There exists one and only one convex function g defined on ]0, +∞[ that satisfies the equation (3) and takes the value 0 for x=1.
First we show that if there is a function g satisfying the conditions stated then it is well-determined on the interval ]0, 1], and consequently on the interval ]0, +∞[. Indeed, for every integer n>1 the gradient of the line joining the point (n,g(n)) to the point (x,g(x)) is an increasing function of x, since g is convex (I, p. 27, prop. 5); one thus must have, for 0<x≤1,
(n−1)−ng(n−1)−g(n)≤(n+x)−ng(n+x)−g(n)≤(n+1)−ng(n+1)−g(n)
that is, by (3),
xlog(n−1)≤g(x+n)−g(n)≤xlogn.
Now, by (3),
g(x+n)−g(n)=g(x)+logx+k=1∑n−1(log(x+k)−logk).
Moreover, one can write logn=∑k=2nlogk−1k so the inequality (4) can be written
xk=2∑n−1logk−1k≤g(x)+logx+k=2∑n(log(x+k−1)−log(k−1))
≤xk=2∑nlogk−1k.
Let us put, for every n≥2,
un(x)=xlogn−1n−log(x+n−1)+log(n−1)
(5)
and
gn(x)=−logx+k=2∑nuk(x).
For 0<x≤1 one then has
gn(x)−xlogn−1n≤g(x)≤gn(x). (6)
Since logn−1n tends to 0 as n tends to +∞ one deduces from (6) that if a solution g exists then it is necessarily equal, on ]0,1], to the limit of the gn(x).
Now one deduces immediately from (5) that for x fixed and >0 one has
un(x)=−xlog(1−n1)−log(1+nx−1)+log(1−n1)∼2n2x(x−1) as
n tends to
+∞, which proves that the series with general term
un(x) converges for every
x>0. Each of the functions
un(x) being convex on
]0,+∞[, as is
−logx, the function
g(x)=−logx+∑n=2∞un(x) is convex on this interval (I, p. 27, prop. 2 and prop. 4); finally, one has
un(1)=0, whence
g(1)=0, and
un(x+1)=un+1(x)+x(logn−1n−lognn+1) whence
g(x+1)=−log(x+1)+xlog2+n=3∑∞un(x)=logx+g(x); in other words,
g satisfies equation (3) of VII, p. 306.
Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.
Nothing.