Module – 2 : The Sturm-Liouville Theory and Fourier Expansions

Lecture – 8 : Sturm-Liouville Problem

1. Boundary-value Problems

2. Characteristic-value Problems

3. Orthogonal Functions

4. Sturm-Liouville Problems
Introduction
In Module 1, we developed the governing equation for heat transfer by condution, along with the
appropriate boundary conditions.

We have discussed one-dimensional steady-state heat conduction problems and observed that
formulations of such problems involve ordinary differential equations, which can be solved easily.
Formulations of
 two- or three-dimensional steady-state heat conduction problems, and
 one-dimensional or multidimensional unsteady-state heat conduction problems,
will involve partial differential equations.

For solutions of such problems, we will explore various solution schemes, such as
 Separation of variables  Duhamel’s theorem
 Integral transforms  The Green’s function method, etc.

In this Module-2, we will study a set of mathematical tools that will be

used extensively in our subsequent analytical solutions.
Review of Some Mathematical Concepts
A differential equation or a boundary condition is said to be linear if, when rationalized and
cleared of fractions, it contains no products of the dependent variable or its derivatives.

1 𝜕𝑇(𝐫, 𝑡) 𝑞 𝜕𝑇(𝑥, 𝑡) 𝜕 𝜕𝑇(𝑥, 𝑡)

2
=𝛻 𝑇+ Nonlinear 𝜌𝑐 = 𝑘(𝑇) + 𝑞(𝑥, 𝑡)
𝜕𝑡 𝜕𝑥 𝜕𝑥
𝛼 𝜕𝑡 𝑘
Linear 𝜕𝑇(𝐫, 𝑡)
Nonlinear 𝜌𝑐 = 𝛻 ∙ 𝑘(𝑇)𝛻𝑇 𝐫, 𝑡 + 𝑞 𝐫, 𝑡
(constant k, ρ, c) 𝜕𝑡

A linear differential equation or a linear boundary condition is said to be homogeneous if it

is not altered when the dependent variable in the equation is multiplied by a constant.

A homogeneous condition is one in which all nonzero terms in the expression contain
the dependent variable or its derivative.
2
2
𝑑 𝑇(𝑥) 𝑑𝑇
𝑑 𝑇(𝑥) 𝑑𝑇 2
+𝑥 − 𝑥𝑇(𝑥) = 𝑓(𝑥)
𝑥 2
+𝑥 − 𝑥𝑇(𝑥) = 0 𝑑𝑥 𝑑𝑥
𝑑𝑥 2 𝑑𝑥
homogeneous nonhomogeneous
Review of Some Mathematical Concepts
𝑑𝑦(𝑎) 𝑑𝑦(𝑎)
1 𝜕𝑇(𝐫, 𝑡) 𝑞 𝑦 𝑎 = 𝐶1 , = 𝐶2 , 𝛼𝑦 𝑎 + 𝛽 = 𝐶3 Linear, nonhomogeneous
𝑑𝑥 𝑑𝑥
= 𝛻2𝑇 +
𝛼 𝜕𝑡 𝑘 𝑑𝑦(𝑏) 𝑑𝑦(𝑏)
𝑦 𝑏 = 0, = 0, 𝛼𝑦 𝑏 + 𝛽 =0 Linear, homogeneous
𝑑𝑥 𝑑𝑥
Linear, nonhomogeneous
(constant k, ρ, c)
Multiply the dependent variable T
2 2 2
𝜕 𝑇 𝜕 𝑇 𝜕 𝑇 𝑞 by a constant c, take c out of the 𝜕2𝑇 𝜕2𝑇 𝜕2𝑇 𝑞
+ + + =0 + + + =0
𝜕𝑥 2 𝜕𝑦 2 𝜕𝑧 2 𝑘 differentiation signs and divide 𝜕𝑥 2 𝜕𝑦 2 𝜕𝑧 2 𝑐𝑘
Linear, nonhomogeneous through by c to obtain

𝜕2𝑇 𝜕2𝑇 𝜕2𝑇 Multiply T by constant c, 𝜕2𝑇 𝜕2𝑇 𝜕2𝑇

+ + =0 + + =0
𝜕𝑥 2 𝜕𝑦 2 𝜕𝑧 2 then divide through by c 𝜕𝑥 2 𝜕𝑦 2 𝜕𝑧 2
Linear, homogeneous
𝜕𝑇 Multiply T by constant 𝜕𝑇 𝑇∞
−𝑘 = ℎ(𝑇 − 𝑇∞ ) −𝑘 =ℎ 𝑇−
𝜕𝑥 c, divide through by c 𝜕𝑥 𝑐
Linear, nonhomogeneous
Note: If T∞ = 0, this boundary condition becomes homogeneous.
Review of Some Mathematical Concepts
Consider 1D, Steady-state heat conduction in a fin of uniform cross section:

𝑑 2 𝑇 ℎ𝑃 The solution of this equation involves two arbitrary integration constants

2
− 𝑇 − 𝑇∞ = 0 which are determined by two conditions, each specified at one boundary.
𝑑𝑥 𝑘𝐴𝑐
Such problems are called Boundary–value problems.

For initial-value problems, all conditions are specified at one value of the
independent variable.

Solution of this unsteady-state equation requires an

1 𝜕𝑇 𝜕 2 𝑇 𝑞
= + initial condition and two boundary conditions. We will
𝛼 𝜕𝑡 𝜕𝑥 2 𝑘 call this an initial-and-boundary-value problem.
Characteristic-Value Problems
Let us consider the following linear and homogeneous boundary-value problem (homogeneous
DE and BC) defined in [a, b]:
Since the differential equation is linear and homogeneous, its
2
𝑑 𝑦 𝑑𝑦 general solution can be written in terms of two linearly
+ 𝑓1 𝑥 + 𝑓2 𝑥 𝑦 = 0
𝑑𝑥 2 𝑑𝑥 independent solutions of the differential equation (y1(x) and y2(x))
𝑑𝑦(𝑎) as follows where C1 and C2 are two arbitrary nonzero constants.
𝛼1 𝑦 𝑎 + 𝛽1 = 0, 𝛼12 + 𝛽12 > 0
𝑑𝑥 𝑦 𝑥 = 𝐶1 𝑦1 𝑥 + 𝐶2 𝑦2 𝑥
𝑑𝑦(𝑏)
𝛼2 𝑦 𝑏 + 𝛽2 = 0, 𝛼22 + 𝛽22 > 0 Application of the boundary conditions yields the following
𝑑𝑥
set of algebraic equations for the two unknown constants:
One possible solution of Eq (8.1) is
C1 = C2 = 0, leading to the trivial 𝐴11 𝐶1 + 𝐴12 𝐶2 = 0
solution y(x) ≡ 0. Eq. (8.1)
𝐴21 𝐶1 + 𝐴22 𝐶2 = 0
By Cramer’s rule, in order to have a where,
nontrivial solution, the determinant of 𝑑𝑦𝑖 (𝑎)
coefficients of C1 and C2 must vanish. 𝐴1𝑖 = 𝛼1 𝑦𝑖 𝑎 + 𝛽1 = 0, 𝑖 = 1, 2
𝑑𝑥
𝑑𝑦𝑖 (𝑏)
𝐴11 𝐴12 𝐴2𝑖 = 𝛼2 𝑦𝑖 𝑏 + 𝛽2 = 0, 𝑖 = 1, 2
=0 𝑑𝑥
𝐴21 𝐴22
Characteristic-Value Problems
The solution of two- and three-dimensional steady-state and the solution of one-dimensional or multidimensional
unsteady-state heat conduction problems, may be reduced to the solution of:
𝑑2 𝑦 𝑑𝑦
+ 𝑓1 𝑥 + 𝑓2 𝑥 𝑦 = 0
𝑑𝑥 2 𝑑𝑥 where coefficients f1(x) and/or f2(x) depend on a parameter (say λ).
𝑑𝑦(𝑎) Hence the solutions y1(x), y2(x) may also depend on λ.
𝛼1 𝑦 𝑎 + 𝛽1 = 0, 𝛼12 + 𝛽12 > 0
𝑑𝑥
𝐴11 𝐴12
𝑑𝑦(𝑎) In such problems, the determinant
𝛼2 𝑦 𝑎 + 𝛽2 = 0, 𝛼22 + 𝛽22 > 0 𝐴21 𝐴22
𝑑𝑥
may vanish only for certain values of λ (say λ1, λ2, λ3, ….).
These values of λ are called characteristic values or
eigenvalues.
For each such value of λ, the particular solution is called characteristic function
or eigenfunction. Such boundary value-problems are called Characteristic
Value Problem or Eigenvalue Problem.

A boundary-value problem may or may not be a characteristic-value problem.

But a characteristic-value problem is always a boundary-value problem.
Example: Characteristic-Value Problems
Consider the following homogeneous differential equation involving a parameter λ subject to homogeneous
boundary conditions:
𝑑2 𝑦 2
+ 𝜆 𝑦 = 0, 𝑦 0 = 0, 𝑦 𝐿 =0 Eq. (8.2)
𝑑𝑥 2

The general solution of Eq. (8.2) is: 𝑦 = 𝐶1 sin(𝜆𝑥) + 𝐶2 cos(𝜆𝑥)

The use of BC y(0) = 0 gives: C2 = 0 and y = C1 sin (λx) The use of BC y(L) = 0 gives: 0 = C1 sin (λL)

We see that the problem has nontrivial solutions only if λ satisfies the equation 0 = sin (λL).

This is possible if λ is equal to 𝑛𝜋 which are the characteristic

𝜆𝑛 = , 𝑛 = 1,2,3, … .
one of the values of 𝐿 values of the problem.

Thus, the given boundary-value problem has no solution other than the trivial
solution y(x) ≡ 0, unless λ assumes one of the characteristic values λn given above.
Example: Characteristic-Value Problems
𝑛𝜋 which are the characteristic
𝑦 = 𝐶1 sin(𝜆𝑥) + 𝐶2 cos(𝜆𝑥) 𝜆𝑛 = , 𝑛 = 1,2,3, … .
𝐿 values of the problem.

Corresponding to each characteristic value λn, the solution of the problem can be written as:
𝑦𝑛 𝑥 = 𝐴𝑛 𝜙𝑛 (𝑥)
𝑛𝜋
where An is an arbitrary nonzero constant and 𝜙𝑛 (𝑥) = sin 𝑥, 𝑛 = 1,2,3, … .
𝐿
is the characteristic function corresponding to the characteristic value λn
Note that no new solutions are obtained when n assumes negative integer values.
Also, n = 0 leads to the trivial solution y(x) ≡ 0

Now, let us consider the 𝑑2 𝑦 2

− 𝜆 𝑦 = 0, 𝑦 0 = 0, 𝑦 𝐿 =0
boundary-value problem: 𝑑𝑥 2
This boundary-value problem has no solution other than the trivial solution y(x) ≡ 0,
corresponding to λ = 0. Hence there does not exist any set of characteristic values
or characteristic functions for this problem. Thus, a boundary-value problem may or
may not be a characteristic-value problem.
Orthogonal Functions
Two real-valued functions ϕm(x) and ϕn(x) are said to be orthogonal with respect to a weight
function (or density function) w(x) on an interval (a, b) if
𝑏
𝜙𝑚 𝑥 𝜙𝑛 𝑥 𝑤 𝑥 𝑑𝑥 = 0, 𝑚≠𝑛 Eq. (8.2)
𝑎

A set of real-valued functions {ϕn(x); n = 1, 2,...} is called orthogonal with respect

to a weight function w(x) on an interval (a, b) if all the pairs of distinct functions in this
set satisfy the above orthogonality condition (Eq. 8.2).

For example, the set {sin(nπ/L)x; n = 1, 2, 3,...} is an orthogonal set with respect to the
weight function unity on the interval (0, L) because
𝐿
𝑚𝜋 𝑛𝜋
sin 𝑥 sin 𝑥 𝑑𝑥 = 0, 𝑚≠𝑛
0 𝐿 𝐿

which can be verified by direct integration.

11
Orthogonal Functions
We can define an orthogonal set of functions as:
𝑏
0, if 𝑚 ≠ 𝑛
𝜙𝑚 𝑥 𝜙𝑛 𝑥 𝑤 𝑥 𝑑𝑥 =
𝑎
𝑁𝑛 if 𝑚 = 𝑛

where Nn is the norm, or normalization integral of 𝜙𝑛 , defined as:

𝑏
2
𝑁𝑛 = 𝜙𝑛 𝑥 𝑤(𝑥)𝑑𝑥
𝑎

Note that the interval of orthogonality is a ≤ x ≤ b.

Orthogonal functions may take many forms, such as:

 Trigonometric functions
 Bessel functions
 Legendre polynomials
 Hermite polynomials
 Laguerre polynomials

12
 Sturm – Liouville Problems
Consider the following Characteristic-Value Problem, composed of the linear and homogeneous second-order
differential equation of the general form
𝑑 𝑑𝑦
𝑝(𝑥) + 𝑞 𝑥 + 𝜆𝑤(𝑥) 𝑦 = 0
𝑑𝑥 𝑑𝑥

and the two homogeneous linear boundary 𝑑𝑦(𝑎)

𝛼1 𝑦 𝑎 + 𝛽1 = 0, 𝛼12 + 𝛽12 > 0
condition prescribed at the ends of the finite 𝑑𝑥
interval (a, b). 𝑑𝑦(𝑏)
𝛼2 𝑦 𝑏 + 𝛽2 = 0, 𝛼22 + 𝛽22 > 0
𝑑𝑥
Note:

1. The functions p(x), q(x) and w(x) are real valued and continuous (including dp/dx).

2. Also, p(x) and w(x) are positive over the entire interval (a, b), including the end
points, while q(x) ≤ 0 in the same interval.

3. Furthermore, α1, α2, β1 and β2 are given real constants, and λ is an unspecified
parameter, independent of x.

Characteristic-value problems of this type are known as Sturm–Liouville problems.

 Sturm – Liouville Problems
𝑑 𝑑𝑦 𝑑𝑦(𝑎)
𝑝(𝑥) + 𝑞 𝑥 + 𝜆𝑤(𝑥) 𝑦 = 0 𝛼1 𝑦 𝑎 + 𝛽1 = 0, 𝛼12 + 𝛽12 > 0
𝑑𝑥 𝑑𝑥 𝑑𝑥
𝑑𝑦(𝑏)
𝛼2 𝑦 𝑏 + 𝛽2 = 0, 𝛼22 + 𝛽22 > 0
𝑑𝑥

Nontrivial solutions of the above problem exist, in general, for a particular set of values λ1, λ2,..., λn
of the parameter λ. These are the characteristic values, (or the eigenvalues), and the corresponding
solutions are the characteristic functions (or the eigenfunctions) of the problem.

Let λm and λn be any two distinct characteristic values, and ϕm(x) and ϕn(x) be the
corresponding characteristic functions, respectively. These functions satisfy the above
differential equation. It can be shown:

𝑏 𝑏
𝜆𝑛 − 𝜆𝑚 𝜙𝑚 𝑥 𝜙𝑛 𝑥 𝑤 𝑥 𝑑𝑥 = 0 ⟹ 𝜙𝑚 𝑥 𝜙𝑛 𝑥 𝑤 𝑥 𝑑𝑥 = 0, 𝑚≠𝑛
𝑎 𝑎

Thus, the characteristic functions of the Sturm–Liouville problem form an

orthogonal set with respect to the weight function w(x) on the interval (a, b).
Sturm – Liouville Problems: Singular End Points
𝑑 𝑑𝑦 𝑑𝑦(𝑎)
𝑝(𝑥) + 𝑞 𝑥 + 𝜆𝑤(𝑥) 𝑦 = 0 𝛼1 𝑦 𝑎 + 𝛽1 = 0, 𝛼12 + 𝛽12 > 0
𝑑𝑥 𝑑𝑥 𝑑𝑥
𝑑𝑦(𝑏)
𝛼2 𝑦 𝑏 + 𝛽2 = 0, 𝛼22 + 𝛽22 > 0
𝑑𝑥

If it so happens that p(a) = 0, then the characteristic functions corresponding to different

characteristic values will be orthogonal with respect to the weight function w(x) on (a, b),
provided that y(x) and dy/dx are both finite at x = a.

In this case, the boundary condition at x = a is replaced by the requirement that y(x) and dy/dx
be finite at x = a when p(a) = 0.

Similarly, if p(b) = 0, we can then replace the second boundary condition (at x = b) by
the requirement that y(x) and dy/dx be finite at x = b.

If both p(a) = 0 and p(b) = 0, then neither boundary condition (at x = a and x = b) as
specified above is needed to ensure the orthogonality of the characteristic functions,
provided that both y(x) and dy/dx are finite at x = a and x = b.
 Sturm – Liouville Problems: Periodic BC
𝑑𝑦(𝑎)
𝑑 𝑑𝑦 𝛼1 𝑦 𝑎 + 𝛽1 = 0, 𝛼12 + 𝛽12 > 0
𝑝(𝑥) + 𝑞 𝑥 + 𝜆𝑤(𝑥) 𝑦 = 0 𝑑𝑥
𝑑𝑥 𝑑𝑥
𝑑𝑦(𝑏)
𝛼2 𝑦 𝑏 + 𝛽2 = 0, 𝛼22 + 𝛽22 > 0
𝑑𝑥
If p(a) = p(b), then the boundary conditions can be replaced by

𝑦 𝑎 = 𝑦(𝑏) These are called periodic boundary conditions and they are
satisfied, in particular, if the solution y(x) is required to be
𝑑𝑦(𝑎) 𝑑𝑦(𝑏)
= periodic, of period (b – a).
𝑑𝑥 𝑑𝑥

Example: Heat Conduction in a Circular Ring:

We consider a thin wire of length 2L which is

bent into the shape of a circle.

Then at the endpoints we must match the

temperature and the flux.
 Sturm – Liouville Problems
𝑑 𝑑𝑦 𝑑𝑦(𝑎)
𝑝(𝑥) + 𝑞 𝑥 + 𝜆𝑤(𝑥) 𝑦 = 0 𝛼1 𝑦 𝑎 + 𝛽1 = 0, 𝛼12 + 𝛽12 > 0
𝑑𝑥 𝑑𝑥 𝑑𝑥
𝑑𝑦(𝑏)
𝛼2 𝑦 𝑏 + 𝛽2 = 0, 𝛼22 + 𝛽22 > 0
𝑑𝑥

Note:
The characteristic values of the Sturm–Liouville problem are all real and
nonnegative, and the corresponding characteristic functions are real. The
parameter λ can, therefore, be replaced by λ2 with no loss in the generality of
the problem.

In addition, there is only one characteristic function ϕn(x) which corresponds

to each characteristic value λn, except when the periodic boundary conditions
are involved.

More importantly, the characteristic functions form a complete orthogonal

set.
1. Boundary-value Problems

2. Characteristic-value Problems

3. Orthogonal Functions

4. Sturm-Liouville Problems
“Heat Conduction” by Sadik Kakac, Yaman Yener, Carolina P. Naveira-Cotta,
5th Ed, CRC Press

“Heat Conduction” by David W. Hahn, M. Necati Ozisik,

3rd Ed, John Wiley and Sons, Inc
Module – 2 : The Sturm-Liouville Theory and Fourier Expansions
Lecture – 9 : Fourier Series
1. Expansion of Arbitrary Functions in Series of Orthogonal Functions

2. Generalized Fourier Series

3. Fourier Sine and Fourier Cosine Series

4. Complete Fourier Series

 Generalized Fourier Series
Consider a set of functions {ϕn(x); n = 0, 1, 2,...} orthogonal with respect to a weight function w(x) on the finite
interval (a, b). We wish to expand an arbitrary function f(x) into a series of these orthogonal functions as

𝑓 𝑥 = 𝐴0 𝜙0 𝑥 + 𝐴1 𝜙1 𝑥 + … +𝐴𝑛 𝜙𝑛 𝑥 + ⋯ Note: The coefficients An are unknown

∞

⟹𝑓 𝑥 = 𝐴𝑛 𝜙𝑛 𝑥 Assuming that such an expansion exists, the unknown coefficients An

can be evaluated by multiplying both sides by ϕn(x)w(x) and then
𝑛=0
integrating the resulting equation over the interval (a, b):
𝑏 𝑏 ∞

𝑓 𝑥 𝜙𝑛 𝑥 𝑤 𝑥 𝑑𝑥 = 𝐴𝑘 𝜙𝑘 𝑥 𝜙𝑛 𝑥 𝑤 𝑥 𝑑𝑥
𝑎 𝑎 𝑘=0

𝑏 𝑏 All of the terms on the

Use orthogonal 2
𝑓 𝑥 𝜙𝑛 𝑥 𝑤 𝑥 𝑑𝑥 = 𝐴𝑛 𝜙𝑛 (𝑥) 𝑤 𝑥 𝑑𝑥 RHS vanish, except the
property: 𝑎 𝑎
one for which k = n.
𝑏 𝑏
1 2
Thus we get, 𝐴𝑛 = 𝑓 𝑥 𝜙𝑛 𝑥 𝑤 𝑥 𝑑𝑥 where 𝑁𝑛 = 𝜙𝑛 (𝑥) 𝑤 𝑥 𝑑𝑥
𝑁𝑛 𝑎 𝑎
normalization integral
of the function ϕn(x)
 Generalized Fourier Series ∞
𝑓 𝑥 = 𝐴0 𝜙0 𝑥 + 𝐴1 𝜙1 𝑥 + … +𝐴𝑛 𝜙𝑛 𝑥 + ⋯ ⟹𝑓 𝑥 = 𝐴𝑛 𝜙𝑛 𝑥
𝑏 𝑛=0
1 𝑏
𝐴𝑛 = 𝑓 𝑥 𝜙𝑛 𝑥 𝑤 𝑥 𝑑𝑥 𝑁𝑛 = 𝜙𝑛 (𝑥) 2
𝑤 𝑥 𝑑𝑥
𝑁𝑛 𝑎 𝑎

The expansion of f(x) in orthogonal functions in 𝜙𝑛 𝑥 with coefficients as defined above, is a formal
series representation of the function f(x) on the interval (a, b) and is referred to as the generalized
Fourier series (or expansion) of f(x) corresponding to the orthogonal set {ϕn (x); n = 0, 1, 2,...}.

The coefficients An are called Fourier constants of the function f(x).

If f(x) is a piecewise-differentiable function in the interval (a, b), and {ϕn(x); n = 0, 1, 2,...} is
the complete set of all the characteristic functions of a Sturm–Liouville problem, then the
series representation of f(x) converges inside (a, b) to f(x) at all points of continuity and
converges to the mean value [f(x+) + f(x–)]/2 of f(x) at any discontinuity.

In addition, this series representation may or may not converge to the given value
of f(x) at one or both end points x = a and x = b.
 Ordinary Fourier Series
The ordinary Fourier series, or simply the Fourier series, are developed from the characteristic functions of
the following characteristic-value problem for different combinations of the boundary conditions:
2
𝑑 𝑑𝑦
𝑑 𝑦 𝑝(𝑥) + 𝑞 𝑥 + 𝜆𝑤(𝑥) 𝑦 = 0
2
+𝜆 𝑦 𝑥 =0 𝑑𝑥 𝑑𝑥
𝑑𝑥 2
𝑑𝑦(0) This characteristic-value problem is a special case of the
2 2
𝛼1 𝑦 0 + 𝛽1 = 0, 𝛼1 + 𝛽1 > 0 Sturm–Liouville system with p(x) = 1, q(x) = 0, w(x) = 1,
𝑑𝑥
and λ replaced by λ2 for convenience (as the system can
𝑑𝑦(𝐿) not have any negative eigenvalue)
𝛼2 𝑦 𝐿 + 𝛽2 = 0, 𝛼22 + 𝛽22 > 0
𝑑𝑥
The characteristic functions ϕn(x) will therefore form a complete orthogonal set in the interval (0, L)
with respect to the weight function unity.
Let the functions ϕn(x) are the characteristic functions (orthogonal) of the characteristic-value
problem. The ordinary Fourier series, or simply a Fourier series, of f(x) on the interval (0, L):
∞
𝐿
1
𝑓 𝑥 = 𝐴𝑛 𝜙𝑛 𝑥 , 0<𝑥<𝐿 with 𝐴𝑛 = 𝑓 𝑥 𝜙𝑛 𝑥 𝑑𝑥
𝑁𝑛 0
𝑛=0
𝐿
2
𝑁𝑛 = 𝜙𝑛 (𝑥) 𝑑𝑥
0
 Ordinary Fourier Series: How Many?
𝑑2 𝑦 There are nine different combinations of the
2
+ 𝜆 𝑦 𝑥 =0 boundary conditions.
𝑑𝑥 2
𝑑𝑦(0) Accordingly, corresponding to each combination
𝛼1 𝑦 0 + 𝛽1 = 0, 𝛼12 + 𝛽12 > 0
𝑑𝑥 there will be a series expansion of the form given
𝑑𝑦(𝐿) by:
𝛼2 𝑦 𝐿 + 𝛽2 = 0, 𝛼22 + 𝛽22 > 0 ∞
𝑑𝑥
𝑓 𝑥 = 𝐴𝑛 𝜙𝑛 𝑥 , 0<𝑥<𝐿
𝑛=0

𝐿
1
𝐴𝑛 = 𝑓 𝑥 𝜙𝑛 𝑥 𝑑𝑥
Note that different series expansions will 𝑁𝑛 0
have different characteristic functions ϕn(x)
and coefficients An 𝐿
2
𝑁𝑛 = 𝜙𝑛 (𝑥) 𝑑𝑥
0
We will discuss the two most common cases:
 Fourier Sine Series
 Fourier Cosine Series
Odd Functions: Even Functions Neither even nor
odd function

Graphically
 A function is even if its graph is symmetric about the 𝑦-axis.
 A function is odd if its graph is symmetric about the origin.
 A function that is neither even nor odd is not symmetric about the 𝑦-axis or the
origin.
 Fourier Sine Series A function is "odd" when:
𝑓 −𝑥 = −𝑓 𝑥 ∀𝑥
Consider the following characteristic-value problem:
𝑑2 𝑦 Special case of Sturm-
+ 𝜆2 𝑦 𝑥 = 0 BC: y(0) = 0, y(L) = 0 Liouville problem with 𝑓 𝑥 = 𝑥3
𝑑𝑥 2
𝜷𝟏 = 𝟎, 𝜷𝟐 = 𝟎
Characteristic functions and characteristic values:
𝑛𝜋
𝜙𝑛 (𝑥) = sin 𝜆𝑛 𝑥 , 𝜆𝑛 = , 𝑛 = 1,2,3, … .
𝐿
∞
𝑛𝜋 𝐿
1 𝑛𝜋
𝑓 𝑥 = 𝐴𝑛 sin 𝑥, 0 < 𝑥 < 𝐿 Eq. (9.1) 𝐴𝑛 = 𝑓 𝑥 sin 𝑥 𝑑𝑥
𝐿 𝑁𝑛 0 𝐿
𝑛=1 with
𝐿
𝑛𝜋 𝐿
Fourier sine series representation 𝑁𝑛 = sin2 𝑥 𝑑𝑥 =
0 𝐿 2
of f(x) on the interval (0, L)

Note that the RHS of Eq. (9.1) is a periodic function of period 2L, and also is an
odd function of x. Therefore, if Eq. (9.1) converges to f(x) in (0, L), it will also
converge to – f(–x) in (–L, 0). In other words, if f(x) is an odd function of x, then
Eq. (9.1) will represent f(x) not only in (0, L), but also in (– L, L).
 Fourier Cosine Series
This is another special case of the
Consider the following characteristic-value problem: Sturm–Liouville problem with α1 = α2 = 0.
𝑑2 𝑦 𝑑𝑦(0) 𝑑𝑦(𝐿)
2
+ 𝜆2 𝑦 𝑥 = 0 BC: = 0, =0 Therefore, the characteristic functions
𝑑𝑥 𝑑𝑥 𝑑𝑥
form a complete orthogonal set with
Characteristic functions and characteristic values: respect to the weight function unity over
the interval (0, L).
𝑛𝜋
𝜙𝑛 (𝑥) = cos 𝜆𝑛 𝑥 , 𝜆𝑛 = , 𝑛 = 0,2,3, … .
𝐿

Note The eigenfunction ϕn(x) = 1 corresponding to λ0 = 0, is a member of the set of

characteristic functions. Thus, the series expansion can be written as
∞ ∞ Fourier cosine series
𝑛𝜋 𝑛𝜋
𝑓 𝑥 = 𝐴𝑛 cos 𝑥 = 𝐴0 + 𝐴𝑛 cos 𝑥, 0<𝑥<𝐿 representation of f(x)
𝐿 𝐿
𝑛=0 𝑛=1 on the interval (0, L)

with coefficients as:

𝐿
𝑛𝜋 𝐿, 𝑛=0
𝐿
1 𝑛𝜋 𝑁𝑛 = cos 2 𝑥 𝑑𝑥 = 𝐿
𝐴𝑛 = 𝑓 𝑥 cos 𝑥 𝑑𝑥 and 𝐿 , 𝑛 = 1,2,3, …
𝑁𝑛 0 𝐿 0
2
 Fourier Cosine Series A function is "even" when:
f(x) = f(−x) for all x
∞ ∞
𝑛𝜋 𝑛𝜋
𝑓 𝑥 = 𝐴𝑛 cos 𝑥 = 𝐴0 + 𝐴𝑛 cos 𝑥, 0<𝑥<𝐿 Eq. (9.2)
𝐿 𝐿
𝑛=0 𝑛=1

Fourier cosine series representation of

f(x) on the interval (0, L)

Note that the RHS of Eq. (9.2) is a periodic function of period 2L, and also is an even
function of x. Therefore, if Eq. (9.2) converges to f(x) in (0, L), it will also converge to f(–x)
in (–L, 0). In other words, if f(x) is an even function of x, then Eq. (9.2) will represent f(x) not
only in (0, L), but also in (– L, L).
Complete Fourier Series
Any function of x, say F(x), can be written as: If f(x) is odd:

1 1
𝐹 𝑥 = 𝐹 𝑥 − 𝐹(−𝑥) + 𝐹 𝑥 + 𝐹(−𝑥) If f(x) is even: 𝑓 −𝑥 = 𝑓(𝑥)
2 2
Odd function Even function
∞ ∞
𝑛𝜋 𝑛𝜋
𝐹 𝑥 = 𝑎𝑛 sin 𝑥+ 𝑏𝑛 cos 𝑥, −𝐿 < 𝑥 < 𝐿
𝐿 𝐿
𝑛=1 𝑛=0

∞ ∞
𝑛𝜋 𝑛𝜋
⟹ 𝐹 𝑥 = 𝑏0 + 𝑎𝑛 sin 𝑥+ 𝑏𝑛 cos 𝑥, −𝐿 < 𝑥 < 𝐿
𝐿 𝐿
𝑛=1 𝑛=1

∞
𝑛𝜋 𝑛𝜋
⟹ 𝐹 𝑥 = 𝑏0 + 𝑎𝑛 sin 𝑥 + 𝑏𝑛 cos 𝑥 , −𝐿 < 𝑥 < 𝐿
𝐿 𝐿
𝑛=1
 Complete Fourier Series
∞
𝑛𝜋 𝑛𝜋
𝐹 𝑥 = 𝑏0 + 𝑎𝑛 sin 𝑥 + 𝑏𝑛 cos 𝑥 , −𝐿 < 𝑥 < 𝐿
𝐿 𝐿
𝑛=1

The coefficients an, bn, b0 are given as

1. Expansion of Arbitrary Functions in Series of Orthogonal Functions

2. Generalized Fourier Series

3. Fourier Sine and Fourier Cosine Series

4. Complete Fourier Series

“Heat Conduction” by Sadik Kakac, Yaman Yener, Carolina P. Naveira-Cotta,
5th Ed, CRC Press

“Heat Conduction” by David W. Hahn, M. Necati Ozisik,

3rd Ed, John Wiley and Sons, Inc
Module – 2 : The Sturm-Liouville Theory and Fourier Expansions
Lecture – 10 : Fourier-Bessel Series
1. Expansion of a Function into Fourier-Bessel Series
 Sturm – Liouville Problems: Review
∞ Fourier Sine Series
𝑑 𝑑𝑦 𝑛𝜋
𝑝(𝑥) + 𝑞 𝑥 + 𝜆𝑤(𝑥) 𝑦 = 0 𝑓 𝑥 = 𝐴𝑛 sin 𝑥, 0<𝑥<𝐿
𝑑𝑥 𝑑𝑥 𝐿
𝑛=1
𝐿
𝑑𝑦(𝑎) 1 𝑛𝜋
𝛼1 𝑦 𝑎 + 𝛽1 = 0, 𝛼12 + 𝛽12 >0 𝐴𝑛 =
𝑁𝑛
𝑓 𝑥 sin
𝐿
𝑥 𝑑𝑥
𝑑𝑥 0
𝐿
𝑑𝑦(𝑏) 𝑛𝜋 𝐿
𝛼2 𝑦 𝑏 + 𝛽2 = 0, 𝛼22 + 𝛽22 >0 𝑁𝑛 = sin 2
𝑥 𝑑𝑥 =
𝑑𝑥 0 𝐿 2

Fourier Cosine Series Complete Fourier Series

∞
𝑛𝜋 1 1
𝑓 𝑥 = 𝐴𝑛 cos 𝑥, 0<𝑥<𝐿 𝐹 𝑥 = 𝐹 𝑥 − 𝐹(−𝑥) + 𝐹 𝑥 + 𝐹(−𝑥)
𝐿 2 2
𝑛=0 Odd function Even function
𝐿
1 𝑛𝜋 ∞
𝐴𝑛 = 𝑓 𝑥 cos 𝑥 𝑑𝑥 𝑛𝜋 𝑛𝜋
𝑁𝑛 0 𝐿 ⟹ 𝐹 𝑥 = 𝑏0 + 𝑎𝑛 sin 𝑥 + 𝑏𝑛 cos 𝑥 ,
𝐿 𝐿
𝐿 𝐿, 𝑛=0 𝑛=1
2
𝑛𝜋 −𝐿 < 𝑥 < 𝐿
𝑁𝑛 = cos 𝑥 𝑑𝑥 = 𝐿
0 𝐿 , 𝑛 = 1,2,3, …
2

3
 Bessel Functions: Review
Bessel’s differential 2
equation of order v, where 𝑥 2 𝑦 + 𝑥 𝑑𝑦 + 𝑚2 𝑥 2 − 𝜈 2 𝑦 = 0
𝑑
m is a parameter and v is 𝑑𝑥 2 𝑑𝑥
any real constant.
Bessel functions of the
𝐽𝜈
General Solution: 𝑦 𝑥 = 𝐶1 𝐽𝜈 𝑚𝑥 + 𝐶2 𝑌𝜈 (𝑚𝑥) first kind of order v

𝑌0
𝑌𝜈 Bessel functions of the
second kind of order v
𝑌1 𝑌2

Bessel functions of the first kind Bessel functions of the second kind
Modified Bessel Functions: Review
Modified Bessel’s differential 𝑑 2
𝑦 𝑑𝑦
2
equation of order v, where m is a 𝑥 2
+ 𝑥 − 𝑚2 𝑥 2 + 𝜈 2 𝑦 = 0
𝑑𝑥 𝑑𝑥
parameter and v is any real constant.

General Solution: 𝑦 𝑥 = 𝐶1 𝐼𝜈 𝑚𝑥 + 𝐶2 𝐾𝜈 (𝑚𝑥) Modified Bessel functions of the

𝐼𝜈
first kind of order v

𝐾𝜈 Modified Bessel functions of the

second kind of order v

𝐼𝜈 𝑥 𝐾𝜈 𝑥
𝐾2
𝐼0 𝐾1
𝐼1 𝐼2 𝐾0

𝑥 𝑥
Modified Bessel functions of Modified Bessel functions of
the first kind the second kind
 Domain: (a, b)
Fourier Bessel Series 𝑟 2
𝑑 2
𝑅
+ 𝑟
𝑑𝑅
+ 𝜆 2 2
𝑟 − 𝜈 2
𝑅=0
Series expansions in terms of Bessel 𝑑𝑟 2 𝑑𝑟
𝑑𝑅(𝑎)
functions arise most frequently in connection 𝛼1 𝑅 𝑎 + 𝛽1 = 0, 𝛼12 + 𝛽12 > 0
with the following characteristic-value problem 𝑑𝑟
𝑑𝑅(𝑏)
(r-direction): 𝛼2 𝑅 𝑏 + 𝛽2 = 0, 𝛼22 + 𝛽22 > 0
𝑅(𝑟) = 𝐶1 𝐽𝜈 𝜆𝑟 + 𝐶2 𝑌𝜈 (𝜆𝑟) 𝑑𝑟
We can write: Solution of Recall Bessel 2
2
𝑑 𝑦 𝑑𝑦 2 2 2
Bessel equation Differential 𝑥 + 𝑥 + 𝑚 𝑥 − 𝜈 𝑦=0
2
𝑑2 𝑅 𝑑𝑅 2 2 2 𝑑𝑥 2 𝑑𝑥
𝑟 + 𝑟 + 𝜆 𝑟 − 𝜈 𝑅=0 Equation:
𝑑𝑟 2 𝑑𝑟
𝑑 𝑑𝑦
𝑑 𝑑𝑅 Compare with: 𝑝(𝑥) + 𝑞 𝑥 + 𝜆𝑤(𝑥) 𝑦 = 0
⟹𝑟 𝑟 + 𝜆2 𝑟 2 − 𝜈 2 𝑅 = 0 𝑑𝑥 𝑑𝑥
𝑑𝑟 𝑑𝑟
(Divide the DE through by r)
This is a special case of the 𝜈2
Sturm–Liouville system with: 𝑝 𝑟 = 𝑟, 𝑞 𝑟 =− , 𝑤 𝑟 =𝑟
𝑟

Hence, the characteristic functions of this problem form a complete

orthogonal set with respect to the weight function r on the interval (a, b).
Fourier Bessel Series
We will consider problems on solid cylinders. The problems on
hollow-cylinder is analogous, but involve more complex algebra.

Problems on Solid Cylinders (r-direction) involve characteristic-

value problems of the following form:

2
𝑑2 𝑅 𝑑𝑅 2 2 2 The general solution can be written as:
𝑟 2
+ 𝑟 + 𝜆 𝑟 − 𝜈 𝑅=0
𝑑𝑟 𝑑𝑟
𝑅(𝑟) = 𝐶1 𝐽𝜈 𝜆𝑟 + 𝐶2 𝑌𝜈 (𝜆𝑟) Recall:
𝑅 0 = finite (finiteness or centerline symmetry at r = 0) Y(0) = ∞
𝑑𝑅(𝑟0 ) Use of BC R(0) = finite: 𝐶2 = 0
2 2
𝛼𝑅 𝑟0 + 𝛽 = 0, 𝛼 +𝛽 > 0
𝑑𝑟

Hence, the characteristic functions are Jv(λr) and the characteristic values λn are
the roots of the characteristic-value equation:
This is obtained by
𝑑𝐽𝜈 (𝜆𝑟0 ) applying the
𝛼𝐽𝜈 𝜆𝑟0 + 𝛽 = 0 ⟹ 𝜆𝑛 , 𝑛 = 1,2,3, … .
𝑑𝑟 boundary condition
at R = r0
 Fourier Bessel Series
𝑑 2
𝑅 𝑑𝑅 𝑑 𝑑𝑅 𝜈2
𝑟2 2 + 𝑟 + 𝜆2 𝑟 2 − 𝜈 2 𝑅 = 0 ⟹𝑟 𝑟 + 𝜆2 𝑟 2 − 𝜈 2 𝑅 = 0 𝑝 𝑟 = 𝑟, 𝑞 𝑟 =− ,
𝑑𝑟 𝑑𝑟 𝑑𝑟 𝑑𝑟 𝑟
(A)
𝑤 𝑟 =𝑟
𝑅 0 = finite
In view of the fact that p(0) = 0 and R(0) = finite (and also
𝑑𝑅(𝑟0 ) dR(0)/dr = finite), the characteristic functions of the system (A)
𝛼𝑅 𝑟0 + 𝛽 = 0, 𝛼 2 +𝛽 2 > 0
𝑑𝑟 form an orthogonal set with respect to the weight function
w(r) = r over the interval (0, r0); that is,
𝑟0
𝐽𝜈 𝜆𝑚 𝑟 𝐽𝜈 𝜆𝑛 𝑟 𝑟𝑑𝑟 = 0, 𝜆𝑚 ≠ 𝜆𝑛
0
and the set {Jv(λnr); n = 1, 2,...} is a complete orthogonal set.

Thus, we can expand an arbitrary function f(r), which is piecewise differentiable on

the interval (0, r0), in a series of these characteristic functions in the same interval
(0, r0) as:
∞
This is known as the Fourier–Bessel
𝑓 𝑟 = 𝐴𝑛 𝐽𝜈 𝜆𝑛 𝑟 , 0 < 𝑟 < 𝑟0
series of f(r) on the interval (0, r0).
𝑛=1
 Fourier Bessel Series
∞
This is known as the Fourier–Bessel series of f(r) on the
𝑓 𝑟 = 𝐴𝑛 𝐽𝜈 𝜆𝑛 𝑟 , 0 < 𝑟 < 𝑟0
interval (0, r0). We need to find An.
𝑛=1

Find the Fourier coefficients by multiplying by Jν(λnr) and by the necessary weighting function w(r) = r,
and then integrating over 0 ≤ r ≤ r0
𝑟0 𝑟0 ∞
Use orthogonal property. All of the terms on the
𝑓 𝑟 𝐽𝜈 𝜆𝑛 𝑟 𝑟𝑑𝑟 = 𝐴𝑘 𝐽𝜈 𝜆𝑘 𝑟 𝐽𝜈 𝜆𝑛 𝑟 𝑟𝑑𝑟 RHS vanish, except the one for which k = n.
0 0 𝑘=1 𝑟0
𝑟0 𝑟0 𝐽𝜈 𝜆𝑚 𝑟 𝐽𝜈 𝜆𝑛 𝑟 𝑟𝑑𝑟 = 0, 𝜆𝑚 ≠ 𝜆𝑛
2 0
⟹ 𝑓 𝑟 𝐽𝜈 𝜆𝑛 𝑟 𝑟 𝑑𝑟 = 𝐴𝑛 𝐽𝜈 𝜆𝑛 𝑟 𝑟 𝑑𝑟
0 0

𝑟0 𝑟0
1
This gives: 𝐴𝑛 = 𝑓(𝑟)𝐽𝜈 𝜆𝑛 𝑟 𝑟𝑑𝑟 where 𝑁𝑛 = 𝐽𝜈2 𝜆𝑛 𝑟 𝑟𝑑𝑟
𝑁𝑛 0 0

The normalization integral Nn can be evaluated as:

2 2
𝑟02 2
𝜈 2
𝑑𝐽𝜈 𝜆𝑛 𝑟0
𝑁𝑛 = 2 𝜆𝑛 − 𝐽𝜈 𝜆𝑛 𝑟0 +
2𝜆𝑛 𝑟0 𝑑𝑟
Fourier Bessel Series: Three Special Cases
∞ 𝑟0
𝑟0
1
𝑓 𝑟 = 𝐴𝑛 𝐽𝜈 𝜆𝑛 𝑟 , 0 < 𝑟 < 𝑟0 𝐴𝑛 = 𝑓(𝑟)𝐽𝜈 𝜆𝑛 𝑟 𝑟𝑑𝑟 𝑁𝑛 = 𝐽𝜈2 𝜆𝑛 𝑟 𝑟𝑑𝑟
𝑁𝑛 0 0
𝑛=1
𝑑𝑅(𝑟0 )
The BC at r = r0: 𝛼𝑅 𝑟0 + 𝛽 = 0, 𝛼 2 +𝛽 2 > 0 Characteristic functions: 𝐽𝜈 𝜆𝑛 𝑟 , 𝑛 = 1,2,3, … .
𝑑𝑟

Case 1: α ≠ 0, β = 0
𝑟02
For this case the characteristic values are the roots of: 𝐽𝜈 𝜆𝑛 𝑟0 = 0 𝑁𝑛 = 𝐽𝜈+1 𝜆𝑛 𝑟0
2

Case 2: α = 0, β ≠ 0

For this case the characteristic values are the roots of:
2
𝑑𝐽𝜈 𝜆𝑛 𝑟0 𝑟02 𝜈
=0 𝑁𝑛 = 1− 𝐽𝜈2 𝜆𝑛 𝑟
𝑑𝑟 2 𝜆𝑛 𝑟0
 Fourier Bessel Series: Three Special Cases
∞ 𝑟0
𝑟0
1
𝑓 𝑟 = 𝐴𝑛 𝐽𝜈 𝜆𝑛 𝑟 , 0 < 𝑟 < 𝑟0 𝐴𝑛 = 𝑓(𝑟)𝐽𝜈 𝜆𝑛 𝑟 𝑟𝑑𝑟 𝑁𝑛 = 𝐽𝜈2 𝜆𝑛 𝑟 𝑟𝑑𝑟
𝑁𝑛 0 0
𝑛=1
𝑑𝑅(𝑟0 )
The BC at r = r0: 𝛼𝑅 𝑟0 + 𝛽 = 0, 𝛼 2 +𝛽 2 > 0 Characteristic functions: 𝐽𝜈 𝜆𝑛 𝑟 , 𝑛 = 1,2,3, … .
𝑑𝑟
Case 2: α = 0, β ≠ 0 (Cont’d) 2
𝑑𝐽𝜈 𝜆𝑛 𝑟0 𝑟02 𝜈
𝑁𝑛 = 1− 𝐽𝜈2 𝜆𝑛 𝑟
=0 2 𝜆𝑛 𝑟0
𝑑𝑟
Note: If ν = 0 then λ0 = 0 is a characteristic value and the corresponding characteristic
function is J0(λ0r) = 1.

Thus, when ν = 0, the function J0(λ0r) = 1 must be included in the set of characteristic functions.

The Fourier–Bessel series of f(r) then becomes:

∞ 𝑟0
2
𝑓 𝑟 = 𝐴0 + 𝐴𝑛 𝐽0 𝜆𝑛 𝑟 𝐴
where 0 = 𝑓(𝑟) 𝑟𝑑𝑟
𝑟02 0
𝑛=1
Fourier Bessel Series: Three Special Cases
∞ 𝑟0
𝑟0
1
𝑓 𝑟 = 𝐴𝑛 𝐽𝜈 𝜆𝑛 𝑟 , 0 < 𝑟 < 𝑟0 𝐴𝑛 = 𝑓(𝑟)𝐽𝜈 𝜆𝑛 𝑟 𝑟𝑑𝑟 𝑁𝑛 = 𝐽𝜈2 𝜆𝑛 𝑟 𝑟𝑑𝑟
𝑁𝑛 0 0
𝑛=1
𝑑𝑅(𝑟0 )
The BC at r = r0: 𝛼𝑅 𝑟0 + 𝛽 = 0, 𝛼 2 +𝛽 2 > 0 Characteristic functions: 𝐽𝜈 𝜆𝑛 𝑟 , 𝑛 = 1,2,3, … .
𝑑𝑟
Case 3: α ≠ 0, β ≠ 0

For this special case the characteristic values are the roots of:

𝛼 𝑑𝐽𝜈 𝜆𝑛 𝑟0
𝐽 𝜆 𝑟 + =0
𝛽 𝜈 𝑛 0 𝑑𝑟

𝑟02 1 𝛼2 𝜈2
𝑁𝑛 = 1− 2 − 𝐽𝜈2 𝜆𝑛 𝑟
2 𝜆𝑛 𝛽 2 𝑟02
Fourier Bessel Series: Table
Reference:
“Heat Conduction” by Sadik Kakac,
Yaman Yener, Carolina P. Naveira-
Cotta, 5th Ed, CRC Press
1. Expansion of a Function into Fourier-Bessel Series
“Heat Conduction” by Sadik Kakac, Yaman Yener, Carolina P. Naveira-Cotta,
5th Ed, CRC Press

“Heat Conduction” by David W. Hahn, M. Necati Ozisik,

3rd Ed, John Wiley and Sons, Inc
Module – 2 : The Sturm-Liouville Theory and Fourier Expansions
Lecture – 11 : Fourier-Legendre Series
1. Legendre Polynomials

2. Fourier-Legendre Series
Legendre Polynomials
When a problem in spherical coordinate system depends on
the cone angle θ, its solution can be reduced to expansion of
an arbitrary function into a series of Legendre Polynomials.

The following linear and homogeneous second-order ordinary

differential equation with variable coefficients is known as
Legendre’s differential equation:
2
𝑑2 𝑦 𝑑𝑦
1−𝑥 − 2𝑥 +𝑛 𝑛+1 𝑦 =0 where n is real and non-negative.
𝑑𝑥 2 𝑑𝑥
𝑑 2
𝑑𝑦
⟹ 1−𝑥 +𝑛 𝑛+1 𝑦 =0
𝑑𝑥 𝑑𝑥

It’s two linearly independent solutions are called Legendre function of the first kind
and Legendre function of the second kind, respectively.

When n is zero or a positive integer (common cases), the Legendre function of the
first kind becomes a polynomial in x of degree n, whereas the second kind can be
expressed as an infinite series.
Legendre Polynomials
The general solution of Legendre’s equation for n = 0, 1, 2, … can be written as

𝑦 𝑥 = 𝐶1 𝑃𝑛 𝑥 + 𝐶2 𝑄𝑛 (𝑥) Pn(x) = Legendre polynomials of degree n

Qn(x) = Legendre functions of the second kind
The Legendre polynomials Pn(x) are given by:
1.3.5 … 𝑛 − 1
𝑃0 (𝑥) = 1 𝑃1 𝑥 = 𝑥 𝑃𝑛 𝑥 = −1 𝑛/2
𝑈𝑛 𝑥 , 𝑛 = 2,4,6, … . n is even, ≥ 2
2.4.6 … . 𝑛
1.3.5 … 𝑛
𝑃𝑛 𝑥 = −1 (𝑛−1)/2
𝑉𝑛 𝑥 , 𝑛 = 3,5,7, … . n is odd, ≥ 3
2.4.6 … . (𝑛 − 1)
where
𝑛 𝑛+1 2 𝑛 𝑛−2 𝑛+1 𝑛+3 4
𝑈𝑛 𝑥 = 1 − 𝑥 + 𝑥 −⋯ (11.1)
2! 4!
(𝑛 − 1) 𝑛 + 2 3 (𝑛 − 1) 𝑛 − 3 𝑛 + 2 𝑛 + 4 5
𝑉𝑛 𝑥 = 𝑥 − 𝑥 + 𝑥 −⋯ (11.2)
3! 5!
 Note that if n is an even positive integer, the series (Un(x), 11.1) terminates with the
term involving xn, and hence is a polynomial of degree n.
 Similarly, if n is an odd positive integer, the series (Vn(x), 11.2) terminates with the
term involving xn.
 Otherwise, these expressions are infinite series convergent only when |x| < 1.
Legendre Polynomials
The general solution of Legendre’s equation for n = 0, 1, 2, … can be written as
Pn(x) = Legendre polynomials of degree n
𝑦 𝑥 = 𝐶1 𝑃𝑛 𝑥 + 𝐶2 𝑄𝑛 (𝑥)
Qn(x) = Legendre functions of the second kind

The first six Legendre polynomials are:

1 1
𝑃0 𝑥 = 1, 𝑃1 𝑥 = 𝑥, 𝑃2 𝑥 = 3𝑥 2 − 1 , 𝑃3 𝑥 = 5𝑥 3 − 3𝑥
2 2
1 1
𝑃4 𝑥 = 35𝑥 4 − 30𝑥 2 + 3 , 𝑃5 𝑥 = 63𝑥 5 − 70𝑥 3 + 15𝑥
8 8
 Legendre Polynomials Pn(x): Some Useful Properties
A) Recurrence formula:
Each of these polynomials is a function of the polynomials preceding and succeeding it. Thus, the
following recurrence formula can be established:
1 1
𝑃0 𝑥 = 𝑃1 𝑥 , 𝑃1 𝑥 = 𝑃0 𝑥 + 2𝑃2 (𝑥)
𝑥 3𝑥
𝑛𝑃𝑛−1 𝑥 + (𝑛 + 1)𝑃𝑛+1 (𝑥)
In general, we can write: 𝑃𝑛 𝑥 = , 𝑛 = 1,2,3, … .
2𝑛 + 1 𝑥

B) 𝑃𝑛 𝑥 = −1 𝑛 𝑃𝑛 (𝑥) Pn(1) = 1 and Pn(–1) = (–1)n

𝑑𝑃𝑛+1 𝑑𝑃𝑛−1
C) − = 2𝑛 + 1 𝑃𝑛 𝑥 , 𝑛 = 1,2,3, …
𝑑𝑥 𝑑𝑥

The Legendre polynomials can also be 1 𝑑𝑛 2 𝑛

D) 𝑃𝑛 𝑥 = 𝑛 𝑥 −1
expressed by Rodrigues’ formula: 2 𝑛! 𝑑𝑥 𝑛
Legendre Polynomials
The general solution of Legendre’s equation for n = 0, 1, 2, … can be written as
𝑦 𝑥 = 𝐶1 𝑃𝑛 𝑥 + 𝐶2 𝑄𝑛 (𝑥) Pn(x) = Legendre polynomials of degree n
Qn(x) = Legendre functions of the second kind

If |x| < 1, the Legendre functions of the second kind Qn(x) are given by:

𝑛/2
2.4.6 … . 𝑛
𝑄0 𝑥 = 𝑉0 (𝑥) 𝑄𝑛 𝑥 = −1 𝑉𝑛 𝑥 , 𝑛 = 2,4,6, … . n is even, ≥ 2
1.3.5 … . (𝑛 − 1)

(𝑛+1)/2
2.4.6 … . (𝑛 − 1)
𝑄1 𝑥 = −𝑈1 (𝑥) 𝑄𝑛 𝑥 = −1 𝑈𝑛 𝑥 , 𝑛 = 3,5,7, … .
1.3.5 … . 𝑛
n is odd, ≥ 3

Note that the Legendre functions of the second kind are infinite series,
which are convergent when |x| < 1 but diverge as x → ±1.
Legendre Polynomials: Transformation
When |x| < 1, if we substitute x = cos θ into the Legendre’s differential 𝑑 𝑑𝑦
1 − 𝑥2 +𝑛 𝑛+1 𝑦 =0
equation, the equation is transformed to: 𝑑𝑥 𝑑𝑥

1 𝑑 𝑑𝑦
sin 𝜃 +𝑛 𝑛+1 𝑦 =0 Equations of such a form arise in
sin 𝜃 𝑑𝜃 𝑑𝜃
connection with solutions of various
heat conduction problems in
𝑑2 𝑦 𝑑𝑦 spherical coordinates.
equivalently, ⇒ + cot 𝜃 +𝑛 𝑛+1 𝑦 =0
𝑑𝜃 2 𝑑𝜃

The general solution of the above equations: 𝑦 𝑥 = 𝐶1 𝑃𝑛 cos 𝜃 + 𝐶2 𝑄𝑛 (cos 𝜃)

Note:
 Qn(cos θ) is not finite when cos θ = ±1, that is, when θ = kπ, k = 0, ±1, ±2, …

 Pn(cos θ) is a polynomial of degree n in cos θ.

Orthogonality of Legendre Polynomials
Consider the Legendre’s differential equation and 𝑑 𝑑𝑦
𝑝(𝑥) + 𝑞 𝑥 + 𝜆𝑤(𝑥) 𝑦 = 0
compare it with the Sturm-Liouville system: 𝑑𝑥 𝑑𝑥
𝑑𝑦(𝑎)
𝑑 𝑑𝑦 𝛼1 𝑦 𝑎 + 𝛽1 = 0, 𝛼12 + 𝛽12 > 0
1 − 𝑥2 +𝑛 𝑛+1 𝑦 =0 𝑑𝑥
𝑑𝑥 𝑑𝑥 𝑑𝑦(𝑏)
𝛼2 𝑦 𝑏 + 𝛽2 = 0, 𝛼22 + 𝛽22 > 0
𝑑𝑥
This is a special case of the Sturm–Liouville system with:
Note: 𝑄𝑛 𝑥 → ∞ as 𝑥 → ±1
𝑝 𝑥 = 1 − 𝑥2, 𝑞 𝑥 = 0, 𝑤 𝑥 = 1, 𝜆2 = 𝑛(𝑛 + 1) Thus we exclude physically
insignificant Qn(x)
Since the function p(x) vanishes at x = ±1, we conclude that any two distinct
solutions of the Legendre’s differential equation, which are finite and have finite
first derivatives at x = ±1, will be orthogonal with respect to the weight function
w(x) = 1 on the interval (–1, 1).

Therefore, no specific boundary conditions are needed for the above Legendre
differential equation to lead to a Sturm–Liouville type characteristic-value
problem over the interval (–1, 1).
Fourier-Legendre Series
Legendre’s differential equation has finite solutions at 𝑑 2
𝑑𝑦
1−𝑥 +𝑛 𝑛+1 𝑦 =0 (11.3)
x = ±1 (in terms of Pn(x), only if n is zero or a positive 𝑑𝑥 𝑑𝑥
integer.

This condition of finiteness determines the permissible values of n : 𝑛 = 0,1,2, …

The corresponding solutions that are finite when x = ±1 Note: 𝑄𝑛 𝑥 → ∞ as 𝑥 → ±1

are the Legendre polynomials Pn(x). Thus we exclude physically
insignificant Qn(x)

Hence, the (Eq. 11.3) is a characteristic-value problem and the characteristic

functions {Pn(x); n = 0, 1, 2,…} form a complete orthogonal set with respect to
the weight function w(x)=1 on the interval (-1, 1).
1
𝑃𝑚 𝑥 𝑃𝑛 𝑥 𝑑𝑥 = 0, 𝑚≠𝑛
−1

Therefore any piecewise differentiable function f(x) can be represented in

the interval (–1, 1) by a series of the Legendre polynomials.
Fourier-Legendre Series
Any piecewise differentiable function f(x) can be represented in the interval (–1, 1) by a series of
the Legendre polynomials as
∞

𝑓 𝑥 = 𝐴0 𝑃0 𝑥 + 𝐴1 𝑃1 𝑥 + 𝐴2 𝑃2 𝑥 + ⋯ ⇒𝑓 𝑥 = 𝐴𝑛 𝑃𝑛 𝑥 , −1 < 𝑥 < 1
𝑛=0

This is called the Fourier–Legendre series expansion of f(x) on the interval (–1, 1).
1
−1
𝑓 𝑥 𝑃𝑛 𝑥 𝑑𝑥
The expansion coefficients An are given by 𝐴𝑛 = 1
𝑃 (𝑥) 2 𝑑𝑥
−1 𝑛
Unlike Fourier or Fourier-Bessel series, the 1
2
denominator does not depend on any boundary 𝑃𝑛 (𝑥) 2 𝑑𝑥 =
−1 2𝑛 + 1
condition. It may be evaluated once and for all.

The expansion coefficients An can be evaluated as:

1 1 𝑛
2𝑛 + 1 2𝑛 + 1 𝑑 𝑓(𝑥)
𝐴𝑛 = 𝑓 𝑥 𝑃𝑛 𝑥 𝑑𝑥 or 𝐴𝑛 = 𝑛+1 1 − 𝑥2 𝑛 𝑛
𝑑𝑥
2 −1 2 𝑛! −1 𝑑𝑥
Fourier-Legendre Series
Pn(x) is an even function of x when n is even, and an odd function when n is odd.

If f(x) is an even function of x, the coefficients An will vanish when n is odd; whereas if f(x) is an odd
function of x, the coefficients An will vanish when n is even.

0, 𝑛 = odd
1
For an even function f(x), 𝐴𝑛 =
(2𝑛 + 1) 𝑓 𝑥 𝑃𝑛 𝑥 𝑑𝑥, 𝑛 = even
0

1
(2𝑛 + 1) 𝑓 𝑥 𝑃𝑛 𝑥 𝑑𝑥, 𝑛 = odd
For an odd function f(x), 𝐴𝑛 = 0
0, 𝑛 = even
 Fourier-Legendre Series
∞ For an even function f(x),
𝑓 𝑥 = 𝐴𝑛 𝑃𝑛 𝑥 , −1 < 𝑥 < 1 (A)
𝑛=0
0, 𝑛 = odd
1 (B)
𝐴𝑛 =
(2𝑛 + 1) 𝑓 𝑥 𝑃𝑛 𝑥 𝑑𝑥, 𝑛 = even
0
The series expansion given by (A),
considered with (B) or (C), represents For an odd function f(x),
f(x) in [0, 1].
1
(2𝑛 + 1) 𝑓 𝑥 𝑃𝑛 𝑥 𝑑𝑥, 𝑛 = odd (C)
𝐴𝑛 = 0
The series expansion given by (A),
0, 𝑛 = even
considered with (B), represents f(-x) in
[-1, 0].
f(x) expressed as (A) + (B): contains
The series expansion given by (A), only even Legendre polynomials
considered with (C), represents -f(-x) in
[-1, 0]. f(x) expressed as (A) + (C): contains
only odd Legendre polynomials
 Fourier-Legendre Series: General Interval
Expansions valid in the more general interval (–a, a) are readily obtained by replacing x
by x/a in the preceding development, which leads to the following expansion.
∞

𝑓 𝑥 = 𝐴𝑛 𝑃𝑛 𝑥 , −1 < 𝑥 < 1
𝑛=0

∞
𝑥
𝑓 𝑥 = 𝐴𝑛 𝑃𝑛 , −𝑎 < 𝑥 < 𝑎
𝑎
𝑛=0

𝑎
2𝑛 + 1 𝑥
where 𝐴𝑛 = 𝑓 𝑥 𝑃𝑛 𝑑𝑥
2𝑎 −𝑎 𝑎
𝑎
2𝑛 + 1 2 2 𝑛
𝑑 𝑛 𝑓(𝑥)
or 𝐴𝑛 = 𝑛+1 𝑎 −𝑥 𝑑𝑥
2 𝑛! 𝑎𝑛+1 −𝑎 𝑑𝑥 𝑛
1. Legendre Polynomials

2. Fourier-Legendre Series
“Heat Conduction” by Sadik Kakac, Yaman Yener, Carolina P. Naveira-Cotta,
5th Ed, CRC Press

“Heat Conduction” by David W. Hahn, M. Necati Ozisik,

3rd Ed, John Wiley and Sons, Inc
Module – 2 : The Sturm-Liouville Theory and Fourier Expansions
Lecture – 12 : Tutorial on Fourier Expansions
1. Example on Fourier Series

2. Example on Fourier-Bessel Series

3. Example on Fourier-Legendre Series

Example-1: Characteristic-Value Problem
Show that the following boundary-value problem 𝑑2 𝑦
2
− 𝜆2 𝑦 = 0, 𝑦 0 = 0, 𝑦 𝐿 =0
cannot have nontrivial solutions for real values of λ. 𝑑𝑥
Solution:
 Example-2: Fourier-Sine Series
Expand f(x) = x in a Fourier sine series of period 2L over the interval (0, L).
∞
Solution: 𝑛𝜋 This is a Fourier sine series expansion of
𝑓 𝑥 =𝑥= 𝐴𝑛 sin 𝑥, 0<𝑥<𝐿
𝐿 f(x) = x of period 2L on the interval (0, L).
𝑛=1

The expansion coefficients An are given by:

𝐿 𝑛𝜋 𝐿 𝑛𝜋
𝑥 sin 𝑥 𝑑𝑥 𝑥 sin 𝑥 𝑑𝑥 2 𝐿
𝑛𝜋 2𝐿 2𝐿 2𝐿 (−1)𝑛+1
𝐴𝑛 =
0 𝐿 =
0 𝐿 = 𝑥 sin 𝑥 𝑑𝑥 = − cos 𝑛𝜋 = − −1 𝑛
=
𝐿
sin 2 𝑛𝜋 𝑥 𝑑𝑥 𝐿 𝐿 0 𝐿 𝑛𝜋 𝑛𝜋 𝜋 𝑛
0 𝐿 2
∞ sin 𝑛𝜋 = 0
𝑛+1
2𝐿 (−1) 𝑛𝜋
Thus, 𝑓 𝑥 =𝑥= sin 𝑥, 0≤𝑥<𝐿 cos(𝑛𝜋) = (−1)𝑛
𝜋 𝑛 𝐿
𝑛=1

Note that at x = 0, f(x) = x = 0. Also, at x = L, f(x) = L = L. This series converges to

f(x) = x at x = 0, but not at x = L as it converges to zero at x = L.
Moreover, since f(x) = x is an odd function, this series is not only valid on the interval
(0, L), but also on the interval (–L, L).
Example-3: Characteristic-Value Problem
2
Show that this boundary-value problem cannot have 𝑑 𝑅 𝑑𝑅
𝑟2 2 + 𝑟 − 𝜆2 𝑟 2 + 𝜈 2 𝑅 = 0
nontrivial solutions for real values of λ. 𝑑𝑟 𝑑𝑟
𝑅 0 = finite 𝑅 𝑟0 = 0
Solution:
Example-4: Fourier-Bessel Series ∞
Consider the expansion of the function f(r) = 1 on the 𝑓 𝑟 =1= 𝐴𝑛 𝐽0 𝜆𝑛 𝑟 , 0 < 𝑟 < 𝑟0
interval (0, r0) in a Fourier-Bessel series of the form 𝑛=1

where λn are the positive roots of J0(λnr) = 0.

𝑟0
Here, 1 𝑟02 2
𝐴𝑛 = 𝐽0 𝜆𝑛 𝑟 𝑟𝑑𝑟 where 𝑁𝑛 = 𝐽1 𝜆𝑛 𝑟0
𝑁𝑛 0 2
2
Therefore, 𝐴𝑛 = , 𝑛 = 1,2,3, … .
𝜆𝑛 𝑟0 𝐽1 𝜆𝑛 𝑟0

The Fourier-Bessel series for f(r) = 1 becomes:

∞
2 1 𝐽0 𝜆𝑛 𝑟
𝑓 𝑟 =1=
𝑟0 𝜆𝑛 𝐽1 𝜆𝑛 𝑟0
𝑛=1

2 𝐽0 𝜆1 𝑟 2 𝐽0 𝜆2 𝑟 2 𝐽0 𝜆3 𝑟
= + + + ⋯.
Evaluate this series. 𝜆1 𝑟0 𝐽1 𝜆1 𝑟0 𝜆2 𝑟0 𝐽1 𝜆2 𝑟0 𝜆3 𝑟0 𝐽1 𝜆3 𝑟0

Solution:
Example-4: Fourier-Bessel Series
Zeros xk of J0(x) and the
The Fourier-Bessel series for f(r) = 1: Evaluate this series. corresponding values of J1(x)
∞
2 1 𝐽0 𝜆𝑛 𝑟 2 𝐽0 𝜆1 𝑟 2 𝐽0 𝜆2 𝑟 2 𝐽0 𝜆3 𝑟
1= = + + + ⋯.
𝑟0 𝜆𝑛 𝐽1 𝜆𝑛 𝑟0 𝜆1 𝑟0 𝐽1 𝜆1 𝑟0 𝜆2 𝑟0 𝐽1 𝜆2 𝑟0 𝜆3 𝑟0 𝐽1 𝜆3 𝑟0
𝑛=1

Here λn are the positive roots of J0(λnr) = 0.

∞
2 1 𝐽0 𝜆𝑛 𝑟 2 𝐽0 𝜆1 𝑟 2 𝐽0 𝜆2 𝑟 2 𝐽0 𝜆3 𝑟
1= = + + + ⋯.
𝑟0 𝜆𝑛 𝐽1 𝜆𝑛 𝑟0 𝜆1 𝑟0 𝐽1 𝜆1 𝑟0 𝜆2 𝑟0 𝐽1 𝜆2 𝑟0 𝜆3 𝑟0 𝐽1 𝜆3 𝑟0
𝑛=1

𝑟 𝑟 𝑟
𝐽0 (𝜆1 𝑟0 ) 𝐽0 (𝜆2 𝑟0 ) 𝐽0 (𝜆3 𝑟0 )
2 𝑟0 2 𝑟0 2 𝑟0
= + + +⋯
𝜆1 𝑟0 𝐽1 𝜆1 𝑟0 𝜆2 𝑟0 𝐽1 𝜆2 𝑟0 𝜆3 𝑟0 𝐽1 𝜆3 𝑟0
Example-4: Fourier-Bessel Series
Zeros xk of J0(x) and the
𝑟 𝑟 𝑟 corresponding values of J1(x)
𝐽0 (𝜆1 𝑟0 ) 𝐽0 (𝜆2 𝑟0 ) 𝐽0 (𝜆3 𝑟0 )
2 𝑟0 2 𝑟0 2 𝑟0
1= + + + ⋯.
𝜆1 𝑟0 𝐽1 𝜆1 𝑟0 𝜆2 𝑟0 𝐽1 𝜆2 𝑟0 𝜆3 𝑟0 𝐽1 𝜆3 𝑟0

𝑟 𝑟 𝑟
2 𝐽0 2.4048 2 𝐽0 5.5201 2 𝐽0 8.6537
𝑟0 𝑟0 𝑟0
= + + + ⋯.
2.4048 (0.5191) 5.5201 (−0.3403) 8.6537 (0.2715)

𝑟 𝑟 𝑟
= 1.6021 𝐽0 2.4048 − 1.0646 𝐽0 5.5201 + 0.8512 𝐽0 8.6537 + ⋯.
𝑟0 𝑟0 𝑟0
Since J0(xk) are even functions, the above series
represents 1 not only on the interval (0, r0), but also
on the symmetrical interval (–r0, r0).

At the end points r = ± r0, however, the series does

not converge to 1, because all the terms in the
series vanish at r = ± r0 (each term ≡ J0 at root)
Example-5: Fourier-Legendre Series
Consider the expansion of the following function in a 𝜋
cos 𝜃, 0<𝜃<
series of Legendre polynomials in cos θ over the 𝑓 𝜃 = 2
𝜋
interval (0, π). Determine the first three coefficients. 0, <𝜃<𝜋
2
1
Given: The first three Legendre polynomials are: 𝑃0 𝑥 = 1, 𝑃1 𝑥 = 𝑥, 𝑃2 𝑥 = 3𝑥 2 − 1
2
1
2𝑛 + 1
The expansion coefficients An can be evaluated as: 𝐴𝑛 = 𝑓 𝑥 𝑃𝑛 𝑥 𝑑𝑥
2 −1
Solution:
 Example-5: Fourier-Legendre Series cos 𝜃, 0<𝜃<
𝜋
2
Consider the expansion of the following function in a series of Legendre 𝑓 𝜃 = 𝜋
polynomials in cos θ over the interval (0, π). Determine the first three coefficients. 0, <𝜃<𝜋
2
1
Given: The first three Legendre polynomials are: 𝑃0 𝑥 = 1, 𝑃1 𝑥 = 𝑥, 𝑃2 𝑥 = 3𝑥 2 − 1
1
2
2𝑛 + 1
The expansion coefficients An can be evaluated as: 𝐴𝑛 = 𝑓 𝑥 𝑃𝑛 𝑥 𝑑𝑥
2 −1
Solution (Cont’d):
 Example-5: Fourier-Legendre Series cos 𝜃, 0<𝜃<
𝜋
2
Consider the expansion of the following function in a series of Legendre 𝑓 𝜃 = 𝜋
polynomials in cos θ over the interval (0, π). Determine the first three coefficients. 0, <𝜃<𝜋
2
1
Given: The first three Legendre polynomials are: 𝑃0 𝑥 = 1, 𝑃1 𝑥 = 𝑥, 𝑃2 𝑥 = 3𝑥 2 − 1
1
2
2𝑛 + 1
The expansion coefficients An can be evaluated as: 𝐴𝑛 = 𝑓 𝑥 𝑃𝑛 𝑥 𝑑𝑥
2 −1
Solution (Cont’d):
MATLAB Functions for Bessel Functions
Bessel function of first kind: 𝐽𝜈 (𝑧)
>> besselj(0,0)
J = besselj(nu,Z) ans =
1
computes the Bessel function of the first
kind Jν(z) for each element in array Z.

Bessel function of second kind: 𝑌𝜈 (𝑧) >> bessely(0,0)

ans =
-Inf
Y = bessely(nu,Z)
MATLAB Functions for Bessel Functions
Modified Bessel function of first kind: 𝐼𝜈 (𝑧)

I = besseli(nu, z) >> >>

besseli(0,0) besseli(0,1)
ans = ans =
1 1.2661

Modified Bessel function of second kind: 𝐾𝜈 (𝑧)

K = besselk(nu, z) >> besselk(0,0)

ans =
Inf
1. Example on Fourier Series

2. Example on Fourier-Bessel Series

3. Example on Fourier-Legendre Series

1. “Heat Conduction” by M. Necati Ozisik, 2nd Ed, John Wiley and Sons, Inc

2. “Heat Conduction” by David W. Hahn, M. Necati Ozisik, 3rd Ed, John Wiley
and Sons, Inc

3. “Heat Conduction” by Sadik Kakac, Yaman Yener, Carolina P. Naveira-Cotta,

5th Ed, CRC Press

4. “Heat Conduction” by Latif Jiji, 3rd Ed, Springer

5. “Boundary Value Problems of Heat Conduction” by

M. Necati Ozisik, International Text Book Company