You are on page 1of 37

Chapter 6

Analytical Methods for Partial Differential


Equations

1 Definitions
Partial differential equations are statements of relationships among various distinct deriva-
tives of a function of more than one variable. A PDE is linear if it contains no products or
powers of the unknown function or its partial derivatives. It is quasi linear if no powers or
products of the highest order derivative occur. A first order linear PDE is, for instance
z z
x +y =z
x y
A solution of this PDE is an expression of the form z = (x, y), (or, alternatively z(x, y) =
F (x, y, z) = 0). This is also called the integral surface of the problem.
Recall that a general solution of a linear ordinary differential equation of order n can
be expressed as a linear combination of n independent functions each of them multiplied by
one of n arbitrary constants. In contrast, a general solution of a PDE involves arbitrary
functions of specific functions. For instance, the arbitrary function z = f (u) of the specific
function u(x, y) = x + 2y is the most general solution which satisfies the PDE
z z
2 =0
x y
An important special case is the expression

z = f (ax + by) + g(cx + dy)

where a, b, c and d are constants such that ad 6= bc is the general solution of the linear PDE
2z 2z 2z
A + B + C =0
x2 xy y 2
where A, B and C are constants depending on a, b, c and d.

1
In practice, rather than looking for general solutions of PDEs and then specializing them
by introducing boundary conditions one determines particular solutions directly and tries to
combine them to satisfy the prescribed conditions. However, it is useful to investigate some
features of the general solutions.

2 First Order Quasi linear Equations


The most general quasi linear first order PDE is
z z
P (x, y, z) + Q(x, y, z) = R(x, y, z)
x y
where z = (x, y). If the integral surface here is z (x, y) = u(x, y, z) = c then (as long as
u/z 6= 0)
z u/x
=
x u/z
and
z u/y
=
y u/z
Substituting back in the original PDE
u u u
P +Q +R = (P i + Qj + Rk) u = 0
x y z
Since u is normal to the surface u = c then the vector V = P i + Qj + Rk is tangent to it.
Curves traced by the local directions of V are called characteristic curves of the PDE.
The general solution of the original PDE is of the form

u2 (x, y, z) = f [u1 (x, y, z)]


where u1 = c1 and u2 = c2 are solutions of any two independent ODEs associated with the
relationships
dx dy dz
= =
P Q R
which result from the fact that the unit tangent vector along a characteristic has the same
direction as V. Further, the intersection of any two of the surfaces u1 = c1 and u2 = c2 is a
characteristic.
For instance, for the PDE
z z
2 =0
x y

2
one has
dx dy dz
= =
2 1 0
giving u1 = x + 2y = c1 and u2 = z = c2 so that the required solution is of the form

z = f (x + 2y)

Further, since x + 2y = c1 and z = c2 are planes in 3D the characteristics of the original


PDE are lines parallel to the xy plane where the two planes intersect. To generalize, for the
linear first order PDE
z z
a +b =0
x y
where a and b are constants, the general solution is of the form

z = f (bx ay)

Consider now the linear first order PDE


z z
P (x, y) + Q(x, y) = R1 (x, y)z + R2 (x, y)
x y
Here dy/dx = Q/P and dz/dx = R1 z + R2 and the required general solution of the original
PDE has the form

z = s1 (x, y)f [s2 (x, y)] + s3 (x, y)

where s1 , s2 and s3 are specific functions and f is an arbitrary function.

3 Second Order Equations


The general form of a quasi linear second order PDE is

2z 2z 2z
a + b + c +F =0
x2 xy y 2
where a, b, c and F may depend on x, y, z, z/x and z/y. If a, b and c are independent
of z, z/x and z/y and F is linear on those same quantities the PDE is linear in z, i.e.

2z 2z 2z z z
a 2 +b +c 2 +d +e + fz = g
x xy y x y
where a, b, c, d and e depend only on x and y.

3
The discriminant is D = b2 4ac. In analogy with conics, the PDE is called hyperbolic
if D > 0, elliptic if D < 0 and parabolic if D = 0.
The most general solution of the linear PDE is the sum of any particular solution and
the most general solution obtained by making g = 0 (called the complementary solution).
Consider now the second order linear PDE of special form

2z 2z 2z
a + b + c =0
x2 xy y 2

where a, b and c are constants. Assume a general solution of this is of the form z = f (y +mx)
where m is a constant which must be determined. Differentiation and substitution gives

am2 + bm + c = 0

If D > 0 this has two real solutions and the problem corresponds to an hyperbolic PDE

2z 2z
=0
x2 y 2
with general solution

z = f (x + y) + g(x y)

Alternatively, if D < 0, the roots are complex conjugate and the PDE is elliptic (Laplaces
equation)

2z 2z
+ =0
x2 y 2
with general solution

z = f (x + iy) + g(x iy)

Finally, if D = 0 one has a perfect square and the PDE is parabolic

2z 2z 2z
2 + =0
x2 xy y 2
with general solution

z = f (x + y) + xg(x + y)

4
4 Characteristics
Consider the linear first order PDE
z z
y x =y
x y
The two associated equations are
dx dy
=
y x

with solution u1 = x2 + y 2 = c1 , and

dz = dx

with solution u2 = z x = c2 . The characteristic curves (or simply the characteristics) are
then the ellipses formed by the intersection of the right circular cylinders x2 + y 2 = c1 with
the planes z x = c2 . The cylinders x2 + y 2 = c1 are called characteristic cylinders while the
circles x2 + y 2 = c1 on the xy plane are the characteristic base curves and are the projections
of the characteristics onto the that plane.
For the general linear second order PDE

2z 2z 2z z z
a + b + c + d + e + fz = g
x2 xy y 2 x y

the characteristic base curves (in the xy plane) satisfy

a(dy)2 b(dx)(dy) + c(dx)2 = 0

For the hyperbolic case D = b2 4ac > 0 and two distinct sets of such curves result. For
the elliptic case D < 0 and no base curves exist. Finally for the parabolic case the two sets
coincide.

5 Partial Differential Equations commonly encountered


in Applications
Partial differential equations encountered in applications are special cases of

2
2 = 2
+ +h
t t
where = f (x, y, z, t).

5
Important examples are: Laplaces equation

2 = 0

This equation is a mathematical representation of the potential function in steady incom-


pressible irrotational flows, the gravitational potential in free space, the electrostatic poten-
tial in steady electric current flows and the steady state temperature in heat conducting
solids.
The non homogenous counterpart to Laplaces equation is Poissons equation

2 = h

describing for example steady incompressible irrotational flows with sources and/or sinks,
steady state temperature in heat conduction with internal energy generation and the stress
function in torsion of prismatic bars.
The wave equation

1 2
2 =
c2 t2
describing the propagation of waves with velocity c as the electric and magnetic vector
fields in electromagnetics, components of displacement in elastic vibrations and the velocity
potential in the theory of sound.
The heat equation
1
2 =
t
describing the transport of thermal energy in solids and the diffusion of solutes.
Solutions to all the above problems must be found under prescribed boundary and initial
conditions. Moreover, solutions must be single valued and bounded.
The approach to solving problems involving PDEs consists of first determining a suitable
set of particular solutions satisfying some of the specified conditions using, for instance, the
method of separation of variables, and then combining sets of the found solutions so as to
satisfy all the prescribed conditions.
Various analytical methods of solution can be employed to solve linear problems involving
partial differential equations. Descriptions are included below of the methods of separation
of variables, decomposition, Duhamel, Fourier and Laplace transforms and variation of pa-
rameters as applied to the solution of selected elliptic, parabolic and hyperbolic problems in
partial differential equations.

6
6 A Physical Illustration: Energy Balance and the Steady
State Heat Equation
In heat conduction theory, the amount of thermal energy crossing a unit boundary area of
material in unit time is

f = K(T ) n

where T is the temperature and K is the thermal conductivity. The rate of accumulation of
thermal energy per unit volume of material per unit time is
T
e = C
t
where is the density and C the specific heat. Considering now a region of size V and total
boundary area S, where the thermal properties are constant, an energy balance gives
I I Z Z Z Z Z Z
2 T
K(T ) nd = K T d = C d
S V V t
But this must be valid regardless the size of V so that in the limit
1 T
2 T =
t
where = K/C is the thermal diffusivity. So, heat conduction problems consist of deter-
mining T as a function of position and time.
At steady state T /t = 0 so that the temperature satisfies Laplaces equation

2 T = 0

7 Steady State or Elliptic Problems


Steady state problem encountered in applications are represented by Laplaces equation.
Problems involving functions that satisfy Laplaces equations as well as suitable boundary
conditions are called elliptic problems.
Solutions to Laplaces equation are called harmonic functions. A harmonic function of
two variables u(x, y) possesses the following characteristic properties:
From Greens theorem, it is required that
I I
u
d = 0
n
which means that the boundary flux of u cannot be chosen arbitrarily but must average
zero over the entire surface.

7
Also, from Greens theorem, if u1 and u2 are two solutions whose values coincide at
the boundary
Z Z Z
[(u2 u1 )]2 d = 0
V

so that u2 u1 = constant. The constant is zero when the problem involves only
prescribed values of u at the boundary (Dirichlet problem) and can be nonzero when
the fluxes u/n are specified (Neumann problem).

7.1 The One-Dimensional Limit of Laplaces Equation


In a three-dimensional rectangular Cartesian system of coordinates (x, y, z) for the function
u(x, y, z), Laplaces equation is

2u 2u 2u
2 u = + + 2 =0
x2 y 2 z

For a one-dimensional system x [0, L] the above yields

d2 u
=0
dx2
which is readily solved to give

u(x) = Ax + B

where the values of the constants A and B must be determined by specifying boundary
conditions at x = 0 and x = L. If u(0) = u0 and u(L) = uL , the solution becomes
uL u0
u(x) = x + u0
L
The method of separation of variables is a powerful approach designed to obtain solutions
of initial and boundary value problems for partial differential equations. The key assumption
made is that the desired solution, being a function of multiple independent variables, can
be represented as a product of simpler functions each of which may depend only on a single
independent variable. As this assumption is introduced into the original partial differential
equation, systems of initial and/or boundary value problems for ordinary differential equa-
tions result. These can in turn be readily solved using methods learned earlier and the results
finally substituted back into the original product function to produce the final solution. The
implementation of the method will be illustrated by solving a few selected elliptic problems.

8
7.2 Laplaces Equation in a Plate
Consider as a first example the problem of steady state heat conduction in a thin rectangular
plate of width l and height d. The edges x = 0, x = l and y = 0 are maintained at u = 0
while at the edge y = d, u(x, d) = f (x). No heat flow along the z direction perpendicular to
the plane of the plate. The required temperature u(x, y) satisfies
2u 2u
+ =0
x2 y 2
The goal is to determine the function u(x, y) that satisfies the above equation as well as all
the boundary conditions.
To find u(x, y) by separation of variables we assume the a particular solution can be
represented as a product of two functions each depending on a single coordinate, i.e.

up (x, y) = X(x)Y (y)

substituting this into Laplaces equation gives


1 d2 X 1 d2 Y
= = k2
X dx2 Y dy 2
where k 2 is a constant called the separation constant.
The above must be true since the LHS is a function of x alone while the RHS a function of
y alone. The constant is selected as k 2 in order to obtain a proper Sturm-Liouville problem
for X (with real eigenvalues). A clue about the appropriate choice of the placement of
the minus sign above is derived on physical grounds. Note that for a fixed value of y, the
solution must be zero at x = 0, go through a maximum at x = l/2 and become zero again
for x = l. Therefore, a particular solution in terms of simple trigonometric functions can be
expected along the x-direction. On the other hand, along the y-direction, at constant x, a
monotonically increasing/decreasing function is expected as y changes from zero to d.
With the above the original PDE problem has been transformed into a system of two
ODEs subject to their own boundary conditions, i.e.

X 00 + k 2 X = 0

subject to X(0) = X(l) = 0 and

Y 00 k 2 Y = 0

subject to Y (0) = 0.
The solution for X(x) is readily obtained as the solution of a Sturm-Liouville, eigenvalue
problem and is
nx
X = Xn = An sin( )
l

9
with eigenvalues
n
kn =
l
for n = 1, 2, 3, ....
The solution for Y (y) is also readily obtained.
ny
Yn = Bn sinh( )
l
Note that only one boundary condition is employed.
Note also that,as a consequence of the eigenvalue structure of the associated ODE prob-
lem, for each value of n one can then write up (x, y) = Xn (x)Yn (x). The most general
solution is simply obtained using the principle of superposition, i.e. by a linear combination
of individual solutions with different values of n, i.e.

X
X nx ny
u(x, y) = Xn (x)Yn (x) = an sin( ) sinh( )
n=1 n=1 l l

Finally, the coefficients an s must now be determined by making the above satisfy the
non homogeneous condition at y = d, i.e.

X nd nx
f (x) = [an sinh( )] sin( )
n=1 l l

which is a Fourier sine series representation of f (x) with coefficients

nd 2Z l nx
cn = an sinh( )= f (x) sin( )dx
l l 0 l
so that the final solution is

X nx sinh( ny
l
)
u(x, y) = cn sin( ) nd
n=1 l sinh( l )

Therefore, as long as f (x) is representable in terms of Fourier series, the obtained solution
converges to the desired solution. Note also that the presence of homogeneous conditions at
x = 0, x = l made feasible the determination of the required eigenvalues.

7.3 Laplaces Equation in a Cylinder


Consider the problem of determining the electrical potential inside a vertical solid cylinder
(radius a, height L) which is maintained at ground potential (u = 0) over its side and

10
bottom surfaces and is subjected to a potential u = 1) at its upper surface. The problem
thus requires determination of the function u(r, z) satisfying
2 u 1 u 2 u
+ + 2 =0
r2 r r z
subject to
u(a, z) = u(r, 0) = 0
and
u(r, L) = 1
Moreover, the resulting function must be finite along the cylinder centerline (r = 0).
Separation of variables starts by assuming that the desired solution has the form
u(r, z) = R(r)Z(z)
Substituting into the differential equation and rearranging yields the following equivalent
system of ODEs
d dR
(r ) + k 2 rR = 0
dr dr

d2 Z
k2Z = 0
dz 2
where k 2 is the separation constant.
Solving the above and incorporating all the homogeneous boundary conditions and the
boundedness condition yields

X
u(r, z) = cn J0 (kn r) sinh(kn z)
n=1

where the kn s (n = 1, 2, 3, ...) are the positive roots of


J0 (kn a) = 0
Finally, the Fourier coefficients cn are determined by imposing the remaining boundary
condition at z = L, i.e.

X
1= [cn sinh(kn L)]J0 (kn r)
n=1

which is the Fourier-Bessel series representation of the number 1. Orthogonality of the


characteristic functions yields
Ra
rJ0 (kn r)dr
0
cn =
sinh(kn L)a2 J12 (kn a)

11
7.4 Laplaces Equation in a Brick
Consider now Laplaces equation for the function u(x, y, z) inside a rectangular parallelepiped
(length l1 , width l2 , height d) subject to u = 0 on all five faces except that at z = d which
is maintained at u(x, y, d) = f (x, y). Laplaces equation in this case is

2u 2u 2u
+ + 2 =0
x2 y 2 z
Assume a particular solution is of the form

up (x, y, z) = X(x)Y (y)Z(z)

With the above one has


2u d2 X
= Y Z = Y ZX 00
x2 dx2

2u d2 Y
2
= XZ 2 = XZY 00
y dy
and
2u d2 Z
= XY = XY Z 00
z 2 dz 2
Substituting into the original PDE yields, after division by XY Z,
1 00 1 00 1 00
X + Y + Z =0
X Y Z
which can be rearranged to
1 00 1 00 1
Y + Z = X 00
Y Z X
But the left hand side is a function of y and z only while the right hand side is a function of
x only, therefore, necessarily
1 00 1 00 1
Y + Z = X 00 = k12
Y Z X
Now, rearrangement of the term on the left hand side yields
1 00 1
Z k12 = Y 00
Z Y

12
but the term on the left hand side is a function of z only while the one on the right hand
side is a function of y only, thus, necessarily
1 00 1
Z k12 = Y 00 = k22
Z Y
Two separation constants are needed in this case giving

X 00 + k12 X = 0

Y 00 + k22 Y = 0

and

Z 00 (k12 + k22 )Z = 0

Note that homogeneous conditions occur at both boundaries in both the x and y directions
leading to
mx
Xm = Am sin(k1 x) = Am sin( )
l1

ny
Yn = Bn sin(k2 y) = Bn sin( )
l2
and

Zmn = Cmn sinh(kmn z)


2
where kmn = k12 + k22 .
The desired solution then becomes
X
X
u(x, y) = Xm (x)Yn (x)Zmn (x) =
m=1 n=1
X
X
mx ny
= amn sin( ) sin( ) sinh(kmn z)
m=1 n=1 l1 l2

where the coefficients amn must be determined by incorporating the non homogeneous con-
dition. Proceeding as before, this leads to a double Fourier series representation and the
coefficients
Z l1 Z l2
1 4 mx ny
amn = f (x, y) sin( ) sin( )dydx
sinh(kmn d) l1 l2 0 0 l1 l2

13
7.5 Non homogeneous Problems
The procedure for non homogeneous problems will be illustrated with an example from
steady state heat conduction theory. Energy is generated internally inside a rectangular
plate of conductivity K and dimensions a b at an rate of Q units of energy per unit time
per unit volume. Two of the plates edges are insulated while the other two are maintained
at zero temperature. The goal is to determine the temperature u(x, y) inside the plate.
The mathematical formulation of the problem consists of the statement of Poissons
equation
2u 2u Q
2
+ 2 =
x y K
The (homogeneous) boundary conditions are, at x = 0 for all y
u(0, y)
=0
x
at y = 0 for all x
u(x, 0)
=0
y
at x = a for all y

u(a, y) = 0

and at y = b for all x

u(x, b) = 0

Here we assume a particular solution exists with the following special form

u(x, y) = (x, y) + (x)

such that the original problem can be reformulated as a superposition of two simpler prob-
lems; the one-dimensional, non homogeneous problem given by
d2 Q
+ =0
dx2 K
subject to
d(0)
=0
dx
and

(a) = 0

14
with the solution
Qa2 x
(x) = [1 ( )2 ]
2K a
And the two-dimensional homogeneous problem given by
2 2
+ 2 =0
x2 y
subject to
(0, y)
=0
x
at y = 0 for all x
(x, 0)
=0
y
at x = a for all y

(a, y) = 0

and at y = b for all x

(x, b) = (x)

which can readily by solved by separation of variables assuming a particular solution of the
form

(x, y) = X(x)Y (y)

to give

2Q X (1)n cos(n x) cosh(n y)
(x, y) =
aK n=0 3n cosh(n b)

where the eigenvalues n are given by


(2n + 1)
n =
2a
with n = 0, 1, 2, ...
Therefore the required solution of the original problem is finally

Qa2 x 2 X (1)n cos(n x) cosh(n y)
u(x, y) = {[1 ( )2 ] 3 }
K a a n=0 3n cosh(n b)

15
8 Time Dependent Problems
While problems involving PDEs of elliptic type must be solved under prescribed boundary
conditions. Initial conditions must be specified in order to solve Hyperbolic PDEs. Parabolic
PDEs require specification of both boundary and initial conditions.
For elliptic problems (b2 < 4ac) the solution is determined everywhere inside a domain
as long as the value of on the boundary of the domain is prescribed. If instead of the value
of , that of /n, or more generally that of c1 + c2 /n = c3 is given, the solution is
determined within an arbitrary constant. Furthermore, requirements of single-valuedness,
boundedness or periodicity may be used in place of boundary values. One thus associates
elliptic equations with boundary value problems.
In contrast, hyperbolic PDEs (b2 > 4ac) subject to prescribed boundary conditions
may have infinitely many solutions. However, if say, the values of (x, 0) = f (x) and
(x, 0)/y = g(x) are prescribed, a well defined solution exists even for not very well
behaved functions f (x) and g(x). The deflection of an infinitely long vibrating string is
governed by the wave equation subject to appropriate initial conditions. However, if the
string is finite, boundary conditions appear to be also required but this is not really the
case; the problem is a purely initial value problem. One thus associates hyperbolic equations
with initial value problems.
Parabolic equations (b2 = 4ac) are of an intermediate nature and require specification
of both initial and well as boundary conditions. While for < x < , Txx = Tt
the prescribed initial distribution of T (x, 0) determines all subsequent values of T (x, t), if
0 < x < L, values of T , T /x or combinations of these are required at x = 0 and x = L.

9 Parabolic Problems
Here we investigate selected solution methods used to determine the function u(x, t) that
satisfies the linear homogeneous parabolic partial differential equation
1 u
2 u =
t
subject to appropriate initial and boundary conditions.
As mentioned before, this equation constitutes a mathematical model of the process of
diffusional transport of atoms in materials and the conduction of thermal energy in solids.
It is commonly called the heat equation. The parameter is the thermal diffusivity in the
case of heat conduction and the diffusion coefficient in the case of diffusional mass transport.

9.1 Fundamental Solutions of the Heat Equation


An important set of solutions of the heat equation are associated with the instantaneous
release of a fixed amount of thermal energy at time t = 0 and a point inside an infinite
medium and its subsequent dissipation by conduction into the medium.

16
Let a fixed amount of energy QCp be released at time t = 0 at the origin of the three
dimensional solid of infinite extent, initially at u = 0 everywhere. The goal is to determine
the function u(r, t).
The heat equation is

1 u 2u 2u 2u
= + + 2
t x2 y 2 z
and this must be solved subject to the initial condition

u(x, y, z, 0) = 0

for all x, y, z. No boundary conditions are required in this special case because the medium
is assumed to be of infinite extent.
By differentiation, one can show that the solution of this problem is
2 2 +z 2
Q x +y
u(x, y, z, t) = e 4t
(4t)3/2

This is called a fundamental solution of the heat equation. The solution is useful in the study
of thermal explosions where a buried explosive load located at r = 0 is suddenly released
at t = 0 and the subsequent distribution of temperature at various distances from the
explosion is measured as a function of time. A slight modification of the solution produced
by the method of reflexion constitutes an approximation to the problem of surface heating
of bulk samples by short duration pulses of high energy beams.
Similarly, if the heat is released instantaneously at t = 0 but along the zaxis, the
corresponding fundamental solution is
Q x2 +y2
u(x, y, t) = e 4t
4t
where QCp is now the amount of heat released per unit length.
Finally, if the heat is instantaneously released at t = 0 but on the entire the y z plane
at x = 0 the corresponding fundamental solution is
Q x2
4t
u(x, t) = e
(4t)1/2

where QCp is now the amount of heat released per unit area.

9.2 Parabolic Problem in a Rod


The side surface of a thin homogeneous rod of length L is perfectly insulated. Initially,
the rod temperature is f (x) but then the temperatures at x = 0 and x = L are fixed to

17
new constant values u1 and u2 , respectively. The goal is to determine the function u(x, t)
representing the time dependent temperature along the rod.
The heat equation for this problem is
2u 1 u
2
=
x t
subject to

u(x, 0) = f (x)

u(0, t) = u1

u(L, t) = u2

Since the boundary conditions are inhomogeneous one must apply a decomposition
method first. The key is to represent the desired solution as a simple linear combinations of
the solutions to two, simpler problems: a steady state problem subject to non-homogeneous
boundary conditions and a transient problem subject to homogeneous boundary conditions,
i.e.

u(x, t) = uS (x) + uT (x, t)

where
x
uS (x) = u1 + (u2 u1 )
L
and uT (x, t) must satisfy

uT (0, t) = uT (L, t) = 0

and be such that uT (x, t) 0 as t .


The related transient problem with homogeneous boundary conditions is then
2 uT 1 uT
2
=
x t
subject to

uT (0, t) = 0

uT (L, t) = 0

and an unspecified initial condition.

18
A particular solution for uT (x, t) is readily obtained by separation of variables as follows.
Assume uT (x, t) = X(x)(t). Taking derivatives and substituting into the PDE yields the
system of ODEs
1 d2 X
= 2
X dx2
subject to X(0) = X(L) = 0 and
1 d
= 2
dt
Where, again, the sign has been selected as indicated since trigonometric type solutions are
expected with respect to x and the temperature must be bounded for any value of time.
The corresponding solution to the Sturm-Liouville problem for X is
Xn (x) = B sin(n x)
with the eigenvalues n = n/L for n = 1, 2, ...
The appropriate solution for is in turn
n (t) = C exp(n2 t)
Linear combination of the products of the above solutions yields the required solution for
the homogeneous problem as

X X n2 2 t nx
uT (x, t) = Xn (x)n (t) = an exp( 2
) sin( )
n=1 n=1 L L
The desired solution for the original problem has then the form

x X n2 2 t nx
u(x, t) = uS (x) + uT (x, t) = u1 + (u2 u1 ) + an exp( ) sin( )
L n=1 L2 L
and the remaining coefficients an must be determined from the given initial condition, i.e.

x X nx
u(x, 0) = f (x) = uS (x) + uT (x, 0) = u1 + (u2 u1 ) + an sin( )
L n=1 L
or equivalently
X
x nx
f (x) [u1 + (u2 u1 ) ] = an sin( )
L n=1 L
which is readily recognized as the Fourier sine series representation of the function f (x)
[u1 + (u2 u1 ) Lx ] with the Fourier coefficients given by
Z L
2 x nx
an = (f (x) [u1 + (u2 u1 ) ]) sin( )dx =
L 0 L L
Z
2 L nx 2
= f (x) sin( )dx + (u2 cos(n) u1 )
L 0 L n

19
9.3 Duhamels Superposition Integral
The boundary conditions in all problems above were independent of time. However, time de-
pendent boundary conditions are often encountered in practice. The Duhamel superposition
integral method can be used to solve such problems.
To develop the method, consider first the heat conduction in a thin rod, initially at
u(x, 0) = 0, with its surface x = 0 maintained at u(0, t) = 0 and that at x = L subjected to
u(L, t) = 1. The solution of this problem is readily obtained by separation of variables and
is

x 2X (1)n nx n2 t/
A(x, t) = + sin( )e
L n=1 n L

where = L2 / 2 = CL2 / 2 K.
Now, if the boundary temperature is still maintained at u(L, t) = 1 but not beginning at
t = 0 but instead from t = 1 > 0, then, necessarily u(x, t) = 0 for t < 1 and

u(x, t) = A(x, t 1 )

when t > 1 .
Next we consider surface temperature variations consisting of sequences of step changes.
The surface temperature at x = L is suddenly raised to F (0) at t = 0. Then at t = 1 there
is another sudden rise of magnitude F1 = F (1 ) F (0) that brings the surface temperature
to the value F (1 ). Then, at t = 2 , the value is increased again and becomes F (2 ). The
process is repeated until the temperature becomes F (n ) at t = n .
The temperature field inside the rod resulting from the sequence of step changes is then
simply given by

u(x, t) = F (0)A(x, t) + F1 A(x, t 1 ) + F2 A(x, t 2 ) + ... + Fn A(x, t n ) =


X
n1
Fk
= F (0)A(x, t) + A(x, t k+1 )( )k
k=0 k

Finally, assuming that the function F ( ) is differentiable and that the number of jumps
n the solution becomes
Z t Z t
A
u(x, t) = F (0)A(x, t) + A(x, t )F 0 ( )d = A(x, 0)F (t) + d =
F ( )
0 0 t
Z t Z t
A x 2X (1)n 2 2 nx
F ( ) d = F (t) + [F (0) + F 0 ( )en / d ]en t/ sin( )
0 t L n=1 n 0 L

This is called Duhamels superposition integral.

20
9.4 Fourier Integral Methods
When looking for solutions of parabolic equations in semi infinite domains the Fourier In-
tegral Method is often useful. Consider the problem of determining the function u(x, t) for
x [0, +] which satisfies

2u 1 u
2
=
x t
subject to

u(x, 0) = f (x)

and

u(0, t) = 0

In direct analogy with the situation in a finite domain it is easy to show that a bounded
solution of the above has the form
2 t
up (x, t) = Aev sin(vx)

where v is an arbitrary constant and A = A(v). The quantity v is analogous to the eigenvalues
that result is finite domain problems except that in this case it assumes continuous rather
than discrete values.
One can now superimpose solutions of the above form to produce a more general solution.
However, an integral rather than a sum is required in this case, i.e.
Z 2 t
u(x, t) = A(v)ev sin(vx)du
0

Since the initial condition must also be satisfied this requires that
Z
f (x) = A(v) sin(vx)du
0

But this is precisely the Fourier sine integral representation of f (x) with Fourier coefficients
given by
2Z
A(v) = f () sin(v)d
0
Therefore the resulting solution becomes
2 Z Z v2 t
u(x, t) = e f () sin(v) sin(ux)ddu
0 0

21
However, since
1
sin(v) sin(vx) = [cos(v( x)) cos(v( + x))]
2
and
Z

a2 x2 b2 /4a2
e cos(bx)dx = e
0 2a
the required solution can be expressed as
1 Z 2 2
u(x, t) = f ()[e(x) /4t e(+x) /4t ]d
2 t 0
Specifically, if f (x) = const = u0 , the solution becomes
Z
2u0 x/2 t 2 x
u(x, t) = ev dv = u0 erf ( )
0 2 t
where
2 Z z v2
erf (z) = e dv
0

is the error function.

9.5 Laplace Transform Methods


The Laplace transform method is based on reducing the complexity of the original PDE prob-
lem for the original variable u(x, t) by converting it into an ODE problem for the transformed
variable u(x, s). The simpler problem is then solved and the solution inverse transformed to
recover the required temperature distribution.

9.5.1 Review of Basic Formulae


Consider a real function F (t). The Laplace transform of F (t), L[F (t)] = F (s) is defined as
Z 0
L[F (t)] = F (s) = est F (t0 )dt0
t0 =0

The inverse transform (to recover F (t) from L[F (t)]) is the contour integral in the complex
plane given by
1 Z +i st
F (t) = e F (s)ds
2i s=i

22
where is large enough that all the singularities of F (s) lie to the left of the imaginary axis.
For the transform and inverse to exist, F (t) must be at least piecewise continuous, of
exponential order and tn F (t) must be bounded as t 0+ .
Laplace transforms have the following properties:
Linearity.

L[c1 F (t) + c2 G(t)] = c1 F (s) + c2 G(s)

Transforms of Derivatives.

L[F 0 (t)] = sF (s) F (0)

L[F 00 (t)] = s2 F (s) sF (0) F 0 (0)

L[F 000 (t)] = s3 F (s) s2 F (0) sF 0 (0) F 00 (0)

L[F (n) (t)] = sn F (s) sn1 F (0) sn2 F (1) (0) sn3 F (2) (0) ... F 0(n1) (0)

TransformsR of Integrals.
Let g(t) = 0t F ( )d (i.e. g 0 (t) = F (t)).
Z t 1
L[ F ( )] = F (s)
0 s

Z t Z 2 1
L[ F (1 )d1 d2 ] = F (s)
0 0 s2

Z t Z n 1
L[ ... F (1 )d1 ...dn ] = F (s)
0 0 sn
Scale change.
Let a > 0 be a real number.
1 s
L[F (at)] = F ( )
a a

1
L[F ( t)] = aF (as)
a
Shift.

23
L[eat F (t)] = F (s a)

Transform of translated functions.


The translated function is
(
F (t a) t > a
U (t a)F (t a) =
0 t<a

where
(
1 t>a
U (t a) =
0 t<a

The Laplace transform of the translated function is

L[U (t a)F (t a)] = eas F (s)

Also
1
L[U (t a)] = eas
s
Transform of the function.

L[(t)] = 1

Transform of convolution.
The convolution of two functions f (t) and g(t) is
Z t
f g = f (t )g( )d = g f
0

The Laplace transform is

L[f g] = f(s)g(s)

Derivatives.
If F (s) is the Laplace transform of F (t) then

dF
= F 0 (s) = L[(t)F (t)]
ds

dn F
= F 0 (s) = L[(t)n F (t)]
dsn

24
Integrals.

Z F (t)
F (s0 )ds0 = L[ ]
s t
Functions of More than One Variable
Let f (x, t) be a function of the two independent variables x and t. The Laplace transform
of f (x, t), L[f (x, t)] = f(x, s) is defined as
Z
f(x, s) = est f (x, t)dt
0

With this, the following useful formulae are obtained

f (x, t) df(x, s)
L[ ]=
x dx

2 f (x, t) d2 f(x, s)
L[ ] =
x2 dx2

f (x, t)
L[ ] = sf(x, s) f (x, 0)
t

2 f (x, t) f (x, 0)
L[ 2
] = s2 f(x, s) sf (x, 0)
t t

9.5.2 Inversion of Laplace Transforms


The inverse Laplace transform of L[F (t)] is

1 Z +i st
F (t) = e F (s)ds
2i s=i
The integration must be carried out in the complex domain and requires a basic understand-
ing of analytic functions of a complex variable and residue calculus. Fortunately, inversion
formulae for many transforms have been computed and are readily available in tabular form.
A short table of useful transforms is included below.

25
F (s) F (t)
1
s
1
1
s2
t
1 tn1
sn (n1)!
as
e F (s) F (t a); t a 0; t < a
k2
ek s k e 4t
2 t 3

1 k s k
e erf c( 2 )
s
aek s ak a2 t
t

s(a+ s)
e e erf c(a t + k

s t
)+ k
erf c( 2 t
)
P 2 2
J0 (ix s) en t/a J0 (n x/a)
sJ0 (ia s)
1 2
n=1 n J1 (n )
; n roots of J0 (n ) = 0

9.5.3 Application of Laplace Transform Method to a Parabolic Problem


Consider the problem of determining the bounded function u(x, t) in the semi infinite domain
x 0 which satisfies

2u 1 u
2
=
x t
subject to the initial condition

u(x, 0) = 0

and the boundary condition

u(0, t) = u0

where u0 is constant.
Taking Laplace transforms of the PDE yields

d2 u(x, s) 1 s
= [su(x, s) u(x, 0)] = u(x, s)
dx2
with general solution

u(x, s) = Aex s/ + Bex s/

But since u(x, t) must be bounded, so is u(x, s) and A = 0. Now taking the Laplace transform
of the boundary condition gives
u0
u(0, s) =
s

26
Combining the above thus yields
u0 (x/)s
u(x, s) = e
s
Finally, inverting the transform gives
x
u(x, t) = u0 erf c( )
2 t
where the complementary error function erf c(z) = 1 erf (z) is given by
Z
2
erf c(z) = exp(z 02 )dz 0
z

9.6 Method of Variation of Parameters


Note that success with the separation of variables method requires homogeneity in all the
boundary conditions or time independence since this reduces the original problem to a set
of characteristic value problems and the desired solution can be obtained by superposition.
When inhomogeneities are not easily removed alternative methods must be used.
In the method of variation of parameters existing inhomogeneities are first assumed
nonexistent and the solution to this problem is expressed as a linear combination of charac-
teristic functions. Next the unknown coefficients are replaced by functions of the independent
variable which is missing in the eigenfunction solution. Finally, from the above one deduces
(ordinary) differential equations for these functions whose solutions are then used to obtain
the desired solution.
Consider the non homogeneous heat conduction problem in a rod
2 u T
= h(x, t)
x2 t
subject to

u(0, t) = u(L, t) = 0

and

u(x, 0) = f (x)

If h = 0 the solution, obtained by separation of variables, is



X 2 2 t/L2 nx
u(x, t) = an en sin( )
n=1 L

where the an s are the Fourier coefficients determined by requiring the solution to satisfy the
prescribed initial condition.

27
The method of variation of parameters consists of replacing the an s by functions Cn (t) =
an (t) exp(n2 2 t/L2 ) which must satisfy the non homogeneous heat equation as well as the
initial condition, i.e.

X nx
u(x, t) = Cn (t) sin( )
n=1 L
with
2ZL nx
Cn (t) = u(x, t) sin( )dx
L 0 L
Now, differentiating, using the heat equation and integrating by parts yields
Z
dCn (t) n2 2 L nx
+ Cn (t) = h(x, t) sin( )dx
dt L 0 L
This is a first order ODE where the constant of integration is determined from the initial
condition giving
2 t/ 2 Z t n2 (t )/ Z L nx
Cn (t) = an en e [ h(x, ) sin( )dx]d
L 0 0 L
where
Z L
2 nx
an = f (x) sin( )dx
L 0 L
Consider now the problem
2 u u
= h(x, t)
x2 t
subject to
u(0, t) = G(t)

u(L, t) = F (t)
and
u(x, 0) = f (x)
First, homogenize by subtracting from u the function (x, t) = (x/L)F (t) + (1 x/L)G(t)
to obtain the new function v(x, t) satisfying
2 v v
2

= h (x, t) =
x t
x 0 x
h(x, t) + F (t) + (1 )G0 (t)
L L

28
homogeneous boundary conditions and the initial condition
x x
v(x, 0) = f (x) = f (x) F (0) (1 )G(0)
L L
The solution for v(x, t) is

X nx
v(x, t) = Cn (t) sin( )
n=1 L

If h = G = f = 0
X
x nx
u(x, t) = F (t) + Cn (t) sin( )
L n=1 L

where
2 t/ 2 Z t 2 / Z L x 0 nx
Cn (t) = en {an e [ F ( ) sin( )dx]d }
L 0 0 L L
with
Z L
2 x nx
an = F (0) sin( )dx
L 0 L L
If a suitable homogenizing function u cannot be determined one assumes instead that
the solution is of the form
2ZL nx
u(x, t) = Kn (t) sin( )
L 0 L
so that
Z L
2 nx
Kn (t) = u(x, t) sin( )dx
L 0 L
From the above, necessarily

dKn (t) n2 2n 2ZL nx


+ = [G(t) + (1)n+1 F (t)] h(x, t) sin( dx
dt L 0 L
subject to the initial condition
Z L
2 nx
Kn (0) = f (x) sin( )dx
L 0 L
Substitution of the solution to this IVP into the expression for T above is the solution to
the problem.

29
10 Hyperbolic Problems
Hyperbolic problems consist of determining functions satisfying an hyperbolic partial differ-
ential equation together with suitable initial and boundary conditions. The standard second
order hyperbolic linear homogeneous partial differential equation is the so called wave equa-
tion. It describes the amplitude u(r, t) of a wave propagating with speed c, as a function of
position and time

1 2u
2 u =
c2 t2
Here we discuss a few selected solutions for this equation subject to appropriate conditions.

10.1 Vibration of a Taut Elastic String


Consider the case of a taut elastic string of length L = 1,which is held firmly at both ends,
deflected initially from its equilibrium position by an amount f (x) and released with zero
velocity. The mathematical formulation of the problem requires determing the function
u(x, t) that satisfies

2u 1 2u
=
x2 c2 t2
where c is the velocity of propagation of elastic waves in the string. Moreover, the applicable
boundary and initial conditions in this case are

u(0, t) = u(1, t) = 0

u(x, 0) = f (x)
u
|(x, 0) = 0
t
Assuming that a solution exists of the form u(x, t) = X(x)T (t), substitution into the
above yields

X 00 + k 2 X = 0

subject to X(0) = X(1) = 0, and

T 00 + k 2 T = 0

subject to T 0 (0) = 0. Admissible solutions for X(x) and T (t) are, respectively

Xn (x) = An sin(nx)

30
Tn (t) = Bn cos(nt)

where n = 1, 2, .... The most general solution then has the form

X
u(x, t) = an sin(nx) cos(nt)
n=1

The remaining initial condition requires that



X
u(x, 0) = f (x) = an sin(nx)
n=1

which yields the Fourier coefficients


R1
f (x) sin(nx)
an = 0R 1 2
0 (sin(nx))

10.2 Vibration of a Circular Membrane


Consider a thin, tightly stretched and uniformly stressed elastic circular membrane subjected
to the distributed normal load resulting from its own inertia after being released from rest
at time t = 0 from an initial deflection F (r, ).
The equation governing the (small) deflection u(r, ) perpendicular to the plane of the
(undamped) membrane is the wave equation

2 u 1 u 1 2u 1 2u
2 u = + + =
r2 r r r2 2 c2 t2
where c = T / and T is the tension in the membrane and is its density.
If one assumes that a particular solution up of the form

up (r, , t) = R(r)()T (t)

exists then, necessarily

T 00 + 2 T = 0

00 + k 2 = 0

2 2
r2 R00 + rR0 + ( r k 2 )R = 0
c2
where 2 /c2 and k 2 are separation constants.

31
The signs in front of the separation constants have been chosen based on physical con-
siderations. The solution is expected to be periodic not only with respect to the azimuthal
direction but also with respect to time t.
General solutions for these ODEs are

T = E cos t + F sin t

= C cos k + D sin k

r r
R = AJk ( ) + BYk ( )
c c
Single-valuedness requires that k = m = 1, 2, ...; boundedness requires that B = 0 and
the boundary condition w(a, , t) = 0 requires that be the positive roots of Jm (a/c) = 0,
mn .
The resulting solution is then
X
X
mn r
u(r, , t) = Jm ( )[(amn cos m + bmn sin m) cos mn t +
m=0 n=1 c
+(cmn cos m + dmn sin m) sin mn t]

The constants must be determined so that the prescribed initial conditions are also
satisfied. If the initial deflection is radially symmetric (i.e. F (r, ) = F (r)) only the m = 0
terms remain and the solution becomes

X n r
u(r, t) = An J0 ( ) cos n t
n=1 c

where
Z a
2 n r
An = rF (r)J0 ( )dr
a [J1 (n a/c)]2
2 0 c

The solution is clearly a superposition of deflection modes with frequencies equal to n /2


where n are the positive roots of
n a
J0 ( )=0
c

10.3 Traveling Waves


Recall that the general solution u(x, t) of the wave equation uxx = (1/c2 )utt has the form

u(x, t) = f (x ct) + g(x + ct)

32
The solution f (xct) can be visualized as a constant curve moving in the positive x-direction
at speed c while g(x + ct) is another constant curve moving in the negative x-direction with
speed c. These are called traveling waves.
Consider a tightly stretched string which at time t = 0 is released from an initial deflection
f (x) with a velocity g(x) (Cauchy conditions). The lateral deflection of the string satisfies
the wave equation
2u 1 2u
= 2 2
x2 c t
subject to
u(x, 0) = f (x)

ut (x, 0) = g(x)
It can be shown that a solution of the above is given by DAlemberts formula
Z
1 1 x+ct
u(x, t) = [f (x + ct) + f (x ct)] + g()d
2 2c xct
Further, time-periodic solutions to the wave equation have the form
u(x, t) = F (x)eit
so that the wave equation becomes
2
F 00 + F =0
c2
leading to the general solution
u(x, t) = c1 ei(x+ct)/c + c2 ei(xct)/c
Which represents periodic plane waves moving along the x-axis.
Time periodic spherical and cylindrical waves can be similarly defined starting from the
one dimensional wave equation in the corresponding system of coordinates. In this case
u(r, t) = F (r)eit
For spherical waves the general solution is
1
u(r, t) = [f (r ct) + g(r + ct)] = F (r)eit =
r
ei(r+ct)/c ei(rct)/c
= c1 + c2
r r
representing inward and outward traveling waves.
While for cylindrical coordinates the general solution is
(1) r (2) r
u(r, t) = c1 eit H0 ( ) + c2 eit H0 ( )
c c
(1) (2)
representing again inward and outward traveling waves. Here H0 and H0 are the Hankel
functions of order zero of first and second kinds, respectively.

33
10.4 Pulsating Cylinder
Consider a rigid cylinder pulsating with frequency (radial speed V0 cos t) inside an ideal
compressible fluid. The velocity potential satisfies the wave equation
2 1 1 2
+ =
r2 r r c2 t2
where /r = V is the radial velocity component. The solution, corresponding to outward
traveling waves as r is the real part of
(2) r
c = F (r)eit = c2 eit H0 ( )
c
where c2 , from the boundary condition, is
c V0
c2 = (2)
H1 (r/c)

so that the velocity field is


(2)
it H1 (r/c)
V = Re{V0 e (2)
}
H1 (a/c)
And as r
r
a ra
V V0 cos[(t )]
r c

10.5 Application of Laplace Transform Method to the Wave Equa-


tion
Consider as a second example of the Laplace transform method the problem of finding the
solution to the the wave equation in 0 x <
2u 1 2u
= 2 2
x2 c t
subject to

u(0, t) = 0

u(x, 0) = 0

u(x, 0)
=1
t

34
Taking Laplace transforms one obtains
d2 u s2 1
u =
dx2 c2 c2
subject to
u(0, s) = 0
The bounded solution of the above ODE is
1 1
u(x, s) = 2 2 esx/c
s s
and applying the inverse transform the desired solution becomes
(
0; t x/c
u(x, t) = x
c
; t x/c

11 Exercises
1.- Consider the following functions z(x, y)
z = A(x2 + y 2 )
z = B exp(x2 + y 2 )
z = C sin(x2 + y 2 )
where A, B, C are constants and show they are all general solutions of the first order partial
differential equation
z z
y x =0
x y

2.- Assume that the function (x, y, z) satisfies the equation 2 = 0. Show then that
the function (x, y, z) = (ax + by + cz)(x, y, z) where a, b, c are constants, satisfies the
equation 4 = 0.

3.- Find a solution u(x, y) to the following problem in x [0, 1], y [0, 1],
2u 2u
+ =0
x2 y 2
subject to
u(0, y) = u(1, y) = u(x, 0) = 0
u(x, 1) = sin(x)

35
using the separation of variables method.

4.- Show that the solution u(x, y) to the following problem in x [0, 1], y [0, ],

2u 2u
+ =0
x2 y 2
subject to

u(0, y) = u(1, y)
u(x, y ) 0
u(x, 0) = f (x)

has the form



X
u(x, y) = cn sin(nx) exp(ny)
n=1

with
Z 1
cn = 2 f (x) sin(nx)dx
0

5.- Write down the expression for the transient, three dimensional heat equation in gen-
eralized orthogonal curvilinear coordinates. Then, introduce the appropriate scale factors
and produce expressions for the heat equation in rectangular Cartesian, cylindrical polar
and spherical coordinates.

6.- Find a solution u(x, t) to the following problem in x [0, 1], t > 0,

2 u u
=0
x2 t
subject to

u(0, t) = u(1, t) = 0
u(x, 0) = 1

using the separation of variables method.

7.- Find a solution u(r, z) to the following problem in r [0, a], z [0, Z],

1 u 2u
(r ) + 2 = 0
r r r z

36
subject to

u(0, z) = u(a, z) = u(r, 0) = 0


r
u(r, Z) = cos( )
2a
using the separation of variables method.

8.- Find a solution u(x, t) to the following problem in x [0, 1], t > 0,

2 u u
=0
x2 t
subject to

u(0, t) = u(x, 0) = 0
u(0, t) = t

using Duhamels method.

9.- Use the Laplace transform method to determine the function u(x, t) satisfying

2u 2u
=
x2 t2
subject to

u(0, t) = u(1, t) = 0

u(x, 0) = sin(x)

10.- Find a solution u(x, y) to the following problem for x [0, 1], y [0, 1],

4u 4u 4u
+ + =1
x4 y 4 x2 y 2
subject to

2u 2u
u(0, y) = u(1, y) = |x=0 = |x=1 = 0
x2 x2
2u 2u
u(x, 0) = u(x, 1) = 2 |y=0 = |y=1 = 0
y y 2

37

You might also like