Let f be a locally Lipschitz function on I×H and (a,b) an arbitrary point of I×H.
1 There exist an interval K⊂I, a neighbourhood of a in I, and a neighbourhood V of b in H such that, for every point (t0,x0) of K×V, there exists an integral (and only one) u(t,t0,x0) defined on K, with values in H and equal to x0 at the point t0 (in other words, J(t0,x0)⊃K for all (t0,x0)∈K×V).
2 The map (t,t0,x0)↦u(t,t0,x0) of K×K×V into H is uniformly continuous.
3 There exists a neighbourhood W⊂V of b in H such that, for every point
(t,t0,x0)∈K×K×W, the equation x0=u(t0,t,x) has a unique solution x on V equal to u(t,t0,x0) ("resolution of the integral with respect to the constant of integration").
1 Let S be a ball with centre b and radius r contained in H, and J0 an interval contained in I, a neighbourhood of a in I, such that f is bounded and Lipschitz (with respect to some constant k) on J0×S; denote by M the supremum of ∥f(t,x)∥ on J0×S. Then there exist (IV, p. 171, th. 1) an interval J⊂J0, a neighbourhood of a in I, and an integral v of x′=f(t,x) defined on J, with values in S and equal to b at a. We shall see that the open ball V with centre b and radius r/2, and the intersection K of J with an interval ]a−l,a+l[, where l is small enough, are as required. Indeed, prop. 7 of IV, p. 174 (applied to the set J0×S and the case where α=0) shows that there exists an integral of x′=f(t,x) defined on K, with values in S, and equal to x0 at a point t0∈K, provided that
∥v(t)−b∥+∥v(t0)−x0∥ek∣t−t0∣<r for every
t∈K. Now, by the mean value theorem, one has
∥v(t)−b∥≤M∣t−a∣≤Ml for every
t∈K; since
∥x0−b∥<r/2 one sees that it suffices to take
l such that
Ml+(Ml+r/2)e2kl<r or the relation (18) to be satisfied for every
(t,t0,x0) of
K×K×V.
2) By the mean value theorem we have
∥u(t1,t0,x0)−u(t2,t0,x0)∥≤M∣t2−t1∣
(20)
for all t0,t1,t2 in K and x0 in V. Now prop. 5 (IV, p. 169) shows that
∥u(t,t0,x1)−u(t,t0,x2)∥≤e2kt ∥x2−x1∥
(21)
for every t and t0 in K, and x1 and x2 in V. Finally, if t1 and t2 are any two points in K,
∥u(t1,t2,x0)−u(t2,t2,x0)∥≤M∣t2−t1∣
by the mean value theorem, that is to say
∥u(t1,t2,x0)−x0∥≤M∣t2−t1∣;
since u(t,t2,x0) is identical to the integral which takes the value u(t1,t2,x0) at the point t1, prop. 5 (IV, p. 169) shows that
∥u(t,t1,x0)−u(t,t2,x0)∥≤Me2kl∣t2−t1∣
(22)
for all t,t1,t2 in K and x0 in V. The three inequalities (20), (21) and (22) thus prove the uniform continuity of the map (t,t0,x0)↦u(t,t0,x0) on K×K×V.
3 By (20), we have ∥u(t,t0,x0)−x0∥≤M∣t−t0∣≤2Ml on
K×K×V.
If l is taken small enough, so that 2Ml<r/4, one then sees that if x0 is any point of the open ball W with centre b and radius r/4, that u(t,t0,x0)∈V for any t and t0 in K. If x=u(t,t0,x0), the function s↦u(s,t,x) is then defined on K and is equal to the integral of (1) which takes the value x at the point t, that is, to u(s,t0,x0); in particular
x0=u(t0,t0,x0)=u(t0,t,x).
Moreover, if y∈V is such that x0=u(t0,t,y), then the integral s↦u(s,t,y) takes the value x0 at t0 so is identical to s↦u(s,t0,x0), which consequently takes the value x at t, which shows that y=x and concludes the proof.
Bourbaki puts the proof directly after the statement, so what is above is the statement and its proof together.
Nothing.