Bourbaki

§ 2. Ordered fields

1. Ordered rings

Definition 1

— Given a commutative ring A\mathbf{A}, we say that an ordering on A\mathbf{A} is compatible with the ring structure of A\mathbf{A} if it is compatible with the additive group structure of A\mathbf{A}, and if it satisfies the following axiom : (OR) The relations x⩾0x \geqslant 0 and y⩾0y \geqslant 0 imply xy⩾0xy \geqslant 0. The ring A\mathbf{A}, together with such an ordering, is called an ordered ring.

Example 1

The rings Q\mathbf{Q} and Z\mathbf{Z}, with the usual orderings, are ordered rings.

Example 2

A product of ordered rings, equipped with the product ordering, is an ordered ring. In particular the ring AE\mathbf{A}^E of mappings from a set EE to an ordered ring A\mathbf{A} is an ordered ring.

Example 3

A subring of an ordered ring, with the induced ordering, is an ordered ring.

In an ordered ring, the relations x⩾yx \geqslant y and z⩾0z \geqslant 0 imply xz⩾yzxz \geqslant yz. Indeed these inequalities are equivalent to x−y⩾0x - y \geqslant 0, z⩾0z \geqslant 0 and (x−y)z⩾0(x - y)z \geqslant 0 respectively. Analogously we can show that the relations x⩽0x \leqslant 0 and y⩾0y \geqslant 0 (resp. y⩽0y \leqslant 0) imply xy⩽0xy \leqslant 0 (resp. xy⩾0xy \geqslant 0). These results are often invoked under the name of sign rules (two elements are said to have the same sign if they are both ⩾0\geqslant 0 or both ⩽0\leqslant 0). They imply that, if A\mathbf{A} is a totally ordered ring, then every square is positive, and in particular that every idempotent (for example the unit element) is positive.

Example

There is only one totally ordered ring structure on Z\mathbf{Z}: indeed 1>01 > 0, whence n>0n > 0 for every natural number n≠0n \neq 0, by induction. In contrast there exist ordered ring structures on Z\mathbf{Z} which are not totally ordered (see below).

Let PP be the set of positive elements of an ordered ring A\mathbf{A}. It is known (VI, p. 3, Prop. 3) that PP determines the ordering on A\mathbf{A}. To say that A\mathbf{A} is an ordered ring is equivalent to saying that PP satisfies the following properties :

(API)P+P⊂P(APII)PP⊂P(APIII)P∩(−P)={0} \begin{align*} (\mathrm{AP}_I) &\quad P + P \subset P \\ (\mathrm{AP}_{II}) &\quad PP \subset P \\ (\mathrm{AP}_{III}) &\quad P \cap (-P) = \{0\} \end{align*}

Indeed (API)(\mathrm{AP}_I) and (APIII)(\mathrm{AP}_{III}) state that the additive group of A\mathbf{A} is an ordered group (VI, p. 3, Prop. 3), while (AP,,)(\mathrm{AP}_{,,}) is a translation of (OR).

Recall that the following condition is necessary and sufficient for the order relation on A\mathbf{A} to be total :

(APIV)P∪(−P)=A. (\mathrm{AP}_{IV}) \quad P \cup (-P) = \mathbf{A}.

Example

In Z\mathbf{Z}, if we take PP to be the set of positive (in the usual sense) even integers, we get a ring which is not totally ordered.

Recall also that, in a totally ordered abelian group, the relation n⋅x=0n \cdot x = 0 (for a natural number n≠0n \neq 0) implies x=0x = 0 (VI, p. 4); this gives us the following result.

Proposition 1

— A totally ordered ring is torsion free as a Z\mathbf{Z}-module (II, p. 313).

2. Ordered fields

Definition 2

— A commutative field, equipped with a total ordering, is called an ordered field if its ordering and its ring structure are compatible.

We restrict ourselves to total order relations on fields because the others are very « pathological » (VI, p. 38, Ex. 6).

Examples. — 1) The field Q\mathbf{Q} of rational numbers is an ordered field. 2) A subfield of an ordered field, with the induced ordering, is an ordered field. 3) \ The field of real numbers is an ordered field. \*

Let KK be an ordered field. For all x∈Kx \in K we put

sgn⁡(x)=1if x>0,sgn⁡(x)=−1if x<0,sgn⁡(x)=0if x=0. \begin{align*} \operatorname{sgn}(x) &= 1 & \text{if } x > 0, \\ \operatorname{sgn}(x) &= -1 & \text{if } x < 0, \\ \operatorname{sgn}(x) &= 0 & \text{if } x = 0. \end{align*}

Then we have sgn⁡(xy)=sgn⁡(x)sgn⁡(y)\operatorname{sgn}(xy) = \operatorname{sgn}(x) \operatorname{sgn}(y); we call sgn⁡(x)\operatorname{sgn}(x) the sign of xx. The map x↦sgn⁡(x)x \mapsto \operatorname{sgn}(x) from K∗K^* to the multiplicative group {−1,+1}\{-1, +1\} is a surjective homomorphism whose kernel, the set of strictly positive elements of KK, is a subgroup of K∗K^* of index 2.

Conversely, if KK is a commutative field and s:K∗→{−1,+1}s : K^* \to \{-1, +1\} is a surjective homomorphism whose kernel is closed under addition, then ss is the sign map for a unique ordered field structure, where the set of strictly positive elements is the kernel of ss.

For all xx and yy in KK we have x=sgn⁡(x)∣x∣x = \operatorname{sgn}(x)|x| and ∣xy∣=∣x∣∣y∣|xy| = |x||y|.

On the other hand every ordered field is of characteristic zero (Prop. 1).

Proposition 2

— *Let AA be a totally ordered integral domain, and let KK be its field of fractions. Then there exists one and only one ordering on KK which restricts to the given ordering on AA and makes KK an ordered field.

Every x∈Kx \in K can be expressed in the form x=ab−1x = ab^{-1}, with aa and bb in AA and b≠0b \neq 0. If xx is positive, then aa and bb have the same sign, and conversely. Thus we see that, if there exists an ordering on KK satisfying the prescribed conditions, then it is unique, and the set PP of positive elements is identical to the set of ab−1ab^{-1}, where aa and bb are elements of AA of the same sign, and b≠0b \neq 0. It remains to show that PP satisfies conditions (API_I), (APII_{II}), (APIII_{III}) and (APIV_{IV}). This is obvious for (APII_{II}) and (APIV_{IV}). For (API_I), consider ab−1+cd−1ab^{-1} + cd^{-1}, where we may assume that aa, bb, cc and dd are positive; this sum is (ad+bc)(bd)−1(ad + bc)(bd)^{-1}, and ad+bcad + bc and bdbd are positive.

To show (APIII)(\mathrm{AP}_{III}), consider an identity of the form ab−1=−cd−1ab^{-1} = -cd^{-1}, so that ad+bc=0ad + bc = 0. If we assume that aa and bb have the same sign and that cc and dd have the same sign, then the sign rules show that adad and bcbc have the same sign; whence ad=bc=0ad = bc = 0, so a=c=0a = c = 0; hence PP does indeed satisfy (APIII)(\mathrm{AP}_{III}).

Example

— Since ZZ admits only one totally ordered ring structure (VI, p. 19, example), the field Q\mathbf{Q} admits only one ordering which makes it an ordered field: this is the usual ordering.

3. Extensions of ordered fields

Definition 3

Let KK be an ordered field. An ordered extension of KK is a pair (E,u)(E, u), where EE is an ordered field and uu is an increasing homomorphism from KK to EE.

Let KK be a field, let EE be an ordered field and let u:K→Eu : K \to E be a homomorphism. The relation

x≤yifu(x)≤u(y) x \leq y \quad \text{if} \quad u(x) \leq u(y)
is a total order relation on KK which gives it an ordered field structure, said to be induced by that of EE. If KK and EE are ordered fields, then a homomorphism u:K→Eu : K \to E is increasing if and only if the ordered field structure of KK is induced by that of EE. We will usually identify KK with its image in EE under uu.

Example

— 1) Every ordered field KK is an ordered extension of Q\mathbf{Q}. Indeed KK is an extension of Q\mathbf{Q}, since it is of characteristic zero, and on the other hand Q\mathbf{Q} can only be ordered in one way, as we have just seen.

2) Let KK be an ordered field, and let K(X)K(X) be the field of rational functions in one indeterminate over KK. Let us define an ordering on the polynomial ring K[X]K[X] by taking the positive elements to be 00 and those polynomials whose leading coefficient is positive. In this way we obtain a totally ordered ring whose ordering extends that of KK. By applying Prop. 2 we give K(X)K(X) the structure of an ordered extension of KK. *For K=RK = \mathbf{R} it can be shown that the order relation defined on K(X)K(X) in this way is that of growth near +∞+\infty (cf. VI, p. 24, Prop. 4).

Theorem 1

For an extension EE of KK to admit the structure of an ordered extension of KK, the following condition is necessary and sufficient:* (OE) The relation p1x12+…+pnxn2=0p_1 x_1^2 + \ldots + p_n x_n^2 = 0 implies

p1x1=…=pnxn=0 p_1 x_1 = \ldots = p_n x_n = 0
*for any finite sequence (xi,pi)(x_i, p_i) of pairs of elements xx, of EE and positive elements pip_i of KK.

Condition (OE) is clearly equivalent to: (OE') The element – 1 is not a sum of elements of the form px2px^2 (x∈E,p∈K,p≥0x \in E, p \in K, p \geq 0).

Condition (OE) is necessary: if E is an ordered extension of K then the elements pixi2p_i x_i^2 are positive in E, so zero if their sum is zero. On the other hand pixi2=0p_i x_i^2 = 0 is equivalent to pixi=0p_i x_i = 0.

Conversely, suppose condition (OE) is satisfied, then we will define an ordering on E by constructing a subset P of E which satisfies conditions (AP_I), (AP_{II}), (AP_{III}) and (AP_{IV}), and which contains the set K+K_+ of positive elements of K. Such a subset P will certainly make E an ordered extension of K, for we will have K∩P=K+K \cap P = K_+; indeed, if P were to contain an element −a<0-a < 0 of K, then a would belong to P∩(−P)P \cap (-P), contradicting (APIII).

To define P, let us consider the set M\mathscr{M} of subsets of E which satisfy (AP_I), (AP_{II}) and (AP_{III}), and which contain the union of K+K_+ and the set C of squares of elements of E. This set M\mathscr{M} is nonempty, for it contains the set P0P_0 of elements of the form ∑pixi2\sum p_i x_i^2 (that P0P_0 satisfies (AP_{III}) follows immediately from (OE)).

Moreover M\mathscr{M} is inductive (Set Theory, III, p. 154, Def. 3). Thus there exists, by Th. 2 of Set Theory, III, p. 154, a maximal element in M\mathscr{M}, which it remains for us to prove satisfies (AP_{IV}); now this follows from the following lemma:

Lemma

— Let P∈MP \in \mathscr{M} and x∉Px \notin P; then there exists P′∈MP' \in \mathscr{M} such that P⊂P′P \subset P' and −x∈P′-x \in P'.

Take P′=P−xPP' = P - xP, and check that P′P' has the required properties. Since 0∈C⊂P0 \in C \subset P, we have P⊂P′P \subset P'. Whence C⊂P′C \subset P' and K+⊂P′K_+ \subset P'. Since 1∈C⊂P1 \in C \subset P we have −x∈P′-x \in P'. We have

P′+P′=P−xP+P−xP=P+P−x(P+P)⊂P−xP=P′, P' + P' = P - xP + P - xP = P + P - x(P + P) \subset P - xP = P',
whence (AP_I). We have
P′P′=(P−xP)(P−xP)⊂⊂PP+x2PP−x(PP+PP)⊂P+CP−xP⊂P−xP=P′, P'P' = (P - xP)(P - xP) \subset \\ \quad \subset PP + x^2PP - x(PP + PP) \subset P + CP - xP \subset P - xP = P',
whence (AP_{II}). Finally, let us check (AP_{III}): suppose given an identity of the form p−xq=−(r−xs)p - xq = -(r - xs) where p,q,r,sp, q, r, s belong to P; we deduce from this the relation x(s+q)=p+rx(s + q) = p + r; if s+q≠0s + q \neq 0 we have
x=(s+q)−2(s+q)(p+r)∈CPP⊂P, x = (s + q)^{-2}(s + q)(p + r) \in CPP \subset P,
contrary to the hypothesis; hence s+q=0s + q = 0, whence p+r=0p + r = 0; since P satisfies (AP_{III}) we deduce that s=q=r=p=0s = q = r = p = 0, which completes the proof.

Corollary 1 (« Artin-Schreier Theorem »)

— A necessary and sufficient condition for there to exist an ordering on a commutative field E which makes it an ordered field, is that the relation x12+…+xn2=0x_1^2 + \ldots + x_n^2 = 0 imply x1=…=xn=0x_1 = \ldots = x_n = 0.

The necessity is obvious. Conversely, the stated condition implies that E is of characteristic zero, hence an extension of Q; then condition (OE)(OE) is satisfied, and Th. 1 shows that there exists on E the structure of an ordered extension of Q, that is an ordered field structure.

There does not exist any ordered field structure on a field E in which −1-1 is a square, in particular on an algebraically closed field.

Corollary 2

Let E be an extension of K admitting the structure of an ordered extension of K. For an element x∈Ex \in E to be positive under every such structure on E, it is necessary and sufficient that x be of the form ∑ipixi2\sum_i p_i x_i^2, where xi∈Ex_i \in E and the pip_i are positive elements of K.

The condition is obviously sufficient; it is also necessary, for (in the notation of the proof of Th. 1), if x∉P0x \notin P_0 there exists a maximal element P of M\mathscr{M} such that x∉Px \notin P; then −x∈P-x \in P by the Lemma, and x is not positive under the ordering defined by P, since x≠0x \neq 0.

4. Algebraic extensions of ordered fields

Let K be an ordered field, and f a polynomial in K[X]. We will say that f changes sign in K if there exist two elements a and b in K such that f(a)f(b)<0f(a) f(b) < 0; then we say that f changes sign between a and b.

Proposition 3

Let K be an ordered field and f an irreducible polynomial over K which changes sign between a and b in K. Then the extension E=K[X]/(f)E = K[X]/(f) of K admits the structure of an ordered extension.

We will argue by induction on the degree n off. For n=1n = 1 the proof is trivial. Suppose the result holds for degrees ≤n−1\leq n - 1, and let us prove it for degree n by contradiction; by Th. 1 we are thus assuming a relation of the form

1+∑ipifi2(X)≡0(modf(X)),where fi∈K[X],pi∈Kandpi≥0 1 + \sum_i p_i f_i^2(X) \equiv 0 \pmod{f(X)}, \quad \text{where } f_i \in K[X], \quad p_i \in K \quad \text{and} \quad p_i \geq 0

Without loss of generality we may suppose that the fif_i have degrees ≤n−1\leq n - 1 (IV, p. 11, Cor). Then

1+∑ipifi2(X)=h(X)f(X) 1 + \sum_i p_i f_i^2(X) = h(X) f(X)

where h≠0h \neq 0 has degree at most n−2n - 2. Replacing X by a and b in the above inequality, we see that h(a)f(a)>0h(a) f(a) > 0 and h(b)f(b)>0h(b) f(b) > 0. Since f changes sign between a and b by hypothesis, we conclude that h(a)h(b)<0h(a) h(b) < 0. Then we have a similar inequality for one of the irreducible factors g(X)g(X) of h(X)h(X): that is g(a) g(b)<0g(a)\ g(b) < 0. But 1+∑iPifi2(X)≡0(modg(X))1 + \sum_i P_i f_i^2(X) \equiv 0 \pmod{g(X)}, which shows that the field K[X]/(g)K[X]/(g) cannot be an ordered extension of KK (Th. 1), contrary to the induction hypothesis.

Remark

There exist irreducible polynomials ff over an ordered field KK which do not change sign in KK, but such that K[X]/(f)K[X]/(f) admits the structure of an ordered extension of KK (cf. VI, p. 43, Ex. 26, c)).

In order to apply the previous proposition we will need the following result:

Proposition 4

— *Let KK be an ordered field and let f∈K[X]f \in K[X]. There exists an interval in KK, in the complement of which ff takes the same sign as its highest degree term.

We can immediately reduce ourselves to the case of a monic polynomial; then one can write f(x)=xn(1+a1x−1+…+anx−n)f(x) = x^n(1 + a_1 x^{-1} + \ldots + a_n x^{-n}) for x≠0x \neq 0. Let

M=sup⁡(1,∣a1∣+⋯+∣an∣). M = \sup (1, |a_1| + \cdots + |a_n|).

For ∣x∣>M|x| > M we have 1+a1x−1+…+anx−n>01 + a_1 x^{-1} + \ldots + a_n x^{-n} > 0, which completes the proof of the proposition.

Corollary 1

— *Every extension of an ordered field of odd finite degree admits the structure of an ordered extension.

Such an extension, being monogenous (V, p. 40, Th. 1), is isomorphic to K[X]/(f)K[X]/(f), where ff is an irreducible polynomial of odd degree. Then it is enough to show that ff changes sign in KK (Prop. 3), which follows immediately from Prop. 4.

Corollary 2

— *If aa is a positive element of an ordered field KK, then every splitting field EE of the polynomial X2−aX^2 - a admits the structure of an ordered extension of KK.

The result is trivial if aa is a square in KK. Otherwise the polynomial f(X)=X2−af(X) = X^2 - a is irreducible and changes sign, since f(0)<0f(0) < 0 and f(x)f(x) has the same sign as x2x^2, so positive, for xx in the complement of some interval of KK. We can now complete the proof by applying Prop. 3.

Remark

When the ordered field KK contains the « square roots » of a positive element aa of KK (roots of the polynomial X2−aX^2 - a) then the notation a\sqrt{a} is generally reserved for the positive square root. If KK does not contain the square roots bb and −b-b of aa in the field EE, then the latter can be made an ordered extension of KK in two ways, each induced from the other via the KK-automorphism which sends bb to −b-b; the choice of one of these orderings determines a\sqrt{a}: it is whichever of the elements bb and −b-b is positive.

If aa and a′a' are two positive elements of KK, whose square roots are in KK, then aa′=aa′\sqrt{aa'} = \sqrt{a} \sqrt{a'}, which follows from the definition of a\sqrt{a} and the sign rule.

5. Maximal ordered fields

Definition 4

— *An ordered field K is maximal if every ordered algebraic extension of K is trivial.

Example

We will see later (Gen. Top., VIII, p. 1) that the field R of real numbers is a maximal ordered field.

The existence of maximal ordered fields is a consequence of the following theorem:

Theorem 2

— *Every ordered field K admits an ordered algebraic extension which is a maximal ordered field.

One can show that this ordered extension is unique up to K-isomorphism (VI, p. 40, Ex. 15).

Let Ω\Omega be an algebraic closure of K, and let N\mathfrak{N} be the set of pairs (A,w)(A, w), where AA is a sub-K-extension of Ω\Omega, and ww is an ordering on AA making AA an ordered extension of KK. Order N\mathfrak{N} by the relation « L is an ordered extension of M » between M and LL. Equipped with this ordering, N\mathfrak{N} is an inductive ordered set: indeed if (Li)(L_i) is a totally ordered family of elements of N\mathfrak{N}, then the field L=⋃LiL = \bigcup L_i, ordered by taking L1=⋃(Li)1L_1 = \bigcup (L_i)_1, is an upper bound for the LiL_i. Then N\mathfrak{N} has a maximal element, by Set Theory, III, p. 154, Th. 2, which completes the proof.

Proposition 5

— *Let K be a maximal ordered field, and let f be a polynomial in K[X] which changes sign between two elements a and b of K (with a<ba < b). Then f has a root x in K such that a<x<ba < x < b.

At least one of the irreducible factors off, say hh, changes sign between aa and bb. Then the field K[X]/(h)K[X]/(h) admits the structure of an ordered extension of KK (VI, p. 23, Prop. 3), and hh has degree 1 (Def. 4). Since h(a)h(b)<0h(a) h(b) < 0, the unique root xx of hh is such that a<x<ba < x < b, since a polynomial function of degree 1 is monotonic.

Proposition 6

— *Every positive element of a maximal ordered field K has a square root in K. Every polynomial of odd degree in K[X] has at least one root in K.

This follows immediately from Cor. 2 and 1 to Prop. 4 of VI, p. 24.

Corollary

— *On a maximal ordered field K there exists only one ordering compatible with the field structure.

Indeed the positive elements of K are determined by its algebraic structure: they are the squares.

6. Characterisation of maximal ordered fields. Euler-Lagrange Theorem

The property expressed by Prop. 6 of VI, p. 25 characterises maximal ordered fields. More precisely:

Theorem 3 (Euler-Lagrange)

— Let K be an ordered field. Then the following three properties are equivalent: a) The field K(i)K(i) is algebraically closed (where i denotes a square root of -1). b) The ordered field K is maximal. c) Every positive element of K is a square, and every polynomial of odd degree in K[X]K[X] has a root in K.

It is clear that a) implies b): indeed K has only two algebraic extensions up to isomorphism, the field K itself and K(i)K(i), which cannot be ordered since -1 is a square.

The fact that b) implies c) is nothing other than Prop. 6 of VI, p. 25.

It remains for us to prove that c) implies a). That will follow from the next two propositions.

Proposition 7

— Let K be an ordered field in which every positive element is a square. Then every element of K(i)K(i) is a square, and every polynomial of degree 2 over K(i)K(i) has a root in K(i)K(i).

Let us show first that the second assertion reduces to the first. One can put the second degree polynomial aX2+bX+caX^2 + bX + c (a≠0a \neq 0) in the following form, often called the canonical trinomial form:

a((X+(b/2a))2−(b2−4ac)/4a2) a((X + (b/2a))^2 - (b^2 - 4ac)/4a^2)

If d is a square root of (b2−4ac)/4a2(b^2 - 4ac)/4a^2, then d−(b/2a)d - (b/2a) is a root of the quadratic polynomial under consideration.

Now we show that every element a+bia + bi (a∈K,b∈Ka \in K, b \in K) is a square; we are looking for an element x+yix + yi such that

(x+yi)2=a+bi; (x + yi)^2 = a + bi;

this translates into x2−y2=ax^2 - y^2 = a and 2xy=b2xy = b. From this we deduce that

(x2+y2)2=a2+b2 (x^2 + y^2)^2 = a^2 + b^2

Let c denote the positive square root of a2+b2a^2 + b^2; then c≥∣a∣,c≥∣b∣c \geq |a|, c \geq |b| and x2+y2=cx^2 + y^2 = c. Whence x2=(c+a)/2x^2 = (c + a)/2 and y2=(c−a)/2y^2 = (c - a)/2. Since c≥∣a∣c \geq |a| these equations are soluble in K, and if x0x_0 and y0y_0 are two solutions then x02−y02=ax_0^2 - y_0^2 = a and 2x0y0=±b2x_0y_0 = \pm b. We obtain the desired square root by taking x=x0x = x_0 and y=b/2x0y = b/2x_0.

Proposition 8

— Let K be a commutative field (of arbitrary characteristic) and let K′K' be a splitting field for the polynomial X2+1∈K[X]X^2 + 1 \in K[X] (V, p. 21). Suppose: a) every polynomial in K[X]K[X] of odd degree has a root in K′K';

b) every polynomial in K′[X]K'[X] of degree 2 has a root in K′K'. Then K′K' is algebraically closed.

Note first that it is enough to prove that every non-constant polynomial in K[X]K[X] has a root in K′K': this is indeed clear if K′=KK' = K; if K′≠KK' \neq K then [K′:K]=2[K':K] = 2; let a↦aa \mapsto a denote the unique KK-automorphism of K′K' distinct from the identity map; if f∈K′[X]f \in K'[X] and if fˉ\bar{f} denotes the polynomial obtained by applying a↦aˉa \mapsto \bar{a} to the coefficients of ff, then ffˉ∈K[X]f \bar{f} \in K[X]; if a∈K′a \in K' is a root of ffˉf \bar{f} then aa is either a root of ff or of fˉ\bar{f}; thus either aa or aa is a root of ff.

Thus let ff be a polynomial over KK of degree 2np2^n p, pp odd. We will proceed by induction on nn, the property being true for n=0n = 0 by hypothesis a). Let E be an extension of KK in which ff splits into linear factors:

f(X)=∏i(X−ai) f(X) = \prod_i (X - a_i)

Let b∈Kb \in K; put yij=ai+aj+baiaj∈Ey_{ij} = a_i + a_j + ba_i a_j \in E and

h(X)=∏i<j(X−yij)∈E[X]. h(X) = \prod_{i < j} (X - y_{ij}) \in E[X].

The coefficients of this polynomial are symmetric functions in the aia_i, with coefficients in KK; it therefore belongs to K[X]K[X] (IV, p. 62, Th. 1); since it has degree 2np(2np−1)/2=2n−1p′2^n p (2^n p - 1)/2 = 2^{n-1} p' (p′p' odd), it has a root yy_{}, in K′K' by inductive hypothesis. If we note that this holds for all b∈Kb \in K, and that KK is an infinite field (indeed a finite field, which has monogenous extensions of arbitrarily large odd degree (V, p. 94, Prop. 3), cannot satisfy a)), then we can deduce the existence of at least one pair (i,j)(i, j) such that

ai+aj+baiaj∈K′andai+aj+b′aiaj∈K′, a_i + a_j + ba_i a_j \in K' \quad \text{and} \quad a_i + a_j + b'a_i a_j \in K',

with b≠b′b \neq b'. Then ai+aja_i + a_j and aiaja_i a_j are elements of K′K', hence so are aia_i and aja_j, since they are the roots of the quadratic equation

x2−(ai+aj)x+aiaj=0. x^2 - (a_i + a_j)x + a_i a_j = 0.

Q.E.D.

For a generalisation and an alternative proof of Prop. 8, based on Galois theory, see VI, p. 46, Ex. 33.

Let KK be an ordered field and let K′=K(i)K' = K(i); for every element z=a+biz = a + bi of K′K', the norm zz‾=a2+b′z \overline{z} = a^2 + b' of zz relative to KK (III, p. 544, example 1) is a positive element of KK, which vanishes only for z=0z = 0. If every positive element in KK is a square (in particular if KK is a maximal ordered field), then the positive square root of the norm zz‾z \overline{z} is called the absolute value of zz, and is written ∣z∣|z|. Since ∣zz′∣2=∣z∣2∣z′∣2|zz'|^2 = |z|^2 |z'|^2 we have ∣zz′∣=∣z∣⋅∣z′∣|zz'| = |z| \cdot |z'|.

Moreover, the triangle inequality

∣z+z′∣≤∣z∣+∣z′∣ |z + z'| \leq |z| + |z'|

holds for every pair of elements z,z′z, z' of K′K'. Indeed, if z=a+biz = a + bi and z′=a′+b′iz' = a' + b'i, then this inequality is equivalent to

(a+a′)2+(b+b′)2≤a2+b2+a′2+b′2+2(a2+b2)(a′2+b′2) (a + a')^2 + (b + b')^2 \leq a^2 + b^2 + {a'}^2 + {b'}^2 + 2 \sqrt{(a^2 + b^2)({a'}^2 + {b'}^2)}

and hence also to

(aa′+bb′)2≤(a2+b2)(a′2+b′2) (aa' + bb')^2 \leq (a^2 + b^2)({a'}^2 + {b'}^2)

which can be written (ab′−ba′)2≥0(ab' - ba')^2 \geq 0.

Th. 3 enables us to determine all the irreducible polynomials over a maximal ordered field:

Proposition 9

— *If K is a maximal ordered field, then the only irreducible polynomials in K[X] are the first degree polynomials, and the second degree polynomials aX2+bX+caX^2 + bX + c such that b2−4ac<0b^2 - 4ac < 0.

Since K(i)K(i) is algebraically closed, every algebraic extension of K, and hence also every irreducible polynomial over K, has degree 1 or 2. To see which second degree polynomials are irreducible, it is enough to consider the canonical form a((X+(b/2a))2−(b2−4ac)/4a2)a((X + (b/2a))^2 - (b^2 - 4ac)/4a^2) (cf. VI, p. 26, Prop. 7).

Remark

— Translating into the canonical trinomial form yields this stronger result: a necessary and sufficient condition for the polynomial aX2+bX+caX^2 + bX + c over a given ordered field K to have constant sign in K is that b2−4ac<0b^2 - 4ac < 0, and then the sign of the polynomial is that of aa.

7. Vector spaces over an ordered field

Let K be an ordered field, and let E be a vector space over KK. The relation « there exists λ>0\lambda > 0 in KK such that y=λxy = \lambda x » between two elements xx and yy in the set E−{0}E - \{0\} is an equivalence relation. The equivalence classes under this relation are called open halflines with origin 0; the union of an open half-line and {0}\{0\} is called a closed half-line (or sometimes simply a halfline) with origin 0. Every vector a≠0a \neq 0 contained in an open (resp. closed) half-line A is called a direction vector of AA, and AA is the set of vectors haha for all scalars A>0A > 0 (resp. A≥0A \geq 0). Every line DD through 0 contains exactly two open (resp. closed) half-lines with origin 0 ; if AA is one of these, then −A-A is the other (called the opposite of AA).

Now if FF is an affine space over KK, and E the space of translations of F, then any subset of F of the form A=a+A0A = a + A_0, where Δ0\Delta_0 is an open (resp. closed) half-line of E, is called an open (resp. closed) half-line with origin a∈Fa \in F. The half-line Δ0\Delta_0 is completely determined by AA (for it is the half-line with direction vector b−ab - a, for any b≠ab \neq a in AA), and is called the direction of AA; a direction vector of AA, is also called a direction vector of AA.

Suppose now that E has finite dimension nn over K ; then it is known (III, p. 518, Cor. 1) that the n-th exterior power ΛnE\Lambda^n E is a vector space of dimension 1 over K, and so the union of two opposite closed half-lines of origin 0. These half-lines are called orientations of E; the space E together with a given one of these half-lines A is said to be oriented; an n-vector z is then called positive (resp. negative) under this orientation if it belongs to A (resp. to −A); it is negative (resp. positive) under the opposite orientation.

An orientation of an affine space F over K is by definition an orientation of the space of translations of F; the space F, together with such an orientation, is called an oriented affine space.

Let E be an oriented vector space over K, of dimension nn; an ordered basis (ai)1≤i≤n(a_i)_{1 \leq i \leq n} of E is called positive or direct (resp. negative or inverse) if the n-vector a1a2…ana_1 a_2 \ldots a_n is positive (resp. negative). If uu is an automorphism of the vector space E then (⋀nu)(z)=det⁡(u)⋅z(\bigwedge^n u)(z) = \det(u) \cdot z for all z∈⋀nEz \in \bigwedge^n E, so a necessary and sufficient condition for ⋀nu\bigwedge^n u to leave invariant the orientation of E (or, as we also say, to preserve the orientation) is that det⁡(u)>0\det(u) > 0; the automorphisms having this property are precisely the automorphisms of E as an oriented vector space; they form a normal subgroup GL+(E)\mathrm{GL}^+(E) of the linear group GL(E)\mathrm{GL}(E), which has index 2 whenever E≠0E \neq 0.

When E=0E = 0 then GL(E)=End(E)\mathrm{GL}(E) = \mathrm{End}(E) contains only the identity map 1, and by definition det⁡(1E)=1\det(1_E) = 1. Note that ⋀nE=⋀0E=K\bigwedge^n E = \bigwedge^0 E = K by definition in this case; the half-line of K formed by the positive scalars is called the canonical orientation of the zero space.

Let M and N be two complementary subspaces of dimensions p and n−pn - p respectively in the vector space E of dimension n; if z′z' (resp. z′′z'') is a nonzero vector in ⋀pM\bigwedge^p M (resp. ⋀n−pN\bigwedge^{n-p} N), then z′∧z′′z' \wedge z'' is a nonzero vector in ⋀nE\bigwedge^n E. Given an orientation on M and an orientation on N, the vectors z′∧z′′z' \wedge z'' for positive z′z' and z′′z'' form an orientation of E, called the product orientation of the orientation of M by the orientation of N (which depends on the order of the factors when p(n−p)p(n-p) is odd). Conversely, given orientations on E and on M, there exists a unique orientation on N such that the given orientation on E is the product of the given orientation on M and this orientation on N (in that order); this orientation is said to be complementary to the orientation of M with respect to that of E. If N′N' is a second complementary subspace to M, the canonical projection N→N′N \to N' parallel to M takes the complementary orientation of N onto that of N′N'. The image of the complementary orientation of N under the canonical map N→E/MN \to E/M is thus independent of the choice of complement N; it is called the quotient orientation on E/M of the orientation of E by that of M.

Exercises

Exercises

All rings under consideration are assumed to be commutative unless explicitly stated otherwise.

See the exercises for § 2.

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