},
n ⇣Xr ⌘ o
B = x i1 · · · x im xi yi 1 yjn · · · yj1 : m, n 0, 0  ik , jl  r ,
i=0
C = the set of products of elements of {x0 , . . . , xr , y0 , . . . , yr } containing
86 2. RINGS AND MODULES

at least one string yi xj (i 6= j) but no string yi xi ,

D = the set of products of elements in {x0 , . . . , xr , y0 , . . . , yr }[
{yi xi 1 : 0  i  r} containing at least one yi xi 1 but no string yi xi .

Let I be the Z-module generated by B [C [D. It is easy to see that xI = Ix ⇢ I for

all x 2 {x0 , . . . , xr , y0 , . . . , yr }. Hence I is the ideal of F generated by the relations
in (2.25). Clearly, (F, +) is generated by A[B[C [D, and hence (A, +) is generated
by A (mod I). It remains to show that A (mod I) is linearly independent over Z.
In fact, we show that A [ B [ C [ D is linearly independent over Z. Assume that
s
X t
X u
X v
X
(2.27) ↵i ai + j bj + k ck + l dl = 0,
i=1 j=1 k=1 l=1

where ↵i , j , k , l 2 Z, ai 2 A, bj 2 B, ck 2 C, dl 2 D. Assume that d1 involves

the largest number of elements in {x0 , . . . , xr , y0 , . . . , yr } (counting multiplicity).
In (2.27), expand each bj and dl as a product of elements in {x0 , . . . , xr , y0 , . . . , yr },
and let d01 denote the element obtained from d1 by replacing each yi xi 1 with
yi xi . Then on the left side of (2.27), 1 d01 is the only term of involving d01 , and
hence 1 = 0. In the same way, 1 = · · · = v = 0. It follows immediately
that 1 = · · · = u = 0. Assume that b1 involves the largest number of elements
in {x0 , . . . , xr , y0 , . . . , yr } (countingPmultiplicity), and let b01 denote the element
r
obtained from b1 by replacing each i=0 xi yi 1 with xr yr . Then on the left side
of (2.27), 1 b01 is the only term of involving b01 , and hence 1 = 0. In the same way,
1 = · · · = t = 0. Finally, ↵1 = · · · = ↵s = 0.

2 By Theorem 2.67, A = End(G) for some countable, reduced, torsion-free

abelian group G. Let !i = xi yi , 0  i  r. Then 1 = !0 +· · ·+!r is a decomposition
of 1 into orthogonal idempotents, hence
r
X
G= !i G.
i=0

Since !i = xi yi and 1 = yi xi , by Lemma 2.68, G ⇠

= !i G. Thus G ⇠
= Gr+1 , hence
m ⇠
G = G if m, n 2 Z and m ⌘ n (mod r).
n +

3 Recall that A (defined in (2.26)) is a Z-basis of A. For xi1 · · · xim yjn · · · yj1 2
A, define
(
1 if m = n and ik = jk , 1  k  m,
T (xi1 · · · xim yjn · · · yj1 ) =
0 otherwise,

and extend T to a Z-linear map T : A ! Z. Then T (!i ) = 1 for 0  i  r 1.

For a = xi1 · · · xim yjn · · · yj1 2 A and x 2 {x0 , . . . , xr , y0 , . . . , yr }, we claim that
T (xa) = T (ax). By symmetry, we may assume that x = xi0 for some 0  i0  r.
Then
(2.28) (
1 if n = m + 1, ik = jk+1 , 0  k  m,
T (xa) = T (xi0 xi1 · · · xim yjn · · · yj1 ) =
0 otherwise,
 2.11. CORNER’S THEOREM 87

and
(2.29) (
T (xi1 · · · xim yjn · · · yj2 ) if i0 = j1 ,
T (ax) = T (xi1 · · · xim yjn · · · yj1 xi0 ) =
0 otherwise.
Comparing (2.28) and (2.29) gives T (xa) = T (ax). It follows that T (xy) = T (yx)
for all x, y 2 A.
4 Assume that Gm ⇠= Gn for some 1  m, n  r. Then
⇣m
X1 ⌘ m
X1 n
X1 ⇣nX1 ⌘
!i G = !i G ⇠
= Gm ⇠
= Gn ⇠
= !i G = !i G.
i=0 i=0 i=0 i=0
Pm 1 Pn 1
By Lemma 2.68, there exist x, y 2 A such that i=0 !i = xy and i=0 !i = yx.
Therefore
⇣mX1 ⌘ ⇣nX1 ⌘
m=T !i = T (xy) = T (yx) = T !i = n.
i=0 i=0
⇤
 CHAPTER 3

Fields

3.1. Field Extensions

Degree of extension. Let F ⇢ K be fields. K is called an extension of F and
the extension is denoted by K/F . The dimension dimF K is called the degree of K
over F and is denoted by [K : F ]. If [K : F ] < 1, K is called a finite extension
over F .
Examples. [C : R] = 2, [R : Q] = @. In general, if F ⇢ K are fields such that
|K| =L1 and |K| > |F |, then [K : F ] = |K|. Let X be a basis of K/F . Then
K⇠= x2X F . Clearly, |X| = 1 and |X|  |K|. Let P0 (X) be the set of all finite
subsets of X. Then
M X
|K| = F  |F ||Y |  |P0 (X)| max{|F |, @0 }
x2X Y 2P0 (X)

= |X| max{|F |, @0 } = max{|F |, |X|}.

Since |K| > |F |, we must have |K|  |X|.
Fact. Let F be a field and let f 2 F [x] be irreducible with deg f = n. Then
K = F [x]/(f ) is an extension field of F with [K : F ] = n. The element x + (f ) 2 K
is a root of f and x0 + (f ), . . . , xn 1 + (f ) is a basis of K over F .
Fact. Let F ⇢ K ⇢ L be fields. Then [L : F ] = [L : K][K : F ].

L Proof. Let A L be a basis of K/F andL B be L a basis of L/K.

L Then K =
a2A F a and L = b2B Kb, whence L = b2B ( a2A F a)b = a2A, b2B F ab.
Hence {ab : a 2 A, b 2 B} is a basis of L/F . ⇤
Notation. Let F ⇢ K be fields and X ⇢ K. We write
F [X] = the smallest subring R ⇢ K such that R F and R X,
F (X) = the smallest subfiled E ⇢ K such that E F and E X.
Equivalently,
F [X] = {f (u1 , . . . , un ) : n 2 N, f 2 F [x1 , . . . , xn ], u1 , . . . , un 2 X},
nu o
F (X) = : u, v 2 F [X], v 6= 0 .
v
If E and F are both subfields of K, the compositum of E and F , denoted by EF ,
is the smallest subfield of K containing E [ F .
Definition 3.1. Let F ⇢ K be fields and u 2 K. If there exists 0 6= f 2 F [x]
such that f (u) = 0, u is called algebraic over F . The monic polynomial m 2 F [x]
of the smallest degree such that m(u) = 0 is called the minimal polynomial of u
over F . (Clearly, the minimal polynomial of an algebraic element is unique.) If u
is not algebraic over F , it is called transcendental over F . The field K is called an
89
90 3. FIELDS

algebraic extension of F if every element of K is algebraic over F ; otherwise, K is

called a transcendental extension over F .
p
Examples. C/R and Q( 19)/Q are algebraic extensions; R/Q is transcen-
dental. Let F be a field and F (x) be the field of rational functions in x over F .
Then F (x)/F is transcendental. If u 2 F (x) \ F , then x is algebraic over F (u).
Write u = f (x)/g(x), where f, g 2 F [x] are such that g 6= 0 and gcd(f, g) = 1. Let
h(y) = f (y) ug(y) 2 (F (u))[y]. Then h 6= 0 and h(x) = 0.
p p
Example. 2 + 3 3 2 R is algebraic over Q with minimal polynomial f (x) =
(x3 + 6x 3)2 2(3x2 + 2)2 .
p p p 3 p p
Proof. Let ↵ = p 2 + 3 3. Then 3 = (↵ 2) = ↵3 3 2↵2 + 6↵ 2 2.
Hence p↵3 +p6↵ 3 = 2(3↵2 + p 2) and 3(↵ + 6↵ 3) =2 2(3↵ + 2)p
3 2 2 2
. On the other
p
hand, 2, 3 2 Q(↵) (since 2 = (↵ + 6↵ 3)/(3↵ + 2) and 3 3 = ↵
3
2).
Therefore 6 | [Q(↵) : Q], whence [Q(↵) : Q] 6. By Fact 3.2 (iii), f is the minimal
polynomial of ↵. ⇤
Complex numbers that are algebraic (transcendental) over Q are usually called
algebraic numbers (transcendental numbers). Let A = {u 2 C : u is algebraic over Q}.
Then |A| = @0 since |Q[x]| = @0 and every 0 6= f 2 Q[x] has finitely Pmany roots.
1
Hence |C \ A| = @. Examples of transcendental numbers: e, ⇡ (§3.11), n=1 1/10n!
(Liouville’s number, Theorem 3.9).
Fact 3.2. Let F ⇢ K be fields.
(i) If u 2 K is transcendental over F , then F (u) ⇠ = F (x).
(ii) Let u 2 K be algebraic over F with minimal polynomial m and let f 2
F [x]. Then f (u) = 0 if and only if m | f .
(iii) Let u 2 K be algebraic over F and f 2 F [x] be monic. Then f is
the minimal polynomial of u over F if and only if f is irreducible and
f (u) = 0. In this case, F (u) = F [u] ⇠ = F [x]/(f ) and [F (u) : F ] = deg f ;
1, u, . . . , udeg f 1 is a basis of F (u)/F .
(iv) u 2 K is algebraic over F if and only if [F (u) : F ] < 1.
(v) If [K : F ] < 1, K/F is algebraic. (The converse is false; see Example 3.4.)
Proof. (ii) ()) Assume that f (u) = 0. Write f = qm + r, where q, r 2 F [x]
and deg r < deg m. Then r(u) = 0, and the minimality of deg m forces r = 0,
whence m | f .
(iii) ()) Assume to the contrary that f is reducible, say, f = f1 f2 , where
f1 , f2 2 F [x] and 0 < deg f1 , deg f2 < deg f . Then either f1 (u) = 0 or f2 (u) = 0,
contradicting the minimality of deg f .
(() Let m be the minimal polynomial of u over F . By (ii), m | f . Since f is
irreducible, we have m = f .
Assume that f is the minimal polynomial of u over F . The map : F [x] !
F [u], g(x) 7! g(u), is an onto ring homomorphism with ker g = (f ). Hence
induces an isomorphism ¯ : F [x]/(f ) ! F [u]. Since f is irreducible, F [x]/(f ) is a
field. Therefore F [u] is a field and hence F [u] = F (u). ⇤
Proposition 3.3 (Relative algebraic closure). Let F ⇢ K be fields and let
A = {u 2 K : u is algebraic over F }.
 3.1. FIELD EXTENSIONS 91

Then A is a subfield of K which is called the (relative) algebraic closure of F in K.

Proof. For any u, v 2 A with v 6= 0, we have [F (u, v) : F ] = [F (u, v) :
F (v)][F (v) : F ] < 1, whence F (u, v) ⇢ A. Thus u v, u/v 2 A, so A is a field. ⇤
Fact. Let F ⇢ K be fields and X ⇢ K be a set of algebraic elements over F .
Then F (X) is algebraic over F .
Proof. Let A be the algebraic closure of F in K. Since F [ X ⇢ A, we have
F (X) ⇢ A. ⇤
Example 3.4. Let A be the algebraic closure of Q in C. Then A/Q is algebraic
but [A : Q] = 1. Let p be a prime and n be any positive integer. By Eisenstein’s
criterion, xn p 2 Q[x] is irreducible. Thus [A : Q] [Q(p1/n ) : Q] = n.
Proposition 3.5. Let F ⇢ K ⇢ L be fields such that K/F and L/K are both
algebraic. Then L/K is algebraic.
Proof. For each u 2 L, since u is algebraic over K, we have un + bn 1u
n 1
+
· · · + b0 = 0 for some b0 , . . . , bn 1 2 K. Then
[F (u) : F ]  [F (b0 , . . . , bn 1 )(u) : F]
= [F (b0 , . . . , bn 1 )(u) : F (b0 , . . . , bn 1 )][F (b0 , . . . , bn 1 ) : F ] < 1.
Hence u is algebraic over F . ⇤
Ruler and compass constructions. On the complex plane C with 0 and 1
given, a point (complex number) is called constructible if it can be obtained through
a sequence of steps; in each step, one uses a ruler and a compass to determine the
intersection point(s) of two curves on C each of which is either a line through
two points already constructed or a circle whose center and radius are already
constructed.
Theorem 3.6.
(i) z 2 C is constructible if and only if there are fields Q = K0 ⇢ K1 ⇢ · · · ⇢
Kn ⇢ C such that [Kj : Kj 1 ] = 2 and z 2 Kn .
(ii) The set of all constructible numbers in C is a field.
Proof. (i) ()) Consider a step in a ruler and compass construction. Let
K ⇢ C be a subfield containing all numbers already constructed. The current step
produces a + bi where (a, b) is a common root of two polynomials in K[x, y], each
of which is of the form cx + dy + e ((c, d) 6= (0, 0)) or the form x2 + y 2 + f x + gy + h.
It is easy to see that [K(a) : K] = 1 or 2 and [K(b) : K] = 1 or 2. So K ⇢ K(a) ⇢
K(a, b) ⇢ K(a, b, i) 3 a + bi, where each extension is of degree 1 or 2. Therefore,
each constructible number is contained in the last field of a tower of extensions
Q = K0 ⇢ K1 ⇢ · · · ⇢ Kn ⇢ C with [Kj : Kj 1 ] = 2, 1  j  n.
(() We only have to p show that every element in Kj is constructible from Kj 1 .
Note
p that K j = K j 1 ( d) for some d 2 K j 1 . Let d = ⇢e i✓
where ⇢ 0. Then
p p
d = ⇢ei✓/2 . The angle ✓/2 is constrctible from ✓. Also, ⇢ is constructible from
p p
⇢; see Figure 3.1.
p So d is constructible from d. p Each element in Kj 1 ( d) is of
the form a + b d with a, b 2 Kj 1 . Clearly, a + b d is constructible from Kj 1 .
(ii) Let z, w 2 C (w 6= 0) be constructible. It suffices to show that z w and z/w
are both constructible. The geometric proof of this is obvious. The algebraic proof
92 3. FIELDS
.
.........
..
p ...
.
⇢ ............................
..
.. ..
. .. ... ............
.
.......... ...... .... ..........
.
. . ........
......
...... ..... .. .......... ......
..
...... ....... ....
. ..... .....
.....
.... ...
. .
. ..... ...
.. .
. . ........
...
. .....
. .
.
. ...
.....
...
...
... ..
.... .
.
. .
....... ...
... ..
.... .
.
. ...
..... ...
.
.... ........ .... .....
.....
...
.
... ...... ... ..... ....
... ...... .
. ..... ..
..... ..
........ ...
. .......
................................................................................................................................................................................................
1 0 ...... ⇢
....
....
p
Figure 3.1. Construction of ⇢, ⇢ 0

is also easy. Let Q = K0 ⇢ K1 ⇢ · · · ⇢ Kn 3 z and Q = L0 ⇢ L1 ⇢ · · · ⇢ Lm 3 w,

where [Ki : Ki 1 ] = 2 and [Lj : Lj 1 ] = 2. Then
Q ⇢ K1 ⇢ · · · ⇢ Kn ⇢ Kn L1 ⇢ · · · ⇢ Kn Lm 3 z, w,
where each extension is of degree 1 or 2. Since z w, z/w 2 Kn Lm , both of them
are constructible. ⇤
Corollary 3.7. If z 2 C is constructible, then [Q(z) : Q] is a power of 2.
Three ancient ruler-compass problems.
(i) Squaring the circle (constructing a square having the same area of a
unit circle). This is impossible since ⇡ is transcendental hence not con-
structible.
(ii) Doubling the cube (constructing a cube with volume 2). This is impossible
since [Q(21/3 ) : Q] = 3.
(iii) Trisection of an arbitrary angle. An angle of 60 is constructible but
cannot be trisected by ruler and compass. Since cos 3↵ = 4 cos3 ↵ 3 cos ↵,
cos 20 is a root of 8x3 6x 1, which is irreducible in Q[x]. Hence
[Q(cos 20 ) : Q] = 3 and cos 20 is not constructible.
The prime field. Let F be a field. The intersection of all subfields of F is
called the prime field of F ; it is
(
⇠ Q if char F = 0,
=
Zp if char F = p.

Proof. Let P be the prime field of F . When char F = 0, the map Q ! P ,

m/n 7! (m · 1F )/(n · 1F ) is an isomorphism; when char F = p, the map Zp ! P ,
a + pZ 7! a1F is an isomorphism. ⇤
Transcendence of Liouville’s number.
Theorem 3.8 (Liouville’s inequality). Let ↵ 2 C be a root of a polynomial of
degree d > 0 in Z[x]. Then for each ✏ > 0, there are only finitely many rational
numbers a/b (a, b 2 Z, b > 0) such that
a 1
↵ < d+✏ .
b b
Proof. Assume that ↵ is a root of f (x) = cd xd + · · · + c0 2 Z[x]. Let a/b 2 Q
be such that |a/b ↵| < 1/bd+✏ but f (a/b) 6= 0. Then
⇣a⌘ ⇣ a ⌘d c d a d + c d 1 a d 1 b + · · · + c 0 bd 1
f = cd + · · · + c0 = .
b b bd bd
 3.2. GALOIS THEORY 93

Write f (x) = (x ↵)g(x), where g(x) = ed 1 xd 1 + · · · + e0 2 C[x]. Note that

⇣a⌘ a d 1
g  |ed 1 | + · · · + |e0 |  |ed 1 |(|↵| + 1)d 1 + · · · + |e0 | = C,
b b
where C does not depend on a/b. Therefore,
1 ⇣a⌘ a ⇣a⌘ C
 f = ↵ g  d+✏ ,
bd b b b b
i.e., b✏  C. There are only finitely many such b. For each such b, there are only
finitely many a 2 Z such that |a/b ↵| < 1. ⇤
P1
Theorem 3.9. Liouville’s number ↵ = n=1 1/10 is transcendental.
n!

PN
Proof. For each N 1, let rN = n=1 1/10n! = aN /10N ! 2 Q. Then for
each D > 0,
X1
1 2 1
|rN ↵| = n!
 (N +1)! <
10 10 (10 ! )D
N
n=N +1
when N is large enough. By Loiuville’s inequality, ↵ is transcendental. ⇤
Remark. Let unP2 {0, . . . , 9}, n 1, be a sequence with infinitely many
1
nonzero terms. Then n=1 un /10n! is transcendental; this is clear from the proof
of the above theorem. Thus we have exhibited @ transcendental numbers.

3.2. Galois Theory

The Galois group. Let F ⇢ K be fields. The group Aut(K/F ) = { 2
Aut(K) : |F = id} is called the Galois group of K over F .
Examples. Aut(C/R) = {id, ( )}, where ( ) is the complex conjugation.
Aut(R/Q) = {id}. Let 2 Aut(R/Q). If a, b 2 R are such that a b,
p 2 p
then (a b) = ( a b ) = ( a b )2 0, hence (a) (b). For each
x 2 R, choose sequences an , bn 2 Q such that an  x  bn and limn!1 an = x =
limn!1 bn . Then an = (an )  (x)  (bn ) = bn for all n. Letting n ! 1 gives
(x) = x.
|Aut(C/Q)| = 2@ . (See Example 3.87.)
Fact. Let F ⇢ K be fields, f 2 F [x] and 2 Aut(K/F ). Then permutes
the roots of f in K. It follows that if [K : F ] < 1, then |Aut(K/F )| < 1.
Subfields and subgroups. Let F ⇢ K be fields and let
F(K/F ) = the set of all fields between F and K,
G(K/F ) = the set of all subgroups of Aut(K/F ).
For L 2 F(K/F ) and H 2 G(K/F ), define
L0 = Aut(K/L) 2 G(K/F ),
H 0 = {x 2 K : (x) = x for all 2 H} 2 F(K/F ).

Proposition 3.10. Let F ⇢ K be fields.

(i) K 0 = {id}, F 0 = Aut(K/F ), {id}0 = K.
(ii) If L, M 2 F(K/F ) and L ⇢ M , then L0 M 0 ; if H, J 2 G(K/F ) and
H ⇢ J, then H 0 J 0 .
94 3. FIELDS

(iii) For L 2 F(K/F ) and H 2 G(K/F ), we have L ⇢ L00 , H ⇢ H 00 , L000 = L0 ,

H 000 = H 0 .
(iv) If H 2 G(K/F ) and |H| < 1, then H 00 = H.
(v) For L, M 2 F(K/F ), (LM )0 = L0 \ M 0 ; for H, J 2 G(K/F ), hH [ Ji0 =
H 0 \ J 0.
Proof. (iii) To show that L000 = L0 , note that L ⇢ L00 implies L0 L000 and
that L0 ⇢ (L0 )00 = L000 .
(iv) By Lemmas 3.14 and 3.15, |H| = [H : {id}] [{id}0 : H 0 ] [H 00 : {id}00 ] =
|H |. Hence H = H 00 .
00

(v) Obviously, (LM )0 ⇢ L0 \ M 0 . Also, (L0 \ M 0 )0 L00 M 00 LM . Thus

L \ M ⇢ (L \ M ) ⇢ (LM ) . Hence (LM ) = L \ M . By a similar argument,
0 0 0 0 00 0 0 0 0

hH [ Ji0 = H 0 \ J 0 . ⇤

Note. In (i), we do not always have Aut(K/F )0 = F . If this happens, K/F is

called a Galois extension.
( )0
K .......................................................................................... {id}
.... ....
... ...
.. ..
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
... ...
.. ..
( )0
F 00 = Aut(K/F )0 .....................................................
......
Aut(K/F ) = F 0
.........
........
... .........
... .
...
..........
.
........
... ........
. ......... ( )0
........
.........
F
Definition 3.11 (Galois extension). Let F ⇢ K be fields. K is called a
Galois extension over F if {x 2 K : (x) = x for all 2 Aut(K/F )} = F , i.e.,
F 00 = F . Equivalently, K/F is Galois if and only if for each x 2 K \ F , there exists
2 Aut(K/F ) such that (x) 6= x.
Example. Q(21/3 )/Q is not Galois since Aut(Q(21/3 )/Q) = {id} and {id}0 =
Q(2 ) 6= Q. Let ⇠ = e2⇡i/3 . Then Q(21/3 , ⇠)/Q is Galois. The complex conjuga-
1/3

tion ( ) is in Aut(Q(21/3 , ⇠)/Q(21/3 )). Also, there exists 2 Aut(Q(21/3 , ⇠)/Q(⇠))

such that (21/3 ) = 21/3 ⇠. The extension Q(21/3 , ⇠)/Q has a basis 2k/3 ⇠ l , 0  k  2,
0  l  1. The elements in Q(21/3 , ⇠) that are fixed by ( ) and are those in Q.
The fundamental theorem of Galois theory.
Theorem 3.12. Let K/F be a finite Galois extension. Then ( )0 : F(K/F ) !
G(K/F ) and ( )0 : G(K/F ) ! F(K/F ) are bijections and are inverses of each
other. Moreover, these bijections have the following properties.
(i) If L, M 2 F(K/F ) and L ⇢ M , then [M : L] = [L0 : M 0 ]; if H, J 2
G(K/F ) and H ⇢ J, then [J : H] = [H 0 : J 0 ].
(ii) For L, M 2 F(K/F ), (L\M )0 = hL0 [M 0 i; for H, J 2 G(K/F ), (H\J)0 =
H 0J 0.
(iii) For every L 2 F(K/F ), K/L is Galois; L/F is Galois if and only if
L0 C F 0 ; when L0 C F 0 , Aut(L/F ) ⇠
= F 0 /L0 = Aut(K/F )/Aut(K/L).
 3.2. GALOIS THEORY 95

( )0
K ............................................................................... {id}
.... ....
... ... ( )0
..
...
..
... K ................................................................................ {id}
... ...
.. .. .. ..
... ...
.. ..
() ............................................................................... () ...
....
always Galois ...
....
.. ..
... ... . .
... ...
.. ..
...
...
dgree = index ...
...
() ............................................................................... ()
... ...
.... ....
... ...
() ................................................................................ () ..
... Galois , 4 ..
...
... ...
.. .. .. ..
... ...
.. ..
...
....
...
.... F ............................................................................... Aut(K/F )
.. ..
. .

F ............................................................................... Aut(K/F )

Proof. The proof relies on two key lemmas (Lemmas 3.14 and 3.15) which
will be proved afterwards.
Since K/F is Galois, F 00 = F . For each L 2 F(K/F ), we have L ⇢ L00 and, by
Lemmas 3.14 and 3.15, [L00 : F ] = [L00 : F 00 ]  [F 0 : L0 ]  [L : F ], whence L00 = L.
For each H 2 G(K/F ), we have |H|  |Aut(K/F )| < 1 since [K : F ] < 1. By
Proposition 3.10 (iv), H 00 = H. (Note: In the proof of H 00 = H, we only used the
fact that [K : F ] < 1.)
(i) Since [L0 : M 0 ]  [M : L] = [M 00 : L00 ]  [L0 : M 0 ], we have [M : L] = [L0 :
0
M ].
(ii) Obviously, (L \ M )0 L0 [ M 0 , so (L \ M )0 hL0 [ M 0 i. Also, hL0 [ M 0 i0 ⇢
L \ M 00 = L \ M , so hL0 [ M 0 i (L \ M )0 . Hence (L \ M )0 = hL0 [ M 0 i.
00

(iii) K/L is Galois since L00 = L. Now we prove that L/F is Galois if and only
if L0 C F 0 .
()) Let 2 L0 and ⌧ 2 F 0 . We want to show that ⌧ 1 ⌧ 2 L0 . It suffices to
show that ⌧ (L) ⇢ L. Let u 2 L and let f 2 F [x] be the minimal polynomial of u
over F . Let u1 (= u), u2 , . . . , ur be all the distinct roots of f in L. Then every ↵ 2
Aut(L/F ) permutes u1 , . . . , ur , hence ↵((x u1 ) · · · (x ur )) = (x u1 ) · · · (x ur ).
Since L/F is Galois, (x u1 ) · · · (x ur ) 2 F [x]. Then ⌧ permutes the roots of
(x u1 ) · · · (x ur ). Therefore, ⌧ (u) = ⌧ (u1 ) = ui 2 L for some i.
(() For each ⌧ 2 F 0 = Aut(K/F ), we have ⌧ (L) ⇢ L. (For each 2 L0 ,
⌧ ⌧ 2 L , i.e., ⌧ (v) = ⌧ (v) for all v 2 L. Hence ⌧ (v) 2 L for all v 2 L.) We
1 0

also have ⌧ 1 (L) ⇢ L, whence ⌧ (L) = L. Thus ⌧ |L 2 Aut(L/F ). Now assume that
u 2 L \ F . Since K/F is Galois, there exists ⌧ 2 Aut(K/F ) such that ⌧ (u) 6= u.
Then ⌧ |L 2 Aut(L/F ) and ⌧ |L (u) 6= u. Hence L/F is Galois.
The map : F 0 ! Aut(L/F ), ⌧ 7! ⌧ |L , is a homomorphism with ker = L0 ;
it induces an embedding ¯ : F 0 /L0 ,! Aut(L/F ). Since |F 0 /L0 | = [L : F ] =
|Aut(L/F )| < 1, ¯ is an isomorphism. ⇤
Proposition 3.13 (Linear independence of characters). Let G be a group and
E be a field. Let 1 , . . . , n be distinct homomorphisms from G to E ⇥ . Then
1 , . . . , n are linearly independent over E as functions from G to E. (A homo-
morphism : G ! E ⇥ is called an E-character of G.)
96 3. FIELDS

Proof. Assume to the contrary that 1 , . . . , n are linearly dependent over

E. Choose a minimal linearly dependent subset of { 1 , . . . , n }, say { 1 , . . . , m }.
Then there are c1 , . . . , cm 2 E ⇥ such that c1 1 + · · · + cm m = 0, i.e.,
(3.1) c1 1 (x) + c2 2 (x) + · · · + cm m (x) =0 for all x 2 G.
Clearly, m 2. Choose y 2 G such that 1 (y) 6= 2 (y). Replacing x by yx in (3.1)
gives
(3.2) c1 1 (y) 1 (x) + c2 2 (y) 2 (x) + · · · + cm m (y) m (x) = 0, x 2 G.
Combining (3.1) and (3.2) gives
c2 ( 1 (y) 2 (y)) 2 (x) + · · · + cm ( 1 (y) m (y)) m (x) = 0, x 2 G.
Then 2, . . . , m are linearly dependent, which is a contradiction. ⇤
Lemma 3.14. Let F ⇢ K be fields and let L, M 2 F(K/F ) be such that
L ⇢ M . If [M : L] < 1, then [L0 : M 0 ]  [M : L].
Proof. Let [M : L] = n and assume to the contrary that [L0 : M 0 ] > n. Let
1 , . . . , n+1 2 L be such that they represent distinct left cosets of M in L . Let
0 0 0

✏1 , . . . , ✏n be a basis of M/L. Then there exists 0 6= (c1 , . . . , cn+1 ) 2 K n+1

such
that 2 32 3
1 (✏1 ) · · · n+1 (✏1 ) c1
6 . .. 76 . 7
6 . 7 6 . 7 = 0.
4 . . 54 . 5
1 (✏n ) · · · n+1 (✏n ) cn+1
For each x 2 M , write
2 3
✏1
6.7
x = [a1 , . . . , an ] 6 .7 aj 2 L.
4 . 5,
✏n
Then 2 3
i (✏1 )
6 .. 7
i (x) = [a1 , . . . , an ] 6
4 . 5,
7 1  i  n + 1.
i (✏n )
Thus
3 2
c1
6 . 7
n+1 (x) = [ 1 (x), . . . , n+1 (x)] 4 .. 5
c1 1 (x) + · · · + cn+1 6 7
cn+1
2 32 3
1 (✏1 ) · · · n+1 (✏1 ) c1
6 . .. 76 . 7
= [a1 , . . . , an ] 6 . 76 . 7
4 . . 54 . 5
1 (✏n ) · · · n+1 (✏n ) cn+1
= 0.
Hence 1 |M , . . . , n+1 |M are linearly dependent over K.
Since 1 , . . . , n+1 belong to different left cosets of M 0 in L0 , 1 |M ⇥ , . . . , n+1 |M ⇥
are distinct K-characters of M ⇥ . By Proposition 3.13, 1 |M ⇥ , . . . , n+1 |M ⇥ are lin-
early independent over K, which is a contradiction. ⇤
 3.2. GALOIS THEORY 97

Lemma 3.15. Let F ⇢ K be fields and let H, J 2 G(K/F ) be such that H ⇢ J.

If [J : H] < 1, then [H 0 : J 0 ]  [J : H].
Proof. Let [J : H] = n and let 1 (= id), . . . , n be a system of representatives
of the left cosets of H in J. Assume to the contrary that [H 0 : J 0 ] > n. Let
✏1 , . . . , ✏n+1 2 H 0 be linearly independent over J 0 and let
2 3
1 (✏1 ) · · · 1 (✏n+1 )
6 . .. 7
A=6 . 7 2 Mn⇥(n+1) (K).
4 . . 5
n (✏1 ) · · · n (✏n+1 )

Let 0 6= c 2 K n+1 have the fewest nonzero components such that Ac = 0. We may
assume that 2 3
1
6 7
6c 2 7
6.7
6.7
6.7
6 7
c=6 7
6 cr 7 , ci 6= 0.
6 7
607
6.7
6.7
4.5
0
The first equation in Ac = 0 is ✏1 + ✏2 c2 + · · · + ✏r cr = 0. Hence not all c2 , . . . , cr
are in J 0 . (Otherwise, ✏1 , . . . , ✏r would be linearly dependent over J 0 .) Assume that
/ J 0 . Choose 2 J such that (c2 ) 6= c2 . Applying to Ac = 0 gives (A) (c) =
c2 2
0. Since 1 H, . . . , n H is a permutation of 1 H, . . . , n H, 1, . . . , n is a
permutation of 1 h1 , . . . , n hn for some h1 , . . . , hn 2 H. Hence 1 |H 0 , . . . , n |H 0
is a permutation of 1 |H 0 , . . . , n |H 0 . Thus (A) = [ i (✏j )] is a row permutation
of A. Therefore, (A) (c) = 0 implies that A (c) = 0. Now, A(c (c)) = 0,
where 2 3
0
6 7
6c2 (c2 )7
6 .. 7
6 7
6 . 7
6 7
c (c) = 6 c
6 r (c ) 7
r 7 6= 0
6 7
6 0 7
6 .. 7
6 7
4 . 5
0
has fewer nonzero components than c, which is a contradiction. ⇤
Corollary 3.16. Let K/F be a finite extension. Then |Aut(K/F )|  [K : F ].
The equality holds if and only if K/F is Galois.
Proof. By Lemma 3.14, |Aut(K/F )| = [F 0 : K 0 ]  [K : F ]. By Proposi-
tion 3.10 (iii) and Lemmas 3.14 and 3.15, [K 00 : F 00 ]  [F 0 : K 0 ] = [F 000 : K 000 ] 
[K 00 : F 00 ]. Hence [F 0 : K 0 ] = [K 00 : F 00 ] = [K : F 00 ]. Thus [F 0 : K 0 ] = [K : F ] ,
[K : F 00 ] = [K : F ] , F 00 = F . ⇤
Theorem 3.17 (Artin). Let K be a field and H < Aut(K). Then K/H 0 is
Galois. If |H| < 1, then Aut(K/H 0 ) = H.
98 3. FIELDS

Proof. Let F be the prime field of K and consider the extension K/F . Since
H 000 = H 0 , K/H 0 is Galois. If |H| < 1, by Proposition 3.10 (iv), Aut(K/H 0 ) =
H 00 = H. ⇤

3.3. Splitting Fields and Normal Extensions

Splitting fields. Let F be a field and S ⇢ F [x] \ F . An extension K F is
called a splitting field of S over F if
(i) every f 2 S splits in K, i.e., every f 2 S is a product of linear polynomials
in K[x];
(ii) K is generated by F and the roots of all f 2 S.
Equivalently, a splitting field of S over F is a smallest extension of F in which all
f 2 S splits.
Algebraically closed fields. A field F is called algebraically closed if every
f 2 F [x] \ F splits in F . The following statements are equivalent.
(i) F is algebraically closed.
(ii) Every f 2 F [x] \ F has a root in F .
(iii) The only algebraic extension of F is itself.
The fundamental theorem of algebra. C is algebraically closed, i.e., every
f 2 C[x] \ C has a root in C.
Proof. Assume to the contrary that f (z) 6= 0 for all z 2 C. Then 1/f (z) is
a bounded entire function. By Liouville’s theorem, 1/f (z) is a constant function,
which is a contradiction. ⇤
Algebraic closure. Let F be a field. The following two conditions on an
extension K/F are equivalent.
(i) K/F is algebraic and K is algebraically closed.
(ii) K is a splitting field of F [x] \ F over F .
(Proof of (ii) ) (i): Let u be an algebraic element over K. Then u is algebraic over
F . Since the minimal polynomial of u over F splits in K, we have u 2 K.) The
field K in (i) and (ii) is called an algebraic closure of F .
Examples. C is an algebraic closure of R. If F ⇢ K and K is algebraically
closed, then the (relative) algebraic closure of F in K is an algebraic closure of F .
The field of all algebraic numbers in C is an algebraic closure of Q.
Theorem 3.18 (Existence of algebraic closure). Every field F has an algebraic
closure.
Proof. To each f 2 F [x] \ F , assign an indeterminate Xf . Let X = {Xf :
f 2 F [x] \ F } and consider the polynomial ring F [X ]. Let I ⇢ F [X ] be the ideal
generated by f (Xf ), f 2 F [x] \ F . We claim that 1 2 / I. Otherwise, there exist
f1 , . . . , fn 2 F [x] \ F and g1 , . . . , gn 2 F [X ] such that
n
X
(3.3) gi fi (Xfi ) = 1.
i=1

Let K/F be an extension such that each fi (1  i  n) has a root ui 2 K. In (3.3),

let Xfi = ui , 1  i  n, and Xf = 0 for f 2 (F [x] \ F ) \ {f1 , . . . , fn }. Then 0 = 1,
which is a contradiction.
 3.3. SPLITTING FIELDS AND NORMAL EXTENSIONS 99

Let M be a maximal ideal of F [X ] containing I and let F1 = F [X ]/M . Then

F1 is an algebraic extension of F and every f 2 F [x] \ F has a root in F1 . By
the same construction, there is an algebraic S extension Fi+1 of Fi such that every
1
f 2 Fi [x] \ Fi has a root in Fi+1 . Then K = i=1 Fi is an algebraic closure of F .
Note: In fact, F1 is already algebraically closed; see Exercise 3.38. ⇤

An alternative proof. 1 . Clearly, if K/F is algebraic, then |K|  @0 |F |.

2 Choose a set S F such that |S| > @0 |F |. Let A be the class of all fields
K such that K ⇢ S and K is an algebraic extension of F . Then A is a set. For
K, L 2 A, define K L if K is a subfield of L. Then (A, ) is a poset in which
every chain has an upper bound (the union of the chain). By Zorn’s lemma, (A, )
has a maximal element E. We claim that E is an algebraic closure of F . Assume
to the contrary that there is an algebraic extension E1 /E such that E1 6= E. Since
E1 /F is algebraic, |E1 |  @0 |F | < |S|. Thus there is a one-to-one map f : E1 ! S
such that f |E = id. Define + and · in f (E1 ) by setting f (a) + f (b) = f (a + b) and
f (a)f (b) = f (ab) for all a, b 2 E1 . Then f (E1 ) 2 A and E f (E1 ), which is a
contradiction.
Note: We cannot simply consider the class of all algebraic extensions of F . It
is too big to be a set. ⇤

Corollary 3.19 (Existence of the splitting field). Let F be a field and S ⇢

F [x] \ F . Then there is a splitting field of S over F .
Proof. Let K be an algebraic closure of F and let R be the set of all roots in
K of all polynomials in S. Then F (R) is a splitting field of F . ⇤

Theorem 3.20 (Uniqueness of the splitting field). Let F be a field and S ⇢

F [x] \ F . Then any two splitting fields of S over F are F -isomorphic. (An
isomorphism between two extensions of F which is identity on F is called an
F -isomorphism.) In particular, the algebraic closure of F is unique up to F -
isomorphism.
Proof. This follows from the next theorem. ⇤

Theorem 3.21. Let : F1 ! F2 be an isomorphism of fields, S1 ⇢ F1 [x] \ F1

and S2 = { f : f 2 S1 } ⇢ F2 [x] \ F2 , where f is the resulting polynomial with
applied to the coefficients of f . Let K1 be a splitting field of S1 over F1 and K2 be a
splitting field of S2 over F2 . Then can be extended to an isomorphism K1 ! K2 .
Proof. Let
A = {(L1 , L2 , ⌧ ) : Li is a field between Fi and Ki and
⌧ : L1 ! L2 is an isomorphism such that ⌧ |F1 = }.
For (L1 , L2 , ⌧ ), 2 A, define (L1 , L2 , ⌧ ) (L01 , L02 , ⌧ 0 ) if L1 ⇢ L01 , L2 ⇢
(L01 , L02 , ⌧ 0 )
L2 and ⌧ |L1 = ⌧ . By Zorn’s lemma, the poset (A, ) has a maximal element
0 0

(E1 , E2 , ↵). It suffices to show that E1 = K1 and E2 = K2 .

Assume to the contrary that E1 6= K1 or E2 6= K2 , say E1 6= K1 . Then there
exists f 2 S1 such that f does not split in E1 . Let g 2 E1 [x] be an irreducible factor
of f with deg g 2 and let u 2 K1 \ E1 be a root of g. Let v 2 K2 be a root of
↵g 2 E2 [x]. By the next lemma, ↵ can be extended to an isomorphism : E1 (u) !
E2 (v). Then (E1 , E2 , ↵) (E1 (u), E2 (v), ), which is a contradiction. ⇤
100 3. FIELDS

Lemma 3.22. Let : F1 ! F2 be an isomorphism of fields. Let Ki be an

algebraic closure of Fi , i = 1, 2. Let f 2 F1 [x] be irreducible, u 2 K1 be a root
of f and v 2 K2 be a root of f . Then can be extended to an isomorphism
⌧ : F1 (u) ! F2 (v) such that ⌧ (u) = v.
Proof. Since f is the minimal polynomial of u over F1 and f is the minimal
polynomial of v over F2 , the map
⌧: F1 (u) ! F2 (v)
g(u) 7 ! ( g)(v), g 2 F1 [x]
is a well-defined isomorphism. ⇤
Proposition 3.23. Let f 2 F [x] \ F and K be the splitting field of f over F .
(i) If f is irreducible, then Aut(K/F ) acts transitively on the roots of f .
(ii) If Aut(K/F ) acts transitively on the roots of f and f has no multiple
roots, then f is irreducible.
Proof. (i) follows from Lemma 3.22 and Theorem 3.21.
(ii) Suppose to the contrary that f = gh, g, h 2 F [x] \ F . Then g and h do not
have common roots. Any 2 Aut(K/F ) maps a root of g to a root of g, not a root
of h, which is a contradiction. ⇤
Proposition 3.24. Let f 2 F [x] \ F and let K be the splitting field of f over
F.
(i) [K : F ] | (deg f )!.
(ii) Let f1 , . . . , fk be the distinct irreducible factors of f . Then [K : F ] |
(deg f1 )! · · · (deg fk )!.
Proof. (i) Use induction on deg f . If f is reducible, say f = gh, g, h 2 F [x]\F ,
let E be the splitting field of g over F . Then K is the splitting field of h over E. Thus
[K : F ] = [K : E][E : F ] | (deg g)!(deg h)! | (deg f )!. If f is irreducible, let u 2 K
be a root of f and write f = (x u)m, m 2 (F (u))[x]. Then [F (u) : F ] = deg f
and [K : F (u)] | (deg m)! since K is the splitting field of m over F (u). Hence
[K : F ] | (deg f )!.
(ii) Let E0 = F and Ei ⇢ K be the splitting field of fi over Ei 1. Then Ek = K
and by (i), [Ei : Ei 1 ] | (deg fi )!. ⇤
Proposition 3.25 (Normal extension). Let K/F be an algebraic extension.
Then the following statements are equivalent.
(i) If f 2 F [x] is irreducible and has a root in K, then f splits in K.
(ii) K is a splitting field over F of some S ⇢ F [x] \ F .
(iii) Let F be an algebraic closure of F containing K. Then for every 2
Aut(F /F ), (K) = K.
The field K in (i) – (iii) is called a normal extension of F .
Proof. (i) ) (ii). Let B be a basis of K/F . For each b 2 B, let fb be the
minimal polynomial of b over F . Then K is the splitting field of {fb : b 2 B} over
F.
(ii) ) (iii). Let X be the set of all roots of all f 2 S. Then K = F (X) and for
each 2 Aut(F /F ), (X) = X. Hence (K) = (F (X)) = F ( (X)) = F (X) =
K.
 3.3. SPLITTING FIELDS AND NORMAL EXTENSIONS 101

(iii) ) (i). By assumption, f has a root u 2 K. Let v 2 F be any root of

f . Let E ⇢ F be the splitting field of f over F . By Proposition 3.23 (i), there
exists ⌧ 2 Aut(E/F ) such that ⌧ (u) = v. By Theorem 3.21, ⌧ can be extended to
2 Aut(F /F ). Thus v = (u) 2 K. Hence f splits in K. ⇤
Proposition 3.26. Let K/F be a normal extension. Then every F -isomorphism
between two intermediate fields L1 and L2 can be extended to an automorphism
of K.
Proof. By Proposition 3.25, K is a splitting field of some S ⇢ F [x] \ F over
F . Then K is also a splitting field of S over L1 and over L2 . By Theorem 3.21,
extends to some ¯ 2 Aut(K/F ). ⇤
The converse of Proposition 3.26 is false: Q(21/3 )/Q is not normal and has no
proper intermediate subfields.
Separability. A polynomial f 2 F [x] is called separable if it has no multiple
roots (in any extension of F ), i.e., if gcd(f, f 0 ) = 1. If f 2 F [x] is irreducible, then
f is separable if and only if f 0 6= 0. Let K/F be an algebraic extension. An element
u 2 K is called separable over F if its minimal polynomial over F is separable. The
extension K/F is called separable if every u 2 K is separable over F .
Fact. If char F = 0, then every algebraic extension over F is separable.
Theorem 3.27 (Characterization of algebraic Galois extensions). Let K/F be
an algebraic extension. The following statements are equivalent.
(i) K/F is Galois.
(ii) K is a normal and separable extension over F .
(iii) K is a splitting field over F of a set of separable polynomials in F [x].
Proof. (i) ) (ii). For each u 2 K, we show that the minimal polynomial f
of u over F is separable and splits in K. Let {u1 , . . . , un } be the Aut(K/F )-orbit
of u. Let g(x) = (x u1 ) · · · (x un ). Then g = g for all 2 Aut(K/F ), whence
g 2 F [x]. Therefore f | g. (In fact, f = g since (u) is a root of f for every
2 Aut(K/F ).) Thus f is separable and splits in K.
(ii) ) (iii). Let B be a basis of K/F . For each b 2 B, let fb 2 F [x] be the
minimal polynomial of b over F . Then fb is separable and K is the splitting field
of {fb : b 2 B} over F .
(iii) ) (i). Let S ⇢ F [x] \ F be a set of separable polynomials such that K is
a splitting field of S over F .
1 Assume that [K : F ] = n < 1. We use induction on n. The case n = 1
needs no proof. Assume that n > 1. There exists f 2 S which does not split
in F . Let g 2 F [x] be an irreducible factor of f with deg g = r 2. Let
u1 , . . . , ur 2 K be the roots of g. For each 1  i  r, there is an F -isomorphism
i : F (u1 ) ! F (ui ) such that (u1 ) = ui . By Proposition 3.26, i can be extended
to an isomorphism ⌧i 2 Aut(K/F ). Clearly, ⌧i 1 ⌧j 62 Aut(K/F (u1 )) for i 6= j.
So ⌧1 , . . . , ⌧r represent different left cosets of Aut(K/F (u1 )) in Aut(K/F ). Thus
[Aut(K/F ) : Aut(K/F (u1 ))] r = [F (u1 ) : F ]. Since [K : F (u1 )] < n, by the
induction hypothesis, K/F (u1 ) is Galois. So
|Aut(K/F )| = [Aut(K/F ) : Aut(K/F (u1 ))]|Aut(K/F (u1 ))|
[F (u1 ) : F ][K : F (u1 )] = [K : F ].
102 3. FIELDS

By Corollary 3.16, K/F is Galois.

2 SFor each T ⇢ S, let KT ⇢ K be the splitting field of T over F . Then
K = T ⇢S, |T |<1 KT . For each u 2 K \ F , there exists T ⇢ S with |T | < 1
such that u 2 KT . Since [KT : F ] < 1, by 1 , there exists 2 Aut(KT /F )
such that (u) 6= u. Since K is the splitting field of S over KT , by Theorem 3.21
(or Proposition 3.26), can be extended to an isomorphism ⌧ 2 Aut(K/F ) and
⌧ (u) 6= u. Hence K/F is Galois. ⇤

Theorem 3.28 (Normal closure). Let K/F be an algebraic extension. Then

there exists an extension L/K such that
(i) L is normal over F ;
(ii) if K ⇢ M ⇢ L is such that M is normal over F , then M = L.
If L1 is another extension of K satisfying (i) and (ii), then L1 is K-isomorphic to
L. The field L is called a normal closure of K over F . Moreover,
(iii) if K/F is separable, then L/F is Galois;
(iv) if [K : F ] < 1, then [L : F ] < 1.

L
..
...
..
....
..

K
.....
...
..
...
.

F
Proof. Let B be a basis of K over F . For each b 2 B, let fb be the minimal
polynomial of b over F . Let L be a splitting field of {fb : b 2 B} over K. Then (i)
– (iv) are satisfied.
Assume that L1 is another extension of K satisfying (i) and (ii). Then L1 is
also a splitting field of {fb : b 2 B} over K. By Theorem 3.20, L and L1 are
K-isomorphic. ⇤

3.4. The Galois Group of a Polynomial

Let f 2 F [x] and K be a splitting field of f over F . The group Aut(K/F )
is called the Galois group of f over F and is also denoted by Aut(f /F ). Let
u1 , . . . , un 2 K be the distinct roots of f . Then the map : Aut(K/F ) !
S{u1 ,...,un } , 7! |{u1 ,...,un } is an embedding, whence Aut(K/F ) ⇢ Sn . If f is
irreducible, Aut(K/F ) is a transitive subgroup of Sn .
Discriminant. Let f 2 F [x] be of degree n Q > 0 and split as f = a0 (x
u1 ) · · · (x un ) in a splitting field K of f . Let = i<j (ui uj ) 2 K and
1
Y
D(f ) = 2 = ( 1) 2 n(n 1) (ui uj ).
i6=j

We claim that D(f ) 2 F . To see this, let D = D(f ) and assume that u1 , . . . , un
are all distinct. (Otherwise, D = 0.) Thus K/F is Galois. For each 2 Aut(K/F ),
( ) = (sign ) , so (D) = D. Thus D 2 F . The element D(f ) is called the
discriminant of f .
 3.4. THE GALOIS GROUP OF A POLYNOMIAL 103

Proposition 3.29. Let f 2 F [x] be a separable polynomial and let K be the

splitting field of f over F .
p
(i) = D(f ) 2 K and Aut(K/F ) \ An ⇢ Aut(K/F ( )).
(ii) Assume that char F 6= 2. Then Aut(K/F ) \ An = Aut(K/F ( )). In
particular, Aut(K/F ) ⇢ An if and only if D(f ) is a square in F , i.e., if
and only if 2 F .

.........
... K ............................................................................... {id}
....
.. ... ...
... ... ...
... .. ..
... .... ....
... .. ..
... ... ...
...
...
Galois F ( ) ................................................................................ F ( )0 = Aut(K/F ) \ An
...
...
...
... . .
... .... ....
... ... ...
... .. ..
... ... ...
... ... ...
... . .
...
....
.......
F ............................................................................... Aut(K/F )

Proof. (i) If 2 Aut(K/F ) \ An , then ( ) = , so 2 Aut(K/F ( )).

(ii) For each 2 Aut(K/F ), we have ( ) = sign( ) . Thus 2 F ( )0 ,

( ) = , sign( ) = 1 , 2 Aut(K/F ) \ An . (Note that 1 6= 1 since
char F 6= 2.) ⇤

Note. Proposition 3.29 (ii) is false if char F = 2. Example: f = x2 + x + 1 2

Z2 [x] is irreducible. Let K be the splitting field of f over Z2 and let ↵ 2 K be a
root of f . Then ↵2 is also a root of f and ↵2 6= ↵. We have = ↵ ↵2 = 1. Hence
Aut(K/F ( )) = Aut(K/F ) = S2 6= Aut(K/F ) \ A2 .

Resultant. Let F be a field. For a = (a0 , . . . , an ) 2 F n+1 and b = (b0 , . . . , bm ) 2

F m+1
, where m + n > 0, define
9
a0 a1 · · · an >
>
>
=
a0 a1 · · · an
m
· · · · · · >
>
>
;
a0 a1 · · · an 9
(3.4) R(a, b) = b0 b1 · · bm >
> .
>
>
b0 b1 · · bm >
>
=
· · · · · n
>
>
· · · · · >
>
>
>
b b · · b ; 0 1 m

If f = a0 xn + a1 xn 1 + · · · + an and g = b0 xm + b1 xm 1 + · · · + bm , where m + n > 0

and a0 , b0 6= 0, then R(a, b) is called the resultant of f and g and is denoted by
R(f, g).

Proposition 3.30. In the above notation, gcd(f, g) 6= 1 if and only if R(f, g) =

0.
104 3. FIELDS

Proof. ()) Let u be a common root of f and g (in some extension of F ).

Then 2 3
a0 a1 · · · an
6 a0 a1 · · · an 7
6 7
6 72 3
6 · · · · · · 7 m+n 1
6 7 u
6 an 7
6 a0 a1 · · · 76 .. 7
6 76 . 7
6 b0 b1 · · bm 76 7 = 0.
6 76 7
6 b0 b1 · · bm 74 u 5
6 7
6 · · · · · 7 1
6 7
6 7
4 · · · · · 5
b0 b1 · · bm
Hence R(f, g) = 0.
(() There exists 0 6= (↵0 , . . . , ↵m 1 , 0 , . . . , n 1 ) 2 F m+n such that
2 3
a0 a1 · · · an
6 a a · · · an 7
6 0 1 7
6 7
6 · · · · · · 7
6 7
6 a a · · · an 7
6 0 1 7
6 7
(3.5) (↵0 , . . . , ↵m 1 , 0 , . . . , n 1 ) 6 b0 b1 · · bm 7 = 0.
6 7
6 b0 b1 · · bm 7
6 7
6 · · · · · 7
6 7
6 7
4 · · · · · 5
b0 b1 · · bm
Let ↵ = ↵0 xm 1 + ↵1 xm 2 + · · · + ↵m 1 and = 0 xn 1 + 1 xn 2 + · · · + n 1 .
Then ↵, are not both 0 and deg ↵ < m, deg < n. Moreover, (3.5) is equivalent
to ↵f + g = 0. Hence gcd(f, g) 6= 1. ⇤
Proposition 3.31. Let x1 , . . . , xn , y1 , . . . , ym , X be independent indeterminates.
In F (x1 , . . . , xn , y1 , . . . , ym ) [X], write
(X x1 ) · · · (X xn ) = X n + a1 X n 1
+ · · · + an ,
(X y1 ) · · · (X y m ) = X m + b1 X m 1
+ · · · + bm ,
where ai = ( 1) sn,i (x1 , . . . , xn ), bj = ( 1) sm,j (y1 , . . . , ym ), sn,i is the ith ele-
i j

mentary symmetric polynomial in n indeterminates. Let a = (1, a1 , . . . , an ) and

b = (1, b1 , . . . , bm ). Then
n Y
Y m
(3.6) R(a, b) = (xi yj ).
i=1 j=1

Proof. When xi = yj , by Proposition 3.30, R(a, b) = 0. Thus in F [x1 , . . . , xn ,

y1 , . . . , ym ], xi yj | R(a, b). So the right side of (3.6) divides R(a, b). Since
deg(x1 ,...,xn ) R(a, b) = m deg(x1 ,...,xn ) an = mn,
Qn Qm
we must have R(a, b) = c i=1 j=1 (xi yj ) for some c 2 F [y1 , . . . , ym ]. Compar-
ing the coefficients of (x1 · · · xn )m gives c = 1. ⇤
 3.4. THE GALOIS GROUP OF A POLYNOMIAL 105

Corollary 3.32. Let f, g 2 F [x] \ F . Suppose that f and g split (in a splitting
field of f g) as
f = a0 (x u1 ) · · · (x un ), a0 2 F ⇥ ,
g = b0 (x v1 ) · · · (x vm ), b0 2 F ⇥ .
Then
n Y
Y m
(3.7) R(f, g) = am n
0 b0 (ui vj ).
i=1 j=1

Proof. In Proposition 3.31, let xi = ui , yj = vj . ⇤

Note. (3.7) can be written as
n
Y m
Y
(3.8) R(f, g) = am
0 g(ui ) = ( 1)mn bn0 f (vj ).
i=1 j=1

These formulas can be generalized as follows.

Corollary 3.33. Let f and g be as in Corollary 3.32 and write f = a0 xn +
· · · + an , g = b0 xm + · · · + bm . Let h = c0 xk + · · · + ck 2 F [x], k > 0. (We do not
assume that c0 6= 0.) Let a = (a0 , . . . , an ), b = (b0 , . . . , bm ), c = (c0 , . . . , ck ). Then
n
Y
(3.9) R(a, c) = ak0 h(ui ),
i=1

m
Y
(3.10) R(c, b) = ( 1)mk bk0 h(vj ).
j=1

Proof. If c0 6= 0, use (3.8). Now assume that c0 = 0. Clearly,

(3.11) R(a, c) = a0 R a, (c1 , . . . , ck ) ,

(3.12) R(c, b) = ( 1)m b0 R (c1 , . . . , ck ), b .

Use (3.11) and (3.12) repeatedly until ci 6= 0. Then use (3.8). ⇤
Theorem 3.34. Let f = a0 xn + a1 xn 1 + · · · + an 2 F [x], n 2, a0 6= 0.
Then
1
D(f ) = ( 1) 2 n(n 1) a0 2n+1 R(a, a0 ),
where a = (a0 , . . . , an ) and a0 = (na0 , (n 1)a1 , . . . , an 1 ).
Proof. Write f = a0 (x u1 ) · · · (x un ). Then by Corollary 3.33,
n
Y n Y
Y 1
R(a, a0 ) = an0 1
f 0 (ui ) = a2n
0
1
(ui uj ) = a2n
0
1
( 1) 2 n(n 1)
D(f ).
i=1 i=1 j6=i

⇤
Examples. For the quadratic f = x2 + bx + c, D(f ) = b2 4c. For the cubic
f = x3 + bx2 + cx + d, D(f ) = b2 c2 4c3 4b3 d 27d2 + 18bcd. If char F 6= 3, we
may write f = y 3 + qy + r, where y = x + b/3, and we have D(f ) = 4q 3 27r2 .
Galois groups of separable irreducible polynomials of degree  4. If
f 2 F [x] is a separable irreducible quadratic, clearly, Aut(f /F ) ⇠
= Z2 .
106 3. FIELDS

Proposition 3.35. Let f 2 F [x] be a separable irreducible cubic.

(i) If char F 6= 2, then
(
A3 if D(f ) 2 F (2) ,
Aut(f /F ) =
S3 if D(f ) 2
/ F (2) ,
where F (2) = {a2 : a 2 F }.
(ii) If char F = 2, we may assume that f = x3 + ax + b. Then
(
A3 if x2 + bx + a3 + b2 has a root in F ,
Aut(f /F ) =
S3 otherwise.
Proof. Since Aut(f /F ) is a transitive subgroup of S3 , we have Aut(f /F ) = S3
or A3 .
(i) follows from Proposition 3.29 (ii).
(ii) Let K be a splitting field of f over F , u1 , u2 , u3 2 K be the roots of f , and
G = Aut(K/F ). Let (
↵ = u1 u22 + u2 u23 + u3 u21 ,
= u1 u23 + u3 u22 + u2 u21 .
Since u1 + u2 + u3 = 0, we have ↵ = u1 u2 u3 , whence ↵ =6 . Moreover,
every 2 G permutes ↵, , and fixes ↵ and if and only if 2 A3 . Therefore
F (↵, )0 = G \ A3 . Let r(x) = (x ↵)(x ) 2 K[x]. Since r(x) is fixed by G, we
have r(x) 2 F [x]. In fact, direct computation shows that
r(x) = x2 + bx + a3 + b2 .
Hence r(x) has a root in F , F (↵, ) = F , G \ A3 = G , G = A3 . ⇤

K ................................................................................ {id}
... ...
... ...
.. ..
... ...
... ...
... ...

F (↵, ) ................................................................................ G \ A3
.. ..
... ...
.. ..
... ...
... ...
... ...
. .

...............................................................................
F G

Lemma 3.36. Let f = x4 + bx3 + cx2 + dx + e 2 F [x] be separable and let K

be a splitting field of f over F . Let u1 , . . . , u4 2 K be the roots of f and let
↵ = u1 u2 + u3 u4 , = u1 u3 + u2 u4 , = u1 u4 + u2 u3 .
(i) ↵, , are distinct and
(x ↵)(x )(x ) = x3 cx2 + (bd 4e)x b2 e + 4ce d2 2 F [x].
This polynomial is called the resolvent cubic of f .
(ii) Aut(K/F (↵, , )) = Aut(K/F ) \ V , where
V = {(1), (1, 2)(3, 4), (1, 3)(2, 4), (1, 4)(2, 3)}.
 3.4. THE GALOIS GROUP OF A POLYNOMIAL 107

Proof. (i) We have ↵ = (u1 u4 )(u2 u3 ) 6= 0. In the same way, ↵, ,

are all distinct. Clearly, S4 permutes ↵, , . The coefficients of (x ↵)(x )(x )
are symmetric functions of ↵, , , and hence are symmetric functions of u1 , . . . , u4 ,
which are polynomials in b, c, d, e. The actual computation of the coefficients of
(x ↵)(x )(x ) is tedious but straightforward.
(ii) If 2 Aut(K/F ) \ V , then fixes ↵, , , i.e., 2 Aut(K/F (↵, , )). It
remains to show that Aut(K/F (↵, , )) ⇢ V . Let 2 Aut(K/F (↵, , )). There
exists 2 V such that (u1 ) = u1 . We claim that = id. (Then = 1 2 V .)
Assume to the contrary that 6= id. Without loss of generality, assume that
(u2 ) = u3 . Then ↵ = (↵) = u1 u3 + u2 u4 = , which is a contradiction. ⇤
Proposition 3.37. Let f = x4 + bx3 + cx2 + dx + e 2 F [x] be irreducible and
separable and let g 2 F [x] be the cubic resolvent of f . Let E be a splitting field of
g over F and let m = [E : F ] = |Aut(g/F )|.
(i) If m = 6, then Aut(f /F ) = S4 .
(ii) If m = 3, then Aut(f /F ) = A4 .
(iii) If m = 1, then Aut(f /F ) = V .
(iv) If m = 2, then
(
⇠ D4 if f is irreducible over E,
Aut(f /F ) =
Z4 if f is reducible over E.
Proof. Let K E be a splitting field of f over F and let G = Aut(K/F ).
By Lemma 3.36 (ii), [G : G \ V ] = [E : F ] = m. Since G is a transitive subgroup
of S4 , we have 4 |G|; hence |G| = 4, 8, 12, 24. More precisely, G = S4 , A4 , V or
G⇠ = Z4 , D4 .

K {id}
..... .....
... ...
.. ..
... ...
. .
...................................................
E G\V
... ...
... ...
.. ..
... ...
.. ..

F G

(i) and (ii). Since 3 |G|, |G| = 12 or 24, so G = A4 or S4 . Then m = [G :

G \ V ] = |G|/4. Thus G = S4 when m = 6 and G = A4 when m = 3.
(iii) Since G ⇢ V , we have G = V .
(iv) Since [G : G \ V ] = 2, we have G ⇠
= D4 or G ⇠= Z4 . Moreover, f is
irreducible over E , Aut(f /E) (= G \ V ) acts transitively on the roots of f
,G\V =V ,G⇠ = D4 . ⇤
Theorem 3.38. Let p be a prime and let f 2 Q[x] be an irreducible polynomial
of degree p with precisely two nonreal roots. Then Aut(f /Q) = Sp .
Proof. Let G = Aut(f /Q). The complex conjugation, as an element of G, is a
transposition. Since G is a transitive subgroup of Sp , p | |G|, and hence G contains
an element of order p, which must be a p-cycle. Without loss of generality, assume
that (1, 2), (1, a2 , . . . , ap ) 2 G. There exists k 2 Z such that (1, a2 , . . . , ap )k =
(1, 2, b3 , . . . , bp ). Since (1, 2) and (1, 2, b3 , . . . , bp ) generate Sp , we have G = Sp . ⇤
108 3. FIELDS

Example 3.39. Let f = x5 4x 2 2 Q[x]. By Eisenstein’s criterion with

p = 2, f is irreducible over Q. It is easy to see that f has precisely 3 real roots.
Hence Aut(f /Q) = S5 .
The inverse Galois problem. Can every finite group G be realized as the
Galois group of some finite Galois extension K/Q? The answer is not known. For
many families of finite groups the answer is affirmative:
• Sn (Proposition 3.49), An ;
• finite solvable groups [28];
• many finite simple groups [23].

3.5. Finite Fields

Existence and uniqueness. Let F be a field with |F | < 1. Its prime field,
being finite, must be Zp for some prime p. Since F is finite, we have [F : Zp ] = n <
1, and hence |F | = pn .
Conversely, given a prime p and an integer n > 0, up to isomorphism, there
exists a unique field F with |F | = pn .
Proposition 3.40. Let p be a prime and n be a positive integer. The splitting
n
field of xp x 2 Zp [x] over Zp has precisely pn elements.
n
Proof. Let f = xp x and let F be the splitting field of f over Zp . Since
gcd(f 0 , f ) = gcd( 1, f ) = 1, f has pn distinct roots in F . Let E = {a 2 F : f (a) =
0}. We claim that F = E. It suffices to show that E is a field. (Then, since F is
the smallest field in which f splits, we have F = E.) In fact,
: F ! F
n
a 7 ! ap
is an automorphism of F , and E is the fixed field of in F . Hence E is a field. ⇤

Theorem 3.41. Given a prime p and an integer n > 0, all finite fields of order
pn are isomorphic.
Proof. Let F be a finite field with |F | = pn . Then Zp ⇢ F . Since F \ {0}
n
is a multiplicative group of order pn 1, we have ap 1 = 1 for all a 2 F \ {0}.
n n
Thus ap = a for all a 2 F , i.e., all elements of F are roots of f = xp x 2 Zp [x].
Therefore, F is a splitting field of f over Zp . Since all splitting fields of f over Zp
are isomorphic, the conclusion of the theorem follows. ⇤

We denote the finite field with pn elements by Fpn . In particular, Fp = Zp .

There is an Fp -module isomorphism (not a ring isomorphism) Fpn ⇠
= Fnp .
The multiplicative group of Fpn .
Proposition 3.42. F⇥ ⇥
pn is cyclic. A generator of Fpn is called a primitive
element of Fpn .
Proof. This follows from the next proposition. ⇤

Proposition 3.43. Let F be any field and G be a finite subgroup of F ⇥ . Then

G is cyclic.
 3.5. FINITE FIELDS 109

Proof. Assume to the contrary that G is not cyclic. By Theorem 1.27, G ⇠ =

G1 ⇥ G2 , where |G1 | = m, |G2 | = n and gcd(m, n) > 1. Let k = lcm(m, n). Then
k < mn = |G| and xk = 1 for all x 2 G. However, this is impossible since xk 1
cannot have more than k roots in F . ⇤
Corollary 3.44. Let p be a prime and n > 0 an integer. Then there is an
irreducible polynomial f 2 Fp [x] of degree n.
Proof. Let ↵ 2 Fpn be a primitive element. Clearly, Fpn = Fp (↵). Let
f 2 Fp [x] be the minimal polynomial of ↵ over Fp . Then f is irreducible and
deg f = [Fp (↵) : Fp ] = n. ⇤
Representation of elements of Fpn . Let f 2 Fp [x] be irreducible of degree
n. Then Fpn = Fp [x]/(f ). Each element in Fpn is uniquely of the form
n 1
c0 + c1 x + · · · + cn 1x + (f ), c i 2 Fp ;
such an element is usually denoted by (c0 , c1 , . . . , cn 1 ) 2 Fnp . See Table 3.1 for the
multiplication table of F23 = F2 [x]/(x3 + x + 1).
n
Let ↵ be a primitive element of Fpn . Then Fpn = {0, 1, ↵, . . . , ↵p 2 }. Elements
of Fpn in this form are convenient for multiplication but not for addition.

Table 3.1. Multiplication table of F23 = F2 [x]/(x3 + x + 1)

· 000 001 010 011 100 101 110 111

000 000 000 000 000 000 000 000 000
001 000 001 010 011 100 101 110 111
010 000 010 100 110 011 001 111 101
011 000 011 110 101 111 100 001 010
100 000 100 011 111 110 010 101 001
101 000 101 001 100 010 111 011 110
110 000 110 111 001 101 011 010 100
111 000 111 101 010 001 110 100 011

Lattice of finite fields.

Theorem 3.45. Let p be a prime and let Fp be the algebraic closure of Fp .
(i) For each integer n > 0, Fp has a unique subfield of order pn .
(ii) Let Fpm ⇢ Fp and Fpn ⇢ Fp . Then Fpm ⇢ Fpn if and only if m | n. In
general,
(3.13) Fpm \ Fpn = Fpgcd(m,n) ,

(3.14) Fpm Fpn = Fplcm(m,n) ,

where Fpm Fpn is the subfield of Fp generated Fpm [ Fpn .
Note. We already know that a finite field of order pn is unique up to isomor-
phism. However, Theorem 3.45 (i) states that in a given algebraic closure of Fp , a
finite field of order pn is not only unique up to isomorphism, but also unique as a
set.
110 3. FIELDS

Proof of Theorem 3.45. (i) By the proof of Theorem 3.41, a subfield of Fp

n
of order pn must be {a 2 Fp : ap = a}.
(ii) If Fpm ⇢ Fpn , then Fpn is an [Fpn : Fpm ]-dimensional vector space over Fpm .
Hence,
pn = |Fpn | = |Fpm |[Fpn :Fpm ] = pm[Fpn :Fpm ] ,
so n = m[Fpn : Fpm ].
m n
On the other hand, if m | n, then pm 1 | pn 1, and hence xp 1 1 | xp 1 1,
pm pn pm
i.e., x x|x x. Therefore, in Fp , the splitting field of x x is contained
n
in the splitting field of xp x, i.e., Fpm ⇢ Fpn .
To prove (3.13), first observe that Fpgcd(m,n) ⇢ Fpm \ Fpn . Let Fpm \ Fpn = Fps .
Since Fps ⇢ Fpm and Fps ⇢ Fpn , we have s | m and s | n, whence s | gcd(m, n).
Therefore Fpm \Fpn = Fps ⇢ Fpgcd(m,n) . Equation (3.14) is proved the same way. ⇤
n
1)/(pm 1)
If ↵ is a primitive element of Fpn and m | n, then ↵(p is a primitive
element of Fpm .
Automorphism group. Define
: Fp n ! Fp n
a 7 ! ap .
Clearly, 2 Aut(Fpn /Fp ); is called the Frobenius map of Fpn over Fp .
Theorem 3.46. The extension Fpn /Fp is Galois and Aut(Fpn /Fp ) = h i.
More generally, if m | n, then the extension Fpn /Fpm is Galois and Aut(Fpn /Fpm ) =
h m i.
n
Proof. Since xp x 2 Fp [x] is separable and since Fpn is the splitting field
pn
of x x over Fp , Fpn is Galois over Fp . Since |Aut(Fpn /Fp )| = [Fpn : Fp ] = n
and 2 Aut(Fpn /Fp ), to prove that Aut(Fpn /Fp ) = h i, it suffices to show that
o( ) = n, or equivalently, o( ) n. Since o( ) = id, we have
o( )
(3.15) 0= o( )
(a) a = ap a for all a 2 Fpn .
o( )
The polynomial xp x has at most po( ) roots in Fpn . Thus (3.15) implies that
n
p p o( )
, i.e., n  o( ).
If m | n, then Fp ⇢ Fpm ⇢ Fpn . Since Fpn /Fp is Galois, so is Fpn /Fpm . More-
over, Aut(Fpn /Fpm ) is a subgroup of Aut(Fpn /Fp ) of order n/m. Since Aut(Fpn /Fp ) =
h i is cyclic, its only subgroup of order n/m is h m i. Hence Aut(Fpn /Fpm ) =
h m i. ⇤
m
The automorphism m 2 Aut(Fpn /Fpm ) is defined by m
(a) = ap , a 2 Fpn ;
it is called the Frobenius map of Fpn over Fpm .
Realization of Sn as a Galois group of a polynomial over Q.
Lemma 3.47. Assume that f (x) 2 F [x] is of degree n with distinct roots
u1 , . . . , un in a splitting field K and let G = Aut(f /F ) < Sn . Let y1 , . . . , yn be
indeterminates and define
Y⇣ X ⌘
g(x, y1 , . . . , yn ) = x u (i) yi 2 K[x, y1 , . . . , yn ].
2Sn i
 3.5. FINITE FIELDS 111

Let R be the set of right cosets of G in Sn and define

Y⇣ X ⌘
(3.16) gR = x u (i) yi , R 2 R.
2R i
Q
Then g = R2R gR , where gR 2 F [y1 , . . . , yn ][x] and is irreducible over F (y1 , . . . , yn ).
Pn
Proof. Since K/F is Galois and since { i=1 u (i) yi : 2 R} is a G-orbit in
K[y1 , . . . , yn ], it follows that gR 2 F [y1 , . . . , yn ][x] and is irreducible over F (y1 , . . . , yn ).
⇤
In the above lemma, write f (x) = xn + an 1 xn 1 + · · · + a0 . Clearly, the coef-
ficients of g(x, y1 , . . . , yn ) are symmetric polynomials in u1 , . . . , un with coefficients
in Z. Thus
X
(3.17) g(x, y1 , . . . , yn ) = ci0 ,...,in (a0 , . . . , an 1 )xi0 y1i1 · · · ynin ,
i0 +···+in =n!

where ci0 ,...,in 2 Z[x0 , . . . , xn 1 ].

Note. In Lemma 3.47, if F = Q and f (x) 2 Z[x] is monic, then gR 2

Z[y1 , . . . , yn ][x]. To see this, note that Z[y1 , . . . , yn ] is a UFD whose fractional field
is Q(y1 , . . . , yn ). Since g 2 Z[y1 , . . . , yn ][x] is monic in x and gR 2 Q(y1 , . . . , yn )[x]
is a monic divisor of g, we have gR 2 Z[y1 , . . . , yn ][x]. (Another way to see the claim:
The coefficients of gR (as a polynomial in x, y1 , . . . , yn ) are in Q (by Lemma 3.47)
and they are integral over Z, hence they belong to Z.)
Proposition 3.48. Let f (x) 2 Z[x] be monic of degree n and p be a prime such
that the image f¯(x) 2 Fp [x] of f (x) has n distinct roots v1 , . . . , vn in a splitting field
E of f¯ over Fp . Then f has n distinct roots u1 , . . . , un in a splitting field K of
f over Q. Moreover, if u1 , . . . , un are ordered suitably and S{u1 ,...,un } is identified
with S{v1 ,...,vn } through the correspondence ui $ vi , then Aut(f¯/Fp ) ⇢ Aut(f /Q).

⇠
=
S{v1 ,...,vn } ............................................... S{u1 ,...,un }
. .
......... .........
.... ....
... ...
... ...
... ...
... ...
. .
.... ...... .... ......
.. ..
..
.....
Aut(E/Fp ) = Aut(f¯/Fp ) ............................................. Aut(f /Q) = Aut(K/Q)

Proof. Denote the reduction from Z to Fp by ( ). Since gcd(f¯, f¯0 ) = 1 in

Fp [x], we have gcd(f, f 0 ) = 1 in Q[x]. (Otherwise, the gcd of f and f 0 in Z[x] is
a monic polynomial of degree > 0. Then gcd(f¯, f¯0 ) 6= 1.) Therefore, f has no
multiple roots in K. Write f = xn + an 1 xn 1 + · · · + a0 and let
Y⇣ X ⌘
g(x, y1 , . . . , yn ) = x u (i) yi 2 K[x, y1 , . . . , yn ],
2Sn i
Y⇣ X ⌘
g(x, y1 , . . . , yn ) = x v (i) yi 2 E[x, y1 , . . . , yn ].
2Sn i
By (3.17), g 2 Z[x, y1 , . . . , yn ], g 2 Fp [x, y1 , . . . , yn ], and ḡ = g. Let G = Aut(f /Q)
and Ḡ = Aut(f¯/Fp ), and let R (resp., R̄) denote the set of right Q cosets of G (resp.,
Ḡ) in Sn . In the notation of Lemma 3.47, gḠ | g = ḡ = R2R gR . Since gḠ
is irreducible in Fp (y1 , . . . , yn )[x], we have gḠ | gR for some R 2 R. In (3.16),
112 3. FIELDS

by ordering u1 , . . . , un suitably, we may assume that id 2 R, i.e., R = G. Then

gḠ | gG . For each 2 Ḡ,
Y⇣ n
X ⌘ Y⇣ Xn ⌘
gḠ (x, y1 , . . . , yn ) = x v⌧ (i) yi = x v⌧ (i) y 1 (i)
⌧ 2Ḡ i=1 ⌧ 2Ḡ i=1

= gḠ (x, y 1 (1) ,...,y 1 (n) ),

which divides gG (x, y 1 (1) , . . . , y 1 (n) ) = gG Q
(x, y1 , . . . , yn ). Thus gḠ divides gG
and gG . However, in E(y1 , . . . , yn )[x], since R2R gR = ḡ = g has no multiple
roots, gR , R 2 R, are pairwise relatively prime. Hence we must have G = G , i.e.,
2 G. Thus we have proved that Ḡ ⇢ G. ⇤
Proposition 3.49. Let n > 3. Let f1 , f2 , f3 2 Z[x] be monic polynomials of
degree n such that
(i) f¯1 2 Z2 [x] is irreducible;
(ii) in Z3 [x], f¯2 = gh, where g is irreducible of degree n 1 and h is linear;
(iii) in Z5 [x],
(
kl if n is odd,
f¯3 =
kl1 l2 if n is even,
where k is irreducible of degree 2, l, l1 , l2 are irreducible of odd degree and
gcd(l1 , l2 ) = 1.
Let f = 15f1 + 10f2 + 6f3 2 Z[x]. Then the Galois group of f over Q is Sn .
Proof. Let G = Aut(f /Q). Note that f is monic of degree n and f ⌘ f1
(mod 2), f ⌘ f2 (mod 3), and f ⌘ f3 (mod 5). By Proposition 3.48, Aut(f¯1 /Z2 ) ⇢
G, so G contains an n-cycle ↵. Since Aut(f¯2 /Z3 ) ⇢ G, G contains an (n 1)-cycle
. Since Aut(f¯3 /Z5 ) ⇢ G, G contains an element of the form ⌧ , where ⌧ is a
transposition, o( ) is odd and ⌧ = ⌧ . It follows that ⌧ 2 G. It remains to
show that h↵, , ⌧ i = Sn . Assume that ⌧ = (1, 2). Replacing by ↵m ↵ m for
a suitable m 2 Z, we may assume that fixes 1. Write = (2, ⇤, . . . , ⇤). Then
⌧ = (1, 2)(2, ⇤, . . . , ⇤) = (1, 2, ⇤, . . . , ⇤) is an n-cycle. Now Sn is generated by (1, 2)
and (1, 2, ⇤, . . . , ⇤). ⇤
In Proposition 3.48, there is a natural way to associate the roots of f with those
of f¯ using algebraic number theory.
Proposition 3.50. Let f (x) 2 Z[x] be monic of degree n and let p be a prime
such that the image f¯(x) 2 Fp [x] of f (x) has no multiple roots. Then f (x) has
n distinct roots u1 , . . . , un 2 C. Let K = Q(u1 , . . . , un ), and let oK be the ring
of integers of K and P be a prime ideal of oK lying above pZ. Then oK /P is a
splitting field of f¯ over Fp and ūi = ui + P 2 oK /P, 1  i  n, are the roots of f¯.
Let D(P|p) = { 2 Aut(K/Q) : (P) = P} be the decomposition group of P | p.
Then we have an isomorphism
: D(P|p) ! Aut((oK /P)/Fp )
7 ! ( ),
where
( ): oK /P ! oK /P
a+P 7 ! (a) + P, a 2 oK .
 3.6. SEPARABILITY 113

Equivalently, when S{u1 ,...,un } is identified with S{ū1 ,...,ūn } through the correspon-
dence ui $ ūi , then Aut(f¯/Fp ) = D(P|p) < Aut(f /Q).
Qn
Proof.Q1 Write f = i=1 (x ui ), ui 2 C, and let K = Q(u1 , . . . , un ).
n
Then f¯ = i=1 (x ūi ). Since ū1 , . . . , ūn are distinct, so are u1 , . . . , un . Let
G = Aut(K/Q), D = D(P|p), and Ḡ = Aut((oK /P)/Fp ). For 2 D, permutes
u1 , . . . , un the same way ( ) permutes ū1 , . . . , ūn . Hence : D ! Ḡ is one-to-one.
2 Let D0 = {z 2 K : (z) = z for all 2 D}. We claim that for each
a 2 oD0 (= oK \ D0 ), there exists l 2 Z such that a ⌘ l (mod P). Let p = P \ D0
and let 1 (= id), 2 , . . . , m be the representatives of the left cosets of D in G.
Since D acts transitively on the prime ideals of oK lying above p, the only prime
ideal of oK lying above p is P. For 2  i  m, i 1 (P) 6= P and hence i 1 (P) 6 p.
Let pi = i 1 (P) \ oD0 , 2  i  m. Then pi 6= p. Thus there exists b 2 oD0
such that b ⌘ a (mod p) and b ⌘ 1 (mod pi ), 2  i  m, i.e., i (b) ⌘ 1 (mod P),
2  i  m. Then
m
Y
a ⌘ i (b) (mod P)
i=1
= ND0 /Q (b) 2 Z,
where ND0 /Q is the norm from D0 to Q.
3 We claim that : D ! Ḡ is onto. It suffices to show that |Ḡ|  |D|, i.e.,
[oK /P : Fp ]  [K : D0 ]. Let ✏1 , . . . , ✏t 2 oK be such that ✏¯1 , . . . , ✏¯t form a basis
of oK /P over Fp . We claim that ✏1 , . . . , ✏t are linearly independent over D0 . If,
to the contrary, a1 ✏1 + · · · + at ✏t = 0 for some a1 , . . . , at 2 D0 which are not all
zero, replacing a1 , . . . , at by ra1 , . . . , rat for a suitable 0 6= r 2 D0 , we may assume
that ai 2 oD0 for all i and ai 6⌘ 0 (mod p) for at least one i. By 2 , there exists
li 2 Z such that ai ⌘ li (mod P). Then li 6⌘ 0 (mod p) for at least one i and
l1 ✏¯1 + · · · + lt ✏¯t = 0, which is a contradiction.
4 Let E = Fp (ū1 , . . . , ūn ). Then the map
D ! Aut(E/Fp )
7 ! ( )|E
is one-to-one since permutes u1 , . . . , un the same way ( ) permutes ū1 , . . . , ūn .
Hence [E : Fp ] = |Aut(E/Fp )| |D| = [oK /P : Fp ]. Thus E = oK /P. ⇤

3.6. Separability
Definition 3.51. An algebraic element u over a field F is called purely insep-
arable over F if the minimal polynomial of u over F is (x u)n for some n > 0. An
extension K/F is called a purely inseparable if every u 2 K is purely inseparable
over F .
Example. Let char F = p and consider the fields F (x) F (xp ). The minimal
polynomial of x over F (xp ) is f (y) = y p xp 2 F (xp )[y]. Since f (y) = (y x)p , x
is purely inseparable over F (xp ).
Fact. If u is both separable and purely inseparable over F , then u 2 F .
Proposition 3.52. Let K/F be an extension with char F = p > 0 and let
n
u 2 K be algebraic over F . Then up is separable over F for some n 0.
114 3. FIELDS

Proof. Let f = a0 + a1 x + · · · be the minimal polynomial of u over F . Use

induction on deg f .
Assume u is not separable over F . Then 0 = f 0 = a1 + 2a2 x + 3a3 x2 + · · · . It
follows that ai = 0 whenever p - i. Hence f (x) = a0 + ap xp + a2p x2p + · · · = g(xp ),
where g 2 F [x] with deg g = (1/p) deg f < deg f . Since g(up ) = 0, by the induction
m m m+1
hypothesis, (up )p is separable over F for some m 0, where (up )p = up . ⇤
Proposition 3.53. Let K/F be an extension with char F = p > 0 and let
u 2 K be algebraic over F . Then the following statements are equivalent.
(i) u is purely inseparable over F .
n
(ii) up 2 F for some n 0.
n
(iii) The minimal polynomial of u over F is of the form xp a.
Proof. (i) ) (iii). Let f = (x u)m 2 F [x] be the minimal polynomial of u
over F . Write m = kpn , where p - k. Then
n n n n n
f = (xp up )k = (xp )k kup (xp )k 1
+ · · · 2 F [x].
n n n n
In particular, ku 2 F , and hence u 2 F . Thus x
p p p
up belongs to F [x] and
n n
divides f . It follows that f = xp up .
n
(iii) ) (ii). We have up = a 2 F .
n n
(ii) ) (i). Let f be the minimal polynomial of u over F . Then f | xp up =
pn
(x u) . Thus f = (x u) for some 1  m  p , i.e., u is purely inseparable
m n

over F . ⇤
Corollary 3.54. Let K/F be a finite purely inseparable extension, where char F =
p > 0. Then [K : F ] is a power of p.
Proof. Use induction on [K : F ]. Assume that [K : F ] > 1. Choose u 2 K \F .
By Proposition 3.53 (iii), [F (u) : F ] = pn for some n > 0. Since K/F (u) is purely
inseparable and [K : F (u)] < [K : F ], by the induction hypothesis, [K : F (u)] is a
power of p. So [K : F ] is a power of p. ⇤
Fact. Let F ⇢ E ⇢ K be fields such that both E/F and K/E are purely
inseparable. Then K/E is purely inseparable.
Proposition 3.55. Let K/F be an algebraic extension where char F = p > 0.
Then the following statements are equivalent.
(i) K is purely inseparable over F .
(ii) If u 2 K is separable over F , then u 2 F .
(iii) K is generated over F by a set of purely inseparable elements over F .
Proof. (i) ) (ii). Since u is separable and purely inseparable over F , we have
u 2 F.
n
(ii) ) (i). Let u 2 K. By Proposition 3.52, up is separable over F for some
n
n 0. By (ii), up 2 F . By Proposition 3.53, u is purely inseparable over F .
(i) ) (iii). Obvious.
(iii) ) (i). Assume that K = F (X), where X ⇢ K is a set of purely insep-
arable elements over F . Let P = {u 2 K : u is purely inseparable over F }. By
Proposition 3.53,
n
(3.18) P = {u 2 K : up 2 F for some n 0}.
 3.6. SEPARABILITY 115

It is clear from (3.18) that P is a subfield of K. Since P F and P X, we have

P F (X) = K. ⇤
Proposition 3.56. Let K/F be an extension and let X ⇢ K be a set of
separable elements over F . Then F (X)/F is separable.
Proof. For each u 2 X, let fu 2 F [x] be the minimal polynomial of u over F .
Then fu is separable. Let L F (X) be a splitting field of {fu : u 2 X} over F . By
Theorem 3.27, L/F is Galois and hence separable. So F (X)/F is separable. ⇤
Theorem 3.57. Let K/F be an algebraic extension. Let
S = {u 2 K : u is separable over F },
P = {u 2 K : u is purely inseparable over F }.
(i) S and P are subfields of K; S is separable over F ; P is purely inseparable
over F .
(ii) K is purely inseparable over S.
(iii) P \ S = F .
(iv) K is separable over P if and only if K = SP .
(v) If K is normal over F , then S/F and K/P are Galois and Aut(S/F ) ⇠ =
Aut(K/P ) = Aut(K/F ).

K
. .....
.... .....
in............. ..... ?
.....
... .....
.... .....
..... .

S P
..... ...
..... .....
..... .....
.. .....
sep ............ ..
...... in
.
.. ....

F
Proof. Assume char F = p > 0 since if char F = 0, all the conclusions are
obvious.
(i) By Propositions 3.56 and 3.55, F (S) ⇢ S and F (P ) ⇢ P , hence S and P
are subfields of K.
n
(ii) For each u 2 K, by Proposition 3.52, there exists n 0 such that up is
n
separable over F , i.e., up 2 S. By Proposition 3.53, u is purely inseparable over
S.
(iii) Obvious.
(iv) ()) Since K is both separable and purely inseparable over SP , we have
K = SP .
(() Every u 2 S is separable over F and hence separable over P . Thus
K = P (S) is separable over P .
(v) 1 Aut(K/F ) = Aut(K/P ). Let 2 Aut(K/F ) and u 2 P . Let f =
(x u)m be the minimal polynomial of u over F . Since (u) is a root of f , we have
(u) = u. Thus 2 Aut(K/P ). (Note: Here we do not need K/F to be normal.)
2 K/P is Galois. Let u 2 K \ P and let f be the minimal polynomial of
u over F . Since u is not purely inseparable over F , f has a root v 2 K such
that v 6= u. There is an F -isomorphism ⌧ : F (u) ! F (v) such that ⌧ (u) = v.
116 3. FIELDS

By Proposition 3.26, ⌧ extends to some 2 Aut(K/F ) = Aut(K/P ). We have

(u) = v 6= u. So K/P is Galois.
3 Aut(S/F ) ⇠= Aut(K/F ). For each 2 Aut(K/F ), clearly, (S) = S, so
|S 2 Aut(S/F ). The group homomorphism

✓: Aut(K/F ) ! Aut(S/F )
7 ! |S
is onto. (Since K/F is normal, every ⌧ 2 Aut(S/F ) extends to some 2 Aut(K/F ).)
To prove that ✓ is one-to-one, let 2 ker ✓. Then |S = id. By 1 , |P = id. Thus
|SP = id. However, by 2 , K/P is separable. By (iv), SP = K. Hence = id.
4 S/F is Galois. For each u 2 S \ F , we have u 2 K \ P (since S \ P = F ).
By 2 , there exists 2 Aut(K/P ) such that (u) 6= u. We have |S 2 Aut(S/F )
and |S (u) 6= u. ⇤

Corollary 3.58. Let F ⇢ E ⇢ K be fields such that both E/F and K/E are
separable. Then K/F is separable.
Proof. Let S = {u 2 K : u is separable over F }. Since K is separable over
E, K is separable over S. By Theorem 3.57 (ii), K is purely inseparable over S.
So K = S. ⇤

K .. ..
.... . ........
..... .. .....
..... .. .....
..
...... ....
. .....
...
. .
. .....
..
. . .....
.
.....
. .
.. .....
...
.... ..
. .....
.....
.. ..
..... . ...
...
S ..... .
.
..
.
.. P
.....
..... ...
. ...
.
.
.....
... .....
..
.......
E .....
...
.....
.....
..... .....
..... ....
... .....
F
Corollary 3.59. Let K/F be an algebraic extension with char F = p > 0.
(i) If K/F is separable, then K = F K p , where K p = {ap : a 2 K}.
(ii) If K = F K p and [K : F ] < 1, then K/F is separable.
(iii) u 2 K is separable over F if and only if F (up ) = F (u).
Proof. (i) Since K is separable over F , K is separable over F K p . Since K is
purely inseparable over K p , K is purely inseparable over F K p . Thus K = F K p .

K
...
...
..
...
.

F K .p..
..
..... .....
..... .....
..... .....
..... ....

F Kp
n
(ii) We have K = F K p for all n 1; see the remark below. Since [K : F ] < 1,
we can write K = F (u1 , . . . , um ) for some u1 , . . . , um 2 K. There exists n > 0 such
n n n n n
that upi is separable over F for all 1  i  m. Then K p = F p (up1 , . . . , upn ) is
n
separable over F . Thus K = F K p is separable over F .
 3.6. SEPARABILITY 117

(iii) In fact,
u is separable over F , F (u)/F is separable
, F (u) = F (F (u)p ) = F (up ) (by (i) and (ii)).
⇤

Remark. Let K/F be an extension with char F = p > 0.

n n
• K = F K p for some n 1 if and only if K = F K p for all n 1.
n n
• For u 2 K, F (up ) = F (u) for some n 1 if and only if F (up ) = F (u)
for all n 1.
n n
Proof. Assume that K = F K p for some n 1. Then K = F K p ⇢ F K p ,
2 2
so K = F K p . Thus K = F (F K p )p = F (F p K p ) = F K p , and the argument can
n n
be repeated. For the second claim, let L = F (u) and note that F Lp = F (up ). ⇤

Simple extensions. An extension K/F is called simple if K = F (a) for some

a 2 K.
Theorem 3.60. Let K/F be an algebraic extension. Then K/F is a simple
extension if and only if there are only finitely many intermediate fields between F
and K.
Proof. (() Let u 2 K be such that F (u) is a maximal simple extension
of F in K. Assume to the contrary that F (u) 6= K. Choose v 2 K \ F (u). If
|F | < 1, then |F (u, v)| < 1, whence F (u, v) is a simple extension over F , which is
a contradiction. Assume that |F | = 1. Among the intermediate fields F (u + av),
a 2 F , at least two are equal, say, F (u + a1 v) = F (u + a2 v), where a1 , a2 2 F ,
a1 6= a2 . Then F (u, v) = F (u + a1 v, u + a2 v) = F (u + a1 v), which is a simple
extension over F . This contradicts the maximality of F (u).
()) Let K = F (u) and let f (x) 2 F [x] be the minimal polynomial of u
over F . For any intermediate field B between F and F (u), let fB (x) = xn +
bn 1 xn 1 + · · · + b0 2 B[x] be the minimal polynomial of u over B. We claim
that B = F (b0 , . . . , bn 1 ). Clearly, B F (b0 , . . . , bn 1 ). Let B 0 = F (b0 , . . . , bn 1 ).
Since B 0 and B are between F and F (u), we have B 0 (u) = B(u) = F (u). Note that
[F (u) : B] = [B(u) : B] = deg fB
= [B 0 (u) : B 0 ] (since fB is also the minimal polynomial of B 0 )
= [F (u) : B 0 ].
It follows that B = B 0 .
Therefore, B is determined by fB , where fB is a monic factor of f (x). Since
f (x) has only finitely many monic factors, there are only finitely many intermediate
fields B between F and F (u). ⇤

Corollary 3.61. Every finite separable extension is a simple extension.

Proof. Let K/F be a finite separable extension. Let L be the normal closure
of K over F . Then L is a finite Galois extension over F . So there are only finitely
many fields between F and L, and of course, only finitely many fields between F
and K. ⇤
118 3. FIELDS

Separable and inseparable degrees. Let K/F be an algebraic extension

and S ⇢ K be the largest separable extension over F ; [K : F ]s := [S : F ] is called
the separable degree of K over F ; [K : F ]i := [K : S] is called the inseparable degree
of K over F . Note that [K : F ] = [K : F ]i [K : F ]s .

K
....
...
.. [K:F ]i
....

S
..
...
.. [K:F ]s
...
..

Let F ⇢ L ⇢ K be fields such that K/F is normal. Let Emb(L/F ) be the set
of all F -embeddings L ! K, i.e., embeddings L ! K which are identity on F . The
set Emb(L/F ) does not depend on K as long as we consider the fields in a fixed
algebraic closure of F .

Lemma 3.62. Let F ⇢ L ⇢ M ⇢ K be fields such that K/F is normal. Then

|Emb(M/F )| = |Emb(M/L)||Emb(L/F )|.

K
...
....

M
..
...
.

L
....
...

Proof. Since K/F is normal, every 2 Emb(L/F ) extends to some ¯ 2

Aut(K/F ). Define

✓: Emb(L/F ) ⇥ Emb(M/L) ! Emb(M/F )

( , ⌧) 7 ! ¯ |⌧ (M ) ⌧.

1 We claim that ✓ is one-to-one. Assume that ¯1 |⌧1 (M ) ⌧1 = ¯2 |⌧2 (M ) ⌧2 ,

where 1 , 2 2 Emb(L/F ) and ⌧1 , ⌧2 2 Emb(M/L). Then

1 = ¯1 |L = (¯1 |⌧1 (M ) ⌧1 )|L = (¯2 |⌧2 (M ) ⌧2 )|L = ¯2 |L = 2.

Now ¯1 |⌧1 (M ) ⌧1 = ¯1 |⌧2 (M ) ⌧2 implies that ¯1 (⌧1 (a)) = ¯1 (⌧2 (a)) for all a 2 M .
So, ⌧1 (a) = ⌧2 (a) for all a 2 M , i.e., ⌧1 = ⌧2 .

2 We claim that ✓ is onto. Given ↵ 2 Emb(M/F ), let = ↵|L 2 Emb(L/F ).

Let ⌧ = ¯ 1 |↵(M ) ↵. Then ⌧ 2 Emb(M/L) and ↵ = ¯ |⌧ (M ) ⌧ . ⇤

Proposition 3.63. Let F ⇢ L ⇢ K be fields such that [L : F ]s < 1 and K/F

is normal. Then |Emb(L/F )| = [L : F ]s .
 3.6. SEPARABILITY 119

K
....
...

L
....
..
.

S
..
...
.

F
Proof. Let S ⇢ L be the largest separable extension over F . Since L/S is
purely inseparable, it is easy to see that Emb(L/S) = {id}. (Let 2 Emb(L/S) and
let u 2 L. Since u is purely inseparable over S, the minimal polynomial of u over
S is f (x) = (x u)m for some m > 0. Since (u) is a root of f , we have (u) = u.)
By Lemma 3.62, |Emb(L/F )| = |Emb(L/S)||Emb(S/F )| = |Emb(S/F )|. Thus it
suffices to show that [S : F ] = |Emb(S/F )|.
By Corollary 3.61, S = F (u) for some u 2 S. Let f be the minimal poly-
nomial of u over F . Then f has n = deg f distinct roots u1 , . . . , un 2 K. Thus
Emb(F (u)/F ) = { 1 , . . . , n }, where i : F (u) ! F (ui ) is the F -isomorphism such
that i (u) = ui . Hence |Emb(F (u)/F )| = n = [F (u) : F ]. ⇤
Proposition 3.63 is false when [L : F ]s = 1.
Proposition 3.64. Let F ⇢ L ⇢ K be fields such that K/F is normal and
[L : F ]s = 1. Then |Emb(L/F )| = 2[L:F ]s .
Proof. For each Y ⇢ L, let C(Y ) ⇢ K be the set of all conjugates of the
elements in Y over F . (Two algebraic elements over F are called conjugates if they
have the same minimal polynomial over F .)
1 Let S be the largest separable extension of F in L and let X be a basis of
S over F . Define Q
✓ : Emb(S/F ) ! ✏2X C(✏)
7 ! (✏) ✏2X .
Clearly, ✓ is one-to-one, so
Y |X|
|Emb(L/F )| = |Emb(S/F )|  C(✏)  @0  (2@0 )|X| = 2@0 |X| = 2|X| = 2[L:F ]s .
✏2X

In the above, |Emb(L/F )| = |Emb(S/F )| since |Emb(L/S)| = 1; see the proof of

Proposition 3.63.
2 Fix s0 2 S. Let
Y= (Y, ) : s0 2 Y ⇢ S,  is a linear order on Y such that s0 is
the minimum element and for each y 2 Y, y 2
/ F C({z 2 Y : z < y}) .
For (Y1 , 1 ), (Y2 , 2 ) 2 Y, say (Y1 , 1 ) (Y2 , 2 ) if Y1 ⇢ Y2 and 1 is the
restriction of 2 . Then (Y, ) is a poset in which every chain has an upper bound.
By Zorn’s lemma, (Y, ) has a maximal element (Y, ).
We claim that |Y | [S : F ] = [L : F ]s . Otherwise,
(
< @0  [S : F ] if |Y | < 1,
[F (C(Y )) : F ]
 |Y |@0 = |Y | < [S : F ] if |Y | = 1.
120 3. FIELDS

So F (C(Y )) ( S. Choose y0 2 S \ F (C(Y )) and define y  y0 for all y 2 Y . Then

(Y [ {y0 }, ) 2 Y, which contradicts the maximality of (Y, ).
For each y 2 Y , since y is separable over F and y 2/ F C({z 2 Y : z < y}) ,
y has a conjugate ȳ 2 K over F C({z 2 Y : z < y}) Q such that ȳ 6= y. Using
Zorn’s lemma, it is easy to see that for every (fy )y2Y 2 y2Y {y, ȳ}, there exists
2 Emb(L/F ) such that (y) = fy for all y 2 Y . Thus
Y
|Emb(L/F )| {y, ȳ} = 2|Y | 2[L:F ]s .
y2Y

Proposition 3.65. Let F ⇢ L ⇢ K be fields such that K/F is algebraic. Then

(3.19) [K : F ]s = [K : L]s [L : F ]s ,

(3.20) [K : F ]i = [K : L]i [L : F ]i .

Proof. 1 We first prove (3.19). Let SK/F be the largest separable extension
of F in K.

K
in......... ...
.... ...
..... ..
...
...
SK/L ...
...
in.......... ..... ...
..... ...
.... .
..... sep ....... ....

(3.21) SK/F L
.....
..... in.......... ....
. ... ...
sep ....... ..... ..
...
...
SL/F ...
...
..... ...
..... ....
.
sep ....... ..

Since [K : F ]s = [SK/F : SL/F ][SL/F : F ], [K : L]s = [SK/L : L], and [L :

F ]s = [SL/F : F ], it suffices to show that [SK/F : SL/F ] = [SK/L : L]. Applying
Theorem 3.57 (iv) to the following diagram gives SK/L = SK/F L.
SK/L
in.......... ..... sep
.....
. .....
..... .....
..... .

(3.22) SK/F L
..... .
..... .....
... ....
sep ........ .....
..... in

SL/F

Let X ⇢ SK/F be linearly independent over SL/F with |X| < 1. We claim
that X is also linearly independent over L. (This means that SK/F and L are
linearly disjoint over SL/F . Also, this implies that [SK/L : L] = [SK/F : SL/F ]; see
Exercise 3.3 (i).) We have the following diagram.
 3.6. SEPARABILITY 121

L(X)
in.......... ..... sep
.....
. .....
..... .....
..... .

SL/F (X) L
.....
..... .....
... .....
sep ........ .....
..... in

SL/F

Extend X to a basis Y of SL/F (X)/SL/F . Since [SL/F (X) : SL/F ] and [L(X) : L]
are finite, by Proposition 3.63 and Lemma 3.62,
(3.23) [L(Y ) : L] = [L(X) : L] = |Emb(L(X)/L)| = |Emb(L(X)/SL/F )|
= |Emb(SL/F (X)/SL/F )| = [SL/F (X) : SL/F ] = |Y |.
Since a product of elements in Y is a linear combination of elements in Y with
coefficients in SL/F , the L-module hY i generated by Y is a ring. By Exercise 3.1
(i), hY i = L(Y ). Then by (3.23), Y is linearly independent over L. So X is linearly
independent over L.
2 We now prove (3.20). It suffices to show that in diagram (3.21), [SK/L :
SK/F ] = [L : SL/F ]. Since we have proved that in diagram (3.22), SK/F and L are
linearly disjoint over SL/F , it follows that L and SK/F are linearly disjoint over
SL/F (Theorem 3.67). So, [SK/L : SK/F ] = [SK/F L : SK/F ] = [L : SL/F ]. ⇤

Corollary 3.66. Let f 2 F [x] be monic and irreducible and let K be a splitting
field of f over F . Let u1 2 K be any root of f . Then
(i) f = [(x u1 ) · · · (x un )][F (u1 ):F ]i , where u1 , . . . , un 2 K are the distinct
roots of f and n = [F (u1 ) : F ]s ;
[F (u ):F ]
(ii) u1 1 i is separable over F .
Proof. We may assume that char F = p > 0.
(i) Let u1 , . . . , un 2 K be all the distinct roots of f . Then
[F (u1 ) : F ]s = |Emb(F (u1 )/F )| = n.
Write f = (x u1 )r1 · · · (x un )rn . For each 1  i  n, there exists an F -
isomorphism i : F (u1 ) ! F (ui ) such that i (u1 ) = ui . Extend i to ¯i 2
Aut(K/F ). Then
(x u1 )r1 · · · (x un )rn = f = ¯i f = (x ¯i (u1 ))r1 · · · (x ¯i (un ))rn .
It follow that ri = r1 . So f = [(x u1 ) · · · (x un )]r1 . Thus nr1 = deg f = [F (u1 ) :
F ] = [F (u1 ) : F ]s [F (u1 ) : F ]i , whence r1 = [F (u1 ) : F ]i .
(ii) In the notation of (i), we have f = (xr1 ur11 ) · · · (xr1 urn1 ) since r1 is a
power of p. Let g = (x ur11 ) · · · (x urn1 ). Then g 2 F [x] and its roots are all
distinct. Since ur11 is a root of g, ur11 is separable over F . ⇤

Linear disjointness. Let F ⇢ K be fields and let L and M be intermediate

fields. We say that L and M are linearly disjoint over F if for every X ⇢ L which
122 3. FIELDS

is linearly independent over F , X is linearly independent over M .

.
K ..
..... .....
..... .....
..... .....
..... .....
..... ....

L .. ..
M
..... ....
..... .....
..... .....
..... ....
.... .....

F
Theorem 3.67. Let F ⇢ K be fields and let L and M be intermediate fields.
Recall that L[M ] denotes the ring generated by M over L.
(i) M and L are linearly disjoint over F if and only if the L-map ↵ : L ⌦F
M ! L[M ], a ⌦ b 7! ab (a 2 L, b 2 M ), is an isomorphism.
(ii) If M and L are linearly disjoint over F , then so are L and M .
P
Proof. (i) Let X be a basis of M over F . Then L[M ] = x2X Lx and
⇣M ⌘ M M
L ⌦F M = L ⌦F Fx ⇠ = (L ⌦F F x) ⇠= Lx.
x2X x2X x2X
L
Moreover, when L ⌦F M is identified with x2X Lx through the above isomor-
phism, we have
M X
↵: Lx ! Lx
x2X x2X
✏x 7 ! x, x 2 X,
where ✏x = (ay y)y2X , ay = 1 if y = x and ay = 0 if y 6= x. Thus ↵ is an L-
isomorphism if and only if X is linearly independent over L. Since ↵ does not
depend on X, we have
↵ is an L-isomorphism
, X is linearly independent over L for all bases X of M over F
, M and L are linearly disjoint over F .
(ii) Consider the M -map : M ⌦F L ! M [L], b ⌦ a 7! ba (b 2 M, a 2 L),
and the F -isomorphism : L ⌦F M ! M ⌦F L, a ⌦ b 7! b ⌦ a (a 2 L, b 2 M ).
Since the following diagram commutes, ↵ is an isomorphism if and only if is an
isomorphism.
L ⌦F M .............↵..................... L[M ]
... ... ...
... ... ...
... ... ...
... ⇠ ... ...
.= ..... .....
..........
.. .. ..

M ⌦F L ................................... M [L]
⇤

3.7. Cyclotomic Extensions

Let F be a field. A splitting field of xn 1 over F is called a cyclotomic
extension of order n over F . If char F = p > 0 and n = mpt , where p - m, then
t
xn 1 = (xm 1)p . Hence the splitting field of xn 1 over F is that of xm 1
over F . Therefore, we assume that char F - n.
Let K be a cyclotomic extension of order n over F (char F - n) and let Un =
{u 2 K : un = 1}. Then |Un | = n since xn 1 has no multiple roots. Since Un is
 3.7. CYCLOTOMIC EXTENSIONS 123

a finite subgroup of K ⇥ , Un is cyclic. A generator of Un is called a primitive nth

root of unity.
Proposition 3.68. Let K be a cyclotomic extension of order n over F , where
char F - n.
(i) K/F is Galois.
(ii) K = F (⇣), where ⇣ is any primitive nth root of unity.
(iii) Let
✓ : Aut(K/F ) ! Z⇥ n
7 ! i,
where (⇣) = ⇣ i . Then ✓ is a one-to-one group homomorphism. In par-
ticular, [K : F ] | (n), where is the Euler function.
Cyclotomic polynomials. Let K = F (⇣), where ⇣ is a primitive nth root of
unity and char F - n. The polynomial
Y
n (x) = (x u)
u2h⇣i
o(u)=n

is called the nth cyclotomic polynomial over F .

Fact.
Q
(i) xn 1= d|n d (x).
(ii)
xn 1 Y
n (x) = Y = (xd 1)µ(n/d) ,
d (x) d|n
d|n, d<n
where µ is the Möbius function.
(iii) If char F = 0, n (x) 2 Z[x]; if char F = p > 0, n (x) 2 Zp [x].
Proof. (i) We have
Y Y Y Y
xn 1= (x u) = (x u) = d (x).
u2h⇣i d|n u2h⇣i d|n
o(u)=d
Q
(ii) The formula n (x) = d|n (x
d
1)µ(n/d) follows from (i) and the Möbius
inversion.
(iii) Assume that char F = 0. (The proof in the case char F = p is the same.)
Use induction on n. We have
Y
xn 1 = n (x) d (x).
d|n
d<n
Q
Since xn 1 2 Z[x] and since d|n, d<n d (x) 2 Z[x] is monic (by the induction
hypothesis), we have n (x) 2 Z[x]. ⇤
Cyclotomic extensions in characteristic 0.
Theorem 3.69. Let ⇣n = e2⇡i/n 2 C.
(i) n (the nth cyclotomic polynomial over Q) is the minimal polynomial of
⇣n over Q.
124 3. FIELDS

(ii) [Q(⇣n ) : Q] = (n) and Aut(Q(⇣n )/Q) ⇠ n.

= Z⇥
Proof. We only have to show that n is irreducible in Q[x]. Let f 2 Q[x]
be a monic irreducible factor of n and write n = f g, where g 2 Q[x] is monic.
Since n (x) 2 Z[x], it follows that f, g 2 Z[x] (Write f = (k/l)f1 , g = (s/t)g1 ,
where k, l, s, t 2 Z+ , gcd(k, l) = 1, gcd(s, t) = 1, and f1 , g1 2 Z[x] are primitive.
Since f and g are monic, we have k = s = 1. Then (1/lt)f1 g1 = f g 2 Z[x], whence
l = t = 1.)
Let p be a prime such that p - n. We claim that if u is a root of f , then so is up .
Suppose to the contrary that up is not a root of f . Then up is a root of g, i.e., u is
a root of g(xp ). Then f (x) | g(xp ). Let f¯ denote the reduction of f in Zp [x]. Then
in Zp [x], f¯(x) | ḡ(xp ) = ḡ(x)p , whence gcd(f¯, ḡ) 6= 1. Then n = f¯ḡ has multiple
roots. Since n | xn 1, it follows that xn 1 2 Zp [x] has multiple roots. But this
is impossible since p - n. So the claim is proved.
By the above claim, if u is a root of f , then so is ur for all r with gcd(r, n) = 1.
Thus deg f (n), and hence f = n . ⇤
Corollary 3.70. Let F be a field with char F = 0. Let ⇣ be a primitive nth
root of unity in some extension of F . Then
(n)
[F (⇣) : F ] = [Q(⇣) : Q(⇣) \ F ] = .
[Q(⇣) \ F : Q]
F (⇣)
... .....
..... .....
..... .....
..... .....
..
......
. .....
.....
.... .....
..... .

Q(⇣) F
... ..... ....
... ...... .....
... ..... .....
... ...... .....
... .....
..... ..
......
... ...
... .... .....
...
...

(n) ......
...
.. Q(⇣) \ F
...
... ....
... ...
... ..
...
... ...
..

Q
Proof. It suffices to show that [F (⇣) : F ] [Q(⇣) : Q(⇣) \ F ]. Let f 2 F [x] be
Qn 1
the minimal polynomial of ⇣ over F . Since f is a factor of xn 1 = i=0 (x ⇣ i ),
we have f 2 Q(⇣)[x], so f 2 Q(⇣) \ F [x]. Thus [Q(⇣) : Q(⇣) \ F ]  deg f =
[F (⇣) : F ]. ⇤
p p
Example. Let ⇣8 = e2⇡i/8 = (1+i)/ 2 and F = Q( 2). Then Q(⇣8 )\F = F .
So [F (⇣8 ) : F ] = (8)/[F : Q] = 2.
Cyclotomic extensions in characteristic p.
Fact. Assume that p - n and let ⇣n be a primitive nth root of unity in some
extension of Fp . Let on (p) be the order of p in Z⇥
n . Then Fp (⇣n ) = Fpon (p) . More
generally, Fpm (⇣n ) = Fplcm(m,on (p)) .
Proof. ⇣n 2 Fpk , n | pk 1 , on (p) | k. ⇤
Corollary 3.71. Let char F = p > 0. Let ⇣ be a primitive nth root of unity
in some extension of F , where p - n. Assume that Fpon (p) \ F = Fpm . Then
[F (⇣) : F ] = on (p)/m.
 3.7. CYCLOTOMIC EXTENSIONS 125

F (⇣)
.... .....
..... .....
..... .....
..... .....
..
...... .....
.....
.... .....
.... .

Fpon (p) = Fp (⇣) F

..... ....
..... .....
..... .....
..... .....
.....
..... ..
.....
.
..... ...
. .....

Fp (⇣) \ F = Fpm

Proof. Same as the proof of Corollary 3.70 ⇤

Abelian extensions. An abelian extension is an algebraic Galois extension

K/F such that Aut(K/F ) is abelian. Subextensions of an abelian extension are
abelian (Exercise 3.10). An extension K of F contained in a cyclotomic extension
of F is a finite abelian extension over F . The converse is true for F = Q.
The Kronecker-Weber Theorem. If K/Q is a finite abelian extension, then
K ⇢ Q(⇣n ) for some n > 0, where ⇣n = e2⇡i/n .
The proof of the above theorem requires algebraic number theory [30, Ch.14].
k
Ruler and compass construction of regular polygons. Let Fk = 22 + 1,
k 0, be the kth Fermat number. The only known primes in the sequence Fk
are F0 , . . . , F4 . For 5  k  32 and many other values of k, Fk are known to be
composite. A prime of the form Fk is called a Fermat prime.
Proposition 3.72. ⇣n = e2⇡i/n is constructible by ruler and compass if and
only if n = 2a p1 · · · ps , where a 0 and p1 , . . . , ps are distinct Fermat primes.
Proof. 1 We first show that ⇣n is constructible if and only if (n) is a power
of 2.
()) By Theorem 3.69 (ii) and Corollary 3.7, (n) = [Q(⇣n ) : Q] is a power of
2.
(() Let (n) = 2m . By Theorem 3.69 (ii), Q(⇣n )/Q is a Galois extension and
Aut(Q(⇣n )/Q) is an abelian group of order 2m . Thus there are subgroups
1 = H0 < H1 < · · · < Hm = Aut(Q(⇣n )/Q)
such that [Hi : Hi 1] = 2. So we have a tower of fields
0
Q = Hm ⇢ · · · ⇢ H10 ⇢ H00 = Q(⇣n )
such that [Hi0 1 : Hi0 ] = 2. Hence by Theorem 3.6, ⇣n is constructible.
2 Let n = 2a pe11 · · · pess , where p1 , . . . , ps are distinct odd primes and ej > 0.
Then
(n) = 2a 1 pe11 1 (p1 1) · · · pess 1 (ps 1).
Therefore, (n) is a power of 2 if and only if e1 = · · · = es = 1 and pj = 2tj + 1,
1  j  s. Note that if 2t + 1 is a prime, then t is a power of 2. (If t = uv, where
u is odd, then 2v + 1 | 2uv + 1.) So pj = 2tj + 1 is a prime if and only if pj is a
Fermat prime. ⇤
126 3. FIELDS

3.8. Trace and Norm, Cyclic Extensions

Let F ⇢ K ⇢ F be fields where [K : F ] < 1 and F is an algebraic closure of
F . Let r = [K : F ]s and Emb(K/F ) = { 1 , . . . , r }. For each u 2 K, define
TrK/F (u) = [K : F ]i 1 (u) + ··· + r (u) ,
[K:F ]i
NK/F (u) = 1 (u) · · · r (u) ,
which are called the trace and the norm of u from K to F , respectively. It follows
from the next proposition that TrK/F (u), NK/F (u) 2 F .
Proposition 3.73. Let [K : F ] < 1 and u 2 K. Let f = xn + an 1x
n 1
+
· · · + a0 2 F [x] be the minimal polynomial of u over F . Then
(3.24) TrK/F (u) = [K : F (u)]an 1,

⇥ ⇤[K:F (u)]
(3.25) NK/F (u) = ( 1)n a0 .
Proof. Let r = [F (u) : F ]s and let Emb(F (u)/F ) = { 1 , . . . , r }. By Corol-
lary 3.66 (i),
⇥ ⇤[F (u):F ]i
f = (x 1 (u)) · · · (x r (u)) .
So,
r
X h Yr i[F (u):F ]i
an 1 = [F (u) : F ]i j (u), a0 = ( 1)r j (u) .
j=1 j=1

Let Emb(K/F (u)) = {⌧1 , . . . , ⌧t }, where t = [K : F (u)]s . Let ¯j 2 Aut(F /F ) be

an extension of j . By the proof of Lemma 3.62, Emb(K/F ) = {¯j |⌧k (K) ⌧k : 1 
j  r, 1  k  t}.
F
..
...
..

K
....
...
.

F (u)
..
...
..

F
Then
X X
TrK/F (u) = [K : F ]i ↵(u) = [K : F ]i ¯j (⌧k (u))
↵2Emb(K/F ) 1jr
1kt
X X
= [K : F ]i · t j (u) = t [K : F (u)]i [F (u) : F ]i j (u)
1jr 1jr
= [K : F (u)]an 1.

The proof of (3.25) is the same. ⇤

Fact. Let [K : F ] < 1.
(i) For u, v 2 K and a, b 2 F ,
TrK/F (au + bv) = a TrK/F (u) + b TrK/F (v),
NK/F (uv) = NK/F (u)NK/F (v).
 3.8. TRACE AND NORM, CYCLIC EXTENSIONS 127

(ii) If u 2 F , then TrK/F (u) = [K : F ]u and NK/F (u) = u[K:F ] .

(iii) (Transitivity) Let F ⇢ K ⇢ L where [L : F ] < 1. Then for each u 2 L,
TrL/F (u) = TrK/F TrL/K (u) ,
NL/F (u) = NK/F NL/K (u) .

Proof. (iii) Let Emb(K/F ) = { 1 , . . . , r } and Emb(L/K) = {⌧1 , . . . , ⌧t }.

Extend j to ¯j 2 Aut(F /F ). Then Emb(L/F ) = {¯j |⌧k (L) ⌧k : 1  j  r, 1 
k  t}.
F̄
...
...
.

L
..
...
...

K
.....
...
.

F
Therefore
X X
TrK/F TrL/K (u) = [K : F ]i j (TrL/K (u)) = [K : F ]i ¯j (TrL/K (u))
1jr 1jr
X ⇣ X ⌘
= [K : F ]i ¯j [L : K]i ⌧k (u)
1jr 1kt
X
= [L : F ]i ¯j (⌧k (u)) = TrL/F (u).
1jr
1kt

Cyclic extensions. A cyclic extension is an algebraic Galois extension K/F

such that Aut(K/F ) is cyclic.
Theorem 3.74. Let K/F be a finite cyclic extension with Aut(K/F ) = h i,
and let u 2 K.
(i) TrK/F (u) = 0 if and only if u = v (v) for some v 2 K.
(ii) (Hilbert’s Theorem 90) NK/F (u) = 1 if and only if u = v/ (v) for some
v 2 K ⇥.
Proof. Let n = [K : F ].
(i) We show that the sequence of F -maps
id TrK/F
0 ! F ,! K !K !F !0
is exact.
1 We claim that TrK/F : K ! F is onto. Since 0 , . . . , n 1 are distinct
automorphisms of K, by Proposition 3.13, they are linearly independent over K as
K-valued functions. Thus TrK/F = 0 + · · · + n 1 6= 0. Hence TrK/F : K ! F is
onto.
2 We have ker(id ) = {v 2 K : (v) = v} = F since K/F is Galois.
128 3. FIELDS

3 Clearly, im(id ) ⇢ ker TrK/F . However, by 1 and 2 ,

dimF (ker TrK/F ) = n 1 = dimF im(id ) .
So im(id ) = ker TrK/F .
(ii) We show that
id/ NK/F
1 ! F ⇥ ,! K ⇥ ! K⇥ ! F⇥
is exact. It suffices to show that ker NK/F ⇢ im(id/ ). Let u 2 ker NK/F . Define
↵: K ! K
x 7 ! u (x).
Then ↵i = u (u) · · · i 1 (u) i and ↵n = id. Since 0 , . . . , n 1 are linearly inde-
pendent over K, so are ↵0 , . . . , ↵n 1 . Hence there exists x 2 K such that
v := (↵0 + · · · + ↵n 1
)(x) 6= 0.
Clearly, ↵(v) = v, i.e., u (v) = v. So u = v/ (v). ⇤
In general, NK/F : K ⇥ ! F ⇥ is not onto. Example: NC/R (z) = |z|2 , z 2 C.
Proposition 3.75. Let F be a field containing a primitive nth root of unity ⇣
(so charF - n).
(i) K/F is a cyclic extension of degree n if and only if K = F (u), where u
is a root of an irreducible polynomial of the form xn a 2 F [x].
(ii) If un 2 F , then Aut(F (u)/F ) ,! Zn .
In Proposition 3.75 (ii), if F does not contain a primitive nth root of unity, it is
possible that F (u)/F is Galois but Aut(F (u)/F ) is not abelian. See Exercise 3.16.
Proof of Proposition 3.75. (i) (() Obviously, K/F is Galois. By (ii),
Aut(K/F ) = Zn .
()) Let Aut(K/F ) = h i. Since NK/F (⇣) = ⇣ n = 1, by Hilbert’s Theorem 90,
⇣ = (u)/u for some u 2 K ⇥ , so (u) = ⇣u. Since i (u) = ⇣ i u, 0  i  n 1,
are distinct conjugates of u over F , [F (u) : F ] n. Thus K = F (u). Since
(un ) = (u)n = (⇣u)n = un , we have un 2 F . Let a = un . Then xn a 2 F [x] is
the minimal polynomial of u over F .
(ii) For each 2 Aut(F (u)/F ), (u) = ⇣ i u for some i 2 Zn . The embedding
Aut(F (u)/F ) ,! Zn is given by 7! i. ⇤
Fact 3.76. If char F = p > 0, a polynomial of the form xp x a 2 F [x] is
either irreducible or splits in F .
Proof. Assume that f = xp x a does not split in F and let u 2 / F be a
root of f . Then u + i, i 2 Fp , are all theQroots of xp x a. Let g 2 F [x] be a
monic irreducible factor of f . Then g = i2S (x (u + i)) for some ; =6 S ⇢ Fp .
Since ⇣ X ⌘
g = x|S| |S|u + i x|S| 1 + · · · ,
i2S
we have |S|u 2 F . Since u 2
/ F , we must have |S| = p. Hence f = g, which is
irreducible. ⇤
 3.9. RADICAL EXTENSIONS 129

Theorem 3.77 (Artin-Schreier). Assume that char F = p > 0. Then K/F is

a cyclic extension of degree p if and only if K = F (u), where u 2 K \ F and u is a
root of a polynomial of the form xp x a 2 F [x].

Proof. (() By Fact 3.76, xp x a is irreducible over F but splits in F (u).

Hence [F (u) : F ] = p and F (u) is the splitting field of xp x a over F . Thus
K/F is Galois of degree p, which must be cyclic.
()) Let Aut(K/F ) = h i. Since TrK/F (1) = p = 0, by Theorem 3.74 (i),
1 = (u) u for some u 2 K. Clearly u 2
/ F . We have
(up u) = (u)p (u) = (u + 1)p (u + 1) = up u,
whence up u 2 F . Let a = up u. Then u is a root of xp x a 2 F [x]. Since
F ( F (u) ⇢ K and [K : F ] = p, we have F (u) = K. ⇤

3.9. Radical Extensions

Definition 3.78. Let K/F be a finite extension. K is called a radical extension
over F if K = F (u1 , . . . , un ) such that for each 1  i  n,
(i) um
i
i
2 F (u1 , . . . , ui 1 ) for some mi > 0 or
p
(ii) ui ui 2 F (u1 , . . . , ui 1 ), where char F = p.

Assume that char F = 0. If K/F is a radical extension, then every element in

p
K can be expressed in terms of elements in F using +, , ⇥, ÷, m . Let f 2 F [x].
If the splitting field of f over F is contained in a radical extension over F , then the
equation f (x) = 0 is solvable by radicals.
An extension K/F (with no assumption on char F ) is called solvable by radicals
if K is contained in a radical extension over F .

Fact 3.79. Let E1 , E2 be intermediate fields of F ⇢ K such that both E1 and

E2 are radical over F . The E1 E2 is also radical over F .

Proof. Let E1 = F (u1 , . . . , um ) and E2 = F (v1 , . . . , vn ), where u1 , . . . , um

and v1 , . . . , vn satisfy the conditions in Definition 3.78. Let (w1 , . . . , wm+n ) =
(u1 , . . . , um , v1 , . . . , vn ). Then E1 E2 = F (w1 , . . . , wm+n ) and w1 , . . . , wm+n satisfy
the conditions in Definition 3.78. ⇤

Theorem 3.80 (Galois). Let K/F be a finite extension and K 0 be the normal
closure of K over F . Then K/F is solvable by radicals if and only if Aut(K 0 /F )
is solvable.

Proof. ()) 1 Assume F ⇢ K ⇢ L, where L is a radical extension over F .

Let L = F (u1 , . . . , un ), where ui satisfies (i) or (ii) in Definition 3.78. We may
assume that each mi in (i) of Definition 3.78 is a prime.
Let N be the normal closure of L over F . We claim that N is radical over F . Let
v1 , . . . , vm be all the conjugates of u1 , . . . , un over F . Then N = F (v1 , . . . , vm ). For
each 1  j  m, there exists u 2 {u1 , . . . , un } such that u and vj are conjugates over
F . Then is an F -isomorphism j : F (u) ! F (vj ). Extend j to ¯j 2 Aut(N/F )
and let Lj = ¯j (L). Then Lj is radical over F . Since vj 2 F (vj ) ⇢ ¯j (L) = Lj , we
have N = L1 · · · Lm . By Fact 3.79, N is radical over F . Replacing L with N , we
130 3. FIELDS

may assume that L/F is radical and normal. We may assume that K 0 ⇢ L.

.........
...
L
.. ..
... ...
... ...
...
...
normal ...
...
...
K0
... ...
...
...
...
. normal
...
.........
F

2 By 1 , Aut(K 0 /F ) ⇠ = Aut(L/F )/Aut(L/K 0 ). So it suffices to show that

Aut(L/F ) is solvable. Let P ⇢ L be the largest purely inseparable extension over
F . Then L/P is Galois and Aut(L/P ) = Aut(L/F ) (Theorem 3.57). Note that
L/P is still radical. Replacing F with P , we may assume that L/F is Galois. Hence
we may assume that the mi ’s are primes 6= char F .

3 Let m be the product of all mi ’s and let ⇣ be a primitive mth root of

unity. Since Aut(L/F ) ⇠ = Aut(L(⇣)/F ) Aut(L(⇣)/L), it suffices to show that
Aut(L(⇣)/F ) is solvable. Since Aut(L(⇣)/F ) Aut(L(⇣)/F (⇣)) ⇠
= Aut(F (⇣)/F ) is
abelian, it suffices to show that Aut(L(⇣)/F (⇣)) is solvable.

L(⇣)
.. .....
..... .....
..... .....
..... .....
..
......
. .....
.....
.... .

L F (⇣)
..... .....
..... .....
..... .....
..... ....
..... ..
..
..... ..
. .....

Let Gi = Aut(L(⇣)/F (⇣, u1 , . . . , ui )). Since F (⇣, u1 , . . . , ui ) is normal over

F (⇣, u1 , . . . , ui 1 ), we have Gi C Gi 1 and Gi 1 /Gi ⇠
= Aut(F (⇣, u1 , . . . , ui )/F (⇣, u1 ,
. . . , ui 1 )). By Proposition 3.75 (ii) and Theorem 3.77, Aut(F (⇣, u1 , . . . , ui )/F (⇣, u1 ,
. . . , ui 1 )) is cyclic. So G0 = Aut(L(⇣)/F (⇣)) is solvable.

(() 1 It suffices to show that K 0 /F is solvable by radicals. Let S ⇢ K 0 be

the largest separable extension over F . Then S is Galois over F and Aut(S/F ) ⇠ =
Aut(K 0 /F ) (Theorem 3.57). The extension K 0 /S is purely inseparable, hence rad-
ical. Thus it suffices to show that S/F is solvable by radicals.

2 Let m be the product of all prime factors of [S : F ] different from char F .

Let ⇣ be a primitive mth root of unity. We claim that [S(⇣) : F (⇣)] | [S : F ]. (By
Corollaries 3.70 and 3.71, we have [S(⇣) : S] | [F (⇣) : F ], so the claim follows.)
We show that S(⇣)/F is radical. It suffices to show that S(⇣)/F (⇣) is radical.
Since both Aut(S(⇣)/S) and Aut(S(⇣)/F )/Aut(S(⇣)/S) ⇠ = Aut(S/F ) are solvable,
Aut(S(⇣)/F ) is solvable. So Aut(S(⇣)/F (⇣)) is solvable. Let 1 = H0 C H1 C
· · · C Hn = Aut(S(⇣)/F (⇣)) be such that Hi /Hi 1 is cyclic of prime order. Then
Hi0 1 /Hi0 is a cyclic extension of prime degree. By Proposition 3.75 (i) and Theo-
rem 3.77, Hi0 1 /Hi0 is radical. Therefore H00 = S(⇣) is radical over Hn0 = F (⇣). ⇤
 3.10. TRANSCENDENTAL EXTENSIONS 131

S(⇣) .................................................. 1
... ...
... ...
.. ..
.... ....
S(⇣) . .
...
.....
.....
..... Hi0 1
.................................................. Hi 1
..... .....
..
......
. .....
.....
.
....
.
....
.... ..... .. ..
.... .

S F (⇣) Hi0 .................................................. Hi

..... .. ... ...
..... .....
..... ..... ... ...
..... ..... .. ..
.....
..... ..
......
. ... ...
. ..... . .

F F (⇣) .................................................. Hn

Proposition 3.81. Let F ⇢ K ⇢ N be such that N/F is normal and Aut(N/F )

is solvable. Then Aut(K/F ) is also solvable.
Proof. Let H = { 2 Aut(N/F ) : (K) = K} < Aut(N/F ). Since N/F is
normal, the map H ! Aut(K/F ), 7! |K , is an onto homomorphism. Since H
is solvable, so is Aut(K/F ). ⇤
Corollary 3.82. If K/F is solvable by radicals, then Aut(K/F ) is solvable.
Proof. Combine Theorem 3.80 and Proposition 3.81. ⇤
Example. Let f = x5 4x 2 2 Q[x] and let K be a splitting field of f over
Q. Then Aut(K/Q) ⇠ = S5 (Example 3.39), which is not solvable. So the equation
f (x) = 0 is not solvable by radicals over Q.
Note. Let K/F be algebraic and K 0 be the normal closure of K over F . If
Aut(K/F ) is solvable, Aut(K 0 /F ) is not necessarily solvable. Example: Let u be a
root of f (x) = x5 4x 2 2 Q[x] and let K = Q(u). Then Aut(K/Q) = {id} but
Aut(K 0 /Q) = S5 . (Proof that Aut(K/Q) = {id}: If K = Q(u) contains more than
one root of f , then [K 0 : Q]  3! [Q(u) : Q] = 3! · 5 < 5!, which is a contradiction.)

3.10. Transcendental Extensions

Algebraic dependence and independence. Let K/F be an extension. A
subset S ⇢ K is called algebraically dependent over F if there exist distinct elements
s1 , . . . , sn 2 S and 0 6= f 2 F [x1 , . . . , xn ] such that f (s1 , . . . , sn ) = 0; S is called
algebraically independent over F if it is not algebraically dependent over F .
Fact. Let S ⇢ K be algebraically independent over F . In the field F (S),
elements of S can be treated as indeterminates.
Transcendence basis. Let K/F be an extension. A transcendence basis of
K over F is a maximal subset of K that is algebraically independent over F . By
Zorn’s lemma, transcendence bases exist.
Proposition 3.83. Let K/F be an extension and S ⇢ K be algebraically
independent over F . Let u 2 K \ S. Then u is transcendental over F (S) if and
only if S [ {u} is algebraically independent over F .
Proof. ()) Assume to the contrary that there exist distinct elements s1 , . . . , sn 2
Pm 0 6= f 2 F [xi 1 , . . . , xn , xn+1 ] such that f (s1 , . . . , sP
S and n , u) = 0. Write f =
m
i=0 f i (x 1 , . . . , x n )x n+1 , where f i 2 F [x 1 , . . . , x n ]. Then i
i=0 fi (s1 , . . . , sn )u =
0. Since u is transcendental over F (S), we have fi (s1 , . . . , sn ) = 0, 1  i  m.
132 3. FIELDS

Since s1 , . . . , sn are algebraically independent over F , fi = 0, 1  i  m. So f = 0,

which is a contradiction.
(() Assume that there exists f 2 F (S)[x] such that f (u) = 0. Write
m
X fi (s1 , . . . , sn )
f (x) = xi ,
i=0
gi (s1 , . . . , sn )
where s1 , . . . , sn 2 S are distinct, fi , gi 2 F [x1 , . . . , xn ] and gi (s1 , . . . , sn ) 6= 0. Let
⇣Ym ⌘Xm
fi (x1 , . . . , xn ) i
h= gi (x1 , . . . , xn ) x 2 F [x1 , . . . , xn , x].
i=0
g (x , . . . , xn )
i=0 i 1

Then h(s1 , . . . , sn , u) = 0, so h = 0. Hence

⇣Y
m ⌘
0 = h(s1 , . . . , sn , x) = gi (s1 , . . . , sn ) f (x) = 0,
i=0

so f = 0. Therefore u is transcendental over F (S). ⇤

Corollary 3.84. Let K/F be an extension. A subset S ⇢ K is a transcendence
basis of K over F if and only if
(i) S is algebraically independent over F and
(ii) K is algebraic over F (S).
Fact. Let K/F be an extension such that F = F (S). Let T be a maximal
subset of S which is algebraically independent over F . Then T is a transcendence
basis of K over F .
Proof. By Proposition 3.83, all elements of S are algebraic over F (T ), so K
is algebraic over F (T ). ⇤
Theorem 3.85. Let K/F be an extension. All transcendence bases of K over
F have the same cardinality.
Proof. Let S and T be two transcendence bases of K/F .
Case 1. Assume that |S| < 1, say, S = {s1 , . . . , sn }.
1 We claim that there exists t1 2 T such that {t1 , s2 , . . . , sn } is a transcendence
basis of K/F .
First, we claim that there exists t1 2 T such that t1 is transcendental over
F (s2 , . . . , sn ). Otherwise, F (s2 , . . . , sn )(T )/F (s2 , . . . , sn ) is algebraic. Since K/F (T )
is algebraic, K/F (s2 , . . . , sn )(T ) is algebraic. So K/F (s2 , . . . , sn ) is algebraic. But
s1 2 K is not algebraic over F (s2 , . . . , sn ), which is a contradiction. By Proposi-
tion 3.83, {t1 , s2 , . . . , sn } is algebraically independent over F .
Next, we claim that s1 is algebraic over F (t1 , s2 , . . . , sn ). Otherwise, t1 , s1 , s2 ,
. . . , sn would be algebraically independent over F , which is a contradiction. Now
F (S) is algebraic over F (t1 , s2 , . . . , sn ), and hence K is algebraic over F (t1 , s2 , . . . , sn ).
By Corollary 3.84, {t1 , s2 , . . . , sn } is a transcendence basis of K/F .
2 Using 1 repeatedly, we find that there exist t1 , . . . , tn 2 T such that
{t1 , . . . , tn } is a transcendence basis of K/F . Thus T = {t1 , . . . , tn }. So |T | =
n = |S|.
Case 2. Assume that |S| = 1 and |T | = 1.
 3.10. TRANSCENDENTAL EXTENSIONS 133

For each s 2 S, s is algebraic over F (T ). Let f (x) = xm + am 1 xm 1 +

·S· · + a0 2 F (T )[x] be the minimal polynomial of s over F (T ). Since F (T ) =
T 0 ⇢T,|T 0 |<1 F (T ), there exists Ts ⇢ T with |Ts | < 1 such that a0 , . . . , am 1 2
0

F (Ts ). So s is algebraic
S over F (Ts ). S
We claim that s2S Ts is a transcendence basis of K/F . First, s2S Ts is alge-
braically independent over F since it is contained S in T . SinceSK/F (S) is algebraic
and everyS element of S is algebraic over F ( ( s2S Ts ) is algebraic.
s2S s ), K/F S
T
Hence s2S Ts is a transcendence basis of K/F . Thus T = s2S Ts . Now we have
X
|T |  |Ts |  |S|@0 = |S|.
s2S

By symmetry, |S|  |T |. ⇤
Transcendence degree. The transcendence degree of K/F , denoted by tr.d. K/F ,
is the cardinality of any transcendence basis of K/F .
Theorem 3.86. Let F ⇢ K ⇢ L be fields. Then
tr.d. L/F = tr.d. L/K + tr.d. K/F.
Proof. Let S be a transcendence basis of K/F and T be a transcendence basis
of L/K. Then clearly S \ T = ;. We claim that S [ T is a transcendence basis
of L/F . Since L is algebraic over K(T ) and K(T ) is algebraic over F (S [ T ), L is
algebraic over F (S [ T ). It remains to show that S [ T is algebraically independent
over F . Assume to the contrary that there are s1 , . . . , sm 2 S (distinct) and
t1 , . . . , tn 2 T (distinct) such that s1 , . . . , sm , t1 , . . . , tn are algebraically dependent
over F . Then there exists
X
0 6= f (x1 , . . . , xm , y1 , . . . , yn ) = fi1 ,...,in (x1 , . . . , xm )y1i1 · · · ynin
i1 ,...,in
2 F [x1 , . . . , xm , y1 , . . . , yn ],
where fi1 ,...,in (x1 , . . . , xm ) 2 F [x1 , . . . , xm ], such that f (s1 , . . . , sm , t1 , . . . , tn ) = 0.
Since t1 , . . . , tn are algebraically independent over K, we have fi1 ,...,in (s1 , . . . , sm ) =
0 for all i1 , . . . , in . Since s1 , . . . , sm are algebraically independent over F , we have
fi1 ,...,in (x1 , . . . , xm ) = 0. Then f = 0, which is a contradiction. ⇤
Example 3.87. tr.d. C/Q = @ and |Aut(C/Q)| = |SC | = 2@ , where SC is the
symmetric group on C.
Proof. Let T be a transcendence basis of C/Q. Clearly, |T | = 1. Since
C/Q(T ) is algebraic, |C|  |Q(T )|@0 = |Q(T )|. Let P0 (T ) be the set of all finite
subsets of T . Then
[ X
|Q(T )| = Q(T 0 )  |Q(T 0 )|  |P0 (T )|@0 = |T |@0 = |T |.
T 0 2P0 (T ) T 0 2P0 (T )

So |C|  |T |. Of course, |T |  |C|. Hence |T | = |C| = @.

Every ⇢ 2 ST induces an automorphism ⇢¯ of Q(T ). Since C is the algebraic
closure of Q(T ), ⇢¯ extends to an automorphism ⇢˜ of C. The mapping ST !
Aut(C/Q), ⇢ 7! ⇢˜ is one-to-one. So |Aut(C/Q)| |ST | = |SC |. Since Aut(C/Q) <
SC , we have |Aut(C/Q)|  |SC |.
It is easy to show that |SC | 2@ . On the other hand, |SC |  |CC | = @@ 
@ @
(2 ) = 2 @·@
= 2 . Hence |SC | = 2 .
@ @
⇤
134 3. FIELDS

Theorem 3.88 (Lüroth). Let F ( E ⇢ F (x) be fields, where x is an indeter-

minate. Then there exists y 2 E \ F such that E = F (y).
Proof. By Exercise 3.40 (i), x is algebraic over E. Let f (z) = z n +an 1 z n 1 +
· · · + a0 2 E[z] be the minimal polynomial of x over E. Write ai = Ai (x)/Bi (x),
where Ai , Bi 2 F [x], gcd(Ai , Bi ) = 1, and let B = lcm(B0 , . . . , Bn 1 ). Then
An 1 (x) n 1 A0 (x)
g(x, z) := B(x)f (z) = B(x)z n + B(x) z + · · · + B(x) 2 F [x][z]
Bn 1 (x) B0 (x)
is a primitive polynomial in z over F [x]. There exists 0  i  n 1 such that
/ F since otherwise, x would be algebraic over F . Let
ai 2
Ai (x)
h(z) = Ai (z) Bi (z) 2 E[z].
Bi (x)
Since h(x) = 0, we have f (z) | h(z) in E[z] and hence in F (x)[z]. Then g(x, z) |
Bi (x)Ai (z) Ai (x)Bi (z) in F [x][z] = F [x, z], whence
(3.26) Bi (x)Ai (z) Ai (x)Bi (z) = c(x, z)g(x, z)
for some c(x, z) 2 F [x, z]. We claim that Bi (x)Ai (z) Ai (x)Bi (z), as a polynomial
in x over F [z], is primitive and has degree max{deg Ai , deg Bi }. The claim follows
from the fact that there are u, v 2 F (the algebraic closure of F ) such that Ai (u) 6=
0, Bi (u) = 0, and Ai (v) = 0, Bi (v) 6= 0. Since
degx g(x, z) max{deg B(x), deg(B(x)Ai (x)/Bi (x))}
max{deg Bi (x), deg Ai (x)}
= degx (Bi (x)Ai (z) Ai (x)Bi (z)),
it follows from (3.26) that c(x, z) = c(z). The left side of (3.26) is a primitive
polynomial in x over F [z], whence c(z) = c 2 F ⇥ . Therefore g(x, z) = g(z, x).
Now we have
[F (x) : F (ai )] = [F (x) : F (Ai (x)/Bi (x))]
= max{deg Ai , deg Bi } (Exercise 3.40 (i))
= degx g(x, z) = degz g(x, z)
= [F (x) : E],
whence E = F (ai ). ⇤

3.11. Transcendence of e and ⇡, Lang’s Theorem

The transcendence of e and ⇡ (over Q) were first proved by Hermite and Lin-
demann, respectively. The proof included below is adopted from Baker [1].
Theorem 3.89. e and ⇡ are transcendental numbers.
Proof. For f (x) 2 R[x] and t 2 C, define
Z t
(3.27) I(f, t) = et e u f (u)du.
0
Integration by parts gives
I(f, t) = I(f 0 , t) + et f (0) f (t).
 3.11. TRANSCENDENCE OF e AND ⇡, LANG’S THEOREM 135

Let m = deg f . The above recursive formula gives

m
X m
X
(3.28) I(f, t) = et f (j) (0) f (j) (t).
j=0 j=0

1 e is transcendental. Assume to the contrary that there exist a0 , . . . , an 2 Z,

a0 > 0, such that
(3.29) a0 + a1 e + · · · + an en = 0.
Let p be a sufficiently large prime and define
(3.30) f (x) = xp 1
(x 1)p · · · (x n)p ,
where m = deg f = (n + 1)p 1. Let
J = a0 I(f, 0) + · · · + an I(f, n).
By (3.28) and (3.29),
m X
X n
J= ak f (j) (k) 2 Z.
j=0 k=0

For 0  j  m and 0  k  n, p! | f (j) (k) unless (j, k) = (p 1, 0), and when

(j, k) = (p 1, 0), ak f (j) (k) = a0 (p 1)!( 1)np (n!)p , which is divisible by (p 1)! but
not by p! (under the assumption that p is sufficiently large). Therefore (p 1)! | J
but p! - J, whence
(3.31) |J| (p 1)!.
On the other hand, it is clear from (3.27) and (3.30) that
(3.32) |I(f, k)|  cp , 0  k  n,
where c > 1 does not depend on p. Hence
|J|  |a0 ||I(f, 0)| + · · · + |an ||I(f, n)|  Acp ,
where A > 0 is independent of p; this contradicts (3.31).
2 ⇡ is transcendental. Assume to the contrary that ✓1 := ⇡i is algebraic and
let ✓1 , ✓2 , . . . , ✓d be the conjugates of ✓1 over Q. Then
(3.33) 0 = (1 + e✓1 ) · · · (1 + e✓d ) (since e✓1 = 1)
X
= e↵(✏) ,
✏=(✏1 ,...,✏d )2{0,1}d

where ↵(✏) = ✏1 ✓1 + · · · + ✏d ✓d . Let E = {✏ 2 {0, 1}d : ↵(✏) 6= 0}. For a sufficiently

large prime p, define
Y
(3.34) f (x) = xp 1 (x ↵(✏))p ,
✏2E

where m = deg f = (|E| + 1)p 1. Let

X
J= I(f, ↵(✏)).
✏2E

Then we have
⇣X ⌘⇣X
m ⌘ m X
X
(3.35) J = e↵(✏) f (j) (0) f (j) (↵(✏)) (by (3.28))
✏2E j=0 j=0 ✏2E
136 3. FIELDS

m
X m X
X
= (|E| 2d ) f (j) (0) f (j) (↵(✏)) (by (3.33)).
j 0 j=0 ✏2E

Note that
(
0 if j 6= p 1,
(3.36) f (j)
(0) =
(p 1)!( 1) |E|p
P p
if j = p 1,
Q
where P = ✏2E ↵(✏), and
(
X 0 if j 6= p,
(3.37) f (j)
(↵(✏)) =
✏2E
p!Q if j = p,
P Q
where Q = ↵2E ↵(✏) p 1
2E\{✏} (↵(✏) ↵( ))p . Both P and Q are symmetric
polynomials with integer coefficients in ↵(✏), ✏ 2 E, and hence are symmetric
polynomials with integer coefficients in ↵(✏), ✏ 2 {0, 1}d . Therefore, P and Q are
symmetric polynomials with integer coefficients in ✓1 , . . . , ✓d , and hence P, Q 2 Q.
Choose l 2 Z+ (independent of p) such that l✓1 is integral over Z, i.e., the minimal
polynomial of l✓1 over Q is a monic polynomial in Z[x]. Then l|E|p P p and l|E|p Q are
integral over Z. Thus l|E|p P p , l|E|p Q 2 Z. When p is sufficiently large, it follows
from (3.36) that the integer l|E|p f (p 1)P(0) is divisible by (p 1)! but not by p!,
and from (3.37) that the integer l|E|p ✏2E f (j) (↵(✏)) is divisible by p! for all j.
Then by (3.35), the integer l|E|p J is divisible by (p 1)! but not by p!, whence
l|E|p |J| (p 1)!, i.e.,
|E|p
(3.38) |J| l (p 1)!.
On the other hand, it is clear from (3.27) and (3.34) that
X
|J|  |I(f, ↵(✏))|  Acp ,
✏2E

where A > 0 and c > 1 are independent of p; this contradicts (3.38). ⇤

Entire functions of order  ⇢ on C. An analytic function on C is called an

entire function. A meromorphic function on C is a quotient of two entire functions
[4, Corollary 5.20]. An entire function f (z) is said to have order  ⇢ (⇢ > 0) if
there exists C > 0 such that
⇢
|f (z)|  eC|z| for all z 2 C.
|z|⇢
Equivalently, |f (z)| = as |z| ! 1 for some C1 > 1. A meromorphic
O(C1 )
function of order  ⇢ on C is of the form f (z)/g(z) where f (z) and g(z) (g 6= 0)
are entire functions of order  ⇢. The following theorem, due to Lang [21], implies,
among other things, the transcendence of e and ⇡.
Theorem 3.90 (Lang). Let K be a number field (a finite extension of Q). Let
f1 , . . . , fN be meromorphic functions on C of order  ⇢ such that
(i) tr.d. K(f1 , . . . , fN )/K 2;
(ii) Df↵ 2 K[f1 , . . . , fN ], 1  ↵  N , where D = d/dz.
Assume that w1 , . . . , wm 2 C are distinct such that f↵ (wj ) 2 K for all 1  ↵  N
and 1  j  m. Then m  10⇢[K : Q].
 3.11. TRANSCENDENCE OF e AND ⇡, LANG’S THEOREM 137

Proof. 1 Notation and assumptions. Let oK be the ring of integers of K.

We may assume that f↵ (wj ) 2 oK . (Otherwise, multiply f↵ by a suitable integer in
oK .) Let t 2 Z+ , n = 2mt2 , r = 2mt. The notation O( ) refers to the asymptotic
bound as t ! +1. Constants, denoted by C1 , C2 , . . . , are positive real numbers
> 1 depending only on the data in the statement of the theorem. For each x 2 K,
let
||x|| = max{| (x)| : 2 Emb(K/Q)},
where Emb(K/Q) is the set of embeddings of K in C. Assume that f1 and f2 are
algebraically independent over K.
2 We claim that there exists a constant C1 > 1 such that
(3.39) ||Dk (f1u f2v )(wj )||  k!rk C1k+r for all k 0, 1  u, v  r, 1  j  m.
Let h1 = · · · = hu = f1 and hu+1 = · · · = hu+v = f2 . Then
(3.40) ✓ ◆
X k
Dk (f1u f2v )(wj ) = Dk1 h1 (wj ) · · · Dku+v hu+v (wj )
k1 , · · · , ku+v
k1 +···+ku+v =k

and hence
(3.41) ||Dk (f1u f2v )(wj )||
 (u + v)k max ||Dk1 h1 (wj )|| · · · ||Dku+v hu+v (wj )|| : k1 + · · · + ku+v = k .
Let 2 3 2 3
f1 P1 (f1 , . . . , fN )
6 . 7 6 .. 7
D6 . 7 6 7, P↵ 2 K[X1 , . . . , XN ].
4 . 5=4 . 5
fN PN (f1 , . . . , fN )
By Theorem 3.92, for each l > 0, (Dl f↵ )(wj ) is a sum of (l 1)!N l 1 terms of the
form
(3.42)
h @ i 1 P↵ 1 @ i l 1 P↵ l 1 i
··· P↵l (f1 (wj ), . . . , fN (wj )),
@X (1,1) · · · @X (1,i1 ) @X (l 1,1) · · · @X (l 1,il 1 )
where i1 + · · · + il = l 1 and 1  ↵⇤ , (⇤, ⇤)  N . Put
1
n @ i P↵
C = max (f1 (wj ), . . . , fN (wj )) :
@X 1 · · · @X i
o
i 0, 1  ↵, 1 , . . . , i  N, 1  j  m .
Then
||Dl f↵ (wj )||  (l 1)!N l 1
C l, l > 0, 1  ↵  N, 1  j  m.
Including the case l = 0, we have
(3.43) ||Dl f↵ (wj )||  l!N l (C + C 0 )l+1 , l 0, 1  ↵  N, 1  j  m,
where
C 0 = max{||f↵ (wj )|| : 1  ↵  N, 1  j  m}.
By (3.41) and (3.43),
||Dk (f1u f2v )(wj )||
 (2r)k max{k1 ! · · · ku+v !N k (C + C 0 )k+u+v : k1 + · · · + ku+v = k}
138 3. FIELDS

 k!rk C1k+r .

3 Choose 0 6= 2 oK such that

@ i P↵
(f1 (wj ), . . . , fN (wj )) 2 oK
@X 1 · · · @X i
for all i 0, 1  ↵, 1, . . . , i  N , 1  j  m. Then we have
(3.44) l l
D f↵ (wj ) 2 oK , l 0, 1  ↵  N, 1  j  m.
For l > 0, (3.44) follows from (3.42); for l = 0, recall that we assumed that f↵ (wj ) 2
oK . It follows from (3.40) and (3.44) that
(3.45) k
Dk (f1u f2v )(wj ) 2 oK , k 0, 1  u, v  r, 1  j  m.
By 2 ,
(3.46) || k
Dk (f1u f2v )(wj )||  || ||k k!rk C1k+r  k!rk C2k+r ,
k 0, 1  u, v  r, 1  j  m.

4 We claim that there exist buv 2 oK (1  u, v  r) not all 0 such that

r
X
(3.47) buv Dk (f1u f2v )(wj ) = 0, 0  k < n, 1  j  m,
u,v=1

and
(3.48) max{||buv || : 1  u, v  r} = O(n2n ).
Write uv,kj = k
Dk (f1u f2v )(wj ) 2 oK . Then (3.47) is equivalent to
r
X
(3.49) uv,kj buv = 0, 0  k < n, 1  j  m.
u,v=1

We know that oK is a free Z-module of rank M := [K : Q]. Let ✏1 , . . . , ✏M be a

basis of oK over Z. Write
M
X
l l
✏ l1 ✏ l2 = l1 l 2 ✏ l , l 1 l2 2 Z,
l=1
M
X
uv,kj = ⇠uv,kj,l ✏l , ⇠uv,kj,l 2 Z,
l=1

M
X
(3.50) buv = cuv,l ✏l , cuv,l 2 Z.
l=1

Then (3.49) is a system of nmM linear equations in r2 M unknowns cuv,l , i.e.,

(3.51) A[cuv,l ] = 0,
where [cuv,l ] is an r M ⇥ 1 column and A is an nmM ⇥ r2 M matrix whose rows are
2

labeled by (kj, l1 ), 0  k < n, 1  j  m, 1  l1  M , whose columns are labeled

by (uv, l2 ), 1  u, v  r, 1  l2  M , and whose ((kj, l1 ), (uv, l2 ))-entry is
M
X
l1
⇠uv,kj,l l l2 .
l=1
 3.11. TRANSCENDENCE OF e AND ⇡, LANG’S THEOREM 139

Let ✏01 , . . . , ✏0M be the dual basis of ✏1 , . . . , ✏M of K/Q with respect TrK/Q . Then
⇠uv,kj,l = TrK/Q ( uv,kj ✏0l ). So, by (3.46),
|⇠uv,kj,l |  C3 || uv,kj ||  k!rk C4k+r  n!rn C4n+r ,
0  k < n, 1  u, v  r, 1  j  m.
Thus all the entries of A have | |  n!rn C5n+r . Let L 2 Z+ to be chosen. Note that
2 2
A : Zr M ! ZnmM maps [ L, L]r M to [ n!rn C5n+r r2 M L, n!rn C5n+r r2 M L]nmM ⇢
[ Ln!rn+2 C6n+r , Ln!rn+2 C6n+r ]nmM . Therefore, if
2
(3.52) (2L + 1)r M
> (2Ln!rn+2 C6n+r + 1)nmM ,
2
(3.51) has a nonzero integer solution [cuv,l ] 2 [ L, L]r M . Inequality (3.52) holds
when
2
(2L)r M > (3Ln!rn+2 C6n+r )nmM ,
i.e.,
2 2
(3.53) 2r M
L(r nm)M
> (3n!rn+2 C6n+r )nmM .
Since r2 nm = nm, (3.53) holds when we choose
1
L = 3n!rn+2 C6n+r = 3n!(2nm) 2 (n+2) C6n+r = O(n2n ).
Then by (3.50),
max{||buv || : 1  u, v  r} = O(n2n ).

5 Define a meromorphic function

r
X
F = buv f1u f2v .
u,v=1

Note that F depends on r and hence on t. By 4 ,

Dk F (wj ) = 0 for all 0  k < n, 1  j  m.
However, F = 6 0 since f1 , f2 are algebraically independent over K. Let s n be
the smallest integer such that
Dk F (wj ) = 0 for all 0  k < s, 1  j  m.
Assume, without loss of generality, that
:= Ds F (w1 ) 6= 0.
By (3.45), s
2 oK ; hence
(3.54) 1  |NK/Q ( s
)|  || ||s[K:Q] |NK/Q ( )|.
By (3.39) and (3.48),
r
X
(3.55) || || = buv Ds (f1u f2v )(wj ) = O(r2 n2n s!rs C1s+r ) = O(s5s ).
u,v=1

By (3.54) and (3.55),

(3.56) 1  || ||s[K:Q] O(s5s )[K:Q] 1
| |.
140 3. FIELDS

6 There exist entire functions p(z) and q(z) of order  ⇢ with p(w1 ) 6= 0 and
q(w1 ) 6= 0 such that pf1 and qf2 are entire functions. Let ✓ = pq. Then ✓ is an
entire function of order  ⇢ and ✓f1 , ✓f2 are both entire. Clearly,
✓(z)2r F (z)
(3.57) H(z) := Qm
j=1 (z wj )s
is an entire function. Let R > 0 be large. When |z| = R,
r
X
2r
⇥ ⇤u ⇥ ⇤v ⇥ ⇤2r (u+v)
|✓(z) F (z)| = buv (✓f1 )(z) (✓f2 )(z) ✓(z)
u,v=1
⇢ ⇢
 r2 O(n2n )C72rR  O(s2s C8rR ).
By the maximum modulus principle,
⇣ s2s C rR⇢ ⌘
8
max{|H(z)| : |z|  R}  O .
Rms
Let R = s1/2⇢ and z = w1 . We have
p
⇣ s2s C rs1/2 ⌘ ⇣ s2s C 2mns ⌘ ⇣ s2s C s ⌘
8 8 9
|H(w1 )|  O =O O .
sms/2⇢ sms/2⇢ sms/2⇢
Since F (z)/(z w1 )s |z=w1 = Ds F (w1 )/s! = /s!, setting z = w1 in (3.57) gives
⇣ s3s C s ⌘
11
(3.58) s
| | = O(s!C10 )|H(w1 )|  O ms/2⇢ .
s
Now we combine (3.56) and (3.58) to have
⇣ 5s([K:Q] 1)+3s ⌘
s s s s(5[K:Q] m/2⇢)
1  O C12  O(C12 s ).
sms/2⇢
So, 5[K : Q] m/2⇢ 0, i.e., m  10⇢[K : Q]. ⇤

Corollary 3.91 (Hermite-Lindemann). If ↵ 2 C \ {0} is algebraic, then e↵ is

transcendental.
Proof. Assume to the contrary that e↵ is algebraic. Let K = Q(↵, e↵ ) and
let f1 (z) = z and f2 (z) = ez , which are entire functions of order  1 satisfying
Theorem 3.90 (i) and (ii). Choosing wj = j↵, j 2 Z, produces a contradiction. ⇤

The transcendence of e and ⇡ follows from Corollary 3.91 immediately. Since

e2⇡i = 1 is algebraic, 2⇡i is transcendental and so is ⇡.
Derivatives of the solution of the Cauchy problem. For k 2 Z+ , let
Ik = (i1 , . . . , ik ) 2 Nk : i1 + · · · + it t, for 1  t  k, i1 + · · · + ik = k .
It is convenient to define I0 to be the set of an empty tuple ;. For (j1 , . . . , jk 1 ) 2
Nk 1 and (i1 , . . . , ik ) 2 Nk , define (j1 , . . . , jk 1 ) (i1 , . . . , ik ) if (i1 , . . . , ik ) =
(j1 , . . . , jl 1 , jl + 1, 0, jl+1 , . . . , jk 1 ) for some 1  l  k 1 or (i1 , . . . , ik ) =
(j1 , . . . , jk 1 , 1). If (j1 , . . . , jk 1 ) (i1 , . . . , ik ), then (j1 , . . . , jk 1 ) 2 Ik 1 if and
only if (i1 , . . . , ik ) 2 Ik . Recall that In denotes the n ⇥ n identity matrix.
 3.11. TRANSCENDENCE OF e AND ⇡, LANG’S THEOREM 141

Theorem 3.92. Let D = d/dz, where z is either a real or a complex variable.

Consider the Cauchy problem
2 3 2 3 2 3 2 3
y1 P1 (y1 , . . . , yn ) y1 (z0 ) b1
6.7 6 . 7 6 . 7 6.7
(3.59) D6 .7 6 .. 7, 6 . 7 = 6 . 7,
4 . 5=4 5 4 . 5 4.5
yn Pn (y1 , . . . , yn ) yn (z0 ) bn
where P1 , . . . , Pn are functions with continuous partial derivatives of total order up
to k in a neighborhood of (b1 , . . . , bn ). Then the (k + 1)-th derivative of a solution
of (3.59) in a neighborhood of z0 is given by
(3.60)
2 3
y1
6.7
Dk+1 6 .7
4 . 5=
yn
X @ i1 P ⇣ @ i2 P ⌘ ⇣ @ ik P ⌘
a(i1 , . . . , ik ) i1 Ini1 1 ⌦ · · · Ini1 +···+ik 1 (k 1) ⌦ P,
@y @y i2 @y ik
(i1 ,...,ik )2Ik

where
(i) 2 3
P1 (y1 , . . . , yn )
6 .. 7
P =6
4 .
7;
5
Pn (y1 , . . . , yn )
(ii) @ i P /@y i is an n ⇥ ni matrix whose rows are indexed by ↵ 2 {1, . . . , n},
whose columns are indexed by ( 1 , . . . i ) 2 {1, . . . , n}i lexicographically,
and whose (↵, ( 1 , . . . , i ))-entry is
@ i P↵
;
@y 1 · · · @y i
(iii) a(i1 , . . . , ik ) 2 Z+ , (i1 , . . . , ik ) 2 Ik , are defined inductively by
( P
a(i1 , . . . , ik ) = (j1 ,...,jk 1 ) (i1 ,...,ik ) a(j1 , . . . , jk 1 ),
a(;) = 1.
Moreover, X
a(i1 , . . . , ik ) = k!.
(i1 ,...,ik )2Ik

Proof. For (i1 , . . . , ik ) 2 Ik , let ik+1 = 0 and let

Y⇣
k+1
@ il P ⌘
(3.61) F(i1 ,...,ik ,0) = Ini1 +···+il 1 (l 1) ⌦ ,
@y il
l=1
where the factors in the product appear from left to right in the order of l =
1, 2, . . . , k + 1. Then (3.60) can be written as
2 3
y1
6 .. 7 X
(3.62) Dk+1 64 5 .
7= a(i1 , . . . , ik )F(i1 ,...,ik ,0) .
(i1 ,...,ik )2Ik
yn
142 3. FIELDS

To prove (3.62), we use induction on k. The initial case k = 0 needs no proof.

Since
⇣ @ i P↵ ⌘ X @ i+1 P↵ X @ i+1 P↵
D = Dy i+1 = P ,
@y 1 · · · @y i @y 1 · · · @y i @y i+1 @y 1 · · · @y i @y i+1 i+1
i+1 i+1

we have ⇣ @iP ⌘ @ i+1 P

D = (I i ⌦ P ).
@y i @y i+1 n
Thus
h @ il P i
(3.63) D Ini1 +···+il 1 (l 1) ⌦
@y il
h @ il +1 P i
= Ini1 +···+il 1 (l 1) ⌦ i +1
(Inil ⌦ P )
@y l
h @ il +1 P i⇥ ⇤
= Ini1 +···+il 1 (l 1) ⌦ Ini1 +···+il 1 +(il +1) l ⌦P .
@y il +1
By (3.61) and (3.63), we have
DF(i1 ,...,ik ,0)
Xh lY1 ⇣
k+1
@ is P ⌘ih ⇣ @ il P ⌘i
= Ini1 +···+is 1 (s 1) ⌦ D Ini1 +···+il 1 (l 1) ⌦
@y is @y il
l=1 s=1
h k+1
Y ⇣ @ is P ⌘i
Ini1 +···+is 1 (s 1) ⌦
@y is
s=l+1
k+1
X
= F(i1 ,...,il 1 ,il +1,0,il+1 ,...,ik ,0)
.
l=1
Therefore, assuming (3.62), we have
2 3
y1 k+1
6 7 X X
k+2 6 .. 7
D 4 5. = a(i 1 , . . . , i k ) F(i1 ,...,il 1 ,il +1,0,il+1 ,...,ik ,0)
(i1 ,...,ik )2Ik l=1
yn
X
= a(j1 , . . . , jk+1 )F(j1 ,...,jk+1 ,0) ,
(j1 ,...,jk+1 )2Ik+1

where X
a(j1 , . . . , jk+1 ) = a(i1 , . . . , ik ).
(i1 ,...,ik ) (j1 ,...,jk+1 )
So the induction is complete.
The integer a(i1 , . . . , ik ) is the number of chains ; = ↵0 ↵1 ··· ↵k =
(i1 , . . . , ik ), where ↵l 2 Il , 0  l  k. For each ↵l 2 Il , the number of ↵l+1 2 Il+1
such that ↵l ↵l+1 is l + 1. Hence
X
a(i1 , . . . , ik ) = k!.
(i1 ,...,ik )2Ik

⇤
Note. For a formula for a(i1 , . . . , ik ), see [14].
 CHAPTER 4

Noncommutative Rings

4.1. The Jacobson Radical

Definition 4.1. Let R be a ring. The Jacobson radical of R is
\
J(R) = I.
I is a max.
left ideal of R

It will be shown that J(R) is a two-sided ideal (Corollary 4.3). A ring R is called
J-semisimple if J(R) = 0.
Fact. J(R/J(R)) = 0, i.e., R/J(R) is always J-semisimple.
Proof. Let I be the set of all maximal left ideals of R. Then {I/J(R) : I 2 I}
is the set of all maximal left ideals of R/J(R). Therefore
\ ⇣\ ⌘
J R/J(R)) = I/J(R) = I /J(R) = 0.
I2I I2I
⇤
Example. Let R = Mn (D) be the ring of n ⇥ n matrices over a division ring
D. For each 1  j  n, let
Jj = {A 2 Mn (D) : the jth column of A is 0}.
Then Jj is a left ideal of R and R/Jj ⇠ = Dn . We claim that R/Jj is a simple
R-module. Let 0 6= ↵ 2 R/Jj . Then ↵ = [0 . . . a . . . 0] + Jj , where 0 6= a 2 Dn .
j
For each x 2 Dn , there exists A 2 R such that Aa = x. Then [0 . . . x . . . 0] + Jj =
j
A↵ 2 R↵. So R↵ T
= R/Jj , hence the claim. Therefore, Jj is a maximal left ideal of
n
R. Thus J(R) ⇢ j=1 Jj = 0.
Proposition 4.2. Let R be a ring and let x 2 R. Then the following statements
are equivalent.
(i) x 2 J(R).
(ii) For every r 2 R, 1 rx has a left inverse in R.
(iii) For each simple module R M , xM = 0.
Proof. (i) ) (ii) Suppose to the contrary that 1 rx does not have a left
inverse. Then R(1 rx) is a proper left ideal of R. So R(1 rx) is contained in
a maximal left ideal I of R. Then 1 = (1 rx) + rx 2 I + J(R) ⇢ I, which is a
contradiction.
(ii) ) (iii) Assume to the contrary that xM 6= 0. Choose m 2 M such that
xm 6= 0. Since M is simple, we have Rxm = M . Thus there exists r 2 R such that
rxm = m, i.e., (1 rx)m = 0. This is not possible since 1 rx is left invertible.
143
144 4. NONCOMMUTATIVE RINGS

(iii) ) (i) Let I be a maximal left ideal

T of R. Then R/I is a simple R-module,
so x(R/I) = 0. Thus x 2 I. Hence x 2 I is a max. left ideal of R I = J(R). ⇤
Corollary 4.3. We have
\
(4.1) J(R) = ann(M ),
RM is a simple
left R-module
where ann(M ) = {r 2 R : rM = 0}. In particular, J(R) is a two-sided ideal of R.
Proof. Proposition 4.2 (i) , (iii). ⇤
Proposition 4.4. Let R be a ring and x 2 R. Then x 2 J(R) if and only if
1 rx is a unit of R for all r 2 R.
Proof. ()) By Proposition 4.2 (ii), there exists u 2 R such that u(1 rx) = 1.
Then u = 1 + urx, which has a left inverse by Proposition 4.2 (ii). Thus u is a unit
of R and 1 rx = u 1 . ⇤
Remark 4.5.
(i) In Proposition 4.2 (ii), the word “left” can be dropped (Proposition 4.4).
(ii) In Proposition 4.4, 1 rx is a unit if and only if 1 xr is a unit (Exer-
cise ??). Hence in Proposition 4.4, 1 rx can be replaced with 1 xr.
Consequently, Definition 4.1, Proposition 4.2 and Corollary 4.3 are all
left-right symmetric.
Notation. Let R be a ring and A1 , . . . , An ⇢ R. Define
nX
m o
A1 · · · An = a1j · · · anj : m 0, aij 2 Ai ⇢ R.
j=1

If A1 is a left ideal of R, then so is A1 · · · An ; if An is a right ideal of R, then so is

A1 · · · An .
Nil and nilpotent ideals. An element a 2 R is called nilpotent if an = 0
for some n > 0. A left (or right) ideal I of R is called nil if all its elements are
nilpotent; I is called nilpotent if I n = 0 for some n > 0, i.e., a1 · · · an = 0 for all
a1 , . . . , an 2 I. If I is nilpotent, then I is nil.
Theorem 4.6 (Levitzki). Let R be a left noetherian ring and I be a left or
right ideal of R. Then I is nil if and only if I is nilpotent.
Proof. ()) 1 Let N be a maximal nilpotent ideal of R. (N exists since R
is left noetherian.) Let R1 = R/N . We claim that 0 is the only nilpotent left (or
right) ideal of R1 . Assume that L is a left (or right) ideal of R with L N such
that L/N is nilpotent. Since N is nilpotent, it follows that L is nilpotent. Then
RLR is a nilpotent ideal with RLR L N . By the maximality of N , RLR = N ,
so L = N , whence L/N = 0.
2 Assume to the contrary that I is not nilpotent. Let I1 = (I + N )/N . Then
I1 6= 0. Since I is nil and N is nilpotent, I + N is nil. So I1 is nil.
3 Choose 0 6= a 2 I1 . Clearly aR1 is a nil right ideal of R1 . For each u 2 aR1 ,
annL (u) := {v 2 R1 : vu = 0} is a left ideal of R1 . Let 0 6= u0 2 aR1 be such
that annL (u0 ) is maximal in {annL (u) : 0 6= u 2 aR1 }. (u0 exists since R1 is left
noetherian.) Then annL (u0 ) = annL (u0 x) for all x 2 R1 such that u0 x 6= 0. We
 4.1. THE JACOBSON RADICAL 145

claim that u0 xu0 = 0 for all x 2 R1 . If u0 x = 0, we are done. Assume that u0 x 6= 0.

Since u0 x 2 aR1 and aR1 is nil, u0 x is nilpotent. Say (u0 x)n = 0 but (u0 x)n 1 6= 0.
Then u0 x 2 annL ((u0 x)n 1 ) = annL (u0 ), i.e., u0 xu0 = 0. So the claim is proved.
Now (R1 u0 )2 = 0. By 1 , R1 u0 = 0, which is a contradiction since u0 6= 0. ⇤

Nil radical. The nil radical of a ring R, denoted by N (R), is the sum of all
nil ideals of R. If R is commutative, N (R) is the set all nilpotent elements of R.
Proposition 4.7.
(i) J(R) contains all nil left (right) ideals of R, in particular, N (R) ⇢ J(R).
(ii) Assume that R is left artinian. Then J(R) is nilpotent and J(R) = N (R).
Moreover, J(R) = N (R) is the largest nil left (right) ideal of R.
Proof. (i) Let I be a nil left ideal of R. For x 2 I and r 2 R, rx 2 I, so
(rx)n = 0 for some n > 0. Then 1 rx has a left inverse since 1 + rx + · · · +
(rx)n 1 (1 rx) = 1. Hence x 2 J(R). Thus I ⇢ J(R).
(ii) We first show that J(R) is a nilpotent ideal. Let J = J(R). Applying DCC
to J J 2 · · · , we have J m = J m+1 for some m > 0. Let I = J m . Then I 2 = I.
It suffices to show that I = 0. Assume to the contrary that I 6= 0. Let A be the
set of all left ideals A of R such that IA 6= 0. Then A 6= ; since I 2 A. Since R is
left artinian, A has a minimal element A0 . Choose a 2 A0 such that Ia 6= 0. Then
I(Ia) 6= 0, i.e., Ia 2 A. By the minimality of A0 , we have Ia = A0 . So there exists
r 2 I such that ra = a. Then (1 r)a = 0, whence 1 r is not left invertible. This
is a contradiction since r 2 I ⇢ J(R).
Since J(R) is nilpotent, J(R) ⇢ N (R). By (i), J(R) = N (R). Since J(R) is
nil and contains all nil left (right) ideals of R, it is the largest nil left (right) ideal
of R. ⇤

Example. Let R be a PID and a = pe11 · · · penn 2 R, where p1 , . . . , pn are

distinct primes in R and ei > 0, 1  i  n. Then
N R/(a) = J R/(a) = (p1 · · · pn )/(a).

Proof. Since (p1 · · · pn )/(a) is the set of all nilpotent elements of R/(a), we
have N (R/(a)) = (p1 · · · pn )/(a). Since R/(a) has DCC, J(R/(a)) = N (R/(a)). ⇤

Note. In the above example, if R is only a UFD, we still have N (R/(a)) =

(p1 · · · pn )/(a), but J(R/(a)) may be larger. Example: Let R = (F [[x]])[y], where
F is a field, and a = y, which is a prime in R. Then R/(a) ⇠
= F [[x]] and J(F [[x]]) =
xF [[x]] 6= 0.
Theorem 4.8 (Nakayama’s lemma). Let R M be a finitely generated R-module
such that J(R)M = M . Then M = 0.
Proof. Assume to the contrary that M 6= 0. Let m1 , . . . , mn be a minimal
set of generators of M . Since J(R)M = M , we have
m1 = r1 m1 + · · · + rn mn , ri 2 J(R).
The (1 r1 )m1 = r2 m2 + · · · + rn mn . Since r1 2 J(R), 1 r1 has a left inverse
u, and hence m1 = ur2 m2 + · · · + urn mn . Then M = hm2 , . . . , mn i, which is a
contradiction. ⇤
146 4. NONCOMMUTATIVE RINGS

4.2. Structure of Semisimple Rings

Definition 4.9. A module R M is called semisimple if it is a direct sum of
simple modules. A ring R is called left semisimple if R R is a semisimple module,
i.e., R R is a direct sum of certain minimal left ideals of R. Note: A ring R is called
right semisimple if RR is a semisimple module. However, it turns out that R is
right semisimple if and only if it is left semisimple (Corollary 4.20).

Proposition 4.10. Let M be a left R-module. The following statements are

equivalent.
(i) M is semisimple.
(ii) M is a sum of simple submodules.
(iii) Every submodule of M is a direct summand of M .

Proof. (i) ) (ii). Obvious.

P
(ii) ) (iii). Assume that M = i2I Mi , where each Mi is a simple submodule
of M . Let N be a submodule Pof M . By Zorn’s P lemma, there exists a maximal
subset
P J ⇢ I such that N + i2J M i = N i2J Mi . It suffices to show that
N M
i2J Pi = M . Assume the contrary.
P Then there exists k 2 I \ P
J such that
Mk 6⇢ N + P2J Mi . Then Mk \ (N i2J M i ) = {0}. So N + (M k + i2J Mi ) =
N (Mk i2J M i ), which contradicts the maximality of J.

(iii) ) (i). 1 We claim that every nonzero submodule A of M contains a

simple submodule. Let 0 6= a 2 A. We may assume that A = Ra (since it suffices
to show that Ra contains a simple submodule). Then A ⇠= R/L, where L = ann(a).
L is contained in a maximal left ideal K of R. Then K/L is a maximal submodule
of R/L. So A contains a maximal submodule B. Write M = B C. Then
A = B (C \ A). Since B is a maximal submodule of A, C \ A is a minimal
submodule of A.
2 We claim that M is semisimple. Let {Mi : i 2 I} be the set of all simple
submodules
P of M . By Zorn’s lemma, there exists
P a maximal subset J ⇢ I such
thatP i2J Mi is a direct sum. We claim that i2J Mi = M . Otherwise, M =
A Pi2J Mi for some nonzero submodule A of M . By 1 , A Mk for some k 2 I.
Then i2J[{k} Mi is a direct sum, which contradicts the maximality of J. ⇤

Proposition 4.11.
(i) Submodules and quotient modules of a semisimple module are semisimple.
(ii) If R is a left semisimple ring, then every left R-module is semisimple.

Proof. (i) Let R M be a semisimple module. Let A be a submodule of M . Let

B be a submodule of A. By Proposition 4.10, M = B C for some submodule C of
M . Then A = B (A \ C), so B is a direct summand of A. Hence A is semisimple.
Also, M = A D for some submodule D of M . Thus M/A ⇠ = D is semisimple.
(ii) Every left R-module is isomorphic to a quotient of a free R-module; the
free R module is semisimple since R R is semisimple. ⇤

Proposition 4.12. If a ring R is left semisimple, then RR has a composition

series.
 4.2. STRUCTURE OF SEMISIMPLE RINGS 147

L
Proof. We have R = i2I Li , where each Li is a minimal left ideal of R.
Write X
1= ei ,
i2I
where ei 2 Li and only finitely many ei 6= 0. For each j 2 I, choose 0 6= r 2 Lj .
We have X X
r=r ei = rei .
i2I i2I
L
Since i2I Li is a direct sum, we have r = rej . So ej 6= 0. Therefore |I| < 1, i.e.,
R = L1 · · · Ln , where each Li is a minimal left ideal of R. Then
{0} ⇢ L1 ⇢ L1 L2 ⇢ · · · ⇢ L1 ··· Ln = R
is a composition series of R R. ⇤
Note. If M is a semisimple R-modules, M may not have a composition series.
A vector space over a division ring D is a semisimple D-module. However, if
dimD V = 1, then D V does not have ACC or DCC.
Theorem 4.13. A ring R is left semisimple if and only if R is left artinian
and J(R) = 0.
Proof. ()) By Proposition 4.12, R is left artinian. By Proposition 4.10,
R = J(R) I, where I is a left ideal of R. Write 1 = e + f , where e 2 J(R) and
f 2 I. Then f = 1 e has a left inverse. So I = R, and hence J(R) = 0.
(() Since R has DCC, R has a minimal left ideal I1 . Since I1 6⇢ {0} = J(R),
there exists a maximal left ideal B1 such that I1 6⇢ B1 . Then R = I1 +B1 = I1 B1 .
If B1 6= 0, B1 contains a minimal left ideal I2 of R. By the same argument, there
exists a maximal left ideal M of R such that R = I2 M . Then it is easy to see
that B1 = I2 (B1 \ M ). Let B2 = B1 \ M . Then R = I1 I2 B2 . Continuing
this way, we have
R = I1 B1 = I1 I2 B2 = · · · ,
where Ii ’s are minimal left ideals of R and R ) B1 ) B2 ) · · · unless Bn = 0 for
some n. Since R has DCC, Bn = 0 for some n. So R = I1 · · · In . ⇤
Simple rings. A ring R is called simple if 1R 6= 0 and R does not have any
nontrivial ideal. If D is a division ring, then Mn (D) is a simple ring.
Example. A simple ring is not necessarily left artinian and hence is not neces-
sarily left semisimple. Let V be a vector space over a field F with a countable basis
x1 , x2 , . . . . Let R = EndF (V ) and I = {f 2 R : dimF f (V ) < 1}. Clearly, I is an
ideal of R. If h 2 R \ I, then dimF h(V ) = 1, and hence there exist u1 , u2 , · · · 2 V
such that h(u1 ), h(u2 ), . . . are linearly independent. Let f, g 2 R be such that
f (xi ) = ui and g(h(ui )) = xi . Then g h f = idV . Thus I is a maximal ideal of R,
whence R/I is aF simple ring. (In fact, I is the only nontrivial ideal of R.) S
Partition
1
{x1 , x2 , . . . } as i=1 Si with |Si | = 1, and let Ji = {f 2 R : ker f j i Sj }.
Each Ji is a left ideal of R and Ji + I ( Ji+1 + I. (Choose f 2 Ji+1 such that
dimF f (hSi i) = 1. Then f 2 / Ji + I.) Hence (Ji + I)/I ( (Ji+1 + I)/I, whence
R/I is not left noetherian. By Theorem 4.26, R/I is not left artinian.
Fact. If R is a simple ring and is left artinian, then R is left semisimple.
Proof. J(R) is a proper ideal of R, so J(R) = 0. ⇤
148 4. NONCOMMUTATIVE RINGS

Endomorphism ring. The endomorphism ring of a module R M is EndR (M ) =

(HomR (M, M ), +, ).
Lemma 4.14 (Schur’s lemma). If RM is a simple R-module, then EndR (M )
is a division ring.
Proof. Let 0 6= f 2 EndR (M ). We show that f is an isomorphism of M .
Since 0 6= f (M ) ⇢ M and M is simple, we have f (M ) = M . Since ker f ( M , we
have ker f = 0. ⇤
n
z }| {
Proposition 4.15. Let R L be an R-module and V = L ··· L. Then
EndR (V ) ⇠
= Mn EndR (L) .
Proof. Let ◆i : L ! L ··· L, x 7! (0, . . . , 0, x, 0, . . . , 0), and ⇡i : L ···
i
L ! L, (x1 , . . . , xn ) 7! xi . Define

✓: EndR (V ) ! Mn EndR (L)

f ! [⇡i f ◆j ]1i,jn .
Then it is easy to see thatP ✓ is an abelian group isomorphism whose inverse is
given by [fij ]1i,jn 7! i,j ◆i fij ⇡j , where fij 2 EndR (L). It remains to show
that ✓ preserves multiplication. For all f, g 2 EndR (V ), we have ✓(f g)ij = ⇡i f g◆j ,
✓(f )ik = ⇡i f ◆k and ✓(g)kj = ⇡k g◆j . Therefore,
⇥ ⇤ X ⇣X ⌘
✓(f ) ✓(g) ij = ⇡i f ◆k ⇡k g◆j = ⇡i f ◆k ⇡k g◆j = ⇡i f g◆j = ✓(f g)ij .
k k

So ✓(f g) = ✓(f )✓(g). ⇤

The opposite ring. Let (R, +, ·) be a ring. The opposite ring R is Rop =
(R, +, ⇤), where a ⇤ b = ba for all a, b 2 R.
Proposition 4.16. Let R be a ring. Then EndR (R R) ⇠
= Rop .
Proof. Define
: EndR (R R) ! Rop
f 7 ! f (1).
1 We claim that is a ring homomorphism. Let f, g 2 EndR (R R). Clearly,
(f + g) = (f ) + (g). Also,
(f g) = (f g)(1) = f (g(1)) = f (g(1)1R ) = g(1)f (1) = (f ) ⇤ (g).
Clearly, (idR ) = 1Rop .
2 We claim that is onto. For each r 2 Rop , let f : R R ! R R, x 7! xr. Then
f 2 EndR (R R) and f (1) = r.
3 We claim that ker = {0}. Let f 2 ker . Then f (1) = 0. For each r 2 R,
we have f (r) = f (r1R ) = rf (1) = 0. So f = 0. ⇤

Proposition 4.17. Let R be a ring. Then Mn (R)op ⇠

= Mn (Rop ).
 4.2. STRUCTURE OF SEMISIMPLE RINGS 149

Proof. Let ⇤ denote the multiplication in ( )op and let ⇧ denote the multipli-
cation in Mn (Rop ). Define
f: Mn (R)op ! Mn (Rop )
A 7 ! AT .
Clearly, f is an abelian group isomorphism. It remains to show that f (A ⇤ B) =
f (A) ⇧ f (B) for all A, B 2 Mn (R)op . Let A = [aij ] and B = [bij ]. Then
⇥ ⇤ X
f (A ⇤ B)ij = f (BA)ij = (BA)T ij = (BA)ji = bjk aki ,
k
⇥ ⇤ X X
T T
f (A) ⇧ f (B) ij
= [A ⇧ B ]ij = aki ⇤ bjk = bjk aki .
k k

So the proof is complete. ⇤

Proposition 4.18. Let R be a ring. The column module Rn is a left Mn (R)-

module. We have EndMn (R) (Rn ) ⇠
= Rop .
Proof. Define
✓: Rop ! EndMn (R) (Rn )
a 7 ! ✓(a),
where
n n
✓(a) : 2R 3 ! 2 R3
x1 x1
6 . 7 6 . 7
6 . 7 7 ! 6 . 7 a.
4 . 5 4 . 5
xn xn
It is easy to see that ✓ is one-to-one ring homomorphism. It remains to show that
✓ is onto. Let f 2 EndMn (R) (Rn ). We have
2 3 2 32 3 2 3 2 3 2 3
1 1 1 1 1 a
⇣6 0
6 7
7⌘ ⇣ 6
6 0
7 6 7⌘ 6
76 70 6 0
7 ⇣6 7 ⌘ 6 7
7 6 70 607
f 6 .
6.7
7 =f 6
6 .
7 6.7 = 6
7 6.7 6 ..
7 f 6.7 = 6.7
7 6.7 6.7
4.5 4 .. 5 4.5 4 . 5 4.5 4.5
0 0 0 0 0 0
for some a 2 R. Then
2 3 2 3
2 3 2 3 1 2 3 1
x1 x 0 ··· 0 6 7 x1 0 ··· 0
⇣6 7 ⌘ ⇣6 1 6 07 ⌘ 6 . ⇣6 7⌘
.
. 7 6 .. .. .. 7 .. .. 7 607
7 6
f 6 . 5 6 .. 7 = 4 ..
76 7 6 7
4 . 5 =f 4 . . . . 5 f 6 .. 7
4.5 4.5
xn xn 0 ··· 0 xn 0 ··· 0
0 0
2 3
2 3 a 2 3
x1 0 ··· 0 6 7 x1
6 . .. .. 7 607 6 . 7
=6 . . 5 6 .. 7 = 4 .. 5 a,
76 7 6 7
4 . .
4.5
xn 0 ··· 0 xn
0
i.e., f = ✓(a). ⇤
150 4. NONCOMMUTATIVE RINGS

Theorem 4.19 (Wedderburn-Artin, structure of semisimple rings). Every left

semisimple ring R is isomorphic to
Mn1 (D1 ) ⇥ · · · ⇥ Mnk (Dk ),
where ni 2 Z+ and Di is a division ring. Moreover, (n1 , D1 ), . . . , (nk , Dk ) are
uniquely determined by R.
Proof. Existence of the isomorphism. Since R is left semisimple, R = J1
· · · Jn , where each Ji is a minimal left ideal of R. Group J1 , . . . , Jn into isomor-
phism classes. We can write
M k M ni
R= Lij ,
i=1 j=1
where {Lij : 1  i  k, 1  j  ni } = {J1 , . . . , Jn } and Lij ⇠
= Li0 j 0 if and only if
Lni Lk
i = i0 . Let Ai = j=1 Lij . Then R = i=1 Ai .
1 We claim that all simple submodules of Ai are isomorphic to Li1 . The module
Ai has a composition series 0 ⇢ Li1 ⇢ Li1 Li2 ⇢ · · · ⇢ Li1 · · · Lini = Ai
whose factors are all ⇠
= Li1 . Let B be a simple submodule of Ai . Then 0 ⇢ B ⇢ Ai
can be refined to a composition series of Ai , and B is a factor of this composition
series. By the Jordan-Hölder theorem, B ⇠ = Li1 .
2 We claim that
EndR (R) = EndR (A1 ··· Ak ) ⇠
= EndR (A1 ) ⇥ · · · ⇥ EndR (Ak ).
Let f 2 EndR (R). We first show that f (Ai ) ⇢ Ai . Assume to the contrary that
f (A1 ) 6⇢ A1 . Let ⇡i : A1 · · · Ak ! Ai be the projection. Then there exists
i > 1 such that ⇡i f (A1 ) 6= 0. Thus there is j such that ⇡i f (L1j ) 6= 0. Since L1j is
simple, ⇡i f |L1j : L1j ! ⇡i f (L1j ) is an isomorphism. Since ⇡i f (L1j ) ⇢ Ai , by 1 ,
⇡i f (L1j ) ⇠
= Li1 ⇠
6 L1j , which is a contradiction.
=
Now define
: EndR (A1 ··· Ak ) ! EndR (A1 ) ⇥ · · · ⇥ EndR (Ak )
f 7 ! (f |A1 , . . . , f |Ak ).
Clearly, f is an isomorphism.
ni
z }| {
3 Since Ai ⇠
= Li1 · · · Li1 , by Proposition 4.15,
EndR (Ai ) ⇠
= EndR (Li1 · · · Li1 ) ⇠
= Mni (EndR (Li1 )) = Mni ( i ),

where i = EndR (Li1 ) is a division ring (Schur’s lemma). Therefore, by Proposi-

tion 4.16,
Rop ⇠= EndR (R) ⇠ = EndR (A1 ) ⇥ · · · ⇥ EndR (Ak )
⇠
= M n1 ( 1) ⇥ · · · ⇥ M nk ( k ).

So by Proposition 4.17,
⇠ Mn ( 1 )op ⇥ · · · ⇥ Mn ( op ⇠ op op
R= 1 k k) = M n1 ( 1 ) ⇥ · · · ⇥ M nk ( k ),
op
where i is also a division ring.
Uniqueness of (n1 , D1 ), . . . , (nk , Dk ). Assume that
R⇠= Mn1 (D1 ) ⇥ · · · ⇥ Mnk (Dk ) ⇠ = Mm1 (D10 ) ⇥ · · · ⇥ Mml (Dl0 ),
 4.3. THEOREMS OF WEDDERBURN, HOPKINS-LEVITZKI AND MASCHKE 151

where mi > 0 and Di0 is a division ring. Let Ri and Ri0 denote the image of Mni (Di )
and Mmi (Di0 ) in R respectively. Then Ri and Ri0 are ideals of R and are simple
rings themselves. We claim that for every 1  i  k, there exists 1  j  l such
that Ri = Rj0 . (Then it follows that k = l and, after a permutation of the indices,
Ri = Ri0 , 1  i  k.) Write 1Ri = a1 + · · · + al , where aj 2 Rj0 . There exists
1  j  l such that aj 6= 0. Then aj = 1Ri 1Rj0 2 Ri \ Rj0 , so Ri \ Rj0 is a nonzero
ideal of Ri and of Rj0 . Thus Ri = Ri \ Rj0 = Rj0 .
Therefore, we have k = l and Mni (Di ) ⇠ = Mmi (Di0 ), 1  i  k. It remains to
⇠
show that if Mn (D) = Mm (D ), where m, n > 0 and D, D0 are division rings, then
0

n = m and D ⇠ = D0 .
Let Li = {[0, . . . , 0, a, 0, . . . , 0] 2 Mn (D) : a 2 Dn }, 1  i  n. Each Li is
i
a minimal left ideal of Mn (D) and Mn (D) = L1 · · · Ln . Using composition
series, it is clear that all minimal left ideals of Mn (D) are ⇠
= L1 ⇠
= Dn . By Propo-
sition 4.18, D = EndMn (D) (L1 ). Under the isomorphism Mn (D) ⇠
op ⇠
= Mm (D0 ),
L1 is isomorphic to a minimal left ideal L of Mm (D ) and by Proposition 4.18,
0 0

D0op ⇠= EndMm (D0 ) (L0 ). So

Dop ⇠
= EndMn (D) (L1 ) ⇠
= EndMm (D0 ) (L0 ) ⇠
= D0op ,
and hence D =⇠ D0 . Finally, n is the length of the composition series of M (D) Mn (D)
n
and m is the length of the composition series of Mm (D0 ) Mm (D0 ), whence n = m. ⇤

Corollary 4.20. A ring R is left semisimple if and only if R is right semisim-

ple.
If R is an algebra over a field F , then all the rings in Lemma 4.14, Proposi-
tions 4.15 – 4.18 and Theorem 4.19 are F -algebras and all the ring isomorphisms
there are F -algebra isomorphisms; in this case, Theorem 4.19 can be stated as
follows.
Theorem 4.21 (Structure of semisimple algebras). Every semisimple algebra
A over a field F is isomorphic to
Mn1 (D1 ) ⇥ · · · ⇥ Mnk (Dk ),
where ni 2 Z+ and Di is a division algebra over F . Moreover, (n1 , D1 ), . . . , (nk , Dk )
are uniquely determined by A. (Note: A semisimple algebra (division algebra) is
an algebra which is a semisimple (division) ring.)
In Theorem 4.21, dimF A < 1 if and only if dimF Di < 1 for all i.

4.3. Theorems of Wedderburn, Hopkins-Levitzki and Maschke

This section contains several classical theorems in ring theory.
• Wedderburn’s theorem: Finite division rings are fields.
• Hopkins-Levitzki’s theorem: For a ring R, DCC on R R implies ACC on
R R.
• Maschke’s theorem: The group ring k[G] of a finite group over a field k is
semisimple if and only if char k - |G|.
Theorem 4.22 (Wedderburn). Every finite division ring D is a field.
152 4. NONCOMMUTATIVE RINGS

Proof. Recall that Z(A) denotes the center of a group or a ring A and CA (a)
denotes the centralizer of a in A.
Let Z = Z(D) = Fq . Assume to the contrary that D is not a field. Then
dimZ D = n > 1. For each a 2 D⇥ \ Z ⇥ , CD (a) is a proper sub division ring of D.
Let d(a) = dimZ CD (a). Then d(a) | n and d(a) < n. We have
|CD⇥ (a)| = |CD (a)| 1 = q d(a) 1.
Let [a1 ], . . . , [ak ] be the conjugacy classes of D not contained in Z(D⇥ ) = Z ⇥ . By
⇥

the class equation of D⇥ ,

k
X Xk
qn 1
(4.2) qn 1 = |D⇥ | = |Z(D⇥ )| + |[ai ]| = q 1+ .
i=1 i=1
q d(ai ) 1
Q
Let ⇣ = e2⇡i/n and let n (x) = 1kn, gcd(k,n)=1 (x ⇣ k ) 2 Z[x] be the nth
Q
cyclotomic polynomial over Q. Since xn 1 = c|n c (x), n | (xn 1)/(xd 1)
in Z[x] for all d | n, d < n. Thus in Z, n (q) | (q n 1)/(q d 1) for all d | n, d < n.
By (4.2), we have n (q) | q 1. However, since |q ⇣ k | > |q 1| for 1  k  n 1,
we have Y Y
| n (q)| = |q ⇣ k | > |q 1| q 1,
1kn 1kn
gcd(k,n)=1 gcd(k,n)=1
which is a contradiction. ⇤
Wedderburn’s theorem has several generalizations. In Theorems 4.23 – 4.25,
the ring is not assumed to have identity.
Theorem 4.23 (Jacobson [17]). Let R be a ring such that for each a 2 R,
there is an integer n(a) > 1 such that an(a) = a. Then R is commutative.
Theorem 4.24 (Herstein [13]). Let R be a ring such that for any x, y 2 R,
there is an integer n(x, y) > 1 such that (xy yx)n(x,y) = xy yx. Then R is
commutative.
Theorem 4.25 (Herstein [12]). Let R be a ring such that for each a 2 R,
there exists p(x) 2 Z[x] such that a a2 p(a) 2 Z(R). Then R is commutative.
Theorem 4.26 (Hopkins-Levitzki). If a ring R is left artinian, it is left
noetherian.
Proof. We show that R R has a composition series. Let J = J(R). By Propo-
sition 4.7, J m = 0 for some m > 0. Since
R = J0 J1 ··· J m = 0,
it suffices to show that for each 0  i  m 1, J i /J i+1 has a composition series.
Clearly, R/J is left artinian. Since J(R/J) = 0, R/J is semisimple. J i /J i+1
is an R/J-module. By Proposition 4.11 (ii), J i /J i+1 is a semisimple R/J-module.
Thus J i /J i+1 is a direct sum of simple R/J-modules. Since J i /J i+1 has DCC as
an R-module, J i /J i+1 has DCC as an R/J-module. Therefore, J i /J i+1 is a direct
sum of finitely many simple R/J-modules. Thus J i /J i+1 , as an R/J-module, has
a composition series
(4.3) J i /J i+1 = M0 ··· Mk = 0.
(4.3) is also a composition series of J /J
i i+1
as an R-module. ⇤
 4.3. THEOREMS OF WEDDERBURN, HOPKINS-LEVITZKI AND MASCHKE 153

Definition 4.27. Let B ⇢ A be R-modules. A projection from A to B is an

onto R-map f : A ! B such that f 2 = f .
If f : A ! B is a projection, then A = B ker f .
Theorem 4.28 (Maschke). Let G be a finite group and k be a field. Then k[G]
is semisimple if and only if char k - |G|.
Proof. (() Let I be a left ideal of k[G]. We show that I is a direct summand
of k[G]. Since I is a k-subspace of k[G], there is a k-module projection ⇡ : k[G] ! I.
Define
1 X
⇢(x) = y⇡(y 1 x), x 2 k[G].
|G|
y2G
It is easy to see that ⇢ : k[G] ! I is also a k-module projection. We claim that ⇢ is
a k[G]-map. It suffices to show that ⇢(ax) = a⇢(x) for all a 2 G and x 2 k[G]. We
have
1 X 1 X 1
⇢(ax) = y⇡(y 1 ax) = a a y⇡((a 1 y) 1 x) = a⇢(x).
|G| |G|
y2G y2G

Therefore k[G] = I ker ⇢ and I is a direct summand of k[G].

()) Let G act on k trivially, that is, ga = a for all g 2 G and a 2 k. Then k
becomes a k[G]-module. Define a k[G]-map
✏: k[G] ! k
X X
ag g 7 ! ag ,
g2G g2G

which is called the augmentation map. Since k[G] is semisimple, we have k[G] =
ker ✏ L for some left ideal L of k[G].
P Since dimk L = |G| dimk ker ✏ = 1, we have
L = kv = k[G]v for some 0 6= v = g2G ag g 2 k[G]. Since v 2 / ker ✏, ✏(v) 6= 0. For
each h 2 G, there exists 2 k such that hv = v. So ✏(v) = ✏(hv) = ✏(v), which
implies that = 1. Then
X X
h ag g = hv = v = ag g for all h 2 G.
g2G g2G

This menas that all the coefficients ag , g 2 G, are the same. Thus |G|a1 = ✏(v) 6= 0,
whence char k - |G|. ⇤