Lecture 2
Lecture 2
Sturm-Liouville Problems
Definition 6.1 (Sturm-Liouville Boundary Value Problem (SL-BVP)) With the notation
· ¸
d dy
L[y] ≡ p(x) + q(x) y, (6.1)
dx dx
where p > 0, r ≥ 0, and p, q, r are continuous functions on interval [a, b]; along with the boundary
conditions
a1 y(a) + a2 p(a)y 0 (a) = 0, b1 y(b) + b2 p(b)y 0 (b) = 0, (6.3)
where a21 + a22 6= 0 and b21 + b22 6= 0.
The problem of finding a complex number µ if any, such that the BVP (6.2)-(6.3) with λ = µ, has
a non-trivial solution is called a Sturm-Liouville Eigen Value Problem (SL-EVP). Such a value µ
is called an eigenvalue and the corresponding non-trivial solutions y(.; µ) are called eigenfunctions.
Further,
(i) An SL-EVP is called a regular SL-EVP if p > 0 and r > 0 on [a, b].
(ii) An SL-EVP is called a singular SL-EVP if (i) p > 0 on (a, b) and p(a) = 0 = p(b), and (ii)
r ≥ 0 on [a, b].
(iii) If p(a) = p(b), p > 0 and r > 0 on [a, b], p, q, r are continuous functions on [a, b], then solving
Sturm-Liouville equation (6.2) coupled with boundary conditions
We are not going to discuss singular SL-BVPs. Before we discuss further, let us completely study
two examples that are representatives of their class of problems.
53
54 6.1. Two examples
Case 1. Let λ < 0. Then λ = −µ2 , where µ is real and non-zero. The general solution of ODE in
(6.5) is given by
y(x) = Aeµx + Be−µx (6.6)
This y satisfies boundary conditions in (6.5) if and only if A = B = 0. That is, y ≡ 0. Therefore,
there are no negative eigenvalues.
Case 2. Let λ = 0. In this case, it easily follows that trivial solution is the only solution of
This is not a SL-BVP. It is a mixed boundary condition unlike the separated BC above. These
boundary conditions are called periodic boundary conditions.
Case 1. Let λ < 0. Then λ = −µ2 , where µ is real and non-zero. In this case, it can be easily
verified that trivial solution is the only solution of the BVP (6.11).
Case 2. Let λ = 0. In this case, general solution of ODE in (6.11) is given by
y(x) = A + Bx (6.12)
This y satisfies the BCs in (6.11) if and only if B = 0. Thus A remains arbitrary.
Thus 0 is an eigenvalue with eigenfunction being any non-zero constant. Note that eigenvalue is
simple. An eigenvalue is called simple eigenvalue if the corresponding eigenspace is of dimension
one, otherwise eigenvalue is called multiple eigenvalue.
Case 3. Let λ > 0. Then λ = µ2 , where µ is real and non-zero. The general solution of ODE in
(6.11) is given by
y(x) = A cos(µx) + B sin(µx) (6.13)
This y satisfies boundary conditions in (6.11) if and only if
This has non-trivial solution for the pair (A, B) if and only if
¯ ¯
¯ sin(µπ) 1 − cos(µπ) ¯¯
¯
¯ 1 − cos(µπ) − sin(µπ) ¯ = 0. (6.14)
That is, cos(µπ) = 1. This further implies that µ = ±2n with n ∈ N, and hence λ = 4n2 with
n ∈ N.
Thus positive eigenvalues are given by
λn = 4n2 , n ∈ N. (6.15)
Note: All the eigenvalues are non-negative. There are two linearly independent eigenfunctions,
namely cos (2nx) and sin (2nx) corresponding to each positive eigenvalue λn = 4n2 . Compare
these properties with that of previous example.
Multiply the ODE in (6.17) wtih y, and multiply that of (6.17) with y, and subtracting
one from the other yields
£ 0 ¤0
p(y y − y 0 y) + (λ − λ)ryy = 0 (6.19)
But LHS of the last equation is zero, since we have both b1 y(b) + b2 p(b)y 0 (b) = 0
and b1 y(b) + b2 p(b)y 0 (b) = 0, we also know that b21 + b22 6= 0, and hence a certain
determinant associated is zero.
Thus we have Z b
(λ − λ) r|y|2 dy = 0. (6.21)
a
Since y, being an eigenfunction, is not identically equal to zero, and integral of non-
negative function (since r > 0) is not zero, the only possibility is that λ = λ. That
is, λ is real. Note that we have used self-adjointness of operator L somewhere!
Proof :
As in the previous proof, writing down the equations satisfied by u and v, and multi-
plying the equation for u with v and vice versa, finally subtracting one from another,
we get £ 0 ¤0
p(u v − v 0 u) + (λ − µ)ruv = 0 (6.23)
Integrating the last equality yields
Z b
¤¯b
[p(u v − v u) ¯a = −(λ − µ)
0 0
r(x)u(x)v(x) dx. (6.24)
a
Reasoning exactly as in the previous proof, LHS of the above equality is zero. Since
λ 6= µ, we get the desired (6.22).
(3) The eigenvalues of a regular SL-BVP are simple. Thus an eigenfunction corresponding
to an eigenvalue is unique up to a constant multiple.
Proof :
Let φ1 and φ2 be two eigenfunctions corresponding to the same eigenvalue λ.
We recall from the section on Green’s functions (the identity (5.44) ) here:
d
£ 0 ¤
By Lagrange’s identity (5.20), we get dx p(φ1 φ2 − φ1 φ02 ) = 0. This implies
Since φ1 and φ2 satisfy the boundary condition U1 [y] = 0, we get the following
¯ ¯
¯ φ1 (a) φ01 (a) ¯
¯ ¯
¯ φ2 (a) φ02 (a) ¯ = 0. (6.26)
Since (φ01 φ2 − φ1 φ02 ) is the wronskian of two solutions of a second order ODE, it is
identically equal to zero. From here, it follows that φ1 and φ2 differ by a constant
multiple.
In the above remark, we only analysed the properties of an eigenvalue or of two eigenfunctions
corresponding to distinct eigenvalues. We have not proved the existence of eigenvalues for a regular
SL-BVP so far. We are not going to do this, since the result follows easily from a much general
theory of a subject known as Functional analysis, to be more specific, the topic is called spectral
theory. Some references where a proof can be found are books on functional analysis by B.V.
Limaye, Yosida and also books on ODE by Coddington & Levinson, Hartman or even
books on PDE by Weinberger. We will only state the result.
¡ ¢
Theorem 6.5 A self-adjoint regular SL-BVP has an infinite sequence of real eigenvalues λn n∈N ,
that are simple satisfying
λ1 < λ 2 < · · · < λ n < . . . (6.27)
with lim λn → ∞.
n→∞
Exercise 6.6 Let h > 0 be a real number. Find eigenvalues and corresponding eigenvectors of the
regular SL-BVP posed on the interval [0, 1]
But LHS of the last equation is zero, since both y and y satisfy the periodic boundary
conditions. Note that this is new argument that we use to replace the correspondign
argument in part (1) of Remark 6.4.
Thus we have Z b
(λ − λ) r|y|2 dy = 0. (6.32)
a
Since y, being an eigenfunction, is not identically equal to zero, and integral of non-
negative function (since r > 0) is not zero, the only possibility is that λ = λ. That
is, λ is real. Note that we have used self-adjointness of operator L somewhere!
(2) The eigenfunctions of a periodic SL-BVP corresponding to distinct eigenvalues are
othogonal w.r.t. weight function r on [a, b], that is, if u and v are eigenfunctions
corresponding to distinct eigenvalues λ and µ respectively, then
Z b
r(x)u(x)v(x) dx = 0. (6.33)
a
Proof :
As in the previous proof, writing down the equations satisfied by u and v, and multi-
plying the equation for u with v and vice versa, finally subtracting one from another,
we get £ 0 ¤0
p(u v − v 0 u) + (λ − µ)ruv = 0 (6.34)
Integrating the last equality yields
Z b
¤¯b
[p(u0 v − v 0 u) ¯a = −(λ − µ) r(x)u(x)v(x) dx. (6.35)
a
Reasoning exactly as in the previous proof, LHS of the above equality is zero. Since
λ 6= µ, we get the desired (6.33).
(3) The eigenvalues of a regular SL-BVP are not simple, and Example 6.3 illustrates this
fact. However, it will be interesting to know why the arguments given in point (3) of
Remark 6.4 can not be modified. In that proof we heavily rely on the form of boundary
conditions for a regular SL-BVP, in deducing 6.26.
In the above remark, we only analysed the properties of an eigenvalue or of two eigenfunctions
corresponding to distinct eigenvalues. We have not proved the existence of eigenvalues for a
periodic SL-BVP so far. We are not going to do this. We will only state the result, and you may
refer to ODE book of Coddington & Levinson.
¡ ¢
Theorem 6.8 A self-adjoint periodic SL-BVP has an infinite sequence of real eigenvalues λn n∈N
satisfying
−∞ < λ1 < λ2 ≤ λ3 ≤ · · · ≤ λn ≤ · · · (6.36)
The first eigenvalue λ1 is simple. The number of linearly independent eigenfunctions corresponding
to any eigenvalue µ is equal to the number of times µ is repeated in the above listing.
Exercise 6.9 Find eigenvalues and eigenvectors of the following periodic SL-BVP posed on the
interval [−π, π]
y 00 + λy = 0, y(−π) = y(π), y 0 (−π) = y 0 (π).