1

Lecture notes on Ordinary Dierential Equations

Annual Foundation School, IIT Kanpur,
Dec.3-28, 2007.
by
S. Sivaji Ganesh
Dept. of Mathematics, IIT Bombay,
Mumbai-76.
e-mail: sivaji.ganesh@gmail.com
Plan of lectures
(1) First order equations: Variable-Separable Method.
(2) Existence and uniqueness of solutions to initial value problems.
(3) Continuation of solutions, Saturated solutions, and Maximal interval of existence.
(4) Continuous dependence on data, Global existence theorem.
(5) Linear systems, Fundamental pairs of solutions, Wronskian.
References
1. V.I. Arnold, Ordinary dierential equations, translated by Silverman, (Printice-Hall of India,
1998).
2. E.A. Coddington, Introduction to ordinary dierential equations, (Prentice-Hall of India,
1974).
3. P. Hartman, Ordinary dierential equations, (Wiley, 1964).
4. M.W. Hirsh, S. Smale and R.L. Devaney, Dierential equations, dynamical systems & Chaos,
(Academic press, 2004).
5. L.C. Piccinini, G. Stampacchia and G. Vidossich, Ordinary dierential equations in R
N
:
problems and methods, (Springer-Verlag, 1984).
6. M.R.M. Rao, Ordinary dierential equations: theory and applications, (Aliated East-West,
1980).
7. D.A. Sanchez, Ordinary dierential equations: A brief eclectic tour, (Mathematical associa-
tion of America, 2002).
8. I.I. Vrabie, Dierential equations, (World scientic, 2004).
9. W.Wolfgang, Ordinary dierential equations, (Springer-Verlag, 1998).
S. Sivaji Ganesh ODE@AFS
2 1.1. BASIC CONCEPTS
Lecture-1
First order equations: Basic concepts
We introduce basic concepts of theory of ordinary dierential equations. A scalar ODE will be
given geometric interpretation and thereby try to gain a geometric understanding of solution
structure of ODE whose vector eld has some invariance. This understanding is then used to
solve equations of variable-separable type.
1.1 Basic concepts
We want to translate the feeling of what should be or what is an Ordinary Dierential Equation
(ODE) into mathematical terms. Dening some object like ODE, for which we have some rough
feeling, in English words is not really useful unless we know how to put it mathematically. As
such we can start by saying Let us look at the following dierential equation . . . . But since
many books give this denition, let us also have one such. The reader is referred to Remark 1.2
for an example of an ODE that we really do not want to be an ODE.
Let us start with
Hypothesis
Let R
n+1
be a domain and I R be an interval. Let F : I R be a function
dened by (x, z, z
1
, . . . z
n
) F(x, z, z
1
, . . . z
n
) such that F is not a constant function in
the variable z
n
.
With this notation and hypothesis on F we dene the basic object in our study, namely, an
Ordinary dierential equation.
Denition 1.1 (ODE) Assume the above hypothesis. An ordinary dierential equation of order
n is dened by the relation
F
_
x, y, y
(1)
, y
(2)
, . . . y
(n)
_
= 0, (1.1)
where y
(n)
stands for n
th
derivative of unknown function x y(x) with respect to the independent
variable x.
Remark 1.2
1. As we are going to deal with only one independent variable through out this course, we use the
terminology dierential equation in place of ordinary dierential equation at times. Also
we use the abbreviation ODE which stands for Ordinary Dierential Equation(s). Wherever
convenient, we use the notation prime

to denote a derivative w.r.t. independent variable
x; for example, y

is used to denote y
(1)
).
2. Note that the highest order of derivative of unknown function y appearing in the relation
(1.1) is called the order of the ordinary dierential equation. Look at the carefully framed
hypothesis above that makes sure the appearance of n
th
derivative of y in (1.1).
3. (Arnold) If we dene an ODE as a relation between an unknown function and its derivates,
then the following equation will also be an ODE.
dy
dx
(x) = y y(x). (1.2)
However, note that our Dention 1.1 does not admit (1.2) as an ODE. Also, we do not like
to admit (1.2) as an ODE since it is a non-local relation due to the presence of non-local
operator composition. On the other hand recall that derivative is a local operator in the
sense that derivative of a function at a point, depends only on the values of the function in
a neighbourhood of the point.
ODE@AFS S. Sivaji Ganesh
3
Having dened an ODE, we are interested in its solutions. This brings us to the question of
existence of solutions and nding out all the solutions. We make clear what we mean by a solution
of an ODE.
Denition 1.3 (Solution of an ODE) A real valued function is said to be a solution of ODE
(1.1) if (
n
(I) and
F
_
x, (x),
(1)
(x),
(2)
(x), . . .
(n)
(x)
_
= 0, x I. (1.3)
Remark 1.4
(1) There is no guarantee that an equation such as (1.1) will have a solution.
(i) The equation dened by F(x, y, y

) = (y

)
2
+y
2
+1 = 0 has no solution. Thus we
cannot hope to have a general theory for equations of type (1.1). Note that F is
a smooth function of its arguments.
(ii) The equation
y

=
_
1 if x 0
1 if x < 0,
does not have a solution on any interval containing 0. This follows from Darbouxs
theorem about derivative functions.
(2) To convince ourselves that we do not expect every ODE to have a solution, let us recall
the situation with other types of equations involving Polynomials, Systems of linear
equations, Implicit functions. In each of these cases, existence of solutions was proved
under some conditions. Some of those results also characterised equations that have
solution(s), for example, for systems of linear equations the characterisation was in
terms of ranks of matrix dening the linear system and the corresponding augmented
matrix.
(3) In the context of ODE, there are two basic existence theorems that hold for equations
in a special form called normal form. We state them in Section 3.1.
As observed in the last remark, we need to work with a less general class of ODE if we expect
them to have solutions. One such class is called ODE in normal form and is dened below.
Hypothesis (H)
Let R
n
be a domain and I R be an interval. Let f : I R be a continuous
function dened by (x, z, z
1
, . . . z
n1
) f(x, z, z
1
, . . . z
n1
).
Denition 1.5 (ODE in Normal form) Assume Hypothesis (H) on f. An ordinary dieren-
tial equation of order n is said to be in normal form if
y
(n)
= f
_
x, y, y
(1)
, y
(2)
, . . . y
(n1)
_
. (1.4)
Denition 1.6 (Solution of ODE in Normal form) A function C
n
(I
0
) where I
0
I is
a subinterval is called a solution of ODE (1.4) if for every x I
0
, the (n + 1)-tuple
_
x, (x),
(1)
(x),
(2)
(x), . . .
(n1)
(x)
_
I and

(n)
(x) = f
_
x, (x),
(1)
(x),
(2)
(x), . . .
(n1)
(x)
_
, x I
0
. (1.5)
Remark 1.7
S. Sivaji Ganesh ODE@AFS
4 1.2. GEOMETRIC INTERPRETATION OF A FIRST ORDER ODE AND ITS SOLUTION
1. Observe that we want equation (1.5) to be satised for all x I
0
instead of for all
x I. Compare now with dention of solution given before in Denition 1.3 which is more
stringent. We modied the concept of solution, by not requiring that the equation be satised
by the solution on entire interval I, due to various examples of ODEs that we shall see later
which have solutions only on a subinterval of I. We dont want to miss them!! Note that the
equation
y

=
_
1 if y 0
1 if y < 0,
does not admit a solution dened on R. However it has solutions dened on intervals (0, ),
(, 0). (Find them!)
2. Compare Denition 1.5 with Denition 1.1. See the item (ii) of Remark 1.2, observe that we
did not need any special eort in formulating Hypothesis (H) to ensure that n
th
derivative
makes an appearance in the equation (1.4).
Convention From now onwards an ODE in normal form will simply be called ODE for brevity.
Hypothesis (H
S
)
Let R
n
be a domain and I R be an interval. Let f : I R
n
be a continuous
function dened by (x, z) f (x, z) where z = (z
1
, . . . z
n
).
Denition 1.8 (System of ODEs) Assume Hypothesis (H
S
) on f . A rst order system of n
ordinary dierential equations is given by
y

= f (x, y) . (1.6)
The notion of solution for above system is dened analogous to Denition 1.5. A result due to
DAlembert enables us to restrict a general study of any ODE in normal form to that of a rst
order system in the sense of the following lemma.
Lemma 1.9 (DAlembert) An n
th
order ODE (1.4) is equivalent to a system of n rst order
ODEs.
Proof :
Introducing a transformation z = (z
1
, z
2
, . . . , z
n
) :=
_
y, y
(1)
, y
(2)
, . . . y
(n1)
_
, we see that z
satises the linear system
z

= (z
2
, . . . , z
n
, f(x, z)) (1.7)
Equivalence of (1.4) and (1.7) means starting from a solution of either of these ODE we can
produce a solution of the other. This is a simple calculation and is left as an exercise.
Note that the rst order system for z consists of n equations. This n is the order of (1.4).
Exercise 1.10 Dene higher order systems of ordinary dierential equations and dene corre-
sponding notion of its solution. Reduce the higher order system to a rst order system.
1.2 Geometric interpretation of a rst order ODE and its
solution
We now dene some terminology that we use while giving a geometric meaning of an ODE given
by
dy
dx
= f (x, y) . (1.8)
We recall that f is dened on a domain D in R
2
. In fact, D = I J where I, J are sub-intervals
of R.
ODE@AFS S. Sivaji Ganesh
5
Denition 1.11 (Line element) A line element associated to a point (x, y) D is a line passing
through the point (x, y) with slope p. We use the triple (x, y, p) to denote a line element.
Denition 1.12 (Direction eld/ Vector eld) A direction eld (sometimes called vector eld)
associated to the ODE (1.8) is collection of all line elements in the domain D where slope of the
line element associated to the point (x, y) has slope equal to f (x, y). In other words, a direction
eld is the collection
_
(x, y, f (x, y) ) : (x, y) D
_
.
Remark 1.13 (Interpretations)
1. The ODE (1.8) can be thought of prescribing line elements in the domain D.
2. Solving an ODE can be geometrically interpreted as nding curves in D that t the direction
eld prescribed by the ODE. A solution (say ) of the ODE passing through a point (x
0
, y
0
)
D (i.e., (x
0
) = y
0
) must satisfy

(x
0
) = f(x
0
, y
0
). In other words,
(x
0
, y
0
,

(x
0
) ) = (x
0
, y
0
, f (x
0
, y
0
) ).
3. That is, the ODE prescribes the slope of the tangent to the graph of any solution (which is
equal to

(x
0
)). This can be seen by looking at the graph of a solution.
4. Drawing direction eld corresponding to a given ODE and tting some curve to it will end up
in nding a solution, at least, graphically. However note that it may be possible to t more
than one curve passing through some points in D, which is the case where there are more
than one solution to ODE around those points. Thus this activity (of drawing and tting
curves) helps to get a rough idea of nature of solutions of ODE.
5. A big challenge is to draw direction eld for a given ODE. One good starting point is to
identify all the points in domain D at which line element has the same slope and it is easy
to draw all these lines. These are called isoclines; the word means leaning equally.
Exercise 1.14 Draw the direction eld prescribed by ODEs where f(x, y) = 1, f(x, y) = x,
f(x, y) = y, f(x, y) = y
2
, f(x, y) = 3y
2/3
and t solution curves to them.
Finding a solution of an ODE passing through a point in D is known as Initial value problem. We
address this in the next section.
1.3 Initial Value Problems
We consider an Initial Value Problem (also called Cauchy problem) for an ODE (1.4). It consists
of solving (1.4) subject to what are called Initial conditions. The two basic theorems we are going
to present are concerning an IVP for a rst order ODE.
Denition 1.15 (Initial Value Problem for an ODE) Let x
0
I and (y
1
, y
2
, . . . , y
n
) be
given. An Initial Value Problem (IVP) for an ODE in normal form is a relation satised by an
unknown function y given by
y
(n)
= f
_
x, y, y
(1)
, y
(2)
, . . . y
(n1)
_
, y
(i)
(x
0
) = y
i
, i = 0, . . . , (n 1). (1.9)
Denition 1.16 (Solution of an IVP for an ODE) A solution of ODE (1.4) (see Deni-
tion 1.6) is said to be a solution of IVP if x
0
I
0
and

(i)
(x
0
) = y
i
, i = 0, . . . , (n 1).
This solution is denoted by (.; x
0
, y
0
, y
1
, . . . , y
n1
) to remember the IVP solved by .
S. Sivaji Ganesh ODE@AFS
6 1.3. INITIAL VALUE PROBLEMS
Denition 1.17 (Local and Global solutions of an IVP) Let be a solution of an IVP for
ODE (1.4) according to Denition 1.16.
1. If I
0
I, then is called a local solution of IVP.
2. If I
0
= I, then is called a global solution of IVP.
Remark 1.18
1. Note that in all our denitions of solutions, a solution always comes with its domain of
denition. Sometimes it may be possible to extend the given solution to a bigger domain.
We address this issue in the next lecture.
2. When n = 1, geometrically speaking, graph of solution of an IVP is a curve passing through
the point (x
0
, y
0
).
Exercise 1.19 Dene an IVP for a rst order system. Reduce an IVP for an n
th
order ODE to
that of an equivalent rst order system.
ODE@AFS S. Sivaji Ganesh
7
Lecture-2
First order equations: Variable-Separable method
Hypothesis (H
VS
)
Let I R and J R be intervals. Let g : I R and h : J R 0 be continuous
functions.
If domains of functions are not specied, then they are assumed to be their natural
domains.
We consider a rst order ODE in Variable-Separable form given by
dy
dx
= g(x)h(y) (2.1)
We gain an understanding of the equation (2.1) and its solution in three simple steps.
2.1 Direction eld independent of y
The equation (2.1) takes the form
dy
dx
= g(x) (2.2)
Geometric understanding: Observe that slope of line element associated to a point depends
only on its x coordinate. Thus the direction eld is dened on the strip I R and is invariant
under translation in the direction of Y axis. Therefore it is enough to draw line elements for
points in the set I 0. This suggests that if we know a solution curve then its translates in the
Y axis direction gives all solution curves.
Let x
0
I be xed and let us dene a primitive of g on I by
G(x) :=
x
_
x
0
g(s) ds, (2.3)
with the understanding that, if x < x
0
, dene G(x) :=
x
0
_
x
g(s) ds.
The function G is a solution of ODE (2.2) by fundamental theorem of integral calculus, since
G

(x) = g(x) on I. Moreover, G(x

0
) = 0.
A solution curve passing through an arbitrary point (, ) I R can be obtained from G and is
given by
y(x; , ) = G(x) + ( G()), (2.4)
which is of the form G(x)+C where C = G(). Thus all solutions of ODE (2.2) are determined.
Moreover, all the solutions are global.
Example 2.1 Solve the ODE (2.2) with g given by (i). x
2
+ sin x (ii). x

1 x
2
(iii). sin x
2
(iv) Have you run out of patience? An explicit formula for the solution does not mean having it
explicitly!
S. Sivaji Ganesh ODE@AFS
8 2.2. DIRECTION FIELD INDEPENDENT OF X
2.2 Direction eld independent of x
The equation (2.1) takes the form
dy
dx
= h(y). (2.5)
This type of equations are called autonomous since the RHS of the equation does not depend
on the independent variable x.
Geometric understanding: Observe that slope of line element associated to a point depends
only on its y coordinate. Thus the direction eld is dened on the strip R J and is invariant
under translation in the direction of X axis. Therefore it is enough to draw line elements for
points in the set 0 I. This suggests that if we know a solution curve then its translates in the
X axis direction gives solution curves. Note that this is one of the main features of autonomous
equations.
Exercise 2.2 Verify that x
3
, (x c)
3
) with c R arbitrary are solutions of the ODE y

= 3y
2/3
.
Verify that (x) 0 is also a solution. However this can not be obtained by translating any of
the other solutions. Does this contradict the observation preceding this exercise?
Exercise 2.3 Formulate the geometric observation if we know a solution curve then its translates
in the X axis direction gives solution curves as a mathematical statement and prove it. Compare
and contrast with a similar statement made in Step 1.
This case is considerably dierent from that of Step 1 where solutions are dened on the entire
interval I. To expect the diculties awaiting us in the present case, it is advisable to solve the
following exercise.
Exercise 2.4 Verify that (x) =
1
(x+c)
are solutions of y

= y
2
on certain intervals. Graph the
solutions for c = 0, 1, 2. Verify that (x) 0 is also a solution on R.
Note from the last exercise that it has a constant solution (also called, rest/equilibrium point
since it does not move!) dened globally on R and also non-constant solutions dened on only a
subinterval of the real line that varies with c. There is no way we can get a non-zero solution from
zero solution by translation. Thus it is a good time to review Exercise 2.3 and nd out if there is
any contradition.
Observation: If J satises h() = 0, then is a rest point and (x) = is a solution of (2.5)
for x R.
Since the function h is continuous, it does not change sign. Therefore we may assume, without
loss of generality (WLOG), that h(y) > 0 for y J. There is no loss of generality because the
other case, namely h(y) < 0 for y J, can be disposed o in a similar manner.
Formally speaking, ODE (2.5) may be written as
dx
dy
=
1
h(y)
.
Thus all the conclusions of Step 1 can be translated to the current problem. But inversion of
roles of x and y needs to be justied!
Let y
0
J be xed and let us dene a primitive of 1/h on J by
H(y) :=
y
_
y
0
1
h(s)
ds. (2.6)
We record some properties of H below.
1. The function H : J R is dierentiable (follows from fundamental theorem of integral
calculus).
ODE@AFS S. Sivaji Ganesh
9
2. Since h > 0 on J, the function H is strictly monotonically increasing. Consequently, H is
one-one.
3. The function H : J H(J) is invertible. By denition, H
1
: H(J) J is onto, and is also
dierentiable.
4. A ner observation yields H(J) is an interval containing 0.
We write
H H
1
(x) = x
A particular solution
The function H gives rise to an implicit expression for a solution y of ODE (2.5) given by
H(y) = x (2.7)
This assertion follows from the simple calculation
d
dx
H(y) =
dH
dy
dy
dx
=
1
h(y)
dy
dx
= 1.
Note that H(y
0
) = 0 and hence x = 0. This means that the graph of the solution y(x) := H
1
(x)
passes through the point (0, y
0
). Note that this solution is dened on the interval H(J) which may
not be equal to R.
Exercise 2.5 When will the equality H(J) = R hold? Think about it. Try to nd some examples
of h for which the equality H(J) = R holds.
Some more solutions
Note that the function H
1
(x c) is also a solution on the interval c + H(J) (verify). Moreover,
these are all the solutions of ODE (2.5).
All solutions
Let z be any solution of ODE (2.5) dened on some interval I
z
. Then we have
dz
dx
= z

(x) = h(z(x))
Let x
0
I
z
and let z(x
0
) = z
0
. Since h ,= 0 on J, integrating both sides from x
0
to x yields
_
x
x
0
z

(x)
h(z(x))
dx =
_
x
x
0
1 dx
which reduces to
_
x
x
0
z

(x)
h(z(x))
dx = x x
0
The last equality, in terms of H dened by equation (2.6), reduces to
H(z)
z
0
_
y
0
1
h(s)
ds = x x
0
i.e., H(z) H(z
0
) = x x
0
(2.8)
From the equation (2.8), we conclude that x x
0
+ H(z
0
) H(J).
Thus, for x belonging to the interval x
0
H(z
0
) + H(J), we can write the solution as
z(x) = H
1
_
x x
0
+ H(z
0
)
_
. (2.9)
Note that a solution of ODE (2.5) are not global in general.
S. Sivaji Ganesh ODE@AFS
10 2.3. GENERAL CASE
Exercise 2.6 Give a precise statement of what we proved about solutions of ODE (2.5).
Exercise 2.7 Prove that a solution curve passes through every point of the domain R J.
Exercise 2.8 In view of what we have proved under the assumption of non-vanishing of h, revisit
Exercise 2.4, Exercise 2.3 and comment.
Exercise 2.9 Solve the ODE (2.5) with h given by (i). ky (ii).
_
[y[ (iii). 1 + y
2
(iv). 3y
2/3
(v). y
4
+ 4 (vi).
1
y
4
+4
2.3 General case
We justify the formal calculation used to obtain solutions of ODE (2.1) which we recall here for
convenience.
dy
dx
= g(x)h(y) (2.10)
The formal calculation to solve (2.10) is
_
dy
h(y)
=
_
g(x) dx + C (2.11)
We justify this formal calculation on a rigorous footing. We introduce two auxiliary ODEs:
dy
du
= h(y) and
du
dx
= g(x). (2.12)
Notation We refer to the rst and second equations of (2.12) as equations (2.12a) and (2.12b)
respectively. We did all the hard work in Steps 1-2 and we can easily deduce results concerning
ODE (2.10) using the auxiliary equations (2.12) introduced above.
Claim Let (x
1
, y
1
) I J be an arbitrary point. A solution y satisfying y(x
1
) = y
1
is given by
y(x) = H
1
_
G(x) G(x
1
) + H(y
1
)
_
. (2.13)
Proof :
Recalling from Step 2 that the function H
1
is onto J, there exists a unique u
1
such that
H
1
(u
1
) = y
1
. Now solution of (2.12a) satisfying y(u
1
) = y
1
is given by (see, formula (2.9))
y(u) = H
1
_
u u
1
+ H(y
1
)
_
(2.14)
and is dened on the interval u
1
H(y
1
) + H(J).
Solution of (2.12b), dened on I, satisfying u(x
1
) = u
1
is given by (see, formula (2.4))
u(x; x
1
, u
1
) = G(x) + (u
1
G(x
1
)).G(x) G(x
1
) H(y
1
) + H(J) (2.15)
Combining the two formulae (2.14)-(2.15), we get (2.13). This formula makes sense for x I such
that G(x) + (u
1
G(x
1
)) u
1
H(y
1
) + H(J). That is, G(x) G(x
1
) H(y
1
) + H(J).
Thus a general solution of (2.1) is given by
y(x) = H
1
_
G(x) c
_
. (2.16)
This ends the analysis of variable-separable equations modulo the following exercise.
Exercise 2.10 Prove that the set
_
x I : G(x) G(x
1
) H(y
1
) + H(J)
_
is non-empty and
contains a non-trivial interval.
Exercise 2.11 Prove: Initial value problems for variable-separable equations under Hypothesis
(H
VS
) have unique solutions. Why does it not contradict observations made regarding solutions
of y

= 3y
2/3
satisfying y(0) = 0?
ODE@AFS S. Sivaji Ganesh
11
Lecture-3
First order equations: Local existence & Uniqueness theory
We discuss the twin issues of existence and uniqueness for Initial value problems corresponding
to rst order systems of ODE. This discussion includes the case of scalar rst order ODE and also
general scalar ODE of higher order in view of Exercise 1.19. However certain properties like
boundedness of solutions do not carry over under the equivalence of Exercise 1.19 which are used
in the discussion of extensibility of local solutions to IVP.
We compliment the theory with examples from the class of rst order scalar equations. We recall
the basic setting of IVP for systems of ODE which is in force through our discussion. Later on
only the additional hypotheses are mentioned if and when they are made.
Hypothesis (H
IVPS
)
Let R
n
be a domain and I R be an interval. Let f : I R
n
be a continuous
function dened by (x, y) f (x, y) where y = (y
1
, . . . y
n
). Let (x
0
, y
0
) I be an
arbitrary point.
Denition 3.1 Assume Hypothesis (H
IVPS
) on f . An IVP for a rst order system of n ordinary
dierential equations is given by
y

= f (x, y) , y(x
0
) = y
0
. (3.1)
As we saw before (Exercise 2.4) we do not expect a solution to be dened globally on the entire
interval I. Recall Remark 1.7 in this context. This motivates the following denition of solution
of an IVP for systems of ODE.
Denition 3.2 (Solution of an IVP for systems of ODE) An n-tuple of functions
u = (u
1
, . . . u
n
) C
1
(I
0
) where I
0
I is a subinterval containing the point x
0
I is called a
solution of IVP (3.1) if for every x I
0
, the (n + 1)-tuple (x, u
1
(x), u
2
(x), , u
n
(x)) I ,
u

(x) = f (x, u(x)) , x I

0
and u(x
0
) = y
0
. (3.2)
We denote this solution by u = u(x; f , x
0
, y
0
) to remind us that the solution depends on f , y
0
and
u(x
0
) = y
0
.
The boldface notation is used to denote vector quantities and we drop boldface for scalar quan-
tities.
Remark 3.3 The IVP (3.1) involves an interval I, a domain , a continuous function f on
I , x
0
I, y
0
. Given I, x
0
I and , we may pose many IVPs by varying the data (f , y
0
)
belonging to the set C(I ) . .
Basic questions
There are three basic questions associated to initial value problems. They are
(i) Given (f , y
0
) C(I ) , does the IVP (3.1) admit at least one solution?
(ii) Assuming that for a given (f , y
0
) C(I ) , the IVP (3.1) has a solution, is the solution
unique?
(iii) Assuming that for each (f , y
0
) C(I) the IVP admits a unique solution y(x; f , x
0
, y
0
)
on a common interval I
0
containing x
0
, what is the nature of the following function?
o : C(I ) C
1
(I
0
) (3.3)
dened by
(f , y
0
) y(x; f , x
0
, y
0
). (3.4)
S. Sivaji Ganesh ODE@AFS
12 3.1. EXISTENCE OF LOCAL SOLUTIONS
We address questions (i), (ii) and (iii) in Sections 3.1, 3.2 and 4.3 respectively. Note that we do not
require a solution to IVP be dened on the entire interval I but only on a subinterval containing
the point x
0
at which initial condition is prescribed. Thus it is interesting to nd out if every
solution can be extended to I and the possible obstructions for such an extension. We discuss this
issue in the next lecture.
3.1 Existence of local solutions
There are two important results concerning existence of solutions for IVP (3.1). One of them
is proved for any function f satisfying Hypothesis (H
IVPS
) and the second assumes Lipschitz
continuity of the function f in addition. As we shall see in Section 3.2, this extra assumption on
f gives rise not only to another proof of existence but also uniqueness of solutions.
Both proofs are based on equivalence of IVP (3.1) and an integral equation.
Lemma 3.4 A continuous function y dened on an interval I
0
containing the point x
0
is a solu-
tion of IVP (3.1) if and only if y satises the integral equation
y(x) = y
0
+
_
x
x
0
f (s, y(s)) ds x I
0
. (3.5)
Proof :
If y is a solution of IVP (3.1), then by denition of solution we have
y

(x) = f (x, y(x)). (3.6)

Integrating the above equation from x
0
to x yields the integral equation (3.5).
On the other hand let y be a solution of integral equation (3.5). Observe that, due to continuity
of the function t y(t), the function t f (t, y(t)) is continuous on I
0
. Thus RHS of (3.5) is
a dierentiable function w.r.to x by fundamental theorem of integral calculus and its derivative
is given by the function x f (x, y(x)) which is a continuous function. Thus y, being equal to
a continuously dierentiable function via equation (3.5), is also continuously dierentiable. The
function y is a solution of ODE (3.1) follows by dierentiating the equation (3.5). Evaluating
(3.5) at x = x
0
gives the initial condition y(x
0
) = y
0
.
Thus it is enough to prove the existence of a solution to the integral equation (3.5) for showing
the existence of a solution to the IVP (3.1). Both the existence theorems that we are going to
prove, besides making use of the above equivalence, are proved by an approximation procedure.
The following result asserts that joining two solution curves in the extended phase space I
gives rise to another solution curve. This result is very useful in establishing existence of solutions
to IVPs; one usually proves the existence of a solution to the right and left of the point x
0
at
which initial condition is prescribed and then one gets a solution (which should be dened in an
open interval containing x
0
) by joining the right and left solutions at the point (x
0
, y
0
).
Lemma 3.5 (Concactenation of two solutions) Assume Hypothesis (H
IVPS
). Let [a, b] and
[b, c] be two subintervals of I. Let u and w dened on intervals [a, b] and [b, c] respectively be
solutions of IVP with initial data (a, ) and (b, u(b)) respectively. Then the concatenated function
z dened on the interval [a, c] by
z
(
x) =
_
u(x) if x [a, b].
w(x) if x (b, c].
(3.7)
is a solution of IVP with initial data (a, ).
ODE@AFS S. Sivaji Ganesh
13
Proof :
It is easy to see that the function z is continuous on [a, c]. Therefore, by Lemma 3.4, it is
enough to show that z satises the integral equation
z(x) = +
_
x
a
f (s, z(s)) ds x [a, c]. (3.8)
Clearly the equation (3.8) is satised for x [a, b], once again, by Lemma 3.4 since u solves IVP
with initial data (a, ) and z
(
x) = u(x) for x [a, b]. Thus it remains to prove (3.8) for x (b, c].
For x (b, c], once again by Lemma 3.4, we get
z(x) = w(x) = u(b) +
_
x
b
f (s, w(s)) ds = u(b) +
_
x
b
f (s, z(s)) ds. (3.9)
Since
u(b) = +
_
b
a
f (s, u(s)) ds = +
_
b
a
f (s, z(s)) ds, (3.10)
substituting for u(b) in (3.9) nishes the proof of lemma.
Rectangles
As I is an open set, we do not know if functions dened on this set are bounded; also we do
not know the shape or size of . If we are looking to solve IVP (3.1) (i.e., to nd a solution curve
passing through the point (x
0
, y
0
)), we must know how long a solution (if and when it exists)
may live. In some sense this depends on size of a rectangle R I centred at (x
0
, y
0
) dened
by two positive real numbers a, b
R = x : [x x
0
[ a y : |y y
0
| b (3.11)
Since I is an open set, we can nd such an R (for some positive real numbers a, b) for each
point (x
0
, y
0
) I . Let M be dened by M = sup
R
|f (x, y|.
Note that rectangle R in (3.11) is symmetric in the x space as well. However a solution may be
dened on an interval that is not symmetric about x
0
. Thus it looks restrictive to consider R
as above. It is denitely the case when x
0
is very close to one of the end points of the interval
I. Indeed in results addressing the existence of solutions for IVP, separately on intervals left and
right to x
0
consider rectangles R

I of the form
R

= [x
0
, x
0
+ a] y : |y y
0
| b (3.12)
3.1.1 Existence theorem of Peano
Theorem 3.6 (Peano) Assume Hypothesis (H
IVPS
). Then the IVP (3.1) has at least one solu-
tion on the interval [x x
0
[ where = mina,
b
M
.
3.1.2 Cauchy-Lipschitz-Picard existence theorem
From real analysis, we know that continuity of a function at a point is a local concept (as it
involves values of the function in a neighbourhood of the point at which continuity of function is
in question). We talk about uniform continuity of a function with respect to a domain. Similarly
we can dene Lipschitz continuity at a point and on a domain of a function dened on subsets of
R
n
. For ODE purposes we need functions of (n + 1) variables and Lipschitz continuity w.r.t. the
last n variables. Thus we straight away dene concept of Lipschitz continuity for such functions.
Let R I be a rectangle centred at (x
0
, y
0
) dened by two positive real numbers a, b (see
equation (3.11)).
S. Sivaji Ganesh ODE@AFS
14 3.1. EXISTENCE OF LOCAL SOLUTIONS
Denition 3.7 (Lipschitz continuity) A function f is said to be Lipschitz continuous on a
rectangle R with respect to the variable y if there exists a K > 0 such that
|f (x, y
1
) f (x, y
1
)| K|y
1
y
2
| (x, y
1
), (x, y
2
) R. (3.13)
Exercise 3.8 1. Let n = 1 and f be dierentiable w.r.t. the variable y with a continuous
derivative dened on I . Show that f is Lipschitz continuous on any rectangle R I .
2. If f is Lipschitz continuous on every rectangle R I , is f dierentiable w.r.t. the
variable y?
3. Prove that the function h dened by h(y) = y
2/3
on [0, ) is not Lipschitz continuous on
any interval containing 0.
4. Prove that the function f(x, y) = y
2
dened on doamin R R is not Lipschitz continuous.
(this gives yet another reason to dene Lipschitz continuity on rectangles)
We now state the existence theorem and the method of proof is dierent from that of Peano
theorem and yields a bilateral interval containing x
0
on which existence of a solution is asserted.
Theorem 3.9 (Cauchy-Lipschitz-Picard) Assume Hypothesis (H
IVPS
). Let f be Lipschitz
continuous with respect to the variable y on R. Then the IVP (3.1) has at least one solution
on the interval J [x x
0
[ where = mina,
b
M
.
Proof :
Step 1: Equivalent integral equation We recall (3.5), which is equivalent to the given IVP below.
y(x) = y
0
+
_
x
x
0
f (s, y(s)) ds x I. (3.14)
By the equivalence of above integral equation with the IVP, it is enough to prove that the integral
equation has a solution. This proof is accomplished by constructing, what are known as Picard
approximations, a sequence of functions that converges to a solution of the integral equation (3.14).
Step 2: Construction of Picard approximations
Dene the rst function y
0
(x), for x I, by
y
0
(x) := y
0
(3.15)
Dene y
1
(x), for x I, by
y
1
(x) := y
0
+
_
x
x
0
f (s, y
0
(s)) ds. (3.16)
Note that the function y
1
(x) is well-dened for x I. However, when we try to dene the next
member of the sequence, y
2
(x), for x I, by
y
2
(x) := y
0
+
_
x
x
0
f (s, y
1
(s)) ds, (3.17)
caution needs to exercised. This is because, we do not know about the values that the function
y
1
(x) assumes for x I, there is no reason that those values are inside . However, it happens
that for x J, where the interval J is as in the statement of the theorem, the expression on RHS
of (3.17) which dened function y
2
(x) is meaningful, and hence the function y
2
(x) is well-dened
for x J. By restricting to the interval J, we can prove that the following sequence of functions
is well-dened: Dene for k 1, for x J,
y
k
(x) = y
0
+
_
x
x
0
f (s, y
k1
(s)) ds. (3.18)
ODE@AFS S. Sivaji Ganesh
15
Proving the well-denedness of Picard approximations is left as an exercise, by induction. In fact,
the graphs of each Picard approximant lies inside the rectangle R (see statement of our theorem).
That is,
|y
k
(x) y
0
| b, x [x
0
, x
0
+ ]. (3.19)
The proof is immediate from
y
k
(x) y
0
=
_
x
x
0
f (s, y
k1
(s)) ds, x J. (3.20)
Therefore, |y
k
(x) y
0
| M[x x
0
[, x J. (3.21)
and for x J, we have [x x
0
[ .
Step 3: Convergence of Picard approximations
We prove the uniform convergence of sequence of Picard approximations y
k
on the interval J, by
proving that this sequence corresponds to the partial sums of a uniformly convergent series of
functions, and the series is given by
y
0
+

l=0
_
y
l+1
(x) y
l
(x)

, (3.22)
Note that the sequence y
k+1
corresponds to partial sums of series (3.22). That is,
y
k+1
(x) = y
0
+
k

l=0
_
y
l+1
(x) y
l
(x)

. (3.23)
Step 3A: Unifrom convergence of series (3.22) on x J
We are going to compare series (3.22) with a convergence series of real numbers, uniformly in
x J, and thereby proving uniform convergence of the series. From the expression
y
l+1
(x) y
l
(x) =
_
x
x
0
_
f (s, y
l
(s)) f (s, y
l1
(s))
_
ds, x J, (3.24)
we can prove by induction the estimate (this is left an exercise):
|y
l+1
(x) y
l
(x)| ML
l
[x x
0
[
l+1
(l + 1)!

M
L
L
l+1

l+1
(l + 1)!
(3.25)
We conclude that the series (3.22), and hence the sequence of Picard iterates, converge uniformly
on J. This is because the above estimate (3.25) says that general term of series (3.22) is uniformly
smaller than that of a convergent series, namely, for the function e
L
times a constant.
Let y(x) denote the uniform limit of the sequence of Picard iterates y
k
(x) on J.
Step 4: The limit function y(x) solves integral equation (3.14)
We want to take limit as k in
y
k
(x) = y
0
+
_
x
x
0
f (s, y
k1
(s)) ds. (3.26)
Taking limit on LHS of (3.26) is trivial. Therefore, for x J, if we prove that
_
x
x
0
f (s, y
k1
(s)) ds
_
x
x
0
f (s, y(s)) ds, (3.27)
S. Sivaji Ganesh ODE@AFS
16 3.2. UNIQUENESS
then we would obtain, for x J,
y(x) = y
0
+
_
x
x
0
f (s, y(s)) ds, (3.28)
and this nishes the proof. Therefore, it remains to prove (3.27). Let us estimate, for x J, the
quantity
_
x
x
0
f (s, y
k1
(s)) ds
_
x
x
0
f (s, y(s)) ds =
_
x
x
0
_
f (s, y
k1
(s)) f (s, y(s))
_
ds (3.29)
Since the graphs of y
k
lie inside rectangle R, so does the graph of y. This is because rectangle R
is closed. Now we use that the vector eld f is Lipschitz continuous (with Lipschitz constant L)
in the variable y on R, we get
_
_
_
_
_
x
x
0
_
f (s, y
k1
(s)) f (s, y(s))
_
ds
_
_
_
_

_
x
x
0
_
_
_
f (s, y
k1
(s)) f (s, y(s))
__
_
ds

(3.30)
L

_
x
x
0
|y
k1
(s) y(s)| ds

L[x x
0
[ sup
J
|y
k1
(x) y(x)| (3.31)
L sup
J
|y
k1
(x) y(x)| . (3.32)
This estimate nishes the proof of (3.27), since y(x) is the uniform limit of the sequence of Picard
iterates y
k
(x) on J, and hence for suciently large k, the quantity sup
J
|y
k1
(x) y(x)| can be
made arbitrarily small.
Some comments on existence theorems
Remark 3.10 (i) Note that the interval of existence depends only on the bound M and not on
the specic function.
(ii) If g is any Lipschitz continuous function (with respect to the variable y on R) in a neigh-
bourhood of f and is any vector in an neighbourhood of y
0
, then solution to IVP with data
(g, ) exists on the interval = mina,
b
M+
. Note that this interval depends on data (g, )
only in terms of its distance to (f , y
0
).
Exercise 3.11 Prove that Picards iterates need not converge if the vector eld does is not locally
Lipschitz. Compute successive approximations for the IVP
y

= 2x x

y
+
, y(0) = 0, with y
+
= maxy, 0,
and show that they do not converge, and also show that IVP has a unique solution. (Hint: y
2n
=
0, y
2n+1
= x
2
, n N).
3.2 Uniqueness
Recalling the denition of a solution, we note that if u solves IVP (3.1) on an interval I
0
then
w
def
= u[
I
1
is also a solution to the same IVP where I
1
is any subinterval of I
0
containing the point
x
0
. In principle we do not want to consider the latter as a dierent solution. Thus we are led to
dene a concept of equivalence of solutions of an IVP that does not distinguish w from u near
the point x
0
. Roughly speaking, two solutions of IVP are said to be equivalent if they agree on
some interval containing x
0
(neighbourhood of x
0
). This neighbourhood itself may depend on the
given two solutions.
Denition 3.12 (local uniqueness) An IVP is said to have local uniqueness property if for
each (x
0
, y
0
) I and for any two solutions y
1
and y
2
of IVP (3.1) dened on intervals I
1
and I
2
respectively, there exists an open interval I
lr
:= (x
0

l
, x
0
+
r
) containing the point x
0
such that y
1
(x) = y
2
(x) for all x I
lr
.
ODE@AFS S. Sivaji Ganesh
17
Denition 3.13 (global uniqueness) An IVP is said to have global uniqueness property if for
each (x
0
, y
0
) I and for any two solutions y
1
and y
2
of IVP (3.1) dened on intervals I
1
and I
2
respectively, the equality y
1
(x) = y
2
(x) holds for all x I
1
I
2
.
Remark 3.14 1. It is easy to understand the presence of adjectives local and global in Deni-
tion 3.12 and Denition 3.12 respectively.
2. There is no loss of generality in assuming that the interval appearing in Denition 3.12,
namely I
lr
, is of the form J
lr
:= [x
0

l
, x
0
+
r
]. This is because in any open interval
containing a point x
0
, there is a closed interval containing the same point x
0
and vice versa.
Though it may appear that local and global uniqueness properties are quite dierent from each
other, indeed they are the same. This is the content of the next result.
Lemma 3.15 The following are equivalent.
1. An IVP has local uniqueness property.
2. An IVP has global uniqueness property.
Proof :
From denitions, clearly (2) =(1). We turn to the proof of (1) =(2).
Let y
1
and y
2
be two solutions of IVP (3.1) dened on intervals I
1
and I
2
respectively. We prove
that y
1
= y
2
on the interval I
1
I
2
; we split its proof in to two parts. We rst prove the equality
to the right of x
0
, i.e., on the interval [x
0
, sup(I
1
I
2
)) and proving the equality to the left of x
0
(i.e., on the interval (inf(I
1
I
2
), x
0
] ) follows a canonically modied argument. Let us consider
the following set
/
r
= t I
1
I
2
: y
1
(x) = y
2
(x) x [x
0
, t] . (3.33)
The set /
r
has the following properties:
1. The set /
r
is non-empty. This follows by applying local uniqueness property of the IVP
with initial data (x
0
, y
0
).
2. The equality sup /
r
= sup(I
1
I
2
) holds.
Proof :
Note that inmum and supremum of an open interval equals the left and right end points of
its closure (the closed interval) whenever they are nite. Observe that sup /
r
sup(I
1
I
2
)
since /
r
I
1
I
2
. Thus it is enough to prove that strict inequality can not hold. On the
contrary, let us assume that a
r
:= sup /
r
< sup(I
1
I
2
). This means that a
r
I
1
I
2
and
hence y
1
(a
r
), y
2
(a
r
) are dened. Since a
r
is the supremum (in particular, a limit point) of
the set /
r
on which y
1
and y
2
coincide, by continuity of functions y
1
and y
2
on I
1
I
2
, we
get y
1
(a
r
) = y
2
(a
r
).
Thus we nd that the functions y
1
and y
2
are still solutions of ODE on the interval I
1
I
2
,
and also that y
1
(a
r
) = y
2
(a
r
). Thus applying once again local uniquness property of IVP but
with initial data (a
r
, y
1
(a
r
)), we conclude that y
1
= y
2
on an interval J
lr
:= [a
r

l
, a
r
+
r
]
(see Remark 3.14). Thus combining with arguments of previous paragraph we obtain the
equality of functions y
1
= y
2
on the interval [x
0
, a
r
+
r
]. This means that a
r
+
r
/
r
and thus a
r
is not an upper bound for /
r
. This contradiction to the denition of a
r
nishes
the proof of 2.
As mentioned at the beginning of the proof, similar statements to the left of x
0
follow. This
nishes the proof of lemma.
S. Sivaji Ganesh ODE@AFS
18 3.2. UNIQUENESS
Remark 3.16 1. By Lemma 3.15 we can use either of the two denitions Denition 3.12 and
Denition 3.12 while dealing with questions of uniqueness. Henceforth we use the word
uniqueness instead of using adjectives local or global since both of them are equivalent.
2. One may wonder then, why there is a need to dene both local and global uniqueness prop-
erties. The reason is that it is easy to prove local uniqueness compared to proving global
uniqueness and at the same retaining what we intuitively feel about uniqueness.
Example 3.17 (Peano) The initial value problem
y

= 3 y
2/3
, y(0) = 0. (3.34)
has innitely many solutions.
However there are sucient conditions on f so that the corresponding IVP has a unique solution.
One such condition is that of Lipschitz continuity w.r.t. variable y.
Lemma 3.18 Assume Hypothesis (H
IVPS
). If f is Lipschitz continuous on every rectangle R
contained in I , then we have global uniqueness.
Proof :
Let y
1
and y
2
be two solutions of IVP (3.1) dened on intervals I
1
and I
2
respectively. By
Lemma 3.4, we have
y
i
(x) = y
0
+
_
x
x
0
f (s, y
i
(s)) ds, x I
i
and i = 1, 2. (3.35)
Subtracting one equation from another we get
y
1
(x) y
2
(x) =
_
x
x
0
_
f (s, y
1
(s)) f (s, y
2
(s))
_
ds, x I
1
I
2
. (3.36)
Applying norm on both sides yields, for x I
1
I
2
|y
1
(x) y
2
(x)| =
_
_
_
_
_
x
x
0
_
f (s, y
1
(s)) f (s, y
2
(s))
_
ds
_
_
_
_
(3.37)

_
x
x
0
_
_
f (s, y
1
(s)) f (s, y
2
(s))
_
_
ds (3.38)
Choose such that |y
1
(s)y
0
| b and |y
2
(s)y
0
| b, since we know that f is locally Lipschitz,
it will be Lipschitz on the rectangle R with Lipschitz constant L > 0. As a consequence we get
|y
1
(x) y
2
(x)| L sup
|xx
0
|
|y
1
(x) y
2
(x)| [x x
0
[ (3.39)
L sup
|xx
0
|
|y
1
(x) y
2
(x)| (3.40)
It is possible to arrange such that L < 1. From here we conclude that
sup
|xx
0
|
|y
1
(x) y
2
(x)| < sup
|xx
0
|
|y
1
(x) y
2
(x)| (3.41)
Thus we conclude sup
|xx
0
|
|y
1
(x) y
2
(x)| = 0. This establishes local uniqueness and global
uniqueness follows from their equivalence.
Example 3.19 The IVP
y

=
_
y sin
1
y
if y ,= 0
0 if y = 0,
y(0) = 0.
has unique solution, despite the RHS not being Lipschitz continuous w.r.t. variable y on any
rectangle containing (0, 0).
ODE@AFS S. Sivaji Ganesh
19
Lecture-4
Saturated solutions & Maximal interval of existence
4.1 Continuation
We answer the following two questions in this section.
1. When can a given solution be continued?
2. If a solution can not be continued to a bigger interval, what prevents a continuation?
The existential results of Section 3.1 provide us with an interval containing the initial point on
which solution for IVP exists and the length of the interval depends on the data of the problem
as can be seen from expression for its length. However this does not rule out solutions of IVP
dened on bigger intervals if not whole of I. In this section we address the issue of extending a
given solution to a bigger interval and the diculties that arise in extending.
Intuitive idea for extension Take a local solution u dened on an interval I
0
= (x
1
x
2
) con-
taining the point x
0
. An intuitive idea to extend u is to take the value of u at the point x = x
2
and consider a new IVP for the same ODE by posing the initial condition prescribed at the point
x = x
2
to be equal to u(x
2
). Consider this IVP on a rectangle containing the point (x
2
, u(x
2
)).
Now apply any of the existence theorems of Section 3.1 and conclude the existence of solution on
a bigger interval. Repete the previous steps and obtain a solution on the entire interval I.
However note that this intuitive idea may fail for any one of the following reasons. They are (i).
limit of function u does not exist as x x
2
(ii). limit in (i) may exist but there may not be any
rectangle around the point (x
2
, u(x
2
)) as required by existence theorems. In fact these are the
principle diculties in extending a solution to a bigger interval. In any case we show that for any
solution there is a biggest interval beyond which it can not be extended as a solution of IVP and
is called maximal interval of existence via Zorns lemma.
We start with a few denitions.
Denition 4.1 (Continuation, Saturated solutions) Let the function u dened on an inter-
val I
0
be a solution of IVP (3.1). Then
1. The solution u is called continuable at the right if there exists a solution of IVP w dened
on interval J
0
satisfying sup I
0
sup J
0
and the equality u(x) = w(x) holds for x I
0
J
0
.
Any such w is called an extension of u. Further if sup I
0
< sup J
0
, then w is called a non-
trivial extension of u. The solution u is called saturated at the right if there is no non-trivial
right-extension.
2. The solution u is called continuable at the left if there exists a solution of IVP z dened on
interval K
0
satisfying inf K
0
< inf I
0
and the equality u(x) = z(x) holds for x I
0
K
0
. The
solution u is called saturated at the left if it is not continuable at the left.
3. The solution u is called global at the right if I
0
x I : x x
0
. Similarly, u is called
global at the left if I
0
x I : x x
0
.
Remark 4.2 If the solution u is continuable at the right, then by concactenating the solutions u
and w we obtain a solution of IVP on a bigger interval (inf I
0
, sup J
0
), where w dened on J
0
is
a right extension of u. A similar statement holds if u is continuable at the left.
Let us dene the notion of a right (left) solution to the IVP (3.1).
Denition 4.3 A function u dened on an interval [x
0
, b) (respectively, on (a, x
0
]) is said to be
a right solution (respectively, a left solution) if u is a solution of ODE y

= f (x, y) on (x
0
, b)
(respectively, on (a, x
0
)) and u(x
0
) = y
0
.
S. Sivaji Ganesh ODE@AFS
20 4.1. CONTINUATION
Note that if u is a solution of IVP (3.1) dened on an interval (a, b), then u restricted to [x
0
, b)
(respectively, to (a, x
0
]) is a right solution (respectively, a left solution) to IVP.
For right and left solutions, the notions of continuation and saturated solutions become
Denition 4.4 (Continuation, Saturated solutions) Let the function u dened on an inter-
val [x
0
, b) be a right solution of IVP and let v dened on an interval (a, x
0
] be a left solution of
IVP (3.1). Then
1. The solution u is called continuable at the right if there exists a right solution of IVP w
dened on interval [x
0
, d) satisfying b < d and the equality u(x) = w(x) holds for x [x
0
, b).
Any such w is called a right extension of u. The right solution u is called saturated at the
right if it is not continuable at the right.
2. The solution v is called continuable at the left if there exists a solution of IVP z dened
on interval (c, x
0
] satisfying c < a and the equality u(x) = z(x) holds for x (a, x
0
]. The
solution u is called saturated at the left if it is not continuable at the left.
3. The solution u is called global at the right if [x
0
, b) = x I : x x
0
. Similarly, u is
called global at the left if (a, x
0
] = x I : x x
0
.
The rest of the discussion in this section is devoted to analysing continuability at the right, saturated at the right as
the analysis for the corresponding notions at the left is similar. We drop sufxing at the right from now on to save
space to notions of continuability and saturation of a solution.
4.1.1 Characterisation of continuable solutions
Lemma 4.5 Assume Hypothesis (H
IVPS
). Let u : I
0
R
n
be a right solution of IVP (3.1) dened
on the interval [x
0
, d). Then the following statements are equivalent.
(1) The solution u is continuable.
(2) (i) d < sup I and there exists
(ii) y

= lim
xd

y(x) and y

.
(3) The graph of u i.e.,
graphu =
_
(x, u(x)) : x [x
0
, d)
_
(4.1)
is contained in a compact subset of I .
Proof :
We prove (1) =(2) =(3) =(2) =(1). The implication (1) =(2) is obvious.
Proof of (2) =(3)
In view of (2), we can extend the function u to the interval [x
0
, d] and let us call this extended
function u. Note that the function x (x, u(x) is continuous on the interval [x
0
, d] and the
image of [x
0
, d] under this map is graph of u, denoted by graph u, is compact. But graphu
graph u I . Thus (3) is proved.
Proof of (3) =(2)
Assume that graphu is contained in a compact subset of I . As a consequence, owing to
continuity of the function f on I , there exists M > 0 such that |f (x, u(x))| < M for all
x [x
0
, d). Also, since I is an open interval, necessarily d < sup I. We will now prove that the
limit in (2)(ii) exists.
Since u is a solution of IVP (3.1), by Lemma 3.4, we have
u(x) = y
0
+
_
x
x
0
f (s, u(s)) ds x [x
0
, d). (4.2)
ODE@AFS S. Sivaji Ganesh
21
Thus for , [x
0
, d), we get
|u() u()|
_

|f (s, u(s))| ds M[ [. (4.3)

Thus u satises the hypothesis of Cauchy test on the existence of nite limit at d. Indeed, the
inequality (4.3) says that u is uniformly continuous on [x
0
, d) and hence limit of u(x) is nite as
x d. This follows from a property of uniformly continuous functions, namely they map Cauchy
sequences to Cauchy sequences. Let us denote the limit by y

. In principle being a limit, y

.
To complete the proof we need to show that y

. This is a consequence of the hypothesis that

graph of u is contained in a compact subset of I and the fact that I and are open sets.
Proof of (2) =(1)
As we shall see, the implication (2) =(1) is a consequence of existence theorem for IVP (Theo-
rem 3.6) and concactenation lemma (Lemma 3.5).
Let w be a solution to IVP corresponding to the initial data (d, y

) I dened on an interval
(e, f) containing the point d. Let w[
[d, f)
be the restriction of w to the interval [d, f) I. Let u
be dened as the continuous extension of u to the interval (c, d]) which makes sense due to the
existence of the limit in (2)(ii). Concactenating u and w[
[d, f)
yields a solution of the original IVP
(3.1) that is dened on the interval (c, f) and d < f.
Remark 4.6 The important message of the above result is that a solution can be extended to a
bigger interval provided the solution curve remains well within the domain I i.e., its right
end-point lies in .
4.1.2 Existence and Classication of saturated solutions
The following result is concerning the existence of saturted solutions for an IVP. Once again we
study saturated at the right and corresponding results for saturated at the left can be obtained
by similar arguments. Thus for this discussion we always consider a soluton as dened on interval
of the form [x
0
, d)
Theorem 4.7 (Existence of saturated solutions) If u dened on an interval [x
0
, d) is a right
solution of IVP (3.1), then either u is saturated, or u can be continued up to a saturated one.
Proof :
If u is saturated, then there is nothing to prove. Therefore, we assume u is not saturated.
By denition of saturatedness of a solution, u is continuable. Thus the set o, dened below, is
non-empty.
o = Set of all solutions of IVP (3.1) which extend u. (4.4)
We dene a relation _ on the set o, called partial order as follows. For w, z o dened on
intervals [x
0
, d
w
) and [x
0
, d
z
) respectively, we say that w _ z if z is a continuation of w.
Roughly speaking, if we take the largest (w.r.t. order _) element of o then by it can not be further
continued. To implement this idea, we need to apply Zorns lemma. Zorns lemma is equivalent to
axiom of choice (see the book on Topology by JL Kelley for more) and helps in asserting existence
of maximal elements provided the totally ordered subsets of o have an upper bound (upper
bound for a subset T o is an element h o such that w _ h for all w o).
Exercise 4.8 Show that the relation _ denes a partial order on the set o. Prove that each totally
ordered subset of o has an upper bound.
By Zorns lemma, there exists a maximal element q in o. Note that this maximal solution is
saturated in view of the denition of _ and maximality q.
S. Sivaji Ganesh ODE@AFS
22 4.1. CONTINUATION
Remark 4.9 Under the hypothesis of previous theorem, if a solution u of IVP (3.1) is continuable,
then there may be more than one saturated solution extending u. This possibility is due to non-
uniqueness of solutions to IVP (3.1). The following exercise is concerned with this phenomenon.
Further note that if solution u to IVP (3.1) is unique, then there will be a unique saturated solution
extending it.
Exercise 4.10 Let f : R R R be dened by f(x, y) = 3 y
2/3
. Show that the solution y :
[1, 0] R dened by y(x) = 0 for all x [1, 0] of IVP satisfying the initial condition y(1) = 0
has at least two saturated solutions extending it.
The longevity of a solution (the extent to which a solution can be extended) is independent of
the smoothness of vector eld, where as existence and uniqueness of solutions are guaranteed for
smooth vector elds.
Theorem 4.11 (Classication of saturated solutions) Let u be a saturated right solution of
IVP (3.1), and its domain of denition be the interval [x
0
, d). Then one of the following alterna-
tives holds.
(1) The function u is unbounded on the interval [x
0
, d).
(2) The function u is bounded on the interval [x
0
, d), and u is global i.e., d = sup I.
(3) The function u is bounded on the interval [x
0
, d), and u is not global i.e., d < sup I
and each limit point of u as x d

lies on the boundary of .

Proof :
If (1) is not true, then denitely (2) or (3) will hold. Therefore we assume that both (1) and
(2) do not hold. Thus we assume that u is bounded on the interval [x
0
, d) and d < sup I. We need
to show that each limit point of u as x d

lies on the boundary of .

Our proof is by method of contradiction. We assume that there exists a limit point u

of u as
x d

in . We are going to prove that lim

xd

u(x) exists. Note that, once the limit exists it

must be equal to u

which is one of its limit points. Now applying Lemma 4.5, we infer that the
solution u is continuable and thus contradicting the hypothesis that u is a saturated solution.
Thus it remains to prove that lim
xd

u(x) = u

i.e., |u(x) u

| can be made arbitrarily small for

x near x = d.
Since is an open set and u

, there exists r > 0 such that B[u

, r] . As a consequence,
B[u

, ] for every < r. Thus on the rectangle R I dened by

R = [x
0
, d] y : |y u

| r, (4.5)
|f (x, y)| M for some M > 0 since R is a compact set and f is continuous.
Since u

is a limit point of u as x d

, there exists a sequence (x

m
) in [x
0
, d) such that x
m
d
and u(x
m
) u

. As a consequence of denition of limit, we can nd a k N such that

[x
k
d[ < min
_

2M
,

2
_
and |u(x
k
) u

| < min
_

2M
,

2
_
. (4.6)
Claim: (x, u(x)) : x [x
k
, d) I B[u

, ].
PROOF OF CLAIM: If the inclusion in the claim were false, then there would exist a point on the
graph of u (on the interval [x
k
, d)) lying outside the set I B[u

, ]. Owing to the continuity of

u, the graph must meet the boundary of B[u

, ]. Let x

> x
k
be the rst instance at which the
trajectory touches the boundary of B[u

, ]. That is, = |u(x

) u

| and |u(x) u

| < for
x
k
x < x

. Thus
= |u(x

) u

| |u(x

) u(x
k
)| +|u(x
k
) u

| <
_
x

x
k
|f (s, u(s))| ds +

2
(4.7)
< M(x

x
k
) +

2
< M(d x
k
) +

2
< (4.8)
ODE@AFS S. Sivaji Ganesh
23
This contradiction nishes the proof of Claim.
Therefore, limit of u(x) as x d exists. As noted at the beginning of this proof, it follows that
u is continuable. This nishes the proof of the theorem.
Theorem 4.12 (Classication of saturated solutions) Let f : I R
n
be continuous on I
and assume that it maps bounded subsets in I into bounded subsets in R
n
. Let u : [x
0
, d) R
n
be a saturated right solution of IVP. Then one of the following alternatives holds.
(1)

The function u is unbounded on the interval [x

0
, d). If d < there exists lim
xd

|u(x)| =
.
(2) The function u is bounded on the interval [x
0
, d), and u is global i.e., d = sup I.
(3)

The function u is bounded on the interval [x

0
, d), and u is not global i.e., d < sup I
and limit of u as x d

exists and lies on the boundary of .

Corollary 4.13 Let f : R R
n
R
n
be continuous. Let u : [x
0
, d) R
n
be a saturated right
solution of IVP. Use the previous exercise and conclude that one of the following two alternatives
holds.
(1) The function u is global i.e., d = .
(2) The function u is not global i.e., d < and lim
xd

|u(x)| = . This phenomenon

is often referred to as u blows up in nite time.
4.2 Global Existence theorem
In this section we give some sucient conditions under which every local solution of an IVP is
global. One of them is the growth of f w.r.t. y. If the growth is at most linear, then we have a
global solution.
Theorem 4.14 Let f : I R
n
R
n
be continuous. Assume that there exist two continuous
functions h, k : I R
+
(non-negative real-valued) such that
|f (x, y)| k(x)|y| + h(x), (x, y) I R
n
. (4.9)
Then for every initial data (x
0
, y
0
) I R
n
, IVP has at least one global solution.
Exercise 4.15 Using Global existence theorem, prove that any IVP corresponding to a rst order
linear system has a unique global solution.
4.3 Continuous dependence
In situations where a physical process is described (modelled) by an initial value problem for a
system of ODEs, then it is desirable that any errors made in the measurement of either initial
data or the vector eld, do not inuence the solution very much. In mathematical terms, this is
known as continuous dependence of solution of an IVP, on the data present in the problem. In
fact, the following result asserts that solution to an IVP has not only continuous dependence on
initial data but also on the vector eld f .
Exercise 4.16 Try to carefully formulate a mathematical statement on continuous dependence of
solution of an IVP, on initial conditions and vector elds.
S. Sivaji Ganesh ODE@AFS
24 4.4. WELL-POSED PROBLEMS
An honest eort to answer the above exercise would make us understand the diculty in formu-
lating such a statement. In fact, many introductory books on ODEs do not address this subtle
issue, and rather give a weak version of it without warning the reader about the principal di-
culties. See, however, the books of Wolfgang [33], Piccinini et. al. [23]. We now state a result on
continuous dependence, following Wolfgang [33].
Theorem 4.17 (Continuous dependece) Let R
n
be a domain and I R be an interval
containing the point x
0
. Let J be a closed and bounded subinterval of I, such that x
0
J. Let
f : I R
n
be a continuous function. Let y(x; x
0
, y
0
) be a solution on J of the initial value
problem
y

= f (x, y) , y(x
0
) = y
0
. (4.10)
Let S

denote the -neighbourhood of graph of y, i.e.,

S

:=
_
(x, y) : |y y(x; x
0
, y
0
)| , x J
_
. (4.11)
Suppose that there exists an > 0 such that f satises Lipschitz condition w.r.t. variable y on
S

. Then the solution y(x; x

0
, y
0
) depends continuously on the initial values and on the vector
eld f .
That is: Given > 0, there exists a > 0 such that if g is continuous on S

and the inequalities

|g (x, y) f (x, y) | on S

, | y
0
| (4.12)
are satised, then every solution z(x; x
0
, ) of the IVP
z

= g (x, z) , z(x
0
) = . (4.13)
exists on all of J, and satises the inequality
|z(x; x
0
, ) y(x; x
0
, y
0
)| , x J. (4.14)
Remark 4.18 (i) In words, the above theorem says : Any solution corresponding to an IVP
where the vector eld g near a Lipschitz continuous vector eld f and initial data (x
0
, )
near-by (x
0
, y
0
), stays near the unique solution of IVP with vector eld f and initial data
(x
0
, y
0
).
(ii) Note that, under the hypothesis of the theorem, any IVP with a vector eld g which is only
continuous, also has a solution dened on J.
(iii) Note that the above theorem does not answer the third question we posed at the beginning of
this chapter. The above theorem does not say anything about the function in (3.3).
4.4 Well-posed problems
A mathematical problem is said to be well-posed (or, properly posed) if it has the EUC property.
(1) Existence: The problem should have at least one solution.
(2) Uniqueness: The problem has at most one solution.
(3) Continuous dependence: The solution depends continuously on the data that are
present in the problem.
Theorem 4.19 Initial value problem for an ODE y

= f (x, y), where f is Lipschitz continuuos

on a rectangle containining the initial data (x, y
0
), is well-posed.
Example 4.20 Solving Ax = b is not well-posed. Think why such a statement could be true.
ODE@AFS S. Sivaji Ganesh
25
Lecture-5
22 Linear systems & Fundamental pairs of solutions
5.1 The Linear System
We consider the two-dimensional, linear homogeneous rst order system of ODE
_
_
y

1
y

2
_
_
=
_
_
a(x) b(x)
c(x) d(x)
_
_
_
_
y
1
y
2
_
_
(5.1)
The above system (5.1), with obvious notations, can be written as
y

= A(x)y (5.2)
An IVP is given by
y

= A(x)y, y(x
0
) =
_
y
10
y
20
_
(5.3)
1. Linear combinations of solutions of ODE (5.1) are also solutions of ODE (5.1).
2. The only solution of IVP (5.3) satisfying y(x
0
) = 0 is the trivial solution.
3. Given two solutions w
1
and w
2
of system (5.1), can the solution y of IVP (5.3) be obtained
as a linear combination of w
1
and w
2
? Supposing that the answer to this question is positive,
we will nd out the properties that the two solutions w
1
and w
2
must satisfy. Therefore
assume that there exist C
1
, C
2
R such that
y(x) = C
1
w
1
(x) + C
2
w
2
(x) (5.4)
The function y solves the IVP if and only if
_
y
10
y
20
_
= y(x
0
) = C
1
w
1
(x
0
) + C
2
w
2
(x
0
) = C
1
_
w
11
(x
0
)
w
12
(x
0
)
_
+ C
2
_
w
21
(x
0
)
w
22
(x
0
)
_
(5.5)
that is,
_
y
10
y
20
_
=
_
w
11
(x
0
) w
21
(x
0
)
w
12
(x
0
) w
22
(x
0
)
__
C
1
C
2
_
(5.6)
The matrix equation (5.6) is solvable for arbitrary initial values
_
y
10
y
20
_
if and only if
det
_
w
11
(x
0
) w
21
(x
0
)
w
12
(x
0
) w
22
(x
0
)
_
is non-zero. This happens if and only if the columns
_
w
11
(x
0
)
w
12
(x
0
)
_
and
_
w
21
(x
0
)
w
22
(x
0
)
_
are linearly independent. In other words, the vectors w
1
(x
0
) and w
2
(x
0
)
are linearly independent.
4. Suppose we want to solve arbitrary initial value problem, that is, (x
0
, y
0
) is arbitrary.
This requires J(x; w
1
, w
2
) = det
_
w
11
(x) w
21
(x)
w
12
(x) w
22
(x)
_
to be non-zero for all x (a, b). It
seems that we are asking too much on solutions w
1
and w
2
. However, it turns out that if
det
_
w
11
(x) w
21
(x)
w
12
(x) w
22
(x)
_
is non-zero at some x, then it will be non-zero for all a < x < b.
Lemma 5.1 The wronskian J(x; w
1
, w
2
) satises the ODE
J

(x; w
1
, w
2
) = traceA(x) J(x; w
1
, w
2
), a < x < b. (5.7)
As a consequence, the function J(x; w
1
, w
2
) is either non-vanishing or identically zero on
a < x < b.
S. Sivaji Ganesh ODE@AFS
26 5.1. THE LINEAR SYSTEM
Proof :
From the denition of J(x; w
1
, w
2
), we get
J

(x; w
1
, w
2
) = (w
11
(x)w
22
(x) w
21
(x)w
12
(x))

(5.8)
= w

11
(x)w
22
(x) + w
11
w

22
(x) w

21
(x)w
12
(x) w
21
(x)w

12
(x) (5.9)
Substituting the values of w

11
, w

22
, w

21
, w

12
from the ODE, we get
J

(x; w
1
, w
2
) = (a(x) + d(x))J(x; w
1
, w
2
) (5.10)
= traceA(x) J(x; w
1
, w
2
) (5.11)
Its general solution is given by
J(x; w
1
, w
2
) = J(x
0
; w
1
, w
2
) exp
__
x
x
0
traceA(s) ds
_
(5.12)
This nishes the proof of lemma.
Denition 5.2 A pair of solutions w
1
and w
2
of system (5.1) is called a fundamental pair of
solutions if any solution y of IVP (5.3) can be obtained as a linear combination of w
1
and w
2
.
That is, there exist C
1
, C
2
R such that
y(x) = C
1
w
1
(x) + C
2
w
2
(x). (5.13)
We answer the following questions concerning fundamental pairs of solutions.
1. Does there exist at least one fundamental pair of solutions?
2. Can we describe all fundamental pairs?
Answers
1. Existence of a fundamental pair follows by taking two solutions w
1
and w
2
of ODE satisfying
IVP with initial data
_
w
11
(x
0
)
w
12
(x
0
)
_
=
_
1
0
_
and
_
w
21
(x
0
)
w
22
(x
0
)
_
=
_
0
1
_
respectively.
2. Let P =
_
p
11
p
12
p
21
p
22
_
be a non-singular matrix. Then p
11
w
1
+ p
21
w
2
, p
12
w
1
+ p
22
w
2
is
another fundamental pair. In fact this is the general form of a fundamental pair of solutions.
Special case of Second order linear equation
Let us now specialise to the second order homogeneous ODE (5.15) which is equivalent to the
system
y

+ p(x)y

+ q(x)y = r(x). (5.14)

y

+ p(x)y

+ q(x)y = 0, (5.15)
_
_
y

1
y

2
_
_
=
_
_
0 1
q(x) p(x)
_
_
_
_
y
1
y
2
_
_
(5.16)
Let us relate a fundamental pair of solutions of second order scalar equation with those of the
equivalent system.
1. Let w
1
, w
2
be a fundamental pair of solutions corresponding to ODE (5.15), then a funda-
mental pair of solutions for the system (5.16) is given by
z
1
(x) =
_
_
w
1
(x)
w

1
(x)
_
_
, z
2
(x) =
_
_
w
2
(x)
w

2
(x)
_
_
(5.17)
Verication of the above assertion is easy.
ODE@AFS S. Sivaji Ganesh
27
2. Similarly, given a fundamental pair of solutions for the system z
1
, z
2
one can produce a
fundamental pair of solutions for the scalar equation by simply taking the rst components
of the vector valued functions z
1
, z
2
.
5.2 Nonhomogeneous equation
In this section we discuss on methods of solving nonhomogeneous equation. We follow the classical
Variation of Parameters (VoP) method to nd solutions of the nonhomogeneous equation starting
from a general solution of the corresponding homogeneous equation. As already noted in the
discussion of rst order ODEs, VoP method involves computation of integrals that need lot of
hard work even for simpler right hand side functions r. At least for second order equation with
constant coecients, as a second method, we prescribe the method of undetermined coecients
that works only for a restricted class of second order linear ODE with constant coecients, and
when the function r in (5.15) is of a special type involving only polynomials, exponentials.
5.2.1 Variation of Parameters method - Revisited.
We are going to exclusively work with the equivalent rst order system of the second order non-
homogeneous ODE (5.14) given by
_
_
y

1
y

2
_
_
=
_
_
0 1
q(x) p(x)
_
_
_
_
y
1
y
2
_
_
+
_
_
0
r(x)
_
_
(5.18)
1. The starting point of VoP method is to have a general solution of homogeneous system.
Let z
1
(x) =
_
_
w
1
(x)
w

1
(x)
_
_
, z
2
(x) =
_
_
w
2
(x)
w

2
(x)
_
_
be a fundamental pair of solutions correspond-
ing to the equivalent system (5.16). Then a general solution y of homogeneous system is
given by
y(x) = C
1
_
_
w
1
(x)
w

1
(x)
_
_
+ C
2
_
_
w
2
(x)
w

2
(x)
_
_
(5.19)
Re-writing the above equality,
y(x) =
_
_
w
1
(x) w
2
(x)
w

1
(x) w

2
(x)
_
_
_
_
C
1
C
2
_
_
= (x)C(x) (5.20)
We assume that (x
0
) is identity matrix and there is no loss of generality in assuming this.
2. The second step is to allow
_
C
1
C
2
_
to be a function of x, and search for a particular solution
y
p
of the nonhomogeneous system in the form
y
p
(x) = (x)C(x) (5.21)
Substitute the ansatz for y
p
in (5.18) to get
y

p
(x) =

(x)C(x) + (x)C

(x) =
_
_
0 1
q(x) p(x)
_
_
y
p
(x) +
_
_
0
r(x)
_
_
(5.22)
It is easy to verify that the matrix valued function satises

(x) =
_
_
0 1
q(x) p(x)
_
_
(x) (5.23)
S. Sivaji Ganesh ODE@AFS
28 5.2. NONHOMOGENEOUS EQUATION
In view of (5.23), the equation (5.22) yields
C

(x) = ((x))
1
_
_
0
r(x)
_
_
(5.24)
Integrating the above equation from x
0
to x, we get
C(x) = C(x
0
) +
_
x
x
0
((s))
1
_
_
0
r(s)
_
_
ds (5.25)
Substituting the value of ((s))
1
in (5.25), we get
C(x) = C(x
0
) +
_
x
x
0
1
J(s; w
1
, w
2
)
_
_
w

2
(s) w
2
(s)
w

1
(s) w
1
(s)
_
_
_
_
0
r(s)
_
_
ds (5.26)
That is,
C(x) = C(x
0
) +
_
x
x
0
1
J(s; w
1
, w
2
)
_
_
w
2
(s)r(s)
w
1
(s)r(s)
_
_
ds (5.27)
Thus, a particular solution is given by
y
p
(x) = (x)C(x) = (x)
_
_
C(x
0
) +
_
x
x
0
1
J(s; w
1
, w
2
)
_
_
w
2
(s)r(s)
w
1
(s)r(s)
_
_
ds
_
_
(5.28)
We will take C(x
0
) = 0 since the term corresponding to C(x
0
) contributes only a solution
to the homogeneous system and we are looking for a particular solution. Thus we get a
solution of IVP for the nonhomogeneous system and is given by
y(x) =
_
_
w
1
(x) w
2
(x)
w

1
(x) w

2
(x)
_
_
_
_

_
_
+ (x)
_
_
_
x
x
0
1
J(s; w
1
, w
2
)
_
_
w
2
(s)r(s)
w
1
(s)r(s)
_
_
ds
_
_
(5.29)
We substitute the value of (x) in (5.29), and take only the rst component of the resulting
equation which will be a solution to the IVP for the corresponding second order equation
and we denote it by y. Thus we have the following formula for y.
y(x) = w
1
(x) + w
2
(x) w
1
(x)
_
x
x
0
w
2
(s)r(s)
J(s; w
1
, w
2
)
ds + w
2
(x)
_
x
x
0
w
1
(s)r(s)
J(s; w
1
, w
2
)
ds (5.30)
Putting it in a more appealing format,
y(x) = w
1
(x) + w
2
(x) +
_
x
x
0
w
2
(x)w
1
(s) w
1
(x)w
2
(s)r(s)
J(s; w
1
, w
2
)
ds (5.31)
We will come back to the above formula while dealing with what are called boundary value
problems.
ODE@AFS S. Sivaji Ganesh
29
5.3 Solving linear planar systems with constant coecients
Consider the system of ODE
_
x

_
=
_
a b
c d
__
x
y
_
=: A
_
x
y
_
. (5.32)
5.3.1 Fundamental matrix
Denition 5.3 (Fundamental matrix) A matrix valued function whose columns are solu-
tions of the system of ODE (5.32) is called a solution matrix. A solution matrix is called a
fundamental matrix if the columns of form a fundamental pair of solutions for the system (5.32).
A fundamental matrix is called the standard fundamental matrix if (0) is the identity matrix.
Remark 5.4 Since the columns of a solution matrix are solutions of (5.32), the matrix valued
function satises the system of ODE

= A. (5.33)
Exercise 5.5 A solution matrix is a fundamental matrix if and only if its determinant is not
zero.
Exercise 5.6 Prove that if is a fundamental matrix then C is also a fundamental matrix for
every constant invertible matrix C. Prove that all fundamental matrices occur this way.
Computation of fundamental matrix
By denition of a fundamental matrix, nding a fundamental pair of solutions to system of ODE
(5.32) is equivalent to nding a fundamental matrix. In view of Exercise 5.6, fundamental matrix
is not unique but the standard fundamental matrix is unique.
(1) Observe that e
t
_
a
b
_
is a non-trivial solution of (5.32) if and only if
_
a
b
_
,=
_
0
0
_
, A
_
a
b
_
=
_
a
b
_
. (5.34)
That is, is an eigenvalue and
_
a
b
_
is an eigenvector corresponding to .
(2) Question Is it possible to nd a fundamental pair, both of which are of the form
e
t
_
a
b
_
?
Answer Supposing that
1
(t) = e

1
t
_
a
b
_
and
2
(t) = e

2
t
_
c
d
_
are two solutions of
(5.32),
1
,
2
form a fundamental pair if and only if

a c
b d

,= 0, (5.35)
since the above determinant is the Wronskian of
1
,
2
at t = 0.
That is, the matrix A should have two linearly independent eigenvectors. Note that
this is equivalent to saying that A is diagonalisable.
S. Sivaji Ganesh ODE@AFS
30 5.3. SOLVING LINEAR PLANAR SYSTEMS WITH CONSTANT COEFFICIENTS
(3) Question What if the matrix A does not have two linearly independent eigenvectors?
This can happen when A has only one eigenvalue of multiplicity two. Inspired by a
similar situation in the context of constant coecient second order linear ODE, we are
tempted to try
1
(t) = e

1
t
_
a
b
_
and
2
(t) = te

1
t
_
a
b
_
as a fundamental pair. But
note that
1
,
2
does not form a fundamental pair since Wronskian at t = 0 will be
zero, also note that
2
is not even a solution of the linear system (5.32). Nevertheless,
we can nd a solution having the form of
1
. Therefore, we try a variant of above
suggestion to nd another solution that together
1
constitutes a fundamental pair.
Let

2
(t) = te

1
t
_
a
b
_
+ e

1
t
_
c
d
_
(5.36)
Then
2
(t) solves the system (5.32) if and only if
(A
1
I)
_
c
d
_
=
_
a
b
_
. (5.37)
One can easily verify that
_
a
b
_
,
_
c
d
_
are linearly independent. Thus,
1
,
2
dened
by

1
(t) = e

1
t
_
a
b
_
,
2
(t) = te

1
t
_
a
b
_
+ e

1
t
_
c
d
_
(5.38)
is a fundamental pair, where
_
a
b
_
,
_
c
d
_
are related by the equation (5.37).
(4) In case the matrix A does not have real eigenvalues, then eigenvalues are complex
conjugates of each other. In this case,
(, v) is an eigen pair if and only if (, v) is also an eigen pair for A. (5.39)
Denoting = r + iq (note q ,= 0), v =
_
+ i
+ i
_
, dene

1
(t) = e
rt
_
cos qt sin qt
cos qt sin qt
_
,
2
(t) = e
rt
_
sin qt + cos qt
sin qt + cos qt
_
. (5.40)
Verifying that the pair of functions dened above constitute a real-valued fundamental
pair of solutions is left as an exercise.
5.3.2 Matrix exponentials
Inspired by the formula for solutions to linear scalar equation y

= ay, given by y(t) = ce

at
, where
y(0) = c, we would like to say that the system (5.32) has solution given by
_
x(t)
y(t)
_
= exp(tA)
_
a
b
_
, where
_
a
b
_
=
_
x(0)
y(0)
_
. (5.41)
First of all, we must give a meaning for the symbol exp(A) as a matrix, which is called the
exponental of matrix B. Then we must verify that the proposed formula for solution, in (5.41), is
indeed a solution of (5.32). We do this next.
Denition 5.7 (Exponential of a matrix) If A is a 2 2 matrix, then exponential of A (de-
noted by e
A
, or exp(A), is dened by
e
A
= I +

k=1
A
k
k!
(5.42)
ODE@AFS S. Sivaji Ganesh
31
We record below, without proof, some properties of matrix exponentials.
Lemma 5.8 Let A be a 2 2 matrix.
(1) The series in (5.42) converges.
(2) e
0
= I.
(3) (e
A
)
1
= e
A
(4) exp(A + B) = exp(A) exp(B) if the matrices A and B satisfy AB = BA.
(5) If J = P
1
AP, then exp(J) = P
1
exp(A)P.
(6)
d
dt
e
tA
= Ae
tA
.
Theorem 5.9 If A is a 22 matrix, then (t) = e
tA
is a fundamental matrix of the system (5.32).
If (x(t), y(t)) is a solution of the system (5.32) with (x(t
0
), y(t
0
)) = (x
0
, y
0
), then (x(t), y(t)) is
given by
_
x(t)
y(t)
_
= e
(tt
0
)A
_
x
0
y
0
_
. (5.43)
Remark 5.10 Thanks to the above theorem, we do not need to struggle to nd a fundamental
matrix as we did earlier. We need to to take just the exponential of the matrix tA and that would
give very easily a fundamental matrix. But it is not as simple as it seems to be. In fact, summing
up the series for exponential of a matrix is not easy at all, even for relatively simpler matrices.
To realise this, solve the exercise following this remark.
Exercise 5.11 Find exponential matrix for
A
1
=
_
2 0
0 1
_
, A
2
=
_
1 1
0 1
_
, A
3
=
_
1 2
2 3
_
. (5.44)
Remark 5.12 After solving the above exercise, one would realise that it is easier to nd funda-
mental matrices by nding fundamental pairs of solutions instead of summing up a series! There
is an alternate method to calculate exponential matrix for tA, via fundamental pairs, by observing
that exponential matrix is nothing but the standard fundamental matrix. So, nd a fundamental
matrix , then exponential matrix e
tA
is given by
e
tA
= [(0)]
1
(t). (5.45)
S. Sivaji Ganesh ODE@AFS
32 5.3. SOLVING LINEAR PLANAR SYSTEMS WITH CONSTANT COEFFICIENTS
Topics for further study
So far we studied the basic theory of ODE, and I followed mainly two books [24, 32].
Topological properties of trajectories of solutions (the values that solutions of ODE take in the
space , called Phase space) are very interesting. Another object associated to ODE is nding
the subsets of which are invariant under ows dened by solutions of ODE. These studies go
under the name of Dynamical systems.
Given that ODEs arise as models of physical phenomena, their understanding helps in under-
standing the systems that ODE model.
Another class of important ODEs are those with periodic coecients.
One can look up books [1, 6, 14, 17, 20, 22] for further study.
ODE@AFS S. Sivaji Ganesh
Bibliography
[1] V.I. Arnold, Ordinary dierential equations, translated by Silverman, (Printice-Hall of India,
1998).
[2] V.I. Arnold, Geometrical methods in the theory of ordinary dierential equations, (Springer,
1988).
[3] G. Birkho and G.-C. Rota, Ordinary dierential equations, (John wiley, 1978).
[4] W.E. Boyce and R.C. Diprima, Elementary dierential equations and boundary value prob-
lems, (Johy wiley, 2003).
[5] F. Brauer and J.A. Nohel, The qualitative theory of Ordinary dierential equations: an in-
troduction, (Dover, 1989).
[6] M. Braun, Dierential equations and their applications : an introduction to applied mathe-
matics, (Springer-Verlag, 1993).
[7] L. Cesari, Asymptotic behavior and stability problems in ordinary dierential equations, 2nd
ed., (Academic Press, 1963).
[8] E.A. Coddington, Introduction to ordinary dierential equations, (Prentice-Hall of India,
1974).
[9] E.A. Coddington and N. Levinson, Theory of ordinary dierential equations, (McGraw-Hill,
1955).
[10] J. Cronin, Dierential equations: Introduction and qualitative theory, (Marcel Dekker, 1980).
[11] S.G. Deo and V. Raghavendra, Ordinary dierential equations and stability theory, (Tata
McGraw-Hill, 1980).
[12] M.S.P. Eastham, Theory of ordinary dierential equations, (Van Nostrand Reinhold, 1970).
[13] N.P. Erugin , Linear systems of ordinary dierential equations with periodic and quasi-periodic
coecients translation edited by Richard Bellman, (Academic Press, 1966).
[14] K.O. Friedrichs, Lecture notes on advanced ordinary dierential equations, (Gordon & Breach,
1965).
[15] M. Golomb and M. Shanks, Elements of ordinary dierential equations, (McGraw-Hill, 1965).
[16] P. Hartman, Ordinary dierential equations, (Wiley, 1964).
33
34 Bibliography
[17] M.W. Hirsh, S. Smale and R.L. Devaney, Dierential equations, dynamical systems & Chaos,
(Academic press, 2004).
[18] W. Hurewicz, Lectures on ordinary dierential equations, (Cambridge : MIT Press, 1958).
[19] E.L. Ince, Ordinary dierential equations, (Dover, 1956).
[20] D.W. Jordan and P. Smith, Nonlinear ordinary dierential equations : an introduction to
dynamical systems, (Oxford, 1987).
[21] T. Myint-U, Ordinary dierential equations, (North-Holland, 1978).
[22] L. Perko, Dierential equations and dynamical systems (Springer-Verlag, 1991).
[23] L.C. Piccinini, G. Stampacchia and G. Vidossich, Ordinary dierential equations in R
N
:
problems and methods, (Springer-Verlag, 1984).
[24] L.S. Pontriagin , Ordinary dierential equations, (Addison-Wesley, 1962).
[25] A.L. Rabenstein, Introduction to ordinary dierential equations, (Academic Press, 1966).
[26] E.D. Rainville and P.E. Bedient, A short course in dierential equations, (IBH, 1973; Macmil-
lan, 1969).
[27] M.R.M. Rao, Ordinary dierential equations: theory and applications, (Aliated East-West,
1980).
[28] W.T. Reid, Sturmian theory for ordinary dierential equations, (Springer-Verlag, 1980).
[29] S.L. Ross, Introduction to ordinary dierential equations, (Waltham : Blaisdell Pub., 1966).
[30] D.A. Sanchez, Ordinary dierential equations and stability theory, (Dover, 1979).
[31] D.A. Sanchez, Ordinary dierential equations: A brief eclectic tour, (Mathematical associa-
tion of America, 2002).
[32] I.I. Vrabie, Dierential equations, (World scientic, 2004).
[33] W.Wolfgang, Ordinary dierential equations, (Springer-Verlag, 1998).
ODE@AFS S. Sivaji Ganesh