Bourbaki

§ 1. Ordered groups. Divisibility

The notions and results presented in this section concern the study of order relations in commutative monoids (I, p. 12, Def. 2), the most important case being that of abelian groups. Unless explicitly stated otherwise, we will use additive notation for the composition law in all groups and monoids under study. On the other hand, as we go along we will present certain important algebraic applications of the theory of ordered groups and monoids, and we will accordingly translate certain of our results into the multiplicative notation which is appropriate for these applications.

1. Definition of ordered monoids and groups

Definition 1

— A commutative monoid structure (written additively) and an ordering (written ≤\leq) on a set MM are said to be compatible if they satisfy the following axiom:

(OM) For each z∈Mz \in M, the relation x≤yx \leq y implies x+z≤y+zx + z \leq y + z.

A set MM equipped with a commutative monoid structure and an ordering which are compatible is called an ordered monoid; if its commutative monoid structure is a group structure, then it is called an ordered group.

In an analogous fashion, we can define the notion of a noncommutative ordered monoid (VI, p. 30, Ex. 1).

If an ordering is compatible with a given monoid structure, then so is the opposite ordering.

Example 1

The additive group of the rational integers, and that of the rational numbers, are ordered groups under the ordering defined in I, pp. 21 and 117. *The same is true for the additive group of the real numbers (Gen. Top., IV, p 3). \

Example 2

The additive group of finite numerical functions defined on a set EE is an ordered group under the ordering « f(x)≤g(x)f(x) \leq g(x) for all x∈Ex \in E », written f≤gf \leq g. This relation says that the graph of the function ff lies below that of the function gg. The reader may find it convenient to think in terms of this graphical interpretation occasionally. \

According to general definitions (Set Theory, IV, p. 264), a bijective map ff from an ordered monoid M onto an ordered monoid M' is called an isomorphism of M onto M' if the structure of M' is obtained from that of M by transporting it along ff. This is equivalent to saying that ff is a mapping onto M' such that

f(x+y)=f(x)+f(y) f(x + y) = f(x) + f(y)

(that is to say a homomorphism of the monoid M onto the monoid M'), and that the relations x≤yx \leq y and f(x)≤f(y)f(x) \leq f(y) are equivalent (whence in particular f(x)=f(y)f(x) = f(y) implies x=yx = y, that is ff is injective).

Cites 1

Cites E IV § 1. Structures and isomorphisms

Proposition 1 (« addition of inequalities »)

— In an ordered monoid M, let (xi)(x_i) and (yi)(y_i) (1≤i≤n1 \leq i \leq n) be two sequences of n elements such that xi≤yix_i \leq y_i for all ii; then one has

x1+⋯+xn≤y1+⋯+yn x_1 + \cdots + x_n \leq y_1 + \cdots + y_n
Cites 2

Cites Definition 5, I § 2, Proposition 2, I § 2

If moreover all the elements xi,yix_i, y_i are cancellable (I, p. 15, Def. 5) (in particular if M is a group), and if xi<yix_i < y_i for some ii, then

x1+⋯+xn<y1+⋯+yn. x_1 + \cdots + x_n < y_1 + \cdots + y_n .

The general case reduces by induction to the case n=2n = 2, using the fact that a sum of cancellable elements is cancellable for the second assertion (I, p. 15, Prop. 2). The first assertion follows from the relations

x1+x2≤x1+y2andx1+y2≤y1+y2, x_1 + x_2 \leq x_1 + y_2 \quad \text{and} \quad x_1 + y_2 \leq y_1 + y_2 ,

which are consequences of the hypotheses and of (OM). This being the case, the relation

x1+x2=y1+y2 x_1 + x_2 = y_1 + y_2

would imply

x1+x2=x1+y2=y1+y2, x_1 + x_2 = x_1 + y_2 = y_1 + y_2 ,

whence x2=y2x_2 = y_2 and x1=y1x_1 = y_1 if xx, and y2y_2 are cancellable, which proves the second assertion.

Proposition 2

In an ordered group G* the relations x≤yx \leq y and x+z≤y+zx + z \leq y + z are equivalent.

Cited by 1

Cited by Corollary

Indeed one can obtain each from the other by adding zz or (−z)(-z) to both sides.

This fact can be expressed by saying that, in an ordered group G, the ordering is translation-invariant. In other words a translation is an automorphism of the ordering of an ordered group.

Corollary

— In an ordered group GG, the relations x≤yx \leq y, 0≤y−x0 \leq y - x, x−y≤0x - y \leq 0 and −y≤−x-y \leq -x are equivalent.

Cites 1

Cites Proposition 2

In fact, we apply Prop. 2, successively taking z=−xz = -x, z=−yz = -y and z=−(x+y)z = -(x + y).

In particular we deduce from this corollary that if GG is an ordered group, the map x↦−xx \mapsto -x from GG to itself takes the ordering of GG to the opposite ordering.

2. Pre-ordered monoids and groups

Recall that if a relation x≤yx \leq y between elements of a set EE is reflexive and transitive, it is called a pre-order relation (Set Theory, III, p. 133). The relation « x≤yx \leq y and y≤xy \leq x » is an equivalence relation SS on EE, compatible with the relation x≤yx \leq y; on passing to the quotient, the relation ≤\leq induces an order relation on E/SE/S, called the order relation associated to ≤\leq.

Definition 2

— A pre-order relation (written ≤\leq) and a commutative monoid structure (written additively) on a set MM are said to be compatible if they satisfy the following axiom:

Cites 1

Cites I § 1. Laws of composition; associativity; commutativity

(POM) For each z∈Mz \in M, the relation x≤yx \leq y implies x+z≤y+zx + z \leq y + z.

A set MM equipped with a commutative monoid structure and a pre-order relation which are compatible is called a pre-ordered monoid.

Let MM be a pre-ordered monoid, and SS the equivalence relation « x≤yx \leq y and y≤xy \leq x ». By virtue of (POM) the relation x≡x′(modS)x \equiv x' \pmod{S} implies x+y≤x′+yx + y \leq x' + y and x′+y≤x+yx' + y \leq x + y for all y∈My \in M, that is x+y≡x′+y(modS)x + y \equiv x' + y \pmod{S}. In other words the equivalence relation SS is compatible with addition in MM (I, p. 11). Thus the quotient by SS of the addition law on MM, together with the ordering associated to ≤\leq, gives M/SM/S the structure of an ordered monoid. In the case where MM is a pre-ordered group, the group M/SM/S is the quotient of MM by the subgroup consisting of all elements xx such that x≤0x \leq 0 and 0≤x0 \leq x.

3. Positive elements

Let GG be a pre-ordered group with pre-order relation ≤\leq; if 0≤x0 \leq x and 0≤y0 \leq y then we deduce that y≤x+yy \leq x + y by (POM), and so 0≤x+y0 \leq x + y by transitivity; this says that the set G+G_+ of elements x∈Gx \in G such that 0≤x0 \leq x is closed under addition; moreover, the relation x≤yx \leq y is equivalent to 0≤y−x0 \leq y - x, that is to y−x∈G+y - x \in G_+. Conversely:

Proposition 3

— If PP is a subset of an abelian group GG, containing 00 and such that P+P⊂PP + P \subset P, then the relation y−x∈Py - x \in P is a pre-order relation compatible with the group structure of GG. For this relation to make GG an ordered group it is necessary and sufficient to have P∩(−P)={0}P \cap (-P) = \{0\}; for GG to be a totally ordered group under this ordering it is necessary and sufficient to have in addition P∪(−P)=GP \cup (-P) = G.

Cited by 2

Cited by Proposition 3, AC VI § 4, Example

It is immediate that the relation y−x∈Py - x \in P is reflexive and transitive, and (if it is written x≤yx \leq y) satisfies the axiom POM. To prove the second assertion it is enough to remark that P∩(−P)P \cap (-P) is the subgroup G′G' consisting of elements xx such that 0≤x0 \leq x and x≤0x \leq 0. Finally, to say that GG is totally ordered means that, for each pair of elements x,yx, y of GG, at least one of the elements x−y,y−xx - y, y - x belongs to PP, which completes the proof.

Definition 3

In an ordered group GG, any element xx such that 0≤x0 \leq x (resp. x≤0x \leq 0) is called a positive (resp. negative) element.

Note that 00 is the only element which is both positive and negative ; any element xx such that 0<x0 < x (resp. x<0x < 0) is called strictly positive (resp. strictly negative).

Example

— In the additive group Z×Z\mathbf{Z} \times \mathbf{Z}, let PP be the set of elements (x,y)(x, y) satisfying two inequalities ax+by≥0ax + by \geq 0 and cx+dy≥0cx + dy \geq 0, where a,b,c,da, b, c, d are integers (* or real numbers *) such that ad−bc≠0ad - bc \neq 0; the « cone » PP satisfies the first two conditions of Prop. 3. In this way various orderings compatible with the group structure of Z×Z\mathbf{Z} \times \mathbf{Z} can be defined ; the group is not totally ordered under any of these orderings.

Cites 1

Cites Proposition 3

Remark

— By virtue of the condition P+P⊂PP + P \subset P, the relation x≥0x \geq 0 in an ordered group implies that nx≥0nx \geq 0 for every natural number nn. If in addition the positive element xx of GG has finite order nn, then −x=(n−1)x-x = (n-1)x is positive ; since

P∩(−P)={0}, P \cap (-P) = \{0\},

this implies that x=0x = 0. In particular, if every element of GG has finite order, then P={0}P = \{0\}; the relation x≤yx \leq y is then equivalent to x=yx = y (the discrete ordering).

4. Filtered groups

Recall (Set Theory, III, p. 145) that an ordered set GG is right (resp. left) filtered\footnote{1} if for each pair (x,y)(x, y) of elements of GG there exists z∈Gz \in G such that x≤zx \leq z and y≤zy \leq z (resp. z≤xz \leq x and z≤yz \leq y). Every right filtered ordered group GG is also left filtered and conversely : indeed, since there exists z∈Gz \in G such that −x≤z-x \leq z and −y≤z-y \leq z, we have −z≤x-z \leq x and −z≤y-z \leq y (VI, p. 3, Cor.). We will therefore speak simply of filtered groups.

Proposition 4

For an ordered group GG to be filtered it is necessary and sufficient that it be generated by its positive elements, that is that every element be the difference of two positive elements.

Indeed if GG is filtered, then for each x∈Gx \in G there exists a positive element zz such that x≤zx \leq z, and xx is the difference of the two positive elements zz and z−xz - x. Conversely, if x=u−vx = u - v and y=w−ty = w - t with u,v,w,tu, v, w, t positive, then the element u+wu + w is greater than xx and greater than yy.

1 The terminology used in loc. cit. is directed.

Proposition 5

— If (xi)(x_i) is a finite family of elements of a filtered group G, then there exists z∈Gz \in G such that xi+zx_i + z *is positive for each i.

If xi=ui−vix_i = u_i - v_i, with uiu_i and viv_i positive, it is enough to take zz to be the sum of the family (vi)(v_i).

5. Divisibility relations in a field

Here we will define certain ordered groups which play an important part in algebra. The notation usually used for these groups is multiplicative; the application of results previously obtained in additive notation to these groups thus presupposes the translation to multiplicative notation — which should present no difficulty to the reader. Throughout this section, the letter A will denote an integral domain and K its field of fractions (l, p. 116).

In the multiplicative group K∗\mathbf{K}^* of nonzero elements of K\mathbf{K}, the set P of nonzero elements of A is closed under multiplication, since A is a ring. Hence it defines a pre-order relation on K∗\mathbf{K}^* by x−1y∈Px^{-1}y \in P, that is « there exists z∈Pz \in P such that y=zxy = zx », which makes K∗\mathbf{K}^* apre-ordered group (written multiplicatively) (VI, p. 3, Prop. 3). Extending the terminology relating to elements of A (I, p. 97) to K∗\mathbf{K}^*, the relation x−1y∈Px^{-1}y \in P can also be expressed as : x divides y, or x is a divisor of y, or y is a multiple of x (relative to the ring A); and we will call the relation x−1y∈Px^{-1}y \in P the divisibility relation in K∗\mathbf{K}^* relative to the ring A. The relation « x 'divides y » is denoted x∣yx \mid y, and its negation x∤yx \nmid y. The elements of P are none other than the multiples of 1.

Remark 1

The divisibility relation in K∗\mathbf{K}^* depends in an essential way on the particular ring A. If A=KA = K we obtain the « trivial » relation, under which x∣yx \mid y for every pair (x,y)(x, y) of elements of K∗\mathbf{K}^*. Let p and q be prime numbers; the rational numbers r/s whose denominators are not multiples of pp (resp. q) form a subring Z(p)\mathbf{Z}_{(p)} (resp. Z(q)\mathbf{Z}_{(q)}) of Q\mathbf{Q}; the divisibility relations in Q∗\mathbf{Q}^* relative to these two rings are distinct if p≠qp \neq q, the number p/qp/q being a multiple of 1 under one relation but not under the other.

Remark 2

We will sometimes extend the definition of the relation x∣yx \mid y to pairs of elements of K\mathbf{K} (rather than only K∗\mathbf{K}^*), this relation being taken to mean « there exists z∈Az \in A such that y=zxy = zx »; hence we will have x∣0x \mid 0 for all x∈Kx \in K. This allows us to state the following results without restrictions : if x∣yx \mid y and x∣zx \mid z then x∣(y−z)x \mid (y - z); if x∣yx \mid y and x∤zx \nmid z then x∤(y−z)x \nmid (y - z). In the same way we can extend all the corresponding terminology.

To obtain an order relation (No. 2) from the divisibility relation we have to pass to the quotient group of K∗\mathbf{K}^* by the subgroup A∗\mathbf{A}^* of elements x∈K∗x \in \mathbf{K}^* such that x∣1x \mid 1 and 1∣x1 \mid x; these are the elements of P which are divisors of 1, that is to say the invertible elements of AA; by abuse of language they are often called the units of the ring A. The quotient group K∗/A∗\mathbf{K}^*/\mathbf{A}^* is then an ordered group. Two elements xx and yy of K∗\mathbf{K}^* which belong to the same coset of A∗\mathbf{A}^* are said to be associate; this means that x∣yx \mid y and y∣xy \mid x. If on the other hand x divides y without y dividing x, we say that x strictly divides y, or that x is a strict divisor of y, or that y is a strict multiple of x.

Note that K∗/A∗K^*/A^* is a filtered group, since K is the field of fractions of A (VI, p. 4, Prop. 4).

To say that two elements x and y of K∗K^* are associate amounts to saying that they have the same multiples in K, by virtue of the transitivity of the divisibility relation. For all x∈Kx \in K, we will denote by AxAx the set of all elements zx with z∈Az \in A; the set AxAx is a submodule of K regarded as an A-module. Extending the terminology relating to the case where x∈Ax \in A, we will call it a principal fractional ideal of the field K relative to the ring A. In contrast the ideals of the ring A are said to be integral.

Note that if A≠KA \neq K then a principal fractional ideal ≠{0}\neq \{0\} is not an ideal of K considered as a ring.

The principal fractional ideal AxAx is also denoted (x). We will write x≡0x \equiv 0 (mody) for x∈Ayx \in Ay and x≡x′x \equiv x' (mody) for x−x′∈Ayx - x' \in Ay; if x≡x′x \equiv x' (mody) then zx≡zx′zx \equiv zx' (mod zyzy) for any z∈Kz \in K.

Note that x≡x′x \equiv x' (mod y) does not imply zx≡zx′zx \equiv zx' (mod y) unless z∈Az \in A. Thus in Q\mathbf{Q}, relative to Z\mathbf{Z}, we have 4≡24 \equiv 2 (mod 2) but not 2≡12 \equiv 1 (mod 2).

The relation x∣yx | y is obviously equivalent to (x)⊃(y)(x) \supset (y). The mapping x↦(x)x \mapsto (x) from K∗K^* onto the set P∗\mathscr{P}^* of principal fractional ideals ≠(0)\neq (0) of K thus defines, on passing to the quotient, a bijective mapping from K∗/A∗K^*/A^* onto P∗\mathscr{P}^*; translating the group structure of K∗/A∗K^*/A^* to P∗\mathscr{P}^* by means of this mapping, we are led to define the product of two principal fractional ideals (x) and (y) to be the ideal (xy ), which depends only on (x) and (y). Equipped with this law and the order relation (x)⊃(y)(x) \supset (y), the set P∗\mathscr{P}^* is an ordered group, isomorphic to K∗/A∗K^*/A^*. By convention we will identify P∗\mathscr{P}^* with K∗/A∗K^*/A^* via the above map.

Note that the relation « x divides y » which, in the case of positive integers, implies that x is smaller than y, corresponds to the inclusion (x)⊃(y)(x) \supset (y), in which the ideal (x) is « greater » than the ideal (y). We can keep this « order reversal » in mind by noting that for example 7 has a more multiples » than 91.

If we extend the relation x∣yx | y to all elements of K, this relation is still equivalent to (x)⊃(y)(x) \supset (y) in the set P\mathscr{P} of all principal fractional ideals of K (in which (0)(0) is the smallest element under the relation of inclusion).

As in the previous sections, we will generally be using additive notation in the sequel. However, terminology relating to divisibility will be introduced following the corresponding additive terminology, in paragraphs preceded by the sign (DIV) (in which it is understood that the notation used is that of the present section). In order to make the reader's task easier, certain results will be translated into the language of divisibility, the translation of Prop. 7, for example, being denoted « PROPOSITION 7 (DIV) ».

6. Elementary operations on ordered groups

Let H be a subgroup of an ordered group G; it is clear that the restriction to H of the ordering of G is compatible with the group structure of H; we will always take H to be ordered in this way, unless otherwise stated. If P is the set of positive elements of G then the set of positive elements of H is H ∩ P.

Let (Gα)(G_\alpha) be a family of ordered groups; according to the definition of the product of ordered sets (Set Theory, III, p. 137) the product group G = ∏Gα\prod G_\alpha is equipped with an ordering, the relation « (xα)≤(yα)(x_\alpha) \leq (y_\alpha) » between two elements of G being by definition the same as « xα≤yαx_\alpha \leq y_\alpha for all α ». It is immediate that this ordering is compatible with the group structure of G; this ordering makes G an ordered group which we call the product of the ordered groups GαG_\alpha. The positive elements of G are those elements all of whose components are positive. In the case where all the factors GαG_\alpha are identical to the same ordered group H, then G is the group H' of maps from the index set I into H, the relation « f ≤ g » between two maps from I into H being the same as « f(α) ≤ g(α) for all α ∈ I »; the positive maps are those which take only positive values. The direct sum of a family (Gα)(G_\alpha) of ordered groups is defined as an ordered subgroup of their product (II, p. 202).

Let (Gi)i∈I(G_i)_{i \in I} be a family of ordered groups whose index set I is well ordered: recall (Set Theory, III, p. 157) that an order relation, called the lexicographic ordering, is defined on the product set G = ∏Gi\prod G_i, the relation « (xi)<(yi)(x_i) < (y_i) » between two elements of G being by definition the same as « if β is the smallest of the indices i such that x_i ≠ y_i, then x_β < y_β ». Recall that the product of a well ordered family of totally ordered sets is totally ordered under the lexicographic ordering. In the general case, the lexicographic ordering on G is compatible with its group structure, as is immediately verified; equipped with this ordering, the group G is thus an ordered group, called the lexicographic product of the well ordered family of ordered groups (Gi)(G_i).

Remark 1

In the commonest cases, the well ordered index set I will be a finite interval (1, n) in N.

Remark 2

The set of positive elements of the lexicographic product G consists of 0 and those nonzero elements whose nonzero component of least index is positive.

7. Increasing homomorphisms of ordered groups

Let G and G' be two ordered groups; among the homomorphisms f from the underlying additive group of G into that of G', it makes sense to consider the increasing maps, that is those such that x ≤ y implies f(x) ≤ f(y). Because of the relation f(y−x)=f(y)−f(x)f(y - x) = f(y) - f(x) it follows that the increasing homomorphisms from G into G' are characterised by the fact that the image of a positive element of G under such a homomorphism is a positive element of G'; if P (resp. P') denotes the set of positive elements of G (resp. G'), this can be written f(P)⊂P′f(P) \subset P'. It is clear that the canonical injection of a subgroup G into an ordered group G', and the projection of a product of ordered groups onto one of its factors, are increasing homomorphisms.

An isomorphism (VI, p. 2) ff from an ordered group G onto an ordered group G' is a bijective homomorphism from G onto G' such that both ff and the inverse homomorphism are increasing, in other words f(P)=P′f(P) = P'.

It can happen that an isomorphism from the underlying group of G onto that of G' can be increasing without the inverse isomorphism also being increasing. This is the case, for example, if G=G′G = G', if ff is the identity map on G, and if P⊂P′P \subset P' but P≠P′P \neq P'. Thus in Z\mathbf{Z} we can take P′P' to be the set of (ordinary) positive integers and P to be the set of even positive integers.

(DIV) Let K be the field of rational functions F2(X)F_2(X) over the field F2F_2 of order 2. The divisibility relations relative to the rings F2[X]=A′F_2[X] = A' and F2[X2,X3]=AF_2[X^2, X^3] = A define two distinct ordered group structures on K∗K^*, such that A⊂A′A \subset A' (they are ordered group structures since 1 is the only unit in A and the only unit in A').

8. Suprema and infima in an ordered group

Recall (Set Theory, III, p. 141) that if the set of upper bounds of a subset F of an ordered set E (that is to say the set of z∈Ez \in E such that x≤zx \leq z for all x∈Fx \in F) has a least element a, then this element, which is then unique, is called the supremum of F. If F is the set of elements in a family (xi)i(x_i)_i, of elements of E, its supremum, if it exists, is denoted sup⁡xi\sup x_i (or sup⁡xi\sup x_i or simply sup⁡(xi)\sup(x_i)) ; if it is a finite family (xi)(x_i) (1≤i≤n1 \leq i \leq n), this supremum is also denoted sup⁡LEI(x1,…,xn)\sup_{\text{LEI}} (x_1, \ldots, x_n). The infimum is defined in an analogous manner and is denoted inf. The operations sup and inf are associative and commutative.

Recall (Set Theory, loc. cit.) that if F is a subset of an ordered set E, and (xi)(x_i) a family of elements of F, then the existence of sup⁡(xi)\sup(x_i) in E (which may be denoted sup⁡E(xi)\sup_E(x_i)) does not imply the existence of a supremum of the xix_i in F (which may be denoted sup⁡F(xi)\sup_F(x_i) when it does exist); if both exist we know only that sup⁡E(xi)≤sup⁡F(xi)\sup_E(x_i) \leq \sup_F(x_i); however if sup⁡E(xi)\sup_E(x_i) exists and belongs to F, then sup⁡F(xi)\sup_F(x_i) exists and is equal to sup⁡E(xi)\sup_E(x_i). For example, in the polynomial ring A=K[X,Y]A = K[X, Y] (K a field), the principal ideals AX and AY have the ideal AX+AYAX + AY as supremum (under the relation ⊂\subset) in the ordered set of ideals of A, but have the ideal A as supremum in the set of all principal ideals of A.

(DIV) An element d of K∗K^* is called a greatest common divisor, or gcd for short, of a family (xi)(x_i) of elements of K∗K^*, if the principal fractional ideal (d) is the supremum in P∗\mathscr{P}^* (under the relation ⊂\subset) of the family of ideals ((xi))((x_i)), or in other words if the relation z∣dz \mid d for z∈K∗z \in K^* is equivalent to « z∣xiz \mid x_i for all ii ». In the same way we will say that m∈K∗m \in K^* is a least common multiple or an lcm of the family (xi)(x_i) if (m)(m) is the infimum in P∗\mathscr{P}^* of the family of ideals ((xi))((x_i)), that is if m | z is equivalent to « x_i | z for all i ». It amounts to the same thing to say that (m)=∩i(xi)(m) = \cap_{i} (x_i); indeed, the condition xi∣zx_i | z for all ii is equivalent to z∈Axz \in A x, for all ii, that is to z∈∩i(xi)z \in \cap_{i} (x_i), and the condition m∣zm | z is equivalent to z∈(m)1z \in (m)^1.

Note that if a principal fractional ideal (d) satisfies (d)=∑(xi)(d) = \sum (x_i) then d is a gcd of the family (xi)(x_i); but conversely a gcd of (xi)(x_i) does not necessarily satisfy the above condition (cf. VI, p. 33, Ex. 24).

The gcd and lcm, if they exist, are well defined modulo the subgroup U of units of K∗K^*, that is two gcd's (or two lcm's) of a given family are associate ; by abuse of language we will often write gcd(xi)(x_i) and lcm(xi)(x_i) for any of the gcd's or lcm's of the family (xi)(x_i), whenever such elements exist.

(DIV) By abuse of language we sometimes extend the notion of gcd to a family (xi)(x_i) of elements of K, some of which may be zero ; this gcd is again defined to be an element d such that the relation z∣dz | d is equivalent to « z∣xiz | x_i for all ii »; clearly d is 0 if all the xix_i are zero ; otherwise d is a gcd of the nonzero xix_i. Similarly the lcm of a family, some elements of which are zero, is 0.

In an ordered group G, an immediate consequence of the invariance under translation of the ordering (VI, p. 2, Prop. 2) is that

(1)

sup⁡(z+xi)=z+sup⁡(xi) \sup(z + x_i) = z + \sup(x_i)
in the sense that, whenever one side is defined, then so is the other and the two are equal. Similarly, it follows from the fact that the map x↦−xx \mapsto -x takes the ordering of G to the opposite ordering (VI, p. 3, Cor.) that

(2)

inf⁡(−xi)=−(sup⁡(xi)), \inf(-x_i) = -(\sup(x_i)),
this relation being understood in the same sense as the previous one.

Proposition 6

— Let (xi)i∈A,(yβ)β∈B(x_i)_{i \in A}, (y_\beta)_{\beta \in B} be two families of elements of an ordered group G, each having a supremum. Then the family (xα+yβ)(α,β)∈A×B(x_\alpha + y_\beta)_{(\alpha, \beta) \in A \times B} has a supremum, and

sup⁡(α,β)∈A×B(xα+yβ)=sup⁡α∈Axα+sup⁡β∈Byβ. \sup_{(\alpha, \beta) \in A \times B} (x_\alpha + y_\beta) = \sup_{\alpha \in A} x_\alpha + \sup_{\beta \in B} y_\beta.
Indeed, from xα+yβ≤zx_\alpha + y_\beta \leq z for all α\alpha and β\beta, we deduce sup⁡(xi)+sup⁡(yβ)≤z\sup(x_i) + \sup(y_\beta) \leq z for all β\beta, and hence sup⁡(xi)+sup⁡(yβ)≤z\sup(x_i) + \sup(y_\beta) \leq z.

Cites 1

Cites I § 8. Rings

1 When A is the ring of integers (resp. the polynomial ring in one indeterminate with coefficients in a field), these definitions coincide with those of I, p. 112 (resp. IV, p. 12, Def. 1).

9. Lattice ordered groups

Recall that an ordered set in which every non empty finite subset has a supremum and an infimum is called a lattice (E, III, p. 13). It is clear that a product of lattice ordered groups, and in particular a product of totally ordered groups, is a lattice ordered group. In contrast a subgroup of a lattice ordered group is not necessarily lattice ordered.

Thus, in the product ordered group Z×ZZ \times Z, the « antidiagonal » (set of pairs (n,n′)(n, n') such that n+n′=0n + n' = 0) is ordered with the discrete ordering, and hence is not a lattice ordered group. The additive group of polynomials in one real variable (VI, p. 1, example 2) is a filtered group (since both p(x)p(x) and q(x)q(x) are less than (p(x))2+(q(x))2+1(p(x))^2 + (q(x))^2 + 1) which one can show is not lattice ordered. \

Proposition 7

— *If x and y are two elements of an ordered group G, and if one of the elements inf⁡(x,y)\inf(x, y), sup⁡(x,y)\sup(x, y) exists, then so does the other, and x+y=inf⁡(x,y)−sup⁡(x,y)x + y = \inf(x, y) - \sup(x, y).

Indeed, according to the relations (1) and (2) (VI, p. 9) we have

sup⁡(a−x,a−y)=a+sup⁡(−x,−y)=a−inf⁡(x,y), \sup(a - x, a - y) = a + \sup(-x, -y) = a - \inf(x, y),

and it suffices to take a=x+ya = x + y.

Proposition 7

— *If a,b∈K∗a, b \in K^*, and if d is a gcd of a and b, and m an lcm of a and b, then the product dm is an associate of ab.

Cited by 1

Cited by Proposition 8

Proposition 8

— *Let P be the set of positive elements of an ordered group G. For G to be lattice ordered, it is necessary and sufficient that G=P−PG = P - P, and that in addition P, equipped with the induced ordering, satisfies one or other of the following conditions:

Cites 1

Cites Proposition 7

a) Each pair of elements of P has a supremum in P. b) Each pair of elements of P has an infimum in P.

The necessity of these conditions is obvious : indeed the relation G=P−PG = P - P says that G is filtered (VI, p. 4, Prop. 4); on the other hand the supremum and infimum in G of two elements of P are positive, so are also their supremum and infimum in P.

Conversely, note first that under hypothesis a) (resp. b)), every pair of elements x, y of P has a supremum (resp. infimum) in G equal to its supremum a (resp. infimum b) in P. This is obvious for a, since any upper bound for x and y is positive; for b, let z∈Gz \in G be a lower bound for x and y; then there exists u∈Pu \in P such that z+u∈Pz + u \in P, since G=P−PG = P - P; now inf⁡P(x+u,y+u)\inf_P(x + u, y + u) is greater than b+ub + u, and so is of the form b+c+ub + c + u (c≥0c \geq 0); since b+cb + c is less than x and less than y, we have c=0c = 0; hence inf⁡P(x+u,y+u)=b+u\inf_P(x + u, y + u) = b + u, which implies z+u≤b+uz + u \leq b + u, thus z≤bz \leq b and b is certainly the infimum of x and y in G. Now if x and y are arbitrary elements of G we translate them into P: let v∈Pv \in P be such that x+vx + v and y+vy + v are positive (VI, p. 5, Prop. 5); under hypothesis a) (resp. b)) there exists a supremum (resp. infimum) for x+vx + v and y+vy + v in PP, and hence also in GG by what we have just seen; by translation xx and yy have a supremum (resp. infimum) in GG; the existence of one of these implies the other, by Prop. 7, and this shows that the conditions are sufficient.

10. The decomposition theorem

Theorem 1 (decomposition theorem)

— Let (xi)1≤i≤p(x_i)_{1 \leq i \leq p} and (yj)1≤j≤q(y_j)_{1 \leq j \leq q} be two finite sequences of positive elements of a lattice ordered group GG such that

∑i=1pxi=∑j=1qyj; \sum_{i=1}^p x_i = \sum_{j=1}^q y_j ;
then there exists a double sequence (zij)1≤i≤p,1≤j≤q(z_{ij})_{1 \leq i \leq p, 1 \leq j \leq q} of positive elements of GG such that xi=∑j=1qzijx_i = \sum_{j=1}^q z_{ij} for all ii, and yj=∑i=1pzijy_j = \sum_{i=1}^p z_{ij} for all jj.

Cited by 1

Cited by Lemma 3, LIE II § 1

1) Let us prove the theorem first for the case p=q=2p = q = 2. Let x,x′,y,y′x, x', y, y' be positive elements of GG such that x+x′=y+y′x + x' = y + y', and put a=sup⁡(0,x−y′)a = \sup(0, x - y'). Since

x−y′=y−x′ x - y' = y - x'
is less than xx and less than yy, it follows that b=x−ab = x - a and c=y−ac = y - a are positive, as is d=a−(x−y′)d = a - (x - y'). Also we have
x=a+b,x′=c+d,y=a+candy′=b+d. x = a + b, \quad x' = c + d, \quad y = a + c \quad \text{and} \quad y' = b + d .

2) Let us now show that if the theorem is true for p<mp < m and q=nq = n (m>2, n≥2m > 2,\ n \geq 2) then it is true for p=mp = m and q=nq = n. By hypothesis we have x,+∑i=1m−1xi=∑j=1nyjx, + \sum_{i=1}^{m-1} x_i = \sum_{j=1}^n y_j. The theorem being true for p=2p = 2 and q=nq = n, there exist two finite sequences (zj′),(zj′′)(z_j'), (z_j'') of nn positive terms such that ∑i=1m−1xi=∑j=1nzj′\sum_{i=1}^{m-1} x_i = \sum_{j=1}^n z_j', xm=∑j=1nzj′′x_m = \sum_{j=1}^n z_j'', and yj=zj′+zj′′y_j = z_j' + z_j'' for 1≤j≤n1 \leq j \leq n. On the other hand, since the theorem is true for p=m−1p = m - 1 and q=nq = n, there exists a double sequence (uij)1≤i≤m−1,1≤j≤n(u_{ij})_{1 \leq i \leq m-1, 1 \leq j \leq n} such that xi=∑j=1nuijx_i = \sum_{j=1}^n u_{ij} for 1≤i≤m−11 \leq i \leq m - 1, and zj′=∑i=1m−1uijz_j' = \sum_{i=1}^{m-1} u_{ij} for 1≤j≤n1 \leq j \leq n. Putting

zij=uijfor1≤i≤m−1,andzmj=zj′′(1≤j≤n), z_{ij} = u_{ij} \quad \text{for} \quad 1 \leq i \leq m - 1 , \quad \text{and} \quad z_{mj} = z_j'' \quad (1 \leq j \leq n) ,
we certainly obtain a double sequence satisfying the conditions of the theorem.

3) By interchanging the xix_i and the yjy_j we see in the same way that, if the theorem is true for p=mp = m and q<nq < n (m≥2, n>2m \geq 2,\ n > 2), then it is true for p=mp = m and q=nq = n. The theorem is thus proved by double induction starting from the case p=q=2p = q = 2, for it is trivially true whenever p≤1p \leq 1 or q≤1q \leq 1.

Corollary

— Let y,x1,x2,…,xny, x_1, x_2, \ldots, x_n be n+1n + 1 positive elements of GG such that y≤∑i=1nxiy \leq \sum_{i=1}^n x_i; then there exist nn positive elements yiy_i (1≤i≤n1 \leq i \leq n) such that yi≤xiy_i \leq x_i and y=∑i=1nyiy = \sum_{i=1}^n y_i.

It is sufficient to apply theorem 1 to the sequence (xi)(x_i) and the sequence consisting of the two elements yy and z=(∑i=1nxi)−yz = \left( \sum_{i=1}^n x_i \right) - y.

11. Positive and negative parts

Definition 4

— In a lattice ordered group GG the positive part (resp. negative part, absolute value) of an element x∈Gx \in G is the element sup⁡(x,0)\sup(x, 0) (resp. sup⁡(−x,0)\sup(-x, 0), sup⁡(x,−x)\sup(x, -x)), which is denoted x+x^+ (resp. x−x^-, ∣x∣|x|).

Despite its name, the negative part x−x^- is a positive element.

Clearly x−=(−x)+x^- = (-x)^+ and ∣−x∣=∣x∣|-x| = |x|. Let us also note the following formulae, the first of which is an immediate consequence of the definitions and of the invariance of the ordering under translation, and the second of which follows from the first by Prop. 7 of VI, p. 10:

{sup⁡(x,y)=x+(y−x)+inf⁡(x,y)=y−(y−x)+ \begin{cases} \sup(x, y) = x + (y - x)^+ \\ \inf(x, y) = y - (y - x)^+ \end{cases}

Proposition 9

— a) For each element xx of a lattice ordered group GG we have x=x+−x−x = x^+ - x^- and inf⁡(x+,x−)=0\inf(x^+, x^-) = 0.

b) For every expression of xx as the difference of two positive elements, say x=u−vx = u - v, we have u=x++wu = x^+ + w and v=x−+wv = x^- + w with w=inf⁡(u,v)w = \inf(u, v). In particular if inf⁡(u,v)=0\inf(u, v) = 0 then u=x+u = x^+ and v=x−v = x^-.

c) The relation « x≤yx \leq y » is equivalent to « x+≤y+x^+ \leq y^+ and x−≥y−x^- \geq y^- ».

d) We have ∣x∣=x++x−≥0|x| = x^+ + x^- \geq 0.

e) For any xx and yy in GG, we have the inequality ∣x+y∣≤∣x∣+∣y∣|x + y| \leq |x| + |y|, and more generally ∣∑i=1nxi∣≤∑i=1n∣xi∣\left| \sum_{i=1}^n x_i \right| \leq \sum_{i=1}^n |x_i| for any finite family (xi)(x_i) of elements of GG.

f) For any xx and yy in GG we have ∣∣x∣−∣y∣∣≤∣x−y∣||x| - |y|| \leq |x - y|.

We will prove a) and b) simultaneously. If x=u−vx = u - v with uu and vv positive, then uu a xx, so u≥sup⁡(x,0)=x+u \geq \sup(x, 0) = x^+, and w=u−x+w = u - x^+ is positive. On the other hand we have

x+−x=sup⁡(x,0)−x=sup⁡(x−x,−x)=x x^+ - x = \sup(x, 0) - x = \sup(x - x, -x) = x

from which it follows that x=x+−x−x = x^+ - x^-, and v−x−=wv - x^- = w. If z≤x−z \leq x^- then z≤x+−xz \leq x^+ - x, and so x≤x+−zx \leq x^+ - z; if also z≤x+z \leq x^+, then xz−zx^z - z is positive, and so xi≤x+−zx^i \leq x^+ - z by definition of xix^i. Hence we have z≤0z \leq 0, which implies inf⁡(x+,x−)=0\inf(x^+, x^-) = 0, whence by translation inf⁡(u,v)=w\inf(u, v) = w.

c) The relation x≤yx \leq y implies sup⁡(y,0)≥x\sup(y, 0) \geq x and sup⁡(y,0)≥0\sup(y, 0) \geq 0, hence xi≤yix^i \leq y^i; similarly if −y≤−x-y \leq -x we deduce x−≥y−x^- \geq y^-. The converse implication follows immediately from x=x+−xx = x^+ - x and y=y+−yy = y^+ - y.

d) Since x≤x+x \leq x^+ and −x≤x−-x \leq x^-, it is clear that

∣x∣=sup⁡(x,−x)≤x++x |x| = \sup(x, -x) \leq x^+ + x
Conversely, if a≥xa \geq x and a≥−xa \geq -x then we deduce from c) that a+≥x+a^+ \geq x^+, a+≥x−a^+ \geq x^-, a−≤x−a^- \leq x^- and a−≤x+a^- \leq x^+; since a−a^- is positive and inf⁡(xi,x−)=0\inf(x^i, x^-) = 0, the last two inequalities imply a−=0a^- = 0 and a=a+a = a^+; the first two inequalities then give a≥sup⁡(x+,x−)a \geq \sup(x^+, x^-), and sup⁡(x+,x−)\sup(x^+, x^-) is equal to x++x−x^+ + x^- by a) and by Prop. 7 of VI, p. 10.

e) Since xx and yy s ∣x∣|x| and yy, we have x+y≤∣x∣+∣y∣x + y \leq |x| + |y|; since −x≤∣x∣-x \leq |x| and −y≤∣y∣-y \leq |y|, we have −x−y≤∣x∣+∣y∣-x - y \leq |x| + |y|; whence the first inequality. The second follows by induction on n.

f) Replacing x and y by y and x - y in e), we obtain

∣x∣−∣y∣≤∣x−y∣ |x| - |y| \leq |x - y|
similarly we have ∣y∣−∣x∣≤∣y−x∣=∣x−y∣|y| - |x| \leq |y - x| = |x - y|; whence the stated result.

Remark

We deduce from d) that ∣x∣=0|x| = 0 implies x=0x = 0 (for x+x^+ and x−x^- are positive); thus x≠0x \neq 0 implies ∣x∣>0|x| > 0.

Proposition 9

— If the group P∗\mathscr{P}^* of principal fractional ideals of K is lattice ordered, then every element x of K∗K^* can be written in the form x=uv−1x = uv^{-1}, where u and v are elements of A such that 1=gcd⁡(u,v)1 = \gcd(u, v); for any other expression x=u′v′−1x = u'{v'}^{-1} of x as the quotient of two elements of A, we have u′=uwu' = uw and v′=vwv' = vw, where w is a gcd of u′,v′u', v'; in particular if 1=gcd⁡(u′,v′)1 = \gcd(u', v') then u' and v' are associates of u and v respectively.

Cited by 2

Cited by Corollary 5, Proposition 13

Such an expression uv−1uv^{-1} for an element x of K∗K^* is often called a reduced fraction.

12. Coprime elements

Definition 5

— In an ordered group two elements x and y are called coprime if inf⁡(x,y)=0\inf(x, y) = 0.

Cites 2

Cites II § 1. Modules, Definition 16, I § 4

In some cases it is natural to define two elements to be coprime if inf⁡(∣x∣,∣y∣)=0\inf(|x|, |y|) = 0 (cf. INT, II, § 1) or to introduce the corresponding terminology in divisibility theory. We shall not do so here.

Two coprime elements are necessarily positive. The positive and negative parts x+x^+ and x−x^- of x are coprime (VI, p. 12, Prop. 9, a)). The elements xix_i of a family (xi)i∈I(x_i)_{i \in I} are said to be setwise coprime if inf⁡i∈Ixi=0\inf_{i \in I} x_i = 0; if the xix_i are all ≥0\geq 0 then it is sufficient for there to exist a finite subset J of I such that the corresponding elements are setwise coprime. The elements of a family (xi)(x_i) are said to be pairwise coprime if inf⁡(xi,xj)=0\inf(x_i, x_j) = 0 for every pair (ℓ,κ)(\ell, \kappa) of distinct indices.

The xix_i can be setwise coprime without being pairwise coprime.

If x and y are coprime, we also say that x is coprime to y, or that y is coprime to X

(DIV) Two elements x and y of K are said to be coprime if the principal ideals (x) and (y) are nonzero and coprime in P∗\mathscr{P}^*; this amounts to saying that 1 is a gcd of x and y, and implies that x and y belong to A. For example the numerator and denominator of a reduced fraction are coprime. The notions of pairwise and setwise coprime elements are defined similarly.

(DIV) When x and y are coprime, they are often said to be « prime to one another »; it is convenient to avoid this terminology, which can lead to confusion with the notion of prime numbers (I, p. 50, Def. 16).

Proposition 10

Let x, y and z be three elements of an ordered group ; for x – z and y – z to be coprime, it is necessary and sufficient that z = inf (x, y).

Cited by 1

Cited by Corollary 5

Indeed the relations z=inf⁡(x,y)z = \inf(x, y) and 0=inf⁡(x−z,y−z)0 = \inf(x - z, y - z) are equivalent.

Proposition 10 (DIV). — Let a, b and c be three elements of K with c ≠0\neq 0; for the quotients ac−1ac^{-1} and bc−1bc^{-1} to be coprime, it is necessary and sufficient that c be a gcd of a and b.

Proposition 11

If (xi),(yj)(x_i), (y_j) are two finite families of positive elements of a lattice ordered group, then

Cited by 1

Cited by Corollary 1

inf⁡(∑ixi,∑jyj)≤∑i,jinf⁡(xi,yj) \inf \left( \sum_i x_i, \sum_j y_j \right) \leq \sum_{i,j} \inf(x_i, y_j)

Arguing by induction on the number of elements in the families (xi)(x_i) and (yj)(y_j), it is enough to prove that, if x, y and z are positive elements, then

inf⁡(x,y+z)≤inf⁡(x,y)+inf⁡(x,z). \inf(x, y + z) \leq \inf(x, y) + \inf(x, z).

Indeed, put t=inf⁡(x,y+z)t = \inf(x, y + z); by VI, p. 12, Cor., we can write t=t1+t2t = t_1 + t_2 with 0≤t1≤y0 \leq t_1 \leq y and 0≤t2≤z0 \leq t_2 \leq z; since t1t_1 and t2t_2 are positive, we also have t1≤xt_1 \leq x and t2≤xt_2 \leq x, whence t1≤inf⁡(x,y)t_1 \leq \inf(x, y) and t2≤inf⁡(x,z)t_2 \leq \inf(x, z).

Corollary 1

If x and y are two coprime elements, and z a positive element, of a lattice ordered group, then inf⁡(x,z)=inf⁡(x,y+z)\inf(x, z) = \inf(x, y + z).

Cites 1 · Cited by 1

Cites Proposition 11

Cited by Proposition 10, AC V § 1

Indeed inf⁡(x,y+z)≤inf⁡(x,z)\inf(x, y + z) \leq \inf(x, z) by Prop. 11, and inf⁡(x,z)≤inf⁡(x,y+z)\inf(x, z) \leq \inf(x, y + z) since y≥0y \geq 0, whence the corollary.

Corollary 2

— In a lattice ordered group, if x and y are coprime and if z ≥0\geq 0 and x≤y+zx \leq y + z, then x≤zx \leq z.

Corollary 3

— In a lattice ordered group, if x is coprime to y and to z, then it is also coprime to y+zy + z.

Cited by 1

Cited by Corollary 4

Corollary 4

— If (xi)1≤i≤n(x_i)_{1 \leq i \leq n} and (yj)1≤j≤m(y_j)_{1 \leq j \leq m} are two finite families of elements of a lattice ordered group G, such that each xix_i is coprime to each yjy_j, then x1+…+xnx_1 + \ldots + x_n is coprime to y1+…+ymy_1 + \ldots + y_m. This follows from Cor. 3 by induction on m and n.

Cites 1 · Cited by 2

Cites Corollary 3

Cited by Corollary 5, Proposition 12

Corollary 5

— For any integer n≥0n \geq 0 we have (nx)+=nx+(nx)^+ = nx^+ and (nx)−=nx−(nx)^- = nx^-; for all n∈Zn \in \mathbf{Z} we have ∣nx∣=∣n∣⋅∣x∣|nx| = |n| \cdot |x|. Indeed nx=nx+−nx−nx = nx^+ - nx^-; since x+x^+ and x are coprime, so are nx+nx^+ and nx−nx^- if n≥0n \geq 0 (Cor. 4); the first assertion follows by Prop. 9 b) of VI, p. 12. The second follows from the first by Prop. 9 d) in the case n≥0n \geq 0; the case n<0n < 0 follows from this as a result of the relation ∣−x∣=∣x∣|-x| = |x|.

Cites 2

Cites Proposition 9, Corollary 4

Proposition 11

— Suppose the set P∗\mathscr{P}^* is a lattice, and let (ai)(a_i), (bj)(b_j) be two finite families of elements of A. Then every gcd of ∏iai\prod_i a_i and ∏jbj\prod_j b_j divides the product ∏i,jgcd⁡(ai,bj)\prod_{i,j} \gcd(a_i, b_j).

Corollary 1

— If a, b, c are three elements of A such that a is coprime to b, then every gcd of a and c is also a gcd of a and bc.

Corollary 2 (Euclid’s lemma)

— Let a, b, c be three elements of A. If a is coprime to b and divides bc, then it divides c.

Corollary 3

— If x is coprime to y and to z, then it is coprime to yz.

Corollary 4

— If (xi)(x_i) and (yj)(y_j) are two finite families of elements of A such that each xix_i is coprime to each yjy_j, then the product of the xix_i is coprime to the product of the yjy_j.

Cited by 1

Cited by Corollary 5

Corollary 5

— If d is a gcd of x and y, then dnd^n is a gcd of xnx^n and yny^n for each positive integer n. Indeed xd1xd^1 and yd1yd^1 are coprime (Prop. 10 (DIV)), and hence so are xnd−nx^n d^{-n} and ynd−ny^n d^{-n} (Cor. 4).

Cites 2

Cites Corollary 4, Proposition 10

Proposition 12

— Let xix_i (1≤i≤n1 \leq i \leq n) be n pairwise coprime elements in a lattice ordered group. Then

sup⁡(x1,…,xn)=x1+⋯+xn. \sup(x_1, \ldots, x_n) = x_1 + \cdots + x_n .

Cites 1

Cites Corollary 4

This follows from the formula u+v=sup⁡(u,v)+inf⁡(u,v)u + v = \sup(u, v) + \inf(u, v) (VI, p. 10, Prop. 7) by induction on nn, using the fact that xix_i is coprime to x1+…+xi−1x_1 + \ldots + x_{i-1} for 2≤i≤n2 \leq i \leq n (Cor. 4 to Prop. 11).

Remark

Prop. 7 of VI, p. 10 also shows that a necessary and sufficient condition for xx and yy to be coprime is that x+y=sup⁡(x,y)x + y = \sup(x, y).

Proposition 12

— *Let aia_i be a finite number nn of pairwise coprime elements of A\mathbf{A}; then the product a1…ana_1 \ldots a_n, is an lcm of a1,…,ana_1, \ldots, a_n.

Proposition 13

— *In a lattice ordered group GG, let (xα)(x_\alpha) be a family having an infimum (resp. supremum), and let zz be an arbitrary element of GG; then the family (sup⁡(z,xα))(\sup(z, x_\alpha)) (resp. (inf⁡(z,xα))(\inf(z, x_\alpha))) has an infimum (resp. supremum) and

inf⁡α(sup⁡(z,xα))=sup⁡(z,inf⁡αxα)sup⁡α(inf⁡(z,xα))=inf⁡(z,sup⁡αxα) \begin{align*} \inf_{\alpha} (\sup(z, x_\alpha)) &= \sup \left( z, \inf_{\alpha} x_\alpha \right) \\ \sup_{\alpha} (\inf(z, x_\alpha)) &= \inf \left( z, \sup_{\alpha} x_\alpha \right) \end{align*}

respectively.

Cites 1 · Cited by 1

Cites Proposition 9

Cited by Remark

Suppose that the family (xα)(x_\alpha) has an infimum yy. We will show that sup⁡(z,y)\sup(z, y) is an infimum of the family (sup⁡(z,xα))(\sup(z, x_\alpha)).

Indeed sup⁡(z,xα)=z+(xα−z)+\sup(z, x_\alpha) = z + (x_\alpha - z)^+ and by translating we can reduce to the case z=0z = 0, that is we must show that (xα+)(x_\alpha^+) has an infimum equal to y+y^+. Since y≤xαy \leq x_\alpha we have y+≤xα+y^+ \leq x_\alpha^+ for all α\alpha (VI, p. 12, Prop. 9, c)). Conversely, if a≤xα+a \leq x_\alpha^+ for all aa, then it follows that a≤xα+xα−a \leq x_\alpha + x_\alpha^- (Prop. 9, a)); now y−≥xα−y^- \geq x_\alpha^- since y≤xαy \leq x_\alpha; hence a≤xα+y−a \leq x_\alpha + y^- for all aa, that is a≤y+y−=y+a \leq y + y^- = y^+.

The other formula follows by changing to the opposite order relation.

Corollary

— *If an element zz of a lattice ordered group GG is coprime to each element xαx_\alpha of a family having an infimum yy, then zz is coprime to yy.

This is an immediate consequence of the second formula (4).

Remark

Applying the formulae of Prop. 13 to a family of two elements (x,y)(x, y), we obtain the following formulae, which express the fact that each of the laws of composition sup, inf in a lattice ordered group is distributive with respect to the other :

sup⁡(z,inf⁡(x,y))=inf⁡(sup⁡(z,x),sup⁡(z,y))inf⁡(z,sup⁡(x,y))=sup⁡(inf⁡(z,x),inf⁡(z,y)). \begin{align*} \sup(z, \inf(x, y)) &= \inf(\sup(z, x), \sup(z, y)) \\ \inf(z, \sup(x, y)) &= \sup(\inf(z, x), \inf(z, y)) . \end{align*}

This distributivity property is peculiar to lattice ordered groups, and does not extend to arbitrary lattices, or even to lattice ordered monoids (cf. VI, p. 33, Ex. 24).

Cites 1

Cites Proposition 13

13. Irreducible elements

Definition 6

— *An element x of an ordered group G is said to be irreducible if it is a minimal element in the set of strictly positive elements of G.

Cites 1

Cites I § 4. Groups and groups with operators

Let x be an irreducible element of the ordered group G; if y is a positive element of G then the element inf(x, y), if it exists, can only be equal to x or to 0. Thus in a lattice ordered group G, every positive y is either greater than or coprime to the irreducible element x; in particular, two distinct irreducible elements are coprime.

(DIV) An element p of A is called irreducible if the ideal (p) is an irreducible element of the ordered group P∗\mathscr{P}^*; this says that p is neither zero nor invertible, and that any element of AA which divides p is associate either to p or to 1. If P∗\mathscr{P}^* is lattice ordered, then every a∈Aa \in A is either coprime to p or a multiple of p.

Examples (DIV). — 1) An integer p>0p > 0 is irreducible in Z\mathbf{Z} if and only if it is prime (I, p. 50).

2) A polynomial in one indeterminate over a field K is irreducible if and only if it is irreducible in the usual sense (IV, p. 13).

Proposition 14

— *For an element x>0x > 0 of an ordered group G to be irreducible it is sufficient that it satisfy the following property :

(P) The relations x≤y+z, y≥0, z≥0x \leq y + z,\ y \geq 0,\ z \geq 0 imply x≤yx \leq y or x≤zx \leq z. This condition is necessary when G is lattice ordered.

If G is lattice ordered and x is irreducible, we have just seen that y is either greater than x or coprime to x; in the latter case Cor. 2 of VI, p. 15 shows that z is greater than x. Conversely, suppose the condition is satisfied: if 0≤y≤x0 \leq y \leq x then it follows, by putting x=y+z (z≥0)x = y + z\ (z \geq 0), that either x≤yx \leq y or x≤zx \leq z; in the first case we have x=yx = y; in the second we have x≤x−yx \leq x - y, so y≤0y \leq 0 and so y=0y = 0; this shows that x is indeed irreducible.

Proposition 14

— *For a nonzero element p of A to be irreducible it is sufficient that it not be a unit, and that it cannot divide a product of two elements of A without dividing one or other of them. This condition is necessary if P∗\mathscr{P}^* is lattice ordered.

Cited by 2

Cited by Remark, Theorem 2

Remark

Proposition 14 (DIV) can also be expressed as follows: if p is a non zero element of A such that the ideal (p) is prime (I, p. 117, Def. 3) then p is irreducible; conversely, if P∗\mathscr{P}^* is lattice ordered and p is irreducible then the ideal (p) is prime.

Cites 2

Cites Definition 3, I § 9, Proposition 14

Proposition 15

— *In an ordered group G, let (pi)i∈I(p_i)_{i \in I} be a family of pairwise distinct positive elements of G satisfying condition (P) (and hence irreducible). Then the map

(ni)i∈I↦∑i∈Inipi (n_i)_{i \in I} \mapsto \sum_{i \in I} n_i p_i

is an isomorphism of the ordered group Z(I)\mathbf{Z}^{(I)}, the direct sum of the ordered groups ZZ (VI, p. 7) onto the ordered subgroup of GG generated by the pιp_\iota.

Cited by 1

Cited by Theorem 2

It is enough to show that the relation ∑ι∈Inιpι≥0\sum_{\iota \in I} n_\iota p_\iota \geq 0 is equivalent to nι≥0n_\iota \geq 0 for all ι\iota, for in particular the relation ∑ι∈Inιpι=0\sum_{\iota \in I} n_\iota p_\iota = 0 will imply nι=0n_\iota = 0 for all ι\iota, hence this will show that the family (pι)(p_\iota) is linearly independent. Now let I′I' (resp. I′′I'') be the finite subset of II consisting of those ι\iota such that nι>0n_\iota > 0 (resp. nι<0n_\iota < 0) ; we have

∑ι∈I′nιpι≥∑ι∈I′′(−nι)pι. \sum_{\iota \in I'} n_\iota p_\iota \geq \sum_{\iota \in I''} (-n_\iota)p_\iota .

In particular, for λ∈I′′\lambda \in I'', this implies that pλ≤∑ι∈I′nιpιp_\lambda \leq \sum_{\iota \in I'} n_\iota p_\iota, and it follows by induction from property (P) that we must have pλ≤pιp_\lambda \leq p_\iota for some ι∈I′\iota \in I'; since pιp_\iota is irreducible, this would imply pλ=pιp_\lambda = p_\iota, which is absurd. Hence I′I' is empty, which proves the proposition.

Theorem 2

— Let GG be a filtered group. Then the following properties are equivalent : a) GG is isomorphic to an ordered group of the form Z(I)\mathbf{Z}^{(I)}. b) GG is lattice ordered and satisfies the following condition : (MIN) Every nonempty set of positive elements of GG has a minimal element. c) GG satisfies condition (MIN) and every irreducible element of GG has property (P). d) GG is generated by its irreducible elements, and every irreducible element of GG has property (P).

Cites 2 · Cited by 1

Cites Proposition 14, Proposition 15

Cited by Lemma 1, AC VII § 1

Let us show first that a) implies b). The group Z(I)\mathbf{Z}^{(I)} is lattice ordered, as the direct sum of totally ordered groups. On the other hand let EE be a nonempty set of positive elements of Z(I)\mathbf{Z}^{(I)} and let x=∑nιeιx = \sum n_\iota e_\iota be an element of EE (where (eι)(e_\iota) denotes the natural basis of Z(I)\mathbf{Z}^{(I)}) ; there are a finite number ∏(nι+1)\prod (n_\iota + 1) of elements yy of Z(I)\mathbf{Z}^{(I)} such that 0≤y≤x0 \leq y \leq x, so the set FF of elements of EE which are less than or equal to xx is aa fortiori finite ; since it is nonempty, it contains a minimal element (Set Theory, III, p. 170, Cor. 2), which is clearly a minimal element of EE.

It is clear that b) implies c), by Prop. 14. Let us show that c) implies d). Since GG is filtered, it is enough (VI, p. 4, Prop. 4) to check that the set FF of positive elements of GG which are sums of irreducible elements is equal to G+−{0}G_+ - \{0\}. If this were not true, it would follow from (MIN) that the complement of FF in G+−{0}G_+ - \{0\} would have a minimal element aa; by definition aa is not irreducible, so is the sum of two strictly positive elements xx and yy; since x<ax < a and y<ay < a, these elements belong to FF, and so are sums of irreducible elements, and it follows that so is aa, which is a contradiction. Finally, d) implies a) by Prop. 15.

We will apply Th. 2 to the theory of divisibility in principal ideal domains (VII, p. 4) and in unique factorisation domains (AC, VII, § 3), as well as to the study of ideals in a Dedekind ring (AC, VII, § 2).

Exercises

See the exercises for § 1.

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