Bourbaki

Proposition 1

Tag 0BVG, permanent. Cite it as [Bourbaki, Tag 0BVG].

General Topology, chapter V, § 2. Measurement of magnitudes

Let E be a linearly ordered set with a smallest element ω; let I be a subset of E such that ω ∈ I and such that the relations x ∈ I, y ≤ x imply y ∈ I; let (x, y) → xy be a law of composition on E, defined for x ∈ I and y ∈ I. Then, if the axioms (GR_I), (GR_{II}), (GR_{III}) and (GR_{IV}) are satisfied, there exists a strictly increasing mapping f of I into the set R_+ of real numbers ≥ 0, such that

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

whenever x ∈ I, y ∈ I and xy ∈ I; moreover, the intersection of f(I) with every interval [0, f(b)] of R is dense in this interval, where b denotes any element of I.

Given any two elements x, y of I such that y ≠ ω, let us denote by (x : y) the largest integer n ≥ 0 such that y^n is defined and ≤ x (); this integer exists by (GR{IV}); if (x : y) = p, then y^{p+1} is defined and > x. If x ∈ I, y ∈ I and xy ∈ I, we have

(I)

(x:z)+(y:z)≤(xy:z)≤(x:z)+(y:z)+1. (x : z) + (y : z) \leq (xy : z) \leq (x : z) + (y : z) + 1.

For let (x : z) = p, (y : z) = q; then we have z^p ≤ x, z^q ≤ y; since xy ∈ I, z^p z^q is defined and belongs to I, therefore z^{p+q} is defined and z^{p+q} = z^p z^q ≤ xy; moreover, if z^{p+q+2} is defined, we have z^{p+q+2} > xy because z^{p+1} > x and z^{q+1} > y.

Next we establish the inequalities

(2)

{(x:y)(y:z)≤(x:z),((x:y)+1)((y:z)+1)≥(x:z)+1. \begin{cases} (x : y) (y : z) \leq (x : z), \\ ((x : y) + 1) ((y : z) + 1) \geq (x : z) + 1. \end{cases}

Let (x : y) = p and (y : z) = q; then y^p ≤ x and z^q ≤ y, so that (z^q)^p is defined and ≤ x; it belongs therefore to I; consequently z^{pq} is defined and we have z^{pq} = (z^q)^p ≤ x, from which the first inequality follows. On the other hand, if z^{(p+1)(q+1)} is defined, we have z^{(p+1)(q+1)} > x, because y^{p+1} > x and z^{q+1} > y; hence the second inequality.

Let X~\tilde{\mathscr{X}} denote the filter of sections of the ordered set of elements > ω in I, with respect to the relation ≥; the intervals ]ω, z[, where z runs through the set of all elements > ω, form a base of X~\tilde{\mathscr{X}}. Given two elements a and x of I such that a > ω, we shall show that the ratio (x:z)(a:z)\frac{(x : z)}{(a : z)}, which is defined for z ≤ a and is a rational number > 0,

(*) When E = I is the set of natural integers, the law of composition being addition, (x : y) is the integral part of x/y.

is a function of zz which has a limit with respect to F\mathfrak{F}. This is obvious if x=ωx = \omega, for then (x:z)=0(x : z) = 0 for all zz. If x>ωx > \omega, we shall show that the image G\mathscr{G} of F\mathfrak{F} under the mapping z→(x:z)(a:z)z \to \frac{(x : z)}{(a : z)} (restricted to the set of those z>ωz > \omega which are ≤x\leq x and ≤a\leq a) is a Cauchy filter base for the uniform structure of the multiplicative group R+∗\mathbf{R}_+^*, and therefore converges to a real number >0> 0. Note first that, u>ωu > \omega being given, (u:z)(u : z) has limit +∞+\infty with respect to F\mathfrak{F}: for there exists z>ωz > \omega such that z2n≤uz^{2^n} \leq u, so that (u:z)≥2n>n(u : z) \geq 2^n > n. Now take a number ε>0\varepsilon > 0 arbitrarily; there exists t>ωt > \omega such that (x:t)≥1/ε(x : t) \geq 1 / \varepsilon and (a:t)≥1/ε(a : t) \geq 1 / \varepsilon. Consider the double inequality

(x:t)(a:t)+1⋅(t:z)(t:z)+1≤(x:z)(a:z)≤(x:t)+1(a:t)⋅(t:z)+1(t:z), \frac{(x : t)}{(a : t) + 1} \cdot \frac{(t : z)}{(t : z) + 1} \leq \frac{(x : z)}{(a : z)} \leq \frac{(x : t) + 1}{(a : t)} \cdot \frac{(t : z) + 1}{(t : z)},

which follows immediately from the inequalities (2). There exists z0>ωz_0 > \omega such that z≤z0z \leq z_0 implies (t:z)≥1/ε(t : z) \geq 1 / \varepsilon, so that

1(1+ε)2(x:t)(a:t)≤(x:z)(a:z)≤(1+ε)2(x:t)(a:t), \frac{1}{(1 + \varepsilon)^2} \frac{(x : t)}{(a : t)} \leq \frac{(x : z)}{(a : z)} \leq (1 + \varepsilon)^2 \frac{(x : t)}{(a : t)},

which shows that G\mathscr{G} is a Cauchy filter base for the multiplicative uniformity.

Fix once and for all the element a>ωa > \omega (the "unit of measure") and for each x∈Ix \in I put

f(x)=lim⁡z∈F(x:z)(a:z). f(x) = \lim_{z \in \mathfrak{F}} \frac{(x : z)}{(a : z)}.

From what has already been proved, we have f(ω)=0f(\omega) = 0, f(x)>0f(x) > 0 for x>ωx > \omega, and f(a)=1f(a) = 1. If we divide the inequality (1) throughout by (a:z)(a : z) and pass to the limit with respect to F\mathfrak{F}, we see that

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

whenever x∈I,y∈Ix \in I, y \in I and xy∈Ixy \in I. Likewise, the relation x≤yx \leq y implies (x:z)≤(y:z)(x : z) \leq (y : z), whence by dividing by (a:z)(a : z) and passing to the limit we have f(x)≤f(y)f(x) \leq f(y), so that ff is increasing on II. We deduce that ff is strictly increasing on II; for if x<yx < y, there exists z>ωz > \omega such that xz≤yxz \leq y, whence f(xz)≤f(y)f(xz) \leq f(y); and since xz∈Ixz \in I,

f(x)+f(z)=f(xz)≤f(y); f(x) + f(z) = f(xz) \leq f(y);

but f(z)>0f(z) > 0, so that indeed f(x)<f(y)f(x) < f(y).

Finally, if b∈Ib \in I, the intersection of f(I)f(I) and the interval [0,f(b)][0, f(b)] of R\mathbf{R} is dense in this interval. For if nn is any integer >0> 0, there exists x>ωx > \omega such that f(x)≤2−nf(x) \leq 2^{-n} (take xx such that x2n≤ax^{2^n} \leq a); if pp is the smallest integer such that xp+1>bx^{p+1} > b, we have (p+1)f(x)>f(b)(p+1)f(x) > f(b) and qf(x)≤f(b)qf(x) \leq f(b) for 1≤q≤p1 \leq q \leq p; therefore every interval contained in [0,f(b)][0, f(b)] and of length >2−n> 2^{-n} contains at least one point of the form qf(x)=f(xq)∈f(I)qf(x) = f(x^q) \in f(I). The proof of Proposition 1 is therefore complete.

Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.

Cited by

Cites

Nothing.

Transcribed from the printed volume. Not translated.