MA-417-ODE

Lecture 21

Debanjana Mitra

Department of Mathematics
Indian Institute of Technology Bombay
Powai, Mumbai - 76

October 1, 2021
Regular Sturm-Liouville’s problem
Regular Sturm-Liouville’s problem
Regular Sturm-Liouville’s problem:
Under the following assumptions:
1. The interval (a, b) is open and bounded, i.e., a and b are distinct
and finite real numbers.
2. The functions p ∈ C 1 ([a, b]), q ∈ C ([a, b]) and r ∈ C ([a, b]) are real
valued. Further, p(t) > 0 and r (t) > 0 on [a, b].
3. The boundary condtions:

U1 (y ) = α1 y (a) + α2 y 0 (a) = 0, U2 (y ) = β1 y (b) + β2 y 0 (b) = 0,

where α1 , α2 , β1 , β2 are given real numbers with α12 + α22 6= 0 and

β12 + β22 6= 0,
consider the eigenvalue problem
d
L(y )(t) = − (p(t)y 0 (t)) + q(t)y (t) = λr (t)y (t), ∀t ∈ (a, b),
dt
with homogeneous boundary conditions:

U1 (y ) = 0, U2 (y ) = 0.
Wronskian for BVP
Theorem
Let φ and ψ be two solutions satisfying the Regular Sturm-Liouville’s
problem for a λ ∈ C : L(y )(t) = λr (t)y (t), on (a, b), with boundary
conditions U1 (y ) = 0 and U2 (y ) = 0, then the Wronskian W (φ, ψ)
satisfies: p(t)W (φ, ψ)(t) = constant, ∀, t ∈ [a, b], where
W (φ, ψ)(t) = φ(t)ψ 0 (t) − φ0 (t)ψ(t).
Proof: We have
L(φ)(t) = λr (t)φ(t), on (a, b), U1 (φ) = 0 = U2 (φ), (1)
L(ψ)(t) = λr (t)ψ(t), on (a, b), U1 (ψ) = 0 = U2 (ψ). (2)
Multiplying (1) with ψ and (2) with φ and taking difference between
them, we get
d d
(p(t)ψ 0 (t))φ(t) − (p(t)φ0 (t))ψ(t) = 0.
dt dt
Now taking integration of the above identity over [a, t], for any t ∈ (a, b],
we get
p(t)W (φ, ψ)(t) = p(a)W (φ, ψ)(a), ∀ t ∈ [a, b],
and hence the result is proved.
Proof of Property 3. in Main theorem (Lecture 20)
Theorem
Each eigenvalue of a regular Sturm-Liouville’s problem is simple, i.e., to
each eigenvalue, there corresponds one and only one eigenfunction, up to
a constant factor, i.e., the dimension of the eigenspace corresponding to
each eigenvalue is 1.
Proof: For any eigenvalue λ, let φ and ψ be eigenfunctions i.e.,

L(φ)(t) = λr (t)φ(t), on (a, b), U1 (φ) = 0 = U2 (φ), (3)

L(ψ)(t) = λr (t)ψ(t), on (a, b), U1 (ψ) = 0 = U2 (ψ). (4)

Now since U1 (φ) = 0 and U1 (ψ) = 0, we get

α1 φ(a) + α2 φ0 (a) = 0,
α1 ψ(a) + α2 ψ 0 (a) = 0.

Since (α1 , α2 ) 6= (0, 0), from above it follows that W (φ, ψ)(a) = 0.
Thus, using previous theorem along with p(t) 6= 0 on [a, b], we get
W (φ, ψ)(t) = 0, for all t ∈ [a, b].
It yields that φ, ψ are linearly dependent, i.e., there exists a c ∈ C such
that ψ(t) = cφ(t), ∀ t ∈ [a, b].
Regular Sturm-Liouville’s problem
Let all assumptions in the regular Sturm-Liouville’s problem hold.
Consider the eigenvalue problem:

L(y )(t) = λr (t)y (t) on (a, b), U1 (y ) = 0 = U2 (y ).

Recall, we have
1. The regular Sturm-Liouville’s problem has an infinite sequence of
real eigenvalues
λ1 < λ2 < · · · < λn < · · · ,
with limn→∞ λn = ∞.
2. The set of normalized eigenfunctions {φn }∞ n=1 , where the
eigenfunction φn corresponds to the eigenvalue λn , form an
orthonormal basis in L2r (a, b). That is any f ∈ L2r (a, b) can be
represented by
∞
X
f = hf , φn iL2r φn , in L2r (a, b).
n=1

2
Further, if f ∈ C ([a, b]) satisfying U1 (f ) = 0 = U2 (f ), the series
converges absolutely and uniformly on [a, b].
Non-homogeneous ODE: Regular Sturm-Liouville’s problem
Let all assumptions in the regular Sturm-Liouville’s problem hold. For
any f ∈ C ([a, b]), consider the BVP:
d
L(y )(t) = − (p(t)y 0 (t)) + q(t)y (t) = λr (t)y (t) + f (t), ∀t ∈ (a, b),
dt
(5)
with homogeneous boundary conditions:

U1 (y ) = α1 y (a) + α2 y 0 (a) = 0, U2 (y ) = β1 y (b) + β2 y 0 (b) = 0,

where α1 , α2 , β1 , β2 are given real numbers with α12 + α22 6= 0 and

β12 + β22 6= 0. Then exactly one of the following alternatives holds:
1. If λ is not an eigenvalue of

L(y )(t) = λr (t)y (t), on (a, b), U1 (y ) = 0 = U2 (y ),

then (5) has a unique solution for every f ∈ C ([a, b]) and this
solution is given by
∞
X hf , φiL2
y (t) = r
φn (t), t ∈ [a, b].
n=1
λn − λ
Contd...

2. If λ = λj , an eigenvalue, then (5) has a solution provided

hf , φj iL2r = 0.

In this case, there are infinitely many solutions given by

∞
X hf , φiL2r
y (t) = cφj (t) + φn (t), t ∈ [a, b],
λn − λ
n=1,n6=j

for any c ∈ C.
The advantage of the above result is the explicit represent formula of the
solution of the Non-homogeneous ODE.

However, there is another approach to obtain the representation formula

of the solution of the Non-homogeneous ODE using Green’s functions.
Green’s function
Let all assumptions in the regular Sturm-Liouville’s problem hold. For
any f ∈ C ([a, b]), consider the BVP:
d
L(y )(t) = − (p(t)y 0 (t)) + q(t)y (t) = λy (t) + f (t), ∀t ∈ (a, b), (6)
dt
with homogeneous boundary conditions:
U1 (y ) = α1 y (a) + α2 y 0 (a) = 0, U2 (y ) = β1 y (b) + β2 y 0 (b) = 0,
where α1 , α2 , β1 , β2 are given real numbers with α12 + α22 6= 0 and
β12 + β22 6= 0.
Assume λ is not an eigenvalue of
L(y )(t) = λy (t) on (a, b), U1 (y ) = 0 = U2 (y ).
Qn. Can we find a functions G (·, ·) defined on [a, b] × [a, b] such that the
solution of the (6) can be written as
Z b
y (t) = G (t, τ )f (τ ) dτ,
a
for any f ∈ C ([a, b])? Such function G is called ‘Green’s function’.
Advantage: The integral representation of the solution is often more
informative for studying the behavior of the solution.
Method of constructing the Green’s function
Step 1. Let φ1 be the solution of the IVP:

L(φ1 )(t) = 0, φ1 (a) = α2 , φ01 (a) = −α1 .

Thus, U1 (φ1 ) = 0.
Let φ2 be the solution of the IVP:

L(φ2 )(t) = 0, φ2 (b) = β2 , φ02 (b) = −β1 .

Thus, U2 (φ2 ) = 0.

Note that, since λ is not an eigenvalue of

L(y )(t) = λy (t), U1 (y ) = 0 = U2 (y ),

the two solutions of the second order ODE, φ1 and φ2 are linearly
independent on [a, b], and hence

W (φ1 , φ2 )(t) 6= 0, ∀ t ∈ [a, b].

Contd...
Step 2. Use variation of parameter: Set
φ(t) = C1 (t)φ1 (t) + C2 (t)φ2 (t), t ∈ [a, b].
Find C1 (t) and C2 (t) such that L(φ)(t) = λφ(t) + f (t) and
U1 (φ) = 0 = U2 (φ).
Calculate and get:
φ0 (t) = C10 (t)φ1 (t) + C20 (t)φ2 (t) + C1 (t)φ01 (t) + C2 (t)φ02 (t).
Impose the condition:

C10 (t)φ1 (t) + C20 (t)φ2 (t) = 0, ∀ t ∈ [a, b]. (7)

Calculate and get:

f (t) = L(φ)(t)
 − λφ(t)   
= C1 (t) L(φ1 )(t) − λφ1 (t) + C2 (t) L(φ1 )(t) − λφ(t2 )
 
−p(t) C10 (t)φ01 (t) + C20 (t)φ02 (t)

Thus, we get
f (t)
C10 (t)φ01 (t) + C20 (t)φ02 (t) = − , t ∈ [a, b]. (8)
p(t)
Contd...
Writting (7)-(8) together,

C10 (t)φ1 (t) + C20 (t)φ2 (t) = 0, ∀ t ∈ [a, b],

f (t)
C10 (t)φ01 (t) + C20 (t)φ02 (t) = − p(t) , t ∈ [a, b].

we get

φ2 (t)f (t) φ1 (t)f (t)

C10 (t) = , C20 (t) = − , t ∈ [a, b].
p(t)W (φ1 , φ2 )(t) p(t)W (φ1 , φ2 )(t)

Note that using W (φ1 , φ2 )(t) 6= 0 for all t ∈ [a, b], to make U1 (φ) = 0,
we need C2 (a) = 0 and to make U2 (φ) = 0, we need C1 (b) = 0.
Thus, solving C1 and C2 , we get
Z b
φ2 (τ )
C1 (t) = − f (τ ) dτ, ∀ t ∈ [a, b],
t p(τ )W (φ1 , φ2 )(τ )
Z t
φ1 (τ )
C2 (t) = − f (τ ) dτ, ∀ t ∈ [a, b].
a p(τ )W (φ1 , φ2 )(τ )
Thus, we get that the solution φ satisfying

L(φ)(t) = λφ(t) + f (t), ∀ t ∈ (a, b), U1 (φ) = 0 = U2 (φ),

is represented by
Z b Z t
φ1 (t)φ2 (τ ) φ1 (τ )φ2 (t)
φ(t) = − f (τ ) dτ − f (τ ) dτ
t p(τ )W (φ1 , φ2 )(τ ) a p(τ )W (φ1 , φ2 )(τ )

Setting G (t, τ ) as

 − φ1 (τ )φ2 (t) , a ≤ τ ≤ t,
p(τ )W (φ1 ,φ2 )(τ )
G (t, τ ) =
 − φ1 (t)φ2 (τ ) , t ≤ τ ≤ b,
p(τ )W (φ1 ,φ2 )(τ )

we get
Z b
φ(t) = G (t, τ )f (τ ) dτ, ∀ t ∈ [a, b].
a
Properties of Green’s function

From the formula of G , the following properties can be obtained:

1. G is continuous over [a, b] × [a, b].
∂ ∂ 1
2. limt→τ + ∂t G (t, τ ) − limt→τ − ∂t G (t, τ ) = − p(τ ).
3. G is symmetric i.e., G (t, τ ) = G (τ, t).