Let f be a regulated function on an interval I, let g be a primitive of f on I, and [x0,x] a compact interval contained in I. For every ε>0 there exists a number ρ>0 such that for every subdivision of [x0,x] into intervals of length ≤ρ one has
g(x)−g(x0)−i=0∑n−1f(ti)(xi+1−xi)≤ε
(1)for every Riemann sum relative to this subdivision.
Indeed, let f1 be a step function such that ∥f(y)−f1(y)∥≤ε for every y∈[x0,x]; one has, denoting a primitive of f1 on I by g1,
It thus suffices to prove the proposition when f is a step function. Let (yk)1≤k≤m be the strictly increasing finite sequence of points of discontinuity of f in [x0,x]. For every subdivision of [x0,x] into intervals of length ≤ρ each of the points yk belongs to at most two of these intervals; there can therefore be no more than 2m intervals on which f is not constant; but, on such an interval [xi,xi+1] one has
∥g(xi+1)−g(xi)−f(ti)(xi+1−xi)∥≤2M(xi+1−xi)
on denoting by M the least upper bound of ∥f∥ on [x0,x]; on the other hand, when f is constant on [xi,xi+1] one has
g(xi+1)−g(xi)−f(ti)(xi+1−xi)=0.
One thus sees that the difference
g(x)−g(x0)−i=0∑n−1f(ti)(xi+1−xi)
cannot exceed 4Mmρ; it therefore suffices to take ρ≤ε/4Mm to obtain (1).
Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.