MA-417-ODE

Lecture 5

Debanjana Mitra

Department of Mathematics
Indian Institute of Technology Bombay
Powai, Mumbai - 76

August 6, 2021
Existence and uniqueness theorem of solutions of ODEs
Uniqueness of solutions of ODEs
Existence theorem under less restricted assumption

Continuation of solutions of an ODE

Existence and uniqueness theorem
Let us consider the ODE:
dy (t)
= f (t, y (t)), y (t0 ) = y0 , (1)
dt
where f : D → R, D is a subset of R2 .

Qn.
I Does there exist a solution of the ODE (IVP)?
I Is the solution unique?
I What is the largest interval of its existence?
Recall. Let I be any interval in R such that t0 ∈ I and (t, y (t)) ∈ D, for
all t ∈ I : Z t
y (t) = y0 + f (s, y (s)) ds, t ∈ I . (2)
t0

Let f ∈ C (D). Then y ∈ C (I ) satisfies (2) if and only if y is a solution of

(1).
Recall

Successive Iterative method: (Picard’s iteration) Let (t0 , y0 ) ∈ R2 and

a closed rectangle R in R2 containing (t0 , y0 ) defined by

R = {(t, y ) ∈ R2 | |t − t0 | ≤ a, |y − y0 | ≤ b},

for some positive constants a, b.

We define a successive iteration (called by Picard’s iteration) by:

I = [t0 − α, t0 + α]
Z t
∀ n ∈ N, yn+1 (t) = y0 + f (s, yn (s)) ds, ∀s ∈ I, y (t0 ) = y0 . (3)
t0

Qn. Can we have a function y ∈ C (I ) such that yn → y converges

uniformly?
Qn. If so, is that y a solution of ODE?
Definition (A Lipschitz condition)
Let D be a subset of R2 and f be defined on D. Then f is said to satisfy
Lipschitz condition with respect to y in D if there exists a positive
constant K (independent of t and y ) such that

|f (t, y1 ) − f (t, y2 )| ≤ K |y1 − y2 |, ∀ (t, y1 ), (t, y2 ) ∈ D.

Notation: f ∈ Lip in D. In addition, if it is know that f ∈ C (D), then

together, the notation ‘f ∈ (C , Lip) in D’ will be used.
Successive iteration
Lemma (Picard’s iteration)
Let f ∈ (C , Lip) on R, a closed rectangle defined by

R = {(t, y ) ∈ R2 | |t − t0 | ≤ a, |y − y0 | ≤ b},

for some positive constants a and b. Let

M = max |f (t, y )|, α = min{a, b/M},

(t,y )∈R

and set I = [t0 − α, t0 + α]. Let K > 0 be such that

|f (t, z1 ) − f (t, z2 )| ≤ K |z1 − z2 |, ∀ (t, z1 ), (t, z2 ) ∈ R.

Then, the followings hold:

1. for all n ∈ N, (t, yn (t)) ∈ R, for all t ∈ I and yn ∈ C 1 (I ).
2. for all n ∈ N, |yn (t) − y0 | ≤ M|t − t0 |, ∀t ∈ I.
n+1
M K (t−t0 )n+1
3. For any n ∈ N ∪ {0}, |yn+1 (t) − yn (t)| ≤ K (n+1)! , for all t ∈ I .
Proof. We prove the above lemma for t ∈ [t0 , t0 + α]. The same
argument also holds for t ∈ [t0 − α, t0 ] and the lemma can be proved.
The proof is using an Induction argument
Step 1. Note that y1 ∈ C 1 ([t0 − α, t0 ]) and from the definition of y1 , it
follows |y1 (t) − y0 | ≤ M(t − t0 ) ≤ b, for all t ∈ [t0 − α, t0 ].
Let us assume that for a n ∈ N, yn ∈ C 1 ([t0 − α, t0 ]) and
|yn (t) − y0 | ≤ M(t − t0 ) ≤ b, for all t ∈ [t0 , t0 + α], i.e., (t, yn (t)) ∈ R for
all t ∈ [t0 , t0 + α]. Thus f (s, yn (s)) is well-defined for all s ∈ [t0 , t0 + α]
and s 7→ f (s, yn (s)) is a continuous map for all s ∈ [t0 , t0 + α].
We want to show the above assertions also hold for n + 1.
From the definition of yn+1 , it follows that yn+1 ∈ C 1 ([t0 , t0 + α])
because it is given that f ∈ C (R) and yn ∈ C ([t0 − α, t0 ]). Now
estimating, it can be derived
Z t
|yn+1 (t) − y0 | ≤ |f (s, yn (s))| ds ≤ M(t − t0 ) ≤ b, ∀ t ∈ [t0 , t0 + α].
t0

Note that for all n ∈ N, f (t, yn (t)) is defined on t ∈ [t0 − α, t0 ] because

of the choice of α = min{a, b/M}.
Step 2. Set for all n ∈ N ∪ {0}, ∆n (t) = |yn+1 (t) − yn (t)| for all
t ∈ [t0 , t0 + α]. Note that, using the boundedness of f and Lipschitz
property of f with respect to y , we get that for all n ∈ N ∪ {0},
Z t
∆n+1 (t) ≤ K ∆n (s) ds, ∀ t ∈ [t0 , t0 + α].
t0

Now observe

∆0 (t) = |y1 (t) − y0 | ≤ M(t − t0 ), ∀ t ∈ [t0 , t0 + α].

Let for a (n − 1) ∈ N, the following estimate holds:

M K n (t − t0 )n
∆n−1 (t) ≤ , ∀ t ∈ [t0 , t0 + α].
K n!
Now we show the estimate for ∆n . Estimating ∆n (t) using its definition,
we get
M K n+1 (t − t0 )n+1
∆n (t) ≤ , ∀ t ∈ [t0 , t0 + α].
K (n + 1)!
Existence and uniqueness theorem

Theorem (Picard-Lindelöf Theorem)

Let f ∈ (C , Lip) on R, a closed rectangle defined by

R = {(t, y ) ∈ R2 | |t − t0 | ≤ a, |y − y0 | ≤ b},

for some positive constants a and b. Let

M = max |f (t, y )|, α = min{a, b/M},

(t,y )∈R

and set I = [t0 − α, t0 + α]. Then there exists a unique y ∈ C 1 (I )

satisfying (1):

y 0 (t) = f (t, y (t)), ∀t ∈ I, y (t0 ) = y0 .

Proof of the theorem.
Step 1. Because of Lemma 2, for all n ∈ N, (t, yn (t)) ∈ R, and hence
f (t, yn (t)) is well-defined. Now, Using 3. in Lemma 2, we get that
M K n+1 αn+1
|yn+1 (t) − yn (t)| ≤ , t ∈ I.
K (n + 1)!
P∞ n+1 αn+1
Since 1 + n=0 K(n+1)! = e αK , using Weirstrass-M test for series, we
P∞
get that n=0 ∆n (·) is uniformly and absolutely convergent on I , where

∆n (t) = |yn+1 (t) − yn (t)|, ∀t ∈ I, ∀ n ∈ N ∪ {0}.

Step 2. For any n ∈ N, rewriting

n−1
X
yn (t) = y0 + [y`+1 (t) − y` (t)], ∀t ∈ I,
`=0

since the RHS is uniformly convergent on I as n → ∞, we obtain that

there exists a y ∈ C (I ) (because ∆n (·) ∈ C (I ) for all n. See 1 in Lemma
2) such that

yn → y uniformly on I as n → ∞.
Step 3.
So far we have y ∈ C (I ). First show that y satisfies (2) and from the
expression, it can be derived that y ∈ C 1 (I ). First we need to show:
|y (t) − y0 | ≤ b for all s ∈ I and thus f (t, y (t)) is defined for all t ∈ I .
Estimating for all t ∈ I ,

|y (t) − y0 | ≤ |y (t) − yn (t)| + |yn (t) − y0 | ≤ |y (t) − yn (t)| + b,

taking n → ∞ in both side, we get |y (t) − y0 | ≤ b, for all t ∈ I .

Rt Rt
Next, we need to show: t0
f (s, yn (s)) − t0
f (s, y (s)) → 0 uniformly
for all t ∈ I as n → ∞. To do it, estimate (enough to do for
t ∈ [t0 , t0 + α], similarly, the estimate will follow for t ∈ [t0 − α, t0 ])
Z t Z t Z t
f (s, yn (s))− f (s, y (s)) ≤ K |yn (s)−y (s)| ds, ∀ t ∈ [t0 , t0 +α],
t0 t0 t0

and it is uniformly convergent on [t0 , t0 + α] due to the uniform

convergence of yn to y .
Now taking limit in both side of
Z t
yn (t) = y0 + f (s, yn (s)) ds,
t0

we get y satisfies Z t
y (t) = y0 + f (s, y (s)) ds.
t0

Since y ∈ C (I ) and f ∈ C (R), then from the expression of y , it follows

that y ∈ C 1 (I ). Thus, from the expression of y , noting y (t0 ) = y0 , and
diferentiating both side with respect to t, it follows that

y 0 (t) = f (t, y (t)), y (t0 ) = y0 .

Error estimate between solution y and nth approximation yn :

∞
X M (K α)n+1 K α
|y (t)−yn (t)| ≤ |y`+1 (t)−y` (t)| ≤ e , ∀ t ∈ [t0 −α, t0 +α].
K (n + 1)!
`=n
Uniqueness of solutions

Lemma (Gronwall’s inequality)

Let C be a non-negative constant and h and g be two continuous
non-negative functions defined on a ≤ t ≤ b satisfying
Z t
h(t) ≤ C + h(s)g (s) ds, ∀ a ≤ t ≤ b,
a

then it follows Rt
g (s) ds
h(t) ≤ Ce a , ∀ a ≤ t ≤ b.

Proof. The proof can be derived in following steps.

Rt
Step 1. Set U(t) = C + a h(s)g (s) ds, for all a ≤ t ≤ b. Then we have

U 0 (t) = h(t)g (t), ∀ a ≤ t ≤ b, U(a) = C .

Proof contd...

Step 2. By assumption, we have h(t) ≤ U(t) for all a ≤ t ≤ b and

g (t) ≥ 0 for all a ≤ t ≤ b. From Step 1 along with the assumptions, we
get
U 0 (t) ≤ U(t)g (t), a ≤ t ≤ b.
It gives that
d − R t g (s) ds
e a U(t) ≤ 0, a ≤ t ≤ b.
dt
Step 3. Integrating over [a, t] for all a < t ≤ b, from the above
inequality, we get
Rt Rt
g (s) ds g (s) ds
U(t) ≤ U(a)e a = Ce a , a ≤ t ≤ b,

since U(a) = C in Step 1. By assumption, we have h(t) ≤ U(t), for all

a ≤ t ≤ b. Thus, from the above estimate, it follows
Rt
g (s) ds
h(t) ≤ Ce a , a ≤ t ≤ b.
Uniqueness of solution in Picard-Lindelöf Theorem
Let us assume all hypothesis in Picard-Lindelöf Theorem holds.
Let φ1 and φ2 be two solutions of IVP-ODE:

y 0 (t) = f (t, y (t)), y (t0 ) = y0 .

The proof will be given for t ∈ [t0 , t0 + α]. Similar argument and hence
the result hold for t ∈ [t0 − α, t0 ].
Denoting h(t) = |φ1 (t) − φ2 (t)| for all t ∈ [t0 , t0 + α], note that
h(t) ≥ 0 for all t ∈ [t0 , t0 + α] and
Z t
h(t) ≤ K h(s) ds, ∀ t ∈ [t0 , t0 + α],
t0

where K is the Lip constant.

Now using Gronwall’s inequality, we get h(t) ≤ 0 for all t ∈ [t0 , t0 + α].
This along with the fact that h is a non-negative function, implies

h(t) = 0, ∀ t ∈ [t0 , t0 + α].

Remark.
I Advantage of the method: Constructive.

I Further sufficient conditions (more restricted than Lip:) f ∈ C 1 (R)

∂f
or ∂y exists and bdd over R.
I Globally Lip in y i.e., f is defined on [t0 − a, t0 + a] × (−∞, ∞) and
|f (t, y1 ) − f (t, y2 )| ≤ k|y1 − y2 |, for all t ∈ [t0 − a, t0 + a] and
y1 , y2 ∈ R. Let f be continuous and bounded on
[t0 − a, t0 + a] × (−∞, ∞). The ODE admits a unique solution in
I = [t0 − a, t0 + a]. (global existence).
Ex. Check if assumptions in Picard-Lindelöf theorem hold for a linear
ODE: where a(·) and b(·) are continuous and bounded on R,

y 0 (t) + a(t)y (t) = b(t), y (t0 ) = y0 .

Justify the existence (global) and uniqueness of solution of the ODE

using Picard-Lindelöf theorem.

Ex. Construct solutions using Picard’s iteration:

I y 0 (t) = ay (t) + b, y (0) = 1.
I y 0 (t) = 5ty (t), y (0) = 1.
I y 0 (t) = y 2 (t), y (0) = 1.
Ex: Picard’s iteration fails in general if the Lipschitz condition (uniform
in t) on f is dropped: y 0 (t) = f (t, y (t)), y (0) = 0 in
D = (−∞, 1] × R, where


 0 t ≤ 0, y ∈ R,
2t 0 < t ≤ 1, y < 0,

f (t, y ) = 4y

 2t − t 0 < t ≤ 1, 0 ≤ y ≤ t 2 ,
−2t 0 < t ≤ 1, t 2 < y < ∞.


Hint. Write the Picard’s iteration. Get y2k−1 (t) = t 2 and y2k (t) = −t 2 .
The sequence does not converge.

Qn. Can we have the existence of solution of an ODE without the

Lipschitz condition on f ?