The Laplace Operator

In mathematics and physics, the Laplace operator or Laplacian, named after

Pierre-Simon de Laplace, is an unbounded differential operator, with many
applications. However, in describing application of spectral theory, we re-
strict the attention to an open subset of Euclidean space Rd .

Definition and Self Adjointness

Let, U ⊂ Rd be open and non empty set and H = L2 (U ) be the space of
all square integrable functions on U . Then we define the Laplace operator
(D(∆), ∆) as follows. The domain D(∆) = Cc∞ (U ) is the space of smooth
and compactly supported functions on U and assumed to be dense in H.
For φ ∈ D(∆), put
n
X ∂2φ
∆φ = −
j=1
∂x2j
which is again smooth, compactly supported function and hence lies in
L2 (U ).
For differential operators in such a domain, we can not expect their self
adjointness. However, we proceed to show that Laplace operator has at least
self adjoint extension.
Proposition :
Let U be open and non empty open subset of Rd , and (D(∆), ∆) be the
Laplace operator defined above. Then,
1. The Laplace operator is symmetric on D(∆) i.e.

h∆φ, ψi = hφ, ∆ψi ∀ φ, ψ ∈ Cc∞ (U )

2. The Laplace operator is positive i.e.

h∆φ, φi ≥ 0 ∀ φ ∈ Cc∞ (U )

1
 Proof. Using integration by parts twice and using the fact that com-
pactly supported functions are zero in an open set, we get

Z
h∆φ, ψi = ∆φ(x)ψ(x)dx
U
n Z
X
= −∂x2j φ(x)ψ(x)dx
j=1 U

Xn Z
= −φ(x)∂x2j ψ(x)dx
j=1 U

= hφ, ∆ψi

for every φ, ψ ∈ Cc∞ (U ) Hence ∆ is symmetric.

Again, for any φ ∈ D(∆), we have
Z
h∆φ, φi = ∆φ(x)φ(x)dx
U
n
XZ
= −∂x2j φ(x)φ(x)dx
j=1 U
Xn Z
= ∂xj φ(x)∂xj φ(x)dx
j=1 U
n Z
∂x φ(x)2 dx
X  
= j
j=1 U

≥ 0 ⇒ ∆ is positive

Note.
1. Every self adjoint linear T : H → H operator is symmetric. On the
other hand, symmetric linear operators need not be self adjoint. The reason
is that T ∗ may be a proper extension if T i.e D(T ) 6= D(T ∗ ). Clearly, this
can not happen if D(T )is all of H. So
For linear operators T : H → H on complex Hilbert space H, the concepts
of symmetry and self adjointness are identical.
2. If there were no negative sign in the definition, the Laplace operator
would have been negative.
Theorem. Let U ⊂ Rd be a non empty open subset. Then the Laplace
operator admits self adjoint extensions. Proof : This follows at once from
the general theorem of Friedrichs.

2
 Theorem. (Friedrichs) Let, H be a Hilbert space,.(D(T ), T ) ∈ DD∗ (H)
a symmetric operator which is positive i.e. such that

hT (v), vi ≥ 0,
for all v ∈ D(T ). Then T admits at least one self adjoint extension, called
the Friedrichs Extension (D(S), S) such that

hS(v), vi ≥ 0, for allv ∈ D(S).

Two Basic Examples

In this unit, we will discuss two examples of Laplace operators acting on
the whole space Rn and on the open cube (0, 1)n and discuss their spectral
properties by finding the explicit representation of self adjoint extension of
∆ as multiplication operators. These operators differ in two aspects. On
Rn , ∆ is essentially self adjoint i.e. the closure is self adjoint and is the
unique self adjoint extension of ∆. Its spectrum is purely continuous and
σ(∆) = [0, ∞). On the other hand, Laplace operator on cube (0, 1)n has
distinct extensions and in particular, we discuss two of them, the Dirichlet
and the Neumanns extension. These two extensions are distinguished by
specific boundary conditions. In this case,∆ turns out to have purely point
spectrum. Also, the eigenvalues can be computed fairly explicitly.
Proposition :
The closure of Laplace operator on Rn is unitarily equivalent with the
multiplication operator (D, T ) acting on L2 (Rn ), where

D = {φ ∈ L2 (Rn )|x → kxk2 φ(x) ∈ L2 (Rn )}

T φ(x) = (2π)2 kxk2 φ(x)

where kxk2 is the Euclidean norm on Rn .
In particular, ∆ is essentially self adjoint on Rn , its spectrum is equal to
[0, ∞) and is entirely continuous spectrum.
Proof.
The main tool for proving this argument is the Fourier Transform,

U : L2 (Rn ) → L2 (Rn )

given by Z
U φ(x) = φ(t)e−2πihx,ti dt
Rn

3
 Integrating by parts, we get

U (∂xj φ)(x) = 2πxj U φ(x)

and from this we derive that

U (∆φ)(x) = (2π)2 kxk2 U φ(x) forφ ∈ D(∆)

We first prove that ∆ is essentially self adjoint by showing Im(∆ + i) is dense

in L2 (Rn ). Let z = i or −i. For showing Im(∆ + z) is dense in L2 (Rn ), we
shall prove its orthogonal complement

Im(∆ + z)⊥ = 0

Let φ ∈ L2 (Rn ) be such that

hφ, (∆ + z)ψi = 0for allψ ∈ D(∆) = Cc∞ (Rn )

Since, Fourier transform is unitary,

0 = hφ, (∆ + z)ψi = hU φ, U (∆ + z)ψi

= U φ, (4π 2 kxk2 + z)U ψ



= (4π 2 kxk2 + z̄)U φ, U ψ for allψ ∈ D(∆)



Since D(∆) is dense in L2 (Rn ), so is U D(∆) and thus we have

(4π 2 kxk2 + z̄)U φ = 0

⇒ Uφ = 0
⇒φ=0

Thus, ∆ is essentially self adjoint.

Also, the formula above shows that Laplace operator D((∆), ∆) is uni-
tarily equivalent with the multiplication operator.

M4π2 k.k2 : φ 7−→ 4π 2 kxk2 φ on U D(∆)

This is essentially self adjoint.

The multiplication operator can be defined on

D = {φ ∈ L2 (Rn )| kxk2 φ(x) ∈ L2 (Rn )}

Indeed, (D, M4π2 kxk2 ) is self adjoint and so is the closure of (U D(∆), M4π2 kxk2 ).

4
 Now, from inverse Fourier transform, it follows that closure of ∆ is uni-
tarily equivalent with (D, 4π 2 kxk2 ).
Finally, the essential range of M4π2 kxk2 is [0, ∞), it follows that σ(∆) =
[0, ∞).
Also, since there is no eigen value of the multiplication operator, the
spectrum is purely continuous.
Example :
Next, we see for U=(o, 1)n The main idea corresponds to expanding of
function φ ∈ L2 (Rn ) in terms of orthonormal basis of H = L2 (U ) formed by
complex exponentials

ek : x → e2πihx,ki for k = (k1 , k2 , ..., kn ) ∈ Z n

Note that (ek ∈ C ∞ (U )). Moreover, viewing ∆ as an differential operator

and using the relation

ek (x1 , x2 , ..., xn ) = e2πik1 x1 .e2πik2 x2 ....e2πikn xn

and y ” = α2 y for y(x) = eαx ,we see that

∆ek = (2π)2 kkk2 ek

However,ek is not compactly supported. So, the domain of this eigen-

function is different from ∆ and such eigenfunction could be many.
For any
α = (α1 , ...., αn ) ∈ C n
fα (x) = eα1 x1 +....+αn xn ,
we have
∆fα = α12 + α22 + ..... + αn2 fα

Using this, we confirm that ∆ is not essentially self adjoint. Namely, for
any α ∈ C n with

α12 + α22 + ..... + αn2 = ±i for instance α = eiπ/4 , 0, 0....0

the function fα is an eigenfunction of ∆ with eigenvalue ±i.

We now check fα ∈ D(∆∗ )and satisfies

∆∗ fα = ±ifα

In fact, for φ ∈ D(α), we have

Z
h∆φ, fα i = ∆φ(x)ehα,xi dx
U

5
 Integrating by parts and using the fact that φ is compactly supported
(despite the fact that it is not for fα shows that ∆ is symmetric. However,
by definition this proves that (D (∆) , ∆) is not self adjoint.
The computation gives a rough idea of self adjoint extension of ∆ by
allowing more general functions rather than compactly supported one so
that boundary conditions will force the integration by parts to get symmetric
operators. But, the domain of such may be delicate to describe.
It is easy to think the situation for n=1. Define three subspaces of
L ([0, 1]) containing D(∆) = CC∞ ((0, 1)) as follows.
2

First, define D̃ to be space of function of φ ∈ C ∞ ((0, 1)) for which every

derivative φ(j) ,j ≥ 0 extends to continuous function on [0,1], where we denote
by φ(j) (0) and φ(j) (1) the corresponding value at the boundary points.
Then let
n ′ ′
o
D1 = φ ∈ D̃|φ(0) = φ(1), φ (0) = φ (1), .....
n o
D2 = φ ∈ D̃|φ(0) = φ(1) = 0
n ′ ′
o
D3 = φ ∈ D̃|φ (0) = φ (1) = 0

These spaces are distinct and have the laplace operators (Di , ∆), each
extending to the laplace operator (D(∆), ∆).
Then integrating by parts
Z 1
h∆φ, ψi = −φ” (x)ψ(x)dx
Z0 1
= −φ(x)ψ ” (x)dx
0
= hφ, ∆ψi ∀φ, ψ ∈ Dj

So, ∆ is symmetric.
Here,D1 , D2 and D3 are called respectively the Laplace operator with
periodic, Dirichlet and Neumann’s boundary conditions.
Proposition :
The three operators (Dj , ∆), 1 ≤ j ≤ 3 are essentially self-adjoint ex-
tensions of the Laplace operator (D(∆), ∆) on D(∆). Moreover, all three
σ (∆j ) = σp (∆j )-(pure point spectrum) and the eigenvalues are given by the
following:
σ (D1 ) = 0, 4π 2 , 16π 2 , ...., 4π 2 k 2 , .....


σ (D2 ) = π 2 , 4π 2 , ...., k 2 π 2 , .....



6
 σ (D1 ) = 0, π 2 , 4π 2 , ...., k 2 π 2 , .....


The spectrum is simple, i.e., the eigen spaces have dimension 1, for D2 and
D3 . For D1 , we have

dim ker (D1 ) = 1, dim ker D1 − (2πk)2 = 2, for k ≥ 1

Corollary :
Let U = (0, 1)n with n ≥ 1. Consider the operators ∆p = (Dp , ∆) and
∆d = (Dd , ∆) extending (D(∆), ∆) with domains given, respectively, by Dp ,
which is the space of restrictions of C ∞ functions on R which are Z n periodic,
and Dd which is the space of function φ ∈ C ∞ (U ) for which every partial
derivative of any order ∂α φ extends to a continuous function on Ū , and
moreover such that φ(x) = 0for x ∈ ∂U , where we use the same notation φ
for the function and its extension to Ū . Then ∆p and ∆d are essentially self-
adjoint. Their closures have pure point spectra, given by the real numbers
of the form
λ = 4π 2 k12 + .... + kn2 , ki ∈ Z

with the condition ki ≥ 1 for Dd The multiplicity of a given λ is the

number of k1 , ..., kn ∈ Z n with

λ = 4π 2 k12 + .... + kn2

for ∆p , and is 2−n times that number for ∆d .

Submitted by : MADHAV WAGLEY

Student Number : 250539918