Field Theory

Pete L. Clark
Thanks to Asvin Gothandaraman and David Krumm for pointing out errors in
these notes.
 Contents

About These Notes 7

Some Conventions 9
Chapter 1. Introduction to Fields 11
Chapter 2. Some Examples of Fields 13
1. Examples From Undergraduate Mathematics 13
2. Fields of Fractions 14
3. Fields of Functions 17
4. Completion 18
Chapter 3. Field Extensions 23
1. Introduction 23
2. Some Impossible Constructions 26
3. Subfields of Algebraic Numbers 27
4. Distinguished Classes 29
Chapter 4. Normal Extensions 31
1. Algebraically closed fields 31
2. Existence of algebraic closures 32
3. The Magic Mapping Theorem 35
4. Conjugates 36
5. Splitting Fields 37
6. Normal Extensions 37
7. The Extension Theorem 40
8. Isaacs’ Theorem 40
Chapter 5. Separable Algebraic Extensions 41
1. Separable Polynomials 41
2. Separable Algebraic Field Extensions 44
3. Purely Inseparable Extensions 46
4. Structural Results on Algebraic Extensions 47
Chapter 6. Norms, Traces and Discriminants 51
1. Dedekind’s Lemma on Linear Independence of Characters 51
2. The Characteristic Polynomial, the Trace and the Norm 51
3. The Trace Form and the Discriminant 54
Chapter 7. The Primitive Element Theorem 57
1. The Alon-Tarsi Lemma 57
2. The Primitive Element Theorem and its Corollary 57
3
4 CONTENTS

Chapter 8. Galois Extensions 61

1. Introduction 61
2. Finite Galois Extensions 63
3. An Abstract Galois Correspondence 65
4. The Finite Galois Correspondence 68
5. The Normal Basis Theorem 70
6. Hilbert’s Theorem 90 72
7. Infinite Algebraic Galois Theory 74
8. A Characterization of Normal Extensions 75

Chapter 9. Solvable Extensions 77

1. Cyclotomic Extensions 77
2. Cyclic Extensions I: Kummer Theory 81
3. The equation tn − a = 0 84
4. Cyclic Extensions II: Artin-Schreier Theory 88
5. Cyclic Extensions III: Witt’s Theory 89
6. Abelian Extensions of Exponent n: More Kummer Theory 89
7. Solvable Extensions I: Simple Solvable Extensions 89
8. Solvable Extensions II: Solvability by Radicals 89

Chapter 10. Computing Galois Groups 91

Chapter 11. Structure of Transcendental Extensions 93

1. Rational Function Fields 93
2. Transcendence Bases and Transcendence Degree 97
3. Applications to Algebraically Closed Fields 99
4. An Axiomatic Approach to Independence 100
5. More on Transcendence Degrees 105
6. Digging Holes 106

Chapter 12. Linear Disjointness 109

1. Definition and First Properties 109
2. Intrinsic Nature of Linear Disjointness 112
3. Linear Disjointness and Normality 114
4. Linear Disjointness and Separability 114
5. Interlude 118
6. Regular Extensions 118

Chapter 13. Derivations and Differentials 121

1. Derivations 121
2. Kähler Differentials 133
3. Applications to One Variable Function Fields 135
4. p-Bases 136

Chapter 14. Ordered Fields 141

1. Ordered Abelian Groups 141
2. Introducing Ordered Fields 144
3. Extensions of Formally Real Fields 147
4. The Grand Artin-Schreier Theorem 150
5. Sign Changing in Ordered Fields 154
 CONTENTS 5

6. Real Closures 155

7. Artin-Lang and Hilbert 156
8. Archimedean and Complete Fields 159
9. The Real Spectrum 163
Bibliography 165
 About These Notes

The purpose of these notes is to give a treatment of the theory of fields. Some
aspects of field theory are popular in algebra courses at the undergraduate or grad-
uate levels, especially the theory of finite degree field extensions and Galois theory.
However, a student of algebra (and many other branches of mathematics which use
algebra in a nontrivial way, e.g. algebraic topology or complex manifold theory)
inevitably finds that there is more to field theory than one learns in one’s stan-
dard “survey” algebra courses.1 When teaching graduate courses in algebra and
arithmetic/algebraic geometry, I often find myself “reminding” students of field-
theoretic facts that they have not seen before, or at any rate not in the form I wish
to use them.
I also wish to fill in some gaps in my own knowledge. Especially, I have long
wished to gain a deeper understanding of positive characteristic algebraic geometry,
and it has become clear that the place to begin understanding the “pathologies”2
of algebraic geometry in characteristic p is the study of finitely generated field ex-
tensions in positive characteristic.

These notes are meant to be comprehensible to students who have taken a basic
graduate course in algebra. In theory one could get away with less – the exposi-
tion is mostly self-contained. As algebraic prerequisites we require a good working
knowledge of linear algebra, including tensor products. The reader should also be
comfortable with – and fond of – groups and rings. Such a benevolent familiarity
is used much more than any specific results of group or ring theory. Our approach
is sufficiently abstract and streamlined that it is probably inappropriate for most
undergraduates. In particular, more often than not our approach proceeds from
the general to the specific, and we make no apologies for this.

1I make no claim that this phenomenon is unique to field theory.

2The term was used by Mumford [Mu61], [Mu62], [Mu67], [Mu75], with evident affection.

7
 Some Conventions

By convention, all of our rings are associative and have a multiplicative unity,
called 1. Again by convention, a homomorphism of rings necessarily carries 1 to 1.

These notes contain many exercises, including some which ask for proofs of stated
results. In general I am not at all opposed to the idea of a text giving complete
details for all of its arguments.3 However, it is in the nature of this particular
subject that there are many more results than proof techniques, to the extent that
giving complete proofs of all results would create a lengthy repetitiveness that may
discourage the reader to read the proofs that we do give.
As a rule, exercises that ask for proofs of stated results are meant to require no
new ideas beyond what was (even recently) exposed in the text. A reader who feels
otherwise should contact me: there may be an unintended gap in the exposition.
On the other hand, if exercises are given at all, it certainly spruces things up to
have some more challenging and interesting exercises. I have also not hesitated to
give exercises which can in principle be solved using the material up to that point
but become much easier after later techniques are learned.
At some point I fell victim to the disease of not liking the look of a paragraph
in which only a few words appear on the last line. Because of this, in the exercises I
have sometimes omitted the words “Show that”. I hope the meaning remains clear.

3In fact I agree with Robert Ash that the prevailing negative reputation of such texts is
undeserved: the royal road to a particular destination may or may not exist, but it seems perverse
to claim that it ought not to exist.

9
 CHAPTER 1

Introduction to Fields

A field is a commutative ring in which each nonzero element has a multiplicative

inverse. Equivalently, a field is a commutative ring R in which the only ideals are
(0) and R itself.
So if F is a field, S is a ring, and ϕ : F → S is a homomorphism of rings,
then since the kernel of ϕ is an ideal of F , ϕ is either injective (if its kernel is 0) or
identically the zero map (if its kernel is F ). Moreover, the latter case implies that
1S = ϕ(1F ) = 0, which happens iff S is the zero ring. So any homomorphism from
a field into a nonzero ring – in particular into any field or domain – is injective.
Thus if ϕ : F → K is a homomorphism between fields, we may equally well speak
of the field embedding ϕ.

Variations on the definition: In older terminology, a field could be non-commutative,

i.e., any ring in which each nonzero element has a two-sided multiplicative inverse.
We now call such things “division rings” or “division algebras.” One also sometimes
encounters non-associative division algebras, e.g. Cayley’s octonions.

The two branches of mathematics in which general fields play a principal role are
field theory (of course) and linear algebra. Most of linear algebra could be devel-
oped over a general division algebra rather than over a general field. In fact for the
most part the theory is so similar that it is not really necessary to consider division
algebras from the outset: one can just check, if necessary, that a certain result which
is true for vectors spaces over a field is also true for left modules over a division
algebra. On the other hand, when one studies things like roots of polynomials and
lattices of finite degree extensions, one immediately finds that non-commutative
division algebras behave in quite different and apparently more complicated ways.

Example 1.1. There are exactly two complex numbers z such that z 2 = −1:
z = i and z = −i. In general, any nonzero polynomial P (t) with coefficients in
a field can have no more solutions than its degree. But in Hamilton’s quaternion
algebra H there are clearly at least three solutions: i2 = j 2 = k 2 = −1, and in
fact there are uncountably many: a quaternion squares to −1 iff it is of the form
xi + yj + zk with x2 + y 2 + z 2 = 1.

Example 1.2. Let K/Q be a quartic field (i.e., a field extension of Q which has
dimension 4 as a Q-vector space). Then there are at most three intermediate sub-
fields Q ( F ( K. (More precisely there is either zero, one or three such fields, and
the first case happens “most of the time.”) However, any noncommutative division
algebra B/Q of degree 4 as a Q-vector space has infinitely many nonisomorphic
quadratic subfields.
11
12 1. INTRODUCTION TO FIELDS

The study of division algebras is closely related to field theory – via Brauer groups
and Galois cohomology – so that one can put one’s understanding of a field F
and its finite degree extensions to excellent use in studying noncommutative divi-
sion algebras over F . In fact, notwithstanding the above two examples, the finite
dimensional, central division algebra over a field F are significantly easier to under-
stand than finite dimensional extension fields of F : e.g. we understand quaternion
algebras over Q far better than quartic number fields.
 CHAPTER 2

Some Examples of Fields

1. Examples From Undergraduate Mathematics

Example 2.1. First of all there is the field of real numbers R. One also en-
counters the complex numbers C = {a + bi | a, b ∈ R, i2 = −1} and the rational
numbers Q = { ab | a ∈ Z, b ∈ Z \ {0}}.
Example 2.2. For a prime p, the ring Fp = Z/pZ of integers modulo p is a
field. In fact it is enough to show that it is a domain, since any finite domain
must be a field: if a is a nonzero element of a finite domain, there must exist
0 < i < j such that ai = aj , and then by cancellation we get 1 = aj−i = aj−i−1 a.
To check that Fp is a domain, suppose that x, y are nonzero elements in Fp such
that 0 = xy. Equivalently, we have integers x, y not divisible by p but such that
p | xy. This contradicts the uniqueness of factorization of integers into primes,
i.e., the “Fundamental Theorem of Arithmetic.”
Nonexample 2.3. The ring of integers Z/nZ is not a field unless n is prime:
if n = n1 · n2 with n1 , n2 > 1, then (n1 (mod n)) · (n2 (mod n)) = 0 (mod n)
exhibits zero divisors.
Let us reflect on this a bit. Any subring of a domain is again a domain (if the larger
ring has no nonzero divisors of zero, neither does the smaller ring). In particular,
any subring of a field must be a domain. Suppose n ∈ Z+ is not prime. Then, since
Z/nZ is not a domain, no ring which contains Z/nZ as a subring can be a domain.
This leads to the concept of characteristic: a ring is said to have characteristic n if it
admits Z/nZ as a subring, and characteristic zero if it does not have characteristic
n for any positive integer n. Equivalently, a ring has characteristic n > 0 iff n is
the least positive integer such that adding 1 to itself n times in the ring yields 0,
and has characteristic zero if there is no such integer. We see therefore that any
domain – and in particular any field – must have characteristic 0 or characteristic
a prime number p.
Exercise 2.1. Let R be a finite ring. Show that R has finite characteristic,
and that the characteristic divides #R.
Example 2.4. Suppose there is a field F4 with four elements. Like all fields,
it has distinct elements 0 and 1. Moreover, by the preceding exercise, it must have
characteristic 2, so 1 + 1 = 0. This leaves two further elements unaccounted for:
x and y. The nonzero elements of any field form a group under multiplication, so
in this case the group would have order 3 and therefore be cyclic. In particular x
has order 3, hence so does x2 , so x2 is equal to neither 0, 1 or x, and therefore
x2 = y = x−1 and y 2 = x = y −1 . Also x + y cannot equal x or y; if x + y = 0,
then x = −y = y since −1 = 1 in F4 . Therefore we must have x + y = 1, i.e.,
13
14 2. SOME EXAMPLES OF FIELDS

y = x − 1 = x + 1 = x2 . We have thus uniquely worked out the addition and

multiplication table for our putative field of order four, and one can check directly
that all the field axioms are satisfied: there is, indeed, a field of order four. There is
a unique such field up to isomorphism. Finally, as suggested by our analysis above,
the map which fixes 0 and 1 and interchanges x and y is an automorphism of the
field. One can think of it as the map a ∈ F4 7→ a2 .
Nonexample 2.5. Suppose F is a field of order 6. By Exercise 2.1, F must
have characteristic 2 or characteristic 3. Suppose it has characteristic 2. Then, by
Sylow’s Theorem, there exists x ∈ (F, +) of order 3: 3x = 0. But also 2x = 0, so
x = 3x − 2x = 0, contradiction.
Exercise 2.2. Let F be a finite field. Show that #F cannot be divisible by two
distinct primes p, q. (Hint: suppose the characteristic is p. Then there exists a ∈ Z+
such that pa | #F, #F a #F
pa is divisible by a second prime q 6= p and gcd(p , pa ) = 1.
By elementary number theory – “Bézout’s Lemma” – there exist integers x, y such
that xpa + y #F
pa = 1. Now argue as above.)

Therefore the order of a finite field F must be a prime power pf . In particular, F

contains Z/pZ as its prime subring (i.e., the subring generated by one).
Exercise 2.3. Give a second proof that a finite field F must have prime power
order: as above, F contains a unique subfield Fp of prime order. Argue that F is a
finite-dimensional vector space over Fp of dimension f = logp #F.
Exercise 2.4. The next largest non-prime prime powers are 8 and 9. Try to
contruct finite fields of these orders from “first principles”, as we did with the case
of order 4 above.
We will see later that for every prime power pa there is a finite field F of order pf ,
that any two finite fields of order pf are isomorphic, and that the automorphism
group of a finite field of order pf is cyclic of order f , generated by the Frobenius
map x 7→ xp .

2. Fields of Fractions
If R is a domain, then one can define a field F whose elements are viewed as
fractions ab with a, b ∈ R, b 6= 0. Formally speaking one considers ordered pairs
(a, b) ∈ R2 , b 6= 0 and introduces the equivalence relation (a, b) ∼ (c, d) ⇐⇒
ad = bc, i.e., exactly the same construction that one uses to define the rational
numbers in terms of the integers. The field F is called the field of fractions, (or,
sometimes, “quotient field”) of the domain R.
Exercise 2.5. (Functoriality of the field of fractions) Let ϕ : R → S be an
injective homomorphism of domains. Show that ϕ extends uniquely to a homomor-
phism from the fraction field F (R) of R to the fraction field F (S) of S.
Exercise 2.6. (Universal property of the field of fractions) Let R be a domain
with fraction field F and let K be a field. For any injective homomorphism ϕ : R →
K, there exists a unique extension to a homomorphism F → K.
Exercise 2.7. Let R be a domain with field of fractions F (R). Show: #R =
#F (R).
 2. FIELDS OF FRACTIONS 15

Thus any method which produces a supply of domains will also produce a supply of
fields (of course distinct domains may have the same fraction field, a rather trivial
example being Z and Q itself; there are in fact uncountably many isomorphism
classes of domains with fraction field Q).
Proposition 2.6. If R is a domain, then the univariate polynomial ring R[t]
is also a domain. Moreover, if F is the fraction field of R, then the fraction field
of R[t] is F (t), the field of all quotients of polynomials with F -coefficients.
Exercise 2.8. Prove Proposition 2.6.

Example 2.7. Applying the Proposition with R = F a field, we get a field F (t)
of rational functions in F . E.g., the field C(t) is the field of meromorphic functions
on the Riemann sphere (see the next section). Moreover, for any field F , F [t] is a
domain, so F [t1 , t2 ] := F [t1 ][t2 ] is also a domain. The fraction field is easily seen
to be F (t1 , t2 ), i.e., the fraction field of F [t1 , . . . , tn ] is F (t1 , . . . , tn ) the field of
rational functions in n indeterminates.
Although successive applications of Proposition 2.6 will yield polynomial rings in
only finitely many indeterminates, nothing stops us from considering larger poly-
nomial rings: let T = {ti } be any set of indeterminates, and R any commutative
ring. One can consider the polynomial ring R[T], defined as the union (or, if you
like, direct limit) of polynomial rings R[S] where S ⊂ T is a finite subset. In other
words, we consider the ring of polynomials in an arbitrary infinite set S of inde-
terminates, but any given polynomial involves only finitely many indeterminates.
One can again show that if R is a domain, so is R[T]. The corresponding fraction
field R(T) is the field of all quotients of polynomials in all these indeterminates.
Exercise 2.9. Let F be a field and T a nonempty set of indeterminates. Show
that the cardinality of the rational function field F (T) is max(ℵ0 , #F, #T).
Another way of manufacturing domains is to start with a commutative ring R and
take the quotient by a prime ideal p. Then we can get a field by (if necessary, i.e.,
if p is not maximal) taking the field of fractions of R/p. For example with R = Z
we get the finite fields Fp .
Example 2.8. Let R = F [T ] and p a nonzero prime ideal. Then, since R is
a PID, p = (f (t)), where f (t) is an irreducible polynomial. Moreover, assuming
f (t) 6= 0, p is maximal, so without having to take quotients we get a field
K = F [t]/(f (t)),
whose dimension as an F -algebra is the degree of f .
A domain R is finitely generated (over Z) if there exist n ∈ Z+ and elements
α1 , . . . , αr ∈ R such that the least subring of R containing all the αi ’s is R itself.
Another way of saying this is that the natural map
Z[T1 , . . . , Tn ] → R, Ti 7→ αi
is surjective. In other words, a domain is finitely generated iff it is, for some n, the
quotient of the ring Z[T1 , . . . , Tn ] by some prime ideal p.
Proposition 2.9. For a field F , the following are equivalent:
a) There are α1 , . . . , αn ∈ F so that the only subfield of F containing all the
αi ’s is F itself.
16 2. SOME EXAMPLES OF FIELDS

b) The field F is the fraction field of Z[x1 , . . . , xn ]/p for some prime ideal p.
Exercise 2.10. Prove Proposition 2.9.
A field satisfying the equivalent conditions of Proposition 2.9 is said to be finitely
generated. Applying part b) and Exercise 2.7 we see that any finitely generated
field is finite or countably infinite. In particular the fields R, C are not finitely
generated. Conversely, a countable field need not be finitely generated: if T is
a countably infinite set of indeterminates, then by Exercise 2.9 the field Q(T) is
countable. Moreover it is both plausible and true that Q(T) is not finitely gener-
ated, but we lack the tools to prove this at the moment: we will return to this later
on in the context of the concept of transcendence degree.

One can also speak of finite generation in a relative sense:

Proposition 2.10. For a subfield F ⊂ K, the following are equiva-
lent:
a) There exist elements α1 , . . . , αn ∈ K such that the only subfield of K
containing F and the αi ’s is K itself.
b) The field K is isomorphic to the fraction field of F [x1 , . . . , xn ]/p for some
prime ideal p.
Exercise 2.11. Prove Proposition 2.10.
If F is a subfield of K and α1 , . . . , αn ∈ K, we write F (α1 , . . . , αn ) for the small-
est subfield of K containing F and the αi ’s. The notation is sensible because this
P (α1 ,...,αn )
field can be described concretely as the set of all rational expressions Q(α 1 ,...,αn )
for P, Q ∈ k[t1 , . . . , tn ]. (In particular there is a unique such smallest subfield.)

So for instance one can speak of fields which are finitely generated over the complex
numbers C, and such fields are especially important in algebraic geometry.
Proposition 2.11. Let F be a field.
a) If F has characteristic 0, there is a unique homomorphism ι : Q → F .
b) If F has characteristic p, there is a unique homomorphism ι : Fp → F .
Proof. For any ring R, there exists a unique ring homomorphism ι : Z → R,
which takes the integer n to n times the multiplicative identity in R. For R = F a
field, the map ι is an injection iff F has characteristic 0. So if F has characteristic 0,
ι is injective, and by Exercise 2.5 it extends uniquely to a homomorphism ι : Q → F .
Any homomorphism from Q to F must restrict to the canonical injection on Z and
therefore be ι. If F has characteristic p > 0, then ι factors through to give a map
ι : Fp → F . The uniqueness of ι can be seen in any number of ways: we leave it to
the reader to find an explanation that she finds simple and convincing. 
It follows that Q (resp. Fp ) is the unique minimal subfield of any field F of charac-
teristic 0 (resp. p > 0). We refer to Q (resp. Fp ) as the prime subfield of F . Note
that since there are no nontrivial automorphisms of either Q or Fp (this follows by
applying the proposition with F = Q or F = Fp ), the prime subfield sits inside F
in an especially canonical way.
Exercise 2.12. Let K be a field and k its prime subfield. Show that the concepts
“K is finitely generated” and “K is finitely generated over k” coincide.
 3. FIELDS OF FUNCTIONS 17

Exercise 2.13. For any field F , there exists a set of indeterminates T and a
prime ideal p of Z[T] such that F is isomorphic to the fraction field of Z[T]/p.
If F is infinitely generated (i.e., not finitely generated over its prime subfield) then
the set T in Exercise 2.13 will of course have to be infinite. In such a case this
“presentation” of F is not, in truth, so useful: e.g., with certain limited exceptions
(to be discussed!) this is not a very insightful way of viewing the complex field C.
Exercise 2.14. Let R be a commutative ring, ι : R → S an injective ring
homomorphism, and α ∈ S. Show that there is a unique minimal subring of S
containing R and α, namely the set of all polynomials P (α), P ∈ R[t]. This subring
is accordingly denoted R[α].

3. Fields of Functions
Let U be a domain – i.e., a nonempty connected open subset – of the complex plane.
In complex analysis one studies the set Hol(U ) of all functions holomorphic (a.k.a.
analytic) on all of U and also the larger set Mer(U ) of all meromorphic functions
on U , i.e., functions which are holomorphic on the complement of a discrete set
X = {xi } and such that for each xi there exists a positive integer ni such that
z ni f is holomorphic at xi . Under the usual pointwise addition and multiplication
of functions, Hol(U ) is a ring (a subring of the ring of all continuous C-valued
functions on U ). Similarly, one can view Mer(U ) as a ring in a natural way.
Theorem 2.12. Let U be a domain in the complex plane. Then:
a) The ring Hol(U ) is a domain.
b) The ring Mer(U ) is a field.
c) The field Mer(U ) is the field of fractions of Hol(U ).
Proof. a) A consequence of the principle of analytic continuation is that the
zero set of a not-identically-zero holomorphic function is discrete in U . For 0 6=
f, g ∈ Hol(U ), the zero set of f g is the union of the zero sets of f and g so is again
discrete and thus certainly a proper subset of U .
b) Because 0 6= f ∈ Hol(U ) has a discrete zero set {xi } and for each xi , there exists
a positive integer ni such that zfni extends to a continuous nonzero function at xi ,
it follows that f1i is meromorphic.
c) This lies deeper: it is a consequence of Weierstrass’ factorization theory, in
particular of the fact that for any discrete subset X = {xi } of U and any sequence
of positive integers {ni } there exists a holomorphic function on U with zero set X
and order of vanishing ni at xi . 
Exercise 2.15. Show: the field Mer(C) is not finitely generated over C.
More generally, if M is a connected complex manifold, there is a ring Hol(M ) of
“global” holomorphic functions on M and a field Mer(M ) of meromorphic functions.
It need not be the case that Mer(M ) is the fraction field of Hol(M ).
Example 2.13. Take M = C ∪ {∞} to be the Riemann sphere. Then the only
holomorphic functions on M are the constant functions, whereas Mer(M ) = C(z),
the rational functions in z.
In various branches of geometry one meets many such “fields of functions”: a very
general example, for the highly trained reader, is that if X is an integral (reduced
18 2. SOME EXAMPLES OF FIELDS

and irreducible) scheme, then the ring of all functions regular at the generic point
η is a field. If X itself is a scheme over a field k, then this field is written k(X) and
called the field of rational functions on X. For example, the field of rational
functions on the complex projective line P1 /C is the rational function field C(t).
This is essentially the same example as the Riemann sphere above, but couched in
more algebraic language.

In general, one must restrict to functions of a rather special kind in order to get a
field of functions. Using the ideas of the previous subsection, it seems fruitful to
first consider rings R of functions on a topological space X. Then we want R to be
a domain in order to speak of fraction field F (R) of “meromorphic functions” on X.

Suppose X is a topological space and consider the ring R = R(X, C) of all continu-
ous functions f : X → C. A moment’s thought indicates that for the “reasonable”
topological spaces one considers in geometry, R will not be a domain. The question
comes down to: do there exist functions f1 , f2 : X → C neither of which is zero on
all of X but such that the product f1 · f2 is identically zero?

Here are some easy observations. First, if X is not connected, the answer is cer-
tainly yes: write X = Y1 ∪ Y2 where Yi are disjoint open sets. Take f1 to be the
characteristic function of Y1 an f2 = 1 − f1 to be the characteristic function of Y2 .

In fact R is not a domain even if X is the Euclidean plane: let D1 , D2 be two

disjoint closed disks, say with centers zi and radii equal to 1. Certainly there exist
continuous functions fi : X → C such that fi (zi ) = 1 and fi (z) = 0 if z lies outside
of Di . Indeed it is well-known that fi may be chosen to be infinitely differentiable,
and the argument generalizes to all manifolds and indeed to paracompact Hausdorff
spaces (the key point being the existence of suitable partitions of unity).

On the other hand, suppose the space X is irreducible: that is, if Y1 , Y2 are two
proper closed subsets of X then Y1 ∪ Y2 6= X. Then, applying this to Yi = fi−1 (0),
we get that the zero set of f1 f2 is Y1 ∪ Y2 6= X, so R(X, C) is a domain, and one
can take its fraction field, which consists of functions which are defined on some
dense (equivalently, nonempty!) open subset of X. If you have never studied al-
gebraic geometry, you will doubtless be thinking, “What kind of crazy topological
space would be irreducible?” However, the Zariski topology on a smooth, connected
algebraic variety over (say) the complex field C is irreducible.

4. Completion
None of the constructions of fields we have discussed so far give rise to either R or
C in a reasonable way. These fields are uncountable, and from a purely algebraic
perspective their structure is quite complicated. The right way to think about them
is via a mixture of algebra and topology, e.g. one thinks of R as the completion of
the field of rational numbers with respect to the standard absolute value.

An absolute value on a field K is a real-valued function x → ||x|| satisfying:

(AV1) ||x|| ≥ 0 for all x ∈ K, with equality iff x = 0.

 4. COMPLETION 19

(AV2) ||xy|| = ||x||||y|| for all x, y ∈ K.

(AV3) ||x + y|| ≤ ||x|| + ||y|| for all x, y ∈ K.

It is immediate that an absolute value gives rise to a metric on K, via d(x, y) =

||x − y||. We can therefore complete the metric space to get a metric space K̂ with
a canonically embedded, dense copy of K. The key point is that K̂ also has a
canonical field structure.

In brief, we consider the set C of Cauchy sequences in K. This becomes a ring

under the operations of pointwise addition and multiplication. (It is far from being
a domain, having many zero divisors and idempotent elements.) Inside this ring we
have c, the collection of sequences converging to 0. It is not hard to check that c is
in fact an ideal of C, so that we may form the quotient C/c. Best of all, this quotient
ring is a field: indeed, a nonzero element of the quotient may be represented by
a Cauchy sequence x• in K which does not converge to 0. It follows that there
are only finitely many indices n such that xn = 0: otherwise a subsequence of x•
converges to 0 and a Cauchy sequence with a convergent subsequence is itself con-
vergent to the same limit as the subsequence. Consider then the sequence y• which
is defined by yn = xn if xn = 0 and yn = x−1 n otherwise. The product sequence
x• y• has all sufficiently large terms equal to 1, so differs from the constant sequence
1 (the identity element of C) by a sequence which has only finitely many nonzero
terms, so in particular lies in c. Therefore the class of y• in C/c is the inverse of x• .

We denote C/c by K̂ and call it the completion of K with respect to || ||. There
exists a natural embedding K ,→ K̂ – namely we map each element of K to the
corresponding constant sequence – and a natural extension of the norm on K to
a norm on K̂, namely ||x• || = limn→∞ ||xn ||, with respect to which ι : K ,→ K̂ is
an isometry of normed spaces in which the image of K in K̂ is dense. For more
details, the reader is invited to consult [Cl-NTII, Chapter 2].
Example 2.14. The completion of Q with the standard Archimedean absolute
value || pq || = | pq | is the real field R.
Remark 2.1. It is sometimes suggested that there is a circularity in this con-
struction, in that the definition of completion refers to a metric and the definition
of a metric refers to the real numbers.1 But one should not worry about this. On the
one hand, from our present point of view we can consider the reals as being already
constructed and then it is a true, non-tautologous statement that the metric com-
pletion of the rationals is the reals. But moreover, a careful look at the construction
in terms of equivalence classes of Cauchy sequences shows that one absolutely can
construct the real numbers in this way, just by being careful to avoid referring to
the real numbers in the course of the completion process. In other words, the real
numbers can be defined as the quotient of the ring of Cauchy sequences of rational
numbers (where the definition of Cauchy sequence uses only the metric as defined
on rational numbers) by the maximal ideal of sequences converging to zero. After
one constructs the real numbers in this way, one notes that the Q-valued metric on
Q extends to an R-valued metric on R: no problem.
1In particular, Bourbaki’s General Topology refrains from making any reference to real num-
bers or metric spaces for many hundreds of pages until the reals can be rigorously constructed.
20 2. SOME EXAMPLES OF FIELDS

Example 2.15. If k is any field, then defining ||0|| = 0 and ||x|| = 1 for all
x 6= 0 gives an absolute value on k. The induced metric is the discrete metric
and therefore k is, in a trivial way, complete and locally compact. This absolute
value (and any other absolute value inducing the discrete topology) is called triv-
ial; such absolute values are usually either explicitly or implicitly excluded from
consideration.
Example 2.16. || ab || = pordp (b)−ordp (a) , where for an integer a, ordp (a) denotes
the largest power of p dividing a. (To get the degenerate cases to work out correctly,
we set ordp (0) = ∞ and p−∞ = 0.) The induced metric on Q is called the p-adic
metric: in this metric, a number is close to zero if, after cancelling common factors,
its numerator is divisible by a high power of p. Since the induced topology has no
isolated points, the completeness of the metric would contradict the Baire category
theorem, hence the completion is an uncountable field, called Qp , the field of p-adic
numbers.
Example 2.17. Let k be any field and K = k(t). Any element r(t) ∈ K can be
P (t)
written as ta Q(t) where P (0)Q(0) 6= 0 for a uniquely determined integer a. Define
||r(t)||∞ := e−a . (There is no particular reason to use the number e = 2.718 . . .;
any real number greater than 1 would serve as well.)
Exercise 2.16. Show: || ||∞ gives an absolute value on K(t).
An element r(t) ∈ K(t) is close to 0 iff it is divisible by a high power of t.
Exercise 2.17. Show: the completion of K(t) with respect to || ||∞ is isomor-
phic
P∞ to thenLaurent series field K((t)), whose elements are formal power series
n=n0 an t with n0 ∈ Z, an ∈ f . (Hint: It is enough to show that the norm || ||∞
extends to all of K((t)) and that K(t) is dense in K((t)) in the induced topology.)
Exercise 2.18. Show: the fields Qp are locally compact in their natural topol-
ogy. Show: K((t)) is locally compact iff K is finite.
Remark 2.2. If k is a field complete with respect to an absolute value | | and
V is a finite-dimensional vector space over k, then viewing V ∼ = k dim V gives V the
canonical structure of a topological space – i.e., we can endow it with the product
topology, and this topology is independent of the choice of basis. In particular, if k
is locally compact, so is V . Moreover it has the canonical structure of a uniform
space, and if k is complete then so is V . In particular, if k ,→ l is a field embedding
such that l is finite-dimensional as a k-vector space, then l is a complete uniform
space and is locally compact iff k is. This implies that any finite degree extension
of the fields R, Qp or Fp ((t)) have a canonical locally compact topology.
Theorem 2.18. (Classification of locally compact valued fields) Let || || be a
nontrivial valuation on a field K. The following are equivalent:
(i) The metric topology on K is locally compact.
(ii) Either (K, || ||) = R or C; or the induced metric is complete and non-
Archimedean and the residue field is finite.
(iii) K is a finite degree extension of R, of Qp or of Fp ((t)).
Proof. See [Cl-NTII, Theorem 5.1]. 
There are more elaborate ways to construct complete fields. For instance, suppose
R is a domain and p is a prime ideal of R. Then in commutative algebra one learns
 4. COMPLETION 21

how to complete R with respect to p, getting a homomorphism R → R̂ in which R̂

is a domain, the image pR̂ is the unique maximal ideal of R̂, and R̂ is complete with
respect to a canonical uniform structure. We can then take the fraction field to get
a complete field K̂. Let us just mention one simple example to give the flavor: let
f be a field and R = f [x1 , . . . , xn ] and p = (x1 , . . . , xn ). Then the completion is
R̂ = f [[x1 , . . . , xn ]], the ring of formal power series in the indeterminates x1 , . . . , xn ,
and its quotient ifeld is f ((x1 , . . . , xn )), the field
P of formal Laurent series in these
indeterminates, i.e., the set of all formal sums I aI xI where I = (i1 , . . . , in ) ∈ Zn
is a multi-index, aI ∈ k, xI = xi1 · · · xin , and the set of indices I in which at least
one ij is negative and aI 6= 0 is finite.

Such fields arise in algebraic and analytic geometry: C((x1 , . . . , xn )) is the field
of germs of meromorphic functions at a nonsingular point P on an n-dimensional
analytic or algebraic variety.
Exercise 2.19. Show: the field k((x1 , x2 )) is properly contained in k((x1 ))((x2 )).
 CHAPTER 3

Field Extensions

1. Introduction
Let K be a field. If ι : K → L is a homomorphism of fields, one says that L is an
extension field of K. As a matter of psychology, it often seems more convenient
to think of L as “lying above K” rather than K as embedding into L. We often
write L/K instead of ι : K → L, notwithstanding the fact that the latter notation
hides important information, namely the map ι.1

Much of field theory is devoted to an understanding of the various extension fields

of a given field K. Since any field K has extensions of all sufficiently large cardi-
nalities – K(T) for any large enough set T – one obviously cannot literally hope
to understand all field extensions of K. However there are two important classes
(sets!) of field extensions that one can at least hope to understand: the first is the
class of all finitely generated field extensions of K, and the second is the class of
all algebraic field extensions of K.

If L/K is a field extension, then L is a K-algebra and in particular a vector space

over K. Therefore it has a well-determined (but possibly infinite) dimension, de-
noted by [L : K]. One says that the extension L/K is finite if [L : K] < ∞, i.e., if
L is a finite-dimensional K-vector space. For instance, one has [C : R] = 2 < ∞,
so C/R is a finite degree field extension.

Warning: The term “finite field extension” is ambiguous: it could presumably

also refer to an extension of fields L/K in which L and K are both finite fields. In
practice, one should expect the term to have the former meaning – i.e., the finite-
ness refers to the degree of the extension, and not to either field – but be prepared
to seek clarification if necessary.

As an immediate application we can rederive the fact that the order of a finite
field is necessarily a prime power. Namely, let F be a finite field, and let Fp be
its prime subfield. Since F is finite, it is certainly finite-dimensional over Fp (any
infinite dimensional vector space over any field is infinite), say of dimension d. Then
F as an Fp -vector space is isomorphic to Fdp , so its cardinality is pd .
Theorem 3.1. (Degree multiplicativity in towers) Let F ⊂ K ⊂ M be field
extensions. Then we have
[M : F ] = [M : K][K : F ].

1Beware: the notation L/K has nothing to do with cosets or quotients!

23
24 3. FIELD EXTENSIONS

Proof. Let {bi }i∈I be an F -basis for K and {aj }j∈J be a K-basis for M . We
claim that {ai bj }(i,j)∈I×J is an F -basis for M . This suffices, since then [K : F ] =
#I, [M : K] = #J, [M : F ] = #(I × J) = #I × #J.
Let c ∈ M .PThen there exist αj ∈ K, all but finitely many of which are zero,
such that c = j∈J αj aj . Similarly, for each j ∈ J, there exist βij ∈ F , all but
P
finitely many of which are zero, such that αj = i,j βij bj , and thus
X X
c= αj aj = βij ai bj ,
j∈J (i,j)∈I×J

so that {ai bj } spans K as an F -vector space. Now suppose the set {ai bj } were
linearly dependent. By definition, this means that there is some finite subset S ⊂
I × J such that {ai bj }(i,j)∈S is linearly dependent, and thus there exist βij ∈ F ,
not all zero, such that X
(βi jbj )ai = 0.
(i,j)∈S
Since
P the ai ’s are K-linearly independent elements of M , we have that for all i,
βij bj = 0, and then similarly, since the bj ’s are linearly independent elements of
K we have βij = 0 for all j. 
Remark 3.1. In general the degree [L : K] of a field extension is a cardinal
number, and the statement of Theorem 3.1 is to be interpreted as an identity of
(possibly infinite) cardinals. On the other hand, when M/K and K/F are finite,
the argument shows that M/F is finite and the result reduces to the usual product
of positive integers. Moreover the finite case is the one that is most useful.
Let L/K be an extension of fields and α ∈ L. We say that α is algebraic over K
if there exists some polynomial P (t) = tn + an−1 tn + . . . + a1 t + a0 ∈ K[t] such that
P (α) = 0. If α is not algebraic over K it is said to be transcendental over K. A
complex number which is algebraic over Q is called an algebraic number.
Example 3.2. The element i is algebraic over R since it satisfies the equation
1
i2 +1 = 0. It is also algebraic over Q for the same reason. Indeed for any a ∈ Q, a n
1
is algebraic over Q. This is almost tautological, since by a n , one generally means
any complex number α such that αn = a, so α satisfies tn − a = 0.
The following exercise gives less trivial examples.
a
Exercise 3.1. Let b ∈ Q. Show cos( ab π) and sin( ab π) are algebraic.
Exercise 3.2. a) Show that the set of all algebraic numbers is countably infi-
nite.
b) More generally, let K be any infinite field and L/K be any field extension. Show
that the cardinality of the set of elements of L which are algebraic over K is equal
to the cardinality of K.
So “most” real or complex numbers are transcendental. This was observed by Can-
tor and stands as a famous early application of the dichotomy between countable
and uncountableP∞sets. Earlier Liouville had constructed particular transcendental
numbers, like n=1 10−n! : an application of the Mean Value Theorem shows that a
number which is “too well approximated” by rational numbers cannot be algebraic.
It is of course a different matter entirely to decide whether a particular, not obvi-
ously algebraic, number which is given to you is transcendental. Let us say only
 1. INTRODUCTION 25

that both e and π were shown to be transcendental in the 19th century; that there
were some
√
interesting results in transcendence theory in the 20th century – e.g.
eπ and 2 2 are transcendental – and that to P this day the transcendence of many
∞
reasonable looking constants – e.g. π e , ζ(3) = n=1 n13 – is much beyond our reach.

The problem of determining whether particular numbers are transcendental, al-

though certainly of interest, has little to do with modern field theory. (Rather it is
part of the theory of Diophantine approximation, a branch of number theory.)
Exercise 3.3. (Universal property of polynomial rings): Let ι : R → S be a
homomorphism of commutative rings, and let α1 , . . . , αn be elements of S. There
is a unique R-algebra homomorphism Φ : R[t1 , . . . , tn ] → S which takes ti 7→ αi .
Now let L/K be a field extension and α ∈ L. By Exercise 3.1.6 there is a unique
K-algebra homomorphism Φ : K[t] → L, t 7→ α. Let I be the kernel of Φ. Since
K[t]/I embeds in L, it is a domain, so I is a prime ideal. Since K[t] is a principal
ideal domain, there are only two choices:

Case 1: I = 0, i.e., Φ embeds K[t] into L. This means precisely that α satisfies
no polynomial relations with K-coefficients, so occurs iff α is transcendental over K.

Case 2: I = (P (t)) is generated by a single irreducible polynomial P (t). Since

the units of K[t] are precisely the nonzero elements of K, it follows that there is a
unique monic polynomial P (t) (i.e., with leading coefficient 1) that generates I. We
call this the minimal polynomial of α. Evidently for Q ∈ K[t] we have Q(α) = 0
⇐⇒ P (t) | Q(t). In particular P (α) = 0, so that α is algebraic, and moreover
Φ induces an embedding K[t]/(P (t)) ,→ L. If P has degree d, then we say α is
algebraic of degree d; moreover, a K-basis for the left-hand side is 1, t, . . . , td−1 , so
[L : K] = d = deg(P ).

Let us summarize:
Theorem 3.3. Let L/K be a field extension and α ∈ L.
a) The following are equivalent:
(i) The element α is algebraic of degree d over K.
(ii) The K-vector space K[α] is finite, of degree d.
(iii) The K-vector space K(α) is finite, of degree d.
b) If α is algebraic of degree d, then K[α] = K(α) ∼ = K[t]/(P (t)), where
P (t) ∈ K[t] is the unique monic polynomial of degree d such that P (α) = 0.
c) If α is transcendental over K, then K[t] ∼= K[α] ( K(α) ∼ = K(t).
P (π)
It follows that the set of all rational expressions Q(π) with P, Q ∈ Q[t] is isomorphic
to the rational function field Q(t)! In other words, there is no genuinely algebraic
distinction to be made between “fields of numbers” and “fields of functions.”

A field extension L/K is algebraic if every α ∈ L is algebraic over K.

Corollary 3.4. A finite degree extension L/K of fields is algebraic.
Proof. We go by contraposition: suppose that L/K is transcendental, and
let α ∈ L be transcendental over K. Then by Theorem 3.3c) we have
[K(α) : K] ≥ [K[α] : K] = [K[t] : K] = ℵ0 ,
26 3. FIELD EXTENSIONS

so
[L : K] = [L : K(α)][K(α) : K] ≥ ℵ0 . 
The converse does not hold: many fields admit infinite algebraic extensions. A
detailed analysis of algebraic field extensions is still ahead of us, but it is easy to
S 1
see that the extension Q[ n≥2 2 n ] is an infinite algebraic extension, since it contains
subextensions of arbitrarily large finite degree.
Exercise 3.4. (Direct Limits) Let (I, ≤) be a directed set: recall that this
means that I is partially ordered under ≤ and for any i, j ∈ I there exists k ∈ I
with i ≤ k and j ≤ k. A directed system of sets is a family of sets {Xi }i∈I
together with maps ι(i, j) : Xi → Xj for all i ≤ j satisfying the natural compatibility
conditions: (i) ιi,i = 1Xi and (ii) for all i ≤ j ≤ k, ι(i, k) = ι(j, k) ◦` ι(i, j). By
definition, the direct limit limI X is the quotient of the disjoint union i∈I Xi by
the equivalence relation (x, Xi ) ∼ (ι(i, j)x, Xj ) for all i ≤ j.
a) Show: there are natural maps ιi : Xi → limI Xi . State and prove a
universal mapping property for the direct limit.
b) Suppose the maps ι(i, j) are all injective. Show that the maps ιi : Xi →
limI Xi are all injective. Explain why in this case limI Xi is often infor-
mally referred to as the “union” of the Xi ’s.
c) In any concrete category C – i.e., a category whose objects are sets, for
which the set of all morphisms from an object A to an object B is a subset
of the set of all functions from A to B, and for which composition and
identity of morphisms coincide with the usual notions of functions – one
has the notion of a directed system {Ai } of objects in C, i.e., we have
sets Ai indexed by the directed set (I, ≤) and for all i ≤ j, the function
ι(i, j) : Ai → Aj is a morphism in C. Give a definition of the direct limit
limI Ai in this more general context. Show that the direct limit exists
in the following categories: monoids, groups, commutative groups, rings,
commutative rings, fields.
d) Give an example of a concrete category in which directed limits do not
necessarily exist.2
e) Show that a field extension L/K is algebraic iff it is the direct limit of its
finite subextensions.

2. Some Impossible Constructions

The results we have derived so far do not look very deep to modern eyes, but
they were recognized in the 19th century to imply negative solutions to several
of the longest standing open problems in mathematics. Namely, the Greeks were
interested in constructibility of quantities using a compass and a straightedge.
We recall the basic setup: one starts out with two distinct points in the plane,
which we may as well view as being a unit distance apart. We have at our disposal
an unmarked straightedge, so that given any two points we may construct the line
passing through them, and a compass, such that given any previously constructed
point P1 and any previously constructed pair of points P2 , P3 , we may draw a
circle whose center is P1 and whose radius is the distance between P2 and P3 .
Let us say that a positive real number α is constructible if we can after a finite

2Suggestion: impose some finiteness condition on one of the above categories.

 3. SUBFIELDS OF ALGEBRAIC NUMBERS 27

sequence of steps construct points P, P 0 with distance α (more precisely, α times

the unit distance we started with), and let us agree that a negative number α is
constructible iff |α| is constructible. Despite the severely constrained toolkit, the
supply of constructible numbers is in some sense rather large.
Exercise 3.5. a) Show: the constructible√numbers form a subfield of R.
b) Show: if α > 0 is constructible, then so is α.
c) Conclude: the field of constructible numbers has infinite degree over Q.
Now let us look more closely: a constructible number is built up in a sequence
of steps: α1 = 1, α2 , . . . αn = α corresponding to a tower of fields F1 = Q,
F2 = F1 (α2 ), . . ., Fn = Fn−1 (αn ). To get from Fi to Fi+1 = Fi (αi ), we are either
intersecting two lines – which corresponds to solving a linear equation with coeffi-
cients in Fi−1 , so Fi = Fi−1 – or intersecting a line defined over Fn−1 with a circle
whose coefficients lie in Fi−1 which yields solutions in either Fi−1 or a quadratic
extension of Fi−1 – or we are intersecting two circles with equations defined over
Fi−1 , which leads to solutions over at worst a quadratic extension of a quadratic
extension of Fi−1 . (Note quadratic, not quartic: any two distinct circles intersect
in at most two points, and thus the common intersection can also be expressed as
the intersection of a line and a circle.)

Thus any constructible number α lies in a field which is at the top of a tower
of quadratic field extensions, so [Q(α) : Q] is a power of 2. The impossibility of
three classically sought after constructions follows easily.
First we cannot double the cube: given a cube with sides of our unit length,
we cannot construct a cube whose √ volume√is twice that of the given cube, because
the length of a side would be 3 2, and [Q( 3 2) : Q] = 3. Similarly we can construct
angles that we cannot trisect; in particular, we can construct
√
an angle of 60 degrees
o 1 o 3
(i.e., we can construct cos 60 = 2 and sin 60 = 2 ), but we cannot construct
cos 20o since it satisfies an irreducible cubic polynomial over Q. Finally, we cannot
square the circle i.e., construct a square whose √ area is that of a unit circle, for
that would involve constructing a side length of π and π is not even algebraic!

3. Subfields of Algebraic Numbers

Let L/K be an arbitrary extension of fields. Consider the set ClL (K) of all elements
of L which are algebraic over K. For example, when K = Q, L = C we are
examining the set of all algebraic numbers, which is certainly a proper subset of C.
Proposition 3.5. The set ClL (K) is a subfield of K.
We often refer to ClL (K) as the algebraic closure of K in L.

Let us this result in a more general context, that of integral extensions of do-
mains. The generalized proof is not much harder and will be extremely useful for
any student of algebra. So: let R be a domain and S a domain which extends R,
i.e., there is an injective homomorphism R → S. We say that α ∈ S is integral
over R if α satisfies a monic polynomial with R-coefficients:
∃ an−1 , . . . , a0 ∈ R | αn + an−1 αn−1 + . . . + a1 α + a0 = 0.
We say that the extension S/R is integral if every element of S is integral over R.
28 3. FIELD EXTENSIONS

Note that if R and S are fields, α ∈ S is integral over R is by definition pre-

cisely the same as being algebraic over R. The next result in fact revisits the basic
finiteness property of algebraic elements in this more general context.
Theorem 3.6. Let R ⊂ T be rings, and α ∈ T . The following are equivalent:
(i) The α is integral over R.
(ii) The ring R[α] is finitely generated as an R-module.
(iii) There is an intermediate ring R ⊂ S ⊂ T such that α ∈ S and S is finitely
generated as an R-module.
(iv) There exists a faithful R[α]-submodule M of T which is finitely generated
as an R-module.
Proof. (i) =⇒ (ii): If α is integral over R, there are a0 , . . . , an−1 ∈ R such
that
αn + an−1 αn−1 + . . . + a1 α + a0 = 0,
or equivalently
αn = −an−1 αn−1 − . . . − a1 α − a0 .
This relation allows us to rewrite any element of R[α] as a polynomial of degree at
most n − 1, so that 1, α, . . . , αn−1 generates R[α] as an R-module.
(ii) =⇒ (iii): Take T = R[α].
(iii) =⇒ (iv): Take M = S.
(iv) =⇒ (i): Let m1 , . . . , mn be a finite set of generators for M over R, and express
each of the elements mi α in terms of these generators:
Xn
αmi = rij mj , rij ∈ R.
j=1

Let A be the n × n matrix αIn − (rij ); then recall from linear algebra that
AA∗ = det(A) · In ,
where A∗ is the “adjugate” matrix (of cofactors). If m = (m1 , . . . , mn ) (the row
vector), then the above equation implies 0 = mA = mAA∗ = m det(A) · In . The
latter matrix equation amounts to mi det(A) = 0 for all i. Thus • det(A) = •0 on
M , and by faithfulness this means det(A) = 0. Since so that α is a root of the
monic polynomial det(T · In − (aij )). 
Lemma 3.7. Let R ⊂ S ⊂ T be rings. If α ∈ T is integral over R, then it is
also integral over S.
Proof. If α is integral over R, there exists a monic polynomial P ∈ R[t] such
that P (α) = 0. But P is also a monic polynomial in S[t] such that P (α) = 0, so α
is also integral over S. 
Lemma 3.8. Let R ⊂ S ⊂ T be rings. If S is finitely generated as an R-
module and T is finitely generated as an S-module then T is finitely generated as
an R-module.
Proof. If α1 , . . . , αr generates S as an R-module and β1 , . . . , βs generates T
as an S-module, then {αi βj }{ 1 ≤ i ≤ r, 1 ≤ j ≤ s} generates T as an R-module:
for α ∈ T , we have X XX
α= bj βj = (aij αi )βj ,
j i j
 4. DISTINGUISHED CLASSES 29

with bj ∈ S and aij ∈ R. 

Corollary 3.9. (Transitivity of integrality) If R ⊂ S ⊂ T are rings such that
S/R and T /S are both integral, then T /R is integral.
Proof. For α ∈ S, let αn + bn−1 αn−1 + . . . + b1 α + b0 = 0 be an integral
dependence relation, with bi ∈ S. Thus R[b1 , . . . , bn−1 , α] is finitely generated over
R[b1 , . . . , bn−1 ]. Since S/R is integral, R[b1 , . . . , bn−1 ] is finite over R. By Lemma
3.8, R[b1 , . . . , bn−1 , α] is a subring of T containing α and finitely generated over R,
so by Theorem 3.6, α is integral over R. 
Corollary 3.10. If S/R is a ring extension, then the set IS (R) of elements
of S which are integral over R is a subring of S, the integral closure of R in S.
Thus R ⊂ IS (R) ⊂ S.
Proof. If α ∈ S is integral over R, R[α1 ] is a finitely generated R-module.
If α2 is integral over R it is also integral over R[α1 ], so that R[α1 ][α2 ] is finitely
generated as an R[α1 ]-module. By Lemmma 3.8, this implies that R[α1 , α2 ] is a
finitely generated R-module containing α1 ± α2 and α1 · α2 . By Theorem 3.6, this
implies that α1 ± α2 and α1 α2 are integral over R. 
If R ⊂ S such that IS (R) = R, we say R is integrally closed in S.
Proposition 3.11. Let S be a ring. The operator R 7→ IS (R) on subrings of
R is a closure operator in the abstract sense, namely it satisfies:
(CL1) R ⊂ IS (R),
(CL2) R1 ⊂ R2 =⇒ IS (R1 ) ⊂ IS (R2 ).
(CL3) IS (IS (R)) = IS (R).
Proof. (CL1) is the (trivial) Remark 1.1. (CL2) is obvious: evidently if R1 ⊂
R2 , then every element of S which satisfies a monic polynomial with R1 -coefficients
also satisfies a monic polynomial with R2 -coefficients. Finally, suppose that α ∈ S
is such that αn + an−1 αn−1 + . . . + a1 α + a0 = 0 for ai ∈ IS (R). Then each ai
is integral over R, so R[a1 , . . . , an ] is finitely generated as an R-module, and since
R[a1 , . . . , an , α] is finitely generated as an R[a1 , . . . , an ]-module, applying Lemma
3.8 again, we deduce that α lies in the finitely generated R-module R[a1 , . . . , an , α]
and hence by Theorem 3.6 is integral over R. 
Proposition 3.12. Let R ⊂ S be an integral extension. If R is a field, so is
S.
Proof: Let L be the fraction field of S. If 0 6= α ∈ S is integral over R, then by
Theorem 3.6, R[α] is a finite-dimensional R-submodule of L, so it is a subfield, i.e.,
is equal to R(α). So R(α) = R[α] ⊂ S, meaning that S contains α−1 .

4. Distinguished Classes
Here is an organizing principle for classes of field extensions due to S. Lang.

A class C of field extensions is distinguished if it satisfies these two properties:

(DC1) (Tower meta-property) For a tower M/K/F , then M/F ∈ C iff M/K ∈ C
and K/F ∈ C.
(DC2) (Base change meta-property) Let K/F be an element of C, let L/F be any
30 3. FIELD EXTENSIONS

extension such that K and L are contained in a common field. Then LK/L ∈ C.

We note that (DC1) and (DC2) imply the following

(DC3) (Compositum meta-property) Let K1 /F and K2 /F be elements of C with

K1 , K2 contained in a common field. Then K1 K2 /F ∈ C.

Indeed, applying (DC2) we get that K1 K2 /K2 ∈ C. Since also K2 /F ∈ C, ap-

plying (DC1) we get that K1 K2 /F ∈ C.
Exercise 3.6. a) Show: the class of all finite degree extensions is distinguished.
b) Show: the class of all algebraic extensions is distinguished.
Some examples of distinguished classes of extensions to come later: finitely gener-
ated extensions, separable algebraic extensions, purely inseparable algebraic exten-
sions, solvable extensions, purely transcendental extensions.

Some nonexamples of distinguished classes of extensions to come later: normal ex-

tensions, Galois extensions, inseparable extensions, abelian extensions, not-necessarily-
algebraic separable extensions.
 CHAPTER 4

Normal Extensions

1. Algebraically closed fields

Let F be a field. A polynomial f ∈ F [t] is split if every irreducible factor has
degree 1. If f ∈ F [t] is a polynomial and K/F is a field extension, we say f splits
in K if f ∈ K[t] is split.
Proposition 4.1. Let F be a field. The following are equivalent:
(i) There is no algebraic extension K ) F .
(ii) There is no finite degree extension K ) F .
(iii) There is no finite degree monogenic extension F (α) ) F .
(iv) If f ∈ F [t] is irreducible, then f has degree 1.
(v) If f ∈ F [t] is nonconstant, then f has a root in F .
(vi) Every polynomial f ∈ F [t] is split.
A field satisfying these equivalent conditions is called algebraically closed.
Proof. (i) =⇒ (ii) =⇒ (iii) is immediate.
¬ (iv) =⇒ ¬ (iii): if f ∈ F [t] is an irreducible polynomial of degree d > 1 then
K := F [t]/(f ) is a finite degree monogenic extension of f of degree d > 1.
¬ (v) =⇒ ¬ (iv): Suppose f is nonsconstant and admits no root in F . Write
f = f1 · · · fm as a product of irreducible polynomials; since linear polynomials have
roots in F , no fi has degree 1.
(iv) ⇐⇒ (v) ⇐⇒ (vi) is easy and familiar.
¬ (i) =⇒ ¬ (iv): If K ) F is a proper algebraic extension, let α ∈ K \ F , and let
f ∈ F [t] be the minimal polynomial of α over F , so f is irreducible. By assumption
f is also split, so it has degree 1 and is thus of the form t − α, contradicting the
fact that α ∈ / F. 
Theorem 4.2 (Fundamental Theorem of Algebra). The complex field C is
algebraically closed.
Because the existence of a nonconstant f ∈ C[t] without a root in C leads to
absurdities in many areas of mathematics, there are many different proofs, e.g.
using degree theory or complex analysis. It is often held that “fundamental theorem
of algebra” is a misnomer, in that the result concerns a structure – the complex
numbers – whose definition is in part analytic/topological. We do not dispute this.
Nevertheless the true algebraist hankers for an algebraic proof, and indeed this is
possible. We may, in fact, view Theorem 4.2 as a special case of the following result.
Theorem 4.3 (Artin-Schreier). Let K be a field such that:
(i) There do not exist n ∈ Z+ and x1 , . . . , xn ∈ K such that −1 = x21 + . . . +
x2n .
(ii) Every polynomial P ∈ K[t] of odd degree has a root in K.
31
32 4. NORMAL EXTENSIONS

(iii) For all x ∈ K × , exactly one of x and −x is a square in K.

√
Then K[ −1] = K[t]/(t2 + 1) is algebraically closed.
Proof. We will prove this result much later, as Theorem 14.31. 
Exercise 4.1. Show: Theorem 4.3 implies Theorem 4.2.
Proposition 4.4. Let L/K be a field extension, with L algebraically closed.
Let ClL (K) be the set of all elements of L that are algebraic over K. Then ClL (K)
is algebraically closed.

Proof. By Proposition 4.1, if ClL (K) is not algebraically closed then there is
a monogenic finite degree extension K(α) ) K. Because α is algebraic over ClL (K)
and ClL (K) is algebraic over K, we have by Corollary 3.9 that α is algebraic over K.
Let f ∈ F [t] be the minimal polynomial of α. By Proposition 4.1, as a polynomial
over L[t] we have
f (t) = (t − α1 )(t − α2 ) · · · (t − αd )
for some α1 , . . . , αd ∈ L. Each αi is algebraic over K so lies in ClL (K). Moreover
the α1 , . . . , αd are the only roots of f in L, and thus for some i we have α = αi ∈
ClL (K), a contradiction. 
Corollary 4.5. The field Q of all algebraic numbers is algebraically closed.
Proof. Since Q is the algebraic closure of Q in C, this follows from Theorem
4.2 and Proposition 4.4. 
Let K be a field. An algebraic closure of K is a field extension K/K that is both
algebraic and algebraically closed. It follows from Proposition 4.1 an algebraic
closure of K is precisely a maximal algebraic extension of K, i..e., an algebraic
extension that is not properly contained in any other algebraic extension of K.
Exercise 4.2. Let K/F be an algebraic field extension. Let L/K be a field
extension. Show: L is an algebraic closure of K iff L is an algebraic closure of F .

2. Existence of algebraic closures

In this section we will show that every field admits at least one algebraic closure,
a basic but nontrivial result. How might one try to prove this? Probably we can
agree to start with the following easy result.
Lemma 4.6. Let F be a field, and let f1 , . . . , fn ∈ F [t] be nonconstant polyno-
mials, of degrees d1 , . . . , dn .
a) There is a finite degree field extension K/F such that each Qnfi has a root
in K. Moreover, we can choose K so as to get [K : F ] ≤ i=1 di .
b) There is a finite degree field extension K/F such that Qneach fi splits in K.
Moreover, we can choose K so as to get [K : F ] ≤ i=1 di !.
Proof. a) Let M be a field, and let f ∈ M [t] be a polynomial of degree d.
Let g be an irreducible factor of f , say of degree d0 ≤ d. Then M [t]/(g) is a
field extension of M of degree d0 ≤ d in which g (and hence also f ) has a root.
By applying this procedure successively to f1 , . . . , fn we generate a tower of field
extensions F ⊂ M1 ⊂ . . . ⊂ Mn such that for all 1 ≤ i ≤ n, the polynomials
f1 , . . . , fi all have a root in Mi and [Mi : F ] ≤ d1 · · · di , so we may take K := Mn .
 2. EXISTENCE OF ALGEBRAIC CLOSURES 33

b) Let M be a field, and let h ∈ M [t] be a polynomal of degree d. Applying part a)

to h, there is a field extension M1 /M of degree at most d in which h has a root α1
and thus we get a factorization h(t) = (t − α1 )h2 (t) ∈ M1 [t]. We apply part a) to
h2 and get a field extension M2 /M1 of degree at most d − 1 in which h2 has a root
α2 and thus we get a factorization h(t) = (t − α1 )(t − α2 )h3 (t). Continuing in this
manner, we end up with a field extension Mn of degree at most d! in which h splits.
Applying this procedure successively to the polynomials
Qn f1 , . . . , fn over the field F
we get a field extension K of degree at most i=1 di ! in which each fi splits. 
Exercise 4.3. Let d1 , . . . , dn ∈ Z+ .
a) Show: there are (necessarily irreducible) polynomials f1 , . . . , fn ∈ Q[t] such that
if K/Q is a number
Qnfield (i.e., a finite degree field extension) such that each fi has
a root in K then i=1 deg fi | [K : Q].
b) Show: there are (necessarily irreducible) polynomialsQf1 , . . . , fn ∈ Q[t] such that
n
if K/Q is a number field in which each fi splits, then i=1 di ! | [K : Q].
(Hint/warning: this is most naturally done using basic algebraic number theory.)
Theorem 4.7. Every field K can be embedded in an algebraically closed field
L. Thus every field has at least one algebraic closure, namely ClL (K).
Proof. Step 1: Let R = K[T] be a polynomial ring over K indexed by a
set of indeterminates tf that are in bijection with the nonconstant polynomials
f ∈ K[t]. Consider the ideal I of R generated by all polynomials of the form f (tf ).
We claim that I is proper: if not, there is a finite subset {f1 , . . . , fn } and elements
g1 , . . . , gn ∈ R such that
g1 f1 (tf1 ) + . . . + gn fn (tfn ) = 1.
By Lemma 4.5, there is a finite degree field extension F/K such that each fi (t) has
a root αi ∈ F . If we evaluate tf1 = α1 , . . . , tfn = αn in the above equation, we get
0 = 1: contradiction. So we may choose a maximal ideal m ⊃ I. Thus K1 := R/m
is a field exetnsion of F in which each tf is a root of f . Thus K1 /K is a field
extension in which each nonconstant polynomial f ∈ K[t] has a root.
Step 2: The natural question here is whether K1 is algebraically closed. The
remainder of the proof consists of a clever evasion of this question! Namely, we
apply the construction of Step 1 to K1 , getting a field extension K2 in which each
polynomial with coefficients in K1 has a root in K2 , and so forth: we generate a
sequence of field extensions
K ⊂ K1 ⊂ . . . ⊂ Kn ⊂ . . . .
S
The union L = n Kn is a field, and any nonconstant polynomial P ∈ L[t], having
only finitely many nonzero coefficients lies in Kn [t] for sufficiently large n, thus has
a root in Kn+1 and therefore also in L. So L is algebraically closed, and then by
Proposition 4.4 the algebraic closure of K in L is an algebraic closure of K. 
Theorem 4.7 lies among the most important results in all of field theory. So we
pause to discuss several aspects of it.
First, the proof of Theorem 4.7 used the Axiom of Choice (AC) in a somewhat
disguised way: in the assertion that a proper ideal in a ring is contained in a max-
imal ideal. In fact the statement that every proper ideal in a commutative ring is
contained in a maximal ideal implies (AC). So it is natural to wonder whether the
existence of an algebraic closure of any field implies (AC). Indeed not: it would be
34 4. NORMAL EXTENSIONS

enough to use that every proper ideal is contained in a prime ideal: this gives us a
domain, and we can take the fraction field. The assertion that every proper ideal
in a commutative ring is contained in a prime ideal is known to be equivalent to
the Ultrafilter Lemma (UL), which does not imply (AC).
It seems to be an open problem whether the existence of an algebraic closure of
every field implies (UL): cf. http://mathoverflow.net/questions/46566. How-
ever, it is known that (AC) is required for Theorem 4.7 to hold in the sense that
there is a model of Zermelo-Fraenkel set theory in which not every field admits an
algebraic closure [Je, Thm. 10.13].

The proof of Theorem 4.7 comes from E. Artin by way of Lang [LaFT]. It is
unnecessarily (though helpfully) slick in several respects. The use of polynomial
rings is a crutch to avoid some mostly set-theoretic unpleasantries: later we wil
see that an algebraic closure of F is essentially the direct limit of all finite degree
normal field extensions K/F : here the essentially means that we want each K/F
to appear exactly once up to F -isomorphism. It just happens that the easiest way
to do that is to realize each K inside a fixed algebraically closed field containing F !
But by the time the reader has made it to the end of this section, she may consider
trying to construct this direct limit directly.

Finally, as we pointed out, the proof constructs an extension K1 /K such that

every nonconstant f ∈ K[t] has a root in K1 and then nimbly evades the question
of whether K1 contains an algebraic closure of K. It turns out that the answer to
this is affirmative. We break this up into two steps. First:
Proposition 4.8. Let L/K be a field extension. Suppose every nonconstant
f ∈ K[t] splits in L. Then the algebraic closure of K in L is algebraically closed.
Proof. Let K be the algebraic closure of K in L. Suppose K is not alge-
braically closed: then by Proposition 4.1 there is a field extension M/K and an
element α ∈ M \ K that is algebraic over K. By Corollary 3.9 we have that α is
algebraic over K, so has a minimal polynomial f ∈ K[t]. By assumption f splits in
K, and since f (α) = 0 one of the factors of f must be t − α and thus α ∈ K. 
As for the second step: we will record the answer now, but we will need to know
more of the structure theory of algebraic field extensions in order to prove it.
Theorem 4.9. (Gilmer [Gi68]) Let L/K be a field extension. If every non-
constant f ∈ K[t] has a root in L, then every nonconstant f ∈ K[t] splits in L.
Exercise 4.4. Let K be a field, and let f ∈ K[t] be a monic polynomial of
degree d ≥ 1. Let K be an algebraic closure of K. Over K, f splits:
f (t) = (t − α1 ) · · · (t − αd ).
We say f is separable if the α1 , . . . , αd are distinct elements of K.
a) Conceivably the above definition depends on the choice of K. However,
let f 0 be the (formal) derivative of f : the unique K-linear endomorphism
of K[t] such that (tn )0 = ntn−1 . Show: f is separable iff gcd(f, f 0 ) = 1.
b) Let K be a field and n ∈ Z+ . If K has positive characteristic p, assume
that gcd(n, p) = 1. Let a ∈ K be arbitrary. Show: the polynomial tn − a
is separable.
 3. THE MAGIC MAPPING THEOREM 35

c) Deduce: if K is a field and n ∈ Z+ is prime to the characteristic of K

if it is positive, then there is a field extension L/K containing n different
nth roots of unity: i.e., distinct z1 , . . . , zn such that zin = 1 for all 1.
d) Deduce: no finite field is algebraically closed.
Exercise 4.5.
a) Show: if K is a field and K is an algebraic closure, then #K = max(ℵ0 , #K).
b) Show: there are algebraically closed fields of all infinite cardinalities.

3. The Magic Mapping Theorem

Theorem 4.10. (Magic Mapping Theorem) Let F be a field. Let K/F be an
algebraic field extension, and let L/F be a field extension with L algebraically closed.
Then there is an F -algebra homomorphism ϕ : K ,→ L.
Proof. Consider the partially ordered set whose elements are are pairs (M, ϕ)
where M is a subextension of K/F and ϕ : M → L is an F -algebra homomorphism.
We say that (M1 , ϕ1 ) ≤ (M2 , ϕ2 ) if M1 ⊂ M2 and the restriction of ϕ2 to M1 is
ϕ1 . In this partially ordered set, any chain has an upper bound given by taking
the union of the elements of the chain. So by Zorn’s Lemma there is a maximal
element (M, ϕ). We claim that M = K. If not, let α ∈ K \ M , and consider
the field extension M (α)/M . Let f ∈ M [t] be the minimal polynomial of α, so
M (α) ∼ = M [t]/(f ). We view L as an M -algebra via ϕ, and thus we may view
f ∈ L[t]. Since L is algebraically closed, there is a root in L, say α. There is
a unique M -algebra homomorphism M (α) → L that maps α to α: it is unique
because M (α) = M [α] is generated as an M -algebra by α, and it exists because
M (α) ∼= M [t]/(f (t)) so the unique M -algebra map M [t] → L that carries t to α
has f (t) in its kernel. It follows that M = K. 

Corollary 4.11 (“Uniqueness” of Algebraic Closure). Let F1 and F2 be two

algebraic closures of a field F . There is an F -algebra isomorphism ϕ : F 1 → F2 .
Proof. We may apply the Magic Mapping Theorem with K = F1 and L = F2
to get an F -algebra homomorphism ϕ : F1 ,→ F2 . Then F2 /ϕ(F1 )) is an algebraic
extension of an algebraically closed field, so it cannot be proper: we have F2 = ϕ(F1 )
and thus ϕ is an F -algebra isomorphism. 

Note the scare quotes around uniqueness. This is because we have shown that
the algebraic closure of F is unique up to F -algebra isomorphism, but given two
algebraic closures of F there is in general no canonical F -algebra isomorphism
between them. If ϕ, ψ : F1 ,→ F2 are two F -algebra isomomorphisms, then ψ −1 ◦ ϕ
is an F -algebra automorphism of F1 , and conversely: the ambiguity in the choice of
isomorphism is precisely measured by the group GF := Aut(F 1 /F ). This group is
called the absolute Galois group of F and is in general a very large, interesting
group. In fact, we should not speak of “the” absolute Galois group of F (though we
will: it is traditional to do so): it is well-defined up to isomorphism, but switching
from one isomorphism F1 → F2 to another gives rise to an inner automorphism
(i.e., a conjugation) of G. More on this later.
Remark 4.1. There are models of Zermelo-Fraenkel set theory – i.e., without
(AC) – in which a field F can admit non-F -isomorphic algebraic closures.
36 4. NORMAL EXTENSIONS

Corollary 4.12. Let K1 /F and K2 /F be two algebraic field extensions. If

ϕ : K1 → K2 is any F -algebra embedding and Ki is any algebraic closure of Ki ,
then ϕ extends to an isomorphism K1 → K2 .
Exercise 4.6. Prove Corollary 5.2.

4. Conjugates
Let K/F be an algebraic field extension. We say that elements α, β ∈ K are con-
jugate over F if α and β have the same minimal polynomial over F . Thus if
[K(α) : K] = d, then the number of conjugates of α lying in K is at most d. We
will see later that whether α has d conjugates in K is is a crucial issue: it will be so
if and only if the extension K(α)/K is both normal (the property we are studying
in this chapter, with the definition to come soon) and separable (the property we
will study in the next chapter).

If K/F is an algebraic extension and F is any algebraic closure of F , then as

we know there is an F -algebra homomorphism ι : K ,→ F . If α ∈ K and f ∈ F [t]
is the minimal polynomial of α, then f splits in F . We call the roots of f in F the
conjugates of α. Notice that the set of conjugates is defined only in terms of the
minimal polynomial, which lies in F , so it is independent of the choice of ι.

For the remainder of this section we fix an algebraic closure F of F and only
consider algebraic extensions K/F that are subextensions of F /F (again, every
algebraic extensions occurs this way up to F -algebra isomorphism). From this per-
spective, being conjugate over F is an equivalence relation on F . Moreover, if σ
is an F -algebra automorphism of F , then for all α ∈ F , we have that σ(α) is a
conjugate of α: indeed, for every polynomial f ∈ F [t], we have
f (α) = 0 ⇐⇒ f (σ(α)) = 0
and thus α and σ(α) have the same minimal polynomial. Conversely, if α, β ∈ F
are conjugate over F , then there is an F -algebra automorphism σ of F such that
σ(α) = β. Indeed, let f ∈ F [t] be the common minimal polynomial of α and β.
Then the field extensions F (α) and F (β) are both isomorphic to F [t]/(f (t)), so
there is an isomorphism
F (α) → F (β),
which by Corollary 5.2 extends to an automorphism of F . We sum up this discussion
as follows.
Lemma 4.13 (Extension Lemma). Let F be an algebraic closure of the field F .
For α, β ∈ F , the following are equivalent:
(i) The elements α and β are conjguate over F : that is, they have the same
minimal polynomial.
(ii) There is σ ∈ Aut(F /F ) such that σ(α) = β.
Remark 4.2. Recall that if a group G acts on a set X, we say that two elements
x, y ∈ X are conjugate if there is g ∈ X such that gx = y. As we just saw, the
terminology of conjugate elements of F is compatible with this: two elements of F
are conjugate iff they are conjugate under the action of Aut(F /F ).
Exercise 4.7. Let Q be an algebraic closure of Q. Show that Aut(Q/Q) is
infinite.
 6. NORMAL EXTENSIONS 37

5. Splitting Fields
It follows from Proposition 4.8 that if K/F is an algebraic field extension such that
every nonconstant f ∈ F [t] splits in K, then K is an algebraic closure of F . This
view on algebraic closure opens the door to a natural and important generalization:
we go from “all polynomials” to “some polynomials.”

Let F be a field, and let S ⊂ F [t] be a set of nonconstant polynomials. A splitting

field for (F, S) is a field extension K/F satisfying the following properties:

(SF1) Every fi ∈ S splits in K.

(SF2) No proper subextension of K satisfies (SF1), i.e., if F ⊂ K 0 ⊂ K and every
fi ∈ S splits in K 0 , then K 0 = K.
Exercise 4.8. Suppose K/F is a splitting field for (F, S), and K 0 is an F -
algebra isomorphic to K. Show: K 0 is also a splitting field for (F, S).
Theorem 4.14. (Existence and “Uniqueness” of Splitting Fields) Let F be a
field and S ⊂ F [t] a set of nonconstant polynomials.
a) Any algebraic closure F contains a unique splitting field for S, namely the
subfield of F obtained by adjoining to F all roots αij of all polynomials
Pi ∈ S.
b) Splitting fields are unique up to F -algebra isomorphism.
Proof. It is no problem to see that the recipe of part a) does indeed construct
a splitting field for F and S: clearly every polynomial in S splits in F (αij ) and
conversely any subfield of F in which all the polynomials in F split must contain
all the αij ’s. One way to see the uniqueness up to isomorphism is to reduce to the
case of uniqueness up to isomorphism of algebraic closures. Namely, let K1 , K2 be
two splitting fields for F and S. It is easy to see that (SF2) implies that Ki /F is
algebraic, so let K i be an algebraic closure of Ki . Since Ki is algebraic over F , K i is
equally well an algebraic closure of F , so by Corollary 4.11 there exists an F -algebra
isomorphism Φ : K 1 → K 2 . Then Φ(K1 ) is a subfield of K 2 which is a splitting
field for F and S, and we just saw that each algebraic closure contains a unique
splitting field, so Φ(K1 ) = K2 and Φ : K1 → K2 is an F -algebra isomorphism. 
√
Exercise 4.9. Show that the field K = Q[ζ3 , 3 2] is the splitting field of f =
t3 − 2. Conclude that if L ⊂ C is such that L 6= K, then L is not isomorphic to K.

6. Normal Extensions
Lemma 4.15. Let F be an algebraic closure of F , let K be a subextension of
F /F , and let σ : K ,→ F be an F -algebra embedding. The following are equivalent:
(i) σ(K) ⊂ K.
(ii) σ(K) ⊃ K.
(iii) σ(K) = K.
Proof. Certainly (iii) implies both (i) and (ii). We will show (i) =⇒ (iii),
and it will be clear how to modify the argument so as to obtain (ii) =⇒ (iii).
(i) =⇒ (iii): Let α ∈ K, and let S be the set of F -conjugates of α that lie in
K. We observe that S is a finite set containing α. For β ∈ K, we have that β is a
conjugate of α iff σ(β) is a conjugate of α, so the set of F -conjugates of α that lie in
38 4. NORMAL EXTENSIONS

σ(K) is precisely σ(S). By hypothesis we have σ(S) ⊂ S; since both are finite sets
of the same cardinality we must have σ(S) = S and thus α ∈ σ(S) ⊂ σ(K). 
Theorem 4.16. Let K/F be an algebraic field extension. Let F be an algebraic
closure of K (hence also of F ). The following are equivalent:
(i) For every F -algebra embedding σ : K ,→ F we have σ(K) = K.
(ii) K/F is the splitting field of a subset S ⊂ F [t].
(iii) Every irreducible polynomial f ∈ F [t] with a root in K splits in K.
(iv) For all α ∈ K, if β ∈ F is an F -conjugate of α, then β ∈ K.
An extension K/F satisfying these properties is called normal.1
Proof. (i) ⇐⇒ (iv): We saw above that for α ∈ K and β ∈ F , β is a
conjugate of α in F iff there is an F -algebra homomorphism σ : K ,→ F such
that σ(α) = β. It follows S that as we range over all F -algebra homomorphisms
σ : K ,→ F , we have that σ σ(K) is the set of all conjugates of all elements of K.
Condition
S (iv) holds iff the set of all conjugates of all elements of K is just K itself
iff σ σ(K) = K iff σ(K) ⊂ K for all σ iff σ(K) = K for all σ(K): condition (i).
(iii) ⇐⇒ (iv) is immediate.
(ii) ⇐⇒ (iv): Condition (ii) can be rephrased by saying that K is generated by
adjoining to F a subset S of F that is stable under conjugation. Thus if (iv) holds,
then (ii) holds with S = K. Conversely, suppose that K is obtained by adjoining to
F a set S that is stable under conjugation, and let x ∈ K. Then x = f (α1 , . . . , αn ) is
a rational function in elements α1 , . . . , αn ∈ S with F -coefficients. Every conjugate
of x in F is of the form σ(x) for some F -automorphism σ of F , and then
σ(x) = f (σ(α1 ), . . . , σ(αn )) ∈ K,
since S is closed under conjugation. 
Exercise 4.10. Let F ⊂ K ⊂ L be field extensions. If L/F is normal, show
that L/K is normal.
Corollary 4.17. Let K/F be a normal field extension. Let L/K be any field
extension and let σ : L → L be any automorphism. Then σ(K) = K.
Proof. Let L be an algebraic closure of L. By Corollary 4.11, we may extend
σ to an automorphism σ : L → L. Let K = ClL (K), the unique algebraic closure
of K contained in L. Since K/F is algebraic, K is also an algebraic closure of F .
Since σ(K) is also an algebraic closure of K contained in L, by the aforementioned
uniqueness we have σ(K) = K. By Theorem 4.16 we have σ(K) = K. 
√
Example 4.18. For each n ≥ 3, the extension K = Q[ n 2]/Q is a non-normal
extension of√ degree n. Indeed, let ζn = e2πi/n ; then the other roots of tn − 2 in
C are ζnk · n 2 with 0 ≤ k < n, which are not even real numbers unless k = 0 or
k = n2 . So tn − 2 does not split over K. In this case, any extension of K which
√
is normal over Q must contain all the roots of tn − 2, hence must contain n 2 and
ζn . Therefore
√ the smallest normal extension is the splitting field of tn − 2, which is
n
M = Q[ 2, ζn ].
Exercise 4.11. a) Suppose K/F is finite of degree at most 2. Show:
K/F is normal.
1Normal field extensions are by definition algebraic.
 6. NORMAL EXTENSIONS 39

√ √
b) Use the example Q ⊂ Q( 2) ⊂ Q( 4 2) to show that if we have fields
F ⊂ K ⊂ L and K/F and L/K are both normal, then L/F need not be
normal. ((Thus normality does not satisfy the tower meta-property.)
Example 4.19. Suppose F has characteristic p > 0. Suppose a ∈ F is such
n
that f (t) = tp − a ∈ F [t] is irreducible. Let K = F [t]/(f (t)), and write α for the
n n
coset t+(f (t)): thus αp = a. Then as an element of K[t] we have f (t) = (t−α)p .
That is, despite the fact that f has degree pn , α is conjugate in F only to itself.
Thus K/F is a normal extension.
Exercise 4.12. Show: a direct limit of normal extensions is normal.
Exercise 4.13. a) Let L/F be an extension and K1 , K2 be subexten-
sions. Show: K1 ∩ K2 is again an extension field of F .
b) As above, but with any collection of intermediate field extensions {Ki }i∈I .
Proposition 4.20. Let L/F be an extension and {Ki }i∈I /F a collection of
algebraic
T subextensions. If each Ki /F is normal, then so is the intersection K =
i K i .
Proof. Without loss of generality we may replace L by the algebraic extension
ClL (F )/F . Let f ∈ F [t] be an irreducible polynomial with a root α in K. Then
for all i ∈ I, f has a root in Ki , thus each fi contains all the F -conjugates of α,
hence so does K, so f splits in K. 

Let K/F be an algebraic field extension, and let K be an algebraic closure of

K. Then K/F is certainly normal. Since the intersection of any family of normal
subextensions of K is normal, it follows that there is a unique smallest subextension
L, F ⊂ K ⊂ L ⊂ K, such that L/F is normal. If we define a normal closure of an
extension K/F to be an extension L/K which is normal over F and such that no
proper subextension is normal over F , then we just constructed a normal closure,
by intersecting all normal subextensions inside an algebraic closure of K.
Exercise 4.14. Let K/F be an algebraic extension. Show that any two normal
closures of K/F are F -isomorphic.
Proposition 4.21. Let K/F be finite of degree n. Then the degree of the
normal closure M of K/F (inside any algebraic closure K) is at most n!
Proof. Put F = F0 . Write K = F (α1 , . . . , αd ) and for 1 ≤ i ≤ d, put
Ki = F (α1 , . . . , αi ) and di := [Ki : Ki−1 ]. An argument almost identical to
that of Lemma 4.6b) yields a field extension M/K containing all the conjugates of
Qd
α1 , . . . , αd and such that [M : F ] = i=1 di !. Thus the normal closure of K/F has
Qd
degree at most i=1 di !. Now
d
Y d
Y
n = [K : F ] = [Ki : Ki−1 ] = di .
i=1 i=1
Qd
It follows that i=1 di ! ≤ n!: for instance take sets S1 , . . . , Sd of cardinalities
Qd Qn
d1 , . . . , dd . Then i=1 di ! is the number of bijections of S := i=1 Si that preserve
each coordinate, while (d1 · · · dn )! is the number of bijectiions of S. 
40 4. NORMAL EXTENSIONS

7. The Extension Theorem

The following result is an immediate consequence of already developed results, but
it is so useful that we state and prove it so as to be able to refer to it later on.
Theorem 4.22 (Extension Theorem). Let K/F be a normal field extension.
For α, β ∈ K, the following are equivalent:
(i) The elements α and β are conjugates over F : they have the same minimal
polynomial over F .
ii) There is an F -algebra automorphism s ∈ Aut(K/F ) such that s(α) = β.
Proof. (i) =⇒ (ii): When K = F is an algebraic closure of F , this is half of
Lemma 4.13. In general, let F be an algebraic closure of F containing K, and let
σ ∈ Aut(F /F ) be such that σ(α) = β. By Theorem 4.16 we have σ(K) = K, so
s := σ|K ∈ Aut(K/F ) and s(α) = β.
(ii) =⇒ (i): For any field extension K/F , if α ∈ K is algebraic over F and β ∈ K
is such that there is some s ∈ Aut(K/F ) such that s(α) = β, then let P ∈ F [t] be
the minimal polynomial of α over F . Then
0 = σ(0) = σ(P (α)) = P (σ(α)) = P (β),
so β is a conjugate of α over F . 

8. Isaacs’ Theorem
The goal of this section is to prove the following result of Isaacs.
Theorem 4.23. (Isaacs [Is80]) Let F be a field. For an algebraic extension
K/F , let P(K) be the set of polynomials f ∈ F [t] having a root in K. Then for
algebraic extensions K1 /F , K2 /F , the following are equivalent:
(i) The F -algebras K1 and K2 are isomorphic.
(ii) We have P(K1 ) = P(K2 ).
Exercise 4.15. a) Show: Isaacs’ Theorem implies that a field extension
L/F that contains a root of every nonconstant f ∈ F [t] contains an alge-
braic closure of F .
b) Deduce Gilmer’s Theorem (Theorem 4.9).
 CHAPTER 5

Separable Algebraic Extensions

Let K/F be an algebraic field extension. We have already explored one desirable
property for K/F to have: normality. Normality can be expressed in terms of
stability under F -homomorphisms into any extension field, and also in terms of
irreducible polynomials: every irreducible polynomial in F [t] with a root in K[t]
must split. There is another desirable property of an algebraic extension L/K called
separability. In some sense it is dual to normality, but this is hard to believe at
first because there is a large class of fields F for which all algebraic extensions K/F
are separable, including all fields of characteristic 0. (For that matter, there are
fields for which every algebraic extension is normal, like R and Fp .) Like normality,
separability can also be expressed in terms of polynomials and also in terms of
embedding conditions. We begin with a study of polynomials.

1. Separable Polynomials
A nonconstant polynomial P ∈ F [t] is separable if over an algebraic closure F ,
P (t) splits into distinct linear factors. Equivalently, if P has degree n, then there
are n distinct elements α1 , . . . , αn ∈ F such that P (αi ) = 0 for all i. Each condition
is easily seen to be independent of the chosen algebraic closure.
Exercise 5.1. Let F be a field and K/F be any extension. Show that a poly-
nomial P ∈ F [t] is separable as a polynomial over F iff it is separable when viewed
as a polynomial over K.
Lemma 5.1. Let F be a field of characteristic p > 0 and α ∈ F × \ F ×p . (That
n
is, α is not a pth power in F .) Then for all n ≥ 1, the polynomial tp − α is
irreducible.
Proof. We shall prove the contrapositive: suppose that for some n ∈ Z+ the
n
polynomial tp − α is reducible; we will show that α is a pth power in F . We may
n
write tp − α = f (t)g(t), where f (t) and g(t) are nonconstant monic polynomials.
n
Let K/F be an extension field containing a root β of tp − α, so that in K[t] we
have n n n n
tp − α = tp − β p = (t − β)p .
Since K[t] is a UFD and f (t) and g(t) are monic, we therefore have f (t) = (t − β)r
for some 0 < r < pn . Write r = pm s with gcd(p, s) = 1. Note that m < n. Then
m m
f (t) = (tp − β p )s ,
m m
so that the coefficient of tp (s−1) is −sβ p . This lies in F and – since s 6= 0 in F
m
– we conclude β p ∈ F . Thus
m n−m n−m
α = (β p )p ∈ Fp ∈ Fp
since m < n. 
41
42 5. SEPARABLE ALGEBRAIC EXTENSIONS

Over any field F it is no trouble to come up with a polynomial that is not sepa-
rable: t2 . What is of more interest is whether there is an inseparable irreducible
polynomial in F [t]. Note that some authors define a polynomial to be separable if
all its irreducible factors are separable and others only discuss in/separability for
irreducible polynomials. Although these conventions certainly “work” as well, I find
the current definition to be more convenient and more thematic. First, Exercise 5.1
shows that with this definition, separability is faithfully preserved by base exten-
sion. Since the way one will check whether an irreducible polynomial is separable is
by considering it over the algebraic closure, where of course it is a product of sep-
arable (linear!) polynomials, our definition seems simpler. Moreover, in the theory
of algebras one does meet reducible polynomials: for any nonconstant P ∈ F [x],
we may consider the finite-dimensional F -algebra AP = F [x]/(P (x)). Then our
definition makes it true that P is separable iff AP is a separable algebra, i.e., an
algebra that is semisimple and remains semisimple after arbitrary base change.

In general it is far from obvious whether the field extension obtained by adjoin-
ing a root of an irreducible polynomial is normal. Fortunately, it is much easier to
determine whether a polynomial, especially an irreducible polynomial, is separable.
Exercise 5.2. Let k be a field.
a) Show: there is a unique k-linear endomorphism f 7→ f 0 of k[t] such that
for all n ∈ N we have (tn )0 = ntn−1 .
b) Show: for all f, g ∈ k[t] we have (f g)0 = f 0 g + f g 0 .
c) Show: for all f, g ∈ k[t] we have (f (g(t)))0 = f 0 (g(t))g 0 (t).
d) Suppose k has characteristic 0. Show: if deg(f ) = n ≥ 1, then deg(f 0 ) =
n − 1. Deduce that {f ∈ k[t] | f 0 = 0} = k.
e) Suppose k has characteristic p > 0. Show:
{f ∈ k[t] | f 0 = 0} = k[tp ].
Proposition 5.2 (Derivative Criterion). Let f ∈ F [t] be a nonconstant poly-
nomial. Then:
a) The polynomial f is separable iff gcd(f, f 0 ) = 1.
b) If f is irreducible, it is separable iff f 0 6= 0.
c) An irreducible polynomial is always separable in characteristic 0. In char-
acteristic p > 0, an irreducible polynomial is inseparable iff there exists
g ∈ F [t] such that f (t) = g(tp ).
Proof. a) Let d ∈ F [t] be a greatest common divisor of f and f 0 . This means
{αf + βf 0 mod α, β ∈ F [t]} = {γd | γ ∈ F [t]}.
If K/F is any field extension, it follows that
{αf + βf 0 mod α, β ∈ K[t]} = {γd | γ ∈ K[t]},
so d is again a greatest common divisor of f and f 0 in K[t]. Moreover, if F is an
algebraic closure of F and K is an algebraic closure of K, then the Magic Mapping
Theorem gives an F -algebra homomorphism F ,→ K, so f ∈ F [t] is separable iff
f ∈ K[t] is separable. Thus both of the conditions of part a) are stable under
replacing F by an extension field, so we may assume that F is algebraically closed
and thus f is split. If f is not separable, then for some α ∈ F we have
f = (t − α)2 g
 1. SEPARABLE POLYNOMIALS 43

and thus
f 0 = (t − α)2 g 0 + 2(t − α)g = (t − α)h
so (t − α) | gcd(f, f 0 ). Conversely, if f is separable, then for every root α of f we
have
f = (t − α)g with g(α) 6= 0,
so
f 0 = (t − α)g 0 + g,
so f 0 (α) = g(α) 6= 0 and thus (t − α) - f 0 . Thus gcd(f, f 0 ) = 1.
b) If f is irreducible, then since gcd(f, f 0 ) | f , if gcd(f, f 0 ) 6= 1 then gcd(f, f 0 ) = f
so f | f 0 . Since deg(f 0 ) < deg(f ), this occurs if and only if f 0 = 0.
c) This follows from part b) and the previous exercise. 

A field F is perfect if every irreducible polynomial over F is separable. By Propo-

sition 5.2, every field of characteristic 0 is perfect. In other words, the entire
discussion of separability is nonvacuous only in positive characteristic, so for the
remainder of this section we assume that all fields are of positive characteristic.
Unless otherwise specified, p shall always denote a prime number which is the char-
acteristic of the field(s) in question.

If F has characteristic p > 0, we consider the Frobenius homomorphism

f : F → F, x 7→ xp .
Let F p = f(F ) be the image, a subfield of F .
Proposition 5.3. a) A field of characteristic p > 0 is perfect iff the Frobenius
homomorphism is surjective: F p = F .
b) Therefore finite fields and algebraically closed fields are perfect.
Proof. Assume F p = F , and let P (t) = i ai tpi be an irreducible inseparable
P
polynomial. We can then write ai = bpi and then
X X
P (t) = (bi )p (ti )p = ( bi ti )p = Q(t)p ,
i i

hence P (t) is not irreducible after all. Therefore F is perfect. Inversely, if the
Frobenius homomorphism is not surjective, then there exists some α ∈ F which is
not a pth power, and then by Lemma 9.20, the inseparable polynomial tp − α is
irreducible, so F is not perfect. This gives part a). As for part a), like any field
homomorphism, the Frobenius map is injective, and an injective map from a finite
set to itself is necessarily surjective. If F is algebraically closed, then for any α ∈ F
the polynomial tp − α has a root in F , i.e., α ∈ F p . 
a
For any positive integer a, we may consider the pa , the map which takes α 7→ αp ,
which can also be described as the ath power of the Frobenius map. We write
a
F p = pa (F ). If F is not perfect then we get an infinite descending chain of proper
subfields
2
F ) Fp ) Fp ) ...
a−1 a−1 a
Indeed, if α ∈ F \ F p , then αp ∈ Fp \ F p . This gives another proof that an
imperfect field is infinite.
44 5. SEPARABLE ALGEBRAIC EXTENSIONS

Exercise 5.3. Let F be a field of characteristic p, with an algebraic closure F .

Define F 1/p = {β ∈ F | β p ∈ F }.
a) Show that F 1/p is a subextension of F /F .
b) Similarly define a tower of subextensions
2 a
F ⊂ F 1/p ⊂ F 1/p ⊂ . . . F 1/p ⊂ . . . ⊂ F ,
and show that if F S is imperfect, all these inclusions are strict.
∞ ∞ a ∞
c) Define F 1/p = a=1 F 1/p . Show that F 1/p is perfect and is the intersection
of all perfect subextensions of F . It is called the perfect closure of F .
Purely inseparable polynomials: Say that a polynomial P (t) ∈ F [t] is purely
inseparable if there exists exactly one α ∈ F such that P (α) = 0. As above, there
are certainly purely inseparable polynomials over F – (t − α)n for any α ∈ F and
n ∈ Z+ – and what is of interest is the purely inseparable irreducible polynomials,
which by the discussion thus far clearly can only exist in characteristic p > 0.
Proposition 5.4. Let F be a field of characteristic p > 0. The irreducible,
purely inseparable monic polynomials P (t) ∈ F [t] are precisely those of the form
a
tp − α for some a ∈ Z+ and some α ∈ F \ F p .
a
Proof. By Lemma 9.20, any polynomial of the form tp − α for α ∈ F \ F p is
irreducible. Conversely, let P (t) ∈ F [t], By Proposition 5.2c), there is a polynomial
P2 (t) such that P (t) = P2 (tp ). Since P is irreducible, so is P2 . If there are distinct
1
α, β ∈ F such that P2 (α) = P2 (β) then there are unique and distinct elements α p ,
1 1 1
β p in F such that P (α p ) = P (β p ) = 0, contradicting the pure inseparability of α.
Therefore P2 must itself be irreducible purely inseparable, and an evident inductive
argument finishes the proof. 
Exercise 5.4. Show that the polynomial t6 −x over the field F3 [x] is irreducible
and inseparable but not purely inseparable.

2. Separable Algebraic Field Extensions

Let F be a field and P (t) an irreducible, inseparable polynomial over F of degree
d > 1. Consider the finite field extension K = F [t]/(P (t)) of F . It exhibits some
strange behavior. First, the only F -algebra embedding σ : K → K is the inclusion
map. Indeed, such embeddings correspond bijectively to the assignments of t ∈ K
to a root α of P in K, and by assumption there are less than d such elements. It
follows that the group Aut(K/F ) of F -algebra automorphisms of F has cardinality
smaller than d.

For an extension K/F , the separable degree [K : F ]s is the cardinality of the set
of F -algebra embeddings σ : K → F .
Exercise 5.5. Show that the separable degree may be computed with respect to
embeddings into any algebraically closed field containing F .
Theorem 5.5. The separable degree is multiplicative in towers: if L/K/F is a
tower of finite degree field extensions, then [L : F ]s = [L : K]s [K : F ]s .
Proof. Let σ : F ,→ C be an embedding of F into an algebraically closed field.
Let {σi }i∈I be the family of extensions of σ to K, and for each i ∈ I let {τij }j∈Ji be
the family of extensions of σi to L. Each σi admits precisely [L : K]s extensions to
 2. SEPARABLE ALGEBRAIC FIELD EXTENSIONS 45

embeddings of L into C: in particular, the cardinality of Ji is independent of i and

there are thus precisely [L : K]s [K : F ]s F -algebra embeddings τij overall. These
give all the F -algebra embeddings L ,→ C, so [L : F ]s = [L : K]s [K : F ]s . 
Corollary 5.6. Let K/F be a finite degree field extension. Then
[K : F ]s ≤ [K : F ].
In particular, the separable degree is finite.
Proof. We employ dévissage: break up K/F into a finite tower of simple
extensions. Each simple extension has finite degree and by Theorem 5.5 the degree
is multiplicative in towers. We are therefore reduced to the case K = F (α) ∼ =
F [t]/(P (t)), where P (t) is the minimal polynomial for α. In this case the result is
clear, since an F -algebra homomorphism of F [t]/(P (t)) into any field M is given by
sending the image of t to a root of P (t) in M , and the degree [K : F ] polynomial
has at most [K : F ] roots in any field. 
In the situation of the proof of Corollary 5.6 we can say more: the separable degree
[F (α) : F ]s is equal to the number of distinct roots of the minimal polynomial P (t)
of α. In particular it is equal to the degreee of the field extension iff P (t) is a
separable polynomial. Let us record this result.
Proposition 5.7. For K/F a field extension and α ∈ K algebraic over F , the
following are equivalent:
(i) The minimal polynomial of α is a separable polynomial.
(ii) [F (α) : F ]s = [F (α) : F ].
More generally:
Theorem 5.8. For a finite degree field extension K/F , the following are equiv-
alent:
(i) Every element of K is separable over F .
(ii) We have [K : F ]s = [K : F ].
A field extension satisfying these equivalent conditions is said to be separable.
Proof. (i) =⇒ (ii): We may write K/F as a finite tower of simple extensions:
F = F0 ⊂ . . . ⊂ Fn = K
such that for all i we have Fi+1 = Fi (αi+1 ). Since αi+1 is separable over F , it is
separable over Fi : indeed, the minimal polynomial for αi+1 over the extension field
divides the minimal polynomial over the ground field. Therefore Proposition 5.7
applies and [Fi+1 : Fi ]s = [Fi+1 : Fi ] for all i. Since both the separable degree and
the degree are multiplicative in towers, we conclude [K : F ]s = [K : F ].
(ii) =⇒ (i): Seeking a contradiction, we suppose that there exists α ∈ K which is
not separable over F . By Proposition 5.7, it follows that [F (α) : F ]s < [F (α) : F ].
Now applying Theorem 5.5 and Corollary 5.6 we get
[K : F ]s = [K : F (α)]s [F (α) : F ]s < [K : F (α)][F (α) : F ] = [K : F ]. 
Corollary 5.9. Finite degree separable extensions are a distinguished class of
field extensions: that is, they satisfy (DC1) and (DC2) of §3.4 and thus also (DC3).
Exercise 5.6. Prove Corollary 5.9.
46 5. SEPARABLE ALGEBRAIC EXTENSIONS

Theorem 5.10.
Let L/F be an algebraic field extension. The following are equivalent:
(i) Every finite subextension of L/F is separable.
(ii) Every irreducible polynomial P ∈ F [t] which has a root in L is separable.
(iii) L is obtained by adjoining to F a set of roots of separable polynomials.
An extension satisfying these equivalent properties is called a separable algebraic
extension.
Exercise 5.7. Prove Theorem 5.10.
Corollary 5.11. Algebraic separable extensions are a distinguished class of
field extensions.
Exercise 5.8. Prove Corollary 5.11.
Corollary 5.12. For a family {Ki /F }i∈I of algebraic field extensions inside
a common algebraically closed field M , the following are equivalent:
(i) For all i ∈ I, Ki /F
Q is a separable algebraic field extension.
(ii) The compositum i Ki is a separable algebraic field extension.
Exercise 5.9. Prove Corollary 5.12.
Corollary 5.12 has the following important consequence: for any field extension
K/F , there exists a unique maximal separable algebraic subextension SepClK (F ),
the separable closure of F in K.

3. Purely Inseparable Extensions

Theorem 5.13. For an algebraic field extension K/F , the following are equiv-
alent:
(i) There is only one F -algebra embedding K ,→ K.
(ii) Every irreducible polynomial P ∈ F [t] with a root in K is purely insepa-
rable.
(iii) K is obtained by adjoining to F roots of purely inseparable polynomials.
(iv) The separable closure of K in F is F .
Exercise 5.10. Prove Theorem 5.13.
An extension satisfying the conditions of Theorem 5.13 is purely inseparable.
Exercise 5.11. a) Show: finite degree purely inseparable extensions form
a distinguished class.
b) Show: the purely inseparable algebraic extensions form a distinguished
class that is closed under composita.
In light of Exercise 5.9b), for any algebraic field extension K/F we may define the
purely inseparable closure of F in K to be the largest subextension of K which
is purely inseparable over F .
Exercise 5.12. Show that the purely inseparable closure of F in an algebraic
∞
closure F is the perfect closure F 1/p .
Corollary 5.14. Let K/F be a purely inseparable extension of finite degree.
Then [K : F ] is a power of p.
 4. STRUCTURAL RESULTS ON ALGEBRAIC EXTENSIONS 47

Proof. One may reduce to the case of a simple extension K = F [α], and then
a
α is purely inseparable over F so has minimal polynomial of the form tp − α for
some a ∈ Z+ . 
Corollary 5.15. A purely inseparable extension is normal.
Proof. This follows immediately from condition (i) of Theorem 5.13. 
The flavor of these results is that many formal properties are common to both
separable and purely inseparable extensions. The exceptions to this rule are the
following: first, purely inseparable extensions are always normal, whereas this is
most certainly not the case for separable extensions. A more subtle difference is
expressed in Theorem 5.13: if K/F is not purely inseparable, then it must have
a nontrivial separable subextension. However, if K/F is not separable, that does
not mean that it has a nontrivial purely inseparable subextension.
Example 5.16. [Mo96, p. 48]: Let k be a field of characteristic 2, F =
k(x, y) (rational function field), u a root in F of the separable irreducible quadratic
√
polynomial t2 +t+x, S = F (u) and K = S( uy). Clearly K/S is purely inseparable
and S/F is separable. But there is no nontrivial purely inseparable subextension of
K/F . Equivalently, we will show that if a ∈ K, a2 ∈ F , then already a ∈ F . An
√ √
F -basis for K is 1, u, uy, u uy. If a2 ∈ F , write
√ √
a = α + βu + γ uy + δu uy, α, β, γ, δ ∈ F.
Since a2 ∈ F , the coefficient of u = 0, i.e.,
β 2 + (γ + δ)2 y + δ 2 xy = 0.
If δ = 0 then β 2 + γ 2 y = 0, so γ = 0 since y is not a square in F . But then β = 0
and a ∈ F . If δ 6= 0, then
β 2 + (γ + δ)2 y γ β
x= 2
= ( + 1)2 + ( )2 y,
δ y δ δ
so that x ∈ F 2 (y), which is not the case. So δ = 0 and a ∈ F .

4. Structural Results on Algebraic Extensions

Proposition 5.17. Suppose an algebraic extension K/F is both separable and
purely inseparable. Then K = F .
Proof. For such an extension, let α ∈ K. Then the minimal polynomial of α
over F is both separable and purely inseparable. The only such polynomials have
degree one, i.e., α ∈ F . 
Proposition 5.18. For any algebraic field extension K/F , the extension K/ SepClK (F )
is purely inseparable.
Proof. Since SepClK (F ) is the maximal separable subextension of K/F , there
cannot be a proper nontrivial separable extension of K/ SepClK (F ), so it is purely
inseparable. 
In general this result is not valid the other way around: an algebraic field extension
K/F need not be separable over its purely inseparable closure. Indeed, in the
example of the previous section the purely inseparable closure Fi was F and K/F
was not separable. The following two results give more information on when K is
separable over Fi .
48 5. SEPARABLE ALGEBRAIC EXTENSIONS

Theorem 5.19. For an algebraic extension K/F , let Fs and Fi be, respec-
tively, the separable and purely inseparable closures of F in K. The following are
equivalent:
(i) K = Fs Fi .
(ii) K is separable over Fi .
Proof. (i) =⇒ (ii): K is obtained by adjoining to Fi roots of separable
polynomials with coefficients in F , hence by polynomials with coefficients in Fs .
(ii) =⇒ (i): If K/Fi is separable, then K/Fi Fs is separable. Similarly, since K/Fs
is inseparable, K/Fi Fs is inseparable. By Proposition 5.17, K = Fi Fs . 
Corollary 5.20. The equivalent conditions of Theorem 5.19 hold when K/F
∞
is normal. In particular they hold for F /F , giving F = F sep F 1/p .
Proof. Let α ∈ K \ Fi . The minimal polynomial P (t) ∈ Fi [t] of α has at least
one other distinct root, say β, in an algebraic closure F . Snce K/F is normal, also
K/Fi is normal, so we have β ∈ K, and the Extension Theorem (Theorem 4.22)
shows that there is an element s ∈ Aut(K/Fi ) such that s(α) = β. This shows that
the set of elements in F that are fixed by every element of Aut(K/Fi ) is Fi .
Later on we will see that we have proved that K/Fi is a Galois extension, which
implies that it is separable. For now we argue by hand: let α = α1 , . . . , αr be all
the Fi -conjugates of α in K, and consider the separable polynomial The
Yr
R(t) := (t − αi ) ∈ K[t].
i=1
The action of Aut(K/Fi ) on K extends to an action on K[t] by
Xn n
X
s( ai ti ) := s(ai )ti .
i=0 i=0
Since element s ∈ Aut(K/Fi ) permutes the roots αi of R, we have σ(R) = R, and
thus every coefficient of R is fixed by every element of Aut(K/Fi ), so R ∈ Fi [t]. It
follows that R = P , so P is separable, and thus K/Fi is (normal and) separable.
The second sentence of the Corollary follows immediately from the first. 
Corollary 5.21. For a finite degree extension K/F , we have
[K : F ]s = [SepClK (F ) : F ] | [K : F ].
Proof. We have [K : F ]s = [K : SepClK (F ) : F ]s [SepClK (F ) : F ]s . But the
separable degree of a purely inseparable extension is 1, so the conclusion follows. 
For a finite degree extension K/F one may therefore define the inseparable de-
gree [K : F ]i of a finite degree extension to be [K : F ]/[K : F ]s = [K : SepClK (F )].

A field is separably closed if it admits no proper separable algebraic field ex-

tension.
Proposition 5.22. The separable closure of a field in any algebraically closed
field is separably closed.
Exercise 5.13. Prove Proposition 5.22.
One often writes F sep for a separable closure of F . Like the algebraic and normal
closures, this extension is unique up to non-canonical F -algebra isomorphism.
 4. STRUCTURAL RESULTS ON ALGEBRAIC EXTENSIONS 49

Corollary 5.23. Let K/F be a normal algebraic extension. Then the sepa-
rable closure F s of F in K is also normal.
Proof. For any embedding σ of K into F , the image σ(F s ) lies in K (by
normality of K) and is evidently also a separable subextension of K/F . Therefore
we must have σ(F s ) = F s . 
Corollary 5.24. A field F is perfect iff its separable closure is algebraically
closed.
Proof. If F is perfect then all algebraic extensions are separable, so the result
is clear. Inversely, suppose that F is not perfect, so there exists α ∈ F \ F p
and a corresponding purely inseparable field extension F [α1/p ]/F defined by the
irreducible inseparable polynomial P = tp − α. By Theorem 5.10, only a separable
irreducible polynomial can acquire a root in a separable field extension, so the
polynomial P remains irreducible over the separable closure of F . 
Corollary 5.25. Let K/F be a finite degree field extension. Then F is perfect
iff K is perfect.
Proof. The result holds vacuously in characteristic 0, since all fields are per-
fect. So suppose that F (hence also K) has characteristic p > 0. If F is perfect,
then every algebraic extension of F is separable, hence every algebraic extension of
every algebraic extension of F is separable, and thus not only is every finite degree
extension of F perfect but every algebraic extension of F is perfect.
The new case is therefore when F is not perfect, so there is x ∈ F \ F p . We
−N −N
claim that there is a largest natural number N such that xp lies in K: if so, xp
lies in K \ K p , so K is not perfect. To establish the claim, we observe that since
x is not a pth power in F , Lemma 9.20 says that for all n ∈ Z+ the polynomials
n −n
tp − x ∈ F [t] are irreducible and thus we have [F (xp ) : F ] = pn . Thus when n
−n
is large enough so that [K : F ] < pn we cannot have xp ∈ K. This establishes
the claim and completes the proof. 
 CHAPTER 6

Norms, Traces and Discriminants

1. Dedekind’s Lemma on Linear Independence of Characters

Theorem 6.1 (Dedekind’s Lemma). Let M be a monoid and K a field. The
set X(M, K) of all monoid homomorphisms M → K × is linearly independent as a
subset of the K-vector space K M of all functions from M to K.
Proof. By definition, a subset of a vector space is linearly independent iff
every nonempty finite subset is linearly independent. So it’s enough to show that
for all N ∈ Z+ , every N -element subset of X(M, K) is linearly independent in K M .
We show this by induction on N . The base case, N = 1, is immediate: the only
one element linearly dependent subset of K M is the zero function, and elements
of X(M, K) are nonzero at all values of M . So suppose N ≥ 2, that every N − 1
element subset of X(M, K) is linearly independent, and let χ1 , . . . , χN be distinct
elements of X(M, K). Let α1 , . . . , αN ∈ K be such that for all x ∈ M , we have
(1) α1 χ1 (x) + . . . + αN χN (x) = 0.
Our goal is to show that α1 = . . . = αN = 0. Since χ1 6= χN , there is m ∈ M such
that χ1 (m) 6= χN (m). Substituting mx for x in (1), we get that for all x ∈ M ,
(2) α1 χ1 (m)χ1 (x) + α2 χ2 (m)χ2 (x) + . . . + αN χN (m)χN (x) = 0.
Multiplying (2) by χ1 (m)−1 and subtracting this from (1), we get
   
χ2 (m) χN (m)
(3) ∀x ∈ M, α2 − 1 χ2 (x) + . . . + αN − 1 χN (x) = 0.
χ1 (m) χ1 (m)
 
By induction, χ2 , . . . , χN are linearly independent, so αN χχN1 (m)
(m)
− 1 = 0 and thus
αN = 0. Thus (1) gives a linear dependence relation among the N − 1 characters
χ1 , . . . , χN −1 , so by induction α1 = · · · = αN −1 = 0. 

2. The Characteristic Polynomial, the Trace and the Norm

Let L/K be a field extension of degree n < ∞. For x ∈ L, the map x• : L → L given
by y ∈ L 7→ xy is an endomorphism of L as a K-vector space. That is, for all α ∈ K
and y1 , y2 ∈L, we have x(αy1 + y2 ) = x(αy1 + y2 ) = αxy1 + xy2 = α(xy1 ) + (xy2 ).
We may therefore analyze the element x ∈ L using tools of linear algebra.

Choose a K-basis b1 , . . . , bn for L. With respect to such a basis, the linear trans-
formation x• is represented by an n × n matrix, say M (x).
Example 6.2. Take K = R, L = C, and the basis (1, i). Let x = a + bi. Then
x • 1 = a · 1 + b · i and x • i = −b · 1 + a · i. Therefore
 
a −b
M (x) = .
b a
51
52 6. NORMS, TRACES AND DISCRIMINANTS

Example 6.3. If x lies in K, then M (x) = mi,j is simply the scalar matrix
diag(x, . . . , x). Proposition 6.4 below gives a generalization.
We define the characteristic polynomial of x:
n
Y
Px (t) = det(tIn − M (x)) = (t − λi ).
i=1

Similarly we define the trace

n
X n
X
TrL/K (x) = tr(M (x)) = mi,i = λi
i=1 i=1

and the norm

n
Y
NL/K (x) = det(M (x)) = λi .
i=1

Proposition 6.4. Let L/K/F be a tower of field extensions with m = [K : F ]

and n = [L : K]. Let x1 , . . . , xm be a basis for K/F and y1 , . . . , yn a basis for L/K.
a) For any element α ∈ K, if M is the matrix representing x• ∈ EndF (K)
with respect to {x1 , . . . , xm }, the matrix representation of x• ∈ EndF (L)
with respect to the basis {xi yj }1≤i≤m,1≤j≤n , reverse lexicographically or-
dered, is the block diagonal matrix diag(M, . . . , M ), i.e., n blocks, each
equal to M .
b) Let f (t) be the characteristic polynomial of α• ∈ EndF (K) and g(t) be the
characteristic polynomial of x• ∈ EndF (L). Then g(t) = f (t)[L:K] .
c) We have NL/F (x) = NK/F (x)[L:K] .
d) TrL/F (x) = [L : K] TrK/F (x).
Pm Pm
Proof. We have αxi = k=1 mki xk and hence αxi yj = k=1 mki (xk yj ).
This establishes part a). The remaining parts follow easily by standard linear
algebraic considerations. 

Corollary 6.5. Let L/F be a finite degree field extension. Let α be an el-
ement of L, let f (t) be the minimal polynomial of α over F , and let g(t) be the
characteristic polynomial of α• ∈ EndF (L). Then g(t) = f (t)[L:F (α)] .
Proof. Put K = F (α). The minimal polynomial f of α over F is the char-
acteristic polynomial of x• ∈ EndF (K). So the result follows from Proposition
6.4. 

Proposition 6.6. Let L/K/F be a tower of finite degree field extensions.

Then:
a) TrK/F : K → F is an F -linear map.
b) For all x, y ∈ K, NK/F (xy) = NK/F (x)NK/F (y).
c) For all c ∈ F and x ∈ K, NK/F (cx) = c[K:F ] NK/F (x).
Proof. Parts a) and b) are standard properties of the trace and determinant of
any F -linear map. Part c) follows by applying part b) and observing that for c ∈ F ,
NK/F (c) is the determinant of the scalar matrix diag(c, . . . , c), i.e., c[K:F ] . 

The following key result identifies the eigenvalues of α• in field-theoretic terms.

 2. THE CHARACTERISTIC POLYNOMIAL, THE TRACE AND THE NORM 53

Theorem 6.7. Let K/F be a field extension of degree n < ∞ and separable
degree ns . Put pe = nns = [K : F ]i . Let K be an algebraic closure of K. Let α ∈ K
and let f (t) be the characteristic polynomial of α• ∈ EndF (K). Let τ1 , . . . , τns be
the distinct F -algebra embeddings of K into K. Then
ns
Y e
f (t) = (t − τi (α))p .
i=1

It follows that
ns
Y e
(4) NK/F (α) = ( τi (α))p
i=1

and
m
X
(5) TrK/F (α) = pe τi (α).
i=1

Proof. Put L = F [α]. Let d = [L : F ], ds = [L : F ] and di = [L : F ]i .

Let σ1 , . . . , σds be the distinct F -algebra homomorphisms from L into F . For each
1 ≤ i ≤ ds , σi extends to ndss F -algebra homomorphisms from K into F . Let
ds
Y
f (t) = ( (t − σi (α)))di
i=1

be the minimal polynomial of α over F , and let g(t) be the characteristic polynomial
of α• on K, so by Corollary 6.5 we have
ds ds
!ni
ns
di n
Y Y
[K:L]
g(t) = f (t) = ( (t − σi (α)) d = ( (t − σi (α)) ds
i=1 i=1

ns
!pi
Y
= (t − τi (α)) .
i=1

Equations (4) and (5) follow immediately. 

Corollary 6.8. Let Fqd /Fq be an extension of finite fields. Then the norm
map N : F×
qd
→ F×
q is surjective.

Proof. Let σ : x 7→ xq , so that Aut(Fqd /Fq ) = h1, σ, . . . , σ d−1 }. Thus for

x ∈ Fqd ,
d−1 d−1
i Pd−1 q d −1
qi
Y Y
N (x) = σ i (x) = xq = x i=0 =x q−1 .
i=0 i=0

Therefore Ker N consists of all elements of the finite cyclic group F× qd

of order
q d −1 q d −1 × ∼ ×
dividing q−1 , so # Ker N = . Since F d / Ker N = N (F d ), we deduce that
q−1 q q
#N (F×
qd
) = q − 1: N is surjective. 
54 6. NORMS, TRACES AND DISCRIMINANTS

3. The Trace Form and the Discriminant

Let F be a field and V a finite-dimensional F -vector space equipped with a bilinear
form, i.e., a function h, i : V × V → F such that for all v, 1, v2 ∈ V and α ∈ F ,
hαv1 + v2 , v3 i = αhv1 , v3 i + hv2 , v3 i
and
hv1 , αv2 + v3 i = αhv1 , v2 i + hv1 , v3 i.
Let V ∨ = Hom(V, K) be the dual space of V . A bilinear form on V induces a linear
map Φ : V → V ∨ , namely
Φ(v) = hv, i.
(Note that a more careful notation would be something like ΦL : v 7→ hv, i, to
distinguish it from the other obvious map ΦR : v 7→ h , vi. We have ΦL = ΦR
iff the bilinear form is symmetric, an assumption which we have not (yet) made.
But in the general case the two maps are equally good, so let us work with Φ = ΦL
for simplicity.) We say that the bilinear form h, i is nondegenerate if Φ : V → V ∨
is an isomorphism. Since Φ is a linear map between two finite-dimensional vector
spaces of the same dimension, Φ is an isomorphism iff it is injective, i.e., for each
v ∈ V , if hv, wi = 0 for all w ∈ V , then v = 0.

Let h, i be a bilinear form on V , and fix a K-basis e1 , . . . , en of V . We define

the Gram matrix M of the bilinear form as M (i, j) = hei , ej i. Then for all
v, w ∈ V , we have
hv, wi = v T M w.
We claim that the nondegeneracy of the form is equivalent to the nonsingularity
of the Gram matrix M . If M is singular, so is M T , so there exists 0 6= v such
that v T M = (M v)T = 0, and thus hv, wi = 0 for all w ∈ V . Conversely, if M is
nonsingular, then for all 0 6= v ∈ V , v T M is nonzero, so it has at least one nonzero
component i, so v T M ei = hv, ei i 6= 0. (Note that this argument also makes clear
that ΦL is an isomorphism iff ΦR is an isomorphism.

Moreover, our fixed basis (e1 , . . . , en ) induces a dual basis (e∨ ∨

1 , . . . , en ), character-
∨
ized by ei (ej ) = δi,j (Kronecker delta) for all 1 ≤ i, j ≤ n. Thus, given a nondegen-
erate bilinear form h, i on V , we may pull back the dual basis (e∨ ∨
1 , . . . , en ) under Φ
−1
1 n j
to get a basis (e , . . . , e ) of V with the characteristic property hei , e i = δi,j . Con-
versely, if a basis (e1 , . . . , en ) of V exists which is dual to the given basis (e1 , . . . , en )
in the above sense, then the bilinear form is easily seen to be nondegenerate. In
summary:
Proposition 6.9. Let V be an n-dimensional vector space over a field K, let
h, i be a bilinear form on V , and let (e1 , . . . , en ) be any K-basis of V . Then the
following are equivalent:
(i) The induced map Φ = ΦL : V → V ∨ given by v 7→ hv, i is an isomorphism.
(ii) The induced map ΦR : V → V ∨ given by v 7→ h , vi is an isomorphism.
(iii) The Gram matrix M (i, j) = hei , ej i is nonsingular.
(iv) There exists a basis (e1 , . . . , en ) of V such that hei , ej i = δi,j .
And now, back to field theory: let K/F be a finite-dimensional field extension.
Define the trace form T : K × K → F , T (x, y) := Tr(x • y•). The bilinearity
of T follows immediately from the linearity of the trace map. Note that T is also
 3. THE TRACE FORM AND THE DISCRIMINANT 55

symmetric in the sense that T (x, y) = T (y, x) for all x, y ∈ K. A natural question
is when the trace form is nondegenerate.
Theorem 6.10. Let K/F be a field extension of finite degree n. The following
are equivalent:
(i) The trace form T : K × K → F is nondegenerate.
(ii) There exists some x ∈ K such that Tr(x) 6= 0.
(iii) The trace function Tr : K → F is surjective.
(iv) The extension K/F is separable.
Proof. The implications (i) =⇒ (ii) =⇒ (iii) may safely be left to the
reader.
(iii) =⇒ (iv): we argue by contraposition. If K/F is not separable, then char(F ) =
p > 0, [K : F ]i = pe is divisible by p, and thus (5) shows that the trace function is
identically zero.
(iv) =⇒ (i): By the Primitive Element Corollary, we have K = F [α] for some
α ∈ K. Then (1, α, . . . , αn−1 ) is an F -basis of K. Let x ∈ K. By Proposition 6.9,
it is enough to show that the Gram matrix M (i, j) = Tr(αi−1 αj−1 ) = Tr(αi+j−2 ) is
nonsingular.
Pn To see this, let α1 , . . . , αn be the distinct F
P-conjugates of α in K. Then
n
Tr(α) = i=1 αi , so that for any N ∈ N, Tr(αN ) = i=1 αiN . Now we introduce
the Vandermonde matrix V = V (α1 , . . . , αn ): V (i, j) = αji−1 . Why? Well, we
Pn
compute that the (i, j) entry of V V T is k=1 αki−1 αkj−1 = M (i, j). Therefore
Y
det M = det V V T = (det V )2 = ( αi − αj )2 6= 0. 
i>j

Example 6.11. (Trace form of a quadratic √ extension): Let F be a field of

characteristic different from 2, and let K = F ( D) be a quadratic field extension.
We√wish to explicitly compute the trace form. A natural choice of F -basis for K is
(1, D). The Gram matrix is then
 √   
Tr(1)
√ Tr( D) 2 0
M= = .
Tr( D) Tr(D) 0 2D
Thus the corresponding quadratic form is (2, 2D), of discriminant D ∈ K × /K ×2 .
 CHAPTER 7

The Primitive Element Theorem

1. The Alon-Tarsi Lemma

Lemma 7.1 (Alon-Tarsi). Let R be a domain, let f ∈ R[t1 , . . . , tn ] be a poly-
nomial, and for 1 Q ≤ i ≤ n let Ai ⊂ R be an infinite subset. Suppose that for all
n
a = (a1 , . . . , an ) ∈ i=1 Ai we have f (a) = 0. Then f = 0.
Proof. We go by induction on n. The n = 1 case is the familiar fact that a
nonzero univariate polynomial over a domain f ∈ R[t] cannot have more roots than
its degree, which follows from the “root-factor theorem” of high school algebra: if
f (a) = 0 then f = (t − a)g for some g ∈ R[t].
Suppose now that n ≥ 2 and that the result holds for polynomials in fewer
than n variables, and let f ∈ R[t1 , . . . , tn ]. Write
f = gd tdn + . . . + g1 tn + g0 , gi ∈ R[t1 , . . . , tn−1 ].
Qn−1
For each a0 = (a1 , . . . , an−1 ) ∈ A0 = i=1 Ai , the polynomial fa0 := f (a0 , tn ) ∈
R[tn ] has the property that for all an ∈ An we have fa0 (an ) = 0. Since An is
infinite, it follows from the base case that fa0 = 0, or in other words that for all
0 ≤ i ≤ d we have gi (a0 ) = 0. Since A0 is also infinite, it follows from our induction
hypothesis that gi = 0 for all 0 ≤ i ≤ d, and thus f = 0. 

2. The Primitive Element Theorem and its Corollary

Theorem 7.2. Let K/F be a finite degree field extension. The following are
equivalent:
(i) The set of subextensions L of K/F is finite.
(ii) K/F is monogenic: there exists α ∈ K such that K = F [α].
Proof. [LaFT, pp. 243-244] Suppose first that K = Fq is finite. Then (i) is
clear, while (ii) holds because K × is cyclic of order q − 1: if α is a generator of the
group K × , then K = F [α]. Henceforth we suppose that K is infinite.
(i) =⇒ (ii): observe that for any subextension E of K/F , since (i) holds for K/F ,
it also holds for E/F . Writing K = F [α1 , . . . , αn ], we see that it is enough to prove
the result in the case of extensions which are generated by two elements: a simple
dévissage/induction argument then recovers the general case.
So suppose that K = F [α, β]. As c ranges over the infinitely many elements
of F , there are only finitely many distinct subfields of K of the form F [α + cb], so
there exist distinct elements c1 , c2 of F such that
E = F [α + c1 β] = F [α + c2 β].
It then follows, successively, that (c1 − c2 )β ∈ E, β ∈ E, α ∈ E, so
F [α + c1 β] = E = F [α, β] = K.
57
58 7. THE PRIMITIVE ELEMENT THEOREM

(ii) =⇒ (i): Suppose K = F [α], and let f (t) ∈ F [t] be the minimal polynomial
for α over F . For each subextension E of K/F , let gE (t) ∈ E[t] be the minimal
polynomial for α over E. Let E 0 be the subextension of K/F generated by the
coefficients of gE . So F ⊂ E 0 ⊂ E ⊂ K; since gE is irreducible over E, it is also
irreducible over E 0 , and thus [K : E 0 ] = [E 0 [α] : E 0 ] = [E[α] : E] = [K : E]. It
follows that E = E 0 . In other words, E can be recovered from gE and thus the
map E 7→ gE is bijective. However, we also have that gE divides f for all E, so gE
is a monic polynomial whose multiset of roots in any algebraic closure is a subset
of the multiset of roots of f . So there are only finitely many possibilities for E. 

Corollary 7.3 (Primitive Element Corollary). Let K/F be a finite degree

separable field extension.
a) The equivalent conditions of Theorem 7.2 hold. In particular, there is
α ∈ K such that K = F [α].
b) Suppose F is infinite, and let S ⊂ F be an infinite subset. PnIf K =
F (x1 , . . . , xn ), then there are s1 , . . . , sn ∈ S such that K = F ( i=1 si xi ).
Proof. If F is finite, then so is K, so there are only finitely many subsets of
K and thus certainly only finitely many subextensions of K/F . So the equivalent
conditions of Theorem 7.2 hold in this case. Henceforth we assume that F is infinite.
We may assume that no xi lies in F , since if xi does then we may take si to be any
nonzero element of S.
Step 1: We claim that if F is an infinite field K = F [α, β] is a finite degree separable
field extension of F with α, β ∈ K \ F and S ⊂ F is any infinite subset, then are
s1 , s2 ∈ S such that [α, β] = F [α + s2 β].
Let σ1 , . . . , σn be the distinct F -algebra embeddings of K into an algebraic
closure F , and consider
Y
P (s, t) := s ((σi (α) − σj (α)) + t(σi (β) − σj (β)) ∈ K[s, t].
i6=j

Since σi and σj are distinct F -algebra embeddings of K = F [α, β], we must have
either σi (α) 6= σj (α) or σi (β) 6= σj (β), so P is nonzero. By Alon-Tarsi (Lemma
7.1) there are s2 ∈ S such that P (s1 , s2 ) 6= 0, which means that for 1 ≤ i ≤ n the
elements σi (s1 α + s2 β) are all distinct, so we have
F ⊂ F [s1 α + s2 β] ⊂ F [α, β]
and
[F [s1 α + s2 β] : F ] ≥ n = F [α, β],
so
F [α, β] = F [s1 α + s2 β].
Step 2: If we run through the above argument with
Y
P (t) := ((σi (α) − σj (α)) + t(σi (β) − σj (β)) ∈ K[t]
i6=j

in place of P (s, t), we get the similar conclusion that there is s2 ∈ S such that
K[α, β] = K[α + s2 β]. Step 3: We now finish by an inductive argument: let
2 ≤ i ≤ n − 1 and suppose that we have found s2 , . . . , si ∈ S such that
F [x1 , . . . , xi ] = F [s1 x1 + s2 x2 + . . . + si xi ].
 2. THE PRIMITIVE ELEMENT THEOREM AND ITS COROLLARY 59

(Step 1 establishes this for i = 2.) Then we have

F [x1 , . . . , xi , xi+1 ] = F [s1 x1 + s2 x2 + . . . + si xi , xi+1 ].
By Step 2, there is si+1 ∈ S such that
F [s1 x1 + s2 x2 + . . . + si xi , xi+1 ] = F [s1 x1 + s2 x2 + . . . + si xi + si+1 xi+1 ].
Taking i = n − 1 we get the desired conclusion. 
What we call the Primitive Element Corollary is often itself referred to as the
Primitive Element Theorem.
Corollary 7.4. (Lang) Let K/F be a separable algebraic extension such that:
there is n ∈ Z+ such that for all α ∈ K, [F (α) : F ] ≤ n. Then [K : F ] ≤ n.
Proof. Let α ∈ K be such that [F (α) : F ] has maximal degree – it is no loss
of generality to assume that this degree is n. We claim that K = F (α), which will
establish the result.
Suppose that K ) F (α), and let β ∈ K \ F (α). Since F (α, β)/F is finite
separable, by the Primitive Element Corollary (Corollary 7.3) there exists γ ∈ K
such that F (α, β) = F (γ). But then we must have [F (γ) : F ] > [F (α) : F ],
contradiction. 
Exercise 7.1. Show that the conclusion of Corollary 7.4 need not hold without
the separability hypothesis.
A more natural proof of Corollary 7.3 would be obtained by taking the normal clo-
sure M of K/F and using the Galois correspondence: the lattice of subextensions
of M/F is anti-isomorphic to the lattice of subgroups of Aut(M/F ), hence there
are certainly only finitely many of the former, which of course implies that there
are only finitely many subextensions of K/F . This brings us to our next topic,
Galois Theory.

Before we get there though, one remark: we have not yet seen a finite degree
field extension that is not monogenic, and Corollary 7.3 shows that such exam-
ples must be relatively exotic: in particular they must be inseparable and thus
can only occur in positive characteristic. Later we will see that in positive charac-
teristic such examples are relatively abundant: for instance, Exercise 13.10 shows
that if k is any field of characteristic p and k(x1 , . . . , xn ) is a rational function field
in n variables, then the field extension k(x1 , . . . , xn )/k(xp1 , . . . , xpn ) has degree pn
and requires n generators. Taking n = 2 for concreteness, we draw the following
remarkable conclusion.
Corollary 7.5. Let k be a field of characteristic p. The degree p2 field exten-
sion k(x, y)/k(xp , y p ) has infinitely many subextensions.
For me, the existence of a finite degree field extension with infinitely many subex-
tensions is the most startling result in these notes and, in fact, one of the most
counterintuitive mathematical results I have met in my adult mathematical life. (I
am fairly confident that I received my PhD in mathematics blissfully ignorant of
the fact that such field extensions existed. In fact I vaguely thought that standard
results in field theory rule them out...as is certainly the case in characteristic 0.)
Thus we find that the subfield lattice of a finite degree field extension, though both
60 7. THE PRIMITIVE ELEMENT THEOREM

Noetherian and Artinian – i.e., satisfying both the ascending and descending chain
conditions – may be infinite.
 CHAPTER 8

Galois Extensions

1. Introduction
For any field extension K/F we define Aut(K/F ) to be the group of F -algebra
automorphisms of K, i.e., the set of all field isomorphisms σ : K → K such that
σ(x) = x for all x ∈ F . This is a group under composition.

Let G be a subgroup of Aut(K/F ), i.e., a group of F -algebra automorphisms of K.

We define the fixed field

K G = {x ∈ K | σ(x) = x ∀σ ∈ G}.
Note that the notation comes from representation theory: if R is a commutative
ring, M an R-module and G is a group, then one has the notion of an R-linear
representation of G on M , i.e., a homomorphism from G to the group of R-module
automorphisms of M . In such a situation one can “take invariants”, i.e., consider
the subset of M on which G acts trivially: this is denoted M G . The present defi-
nition is an instance of this with R = F , M = K.

It is immediate to check that K G is a subextension of K/F . (In fact in the more

general setting detailed above, one checks that M G is an R-submodule of M .)

A field extension K/F is weakly Galois if K Aut(K/F ) = F . Equivalently, for

any element x ∈ K \ F , there exists σ ∈ Aut(K/F ) such that σ(x) 6= x.

A field extension K/F is Galois if for all subextensions L of K/F , K Aut(K/L) = L.

Remark: The terminology “weakly Galois” is not standard. In fact, it is usual

to consider Galois theory only for algebraic extensions and in this case it will turn
out to be the case that the notions of weakly Galois and Galois coincide.

This “top down” definition of a weakly Galois extension is the generalization to

arbitrary extensions of a definition of E. Artin for finite degree extensions. It has
the merit of making it easy to exhibit a large class of weakly Galois extensions: if
K is any field and G is any group of automorphisms of K, then K/K G is, tauto-
logically, a Galois extension.
Example 8.1. Let G be the 2-element subgroup of the complex numbers gener-
ated by complex conjugation. Then CG = R, so C/R is a Galois extension.
Example 8.2. Let L/K be a separable quadratic extension, so that L = K[t]/(P (t)),
where P (t) is a separable polynomial. Then P (t) splits over L into (t − α)(t − α),
61
62 8. GALOIS EXTENSIONS

so that the automorphism group of L/K has order 2, the nontrivial element being
the unique K-automorphism σ of L which sends α 7→ α. Since LAut(L/K ) is a
subextension of the degree 2 extension L/K, it could only be L or K, and since
σ(α) = α 6= α, we conclude that the fixed field is K and the extension is Galois. In
contrast the automorphism group of an inseparable quadratic extension is trivial, so
this extension is not Galois.
√
Example 8.3. Let K = Q[t]/(t3 −2) = Q[ 3 2]. Since K contains exactly one of
the three roots of t3 − 2 in Q, Aut(K/Q) is the trivial group and K/Q is not Galois.√
On the other hand, the automorphism group of the normal closure M = Q[ζ3 , 3 2]
of √
K/Q has order 6: since everything is separable, there are three embeddings of
Q[ 3 2] into M , and each of these extends in two ways to an automorphism of M .
Any automorphism s of M is determined by an i ∈ {0, 1, 2} and j ∈ {0, 1} such
that √ √ j
s : 2 7→ ζ3i 2, ζ3 7→ (ζ3 )(−1) .
3 3

Since there are six possibilities and six automorphisms, all of these maps must
indeed give √automorphisms.
√ In particular, there is an order 3 automorphism√σ
which takes 3 2 7→ ζ3 3 2 and fixes ζ3 and an order 2 automorphism τ which fixes 3 2
and maps ζ3 7→ ζ3−1 . One checks that τ στ = τ στ −1 = σ −1 , i.e., Aut(L/Q) ∼ = S3 ,
the symmetric group on three elements. Indeed, these three elements can be viewed
as the three roots of t3 − 2 √ in M . Finally, the subgroup fixed by {1, σ} is precisely
K, whereas the generator 3 2 of K/Q is not fixed by σ, so that we conclude that
M Aut(M/Q) = Q and M/Q is Galois.
These examples already suggest that a finite degree extension K/F is Galois iff it
is normal and separable, and in this case # Aut(K/F ) = [K : F ]. We will show in
the next section that these conditions are all equivalent.
Example 8.4. The extension Q/Q is Galois. We cannot show this by some sort
of direct computation of GQ := Aut(Q/Q): this group is uncountably infinite and
has a very complicated structure. Indeed, as an algebraic number theorist I am more
or less honorbound to inform you that the group GQ is the single most interesting
group in all of mathematics! We will see that the Galois theory of infinite algebraic
extensions cannot be developed in exactly the same way as in the finite case, but is,
in theory, easily understood by a reduction to the finite case.
Example 8.5. The extension C/Q is Galois, as is C/Q. In particular the
automorphism group of the complex field is (much) larger than just {1, c}. In fact we
will show that if F has characteristic zero and K is algebraically closed, then K/F
is Galois. These results are not part of “Galois theory” as it is usually understood,
but rather are facts about automorphism groups of transcendental extensions. These
results will be shown in §10.1.
Example 8.6. For any field F , Aut(F (t)/F ) is the group  of linear fractional
a b
transformations: the group GL2 (F ) of 2×2 matrices with ad 6= bc acts by
c d
automorphisms on F (t), via t 7→ at+b
ct+d . Scalar matrices – those with b = d = 0, a =
c – act trivially, so the action factor through to the quotient P GL2 (F ) of GL2 (F )
by the subgroup F × of scalar matrices. It is a standard fact (more in the vein
of algebraic geometry than pure field theory) that this is the entire automorphism
group of F (t).
 2. FINITE GALOIS EXTENSIONS 63

Proposition 8.7. The extension F (t)/F is weakly Galois iff F is infinite.

Proof. We will need to use a fact from the next section: if G is a finite group
of automorphisms acting on a field K, then [K : K G ] = #G < ∞. Therefore if
F is finite, F (t)Aut(F (t)/F ) has finite index in F (t), so is certainly not equal to F .
Conversely assume F is infinite. . . 

Remark Aside: I am not aware of a simple necessary and sufficient condition for
an extension K/F which is finitely generated, but of infinite degree, to be Galois.
When K/F is regular of transcendence degree 1 (two terms which we have not
yet defined), one can give such a criterion in terms of the Jacobian J(C) of the
corresponding algebraic curve C/F , namely K/F is Galois iff dim J(C) = 0 or
(dim J(C) = 1 and J(C)(F ) is infinite). In particular no such field of genus g ≥ 2
is Galois. One can give some examples of Galois extensions of higher transcendence
degree – e.g. the proof of Proposition XX easily adapts to show that F (t1 , . . . , tn )/F
is Galois if F is infinite – but the general problem seems to be a quite subtle one
in birational arithmetic geometry.

2. Finite Galois Extensions

Theorem 8.8. If K/F is a finite degree field extension, Aut(K/F ) is a finite
group of cardinality at most [K : F ].
Proof. First recall that the set of F -algebra embeddings σ of K into an al-
gebraic closure F is finite, so in particular the subset of such with σ(K) = K is
finite. This holds because K = F (α1 , . . . , αn ), and an embedding σ is determined
by sending each αi to one of the at most di = [F [αi ] : F ] roots of the minimal
polynomial of αi over F in F . Therefore the set of such embeddings has cardinality
at most d1 · · · dn . Note that when K = F [α] is simple this is exactly the bound we
want, so that e.g. if K/F is separable we are already done.
Now for the general case. Let Aut(K/F ) = {σ1 , . . . , σN } and suppose, for a
contradiction, that N > m = [K : F ]. Let α1 , . . . , αm be an F -basis for K, and
consider the N × m matrix A whose (i, j) entry is σi (αj ). This matrix has rank at
most m < N , so that its rows are K-linearly dependent: there exist c1 , . . . , cN ∈ K,
not all 0, such that for all 1 ≤ j ≤ m we have
X
ci σi (αj ) = 0.
i
×
P
For each x ∈ K , there exist a1 , . . . , am in F such that x = j aj αj . Then
X X X X X
ci σi (x) = ci σi ( aj αj ) = ci (aj σj (αj ))
i i j i j
X X
= aj ( ci σi (αj )) = 0.
j i

But taking M = K all the automorphisms σi give characters M → K × hence are

×

K-linearly independent. Therefore in the last equation we must have ci = 0 for all
i, a contradiction. 

Proposition 8.9 (Artin). Let K be a field and G a finite group of automor-

phisms of K, of cardinality n. Then we have [K : K G ] = n.
64 8. GALOIS EXTENSIONS

Proof. Step 1: We show that K/K G has finite degree.1

Let α ∈ K, and let S = σ1 , . . . , σr be a maximal subset of G such that the
elements σi (α) are distinct in K. It follows that for all τ ∈ G, the r-tuple v =
(τ σ1 α, . . . , τ σr α) differs from w = (σ1 α, . . . , σr α) by a permutation: indeed, since
τ is injective, the components of w are all distinct, and if they were not simply
a reordering of the components of v, this would contradict the maximality of S.
Therefore α is a root of the polynomial
t
Y
f (t) = (t − σi α),
i=1

a polynomial with coefficients in K G . Moreover, f (t) is separable, and thus K/K G

is separable. Corollary 7.4 applies to show that K/K G has finite degree, indeed
degree equal to the maximal degree [K G (α) : K G ] of an element α ∈ K.
Step 2: Above, for each α we constructed a polynomial satisfied by α of degree
r ≤ n, it follows that [K G : K] ≤ n. On the other hand, by Theorem 8.8 we
have n = #G ≤ # Aut(K/K G ) ≤ [K G : K]. We conclude [K : K G ] = n and
G = Aut(K/K G ). 
Theorem 8.10 (Omnibus Theorem for Finite Galois Extensions). Let K/F be
a finite degree extension. The following are equivalent:
(i) K Aut(K/F ) = F (“K/F is Galois.”)
(ii) # Aut(K/F ) = [K : F ].
(iii) K/F is normal and separable.
(iv) K/F is the splitting field of a separable polynomial.
Proof. Let G = Aut(K/F ). (i) implies (ii) by Proposition 8.9. (ii) implies
(i): we have F ⊂ K G ⊂ K, and [K : K G ] = #G = [K : F ], so K G = F .
(iii) implies (iv): if K/F is separable then by the Primitive Element Theorem
K = F [t]/(P (t)) for some irreducible, separable polynomial P . Since it is normal,
P splits in K and therefore K/F is the splitting field of the separable polynomial P .
(iv) implies (iii) is essentially the same: since K/F is a splitting field, it is normal;
since it is obtained by adjoining roots of separable polynomials, it is separable.
(iv) ⇐⇒ (ii): We know that the number of embeddings of K into F is equal
to the separable degree of K/F and that this equals [K : F ] iff K/F is separable;
moreover, every F -algebra embedding s : K → F has s(K) = K – i.e., gives an
automorphism of K iff K/F is normal. 
Corollary 8.11. Let K/F be a finite degree field extension.
a) The extension K is a subextension of a finite degree Galois extension L/F
iff K/F is separable.
b) If K/F is separable, any algebraic closure F of K contains a unique min-
imal extension M of K such that M/F is Galois, namely the normal
closure of K/F in F .
Proof. Since Galois extensions are separable and subextensions of separable
extensions are separable, for K/F to be contained in a finite Galois extension it
is clearly necessary for it to be separable. If so, then the normal closure M of
1In many standard treatments of finite Galois theory, the finiteness of K/K G is an additional
assumption. Our source for this stronger version is Lang’s Algebra.
 3. AN ABSTRACT GALOIS CORRESPONDENCE 65

K/F , being a compositum of the separable extensions s(K) as s ranges over the
finite set of distinct F -algebra embeddings of K into F is separable and normal,
hence Galois. M/K is even the minimal extension of K which is normal over F , so
certainly it is the minimal such Galois extension. 

In view of Corollary 8.11, it is reasonable to call the normal closure of a finite degree
separable field extension the Galois closure.
Theorem 8.12 (Natural Irrationalities). Let K/F be a finite degree Galois
extension, and let L/F be an arbitrary field extension. Then:
a) The field extension KL/L is Galois.
b) The restriction map r : Aut(KL/L) → Aut(K/K ∩ L) is an isomorphism.
c) We have [KL : L] = [K : K ∩ L].
Proof. a) This is the assertion that finite Galois extensions have the base
change meta-property. But all of the following properties have the base-change
meta property: being of finite degree, normality and separability. Alternately, since
K/F is finite Galois, it is the splitting field of the separable polynomial f ∈ F [x].
Then KL/L is the splitting field of the polynomial f ∈ L[x], which is still separable
because of the Derivative Criterion.
b) Let σ ∈ Aut(KL/L), and let r(σ) denote the restriction of σ to K. Since σ
fixes L pointwise and F ⊂ L, also σ fixes F pointwise. So for all x ∈ K, r(σ)(x)
is an F -conjugate of x; since K/F is normal, this implies r(σ)(x) ∈ K and thus
r(σ) ∈ Aut(K/F ). Indeed, because σ pointwise fixes L, r(σ) pointwise fixes K ∩ L
and r(σ) ∈ Aut(K/K ∩ L). This defines a map
r : Aut(KL/L) → Aut(K/K ∩ L).
That r is a group homomorphism is immediate. Moreover, the kernel of r consists
of the set of automorphisms α of KL that pointwise fix both K and L and thus
also pointwise fix KL: r is injective. Finally we must show that α is surjective.
Its image is a subgroup of Aut(K/K ∩ L), which by the Galois correspondence is
therefore of the form Aut(K/E) for some K ∩ L ⊂ E ⊂ K. Now observe that E
is pointwise fixed by every α ∈ Aut(KL/L), so hence E ⊂ (KL)Aut(KL/L) = L. It
follows that E ⊂ K ∩ L and thus E = K ∩ L and α is surjective.
c) By part b) we have
[KL : L] = #Aut(KL/L) = # Aut(K/K ∩ L) = [K : K ∩ L]. 

3. An Abstract Galois Correspondence

Let X be a set and G a group of automorphisms of X, i.e., a subgroup of the group
Sym(S) of all bijections s : X → X. Let Λ(X) be the collection of all subsets of
X and Λ(G) be the collection of all subgroups of G. Both Λ(X) and Λ(G) are
partially ordered sets under inclusion.

For a subset Y ⊂ X, we define

GY = {g ∈ G | gy = y∀y ∈ Y },
which is a subgroup of G. Dually, for a subgroup H of G, we define
X H = {x ∈ X | gx = x∀g ∈ H},
66 8. GALOIS EXTENSIONS

which is a subgroup of H. (We could define in the same way X S for any subset
S ⊂ G, but one checks immediately that if H is the subgroup generated by S,
X S = X H , so this extra generality leads nowhere.) To be very formal about it, we
have thus defined a map
Φ : Λ(X) → Λ(G), Y 7→ GY
and a map
Ψ : Λ(G) → Λ(X), H 7→ X H .
Let us explore what can be said about these two maps in this extreme level of gen-
erality. Statements that we do not prove are exercises in unwinding the definitions
and left to the reader. (We do recommend that the reader peform these exercises!)

First, both Φ and Ψ are anti-homomorphisms of the partially ordered sets, i.e.,
if Y1 ⊂ Y2 , then Φ(Y2 ) ⊂ Φ(Y1 ), and similarly if H1 ⊂ H2 then Ψ(H2 ) ⊂ H1 . This
implies that Ψ ◦ Φ : Λ(X) → Λ(X) and Φ ◦ Ψ : Λ(G) → Λ(G) are homomorphisms
of partially ordered sets:
Y1 ⊂ Y2 =⇒ X GY1 ⊂ X GY2 ,

H1 ⊂ H2 =⇒ GX H1 ⊂ GX H2 .
Moreover, for all Y ⊂ X and H ⊂ G we have

(GC) Y ⊂ X H ⇐⇒ H ⊂ GY .

Indeed, both containments assert precisely that every element of H acts trivially on
every element of Y . If H = GY we certainly have the second containment, therefore
by (GC) we have
(6) Y ⊂ X GY .
Dually with Y = X H we certainly have the first containment hence (GC) gives
(7) H ⊂ GX H .
Proposition 8.13. Let H be a subgroup of G, Y a subset of X and σ ∈ G.
We have:
a) σGY σ −1 = GσY .
−1
b) σX H = X σHσ .
Proof. We have g ∈ GσY ⇐⇒ ∀y ∈ Y, gσy = σy ⇐⇒ ∀y ∈ Y, σ −1 gσy =
y ⇐⇒ σ −1 gσ ∈ GY ⇐⇒ g ∈ σGY σ −1 . Similarly, y ∈ σX H ⇐⇒ σ −1 y ∈
X H ⇐⇒ ∀h ∈ h, hσ −1 y = σ −1 y ⇐⇒ ∀h ∈ H, (σhσ −1 )y = y ⇐⇒ y ∈
σHσ −1 . 

Let us now introduce the following simplified (and symmetric) notation: for Y ⊂ X,
we write Y 0 for GY ; for H ⊂ G, we write H 0 for X H . Equations (6) and (7) now
read as Y ⊂ Y 00 and H ⊂ H 00 . Let us call a subset Y of X (resp. a subgroup H of
G) closed if Y 00 = Y ( resp. if H 00 = H).
Proposition 8.14. For any Y ∈ Λ(X) and H ∈ Λ(G), we have Y 0 = Y 000 and
H = H 000 . Hence Y 0 is a closed subgroup of G and X 0 is a closed subset of X.
0
 3. AN ABSTRACT GALOIS CORRESPONDENCE 67

Proof. By (6) we have Y 0 ⊂ (Y 0 )00 and Y ⊂ Y 00 . Applying a prime to the

latter containment reverses it and hence gives Y 0 ⊃ (Y 00 )0 . Therefore Y 0 = Y 000 .
The argument for H is identical. 
Remark: This shows that the operators 00 on the posets Λ(X) and Λ(G) are what
are called closure operators. In general, if (S, ≤) is a partially ordered set, then a
map c : S → S is a closure operator if for all s ∈ S, s ≤ c(s), s ≤ t =⇒ c(s) ≤ c(t)
and c(c(s)) = c(s) for all s ∈ S.
Corollary 8.15. Let Λc (X) be the closed subsets of X and Λc (G) be the closed
subgroups of G. Let Φc be Φ restricted to Λc (X) and Ψc be Ψ restricted to Λc (G).
Then
Φc : Λc (X) → Λc (G), Ψc : Λc (G) → Λc (X)
give mutually inverse anti-automorphisms of posets.
In fact the proof is immediate from the previous result; again, it is a good exercise
for the reader to chase through the definitions and notation to see this.
Corollary 8.16. a) A closed subgroup H of G is normal iff its corresponding
closed subset Y = H 0 = H 000 is stable under all automorphisms of G: for all σ ∈ G,
σY = Y .
b) A closed subset Y of X is stable under all automorphisms of G iff the corre-
sponding closed subgroup H = Y 0 = Y 000 is normal in G.
Again, this follows immediately from Proposition XX.
Exercise 8.1. Show that if H is a normal subgroup of G, so is its closure H 00 .
Example 8.17. Suppose #X > 1; let x ∈ X and take Y = X \ x. Then Y is
not a closed subset of X, since any group of automorphisms of X which fixes every
element of Y must also fix x.
Example 8.18. If X = {1, 2} and G = S2 is the full symmetry group of X,
then the closed subsets are ∅ and X; the corresponding closed subgroups are G and
the trivial subgroup e. In particular all subgroups are closed. If X = {1, 2, 3}
and G is the full symmetry group S3 . The closed subsets are ∅, {1}, {2}, {3}
and X. The corresponding closed subgroups are S3 , h(23)i, h(13)i, h(12)i and the
trivial subgoup e. In particular the (unique) subgroup H = h(123)i of order 3 is not
closed: H 0 = ∅ and H 00 = S3 . If X is any set, the only subsets invariant under
G = Sym(X) are ∅ and X itself, so Sym(S) does not have any nontrivial, proper
closed normal subgroups. On the other hand, if #X ≥ 3 then Sym(S) always has a
nontrivial, proper normal subgroup (i.e., it is not a simple group): if S is finite, so
Sym(S) ∼ = Sn take the alternating group An (the only possible choice if n ≥ 5); if
S is infinite, take the the subgroup H of elements g ∈ Sym(S) such that X \ X hgi
is finite. Then #H = #S while # Sym(S) = 2# Sym(S) .
Example 8.19. Let K/F be a field extension, X = K and G = Aut(K/F ).
Then every closed subset of X is a subextension K H of K/F . Corollary XX shows
that there is a bijective correspondence between the closed subextensions of K/F
and the closed subgroups of Aut(K/F ). Of course the key word in the previous
sentence is “closed”: if e.g. Aut(K/F ) is the trivial group (e.g. K = R) then the
statement is completely vacuous. In the next section we will show that if K/F is a
finite Galois extension, the best possible behavior occurs.
68 8. GALOIS EXTENSIONS

Exercise 8.2. Let Λ and Λ0 be two partially ordered sets. A Galois connec-
tion between Λ and Λ0 is a pair of order-reversing maps Φ : Λ → Λ0 , Ψ : Λ0 → Λ
satisfying the analogoue of identity (GC)2 above: for x ∈ Λ, y ∈ Λ0 , Φ(x) ≤ y ⇐⇒
x ≤ Ψ(y).
a) Check that the entire discussion (except for the bit about conjugation and nor-
mality) goes through in this level of generality: we get closure operators on Λ and
Λ0 such that Φ and Ψ give mutually inverse anti-automorphisms on the subsets of
∼ ∼
closed elements: Φ : Λc → Λ0c , Ψ : Λ0c :→ Λc .
b) Look for Galois connection. in your everyday (mathematical) life, paying spe-
cial attention to the closure process. For example, consider the polynomial ring
R = k[t1 , . . . , tn ] over an algebraically closed field k. Let Λ be the set of ideals I
of R, and let Λ0 be the set of algebraic
Tr subsets of affine n-space An over k: that
n −1
is, the subsets of k of the form i∈I Pi (0), where {Pi }i∈I is a set of elements
of R. Define Φ : Λ → Λ0 by I 7→ V (I), the set of points of (a1 , . . . , an ) such that
P (a1 , . . . , an ) for all P ∈ I. Define Ψ : Λ0 → Λ by S 7→ I(S), the ideal of all
elements of R which vanish at every (x1 , . . . , xn ) ∈ S. It is no problem to see that
this gives a Galois connection. What are the closed ideals? What are the closed
algebraic subsets?

4. The Finite Galois Correspondence

Let K/F be a finite Galois extension, so that by the general nonsense of the previous
section, we get a bijective correspondence between closed subextensions L of K and
closed subgroups of G = Aut(K/F ).
Theorem 8.20. (Fundamental theorem of Galois theory) If K/F is finite Ga-
lois, then every subgroup H of G = Aut(K/F ) is closed, i.e., of the form H =
Gal(K/L) for a unique subextension L/K. Conversely, every subextension L is
closed, i.e., of the form K H for a unique subgroup H of G. Therefore the maps
L 7→ Gal(K/L) and H 7→ K H give mutually inverse inclusion-reversing bijections
between the set of subextensions of L/K and the set of subgroups of G. Moreover,
a subextension L is Galois over F iff the corresponding subgroup Gal(K/L) is nor-
mal in G, and in this case Aut(L/F ) is canonically isomorphic to the quotient
Aut(K/F )/ Aut(K/L).
Proof: Let L be a subextension of K/F . It is clear that L ⊂ K Aut(K/L) . But
by XXX we know that [K : K Aut(K/L) ] = # Aut(K/L). Since K/F is Galois, so
is K/L, hence # Aut(K/L) = [K : L]. Therefore we must have K Aut(K/L) = L.
Moreover, if H is a subgroup of G, we again clearly have H ⊂ GK H ; but we also
#G
have [G : GK H ] = # Aut(K/K H ) = [G : H], so H = GK H . This shows the Galois

correspondence is perfect. Now applying Corollary XX we get that H = Gal(K/L)

is normal in G iff L is stable under all F -algebra automorphisms σ of K. Since
K/F is itself normal, this holds iff L is stable under all F -algebra embeddings into
an algebraic closure F , i.e., iff L/F is normal. Finally, suppose that L/F is normal.
Then every F -automorphism of K restricts to an F -automorphism of L, giving a
natural map Aut(K/F ) → Aut(L/F ) which is easily checked to be a homomor-
phism of groups. The map is surjective by the Extension Theorem XX. Its kernel is
the subgroup of F -algebra automorphisms of K which fix every element of L, i.e.,

2In particular “GC” stands for Galois Connection.

 4. THE FINITE GALOIS CORRESPONDENCE 69

Aut(K/L).

This theorem is probably the single most important result in field theory. It re-
duces the study of the lattice of subextensions of a finite Galois extension K/F
to the corresponding lattice of subgroups of the finite group Aut(K/F ), which is
much easier to study, e.g. is a priori finite. Indeed, if K/F is any finite separable
extension, then one may – and should! – apply the Galois correspondence to the
Galois closure M/F .
Exercise 8.3. Use the Galois Correspondence to give a more natural proof of
the Primitive Element Corollary (7.3).
When K/F is Galois, we write Gal(K/F ) for Aut(K/F ) and speak of Gal(K/F ) as
the Galois group of K/F . We note that some authors (e.g. Shifrin, Kaplansky)
use the notation Gal(K/F ) for the automorphism group of an arbitrary field exten-
sion, but from the perspective of infinite Galois theory (coming up!) and modern
number theory this seems dangerously misleading. Namely, it would then be tempt-
ing to call any automorphism group of a finite degree extension “a Galois group”
and this is most certainly at odds with contemporary terminology. Indeed, perhaps
the single outstanding problem in field theory is to decide whether, for any finite
group G, there is a Galois extension K/Q such that Gal(K/Q) ∼ = G. However, the
corresponding statement that any finite group is the automorphism group of some
finite extension K/Q – possibly with [K : Q] > #G – is a much weaker one, and
indeed this is a known theorem of E. Fried and J. Kollar [FK78]

Composita of Galois extensions: let F be a field and K1 , K2 two Galois ex-

tensions of F . After choosing an algebraic closure F of F , since K1 and K2
are splitting fields, there is a unique F -algebra embedding of Ki into F . Since
composita of normal (resp. separable) extensions are normal (resp. separable),
the compositum K = K1 ∨ K2 is a finite Galois extension. What is the rela-
tionship of Gal(K/F ) to Gal(K1 /F ) and Gal(K2 /F )? As above we get surjec-
tive restriction maps ιi : Gal(K/F ) → Gal(Ki /F ), and hence a diagonal map
ι = (ι1 , ι2 ) : Gal(K/F ) → Gal(K1 /F ) × Gal(K2 /F ). This composite homomor-
phism ι need not (of course?) be surjective: e.g. it will not be if K1 = K2 are
nontrivial extensions of F . Rather ι is always injective: since K is generated as a
field by K1 and K2 , a pair of automorphisms σi of Ki can extend in at most one
way to an automorphism of K. Therefore Gal(K/F ) can naturally be viewed as a
subgroup of the product Gal(K1 /F ) × Gal(K2 /F ).

This is in fact rather useful: let C be any class of finite groups which is closed under
formation of direct products and passage to subgroups, and suppose that Ki /F are
two C-Galois extensions, i.e., finite Galois extensions whose Galois groups lie in C.
Then the compositum K1 ∨ K2 is a C-Galois extension. E.g. we may profitably
take C to be the class of all finite abelian groups, or the class of all finite solvable
groups. When we turn to infinite Galois theory we will see that we are allowed to
take infinite composita as well, and this observation will show that any field admits
a maximal C-Galois extension.
Exercise 8.4. Let K1 , K2 /F be two finite Galois extensions, and K = K1 K2
their compositum. Let H be the image of the map ι : Gal(K/F ) → Gal(K1 /F ) ×
70 8. GALOIS EXTENSIONS

Gal(K2 /F ). Show that H is normal in Gal(K1 /F ) × Gal(K2 /F ), and that the

quotient (Gal(K1 /F ) × Gal(K2 /F ))/H ∼
= Gal(K1 ∩ K2 /F ). In particular, ι is an
isomorphism iff K1 ∩ K2 = F .

5. The Normal Basis Theorem

Let K/F be a finite degree field extension. Then a basis {α1 , . . . , αn } of K as an F -
vector space is a normal basis if all of its elements lie in the same Aut(K/F )-orbit,
i.e., if for all 1 ≤ i ≤ n there exists σ ∈ Aut(K/F ) such that αi = σα1 .
Exercise 8.5. If a finite degree extension K/F admits a normal basis, then
K/F is Galois.
The main result of this section is the converse: every finite Galois extension admits
a normal basis. A lot of literature has been written on this result. Our treatment
follows [CW50] and [We09, §3.6]. It is certainly not the shortest treatment avail-
able, but it proceeds by establishing several preliminary results which are of some
interest in their own right.

Every known proof of the existence of normal bases must negotiate a fundamental
dichotomy between finite fields and infinite fields. This dichotomy comes up several
times in field theory, algebra and algebraic geometry (another good example of a
theorem for which the finite field case must be taken separately is the Noether
Normalization Theorem), but often without much fanfare our explanation. To
our mind at least, the source of the trouble is the different behavior of the evalu-
ation map on polynomials over finite domains versus infinite integal domains. (A
geometer might point to the fact that for any n ∈ Z+ , a field K is infinite iff the
K-rational points of affine n-space over K are Zariski dense, but in fact this comes
down to the same algebraic observation.)
Lemma 8.21. Let R ⊂ S be an extension of domains and n ∈ Z+ . The following
are equivalent:
(i) For all f ∈ S[t1 , . . . , tn ], f (a1 , . . . , an ) = 0 for all (a1 , . . . , an ) ∈ Rn =⇒ f = 0.
(ii) R is infinite.
Proof. (i) =⇒ (ii): We prove the contrapositive. Note that any finite domain
is a field, so suppose R = Fq . Let f (t) = tq1 − t1 . Then for all a = (a1 , . . . , an ) ∈ Fnq ,
f (a) = aq1 − a1 = 0.
(ii) =⇒ (i): We go by induction on n.
Base Case (n = 1): suppose f ∈ S[t] is a polynomial which is not the zero polyno-
mial. Then it has degree d ≥ 0 and by the Root-Factor Theorem has at most d roots
in the fraction field of R, hence a fortori at most d roots in R. But #R ≥ ℵ0 > d,
so there exists a1 ∈ R with f (a1 ) = 0.
Induction Step: Suppose n > 1 and that every polynomial in n − 1 vari-
ables with S-coefficients which is not the zero polynomial has a R-rational root.
Let f (t1 , . . . , tn−1 , z) ∈ S[t1 , . . . , tn−1 , z]. Put S 0 = S[t1 , . . . , tn−1 ], so f may
be identified with a nonzero polynomial g(z) ∈ S 0 [z]. Applying the Base Case,
there exists A ∈ R0 such that 0 6= g(A) ∈ R0 . Now g(A) is a nonzero element
of S 0 = S[t1 , . . . , tn−1 ], so by induction there exist a1 , . . . , an−1 ∈ R such that
g(A(a1 , . . . , an−1 )) 6= 0. Putting an = A(a1 , . . . , an−1 ) we have
f (a1 , . . . , an−1 , an ) = g(A(a1 , . . . , an−1 )) 6= 0.
 5. THE NORMAL BASIS THEOREM 71


Proposition 8.22. Any finite cyclic extension K/F admits a normal basis.
Proof. Let K/F be cyclic of degree n with Gal(K/F ) = hαi. We may en-
dow K with the structure of an F [t]-module extending its F -module structure by
putting t · x = σ(x) for all x ∈ K. Then tn − 1 annihilates K; moreover, by linear
independence of characters, no smaller degree polynomial does so. It follows that
as an F [t]-module, K is isomorphic to F [t]/(tn − 1). Thus there exists α ∈ K
such that ann(α) = (tn − 1) – take, e.g., the preimage of 1 (mod tn − 1) under an
isomorphism – so the elements α, σα, σ 2 α, . . . , σ n−1 α are F -linearly independent
and thus give a normal basis. 
Lemma 8.23. Let K/F be a degree n Galois extension, and write Aut(K/F ) =
{σi }ni=1 . For α1 , . . . , αn ∈ K, the following are equivalent:
(i) α1 , . . . , αn is an F -basis of K.
(ii) The matrix A ∈ Mn (K) with Aij = σi αj is nonsingular.
Proof. (i) =⇒ (ii) follows almost immediately from the (K-)linear indepen-
dence of the characters σ1 , . . . , σn : details are left to the reader.
(ii) =⇒ (i): We argue by contraposition: suppose α1 , . . . , αn is not an F -basis for
K, so there exist a1 , . . . , an ∈ F , not all zero, with a1 α1 + . . . + an αn = 0. Then
for all i we have
X n Xn Xn
aj Aij = aj σi αj = σi ( ai αj ) = 0,
j=1 j=1 j=1

which shows that the columns of the matrix A are linearly dependent. 
By linear independence of characters, for any field extension K/F , any finite set
of automorphisms σ1 , . . . , σn ∈ Aut(K/F ) is K-linearly independent. If K/F is a
Galois extension and F is infinite, we have the following significantly stronger
independence result.
Theorem 8.24. Let K/F be a finite degree Galois extension of infinite fields.
Then the elements σ1 , . . . , σn of Aut(K/F ) are algebraically independent – if
0 6= f (t1 , . . . , tn ) ∈ K[t1 , . . . , tn ], there exists α ∈ K such that f (σ1 (α), . . . , σn (α)) 6=
0.
Proof. As a matter of notation, for an n-tuple (x1 , . . . , xn ) ∈ K n , we will
denote by (x1 , . . . , xn )• the corresponding column vector, i.e., element of Mn,1 (K).
If it brings no confusion, we will suppress indices by writing x• for (x1 , . . . , xn )• .
Let α1 , . . . , αn be a basis for K/F . Define A ∈ Mn (K) by Aij = σi αj . By
Lemma 8.23, A is nonsingular. Now let c = (c1 , . . . , cn ) ∈ F n and put
n
X
α= cj α j .
j=1

Then for all 1 ≤ i ≤ n,

n
X
σi (α) = Aij cj ,
j=1
so
σ(α)• = Ac• .
72 8. GALOIS EXTENSIONS

Seeking a contradiction, we suppose that for all σ ∈ K, f (σ1 (α), . . . , σn (α)) = 0.

By the above, this can be reexpressed as
0 = f (σ(α)• ) = f (Ac• )
for all c ∈ F n . Thus the polynomial
g(t) = g(t1 , . . . , tn ) = f (At• ) ∈ K[t1 , . . . , tn ]
vanishes at every c ∈ F n , so by Lemma 8.21, g = 0. So f (t) = g(A−1 t• ) = 0. 
Exercise 8.6. Show that Theorem 8.24 fails for every finite degree extension
of finite fields.
Theorem 8.25. (Normal Basis Theorem) Let K/F be a finite Galois extension
of degree n. Then there exists α ∈ K such that the set {σα}σ∈Gal(K/F ) is a basis
of K as an F -vector space.
Proof. By Proposition 8.21 we may assume that F , and hence also K, is
infinite. Write out the elements of Aut(K/F ) as 1 = σ1 , σ2 , . . . , σn . Let t1 , . . . , tn
be independent indeterminates, and consider the matrix B with Bij = tk , where
σi σj = σk . In this matrix each ti appears exactly once in each row and column,
so the specialization t1 = 1, ti = 0 for all i > 1 gives rise to a permutation matrix
with determinant ±1. It follows that d(t1 , . . . , tn ) = det B is a nonzero element of
the polynomial ring K[t1 , . . . , tn ]. Applying Theorem 8.24, there exists α ∈ K such
that d(σ1 (α), . . . , σn (α)) 6= 0.
For 1 ≤ j ≤ n, put αj = σj (α). Then the matrix A with Aij = σi αj = σi σj α =
σk α nonsingular, so by Lemma 8.23 σ1 α, . . . , σj α is an F -basis of K. 
Exercise 8.7. Explain how the Normal Basis Theorem gives a stronger result
than the Primitive Element Corollary in the case of a Galois extension.

6. Hilbert’s Theorem 90
Let G be a group, and let M be a G-module, i.e., commutative group on which
G acts Z-linearly: that is, we are given a homomorphism G → AutZ (M ). Let
Z 1 (G, M ) be the set of all maps f : G → M which satisfy the cocycle condition:
∀σ, τ ∈ G, f (στ ) = f (σ) + σ(f (τ )).
1
Let B (G, M ) be the set of maps f : G → M such that there is a ∈ M with
f (σ) = σ(a) − a for all σ ∈ G.
Exercise 8.8. a) Show that Z 1 (G, M ) and B 1 (G, M ) are commutative groups
under pointwise addition.
b) Show that B 1 (G, M ) ⊂ Z 1 (G, M ).
We may therefore define
H 1 (G, M ) = Z 1 (G, M )/B 1 (G, M ),
the first cohomology group of G with coefficients in M.
Exercise 8.9. Suppose that G acts trivially on M . Show that H 1 (G, M ) =
Hom(G, M ), the group of all homomorphisms from G to M .
Now observe that if K/F is a field extension and G = Aut(K/F ), then both K (as
an additive group) and K × (as a multiplicative group) are G-modules.
 6. HILBERT’S THEOREM 90 73

Theorem 8.26. Let K/F be a finite Galois extension, with Galois group G =
Aut(K/F ).
a) H 1 (G, K) = 0.
b) H 1 (G, K × ) = 0.
Proof. a) Let f : G → K be a 1-cocycle. Since K/F is finite separable, by
Theorem X.X there is c ∈ K with TrK/F (c) = 1. Put
X
b= f (σ)σ(c),
σ∈G
so
X
τ (b) = τ (f (σ))(τ σ)(c)
σ∈G
X X X
= (f (τ σ) − f (τ )) (τ σ)(c) = f (τ σ)(τ σ)(c) − f (τ )(τ σ)(c)
σ∈G σ∈G σ∈G
!
X
= b − f (τ ) · τ σ(c) = b − f (τ ).
σ∈G

Thus f (τ ) = b − τ (b) for all τ ∈ G, so f ∈ B 1 (G, K).

G → K × be a 1-cocycle. By independence ofP
b) Let f : P characters, there is c ∈ K
such that σ∈G f (σ)σ(c) 6= 0; fix such a c and put b = σ∈G f (σ)σ(c). Then
X
τ (b) = τ (f (σ))(τ σ)(c),
σ∈G
so
X X
f (τ )τ (b) = f (τ )τ (f (σ)) · (τ σ)(c) = f (τ σ) · (τ σ)(c) = b,
σ∈G σ∈G

i.e., f (τ ) = b/τ (b). So f ∈ B 1 (G, K × ). 

The following is a basic result from group cohomology.

Theorem 8.27. Let n ∈ Z+ , and let G = hσ | σ n = 1i be a finite cyclic group.
For any G-module M , we have
H 1 (G, M ) ∼
= {x ∈ M | (1 + σ + . . . + σ n−1 )(x) = 0}/{σx − x | x ∈ M }.
Combining Theorems 8.26 and 8.27 we immediately deduce the following famous
result of D. Hilbert, the 90th theorem in his Zahlbericht. However, because our
focus here is on field-theoretic methods, we will not give a proof of Theorem 8.27
but rather a purely field-theoretic proof of Hilbert’s Satz 90.
Theorem 8.28. (Hilbert’s Satz 90) Let K/F be a finite Galois extension with
cyclic Galois group G = hσ | σ n = 1i.
a) For c ∈ K, the following are equivalent:
(i) TrK/F (c) = 0.
(ii) There is a ∈ K such that c = a − σ(a).
b) For c ∈ K, the following are equivalent:
(i) NK/F (c) = 1.
(ii) There is a ∈ K × such that c = σ(a)
a
.
74 8. GALOIS EXTENSIONS

Proof. Step 1: Because Galois conjugate elements have the same norm and
trace, in both parts a) and b) the implications (ii) =⇒ (i) are immediate.
Step 2: Let c ∈ K be such that TrK/F (c) = 0. Since K/F is separable, by Theorem
6.10 there is b ∈ K with TrK/F (b) = 1.3
Put
a = cb + (c + σ(c))σ(b) + . . . + (c + σ(c) + . . . + σ n−2 (c))σ n−2 (b).
Then
σ(a) = σ(c)σ(b) + (σ(c) + σ 2 (c))σ 2 (b) + . . . + (σ(c) + . . . + σ n−1 (c))σ n−1 (b).
Since TrK/F (c) = c + σ(c) + . . . + σ n−1 (c) = 0, we have
a − σ(a) = cb + cσ(b) + . . . + cσ n−1 b = c TrK/F (b) = c.
Step 3: Let c ∈ K be such that NK/F (c) = 1. By Dedekind’s linear independence
of characters, there is b ∈ K with
a = b + cσ(b) + cσ(c)σ 2 (b) + . . . + cσ(c) · · · σ n−2 (c)σ n−1 (b) 6= 0.
Then
cσ(a) = cσ(b) + cσ(c)σ 2 (b) + . . . + cσ(c) · · · σ n−1 (c)b = a,
so
a
c= .
σ(a)

We will use Theorem 8.28 later on in our study of cyclic extensions.

include application to Pythagorean triples

7. Infinite Algebraic Galois Theory

Theorem 8.29. For an algebraic field extension K/F , the following are equiv-
alent:
(i) K Aut(K/F ) = F . (“K/F is Galois.”)
(ii) K is normal and separable.
(iii) K is the splitting field of a set (possibly infinite) of separable polynomials.
Proof. The equivalence of (ii) and (iii) follows from our characterization of
normal and separable algebraic extensions.
(i) =⇒ (ii): (Morandi, p. 40something) FIXME!!!
(ii) =⇒ (i): Let α ∈ K \ F . Then the minimal polynomial P for α over K
splits in K and has at least one other distinct root β. There is a unique F -algebra
embedding σ : F [α] → K that sends α to β; as usual, we can extend σ to an
automorphism of F and then the restriction of σ to K is an automorphism of K
(since K is normal) for which σ(α) 6= α. Therefore K Aut(K/F ) = F . 
Let us now revisit the abstract setting of section XX in the somewhat less triv-
ial present framework: X = K = F sep , G = Aut(K/F ). Then the maps L 7→
Gal(K/L) and H 7→ K H give a bijective correspondence between closed subexten-
sions L of K/F and closed subgroups H of G. The key fact is the following
3Alternately, since K/F is Galois, we have Tr n−1 (x). It follows
K/F (x) = x + σ(x) + . . . + σ
from Dedekind’s linear independence of characters that TrK/F is not identically zero, and since
it is an F -linear functional it must then be surjective.
 8. A CHARACTERIZATION OF NORMAL EXTENSIONS 75

Lemma 8.30. Every subextension L of K/F is closed, i.e. K Gal(K/L) = L.

Proof. FIX ME!!! 

8. A Characterization of Normal Extensions

Lemma 8.31. a) Let K be a field with algebraic closure K. Let L/K be a purely
inseparable extension, and let σ : L ,→ K be a K-algebra embedding. Then σ is an
L-algebra embedding, i.e., for all x ∈ L, σ(x) = x.
b) Let K/F be an algebraic field extension and Fi the purely inseparable closure of
F in K. Then Aut(K/Fi ) = Aut(K/F ).
Proof. a) Any element y ∈ L satisfies a purely inseparable polynomial P (t) =
n
tp − x for some x ∈ K. The map σ must send y to some root of P (t), of which
there is only one.
b) Choose an algebraic closure K of K. Let σ ∈ Aut(K/F ); by X.X σ extends to
an automorphism of K, which we continue to denote by σ. Applying part a) to σ
with Ki = L, we get that σ fixes Ki pointwise, qed. 
Theorem 8.32. For an algebraic extension K/F , the following are equivalent:
(i) The extension K Aut(K/F ) /F is purely inseparable.
(ii) K/F is normal.
Proof. (i) =⇒ (ii): Put L = K Aut(K/F ) . Let F be an algebraic closure
of K and let σ : K → F be an F -algebra embedding, which we may extend to
an automorphism of F . Since L/F is purely inseparable, by Lemma 8.31b) we
have σ ∈ Aut(K/L). In other words, σ fixes L pointwise. But K/L is Galois,
hence normal, so for any embedding σ : K ,→ F which fixes L pointwise we have
σ(K) = K.
(ii) =⇒ (i): Let Fi be the purely inseparable closure of F in K. Since K/F
is normal, so is K/Fi . Moreover, by Theorem X.X and Corollary X.X, K/Fi is
separable. Thus K/Fi is Galois, so (applying Lemma) 8.31) we get
K Aut(K/F ) = K Aut(K/Fi ) = Fi .

 CHAPTER 9

Solvable Extensions

1. Cyclotomic Extensions
1.0.1. Basics. Let K be a field. An element x ∈ K × is a root of unity if there
is n ∈ Z+ such that xn = 1; equivalently, x lies in the torsion subgroup of K × . We
put
µn (K) = {x ∈ K | xn = 1},
S
the nth roots of unity in K. We put µ(K) = n≥1 µn (K). Thus µn (K) and µ(K)
are subgroups of K × and µ(K) = K × [tors].
Lemma 9.1. For any field K and n ∈ Z+ , we have #µn (K) ≤ n.
Proof. The elements of µn (K) are the roots of the polynomial tn − 1 over K,
and a nonzero polynomial over a field cannot have more roots than its degree. 

Lemma 9.2. For any field K and n ∈ Z+ , the group µn (K) is finite cyclic.
Proof. By Lemma 9.2, µn (K) is finite. We use the Cyclicity Criterion
[CL-NT, Thm. B.9]: a finite group G is cyclic iff for all d ∈ Z+ there are at most
d element of order n in G. This holds in µn (K) since the polynomial td − 1 can
have no more than d roots. 
2πki
Example 9.3. Fix n ∈ Z. For 0 ≤ k < n, the elements e n are distinct nth
roots of unity in C. So #µn (C) = n.
Exercise 9.1. Let K be an ordered field. Show that µ(K) = {±1}.
An element of K × of exact order n is called a primitive nth root of unity.
Proposition 9.4. Let K be an algebraically closed field. For n ∈ Z+ , the
following are equivlaent:
(i) char K - n.
(ii) #µn (K) = n.
(iii) K admits a primitive nth root of unity.
(iv) K admits precisely ϕ(n) primitive nth roots of unity.
Proof. (i) ⇐⇒ (ii): Let f (t) = tn − 1. Then f 0 (t) = ntn−1 . Thus char K -
n ⇐⇒ gcd(f, f 0 ) = 1 ⇐⇒ tn − 1 has n distinct roots ⇐⇒ #µn (K) = n.
(ii) ⇐⇒ (iii): By Lemma 9.2, µn (K) is a finite, cyclic n-torsion abelian group.
Thus it has order n iff it has an element of order n.
(ii) =⇒ (iv): (ii) holds ⇐⇒ µn (K) is cyclic of order n, in which case it has
precisely ϕ(n) generators.
(iv) =⇒ (iii): Since for all n ∈ Z+ , ϕ(n) ≥ 1, this is clear. 
77
78 9. SOLVABLE EXTENSIONS

Exercise 9.2. a) Let K be an algebraically closed field of characteristic zero.

Show that µ(K) ∼= lim + Z/nZ.
−→ n∈Z
b) Let K be an algebraically closed field of characteristic p ≥ 0. Show that µ(K) ∼
=
lim Z/nZ.
−→ n∈Z+ , p-n

Exercise 9.3. Show that for any field K, µ(K sep ) = µ(K).
Henceforth we only consider µn for char K - n.

For a field K, we denote by K cyc the field obtained by adjoining to K all roots
of unity in a fixed algebraic closure K. Then K cyc is the splitting field of the set
{tn − 1}char K-n of separable polynomials, so is an algebraic Galois extension, the
maximal cyclotomic extension of K. For n ∈ Z+ with char K - n, let K(µn )
be the splitting field of the separable polynomial tn − 1, the nth cyclotomic ex-
tension. Thus K cyc = limK(µn ).
−→

For a field K, it is traditional to denote by ζn a primitive nth root of unity in

2πi
K sep . When K = C, the standard choice is ζn = e n . There is an advantage to
this choice: for all m | n, we have
n
(8) ζnm = ζm .
Exercise 9.4. Let K be any algebraically closed field.
a) Show that one may choose, for all n ∈ Z+ with char K - n, a primitive nth root
of unity ζn such that the compatibility relation (8) holds.
b) In how many ways is it possible to do this?
(Suggestion: express your answer as an inverse limit of finite sets.)
Proposition 9.5. Let K be a field and n ∈ Z+ with char K - n.
a) We have K(µn ) = K(ζn ).
b) There is a canonical injection an : Aut(K(ζn )/K) ,→ (Z/nZ)× .
Proof. a) In other words, the assertion is that by adjoining any one primitive
root of unity, we get the splitting field of the polynomial tn − 1. Since every nth
root of unity is a power of ζn , this is clear.
b) For σ ∈ Aut(K(ζn )/K), σ(ζn ) is a primitive nth root of unity: any automorphism
of a field preserves the order of elements of the multiplicative group of that field.
a (σ)
Thus σ(ζn ) = ζnn for a unique an (σ) ∈ (Z/nZ)× . It is immediate that σ 7→ an (σ)
is a group homomorphism. Finally, if an (σ) = 1, then σ(ζn ) = ζn , so σ fixes K(ζn )
and is thus trivial. 
Exercise 9.5. a) In order to define an we chose a primitive nth root of unity
ζn ∈ K sep . Show that the homomorphism an is in fact independent of this choice.
b) Suppose that m | n. Show that we have a commutative diagram
a
n
Aut(K(ζn )/K) → (Z/nZ)×
a
m
Aut(K(ζm )/K) → (Z/mZ)× ,
where the map (Z/nZ)× → (Z/mZ)× is the induced map on units of the quotient
map Z/nZ → Z/mZ.
c) Deduce that there is an injection
a : Aut(K cyc /K) ,→ lim (Z/nZ)× .
←− n∈Z+ ,char K-n
 1. CYCLOTOMIC EXTENSIONS 79

d) In particular, for any prime ` 6= char K, there is an injection

χ` : Aut(limK(µ`n )/K) → Z×
` ,
−→
called the `-adic cyclotomic character. When char K = 0, there is an injection
χ : Aut(K cyc /K) ,→ Ẑ× ,
the adelic cyclotomic character.
1.0.2. Cyclotomic Polynomials. For n ∈ Z+ , let Φn (t) be the unique monic
polynomial with roots the primitive nth roots of unity in C.1
Proposition 9.6. a) For all n ∈ Z+ , we have
Y
(9) Φd (t) = tn − 1.
d|n
+
b) For all n ∈ Z , Φn (t) ∈ Z[t].
c) For all n ∈ Z+ , we have
n
Y
(10) Φn (t) = (td − 1)µ( d ) .
d|n

Proof. a) Both sides of (9) are monic polynomials with C coefficients whose
roots are precisely the nth roots of unity in C. So they are equal.
b) By strong induction on n. The base case is clear: Φ1 (t) = t − 1. Now sup-
Q n > 1 and that Φd (t) ∈ Z[t] for all proper divisors d ofn n. Then Q(t) =
pose
d|n, d6=n Φd (t) ∈ Z[t] is a monic polynomial and Q(t)Φn (t) = t − 1. Now imagine
actually performing polynomial long division of tn − 1 by Q(t) to get Φn (t): since
tn − 1, Φn (t) ∈ Z[t] are monic, the quotient Φn (t) has Z-coefficients.
c) This follows from part a) by the Möbius Inversion Formula applied in the com-
mutative group Q(t)× .2 
Theorem 9.7. Let n ∈ Z+ and let K be a field of characteristic p. Regard
Φn (t) ∈ Fp [t] ⊂ K[t]. Then Φn (t) is a separable polynomial whose roots in K[t] are
precisely the primitive nth roots of unity.
Proof. By the Derivative Criterion tn − 1 ∈ K[t] is separable; by (8) so is
Φn (t). It is clear that the ϕ(n) roots of Φn (t) in K are nth roots of unity; that they
are the ϕ(n) primitive nth roots of unity follows by an easy induction argument. 
Exercise 9.6. Let p be a prime number and a ∈ Z+ .
a) Show that Φp (t) = 1 + t + . . . + tp−1 .
b) Show that Φ2p (t) = 1 − t + . . . + (−t)p−1 .
a−1
c) Show that Φpa (t) = Φp (tp ).
Exercise 9.7. For n ∈ Z+ , let r(n) = p|n p. Show:
Q
n
Φn (t) = Φr(n) (t r(n) ).
Exercise 9.8. Let n ∈ Z+ .
a) Show: for all n ≥ 2, the constant coefficient of Φn (t) is 1.
b) Show: for all n 6= 2, the product of the primitive nth roots of unity in C is 1.
1The use of C here is somewhere between tradition and psychology: any algebraically closed
field of characteristic zero – e.g. Q – would serve as well.
2In fact we don’t need this in what follows – it is just a pretty formula.
80 9. SOLVABLE EXTENSIONS

Theorem 9.8. (Gauss-Kronecker) For all n ∈ Z+ , Φn (t) ∈ Q[t] is irreducible.

Proof. Since Φn (t) ∈ Z[t] is monic and Z is a UFD, by Gauss’s Lemma it is
equivalent to show that Φn is irreducible in Z[t]. We may write Φn (t) = f (t)g(t)
with f, g ∈ Z[t] monic and f irreducible, and the goal is to show g = 1.
Step 1: Let α be a root of f (t) ∈ Q (hence a primitive nth root of unity) and let p
be a prime number not dividing n. We claim that αp is also a root of f (t).
proof of claim: Suppose not; then, since p - n, αp is a primitive nth root of unity,
so αp is a root of g. Thus α is a root of h(tp ). Since f is monic irreducible and
f (α) = 0, f is the minimal polynomial for α, so there is h ∈ Z[t] with f (t)h(t) =
g(tp ). Now apply the homomorphism Z[t] → Z/pZ[t], f 7→ f : we get
Φn = f h.
For any polynomial a(t) ∈ F/pZ[t] we have a(tp ) = a(t)p , and thus
g p = f h.
Let q be an irreducible factor of f . Then q | f | g p , so q | g. It follows that
g 2 | f g = Φn . But since p - n, the Derivative Criterion still holds to show that
Φn ∈ Z/pZ[t] is separable: contradiction.
Step 2: Let β be any root of Φn (t) in Q. Then β and α are both primitive nth
roots of unity, so that there is a sequence of (not necessarily distinct) prime numbers
p1 , . . . , pr with gcd(p1 , . . . , pr , n) = 1 and αp1 ···pr = β. Applying Step 1 successively
to α, αp1 , . . . , αp1 ···pn−1 we find that β is also a root of f (t). Thus f has as its roots
all primitive nth roots of unity, i.e., f = Φn , and Φn is irreducible. 
1.0.3. Some Applications.
Corollary 9.9. For any n ∈ Z+ , the extension Q(µn )/Q) is Galois, with
Aut(Q(µn )/Q) canonically isomorphic to (Z/nZ)× .
Exercise 9.9. Prove it.
Exercise 9.10. Let m, n ∈ Z+ with m | n.
a) Show: Q(µm ) ⊆ Q(µn ).
b) Show: Q(µm ) = Q(µn ) iff m = n or (m is odd and n = 2m). c) Show:
(11) Q(µm , µn ) = Q(µlcm(m,n) ).

(12) Q(µm ) ∩ Q(µn ) = Q(µgcd(m,n) ).

Theorem 9.10. Let n ∈ Z+ . There are infinitely many primes p with p ≡ 1
(mod n).
Proof. We may assume n ≥ Q 2. Let S be a finite set (possibly empty) of
primes p ≡ 1 (mod n), and let q = p∈S p. For sufficiently large k ∈ Z, we have
N = Φn (knq) > 1.
Since the constant term of Φn is 1, for any prime p | knq, N ≡ 1 (mod p). Since
N > 1, there is a prime p with Φn (knq) = N ≡ 0 (mod p), so p - knq: in particular
p∈/ S. By Theorem 9.7, knq ∈ Fp is a primitive nth root of unity. By Lagrange’s
Theorem, n | p − 1. We’ve produced a prime p ∈ / S with p ≡ 1 (mod n). 
Lemma 9.11. Let G be a finite abelian group. Then there are k, n ∈ Z+ and a
surjective homomorphism of groups (Z/nZ)k → G.
 2. CYCLIC EXTENSIONS I: KUMMER THEORY 81

Exercise: Prove it.

Corollary 9.12. For any finite abelian group G, there is a Galois extension
L/Q with Aut(L/Q) ∼
= G.
Proof. Step 1: By Lemma 9.11, G is a quotient of (Z/nZ)k for some k, n ∈ Z+ .
Since any group which is a quotient of a finite Galois group over a field K is also a
finite Galois group over that field, it suffices to treat the case G = (Z/nZ)k .
Step 2: By Theorem 9.10, there are prime numbers p1 , . . . , pk such that n | (pi − 1)
for 1 ≤ i ≤ k. The group (Z/pi Z)× is cyclic of order ϕ(pi ) = pi − 1, so there is a
surjection qi : (Z/pi Z)× → Z/nZ. Let
k
Y
q = (q1 , . . . , qk ) : (Z/pi Z)× → (Z/nZ)k ,
i=1

a surjective group homomorphism. Put N = p1 · · · pk . Since the pi ’s are distinct,

by the Chinese Remainder Theorem there is an isomorphism
k
∼
Y
Z/N Z → Z/pi Z
i=1

and thus, passing to unit groups, an isomorphism

k
∼
Y
Φ : (Z/N Z)× → (Z/pi Z)× .
i=1

Thus we get a surjective map

k
∼ Φ
Y q
Aut(Q(µN )/Q) → (Z/N Z)× → (Z/pi Z)× → (Z/nZ)k .
i=1

By Galois Theory, there is a subextension L of Q(µN )/Q with Aut(L/Q) ∼

= (Z/nZ)k .

Exercise 9.11. Show: for any number field K and any finite abelian group G,
there is a Galois extension L/K with Aut(L/K) ∼
= G.
Exercise 9.12. Let n ∈ Z+ .
a) (Parker: [Pa74]) Show: there is a number field K ⊂ R such that K/Q is Galois
and Aut(K/Q) ∼ = Z/nZ.
b) Prove or disprove: for every finite abelian group G, there is a number field K ⊂ R
such that K/Q is Galois and Aut(K/Q) ∼ = G.

2. Cyclic Extensions I: Kummer Theory

A field extension K/F is cyclic if it is of finite degree and Aut(K/F ) is a cyclic
group of order [K : F ]. In particular a cyclic extension is necessarily Galois. By
a generator of a cyclic extension L/K, we mean an element σ which generates
Aut(L/K). (Of course σ is not unique if n > 2.)
Example 9.13. A quadratic extension K/F is cyclic iff it is separable. Thus if
F does not have characteristic 2 then every quadratic extension K/Fp is cyclic, and
moreover – as the quadratic formula holds here – is of the form F ( a) for some
a ∈ F \ F 2.
82 9. SOLVABLE EXTENSIONS

Let F be a field of characteristic 0. Since adjunction of square roots of elements

of F yields cyclic extensions, it is natural to try to construct cyclic extensions of
degree n by adjunction of nth roots. This is a good idea, but it works only under
certain restrictions.
Example 9.14. We revisit Example 4.18. For n ≥ 3, let pn (t) = tn − 2, and
let Fn = Q[t]/(p
√ n (t)). We may embed Q ,→ R ,→ C, and then, since pn has a
unique root n 2 in R, and in such a way √ we view Fn ,→ √R. If ζn is a primitive nth
i n
root of unity, then the conjugates of n 2 over
√ Q are ζ n 2 for 0 ≤ i < n. The only
conjugate that lies in R, let alone Fn , is n 2, so Fn /Q is not normal (so certainly
not cyclic). The splitting field of Fn /Q is
√n
Kn := Q(ζn , 2).
Because the subgroup Aut(Kn /Fn ) of Aut(Kn /Q) is not normal, the group Aut(Kn /Q)
is not commutative, hence certainly not cyclic.
Now let n = 3. Then the polynomial p3 (t) remains irreducible over Q(ζ3 ): in-
deed, every irreducible cubic polynomial remains irreducible over a quadratic field
extension, so K3 /Q(ζ3 ) is Galois of degree 3, hence √
cyclic. A √ generator for its
automorphism group is the automorphism that sends 3 2 to ζ3 3 2. We also com-
pute in this way that the automorphism group Aut(K3 /Q) √ is noncommutative
√ √ of
order 6 and thus, as a permutation group on the conjugates 3 2, ζ3 3 2, ζ32 3 2, is the
full symmetric group S3 . Indeed, it has order [K3 : Q(ζ3 )][Q(ζ3 ) : Q] = 6 and is
noncommutative since the order 2 subgroup Aut(K3 /F3 ) is not normal.
Proposition 9.15. Let K be a field of characteristic p ≥ 0, let n ∈ Z+ , and
let a ∈ K be such that√ the polynomial f (t) = tn − a is irreducible in K[t]. Let
L := K[t]/(f (t)) = K( n a). The following are equivalent:
(i) The extension L/K is cyclic.
(ii) The field K contains a primitive nth root of unity. (In particular, p - n).
Proof. We have f 0 (t) = ntn−1 , so by the Derivative Criterion, L/K is sep-
arable iff p - n. It follows that if p | n then neither (i) nor (ii) holds, so we may
assume henceforth√ that p - n. In this case the roots of f (t) in a splitting field are
of the form ζni n a, where ζn is √a primitive nth root of unity.
n a
(i) =⇒ (ii): In particular, ζn√ n a = ζn lies in any splitting field for f , so if L/K is
normal then ζn lies in L.
(ii) =⇒ (i): The above discussion shows that if K contains a primitive nth root
of unity – say ζn – then L/K is normal and separable, thus Galois.
It remains to show that the group Aut(L/K) √ is cyclic.
√ For this, observe that
there is a unique σ ∈ Aut(L/K) such that σ( n a) = ζn n a: such an automorphism
exists because the automorphism group of a Galois extension √ K[t]/(f )/K acts tran-
sitively
√ on the √ roots of f , and it is unique because L = K( n
a). For any i ∈ Z+ ,
i i n
σ : a 7→ ζn a, and thus the order of σ is
n

hσi = n = [L : K] = # Aut(L/K). 
There is an important converse to Proposition 9.15. To prove it, we need first the
following result, which despite its innocuous appearance is actually quite famous.
Lemma 9.16. Let K be a field, ζn ∈ K a primitive nth root of unity. Let
L/K be a cyclic extension of degree n, with generator σ. There is α ∈ L such that
ζn = σ(α)
α .
 2. CYCLIC EXTENSIONS I: KUMMER THEORY 83

Proof. Equivalently, we need to show that ζn is an eigenvalue for the K-

linear endomorphism σ : L → L. Since σ has order n, by Dedekind’s Theorem the
transformations 1, σ, . . . , σ n−1 are all K-linearly independent, and therefore the
minimal polynomial of σ is indeed p(t) = tn − 1. Thus ζn is a root of the minimal
polynomial for σ and therefore also a root of its characteristic polynomial. 
Theorem 9.17. Let n ∈ Z+ , and let K be a field containing a primitive nth
root of unity ζn . Let L/K be cyclic of degree n with generator σ.
√ √ √
a) There exists a ∈ K such√that σ( n√ a) = ζn n a and √
L = K( n a).
b) If b ∈ K is such that σ( n b) = ζn n b and L = K( n b), then ab ∈ K n .
Proof. a) By Lemma 9.16, there is α ∈ L such that σ(α) = ζn α. Thus for all
i ∈ Z+ σ(αi ) = ζni α. In particular a = αn ∈ K, and the subgroup of √ Aut(L/K)
fixing K(α) pointwise is the identity.
√
It follows that L = K(α) = K( n
a).
n a
b) We have σ( n ab ) = n ab , so √ = u ∈ K. Take nth powers: ab = un ∈ K n . 
p p
n
b

Proposition 9.18. Let K be a field containing a primitive nth root of unity

ζn , and let L/K be a field extension such that L = K(α) and αn = a ∈ K.
a) L/K is a cyclic extension.
b) The degree m = [L : K] is equal to the order of the image of a in K × /K ×n .
c) There exists b ∈ K such that the minimal polynomial of α over K is tm −b.
Proof. a) Since K contains a primitive nth root of unity, in characteristic
p > 0 we must have p - n, so the polynomial f = tn − a ∈ K[t] is separable.
Therefore the splitting field of f is a finite Galois extension of K. But the roots of
f are {ζni α | 0 ≤ i < n} and since ζn ∈ K, all of these roots lie in L. This shows
that L is the splitting field of the separable polynomial f , so L/K is Galois.
a(σ)
For all σ ∈ Aut(L/K), there is a unique a(σ) ∈ Z/nZ such that σ(α) = ζn α,
so σ 7→ a(σ) defines a map
a : Aut(L/K) → Z/nZ.
We claim a is an injective group homomorphism. For σ, τ ∈ Aut(L/K) we have
(στ )(α) = σ(ζna(τ ) α) = ζna(τ ) σ(α) = ζna(τ ) ζna(σ) α = ζna(σ)+a(τ ) α,
which shows that a(στ ) = a(σ) + a(τ ). Moreover we have a(σ) = 0 iff σ(α) = α
iff σ = 1, since α generates L/K. It follows that Aut(L/K) is a subgroup of Z/nZ
hence is cyclic of order M for some M | n.
b) If a has order m in K × /K ×n , there is b ∈ K × such that am = bn . Then
(αm )n = (αn )m = am = bn ,
so there is an nth root of unity ζ such that αm = ζb. Since K contains a primitive
nth root of unity, this shows that αm ∈ K. It therefore follows that for all σ ∈
Aut(L/K) we have ma(σ) = 0 ∈ Z/nZ and thus M | m. Conversely, the L/K
conjugates of α are {σ(α) | σ ∈ Aut(L/K)}, so the minimal polynomial of α is
Y
(t − ζα) = tM − αM ,
ζ∈µM

so α M
∈ K and thus a M
= αM n ∈ K ×n , so m | M . It follows that
m = M = # Aut(L/K) = [L : K].
c) Indeed, we just saw that we may take b = αm . 
84 9. SOLVABLE EXTENSIONS

Proposition√ 9.19. Let K be a field containing a primitive nth root of unity,

and√let L = K( n a) for a ∈ K. Then any subextension M of L/K is of the form
K( d a) for some divisor d of n.
Proof. Let m tbe the order of a in K × /K ×n . By Proposition 9.18m the
extension√
L/K is cyclic of degree m. Let σ be a generator of Aut(L/K), and put
σ( n a) +
√ √
ζ := n a . Since for all d ∈ Z
√ we have σ d ( n a) = ζ d n a, we have that ζ is a
primitive mth root of unity. Since G := Aut(L/K) is cyclic of order m, for all d | m
there is a unique subgroup Hd := hσ d i of order m d and these are the only subgroups
of G. In particular, if d0 (m) denotes the number of positive divisors of m, then
d0 (m) is the number of subgroups of G and thus, by Galois theory, also the number
of subextensions of L/K. √
For d | m and i ∈ Z+ , we have that σ i fixes K( d a) pointwise iff
√ √ √ √
d
a = ( n a)n/d = σ i (( n a)n/d ) = ζ in/d d a
in
iff m | d . The least i ∈ Z+ that satisfies this condition is gcd( nd , m), so
√
Aut(L/K( d a)) = Hgcd( nd ,m) .
As d ranges over all divisors of n, the values of gcd( nd , m) are precisely all the
√
divisors of m, so the set {K( d a) | d | n} consists of d0 (m) different subextensions
of L/K, which by the above, means they account for all the subextensions. 
If K is a field of characteristic not dividing n but not containing a primitive nth root
of unity, there is in general no simple description of the degree n cyclic extensions
of K. A lot of work has been done on special cases: for instance global class
field theory gives a kind of description of all abelian extensions of a number
field or function field in one variable over a finite field. Cyclic extensions have a
distinguished role to play in this theory (e.g. via the Hasse Norm Theorem),
but restricting class field theory to the cyclic case does not make it easier.

3. The equation tn − a = 0
In this section we analyze the structure of the splitting field of a polynomial tn −a =
0 without assuming that the ground field contains a primitive nth root of unity.
We closely follow [LaFT, §VI.9].
Lemma 9.20. Let F be a field of characteristic p > 0 and a ∈ F × \ F ×p . Then
n
for all n ≥ 1, the polynomial tp − a is irreducible.
Proof. We shall prove the contrapositive: suppose that for some n ∈ Z+ the
n
polynomial tp − α is reducible; we will show that α is a pth power in F . We may
n
write tp − α = f (t)g(t), where f (t) and g(t) are nonconstant monic polynomials.
n
Let K/F be an extension field containing a root β of tp − α. Then in K[t] we have
n n n n
tp − α = tp − β p = (t − β)p .
Since K[t] is a UFD and f (t) and g(t) are monic, we therefore have f (t) = (t − β)r
for some 0 < r < pn . Write r = pm s with gcd(p, s) = 1. Note that m < n. Then
m m
f (t) = (tp − β p )s ,
m m
so that the coefficient of tp (s−1) is −sβ p . This lies in F and – since s 6= 0 in F
m
– we conclude β p ∈ F . Thus
m n−m n−m
α = (β p )p ∈ Fp ∈ Fp
 3. THE EQUATION tn − a = 0 85

since m < n. 
Theorem 9.21. Let n ≥ 2, let F be a field, and let a ∈ F × . We suppose:
• For all prime numbers p | n, we have a ∈ / F p , and
4
• If 4 | n, then a ∈
/ −4F .
Then f (t) := tn − a is irreducible in F [t].
Proof. We begin by establishing several special cases.
Step 1: Suppose n = pe is a prime power, a ∈ F \ F p and p is the characteristic of
F . This case is covered by Lemma 9.20.
Step 2: Suppose n = pe is a prime power, a ∈ F \ F p and p is not the characteristic
of F . First we claim that tp − a is irreducible. Otherwise, there is some root α ∈ F
of tp − a such that [F (α) : F ] = d < p. Let N denote the norm map from F (α) to
F : since αp = a, we have
N (α)p = N (a) = ad .
Since gcd(d, p) = 1, there are x, y ∈ Z such that xd + yp = 1, and thus
a = axd aup = (N (α)x au )p ∈ F,
contradiction. Now write
p
Y
tp − a = (t − αi ),
i=1
with α1 , . . . , αp ∈ F and α1 = α. We may thus also write
p
e Y e−1
tp − a = (tp − αi ).
i=1
e−1
Suppose first that α ∈/ F (α)p . Let A be root of tp − α. If p is odd, then by
induction A has degree pe−1 over F (α) and thus degree pe over F and it follows
e
that tp − a is irreducible. If p = 2, suppose α = −4β 4 for some β ∈ F (α). Again
let N be the norm from F (α) to F . Then
−a = N (α) = 16N (β)4 ,
√ √
so −a ∈ F 2 . Since p = 2 it follows that −1 ∈ F (α) but α = ( −12β 2 )2 , a
contradiction. By induction, A has degree pe over F . So we may assume that there
is β ∈ F (α) such that β p = α. . . .

The following is an immediate consequence.
Corollary 9.22. Let p be a prime number, F a field, and a ∈ F \ F p . If p is
either odd or equal to the characteristic of F , then for all n ∈ Z+ the polynomial
n
tp − a is irreducible in F [t].
Let F be a field. Let n be a positive integer that is not divisible by the characteristic
of F , let a ∈ F × , and let K be the splitting field of the separable polynomial p(t) =
tn −a. We address the following question: what is the Galois group G := Aut(K/F )?
Let α be a root of p(t) in K, so K = (α, ζn ). Then an element σ ∈ G is determined
by its action on α and ζn , and we have
σ(α) = ζ b(σ) α, b(σ) ∈ Z/nZ,
σ(ζn ) = ζnd(σ) , d(σ) ∈ (Z/nZ)× .
86 9. SOLVABLE EXTENSIONS

Consider the group

  
1 0
G(n) := ∈ GL2 (Z/nZ) .
b d
The identity      
1 0 1 0 1 0 1 0
= .
0 d b 1 0 d−1 bd 1
shows that the subgroup
  
1 0
N= ∈ GL2 (Z/nZ)
b 1
is normal. It also cyclic of order n, and it follows easily that
G(n) ∼= Z/nZ o (Z/nZ)× ,
with the homomorphism given by the canonical isomorphism
ϕ : (Z/nZ)× → Aut Z/nZ.
A straightforward computation shows that the commutator subgroup of G(n) is
contained in N ; since G(n)/N ∼= (Z/nZ)× is commutative, N must be the com-
mutator subgroup of G(n). The map σ 7→ d(σ) is precisely the mod n cyclotomic
character, so ζn ∈ F ⇐⇒ G ⊂ N . In general, let Cn ⊂ (Z/nZ)× be the  image
 of
1 0
the cyclotomic character, viewed as a subgroup of diagonal matrices as
0 d
above. Then
G ⊂ Z/nZ o Cn .
On the other hand, if p(t) = tn − a is irreducible then K contains F [t]/(p(t)) hence
n | #G. So this gives us the answer in some cases.
Proposition 9.23. Suppose tn − a is irreducible and gcd(n, ϕ(n)) = 1. Then
G∼ = Z/nZ o Cn .
Proof. We know that G is a subgroup of Z/nZ o Cn , of corder n#Cn . As
above, irreducibility implies n | #G. We also have Cn ⊂ G, so #Cn | #G. Since
gcd(n, ϕ(n)) = 1 and #Cn | ϕ(n) = 1, also gcd(n, #Cn ) = 1 and thus n#Cn | #G.
It follows that G = Z/nZ o Cn . 
Theorem 9.24. Let n be an odd positive integer prime to the characteristic
of F , and suppose that [F (ζn ) : F ] = ϕ(n): equivalently, the mod n cyclotomic
character is surjective. Let a ∈ F be such that a ∈ F \ F p for all primes p - n.
Let K be the splitting field of tn − a over F , and let G := Aut(K/F ) be its Galois
group. Then G = G(n), and the commutator subgroup of G is Aut(K/F (ζn )).
Proof. Noe first that since n is odd, by Theorem 9.21 the polynomial tn − a
is irreducible in F . Let α ∈ K be a root, so [F (α) : F ] = n.
Step 1: Suppose n = p is prime. Since gcd(p, ϕ(p)) = gcd(p, p − 1) = 1, Proposition
9.23 applies to give G = G(n). The commutator subgroup is N , which is precisely
the set of automorphisms that pointwise fix ζn , so the commutator subgroup is
Aut(K/F (ζn )). (This latter argument holds in the general case.)
Step 2: Now suppose that n is composite; we may write n = pm with p prime.
Since the mod n cyclotomic character is surjective and m | n, also the mod m
 3. THE EQUATION tn − a = 0 87

cyclotomic character is surjective. Put β := αp , so of course β is a root of tm − a,

and by induction the result applies to tm − a. In particular we have
n = pm = [F (α) : F ] = [F (α) : F (β)][F (β) : F ],
so [F (α) : F (β)] = p. This implies that tp − β is irreducible over F (β): otherwise,
the minimal polynomial of α over F (β) would have degree less than p, contradiction.
Consider the subfield
L := F (α) ∩ F (β, ζn ) ⊂ K.
Certainly F (β) ⊂ L. On the other hand, L/F (β) is an abelian extension. On
the other hand, L is also the splitting field of tp − β over F (β), so by Step 1, the
maximal abelian subextension of K/F (β) is F (β, ζp ), and thus
L ⊂ F (α) ∩ F (β, ζp ) = F (β) :
if it were any larger, then F (α) would contain a nontrivial subextension of F (ζp )/F ,
contradicting [F (ζn ) : F ] = ϕ(n). Thus
[F (α, ζn ) : F (β, ζn )] = p :
if not, then these fields would be equal and thus
F (β) ⊂ F (α) ⊂ F (β, ζn ),
so F (α)/F (β) would be abelian, again contradicting Step 1. An argument identical
to the above but using induction instead of Step 1 shows that
F (ζn ) ∩ F (β) = F
and then using Natural Irrationalities we get
[F (β, ζn ) : F (β)] = [F (ζn ) : F ] = ϕn .
It follows that
[K : F ] = [K : F (β, ζn )][F (β, ζn ) : F (ζn )][F (ζn ) : F ] = nϕ(n) = #G(n),
so Aut(K/F ) = Gn . The conclusion on commutator subgroups follows. 

Exercise 9.13.
√ a) Let f (t) = x8 − 2 ∈ Q[t]. Show: the splitting field of
8
f is Q( 2, ζ4 ).
b) Observe that f satisfies all of the hypotheses of Theorem 9.24 except that
8 is not odd, and that the conclusion does not hold: [K : F ] = nϕ(n)
2 , not
nϕ(n).
We remark that the essential content content of Theorem 9.24 lies in the assertion
that (under the hypotheses and notation used therein) F (ζn ) ∩ F (α) = F , since
by Natural Irrationalities this implies that [F (ζn , α) : F (ζ)] = n. It is also natural
to think in terms of linear disjointness (cf. §12): because F (ζn )/F is Galois, the
identity F (ζn ) ∩ F (α) = F holds iff F (ζn ) and F (α) are linearly disjoint over F .
Since this holds iff
F (ζn ) ⊗F F (α) = F (ζn )[t]/(tn − a)
is a field, another equivalent condition is that the polynomial tn − a remains irre-
ducible over F (ζn ). In the situation of the above exercise we have
√
8
√
Q(ζ8 ) ∩ Q( 2) = Q( 2)
88 9. SOLVABLE EXTENSIONS

and thus the polynomial t8 − 2, which is irreducible over Q, becomes reducible over
Q(ζ8 ): indeed we have √ √
t8 − 2 = (t4 − 2)(t4 + 2).

4. Cyclic Extensions II: Artin-Schreier Theory

Let p be a prime number, and let F be a field of characteristic p. Our goal is
to study extensions K/F that are cyclic of degree p. In this regard the Kummer
Theory of §9.2 breaks down completely: if ζ ∈ F is such that ζ p = 1, then since
are in characterstic p we have
0 = ζ p − 1 = (ζ − 1)p ,
so ζ = 1. It follows that the only p-power root of unity in characteristic p is 1. This
definitively kills the Kummer-theoretic approach to cyclic extensions of §9.2.

Happily, there is something to take its place. We consider the map

℘ : F → F, x 7→ xp − x.
This map is not a field homomorphism. However, for all x, y ∈ F we have
℘(x + y) = (x + y)p − (x + y) = xp + y p − x − y = ℘(x) + ℘(y),
so it is an endomorphism of the additive group (F, +) of F . Its kernel is {x ∈
F | xp − x = 0}, which is the prime subfield Fp . (The existence of a nontrivial
kernel shows that ℘ is not a field homomorphism.) Borrowing somewhat geometric
language, we call the ℘ the Artin-Schreier isogeny.
Theorem 9.25. Let F be a field of characteristic p > 0, and let a ∈ F .
a) The following are equivalent:
(i) The polynomial fa := tp − t − a ∈ F [t] is irreducible.
(ii) We have a ∈/ ℘(F ): that is, there is no α ∈ F such that αp − α = a.
b) When the equivalent conditions hold, the splitting field K/F of fa is a
cyclic Galois extension of degree p.
Proof. a) (i) =⇒ (ii): This says that an irreducible polynomial of degree
p > 1 has no rational root, which is clear.
(ii) =⇒ (i): Suppose that fa has no root in F and let α ∈ F be a root of f . Since
℘ is an additive group homomorphism with kernel Fp , if α ∈ K is a root of fa , then
for all i ∈ Fp we have
fa (α + i) = ℘(α + i) − a = ℘(α) + ℘(i) − a = ℘(α) − a = fa (α) = 0.
This shows that the roots of fa in K are the p distinct elements
αi := α + (i − 1), 1 ≤ i ≤ p.
If fa factors into irreducible factors p1 · · · pr then for all 1 ≤ j ≤ r the degree of pj
is equal to [F (β) : F ], where β is any root of pj . But the above calculation shows
that if L is the splitting field of fa , then for all 1 ≤ i ≤ p we have F = F (αi ), hence
for each root β we have [F (β) : F ] =P[K : F ] and thus all the polynomials pi have a
r
common degree d > 1. Since dr = j=1 deg(pj ) = deg(fa ) = p, we conclude that
d = p and r = 1: that is, f is irreducible.
b) Suppose now that the equivalent conditions of part a) hold. It follows from our
analysis that if α is any root of fa in F then we have K = F (α), so K/F is the
degree p splitting field of the separable polynomial fa , so is cyclic. 
 8. SOLVABLE EXTENSIONS II: SOLVABILITY BY RADICALS 89

Exercise 9.14. Let a ∈ F \ ℘(F ), let α ∈ K be a root of fa , and let K = F (α).

Show: there is a unique σ ∈ Aut(K/F ) such that σ(α) = α+1, so Aut(K/F ) = hσi.
Let a ∈ F \ ℘(F ). It follows from Theorem 9.25 that ta = tp − t − a ∈ F [t] is
irreducible and that for any two elements α, β ∈ F such that ℘(α) = ℘(β) = a
we have that F (α) = F (β) is the splitting field for fa . Because of this we write
F (℘−1 (a)) for this common field: although ℘−1 (a) consists of p different elements,
they each generate the same field. Conversely, every cyclic field extension of degree
p in characteristic p is generated by an “Artin-Schreier root”:
Theorem 9.26. Let F be a field of characteristic p, and let K/F be cyclic of
degree p. Then there is a ∈ F \ ℘(F ) such that K = F (℘−1 (a)).
Proof. Let σ be a generator of the cyclic group Aut(K/F ). Since TrK/F (−1) =
−p = 0, by Theorem 8.28 there is α ∈ K such that −1 = α − σ(α), or equivalently
σ(α) = α + 1.
It follows that α ∈
/ F and since K/F has degree p we must have K = F (α). Put
a := ℘(α) = αp − α.
Then
σ(a) = σ(α)p − σ(α) = (α + 1)p − (α + 1) = αp − α = a,
so a ∈ F and K = F (℘−1 (a)). 
Proposition 9.27. Let K be a field of characteristic p, with algebraic closure
K. For a, b ∈ K, show that the following are equivalent:
(i) We have K(℘−1 (a)) = K(℘−1 (b)).
(ii) The elements a and b generate the same cyclic subgroup of K/℘(K).
Proof. (i) =⇒ (ii): . . .
(ii) =⇒ (i): If a and b generate the same cyclic subgroup of K/℘(K), then there is
n ∈ Z+ with gcd(n, p) = 1 and z ∈ K such that na − b = z p − z. Choose α, β ∈ K
such that ℘(α) = a, ℘(β) = b. Then ℘(nα−β) = n℘(α)−℘(β) = na−b = ℘(z p −z),
so there is i ∈ Fp such that
nα − β = z p − z + i ∈ K,
which implies that
K(℘−1 (a)) = K(α) = K(β) = K(℘−1 (b)). 

5. Cyclic Extensions III: Witt’s Theory

6. Abelian Extensions of Exponent n: More Kummer Theory
7. Solvable Extensions I: Simple Solvable Extensions
8. Solvable Extensions II: Solvability by Radicals
 CHAPTER 10

Computing Galois Groups

91
 CHAPTER 11

Structure of Transcendental Extensions

1. Rational Function Fields

1.1. The degree of [k(t) : k(f )].
Theorem 11.1. Let k be a field, and let K = k(t) be the field of rational
functions in one variable over k. Let f ∈ k(t) \ k, which we may write as f = p(t)
q(t)
with gcd(p, q) = 1. Then:
a) We have that f is transcendental over k.
b) We have [K : k(f )] = max(deg(p), deg(q)).
Proof. Put d := max deg p, deg q, and consider the polynomial
P := p(X) − f q(X) ∈ k[f ][X] ⊂ k(f )[X].
Then deg P ≤ d. In fact, since f ∈
/ k, if p(X) and q(x) have the same degree, the
leading terms of p(X) and f q(X) cannot cancel and thus deg P = d. We have
P (t) = p(t) − f q(t) = 0,
so t satisfies a degree d polynomial equation with coefficients in k(f ). This shows
that t is algebraic over k(f ) of degree at most d. It follows that if f were algebraic
over k then t would be algebraic over k, which it certainly is not. This shows part
a) and also that [K : k(f )] ≤ d.
To complete the proof it suffices to show that P ∈ k(f )[X] is irreducible. Since
P ∈ k[f ][X] = k[f, X] = k[X][f ] has degree 1 in f , in the polynomial ring k[f, X]
the only possible factorization could be into c(X)`(X, f ), where c(X) ∈ k[X] and
`(X, f ) has degree 1 in X and thus c(X) | gcd(P (x), Q(x)), hence c(X) ∈ k × and
the factorization is trivial. So P ∈ k[f ][X] is irreducible. Since k[X] is a UFD,
by one version of Gauss’s Lemma [Cl-CA, Cor. 15.25a)] we have P ∈ k(f )[X] is
irreducible, completing the proof. 
Let L/K be a field extension. We say that K is algebraically closed in L if
for all x ∈ L, if x is algebraic over K then x ∈ K. Otherwise put, L/K is not
algebraically closed in L if there is a subextension M of L/K such that M/K is
nontrivial algebraic. In particular, if L ) K and K is algebraically closed in L then
L/K is transcendental.

Many find this terminology confusing at first, because it is so close to the ter-
minology “algebraically closed” and yet means something quite different. There is
however some connection:
Exercise 11.1. For a field K, show that the following are equivalent:
(i) For every field extension L/K we have that K is algebraically closed in L.
93
94 11. STRUCTURE OF TRANSCENDENTAL EXTENSIONS

(ii) The field K is algebraically closed.

As an important example, Theorem 11.1a) is precisely the assertion that for every
field k we have that k is algebraically closed in k(t). One can easily generalize this
as follows:
Proposition 11.2. Let k be a field, and let k[{ti }i∈I ] be the polynomial ring
over k in a (possibly infinite) set of indeterminates. Let K be the fraction field of
k[{ti }i∈I ], a rational function field. Then k is algebraically closed in K.
Proof. Step 1: Suppose that I is finite, in which case we may as well take
I = {1, . . . , n}. We will prove that k is algebraically closed in K = k(t1 , . . . , tn )
by induction on n, the base case n = 1 being Theorem 11.1a). Suppose the result
holds for rational function fields in n − 1 variables, and let x ∈ K be algebraic
over k. Certainly then x is algebraic over k(t1 , . . . , tn−1 ), and since k(t1 , . . . , tn ) =
k(t1 , . . . , tn−1 )(tn ), it follows from the base case that x ∈ k(t1 , . . . , tn−1 , and then
it follows from the induction hypothesis that x ∈ k.
Step S 2: Suppose that I is infinite, and let x ∈ K be algebraic over k. We have
K = S⊂T k({ti | i ∈ S}) as S ranges over all finite subsets of T , so x lies in some
subfield that is a rational function field in finitely many indeterminates, and thus
being algebraic over k we get from Step 1 that x lies in k. 
1.2. Aut(K(t)/K). We now compute the automorphism group Aut(K(t)/K)
of the transcendental extension K(t)/K. We will see in particular that this group
if infinite iff K is infinite, which gives our first example of an infinite automorphism
group of a field.

If L/K is a field extension, S is a set of generators for L/K, M/K is another

field extension and ι1 , ι2 : L ,→ M are K-algebra homomorphisms, then ι1 = ι2 iff
ι1 (x) = ι2 (x) for all x ∈ S. It follows that if σ, τ ∈ Aut(K(t)/K), then we have
σ = τ iff σ(t) = τ (t). So the main question is for which f ∈ K(t) we can extend
t 7→ f to a K-algebra automorphism of K(t). By the universal property of polyno-
mial algebras [Cl-CA, Thm. 5.37], for any f ∈ K(t) there is a unique K-algebra
homomorphism ϕ : K[t] → K(t) such that ϕ(t) = f . If f = a ∈ K, then the
kernel of ϕ is (t − a), so ϕ cannot extend to a homomorphism K(t) → K(t) because
field homomorphisms are injective. If f ∈ K(t) \ K, then since K is algebraically
closed in K(t), the element f is transcendental. It follows that for p ∈ K[t], if
p = an tn + . . . + a1 t + a0 ∈ Ker ϕ then 0 = ϕ(p) = an f n + . . . + a1 f + a0 and thus
p = 0 since f is transcendental. So ϕ is an injective K-algebra homomorphism of
domains and thus uniquely extends to a homomorphism on fraction fields
Φ : K(t) → K(t).
The image of Φ is the subfield k(f ) of K(t). By Theorem 11.1, if we write f ∈ K(t)\
K as f = pq with gcd(p, q) = 1, then we have [K(t) : Φ(K(t))] = [K(t) : K(f )] =
max(deg(p), deg(q)). It follows that the field homomorphism Φ : K(t) → K(t) is a
field automorphism iff max(deg(p), deg(q)) = 1.
It follows that evaluation at t gives a bijection from Aut(K(t)/K) to { at+b ct+d |
at+b
a, b, c, d ∈ K | ad − bc 6= 0} – the latter condition is because ad − bc = 0 iff ct+d ∈ k.
Thus we can represent each σ ∈ Aut(K(t)/K) as a matrix
 
a b
M (σ) := ∈ GL2 (K).
c d
 1. RATIONAL FUNCTION FIELDS 95

Two matrices determine the same automorphism iff they are scalar multiples of
each other, so if we define PGL2 (K) as the quotient
 of GL2(K) modulo the (central,
a 0
hence normal) subgroup K × of scalar matrices { | a ∈ K × } then we get
0 a
a canonical bijection from Aut(K(t)/K) to PGL2 (K). In the following theorem we
summarize the above discussion and add one more piece.
Theorem 11.3. The map Aut(K(t)/K) → PGL2 (K) given by
 
at + b a b
s ∈ Aut(K(t)/K) 7→ s(t) = 7→ (mod K × )
ct + d c d
is an isomorphism of groups.
Exercise 11.2. Complete the proof of Theorem 11.3 by showing that the above
bijection from Aut(K(t)/K) to PGL2 (K) is an isomorphism of groups.
Exercise 11.3. Let K be a field.
a) Suppose that K ∼
= Fq . Show: # Aut(K(t)/K) = (q + 1)(q 2 − q).
b) Suppose that K is infinite. Show: # Aut(K(t)/K) = #K.
1.3. Lüroth’s Theorem.
Theorem 11.4 (Lüroth’s Theorem). Let K be a field, and let L be a field such
that K ( L ⊂ K(t). Then there is f ∈ K(t) such that L = K(f ).
Proof. Let v ∈ L \ K. By Theorem 11.1 we know that K(t)/K(v) has finite
degree, so [K(t) : L] is finite. Let
f (x) := xn + `n−1 xn−1 + . . . + `1 x + `0 ∈ L[x]
be the minimal polynomial of t over L. Since t is transcendental over K, there is
0 ≤ j ≤ n − 1 such that lj lies in L \ K: we fix one such index and call it J. Put
u := `J ;
we will in fact show that L = K(u). Put
m := [K(t) : K(u)].
Since K(u) ⊂ L we have m ≥ n, so to show L = K(u) it suffices to show n ≥ m.
We may scale the coefficients of f so as to get a primitive polynomial in K[t][x]:
choose c0 (t), . . . , cn−1 (t), d(t) ∈ K[t] such that gcd(c0 , . . . , cn−1 , d) = 1 and for all
c
0 ≤ j ≤ n − 1 we have `j = dj . In particular we have
cJ
u = `J = .
d
Put
M := max(deg(cJ ), deg(d)).
By Theorem 11.1 we have
m = [K(t) : K(u)] ≤ M.
The inequality is because of the possibility that cJ and d are not coprime. Put
F (x, t) := d(t)f (x) = d(t)xn + cn−1 (T )xn−1 + . . . + c0 (T ) ∈ K[x, t].
96 11. STRUCTURE OF TRANSCENDENTAL EXTENSIONS

Then F (x, t) is primitive as a polynomial in x, has x-degree degx F = n and has

t-degree degt F ≥ M ≥ m. Now t is a root of the polynomial cJ (x) − ud(x) ∈ L[x],
so there is q ∈ L[x] such that
cJ (x) − ud(x) = q(x)f (x) ∈ L[x].
cJ
Substituting u = d and clearing denominators, we get
cJ (x)d(t) − cJ (t)d(x) = d(t)q(x)f (x) = q(x)F (x, t).
Since the left hand side lies in K[t, x] and F (x, t) is primitive in x, Gauss’s Lemma
implies that
q(x) = r(x, t) ∈ K[x, t]
and thus
(13) cJ (x)d(t) − cJ (t)d(x) = r(x, t)F (x, t).
In (13), the t-degree of the left hand side is at most max(deg(d(t), deg(cJ (t)) = m,
whereas for the right hand side we have
degt (r(x, t)F (x, t)) degt (r(x, t)) + degt (F (x, t)) ≥ degt (r(x, t)) + M.
It follows that degt (r(x, t)) = 0, i.e., r(x, t) = r(x) ∈ K[x]. It follows that
rF = cJ (x)d(t) − cJ (t)d(x) is primitive in K[t][x], and then the symmetric form
of cJ (x)d(t) − cJ (t)d(x) implies that rF is also primitive in K[x][t], which implies
that r ∈ K is constant. We conclude
n = degx (F ) = degx (rF ) = degx (cJ (x)d(t) − cJ (t)d(x)) ≥ M ≥ m.
It follows that n = M and L = K(u). 
Whereas I find that Corollary 7.5 is the most surprising result of these notes, per-
haps Theorem 11.4 is the most misleading result. More generally, the Lüroth
Problem asks whether if K is a field and K(t1 , . . . , tn ) is a rational function field
over K in n indeterminates and L is a field with K ⊂ L ⊂ K(t1 , . . . , tn ) such that
K(t1 , . . . , tn )/L has finite degree, then we must have L ∼
= K(t1 , . . . , tn ). Theorem
11.4 yields an affirmative answer to the Lüroth Problem for n = 1 and all fields K.1
For n ≥ 2 the Lüroth Problem is really one of the most basic (and deep)
questions of higher-dimensional birational algebraic geometry. There is only one
more case of an affirmative answer: when K is algebraically closed of character-
istic 0 (e.g. K = C, the most important field for classical algebraic geometry)
Castelnuovo showed that when n = 2 the Lüroth Problem again has an affirmative
answer: that is, any finite index K-subalgebra of K(t1 , t2 ) is again K-isomorphic to
K(t1 , t2 ). On the other hand, for any prime number p, and any algebraically closed
field K of charcteristic p, Zariski gave counterexamples to the Lüroth Problem over
K with n = 2 [Za58]. Moreover there are counterexamples to the Lüroth Problem
for n = 2 over many non-algebraically closed fields of characteristic 0, including
K = Q. For each n ≥ 3 there are counterexamples to the Lüroth Problem even
over K = C, the first such examples being given by Clemens and Griffiths [CG72].

These higher dimensional counterexamples to the Lüroth Problem are far beyond
our means in these notes. Indeed the methods of pure field theory are not even
1When n = 1 it follows from Theorem 11.1 that for any L with K ( L ⊂ K(t), then K(t)/L
must have finite degree. This is clearly false when n ≥ 2, hence we impose that hypothesis
explicitly.
 2. TRANSCENDENCE BASES AND TRANSCENDENCE DEGREE 97

well equipped to give an example of a finite degree field extension K/C(t) that is
not isomorphic to C(t), although such examples are given by meromorphic function
fields of any compact Riemann surface of genus g ≥ 1. The situation is highly
analogous (and actually more than analogous!) to what one encounters in general
topology, where the methods developed are “too general” to provably exhibit a
compact, connected topological surface that it not homeomorphic to the 2-sphere–
even though the torus presents itself as a very plausible candidate, one needs the
methods of a different subject, algebraic topology, to confirm this.

Theorems 11.1 and 11.4 give a good picture of the subfield lattice of K(t)/K for any
field K. Namely, this lattice is Noetherian – for any subextension K ( L ⊂ K(t)
we have [K(t) : L] is finite, so there are no infinite ascending chains. On the other
hand, the lattice is not Artinian, as there are infinite descending chains, e.g.
n
K(t) ) K(t2 ) ) K(t4 ) ) . . . ) K(t2 ) ) . . . .
This observation can be generalized.
Exercise 11.4. Let L/K be a transcendental field extension. Show: the subfield
lattice of L/K is not Artinian.
It turns out that the subfield lattice of a field extension L/K is Noetherian iff L/K
is finitely generated: this is Exercise 11.11, and at the point at which it is assigned
it will be more clear how to show it.
Exercise 11.5. Show that there is an algebraic extension L/K that is not
finitely generated (equivalently, is of infinite degree) such that the subfield lattice is
Artinian.
Exercise 11.6. Combining Exercise 11.3 with Lüroth’s Theorem (Theorem
11.4) we deduce the following intriguing consequence: for a finite field Fq , there is
a function f ∈ Fq (t) such that
Fq (t)Aut(Fq (t)/Fq ) = Fq (f )
and Fq (t)/Fq (f ) is finite degree Galois with automorphism group PGL2 (Fq ). Can
you explicitly write down this rational function f ?2

2. Transcendence Bases and Transcendence Degree

Let K/F be an extension. A finite set S = {x1 , . . . , xn } ⊂ K is algebraically
independent over F if for the only polynomial P (t1 , . . . , tn ) ∈ F [t1 , . . . , tn ] such
that P (x1 , . . . , xn ) = 0 is P = 0. An arbitrary set S ⊂ K is algebraically inde-
pendent if all of its finite subsets are algebraically independent. (To be precise,
we must impose some ordering on the elements of S in order to substitute them in
as values of an n-variable polynomial, but the definition is obviously independent
of the chosen ordering.) We say that K/F is purely transcendental if it is of the
form F (S) for some algebraically independent subset S of K.
Proposition 11.5. Let K/F be an extension and S = {xi } be an ordered set
of elements of K. The following are equivalent:
(i) The natural map Φ : F [{ti }] → K given by ti 7→ xi is an injection.
2This question is answered in [Ho20]. In fact for any subgroup H ⊂ PGL (F ), Hou explicitly
2 q
computes fH such that Fq (t)H = Fq (fH ).
98 11. STRUCTURE OF TRANSCENDENTAL EXTENSIONS

(ii) The map Φ extends uniquely to an isomorphism F ({ti }) → F (S).

(iii) S is algebraically independent over F .
A subset S of K/F is a transcendence basis if it is algebraically independent
and K/F (S) is algebraic. In other words, a transcendence basis for K/F effects a
decomposition of K/F into a tower K/F (S)/F of a purely transcendental extension
followed by an algebraic extension.
Example 11.6. The empty set is – perhaps by definition – always algebraically
independent. If K/F is algebraic, then the only algebraically independent subset is
the empty set, which is a transcendence basis.
Lemma 11.7. Let K/F be an extension, S ⊂ K be algebraically independent,
and x ∈ K. Then S ∪ {x} is algebraically independent iff x is transcendental over
F (S).
Proof. f S is an algebraically independent subset and x ∈ K is transcenden-
tal over F (S), then suppose for a contradiction that S ∪ {x} were dependent: i.e.,
there exists finite ordered subset Sn = (x1 , . . . , xn ) of S and a nonzero polyno-
mial P ∈ F [t1 , . . . , tn , tn+1 ] such that P (x1 , . . . , xn , x) = 0. But the transcendence
of x over F (S) implies that the polynomial P (x1 , . . . , xn , tn+1 ) is identically zero,
so that the polynomial Q(t1 , . . . , tn ) := P (t1 , . . . , tn , 0) is not identically zero and
Q(x1 , . . . , xn ) = 0, contradicting the independence of (x1 , . . . , xn ). The other di-
rection is even easier. 
Corollary 11.8. a) An algebraically independent subset S of K is a transcen-
dence basis iff it is not properly contained in any other algebraically independent
set.
b) Every algebraically independent subset of K is contained in a transcendence basis.
Proof. Part a) follows immediately from Lemma 11.7: a maximal algebraically
independent set S is precisely one for which K/F (S) is algebraic, i.e., a transcen-
dence basis. Moreover the union of a chain of algebraically independent sets is
algebraically independent, so part b) follows from part a) by Zorn’s Lemma. 
Applying Corollary 11.8 to S = ∅, we deduce that every field extension K/F admits
a transcendence basis.
Exercise 11.7. Let {xi }i∈S be a transcendence basis for the (nonalgebraic)
field extension K/F . Let n• : S → Z+ be any function. Show that {xni i } is also a
transcendence basis.
The transcendence degree of a field extension K/F is the minimum cardinality
of a transcendence basis.

The transcendence degree of an extension is related to #K and #F as follows:

Proposition 11.9. Let K/F be a transcendental field extension, with tran-
scendence degree κ. Then
#K = max(#F, κ, ℵ0 ).
Proof. Since K/F is transcendental, K is infinite. Moreover, κ and #F are
cardinalities of subsets of K, so clearly #K ≥ max(#F, κ, ℵ0 ). Conversely, let
S be a transcendence basis; then F (S) has cardinality max(#, κ) and K/F (S) is
algebraic and F (S) is infinite, so #K = #F (S). 
 3. APPLICATIONS TO ALGEBRAICALLY CLOSED FIELDS 99

3. Applications to Algebraically Closed Fields

Theorem 11.10 (Automorphism Extension Theorem). Let K be an extension
of F , with K algebraically closed. Then every automorphism of F can be extended
to at least one automorphism of K.
Proof. Let {xi }i∈S be a transcendence basis for K/F . There is a unique au-
tomorphism of F (S) which extends ι and maps each xi to itself. Since K is the
algebraic closure of F (S), by Corollary XX we can further extend to an automor-
phism of K. 
For any field K, let F be its prime subfield. An absolute transcendence basis
for K is a transcendence basis for K/F .
Corollary 11.11. a) Two algebraically closed fields K1 and K2 are iso-
morphic iff they have the same characteristic and the same absolute tran-
scendence degree.
b) Suppose K1 , K2 are two algebraically closed fields of the same character-
istic and #K1 = #K2 is uncountable. Then K1 ∼ = K2 .
∼
Proof. Evidently any pair of isomorphic fields K1 = K2 have the same char-
acteristic and absolute transcendence degree. If K1 is algebraically closed with
prime subfield F and transcendence degree κ, then for a set S of indeterminates of
cardinality κ, then K1 is isomorphic to the algebraic closure of F(S), which shows
that the characteristic and the absolute transcendence degree determine the iso-
morphism class of an algebraically closed field. Proposition ?? implies that the
absolute transcendence degree of any uncountable field is equal to its cardinality,
and part b) then follows immediately from part a). 
Remark: The fact that any two algebraically closed fields of given cardinality and,
say, continuum cardinality, are isomorphic has important applications in model the-
ory: via the Tarski-Vaught test, it shows that the first order theory of algebraically
closed fields of a given characteristic is complete.
Theorem 11.12. Let K/F be an extension of fields, of transcendence degree
κ. The following are equivalent:
(i) For any extension field K 0 of F with transcendence degree κ0 ≤ κ, there exists
an F -algebra embedding K 0 ,→ K.
(ii) K is algebraically closed.
Exercise 11.8. Prove Theorem 11.12.
Theorem 11.13. Let K be an algebraically closed field. The group Aut(K) of
all automorphisms of K has cardinality 2#K .
Proof. Step 0: Note that 2#K is also the cardinality of the set of all functions
from K to K, so is the largest conceivable value of # Aut(K).

Step 1: We must check the result for Fp and Q. In the former case we have identified
the automorphism group as Ẑ, which indeed has cardinality c = 2ℵ0 = 2#Fp . In the
latter case we can by no means “identify” Aut(Q), but to see that it has continuum
cardinality it suffices, by the automorphism extension theorem, to exhibit a simpler
Galois extension K/Q which has continuum cardinality. Indeed one can take K to
be quadratic closure of Q, i.e., the compositum of all quadratic field extensions of
100 11. STRUCTURE OF TRANSCENDENTAL EXTENSIONS

Q. The automorphism group here is (Z/2Z)ℵ0 = c.

Step 2: By the automorphism extension theorem, the cardinality of the automor-

phism group of any algebraically closed field is at least that of the continuum, which
by Step 0 gives the answer for all countable fields, i.e., for all fields of countable
absolute transcendence degree.

Step 3: Otherwise K is uncountable so there exists an absolute transcendence

basis S with #S = #K. Now the natural action of Sym(S) on S gives rise to an
injection Sym(S) ,→ Aut(F(S)), i.e., by permutation of indeterminates. By the au-
tomorphism extension theorem, this shows that # Aut(K) ≥ # Sym(S) = 2#S . 
Corollary 11.14. Suppose K/F is an extension with K algebraically closed.
Then K Aut(K/F ) is the purely inseparable closure of F in K. In particular, K Aut(K/F ) =
F iff F is perfect.
Proof. If x lies in the purely inseparable closure of F in K, then for some
e
e ∈ Z+ , xp ∈ F . Since x has no Galois conjugates, we must have σ(x) = x for
every σ ∈ Aut(K/F ). Let F be the algebraic closure of F in K. By the usual Galois
Aut(F /F )
theory we have F is the purely inseparable closure of F in F , and by the
automorphism extension theorem we conclude that K Aut(K/F ) ∩ F is the purely
inseparable closure of F in K. If x ∈ K is transcendental over F , then by Theorem
X.X.X there exists an ordered transcendence basis S = (x, {xα }) containing x. By
Exercise X.X.X, S 0 = (x2 , {xα }) is also a transcendence basis hence there exists
an automorphism F (S) → F (S 0 ) sending x 7→ x2 , which, as usual, extends to an
F -algebra automorphism σ of K with σ(x) = x2 6= x. 
Another fact which is true about automorphism groups of algebraically closed field
extensions K/F is that any bijection ϕ between algebraically independent subsets
I and I 0 of K extends to an F -automorphism of F . For this it is necessary and
sufficient that ϕ extend to a bijection on transcendence bases S ⊃ I, S 0 ⊃ I 0 . A
moment’s thought shows that this holds provided that all transcendence bases of
K/F have the same cardinality and need not hold otherwise. This brings us to the
next section.

4. An Axiomatic Approach to Independence

We wish to prove the following result.
Theorem 11.15. Let K/F be a field extension. Then any two transcendence
bases for K/F have the same cardinality, so that the transcendence degree of K/F
is the cardinality of any transcendence basis.
Of course this is strikingly similar to the situation in ordinary linear algebra. We
could therefore go back to our linear algebra texts, consult the proof of the car-
dinality independence of bases in vector spaces, and attempt to mimic it in the
present context. This approach will succeed. Of course in order to do this we will
have to find some sort of precise analogy between linear independence and alge-
braic independence. In mathematics, once we determine that situations A and B
are analogous (to the extent that certain proofs can be carried over from one con-
text to the other), do we just dutifully copy down the similar proofs and keep the
 4. AN AXIOMATIC APPROACH TO INDEPENDENCE 101

analogy in the back of our mind in case we need it later? Depending on taste, this
is a reasonable approach to take, perhaps more reasonable for the mind which is
able to quickly remember what it once knew. As for myself, I would at the same
time worry that it would take me some time and energy to recreate the analogy if
I hadn’t written it down, and I would also be curious whether A and B might be
common instances of a more general construction that it might be interesting or
useful to know explicitly. So we shall follow the second course here, with apologies
to those with different tastes.

Let us begin by placing alongside the analogies between linear independence of

a subset S of an F -vector space V and algebraic independence of a subset S of an
F -algebra K.

In both contexts we have a set, say X, and a collection of subsets S of X that

we are calling independent, subject to:

(LI1) The empty set is independent.

(LI2) A set is independent iff all its finite subsets are independent.
(LI3) Any subset of an independent set is independent.
S
Notice that it follows from (LI2) and (LI3) that the union S = i Si of any chain
of independent subsets is independent: if not, there would exist a finite dependent
subset S 0 of S, but S 0 would have to be a subset of some Si , contradicting the
independence of Si . Combining this with (LI1) and aplying Zorn’s Lemma, we get

(A) Maximal independent sets exist, and every independent set is contained in
some maximal independent set.

Could it be that (LI1) through (LI3) imply the following desirable property?

(B) All maximal independent sets have the same cardinality.

Unfortunately this is not the case. Suppose we have a set X which is partitioned
into disjoint subsets:
a
X= X.
i

Call a subset S ⊂ X independent iff it is contained in Xi for some i. Then (LI1)

through (LI3) are satisfied and the maximal independent sets are simply the Xi ’s,
which we are evidently not entitled to conclude have the same cardinality.

So we need another axiom. Consider the following:

(LI4) If S1 and S2 are independent

S subsets of X with #S1 < #S2 , then there
exists x ∈ X \ S1 such that S1 {x} is independent.

A set X equipped with a family of subsets {Si } satisfying axioms (LI1) through
(LI4) is called an independence space.
102 11. STRUCTURE OF TRANSCENDENTAL EXTENSIONS

In an independence space, if S1 and S2 are independent sets with #S1 < #S2 ,
then S1 is non-maximal. Therefore a maximal independent set has cardinality at
least as large as any other independent set, so by symmetry all maximal indepen-
dent sets have the same cardinality: independence spaces satisfy (B). Conversely,
(LI1) through (LI3) and (B) clearly imply (LI4).

In this new language, Theorem 11.15 takes the form

Theorem 11.16. If K/F is a field extension, then the collection of algebraically
independent subsets of K is an independence space.
Unfortunately it is not so obvious how to show that the collection of algebraically
independent subsets of K satisfies (LI4). So let us try a different approach, in terms
of something called spanning sets. We notice that to each subset S of a vector space
its linear span S gives an abstract closure operator: namely we have

(SO1) = (CL1) S ⊂ S
(SO2) = (CL2) S ⊂ S 0 =⇒ S ⊂ S 0
(SO3) = (CL3) S = S.

But the linear span satisfies two other properties, the first of which is not sur-
prising in view of what has come before:

(SO4) if x ∈ S, there exists a finite subset S 0 ⊂ S such that x ∈ S 0 .

Famously, linear span also satisfies the following Exchange Lemma:3

(SO5) If y ∈ S ∪ x and y is not in S, then x ∈ S ∪ y.

(Proof: If y ∈ S ∪ x, there exist s1 , . . . , sn ∈ S and scalars a1 , . . . , an , a such

that y = a1 s1 + . . . + an sn + ax. If y is not in the span of S, then a 6= 0, so
x = y − −a −an
a s1 + . . . + a sn ∈ S ∪ y.)
1

Now, suppose K/F is a field extension and S is a subset of K. We will define

S to be the algebraic closure of F (S) in K. It is immediate that this “algebraic
closure” operator satisfies (SO1) through (SO4). Let us check that it also satisfies
(SO5): suppose y ∈ S ∪ x and y is not in the algebraic closure of S. Then there
exists a finite subset x1 , . . . , xn of S such that y is algebraic over F (x1 , . . . , xn , x):
i.e., there exists a polynomial f (t1 , . . . , tn , tn+1 , tn+2 ) with F -coefficients such that
f (x1 , . . . , xn , x, tn+2 ) 6= 0 and f (x1 , . . . , xn , x, y) = 0. Writing
g
X
f (x1 , . . . , xn , tn+1 , tn+1 ) = Ai (x1 , . . . , xn , tn+2 )tin+1 ,
i=0
observe that not all the polynomials Ai (x1 , . . . , xn , tn+2 ) can be zero. Since y is
not algebraic over F (S), it follows that not all of the elements A(x1 , . . . , xn , y) are
3This is an absolutely prototypical example of a lemma: the exchange lemma is the essential
kernel of content in the theory of linearly independence, and yet it is itself not very memorable or
appealing, so is doomed to be overshadowed by the figurehead theorems that it easily implies.
 4. AN AXIOMATIC APPROACH TO INDEPENDENCE 103

zero, and therefore f (x1 , . . . , xn , tn+1 , tn+1 , y) 6= 0. Since f (x1 , . . . , xn , x, y) = 0, it

follows that x is algebraic over F (S, y) as asserted.

Suppose again that X is any set equipped with a spanning operator S 7→ S,

i.e., an operator satisfying the three closure axioms (CL1) through (CL3) and also
(CL4) and (CL5). A subset S of X is a spanning set if S = X. A subset S of
X is independent if for all s ∈ S, s is not in S \ s. A basis is an independent
spanning set.

Note that it is immediate to show that the independent sets for a spanning op-
erator satisfy (LI1) through (LI3). In particular, we have (A), that bases exist and
any independent set is contained in a basis. Again it is not obvious that (LI4) is
satisfied. Rather we will show (B) directly – which is what we really want anyway
– and by the above remarks that implies (LI4).

In the following results X is always a set equipped with a spanning operator S 7→ S.

Proposition 11.17. For a subset S ⊂ X, the following are equivalent:
(i) S is a minimal spanning set of X.
(ii) S is a maximal independent set of X.
(iii) S is a basis.
Proof. (This is the usual thing.) (i) =⇒ (iii): Suppose S is minimal spanning
but not dependent; then by definition there exists s ∈ S such that x ∈ S \ s), so
that S \ s, being a closed set containing S, also contains the closure of S, i.e., X,
and we found a smaller spanning set. (iii) =⇒ (ii): if S is a basis and S ∪ {x}
is independent then x does not lie in S which is absurd since S is a spanning set.
(ii) =⇒ (i) is similar: if S were a maximal independent set but not a spanning
set, then there exists x ∈ X \ S and then S ∪ {x} is independent. 
Theorem 11.18. Let S be an independent subset of X and T a spanning set.
There exists a subset T 0 ⊂ T such that S ∪ T 0 is a basis and S ∩ T 0 = ∅.
Proof. Let I be the collection of all subsets T 0 of T such that S ∩ T 0 = ∅ and
S ∪ T 0 is indepdendent. Observe that ∅ ∈ I, so I is not itself empty. As usual,
I is closed under unions of increasing chains so by Zorn’s Lemma has a maximal
element T 0 . Let x ∈ T , and suppose that x is not in S ∪ T 0 . Then T 00 := T 0 ∪ {x}
is a strictly larger subset of T such that S ∪ T 00 is still independent, contradicting
the maximality of T 0 . Therefore
X = T ⊃ S ∪ T 0 = S ∪ T 0,
so S ∪ T 0 is a basis. 
Corollary 11.19. If X admits a finite spanning set, it admits a finite basis.
Proof. Apply Theorem 11.18 with S = ∅. 
0
Theorem 11.20. Any two bases B, B of X have the same cardinality.
Proof. Case 1: Suppose B = {x1 , . . . , xn } is a finite basis, and let B 0 be any
other basis. Let m = #B ∩ B 0 . If m = n then B ⊂ B 0 and by Proposition 11.17
distinct bases are at least incomparable, so B = B 0 . So suppose (WLOG) that
104 11. STRUCTURE OF TRANSCENDENTAL EXTENSIONS

B ∩ B 0 = {x1 , . . . , xm } with m < n. The set B \ xm+1 cannot be a spanning set,

whereas B 0 is, so there exists y ∈ B 0 B \ xm+1 . The set B1 := (B \xm+1 )∪y is inde-
pendent. By the Exchange Lemma (SO5), xm+1 ∈ (B1 ). Hence B ⊂ B1 , and since
B is a spanning set, so is B1 . Thus B1 is a basis. Notice that B1 has n elements
and also {x1 , . . . , xm , y} ⊂ B1 ∩ B 0 , so that we have replaced B by another basis
of the same cardinality and sharing at least one more element with B 0 . Repeating
this procedure will produce a finite sequence of bases B2 , B3 , each of cardinality n,
such that the last basis Bk is contained in, and thus equal to, B 0 .

Case 2: We may now suppose that B and B 0 are both infinite. For every x ∈ X,
we claim the existence of a subset Ex with the property that x ∈ Ex and for any
subset E of B such that x ∈ E, Ex ⊂ E. Assuming S the claim for the moment,
we complete the proof. Consider the subset S = x∈B 0 Ex of B. Since each Ex
is finite, #S ≤ #B 0 . On the other hand, for all x ∈ B 0 , x ∈ Ex ⊂ S, so B 0 ⊂ S
and therefore S ⊃ B 0 = X. Therefore S is a spanning subset of the basis B, so
S = B and thus #B ≤ #B 0 . By reversing the roles of B and B 0 in the argument
we conclude #B = #B 0 .
It remains to prove the claim on the existence of Ex . In turn we claim that if
E 0 and E 00 are two subsets of B such that x ∈ E 0 ∩ E 00 and x is not in the span
of any proper subset of E 0 , then E 0 ⊂ E 00 ; this certainly suffices. Assuming to the
contrary that there exists y ∈ E 0 \ E 00 . Then x is not in the span of E 0 \ y and is in
the span of (E 0 \ y) ∪ y, so by (SO5) y is in the span of (E 0 \ y) ∪ x. Since x is in
the span of E 00 , we get that y is in the span of (E 0 \ y) ∪ E 00 . But this contradicts
the fact that the (E 0 \ y)cupE 00 ∪ {y}, being a subset of B, is independent. 

Remark: A set X endowed with a spanning operator as above is often called a

finitary matroid. (The word “finitary” refers to (SO4).) Combinatoricists are es-
pecially interested in finite matroids, which includes the class of finite-dimensional
vector spaces over finite fields but not that of independent subsets of a field exten-
sion (except in the trivial case of an algebraic field extension).

For future reference, for a field extension L/K, we will refer to the matroid with
sets the subsets of L, spanning operator S 7→ S the algebraic closure of K(S) in
L and (it follows) with independent sets the algebraically independent subsets the
transcendence matroid of L/K.

We saw above how to go from a finitary matroid to an independence space, namely

by decreeing a subset S ⊂ X to be dependent if there exists x ∈ S such that
x ∈ S \ x. Conversely, to every independence space we can associate a finitary
matroid: define the span Y of a subset Y to be the set of x ∈ X such that S ∪ x
is dependent. This complete equivalence between concepts of linear independence
and spanning seems a bit unexpected, even in the context of vector spaces.

For finite matroids, combinatorialists know at least half a dozen other equivalent
axiomatic systems: e.g. in terms of graphs, circuits, “flat” subspaces and pro-
jective geometry. As above, demonstrating the equivalence of any two of these
systems is not as easy as one might expect. This phenomenon of multiple nonob-
viously equivalent axiomatizations has been referred to, especially by G. Rota, as
 5. MORE ON TRANSCENDENCE DEGREES 105

cryptomorphism. Of course every twenty-first century student of mathematics

has encountered crytomorphism (although it seems that the multiplicity is espe-
cially large for finite matroids!). In several essays, Rota saw cryptomorphism as a
warning not to take any particular axiomatization of a theory or structure too seri-
ously. This seems fair, but since the different axiomatizations can lead to different
and possibly easier proofs, perhaps it should also be viewed as an instance of the
inherent richness of mathematical concepts.

5. More on Transcendence Degrees

Proposition 11.21. Let L/K be a field extension and T a subset of L such
that L = K(T ). Then trdeg(L/K) ≤ #T .
Proof. In the transcendence matroid of L/K, T is a spanning set. According
to Theorem 11.18 with S = ∅, some subset T 0 of T is a basis for the matroid, i.e.,
a transcendence basis for L/K. Thus
trdeg(L/K) = #T 0 ≤ #T. 
Theorem 11.22. Let F ⊂ K ⊂ L be a tower of field extensions.
a) If {xi }i∈I is a transcendence basis for K/F and {yj }j∈J is a transcendence basis
for L/K, then {xi , yj } is a transcendence basis for L/F .
b) We have trdeg(L/F ) = trdeg(L/K) + trdeg(K/F ).
Proof. a) We first show that {xi , yj } is an algebraically independent set.
Choose any finite subsets of {xi } and {yj }: for ease of notation, we rename the ele-
ments x1 , . . . , xm , y1 , . . . , yn . Suppose there exists a polynomial P ∈ F [t1 , . . . , tm+n ]
such that P (x1 , . . . , xm , y1 , . . . , yn ) = 0. Put Q(t1 , . . . , tn ) = P (x1 , . . . , xm , t1 , . . . , tn ) ∈
K[t]. Then Q(y1 , . . . , yn ) = 0 implies Q(t1 , . . . , tn ) = 0. Each coefficient of this
polynomial is a polynomial expression in x1 , . . . , xm with F -coefficients, and the
algebraic independence of the xi ’s implies that each of these coefficients is equal to
0. Thus P = 0. Let K0 = F ({xi }), so K/K0 is algebraic. Let L0 = K({yj }, so
L/L0 is algebraic. Let z ∈ L. Then z satisfies a polynomial equation with coeffi-
cients in L0 . Since K/K0 is algebraic, z also satisfies a polynomial equation with
coefficients in K0 ({yj }) = F ({xi , yj }).
b) By part a), {xi , yj } is a transcendence basis for L/F , of cardinality #I + #J =
trdeg(K/F ) + trdeg(L/K). 
Exercise 11.9. Let M/F be a field extension, and let K, L be subextensions of
M/F . Suppose K/F is finite and L/F is purely transcendental. Show [LK : L] =
[K : F ].
(Suggestion: reduce to the case K = F [t]/(p(t)) and L = F (t). For this case, if
the polynomial p(t) factors over F (t), then by taking t = a for a ∈ F we get a
factorization over F . One has to be a little careful here in order to avoid values a
which make the denominator of one of the rational functions equal to 0.)4
Theorem 11.23.
For F ⊂ K ⊂ L a tower of field extensions, the following are equivalent:
(i) K/F and L/K are both finitely generated.
(ii) L/F is finitely generated.
4This result will become much more clear following our later discussion of linear disjoint-
ness. The reader may prefer to defer the exercise until then.
106 11. STRUCTURE OF TRANSCENDENTAL EXTENSIONS

Proof. (i) =⇒ (ii): If K = F (x1 , . . . , xm ) and L = K(y1 , . . . , yn ), then

L = F (x1 , . . . , xm , y1 , . . . , yn ).
(ii) =⇒ (i): It is immediate that if L/F is finitely generated then so is L/K
for any subextension K of L/F : any finite generating set for L/F is also a finite
generating set for L/K. Let z1 , . . . , ze be a transcendence basis for K/F . Then
F (z1 , . . . , ze )/F is finitely generated, so it suffices to show that the algebraic ex-
tension K/F (z1 , . . . , ze ) is finitely generated. Moreover, L/F (z1 , . . . , ze ) is finitely
generated, so it is enough to prove the result with F (z1 , . . . , ze ) in place of F and
thus we may assume that K/F is algebraic.
We are thus reduced to showing: if L/K(t1 , . . . , tn ) is a finite extension of a
rational function field and K/F is an algebraic extension, then L/F finitely gen-
erated implies K/F finitely generated – or, equivalently since K/F is algebraic –
that K/F is finite. But suppose not: then for all d ∈ Z+ there exists a subex-
tension Kd of K/F such that [Kd : F ] ≥ d. By the preceding exercise we have
[Kd (t1 , . . . , tn ) : F (t1 , . . . , tn )] = [Kd : F ] ≥ n. Thus L/F (t1 , . . . , tn ) is an algebraic
extension but
[L : F (t1 , . . . , tn )] ≥ [K(t1 , . . . , tn ) : F (t1 , . . . , tn ) ≥ ℵ0 ,
so it is algebraic of infinite degree, hence not finitely generated: contradiction! 
Exercise 11.10. Let k be any field. Consider the polynomial ring R = k[x, y]:
note that it is finitely generated as a k-algebra. Show that there is a k-subalgebra of
R which is not finitely generated. (Thus Theorem 11.23 exhibits a property of field
extensions without analogue in the study of commutative rings.)
Exercise 11.11. For a field extension L/K, show the following are equivalent:
(i) The subfield lattice of L/K is Noetherian: there are no infinite chains
K = F0 ( F1 ( . . . Fn ( . . . ⊂ L.
(ii) The extension L/K is finitely generated.

6. Digging Holes
Lemma 11.24. Let K/k be an extension of commutative rings, and let S be a
subset of K that is disjoint from k. Then there is a ring l such that k ⊂ l ⊂ K, l
is disjoint from S and l is maximal with respect to these properties.
Exercise 11.12. Prove Lemma 11.24. (Use Zorn’s Lemma!)
In this section we are interested in the case of Lemma 11.24 in which K/k is a a field
extension, and K is algebraically closed, so henceforth we impose these hypotheses.
Proposition 11.25 (Quigley-McCarthy [Qu62] [Mc67]). Let K/k be an ex-
tension of fields with K algebraically closed, let S ⊂ K be a finite subset that is
disjoint from S, and let l be a subextension of K/k that is maximal with resepect
to the exclusion of S. Then K/l is an algebraic field extension.
Proof. Seeking a contradiction, let t ∈ K be transcendental over l. By our
hypothesis on l, there is an element a1 ∈ l(t) ∩ S. The element a1 is transcendental
over l – otherwise the rational function field l(t) contains a nontrivial finite degree
field extension l(a1 )/l, and tensoring the inclusion l(a1 ) ,→ l(t) with l(a1 ) gives
a contradiction. It follows that l ( l(a21 ) and a2 ∈ / l(a21 ), so l(a21 ) contains some
 6. DIGGING HOLES 107

element a2 ∈ S \ {a1 }. By Lüroth’s Theorem, l(a2 )/l is purely transcendental. If

#S = n, we may repeat this argument, getting finally
l ( l(a2n ) ( l(a2n−1 ) ( . . . l(a21 )
/ l(a2n ) for all 1 ≤ i ≤ n. But then
and ai ∈
l ( l(a2n ) ⊂ K
and l(a2n ) ∩ S = ∅, contradicting the defining property of l. Thus l/k is algebraic.

 CHAPTER 12

Linear Disjointness

1. Definition and First Properties

Let E/F be a field extension, and let R, S be F -subalgebras of E. We say that R
and S are F-linearly disjoint in E if the canonical map R ⊗F S → E is injective.
(If the ambient field E is understood, we will just say that R, S are F -linearly
disjoint, or that they are linearly disjoint over F. In fact the depenence on E is
often suppressed, for reasons that will be explored soon enough.)
Lemma 12.1. Let E/F be a field extension, and let K, L be subextensions of
finite degree over F . Then K and L are linearly disjoint over F iff
[KL : F ] = [K : F ][L : F ].
Proof. Since K and L are finite-dimensional over F , the compositum KL is
the F -algebra generated by K and L, so the canonical map τ : K ⊗F L → KL is
always surjective. Since its source and target are both finite-dimensional F -vector
spaces, τ is injective iff
[K : F ][L : F ] = dimF K ⊗F L = dimF LK. 
Exercise 12.1. Let K, L be finite degree extensions of a field F of coprime
degrees. Show: K, L are F -linearly disjoint.
Lemma 12.2. If R, S are F -linearly disjoint in E, then R ∩ S = F .
Proof. By contraposition: suppose there exists u ∈ (R ∩ S) \ F . We may then
choose F -bases A of R and B of S such that {1, u} ⊂ A ∩ B. The elements 1 ⊗ u
and u ⊗ 1 are then F -linearly independent in R ⊗F S but under ι : R ⊗F S → E
they both get mapped to u, so ι is not injective. 
√
3
Exercise 12.2. a) Let F = Q and E = C. Show that K = Q( 2) and
2πi √
3
L = Q(e 3 2) are not linearly disjoint over F , even though K ∩ L = F .
b) Try to generalize the result of part a), for instance as follows: if K/F is alge-
braic and not normal, then inside any algebraic closure E of K there exists a field
extension L/F such that K ∩ L = F but K, L are not F -linearly disjoint in E.
Exercise 12.3. Let R, S be F -subalgebras of E/F . Show that the following are
equivalent:
(i) R and S are linearly disjoint over F .
(ii) For all F -linearly independent subsets {ai }i∈I of R and {bj }j∈J of S,
{ai bj }(i,j)∈I×J is F -linearly independent in E.
(iii) For all positive integers m and n, if a1 , . . . , am are F -linearly independent
in R and b1 , . . . , bn are F -linearly independent in S, then a1 b1 , . . . , am , b1 , a2 b1 , . . . , am bn
are F -linearly indpendent in E.
109
110 12. LINEAR DISJOINTNESS

Exercise 12.4. (Linear disjointness is preserved by direct limits) Let R be an

F -subalgebra of E/F . Suppose R = lim Ri is a direct limit of a family {Ri }i∈I of
−→
F -subalgebras. Show: for any F -subalgebra S of E/F , R and S are linearly disjoint
iff for all i ∈ I, Ri and S are linearly disjoint.
Exercise 12.5. Suppose R, S are linearly disjoint subalgebras of E/F . Let
R0 ⊂ R and S 0 ⊂ S be F -subalgebras. Show: R0 and S 0 are linearly disjoint over
F.
Lemma 12.3. Two subalgebras R and S of E/F are linearly disjoint over F iff
the subfields they generate, say K and L, are linearly disjoint over F .
Proof. Suppose that R and S are linearly disjoint over F . It is enough to
show that if k1 , . . . , km are F -linearly independent elements of K and l1 , . . . , ln
are F -linearly independent elements of L, then {ki lj }1≤i≤m,1≤j≤n are F -linearly
independent in E. There exist a, a1 , . . . , am ∈ R such that ki = aai for all i, and
b
simiarly there exist b, b1 , . . . , bn ∈ S such that lj = bj for all j. Then if αij ∈ F is
P ai bj P
such that i,j αij ab = 0, then multiplying by ab gives i,j αij ai bj = 0, and by
assumption αij = 0 for all i and j.
The converse is immediate from Exercise 12.5. 
Thus it is no loss of generality to speak of linear disjointness of subfields of E/F ,
but it is often convenient to phrase things in terms of subdomains of these fields.
Proposition 12.4. Let K, L be subextensions of a field extension E/F . The
following are equivalent:
(i) K and L are linearly disjoint over F .
(ii) Every F -linearly independent subset S of K is L-linearly independent in E.
(ii0 ) Every F -linearly independent subset T of L is K-linearly independent in E.
(iii) There is an F -basis A of K which is L-linearly independent as a subset of E.
(iii0 ) There is an F -basis B of L which is K-linearly independent as a subset of E.
Proof. (i) =⇒ (ii): Let A be F -linearly independent in K. Consider any
finite subset of elements of A, say k1 , . . . , kn , and let β1 , . . . , βn ∈ L be such that
(14) β1 k1 + . . . + βn kn = 0.
an F -basis {lj }j∈J for L, so that there are unique αij ∈ F such that for all
Choose P
i, βi = j αij lj . Substituting this into (14) gives
X
αij ki lj = 0.
i,j

By Exercise 12.3 this forces αij = 0 for all i, j and thus βj = 0 for all k, so the ki ’s
are L-linearly independent.
(i) =⇒ (ii0 ): The above proof works with the roles of K and L reversed.
(ii) =⇒ (i): By Exercise 12.3, it is enough to fix m, n ∈ Z+ let k1 , . . . , km
be F -linearly independent elements of K and l1 , . . . , ln be F -linearly independent
elements of L and show that {ki lj P } are F -linearly independent elements of E.
Suppose that αij ∈ F are such that i,j αij ki lj = 0. But we may rewrite this as
(α11 l1 + . . . + α1n ln )k1 + . . . + (αm1 l1 + . . . + αmn ln )km = 0.
By hypothesis the ki ’s are L-linearly independent, so this forces all the coefficients
of the above equation to be equal to zero, which in turn, since the lj ’s are F -linearly
 1. DEFINITION AND FIRST PROPERTIES 111

independent, forces all the αij ’s to be zero.

(ii0 ) =⇒ (i) in the same way.
(ii) =⇒ (iii) and (ii0 ) =⇒ (iii0 ) are immediate.
(iii) =⇒ (ii): Let S be an F -linearly independent subset of K, and complete it
to a basis A0 of K. Let ϕ : A0 → A be a bijection and Φ the induced F -linear
automorphism of K. Suppose that A0 is not L-linearly independent, i.e., there
exists a finite subset a01 , . . . , a0n of A’ and β1 , . . .P
, βn ∈ L, not all zero, such that
0
P
i β i ai = 0. Applying Φ to this relation gives i βi ai = 0, so that A is not L-
linearly independent, contradiction. Thus A0 is L-linearly independent independent
and a fortiori so is its subset S.
(iii0 ) =⇒ (ii0 ) in the same way. 
Remark: Some source take condition (ii) of Proposition 12.4 to be the definition
of linear disjointness. This has the advantage of not requiring any knowledge of
tensor products on the part of the reader. All the other advantages, however, seem
to lie with the tensor product definition. For instance, it is clearly symmetric with
respect to K and L.
Exercise 12.6. Let K, L be subfields of E/F , and let R be an F -subalgebra of
K with fraction field K. Suppose that there exists a K-basis of R which is L-linearly
independent in E. Show that K, L are F -linearly disjoint in E.
Theorem 12.5. Let C/F be a field extension, and let K, L, M be subextensions
of C/F with K ⊂ M . The following are equivalent:
(i)) We have that M and L are linearly disjoint over F .
(ii) We have:
a) The fields K and L are linearly disjoint over F , and
b) The fields M and KL are linearly disjoint over K.
Proof. We consider the F -algebra maps
ι ϕ1 ϕ2
M ⊗F L → M ⊗K (K ⊗F L) → M ⊗K K[L] → M [L].
The map ι is an isomorphism: this a special case of the “telescoping tensor identity”
M ⊗T (T ⊗R P ) ∼ = M ⊗R P
valid for a homomorphism R → T of commutative rings, an R-module P and a T -
module M . In this particular case it can be seen as follows: if {mi }i∈I is a K-basis
for M and {αj }j∈J is an F -basis for K, then {αj mi | (i, j) ∈ I ×J} is an F -basis for
M , so if {lx }x∈X is an F -basis for L, then {(αj mi ) ⊗ lx | (i, j, x) ∈ I × J × X} is an
F -basis for M ⊗F L. This maps under ι to (αj mi )⊗(1⊗lx ) = αj (mi ⊗(1⊗lx )), which
is an F -basis for M ⊗K (K ⊗F L) since mi ⊗ (1 ⊗ lx ) is a K-basis for M ⊗ (K ⊗F L).
The composition of these maps is the natural multiplication map
ϕ : M ⊗F L → M [L],
so in all cases we have that ϕ1 , ϕ2 and ϕ are surjective.
(i) =⇒ (ii): Suppose M and L are linearly disjoint over F : thus ϕ is an isomor-
phism. Since we have that ϕ = ϕ2 ◦ ϕ1 ◦ ι, that ι is an isomorphism and that ϕ1
and ϕ2 surjective, we deduce that ϕ2 ◦ ϕ1 is an isomorphism, hence ϕ1 is injective,
hence ϕ1 is an isomorphism, hence ϕ2 is an isomorphism. Because M is a nonzero
K-vector space, the injectivity of ϕ1 implies the injectivity of K ⊗F L → K[L], so
K and L are linearly disjoint over F . The injectivity of ϕ2 implies that M and
112 12. LINEAR DISJOINTNESS

K[L] are linearly disjoint over K, which by Lemma 12.3 implies that M and KL
are linearly disjoint over F .
(ii) =⇒ (i): If K and L are linearly disjoint over F , then ϕ1 is injective. If M and
KL are linearly disjoint over K then by Exercise 12.5 we have that M and K[L]
are linearly disjoint over K, so ϕ2 is injective. Therefore ϕ is injective, so M and
L are linearly disjoint over F . 

2. Intrinsic Nature of Linear Disjointness

The definiton of linear disjointness is initially hard to process because it involves
four different algebras. In fact the dependence of the definition on the “ambient”
field E is in many cases rather weak. One easy of instance of this is given in the
following exercise.
Exercise 12.7. Let K, L be subextensions of a field extension E/F , and let
E 0 /E be any field extension. Show: K, L are F -linearly disjoint as subfields of E
iff they are F -linearly disjoint as subfields of E 0 .
We now look more deeply into the dependence on the ambient field E, following
a MathOverflow discussion led by Andrew Critch. Let F be a field, and let K, L
be field extensions of F . We say that K, L are somewhere linearly disjoint
over F if there exists a field extension E/F and F -algebra embeddings of K and
L into E such that K, L are F -linearly disjoint in E. Further, we say that K, L
are everywhere linearly disjoint over F if for all field extensions E/F and all
F -algebra embeddings of K, L into E, K, L are F -linearly disjoint in E.
Certainly we want everywhere linearly disjoint over F to imply somewhere
linearly disjoint over F . To see this there is a minor technicality to be disposed of,
which is treated in the next exercise.
Exercise 12.8. a) Let F be a field and K, L be field extensions of F . Show:
there exists a field extension E and F -algebra embeddings of K and L into E.
Show that for instance one may take E to be any algebraically closed field such that
trdeg(E/F ) ≥ max trdeg(K/F ), trdeg(L/F ).
b) Deduce: if K, L are everywhere linearly disjoint over F then they are somewhere
linearly disjoint over F .
Exercise 12.9. Let F be any field, and put K = L = F (t).
a) Take E = F (t) to show that K, L are not everywhere linearly disjoint.
b) Take E = F (a, b) (rational function field in two variables) to show that K and
L are somewhere linearly disjoint.
Proposition 12.6. Let F be a field, and let K and L be field extensions of F .
The following are equivalent:
(i) K, L are somewhere F -linearly disjoint.
(ii) The tensor product K ⊗F L is a domain.
Proof. If K⊗F L can be embedded into a field, then it is a domain. Conversely,
if K ⊗F L is a domain, it can be embedded into its fraction field. 
Proposition 12.7. Let F be a field, and let K and L be field extensions of F .
The following are equivalent:
(i) K, L are everywhere F -linearly disjoint.
(ii) K ⊗F L is a field.
 2. INTRINSIC NATURE OF LINEAR DISJOINTNESS 113

Proof. (i) =⇒ (ii): In order to show that the (evidently nonzero, since it
contains F ) ring R = K ⊗F L is a field, it suffices to show that the only maximal
ideal is (0). So let m be a maximal ideal of R. Then E = R/m is a field extension
of K, L and the induced map K ⊗F L → E is precisely the quotient map R → R/m.
Since this map is injective, m = (0).
(ii) =⇒ (i): If R = K ⊗F L is a field, then every homomorphism into a nonzero
ring – and in particular, any F -algebra homomorphism – is injective. 
Theorem 12.8. Let K, L be field extensions of F .
a) Suppose that K, L are everywhere F -linearly disjoint. Then at least one
of K, L is algebraic over F .
b) Conversely, suppose that at least one of K, L is algebraic over F . Then
K, L are somewhere F -linearly disjoint iff they are everywhere F -linearly
disjoint.
Proof. a) If K and L are transcendental over F , then they admit subexten-
sions K 0 = F (a), L0 = F (b). By Exercise 12.5, it suffices to show that F (a) and
F (b) are not everywhere F -linearly disjoint over F . To see this take E = F (t) and
map K 0 → E by a 7→ t and L0 → E by b 7→ t and apply Lemma 12.2.
b) Because every algebraic extension is a direct limit of finite degree extensions, by
Exercise 12.4 it is no loss of generality to assume that K/F is finite, and in light
of Propositions 12.6 and 12.7, we must show that if K ⊗F L is a domain then it is
a field. But if {k1 , . . . , kn } is a basis for K/F , then k1 ⊗ 1, . . . , kn ⊗ 1 is a basis for
K ⊗F L over L, so K ⊗F L is a domain and a finitely generated L-module. Therefore
it is a field, by an elementary argument which we have seen before (and which is a
special case of the preservation of Krull dimension in an integral extension). 
In conclusion: the notion of F -linear disjointness of two field extensions K, L is in-
trinsic – independent of the embeddings into E – iff at least one of K, L is algebraic
over F . In most of our applications of linear disjointness this hypothesis will be
satisfied, and when it is we may safely omit mention of the ambient field E.

Here is a first result with our new convention in force.

Theorem 12.9. Let K/F be purely transcendental and L/F be algebraic. Then
K, L are F -linearly disjoint.
Proof. By Exercise 12.4 and Lemma 12.3 it is enough to show that for all
n ∈ Z+ , F [x1 , . . . , xn ], L are F -linearly disjoint. By Corollary 4.5, this holds
iff F [x1 , . . . , xn ] ⊗F L is a domain. It is clear that the F -basis of F [x1 , . . . , xn ]
consisting of monomials remains L-linearly independent in L[x1 , . . . , xn ] and by
Proposition 12.4 this implies that F [x1 , . . . , xn ] and L are F -linearly disjoint. In
particular, the natural map F [x1 , . . . , xn ] ⊗K L → L[x1 , . . . , xn ] is an isomorphism
of L-algebras. 
Theorem 12.10. Let K, L be two field extensions of F with K/F purely tran-
scendental. Then K ⊗F L is a domain.
Proof. The F -algebra K ⊗F L is the direct limit of the F -algebras Ki ⊗F Li
as Ki ranges over finitely generated subextensions of K/F and Li ranges over
finitely generated subextensions of L/F . Since the direct limit of domains is a
domain, we have reduced to the case in which K and L are finitely generated
114 12. LINEAR DISJOINTNESS

over F , say E2 = F (s1 , . . . , sm ), and E1 = F (t1 , . . . , tn , x1 , . . . , xp ), where the

ti ’s are independent indeterminates over F and for all 1 ≤ k ≤ p, the field exten-
sion F (t1 , . . . , tn , x1 , . . . , xk )/F (t1 , . . . , tn , x1 , . . . , xk−1 ) has finite degree. Put K1 =
F (t1 , . . . , tn ). Let E be the algebraic closure of the fraction field of F [s1 , . . . , sm ]⊗F
F [t1 , . . . , tn ]. We may embed E2 and L in E, and then E2 and K1 are linearly dis-
joint over F . Since K1 E2 /K1 is purely transcendental and E1 /K1 is algebraic, by
Theorem 12.13 K1 E2 and E1 are linearly disjoint over K1 . By Theorem 12.5, the
fields E1 and E2 are linearly disjoint over F , hence E1 ⊗F E2 is a domain. 

3. Linear Disjointness and Normality

Proposition 12.11. Let E/F be a field extension, and let K, L be two finite
degree subextensions, with K/F and L/F both Galois extensions. Then K, L are
F -linearly disjoint (in E, but by Theorem 12.8 this does not matter) iff K ∩ L = F .
Proof. The forward direction holds for any pair of subextensions by Lemma
12.2. Conversely, assume K ∩ L = F . The image of K ⊗F L in E is the compositum
KL, which is finite Galois over F since normality, separability and finiteness of
degree are all preserved by finite composita. Let d = [KL : F ], dK = [K : F ]
and dL = [L : F ]. We have a surjective F -linear map ι : K ⊗F L → KL between
two finite-dimensional F -vector spaces, so ι is injective iff [KL : F ] = [K ⊗F L :
F ] = dK dL . Let HK = Aut(KL/K) and HL = Aut(KL/L). Since K/F and L/F
are Galois, HK and HL are normal in G = Aut(L/K). Moreover, since KL is the
compositum of (KL)HK and (KL)HL , KL = KLHK ∩HL , i.e., HK ∩ HL = {e}.
Therefore HK HL is a subgroup of G, the internal direct product of HK and HL .
Moreover,
F = K ∩ L = (KL)HK ∩ (KL)HL = KLhHK ,HL i = KLHK HL ,
so HK HL = G. It follows that G = HK × HL and therefore
d d
d = #G = #HK #HL =
dK dL
and thus [K ⊗ L : F ] = dK dL = d = [KL : F ]. 

Theorem 12.12. Let E/F be a field extension, and let K, L be two algebraic
subextensions such that K/F is Galois. Then K, L are F -linearly disjoint (in E,
but by Theorem 12.8 this does not matter) iff K ∩ L = F .
Proof. By a now familiar argument involving Exercise 12.5 and Proposition
12.7, we reduce to the case in which K/F is finite Galois. Now by the theorem of
Natural Irrationalities, we have
[KL : F ] = [KL : L][L : F ] = [K : K ∩ L][L : F ],
so
[KL : F ] = [K : F ][L : F ] ⇐⇒ K ∩ L = F. 

4. Linear Disjointness and Separability

Lemma 12.13. Let K/F be a separable field extension of characteristic p > 0,
and let a1 , . . . , an be F -linearly independent elements of K. Then for all e ∈ Z+ ,
e e
ap1 , . . . , apn are F -linearly independent.
 4. LINEAR DISJOINTNESS AND SEPARABILITY 115

Proof. By replacing K by F (a1 , . . . , an ), we may assume that a1 , . . . , an is a

basis for K and thus [K : F ] = n.
e e
Step 1: Let K 0 = F (ap1 , . . . , apn ), so that K 0 is a subextension of K/F . Observe
that for all i, the element ai is both separable and purely inseparable over K 0 , so
ai ∈ K 0 for all i and thus K 0 = K and [K 0 : F ] = n.
e e
Step 2: Let V be the F -subspace spanned by ap1 , . . . , apn . It is enough to show
that V is closed under multiplication: then it is a subring of a field which is finite
dimensional as an F -algebra and therefore a field and therefore the F -subalgebra
1 e
generated by ap1 , . . . , apn . By Step 1, this means V = K 0 = K and thus [V : F ] = n.
e e
Therefore the n-element spanning set ap1 , . . . , apn is linearly independent.
Step 3: To show that V is a subalgebra, it is enough to show that the product of
two basis elements is an F -linear combination of the basis elements. To see this, fix
any 1 ≤ i, j ≤ n. Since a1 , . . . , an span K over F , there exist α1 , . . . , αn such that
X
ai aj = αi ai
i
e
Raising both sides to the p th power gives
e e X pe pe
api apj = αi ai ,
i
e e
which shows that api apj lies in V . 

Proposition 12.14. Let K/F be a separable algebraic extension. Then K and

−∞
Fp are F -linearly disjoint.
−∞ −e
Proof. Since F p = lim F p , by Exercises 12.3, 12.4 and Proposition 12.4
−→
it is enough to show that for all e, m ∈ Z+ , if a1 , . . . , an are F -linearly independent
−e
elements of K, they are also F p -linearly independent. But this last statement
pe e
holds iff a1 , . . . , apn are F -linearly independent, which they are by Lemma 12.13.

The natural question to ask at this point is: can an inseparable extension K/F be
−∞
linearly disjoint from F p ? It follows from what we already know about sepa-
rable extensions that the answer is no if K/F is inseparable and normal, for then
by Corllary 5.20 it contains a nontrivial purely inseparable subextension and thus
−1 −∞
F ( K ∩ F p ⊂ K ∩ F p . In fact, as we are about to see, among algebraic field
−∞
extensions K/F , being linearly disjoint from F p characterizes separable exten-
sions. But actually we can go further, with the following definitions.

A separating transcendence basis for a field extension K/F is an algebraically

independent subset S of K such that K/F (S) is separable algebraic.

It is clear that separating transcendence bases need not exist, e.g. an insepara-
ble algebraic extension will not admit a separating transcendence basis. On the
other hand, it is clear that separable algebraic extensions and purely transcenden-
tal extensions both admit separating transcendence bases: as with being linearly
disjoint from the perfect closure, this is something that these apparently very dif-
ferent classes of extensions have in common.
116 12. LINEAR DISJOINTNESS

We say that a field extension K/F is separably generated if it admits a sep-

arating transcendence basis.
Exercise 12.10. Give an example of a separably generated field extension ad-
mitting a transcendence basis that is not a separating transcendence basis.
An arbitrary field extension K/F is separable if every finitely generated subex-
tension admits a separating transcendence basis.

And now the main theorem on separable extensions.

Theorem 12.15. (Mac Lane) Let F be a field of characteristic p > 0, and let
E/F be a field extension. The following are equivalent:
(i) The extension E/F is separable: every finitely generated subextension is
separably generated.
−∞
(ii) The extensions E and F p are F -linearly disjoint.
−1
(iii) The extensions E and F p are F -linearly disjoint.
Proof. (i) =⇒ (ii): Since every field extension is the direct limit of its fi-
nite generated subextensions, by Exercise LD2 we may assume that E/F is finitely
generated and thus separably generated, so let B be a transcendence basis for E/F
−∞
such that E/F (B) is separable algebraic. By Proposition 12.14, E and F (B)p
−∞ −∞ −∞
are F (B)-linearly disjoint. Since F (B)p ⊃ F p (B), it follows that F p (B)
−∞
are F (B)-linearly disjoint. Theorem 12.5 implies that E and F p are F -linearly
disjoint.
(ii) =⇒ (iii) is immediate.
−1
(iii) =⇒ (i): Suppose that E and F p are F -linearly disjoint. We will prove
by induction on n that for all n ∈ N, if K = F (a1 , . . . , an ) is a finitely gener-
ated subextension of E/F then there exists a subset S ⊂ {a1 , . . . , an } which is a
separating transcendence basis for K/F . When n = 0, K = F and the result is
trivial. The result is also clear if a1 , . . . , an are algebraically independent. Hence
we may assume (after relabelling) that there exists r < n such that a1 , . . . , ar are a
transcendence basis for K/F . Let f ∈ K[t1 , . . . , tr+1 ] be a polynomial of minimal
total degree such that f (a1 , . . . , ar+1 ) = 0; necessarily f is irreducible.
We claim f is not of the form g(tp1 , . . . , tpr ). If it were, there would exist h ∈
−1
F p [t1 , . . . , tr+1 ] such that g(tp1 , . . . , tpr ) = h(t1 , . . . , tr )p with h(a1 , . . . , ar+1 ) = 0.
Let {mi } be the monomials occurring in h. Then the elements mi (a1 , . . . , ar+1 )
−1
are F p -linearly dependent, so by hypothesis they are F -linearly dependent. This
gives a nontrivial polynomial relation in the ai of degree less than the degree of h,
contradiction.
It follows that there is at least one i, 1 ≤ i ≤ r + 1, such that f (t1 , . . . , tr+1 )
is not a polynomial in tpi . Then ai is algebraic over F (a1 , . . . , ai−1 , ai+1 , . . . , ar+1 )
and thus {a1 , . . . , ai−1 , ai+1 , . . . , ar+1 } is a transcendence basis for K/F . So
F [a1 , . . . , ai−1 , t, ai+1 , . . . , ar+1 ] ∼
= F [t1 , . . . , tr+1 ],
so f (a1 , . . . , ai−1 , t, ai+1 , . . . , ar+1 ) is ireducible in F [a1 , . . . , ai−1 , t, ai+1 , . . . , ar+1 ],
so by Gauss’s Lemma it is irreducible in F (a1 , . . . , ai−1 , ai+1 , . . . , ar+1 )[t]. Since
ai is a root of f (a1 , . . . , ai−1 , t, ai+1 , . . . , ar+1 ) and this is not a polynomial in tp ,
ai is separable algebraic over F (a1 , . . . , ai−1 , ai+1 , . . . , ar+1 ) and hence over L :=
F (a1 , . . . , ai−1 , ai+1 , . . . , an ). The induction hypothesis applies to L to give a subset
 4. LINEAR DISJOINTNESS AND SEPARABILITY 117

{ai1 , . . . , air } of {a1 , . . . , ai−1 , ai+1 , . . . , an } that is a separating transcendence base

for L/F . Since ai is separable algebraic over L, it is separable algebraic over
F (ai1 , . . . , air ). So {ai1 , . . . , air } is a separating transcendence basis for K/F . 
Remark 12.1. By our earlier results on linear disjointness, the conditions in
Theorem 12.15 are also equivalent to any of the following ones:
−1
(iv) E ⊗F F p is a field.
−1
(v) E ⊗F F p is a domain.
−∞
(vi) E ⊗F Fp is a field.
−∞
p
(vii) E ⊗F F is a domain.
Example 12.16. (Mac Lane): Let F be any field of characteristic p > 0, F (t)
−1 −2
a rational function field. Let E = F (t, tp , tp , . . .). Then any finitely generated
subextension of E/F is isomorphic to F (t) and thus separably generated. But E
itself does not admit a separating transcendence basis. Thus E/F is separable but
not separably generated.
Exercise 12.11. Let K/F be a field extension in characteristic p > 0. Show
that K/F is separable iff: for every F -linearly subset S ⊂ K, the set S p := {sp |
s ∈ S} is also F -linearly independent.
Corollary 12.17. Let F ⊂ K ⊂ L be a tower of field extensions in charac-
teristic p > 0.
a) If L/F is separable, then K/F is separable.
b) If K/F and L/K are separable, then L/F is separable.
c) If L/F is separable and K/F is algebraic, then L/K is separable.
Proof. a) We use Theorem 12.15 and Remark 12.1: if L/F is separable, then
−1 −1
L ⊗F F p is a domain, hence so its its subring K ⊗F F p , so K/F is separable.
−1
b) If K/F and L/K are separable, then K and F p are linearly disjoint over F
−1
and L and K p are linearly disjoint over K. Since F 1/p K ⊂ K 1/p , we get that
F 1/p K and L are linearly disjoint over F , so L and F 1/p are linearly disjoint over
F by Theorem 12.5. Thus L/F is separable.
c) Suppose L/F is separable and K/F is algebraic. Part a) gives that K/F is separa-
ble algebraic. Let K(x1 , . . . , xn ) be a finitely generated subextension of L/K. Then
F (x1 , . . . , xn ) is a finitely generated subextension of the separable extension L/F ,
so it admits a separating transcendence basis: there are y1 , . . . , ym ∈ L such that
F (x1 , . . . , xn )/F (y1 , . . . , ym ) is separable algebraic. Since K/F is separable alge-
braic, so is K(x1 , . . . , xn )/F (x1 , . . . , xn ), and therefore K(x1 , . . . , xn )/F (t1 , . . . , tm )
is separable algebraic. Thus t1 , . . . , tm is a separating transcendence basis for
K(x1 , . . . , xn )/F , so K(x1 , . . . , xn )/F is separably generated. 
Example 12.18. Let F be a field of characteristic p, let L = F (t) and let
K = F (tp ). Then L/F is separable but L/K is not. This shows that in Corollary
12.17, the hypothesis that K/F be algebraic cannot be removed.
Exercise 12.12. Show: a separably generated extension is separable.
Exercise 12.13. a) Show: separably generated extensions do not satisfy the
base change property (DC2).
(Suggestion: let F be a field of characteristic p > 0, let K = F (t1/p ) and L = F (t).)
118 12. LINEAR DISJOINTNESS

b) Conclude that separably generated extensions – and thus also separable extensions
– do not form a distinguished class in the sense of Lang.
c) Prove or disprove: the compositum of two separably generated extensions is sep-
arably generated.
d) Prove or disprove: the compositum of two separable extensions is separable.

5. Interlude
For later use we record the following result, whose proof will use methods of
commutative algebra rather than field theory.1
Theorem 12.19 (Integrality of Products). Let k be an algebraically closed field,
and for i = 1, 2, let Ri be a domain that is finitely generated as a k-algebra. Then
R1 ⊗k R2 is a domain.
Proof. The statement holds trivially if either R1 = k or R2 = k, so assume
that each of R1 and R2 have positive transcendence degree over k. Seeking a
contradiction, suppose that we have nonzero elements
m
X n
X
x= ai ⊗ bi , y = cj ⊗ dj ∈ R1 ⊗k R2
i=1 j=1

such that xy = 0. We may assume that each of b1 , . . . , bm and d1 , . . . , dn are k-

linearly independent in R2 , and we may also assume that a1 c1 6= 0. Because R1
is a Hilbert-Jacobson ring [Cl-CA, Prop. 11.3], there is a maximal ideal m of R1
such that a1 c1 ∈ R1 \ m. Denote the quotient map R1 → R1 /m by a 7→ a. Because
k is algebraically closed, we have R1 /m = k [Cl-CA, Thm. 11.5]. Then under the
homomorphism R1 ⊗k R2 → R1 /m ⊗k R2 = R2 the relation xy = 0 becomes
X X
( ai bi )( cj dj ) = 0,
i j

which exhibits a product of two nonzero elements in the domain R2 being zero, a
contradiction. 
Although our motivation for proving Theorem 12.19 is field theoretic – it will be
used in the proof of Theorem 12.22 giving an important equivalent condition on a
class of field extensions – for those with some familiarity with algebraic geometry
it is hard not to notice that it is saying precisely that over an algebraically closed
field, the product of two integral affine varieties remains integral.
Exercise 12.14. Show that if the conclusion of Theorem 12.19 holds for a field
k, then k is algebraically closed.

6. Regular Extensions
A field extension K/F is regular if for all field extensions L/F , the F -algebra
K ⊗F L is a domain.

As an important example, in this new termionlogy Theorem 12.10 says precisely

that purely transcendental extensions are regular.
Exercise 12.15. Show: If K/F is regular and algebraic, then K = F .
1The proof comes from https://stacks.math.columbia.edu/tag/020C.
 6. REGULAR EXTENSIONS 119

Lemma 12.20. If K/F is regular, then for all algebraic extensions L/F , we
have that K ⊗F L is a field.
Proof. We have by definition that K ⊗F L is a domain, so by Proposition
12.6 we have that K and L are somewhere linearly disjoint over F . Since L/F is
algebraic, Theorem 12.8 gives that K and L are everwhere linearly disjoint over F ,
and then Proposition 12.7 gives that K ⊗F L is a field. 

Lemma 12.21. Let K and L be commutative F -algebras. If K ⊗F L is a field,

then K and L are fields.
Proof. Tensoring F ,→ L with the (flat, since F is a field) F -module K gives
an injection K ,→ K ⊗F L. This shows that K is a domain. Let t ∈ K • , and let C
be the cokernel of multiplication by t on K. Tensoring the exact sequence
·t
0→K→K→C→0
with L, we get
·t⊗1
0 → K ⊗F L → K ⊗F L → C ⊗F L → 0.
Since K ⊗F L is a field, multiplication by t ⊗ 1 is an isomorphism, so C ⊗F L = 0,
so C = 0. This shows that K is a field. Interchanging K and L, the same argument
shows that L is a field. 

Theorem 12.22. For a field extension K/F , the following are equivalent:
(i) We have that K/F is regular.
(ii) For every algebraic extension L/F , we have that K ⊗F L is a field.
(iii) For any algebraic closure F of F , we have that K ⊗F F is a field.
(iv) We have that F is algebraically closed in K and K/F is separable.
Proof. (i) =⇒ (ii): This follows from Lemma 12.20.
(ii) =⇒ (iii) is immediate.
(iii) =⇒ (ii): Let L/F be an algebraic field extension. We have by assumption
that K ⊗F F = (K ⊗F L)⊗L F is a field, so by Lemma 12.21 we deduce that K ⊗F L
is a field.
(ii) =⇒ (iv): Let L be the algebraic closure of F in K, so by assumption we have
that K ⊗F L is a field. Thus the F -subalgebra L ⊗F L is a domain, and since L/F
is algebraic this implies L = F . The separability is automatic in characteristic 0
−1
and in characteristic p we have that K ⊗F F p is a field, so K/F is separable by
Theorem 12.15.
(iv) =⇒ (iii): Let F sep be the maximal separable subextension of F /F . Then
F sep /F is Galois and F sep ∩K = F , so K ⊗F F sep is a field. Since K/F is separable,
K ⊗F F sep /F sep is separable (indeed one can choose the same separating transcen-
dence basis), hence K is F -linearly disjoint from the maximal purely inseparable
extension of F sep , which is F .
(iii) =⇒ (i): We must show that K ⊗F L is a domain for every field extension
L/F . For this it is harmless to replace L with a larger field extension, so suppose
that L is algebraically closed, and let F be the algebraic closure of F in L (which is
indeed algebraically closed!). By assumption, we have that K ⊗F F is a field, and
since
K ⊗F L = f (K ⊗F F ) ⊗F L,
120 12. LINEAR DISJOINTNESS

we may replace F by F and K by K ⊗F F and thereby assume that F is algebraically

closed. In order to show that K ⊗F L is a domain it is enough to show that for any
finitely generated F -subalgebra R1 and any finitely generated R2 -subalgebra R2 of
L, the ring R1 ⊗F R2 is a domain. (This is essentially Exercise 12.4, and at any
rate it is not difficult: just write out an arbitrary element of the tensor product to
see that it has this form.) This is a nontrivial fact, but fortunately for us we have
just proven it: Theorem 12.19. 
Corollary 12.23. For a field F , the following are equivalent:
(i) Every field extension K/F is regular.
(ii) The field F is algebraically closed.
Exercise 12.16. Prove Corollary 12.23.
Exercise 12.17. Let F ⊂ K ⊂ L be a tower of field extensions.
a) Show: If K/F and L/K are regular, then so is L/F .
b) Show: if L/F is regular, then K/F is regular.
c) If L/F is regular, must L/K be regular?
 CHAPTER 13

Derivations and Differentials

1. Derivations
1.1. Definitions and First Results. Let R be a commutative ring, and let
M be an R-module. A derivation of R into M is a map D : R → M satisfying
both of the following:

(D1) For all x, y ∈ R, D(x + y) = D(x) + D(y)

(i.e., D is a homomorphism of additive groups),
(D2) For all x, y ∈ R, D(xy) = xD(y) + D(x)y (“Leibniz rule”).
Exercise 13.1. Let D : R → M be a derivation, let x ∈ R and let n ∈ Z+ .
Show that D(xn ) = nxn−1 D(x).
Suppose we are given a subring k of R. Then a k-derivation is a derivation
D : R → M satisfying the additional property

(D3) For all x ∈ k, D(x) = 0.

We often have M = R and then we speak of derivations and k-derivations on

R. We denote the set of all k-derivations on R by Derk (R).
Exercise 13.2. a) Show that any k-derivation D : R → M is a k-linear
map: for all c ∈ k and x ∈ R, D(cx) = cD(x).
b) Show that Derk (R) is a k-submodule of Homk (R, R).
c) Show that Derk (R) in fact has the structure of an R-module: if D ∈
Derk (R) and α ∈ R, then αD ∈ Derk (R).
d) Show that if D1 , D2 ∈ Derk (R), D1 ◦ D2 need not be a derivation of R.
e) Show that if D1 , D2 ∈ Derk (R), then the map [D1 , D2 ] : R → R defined
by [D1 , D2 ] : x 7→ D1 (D2 (x)) − D2 (D1 ((x)) is a k-derivation of R.
f) Suppose that k is a field of characteristic p > 0. Show that for any D ∈
Derk (R), the p-fold composition D◦p is a k-derivation on R.
Exercise 13.3. Let D : R → M be a derivation, and let C = {x ∈ R | D(x) =
0} be its kernel. Show that C is a subring of R and is in fact the unique maximal
subring of k of R such that D is a k-derivation. (It is sometimes called the constant
subring of R.)
Example 13.1. Let k be a field and R = k[t]. The usual polynomial derivative
f 7→ f 0 is a k-derivation on R; we will denote it by ∂. The derivation ∂ is the
unique k-derivation D such that D(t) = 0.
Exercise 13.4. Compute the constant subring of ∂ : k[t] → k[t]. Note that the
answer in positive characteristic is very different from characteristic zero!
121
122 13. DERIVATIONS AND DIFFERENTIALS

Exercise 13.5. Let k be a domain, let n ∈ Z+ , and let R = k[t1 , . . . , tn ] be

the polynomial ring in n variables over k. Show that for each 1 ≤ i ≤ n there is a
unique k-derivation ∂i on R such that ∂i (tj ) = δij .
At least when k = R, it is well known that we may differentiate not only polynomials
but also rational functions. This generalizes nicely to our abstract algebraic context.
Theorem 13.2. Let R be a domain with fraction field K, and let D ∈ Der(R).
a) There is a unique extension of D to a derivation on K, given by
 
x yD(x) − xD(y)
(15) DK = .
y y2
b) If D is a k-derivation for some subring k of R with fraction field f (k),
then DK is an f (k)-derivation.
Proof. a) Our first order of business is to show that DK is well-defined, i.e.,
if x1 , x2 , y1 , y2 ∈ R are such that y1 y2 6= 0 and x1 y2 = x2 y1 , then
y1 D(x1 ) − x1 D(y1 ) y2 D(x2 ) − x2 D(y2 )
2 = .
y1 y22
We check this by a straightforward if somewhat unenlightening calculation:
y22 (x1 D(y1 ) − y1 D(x1 )) − y12 (x2 D(y2 ) − y2 D(x2 )

= y22 x1 D(y1 ) − y22 y1 D(x1 ) − y12 x2 D(y2 ) − y12 y2 D(x2 )

= (y2 x1 d(y1 y2 ) − y1 y2 D(x1 y2 )) − (y1 x2 D(y1 y2 ) + y1 y2 D(y1 x2 ))
= (x1 y2 − x2 y1 )D(y1 y2 ) − y1 y2 D(x1 y2 − x2 y1 ) = 0.
Next we check that DK is a derivation:
   
x1 x2 y1 D(x1 ) − x1 D(y1 ) y2 D(x2 ) − x2 D(y2 )
DK + DK = +
y1 y2 y12 y22
y1 y22 D(x1 ) + y12 y2 D(x2 ) − x1 y22 D(y1 ) − x2 y12 D(y2 )
= .
y12 y22
On the other hand we have
   
x1 x2 x1 y1 + x2 y1 N (x1 , x2 , y1 , y2 )
DK + = DK = ,
y1 y2 y1 y2 y12 y22
where
N (x1 , x2 , y1 , y2 ) = y1 y2 DK (x1 y2 + x2 y1 ) − x1 y2 DK (y1 y2 ) − x2 y1 DK (y1 y2 )
= y1 y22 D(x1 )+y1 y22 DK (x2 )+(x2 y1 y2 −x1 y22 −x2 y1 y2 )DK (y1 )+(x1 y1 y2 −x1 y1 y2 −x2 y12 )DK (y2 )
= y1 y22 DK (x1 ) + y12 y2 DK (x2 ) − x1 y22 DK (y1 ) − x2 y12 DK (y2 ),
establishing the additivity of DK . Similarly we have
 
x1 x2 y1 y2 DK (x1 x2 ) − x1 x2 DK (y1 y2 )
DK =
y1 y2 y12 y22
x2 y1 y2 DK (x1 ) + x1 y1 y2 DK (x2 ) − x1 x2 y2 DK (y1 ) − x1 x2 y1 DK (y2 )
= ,
y12 y22
while    
x1 x2 x1 x2
DK + DK
y1 y2 y1 y2
 1. DERIVATIONS 123

x1 y1 (y2 DK (x2 ) − x2 DK (y2 )) + x2 y2 (y1 DK (x1 ) − x1 DK (y1 ))

=
y12 y22

x2 y1 y2 DK (x1 ) + x1 y2 y2 DK (x2 ) − x1 x2 y2 DK (y1 ) − x1 x2 y1 DK (y2 )

= ,
y12 y22
establishing the Leibniz rule.
Let D be any derivation on K extending D. For x, y ∈ K with y 6= 0, we have
   
x x x
D(x) = D · y = D(y) + yD ,
y y y
so
   
x D(x) xD(y) yD(x) − xD(y) x
D = − = = DK ,
y y y2 y2 y
completing the proof of part a).
b) Since DK extends D and D(x) = 0 for all x ∈ k, certainly DK (x) = 0 for all
x ∈ k. Using (15) it follows that for all x, y ∈ k with y 6= 0, D xy = 0. 

Proposition 13.3. Let L/K be a field extension, and let D ∈ DerK (L). Let
f ∈ K[t1 , . . . , tn ] and a = (a1 , . . . , an ) ∈ Ln . Then
n
X
D(f (a)) = ∂i f (a1 , . . . , an )D(ai ).
i=1

Exercise 13.6. Prove Proposition 13.3.

Let L/K be a field extension and S ⊂ L. Any derivation D on L restricts to a

function DS : S → L. We say that D is S-finite if {x ∈ S | D(x) 6= 0} is finite. Of
course S-finiteness is automatic if S itself is a finite set. The S-finite derivations
form an L-subspace of DerK (L) which we will denote by DerSK (L).

Proposition 13.4. Let L/K be a field extension, and let S ⊂ L be such that
L = K(S).
a) We have
dimL DerSK (L) ≤ #S.
b) In particular if L can be generated as a field extension by n < ∞ elements, then
dimL DerK (L) ≤ n.

Proof. Let L(S) be the set of all finitely nonzero functions from S to L. This
is an L-vector space with basis canonically in bijection with S: indeed, for s ∈ S,
let δs be the function which takes the value 1 at s and zero elsewhere. Then {δs }s∈S
is an L-basis for L(S) .
The natural restriction map DerSK (L) → L(S) is L-linear and injective. The L-
linearity is a triviality: the injectivity follows from the fact that every element
of L is a rational function in the elements of S with coefficients in K. Since
dim L(S) = #S, part a) follows immediately! Part b) is also immediate from the
observation that S-finiteness is a vacuous condition when S itself is a finite set. 
124 13. DERIVATIONS AND DIFFERENTIALS

1.2. The Derivation Extension Theorem. Let L = K(x1 , . . . , xn )/K be

a finitely generated field extension. Let I(x1 , . . . , xn ) be the ideal of K[t1 , . . . , tn ]
consisting of all polynomials f such that f (x1 , . . . , xn ) = 0. Thus I(x1 , . . . , xn ) is
the kernel of the natural surjective K-algebra map
K[t1 , . . . , tn ] → K[x1 , . . . , xn ], ti 7→ xi ,
so I(x1 , . . . , xn ) is a prime ideal.
Exercise 13.7. Let K be a field, and let D ∈ Der(K).
a) For a polynomial f ∈ K[t1 , . . . , tn ], we denote by f D the polynomial ob-
tained from f by applying D to each coefficient:
X X
f= aI ti11 · · · tinn 7→ f D = D(aI )ti11 · · · tinn .
I i
Show: this defines an extension of D to a derivation on K[t1 , . . . , tn ].
b) Show: there is a unique derivation D on K(t1 , . . . , tn ) such that D|K = D
and D(ti ) = 0 for all 1 ≤ i ≤ n. The restriction of D to kK[t1 , . . . , tn ] is
the derivation of part a).
Suppose D is a derivation on K and D is a derivation on K(x1 , . . . , xn ) extending
D. Then for all f ∈ I(x1 , . . . , xn ) we have
X
0 = D(f (x1 , . . . , xn )) = D( aI xi11 · · · xinn ) =
I
n
!
ik−1 ik+1
X X X
D(ai )xi11 · · · xinn + ai xi11 · · · xk−1 xk+1 · · · xinn ik xkik −1 D(xk )
I I k=1
n
X
= f D (x) + (∂i f )(x)D(xi ).
i=1
Thus each f ∈ I(x1 , . . . , xn ) gives a linear equation satisfied by the values D(x1 ), . . . , D(xn ) ∈
K(x1 , . . . , xn ) . The next result – probably the most important one in this chap-
ter – says that these linear relations give the only constraints in extending D to a
derivation on K(x1 , . . . , xn ).
Theorem 13.5 (Derivation Extension Theorem). Let K be a field, let L =
K(x1 , . . . , xn ) be a finitely generated field extension, and let {fj }j∈J be a set of
generators for the ideal I(x1 , . . . , xn ) of K[t1 , . . . , tn ]. Let D be a derivation on K.
Let u1 , . . . , un ∈ L be such that for all j ∈ J, we have
n
X
(16) 0 = fjD (x1 , . . . , xn ) + (∂i fj )(x1 , . . . , xn )ui .
i=1

Then there is a unique derivation D on L that extends D and satisfies D(xi ) = ui

for all 1 ≤ i ≤ n.
Proof. We saw above that for every derivation D on L that extends K, the
equation (16) holds with ui = D(xi ) for all 1 ≤ i ≤ n, establishing the uniqueness.
Conversely, suppose (16) holds for all fj in a set of generators for I(x1 , . . . , xn ).
Then for all j1 , j2 ∈ J we have
n
X
(fj1 + fj2 )D (x1 , . . . , xn ) + (∂i (fj1 + fj2 )(x1 , . . . , xn )ui =
i=1
 1. DERIVATIONS 125

n
! n
!
X X
fjD1 (x1 , . . . , xn ) + D
(∂i fj1 (x1 , . . . , xn )ui + fj2 (x1 , . . . , xn ) + (∂i fj2 (x1 , . . . , xn )ui
i=1 i=1
= 0 + 0 = 0.
For j ∈ J and g ∈ K[t1 , . . . , tn ], we have
n
X
D
(gfj ) (x1 , . . . , xn ) + (∂i (gfj ))(x1 , . . . , xn )ui
i=1
n
!
X
= g(x1 , . . . , xn ) fjD (x1 , . . . , xn ) + (∂i fj )(x1 , . . . , xn )ui
i=1
n
!
X
D
+fj (x1 , . . . , xn ) g (x1 , . . . , xn ) + (∂i g)(x1 , . . . , xn )ui
i=1
n
!
X
D
= g(x1 , . . . , xn ) · 0 + 0 · g (x1 , . . . , xn ) + (∂i g)(x1 , . . . , xn )ui = 0.
i=1
This shows that (16) holds for all f ∈ I(x1 , . . . , xn ). Now we define D on K[x1 , . . . , xn ]
by putting, for g ∈ K[t1 , . . . , tn ],
n
X
D(g(x1 , . . . , xn )) := g D (x1 , . . . , xn ) + (∂i g)(x1 , . . . , xn )ui .
i=1

The point here is that the same element of K[x1 , . . . , xn ] can in general by expressed
as g(x1 , . . . , xn ) for several polynomials g ∈ K[t1 , . . . , tn ], but any two such polyno-
mials differ by an element of I(x1 , . . . , xn ), so our calculation just above has checked
that if g1 (x1 , . . . , xn ) = g2 (x1 , . . . , xn ) then D(g1 (x1 , . . . , xn )) = D(g2 (x1 , . . . , xn )).
The fact that D is a derivation now follows from the fact that g 7→ g D and g 7→ ∂i g
are derivations. Thus we have succeeded in extending D to a derivation D of
K[x1 , . . . , xn ]. By Theorem 13.2a), there is a unique extension of D to a derivation
on K(x1 , . . . , xn ). Finally, for 1 ≤ i ≤ n, taking g = ti gives
n
X
D(xi ) = D(g(x1 , . . . , xn )) = g D (x1 , . . . , xn ) + (∂i g)(x1 , . . . , xn )ui
i=1
= 0 + 0 · u1 + . . . + 0 · ui−1 + 1 · ui + 0 · ui+1 + . . . + 0 · un = ui . 
1.3. The structure of DerK (L) for a finitely generated extension L/K.
Let L/K be any finitely generated field extension. In this section we will apply
Theorem 13.5 to give a good description of the L-vector space DerK (L). In par-
ticular its dimension is an interesting invariant of the extension L/K and will be
computed. What we know so far is that, by Proposition 13.4, dimL DerK (L) is
bounded above by the minimal number of generators of L/K, and (as we are about
to nail down explicitly) it follows easily from Exercise 13.5 that equality holds if
L/K is purely transcendental. It turns ou that in characteristic 0 the dimension
is always equal to trdeg(L/K) whereas in positive characteristic separability issues
intervene to make things more interesting.

The easiest application of Theorem 13.5 is when L = K(t1 , . . . , tn ) is a ratio-

nal function field. In this case the ideal I(t1 , . . . , tn ) is the zero ideal, so in this
case Theorem 13.5 says that we may extend D ∈ Der(K) to a derivation D on
126 13. DERIVATIONS AND DIFFERENTIALS

K(t1 , . . . , tn ) by freely choosing the values of D(t1 ), . . . , D(tn ). We have met spe-
cial cases of this result before: Exercise 13.7 treats the case D(ti ) = 0 for all i and
Exercise 13.5 fixes 1 ≤ j ≤ n and treats the case D(ti ) = δ(i, j) (Kronecker δ).
This gives part a) of the following result, and part b) will follow easily.
Corollary 13.6. Let L = K(t1 , . . . , tn ) be a purely transcendental extension.
a) Let D ∈ Der K. For each 1 ≤ i ≤ n, there is a unique Di ∈ Der L that
extends D and such that for all 1 ≤ i ≤ n we have
(
1 i=j
Di (tj ) = .
0 otherwise

( each 1 ≤ i ≤ n there is a unique δi ∈ DerK (L) such that δi (tj ) =

b) For
1 i=j
. The set {δ1 , . . . , δn } is an L-basis for DerK (L), so
0 i 6= j

dimL DerK (K(t1 , . . . , tn )) = n.

Proof. a) This restates Exercise 13.5 and is immediate from Theorem 13.5.
b) Applying part a) with D = 0 ∈ Der K gives the δ1 , . . . , δn . Let S = {t1 , . . . , tn }.
Then the map DerK L → LS given by restricting any D ∈ DerK L to S is a bijection.
As seen in Proposition 13.4, this restriction map is L-linear, hence it is an L-vector
space isomorphism. 

Corollary 13.7. Let L/K be a separable algebraic extension.

a) Every derivation on K extends uniquely to a derivation on L.
b) We have DerK (L) = 0.
Proof. a) Step 1: suppose that [L : K] is finite. By the Primitive Element
Corollary (Corollary 7.3), we have L = K[x] for some x ∈ L. The minimal polyno-
mial f ∈ K[t] of x is a generator of the ideal I(x) of K[t]. By Theorem 13.5a), if DL
extends K to L, then we have 0 = f D (x) + f 0 (x)DL (x). Since K(x) is separable,
we have f 0 (x) = 0, and thus we must have
−f D (x)
(17) DL (x) = .
f 0 (x)
By Theorem 13.5b), there is a unique DL ∈ Der(L) extending D and satisfying
(17).
Step 2: Suppose that L/K is an infinite degree separable extension. It is therefore
the direct limit of its finite separable subextensions Lα . By Step 1, there exists a
unique Dα ∈ DerK (Lα ) extending D. Because of the uniqueness, it is automatic
that these derivations fit together to give a derivation DL on L: that is, for any
x ∈ L, we choose α such that x ∈ Lα and put DL (x) = DLα (x). If x ∈ Lα ∩ Lβ
then the uniqueness forces DLα (x) = DLα Lβ (x) = DLβ (x).
b) Let D ∈ DerK (L). Then D extends 0 ∈ Der(K), as does 0 ∈ Der L. By part a),
we must have D = 0. 

Exercise 13.8. Let F ⊂ K ⊂ L be a tower of fields with L/K separable

algebraic, and let M be a subextension of L/K. Show that for any F -derivation D
of L such that D(K) ⊂ K, we have also D(M ) ⊂ M .
 1. DERIVATIONS 127

Example 13.8. Let K have characteristic p > 0, and let L/K be a monogenic
purely inseparable extension, so there is a ∈ K, x ∈ L and n ∈ Z+ such that L =
n n
K(x) and xp = a. Let D ∈ Der(K). The ideal I(x) is generated by f (t) = tp − a.
n
We have f D (x) = D(1)tp − D(a) = −D(a), so by Theorem 13.5 there is a unique
extension of D to a derivation on L with D(x) = u for each u ∈ L such that
0 = f D (x) + f 0 (x)u = −D(a) + 0 · u = −D(a).
Thus: if D(a) 6= 0, there is no extension of D to L. If D(a) = 0, then such
extensions correspond to elements of L. In particular, we have dimL DerK L = 1:
a basis is given by {Dx }, where Dx is the unique K-derivation with Dx (x) = 1.
Exercise 13.9. Let L = K(x) be a monogenic inseparable algebraic extension.
Show that once again dimL DerK L = 1.
Exercise 13.10. Let k be a field of characteristic p > 0, let x1 , . . . , xn be inde-
pendent indeterminates over k, let L := k(x1 , . . . , xn ) and let K := k(xp1 , . . . , xpn ).
a) Show: [L : K] = pn .
(Suggestion: use the tower k(xp1 , . . . , xpn ) ( k(x1 , xp2 , . . . , xpn ) ( . . . (
k(x1 , . . . , xn ).)
b) Let I = I(x1 , . . . , xn ) be the ideal of all f ∈ K[t1 , . . . , tn ] such that
f (x1 , . . . , xn ) = 0. Show that K[t1 , . . . , tn ]/I is isomorphic to L and thus
#K[t1 , . . . , tn ]/I = pn .
c) Let J be the ideal of K[t1 , . . . , tn ] generated by tp1 − xp1 , . . . , tpn − Lxpn . Show
that # dim K[t1 , . . . , tn ]/J ≤ pn . Deduce that I = J.
d) Let S = {x1 , . . . , xn }. Use Theorem 13.5 to show that the restriction-to-S
map r : DerK (L) → LS is an L-vector space isomorphism.
e) Deduce that the minimal number of generators for the degree pn field ex-
tension L/K is n.
f) Let s1 , . . . , sm be independent indeterminates over L, and put
M := L(s1 , . . . , sm ) = K(x1 , . . . , xn , s1 , . . . , sm ).
Let T := {x1 , . . . , xn , s1 , . . . , sm }. Use Theorem 13.5 to show that the
restriction-to-T map r : DerK (M ) → T M is an M -vector space isomor-
phism. Deduce the that the minimal number of generators of the field
extension M/K is m + n.
Taking m = 1 in Exercise 13.10 answers a question I raised in my Fall 2020 course
on algebraic function fields: whereas any separable finitely generated field extension
of transcendence degree 1 can be generated by 2 elements, for all n ∈ Z+ there is
a finitely generated field extension M/K of transcendence degree 1 in which the
minimal number of generators is 1. As of now I still wonder whether such examples
exist with K algebraically closed in M .
Lemma 13.9. Let K be a field of characteristic p > 0, and let L be a field with
K ⊂ L ⊂ K 1/p . If S is a set of generators for L/K and D ∈ Der(K) is such that
D(xp ) = 0 for all x ∈ S, then D can be extended to a derivation on L.
Proof. Consider the set S of pairs (M, DM ) such that M is a field with
K ⊂ M ⊂ L and DM ∈ Der(M ) is such that DM |K = D. On S we define the
relation (M, DM ) ≺ (M 0 , DM 0 ) if M ⊂ M 0 and DM 0 |M = DM . This makes S into
a partially ordered set in which the union over any chain is an upper bound, so by
128 13. DERIVATIONS AND DIFFERENTIALS

Zorn’s Lemma the set S has a maximal element (M, DM ). If M 6= L, then there
is x ∈ S \ M . Then xp ∈ K and DM (xp ) = D(xp ) = 0, so by Example 13.8 we
may extend DM to a derivation DM (x) on M (x), contradicting the maximality of
(M, DM ). Thus M = L and DM is the desired extension of D to L. 
Corollary 13.10. Let K be a field of characteristic p > 0, and let L/K be a
finite degree extension such that K ⊂ L ⊂ K 1/p . Put [L : K] = ps . Then:
a) We have dimL DerK (L) = s.
b) There are elements x1 , . . . , xs ∈ L and D1 , . . . , Ds ∈ DerK (L) such that
L = K(x1 , . . . , xs ), for all 1 ≤ i ≤ n we have xi ∈ / K(x1 , . . . , xi−1 ) and
(
1 i=j
∀1 ≤ i, j ≤ s, Di (xj ) = .
0 i 6= j
Proof. Of course the result holds trivially if L = K, so we may assume
that K ( L and choose x1 ∈ L \ K. Then xp1 ∈ K, so [K(x1 ) : K] = p.
If K(x1 ) ( L then we may choose x2 ∈ L \ K(x1 ) and once again we have
xp2 ∈ K ⊂ K(x1 ), so [K(x1 , x2 ) : K(x1 )] = p and thus [K(x1 , x2 ) : K] = p2 .
Since [L : K] = ps , this process terminates precisely after we have chosen x1 , . . . , xs
such that [K(x1 , . . . , xi ) : K(x1 , . . . , xi−1 )] = p for all 1 ≤ i ≤ s.
It follows from Example 13.12 that for any K ⊂ M ⊂ L, if D ∈ DerK (M )
and xi ∈ / M , then for all α ∈ M (xi ) there is a unique extension of D to a
derivation on M (xi ) such that D(xi ) = α. Fix 1 ≤ j ≤ s. Applying this
observation with M = K(x1 , . . . , xj−1 ) and D = 0, we get a unique derivation
Dj0 ∈ DerK(x1 ,...,xj−1 ) (K(x1 , . . . , xj )) with Dj0 (xj ) = 1 and then applying the ob-
servation again several more times we get a unique extension of Dj0 to Dj ∈ Der(L)
such that Dj (xj+1 ) = . . . = Dj (xs ) = 0. These elements D1 , . . . , Ds ∈ DerK (L)
indeed have the desired property that Di (xj ) = δ(i, j) (Kronecker delta) for all
1 ≤ i, j ≤ s, which establishes part b).
Upon restriction to S = {x1 , . . . , xj }, the derivations D1 , . . . , Ds yield the stan-
dard basis for the L-vector space LS of all functions from S to L, so they are cer-
tainly linearly dependent as functions from L to L, which shows that dimL DerK (L) ≥
s. Because L = K(x1 , . . . , xs ) we have dimL DerK (L) ≤ s, and thus we have
dimL DerK (L) = s, establishing part a). 
Proposition 13.11. For a finitely generated field extension L/K, the following
are equivalent:
(i) The extension L/K is separable algebraic.
(ii) We have DerK (L) = 0.
Proof. (i) =⇒ (ii): This is Corollary 13.7.
(ii) =⇒ (i): If L/K is not separably generated (equivalently, is not separable),
then there is a subextension F1 of L/K such that L/F1 is finite degree inseparable.
By decomposing L/F1 is purely inseparable over separable, one sees that there is
a subextension M of L/F1 such that L/M is purely inseparable monogenic. By
Example 13.12 we have
(0) ( DerM (L) ⊂ DerK (L).
Otherwise L/K is separably generated and transcendental; let x1 , . . . , xd be a sep-
arable transcendence basis, so L/K(x1 , . . . , xd ) is separable algebraic (in fact of
 1. DERIVATIONS 129

finite degree, though this is not needed). Then δ1 ∈ DerK (L(x1 , . . . , xn )) \ {0}, and
by Corollary 13.7 δ1 extends (uniquely) to a nonzero element of DerK (L). 
The following result is an interesting instance of derivations working for us rather
than we for them.
Corollary 13.12. Let L = K(x1 , . . . , xn ) be a finitely generated field etension.
Suppose there are polynomials f1 , . . . , fn ∈ I(x1 , . . . , xn ) ⊂ K[x1 , . . . , xn ] such that
the matrix (∂i fj )(x1 , . . . , xn ) ∈ Mn (L) is nonsingular. Then L/K is separable
algebraic.
Proof. Let D ∈ DerK (L). Since fj (x) = 0, we have
n
X n
X
0 = D(fj (x)) = f D (x) + (∂i fj )(x)D(xi ) = (∂i fj )(x)D(xi ).
i=1 i=1

The assumed nonsingularity of the matrix therefore gives D(x1 ) = . . . = D(xn ) = 0,

so D = 0. By Proposition 13.11, we have that L/K is separable algebraic. 
Theorem 13.13. Let L/K be a finitely generated separable field extension.
a) If {x1 , . . . , xn } is a separating transcendence basis for L/K, then there
is an L-basis {Di }1≤i≤n for DerK (L) such that for all 1 ≤ i ≤ n, the
restriction of Di to K(x1 , . . . , xn ) is ∂i .
b) We have trdeg(L/K) = dimL DerK (L).
Proof. a) Let {x1 , . . . , xn } be separating transcendence basis for L/K, and
put M := K(x1 , . . . , xn ). Let δ1 , . . . , δn be the M -basis for DerK (M ) of Corollary
13.6. By Corollary 13.7, each δi extends uniquely to an element Di of DerK (L).
We claim that {D1 , . . . , Dn } is an L-basis for DerK (L).
P Di 6= 0), for some 2 ≤
If D1 , . . . , Dn were L-linearly dependent, then (since each
m ≤ n we would have elements ai ∈ L such that Dm = 1≤i<m ai Di . Evaluating
this equation at tm gives 1 = 0, a contradiction.
PnD ∈ DerK (L), and for 1 ≤ i ≤ n,Pput
Let
n
ai := D(ti ). Then the restriction of
D − i=1 ai Di to M is zero, hence D = i=1 ai Di by Corollary 13.7.
b) This is immediate from part a). 
Proposition 13.14. Let K be a field of characteristic p > 0. For L/K a
finitely generated field extension and x ∈ L, the following are equivalent:
(i) For all D ∈ DerK (L) we have D(x) = 0.
(ii) We have x ∈ KLp .
Proof. (i) =⇒ (ii): We show the contrapositive: for x ∈ L \ KLp , we will
show that there is D ∈ DerK (L) such that D(x) 6= 0. The field extension L/KLp
is inseparable algebraic and finitely generated, hence it is finite of some degree pa .
There is a tower of degree p (hence monogenic) purely inseparable field extensions
F0 = KLp ( F1 = KLp (x) ( F2 ( Fa = L.
By Example 13.8 there is D ∈ DerKLp (KLp (x)) such that D(x) = 1. Inductively,
having extended D to Fi , we may write Fi+1 = Fi (xi+1 ), so xpi+1 ∈ Fi . But also
xpi+1 ∈ KLp , so D(xpi+1 ) = 0. By Example 13.8 again we may extend D to Fi+1 .
Eventually we extend D to Fa = L.
(ii) =⇒ (i): Let D ∈ DerK (L). Then the kernel of the K-linear map D is a subfield
130 13. DERIVATIONS AND DIFFERENTIALS

of L that contains K (by assumption) and Lp (since D(xp ) = pD(x)p−1 = 0 in

characteristic p), so the kernel contains KLp . 

Finally we come to the general computation of dimL DerK (L) that was promised.

Theorem 13.15. Let L = K(x1 , . . . , xn ) be a finitely generated field extension,

and put
d := dimL DerK (L).
a) Let t be the minimal cardinality of a set of elements y1 , . . . , ym such that
L/K(y1 , . . . , ym ) is separable algebraic. Then d = t.
b) If K has characteristic p > 0 then we have

pd = [L : KLp ].

c) The extension L/K is separably generated iff d = trdeg(L/K).

Proof. a) If L/K(y1 , . . . , ym ) is separable algebraic, then D ∈ DerK (L) is de-

termined by its values on y1 , . . . , ym , which shows that d ≤ t. To complete the proof
of part a) we must find elements y1 , . . . , yd such that L/K(y1 , . . . , yd ) is separable
algebraic. If K has characteristic 0, then L/K is separable and by Theorem 13.13b)
we have d = trdeg(L/K), so we may take y1 , . . . , yd to be any transcendence basis
for L/K. So suppose K has characteristic p > 0. We have DerK (L) = DerKLp (L),
so d = dimL DerKLp (L). The extension L/KLp is finitely generated and L ⊂
(KLp )1/p , so Corollary 13.10 applies to show that there are elements y1 , . . . , yd of
L and D1 , . . . , Dd ∈ DerK (L) such that Di (yj ) = δ(i, j) (Kronecker delta). Then
Pd
{D1 , . . . , Dd } is an L-basis for DerK (L), so if D = i=1 αi Di ∈ DerK (L) vanishes
on y1 , . . . , yd , then D = 0. This shows that DerK(y1 ,...,yd ) (L) = 0, so by Proposition
13.11 we get that L/K(y1 , . . . , yd ) is separable algebraic.
b) In the proof of part a), we saw that in characteristic p > 0 we have d =
dimL DerKLp (L), so by Corollary 13.10 we have [L : KLp ] = pdimL DerKLp (L) = pd .
c) Again, if L/K is separably generated, then Theorem 13.13b) gives d = trdeg(L/K).
Conversely, if trdeg(L/K) = m and there are elements y1 , . . . , ym ∈ L such that
L/K(y1 , . . . , ym ) is separable algebraic then y1 , . . . , ym must be algebraically in-
dependent over K hence yield a separating transcendence basis, and thus L/K is
separably generated. 

Corollary 13.16. Let K be a perfect field of characteristic p > 0, and let

L/K be finitely generated of transcendence degree d. Then for all n ∈ Z+ we have
n
[L : Lp ] = pdn .

Proof. Step 1: First suppose that n = 1. Since K is perfect, we have that

L/K is separably generated and KLp = K p Lp = Lp . By parts b) and c) of Theorem
13.15 we get pd = [L : Lp ].
Step 2: Applying Step 1 to each extension in the tower
n n−1
Lp ⊂ Lp ⊂ . . . ⊂ Lp ⊂ L

yields the general case. 

 1. DERIVATIONS 131

1.4. Derivations in Infinitely Generated Field Extensions.

Proposition 13.17. Let K be a field, S a set and L = K({ts }s∈S be the
rational function field over K.
a) For s ∈ S, there is a unique K-derivation δs of L such that δs (ts ) = 1
and δs (ts0 ) = 0 for all s0 6= s.
b) The set {δs }s∈S is an L-basis for DerSK (L).
Proof. a) For each finite subset W of S that contains s, there is a unique
K-derivation DW on KW := K({ts | s ∈ W }) such that DW (ts ) = 1 and for each
w ∈ W \ {s} we have DW (tw ) = 0. When W1 ⊂ W2 , we have KW1 ⊂ KW2 and
(DW2 )|K(W1 ) = DW1 . Since L is the direct limit of the subfields KW , it follows
that there is a unique derivation D on L that extends this compatible family of
derivations on the subfields. This derivation satisfies the required properties of δs ,
and it is unique because L is generated by K and the elements ti for i ∈ S.
b) In Proposition 13.4, we saw that the natural map r : DerSK (L) → LS obtained
by restricting each derivation to S was injective and L-linear. Under this map r,
the derivations δs map to an L-basis for LS . It follows that r is an L-vector space
isomorphism and thus {δs }s∈S is an L-basis. 
Exercise 13.11. Let K be a field, S an infinite set, and let L = K({ts | s ∈ S})
be the rational function field in a set of indeterminates parameterized by S. Let LS
be the L-vector space of all functions from S to L. Show that restriction to S gives
an L-isomorphism
∼
DerK (L) → LS .
Deduce that dimL DerK L > #S.
Theorem 13.18. For a field extension K/F , the following are equivalent:
(i) The extension K/F is separable.
(ii) Every derivation on F extends to a derivation on K.
Proof. (i) =⇒ (ii): First suppose that F has characteristic 0, so there is a
separating transcendence basis {xi }i∈I for K/F . If D ∈ Der F , then for any finite
subset J ⊂ I, by Corollary 13.6a) there is a unique extension of D to F ({xi | i ∈ J})
such that D(xi ) = 0 for all i ∈ J. These derivations piece together to give a unique
extension of D to F ({xi | i ∈ I}) such that D(xi ) = 0 for all i ∈ I. By Corollary
13.7, since K/F ({xi | i ∈ I}) is separable algebraic, D extends uniquely to a
derivation on K.1
Now suppose that F has characteristic p > 0. Since K/F is separable, the
fields F 1/p and K are linearly disjoint over F ; applying the pth power map to all
these fields, we get that F and K p are linearly disjoint over F p . Let {ui }i∈I be
a basis for K p as an F p -vector space, one of whose elements is 1. Then there are
elements γi,j,k ∈ F p such that
X
∀i, j ∈ I, ui uj = γi,j,k uk .
k

By linear disjointness, {ui }i∈I is an F -linearly independent subset of K, and its

F -span is the ring F [K p ].

1As should be clear, this argument is valid whenever K/F is separably generated.
132 13. DERIVATIONS AND DIFFERENTIALS

p
P D ∈ Der(F ). Every element x ∈ F [K ] has a unique expression as a finite
Let
sum i∈I xi ui with xi ∈ F . We put
X
D(x) := D(xi )ui .
i∈I
p
P
The map D : F [K ] → K is clearly additive. Moreover, for x = i∈I xi ui , y =
p p
P
i∈I yi ui ∈ F [K ], then since D(γi,j,k ) = 0 for all γi,j,k ∈ F we have
X
D(xy) = D( xi yj γi,j,k uk )
i,j,k
X
= (D(xi )yj + si D(yj ))γi,jk )uk = xD(y) + D(x)y.
i,j,k

Thus D is a derivation on F [K p ], and it extends D since one of the Pui ’s is 1. By

Theorem 13.2 the derivation D extends uniquely to F K p . If x = i xi ui ∈ K p ,
then each xi lies in F p so D(xi ) = 0 and thus D(x) = i D(xi )ui = 0, so D ∈
P

DerK p (F K p ). Lemma 13.9 now implies that D extends to a derivation on K.

(ii) =⇒ (i): Suppose that every derivation on F extends to a derivation on
K. We may of course assume that F and K have characteristic p > 0, for in
characteristic 0 every field extension is separable. By Exercise 12.11 it suffices to
show that if S is an F -linearly independent subset of K, then S p = {xp | x ∈ S}
is also F -linearly independent. Assuming not, let n be the minimal cardinality of
an F -linearly dependent subset of S p . Then there are elements x1 , . . . , xn ∈ S and
a2 , . . . , an ∈ F × such that
(18) xp1 + . . . + an xpn = 0.
Let D ∈ Der(F ) and extend it to a derivation on K, which we continue to denote
by D. Since for all x ∈ K we have D(xp ) = 0, applying D to (18) gives
0 = D(1)xp1 + D(a2 )xp2 + . . . + D(an )xpn = D(a2 )xp2 + . . . + D(an )xpn .
By the minimality of n we have D(a2 ) = . . . = D(an ) = 0. Applying Propo-
sition 13.14 with K = Fp and L = F , we get that a2 , . . . , an ∈ F p , i.e., there
are b2 , . . . , bn ∈ F × such that ai = bpi for all 2 ≤ n, and then (18) implies that
x1 +b2 x2 +. . .+bn xn = 0, contradicting the F -linear independence of x1 , . . . , xn . 
Exercise 13.12. In this exercise we will determine the fields F for which
Der(F ) = 0.
a) Suppose that F has characteristic 0. Show: Der(F ) = 0 iff F/Q is alge-
braic.
b) Suppose that F has characteristic p > 0. Show: Der(F ) = 0 iff F is
perfect.
The next two exercises explore whether Proposition 13.11 extends to infinitely gen-
erated field extensions.
Exercise 13.13. Let K be a field of characteristic 0, and let L/K be an ex-
tension such that DerK (L) = 0. Show: L/K is algebraic.
Exercise 13.14. Let p be a prime number.
a) Find a field K of characteristic p and an inseparable algebraic field exten-
sion L/K such that DerK (L) = 0.
 2. KÄHLER DIFFERENTIALS 133

b) Let κ ≥ 1 be a cardinal number. Find a field K of characteristic p and a

separable extension L/K of transcendence degree κ such that DerK (L) =
0.

2. Kähler Differentials
Let B ⊂ A be an extension of commutative rings. We define the module of
Kähler differentials ΩA/B to be the quotient of the free A-module à on the set
{da | a ∈ A} (here we understand that for each a ∈ A we have a formal symbol da,
such that the map a 7→ da is injective) via the following submodule R of relations:
∀a, b ∈ A, d(a + b) − da − db,
∀a, b ∈ A, d(ab) − (adb + bda),
∀α ∈ B, dα.
The effect of this is that we get a map
d : A → ΩA/B , a 7→ da.
It is easy to check that d is a B-derivation. Moreover it is universal among all
B-derivations into an A-moduule M in the following sense.
Proposition 13.19. If M is an A-module and D : A → M is a B-derivation,
then there is a unique A-module homomorphism f : ΩA/B → M such that D = f ◦d.
Proof. There is a unique A-module homomorphism f˜ : Ã → M such that for
all a ∈ A, f˜(da) = D(a). For all a, b ∈ A, we have
f˜(d(a+b)−d(a)−d(b)) = f˜(d(a+b))−f˜(d(a))−f˜(d(b)) = D(a+b)−D(a)−D(b) = 0,
f˜(d(ab) − adb − bda) = f˜(d(ab)) − af˜(db) − bf˜(da) = D(ab) − aD(b) − bD(a) = 0,
and for all α ∈ B we have
f˜(dα) = D(α) = 0.
Thus f˜ factors through f : ΩA/B → M and has the property that for all a ∈ A,
f (da) = D(a), so D = f ◦ d. Conversely, any such f satisfies f (da) = D(a) for all
a ∈ A, and since {da}a∈A generate ΩA/B as an A-module, the map is unique. 
Exercise 13.15. Let B ⊂ A be a ring extension.
(i) If S is a set of generators for B as an A-algebra (that is, every element
of B is an A-linear combination of products of elements of S), then {ds |
s ∈ S} is a set of generators for ΩA/B as an A-module.
(ii) Suppose that A and B are fields and A = B(S). Show that the elements
{ds | s ∈ S} span ΩA/B as an A-vector sapce.
We can restate Proposition 13.19 as giving a natural A-module isomorphism
DerB (A, M ) = HomA (ΩA/B , M ).
In particular, taking M = A, we get
(19) DerB (A) = HomA (ΩA/B , A) = Ω∨
A/B .

That is, the B-derivations of A are the A-linear functionals on ΩA/B . In particular:
Corollary 13.20. a) Suppose ΩA/B is finitely generated and free as an
A-module. Then DerB (A) is finitely generated and free as an A-module
and rank DerB (A) = rank ΩA/B .
134 13. DERIVATIONS AND DIFFERENTIALS

b) Let L/K be a finitely generated field extension. Then

dimL ΩL/K = dimL DerK (L) < ∞.
Proof. a) This just uses the fact that the dual of a finitely generated free
module is a finitely generated free module of the same rank.
b) If ΩL/K were infinite-dimensional, then so would be its dual space and thus
DerK (L) would be infinite-dimensional, but the number of generators of L/K is an
upper bound on dimL DerK (L). So ΩL/K is a finite-dimensional L-vector space,
and we may now apply part a). 
We say a ring extension B ⊂ A is omega-finite if ΩA/B is a finitely generated
A-module and omega-free if ΩA/B is a free A-module. Evidently omega-freeness
automatic when A is a field. For a field extension L/K, we saw above that omega-
finiteness holds iff DerK (L) is a finite-dimensional L-vector space, hence it holds if
L/K is either finitely generated or separably generated and of finite transcendence
degree. In this case, we have
ι
ΩL/K → Ω∨∨ ∨
L/K = DerK (L) ,

where ι is the natural isomorphism of a finite-dimensional vector space with its

second dual space given
Pby evaluating linear functionals on V at points of V . Under
n
this isomorphism, for i=1 ai dxi ∈ ΩL/K and D ∈ DerK (L) we have
Xn n
X
( ai dxi )(D) = ai D(xi ).
i=1 i=1

Theorem 13.21. Let L/K be a finitely generated field extension, and let x1 , . . . , xn ∈
L.
a) If {x1 , . . . , xn } is a separating transcendence basis for L/K, then {dx1 , . . . , dxn }
is an L-basis for ΩL/K .
b) Suppose that L/K is separably generated. If {dx1 , . . . , dxn } is an L-basis
for ΩL/K , then {x1 , . . . , xn } is a separating transcendence basis for L/K.
Proof. a) If x1 , . . . , xn is a separating transcendence basis for L/K, then for
all 1 ≤ i ≤ n there is a unique Di ∈ DerK (L) such that Di (xj ) = δ(i, j) and
D1 , . . . , Dn is an L-basis of DerK (L). Since DerK (L) = Ω∨ L/K , this basis is the dual
basis of a unique L-basis b1 , . . . , bn of ΩL/K : that is, for all 1 ≤ i, j ≤ n, we have
Di (bj ) = δ(i, j). We have identified Di with an L-linear functional `i on ΩL/K :
this functional is the unique one such that for all x ∈ L we have λi (dx) = Di (x).
Thus for all 1 ≤ i, j, ≤ n we have λi (dxj ) = Di (xj ) = δ(i, j), and it follows that we
have b1 = dx1 , . . . , bn = dxn .
b) Suppose dx1 , . . . , dxn is an L-basis for ΩL/K = DerK (L)∨ . First of all this
gives that dim DerK (L) = n, and since L/K is finitely generated and separable, by
Theorem 13.13 we get that trdeg(L/K) = n.
Now, let D ∈ DerK (L) be such that for all 1 ≤ i ≤ n we have 0 = dxi (D) =
D(xi ). Since the dxi ’s span DerK (L)∨ this implies D = 0, which shows that
DerK (L) = DerK(x1 ,...,xn ) (L). Using Proposition 13.11 we get that L/K(x1 , . . . , xn )
is separable algebraic, so x1 , . . . , xn is a separating transcendence basis for L/K. 
Exercise 13.16. Show that Theorem 13.21 holds without the hypothesis that
L/K is finitely generated.
 3. APPLICATIONS TO ONE VARIABLE FUNCTION FIELDS 135

Example 13.22. ([Mo96, Ex. 23.18]) Let k be a field of characteristic p > 0,

let L = k(x, y) be a rational function field in two variables, and let K = k(xp , y p ).
By Exercise 13.10 we have dimL DerK (L) = 2, so by Corollary 13.21 we have
dimL ΩL/K = 2. We claim that {dx, dy} is an L-basis for ΩL/K . This will show
that in Theorem 13.21b) if we remove the hypothesis that L/K is separably generated
is needed, then {x, y} need not even be a transcendence basis for L/K: indeed in
this case L/K is algebraic!
Because dimL ΩL/K = 2, it is enough to show that dx and dy are L-linearly
independent. Suppose that there are a, b ∈ L such that adx + bdy = 0. Let δ1 , δ2 ∈
DerK L be the usual K-derivations. Since in characteristic p every derivation kills
every P th power, we have Derk L = DerK L. Let `1 , `2 ∈ Ω∨ L/K be the linear
functionals corresponding to δ1 and δ2 , so δi (df ) = ∂i (f ). Applying `1 we get
0 = `1 (0) = `1 (adx + bdy) = a`1 (dx) + b`1 (dy) = aδ1 (x) + bδ1 (y) = a,
and similarly applying `2 we get b = 0.

3. Applications to One Variable Function Fields

Corollary 13.23. Let K be a field of characteristic p > 0, and let L/K be
finitely generated and separable, of transcendence degree 1.
a) For x ∈ L, the following are equivalent:
(i) We have that x is a separating element for L/K – i.e., x is a
separating transcendence basis for L/K.
(ii) We have dx 6= 0.
(iii) We have that x ∈/ KLp .
/ Lp .
If K is perfect, the conditions are also equivalent to x ∈
b) For each separating element x of L/K, there is a unique derivation δx ∈
DerK (L) such that δx (x) = 1.
c) For elements x, y of L/K with y separating, we have
δy = δy (x)δx .
d) For y ∈ K, we have δx (y) 6= 0 iff y is a separating element of L/K.
Proof. a) Theorem 13.21 shows (i) ⇐⇒ (ii). Since ΩL/K is the L-dual of
the one-dimensional L-vector space DerK (L), for any x ∈ L we have dx = 0 iff
D(x) = 0 for all D ∈ DerK (L) iff x ∈ KLp by Proposition 13.14, showing (ii) ⇐⇒
(iii).
b) Theorem 13.13a) supplies δx ∈ DerK (L) such that δx (x) = 1, while Theorem
13.13b) shows that dimL DerK (L) = 1, which implies that this derivation is unique.
c) Since δy is nonzero in the one-dimensional vector space DerK (L), there is a unique
α ∈ L× such that δy = αδx . Evaluating at x gives α = δy (x).
d) If y is a separating element of L/K then it follows from part c) that δx (y) =
1
δy (x) 6= 0. If y is not separating, then by part a) we have dy = 0 and thus

0 = dy(δx ) = δx (y). 
Exercise 13.17. Let K be a perfect field of chracteristic p > 0, and let L/K
be a finitely generated field extension.
a) Show: [L1/p : L] is finite.
b) Suppose that L/K has transcendence degree 1. Show: [L1/p : L] = p.
136 13. DERIVATIONS AND DIFFERENTIALS

Corollary 13.24. Let k be a perfect field of characteristic p > 0, let K/k be

finitely generated of transcendence degree 1, and let L/K be a finite degree insepa-
rable field extension. Then L ⊃ K 1/p .
Proof. Let Ks be the maximal separable subextension of L/K. Then all the
1/p
hypotheses apply to L/Ks in place of L/K, and if we can show that L ⊃ Ks
1/p
then that suffices, since Ks ⊃ K 1/p . Because of this we may assume that L/K
is a nontrivial purely inseparable extension, and thus there is α ∈ L \ K such that
x = αp ∈ K. We have
K ( K(α) ⊂ K 1/p ,
so by Exercise 13.17 we get
L ⊃ K(α) = K 1/p . 
Exercise 13.18. Let k be a perfect field of characteristic p > 0, let K/k be
finitely generated of transcendence degree 1, and let L/K be a finite degree extension.
Let Ks be the maximal subextension of L/K, so [L : Ks ] = pa for some a ∈ N.
−a
Show: L = K p .
We have developed our field theory to the point where it is essentially algebraic
geometry in disguise. For instance, Corollary 13.24 carries all the content of the
following fact: if f : C1 → C2 is a finite morphism of nice curves defined over a
a
perfect field k of characteristic p > 0, then it factors as a Frobenius map C1 → C1p
a
followed by a separable morphism C1p → C2 [Si, Cor. II.2.12]. (It also calls
attention to the fact that the ground field k should be perfect for this to hold, a
fact which is somewhat softly voiced in Silverman’s exposition.)

4. p-Bases
Throughout this section we work with fields of a fixed characteristic p > 0.
Lemma 13.25. Let F be a field of characteristic p, and let S ⊂ F . Let S be the
set of all F p -linear combinations of monomials xi11 · · · xinn with x1 , . . . , xn ∈ S and
0 ≤ i1 , . . . , in < p. Then S = F p (S).
Proof. We have F p (S) = lim F p (T ) as T ranges over finite subsets of S
−→
and S = lim T as T ranges over all finite subets of S. So it suffices to show
−→
that S = F p (S) for all finite subsets S ⊂ F . The inclusion S ⊂ F p (S) is clear.
Conversely, because S is finite and consists of elements algebraic over F p , we have
F p (S) = F p [S]: that is, each element of F p (S) is an F p -linear combination of
monomials xa1 1 · · · xann with x1 , . . . , xn ∈ S and a1 , . . . , an ∈ N. For 1 ≤ i ≤ n, write
ai = pbi + ri with 0 ≤ ri < p. Then
xa1 1 · · · xann = (xb11 · · · xbnn )p xr11 · · · xrnn = αxr11 · · · xrnn
with α ∈ F p . This completes the proof. 
A p-spanning subset is a subset S ⊂ F such that S = F . On the other hand:
Lemma 13.26. For a subset S ⊂ F , the following are equivalent:
(i) For every finite subset {s1 , . . . , sn } ⊂ S, the set of monomials si11 · · · sinn
with 0 ≤ i1 , . . . , in < p is F p -linearly independent.
(ii) For all s ∈ S, we have s ∈ / S \ {s}.
 4. p-BASES 137

A subset satisfying these equivalent conditions is called a p-independent subset.

Proof. We easily reduce to the case where S is finite. (i) =⇒ (ii): If
(ii) fails, then we may write S = {s1 , . . . , sn , x} such that x ∈ I si11 · · · sinn for
P
0 ≤ i1 , . . . , in < p. It is clear that this violates the condition in (i).
(ii) =⇒ (i): Let S = {s1 , . . . , sn }. A nontrivial F p -linear dependence relation
among the monomials si11 · · · sinn yields, after reordering the si ’s if necessary, a
nonzero polynomial f ∈ F p (s1 , . . . , sn−1 ) of degree less than p satisfied by sn .
This means that sn is both separable and purely insepable over F p (s1 , . . . , sn−1 ),
so sn ∈ F p (s1 , . . . , sn−1 ) = S \ {s}, a contradiction. 

Probably the reader suspects what is coming next: we claim that for subsets S of
F , the mapping S 7→ S = F p (S) ⊂ F is a spanning operator in the sense of §11.4.
The properties (SO1) through (SO4) hold immediately: in fact, for any subfield A
of a field F , the operator S ⊂ F 7→ A(S) satisfiies these properties. We now check
(SO5) (“Exchange Lemma”): for x, y ∈ F and S ⊂ F , if y ∈ S ∪ {x} \ S, we must
show that x ∈ S ∪ {y}.

If y ∈ S ∪ {x} then there are s1 , . . . , sn ∈ S and for all I = (i1 , . . . , in+1 ) ∈

{0, . . . , p − 1}n+1 an element αI ∈ F p such that
X
y= αI si11 · · · sinn xin+1 .
I

Because y ∈ / S, there is at least one I with in+1 6= 0 such that αI 6= 0 and this
shows that x satisfies a polynomial relation of degree less than p with coefficients
in F p (S, y). In other words, x is separable algebraic over F p (S, y), but it is also
purely inseparable over F p (S, y), so x ∈ F p (S, y) = S ∪ {y}.

A p-basis for F is a subset S ⊂ F that is both p-independent and p-spanning.

By the abstract theory developed in §11.4, we know: a subset S ⊂ F is a p-basis
iff it is a maximal p-indpendent subset iff it is a minimal p-spanning subset. More-
over, every p-independent subset is contained in a p-basis, every p-spanning usbset
contains a p-basis, and any two p-bases have the same cardinality, which we call
the p-dimension dimp F of F .
Corollary 13.27. Let F be a field of characteristic p > 0.
a) The following are equivalent:
(i) The extension F/F p has finite degree.
(ii) The field F has finite p-dimension.
b) When the equivalent conditions of part a) hold, we have [F : F p ] = pdimp F .
Proof. a) The extension F/F p has finite degree iff F has a finite p-spanning
sett iff F has a finite p-basis.
b) If S = {s1 , . . . , sn } is a p-basis then {si11 · · · sinn | 0 ≤ i1 , . . . , in < p} is an
F p -basis for F . 

Exercise 13.19. Let F be a field of characteristic p > 0 such that dimp (F ) is

infinite. Show: [F : F p ] = dimp (F ).
We can use derivations to give a characterization of p-independent subsets:
138 13. DERIVATIONS AND DIFFERENTIALS

Theorem 13.28. Let F be a field of characteristic p > 0, and let S be a subset

of F . The following are equivalent:
(i) The subset S is p-independent.
(ii) For all x(∈ S there is a Dx ∈ DerF p (F ) such that for all s ∈ S we have
1 s=x
Dx (s) = .
0 s 6= x

Proof. (i) =⇒ (ii): Suppose that S is p-independent, and let S be the set
of subsets T ⊂ S such that for all t ∈ T there is DT,t ∈ DerF p (F p (T )) such that
DT,t (t) = 1 and DT,t (t0 ) = 0 for all t0 ∈ST \ {t}, partially ordered under inclusion.
For any chain {Ti }i∈I in S, the union i∈I Ti is an upper bound: the derivation
DT,t is uniquely determined by its properties, S so if t ∈ T1 ⊂ T2 then the restriction
of DT2 ,t to F p (T1 ) is DT1 ,t . For any t ∈ i∈I Ti , the existence and uniquenessf of
the derivation DSi∈I Ti ,t follows easily from this. By Zorn’s Lemma the set S has
a maximal element T . If T ( S, choose s ∈ S \ T ; by the p-independence of S we
have s ∈ / F p (T ) and sp ∈ F p ⊂ F p (T ), so by Example 13.8, for all α ∈ F P (T ∪{s}),
each derivation D ∈ DerF p (F p (T )) admits a unique extension to F p (T ∪ {s}) such
that D(s) = α. For all t ∈ T , extending DT,t by D(s) = 0 yields the derivation
DS,t , while extending the 0 derivation on F p (T ) to F p (T ∪ {s}) by D(s) = 1 yields
the derivation DS,s . This shows that T ∪ {s} ∈ S, contradicting the maximality of
T . It follows that T = S.
¬ (i) =⇒ ¬ (ii): If the subset S is not p-independent, then there is s ∈ S such
that s ∈ F p (S \ {s}). It follows that every D ∈ DerF p (F ) such that D(t) = 0 for
all t 6= s also satisfies D(s) = 0. 
Corollary 13.29. Let F be a field of characteristic p > 0. For S ⊂ F , the
following are equivalent:
(i) The subset S is a p-basis for F .
(ii) The map s 7→ ds is injective and {ds | s ∈ S} is a basis for ΩF/F p .
Proof. (i) =⇒ (ii): Let S be a p-basis for F .
Since S is a p-spanning subset of F we have F = F p (S), so by Exercise 13.15
the set {ds | s ∈ S} spans ΩF/F p .
Since S is a p-independent subset, applying Theorem 13.28 gives a family of
derivations {Dx | x ∈ S} such that for all y ∈ S, we have
(
1 y=x
Dx (y) = .
0 y 6= x
Regarding DerF p (F ) as the F -vector space dual of ΩF/F p , we have
(
1 y=x
∀x, y ∈ S, Dx (dy) = Dx (y) = .
0 y 6= x
Thus if dx were an F -linear combination of elements dy with y ∈ S \{x}, evaluating
at Dx gives 1 = 0, a contradiction. So the dx’s are distinct elements and form an
F -linearly independent subset of ΩF/F p .
(ii) =⇒ (i): Suppose that x ∈ S 7→ dx is an injection and that {dx | x ∈ S}
is an F -basis for ΩF/F p . Because DerF/F p = Ω∨F/F p , there is a unique family of
derivations {Dx | x ∈ S} such that Dx ∈ DerF p (F ) and for all x, y ∈ S we have
 4. p-BASES 139

(
1 y=x
Dx (y) = . Theorem 13.28 implies that S is a p-independent subset. If it
0 y 6= x
were not a p-basis, there would be a strictly larger p-independent subset S̃ = S∪{s},
and Theorem 13.28 implies that there is Ds ∈ DerF p (F ) such that Ds (s) = 1 and
Ds (x) = 0 for all x ∈ S. But then on the one hand we would have Ds (ds) =
Ds (s) = 1, and on the other hand, since ds is an F -linear combination of dx for
x ∈ S, we would have Ds (ds) = 0, a contradiction. 
Corollary 13.29 is an example of how Kähler differentials behave better than deriva-
tions in non-omega finite extensions. Indeed, when ΩK/F is infinite-dimensional as
a K-vector space, its dual vector space DerF (K) is an infinite vector space of larger
dimension. If we choose a differential basis for K/F – i.e., a subset S of K such
that s 7→ ds ∈ ΩK/F is injective and {ds | s ∈ S} is a K-basis for ΩK/F – then there
is a “dual basis” {D
( s | s ∈ S} of F -derivations of K characterized by the usual
1 x=y
relation Dx (y) = for all x, y ∈ S. However this dual basis is a basis
0 x 6= y
not for DerF (K) but for the subspace DerSF (K) of S-finite derivations, a subspace
which depends strongly on the choice of S.
Exercise 13.20. An extension of characteristic p fields F ⊂ K is p-radical if
K ⊂ F 1/p : in other words, K is obtained by adjoining pth roots of elements of F .
For any p-adical extension K/F and any subset S ⊂ K, define S :== F (S).
a) Show that S is the F -span of monomials si11 · · · sinn with s1 , . . . , sn ∈ S and
0 ≤ i1 , . . . , in < p.
b) Extend the theory of p-spanning subsets, p-independent subsets and p-bases
to p-radical extensions K/F . Show in particular:
(i) The extension K/F has a p-basis – a subset S such that K = F (S)
and for all s ∈ S, s ∈ / F (S \ {s}) – and any two p-bases have the
same cardinality.
(ii) If S is a finite p-basis for K/F , then [K : F ] = p#S , while if S is an
infinite p-basis for K/F , then [K : F ] = #S.
(iii) A subset S ⊂ F is a p-basis iff s 7→ ds ∈ ΩK/F is injective and
{ds | s ∈ S} is an F -basis for ΩK/F .
(iv) For x ∈ F , we have dx = 0 ∈ ΩK/F iff x ∈ F .
 CHAPTER 14

Ordered Fields

1. Ordered Abelian Groups

An ordered abelian group (G, +, <) is an abelian group (G, +) equipped with a
total ordering < which is compatible with the group law in the sense that

(OAG) For all x, y, z ∈ G, x ≤ y =⇒ x + z ≤ y + z.

A homomorphism of ordered abelian groups f : (G, <) → (H, <) is a group homo-
morphism which is isotone: for all x1 ≤ x2 , f (x1 ) ≤ f (x2 ).
Lemma 14.1. For x, y, z in an ordered abelian group G, if x < y then x + z <
y + z.
Proof. Since x < y, certainly x ≤ y, so by (OAG) we have x + z ≤ y + z. If
x + z = y + z then adding −z to both sides gives x = y, a contradiction. 
Lemma 14.2. Let x1 , x2 , y1 , y2 be elements of an ordered abelian group G with
x1 ≤ x2 and y1 ≤ y2 . Then x1 + y1 ≤ x2 + y2 .
Proof. Applying (OAG) with x1 , x2 , y1 gives x1 + y1 ≤ x2 + y1 . Applying
(OAG) with y1 , y2 , x2 gives x2 + y1 = y1 + x2 ≤ y2 + x2 . By transitivity we have
x1 + y1 ≤ x2 + y2 . 
To an ordering on a commutative group we associate its positive cone:
G+ = {x ∈ G | x > 0}.
Elements of G+ are called positive. We also define
G− = {x ∈ G | x < 0}.
Elements of G− are called negative.
Lemma 14.3. Let x be a nonzero element of`the ordered abelian group G. Then
exactly one of x, −x is positive. Thus G = {0} G+ G− .
`

Proof. If x > 0 and −x > 0 then adding gives 0 > 0, a contradiction.

If x is not positive then x < 0. By Lemma 14.1 we may add −x to both sides,
getting 0 = x + (−x) < 0 + x = −x. 
Lemma 14.4. Let x1 , x2 be elements of an ordered abelian group.
a) If x1 , x2 ∈ G+ , then x1 + x2 ∈ G+ .
b) If x1 , x2 ∈ G− , then x1 + x2 ∈ G− .
Proof. a) Since x1 > 0 and x2 > 0, then by Lemma 14.1 we have
0 < x1 < x1 + x2 .
141
142 14. ORDERED FIELDS

b) If x1 < 0 and x2 < 0, then by Lemma 14.3 we have −x1 , −x2 > 0. Now part a)
gives −x1 − x2 = −(x1 + x2 ) > 0, so by Lemma 14.1 again we have x1 + x2 < 0. 
In an ordered abelian group we define |x| to be x if x ≥ 0 and −x otherwise.
Exercise 14.1. Let x, y be elements of an ordered abelian group G.
a) Suppose x ≤ y and n ∈ N. Show that nx ≤ ny.
b) Suppose x ≤ y and n is a negative integer. Show that nx ≥ ny.
Example 14.5. Let (G, <) be an ordered abelian group and H a subgroup of
G. Restricting < to H endows H with the structure of an ordered abelian group.
Example 14.6 (Lexicographic Ordering). Let {Gi }i∈I be a nonempty indexed
family of ordered abelian groups. Suppose that we are given Q a well-ordering on the
index set I. We may then endow the direct product G = i∈I Gi with the structure
of an ordered abelian group, as follows: for (gi ), (hi ) ∈ G, we decree (gi ) < (hi ) if
for the least index i such that gi 6= hi , gi < hi .
Theorem 14.7. (Levi [Lev43])
For an abelian group G, the following are equivalent:
(i) G admits at least one ordering.
(ii) G is torsionfree.
Proof. (i) =⇒ (ii) Let < be an ordering on G, and let x ∈ G• . By Lemma
14.4 we have nx 6= 0 for all n ∈ Z+ .
(ii) =⇒ (i): Let G be a torsionfree abelian group. Then G is a flat Z-module.
Tensoring the injection Z ,→ Q gives us an injection G ,→ L G ⊗ Q. Since Q is a
field, the Q-module G ⊗ Q is free, i.e., it is isomorphic to i∈I Q. Choose a total
ordering on I. Give each copyL of Q its standard ordering as a subfield of R and put
the lexicographic ordering on i∈Q Q ∼ = G ⊗ Q. Via the injection G ,→ G ⊗ Q this
induces an ordering on G. 
An anti-isomorphism of abelian groups is an order-reversing group isomophism.
For every ordered abelian group (G, <), the inversion map x ∈ G 7→ −x is an
anti-isomorphism of G.
Exercise 14.2. a) Show that the abelian group Z admits exactly two orderings
∼
<1 and <2 , such that inversion gives an isomorphism (Z, <1 ) → (Z, <2 ).
b) Give an example of an abelian group G admitting orderings <1 and <2 such that
(G, <1 ) is not isomorphic or anti-isomorphic to (G, <2 ).
The comparability quasi-ordering: For x, y ∈ G, we write x ≺ y if there exists
n ∈ Z+ such that |x| ≤ n|y|. We claim that ≺ is a quasi-ordering on G, i.e., a
reflexive, transitive but not necessarily anti-symmetric binary relation. Indeed the
reflexivity is immediate; if x ≺ y and y ≺ z then there exist n1 , n2 ∈ Z+ such that
|x| ≤ n1 |y| and |y| ≤ n2 |z|, and thus |x| ≤ n1 n2 |z|.

As is the case for any quasi-ordering, the relation x ≺ y and y ≺ x is an equiva-

lence relation, and the quasi-ordering descends to a partial ordering on equivalence
classes. Write x ≈ y for the resulting equivalence relation on the ordered group G:
explicitly, there exist n1 , n2 ∈ Z+ such that |x| ≤ n1 |y| and |y| ≤ n2 |x|.
Exercise 14.3. Show that the resulting partial ordering on G/ ≈ is a total
ordering.
 1. ORDERED ABELIAN GROUPS 143

In any ordered abelian group G, {0} is its own ≈-equivalence class, hence any non-
trivial ordered abelian group has at least two ≈-equivalence classes. We refer to
nonzero ≈-equivalence classes as Archimedean equivalence classes and denote
the set of all such equivalence classes as Ω(G).
An ordered abelian group with #Ω(G) ≤ 1 is called Archimedean. Equiva-
lently, for all x, y ∈ G• , there are n1 , n2 ∈ Z+ such that |x| ≤ n1 |y| and |y| ≤ |x|.
Example 14.8. The group (R, +) is Archimedean. That is, for any x ∈ R>0
there are positive integers n1 and n2 such that n11 ≤ x ≤ n2 . Indeed the second
inequality follows from the least upper bound axiom: if this were not the case then
the set Z+ of positive integers would be bounded above in R, and this set cannot
have a least upper bound. The first inequality follows from the second upon taking
reciprocals.
Example 14.9. A subgroup of an Archimedean ordered abelian group is Archimedean.
In particular, any subgroup of (R, +) is Archimedean in the induced ordering.
Rather remarkably, the converse is also true.
Theorem 14.10. (Hölder [Hö01]) Let (G, +) be an ordered abelian group. If
G is Archimedean, there exists an embedding (G, +) ,→ (R, +).
Proof. We may assume G is nontrivial. Fix any positive element x of G. We
will construct an order embedding of G into R mapping x to 1.
Namely, let y ∈ G. Then the set of integers n such that nx ≤ y has a maximal
element n0 . Put y1 = y − n0 x. Now let n1 be the largest integer n such that
nx ≤ 10y1 : observe that 0 ≤ n1 < 10. Continuing in this way we get P a set of integers
∞ nk
n1 , n2 , . . . ∈ {0, . . . , 9}. We define ϕ(y) to be the real number n0 + k=1 10 k . It is

not hard to show that ϕ is isotone – y ≤ y 0 =⇒ ϕ(y) ≤ ϕ(y 0 ) – and also that ϕ is
injective: we leave these tasks to the reader.
But let us check that ϕ is a homomorphism of groups. For y ∈ G, and r ∈ Z+ ,
let 10nr be the rational number obtained by truncating ϕ(y) at r decimal places.
The numerator n is characterized by nx ≤ 10r y < (n + 1)x. For y 0 ∈ G, if
n0 x ≤ 10r y 0 ≤ (n0 + 1)x, then
(n + n0 )x ≤ 10r (y + y 0 ) < (n + n0 + 2)x,
so
2
ϕ(y + y 0 ) − (n + n0 )10−r <
10r
and thus
4
|ϕ(y + y 0 ) − ϕ(y) − ϕ(y 0 )| < .
10r
Since r is arbitrary, we conclude ϕ(y + y 0 ) = ϕ(y) + ϕ(y 0 ). 
Proposition 14.11. Let G be a nontrivial Archimidean ordered abelian group.
Then exactly one of the following holds:
(ii) G is order-isomorphic to Z.
(ii) The ordering on G is dense.
Proof. Step 1: Suppose G+ has a least element x. Let y ∈ G+ . Since
the ordering is Archimedean there is a largest n ∈ Z+ such that nx ≤ y. Then
y − nx ≥ 0; if y > 0 then y − nx ≥ x so y ≥ (n + 1)x, contradicting the maximality
of n. Thus y = nx, i.e., every positive element of G+ is a multiple of x. It follows
144 14. ORDERED FIELDS

that there is a unique order isomorphism from G to (Z, <) carrying x to 1.

Step 2: Suppose G is not isomorphic to (Z, <), so there is no least positive element.
In other words, given any positive element x there exists 0 with 0 < y < x. Now let
a, b ∈ G with a < b. If 0 < y < b − a then a < y < b. So the ordering is dense. 
Theorem 14.12. (Pierce) Let (G, <) be an ordered abelian group. Let GD be
the Dedekind completion of the linearly ordered set (G, <).
a) There is a unique commutative monoid structure + on GD such that (GD , +) is
an ordered commutative monoid and the natural map (G, +, <) ,→ (GD , +, <) is a
homomorphism of ordered commutative monoids.
b) The following are equivalent:
(i) GD is a group.
(ii) (G, <) is Archimedean.
1.0.1. Hahn Embedding Theorem. The following generalization of Hölder’s The-
orem was proven by H. Hahn.
Theorem 14.13. (Hahn Embedding Theorem) Let G be an ordered abelian
group. Then there is an embedding of ordered abelian groups
h : G ,→ RΩ(G) .
Here Ω(G) denotes the Archimedean equivalence classes of G• and RΩ(G) is lexico-
graphically ordered.

2. Introducing Ordered Fields

An ordered ring is a ring (R, +, ·) together with a total ordering ≤ on R compat-
ible with the commutative group (R, +) and satisfying the additional property

(OR) ∀x, y ≥ 0, xy ≥ 0.

In these notes the ordered rings we will study are ordered fields.
Example 14.14. The real numbers R with the standard < form an ordered
field.
√
Example 14.15. Let F = Q( 2). There are two embeddings F ,→ R √ which
differ
√ from each other by the nontrivial
√ automorphism of F , which carries 2 7→
− 2. In one of these embeddings, 2 goes to the positive real number whose square
is 2, and in the other one it goes to the negative real number whose square is 2.
Thus the two embeddings give different orderings, and it is easy to check that these
are the only two orderings of F .
A homomorphism f : (F, <) → (F 0 , <0 ) is a monotone field homomorphism: i.e., a
field homomorphism such that x < y =⇒ f (x) < f (y).
Exercise 14.4. Let (K, <) be an ordered field and let F be a subfield of K.
Denote by <F the restriction to F of <. Show that (F, <F ) is an ordered field and
the inclusion of F into K is an homomorphism of ordered fields.
We denote by X(K) the set of all field orderings on K.
Exercise 14.5. Show that there is a natural action of Aut(K) on X(K). Give
an example where the orbit space Aut(K)\X(K) consists of more than one element.
 2. INTRODUCING ORDERED FIELDS 145

Proposition 14.16. Every ordered field (K, ≤) has characteristic 0.

Proof. Apply Theorem 14.7 to (K, +). 
For a subset S ⊂ K, put S • = S \ {0}.

We consider the following conditions on a subset P of a field K:

(PO1) P + P ⊂ P , and P P ⊂ P .
(PO2) Σ (K) = {x21 + . . . + x2n | xi ∈ K} ⊂ P .
(PO3) −1 ∈ / P.
(PO30 ) P ∩ (−P ) = {0}.
(PO300 ) P • + P • ⊂ P • .
(PO3000 ) P 6= K.
(PO4) P ∪ (−P ) = K.
Exercise 14.6. Let P ⊂ K satisfy (PO1) and (PO2).
a) Show: (PO3), (PO30 ), and (PO300 ) are equivalent conditions on P .
b) Suppose char K 6= 2. Show: (PO3000 ) and (PO3) are equivalent conditions on P .

2
2
(Hint: x = x+1
2 − x−12 .)
c) Suppose char K = 2. Show: P satisfies (PO1) and (PO2) iff P is a subfield of
K containing K 2 .
Exercise 14.7. Let P ⊂ K satisfy (PO1) and (PO4). Show: P satisfies
(PO2).
Lemma 14.17. Let K be a field.
a) If ≤ is a field ordering on K, put P = K ≥0 . Then K satisfies (PO1) through
(PO4) above, and also 1 ∈ P .
b) Let P ⊂ K satisfy (PO1) through (PO4). Define a relation ≤ on K by x ≤
y ⇐⇒ y − x ∈ P . Then ≤ is a field ordering on K.
Proof. a) By Proposition X.X, K has characteristic 0. Lemma 146 implies P
satisfies (PO1), and Lemma 147 implies P satisfies (PO30 ) and (PO4). By Exercise
X.X, P satisfies (PO2). Finally, by (PO4), exactly one of 1, −1 lies in P . But if
−1 ∈ P , then (−1)2 = 1 ∈ P , so 1 ∈ P .
b) By (PO3) and (PO4), ≤ is a total ordering on K. Given x, y, z ∈ K with x ≤ y,
then (y + z) − (x + z) = y − x ∈ P , so x + z ≤ y + z: (K, ≤) is an ordered abelian
group. Finally, if x, y ≥ 0 then x, y ∈ P , so by (PO1) xy ∈ P , i.e., xy ≥ 0. 
In view of this result, we refer to a subset P ⊂ K satisying (PO1), (PO2) and
(PO3) as being an ordering on K, and we often refer to the ordered field (K, P ).
Exercise 14.8. Let P1 , P2 be two orderings on a field K. Show: P1 ⊂ P2 =⇒
P1 = P2 .
The alert reader may now be wondering why we have introduced (PO2) at all since
it is implied by the other axioms for an ordering.1 The reason is that it is a key
idea to entertain a more general structure.

A subset P ⊂ K satisying (PO1), (PO2) and (PO3) is called a preordering

1The less than alert reader may now be asleep, and we owe him our apologies: things will
liven up shortly!
146 14. ORDERED FIELDS

of K. (Note that our choice of (PO3) instead of (PO3000 ) shows that a field of
characteristic 2 admits no preorderings.)
Exercise 14.9. Let T be a preordering on F and x, y ∈ T .
Show that x, y ∈ T, x + y = 0 =⇒ x = y = 0.
A field K is formally real if −1 ∈
/ Σ (K).
Lemma 14.18. Suppose char K 6= 2. Then the following are equivalent:
(i) K is formally real.
(ii) Σ (K) ( K. (iii) Σ (K) is a preordering on K.
(iv) For all n ≥ 1 and all x1 , . . . , xn ∈ K, x21 +. . .+x2n = 0 =⇒ x1 = . . . = xn = 0.
Exercise 14.10. Prove it.
Remark: Condition (iv) above makes a connection with quadratic form theory. A
quadratic form q(x) = a1 x21 + . . . + an x2n over a field K is isotropic if there exists
some nonzero x ∈ K n with q(x) = 0 and otherwise anisotropic. Thus by Lemma
X.X, a field is formally real iff for each n the sum of n squares form is anisotropic.
Lemma 14.19. Let F be a field such that Σ (F ) ∩ (−Σ (F )) = ∅ and Σ (F ) ∪
(−Σ (F )) = F × . Then P = Σ (F ) is the unique ordering on F .
Exercise 14.11. Prove Lemma 14.19.
Exercise 14.12. Use Lemma 14.19 to show that each of the following fields
admits a unique ordering: R, Q, the field of constructible numbers.
Proposition 14.20. If (F, P ) is an ordered field, F is formally real.
Proof. The contrapositive is clear: if F is not formally real, then −1 is a sum
of squares, so it would be – along with 1 – in the positive cone of any ordering. 
It follows that any ordered field has characteristic 0.

Much more interestingly, the converse of Proposition 14.20 is also true. In order to
prove this celebrated result we need the following innocuous one.
Lemma 14.21. Let F be a field, T ⊆ F a preordering on F , and a ∈ F × . The
following are equivalent:
(i) The set T [a] := {x + ya | x, y ∈ T } is a preordering.
(ii) a is not an element of −T .
Proof. Since by (PC4) no preordering can contain both a and −a, (i) =⇒
(ii) is clear. Conversely, assume (ii). It is immediate to verify that T [a] satisfies
(PC1), (PC2) and (PC5), so it suffices to show that there is no x ∈ F × such that
x and −x both lie in T [a]. If so, we deduce
1
−1 = −x · x · ( )2 ∈ T [a].
x
But now suppose −1 = x + ya for x, y ∈ T . Then −ya = 1 + x is a nonzero element
of T , so a = (−y)−2 (−y)(1 + x) ∈ −T , a contradiction. 
Theorem 14.22. Let t be a preordering on a field K. Then:
a) The set t is the intersection of all orderings P ⊃ t.
b) (Artin-Schreier) If K is formally real, then it admits an ordering.
 3. EXTENSIONS OF FORMALLY REAL FIELDS 147

Proof. a) Step 1: Let S be the set of all preorderings on K containing t. The

union of a chain of preorderings is again a preordering. Applying Zorn’s Lemma,
we get a maximal element T ⊃ t. By Lemma 14.21 we have that for all a ∈ F , if
−a 6∈ T then a ∈ T , so T satisfies (PO4) and is therefore an order.
Step 2: Let b ∈ K \ t. We must construct an ordering P ⊃ t with b ∈ / P . But by
Lemma 14.21, t[−b] is a preordering, which by Step 1 extends to an ordering P ,
and since −b ∈ P , b ∈/ P.
b) If K is formally real then Σ (K) is a preordering on K. In particular, by (PO3)
a preordering is a proper subset of K, whereas by part a) if there were no orderings
on K then the intersection over all orderings containing Σ (T ) would be the empty
intersection, and thus would equal K. 
The following special case of Theorem X.Xa) is important in its own right.
Corollary 14.23. (Artin) For x ∈ K × , char(K) 6= 2, the following are
equivalent:
(i) For every ordering P on K, x ∈ K.
(ii) The element x is a sum of squares.
Remark: Corollary 14.23 is an important step towards the solution of Hilbert’s 17th
problem: show that any positive semidefinite polynomial f ∈ R[t1 , . . . , tn ] is a sum
of squares of rational functions.

Remark: Corollary 14.23 does not extend to all fields of characteristic 2. Indeed,
for a field F of characteristic 2, we simply have Σ (F ) = F 2 , so every element of
F is a sum of squares iff F is perfect. (In no case are there any orderings on F .)

3. Extensions of Formally Real Fields

Let L/K be a field extension. If L is formally real, then by Artin-Schreier it admits
an ordering P , which restricts to an ordering p on K.2 However, there is a related
but much more subtle question: suppose p is an ordering on a field K and L/K is
an extension field. Can the ordering p be extended to L?
An obvious necessary condition is that L be formally real: if not it admits no
orderings at all, let alone an extension of p. But this condition is not sufficient: let
K = R(t). By Example X.X above there is a unique ordering p on K extending √
the unique ordering on R and such that x ≤ t for all x ∈ R. Take L = K( −t).
Clearly p does not extend to L,√since if so the negative element −t would be a
square. However, the element√ −t is transcendental over K, so there is a K-
algebra automorphism K( −t) → K(t), and thus L is certainly formally real.

In general the extension problem for orderings is a rich one with a large liter-
ature. But we will give one fundamental and useful result, an extension of the
Artin-Schreier Theorem. First:
Lemma 14.24. For an ordered field (K, p), an extension L/K, and c ∈ L, the
following are equivalent:
(i) There are a1 , . . . , an ∈ p• and x1 , . . . , xn ∈ L such that
c = a1 x21 + . . . + an x2n .
2particular, a subfield of a formally real field is formally real. But that was clear anyway
from the definition.
148 14. ORDERED FIELDS

T
(ii) c ∈ P ⊃p P, the intersection being over all orderings of L extending p.
Proof. Let
t = {a1 x21 + . . . + an x2n | ai ∈ p, xi ∈ L},
T
and note that the desired equivalence can be rephrased as t = P ⊃p P . Moreover
t satisfies (PO1) and (PO2), and an ordering P of L contains t iff it contains
T p.
Case 1: Suppose −1 ∈ / t. Then t is a preordering, and by Theorem X.X, t = P ⊃p P .
Case 2: If −1 ∈ t, there is no ordering on L extending p. Then – since KThas ordered
and thus not of characteristic 2! – by Exercise X.X, we have t = K = P ⊃p P . 
We are now ready for one of our main results.
Theorem 14.25. For an ordered field (K, p) and an extension field L/K, the
following are equivalent:
(i) There is an ordering on L extending p.
(ii) For all a = (a1 , . . . , an ) ∈ pn , the quadratic form
qa (x) = a1 x21 + . . . + an x2n
is anisotropic over L: if x = (x1 , . . . , xn ) ∈ Ln is such that q(x) = 0, then x = 0.
Proof. (i) =⇒ (ii) is immediate.
(ii) =⇒ (i): If for any a ∈ pn the quadratic form qa (x) represents −1, then
the form qa,1 (x) = a1 x21 + . . . + an x2n + x2n+1 would be isotropic, contrary to our
hypothesis. It follows that
/ t = {a1 x21 + . . . + an x2n | ai ∈ p, xi ∈ L},
−1 ∈
so – as in the proof of the previous result – t is a preordering of L containing p. By
Theorem X.X, t must extend to at least one ordering of L. 
Exercise 14.13. Deduce from Theorem 14.25 that every formally real field L
admits an ordering. (Hint: we wrote L, not K!)
We will now deduce several sufficient conditions for extending orderings.
√
Theorem 14.26. Let (K, p) be an ordered field, and let L = K({ x}x∈p ) be
the extension obtained by adjoining all square roots of positive elements. Then the
ordering p extends to L.
Proof. By Theorem 14.25, it suffices√ to show √ that for any n, r ∈ Z+ and any
b1 , . . . , br , c1 , . . . , cn ∈ p, if x1 , . . . , xn ∈ F ( b1 , . . . , br ) are such that
(20) c1 x21 + . . . + cn x2n = 0,
then x1 = . . . = xn = 0. For any fixed n, we prove this by induction on r. Suppose
by induction that the equation c1 x21 + . . . + cn x2n = 0 has no nontrivial solutions
√
over Kr−1 , and let (z1 , . . . , zn ) ∈ Krn be a solution to (20). Write zi = xi + br yi ,
with xi , yi ∈ Kr−1 . Then equating “rational parts” in the equation
X X X X p
0= ci zi2 = ci x2i + br ci yi2 + 2 ci xi yi br
2n
shows that (x1 , . . . , xn , y1 , . . . , yn ) ∈ Kr−1 is a solution of
c1 t21 + . . . + cn t2n + br c1 t2n+1 + . . . + br cn t22n = 0.
By induction, x1 = . . . = xn = y1 = . . . = yn = 0, i.e., z1 = . . . = zn = 0. 
 3. EXTENSIONS OF FORMALLY REAL FIELDS 149

To obtain further results we take a perspective arising from quadratic form theory.
Let us say a field extension L/K is anistropic if every anisotropic quadratic form
q(x1 , . . . , xn ) ∈ K[x1 , . . . , xn ] remains aniostropic when extended to L. (In the
algebraic theory of quadratic forms one studies the Witt kernel of a field extension:
the kernel of the natural ring homomorphism W (K) → W (L). An anisotropic
extension is precisely one in which the Witt kernel is trivial.) From Theorem 14.25
we immediately deduce the following result.
Corollary 14.27. If (K, p) is an ordered field and L/K is an anisotropic
extension, then the ordering p extends to L.
Exercise 14.14. a) Let K be a field and let {Li }i∈I be a directed system of
anisotropic extensions of K. Show that lim Li /K is an anisotropic extension.
−→
b) Let (K, p) be an ordered field and L/K a field extension. Suppose that p extends
to an ordering on any finitely generated subextension of L/K. Show that p extends
to an ordering on L.
The next results give the two basic examples of anisotropic extensions.
Theorem 14.28. A purely transcendental extension L/K is anisotropic.
Proof. Step 0: It suffices to prove that K(t)/K is anisotropic. Indeed, if
so then an immediate induction gives that K(t1 , . . . , tn )/K is anisotropic, and we
finish by applying Exercise X.X.
Step 1: Let K be any field, and let (f1 , . . . , fn ) ∈ K(t)n be an n-tuple of ra-
tional functions, not all zero. Then there exists a nonzero rational function f
such that (f f1 , . . . , f fn ) is a primitive vector in K[t], i.e., each f fi ∈ K[t] and
gcd(f f1 , . . . , f fn ) = 1. Indeed this holds with K[t] and K(t) replaced by any UFD
and its fraction field.

Step 2: Let q = a1 x21 + . . . + an x2n be a nonsingular quadratic form over K such

that qK(t) is isotropic: that is, there exist rational functions f1 , . . . , fn , not all zero,
such that
a1 f12 + . . . + an fn2 = 0.
Let f ∈ K(t)× be the rational function as in Step 1; then multiplying through by f 2
we get a primitive polynomial solution, i.e., there exist polynomials p1 (t), . . . , pn (t) ∈
K[t] with gcd(p1 (t), . . . , pn (t)) = 1 and
a1 p1 (t)2 + . . . + an pn (t)2 = 0.
Now we substitute t = 0 (or any value of K): we cannot have p1 (0) = . . . = pn (0) =
0, because then all of the pi ’s would be divisible by t, contradicting primitivity.
Therefore q(p1 (0), . . . , pn (0)) = 0 shows that q is isotropic over K. 

Remark: The proof of Proposition X.X used only that q was a form – i.e., a homo-
geneous polynomial – not that it was a quadratic form. Indeed any system of ho-
mogeneous polynomials would work as well, so the argument really shows: if V/K is
a projective variety which has a K(t)-rational point, then it has a K-rational point.3

3The same conclusion holds for arbitrary varieties over any infinite field, or for complete
varieties over a finite field. But taking the projective line over Fq and removing its Fq -rational
points shows that some hypothesis is necessary!.
150 14. ORDERED FIELDS

The following was conjectured by Witt in 1937 and proven by Springer in 1952.4
Theorem 14.29. (Springer [Sp52]) Let L/K be a field extension of finite odd
degree d. Then L/K is anisotropic.
Proof. We go by induction on the degree, the case d = 1 being trivial. Sup-
pose the result holds for all field extensions of odd degree less than d, and L/K
be an extension of odd degree d. If L/K had any proper subextension, then we
would be done by a dévissage argument. So we may assume in particular that L is
monogenic over K: L = K[x]. Let p(t) ∈ K[t] be the minimal polynomial of x. Let
q be an anisotropic quadratic form over K which becomes isotropic over L: i.e.,
there exists an equation
(21) q(g1 (t), . . . , gn (t)) = h(t)p(t)
with polynomials gi , h ∈ K[t], not all gi = 0, and M := max deg gi ≤ d − 1.
As in the proof of Proposition 14.28, we may also assume that (g1 , . . . , gn ) is a
primitive vector in K[t]. Since q is anisotropic, the left hand side of (21) has degree
2M ≤ 2d − 2, so deg h is odd and at most d − 2. In particular, h has an irreducible
factor h̃ of odd degree at most d − 2; let y be a root of h̃ in K. Taking t = y in (21),
we see that q(g1 (y), . . . , gn (y)) = 0. Note that since K[t] is a PID, the condition
gcd(g1 , . . . , gn ) = 1 is equivalent to the fact that 1 ∈ hg1 , . . . , gn i, which implies
that the polynomials g1 , . . . , gn remain setwise coprime as elements of K[y][t]. In
particular, not all gi (y) are equal to 0, so that qK[y] is isotropic. By induction, this
implies that q was isotropic, contradiction! 
Exercise 14.15. Let (K, p) be an ordered field. Show that the formal power
series field K((t)) admits a unique ordering extending p in which 0 < t < x for all
x ∈ K.
Exercise 14.16. In the algebraic theory of quadratic forms it is shown that the
√
Witt kernel of a quadratic extension L = K( p)/K is the principal ideal generated
by α = h1, −pi: [Cl-QF, Thm. II.20]. In other words, it consists of quadratic
forms a1 x21 + . . . + an x2n − pa1 x21 − . . . − pan x2n for a1 , . . . , an ∈ K × . Use this (and
induction) to give another proof of Theorem 14.26.

4. The Grand Artin-Schreier Theorem

A field F is real-closed if it is formally real and admits no proper formally real
algebraic extensions. For instance, R√is evidently real-closed since its unique non-
trivial algebraic extension is C = R( −1), which is not formally real.
S 1
Example 14.30. (Puiseux series): The Puiseux series field n∈Z+ R((t n ))
is real-closed.
The previous examples √ of real-closed fields F were obtained by showing that F is
formally real and F ( −1) is algebraically closed. In fact this is a characterization
of real-closed fields. In particular the absolute Galois group of a real-closed field
is finite and nontrivial. Remarkably, this too is a characterization of real-closed
fields! These assertions are part of the following result, one of the most striking
and celebrated theorems in all of field theory.
4
According to D. Hoffmann, Artin orally conveyed a proof of Witt’s conjecture to Witt in
1939: he calls the result the Artin-Springer Theorem.
 4. THE GRAND ARTIN-SCHREIER THEOREM 151

Theorem 14.31. (Grand Artin-Schreier Theorem) For a field F , the following

are equivalent:
(i) F is formally real and admits no proper formally real algebraic extension.
(ii) F is formally real, every odd degree polynomial over F has a root, and for each
x ∈ F × , one of x, −x is a square.
√
(iii) F is formally real and F ( −1) is algebraically closed.
(iv) The absolute Galois group of F is finite and nontrivial.
The proofs of (i) =⇒ (ii) =⇒ (iii) =⇒ (iv) follow relatively easily from what
we have already done. We give these first and then tackle (iv) =⇒ (i), the hardest
implication.5
Proof. (i) =⇒ (ii): Since an odd degree polynomial has an odd degree irre-
ducible factor, an odd degree polynomial without a root would yield a proper odd
degree extension K/F . By Proposition X.X, K would be formally real, contradict-
ing the definition of real closure. Suppose that neither x nor −x is √a square. One
of them is positive; WLOG say it is x. By Proposition X.X, F ( x) is a proper
formally real extension field, contradiction. √
(ii) =⇒ (iii) Since F √ is formally real, certainly [F ( −1) : √F ] = 2. Let F be an
algebraic clsoure of F ( −1): we wish to show that F = F ( −1). By hypothesis
on odd degree polynomials having a root, the absolute√ Galois group√of F is a pro-2-
group, and thus so is the absolute Galois group of F ( −1).√If F ( −1) 6= F then,
we are entitled
√ to a proper finite degree extension M of F ( −1), which is Galois
over F ( −1) and has degree a power of 2. By the basic theory of 2-groups together
√
√ correspondence, there must exist a subextension G of M/F (√−1)
with the Galois
with [G : F ( −1)] = 2. But we claim that the hypotheses on F imply that F ( −1)
is quadratically closed. Indeed, let a, b be arbitrary elements of F . We claim that
there are c, d ∈ F such that
√ √
a + b −1 = (c + d −1)2 .
b
This amounts to the system a = c2 − d2 , b = 2cd. Substituting d = 2c , we get the
2 2
b b
equation c2 = a + 4c2 , or c4 − ac2 − 4 = 0. The quadratic formula gives
√
2 a ± a2 + b2
c = .
2
Since inside the radical we have a sum of squares, the squareroot does exist in F .
If we choose the plus sign in the squareroot, it is easy to see that the expression is
again non-negative, so we can solve for c in our field F .
(iii) =⇒ (iv) is immediate. 
Now we begin the proof of (iv) =⇒ (i), so suppose that F is a field with algebraic
closure F such that 1 < [F : F ] < ∞.

Step 1: We claim that F /F is Galois.

Proof: Certainly F /F is normal, so it suffices to show that it is separable. If

F has characteristic 0 (which it cannot, in fact, but we haven’t shown that yet),
then there is nothing to say, so suppose F has characteristic p > 0. We claim that
the hypotheses imply that F is perfect, and thus that every algebraic extension of
5Our proof of (iv) =⇒ (i) closely follows lecture notes of Keith Conrad.
152 14. ORDERED FIELDS

F is separable. Indeed, if F is not perfect, then there exists α ∈ F \ F p and then by

n
Lemma 9.20, the polynomials tp − α are irreducible for all n ∈ Z+ so [F : F ] = ∞.

Step 2: Let G = Gal(F /F ). We wish to show that #G = 2. If not, then by

Sylow theory there exists a subgroup H of order either 4 or an odd prime `. We
wish to derive a contradiction.

We will consider the cases #G = ` a prime number and #G = 4 in turn. First we

suppose #G = ` and let σ be a generator of the cyclic group G.

Step 3: We claim that the characteristic of F is not equal to `. If it were, then

Artin-Schreier theory would apply, so that F = F (α), where α is a root of an
Artin-Schreier polynomial tp − t − a ∈ F [t]. We may write any element b ∈ F as
b = b0 + b1 α + . . . + b`−1 α`−1
for unique b0 , . . . , b`−1 ∈ F . Thus
`−1
X `−1
X
b` − b = b`i α`i − bi αi = b`i (α + a)i − bi αi = bpp−1 − bp−1 αp−1 + O(αp−2 ),


i=0 i=0
p−2
where by O(α ) we mean a polynomial in α of degree at most p − 2. Choose
b ∈ F such that bp − b = aαp−1 , and then equating coefficients of αp−1 gives
bpp−1 − bp−1 − a = 0. Since bp−1 ∈ F , this contradicts the irreducibility of tp − t − a.

Step 4: Since the characteristic of F is not ` = #G, F contains a primitive `th

root of unity ζ. Indeed, since [F (ζ) : F ] ≤ ` − 1 and (` − 1, `) = 1, we must have
ζ ∈ F . Therefore Kummer Theory applies to give F = F (γ), where γ ` = c ∈ F .
2 2 2 2
Choose β ∈ F such that β ` = γ, so β ` = c. Thus β ` = σ(β ` ) = (σβ)` , so
2
σ(β) = ωβ with ω ` = 1. Then ω ` , being an `th root of unity, lies in F . If ω ` = 1,
then (σ(β))` = β ` , so σ(β ` ) = β ` and then β ` = γ ∈ F , contradiction. So ω is a
primitive (`2 )th root of unity. It follows easily that there exists k ∈ Z such that
σω = ω 1+`k .
From σβ = ωβ, we get
`−1
β = σ p β = σ `−1 ωβ = ωσ(ω) · · · σ `−1 (ω)β = ω 1+(1+`k)+...+(1+`k) β.
From this we deduce
`−1
X
1 + (1 + `k) + . . . + (1 + `k)`−1 ≡ 0 (mod `2 ).
i=0

Expanding out the binomial and reducing modulo `2 , we get

`−1
X (` − 1)(`)
0≡ (1 + i`k) ≡ ` + (`k) (mod `2 ).
i=0
2
If ` is odd, this gives 0 ≡ ` (mod `2 ), a contradiction. When ` = 2, we get
2 + 2k ≡ 0 (mod 4),
so that k is odd. In this case ω has order 4 and σω = ω 1+2k = ω 3 , so σω 6= ω and
ω 6∈ F . Let us write ω as i. In summary: if #G is prime, then it equals 2, i 6∈ F
 4. THE GRAND ARTIN-SCHREIER THEOREM 153

and F does not have characteristic 2.

Step 5: Now suppose that #G = 4. Then there exists at least one subexten-
sion K of F /F with [F : K]. Then the above reasoning shows that i 6∈ K, hence
not in F , but then F (i) is a subfield of F with [F : F (i)] = 2 and containing a 4th
root of unity, contradicting the above analysis.

In summary, we have shown so far that if 1 < [F : F ] < ∞, then F does not
have characteristic 2 and F = F (i). It remains to be shown that F is formally real,
and this is handled by the following result.

Lemma 14.32.
√ which −1 is not a square and such that every
Let F be a field in √
element of F ( −1) is a square in F ( −1). Then:
a) Σ (F ) = F 2 ,
b) char(F ) = 0, and
c) F is formally real.
√
Proof. Put i = −1. To show part a), it is enough to see that the sum of
two squares in F (i) is again a square in F (i). Let a, b ∈ F . By hypothesis, there
are c, d ∈ F such that (a + bi) = (c + di)2 , so a = c2 − d2 and b = 2cd and thus
a2 + b2 = (c2 + d2 )2 .
b) If F had positive characteristic p, then −1 is a sum of p − 1 squares but not
itself a square, contradicting part a).
c) Since −1 is not a square, F does not have characteristic 2, and thus by part a)
−1 is not a sum of squares and F is formally real. 

The following exercises give strengthenings and variations on the Artin-Schreier

theorem.

Exercise 14.17. (E. Fried): Let F be a field. Suppose that there exists a
positive integer d such that for every irreducible polynomial P ∈ K[t], deg(P ) ≤ d.
Show that F is real-closed or algebraically closed.

Exercise 14.18. (Knopfmacher-Sinclair) Let F be a field. Suppose that the set

of isomorphism classes of finite-dimensional field extensions of F is finite. Show
that F is real-closed or algebraically closed.

Exercise 14.19. (K. Conrad): A field K is real-closed iff 1 < [K sep : K] < ∞.

Exercise 14.20. (E. Fried): Let C be an algebraically closed field and K a

subfield of C with K 6= √
C. Suppose that C is finitely generated over K. Then K is
real-closed and C = K( −1).

Corollary 14.33. Let R be a real-closed field and K be a subfield of R. Let

K 0 be the algebraic closure of K in R. Then K 0 is real-closed.

Proof. Certainly K 0 is formally real. If P (t) ∈ K 0 [t] is an irreducible poly-

nomial of odd degree, then K 0 [t]/(P ) is formally real, so P has a root in R and
therefore also in K 0 . Moreover, if 0 6= α ∈ K 0 , then exactly one of α, −α is a square
in R, so that t2 ± α has a root in R and thus in K 0 . By Theorem 14.31, K 0 is
real-closed. 
154 14. ORDERED FIELDS

5. Sign Changing in Ordered Fields

Let (K, p) be an ordered field, and let f ∈ K[t] be a polynomial. If for a, b ∈ K we
have f (a)f (b) < 0, then we say f changes sign between a and b. If such a, b
exist we say f changes sign.
Lemma 14.34. Let (K, p) be an ordered field.
a) Every odd degree f ∈ K[t] changes sign.
b) For all a > 0, the polynomial t2 − a changes sign.
Exercise 14.21. Prove Lemma 14.34.
Proposition 14.35. For an ordered field (F, p), the following are equivalent:
(i) (Polynomial Intermediate Value Theorem) Let f ∈ F [t] and let a < b ∈ F
be such that f (a)f (b) < 0. Then there is c ∈ F such that a < c < b and f (c) = 0.
(ii) F is real-closed.
Proof. (i) =⇒ (ii): Suppose the Polynomial Intermediate Value Theorem
holds in F . By Lemma 14.34, every odd degree polynomial f ∈ K[t] change sign
hence has a root. Similarly, if a ∈ F × , then either a or −a is positive; without loss
of generality a > 0, and by Lemma 14.34, t2 − a changes sign so has a root. Thus
there is b ∈ F with b2 = a. By Theorem 14.31 F is real-closed.
(ii) =⇒ (i): Without loss of generality we may assume that f (a) < 0,, f (b) > 0
and that f is monic irreducible. By Theorem 14.31 f has degree 1 or 2. At this
point the proof is an amusing  callback to high school algebra. If f has degree 1
f (b)−f (a)
then it is f (a) + b−a x, so it has a unique root and is moreover increasing,
so its unique root must occur in (a, b). Otherwise f (t) = t2 + ct + d, so by the
quadratic formula if it does
 not 2have
 a root then c2 − 4d < 0, but then for all
2
x ∈ K, f (x) = x + 2c + d − c4 > 0, contradiction!




Proposition 14.36. Let (K, p) be an ordered field, and let f ∈ K[t] be an

irreducible polynomial which changes sign. Then the field L = K[t]/(f ) admits an
ordering extending p.
Proof. We go by induction on n = deg f , the base case n = 1 being trivial.
So suppose n ≥ 2, that the result holds for all smaller degrees and – seeking a
contradiction – that it fails for some irreducible f of degree n. By Theorem 14.25
then there are ai ≥ 0 and fi ∈ K[t], each of degree at most n − 1, such that
X
1+ ai fi (t)2 ≡ 0 (mod f )
i
and thus there is 0 6= h ∈ K[t] with deg h ≤ n − 2 such that
X
1+ ai fi (t)2 = f (t)h(t).
i
Plugging in t = a and t = b we find f (a)h(a) > 0 and f (b)h(b) > 0 and thus
h(a)h(b) < 0. There must then be at least one irreducible factor g(t) of h(t) such
that g(a)g(b) < 0. Since
deg g ≤ deg h ≤ n − 2 < n = deg f
and X
1+ ai fi (t)2 ≡ 0 (mod g),
i
 6. REAL CLOSURES 155

this contradicts our induction hypothesis. 

Exercise 14.22. Use Proposition 14.36 to deduce new proofs of many (as many
as possible!) of the results of § 16.3.

6. Real Closures
Proposition 14.37. For every formally real field K, there exists an algebraic
extension K rc which is real-closed.
Proof. Let K be an algebraic closure of K, and consider the partially ordered
set of formally real subextensions of K/K. Since the union of a chain of formally
real fields is formally real, Zorn’s Lemma applies to give a maximal formally real
subextension, which is by definition real-closed. 
Definition: A real closure of a formally real field K is a real-closed algebraic
extension of K.
Lemma 14.38. Let K be a field, let R/K be a real-closed extension field of K,
and let R0 be the algebraic closure of K in R. Then R0 is a real closure of K.
Exercise 14.23. Prove Lemma 14.38.
Thus we have shown the existence of real closures for formally real fields. What
about uniqueness? By comparison with the case of algebraically closed fields, one
might guess that any two real closures of a given formally real field K are isomorphic
as K-algebras. However, this is in general very far from being the case!
Example 14.39. Let K = Q(t). There is a unique embedding ι : K → R
in which t gets sent to π. Let K1 be the algebraic closure of ι(K) in R. On the
S 1
other hand, let ι2 : K → n R(t n ) be the natural embedding of K into the Puiseux
S 1
series field, and let K2 be the algebraic closure of ι2 (K) in n R(t n . By Corollary
14.33, K1 and K2 are both real-closed fields. In particular, they each admit a unique
ordering, in which the positive elements are precisely the nonzero squares. However,
the ordering on K1 is Archimedean and the ordering on K2 is not, since t is an
infinitesimal element. Therefore K1 and K2 are not isomorphic as fields, let alone
as K-algebras.
Theorem 14.40. Let (F, p) be an ordered field. Then there is an algebraic
extension R/F which is real-closed and such that the unique ordering on R extends
p.
√
Proof. Let K = F ({ x}x∈p ). By Theorem 14.26, K is formally real, and
now by Proposition 14.37, there exists a real-closed algebraic extension R of K.
Let P = {x2 | x ∈ R× } be the unique ordering on R. Every x ∈ p is a square
in K and hence also in R: that is, p ⊂ P ∩ F . Conversely, if x ∈ F × \ p, then
−x ∈ p ⊂ P ∩ F ⊂ P , so that x 6∈ P , hence x 6∈ P ∩ F . Thus P ∩ F = p. 
In the above situation, we say R is a real closure of the ordered field (F, p).
Exercise 14.24. Use Theorem 14.40 to give a third proof of X.X and X.X.
Theorem 14.41. (Sylvester) Let (K, p) be an ordered field, and let (R, P ) be a
real-closed extension. Let f ∈ K[t] be a nonzero monic separable polynomial, and
put A = K[t]/(f ). Let Bf be the trace form on the K-algebra A, i.e., the bilinear
156 14. ORDERED FIELDS

√
form hx, yi = TrA/K (x, y). Let C = R( −1). Then:
a) The number of roots of f in R is equal to the signature of Bf .
b) Half the number of roots of f in C \ R is equal to the number of hyperbolic planes
appearing in the Witt decomposition of Bf .
Proof. Let f (t) = f1 (t) · · · fr (t) be the factorization
√ of f over R[t]. Since f is
separable, the polynomials fi are distinct, and since R( −1) is algebraically closed,
each fi has degree 1 or 2. Since A ⊗K R ∼ = R[t]/(f ), the trace form of A ⊗K R
is simply the scalar extension to R of the trace form Bf . Further, by the Chinese
Remainder Theorem
r
R[t]/(f ) ∼
Y
= R[t]/(fi ),
i=1
so
r
(Bf )/R ∼
M
= Bfi .
i=1
It is easy to see that if deg fi = 1 then the trace form is just h1i, whereas the
computation at the end of Section 7 shows that when deg fi = 2 – so that R[t]/(fi ) ∼
=
C – the trace form is congruent to h2, −2i ∼ = h1, −1i = H, the hyperbolic plane.
Both parts of the theorem follow immediately. 

Sylvester’s Theorem may look rather specific and technical at first glance. Let
us explicitly extract from it the following key consequence: let f ∈ K[t] by a
polynomial defined over an ordered field (K, P ). Then if f has a root in one
real-closed field extending (K, P ), it has a root in every real-closed field extending
(K, P ). This is a very special case of Tarski’s transfer principle, which a logician
would express in the form “The theory of real-closed fields is model complete.”
Although it is a very special case, it has enough teeth to be the driving force
behind the powerful theorems we will now establish.
Theorem 14.42. Let (E, P )/(K, p) be an algebraic extension of ordered fields.
Let R be a real-closed field, and let σ : K → R be an ordered field embedding. Then
there is a unique order embedding ρ : E ,→ R extending σ.
Corollary 14.43. Let (K, P ) be an ordered field, and for i = 1, 2 let σ: (K, P ) →
Ri be real closures. There is a unique K-algebra isomorphism ρ : R1 → R2 .
Proof. Applying Theorem 14.42 with R1 = E and R2 = R, σ2 = σ, there is a
unique order embedding ρ : R1 → R2 extending σ2 . Since R2 /ρ(R1 ) is an algebraic
extension of real-closed fields we must have ρ(R1 ) = R2 . Finally, if τ : R1 → R2 is
any K-algebra homomorphism, then for all α > 0 in R1 , we have
√
τ (α) = τ ( α)2 > 0.
Thus τ is order-preserving, so τ = ρ. 

7. Artin-Lang and Hilbert

Lemma 14.44. Let K be real-closed, and let h1 , . . . , hn ∈ K[t]• . Let P be an
ordering on K(t). Then there are infinitely many a ∈ K such that
∀1 ≤ i ≤ n, sgn(hi ) = sgn(hi (a)).
 7. ARTIN-LANG AND HILBERT 157

Proof. Let h ∈ K[t]• . Then we may write

h = u(t − c1 ) · · · (t − cr )q1 (t) · · · qs (t)
with u, c1 , . . . , cr ∈ K × and qj (t) a monic irreducible quadratic for all 1 ≤ j ≤ s.
For any j,
b b2
qj (t) = q(t) = t2 + bt + c = (t + )2 + (c − ),
2 4
2
and since qj is irreducible over the real-closed field K, c − b4 > 0. It follows that
q > 0 and that for all a ∈ K, q(a) > 0. Thus
r
Y
sgn h = sgn u sgn(t − ci ),
i=1
r
Y
∀a ∈ K, sgn h(a) = sgn u sgn(a − ci ).
i=1
Qn
We may thus assume that each hi is monic and i=1 hi has distinct roots in K.
Let c be the smallest root which is strictly greater than t, or ∞ if there is no such
root. Then for all a ∈ (t, c), sgn hi (a) = sgn hi (t) for all i: this is an infinite set. 
Theorem 14.45 (Artin-Lang Homomorphism Theorem). Let R be a real-closed
field, and let E = R(x1 , . . . , xm ) be a finitely generated field extension. If E is
formally real, then there is an R-algebra map R[x1 , . . . , xm ] → R.
Proof. Let d be the transcendence degree of E/R. The case d = 0 is trivial:
then E = R[x1 , . . . , xm ] = R.
Step 1: We reduce to the d = 1 case. Indeed, let E 0 be a subextension of E/R of
transcendence degree 1. Let R be a real-closure of E, and let R0 be the algebraic
closure of E 0 in R, so by Lemma 14.38 R0 is real-closed. Assuming the result in
transcendence degree 1, there is a homomorphism of R0 -algebras
ϕ : R0 [x1 , . . . , xm ] → R0 .
Then
trdeg(K(ϕ(x1 ), . . . , ϕ(xm ))/K) ≤ trdeg(R0 /K) = trdeg(E 0 /K) = trdeg(E/K) − 1,
so by induction on d we may assume there is a K-algebra map K[ϕ(x1 ), . . . , ϕ(xm )] →
K. Composing with the restriction of ϕ to K[x1 , . . . , xm ] we get a K-map to K.
Step 2: Suppose E = K(x, y1 , . . . , yr ), with x transcendental over K and y1 , . . . , yr
algebraic over K. We want a K-algebra map K[x, y1 , . . . , yr ] → K. By the Prim-
itive Element Corollary, there is y ∈ E such that E = K(x)[y]; further, we may
take y to be integral over K[x]. Then:
gi (x, y)
∃g1 , . . . , gr ∈ K[x, y], h ∈ K[x]• such that ∀1 ≤ i ≤ r, yi = .
h(x)
If ϕ : K[x, y] → K is such that ϕ(h) 6= 0, then ϕ induces a K-algebra map
K[x, y1 , . . . , yr ] → K. Thus it is enough to show: there are infinitely many K-
algebra maps ϕ : K[x, y] → K. Indeed, if 0 = ϕ(h) = h(ϕ(x)), then ϕ maps x to
one of the finitely many roots of h in K; since y is algebraic over K(x), having fixed
ϕ(x) there are only finitely many choices for ϕ(y).
Step 3: Let
f = (x, Y ) = Y n + cn−1 (x)Y n−1 + . . . + c0 (x)
158 14. ORDERED FIELDS

be the minimal polynomial for y over K(x). Since y is integral over K[x], we have
ci (x) ∈ K[x] for all i. For a ∈ K, put fa (Y ) = f (a, Y ) ∈ K[Y ]. We look for roots of
fa in K. For if b ∈ K is such that fa (b) = f (a, b) = 0, there is a unique K-algebra
map ϕ : K[x, y] → K with ϕ(x) = a, ϕ(y) = b. So it is enough to show: there are
infinitely many a ∈ K such that there is b ∈ K with fa (b) = 0.
Step 4: Finally we use that E is formally real! Let P be an ordering on E and
let R be a real-closure of (E, P ). Then f (x, Y ) ∈ K[x][Y ] has a root in R, namely
y ∈ E ⊂ R. By Sylvester’s Theorem, sgn(Bf )/K(x) > 0. If we can show that
there are infinitely many a ∈ K such that sgn((Bf )a )/K) > 0, then applying
Sylvester’s Theorem again we will get infinitely many a such that fa (Y ) has a root
in K and be done. We may diagonalize the quadratic form corresponding to Bf
as hh1 (x), . . . , hn (x)i, say. Staying away from the finitely many a such that hi (a)
is zero or undefined for some i, we have that Bfa ∼ = hh1 (a), . . . , hn (a)i. By Lemma
14.44 there are infinitely many a such that sgn Bfa = sgn Bf > 0, and we’re done.


Actually Lang proved a stronger result, giving in particular a necessary and suffi-
cient condition for E to be formally real. His result uses the language of arithmetic
geometry, so unfortunately will probably not be accessible to all readers of these
notes, but here it is anyway.
Theorem 14.46 (Lang [La53]). Let V/R be a geometrically integral algebraic
variety over a real-closed field R, with function field E = R(V ). Then E is formally
real iff V has a nonsingular R-point.
The Artin-Lang homomorphism theorem is powerful enough to yield a quick proof
of the following result, which when one takes K = R = R, was the 17th of Hilbert’s
Problems proposed to the worldwide mathematical community in 1900.
Theorem 14.47 (Artin). Let K be a formally real field admitting a unique
ordering, and let R be a real closure of K. If f ∈ K[t1 , . . . , tm ] is such that
f (a1 , . . . , an ) ≥ 0 ∀(a1 , . . . , an ) ∈ Rn ,
then f is a sum of squares in K(t1 , . . . , tm ).
Proof. We argue by contraposition: suppose f ∈ K[t1 , . . . , tm ] is not a sum
of squares in K(t1 , . . . , tm ). By Corollary 14.23, there is an ordering P on E =
K(t1 , . . . , tm ) such that f <P 0. Let R be a real closure of (E, P ). Then f < 0
in R, so there is w ∈ R with w2 = −f . By Lemma 14.38, the algebraic closure
R0 of K in R is real-closed, hence is a real-closure of the ordered field K since K
admits exactly one ordering. By uniqueness of real closures R0 = R. The field
R(t1 , . . . , tm , w) is a subfield of the real-closed field R, hence by Artin-Lang there
is an R-algebra map
1
ϕ : R[t1 , . . . , tn , w, ] → R.
w
Note that the effect of including w1 is that ϕ(w)ϕ( w1 ) = 1, hence ϕ(w) 6= 0. For
1 ≤ i ≤ n, put ai = ϕ(ti ); then (a1 , . . . , an ) ∈ Rn and
f (a1 , . . . , an ) = ϕ(f ) = −ϕ(w)2 < 0.

 8. ARCHIMEDEAN AND COMPLETE FIELDS 159

Exercise 14.25. Let (K, P ) be an ordered field with real-closure R. Suppose

f ∈ K[t1 , . . . , tn ] has the property that f (a) ≥ 0 for all a ∈ Rn . Show that there is
a positive definite quadratic form q/K such that q represents f over K(t1 , . . . , tn ):
there are x1 , . . . , xn ∈ K(t1 , . . . , tn ) such that q(x1 , . . . , xn ) = f .

8. Archimedean and Complete Fields

As usual, a subset S of an ordered field F is called bounded above if there exists
a single element x ∈ F such that s ≤ x for all s ∈ S; bounded below is defined
similarly.

An ordered field F is Archimedean if the subfield Q is not bounded above.

Example 14.48. If x is any rational number, then x + 1 is a larger rational
number. Thus the field Q is Archimedean.
Exercise 14.26. Show that all of the following conditions on an ordered field
are equivalent to the Archimedean property.
(i) For all x ∈ F , there exists n ∈ Z+ with n > x.
(ii) If x, y ∈ F with x > 0, then there eixsts n ∈ Z+ with nx > y.
(iii) If x ∈ F is non-negative and such that x < n1 for all n ∈ Z+ , then x = 0.
An ordered field is non-Archimedean if it is not Archimedean.
Exercise 14.27. Show that any subfield of an Archimedean ordered field is
Archimedean, but a subfield of a non-Archimedean ordered field may be Archimedean.
An element x of an ordered field is infinitely large if x > n for all n ∈ Z+ and
infinitesimal if 0 < x < n1 for all n ∈ Z+ . Thus x is infinitely large iff x1 is
infinitesimal, and by Exercise X.X, the ordering is non-Archimedean iff infinitely
large elements exist iff infinitsimal elements exist.
Exercise 14.28. Suppose x is an infinitely large element of an ordered field.
Show that for all y ∈ Q, x − y is infinitely large.
Exercise 14.29. Let K be an ordered field; consider the rational function field
K(t).
a) Observe that Proposition ?? shows that K(t) admits at least one non-Archimedean
ordering. Show that in fact K(t) admits at least four non-Archimedean orderings.
Can you improve upon 4?
b) Use part a) to show that for every infinite cardinal κ, there exists a non-
Archimedean ordered field of cardinality κ.
A partially ordered set (S, ≤) is Dedekind complete if every nonempty subset
which is bounded above has a least upper bound.
Exercise 14.30. Show that a partially ordered set is Dedekind complete iff
every subset that is bounded below has a greatest lower bound.
Proposition 14.49. Let F be a Dedekind complete ordered field. Then the
ordering is Archimedean.
Proof. We go by contraposition: if F is non-Archimedean, then the subset
Z+ is bounded above, and the set of upper bounds is precisely the set of infinitely
large elements. However, Exercise X.X shows in particular that the set of infinitely
large elements has no least element: if x is infinitely large, so is x − 1. 
160 14. ORDERED FIELDS

Famously, R satisfies the least upper bound axiom, i.e., its ordering is Dedekind
complete. So by Proposition 14.49 the ordering on R is Archimedean. (Probably
the reader was not in doubt of this, but this is an especially clean approach.) Thus
every subfield of R is Archimedean.

The order topology: let (S, ≤) be any linearly ordered space. Recall that we can
use the ordering to endow S with a topology, the order topology, in which a base
of open sets consists of all open intervals.6 Order topologies have several pleasant
properties: for instance, any order topology is a hereditarily normal space (i.e.,
every subspace is normal: for us, this includes Hausdorff).
Proposition 14.50. Let K be an ordered field. Then the order topology endows
K with the structure of a topological field. That is, the addition and multiplication
operations are continuous as functions from K × K to K.
Exercise 14.31. Prove Proposition 14.50. (Suggestion: use the characteriza-
tion of continuous functions as those which preserve limits of nets.)
Proposition 14.51. For any Archimedean ordered field F , Q is dense in the
order topology on F .
Proof. It is sufficient to show that for a, b ∈ F with 0 < a < b, there exists
x ∈ Q with a < x < b. Because of the nonexistence of infinitesimals, there exist
x1 , x2 ∈ Q with 0 < x1 < a and 0 < x2 < b−a. Thus 0 < x1 +x2 < b. Therefore the
set S = {n ∈ Z+ | x1 + nx2 < b} is nonempty. By the Archimedean property S is
finite, so let N be the largest element of S. Thus x1 + N x2 < b. Moreover we must
have a < x1 + N x2 , for if x1 + N x2 ≤ a, then x1 + (N + 1)x2 = (x1 + N x2 ) + x2 <
a + (b − a) = b, contradicting the definition of N . 

Exercise 14.32. Deduce from Proposition 14.51 that the order topology on any
Archimedean ordered field is second countable. (Hint: show in particular that open
intervals with rational endpoints form a base for the topology.) From the normality
of all order topologies cited above and Urysohn’s Metrization Theorem, it follows
that the order topology on an Archimedean ordered field is metrizable.7
The order topology on K endows (K, +) with the structure of a commutative
topological group. In such a situation we can define Cauchy nets, as follows: a
net x• : I → G in a commutative topological group G is Cauchy if for each
neighborhood U of the identity 0 ∈ G there exists i ∈ I such that for all j, k ≥ i,
xj − xk ∈ U . A topological group is complete if every Cauchy net converges.
Let F be an ordered field. We define the absolute value function from F to
F ≥0 , of course taking |x| to be x if x ≥ 0 and −x otherwise.
Exercise 14.33. Let F be an ordered field. Show that the triangle inequality
holds: for all x, y ∈ F , |x + y| ≤ |x| + |y|.

6If there is a bottom element b of S, then the intervals [b, b) are deemed open. If there is a
top element t of S, then the intervals (a, t] are deemed open. Of course, no ordering on a field has
either top or bottom elements, so this is not a relevant concern at present.
7However, we are not going to use this fact in our discussion. Rather, as will become clear,
an ordered field K comes with a canonical “K-valued metric”, which will be just as useful to us
as an “R-valued metric” – a special case!
 8. ARCHIMEDEAN AND COMPLETE FIELDS 161

Thus for any ordered field F , one can define the function ρ : F × F → F ≥0 by
ρ(x, y) = |x − y| and this has all the formal properties of a metric except that it
is F -valued. In particular, for any net x• in F we have x• → x iff |x• − x| → 0.
In general it can be of some use to consider “F -valued metrics” where F is a
non-Archimedean ordered field. But here is the key point: if the ordering on F
is Archimedean, then the convergence can be expressed by inequalities involving
rational numbers (rather than the infinitesimal elements that would be required
in the non-Archimedean case): namely, for an Archimedean ordered field F , a net
x• : I → F converges to x ∈ F iff for all n ∈ Z+ , there exists in ∈ I such that
j ≥ i =⇒ |xj − x| < n1 . Topologically speaking, we are exploiting the fact that
the topology of an Archimedean ordered field has a countable neighborhood base
at each point. Thus it is sufficient to replace nets by sequences. In particular we
have the following simple but important result.
Lemma 14.52. Let K be an Archimedean ordered field. Then the following are
equivalent:
(i) Every Cauchy net in K is convergent.
(ii) Every Cauchy sequence in K is convegent.
Proof. Of course (i) =⇒ (ii). Now suppose that every Cauchy sequence in
K converges, and let x• : I → K be a Cauchy net. We may assume that I has no
maximal element, for otherwise the net is certainly convergent. Choose i1 ∈ I such
that j, k ≥ i1 implies |xj − xk | < 1. Now pick i2 ∈ I such that i2 > I1 and j, k ≥ i2
implies |xj − xk | < 12 . Continuing in this manner we get an increasing sequence
{in } in I such that for all n, if j, j ≥ in , |xj − xk | < n1 . Thus from the net we have
extracted a Cauchy subsequence, which by hypothesis converges, say to x. From
this it follows immediately that the net x• converges to x. 
Remark: The proof here is based on [Wi, Thm. 39.4], which asserts that the
uniform structure associated to a complete metric is a complete uniform structure
iff the metric is a complete metric.
Theorem 14.53. For an Archimedean ordered field K, the following are equiv-
alent:
(i) The ordering on K is Dedekind complete: every nonempty subset which
is bounded below has a greatest lower bound.
(ii) (K, +) is a Cauchy-complete topological group: every Cauchy net con-
verges.
Proof. (i) =⇒ (ii): Dedekind complete implies Archimedean implies second
countable implies first countable implies it is enough to look at Cauchy sequences.
The argument is then the usual one from elementary real analysis: suppose K is
Dedekind complete, and let xn be a Cauchy sequence in K. Then the sequence is
bounded, so there exists a least upper bound x. We can construct a subsequence
converging to x in the usual way: for all k ∈ Z+ , let xnk be such that |xnk − x| < n1 .
(That this implies that the subseuence converges is using the the Archimedean
property that for all x > 0, there exists n ∈ Z+ with n1 < x.) Then, as usual, a
Cauchy sequence with a convergent subsequence must itself be convergent.
(ii) =⇒ (i): let S ⊂ K be nonempty and bounded below. Let B be the set of all
lower bounds of S, with the ordering induced from K. What we want to show is
that B has a greatest element: we will prove this by Zorn’s Lemma. Let C be a
162 14. ORDERED FIELDS

nonempty chain in B. We may view this as a net x : C → K. We claim that it is

Cauchy: i.e., for every open neighborhood U of 0, there exists an index i such that
for all j, k ≥ i, xi − xj ∈ U . Because the ordering is Archimedean, this is equivalent
to |xi − xj | <  for some positive rational number . But since C is a set of lower
bounds for the nonempty set S, it is certainly bounded above, and if the desired
conclusion were false there would exist infinitely many pairs of indices (i, j) with
j > i and xj − xi ≥ , and by the Archimedean nature of the ordering this would
imply that C is unbounded above, contradiction! Therefore the net x• is Cauchy
and converges by assumption to x ∈ K. This element x is an upper bound for C
and a lower bound for S. Thus by Zorn’s Lemma B has a maximal element, i.e., S
has a greatest lower bound. 
An Archimedean ordered field satisfying the equivalent conditions of Theorem 14.53
will simply be said to be complete.
Proposition 14.54 (Strong Rigidity for Archimedean Ordered Fields). Let K
be an Archimedean ordered field and let f : K → K be an order-preserving field
homomorphism. Then f = 1K is the identity map.
Proof. Suppose not, and let x ∈ K be such that f (x) 6= x. Without loss
of generality we may suppose that x < f (x), and then by Proposition 14.51 there
exists q ∈ Q with x < q < f (x). Applying the isotone map f gives f (x) < f (q) = q,
a contradiction! 
Lemma 14.55. Let R and S be topological rings and D a dense subring of R.
Suppose that f : R → S is a continuous set map from R to S which upon restriction
to D is a homomorphism of rings. Then f is itself a homomorphism of rings.
Exercise 14.34. Prove Lemma 14.55. (Hint: use the net-theoretic character-
ization of dense subspaces: for any x ∈ R, there exists a net x• : I → D which
converges to x.)
Theorem 14.56 (Main Theorem on Archimedean Ordered Fields).
A complete Archimedean field R is a final object in the category of fields. That is:
(i) For any Archimedean field K and Dedekind complete field R, there exists
a unique embedding of ordered fields K ,→ R.
(ii) Any two Dedekind complete fields are canonically – even uniquely! – iso-
morphic.
Proof. (i) The idea here is that we have copies of Q inside both K and L and
that in an Archimedean ordered field an element is uniquely specified by all of its
order relations with elements of Q. Formally, we define a map ϕ : K → L as follows:
we map x to sup{q ∈ Q | q < x}. As above, it is clear that ϕ is order-preserving.
When restricted to the dense subring Q it is certainly a homomorphism, so in order
to apply Lemma 14.55 we need only check that ϕ is continuous. But again, a base
for the topology of any Archimedean field is given by open intervals (a, b) with
a, b ∈ Q. Evidently ϕ maps the interval (a, b) of K to the interval (a, b) of L, so it
is therefore continuous: done.
(ii) Let R1 and R2 be complete Archimedean fields. By (i), there exist embeddings
of ordered fields ϕ : R1 → R2 and $ : R2 → R1 . Applying Proposition 14.54 to
the endomorphisms $ ◦ ϕ and $ ◦ ϕ, we get $ ◦ ϕ = 1R1 and ϕ ◦ $ = 1R2 , thus $
and ϕ are mutually inverse isomorphisms: so R1 ∼ = R2 as ordered fields. Moreover
 9. THE REAL SPECTRUM 163

the same argument applies to show that any two isomorphisms ϕ1 , ϕ2 from R1 to
R2 are inverses of the isomorphism $, so ϕ1 = ϕ2 : there is only one isomorphism
from R1 to R2 . 
We have already identified the real numbers R as a complete Archimedean field, so
we know that the final object referred to in Theorem 14.56 indeed exists. Let us
restate things in a more concrete fashion using R.
Corollary 14.57. For any Archimedean ordered field K, there is a unique
embedding of ordered fields K ,→ R. Thus we may identify the Archimedean ordered
fields – up to unique isomorphism – as subfields of R with the inherited ordering.
It may be worth asking at this point: exactly how do we know that this field “of
real numbers” we’ve heard so much about actually exists? We’ve proven some fairly
remarkable facts about it: maybe rumors of its existence are greatly exaggerated!

The previous paragraph is silly. A rigorous construction of R was first given by

R. Dedekind in the late 19th century. Accounts of his method (using what are
now called) “Dedekind cuts” may be found in many texts. However, our Cauchy-
theoretic perspective also gives an easy answer to this question. Namely, one has the
notion of Cauchy completion of any commutative topological group G: namely,
given G there exists a complete topological group Ĝ and a homomorphism of topo-
logical groups G → Ĝ which is universal for homomorphisms from G into a complete
topological group (If G is Hausdorff the map to the completion is an embedding.)
The construction can be given in terms of an equivalence relation on the class of
Cauchy nets on G, for instance. Moreover, when G is the additive group of an
ordered field F , it is not hard to show that F̂ is also an ordered field. Note well
that we can therefore construct many Cauchy complete non-Archimedean or-
dered fields. However what we want is a Dedekind complete ordered field, and for
this, according to Theorem 14.53 it is sufficient – and clearly also necessary – to
complete an Archimedean ordered field, like Q.

The construction of the Cauchy completion of a commutative topological group

is more abstruse than is necessary for this application, though. As in Lemma 14.52
above, we can get away with Cauchy sequences rather than Cauchy nets. Thus
we may construct R from Q in the following appealingly algebraic way: take the
ring C(Q) of all Cauchy sequences in Q and mod out by the maximal ideal c0 of
sequences converging to 0. Therefore the quotient is a Cauchy complete field, say
R. It is easy to check that the ordering on Q extends to R and that Q is dense in R
in the order topology, which implies that the ordering on R is Archimedean. Thus
R is a Cauchy complete, Archimedean ordered field, so it is Dedekind complete.

9. The Real Spectrum

For a field F , let X(F ) be the set of all orderings on F . There is a natural topology
on X(F ): namely the open sets are given by finite intersections of (subbasic) open
sets of the form
H(a) = {P ∈ X(F ) | a ∈ P }
as a ranges through nonzero elements of F : that is, H(a) is the set of orderings
which regard a as positive. Note that H(−a) = X(F ) \ H(a), so that the H(a)
164 14. ORDERED FIELDS

and (and hence also all the basis elements) are closed as well open: this implies
that X(F ) is totally disconnected and Hausdorff. It is also compact. To see this,
×
note that an ordering P of F gives rise to an element of Y = {±1}F , namely for
each nonzero element a, we assign +1 if a ∈ P and −1 if −a ∈ P . Giving {±1}
the discrete topology and Y the product topology, it is a compact Hausdorff totally
disconnected space by Tychonoff’s theorem. It remains to be shown first that the
topology on X(F ) defined above is the same as the topology it gets as a subspace
of Y 8, and second that X(F ) is closed as a subspace of Y . Neither of these is very
difficult and we leave them to the reader.

If F1 ,→ F2 is a field embedding, then the aforementioned process of restricting

orders on F2 to orders on F1 gives a map X(F2 ) → X(F1 ) which is easily seen to
be continuous.

A topological space which is compact Hausdorff and totally disconnected is of-

ten called Boolean, since these are precisely the spaces which arise as spectra of
maximal ideals of Boolean algebras.
Theorem 14.58. (Craven [Cr75]) Any Boolean space X is homeomorphic to
X(F ) for some field F .
The following exercises develop a proof in the special case in which X is second
countable.
Exercise 14.35. Let F = lim Fα be a direct limit (i.e., directed union) of
−→ α
fields. Show X(F ) = lim X(Fα ) as topological spaces.
←− α

Exercise 14.36. Let F/Q be a (possibly infinite) formally real Galois exten-
sion. Show that Aut(F ) = Gal(F/Q) acts continuously and simply transitively on
X(F ), and conclude that in this case X(F ) is homeomorphic to the underlying
topological space of a profinite group. In particular, if F/Q is infinite, X(F ) is
an infinite profinite space without isolated points and with a countable basis, so is
√
homeomorphic to the Cantor set. (A good example is F = Q({ p}) as p ranges
over all the prime numbers: here Aut(F ) ∼ = (Z/2Z)ℵ0 really looks like the Cantor
set.)
Exercise 14.37. Use weak approximation of valuations to show that any in-
verse system
. . . → Sn+1 → Sn → . . . → S1
of finite sets can be realized as the system of X(Fn )’s where
F1 . . . ,→ Fn ,→ Fn+1 ,→ . . .
is a tower of number fields. Conclude that any profinite space with a countable basis
arises as the space of orderings of an algebraic field extension of Q.

8One might wonder why we didn’t save ourselves the trouble and define the topology on X(F )
in this latter way. It turns out that the sets H(a), called the Harrison subbasis, are important in
their own right.
 Bibliography

[Ar44] E. Artin, Galois Theory. Second edition. Notre Dame Mathematical Lectures, no. 2.
University of Notre Dame, Notre Dame, Ind., 1944.
[BAI] N. Jacobson, Basic algebra. I. Second edition. W. H. Freeman and Company, New York,
1985.
[BAII] N. Jacobson, Basic algebra. II. Second edition. W. H. Freeman and Company, New
York, 1989.
[Bar51] D. Barbilian, Solution exhaustive du problème de Steinitz. Acad. Repub. Pop. Romn̂e.
Stud. Cerc. Mat. 2, (1951). 195–259 (misprinted 189–253).
[BJ01] F. Borceux and G. Janelidze, Galois theories. Cambridge Studies in Advanced Mathe-
matics, 72. Cambridge University Press, Cambridge, 2001.
[CG72] C.H. Clemens and P.A. Griffiths, The intermediate Jacobian of the cubic threefold.
Ann. of Math. (2) 95 (1972), 281–356.
[Ch] P.M. Cohn, Basic algebra. Groups, rings and fields. Springer-Verlag London, Ltd.,
London, 2003.
[Ch70] A. Charnow, The automorphisms of an algebraically closed field. Canad. Math. Bull.
13 (1970), 95–97.
[Cl-CA] P.L. Clark, Commutative Algebra. http://math.uga.edu/~pete/integral2015.pdf
[CL-NT] P.L. Clark, Number Theory: A Contemporary Introduction. http://math.uga.edu/
~pete/4400FULL.pdf
[Cl-NTII] P.L. Clark, Algebraic Number Theory II: Valuations, Local Fields and Adeles. http:
//math.uga.edu/~pete/8410FULL.pdf
[Cl-QF] P.L. Clark, Lecture notes on quadratic forms. http://alpha.math.uga.edu/~pete/
quadraticforms0.pdf
[Cr75] T.C. Craven, The Boolean space of orderings of a field. Trans. Amer. Math. Soc. 209
(1975), 225–235.
[CW50] J.W.S. Cassels and G.E. Wall, The normal basis theorem. J. London Math. Soc. 25
(1950), 259–264.
[DG94] M. Dugas and R. Göbel, Automorphism groups of fields. Manuscripta Math. 85 (1994),
no. 3-4, 227–242.
[DG97] M. Dugas and R. Göbel, Automorphism groups of fields. II. Comm. Algebra 25 (1997),
3777–3785.
[Di74] J. Dieudonné, Sur les automorphismes des corps algébriquement clos. Bol. Soc. Brasil.
Mat. 5 (1974), 123–126.
[FK78] E. Fried and J. Kollar, Automorphism groups of algebraic number fields. Math. Z. 163
(1978), 121–123.
[Gi68] R. Gilmer, Classroom Notes: A Note on the Algebraic Closure of a Field. Amer. Math.
Monthly 75 (1968), 1101—1102.
[Hö01] O. Hölder, Die Axiome der Quantitat und die Lehre vom Mass. Ber. Verh. Sachs. Ges.
Wiss. Leipzig, Math.-Phys. Cl. 53 (1901), 1–64.
[Ho20] X.-D. Hou, PGL2 (Fq ) acting on Fq (x). Comm. Algebra 48 (2020), 1640–1649.
[Is80] I.M. Isaacs, Roots of polynomials in algebraic extensions of fields. Amer. Math. Monthly
87 (1980), 543–544.
[Ja1] N. Jacobson, Basic algebra. I. Second edition. W. H. Freeman and Company, New York,
1985.
[Ja2] N. Jacobson, Basic algebra. II. Second edition. W. H. Freeman and Company, New
York, 1989.

165
166 BIBLIOGRAPHY

[Je] T.J. Jech, The axiom of choice. Studies in Logic and the Foundations of Mathematics,
Vol. 75. North-Holland Publishing Co., New York, 1973.
[Kap95] I. Kaplansky, Fields and rings. Reprint of the second (1972) edition. Chicago Lectures
in Mathematics. University of Chicago Press, Chicago, IL, 1995.
[Ka89] G. Karpilovsky, Topics in field theory. North-Holland Mathematics Studies, 155. Notas
de Matematica [Mathematical Notes], 124. North-Holland Publishing Co., Amsterdam,
1989.
[Kr53] W. Krull, Über eine Verallgemeinerung des Normalkörperbegriffs. J. Reine Angew.
Math. 191 (1953), 54–63.
[Ku65] W. Kuyk, The construction of fields with infinite cyclic automorphism group. Canad.
J. Math. 17 (1965), 665–668.
[La53] S. Lang, The theory of real places. Ann. of Math. (2) 57 (1953), 378–391.
[LaFT] S. Lang, Algebra. Revised third edition. Graduate Texts in Mathematics, 211. Springer-
Verlag, New York, 2002.
[Las97] D. Lascar, The group of automorphisms of the field of complex numbers leaving fixed
the algebraic numbers is simple. Model theory of groups and automorphism groups
(Balubeuren, 1995), 110-114, London Math. Soc. Lecture Note Ser., 244, Cambridge
Univ. Press, Cambridge, 1997.
[Le55] H. Leptin, Ein Darstellungssatz für kompakte, total unzusammenhngende Gruppen.
Arch. Math. (Basel) 6 (1955), 371–373.
[Lev43] F.W. Levi, Contributions to the theory of ordered groups. Proc. Indian Acad. Sci., Sect.
A. 17 (1943), 199–201.
[LoI] F. Lorenz, Algebra. Vol. I. Fields and Galois theory. Translated from the 1987 German
edition by Silvio Levy. With the collaboration of Levy. Universitext. Springer, New
York, 2006.
[LoII] F. Lorenz, Algebra. Vol. II. Fields with structure, algebras and advanced topics. Trans-
lated from the German by Silvio Levy. With the collaboration of Levy. Universitext.
Springer, New York, 2008.
[Mac39] S. Mac Lane, Steinitz field towers for modular fields. Trans. Amer. Math. Soc. 46,
(1939). 23–45.
[Mc67] P.J. McCarthy, Maximal fields disjoint from certain sets. Proc. Amer. Math. Soc. 18
(1967), 347–351.
[Mo96] P. Morandi, Field and Galois theory. Graduate Texts in Mathematics, 167. Springer-
Verlag, New York, 1996.
[Mu61] D. Mumford, Pathologies of modular algebraic surfaces. Amer. J. Math. 83 (1961),
339–342.
[Mu62] D. Mumford, Further pathologies in algebraic geometry. Amer. J. Math. 84 (1962),
642–648.
[Mu67] D. Mumford, Pathologies. III. Amer. J. Math. 89 (1967), 94–104.
[Mu75] D. Mumford, Pathologies IV. Amer. J. Math. 97 (1975), 847–849.
[Pa74] T. Parker, Some applications of Galois theory to normal polynomials. Amer. Math.
Monthly 81 (1974), 1009–1011.
[Pi09] D. Pierce, https://dialinf.wordpress.com/2009/08/28/
completions-and-the-archimedean-property/
[Qu62] F. Quigley, Maximal subfields of an algebraically closed field not containing a given
element. Proc. Amer. Math. Soc. 13 (1962), 562–566.
[Rom06] S. Roman, Field theory. Second edition. Graduate Texts in Mathematics, 158. Springer,
New York, 2006.
[Rot98] J. Rotman, Galois theory. Second edition. Universitext. Springer-Verlag, New York,
1998.
[Sc92] B. Schnor, Involutions in the group of automorphisms of an algebraically closed field.
J. Algebra 152 (1992), 520–524.
[Si] J.H. Silverman, The arithmetic of elliptic curves. Second edition. Graduate Texts in
Mathematics, 106. Springer, Dordrecht, 2009.
[Sp52] T.A. Springer, Sur les formes quadratiques d’indice zéro. C. R. Acad. Sci. Paris 234
(1952), 1517–1519.
[St10] E. Steinitz, Algebrasiche Theorie der Körper. Journal für die reine und angewandte
Mathematik, 1910.
 BIBLIOGRAPHY 167

[Su99] B. Sury, On an example of Jacobson. Amer. Math. Monthly 106 (1999), 675-–676.
[Wa74] W.C. Waterhouse, Profinite groups are Galois groups. Proc. Amer. Math. Soc. 42
(1974), 639–640.
[Wa94] W.C. Waterhouse, The normal closures of certain Kummer extensions. Canad. Math.
Bull. 37 (1994), 133–139.
[Wa04] W.C. Waterhouse, Intersections of two cofinite subfields. Arch. Math. (Basel) 82 (2004),
298–300.
[We09] S.H. Weintraub, Galois theory. Second edition. Universitext. Springer, New York, 2009.
[Wi] S. Willard, General topology. Reprint of the 1970 original. Dover Publications, Inc.,
Mineola, NY, 2004.
[Ya66] P.B. Yale, Automorphisms of the Complex Numbers. Math. Magazine 39 (1966), 135–
141.
[Za58] O. Zariski, On Castelnuovo’s criterion of rationality pa = P2 = 0 of an algebraic
surface. Illinois J. Math. 2 (1958), 303–315.
[ZSI] O. Zariski and P. Samuel, Commutative algebra. Vol. 1. With the cooperation of I. S.
Cohen. Corrected reprinting of the 1958 edition. Graduate Texts in Mathematics, No.
28. Springer-Verlag, New York-Heidelberg-Berlin, 1975.