H.

Ammari Summer Term 2021

L. Baldassari ETH Zürich
A. Scapin Numerical Analysis II D-MATH

Problem Sheet 3

Problem 3.1 Lipschitz condition

Check whether the following functions satisfy the Lipschitz condition on the respective intervals.
If so, find a suitable Lipschitz constant.

(3.1a)
f (t, x) = 2tx−4 , (t, x) ∈ [0, ∞) × [1, ∞);

Solution: It does not satisfy the Lipschitz condition. Note that

|f (t, x1 ) − f (t, x2 )| x4 − x4 1 (x1 + x2 )(x21 + x22 )
= 2t 1 4 4 2 = 2t .
|x1 − x2 | x1 x2 |x1 − x2 | x41 x42
For any fixed x1 , x2 ∈ [1, ∞), this is not bounded when t → ∞.

(3.1b)
2
f (t, x) = e−x tan−1 t, (t, x) ∈ [0, ∞) × [1, ∞);

Solution: It satisfies the Lipschitz condition. By the mean value theorem and the fact that
2 2
(e−x )0 = −2xe−x , there exists some x∗ ∈ (x1 , x2 ) such that
2 2
|f (t, x1 ) − f (t, x2 )| e−x1 − e−x2
= tan−1 t
|x1 − x2 | |x1 − x2 |
2
= tan−1 t 2x∗ e−x∗ .
2
Then, since 2xe−x < 1 for x ∈ [1, ∞) and tan−1 t < π/2 for t ∈ [0, ∞), we obtain
|f (t, x1 ) − f (t, x2 )| π
< .
|x1 − x2 | 2
Thus, f is Lipschitz with constant π/2.

(3.1c)
f (t, x) = 2t(1 + t2 )−1 (1 + e−x ), (t, x) ∈ [0, ∞) × [0, ∞).

Solution:
It satisfies the Lipschitz condition. By the mean value theorem and the fact that (e−x )0 = −e−x ,
there exists some x∗ ∈ (x1 , x2 ) such that
|f (t, x1 ) − f (t, x2 )| 2t |e−x1 − e−x2 | 2t −x∗
= 2
≤ e .
|x1 − x2 | 1 + t |x1 − x2 | 1 + t2

Problem Sheet 3 Page 1 Problem 3.1

Then, since 2t/(1 + t2 ) ≤ 1 for t ∈ [0, ∞) (it has a local maximum at t = 1) and e−x ≤ 1 for
x ∈ [0, ∞), we obtain

|f (t, x1 ) − f (t, x2 )|
≤ 1.
|x1 − x2 |

Thus, f is Lipschitz with constant 1.

Problem 3.2 Infinitely many solutions to an ODE

Suppose that m is a fixed positive integer. Show that
dx 2m
= x 2m+1 , x(0) = 0,
dt
has infinitely many continuously differentiable solutions x = x(t). Why does this not contradict
Cauchy-Lipshitz theorem (there exists a unique solution to dx/dt = f (x, t) when f is Lipschitz)?
Solution: For any c ∈ [0, ∞), let us define
(
(2m + 1)−2m−1 (t − c)2m+1 , t ≥ c,
xc (t) =
0, 0 ≤ t < c.

It is clear that xc is a continuously differentiable function on [0, ∞) and that xc (0) = 0. For t ≤ c,
it is obvious that xc is a solution. For t > c, we have
2m
dxc h i 2m+1 2m
= (2m + 1)−2m (t − c)2m = (2m + 1)−(2m+1) (t − c)2m+1 = x(t) 2m+1 . (3.2.1)
dt
Since c can be any positive constant, there are infinitely many continuously differentiable solu-
tions.

Problem 3.3 Lipschitz condition and Gronwall lemma

Let f (t, x) be uniformly Lipschitz in x, i.e. there exists C0 > 0 such that

|f (t, x) − f (t, y)| ≤ C0 |x − y| (3.3.1)

for all t ≥ 0, x, y ∈ Rn .
The flow φt (x0 ) : R+ × Rn → Rn is defined by φt (x0 ) = x(t) where x(t) satisfies

 dx = f (t, x(t)), t ≥ 0
dt
x(t = 0) = x0


(3.3a) Prove that φt (x0 ) is well-defined (that it exists and is unique).

Solution:
Since f is uniformly Lipschitz, the Cauchy-Lipschitz theorem says that there exists a unique
solution x(t) on t ∈ R+ for each x0 . Thus, φt (x0 ) is well-defined.

Problem Sheet 3 Page 2 Problem 3.2

(3.3b) (Gronwall’s lemma)
Let a, b ≥ 0 and u : R+ → R be a continuous function such that
Z t
u(t) ≤ a + b u(s)ds, (3.3.2)
0

for t ≥ 0. Prove that u(t) ≤ a exp(bt).

Rt
Solution: Let ψ(t) := a + b 0 u(s)ds. Notice that ψ ∈ C 1 , so for t ∈ R+

dψ
(t) = bu(t).
dt
Then, by (3.3.2) we have that
dψ
≤ bψ(t).
dt
Let f (t) := exp(−bt)ψ(t). Then

df dψ
= −b exp(−bt)ψ(t) + exp(−bt)
dt dt
dψ
= exp(−bt)(−bψ(t) + )
dt
≤0

Notice that f (0) = ψ(0) = a, which means that f (t) ≤ a for t ∈ R+ . So ψ(t) ≤ a exp(bt) for
t ≥ 0, and u(t) ≤ ψ(t) ≤ a exp(bt).

(3.3c) Deduce that φt (x0 ) is Lipschitz with a Lipschitz constant given by exp(C0 t).
Solution: Notice that Z t
φt (x0 ) = x0 + f (s, φs (x0 ))ds,
0
so
Z t
|φt (x0 ) − φt (y0 )| ≤ |x0 − y0 | + |f (s, φs (x0 )) − f (s, φs (y0 ))|ds
0
Z t
≤ |x0 − y0 | + C0 |φs (x0 ) − φs (y0 )|ds
0

Let a = |x0 − y0 |, b = C0 , then by Gronwall’s lemma,

|φt (x0 ) − φt (y0 )| ≤ exp(C0 t)|x0 − y0 |,

which proves our conclusion.

Problem Sheet 3 Page 3 Problem 3.3

Problem 3.4 Transforming the Thomas-Fermi Differential Equation
Consider the Thomas-Fermi differential equation, given by
3
y 2 (t)
ÿ(t) = 1 , y(0) = y0 6= 0, ẏ(0) = z0 . (3.4.1)
t2
This form does not satisfy the local Lipschitz condition and is furthermore not suitable as an entry
for numerical integration.
Transform (3.4.1) into a system of two first-order ODEs that satisfies the Lipschitz condition.
H INT: Use the substitution
1 w0 (s)
s = t2 , y(t) = w(s), u(s) = .
s

Solution: Notice that the original system is not Lipschitz-continuous at t = 0.

1 w0 (s)
Let s = t 2 , y(t) = w(s), u(s) = s
. Using the chain rule, it follows that
d d d 1 1
w(s(t)) = (w(s(t))) · s(t) = w0 (s(t)) · = u(s(t)),
dt ds dt 2s 2
and further
d2 1d 1 d d u0 (s(t))
(w(s(t))) = u(s(t)) = (u(s(t))) · s(t) = .
dt2 2 dt 2 ds dt 4s
Thus,
d2 3
u0 (s) = 4s 2
(w(s(t))) = 4sÿ(t) = 4w 2 (s),
dt
so transformed system is
(
w0 (s) = su(s), w(0) = y0 ,
3
u0 (s) = 4w 2 (s), u(0) = 2z0 ,
where the initial value u(0) follows from the fact that
dy d u(s(t))
= w(s(t)) = .
dt dt 2
The right hand side is locally Lipschitz-continuous, as
 
su(s)
f (s, u(s), w(s)) = 3
4w 2 (s)
and  
s 0
Du,w f (s, u(s), w(s)) = 1
0 6w 2 (s)
are continuous on the closed set Ω = (0, s) × (y0 , u(s)) × (2z0 , w(s)).

Published on 10 March 2021.

To be submitted by 17 March 2021.
Last modified on March 11, 2021

Problem Sheet 3 Page 4 Problem 3.4