MA4255 Numerical Methods in Differential Equations

Chapter 1: Introduction/Preliminaries (Part I: ODEs)

1 / 14
What are Ordinary Differential Equations (ODEs)?
An ODE is an equation for a fct y = y(x) containing derivatives of y, e.g.,
y ′ (x) = 3y(x),
(y(x) + y ′ (x) + y ′′′′ (x))7 = y ′′ (x) − y ′′′′ (x) + x3 .

Explicit ODE of order n:

y (n) (x) = f (x, y(x), y ′ (x), . . . , y (n−1) (x)).

Implicit ODE of order n:

F (x, y(x), y ′ (x), . . . , y (n) (x)) = 0.

In Part I of this course, we focus on explicit first-order ODEs, i.e.,

y ′ (x) = f (x, y(x)).

2 / 14
Initial-value Problems (IVPs)
Ex.: y ′ (x) = 3y(x) has infinitely many solns: y(x) = c e3x , c ∈ R.

=⇒ To get a unique solution, we need to specify y at some value x0 ∈ R.

y ′ (x) = 3y(x), y(x0 ) = y0 =⇒ y(x) = y0 e3(x−x0 ) .

y(x0 ) = y0 is called an initial condition (i.c.). Problems of the form

y ′ (x) = f (x, y(x)), y(x0 ) = y0 (1)

are called Initial-value Problems (IVPs).

Note: Not all IVPs have a unique solution. E.g.,

2
y ′ (x) = [y(x)] 3 , y(0) = 0
2
1 3
has multiple solutions: y ≡ 0 and y(x) = 27 x . (Here, f (x, z) := z 3 )
=⇒ f (·, ·) continuous not sufficient for uniqueness. So what is sufficient?
Rk: f continuous =⇒ (1) has at least one solution (Peano Existence Thm)
3 / 14
Picard’s Theorem for IVPs y ′ (x) = f (x, y(x)), y(x0 ) = y0
Theorem (Picard’s Theorem)
Suppose f (·, ·) is continuous in a region U ⊆ R2 containing the rectangle
R := [x0 , XM ] × [y0 − YM , y0 + YM ], where XM > x0 , YM > 0, suppose

∃L > 0 : |f (x, z) − f (x, z̃)| ≤ L|z − z̃| ∀(x, z), (x, z̃) ∈ R, (2)

and with M := max(x,z)∈R |f (x, z)|, suppose M (XM − x0 ) ≤ YM . Then,

∃ a unique continuously differentiable fct y : [x0 , XM ] → R s.t.

y ′ (x) = f (x, y(x)) for x ∈ (x0 , XM ), y(x0 ) = y0 .

(2) is called Lipschitz condition (in 2nd argument of f ), L is called

Lipschitz constant.
Proof: See any introductory ODE course or book. Idea: Picard iteration:
Z x
y0 (x) ≡ y0 , yn (x) := y0 + f (t, yn−1 (t)) dt, n ∈ N.
x0
Then, yn → y in C([x0 , XM ]) and y solves the IVP. 4 / 14
Theorem (Picard’s Theorem)
Suppose f (·, ·) is continuous in a region U ⊆ R2 containing the rectangle
R := [x0 , XM ] × [y0 − YM , y0 + YM ], where XM > x0 , YM > 0, suppose

∃L > 0 : |f (x, z) − f (x, z̃)| ≤ L|z − z̃| ∀(x, z), (x, z̃) ∈ R,

and with M := max(x,z)∈R |f (x, z)|, suppose M (XM − x0 ) ≤ YM . Then,

∃ a unique continuously differentiable fct y : [x0 , XM ] → R s.t.

y ′ (x) = f (x, y(x)) for x ∈ (x0 , XM ), y(x0 ) = y0 .

Observation: the graph of y lies in R, i.e., (x, y(x)) ∈ R ∀x ∈ [x0 , XM ].

Pf of observation: Suppose this were not true. Then, by continuity of y,
∃x∗ ∈ (x0 , XM ) : |y(x∗ ) − y0 | = YM , |y(x) − y0 | < YM ∀x ∈ [x0 , x∗ ).
Z x∗ Z x∗
′
⇒ |y(x∗ ) − y0 | = y (x) dx ≤ |y ′ (x)| dx
x0 x0
Z x∗
= |f (x, y(x))| dx ≤ M (x∗ − x0 ) < YM . 
x0 5 / 14
IVPs for systems of ODEs
IVPs for systems of m ODEs:
 ′
 y1 (x) = f1 (x, y1 (x), y2 (x), . . . , ym (x)), y1 (x0 ) = y0,1 ,

 y2′ (x) = f2 (x, y1 (x), y2 (x), . . . , ym (x)), y2 (x0 ) = y0,2 ,


..


 .

 ′
ym (x) = fm (x, y1 (x), y2 (x), . . . , ym (x)), ym (x0 ) = y0,m ,
or equivalently,
y′ (x) = f(x, y(x)), y(x0 ) = y0
with given
   
y0,1 f1
x0 ∈ R, y0 =  ...  ∈ Rm , f =  ...  : [x0 , XM ] × Rm → Rm ,
   

y0,m fm

and we are seeking a soln y = (y1 , . . . , ym )T : [x0 , XM ] → Rm .

6 / 14
Picard’s Thm for systems y′ (x) = f(x, y(x)), y(x0 ) = y0
Introduce the Euclidean norm ∥ · ∥ : Rm → [0, ∞) on Rm by
v
um
uX
∥u∥ := t |ui |2 , for u = (u1 , . . . , um )T ∈ Rm .
i=1

Theorem (Picard’s Theorem (version for systems))

Suppose that f (·, ·) is a continuous in a region U ⊆ R1+m containing

R = {(x, z) ∈ R × Rm : x ∈ [x0 , XM ], ∥z − y0 ∥ ≤ YM },

where XM > x0 , YM > 0, suppose that

∃L > 0 : ∥f (x, z) − f (x, z̃)∥ ≤ L∥z − z̃∥ ∀(x, z), (x, z̃) ∈ R,

and with M := max(x,z)∈R ∥f (x, z)∥, suppose M (XM − x0 ) ≤ YM . Then,

∃ a unique continuously differentiable fct y : [x0 , XM ] → Rm s.t.

y′ (x) = f(x, y(x)), y(x0 ) = y0 .

7 / 14
A sufficient condition guaranteeing the Lipschitz property
Recall Lipschitz condition for IVPs:
∃L > 0 : |f (x, z) − f (x, z̃)| ≤ L|z − z̃| ∀(x, z), (x, z̃) ∈ R. (3)
This is automatically satisfied if f is cts on R, differentiable in int(R), and
∂f
∃C > 0 : |fz (x, z)| := (x, z) ≤ C ∀(x, z) ∈ int(R). (4)
∂z
Indeed, suppose that (4) holds. By the Mean-Value Thm, for any
(x, z) ∈ R we have f (x, z) − f (x, z̃) = fz (x, ξ)(z − z̃) for some ξ between
z and z̃. So, we obtain (3) with L := C.
Recall Lipschitz condition for IVPs for systems:
∃L > 0 : ∥f (x, z) − f (x, z̃)∥ ≤ L∥z − z̃∥ ∀(x, z), (x, z̃) ∈ R.
This is automatically satisfied with L = C if f cts on R, diff. in int(R), and
∂f
∃C > 0 : (x, z) ≤ C ∀(x, z) ∈ int(R), (5)
∂z
where ∥ · ∥ in (5) is the matrix norm induced by the Euclidean vector norm.
8 / 14
Warning: The converse is not necessarily true!

E.g., consider

f : R2 → R, f (x, z) := |z|

and R := [x0 , XM ] × [y0 − YM , y0 + YM ] with x0 := y0 := 0 and any

chosen XM , YM > 0. Then,

f satisfies the Lipschitz condition with L = 1:

|f (x, z) − f (x, z̃)| = ||z| − |z̃|| ≤ |z − z̃| ∀(x, z), (x, z̃) ∈ R,

but f is not differentiable in int(R) (as z 7→ |z| is not diff. at z = 0).

9 / 14
Stability
Definition (Stability)
Consider an IVP for a system of ODEs y′ (x) = f(x, y(x)), y(x0 ) = y0 .
(i) A solution y = v(x) is called stable on [x0 , XM ] if ∀ε > 0 ∃δ > 0
s.t. for all z ∈ Rm satisfying ∥y0 − z∥ < δ the solution w to

w′ (x) = f (x, w(x)), w(x0 ) = z

is defined in [x0 , XM ] and satisfies ∥v(x) − w(x)∥ < ε ∀x ∈ [x0 , XM ].

(ii) A soln y = v(x) which is stable on [x0 , ∞) (i.e. stable on [x0 , XM ]
for each XM and with δ independent of XM ) is called stable in the
sense of Lyapunov.
(iii) If in addition to (ii) there holds

lim ∥v(x) − w(x)∥ = 0,

x→∞

then the solution y = v(x) is called asymptotically stable.

10 / 14
Stability: Example
Recall defn: A solution y = v(x) is called stable on [x0 , XM ] if ∀ε > 0 ∃δ > 0 s.t. for all
z ∈ Rm satisfying ∥y0 − z∥ < δ the solution w to w′ (x) = f (x, w(x)), w(x0 ) = z is defined in
[x0 , XM ] and satisfies ∥v(x) − w(x)∥ < ε ∀x ∈ [x0 , XM ].
For some fixed λ ∈ R, consider the IVP
y ′ (x) = λy(x), y(0) = 1
with unique soln y = v(x) with v(x) := eλx .
Question: For what values of λ is the soln y = v(x) stable on [0, ∞) (i.e.,
stable on [0, XM ] for any XM > 0 and δ independent of XM )?
For z ∈ R, the problem
w′ (x) = λw(x), w(0) = z
has the unique soln w(x) := zeλx . Note that |v(x) − w(x)| = |1 − z|eλx .
Case λ ≤ 0: Then, |v(x) − w(x)| ≤ |1 − z| ∀x ∈ [0, ∞).
=⇒ y = v(x) stable on [0, ∞) when λ ≤ 0.
Case λ > 0: Note maxx∈[0,XM ] |v(x) − w(x)| = |1 − z|eλXM .
=⇒ y = v(x) is unstable on [0, ∞) when λ > 0.
However, y = v(x) is stable on [0, XM ] for fixed XM . 11 / 14
Main result on stability
Theorem (Stability under assumptions of Picard’s Thm)
Under assumptions of Picard’s Thm, the unique solution y = v(x) to
y′ (x) = f(x, y(x)), y(x0 ) = y0 is stable on [x0 , XM ].

Proof: Let w be the ′

R x soln to w (x) = f(x, w(x)), w(x R x0 ) = z. Then, we
have v(x) = y0 + x0 f (t, v(t)) dt and w(x) = z + x0 f (t, w(t)) dt.
Z x
∥v(x) − w(x)∥ ≤ ∥y0 − z∥ + (f (t, v(t)) − f (t, w(t))) dt
x0
Z x
≤ ∥y0 − z∥ + ∥f (t, v(t)) − f (t, w(t))∥ dt
x0
Z x
≤ ∥y0 − z∥ + L ∥v(t) − w(t)∥ dt.
x0
Rx
Gronwall Lemma: A(x) ≤ a + L x0 A(t) dt ∀x =⇒ A(x) ≤ aeL(x−x0 ) ∀x.
∥v(x) − w(x)∥ ≤ eL(x−x0 ) ∥y0 − z∥ ≤ eL(XM −x0 ) ∥y0 − z∥.
Given ε > 0, set δ = e−L(XM −x0 ) ε: ∥v(x) − w(x)∥ < ε if ∥y0 − z∥ < δ. 12 / 14
Proof of Gronwall Lemma: Write I := [x0 , XM ]. We need to show that
Z x
A(x) ≤ a + L A(t) dt ∀x ∈ I =⇒ A(x) ≤ aeL(x−x0 ) ∀x ∈ I.
x0
Rx
Multiplying the inequality −a − L x0 A(t) + A(x) ≤ 0 by e−Lx :
 Z x 
d a −Lx −Lx
e +e A(t) dt ≤ 0 ∀x ∈ I.
dx L x0

So, the function in the brackets is non-increasing. Thus,

Z x
a −Lx −Lx a
e +e A(t) dt ≤ e−Lx0 ∀x ∈ I.
L x0 L

Multiply by LeLx :
Z x
a+L A(t) dt ≤ a eL(x−x0 ) ∀x ∈ I.
x0
Rx
It follows that A(x) ≤ a + L x0 A(t) dt ≤ aeL(x−x0 ) ∀x ∈ I.
13 / 14
End of “Chapter 1: Introduction/Preliminaries”.

14 / 14