LECTURE NOTES ON PARTIAL

DIFFERENTIAL EQUATIONS
Dr. E. Natarajan
Department of Mathematics
Indian Institute of Space Science and Technology
Thiruvananthapuram
thanndavam@iist.ac.in
Contents
1 Modeling partial dierential equations 5
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2 Vibrating string problem . . . . . . . . . . . . . . . . . . . . . 5
1.3 Heat equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.4 Well-posed PDE . . . . . . . . . . . . . . . . . . . . . . . . . 8
1.5 Solution of PDEs by Fourier transform . . . . . . . . . . . . 9
2 Second order linear PDEs 13
2.1 Classication . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.1.1 Denitions . . . . . . . . . . . . . . . . . . . . . . . . 13
2.2 Canonical forms . . . . . . . . . . . . . . . . . . . . . . . . . 14
2.2.1 Canonical form of hyperbolic PDE . . . . . . . . . . . 15
2.2.2 Canonical form of parabolic PDE . . . . . . . . . . . . 16
2.2.3 Canonical form of elliptic PDE . . . . . . . . . . . . . 17
3 Method of Separation of variables 19
3.1 Heat equation . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.2 Wave equation . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.3 Mathematical justication of the method . . . . . . . . . . . 23
4 First order partial dierential equations 25
4.1 Method of Characteristics . . . . . . . . . . . . . . . . . . . . 26
4.1.1 Transversality condition . . . . . . . . . . . . . . . . . 28
4.2 Lagranges Method . . . . . . . . . . . . . . . . . . . . . . . . 31
3
Chapter 1
Modeling partial dierential
equations
1.1 Introduction
Partial dierential equations is a relation between an unknown function and
its partial derivatives
F(x
1
, x
2
, ..., x
n
, u
x
1
, u
x
2
, ..., u
x
11
) = 0 (1.1)
where x
1
, x
2
, ..., x
n
are independent variables and u(x
1
, x
2
, ..x
n
) dependent
variable u
x
i
=
u
x
i
Order of the partial dierential equation is the order of the highest
partial derivative. u
tt
u
xx
= f(x, t) is a second order PDE whereas
u
t
+u
xxxx
= 0 is a fourth order PDE.
An equation is called linear if in Eq(1.1) F(.) is a linear function of the
unknown function u and its partial derivatives. For example x
2
u
x
+
xy u
y
+sin(x
2
+y
2
) u = x
3
is a linear equation whereas u
xx
2
+u
yy
2
= 0
is a non-linear equation. An equation is quasilinear if it is linear in
the highest partial derivative u
xx
+ u
yy
= |u|
2
u and semilinear if it
is non-linear in its unknown function u
xx
+u
yy
= u
3
.
1.2 Vibrating string problem
Let (x, t) measures the vertical displacement of the string from equilibrium.
Length of this piece
_
x
2
+
2
where = (x + x, t) (x, t) with
5
6 CHAPTER 1. MODELING PARTIAL DIFFERENTIAL EQUATIONS
( x,t)
x
Figure 1.1: Vertical displacement of string
Density (mass per unit length) is the mass of the string particle.

_
x
2
+
2
= x

1 +O
_

x
_
2
(1.2)
Displacement of the string is slight, so terms of order 2 for and

x
are neglected.
Mass x and acceleration in the vertical direction is

2

t
2
.
Force acting on the string particle is Tension T. This acts parallel to the
string i.e., tangentially. Tension is assumed to be uniform. This gives
T(x, t) = T.

i +

(x)

j
_
1 + (

(x))
2
(1.3)
Vertical component, T(x, t) =
T

x
1 + (

(x))
2
T

x
Net vertical tensile force is, T(x + x, t) T(x, t) T

2

x
2
+O(x
2
)
Using Newtons second law
T

2

x
2
x +O
_
x
2
_
= x

2

t
2
(1.4)
Dividing by x and letting x 0 we get

2

t
2
=
T

x
2
= c
2

2

x
2
where
c =

(dimensions of velocity).
1.3. HEAT EQUATION 7
Dirichlet
Neumann Robin
Figure 1.2: Dirichlet, Neumann and Robin boundary condition for left end
of the string
Let f be twice-dierentiable then (x, t) = f(x ct) satises the wave
equation. It is the wave propagating to the left at speed c and f(x +ct) is
the wave propagating to the right at speed c so the solution will be
(x, t) = f(x ct) +f(x +ct) where f and g are arbitrary.
In order to determine the motion completely (x, 0) = F(x) and

t
(x, 0) =
G(x) are given as initial displacement and initial velocity.
(x, t) =
F(x c t) +F(x +c t)
2
+
1
2 c
_
x+ct
xct
G() d (1.5)
This is known as the DAlemberts form of the solution. This says that to
determine (x, t) we need to know F and G within c t distance.
Types of boundary condition for the string tied at left boundary:
Dirichlet boundary condition (0, t) = c
Neumann boundary condition

x
(0, t) = d
Robin boundary condition (0, t) +

x
(0, t) = l
1.3 Heat equation
Consider heat owing through a uniform thin rod. Experiment shows that
Q = c u(x, t) V for a uniform cross section V = Ax between small
8 CHAPTER 1. MODELING PARTIAL DIFFERENTIAL EQUATIONS
Figure 1.3: Heat transfer in a uniform rod
portion x [a, b] where c is the concentration, is the density, u(x, t) is the
temperature and V is the small volume element.
Q(x, t) =
_
b
a
c u(x, t) Adx
dQ
dt
=
_
b
a
c
u(x, t)
t
Adx
This gives the rate at which heat accumulates in this segment. According
to Newtons Law of Cooling the rate of change of the temperature of an
object is proportional to the temperature gradient. Also heat energy ows
from warmer region to cooler region. This gives
dQ
dt
=

k [Rate in Rate out] A (1.6)
dQ
dt
=

k
_
u
x
(b, t)
u
x
(a, t)
_
A (1.7)
In the integral form this gives
dQ
dt
=
_
b
a

k A

2
u
x
2
dx. Equating similar terms
we get
_
b
a
_
c
u(x, t)
t

k

2
u
x
2
_
Adx = 0 (1.8)
Under suitable conditions it reduces to
u
t
= k

2
u
x
2
where k =

k
c
is the
thermal diusivity.
1.4 Well-posed PDE
Any physical phenomena can be modeled and it gives rise to partial dier-
ential equation. Suppose motion of a pendulum is modeled as a dierential
equation, denitely basic problem has a sure motion whereas the modeled
equation need not have solution. This doesnt contradict with the physics of
the problem but it says that the modeled equations are not sucient enough
to talk about the physics of the problem.
1.5. SOLUTION OF PDES BY FOURIER TRANSFORM 9
Jacques Hadamard came up with a set of conditions PDE has to satisfy
so that the equation can undoubtedly talk about the physics of the problem
and the PDEs satifying those conditions are known as Well-posed problems.
They are
(i) Existence of a solution
(ii) Uniqueness of the solution (there exists only one solution)
(iii) Solution is stable (continuous dependence of data)
We will illustrate the third point with an example.
u
t
= k u
xx
, < x < (1.9)
u(x, 1) = c sin
_
x
c

k
_
(1.10)
where c is small. This can be solved by taking u(x, t) = c sin
_
x
c

k
_
g(t)
with g(1) = 1. ( Try it!). The solution is u(x, t) = c e
(1t)
c
2
sin
_
x
c

k
_
. Note
that when c 0 when 0 < t < 1 the solution u(x, t) . A small change
in the initial condition leads to a large change in the solution thus killing
the stability of the problem.
1.5 Solution of PDEs by Fourier transform
Heat conduction in an innite rod
u
t
=
2
u
xx
, < x < , 0 < t <
u(x, 0) = f(x), < x <
Note: This problem can be solved by separation of variables if f(x) is dened
in nite interval or even if f is dened in innite interval provided if it is
periodic. Suppose f is dened in innite interval and aperiodic like f(x) =
e
|x|
. Let us take fourier transform of heat equation with respect to x.
F{u
t
} =
2
F{u
xx
}
_

u
t
e
i x
dx =
2
(i )
2
u
d
dt
_

u(x, t) e
i x
dx =
2

2
u
Under the assumption (u 0, u
x
0, x )
10CHAPTER 1. MODELING PARTIAL DIFFERENTIAL EQUATIONS
d u
dt
+
2

2
u = 0
The solution is u(x, t) = A()e

2
t
. We also need to transform the initial
condition. F(u(x, 0)) = F(f(x)) u(, 0) =

f(). Using this in the
solution we get A =

f(). Finally we get
u(, t) =

f() e

2
t
u(x, t) = f(x)
1
2

t
e

x
2
4
2
t
u(x, t) =
1
2

t
_

f() e

(x)
2
4
2
t
d
Case 1: If f is constant F then u(x, t) = F which violates the assumption
u , x .
Case 2: f(x) =
_
F, x > 0
0, x < 0
= F H(x)
u(x, t) =
F
2
_
1 + erf
_
x
2

t
__
, where erf(x) =
2

_
x
0
e

2
d.
In a more formidable way our solution can be written as
u(x, t) =
_

f() K( x; t) d
where
K( x; t) =
e

(x)
2
4
2
t
2

t
is called as Kernel of the integral. If K( x; t) ( x) as t 0 then
lim
t0
u(x, t) = lim
t0
_

f() K( x; t) d = f(x)
Let f(x) = (x x
0
). Then
u(x, t) =
_

( x
0
) K( x; t) d
= K(x
0
x; t) = K(x x
0
; t)
If initial temperature distribution is (xx
0
) then K(xx
0
; t) is the solution
to the heat equation. Diusion process smoothes out the solution.
1.5. SOLUTION OF PDES BY FOURIER TRANSFORM 11
Example 1.5.1. u
t
= u
xx
, < x < , t > 0, u(x, 0) = F(x), <
x < . By solving this problem using Fourier transform we get
u(x, t) =
1

4 t
_

F(y) e

(xy)
2
4 t
dy
where e
(xy)
2
4 t
> 0 x, y. For t however small u depends on F(x), <
x < which means the speed of propagation is innite. But energy cannot
propagate faster than speed of light. This model of heat equation has a aw.
Chapter 2
Second order linear PDEs
2.1 Classication
Let us consider the general form of second order linear PDEs in two variables
A(x, y) u
xx
+ 2 B(x, y) u
xy
+C(x, y) u
yy
= (x, y, u
x
, u
y
, u) (2.1)
where the left hand side of Eq (2.1) denotes the principal part of the PDE.
2.1.1 Denitions
(a) Equation 2.1 is said to be hyperbolic if B
2
AC > 0
(b) Equation 2.1 is said to be parabolic if B
2
AC = 0
(c) Equation 2.1 is said to be elliptic if B
2
AC < 0
What do these classication do? Why is it needed?
These classications can be related with equation of general conic section
a x
2
+ 2 b xy +c y
2
+d x +e y +f = 0 (2.2)
(a) Equation 2.2 represents eqn of hyperbola if b
2
ac > 0
(b) Equation 2.2 represents eqn of parabola if b
2
ac = 0
(c) Equation 2.2 represents eqn of ellipse if b
2
ac < 0
The need for this classication is any second order PDEs in two variables
can be reduced to either hyperbolic, parabolic or elliptic. Once the behaviour
13
14 CHAPTER 2. SECOND ORDER LINEAR PDES
of the solutions of hyperbolic,parabolic and elliptic PDEs are known then
classication to one of the three forms helps in understanding the problem
apriori.
Examples:
u
tt
= c
2
u
xx
where A = 1, B = 0, C = c
2
gives B
2
AC > 0
then the equation is said to be hyperbolic.
u
t
= k u
xx
where A = 0, B = 0, C = k gives B
2
AC = 0 then
the equation is said to be parabolic.
u
xx
+u
yy
= 0 where A = 1, B = 0, C = 1 gives B
2
AC < 0 then
the equation is said to be elliptic.
2.2 Canonical forms
Any second order PDE in two variables can be reduced to canonical
form of hyperbolic, parabolic and elliptic PDEs under suitable trans-
formation. Let us convert the PDE Equation 2.1 using the coordinate
transformation (x, y), (x, y). Also take w(, ) = u(x(, ), y(, ))
u
x
= w

x
+w

x
u
y
= w

y
+w

y
Similarly compute u
xx
, u
yy
and u
xy
in terms of w

, w

and w

. After
substitution of these terms into the Equation 2.1 gives
a(, ) w

+ 2 b(, ) w

+c(, ) w

= (w

, w

, w, , )
where
a(, ) = A
x
2
+ 2 B
x

y
+C
y
2
b(, ) = A
x

x
+B(
x

y
+
y

x
) +C
y

y
c(, ) = A
x
2
+ 2 B
x

y
+C
y
2
We need a justication that the form of the PDE remains invari-
ant even after the coordinate transformation i.e., hyperbolic remains
as hyperbolic, parabolic remains parabolic and vice versa. It can be
observed that
_
a b
b c
_
=
_

x

y

x

y
_ _
A B
B C
_ _

x

x

y

y
_
2.2. CANONICAL FORMS 15
Taking the determinant on both sides gives
b
2
a c = (
x

x
)
2
_
B
2
AC
_
where (
x

x
)
2
is nothing but the square of the Jacobian J
2
.
The canonical form of the PDE after transformation reduces to
(i) w

= G(w

, w

, , ) if it is hyperbolic.
(ii) w

= G(w

, w

, , ) or w

= G(w

, w

, , ) if it is parabolic.
(iii) w

+w

= G(w

, w

, , ) if it is elliptic.
2.2.1 Canonical form of hyperbolic PDE
If A = C = 0 then u
xy
=
(u
x
, u
y
, u, x, y)
2B
in which case there is no
need of transformation.
Let us assume A = 0 and in order to make b
2
a c > 0 we take
a = c = 0. This gives
A
x
2
+ 2 B
x

y
+C
y
2
= 0
A
x
2
+ 2 B
x

y
+C
y
2
= 0
After factorizing
1
A
_
A
x
+
_
B

B
2
AC
y
__ _
A
x
+
_
B

B
2
+AC
y
__
= 0
A
x
+
_
B +

B
2
AC
_

y
= 0
A
x
+
_
B

B
2
AC
_

y
= 0
A
x
+
_
B +

B
2
AC
_

y
= 0
A
x
+
_
B

B
2
AC
_

y
= 0
We get totally two equations for and two equations for but we need
one solution for and one for . We take
A
x
+
_
B +

B
2
AC
_

y
= 0
A
x
+
_
B

B
2
AC
_

y
= 0
16 CHAPTER 2. SECOND ORDER LINEAR PDES
The above equations are itself PDE in terms of (x, y). On the
characteristic curves (x, y) = constant we get d = 0
x
dx+
y
dy =
0. Similarly d = 0
x
dx +
y
dy = 0. Equating the like terms gives
dy
dx
=

y
=
B +

B
2
AC
A
dy
dx
=

y
=
B

B
2
AC
A
Solving the above two ordinary dierential equations we get two curves
(x, y) = C
1
and (x, y) = C
2
. Thus we get the coordinate transfor-
mations by which we can reduce the PDES to canonical form.
Example 2.2.1. Determine the canonical form of u
tt
c
2
u
xx
= 0
B
2
AC = c
2
> 0 so the eqn is hyperbolic.
We get
dx
dt
= c and
dx
dt
= c
This gives (x, y) = x ct and (x, y) = x + ct. Using this we get the
canonical form w

= 0 w(, ) = f() +g().
So u(x, y) = f(x ct) +g(x +ct).
2.2.2 Canonical form of parabolic PDE
In order to make b
2
a c = 0 we need either b = c = 0 or b = a = 0 Let
us take c = 0 because b = 0 will be automatically forced by parabolic
condition.
Let us assume A = 0. This gives
A
x
2
+ 2 B
x

y
+C
y
2
= 0
Substituting C =
B
2
A
and factorizing we get
A
x
2
+ 2 B
x

y
+
B
2
A

y
2
= 0
1
A
(A
x
+B
y
)
2
= 0
A
x
+B
y
= 0 ( A = 0)
2.2. CANONICAL FORMS 17
On the characteristic curve = const we have d = 0
x
dx +

y
dy = 0. This gives the transformation (x, y) from

y
=
dy
dx
=
B
A
.
Now we have the freedom to choose the transformation (x, y) such
that
x

y

y

x
= 0. Thus we can reduce the PDE to its canonical
form.
Example 2.2.2. Determine the canonical form of x
2
u
xx
2 xy u
xy
+
y
2
u
yy
+xu
x
+y u
y
= 0.
B
2
AC = 0 so the eqn is parabolic.
We get
dy
dx
=
y
x
xy = const
This implies (x, y) = xy. Now let us take (x, y) = x so that
x

x
= x = 0. You can choose any satisfying above condition.
The canonical form is
2
w

+ w

= 0. Solving this nally we get

w(, ) = ln() g() +f() u(x, y) = ln(x) g(xy) +f(xy).
2.2.3 Canonical form of elliptic PDE
For the elliptic case we need to make b
2
a c = 0, we can make a =
c and b = 0. This gives
A
x
2
+ 2 B
x

y
+C
y
2
= A
x
2
+ 2 B
x

y
+C
y
2
A
x

x
+B(
x

y
+
y

x
) +C
y

y
= 0
This is in itself coupled PDE in (x, y) and (x, y). Let us try to convert
these two equations in complex form by taking = + i . Then we
can express the above two equations as
A(
x
2

x
2
) + 2 B(
x

y
) +C (
y
2

y
2
) = 0
A
x
i
x
+B(
x
i
y
+
y
i
x
) +C
y
i
y
= 0
In terms of ,
A
x
2
+ 2 B
x

y
+C
y
2
= 0,
18 CHAPTER 2. SECOND ORDER LINEAR PDES
Factorizing gives
_
A
x
+
_
B +i

AC B
2
_

y
__
A
x
+
_
B i

AC B
2
_

y
_
= 0
On = constant we have d = 0
x
dx +
y
dy = 0.
Equating

y
=
dy
dx
to the factorization above gives
dy
dx
=
B +i

AC B
2
A
dy
dx
=
B i

AC B
2
A
This gives (x, y) and (x, y) both are complex functions. From this
we can extract real valued function (x, y) and (x, y). We will try to
understand by an example
Example 2.2.3. Determine the canonical form of u
xx
+x
2
u
yy
= 0
B
2
AC = x
2
< 0 so the eqn is elliptic
We have
dy
dx
=
i x
2
,
dy
dx
=
i x
2
After solving these two ODES we get
(x, y) =
x
2
2
+i y and (x, y) =
x
2
2
i y
Taking =
+
2
and =

2 i
gives =
x
2
2
and = y. Using
this and the canonical form reduces to w

+w

=
w

2
.
Chapter 3
Method of Separation of
variables
3.1 Heat equation
Let us consider the heat conduction problem
u
t
k u
xx
= 0 0 < x < L, t > 0 (3.1)
u(0, t) = u(L, t) = 0, t 0 (3.2)
u(x, 0) = f(x), 0 x L (3.3)
Eq (3.2) and (3.3) implies f(0) = f(L) = 0. Let us assume the solution
u(x, t) = X(x) T(t). This gives XT

= k X

T.
T

k T
=
X

X
= gives
d
2
X
dx
2
= X, 0 < x < L,
dT
dt
= k T, t > 0
First let us solve the ODE for X(x). We need to nd the boundary con-
dition for X(x) at x = 0 and x = L. Using u(0, t) = 0 X(0) T(t) =
0 X(0) = 0 and u(L, t) = 0 X(L) T(t) = 0 X(L) = 0.
d
2
X
x
2
+ X = 0, 0 < x < L (3.4)
X(0) = X(L) = 0 (3.5)
This is an eigenvalue problem with eigenvalue and eigenfunction X

.
19
20 CHAPTER 3. METHOD OF SEPARATION OF VARIABLES
( < 0) then X(x) = a cosh(

x)+b sinh(

x) with X(0) = X(L) =

0 X(x) = 0.
( = 0) then X(x) = a +b x with X(0) = X(L) = 0 X(x) = 0.
( > 0) then X(x) = a cos(

x) +b sin(

x) with X(0) = X(L) = 0

a = 0 and sin

L = 0 =
_
n
L
_
2
, n = 1, 2, 3..
Solving ODE for T(t) gives T
n
(t) = e
k (
n
L
)
2
t
, n = 1, 2, 3, .... For
each
n
the product of X
n
(x)T
n
(t) is a solution so by the superposition
principle we get u(x, t) =

n=1
A
n
sin
nx
L
e
k (
n
L
)
2
t
. In order to deter-
mine the constants A
n
we need to use the condition u(x, 0) = f(x).
This gives

n=1
A
n
sin
nx
L
= f(x), Multiplying both sides by sin
mx
L
and integrating from 0 to L gives A
m
=
2
L
_
L
0
sin
mx
L
f(x) dx, m =
1, 2, 3, ...
Let us solve a problem with some simple f(x).
Example 3.1.1. Consider the problem
u
t
u
xx
= 0, 0 < x < , t > 0
u(0, t) = u(, t) = 0, t 0
u(x, 0) =
_
x, 0 x

2
x,

2
x
We get the solution
u(x, t) =

n=1
A
n
sin(nx) e
k n
2
t
Using the condition of u(x, 0) we can obtain A
n
=
4
n
2
sin
n
2
. The
nal solution is
u(x, t) =
4

n=1
(1)
n+1
(2n 1)
2
sin((2n 1) x) e
(2n1)
2
t
3.2. WAVE EQUATION 21
Our solution is an innite series of functions. So in order for the solu-
tion to be well dened we need to discuss the convergence of the solution.
We can use Weierstrass M-test to check the convergence of series of
functions

n=1
u
n
(x, t) where we try to bound u
n
(x, t) < M
n
then if the
series of real numbers

n=1
M
n
converges then the real series

n=1
u
n
(x, t)
converges uniformly. In general

x

n=1
u
n
(x, t) =

n=1

x
u
n
(x, t) may
not be equal so the convergence of

n=1
u
n
t
and

n=1

2
u
n
x
2
should also be
veried. I will leave it as an exercise.
3.2 Wave equation
Let us consider the problem
u
tt
c
2
u
xx
= 0, 0 < x < L, t > 0
u
x
(0, t) = u
x
(L, t) = 0, 0 x L
u(x, 0) = f(x), u
t
(x, 0) = g(x), 0 x L
Compatibility conditions
f

(0) = f

(L) = g

(0) = g

(L) = 0
Let u(x, t) = X(x) T(t) which gives X T

= c
2
X

T
T

c
2
T
=
X

X
=
From this we get two ordinary dierential equations
X

+ X = 0, T

+ c
2
T = 0
This is an eigenvalue problem. The boundary condition gives u
x
(0, t) =
0 X

(0) T(t) = 0 X

(0) = 0. Similarly X

(L) = 0.
( < 0) then X(x) = a cosh(

x)+b sinh(

x) with X

(0) = X

(L) =
0 b = 0 and =
_
n
L
_
2
which cannot happen since is nega-
tive.
22 CHAPTER 3. METHOD OF SEPARATION OF VARIABLES
( = 0) then X(x) = a +b x with X

(0) = X

(L) = 0 X
0
(x) = const.
( > 0) then X(x) = a cos(

x) + b sin(

x) with X

(0) = X

(L) =
0 b = 0 and sin

L = 0 =
_
n
L
_
2
, n = 1, 2, 3....
X
n
(x) = a
n
cos
_
n x
L
_
.
Solving the equation T

+ c
2
T = 0
( > 0) then T
n
(t) =
n
cos(
_

n
c
2
t) +
n
sin(
_

n
c
2
t), n = 1, 2, 3., .
( = 0) then T

= 0 T
0
(t) =
0
+
0
t.
Finally u(x, t) = X
0
(x) T
0
(t) +

n=1
X
n
(x) T
n
(t)
u(x, t) =
A
0
+B
0
t
2
+

n=1
_
A
n
cos
_
cnt
L
_
+B
n
sin
_
cnt
L
__
cos
n x
L
, n =
1, 2, 3, ..
Example 3.2.1. Consider the problem
u
tt
4 u
xx
= 0, 0 < x < 1, t > 0
u
x
(0, t) = u
x
(1, t) = 0, t 0
u(x, 0) = cos
2
( x), 0 x 1
u
t
(x, 0) = sin
2
( x) cos( x), 0 x 1
Using the solution of the previous problem
u(x, t) =
A
0
+B
0
t
2
+

n=1
(A
n
cos(2n t) +B
n
sin(2n t)) cos(n x)
Using the conditions of intial displacement and intial velocity we get,
u(x, 0) =
A
0
2
+

n=1
A
n
cos(n x) = cos
2
x =
1 + cos(2 x)
2
This gives A
0
= 1, A
2
=
1
2
, A
n
= 0, n = 0, 2
u
t
(x, 0) =
B
0
2
+

n=1
B
n
(2 n) cos(n x) = sin
2
( x) cos( x) =
cos x
4

cos 3 x
4
3.3. MATHEMATICAL JUSTIFICATION OF THE METHOD 23
This gives B
1
=
1
8
, B
3
=
1
24
, B
n
= 0, n = 1, 3
u(x, t) =
1
2
+
1
8
sin(2 t) cos( x)+
1
2
cos(4 t) cos(2 x)
1
24
sin(6 t) cos(3 x).
Example 3.2.2.
u
tt
u
xx
= cos(2 x) cos(2 t), 0 < x < 1, t > 0
u
x
(0, t) = u
x
(1, t) = 0, t 0
u(x, 0) = cos
2
( x), 0 x 1
u
t
(x, 0) = 2 cos( x), 0 x 1
We have X
n
(x) = cos(n x),
n
= (n)
2
, n = 0, 1, 2, .... Let us as-
sume the solution as
u(x, t) =
T
0
(t)
2
+

n=1
T
n
(t) cos(n x)
Substituting this expression in the PDE and using the initial condi-
tions nally we get the solution
u(x, t) =
1
2
+
_
cos(2 t)
2
+
t + 4
4
sin(2 t)
_
cos(2 x).
3.3 Mathematical justication of the method
The method of separation of variables has limitations and the points
to be noted before applying the method are as follows.
(a) In the partial dierential equation Lu = f(x, y)
_
Ex : For wave equation L =

2
t
2
+

2
x
2
_
L must be separable.
(x, y) such that
L(X(x) Y (y))
(x, y) X(x) Y (y)
= F(x) G(y) for some F and
G functions of x and y.
(Ex: L(u) =

2
u
x
2
+

2
u
xy
+

2
u
y
2
, substituting u(x, y) = X(x) Y (y)
L(X(x) Y (y)) = X

Y +X

+X Y

cannot be written as F(x)+

G(y). So the PDE is not separable.
24 CHAPTER 3. METHOD OF SEPARATION OF VARIABLES
(b) All initial and boundary conditions must be on lines x = const
and y = const. (Ex: Suppose a domain not rectangular with sides
parallel to x and y axes.)
(c) Linear operator boundary conditions x = const involve no partial
derivative of u w.r.t x. Similarly y = const involve no partial
derivative of u w.r.t y. (Ex: Suppose at x = 0 the condition is
given as
u
x
+
u
y
= 0 ).
Chapter 4
First order partial
dierential equations
In this chapter we will see the methods to solve rst order PDES which
are Quasilinear.
a(x, y, u) u
x
+b(x, y, u) u
y
= c(x, y, u) (4.1)
Let us start solving this PDE
Example 4.0.1.
u
x
= c
0
u +c
1
(x, y), u(0, y) = y (4.2)
Eventhough this a PDE since there is no term with u
y
this can be looked
as a rst-order ODE and then solved by method of integrating factors.
dy
dx
+P(x) y = Q(x)
u(x, y) e
c
0
x
=
_
x
0
c
1
(, y) e
c
0

d + T(y). By using the condition

u(0, y) = y. We nd that T(y) = y. So this problem has a unique
solution.
In solving the above problem by xing y = const we see that the
PDE gets converted to set of ODEs on each y = const line. So we
have solved the set of ODESs. These are solutions of ODEs on the
line y = const. We then call the line as characteristic curve and the
solution as characteristic solution.
25
26CHAPTER 4. FIRSTORDER PARTIAL DIFFERENTIAL EQUATIONS
X
Y
Characteristic curves
Initial Curve
y = const
y=const
Now it is not always true that a rst-order PDE with given condition
will have a solution which is unique. Let us see few examples on this
issue.
Example 4.0.2. Consider the PDE u
x
= c
0
u with u(x, 0) = 2 x.
The solution is u(x, y) = e
c
0
x
T(y) by using the condition we see
that T(0) = 2 xe
c
0
x
which is impossible. So the PDE has no solution.
Example 4.0.3. Consider the same PDE u
x
= c
0
u with u(x, 0) =
2 e
c
0
x
. we see that T(0) = 2 which means there are innitely many
functions T(y) satisfying the above condition. So the PDE has innitely
many solutions.
4.1 Method of Characteristics
Consider the quasi-linear rst-order partial dierential equation
a(x, y, u) u
x
+b(x, y, u) u
y
= c(x, y, u)
The initial condition is given on some curve (s) : (x
0
(s), y
0
(s), u
0
(s))
in the parametric form where s (, ). This can be written as
(a, b, c) . (u
x
, u
y
, 1) = 0 where (u
x
, u
y
, 1) is normal to the surface
4.1. METHOD OF CHARACTERISTICS 27
Initial curve
Characteristic curve
x
y
u
u(x, y) u = 0 on the (x, y, u) plane. Therefore (a, b, c) lies in the
tangent plane. So we get the equations
dx
dt
= a(x(t), y(t), u(t))
dy
dt
= b(x(t), y(t), u(t))
du
dt
= c(x(t), y(t), u(t))
We call this set of ODEs as characteristic equation. Solving this set
of ODEs gives characteristic curves (x(t, s), y(t, s), u(t, s)) where each
curve emanates from dierent points on (s) for the given initial con-
dition x(0, s) = x
0
(s), y(0, s) = y
0
(s), u(0, s) = u
0
(s).
Example 4.1.1. Solve the PDE u
x
+u
y
= 2, u(x, 0) = x
2
The set of characteristic equations are
dx
dt
= 1,
dy
dt
= 1,
du
dt
= 2
where x(0, s) = s, y(0, s) = 0, u(0, s) = s
2
. Solving this set we get
x(t, s) = t + s, y(t, s) = t, u(t, s) = 2 t + s
2
. This gives u(x, y) =
28CHAPTER 4. FIRSTORDER PARTIAL DIFFERENTIAL EQUATIONS
2 y +(x y)
2
. To see the characteristics in the x y plane we can use
the ODE
dy
dx
=
b(x, y)
a(x, y)
which gives y = x +c for the current problem.
4.1.1 Transversality condition
In order for the PDE to have a unique solution near the initial curve
(s) the transformation x(t, s), y(t, s) has to be invertible. That is the
jacobian should be non-zero on the points of initial curve .
(x, y)
(t, s)
=

x
t
y
t
x
s
y
s

a b
(x
0
)
s
(y
0
)
s

= 0 Geometrically this means

projection of on the x y plane is tangent at this point to the
projection of the characteristic.
Example 4.1.2. Solve the equation u
x
= 1 subject to u(0, y) = g(y).
dx
dt
= 1,
dy
dt
= 0,
du
dt
= 1
where x(0, s) = 0, y(0, s) = s, u(0, s) = g(s). Solving this set we
get x(t, s) = t, y(t, s) = s, u(t, s) = t + g(s). This gives u(x, y) =
x + g(y). If the initial condition is changed to u(x, 0) = h(x). we get
where x(0, s) = s, y(0, s) = 0, u(0, s) = h(s). Solving for this condi-
tion we get x(t, s) = t +s, y(t, s) = 0, u(t, s) = t +h(s). The transver-
sality condition

1 0
1 0

= 0. The solution has a problem with unique-

ness. The characterisitc curves for this equation is
dy
dx
= 0 y = c.
The initial curve (s) is the x axis which is also the characteristic curve
on y = 0. So the initial curve and the characteristic curve coincides.
So we loose the uniqueness near the initial curve.
Example 4.1.3. Solve the equation u
x
+ u
y
+ u = 1 subject to u =
sinx, on y = x +x
2
, x > 0.
dx
dt
= 1,
dy
dt
= 1,
du
dt
= 1 u
where x(0, s) = s, y(0, s) = s +s
2
, u(0, s) = sins. Solving this set we
get x(t, s) = t+s, y(t, s) = t+s+s
2
, u(t, s) = 1+(sins 1) e
t
. This
gives u(x, y) = 1 +
_
sin

y x 1
_
e
(x

yx)
. This solution is valid
4.1. METHOD OF CHARACTERISTICS 29
5 4 3 2 1 0 1 2 3 4 5
10
5
0
5
10
15
20
25
30
Initial curve
characteristic
curves
in the domain D = {(x, y)|0 < x < y} {(x, y)|x 0 and x +x
2
< y}
The transversality condition

1 1
1 1 + 2s

= 2 s = 0 when s = 0. If you
observe clearly the solution u is not dierentiable at the origin. The
characteristics are
dy
dx
= 1 y = x + c. We can observe that the
Slope of the initial curve at the origin = Slope of the characteristic
at the origin. When choosing s we have omitted s =

y x. So
the solution near the curve in the region {(x, y)|x < 0 and y = x +x
2
}
gives non-uniqueness of the solution.
Example 4.1.4. Solve the equation y u
x
+xu
y
= u subject to u(x, 0) =
(x).
dx
dt
= y,
dy
dt
= x,
du
dt
= u
where x(0, s) = s, y(0, s) = 0, u(0, s) = (s). Solving this set we get
x(t, s) = s cos t, y(t, s) = s sint, u(t, s) = e
t
(s). This gives u(x, y) =

_
_
x
2
+y
2
_
e
tan
1
(
y
x
)
. The transversality condition

1 s
1 0

= s =
0 when s = 0. In choosing
_
_
x
2
+ y
2
_
we have assumed x > 0.
Each characteristic intersects the initial curve twice.
30CHAPTER 4. FIRSTORDER PARTIAL DIFFERENTIAL EQUATIONS
2 1.5 1 0.5 0 0.5 1 1.5 2
2
1.5
1
0.5
0
0.5
1
1.5
2
Initial curve
characteristic
curve
Example 4.1.5. Solve the equation u
x
+3 y
2
3
u
y
= 2 subject to u(x, 1) =
1 +x.
dx
dt
= 1,
dy
dt
= 3 y
2
3
,
du
dt
= 2
where x(0, s) = s, y(0, s) = 1, u(0, s) = 1 + s. Solving this set we
get x(t, s) = t + s, y(t, s) = (t + 1)
3
u(t, s) = 2 t + s + 1. This gives
u(x, y) = x + y
1
3
. The transversality condition

1 3
1 0

= 3 = 0. The
characteristic equation
dy
dt
= 3 y
2
3
with y(0, s) = 1 is not Lipchitz con-
tinuous at the origin. This gives y = t
3
and y = 0 as solutions. So this
does not have unique solution. Hence y = 0 is also a characterisitic.
We also get y = (x s + 1)
3
. For each point s on y = 1 we get dif-
ferent characteristic curve which intersects with another characteristic
y = 0. Thus u may not have unique solution there. Hence the solution
u(x, y) = x +y
1
3
is singular on the x axis.
4.2. LAGRANGES METHOD 31
y=1
Initial curve
Characteristic
curve
4.2 Lagranges Method
Consider the quasilinear PDE a(x, y, u) u
x
+b(x, y, u) u
y
= c(x, y, u).
The characteristic equations are
dx
dt
= a(x, y, u),
dy
dt
= b(x, y, u),
du
dt
= c(x, y, u)
From this we can extract two set of equations
dy
dx
=
b(x, y, u)
a(x, y, u)
,
du
dx
=
c(x, y, u)
a(x, y, u)
We get two family of curves (x, y, u) = , (x, y, u) = . So the
general solution to the quasilinear PDE is F((x, y, u), (x, y, u)) = 0.
Lemma 4.2.1. Let = (x, y, u), = (x, y, u). If f(, ) = 0 then
u satises
u
x
(, )
(y, u)
+
u
y
(, )
(u, x)
=
(, )
(x, y)
32CHAPTER 4. FIRSTORDER PARTIAL DIFFERENTIAL EQUATIONS
Proof. We know that f(, ) = 0. Dierentiating with respect to x
and y we get
f

x
+

u
u
x
_
+
f

x
+

u
u
x
_
= 0
f

y
+

u
u
y
_
+
f

y
+

u
u
y
_
= 0

x
+u
x

u

x
+u
x

y
+u
y

u

y
+u
y

= 0
which gives
u
x
(, )
(y, u)
+
u
y
(, )
(u, x)
=
(, )
(x, y)
Theorem 4.2.1. The general solution of PDE a(x, y, u) u
x
+b(x, y, u) u
y
=
c(x, y, u) is f(, ) = 0, where (x, y, u) = c
1
, (x, y, u) = c
2
are the
solution curves of
dx
a(x, y, u)
=
dy
b(x, y, u)
=
du
c(x, y, u)
Proof. Since (x, y, u) = c
1
and (x, y, u) = c
2
d =
x
dx +
y
dy +
u
du = 0
d =
x
dx +
y
dy +
u
du = 0
Using
dx
a(x, y, u)
=
dy
b(x, y, u)
=
du
c(x, y, u)
a
x
+ b
y
+c
u
= 0
a
x
+b
y
+c
u
= 0
Solving for a, b, c we get
a
(,)
(y,u)
=
b
(,)
(u,x)
=
c
(,)
(x,y)
The proof can be deduced by using the above lemma.
4.2. LAGRANGES METHOD 33
Example 4.2.1. Find the general solution of xu
x
+ y u
y
= u. The
set of equations are
dx
x
=
dy
y
=
du
u
We get (x, y, u) =
y
x
= c
1
, (x, y, u) =
u
x
= c
2
. Thus the general
solution is f
_
y
x
,
u
x
_
= 0. This can be written as u(x, y) = xg
_
y
x
_
.
Example 4.2.2. Find the general solution of x
2
u
x
+y
2
u
y
= (x+y) u.
The set of equations are
dx
x
2
=
dy
y
2
=
du
(x +y) u
dx
x
2
=
dy
y
2
gives
y x
xy
= c
1
and
dx dy
x
2
y
2
=
du
(x +y) u

x y
u
= c
2
.
The general solution of the above PDE is f
_
y x
xy
,
x y
u
_
= 0.
Example 4.2.3. Find the general solution of PDE u (x+y) +u (x
y) u
y
= x
2
+y
2
, u = 0on y = 2x. The set of equations are
dx
u (x +y)
=
dy
u (x y)
=
du
x
2
+y
2
y dx +xdy u du
0
=
xdx y dy u du
0
d
__
xy
u
2
2
__
= 0, d
_
x
2
y
2
u
2
2
_
= 0 which gives u
2
x
2
+y
2
= c
1
and 2 xyu
2
= c
2
. The general solution is f
_
x
2
+y
2
u
2
, 2 xy u
2
_
=
0. Using the initial condition u = 0 on y = 2x. We get 4 c
1
= 3 c
2

7 u
2
= 6 xy + 4 (x
2
y
2
).
Tutorial I
Exercise 4.2.1. Using DAlemberts form of solution for wave equation
nd the solution of
(a)

2
u
t
2
c
2

2
u
x
2
= 0, < x < , t > 0
u(x, 0) = e
x
2
, < x <
u
t
(x, 0) = 0, < x < (4.3)
(b)

2
u
t
2
9

2
u
x
2
= 0, < x < , t > 0
u(x, 0) =
_
1, if |x| 2
0, if |x| > 2
u
t
(x, 0) =
_
1, if |x| 2
0, if |x| > 2
(4.4)
(i) Find u(0,
1
6
)?
(ii) Find large time behaviour of the solution lim
t
u(, t)? for xed
R.
Exercise 4.2.2. Thermal conductivity of an aluminium alloy is 1.5
W/cm K. Calculate the steady state temperature of an aluminium bar
of length 1m (insulated along the sides) with its left end xed at 20

C
and right end xed at 30

C.
35
36CHAPTER 4. FIRSTORDER PARTIAL DIFFERENTIAL EQUATIONS
Exercise 4.2.3. (a) Show that the function u(x, t) = e

k
2
t
c
sin(x) is
a solution to the homogeneous heat equation c
u
t
k

2
u
x
2
= 0,
0 < x < l, t.
(b) What values of will cause u to also satisfy homogeneous Dirichlet
conditions at x = 0 and x = l?
Exercise 4.2.4. Classify the PDE and reduce to its canonical form
(a)

2
u
x
2
2 sinx

2
u
xy
cos
2
x

2
u
y
2
cos x
u
y
= 0
(b)
y
5

2
u
x
2
y

2
u
y
2
+ 2
u
y
= 0, y > 0
(c)

2
u
x
2
+ (1 +y
2
)
2

2
u
y
2
2y (1 +y
2
)
u
y
= 0
Exercise 4.2.5. Consider the equation x

2
u
x
2
y

2
u
y
2
+
1
2
_
u
x

u
y
_
=
0
(a) Find the domain where the equation is elliptic and the domain
where it is hyperbolic.
(b) For each of the above two domains, nd the corresponding canon-
ical transformation.
Exercise 4.2.6. Using Separation of variables nd the solution to the
following PDEs
(a)

2
u
t
2
=

2
u
x
2
, 0 < x < , t > 0
u(0, t) = u(, t) = 0, t 0
u(x, 0) = sin
3
x, 0 x
u
t
(x, 0) = sin2x, 0 x (4.5)
4.2. LAGRANGES METHOD 37
(b) Solve the heat equation
u
t
= 12

2
u
x
2
, 0 < x < , t > 0 subject to
the following boundary and initial condition
u
x
(0, t) =
u
x
(, t) = 0, t 0
u(x, 0) = 1 + sin
3
x, 0 x (4.6)
Find lim
t
u(x, t) for all 0 < x < and explain physical interpre-
tation of your result.
(c) Consider the heat equation
u
t

2
u
x
2
= 0, x R, t 0. Find
the transformation (x, t) by solving an ODE of the form ().
Tutorial II
Exercise 4.2.7. Solve the PDE subject to the given Cauchy condition
on the Cauchy data curve
(1) 2 u
x
5 u
y
= 4, subject to the Cauchy condition u(x, 0) = x.
(2) u
x
+ 3 u
y
= u + 2, subject to the Cauchy condition u(0, y) = y.
(3) xu
x
+ 2y u
y
= 3 u, subject to the Cauchy condition u(1, y) =
cos y for y > 1 .
(4) u
x
2 u
y
= u1, subject to the Cauchy condition u = 2y on x =
ky, giving reasons for any restriction that must be placed on k.
(5) Solve by the method of characteristic u
x
+ 2 xu
y
= 2 xu, given
that u = x
2
on the initial curve with the equation y =
x
2
2
.
(6) u
t
+ 2 u u
x
= 0, subject to the Cauchy condition u(x, 0) = tanh x.
(7) u
t
u u
x
= e
t
, subject to the Cauchy condition u(x, 0) = x.
(8) u u
t
+u
x
= 0, subject to the Cauchy condition u(x, 1) =
1
x
for x
1.
(9) u
t
+u u
x
= t, subject to the Cauchy condition u(x, 0) = 2x. Find
the domain in the upper half of (x, t) plane where the solution is
valid.
(10) Solve the PDE u
t
+ 3 u
3
u
x
= 0 subject to the Cauchy condition
u(x, 0) =
_

_
1, < x < 1
x, 1 x 4
4, 4 < x <
39
40CHAPTER 4. FIRSTORDER PARTIAL DIFFERENTIAL EQUATIONS
Exercise 4.2.8. Solve the following PDE by Lagranges method
(1) 3 u
x
+ 2 u
y
= 0 with u(x, 0) = sin(x)
(2) y u
x
+xu
y
= 0 with u(0, y) = e
y
2
(3) y u
x
+ xu
y
= xy, x 0, y 0 with u(0, y) = e
y
2
for y >
0, u(x, 0) = e
x
2
, for x > 0
(4) 2 xy u
x
+ (x
2
+ y
2
) u
y
= 0, u = e
x
xy
on x +y = 1.