Bourbaki

Proposition 2

Tag 0841, permanent. Cite it as [Bourbaki, Tag 0841].

Functions of a Real Variable, chapter I, § 2. THE MEAN VALUE THEOREM

Let ff be a real function which is finite and continuous on the closed bounded interval I=[a,b]I = [a, b] (where a<ba < b*) and has a right derivative* (finite or not) at all the points of the relative complement in [a,b)[a, b) of a countable subset AA of this interval. If fd′(x)≥0f'_d(x) \geq 0 at every point of [a,b[[a, b[ not belonging to AA, then one has f(b)≥f(a)f(b) \geq f(a); if, further, fd′(x)>0f'_d(x) > 0 for at least one point of [a,b[[a, b[, then f(b)>f(a)f(b) > f(a).

Let ε>0\varepsilon > 0 be arbitrary, and denote by (an)n≥1(a_n)_{n \geq 1} a sequence obtained by listing the countable set AA. Let JJ be the set of points y∈Iy \in I such that one has

f(x)−f(a)≥−ε(x−a)−ε∑an<x12n f(x) - f(a) \geq -\varepsilon(x - a) - \varepsilon \sum_{a_n < x} \frac{1}{2^n}

(1)

for all xx with a≤x≤ya \leq x \leq y, the sum in the second term of the right-hand side being taken over all indices nn for which an<xa_n < x. We shall show that if fd′(x)≥0f'_d(x) \geq 0 at every point of [a,b[[a, b[ distinct from the ana_n, then J=IJ = I.

It is clear that JJ is not empty, since a∈Ja \in J; moreover the definition of this set shows that if y∈Jy \in J one has x∈Jx \in J for a≤x≤ya \leq x \leq y, so JJ is an interval with left-hand endpoint aa (Gen. Top., IV, p. 336, prop. 1); let cc be its right-hand endpoint. One has c∈Jc \in J; this is clear if c=ac = a; if not, for every x<cx < c we have the inequality (1), and a fortiori

f(x)−f(a)≥−ε(c−a)−ε∑an<c12n f(x) - f(a) \geq -\varepsilon(c - a) - \varepsilon \sum_{a_n < c} \frac{1}{2^n}

from which it follows, on letting xx tend to cc in this inequality (since ff is continuous), that cc satisfies (1).

This being so, we shall see that we must have c=bc = b. Indeed, if one had c<bc < b, then certainly one would have c∉Ac \notin A; now fd′(c)f'_d(c) exists, and since fd′(c)≥0f'_d(c) \geq 0 by hypothesis, there exists a yy such that c<y≤bc < y \leq b and such that for c≤x≤yc \leq x \leq y one has

f(x)−f(c)≥−ε(x−c) f(x) - f(c) \geq -\varepsilon(x - c)

from which, taking account of (1), where xx is replaced by cc,

f(x)−f(a)≥−ε(x−a)−ε∑an<c12n≥−ε(x−a)−ε∑an<x12n f(x) - f(a) \geq -\varepsilon(x - a) - \varepsilon \sum_{a_n < c} \frac{1}{2^n} \geq -\varepsilon(x - a) - \varepsilon \sum_{a_n < x} \frac{1}{2^n}

which signifies that y∈Jy \in J, contradicting the definition of cc. Thus we have c=akc = a_k for some index kk; since ff is continuous at the point aka_k there is a yy such that c<y≤bc < y \leq b and such that for c<x≤yc < x \leq y one has

f(x)−f(c)≥−ε2k f(x) - f(c) \geq -\frac{\varepsilon}{2^k}
from which, taking account of (1), where xx is replaced by cc,
f(x)−f(a)≥−ε(c−a)−ε∑an<x12n≥−ε(x−a)−ε∑an<x12n f(x) - f(a) \geq -\varepsilon(c-a) - \varepsilon \sum_{a_n < x} \frac{1}{2^n} \geq -\varepsilon(x-a) - \varepsilon \sum_{a_n < x} \frac{1}{2^n}
which again leads to a contradiction; we thus have c=bc = b, and in consequence
f(b)−f(a)≥−ε(b−a)−ε∑an<b12n≥−ε(b−a)−ε.(2) f(b) - f(a) \geq -\varepsilon(b-a) - \varepsilon \sum_{a_n < b} \frac{1}{2^n} \geq -\varepsilon(b-a) - \varepsilon. \tag{2}
Since ε>0\varepsilon > 0 is arbitrary we deduce from (2) that f(b)≥f(a)f(b) \geq f(a), which demonstrates the first part of the proposition.

We remark now that this result applied to an interval [x,y][x, y] where a≤x<y≤ba \leq x < y \leq b proves that ff is increasing on I; if one had f(b)=f(a)f(b) = f(a) one could deduce that ff is constant on I, and then that fd′(x)=0f'_d(x) = 0 at every point of [a,b[[a, b[; the second part follows from this.

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

Cited by

Cites

Transcribed from the printed volume. Not translated.