CE603 : Partial differential equations (instructor’s scribble)

Expected outcome:

1. You should be able to characterize partial differential equations

2. Learn techniques to solve mainly 1st and 2nd order partial differential equations analytically

Sub-Topics:

Types of partial differential equations that we encounter in engineering problems

General and particular solution of partial differential equations

Learn the following techniques to solve PDE

Method of characteristic curve

Separation of variables

Integral transform method

Solution of specific types of PDE

Parabolic

Elliptic

Hyperbolic

Week 9: we will only cover the method of characteristic curve for 1 st order PDE that will be in your
quiz. However, the method of characteristic for 2nd order PDE will not be in next week quiz.

General introduction

Differential equations with more than one independent variables are known as partial differential equa-
tions (PDEs). Similar to ODEs PDEs also model the behaviour of a system with underlying physical
and chemical processes. PDEs for describing engineering processes are generally derived by using the
following laws that are as follows

i. Mass-balance
ii. Linear momentum balance
iii. Angular momentum balance
iv. First law of thermodynamics or energy balance
v. Second law of thermodynamics or entropy imbalance

1
Referring to our discussion on ODE, we have seen the application of the first law when we were mod-
elling the solute inflow and outflow in tank. Linear momentum and angular momentum balance was
used to derive the governing equation of the pulley problem, the 4 th principle was used to study heat
transfer. We have not seen any application of the fifth law. Fifth law is only required when you want to
model some type of dissipation in the system that can be degradation, damage, decay etc. In multi-
physics problem with dissipation we will need all 5 laws. At micron, sub-micron, or nano-scales you
may need to adopt more fundamental lawos to derive PDEs governing the behavior of the system one
such example is Schrödinger equation .

Some important partial differential equations

The wave equations

2 1 ∂2
∇ w (r)= 2 2
w( r) it is the wave equation in co-ordinate independent form i.e. it is valid irre-
c ∂t
spective of the coordinate systems. ∇ 2=∇ . ∇ the Laplacian operator is dot between the gradient
operators.For 3D Cartesian coordinate system we will have the followings

x 1 , x 2 , x3 are the three Cartesian coordinates that will denote the position co-ordinate r given by
r =x 1 e 1 + x2 e2 +x 3 e3 where e 1 , e2 , e 3 are unit vectors in the three Cartesian coordinate directions,
the dependent variable w ( x1 , x 2 , x 3 ) , in 3D Cartesian coordinate frame is as follows

2 2 3
∇ 2= ∂ 2 + ∂ 2 + ∂ 3 , and ∇= ∂ e1 + ∂ e 2 + ∂ e 3 . w is the dependent variable that is dis-
∂ x1 ∂ x 2 ∂ x 2 ∂ x1 ∂ x2 ∂ x3
placement from equilibrium for vibrating solid, gas or liquid, electric or magnetic field in electromag-
netic wave or the current or voltage along a transmission line. The constant c represents the speed of
propagation of the wave. Later I will show you how to transform these differential equations from one
coordinate system to another.

Given next shows the transverse vibration of a pre-stressed membrane

( )
2 2 2
∂ w ∂ w ∂ w
2+ 2 +q( x 1 , x 2 ,t)= ρ ( x1 , x 2)
T where w is the transverse deflection, T stress in
∂ x1 ∂ x2 ∂ t2
the membrane, q transverse loading, and ρ the density of membrane material.

Diffusion/Heat equation without external heat flux

∂u
κ ∇ 2 u=
∂t

where κ is the thermal conductivity or diffusion constant, u is temperature in heat equation, or

chemical concentration for diffusion problem.

Transient heat equation with external heat source is given as

2
 ∂T
κ ∇ 2 T + f ( x1 , x 2 , x 3 )=s ρ
∂t
where κ is thermal conductivity, f external heat source throughout the body, s specific heat,
ρ density.

Laplace equation
2
∇ u=0

∂T
It represents steady state temperature distribution that we obtain by setting =0 in transient heat
∂t
equation, it also describes electrostatic potential in a charge free region, flow of in-compressible fluid
with no source, sink, or vortices for which u is velocity potential v =∇ u

Poisson’s equation

It is in-homogeneous Laplace equation given as ∇ 2 u=f ( x 1 , x2 , x 3 ) and f the source densitiy.

Bending of thin plate is also a Poisson’s type PDE that is given as

2 q
∇ P= and ∇ 2 w=P where w is transverse deflection, q transverse load on the plate,
D
D flexural rigidity of the plate .

Helmholtz’s equations
2
∇ u+ λ u=0 used in acoustic problems.

All of the above PDEs are linear PDEs that can be written as L(u)=f where L is the linear operator.
These differential equations differ by the order of differentiation with respect to time.

Some nonlinear PDEs are given below for your reference only

The Fisher equation

∂u
∂t
−ν
∂2 u
∂x
2
( u2
)
=α u− κ , where x is real and t >0
It is used to study wave propagation in a large number of biological and chemical systems.

The Burgers equations

2
∂u ∂u ∂u
+u =ν , where x is real and t >0
∂t ∂x ∂x
2

3
It describes one dimensional turbulence, sound wave in viscous medium etc.

If you are curious to know more about nonlinear PDEs you are requested to read Nonlinear partial dif-
ferential equations for scientists and engineers by Lokenath Debnath.

Example 1 change of coordinate from 3D Cartesian to 3D cylindrical

You transform these equation in different coordinate systems by choosing appropriate transformation
rule from one coordinate system to another. For instance in cylindrical coordinate system
x 1=r cos θ , x 2 =r sin θ , x 3= z . Then the gradient operator in cylindrical coordinate system is
∇= ∂ e r + ∂ eθ + ∂ e z where e r ,eθ and e z are unit vectors in cylindrical coordinate system
∂r r ∂θ ∂z
along the radial, tangential, and longitudinal directions respectively. And, e r =cos θ e1 +sin θ e 2 ,
eθ =−sin θ e 1+cos θ e 2 , e z=e 3 .

The Laplacian operator in Cylindrical coordinate becomes

∂r( ∂z ∂r )(
∇ . ∇= ∂ e r + ∂ eθ + ∂ e z . ∂ e r + ∂ eθ + ∂ e z
r ∂θ r ∂θ ∂z )
See if you can prove (differentiating term by term)

1 2 2 2
∇ . ∇= ∂ 2 + ∂ + 2 ∂ 2 + ∂ 2
∂ r r ∂ r r ∂θ ∂ z

∂ eθ ∂ er
While differentiating keep in mind =e r and =eθ
∂θ ∂θ

General and particular solution of first order PDEs using method of characteristics

Before learning the techniques to characterize and solve the above-mentioned PDEs we will look into
general and particular solution of PDEs starting with 1st order PDEs. For simplification, we will look
into first order PDEs with two independent variables namely x 1 , x 2 and one dependent variable
u( x 1 , x 2 ) . Please note that u can be a vector valued function but we will consider a scalar func-
tion.

Most general form of a 1st order non-homogeneous PDE is given as

∂u ∂u
C 1 (u , x 1 , x 2 ) +C 2 (u , x 1 , x 2 ) =F (u , x1 , x 2) (quasi-linear PDE)
∂ x1 ∂ x2

∂u ∂u
C1 (x 1 , x 2 ) +C 2 ( x1 , x 2) =F (u , x 1 , x 2 ) (semi-linear PDE)
∂ x1 ∂ x2

∂u ∂u
C1 (x 1 , x 2 ) +C 2 ( x1 , x 2) + C0 (x 1 , x 2 )u=F (x 1 , x 2) (linear PDE)
∂ x1 ∂ x2

4
 F(u , x 1 , x 2 ) is a continuous function of its arguments x 1 , x 2 ,u and the coefficients C 0 ,C 1 ,C 2
are continuous functions of the independent variables. Please note a semi linear PDE gets converted to
linear PDE if F(u , x 1 , x 2 ) is replaced by f (x 1 , x 2)u +g ( x 1 , x 2 ) .

A general solution u( x 1 , x 2 ) satisfies a PDE. Once we impose the boundary conditions on the gen-
eral solution we obtain he particular solution. In the method of characteristics we convert the PDE to
ODEs that we can solve easily. There are a few ways to look into such conversions that I will discuss
next. Later, we will learn the technique to obtain the general solution of 1 st order PDEs through some
examples.

Approach1: To convert PDEs to a set of ODEs

Solution to a PDE in 2-dimension is a surface defined by x 1 , x 2 , u(x 1 , x 2) where u( x 1 , x 2) is the

dependent variable. Presently, we have two independent variables x 1 , x 2 . Say we have parametric
representation of the coordinates given as x 1 (r) , x 2 (r) . Then the solution of the differential equation
is given by u( x 1 (r ), x 2 (r )) . Now, we seek to find the parametric curve s that relates x 1 , x 2 . and
u . Let me first re-write the PDE as follows

∂u ∂u
C1 (u , x 1 , x 2) +C2 (u , x 1 , x 2) =F (u , x1 , x2 ) Eq A
∂ x1 ∂ x2

Assuming that we can represent the solution using a parametric curve we write
du ∂ u dx 1 ∂ u dx 2 Eq B
= +
dr ∂ x 1 dr ∂ x 2 dr
Re-arrange Eq. B to write
dx 1 ∂ u dx 2 ∂ u du
+ = Eq C
dr ∂ x 1 dr ∂ x 2 dr
Compare the terms on the left side of Eq A with those on left side of Eq C, and also compare the terms
on right side of Eq A with the one right side of Eq C. Then w get the following relationships
dx 1 dx 2 du
=C 1 (u, x1 , x 2) , =C 2 (u , x1 , x 2 ) , =F (u , x 1 , x 2 ) Eq C
dr dr dr
Now, the original PDE is replaced by 3 ODEs. The first two equations gives the parametric relation be-
tween x 1 and x 2 and these parametric curves relating x 1 , x 2 are known as characteristic curves.
We can also write the above relation in differential form as follows (See if you can prove it. Hint:
dx 1 Δ x1
=lim )
dr r →0 Δ r
d x1 d x2 d u
= =
C1 C2 F

5
 dx 1 dx 2
Depending on the problem either =C 1 (u , x1 , x 2) , =C 2 (u , x1 , x 2 ) , and
dr dr
d x1 d x2 d u
du dr dr dr d x1 d x2 d u
=F (u , x 1 , x 2 ) , or = = , or = = may become useful.
dr C1 C2 F C1 C2 F
In case of homogeneous equation the first form is easier to work with.

Without introducing too many definitions we are going to look into some examples after I show an-
other way of deriving the above equations.

Approach 2: Cauchy’s method of characteristics with help of differential geometry (you may find
it a bit abstract, it is for your information only)
Let us try writing any of the equation in the following format

{}
∂u

{}
T
C1 ∂ x1
C2 ∂ u =0
F ∂ x2
−1

Say the solution to this PDEs is given by z=u (x 1 , x 2) where z can take any value. Now, we can
imagine the solution space as a surface and point on this surface has 3 coordinates x 1 , x 2 ,u ( x1 , x 2 ) .
Thus the equation of the surface is F( x 1 , x2 , z)=z−u ( x1 , x 2)=0 that means each point on the sur-
face represents a solution of the PDE defined by x 1 , x 2 . From vector calculus the normal to this sur-

{ } { }
T T
∂F ∂F ∂F ∂u ∂u
face is ∇ F= = − − 1 .
∂ x1 ∂ x2 ∂ z ∂ x1 ∂ x2
(Let me give you an analogy. Say you have a sphere of radius R then each point on the sphere is
given by {x 1 , x 2 ,u ( x1 , x 2)} where the x 3 coordinate is given by the function
u( x 1 , x 2 )= √ R − x 1−x 2 for a sphere of radius R . You may be familiar with another expression of
2 2 2

normal to the surface of the sphere where we define the spherical surface as f =x 21 + x 22 + x23 −R 2 =0 ,

{ }
T
∂f ∂f ∂f 2 T
={2 x 1 2 x 2 2 x 3 } ={2 x 1 2 x 2 2 √ R − x1 −x 2 }
T 2 2
then the normal vector ∇ f =
∂ x 1 ∂ x 2 ∂ x3

{√ }
T
x1 x2
that we can write as √ R − x −x
2 2 2
1 . So both approaches give the
R 2− x 21− x22 √ R2 −x 21− x 22
1 2

same normal.) A pictorial illustration is also given next for your reference.

6
 Figure 1: A pictorial representation of the solution of 1st order PDE
In short, the solution is given by a surface defined as z=u (x 1 , x 2) , each point on the surface has 3
co-ordinates namely x 1 , x 2 ,u ( x1 , x 2 ) . Now let us recall the differential equation written in the fol-
lowing form

{}
∂u

{}
T
C1 ∂ x1
C2 ∂ u =0
F ∂ x2
−1

{ }
T
∂u ∂u
Evidently, −1 is normal to the surface, then {C 1 C 2 F } must represent a tan-
∂ x1 ∂ x2
gent to a curve lying on the surface F( x 1 , x2 , z)=z−u ( x1 , x 2)=0 and this curve is known as charac-
teristic curve, let the parametric representation of this characteristic curve be represented by r . Then
the tangent vector is t = { d x1 d x2 d u
dr dr dr } . Thus we can say t = {d x1 d x2 d u
dr dr dr }
and
{C 1 C 2 F} are parallel. Hence, we write

d x1 d x2 d u
dr dr dr
= =
C1 C2 F

We can also write the above relation in differential form as follows (See if you can prove it. Hint:
dx 1 Δ x1
=lim )
dr r →0 Δ r
d x1 d x2 d u
= =
C1 C2 F

7
 dx 1 dx 2
Depending on the problem either =C 1 (u , x1 , x 2) , =C 2 (u , x1 , x 2 ) , and
dr dr
d x1 d x2 d u
du dr dr dr d x1 d x2 d u
=F (u , x 1 , x 2 ) , or = = , or = = may become useful.
dr C1 C2 F C1 C2 F
In case of homogeneous equation the first form is easier to work with.

Now let us see a few examples

Example 1: Linear homogeneous ODE: 1 D transport equation

∂u ∂u
+C =0 with boundary condition u(x ,t=0)=2 x 2 (the boundary condition may not make
∂t ∂x
sense as I just chose it arbitrarily). The solution u(x (r ),t (r )) where r is the parametric relation-
ship between x , t .

From previous derivations

dt dx du
=1, =C , =0 , it is an initial value problem for which the boundary condition is
dr dr dr
u(x (r ),t (r )=0)=2 x 2( r) . After integrating we calculate t( r)=r + K 1 , x (r)=Cr+ K 2 ,
u(r )=K 3 where the integration constants are K 1 , K 2 , K 3 .

The equation of characteristic curve is obtained by combining t( r)=r + K 1 , x (r)=Cr+ K 2 to write

the equation of characteristic curve is given as x−Ct=K 2−C K 1 . There are many characteristic
curves depending on the values of K 2−C K 1 , collection of these characteristic curves is the charac-
teristic surface.

General homogeneous solution is given by f ( x−Ct) where f can be any function. For this prob-
lem, as we know the boundary conditions we can obtain f as follows. Given that,
2
u(x (r ),t (r )=0)=2 x ( r)
The boundary condition when t( r)=0→r=−K 1
Thus, x (r=−K 1 )=−CK 1+ K 2
We have, u(x (r ),t (r )=0)=2(−C K 1+ K 2)2=K 3
Thus, the solution is u(x (r ),t (r ))=2(−C K 1 + K 2)2 , now we can replace the constants using x , t
to obtain
2
u( x ,t )=2(x−Ct )

For visualization create a 3d plot of the characteristic curve and the solution.

8
Example 2: non-homogeneous ODE: 1 D transport equation

∂u ∂u
+C =1 with boundary condition u(x , 0)=sin x
∂t ∂x

After integrating we calculate t(r)=r + K 1 , x (r)=Cr+ K 2 , u(r )=K 3 where the integration con-
dt dx du
stants are K 1 , K 2 , K 3 . Therefore, =1, =C , =1 for which the boundary condition is
dr dr dr
u(x (r ),t (r )=0)=sin (x( r)) . After integrating we calculate t( r)=r + K 1 , x (r)=Cr+ K 2 ,
u(r )=r + K 3 where the integration constants are K 1 , K 2 , K 3 .

The equation of characteristic curve is obtained by combining t( r)=r + K 1 , x (r)=Cr+ K 2 to write

the equation of characteristic curve is given as x−Ct=K 2−C K 1 .

The boundary condition when t( r)=0→r=−K 1 (basically we found the value of r for which t = 0
Thus, x (r=−K 1 )=−CK 1+ K 2

We have, u(x (r ),t (r )=0)=sin (−C K 1 + K 2)=−K 1+ K 3 →K 3=sin(−CK 1 + K 2)+ K 1

Thus, the complete solution, u=r + sin(−C K 1 + K 2 )+ K 1=t+ sin(x−Ct )

Example 3: 1 D transport equation with decay

∂u ∂u 2

+3 =−u where the boundary condition is u(x , 0)=e−x

∂t ∂x
We have,
dt dx du −r+ K
=1→t=r + K 1 , =3 →x=3 r + K 2 , =−u→ln u=−r + K 3 →e =u 3

dr dr dr

Characteristic curve is x−3 t=K 2−3 K 1

When t=0 , r=−K 1 , x=K 2−3 K 1

2
−(K −3 K )
Now, u( K 2−3 K 1 ,−K 1)=e 2
=e(K +K ) → K 3=−( K 2−3 K 1)2−K 1
1 1 3

2 2
−r+ K 3 −r−( K 2−3 K 1) −K 1 −t −( x−3 t )
Final solution u=e =e =e e

Example 4: Solve the following

∂u ∂ u
+u =u subjected to u( x , 0)=2 x for 1≤x≤2
∂t ∂x

dt du dx
=1, =u , =u
dr dr dr

9
We have,
dt du r+C dx r +C r+C
=1→t=r + K 1 , =u→ln u=r+ C1 →e =u , then 1
=e → x=e +C 2
1 1

dr dr dr

Without loss of generality, to avoid too much typing, let us assume t=0 when r=0 then
K 1=0 .

t +C 1 t +C 1
Thus, u=e , x=e +C 2

So, we have three constants K 1 ,C 1 , C 2 and here the characteristic curve will be implicit.

Now let us impose the boundary conditions

C C 1 C
When, t=0, u(x ,t=0)=2 x →e =2 e +2 C2 →C3 =− e 1 1 1

2
t C 1 C C 2x
x=e e − e 1
e = t 1 1

2 2 e −1

t +C 1 2 x et
Therefore, the solution is u=e =
2 et −1

Please also refer to the PDF containing the more examples for you to practice.

Solve the followings for practice

∂u ∂u
1. x 2 − x1 =3 x 1
∂ x1 ∂ x2

with boundary condition u( x 1 , 0)=x 21

Although the previous approach is feasible but to make it easier take help of

d x1 d x2 d u
= =
C1 C2 F
Following the steps show that the general solution is u( x 1 , x 2 )=−3 x 2 + f ( x21 + x 22 )

And the particular solution is u( x 1 , x 2 )=−3 x 2 +( x 21 + x22 )

2. Solve
∂u ∂u −u
x1 + x2 =x e subjected to boundary condition u( x 1 , x 21 )= x 1
∂ x1 ∂ x2

10
 u
Show the general solution is e = x1 + f
x2
( )x2
x1
u x2
x1
Particular solution is , e = x1 +e −
x1
∂u ∂ u
3. Solve the Burger’s equation u + =0 , such that u( x 1 ,0)=0
∂ x1 ∂ x2

d x1 d x2 d u
dr dr dr
Hint: Here, = = may become handy. However, while using this form make use of
C1 C2 F
du
=lim F
dr F→ 0
Show that the characteristic curve is a line.
∂u ∂u
4. + √ x1 =0
∂ x1 ∂ x2

( )
3
2
Show that the solution is u( x 1 , x 2 )=f y− x 2 +C where C is an arbitrary constant.
3

The concept of characteristic curve will be useful to classify PDEs.

xxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxxx

Classification of 2nd order PDEs

11
Appendix: Linear PDEs (you can skip this part because it is a special case of the previous deriva -
tions)

1st order ODE with constant coefficients

a) Homogeneous case

∂u ∂u
Let us choose, a +b =0
∂ x1 ∂ x2
∂u ∂u ∂r ∂u ∂ s ∂u ∂u
We choose r=a x 1 +b x 2 and s=bx 1−ax2 , Thus = + =a +b and
∂ x 1 ∂ r ∂ x1 ∂ s ∂ x 1 ∂r ∂s
∂u ∂u ∂r ∂u ∂ s ∂u ∂u
= + =b −a
∂ x 2 ∂r ∂ x 2 ∂ s ∂ x 2 ∂r ∂s
Substituting the above in the PDE

2 2 ∂u
(a +b ) =0
∂r

So we have, uc (r , s)=p (s)

Thus the solution u( x 1 , x 2)=p (b x 1−ax 2) is the general solution of the problem.

What we basically did is to cleverly choose the transformation from x 1 , x 2 to r , s so that we could
get rid of one of the partial. Please note r , s are called characteristic coordinates.

b) Non-homogeneous case

∂u ∂u
a +b =f ( x 1 , x 2 )
∂ x1 ∂ x2

As usual the first step is to solve the homogeneous case for which we got uh (x 1 , x 2)=p (b x 1−a x 2) .
With help of r , s we can convert the non-homogeneous PDE to a non-homogeneous ODE as

2 2 ∂u
(a +b ) =g(r , s) you can integrate it as there is only 1 variable r now.
∂r

g( r , s)
u p (r , s )=∫ 2 2
dr + q(s)
a +b

g (r , s)
Complete solution u(r , s )=uc +u p= p(s)+∫ 2 2
dr +q( s)
a +b
Particular solution can be obtained by substituting the boundary conditions.

12