Bourbaki

§ 14. Transcendental extensions

1. Algebraically free families. Pure extensions

Let us recall (IV, p. 4) the following definition:

Definition 1

Let E be an extension of a field K; a family x=(xi)i∈I\mathbf{x} = (x_i)_{i \in I} of elements of E is said to be algebraically free over K if the monomials xa=∏i∈Ixiai\mathbf{x}^a = \prod_{i \in I} x_i^{a_i} with respect to the xix_i (for a=(αi)i∈Ia = (\alpha_i)_{i \in I}, in N(I)\mathbf{N}^{(I)}) are linearly independent over K. In the contrary case the family is said to be algebraically related over K.

Definition 1 may also be expressed as follows:

Proposition 1

For a family (xi)i∈I(x_i)_{i \in I} of elements of an extension E of a field K to be algebraically free over K it is necessary and sufficient for the relation f((xi))=0f((x_i)) = 0, where ff is a polynomial in K[Xi]i∈IK[X_i]_{i \in I}, to imply f=0f = 0.

Definition 2

Let E be an extension of a field K. A family (xi)i∈I(x_i)_{i \in I} of elements of E is called a pure basis of E (over K) if it is algebraically free and E=K(xi)i∈IE = K(x_i)_{i \in I}. The extension E of K is also called pure if it possesses a pure basis.

The empty family is algebraically free, hence K is a pure extension of itself. With the notation of Def. 2, every element xix_i is transcendental over K; if II is non-empty, E is thus a transcendental extension of K.

Proposition 2

Let E and E′E' be two fields and u an isomorphism of a subfield K of E onto a subfield K′K' of E′E'. Let x=(xi)i∈I\mathbf{x} = (x_i)_{i \in I} (resp. x′=(xi′)i∈I\mathbf{x}' = (x'_i)_{i \in I}) be a family of elements of E (resp. E′E') which is algebraically free over K (resp. K′K'). Then there exists a unique isomorphism vv of L=K(xi)i∈IL = K(x_i)_{i \in I} onto L′=K′(xi′)i∈IL' = K'(x'_i)_{i \in I}, which induces uu on KK and maps xix_i to xi′x'_i for each i∈Ii \in I.

The uniqueness of vv is clear. Let us put A=K[xi]i∈IA = K[x_i]_{i \in I} and A′=K′[xi′]i∈IA' = K'[x'_i]_{i \in I}. By hypothesis the monomials x′′=∏i∈Ixiαix'' = \prod_{i \in I} x_i^{\alpha_i} (for α=(αi)i∈I\alpha = (\alpha_i)_{i \in I}, in N(I)\mathbf{N}^{(I)}) form a basis of the vector K-space AA and there is a similar property of A′A'. Hence there exists a ring isomorphism w:A→A′w : A \to A' mapping each element ∑α∈N(I)cαxα\sum_{\alpha \in \mathbf{N}^{(I)}} c_\alpha x^\alpha to ∑α∈N(I)u(cα)x′α\sum_{\alpha \in \mathbf{N}^{(I)}} u(c_\alpha) {x'}^\alpha. Since LL is the field of fractions of AA and L′L' that of A′A', the isomorphism ww extends to an isomorphism vv of the field LL onto the field L′L'.

Corollary

— *For an extension E of a field K to be pure it is necessary and sufficient for E to be K-isomorphic to a field of rational fractions over K. More precisely, if the family (xi)i∈I(x_i)_{i \in I} is a pure basis of E, then there exists a unique K-isomorphism of K(Xi)i∈IK(X_i)_{i \in I} onto E which maps XiX_i to xix_i for each i∈Ii \in I.

Remark

It is clear that in an extension E of K an algebraically free family over K consists of linearly independent elements over KK (hence pairwise distinct); in other words, it is also a free family for the vector space structure of EE (with respect to K). But the converse is false, for if E is an algebraic extension of K, any non-empty family of elements of E (and a fortiori a non-empty family of linearly independent elements over K) is never algebraically free over K. When there is a risk of confusion, we shall say that a subset of an extension E of K which is free for the vector space structure of E with respect to K is linearly free over K.

Let E be an extension of a field K. A subset S of E is said to be algebraically free (over KK) if the family defined by the identity mapping of S onto itself is algebraically free. The elements of an algebraically free subset of E are also called algebraically independent. If a subset of E is not algebraically free, it is said to be algebraically related and that its elements are algebraically dependent. For a family (xi)i∈I(x_i)_{i \in I} of elements of E to be algebraically free it is necessary and sufficient that i↦xii \mapsto x_i should be a bijection of I onto an algebraically free subset of E.

Every subset of an algebraically free set is algebraically free. Furthermore:

Proposition 3

— *For a family (xi)i∈I(x_i)_{i \in I}, of elements of an extension E of a field K to be algebraically free over K it is necessary and sufficient that every finite subfamily of (xi)i∈I(x_i)_{i \in I} should be algebraically free over K.

This proposition follows immediately from Def. 1.

2. Transcendence bases

Proposition 4

— *Let E be an extension of a field K and S and T two subsets of E. Then the following properties are equivalent: a) S∪TS \cup T is algebraically free over K and S∩T=∅S \cap T = \varnothing.

b) SS is algebraically free over KK and TT is algebraically free over K(S)K(S). c) TT is algebraically free over KK and SS is algebraically free over K(T)K(T).

Evidently it is enough to prove that a) and b) are equivalent.

a)⇒b)a) \Rightarrow b): Suppose that a) holds. Since SS is contained in S∪TS \cup T, it is algebraically free over KK. If TT is not algebraically free over K(S)K(S), there exists (Prop. 3) a finite family (yj)1≤j≤n(y_j)_{1 \leq j \leq n} of distinct elements of TT which is algebraically related over K(S)K(S). Hence there is a non-zero polynomial ff in the ring K(S)[Y1,...,Yn]K(S)[Y_1, ..., Y_n] such that f(y1,...,yn)=0f(y_1, ..., y_n) = 0; after multiplying ff if necessary by a non-zero element of K[S]K[S] we may suppose that all the coefficients off belong to K[S]K[S]. The coefficients of fare polynomials in a finite number of distinct elements xi (1≤i≤m)x_i \ (1 \leq i \leq m ) of SS, with coefficients in KK. The elements x1,...,xm,y1,...,ynx_1, ..., x_m, y_1, ..., y_n are pairwise distinct because S∩T=(21S \cap T = (21. The relation f(y1,...,yn)=0f(y_1, ..., y_n) = 0 may thus be written

g(x1,...,xm;y1,...,yn)=0, g(x_1, ..., x_m ; y_1, ..., y_n) = 0 ,

where gg is a non-zero polynomial of K[X1,...,Xm,Y1,...,Yn]K[X_1, ..., X_m, Y_1, ..., Y_n], and such a relation contradicts the hypothesis that S∪TS \cup T is algebraically free.

b)⇒a)b) \Rightarrow a): Suppose that b) holds. In the first place it is clear that T∩K(S)=∅T \cap K(S) = \varnothing and afortiori S∩T=(21a fortiori \ S \cap T = (21. It suffices to show that if xi (1≤i≤m)x_i \ (1 \leq i \leq m ) are distinct elements of SS, finite in number, and yj (1≤j≤n)y_j \ (1 \leq j \leq n ) distinct elements of TT finite in number, then the set of the xix_i and yjy_j is algebraically free over KK (Prop. 3). Consider a polynomial f∈K[X1,...,Xm,Y1,...,Yn]f \in K[X_1, ..., X_m, Y_1, ..., Y_n] such that f(x1,...,xm,y1,...,yn)=0f(x_1, ..., x_m, y_1, ..., y_n) = 0 and put f=∑φαY1α1...Ynαnf = \sum \varphi_\alpha Y_1^{\alpha_1} ... Y_n^{\alpha_n} with φα∈K[X1,...,Xm]\varphi_\alpha \in K[X_1, ..., X_m] for all α=(α1,...,αn)∈Nn\alpha = (\alpha_1, ..., \alpha_n) \in \mathbf{N}^n. Let g=f(x1,...,xm,Y1,...,Yn)g = f(x_1, ..., x_m, Y_1, ..., Y_n); then gg is a polynomial in the ring K[S][Y1,...,Yn]K[S][Y_1, ..., Y_n] and the relation f(x1,...,xm,y1,...,yn)=0f(x_1, ..., x_m, y_1, ..., y_n) = 0 may be written g(y1,...,yn)=0g(y_1, ..., y_n) = 0. Since TT is algebraically free over K(S)K(S), each of the coefficients φα(x1,...,xm)\varphi_\alpha(x_1, ..., x_m) of gg is zero; since SS is algebraically free over KK, we have φα=0\varphi_\alpha = 0 for all α∈Nn\alpha \in \mathbf{N}^n, and hence f=0f = 0.

Corollary

— *Let E be an extension of a field K and S a subset of E which is algebraically free over K. If x \in E is transcendental over K(S), then S U {x} is algebraically free over K.

Proposition 5

— *Let E be an extension of the field K. For a subset S of E to be algebraically free over K it is necessary and sufficient that for each x \in S, the element x should be transcendental over the field K(S - {x}).

The condition is necessary by Prop. 4.

To prove the sufficiency it is enough (Prop. 3) to show that every finite sequence (x1,...,xn)(x_1, ..., x_n) of distinct elements of SS is algebraically free. Now by hypothesis xix_i is transcendental over K(x1,...,xi−1)K(x_1, ..., x_{i-1}) for 1≤i≤n1 \leq i \leq n, and now our assertion follows by induction on nn from the Cor. to Prop. 4.

Proposition 6

— *Let E be an extension of a field K and X an algebraically free subset of E over K. If K' \subset E is an algebraic extension of K then X is algebraically free over K'.

Arguing by contradiction, let us suppose that X is algebraically related over K'. By Prop. 5 there exists an element x∈Xx \in X which is algebraic over the field K′(M)K'(M), where M=X−{x}M = X - \{x\}. Since K′(M)=K(M)(K′)K'(M) = K(M)(K') and K′K' is algebraic over KK, Cor. 2 of V, p. 18 shows that K′(M)K'(M) is algebraic over K(M)K(M); since xx is algebraic over K′(M)K'(M), it is therefore algebraic over K(M)=K(X−{x})K(M) = K(X - \{x\}), by Prop. 3 of V, p. 19. Now Prop. 5 shows X to be algebraically related over K and we have a contradiction.

Definition 3

A subset B of an extension E of a field K is a transcendence basis of E (over K) if B is algebraically free over K and E is algebraic over K(B).

Example

— *A pure basis is a transcendence basis. On the other hand, if E is a pure extension of K, a transcendence basis of E over K is not always a pure basis of E. For example, in K(X)K(X), {X2}\{X^2\} is a transcendence basis but does not generate K(X)K(X).

Proposition 7

Let E be an extension of a field K. Every transcendence basis of E is a maximal element of the set (ordered by inclusion) of subsets of E algebraically free over K. Conversely, if S is a subset of E such that E is algebraic over K(S)K(S), every maximal algebraically free subset of S is a transcendence basis of E.

Let B be a transcendence basis of E over K and x∈E−Bx \in E - B. Then xx is algebraic over K(B)K(B); by V, p. 107, Prop. 4, the subset B∪{x}B \cup \{x\} of E is not algebraically free over K, whence the first part of the proposition. Secondly, if E is algebraic over K(S)K(S) and B is a maximal algebraically free part of S, it follows from the Cor. of Prop. 4 that every x∈Sx \in S is algebraic over K(B)K(B); hence (V, p. 18, Cor. 1) K(S)K(S) is algebraic over K(B)K(B), and so (V, p. 19, Prop. 3), E is algebraic over K(B)K(B).

Theorem 1 (Steinitz)

— *Every extension E of a field K admits a transcendence basis over K. In other words, every extension of a field K is an algebraic extension of a pure extension of K.

By contrast, an extension is not always a pure extension of an algebraic extension (V, p. 171, Ex. 2).

This theorem is a consequence of the following more precise result:

Theorem 2

Let E be an extension of a field K, S a subset of E such that E is algebraic over K(S)K(S) and T a subset of S which is algebraically free over K; then there exists a transcendence basis B of E over K such that T⊂B⊂ST \subset B \subset S.

For the set of algebraically free subsets of S, ordered by inclusion, is a set of finite character (E, III, p. 34) by Prop. 3. By Th. 1 of E, III, p. 35 it has a maximal element B containing T, and B is a transcendence basis of E over K, by Prop. 7.

Corollary (« Exchange Theorem »)

— Let E be an extension of K, S a subset of E such that E is algebraic over K(S), T a subset of E which is algebraically free over K; then there exists a subset S' of S such that T ∪ S' is a transcendence basis of E over K and T ∩ S' = ∅.

For E is algebraic over K(T U S) and we have T ⊂ T U S.

3. The transcendence degree of an extension

Theorem 3

— Let E be an extension of a field K. All transcendence bases of E over K have the same cardinal.

It is enough to prove the inequality Card(B) ≥ Card(B'), when B and B' are two transcendence bases of E over K, and we may suppose that B' is not empty. We first suppose that B is finite and use induction on its cardinal n; for n = 0, E is algebraic over K and B' is empty. Suppose now that n ≥ 1; given x ∈ B', the exchange theorem provides a subset C of B such that x ∉ C and {x} U C is a transcendence basis of E over K; we have C ≠ B by Prop. 7 whence Card C < n. Put K_1 = K(x) and C' = B' - {x}; then C and C' are algebraically free over the field K_1 (V, p. 107, Prop. 4) and E is algebraic both over K_1(C) = K(C U {x})) and K_1(C') = K(B'). In other words, C and C' are two transcendence bases of E over K_1. Since Card(C) < n, the induction hypothesis implies the inequality Card(C') ≤ Card(C) ≤ n - 1, whence Card(B') ≤ n = Card(B).

Suppose now that B is infinite. Every x ∈ B is algebraic over K(B') and so there exists a finite subset S(x) of B' such that x is algebraic over K(S(x)). Write S = U_{x ∈ B} S(x), then S ⊂ B' and since B is infinite, we have Card(S) ≤ Card(B) (E, III, p. 49, Cor. 3). But since every element of B is algebraic over K(S) and E is algebraic over K(B), we conclude that E is algebraic over K(S) (V, p. 19, Prop. 3). Now Prop. 7 implies that S = B', whence the desired inequality Card(B') ≤ Card(B).

Definition 4

— Let E be an extension of a field K. The cardinal of every transcendence basis of E over K is called the transcendence degree of E over K and written tr. deg_K E.

Th. 2 and 3 and Def. 4 imply the following corollaries, where E denotes an extension of a field K, of finite transcendence degree n.

Corollary 1

— Let S be a subset of E such that E is algebraic over K(S). Then Card(S) ≥ n; if the cardinal of S is equal to n, then S is algebraically free over K (hence it is then a transcendence basis of E over K).

Corollary 2

— Suppose that E = K(x_1, ..., x_m), then m ≥ n; if moreover m = n, then (x_1, ..., x_m) is a pure basis of E over K, and E is then a pure extension of K.

Corollary 3

— Every subset of EE which is algebraically free over KK has at most nn elements, and if it has exactly nn elements, it is a transcendence basis of EE over KK.

Theorem 4

— Let K,EK, E and FF be three fields such that K⊂E⊂FK \subset E \subset F. If SS is a transcendence basis of EE over KK and TT a transcendence basis of FF over EE, then S∩TS \cap T is empty and S∪TS \cup T is a transcendence basis of FF over KK.

For FF is algebraic over E(T)E(T); moreover, E(T)E(T) is algebraic over the field K(S∪T)=K(S)(T)K(S \cup T) = K(S)(T), because EE is algebraic over K(S)K(S) (VV, p. 18, Cor. 2); therefore (VV, p. 19, Prop. 3) FF is algebraic over K(S∪T)K(S \cup T). On the other hand, TT being algebraically free over EE is so afortioria fortiori over K(S)K(S), hence (VV, p. 107, Prop. 4) S∪TS \cup T is algebraically free over KK and S∩T=∅S \cap T = \varnothing.

Corollary

— Let K,EK, E and FF be three fields such that K⊂E⊂FK \subset E \subset F. Then we have

tr. degKF=tr. degKE+tr. degEF \text{tr. deg}_K F = \text{tr. deg}_K E + \text{tr. deg}_E F

4. Extension of isomorphisms

Proposition 8

— Let Ω\Omega be an algebraically closed extension of a field KK, EE and FF two subextensions of Ω\Omega and uu a KK-isomorphism of EE onto FF. For a KK-automorphism vv of OO to exist which extends uu, it is necessary and sufficient for Ω\Omega to have the same transcendence degree over EE and FF.

The condition is clearly necessary.

Suppose then that Ω\Omega has the same transcendence degree over EE and FF and let us choose a transcendence basis BB of Ω\Omega over EE and a transcendence basis CC of Ω\Omega over FF. Since BB and CC are equipotent, Prop. 2 (VV, p. 106) shows that uu extends to a KK-isomorphism u′u' of E(B)E(B) onto F(C)F(C). Since C?C? is an algebraic closure of E(B)E(B) and F(C)F(C), the Cor. of VV, p. 23 shows that u′u' extends to an automorphism vv of Ω\Omega.

Corollary 1

— Let OO be an algebraically closed extension of a field KK and EE a subextension of Ω\Omega. Every KK-automorphism of EE extends to a KK-automorphism of Ω\Omega.

Corollary 2

— Let Ω\Omega be an algebraically closed extension of a field KK, EE and FF two subextensions of Ω\Omega and uu a KK-isomorphism of EE onto FF. If the transcendence degree of EE over KK is finite (in particular if EE is algebraic over KK) then there exists a KK-automorphism of Ω\Omega extending uu.

Let us denote by n,d(E)n, d(E) and d(F)d(F) the transcendence degrees of EE over KK, of OO over EE and of C?C? over FF respectively. The existence of the KK-isomorphism uu shows that the transcendence degree of FF over KK is equal to nn. By the Cor. of Th. 4 the transcendence degree of Ω\Omega over KK is equal to d(E)+nd(E) + n and also to d(F)+nd(F) + n. Therefore (EE, III, p. 28, Prop. 8) we have d(E)=d(F)d(E) = d(F) and we can apply Prop. 8.

Proposition 9

— Let K be a field and Ω\Omega an algebraically closed extension of K. Suppose that Ω\Omega is not algebraic over K; then the set of elements of Ω\Omega which are transcendental over K is infinite. Moreover, if x and y are two elements of Ω\Omega which are transcendental over K, then there exists an automorphism u of Ω\Omega over K such that u(x)=yu(x) = y.

Since Ω\Omega is not algebraic over K, there exists an element x of Ω\Omega transcendental over K; then the elements xnx^n (n∈Nn \in \mathbf{N}) are distinct and transcendental over K. Suppose that x and y are distinct and transcendental over K; by Prop. 2 (V, p. 106) there exists a K-isomorphism u~\tilde{u} of K(x)K(x) onto K(y)K(y) such that u~(x)=y\tilde{u}(x) = y; since K(x)K(x) is of transcendence degree 1 over K, Cor. 2 of Prop. 8 shows that u~\tilde{u} extends to a K-automorphism u of Ω\Omega.

Proposition 10

— Let K be a field, Ω\Omega an algebraically closed extension of K and G the group of K-automorphisms of Ω\Omega. Let x∈Ωx \in \Omega.

a) For x to be algebraic over K it is necessary and sufficient for the set of elements u(x)u(x) as u runs over G to be finite,

b) For x to be p-radical over K it is necessary and sufficient that u(x)=xu(x) = x for all u∈Gu \in G.

In particular if K is perfect, the set of invariants of the group G is equal to K.

Suppose first that x is transcendental. By Prop. 9 the set T of elements of Ω\Omega transcendental over K is infinite, and for each y∈Ty \in T there exists u∈Gu \in G such that u(x)=yu(x) = y. Hence the set of elements u(x)u(x) as u runs over G is infinite.

Suppose now that x is algebraic over K and denote by fits minimal polynomial over K; the set of roots of f in Ω\Omega is finite and for each u∈Gu \in G we have f(u(x))=u(f(x))=0f(u(x)) = u(f(x)) = 0. Hence the set of elements u(x)u(x) as u runs over G is finite. This proves a).

Let L be the set of elements y of Ω\Omega such that u(y)=yu(y) = y for all u∈Gu \in G, and let Kˉ\bar{K} be the relative algebraic closure of K in Ω\Omega. By what has been said, L is a subextension of Kˉ\bar{K} over K. Further (Cor. 1 to Prop. 8) every K-automorphism of Kˉ\bar{K} is the restriction to Kˉ\bar{K} of an element of G. Assertion b) of Prop. 10 now follows from Cor. 3 of V, p. 53.

5. Algebraically disjoint extensions

Definition 5

— Let L be an extension of a field K, and E and F two subextensions of L. Then E and F are said to be algebraically disjoint (over K) and E is said to be algebraically disjoint from F over K, if for every subset A (resp. B) of E (resp. F) algebraically free over K, A and B are disjoint and A∪BA \cup B is algebraically free over K.

Remark 1

If E is a subextension of L which is algebraic over K, then it is algebraically disjoint from every subextension F of L. For an extension of K to be algebraic it is necessary and sufficient that it should be algebraically disjoint from itself.

Remark 2

It may happen that E is algebraically disjoint from F over K, but not over a subfield K∥K_{\|} of K. \* For example C is algebraically disjoint from itself over R but not over Q\mathbf{Q}. \

Remark 3

It is clear that if E is algebraically disjoint from FF over K, when E and F are considered as subextensions of L, the same is true when they are considered as subextensions of K(E∪F)K(E \cup F) and conversely.

Proposition 11

— If E and F are algebraically disjoint over K then E∩FE \cap F is algebraic over K.

This follows from Def. 5.

Proposition 12

— Let L be an extension of a field K and E, F subextensions of L. Then the following conditions are equivalent:

a) E and F are algebraically disjoint;

b) there exists a transcendence basis of E over K which is algebraically free over F;

c) every subset of E which is algebraically free over K is algebraically free over F.

Let us introduce the following conditions:

b′b'') there exists a transcendence basis of F over K which is algebraically free over E;

c′c'') every subset of F which is algebraically free over K is algebraically free over E.

a)⇒b′a) \Rightarrow b'') : Suppose that E and F are algebraically disjoint. Let B (resp. C) be a transcendence basis of E (resp. FF) over K. Then B∩C=∅B \cap C = \varnothing and B∪CB \cup C is algebraically free over K, hence C is algebraically free over K(B)K(B) (V, p. 107, Prop. 4); since E is algebraic over K(B)K(B), Prop. 6 of V, p. 108 shows C to be algebraically free over E.

b′b'') ⇒\Rightarrow c) : Suppose that there exists a transcendence basis C of F over K which is algebraically free over E. Let A be a subset of E algebraically free over K. Then C is algebraically free over K(A)K(A), hence A is algebraically free over K(C)K(C) (V, p. 107, Prop. 4) and therefore over FF (V, p. 108, Prop. 6), because F is algebraic over K(C)K(C).

c) ⇒\Rightarrow a) : This follows at once from Prop. 4 (V, p. 107).

The implications a)⇒b)⇒c′)⇒a)a) \Rightarrow b) \Rightarrow c' ) \Rightarrow a) may be proved in the same way.

Corollary

— Let E and F be algebraically disjoint over K. Let E' be the relative algebraic closure of E in L and F' that of F (V, p. 19). Then E' and F' are algebraically disjoint over K.

Let B be a transcendence basis of E over K; this is also a transcendence basis of E' over K. Since E is algebraically disjoint from F over K, B is algebraically free over F, hence over F' (V, p. 108, Prop. 6); now we can apply Prop. 12.

Proposition 13

— Let L be an extension of a field K and E, F two subextensions of L.

a) We have tr.deg⁡LF(E)≤tr.deg⁡KE\operatorname{tr. deg}_L F(E) \leq \operatorname{tr. deg}_K E. When E and F are algebraically disjoint over K, then every transcendence basis of E over K is a transcendence basis of

F(E) over F and we have tr . deg_F F(E) = tr . deg_K E. Conversely, this equality implies that E and F are algebraically disjoint over K when tr . deg_K E is finite.

b) We have tr . deg_K K(E U F) ≤ tr . deg_K E + tr . deg_K F. When E and F are algebraically disjoint over K, we have tr . deg_K K(E U F) = tr . deg_K E + tr . deg_K F. Conversely, this equality implies that E and F are algebraically disjoint over K when E and F are of finite transcendence degree over K.

a) Let B be a transcendence basis of E over K; then E is algebraic over K(B), and Cor. 2 of V, p. 18 shows that F(E) is algebraic over F(K(B)) = F(B). By Th. 2 (V, p. 109) B contains a transcendence basis of F(E) over F; when E is algebraically disjoint from F over K, B is algebraically free over F (Prop. 12) and this is then a transcendence basis of F(E) over F. The three first assertions of a) follow from this. Suppose now that E is of finite transcendence degree over K, equal to that of F(E) over F; since F(E) is algebraic over F(B) and Card B = tr . deg_F F(E), Cor. 1 of V, p. 110 shows B to be algebraically free over F, so E is algebraically disjoint from F over K (Prop. 12).

b) We have K(E U F) = F(E) and so the Cor. of V, p. 110 implies the equality:

tr . degKK(E∪F)=tr . degFF(E)+tr . degKF. \text{tr . deg}_K K(E \cup F) = \text{tr . deg}_F F(E) + \text{tr . deg}_K F .

Now b) follows at once from a) and this equality.

Proposition 14

— *Let L be an extension of a field K, E and F two subextensions of L and B a transcendence basis of E over K. For E and F to be algebraically disjoint over K it is necessary and sufficient that K(B) and F should be linearly disjoint over K.

For E and F to be algebraically disjoint over K it is necessary and sufficient for B to be algebraically free over F (Prop. 12), that is, for the monomials in the elements of B to be linearly independent over F. Since these monomials form a basis of the vector K-space K[B], it comes to the same to say that K[B] and F are linearly disjoint over K. Finally, since K(B) is the field of fractions of K[B], Prop. 6 of V, p. 14 shows that K[B] and F are linearly disjoint if and only if this is so for K(B) and F.

Corollary 1

— *If E and F are linearly disjoint, then E is algebraically disjoint from F over K. Conversely, if E is a pure extension of K and is algebraically disjoint from F over K, then E and F are linearly disjoint over K.

Corollary 2

— *Every pure extension of K is linearly disjoint from every algebraic extension of K; in particular, K is relatively algebraically closed in every pure extension of K.

6. Algebraically free families of extensions

Definition 6

— Let LL be an extension of a field KK and (Ei)i∈I(E_i)_{i \in I} a family of subextensions of LL. The family (Ei)i∈I(E_i)_{i \in I} is said to be algebraically free if the following condition is satisfied:

(AF) For each i∈Ii \in I let AiA_i be a subset of EiE_i which is algebraically free over KK. Then Ai∩Aj=∅A_i \cap A_j = \varnothing for i≠ji \neq j and ⋃i∈IAi\bigcup_{i \in I} A_i is algebraically free over KK.

Remark

By Prop. 3 (VV, p. 107) it is enough to verify the condition (AF) for finite subsets AiA_i. We thus obtain the following result : if (Ei)i∈I(E_i)_{i \in I} is an algebraically free family, the same is true of (Ei′)i∈I(E'_i)_{i \in I} if Ei′E'_i is a subextension of EiE_i for each i∈Ii \in I; conversely, if every family (Ei′)i∈I(E'_i)_{i \in I}, where Ei′E'_i is a finitely generated subextension of EE for each i∈Ii \in I is algebraically free, then (Ei)i∈I(E_i)_{i \in I} is algebraically free. On the other hand, for (Ei)i∈I(E_i)_{i \in I} to be algebraically free it is necessary and sufficient that (Ei)i∈J(E_i)_{i \in J} should be algebraically free for every finite subset JJ of II. Speaking intuitively we may say that the algebraic independence of extensions is a property « of finite character ».

Proposition 15

— Let (Ei)i∈I(E_i)_{i \in I} be a family of subextensions of a given extension LL of a field KK. Then the following conditions are equivalent :

a) The family (Ei)i∈I(E_i)_{i \in I} is algebraically free.

b) For each i∈Ii \in I the extension EiE_i is algebraically disjoint over KK from the extension FiF_i generated by the EjE_j for j≠ij \neq i.

c) There exists a family (Bi)i∈I(B_i)_{i \in I} of disjoint subsets of LL, such that BiB_i is a transcendence basis of EiE_i over KK for each i∈Ii \in I, and B=⋃i∈IBiB = \bigcup_{i \in I} B_i is algebraically free over KK.

It is clear that a) implies c).

Assuming c), let us choose ii in II; put Ci=⋃j≠iBjC_i = \bigcup_{j \neq i} B_j. For each j≠ij \neq i every element of EjE_j is algebraic over K(Bj)K(B_j) and a fortiori over K(Ci)K(C_i). By Cor. 1 of VV, p. 18, the field FiF_i is therefore algebraic over K(Ci)K(C_i). Further, we have Bi∩Ci=∅B_i \cap C_i = \varnothing and B−Bi∪CiB - B_i \cup C_i is algebraically free over KK; therefore BiB_i is algebraically free over K(Ci)K(C_i) (VV, p. 107, Prop. 4), hence also over FiF_i (which is algebraic over K(Ci)K(C_i)) by Prop. 6 of VV, p. 108. Thus we have proved that EiE_i is algebraically disjoint from FiF_i over KK (VV, p. 113, Prop. 12), hence c) implies b).

Let us now assume b) and prove a). It is enough to show that if i1,…,ini_1, \ldots, i_n are distinct elements of II, then the family of extensions (Ei1,…,Ein)(E_{i_1}, \ldots, E_{i_n}) is algebraically free; we argue by induction on nn, the case n=1n = 1 being trivial. Suppose then that n>1n > 1 and that the family (Ei1,…,Ein−1)(E_{i_1}, \ldots, E_{i_{n-1}}) is algebraically free; for 1≤k≤n1 \leq k \leq n choose a subset AkA_k of EikE_{i_k} algebraically free over KK and put

B = A, U ... U A_{n-1}. By the induction hypothesis the subsets A_1, ..., A_{n-1} are pairwise disjoint and B is algebraically free over K; by b) E_{i_n} is algebraically disjoint from F_i and since B is contained in F_{i_n} we have B \cap A_i = \varnothing and B \cup A_i = A, U ... U A_i is algebraically free over K. We have thus shown that the family (E_{i_1}, ..., E_{i_n}) is algebraically free.

The following proposition generalizes part b) of Prop. 13 (V, p. 113).

Proposition 16

— Let (E_i)_{i \in I} be a family of subextensions of an extension of a field K, and let E be the field generated by ⋃i∈IEi\bigcup_{i \in I} E_i.

a) We have tr . deg_K E \leq \sum_{i \in I} \text{tr . deg}_K E_i, with equality when the family (E_i)_{i \in I} is algebraically free over K.

b) Conversely, assume that tr . deg_K E = \sum_{i \in I} \text{tr . deg}_K E_i and that tr . deg_K E is finite; then the family (E_i)_{i \in I}, is algebraically free over K.

For each i∈Ii \in I let B_i be a transcendence basis of E_i over K and put B=⋃i∈IBiB = \bigcup_{i \in I} B_i. For each i∈Ii \in I, each element of E_i is algebraic over K(B_i), hence over K(B). Now Cor. 1 of V, p. 18 shows that E is algebraic over K(B); by V, p. 109, Th. 2, B thus contains a transcendence basis of E over K. If moreover the family (E_i)_{i \in I} is algebraically free over K, then the B_i are disjoint and the set B is algebraically free over K. This establishes a) (Set Theory, III, p. 160, Cor. of Prop. 4).

Under the hypotheses of b), E is algebraic over K(B) and of finite transcendence degree over K, and we have Card(B) \leq \text{tr . deg}_K E. By Cor. 1 of V, p. 110, B is algebraically free over K and the B_i are disjoint. Now Prop. 15 shows that the family (E_i)_{i \in I} is algebraically free over K.

Before stating the next theorem let us remark that there exist algebraically closed extensions of K of arbitrary transcendence degree, for example an algebraic closure of an appropriate field of rational fractions.

Theorem 5

— Let (E_i)_{i \in I}, be a family of extensions of a field K and \Omega an algebraically closed extension of K. *Suppose that the inequality

tr . degKΩ≥∑i∈Itr . degKEi \text{tr . deg}_K \Omega \geq \sum_{i \in I} \text{tr . deg}_K E_i

holds. Then there exists an algebraically free family (F_i)_{i \in I}, of subextensions of \Omega such that F_i is K-isomorphic to E_i for all i \in I.

For each i∈Ii \in I let B_i be a transcendence basis of E_i over K. Let B be a transcendence basis of \Omega over K. By (2) we have Card B \geq \sum_{i \in I} \text{Card} B_i; hence there exists a family (B'_i)_{i \in I}, of pairwise disjoint subsets of B and bijections u_i : B_i \to B'_i for i \in I. By Prop. 2 of V, p. 106, u, extends to a K-isomorphism v_i of K(B_i) onto K(B'_i); since \Omega is algebraically closed and E_i algebraic over

K(B_i), the Cor. (V, p. 23) shows that viv_i extends to a K-isomorphism of EiE_i onto a subextension FiF_i of Ω\Omega. By construction Bi′B'_i is a transcendence basis of FiF_i over K, and Prop. 15 (V, p. 115) shows that the family (Fi)i∈I(F_i)_{i \in I} of subextensions of Ω\Omega is algebraically free over K.

Corollary 1

— *Let E and R be two extensions of a field K. Suppose that Ω\Omega is algebraically closed, of transcendence degree at least equal to that of E. Then E is K-isomorphic to a subextension of R.

Corollary 2

— *Let R be an algebraically closed field of infinite transcendence degree over its prime subfield. Then every field of the same characteristic as Ω\Omega is the ascending directed union of fields isomorphic to subfields of R.

For each field is the ascending directed union of finitely generated subfields over its prime field, and now it is enough to apply Cor. 1. \* Example. — This applies particularly in characteristic 0 in taking Ω=C\Omega = \mathbf{C} (« Lefschetz' principle »). \*

7. Finitely generated extensions

Proposition 17

— *Let E be an extension of a field K and B a transcendence basis of E over K. For E to be finitely generated over K (V, p. 11, Def. 2) it is necessary and sufficient that B be finite and that the degree [E:K(B)][E : K(B)] be finite.

Suppose that E is finitely generated over K and let S be a finite subset of E such that E=K(S)E = K(S). By Th. 2 (V, p. 109) S contains a transcendence basis B′B' of E over K and this has the same cardinal as B (V, p. 110, Th. 3). Therefore B is finite. Put K′=K(B)K' = K(B); then E is algebraic over K′K' and we have E=K′(S)E = K'(S); since S is finite, Th. 2 of V, p. 18 shows that [E:K′][E : K'] is finite.

Conversely, suppose that B is finite and [E:K(B)][E : K(B)] is finite. If C is a (finite) basis of the vector space E over K(B)K(B), we have E=K(B)(C)=K(B∪C)E = K(B)(C) = K(B \cup C) and so E is a finitely generated extension of K.

Corollary 1

— *Suppose that E is a finitely generated extension of K, and let K′K' be the relative algebraic closure of K in E (V, p. 19). Then K′K' is of finite degree over K.

Let B be a transcendence basis of E over K. By Cor. 2 of V, p. 114, K′K' is linearly disjoint from K(B)K(B) over K, whence [K′:K]=[K′(B):K(B)]≤[E:K(B)][K' : K] = [K'(B) : K(B)] \leq [E : K(B)] and now the finiteness of [K′:K][K' : K] follows from that of [E:K(B)][E : K(B)].

Corollary 2

— *A field that is finitely generated over its prime subfield contains only finitely many roots of unity.

By Cor. 1 we are reduced to proving that a field L which is an extension of finite degree of its prime subfield possesses only a finite number of roots of unity. This is clear when L is of characteristic ≠0\neq 0 because it is then finite. If L is of characteristic 0 and contains infinitely many roots of unity, then it contains primitive roots of unity of arbitrarily high order. By V, p. 84, Th. 2 there exist therefore an infinity of integers n>0n > 0 such that φ(n)≤[L:Q]\varphi(n) \leq [L : \mathbf{Q}], which is absurd (V, p. 80, Formulae (2) and (3)).

Corollary 3

— *If E is a finitely generated extension of a field K, then every subextension E' of E is finitely generated.

Let B' be a transcendence basis of E' over K. By V, p. 109, Th. 2, B' is contained in a transcendence basis B of E over K and so is finite, by Prop. 17. Since E' is algebraic over K(B') and E is a finitely generated extension of K(B'). Cor. 1 shows [E′:K(B′)][E' : K(B')] to be finite. Now Prop. 17 shows that E' is finitely generated over K.

Prop. 17 may be paraphrased by saying that a finitely generated extension of K is an algebraic extension of finite degree of a purely transcendental extension K(x1,…,xn)K(x_1, \ldots, x_n).

Exercises

See the exercises for § 14.

Transcribed from the printed edition, 2003, Springer. Not translated.