Theory of
Commutative Fields
Translations of
MATHEMATICAL
MONOGRAPHS
Volume 125
Theory of
Commutative Fields
Masayoshi Nagata
Translated by
Masayoshi Nagata
0. Basic symbols 1
1. Mappings 1
2. Ordered sets 3
3. Partitions and equivalence relations 4
Exercises 37
vii
viii CONTENTS
Appendix 1 73
Appendix 2 74
Exercises 75
Exercises 126
Exercises 182
Masayoshi Nagata
September 1992
Xi
Preface to the New Japanese Edition
It has been 18 years since the manuscript of the original Japanese edition
was completed. After its publication, the author noticed several points that
should be improved. Because of this and in view of the length of time since
the publication of the original edition, the author proposed to write a revision.
The author wishes to express his thanks to the publisher for accepting this
proposal.
The main goal of this new Japanese edition is the same as that of the
original and can be stated in three parts.
(1) The prerequisites should be as few as possible. (The prerequisite results
on set theory are stated in Chapter 0 without proofs.)
(2) All results considered by the author to be important and fundamental
in the theory of commutative fields are included.
(3) Chapter I consists of the basic results on group theory and the theory
of commutative rings that are needed to achieve the purpose stated in (2).
In this new Japanese edition, the author has improved several points in
the first edition and added some new topics.
Masayoshi Nagata
March 1985
Xlll
Preface to the Original Japanese Edition
Masayoshi Nagata
December 1966
XV
CHAPTER 0
0. Basic symbols
We assume the readers are familiar with basic notions such as sets, ele-
ments of a set, subsets, empty set, intersection and union of subsets.
The following symbols related to sets are used freely.
E) If a is an element of M, then a c M or M D a.
0 If a is not an element of M, then a V M or M a.
C D If N is a subset of M, then N C M or M D N.
C D If N C M, N# M, then N C M or M D N. (Note that in some
literature, our C is expressed by c ; our c is expressed by C.)
1. Mappings
A mapping f of a set M to a set N is a correspondence which associates
with each m E M a single element n E N. If f is such a mapping, then n
is called the image of m, and there are two types of expression of the image.
One is fm (or f (m)) and the other is mf If M1 is a subset of M, then
.
1
2 0. PREREQUISITES FROM SET THEORY
If we associate with each element (a, , ... , an) of the direct product MI x
x Mn of sets Mi its ith component al , then it is a mapping of the direct
product onto Mi , and this mapping is called the projection of H M to Mi.
jf
The projection, in the case of the product of infinitely many sets, is defined
similarly.
If f is a mapping of a set MI to a set M2 and if g is a mapping of M2
to a set M3, then we obtain a mapping which associates with each m E MI
the image of fm (or mf) under g. This mapping is called the product or
the composite of f and g, and is denoted by g f (in the case of fm-type
expression) or f g (in the case of mf -type expression), because the image of
m is expressed by g f m or mfg , respectively.
To each set, the notion of cardinality, which expresses something like the
number of elements, is well defined. The cardinality of a set M is denoted
by #(M). For a finite set (namely, a set consisting of a finite number of
elements), its cardinality is the number of elements. (The cardinality of the
empty set is 0) . The cardinality of the set of natural numbers is said to be
countably infinite. Countable means finite or countably infinite. The cardinal-
ity of the set of real numbers is called the cardinality of the continuum . The
product #(M) x #(N) of cardinalities is defined to be the cardinality of the
direct product M x N. If M, N are finite sets, then the product coincides
with the product as numbers. As for the product of infinite cardinalities, see
Theorem 0.1.3 below.
If there is a one-to-one mapping of M onto N, then the cardinality of
M is defined to be the same as that of N. If there is a one-to-one mapping
of M onto a subset of N, then we have that #(N) > #(M). Then we have
the
COMPARABILITY THEOREM FOR CARDINALITIES. For any two sets M, N :
(i) Either #(M) > #(N) or #(N) > #(M).
(ii) If both of these inequalities hold, then #(M) = #(N).
We have some other results as follows.
THEOREM 0.1.1. Let S(M) be the set of all subsets of a given set M. Then
#(S(M)) > #(M).
THEOREM 0.1.2. If #(M) is countably infinite, then #(S(M)) is the car-
dinality of continuum. Therefore, the cardinality of the set of real numbers is
not countable.
THEOREM 0.1.3. If M is an infinite set (namely, #(M) is not finite), then
the cardinality of the direct product M x M coincides with that of M.
One can find proofs of these results in standard text books on set theory.
EXERCISE. In the case where #(M) is countably infinite, give a direct
proof of Theorem 0.1.3 using the following idea. Assigning a number to each
element of M , we arrange all elements of M as m , 5 ... , mn , ... , and then
2. ORDERED SETS 3
2. Ordered sets
If a relation > is defined on a set M (namely, for each pair (a, b) of
elements a, b of M, it is well defined whether a > b or not), and if >
satisfies the following three conditions (1), (2), (3), then we call > an order,
and we say that M is an ordered set.
(1) a > a for any aEM.
(2) If a > b and b > a , then a = b .
, then the subset {ai I i E N} does not have a maximal element. Thus,
such that al = bi for every j < i and ai > bi . Namely, we look for the
least i such that ai # bi , and then (a , ... , an) > (bt , ... , bn) is defined
1
by ai > bi . The order defined in this way is called the lexicographical order
on the direct product M1 x ... X Mn M.
EXERCISE. Prove under the above circumstances that:
(i) If M1 , ... , Mn are all linearly ordered, then M1 x x Mn is linearly
ordered.
(ii) If M1 , ... , Mn are well-ordered, then M1 x . x Mn is well-ordered.
1. Groups
Assume that M is a set and that a mapping 0 of the product set M x M
into M is given. Then we can write 0(a, b) = c to indicate that an ordered
pair (a, b) of elements a, b of M defines an element c. If we regard
such a mapping 0 as a rule that to each ordered pair of elements of M
associates an element of M, then we say that M is a binary operation, or
simply an operation, on M. If such an operation is called multiplication,
then the element c = 0(a, b) defined by a, b is called the product of a, b
and is denoted by ab. If the operation is called addition, then c = 0(a, b)
is called the sum of a, b and is denoted by a + b. If 0(a, b) = 0(b, a)
holds for any a, b in M, then we say that the operation 0 is commutative.
Assume that a binary operation, say, a multiplication, is defined on a set
G, and the operation satisfies the following three conditions, then we say that
G is a group. (See the example at the end of this section.)
(1) (ab)c = a(bc) for all a, b, c E G (associative law).
(2) There is an element e in G such that, for all a E G, ea = ae = a.
This element e is called the identity and is usually denoted by 1. If we wish
to make it clear that this is the identity of G, then we write G'
1
(3) For each a E G, there is an element a' E G such that aa' = a'a = 1 .
7
8 I. GROUPS, RINGS, AND FIELDS
PROOF. Assume that 1 and 1' are identities. Then 1 = 1 1' = 1', so
1 = 1' . Assume now that a' and a" are inverses of a. Then a' = a'(aa") _
(a'a)a" = all, so a' = a" . From the uniqueness of the inverse, we see that
a is the inverse of a-' . abb-' a-' = 1 = b-' a-lab , and we see that
(ab)-I = b-Ia-I . Q.E.D.
If a subset H of a group G forms a group with the operation on G, then
we say that H is a subgroup of G. Submodules are defined similarly.
If H. (with . running through a set A) are subgroups of a group G, then
the intersection n2EA HA is also a subgroup of G. Therefore, if a subset S
of G is given, the intersection of all subgroups of G which contain S is the
smallest subgroup of G containing S. This subgroup is called the subgroup
generated by S. The group generated by a single element is called a cyclic
group, and if its generator is a, then the cyclic group is denoted by (a) .
If H, and H2 are subgroups, then the subgroup generated by HI U H2 is
denoted by Hl V H2. This notation is applied also to the subgroup generated
by many subgroups (HI V H2 V V Hn , VAEA H2 , etc.).
Then
(i) aH = bH iff b-'a E H.
(ii) Ha = Hb iff ab-' E H.
(iii) aH bH iff aH n bH is the empty set.
(iv) Ha Hb iff Ha n Hb is the empty set.
1. GROUPS 9
in the form
1 2 3 4 1 2 3 4 1 2 3 4 1 2 3 4
(1) 2 1 4 3 4 3 2 1 (2) 2 3 4 1 2 1 4 3
1 2 3 4 5 2 (4) 1 2 3 4 5 2
(3) 2 3 4 5 1 2 1 4 5 3
PROOF. If g E G, then
g-1[a, b]g = [g-Iag, g-1bg]
and
-11
9 [ n ,, n2]Ini E N1}g = {[n1 , n2]I nl E N }.
From this, it follows that [N1 , N2] is a normal subgroup of G. [n1 , n2] _
ni 1(n21 n1 n2) and n1 E N1 implies [n1, n2] E N1 . Thus [N1 , N2] c N1 .
Similarly, [N1 , N2] c N2 , so [N1 , N2] c N1 n N2 . Q.E.D.
COROLLARY 1.2.5. For a given group G, let Do = G, Di = [D1_ I, D._ ] . 1
element a o 1, then a = h1 , ht_1 (hi E Hj), which shows there are two
ways toexpressa (1 EH(j i), aEH, and h1 EH (j<i), 1EHn
(m > i)), contradicting the definition. Thus H1 . . H_ I n H, = { 1}. In
.
(g'1,...,5'r)(S'I,...,gr)=(glg1,...,grgr
and G1 x x Gr is a group. This group is also called the direct product
of G1 , ... , Gr . If we identify each element g, of G, with (1 , ... , , g, , 1
1, ... , 1), then each G. is a normal subgroup of the direct product, and
GI x x Gr is the direct product of these normal subgroups in the sense of
the previous definition. (As for modules, we may say direct sum, as explained
previously.)
16 I. GROUPS, RINGS, AND FIELDS
We may say that the previous definition is from the view point of the
structure of a group and that this definition is from the view point of the
construction of a group. These two definitions are substantially the same.
3. Rings and fields
A set R with two operations, say, multiplication and addition, is called
a ring if these operations satisfy the following three conditions. (As will be
explained at the end of this section, after this section, a ring will mean a ring
under stronger conditions (namely, commutative and with identity).)
(1) R is a commutative group under the addition +. (Associativity,
existence of 0, and minus of each element, and commutativity.)
(2) The multiplication satisfies the associative law ((ab)c = a(bc)) .
(3) (Distributive law) For all a, b, c E R, we have
(a + b)c = ac + bc, c(a + b) = ca + cb.
If the multiplication is commutative in a ring R (namely, ab = ba for all
a, b E R) , then we say that R is a commutative ring. On the other hand, if a
ring R has a nonzero element and also an element e satisfying ae = ea = a
for every a e R, then e is called the identity of R and is usually denoted
by 1 or by 1 R as in the case of a group.
THEOREM 1.3.1. If R is a ring, then
(i) aER=a0=0a=0,
(ii) a, b E R = (-a)b = a(-b) _ -ab (-a) (-b) = ab
, .
0. Q.E.D.
Let R be a ring with identity. If an element a has an inverse a-' under
multiplication (namely, a-' a = as-' = 1) , then we say that a is an invert-
ible element or a unit. (The set of all units in R forms a group. Readers are
advised to prove this as an exercise.)
If a is an element of a ring R and if n is a natural number, by na we
mean the sum a + + a (n times). If the set N = {natural numbers
nI for all x E R, nx = 0} is not empty, then the smallest number in N
is called the characteristic of R. If N is empty, then the characteristic of
R is defined to be 0. Note that if R has 1, then N = {natural numbers
nlnl =0}.
Next, we define fields. A ring R with identity is called a field, if every
nonzero element of R has an inverse. In this case, the set of nonzero el-
ements of R forms a group, which is sometimes called the multiplicative
3. RINGS AND FIELDS 17
two-sided) ideal. The n-times product of a left (or right, or two-sided) ideal
I is denoted by In .
Since the main objects of this book are commutative fields and rings, we
will be working with commutative rings with identity. Therefore, hereafter
in this book, by a ring, we mean a commutative ring with identity.
EXERCISE. Let R be a ring (commutative and with 1). Prove that the
ideal generated by an element a of R is aR = {ar1r E R} and that the ideal
generated by elements a1 , ... , an of R is a1 R+ . +a,,R = {>i a1ri I rri E R1.
A principal ideal is defined to be an ideal generated by a single element.
Here, we shall discuss direct sums of rings. When rings R , ... , Rn are
1
(a1,...,an)+(b1,...,bn)=(a1+b1,...,an+bn),
then we obtain a new ring ((1 , ... , 1) is the identity and (0, ... , 0) is zero).
This new ring is called the direct sum, or the direct product, of R , ... , Rn .
1
(This is the definition from the view point of construction of a ring.) Assume
that I, , ... , In are ideals such that 11 +...+In = R, (It + ..+Is)nIs+1 = {0}
for each s = 1, 2, ... , n - , and Is {0} for each s = 1 , 2, ... , n , then
1
R is the direct sum of II , ... , in as modules (cf. 2). If we write the identity
1 of R in the form e1 +. - .+e,, (ei E I) , then (i) e,ej = 0 for i : j, because
I,I c I n I = 101, and (ii) for x c I we have x l = xet + . + xe,, . Then,
. -
by the property of direct sum, it follows that x = xej (xe1 = 0 for i 0 j).
Therefore, ei is the identity of and we say that R is the direct sum (or
ii,
direct product) of rings I, , ... , In (the definition from view point of the
structure of rings; cf. the two definitions of direct product of groups in 2).
The symbols x , are used similarly as in the case of groups and modules.
20 I. GROUPS, RINGS, AND FIELDS
the direct sum S of R/I1 , ... , R/I defined by q5a = (q1a, ... , qsa) (a =
(a mod 11 n .. n IS) with a E R) . a is in the kernel of 0 iff a E I for all j
that is, a = 0. Thus, 0 is an injection. Now, it suffices to show that 0 is an
onto mapping. Let (a1 , ... , as) be an arbitrary element of the direct sum
S, where aj _ (aj mod I) . Since I1 + I2 Is = R by (ii), we may assume
that a 1 E I2 Is I. Similarly, we may assume that ai E I ... Ii-1 Ii+
1
Is
1 .
(Ea1ml) (Eb1m1) = E
m,
T,
mI mA=n:+
ajbk m1.
24 I. GROUPS, RINGS, AND FIELDS
where c. ER. Then 0 is a homomorphism from the ring R[XI, ... , Xn]
onto the ring R[a1 , ... , an] .
The proof is easy and the reader is advised to prove this theorem.
Under these circumstances, we say that aI , ... , an are algebraically inde-
pendent over R if 0 is an isomorphism. Otherwise, we say that aI , ... , an
are algebraically dependent over R. If aI , ... , an are algebraically inde-
pendent over R, we may say that R[al , ... , an] is the polynomial ring in
aI , ... , an over R. On the other hand, the kernel of 0 is called the relation
ideal of aI , ... , an over R.
THEOREM 1.5.2. If R is an integral domain, then the polynomial ring R[X]
over R is an integral domain.
PROOF. Let f = Emo ai X ` , g = E a bi X' be nonzero elements of
R[X]. We may assume that am 0 0, bn 0 0. Then the term of degree
m + n in the product fg is ambnXm+n , and this is different from 0 because
R is an integral domain. Therefore R[X] is an integral domain. Q.E.D.
The remainder theorem and the factor theorem hold also over a general
ring.
THEOREM 1.5.3. If f (x) is a polynomial in one variable x over a ring
R and if a E R, then there is a polynomial q(x) E R such that f (x) =
(x - a)q(x) + r, where r = f (a) . In particular, if f (a) = 0, then f (x) is
divisible by x - a. (The first half is the remainder theorem, and the latter
half is the factor theorem.)
PROOF. We see easily the existence of q(x)(The reader is advised to
.
6. Unique factorization
A nonzero element p of a ring R is called a prime element if the principal
ideal pR is a prime ideal of R. In the ring of rational integers, prime
numbers are prime elements. For elements a, b of a ring R, if there is an
element c such that a = be (namely, if a E bR) , then we say that a is
a multiple of b and that b is a divisor of a. If there is such an element
c which is invertible (if R is an integral domain, then this condition is
equivalent to aR = bR) , then we say that b is an associate of a.
Let a be a noninvertible element of R. It is obvious that invertible
elements and associates of a are divisors of a. If a has no other divisors,
then we say that a is irreducible. (If R is an integral domain, then this is
equivalent to the condition that if a = be (b, c E R) then one of b, c is
invertible.)
THEOREM 1.6.1. In an integral domain, every prime element is irreducible.
PROOF. Let p be a prime element in an integral domain R, and assume
that p = be with b, c E R. Since be E pR (a prime ideal), either b c pR
or c E pR. We can assume that b E pR. Then b = pd (d c R), and
p = be = pdc. It follows that p(1 - dc) = 0. Since R is an integral
domain, we have that 1 - dc = 0, namely, dc = 1 , which means that c is
invertible. Q.E.D.
A ring R is called a unique factorization domain, if R is an integral
domain and if every nonzero, noninvertible element of R is the product
of a finite number of prime elements. (The reason for the use of the word
"unique factorization" will be seen from Theorem 1.6.2.) In this case, every
irreducible element is a prime element (because it is a product of prime
elements).
THEOREM 1.6.2. An integral domain R is a unique factorization domain
if and only if the following two conditions (i), (ii) hold-
(i) Every nonzero, noninvertible element of R can be expressed as the prod-
uct of a finite number of irreducible elements.
(ii) If a = b1 b,n = c1 cn (b1, cj are all irreducible) then m = n,
and after suitable renumbering of cj, we have that each bl is an associate of
cl (a more rigorous statement might be "there is a permutation n on the set
{ 1, ... , n} so that each bi is an associate of cn1) .
Under these circumstances, every irreducible element is a prime element.
PROOF. Assume first that R is a unique factorization domain. Then (i)
obviously holds, because prime elements are irreducible. We noted just be-
fore the theorem that every irreducible element is a prime element. We shall
prove (ii) by induction on m. If m = 1 , then a is irreducible, and (ii)
26 I. GROUPS, RINGS, AND FIELDS
holds in this case. Assume now that n > 1 . Since bI R is a prime ideal and
c, cn E b, R , we see that one of ci, say c,, is in b, R . Then c, =b , ci .
lows now that b2 bin = (cl C2) C3 . . cn . Here, c', c2 is an associate of C2 , and
.
Thus, I = aR, and R is a principal ideal ring. This completes the proof of
the first assertion.
Assume next that D is a principal ideal domain. As for the second asser-
tion, it suffices to show that every nonzero, noninvertible element a can be
expressed as the product of a finite number of prime elements. There is a
maximal ideal MI containing aD. By assumption, Ml = pID (p1 (=- D).
Since MI is a prime ideal, pI is a prime element. Since aD is contained
in MI , a = plat (a, E D). If aI is not invertible, then the same argu-
ment is applied to aI , and so on. Thus, we have a = pI . . p,ta,, with prime
.
REMARK 2. It follows from the lemma that if f(X), g(X) are primitive
polynomials, then the product f(X)g(X) is also primitive. (The reader is
advised to prove this.)
LEMMA 1.6.6. Under the circumstances, let K be the field of fractions
of R. Assume that f(X), g(X) E R[X] are such that g(X) is primitive
and such that f(X) is divisible by g(X) as polynomials in K[X], namely,
f(X) = g(X)h(X) with h(X) E K[X]. Then f(X) is divisible by g(X) as
polynomials in R[X].
PROOF. Considering the common denominator of the coefficients of h (X) ,
we see that there is a nonzero element a of R such that ah(X) E R[X].
Considering the common divisors of the coefficients of ah (X) , we can write
ah(X) = bh'(X) with b E R and primitive h'(X). By cancellation of the
common divisors of a, b, we can assume that a and b have no common
divisor (except for invertible elements). Then af(X) = bg(X)h'(X). If a
prime element p is a divisor of a, then p is a prime element in R[X], and
therefore, p must divide one of b, g(X), h'(X). This is a contradiction.
Therefore a is invertible and h(X) E R[X]. Q.E.D.
COROLLARY 1.6.7. Under the circumstances, a primitive polynomial p(X)
is a prime element in R[X], if p(X) is a prime element in K[X].
PROOF. Assume that h(X), g(X) E R[X] and h(X)g(X) E p(X)R[X].
Since p (X) is a prime element in K [X ] , one of h (X) , g (X) , say h (X) , is
divisible by p (X) in K[X ] . Since p (X) is primitive, h (X) is divisible by
p(X) in R[X]. Q.E.D.
PROOF OF THEOREM 1.6.4. Let f (X) be a nonzero element of R[X].
Then f (X) = a fi (X) with a E R and primitive fl (X) . Lemma 1.6.5
shows that a is the product of a finite number of prime elements (in R[X]).
Since K[X] is a unique factorization domain by Theorem 1.6.3, f1(X) is
the product of a finite number of prime elements, say p (X), ... , pr (X )
1
These cl are called the roots of f (X) in K , and the multiplicity of a root
c appearing in the sequence is called the multiplicity of the root c. A root
with multiplicity > 2 is called a multiple root and a root of multiplicity > m
is called an m-ple root.
PROOF. The first half follows immediately from the uniqueness of factor-
ization, and the latter part follows from the factor theorem. Q.E.D.
7. Modules
Let R be a ring. A module M is called a left R-module if an operation
(multiplication) of R on M is defined in such a way that for any r E R and
for any m E M their product rm (E M) is defined and the multiplication
satisfies the following three conditions:
(i) r(m I + m2) = rm 1 + rm2 for all r E R, m , m2 E M'
(ii) r1(r2m) = (r1 r2)m and (r1 + r2)m = rim + r2m for all r1 , r2 E R1
mEM.
(iii) 1 m = m for 1 , the identity of R, and for any m in M.
We define a right R-module similarly with multiplication of elements of
R to elements of M from the right-hand side.
When a left R-module M is given, we can define such a multiplication by
r E R, m E M = mr = rm (because of the commutativity of R), and M
becomes a right module. Therefore, from now on, we do not distinguish left
and right for R-modules. So, we simply say that M is an R-module or M
is a module over R. When M is an R-module, we mean by a submodule a
submodule which is an R-module, unless otherwise specified. (So, it may be
proper to call an R-submodule.)
Let M , N be R-modules (R being a ring). Then a homomorphism f of
additive groups M to N is called an R-homomorphism, or a homomorphism
of R-modules, if it satisfies the condition that r(f m) = f (rm) for all r E
R, m E M. If it is clear that we are speaking of R-modules, then by a
homomorphism we mean an R-homomorphism. Similarly for isomorphisms;
we consider isomorphisms of modules which are R-homomorphisms. Then,
we obtain results similar to those we proved in 2. For instance, Theorem
1.2.1-Theorem 1.2.3 are modified as follows.
THEOREM 1.7.1. (i) If N is a submodule of an R-module M, then the
natural homomorphism from M onto MIN is an R-homomorphism.
(ii) If 0 is a homomorphism from an R-module M to an R-module M',
then the kernel N of 0 is a submodule of M (as an R-module) and OM
is isomorphic to MIN. In this case, there is a one-to-one correspondence
between submodules H of M containing N and submodules H' of q5M in
such a way that each H corresponds H' = OH.
(iii) If N, H are submodules of an R-module M, then the natural homo-
morphism from M onto M/N gives an isomorphism from H/(N fl H) to
(H + N)/N IN.
30 I. GROUPS, RINGS, AND FIELDS
(iii) is called the isomorphism theorem for modules. (The reader is advised
to prove these results.)
EXERCISE. State the modification of Theorem 1.2.9 to the case of modules
over a ring.
An R-module (R being a ring) which is generated by a finite number of
elements (namely, a module which can be written in the form a R + +anR)
I
is called a finitely generated R-module, or simply, a finite module.
Elements mI , ... , mn of an R-module M are called linearly independent
(over R) if En I rlmi = 0 with ri E R, implies rr = 0 (i = 1 , ... , n). A
subset S of M is said to be linearly independent if every finite subset of S
consists of linearly independent elements.. For convenience, we regard that
the empty set is linearly independent.
If an R-module M is generated by a linearly independent subset S, then
we say that S is a linearly independent base, or a free base, of M and that
M is a free R-module, or simply, a free module.
THEOREM 1.7.2. Every module over a field K is a free module.
PROOF. Let F be the set of all linearly independent subsets of a K-module
M. Then F is an inductive set by the inclusion relation. This is proved as
follows. (i) F is not empty because the empty set belongs to F. (ii) Let
{S2IA, E Al be a well-ordered subset of F. If bI , ... , bin are mutually
distinct elements of S* = US,,, then 1% E A, {bI, ... , b,1 } c S.. This
means that S* is linearly independent and S* E F. Obviously if S* D S2
(for all A). If S D S. (for all A), then S D US2 = S* . Therefore S* is
the supremum of {S2 JA E A}. Thus, we proved that F is an inductive set.
Therefore, by Zorn's Lemma, there is a maximal member B of F. Let N be
the submodule generated by B over K. It suffices to show that N = M. Let
m be an arbitrary element of M. We are to show that m E N. If M E B,
then m E N. Assuming that m is not in B, we consider B U fm}. By the
maximality of B, B U {m} is not linearly independent, which means that
there is a finite subset S of BU{m} which is not linearly independent. Since
B is linearly independent, m ES, and S= fm, bI, ... , bn } with b, E B.
Then there are ci E K such that co m + E cib; = 0, (c0, cI , ... , cn) 54
(0, ... , 0) . Since bI , ... , bn are linearly independent, we have co 54 0.
Multiplying -co we may assume that co = -1 . Then m = cibi and
m E N. Q.E.D.
We can see the following result by the proof above.
THEOREM 1.7.3. If M is a module over a field K and if S is a linearly
independent subset of M, then there is a linearly independent baseof M con-
taining S.
THEOREM 1.7.4. If a module M over a field K has a linearly independent
base mI , ... , mn consisting of n elements, then every linearly independent
base of M consists of n elements.
7. MODULES 31
sr = E xi xi ...xi
I 2 r
Sn = XI X2 ... Xn
tuple d (M) = (j j=I ij, i 1, ... , in) , and we introduce an order on the set of
monomials by the lexicographical order of d (M) . Namely, for monomials
M1 , M2, MI < M2 means that, letting d(M1) _ (d0, ... , do j) (j = 1, 2),
d01 < d02 , or d01 = d02 and d11 < d12 , or , namely, for some j ,
dj1 < dj2 and dkl = dk2 if k < j. We shall prove the assertion by induction
on d (M) where M is the greatest monomial appearing in a given symmetric
form f (x1 , ... , xn) .
If only the least monomial 1 appears in f , then f is an element of R , and
the assertion is obvious. Let M be the greatest monomial among the mono-
mials appearing in f . Then our induction hypothesis is that if h is a sym-
metric form such that d (M) > d (N) for every monomial N which appears
in h, then h E R[s1 , ... , sn]. Write M as xi' . xn" . Since f is symmet-
ric, xn'1 xn'n appears in f for any U E Sn . Therefore, by the greatestness
of M , we have e1 > e2 > . . > en . Consider g = Sn"sn"-1 -e" S22 -e 3s1 I -e
.
2
8. SYMMETRIC AND ALTERNATING FORMS 35
=a
(a! a2 .. an
Q.E.D.
a1 a2 ...
an
ri-1
Q m) . Suppose that we have defined cyclic permutations n , ... , 71c-1 1 1
36 I. GROUPS, RINGS, AND FIELDS
taking a letter k such that ak # k , but k does not appear in any of it1 ,
we define nc similarly. If we proceed in this way, then those k such that
ak 54 k are exhausted, and we complete the proof. Q.E.D.
This equality and Lemma 1.8.4. imply the first half. Consider the difference
product A. Then, as we show below, with any transposition (i, j) (1 < i <
j < n) , we have (i, j )A = -A. Therefore, a given permutation a is even or
odd according to whether aO = 0 or aA = -A.
Proof of (i , j )A = -0 . Those factors (xs - xl) of A such that (i , j) (xs -
XI) (xs - x,) are divided into 5 classes as follows.
(1) (xs -x,), (xs -xj) with s < i,
(2) (xj - xe) , (x1- x,) with t > j ,
(3) (x,-xt) with i<t<j,
(4) (xs - xx) with i < s < j ,
Exercises
6
1. Generalize Lemma 1.6.5 as follows: Let P be a prime ideal of a ring R,
and consider the polynomial ring R[X1 , ... , Xn] over R. Prove that
the ideal of the polynomial ring generated by P is a prime ideal.
2. Under the assumptions in 1., prove that if I is an ideal of R, then
R[X1 , ... , Xn]/IR[X1, ... , Xn] is isomorphic to (R/I)[X1 , ... , Xn].
3. Prove that prime ideals of the ring Z of rational integers are {0} or pZ
with prime numbers p. Prove also that if p is a prime number, then
pZ is a maximal ideal and Z/pZ is a field consisting of p elements.
4. Consider the polynomial ring R[X] in one variable over a unique factor-
ization domain R. Prove that an element f of R[X] is a prime element
if and only if f is either a prime element in R or a primitive polynomial
which is a prime element in the polynomial ring in X over the field of
fractions of R.
7
1. Basic notions
Let K be a subfield of a field L, and let S be a subset of L. Then the
least subfield containing K and S is denoted by K(S) and is called the field
generated by S over K or the field adjoined S to K . If S = {al , ... , an } ,
then K(S) is denoted by K(al, ... , an). If Kl K2 are subfields of L,
,
43
44 II. ALGEBRAIC EXTENSIONS OF FINITE DEGREES
(i) We say that an element a of I' is algebraic over I if there are elements
co (34 0), c1, ...,cn of I such that
(ii) If we can choose ci so that co = 1 , then a is said to be integral over
I.
(iii) We say that I' is algebraic, or integral, over I, if every element of
I' is algebraic, or integral, respectively, over I.
(iv) An element, or I' is said to be transcendental over I, if it is not
algebraic over I.
The condition in (ii) follows if co in (i) is chosen from the invertible
elements of I. Therefore, when I is a field, "algebraic" and "integral" are
equivalent to each other, and we mainly use the term "algebraic".
If a field K' contains another field K , we say that K' is an extension
field, or simply, an extension, of K. If K' is algebraic over K, then K' is
called an algebraic extension of K. An extension field that is not algebraic
over K is called a transcendental extension of K.
In general, if a ring R and its subring S share the same identity, then the
multiplication of elements of R by elements of S is well defined. Thus, R
is an S-module.
If an extension field K' of a field K is of finite length as a K-module,
then the length is called the degree of K' over K and is denoted usually by
[K': K], (K' IK) , or lengthK K' . In this book, we use [K': K1. In this case,
K' is called a finite algebraic extension, or an algebraic extension of finite
degree, of K. On the other hand, if K' contains a transcendental element
x over K, then K[x] is isomorphic to a polynomial ring. Therefore, the
finiteness of degree implies that K' is algebraic over K.
If a field K consists of only a finite number of elements, then K is called
a finite field. A field which is not a finite field is called an infinite field.
THEOREM 2.1.2. Let K be a finite field. Then the number of elements of
K is a power pe of its characteristic p and the exponent e coincides with
[K: n] , where it is the prime field of K K. e
independent. Since a" = - (c1 a"-' + . . + c,), we see that , a, ... , a"-'
. 1
PROOF. There are elements bI , ... , b" such that L = K(bI , ... , b,,) . Let
fI (xi) be the minimal polynomial for bI over K. In general, we take f =
fj(x1 , ... , xj) E K[x1, ... , xj] so that f (b1 , ... , bj_I, x) is the minimal
polynomial for bj over K(b1 , ... , bj_I) . If we regard K[x1 , ... , xj]/I as
a homomorphic image of K[a1 , ... , aj_I ][xj], then the kernel is generated
by f (a1 , ... , aj_I , xj). Therefore, we see easily that K[a1 , ... , ai_I] is
a field and that K[a1 , ... , aj] is K-isomorphic to K[b1 , ... , bj] such that
ak -* bk by induction on j. Q.E.D.
The following lemma is proved easily by using Theorem 2.1.7.
2. Splitting fields
Let f (x) be a polynomial in one variable x over a field K. If f (x)
is the product of linear factors as a polynomial over an algebraic extension
field L of K , then we call L a splitting field of f (x) . Then f (x) = a(x -
cl) . (x - c") (n = deg f (x) , cl E L , and a is the coefficient of x" in
f (x)) . Therefore, a subfield M of L is a splitting field of f (x) if and
only if all roots c, are in M. Among such M, K(c1 , ... , c") is the least
subfield, and we call K(c1 , ... , c") a minimal splitting field of f (x) over
K.
THEOREM 2:2.1. For any given polynomial f (x) of one variable over afield
K, there is a minimal splitting field of f (x) .
PROOF OF THE LEMMA. The derivative does not change even if we con-
sider it over a splitting field L. Therefore, we assume that f (x) = ao(x -
c1) ... (x-cn) . Then f'(x) = ao E "= ... (x-c;-1)(x-c;+I) ... (x-cn) .
Assume first that c is a multiple root of f (x) . Then we may assume that
c1 = c2 = c. Then the derivative is divisible by x - c, and f'(c) = 0. Con-
versely, assume that c is not a multiple root of f (x) . We may assume that
c1 = c, ci c (i > 2). Then f'(c) = ao(c - c2) ... (c - cn) 0. Q.E.D.
THEOREM 2.3.2. Any field of characteristic 0 is a perfect field.
PROOF. It suffices to show that if f (x) is an irreducible polynomial over a
field K of characteristic 0, then f (x) has no multiple root. The derivative
f'(x) of f (x) is a polynomial over K. If f (x) and f'(x) have a com-
mon root c, then f (x) and f4(x) must have a nontrivial common divisor,
and f'(x) must be divisible by f (x) because f (x) is irreducible. Since
deg f'(x) < deg f (x) , f'(x) must be 0. But, from f (x) = aox :' + + an
(ao 0), it follows that f4(x) = naox"- I + 0 0. Q.E.D.
Let us examine the proof above in the case of characteristic p. The proof
goes well except for the last step, and we see that if an irreducible polynomial
f (x) over a field K of characteristic p 0 0 has a multiple root, then the
derivative of f (x) must be 0. This implies, by the definition of the deriva-
tive, that f (x) is a polynomial in x' . Conversely, if f (x) is a polynomial
50 II. ALGEBRAIC EXTENSIONS OF FINITE DEGREES
in x' , then its derivative f'(x) is 0 , and every root of f (x) is a multiple
root by Lemma 2.3.1.
REMARK. If Ax) = bnx"p + + brxrp + + b0 , then taking pth root ci
of b.bn 1 for each i, we have f (x) = bn (x" + C"_1 + + c0)p by Theorem
2.1.3. Therefore, every root has a multiplicity that is a multiple of p.
By the observation above, we have the following theorem.
REMARK. As one can see from the first part of the proof above, each el-
ement a has a unique pth root in the characteristic p # 0 case. Conse-
quently, a has a unique peth root for each natural number e.
THEOREM 2.3.5. Every finite field is a perfect field.
over K, the number of such a, is exactly n = [L: K]. Our observation can
be applied even if a is inseparable, and in that case, the number of mutually
distinct roots of f (x) is less than n , and the number of these injections is
less than [L: K1. Thus, we have proved that (iii) implies (iv) and also the
converse of this implication.
Next, we shall show that (ii) implies (iii). If we prove this in the case where
r = 2, then we can prove the general case by induction on r. Therefore, we
assume r = 2. By our proof of Theorem 2.5.1 above, there is an element a
such that K(al , a2) = K(a), and it suffices to show the separability of a. By
what we proved above, we have only to show the validity of (iv). There are K-
isomorphisms of , ... , aS (s = [K(al) : K]) from K(ai) into Q. We denote
5. SIMPLE EXTENSIONS 53
z
implies that a;+I E Ls (a1 , ... , a; , a p I , a +2 , , a ) . We see similarly that
a;+ I E Ls (a I , ... , a; , ap+"j, a p2 , ... , a p) for any natural number n. But,
there is an n such that ap I E Ls , which contradicts the minimality of r.
Thus, L;+, = L;(a;+I), and L; # L;+1 . We see that [L: Ls] = pr .) Since
Ls is not a perfect field, LS is not a finite field. For each c c Ls , set y(c) =
aI + cat . Then [Lo(y(c)) : Lo] = p. If c 54 c', then Lo(y(c)) 54 Lo(y(c')).
(Indeed, if Lo(y(c)) = Lo(y(c')), then y(c) and y(c') are in Lo(y(c)) and
(c - c')a2 = y(c) - y(c') E Lo(y(c)). Then a2 E Lo(y(c)), which implies
aI E Lo(y(c)) and [Lo(y(c)) : Lo] > p2 , a contradiction.) Since LS is an
infinite field, there are infinitely many intermediate fields. Q.E.D.
6. Normal extensions
We say that an algebraic extension field L of a field K is a normal ex-
tension of K if, for any a E L, is fa(x) is the minimal polynomial for a
over K then L is a splitting field of fa(x) , namely, all conjugates of every
element of L over K (considered in a sufficiently large field containing L)
are in L. The theorem below shows that this condition is satisfied if the
same condition is satisfied by the elements of a set of generators of L over
K
We add here the definition that a normal extension L of K is called a
Galois extension if L is separable over K.
THEOREM 2.6.1. Let L = K(a1, ... , a,,) be an algebraic extension of a
field K. Then L is a normal extension of K if and only if all conjugates of
every a; (i = 1, ... , n) are in L.
7. INVARIANTS OF A FINITE GROUP 55
PROOF. The only if part is obvious and we shall show the if part. Let f(x)
be the minimal polynomial for ai over K and set f (x) = Hi fi(x). Then
L is the minimal splitting field of f (x) . Let b be an arbitrary element of
L, and consider an arbitrary conjugate b' of b in a sufficiently large field
Q containing L. Then there is a K-isomorphism a from K(b) into 92
such that ab = b'. a can be extended to a K-isomorphism from L into
92. Then aL is also a minimal splitting field of 1(x), and aL = L. Thus
b' E L. Q.E.D.
If a field L contains a field K, the set AutK L of K-automorphisms of
L forms a group (with multiplication defined by (ar)a = a(ra)). If L is a
normal extension of K, then we call this group the Galois group of L over
K and denote it by G(L/K). In this case, our proof of the theorem above
shows that for any K-isomorphism from L into a field containing L, L
is mapped to L itself. If, furthermore, L is of finite degree over K, then
Corollary 2.5.4 shows that #(G(L/K)) coincides with the separable factor
of the degree of the extension. Therefore we have
THEOREM 2.6.2. If L is a normal extension of finite degree over a field
K, then the order of the Galois group G(L/K) coincides with the separable
factor of the degree of the extension. Let LS be the separable closure of K
in L. Then LS is a Galois extension of K and G(L/K) ^' G(Ls/K) by
a natural isomorphism such that a - al L s (the restriction of a of Ls) for
a E G(L/K).
From our proof of Theorem 2.6.1, we also have
THEOREM 2.6.3. Let L be a normal extension of finite degree over a field
K. Then the set of conjugates of an element a of L over K coincides with
{aala E G(L/K)}, and the set of conjugates over K of an intermediate field
M between K and L coincides with {aMla E G(L/K)}.
Q.E.D.
COROLLARY 2.7.2. Let L be a normal extension of a field K of finite
degree. Consider the separable closure LS of K in L and also the set Li of
elements of L which are purely inseparable over K. Then
(i) L. is a field and L is generated by Li and LS L.
(ii) L is a Galois extension of Li and G(L/K) = G(L/Li) .
PROOF. (i) Let f (x) be the minimal polynomial for a (E L) over K' n L
with a such that L = K(a). Let aI , ... , ar (r = deg f (x)) be the roots
of f (x) . Since L is Galois over K, conjugates of a over K are in L.
Therefore, these ai are in K'(L). Since K'(a) = K'(L), we see that K'(L)
is a Galois extension over K'. Suppose that f (x) is reducible over K' By .
This F (x) is called the nth cyclotomic polynomial. The degree of this
polynomial is 0(n), with 0 the Euler's function.
For the proof of this theorem, we note the following lemma.
LEMMA 2.9.7. If n is the prime field of characteristic p 54 0, then f (x)p =
f (xp) for any f (X) E 7r[X]
PROOF. Let f (x) _ aix` . Then f (x)p = > apx`p . Since ai are in 7r,
we have ap = ai and f (x)p = f (xp). Q.E.D.
9. ROOTS OF UNITY AND CYCLIC EXTENSIONS 61
C r a where C is an nth root of unity. (2) In the proof below, for a E Aut L
and b c L, we use b-type notation, and we mean by b 1 +'+'2+' the
bbaba2 ba,
product . . . .
of Kn_1 (Theorem 2.9.2), and the degree of the extension is less than n,
because x" - 1 is reducible. In the Galois group G = G(Kn/Kn_1), we take
a sequence G = Gr D Gr_I D D G1 D Go = {1} such that each Gi/Gi_I is
a cyclic group (take a cyclic subgroup G1 first, then considering G/GI , take
G2 so that G2/G1 is cyclic, and so on). Let Mi be the intermediate field
corresponding to Gi . Then Mo = Kn, 111, = Kn_ , and each Mi (i < r) is
1
PROOF. Conjugates of Ca over K' are of the form C'. Ca. Let these be
`' Ca, ... , C'S ra. Their product is equal to be (r a)' with c = > ii .
64 II. ALGEBRAIC EXTENSIONS OF FINITE DEGREES
hard to show, by a proof similar to the one above that if m = #(G), then
Km(L) is an extension by radicals of the field K.)
To prove the only if part, assume that L C_ K( VaI , ... , a,.) = L,.,
where ai+1 E Li = K(' a1 , ... , ai) and xni+' - ai+1 is irreducible over
Li . Let L* be the smallest Galois extension of K containing L,.. Then it
suffices to show that G(L*/K) is solvable, since G(L/K) ^_' G(L*/K)/G(L*/L)
and any homomorphic image of a solvable group is solvable. We use an
induction argument on r. If r = 1 , then the conjugates of Ca (n =
n1 , a = a1) are " a, ... , b=n-1 " a , " a , where C is a primitive nth
root of unity. Therefore, L* K(C, " a) K(C) is an Abelian extension
.
a /3 y
QTa QT/3 aTy
7C(ar).
66 II. ALGEBRAIC EXTENSIONS OF FINITE DEGREES
and so on) forms a normal subgroup N of order 4, and S4/N ' S3 (we
omit the proof). Let us pay attention to the case G(L/K) = S4 (the existence
of this case is seen in (D) below). Then we see that we have to solve a cubic
equation in order to obtain the intermediate field corresponding to N. (Then
we solve two quadratic equations.) This is the reason why we have to solve
a cubic equation in solving a quartic equation.
11. CONSTRUCTION PROBLEMS 67
we may assume that 2 points are given and that the distance between them
defines the unit of length.) But, to take an arbitrary point on each edge will
cause some difficulty because we know nothing about the distances of these
points from the vertex. Similar comments apply to points on given lines,
given half lines, or circles.
In this section, we use Greek letters a, /3 , ... to denote complex numbers;
the symbol denotes the complex conjugate. Q denotes the rational number
field.
THEOREM 2.11.1. If 0, 1 , a I , a2 , ... , an are the given points, then a
point /3 is constructed if and only if there is a sequence Ko c KI c c Kt
of fields K. such that Ko = Q(a I , ... , an , dl , ... , an ) and such that
(i) K,D/3,
(ii) [K,:K;-I]=2 (i 1,2,...,t).
PROOF. To prove the if part, note that the point of symmetry of a given
point with respect to a given line can be constructed. Therefore, ai (i =
1) 2 , ... , n) are constructed. The product (or ratio) of two complex num-
bers is obtained by adding (or subtracting) arguments and multiplying (or
dividing) absolute values. The sum of two complex numbers is the sum
as vectors. Therefore, the product and the sum of constructible complex
numbers (i.e., points) are constructible. If a is constructible, then na,
-1
n a (with n a natural number) and -a are constructible. We see that
elements of Ko are constructible. So, we shall show that elements of K, are
constructible by induction on i. Assume that elements of K._ are con- I
such that Ki = K,_1 (J) . The argument and the absolute value of
are one half and the square root of those of /3 (note that the argument is
unique modulo 27r, and therefore, there are two choices for the argument of
modulo 27r). Therefore, is constructible. Thus, the elements of
Ki are constructible.
To prove the only if part, let us see how to construct a new point. There
are 3 types of constructions:
(1) the intersection of 2 lines.
(2) one of the intersection points of a line with a circle,
(3) one of the intersection points of two circles.
Therefore, assuming that K is a field consisting of constructible com-
plex numbers and satisfying the condition that a c K for all a c K (note
that Ko satisfies these conditions), we consider these three types of construc-
tions to obtain a new point 6. If we can show that [K(8) : K] < 2 and
[K(8 , j): K(8)] < 2, then we can take successive field extensions, satisfying
(ii), until we obtain /3 .
that these are a primitive PI pr_ i th root of unity and a primitive Prth
root of unity, respectively). If t, u are rational integers, then the argument
of C1 C2 is 2n(t/(p1 . p . ) ) + u/pr) = 2n(tpr + upI ... pr- I )l n . Since pr
1
and PI . . Pr- I are coprime, a primitive nth root of unity is obtained in the
.
form C' (2 Since (I , (2 are constructible, we see that primitive nth roots
.
only to show that any regular p2-gon is not constructible with p = pi. This
-
follows from the equality q5(p2) = p(p 1) and Corollary 2.11.3. Q.E.D.
COROLLARY 2.11.6. There is an angle whose trisection is not constructible.
More generally, if p is an odd prime number, then there is an angle whose
p-section is not constructible.
PROOF. A regular p2-gon is not constructible. This implies that p-section
of either 2mr or 2n/p is not constructible. Q.E.D.
COROLLARY 2.11.7. n-section of every angle is constructible iff n is a power
of 2.
PROOF. If the n-section of an angle is constructible, then for any divisor
n' of n , the n'-section of the angle is constructible. Therefore, the assertion
follows from Corollary 2.11.6. Q.E.D.
12. ALGEBRAICALLY CLOSED FIELDS 71
order, the fact that these Li form an ascending sequence and that L is
K.
a field, we see that a is a well-defined K-isomorphism from L into K I
Appendix 1
THEOREM 2.12.4. The complex number field C is algebraically closed.
A well-known proof of this theorem uses two facts: If f is a real valued
continuous function on a bounded closed set, then f takes a minimum value
at some point, and that for a polynomial f(x) in one variable x over C,
if F(x) = If(x)I has a nonzero value at a E C, then F(x) takes a smaller
value at a point on a sufficiently small circle with center a. Here, we shall
give another proof as an application of the fundamental theorem of Galois.
We use the existence theorem of Sylow subgroups (Chapter I, 2) and the
fact that if a polynomial g(x) in x over the real number field R is of odd
degree, then g(x) has a linear factor (as is obvious by the continuity of the
real numbers).
PROOF. Take an algebraic element c over C. We have only to show that
c E C. Let L be a Galois extension of finite degree over R, such that
C(c) C_ L. Let S be a 2-Sylow subgroup of G = G(L/R) and let M be
the intermediate field corresponding to S. S = G(L/M), #(S) = [L: M],
and therefore, [M: R] = #(G)/ #(S) , which is odd. By the reducibility of
polynomials of odd degree, we have M = R, and #(G) is a power of 2.
If #(G) = 2, then L = C. Assume for a moment that #(G) = 2n with
n > 1 . Letting H be the subgroup of G corresponding to C, we take a
subgroup H* of H of index 2 (the existence of H* follows from the fact
that H is nilpotent (Theorem 1.2.6), cf. Exercise 1.2.9). Let C* be the
field corresponding to H*. Then C* is a quadratic extension of C, and is
generated by a root of some quadratic equation x2 + tax + b = 0 (a, b E C).
The roots are -a f (a2 - b). Therefore, if we see that the square roots of
an arbitrary complex number r + 2s (r, s E R) are in C, then C* = C,
contradicting n > 1 , which would complete the proof. As for the square roots
of r + 2s , the existence is obvious if s = 0 , so we assume that s 0 0 . It
suffices to find real numbers x, y such that (x + y )2 = r + 2sV-_l . The
conditions are x2 - y2 = r and xy = s. Since x4 - x2y2 = rx2 , we have
x4 - rx2 - s2 = 0. Then, since x = V(r + r2 + 4s2)/2 E R and y = s/x,
we obtain square roots (x + yam) of r + 2s . Q.E.D.
74 II. ALGEBRAIC EXTENSIONS OF FINITE DEGREES
Appendix 2
Exercises
1
1. Assume that K1 , K2 are finite fields such that K1 C_ K2, n = [K2: K1].
Prove that #(K2) = (#(K1))" .
2. Prove the following equalities on the degrees of extension fields of the
rational number field Q. Here, co denotes an imaginary cubic root of
unity.
[Q(vT) : Q] = 2, [Q(w) : Q] = 2, [Q(' 2) : Q] = 3.
3. If a is an algebraic element over a field K, by the degree of a over K
we mean the degree of the minimal polynomial for a over K. Prove
that if a is an element of a finite algebraic extension field L of K, then
the degree of a over K is a divisor of [L: K].
4. Let K, L, M be fields such that L, M are algebraic extensions of K, L,
respectively. Prove that M is algebraic over K.
5. Let K be a field of characteristic p 54 0 and let q be p" with a natural
number n. Prove that Kq = {xq,X E K} forms a subfield of K.
2
1. Find the minimal splitting field for each of the following polynomials
over the rational number field Q.
(1) X4 - X2 +4, (2) x3 - 1 , (3) x3 - 2.
2. Find the degree of extension of each field obtained above.
3
1. Determine which of the following polynomials have multiple roots. We
assume that the coefficients are in some field K. Note that the answers
depend on the characteristic of K. In (4), we assume that p is the
characteristic of K and that p 54 0.
(1) X5+x, (2) x4+X+ 1, (3) x4+2X3+3x2+8x+ 1,
n n-1
(4) cOXp +C1Xp + +Cn_1XP+C,i+dx (C1) d E K, co 54 O).
2. Give a proof of Theorem 2.3.5 along the following lines. If K is a
finite field consisting of pe elements (p being the characteristic), then
K* = K - {0} is a group of order pe _1 , and therefore the order of each
element a is coprime to p. By using the fact, with some power at of
a, show that (a`)p = a.
3. Let K be a field of characteristic p 54 0. Prove that if t is a transcen-
dental element over K, then K(t) is not a perfect field.
4. Let L be an extension field of a field K. Prove that the set Li of
elements of L, which are purely inseparable over K, forms a field.
5. When a natural number n and field K of characteristic p 0 is given,
we denote by K' the set of pth " roots of elements of K (in a suffi-
76 II. ALGEBRAIC EXTENSIONS OF FINITE DEGREES
4. Let Q denote the rational number field. Show the following: Q(v ) is
a normal extension of Q and Q(" 3) is a normal extension of Q(/) ,
but Q(4 3) is not a normal extension of Q.
5. For each of the following fields, find the smallest normal extension con-
taining it over the rational number field Q. Find, also, the Galois group
of each of the normal extensions.
(1) Q(v) (2) Q(v, /) (3) Q(3 2) (4) Q(4 2)
7
1. Find the structure of the following groups, where Q denotes the rational
number field.
(1) AutQ(3 2) , (2) AutQ(, r2),
(3) AutK(X2) K(x) (where x is transcendental over a field K) ,
(4) AutK(X3) K(x) (with K, x as above).
2. Find the fields of invariants of the groups in 1 . above.
3. Let a field SZ be a finite normal extension of a field K, and let L be
an intermediate field. Let s, t be the separable and inseparable factors
of degree of L over K. Let oI , ... , ors be a set of representatives
of G(S /L)\G(S2/K) . For each element a of L, we consider a system
consisting of a,a such that each a.a appears exactly t times. We call
such a system the complete system of conjugates of a in L over K (we do
not care in what order they are arranged). Let the system be aI , ... , an
(n = st). Then Ei= a! , fJ i= I a, are called the trace and the norm of a,
I
respectively, and they are denoted by TrL/K (a) , NL/K (a) , respectively.
Prove the following.
(1) TrL/K (a) E K, NL/K (a) E K for all a c L.
(2) TrL/K(a+b)=TrL/K(a)+TrL/K(b) and NL/K(ab)=NL/K(a)NL/K(b)
for all a, b c L.
(3) Let b1 , ... , bn be a linearly independent base of L as a K-module.
For each a E L, consider the linear transformation of L given
by x -> ax. Let p(a) be the matrix which represents this linear
transformation with respect to b1 , ... , bn . Show that TrL/K (a) is
the trace (i.e., the sum of diagonal entries) of p(a) and that NL/K(a)
is the determinant of p(a) .
(4) Prove that if L' is a finite algebraic extension of L, then
TrL/K TrL,/L = TrL?/K , NL/K NL'IL = NL,/K
(5) Prove that TrL/K(a) = 0 for every a E L if and only if L is
inseparable over K.
4. Let x be a transcendental element over the complex number field C,
and consider the field C(x). For a given natural number n, let C be
(i.e., the primitive nth root of unity with argument 2n/n) .
78 II. ALGEBRAIC EXTENSIONS OF FINITE DEGREES
(1) Prove that there is a unique pair a, r c Autc C(x) such that ax =
I,
x - 'Cx=Cx.
(2) Show that these a and T satisfy a2 = 1 , T" = 1 , ara- = T-I
1. Prove the converse of Theorem 2.9.8, i.e., that if a field K contains all
nth roots of unity, if the characteristic p of K is either 0 or prime to
the given natural number n and if x" - a (a c K) is an irreducible
polynomial over K, then K(" a) is a cyclic extension of degree n over
K.
2. A field L is called a Kummer extension of a field K if (i) L is an Abelian
extension of K of finite degree and (ii) x" - 1 has no multiple root and
K is a splitting field of x" - 1 , where n is the maximum of the orders
of elements of G(L/K).
Prove the statement below using the fact that every finite commutative
group is the direct product of cyclic groups.
A Kummer extension L, as above, is obtained as
K(., aI , 'V a2 , ... , ,Var )
with a! E K, n is the L.C.M. of n , ... , nr and each x -ai
ai_I , n,fai+I , ... , a).
I
is irreducible over
Prove the converse of this statement assuming the same condition on
K.
3. Prove that if n is a power of an odd prime number, then the cyclotomic
field of order n over an arbitrary field K is a cyclic extension of K.
Describe the case where n is a power of 2. [Hint. Use Exercise 2.4. 1.1
4. Let L be a separable algebraic extension of degree n over a field K .
Let aI , ... , a" be the mutually distinct K-isomorphisms from L into a
sufficiently large field 0, and let wI , ... , w" be a linearly independent
EXERCISES 79
10
12
1. Show that the fact that every irreducible polynomial in one variable over
the real number field R is of degree at most 2 follows from the facts that
the complex number field C is algebraically closed and that [C: R] = 2.
2. Let K be the algebraic closure of a field K. Prove that the field K* of
G(K/K)-invariants in K is the smallest perfect field containing K.
3. Let 7C be the prime field of characteristic p i4 0. Show that there is a
field K such that K is algebraic over n and #(K) = oc, but K is not
algebraically closed.
CHAPTER III
Transcendental Extensions
1. Transcendence bases
Among transcendental extensions of a field, there are special extensions
called purely transcendental extensions, which are defined as follows: A field
L is called a purely transcendental extension of a field K if L is generated
by an algebraically independent set S over K. L is also called the rational
function field of S over K. L is the field of fractions of the polynomial
ring in elements of S (which may be infinite) over K.
As we saw in the proof of Theorem 2.12.5, we have the following result.
THEOREM 3.1.1. If K is a subfield of a field L, then there is a maximal
member S among algebraically independent subsets of L over K and L is
algebraic over K(S).
This S is called a transcendence base of L over K. If L is separable
over K(S), then S is called a separating transcendence base (which may not
exist).
THEOREM 3.1.2. If K is a subfield of a field L and if both S and S' are
transcendence bases of L over K, then #(S) = #(S').
This #(S) is called the transcendence degree, or the dimension, of L over
K . We denote this by trans. degK L.
If A D I are integral domains, then the transcendence degree, denoted
by trans. deg, A , of A over I is defined to be trans. degK L with fields of
fractions L, K of A, I, respectively.
PROOF. (1) In the case where #(S) is finite, we may assume that n =
#(S) < #(S'). Let S be {x1 , ... , xn} . Take mutually different elements
yl , .. , yn of S' S. It suffices to show that, by a suitable renumbering of
xi , L is algebraic over Ki = K(y1 , ... , y; , xi+1 , ... , xn) for each i =
0, 1 , ... , n, because, then L must be algebraic over K(y.1 , ... , yn) and
S' = {y1 , ... , yn} . This assertion is true if i = 0. Therefore, assuming that
L is algebraic over K5_ , we consider the case over KS . Since ys is algebraic
1
over K,-, , there are co, ... , cl E K[y1 , ... , ys-1 , xs, ... , xn] such that
co 54 0, coys + clys-1 + + ct = 0. Since y1 , ... , ys are algebraically
independent over K, at least one xj (j > s) must be contained in some
81
82 III. TRANSCENDENTAL EXTENSIONS
[M:K]<[L:K(xl,...,xn)].
(ii) In the general case, let xI , ... , xr be a transcendence base of M over
K, and apply the proof above, by taking a transcendence base of L contain-
ing xI , ... , xr and by considering K(x1, ... , xr) instead of K. Then we
see that [M: K(x1, ... , xr)] is finite. Q.E.D.
We add here the following lemma to be used later.
LEMMA 3.1.5. Let S2 be an algebraically closed field containing afield K.
Assume that aI , ... , ar, bI , ... , br c S2 and that there is a K-isomorphism
a from an algebraic extension L (c S2) of K(a1 , ... , ar) to an algebraic ex-
tension M (c 12) of K(b1 , ... , br), then or is extended to an automorphism
of S2.
PROOF. We can assume that aar = bi and that al , ... , ar are alge-
braically independent over K. Take a transcendence base A = {cjjj E Al of
12 containing al , ... , ar . Consider subsets S of A such that {b1 , ... , br}U
S is algebraically independent and {b1 , ... , br} n S is empty. Let So be a
2. TENSOR PRODUCTS OVER A FIELD 83
and N are often identified with 0IM and 02N, respectively. The following
assertion is clear (and we omit the proof).
THEOREM 3.2.3. If, furthermore, M, N are subrings of a ring R, then
there is a unique ring homomorphism 0 from M N to R such that O(m (9
n) = mn.
This 0 is called the natural mapping of M N to R (or, to the subring
M[N] generated by M, N).
If M, N are extension fields of a field K, then the ring of total fractions
of M K N is called the local tensor product of M and N over K and is
denoted by M xK N or simply by M x N. (Though the local tensor product
can be defined in more general way, we give here only this special definition,
because we do not treat the general case. As for the notation M x N, we
must be careful not to confuse this with the product set. There will be no
confusion if we say the ring M x N.) Assume that the fields M, N are
subfields of a field Q. If the natural mapping 0 of M N to the ring M[N]
generated by M, N in S2 is an isomorphism, then we say that M and N
are linearly disjoint over K. In this case 0 extends to an isomorphism
from M xK N to the field M(N). This extension is also called the natural
mapping.
THEOREM 3.2.4. Let K be a field, let M be a K-module, and let N be a
ring containing K. M is identified with M 1 c M K N by the canonical
2. TENSOR PRODUCTS OVER A FIELD 85
We say that M and N are free over K if the conditions in (ii) are
satisfied.
PROOF. (i) Assume that Q is a nonmaximal prime ideal of M xK N. Let
I be a maximal ideal containing Q. Choose an element m E M N that
is in I but not in Q. Then there is a subfield M' of M that is finitely
generated over K such that m E M' x K N. II = I f1 (M' XK N) and
QI = Q rl (M' XK N) are prime ideals of M xK N, and Ii # Qi because
m E II , a contradiction since M' x N is an Artin ring by Theorem 3.2.7.
(ii) It is obvious that (1) implies (2). Assume that (1) is not satisfied. Then
P contains an element f that is not a zero-divisor (cf. Theorem 3.6.12),
so we can choose a finitely generated subfield M' of M over K, such that
f E M'(9 N. Then f is not a zero-divisor and is in the kernel of the natural
mapping of M' N to N(M'). Therefore, we can assume that M is finitely
3. DERIVATIONS 87
that the equality is true for n = 0, 1 , ... , n - . Then Dfn = D(f"- ' f) _
1
fDfn-l+fn-1Df =
nfn-1Df . In the case n < 0, setting m = -n, we have
Dfn = D(1/f'n) = (fin D1 - Dfm)/ f2in = -(m fm-1Df )/f2n = nfn-1Df .
in
Etl
ai,...i,, x1
. .
xn -* >( ) !'
ai1...i , 1 xl
. . t,i
. xn
PROOF. To prove the only if part, note that f (al , ... , an) = 0 implies
D' f (a , ... , an) = 0. On the other hand, we see that D' fj (a , ... , an) _
.
1
1
To prove the if part, let f E I, and write f = > hj f (hj E K[x, , ... ,
x,]) , then
Ff = fjDhj + hjDfj + E fj((a/axl)hj)vi + hj((a/ax1) fj)vi ;
hence Ff(al, ... , an) = 0. Thus, if g(a,, ... , an) = g'(a,, ... , an) with
g, g'EK[x,,...,xn],then g - g'EI and Fg(a,,...,an)=Fg,(a, ,...I
an). Therefore, the mapping D' of K[x, , ... , xn] to L defined by
DI(g(al,...,an)) = Fg(a,, ... , an) for g(a,,...,an) E K[a,,...,an]
satisfies conditions (1) and (2) in Theorem 3.3.2, and therefore, the required
extension exists. Q.E.D.
REMARK. For elements fl , ... , f. of K(x, , ... , xn) , the (m, n)-matrix
whose (i, j) entry is (a/axj) f is called the Jacobian matrix of f , ... , f,n
and is denoted by J(f , ... , A similar definition is applied to infinitely
4n).
many elements of K(x, , ... , x,,) . On the other hand, if A is a matrix with
entries in K(x, , ... , xn) , then A(X =a) denotes the matrix obtained from A
setting xi = ai (i = 1 , ... , n). Then the condition in the theorem proved
above can be expressed as
(Df,)
J(f, ,... I fjn I ... (X =a, )
__ - I
I (X,=a,)
Therefore, we can conclude that
(i) If the rank of J(f, , ... , fin, ... )(X,-ai) is equal to n, then there exists
a unique extension of D to L.
(ii) Assume that the rank of J(f , ... , 1,,, , ... )(X;=aj) is n - r.
(1) If an extension D" of D to L exists, then the extensions of D to
L are in one-to-one correspondence with vectors of the form
c, b, , br,
+d, J+ +dr
cn b1 n brn
where ci = D"a, (i = 1, ... , n) and the r vectors (b j;) (j
, ... , r) are linearly independent solutions of
1
X1
if D*al = c .
D - > u.(a/ax;). Then D'xl = 0 and D'(K(x1 , ... , xn)) = 0, i.e., D' = 0 .
independence of these si , we see that each term of f (xl , ... , xn) must be
of degree a multiple of p for every xi Thus, there is g(xl , ... , xn) E
.
KP [xl , ... , xn] such that f (xl , ... , xn) = g(xl , ... , xn)p. Let g =
E ci 1 n
xi' xn'l . Then the element b = E(si' S' ci ... E L Kp
I n
is
not 0, because K[sl, ... , sn] KP ^' KP [xl, ... , xn]/IKp [xl , ... , xn] ,
1
0
length(L (9 Kp )/P1)
1 KP-1
Kp . On one hand, i(L/K) = length L K
00
Kp and L K Kp 00 = (L K Kp 1) KP_, Kp . On the other hand,
PI (L (& Kp-00) is the ideal generated by P1 KP_, Kp "O and is different from
101. Therefore, length L Kp > length(L (& Kp )/P1(L Kr-0)
00
length((L (9 K Kp-1 /PJ) KP-1 Kp .
5. Regular extensions
We say that an extension L of a field K is a regular extension of K if
L K K, with K the algebraic closure of K, is an integral domain (hence, a
field). An integral domain R containing K is defined to be regular over K
if the field of fractions of R is a regular extension of K. The following is
immediate from our definition of a regular extension.
THEOREM 3.5.1. If L is a regular extension of a field K, then every inter-
mediate field between L and K is a regular extension of K. Every extension
of an algebraically closed field K is a regular extension of K.
THEOREM 3.5.2. For a subfield K of a field L, the following three condi-
tions are equivalent.
(i) L is a regular extension of K .
Then, L1 x K(b2 , ... , bu) is not an integral domain, and therefore, we may
rechoose Lo to be K(b2 , ... , bu) . q(L1 x Lo) can be regarded as a field
generated by L1 and Lo .
First, assume that Lo = K(b2 , ... , bu) is not separably algebraic over K.
Then there is a nonzero derivation D E Der(L0/K). Since LI has a separat-
ing transcendence base z , ... , zt over K , z 1 , ... , zt form a separating
1
such that a. E for each i = 1 , ... , t . Then, with s such that s > n(i)
for every i , all ai are in N , which shows that Ns = N* . Thus, the validity
of the ascending chain condition is proved. Q.E.D.
THEOREM 3.6.2. If R is a Noetherian ring, then every finitely generated
module M over R is Noetherian.
PROOF. Let al , ... , a,, be a set of generators of M we use an induction
argument on r. In the case r = 1 , for a submodule N, we consider I =
{x E RIa1x E NJ. Then N = all. Since R is Noetherian, I is generated
by a finite number of elements, say, bl , ... , bt . Then N is generated by
albs, ... , alb, . (Remark. As an R-module, M = a1R R/{x E RIalx =
0} .)
In the general case, our induction hypothesis is that Ml =- i a R is
Noetherian. For a submodule N of M, set Ni = Ml n N. NI is finitely
generated because Ml is Noetherian. Consider the ideal J = {x E RIarx E
M1 + NJ. Let cl , ... , cs be a set of generators of J. For each c, , we
take an element ni of N of the form arc, + Er-1 aib1j (b1j c R). Then it
suffices to show that N is generated by n , ... , ns and Nl If n E N, then
1 .
P c Pn for some i < n , then we can disregard P1 , and we have the case of
n - 1 prime ideals. So, we assume that every P1, for i < n , is not contained
in Pn. By our induction hypothesis, there is an element a1 of I which is
not in any of the P , i < n. If a1 is not in Pn , then we may set a = a1 .
Assume that a E Pn P. If I P1 . . Pn-1 c Pn, then P1 c Pn for some i < n,
1
.
element. Q.E.D.
THEOREM 3.6.10. Every ideal I in a Noetherian ring R can be written
as the intersection of a finite number of primary ideals. Here, we regard R
100 III. TRANSCENDENTAL EXTENSIONS
N
Qi : a R = R because aN E Qi. Similarly, Q : aNR=R for j < t For .
bers in the set of prime divisors of I are called minimal prime divisors or
minimal associated prime ideals of I.
THEOREM 3.6.11. Let I be an ideal of a Noetherian ring R (I R). If
a prime ideal P contains 1, then P contains a minimal prime divisor of I.
Consequently, a minimal prime divisor is characterized by the property that it
is minimal in the set of prime ideals containing I.
PROOF. I = Q1 n . . n Qn with primary ideals Qi. Then Q1 ... Qn C
.
matrix (adij-c,j) , where alb is the Kronecker o function (i.e., al, = 1 , (5ij =
0 if i # j). Lemma 3.7.1 shows that Dbi = 0 (i = 1, ... , n) with D =
det(a8, - c.1) . Thus, DT = 0 and D = 0. If we expand D, then we
have D = an + e1 an-1 + + en with e1 E S. Therefore, a is integral over
S. Q.E.D.
COROLLARY 3.7.3. If S is a subring of a ring R sharing the identity, then
the set S' of elements of R which are integral over S forms a ring.
This S' is called the integral closure of S in R. If S' = S, we say that
S is integrally closed in R.
PROOF. If a, b E S', then S[a] is a finitely generated S-module and
S[a, b] is a finitely generated S[a]-module, which implies that the ring
S[a, b] is a finitely generated S-module. Therefore, every element of S[a, b]
is integral over S (Theorem 3.7.2). In particular, a + b and ab are integral
over S, hence, they are in S'. Q.E.D.
COROLLARY 3.7.4. If a ring R' is integral over a ring S and if a ring R"
contains R' and is integral over R', then R" is integral over S. Conse-
quently, S' in Corollary 3.7.3 is integrally closed in R.
PROOF. For a E R", there are ci E R' such that an + c1 an-' +... + c,t = 0.
Then the ring S[c1 , ... , cn , a] is a finitely generated S-module (cf. the
proof of Corollary 3.7.3). Therefore a is integral over S. Q.E.D.
If an integral domain R is integrally closed in its filed of fractions, we
say that R is an integrally closed integral domain or a normal ring. For a
ring R, the integral closure in its ring of total fractions is simply called the
integral closure of R.
LEMMA 3.7.5. Assume that I CK are integral domains such that K is
integral over I. If one of I, K is afield, then the other is also a field.
PROOF. Assume I is a field. If 0 54 a E K, then there c; E I such that
an-1
an +C1 + +Cn Since K is an integral domain, we may assume that
=0.
Therefore, K is a field.
Assume K is a field. Let(a-1)n
a be a nonzero
(a-1n-1
element of I. a-' E K and
there are c, E I such that + c1 + + cn = 0. By multiplying
by an-1 , we have a-' _ -(c1 + c2a + + cnai-1) E I. Therefore, I is a
field. Q.E.D.
LEMMA 3.7.6. If a ring R is integral over its subring S and if 0 is a
homomorphism of R to some ring, then OR is integral over OS.
PROOF. If a E R, then there are ct , such that an + c1 ai-1 + + cn = 0,
and we have
+ ... +
(Oci)(0a)"-1
0. Q.E.D.
(Oa)n +
104 III. TRANSCENDENTAL EXTENSIONS
also a field by the lemma above. This implies that 101, i.e., P = Q.
Therefore, P' lies over P. Q.E.D.
COROLLARY 3.7.8 (lying-over theorem). If a ring R is integral over its
subring S, then
(i) For each prime ideal P of S, there is a prime ideal of R lying over P.
(ii) If P1 and P2 are prime ideals of R such that P1 c P2, then
P1nSCP2nS.
The proof is straightforward by the theorem above.
COROLLARY 3.7.9 (going-up theorem). Assume that a ring R is integral
over its subring S. If P1 c P2 c c P, is an ascending chain of prime
ideals in S, then there is an ascending chain P1 c PP c c Pn of prime
ideals in R such that P,' n S = P1. (i = I, ... , n).
PROOF. We use an argument by induction on n . The case n = 1 is
covered in the corollary above. Assume that P1 , ... , P' satisfying the
requirement, exist. Then R = R/P'_ is integral over S = SIP,,_, and
1
Here, we shall define the notions of the height of an ideal and the Krull
dimension of a ring. For a prime ideal P of a ring R, we consider descending
chains of prime ideals that begin with P. If the chain of length n : P = P0 D
P1 D D Pn is the longest among them, then the height of P is defined to
be n ; otherwise, we say that the height of P is oo. For an ideal I (54 R),
the minimum of the heights of prime ideals containing. I is called the height
of I and is denoted by ht I. The Krull dimension of a ring R, denoted by
Krull dim R , is defined to be (i) the maximum of the heights of prime ideals
if it exists, (ii) 00 otherwise. Then we have the following corollary.
COROLLARY 3.7.10. If a ring R is integral over its subring S, then
KrulldimR = KrulldimS.
PROOF. Krull dim R > Krull dim S follows from Corollary 3.7.9. Con-
versely, if P' D Pi D . D P' is a descending chain of prime ideals in R,
then (PonS) D (Pi nS) D D (PnS) is a descending chain of prime ideals
in S (Corollary 3.7.8, (ii)). Therefore, Krull dim R < Krull dim S. Q.E.D.
Corollary 3.7.9 does not guarantee that if we first specify a prime ideal P"
that lies over Pn , then there is a chain P1 c c Pn such that Pn = P" .
root of a polynomial over K[Y] of the form X"(M') - (terms of lower degree
-
in X) a-I Yj . Thus, X1 is integral over K[Y], and K[X] is integral over
K[ Y] because Xl E K[YI , ... , Y , X1] (i > 2).
108 III. TRANSCENDENTAL EXTENSIONS
(ii) Consider Y = Xi + c,X1 (i > 2). By the proof of (i), we see that it
suffices to determine c1 so that, in the expansion of f as a polynomial in
XI , Y2, ... , Yn , the term Xd (d = deg f) appears with nonzero coefficient.
This coefficient is equal to fd(1 , -c2, ... , -cn) where fd(XI , ... , Xn) is
the degree d part of f . Since fd is a nonzero homogeneous form, fd (I, X2,
... , Xn) is a nonzero polynomial, and therefore, we see the existence of
c2 , ... , cn as required. Q.E.D.
THEOREM 3.8.2 (normalization theorem for a polynomial ring). If I is an
ideal of the polynomial ring K[X] = K[XI , ... , Xn] and if htl = r, then
there are elements YI , ... , Yn of K[X] satisfying the following conditions.
Here n denotes the prime subfield.
(1) K[X] is integral over K[Y] = K[YI, ... , Y ] .
(2) I n K[Y] is generated by Y1 , ... , Y .
(3) Y+j = X +t + f with f E 7r n [XI , ... , Xr] (if the characteristic p 0,
then f En[Xf,...,Xp]) for i=1,...,n-r.
PROOF. We use an induction argument on r. If r = 0, then I = 101,
and we can set Y = Xj. Assume r > 1 Choose an ideal I' CI such that
.
(ii) The general case. We can apply (i) for each prime ideal P of A and
we see that the intersection of maximal ideals containing P coincides with
P, hence D C P. Thus, D c nP = {0} = 10}. Q.E.D.
THEOREM 3.8.7. Let A be a ring generated by a finite number of elements
over a field K. If an ideal I (0 A) of A is generated r elements and if P
is a minimal prime divisor of I, then ht P < r. Consequently, ht I < r.
REMARK. This assertion holds under a weaker condition that A is a
Noetherian ring. For the proof, we need some results on so-called local rings
and we restrict ourselves to this special case.
PROOF. We use an induction argument on r. Assume first that r = 1 .
it suffices to show that htP/Po < 1 for any prime ideal P0 (c P). Thus,
considering A/Po , we can assume that A is an integral domain. Let I be f A
and let Q1, ... , QS be the minimal prime divisors of I other than P. For
each Q i , we take qi E Qi , which is not in P. Set A 1- A [ql , ... , q-1 ] .
S
Then PAI is a prime ideal by Theorem 3.6.12. If there is a prime ideal P' of
Al such that f E P' c PAI , then f c P' n A c P. By the minimality of P,
PAI is a minimal prime divisor of fAI . Similarly, if Q' is a minimal prime
divisor of fAI , then Q' n A is a minimal prime divisor of f A . Thus PA I
over K and A is integral over K[z] = K[z1 , ... , z1] (Corollary 3.8.3).
Take the least normal extension L of K(z) containing A, and let A* be
the integral closure of A in L. Let P* be a prime ideal of A* lying over
P (Theorem 3.7.7). We see that P* is a minimal prime divisor of fA* by
Corollary 3.7.8. Let q be the inseparable factor of the degree of the extension
L over K(z), and set G = G(L/K(z)) (= G(A*/K[z]), fo = (l] EC C P )"
Since K[z] is an integrally closed integral domain (cf. Exercise 3.7.7), we see
that fo c K(z) n A* = K[z]. Let Q be a minimal prime divisor of foK[z],
and let Q* be a prime ideal of A* lying over Q. Then fo E Q* implies
that c f c Q* for some a E G and f c a-1 Q* . Since the minimal prime
divisors of fA* are conjugates of P* (Theorem 3.7.11), we can assume that
P* c a_I Q* Then we have htP* < hta_IQ* = htQ* = ht(Q* n K[z]) =
.
divisor of the principal ideal generated by (fr mod P') and ht P/P' < 1
Therefore, ht P < r by Corollary 3.8.5, (ii). Q.E.D.
9. Integral closures
LEMMA 3.9.1. Let R be a Noetherian ring, and let Q be the ring of total
fractions of R. An R-submodule M of Q is finitely generated if there is a
non-zero divisor d of R such that d M C R.
REMARK. R-module like this M is often called a fractional ideal of R
under certain restrictions such as M contains a non-zero-divisor.
PROOF. M C Rd-1 , and Rd-' is a finitely generated R-module. Since
R is Noetherian, Rd- is a Noetherian module (Theorem 3.6.2). Q.E.D.
THEOREM 3.9.2. Let f (x) be a monic polynomial over an integrally closed
integral domain R, and let a be a root of f (x) (in some field containing
R) . Let R* be the integral closure of R[a] in its field of fractions, and let
f'(x) be the derivative (d/ dx) f (x). Then we have f (a)R* C R[a]. If d is
the discriminant of f (x) , then we also have dR* C R[a].
PROOF. Let the roots of f (x) be u1 = a, u2 , ... , ur (r = deg f (x)) ,
and we set gi(x) = f(x)/(x - u,) (i = 1, ... , r). Then f4(x) = > g,(x)
and f4(a) = g1(a) . Therefore, it suffices to prove the assertion under an
additional assumption that f (x) is irreducible. If a is inseparable, then
f'(a) = d = 0, and we assume that a is separable. Let R' be the smallest
Galois extension of R containing a, and we set G = G(R'/R) (note that
R' is integral over R and is integrally closed). Let H be the subgroup
corresponding to R* (namely, H is the set of elements of G which leave
elements of R* invariant). We take of E G such that Qi u 1 = ui for each
i = 1, ... , r (a1 = 1). Then G is the union of of H (i = 1 , ... , r)
(note that as = ui if a E a,H). o f (x) = f (x) , and therefore, gi (x) =
f(x)/(x - a1u1) = a g1(x) We write g1(x) = c,._
.
+ ... + co (ci E R[a]
1xr-1
,
r
b.f (a) = bg1(a) _ aibgi(a) _ (aib)(a,g1(a))
Then Corollary 3.9.4 and Theorem 3.9.3 imply that the integral closure At
of A" = K'[z1 , ... , z1] in L'(K') is finite over A". Since A" is finite over
A, A* is finite over A and A* is a Noetherian A-module. Therefore, its
submodule A' is a finitely generated A-module. Q.E.D.
REMARK. In this proof, we have to take a separating transcendence base of
L'(K') over K'. There is another way to prove this assertion. Namely, if we
take a suitable power q of the characteristic p , then L' (K') (z 11q , ... , z 1q)
is separably algebraic over K ... , zr q) Thus, we consider the integral
.
Therefore, we usually develop the theory first in the case where K is al-
gebraically closed and then in the other cases, we apply the results over an
algebraically closed field S2 containing K. For the purpose, we need some
notions as follows.
If K' is a subfield of K, we regard S" (K') as a subset of S" (K) and
points in S"(K') are called K'-rational points. In dealing with an algebraic
set A' in S" (K') , we first consider an algebraic set A in SN(K) , or in
S" (a) , with a big algebraically closed field 52 containing K, defined by the
same set of polynomials, and we regard A' as the set of K'-rational points in
A. In this sense, though the algebraic set in S" (R) defined by X + +X2l _
0 coincides as a point set with the one defined by X1 = . . = X" = 0 , we
.
distinguish these two as algebraic sets. More precisely, algebraic sets should
be interpreted not as point sets by themselves, but as pairs of point sets and
systems of equations (or ideals) which define them. For instance, though Xl
and Xi determine the same point set, we should understand that XI defines
a hyperplane and X2 defines a special quadratic hypersurface. However, in
this book, we intend to give some primitive notions on algebraic sets, and
therefore, we treat them mainly as point sets.
In the following, we fix an algebraically closed field S2 and consider points
in S" (a) . Namely, we restrict coordinates of points to elements of Q, and
we call this S2 the universal domain. To have enough freedom, 52 should
be big enough. Thus, we usually assume that the transcendence degree of S2
over its prime field is infinity.
An algebraic set V in S" (52) is said to be irreducible, or an algebraic vari-
ety, if it is defined by a prime ideal P of 52[X1, ... , X,1] . trans. deg, 52[X]/P
is called the dimension of V and is denoted by dim V.
THEOREM 3.10.3. Each algebraic set A in S" (52) is the union of a finite
number of algebraic varieties Vl , ... , Vt. If the union is irredundant, then
Vl , ... , V are uniquely determined by A (disregarding the order).
(i) A hyperplane is the one defined by a linear form E`'=1 a;Xi - b ((a1 ,
.) an) 0(0,...,0)).
(ii) A hypersurface is the one defined by a polynomial f (X1 , ... , X").
deg f is called the degree of the hypersurface.
(iii) A linear variety is the one defined by several linear forms f(X)
Ej a1jXj -b; (i = 1, ...,r).
(iv) A linear variety of dimension one is called a line and a linear variety
of dimension two is called a plane.
(v) A one-dimensional algebraic variety is called a curve and a two-
dimensional algebraic variety is called a surface.
REMARK. The terms "curve" and "surface" may be used under weaker
conditions. Namely, an algebraic set is called a curve or a surface if every
irreducible component is a curve or a surface, respectively.
REMARK. Consider the case where n is the complex number field C.
Then a line (which may be called a complex line) is two dimensional over
the real number field R ; for instance, S I (C) is the so-called Gauss-plane
(S2(R) E) (a, b) a+b E C). Therefore, it is quite hard to have
intuitive images of planes, curves, surfaces, and so on over C. Therefore,
the readers are advised to have intuitive images in S3(R) .
The following two theorems follow easily from our results in 8.
THEOREM 3.10.4. If V and V' are algebraic varieties in S" (S2) and if
V c V', then dim V < dim W. Furthermore, there are algebraic varieties U1
(i 1, ... , r - 1, with r = dim V'- dim V) such that
Ur = V' and dim Ui = dim V + i .
ci.l
E K, then K is a field of definition for V. Conversely, assume that K
is a field of definition for V. Since S2[X]/I(V) ^_' K[X]/(I(V) flK[X]) K S2
and since the mi are elements of K[X]/(I(V) n K[X]), we can consider
M Then we have Mn Al K S2 and mi c M* , if i E B .
(iii) These V are defined over the algebraic closure K of K, and for each
z c G(K/K), there is a c AutK S2 such that a I K = r and aA = A.
For the proof, we need two lemmas.
LEMMA 3.10.10. Let f1 , ... , ft be elements of a ring R and let J be an
ideal of R. If all of the elementary symmetric forms sl , ... , st of f1 , ... , f
are in V-T then the f are in v'j-.
PROOF. Since the f are roots of F(X) = X` - s1 X`- I + + (-1)'st , we
have ft-sl(f )t-I +- +(-1)tst = 0. Thus, ft c V7, and f c . Q.E.D.
LEMMA 3.10.11. Let ko be the smallest field of definition for an algebraic
variety V in S" (S2) . Then for a a C Aut K2,
aV=V iff as=a forallaeko.
PROOF. The elements c. j in Theorem 3.10.7 are uniquely determined by
V. Therefore, a V = V implies ac, j = ci j for all cj . Thus, as = a for
all a E ko, because ko is generated by c.
Conversely, if as = a for all
a E ko, then a certain system of generators of I (V) is invariant under a
and aI(V) = I(V) ; hence, aV = V. Q.E.D.
PROOF OF THEOREM 3.10.9. Let Ki be the smallest field of definition for
over K.
To prove that (i) implies (iii), suppose, for a moment, that some Kl, say
K1 , is not algebraic over K. Take a transcendence base z , ... , zt of K11
{(al,...,am,bl,...,bn)ESm+,l(1)I(al,...,a,n)EA,
(bl,...,b,t)EB}
is called the product of A, B and is denoted by A x B (A x B C S'"+" (0))
THEOREM 3.10.12. If A
CS'" p), B C S" (0) are algebraic sets defined
respectively, by ideals JA of SZ[X1 , ... , JB of '2[Yl , ... , Yn] Then
Xm] , .
fI( X('+1)
n+1
X('+1)
2n )' "' ' fr( X('+1)
11+1
X('+1)
2n
f(XU+1) ,... ,X('+1)),0,...,0)
l (tj-I)n+1 ,...,X('+1)),...,
tfn f (X('+1)
r (Ui - I)n+ tin
120 III. TRANSCENDENTAL EXTENSIONS
-
variables yuv of "degree" d. Since n d'+' > (d ` (r + 1) - n) t-1 , we have
n (t+ 1) > d ` (d t+r+ 1), and therefore, fo = 0, ,
fdt+r = 0 have a common
nontrivial solution (r1 uv) in K . Then u = Ev=0 quv zv (u = 1 , ... , n) give
a nontrivial solution in K(z) of 0 = 0. Q.E.D.
THEOREM 3.11.8. Every finite field is a C'-field.
For the proof of this theorem, it suffices to prove the following theorem
(because every polynomial equation having no constant term has a trivial
solution).
THEOREM 3.11.9. Let F be a finite field of characteristic p and set q =
#(F) Let f (XI , ... , Xn) be a polynomial of degree d in n variables Xi
. .
Thus, we have APES mi (P) = flj (UCEF Ce') Each factor of this last .
Since n > d and deg mi < d (q - 1) , at least one of these factors is of the
case (ii), and we have completed the proof. Q.E.D.
12. The theorem of Luroth
-
PROOF. g(X)y f(X) is linear in y, hence is irreducible as a polynomial
in y over K(X). Since K[X] a unique factorization domain, g(X)y -
f (X) is irreducible as a polynomial in X, y, by Corollary 1.6.7, hence is
irreducible as a polynomial in X over K(y). Q.E.D.
THEOREM 3.12.2 (theorem of Luroth). If x1 , ... , xr are algebraically in-
dependent over a field K and if L is a field such that K C L CK(x1 ) ... , x,.)
and trans. degK L = 1, then there is an element t such that L = K(t).
PROOF. In the case r = 1 , let h(X) = X" + +c,, (ci E L) be
+c1Xn-i
PROOF. Write K1 = k(y1 , ... , Yin), K2 = k(z1 , ... , zn). There are ra-
tional expressions of y, ,zi such that y. = f (z1 , ... , z,t , x) and zi =
gj(y1,...,y,n,x). Then yj = zi =
gi (f1(z , x) , ... , fin(z , x) , x) Since #(k) is infinite, there is an element
.
a of k such that the denominators of f,., gj and also those of the new ex-
pressions of y, , zi do not vanish by the substitution of a for x. We
set yi = f (z1 , ... , zn , a) , zi = gj(y1 , ... , Yin , a). Since x is tran-
scendental over k(y), we have y, = f (g1(y , a), ... , gn (y , a), a), namely,
y; = f (z; , ... , zn , a). Similarly, zj = gj(Y1 , ... , y,,, , a). Thus, K1 =
k(y1 , ... , Yin) = k(zi , ... , z'1) and K2 = k(z1, ... , zn) = k(yi , ... , y,,,) .
Hence, trans. degk K1 =trans. degk K2. Further, if F (Y1 , ... , Y,) = 0 is
a relation satisfied by y1 , ... , Yin , then F(f1(z , x) , ... , f;,,(z , x)) = 0 .
Since x is transcendental over k (z , ... , z,) , we have F (y' , ... ,
1
0 .
REMARK. We can choose x so that V[x] c Vx. In this case our proof
below shows 1 can be chosen as the residue class of the element t as follows
(in the following, - denotes the residue class modulo Mx) .
For each element h of K(x), d*(h) is defined as the maximum of the
degrees of the numerator and the denominator of the irreducible fractional
expression of h . Then t is an element of T = j h E Vx h is transcendental
over k} such that d*(t) < d*(h) (Vh E T).
PROOF. Let t be as in the remark above and let t = f(x)/g(x) be the
irreducible fractional expression of t. First, we reduce to the case where
deg f > deg g without changing k (t) . If deg f < deg g , we take t- I instead
of t. Assume that deg f = deg g . Then f = ag + r (a E K, deg r <
deg g) . (i) If a E V, we take (t - a)-1 (= g/r). (ii) Otherwise, we take
(a- I - t- I) -1(= at/(t - a) = a f /r) Thus, we have the case deg f > deg g.
.
a, , bi are algebraic over k , because d * (a;) , d* (bj) are less than n Since
.
APPENDIX. A THEOREM ON VALUATION RINGS AND ITS APPLICATIONS 125
(R0)pI and the similar discrete valuation ring of K(x2 , , ... , Xr) , and so on.
Let v be an extension of vo to L(t) , and let v' be the restriction of
v to L. Then v' is of rank r because the residue class field of vo is K
and since x; E L. Since trans. degk L(t) = trans. degk K + r, and since
trans. degk L > trans. degk (the residue class field of v') + rank v' (cf. Exer-
cise 4.9.6), the residue class field of v (which is a finite algebraic extension
of K ; cf. Theorem 4.9.6) is not algebraic over the residue class field of v'.
Therefore, the residue class field of v is of the form k'(?) with a field k'
126 III. TRANSCENDENTAL EXTENSIONS
containing the residue class field of v' and with t transcendental over k' ,
a contradiction. Q.E.D.
THEOREM 3.A.4. Let L be a finitely generated field over a field k. If
intermediate fields K1, K2 are antirational over k , then the field K1(K2) is
also antirational over k.
PROOF. Suppose for a moment that K3 = K1(K2) is not antirational over
k. Then there is a finite algebraic extension K3 of K3 such that K3 = k'(t)
with a field k' containing k and a transcendental element t over k'. By
the theorem above, Kl CK3 = k'(t) implies K, c k', which shows that
K3 is not algebraic over K3 = K1(K2) , a contradiction. Thus, K1(K2) is
antirational over k. Q.E.D.
COROLLARY 3.A.5. In L above, there is the largest subfield among those
which are antirational over k.
It is difficult to see if a given field extension is antirational or not. We add
here some known cases without proof.
(1) When k is algebraically closed, L is finitely generated over k and
trans. degk L = 1 , we have that L is not antirational over k if L = k (x)
for some x.
(2) When L is a regular extension of a field k and trans. degk L = 1 , we
have that L is not antirational over k if either L = k(x) or L = k(x, y)
with x, y such that axe + bye = 1 (a, b E k).
(3) When k is an algebraically closed field of characteristic 0 and
trans. degk L = 2, we have that L is not antirational over k if L = K(x)
with a field K such that trans. degk K = 1 .
Exercises
1
1. Let M, N, M' , N' be modules over a field K, and let 0: M -> M',
v: N -+ N' be K-homomorphisms. Then we define the tensor product
of q and y/, which is denoted by q V, as a mapping of M (9 N to
M' N' determined by
V) mi n _ 0(m,) yr(n1) .
statements.
(i) Ro is a subring of R.
(ii) M = 1:1 R, is an ideal of R. If M is generated by a finite
number of homogeneous elements f1 , ... , fN, then R is generated
by f1 , ... , fN over Ro as a ring.
130 III. TRANSCENDENTAL EXTENSIONS
n
Y+, = >2 c11X3 (c1j E K) . (3')
j=1
10
closed, then, though Al nA2 is an algebraic set, the equality I(A1 nA2) _
I -+I can be false.
2. Consider S" (S2) with an algebraically closed field Q. Let V be an al-
gebraic variety in S' (Q) of dimension r > 0. Prove that if H is a
hypersurface in S" (S2) and if H n V (the empty set), then the dimen-
sion of each irreducible component of H n V is r - 1 or more.
3. Prove that, in the affine space S" (K) over an arbitrary field K , the
minimum condition on algebraic sets holds good.
4. Let A be an algebraic set in the affine space S" (S2) over an algebraically
closed field K2, and let K be a subfield of 0. We say that A contains
sufficiently many K-rational points, if A is the smallest algebraic set
containing the set of K-rational points in A .
Examine whether the following quadratic hypersurfaces in S" (C) have
sufficiently many R-rational points, where C and R are the complex
number field and the real number field, respectively.
(i) c1X1+ +CrXr = Cr+1X+1 + +CSXs +1 (C1 E R, ci > 0, s > 1 ,
0<r<s<n),
(ii) 1X1
+...+CrXr2 = 2 2
Cr+1X+1+...+Cs-1Xs I+CsXs (C! E R, c > 0,
0<r<s<n).
5. Let A be an algebraic set in S" (S2) with an algebraically closed field Q.
Assume that A is normally algebraic over a subfield K of Q. Prove
that if Vl , V2 are.components of A which are conjugate over K, then
there is a Q E G(K/K) (K being the algebraic closure of K) , such
that, under an extension a' of a to an element of AutK 92, we have
a' Vl = V2. Prove also that u' Vl = V2 for any extension a' of a.
6. Let A and V be an algebraic set and an algebraic variety, respectively,
in S" (S2) . Prove that if V is contained in A , then V is contained in
an irreducible component of A.
7. Prove that if A, B are algebraic varieties in S"' (0), S"(), respec-
tively, then dim A x B = dim A + dim B.
11
1. Prove that every subfield of the real number field does not satisfy any
Cl-condition.
2. For each i = 1, 2, ... , give an example of a field K such that K is
not a Cl-field, and such that there is a finite algebraic extension which is
a C!-field but not a C1_ 1-field, for each i = 1 , 2, ... .
12
Valuations
1. Multiplicative valuations
If a mapping v of a field K to the set R of real numbers satisfies three
conditions (1)-(3) below, then v is called a multiplicative valuation, or sim-
ply, a valuation of K. {vala E K, a 0 0} forms a multiplicative group
consisting only of positive real numbers and is called the value group of v .
135
136 IV. VALUATIONS
b' , c' similar to b, c above, then p"c/b = p'"c'/b'. Assume for a moment
that n > m. Then bc' = p cb' E pR, which contradicts pR being a
prime ideal. Therefore m = n.
(ii) Next we check the conditions for a valuation are satisfied. It is clear
that (1) and (2) are satisfied. Assume that va = e-" < vb = e-'" . Then
a = p"c/b , b = p'"c'/b' (b, c, b', c' are not divisible by p) . Since e > 1
we have n > m and a + b = p'" (p"-'" cd' + d c') /d d' . Therefore, a + b =
prC"/b" with r > m and v(a + b) = e-r < e-'" = vb. Q.E.D.
The valuation v obtained above is called the p-adic valuation.
In Example 3 above, to change the value of e is analogous to changing
the unit of length in a scale; hence, is not essential. To clarify the situation,
we define the notion of an equivalence as follows.
Two valuations v1 , v2 of a field K are equivalent to each other if there
is a positive real number t such that v2 = vi (this equality means that
v2a = (vta)' for every a E K). In Example 3, we obtain another valuation
V' , by changing e to another e' (> 1) Then e' = e' with t = loge e' and
.
PROOF. It is obvious that both (i) and (ii) are necessary conditions. The
sufficiency of (i) follows from that of (ii), and therefore, it is enough to show
the sufficiency of (ii). Take an element a 0 0 such that va < 1 and set
e = va, e' = v'a , t = loge, e. Then we consider v" instead of v'. Thus,
we have only to show that v = v' assuming that va = v'a < 1 . Let b be
an arbitrary nonzero element of K.
1. MULTIPLICATIVE VALUATIONS 137
(1) The case vb < 1 . Consider I = {(r, s)l(vb)s < (va)r, (vb)s > v(a)'.+1
(va)r/s.
with natural numbers r, s} . Then vb <
(va)(r+1)/s <
Therefore,
vb = lims , er/S . On the other hand, since v(ar+1) < v(bs) < v (d), we
have v(bs/ar) < 1 and v'(bs) < v'(ar) by (ii). Similarly, we have v'(ai+2) <
v'(bs) , because v(ar+2) < v(ar+1) < v(bs). Thus, < v'b < (v'a)('+2)/s
= vb.
er/s
(2) The case vb > 1. v(b-I) < 1 and v'(b-I) = v(b-I) by (1). There-
fore, v'b = vb.
(3) The case vb = 1. v(ab) < 1 and v'(ab) = v(ab) by (1), which
implies (v'b)(v'a) = (vb)(va), and therefore, v'b = vb. Q.E.D.
Hereafter in this section, v denotes a valuation of a field K and C
denotes the constant in condition (3).
LEMMA 4.1.3. If a1 , ... , an E K, vai < e E R (the real number field),
and n < 2s with s a natural number, then v(al + + an) < Cse.
PROOF. We use an induction argument on s. If s = 1 , then the assertion
is obvious. Assume that s > 1 . Set m = 2s- I . Then v (al + + a,n) <
Cs-1 e, v (a,n+l + + an) < CS-1 e , and therefore, v (al + + an) < C
CS-le = Cse. Q.E.D.
LEMMA 4.1.4. Assume that C < 2. Then v n < n for every natural number
n (where n in vn means n-times of the identity). More generally, if vai < e
for every i = 1, ... , n, then v (a 1 + + an) < ne.
PROOF. It suffices to prove the latter half, because the former half follows
from the case where all the ai are 1. For each natural number t, we take
a natural number s(t) such that 2s(t)-1 < n' < 2S(') Then (a1 + +
is the sum of n` monomials in al , ... ) an , and each monomial is of the
form a"-..an (E ti =t). v (a;' an-) < e` and Lemma 4.1.3 implies that
v(a1 + + an)' = v((aI + + an)') < 2S(')e' . Therefore, v(aI + + an) <
2s(t)/te
and v(aI + + an) < limt,. 2S(')1'e = e = ne. Q.E.D. 2"92n
v(a + b) < "fn + 1 d . Therefore, v(a + b) < d , which shows (i). Thus, (i),
(ii), and (iii) are equivalent to each other. Since condition (iii) is a condition
only on the prime field, we see that (iv) is also equivalent to the conditions
above. Q.E.D.
or 0. Therefore, (vn)p-1
< 1 and vn < 1 . Q.E.D.
vm< 1-tu)+1
Note that limn t(u)/u = log,, in. Since t(u)/u < log,, in, we have
,
11--+00
" t(u) + 1 = lim " u V in + u- = 1 Thus, we have
.
Q.E.D.
COROLLARY 4.2.3. As for a valuation v of a field, v is non-Archimedean
iff there is a natural number n > 1 such that v n < 1 .
3. Topology
We say that a topology is defined on a set S if a family of subsets called
open sets are given so that the family S2 of open sets satisfies the following
conditions.
(i) SES2.
(ii) The empty set E S2 .
REMARK. Readers who are familiar with open sets and closed sets in a
Euclidean space are advised to compare topological spaces with Euclidean
spaces.
Let M be a subset or a point of a topological space S. By an open neigh-
borhood, or simply, a neighborhood, of M, we mean an open set containing
M. (In some literature, a neighborhood of M means a subset containing
an open neighborhood of M. But, in this book, we consider open neighbor-
hoods only.) The family v(P) of all neighborhoods of a point P is called
the system of neighborhoods of P. The following properties are obvious:
(1) S E v(P)
(2) If UEV(P),then PEU.
(3) If Ul , U2 E v(P) , then U1 n U2 E v(P) ,
(4) If Ui E v(PP) (i E A), then UEA Ui E v(F) (for all j c A).
(5) If U E v(P) and Q E U, then U E v(Q).
THEOREM 4.3.1. The conditions (1)-(5) above.form a (necessary and) suf-
ficient condition for systems of neighborhoods. That is, if a family v(P) of
3. TOPOLOGY 141
called the distance between P and F2. We shall show that d (P) > 0. As-
sume for a moment that d (P) = 0 . Then there is a sequence {d } such that
d,, E D(P) and 1im,. d = 0. Take Q, E F2 such that d = p(P, Q) .
3. TOPOLOGY 143
Then for any positive number e, these points Q,, are in the 8-neighborhood
UE(P) for sufficiently large n. This means that every neighborhood of P
meets F2, which contradicts the assumption that P is not in the closed set
F2. Therefore, d (P) > 0.
Set U1 = UPEF, Ud'(P)(P) with d'(P) = d(P)/2. In the same manner,
we take U2 for F2. Then U1 , U2 are neighborhoods of F1 , F2, respec-
tively. We shall show that Ul , U2 has no common point. Assume that
Q E U1 n U2. Then there are Pi E F (i = 1, 2), p(Pi, Q) < d(Pi)12,
and therefore p(P1, P2) < p(P1, Q) + p(Q, P2) < (d (P1) + d (P2))12 <
(p(P1, P2) + p(P1, P2))/2 = p(P1, P2), a contradiction. Q.E.D.
In a topological space S, a family Q0 of subsets is called a subbase of
open sets if the family of open sets of S is the smallest set containing 920
and satisfying the conditions (i)-(iv) on the family of open sets. Note that
if S is a set and if S20 is a family of subsets of S, then there is a unique
topology on S which has 00 as a subbase of open sets.
A mapping f of a topological space S1 to a topological space S2 is said
to be continuous if for any given P E S1 and any neighborhood U' of f P ,
there is a neighborhood U of P with f U C U'.
PROOF. Ul x x Un = Ii (S1 x
I
x Si_ 1 x Ui X Si+1 x . . . X Sn) , and
Ul x x Un is an open set. Therefore the union of subsets of this form is
an open set. Conversely, we consider the family 0 of open sets of S* . The
subbase of open sets of the topological space S* is the family E20 consisting
of subsets of the form Sl x x Si_1 X Ui X Si+1 x . . X Sn . Hence, if
.
S2' is the family of subsets that are expressed as the union of sets of the
form U1 x x Un with open sets Ui, then S20 C Q'. On the other hand,
if U, V E Q', then U, V are expressed as U = Ui U1 i x x Uni, V =
U j V, j x x Vn j with open sets Uki , Vk j in Sk . Thus,
U n V= U(U1i x ... X Uni n Vl j x ... x Vl j)
ij
=U n(S1x...xSk-1xUkixSk+lx...XSFl)
k
Ulinvj)x...X(UninVnj)ES2'.
U(
Consequently, S2' contains S20 and satisfies conditions (i)-(iv) on open
sets. Therefore, S2' = K2. Q.E.D.
For a given subset X in a topological space S, the intersection X of all
closed sets containing X is the smallest closed set that contains X and is
called the closure of X. If the closure of X coincides with S, then we say
that X is dense in S.
THEOREM 4.3.6. Under these circumstances, a point P E (the closure X
of X) iff every neighborhood U(P) of P meets X.
PROOF. P X if there is a closed set F with P V F D X. This, in
turn, is equivalent to the condition that there is a neighborhood U(P) of P
such that U(P) n X is empty. Q.E.D.
Consider the case where two topologies p1 , p2 are defined on the same
set S. We denote by Si (i = 1, 2) the topological space S with topology
pi . If the identity mapping f : x -* x gives a continuous mapping of SI to
S2, (or, open sets under p2 are also open sets under pl ,) then we say that
the topology p1 is stronger than the topology p2 (or, precisely speaking, we
should say "stronger or equal") and that p2 is weaker than p1 The strongest
.
4. TOPOLOGICAL GROUPS AND TOPOLOGICAL FIELDS 145
topology on S is the one such that every subset of S is an open set. This
topology is called the discrete topology. The weakest topology is such that
the only open sets are S and the empty set. Then, obviously, S is not even
a To-space (provided that #(S) > 1) ; hence, this topology is not important.
Among further important notions on topological spaces, we mention only
those of compactness and completeness. As for completeness, we shall study
it later in 5. Here we shall define the notion of compactness.
A topological space S is said to be compact if S satisfies the conditions
in the following theorem.
THEOREM 4.3.7. In a topological space S, the following two conditions are
equivalent.
(i) If U1 (i E M) are open sets and if U1 UI = S, then there are a finite
number of elements i , ... , i,1 of M such that S = U;-1 U,
1 .
(ii) If Fl (i E M) are closed sets and if, for any finite number of elements
nn=1
i1 , ... , in of M, the intersection FJ is not empty, then nIEM F is not
empty.
REMARK. In some literature, compactness is defined with the additional
condition that the space should be a Hausdorff space.
PROOF. To prove that (i) implies (ii), set V = S - F If U; V = S, then
.
Q.E.D.
REMARK. The converse of (i) above is false. For instance, consider the
additive group G of rational integers with weakest topology among those
satisfing the T1-condition. Namely, a subset U is an open set if either U
is empty or the complement of U is a finite set. Then rg(= 1g) maps the set
of open sets onto itself, and therefore, rg is a homeomorphism. But, for any
two neighborhoods U1 , U2 of 0, we have that U1 + U2 = G, and therefore,
the addition is not continuous.
(Another proof. Obviously, this G is not a T2-space. If the addition is
continuous, then we see easily that G is a topological group. But, as will be
shown in the latter half of Corollary 4.4.2 below, every topological group is
a T2-space.)
COROLLARY 4.4.2. Let G be a topological group.
(i) For each a E G, we denote by o(a) the family of neighborhoods of a.
Then v(a) = v(1)a = ao(1), where v(1)a = {NaIN E v(l)} and similarlyfor
av(1).
(ii) G is a T2-space (a Hausdorff space).
REMARK. It is known that every topological group is a regular space (and,
furthermore, a completely regular space).
PROOF. (i) is obvious because ra , la are homeomorphisms.
(ii) If a, b E G with a b, then for some U E v(1) , a 0 b U or
b 0 a U (because of To-space), and hence for some V E V(1), VV' C_ U
(by continuity). Suppose that aVflbV is not empty. Then ay = bz for some
y-1 -1
y, z E V and therefore a = b z EbVV c W. Similarly, b E a U.
These contradict our choice of U. Therefore, G is a T2-space. Q.E.D.
Let G be a topological group. A family v of neighborhoods of 1 is called a
fundamental system of neighborhoods of the identity if it satisfies the condition
that if an open set U contains 1 , then V C U for some V E v V.
It is obvious that v(1) , the family of neighborhoods of 1 , is a fundamental
system of neighborhoods of the identity.
4. TOPOLOGICAL GROUPS AND TOPOLOGICAL FIELDS 147
PROOF. Consider the first half. (i) is obvious and (ii) follows from the
T1-condition (in view of Corollary 4.4.2, (ii)). (iii) follows since U1 n U2 is
an open set containing 1. (iv) follows from continuity, namely, considering
the case a = b = 1 in the definition, there are neighborhoods U1 , U2 of 1
such that U1 U2 C_ U , and for some V E v , we have V c U1 n U2. (v) follows
similarly from the fact that the mapping 0 (qa = a- I) is a homeomorphism.
(vi) follows from the fact that that a Ub is an open set containing 1.
To prove the latter half, let 12 be the family of subsets which are written
as the union of the members of { a U j a E G, U E V j . Similarly, we define S2'
by using { Ua I a E G, U E 01. If X E n', b E X, then for some Ua E Q',
we have b E Ua C X, and hence, 1 E b-1 Ua. Thus, there is a V E v with
V c b-1 Ua (by (iv)), so bV c Ua. Hence, if we consider the union of all
b V obtained for each b E X, then the union coincides with X, and X E 0 .
Thus, we see that S2' C 92. Similarly, we have S2 c_ S2' , and SZ = S2' If .
does not change. Therefore, the topology does not change. Q.E.D.
THEOREM 4.4.7. Let K and v be as above. If K' is a subfield of K and
if v' is the restriction of v to K', then the topological field K' defined by v'
is a subspace of K.
5. COMPLETIONS 149
PROOF. (i) is clear. (ii) For each natural number n , let N(n) be such that
m, m' > N(n) implies p(P,n, P,n') < 2-n , and we take natural numbers
e1 < e2 < < en < . so that en > N(n) for each n. Set Qn = Pe Then
. . .
PROOF. First, we prove this under the assumption that v satisfies the
triangle inequality. Let S be the set of Cauchy sequences in K and introduce
a multiplication and addition in S by
{an}{bn} = {anbn} , {a11} + {bn} = {an + bn}.
Then S is a ring in which 0, 0, ... , 0, ... and 1 , 1 , ... , , ... are 0
1
and 1 , respectively. (As is easily seen, the sum and the product of two
Cauchy sequences are again Cauchy sequences.) Let N be the set of Cauchy
sequences that converge to 0. Members of N are called null sequences. N is
an ideal of S. Furthermore, {an} is equivalent to {bn} if E N.
Therefore, S/N is the set of equivalence classes of Cauchy sequences in K.
We set K* = S/N and identify each element a of K with the class of the
sequences a, a, ... , a, .... Then the mapping a - a, a, ... , a ... is an
isomorphism from K into S, and therefore, the mapping a --> the class of
a, ... , a... is an isomorphism from K into K* . Note that if {an } is a
Cauchy sequence in K, then {v (an) } is a Cauchy sequence in the set of real
numbers, by the triangle inequality I v (an) - v (am) l < v (an - am) .
(1) To see that K* is a field, consider 0 0 a* E K* . Then a* is the
class of a Cauchy sequence {an} , which is not a null sequence. Thus,
v (an) > a for some 8 > 0 (for sufficiently large n) . Here, considering
a suitable equivalent sequence, we may assume that v (an) > 6. Then, since
p(an I , am I) = v ((am - an )an I am I) < v (am - an) a2 , we see that the sequence
{an is a Cauchy sequence, and its class is the inverse of a* . Thus, K* is
a field.
(2) We define a mapping v* of K* to be the set of real numbers as
follows. If a* is the class of {an } , then we set v * (a*) = lim v (a,1) This .
positive number e, there is a number N = N(e) such that for all m, n > N,
v(an - am) < e. Now, v*(a* - aj) = lim v(an - aj) <e, whenever j > N.
Thus a* = lim an .
{an } is a Cauchy sequence, given any e > 0, there is an N = N(e) such that
-
v*(a* a*) < e for all m, n > N. This implies that if t is sufficiently large,
then v (amt - ant) < 8. Since the sequences {ajn } are regular, v (aj,n -
2-n for all m > n , and we have, for sufficiently large t,
v(amm - anti) = v(a,n,n - amt + amt - ant + ant - anti)
v(amm - ant) + v(ant - anti)
< 2-m + 2-n + v(a,nt - ant).
Therefore, the sequence {anti} is a Cauchy sequence. Let b* be its class.
Then b* = limann and v*(a*-b*) = lim n-+oo v(a .n -ann) = limn-.oo v(ajn -
aii + aii - anti) < 2-j + limn,, v(aii - anti) . Thus, if j -> oo , then v*(a* -
b*) -> 0 and b* is the limit of {an} I.
(5) To prove uniqueness; let K, v be another pair of completions of K,
v , respectively. Note first that if {an } is a Cauchy sequence in K , then
{v (an) } is a Cauchy sequence, and therefore, v (lim an) must coincide with
lim v (an) . Consider the mapping 0 from S to K such that q{ati } = (lim an
in K). Since K is dense in K, we see that q5S = K . O{an } = 0 if
0= (limati in X), that is, {an } E N, and therefore, K ^ K*. By what was
noted above, v and v* are compatible with the isomorphism.
To complete our proof, we consider the case where v does not satisfy the
triangle inequality. There is a valuation vo , which is equivalent to v that
satisfies the triangle inequality. Since vo = v' with t a positive number,
we see that the completion of K with respect to vo is the completion of K
with respect to v. Q.E.D.
THEOREM 4.5.3. Let K be a subfield of a field L with a multiplicative
valuation v', and let v be the restriction of v' to K. Assume that L is
complete with respect to v'. Then the closure K* of K in L (as a topological
space) is the completion of K.
PROOF. Each element of K* is obtained as the limit (in L) of a Cauchy
sequence in K, and we see that K* is a field by our proof of Theorem 4.5.2.
Since K is a subspace of the topological space K* (Theorem 4.4.7), we see
that K* is the completion of K. Q.E.D.
As for the value groups of the completions of valuations, there is a clear
difference between those of Archimedean valuations and non-Archimedean
valuations. Namely,
THEOREM 4.5.4. Let K, v, K*, v* be as in Theorem 4.5.2. Then
(i) If v is non-Archimedean, then, for any Cauchy sequence {an } that is
not a null sequence, limti--+00 v(an) coincides with v(at) for sufficiently large
j. Consequently, the value group of v* coincides with that of v.
(ii) If v is Archimedean, then the value group of v* coincides with the
multiplicative group of positive real numbers.
152 IV. VALUATIONS
PROOF. (i) 6 = lim v (an) is different from 0. Take r such that 8/2 >
e > 0. Then there is an N such that v (an - a,n) < E for all m , n > N N.
Then v (an) > 6 - e > e and v (an) = v (an - a,n + am) = v (an) , because
v (an - a,) < v (am) . Thus, v (an) is constant for large n , and 6 = v (an) .
(ii) K contains the rational number field Q, and the restriction of v
to Q is equivalent to the absolute value I (Theorem 4.2.1). Therefore,
I
v(/3 - c)8"-I , and therefore, v(/3 - c"//3"-1) > v(/3 _C). It follows that
v/3 + > v(/3 - c) = f(b - c). Since vc < v/3 , we have
v(c")/v(/3"-I)
Nz = 1 and v' z < 1 . Hence, considering elements of the form z3 for many
j, we have a sequence of elements zi = al + bi such that N(zi) = 1
and v' (zi) < 2-' (i = 1, 2, ...). Taking zi if necessary, we may
assume that v (ai) > v (bi) . Since v' (zi (ai - bi)) = 1 , we have that
2` < v'(ai - bid) < C v'(ai) , where C is the constant in condition (3)
of the definition of a valuation. Therefore, v (ai) = v' (ai) tends to oo when
i tends to oo. Set yi = zi/ai (= 1 + (bi/ai) ) . Then, since N(zi) = 1 ,
coincides with v .
Furthermore, it is easy to see that N(zz') = (Nz) (Nz') . Hence, v'(zz') _
(v'z)(v'z') for all z, z' E K(im). Nz = 0 if z = 0, and consequently,
v' (z) = 0 if z = 0. Thus, it remains only to show the existence of C such
that v'(1 + z) < C for v'z < 1 . Assume for a moment that there is no such
C. Then there are z1 E K(im) , v'(z1) < 1 , v'(1 + z1) > 2' We write
.
Z (a1, b1 E K) . Then
v(N(z1)) = v(a2 + b2) < 1, v(N(1 + z1)) = v((1 + a1)2 + b2) >
221.
Since v satisfies the triangle inequality, we have v(1 + 2a1) > 22i - 1
Thus, v (a1) > - 2 > 22(1-1) Set y1 = z, /a1. Then {y, } converges to
221-1 .
additive valuation, we can use a more general one than the additive group of
real numbers R ; a simple example is the n-ple product R(") = R x xR
of R with lexicographical order.
One aspect of the importance of additive valuations is its applications to
the theory of commutative rings. But, we do not go in this direction in this
book.
If an additive valuation w of a field K is given, then R. = j a E K j w a >
0} forms a ring, which we call the valuation ring of w. M. = {a E Kjwa >
0} forms an ideal of Rw . Obviously, a (E Rw) is not invertible in R. if
wa > 0. This shows that Mw is the set of noninvertible elements in R. ;
hence, Mw is the unique maximal ideal of Rw and is called the valuation
ideal of w . R. /Mw is called the residue class field of w , or, of Rw The
.
PROOF. As for the first assertion, we saw this already, except for the result
that w is of rank 1 . Let a be an element such that .0 0 va < 1 Then a
.
I(e(n, b)/n) - (e(m, b)/m)I < 11n, limn-,O e(n , b)/n exists and is well
determined by the value wb. We denote the limit by u(b).
We define a mapping 0 of G to R by O(wb) = u(b). We shall prove that
0 is an order-preserving isomorphism from G into R. First, let y E G.
Then there are b, c E K such that /3 = wb, y = wc, and /3 + y = w (bc) .
Now bn ae(n'b)R, bn E ae(n,b)-1R, Cn ae(n'c)R, c" E ae(n'c)-1R.
ae(n , b)+e(n , c)R b)+e(n , c)-2R
Thus (bc)n , (bC)n c ae(n , , which implies that
have P n S = P. The last equality follows from the fact that prime ideals of
R are prime ideals of S (Corollary 4.7.4). Q.E.D.
Here we give the definition of a place of a field and state the relation
between the notion and valuation rings. A place 0 of a field K is defined
as follows. Suppose we have a field L, an element denoted by oo, and a
mapping 0 of K to L U {oo} satisfying the following conditions.
(i) a, b E K, q5a, Ob oo implies d(ab) = (qa) (qb) , q(a+b) = qa+cbb .
(ii) qa 0, qSb = oo implies q(ab) = oo.
(iii) Oa # oo, Ob = oo implies 0(a + b) = oo.
(iv) qa = oo implies 0(a-') = 0.
In other words, we define operations on oo and elements a' of L by
a' + oo = oo ; a' oo = oo for a' 0; oo-1 =0 (0. oo , oo oo are
8. APPROXIMATION THEOREMS 159
any given elements a , ... , an of K and for any positive numbers e , ... , en
1 1
that vn(c1) > 1 , v2(c2) > 1 . If vn(c1) = v2(c2) = 1 , then we can take
clc2 as dI . So, we assume that v2(c2) > 1 , v2(c2) > v11(cI) (because of
symmetry between 2 and n). Taking some powers of cI , c2 , if necessary,
we can assume that vI (c1) > 2, v(c1) < 2-' for j = 2, ... , n - 1 , and
vI (c2) > 2, v j(c2) < 2-' for j = 3, ... , n , and furthermore, v2(c2) > 2,
vn (cI c2) < 2-' Then d, = cI c2 (1 + c2) -' is the required element (Lemma
.
4.8.2).
Next, we show that there are ei E K (i = 1 , ... , n), vi(ei - 1 ) <8 i ,
v j (ei) < e j (i j) . Indeed, for each i , we take a sufficiently large real
number t so that vi (e`.) > Ei ' + 1 , v j (d,) < 2e1, then e, = d `. (1 + d')-
is the required element (Lemma 4.8.2).
Finally, we take positive numbers Ei such that Enj=1 v.(aj)e < Ei , and
I
which are of rank 2 and having a common prime ideal P 0 101. Then
(RI )P = (R2)P, RI /P R2/P . In this case, if we take aI E P and a2 P,
then we cannot have an element b such that the elements b-ai are "near" to
0. We shall give an approximation theorem applicable to additive valuations.
For this purpose, we give some preliminary results.
LEMMA 4.8.3. Let RI , ... R11 be valuation rings of a field K such that
,
(1 - as)/(1 - a).
(4) In the case v (a - 1) > 0, these elements are in R. , if s is not a
multiple of the characteristic of the residue class field of R..
(5) In the case v (a - 1 ) = 0, v (a` - 1) > 0 (for some t > 2), let m be
the smallest among such t. Then other t are multiples of m. Therefore,
these elements are in R; if s is not a multiple of m .
Thus, the condition that these two elements are in all the R, is obtained
in the form that s (> 2) is not a multiple of certain special natural numbers.
Therefore, such s exists and the assertion is proved. Q.E.D.
THEOREM 4.8.4 (Independence theorem of additive valuations). Let
R1 , ... , Rn be valuation rings of a field K such that R. Ri if i 0 j I.
Let Pi be the maximal ideal of R, and set D= (l; R. , M1 = P. n D. Then
we have
(i) DM = RI .
(ii) The maximal ideals of D are exactly M1 , ... , M1.
PROOF. DM C Ri because D C Ri . For each a E R., there is an s (> 2)
with 1/(l + a + + as-'), a/(1 + a + + as-') E D by the lemma above.
Since a ER1, 1 ERl, and therefore, 1/(l
is an invertible element in R. , and hence, is not in M1. Thus a = (a/(1 +
a + + as-')/(1 /(1 + a + + as-') E DM , which implies (i). DMi OD ,
Mi
for i 54 j, which implies that, for i 54 j, M1 Mj in view of the condition
Rj % R.. Therefore, to complete the proof of (ii), it suffices to show that
if M is a maximal ideal of D, then M C M, for some i. Assume for a
moment that M is not contained in any Mi. Then there is a b c M with
b 0 M by Lemma 1.4.5. Since b is not in any Mi, vl (b) = 0 for every
valuation vi defined by R.. Thus b-' E R, for every i and b c D. Thus
b is invertible in D, which contradicts b E M. Q.E.D.
COROLLARY 4.8.5. Under these circumstances, if Q, is an ideal of each Ri
such that \ % Rj f o r any j i, then we have D/(n! Ql) ' R /Q1 ... ED
i
Rnl Qn .
PROOF. Set Ql = Qi n D. Assume that a maximal ideal M. contains
Ql + Q (i 54 j). Since D Q' DM, = Ri, we have D, = (R1) , .
other hand, we have shown above that M;/Q' is the unique maximal ideal
of D/Q'. Consequently, the class of every element of D which is not in
Mi is invertible in D/Ql. Thus, D/Q' = DM,/Q'DM, = Ri/Q'Ri = Ri/Qi.
Q.E.D.
From this result, we obtain an approximation theorem.
THEOREM 4.8.6. Let vI , ... , vn be additive valuations of a field K whose
valuation rings are RI , ... , Rn , respectively. Assume that R. Z Rj for i 34 j.
Let Qi.l
be the largest member among common prime ideals of Ri and Rj
(the existence will be shown in the proof). Let Qi be the largest among
Qil , - -- , Qi,i-1 , Qi,i+l , "' , Qin . If al , ... , an ; bI, ... ,
b,1 are elements
of K such that bi E Ri , bi Qi, and there are ti with vi(b'iai) > 0, then
there is an element c of K such that vi (c - ai) > vi (bi) for every i.
PROOF. The existence of QiJ follows from the facts that in a valuation
ring, the family of ideals forms a linearly ordered set and that if I (j E A)
form a family of prime ideals of a ring which is linearly ordered, then the
union of all I is a prime ideal.
The assertion becomes stronger if we change bi to those with bigger values
under vi . Therefore, we may assume that the bi are not invertible in Ri .
(i) In the case where all ai are in Ri , set D= n Ri , and apply the corol-
lary above to D/(nb?Ri) Then D/((l b?Ri) '" RI/biRI ... Rn/bnRn ,
.
and we have a representative c E D of > (ai mod b2Ri) (in the right-hand
side). Then c - ai E b2Ri and vi (c - ai) > vi (b?) > vi (bi) .
(ii) In the general case, apply the case (i) above to the case where both ai
and bi are b`.' , and we see that there is a d E D , vi (d - b;') > vi (b'i) . Then
d # 0, and vi (d) = vi (b;') . Set ai =aid , bl =bid , and apply (i) to ai , b; .
Then there is an e E D such that vi (e - aid) > vi (bid) . Consequently,
c = e/d is the required element. Q.E.D.
9. Prolongations of a valuation
In this section, we shall prove the following and some related theorems.
THEOREM 4.9.1. Let L be a field algebraic over its subfield K. If w is
an additive or a multiplicative valuation of K, then there is a valuation w'
of L which is a prolongation of w. All such w' are obtained as w, defined
below.
We fix a prolongation w of w to the algebraic closure K of K. For
each K-isomorphism 0 from L into K, we define a valuation wo of L by
w,(a) = w(a) (0 0' may not imply w. 0 w,,).
Consequently, the number of mutually distinct prolongations of w to L
does not exceed [L: K]S. If w is an additive valuation or a non-Archimedean
valuation, then this number coincides with the number of maximal ideals of
the integral closure of the valuation ring of w in L.
9. PROLONGATIONS OF A VALUATION 163
can assume that L = K. Let R be the valuation ring of w and let D be the
integral closure of R in T. Then the valuation rings of w., w' are DP ,
DM with maximal ideals P, M of D. Hence, there is a U E G(K/K) such
that aP = M (Theorem 3.7.11). Thus, the valuation ring of w' coincides
with that of wQ , and we can regard that w' = w, (In fact, first, we can
.
the other hand, if w' is an extension of w , then Theorem 4.6.1 shows that
there is a pair consisting of a positive real number t and a K-isomorphism
u of L into C such that (w'a)t = j oral (for all a E L). Since w'a = dual
for any a E K, we see that t = 1 . Q.E.D.
THEOREM 4.9.6. Let w be an additive valuation of a field K and let L be
a finite algebraic extension of K . Assume that mutually distinct prolongations
o f w to L are wt , ... , wn . Let R, R! ; P, P1 ; G, G. are valuation rings,
valuation ideals, and value groups of w, w,, respectively. Set e; = #(G1/G),
f = [R1lP : RIP]. Then E1 elf < [L: K1. Consequently, all et , fi are
finite.
el , f are called the ramification exponent and the degree of wl over w ,
respectively. wl is said to be unramified if e = 1 .
PROOF. (i) First, we shall show that elf < [L: K] for each i. Assume
that at , ... , as E R, are such that their residue classes modulo Pi are lin-
early independent over R/P and that at , ... , at E G. are such that their
residue classes modulo G are mutually distinct. Take b. E L such that
w (bl) = al . Assume that > Cjk ajbk = 0 with Cjk E K such that not all of
them are 0. Let I be the set of (j, k) such that wl (Cjk ajbk) is the small-
est. Then, since wl(C1) < w1(C2) implies w.(C1 + C2) = w.(c1) , we see that
wl(>(i,k)EI Cikaibk) > wl(Cikaibk) if (j, k) E I. Thus, #(I) > 2 .
But this follows from the following fact. Set {wj(b)lb is invertible
in (Rj)Q } and G" = {wblb is invertible in RQ}. Then ej" = #(G"J/G")
9. PROLONGATIONS OF A VALUATION 167
and the value groups of the valuations defined by (Rj)Qj , RQ are Gj/G",
GIG", respectively.
(v) For the general case, let gl , ... , g,,, be elements of L such that
L = K(g1 , ... , gm) . Take a subfield Ko of K such that these g, are
algebraic over K0. Let w0 , w0, be the restrictions of w to KD , of wi
to Lo = K0(g1 , ... , gm), respectively. Then w0i are prolongations of w0 .
Therefore, we take K0 above so that it is finitely generated over the prime
field and such that K0 satisfies the following 4 conditions:
(1) i 54 j implies w0i # w0i . (For each pair (i, j), take h.j E L such
that h .1 E P , h. P , and choose K0 so that h.j E L0 . )
(2) A representative of every residue class in GjIG is in the value group
of w0i (For each class, take a representative aid and an element bij such
.
Before proving this, we prove the following theorem, which is called the
lemma of Krull-Azumaya (some people call this Nakayama's lemma):
with coefficient matrix (8;t - ,alt) , where 8;t is Kronecker 6. Let d be the
determinant of this coefficient matrix. Then d - 1 E J ; hence, d is not in
any maximal ideal, i.e., d is invertible in R. On the other hand, d mi = 0
(for all i) by Lemma 3.7.1, and ml = 0. Thus, M = {0} .
For the general case, let M = MIN. Then M = JM, and M = 101,
namely, M = N. Q.E.D.
PROOF OF THEOREM 4.9.7. (i) We take a generator a (> 0) of G. If
a' E G. , then e,a' E G. This implies that w, is also discrete. Take
a generator a.(> 0) of G.. Then e.a, = a. Choose al c D so that
wi(a;) = a , wj(a;) > a (j 0 i) (approximation theorem). Take also a
.
THEOREM 4.10.1. (i) The product formula holds in the rational number
field Q with respect to V = {vp Ip runs through prime numbers and oo},
where v.(x) = JxJ and vp with p a prime number is the p-adic valuation
such that vp(p) =p -1 .
(ii) The product formula holds in the rational function fields in one variable
K(x) over afield K, with respect to V = {vp Ip runs through oo and monic
irreducible polynomials in K[x]} , where vo. and vp with a monic irreducible
polynomial p are defined as follows.
We fix a positive real number c such that c < 1 Then v.0 is the x-1-adic
.
0 ... 0 do d1
=
R). The coefficients of xn-'g (i < n) form the ith row of A(g , h) , and
the coefficients of xm-'h form the (n + i)th row of A(g, h) . Therefore,
that d = gk + hk' E R means that if we multiply the ith row by ei and
take the sum of these, then (a) every entry except the last one is 0 and (b)
d = c,nen + If we write down this fact in the form of a system of
linear equations in these ei , then we see by the property of cofactors that the
11. HENSEL'S LEMMA 171
degree m n .
Next, we shall show that det(A (c, 'g', d; h') is divisible by fJ,, l (vj -
1
u,) . Indeed, we substitute one vjo by ulo , then g' , h' are changed so
that they have a common root u . Then the system of homogeneous linear
.0
solution (ulo+n , u o+n-1 ulo , 1). Hence det(A(c1 g' , d- h') vanishes
1 1
sequences, and using limits c* = lim1 cij and el = lim1 eij , we see that
xr- I
g = xr + C1 + + cr , h = eo xs + + es satisfy the conditions (1), (2),
11. HENSEL'S LEMMA 173
(3). Now we show how to construct such sequences. Take e E P such that
f - goho E d 2eR[x] , and set di = de'. Then the requirement (1) is satisfied.
Assume that gr , h, are constructed for t < i. Since f - gihi E d. R[x],
f - gihi = d2e21elki(x) (e' E R, ki(x) E R[x]). It follows from (ii) that
d E g0R[x] + h0R[x] + dPR[x] = giR[x] + hiR[x] + dPR[x]. Therefore,
there are 0i , Y/., i E R[x] and e" E P such that
d=gi0i+hiy/i+de
2i-2 ,2 / /
2 2i , 2 ,
i
= di+1(ei i - e ei CbiVi).
Thus the proof is complete except for the case i = 0. In the case i = 0,
we can choose eo to be e, and f - g1 h 1 E di R[x]. Therefore, these gi, hi
are the required polyomials. Q.E.D.
A ring R is called a Hensel ring if R satisfies the following two conditions.
(i) R has only one maximal ideal (namely, the set of noninvertible ele-
ments forms an ideal).
(ii) Denote the maximal ideal by P. Then if, for a given monic polynomial
f (x) over R , there are monic polynomials go(x) , ho(x) such that go and
h0 have no common factor modulo P and f - goho E PR[x], then there
are monic polynomials g(x), h(x) such that f = gh and g - go E PR[x],
h - h0 E PR[x].
There are many results known on Hensel rings (see, for instance, Nagata,
Local rings, John Wiley, New York, 1962). In this book, we deal with several
results which are closely related to valuation rings. For this reason, we often
state some results under stronger conditions. For instance, Theorem 4.11.6
below is known to be true without assuming integral closedness.
THEOREM 4.11.6. Assume that an integrally closed integral domain R has
only one maximal ideal M. Then the following four conditions are equivalent.
(i) R is a Hensel ring.
(ii) Every integral domain R', which is integral over R, has only one
maximal ideal.
(iii) If f(x) = X'.+C1X'.-1+ - +cr (ci E R) satisfies c1 V M, C2,...,CrE
M, then f has a linear factor x + c' (c1 - c' E M).
174 IV. VALUATIONS
(Lemma 1.4.5).
Let f (x) , g(x) be the minimal polynomials for c, d over R, respec-
tively. Let al = 1 , a2, ... , ar be a set of representatives of G/H. Since
ac = c (for all a E H) , the conjugates of c are a1 C, ... , arc (or mutually
distinct members out of these; as is seen later, these are exactly the conju-
gates of c). By our choice of c, we have c c Mi' C M1 (for all i 1).
Since a^ : M1 (for j 54 1), a^ = M1 for some i. Thus, ajc E M1
(for j : 1) (and therefore, a1 c , ... , arc are the conjugates of c) This fact
.
modulo M) . This shows that R' has at least two maximal ideals, which
proves that (ii) implies (i). Q.E.D.
REMARK. When R' is a Galois extension of an integral domain R, and
P' is a prime ideal of R', we call Hp, = {a E G(R'/R)IQP' = P'} and
{a E R'IQa = a (for all Q E Hp,)} , the splitting group and the splitting ring
of P', respectively. The proofs that (iii) implies (ii) and (iv) implies (ii)
above show the following facts.
Assume that G(R'/R) is a finite group. (1) Hp, # G(R'/R) if the splitting
ring R* of P' has at least two prime ideals lying over P' n R, and in this
case, (2) there are elements a, b E R* whose respective minimal polynomials
are of the forms
.f (x) = xr + C
JCr-1 + ....+ Cr, C1 o P' n R, C2 , ... , Cr E P' n R,
g(x)=xr+dlxr-1+...+dr, d,.-1 P'nR, d,EP'nR,
where r = [G(R'/R): Hp,] = the degree of the field of fractions of R* over
that of R.
Assume that a given integrally closed integral domain R with field of frac-
tions K has only one maximal ideal P. Let K* be the separable algebraic
closure of K and R* the integral closure of R in K*. Let P* be a max-
imal ideal of R* , and let k be the splitting ring of P*. Furthermore, set
P = P* n k, R = RP . Since maximal ideals of R* are conjugate to each
other, k is uniquely determined up to R-isomorphisms, and we _call R the
Henselization of R. Obviously, the maximal ideal of k is P =PR .
THEOREM 4.11.7. Under the above circumstances
(i) The Henselization R of R is a Hensel ring.
(ii) If yr is a homomorphism from R to a Hensel ring R" and if VIP
is contained in the maximal ideal M" of R", then yr is extended to a ho-
momorphism from the Hen_selization k to R" so that VP c M" Such an .
say Ro .
It suffices to show R = Ro . Assume the contrary. Let K0 be the field
of fractions of Ro and Ro the integral closure of R in K0. Then Ro =
(RO)pnR . Set So = Ro - (P* n Ro). Then the integral closure of Ro in K*
o
coincides with Rso . From the definition of k, Rso has at least two maximal
ideals and therefore there is a finite Galois extension Ko of K0 such that, by
setting Ro = Rso n KK , PP = P* n Ro , the ring Ro has at least two maximal
ideals and the splitting ring RO of PP contains RO Set PO = P* n Ro Then,
. .
by Theorem 4.9.8.
(ii) For any element c E P n Q and an element b as above, we have
b+cER1 by (i), and cER1. Thus we have seen that PnQcR1.
(iii) In the general case, we can assume that b E PRP because R/P =
RIP. Since P c (P n Q)RT with T as above, b can be written as E cr t
with c, E P n Q, t. E T. Then c, E R1 by (ii), and we have b E R1 Thus .
R1 = RP. Q.E.D.
In the rest of this section, we shall consider valuation rings only.
178 IV. VALUATIONS
other maximal ideal of R , then the minimal polynomial for a" may have
1
replace R with R1(P,nR,). Then we have the case with smaller s and the
proof is completed by induction Q.E.D.
In the case of rank one valuation, there is an interesting relation between
completion and Henselization as follows.
THEOREM 4.11.11. Let R be the valuation ring of a non-Archimedean val-
uation v of a field K and let R , R be the Henselization and the completion
of R, respectively. Then R can be identified with the set of separably algebraic
elements of R over R.
PROOF. The canonical inclusion mapping R --> R can be extended to a
homomorphism from k to R (Theorem 4.11.7), and we may regard k to
be a subring of K. Since every element of k is separably algebraic over
R, we have only to show that if a E R_ is separably algebraic over R, then
a E k. The prolongations of v to R , R are also denoted by the same
symbol v .
Let k be the field of fractions of R and set R' = R n K(a) . Then R'
is a valuation ring, j?' C R' CR and the valuations defined by F?, R', R_
share the same value group. On the other hand, the integral closure R" of R
in K(a) has only one maximal ideal. Hence, R" is a valuation ring, which
coincides with R'. Thus, a is integral over R . Let f (X) = x" +Cix n-1 +- - +
c" be the minimal polynomial for a over R. If n = 1 , then a E R. Assume
for a moment that n > 1 . Since a is separable, the discriminant d of f (x)
is not 0. Take a nonzero element e of R such that ve < vd. Since a E A,
a - b E elk n R' = eR' for some b E R. Since d is also the discriminant
of the minimal polynomial for a - b, we may replace a with a - b. Thus,
we may assume that va < ve. Let the conjugates of a over k be a1 = a,
a2 , ... , an, and let v' be the unique prolongation of v to K(a1 , ... , an) .
180 IV. VALUATIONS
Q.E.D.
We can prove the following theorem from the result above.
THEOREM 4.11.12. Let R be a Hensel valuation ring. Consider polyno-
mials f(x), go(x), ho(x) in one variable x over R and an element d
of R. Assume that go is monic. Q denotes (i) the maximal ideal P of
R if d is invertible in R , (ii) d R otherwise. If f - g0h0 E d 2QR[x] ,
,
d E g0R[x] + h0R[x] + d QR[x] , then there are g(x), h(x) E R[x] such that
-
f = gh, g is monic, g - g0 E d QR[x] , h h0 E d QR[x] .
PROOF. (i) In the case where Krull dim R = r < oo, we use an induction
argument on r. The assertion is clear if r = 0. In the case r = 1 , we can
choose such g, h with coefficients in the completion R of R (Theorem
4.11.5), and it suffices to show that these coefficients are separably algebraic
over R (Theorem 4.11.11). Let al, ... , an ; $1 , ... , /3S be the roots of g,
h respectively. Then, a , ... , a,, , 91 , ... , /3S are the roots of f . Since
, I
g1 - g0 E d QR[x] , h1 = h0 E d QR[x] .
Furthermore, (g1 mod Q1) and (h1 mod Q1) have no common root (indeed,
if there is a common root, then we have a contradiction similar to the one we
observed in the case r = 1) . Let h2 be the polynomial obtained from h by 1
1 1. HENSEL'S LEMMA 181
f - goho=d q1
2
, d = g000 + ho yro + d g2Co.
Then for some n , qn = dq' (for some qi E R) . Let Ko be the field generated
by d, q1 , q2, ql , q2 and also coefficients of f , go , ho , q , 0o , wo Co) and
set Ro = R n Ko . Since Ko is finitely generated, Krull dim Ro is finite.
Thus, by our result in (i), the required g, h exist with coefficients in the
Henselization Ro of Ro , and Ro is contained in R. Q.E.D.
We close this section with an interesting application of the theorem above
and the approximation theorem.
THEOREM 4.11.13. Let R be a Hensel valuation ring of a field K such that
R # K. Suppose that, for any nonmaximal prime ideal Q of R, the field of
fractions of R/Q has a proper separably algebraic extension. Then if R' is a
Hensel valuation ring of K, either R' C R or R' D R.
PROOF. Assume that neither R' C R nor R' D R Set D = R' n R and let
.
Exercises
1
...Xn
E i,iXl ...Xn + Ed11...1 X1
EE
(ii...i.) je+ke=re
C'1-41 dki ...k
X`. .
n
(i) Prove that the set R[[X1 , ... , Xn]] of such formal power series
forms a ring.
This ring R[[X1 , ... , Xn]] is called the formal power series ring
in X1 , ... , Xn over R.
(ii) Prove that if R is a field and if n = 1 , then R[[X1 ]] is a com-
plete valuation ring and is the completion of the valuation ring
R[X1]X R[xi] .
3. Assume that R is a complete valuation ring whose maximal ideal is gen-
erated by an element n and such that K = R/irR is a perfect field of
characteristic p 54 0.
(i) For each 0 54 a E K, we take a representative an E R of a'
for each natural number n. Prove that the sequence lap } is a
convergent sequence.
(ii) Define q5a to be limn an . Prove that 0 is a group isomorphism
from K-{0} to K*={$aIO#ii EK} (c R).
(iii) Prove that every element a of R is expressed uniquely in the form
00 n
tor of length n. Thus, R and R/p"R are called Witt rings of length
infinity and of length n, respectively. As for the operations among
Witt vectors, the following facts are known.
For each m = 1, 2, ... , there are polynomials fn (xo , x1 , ... ,
x,nIYp,Y1,...,Yin) , gn(x0,x1,... ,x,n,Yp,Y1,...,Yin) in
2m + 2 variables over the prime field of characteristic p, so that
for two Witt vectors
6
1. Show that the rational function field C(t) in one variable over the com-
plex number field C has an Archimedean valuation and that its restriction
to C is not equivalent to I (absolute value).
I
1. Prove that any valuation ring of Krull dimension > 2 is not Noetherian.
(Cf. exercise 4.7.7.)
2. Prove that if Rj (j E M) are valuation rings of a field K, then the
intersection njEM Rj is an integrally closed integral domain.
3. Prove that if an additive group G has a subset P satisfying three con-
ditions below, then we can give an order on G so that G is an ordered
additive group and a E P if a > 0. Prove also that such an order is
uniquely determined by P.
(i) If a, b P, then a +bEP.
(ii) If 0 a E G and a P, then -a E P.
(iii) 0 0 P.
EXERCISES 187
8
1. Let v1 , ... , vn be additive valuations of a field K. Find necessary and
sufficient conditions on the relation among the valuation rings R , ... , 1
prime ideal P that is not contained in any other Ri . Prove that there is
an element a of P which is invertible in any Ri (i > 1).
11
191
192 V. FORMALLY REAL FIELDS
obviously satisfies condition (ii) in Lemma 5.1.6. Theorem 5.1.5 shows that
P also satisfies condition (iii). As for (i), given a, b c P, we have a = c2
and b = d 2 , for c, d E K with cd 0. Hence, ab E P. Assume for a
moment that a + b P. Then -(a + b) E P U 101, so for some e E K, we
have c2 + d 2+ e2 = 0, a contradiction. Thus, a, b E P implies a+ b E P,
and K is an ordered field.
If K is an ordered field, then every element of P must be a positive
element, since it is the square of a nonzero element. If a nonzero element b
of K is not in P, then -b E P by Theorem 5.1.5. Thus, P is exactly the
set of positive elements. Therefore, the uniqueness of the order follows from
Lemma 5.1.6. Any automorphism of K, maps P to P, and therefore, the
order is preserved. Q.E.D.
PROOF OF THE LATTER HALF OF THEOREM 5.1.2. There is a real closed field
K* that contains K (Theorem 5.1.3) and K* is an ordered field (Theorem
5.1.7). Therefore, restricting the order on K* to K, we see that K is an
ordered field. Q.E.D.
THEOREM 5.1.8. If an ordered field K satisfies conditions (i) given a > 0,
a = b2 for some b c K and (ii) for f (x) E K[x] with deg f (x) odd, there
is an element c e K with f (c) = 0, then K is real closed and K(im) is
algebraically closed.
PROOF. The fact that K(im) is algebraically closed can be shown by the
same way as in the proof that the complex number field is algebraically closed
(Theorem 2.12.4). Since K is an ordered field, K is formally real. Further-
more, the only algebraic extensions of K are K and K(im). Therefore,
K is real closed. Q.E.D.
THEOREM 5.1.9. The conditions (i) and (ii) in the theorem above charac-
terize real-closedness of an ordered field K.
PROOF. We have only to show that (i) and (ii) follow from the condition
that K is real closed. Since the order on K is unique (Theorem 5.1.7), (i)
is straightforward by Theorem 5.1.5. As for (ii), let f (x) = cox" + + c,t E
K[x] (c1 E K, co 0) be a polynomial of odd degree n. To prove the
existence of a root in K , we may assume that co = 1 . We use an induction
argument on n . Existence is obvious if n = 1 , and so, we assume that
n > 1 . If f is reducible, then one factor is of odd order, which has a root
194 V. FORMALLY REAL FIELDS
a, , /3, V K, because a2
Letting al , a2 , f , 92 be
the roots of g, we may assume that a? = a+ b , f? = a -
V K K. a1 a2 = -(a + b
, /3?
.
) , and therefore,
I
b
a2 is not a conjugate of a1 . Hence, we may assume that /31 is conjugate to
a1 . Then al /31 E K and (a1 /31)2 = a 2 = a2 + b2 Thus, we have shown
1131 .
K(a1 , ... , al) are the required fields.) Assume that aap-1 E (K(a)). Then
for some b = c0 + c1 a + + cp_ 1 ap-1 (c, E K), aap-1 = by - b. Since
cap = a + a, we have
aap-' cp(a+a)`-E cla`.
Since 1, a , ... , ap-1 are linearly independent over K, by comparing the
ap-1
coefficients of , we have a= p_ 1 - cp_ 1 Thus, a E OK, a contradic-
.
tion. Q.E.D.
THEOREM 5.1.13. Assume that the field K is not algebraically closed and
that the algebraic closure K of K is of finite degree over K. Then K is real
closed and consequently, K = K(im).
PROOF. If K = K(im) , then the assertion follows from Theorem 5.1.10.
Assume for the moment that K(im) c K.
If K is not a perfect field, then the characteristic p of K is not 0 and
Kp' D K. Since Kp ' -= K, it follows that K -A Kp ' # Kp # , and 2
(The existence of r follows from the fact that [n(C2): iv] < q2 and that
[n(Cr): 7L] is big for large r (cf. Theorem 2.9.6).) The minimal polynomial
r+I
g(x) for Cr+l over L is a factor of fl (x - Cr+I) and therefore, g E
I
i.e.,
2
cia +Ecib,n/32 = 0, 2Eciaifi = 0.
2. REAL CLOSURES 197
From the left-hand side equality, we have ai = /3i = 0 for all i. Thus,
ai = 0 for all i. Q.E.D.
THEOREM 5.2.2. If K is an ordered field, then there is a real closure K*.
Real closures of K are essentially unique. In particular, if K" is another real
closure of K, then there is an order isomorphism from K" onto K* over K K.
PROOF. Let K' be the field generated by square roots of positive elements
of K. Since K' is formally real, there is a real closed field K* that is al-
gebraic over K' (cf. Corollary 5.1.11). For a E K with a > 0, we have
Va- E K' , so a = (/)2 is positive in K* This shows that the set P of pos-
.
t such that al < < at < a, b < at+1 < ... < ar , we see that both
f (a)(-1)r-' and f (b)(-1)r-t are positive, which is not the case. Q.E.D.
For a given polynomial f (x) E K[x] (K a field and f (x) K)) , the stan-
dard sequence f0 = f , f1 , ... , f, is defined as follows: f1 (x) = (d /d x) f (x)
(i.e., if f = E cix' , then f1 = E icixt-1) When f0 , ... , f are defined
.
have (iii).
(iv) Assume that go(c) = 0, a < c < b. Write f = (x - c)eho(x)
(ho(c) 0), then f1 = e(x - c)e-I ho(x) + (x - c)e((d/dx)ho(x)) . Hence,
go(x) = (x - c)h(x) with h(x) E K[x] (h(c) 0), and g1(x) = eh(x) +
(x - c)k(x) with k(x) E K[x]. Since h(c) 0, we have g1 (c) 0, and
we can choose d1 < c < d2 so that both h(x) and g1(x) do not vanish
in the interval [d1 , d2] , because the numbers of roots of h and of g1 are
finite. Lemma 5.2.4 shows that each of h, g1 does not change its sign in
[d1 , Therefore g1h takes positive value in the same interval, because
d2] .
For the proof of our theorem, it suffices to show that the number of distinct
roots of go in the interval [a, b] is the difference of the numbers V, Vb of
variations of signs of go (a) , ... , gs(a) ; go (b) , ... , gs(b) .
If we take a = ao < a 1 < < an = b such that go (ai) 0, then
Va - Vb = 01(Vai - Vas+1 ) where Vas is the number of variations of signs of
go(ad) , ... , gs(a,.) . Therefore, we may count the number of roots of go in
each interval [ai , a,1], after choosing ai so that they satisfy the following
four conditions: (I) In each interval [ai , ai+1 ] , go has either no root or only
one (single or multiple) root. (II) Every root of gi in the interval [a, b] is
either some ai or a root of go. (III) If go(c) = 0 and a, < c < a,+1 , then
ai, ai+1 satisfy the condition on d1 , d2 in (iv) above. (IV) go(ai) 0 for
all i.
Since the number of roots of gi is finite for each i, we can choose such
ai . Now, it remains to show that Va - Va = 1 or 0 depending on whether
the interval [ai, ai+1 ] contains a root of go or not.
( 1 ) The case where go(c) = 0 for some c c [ai, a , ] : By condition (III),
1
interval by the conditions (iii) and (II), and g1 , ... , gs is a Sturm sequence.
Therefore, our assertion follows from the remaining case (2).
(2) The case where go does not vanish in [ai, a,1]. If there is a gj
(s > j > 0) that does not vanish in this interval, then we obtain the claim
by observing the two shorter Sturm sequences go , ... , gi and gi , ... , gs .
Thus, we may assume that if s > j > 0, then gj vanishes at ai or ai+1 .
When gi vanishes at both ai and ai+1 , then by dividing the interval into
two intervals, we may assume that each gj vanishes at only one of ai , ai+1 .
(i) If s = I , then Var = Var+i = 0, and the assertion is clear.
(ii) Assume that s = 2. If g1(ai) = 0, then Q = 1 (Condition (iii)).
Vaf+, = 1 , because neither of go, g2 vanishes and gog2 < 0 in this interval.
The case g1(ai+1) = 0 is proved similarly.
(iii) If s > 2, then by a suitable division of the interval, we can reduce
to the case where s < 2, because there is no j (0 < j < s) such that
gj(ai) = g1+1(ai) = 0. Q.E.D.
LEMMA 5.2.6. Let K be an ordered field. If a E K is a root of xn+c1 xi-1 +
+cn (ci E K), then -m < a < m, where m = max{ 1, Ic I+Ic2I+ +I cn I } 1
PROOF. If Ial < 1 , then there is nothing to prove. Assume that Ial > 1 .
c2a-1 cna-(n-1
It follows from -a = c1 + + + ) that
IC I
+ Ic2I + ... + Icel. Q.E.D.
200 V. FORMALLY REAL FIELDS
plies V. Va < 1 . Thus, f has at most one root in each of the intervals
[c1_ 1, a1] . This completes the proof, since f is assumed to have no root in
each of the intervals [al , c,] . Q.E.D.
PROOF OF THEOREM 5.2.3. For f (x) = coxn
+c1xn-I
+ +cn E K[x], we
denote by Of (x) the polynomial (q5co)x" + (gc1)xn- I + + c,t . Note that
if fo, f1 , ... , fs is the standard sequence for f (x), then q5fo , Oft , , Ofs
is the standard sequence for Of (x) .
When a monic polynomial f (x) is given, taking m in Lemma 5.2.6, we
apply Theorem 5.2.5 to the case a < -m, m < b. Then we see (i) and (ii)
below.
(i) (The number of mutually distinct roots of f(x) in K*) = Va - Vb =
Va - VG = (the number of mutually distinct roots of Of(x) in K'*), where
V and V * denote the number of variations of signs of standard sequences
of f and Of, respectively.
(ii) If a1 < ... < an (a1 E K*) , then 0 is extended to an isomorphism (not
necessarily an order isomorphism) yr of K(a1 , ... , an) into K'* such that
t/ra1 < yra2 < < yran . Indeed, bl = a1 - a1_I E K* (i = 2, ... , n),
and we have a subfield K1 = K(a1 , ... , an , b2 , ... ,
Take an element c
b,1) .
such that K1 = K(c) (Theorem 2.5.2), and let f (x) be the minimal polyno-
mial for c over K. (i) above shows that there is a root c' of Of (x) in K'* .
Then we have an extension yr of 0 such that yrc = c'. Then yrbl E yrK(c)
and 0 < yrb2 = Oat - dal_ 1 .
Now, we shall define a mapping a of K* to K'*. For each a E K*, we
take the minimal polynomial fa(x) for a over K and let a1 < a2 < . . < an .
be the roots of fa (x) in K * . Let a i < a? < < an be the roots of cb fa (x)
in K'* (cf. (i) above), and we define a by Qa = aj , with j is such that
a=ai . Then
(1) as = Oa for a E K (because fa (x) = x - a).
2. REAL CLOSURES 201
isomorphism of fields.
(4) That as > 0 for 0 < a c: K* follows from the properties that E K*
and as = (a/)2 > 0. This and (3) show that a is an order isomorphism
since for a > b, we have a - b > 0, and hence, a(a - b) = as - ab > 0, so
as > ab.
(1), (2), (3), and (4) above show that qS can be uniquely extended to an
order isomorphism a. In particular, if q is the identity mapping, then
a must be the identity mapping. Thus, we have completed the proof of
Theorem 5.2.3. Q.E.D.
Note that we obtain the following by Corollary 5.1.11.
THEOREM 5.2.8. If an ordered field K is a subfield of a real closed field L
as an ordered field, then the algebraic closure of K in L is the real closure of
K.
Using this result, we can prove.
THEOREM 5.2.9. Let K be an ordered field that is a subfield of a real closed
field L as an ordered field. Let K' be an algebraic extension of K, and let
H be the set of K-isomorphisms from K' into L.
(i) Assume that H is not empty. Then, fixing an element a of H, we
define an order on K' by a > b iff as > ub, with this order, K' becomes
an ordered field.
(ii) Conversely, assume that we can extend the order on K to K' so that
K' is an ordered field. Then the order is the one defined by some a E H as
in (i). Furthermore, this correspondence between orders on K' and a E H is
a one-to-one correspondence.
PROOF. (i) is clear.
(ii) Since K' contains K as ordered fields, the real closure K'* of K' is
also the real closure of K. By Theorem 5.2.3 and Theorem 5.2.8, we see
that the algebraic closure K* of K in L is order isomorphic to K'*. The
order on each of K* , K'* is unique (Theorem 5.1.7). Thus, with the unique
order K-isomorphism 0 from K'* to K*, its restriction 5IK' to K' is an
202 V. FORMALLY REAL FIELDS
PROOF. The if part is clear and we consider only if part. Assume that a
is. not the sum of any square elements. Let F be the family of intermediate
fields L between K and its algebraic closure K such that a is not the
sum of any square elements in L. Then F is an inductive set and has a
maximal member L*. Lemma 5.1.4 shows that L* is formally real, and
therefore, we can define an order on L* so that L* is an ordered field. We
shall prove a < 0 in the ordered field L* , and it suffices to show that -a
is a square element in L*. Assume the contrary. Then L* (/) : L* .
5.1.4 shows that a is the sum of certain square elements, which is not the
case. Q.E.D.
LEMMA 5.3.2. Let xi , ... , xn be algebraically independent elements over
an ordered field K. Then we can extend the order on K to K(xi , ... , x,,)
so that K(x, , ... , xn) becomes an ordered field.
PROOF. Let K* be the real closure of K. Then K* (xi , ... , xn) is for-
mally real (see Exercise 5.1.6). Thus, K*(x, , ... , xn) can be made an or-
3. HILBERT'S 17TH PROBLEM 203
ym(x) (y1 E K(x)) be the minimal polynomial for 0 over K(x). Then
there are bi(x, y), b'(x, y), cj(x, y) E K(x)[y] such that al = b; (x , 0),
aj+1 - aj = b(x, 0)cj(x, 0) = 1 Since f (x, al) = 0, we
.
fl, ... , fs
By our inductive hypothesis, there is an n-tuple (a) = (a1 , ... , an) of
rational numbers ai such that {g1(x) , ... , gt(x)} has the same signs as
{g1 (a), ... , g,(a)}. Since fs+1 , ... , fr are among these gi , { f +1(x) , ... ,
fr(x)} has the same signs as { f +1(a) , ... , f,.(a)}.
Now, consider f (i < s). Let a < <a. be the roots of f in
.
among y and these ai . (Note that K(x) * can be embedded in the real
closure K(x, y)* of K(x, y) by Theorem 5.2.8.) Just for convenience, we
set ao = flo = -oo , as+1 = fls+I = oo , and choose i such that ai < y < ai+I .
Then a rational number an+1 such that /3i < an+I < Qi+I is the required last
member. Q.E.D.
The next theorem, which is due to Artin, gives an affirmative answer to a
generalization of the Hilbert's 17th Problem. It is clear that this theorem is
a generalization, because the rational number field, as an ordered field, has
unique order (cf. Exercise 5.1.3).
THEOREM 5.3.6. Assume that a formally real field K has the property that
whenever K becomes an ordered field under some order, then K is order
206 V. FORMALLY REAL FIELDS
isomorphic to a certain subfeld of the real number field. Then, for an element
f(x) of the purely transcendental extension K(x) = K(x1 , ... , xn) of K,
it follows that there are elements gi (x) E K(x) with f (X) _ E i gi (x)2 iff
f (a) > 0 for any order on K that makes K an ordered field and for any
(a) = (a, , ... , an) with rational numbers al whenever f (a) is well defined.
PROOF. The only if part is obvious. Assume that f (x) is not the sum of
any square elements. Then Theorem 5.3.1 shows that there is an order on
K(x) which makes K(x) an ordered field with f (x) < 0. Then there are
rational numbers ai such that f (a) < 0 by Theorem 5.3.5. Q.E.D.
COROLLARY 5.3.7. Let K be a subfield of the real number field R such
that K has a unique order as an ordered field. Let K(x) be as above, and
consider f (x) E K(x). Then f (x) is the sum of some square elements iff
f(aI , ... , an) > 0 for any rational numbers ai whenever f(a1 , ... , an) is
well defined.
Note that this corollary can be applied to the rational number field or real
closed subfields of R (cf. Exercise 5.1.3 and Theorem 5.1.7).
4. A valuation corresponding to an order
THEOREM 5.4.1. Let S be a subfield of an ordered field K. Then V =
{x E KI for some a c S, -a < x < a} forms a valuation ring of K with
maximal ideal M = {x E KI if a E S and a > 0, then -a < x < al.
Furthermore, letting yr be the natural homomorphism from V to VIM, we
see that VV = VIM is an ordered field in a natural way such that yra > 0
iffa>0 and a V M.
This V is called the valuation ring of the ordered field K with respect to
S
PROOF. Assume that x E K, x V. If x > 0, then x > a (for all
I
a (=- S). Hence, x- < b for any positive element b of S. Consequently,
x-I E M in this case. The case x < 0 can be proved similarly. Thus, V is
a valuation ring of K and M is the maximal ideal of V.
Set P' _ { yrx Ix > 0, x V MI. Obviously, 0 P'. If a > 0, b > 0,
and a, b M, then there is an element c E S, with c > 0, a > c, b > c,
so a + b > c and ab > c2 . This fact shows that if a', b' E P' , then a' + b' ,
a'b' E P'. Therefore, V/M is an ordered field with the set P' of positive
elements by Lemma 5.1.6. Q.E.D.
The main aim of this section is to prove a certain converse of this theorem.
However, for simplicity, we consider the case where S is the rational number
field Q. Then the result can be stated as follows. (For the general case, see
Exercise 5.4.2.)
THEOREM 5.4.2. Let V be a valuation ring of a field K. Assume that there
is a homomorphism yr from V to the real number field R whose kernel is
4. A VALUATION CORRESPONDING TO AN ORDER 207
Exercises
1
degree over K, then [G(L/K) : G(L/L1)] = [L1 : K]5 < [L1 : K].
PROOF. Obviously G(L/L1) C G(L/K) by definition. Let 0 be the map-
ping of G(L/K) to G(L1 /K) that associates a ((=- G(L/K)) to its restric-
tion on L1 Ou is well defined, because L1 is normal over K. Since
.
a1Hj = QHj (for all H E C). Assume for a moment that a F for some
F E C. Then there is a neighborhood U of 1 such that a U n F is empty.
Take H. E S such that H c U (property (iv)). Then, aH c uU and
QHj n F is empty, a contradiction. Thus a E nFEC F. Q.E.D.
THEOREM 6.1.4. Let L be a normal algebraic extension of a field K. Then
the Galois group G = G(L/K) is a compact group. If K' is an intermedi-
ate field between K and L, then G(L/K') is a closed subgroup of G. If
furthermore, [K' : K] is finite, then G(L/K') is an open set of G.
PROOF. Consider K21 in Lemma 6.1.2. If N E S21 , then G/N is a finite
group; hence, G/N is compact. Therefore, G is compact by Lemma 6.1.2
and Theorem 6.1.3. As for G(L/K') with a finite extension K', there is
a finite normal extension L' of K containing K'. Then N = G(L/L') E
K21 , and N is an open set. G(L/K') is the union of subsets of the form
aN; hence, G(L/K') is an open set. Consequently, G(L/K') is also closed
by Theorem 4.4.5. As for G(L/K') in the general case, K' is generated
by a certain number of finite extensions Kj of K, and G(L/K') is the
intersection of G(L/K) (Lemma 6.1.2). Since each G(L/Kj) is a closed
set, we see that G(L/K') is a closed set. Q.E.D.
THEOREM 6.1.5. Under the above circumstances, the topology on G(L/K')
coincides with induced topology (i.e., the topology as a subspace of G(L/K)).
The proof is clear by the relationship between the family of fields of the
form K'(L') with finite normal extensions L' of K and the family of finite
normal extensions of K'.
2. The fundamental theorem of Galois
Let L be a Galois extension, namely, a separable normal algebraic exten-
sion of a field K. In view of the topology defined in 1, we can state the
fundamental theorem of Galois as follows.
THEOREM 6.2.1. There is a one-to-one correspondence between the family
F of closed subgroups of G and the family 0 of intermediate fields between
K and L as follows.
If H E I' corresponds to M E A, then
(i) H = G(L/M),
218 VI. GALOIS THEORY
field M", which is finite over M and containing a, we see that there is an
element a' E G(M"/M) such that a'a a. A prolongation a of a' to an
automorphisms of L belongs to H, contradicting a E M'. Q.E.D.
3. Splitting fields, inertia fields, and ramification fields
Let R be an integrally closed integral domain with field of fractions K,
and let L be a normal algebraic extension of K. If we take the integral
closure RL of R in L, then AutR RL is identified with G = G(L/K) and
called the Galois group of RL over R (cf. Theorem 3.7.11). Futhermore,
the splitting group HP of a prime ideal P of RL is {a E GI QP = P} (cf.
Remark after Theorem 4.11.6). Here we shall define some others. For an
ideal I of RL , VI = {Q E GIQx - x E I for all x E RL} is called the
ramification group of I. Ramification groups of primary ideals belonging to
a prime ideal P are called ramification groups of P. If I is a prime ideal,
then VI is called the inertia group of I.
THEOREM 6.3.1. The subgroups HP , VI above are closed subgroups of the
Galois group G.
PROOF. Take an arbitrary a not in HP (a E G). Then uP 0 P, and
ab P for some b c: P. Then N = G(L/K(b)) is a neighborhood of
1 by Theorem 6.1.4; hence, QN is a neighborhood of a. T E N implies
arb = ab 0 P. Thus, uN and HP have no common element, which shows
that a is not in the closure of HP and HP is a closed set.
Next we consider VI. Take a VI (a c G). Then for some element
c E RL , ac - c I. Consider N = G(L/K(c)). TEN implies QTC - C =
cc - c 0 1. Thus, aN n VI is empty, and we see that VI is a closed
set. Q.E.D.
We defined the splitting field and the splitting ring as field and ring cor-
responding to the splitting group HP In the same manner, we define the
.
inertia field and the inertia ring as the field and the ring ({x E Lax = x
(for all a c VP)} , {x (=- RLIa, = x (for all a E VP)}) corresponding to
the inertia group VP . Similarly, we define the ramification field and the
ramification ring of an ideal I by using the ramification group of I.
THEOREM 6.3.2. Under the above circumstances, let L and K be the fields
of fractions of RLIP and R/(P n R), respectively. Then, the inertia group VP
is a normal subgroup of the splitting group HP , and L is a normal extension
of K with Galois group G(L/K) naturally identified with HPIVP .
fractions of RI/(P n RI) Therefore, we may take the inertia field instead of
.
over RSl(Pf1Rs). On the other hand, RLIP is purely inseparable over RI/(Pfl
RI).
LEMMA 6.3.4. Let R and RL be as before. Let I be an ideal of RL and
let T be a subset of R such that no element of T is a zero-divisor modulo
I (i.e., b E T, c E RL , be (=- I implies c c I). Then, taking RT , (RL)T'
and I' = I(RL)T instead of R, RL, and I, respectively, we can consider the
ramification group VI, of P. This VI, coincides with the ramification group
VI of I .
(2) Assume now that aI = 1. T'QT-` is of the form (j, k). For a
suitable i, 1 appears among j or k. Then we can apply (1), to such a
T'QT-! and q. Q.E.D.
Incidentally, Theorem 6.3.2, in the rational coefficient case, can be stated
as follows.
THEOREM 6.4.2. Assume that the coefficients ci of f (x) = xn + cI xn-' +
c2xn-2 + + cn are all rational numbers. Let R be the ring of algebraic
integers in the minimal splitting field L of f (x) and let P, p be a maxi-
mal ideal of R and the prime number contained in P (i.e., P n Z = pZ),
respectively. Each element of G(L/Q) gives an automorphism of R and each
element of H = {a E G(L/Q)IQP = P} induces an automorphism of RIP.
Furthermore, R/P is a Galois extension of Z/pZ whose Galois group consists
of automorphisms induced by elements of H.
Now we shall achieve our aim by proving the following theorem.
THEOREM 6.4.3. For each natural number n > 2, there exists a polynomial
f (x) = xn + clxn-' + + cn with rational integral coefficients ci such that
its Galois group is the symmetric group Sn of degree n.
PROOF. If n = 2, then any irreducible f (x) over Q satisfies the require-
ment. The n = 3 case is also easy and we leave this case to exercise 6.4.1. So
we shall consider the case n > 4. We choose three mutually distinct prime
numbers p, q, r and three polynomials g(x), h(x), k(x) with rational
integral coefficients satisfying the following conditions.
g(x) = xn + a,xn-' + a2xn-2 + + an is irreducible modulo p (over
Z/pZ).
h (x) = xn + b1 xn-' + b2xn-2 + - + bn modulo q has an irreducible factor
of degree n - 1.
k(x) = xn + elxn-' + e2xn-2 + + en modulo r decomposes into the
product of an irreducible quadratic polynomial and n - 2 linear factors and
has no double root modulo r.
By Theorem 1.3.5, there are rational integers ci (i = 1 , ... , n) such that
ci = ai ( mod p) , ci - bi (mod q) , ci - ei ( mod r) . It suffices to show
4. ALGEBRAIC EQUATIONS OF HIGH DEGREES 223
Thus we can see that there are many polynomials (E Q[x]) of degree n
whose Galois group is Sn .
We shall give a concrete example in the case n = 5, following the proof
above.
EXAMPLE. Let p = 5, q = 2, r = 3 and
g(x) = x5 - x - 1,
h(x) = (x4+x+ 1)(x+ 1) x5 +x4+x2 + 1 (mod 2),
k(x) _ (x2 + x - 1)(x + 1)(x - 1)x - x5 + x4 + x3 - x2 + x (mod 3).
Then, one example of f (x) is
f (x) = x5 + 25x4 + 10x3 + 5x2 + 4x + 9.
Exercises
1
3. Let F be a finite field, and let F be its algebraic closure. Prove the
following.
(i) The additive group of the set Z of rational integers can be made a
topological group by letting N = {nZJn is a natural number} be a
fundamental system of neighborhoods of 0.
(ii) There is a homomorphism 0 from the additive group Z to G(F/F)
such that
(1) the topological group Z above is isomorphic to the subgroup
q5Z of G(F/F) with topology as a subspace of G(F/F),
(2) c(Z) is dense in G(F/F) ; hence, G(F/F) is the completion
of 0z'
(3) the element 01 = a has the property that, for any a E F ,
as = aq with q = #(F).
(iii) G(F/F) is a commutative group. Furthermore, if F' is another fi-
nite field and F' is its algebraic closure, then G(F'/F') G(F/F).
2
1. Let K be the algebraic closure of a field K and let H be the set of order-
finite elements a E G(K/K) Prove that either H = { 1 } or, for any a in
.
1. Prove the case n = 3 in Theorem 6.4.3 in the following form. The Galois
group of f (x) = x3 + ax2 + bx + c (a, b, c being rational integers) is
the symmetric group of degree 3, if there are two prime numbers p, q
such that
(i) f (x) (mod p) is irreducible over Z/pZ, and
(ii) f (x) (mod q) is the product of a linear factor and an irreducible
quadratic factor over Z/qZ.
2. Give a concrete example of f (x) , as in the last part of 4, in the case
n=5,p=3,q=2,r=7.
Answers and Hints
CHAPTER I
227
228 ANSWERS AND HINTS
- I J C- I n J.
3. (1) (i) If I, J are ideals of a ring R, then (I + J)(I n J) C
(ii) R=(I1+I2)2cI +I2 and Ii +I2=R.
(2) Use (1) and an induction argument on n.
4
(3) deg(f1 + f2) < max{deg f1 , deg f2 } ; if def fl deg f2, then deg(f1 +
f2) = max{deg f1 , deg f2} .
6
lengths R = sic(i)
REMARK 1. .
1. The if part is obvious. As for the only if part, when T, T' are odd
permutations, TT', T- I T' are in An , and TT' f = f , T f = T' f . Thus,
g = f - T f is independent of T, and Tg = T f - f = -g. Hence, g is an
alternating form. h = f+Tf is a symmetric form, and f = (h/2)+(g/2) .
p = 2, then the group is the direct product of the cyclic group (or order
2) consisting of the residue classes of , -1 and the cyclic group (or
1
not a simple extension. For the proof, first show that the extension is of
degree p2 , and then adapt the last part of the proof of Theorem 2.5.5.
For the degree part, cf. the hint for exercise 2.3.3 in the cases:
(i) Apply it to ko(t, u) = ko(t)(u) and obtain [ko(t, up-') : ko(t, u)] _
p,
(ii) Apply it to ko(t, up-') = ko(up ')(t) and obtain
[ko(tp , up ko(t, up )] = p
2. Let L = K(a) , and consider the minimal polynomial f (x) = fl.' (x-ai ) 1
7
a(0) = r2 , T(4 2) =
1. (1) {1 } . (2) See exercise 2.6.5, (4) . (3) 111 in the characteristic 2
0,
231
mal polynomial for a over K and let d1 , ... , d,U be a linearly inde-
pendent base of L over K(a) (rv = n). Now, let b1 , ... , b be 1
a , ... , ar-l, d1 , d1a, ... , dear-1, ... dia', ... , d,U , da, ... , da"-'
3. If we can prove this in the case K = Q (the rational number field), then
the general case can be proved as in Theorem 2.9.4. The K = Q case
follows from Theorem 2.9.5 and exercise 2.4.1. If n = 2S with natural
number s > 2 and if K = Q, then the extension is not cyclic.
4. Take a such that L = K(a). Then 1, a, ... , an-1 form a linearly
independent base of L over K. On the other hand, det(ai (a')) 0,
because
xn-1
1 X 1
x21 1
... x2-1
1 x x2
.. ... ... ... 2..
= II(xi - xi).
Xn-1 i>j
1 xn X2n n
10
2. We can reduce the problem to the case where LI , L2 are finitely gener-
ated. Then we can use separating transcendence bases.
3. [KI : K] < i(L/K) is easy by considering L K KI . i(L/K) < [L
K(x1 , ... , x")]1 follows from Theorem 3.4.3 and the fact that [L(K" ) :
K -00 (x1 , ... , x")] > [L : K(x1 , ... , x,,)],. For this last fact, use Theo-
rem 2.7.3, (ii).
5
1. For the first half, use the zero-point theorem of Hilbert and the fact that
V(I(AI) + I(A2)) = AI n A2. For the last part, consider the polynomial
ring P = R[xI , ... , xn ] (n > 2) over the real number field R. Then,
with A1 = V (x1) , A2 = V (-xi + En 2x2 + 1) , we have II(A1)+I(A2)
(A _
xI P + (> 2x2 + I )P = I (AI) + 1(A2), but A n A2 is empty. I
1. Consider xi + + xn
2. Set K = R(t , ... , ti) and L = C(t , ... , ti) with algebraically indepen-
I I
dent elements tt , ... , ti over the real number field R. Here C denotes
the complex number field. K is not a C,7-field for any n and L is a
Ci-field but not a Ci _ 1 -field.
ANSWERS AND HINTS 237
12
by=c=ap
2 2+bq 2 . From this relation, we have y E K(t).) If (p, q)
We can assume that s > u > 0 (> 0 follows from the nonexistence of
p). Then f (O) -A 0 and a(f (t)/g1 (t))2 + b(ts-"h(t)/k1 (t))2 = cts By .
1. Take a valuation ring V.. of K(x1 , ... , xr) containing K and such that
V has prime ideals PI P2 D D P,. D {0} with the property that
Pi = x(V) p for each i. Then adapt Lemma 3.A.2, considering V n
K' (XI, ... , xr- I) .
2. Here is a proof of the theorem stated in the hint. If n = r, then there
is nothing to prove. Assume that n > r. We use the normalization
theorem for polynomial rings using exercise 3.8.1, and we see that there
are linear combinations yI , ... , of xI , ... , x1, with coefficients in
K such that L[xI , ... , xn] is integral over L[y1 , ... , yn-r] Choose -
aI , ... , as E L such that L = K(a1 , ... , as). Then, x, , ... , x11 are in-
tegral over K[a1 , ... , as, yi , ... , yn_r]. There is c E K such that, when
we write aI , ... , as in fractional forms of xI , ... , xn , no denominator
is divisible by yI -c. Consider the ring V =K[x1 , ... , x,,]3, _
Since aI E V, V contains K[ai , ... Set P =
, as , x1 , ... , xn] .
is a contradiction.
3
1. (i) Let M be a set with at least two elements and let open sets on M be
only the empty set and M itself.
(ii) With M as above, let the open sets on M be M itself and all subsets
not containing a fixed element a.
(iii) Let M be a set containing infinitely many elements and define that
a subset S is an open set if either S is the empty set or the complement
of S consists of a finite number of elements.
(iv) We fix a line L and a point A which is not on L, on the Euclidean
plane P. Then, a new topology is defined on P by letting the following
family C of subsets be a subbase of open sets. C = BI U B2 U B3 , where
BI={{x}jxE P, x0 A, x V L}, B2={UIU is an open set in the
usual topology not containing Al , B3 = {P - (L U U) I U is the union of
a finite number of circular disks}. Then L is a closed set, and there is
no pair of neighborhoods of L and of A separating each other.
(v) On the plane P, we set U(a, b, 8,J) = {(x, y) E Pea < x < a +
e, b < y < b + 61 with real numbers a, b, e, 8 We define a new
.
H,
2. The space is not a TI -space. Indeed, there is P E G, such that P
and any neighborhood U(P) of P meets H. This implies that any
ANSWERS AND HINTS 239
equality follows from the fact that M is a metric space). For the only
if part, use an induction argument on n Assume that a1I , ... , ant ,
.
Let t(i) be natural numbers such that t(l) < t(2) < . Then di =
`1J=2 Ku j , and {di} is a Cauchy sequence converging to 0.
mt(i) - mi E
Therefore, by our induction hypothesis, {cij = at(i) j - ai j I i = 1 , 2, . .. }
is a Cauchy sequence converging to 0 for each j > 2. It follows now that
{ai j I i = 1, 2, ...1 is a Cauchy sequence. Set a = limi-+00 aij . Then
u1 + a2 u2 + + a*u n = mi = 0, which contradicts the linear
independence of u 1 , ... , un .
6
1. If i 0 j, then R, [Rj] = K. (In this case, we say that R, , ... , R,, (or,
ANSWERS AND HINTS 241
an inductive set and has a maximal member (M*, vM. , 0M.) . Then the
residue class field of vM* is isomorphic to K. Hence, the completion of
vM* is the required one.
10
1. Show, on the one hand, that we have a contradiction if there is a v E VL
such that no member of V' is equivalent to v . Show, on the other
hand, if V' = {vt(v) l v E VL} contains v1 , v2 such that t(v1) 54 t(v2) ,
then since the product formula holds with respect to {vt(vl),v E VL} , we
obtain another V' with no member equivalent to v1 .
11
CHAPTER V. EXERCISES
1. When b > 0, d > 0, we have alb > c/d iff ad > be.
3. 1 > 0 because 12 = 1 . Consequently, every natural number > 0.
4. Cf. the proof of Theorem 4.7.5.
5. (i) xEK, aeL, a>x<0 implies xER.
(ii) (a + P) < (b + P) implies there is a d c L such that b - a > d > 0.
7. Use Exercise 5.1.6.
2
being the rational number field). Let K be an ordered field with an order
such that / < 0.
4
1. The condition (i) implies that the order of the Galois group G is a mul-
tiple of 3 and the condition (ii) implies that #(G) is an even number.
Since G is a subgroup of S3, we have G = S3 .
5
243
Subject Index
245
246 SUBJECT INDEX