Ordinary Differential Equations

MTH-114M

T. Muthukumar & Akash Anand

(In memory of Prof. Arbind Kumar Lal)

June 16, 2023

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 1 / 126
1 Course Plan
2 References
3 15–19 May 2023
Lecture One
Lecture Two
Lecture Three
4 29 May – 02 June 2023
Lecture Four
Lecture Five
Lecture Six
5 5–9 June 2023
Lecture Seven
Lecture Eight
Lecture Nine
6 12–16 June 2023
Lecture Ten
Lecture Eleven
Lecture Twelve
T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 2 / 126
Course Plan

Introduction to Ordinary Differential Equations (ODEs):

definition, order and linearity of ODE, types of solutions, boundary
and initial value problem
First order ODE:
I Picard’s existence and uniqueness results (without proof)
I Solution of some special classes of first order ODEs:
F separable equations,
F homogeneous and exact ODE,
F integrating factors,
F first order linear ODE,
F Bernoulli equation.
F coefficients with one degree polynomial.
I Approximate methods:
F isocline method: orthogonal and oblique trajectories,
F iterative method: Picards’ iterative process,
F numerical method: Euler method, improved Euler method.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 3 / 126
Course Plan Contd...

Higher order linear ODE:

I Linear homogeneous equations: linear independence of solutions,
Wronskian, reduction order technique,
I Linear homogeneous equations with constant coefficients: characteristic
equations with real distinct roots, complex roots, repeated roots.
I Linear inhomogeneous equations with constant coefficients: method of
undetermined coefficients,
I Linear homogeneous equations with variable coefficients: variation of
parameters, Euler-Cauchy equation.
Series solutions of ODE: ordinary points, power series solution,
regular singular points, Frobenius method.
Laplace transform: definition, existence, first shifting theorem,
second shifting theorem, differentiation and integration of transforms,
convolution and applications.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 4 / 126
Reference text for the Course

Ordinary Differential Equations by Morris Tenenbaum and Harry

Pollard.
Elementary Differential Equations and Boundary Value Problems by
William E. Boyce and Richard C. DiPrima.
Introduction to Ordinary Differential Equations by Shepley L. Ross.
Differential Equations and Linear Algebra by Gilbert Strang.
Differential Equations with Applications and Historical Notes by
George F. Simmons.
The above references are indicative, for further reading and not exhaustive.
The contents of the course is not a linear adaptation of above references
but intersect.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 5 / 126
How Equations Arise?

The process of understanding natural phenomena may be viewed in three

stages:
(i) Modelling the phenomenon as a mathematical equation (algebraic,
differential or integral equation) using physical laws such as Newton’s
law, momentum, conservation laws, balancing forces etc.
(ii) Solving the equation! This leads to the question of what constitutes
as a solution to the equation?
(iii) Properties of the solution, especially in situations when exact solution
is not within our reach.
In this course, we are mostly interested in differential equations in
dimension one, i.e. one independent variable!

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 6 / 126
Algebraic Equations to Differential Equations

(i) Algebraic equations are relations between (one or more) unknown

quantities/numbers and known quantities.
(ii) The Linear algebra (earlier part of the course) may also be applied to
solve system of linear (algebraic or differential) equations.
(iii) Differential equations are relations between (one or more) unknown
functions along with its derivatives and known quantities.
(iv) Modelling into differential equations was facilitated by the invention
of differential calculus in sixteenth century.
(v) The first ordinary differential equation (ODE) probably comes from
the Newton’s second law: How much a body of mass m will be
displaced when acted up on by a force F !
(vi) The first partial differential equation (PDE) was written down to
study the vibrating strings, now known as the (one space dimension)
wave equation.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 7 / 126
Simplest Differential Equation

If I ⊂ R is an open interval, not necessarily of finite length, and

f : I → R is a given continuous function then find y : I → R that
satisfies the ODE y 0 (x) = f (x) for all x ∈ I .
We are given the data I and f and the relation of the unknown y
with the given data.
The fundamental theorem of calculus
Rx hints that
R x we should integrate
both sides of the ODE to get x0 y 0 (s) ds = x0 f (s) ds.
Thus, for all x ∈ I ,
Z x
y (x) = y (x0 ) + f (s) ds
x0

is the infinitely many solutions/curves of the ODE, each passing

through the initial value (x0 , y (x0 )).
In fact, every solution is of the above form.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 8 / 126
Ordinary Differential Equation

Definition
Let I be an open interval of R. A k-th order ordinary differential equation
of an unknown function y : I → R is of the form
 
F y (k) , y (k−1) , . . . y 0 (x), y (x), x = 0, (3.1)

for each x ∈ I , where F : Rk+1 × I → R is a given map such that F

depends on the k-th order derivative y and is independent of (k + j)-th
order derivative of y for all j ∈ N.

The word ordinary refers to the situation of exactly one independent

variable x. The unknown dependent variable y can be more than one,
leading to a system of ODE.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 9 / 126
Example

Example
The DE y 0 + y + x = 0 comes from the F (u, v , w ) = u + v + w .
The DE y 00 + xy (y 0 )2 = 0 comes from the F (t, u, v , w ) = t + wvu 2 .
The DE y (4) + 5y 00 + 3y = sin x comes from the
F (r , s, t, u, v , w ) = r + 5t + 3v − sin w .
∂u ∂u
∂x+ ∂y = 0 is a partial diferential equation of a two variable
unknown function u(x, y ).
∂2u ∂ u 2 ∂ u 2
+ ∂y
∂x 2 2 + ∂z 2 = 0 is a partial diferential equation of a three

variable unknown function u(x, y , z).

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 10 / 126
Explicit Solution of ODE

Definition
We say u : I → R is an explicit solution to the ODE (3.1) on I , if u (j) (x)
exists for all j explicitly present in (3.1), for all x ∈ I , and u satisfies the
equation (3.1) in I .

Example
The function y : R → R defined as y (x) := 2 sin x + 3 cos x is an explicit
solution of the ODE y 00 + y = 0 in R.

Roughly speaking, an explicit solution curve is a graph of the solution in

the plane!

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 11 / 126
Implicit Solution of ODE

Definition
We say a relation v (x, y ) = 0 is an implicit solution to the ODE (3.1) on
I , if it defines at least one real function u : I → R of x variable which is a
explicit solution of (3.1) on I .

Example
The relation v (x, y ) := x 2 + y 2 − 25 = 0 is an implicit solution of the
ODE yy 0 + x = 0 in√(−5, 5). The given relation √ v defines two real valued
functions u1 (x) := 25 − x 2 and u2 (x) := − 25 − x 2 in (−5, 5) and
both are explicit solutions of the given ODE.

Roughly speaking, an implicit solution curve is not a graph of the solution

but some part (locally) of the solution curve can be expressed as the graph
of a function.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 12 / 126
Formal Solutions

In the previous example, could we have just differentiated the relation

x 2 + y 2 − 25 = 0 with respect to x to obtain 2x + 2yy 0 = 0 and
concluded that it is implicit solution?
The answer is a ‘no’ !
For instance, consider x 2 + y 2 + 25 = 0.
It formally satisfies the ODE on implicit differentiation
√
but is not an implicit solution because y (x) := ± −25 − x 2 cannot
be expressed as (even locally) graph of a function.
Complex valued functions can be solutions to ODE with real
coefficients (as above) or (y 0 )2 + 1 = 0 has the family of solutions
y (x) = ıx + c.
Henceforth, explicit, implicit and formal solution will be simply
referred to as ‘solution’.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 13 / 126
Do solutions always exist?

Do all ODE have solutions?

If yes, how many?

Example
The ODE |y 0 | + |y | + 1 = 0 admits no real (or complex) valued function y
as a solution because sum of positive quantities cannot be a negative
number.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 14 / 126
Family of Solutions

Example
Observe that on differentiation of the one parameter family of solutions
yc (x) := x 2 + c, indexed by c ∈ R, we obtain the first order ODE y 0 = 2x.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 15 / 126
Solutions Outside Family of Solutions

In the earlier example, the ODE had a unique family of solutions! But this
may not be the case always.
Example
Consider the first order ODE (y 0 )2 − 4y = 0.
The one parameter family yc (x) := (x + c)2 are solution to the ODE.
In addition, y ≡ 0 is also a solution, not included in the above family
of solutions.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 16 / 126
Initial/Boundary Conditions

In practice, to choose a desirable solution from the family of solutions,

some additional conditions are imposed along with the ODE.
For instance, in the earlier example y 0 = 2x, on integration from x0
and x, we get the family y (x) = x 2 − x02 + y (x0 ). Thus, the additional
information of (x0 , y (x0 )) will help fix a solution from the family.

(x0 , y0 )

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 17 / 126
Initial Value Problem

Definition
An initial value problem (IVP) corresponding to a k-th order ODE is given
as
F (y (k) , . . . , y , x) = 0 on I


y (j) (x0 ) = yj for all 0 ≤ j ≤ k − 1

where k initial conditions are prescribed at some x0 ∈ I¯ and I¯ is the union
of I with its end-points.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 18 / 126
Examples for IVP

Example
The first order IVP
y 0 = 2x

in R
y (1) = 4.

Example
The second order IVP
 00
 y +y =0 in R
y (1) = 3
y 0 (1) = −4.


T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 19 / 126
II order Boundary Value Problem (BVP)

The additional conditions could be either an initial condition or boundary

condition or their mixture.
A first order boundary value problem is same as an initial value problem!
Definition
A boundary value problem corresponding to a second order ODE
F (y 00 , y 0 , y , x) = 0 on I is prescribed with one of the following additional
conditions: For any x0 , x1 ∈ I¯
(i) y (x0 ) = y0 and y (x1 ) = y1 .
(ii) y 0 (x0 ) = y0 and y 0 (x1 ) = y1 .
(iii) c1 y (x0 ) + c2 y 0 (x0 ) = y0 and d1 y (x1 ) + d2 y 0 (x1 ) = y1 for given
ci , di ∈ R and i = 1, 2.
(iv) y (x0 ) = y (x1 ) and y 0 (x0 ) = y 0 (x1 ).

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 20 / 126
Examples for BVP

Example
The second order BVP
 00
 y + y = 0 in R
y (0) = 1
y ( π2 ) = 5.


Example
The second order BVP
 00
 y + y = 0 in R
y (0) = 1 (3.2)
y (π) = 5.


T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 21 / 126
First Order ODE
Let us begin solving some ODEs and let’s start with first order ODEs!
A general first order ODE is of the form F (y 0 , y , x) = 0. However, we
shall restrict our attention to I order ODE of the form y 0 = f (x, y ).
(Implicit Function Theorem!)
Observe that any first order ODE of the form y 0 = f (x, y ) can be
equivalently written in the differential form as

M(x, y )dx + N(x, y )dy = 0. (3.3)

The differential is motivated from the chain rule structure! If we

imagine x and y as functions of t and F as functions of x and y then
the chain rule says that
dF ∂F dx ∂F dy
= + .
dt ∂x dt ∂y dt

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 22 / 126
Total Differential

Definition
Let F : Ω ⊂ R2 → R be a two-variable function that admit continuous
first order partial derivatives in Ω. The total differential, denoted as dF , is
defined as
∂F ∂F
dF = dx + dy .
∂x ∂y

The definition can be extended to n variable function in the obvious

way.The total differential signifies the changes in the function as
linear combination of changes in each variable.
The advantage of differential form version is that the roles of x and y
can be interchanged.
It should be observed that solutions may be gained or lost while
performing multiplication/division operations. We shall illustrate this
phenomenon in the following example.
T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 23 / 126
Example (loss of solution)

Example
4
Consider the ODE y 0 = x(x−4)y
3 (y 2 −3) . Observe that y ≡ 0 is one possible

solution to the ODE. Rewriting it in the differential form, we obtain

(x − 4)y 4 dx − x 3 (y 2 − 3)dy = 0.

Dividing the equation above by x 3 y 4 , we obtain

(x −2 − 4x −3 )dx − (y −2 − 3y −4 )dy = 0.

On integration, we have the family of solutions

−x −1 + 2x −2 + y −1 − y −3 = c.

Note that the solution y = 0 (a solution of y 4 = 0) is not a member of the

above family which got eliminated with the restriction y 4 6= 0.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 24 / 126
Example (no loss of solution)

Example
Consider the ODE y 0 = (x 2−x sin y
+1) cos y
. Observe that the roots of sin y = 0,
i.e. y = kπ for k ∈ Z are constant solutions to the ODE. Rewriting it in
the differential form, we obtain

x sin y dx + (x 2 + 1) cos y dy = 0.

Dividing the equation above by (x 2 + 1) sin y , we obtain

x cos y
dx + dy = 0.
x2 + 1 sin y
On integration, we have the family of solutions
1 2 2 2 2
2 ln(x + 1) + ln | sin y | = ln |c| or (x + 1) sin y = c . Note that the
solution y = kπ, for all k ∈ Z (a solution of sin y = 0) is also a member of
the above family (for c = 0) in spite of imposing the restriction sin y 6= 0.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 25 / 126
Separable type ODE

Definition
The first order ODE is said to be of separable type if f (x, y ) is of the form
− M(x)
N(y ) . Equivalently, M and N are independent of the y and x variable,
respectively, in the differential form M(x)dx + N(y )dy = 0.

The family Rof solutions for

R a separable equation is
F (x, y ) := M(x) dx + N(y ) dy = c.

Example
Consider the ODE (x −2 − 4x −3 )dx − (y −2 − 3y −4 )dy = 0. Then, on
integration, we obtain the one parameter family of solutions
1 2 1 1
− + 2 + − 3 = c.
x x y y

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 26 / 126
Homogeneous ODE

Definition
A function v : R2 → R is said to be homogeneous of degree n if
v (tx, ty ) = t n v (x, y ) for all non-zero t ∈ R.

For instance, v (x, y ) := x 2 + y 2 is homogeneous of degree 2.

Definition
A first order ODE of the form (3.3) is homogeneous if both M and N are
homogeneous of same degree. Equivalently, y 0 = f (x, y ) is homogeneous if
f (tx, ty ) = f (x, y ).
p
For instance, (y + x 2 + y 2 )dx − xdy = 0 is homogeneous.
The reason for considering homogeneous ODEs is that it can be
transformed to ODE with separated variables.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 27 / 126
Homogeneous to Separable Form

Theorem
If y 0 = f (x, y ) is homogeneous then the change of variable y = zx
transforms the ODE into a separable type ODE in x and z variables.

Proof.
By homogeneity, since f (tx, ty ) = f (x, y ) for any t, we choose in
particular t = 1/x and obtain f (1, y /x) = f (x, y ). Thus, the given ODE
becomes y 0 (x) = f (1, y /x). Using the change of variable y (x) = xz(x) in
the ODE, we get z + xz 0 = f (1, z) or equivalently, z 0 = f (1,z)−z
x is in the
separable form.

Note that such a reduction is valid in region excluding {x = 0}.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 28 / 126
Solution to Homogeneous ODE

A homogeneous ODE y 0 = f (x, y ) can be reduced to the separable

type z 0 = f (1,z)−z
x .
The solution to the separable type form is given as
Z Z
dz dx
+ = c.
z − f (1, z) x

Thus, in terms of x and y variable, the family of solutions is

F (y /x) + ln |x| = c where
Z
dz
F (z) = .
z − f (1, z)

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 29 / 126
Example

Example
Consider the ODE (x 2 − 3y 2 )dx + 2xydy = 0 or, equivalently,
2 −x 2
y 0 = 3y2xy .
This ODE is homogeneous because M and N are homogeneous of
degree 2, i.e. M(tx, ty ) = t 2 (x 2 − 3y 2 ) = t 2 M(x, y ) and
N(tx, ty ) = t 2 2xy = t 2 N(x, y ).
3z 2 −1
Using y = zx substitution, we get z + xz 0 = 2z or, equivalently
2
xz 0 = z 2z−1 .
Thus, the separable type ODE we obtain is
2z dx
dz − = 0.
z2 −1 x

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 30 / 126
Example
Example
On integration, the family of solutions is

ln |z 2 − 1| − ln |x| = ln |c|

or, equivalently, |z 2 − 1| = |cx|. Finally, |y 2 − x 2 | − |cx|x 2 = 0 is the

family of solutions.

Example
Consider the IVP
 p
(y + x 2 + y 2 )dx − xdy = 0
y (1) = 0.
√
y+ x 2 +y 2
The ODE can be rewritten as = x y0 . We have already
checked that this ODE is homogeneous!
T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 31 / 126
Homogeneous IVP

Example
√
Using y = zx substitution,
√ we get z + xz 0 = z + 1 + z 2 or,
0
equivalently xz = 1 + z .2

Thus, the separable type ODE we obtain is

dz dx
√ − = 0.
2
z +1 x

On integration,
√ the family of solutions is √
ln |z + z 2 +p1| − ln |x| = ln |c| or, equivalently, |z + z 2 + 1| = |cx|.
Finally, |y + y 2 + x 2 | − |c|x 2 = 0 is the family of solutions.
Using the initial conditions, we get |c| = 1.
p
The solution is |y + y 2 + x 2 | − x 2 = 0.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 32 / 126
One degree Polynomial Coefficients

Theorem
Consider the ODE (a11 x + a12 y + b1 )dx + (a21 x + a22 y + b2 )dy = 0 for
given constants aij and bi with i, j = 1, 2. Let A = (aij ) be the 2 × 2
matrix.
If A is invertible and (h, k) = A−1 (b1 , b2 ) then the transformation
w = x + h and z = y + k reduces the ODE to a homogeneous ODE
in w and z variable of the form

(a11 w + a12 z)dw + (a21 w + a22 z)dz = 0.

If A is not invertible then the transformation z = a11 x + a12 y reduces

the ODE to a separable ODE in z and x variables.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 33 / 126
Example

Example
Consider the ODE (x − 2y + 1)dx + (4x − 3y − 6)dy = 0.
Determinant of the matrix A is 5 6= 0. Then its inverse is
 
−1 1 −3 2
A = .
5 −4 1

Therefore (h, k) = A−1 (1, −6) = (−3, −2). Thus, w = x − 3 and

z = y − 2.
Using this transformation in the ODE we obtain
(w − 2z)dw + (4w − 3z)dz = 0 or z 0 = 4w
2z−w
−3z . This is homogeneous!
2ζ−1
Use the transformation z = ζw to get ζ + w ζ 0 = 4−3ζ .

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 34 / 126
Example

Example
or, equivalently,
4 − 3ζ dw
dζ = .
3ζ 2 − 2ζ − 1 w
On integration, we obtain
 
3  3ζ − 3  1
ln − ln |3ζ 2 − 2ζ − 1| = ln |w | + ln |c0 |
4  3ζ + 1  2

 3ζ−3 3
 
or, equivalently, ln  3ζ+1  − ln |3ζ 2 − 2ζ − 1|2 = ln |wc0 |4 or
   
 ζ−1  4  ζ−1  4
ln  (3ζ+1)5  = ln |wc0 | or  w 4 (3ζ+1)5  = c0 .
 
 z−w 
Replacing ζ = z/w , we get  (3z+w )5 
= c04 .
 
 y −x+1 
Going back to x and y variables, we get  (3y +x−9)5  = c04 .

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 35 / 126
Example

Example
Consider the ODE (x + 2y + 3)dx + (2x + 4y − 1)dy = 0.
Determinant of the matrix A is 0.
The matrix is not invertible.
Using the transformation z =
x + 2y in the ODE we obtain
(z + 3)dx + (2z − 1) dz−dx
2 = 0 or 7dx + (2z − 1)dz = 0. This is
separable!
Integrating, we get 7x + z 2 − z = c.
Going back to x and y variables, we get
7x + (x + 2y )2 − (x + 2y ) = c or x 2 + 4xy + 4y 2 + 6x − 2y = c.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 36 / 126
Exact differential Equation

Definition
An ODE of the form (3.3) is said to be an exact differential equation if
there exists a two variable function F such that ∂x F (x, y ) = M(x, y ) and
∂y F (x, y ) = N(x, y ).

The motivation for the above definition is that an exact differential

equation can be rewritten as dF (x, y ) = 0 and solved as F (x, y ) = c for
any constant c ∈ R.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 37 / 126
Examples

Example
The ODE y 2 dx + 2xydy = 0 is an exact differential equation because
F (x, y ) = xy 2 satisfies the requirement.

Example
The ODE ydx + 2xdy = 0 is not exact!

Now, showing exactness of an ODE is not that easy!

We either need to find a F or show the non-existence of a F meeting
the requirement.
Thus, we need a test to determine the exactness of ODE.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 38 / 126
Necessary and Sufficient Condition for Exactness

Theorem
Let M and N admit continuous first order partial derivatives in a
rectangular domain Ω ⊂ R2 . The ODE (3.3) is exact in Ω iff
∂y M(x, y ) = ∂x N(x, y ) for all (x, y ) ∈ Ω.

Proof.
The necessary part is obvious! The sufficiency part involves finding
suitable F . The arguments are shown by examples. Abstract proof follows
the same line as you see in examples.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 39 / 126
Examples

Example
The ODE y 2 dx + 2xydy = 0 is exact in every domain Ω ⊂ R2 because
M(x, y ) = y 2 and N(x, y ) = 2xy with ∂y M(x, y ) = 2y = ∂x N(x, y ).

Example
The ODE ydx + 2xdy = 0 is not exact because M(x, y ) = y and
N(x, y ) = 2x with ∂y M(x, y ) = 1 6= 2 = ∂x N(x, y ).

Example
The ODE (2x sin y + y 3 e x )dx + (x 2 cos y + 3y 2 e x )dy = 0 is exact because

∂y M(x, y ) = 2x cos y + 3y 2 e x = ∂x N(x, y ).

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 40 / 126
How to Solve Exact ODE?

Theorem
If the ODE (3.3) is exact in Ω then there is a one parameter family of
solutions to (3.3) given by F (x, y ) = c where c is arbitrary constant and
F is such that ∂x F = M and ∂y F = N in Ω.

Proof.
By the definition of exact ODE there is a F satisfying ∂x F = M and
∂y F = N. Thus, the ODE becomes dF (x, y ) = 0. Hence, F (x, y ) = c for
arbitrary c are solutions.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 41 / 126
How to find the Solution F ?

We shall present two methods to identify F

The “standard” method
Method of grouping

Example (Standard Method)

Consider the ODE (3x 2 + 4xy )dx + (2x 2 + 2y )dy = 0. It is exact because
M(x, y ) = 3x 2 + 4xy and N(x, y ) = 2x 2 + 2y with
∂y M(x, y ) = 4x = ∂x N(x, y ). We seek F such that

∂x F (x, y ) = M(x, y ) = 3x 2 + 4xy and ∂y F (x, y ) = N(x, y ) = 2x 2 + 2y .

From the first identity,

Z
F (x, y ) = M(x, y ) dx + g (y ) = x 3 + 2x 2 y + g (y ).

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 42 / 126
Standard Method to find F

Example
From the second identity, we must have

2x 2 + 2y = 2x 2 + g 0 (y ).

Therefore, g 0 (y ) = 2y and g (y ) = y 2 + c0 . Thus,

F (x, y ) = x 3 + 2x 2 y + y 2 + c0 . Since F (x, y ) = c is the one paramater
family of solutions, we get

x 3 + 2x 2 y + y 2 + c0 = c

or, equivalently, x 3 + 2x 2 y + y 2 = c is the one parameter family of

solutions.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 43 / 126
Method of grouping to find F

Example
Consider the same ODE (3x 2 + 4xy )dx + (2x 2 + 2y )dy = 0. We have
already checked it is exact! We rewrite the ODE as follows:

3x 2 dx + (4xydx + 2x 2 dy ) + 2ydy = 0
d(x 3 ) + d(2x 2 y ) + d(y 2 ) = d(c)
d(x 3 + 2x 2 y + y 2 ) = d(c).

Thus, x 3 + 2x 2 y + y 2 = c is a family of solutions.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 44 / 126
Exact IVP

Example
Consider the IVP
(2x cos y + 3x 2 y )dx + (x 3 − x 2 sin y − y )dy = 0


y (0) = 2.

The equation is exact in R2 because

∂y M(x, y ) = −2x sin y + 3x 2 = ∂x N(x, y ) in R2 . We seek F such that

∂x F (x, y ) = M(x, y ) = 2x cos y + 3x 2 y and

∂y F (x, y ) = N(x, y ) = x 3 − x 2 sin y − y .

From the first identity,
Z
F (x, y ) = M(x, y ) dx + g (y ) = x 2 cos y + x 3 y + g (y ).

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 45 / 126
Example
From the second identity, we must have

x 3 − x 2 sin y − y = −x 2 sin y + x 3 + g 0 (y ).
2
Therefore, g 0 (y ) = −y and g (y ) = − y2 + c0 . Thus,
2
F (x, y ) = x 2 cos y + x 3 y − y2 + c0 . Since F (x, y ) = c is the one
paramater family of solutions, we get

y2
x 2 cos y + x 3 y − + c0 = c
2
2
or, equivalently, x 2 cos y + x 3 y − y2 = c is the one parameter family of
solutions. Now, using the initial value we get c = −2 and
2
x 2 cos y + x 3 y − y2 + 2 = 0 is a solution.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 46 / 126
Same Example via Grouping Method

Example
Consider the same IVP
(2x cos y + 3x 2 y )dx + (x 3 − x 2 sin y − y )dy = 0


y (0) = 2.

We have already checked it is exact! We rewrite the ODE as follows:

(2x cos ydx − x 2 sin ydy ) + (3x 2 ydx + x 3 dy ) − ydy = 0

 2
2 3 y
d(x cos y ) + d(x y ) − d = d(c).
2
2
Thus, x 2 cos y + x 3 y − y2 = c is a family of solutions and using the initial
condition we get c = −2.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 47 / 126
Integrating Factor
Some non-exact ODEs can be transformed into an exact ODE using
integrating factor!
Definition
Let (3.3) be not exact in Ω. If there exists a function µ : Ω → R such that

µ(x, y )M(x, y )dx + µ(x, y )N(x, y )dy = 0

is exact in Ω then µ is said to be an integrating factor of (3.3).

Please note that multiplication (by integrating factor) may lead to

gain/loss of solutions!
In general, finding the integrating factor µ is not easy. (It appears as
a solution of partial differential equation involving M and N).
However, for some special ODEs computing integrating factor is
easier!
T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 48 / 126
Integrating Factor for some Non-Exact ODE

Theorem
Consider the ODE M(x, y )dx + N(x, y )dy = 0 and let

∂M ∂N
d(x, y ) := (x, y ) − (x, y ).
∂y ∂x
R 
d(x,y ) d(x,y )
(i) If N(x,y ) depends only on x then exp N(x,y ) dx is an I.F. of (3.3).
R 
(ii) If −d(x,y )
M(x,y ) depends only on y then exp
−d(x,y )
M(x,y ) dy is an I.F. of
(3.3).

It is possible that the situations are different from above possibilities in

which case we seek other means!

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 49 / 126
Example

Example
Consider the ODE (2x 2 + y )dx + (x 2 y − x)dy = 0. This ODE is not
separable, not homogeneous, not exact. Then d(x, y ) := 2(1 − xy ). Now,

d 2(1 − xy ) −2
= 2 =
N x y −x x
is dependent only on x. Thus, the I.F. is
 Z 
1 1
exp −2 dx = exp(−2 ln |x|) = 2 .
x x

Multiplying the I.F. in the equation we get the exact equation

(2 + xy2 )dx + (y − x1 )dy = 0.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 50 / 126
Derivative as a Linear Map

dk
The ordinary differential operator of order k is denoted as dx k
.
A k-th order ordinary differential operator can be viewed as a map
dk k ¯ ¯ ¯ k ¯
from dx k : C (I ) → C (I ) where I = [a, b] and C (I ) is the set of all
k-times differentiable function in I with the k-the derivative being
continuous and extended continuously to I¯.
(Exercise) Verify that C (I¯) is a vector space over R.
(Exercise!) Verify that C k (I¯) ⊂ C (I¯) for k ∈ N. Also, find examples
of differentiable functions whose derivative is not continuous!

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 51 / 126
Classification of ODE in terms of Linearity

The level of difficulty in solving a ODE may depend on its order k and
linearity of F .
Definition
A k-th order ODE is linear if F in (3.1) has the form

F := Ly − f (x)

where Ly (x) := ki=0 ai (x)y (i) (x) for given functions f and ai ’s such that
P
ak is not identically zero. Otherwise the ODE is nonlinear. If, in addition
to being linear, one has f ≡ 0 then the ODE is linear homogeneous.

It is called linear because L is linear in y , i.e.,

L(cy1 + dy2 ) = cL(y1 ) + dL(y2 ) for all c, d ∈ R.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 52 / 126
Examples

Example
2
(i) y 00 + 5y 0 + 6y = 0. Here L := dx
d d
2 + 5 dx + 6. and is linear,
homogeneous because f ≡ 0 and second order with constant
coefficients.
(ii) x (4) + 5x (2) + 3x = sin t is linear and fourth order with constant
coefficients.
(iii) y (4) + x 2 y (3) + x 3 y 0 = xe x is linear and fourth order with variable
coefficients.
(iv) y 00 + xy (y 0 )2 = 0 is nonlinear and second order.
(v) y 00 + 5y 0 + 6y 2 = 0 is nonlinear and second order.
(vi) y 00 + 5(y 0 )3 + 6y = 0 is nonlinear and second order.
(vii) y 00 + 5yy 0 + 6y = 0 is nonlinear and second order.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 53 / 126
Linear First Order ODE

Definition
A first order ODE is linear in the y -variable (unknown) if it is of the form
P(x)y 0 + Q(x)y (x) = R(x).

The differential form is: [Q(x)y − R(x)]dx + P(x)dy = 0.

A linear ODE is exact iff Q(x) = P 0 (x). In that case the ODE
becomes (P(x)y )0 = R(x) which is already in the separable form.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 54 / 126
Family of Solution to Linear FODE

Theorem
The linear ODER y 0 + Q(x)y (x) = R(x) has the integrating factor
µ(x, y ) := e Q(x) dx and its one parameter family of solutions is

R
Z R 
− Q(x) dx Q(x) dx
y (x) = e e R(x) dx + c .

Proof.
The differential form of the given linear ODE is
[Q(x)y − R(x)] dx + dy = 0. This is exact iff Q ≡ 0. If Q is non-zero then
d(x, y ) = Q(x) and the ratio Nd = Q depends only R on x. By the theorem
on I.F. for non-exact ODE its I.F. is µ(x) := exp( Q) which depends only
on x. With this choice of µ the linear ODE becomes (µy )0 = R(x)µ.
Thus, the family of solutions as given in the statement is obtained.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 55 / 126
Example

Example
Consider y 0 + 2x+1
y = e −2x .


x
Its integrating factor is
Z   
2x + 1
µ(x) := exp dx = exp(2x + ln |x|)
x
= e 2x e ln |x| = xe 2x .

Using the I.F. in the ODE we get xe 2x y 0 + e 2x (2x + 1)y = x or,

equivalently, (xe 2x y )0 = x.
x2
Thus, xe 2x y − 2 = c is a family of solutions.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 56 / 126
IVP Example

Example
Consider the IVP
(  
y0 + 4x
x 2 +1
y = x
x 2 +1
y (2) = 1.
R 
4x
The I.F. is µ(x) = exp x 2 +1
dx = exp[ln(x 2 + 1)2 ] = (x 2 + 1)2 .
Using the integrating factor in the ODE we get [(x 2 + 1)2 y ]0 = x 3 + x.
x4 x2
Thus, its family of solutions is (x 2 + 1)2 y = 4 + 2 + c.
Using the initial condition, we get c = 19. Thus, the solution is
4 2
(x 2 + 1)2 y = x4 + x2 + 19.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 57 / 126
Interchange of x and y variable in differential form

Example
y 2
Consider the ODE y 0 = 1−3xy . This example does not fall in any of
the category we have seen so far: it is not separable, not
homogeneous, not exact and not linear.
However, treating the y as independent variable and x as dependent
variable, the ODE x 0 = 1−3xy
y2
is linear in x. (The 0 now denotes the
derivative with respect to y variable).
The justification of this interchange is made via the differential form
of the ODE. R 
3
The I.F. for the linear ODE is µ(y ) = exp y dy = y 3.
Multiplying the I.F in the linear ODE we get (y 3 x)0 = y .
y2
Thus, the family of solutions is y 3 x − 2 = c.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 58 / 126
Bernoulli Equations

Definition
An equation of the form y 0 + Q(x)y = R(x)y α , for a given fixed real
number α, is called a Bernoulli differential equation.

For α equal to zero or one, the Bernoulli equation is linear.

For other values of α the Bernoulli equation can be transformed to a
linear equation

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 59 / 126
Transforming to Linear

Theorem
Let α 6= 0, 1. The transformation z = y 1−α reduces the Bernoulli equation
to a linear equation in z.

Proof.
We first divide y α to the Bernoulli equation to obtain
y −α y 0 + Q(x)y 1−α = R(x).
Set z = y 1−α then z 0 = (1 − α)y −α y 0 .
Using this transformation in the ODE above, we get the linear ODE

z 0 + (1 − α)Q(x)z = (1 − α)R(x).

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 60 / 126
Example

Example
Consider the Bernoulli equation y 0 + y = xy 3 .
We divide by y 3 to obtain y −3 y 0 + y −2 = x.
Set z = y −2 . Then z 0 = −2y −3 y 0 and the transformed linear ODE
obtained is z 0 − 2z = −2x.
Its I.F. is µ(x) = e −2x .
Using the I.F. in the linear ODE we get (e −2x z)0 = −2xe −2x .
Integrating both sides, e −2x z = 12 e −2x (2x + 1) + c. Thus,
e −2x y −2 = 21 e −2x (2x + 1) + c.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 61 / 126
Bernoulli Equation as a Special case

Bernoulli equation is a special case of g 0 (y )y 0 + Q(x)g (y ) = R(x)

with g (y ) = y 1−α .
Set z = g (y ). Then z 0 = g 0 (y )y 0 .
Thus, the above ODE is transformed to a linear ODE in the z as
z 0 + Q(x)z = R(x) which can be solved using I.F.

Example
Consider cos yy 0 + x −1 sin y = 1 where g (y ) = sin y .
Use z = sin y . Then we obtain the linear ODE in z, z 0 + x −1 z = 1.
Its I.F. is µ(x) = |x|.
x2
For x > 0, x sin y − 2 = c.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 62 / 126
Existence for First Order of the form y 0 = f (x, y )

Since all ODE does not admit solution in Exact form let us answer the
question of existence of solution.

Theorem (Picard)
Let Ω be an open, connected subset (domain) of R2 and (x0 , y0 ) ∈ Ω. If
f : Ω → R is continuous then the first order IVP
y 0 = f (x, y )

(4.1)
y (x0 ) = y0

admits at least one solution.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 63 / 126
Non-uniqueness of Solution

In general the IVP (4.1) may admit more than one solution!

Example
Consider the IVP y 0 = y 1/3 in R with the initial conditions y (0) = 0. Note
that y ≡ 0 is one solution. Also, being separable type ODE, we obtain

 2(x−c) 3/2

3 x ≥c
yc (x) :=
0 x ≤c

is also a solution for all c ≥ 0! Thus, this problem admits infinitely many
solutions.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 64 / 126
Picard’s Uniqueness Theorem

Theorem (Uniqueness)
Let Ω be a domain of R2 and (x0 , y0 ) ∈ Ω. If both f : Ω → R and
∂f
∂y : Ω → R are continuous then there exists a h > 0 such that the first
order IVP (4.1) admits a unique solution in the interval [x0 − h, x0 + h].

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 65 / 126
Examples for Picard’s Theorem

Since we shall not be proving the theorem, let us verify the theorem with
some examples.
Example
Consider the IVP y 0 = x 2 + y 2 with y (1) = 3.
∂f
Note that f (x, y ) = x 2 + y 2 and ∂y = 2y .
Both are continuous in every domain of R2 .
By Picards’ theorem, for every domain Ω that contains (1, 3) there is
a h > 0 such that the IVP has a unique solution in [1 − h, 1 + h].

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 66 / 126
Examples for Picard’s Theorem

Example
Consider the ODE y 0 = yx −1/2 . Note that f (x, y ) = yx −1/2 and
∂f −1/2 . They are continuous in the open right half plane, i.e
∂y = x
{(x, y ) | x > 0}.
If the initial condition imposed is y (1) = 2 then for every domain Ω
that contains (1, 2) and does not contain the y -axis there is a h > 0
such that the IVP has a unique solution in [1 − h, 1 + h].
If the initial condition imposed is y (0) = 2 then both f and ∂y f are
not continuous (undefined) at (0, 2). Thus, a comment on the
existence and uniqueness of solution is inconclusive!

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 67 / 126
Motivation for Picard’s Successive Approximation
Consider the IVP (4.1), i.e.

y 0 = f (x, y ) in I


y (x0 ) = y0

such that x0 ∈ I¯.

Integrating both sides of the differential equation
Rx between x0 and x,
we get the integral equation y (x) = y0 + x0 f (t, y (t)) dt.
Solving for y in the ODE is same as solving for y in the integral
equation.
Rx
Define the map T : C (I¯) → C (I¯) as Ty (x) = y0 + x0 f (t, y (t)) dt
(ignore concerns related to the well-definedness of T for this course).
Then the solution to the integral equation y = Ty is a fixed point of
T.
Picard’s approximation gives an algorithm to find the fixed point of T
and, in turn the solution of the IVP (4.1).
T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 68 / 126
Picard’s Successive Approximation

Consider the IVP (4.1), i.e.

y 0 = f (x, y )


y (x0 ) = y0

Choose a function u0 (zeroth approximation) such that u0 (x0 ) = y0 .

Next choose u1 (first approximation) as a solution to

u10 = f (x, u0 )


u1 (x0 ) = y0

or, equivalently, for ‘nice’ f

Z x
u1 (x) = y0 + f [s, u0 (s)] ds.
x0

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 69 / 126
Picard’s Successive Approximation

Similarly, the k-th approximation uk is chosen such that

Z x
uk (x) = y0 + f [s, uk−1 (s)] ds.
x0

Under suitable conditions, it can be shown that {uk } will converge to

some u which will be a solution of the IVP.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 70 / 126
Example

Example
Consider the IVP
y0 = x2 + y2


y (0) = 1

Choose u0 ≡ 1 and, for k ≥ 1,

Z x  2 2

uk (x) = 1 + s + uk−1 (s) ds.
0

x3
On integration, we obtain, u1 (x) = 1 + x + 3 .
Similarly,

2x 3 x 4 2x 5 x 7
u2 (x) = 1 + x + x 2 + + + +
3 6 15 63

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 71 / 126
Example

Example
Consider the IVP √
y0 = 2 x


y (0) = 1

Choose u0 ≡ 1 and, for k ≥ 1,

Z x √ 4 3
uk (x) = 1 + 2 s ds = 1 + x 2 .
0 3

Thus, we have a constant sequence {uk }, for k ≥ 1 converging to the

exact solution obtained by integrating (separable ODE).

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 72 / 126
Integral Curves

Observe that the solution y : I → R of an ODE can be geometrically

viewed as the curve (x, y (x)) (graph of y , at least locally) in R2 .
Thus, a one parameter family of solutions for a first order ODE
y 0 = f (x, y ) will correspond to a one parameter family of curves in R2 .
These curves have the property that the slope of their tangents at
(x0 , y0 ) is f (x0 , y0 ) and such curves are called the integral curves of
the ODE y 0 = f (x, y ).
For instance, the family of parabolae x 2 + c are the integral curves of
the ODE y 0 = 2x.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 73 / 126
Graphical Method: An approximate Method

The family of solutions F (x, y , c) = 0 of y 0 = f (x, y ) is not always

computable exactly.
Note that y 0 = f (x, y ) gives the slope f (x, y ) of the integral curves at
each point (x, y ). For instance, the solution curve of y 0 = 2x + y has
slope 4 at (1, 2) because f (1, 2) = 4.
Thus, at (1, 2) one can draw a small line element of slope 4. Doing
this at all points gives the direction field of the ODE.
Smooth curves drawn tangent to the line elements provides the
approximate ‘graph’ of the solution curve.
This process is tedious, time-consuming and does not provide
analytical expressions for solutions!

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 74 / 126
Method of Isoclines

Definition
An isocline of the ODE y 0 = f (x, y ) is a curve along which the slope
f (x, y ) has a constant value c.

For different values of c, the curves f (x, y ) = c are the family of isoclines
of the ODE.
For each fixed c, draw the line elements with slope c (or angle θ such
that tan θ = c) at each (x, y ) in the curve f (x, y ) = c.
Repeat the procedure for different values of c.
Smooth curves drawn tangent to the line elements provides the
approximate ‘graph’ of the solution curve.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 75 / 126
Example
Consider the ODE y 0 = x. The isoclines of the ODE are the vertical lines
{x = c}.

-1

-2

-3

-3 -2 -1 0 1 2 3

Figure: Line Elements (Image Courtesy: Prof. S. Ghorai)

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 76 / 126
 3

-1

-2

-3

-3 -2 -1 0 1 2 3

Figure: Approximate solution curves (Image Courtesy: Prof. S. Ghorai)

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 77 / 126
Euler’s Method

If φ solves the IVP y 0 = f (x, y ) with y (x0 ) = y0 which cannot be

determined exactly, we wish to approximate φ.
Let h > 0 be the incremental value chosen for x. Say xk+1 = xk + h
for all k ∈ N ∪ {0}.
Since φ is a solution, we know that
Z xk+1
φ(xk+1 ) = φ(xk ) + f (x, φ(x)) dx.
xk

The second term is the area under the function f (x, φ(x)) in the
interval (xk , xk+1 ).

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 78 / 126
Left End-point Approximation
If the variation of f (x, y ) is small in (xk , xk+1 ) then we can replace f (x, y )
for all x ∈ (xk , xk+1 ) by f (xk , y (xk )), its value at left end point xk , which
approximates φ(xk+1 ), i.e.
φ(xk+1 ) ≈ yk+1 := yk + hf (xk , yk ).

Figure: Euler Approximation

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 79 / 126
Geometric Interpretation and Approximation Error

Let φ represents the graph of the solution of the IVP passing through
A := (x0 , y0 ).
Then y = y0 + f (x0 , y0 )(x − x0 ) is the equation of the tangent at A
with slope f (x0 , φ(x0 )).
The point B := (x1 , y1 ) approximates the solution at B̃ := (x1 , φ(x1 )).
The error is the length of the segment B B̃ = |y1 − φ(x1 )|.
T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 80 / 126
Example

Example
Consider the linear IVP y 0 = 2x + y with y (0) = 1.
The exact solution is φ(x) := 3e x − 2(x + 1).
Here f (x, y ) = 2x + y and y0 = 1.
Let us choose h = 14 then the points of evaluation are x0 = 0,
x1 = 1/4, x2 = 1/2, x3 = 3/4 and x4 = 1.
Thus, y1 = y0 + hf (x0 , y0 ) = 1 + 1/4 = 5/4.
Similarly, y2 = 5/4 + 1/4(7/4) = 27/16.
|yk −φ(xk )|
xk φ(xk ) yk |yk − φ(xk )| |φ(xk )|
1/4 1.3521 1.25 0.1021 0.0755
1/2 1.9462 1.6875 0.2587 0.1329
3/4 2.8510 2.3594 0.4916 0.1724
1.0 4.1548 3.3242 0.8306 0.1999

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 81 / 126
Piecewise Linear Approximation

Figure: Approximate Solution

For each chosen h, we construct a Euler approximate solution of the IVP

which is a piecewise linear graph obtained by joining succesive points
ABCD . . ..
T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 82 / 126
Better Approximations
Observe that if h is chosen small then the error B B̃ becomes smaller and
the approximation is better!

Figure: Smaller h

If h is not very small then the errors may be large. If h is very small then
the errors are small but the number of computations increase making the
method tedious and time-consuming.
T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 83 / 126
Improved Euler’s Method
In the Improved Euler we approximate the integral with area of a
Trapezoidal region.
Set ỹk+1 := yk + hf (xk , yk ) and
yk+1 := yk + h2 [f (xk , yk ) + f (xk+1 , ỹk+1 )].

Figure: Improved Euler Approximation

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 84 / 126
Example

Example
Consider the IVP y 0 = 2x + y with y (0) = 1.
Let us chooseh = 14 then the points of evaluation are x0 = 0,
x1 = 1/4, x2 = 1/2, x3 = 3/4 and x4 = 1.
Thus, ỹ1 = y0 + hf (x0 , y0 ) = 1 + 1/4 = 5/4
and y1 = 1 + 81 [1 + 7/4] = 43/32.
Similarly, ỹ2 = 43/32 + 1/4(59/32)
and y2 = 1.9248.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 85 / 126
Rate of Convergence

Let’s compare the convergence of error, for instance, at

φ(1) = 3e − 4 = 4.15484548.
h E φh (1) E Eh Eh /Eh/2 IE φh (1) IE Eh Eh /Eh/2
1 2 0.5186 3.5 0.1576
1/2 2.75 0.3381 1.5339 3.9219 0.0561 2.8109
1/4 3.3242 0.1999 1.6913 4.0846 0.0169 3.3150
1/8 3.6974 0.1101 1.8156 4.1355 0.0047 3.6373
1/16 3.9138 0.0580 1.8978 4.1498 0.0012 3.8148
1/32 4.0310 0.0298 1.9460 4.1535 3.120e-04 3.9067
1/64 4.0920 0.0151 1.9722 4.1545 7.893e-05 3.9532
1/128 4.1232 0.0076 1.9859 4.15476 1.985e-05 3.9766
1/256 4.1390 0.0038 1.9929 4.15482 4.977e-06 3.9883
1/512 4.1469 0.0019 1.9964 4.154840 1.246e-06 3.9941
1/1024 4.1509 0.0010 1.9982 4.154844 3.117e-07 3.9971

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 86 / 126
Application: Oblique and Orthogonal Trajectories

Definition
Let F (x, y , c) = 0 be a one parameter family of curves in R2 . The oblique
trajectory of the given family is any curve that intersects the curves in the
given family at a constant angle θ. If θ = 90◦ i.e. the intersection is at
right angles, then the curve is said to be an orthogonal trajectory of the
given family.

Example
Consider the family of circles x 2 + y 2 = c 2 with origin as centre and radius
c. Then every straight line through the origin y = mx is an orthogonal
trajectory of the given family of circles. Conversly, each circle is an
orthogonal trajectory of the family of straight lines passing through its
centre.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 87 / 126
Finding Oblique and Orthogonal Trajectories
Given a family of curves F (x, y , c) = 0, differentiate w.r.t x and
eliminate the constant c to obtain a first order ODE y 0 = f (x, y ).
y 0 = f (x, y ) gives the slope f (x, y ) of the integral curves at (x, y ).
Thus, the angle of inclination of the tangent line is arctan[f (x, y )].
The tangent line of the oblique trajectory intersecting the given
family of curves at a constant angle θ will have the inclination
arctan[f (x, y )] + θ at (x, y ).
The slope of the tangent line of the oblique trajectory at (x, y ) is
given as
f (x, y ) + tan θ
tan [arctan[f (x, y )] + θ] = .
1 − f (x, y ) tan θ
f (x,y )+tan θ
Solve the ODE y 0 = 1−f (x,y ) tan θ to obtain the one parameter family
of oblique curves, say G (x, y , c) = 0.
Solve the ODE y 0 = − f (x,y
1
) to obtain the one parameter family of
orthogonal curves, say G (x, y , c) = 0.
T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 88 / 126
Example

Example
Consider the family of circles x 2 + y 2 = c 2 centred at origin.
−x
Differentiating the family we get x + yy 0 = 0 or y 0 = y .
Then, the ODE for the orthogonal family is given by y 0 = yx which is
separable.
Solving, we get y = mx a one parameter family of orthogonal curves.
Caution: The orthogonal (vertical) line {x = 0} is not obtained in the
above family of curves and should be observed by inspection.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 89 / 126
Example

Example
Consider the family of parabolae y = cx 2 .
Differentiating the family we get y 0 = 2cx. Eliminating the parameter
c between the ODE and the family of curves, we get y 0 = 2y x .
Then, the ODE for the orthogonal family is given by y 0 = − 2y
x
which
is separable.
Solving, we get 2y 2 + x 2 = b 2 a one parameter family of ellipses
centred at origin with x-axis as the major axis.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 90 / 126
Example: Oblique Trajectory

Example
Consider the family of straight lines y = mx. Let us find the oblique
trajectory intersecting the given family at constant angle of 45◦ .
Differentiating the family we get y 0 = m. Eliminating the parameter
m between the ODE and the family of curves, we get y 0 = yx .
Now,
y
f (x, y ) + tan 45◦ x +1 x +y
◦
= y = .
1 − f (x, y ) tan 45 1− x x −y

Then, the ODE for the oblique family with angle 45◦ is given by
x+y
y 0 = x−y which is homogeneous.
Setting, y = zx, we get z + xz 0 = 1+z
1−z .

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 91 / 126
Example: Oblique Trajectory

Example
Thus, solving the separable ODE (z−1) z 2 +1
dz = − dx
x , we get
1 2
2 ln(z + 1) − arctan z = − ln |x| − ln |c| or
ln c 2 (x 2 + y 2 ) − 2 arctan(y /x) = 0 is a one parameter family of
oblique trajectories.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 92 / 126
Topics to be covered in Higher order

Linear Independence of Solutions.

Reduction Order technique (Linearly independent solution from a
given solution).
Linear homogeneous with constant coefficients ODE
Linear inhomogeneous with constant coefficients ODE (Method of
Undetermined Coefficients)
Linear ODE with variable coefficients (Method of Variation of
Parameter)
Power Series Solution

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 93 / 126
k-th order to System of first order ODE

Consider a k-th order ODE of the form y (k) = f (x, y , y 0 , . . . y (k−1) )

For 1 ≤ i ≤ k, introduce k unknowns yi := y (i−1) where y = y1 .
We have the system of k first order ODEs
 0
yi = yi+1 (1 ≤ i ≤ k − 1)
yk0 = f (x, y1 , y2 , . . . , y(k−1) ).

Thus, the existence and uniqueness queries for above k-th order ODE
is equivalent to similar queries posed for first order system of ODE.
Picard’s theorems on existence and uniqueness can be extended to
first order system of ODEs — a topic beyond the purview of this
course.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 94 / 126
Existence and Uniqueness for k-th order Linear ODE

Theorem
Consider the linear k-th order IVP
 Pk (i)
i=0 ai (x)y (x) = f (x) I ⊂ R
y (j) (x0 ) = yj for 0 ≤ j ≤ k − 1

for a given x0 ∈ I¯ and yj ∈ R such that ak (x) 6= 0 for all x ∈ I . If

ai , f : I¯ → R are continuous, for all 0 ≤ i ≤ k, then the IVP admits a
unique solution in I¯.

If f ≡ 0 and yj = 0, for all 0 ≤ j ≤ k − 1, then zero function is the

unique solution.
Denoting Ly (x) := ki=0 ai (x)y (i) (x), one can view L : C (I¯) → C (I¯)
P
as a linear map where C (I¯) is the set of all continuous real-valued
functions on I¯.
Show that C (I¯) forms a vector space over R (Exercise!).
T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 95 / 126
Null Space of the Linear ODE

Definition
The null space of L, denoted as N(L), is defined as the class of all
functions y ∈ C k (I¯) such that Ly = 0 where C k (I¯) is the set of all k-times
differentiable functions on I such that the k-th derivative can be
continuously extended to the end-points of I .

Example
All constant functions in C (I¯) is in the null space of the first order linear
d
L := dx .

Example
All linear functions of the form ax + b in C (I¯) is in the null space of the
d2
second order linear L := dx 2.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 96 / 126
Null Space is a subspace

Theorem
The null space, N(L), is a subspace of C (I¯).

Proof.
Obviously, the constant function zero is in N(L). Further, if u1 , . . . , um are
in N(L),i.e. Lui = 0 for 1 ≤ i ≤ m.Then

L(α1 u1 + . . . + αm um ) = α1 Lu1 + . . . + αm Lum = 0.

Thus, the linear combination of any solutions in N(L) is also in N(L).

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 97 / 126
Example

Example
Consider the second order differential operator Ly := y 00 + y . Observe that
sin x and cos x are in the null space of L. Also, any linear combination
α sin x + β cos x is also in the null space of L, for any choice of α, β ∈ R.

Example
Consider the third order differential operator Ly := y (3) − 2y 00 − y 0 + 2y .
Observe that e x , e −x and e 2x are in the null space of L. Also, any linear
combination α1 e x + α2 e −x + α3 e 2x is in the null space of L, for any
choice of αi ∈ R, for all i = 1, 2, 3.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 98 / 126
Linear Dependence and Independence of Solutions

Definition
A collection of functions u1 , . . . , um are said to be linearly independent on
I¯ if
α1 u1 + . . . + αm um = 0
implies αi = 0 for all i. Otherwise, the collection is said to be linearly
dependent.

Example
The two functions x and 2x are linearly dependent on [0, 1] because
α1 x + 2α2 x = 0 for the choice α1 = 2 and α2 = −1.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 99 / 126
Example

Example
The three functions sin x, 3 sin x and − sin x are linearly dependent on
[−1, 2] because α1 sin x + 3α2 sin x − α3 sin x = 0 for the choice
α1 = α2 = 1 and α3 = 4.

Example
The two functions x and x 2 are linearly independent on [0, 1]. For some
α1 and α2 , consider α1 x + α2 x 2 = 0. Differentiate w.r.t x to obtain
α1 + 2α2 x = 0 for all x ∈ [0, 1]. Then α1 x + 2α2 x 2 = 0 for all x ∈ [0, 1].
Now solving for α1 and α2 in the two equations α1 x + α2 x 2 = 0 and
α1 x + 2α2 x 2 = 0, we get α2 = 0 = α1 .

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 100 / 126
Dimension of the Null Space N(L)

Theorem
Let Ly (x) := ki=0 ai (x)y (i) (x) be the k-th order linear differential map
P
L : C (I¯) → C (I¯). Then its null space N(L) is a k-dimensional subspace of
C (I¯), i.e. there exists k linearly independent solutions to Ly = 0 and every
other solution can be expressed as a linear combination of the k solutions
with appropriate choice of constants.

Definition
The general solution of the k-th order linear, homogeneous ODE Ly = 0 is
the linear combination of k linearly independent solutions of Ly = 0.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 101 / 126
Example

Example
For the second order differential operator Ly := y 00 + y , sin x and cos x are
two linearly independent solutions and its general soluton is given as the
linear combination y (x) := α sin x + β cos x.

Example
For the third order differential operator Ly := y (3) − 2y 00 − y 0 + 2y , e x ,
e −x and e 2x are three linearly independent solutions of Ly = 0. Thus, its
general solution is the linear combination y (x) := α1 e x + α2 e −x + α3 e 2x .

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 102 / 126
Wronskian

Definition
For any collection of k functions u1 , . . . , uk in C (k−1) (I¯), the Wronskian of
the collection, denoted as W (u1 , . . . , uk ), is defined as the determinant of
the k × k matrix
 
 u1 (x) . . . u k (x) 
 u10 (x) 0
 
... uk (x) 
W (u1 , . . . , uk )(x) :=  .. ..

..
.


 (k−1). . 
(k−1)

 u (x) . . . uk (x). 
1

The Wronskian can be viewed as a map W (u1 , . . . , uk ) : I → R.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 103 / 126
Sufficiency Condition for Linear Independence

Theorem
If u1 and u2 are linearly dependent in I then W (u1 , u2 )(x) = 0 for all
x ∈ I . Equivalently, if W (u1 , u2 ) 6= 0 for some x ∈ I then u1 and u2 are
linearly independent.

Proof.
Suppose u1 and u2 are linearly dependent in I then there exists a non-zero
pair (α, β) such that αu1 (x) + βu2 (x) = 0. On differentiation, we have
the matrix equation
    
u1 (x) u2 (x) α 0
= .
u10 (x) u20 (x) β 0

Since we have a non-zero pair (α, β) as a solution, the determinant of the

matrix should be zero for all x ∈ I .

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 104 / 126
Converse Not True

Example
Consider u1 = x|x| and u2 = x 2
with derivative u10 = 2|x| and u20 = 2x.
Its Wronskian is zero everywhere.
But u1 and u2 are linearly independent
because αx|x| + βx 2 = 0 implies that
for x < 0, we get β − α = 0
and for x > 0, we get α + β = 0
implying that both α = β = 0.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 105 / 126
Converse is True for Linearly Independent Solutions
Theorem
The k solutions u1 , . . . , uk of the k-order linear, homogeneous ODE Ly = 0
in I is linearly dependent in I iff W (u1 , . . . , uk )(x) = 0 for some x ∈ I .

second order proof.

One way is already proved in Theorem 92! Conversely, suppose
W (u1 , , u2 )(x0 ) = 0 for some x0 ∈ I . Consider the equation
αu1 (x0 ) + βu2 (x0 ) = 0. We claim that at least either of α or β is
non-zero. On differentiation we have the additional equation
αu10 (x0 ) + βu20 (x0 ) = 0. Since W (u1 , , u2 )(x0 ) = 0, we have a non-zero
pair (α, β). Set y (x) = αu1 (x) + βu2 (x). Then Ly = 0 and
y (x0 ) = y 0 (x0 ) = 0. By the uniqueness of IVP, we have y ≡ 0 for non-zero
pair (α, β) giving the linear dependence of u1 and u2 .

Note that for linearly independent solutions of the linear ODE the
Wronskian is non-zero everywhere.
T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 106 / 126
Vanishing Property of Wronskian

Theorem
If u1 , . . . , uk are solutions of the k-order linear, homogeneous ODE Ly = 0
in I then the Wronskian W (u1 , . . . , uk ) is either identically zero on I or is
never zero on I .

Proof.
If the Wronskian vanishes at some point x0 ∈ I , then the proof of
Theorem 94 implied that u1 and u2 are linearly dependent. But
Theorem 92 implied that for linear dependent functions the Wronskian
vanishes everywhere!

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 107 / 126
Example

Example
The Wronskian of u1 (x) := x and u2 (x) := sin x is x cos x − sin x. This
Wronskian is non-zero, for instance at x = π, thus the functions are
linearly independent by Theorem 92. However, the Wronskian is zero at
x = 0.It is no contradiction to above theorem because x and sin x cannot
span solutions of a second order ODE (see assignment problem)!

Example
For the second order differential operator Ly := y 00 + y , sin x and cos x are
two linearly independent solutions because its Wronskian
 
 sin x cos x 
W (sin x, cos x)(x) :=   = − sin2 x − cos2 x = −1 6= 0
cos x − sin x 

for all x ∈ R.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 108 / 126
Example

Example
For the third order differential operator Ly := y (3) − 2y 00 − y 0 + 2y , e x ,
e −x and e 2x are three linearly independent solutions because its Wronskian
−x
 x
e 2x 
  
 e
 x e −x
 1 1 1 
x −x 2x
2e 2x  = e 2x  1 −1 2 
 
W (e , e , e )(x) :=  e −e 
 e x e −x 4e 2x   1 1 4 
= −6e 2x 6= 0

for all x ∈ R.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 109 / 126
Reduction of Order

This method is used to find a linearly independent solution corresponding

to a given solution.
Theorem
Let u1 be a non-trivial solution of the second order homogeneous linear
ODE a2 (x)y 00 + a1 (x)y 0 + a0 (x)y = 0. Then u2 := u1 (x)v (x) is a linearly
independent solution where
Z  Z 
1 a1 (x)
v (x) := exp − dx dx. (6.1)
u12 (x) a2 (x)

Proof: Set u2 (x) := u1 (x)v (x) where v shall be chosen appropriately.

Then by product rule we have

u20 = u1 v 0 + u10 v and u200 = u1 v 00 + 2u10 v 0 + u100 v .

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 110 / 126
Proof
Since u2 is required to be a solution to the given ODE, we consider

0 = a2 (x)u200 + a1 (x)u20 + a0 (x)u2

= a2 u1 v 00 + (2a2 u10 + a1 u1 )v 0 + (a2 u100 + a1 u10 + a0 u1 )v .

The coefficient of v is zero because u1 is given to be a solution. Thus, if v

is chosen as the solution to the ODE
dv 0
 0 
00 0 0 u 1 a1
a2 u1 v + (2a2 u1 + a1 u1 )v = 0 or 0 = − 2 +
v u 1 a2

then u2 is a solution to the given ODE. Then v is as given in (6.1). Now,

u1 and u2 are linearly independent because
   Z 
 = u12 (x)v 0 (x) = exp − a1 (x) dx 6= 0.
 u1 u 1 v 
W (u1 , u2 ) =  0
u1 u1 v 0 + u10 v  a2 (x)

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 111 / 126
Example

Example
Consider the ODE (x 2 + 1)y 00 − 2xy 0 + 2y = 0. Note that u1 (x) = x is a
solution to the ODE. Now
 Z  Z 2
−2x
Z
1 x +1 1
v (x) = exp − dx dx = dx = x − .
x2 x2 + 1 x2 x

Then u2 (x) := x(x − x −1 ) = x 2 − 1 and the general solution is

y (x) = α1 x + α2 (x 2 − 1).

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 112 / 126
Homogeneous Constant Coefficients

Theorem
A linear, homogeneous
Pk k-th order ODE with constant coefficients is given
(i)
by Ly := i=0 ai y = 0 and its characteristic equation (CE) is the k-th
degree polynomial of m, ki=0 ai mi = 0.
P

If (CE) admits k distinct real roots {mi }k1 then y (x) := ki=1 αi e mi x
P
is a general solution of the ODE Ly = 0.
If (CE) admits
P` ` repeated realP
roots m and the rest are distinct then
y (x) := ( i=1 αi x )e + ki=`+1 αi e mi x .
i−1 mx

If (CE) admits non-repeated pair of complex roots a ± ıb the rest are

distinct real roots then
y (x) := e ax (α1 sin bx + α2 cos bx) + ki=3 αi e mi x .
P

If (CE) admits repeated pair of ` complex roots a ± ıb then the

corresponding
P` part of general
P` solutioni−1
is written as
ax
e [ i=1 αi x i−1 sin bx + i=1 α`+i x cos bx].

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 113 / 126
Proof of Distinct Real Roots

Proof.
Observe that the derivative of exponential is constant multiple of
dk mx ] = mk e mx .
itself, i.e. dx k [e

Thus, it is natural to expect the solution of constant coefficients

linear ODE in the form of exponential.
Let y (x) = e mx be a solution of the ODE where m is to be
determined appropriately.
Using it in the ODE, we get ( ki=0 ai mi )e mx = 0.
P

Thus, the possible choices of m are the roots of the k degree

polynomial and the general solution is the linear combination of the
linearly independent solutions.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 114 / 126
Example

Example
Consider y 00 − 3y 0 + 2y = 0.
The CE is m2 − 3m + 2 = 0 whose roots are m1 = 1 and m2 = 2.
The roots are real and distinct.
The corresponding solutions are e x and e 2x .
They are linearly independent because
 x
e 2x

x 2x
 e
 = e 3x 6= 0.

W (e , e ) =  x
e 2e 2x 

Thus, the general solution is y (x) := α1 e x + α2 e 2x .

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 115 / 126
Example
Consider y (3) − 4y 00 + y 0 + 6y = 0.
The CE is m3 − 4m2 + m + 6 = 0 whose roots are m1 = −1, m2 = 2
and m3 = 3.
The roots are real and distinct.
The corresponding solutions are e −x , e 2x and e 3x .
They are linearly independent (exercise!).
Thus, the general solution is y (x) := α1 e −x + α2 e 2x + α3 e 3x .

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 116 / 126
Proof of Repeated Real Roots

Proof.
If m is a repeated root, say twice, then e mx is one solution.
Now, using the existing solution u1 := e mx , we seek a linearly
independent solution e mx v (x) for a suitable choice of v .
It can be shown that v (x) = x and xe mx is a linearly independent
solution.
For three repeated roots we have e mx , xe mx and x 2 e mx are the
linearly independent solutions.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 117 / 126
Example

Example
Consider the ODE y 00 − 6y 0 + 9y = 0.
The CE is m2 − 6m + 9 = 0 with two repeated roots m1 = m2 = 3.
The corresponding solution is e 3x .
A linear independent solution is xe 3x and the general solution is
y (x) := (α1 + α2 x)e 3x .

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 118 / 126
Example

Example
Consider y (3) − 4y 00 − 3y 0 + 18y = 0.
The CE is m3 − 4m2 − 3m + 18 = 0 whose roots are m1 = m2 = 3
and m3 = −2.
The two of the roots are repeated.
The corresponding solutions are e 3x and e −2x .
Thus, the general solution is y (x) := (α1 + α2 x)e 3x + α3 e −2x .

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 119 / 126
Example

Example
Consider y (4) − 5y (3) + 6y 00 + 4y 0 − 8y = 0.
The CE is m4 − 5m3 + 6m2 + 4m − 8 = 0 whose roots are
m1 = m2 = m3 = 2 and m4 = −1.
Some roots are repeated!
Thus, the general solution is y (x) := (α1 + α2 x + α3 x 2 )e 2x + α4 e −x .

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 120 / 126
Proof of Conjugate Complex Roots

Proof.
If a ± ıb are the conjugate pair of complex roots then
e (a+ıb)x = e ax (cos bx + ı sin bx) and e (a−ıb)x = e ax (cos bx − ı sin bx)
are complex solutions of the ODE.
We seek real linearly independent solutions.
We sum the above two complex solutions and divide by 2 to obtain
e ax cos bx.
Similarly, on subtraction and dividing by 2ı, we get e ax sin bx.
Both are real linearly independent solutions and the general solution is
e ax [α1 sin bx + α2 cos bx].

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 121 / 126
Example

Example
Consider the ODE y 00 + y = 0.
The CE is m2 + 1 = 0 with pair of conjugate complex roots m1 = ı
and m2 = −ı.
The general solution is y (x) := α1 sin x + α2 cos x.

Example
Consider y 00 − 6y 0 + 25y = 0.
The CE is m2 − 6m + 25 = 0 whose roots are m1 = 3 + ı4 and
m2 = 3 − ı4.
Thus, the general solution is y (x) := e 3x (α1 sin 4x + α2 cos 4x).

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 122 / 126
Example

Example
Consider y (4) − 4y (3) + 14y 00 − 20y 0 + 25y = 0.
The CE is m4 − 4m3 + 14m2 − 20m + 25 = 0 whose roots are
m1 = m2 = 1 + ı2 and m3 = m4 = 1 − ı2.
The complex pair of roots are repeated.
Thus, the general solution is

y (x) := e x [(α1 + α2 x) sin 2x + (α3 + α4 x) cos 2x] .

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 123 / 126
Example IVP: Linear Homogeneous

Example
Consider the IVP
 00
 y − 6y 0 + 25y = 0
y (0) = −3
y 0 (0) = −1.


The CE is m2 − 6m + 25 = 0 whose roots are m1 = 3 + ı4 and

m2 = 3 − ı4.
The general solution is y (x) := e 3x (α1 sin 4x + α2 cos 4x).
Then y 0 (x) = e 3x [(3α1 − 4α2 ) sin 4x + (4α1 + 3α2 ) cos 4x].
Using the initial condition y (0) = −3, we get α2 = −3.
Similarly, using y 0 (0) = −1, we get α1 = 2.
Thus the solution is y (x) := e 3x (2 sin 4x − 3 cos 4x).

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 124 / 126
Linear, Non-Homogeneous ODE

Theorem
A linear,
Pknon-homogeneous k-th order ODE is given as
(k)
Ly := i=0 ai (x)y = f (x). If yp is a given particular integral (not
involving arbitrary constants) of the non-homogeneous ODE then yc + yp
is the general solution of the non-homogeneous ODE where yc (called the
complementary function) is the general solution of the corresponding
homogeneous ODE (f ≡ 0).

Proof.
Any element of N(L) is denoted as yc . Then the general solution of
Ly = f is yc + yp , translation of the null space N(L) by yp .

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 125 / 126
Example

Example
Consider the non-homogeneous ODE y 00 + y = x. Its complementary
function is the solution corresponding to the homogeneous ODE
y 00 + y = 0, i.e. yc = α1 sin x + α2 cos x. A particular integral to
y 00 + y = x is yp (x) = x. Then y (x) = α1 sin x + α2 cos x + x is the
general solution for the non-homogeneous ODE.

T. Muthukumar & Akash Anand Ordinary Differential Equations June 16, 2023 126 / 126