MM 1

Mathematical Methods for Physics
Peter S. Riseborough
June 18, 2018
Contents
1 Mathematics and Physics 5
2 Vector Analysis 6
2.1 Vectors . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.2 Scalar Products . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.3 The Gradient . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.4 The Divergence . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.5 The Curl . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.6 Successive Applications of ∇ . . . . . . . . . . . . . . . . . . . . 12
2.7 Gauss’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 13
2.8 Stokes’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
2.9 Non-Orthogonal Coordinate Systems . . . . . . . . . . . . . . . . 16
2.9.1 Curvilinear Coordinate Systems . . . . . . . . . . . . . . . 18
2.9.2 Spherical Polar Coordinates . . . . . . . . . . . . . . . . . 19
2.9.3 The Gradient . . . . . . . . . . . . . . . . . . . . . . . . . 21
2.9.4 The Divergence . . . . . . . . . . . . . . . . . . . . . . . . 21
2.9.5 The Curl . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.9.6 Compounding Vector Differential Operators in Curvilin-
ear Coordinates . . . . . . . . . . . . . . . . . . . . . . . . 23
3 Partial Differential Equations 27

3.1 Linear First-Order Partial Differential Equations . . . . . . . . . 31
3.2 Classification of Partial Differential Equations . . . . . . . . . . . 37
3.3 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . 48
3.4 Separation of Variables . . . . . . . . . . . . . . . . . . . . . . . . 48
4 Ordinary Differential Equations 64

4.1 Linear Ordinary Differential Equations . . . . . . . . . . . . . . . 66
4.1.1 Singular Points . . . . . . . . . . . . . . . . . . . . . . . . 67
4.2 The Frobenius Method . . . . . . . . . . . . . . . . . . . . . . . . 68
4.2.1 Ordinary Points . . . . . . . . . . . . . . . . . . . . . . . 69
1
4.2.2 Regular Singularities . . . . . . . . . . . . . . . . . . . . . 74
4.3 Linear Dependence . . . . . . . . . . . . . . . . . . . . . . . . . . 86
4.3.1 Linearly Independent Solutions . . . . . . . . . . . . . . . 90
4.3.2 Abel’s Theorem . . . . . . . . . . . . . . . . . . . . . . . . 90
4.3.3 Other Solutions . . . . . . . . . . . . . . . . . . . . . . . . 91
5 Stürm Liouville Theory 93

5.1 Degenerate Eigenfunctions . . . . . . . . . . . . . . . . . . . . . . 98
5.2 The Inner Product . . . . . . . . . . . . . . . . . . . . . . . . . . 99
5.3 Orthogonality of Eigenfunctions . . . . . . . . . . . . . . . . . . . 100
5.4 Orthogonality and Linear Independence . . . . . . . . . . . . . . 102
5.5 Gram-Schmidt Orthogonalization . . . . . . . . . . . . . . . . . . 102
5.6 Completeness of Eigenfunctions . . . . . . . . . . . . . . . . . . . 106
6 Fourier Transforms 112

6.1 Fourier Transform of Derivatives . . . . . . . . . . . . . . . . . . 113
6.2 Convolution Theorem . . . . . . . . . . . . . . . . . . . . . . . . 116
6.3 Parseval’s Relation . . . . . . . . . . . . . . . . . . . . . . . . . . 117
7 Fourier Series 119

7.1 Gibbs Phenomenon . . . . . . . . . . . . . . . . . . . . . . . . . . 123
8 Bessel Functions 133

8.0.1 The Generating Function Expansion . . . . . . . . . . . . 133
8.0.2 Series Expansion . . . . . . . . . . . . . . . . . . . . . . . 134
8.0.3 Recursion Relations . . . . . . . . . . . . . . . . . . . . . 134
8.0.4 Bessel’s Equation . . . . . . . . . . . . . . . . . . . . . . . 136
8.0.5 Integral Representation . . . . . . . . . . . . . . . . . . . 137
8.0.6 Addition Theorem . . . . . . . . . . . . . . . . . . . . . . 138
8.0.7 Orthonormality . . . . . . . . . . . . . . . . . . . . . . . . 144
8.0.8 Bessel Series . . . . . . . . . . . . . . . . . . . . . . . . . 146
8.1 Neumann Functions . . . . . . . . . . . . . . . . . . . . . . . . . 153
8.2 Spherical Bessel Functions . . . . . . . . . . . . . . . . . . . . . . 157
8.2.2 Orthogonality Relations . . . . . . . . . . . . . . . . . . . 159
8.2.3 Spherical Neumann Functions . . . . . . . . . . . . . . . . 159
9 Legendre Polynomials 162

9.0.4 Generating Function Expansion . . . . . . . . . . . . . . . 162
9.0.5 Series Expansion . . . . . . . . . . . . . . . . . . . . . . . 163
9.0.7 Legendre’s Equation . . . . . . . . . . . . . . . . . . . . . 168
9.0.8 Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . 169
9.0.9 Legendre Expansions . . . . . . . . . . . . . . . . . . . . . 171
2
9.1 Associated Legendre Functions . . . . . . . . . . . . . . . . . . . 181
9.1.1 The Associated Legendre Equation . . . . . . . . . . . . . 181
9.1.4 Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . 188
9.2 Spherical Harmonics . . . . . . . . . . . . . . . . . . . . . . . . . 192
9.2.1 Expansion in Spherical Harmonics . . . . . . . . . . . . . 192
9.2.2 Addition Theorem . . . . . . . . . . . . . . . . . . . . . . 193
10 Hermite Polynomials 197

10.0.4 Hermite’s Differential Equation . . . . . . . . . . . . . . . 197
10.0.5 Orthogonality . . . . . . . . . . . . . . . . . . . . . . . . . 198
11 Laguerre Polynomials 201

11.0.7 Laguerre’s Differential Equation . . . . . . . . . . . . . . 202
11.1 Associated Laguerre Polynomials . . . . . . . . . . . . . . . . . . 202
12 Inhomogeneous Equations 208

12.1 Inhomogeneous Differential Equations . . . . . . . . . . . . . . . 208
12.1.1 Eigenfunction Expansion . . . . . . . . . . . . . . . . . . 209
12.1.2 Piece-wise Continuous Solution . . . . . . . . . . . . . . . 209
12.2 Inhomogeneous Partial Differential Equations . . . . . . . . . . . 218
12.2.1 The Symmetry of the Green’s Function. . . . . . . . . . . 219
12.2.2 Eigenfunction Expansion . . . . . . . . . . . . . . . . . . 230
13 Complex Analysis 242

13.1 Contour Integration . . . . . . . . . . . . . . . . . . . . . . . . . 245
13.2 Cauchy’s Integral Theorem . . . . . . . . . . . . . . . . . . . . . 248
13.3 Cauchy’s Integral Formula . . . . . . . . . . . . . . . . . . . . . . 254
13.4 Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 256
13.5 Morera’s Theorem. . . . . . . . . . . . . . . . . . . . . . . . . . . 258
14 Complex Functions 261

14.1 Taylor Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 261
14.2 Analytic Continuation . . . . . . . . . . . . . . . . . . . . . . . . 262
14.3 Laurent Series . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 265
14.4 Branch Points and Branch Cuts . . . . . . . . . . . . . . . . . . . 267
14.5 Singularities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 271
3
15 Calculus of Residues 275
15.1 Residue Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . 275
15.2 Jordan’s Lemma . . . . . . . . . . . . . . . . . . . . . . . . . . . 277
15.3 Cauchy’s Principal Value . . . . . . . . . . . . . . . . . . . . . . 278
15.4 Contour Integration . . . . . . . . . . . . . . . . . . . . . . . . . 282
15.5 The Poisson Summation Formula . . . . . . . . . . . . . . . . . . 294
15.6 Kramers-Kronig Relations . . . . . . . . . . . . . . . . . . . . . . 298
15.7 Integral Representations . . . . . . . . . . . . . . . . . . . . . . . 300
4
1 Mathematics and Physics
Physics is a science which relates measurements and measurable quantities to a
few fundamental laws or principles.
It is a quantitative science, and as such the relationships are mathematical.

The laws or principles of physics must be able to be formulated as mathematical
statements.
If physical laws are to be fundamental, they must be few in number and must
be able to be stated in ways which are independent of any arbitrary choices.
In particular, a physical law must be able to be stated in a way which is in-
dependent of the choice of reference frame in which the measurements are made.
The laws or principles of physics are usually formulated as differential equa-

tions, as they relate changes. The laws must be invariant under the choice of
coordinate system. Therefore, one needs to express the differentials in ways
which are invariant under coordinate transformations, or at least have definite
and easily calculable transformation properties.
It is useful to start by formulating the laws in fixed Cartesian coordinate

systems, and then consider invariance under:-
(i) Translations
(ii) Rotations
(iii) Boosts to Inertial Reference Frames
(iv) Boosts to Accelerating Reference Frames
Quantities such as scalars and vectors have definite transformation proper-

ties under translations and rotations.
Scalars are invariant under rotations.
Vectors transform in the same way as a displacement under rotations.
5
2 Vector Analysis
2.1 Vectors
Consider the displacement vector, in a Cartesian coordinate system it can be
expressed as
→
−
r = êx x + êy y + êz z (1)
where êx , êy and êz , are three orthogonal unit vectors, with fixed directions.
The components of the displacement are (x, y, z).
In a different coordinate system, one in which is (passively) rotated through

an angle θ with respect to the original coordinate system, the displacement vec-
tor is unchanged. However, the components (x0 , y 0 , z 0 ) defined with respect to
the new unit vectors ê0x , ê0y and ê0z , are different.
A specific example is given by the rotation about the z-axis

→
−
r = ê0x x0 + ê0y y 0 + ê0z z 0 (2)
The new components are given in terms of the old components by

 0     
x cos θ sin θ 0 x
 y 0  =  − sin θ cos θ 0   y  (3)
z0 0 0 1 z
Hence,
→
−
r = ê0x ( x cos θ + y sin θ ) + ê0y ( y cos θ − x sin θ ) + ê0z z 0 (4)
The inverse transformation is given by the substitution θ → − θ,

→
−
r = êx ( x0 cos θ − y 0 sin θ ) + êy ( y 0 cos θ + x0 sin θ ) + êz z (5)
→
−
Any arbitrary vector A can be expressed as
→
−
A = êx Ax + êy Ay + êz Az (6)
where êx , êy and êz , are three orthogonal unit vectors, with fixed directions.
The components of the displacement are (Ax , Ay , Az ). The arbitrary vector
transforms under rotations exactly the same way as the displacement
→
−
A = ê0x ( Ax cos θ + Ay sin θ ) + ê0y ( Ay cos θ − Ax sin θ ) + ê0z A0z (7)
6
2.2 Scalar Products
Although vectors are transformed under rotations, there are quantities asso-
ciated with the vectors that are invariant under rotations. These invariant
quantities include:-
(i) Lengths of vectors.
(ii) Angles between vectors.
These invariant properties can be formulated in terms of the invariance of a
scalar product.
The scalar product of two vectors is defined as

→
− → −
A . B = Ax Bx + Ay By + Az Bz (8)
The scalar product transforms exactly the same way as a scalar under rotations,
and is thus a scalar or invariant quantity.
→
− → −
A . B = Ax B x + Ay B y + Az B z
= A0x Bx0 + A0y By0 + A0z Bz0 (9)
2.3 The Gradient

The gradient represents the rate of change of a scalar quantity φ(→
−
r ). The gra-
dient is a vector quantity which shows the direction and the maximum rate of
change of the scalar quantity. The gradient can be introduced through consid-
eration of a Taylor expansion
∂φ ∂φ ∂φ
φ(→
−
r +→
−
a) = φ(→
−
r ) + ax + ay + az + ...
∂x ∂y ∂z
→
−
= φ(→
−
r) + →−
a . ∇ φ(→
−
r ) + ... (10)
The change in the scalar qunatity is written in the form of a scalar product of
the vector displacement →
−
a given by
→
−
a = êx ax + êy ay + êz az (11)
with another quantity defined by
→
− ∂φ ∂φ ∂φ
∇φ = êx + êy + êz (12)
∂x ∂y ∂z
The latter quantity is a vector quantity, as follows from the scalar quantities
φ(→
−r ) and φ(→
−
r +→ −a ) being invariant. Thus, the dot product in the Taylor ex-
→
−
pansion must behave like a scalar. This is the case if ∇φ is a vector, since the
scalar product of the two vectors is a scalar.
7
The gradient operator is defined as
→
− ∂ ∂ ∂
∇ = êx + êy + êz (13)
∂x ∂y ∂z
The gradient operator is an abstraction, and only makes good sense when the
operator acts on a differentiable function.
The gradient specifies the rate of change of a scalar field, and the direction
of the gradient is in the direction of largest change.
An example of the gradient that occurs in physical applications is the rela-

tionship between electric field and the scalar potential
→
− →
−
E = − ∇ φ (14)
in electro-statics. This has the physical meaning that a particle will move (ac-
celerate) from regions of high potential to low potential,. The particle always
accelerates in the direction of the maximum decrease in potential.
For a point charge of magnitude q the potential is given by

q
φ = (15)
r
and the electric field is given by
→
− q→−r
E = + 3 (16)
r
2.4 The Divergence

The gradient operator, since it looks like a vector, could possibly be used in a
scalar product with a differentiable vector field. This can be used to define the
divergence of a vector as the scalar product
→
− → −
∇ . A (17)
The divergence is a scalar quantity.
In Cartesian coordinates, the divergence is evaluated from the scalar product

as
→
− → − ∂Ax ∂Ay ∂Az
∇ . A = + + (18)
∂x ∂y ∂z
8
→
−
Consider a vector quantity A of the form
→
−
A = →
−
r f (r) (19)
which is spherically symmetric and directed radially from the origin. The di-
→
−
vergence of A is given by
∂ x f (r) ∂ y f (r) ∂ z f (r) x2 + y 2 + z 2 ∂f (r)

+ + = 3 f (r) +
∂x ∂y ∂z r ∂r
∂f (r)
= 3 f (r) + r (20)
∂r
It is readily seen that the above quantity is invariant under rotations around
the origin, as is expected if the divergence of a vector is a scalar.
→
−
Another example is given by the vector t in the x − y plane which is
perpendicular to the radial vector →
−
ρ . The radial vector is given by
→
−
ρ = êx x + êy y (21)
then tangential vector is founds as

→
−
t = − êx y + êy x (22)
since it satisfies
→
− →
t .−
ρ = 0 (23)
Therefore, the divergence of the tangential vector field is zero
→
− → −
∇ . t = 0 (24)
→
−
In this example, the vector field t is flowing in closed circles, and the diver-
gence is zero.
→
−
Given a differentiable vector field A , which represents the flow of a quantity,
then the divergence represents the net inflow of the quantity to a volume and,
as such, is a scalar.
A physical example of the divergence is provided by the continuity equation

∂ρ →
− → −
+ ∇ . j = 0 (25)
∂t
→
−
where ρ is the density and j is the current density. The continuity equation
just states that the accumulation of matter in a volume (the increase in density)
9
is equal to the matter that flows into the volume.
The presence of a non-zero divergence represents a source or sink for the

flowing quantity. In electro-magnetics, electric charge acts as a source for the
electric field
→
− → −
∇ . E = 4πρ (26)
For the example of a point charge at the origin, the electric field is given by
→
− q→−r
E = (27)
r3
For r 6= 0, the divergence is found as

→
− → − ∂ x ∂ y ∂ z
∇ . E = q + +
∂x r3 ∂y r3 ∂z r3
2 2 2

3 x + y + z
= q − 3 = 0 (28)
r3 r5
which is not defined at r = 0. By consideration of Gauss’s theorem, one can see
that the divergence must diverge at r = 0. Thus, the electric field accumulates
at the point charge.
There is no source for magnetic induction field, and this shows up in the
Maxwell equation
→
− → −
∇ . B = 0 (29)
The finite magnetic induction field is purely a relativistic effect in that it rep-
resents the electric field produced by compensating charge densities which are
in relative motion.
2.5 The Curl

Given a differentiable vector field representing a flow, one can define a vector
(pseudo-vector) quantity which represents the rotation of the flow. The curl is
defined as the vector product
→
− →
−
∇ ∧ A (30)
which is evaluated as

∂ ∂ ∂
êx + êx + êx ∧ êx Ax + êy Ay + êz Az (31)
∂x ∂x ∂x
or

êx êy êz
= êx ∂Az − ∂Ay − êy ∂Az − ∂Ax + êz ∂Ay − ∂Ax
∂ ∂ ∂

= ∂x ∂y ∂z
Ax Ay Az ∂y ∂z ∂x ∂z ∂x ∂y
(32)
10
→
−
The curl of a radial vector A = → −
r f (r) is evaluated as

êx êy êz
→
− →
− ∂ ∂ ∂

∇ ∧ A =
∂x ∂y ∂z
= 0
(33)
x f (r) y f (r) z f (r)
→
−
The curl of a tangential vector t given by
→
−
t = ( êx y − êy x ) (34)
is evaluated as
→
− →
−
∇ ∧ t = − 2 êz (35)
The tangential vector represents a rotation about the z-axis in a clockwise (neg-
ative) direction.
A physical example of the curl is given by the relationship between a mag-

→
− →
−
netic induction B and the vector potential A
→
− →
− →
−
B = ∇ ∧ A (36)
The vector potential
→
− Bz
A = x êy − y êx (37)
2
produces a magnetic field

→
− →
− Bz
B = ∇ ∧ x êy − y êx
2
= êz Bz (38)
which is uniform and oriented along the z-axis.
Another example is that of a magnetic induction field produced by a straight

long current carrying wire. If the wire is oriented along the z-axis, Ampere’s
law yields
→
− I
B = x êy − y êx (39)
2 π ρ2
where
ρ2 = x2 + y 2 (40)
→
− →
−
The vector potentials A that produces this B can be found as a solution of
→
− →
− →
−
∇ ∧ A = B (41)
The solutions are not unique, one solution is given by
→
− I
A = − êz ln ρ (42)
2π
11
2.6 Successive Applications of ∇
The gradient operator can be used successively to create higher-order differen-
tials. Frequently found higher-derivatives include the divergence of a gradient
of a scalar φ
→
− → −
∇ . ∇ φ = ∇2 φ (43)
which defines the Laplacian of φ. In Cartesian coordinates one has
2
∂2 ∂2

→
− → − ∂
∇ . ∇ φ = + + φ (44)
∂x2 ∂y 2 ∂z 2
The Laplacian appears in electrostatic problems. On combining the defini-

tion of a scalar potential φ in electro-statics
→
− →
−
E = − ∇φ (45)
with Gauss’s law

→
− → −
∇ . E = 4πρ (46)
one obtains Laplace’s equation
∇2 φ = − 4 π ρ (47)
which relates the electrostatic potential to the charge density.
Another useful identity is obtained by taking the curl of a curl. It can be

shown, using Cartesian coordinates, that the curl of the curl can be expressed
as
→
− →
− →
− →
− →
− → − → −
∇ ∧ ( ∇ ∧ A ) = − ∇2 A + ∇ ( ∇ . A ) (48)
This identity is independent of the choice of coordinate systems.
This identity is often used in electromagnetic theory by combining

→
− →
− →
−
B = ∇ ∧ A (49)
and the static form of Ampere’s law

→
− →
− 4π →
−
∇ ∧ B = j (50)
c
which yields the relation between the vector potential and the current density
→
− →
− → − → − 4π →
−
− ∇2 A + ∇ ( ∇ . A ) = j (51)
c
The above equation holds only when the fields are time independent.
12
Other useful identities include
→
− →
−
∇ ∧ ( ∇ φ) = 0 (52)
and
→
− →
− →
−
∇ .( ∇ ∧ A ) = 0 (53)
2.7 Gauss’s Theorem

Gauss’s theorem relates the volume integral of a divergence of a vector to the
surface integral of the vector. Since the volume integral of the divergence is a
scalar, the surface integral of the vector must also be a scalar. The integration
over the surface must be of the form of a scalar product of the vector and the
normal to the surface area.
→
− → −
Consider the volume integral of ∇ . A
Z
→
− → −
d3 →
−
r ( ∇ . A ) (54)
For simplicity, consider the integration volume as a cube with faces oriented
parallel to the x, y and z axes. In this special case, Gauss’s theorem can be
easily proved by expressing the divergence in terms of the Cartesian components,
and integrating the three separate terms. Since each term is of the form of a
derivative with respect to a Cartesian coordinate, the corresponding integral
can be evaluated in terms of the boundary term
Z x+ Z y+ Z z+
∂Ax ∂Ay ∂Az
dx dy dz + +
x− y− z− ∂x ∂y ∂z
Z y+ Z z+ x+ Z x+ Z z+ y+ Z x+ Z y+ z+

= dy dz Ax
+ dx dz Ay
+ dx dy Az (55)
y− z− x− x− z− y− x− y− z−
The last six terms can be identified with the integrations over the six surfaces of
→
−
the cube. It should be noted that the for a fixed direction of A the integration
over the upper and lower surfaces have different signs. If the normal to the
surface is always chosen to be directed outwards, this expression can be written
as an integral over the surface of the cube
Z
→
− → −
d2 S . A (56)
Hence, we have proved Gauss’s theorem for our special geometry

Z Z
→
− → − →
− → −
d3 →
−
r ( ∇ . A ) = d2 S . A (57)
V S
13
→
−
where the surface S bounds the volume V .
The above argument can be extended to volumes of arbitrary shape, by con-

sidering the volume to be composed of an infinite number of infinitesimal cubes,
and applying Gauss’s theorem to each cube. The surfaces of the cubes internal
to the volume occur in pairs. Since the surface integrals are oppositely directed,
they cancel in pairs. This leaves only the integrals over the external surfaces of
the cubes, which defines the integral over the surface of the arbitrary volume.
————————————————————————————————–
Example:
Using Gauss’s theorem, show that the divergence of the electric field caused
by a point charge q is proportional to a Dirac delta function.
Solution:
The electric field of a point charge is given by
→
− q→−
r
E = (58)
r3
and one finds that
→
− → −
∇ . E = 0 (59)
for r 6= 0. Then, since Gauss’s theorem applied to a volume containing the
point charge becomes
Z Z
3→
− →
− → − →
− → −
d r ∇ . E = d2 S . E
− q→ −
Z
→ r
= d2 S . 3
r
= 4πq (60)
So one must have

→
− → −
∇ . E = 4 π q δ 3 (→
−
r) (61)
Therefore, since the charge density is given the divergence of the electric field,
the charge density must be represented by a Dirac delta function as is expected
for a point charge.
————————————————————————————————–
14
2.8 Stokes’s Theorem
Stoke’s theorem relates the surface integral of the curl of a vector to an integral
of the vector around the perimeter of the surface.
Stokes’s theorem can easily be proved by integration of the curl of the vector
over a square, with normal along the z-axis and sides parallel to the Cartesian
axes
Z
2→
− →
− →
−
d S . ∇ ∧ A
Z x+ Z y+
∂Ay ∂Ax
= dx dy êz . êz − (62)
x− y− ∂x ∂y
The scalar product with the directed surface selects out the z-component of the
curl. One integral in each term can be evaluated, yielding
Z y+ x+ Z x+ y+ I
→
−
d→−

= dy Ay − dx Ax = r . A (63)
y− x− x− y−
which is of the form of an integration over the four sides of the square, in which
the loop is traversed in a counterclockwise direction.
Stokes’s theorem can be proved for a general surface, by subdividing it into a

grid of infinitesimal squares. The integration over the interior perimeters cancel,
the net result is an integration over the loop bounding the exterior perimeters.
A physical example of Stokes’s theorem is found in the quantum mechanical

description of a charged particle in a magnetic field. The wave function can be
composed of a super-position of two components stemming from two paths. The
two components have a phase difference which is proportional of the integral of
the vector potential along the two paths :-
Z
→
−
d→
−
r . A (64)
path1
and Z
→
−
d→
−
r . A (65)
path2
If these two paths are traversed consecutively, but the second path is traced in
the reverse direction, then one has a loop. If the phase of the wave function at
the origin is unique, up to multiples of 2 π, then the loop integral
I
→
−
d→−r . A (66)
15
must take on multiples of a fundamental value φ0 . Stokes’s theorem leads to
the discovery that magnetic flux must be quantized as
I
→
−
d→
−r . A = n φ0
Z
→
− → −
d2 S . B = n φ 0 (67)
where n is an arbitrary integer and φ0 is the fundamental flux quantum . This

phenomenon, discovered by Dirac, is known as flux quantization.
2.9 Non-Orthogonal Coordinate Systems

One can introduce non-Cartesian coordinate systems. The most general coor-
dinate systems do not use orthogonal unit vectors. As an example, consider a
coordinate system for a plane based on two unit vectors (of fixed direction) ê1
and ê2 . The position of a point on the plane can be labeled by the components
x1 and x2 where
→
−r = ê1 x1 + ê2 x2 (68)
The length l of the vector is given by
l2 = x21 + x22 + 2 ê1 . ê2 x1 x2 (69)
The expression for the length can be written as

X
l2 = g i,j xi xj (70)
i,j
where g i,j is known as the metric tensor and is given by
g i,j = êi . êj (71)
In this prescription, the components have been given as the successive displace-
ments needed to be traversed parallel to the unit vectors to arrive at the point.
This is one way of specifying a vector. The components x1 and x2 are known
as the co-variant components.
Another way of specifying the same vector is given by specifying the com-
ponents x1 , x2 as the displacements along the unit vectors, if the point is given
by the intersection of the perpendiculars subtended from the axes. The compo-
nents x1 and x2 are the contra-variant components.
————————————————————————————————–
16
Example:
What is the relationship between the co-variant and contra-variant compo-

nents of the vector? Express the relationship and inverse relationship in terms
of the components of the metric.
Solution:
Let θ be the angle between ê1 and ê2 , such that
ê1 . ê2 = cos θ (72)
Consider a vector of length l which is oriented at an angle φ relative to the unit

vector ê1 .
The relationship between the Cartesian components of the vector and the
covariant components is given by
l cos φ = x1 + x2 cos θ
l sin φ = x2 sin θ (73)
The contra-variant components are given by
l cos φ = x1
l cos(θ − φ) = x2 (74)
Hence, we find the relationship
x1 = x1 + x2 cos θ
2
x = x2 + x1 cos θ (75)
which can be summarized as

X
xi = g i,j xj (76)
j
The inverse relation is given by

1
x1 = ( x1 − x2 cos θ )
sin2 θ
1
x2 = ( x2 − x1 cos θ ) (77)
sin2 θ
which can be summarized as
X
xi = (g i,j )−1 xj (78)
j
17
————————————————————————————————–
Example:
How does the length get expressed in terms of the contra-variant compo-
nents?
Solution:
The length is given in terms of the contra-variant components by

X
l2 = (g i,j )−1 xi xj (79)
i,j
It is customary to write the inverse of the metric as

(g i,j )−1 = gi,j (80)
so that the sub-scripts balance the superscripts when summed over.
————————————————————————————————–
2.9.1 Curvilinear Coordinate Systems

It is usual to identify generalized coordinates, such as (r, θ, ϕ), and then define
the unit vectors corresponding to the direction of increasing generalized coordi-
nates. That is êr is the unit vector in the direction of increasing r, êθ as the unit
vector in the direction of increasing θ and êϕ as the unit vector in the direction
of increasing ϕ.
If we denote the generalized coordinates as qi , then an infinitesimal change

in a Cartesian coordinate can be expressed as
X ∂xi
dxi = dqj (81)
j
∂qj
The change in length dl can be expressed as

X
dl2 = dx2i (82)
i
which becomes
X X ∂xi ∂xi
2
dl = dqj dqj 0 (83)
0 i
∂qj ∂qj 0
j,j
18
Thus, the metric is found to be
X
j,j 0 ∂xi ∂xi
g = (84)
i
∂qj ∂qj 0
The three unit vectors of the generalized coordinate system are proportional
to
X ∂xi
êqj ∝ êi (85)
i
∂qj
In general, the direction of the unit vectors depends on the values of the set of
three generalized coordinates qj ’s.
In orthogonal coordinate systems, the coordinates are based on the existence

of three orthogonal unit vectors. The unit vectors are orthogonal when the scalar
products of the unit vectors vanish, which gives the conditions
X ∂xi ∂xi
= 0 (86)
i
∂qj ∂qj 0
for j 6= j 0 . Thus, for orthogonal coordinate systems, the metric is diagonal.

0
g j,j ∝ δj,j 0 (87)
The metric is positive definite as the non-zero elements are given by

X ∂xi 2
g j,j = > 0 (88)
i
∂qj
The inverse of the metric is also diagonal and has non-zero elements g1i,i . Thus,
in this case, the co-variant and contra-variant components of a vector are simply
related by
1
xi = i,i xi (89)
g
An example of orthogonal curvilinear coordinates is given by the spherical polar
coordinate representation.
2.9.2 Spherical Polar Coordinates

In the spherical polar representation, an arbitrary displacement vector is speci-
fied by the generalized coordinates (r, θ, ϕ), such that
→
−
r = êx r sin θ cos ϕ + êy r sin θ sin ϕ + êz r cos θ (90)
19
The unit vectors are denoted by (êr , êθ , êϕ ) and are in the direction of increasing
coordinate. Thus,
∂→−
r
êr =
∂r
= êx sin θ cos ϕ + êy sin θ sin ϕ + êz cos θ (91)
and
∂→−
r
êθ ∝
∂θ
= êx r cos θ cos ϕ + êy r cos θ sin ϕ − êz r sin θ (92)
The unit vector êθ is given by
êθ = êx cos θ cos ϕ + êy cos θ sin ϕ − êz sin θ (93)
Finally, we find the remaining unit vector from

∂→
−r
êϕ ∝
∂ϕ
= − êx r sin θ sin ϕ + êy r sin θ cos ϕ (94)
which is in the x − y plane. The unit vector êϕ is given by normalizing the
above vector, and is
êϕ = − êx sin ϕ + êy cos ϕ (95)
As can be seen by evaluation of the scalar product, these three unit vectors are
mutually perpendicular. Furthermore, they form a coordinate systems in which
êr ∧ êθ = êϕ (96)
Due to the orthogonality of the unit vectors, the metric is a diagonal matrix
and has the non-zero matrix elements
g r,r = 1
θ,θ
g = r2
g ϕ,ϕ = r2 sin2 θ (97)
In terms of the metric, the unit vectors are given by

1 ∂→
−
r
êqj = p (98)
g j,j ∂qj
20
2.9.3 The Gradient
In curvilinear coordinates, the gradient of a scalar function φ is given by consid-
eration of the infinitesimal increment caused by the change in the independent
variables qj
X ∂φ
dφ = dqj (99)
j
∂qj
which, for orthogonal coordinate systems, can be written as

X
X 1 ∂φ p
dφ = êqj p . êqj0 g j 0 ,j 0 dqj 0
g j,j ∂qj
j j0

1 ∂φ
. d→−
X
= êqj p 0 r (100)
j g j,j ∂q j
Thus, in orthogonal coordinate systems, the gradient is identified as

→
− X 1 ∂φ
∇ φ = êqj p (101)
j g j,j ∂qj
In spherical polar coordinates, the gradient is given by

→
− ∂φ 1 ∂φ 1 ∂φ
∇ φ = êr + êθ + êϕ (102)
∂r r ∂θ r sin θ ∂ϕ
2.9.4 The Divergence

Gauss’s theorem can be used to define the divergence in a general orthogonal
coordinate system. In particular, applying Gauss’s theorem to an infinitesimal
volume, one has
→
− → − →
− → −
d3 →
−
r ∇ . A = d2 S . A (103)
where the elemental volume is given by
d3 →
−
p
r = Πj dqj g j,j
p
= Det g j,j Πj dqj (104)
and the elemental surface areas are given by

→
− p
d2 S i = Πj6=i dqj g j,j
1 p
= p Det g j,j Πj dqj (105)
dqi g i,i
21
Hence, from Gauss’s theorem, the divergence is given by the the sums of the
scalar product of the vector with the directed surface areas divided by the
volume element. Since the surfaces with normals in the direction of êqi occur in
pairs and are oppositely directed, one finds the divergence as the derivative
→
− → − 1 X ∂ Ai p
∇ . A = p p Det g j,j (106)
Detg j,j i ∂qi g i,i
For spherical polar coordinates, the divergence is evaluated as

→
− → − 1 ∂ 2 1 ∂
∇ . A = r sin θ A r + r sin θ A θ
r2 sin θ ∂r r2 sin θ ∂θ

1 ∂
+ r Aϕ (107)
r2 sin θ ∂ϕ
2.9.5 The Curl

In a generalized orthogonal coordinate system, Stokes’s theorem can be used to
define the curl. We shall apply Stokes’s theorem to an infinitesimal loop integral
Z I
2→− →
− →
− →
−
d S . ∇ ∧ A = d→
−
r . A (108)
The components of the curl along the unit directions êqj in the direction of
→
−
increasing qi can be evaluated over the surface areas d2 S with normals êqj .
Then, we have
p
Detg i,i
Z
2→
− →
− →
− →
− →
−
d Sj . ∇ ∧ A = ∇ ∧ A p Πi6=j dqi (109)
j g j,j
For the surface with normal in direction 1, this is given by

Z p
→
− →
− →
− →
− →
− p
d2 S 1 . ∇ ∧ A = ∇ ∧ A g 2,2 g 3,3 dq2 dq3 (110)
1
The loop integral over the perimeter of this surface becomes

I
→
− ∂
d→
−
p p p
r . A = A2 g 2,2 dq2 + A3 g 3,3 + dq2 A3 g 3,3 dq3
∂q2

p ∂ p p
− A2 g 2,2 + dq3 A2 g 2,2 dq2 − A3 g 3,3 dq3
∂q3
(111)
22
where we have Taylor expanded the vector field about the center of the in-
finitesimal surface. The lowest-order terms in the expansion stemming from the
opposite sides of the perimeter cancel. Hence, the component of the curl along
the normal to the infinitesimal surface is given by the expression

→
− →
− 1 ∂ p 3,3 ∂ p 2,2
∇ ∧ A = p g A3 − g A2 (112)
1 g 2,2 g 3,3 ∂q2 ∂q3
The expression for the entire curl vector can be expressed as a determinant
p p p
ê1 g 1,1 ê2 g 2,2 ê3 g 3,3

→
− →
− 1 ∂ ∂ ∂

∇ ∧ A = p
∂q1 ∂q2 ∂q3
(113)
Detg j,j

p p p
g 1,1 A1 g 2,2 A2 g 3,3 A3
In spherical polar coordinates, the curl is given by

êr r êθ r sin θ êϕ

→
− →
− 1 ∂ ∂ ∂

∇ ∧ A = 2 (114)
r sin θ ∂r ∂θ ∂ϕ

Ar r A θ r sin θ Aϕ
2.9.6 Compounding Vector Differential Operators in Curvilinear Co-

ordinates
When compounding differential operators, it is essential to note that operators
should be considered to act on an arbitrary differentiable function. Thus, since
∂ ∂
x f (x) = x f (x) + f (x) (115)
∂x ∂x
∂
one can compound ∂x and x via
∂ ∂
x = x + 1 (116)
∂x ∂x
The order of the operators is important. The differential operator acts on every-
thing in front of it, which includes the unit vectors. In Cartesian coordinates,
the directions of the unit vectors are fixed thus,
∂ ∂ ∂
êx = êx = êx = 0 (117)
∂x ∂y ∂z
23
etc. For curvilinear coordinates this is no longer true. For example, in spherical
polar coordinates although the directions of the unit vectors are not determined
by the radial distance
∂ ∂ ∂
êr = êθ = êϕ = 0 (118)
∂r ∂r ∂r
the other derivatives of the unit vectors are not zero, as
∂
êr = êθ
∂θ
∂
êθ = − êr
∂θ
∂
êϕ = 0 (119)
∂θ
and
∂
êr = sin θ êϕ
∂ϕ
∂
êθ = cos θ êϕ
∂ϕ

∂
êϕ = − sin θ êr + cos θ êθ (120)
∂ϕ
The Laplacian of a scalar φ can be evaluated by computing the divergence

of the gradient of φ, i.e.,
→
− →
−
∇ . ∇ φ (121)
Here it is important to note that the differential operator acts on the unit vec-
tors, before the scalar product is evaluated.
Problem:2.1
Find an explicit expression for the angular momentum operator

→
−̂ →
−
L = −i→
−
r ∧ ∇
in spherical polar coordinates.
Problem:2.2
How are the Laplacian of a scalar ∇2 ψ and the square of the angular mo-
−̂ 2
→
mentum L ψ related in spherical polar coordinates?
24
Problem:2.3
The two dimensional x − y model consists of a set of classical spins located

→
−
at the sites of a square lattice. Pairs of spins at neighboring lattice sites R and
→
− →
R +− a , have an interaction energy
− J cos(θ−→ − θ− → → )
R R +−a
→
−
where θ−→ is the angle that the spin at site R subtends to a fixed axis. For
R
positive J, the interaction is minimized if the neighboring spins are all parallel.
In the continuum limit, the interactions can be Taylor expanded in the lattice
constant a. For small a, angle θ can be considered as a continuous field θ(→ −r ),
defined at every point in the two-dimensional space. The energy of the field is
approximately given by
Z
→
− → −
E= − J a 2
d2 →
−
r ∇θ . ∇θ
The fields that extremalize the energy satisfy Laplace’s equation
∇2 θ = 0
(i) Find an expresssion for the Laplacian and the gradient in circular polar
coordinates (r, ϕ).
(ii) Show that
θ(→−
X
r) = rm ( Am cos mϕ + Bm sin mϕ )
m
satisfies Laplace’s equation everywhere.

(iii) Present arguments as to why the m should be restricted to the set of posi-
tive integers.
(iv) Show that that the energy density of these solutions diverges at the bound-
ary of the circle at R → ∞, and so do not represent low-energy excitations.
(v) Show that
θ = ln r + C
θ = ϕ + C
are solutions of Laplace’s equation, except at the point r = 0. Since the contin-
uum appozimation is valid for distances greater than a, these may be considered
good solutions, if a cut-off is introduced for distances smaller than a.
(vi) Show that

y − yi
θ(→
−
X
r) = ni tan−1 + θ0
i
x − xi
25
can also be considered at a good solution, except in the vicinity of a finite num-
ber of points (xi , yi ) where the solution is singular.
(vii) Why should the ni be restricted to positive and negative integer values?
These singular solutions have topological characters. The thermally induced

the binding or unbinding of pairs of excitations with singularities at (xi , yi ) with
(xj , yj ) which have opposite values of ni , such that of ni + nj = 0, give rise to
the topological Kosterlitz-Thouless transition.
26
3 Partial Differential Equations
The dynamics of systems are usually described by one or more partial differential
equations. A partial differential equation is characterized as being an equation
for an unknown function of more than one independent variable, which expresses
a relationship between the partial derivatives of the function with respect to the
various independent variables. Conceptually, a solution may be envisaged as
being obtained by direct integration. Since integration occurs between two lim-
its, the solution of a partial differential equation is not unique unless its value
is given at one of the limits. That is, the solution is not unique unless the
constants of integration are specified. These are usually specified as boundary
conditions or initial conditions.
Important examples are provided by the non-relativistic Schrödinger equa-

tion
h̄2 ∂ψ
− ∇2 ψ + V (→ −r ) ψ = i h̄ (122)
2m ∂t
in which the wave function ψ(→−r , t) is usually a complex function of position and
time. The one particle wave function has the interpretation that | ψ(→ −
r , t) |2
→
−
is the probability density for finding a particle at position r at time t. In
order that ψ(→−
r , t) be uniquely specified, it is necessary to specify boundary
conditions. These may take the form of specifying ψ(→ −r , t) or its derivative
→
−
with respect to r on the boundary of the three-dimensional region of interest.
Furthermore, since this partial differential equation contains the first derivative
with respect to time, it is necessary to specify one boundary condition at a
temporal boundary, such as the initial time t0 . That is the entire wave function
must be specified as an initial condition, ψ(→ −
r , t0 ).
Another important example is given by the wave equation
1 ∂2φ
∇2 φ − = f (→
−
r , t) (123)
c2 ∂t2
where φ(→ −
r , t) describes the wave motion, c is the phase velocity of the wave and
the force density f (→ −r , t) acts as a source for the waves, inside the region of in-
terest. Again appropriate boundary conditions for four-dimensional space-time
(→
−r , t) need to be specified, for the solution to be unique. Since this equation
is a second-order equation with respect to time, it is necessary to specify φ at
two times. Alternatively, one may specify φ(→ −
r , t0 ) at the initial time and its
∂φ(→−r ,t)
derivative ∂t t0 | at the initial time.
Poisson’s equation is the partial differential equation
∇2 φ = − 4 π ρ (124)
27
which specifies the scalar or electrostatic potential φ(→
−
r ) produced by a charge
→
−
density ρ( r ). The boundaries of the spatial region in which φ is to be deter-
mined may also involve charge densities on the boundary surfaces or they may
be surfaces over which φ is specified. The charge density is to be regarded as a
source for the electric potential φ.
Maxwell’s theory of electromagnetism is based on a set of equation of the

form
→
− →
−
∇ . E = 4πρ
→
− →
−
∇ . B = 0
→
−
→
− →
− 4π →− 1 ∂E
∇ ∧ B = j +
c c ∂t
→
−
→
− →
− 1 ∂B
∇ ∧ E = − (125)
c ∂t
→
− →
− →
−
for the two vector quantities E and B , where ρ and j are source terms, respec-
tively representing the charge and current densities. These can be considered
→
− →
−
as forming a set of eight equations for the six components of E and B . In gen-
eral specifying more equations than components may lead to inconsistencies.
However, in this case two equations can be thought of specifying consistency
conditions on the initial data, such as the continuity of charge or the absence
of magnetic monopoles. Since these equations are first-order in time, it is only
→
− →
−
necessary to specify one initial condition on each of the E and B fields. This is
in contrast to the wave equation, which is obtained by combining the equations
→
− →
−
for E and B , which is a second-order partial differential equation. That is, on
combining Maxwell’s equations, one can find a second-order partial differential
→
−
equation for the unknown E vector-field
→
− →
−
→
− →
− →
− 1 ∂2 E 4π ∂ j
∇ ∧ ( ∇ ∧ E ) + 2 = − (126)
c ∂t2 c2 ∂t
which in the absence of a charge density, leads to an inhomogeneous wave equa-
tion. The two initial conditions required to solve the wave equation correspond
→
− →
−
to specifying E and the derivative of E with respect to t, the last condition is
→
−
equivalent to specifying B in Maxwell’s equations.
All of the above equations posses the special property that they are linear
partial differential equations. Furthermore, they are all second-order linear par-
tial differential equations since, the highest order derivative that enters is the
second-order derivative.
Consider the homogeneous equation which is obtained by setting the source

terms to zero. In the absence of the source terms, each term in these equations
28
only involve the first power of the unknown function or the first power of a (first
or higher-order) partial derivative of the function. The solution of the partial
differential equation is not unique, unless the boundary conditions are specified.
That is, one may find more than one (linearly independent) solution for the
unknown function, such as φi for i = 1 , 2 , . . . , N . Due to the linearity,
the general solution of the homogeneous equation φ can be expressed as a linear
combination
XN
φ = C i φi (127)
i=1
where the Ci are arbitrary (complex) numbers. The constant Ci may be deter-
mined if appropriate boundary conditions and initial conditions are specified.
This is often referred to as the principle of linear superposition.
Now consider the inhomogeneous equation, that is, the equation in which the
source terms are present. If a particular solution of the inhomogeneous equation
is found as φp , then it can be seen that due to the linearity it is possible to find
a general solution as the sum of the particular solution and the solutions of the
homogeneous equation
XN
φ = φp + Ci φi (128)
i=1
The solution may be uniquely determined if appropriate boundary and initial

conditions are specified.
Non-Linear Partial Differential Equations.
By contrast, a non-linear partial differential equation involves powers of

different orders of the unknown function and its derivatives. Examples are
given by the sine-Gordon equation
1 ∂2φ
∇2 φ − − m2 sin φ = 0 (129)
c2 ∂t2
which is a second-order non-linear partial differential equation or the Korteweg-
de Vries equation
∂φ ∂φ ∂3φ
+ φ + = 0 (130)
∂t ∂x ∂x3
which describes shallow water waves in a one-dimensional channel. The Kortweg-
de Vries equation is a third-order non-linear partial differential equation as the
highest-order derivative that it contains is third-order. In these non-linear equa-
tions, the principle of linear superposition does not hold. One can not express
the general solution of the equation as a linear sum of individual solutions φi .
Both these non-linear equations are special as they have travelling wave like so-
lutions which propagate without dispersion. These special solutions are known
29
as soliton solutions.
For the Korteweg-de Vries equation one can look for soliton solutions which
propagate with velocity c. These are of the form
φ(x, t) = φ(x − c t) (131)
Substituting this form of the solution into the partial differential equation leads
to
∂φ ∂3φ
(φ − c) + = 0 (132)
∂x ∂x3
which can be integrated to yield
φ2 ∂2φ
− cφ + = κ (133)
2 ∂x2
The constant of integration κ is chosen to be zero, by specifying that φ → 0
when | x − c t | → ∞. On identifying an integrating factor of
∂φ
(134)
∂x
and multiplying the differential equation by the integrating factor, one obtains
∂φ φ2 ∂φ ∂φ ∂ 2 φ
− c φ + = 0 (135)
∂x 2 ∂x ∂x ∂x2
This can be integrated again to yield
2
φ3 φ2

1 ∂φ
− c + = γ (136)
6 2 2 ∂x
The boundary conditions can be used again to find γ = 0. The square root of
the equation can be taken, giving the solution as an integral with z = 3φc
φ(x,t)
Z
√
Z x−ct
3 c dz
√ = c dx0 (137)
z 1 − z
The integral can be evaluated, using the substitution
z = sech2 x (138)
and
dz = − 2 sech2 x tanh x dx (139)
giving
3c
φ(x, t) = (140)
√ x

cosh2 c − c t
2
30
This non-linear solution has a finite spatial extent and propagates with velocity
c, and does not disperse or spread out.
The stability of shape of the soliton solution is to be contrasted with the

behavior found from either linear equations with non-trivial dispersion relations
or with the non-linear first-order differential equation
∂φ ∂φ
= −φ (141)
∂t ∂x
which has the solution
φ = f( x − φ t ) (142)
where f (x) is the (arbitrary) initial condition. This can be solved graphically.
As the point with largest φ moves most rapidly, the wave must change its shape.
It leads to a breaker wave and may give rise to singularities in the solution after
a finite time has elapsed. That is, the solution may cease to be single-valued
after the elapse of a specific time.
3.1 Linear First-Order Partial Differential Equations

Consider the homogeneous linear first-order partial differential equation
∂φ ∂φ
+ a(x, t) = 0 (143)
∂x ∂t
with initial conditions φ(x, 0) = f (x). This is known as a Cauchy problem.
We shall solve the Cauchy problem with the method of characteristics. The
method of characteristics is a powerful method that allows one to reduce any
linear first-order partial differential equation into an ordinary differential equa-
tion. The characteristics are defined to be curves in the (x, t) plane, x(t), which
satisfy the equation
dx(t)
= a(x(t), t) (144)
dt
The solution of this equation defines a family or a set of curves x(t). The
different curves correspond to the different constants of integration, or initial
conditions x(0) = x0 .
The solution φ(x, t), when evaluated on the characteristic x(t), yields φ(x(t), t).
This has the special property that φ has a constant value along the curve x(t).
This can be shown by taking the derivative along the characteristic curve
d dx(t) ∂φ(x, t) ∂φ(x, t)

φ(x(t), t) = + (145)
dt dt ∂x ∂t
31
and since the characteristic satisfies
dx(t)
= a(x(t), t) (146)
dt
one has
d ∂φ(x, t) ∂φ(x, t)
φ(x(t), t) = a(x, t) + = 0 (147)
dt ∂x ∂t
as φ(x, t) satisfies the homogeneous linear partial differential equation. Thus,
φ(x, t) is constant along a characteristic. Hence,
φ(x(t), t) = φ(x0 , 0)
= f (x0 ) (148)
This means that if we can determine the characteristic, one can compute the
solution of the Cauchy problem.
Essentially, the method of characteristics consists of finding two non-orthogonal

curvilinear coordinate systems ξ and η. The initial data is specified on one co-
ordiate system ξ = 0, and an η is found such that the system conserves the
initial values along the curves of constant η.
————————————————————————————————–
Example: 3.1.1
Consider the Cauchy problem

∂φ ∂φ
+ c = 0 (149)
∂t ∂x
with initial conditions
φ(x, 0) = f (x) (150)
Solution:
The characteristic is determined from the ordinary differential equation (it

has only one variable)
dx
= c (151)
dt
which has solutions
x(t) = c t + x0 (152)
Thus, the characteristic consist of curves of uniform motion with constant ve-
locity c.
32
The solution φ(x, t) is constant on the curve passing through (x0 , 0), and is
determined from
φ(x, t) = φ(x(t), t) = f (x0 ) (153)
However, on inverting the equation for the characteristic one finds
x0 = x − c t (154)
so one has
φ(x, t) = f ( x − c t ) (155)
which clearly satisfies both the initial condition and the differential equation.
The above solution corresponds to a wave travelling in the positive direction
with speed c.
————————————————————————————————–
Example: 3.1.2
Solve the Cauchy problem

∂φ ∂φ
+ x = 0 (156)
∂t ∂x
with initial conditions
φ(x, 0) = f (x) (157)
Solution:
The characteristic is determined by the ordinary differential equation

dx
= x (158)
dt
which has a solution
x(t) = x0 exp[ t ] (159)
On inverting the equation for the characteristic and the initial condition, one
has
x0 = x(t) exp[ − t ] (160)
Then, as the solution is constant along the characteristic curve
φ(x(t), t) = f (x0 ) (161)
one has
φ(x, t) = f ( x exp[ − t ] ) (162)
33
This is the solution that satisfies both the partial differential equation and the
initial condition.
————————————————————————————————–
Inhomogeneous First-Order Partial Differential Equations.
Inhomogeneous linear first-order partial differential equations can be solved

by a simple extension of the method of characteristics. Consider, the partial
differential equation
∂φ ∂φ
+ a(x, t) = b(x, t) (163)
∂t ∂x
where the inhomogeneous term b(x, t) acts as a source term. The initial condi-
tion is given by
φ(x, 0) = f (x) (164)
The characteristics are given by the solution of
dx
= a(x, t) (165)
dt
and x(0) = x0 . The solution along the characteristic is not constant due to
the presence of the inhomogeneous term, but instead satisfies
dφ(x(t), t) ∂φ dx(t) ∂φ
= +
dt ∂t dt ∂x
∂φ ∂φ
= + a(x, t)
∂t ∂x
= b(x, t) (166)
However, the solution can be found by integration along the characteristic curve.
This yields, Z t
φ(x(t), t) = f (x0 ) + dt0 b(x(t0 ), t0 ) (167)
0
On inverting the relation between x(t) and x0 , and substituting the resulting
relation for f (x0 ) in the above equation, one finds the solution.
————————————————————————————————–
Example: 3.1.3
Consider the inhomogeneous Cauchy problem

∂φ ∂φ
+ c = λx (168)
∂t ∂x
34
where the solution has to satisfy the initial condition φ(x, 0) = f (x).
Solution:
The characteristics are found as
x(t) = x0 + c t (169)
and the solution of the partial differential equation along the characteristic is
given by
Z t
φ(x(t), t) = f (x0 ) + λ dt0 x(t0 )
0
c 2
= f (x0 ) + λ ( x0 t + t ) (170)
2
Since the characteristic can be inverted to yield the initial condition as x0 =
x − c t, one has the solution

ct
φ(x, t) = f ( x − c t ) + λ t x − (171)
2
which completely solves the problem.
————————————————————————————————–
Example: 3.1.4
Find the solution of the inhomogeneous Cauchy problem

∂φ ∂φ 1
+ c = − φ(x, t) (172)
∂t ∂x τ
subject to the initial condition
φ(x, 0) = f (x) (173)
Solution:
The characteristic is given by
x(t) = x0 + c t (174)
The ordinary differential equation for φ(x(t), t) evaluated on a characteristic is

dφ(x(t), t) 1
= − φ(x(t), t) (175)
dt τ
35
t
φ(x(t), t) = φ(x(0), 0) exp[ − ]
τ
t
= f ( x − c t ) exp[ − ] (176)
τ
which is a damped forward travelling wave.
————————————————————————————————–
Example:3.1.5
Find the solution of the equation

∂φ ∂φ
x − t + t2 φ = t2 (177)
∂x ∂t
for x 6= 0 and t 6= 0.
Solution:
The characteristics x(t) satisfy

dx x
= − (178)
dt t
xt = η (179)
where η is an arbitrary constant. The ordinary differential equation for φ(x, t(x))
on the characteristic t(x) is given by
dφ η2 η2
x + 2 φ = 2 (180)
dx x x
for constant η. On multiplying by the integrating factor
η2

exp − (181)
2 x2
and integrating, one finds that the general solution is given by
2
η2

η
φ(x, t(x)) = exp f (η) + exp − (182)
2 x2 2 x2
Hence, on substituting for η, one obtains the solution
2
t
φ(x, t) = exp f (xt) + 1 (183)
2
36
Alternatively, the characteristic curve x(t) could have been specified para-
metrically in terms of the variable ξ
dx
= x (184)
dξ
dt
= −t (185)
dξ
which can be solved independently to yield
x = exp[ξ]
t = η exp[−ξ] (186)
In terms of ξ, the partial differential equation becomes

dφ
+ η 2 exp[−2ξ] φ = η 2 exp[−2ξ] (187)
dξ
On multiplying the equation by the integrating factor
η2

exp − exp[−2ξ] (188)
2
The resulting equation can be solved by integration and substitution of the t = 0
boundary condition.
3.2 Classification of Partial Differential Equations

Second-order partial differential equations can be classified into three types, El-
liptic, Parabolic and Hyperbolic. The classification is based upon the shape of
surfaces on which the solution φ is constant, or on which φ has the maximal
variation. These surfaces correspond to the wave fronts in space-time (→ −
r , t),
and the normals to the surfaces which correspond to the direction of propaga-
tion of waves and are called the characteristics. That is, one finds coordinates
that correspond to either the wave fronts or the light rays of geometrical optics.
To motivate the discussion, consider a general second-order partial differen-

tial equation
∂2φ ∂2φ ∂2φ ∂φ ∂φ

A + 2 B + C + D + E + F = 0 (189)
∂x2 ∂x ∂t ∂t2 ∂x ∂t
where A, B, C, D, E and F are smooth differentiable functions of x and t. Sup-
pose we are trying to obtain a Frobenius series of φ in the independent variables
37
∂ n φ(x,0)
x and t. If φ(x, 0) is given, then all the derivatives ∂xn can be obtained by
∂φ(x,t)

direct differentiation. If the first-order derivative ∂t is given, then the
t=0
n+1
derivatives ∂ ∂t ∂x
φ(x,t)

n can also be obtained by direct differentiation.
t=0
∂ 2 φ(x,t)

These two pieces of initial data allow the second derivative ∂t2 to
t=0
be evaluated, by using the differential equation,
∂2φ A ∂2φ B ∂2φ D ∂φ E ∂φ F

2
= − 2
− 2 − − − (190)
∂t C ∂x C ∂x ∂t C ∂x C ∂t C
Also by repeated differentiation of the differential equation with respect to t, all
the higher-order derivatives can be obtained. Hence, one can find the solution
as a Taylor expansion
2
t0 ∂ 2 φ(x, t)

0 0 ∂φ(x, t)

φ(x, t ) = φ(x, 0) + t + + ... (191)
∂t t=0 2 ∂t2 t=0
which gives a solution

within a radius of convergence. Thus, if φ(x, 0) and the
∂φ(x,t)

derivative ∂t are specified, then φ(x, t) can be determined. However, it
t=0
is essential that all the appropriate derivatives exist. In this method, the initial
conditions where specified in a special coordinate system t = 0.
Generally, initial conditions may be specified in more general coordinate sys-

tems (ξ(x, t), η(x, t)), at ξ = 0.
Characteristics.
Consider a family of curves ξ(x, t) = const. together with another family

of curves η(x, t) = const. which act as a local coordinate system. These set of
curves must not be parallel, therefore we require that the Jacobian be non-zero
or
ξ, η ∂ξ ∂η ∂ξ ∂η
J = − 6= 0 (192)
x, t ∂x ∂t ∂t ∂x
The family of curves do not need to be mutually orthogonal.
The differential equation can be solved in the (ξ, η) local coordinate system.
We shall define the solution in the new coordinates to be
φ(x, t) = φ̃(ξ, η) (193)
Let us assume that the differential equation has boundary conditions in which
38

˜
∂ φ(ξ,η)
the function φ̃(ξ = 0, η) and its derivative ∂ξ
are given. By differentia-
ξ=0
tion with respect to η, one can determine
∂ φ̃(0, η)
∂η
∂ 2 φ̃(0, η)
(194)
∂η 2
and

∂ φ̃(ξ, η)
∂ξ ξ=0
∂ 2 φ̃(ξ, η)

(195)
∂ξ ∂η ξ=0
To obtain all the higher-order derivatives needed for a Taylor series expansion
of the solution, one must express the partial differential equation in terms of
the new variables. Thus, one has

∂φ ∂ φ̃ ∂ξ ∂ φ̃ ∂η
= + (196)
∂t x ∂ξ η ∂t x ∂η ξ ∂t x
and
2 2 2 2
∂ 2 φ̃
2
∂ φ ∂ξ ∂ φ̃ ∂ξ ∂η ∂ φ̃ ∂η
= + 2 +
∂t2 x ∂ξ 2 ∂t x ∂ξ ∂η ∂t x ∂t x ∂η 2 ∂t x
2 2
∂ φ̃ ∂ ξ ∂ φ̃ ∂ η
+ + (197)
∂ξ η ∂t2 x ∂η ξ ∂t2 x
etc. Therefore, the differential equation can be written as
2 2 2
∂ φ̃ ∂ξ ∂ξ ∂ξ ∂ξ
2
A + 2 B + C
∂ξ ∂x ∂x ∂t ∂t
2
∂ φ̃ ∂ξ ∂η ∂ξ ∂η ∂η ∂ξ ∂ξ ∂η
+ 2 A + B + + C
∂ξ ∂η ∂x ∂x ∂x ∂t ∂x ∂t ∂t ∂t
2 2 2
∂ φ̃ ∂η ∂η ∂η ∂η
+ 2
A + 2B + C
∂η ∂x ∂x ∂t ∂t
2 2 2
∂ φ̃ ∂ ξ ∂ ξ ∂ ξ ∂ξ ∂ξ
+ A + 2 B + C + D + E
∂ξ ∂x2 ∂x ∂t ∂t2 ∂x ∂t
2 2 2
∂ φ̃ ∂ η ∂ η ∂ η ∂η ∂η
+ A + 2B + C + D + E
∂η ∂x2 ∂x ∂t ∂t2 ∂x ∂t
+ F φ̃ + G = 0 (198)
39
The ability to solve this equation with the initial data given on ξ = 0, rests
on whether or not the second derivative
2
∂ φ̃
(199)
∂ξ 2
can be determined from the differential equation, since all other quantities are
assumed to be known. The second derivative w.r.t. ξ can be found if its coeffi-
cient is non-zero
2 2
∂ξ ∂ξ ∂ξ ∂ξ
A + 2B + C 6= 0 (200)
∂x ∂x ∂t ∂t
If the above expression never vanishes for any real function ξ(x, t), then the so-
lution can be found. All higher-order derivatives of φ̃ can be found by repeated
differentiation.
On the other hand, if a real function ξ(x, t) exists for which

2 2
∂ξ ∂ξ ∂ξ ∂ξ
A + 2B + C = 0 (201)
∂x ∂x ∂t ∂t
then the problem is not solvable when the initial conditions are specified on the
curve ξ = 0.
The condition of the solvability of the partial differential equation is that

the quadratic expression
2 2
∂ξ ∂ξ ∂ξ ∂ξ
A + 2B + C (202)
∂x ∂x ∂t ∂t
is non-vanishing. The types of roots of the quadratic equation are governed by
the discriminant
B2 − A C (203)
and, therefore, results in three different types of equations.
The above analysis can be extended to the case in which the coefficients
A, B and C are well-behaved functions of (x, t), since only the second-order
derivatives w.r.t. to (ξ, η) play a role in the analysis.
Hyperbolic Equations.
The case where B 2 − A C > 0 corresponds to that of hyperbolic equations.

In this case, one finds that the solvability condition vanishes on the curve
√

− B ± 2
B − AC
∂ξ ∂ξ
/ = (204)
∂x ∂t A
40
However, the slope of the curve ξ(x, t) = const. is given

∂ξ ∂ξ
dx + dt = 0 (205)
∂x ∂t
Hence, the slopes of the family of curves are given by

∂ξ
∂t

dx
= − (206)
dt ∂ξ
∂x
or
√

B ∓ B2 − A C
dt
= (207)
dx A
The solution can not be determined from Cauchy initial boundary conditions
specified on these curves. The set of curves are the characteristics of the equa-
tion. In the local coordinate system corresponding to the positive sign, we see
that 2 2
∂ξ ∂ξ ∂ξ ∂ξ
A + 2B + C = 0 (208)
∂x ∂x ∂t ∂t
The other coordinate η(x, t) can be found by choosing the negative sign. On
this family of curves, one also finds
2 2
∂η ∂η ∂η ∂η
A + 2B + C = 0 (209)
∂x ∂x ∂t ∂t
In this special coordinate system, one finds the partial differential equation
simplifies and has the canonical form
∂ 2 φ̃ ∂ φ̃ ∂ φ̃
+ α + β + γ φ̃ + δ = 0 (210)
∂ξ ∂η ∂ξ ∂η
where α, β, γ and δ are functions of ξ and η.
An example of a hyperbolic equation is given by the wave equation

2 2
∂ φ 1 ∂ φ
2
− 2
= 0 (211)
∂x c ∂t2
where A = 1 and B = 0 and C = − c12 . The equation for the characteristics

is given by
2 2
∂ξ 1 ∂ξ
− 2 = 0 (212)
∂x c ∂t
41
which factorizes as
∂ξ 1 ∂ξ
+ = 0 (213)
∂x c ∂t
and
∂η 1 ∂η
− = 0 (214)
∂x c ∂t
which has solutions
ξ = x − ct (215)
and
η = x + ct (216)
In this new coordinate system, the wave equation reduces to
2
∂ φ̃
= 0 (217)
∂ξ ∂η
∂ φ̃
A solution can be found by integrating the equation w.r.t. ξ, yielding ∂η in
∂g(η)
terms of an arbitrary function of integration η, which we will denote by ∂η .
If this is followed by integration of the differential equation w.r.t η, one finds
∂ φ̃
∂ξ in terms of an arbitrary function of integration ξ, which we will denote by
∂f (ξ)
∂ξ . By integrating the resulting two equations again to obtain expressions
for φ̃(ξ, η) and equating the forms of the two solutions, one finds that
φ̃(ξ, η) = f (ξ) + g(η)

= f ( x − c t ) + g( x + c t ) (218)
which corresponds to a forward and backward travelling wave.
————————————————————————————————–
Example:3.2.1
Find an expression for the solution of the wave equation
1 ∂2φ ∂2φ
− = 0 (219)
c2 ∂t2 ∂x2
subject to the general boundary conditions
φ(x, 0) = f (x)

∂φ
= g(x) (220)
∂t t=0
42
Solution:
The method of characteristics shows that the solution is of the form
φ(x, t) = ψF (x − ct) + ψB (x + ct) (221)
Then. at t = 0, one has
f (x0 ) = ψF (x0 ) + ψB (x0 ) (222)
Substituting x0 = x + ct, yields
f (x + ct) = ψF (x + ct) + ψB (x + ct) (223)
while, with x0 = x − ct, one obtains
f (x − ct) = ψF (x − ct) + ψB (x − ct) (224)
The derivative of φ w.r.t. t is given by
1 ∂φ dψF (x − ct) dψB (x + ct)

= − + (225)
c ∂t dx dx
Hence, at t = 0
1 dψF (x0 ) dψB (x0 )
g(x0 ) = − + (226)
c dx0 dx0
Integrating w.r.t. x0 between the upper limits x + ct and lower limits x − ct,
yields
Z x+ct
1
ψF (x−ct) + ψB (x+ct) − ψF (x+ct) − ψB (x−ct) = dx0 g(x0 ) (227)
c x−ct
Substituting equations (223) and (224) for ψF (x + ct) and ψB (x − ct) in the
above expression, results in the equation
Z x+ct
1
2 ψF (x−ct) + 2ψB (x+ct) − f (x+ct) − f (x−ct) = dx0 g(x0 ) (228)
c x−ct
Hence, after rearranging, we find the solution is given by

Z x+ct
1 1
φ(x, t) = f (x + ct) + f (x − ct) + dx0 g(x0 ) (229)
2 2 c x−ct
43
————————————————————————————————–
Example:3.2.2
Solve the wave equation equation
1 ∂2φ ∂2φ
− = 0 (230)
c2 ∂t2 ∂x2
subject to the boundary conditions
φ(x, 0) = exp[x/ξ]

∂φ
= c/ξ sin x/ξ (231)
∂t t=0
Solution:
Following from the analysis of the previous example, one finds that the
solution is
φ(x, t) = exp[x/ξ] cosh[ct/ξ] − cos(x/ξ) cos(ct/ξ) (232)
————————————————————————————————–
Example:3.2.3
Solve the equation
∂2φ ∂2φ ∂2φ

2
− 3c − 4 c2 = 0 (233)
∂t ∂t∂x ∂x2
φ(x, 0) = x2 ξ −2

∂φ
= c exp[x/ξ] (234)
∂t t=0
————————————————————————————————–
Parabolic Equations.
44
If the discriminant vanishes in a finite region, B 2 − A C = 0, then it
is clear that neither A nor C can vanish as in this case B must also vanish,
and one would only have an ordinary differential equation. If A 6= 0, and the
coefficient of the second derivative of φ̃ with respect to ξ is factorized as
2
1 ∂ξ ∂ξ
A + B (235)
A ∂x ∂t
then the second derivative of φ̃ w.r.t. ξ vanishes when

∂ξ ∂ξ
/ = − B/A (236)
∂x ∂t
corresponding to the double root. Hence, there is only a single family of char-
acteristics ξ defined by
dx A
= (237)
dt B
On transforming to the coordinate system determined by the characteristic

2 2
ξ, we have found that the coefficient of ∂∂ξφ̃2 vanishes, but the coefficient of ∂ξ∂ ∂η
φ̃
also vanishes, since the discriminant vanishes. This can be seen by examining
2
the coefficient of ∂ξ∂ ∂η
φ̃
in the differential equation and noticing that
∂ξ ∂ξ
0 = C (A + B )
∂x ∂t
∂ξ ∂ξ
= B(B + C ) (238)
∂x ∂t
∂ 2 φ̃
Thus, all the factors in the coefficient of ∂ξ ∂η vanish. Hence, parabolic equa-
tions can be put in the canonical form
∂ 2 φ̃ ∂ φ̃ ∂ φ̃
+ α + β + γ φ̃ + δ = 0 (239)
∂η 2 ∂ξ ∂η
An example of a parabolic partial differential equation is given by the diffu-

sion equation,
∂2φ ∂φ
− = 0 (240)
∂x2 ∂t
in which A = 1 and B = C = 0.
Elliptic Equations.
45
In the case when B 2 − A C < 0, the partial differential equation is elliptic.
In the case, when the discriminant is negative, there are no real roots and
thus there are no real solutions of the equation. Therefore, there is always a
solution of the partial differential equation, if the initial conditions are given.
The elliptic equation can be put in canonical form. First, the coefficient of
∂ 2 φ̃
(241)
∂ξ ∂η
is chosen to be put equal to zero. This requires that
" # " #
∂ξ ∂η ∂η ∂ξ ∂η ∂η
A +B + B +C = 0 (242)
∂x ∂x ∂t ∂t ∂x ∂t
The curve ξ that does this, satisfies

∂ξ
∂x
∂η
B ∂x + C ∂η
∂t
dt
= − = (243)
dx ∂ξ ∂η
∂t A ∂x + B ∂η
∂t
∂ 2 φ̃
Secondly, the equation is transformed such that the coefficients of ∂ξ 2 and
∂ 2 φ̃ ∂ 2 φ̃
∂η 2 are made equal, while the coefficient of is maintained to be zero. For
∂ξ∂η
convenience, we shall relabel the coordinates found from the first transformation
as (x, t), and the coefficients as A, B = 0 and C. The required transformation is
determined by the condition that the coefficients of the second-order derivatives
are equal
2 2 2 2
∂ξ ∂ξ ∂η ∂η
A + C = A + C (244)
∂x ∂t ∂x ∂t
and the coefficient of the mixed second-order derivative is kept zero

∂ξ ∂η ∂ξ ∂η
A + C = 0 (245)
∂x ∂x ∂t ∂t
where one has used B = 0. Furthermore as A C > B 2 , then A and C have

the same sign. These equations can be combined to yield
" #2 " #2
∂ξ ∂η ∂ξ ∂η
A + i = −C + i (246)
∂x ∂x ∂t ∂t
46
This condition is equivalent to
√ √

∂ξ ∂η
A = C
∂x ∂t
√ √

∂ξ ∂η
C = − A (247)
∂t ∂x
The existence of a solution of the equations for ξ and η is addressed by

assuming that η is determined. Once η is determined, then ξ can be determined
from
Z x,t " r r #
C ∂η A ∂η
ξ(x, t) = const. + dx − dt (248)
x0 ,t0 A ∂t C ∂x
The solution exists if the integral is independent of the path of integration. This
requires "r # "r #
∂ C ∂η ∂ A ∂η
= − (249)
∂t A ∂t ∂x C ∂x
which is just the condition that ξ is an analytic function, so a solution can al-
ways be found.
An example of an elliptic equation is given by Laplace’s equation

∂2φ ∂2φ
2
+ = 0 (250)
∂x ∂y 2
for which A = C = 1 and B = 0.
The characteristic equation is

2 2
∂ξ ∂ξ
0 = +
∂x ∂y
" #" #
∂ξ ∂ξ ∂ξ ∂ξ
= + i − i (251)
∂x ∂y ∂x ∂y
which has solutions

ξ = x + iy
η = x − iy (252)
In terms of these complex characteristics, the equation reduces to
2
∂ φ̃
= 0 (253)
∂ξ ∂η
47
and thus the solution is given by
φ̃(ξ, η) = F (ξ) + G(η)

= F ( x + i y ) + G( x − i y ) (254)
These are harmonic functions. The real and imaginary part of these solutions
separately satisfy Laplace’s equation.
3.3 Boundary Conditions

Boundary conditions usually take three forms:-
Cauchy Conditions.
Cauchy conditions consist of specifying the value of the unknown function

and the normal derivative (gradient) on the boundary.
Dirichlet Conditions.
Dirichlet conditions correspond to specifying the value of the unknown func-

tion on the boundary.
Neumann Conditions.
Neumann conditions consist of specifying the value of the normal derivative

(gradient) on the boundary.
Hyperbolic equations possess unique solutions if Cauchy conditions are spec-

ified on an open surface.
Elliptic equations possess unique solutions if either Dirichlet or Neumann

conditions are specified over a closed surface.
Parabolic equations posses unique solutions if either Dirichlet or Neumann

conditions are specified over an open surface.
3.4 Separation of Variables

A partial differential equation may some time be solvable by the method of
separation of variables. In this case, the partial differential equation can be bro-
ken down into a set of differential equations, one equation for each independent
48
variable.
For example, the wave equation in Cartesian coordinates takes the form
2 2 2 2
∂ φ ∂ φ ∂ φ 1 ∂ φ
+ + − = 0 (255)
∂x2 ∂y 2 ∂z 2 c2 ∂t2
This can be reduced to four one-dimensional differential equations by assuming
that
φ(x, y, z, t) = X(x) Y (y) Z(z) T (t) (256)
which, on substituting into the differential equation leads to
" 2 2 2 #
1 1 ∂2T

1 ∂ X 1 ∂ Y 1 ∂ Z
X(x) Y (y) Z(z) T (t) + + − 2 = 0
X(x) ∂x2 Y (y) ∂y 2 Z(z) ∂z 2 c T ∂t2
(257)
or on diving by φ one has
" 2 2 2 #
1 1 ∂2T

1 ∂ X 1 ∂ Y 1 ∂ Z
+ + = 2 (258)
X(x) ∂x2 Y (y) ∂y 2 Z(z) ∂z 2 c T ∂t2
The left-hand-side is independent of t and is a constant as t is varied while the

right-hand-side is only a function of t. Hence, we have
1 1 ∂2T

= K (259)
c2 T ∂t2
where K is the constant of separation. One then finds that the t dependence is
governed by 2
1 ∂ T
2
= K T (t) (260)
c ∂t2
and is given by the solution of a one-dimensional problem. The equation for the
spatial terms is given by
" 2 2 2 #
1 ∂ X 1 ∂ Y 1 ∂ Z
2
+ 2
+ = K (261)
X(x) ∂x Y (y) ∂y Z(z) ∂z 2
The function X(x) can be determined by noting that

" 2 2 # 2
1 ∂ Y 1 ∂ Z 1 ∂ X
+ − K = − (262)
Y (y) ∂y 2 Z(z) ∂z 2 X(x) ∂x2
Hence, the left-hand-side is constant while the independent variable x is varied.

Thus, it is a constant L 2
1 ∂ X
= L (263)
X(x) ∂x2
49
which is the one-dimensional problem
2
∂ X
= L X(x) (264)
∂x2
which can easily be solved. The remaining terms are governed by

" 2 2 #
1 ∂ Y 1 ∂ Z
2
+ = K − L (265)
Y (y) ∂y Z(z) ∂z 2
which involves the two constants of separation K and L. This equation can be
solved by noting that
2 2
1 ∂ Z 1 ∂ Y
− K + L = − (266)
Z(z) ∂z 2 Y (y) ∂y 2
which can be reduced from a two-dimensional problem to two one-dimensional

problems by noting that the left-hand-side is independent of y and is thus a
constant, say, − M . Then
2
∂ Y
= M Y (y) (267)
∂y 2
which can be solved. Finally, one has

2
1 ∂ Z
= K − L − M (268)
Z(z) ∂z 2
which can also be solved.
Thus, after solving the one-dimensional problems, we will have found a spe-
cific solution of the form
φK,L,M (x, y, z.t) = TK (t) XL (x) YM (y) ZK−L−M (z) (269)
Since we have a linear partial differential equation, the general solution is of the
form of a linear superposition
X
φ(x, y, z, t) = CK,L,M φK,L,M (x, y, z, t)
K,L,M
X
= CK,L,M TK (t) XL (x) YM (y) ZK−L−M (z) (270)
K,L,M
The constant coefficients CK,L,M are chosen such as to satisfy the boundary
conditions.
50
————————————————————————————————–
Example:3.4.1
The quantum mechanical three-dimensional harmonic oscillator has energy

eigenstates described by the equation
h̄2 m ω2 2
− ∇2 Ψ + r Ψ = EΨ (271)
2m 2
Solution:
This can be solved in Cartesian coordinates by using

r 2 = x2 + y 2 + z 2 (272)
and the ansatz for the eigenfunction
Ψ(x, y, z) = X(x) Y (y) Z(z) (273)
which leads to three equations
h̄2 ∂ 2 X(x) m ω2 2
− + x X(x) = Ex X(x)
2 m ∂x2 2
h̄2 ∂ 2 Y (y) m ω2 2
− + y Y (y) = Ey Y (y)
2 m ∂y 2 2
h̄2 ∂ 2 Z(z) m ω2 2
− + z Z(z) = Ez Z(z) (274)
2 m ∂z 2 2
where the constants of separation satisfy E = Ex + Ey + Ez .
————————————————————————————————–
Example:3.4.2
The energy eigenfunction for a particle of mass m with a radial potential

V (r) can be written as
h̄2
− ∇2 Ψ + V (r) Ψ = E Ψ (275)
2m
which, in spherical polar coordinates, reduces to
" #
h̄2 ∂2Ψ

1 ∂ 2 ∂Ψ 1 ∂ ∂Ψ 1
− r + 2 sin θ + 2 + V (r)Ψ = E Ψ
2 m r2 ∂r ∂r r sin θ ∂θ ∂θ r sin2 θ ∂ϕ2
(276)
51
Solve this equation.
Solution:
The energy-eigenvalue equation can be reduced to three one-dimensional

problems by separation of variables. The eigenfunction is assumed to be of the
form
Ψ(r, θ, ϕ) = R(r) Θ(θ) Φ(ϕ) (277)
On substituting this form into the equation and multiplying by r2 sin2 θ and
dividing by Ψ, it is found that the azimuthal function satisfies
∂2Φ
= − m2 Φ(ϕ) (278)
∂ϕ2
where the constant of separation is identified as m2 . The solution is denoted

by Φm (ϕ). On substituting for the ϕ dependence in terms of the constant of
separation and multiplying by sin θ Θ(θ), one finds that θ dependence is also
separable and is governed by
m2

1 ∂ ∂Θ
sin θ − Θ(θ) = − l ( l + 1 ) Θ(θ) (279)
sin θ ∂θ ∂θ sin2 θ
where the constant of separation is written as l ( l + 1 ). The solution depends

on the values of (l, m) and is denoted by Θl,m (θ). The radial dependence R(r)
satisfies the one-dimensional problem
h̄2 1 ∂ h̄2 l ( l + 1 )

2 ∂R
− r + R(r) + V (r) R(r) = E R(r) (280)
2 m r2 ∂r ∂r 2 m r2
It is seen that R(r) depends on l but not on m, and is denoted by Rl (r). Thus, in
this case, the solution of the partial differential equation can be reduced to the
solution of three one-dimensional problems described by ordinary differential
equations, which involves the two constants of separation (l, m). The general
solution can be written as the linear superposition
X
Ψ(r, θ, ϕ) = Cl,m Rl (r) Θl,m (θ) Φm (ϕ) (281)
l,m
of solutions with the different possible (l, m) values, and with expansion coeffi-
cients Cl,m .
————————————————————————————————–
Example:3.4.3
52
The solutions of the Dirac equation are also solutions of the Klein-Gordon
equation, but not vice versa. Consider the mass m(z) in the Dirac equation
to be a function of the z-spatial coordinates, where m(z) change sign at z = 0
since it iis given by
n h̄ z
m(z) = tanh (282)
ξ ξ
where n is an integer. With an appropriate boundary condition, one can derive
a Klein-Gordon equation of the form
2
E 2 2 2 2 ∂m
2
+ h̄ ∇ − m(z) c Φ = − h̄ c Φ (283)
c ∂z
Find the solutions of this equation localized near the plane z = 0.
Solution
The solution can be obtained by the method of separation of variables, by

setting
Φ = X(x) Y (y) Z(z) (284)
The equation can be reduced to the three first-order ordinary differential equa-
tions
∂2X
= − kx2 X(x)
∂x2
∂2Y
= − ky2 Y (y) (285)
∂y 2
and
n2 h̄2 h̄2
2 2

E 2 2 2 2 ∂ 2 z
− h̄ (kx +ky ) − Z(z) = − h̄ − n(n+1) sech Z(z)
c2 ξ2 ∂z 2 ξ 2 ξ
(286)
The first two ordinary differential equations have the solutions
X(x) = Ax exp[ikx x] + Bx exp[−ikx x]
Y (y) = Ay exp[iky y] + By exp[−iky y] (287)
The equation for Z(z) is the Bargmann equation. It has bound state (localized)
solutions with eigenvalues
n2 h̄2 m2 h̄2
2
E 2 2 2
− h̄ (kx + ky ) − = − (288)
c2 ξ2 ξ2
for n ≥ m > 0. The lowest-energy bound state is

z
φn (z) = sechn (289)
ξ
53
which is exponentially localized around z = 0. The decay length is given by
m0 c
h̄ . Since m = n, the dispersion relation describes massless excitations
E 2 = c2 h̄2 ( kx2 + ky2 ) (290)

These states are the topologically protected surface states of the Dirac equation.
The next lowest-energy state has the wave function

n−1 z z
φn−1 (z) = sech tanh (291)
ξ ξ
which is also a surface state. These excited states describe massive excitations
since the dispersion relation is given by
E2 h̄2
= ( 2 n − 1 ) + h̄2 (kx2 + ky2 ) (292)
c2 ξ2
nh̄
However, they are in-gap states since their masses are less than ξ .
————————————————————————————————–
Problem:
Consider the first-order partial differential equation with constant coeffi-

cients
→
− →
−
a . ∇φ = 0
where →−
a = (ax , ay ), from a geometric point of view. Solve the equation using
the method of characteristics when the initial conditions are specified on curves
that are perpendicular to the characteristics.
The equation indicates that the characteristics curves have tangents given
by →
−a . Hence, since the initial conditions are specified on curves perpendicular
to ξ, one has
ξ = →
−a .→
−
r
η = ( a ∧ êz ) . →
→
− −
r
The characteristics are found as
ξ = ax x + ay y
η = ay x − ax y
As long as

a ay
J = x
ay −ax
6= 0
54
the equation reduces to
∂φ →
−
|→
−
a |2 = →
−
a . ∇φ
∂ξ
= 0
in the transformed coordinates. The first-order linear differential equation can

be integrated w.r.t. ξ to yield
φ(ξ, η) = f (η)
where f (η) is an arbitrary function of integration.
Problem:
Consider the inhomogeneous linear partial differential equation

∂φ ∂φ
ax + ay + c φ = g(x, y)
∂x ∂y
Solve the equation using the method of characteristics. Assume that the initial
conditions are specified on curves which are perpendicular to ξ.
The equation can be transformed into

∂φ
|→
−
a |2 + c φ = g(ξ, η)
∂ξ
The solution can be written as the sum of the general solution of the homoge-
neous equation and a particular solution of the inhomogeneous equation. The
homogeneous equation can be solved by identifying the integrating factor

cξ
exp →
|−
a |2
Therefore, the homogeneous equation reduces to

∂ cξ
exp → φ = 0
|−
h
∂ξ a |2
and has the solution

cξ
φh = exp − → f (η)
|−
a |2
The particular solution is found from

∂ cξ g(ξ, η) cξ
exp → φ = exp
|− |→
− |→
−
p
∂ξ a |2 a |2 a |2
55
which is solved as
ξ
g(ξ 0 , η) c ξ0
Z
cξ
φp = exp − → dξ 0 exp
|−
a |2 |→
−
a |2 |→
−a |2
The general solution is given by φh + φp , thus
Z ξ 0
c ξ0

cξ 0 g(ξ , η)
φ(ξ, η) = exp − → f (η) + dξ exp →
|−
a |2 |→
−
a |2 |−a |2
Problem:
Consider the homogeneous linear partial differential equation with variable

coefficients
∂φ ∂φ
ax (x, y) + ay (x, y) + c(x, y) φ = 0
∂x ∂y
Describe the method of solution by generalizing the method of characteristics
from the case considered in the previous problems in which the coefficients were
constants.
Expressing the partial differentials of φ in terms of the characteristics ξ(x, y)

and η(x, y)

∂φ ∂φ ∂ξ ∂φ ∂η
= +
∂x y ∂ξ η ∂x ∂η ξ ∂x

∂φ ∂φ ∂ξ ∂φ ∂η
= +
∂y x ∂ξ η ∂y ∂η ξ ∂y
and substituting into the partial differential equation results in

∂ξ ∂ξ ∂φ ∂η ∂η ∂φ
ax (x, y) + ay (x, y) + ax (x, y) + ay (x, y) = 0
∂x ∂y ∂ξ ∂x ∂y ∂η
∂φ
The characteristics are chosen by first demanding that the coefficient of ∂η
vanishes. That is,

∂η ∂η
ax (x, y) + ay (x, y) = 0
∂x ∂y
The above equation defines the characteristic η(x, y(x)), on which

d ∂η ∂η dy
η(x, y(x)) = +
dx ∂x y ∂y x dx
= 0
56
when y(x) satisfies
dy ay (x, y)
=
dx ax (x, y)
Then, ξ can be chosen in any way such that the Jacobian is non-zero
∂(ξ, η)
J =
∂(x, y)

∂ξ ∂ξ
∂x ∂y
=

∂η ∂η
∂x ∂y
6= 0
In this case, the equation reduces to an ordinary first-order differential equation

of the form
∂φ
b(ξ, η) + c(ξ, η) φ = 0
∂ξ
where b(ξ, η) 6= 0. After identifying an integrating factor, the equation can be

re-written as
Z ξ
c(ξ 0 , η)

∂
b(ξ, η) φ exp dξ 0 = 0
∂ξ b(ξ 0 , η)

ξ
c(ξ 0 , η)
Z
0
φ(ξ, η) = exp − dξ f (η)
0 b(ξ 0 , η)
where the initial condition is given by
φ(0, η) = f (η)
Problem:
Solve the first-order linear partial differential equation

p ∂φ ∂φ
1 − x2 + = 0
∂x ∂y
subject to the boundary condition
φ(0, y) = y
57
Problem:

∂φ ∂φ
y + x = 0
∂x ∂y
φ(0, y) = exp[−y 2 ]
In which region of the (x, y) plane is the solution uniquely determined?
Problem:

∂φ ∂φ
2xt + = φ
∂x ∂t
φ(x, 0) = x
∂φ
The characteristics can be found by requiring that the coefficient of ∂η van-
ishes
∂η ∂η
2xt + = 0
∂x ∂t
with
dx
= 2xt
dt
The lines of constant η satisfy
dx
= 2 t dt
x
and has the solution
x = η exp[t2 ]
Hence, one finds that
η = x exp[−t2 ]
58
which satisfies our requirement that the coefficient vanishes. On choosing ξ = t,
the partial differential equation reduces to the ordinary differential equation
∂φ
= φ
∂ξ
Therefore, one obtains the form of the solution as
φ(ξ, η) = A(η) exp[ξ]
On imposing the boundary condition at ξ = 0 on this form, one obtains the

solution
φ(x, t) = x exp[−t2 ] exp[t]
that obeys the boundary condition.
Problem:
Solve the first-order non-linear partial differential equation

∂φ ∂φ
φ2 + = 0
∂x ∂t
using the method of characteristics, with the initial condition
√
φ(x, 0) = x
∂φ
The characteristics can be found by setting the coefficient of ∂η to zero.
This yields
∂η ∂η
φ2 + − 0
∂x ∂t
This slope of the lines with constant η are given by
dx
= φ2
dt
The characteristic η is found as
x = φ2 t + η
or
η = x − φ2 t
59
On setting ξ = t, the partial differential equation reduces to
∂φ
= 0
∂ξ
φ(ξ, η) = f (η)
On imposing the boundary conditions on this form, one obtains an implicit
equation for φ
p
φ(x, t) = x − φ2 t
which can be solved to yield
r
x
φ(x, t) = ±
1 + t
This solution is valid for x > 0 and t > 0.
Problem:
Consider the quasi-linear first-order partial differential equation

∂φ ∂φ
+ c = φ2
∂t ∂x
with boundary conditions
φ(x, 0) = cos(x)
Solve this equation using the method of characteristics. Hint: The equation can
be solved exactly the same way as an inhomogeneous equation. However, the
characteristic equations are no-longer uncoupled from φ.
The inhomogeneous equation can be written as

∂φ ∂φ
+ c = a(x, t)
∂t ∂x
where a(x, t) = φ2 . The equation can be expressed in terms of a scalar product of
a vector (1, c, a(x, y)) with the gradient ( ∂ψ ∂ψ ∂ψ
∂t , ∂x , − ∂ψ ). It should be understood
that the solution φ is a surface in (t, x, ψ) space. If one introduces a parametric
representation of the characteristics, then
∂t
1 =
∂ξ
∂x
c =
∂ξ
∂ψ
a(x, y) =
∂ξ
60
then the equation is identically satisfied since
∂ψ ∂ψ
=
∂ξ ∂ξ
The three Lagrange-Charpit equations can separately be integrated w.r.t. ξ.
The first two lead to
t(ξ, η) = ξ + c1 (η)
x(ξ, η) = c ξ + c2 (η)
where c1 (η) and c2 (η) are arbitrary functions of integration. The third equation
becomes
dψ
= dξ
a(ξ, η)
with a(x, y) = ψ 2 , which can be integrated to yield

1
− = ξ + c3 (η)
ψ
On imposing boundary conditions on ξ and η, one has
t = ξ
x = ct + η
while imposing the Cauchy boundary condition on ψ at ξ = 0 yields

1 1
= −ξ +
ψ cos η
The solution can be re-written as
cos( x − c t )
φ(x, t) =
1 − t cos( x − c t )
Problem:
Solve the quasi-linear partial differential equation

∂φ ∂φ
x − 2y = φ2
∂x ∂y
φ(x, x) = x3
61
Problem:
Solve the quasi-linear partial differential equation

∂φ ∂φ
y − x = exp[φ]
∂x ∂y
φ(0, y) = y2 − 1
Problem:
Classify the second-order linear partial differential equation
∂2φ ∂φ
+ = 0
∂x∂t ∂x
and find a general solution.
The equation is hyperbolic and can be expressed as an ordinary differential

equation for φx = ∂φ
∂x . That is, we may write the equation as
∂φx
= − φx
∂t
which has the solution for φx
φx (x, t) = φx (x, 0) exp[−t]
On integrating w.r.t. x, one finds that the solution for φ is
φ(x, t) = φ(x, 0) exp[−t] + g(t)
Problem:
Solve the non-linear partial differential equation
∂2φ ∂φ ∂φ
φ=
∂x∂y ∂x ∂y
62
by using the method of separation of variables. That is, assume the solution is
of the form
φ(x, y) = f (x) g(y)
where f (x) and g(y) are differentiable functions.
Problem: 1
The quantum mechanical three-dimensional harmonic oscillator has energy

eigenstates which are described by the equation
h̄2 m ω2 2
− ∇2 Ψ + r Ψ = EΨ
2m 2
Use the method of separation of variables in spherical polar coordinates to

reduce the equation into three one-dimensional equations for the r, θ and ϕ
dependence of ψ.
63
4 Ordinary Differential Equations
An ordinary differential equation (ODE) is an equation involving one unknown
function φ of a single variable and it’s derivatives. In physical applications, the
variable is either a position coordinate x or time t. Since the equation involves
a derivative of first or higher order, finding solution requires the equivalent of
at least one integration of the unknown function φ(x). Hence, in order that
the constants of integration be uniquely specified the equation must be sup-
plemented by at least one boundary condition. Generally, as many derivatives
occur in the ODE, many different boundary conditions are also needed. The
number of boundary conditions needed usually correspond to the degree of the
highest-order derivative. The boundary conditions usually consist of specifying
a combination of the unknown function and its derivatives at a particular value
of the variable, x0 .
Ordinary differential equations may be either linear or non-linear differential

equations. A linear ODE is an equation in which each term in the equation
only contains terms of first-order in the unknown function or its derivative. For
example,
∂2φ
+ sin ω t φ(t) = cos ω t (293)
∂t2
is a linear differential equation. An equation is non-linear if it contains powers
of the unknown function or its derivatives other than linear. For example,
∂2φ
− m2 sin φ(x) = 0 (294)
∂x2
is a non-linear differential equation. It occurs in connection with the sine-
Gordon field theory, which involves the temporal and spatial variation of the
physical field φ. Non-linear differential equations do not have systematic meth-
ods of finding solutions. Sometimes a specific solution can be found by intro-
ducing an integrating factor. In the above case, an integrating factor of ∂φ
∂x can
be used to convert the equation into the form of an exact integral
∂φ ∂ 2 φ ∂φ
− m2 sin φ(x) = 0 (295)
∂x ∂x2 ∂x
which can be integrated to yield
2
1 ∂φ
+ m2 cos φ = C (296)
2 ∂x
where C is a constant of integration. The constant of integration must be
determined from the boundary conditions. A suitable boundary condition may
be given by
lim φ(x) → 0 (297)
x→ ∞
64
In this case one may identify C = m2 . Hence, the non-linear differential
equation of second-order in the derivative has been reduced to a non-linear dif-
ferential equation only involving the first-order derivative of φ.
On using the trigonometric identity

φ
( 1 − cos φ ) = 2 sin2 (298)
2
one can take the square root of the equation to yield

∂φ φ
= ± 2 m sin (299)
∂x 2
This is a first-order non-linear differential equation. This can be solved by
writing
( ∂φ
∂x )
= 2m (300)
sin φ2
which can be integrated by changing variable to t = tan( φ4 ). A limit of the
integration can be chosen to coincide with the point at which the boundary
condition is specified
x
φ
2 ln tan = ±2mx (301)
4 0
which involves the term ln tan( φ(0)

4 ) which is another constant of integration.
If this is specified to have a value ln A, the above equation can be inverted to
yield
!
−1
φ(x) = 4 tan A exp ± m x (302)
This solution represents a static soliton excitation. Solitons of this kind exist in
a system of infinitely many gravitational pendula. In this case, the pendula are
joined rigidly to one wire. The rotation of a pendulum provides a twist in the
supporting wire which transmits a torsional force on the neighboring wire. The
angle that the pendula make with the direction of the gravitation is denoted by
φ. The gravitational force is periodic in the angle φ and is responsible for the
non-linear term. The above solution represents a rotation of successive pendula
over the top. In addition to this, one expects small amplitude oscillations of the
weights around the bottom. Since φ is expected to be small one can linearize
the equation, that is we may be able to write b
sin φ ≈ φ (303)
to obtain
∂2φ
− m2 φ(x) = 0 (304)
∂x2
65
This is a second-order linear differential equation. It can be solved to yield the
solution
φ(x) = A sin m ( x − x0 ) (305)
involving two constants of integration which are provided by two boundary
conditions. This solution represents a snapshot of the ripples of the coupled
pendula. We have assumed that the amplitude of the ripples are such that
A 1.
4.1 Linear Ordinary Differential Equations

Linear differential equations have the form
n=N
X ∂nφ
an (x) = f (x) (306)
n=1
∂xn
where the set of functions an (x) and f (x) are known. The largest value of n for
which the function an (x) is non-zero is denoted by N , then the linear differential
equation is of N -th order. If the term f (x) is zero, the equation is homogeneous
Whereas if the term f x) is non-zero, the equation is inhomogeneous.
Homogeneous linear differential equations satisfy the principle of superpo-

sition. Namely if φ1 (x) satisfies the equation and φ2 (x) also satisfies the same
equation, then the linear combination
φ(x) = C1 φ1 (x) + C2 φ2 (x) (307)
also satisfies the same linear differential equation. In general, the non-uniqueness
of solutions to an equation is due to the failure to specify appropriate boundary
conditions on φ(x). This non-uniqueness can be utilized to construct solutions
that satisfy the appropriate boundary conditions.
A particular solution of an inhomogeneous linear differential equation, φp (x),

is not unique as, if any solution of the homogeneous equation is added to the
particular solution is added it
X
φ(x) = φp (x) + Cn φn (x) (308)
n
then this φ also satisfies the same inhomogeneous equation. The non-uniqueness
does not hold when appropriate boundary conditions are specified.
For an ordinary differential equation of N -th order it is usually necessary

to specify N boundary conditions in order to obtain a unique solution. These
66
boundary conditions eliminate the N constants of integration that are found
when integrating the equation N times. Exceptions to this rule may occur if
the coefficient of the highest differential vanish at a point. This, essentially,
reduces the order of the differential equation locally. If the boundary conditions
are not used, one can develop N independent solutions of the N -th order linear
differential equation.
4.1.1 Singular Points

A homogeneous linear differential equation can be re-written such that the co-
efficient of the highest-order derivative is unity. That is, it can be written in
the form
X an (x) ∂ n φ
= 0 (309)
n
aN (x) ∂xn
For second-order differential equations, we shall use the notation
d2 φ dφ
+ P (x) + Q(x) φ = 0 (310)
dx2 dx
If both P (x) and Q(x) are finite at the point x = x0 , the point x0 is an
ordinary point. If either P (x), Q(x) or both diverge at the point x = x0 ,the
point x0 is a singular point. Example of a singular point occurs in quantum
mechanics, for example, the behavior of the radial part of the Hydrogen atom
wave function is governed by the ordinary differential equation
h̄2 1 h̄2 l ( l + 1 )

∂ 2 ∂
− r R(r) + V (r) + R(r) = E R(r)
2 m r2 ∂r ∂r 2 m r2
(311)
where the centrifugal potential diverges as r → 0. This is an example of a
singular point. Physically, what is expected to occur is that the terms V (r) and
E are negligible, near this singular point, and that the radial kinetic energy will
balance the centrifugal potential.
h̄2 1 h̄2 l ( l + 1 )

∂ 2 ∂
− r R(r) + R(r) ≈ 0 (312)
2 m r2 ∂r ∂r 2 m r2
Then, one expects that for r ∼ 0 the radial wave function will be such that
lim R(r) ∝ rα (313)

r→0
where α = l or α = − ( l + 1 ). Another example occurs in the one-

dimensional harmonic oscillator which is governed by
h̄2 ∂ 2 ψ m ω 2 x2
− + ψ = Eψ (314)
2 ∂x2 2
67
when x → ±∞. In this case, one expects that the kinetic energy term should
balance the potential energy term. This is usually handled by introducing the
variable z = x1 .
The type of singularity can be classified as being regular or irregular.
A regular singular point x0 is a singular point in which both ( x − x0 ) P (x)

and ( x − x0 )2 Q(x) remain finite in the limit x → x0 .
An irregular point x0 is a singular point in which either ( x − x0 ) P (x) or

( x − x0 )2 Q(x) diverge in the limit x → x0 .
4.2 The Frobenius Method

Given a linear differential equation one can generate an approximate solution
around a point x0 in the form of a power series in ( x − x0 ). This is the Frobe-
nius method. The idea is to develop P (x) and Q(x) in a power series around
x0 and then express the solution as a power series with arbitrary coefficients
Cn . The arbitrary coefficients are determined by demanding that each power
of ( x − x0 )n vanish identically when the series solution is substituted into
the equation. Essentially, we are demanding that the power series satisfies the
differential equation by yielding zero, and that zero only can be expanded as
a polynomial if all the coefficients are zero. This produces a set of linear alge-
braic equations, one for each value of n, which determine the set of unknown
coefficients Cn . For simple equations, these coefficients can be determined from
a recursion relation.
We shall consider the series expansion of the solution φ about the point x0 ,
in the form X
φ(x) = ( x − x0 )α Cn ( x − x0 )n (315)
n=0
Then one has

∂φ(x) X X
= α ( x − x0 )α−1 Cn ( x − x0 )n + ( x − x0 )α n Cn ( x − x0 )n−1
∂x n=0 n=1
X
= Cn ( x − x0 )α+n−1 n + α (316)
n=1
and
∂ 2 φ(x) X
= α ( α − 1 )( x − x0 )α−2 C n ( x − x0 ) n
∂x2 n=0
68
X
+ 2 α ( x − x0 )α−1 n Cn ( x − x0 )n−1
n=1
X
α
+ ( x − x0 ) n ( n − 1 ) Cn ( x − x0 )n−2
n=2
X
α+n−2
= Cn ( x − x0 ) n(n − 1) + 2nα + α(α − 1)
n=0
X
= Cn ( x − x0 )α+n−2 ( n + α ) ( n + α − 1 ) (317)
n=0
etc.
4.2.1 Ordinary Points

If x0 is an ordinary point then P (x) and Q(x) can be expanded around x0
X
P (x) = Pm ( x − x0 )m
m=0
X
Q(x) = Qm ( x − x0 )m (318)
m=0
On substituting the series expansions for P (x), Q(x) and φ(x) into the differ-
ential equation, one obtains the equation
X
α+n−2
0 = ( x − x0 ) Cn n + α n + α − 1
n=0
X m=n
X
+ ( x − x0 )α+n−1 Pm Cn−m n − m + α
n=0 m=0
X m=n
X
+ ( x − x0 )α+n Qm Cn−m (319)
n=0 m=0
On relabelling n by n + 2 in the first line of the equation and n by n + 1 in

the second line, so that the summations have formally similar exponents, one
obtains
X
0 = ( x − x0 )α+n Cn+2 n + 2 + α n + α + 1
n=0
X m=n+1
X
+ ( x − x0 )α+n Pm Cn−m+1 n − m + α + 1
n=0 m=0
X m=n
X
+ ( x − x0 )α+n Qm Cn−m
n=0 m=0
69
+ ( x − x0 )α−2 C0 α ( α − 1 )

+ ( x − x0 )α−1 α C1 ( α + 1) + P0 C0 (320)
On equating two polynomials, the coefficients of the various powers are to be

equated. Therefore, we have the set of algebraic equations consisting of two
equations
0 = α ( α − 1 ) C0

0 = α C1 ( α + 1) + P0 C0 (321)
and the set of equations valid for integer n > 0

0 = Cn+2 n + 2 + α n + α + 1
m=n+1
X
+ Pm Cn−m+1 n − m + α + 1
m=0
m=n
X
+ Qm Cn−m (322)
m=0
The first equation is known as the indicial equation, and has solutions α = 0 or
α = 1. The indicial equation can be obtained directly be examining only the
lowest-order mono-nomial. Due to the principle of superposition, the coefficient
C0 can be chosen arbitrarily.
In the case α = 0, the coefficient C1 is arbitrary.
For α = 1 the coefficient C1 is determined uniquely in terms of C0 by

P0 C0 P0 C0
C1 = − = − (323)
α + 1 2
For each positive index n, the set of equations determines Cn+2 in terms of
all the lower order expansion coefficients. For example, for n = 0 one has

1
C2 = − (α + 1) P0 C1 + (P1 α + Q0 ) C0 (324)
(α + 2)(α + 1)
For α = 0, the first few terms of the solution can be written as
Q0
φa (x) = C0 ( 1 − ( x − x0 )2 + . . . )
2!
P0
+ C 1 ( x − x0 ) ( 1 − ( x − x0 ) + . . . ) (325)
2!
70
while for α = 1, one has
P0
φb (x) = C 0 ( x − x0 ) ( 1 − ( x − x0 ) + . . . ) (326)
2!
Thus, we have found only two independent solutions of the second-order differ-
ential equation. A general solution can be expressed as a linear combination of
φa (x) with C1 = 0 and φb (x).
————————————————————————————————–
Example:
An example of an equation that can be solved by the Frobenius method is

Airy’s equation. It arises in the context of a quantum mechanical particle in
a uniform electric field. The equation for the motion of the particle along the
field direction is given by
h̄2 ∂ 2
− Ψ(z) − e Ez z Ψ(z) = E Ψ(z) (327)
2 m ∂z 2
where Ez is the field strength and E is the energy.
Solution:
Before solving this equation, it is useful to change coordinates. Let us note

that the classical turning point is determined by the point at which the potential
− e Ez z is equal to the energy E. That is, if the turning point is denoted by
z0 , then the turning point is determined from
− e Ez z0 = E (328)
We shall change variables so that the position s is measured relative to the
turning point
s = z − z0 (329)
in which case the equation takes the form
h̄2 ∂ 2
− Ψ − e Ez s Ψ = 0 (330)
2 m ∂s2
The form of the equation can be further simplified by identifying a length scale
ξ defined by
h̄2
ξ3 = (331)
2 m | e | Ez
and then introducing a dimensionless length x = ξs , one obtains
∂2
φ(x) − x φ(x) = 0 (332)
∂x2
71
where φ(x) = φ( z−z
ξ ) = Ψ(z).
0
Airy’s equation has no singularities at finite values of x, except at infinity.

We shall find a solution by series expansion around x = 0. Let us assume that
the solution has the form
∞
X
φ(x) = xα Cn xn
n=0
∞
X
= Cn xn+α (333)
n=0
which is substituted into the differential equation. One finds

∞
X ∞
X
( n + α ) ( n + α − 1 ) Cn xn+α−2 − Cn xn+α+1 = 0 (334)
n=0 n=0
The value of α is determined from the indicial equation. It is found by

examining the terms of lowest power in the above equation. This, by definition,
has to be a term in which n = 0. The term of lowest power is identified as the
term xα−2 and the coefficients are given by
α(α − 1) = 0 (335)
Thus, we either have α = 0 or α = 1.
For α = 0, the equation becomes

∞
X ∞
X
n−2
n ( n − 1 ) Cn x − Cn xn+1 = 0 (336)
n=0 n=0
Or since the first two terms on the left-hand-side are zero, one has
∞
X ∞
X
n ( n − 1 ) Cn xn−2 − Cn xn+1 = 0 (337)
n=2 n=0
Changing the summation variable from m = n − 3, one has

∞
X ∞
X
( m + 3 ) ( m + 2 ) Cm+3 xm+1 − Cn xn+1 = 0 (338)
m=−1 n=0
or
∞
X
2 C−1 + ( m + 3 ) ( m + 2 ) Cm+3 − Cm xm+1 = 0 (339)
m=0
72
On equating the coefficients of the various powers to zero, one obtains
C−1 = 0 (340)
which is consistent with our assumption that the series expansion starts with
C0 . The higher-order coefficients are related via
Cm
Cm+3 = (341)
( m + 3 ) ( m +2 )
Hence, we have a relation between the terms in which the powers of x are of the
form x3r . Iterating this relation yields
C3r−3
C3r =
(3r)(3r − 1)
C3r−6
= (342)
(3r)(3r − 1)(3r − 3)(3r − 4)
This leads to the unique determination of all the coefficients whose orders are
divisible by three.
The other solution of the indicial equation is α = 1. This solution cor-

responds to the terms in the series of which start with C1 . The coefficients
satisfy
Cm
Cm+3 = (343)
( m + 3 ) ( m +2 )
or on writing m = 3 r + 1, one has
C3r−2
C3r+1 =
(3r + 1)(3r)
C3r−5
= (344)
(3r + 1)(3r)(3r − 2)(3r − 3)
The general solution is given by

x3 x6 x3r

φ(x) = C0 1 + + + ... + + ...
2.3 2.3.5.6 2.3. . . . .( 3 r − 1 ) ( 3 r )
4 7
x3r+1

x x
+ C1 x + + + ... +
3.4 3.4.6.7 3.4. . . . .( 3 r ) ( 3 r + 1 )
∞ 3r

X x
= C0 1 +
r=1
2.3. . . . ( 3 r − 1 ) ( 3 r )
∞
x3r+1
X
+ C1 x + (345)
r=1
3.4. . . . .( 3 r ) ( 3 r − 1 )
73
Because of the denominator of the general term in the series, the series converges
for all values of x. The solutions show oscillations with increasing frequency and
decreasing amplitude for increasingly negative values of x. The solutions are
monotonic for positive values of x.
————————————————————————————————–
4.2.2 Regular Singularities

The Frobenius method, when applied at a regular singular point, x0 , yields an
indicial equation that has a non-trivial solution. The series expansion shows a
non-analytic behavior at x0 . Fuchs’s theorem states that it is always possible
to find one series solution for ordinary or regular singular points. For a regular
singular point, where
p−1
P (x) = + p0 + . . . (346)
( x − x0 )
and
q−2 q−1
Q(x) = + + q0 + . . . (347)
( x − x0 ) 2 ( x − x0 )
the indicial equation can be found by examining the lowest term in the expansion
C0 ( x − x0 )α . The indicial equation is found as
α (α − 1 ) + p−1 α + q−2 = 0 (348)
which is a quadratic equation and gives a solution

1 p
α = 1 − p−1 ± ( 1 − p−1 )2 − 4 q−2 (349)
2
————————————————————————————————–
Example:
Consider the differential equation
d2 R 2 dR l(l + 1)
− − + R = 0 (350)
dr2 r dr r2
which has a regular singular point at r = 0.
Solution:
74
The equation can be solved by applying the Frobenius method by assuming
the solution is of the form
X
R(r) = rα Cn r n (351)
n=0
The indicial equation can be found directly by examining the coefficient of the
lowest-order mononomial C0 rα . Thus, we have
α l(l + 1)
− α ( α − 1 ) C0 rα−2 − 2 C0 rα−1 + C0 rα = 0 (352)
r r2
Therefore, the indicial equation leads to
α(α + 1) = l(l + 1) (353)
or α = l or α = − ( l + 1 ). The singularity governs the behavior of

the independent solutions at r = 0. In this case, only C0 is non-zero. All the
higher-order expansion coefficients satisfy equations which only have the solu-
tions Cn = 0. This is because the equation is homogeneous in r. That is, if we
count the powers of r in each term where each derivative ∂r∂
is counted as r−1 ,
then each term has the same power. If the equation is homogeneous it does not
mix up the different powers, so the solutions must be simple powers.
————————————————————————————————–
Example:
Solve the equation
d2 R 2 dR l(l + 1)
− − + R = k2 R (354)
dr2 r dr r2
Solution:
It is convenient to change variables to x = k r. The equation for R(x/k)

becomes
d2 R 2 dR l(l + 1)
− 2
− + R = R (355)
dx x dx x2
The solution is assumed to have the form
∞
X
R(x/k) = xα Cn xn (356)
n=0
75
This ansatz is substituted into the differential equation, leading to the polyno-
mial equation
∞ ∞
X 2 X
− ( n + α ) ( n + α − 1 ) Cn x(α+n−2) − ( n + α ) Cn xn+α−1)
n=0
x n=0
X ∞
l(l + 1)
+ − 1 Cn x(n+α) = 0
x2 n=0
(357)
Combining the terms of the same power, we have
∞
X ∞
X
( n + α ) ( n + α + 1 ) − l ( l + 1 ) Cn x(α+n−2) = Cn xn
n=0 n=0
∞
X X∞
( n + α − l ) ( n + α + l + 1 ) Cn x(α+n−2) = Cn−2 xn−2
n=0 n=2
(358)
If one were to move all the non-zero terms to one side of the equation, it is
seen that this equation is equivalent to equating a polynomial with zero. A
polynomial is only equal to zero, for all values of a continuous variable x, if the
coefficients are of the polynomial are all identical to zero. That is, the coefficients
of the terms with like powers must be identical. On equating coefficients of the
same power, one finds
( n + α − l ) ( n + α + l + 1 ) Cn = − Cn−2 (359)
Then, if one assumes that C0 6= 0, on equating the terms with n = 0, one

finds the indicial equation,
α(α + 1) − l(l + 1) = 0 (360)
with solutions α = l or α = − ( l + 1 ). On equating the coefficients with
n = 1, i.e. the coefficients of the terms proportional to xα+1 , then one finds

( α + 2 ) ( α + 1 ) − l ( l + 1 ) C1 = 0 (361)
Since, for α = l or α = − ( l + 1 ) the term in the parenthesis is non-zero

one must have C1 = 0, except if l = 0.
However, the general set of linear equations reduces to the recursion relation
( n + α − l ) ( n + α + l + 1 ) Cn = ( − 1 ) Cn−2 (362)
76
This series does not terminate, and so one has
1
Cn = ( − 1 ) Cn−2 (363)
( n + α − l ) ( n +α + l + 1 )
which can be iterated to yield
n ( α − l )!! ( α + l + 1 )!!
Cn = ( − 1 ) 2 C0 (364)
( n + α − l )!! ( n + α + l + 1 )!!
where the double factorial is defined as
n!! = n ( n − 2 )!! (365)
and for even n, n = 2 m, can be written as
( 2 m )!! = 2m m! (366)
and for odd n, n = 2 m + 1, one has
( 2 m + 1 )!
( 2 m + 1 )!! = (367)
2m m!
Instead of using the explicit expression for odd n, we shall use the relation
for even n to define the factorial expression for half integer n. That is, we shall
define the half-integer factorial via

m+ 21 1
2 m + 1 !! = 2 m+ ! (368)
2
Thus, the solution regular at the origin can be represented by an infinite

power series
∞
X ( 2 l + 1 )!!
Rl (r) = xl C0 ( − 1 )m x2m
m=0
( 2 m )!! ( 2 m + 2 l + 1 )!!
∞
X ( 2 l + 1 )!!
= xl C0 ( − 1 )m x2m
m=0
2m m! ( 2 m + 2 l + 1 )!!
∞
X l!
= xl C0 1 ( − 1 )m x2m
m=0
22m m! ( m + l + 2 )!
∞ 2m
X l! x
= xl C0 1 ( − 1 )m (369)
m=0
m! ( m + l + 2 )! 2
77
where we have defined the factorial for half-integer n. It can be seen, from the
ratio test, that this series converges as the ratio of successive terms is − m12 , for
large m. The first few terms of the solution are given by
( k r )2

l
Rl (r) = C0 ( k r ) 1 − + ... (370)
2 (2 l + 3 )
This series solution is related to the Bessel function of half-integer order, also
known as the spherical Bessel function.
————————————————————————————————–
Problem:
Show that the series solution for Rl (x) satisfies the recursion relations
2l + 1
Rl+1 (x) + Rl−1 (x) = Rl (x) (371)
x
and
∂Rl (x)
l Rl−1 (x) − ( l + 1 ) Rl+1 (x) = ( 2 l + 1 ) (372)
∂x
Problem:
Verify that
sin x
j0 (x) = (373)
x
and
sin x − x cos x
j1 (x) = (374)
x2
are solutions of the differential equation for l = 0 and l = 1. Show that
on expanding these solutions in powers of x, one recovers (up to an arbitrary
multiplicative constant) the Frobenius series solutions which are non-singular
at x = 0.
Problem:
Using the above recursion relations, find explicit expressions for the solutions
for l = 2 and l = 3.
Homework: 8.5.5
Homework: 8.5.6
78
Homework: 8.5.12
————————————————————————————————–
Example:
Solve the differential equation
∂2φ ∂φ
( 1 − x2 ) − 2x + l(l + 1)φ = 0 (375)
∂x2 ∂x
Solution:
This equation has regular singular points at x = ± 1. We require that the

solution is finite for all values of x in the interval (−1, 1). The Frobenius series
is assumed to be of the form
∞
X
φ(x) = xα Cn xn (376)
n=0
Substitution of this form in the equation leads to

∞
X
2
(1 − x ) Cn ( n + α ) ( n + α − 1 ) x(n+α−2)
n=0
∞
X ∞
X
−2x Cn ( n + α ) x(n+α−1) + l ( l + 1 ) Cn x(n+α) = 0
n=0 n=0
(377)
or
∞
X
Cn ( n + α ) ( n + α − 1 ) x(n+α−2) =
n=0
∞
X
= Cn−2 (n + α − 2)(n + α − 1) − l(l + 1) x(n+α−2)
n=2
(378)
Hence, the recursion relation is given by
Cn ( n + α ) ( n + α − 1 ) = Cn−2 ( n + α − l − 2 ) ( n + α + l − 1 )
(379)
where we have defined C−2 = C−1 = 0.
79
The indicial equation is found by setting n = 0, and yields
α(α − 1) = 0 (380)
On choosing the root α = 0, one obtains solutions of even symmetry, and

the coefficients are related via
(n − l − 2)(n + l − 1)
Cn = Cn−2 (381)
n(n − 1)
For large n, the coefficients are of the same order of magnitude and the ratio
test
Cn+2 xn+2
lim = x2 (382)
n → ∞ Cn xn
provides an estimate of the radius of convergence. In fact, it can be shown that
the series diverges when x = ± 1, unless the coefficients are zero for large n.
However, when l is an integer, say N , the series truncates
(N − l)(N + l + 1)
0 = CN +2 = CN (383)
(N + 2)(N + 1)
Therefore, the solution with α = 0 is an even polynomial

C2 2 C4 4
φl (x) = C0 1 + x + x + ... (384)
C0 C0
in which the highest term is of power N = l.
For α = 1 the solution is odd in x and we find that the coefficients are
related via
(n − l + 1)(n + l)
Cn = Cn−2 (385)
n(n + 1)
For large n, the coefficients are of the same order of magnitude. So by the
ratio test, one expects that the series may diverge when x = ± 1, unless the
coefficients are zero for large n. In fact, this is the case for arbitrary l. However,
when l is an integer, say N , the series truncates,
(N − l + 1)(N + l)
0 = CN +2 = CN (386)
(N + 2)(N + 1)
and the solution is x multiplied by an even polynomial in which the highest
term is of power N = l − 1.

C2 2 C4 4
φl (x) = x C0 1 + x + x + ... (387)
C0 C0
80
Hence, in the case α = 1, the solution is an odd polynomial of order l.
————————————————————————————————–
Problem:
Show that the solution for finite l can be written as

sX
max
φl (x) = As x(l−2s) (388)

s=0
where
l

2 l even
smax = l −2 (389)
2 l odd
and the coefficients are given by
(l − 2s + 1)(l − 2s + 2)
As = − As−1
2s(2l − 2s + 1)
l!
As = ( − 1 )s A0
( l − 2 s )! 2s s! ( 2 l − 2 s + 1 ) . . . ( 2 l − 1 )
l! ( 2 l − 2 s )! 2l l!
As = ( − 1 )s A0 (390)
( l − 2 s )! 2s s! ( 2 l )! 2(l−s) ( l − s )!
Thus, the coefficients are given by
( − 1 )s ( l! )2 ( 2 l − 2 s )!
As = A0 (391)
s! ( 2 l )! ( l − 2 s )! ( l − s )!
and the series solution is determined as

sX
( l )!2 max
( − 1 )s ( 2 l − 2 s )!
φl (x) = A0 x(l−2s) (392)
( 2 l )! s=0
s! ( l − 2 s )! ( l − s )!
Problem:
Show that the series solution, for integer l, satisfies the recursion relation
( 2 l + 1 ) x φl (x) = ( l + 1 ) φl+1 (x) + l φl−1 (x) (393)
————————————————————————————————–
81
Example:
Solve the following eigenvalue equation using the Frobenius method

∂2φ

1 ∂φ m
2
+ − 1 + φ = 0 (394)
∂x x ∂x x
Solution:
The eigenvalue equation is Laguerre’s differential equation. The Frobenius

series represents an eigenfunction when the series truncates to yield a polyno-
mial, as certain boundary conditions have to be satisfied as x → ∞. In the
case that the boundary conditions are satisfied because the series truncates, the
parameter m is the eigenvalue. The point x = 0 is a regular point.
The Frobenius solution is found by assuming a solution in the form of an

expansion
X∞
φ(x) = xα Cn xn (395)
n=0
On substituting the assumed form of the solution into the differential equation,
one finds a polynomial equation which is equal to zero
X∞
0 = Cn n + α n + α − 1 xα+n−2
n=0
∞
X ∞
X
+ Cn n + α ( xα+n−2 − xα+n−1 ) + m Cn xα+n−1
n=0 n=0
(396)
On collecting like terms, the polynomial equation reduces to
∞
X ∞
X
0 = Cn ( n + α )2 xα+n−2 + ( m − n − α ) Cn xα+n−1 (397)
n=0 n=0
0
On writing n = n − 1 in the second term and noting that the sum starts at
n0 = 1, one finds
∞ ∞
X X 0
0 = Cn ( n + α )2 xα+n−2 + ( m − n0 − α + 1 ) Cn0 −1 xα+n −2
n=0 n0 =1
(398)
or
∞
X
0 = Cn ( n + α )2 + ( m − n − α + 1 ) Cn−1 xα+n−2
n=0
(399)
82
if one defines C−1 = 0. This polynomial is only zero if the coefficients of the
various powers are all equal to zero.
Examining the coefficient of the lowest power, xα−2 , and equating the coef-
ficient of the mononomial to zero, one finds the indicial equation
α 2 C0 = 0 (400)
which only involves the lowest-order expansion coefficient C0 . Since, by defini-

tion C0 is the coefficient of the lowest-order term in the expansion, it is non-zero,
otherwise all the coefficients are zero. Hence, we must have
α2 = 0 (401)
which only yields one solution for α.
On examining the higher-order coefficients of the polynomial, i.e. the coef-

ficient of the power in xn−2 , and equating the coefficient to zero, one has
0 = n2 Cn + ( m − n + 1 ) Cn−1 (402)
Hence, we have the recursion relation

(n − m − 1)
Cn = Cn−1 (403)
n2
This yields the higher order coefficients in terms of the lower order coefficients.
In fact, on iterating, one can obtain the formal relation
m!
Cn = ( − 1 ) n C0 (404)
( m − n )! (n!)2
for m > n. The Frobenius method only allows us to find this one solution,
since there is only one solution for α and C1 is uniquely determined by C0 .
For large n the recursion relation can be approximated by

1
Cn ∼ Cn−1 (405)
n
or
C0
Cn ∼ (406)
n!
Hence, for large x, the series can be approximated by an exponential as
∞
X xn
φ(x) ∼ C0
n=0
n!
∼ C0 exp[ x ] (407)
83
If the series truncates at the N -th order term, one must have CM = 0 for
all M > N . This can only be the case if CN +1 = 0 and CN 6= 0. In this
case, the recursion relation becomes
(N − m)
0 = CN +1 = CN (408)
( N + 1 )2
which requires N = m. In the case of integer m, the Frobenius series trun-

cates to an m-th order polynomial, and m is the eigenvalue. The polynomial
converges and therefore represents a good solution.
————————————————————————————————–
Example:
Solve the equation
∂ 2 φm (x) ∂φm (x)

− 2x + 2 m φm (x) = 0 (409)
∂x2 ∂x
for the Hermite polynomials φm (x) on the interval (−∞, +∞).
Solution:
The equation can be solved using the Frobenius method with the ansatz
∞
X
φm (x) = xα C n xn (410)
n=0
Substituting the ansatz into the differential equation yields

∞
X ∞
X
Cn ( n + α ) ( n + α − 1 ) xn+α−2 + 2 Cn ( m − n − α ) xn+α = 0
n=0 n=0
(411)
Setting n = n0 − 2 in the last term, gives
∞ ∞
X X 0
Cn ( n + α ) ( n + α − 1 ) x n+α−2
+2 Cn0 −2 ( m + 2 − n0 − α ) xn +α−2 = 0
n=0 n0 =2
(412)
We shall define C−1 = C−2 = 0, and then find that
∞
X
Cn ( n + α ) ( n + α − 1 ) +2 Cn−2 ( m + 2 − n − α ) xn+α−2 = 0
n=0
(413)
84
The indicial equation is found from examining the coefficient of lowest non-
zero mononomial xα−2 , where n = 0,
α ( α − 1 ) C0 = 0 (414)
hence, either α = 0 or α = 1.
For α = 0, one finds the recursion relation

n − 2 − m
Cn = 2 Cn−2 (415)
n(n − 1)
Thus, the solution is even in x. On writing n = 2 s, one obtains

2s − 2 − m
C2s = 2 C2s−2
2s(2s − 1)
( m − 2 s + 2 ) ( m − 2 s + 4 ) ... ( m − 2 ) m s
= ( − 1 )s 2 C0
(2s)!
(m)!!
= ( − 1 )s 2s C0
( m − 2 s )!! (2s)!
( m )!
= ( − 1 )s m 2 22s C0 (416)
( 2 − s)! (2s)!
It can be seen that the series truncates when

N − m
0 = CN +2 = 2 CN (417)
(N + 2)(N + 1)
that is, the series terminates if m = N in which case the series is a polynomial
of order N .
We note that if the series does not truncate the large n behavior of the
coefficients is governed by
2
Cn ∼ Cn−2 (418)
n
so one has
2s
C2s ∼ C0 (419)
s!
and the series would exponentiate at large x.
85
4.3 Linear Dependence
In the Frobenius method we look for a power series solution, by substituting
the power series into the equation. Let us examine only the first N terms in
the series solution. After some re-organization, one finds a polynomial which
is equated to zero. This polynomial equation is solved by insisting that if the
polynomial in x is equal to zero, then all the coefficients of the polynomial are
zero. This condition then leads to the recursion relation etc. If the series con-
verges in the limit N → ∞, one has a solution.
A crucial point of the procedure is the insistence that if the polynomial is

equal to zero then the only solution is that all the coefficients of the various
mononomials are zero. This is a statement that the various mononomials xn
are linearly independent.
The linear independence can be proved by examining a general N -th order

polynomial which is equal to zero
n+N
X
Cn xn = 0 (420)
n=0
The above equation is supposed to hold over a continuous range of x values.

Then as this holds for x = 0, one finds that the coefficient C0 must be zero.
On taking the differential of the polynomial one obtains
n=N
X
n Cn xn−1 = 0 (421)
n=1
which must also hold at x = 0. Hence, one obtains C1 = 0. By considering

all N higher order differentials, the analysis can extended to show that all the
coefficient Cn satisfy Cn = 0. Thus, we have shown that if the polynomial
is equal to zero, the coefficients of the all the various mononomials must equal
zero. Therefore, it is impossible to write
X−1
n=N
kN xN = kn xn (422)
n=0
for any set of kn which contains non-zero values. This is formalized by the
statement that a set of functions φn = xn are linearly independent if the only
solution of X
k n φn = 0 (423)
n
is that kn = 0 for all n. On the other hand, if one has a set of functions φα
such as the set containing all the N-th lowest-order mononomials, xn , and the
86
function φN +1 is expressible as φN +1 = a x2 + b x + c then there are
non-zero or non-trivial solutions of the equation
N
X +1
k α φα = 0 (424)
α=0
In particular, one has
k0 = c
k1 = b
k2 = a
kN +1 = 1 (425)
In this case, the set of functions φα are linearly dependent at least one of them
can be expressed as a linear combination of the others.
The concept of linear dependence and linear independence can be extended

to other sets of functions. A set of N functions φα are linearly independent if
the only solution of the equation
N
X
kα φα = 0 (426)
α=1
is that
kα = 0 ∀α (427)
Otherwise, if a non-trivial solution exists the set of functions are linearly de-
pendent.
A simple test of linear dependence is provided by the Wronskian. Consider

the generalization of the proof that the mono-nomials are linearly independent.
Namely consider
N
X
kα φα = 0 (428)
α=1
and the successive derivatives with respect to x starting with

N
X ∂φα
kα = 0 (429)
α=1
∂x
and
N
X ∂ 2 φα
kα = 0 (430)
α=1
∂x2
87
etc. The set of all these equations must be satisfied if the original equation
is satisfied for a continuous range of x, since the derivatives of the equation
can be found by subtracting the equation at infinitesimally different values of
x ie, x + dx and x, and then dividing by dx. The set of N coefficients kα are
completely determined from N independent equations, so we shall truncate the
set of equations with
N
X ∂ N −1 φα
kα = 0 (431)
α=1
∂xN −1
Since the functions φα and the derivatives are known, this is a set of N linear
equations for N unknowns and can be written as a matrix equation
φ1 φ2 ... ... φN
     
k1 0
 ∂φ1 ∂φ2
. . . . . . ∂φN   k2   0 
 ∂x ∂x ∂x     
 ... ... ... ... ...   .  =  .  (432)
     
 ... ... ... ... ...   .   . 
∂ N −1 φ1 ∂ N −1 φ2 N −1
∂x N −1 ∂x N −1 . . . . . . ∂ N −1
∂x
φN kN 0
This matrix equation has a solution given by

−1 
φ1 φ2 ... ... φN
   
k1 0
 k2   ∂φ1 ∂φ2
... ... ∂φN   0 
   ∂x ∂x ∂x   
 .  =  ... . . . ... ... ...   .  (433)
     
 .   ... ... ... ... ...   . 
kN ∂ N −1 φ1 ∂ N −1 φ2 ∂ N −1 φN 0
N −1
∂x N −1
∂x
. .. ... ∂xN −1
The matrix has an inverse if it’s determinant is non-zero. In this case, the only
solution consists of the trivial solution kn = 0 for all n. If the determinant is
non-zero, the set of functions φn are linearly independent.
If the determinant is zero, then non-zero solutions exist for kn for some n
and the set of functions φn are linearly dependent.
The determinant of the N by N matrix composed of the N functions and

their first N − 1 derivatives is known as the Wronskian, W

φ φ2 ... ... φN
∂φ11

∂φ2 ∂φN

∂x ∂x ... ... ∂x

W = . . . ... ... ... ...
(434)
... ... ... ... ...
N −1
∂ N −1 φ2 N −1

∂ φ1
... ... ∂ φN
∂xN −1 ∂xN −1 ∂xN −1
As examples one can find that the functions sin x and cos x are linearly
independent as
sin x cos x
cos x − sin x = − 1 (435)

88
which is non zero. Although, sin x is related to cos x via
p
sin x = ± ( 1 − cos2 x ) (436)
this relation is non-linear.
It is easy to show that the functions exp x, exp −x and cosh x are linearly
dependent as
exp x
exp −x cosh x
exp x
− exp −x sinh x = 0 (437)
exp x exp −x cosh x
which shows that it is possible to find a linear relationship between the functions
such as
exp x + exp −x = 2 cosh x (438)
The concept of linear dependence and linear independence is common to
the theory of vectors in d-dimensional spaces. A set of vectors φ̂n are linearly
independent if non-zero constants kn can not be found such that
n=d
X
kn φ̂n = 0 (439)
n=1
In such cases, the set of φ̂n can be used as a basis for the d-dimensional space,
→
−
in that an arbitrary vector Φ can be expanded as
n=d
→
− X
Φ = Cn φ̂n (440)
n=1
→
−
That is, the vector Φ and the set of basis vectors are form a linearly depen-
dent set. To see the relation we should identify kd+1 = 1 as the multiplier
→
−
of φ̂d+1 = Φ and kn = Cn as the multipliers for the rest of the φ̂n . Thus,
the kn or Cn can be considered as the components of the vector and the set of
linearly independent φ̂n as providing the basis vectors.
This is to be considered as analogous to he expansion of an arbitrary poly-

nomial Φ(x) of degree d in terms of the mononomials φn = xn so that
n=d
X
Φ(x) = C n xn
n=1
n=d
X
= C n φn (441)
n=1
where the set of linearly independent mononomials can be considered as forming

the basis functions φn .
89
4.3.1 Linearly Independent Solutions
For an N -th order linear differential equation one expects that there exist N
linearly independent solutions φn . The general solution is expected to be able
to be written as a linear superposition of the N linearly independent solutions
as
n=N
X
φ = C n φn (442)
n=1
where the N arbitrary coefficients Cn roughly correspond to the information

contained in the N constants of integration. If a set of appropriate boundary
conditions are applied, the coefficients Cn can be determined such that the so-
lution φ satisfies the boundary conditions.
Given an N -th order differential equation for φ

n=N
X ∂nφ
an (x) = 0 (443)
n=0
∂xn
then it is easy to show that N is the maximum number of linearly independent

solutions. This can be proved by contradiction. Assume that there are N + 1 or
more linearly independent solutions represented by the set of N + 1 functions
φn . Then form the Wronskian as the determinant of an N + 1 by N + 1 matrix
φ1 φ2 . . . . . . φN +1

. . . . . . ∂φ∂x
∂φ1 ∂φ2
N +1

∂x ∂x
W = . . . ... ... ... . . . (444)
... . . . . . . . . . . . .
N
∂ φ1 ∂ N φ2 . . . . . . ∂ φN +1

∂xN ∂xN ∂x
On substituting from the differential equation for the N -th order derivative into
the determinant, one finds the Wronskian is zero as the last row can be ex-
pressed as a linear combination of the (N − 1) higher rows. Hence, the set of
N + 1 solutions are linearly dependent contrary to our initial assumption. Thus,
at most, there are only N linearly independent solutions of the general N -th
order differential equation.
4.3.2 Abel’s Theorem

If there are two linearly independent solutions, φ1 and φ2 , of the second-order
∂2φ ∂φ
+ p(x) + q(x) φ = 0 (445)
∂x2 ∂x
90
then their Wronskian W is defined as a function of x

φ
1 φ2 ∂φ2 ∂φ1
W = ∂φ1 ∂φ2 = φ1 − φ2
∂x ∂x ∂x ∂x
Abel’s theorem states that the x-dependence of the Wronskian W (x) is deter-
mined from the differential equation through p(x). On taking the derivative of
the Wronskian, one finds

∂W ∂ ∂φ2 ∂ ∂φ1
= φ1 − φ2
∂x ∂x ∂x ∂x ∂x
2 2
∂ φ2 ∂ φ1
= φ1 − φ2 (446)
∂x2 ∂x2
On substituting the second-order derivatives found from the differential equation
∂2φ ∂φ
= − p(x) − q(x) φ (447)
∂x2 ∂x
into the expression for the derivative of the Wronskian, one obtains
∂W ∂φ2 ∂φ1
= − φ1 p(x) + φ2 p(x)
∂x ∂x ∂x
= − p(x) W (x) (448)
The above differential equation for W can be integrated to yield

Z x
W (x) = W (a) exp − dt p(t) (449)
a
Thus, if W (a) is non-zero because the solutions are linearly independent, then
as long as p(t) is not complex one finds that W (x) is always non-zero and the
solutions are always linearly independent. For linearly independent solutions,
the Wronskian is determined up to an arbitrary multiplicative constant since the
solutions φ are only determined up to a multiplicative constant. Furthermore,
if p(x) = 0 then the Wronskian is simply a constant.
4.3.3 Other Solutions

Given of N − 1 linearly independent solutions φn and the Wronskian W of an
N -th order differential, one can reduce the order of the equation to an N − 1-th
ordinary differential equation. This is most useful for the case where N = 2.
In this case given a solution φ1 and a Wronskian W , one can find a second
solution φ2 . Starting with W (x)
∂φ2 ∂φ1
W = φ1 − φ2
∂x ∂x
91
∂φ2
2 φ2 ∂φ1
∂x
= ( φ1 (x) ) − 2
φ1 φ ∂x
1
∂ φ2
= ( φ1 (x) )2 (450)
∂x φ1
Thus,

∂ φ2 W (x)
= (451)
∂x φ1 φ1 (x)2
and so one obtains the second solution of the second-order differential equation
through one integration
Z x
W (t)
φ2 (x) = φ1 (x) dt (452)
φ1 (t)2
Since the Wronskian is only known up to an arbitrary multiplicative constant,

the second solution is also determined up to a multiplicative constant.
A simple example is given by the solution of
∂2φ
+ φ = 0 (453)
∂x2
which if knows one solution φ1 = sin x, then one can find a second solution.
Since p(x) = 0 the Wronskian is just a constant W , hence
Z x
dt W
φ2 (x) = sin x
sin2 t

= W sin x − cot x
= − W cos x (454)
Thus, cos x is a second linearly independent solution of the equation.
92
5 Stürm Liouville Theory
Consider the Stürm-Liouville eigenvalue equation for the unknown function φ
in the form
∂2φ ∂φ
a2 (x) 2
+ a1 (x) + a0 (x) φ = λ u(x) φ (455)
∂x ∂x
where λ is an unknown number, and all the functions are defined on an interval
between x = a and x = b. The solutions are expected to satisfy boundary
conditions. The functions a0 (x) and u(x) are non zero in the interval (a, b). For
convenience we shall consider the case where u(x) > 0. The case of negative
u(x) can be treated by changing the definition of λ to accommodate the sign
change. The number λ is regarded as unknown and is called the eigenvalue. The
possible values of the eigenvalue are found from the condition that a solution
φ(x) exists which satisfies the boundary conditions. When the solutions φλ (x)
are found to exist the eigenvalues have particular values. The set of eigenvalues
may take on either discrete or continuous values. The function φλ (x) is called
an eigenfunction, and each eigenfunction corresponds to a particular eigenvalue
λ. Sometimes a particular eigenvalue λ may correspond to two or more different
eigenfunctions. In this case, the eigenvalue is said to be degenerate. Any lin-
ear combination of the degenerate eigenfunctions is also an eigenfunction with
the same eigenvalue. The number of linearly independent eigenfunctions corre-
sponding to this eigenvalue is the degeneracy of the eigenvalue.
It is usual to re-write the Stürm-Liouville equation in the form

∂φ ∂φ
p(x) + q(x) φ = λ w(x) φ (456)
∂x ∂x
which is equivalent to
∂2φ ∂p(x) ∂φ
p(x) + + q(x) φ = λ w(x) φ (457)
∂x2 ∂x ∂x
Hence, we can identify
∂p
a1 (x)
= ∂x
a2 (x) p(x)
a0 (x) q(x)
=
a1 (x) p(x)
u(x) w(x)
= (458)
a2 (x) p(x)
Thus, we can find p(x) by integration between two limits of integration in the
interval x Z x
a1 (t)
ln p(t) = dt (459)
a a a2 (t)
93
or Z x
a1 (t)
p(x) = p(a) exp dt (460)
a a2 (t)
It is because of this that we required that a2 (t) be non-vanishing in (a, b).
A well-known example of a Stürm-Liouville eigenvalue equation is provided

by
∂2φ
= λφ (461)
∂x2
on the interval (0, L) where the boundary conditions are that the functions φ
must vanish at the boundaries
φ(0) = φ(L) = 0 (462)
In this case, we see that p(x) = w(x) = 1 and q(x) = 0. For each arbitrary
real value of λ one can find two solutions of the differential equation which do
not satisfy the two boundary conditions
√

f±λ (x) = exp ± λ x (463)
The functions f±λ (x) can be combined to yield functions φλ (x) that satisfies
the boundary condition at x = 0 as
√ √

φλ (x) = exp + λ x − exp − λ x (464)
so φ(0) = 0. The second boundary condition is not satisfied for arbitrary

values of λ as √
φλ (L) = 2 sinh λ L (465)
√
which is non-zero for real values of λ or positive λ. However, for negative
values of λ one finds that this equation can have solutions
p
φλ (L) = 2 i sin |λ| L (466)
for values of λ such that p

|λ| L = n π (467)
for integer n. The eigenfunctions can be written as
nπx
φn (x) = sin (468)
L
and the eigenvalues form a discrete set of negative numbers
2
nπ
λn = − (469)
L
94
Thus, the values of the eigenvalues λn are determined by the condition that
φn (x) exists and satisfy the boundary conditions.
Problem:
Find the general solutions of the above equation, using the Frobenius method,
without imposing boundary conditions.
If boundary conditions are applied, can the series be used directly to find
the possible set of eigenvalues?
————————————————————————————————–
Example:
An example is given by the Stürm-Liouville eigenvalue equation
∂2φ
− x2 φ = λ φ (470)
∂x2
subject to the boundary conditions that φ(x) vanishes in the limits x → ± ∞.
Solution:
Here p(x) = w(x) = 1 and q(x) = − x2 . The eigenvalue equation can be
solved near the origin by the Frobenius method. The indicial equation is simply
α(α − 1) = 0 (471)
and thus the solutions have the form of either

X
φ0 = Cn xn (472)
n=0
or
X
φ1 = x C n xn (473)
n=0
Furthermore, from the recursion relation one finds that the odd coefficients
vanish and that the even coefficients must satisfy
n ( n − 1 ) Cn − Cn−4 = λ Cn−2 (474)
Thus, the solutions are either even φ0 (−x) = φ0 (x) or odd φ1 (−x) = − φ1 (x)
in x. The Frobenius method can not be expected to converge in the limits
x → ± ∞. Therefore, we need to examine the asymptotic large x behavior.
95
In this case, one can neglect the eigenvalue compared to the x2 term. The
approximate equation is given by
∂2φ
− x2 φ ∼ 0 (475)
∂x2
This equation can be approximately factorized as

∂ ∂
− x + x φ ∼ 0 (476)
∂x ∂x
Then, we can expect that

∂φ
+ xφ ∼ 0 (477)
∂x
so that the asymptotic large x variation is given by
x2

φ(x) ∼ exp − (478)
2
which satisfies the boundary conditions.
Thus, we may look for solutions of the forms of power series in increasing
powers of x2 times a decreasing exponential function. That is we either have
the even function
x2
X
φ0 (x) = P2n x2n exp − (479)
n=0
2
or the odd function

x2
X
φ1 (x) = x P2n x2n exp − (480)
n=0
2
Using this form we can determine the power series with coefficients Pn by using
the Frobenius method.

x2
To simplify the solution we shall substitute φ(x) = p(x) exp − 2 into
the equation and after cancelling out the common exponential factor one finds
∂ 2 p(x) ∂p(x)
2
− 2x − p(x) = λ p(x) (481)
∂x ∂x
Then the indicial equation for p(x) is
α(α − 1) = 0 (482)
96
The recursion relation becomes
( 2 n + 2 + α ) ( 2 n + 1 + α ) P2n+2 − ( 2 n + 2 α + λ + 1 ) P2n = 0 (483)
For large n, such n λ the coefficients are related by

P2n
P2n+2 ∼ (484)
2n
so
P0
P2n ∼ (485)
2n(n)!
In this case the series can be approximately summed to yield
x2

p(x) ∼ p0 exp + (486)
2
which diverges exponentially. Thus, we note that the recursion relation for
Pn must truncate at some finite value of N in order that the solution has the
correct exponentially decaying behavior for asymptotically large values of x.
However, the series only truncates for particular values of λ. The value of λ
can be determined by examining the recursion relation of the largest term in
the polynomial. Physically this is because the term proportional to xN , and the
exponential factor, dominates the function φ(x) at large x. Hence, we find that
if P2N +2 = 0 then
λN = − ( 2 N + 2 α + 1 ) (487)
Therefore, the value of λ is determined by the highest power in the finite poly-
nomial.
————————————————————————————————–
In the above two examples, the set of eigenfunctions share the common
property that
Z b
dx φn (x) w(x) φm (x) = 0 (488)
a
where λn 6= λm . This is a general property of a set of eigenfunctions of the
Stürm-Liouville equation. This theorem implies that the successive eigenfunc-
tions change sign an increasing number of times. That is the eigenfunctions can
be classified by their number of nodes.
Problem:
Show that the set eigenfunctions

r
2 nπx
φm (x) = sin (489)
L L
97
obey the equation
Z L
dx φm (x) φn (x) = 0 (490)
0
if λm 6= λn .
5.1 Degenerate Eigenfunctions

If more than one eigenfunction corresponds to the same eigenvalue, the eigen-
value is said to be degenerate. The number of linearly independent eigenfunc-
tions are the degeneracy of the eigenvalue. The set of eigenfunctions are said to
be degenerate.
————————————————————————————————–
Example:
An example of degenerate eigenfunctions is given by the solutions of

∂φ
= − k2 φ (491)
∂x
without boundary conditions.
Solution:
The eigenvalue − k 2 is a real negative number and is twofold degenerate, as

it corresponds to the two linearly independent eigenfunctions
φ1 (x) = sin k x
φ2 (x) = cos k x (492)
Since any linear combinations of these are also eigenfunctions, one finds that
the complex functions

φ01 (x) = exp + i k x

φ02 (x) = exp − i k x (493)
are also eigenfunctions corresponding to the same eigenvalue.
————————————————————————————————–
98
5.2 The Inner Product
The inner product of two functions Ψ(x) and Φ(x) is defined as the weighted
integral
Z b
dx Ψ∗ (x) w(x) Φ(x) (494)
a
Two functions are said to be orthogonal if their inner product is zero. The inner
product of a function with itself is the normalization of the function. If w(x) is
a real, non-zero, function the normalization is a real number. It is customary to
demand that all functions Ψ(x) are normalized to unity. Thus, we insist that
Z
dx Ψ∗ (x) w(x) Ψ(x) = 1 (495)
A function can always be normalized as, according to the principle of linear

superposition it can always be multiplied by a constant complex number C.
This appears as a factor | C |2 in the inner product. The magnitude of the real
number | C |2 is chosen such that the functions are normalized.
This inner product is analogous to the scalar product of two vectors

→
− → −
A . B (496)
which is usually evaluated in terms of its components. That is on expanding

the vector in terms of the basis vectors ên via
d
→
− X
A = An ên (497)
n=1
and noting that the basis vectors form an orthogonal set so that
ên . êm = δn,m (498)
one has
d
→
− → − X
A . B = An B n (499)
n=1
The inner product of two functions can be evaluated in the same way. First
the functions are expanded in terms of the basis functions φn (x) with compo-
nents An and Bn
Xd
Φ(x) = An φn (x) (500)
n=1
and
d
X
Ψ(x) = Bn φn (x) (501)
n=1
99
Then, if the basis functions form an orthonormal set so that
Z b
dx φ∗n (x) w(x) φm (x) = δn,m (502)
a
then the inner product is evaluated as

Z b X
dx Ψ∗ (x) w(x) Φ(x) = Bn∗ δn,m Am
a n,m
X
= Bn∗ An (503)
n
which is similar to the scalar product of two vectors.
5.3 Orthogonality of Eigenfunctions

The orthogonality property of eigenfunctions of a Stürm-Liouville equation can
be derived by examining the properties of two solutions, φ1 and φ2 corresponding
to the respective eigenvalues λ1 and λ2 . Then

∂ ∂φ1
p(x) + q(x) φ1 = λ1 w(x)φ1 (504)
∂x ∂x
and
∂ ∂φ2
p(x) + q(x) φ2 = λ2 w(x)φ2 (505)
∂x ∂x
The complex conjugate of the second equation is just
∂φ∗2

∂
p(x) + q(x) φ∗2 = λ∗2 w(x)φ∗2 (506)
∂x ∂x
as p(x), q(x) and w(x) are real. Pre-multiplying the above equation by φ1 (x)
and pre-multiplying the first equation by the complex conjugate φ∗2 (x) and sub-
tracting one obtains
∂φ∗2 (x)

∂ ∂ ∂φ1 (x)
φ1 (x) p(x) − φ∗2 (x) p(x) = ( λ∗2 − λ1 ) φ1 (x) w(x) φ∗2 (x)
∂x ∂x ∂x ∂x
(507)
In this the terms proportional to q(x) have cancelled identically. Taking the
above equation and integrating over the interval between a and b, one has
Z b
∂φ∗2 (x)

∂ ∗ ∂ ∂φ1 (x)
dx φ1 (x) p(x) − φ2 (x) p(x)
a ∂x ∂x ∂x ∂x
Z b
= ( λ∗2 − λ1 ) dx φ∗2 (x) w(x) φ1 (x)
a
(508)
100
where the term on the right-hand-side is recognized as the inner product. The
term on the left-hand-side is evaluated by splitting into two parts and integrating
each term by parts. The terms of the form
∂φ∗2 ∂φ1
p(x) (509)
∂x ∂x
cancel identically, leaving only the boundary terms
b b
∂φ∗2 (x) ∗ ∂φ1 (x)
φ1 (x) p(x) − φ2 (x) p(x)
∂x a ∂x a
Z b
= ( λ∗2 − λ1 ) dx φ∗2 (x) w(x) φ1 (x) (510)
a
However, on either using boundary conditions such that the functions vanish on
the boundaries
φ(a) = φ(b) = 0 (511)
or boundary conditions where the derivative vanishes at the boundaries

∂φ ∂φ
= = 0 (512)
∂x a ∂x b
then one finds that
Z b
( λ∗2 − λ1 ) dx φ∗2 (x) w(x) φ1 (x) = 0 (513)
a
This is the central result. It proves the theorem. First if φ2 = φ1 then the
normalization integral is finite and non-zero, so λ∗1 = λ1 . Thus, the eigenvalues
of the Stürm-Liouville equation are real. Using this we have
Z b
( λ2 − λ1 ) dx φ∗2 (x) w(x) φ1 (x) = 0 (514)
a
Thus, if λ1 6= λ2 one must have

Z b
dx φ∗2 (x) w(x) φ1 (x) = 0 (515)
a
Therefore, the eigenfunctions belonging to different eigenvalues are orthogonal.

As we shall see later, if an eigenvalue is degenerate one can use the proper-
ties of the degeneracy to construct a set of mutually orthogonal eigenfunctions
corresponding to the same eigenvalue. The maximum number of the mutually
orthogonal functions is equal to the degeneracy. This means that the eigenfunc-
tions of the Stürm-Liouville equation can be used to create a very convenient
set of basis functions, in function space.
101
5.4 Orthogonality and Linear Independence
Given a set of mutually orthogonal functions φn (x), one can easily show that
they are linearly independent. For if one has
X
kn φn (x) = 0 (516)
n
then one can take the inner product with any one of them, say φm (x), to find
X Z b
kn dx φ∗m (x) w(x) φn (x) = 0
n a
X
kn δn,m = 0
n
km = 0 (517)
Thus, the only solution of the equation is that all the km are identically zero.
Hence, any set of mutually orthogonal functions are linearly independent.
5.5 Gram-Schmidt Orthogonalization

If we have a set of eigenfunctions of a Stürm-Liouville equation, some of the
eigenvalues might be degenerate. The eigenfunctions corresponding to the
same eigenvalue generally might not be orthogonal. The Gram-Schmidt or-
thogonalization process can be used to construct a set of mutually orthogonal
eigenfunctions. Consider the set of normalized eigenfunctions φ1 (x), φ2 (x),
φ3 (x),....corresponding to an eigenvalue λ. The method produces a set of or-
thogonal and normalized eigenfunctions ψ1 (x), ψ2 (x), ψ3 (x),.... The first eigen-
function is chosen to be
ψ1 (x) = φ1 (x) (518)
The second eigenfunction is to be constructed from φ2 (x). However, since the
inner product with φ1 (x) is non-zero, there is a component of φ2 (x) which is
parallel to φ1 (x). Thus, we can subtract the component of φ2 (x) which is parallel
to φ1 (x) and then normalize the function by multiplying by a constant C2
Z b
∗
ψ2 (x) = C2 φ2 (x) − ψ1 (x) ( dt ψ1 (t) w(t) φ2 (t) ) (519)
a
This is an eigenfunction corresponding to the eigenvalue λ as the eigenvalue

equation is linear. The constant C2 or rather | C2 |2 in ψ2 (x) is determined
from the normalization condition
Z b
dx ψ2∗ (x) w(x) ψ2 (x) = 1 (520)
a
102
It can be seen that ψ2 (x) is orthogonal to ψ1 (x) by direct substitution in
Z b
dx ψ1∗ (x) w(x) ψ2 (x) = 0 (521)
a
and using the fact that ψ1 (x) is normalized to unity

Z b
dx ψ1∗ (x) w(x) ψ1 (x) = 1 (522)
a
The next eigenfunction ψ3 (x) is constructed from φ3 (x) by orthogonalizing it

to ψ2 (x) and ψ1 (x). That is, we write
Z b Z b
ψ3 (x) = C3 φ3 (x) − ψ2 (x) dt ψ2∗ (t) w(t) φ3 (t) − ψ1 (x) dt ψ1∗ (t) w(t) φ3 (t)
a a
(523)
The eigenfunction ψ3 (x) has no components parallel to ψ1 (x) or to ψ2 (x), since
ψ2 (x) and ψ1 (x) have been normalized previously. The constant C3 is then
determined from the normalization condition
Z b
dx ψ3∗ (x) w(x) ψ3 (x) = 1 (524)
a
This procedure is iterated, by orthogonalizing φn (x) to all the previous or-

thonormal functions ψm (x) for n > m
m=n−1
X Z b
∗
ψn (x) = Cn φn (x) − ψm (x) dt ψm (t) w(t) φn (t) (525)
m=1 a
and then determining the normalization | Cn |2 from the condition

Z b
dx ψn∗ (x) w(x) ψn (x) = 1 (526)
a
If the set of eigenfunctions is composed of M functions corresponding to the

same eigenvalue this procedure only has to be iterated M times.
The method does not have to be applied to a set of linearly independent

eigenfunctions. If it is applied to a set of M linearly dependent eigenfunctions,
where M > N and the degeneracy of the eigenvalue is N , then the Gram-
Schmidt orthogonalization procedure will lead to a maximum of N orthonormal
functions. The way this happens is that if the n-th initial eigenfunction φn (x)
is linearly dependent on the previous set of initial eigenfunctions, the orthogo-
nalization procedure will lead to ψn (x) = 0.
103
Therefore, it is always possible to construct a set of orthonormal basis func-
tions from the set of Sturm-Liouville eigenfunctions. We shall always assume
that the set of eigenfunctions of a Stürm-Liouville equation have been chosen
as an orthonormal set.
As an example of Gram-Schmitt orthogonalization, consider the set of poly-

nomials, φn (x) 1, x, x2 , x3 , .... defined on the real interval between (−1, 1) and
with weight factor one. Since the interval and the weighting factor is evenly dis-
tributed around x = 0, we shall see that the orthogonal polynomials ψn (x) are
either even or odd, because the even polynomials are automatically orthogonal
with the odd polynomials.
The method starts with normalizing φ0 (x) to yield ψ0 (x) as

ψ0 (x) = C0 φ0 (x) = C0 1 (527)
The normalization constant C0 is determined from
Z 1
1 = dx | ψ0 (x) |2
−1
Z 1
= | C 0 |2 dx | φ0 (x) |2
−1
Z 1
= | C 0 |2 dx
−1
= | C 0 |2 2 (528)
or | C0 | = √1 . Thus, since all our quantities are real we might as well choose
2
all the phases to be real and have
1
ψ0 (x) = √ (529)
2
The lowest-order normalized polynomial ψ0 (x) is an even function of x.
Proceeding to construct the next orthogonal polynomial from φ1 (x) = x,

one has
Z 1
ψ1 = C1 φ1 (x) − ψ0 (x) dt φ1 (t) ψ0 (t)
−1
Z 1
1
= C1 x − dt t
2 −1
= C1 x (530)
since x is odd and 1 is an even function of x. The normalization is found to be
r
3
C1 = (531)
2
104
q
3
Hence, the orthogonal polynomial ψ1 (x) = 2 x and is an odd function of x.
The next polynomial is constructed from φ2 (x) = x2 this is an even function

and is automatically orthogonal to all odd functions of x. Thus, we only need
to orthogonalize it against ψ0 (x)
Z 1
2 1 2 1
ψ2 = C2 x − √ dt t √
2 −1 2

1
= C2 x2 − (532)
3
and C2 is found as r
45
C2 = (533)
8
The orthogonalization of the next polynomial is non-trivial. The process
starts with φ3 (x) = x3 , so one finds
Z 1
3 3 4
ψ3 (x) = C3 x − x dt t
2 −1

3
= C3 x3 − x (534)
5
etc.
This set of polynomials, apart from multiplicative constants, are the same
as the set of Legendre polynomials Pn (x)
P0 (x) = 1
P1 (x)= x
3 x2 − 1
P2 (x) =
2
5 x3 − 3 x
P3 (x) =
2
(2n)! xn + . . .
Pn (x) = (535)
2n (n!)2
The Legendre polynomials are normalized differently. The Legendre polynomi-

als are normalized by insisting that Pn (0) = 1.
Problem:
Construct the first four orthogonal polynomials on the interval (−∞, +∞)
with weight factor w(x) = exp[ − x2 ]. Show, by direct substitution in the
105
equation, that these are the same polynomials p(x) that occur in the eigenfunc-
2
tions φ(x) = p(x) exp[ − x2 ] of
∂2φ
− x2 φ = λ φ (536)
∂x2
subject to the boundary conditions that the functions φ(x) vanishes in the limits
x → ± ∞.
Problem:
Use the Gram-Schmidt procedure to obtain an orthonormal set of functions

ψn (x) from the degenerate eigenfunctions on the interval (0, L)

nπx
φ0 (x) = exp i
L
nπx
φ1 (x) = cos
L
nπx
φ2 (x) = sin (537)
L
Are the initial set φn (x) and the final set ψn (x) linearly independent?
5.6 Completeness of Eigenfunctions

A set of linearly independent basis vectors in a vector space is defined to be
complete if any vector in the space can be expressed as a linear combination of
the basis vectors. The set of linearly independent basis vectors are incomplete if
they are not complete. In the case of an incomplete set of basis vectors, then a
linear combination of the basis vectors can only describe a subset of the vectors
residing in the vector space.
In general the number of linearly independent basis vectors that form a com-
plete set is equal to the dimension of the vector space. Also, if the basis is not
complete, then the set of vectors that can be expressed as a linear combination
of the incomplete basis set forms a vector space which has dimensions equal to
the number of linearly independent basis vectors in the incomplete set.
For example, in two dimensions, any vector can be expanded in terms of a

basis composed of two non-collinear unit vectors, ê1 and ê2 . The non-collinearity
condition is an expression of the linear independence of the basis vectors. Thus,
any vector in the two dimensional plane can be written as
→
−
C = C1 ê1 + C2 ê2 (538)
106
For the two dimensional space, the set of the two basis vectors span the en-
tire vector space and the basis is said to be complete. However, in a three-
dimensional vector space, the above combination only describes a two dimen-
sional plane in the three-dimensional volume. The two-dimensional basis vectors
do not span the three-dimensional vector space and the set of two basis vectors
is therefore said to be incomplete.
To be able to express any vector in the three-dimensional volume, it is nec-

essary to expand our basis by adding one non-coplanar unit vector. In this case,
one can express a general vector as
3
→
− X
A = Ai êi (539)
i=1
In this case of a three dimensional vector space, the set of the three basis vectors
is complete.
One can generalize the above definitions to a linearly independent set of

basis functions. Thus, for example, given a set of mononomials φm = xm
which serve as basis functions one can describe polynomials as residing in the
space of polynomials. Therefore, with a complete set of basis functions 1, x and
x2 , one can describe any polynomial in a three-dimensional space as a linear
combination of the three linearly independent basis mononomials
p(x) = p0 + p1 x + p2 x2 (540)
where the expansion coefficients (p0 , p1 , p2 ) are the ”components” or ”coordi-

nates” of the polynomial.
Likewise, if we use the set of functions given by the eigenfunctions φm (x)

of a Stürm-Liouville equation one can, if needed through the Gram-Schmidt
process, construct an orthogonal set of basis functions. We shall always assume
that the basis functions are orthonormal. Then a well behaved general function
can be expressed as a linear combination of the basis functions
X
Ψ(x) = Cn φn (x) (541)
n
By a general well behaved function, we require that our function satisfy the
same boundary conditions as the basis functions. Also, the inner product of
Ψ(x) with itself must exist. These functions lie in the space C2 , i.e., the space
of square integrable complex functions.
It is easy to show that a function Ψ(x) in this space has a unique expansion
in terms of the basis. That is the expansion coefficients are uniquely determined
107
since by use of the inner product of Ψ(x) with any one of the basis functions,
say φm (x), one can uniquely determine the expansion coefficient Cm .
Z b X Z b
∗
dx φm (x) w(x) Ψ(x) = Cn dx φ∗m (x) w(x) φn (x)
a n a
X
= Cn δn,m
n
= Cm (542)
That is, the expansion coefficient Cm is uniquely determined by the inner prod-
uct of Ψ(x) and φm (x).
For an arbitrary well-behaved function Ψ(x) one can write

X
Ψ(x) = Cn φn (x)
n
X Z b
= dt φ∗n (t) w(t) Ψ(t) φn (x)
n a
Z b X
= dt Ψ(t) φ∗n (t) w(t) φn (x) (543)
a n
Thus, the delta function δ(x − t) is identified as the sum

X
δ(x − t) = φ∗n (t) w(t) φn (x) (544)
n
Alternatively, if the delta function δ(x − x0 ) is expanded as

X
δ(x − x0 ) = Cn φn (x) (545)
n
then we have Z b
dx φ∗m (x) w(x) δ(x − x0 ) = Cm (546)
a
which is simply evaluated as
Cm = φ∗m (x0 ) w(x0 ) (547)
Therefore, the Dirac delta function has an expansion of the form
X
δ(x − x0 ) = φ∗n (x0 ) w(x0 ) φn (x) (548)
n
Since the delta function is symmetrical in x and x0 one can also write
X
δ(x − x0 ) = φ∗n (x0 ) w(x) φn (x) (549)
n
108
because the left-hand-side is zero unless x = x0 .
An example of the completeness relation is given by the doubly-degenerate

eigenfunctions
1
φk (x) = √ exp + i k x (550)
2π
of the eigenvalue equation
∂2φ
= − k2 φ (551)
∂x2
Degenerate eigenfunctions corresponding to positive and negative values of k.
The completeness relation is given by
Z +∞
0 1 0
δ(x − x ) = dk exp − i k x exp + i k x (552)
2 π −∞
The integral can be evaluated by breaking it into two segments
Z +∞
0 1 0
δ(x − x ) = dk exp − i k x exp + i k x
2π 0
Z +∞
1
+ dk exp + i k x0 exp − i k x
2π 0
(553)
and then a convergence factor exp[ − k ] is added and the limit → 0 is

taken. Thus,
Z +∞
0 1 0
δ(x − x ) = dk exp + i k ( x − x ) − k
2π 0
Z +∞
1
+ dk exp − i k ( x − x0 ) − k
2π 0
1 1 1 1
= − 0
+
2π i(x − x ) − 2 π i ( x − x0 ) +
1
= (554)
π ( x − x0 )2 + 2
This agrees with the definition of the delta function as a limit of a series functions
1
δ(x − x0 ) = lim 0
(555)
→ 0 π ( x − x )2 + 2
The theory of the Fourier integral transform consists of the expansion of an
arbitrary function f (x)
Z +∞
˜ 1
f (x) = dk f (k) √ exp + i k x (556)
−∞ 2π
109
and the inverse transform
Z +∞
˜ 1
f (k) = dx √ exp − i k x f (x) (557)
−∞ 2π
Completeness of the eigenfunctions of a Stürm-Liouville equation has to be
proved on a case by case basis. The proof usually consists of showing that the
expansion in a series of eigenfunctions
N
X
Cn φn (x) (558)
n=0
converges to f (x), when N → ∞ at each point x. However it is possible, by

relaxing the condition for convergence at every point x, to extend the space of
functions f (x) to functions which have a finite number of discontinuities and do
not satisfy the boundary conditions. In this case, one requires convergence in
the mean. This just requires that
Z b XN 2

lim dx f (x) −
Cn φn (x) w(x) → 0 (559)
N → ∞ a n
Since, for finite N one has

Z b N
X 2

dx f (x) − Cn φn (x) w(x) ≥ 0

a n
Z b X Z b
= dx | f (x) |2 w(x) − Cn∗ φ∗n (x) w(x) f (x)
a n a
X Z b X
− Cn f ∗ (x) w(x) φn (x) + | Cn |2 ≥ 0
n a n
Z b X
= dx | f (x) |2 w(x) − | Cn |2 ≥ 0 (560)
a n
one finds that convergence requires that the equality sign in Bessel’s inequality
holds as N → ∞.
Another useful inequality is the Schwartz inequality

Z b Z b Z b 2
2 2
dt f ∗ (t) w(t) φn (t)

dx | f (x) | w(x) dy | φn (y) | w(y) ≥

a a a
(561)
which is evaluated as
Z b
dx | f (x) |2 w(x) ≥ | Cn |2 (562)
a
110
which is a statement that the squared length of a vector must be greater than
the square of any one component.
We shall now examine a few physically important examples of Stürm-Liouville

equations.
111
6 Fourier Transforms
The plane waves are eigenfunctions of the Stürm-Lioville equation
∂ 2 φk
= − k2 φ (563)
∂x2
on the interval (−∞, ∞) has solutions

1
φk (x) = √ exp ikx (564)
2π
The eigenfunctions satisfy the orthogonality relations
Z ∞ Z ∞
dx
dx φ∗k0 (x) φk (x) = exp i ( k − k 0 ) x
−∞ −∞ 2 π
= δ( k − k 0 ) (565)
These eigenfunctions form a complete set, so an arbitrary function f (x) can be

expanded as a linear superposition of the type
Z ∞
1
f (x) = dk f˜(k) √ exp + i k x (566)
−∞ 2π
This expansion coincides with the inverse Fourier Transform. The expansion
coefficient f˜(k) is given by the Fourier Transform
Z ∞
˜ 1
f (k) = dx f (x) √ exp − i k x (567)
−∞ 2π
————————————————————————————————–
Example:
The response of a system to a delta function pulse at time t = 0 can be

represented by a response function χ(t). For a particular system the response
is given by
χ(t) = χ0 exp i ω0 t exp − Γt Θ(t) (568)
where Θ(t) is the Heaviside step function which expresses causality
Θ(t) = 1 for t > 0

Θ(t) = 0 for t < 0 (569)
112
Find the frequency-dependent response function given by the Fourier Transform
χ̃(ω). The divergences only occur in the lower-half complex frequency plane.
The frequency-dependent response is obtained from

Z ∞
1
χ̃(ω) = √ dt χ(t) exp + i ω t
2 π −∞
Z ∞
1
= √ dt χ0 exp − i ω0 t exp − Γ t exp + i ω t
2π 0
i χ0
= √ (570)
2 π ω − ω0 + i Γ
The frequency dependent response is finite for real frequencies, but diverges at
ω = ω0 − i Γ in the lower-half complex plane.
————————————————————————————————–
6.1 Fourier Transform of Derivatives

The Fourier transform of a derivative of a function can be simply expressed in
terms of the Fourier transform of the function. Let f (x) be a function which
has the Fourier transform f˜(k), where
Z ∞
1
f˜(k) = dx f (x) √ exp − i k x (571)
−∞ 2π
Then the Fourier transform of the derivative is defined by
Z ∞
∂f (x) 1
dx √ exp − i k x (572)
−∞ ∂x 2π
Integrating by parts, one obtains
∞ Z ∞
1 1
= f (x) √ exp − i k x −ik dx f (x) √ exp −ikx

2π −∞ 2π
−∞
(573)
which if f (x) vanishes as x → ± ∞ yields the Fourier transform of the derivative
as
Z ∞
1
= −ik dx f (x) √ exp − i k x
−∞ 2π
= i k f˜(k) (574)
113
The Fourier transform of the n-th order derivative can be obtained by re-
peated differentiation by parts and is given by
Z ∞ n
∂ f (x) 1
dx √ exp − i k x
−∞ ∂xn 2π
n ˜
= ( i k ) f (k) (575)
————————————————————————————————–
Example:
The wave equation can be solved by Fourier transformation. Consider the

wave equation
∂2φ 1 ∂2φ
= (576)
∂x2 c2 ∂t2
subject to an initial conditions
φ(x, 0) = f (x) (577)
and
∂φ(x, t)
= g(x) (578)
∂t
t=0
Fourier transforming the equation with respect to x, one has
Z +∞ 2 Z +∞ 2
1 ∂ φ 1 1 ∂ φ
dx √ exp − i k x 2
= 2
dx √ exp − i k x
−∞ 2π ∂x c −∞ 2π ∂t2
1 ∂ 2 φ̃(k, t)
− k 2 φ̃(k, t) = (579)
c2 ∂t2
At t = 0 the Fourier transform of the initial data are given by

Z +∞
1
φ̃(k, 0) = dx √ exp − i k x φ(x, 0)
−∞ 2π
Z +∞
1
= dx √ exp − i k x f (x)
−∞ 2π
= f˜(k) (580)
and
Z +∞
∂ φ̃(k, t) 1 ∂φ(x, t)
= dx √ exp − ikx
∂t
t=0 −∞ 2π ∂t
t=0
114
Z +∞
1
= dx √ exp − ikx g(x)
−∞ 2π
= g̃(k) (581)
Hence, we have to solve the second-order differential equation
1 ∂ 2 φ̃(k, t)
− k 2 φ̃(k, t) = (582)
c2 ∂t2
subject to the two initial conditions given by
φ̃(k, 0) = f˜(k) (583)
and
∂ φ̃(k, t)
= g̃(k) (584)
∂t
t=0
The second-order ordinary differential equation has the general solution

φ̃(k, t) = A exp i c k t + B exp − i c k t (585)
where A and B are arbitrary constants. The initial conditions determine A and
B through
f˜(k) = A + B (586)
and
g̃(k) = i c k ( A − B ) (587)
Hence, we have determined the constants as

1 ˜ g̃(k)
A = f (k) − i (588)
2 ck
and
1 g̃(k)
B = f˜(k) + i (589)
2 ck
Thus, the Fourier Transform of the solution is given by
" #
1 ˜ g̃(k) ˜ g̃(k)
φ̃(k, t) = f (k) − i exp + i c k t + B f (k) + i exp − i c k t
2 ck ck
(590)
and then the solution is given by the inverse Fourier Transform
Z +∞
1
φ(x, t) = dk φ̃(k, t) √ exp + i k x (591)
−∞ 2π
115
or
Z +∞
1 g̃(k)
1
φ(x, t) = dk f˜(k) − i √
exp i k ( x + c t )
2 −∞ ck 2π
Z+∞
1 g̃(k) 1
+ dk f˜(k) + i √ exp i k ( x − c t )
2 −∞ ck 2π
(592)
The integrals can be evaluated from the definition of the inverse Fourier Trans-
form yielding
Z x+ct
1 1
φ(x, t) = f( x + c t ) + dz g(z)
2 c a
Z x−ct
1 1
+ f( x − c t ) − dz g(z) (593)
2 c a
where the arbitrary constant of integration cancels. In fact, using this cancella-
tion, one can write the solution as

1
φ(x, t) = f( x + c t ) + f( x − c t )
2
Z x+ct
1
+ dz g(z) (594)
2 c x−ct
This is D’Alembert’s solution of the wave equation and corresponds to a super-
position of a backward and forward travelling wave.
————————————————————————————————–
6.2 Convolution Theorem

A convolution of two functions f (x) and g(x) is defined as the integral
Z ∞
1
√ dt g(t) f ( x − t ) (595)
2π −∞
The convolution theorem expresses a convolution as the Fourier Transform

of the product of the two functions
Z ∞
1
√ dt g(t) f ( x − t )
2π −∞
Z ∞ Z ∞
1 ˜ 1
= √ dt g(t) dk f (k) √ exp + i k ( x − t )
2π −∞ −∞ 2π
116
Z ∞ Z ∞
1 1
= dk dt g(t) √ exp − i k t f˜(k) √ exp + i k x
−∞ −∞ 2π 2π
Z ∞
1
= dk g̃(k) f˜(k) √ exp + i k x (596)
−∞ 2π
which is the Fourier Transform of the product of Fourier Transforms. Inversely,
the inverse Fourier transform of a convolution is merely the product of Fourier
transforms.
————————————————————————————————–
Example:
The time-dependent response of a system A(t) to an applied time-dependent

field B(t) is given in terms of a response function χ(t − t0 ) such that
Z t
A(t) = dt0 χ(t − t0 ) B(t0 ) (597)
−∞
where the stimulus occurs at a time t0 that is earlier than the response time.
This expresses causality. The integral over t0 in this relation can be extended
to ∞ as Z ∞
A(t) = dt0 χ(t − t0 ) B(t0 ) (598)
−∞
if we define
χ(t − t0 ) = 0 for t0 > t (599)
Thus, the response has the form of the convolution.
The applied field can be Fourier Transformed into its frequency components,
and also the response can be frequency resolved into the components Ã(ω). The
relation between the frequency components of the response and the applied can
be obtained by Fourier Transforming the linear relation, which yields
Ã(ω) = χ̃(ω) B̃(ω) (600)
Therefore, the response relation simplifies in the frequency domain.
————————————————————————————————–
6.3 Parseval’s Relation

Parseval’s relation is given by
Z +∞ Z +∞
dk f˜(k) g̃(k) = dx f (x) g ∗ (x) (601)
−∞ −∞
117
and can be derived using the completeness relation
Z +∞ Z +∞ Z +∞
∗ ˜ 1
dx f (x) g (x) = dx dk f (k) √ exp + i k x g ∗ (x)
−∞ −∞ −∞ 2π
Z +∞ Z +∞ Z +∞
1 1
= dx dk f˜(k) √ exp + i k x dk 0 g̃ ∗ (k 0 ) √ exp − i k 0 x
−∞ −∞ 2π −∞ 2π
Z +∞ Z +∞ Z +∞
dx
= dk f˜(k) dk 0 g̃ ∗ (k 0 ) exp + i ( k − k 0 ) x
−∞ −∞ −∞ 2 π
Z +∞ Z +∞
= dk f˜(k) dk 0 g̃ ∗ (k 0 ) δ( k − k 0 )
−∞ −∞
Z +∞
= dk f˜(k) g̃ ∗ (k) (602)
−∞
which is Parseval’s relation.
118
7 Fourier Series
The Stürm-Liouville equation
∂2φ
+ n2 φ = 0 (603)
∂x2
with eigenvalue − n2 has solutions cos nx and sin nx that can be used to form
an orthonormal set, as
Z 2π
dx sin mx sin nx = π δn,m m = 6 0
0
Z 2π
dx cos mx cos nx = π δn,m m 6= 0
0
Z 2π
dx sin mx sin nx = 0 (604)
0
The basis function corresponding to n = 0 is non-degenerate and can be taken

to be
1
√ (605)
2π
The other basis functions, with n > 0, are degenerate and can be taken to be
1
√ sin nx
π
1
√ cos nx (606)
π
This set of eigenfunctions generates the finite Fourier series expansion, whereby
any well behaved function on the interval (0, 2π) can be expanded as
∞
1 X 1 1
f (x) = a0 √ + an √ cos nx + bn √ sin nx (607)
2π n=1
π π
where the coefficients (an < bn ) are calculated from

Z 2π
1
a0 = dt f (t) √
0 2π
Z 2π
1
an = dt f (t) √ cos nt
0 π
Z 2π
1
bn = dt f (t) √ sin nt (608)
0 π
119
This leads to an explicit form of the completeness condition, in which we define
the dirac delta function restricted to the interval (0, 2π) to be ∆(x − t) so
1 1 X
∆(x − t) = + ( cos nx cos nt + sin nx sin nt )
2π π n=1
1 1 X
= + cos n ( x − t ) (609)
2π π n=1
It can be shown that the set of Fourier expansion coefficients (an , bn ) in the
expansion of f (x) are the coefficients that minimize the difference
N
a0 X 1
χ(x) = f (x) − √ − √ an cos nx + bn sin nx (610)
2π n=1
π
as they can be determined from the χ2 minimization scheme in which

Z 2π
I = dx χ2 (x) ≥ 0 (611)
0
and
∂I ∂I
= = 0 ∀n ≤ N (612)
∂an ∂bn
This leads to the equations
Z 2π
1
a0 = √ dt f (t)
2π 0
Z 2π
1
an = √ dt f (t) cos nt
π 0
Z 2π
1
bn = √ dt f (t) sin nt (613)
π 0
For these values of the expansion coefficients, the difference χ(x) only vanishes
for almost all values of x if
I = 0 (614)
This condition allows the function to deviate from the Fourier series only at a
set of isolated points that contribute zero to the integral. The condition I = 0
is equivalent to Bessel’s inequality being satisfied, since
Z 2π N
X
2
I = dt f (t) − a20 − a2n + b2n (615)
0 n=1
The quantity a2n + b2n (which is a function of n) is known as the power spectrum
of f (x).
120
The basis functions φn (x) are continuous at every point in the interval (0, 2π)
yet, it is possible to expand a square integrable f (x) with a finite number of
discontinuities as a Fourier series. If f (x) has a discontinuity at x = x0 it can
be proved that the series converges to the value of

1
lim f (x0 + ) + f (x0 − ) (616)
→ 0 2
This can be seen by evaluating the Fourier series expansion of
f (x) = x 0 ≤ x < π
f (x) = x − 2π π < x ≤ 2π (617)
In this case, it is easily seen that the constant term is zero

Z 2π
1
a0 = √ dt f (t)
2π 0
Z π Z 2π
1
= √ dt f (t) + dt f (t)
2π 0 π
Z π Z 2π
1
= √ dt t + dt ( t − 2 π )
2π 0 π
2
4 π2 π2

1 π
= √ + − − 2ππ
2π 2 2 2
= 0 (618)
Using integration by parts one finds that the coefficients of the cosine terms are
also zero
Z 2π
1
an = √ dt f (t) cos nt
π 0
Z π Z 2π
1
= √ dt f (t) cos nt + dt f (t) cos nt
π 0 π
Z π Z 2π
1
= √ dt t cos nt + dt ( t − 2 π ) cos nt
π
0 π

1 cos nπ − 1 cos nπ − 1
= √ −
π n2 n2
an = 0 (619)
However, the coefficients of the sine terms are evaluated as

Z 2π
1
bn = √ dt f (t) sin nt
π 0
121
Z π Z 2π
1
= √ dt f (t) sin nt + dt f (t) sin nt
π 0 π
Z π Z 2π
1
= √ dt t sin nt + dt ( t − 2 π ) sin nt
π 0 π

1 1 1 − cos nπ
= √ 2π − +
π n n
√ cos nπ
= −2 π
n
√ 2
bn = π ( − 1 )n (620)
n
and are non-zero. Hence, we have the Fourier series expansion
∞
X ( − 1 )n
f (x) = 2 sin nx (621)
n=1
n
which just consists of an expansion in terms of the sine functions.
Direct evaluation of the series at x = π yields a sum of zero since sin nπ =

0. The original function is undefined at the discontinuity, but has a value of
π just below the discontinuity and a value of − π just above the discontinuity.
Thus, we see that
1
0 = π + ( −π ) (622)
2
as expected.
Homework: 14.1.4
Problem:
Expand x2 and x4 in a Fourier series and then evaluate the series at the
point x = π.
A numerical evaluation of the Fourier series in the vicinity of a discontinuity,

shows the Gibbs phenomenon. Consider a square wave train
f (x) = 1 for 0 < x < π

f (x) = 0 for π < x < 2 π (623)
this has a discontinuity at x = π. The Fourier series expansion is given by
1 2 X sin(2n + 1)x
f (x) = + (624)
2 π n=0 2n + 1
122
The Fourier series assigns a value to the function at the discontinuity, which is
the mean from just above and below. Furthermore, the Fourier series truncated
after N terms, also over-estimates the function or overshoots it just before the
discontinuity. This is the Gibbs phenomenon.
7.1 Gibbs Phenomenon

In order to discuss the Gibb’s phenomenon, it is necessary to sum the Fourier
series. It can be shown that
∞
1 1 X
∆(x − t) = + cos n ( x − t ) (625)
2π π n=1
or, equivalently
∞
1 1 X
∆(x − t) = + Real exp i n ( x − t ) (626)
2π π n=1
then, on summing only the first N terms, one has

N
1 1 X
∆N (x − t) = = + Real exp i n ( x − t )
2π π n=1

exp i ( N + 1 ) ( x − t ) − exp i ( x − t ) !
1 1
= + Real
2π π
exp i ( x − t ) − 1
( N + 1 ) ( x − t )
!
1 1 cos 2 sin N ( x2 − t )
= +
2π π sin ( x −
2
t )
" #
1 sin ( N + 12 ) ( x − t )
= (627)
2π sin ( x − t )
2
Hence, on defining the sum of the first N terms of the Fourier series for f (x) as
fN (x) one has
Z 2 π
fN (x) = dt f (t) ∆N (x − t)
0
" #
Z 2π 1
1 sin ( N + 2 ) ( x − t)
= dt f (t) x − t )
2π 0 sin ( 2
(628)
123
The Gibbs phenomenon can be demonstrated by the square wave in which case
Z π
fN (x) = dt 1 ∆N (x − t)
0
" #
sin ( N + 12 ) ( x − t )
Z π
1
= dt (629)
2π 0 sin ( x − t ) 2
At the discontinuity at x = π, one has

" #
Z π 1
1 sin ( N + 2 ) ( π − t)
fN (π) = dt π − t )
(630)
2π 0 sin ( 2
which, on introducing the variable y = π − t, can be written as

" #
sin ( N + 12 ) y
Z π
1
fN (π) = dy (631)
2π 0 sin y2
1
The integral can be evaluated in the limit N → ∞ by writing z = ( N + 2 ) y,
so that
Z ∞ " #
1 sin z
lim fN (π) = dz
N →∞ π 0 z
1
= (632)
2
The over shoot at x = 0 can be estimated by rewriting the partial sum, by
shifting the variable of integration to s = x − t
!
sin ( N + 12 ) s
Z x
1
fN (x) = ds (633)
2 π x−π sin 2s
N+ 1
The integrand is symmetrical in s and has a maximum value of π 2 at s = 0
and first falls to zero at s = ± N π+ 1 . We shall examine the behavior of the
2
series at the discontinuity at x = 0. At x = 0 the integral over s starts with a
small value of the integrand which oscillates about zero and the integrand then
attains its maximum value at the upper limit of integration. However, it is clear
to see that the integral will be greater if the range of integration of s covers both
positive regions around the central maximum. This occurs when x = N π+ 1 .
2
In this case, one estimates the maximum value of the partial sum is given by
Z π1 " #
1
1 N+
2
sin ( N + 2 ) s
fN (x) ∼ ds
max 2 π −π sin 2s
124
Z π1 " #
1
1 1 N+
2
sin ( N + 2 )s
= + ds
2 2π 0 sin 2s
Z π " #
1 1 sin z
∼ + dz (634)
2 π 0 z
where
1
z = (N + )s (635)
2
The second integral is greater than 0.5, its’ value is 0.588. The truth of this in-
equality can be seen by plotting the integrand. The integrand when considered
as function of s is oscillatory, with a constant period ∼ 2π N . As x is increased
in steps of 2π
N , the upper limit of s includes one more cycle of the integrand,
and the upper limit of the z integration increases by a multiple of 2π. Each
successive half cycle yields a contribution of opposite sign and smaller magni-
tude than the previous half cycle. Thus, when taken in pairs, the contribution
from the entire z interval of (0, ∞) [which contributes 0.5] is smaller than the
contribution from the first half cycle. Therefore, the series over shoots at the
discontinuity by 8%.
An alternative formulation of Fourier series makes use of another linear com-

bination of the degenerate eigenfunctions of the Stürm-Liouville equation with
periodic boundary conditions

1
φm (ϕ) = √ exp + i m ϕ
2π

1
φ−m (ϕ) = √ exp − i m ϕ (636)
2π
and the non-degenerate function
1
φ0 (ϕ) = √ (637)
2π
This is the complex representation, in which the eigenfunctions are symmetri-
cally normalized via
Z 2π
dϕ φ∗m (ϕ) φn (ϕ) = δn,m (638)
0
and any complex function Φ(ϕ) on the interval 2 π > ϕ > 0 can be expanded
as X
Φ(ϕ) = Cm φm (ϕ) (639)
m
125
The complex coefficients Cm are given by
Z 2π
1
Cm = √ dϕ exp − i m ϕ Φ(ϕ) (640)
2π 0
If the function Φ is real, then one must have
∗
C−m = Cm (641)
These functions often occur as solutions of Stürm-Liouville equations in the az-

imuthal angle in spherical polar coordinates.
————————————————————————————————–
Example:
Consider a cylindrical metal sheet of radius a and infinite length that has
been cut across a diameter. The two sheets are almost touching. One half of
the cylinder is kept at a potential φ0 and the other half is kept at a potential
−φ0 . Determine the potential at an arbitrary point (z, r, ϕ) inside the cylinder.
Assume, that the potential φ(z, r, ϕ) inside the cylinder is governed by the
equation
∇2 φ = 0 (642)
or in cylindrical coordinates
∂2φ 1 ∂2φ

1 ∂ ∂φ
2
+ r + = 0 (643)
∂z r ∂r ∂r r2 ∂ϕ2
Solution:
This can be solved by noting that the problem is invariant under transla-
tions along the cylinder’s axis. Thus, the potential is only a function of (r, ϕ)
alone. Furthermore, one can assume that the potential can be expanded in a
discrete Fourier series in ϕ with arbitrary coefficients that depend on r, since
the potential is periodic in ϕ.
X Cm (r)
φ(r, ϕ) = √ exp i m ϕ (644)
m
2π
On substituting the series expansion into Laplace’s equation, and on multiplying

by the complex conjugate function

1
√ exp − i n ϕ (645)
2π
126
and integrating over ϕ between 0 and 2 π, one finds that the expansion coefficient
Cn (r) satisfies the equation
n2

1 ∂ ∂Cn (r)
r − 2 Cn (r) = 0 (646)
r ∂r ∂r r
Thus, we find that

Cn (r) = An rn + Bn r−n (647)
so that the potential is of the form
∞
1 X
n −n
φ(r, ϕ) = √ An r + Bn r exp inϕ (648)
2π n=−∞
Since, the potential must be finite at the center of the cylinder (r = 0) one
can set Bn = 0. Furthermore, from the boundary condition at r = a one has
φ(a, ϕ) = φ0 for 2 π > ϕ > π

φ(a, ϕ) = − φ0 for π > ϕ > 0 (649)
Then, using the Fourier series expansion of φ(a, ϕ) and the orthogonality con-
dition one has
Z 2π Z π
φ0
√ dϕ exp − i n ϕ − dϕ exp − i n ϕ = An an (650)
2π π 0
which uniquely determines An as

√ sin n2π

−n nπ
An = − 2 π φ0 a exp − i n π
2 2
(651)
Since, for even n one has sin nπ

2 = 0, it is useful to re-write the coefficients as
4i φ0
A2n+1 = √ a−(2n+1)
2 π 2n + 1
A2n = 0 (652)
which are only non-zero for odd n. Hence, the expansion only contains terms
that have odd n. The potential is given by
∞ 2n+1
φ0 X 8 r
φ(r, ϕ) = − sin n ϕ (653)
2 π n=0 2 n + 1 a
127
————————————————————————————————–
Example:
An interesting example is given by Laplace’s equation for a potential φ in a

two dimensional region in the shape of a wedge, of angle πb , where b > 21 . The
boundary conditions used will be
π
Θ(r, ) = Θ(r, 0) = 0 (654)
b
Solution:
Laplace’s equation in two dimensions becomes
1 ∂2φ

1 ∂ ∂φ
r + 2 = 0 (655)
r ∂r ∂r r ∂θ2
and this can be solved by separation of variables, and then series expansion.
The solutions are sought in the form
φ(r, θ) = R(r) Θ(θ) (656)
On substitution of this ansatz into Laplace’s equation and writing the separation
constant as − µ2 , one obtains the two ordinary differential equations. These
consist of the differential equation for the angular part
∂2Θ
= − µ2 Θ (657)
∂θ2
and the radial equation
µ2

1 ∂ ∂R
r − R = 0 (658)
r ∂r ∂r r2
The angular equation has solutions
Θ(θ) = A sin µ θ
Θ(θ) = B cos µ θ (659)
However, due to the boundary conditions, the only allowable solutions are
Θ(θ) = A sin µ θ (660)
where
π
µ = mπ (661)
b
128
and m is any positive integer. That is, the allowable values of µ are given by
µ = mb (662)
The functions are normalized such that

Z πb Z π
A2
A2 dθ sin2 m θ b = dx sin2 m x
0 b 0
Z π
A2
= dx ( 1 − cos 2 m x )
2b 0
π
= A2
2b
= 1 (663)
The radial equation becomes

∂ ∂R
r r = m2 b2 R (664)
∂r ∂r
The radial equation has solutions
R(r) = C rµ + D r−µ
= C rmb + D r− mb
(665)
As the solution must be regular at r = 0 one has
R(r) = Cm rmb (666)
and the potential can be expanded as a Fourier series

∞ r
X
mb 2b
φ(r, θ) = Cm r sin m b θ (667)
m=1
π
The coefficients Cm have to be determined from additional boundary conditions.
The above solution has the interesting property that when b > 1, which
means that the angle between the conducting planes is obtuse, then the radial
component of the electric field is given
∞ r
X
mb−1 2b
Er = − m b Cm r sin m b θ (668)
m=1
π
The Fourier component of the field with m = 1 varies as

r
b−1 2b
− C1 b r sin b θ (669)
π
129
which is unbounded for b < 1 as r → 0. Thus, electric fields can be excep-
tionally large and have corner singularities close to the edge on a metal object
with an acute angle, such as lightning rods.
————————————————————————————————–
As an example of the occurrence of a high dimensional eigenvalue equation,

consider Laplace’s equation in three-dimensions
∇2 Φ(x, y, z) = 0 (670)
Laplace’s equation can be considered as an eigenvalue equation with eigenvalue

λ = 0. The solutions should form a complete set. Since one suspects that
all functions can be expanded as polynomials in the variables (x, y, z), one can
look for polynomial solutions. Some solutions are easily identified, for example
φ0 (x, y, z) = 1 (671)
or one has the three linearly independent linear functions
φ1,c (x, y, z) = x
φ1,s (x, y, z) = y
φ1,0 (x, y, z) = z (672)
Although there are six quadratic form there are only five linearly independent
quadratic forms which satisfy Laplace’s equation. These are
φ2,s (x, y, z) = yz
φ2,c (x, y, z) = xz
φ2,2s (x, y, z) = xy
φ2,2c (x, y, z) = x2 − y 2
φ2,0 (x, y, z) = 2 z 2 − x2 − y 2 (673)
Thus, the zero eigenvalue of the Laplace operator ∇2 is highly degenerate. In

spherical polar coordinates (r, θ, ϕ) the eigenfunctions are
φ0 (r, θ, ϕ) = 1 (674)
and
φ1,c (r, θ, ϕ) = r sin θ cos ϕ

φ1,s (r, θ, ϕ) = r sin θ sin ϕ
φ1.0 (r, θ, ϕ) = r cos θ (675)
130
The linear functions, for fixed (r, θ), form a set of orthogonal functions of ϕ.
Likewise, for the set of quadratic functions
φ2,s (r, θ, ϕ) = r2 sin θ cos θ sin ϕ

φ2,c (r, θ, ϕ) = r2 sin θ cos θ cos ϕ
φ2,2s (r, θ, ϕ) = r2 sin2 θ cos ϕ sin ϕ
φ2,2c (r, θ, ϕ) = r2 sin2 θ ( cos2 ϕ − sin2 ϕ )
φ2,0 (r, θ, ϕ) = r2 ( 2 cos2 θ − sin2 θ ) (676)
which again form a set orthogonal functions of ϕ, because the integral over ϕ
vanishes Z 2π
dϕ φ∗2,m (θ, ϕ) φ2,n (θ, ϕ) = 0 (677)
0
if n 6= m. Furthermore, even if the eigenfunctions have the same ϕ dependence,
they are orthogonal to the linear functions with weight factor sin θ as the integral
Z π
dθ φ∗2,m (θ, ϕ) sin θ φ1,m (θ, ϕ) = 0 (678)
0
vanishes. For example, when m = 0, and on substituting x = cos θ our

functions are recognized as being orthogonal polynomials which are proportional
to the Legendre polynomials Pn (x). Also note that there are only five linearly
independent functions φ2,m , since the remaining linear combination is just
x2 + y 2 + z 2 = r 2 (679)
which has the same θ and ϕ dependence as φ0 .
————————————————————————————————–
Example:
Find all (7) real linearly cubic expressions that are solutions of Laplace’s
equation, and then express them in terms of spherical polar coordinates.
Hint: Enumerate all cubic mononomials e.g x3 . . . xyz. Express a general

cubic polynomial in terms of these mononomials. Laplace’s equation yields a re-
lationship between the coefficients of the mononomials. Use the Gram Schmidt
method, to find the linearly independent solutions of Laplace’s equation.
Solution:
131
By substitution of the form
φ = γ x y z + α1 x3 + α2 y 3 + α3 z 3
+ β1 x y 2 + β 2 x z 2 + β3 y x 2 β4 y z 2 + β5 z x 2 + β6 z y 2
(680)
one finds the seven linearly independent solutions
φ3,2s = xyz
φ3,2c = z ( x2 − y 2 )
φ3,0 = 2 z 3 − 3 z ( x2 + y 2 )
φ3,3s = x3 − 3 x y 2
φ3,3c = y 3 − 3 y x2
φ3,1s = y 3 + y x2 − 4 y z 2
φ3,1c = x3 + x y 2 − 4 x z 2 (681)
These combinations are recognized as
φ3,2s = r3 cos θ sin2 θ sin ϕ cos ϕ

φ3,2c = r3 cos θ sin2 θ ( cos2 ϕ − sin2 ϕ )
φ3,0 = r3 ( 2 cos3 θ − 3 cos θ sin2 θ )
φ3,3s = r3 sin3 θ ( cos3 ϕ − cos ϕ sin2 ϕ )
φ3,3c = r3 sin3 θ ( sin3 ϕ − sin ϕ cos2 ϕ )
φ3,1s = r3 ( sin2 θ − 4 cos2 θ ) sin θ sin ϕ
φ3,1c = r3 ( sin2 θ − 4 cos2 θ ) sin θ cos ϕ (682)
which are orthogonal with each other, when considered as functions of ϕ.
————————————————————————————————–
132
8 Bessel Functions
Bessel’s Differential Equation is
d2 yn (x) dyn (x)

x2 + x + ( x2 − n 2 ) yn = 0 (683)
dx2 dx
This has a singularity at x = 0. This is a regular singular point. The solution
either diverges as x−n or xn . The two solutions are usually determined by the
behavior near x = 0.
8.0.1 The Generating Function Expansion

The solutions for the different values of n, which are convergent at x = 0, can
be determined from a generating function g(x, t)

x −1
g(x, t) = exp (t − t ) (684)
2
The generating function expansion is given by

n=∞
x X
exp ( t − t−1 ) = yn (x) tn (685)
2 n=−∞
The generating function is symmetric under the transformation
t → − t−1 (686)
which implies that yn (x) and y−n (x) are related via
y−n (x) = ( − 1 )n yn (x) (687)
On replacing t = 1 in the generating function expansion one then finds the

sum rule
n=∞
X
1 = yn (x) (688)
n=−∞
or
∞
X
1 = y0 (x) + 2 y2n (x) (689)
n=1
for all values of x.
133
8.0.2 Series Expansion
A series expansion ( in x ) for the functions yn (x) can be found by writing

xt x
g(x, t) = exp exp −
2 2t
∞
X (xt) X r ∞ s −s
s x t
= ( − 1 )
r=0
2r r! s=0 2s s!
∞
X xr+s tr−s
= ( − 1 )s
r,s=0
2r+s r! s!
(690)
This series expansion can be compared with the generating function expansion
∞
X
g(x, t) = yn (x) tn (691)
n=−∞
if one identifies n = r − s or alternatively
r = n + s (692)
The function yn (x), for positive n, is given by the coefficient of tn and is iden-
tified as a polynomial in x where the various powers correspond to the terms in
the sum over all values of s. The function is found as
∞
X xn+2s
yn (x) = ( −1 )s
s=0
2n+2s s! (n + s)!
∞
xn X x2s
= n
( − 1 )s 2s (693)
2 s=0 2 s! (s + n)!
which is a series expansion in x which vanishes at the origin for n > 0. For neg-
ative n, the properties of the generating function yield yn (x) = ( − 1 )n y−n (x)
which also vanishes at the origin.
8.0.3 Recursion Relations

On differentiating the generating function expansion with respect to t the func-
tions yn (t), for different values of n, are found to satisfy
n=∞
x x X
( 1 + t−2 ) exp ( t − t−1 ) = yn (x) n tn−1 (694)
2 2 n=−∞
134
On using the generating function expansion in the left-hand-side
n=∞ n=∞
x X X
( 1 + t−2 ) yn (x) tn = yn (x) n tn−1 (695)
2 n=−∞ n=−∞
and identifying the powers of tn−1 one finds the recursion relation
2n
yn−1 (x) + yn+1 (x) = yn (x) (696)
x
This recursion relation can be used to determine yn+1 (x) in terms of yn (x) and
yn−1 (x). However, since the errors increase with increasing n it is more efficient
to use the equation for yn (x) and yn+1 (x) to determine yn−1 (x) with lower val-
ues of n.
Another recursion relation can be found by taking the derivative of the

generating function with respect to x
n=∞
1 x X ∂yn (x) n
( t − t−1 ) exp ( t − t−1 ) = t (697)
2 2 n=−∞
∂x
which leads to
∂yn (x) 1
= yn−1 (x) − yn+1 (x) (698)
∂x 2
These two recursion relations can be combined to yield either
n ∂yn (x)
yn (x) + = yn−1 (x) (699)
x ∂x
or
n ∂yn (x)
yn (x) − = yn+1 (x) (700)
x ∂x
respectively, if one eliminates either yn+1 (x) or yn−1 (x).
Compact alternate forms of the above two recursion relations can be found.
For example, starting with
n ∂yn (x)
yn (x) + = yn−1 (x) (701)
x ∂x
one can multiply by xn and find that

∂
x yn (x) = xn yn−1 (x)
n
(702)
∂x
135
and likewise, starting from the other one of the above pair of relations, one can
show that
∂ −n
x yn (x) = − x−n yn+1 (x) (703)
∂x
8.0.4 Bessel’s Equation

These set of recursion relations can be used to show that the functions yn (x)
satisfy Bessel’s equations. On differentiating the recursion relation
∂yn (x)
n yn (x) + x = x yn−1 (x) (704)
∂x
one obtains
∂yn (x) ∂ 2 yn (x) ∂yn−1 (x)
(n + 1) + x = x + yn−1 (x) (705)
∂x ∂x2 ∂x
Multiplying the above equation by x and subtracting n times the recurrence
relationship for the derivative of yn (x), one obtains
∂ 2 yn (x) ∂yn (x) ∂yn−1

x2 +x − n2 yn (x) = x2 − ( n − 1 ) x yn−1 (x) (706)
∂x2 ∂x ∂x
The right-hand-side is identified as −x2 yn (x), by using the other recursion
relation for the derivative of yn−1 (x). Thus, we find that yn (x) does indeed
satisfy Bessel’s equation
∂ 2 yn (x) ∂yn (x)
x2 2
+ x + ( x2 − n2 ) yn (x) = 0 (707)
∂x ∂x
As Bessel’s equation is a second-order differential equation it has two solutions.
The two solutions can be found from the Frobenius method as series expansions.
The indicial equation yields
α 2 = n2 (708)
The solutions which remain finite or vanish at x = 0, and have the leading
term xn are known as the integer Bessel functions, and are denoted by Jn (x),
while the solutions that diverge as x−n at the origin are known as the Neumann
functions Nn (x). However, the generating function leads to a unique form of
yn (x), which has been found as an expansion in x. By comparison, one finds that
the functions yn (x) are proportional to the Bessel functions. The normalization
is chosen such that
yn (x) = Jn (x) (709)
136
8.0.5 Integral Representation
The series for the Bessel functions can be expressed in terms of an integral. This
can be shown by starting with the generating function expansion
∞
X
g(x, t) = J0 (x) + ( Jn (x) tn + J−n (x) t−n )
n=1
X∞
= J0 (x) + Jn (x) ( tn + ( − 1 )n t−n )
n=1
(710)
Changing variable from t to θ defined by t = exp[ iθ ] one finds that the

expansion takes the form
X∞
n
g(x, exp[ i θ ]) = J0 (x) + Jn (x) exp[ + i n θ ] + ( − 1 ) exp[ − i n θ ]
n=1
X∞ ∞
X
= J0 (x) + J2n (x) 2 cos 2 n θ + J2n+1 (x) 2 i sin ( 2 n + 1 ) θ
n=1 n=0
(711)
where we have separated out the even and odd terms of the series. When
expressed in terms of θ, the generating function reduces to

g(x, exp[ i θ ]) = exp i x sin θ (712)
On splitting the complex form of the generating function expansion into equa-
tions for the real and imaginary parts one finds the two equations
∞
X
cos ( x sin θ ) = J0 (x) + 2 J2n (x) cos 2 n θ
n=1
∞
X
sin ( x sin θ ) = 2 J2n+1 (x) sin ( 2 n + 1 ) θ (713)
n=0
The Bessel functions can be obtained by using the orthogonality properties

of the trigonometric function on the interval (0, π). That is, we have
Z π
π
dθ cos nθ cos mθ = δn,m
0 2
Z π
π
dθ sin nθ sin mθ = δn,m (714)
0 2
137
Thus, on multiplying the real and imaginary part of the generating function by
either sin nθ or cos nθ and integrating, we have the four equations
Z 2π
1
J2n (x) = dθ cos( x sin θ ) cos 2nθ
π 0
Z 2π
1
J2n+1 (x) = dθ sin( x sin θ ) sin (2n + 1) θ
π 0
Z 2π
1
0 = dθ sin( x sin θ ) cos 2nθ
π 0
Z 2π
1
0 = dθ cos( x sin θ ) sin (2n + 1) θ (715)
π 0
depending on whether n is even or odd. Combining these equations in pairs, for
odd n and even n, one has
Z 2π
1
Jn (x) = dθ cos( x sin θ ) cos nθ + sin( x sin θ ) sin nθ
π 0
Z 2π
1
Jn (x) = dθ cos ( x sin θ − n θ ) (716)
π 0
for non zero n. The above equations are integral representations of the Bessel
functions.
8.0.6 Addition Theorem

An addition theorem can be found for the Bessel functions by noting that the
generating functions satisfy the relation
g(x + y, t) = g(x, t) g(y, t) (717)
so
∞
X ∞
X ∞
X
Jn (x + y) tn = Jm (x) tm Jl (y) tl (718)
n=−∞ m=−∞ l=−∞
or
∞
X ∞
X ∞
X
n
Jn (x + y) t = Jm (x) Jn−m (y) tn (719)
n=−∞ n=−∞ m=−∞
Hence, on equating the coefficients of tn , one has the Bessel function addition
theorem
∞
X
Jn (x + y) = Jm (x) Jn−m (y) (720)
m=−∞
138
The Bessel functions often occur in problems that involve circular planar
symmetry, like a vibrating drum head. In this case, one describes the location
of a point on the circular surface of the drum by planar polar coordinates (r, ϕ).
Another application occurs in the theory of diffraction.
Example:
Consider a light wave falling incident normally on a screen with a circular

opening of radius a. It will be seen that the intensity of the transmitted light
falling on a distant screen will form a circular pattern. The intensity of the
transmitted light may fall to zero on concentric circles. The strength of the
electric field passing through the circular aperture and arriving at a distant
point on the other side of the aperture, directed an angle α normal to the
aperture is given by
Z a Z 2π
2π
E ∼ dr r dϕ exp i r sin α cos ϕ (721)
0 0 λ
This electric field is the sum of the amplitudes of the light originating from
the points (r, ϕ) inside the circular aperture, weighted by the phase from their
differences in optical path length r cos ϕ sin α. The integration over ϕ can be
performed using the integral representation of the Bessel function
Z a
2π
E ∼ dr r 2 π J0 r sin α (722)
0 λ
The integration over r can be performed by using the recursion relations, leading
to
λa 2πa
E ∼ J1 sin α (723)
sin α λ
The intensity of the light, I, arriving at the distant screen is proportional to
| E |2 , so
2
λa 2πa
I ∼ J12 sin α (724)
sin α λ
The angle α lies between 0 and π2 so the factor sin α vanishes at 0. Thus,
the denominator vanishes for light transmitted in the normal direction, but
fortunately so does the Bessel function J1 , as can be seen from the Frobenius
series. Therefore, on using l’Hopital’s rule, the intensity is non-vanishing for
a point directly in front of the aperture, as expected. However, the intensity
falls to zero in specific directions α determined by the vanishing of the Bessel
function J1
2πa
J1 sin α = 0 (725)
λ
139
For a sufficiently large aperture, a λ, the intensity falls to zero at the angles.
The first zero is found at the angle α determined by
2π a
sin α = 3.8137 (726)
λ
For a macroscopic opening a = 0.5 × 10−2 m and λ ∼ 10−7 m one has
α ∼ 10−5 rad (727)
Thus, the first circle of zero intensity makes has an extremely small angular
spread. The outer circles are determined by the higher order zeroes of the
Bessel function, and for large n are approximated by
2πa
sin α ∼ n π (728)
λ
which is similar to the Bragg condition. The outer dark rings are hard to ob-
serve as the intensity between the rings is very low.
Example:
A particle of mass m is confined within a cylinder of radius a and length l.

The particle is in an energy eigenstate described by the equation
h̄2
− ∇2 Ψ = E Ψ (729)
2m
where the allowed value of the energy is denoted by E. The wave function
Ψ(r, θ, z) expressed in terms of cylindrical coordinates satisfies the boundary
conditions
Ψ(a, θ, z) = 0
Ψ(r, θ, 0) = Ψ(r, θ, l) = 0
(730)
Find an expression for the wave functions and the allowed energies.
The eigenvalue equation can be written in the form
1 ∂2Ψ ∂2Ψ

1 ∂ ∂Ψ 2mE
r + 2 + = − Ψ (731)
r ∂r ∂r r ∂θ 2 ∂z 2
h̄2
On substituting the ansatz for the eigenfunction
Ψ(r, θ, z) = R(r) Θ(θ) Z(z) (732)
140
into the equation and diving by Ψ one finds
1 ∂2Θ 1 ∂2Z

1 ∂ ∂R 2mE
r + 2 2
= − 2 − (733)
r R ∂r ∂r r Θ ∂θ h̄ Z ∂z 2
Since the z dependence is entirely contained in the right-hand-side, and the
left-hand-side is constant as the independent variable z is changed, we must
have
∂2Z
= − κ2 Z (734)
∂z 2
where − κ2 is an arbitrary constant. This equation has a general solution
Z(z) = A cos κz + B sin κz (735)
Since Ψ must vanish on the two surfaces z = 0 and z = l one determines the
constants as
A = 0
B sin κl = 0 (736)
Hence, we have the allowed solutions
nz π z
Z(z) = B sin (737)
l
where nz is an arbitrary positive integer, and κ = nzl π . On substituting Z(z)
back into the differential equation, and multiplying by r2 one has
1 ∂2Θ n2z π 2

r ∂ ∂R 2 2mE
r + = −r − (738)
R ∂r ∂r Θ ∂θ2 h̄2 l2
From the above equation one can recognize that the θ dependent term must
also be a constant, say − m2 , and therefore satisfies the differential equation
∂ 2 Θ(θ)
= − m2 Θ (739)
∂θ2
which has solutions of the form

Θ(θ) = exp imθ (740)
Since the wave function has a unique value at any point, one must have
Θ(θ + 2 π) = Θ(θ) (741)
which implies that m must be a positive or negative integer.
Finally, we find that the radial function satisfies the equation
n2z π 2

∂ ∂R 2mE
r r − m2 R = − r 2 − R (742)
∂r ∂r h̄2 l2
141
We shall put this equation into a dimensionless form by introducing the variable
x = k r, where k is to be determined. The differential equation of R(x/k)
becomes
x2 n2z π 2

∂ ∂R 2mE
x x − m2 R = − 2 − R (743)
∂x ∂x k h̄2 l2
If we choose the value of k to be given by
n2z π 2

2mE
k2 = − (744)
h̄2 l2
the differential equation has the form of Bessel’s equation

∂ ∂R
x x + ( x2 − m2 ) R = 0 (745)
∂x ∂x
and has the solutions

R(r) = Jm (kr) (746)
which are regular at the origin. Since, the wave function must vanish at the
walls of the cylinder, the value k must satisfy
Jm (ka) = 0 (747)
and so the allowed values of k are given by

zm,n
km,n = (748)
a
where zm,n is the n-th zero of the m-th Bessel function, i.e.
Jm (zm,n ) = 0 (749)
Thus we have shown that the eigenfunctions are of the form of

nz π z zm,n r
Ψ(r, θ, z) = sin exp i m θ Jm ( ) (750)
l a
and the allowed energy eigenvalues are simply given by

2
h̄2 zm,n n2z π 2

E = + (751)
2m a2 l2
Example:
142
Consider the electromagnetic field inside a cylindrical metal cavity of radius
a and length l. Maxwell’s equations are given by
→
−
→
− →
− 1 ∂B
∇ ∧ E = − (752)
c ∂t
and →
−
→
− →
− 1 ∂E
∇ ∧ B = + (753)
c ∂t
Maxwell’s equations can be combined to yield the wave equation governing the
electric field →
−
1 ∂2 E

→
− →
− →
−
∇ ∧ ∇ ∧ E = − 2 (754)
c ∂t2
and since Gauss’s law holds
→
− → −
∇ . E = 0 (755)
one has →
−
→
− 1 ∂2 E
− ∇2 E = − 2 (756)
c ∂t2
On representing the time dependence of the electric field as a wave of frequency
ω via
→
− →− →
− →−
E ( r , t) = E ( r ) Real exp[ i ω t ] (757)
then one finds that the spatial dependence of the field satisfies
→
− − ω2 →
− −
− ∇2 E (→
r ) = 2 E (→
r) (758)
c
In cylindrical coordinates the z component of the electric field satisfies the Lapla-
cian eigenvalue equation
1 ∂ 2 Ez ∂ 2 Ez ω2

1 ∂ ∂Ez
r + 2 2
+ 2
= − 2 Ez (759)
r ∂r ∂r r ∂θ ∂z c
On using the method of separation of variables we assume that the z-component
of the electric field can be written as the product
Ez = R(r) Θ(θ) Z(z) (760)
On substituting this ansatz into the partial differential equation one finds that it
separates into three ordinary differential equations. The z-dependent function
Z(z) satisfies
∂ 2 Z(z)
= − kz2 Z(z) (761)
∂z 2
and Θ(θ) satisfies
∂ 2 Θ(θ)
= − m2 Θ(θ) (762)
∂θ2
143
Using the values of the separation constants, one finds that the Radial function
R(r) has to satisfy
m2
2
1 ∂ ∂R ω 2
r − 2 R = − − kz R (763)
r ∂r ∂r r c2
which is Bessel’s equation. Thus, the general solution of the partial differential
equation has been found to be of the form

→
−
X
Ez ( r ) = Jm (kr r) exp i m θ am,kr sin kz z + bm,kr cos kz z (764)
m,kr
where the constants of separation are related by

ω2
= kz2 + kr2 (765)
c2
The boundary condition at the cylindrical walls is that the component of the
electric field parallel to the axis vanishes. This implies that
Jm (kr a) = 0 (766)
Hence, the allowed values of kr are given in terms of the zeros of the m-th Bessel
function. The allowed k values are given by
zm,n
kr = (767)
a
where zmn stands for the n-th zero of the m-th Bessel function, i.e. Jm (zm,n ) =
0. The electromagnetic field must also satisfy boundary conditions at the two
ends of the cylinder. The boundary conditions at the ends of the cylinder z = 0
and z = l are satisfied by setting am,kr = 0 and kz = nzl π for integer and
zero values of nz . In this case, the tangential components of the field Er and
Eθ vanish at z = 0 and z = l. This leads to the magnetic induction field
being purely transverse. The allowed values of the frequencies are given by
r
2
zm,n n2 π 2
ω = c 2
+ z2 (768)
a l
8.0.7 Orthonormality
The orthonormality relations for the Bessel functions can be proved by starting
from Bessel’s equation
∂ 2 φν (x) ∂φν (x)
x2 + x + ( x2 − ν 2 ) φν (x) = 0 (769)
∂x2 ∂x
144
Due to physical reasons, we shall write x = k r and use zero boundary condi-
tions at the center and edge of the circular area r = 0 and at r = a. Thus,
we demand that our solution must satisfy
φν (0) = φν (ka) = 0 (770)
In this case, Bessel’s equation has the form
∂2 ν2

1 ∂ 2
φ ν (kr) + φ ν (kr) + k − φν (kr) = 0
∂r2 r ∂r r2
ν2

1 ∂ ∂
r φν (kr) + k2 − 2 φν (kr) = 0 (771)
r ∂r ∂r r
where k 2 can be regarded as the eigenvalue. The eigenvalue, or rather the value
km , is determined by demanding that the solution of Bessel’s equation φν (km r)
satisfies the boundary condition
φν (km a) = 0 (772)
The orthogonality of φν (km a) and φν (kn a) can be proved by multiplying the

equation for φν (km a) by φν (kn a) and subtracting this from the analogous equa-
tion with m and n interchanged

1 ∂ ∂ 1 ∂ ∂
φν (km r) r φν (kn r) − φν (kn r) r φν (km r)
r ∂r ∂r r ∂r ∂r
2
= φν (kn r) ( km − kn2 ) φν (km r)
(773)
On multiplying by the weighting factor r ( which, due to the two dimensional

circular symmetry, is related to the infinitesimal area element given by dr r dϕ
) and integrating by parts, one finds that
a a
∂ ∂
φν (km r) r φν (kn r) − φν (kn r) r
φν (km r)
∂r 0 ∂r 0
Z a
2 2
= ( km − kn ) dr r φν (kn r) φν (km r) (774)
0
Using the boundary conditions at r = a and noting that the solutions are
finite at r = 0, one finds that the eigenfunctions corresponding to different
eigenvalues are orthogonal, as
Z a
2
( km − kn2 ) dr r φν (kn r) φν (km r) = 0 (775)
0
145
The normalization can be found by setting km = kn + , hence, on using
the boundary condition at r = 0 one has
a
0 0

φν (kn a + a) a φν (kn a) − φν (kn a) a φν (kn a + a)
0
Z a
= 2 dr r φν (kn r) φν (kn r)
0
(776)
and using the boundary conditions one finds

Z a
2 dr r φν (kn r) φν (kn r) = a2 kn φ02
ν (kn a) (777)
0
Furthermore, since
ν
φν+1 (kn a) = φν (kn a) − φ0ν (kn a) (778)
kn a
one has a
a2
Z
dr r φ2ν (kn r) = φ2ν+1 (kn a) (779)
0 2
In the limit a → ∞, the normalization condition becomes

Z ∞
1
dr φν (kr) r φν (k 0 r) = δ(k 0 − k) (780)
0 k
8.0.8 Bessel Series

Since the set of Bessel functions φν (kn a) for fixed ν and different km values form
a complete set, then an arbitrary function Φ(r) can be expanded as a Bessel
Series X
Φ(r) = Cm φν (km r) (781)
m
for a > r > 0. The expansion coefficients Cm can be found via

Z a
2
Cm = 2 2 dr Φ(r) r φν (km r) (782)
a φν+1 (km a) 0
146
Example:
The amplitude u(r, θ, t) of the vertical displacements of a circular drumhead

of radius a satisfies the wave equation
1 ∂2 u
∇2 u − = 0 (783)
c2 ∂t2
where c is the velocity of the sound waves on the drumhead. Since the drumhead
is fixed tightly to the circular rim, the boundary conditions is that there is no
vertical displacement so
u(a, θ, t) = 0 (784)
In circular coordinates one has
1 ∂2u 1 ∂2u

1 ∂ ∂u
r + 2 = (785)
r ∂r ∂r r ∂θ2 c2 ∂t2
This partial differential equation can be solved by assuming that the solution
can be expressed as a Fourier series in the time variable. For any fixed time
t, the solution can be considered to be a function of (r, θ). The θ dependence
can be expressed as a series in Legendre polynomials, where the coefficients are
functions of r. The undetermined coefficients which are functions of r can then
be expanded as a Bessel series. The general term in this multiple expansion is
given by the ansatz

u(r, θ, t) = R(r) Θ(θ) exp i ω t (786)
On substituting this ansatz into the equation, cancelling the common time-
dependent terms, the equation reduces to
R(r) ∂ 2 Θ(θ) ω2

1 ∂ ∂R(r)
Θ(θ) r + 2 2
= − 2 R(r) Θ(θ) (787)
r ∂r ∂r r ∂θ c
On multiplying by r2 R−1 (r) Θ−1 (θ) one has
1 ∂ 2 Θ(θ) ω2

r ∂ ∂R(r)
r + 2
= − 2 r2 (788)
R(r) ∂r ∂r Θ(θ) ∂θ c
This equation can be written such that one side depends on r alone and another
side which depends on θ alone
ω2 1 ∂ 2 Θ(θ)

r ∂ ∂R(r)
r + 2 r2 = − (789)
R(r) ∂r ∂r c Θ(θ) ∂θ2
147
Therefore, the solution can be obtained by the method of separation of variables.
It can be seen that, since the left-hand-side is independent of θ and the right-
hand-side is independent of r, they must be equal to a constant, ( say − m2 ).
Thus, we have
∂ 2 Θ(θ)
= − m2 Θ(θ) (790)
∂θ2
which has solutions of the form
Θ(θ) = A cos mθ + B sin mθ (791)
Since, one can identify the point (r, θ) as (r, θ + 2π), then one requires that
Θ(θ + 2π) = Θ(θ) (792)
which requires that m is an integer. Furthermore, the other side of the equation
is equal to the same constant
ω2

∂ ∂R(r)
r r + 2 r2 R(r) = m2 R(r) (793)
∂r ∂r c
which can be recognized as Bessel’s equation of integer order
∂ 2 Jm (x) ∂Jm (x)

x2 + x + ( x2 − m2 ) Jm (x) = 0 (794)
∂x2 ∂x
if the variable x = ωc r. The solutions Jm (x) are retained and the Neumann
functions are discarded as the amplitude is expected to not diverge at r = 0.
Thus, for a fixed Fourier component ω, the solution can be written as
∞
X ω
uω (r, θ, t) = exp i ω t Jm r ( Am (ω) cos mθ + Bm (ω) sin mθ )
m=0
c
(795)
The allowed values of ω, ωn , are determined by the boundary condition, which
becomes
ωn
Jm a = 0 (796)
c
Therefore, the frequencies of the normal modes of the drum head are determined
by the zeroes of the Bessel functions Jm (x). The general solution, is a sum over
the different frequencies ωn , and thus is in the form of a Bessel series
∞ X
X ωn
u(r, θ, t) = exp i ωn t ( Am (ωn ) cos mθ + Bm (ωn ) sin mθ ) Jm r
m=0 ω
c
n
(797)
The expansion coefficients have to be determined from the initial conditions.
This can be done, as an arbitrary initial condition can always be expanded in
148
terms of a Bessel series.
Example:
A disk of radius R in the x-y plane ( z = 0 ) is kept at a constant potential

φ0 and the rest of the plane z = 0 is kept at zero potential. Find the potential
for z > 0.
The potential satisfies Laplace’s equation
∇2 φ = 0 (798)
which in cylindrical symmetry, where the potential is independent of θ, becomes
∂2φ

1 ∂ ∂φ
r + = 0 (799)
r ∂r ∂r ∂z 2
Using the method of separation of variables one seeks solutions of the homoge-
neous equation in the form
φ(r, z) = R(r) Z(z) (800)
On substitution of the ansatz for the solution, one finds that R and Z must
satisfy
1 ∂2Z

1 ∂ ∂R
r = − (801)
r R ∂r ∂r Z ∂z 2
Since the left and right-hand-side are functions of independent variables r and
z, they must be equal to a constant ( say − k 2 ). Then the differential equation
is equivalent to the pair of ordinary differential equations
∂2Z
= k2 Z (802)
∂z 2

1 ∂ ∂R
r + k2 R = 0 (803)
r ∂r ∂r
The equation for Z(z) has the solution

Z(z) = A exp − k z + B exp + k z (804)
but since the potential must vanish as z → ∞ one has B = 0. Likewise, the
radial part of the potential is given by
R(r) = J0 (kr) (805)
149
as the potential remains finite at the center of the disk, r = 0. Thus, we
find the general solution of the partial differential equation is of the form of the
linear superposition
Z ∞
φ(r, z) = dk A(k) J0 (kr) exp − k z (806)
0
where A(k) has to be determined by the boundary condition at z = 0.
Imposing the boundary condition at z = 0 yields

Z ∞
φ(r, 0) = dk A(k) J0 (kr) (807)
0
On multiplying by J0 (k 0 r) and integrating over r with a weight factor of r one

has
Z ∞ Z ∞ Z ∞
dr r J0 (k 0 r) φ(r, 0) = dr r J0 (k 0 r) dk A(k) J0 (kr) (808)
0 0 0
On interchanging the order of integration and using the continuum form of the
orthonormality relation
Z ∞
1
dr r J0 (kr) J0 (k 0 r) = δ( k − k 0 ) (809)
0 k
one determines the coefficients A(k 0 ), as the integral
Z R
1 0
A(k ) = φ 0 dr r J0 (k 0 r) (810)
k0 0
However, due to the Bessel function recursion relation

∂
r J1 (k 0 r) = k 0 r J0 (k 0 r) (811)
∂r
one can perform the integration and find
A(k 0 ) = φ0 R J1 (k 0 R) (812)
Hence, we have the potential as

Z ∞
φ(r, z) = φ0 R dk J1 (kR) J0 (kr) exp − k z (813)
0
150
The potential above the center of the disk ( r = 0 ) is given by the integral
Z ∞
φ(0, z) = φ0 R dk J1 (kR) exp − k z
0
z
= φ0 1 − √ (814)
z 2 + R2
This can be seen by noting the special case of the Bessel function recurrence
relations yields
∂J0 (x)
J1 (x) = − (815)
∂x
This relation can be used in the integration of the expression for the potential.
On integrating by parts with respect to k, one finds the equation
k=∞ Z ∞

φ(0, z) = φ0 − J0 (kR) exp − k z − z dk J0 (kR) exp − k z
k=0 0
Z ∞
φ(0, z) = φ0 1 − z dk J0 (kR) exp − k z
0
(816)
The last integral can be evaluated by using the integral representation of the
Bessel function.
Z ∞ Z ∞ Z π
2 2
dk J0 (kR) exp − k z = dk dϕ cos( kR sin ϕ ) exp − k z
0 0 π 0
Z π Z ∞
2 2
= dϕ dk cos( kR sin ϕ ) exp − k z
π 0 0
Z π
2 2 z
= dϕ 2
π 0 z + R2 sin2 ϕ
1
= √ (817)
z + R2
2
Thus, the potential on the axis is given by

z
φ(0, z) = φ0 1 − √ (818)
z + R2
2
Example:
A cylinder, of length l and radius a, has the top plate held at a potential
φ(l, r) and the bottom plate is held at a potential φ(0, r). The potential between
the plates is given by
∇2 φ = 0 (819)
151
which reduces to
∂2φ

1 ∂ ∂φ
r + = 0 (820)
r ∂r ∂r ∂z 2
From separation of variables one finds the solutions of the homogeneous equation
in the form
φ(z, r) = R(r) Z(z) (821)
On substituting this form into the partial differential equation, one finds that
1 ∂2Z

1 ∂ ∂R
r = − (822)
r R ∂r ∂r Z ∂z 2
Thus, the equations for the r dependence and z dependence separates. On

assuming a separation constant of − 2 , one finds that the differential equation
for R and Z is equivalent to the pair of ordinary differential equations
∂2Z
= k2 Z (823)
∂z 2
and
1 ∂ ∂R
r + k2 R = 0 (824)
r ∂r ∂r
The equation for Z(z) has a general solution of the form
Z(z) = Ak cosh[ k z ] + Bk sinh[ k z ] (825)
while the solution of the radial equation which is regular at the center of the
cylinder, r = 0, is given by
R(r) = J0 (kr) (826)
The Bessel function of zero-th order occurs since the potential is invariant under
arbitrary rotations about the cylinder’s axis. The boundary condition at the
surface of the cylinder is equivalent to requiring
J0 (ka) = 0 (827)
which yields the allowed values of k, in terms of the zeroes of the zeroth order
Bessel function, kn = zan . The solution can be expressed as
X
φ(z, r) = Ak cosh[ k z ] + Bk sinh[ k z ] J0 (kr) (828)
k
152
The coefficients Ak and Bk can be obtained, by first finding the coefficients
in the Bessel expansion of φ(0, r) and φ(l, r). At the top plate, ( z = l ), we
expand the boundary value as a Bessel series
X
φ(l, r) = φ̃k (l) J0 (kr) (829)
k
where the expansion coefficients are found from

Z a
2
φ̃k (l) = 2 2 dr r φ(l, r) J0 (kr) (830)
a J1 (ka) 0
and likewise at z = 0 one can also expand φ(0, r). Then, at the bottom plate,
( z = 0 ), we have X
φ(0, r) = φ̃k (0) J0 (kr) (831)
k
where the coefficients are found from
Z a
2
φ̃k (0) = dr r φ(0, r) J0 (kr) (832)
a2 J12 (ka) 0
Then finally, we have the solution in the form of a sum

X sinh k (l − z ) sinh k z

φ(z, r) = φ̃k (0) + φ̃k (l) J0 (kr) (833)
sinh k l sinh k l
k
where the allowed values of k are given by the zeroes of the Bessel function.
Homework: 11.2.3
Homework: 11.2.9
8.1 Neumann Functions

The Neumann functions Nn (x) are also solutions of Bessel’s equation, and are
known as Bessel functions of the second kind. They are distinguished from the
Bessel functions of the first kind Jn (x) in that they diverge as x−n in the limit
x → 0. In particular, for non-integer ν they are defined in terms of Jν (x) and
J−ν (x) via
cos νπ Jν (x) − Jν (x)
Nν (x) = (834)
sin νπ
Thus, the Neumann functions are particular linear combinations of the Bessel
functions. On substituting the power series expansion for Jν (x) one finds
ν
(ν − 1)! 2
Nν (x) = − + ... (835)
π x
153
for ν > 0.
Example:
An example which involves Neumann functions is the co-axial wave guide.

The co-axial wave guide or cable is composed of two metal cylinders, with
the same axis. The radius of the outer cylinder is b and the radius of the
inner cylinder is a. The electromagnetic field is confined in the region enclosed
between the two cylinders. We shall consider the component of the electric field
Ez , along the direction of the axis, êz . This satisfies
→
−
1 ∂2 E

→
− →
− →
−
∇ ∧ ∇ ∧ E = − (836)
c2 ∂t2
or on Fourier Transforming with respect to time
ω2 →

→
− →
− →
− −
∇ ∧ ∇ ∧ E = 2 E (837)
c
However, one has the identity

→
− →
− →
− →
− →
− → − → −
∇ ∧ ∇ ∧ E = − ∇2 E + ∇ ( ∇ . E ) (838)
→
− → −
and Gauss’s law reduces to ∇ . E = 0 in vacuum with no charges present.
Thus, the z-component of the electric field is governed by
2
ω
∇2 Ez + Ez = 0 (839)
c
In cylindrical coordinates this partial differential equation has the form

2
1 ∂ 2 Ez ∂ 2 Ez

1 ∂ ∂Ez ω
r + 2 + + Ez = 0 (840)
r ∂r ∂r r ∂θ2 ∂z 2 c
On using separation of variables one assumes that the E.M. waves have the form
X
Ez (r, θ, z, t) = Am (k) Rm (r) Θ(θ) Z(z) exp − i ω t (841)
m,k
On solving for the (θ, z) dependence, and using the condition that Ez is 2 π
periodic in θ one has travelling waves of the form
X
Ez (r, θ, z, t) = Cm (k) Rm (r) exp − i m θ exp i ( k z − ω t ) (842)
m,k
154
where the radial function satisfies
m2
2
1 ∂ ∂Rm ω 2
r − 2 Rm + − k Rm = 0 (843)
r ∂r ∂r r c2
The radial equation has a solution for b > r > a given by
Rm (r) = Am Jm (αr) + Bm Nm (αr) (844)

2
where α2 = ωc2 − k 2 . The Neumann functions are allowed since the region
where they diverge, ( r = 0 ), has been excluded. The allowed values of α
are determined by the boundary conditions at r = a and r = b where the
tangential field Ez must vanish. Hence, the ratio of the two coefficients and α
are determined from the boundary conditions
Rm (a) = Am Jm (αa) + Bm Nm (αa) = 0 (845)
and
Rm (b) = Am Jm (αb) + Bm Nm (αb) = 0 (846)
The allowed values of α are determined from the transcendental equation
Nm (αa) Jm (αb) = Nm (αb) Jm (αa) (847)
The frequencies of the E.M. waves are given by
ω2
= α2 + k 2 (848)
c2
Since k 2 must be positive if the wave is to be transmitted, the minimum fre-
quency is given by ωmin = c α.
The Hankel functions are defined as solutions of Bessel’s equation which are
the superpositions
Hν+ (x) = Jν (x) + i Nν (x) (849)
and
Hν− (x) = Jν (x) − i Nν (x) (850)
The Hankel functions have the asymptotic forms of

±
lim Hν (x) → exp ± i x (851)
x→∞
and since they diverge at the origin has a natural interpretation in terms of
outgoing cylindrical waves or incoming cylindrical waves.
Example:
155
Consider the diffraction of an electromagnetic plane wave E0 with frequency
ω, by a metal wire of radius a. The electric vector is parallel to the cylindrical
axis, and the direction of propagation is perpendicular to the cylindrical axis.
The incident field is

êz E0 exp i ( k r cos θ − ω t ) (852)
The amplitude of the diffracted electric field satisfies Helmholtz’s equation

1 ∂ 2 Esc

1 ∂ ∂Esc
r + 2 + k 2 Esc = 0 (853)
r ∂r ∂r r ∂θ2
The amplitude must also satisfy the boundary condition that the tangential field
vanishes at the surface of the conductor

Esc (a, θ) + E0 exp i k a cos θ = 0 (854)
The boundary condition at infinity is given by the radiation condition

∂Esc
lim + i k Esc = 0 (855)
r→∞ ∂r
On applying the method of separation of variables one finds the solution as

a series
∞
X
+ −
Esc (r, θ) = Am cos m θ + Bm sin m θ Cm Hm (kr) + Dm Hm (kr)
m=0
(856)
On examining the boundary condition at r → ∞ one finds that the expansion
coefficient Dm = 0. From the symmetry condition θ → − θ, the solution
must be an even function of θ so one has Bm = 0. Thus, the solution is of the
form
X∞
+
Esc (r, θ) = Cm cos m θ Hm (kr) (857)
m=0
From the boundary condition at r = a one has
∞
X
+
Cm cos m θ Hm (ka) + E0 exp i k a cos θ = 0 (858)
m=0
Furthermore, as
∞
X
exp i k a cos θ = J0 (ka) + 2 ( i )m Jm (ka) cos m θ (859)
m=1
156
one can determine the coefficients Cm uniquely. In particular
C0 H0+ (ka) = − E0 J0 (ka) (860)
and
Cm H0+ (ka) = − 2 ( i )m E0 Jm (ka) (861)
Thus, the scattered wave is given by
∞
J0 (ka) +
X
m Jm (ka) +
Esc = − E0 H (kr) + 2 (i) Hm (kr) cos m θ
H0+ (ka) 0 m=1
+
Hm (ka)
(862)
8.2 Spherical Bessel Functions

The spherical Bessel Functions jn (x) are related to Bessel functions of half inte-
ger order. The spherical Bessel functions occur in three-dimensional problems
with spherical symmetry. For example, in three-dimensions the radial part of
the solution of Laplace’s equation can be written as

1 ∂ 2 ∂R(r) 2 l(l + 1)
r + k − R(r) = 0 (863)
r2 ∂r ∂r r2
On substituting the form of the radial function as
φ(kr)
R(r) = 1 (864)
( k r )2
one obtains Bessel’s equation of half integer order
2

2 ∂ φ ∂φ 2 2 1 2
r + r + k r − (l + ) φ = 0 (865)
∂r2 ∂r 2
Thus one has a solution for the radial function of the form
Jl+ 21 (kr)
R(r) = √ (866)
kr
The spherical Bessel functions are defined as the solution
r
π
jm (x) = J 1 (x) (867)
2 x m+ 2
incorporating a specific constant of proportionality. The spherical Neumann
functions are defined analogously by
r
π
ηm (x) = N 1 (x) (868)
2 x m+ 2
157
It can be seen that the zeroth order spherical Bessel function j0 (x) is given by
sin x
j0 (x) = (869)
x
The recurrence relations for the spherical Bessel functions are given by
2n + 1
jn−1 (x) + jn+1 (x) = jn (x)
x
n jn−1 (x) − ( n + 1 ) jn+1 (x) = ( 2 n + 1 ) jn0 (x)
(870)
The recurrence relations can be used to evaluate the low order spherical Bessel
functions
sin x cos x
j1 (x) = 2
−
x x
3 − x2 3
j2 (x) = sin x − 2 cos x (871)
x3 x
The general form of the spherical Bessel functions can be obtained from the
recursion relations by combining them into the form

∂ n+1
x jn (x) = xn+1 jn−1 (x)
∂x

∂ −n
x jn (x) = − x−n jn+1 (x) (872)
∂x
By induction one can establish the Rayleigh formula

n
n n 1 ∂
jn (x) = ( − 1 ) x j0 (x) (873)
x ∂x
for the spherical Bessel function of order n. The asymptotic, large x, behavior
is given by
1 π
jn (x) = sin x − n (874)
x 2
This follows as the asymptotic large x behavior is governed by the terms of
lowest power in x1 . The derivatives present in the Rayleigh formula produce a
faster decay and hence negligible contributions when they act on the powers of
x. Thus the leading term of the large x behavior is determined by the term
where the n derivatives in the Rayleigh formula for j0 (x) all act on the factor
sin x.
158
8.2.2 Orthogonality Relations
The orthogonality relations for the spherical Bessel functions can be expressed
as Z a 2
a3

2 0
dr jn (kr) r jn (k r) = jn+1 (ka) δk,k0 (875)
0 3
where k and k 0 are solutions of the equation
jn (ka) = 0 (876)
expressing the boundary condition.
8.2.3 Spherical Neumann Functions

The spherical Neumann functions can also be evaluated explicitly as
cos x
η0 (x) = −
x
cos x sin x
η1 (x) = − 2
−
x x
3 − x2

3
η2 (x) = − cos x − 2 sin x (877)
x3 x
etc. It can be seen that the spherical Neumann functions diverge at the origin.
Example:
The free particle energy eigenvalue equation in three-dimensions
h̄2
− ∇2 Ψ = E Ψ (878)
2m
has an energy eigenfunction Ψ that can be expressed as
Ψ(r, θ, ϕ) = R(r) Θ(θ) Φ(ϕ) (879)
The wave function Ψ(r, θ, ϕ) with angular momentum l has a radial wave func-
tion Rl (kr) which satisfies the differential equation

∂R(r) 2 ∂R(r) 2 l(l + 1)
+ = k − R(r) = 0 (880)
∂r2 r ∂r r2
h̄2 k2
where the energy eigenvalue is given by E = 2 m . The radial function has
the general solution
Rl (r) = Al jl (kr) + Bl ηl (kr) (881)
159
The solution which is regular at r = 0 corresponds to the case where Bl = 0.
Example:
The energy eigenfunction for a particle, of angular momentum l, moving in

a short ranged potential V (r) such that V (r) = 0 for r > a has a radial wave
function
for r > a. The wave functions have the asymptotic form

1 π
jl (kr) ∼ sin k r − l
kr 2

1 π
ηl (kr) ∼ − cos k r − l
kr 2
(883)
Bl
On writing Al = − tan δl (k), one finds that

1 π
Rl (r) = Al sin kr − l + δl (k) (884)
kr 2
for r > a. The energy eigenfunction is of the form of a standing spherical

wave, in which the effect of the potential is contained in the phase shift δl (k).
Example:
A quantum mechanical particle in a stationary state experiences a strong

repulsive potential V (r) from a nucleus which prevents the particle from entering
the nucleus. The potential V (r) is spherically symmetric and can be represented
by

∞ r < a
V (r) = (885)
0 r > a
The wave function Ψ, which expresses the state of the particle, satisfies the
energy eigenvalue equation
h̄2
− ∇2 Ψ + V (r) Ψ = E Ψ (886)
2m
The effect of the potential is to demand that the wave function Ψ(r, θ, ϕ) satisfies
the boundary condition
Ψ(a, θ, ϕ) = 0 (887)
160
at r = a. This corresponds to the condition that the particle does not enter
the nucleus. Due to the spherical symmetry of the potential, we have a solution
of the form
Ψl,m (r, θ, ϕ) = Rl (r) Ylm (θ, ϕ) (888)
The radial function Rl (r) is given by
where Al and Bl are arbitrary constants, and the energy eigenvalue is given by
2
k2
E = h̄2 m . The boundary condition at the surface of the spherical nucleus
determines the ratio of the constants
0 = Al jl (ka) + Bl ηl (ka) (890)
Therefore, one has

Bl jl (ka)
= − (891)
Al ηl (ka)
In this case, the phase shift δl (k) is determined by the energy and the radius of
the nucleus
jl (ka)
tan δl (k) = (892)
ηl (ka)
Thus, determination of the angle and energy dependence of the scattering cross-
section can be used to determine the phase shift, which can then be used to
determine the characteristic properties of the scattering potential.
Homework: 11.7.22
An important example is given by the expansion of a plane wave in terms of

spherical Bessel functions
∞
X
exp i k r cos θ = Cn jn (kr) (893)
n=0
The coefficient Cn is evaluated as
Cn = in ( 2 n + 1 ) Pn (cos θ) (894)
where Pn (x)are the Legendre Polynomials. It should be expected that plane

waves can be expressed in terms of spherical Bessel functions with the same k
values as both the spherical Bessel functions and the plane wave are both free
particle energy eigenfunctions.
161
9 Legendre Polynomials
9.0.4 Generating Function Expansion
The generating function g(x, t) yields an expansion for the Legendre polynomials
Pn (x). The generating function expansion is simply
− 12
2
g(x, t) = 1 − 2xt + t
∞
X
= Pn (x) tn (895)
n=0
This has direct application in electrostatics, where the potential φ(→−r ) due to
→
−
a charge distribution ρ( r ) is given by the solution of Poisson’s equation. The
solution can be expressed in terms of the Green’s function and the charge den-
sity, and the Green’s function can be expanded using the generating function
expansion
→
−
3→ρ( r0 )
−0
Z
φ(→
−
r) = d r →
−
|→
−r − r0 |
Z ∞ →
−
ρ( r0 )
Z π Z 2π
= dr0 r02 dθ sin θ dϕ √
0 0 0 r2 − 2 r r0 cos θ + r02
∞ π Z 2π ∞ n
→
− X 1
Z Z
r<
= dr0 r02 dθ sin θ dϕ ρ( r0 ) Pn (cos θ)
0 0 0 n=0
r> r>
(896)
→
−
where θ is the angle between → −r and r0 , and r< and r> are, respectively, the
0
smaller and large values of (r, r ).
Problem:
Evaluate the potential due to a spherically symmetric charge density
ρ(r) = 0 for r > a

ρ(r) = ρ0 for r < a (897)
Assume that the Legendre polynomials Pl (cos θ) are orthogonal, with weight
factor sin θ.
162
9.0.5 Series Expansion
The generating function can be used to find the series expansion of the Legendre
polynomials. First we expand in powers of 2 x t − t2 , via
1
g(x, t) = √
1 − 2 x t + t2
∞
X (2s)!
= 2s (s!)2
( 2 x t − t2 )s (898)
s=0
2
and then expand in powers of t

∞
X (2s)!
g(x, t) = ts ( 2 x − t )s
s=0
22s(s!)2
∞ ∞
X (2s)! s
X s!
= 2s 2
t ( − 1 )r ( 2 x )s−r tr
s=0
2 (s!) r=0
r! (s − r)!
(899)
On writing n = s + r and keeping n fixed s = n − r, so
n
[2]
∞ X
X (2n − 2r)!
g(x, t) = tn ( − 1 )r ( 2 x )n−2r
n=0 r=0
22n−2r r! (n − r)! (n − 2r)!
n
[2]
∞ X
X (2n − 2r)!
= tn ( − 1 )r ( x )n−2r
n=0 r=0
2n r! (n − r)! (n − 2r)!
(900)
Thus, the Legendre polynomials are given by
n
[2]
X (2n − 2r)!
Pn (x) = ( − 1 )r ( x )n−2r (901)
r=0
2n r! (n − r)! (n − 2r)!
Hence, the Legendre polynomials have the highest power of xn , and terms which
decrease in powers of x2 . Therefore, for odd n the series only contains odd terms
in x, whereas for even n the series is even in x. The series in decreasing pow-
ers of x−2 terminates when n = 2 r for even n and when n = 2 r + 1 for odd n.

Recursion relations for the Legendre polynomials can be found by taking the
derivative of the generating function with respect to t. Thus, starting from
− 12
2
g(x, t) = 1 − 2xt + t
163
∞
X
= Pn (x) tn
n=0
(902)
one obtains
∂ x − t
g(x, t) = 3
∂t ( 1 − 2 x t + t2 ) 2
∞
X
= Pn (x) n tn−1 (903)
n=0
The above equation can be written as

∞
X
( x − t ) g(x, t) = ( 1 − 2 x t + t2 ) n Pn (x) tn−1
n=0
∞
X X∞
(x − t) Pm (x) tm = ( 1 − 2 x t + t2 ) n Pn (x) tn−1
m=0 n=0
(904)
On equating like powers of t one has
( 2 n + 1 ) x Pn (x) = ( n + 1 ) Pn+1 (x) + n Pn−1 (x) (905)
This recursion relation can be used to construct the higher order Legendre poly-
nomials starting from P0 (x) = 1 and P1 (x) = x.
An alternate form of the recursion relation can be obtained by taking the

derivative of the generating function expansion with respect to x
∂ t
g(x, t) = 3
∂x ( 1 − 2 x t + t2 ) 2
∞
X ∂
= Pn (x) tn (906)
n=0
∂x
or
∞
X ∂
t g(x, t) = ( 1 − 2 x t + t2 ) Pn (x) tn
n=0
∂x
∞ ∞
X X ∂
t Pn (x) tn = ( 1 − 2 x t + t2 ) Pn (x) tn (907)
n=0 n=0
∂x
Equating like powers of t we obtain the recursion relation

∂ ∂ ∂
2x Pn (x) + Pn = Pn+1 (x) + Pn−1 (x) (908)
∂x ∂x ∂x
164
This is not very useful.
A more useful relation can be obtained, if one takes the derivative of
( 2 n + 1 ) x Pn (x) = ( n + 1 ) Pn+1 (x) + n Pn−1 (x) (909)
to give

∂ ∂ ∂
(2n+1) x Pn (x) + Pn (x) = (n+1)Pn+1 (x) + n Pn−1 (x)
∂x ∂x ∂x
(910)
This relation can be used to eliminate the derivative of Pn (x) in the previous
equation, by multiplying by 2 and subtracting ( 2 n + 1 ) times the previous
equation, leading to
∂ ∂
( 2 n + 1 ) Pn (x) = Pn+1 (x) − Pn−1 (x) (911)
∂x ∂x
On eliminating the derivative of Pn−1 (x), by multiplying by n and subtract-

ing, one obtains
∂ ∂
x Pn (x) + ( n + 1 ) Pn (x) = Pn+1 (x) (912)
∂x ∂x
Alternatively, on eliminating the derivative of Pn+1 (x) by multiplying by (n+1)
and subtracting, one obtains
∂ ∂
x Pn (x) − n Pn (x) = Pn−1 (x) (913)
∂x ∂x
The above two equations are to be combined. One is first put in the form
n → n − 1 so it becomes
∂ ∂
x Pn−1 (x) + n Pn−1 (x) = Pn (x) (914)
∂x ∂x
and this is combined with x times the relation
∂ ∂
x Pn (x) − n Pn (x) = Pn−1 (x) (915)
∂x ∂x
∂
so as to eliminate ∂x Pn−1 (x). On performing these manipulations one obtains
the recursion relation
∂
( 1 − x2 ) Pn (x) = n Pn−1 (x) − n x Pn (x) (916)
∂x
165
Finally, if one uses the above recursion relation together with
( 2 n + 1 ) x Pn (x) = ( n + 1 ) Pn+1 (x) + n Pn−1 (x) (917)
one obtains a recursion relation in the form

∂
( 1 − x2 ) Pn (x) = ( n + 1 ) x Pn (x) − ( n + 1 ) Pn+1 (x) (918)
∂x
————————————————————————————————–
Example:
The multi-pole expansion can be derived, using spherical polar coordinates

where z = r cos θ, from the properties of Legendre polynomials. The mathe-
matical basis for this expansion is found in the identity

∂ Pn (cos θ) Pn+1 (cos θ)
(n+1)
= −(n + 1) (919)
∂z r r(n+2)
This can be proved starting with the expression

∂ ∂r ∂ ∂ cos θ ∂
= + (920)
∂z x,y ∂z ∂r ∂z ∂ cos θ
p z
and with r = x2 + y 2 + z 2 and cos θ = √ one has
x2 + y 2 + z 2
2
r − z2

∂ z ∂ ∂
= + (921)
∂z x,y r ∂r r3 ∂ cos θ
Hence,

∂ Pn (cos θ) Pn (cos θ) 2 ∂ Pn (cos θ)
= − ( n + 1 ) cos θ + sin θ
∂z r(n+1) r(n+2) ∂ cos θ r(n+2)
(922)
but one has the recursion relation
∂Pn (cos θ)
sin2 θ = ( n + 1 ) cos θ Pn (cos θ) − ( n + 1 ) P(n+1) (cos θ) (923)
∂ cos θ
Thus, one has proved the identity

∂ Pn (cos θ) Pn+1 (cos θ)
(n+1)
= −(n + 1) (924)
∂z r r(n+2)
166
————————————————————————————————–
Alternatively, the above identity can be shown to be true by starting from

the generating function expansion
∞
1 X Pn (cos θ)
g(cos θ, r) = √ = (925)
1 + r2 − 2 r cos θ n=0
r(n+1)
On taking the derivative with respect to z, keeping (x, y) fixed, one has
∂ 1 1 − z
p = 3
∂z 1 + x + y2 + z2 − 2 z
2 (1 + x2+ y2 + z2 − 2 z ) 2
( 1 − r cos θ )
= 3 (926)
( 1 + r2 − 2 r cos θ ) 2
Hence, we have
X∞
∂ Pn (cos θ) ( 1 − r cos θ )
(n+1)
= 3 (927)
∂z n=0
r ( 1 + r2 − 2 r cos θ ) 2
which can be written as
X∞
∂ Pn (cos θ) ( 1 − r cos θ )
( 1 + r2 − 2 r cos θ ) =
r(n+1)
p
∂z n=0 ( 1 + r2 − 2 r cos θ )
∞
X Pn (cos θ)
= ( 1 − r cos θ )
n=0
r(n+1)
(928)
On substituting the identity that is being verified, we find
∞ ∞ ∞
X P(n+1) (cos θ) X cos θ Pn (cos θ) X Pn (cos θ)
( 1 + r2 − 2 r cos θ ) ( n + 1 ) = −
n=0
r(n+2) n=0
rn n=0
r(n+1)
(929)
or on changing the index of summation
∞ ∞ ∞
X P(n−1) (cos θ) X P(n+1) (cos θ) X cos θ Pn (cos θ)
n n
+ ( n + 1 ) n
= (2n + 1)
n=0
r n=0
r n=0
rn
(930)
which is satisfied identically, because of the recursion relation
( 2 n + 1 ) cos θ Pn (cos θ) = n P(n−1) (cos θ) + ( n + 1 ) P(n+1) (cos θ)
(931)
167
Thus, we have again verified the identity in question.
9.0.7 Legendre’s Equation

The Legendre polynomials satisfy Legendre’s equation. Legendre’s equation can
be derived starting with the recursion relation
∂
( 1 − x2 ) Pn (x) = n Pn−1 (x) − n x Pn (x) (932)
∂x
Differentiating the recursion relation with respect to x one obtains
∂2 ∂ ∂ ∂
( 1 − x2 ) Pn (x) − 2 x Pn (x) = n Pn−1 (x) − n x Pn (x) − n Pn (x)
∂x2 ∂x ∂x ∂x
(933)
and then using
∂ ∂
x Pn (x) − n Pn (x) = Pn−1 (x) (934)
∂x ∂x
to eliminate the derivative of Pn−1 (x) one obtains Legendre’s equation
∂2 ∂
( 1 − x2 ) Pn (x) − 2 x Pn (x) + n ( n + 1 ) Pn (x) = 0 (935)
∂x2 ∂x
Legendre’s differential equation has singular points at x = ± 1, where the
solution Pn (x) may become infinite. The value of n must be an integer if the
solution is to remain finite. If n is integer the Frobenius series expansion termi-
nates and the solution becomes a polynomial.
If the solution is assumed to be of the form

∞
X
Pn (x) = xα C n xn (936)
n=0
then it can be shown that the indicial equation yields
α = 0 (937)
and the coefficient of xm satisfies the recursion relation

( m + 2 ) ( m + 1 ) Cm+2 = m ( m − 1 ) + 2 m − n ( n + 1 ) Cm

= m ( m + 1 ) − n ( n + 1 ) Cm
(938)
168
which terminates after n terms when n is an integer.
Legendre’s equation is most frequently seen in a form where the independent

variable is in the form x = cos θ. In this case, the first-order derivative with
respect to x is given by
∂ 1 ∂
= − (939)
∂x sin θ ∂θ
and the second-order derivative is evaluated as
∂2

1 ∂ 1 ∂
=
∂x2 sin θ ∂θ sin θ ∂θ
2
1 ∂ cos θ ∂
= − (940)
sin2 θ ∂θ2 sin3 θ ∂θ
Thus, Legendre’s equation takes the form
∂2 cos θ ∂
2
Pn (cos θ) + Pn (cos θ) + n ( n + 1 ) Pn (cos θ) = 0
∂θ sin θ ∂θ
1 ∂ ∂
sin θ Pn (cos θ) + n ( n + 1 ) Pn (cos θ) = 0
sin θ ∂θ ∂θ
(941)
9.0.8 Orthogonality
Legendre’s differential equation in the variable x can be written as

∂ ∂
( 1 − x2 ) Pn (x) + n ( n + 1 ) Pn (x) = 0 (942)
∂x ∂x
which is a Stürm-Liouville equation, with eigenvalue n ( n + 1 ) and weighting
factor unity. The boundary conditions are imposed at x = ± 1, which are
regular singular points as ( 1 − x2 ) = 0 at the ends of the interval.
The orthogonality condition is found by multiplying by Pm (x) as the weight

factor is unity, and subtracting this from the equation with n and m inter-
changed. After integrating over x between +1 and −1 one obtains
Z +1
∂ 2 ∂ ∂ 2 ∂
dx Pm (x) (1 − x ) Pn (x) − Pn (x) (1 − x ) Pm (x)
−1 ∂x ∂x ∂x ∂x
Z +1
= n(n + 1) − m(m + 1) dx Pm (x) Pn (x)
−1
(943)
169
On integrating by parts, and noting that the factor ( 1 − x2 ) vanishes at both
the boundaries, then as long as Pn (±1) is not infinite one has
Z +1
n(n + 1) − m(m + 1) dx Pm (x) Pn (x) = 0 (944)
−1
Thus, the Legendre functions are orthogonal as long as the eigenvalues are not
equal n ( n + 1 ) 6= m ( m + 1 ).
The Legendre functions are defined such that Pn (1) = 1. The normalization
integral can be evaluated from the square of the generating function expansion
n=∞
X 2
( 1 − 2 x t + t2 )−1 = Pn (x) tn (945)
n=0
On integrating over x from −1 to +1 one has

Z +1 n=∞ Z +1
1 X
dx = t2n dx Pn2 (x) (946)
−1 1 − 2 x t + t2 n=0 −1
where we have used the orthogonality of the Legendre functions. The integral
can be evaluated as
n=∞ Z +1
1 1 − 2 t + t2 X
2n
− ln = t dx Pn2 (x) (947)
2t 1 + 2 t + t2 n=0 −1
The left-hand-side can be expanded in powers of t as

n=∞
1 1 + t X t2n
ln = 2 (948)
t 1 − t n=0
2n + 1
On equating like coefficients of t2n one has

Z +1
2
dx Pn2 (x) = (949)
−1 2n + 1
Hence, we can write the orthonormality condition as

Z +1
2
dx Pn (x) Pm (x) = δn,m (950)
−1 2n + 1
170
9.0.9 Legendre Expansions
Due to the orthogonality conditions and the completeness of the eigenfunctions
of a Stürm-Liouville equation, any function can be expanded on the interval
(−1, +1) as a Legendre series
n=∞
X
f (x) = Cn Pn (x) (951)
n=0
where the coefficients Cm can be evaluated from

Z +1
2
Cm = dt Pm (t) f (t) (952)
2m + 1 −1
The completeness condition can be expressed as

n=∞ Z +1
X 2n + 1
f (x) = Pn (x) dt Pn (t) f (t)
n=0
2 −1
Z +1 n=∞
X 2n + 1
= dt f (t) Pn (x) Pn (t) (953)
−1 n=0
2
Hence, we may expand the delta function, on the interval (−1, +1) as
n=∞
X 2n + 1
δ( x − t ) = Pn (x) Pn (t) (954)
n=0
2
————————————————————————————————–
Example:
Find the electrostatic potential for a point charge with charge q, located at a
distance a from the center, inside a uniform conducting spherical shell or radius
R.
The potential inside the sphere satisfies Poisson’s equation
− ∇2 φ = 4 π ρ (955)
where ρ is only non-zero at the position of the point charge. Elsewhere, Poisson’s
equation simplifies to
∇2 φ = 0 (956)
171
We shall assume that the polar axis, ( θ = 0 ) , runs through the center and
the charge q. Then the potential is invariant under changes of the azimuthal
angle ϕ, and so φ is independent of ϕ. Laplace’s equation takes the form

1 ∂ 2 ∂φ 1 ∂ ∂φ
r + 2 sin θ = 0 (957)
r2 ∂r ∂r r sin θ ∂θ ∂θ
This has solutions of the form

∞
X
φ(r, θ) = Al rl + Bl r−(l+1) Pl (cos θ) (958)
l=0
where Al and Bl are arbitrary constants that are to be determined by the bound-
ary conditions.
In order for the shell to be at constant potential one must have

∞
X
φ(R, θ) = Al Rl + Bl R−(l+1) Pl (cos θ) = φ0 (959)
l=0
Hence, for l 6= 0 one finds the relation between the expansion coefficients
Al = − Bl R−(2l+1) (960)
and for l = 0 one has

B0
A0 + = φ0 (961)
R
Thus, one can express the potential in terms of the coefficients Bl via
∞
R(l+1) rl

1 1 X
φ(r, θ) = φ0 + B0 − + Bl R−(l+1) (l+1)
− l Pl (cos θ)
r R r R
l=1
(962)
for r > a. The coefficients Bl can be found from the boundary condition at
the point (r = a, θ = 0). Near this point, the potential should be dominated by
the singular behavior of the point charge
∞
q q X al
lim φ(r, θ) ∼ = (963)
r →a (r − a) r rl
l=0
Hence, one finds the coefficients as
B0 = q
Bl = q al (964)
172
Thus, we have we have identified the contribution from the point charge q as
∞
q q X al
φq (r, θ) = √ = Pl (cos θ) (965)
r2 + a2 − 2 r a cos θ r rl
l=0
The induced contribution to the potential from the charge on the conducting
surface is found, from the principle of linear superposition, as
∞ l
q X ar
φind (r, θ) = − Pl (cos θ)
R R2
l=0
q 1
= − s
R 2
r a r a
1 − 2 R2 cos θ + R2
qR 1
= − s (966)
a 2
R2 R2
r2 + a − 2r a cos θ
which is just the contribution from a point image charge of magnitude q 0 =

R2
− q Ra located outside the shell at a distance a from the center. Since the
expansions converge, we have found the solution.
————————————————————————————————–
Example:
Consider a metallic sphere of radius a in a uniform electric field, E0 . The

potential φ in the vacuum satisfies Laplace’s equation
∇2 φ = 0 (967)
The electric field is chosen to be oriented along the polar axis, so the problem
has azimuthal symmetry and the potential is independent of ϕ. We shall use the
method of separation of variables and assume that the potential can be found
in the form of a series expansion
X
φ(r, θ) = Cl Rl (r) Θl (θ) (968)
l
which satisfies the partial differential equation

1 ∂ 2 ∂ φ 1 ∂ ∂φ
r + sin θ = 0 (969)
r2 ∂r ∂r r2 sin θ ∂θ ∂θ
173
On writing the separation constant as l ( l + 1 ) one has the radial function
Rl (r) satisfying the equation

∂ ∂R
r2 = l(l + 1)R (970)
∂r ∂r
Rl (r) = Al rl + Bl r−(l+1) (971)
The angular dependence is given by
Θl (θ) = Pl (cos θ) (972)
where l is an integer, for Θl (θ) to be non-singular along the poles.
Therefore, one has the solution in the form of a series expansion

X∞
l −(l+1)
φ(r, θ) = Al r + B l r Pl (cos θ) (973)
l=0
The coefficients Al and Bl are to be determined from the boundary conditions.
As r → ∞, the potential has to reduce to the potential of the uniform

electric field
lim φ(r, θ) → − E0 r cos θ
r→∞
= − E0 r P1 (cos θ)
(974)
Thus, only the two AL coefficients A0 and A1 can be non-zero. The potential
has the form
∞
X
φ(r, θ) = A0 − E0 r P1 (cos θ) + Bl r−(l+1) Pl (cos θ) (975)
l=0
The metallic sphere is held at a constant potential so

φ(a, θ) = const (976)
As the Legendre polynomials are linearly independent, the coefficients of the
Legendre polynomials must vanish at the surface of the sphere. Hence, we have
the set of equations
0 = − E0 a + B1 a−2
0 = Bl for l > 1
(977)
174
which leads to the potential having the form
a3

B0
φ(r, θ) = A0 + − E0 r P1 (cos θ) 1 − 3 (978)
r r
If the sphere is assumed to be uncharged we can set B0 = 0.
————————————————————————————————–
Example:
Consider the solution of Poisson’s equation
∇2 φ = − 4 π ρ(→
−
r) (979)
in which the charge density is distributed uniformly on a circle of radius a cen-

tered on the origin, and in the plane z = 0. The total charge on the ring
is q. We shall solve Poisson’s equation in a region that does not include the
charge distribution. The effect of the charge distribution, is to be included by
specifying a boundary condition.
For points not in the plane z = 0, the charge density is zero, and so one
has Laplace’s equation
∇2 φ = 0 (980)
Since, the geometry has an axial symmetry, the potential φ is only a function
of (r, θ). In spherical polar coordinates, one has
∂2φ

1 ∂ 2 ∂φ 1 ∂ ∂φ 1
r + sin θ + = 0 (981)
r2 ∂r ∂r r2 sin θ ∂θ ∂θ r2 sin2 θ ∂ϕ2
π
if θ = 2. Due to the axial symmetry, this reduces to

1 ∂ 2 ∂φ 1 ∂ ∂φ
r + 2 sin θ = 0 (982)
r2 ∂r ∂r r sin θ ∂θ ∂θ
This partial differential equation can be solved by the method of separation of

variables,
φ(r, θ) = R(r) Θ(θ) (983)
which, on dividing by φ, leads to the equation

1 ∂ 2 ∂R 1 ∂ ∂Θ
r = − sin θ (984)
R ∂r ∂r Θ sin θ ∂θ ∂θ
Then since one side of the equation is a function of r alone, and the other side
is a function of θ alone, both sides of the equation must be equal to a constant
175
C

∂ 2 ∂R
r = C R(r)
∂r ∂r

∂ ∂Θ
sin θ = − C sin θ Θ(θ)
∂θ ∂θ
(985)
On writing C = n ( n + 1 ), we find that
Rn (r) = An rn + Bn r−(n+1) (986)
and Θ(θ) is given by

Θn (θ) = Pn (cos θ) (987)
Thus a general solution, that does not diverge at infinity, is given by the series
X
φ(r, θ) = An r−(n+1) Pn (cos θ) (988)
n
This series is unique and the coefficients An can be determined from the appro-
priate boundary condition.
A boundary condition that can be easily determined is given by the potential

along the z-axis. Since all the elemental charges are located at equal distances
d from the point on the z axis
p
d = z 2 + a2 (989)
and in spherical polar coordinates r = z, θ = 0. The potential at a point on the

z-axis (r, 0) is just given by the sum of the contributions from each elemental
charge, the sum is just
1
φ(r, 0) = q √ (990)
r 2 + a2
where q is the total charge on the ring. The potential has the expansion
∞
X (2n)! a2n
φ(r, 0) = q ( − 1 )n (991)
n=0
22n (n!)2 r(2n+1)
On noting that P (cos 0) = 1, one can uniquely identify the coefficients An by

comparing the expansions in inverse powers of r as
(2n)!
A2n = q ( − 1 )n a2n (992)
22n (n!)2
while
A2n+1 = 0 (993)
176
Hence, we obtain the potential φ(r, θ) as an expansion
∞ 2n
q X (2n)! a
φ(r, θ) = ( − 1 )n 2n Pn (cos θ) (994)
r n=0 2 (n!)2 r
————————————————————————————————–
Example:
As another example consider a sphere that is cut into two hemispheres of

radius a. The hemisphere at r = a and with π > θ > π2 is held at a potential
φ0 , while the other hemisphere at r = a and π2 > θ > 0 is held at a potential
− φ0 .
Since the problem is symmetric under rotations around the z-axis, the po-
tential φ(r, θ) is independent of ϕ. The potential inside the sphere satisfies
Laplace’s equation
∇2 φ = 0 (995)
which in spherical polar coordinates, and using the azimuthal symmetry, be-
comes
1 ∂ 2 ∂φ 1 ∂ ∂φ
r + sin θ = 0 (996)
r2 ∂r ∂r r2 sin θ ∂θ ∂θ
The solution is found by the method of separation of variables
φ(r, θ) = R(r) Θ(θ) (997)
after which it is found that the radial function R(r) must satisfy an eigenvalue
equation
∂ 2 ∂R
r = n(n + 1)R (998)
∂r ∂r
Then, for fixed n we have the solution

Rn (r) = An rn + Bn r−(n+1) (999)
The angular part must satisfy Legendre’s equation

∂ ∂Θ(θ)
sin θ + n ( n + 1 ) Θ(θ) = 0 (1000)
∂θ ∂θ
which forces n to be an integer. The solution is given by the Legendre polynomial

in cos θ
Θn (θ) = Pn (cos θ) (1001)
177
Thus, we can form a general solution as a series expansion
∞
X
φ(r, θ) = An rn + Bn r−(n+1) Pn (cos θ) (1002)
n=0
The coefficients Bn are zero as the potential φ(0, θ) is expected to be finite.

The coefficients An can be obtained from the boundary condition at r = a.
That is, since one has
∞
X
φ(a, θ) = An an Pn (cos θ) (1003)
n=0
then, on using the orthogonality of the Legendre polynomials one has

Z π
2 n
An a = dθ sin θ φ(a, θ) Pn (cos θ)
2n + 1 0
Z π Z π2
= φ0 dθ sin θ Pn (cos θ) − dθ sin θ Pn (cos θ)
π
2 0
(1004)
When this integral is evaluated, since the function φ(a, θ) is odd only the Legen-
dre polynomials with odd values of n survive, as P2n (cos θ) is an even function
of cos θ. Then
∞
X (4n + 3) (2n − 1)!!
A2n+1 a2n+1 = φ0 ( − 1 )n+1 (1005)
n=0
(2n + 2)!!
Therefore, the potential is given as an expansion

∞ 2n+1
X
n+1 (4n + 3) (2n − 1)!! r
φ(r, θ) = φ0 (−1) P2n+1 (cos θ)
n=0
(2n + 2)!! a
(1006)
————————————————————————————————–
Example:
A plane wave can be expanded as

∞
→
− −
i k .→
X
exp r = Cl jl (kr) Pl (cos θ) (1007)
l=0
178
where θ is the angle between the position and the direction of the momentum.
This form of the expansion is to be expected as both the Bessel function expan-
sion and the plane wave are eigenstates of the Laplacian, with eigenvalue − k 2 .
It can be shown that Cl = il ( 2 l + 1 ).
On differentiating the expansion with respect to kr, one obtains

∞
→
− −
i cos θ exp i k . →
X
r = Cl jl0 (kr) Pl (cos θ) (1008)
l=0
or on using the expansion again

∞
X ∞
X
i cos θ Cl jl (kr) Pl (cos θ) = Cl jl0 (kr) Pl (cos θ) (1009)
l=0 l=0
The recurrence relation can be used to express cos θ Pn (cos θ) entirely in terms
of the Legendre polynomials via
( 2 l + 1 ) cos θ Pl (cos θ) = ( l + 1 ) Pl+1 (cos θ) + l Pl−1 (cos θ) (1010)
On multiplying by Pm (cos θ) sin θ and integrating one obtains

m (m + 1) 0
i Cm−1 jm−1 (kr) + Cm+1 jm+1 (kr) = C m jm (kr)
2m − 1 2m + 3
(1011)
Then on using the recurrence relations for the derivative of the spherical Bessel
functions
0 m (m + 1)
jm (kr) = jm−1 (kr) − jm+1 (kr) (1012)
2m + 1 2m + 1
one finds

m (m + 1)
i Cm−1 jm−1 (kr) + Cm+1 jm+1 (kr)
2m − 1 2m + 3

m (m + 1)
= Cm jm−1 (kr) − jm+1 (kr)
2m + 1 2m + 1
(1013)
Since the two Bessel functions, of order m − 1 and m + 1, can be considered to

be linearly independent, one can equate the coefficients to yield the recursion
relation
(2m + 1)
Cm = i Cm−1
(2m − 1)
(1014)
179
The first coefficient can be obtained by examining the limit k → 0 since in
this case only j0 (0) survives, and has a magnitude of unity. Hence,
C0 = 1 (1015)
and we have the expansion coefficients as
Cl = il ( 2 l + 1 ) (1016)
Thus, we have found the expansion of the plane wave in terms of the spherical
Bessel functions.
————————————————————————————————–
Example:
In quantum mechanics, the wave function in a scattering experiment is given

by
→
− − f (k, θ)
ψ(→
−
r ) = exp i k .→
r + exp ikr (1017)
r
→
−
which is a superposition of the incident or unscattered wave of momentum h̄ k ,
with a spherical outgoing scattered wave. The amplitude of the scattered wave,
with a central scattering potential, is given by
∞
1 X
f (k, θ) = ( 2 l + 1 ) exp i δl (k) sin δl (k) Pl (cos θ) (1018)
k
l=0
where θ is the scattering angle, h̄ k is the magnitude of the incident momentum,

and δl (k) is the phase shift produced by the short ranged scattering potential.
dσ
The differential scattering cross-section, dΩ multiplied by the solid angle sub-
tended by the detector dΩ governs the relative probability that a particle of
momentum k will be scattered through an angle θ into a detector that subtends
a solid angle dΩ to the target. The dependence on the size of the detector is
given by the factor dΩ. The differential scattering cross-section can be shown
to be given by the scattering amplitude through
dσ
(k, θ) = | f (k, θ) |2 (1019)
dΩ
The total scattering cross-section is determined as an integration over all

solid angles of the differential scattering cross-section
Z
dσ
σT (k) = dΩ (k, θ)
dΩ
180
Z 2π Z π
= dϕ dθ sin θ | f (k, θ) |2
0 0
(1020)
On substituting, the expansion in terms of the Legendre series, and integrating,

one can use the orthogonality relations and obtain the total scattering cross-
section as
∞
4π X
σT (k) = ( 2 l + 1 ) sin2 δl (k) (1021)
k2
l=0
9.1 Associated Legendre Functions

The Asssociated Legendre functions, Pnm (cos θ), satisfy the associated Legendre
m2

1 ∂ ∂ m
sin θ P (cos θ) + n(n + 1) − Pnm (cos θ) = 0
sin θ ∂θ ∂θ n sin2 θ
(1022)
or with the substitution x = cos θ one has
∂2 m2

2 m m
(1 − x ) P (x) − 2 x Pn (x) + n ( n + 1 ) − Pnm (x) = 0
∂x2 n 1 − x2
(1023)
This also has regular singular points at x ± 1. When m2 = 0, the associated
Legendre differential equation reduces to Legendre’s differential equation.
The associated Legendre functions are given by the expression

m ∂m
Pnm (x) = ( 1 − x2 ) 2 Pn (x) (1024)
∂xm
where Pn (x) are the Legendre polynomials. It can be proved that these func-
tions satisfy the associated Legendre differential equation. Let us note that the
associated Legendre functions are zero if m > n, since the polynomial Pn (x)
is a polynomial of order n.
9.1.1 The Associated Legendre Equation

The associated Legendre functions satisfy the associated Legendre equation, as
can be shown by by starting from Legendre’s equation
∂2 ∂
( 1 − x2 ) 2
Pn (x) − 2 x Pn (x) + n ( n + 1 ) Pn (x) = 0 (1025)
∂x ∂x
181
and then differentiating m times. On using the formula
n=m
∂m ∂ m−n ∂n
X
A(x) B(x) = C(m, n) A(x) B(x) (1026)
∂xm n=0
∂xm−n ∂xn
one obtains
∂ 2+m ∂ 1+m
( 1 − x2 ) 2+m
Pn (x) − 2 x ( m + 1 ) Pn (x)
∂x ∂x1+m
∂m
+(n − m)(n + m + 1) Pn (x) = 0
∂xm
(1027)
since ( 1 − x2 ) only has two non-vanishing orders of derivatives.
On setting
∂m m
m
Pn (x) = ( 1 − x2 )− 2 Pnm (x) (1028)
∂x
and then differentiating, one obtains
∂ m+1 m ∂ mx m
Pn (x) = ( 1 − x2 )− 2 Pnm (x) + ( 1 − x2 )− 2 Pnm (x)
∂xm+1 ∂x 1 − x 2
(1029)
On differentiating a second time, one obtains
∂ m+2 m ∂2
Pn (x) = ( 1 − x2 ) − 2 P m (x)
∂xm+2 ∂x2 n
mx m ∂
+ 2 ( 1 − x2 ) − 2 P m (x)
1 − x 2 ∂x n
m ( m + 2 ) x2 m
+ 2 2
( 1 − x2 )− 2 Pnm (x)
(1 − x )
m m
+ 2
( 1 − x2 )− 2 Pnm (x)
(1 − x )
(1030)
These expressions are substituted into the m-th order differential relation, and
m
after cancelling the common factor of ( 1 − x2 )− 2 one has
∂2 m m2

∂ m
( 1 − x2 ) P (x) − 2 x P (x) + n ( n + 1 ) − Pnm (x) = 0
∂x2 n ∂x n 1 − x2
(1031)
Expressing the variable x in terms of the variable appropriate to the polar angle
x = cos θ, the Associated Legendre equation becomes
m2

1 ∂ ∂ m
sin θ Pn (cos θ) + n ( n + 1 ) − Pnm (x) = 0 (1032)
182
This equation reduces to Legendre’s differential equation when m = 0 and the
associated Legendre function of order m equal to zero is the Legendre polyno-
mial.
The Legendre polynomial are non zero for the range of n values such that
n ≥ m ≥ 0, but can also be defined for negative m values in which case
n ≥ m ≥ −n (1033)
and
(n − m)! m
Pn−m (x) = ( − 1 )m P (x) (1034)
(n + m)! n
since the differential equation only involves m2 and does not depend on the sign
of m.
The associated Legendre equation is often encountered in three-dimensional

situations which involve the Laplacian, in which the azimuthal dependence does
not vanish. In this case, one has
∂2ψ

1 ∂ 2 ∂ψ 1 ∂ ∂ψ 1
r + sin θ + 2 = − k2 ψ
r2 ∂r ∂r r2 sin θ ∂θ ∂θ r2 sin θ ∂ϕ2
(1035)
On using the ansatz for separation of variables
ψ(r, θ, ϕ) = R(r) Θ(θ) Φ(ϕ) (1036)
and diving by ψ one obtains
∂2Φ

1 ∂ 2 ∂R 1 ∂ ∂Θ 1
r + sin θ + = − k2
r2 R ∂r ∂r r2 sin θ Θ ∂θ ∂θ r2 sin2 θ Φ ∂ϕ2
(1037)
On multiplying by r2 sin2 θ, one recognizes that since all the ϕ dependence is
contained in the term
1 ∂2Φ
(1038)
Φ ∂ϕ2
this must be a constant. That is, the azimuthal dependence must satisfy
∂2Φ
= − m2 Φ (1039)
∂ϕ2
where m2 is an arbitrary constant, (not necessarily integer). The above equation

has solutions of the form

1
Φm (ϕ) = √ exp i m ϕ (1040)
2π
183
Since, with fixed (r, θ), the values of ϕ and ϕ + 2 π represent the same physical
point, one must have
Φm (ϕ) = Φm (ϕ + 2π) (1041)
which implies that m satisfies the condition

exp i m 2 π = 1 (1042)
and therefore m must be an integer. On substituting the constant back into the
original equation one obtains
m2

1 ∂ 2 ∂R 1 ∂ ∂Θ
r + sin θ − = − k 2 (1043)
r2 R ∂r ∂r r2 sin θ Θ ∂θ ∂θ r2 sin2 θ
This equation can be rewritten as
m2

1 ∂ 2 ∂R 2 2 1 ∂ ∂Θ
r + k r = − sin θ + (1044)
R ∂r ∂r sin θ Θ ∂θ ∂θ sin2 θ
The two sides of the equation depend on the different independent variables
(r, θ), and therefore must be constants, say, l ( l + 1 ). Then it can be seen
that the θ dependence is given by the Θm l (θ) which satisfy the equation
∂Θm m2

1 ∂
sin θ l
+ l(l + 1) − Θm
l (θ) = 0 (1045)
The above equation is related to the equation for the associated Legendre func-
tions. That is, the solution is given by
Θm m
l (θ) = Pl (cos θ) (1046)
and l must be integer.

The generating function expansion for the associated Legendre functions can be
obtained from the generating function expansion of the Legendre polynomials
∞
1 X
√ = Pn (x) tn (1047)
1 − 2 x t + t2 n=0
and the definition

m ∂m
Pnm (x) = ( 1 − x2 ) 2 Pn (x) (1048)
∂xm
184
m
By differentiating the recursion relation m times, and multiplying by ( 1 − x2 ) 2
one obtains
m ∞
(2m)! ( 1 − x2 ) 2 t m X
= Pnm (x) tn (1049)
2m m! ( 1 − 2 x t + t2 )m+ 21 n=0
However, the first m − 1 associated Legendre functions in the sum are zero.
Hence we have
m ∞
(2m)! ( 1 − x2 ) 2 X
= Pnm (x) tn−m
2 m! ( 1 − 2 x t + t2 )m+ 12
m
n=m
2 m ∞
(2m)! (1 − x ) 2 X
m
= Pm+s (x) ts (1050)
2m m! ( 1 − 2 x t + t2 )m+ 12 s=0

The associated Legendre functions also satisfy recursion relations. The recursion
relations become identical to the recursion relations of the Legendre polynomials
when m = 0. The recursion relations for the associated Legendre functions
can be derived from the recursion relations of the Legendre polynomials, by
differentiating with respect to x, m times. For example, consider the recursion
relation for the polynomial, Pn (x),
( 2 n + 1 ) x Pn (x) = ( n + 1 ) Pn+1 (x) + n Pn−1 (x) (1051)
and differentiate with respect to x m times
∂m ∂ m−1
(2n + 1)x Pn (x) + m ( 2 n + 1 ) Pn (x)
∂xm ∂xm−1
∂m ∂ m
= (n + 1) Pn+1 (x) + n Pn−1 (x) (1052)
∂xm ∂xm
On identifying the associated Legendre polynomial as
∂m m
Pn (x) = ( 1 − x2 )− 2 Pnm (x) (1053)
∂xm
the above relation reduces to
1
( 2 n + 1 ) x Pnm (x) + m ( 2 n + 1 ) ( 1 − x2 ) 2 Pnm−1 (x) = ( n + 1 ) Pn+1
m m
(x) + n Pn−1 (x)
(1054)
Likewise, on starting with

∂ ∂
( 2 n + 1 ) Pn (x) = Pn+1 (x) − Pn−1 (x) (1055)
∂x ∂x
185
and differentiating m times, one obtains
∂m ∂ m+1 ∂ m+1
(2n + 1) P n (x) = P n+1 (x) − Pn−1 (x)
∂xm ∂xm+1 ∂xm+1
(1056)
or
1
m+1 m+1
( 2 n + 1 ) ( 1 − x2 ) 2 Pnm (x) = Pn+1 (x) − Pn−1 (x) (1057)
If m is replaced by m − 1 in the above equation, it can be combined with

the previous recursion relation to yield the recursion relation
( 2 n + 1 ) x Pnm (x) = ( n + 1 − m ) Pn+1

m m
(x) + ( n + m ) Pn−1 (x) (1058)
The above recurrence relation for the Associated Legendre functions reduces to
the corresponding recursion relation for the Legendre polynomials when m = 0.
————————————————————————————————–
Example:
The above recursion relation has an important application in quantum me-

chanics. The probability for an electron in an atom to undergo a transition
between a state of orbital angular momentum (l, m) to an electronic state or-
bital angular momentum (l0 , m0 ) is proportional to the modulus squared of the
integrals
Z π
dθ sin θ Plm m
0 (cos θ) ( cos θ ) Pl (cos θ)
Z π0
dθ sin θ Plm+1
0 (cos θ) ( sin θ ) Plm (cos θ)
0
Z π
dθ sin θ Plm−1
0 (cos θ) ( sin θ ) Plm (cos θ) (1059)
0
for the case m0 = m and m0 = m ± 1 respectively. Basically, in the

dipole approximation, the electromagnetic field is represented by a complex
time-dependent vector
∝→−r exp i ω t (1060)
In a non-relativistic approximation one would expect that the electromagnetic

field would be represented by a vector, with three linearly independent basis
186
vectors, or three polarizations. The three components of this complex vector
are

r sin θ cos ϕ i ω t

r sin θ sin ϕ i ω t

r cos θ i ω t (1061)
The θ dependence of the vector is responsible for the factors in parenthesis

sandwiched between the Legendre functions. It can be seen, that the first pair
of these integrals can be expressed as
Z π
1 m m m
dθ sin θ Pl0 (cos θ) ( l + m ) Pl−1 (cos θ) + ( l − m + 1 ) Pl+1 (cos θ)
2l + 1 0
Z π
1 m+1 m+1 m+1
dθ sin θ Pl0 (cos θ) Pl+1 (cos θ) − Pl−1 (cos θ)
2l + 1 0
(1062)
by using the above two recursion relations. As the associated Legendre functions
are expected to be orthogonal, since they are eigenvalues of a Stürm-Liouville
equation, these integrals are only non-zero if l0 ± 1 = l. This condition is called
a selection rule. The vanishing of the integral means that the only transitions
which are allowed are those in which the magnitude of the electrons angular mo-
mentum is not conserved! The resolution of this apparent paradox is that the
electromagnetic field carries a unit of angular momentum. The photon, which
is the quantum particle representing an electromagnetic field has a quantized
angular momentum, or spin of magnitude 1 h̄. Likewise, a similar consideration
of the superscript, m, also yields a selection rule of m0 − m = ± 1 , or 0. The
index m corresponds to the z component of the angular momentum. The m
selection rule corresponds to the statement that the angular momentum of the
photon may be oriented in three directions, or have three polarizations. In the
non-relativistic limit, the three polarizations correspond to the three possible
values of m for which P1m (cos θ) is non-vanishing.
Problem:
Derive a recursion relation which will reduce the last expression to the sums
of integrals which only consist of the product of associated Legendre functions
with index m − 1, and the weighting factor sin θ.
187
9.1.4 Orthogonality
Associated Legendre functions with the same fixed m value are orthogonal, for
different values of n. This can easily be proved from the differential equation
m2

1 ∂ ∂
sin θ Pnm (cos θ) + n ( n + 1 ) − Pnm (cos θ) = 0
(1063)
by multiplying by Pnm0 (cos θ) and subtracting the analogous equation with n
and n0 interchanged. Since m is fixed, the terms involving m2 sin−2 θ identically
cancel, and on multiplying by the weight factor, one is left with

m ∂ ∂ m m ∂ ∂ m
Pn0 (cos θ) sin θ P (cos θ) − Pn (cos θ) sin θ P 0 (cos θ)
∂θ ∂θ n ∂θ ∂θ n

= n0 ( n0 + 1 ) − n ( n + 1 ) Pnm0 (cos θ) sin θ Pnm (cos θ)
(1064)
which on integrating over θ from 0 to π, one obtains

Z π
n0 ( n0 + 1 ) − n ( n + 1 ) dθ Pnm0 (cos θ) sin θ Pnm (cos θ) = 0 (1065)
0
since the boundary terms vanish identically, if the associated Legendre functions
and their derivatives are finite at the boundaries.
One finds that the associated Legendre functions have the normalization
integrals given by the value
Z π
2 (n + m)!
dθ Pnm0 (cos θ) sin θ Pnm (cos θ) = δn,n0 (1066)
0 2 n + 1 (n − m)!
Problem:
Evaluate the normalization integral for the case n = n0 .
12.5.16
Example:
One example of the occurrence of the associated Legendre functions occurs

→
−
in the vector potential A from a current loop.
188
Consider a circular current carrying loop of radius a in the equatorial plane
(θ = π2 ), in which the current is constant and has a value I. The magnetic
→
−
induction field, B is given by the solution of
→
− →
− 4π →
−
∇ ∧ B = j (1067)
c
→
−
where j is the current density. Together with the definition of the vector
potential
→
− →
− →
−
B = ∇ ∧ A (1068)
one has
→
− →
− →
− 4π → −
∇ ∧ ∇ ∧ A = j (1069)
c
→
−
Since, the problem has azimuthal symmetry B should be independent of ϕ
and have no component in the direction êϕ . The vector potential, therefore, is
entirely in the direction êϕ
→
−
A = Aϕ êϕ (1070)
Hence, one finds that the magnetic induction is given by

→
− 1 ∂ ∂ ∂

B = 2 (1071)
r sin θ ∂r ∂θ ∂ϕ
0 0 r sin θ Aϕ
Then, the magnetic induction is given by

" #
→
− 1 ∂ ∂
B = êr r sin θ Aϕ − êθ r r sin θ Aϕ
r2 sin θ ∂θ ∂r
(1072)
which is independent of ϕ.

∂ ∂ ∂

→
− →
− 1
∇ ∧ B = ∂r ∂θ ∂ϕ
2
r sin θ
1 ∂ ∂
r sin θ ∂θ sin θ Aϕ − ∂r r Aϕ 0

" #
1 ∂2

1 ∂ 1 ∂
= − êϕ r Aϕ + 2 sin θ Aϕ
r ∂r2 r ∂θ sin θ ∂θ
(1073)
Thus, we have the equation

" #
1 ∂2

1 ∂ 1 ∂ 4π
− r Aϕ + 2 sin θ Aϕ = jϕ (1074)
r ∂r2 r ∂θ sin θ ∂θ c
189
which off the θ = π2 plane becomes
" #
1 ∂2

1 ∂ 1 ∂
r Aϕ + 2 sin θ Aϕ = 0 (1075)
r ∂r2 r ∂θ sin θ ∂θ
On separating variables using

Aϕ = R(r) Θ(θ) (1076)
one has
∂2

r 1 ∂ 1 ∂
r R(r) = − sin θ Θ(θ) (1077)
R(r) ∂r2 Θ(θ) ∂θ sin θ ∂θ
On writing the separation constant as equal to n ( n + 1 ) one has
Rn (r) = An rn + Bn r−(n+1) (1078)
and the angular dependence is given by the solution of the equation

∂ 1 ∂
sin θ Θ(θ) + n ( n + 1 ) Θ(θ) = 0 (1079)
∂θ sin θ ∂θ
or

1 ∂ ∂ 1
sin θ Θ(θ) + n(n + 1) − Θ(θ) = 0 (1080)
Θ(θ) = Pn1 (cos θ) (1081)
Therefore, the solution is of the general form
X∞
Aϕ (r, θ) = An rn + Bn r−(n+1) Pn1 (cos θ) (1082)
n=0
The expansion coefficient An = 0 for r > a as the vector potential must fall
to zero as r → ∞.
The coefficients Bn are determined from the boundary condition on the z-

→
−
axis where it is easy to calculate B as
π 2π a2
Bz (r, ) = I 3 (1083)
2 c ( a2 + r2 ) 2
Using
" #
→
− 1 ∂ ∂
B = êr r sin θ Aϕ − êθ r r sin θ Aϕ
r2 sin θ ∂θ ∂r

cot θ 1 ∂Aϕ 1 ∂r Aϕ
= êr Aϕ + − êθ (1084)
r r ∂θ r r
190
one finds the general expression for the induction field. Since the derivative of
the associated Legendre function can be expressed as
∂ ∂
P 1 (cos θ) = − sin θ P 1 (cos θ) (1085)
∂θ n ∂ cos θ n
and on using the recursion relation
1 ∂ m 1 m+1 1
( 1 − x2 ) 2 P (x) = P (x) − (n+m) (n−m+1) Pnm−1 (x) (1086)
∂x n 2 n 2
and also

2mx
Pnm+1 (x) = √ Pnm (x) + m(m − 1) − n(n + 1) Pnm−1 (x)
1 − x2
(1087)
one has
mx 1 ∂ m
√ Pnm (x) − ( 1 − x2 ) 2 Pn (x) = ( n + m ) ( n − m + 1 ) Pnm−1 (x)
1 − x 2 ∂x
(1088)
On substituting m = 1 one finds that
∞
1 X
Br (r, θ) = An n ( n + 1 ) r−n Pn (cos θ) (1089)
r2 n=0
and
∞
1 X
Bθ (r, θ) = n An r−n Pn1 (cos θ) (1090)
r2 n=0
On substituting θ = 0 one has Pm (1) = 1 and Pn1 (1) = 0, thus the field on
the z-axis takes both the forms
∞
1 X
Br (r, 0) = An n ( n + 1 ) r−n
r2 n=0
2π a2
= I 3 (1091)
c ( a2 + r2 ) 2
From this one finds that, on expanding in powers of r−1 only the terms odd
in r−n and thus only the odd n coefficients (A2n+1 ) are finite and are uniquely
given by
2πI (2n)!
A2n+1 = ( − 1 )n (1092)
c 2(2n+1) n! (n + 1)!
This completes the solution for the vector potential.
191
9.2 Spherical Harmonics
The spherical harmonics are solutions for the angular part of Laplace’s equation.
They are composed of a product of an azimuthal function Φ(φ) and a polar part
Θ(θ) and are normalized to unity.
The azimuthal part is given by the solution of
∂2
Φ(ϕ) = − m2 Φ(ϕ) (1093)
∂ϕ2
which has solutions

1
Φm (ϕ) = √ exp imϕ (1094)
2π
The value of m is integer since the function Φm (ϕ) is periodic in ϕ with period
2 π.
The polar angle dependence is given by

s
m 2n + 1 (n − m)! m
Θn (θ) = P (cos θ) (1095)
2 n + m)! n
where, n > m > − n.
The spherical harmonics are defined by the product
Ynm (θ, ϕ) = ( − 1 ) n Θm (θ) Φm (ϕ)

sn
2n + 1 (n − m)! m 1
= ( − 1 )n Pn (cos θ) √ exp i m ϕ
2 (n + m)! 2π
s
n 2n + 1 (n − m)! m
= (−1) P (cos θ) exp i m ϕ (1096)
4 π (n + m)! n
The phase factor ( − 1 )n is purely conventional and is known as the Condon -

Shortley phase factor.
9.2.1 Expansion in Spherical Harmonics

Any function of the direction (θ, ϕ) can be expanded as a double series of the
spherical harmonics X
f (θ, ϕ) = Clm Ylm (θ, ϕ) (1097)
l,m
192
as the spherical harmonics form a complete set. Furthermore, the expansion
coefficients can be obtained from
Z π Z 2π
m
Cl = dθ dϕ sin θ Ylm (θ, ϕ)∗ f (θ, ϕ) (1098)
0 0
The completeness relation can be expressed as

l=∞
X m=l
X
δ( Ω1 − Ω2 ) = Ylm (θ1 , ϕ1 )∗ Ylm (θ2 , ϕ2 ) (1099)
l=0 m=−l
where
δ( Ω − Ω0 ) = δ( ϕ − ϕ0 ) δ( cos θ − cos θ0 ) (1100)
since this is equal to unity on integrating over all solid angles dΩ = dϕ dθ sin θ
and θ is uniquely specified by the value of cos θ for π > θ > 0.
Alternately one can re-write the completeness relation as

∞
X 2l + 1
δ( Ω1 − Ω2 ) = Pl (cos γ) (1101)
4π
l=0
where γ is the angle between (θ1 , ϕ1 ) and (θ2 , ϕ2 ). This can be proved by
choosing the spherical polar coordinate system such that (θ2 , ϕ2 ) is directed
along the polar axis. In this case θ2 = 0, and so sin θ2 = 0, therefore,
the only non-zero associated Legendre functions and hence non-zero spherical
harmonics, Ylm (θ2 , ϕ2 ), are those corresponding to m = 0. Then as we have
Pl0 (1) = 1, the completeness relation simply takes the form
l=∞
X 1 (2 l + 1 ) 0
δ( Ω1 ) = Pl (cos θ1 ) (1102)
2π 2
l=0
where cos θ1 = cos γ.
If these equivalent forms of the completeness relations can be identified for

each term in the sum over l we have the spherical harmonic addition theorem.
9.2.2 Addition Theorem

Given two vectors with directions (θ1 , ϕ1 ) and (θ2 , ϕ2 ) have an angle γ between
them. The angle is given by the vector product
cos γ = cos θ1 cos θ2 + sin θ1 sin θ2 cos(ϕ1 − ϕ2 ) (1103)
193
The addition theorem states
m=l
4π X
Pl (cos γ) = Ylm (θ1 , ϕ1 )∗ Ylm (θ2 , ϕ2 ) (1104)
2l + 1
m=−l
Example:
The multi-pole expansion of a potential φ(→−

r ) obtained for a charge distri-
→
− 0
bution ρ( r ), can be obtained from the generating function
∞
1 X r0l
→
− →
− 0 = Pl (cos θ) (1105)
| r − r | rl+1
l=0
for r0 < r. Using the addition theorem one has

1 X r0l 4π
→
− →
− 0 = Y m ∗ (θ0 , ϕ0 ) Ylm (θ, ϕ) (1106)
| r − r | rl+1 2 l + 1 l
l,m
Hence, one may write

4π Y m (θ, ϕ)
φ(→
−
X
r) = qlm l l+1 (1107)
2l + 1 r
l,m
where Z
qlm = d3 →
−
r 0 Ylm ∗ (θ0 , ϕ0 ) r0l ρ(→
−
r 0) (1108)
One has the symmetry

ql−m = ( − 1 )m qlm ∗ (1109)
The multipole expansion provides a nice separation of the potential into pieces,
each of which has a unique angular dependence and a dependence on r.
The multipole moments qlm characterize the charge distribution ρ(→

−
r 0 ). The
low order moments are given by
Z
1 q
0
q0 = √ d3 →
−
r 0 ρ(→
−
r 0) = √ (1110)
4π 4π
where q is the total charge. Also
r Z
3
1
q1 = − d3 →
−
r 0 r0 sin θ0 exp − i ϕ ρ(→
−
r 0)
8π
r Z
3 3→
−0
= − d r 0
x − iy 0
ρ(→
−
r 0)
8π
r
3
= − px − i py (1111)
8π
194
and
r Z
3
q10 = − d3 →
−
r 0 r0 cos θ0 ρ(→
−
r 0)
4π
r
3
= − pz (1112)
4π
where the dipole moment is given by
Z
→
−
p = d3 →
−
r0 →
−
r 0 ρ(→
−
r 0) (1113)
The next order contributions are given by

r Z
1 15
d3 →
− ρ(→
−
2
2
q2 = r 0 r0 sin2 θ0 exp − i 2 ϕ0 r 0)
4 2π
r Z 2
1 15 3→
−0
= d r 0
x − iy 0
ρ(→
−
r 0)
4 2π
r
1 15
= Q1,1 − 2 i Q1,2 + Q2,2 (1114)
12 2π
where we are introducing the notation which labels the (x, y, z) components
by (x1 , x2 , x3 ), and have introduced a tensor Qi,j win which the components
are labeled by the subscripts corresponding to (xi , xj ). The multi-pole moment
corresponding to l = 2 and m = 1 is given by
r Z
15
d3 →
−
r 0 r0 sin θ0 cos θ0 exp − i ϕ0 ρ(→ −
2
1
q2 = − r 0)
8π
r Z
15 3→−
= − d r 0 0
x − iy 0
z 0 ρ(→
−
r 0)
8π
r
1 15
= − Q1,3 − i Q2,3
3 8π
(1115)
and finally the l = 2 , m = 0 component is given by

r Z
1 5
d3 →
− 3 cos2 θ0 − 1 ρ(→ −
2
0
q2 = r 0 r0 r 0)
2 4π
r Z
1 5 3→−
= d r 0 02
3z − r 02
ρ(→
−
r 0)
2 4π
r
1 5
= Q3,3 (1116)
2 4π
195
in which the quadrupole moment is represented by the traceless tensor
Z
3→
−
Qi,j = d r 0 0 0
3 xi xj − r 02
ρ(→
−
r 0) (1117)
where →
−
r 0 is a position variable associated with the charge distribution.
In terms of the multi-pole moments, the potential at a position →

−
r far away
from the charge distribution is given by
q →
−
p .→ −
r xi xj
φ(→
−
X
r) = + + Qi,j + ... (1118)
r r3 i,j
r5
This series is expected to converge rapidly if the →

−
r is far away from the charges.
196
10 Hermite Polynomials
The generating function expansion for the Hermite Polynomials is given by
∞
tn
X
g(x, t) = exp − t2 + 2 x t = Hn (x) (1119)
n=0
n!

The generating function can be used to develop the recurrence relations
Hn+1 (x) = 2 x Hn (x) − 2 n Hn−1 (x) (1120)
and
∂
Hn (x) = 2 n Hn−1 (x) (1121)
∂x
10.0.4 Hermite’s Differential Equation

Also one can use the recurrence relations to establish the differential equation
for the Hermite polynomials. First on using
∂
Hn (x) = 2 n Hn−1 (x) (1122)
∂x
combined with
2 n Hn−1 (x) = 2 x Hn (x) − Hn+1 (x) (1123)
one has
∂
Hn (x) = 2 x Hn (x) − Hn+1 (x) (1124)
∂x
Differentiating this with respect to x yields
∂2 ∂Hn (x) ∂Hn+1 (x)
Hn (x) = 2 Hn (x) + 2 x − (1125)
∂x2 ∂x ∂x
Finally, on increasing the index by unity one has
∂
Hn+1 (x) = 2 ( n + 1 ) Hn (x) (1126)
∂x
which can be used to eliminate the derivative of Hn+1 (x). This leads to the
Hermite’s differential equation
∂2 ∂
Hn (x) − 2 x Hn (x) + 2 n Hn (x) = 0 (1127)
∂x2 ∂x
197
This is not of the form of a Stürm-Liouville equation, unless one introduces a
weighting function
w(x) = exp − x2 (1128)
This is found by examining the ratio’s of the first two coefficients in the Stürm-
Liouville form
∂2φ ∂p(x) ∂φ
p(x) + + q(x) φ(x) = λ w(x) φ(x) (1129)
∂x2 ∂x ∂x
in which case one identifies the ratio
∂p(x)
∂x
= −2x (1130)
p(x)
or on integrating
ln p(x) = − x2 (1131)
Since the integrating factor is

2
p(x) = exp − x (1132)
one finds the above weighting factor and the differential equation has the form

∂ ∂
exp − x2 Hn (x) + 2 n exp − x2 Hn (x) = 0 (1133)
∂x ∂x
10.0.5 Orthogonality
The Hermite polynomials form an orthogonal set
Z ∞
2
dx Hn (x) exp − x Hm (x) = 0 (1134)
−∞
if n 6= m. For n = m, one can obtain the normalization by use of the gener-

ating function. Integrating the product of two generating functions multiplied
by the weight function
Z ∞
dx g(x, s) exp − x2 g(x, t)
−∞
∞ X
∞ Z ∞
sn tm
X
= dx exp − x2 Hn (x) Hm (x)
n=0 m=0 −∞ n! m!
∞ ∞
( s t )n
X Z
= dx exp − x2 Hn (x) Hn (x) (1135)
n=0 −∞ (n!)2
198
where we have used the orthogonality relation. However, one also has
Z ∞
dx g(x, s) exp − x2 g(x, t)
−∞
Z ∞
= dx exp − x2 exp − t2 + 2 x t exp − s2 + 2 x s
−∞
Z ∞
= dx exp − x2 + 2 x ( s + t ) exp − t2 − s2
−∞
√

= π exp ( s + t )2 exp − s2 − t2
√

= π exp 2 s t (1136)
On expanding in powers of ( s t ) one finds

Z ∞ ∞
√ X ( 2 s t )n

dx g(x, s) exp − x2 g(x, t) = π
−∞ n=0
n!
(1137)
Hence on equating the coefficients of the powers of ( s t ), one has the equality
Z ∞
√

dx exp − x2 Hn (x) Hn (x) = π 2n n! (1138)
−∞
which gives the desired normalization.
Example:
The one-dimensional quantum mechanical harmonic oscillator is a state with

energy E is governed by the equation
h̄2 ∂ 2 ψ 1
− + m ω 2 x2 ψ = E ψ (1139)
2 m ∂x2 2
The asymptotic large x behavior is found as

mω 2
lim ψ(x) → exp − x (1140)
x → ∞ 2 h̄
and so we seek a solution of the form

mω 2
ψ(x) = exp − x φ(x) (1141)
2 h̄
From this, we we find that the first-order derivative is given by

∂ψ mω 2 mω ∂
= exp − x − x + φ(x) (1142)
∂x 2 h̄ h̄ ∂x
199
and the second-order derivative is
2
∂2ψ ∂2

mω 2 mω 2 mω ∂
= exp − x x − (1 + 2 x )+ φ(x)
∂x2 2 h̄ h̄ h̄ ∂x ∂x2
(1143)
Inserting this last expression into the eigenvalue equation, we find that φ satisfies
the differential equation
∂2φ mω ∂φ m ( 2 E − h̄ ω )
− 2 x = − φ (1144)
∂x2 h̄ ∂x h̄2
This differential equation can be put into Hermite’s form by introducing a di-
mensionless variable r
h̄
z = x (1145)
mω
Thus, we have
∂2φ

∂φ 2E
− 2z + − 1 φ (1146)
∂z 2 ∂z h̄ ω
which has the Hermite polynomial Hn (z) as a solution when

2E
2n = − 1 (1147)
h̄ ω
Therefore, the allowed values of the energy are given by
1
E = h̄ ω ( n + ) (1148)
2
Problem:
In quantum mechanics one encounters the Hermite polynomials in the con-

text of one-dimensional harmonic oscillators. If a system in the n-th state is
perturbed by a potential V (x), the probability of a transition from the n-th
state to the m-th state is proportional to
Z ∞ 2
1 2

dx exp − x Hn (x) V (x) Hm (x) (1149)
π 2n+m n! m! −∞
Evaluate the integrals
Z ∞
2
dx exp − x x Hn (x) Hm (x) (1150)
−∞
and Z ∞
2
dx exp − x x2 Hn (x) Hm (x) (1151)
−∞
200
11 Laguerre Polynomials
The generating function expansion for the Laguerre polynomials is given by

exp − 1 x−t t X∞
g(x, t) = = Ln (x) tn (1152)
1 − t n=0
From which one obtains the series expansion as

n
X n!
Ln (x) = ( − 1 )n−s xn−s (1153)
s=0
(n − s)!2 s!

The recursion relations for the Laguerre polynomials can be obtained from the
generating function expansion. By differentiating the generating function with
respect to t one obtains the
∞
(1 − x − t) X
g(x, t) = Ln (x) n tn−1 (1154)
( 1 − t )2 n=0
and hence on multiplying by ( 1 − t )2 to obtain the recurrence relation

( n + 1 ) Ln+1 (x) + n Ln−1 (x) = ( 2 n + 1 − x ) Ln (x) (1155)
Also, on differentiating with respect to x and then multiplying by ( 1 − t ) one
obtains
∞
X ∂
− t g(x, t) = ( 1 − t ) Ln (x) tn (1156)
n=0
∂x
and hence the recursion relation is found as
∂ ∂
− Ln (x) = Ln+1 (x) − Ln (x) (1157)
∂x ∂x
Differentiating the first recursion relation yields
∂ ∂ ∂
(n + 1) Ln+1 (x) + n Ln−1 (x) = ( 2 n + 1 − x ) Ln (x) − Ln (x)
∂x ∂x ∂x
(1158)
The previous relation can be used ( twice, once with index n and the second time
with index n − 1 ) to eliminate the differential of Ln+1 (x) and the differential
of Ln−1 (x), giving the recursion relation
∂
x Ln (x) = n Ln (x) − n Ln−1 (x) (1159)
∂x
201
11.0.7 Laguerre’s Differential Equation
The Laguerre polynomials satisfy Laguerres differential equation. The differ-
ential equation for the Laguerre polynomials can be derived by combining the
recursion relations. This is done by differentiating the prior relation
∂2 ∂ ∂ ∂
x 2
Ln (x) + Ln (x) = n Ln (x) − n Ln−1 (x) (1160)
∂x ∂x ∂x ∂x
and subtracting the previous equation one has
∂2 ∂ ∂ ∂
x Ln (x) + ( 1 − x ) Ln (x) = n Ln (x) − n Ln−1 (x) − n Ln (x) + n Ln−1 (x)
∂x2 ∂x ∂x ∂x
(1161)
Finally, the right-hand-side can be identified as − n Ln (x). Thus, we have
Laguerre’s differential equation
∂2 ∂
x 2
Ln (x) + ( 1 − x ) Ln (x) + n Ln (x) = 0 (1162)
∂x ∂x
This equation is not in Stürm-Liouville form, but can be put in the form by
multiplying by exp[ − x ]. Hence, we obtain

∂ ∂
x exp[ − x ] Ln (x) + n exp[ − x ] Ln (x) = 0 (1163)
∂x ∂x
The solution is defined on the interval (0, ∞). Therefore, the Laguerre functions
are orthogonal with weighting factor exp[ − x ].
11.1 Associated Laguerre Polynomials

The associated Laguerre polynomials are defined by
∂p
Lpn (x) = ( − 1 )p Ln+p (x) (1164)
∂xp

A generating function can be obtained by differentiating the generating func-
tion for the Laguerre polynomials p times. This yields the generating function
expansion
x t
exp − 1 − t ∞
X
= Lpn (x) tn (1165)
( 1 − t )p+1 n=0
A pair of recursion relations can be derived as
( n + 1 ) Lpn+1 (x) = ( 2 n + p + 1 − x ) Lpn (x) − ( n + p ) Lpn−1 (x) (1166)
202
and
∂ p
x L (x) = n Lpn (x) − ( n + p ) Lpn−1 (x) (1167)
∂x n
From these, or by differentiating Laguerre’s differential equation p times, one
finds
∂2 p ∂ p
x Ln (x) + ( p + 1 − x ) L (x) + n Lpn (x) = 0 (1168)
∂x2 ∂x n
The weighting function is found to be
xp exp[ − x ] (1169)
The orthogonality condition is found to be

Z ∞
(n + p)!
dx xp exp[ − x ] Lpn (x) Lpm (x) = δm,n (1170)
0 n!
Example:
The associated Laguerre polynomials occur in the solution of the Schrödinger

equation for the hydrogen atom. The Schrödinger equation for the electron wave
function has the form
h̄2 Z e2
− ∇2 ψ − ψ = Eψ (1171)
2m r
which involves the spherically symmetric Coulomb potential of the nucleus of
charge Z e acting on the electrons. On separation of variables, and finding that
due to the spherical symmetry, the angular dependence is given by the spherical
harmonic Ylm (θ, ϕ), then the remaining radial function R(r) is given by
h̄2 1 ∂ h̄2 l ( l + 1 ) Z e2

2 ∂R
− 2
r + 2
R − R = E R (1172)
2 m r ∂r ∂r 2m r r
where the second term is the centrifugal potential. This equation can be put
in a dimensionless form by defining a characteristic length scale r0 by equating
the centrifugal potential with the Coulomb attraction. (This just corresponds
to the radius of the circular orbit in Classical Mechanics.) The radius is found
as
h̄2
r0 = (1173)
2 m Z e2
Then, introducing a dimensionless radius ρ as the ratio of the radius to that of
the circular orbit
r
ρ = (1174)
r0
203
one sees that the on expressing the Laplacian in dimensionless form the energy
term is given by the following constant
2 m E r02 E h̄2 1
2 = 2 4
= − (1175)
h̄ 2mZ e 4 α2
which must also be dimensionless. The second equality was found by substitut-
ing the expression for r0 . The dimensionless constant must be negative, since
an electron that is bound to the hydrogen atom does not have enough energy
to escape to infinity. The minimum energy required to escape to infinity is zero
energy, since the potential falls to zero at infinity and the minimum of the ki-
netic energy is also zero. That is the minimum energy E is that in which the
electron comes to rest at infinity. Thus, since E is less than this minimum it is
negative. In terms of the variable, ρ and the constant α one has the differential
equation

1 ∂ 2 ∂R(ρ r0 ) 1 l(l + 1) 1
ρ + − − R(ρ r0 ) = 0 (1176)
ρ2 ∂ρ ∂ρ ρ ρ2 4 α2
The form of the solution can be found by examining the equation near ρ →
0. In this case, the centrifugal potential is much larger than the Coulomb
potential so one can neglect the Coulomb potential. Also, for the same reason,
the energy constant 4 1α2 is also negligible compared with the centrifugal barrier.
The equation reduces to

1 ∂ 2 ∂R(ρ r0 ) l(l + 1)
ρ − R(ρ r0 ) = 0 (1177)
ρ2 ∂ρ ∂ρ ρ2
near the origin, and has a solution
R(r) ∝ ρl (1178)
since the other solution is proportional to ρ−(l+1) which diverges at the origin.
It is not acceptable to have a solution that diverges too badly at the origin.
The form of the equation at ρ → ∞ simplifies to

1 ∂ 2 ∂R(ρ r0 ) 1
2
ρ − R(ρ r0 ) = 0 (1179)
ρ ∂ρ ∂ρ 4 α2
which can be written as

∂ 2 R(ρ r0 ) 2 ∂R(ρ r0 ) 1
+ − R(ρ r0 ) = 0 (1180)
∂ρ2 ρ ∂ρ 4 α2
204
which for large ρ simply becomes
∂ 2 R(ρ r0 ) 1
− R(ρ r0 ) = 0 (1181)
∂ρ2 4 α2

ρ
R(ρ r0 ) ∝ exp − (1182)
2α
as the other solution which diverges exponentially as

ρ
R(ρ r0 ) ∝ exp + (1183)
2α
is physically unacceptable for an electron that is bound to the nucleus at a dis-
tance r0 . Thus, we have found that electrons are bound at a distance of 2 α r0
from the origin.
This motivates one looking for a solution with the right forms at the origin
and at infinity. This form can be expressed as

ρ ρ
R(ρ r0 ) = exp − ρl L( ) (1184)
2α α
To substitute this form into the differential equation, one needs to evaluate the
first and second order derivatives of the form. First note that

∂R(ρ r0 ) ρ l 1 l ρ 1 0 ρ
= exp − ρ − + L( ) + L ( ) (1185)
∂ρ 2α 2α ρ α α α
and then note that

∂ ∂R(ρ r0 ) ρ
ρ2 = exp − ρl+2 ×
∂ρ ∂ρ 2α

1 (l + 1) l(l + 1) ρ
× 2
− + 2
L( )
4α αρ ρ α

1 2l + 2 ρ 1 ρ
+ − 2 + L0 ( ) + 2 L00 ( ) (1186)
α ρα α α α
Thus, one finds that L satisfies the equation

ρ 00 ρ 1 0 (l + 1)
L + ( 2 l + 2 − ) L + 1 − L = 0 (1187)
α2 α α α
or

ρ 00 ρ
L + (2l + 2 − ) L0 + α − (l + 1) L = 0 (1188)
α α
205
(2l+1)
Hence L(x) is the associated Laguerre polynomial Lα − l − 1 (x), when α is an
integer. If the function is to be normalizable, it is necessary for the series to
terminate, and hence α − l − 1 must be an integer. It is usual to set n = α,
and have the condition n ≥ l + 1. The radial wave function is given by

l r (2l+1) r
R(r) ∝ r exp − L(n−l−1) (1189)
2 n r0 n r0
Example:
The energy eigenvalue equation for a three-dimensional harmonic oscillator

is given by
h̄2 1
− ∇2 ψ + m ω 2 r2 ψ = E ψ (1190)
2m 2
On using separation of variables
ψ = R(r) Ylm (θ, ϕ) (1191)
one has
h̄2 1 ∂ h̄2 l ( l + 1 ) m ω 2 r2

2 ∂R
− r R + R + R = ER
2 m r2 ∂r ∂r 2 m r2 2
(1192)
On writing
l ρ
R(r) = ρ 2 exp − L(ρ) (1193)
2
with a dimensionless variable
mω 2
ρ = r (1194)
h̄
one has
r
∂R(r) m ω l+1 ρ l 1 ∂
= 2 ρ 2 exp − − + L (1195)
∂r h̄ 2 2ρ 2 ∂ρ
and the second derivative is given by
∂ 2 R(r)

m ω l+2 ρ l(l − 1) 2l + 1 1
= 4 ρ 2 exp − − +
∂r2 h̄ 2 4 ρ2 4ρ 4
2

2l + 1 ∂ ∂
+ − 1 + L (1196)
2ρ ∂ρ ∂ρ2
206
This leads to the equation
∂2L

2l + 3 ∂L E 2l + 3
−ρ − − ρ = − L (1197)
∂ρ2 2 ∂ρ 2 h̄ ω 4
which is the differential equation for the Associated Laguerre polynomials in

which p = 2 l 2+ 1 and the quantum number n is given by

E 2l + 3
n = − (1198)
2 h̄ ω 4
Problem:
A quantum mechanical analysis of the Stark effect in parabolic coordinates

leads to the differential equation
m2

∂ ∂u 1 1
x + Ex + l − − F x2 u = 0 (1199)
∂x ∂x 2 4x 4
where F is a measure of the strength of the electric field. Find the unperturbed
wave function in terms of the Associated Laguerre polynomials.
207
12 Inhomogeneous Equations
12.1 Inhomogeneous Differential Equations
Inhomogeneous linear differential equations can be solved by Green’s function
methods. For example, if one has an equation of the type

∂ ∂φ
p(x) + ( q(x) − λ w(x) ) φ(x) = f (x) (1200)
∂x ∂x
where λ is a control parameter, and the solution is defined on an interval (a, b)

and which satisfies boundary conditions at the ends of the interval, say of the
type
φ(a) = φ(b) = 0 (1201)
The solution of this equation can be found using Green’s functions. The Green’s
function G(x, x0 ) is defined as the solution of the equation
∂G(x, x0 )

∂
p(x) + ( q(x) − λ w(x) ) G(x, x0 ) = δ(x − x0 ) (1202)
∂x ∂x
in which the inhomogeneous term is replaced by a delta function which is non-

zero at the value of x given by x = x0 . The Green’s function must also satisfy
the same boundary conditions as φ , which in this case is
G(a, x0 ) = G(b, x0 ) = 0 (1203)
Since we can write the inhomogeneous term of our original equation in the form
of an integral
Z b
f (x) = dx0 δ(x − x0 ) f (x0 ) (1204)
a
then we can view the original inhomogeneous equation as a linear superposition
of equations with δ function source terms, but in which the various terms are
weighted with a factor of f (x0 ). That is, the solution can be expressed as
Z b
φ(x) = dx0 G(x, x0 ) f (x0 ) (1205)
a
This expression satisfies the equation for φ(x), as can be seen by direct substi-
tution. This means that once the Green’s function has been found, the solution
of the inhomogeneous equation, for any reasonable f (x), can be found by inte-
gration.
208
12.1.1 Eigenfunction Expansion
One method of finding the Green’s function is based on the completeness re-
lation for the eigenvalues of the Stürm-Liouville equation in which the control
parameter λ has been set to zero
X
δ(x − x0 ) = φ∗n (x0 ) w(x) φn (x) (1206)
n
Due to the completeness relation one can find an expansion of the Green’s
function in the form
X
G(x, x0 ) = Gn (x0 ) φn (x) (1207)
n
0
where Gn (x ) is an unknown coefficient. As φn (x) satisfies the same boundary
conditions as G(x, x0 ), this expansion is satisfies the boundary conditions. On
substituting these two expansions in the equation one obtains
X X
( λn − λ ) Gn (x0 ) w(x) φn (x) = φ∗n (x0 ) w(x) φn (x) (1208)
n n
where we have used the Stürm-Liouville eigenvalue equation. On multiplying

by φ∗m (x) and integrating with respect to x and using the orthogonality of the
eigenfunctions (where in the cases of degeneracies the eigenfunctions have been
constructed via the Gram Schmidt process), one finds that
( λm − λ ) Gm (x0 ) = φ∗m (x0 ) (1209)
Therefore, if the control parameter is chosen such that λn 6= λ for all n, we have
found that the Green’s function is given in terms of a sum of the eigenfunctions
X φ∗ (x0 ) φn (x)
G(x, x0 ) = n
(1210)
n
λn − λ
12.1.2 Piece-wise Continuous Solution

Alternatively, for one-dimensional problems the Green’s function which satisfies
∂G(x, x0 )

∂
p(x) + ( q(x) − λ w(x) ) G(x, x0 ) = δ(x − x0 ) (1211)
∂x ∂x
can be obtained from knowledge of the solutions of the homogeneous equation

∂ ∂φ(x)
p(x) + ( q(x) − λ w(x) ) φ(x) = 0 (1212)
∂x ∂x
209
utilizing the arbitrary constants of integration. In one-dimension, the point
x0 separates the interval into two disjoint regions. The solution of the Green’s
function in the two regions b x > x0 and x0 > x > a coincide with the solution
of the homogeneous equation with the appropriate boundary conditions. Then
in the first region, b > x > x0 , one can have
G(x, x0 ) = C1 φ1 (x) (1213)
which satisfies the boundary condition at x = b. In the second region one can
have x0 > x > a
G(x, x0 ) = C2 φ2 (x) (1214)
where φ2 (x) satisfies the boundary condition at x = a. The arbitrary constants
C1 and C2 can be obtained from consideration of the other boundaries of the two
intervals, that is the point x = x0 . The Green’s function must be continuous
at x = x0 , which requires that
C1 φ1 (x0 ) = C2 φ2 (x0 ) (1215)
Furthermore, the Green’s function must also satisfy the equation at x = x0 . On

integrating the differential equation for the Green’s function from x = x0 −
and x = x0 + and taking the limit → 0
0 0

0 ∂G(x, x ) 0 ∂G(x, x )
p(x ) − p(x ) = 1 (1216)
∂x
x=x0 + ∂x
x=x0 −
which leads to

∂φ1 (x) ∂φ2 (x) 1
C1 − C 2 = (1217)
∂x x=x0 ∂x x=x0 p(x0 )
The above pair of equations for the constants C1 and C2 has a solution if the
Wronskian determinant is non-zero
φ1 (x0 ) φ2 (x0 )

W (φ1 , φ2 ) = ∂φ1 (x0 ) ∂φ2 (x0 )
(1218)
∂x ∂x
However, the Wronskian is given by

C
W (φ1 , φ2 ) = (1219)
p(x0 )
where C is yet another constant. From linear algebra one has

∂φ2
− φ2 (x0 )

∂x |x
0
∂φ1
∂x |x − φ1 (x0 )

C1 0 0
= 1 (1220)
C2 W (φ1 , φ2 ) p(x0 )
210
So the constants C1 and C2 are found as
φ2 (x0 )
C1 = −
C
φ1 (x0 )
C2 = −
C
(1221)
and so the Green’s function is given by
1
G(x, x0 ) = − φ1 (x) φ2 (x0 ) (1222)
C
for b > x > x0 and
1
G(x, x0 ) = − φ1 (x0 ) φ2 (x) (1223)
C
for x0 > x > a . Therefore, the Green’s function is symmetric under the
interchange of x and x0 .
The solution of the inhomogeneous equation is found as

Z b
φ(x) = dx0 G(x, x0 ) f (x0 )
a
Z x Z b
1 0 0 1 0
= − dx φ1 (x) φ2 (x ) f (x ) − dx0 φ1 (x0 ) φ2 (x) f (x0 )
C a C x
Z x Z b
1 1
= − φ1 (x) dx0 φ2 (x0 ) f (x0 ) − φ2 (x) dx0 φ1 (x0 ) f (x0 )
C a C x
(1224)
Example:
Show that the Green’s function for the equation

∂2
φ(x) = f (x) (1225)
∂x2
φ(0) = 0

∂
φ(x) = 0 (1226)
∂x x=1
is given by
G(x, t) = −x for 0 ≤ x < t
G(x, t) = −t for t < x ≤ 1 (1227)
211
This is solved by noting that for x 6= x0 the equation for the Green’s
function is given by the solution of
∂ 2 G(x, x0 )
= 0 (1228)
∂x2
This has a general solution
G(x, x0 ) = a x + b (1229)
The Green’s function in the region x0 > x ≥ 0 must satisfy the boundary
condition at x = 0. Thus, the Green’s function has to satisfy the boundary
condition
G< (0, x0 ) = a< 0 + b< = 0 (1230)
Hence, we have b< = 0 so the Green’s function in this region is given by
G< (x, x0 ) = a< x (1231)
In the second region, 1 ≥ x > x0 , we have the boundary condition
∂G> (x, x0 )

∂x = 0 (1232)
1
which leads to
∂G> (x, x0 )

∂x = a> = 0 (1233)
1
G> (x, x0 ) = b> (1234)

Continuity of the Green’s function at x = x0 yields
a< x0 = b> (1235)
Furthermore, on integrating the differential equation between the limits (x0 −

, x0 + ) we have
x0 + Z x0 +
∂ 2 G(x, x0 )
Z
dx = δ( x − x0 )
x0 − ∂x2 0
x −
0
0 x +
∂G(x, x )
= 1 (1236)
∂x 0
x −
This leads to the second condition
0 − a< = 1 (1237)
212
Hence, we have
G< (x, x0 ) = −x
0
G> (x, x ) = − x0 (1238)
Example:
Find the solution of the equation

∂2
φ(x) + φ(x) = f (x) (1239)
∂x2
φ(0) = 0

∂
φ(x) = 0 (1240)
∂x x=1
The Green’s function can be obtained from knowledge of the solutions of the
homogeneous equation. The general solution of the homogeneous equation is
given by
G(x, x0 ) = a sin x + b cos x (1241)
In the first region ( < ) where x0 > x > 0 the Green’s function has to satisfy
the boundary condition
G< (0, x0 ) = a< sin 0 + b< cos 0 = 0 (1242)
Hence,
G< (x, x0 ) = a< sin x (1243)
0
In the second region ( > ) where 1 > x > x we have
∂G> (x, x0 )

= a> cos 1 − b> sin 1 = 0 (1244)
∂x
x=1
Continuity, leads to the condition

G< (x, x0 ) = G> (x, x0 )
a< sin x0 = a> sin x0 + b> cos x0 (1245)
and since G(x, x0 ) satisfies the differential equation at x = x0 one finds on
integrating the differential equation
∂G> (x, x0 ) ∂G< (x, x0 )

∂x 0 − ∂x 0 = 1
x x
a< cos x0 − a> cos x0 + b> sin x0 = 1 (1246)
213
Eliminating a> , using the relation a> = b> tan 1, one obtains the pair of
equations
a< sin x0 cos 1 = b> cos(x0 − 1)

a< cos x0 cos 1 + b> sin(x0 − 1) = cos 1
(1247)
On solving these one finds the coefficients as

cos(x0 − 1)
a< =
cos 1
b> = sin x0
sin x0 sin 1
a> = (1248)
cos 1
Hence we have the Green’s function as
sin x cos(x0 − 1)
G< (x, x0 ) =
cos 1
sin x0 cos(x − 1)
G> (x, x0 ) = (1249)
cos 1
The solution of the inhomogeneous differential equation is then given by
Z 1 Z x
φ(x) = dx0 G< (x, x0 ) f (x0 ) + dx0 G> (x, x0 ) f (x0 ) (1250)
x 0
Example:
Find the Green’s function G(x, x0 ) that satisfies the equation
∂G(x, x0 )

∂
x = δ( x − x0 ) (1251)
∂x ∂x
subject to the boundary conditions | G(0, x0 ) | < ∞ and G(1, x0 ) = 0.
The homogeneous equation has the solution
G(x, x0 ) = a ln x + b (1252)
The solution has to satisfy the boundary conditions
G< (0, x0 ) = a< ln 0 + b< 6= ± ∞ (1253)
and
G> (1, x0 ) = a> ln 1 + b> = 0 (1254)
214
The continuity condition leads to
G> (x0 , x0 ) = G< (x0 , x0 ) (1255)
or
a> ln x0 = b< (1256)
0
On integrating over a region of infinitesimal width around x , yields
Z x0 + Z x0 +
∂G(x, x0 )

∂
dx x = dx δ( x − x0 )
x0 − ∂x ∂x x0 −
∂G> (x, x0 ) ∂G< (x, x0 )

0
x
∂x 0 − ∂x 0 = 1 (1257)
x x
Thus, a> = 1 and from the continuity condition one has b< = ln x0 . Hence,
we find
G< (x, x0 ) = ln x0
G> (x, x0 ) = ln x (1258)
Example:
Find the series solution for the Green’s function which satisfies the equation
∂ 2 G(x, x0 )
= δ( x − x0 ) (1259)
∂x2
with boundary conditions
G(0, x0 ) = G(1, x0 ) = 0 (1260)
The eigenvalue equation
∂ 2 φ(x)
= λ φ(x) (1261)
∂x2
satisfying the same boundary conditions has eigenfunctions φn (x) given by
√
φn (x) = 2 sin nπx (1262)
and eigenvalue λ = − π 2 n2 . On writing the Green’s function as a series
expansion in terms of a complete set of functions
∞
X
G(x, x0 ) = Gn (x0 ) φn (x) (1263)
n=1
then the expansion coefficients can be found by substitution into the differential
equation
X∞
− π 2 n2 Gn (x0 ) φn (x) = δ( x − x0 ) (1264)
n=1
215
On multiplying by φm (x) and integrating over x we obtain
φm (x0 )
Gm (x0 ) = − (1265)
π 2 m2
Thus, we have
∞
X sin nπx sin nπx0
G(x, x0 ) = − 2 (1266)
n=1
π 2 n2
This can be compared with the expression
G< (x, x0 ) = x ( x0 − 1 )
G> (x, x0 ) = ( x − 1 ) x0 (1267)
————————————————————————————————–
Example:
Consider a bowed stretched string. The bowing force is assumed to be trans-

verse to the string, and has a force per unit length, at position x given by f (x, t).
Then, the displacement of the string, u(x, t) is governed by the partial differen-
tial equation
∂2u 1 ∂2u
2
− 2 = f (x, t) (1268)
∂x c ∂t2
We assume a sinusoidal bowing force

f (x, t) = f (x) Real exp − i ω t (1269)
The forced response is given by

u(x, t) = u(x) Real exp − iωt (1270)
where u(x) is determined by

∂2u ω2
2
+ 2 u(x) = f (x) (1271)
∂x c
The string is fixed at the two ends, so
u(0) = u(l) = 0 (1272)
The Green’s function G(x, x0 ) satisfies the equation

∂ 2 G(x, x0 ) ω2
+ G(x, x0 ) = δ(x − x0 ) (1273)
∂x2 c2
216
along with the boundary conditions
G(0, x0 ) = G(l, x0 ) = 0 (1274)
The solution can be found as

x < x0
 
 A sin k x 
G(x, x0 ) = (1275)
B sin k ( x − l ) x > x0
 
where k = ωc . On integrating the differential equation over an infinitesimal

region around the delta function yields
x0 +
∂ 0

G(x, x ) = 1 (1276)
∂x x0 −
Thus,
k B cos k ( x0 − l ) − k A cos k x0 = 0 (1277)
0
Continuity of the Green’s function at x = x yields
G(x0 + , x0 ) = G(x0 − , x0 ) (1278)
or
B sin k ( x0 − l ) = A sin k x0 (1279)
This, leads to the determination of A and B as
sin k ( x0 − l )
A =
k sin k l
sin k x0
B = (1280)
k sin k l
Hence, the Green’s function is given by
sin k x< sin k ( x> − l )
G(x, x0 ) = (1281)
k sin k l
which is symmetric in x and x0 . The solution of the forced equation is given by
Z l
u(x) = dx0 G(x, x0 ) f (x0 )
0
Z x
sin k ( x − l )
= dx0 sin k x0 f (x0 )
k sin k l 0
Z l
sin k x
+ dx0 sin k ( x0 − l ) f (x0 ) (1282)
k sin k l x
————————————————————————————————–
217
12.2 Inhomogeneous Partial Differential Equations
The Greens’ function method can also be used for inhomogeneous partial dif-
ferential equations. Examples of inhomogeneous partial differential equations
which often occur in physics are given by
∇2 φ(→−r) = − 4 π ρ(→
−
r)
2 →−
1 ∂ φ( r , t)
∇2 φ(→
−
r , t) − 2 = − 4 π ρ(→
−
r , t)
c ∂t2
→
−
1 ∂φ( r , t)
∇2 φ(→
−r , t) − = − 4 π ρ(→
−
r , t) (1283)
κ ∂t
subject to appropriate boundary conditions. These three equations can be
solved, for arbitrary source terms ρ, from knowledge of the respective Green’s
functions which satisfy the differential equations
∇2 G(→ −r ,→− r 0 ) = δ 3 (→
−
r −→
−
r 0)
1 ∂ 2 G(→
−r , t; →
−r 0 , t0 )
∇2 G(→
−
r , t; →
−
r 0 , t0 ) − 2 = δ 3 (→
−
r −→
−
r 0 ) δ(t − t0 )
c ∂t2
1 ∂G(→ −r , t; →
−r 0 , t0 )
∇2 G(→
−r , t; →
−
r 0 , t0 ) − = δ 3 (→
−
r −→
−
r 0 ) δ(t − t0 )
κ ∂t
(1284)
with the same boundary conditions. In these equations we have introduced a
three-dimensional delta function. These are defined via
Z
d3 →
−
r δ 3 (→
−
r −→
−
r 0 ) f (→
−
r ) = f (→
−
r 0) (1285)
The three-dimensional delta function includes the weighting function appropri-

ate for the coordinate systems. In Cartesian coordinates, one has
δ 3 (→
−
r −→
−
r 0 ) = δ(x − x0 ) δ(y − y 0 ) δ(z − z 0 ) (1286)
while in spherical polar coordinates
δ(r − r0 ) δ(θ − θ0 )
δ 3 (→
−
r −→
−
r 0) = δ(ϕ − ϕ0 ) (1287)
r2 sin θ
It can be seen that on integrating the delta function over all three-dimensional,
the denominator cancels with the weight factors giving a result of unity. On
finding the time-dependent Green’s function one obtains the solution of the
inhomogeneous equation as
Z ∞ Z
φ(→
−
r , t) = − 4 π dt0 d3 →
−
r 0 G(→
−
r , t; →
−
r 0 , t0 ) ρ(→
−
r 0 , t0 ) (1288)
−∞
which is valid for any arbitrary source, ρ(→

−
r , t).
218
12.2.1 The Symmetry of the Green’s Function.
The Green’s function is symmetric in its arguments (→
−
r ,→
−
r 0 ), such that
G(→
−
r , t; →
−
r 0 , t0 ) = G(→
−
r 0 , t0 ; →
−
r , t) (1289)
This can be proved by examining the Stürm-Liouville Green’s function equation

with a source at (→
−
r 0 , t0 )

→
− →
−
∇ . p( r , t) ∇ G( r , t; r , t ) + λ q(→
→
− →
− →
− 0 0 −
r , t) G(→
−
r , t; →
−
r 0 , t0 )
1 ∂ 2 G(→ −
r , t; →
−
r 0 , t0 )
− = δ 3 (→−
r −→−r 0 ) δ(t − t0 ) (1290)
c2 ∂t2
and the similar equation with a source at (→ − r ”, t”). Multiplying the equation for
G( r , t; r , t ) by G( r , t; r ”, t”) and subtracting it from G(→
→
− →
− 0 0 →
− →
− −r , t; →
−
r 0 , t0 ) times
→
− →
−
the equation for G( r , t; r ”, t”), one obtains

→
− →
− →
− →
−
G(→
−
r , t; →
−
r ”, t”) ∇ . p(→
−
r , t) ∇ G(→
−
r , t; →
−
r 0 , t0 ) − G(→
−
r , t; →
−
r 0 , t0 ) ∇ . p(→
−
r , t) ∇ G(→
−
r , t; →
−
r ”, t”)
1 →
− →
− ∂ 2 G(→
−
r , t; →
−
r 0 , t0 ) 1 →− →
−
2 →
− → −
0 0 ∂ G( r , t; r ”, t”)
− G( r , t; r ”, t”) + G( r , t; r , t )
c2 ∂t2 c2 ∂t2
→
− →
− 3 →
− →
− 0 0 →
− →
− 0 0 3 →− →
−
= G( r , t; r ”, t”) δ ( r − r ) δ(t − t ) − G( r , t; r , t ) δ ( r − r ”) δ(t − t”)
(1291)
On integrating over →
−r and t, one obtains
Z ∞ Z
→
− →
−
dt d2 S G(→−
r , t; →
−
r ”, t”) . p(→−
r , t) ∇ G(→
−
r , t; →
−
r 0 , t0 )
−∞
Z ∞ Z
2→
− →
− →
− 0 0 →
− →
− →
− →
−
− dt d S G( r , t; r , t ) . p( r , t) ∇ G( r , t; r ”, t”)
−∞
∂G(→
−r , t; →
− t=∞
r 0 , t0 )
Z
1 3→
− →
− →
−
− d r G( r , t; r ”, t”)
c2 ∂t
t=−∞
1
Z
∂G(→
−r , t; →
−
r ”, t”)
t=∞
d3 →
−
r G(→ −
r , t; →
−
r 0 , t0 )

+
c2 ∂t
t=−∞
→
− 0 0 →− →
− →
−
= G( r , t ; r ”, t”) − G( r ”, t”; r , t )0 0
(1292)
On imposing appropriate boundary conditions, the surfaces terms in space time

vanish and one finds that the Green’s function is symmetric in its variables
G(→
−
r 0 , t0 ; →
−
r ”, t”) = G(→
−
r ”, t”; →
−
r 0 , t0 ) (1293)
219
We shall now illustrate the solution of the equation for the Green’s function
of Poisson’s equation and the Wave Equation, in several different geometries.
Poisson’s Equation
————————————————————————————————–
Example:
The Green’s function for Poisson’s equation inside a sphere is given by the
solution of
∇2 G(→−r ,→
−
r 0 ) = δ 3 (→
−
r −→
−r 0) (1294)
In spherical coordinates the delta function can be written as
δ(r − r0 ) δ(θ − θ0 )
δ 3 (→
−
r −→
−
r 0) = δ(ϕ − ϕ0 )
r2 sin θ
∞ m=l
δ(r − r0 ) X X
= Y ∗m 0 0 m
l (θ , ϕ ) Yl (θ, ϕ)
r2
l=0 m=−l
(1295)
where we have used the completeness condition for the spherical harmonics. The
Green’s function can also be expanded in powers of the spherical harmonics
∞ m=l
G(→
−
r ,→
− Gl,m (r, →
−
X X
r 0) = r 0) Y ∗m
l (θ, ϕ) (1296)
l=0 m=−l
On substituting the expansion for the Green’s function into the differential equa-
tion one obtains
X 1 ∂
2 ∂ l(l + 1)
r Gl,m − Gl,m Ylm (θ, ϕ)
r2 ∂r ∂r r2
l,m
δ(r − r0 ) X ∗ m
= Y l (θ, ϕ) Ylm (θ, ϕ) (1297)
r2
l,m
On multiplying by the spherical harmonics, and integrating over the solid angle,
the orthogonality of the spherical harmonics yields
δ(r − r0 ) ∗ m 0 0

1 ∂ 2 ∂ l(l + 1)
r G l,m − G l.m = Y l (θ , ϕ ) (1298)
r2 ∂r ∂r r2 r2
Since the left-hand-side has no explicit dependence on the angle (θ0 , ϕ0 ), one can
separate out the angular dependence
Gl,m (r, →
−
r 0 ) = gl (r, r0 ) Y ∗ m 0 0
l (θ , ϕ ) (1299)
220
and find
δ(r − r0 )

1 ∂ 2∂ l(l + 1)
r gl,m − gl.m = (1300)
r2 ∂r ∂r r2 r2
This is the equation for a one-dimensional Green’s function. For r 6= r0 , the
inhomogeneous term vanishes so one has solutions of the form
(
0
1
A rl + B r(l+1) r < r0
gl (r, r ) = (1301)
l 1
C r + D r(l+1) r > r0
The boundary conditions that G(→

−
r ,→
−
r 0 ) is finite at r = 0 and vanishes at
r = a implies that
B = 0
1
C al + D = 0 (1302)
al+1
Thus, the purely radial part of the Green’s function is given by
A rl r < r0
0

gl (r, r ) = 2l+1 (1303)
C rl − ra(l+1) r > r0
The two remaining coefficients A and C are determined by continuity at r = r0

and by the discontinuity in the slope. That is
gl (r0 + , r0 ) = gl (r0 − , r0 ) (1304)
and by integrating the differential equation over the singularity one obtains
r0 +
∂ 0
1
gl (r, r ) = 02 (1305)
∂r r 0 − r
From these we, obtain the matching conditions
a2l+1

0l
C r − 0l+1 − A r0l = 0
r
a2l+1

0l−1 1
C l r + ( l + 1 ) 0l+2 − A l r0l−1 = 02
r r
(1306)
These two equations can be solved for A and C, yielding
r0l
C =
(2l + 1) a2l+1
2l+1
r0l

a
A = 1 − (1307)
(2l + 1) a2l+1 r0
221
Hence, the radial part of the Green’s function is
 2l+1 
0
1 − ra0 r < r 

 
 
l 0l
 
r r  
gl (r, r0 ) = (1308)
(2l + 1) a2l+1 
 2l+1 

1 − ar r > r0

 

 
In terms of this function, one can write the Green’s function as

∞ m=l
G(→
−
r ,→
−
X X
r 0) = gl (r, r0 ) Y ∗ m 0 0 m
l (θ , ϕ ) Yl (θ, ϕ)
l=0 m=−l
∞
X m=l
X (2l + 1)
= gl (r, r0 ) Pl (cos γ) (1309)
4π
l=0 m=−l
where we have used the spherical harmonic addition theorem and γ is the angle
between →
−
r and →−r 0.
In the limit that the boundaries are removed to infinity, a → ∞, the Green’s
function simplifies to
 2l+1 
1 0 
r < r

 
 r0 
l 0l
 
0 r r  
gl (r, r ) = −
(2l + 1)  2l+1 

1 0 
 


r r > r 
l
1 r<
= − (1310)
(2l + 1) r> r<
Thus, the Green’s function in infinite three-dimensional space is given by the
familiar result
∞ l
1 r<
G(→
−
r ,→
−
X
r 0) = − Pl (cos γ)
4 π r> r<
l=0
1
= − p
4π r2 − 2 r r0 cos γ + r02
1
= − (1311)
4π| r − →
→
− −
r0 |
which is the Green’s function for Laplace’s equation.
————————————————————————————————–
222
Example:
The solution of Poisson’s equation in two dimensions, for the potential φ(r)
confined by a conducting ring of radius a can be obtained from the Green’s
function. The Green’s function satisfies the partial differential equation
∂G(→−r ,→
−r 0) 1 ∂ 2 G(→
−
r ,→
−r 0) δ( r − r0 )

1 ∂
r + 2 = δ( θ − θ0 ) (1312)
r ∂r ∂r r ∂θ2 r
On using the completeness relation
∞
1 X
δ( θ − θ0 ) = exp i m ( θ − θ0 ) (1313)
2π m=−∞
and the series expansion

∞
1
G(→
−
r ,→
− Gm (r, →
−
X
r 0) = r 0) √ exp imθ (1314)
m=−∞
2π
one finds that the coefficient Gm must satisfy the ordinary differential equation
∂Gm (r, →
−r 0) m2 δ( r − r0 )

1 ∂ →
− 0 1 0
r − 2 Gm (r, r ) = √ exp − i m θ
r ∂r ∂r r r 2π
(1315)
Introducing the definition

→
− 0 0 1 0
Gm (r, r ) = gm (r, r ) √ exp − i m θ (1316)
2π
which isolates the angular dependence of Gm (r, → −

r 0 ), one finds that gm (r, r0 )
satisfies the ordinary differential equation
∂gm (r, r0 ) m2 δ( r − r0 )

1 ∂
r − 2 gm (r, r0 ) = (1317)
r ∂r ∂r r r
For m = 0 and r 6= r0 the differential equation has the general solution
g0 (r, r0 ) = A0 + B0 ln r (1318)
while for m 6= 0 and r 6= r0 the equation has the general solution
gm (r, r0 ) = Am rm + Bm r−m (1319)
The boundary conditions at r = 0 leads to the vanishing of the coefficients of

r−m . Hence, the solution close to the origin must be of the form
gm (r < r0 ) = Am rm (1320)
223
The boundary condition at r = a leads to a relation between the coefficients
of rm and r−m . The solution close to the edge of the ring has the form
" m #
m
0 r a
gm (r > r ) = Cm − (1321)
a r
for m 6= 0. The solution for the Green’s function for m = 0 is given by
g0 (r < r0 ) = A0 (1322)
close to the origin and

r
g0 (r > r0 ) = C0 ln (1323)
a
close to the ring. The boundary conditions at r = r0 are
gm (r0 + , r0 ) = gm (r0 − , r0 )
r0 +
∂gm (r, r0 ) 1
= (1324)
∂r
r 0 − r0
The boundary conditions at the intersection of the two intervals determine the
remaining two coefficients via
" m m #
r0

0m a
Am r = Cm − (1325)
a r0
and " m m #
r0

a
m Cm + − m Am r0m = 1 (1326)
a r0
These two equations can be solved resulting in
0 m
1 r
Cm = (1327)
2m a
and " m m #
a−m r0

a
Am = − (1328)
2m a r0
Hence, the solution is given by
m " m m #
r0

0 1 r a
gm (r < r ) = −
2m a a r0
m " m m #
r0

0 1 r a
gm (r > r ) = − (1329)
2m a a r
224
and
r0
g0 (r < r0 ) = ln
a
r
g0 (r > r0 ) = ln (1330)
a
The Green’s function is given by
∞
1
G(→
−
r ,→
−
X
r 0) = 0
gm (r, r ) exp 0
im(θ − θ )
2π m=−∞
∞
1 0
X
0 0
= g0 (r, r ) + 2 gm (r, r ) cos m ( θ − θ )
2π m=1
(1331)
In the limit that the boundaries are removed to infinity a → ∞ one has
m
0 1 r<
gm (r, r ) = − (1332)
2m r>
so
∞ m
1 r> 1 r<
G(→
−
r ,→
−
X
r 0) = ln − 0
cos m ( θ − θ )
2π a m=1
m r>
r + r02 − 2 r r0 cos ( θ − θ0 )
2
1
= ln
4π a2
1 |→−
r − →−
r0 |
= ln (1333)
2π a
which is the Green’s function for the Laplacian operator in an infinite two-
dimensional space.
The Wave Equation.
————————————————————————————————–
Example:
Consider the forced drum head, described by the inhomogeneous partial

1 ∂2φ

1
∇2 φ − 2 2
= − 2 f (→ −r , t) (1334)
c ∂t c σ
225
where f is the force per unit area normal to the drumhead and σ is the mass
density. The boundary condition on the displacements is φ(→ −r , t) = 0. We
shall assume that the driving force has a temporal Fourier decomposition
Z ∞
→
− dω →
−
f ( r , t) = √ f ( r , ω) exp − i ω t (1335)
−∞ 2π
Due to the linear nature of the equation one can find the solution φ(→−
r , t) from
→
−
the temporal Fourier transform φ( r , ω) defined via
Z ∞
→
− dω →
−
φ( r , t) = √ φ( r , ω) exp − i ω t (1336)
−∞ 2π
On Fourier transforming the differential equation with respect to t, one has
Z ∞
1 ∂2φ

dt 2
√ ∇ φ − 2 exp + i ω t
−∞ 2π c ∂t2
Z ∞
1 dt →
−
= − 2 √ f ( r , t) exp + i ω t (1337)
c σ −∞ 2π
On integrating the second derivative with respect to time by parts, twice, and
assuming that limt→±∞ φ(→ −r , t) = 0, one obtains
ω2 →

→
− − 1
2
∇ φ( r , ω) + 2 φ( r , ω) = − 2 f (→ −r , ω) (1338)
c c σ
To solve this equation we introduce the frequency dependent Green’s func-

tion, G(→
−r ,→
−
r 0 ; ω), which satisfies
ω2

→
− →
− →
− →
− 1
2 0
∇ G( r , r ; ω) + 2 G( r , r ; ω) 0
= √ δ 2 (→
−
r −→
−
r 0) (1339)
c 2π
which is the Fourier transform of the equation for the real time Green’s function.
Note that this holds only because the homogeneous equation is only a function
of t − t0 and does not depend on t and t0 independently. This is expected to
be true for most physical phenomenon due to the homogeneity of space time.
The Green’s function satisfies the boundary condition G(→ −r ,→
−
r ; ω) = 0, at the
edge of the drum head.
The solution for φ(→ −

r ; ω) is found in terms of the Green’s function. From
the partial differential equations one has
Z
d2 →
−
r φ(→
−r , ω) ∇2 G(→ −r ,→
−r 0 ; ω) − G(→
−
r ,→
−
r 0 ; ω) ∇2 φ(→
−
r , ω)
φ(→
−
r 0 , ω)
Z
1
= √ + 2 d2 →
−
r G(→
−
r ,→
−
r 0 ; ω) f (→
−
r , ω)
2π c σ
(1340)
226
Also, using the two dimensional version of Green’s theorem, this is equal to
∂G(→−
r ,→
−r 0 ; ω) ∂φ(→
−
Z
→
− →
− →
− 0 r ; ω)
= dl φ( r , ω) − G( r , r ; ω) (1341)
∂n ∂n
where the integral is over the perimeter of the two dimensional area and the
derivative with respect to n ia taken along the normal to the perimeter. For a
∂ ∂
circular area of radius a, one has dl = a dθ and ∂n = ∂r . Since both terms
vanish on the perimeter of the drumhead, φ( r , ω) = 0 and G(→
→
− −
r ,→
−
r 0 ; ω) = 0,
one has
φ(→−r 0 , ω)
Z
1
√ = − 2 d2 →
−
r G(→
−r ,→
−
r 0 ; ω) f (→
−
r , ω) (1342)
2π c σ
or using the symmetry of the Green’s function
φ(→
−r 0 , ω)
Z
1
√ = − 2 d2 →
−
r G(→−
r 0, →
−
r ; ω) f (→
−
r , ω) (1343)
2π c σ
Thus, the solution of the forced equation can be obtained from the Green’s
function and the forcing function. In particular one has
Z ∞
dω
φ(→
−r , t) = √ φ(→−
r , ω) exp − i ω t
−∞ 2π
Z ∞ Z
1
= − 2 dω d2 →
−
r 0 G(→
−
r ,→
−
r 0 ; ω) f (→
−
r 0 , ω) exp − i ω t
c σ −∞
(1344)
The Green’s function can be obtained by expanding it in terms of a Fourier

series in θ,
∞
→
− →
− 0
X
→
− 0 1
G( r , r ; ω) = Gm (r, r ; ω) √ exp i m θ (1345)
m=−∞
2π
The two dimensional Green’s function can be expressed as

δ(r − r0 )
δ 2 (→
−
r −→
−
r 0) = δ(θ − θ0 )
r
∞
δ(r − r0 ) 1 X
= exp i m ( θ − θ0 )
r 2 π m=−∞
(1346)
where we have used the completeness relation for the Fourier series to expand
the delta function of the angle.
227
On substituting these into the equation for the Green’s function, multiplying
by the complex conjugate basis function

1
√ exp − i m θ (1347)
2π
and integrating over θ, one obtains
∂Gm (r, →
−r 0) δ(r − r0 )
2
m2

1 ∂ ω 1
r + − 2 Gm (r, →
−
r 0) = exp − imθ 0
r ∂r ∂r c2 r 2π r
(1348)
On factoring out the θ0 dependence via

1
Gm (r, →
−
r 0 ) = gm (r, r0 ) √ exp − i m θ0 (1349)
2π
one finds that gm (r, r0 ) satisfies the equation
∂gm (r, r0 ) δ(r − r0 )

2
m2

1 ∂ ω 1
r + 2
− 2
gm (r, r0 ) = √
r ∂r ∂r c r 2π r
(1350)
The inhomogeneous term is zero for both regions r > r0 and r0 > r, so that
the solution is
ω
gm (r, r0 ) = A Jm ( r) r < r0
c
ω ω
gm (r, r0 ) = F Jm ( r) + G Nm ( r) r > r0
c c
(1351)
since the Green’s function must be regular at r = 0. The boundary condition

at r = a is
gm (a, r0 ) = 0 (1352)
Hence, we have
ω
gm (r, r0 )= A Jm ( r) r < r0
c
0 ω ω ω ω
gm (r, r ) = B Nm ( a) Jm ( r) − Jm ( a) Nm ( r) r > r0
c c c c
(1353)
228
The matching conditions at r = r0 are given by
gm (r0 + , r0 ) = gm (r0 − , r0 )
r0 +
∂gm (r, r0 ) 1 1
= √ (1354)
∂r
r 0 − 2 π r0
Hence, the coefficients A and B satisfy the linear equations

ω ω ω ω ω
A Jm ( r0 ) = B Nm ( a) Jm ( r0 ) − Jm ( a) Nm ( r0 )
c c c c c

1 1 ω ω 0 ω 0 ω 0 ω 0 0 ω 0
√ = B N m ( a) J m ( r ) − J m ( a) N m ( r ) − A J m ( r )
2 π r0 c c c c c c
(1355)
The solution of these equations can be simplified by noting that the Wronskian
of the solutions of the Bessel functions is given by
0 0 2
Jm (z) Nm (z) − Jm (z) Nm (z) = (1356)
πz
The solution can be written as

ω ω 0 ω ω 0
Nm ( c a) Jm ( c r ) − Jm ( c a) Nm ( c r )
1 π
A = − √
2π 2 Jm ( ωc a)
1 π Jm ( ωc r0 )
B = − √ ω (1357)
2 π 2 Jm ( c a)
The radial Green’s function can be written as
π Jm ( ωc r)

1 ω ω 0 ω ω 0
gm (r, r0 ) = − √ ω N m ( a) J m ( r ) − J m ( a) N m ( r ) r0 > r
2 π 2 Jm ( c a) c c c c
ω 0
1 π Jm ( c r ) ω ω ω ω
gm (r, r0 ) = − √ ω N m ( a) J m ( r) − Jm ( a) N m ( r) r0 < r
2 π 2 Jm ( c a) c c c c
(1358)
which is symmetric under the interchange m ) * − m. Then, the frequency

dependent Green’s function can be written as
∞
→
− →
− 0
X
0 1 0
G( r , r ; ω) = gm (r, r ) exp i m ( θ − θ )
m=−∞
2π
∞
1 X
= g0 (r, r0 ) + 2 gm (r, r0 ) cos m ( θ − θ0 )
2π m=1
(1359)
229
The time-dependent Green’s function is given by
Z ∞
→
− →
− 0 0 dω →
− →
− 0 0
G( r , r ; t − t ) = √ G( r , r ; ω) exp − i ω ( t − t ) (1360)
−∞ 2π
12.2.2 Eigenfunction Expansion

The Green’s function can also be obtained by expanding the Green’s function in
terms of eigenfunctions of an appropriate operator and using the completeness
relation for the delta function. The expansion coefficients can then be deter-
mined directly, using the orthogonality relation.
Poisson’s Equation.
We shall consider examples of the eigenfunction expansion solution for the

Green’s functions for Poisson’s equation, in various geometries.
————————————————————————————————–
Example:
The Green’s function for Poisson’s equation in an infinite three-dimensional

space can be obtained by an eigenfunction expansion method. The Green’s
function satisfies the equation
∇2 G(→
−
r ,→
−
r 0 ) = δ 3 (→
−
r −→
−
r 0) (1361)
Solution:
The eigenfunction of the Laplacian satisfy
→ (→
∇2 φ− − → (→
→ φ−
r ) = λ− −
r) (1362)
k k k
and has eigenfunctions

1 →
− −
→ (→
φ− −
r) = 3 exp i k .→
r (1363)
k ( 2 π )2
and have eigenvalues

→ = − k2
λ− (1364)
k
230
The Green’s function can be expanded as
Z
→
−
G(→−
r ,→
−
r 0) = → (→
d3 k G− − → (→
r 0 ) φ− −
r) (1365)
k k
and the delta function can be expanded using the completeness relation for the
eigenfunctions
Z
→
− ∗ →
δ 3 (→
−
r −→
−r 0) = − (−
d3 k φ→ → (→
r 0 ) φ− −
r) (1366)
k k
Inserting both of these expansions in the equation for the Green’s function we
find the expansion coefficients of the Green’s function satisfy the equation
Z Z
→
− →
− ∗ →
− → (→
d3 k k 2 G− − → (→
r 0 ) φ− −
r) = d3 k φ→− (− → (→
r 0 ) φ− −
r) (1367)
k k k k
Multiplying by φ→∗ →
− →
−
− 0 ( r ) and integrating over r one has the orthogonality re-
k
lation Z
→
− → −0
d3 →
−
r φ→− 0 (→
∗ −
r ) φ−→ (→−
r ) = δ3 ( k − k ) (1368)
k k
one projects out the coefficient G→ − 0 (→
−
r ) so
k
− 0 (→
− k 02 G→ − − 0 (→
∗
r ) = φ→ −
r 0) (1369)
k k
Hence, the Green’s function is given by
∗ →
Z − (−
φ→ r 0 ) φ−→ (→
−r)
3→
− k k
G(→
−
r ,→
−
r 0) = − d k
k2

→
− →
− →
− 0
→
− exp i k . ( r − r )
d3 k
Z
= −
( 2 π )3 k2
(1370)
It is seen that the Green’s function only depends on the difference of the po-
sitions, →
−
r −→ −r 0 , and not on →
−
r and →
−r 0 separately. This is because we have
assumed that space is homogeneous. To simplify further calculations we shall
re-center our coordinates on the source point at →−r 0 . In this case, the Green’s
function only depends on r .→
−
In spherical polar coordinates, one has

→
− → −
− exp i k . r
→
d3 k
Z
G(→−r) = −
( 2 π )3 k2
231

2π π ∞ exp i k r cos θ
dk k 2
Z Z Z
= − dϕ d θ sin θ
0 0 0 ( 2 π )3 k2
(1371)
where the polar axis has been chosen along the direction of → −
r . The integral
over ϕ and cos θ can be performed as
Z 2π Z 1 Z ∞
→
− dk
G( r ) = − dϕ d cos θ exp i k r cos θ
0 −1 0 ( 2 π )3

Z ∞ exp i k r − exp − i k r
dk
= −
0 ( 2 π )2 ikr
(1372)
On writing k = − k in the second term, one can extend the integration to

(−∞, ∞), yielding

Z ∞ exp i k r
dk
G(→
−
r) = − 2
(1373)
−∞ ( 2 π ) ikr
where the integral over the singularity at k = 0 is interpreted as being just the
principal part. That is, the integral is evaluated over two regions (−∞, −ε) and
(ε, ∞), which end symmetrically around the singularity at k = 0. The integral
is evaluated by taking the limit ε → 0. The integration can be evaluated by
Cauchy’s theorem to yield π i times the residue. Hence, we obtain
1
G(→
−
r) = − (1374)
(4πr)
or re-instating the source position at →

−
r0
1
G(→
−
r ,→
−
r 0) = − (1375)
4π| r − →
→
− −
r0 |
as expected from Coulomb’s law.
————————————————————————————————–
Example:
The Green’s function for the Laplacian in an infinite d-dimensional space

satisfies
∇2 G(→−r ,→
−
r 0 ) = δ d (→
−
r −→
−
r 0) (1376)
232
where in ultra-spherical polar coordinates one has the delta function
0 0
δ(r − r0 ) δ(θd−1 − θd−1 ) δ(θd−2 − θd−2 ) δ(θ2 − θ20 )
δ d (→
−
r −→
−
r 0) = . . . δ(ϕ−ϕ0 )
rd−1 sind−2 θd−1 sind−3 θd−2 sin θ2
(1377)
One can solve for the Green’s function using the eigenfunction expansion method.
The Green’s function is expanded as
→
−
dd k
Z
→
− →
− 0 →
− 0 →
− → −
G( r , r ) = d G−→ ( r ) exp i k . r (1378)
( 2 π )2 k
The delta function can also be expanded using the completeness relation
→
−
dd k
Z
d →
− →
− 0 →
− →
− →
− 0
δ (r − r ) = exp i k . ( r − r ) (1379)
( 2 π )d
These are substituted into the equations of motion, which then gives an equation
involving all the Green’s function expansion coefficients. The orthogonality
relation of the eigenfunctions
dd →
−
Z
r →
− →
− →
−0 →
− → −0
exp i r . ( k − k ) = δd ( k − k ) (1380)
( 2 π )d
can be used to obtain the expansion coefficients as

→
−0 → − 0
exp − i k . r
G→− 0 (→
−
r 0) = d (1381)
k ( 2 π )2
Hence, we find that due to the isotropy and translational invariance of space
the Green’s function is only a function of the spatial separation →
−
r −→−
r 0 and
→
− →
− 0
does not depend on r and r separately

→
− →
− →
− 0
− exp i k . ( r − r )
→
dd k
Z
G(→−r −→−r 0) = − (1382)
( 2 π )d k2
or
→
− −
− exp
→ i k .→
r
dd k
Z
G(→
−
r) = − (1383)
( 2 π )d k2
The integration will be evaluated in spherical polar coordinates, in which the
polar axis is chosen along the direction of →−
r . The integration over the d-
d→
−
dimensional volume d k can be expressed in spherical polar coordinates as
→
−
dd k = dk k d−1 dθd−1 sind−2 θd−1 dθd−2 sind−3 θd−2 . . . dθ2 sin θ2 dϕ (1384)
233
Thus, we have

∞ exp i k r cos θd−1
dk k d−1 π
Z Z
G(→
−
r) = − dθ d−1 sind−2
θ d−1
0 ( 2 π )d 0 k2
Z π Z π Z 2π
× dθd−2 sind−3 θd−2 . . . dθ2 sin θ2 dϕ
0 0 0
(1385)
The angular integrations which only involve the weight factors can be performed
using the formula
√ Γ( m+1
Z π
2 )
dθ sinm θ = π m+2 (1386)
0 Γ( 2 )
where the Γ function is the generalized factorial function defined by
Z ∞
Γ(z) = dx exp − x xz−1 (1387)
0
One can show that, by integrating by parts with respect to x, the Γ function
satisfies the same recursion relation as the factorial function
Γ(z) = ( z − 1 ) Γ(z − 1) (1388)
and as Z ∞
Γ(1) = dx exp − x = 1 (1389)
0
Thus, for integer n the Γ function reduces to n factorial
Γ(n + 1) = n! (1390)
Hence, the Green’s function is expressed as

Z ∞ Z π
dk k d−3

G(→
−r) = − dθ d−1 sind−2
θ d−1 exp i k r cos θ d−1
0 ( 2 π )d 0
√ Γ( d−2
2 )
√ Γ( 22 )
× π ... π 2π
Γ( d−1
2 )
Γ( 23 )
(1391)
The integration over the angle θd−1 can be performed using
π d−2
√

d−1
Z
d−2 2 2
dθd−1 sin θd−1 exp i k r cos θd−1 = π Γ( ) J d−2 (kr)
0 2 kr 2
(1392)
234
Thus, we find
∞ d−2
dk k d−3
Z
1 2
G(→
−
r) = − d J d−2 (kr)
0 ( 2 π )2 kr 2
d−2 Z ∞
1 2
dk d−4
= − d k 2 J d−2 (kr) (1393)
r 0 (2π) 2 2
Finally, the integral over k can be evaluated with the aid of the formula
∞
2µ Γ( µ+ν+1 )
Z
dk k µ Jν (kr) = 2
(1394)
0 rµ+1 Γ( µ−ν+1
2 )
which yields the Green’s function for the d-dimensional Laplacian as
Γ( d−2

→
− 2 ) 1
G( r ) = − d (1395)
4 π2 rd−2
The d-dimensional Green’s function for the Laplacian operator reduces to the
three-dimensional case previously
√ considered, as can be seen by putting d = 3
and noting that Γ( 12 ) = π.
The Wave Equation
The Green’s function for the wave equation can be obtained by expansion
in terms of eigenfunctions.
————————————————————————————————–
Example:
The Green’s function for the wave equation, inside a two dimensional area,
satisfies the inhomogeneous equation
1 ∂2

2
∇ − 2 G(→−r ,→
−
r 0 ; t − t0 ) = δ 2 (→
−
r −→
−
r 0 ) δ(t − t0 ) (1396)
c ∂t2
Solution:
On Fourier transforming the equation for the Green’s function with respect
to t according to
Z ∞
→
− →
− 0 dt →
− →
− 0 0 0
G( r , r ; ω) = √ G( r , r ; t − t ) exp + i ω ( t − t ) (1397)
−∞ 2π
235
yields
ω2

1
∇2 + G(→
−
r ,→
−
r 0 ; ω) = √ δ 2 (→
−
r −→
−
r 0) (1398)
c2 2π
The eigenfunctions of the operator satisfy the eigenvalue equation
ω2

2
∇ + 2 φk,m (→
−r ) = λk,m φk,m (→ −
r) (1399)
c
where k, m label the eigenfunctions. The explicit form of the eigenvalue equation
is given by
1 ∂ 2 φk,m ω2

1 ∂ ∂φk,m
r + 2 + φ k,m = λk,m φk,m (1400)
r ∂r ∂r r ∂θ2 c2
The eigenfunctions that satisfy the boundary conditions at the origin and vanish
on a circle of radius a are given by

1
φk,m (r, θ) = Nm Jm (kr) √ exp i m θ (1401)
2π
where the normalization Nm is given by
−1
a 0
Nm = √ Jm (ka) (1402)
2
and the number k is given by the zeroes of the Bessel function
Jm (ka) = 0 (1403)
The eigenvalues are given by
ω2
λk,m = − k2 (1404)
c2
The frequency dependent Green’s function is given by

∞ X N 2 Jm (kr) Jm (kr0 )
1
G(→
−
r ,→
−
X
r 0 ; ω) = √ m
ω2
0
exp i m ( θ − θ )
2π m=−∞
2π c2 − k
2
k
1 X N 2 J0 (kr) J0 (kr0 )
G(→
−
r ,→
−
r 0 ; ω) = √ 0
ω2
2π k 2π c2 − k
2
∞ X
1 X Nm2
Jm (kr) Jm (kr0 )
+ √ ω2
cos m ( θ − θ0 )
2π m=1
π c2 − k 2
k
(1405)
236
where we have used the symmetry of the product Jm (x) Jm (y) under the inter-
change of m and − m.
————————————————————————————————–
Example:
The Green’s function for the wave equation in an infinite three-dimensional

space can be found from the temporal Fourier transform
ω2

1
∇2 + 2 G(→
−r ,→
−
r 0 ω) = √ δ 3 (→
−
r −→−r 0) (1406)
c 2π
Solution:
The Green’s function is expanded in terms of the plane wave eigenfunctions

which satisfy
ω2 ω2

∇2 + 2 φ− → (→−
r ) = − k 2
+ → (→
φ− −r) (1407)
c k c2 k
in which the eigenfunction is found as

1 →
− −
→ (→
φ− −
r) = 3 exp + i k .→
r (1408)
k ( 2 π )2
One obtains the expression

→
−
→
− exp i k .(→
−
r − →
−
r0 )
d3 k
Z
1
G(→
−
r ,→
−
r 0 ; ω) = √ ω2
(1409)
2π ( 2 π )3 c2 − k2
and hence

→
−
∞ − exp
→ i k .(→
−
r − →
−
r 0 ) − i ω ( t − t0 )
d3 k
Z Z
dω
G(→
−
r ,t : →
−
r 0 , t0 ) = ω2
−∞ 2π ( 2 π )3 c2 − k2
(1410)
→
−
The angular part of the integration over k can be performed by using r − →
→
− −r0
as the polar axis. In this case one finds

Z ∞ →
− →
− 0 →
−
exp + i k | r − r | − exp − i k | r − r |→
− 0
dk k 2
G(→−r ,t : →
−
r 0 , t0 ) =
0 ( 2 π )2 ik|→−r − →−r0 |

Z ∞ exp − i ω ( t − t0 )
dω
× ω2
−∞ 2 π
2
c2 − k
237

Z ∞ exp + ik|→
−
r − →
−
r0 |
dk k
=
−∞ ( 2 π )2 |→
−
r − →
−
r0 |

Z ∞ exp − i ω ( t − t0 )
dω
× ω2
(1411)
−∞ 2πi c2 − k2
The integration over ω can be performed using Cauchy’s theorem of contour
integration. The contour has to be closed in the lower-half complex ω plane for
positive t, and in the upper-half complex plane for negative t. There are poles
at ω = ± c k in the lower-half plane. This leads to
Z ∞
c Θ(t − t0 )

dk
G(→−r ,t : →
−
r 0 , t0 ) = − i → exp i k | →
−r − →
−
r 0
| sin c k ( t − t0 )
|−r − → −
r 0 | −∞ ( 2 π )2

c →
− →
− 0 0
= − δ | r − r | − c ( t − t ) (1412)
4π|→ −
r − → −r0 |
————————————————————————————————–
Example:
The Green’s function for the wave equation in an infinite d-dimensional space
can also be obtained using this method. First, the Green’s function equation
is Fourier transformed with respect to time. The frequency dependent Green’s
function is then defined as the solution of the inhomogeneous partial differential
equation
ω2

1
∇2 + 2 G(→−r ,→
−r 0 ; ω) = √ δd ( →
−
r − →−r0 ) (1413)
c 2π
Solution:
The Green’s function can be found by using the completeness relation of the
plane wave eigenfunctions of the Laplace operator. On expanding the Green’s
function in terms of plane waves and also using the representation of the d-
dimensional delta function
→
−
dd k
Z
→
−
δd ( →
−
r − →−
r0 ) = exp i k . ( →
−
r − →
−r 0
) (1414)
( 2 π )d
one finds that the frequency dependent Green’s function can be written as

→
−
Z exp i k . ( →
−
r − → −
r0 )
1 →
−
G(→−r ,→
−
r 0 , ω) = (d+ 1
)
dd k ω2
(1415)
2
(2π) c2 − k
2
238
The time-dependent Green’s function is given by the inverse Fourier Transform
of the frequency dependent Green’s function

→
− →
− →
− 0 0
Z ∞ Z exp i k . ( r − r ) − i ω ( t − t )
1 d→
−
G(→−r ,→
−
r 0 , t−t0 ) = dω d k ω2
( 2 π )(d+1) −∞ c2 − k
2
(1416)
Note that because space time is homogeneous ( we have no boundaries ) the
Green’s function is invariant under translations through time and space.
→
−
The integral over the d-dimensional volume dd k in ultra-spherical polar
coordinates is given by
Z ∞ Z π Z π Z π Z 2π
dk k d−1 dθd−1 sind−2 θd−1 dθd−2 sind−3 θd−2 . . . dθ2 sin θ2 dϕ
0 0 0 0 0
(1417)
The integral over the solid angle involves an integral over the principal polar
angle of the form
d−2
√

d−1
Z
(d−2) 2 2
dθ sin θ exp i x cos θ = π Γ( ) J d−2 (x) (1418)
2 x 2
where Γ(x) is the factorial function
Γ(x + 1) = x Γ(x) (1419)
and for integer n, with Γ(1) = 1, yields
Γ(n + 1) = n! (1420)
The subsequent angular integrations are performed using the formula

π √ Γ( m+1
2 )
Z
dθ sinm θ = π m+2 (1421)
0 Γ( 2 )
Thus, the angular integrations produce the factor

d−2
√ d−1 2 2 √ Γ( d−2
2 )
√ Γ( d−3
2 )
√ Γ( 22 )
π Γ( ) J d−2 (x) π π . . . π 2π
2 x 2 Γ( d−1
2 ) Γ( d−2
2 )
Γ( 32 )
d−2
d 2 2
= 2 π2 J d−2 (x)
x 2
d d−2
= ( 2 π ) 2 x− 2 J d−2 (x) (1422)
2
239
as the Γ function is defined such that Γ(1) = 1.
Also the integral over ω can be performed, using contour integration. Since
we only want the causal part of the Green’s function for physical reasons, the
poles on the real ω-axis at ω = ± c k are displaced into the lower-half complex
plane, giving

Z ∞ exp − i ω t
2πc
dω ω2
= Θ(t) sin c k t (1423)
−∞ c2 − k
2 k
where Θ(t) is the Heaviside step function defined by
t > t0

1
Θ(t) = (1424)
0 t < t0
This leads to the expression

∞
c Θ(t − t0 )
Z
G(→
−
r −→
− J d−2 (k| →
−
r −→
−
d−2
r 0 ; t−t0 ) = − dk k 2 r 0 |) sin c k ( t − t0 )
( 2 π )2 | →
−
r − → − d−2
r0 | 2
d
2
0
(1425)
For convenience of notation we re-center the origin of our coordinate system on
the source at (→−r 0 , t0 ), so the Green’s function only explicitly depends on the
field point coordinates. The integral over k can be evaluated with the aid of the
formula
Z ∞
( 2 α )ν Γ(ν + 21 )

dx xν Jν (αx) exp − β x = √ 1 (1426)
0 π ( α2 + β 2 )ν+ 2
where Γ(n + 1) = n! is the factorial function. However, to ensure that the

integrals over k converge we add a convergence factor of

lim exp − k ε (1427)
ε→0
to the integrals. Thus, one finds the integral

Z ∞
lim dk k ν Jν (kr) sin kr exp − k ε =
ε→0 0
" #
2ν+1
ν Γ( 2 ) 1 1
= lim ( 2 r ) √ 2ν+1 − 2ν+1
ε→0 2i π [ r 2 + ( − i c t + ε )2 ] 2 [ r2 + ( i c t + ε )2 ] 2
(1428)
Then on defining
s = r2 + ( − i c t + ε )2 (1429)
240
and using the identity
1 ∂ n−1 −1
s−n = ( − 1 )n s (1430)
(n − 1)! ∂sn−1
one has
−n n−1 −1
2 2 1 1 ∂ 2 2
r + (−ict + ε) = − r + (−ict + ε)
(n − 1)! 2 r ∂r
(1431)
After some elementary manipulations one obtains the time-dependent Green’s

function
d−3
c Θ(t) 1 ∂ 2
δ( r − c t )
G(r; t) = − − (1432)
4π 2 π r ∂r r
The term involving a second delta function has been dropped as it does not
contribute, due to the presence of the Heaviside function. The Green’s function
is retarded, in the sense that the solution at point (→ −r , t) only experiences the
effect of the source at the point (→ −r 0 , t0 ) only if t is later than t0 . Thus, the re-
tarded Green’s function expresses causality. However, it also shows that the an
effect at point (→
−
r , t) only occurs if the signal from the point source at (→ −
r 0 , t0 )
exactly reaches the point (→ −
r , t), and that in between the signal travels with
speed c.
241
13 Complex Analysis
Consider a function of a complex variable z = x + i y. Then the function
f (z) of the complex variable also has a real and imaginary part, so
f (z) = u(x, y) + i v(x, y) (1433)
where u and v are real functions of x and y. The derivative of a complex function
is defined by the limit
df (z) f (z + ∆z) − f (z)
= lim (1434)
dz ∆z→0 ∆z
provided that the limit exists and is independent of the path in which the limit
is taken. For example, if one considers
∆z = ∆x + i ∆y (1435)
then one can take two independent limits, either ∆x → 0 or ∆y → 0. In

general
∆f ∆u + i ∆v
= (1436)
∆z ∆x + i ∆y
On taking the limit ∆x → 0 one obtains

∆f ∂u ∂v
lim = + i (1437)
∆x→0 ∆x ∂x ∂x
while on the path ∆z = i ∆y the derivative takes the value

∆f ∂u ∂v
lim = −i + (1438)
∆y→0 i ∆y ∂y ∂y
If the derivative is to be well defined, these expressions must be equal, so one
must have
∂u ∂v
= (1439)
∂x ∂y
and
∂u ∂v
= − (1440)
∂y ∂x
These are the Cauchy-Riemann conditions. They are necessary conditions, for if
a unique derivative is to exist the Cauchy Riemann conditions must hold. Note
that a function that satisfies the Cauchy Riemann conditions automatically is
a solution of Laplace’s equation.
Conversely, if the Cauchy Riemann conditions are satisfied then f (z) is con-
tinuous, and the derivative exists. This can be formulated as a theorem.
242
Theorem.
Let u(x, y) and v(x, y) be the real and imaginary parts of a function of a
complex variable f (z), which obey the Cauchy Riemann conditions and also
posses continuous partial derivatives ( with respect to x and y ) at all points in
some region of the complex plane, then f (z) is differentiable in this region.
Proof.
Since u and v have continuous first partial derivatives there exist four num-
bers, 1 , 2 , δ1 , δ2 that can be made arbitrarily small as ∆x and ∆y tend to zero,
such that

∂u ∂u
u(x+∆x, y+∆y) − u(x, y) = ∆x + ∆y + 1 ∆x + δ1 ∆y (1441)
∂x ∂y
and

∂v ∂v
v(x+∆x, y+∆y) − v(x, y) = ∆x + ∆y + 2 ∆x + δ2 ∆y (1442)
∂x ∂y
Multiplying the second of these equations by i and adding them one has

∂u ∂u ∂v ∂v
f (z + ∆z) − f (z) = ∆x + ∆y + i ∆x + i ∆y
∂x ∂y ∂x ∂y
+ 1 ∆x + δ1 ∆y + i 2 ∆x + i δ2 ∆y (1443)
Using the Cauchy Riemann conditions one has

∂u ∂v ∂v ∂u
f (z + ∆z) − f (z) = ∆x − ∆y + i ∆x + i ∆y
∂x ∂x ∂x ∂x
+ 1 ∆x + δ1 ∆y + i 2 ∆x + i δ2 ∆y
" #" #
∂u ∂v
= + i ∆x + i ∆y
∂x ∂x
+ 1 ∆x + δ1 ∆y + i 2 ∆x + i δ2 ∆y (1444)
Thus, on dividing by ∆z, and taking the limit one has

df (z) ∂u ∂v
= + i (1445)
dz ∂x ∂x
————————————————————————————————–
243
Example:
The function f (z) = z 2 satisfies the Cauchy Riemann conditions, at all

points in the complex plane as
u(x, y) = x2 − y 2 (1446)
and
v(x, y) = 2 i x y (1447)
Hence,
∂u ∂v
= 2x = (1448)
∂x ∂y
and
∂u ∂v
= −2y = − (1449)
∂y ∂x
However, the function f (z) = (z ∗ )2 does not satisfy the Cauchy Riemann
conditions, but is still continuous.
On applying the Cauchy Riemann conditions to a function which is assumed

to be expandable as a Taylor series in some region around (x0 , y0 ) one can show
that
u(x, y) + i v(x, y) − u(x0 , y0 ) − i v(x0 , y0 ) =

∞ n n
X 1 ∂
= x + i y − x0 − i y 0 u(x0 , y0 ) + i v(x0 , y0 )
n=1
n! ∂xn0
(1450)
That is, the function only depends on z = x + i y and not on the complex
conjugate z ∗ = x − i y. It should be noted that not all functions can be Taylor
expanded about an arbitrary point, for example the functions
1
f (z) = (1451)
z
and
1
f (z) = exp − (1452)
z
can not be Taylor expanded around the point z = 0. The partial derivatives
do not exist, and the Cauchy-Riemann conditions do not hold at z = 0. The
radius of convergence of the Taylor expansion about the origin is zero.
Analytic Functions
244
A function f (z) is analytic at the point z0 if it satisfies the Cauchy Riemann
conditions at z0 . If f (z) is analytic at all points in the entire complex z plane
it is defined to be an entire function.
13.1 Contour Integration

The integration of a complex function, is defined as an integration over a curve in
the complex plane. The curve is known as a contour, and may be parameterized
by
x = x(s)
y = y(s) (1453)
The integral of f (z) over a specific contour C can be evaluated as

Z Z s1
dx(s) dy(s)
dz f (z) = ds + i f (z(s))
C s ds ds
Z 0s1
dx(s) dy(s)
= ds u(x(s), y(s)) − v(x(s), y(s))
s0 ds ds
Z s1
dx(s) dy(s)
+ i ds v(x(s), y(s)) + u(x(s), y(s))
s0 ds ds
(1454)
which reduces the contour integration to the sum of two Riemann integrations.
————————————————————————————————–
Example:
Consider the integration of the function f (z) = z 2 over an open contour

from z = 0 to z 1 + i along two different paths, the contour C1 which is a
parabola
x(s) = s
y(s) = s2 (1455)
and a second contour C2 which consists of a straight line segment
x(s) = s
y(s) = s (1456)
245
The integral over contour C1 is evaluated as
Z Z 1
dz(s) 2 2
dz f (z) = ds (x − y ) + i2xy
C1 0 ds
Z 1
dx(s) dy(s)
= ds + i ( x2 − y 2 ) + i 2 x y
0 ds ds
Z 1
= ds 1 + i 2 s ( s2 − s4 ) + i 2 s3
0
Z 1
2 4 3 5
= ds (s − 5s ) + i(4s − 2s )
0
s=1
1 3 2 6
s − s5 ) + i ( s4 −

= ( s )
3 6
s=0

2 2
= − + i (1457)
3 3
The integral over the contour C2 is evaluated as

Z Z 1
dz(s)
dz f (z) = ds ( x2 − y 2 ) + i 2 x y
C2 0 ds
Z 1
dx(s) dy(s) 2 2
= ds + i (x − y ) + i2xy
0 ds ds
Z 1
= ds 1 + i ( s2 − s2 ) + i 2 s2
0
Z 1
= ds − 2 s2 + i 2 s2
0
s=1
2 3 2
s + i s3

= −
3 3
s=0

2 2
= − + i (1458)
3 3
Thus, the integration over an entire function appears to be independent of the
contour on which it is evaluated.
————————————————————————————————–
Example:
Important examples of contour integration are given by the integration of

a mono-nomial z n over a closed circle of radius r centered on the origin of the
246
complex z plane. The integral can be evaluated as
Z 2π
rn
Z
1
dz z n = dθ exp i ( n + 1 ) θ (1459)
2πi C 2π 0
where we have used the polar form for the complex number

z = r exp i θ (1460)
and on the contour one has

dz = r exp iθ i dθ (1461)
The integral is easily evaluated for integer values of n as

Z 2π
rn
Z
1 n
dz z = dθ exp i ( n + 1 ) θ
2πi C 2π 0
2π
rn+1
Z
= dθ exp i ( n + 1 ) θ
2π 0
2π
rn+1

= exp i ( n + 1 ) θ
2πi(n + 1)
0
= 0 (1462)
for integer n, n 6= − 1. This occurs since the exponential is 2 π periodic, and
the result is independent of the radius r of the contour.
————————————————————————————————–
A second important example is found by examining the case where n = − 1.

In this case, the closed contour integral is evaluated as
Z Z 2π
1 1
dz z −1 = dθ
2πi C 2π 0
2π
1
= θ
2π 0
= 1 (1463)
It is noteworthy, that the function z −1 does not satisfy the Cauchy Riemann
conditions at the origin, and the contour wraps around the origin once. The
function and the derivatives are not defined at the origin.
Both of these examples yield results which are independent of r, the radius
of the closed circular contour.
247
A contour integral of function, that satisfies the Cauchy Riemann conditions
at every point of a region of the complex plane which contains the function, is
independent of the path of integration. There is an analogy between integration
of analytic functions and conservative forces in Mechanics; the analytic func-
tion plays the role of a conservative force and the integral plays the role of the
potential. Alternatively, there is an analogy between integration of an analytic
function and functions of state in thermodynamics.
The contour integral of an analytic function f (z) can be calculated as the

difference of a function F (z) evaluated the end points,
Z
dz f (z) = F (z1 ) − F (z0 ) (1464)
C
where the function F (z) satisfies
dF (z)
= f (z) (1465)
dz
The function F (z) is the primitive of the function f (z). This is the content of
Cauchy’s Integral Theorem.
13.2 Cauchy’s Integral Theorem

If a function f (z) is analytic and its partial derivatives are continuous every-
where inside a (simply connected) region, then the contour integration over any
closed path entirely contained within the region is zero
I
dz f (z) = 0 (1466)
C
A simply connected region means a region in which there are no holes.
Stokes’s Proof.
Consider the integration over the closed loop

I I
dz f (z) = u + iv dx + i dy
C C
I I
= u dx − v dy + i v dx + u dy
C C
(1467)
248
The integration consists of two line integrals. Each line integral can be thought
of as a integration of a scalar product of a vector with the vector displacement
around the loop in the (x, y) plane. That is, the integration can be considered
as an integration representing the work performed by a force.
The real part of integration involves the scalar product of a vector

→
−
A = ê1 u − ê2 v (1468)
and a displacement. The displacement is represented by
d→
−
r = ê1 dx + ê2 dy (1469)
and the integration is given by
I
→
−
d→
−
r . A (1470)
C
The contour integration can be evaluated by Stokes’s theorem

I Z
→
− →
− →
− →
−
d→−
r . A = d2 S . ( ∇ ∧ A ) (1471)
C
in terms of an integral inside the area of the (x, y) enclosed by the loop or
I Z Z
∂v ∂u
u dx − v dy = − dx dy + (1472)
C ∂x ∂y
The imaginary part can be evaluated in the same way, but this time the
→
−
vector B in the loop integral
I
→
−
d→
−r . B (1473)
C
is identified as
→
−
B = ê1 v + ê2 u (1474)
leading to
I Z Z
∂u ∂v
v dx + u dy = dx dy − (1475)
C ∂x ∂y
where the last integral is evaluated over the area enclosed by the loop.
Since the functions u and v satisfy the Cauchy conditions inside the region
enclosed by the loop, both integrands vanish. Hence, we have Cauchy’s integral
theorem I
dz f (z) = 0 (1476)
C
249
if f (z) is analytic at all points enclosed by the loop.
Since the contour integration around a loop contained entirely in a simply

connected region where the function is analytic, one finds that the integration
along two segments C1 and C2 of our arbitrary closed loop cancel
I Z B Z A
dz f (z) = dz f (z) + dz f (z) = 0 (1477)
C A C1 B C2
Then, on reversing the direction of the integration on one segment, the integrals
in the same direction are equal
Z B Z B
dz f (z) = dz f (z) (1478)
A C1 A C2
Since the integral of the analytic function between A and B is independent of

the path, the integral can only depend upon the end points. Hence, we have
Z B
dz f (z) = F (zB ) − F (zA ) (1479)
A
which is independent of the path of integration (as long as it is entirely con-

tained inside the simply connected region where f (z) is analytic).
Multiply Connected Regions.
A contour that lies within a simply connected region can be continuously

shrunk to a point, without leaving the region. In a multiply connected region
there exists contours that can not be shrunk to a point without leaving the
region. Cauchy’s integral theorem is not valid for a contour that can not be
shrunk to a point without the contour leaving the region of analyticity.
A multiply connected region can be reduced to a simply connected region

by introducing one or more lines connecting the disjoint regions in which the
function is non-analytic, and demanding that the contours can not cross the
lines. That is the lines are considered to cut the plane, so that the region is
simply connected. Due to the cut lines, Cauchy’s theorem can be applied.
————————————————————————————————–
Example:
Consider the integral I

1
dz (1480)
C z(z + 1)
250
around a contour C circling the origin counterclockwise with | z | > 1.
Solution:
The Cauchy Riemann conditions are not satisfied at two point z = and
z = − 1 where the derivative is not defined. The function is analytic at
all points in the surrounding area. The region is multiply connected. Two
cut lines can be chosen linking the singularities to infinity, making the region
simply connected. We shall choose the line running along the positive imagi-
nary axis from the origin and the parallel line running from z = − 1 to infinity.
Cauchy’s theorem can be applied to the contour C and its completion run-
ning along oppositely directed segments on each side of the line cuts extending
to the singularities, and two small circles C1 and C2 of radius r circling the
singularities in a clockwise direction.
Then, from Cauchy’s theorem one has

I I I
1 1 1
dz + dz + dz = 0 (1481)
C z ( z + 1 ) C1 z ( z + 1 ) C2 z ( z + 1)
since the contour is in a singly connected region in which the function is ana-
lytic. The contributions from the anti-parallel line segments cancel in pairs, as
the function is single-valued on these lines.
The integral around C1 enclosing z = − 1 in a clockwise direction is

evaluated on the contour
z = − 1 + r exp[ i θ ] (1482)
and θ runs from 0 to − 2 π. The integral around C1 is given by the limit r → 0

I Z −2π
1 1
dz = lim i dθ r exp[ i θ ]
C1 z ( z + 1 ) r→0 0 ( − 1 + r exp[ i θ ] ) r exp[ i θ ]
Z −2π
1
= lim i dθ
r→0 0 ( − 1 + r exp[ i θ ] )
= i2π (1483)
The integral around the contour C2 running clockwise around the origin is
evaluated along the curve
z = r exp[ i θ ] (1484)
251
and θ runs from 0 to − 2 π. The integral is evaluated as
I Z −2π
1 1
dz = lim i dθ r exp[ i θ ]
C2 z ( z + 1 ) r→0 0 r exp[ i θ ] ( 1 + r exp[ i θ ] )
Z −2π
1
= lim i dθ
r→0 0 ( 1 + r exp[ i θ ] )
= −i2π (1485)
Thus, we have the final result

I
1
dz + 2πi − 2πi = 0 (1486)
C z(z + 1)
or I
1
dz = 0 (1487)
C z(z + 1)
This can be verified by explicitly evaluating the integral over a circular contour
z = R exp[ i θ] of very large radius, R → ∞. In this case one has
I Z 2π
1 i
dz ∼ dθ exp[ − i θ ] + O(R−2 )
C z(z + 1) R 0
= 0 (1488)
————————————————————————————————–
Example:
Consider the integral I

2z + 1
dz (1489)
C z(z + 1)
around a contour C circling the origin counterclockwise with | z | > 1.
Solution:
The Cauchy Riemann conditions are not satisfied at two point z = and
z = − 1 where the derivative is not defined. The function is analytic at all
points in the surrounding area. The region is multiply connected. Two cuts
can be chosen linking the singularities to infinity, cutting the region so that it
is simply connected. We shall choose the cut running along the positive imagi-
nary axis from the origin to i ∞ and a parallel cut running from z = − 1 to i ∞.
252
Cauchy’s theorem can be applied to the contour C and its completion run-
ning along oppositely directed segments on each side of the cuts, and two small
circles C1 and C2 of radius r circling the singularities in a clockwise direction.
Then, from Cauchy’s theorem one has

I I I
2z + 1 2z + 1 2z + 1
dz + dz + dz = 0 (1490)
C z(z + 1) C1 z(z + 1) C2 z(z + 1)
since the contour is in a singly connected region in which the function is ana-
lytic. The contributions from the anti-parallel line segments cancel in pairs, as
the function is single-valued on these lines.
The integral around C1 enclosing z = − 1 in a clockwise direction is

evaluated on the contour
z = − 1 + r exp[ i θ ] (1491)
and θ runs from 0 to − 2 π. The integral around C1 is given by the limit r → 0

Z −2π
− 1 + 2 r exp[ i θ ]
I
2z + 1
dz = lim i dθ r exp[ θ ]
C1 ( z + 1 ) r→0 0 ( − 1 + r exp[ i θ ] ) r exp[ i θ ]
Z −2π
− 1 + 2 r exp[ i θ ]
= lim i dθ
r→0 0 ( − 1 + r exp[ i θ ] )
= −i2π (1492)
The integral around the contour C2 running clockwise around the origin is
evaluated along the curve
z = r exp[ i θ ] (1493)
and θ runs from 0 to − 2 π. The integral is evaluated as
I Z −2π
2z + 1 1 + 2 r exp[ i θ ]
dz = lim i dθ r exp[ θ ]
C2 z(z + 1) r→0 0 r exp[ i θ ] ( 1 + r exp[ i θ ] )
Z −2π
1 + 2 r exp[ i θ ]
= lim i dθ
r→0 0 ( 1 + r exp[ i θ ] )
= −i2π (1494)

I
2z + 1
dz − 2πi − 2πi = 0 (1495)
C z(z + 1)
253
or I
2z + 1
dz = 4πi (1496)
C z(z + 1)
————————————————————————————————–
13.3 Cauchy’s Integral Formula

We shall assume that f (z) is a function that is analytic in a simply connected
region containing a contour C. Cauchy’s integral formula states that
I
1 f (z)
dz = f (z0 ) (1497)
2πi C z − z0
if z0 is a point inside the contour C.
Although the function f (z) is analytic in the region, the function

f (z)
(1498)
z − z0
is not analytic everywhere inside the region. The function diverges at z = z0
if f (z0 ) 6= 0 and the function and derivative is not defined at z0 . Cauchy’s
theorem can not be applied until the region is converted into a simply con-
nected region. This is performed by introducing a line from z0 to ∞, which
runs through the region of analyticity.
A contour can be constructed that runs around a contour C that almost

encloses the point z0 , but continuous along both sides of the line towards z0 and
traverses around a small circle C” of radius r in the opposite sense of rotation.
Thus, the contour excludes the point z0 .
The contour integration lies within the simply connected region of analyticity
and, thus, Cauchy’s theorem can be applied. The integral is evaluated as four
segments
I Z z+
f (z) f (z)
0 = dz + dz
C z − z0 z z − z0
I Z0 z0
f (z) f (z)
+ dz + dz (1499)
C” z − z0 z− z − z0
Since f (z) is single-valued, the integral along both sides of the line cancel. Hence
we have
I I
f (z) f (z)
0 = dz + dz (1500)
C z − z0 C” z − z0
254
On reversing the direction of the contour C” around z0 such that C” = − C 0 ,
then I I
f (z) f (z)
dz = dz (1501)
C z − z 0 C 0 z − z0
where both contours are traversed in the same sense of rotation. The contour
C 0 is evaluated over a path of radius around z0 , i.e.
z = z0 + r exp[ i θ ] (1502)
We shall assume that the contour C and hence C 0 both run in a counter clockwise
direction, so that θ runs from 0 to 2 π. Hence, in the limit r → 0 one has
I Z 2π
f (z) f ( z0 + r exp[ i θ ] )
dz = i dθ r exp[ i θ ]
C z − z 0 0 r exp[ iθ ]
Z 2π
= i dθ f (z0 )
0
= 2 π i f (z0 ) (1503)
since f (z) is analytic and continuous at z0 . Thus, Cauchy’s integral theorem

has been proved.
Cauchy’s theorem can be expressed as

I
1 f (z) f (z0 ) z0 interior
dz = (1504)
2πi C z − z0 0 z0 exterior
corresponding to z0 either inside or outside the contour C.
————————————————————————————————–
Example:
Evaluate the integral

I I
1 f (z) 1 z + 3
dz = dz (1505)
2πi C z − 1 2πi C z − 1
over two contours which run counterclockwise, the contour C 0 enclosing the
point z = 1 and another C” which does not enclose z = 1.
Solution:
From Cauchy’s integral formula one obtains the result

C0
I
1 f (z) 4
dz = (1506)
2πi C z − 1 0 C”
255
which can be verified by explicit integration.
Thus, Cauchy’s integral formula avoids any need to explicitly construct and
evaluate the integral.
————————————————————————————————–
13.4 Derivatives
Cauchy’s integral formula may be used to express the derivative of a function
f (z) at a point z0 in a region which is analytic. The point z0 is assumed to be
enclosed by a loop C contained inside the region of analyticity.
The derivative is defined as

f (z0 + ∆z) − f (z0 )
lim (1507)
∆z →0 ∆z
on using Cauchy’s integral formula at z0 + ∆z
I
1 f (z)
dz = f (z0 + ∆z) (1508)
2πi C z − z0 − ∆z
and z0 I
1 f (z)
dz = f (z0 ) (1509)
2πi C z − z0
So one has
" #
f (z0 + ∆z) − f (z0 )
I
1 f (z) f (z)
lim = lim dz −
∆z →0 ∆z ∆z → 0 2 π i ∆z C z − z0 − ∆z z − z0
I " #
1 f (z)
= lim dz
∆z → 0 2 π i C ( z − z0 − ∆z) ( z − z0 )
I " #
1 f (z)
= dz (1510)
2πi C ( z − z0 )2
Thus, we have the derivative given by the contour integral

I " #
df (z0 ) 1 f (z)
= dz (1511)
dz0 2πi C ( z − z0 )2
where C encloses z0 , and f (z) is analytic at all points of the region enclosed by C.
256
The higher order derivatives can also be evaluated in the same way. This
procedure leads to
" #
dn f (z0 )
I
n! f (z)
= dz (1512)
dz0n 2πi C ( z − z0 )n+1
which is basically the n-th order derivative of the Cauchy integral formula with
respect to z0 .
Thus, Cauchy’s integral formula proves that if f (z) is analytic then all the
derivatives are also analytic.
Example:
Explicitly evaluate the integral

zn
I
1
dz (1513)
2πi C ( z − z0 )2
where the integral runs over a clockwise contour enclosing z0 and compare the
result with Cauchy’s integral formula for the derivative.
The integral can be evaluated on the circular contour
z = z0 + r exp[ i θ ] (1514)
traversed in the counter clockwise direction. The integral is evaluated through

the binomial expansion of z n about z0
n
2π z 0 + r exp[ i θ ]
zn
I Z
1 1
dz 2
= dθ
2πi C ( z − z0 ) 2π 0 r exp[ i θ ]
Z 2π n
1 X
= dθ C(n, m) z0n−m rm−1 exp[ i ( m − 1 ) θ ]
2π 0 m=0
n
X
= C(n, m) z0n−m rm−1 δm,1
m=0
= C(n, 1) z0n−1
= n z0n−1
dz0n
= (1515)
dz0
since only the term with m = 1 is non-zero. The result coincides with the
derivative of z n evaluated at z0 .
257
Example:

I
1 1
dz (1516)
2πi C z2 ( z + 1 )
over a contour that contains the point z = 0 but excludes the point z = − 1.
This can be evaluated by using Cauchy’s integral formula for derivatives as

I
1 1 1
dz 2 = − = −1 (1517)
2πi C z (z + 1) ( z + 1 )2 z=0
Alternatively, on expressing the integrand in terms of partial fractions

1 1 1 1
= − + (1518)
z2 (z + 1) z 2 z z + 1
and then performing a direct integration over a circle of radius ε around z = 0
one obtains the result
I
1 1
dz 2 = −1 (1519)
2πi C z (z + 1)
which agrees with the previous result.
Cauchy’s integral formula for derivatives automatically provides the result

of the integral.
13.5 Morera’s Theorem.

Morera’s Theorem is the converse of Cauchy’s theorem. That is, if a function is
continuous in a simply connected region and
I
dz f (z) = 0 (1520)
C
for every closed contour C in the region, then f (z) is analytic in the region.
This is proved by noting that since the integral on every closed contour is
zero, an integral on an open contour (in the region) can only depend upon the
end points Z zf
F (zf ) − F (zi ) = dz f (z) (1521)
zi
258
Furthermore, as Z zf
f (zi )
f (zi ) = dz (1522)
zf − zi zi
one has the identity
zf
F (zf ) − F (zi ) f (z) − f (zi )
Z
− f (zi ) = dz (1523)
zf − zi zi zf − zi
In the limit zf → zi the right-hand-side goes to zero as f (z) is continuous.
Hence, the left-hand-side is equal to zero. Thus,
F (zf ) − F (zi )
lim − f (zi ) = 0 (1524)
zf → zi zf − zi
and
F (zf ) − F (zi ) dF (z)
lim = = f (zi ) (1525)
zf → zi zf − zi dz zi
Since F (z) is analytic, then Cauchy’s integral formula for the derivatives shows
that all the derivatives of F (z) are analytic, in particular the first derivative
f (z) is also analytic.
Cauchy’s Inequality
If one has a Taylor series expansion of f (z)

X
f (z) = Cn z n (1526)
n
and the function is bounded such that

| f (z) | < M (1527)
for all values of z within a radius r from the origin and where M is a positive
constant, then Cauchy’s inequality provides an upper bound on the coefficients
| Cn | rn < M (1528)
This can be proved using the Cauchy integral

I
1 f (z)
Cn = dz n+1 (1529)
2 π i |z|=r z
since only the term proportional to z −1 gives a finite contribution to the integral.
Then I
1 f (z) 2πr
| Cn | = dz n+1 < M (1530)
2 π |z|=r z 2 π rn+1
259
which leads to the inequality
| Cn | r n < M (1531)
Liouville’s Theorem
Liouville’s theorem states that if f (z) is analytic and bounded in the com-
plex plane then f (z) is a constant.
This is easily proved by using the Cauchy inequality.

M
| Cn | < (1532)
rn
and let r → ∞ be an arbitrarily large radius in the complex plane. Then for
n > 0 one has
Cn = 0 (1533)
Hence, the only non-zero coefficient is C0 and the function is just a constant.
Conversely, if the function deviates from a constant the function must have
a singularity somewhere in the complex plane. In the next section we shall con-
sider properties of more general functions of a complex variable z
260
14 Complex Functions
We shall first examine the properties of a function that is analytic throughout
a simply connected region.
14.1 Taylor Series

Consider a function that is analytic in a simply connected region containing two
point z and z0 . Then Cauchy’s integral formula at the point z yields
f (z 0 )
I
1
f (z) = dz 0 0 (1534)
2πi C z − z
The contour C has to enclose z and we choose it such that it also encloses z0 .
Then on writing
f (z 0 )
I
1
f (z) = dz 0 0
(1535)
2πi C ( z − z0 ) − ( z − z0 )
The denominator may be expanded in powers of ( z − z0 ) if the contour is

such that for all points z 0 we have | z − z0 | < | z 0 − z0 |. In this case
the series may be expected to converge. From this one finds the Taylor series
expansion
∞
f (z 0 )
I
1 X
f (z) = dz 0 ( z − z0 )n
2πi C n=0
( z0 − z0 )n+1
∞
X 1 dn f (z0 )
= n ( z − z0 )n (1536)
n=0
n! dz 0
where Cauchy’s integral formula for the n-th order derivative has been used. If
f (z) is analytic at z0 the Taylor expansion exists and is convergent for z suffi-
ciently close to z0 .
Schwartz Reflection
The Schwartz reflection principle states that if a function f (z) is analytic

over a region that contains the real axis, and f (z) is real when z is real, i.e,
z = x, then
f (z ∗ ) = f ∗ (z) (1537)
That is, in this case the complex conjugate of the function of z is a function of
the complex conjugate variable z ∗ .
261
The Schwartz reflection principle can be proved by using the Taylor expan-
sion about some point x0 on the real axis. Then
X 1 dn f (x0 )
f (z) = ( z − x0 )n (1538)
n=0
n! dxn0
Since f (x) is real on the real axis, then so are all the derivatives. Hence,
X 1 ∗ n
∗ n d f (x0 )
f (z) = ( z − x0 )
n=0
n! dxn0
X 1 dn f (x0 )
= ( z ∗ − x0 )n
n=0
n! dxn0
= f (z0∗ ) (1539)
where the last line is obtained by identifying the Taylor expansion as f (z) with
the replacement z = z ∗ .
14.2 Analytic Continuation

If two functions f1 (z) and f2 (z) are analytic in a region R and agree in a smaller
region R0 that is contained in R, then the functions are equal everywhere in R.
This allows the function f1 (z) which is only known explicitly in some smaller
region to be found in a much larger region.
A function f (z) may be analytic in some region of space, around a point z0 .

The region, by definition does not include the closest singularity, which may be
at zs . In the region around z0 the function may be Taylor expanded, however it
can not be expanded for values of | z − z0 | which are larger than | zs − z0 |
since the contour C involved in the proof of Taylors theorem must encloses a
point z 0 which is nearer to z0 than the singularity at zs . That is there is a radius
of convergence which is given by the distance from z0 to the closest singularity
zs .
However, since f (z) is analytic inside the radius of convergence we may

choose to Taylor expand about another point, z2 inside the radius of conver-
gence of the expansion about z0 . The point z2 can be chosen to be further from
the singularity than z0 is. Hence, the expansion about z2 may have a larger
radius of convergence.
Thus we have a region in which the both Taylor series converge, and yield the
function. However, the second expansion defines the function at points which
lie outside the original circle of convergence. Thus, the function f (z) has been
262
expressed as an analytic function in a region that contains points not included
in the original circle. Thus, the analytic properties of the function has been
continued into a different region of the complex plane.
This process is known as analytic continuation and may be repeated many

times.
It should be noted that analytic continuation can take many different forms
and is not restricted to series expansion.
Example:
Consider the function

1
f (z) = (1540)
z + 1
which can be expanded around z0 = 0 as
∞
X
f (z) = ( − 1 )n z n (1541)
n=0
and the radius of convergence is determined by the distance to the nearest sin-
gularity, which is at zs = − 1. The radius of convergence of the expansion
about z0 = 0 is unity. The Taylor expansion converges for all the points inside
the unit circle.
This function can be analytically continued, by expanding about any other

point inside the circle of radius 1 from the origin.
Let us first choose to expand about the new point z2 = + 1. The function
can be expanded about z2 = 1 by writing
1
f (z) = (1542)
2 + (z − 1)
The series is found as
∞
X 1
f (z) = ( − 1 )n ( z − 1 )n (1543)
n=0
2n+1
and has a radius of convergence determined by the distance to the closest sin-
gularity which is at zs = −1, so the radius of convergence is 2. Thus, the
analytic properties of the function been extended to a circle of radius 2. The
new circle of convergence includes the point z = 3. We can now iterate the
process, an define the function over the whole positive real axis.
263
Alternatively, one may wish to examine the function on the negative real
axis, or some other region. In this case, one might choose another point inside
the original circle of convergence, say z2 = i. The expansion about this point
is expanded about z2 = 1 via
1
f (z) = (1544)
1 + i + (z − i)
as
∞
X 1
f (z) = ( − 1 )n ( z − i )n (1545)
n=0
( 1 + i )n+1
which has a radius of convergence determined by the distance to the closest
singularity which is at zs = −1. The series converges within the region
√
| z − i | < | 1 + i | = | zs − i | = 2 (1546)
√
so the radius of convergence is 2. By choosing a suitable series of overlapping
circles, one can analytically continue the function to the negative real axis, but
the circles always have to exclude the point z = − 1.
It should be noted that a function f (z) may have more than one singularity,
in which case the circle of convergence for an expansion about some point is
limited by the closest singularity.
Example:
A function that can not be analytic continued is given by the Taylor expan-
sion about z = 0, of the function
∞
X n
f (z) = 1 + z2
n=0
= 1 + z 2 + z 4 + z 8 + z 16 + . . . (1547)
It can be seen that the function satisfies the relation
f (z) = z 2 + f (z 2 ) (1548)
Furthermore, at the point z = 1 one has
f (1) = 1 + f (1) (1549)
which makes f (z) singular at z = 1. Therefore, since
f (z) = z 2 + f (z 2 ) (1550)
264
and the term f (z 2 ) on the right-hand-side is singular when z 2 = 1, one finds
that f (z) is singular at all the roots of z 2 = 1. This process can be iterated to
m
show that f (z) is singular at all the roots of z 2 = 1 for each integer value of
m. There are 2m such singularities for each integer value of m and the singulari-
ties are uniformly distributed on the unit circle. The singularities corresponding
to m → ∞ are separated by infinitesimally small distances. Hence, the ra-
dius of convergence for z values on the unit circle is infinitesimally small. Thus,
it is impossible to analytically continue the function f (z) outside the unit circle.
14.3 Laurent Series

Some functions are analytic in a finite area but is not analytic in some inner
region. That is the area of analyticity may not be simply connected. For
example, one may have
R0
f (z) = + 2 + z2 + . . . (1551)
z − z0
in which case one can not obtain a simple Taylor series expansion about the
point z0 .
More generally, the function may be analytic inside a ring shaped area of
inner radius r and outer radius R, where R > r. The the area is not simply
connected. It is possible to introduce a line which cuts the plane making it into
a simply connected region. A particular line can be chosen which runs from
a point on the inner perimeter to a point on the outer perimeter. Since the
area is now simply connected one can apply Cauchy’s integral formula to any
contour in the annular region which does not cross the cut we have chosen. In
particular let us choose a contour that consists of two circular segments which
are concentric, the center of the circles is denoted as the point z0 . The radius
of the circles r1 and r2 are such that R > r1 > r2 > r. The contour is
completed by anti-parallel paths running along opposite sides of the cut.
For any point z inside the singly connected region where f (z) is analytic one
has
f (z 0 ) f (z 0 )
I I
1 0 1
f (z) = dz 0 + dz 0 (1552)
2πi C1 z − z 2πi C2 z0 − z
where the first circular contour is traversed in the counter clockwise direction
and the second contour is traversed in the counterclockwise direction. The con-
tribution from the oppositely directed paths along a segment of the cut has
cancelled, since our function is analytic and single-valued in the entire annular
region. That is the function f (z) has the same value at points immediately
265
adjacent to the cut.
The denominator in both of the integrands of the above expression can be

written as
z 0 − z = ( z 0 − z0 ) − ( z − z0 ) (1553)
and can be expanded. However, it is important to note that for C1 one has
| z 0 − z0 | = r1 > | z − z0 | (1554)
while for the Contour C2 one has
| z − z0 | > | z 0 − z0 | = r2 (1555)
This has the consequence the expansion on contour C1 only converges if it

consists of a power series in !
z − z0
(1556)
z 0 − z0
whereas the expansion on contour C2 is in powers of
!
z 0 − z0
(1557)
z − z0
Hence, we have
∞ I
1 X ( z − z0 )n
f (z) = dz 0 f (z 0 )
2 π i n=0 C1 ( z 0 − z0 )n+1
∞ I
1 X ( z 0 − z0 )n
− dz 0 f (z 0 )
2 π i n=0 C2 ( z − z0 )n+1
∞ ∞
X X Bn
= An ( z − z0 )n + (1558)
n= n=0
( z − z0 )n+1
where
f (z 0 )
I
1
An = dz 0 (1559)
2πi C1 ( z0 − z0 )n+1
in which the contour C1 runs clockwise. One can define
Bn−1 = A−n (1560)
and have A−n given by

I
1
A−n = − dz 0 f (z 0 ) ( z 0 − z0 )n−1 (1561)
2πi C2
266
and the contour C2 runs clockwise. Thus, we have the Laurent expansion of the
function
X∞
f (z) = An ( z − z0 )n (1562)
n=−∞
Example:
The function
1
f (z) = (1563)
z(z − 1)
has a Laurent series expansion about z = 0. The simplest method of obtaining
the expansion consists of expressing the function in partial fractions and then
expanding about z = 0
1 1
f (z) = −
z − 1 z
∞
1 X
= − − zn
z n=0
(1564)
Hence, the non zero coefficients are A−1 and An for n ≥ 0 and are all unity.
A−1 = A0 = A1 = . . . = 1 (1565)
The Laurent series differs from the Taylor series due to the occurrence of nega-
tive powers of ( z − z0 ), which makes the series diverge at z = z0 .
Example:
The Fermi-Dirac distribution function of Statistical Mechanics has the form

1
f (z) = (1566)
exp[z] − 1
find the Laurent expansion about z = 0.
14.4 Branch Points and Branch Cuts

In the above analysis, the functions f (z) were single-valued. That is they have
a unique value at each point in the complex plane. This is ensured by the func-
tion only having positive or negative integral powers in the Laurent or Taylor
expansions. Certain functions of a complex variable are multi-valued, in that
more than one value of f (z) is assigned to each point in the complex plane.
267
As a primitive example, consider the function
1
f (z) = z m (1567)
for integer m. What this equation actually means is that f (z) is described by
the solutions of m
f (z) = z (1568)
The solutions are found by expressing the complex number z in polar form

z = rz exp i θz (1569)
and the solution are found in the same form

f (z) = rf exp i θf (1570)
This leads to the relation between the positive magnitudes
rfm = rz (1571)
which has a unique solution

1
rf = rzm (1572)
However, the phases are not uniquely defined. They are only defined up to a
multiple of 2 π. Hence, on equating the phases one has
m θ f = θz + 2 π n (1573)
where n is an undetermined positive or negative integer. From this we see that

the function is given by

1 θz 2πn
f (z) = rzm exp i exp i (1574)
m m
which involves the undetermined integer n. Hence, there exists more than one
solution. One can see that there are m different solutions corresponding to
n = 0 and every integer up to and including n = m − 1. These solutions
1
form m equally spaced points on a circle of radius rzm in the complex z plane.
1
Thus, the function z m is multi-valued.
In general a function such as z α , where α is either an arbitrary rational or

irrational number, the function is expected to be multi-valued.
1
The function z m has m different values, but on changing z continuously each
of the values change continuously. That is, the function has m branches which
268
are continuous in z. Each branch can be considered to define a single-valued
function, as long as two branches do not give the same value at one point. How-
ever, we see that if z traverses a contour wrapping around the origin exactly
once, then θz increases by exactly 2 π, and one branch merges with the next as
the phase of f (z) has increased by 2mπ . In this case, the origin is a branch point
of the function.
The function can be converted into a single-valued function by introducing a

branch cut and agreeing that the contour shall not cross the branch cut. How-
ever, it should be noted that the function does not have the same value on either
side of the branch cut, as the phases are different by ± 2mπ .
Since a function is not single-valued on any contour that wraps around

z = 0, the branch cut is chosen to start at the branch point z = 0 and
runs to infinity, where the function is not analytic. For convenience we might
choose the cut to run along the positive real axis. In any case, once a branch
cut has been chosen θz is restricted to be greater than 0 and less than 2 π so
there should be no confusion as to which value of the function belongs to which
branch. With this convention, each of the branches define singly valued func-
tions. Again, it should be stressed that the function does not have the same
value on both sides of the branch cut as the phases are different.
If the contour where chosen to cross the branch cut, the function would be
identified with the value on the next branch etc. In this case, one could de-
fine the complex plane to consist of several copies, (in which the phase of z is
advanced by 2 π between otherwise identical points in successive copies). The
function in this enlarged complex plane is single-valued. This manner of ex-
tending the complex plane to different phases, is akin to cutting identical sheets
representing the complex plane, and joining the edge of the cut adjoining the
lower-half complex plane of one sheet to the edge of the cut adjoining the upper-
half complex plane of the next sheet. Thus, one has a spiral of continuously
1
connected copies of the complex plane. The function z m is single-valued on this
spiral sheet, and also repeats after m sheets have been crossed. The m-th sheet
is identified with the zeroth sheet, and this construct is the Riemann surface of
1
z m . The Riemann surface consists of m sheets.
If a contour is chosen to wind around the origin by 2 π then, starting with

the phase given by n = 0, one has
I Z 2π
1 m+1 (m + 1)
dz z m = ir m dθ exp i θ
c 0 m
!
m+1 m 2π
= r m exp i − 1
m + 1 m
269
!
1 m 2π
= z m r exp i − 1
m + 1 m
(1575)
This integral is non zero if m 6= 1 and is zero when m = 1 for which the func-
tion is single-valued. The multi-valued nature of functions can make Cauchy’s
theorem inapplicable.
Example:
Another important example is given by
f (z) = ln z (1576)
or in polar coordinates
f (z) = ln r + i θ (1577)
Thus, ln z is infinitely multi-valued. Furthermore, it is the multi-valued nature
of ln z that makes the integral
I
dz
= ln z = 2 π i (1578)
z
when integrating around a contour circling about the origin once.
Example:
The function
1
f (z) = ( z 2 − 1 ) 2 (1579)
can be factorized as
1 1
f (z) = ( z − 1 ) 2 ( z + 1 ) 2 (1580)
The first factor has a branch point at z = 1 and the second factor has a branch
point at z = − 1. The branch cut has to connect the two branch points. For
convenience one can represent

z − 1 = r1 exp i ϕ1 (1581)
and
z + 1 = r2 exp i ϕ2 (1582)
Then the phase of f (z) is given by

1
( ϕ1 + ϕ2 ) (1583)
2
270
The introduction of the branch cut makes the function single-valued. Con-
sider a contour starting at + 1 running down to − 1 in the upper-half complex
plane and returning to + 1 below in the lower-half complex plane just below
the branch cut.
As the contour rotates around the branch point + 1 the phase ϕ1 changes
from 0 to π while ϕ2 = 0. As the contour rounds the branch point − 1, the
phase ϕ2 changes by 2 π. Secondly, as the contour returns to the point + 1 and
circles around it back to the start ϕ1 changes from π to 2 π. In this loop the
total phase changes by 2 π, hence, the function is single-valued.
14.5 Singularities
An point z0 where a function is not analytic or single valued, but is analytical
at all neighboring points is known as an isolated singularity. In this case one
can draw a contour around the singularity on which the function only has no
other singularity within it. The Laurent expansion yields a series which exhibits
isolated singularities.
A function that is analytic in the entire complex plane except at isolated

singularities is known as a meromorphic function. Examples of meromorphic
functions are given by entire functions, ( as they have no singularities in the
finite complex plane ), functions which are ratios of polynomials ( as they only
have isolated singularities ) and functions like tan z and
1
f (z) = (1584)
exp[z] − 1
This function has isolated singularities at
z = i2πn (1585)
for integer n.
Poles
In the Laurent expansion one has

∞
X
f (z) = an ( z − z0 )n (1586)
n=−∞
Let the highest negative order for which the coefficient is non zero be − n. That
is a−n 6= 0, and the higher order terms a−m are zero for − n > − m. In this
271
case one has a pole of order n. In this case, one has
an an−1
f (z) = n
+ + . . . +a0 + a1 ( z − z0 ) + . . . (1587)
( z − z0 ) ( z − z0 )n−1
A pole of order one is called a simple pole. It should be noted that the function
is single valued on a contour going round a pole, as it does return to its original
value after it has completed the contour, and f (z) varies continuously on the
contour.
————————————————————————————————–
Example:
The function
sin z
f (z) = (1588)
z6
has a pole of order 5 as
∞
X z 2n+1
sin z = ( − 1 )n (1589)
n=0
( 2 n + 1 )!
so
∞
sin z X
n z 2n−5
= ( − 1 ) (1590)
z6 n=0
( 2 n + 1 )!
which leads to a pole of order 5.
Essential Singularities
If the Laurent expansion has terms of negative orders running up to − ∞

then the function has an essential singularity. The behavior of a function near
an essential singularity is pathological.
————————————————————————————————–
Example:
The function
1
f (z) = exp (1591)
z
has an essential singularity at z = 0. The function has the expansion
X 1 1 n
f (z) = (1592)
n=0
n! z
272
which contains terms of arbitrarily high negative orders. The pathological be-
havior of this function is seen by examining how the function varies as z tends
to zero. If z approaches zero from the positive axis then f (z) tends to infinity,
on the other hand if z tends to zero along the negative real axis, then f (z) tends
to zero. In fact by choosing an arbitrary path to the point z = 0 one can have
limz→0 f (z) = b for any complex b. For example, on inverting the equation
one can find solutions
1
z = (1593)
2 π i n + ln b
where n are integers, as n approaches to infinity z approaches zero, while the
function remains equal to b.
Branch Points
A point z0 is a branch point of a function if, when z moves around the point
z0 on a contour of small but non-zero radius, the function does not return to
its original value on completing the contour. It is assumed that f (z) varies
continuously on the contour.
————————————————————————————————–
Example:
The function p
f (z) = z 2 − a2 (1594)
has branch points at z = ± a. The function does not return to the same value
if the contour warps around one pole once, say z = a, but does return to the
same value if the contour encloses both branch points.
Singularities at Infinity
If one makes the transformation

1
z = (1595)
ξ
and
f (z) = g(ξ) (1596)
then, if the function g(ξ) has a singularity at ξ = 0 the function has a singu-
larity at infinity.
————————————————————————————————–
273
Example:
The function
f (z) = sin z (1597)
has an essential singularity at infinity since
1
g(ξ) = sin
ξ
∞
X ( − 1 )2n+1 1
= (2n+1)
(1598)
n=0
( 2 n + 1 )! ξ
has terms of all negative orders.
274
15 Calculus of Residues
15.1 Residue Theorem
If a function f (z) which has a Laurent expansion
n=∞
X
f (z) = An ( z − z0 )n (1599)
n=−∞
then if f (z) is integrated around a counterclockwise contour that only encloses

the isolated singularity at z0 one has
I
1
dz f (z) = A−1 (1600)
2πi c
where A−1 is the coefficient of ( z − z0 )−1 in the Laurent expansion is known

as the residue of f (z) at the pole z0 . We shall denote the residue at the polez0
by R(z0 ).
The residue theorem states that the integral of a function around a contour
C that only encloses a set of isolated singularities, then the integral
I
1 X
dz f (z) = R(zn ) (1601)
2πi C n
is just the sum of the residues of the enclosed poles.
The Residue Theorem can be proved by using Cauchy’s theorem on the

counterclockwise contour C supplemented by segments Cn that run from C to
the isolated poles and circle them in the opposite sense of rotation. The straight-
line portion of the segments Cn cancel in pairs, leaving only the contributions
circling the poles in the opposite sense of rotation. The circular integral around
any singularity Cn can be evaluated on an infinitesimal circle around the pole,
leading to the result − R(zn ). The negative sign occurs, because we have
evaluated the integral on a clockwise contour. Hence, Cauchy’s theorem leads
to I
1 X
dz f (z) − R(zn ) = 0 (1602)
2πi C n
Thus, the problem of evaluating the contour integral is reduced to an algebraic

problem of summing the residues.
————————————————————————————————–
Example:
275
Evaluate the integral Z ∞
Γ
dx (1603)
−∞ x2 + Γ2
Solution:
The integral can be evaluated by decomposing the integrand as partial frac-

tions
" #
Γ 1 1 1
f (x) = 2 = − (1604)
x + Γ2 2i x − iΓ x + iΓ
1
which shows the poles at x = ± i Γ with residues ∓ 2 i.
The contour integral can be evaluated by closing the contour on a semi-circle

at infinity in either the upper or lower half complex plane. The contribution
from the semi-circular contour at infinity vanishes since if

z = R exp i θ (1605)
then
Γ
lim | f (z) | ∼ → 0 (1606)
R → ∞ R2
Hence, the contour at infinity is given by
Z π
Γ
R dθ | f (z) | ∼ π (1607)
0 R
which tends to zero as R → ∞.
One completing the contour in the upper-half complex plane one encloses
the pole at z = i Γ in a counter clockwise sense and obtains the final result
Z ∞
Γ 2πi
dx 2 2
= 1 = π (1608)
−∞ x + Γ 2i
If the contour had been evaluated in the lower-half plane, the contour would
enclose the pole at z = − i Γ in a clockwise sense, since the residue at this
pole is negative, and the clockwise contour produces another negative sign, we
obtain the same result.
276
15.2 Jordan’s Lemma
Consider the integral on an open curve CR consisting of the semi-circle of radius
R in the upper-half complex plane. The semi-circle is centered on the origin.
Let the function f (z) be a function that tends to zero as | M
z | as | z | → ∞,
then I
1
lim dz exp i k z f (z) → 0 (1609)
R → ∞ 2 π i CR
where k is a positive number.
Proof
On the contour one has

z = R exp iθ (1610)
so the integral can be expressed as

I
1

2πi dz exp i k z f (z)
C
Z πR
R
= dθ exp − k R sin θ + i k R cos θ f (R exp[ i θ ])
2π 0
Z π
R
< dθ exp − k R sin θ f (R exp[ i θ ])

2π 0
Z π
R 2
< dθ exp − k R θ f (R exp[ i θ ])

2π 0 π
!
M
< 1 − exp − k R (1611)
2πkR
which tends to zero as R → ∞.
It should be noted that any attempt to evaluate the integration on a semi-

circle in the lower complex plane may diverge if k is positive. Likewise, if k
is a negative number the integration may diverge if evaluated on a semi-circle
in the upper-half complex plane of radius infinity, but will tend to zero on the
semi-circle in the lower-half complex plane.
277
15.3 Cauchy’s Principal Value
The Principal Value of an integral of a function f (x) which has only one simple
pole at x0 on the real axis, is defined as
Z +∞ Z x0 −ε Z ∞
Pr dx f (x) = lim dx f (x) + dx f (x) (1612)
−∞ ε→0 −∞ x0 +ε
The integral is independent of ε since the contributions to the integral within ε

of the pole are of the order of the residue and have opposite sign. Furthermore,
as the intervals are symmetrically located about the pole the contributions can-
cel to order ε.
Integrals of this kind can often be evaluated using contour integration.
For example, if the function only has simple poles in the upper-half complex
plane and can be evaluated on the semi-circle CR at infinity in the upper-half
complex plane. The principal value integral can be evaluated by integrating over
a contour that runs from − ∞ to ( x0 − ε ) then follows a small semi-circle of
radius ε around the pole at x0 , joins the real axis at ( x0 + ε ) then follows
the real axis to + ∞ and then completes the contour by following a semi-circle
of infinite radius (R → ∞) in the upper-half complex plane.
Then the principal value integral and the integral over the small semi-circle
of radius ε around x0 is equal to 2 π i times the sum of the residues enclosed
by the contour.
For example, if f (z) is analytic in the upper-half complex plane on the semi-
circle at infinity in the upper-half complex plane we have
Z +∞ I
f (x) f (z)
Pr dx + dz = i π f (x0 ) (1613)
−∞ x − x 0 CR z − x0
It does not matter if the semi-circle of radius ε is taken to be in the upper-half
complex plane or the lower-half complex plane.
If the small semi-circle is chosen in the lower-half complex plane the inte-
gral runs counter clockwise and runs half way around the pole, and contributes
π i f (x0 ) to the left-hand-side, but the right-hand-side has the contribution of
2 π f (x0 ) as the pole at x0 is enclosed by the contour.
On the other hand if the contour is closed in the upper-half complex plane,
the semi-circle is in the clockwise direction and contributes − π i f (x0 ) to the
left-hand-side and the right-hand-side is zero, as the pole at x0 is not enclosed
by the contour.
278
————————————————————————————————–
Example:

sin x
I = dx (1614)
0 x
Solution:
The integral is evaluated as

Z ∞
1 exp[ + i x ] − exp[ − i x ]
I = dx
2i 0 x
Z +∞
1 exp[ i x ]
= Pr dx (1615)
2i −∞ x
However, on using Jordan’s Lemma and the Cauchy Principal value formula one
has Z +∞
exp[ i x ]
Pr dx = i π exp[ 0 ] = i π (1616)
−∞ x
Hence, we have shown that
Z ∞
sin x π
I = dx = (1617)
0 x 2
————————————————————————————————–
An alternate method of evaluating a Cauchy Principal integral consists of

deforming the contour. The contour is deformed to lie ε above the real axis. In
this case, the integral becomes
Z +∞
f (x + iε)
dx (1618)
−∞ x + i ε − x0
on changing variable from x to z one has
Z +∞ Z +∞
f (z + iε) f (z)
dz = dz (1619)
−∞ z − x 0 + i ε −∞ z − x0 + i ε
since f (z) is continuous at all points on the contour.
This is equivalent to the principal value integral and the small semi-circle,
or Z +∞
f (z)
Pr dx − i π f (x0 ) (1620)
−∞ z − x0
279
Hence, we have
Z +∞ Z +∞
f (z) f (z)
dz = Pr dz − i π f (x0 ) (1621)
−∞ z − x0 + i ε −∞ z − x0
which yields the identity
1 1
lim = Pr − i π δ( z − x0 ) (1622)
ε → 0 z − x0 + i ε z − x0
The truth of this result can also be seen by writing
1 z − x0 ε
= − i (1623)
z − x0 + i ε ( z − x0 ) 2 + ε2 ( z − x0 )2 + ε2
in which the first term on the right not only does not diverge at z = x0 but
is zero. This term cuts off at a distance ε from the pole, and thus corresponds
to the Principal value. The second term on the right is a Lorentzian of width ε
centered at x0 . As the integrated weight is π, this represents the delta function
term in the limit ε → 0.
————————————————————————————————–
Example:
Evaluate the function

Z +∞ exp ikx
1
Θ(k) = (1624)
2πi −∞ x − iε
where ε > 0.
Solution:
For k > 0 the integral can be evaluated by completing the contour in

the upper-half complex plane as the semi-circle at infinity vanishes, whereas for
k < 0 the contour can be completed in the lower-half complex plane. If k > 0
we close the contour in the upper-half complex plane, and the integral encloses
the pole at z = iε. Hence, we have

Z +∞ exp i k x I exp i k z
1 1
+ = exp i k ε (1625)
2 π i −∞ x − iε 2 π i CR z − iε
or on using Jordan’s Lemma

Z +∞ exp ikx
1
+ 0 = 1 (1626)
2πi −∞ x − iε
280
Hence, for k > 0 one has
Θ(k) = 1 (1627)
On the other hand, when k < 0, the integral must be closed in the lower-half
complex plane. The contour does not enclose the pole and Jordan’s Lemma
shows the contour at infinity yields a vanishing contribution. Hence for k < 0
one has
Θ(k) = 0 (1628)
Thus, Θ(k) is the Heaviside step function.
————————————————————————————————–
Example:

Z ∞ exp itx
dx (1629)
−∞ x2 − k 2
Solution:
First we write
Z ∞ exp itx
dx (1630)
−∞ (x − k)(x + k)
The integrand has two poles, at x = ± k. The integral can be evaluated as a
principal value integral and the contribution on two small semicircles of radius
ε about x = ± k.
For t > 0 the integral can be evaluated by completing the contour in the
upper-half complex plane, in which case no poles are enclosed. The contour at
infinity yields zero by Jordan’s Lemma. Hence, one has

Z ∞ exp i t x exp − i t k exp + i t k
Pr dx − π i − π i = 0
−∞ x2 − k 2 −2k 2k
(1631)
Hence, for t > 0, we have

Z ∞ exp itx
π sin t k
Pr dx = − (1632)
−∞ x2 − k2 k
281
For t < 0 the integral can be evaluated by completing the contour in the
lower-half complex plane, in which case no poles are enclosed. The contour at
infinity yields zero by Jordan’s Lemma. Hence, one has

Z ∞ exp i t x exp − i t k exp + i t k
Pr dx + π i + π i = 0
−∞ x2 − k 2 −2k 2k
(1633)
Hence, for t < 0, we have

Z ∞ exp itx
π sin t k
Pr dx = + (1634)
−∞ x2 − k2 k
Thus, combining these results we have the single formula

Z ∞ exp i t x
π sin | t | k
Pr dx 2 − k2
= − (1635)
−∞ x k
15.4 Contour Integration

————————————————————————————————–
Example:

Z 2π
dθ
I = (1636)
0 a + b cos θ
Solution:
Using the substitution

z = exp iθ (1637)
282
one can write the integral as a contour integral around the unit circle in the
complex plane
I
dz 1
I = −i b
C z a + 2 ( z + z −1 )
I
2
= −i dz 2 + 2 a z + b
C b z
I
2i 1
= − dz 2 a (1638)
b C z + 2 b z + 1
The denominator has zeroes at

s
2
a a
z = − ± − 1 (1639)
b b
For a > b the solution

s
2
a a
z = − + − 1 (1640)
b b
lies inside the unit circle. As can be seen directly from the coefficients of the
quadratic equation, the product of the solutions is equal to unity. Thus, the
second solution has to lie outside the unit circle. In the case a > b, one has
I
2i 1
I = − dz a
p a
( b ) − 1 ) ( z + ab −
p a
b C (z + b + 2 ( b )2 − 1 )
2i 2πi
= − p a
b 2 ( b )2 − 1
2π
= √ (1641)
a − b2
2
————————————————————————————————–
Example:

Z 2π
I = dθ cos2n θ (1642)
0
by contour integration.
283
Solution:
On the unit circle

z = exp iθ (1643)
one has
2n
z + z −1
I
dz
I = −i
C z 2
C(2n, n)
= 2π
22n
π (2n)!
= (1644)
22n−1 n! n!
where we have used the binomial expansion and noted the integral only picks
up the term which gives rise to the pole at z = 0. The binomial coefficient of
the term z 0 in the expansion is C(2n, n), which leads to the final result.
————————————————————————————————–
Example:

Z 2π
1
dθ (1645)
0 a + b sin θ
with a > b.
Solution:
We shall change variables from θ to z via

z = exp i θ (1646)
so the integral can be written as a contour integral over the unit circle
I
dz 1
I =
i z a + b ( z − z−1 )
2 i
I
2 1
= dz 2 (1647)
b z + 2 i ab z − 1
The denominator has poles at
s
2
a a
z = −i ± i − 1 (1648)
b b
284
The pole at s
2
a a
z = −i + i − 1 (1649)
b b
is inside the unit circle. The integral can be written as
I
2 1
I = dz s s
b 2 2
z + i ab − i a
b − 1 z + i a
b + i a
b − 1
I " #
1 1 1
= √ dz s − s
i a2 − b2 2 2
a a a a
z + i b − i b − 1 z + i b + i b − 1
2π
= √ (1650)
b2 − a2
The last line of the integral is evaluated by noting that only the pole of the first
term is inside the unit circle.
————————————————————————————————–
Example:

Z 2π
dθ
I = (1651)
0 ( a + b cos θ )2
Solution:
Using the substitution

z = exp iθ (1652)
one can write the integral as a contour integral around the unit circle in the
complex plane
I
dz 1
I = −i 2
z

C
a + 2b ( z + z −1 )
I
4z
= −i dz 2
C
b z2 + 2 a z + b
285
I
4i z
= − 2 dz 2 (1653)
b C a
z2 + 2 b z + 1
The denominator has double zeroes at

s
2
a a
z = − ± − 1 (1654)
b b
For a > b the pole at

s
2
a a
z = zα = − + − 1 (1655)
b b
lies inside the unit circle. The other pole is located at the inverse of the position
of the above pole, and therefore lies outside the unit circle. In this case, one has
I
4i z
I = − 2 dz a
p a
( b ) − 1 ) ( z + ab −
p a
b C (z + b + 2 2 ( b )2 − 1 )2
!
8π d z
=

b2 dz ( z + ab + ( ab )2 − 1 )2
p
zα
!
a
p a
+ ( ) 2 − 1 − z
8π b b

=

( z + ab +
p a
b2 ( b )2 − 1 )3

zα
2πa
= 3 (1656)
( a2 − b2 ) 2
where Cauchy’s integral formula for derivatives has been used to evaluate the
contribution from the pole of order 2.
————————————————————————————————–
Example:
The function
2 sin2 ω T
(1657)
π ω2 T
occurs in the derivation of the Fermi-Golden transition rate, in second order
time-dependent perturbation theory. In the limit T → ∞ this becomes a
Dirac delta function expressing energy conservation. Evaluate the integral
Z ∞
sin2 ω T
dω (1658)
−∞ ω2 T
286
Solution:
The integral can be expressed as

∞ Z ∞ Z ∞
1 − cos 2 ω T 1 − exp[ + i ω T ] 1 − exp[ − i ω T ]
Z
dω 2 T
= dω 2 T
+ dω
−∞ 2 ω −∞ 4 ω −∞ 4 ω2 T
(1659)
The two contributions can be expressed as contour integrals, the first can be
closed by the semi-circle at infinity in the upper-half complex plane, and a small
semi-circle of radius r around the double pole
1 − exp[ + i z T ]
I
dz
C 4 z2 T
Z ∞ Z 0
1 − exp[ + i ω T ] 1
= dω + i dθ exp[ − i θ ] − i r T exp[ + i θ ]
−∞ 4 ω2 T π 4rT
= 0 (1660)
where we have Taylor expanded the function about r = 0, and the contour
does not enclose the pole. Hence, we have
Z ∞
1 − exp[ + i ω T ] π
dω 2 T
= i (1661)
−∞ 4 ω 4
Likewise, the other term can be evaluated by closing the contour in the lower-
half complex plane
Z ∞
1 − exp[ − i ω T ] π
dω 2 T
= i (1662)
−∞ 4 ω 4
On adding these two contributions one has the final result
Z ∞
sin2 ω T π
dω 2
= (1663)
−∞ ω T 2
which shows that the function has weight unity. Furthermore, the weight is
distributed in a frequency interval of width given by T1 , so for T → ∞ one
obtains the energy conserving delta function
2 sin2 ω T

lim = δ( ω ) (1664)
T →∞ π ω2 T
————————————————————————————————–
287
Example:

∞
sin3 x
Z
dx (1665)
−∞ x3
Solution;
As the function is analytic at x = 0 the integration along the real axis can
be deformed slightly near the origin in a semi-circle of radius ε in the lower-half
complex plane. This contour does not contribute to the integral in the limit
ε → 0. The integrand can be expressed as
exp[ + 3 i x ] − 3 exp[ + i x ] + 3 exp[ − i x ] − exp[ − 3 i x ]
(1666)
8 i 3 x3
and the contribution to the small semi-circle from the terms with the positive
phases cancel to order ε with those with the negative phases.
The contour can be closed in the upper-half complex plane for the terms with
the positive phases. The contour encloses the pole of order three at x = 0,
while the contour for the term with the negative phases must be closed in the
lower-half complex plane. This contour excludes the pole at x = 0. The end
result is that the integration is determined by the residue at x = 0 which is
evaluated from Cauchy’s formulae for derivatives as
6
(1667)
16 i
Hence, the integral is evaluated as
Z ∞
sin3 x 3π
dx 3
= (1668)
−∞ x 4
————————————————————————————————–
Example:

Z ∞
exp[ a x ]
I = dx (1669)
−∞ exp[ x ] + 1
where 0 < a < 1.
288
Solution:
The integral is convergent as when x → ∞ the fact that a < 1 makes the
integrand tends to zero. Also, when x → − ∞ the integrand also tends to zero
as a > 0.
The integration is evaluated on a contour which runs along the real axis
between (−R, R) then up to R + 2 π i and then runs anti-parallel to the real
axis from R + 2 π i to − R − 2 π i and then back down to − R. This particular
contour is chosen due to the fact that the integration on the upper line has
1 1
= (1670)
exp[ x + i 2 π ] − 1 exp[ x ] − 1
The two vertical segments vanish as R → ∞ as the integrand vanishes as
exp[ − a R ] or exp[ − ( 1 − a ) R ]. Thus, the contour integral is given by
I Z ∞ Z ∞
exp[ a z ] exp[ a x ] exp[ a x + i 2 π a ]
dz = dx − dx
exp[ z ] + 1 −∞ exp[ x ] + 1 −∞ exp[ x ] + 1
Z ∞
exp[ a x ]
= ( 1 − exp[ i 2 π a ] ) dx (1671)
−∞ exp[ x ] + 1
This contour encloses the pole at z = π i, which has a residue given by

exp[ ( a − 1 ) π i ] = − exp[ i π a ] (1672)
Hence, we have
Z ∞
exp[ a x ]
1 − exp[ i 2 π a ] dx = − 2 π i exp[ i π a ]
−∞ exp[ x ] + 1
(1673)
or in final form, one has
Z ∞
exp[ a x ] π
dx = (1674)
−∞ exp[ x ] + 1 sin π a
————————————————————————————————–
Example:
The gamma function is defined by

Z ∞
Γ(x + 1) = dt tx exp − t (1675)
0
289
for x > − 1. Show that
Z ∞
πx
dt tx sin t = Γ(x + 1) cos
0 2
Z ∞
πx
dt tx cos t = − Γ(x + 1) sin (1676)
0 2
Solution:
Consider the contour integral defined by
I
t
dz z exp − z (1677)
C
where C starts at the origin and runs to infinity, then follows a segment of a
circular path until the positive imaginary axis is reached. The contour is closed
by the segment running down the positive imaginary axis back to the origin.
This integral is zero, as no singularities are enclosed

I
t
dz z exp − z = 0 (1678)
C
In addition, due to the presence of the exponential factor

exp − R cos θ (1679)
in the integrand, the integral over the quarter circle of radius R → ∞ vanishes.
Thus,
Z ∞ Z ∞
dx xt exp − x = i dy it y t exp − i y (1680)
0 0
or on using the definition of the Gamma function

Z ∞
πt
Γ(t + 1) = i exp i dy y t exp − i y (1681)
2 0
On multiplying by a factor of

πt
exp − i (1682)
2
one obtains
Z ∞
πt
exp − i Γ(t + 1) = i dy y t exp − i y (1683)
2 0
290
and, on taking the real and imaginary parts, one obtains the results
Z ∞
πx
dt tx sin t = Γ(x + 1) cos
2
Z 0∞
πx
dt tx cos t = − Γ(x + 1) sin (1684)
0 2
as was to be shown.
————————————————————————————————–
Example:

∞
xα
Z
I = dx (1685)
0 1 + x2
for α < 1.
Solution:
The function has a branch point at x = 0. A branch cut can be drawn along
the positive real axis. A contour can be drawn which performs a circle of radius
R at infinity and a counter clockwise contour at radius r and two anti-parallel
segments on opposite sides of the real axis.
The contribution from the circular contour at infinity vanishes as the inte-
grand vanishes as Rα−2 . The contribution from the clockwise contour around
the origin vanishes as rα+1 as r → 0. The contribution from the two anti-
parallel segments do not cancel as the function just below the real axis differs
from the function just above the real axis by the phase exp[ i 2 π α ]. Hence,
the integral around the contour reduces to
Z ∞
zα xα
I
dz 2 = 1 − exp i 2 π α dx 2 (1686)
C z + 1 0 x + 1
The contour encloses the two poles at x = ± i which have residues

( 2 ∓ 1 )
exp i 2 πα
R± = (1687)
±2i
291
Hence, one has
∞ !
xα
Z
πα3πα
1 − exp i 2 π α dx 2 = π − exp i
exp i
0 x + 1 2 2
(1688)
and on factoring out exp[ i π α ] from both sides, one has the result
Z ∞ !
xα sin π2α
dx 2 = π
0 x + 1 sin π α
!
π
= (1689)
2 cos π2α
————————————————————————————————–
Example:
Prove that ∞
xα π(1 − α)
Z
dx = (1690)
0 ( 1 + x2 ) 2 4 cos π2α
————————————————————————————————–
Example:

ln x
dx (1691)
0 x2 + a2
Solution:
The integrand has a branch point at x = 0. A branch cut can be introduced

along the positive real axis. The integral can be represented as the sum of
a clockwise semi-circular contour around the branch point of radius r and a
counter clockwise semi-circular contour at infinity of radius R, and an integral
just over the entire real axis. The contour at infinity yields a result of the order
R−1 which vanishes as R → ∞ . The contour around the semi-circle of radius
r has the magnitude r ln r which vanishes as r → 0. The remaining integral
over the real axis results in the expression
I Z ∞ Z ∞
ln z ln ρ 1
dx 2 2
= 2 dρ 2 2
+ i π dρ 2 (1692)
C z + a 0 ρ + a 0 ρ + a2
292
where we have used the polar representation of the complex number z =
ρ exp[ i θ ].
The contour encloses the pole in the upper-half complex plane at z = i a.

The residue at the pole is

1 π
R = ln a + i (1693)
2ia 2
Hence, the contour integration is evaluated as
Z ∞
ln ρ π π π
2 dρ 2 + iπ = ln a + i (1694)
0 ρ + a2 2a a 2
Z ∞
ln x π
dx = ln a (1695)
0 x2 + a2 2a
where a > 0.
————————————————————————————————–
Example:
Prove that
∞
ln2 x π2
Z
dx = (1696)
0 1 + x2 8
————————————————————————————————–
Example:
Show that ∞
( nπ )
Z
1
dx = (1697)
0 1 + xn sin ( nπ )
by integrating over a contour composed of the real axis an arc at infinity of
length R ( 2nπ ), and a straight line back to the origin.
Solution:
The segment of the contour at infinity vanishes. The two straight line seg-
ments are evaluated as
Z ∞ Z ∞
1 2π 1
dx − exp i ds (1698)
0 1 + xn n 0 1 + sn
293
where
2π
z = s exp i (1699)
n
along the second segment. The contour integral is equal to the residue of the
pole at
π
z = exp i (1700)
n
which is enclosed by the contour. The residue has the value

1 1 π
R = = − exp i (1701)
n z n−1 n n
Thus, we have
Z ∞
2π 1 2πi π
1 − exp i dx = − exp i (1702)
n 0 1 + xn n n
which on dividing by the prefactor can be re-written as
Z ∞
( nπ )

1
dx = (1703)
0 1 + xn sin ( nπ )
as was to be shown.
15.5 The Poisson Summation Formula

The sum
∞
X
f (n) (1704)
n=0
can be evaluated as a contour integration using the Poisson summation formula.
The method is based on the fact that the function

exp i 2 π z + 1
1
= (1705)
i tan π z
exp i 2 π z − 1
has simple poles at z = n for positive and negative integer values of n, with
residues
1
Rn = (1706)
πi
The summation can be written as a contour integration, where the contour
encircles the poles and avoids the singularities of f (z). Thus, the summation
can be expressed as
∞ I
X 1 1
f (n) = dz f (z) (1707)
n=0
2 i C tan πz
294
and the contour can be deformed as convenient. If the contour at infinity van-
ishes, the integration can be evaluated along the imaginary axis.
————————————————————————————————–
Example:
Express the sum

∞
X 1
(1708)
n=1
nm
for m ≥ 2, as an integral.
Solution:
One can express the sum as a contour integrations around all the positive
poles of cot π z,
∞ I
X 1 1 1 1
m
= dz m
(1709)
n=0
n 2 i C tan π z z
The function f (z) has no poles except the simple pole at z = 0, so the contour
of integration can be deformed. The integrations can be deformed from C to
C 0 which is an integration along both sides of the positive real axis, running
from 1 to ∞. The small anti-parallel segments cancel, so the infinite number of
circles of C have been joined into the closed contour C 0 . The contour can then
be furthered deformed to an integral parallel to the imaginary axis
1
z = + iy (1710)
2
and a semi-circle at infinity. The contour at infinity vanishes as m ≥ 2. Hence
we have
∞ Z ∞
X 1 i 1
m
= dy tanh π y 1
n=0
n 2 −∞ ( 2 + i y )m
Z ∞ " #
i 1 1
= dy tanh π y −
2 0 ( 12 + i y )m ( 21 − i y )m
Z ∞ " #
1
= dy tanh π y Im 1 (1711)
0 ( 2 − i y )m
Thus, for example with m = 2 one has

∞ Z ∞
X 1 y
= dy tanh π y 1 (1712)
n=0
n2 0 ( 4 + y 2 )2
295
————————————————————————————————–
Example:
Evaluate
∞
X ( − 1 )n
(1713)
n=−∞
( a + n )2
where a is a non-integer number.
Solution:
The integral can be evaluated by contour integration

∞
( − 1 )n
I
X 1 1
2
= dz csc π z (1714)
n=0
( a + n ) 2 i C ( a + z )2
in which the contour runs on both sides of the real axis, but does not enclose
the double pole at z = − a. The function csc π z has poles at z = π n and
has residues
1
( − 1 )n (1715)
π
The contour can be deformed to infinity, and an excursion that excludes the
double pole. The excursion circles the double pole in a clockwise direction. The
contribution from the contour at infinity vanishes. Thus, we find that the sum
is equal to a contribution from the clockwise contour around the double pole.
The residue at the double pole z = − a is
− π csc π a cot π a (1716)
Hence, we have evaluated the sum as

∞
X ( − 1 )n
= π 2 csc π a cot π a (1717)
n=0
( a + n )2
————————————————————————————————–
Example:
Evaluate the sum

∞
X 2x
(1718)
n=1
x2 + n2 π 2
296
Solution:
Let
∞
X 2x
f (z) = (1719)
n=1
x2 + z 2 π 2
then
∞ I
X 2x 1 X
= dz cot π z f (z) − Res π cot π z f (z)
n=−∞
( x2 2 2
+ n π ) 2πi C
poles of f (z)
(1720)
where C is a closed contour enclosing the real axis and the poles of f (z). The
contour integral vanishes as the contour is deformed to infinity since | f (z) | →
0. Hence, as the poles of f (z) are located at
x
z = ±i (1721)
π
and the residues are coth x, one has
∞
X 2x
= 2 coth x (1722)
n=−∞
( x2 + z2 π2 )
or
∞
1 X 2x
coth x − = 2 + n2 π 2
(1723)
x n=1
x
On integrating this relation from x = 0 to x = θ one obtains

∞
θ2
X
ln sinh θ − ln θ = ln 1 + 2 2 (1724)
n=1
n π
or
∞
θ2

sinh θ Y
= 1 + 2 2 (1725)
θ n=1
n π
On analytically continuing this to complex θ, such that θ = i ϕ one obtains a
formulae for sin ϕ as a product
∞
ϕ2
Y
sin ϕ = ϕ 1 − 2 2 (1726)
n=1
n π
in which the zeroes appear explicitly.
————————————————————————————————–
297
15.6 Kramers-Kronig Relations
The Kramers-Kronig relation expresses causality. The response of a system can
be expressed in terms of a response function. The response of the system A(t)
only occurs after the system has been perturbed by an applied field B(t0 ) at an
earlier time t0 . The general relation between the response and applied field is
given by linear response theory and involves the convolution
Z ∞
A(t) = dt0 χ(t − t0 ) B(t0 ) (1727)
−∞
where the response function is defined to be zero for t0 > t, i.e.

χ(t) = 0 t < 0 (1728)
The response χ(t) represents the time variation of A(t) if the applied field, B(t),
consists of a delta function pulse of strength unity occurring at time t0 = 0.
When Fourier transformed with respect to time, the linear response relation
becomes
A(ω) = χ(ω) B(ω) (1729)
which relates the amplitude of a response A(ω) to the application of a time-
dependent applied field with frequency ω and amplitude B(ω). The quantity
χ(ω) is the response function. Causality leads to the response function being
analytic in the upper-half complex plane and on the real frequency axis. The
response function χ(ω) usually satisfies the Kramers-Kronig relations.
Consider a function χ(z) which is analytic in the upper-half complex plane

and real on the real axis. Furthermore, we assume that
lim χ(z) → 0 (1730)
|z| →∞
so that the integral on the semi-circle at infinity in the upper-half complex plane
is zero. The Cauchy integral formula is given by
I
1 χ(z)
χ(z0 ) = dz (1731)
2πi C z − z0
if z0 is in the upper-half complex plane. Since the semi-circle at infinity vanishes
Z ∞
1 χ(z)
χ(z0 ) = dz (1732)
2 π i −∞ z − z0
On varying z0 from the upper-half complex plane to a value ω0 on the real axis,
the integral becomes the sum of the Principal Value and a contour of radius ε
around the pole. Hence,
Z ∞
Pr χ(z) χ(ω0 )
χ(ω0 ) = dz + (1733)
2 π i −∞ z − ω0 2
298
or Z ∞
Pr χ(z)
χ(ω0 ) = dz (1734)
π i −∞ z − ω0
On taking the real and imaginary part of this equation one has the two equations
Z ∞
Pr Im χ(ω)
Real χ(ω0 ) = dω (1735)
π −∞ ω − ω0
and Z ∞
Pr Real χ(ω)
Im χ(ω0 ) = − dω (1736)
π −∞ ω − ω0
which are the Kramers-Kronig relations.
Symmetry Relations
An applied field of frequency ω can usually be expressed in terms of an

amplitude A(ω) or A(−ω), since for a real field A(t) one has
A∗ (ω) = A(−ω) (1737)
Hence, one expects that the response function may also satisfy
χ∗ (ω) = χ(−ω) (1738)
This implies that the real and imaginary parts, respectively, are even and odd
as
χ(ω) = Real χ(ω) + i Im χ(ω) (1739)
so
χ∗ (ω) = Real χ(ω) − i Im χ(ω) (1740)
Hence
χ(−ω) = χ∗ (ω) = Real χ(ω) − i Im χ(ω) (1741)
and thus
Real χ(−ω) = Real χ(ω)
Im χ(−ω) = − Im χ(ω) (1742)
These symmetry relations can be used in the Kramers-Kronig relations
Z ∞
Pr Im χ(ω)
Real χ(ω0 ) = dω
π −∞ ω − ω0
Z ∞ Z ∞
1 Im χ(ω) 1 Im χ(−ω)
= dω − dω
π 0 ω − ω0 π 0 ω + ω0
Z ∞ Z ∞
1 Im χ(ω) 1 Im χ(ω)
= dω + dω
π 0 ω − ω0 π 0 ω + ω0
Z ∞
2 ω Im χ(ω)
= dω (1743)
π 0 ω 2 − ω02
299
and similarly for the imaginary part, one can show that
Z ∞
Pr Real χ(ω)
Im χ(ω0 ) = − dω
π −∞ ω − ω0
Z ∞
2 ω0 Real χ(ω)
= − dω (1744)
π 0 ω 2 − ω02
This representation of the Kramers-Kronig relation is useful as perhaps only ei-
ther the real or imaginary part of the response function may only be measurable
for positive ω.
15.7 Integral Representations

The generating functions can be used to develop an integral representation of
the special functions. For example, if
X
g(x, t) = fn (x) tn (1745)
n
then the function fn (x) can be identified as a contour integral

I
1
fn (x) = dt g(x, t) t−(n+1) (1746)
2πi C
around a contour C running around the origin, but excluding any singularities
of g(x, t).
Hence, one can express the Bessel functions as

I
1 x
Jn (x) = dt exp ( t − t−1 ) t−(n+1) (1747)
2πi 2
The Legendre functions can be expressed as
I
1 1
Pn (x) = dt √ t−(n+1) (1748)
2πi 1 − 2 x t + t2
The Hermite polynomials can be written as
I
1
Hn (x) = dt exp − t2 + 2 x t t−(n+1) (1749)
2πi
and the Laguerre polynomials can be written as

I exp − 1 x−t t
1
Ln (x) = dt t−(n+1) (1750)
2πi 1 − t
300

MM 1

Uploaded by

Copyright:

Available Formats

MM 1

Uploaded by

Document Information

Original Description:

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

MM 1

Uploaded by

Copyright:

Available Formats

Mathematical Methods for Physics

3 Partial Differential Equations 27

4 Ordinary Differential Equations 64

5 Stürm Liouville Theory 93

6 Fourier Transforms 112

7 Fourier Series 119

8 Bessel Functions 133

9 Legendre Polynomials 162

10 Hermite Polynomials 197

11 Laguerre Polynomials 201

12 Inhomogeneous Equations 208

13 Complex Analysis 242

14 Complex Functions 261

It is a quantitative science, and as such the relationships are mathematical.

The laws or principles of physics are usually formulated as differential equa-

It is useful to start by formulating the laws in fixed Cartesian coordinate

Quantities such as scalars and vectors have definite transformation proper-

Scalars are invariant under rotations.

Vectors transform in the same way as a displacement under rotations.

In a different coordinate system, one in which is (passively) rotated through

A specific example is given by the rotation about the z-axis

The new components are given in terms of the old components by

The inverse transformation is given by the substitution θ → − θ,

The scalar product of two vectors is defined as

2.3 The Gradient

with another quantity defined by

An example of the gradient that occurs in physical applications is the rela-

For a point charge of magnitude q the potential is given by

2.4 The Divergence

The divergence is a scalar quantity.

In Cartesian coordinates, the divergence is evaluated from the scalar product

∂ x f (r) ∂ y f (r) ∂ z f (r) x2 + y 2 + z 2 ∂f (r)

then tangential vector is founds as

A physical example of the divergence is provided by the continuity equation

The presence of a non-zero divergence represents a source or sink for the

2.5 The Curl

A physical example of the curl is given by the relationship between a mag-

Another example is that of a magnetic induction field produced by a straight

The Laplacian appears in electrostatic problems. On combining the defini-

with Gauss’s law

which relates the electrostatic potential to the charge density.

Another useful identity is obtained by taking the curl of a curl. It can be

This identity is often used in electromagnetic theory by combining

and the static form of Ampere’s law

2.7 Gauss’s Theorem

Hence, we have proved Gauss’s theorem for our special geometry

The above argument can be extended to volumes of arbitrary shape, by con-

The electric field of a point charge is given by

So one must have

Stokes’s theorem can be proved for a general surface, by subdividing it into a

A physical example of Stokes’s theorem is found in the quantum mechanical

where n is an arbitrary integer and φ0 is the fundamental flux quantum . This

2.9 Non-Orthogonal Coordinate Systems

l2 = x21 + x22 + 2 ê1 . ê2 x1 x2 (69)

The expression for the length can be written as

where g i,j is known as the metric tensor and is given by

g i,j = êi . êj (71)

What is the relationship between the co-variant and contra-variant compo-