Picards Theorem

Picard’s Theorem
John Cook
January 26, 1993
Picard’s Theorem
Theorem 1. Let X be a Banach space and let u0 ∈ X be given. Let B be the

ball of radius r centered at uo . Suppose f : [0, T ] × X → X is continuous and
satisfies the Lipschitz condition
kf (t, u) − f (t, v)k ≤ Lku − vk
for all u, v ∈ B and t ∈ [0, T ]. Suppose also that kf k ≤ M on [0, T ] × B. Let

r
c = min(T, M ). Then the equation
u0 (t) + f (t, u(t)) = 0
with initial condition u(0) = u0 is well-posed in C 1 (0, c; X).

If X is a finite dimensional space, the Lipschitz condition is not necessary
for existence, but it is necessary for uniqueness.
Proof. (Existence) The given differential equation is equivalent to the integral
equation
Z t
u(t) = u0 + f (s, u(s)) ds. (1)
0
This suggests the iteration procedure (Picard iteration) given by u0 = u0 and

Z t
n+1
u = u0 + f (s, un (s)) ds. (2)
0
We show that the un converge uniformly to a solution u.

First of all,
Z t
ku1 (t) − u0 (t)k = f (s, u0 (s)) ds ≤ M t.
0
1
r
This is where we need the hypothesis t ≤ M , i.e. this condition is necessary to
keep our iterates inside B. Next we notice
Z t
ku2 (t) − u1 (t)k ≤ kf (s, u1 (s)) − f (s, u0 (s))k ds
0
Z t
≤ Lku1 (s) − u0 (s)k ds
0
Z t
t2
≤ LM s ds = LM .
0 2
In general,
M (Lt)n+1
kun+1 (t) − un (t)k ≤ .
L(n + 1)!
Since
m−1
X
um (t) − un (t) = (ui+1 (t) − ui (t)),
i=1
it follows that
m−1
X (Lc)i+1
kum (t) − un (t)k .
i=1
(i + 1)!
n
Thus, the sequence {u } is uniformly Cauchy by comparison to the power series
for eLc .
By the dominated convergence theorem, it follows from equation 2 that the
limit u is a solution to equation 1. Here we only needed f (t, x) to be continuous
with respect to x and measurable with respect to t.
The following simple version of Gronwall’s inequality is necessary to show
uniqueness and continuous dependence on initial conditions.
Lemma 1. If w ∈ L1 [0, T ] and
Z t
w(t) ≤ C + L w(s) ds
0
then
w(t) ≤ CeLt .
Rt
Proof. Let v(t) = C + L 0
w(s) ds. Since v ≥ w by hypothesis, we see that
e−Lt v is non-increasing:
d −Lt
e v(t) = e−Lt (Lw − Lv) ≤ 0.
dt
So e−Lt v(t) ≤ v(0) = C and w(t) ≤ v(t) ≤ CeLt .
The following establishes a growth estimate on solutions which also proves
uniqueness and continuous dependence on initial conditions.
2
Proof. (Uniqueness) Suppose
Z t
u(t) = u0 + f (s, u(s)) ds
0
and Z t
v(t) = v0 + f (s, u(s)) ds.
0
Let w(t) = ku(t) − v(t)k. The Lipschitz condition on f insures that
Z t
w(t) ≤ w(0) + L w(s) ds.
0
The complex and real analytic analogs of Picard’s theorem are also true:
if f is complex (real) analytic, the solutions are complex (real) analytic. The
basic idea of the proof is to use the real version of Picard’s theorem on the
real and imaginary parts. The integral operator in the existence proof preserves
analyticity by Morera’s theorem. See Garrett Birkhoff and Gian-Carlo Rota’s
text Ordinary Differential Equations for details.
In the case of X being a Hilbert space, the following theorem completes the
proof that the problem
u0 (t) + f (t, u(t)) = 0, u(0) = u0
is well-posed. Notice that this theorem independently establishes uniqueness

and continuous dependence on initial conditions.
Theorem 2. Let X be a Hilbert space. Let g : [0, T ] × B → X be a perturbation
of f in the sense that
kf (t, w) − g(t, w)k ≤ ε
for every t ∈ [0, T ] and w ∈ B. If v is a solution to
v 0 (t) + g(t, v(t)) = 0
then for all t ∈ [0, t],

ε Lt
ku(t) − v(t)k ≤ (e − 1) + eLt ku(0) − v(0)k.
L
Proof. Define σ(t) = ku(t) − v(t)k2 . Then
1 0
σ (t) = hu(t) − v(t), u0 (t) − v 0 (t)i
2
= hu(t) − v(t), f (t, u(t)) − g(t, v(t))i
≤ hu(t) − v(t), f (t, u(t)) − f (t, v(t))i + hu(t) − v(t), f (t, v(t)) − g(t, v(t))i
≤ Lku(t) − v(t)k2 + εku(t) − v(t)k
√
= Lσ + ε σ.
3
A Gronwall-type inequality shows that
p ε p
σ(t) ≤ eLt σ(t) + eLt σ(0).
L
Notice that since this discussion has not involved any notion of accretiveness
or monotonicity, the results hold on [−T, T ].
Picard’s theorem has a number of important special cases; in fact, most of
what is known about the general theory of ODE’s is a special case or a variation
of this theorem. For example, if X = IRn and f (t, u(t)) = A(t)u(t) + g(t) where
A is a matrix whose entries are continuous functions and g is a continuous
function, then f is bounded on [0, T ] × B for any bounded B. Thus any linear
initial value problem is well-posed and solutions exist for all time. If we assume
g and the components of A are L∞ functions then the same result holds, except
that the differential equation is satisfied only for almost every t ∈ [0, T ].

Picards Theorem

Uploaded by

Copyright:

Available Formats

Picards Theorem

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Picards Theorem

Uploaded by

Copyright:

Available Formats

Picard’s Theorem

Theorem 1. Let X be a Banach space and let u0 ∈ X be given. Let B be the

kf (t, u) − f (t, v)k ≤ Lku − vk

for all u, v ∈ B and t ∈ [0, T ]. Suppose also that kf k ≤ M on [0, T ] × B. Let

u0 (t) + f (t, u(t)) = 0

with initial condition u(0) = u0 is well-posed in C 1 (0, c; X).

This suggests the iteration procedure (Picard iteration) given by u0 = u0 and

We show that the un converge uniformly to a solution u.

u0 (t) + f (t, u(t)) = 0, u(0) = u0

is well-posed. Notice that this theorem independently establishes uniqueness

v 0 (t) + g(t, v(t)) = 0

then for all t ∈ [0, t],

You might also like