CHAPTER II

SOBOLEV SPACES – DEFINITIONS AND MAIN PROPERTIES

DEFINITION 2.1:

Given an open set Ω ⊆ R N , a multi-index α ∈ N 0N \{ 0 }, and 1 ≤ p ≤ ∞ , we say that a

function u ∈ L1loc ( Ω ; R M ) admits a weak or distributional αth derivative in L p ( Ω ; R M ) if there

exists a function v α ∈ L p (Ω; R M ) such that

∫ u ∂ α ϕdx=(−1 )|α|∫ v α ϕdx (2.1)

Ω Ω

α
∂ T u (ϕ)=T v (ϕ )α

|α|
∂ u
for all ϕ ∈C ∞c ( Ω ) . The function v α is denoted ∂ α u = α
≔v α .
∂x

p
A similar definition can given when L p (Ω; R M ) is replaced by Lloc ( Ω ; RM ) and the weak
derivatives are unique.

DEFINITION 2.2:

Given an open set Ω ⊆ R N , a vector γ ∈ R N \{ 0 }, n∈ N and 1 ≤ p ≤ ∞ , we say that a

function u ∈ L1loc ( Ω ; R M ) admits a weak or distributional nth directional derivative in

L (Ω; R ) in the direction γ if there exists a function v γ , n ∈ L ( Ω; R ) such that

p M p M

n
∫ u ∂∂ γϕn dx= (−1 )n∫ v γ , n ϕdx
Ω Ω

n
∂ u
for all ϕ ∈C ∞c ( Ω ) . The function v γ , n is denoted n
.
∂γ

DEFINITION 2.3:

Given an open set Ω ⊆ R N , m, M ∈ N , and 1 ≤ p ≤ ∞ , the Sobolev space W m , p ( Ω; R M ) is

the space of all functions u ∈ L p ( Ω ; R M ) which admit weak derivatives ∂ α u in L p (Ω; R M ) for
every α ∈ N 0N with 1 ≤|α|≤ m. The space W m , p ( Ω ; R M ) is endowed with the norm

‖u‖W m, p
(Ω ; R M) ≔ ‖u‖L (Ω ; R )+
p M ∑ ‖∂ α u‖L ( Ω ;R p M
).
1 ≤|α|≤m

The space W mloc, p (Ω; R M ) is defined as the space of all functions u ∈ L p ( Ω; R M ) which
(Ω; R ) for every α ∈ N 0 with 1 ≤|α|≤ m.
p M N
admit weak derivatives ∂ α u in Lloc

When M = 1 we write W m , p (Ω) and W mloc, p (Ω) for W m , p (Ω; R) and W mloc, p ( Ω; R ) ,
respectively. Also, we write ‖u‖L and ‖u‖W for ‖u‖L (Ω ) and ‖u‖W
p m, p p m, p
(Ω ) , respectively.

REMARK:

Given Ω ⊆ R N open, m ∈ N , with m ≥2 , and 1 ≤ p ≤ ∞ , u ∈W m , p ( Ω ) , in view of (2.1) and

Schwartz theorem on the order of differentiation for smooth functions applied to ϕ ∈C ∞c ( Ω ) ,

2 2 2 2
∫ ∂ x∂ ∂ux ϕdx=∫ u
∂ ϕ
∂ xi ∂ x j
dx =∫ u
∂ ϕ
∂ x j ∂ xi
dx=∫
∂ u
ϕdx ,
Ω i j Ω Ω Ω ∂ x j ∂ xi

2 2
∂ u ∂ u
which shows that = . Thus for distributional derivatives the order in which one
∂ xi ∂ x j ∂ x j ∂ x i
differentiates is not important.

For u ∈W m , p ( Ω ) and 1 ≤ k ≤ m , we set ∇ k u to be the vector of components ∂ α u , |α|=k .

Then an equivalent norm in W m , p ( Ω) is given by

m
‖u‖L (Ω )+ ∑ ‖∇ k u‖L (Ω ;R ) ,
p p Mk

k=1

where M k is the number of multi-indicesα ∈ N 0N with |α|=k . For simplicity, we will write
‖ ∇ u‖L ( Ω ) for ‖∇k u‖L ( Ω ;R ) .
k
p p Mk

Another equivalent norm which we will sometimes use is given by

( ∑ ‖∂α u‖L ( Ω )) p
p 1
p
‖u‖W m, p ‖‖
(Ω ) ≔ u L (Ω )+
p p

1≤|α|≤ m
for 1 ≤ p< ∞ , and

‖u‖W m ,∞
( Ω) ≔max {‖∂ α u‖L ∞
(Ω ) : 0 ≤|α|≤ m } ,

where ∂0 u ≔u , for p=∞ .

PROPERTY 2.1:

Let Ω ⊆ R N be an open set, let m ∈ N , with m ≥2 , and let1 ≤ p ≤ ∞ . Then u ∈ L p ( Ω )

belongs to W m , p ( Ω ) if and only if ∇ u ∈ W m−1 , p ( Ω, R N ) .

Proof:

Assume that,u ∈ L p ( Ω) belongs to W m , p ( Ω ) .

(i.e)., u ∈W m , p (Ω; R N )

To prove,∇ u ∈ W m−1 , p ( Ω; R N ) .

Then by definition, for ant multi-indexα with |α|≤ m, the partial derivatives ∂ α u exists
and are in L p ( Ω ) .

β
Now, consider ∂i u for |β|=m−1.
1

By the fundamental theorem of calculus,

We can express ∂ β uas the integral of its gradient over straight line segment connecting
β p
points in Ω . Since ∂ β u ∈ L p ( Ω ) , its gradient ∇ ∂β u is also in L p ( Ω ; R N ) ,and thus ∂i ∂ u ∈ L ( Ω ) .
1

Thus, ∇ u ∈ W m−1 , p ( Ω; R N ) .

Conversely,

Assume that,∇ u ∈ W m−1 , p ( Ω; R N ) .

This implies that u ∈W m−1 , p ( Ω ; R N ) .

To prove, u ∈W m , p ( Ω ; R N ) .
 Then for any multi-index α with |α|≤ m−1 , the partial derivatives ∂ α u exist and are in
Lp (Ω ; RN ) .

β
Now, consider the function ∂ ∂i u for |β|=m.
1

By repeated applications of the product rule and the assumptionu ∈W m−1 , p ( Ω ; R N ) , we

have ∂ ∂i u ∈ L ( Ω; R ) .
β p N
1

Thus, u ∈W m , p ( Ω ; R N ) .

Hence proved.

EXAMPLE:

Show that the function u : B ( 0 ,1 ) → R , is defined by

u ( x )=u ( x 1 , … , x N ) ≔
{
1 if x N >0 ,
0 if x N <0 ,

does not belong to W 1 , p (B ( 0 , 1 ) ) for any 1 ≤ p ≤ ∞ .

Soln:

To show that the function u ( x )= {

1if x N >0 ,
0 if x N < 0 ,
does not belong to W 1 , p ( B ( 0 , 1 ) )for any

1 ≤ p ≤ ∞ , we demonstrate this its partial derivatives are not in L p ( B ( 0 , 1 ) ) .

Consider the partial derivative with respect to x N , denoted as ∂ x u . N

Since u(x) is discontinuous at x N =0 , its derivative ∂ x u will have a jump discontinuity at

N

x N =0.

Thus ∂ x u will have a singularity on the hyperplane x N =0 , making it non-integrable on

N

any neighborhood of 0.

Therefore, u(x) fails to satisfy the condition required for functions in W 1 , p (B ( 0 , 1 ) ) since
at least one of its partial derivatives is not in L p ( B ( 0 , 1 ) ) .
PROPERTY 2.2:

Let Ω ⊂ R N be an open bounded set, let x 0 ∈ Ω , and let 1 ≤ p ≤ ∞ . If

u ∈C ( Ω ) ∩C (Ω¿ x 0 }) is such that the (classical) gradient ∇ u belongs to L p ( Ω ; R N ) , then
1

1,p
u ∈W (Ω).

Proof:

To prove this, we’ll utilize the definition of the Sobolev space W 1 , p (Ω) and the properties
of continuous and continuously differentiable functions.

Given u ∈C ( Ω ) ∩C 1 (Ω¿ {x 0 ¿ }) with ∇ u ∈ L p ( Ω; R n ) .

We need to show that u ∈W 1 , p ( Ω ) . It is enough to show that u is absolutely continuous on

almost every line segment in Ω and its weak derivative belongs to L p ( Ω ) .

(i) Absolutely Continuity of u:

Since u is continuous on Ω , its already absolutely continuous on any line segment

contained entirely within Ω . The only potential issue arises near x 0 .

However, since u is continuously differentiable on Ω ¿ x 0 }, its locally Lipschitz around

x 0 ,ensuring absolute continuity even there.

(ii) Weak derivative of u:

∂ϕ
We need to show that for any test function ϕ ∈C ∞0 ( Ω ) , the integral ∫ u dx exists and
Ω
∂ xi
is finite.

This follows that from the fact that ∇ u ∈ L p (Ω ; RN ) and the definition of the weak
derivative.

Thus, u satisfies the conditions of W 1 , p ( Ω ) , implying u ∈W 1 , p ( Ω ) .

PROPERTY 2.3:

Let Ω ⊆ R N be an open set and let C ⊂ Ω be a closed set with H N−1 (C )=0 , where we
recall that H N−1 is the (N-1)-dimensional Hausdorff measure. Let m ∈ N and 1 ≤ p ≤ ∞ and let
m, p
u ∈W ( Ω ¿ ) . Then u ∈W m , p ( Ω ) .

Proof:

Assume that,

Let Ω ⊆ R N be an open set and let C ⊂ Ω be a closed set with H N−1 (C )=0 , where we
recall that H N−1 is the (N-1)-dimensional Hausdorff measure.

Let m ∈ N and 1 ≤ p ≤ ∞ and let u ∈W m , p ( Ω ¿ ) .

To prove that, u ∈W m , p ( Ω ) when u ∈W m , p (Ω¿), we can start by considering the case when m =
1.

For each i=1 , … , N , let ∏ be the projection into the hyperplane orthogonal to the x i−¿
i

axis.

First, we’ll prove that H N−1 ¿ for each i. Since C has H N−1 (C )=0 , its projection onto any
hyperplane also has H N−1 ¿

Since, H N−1 ¿ for almost every x ∈ Ω , the slice (x ¿¿ i=x)∩Ω ¿ intersects C in a set of
measure zero.

Apply Fubini’s theorem, we have for almost every x ∈ Ω , the function u is in

1, p
W ( x i=x ) ∩ Ω.

Now, we can integrate the W 1 , p ‖u ‖ over each slice ( x i=x ) ∩Ω , and since this holds for
almost every x ∈ Ω , by integrating over all x i , we obtain that u ∈W 1 , p ( Ω ) .

For higher m, we can use induction. If u ∈W m , p ( Ω ¿ ) , then u ∈W m−1 , p (Ω¿) by definition.

Apply the same argument as above to show that u ∈W m−1 , p ( Ω ) .
 Repeat this process m times to conclude that u ∈W m , p ( Ω ) .

Hence proved.

PROPERTY 2.4:

Let Ω ⊆ R N be an open set, let m ∈ N , and let 1 ≤ p ≤ ∞ . If u ∈W m , p (Ω) and

T : R → R is a rigid motion, then u ∘T ∈ W ( U ; R ) , where U ≔T ( Ω ) and compute

N N m, p N −1

‖u ∘ T‖W m ,p
(U) .

Proof:

Assume that,

Let Ω ⊆ R N be an open set, let m ∈ N , and let 1 ≤ p ≤ ∞ . If u ∈W m , p (Ω) and

N N
T : R → R is a rigid motion

To prove that, u ∘T ∈ W m , p ( U ; R N ) , where U ≔T −1 ( Ω ).

We need to show that u ∘T has weak derivatives up to order m in L p ( U ; R N ) .

Since u ∈W m , p ( Ω ) , it means that all partial derivatives of u up to order m are in

Lp (Ω ; RN ) .

Now, let’s consider the composition u ∘T .

By the chain rule,

the derivatives of u ∘T can be expressed in terms of the derivatives of u and the

derivatives of T.

Since T is a rigid motion, its derivatives are also rigid motions, which preserve distances
and angles. Therefore, the composition u ∘T inherits the properties of u in terms of weak
differentiability.

Thus, u ∘T has weak derivatives up to order m, and these derivatives are in L p ( U ; R N ) .

 Hence, u ∘T ∈ W m , p ( U ; R N ) .

Next to compute that, ‖u ∘ T‖W m ,p

(U) .

We can use the chain rule again to relate the ‖u‖ ∈W m , p ( Ω) to the ‖u ∘ T‖∈ W m , p ( U ) .
Since T is a rigid motion, it doesn’t affect the norm.

Therefore, ‖u ∘ T‖W m ,p
(U) =‖u‖W m ,p
( Ω) .

THEOREM 2.1:

Let Ω ⊆ R N be an open set, let m ∈ N , and let 1 ≤ p ≤ ∞ . Then the space W m , p (Ω) is a
Banach space.

Proof:

Assume that, Let Ω ⊆ R N be an open set, let m ∈ N , and let 1 ≤ p ≤ ∞ .

To prove that, the space W m , p ( Ω) is a Banach space.

Let { u n }n be a Cauchy sequence in W m , p ( Ω ) .

(i.e)., 0=l ,nlim

→∞
‖u l−un‖W m ,p

¿ lim
l ,n → ∞ (‖ul −un‖L + ∑ ‖∂α u l−∂α un‖L .
p

1 ≤|α|≤ m
p
)
Since L p (Ω) is a Banach space, there exists u , v α ∈ L p ( Ω) such that

lim ‖u l−un‖L =0 , lim ‖∂α u n−v α‖L =0

p p

n→∞ n→∞

for every multi-index α with 1 ≤|α|≤ m. Fix a multi-index α with 1 ≤|α|≤ m.

We claim that ∂ α u=v α . To show this, let ϕ ∈C ∞c (Ω) and

∫ ϕ ∂ α un dx=(−1 )|α|∫ u n ∂α ϕdx .

Ω Ω

Letting n → ∞ in the above equality which yields,

 ∫ ϕ v α dx= (−1 )|α|∫ u ∂α ϕdx
Ω Ω

for all ϕ ∈C ∞c ( Ω ) , which proves that the claim ∂ α u=v α .

Thus, u ∈W m , p ( Ω )

Hence, the space W m , p ( Ω) is a Banach space.

PROPERTY 2.5:

Let Ω ⊆ R N be an open set and let 1 ≤ p ≤ ∞ . Then

(i) A subset of a separable metric space is separable.

m,p
(ii) W (Ω) is separable.

Proof:

Assume that, Let Ω ⊆ R N be an open set and let 1 ≤ p ≤ ∞ .

(i) To prove that a subset of a separable metric space is separable :

To show this, we use the fact that a separable metric space contains a countable dense
subset.

Let X be a separable metric space, and let A ⊆ X be a subset of X.

Since D is a dense in X, every point in A is arbitrarily close to some point in D.

Thus, A ∩ D is dense in A, and since A ∩ D is countable (as a subset of D), A is

separable.

(ii) To prove W m , p ( Ω) is separable:

To show this, we can consider the mapping:

m,p
W ( Ω ) → L p ( Ω ) × … × L p ( Ω)
 u ↦ ( u , ∂α u , … ,∂ α u ) ,where α 1 , … , α M are all the multi-indices α with 1 ≤|α|≤ m.
1 M

Each component of this mapping is a function in L p (Ω).

Since L p ( Ω) is separable (as Ω is open and thus has a countable dense subset), the
Cartesian product of countably many separable spaces is also separable.

Therefore, W m , p (Ω) is separable.

Hence proved.

PROPERTY 2.6:

Let Ω ⊆ R N be an open set. Then W m ,∞ ( Ω) is not separable.

is not separable.

Proof

Assume that, Let Ω ⊆ R N be an open set, let m ∈ N and let 1 ≤ p ≤ ∞ . The space W mo , p (Ω)
is defined as the closure of the space C ∞o (Ω) in W m , p (Ω) (with respect to the topology of
m,p
W (Ω) ¿ .

To prove that, W m ,∞ ( Ω) is not separable.

It is enough to prove that, it contains uncountably many disjoint balls.

Consider the open set Ω=( 0 ,1 ) ∈ R N and the function f n ( x )=x n for n ∈ N . And each of
these functions is in W m ,∞ ( Ω ) . Since they are continuous on Ω and have bounded derivatives.

Now, consider the balls B(f n , ϵ)centered at each of these functions, where ϵ <1. Each of
these balls is disjoint, and there exists an uncountable family of such balls indexed by n and ϵ .

Since the family of balls is uncountable and each ball contains a function in W m ,∞ ( Ω ) ,
thereforeW m ,∞ ( Ω ) cannot be separable.

Hence proved.
REMARK:

It is important to observe that since uniform convergence preserves continuity, functions

in W m0 ,∞ ( Ω ) are necessarily of class C m ( Ω ) . In particular, piecewise affine functions do not belong
to W 10 , ∞ ( Ω ) . This is the reason why some refer to W 10 , ∞ (Ω) as the closure of the space C ∞c (Ω) in
(Ω) with respect to the weak star topology of W 1 , ∞ ( Ω ) , rather than the strong topology.
1,∞
W
These two topologies are not equivalent which needed when we applying results for W 10 , ∞ ( Ω ) .

When we working with domains Ω of infinite measure, the Sobolev space is right choice.

The below property shows, there exist solutions of the Dirichlet problem for the Laplacian with
p
the condition that u ∈ Lloc ( Ω ) , ∇ u ∈ L p ( Ω ; R N ) ,but W 1 , p ( Ω ) .

PROPERTY 2.7:

Let Ω ≔{x ∈ R N :‖x‖> 1} and let u ( x ) ≔1−‖x‖

2− N
, x ∈ Ω , if N ≥3 ,and

u ( x ) ≔ log ‖x‖, x ∈ Ω , if N=2. Then

(i) u is a classical solution of the Dirichlet problem Δ u=0∈Ω , u=0 on ∂ Ω.

q
(ii) u ∈ Lloc (Ω) for all 1 ≤ q<∞ , but u ∉ Lq ( Ω ) for any 1 ≤ q<∞ .
N
(iii) ∇ u ∈ L p ( Ω; R N ) for all < p <∞ .
N −1
2
∂ u p
(iv) ∈ L ( Ω) for all 1< p< ∞ .
∂ xi ∂ x j

Proof:

Assume that, Let Ω ≔{x ∈ R N :‖x‖> 1} and let u ( x ) ≔1−‖x‖

2− N
, x ∈ Ω , if

N ≥3 ,and u ( x ) ≔ log ‖x‖, x ∈ Ω , if N=2.

To prove (i):

We have show that, u is a classical solution of the Dirichlet problem.

 It is enough to prove that, u is twice continuously differentiable and satisfies
Δ u=0∈Ω , u=0 on ∂ Ω.

For N ≥3 ,u ( x )=1−‖x‖2− N is twice continuously differentiable in Ω ,∧ Δ u=0∈Ω, since u

is harmonic. Also, u=0 on ∂ Ω since |x|=1 on ∂ Ω .

For N=2, u ( x )=log|x| is twice continuously differentiable in Ω ,∧ Δ u=0∈Ω since u is

harmonic. Also, u=0 on ∂ Ω since |x|=1 on ∂ Ω.

To prove (ii):

To show that u ∈ Lqloc (Ω) for all 1 ≤ q<∞ , we note the fact that u is locally integrable since
it is bounded on compact subsets of Ω .

Now we have to claim that, u ∉ Lq ( Ω ) for any 1 ≤ q<∞ , we consider the behavior of u
near the singularity at |x|=1.

As |x|→1 , u ( x )grows without bound, implying that the integral of u over Ω diverges.
Therefore, u ∉ Lq (Ω) for any 1 ≤ q<∞ .

To prove (iii):

N
To show that, ∇ u ∈ L p ( Ω; R N ) for all < p <∞ , we have that ∇ u is continuous on Ω ,
N −1
1
and for N ≥3 , it is bounded since u is bounded. For N=2, |∇ u|= is integrable near |x|=1, so
|x|
∇ u ∈L p ( Ω; R N ) .

To prove (iv):

2 2
∂ u p ∂u
To show that, ∈ L (Ω) for all 1< p< ∞ , we note that for N ≥3 , is
∂ xi ∂ x j ∂ xi ∂ x j
continuous on Ω , and for N=2, it is integrable near |x|=1.

2
∂ u p
Thus, ∈ L ( Ω) .
∂ xi ∂ x j
 Hence proved.

DEFINITION 2.4:

Let Ω ⊆ R N be an open set, let m ∈ N , and 1 ≤ p< ∞ . The homogenous Sobolev space
m,p 1
Ẇ (Ω) is the space of all functions u ∈ Lloc (Ω) whose αth weak derivative ∂ α u belongs to
L (Ω) for every α ∈ N 0 with |α|=m. And the inclusion
p N

m,p
W ( Ω ) ⊆ Ẇ m , p (Ω)

The space Ẇ m , p ( Ω) is equipped with semi-norm

|u|❑Ẇ m ,p
( Ω)
≔‖∇m u‖L ( Ω) .
p

Sometimes we will also use the equivalent semi-normu ↦ ∑ ‖∂ u‖L ( Ω ) . And by

α
p

|α|=m

observing ¿|u|❑Ẇ m ,p
( Ω)
=0if and only if u is a polynomial of degree less than or equal to m-1 in

each connected component of Ω . When Ω has finitely many connected components and, in
particular, when it is connected, then we can define a norm in Ẇ m , p ( Ω ) .

PROPERTY 2.8:

Let Ω ⊆ R N be an open set with finitely many connected components Ω1 , … , Ωl , let

m ∈ N , and let 1 ≤ p< ∞ . For each i=1 , … , l consider a ball Bi ⋐ Ωi and let U ≔¿ i=1 ¿ l Bi . Then
Banach space with the norm

m−1
|u|❑Ẇ m ,p
( Ω)
≔ ∑ ‖∇k u‖L ( U ) +¿‖∇m u‖L ( Ω) , ¿ where ∇ 0 u≔u .
1 p

k=0

Proof:

Assume that, Let Ω ⊆ R N be an open set with finitely many connected components
Ω1 , … , Ωl , let m ∈ N , and let 1 ≤ p< ∞ . For each i=1 , … , l consider a ball Bi ⋐ Ωi and let
U ≔¿ i=1 ¿ l Bi .
 To prove that, Ẇ m , p (Ω) is a Banach space with the given norm, we first need to show
that it’s a complete normed space, (i.e)., a Banach soace.

(i) Completeness:

Let {u n } be a Cauchy sequence in Ẇ m , p ( Ω), (i.e)., for every ϵ >0 , there exists N such that
for all n , m> N , we have ‖u n−um‖Ẇ m ,p
( Ω)
<ϵ .

Now, decompose { u n } into its sequence of derivatives up to order m, un , ∇ u n , … , ∇ m u n .

Since L p ( Ω) is complete, the sequence ∇ k un converges in L p ( Ω) for k =0 ¿ m.

Hence, the sequence {u n } converges to some limit u in Ẇ m , p ( Ω ) , since all the derivatives
converge in L p ( Ω ) .

(ii) Norm properties:

(a) Non-negativity:

‖u‖Ẇ m, p
(Ω ) ≥ 0 for all u ∈ Ẇ m , p ( Ω ) , and ‖u‖Ẇ m, p
(Ω ) =0 if and only if u=0

almost everywhere in Ω .

(b) Homogeneity:

‖αu‖Ẇ m ,p
(Ω) =|α |.‖u‖Ẇ m ,p
( Ω) for all u ∈ Ẇ m , p ( Ω )∧α ∈ R .

(c) Triangle Inequality:

‖u+ v‖Ẇ m ,p
( Ω) ≤‖u‖Ẇ m ,p
( Ω) +‖v‖Ẇ m ,p
(Ω) for all u , v ∈ Ẇ m , p ( Ω ) .

Thus, Ẇ m , p ( Ω) is a Banach space with the given norm

m−1
|u|❑Ẇ m ,p
( Ω)
≔ ∑ ‖∇k u‖L ( U ) +¿‖∇m u‖L ( Ω) , ¿ where ∇ 0 u≔u .
1 p

k=0

Hence proved.
The following property shows that when Ω has infinitely many connected components, then
there is no norm in Ẇ m , p (Ω) which is compatible with the natural convergence in Ẇ m , p ( Ω ) ,
which is strong convergence in L p ( Ω) for the partial derivatives of order m and convergence in
1
Lloc (Ω) for the functions and their derivatives of order less than m.

PROPERTY2.9:

Let Ω ⊆ R N be an open set with infinitely many connected componentsΩn , n ∈ N , and let
1 ≤ p< ∞ . Then there cannot exist a norm‖. ‖Ẇ 1 ,p
( Ω) ∈ Ẇ 1 , p ( Ω )compatible with the convergences
un →u in L1loc ( Ω ) and ∇ u n → ∇ u in L p ( Ω ; R N ) .

Proof:

Assume that, for contradiction, that there exists a norm ‖. ‖❑Ẇ 1 ,p

(Ω) in Ẇ 1 , p (Ω) compatible

(Ω) and ∇ u → ∇ u in L p ( Ω ; R N ) .
p
with the convergences un →u in Lloc

For each n ∈ R N , consider a ball Bn ⊂ Ωn , where Ωn is one of the infinitely many

connected components of Ω .

Now, for each n, let’s construct a piecewise constant function un supported in Bn such that
‖u n‖Ẇ 1 ,p =1.

By the compatibility of the norm with the convergences, we can find such un such that
(Ω) and ∇ u n → 0 in L p ( Ω; R N ) as n → ∞ .
p
un → 0 in Lloc

Now, let u be the limit of un in Ẇ 1 , p ( Ω ) . Since ‖u n‖❑Ẇ 1, p

( Ω)
=1 for all n, by the continuity

of the norm, ‖u‖Ẇ =1. Thus, we have constructed a function u ∈ Ẇ 1 , p (Ω) such that ‖u‖Ẇ =1,
1 ,p 1 ,p

contradicting the assumption that such a norm exists.

Therefore, there cannot exist a norm ‖. ‖Ẇ in Ẇ 1 , p ( Ω ) compatible with the given
1 ,p

convergences.

PROPERTY 2.10:
 Let Ω ⊆ R N be an open set and let m ∈ N .

ν 1 +t ν 2
(i) Let ν 1 , ν 2 ∈ S N−1 and let ν ≔ for t ∈ R ,with ν 1+t ν 2 ≠ 0. Then
‖ν 1+ tν2‖

( )
m m m−k k
∂ u
=∑ m
∂ ∂ u k
()
m
‖ν 1+ tν2‖ m m−k k
t
∂ ν k=0 k ∂ ν 1 ∂ ν2

for every u ∈C m ( Ω ) .

(ii) By taking a finite number of different values of t ∈ R, then for each k =1 , … ,m ,

( )
m−k k
∂ u ∂ u
m−k k can be written as a linear combination of a finite number of directional
∂ ν1 ∂ ν2
m
∂ u
derivatives ,
m with as in part (i).
∂ν

(iii) Finally there exists a finite set Sm ⊂ S N−1 such that for every u ∈C m ( Ω ) ,
m
∂ u
∂ u= ∑ c α , ν
α
m
ν∈ Sm ∂ν

for every multi-index α ∈ N 0N , with |α|=m, and for some constants c α , ν ∈ R independent of u.

PROPERTY 2.11:

Let Ω ∈ R N be an open set, let m ∈ N , and let 1 ≤ p ≤ ∞ . Then the following are equivalent
semi-norms in Ẇ m , p (R N ):

( ‖ ‖ ) ‖ ‖ ‖ ‖
p 1
m m m
∂ u ∂ u ∂ u
∫ ,∑
N −1 p
dH (ν) , ¿ ,
S
N−1 ∂ νm L
p ν ∈S
N −1
∂ νm L
p ν ∈Sm ∂ νm L
p

Where Sm is the finite set of S N−1(which is the unit sphere ∂ B ( 0 ,1 ) ¿ .

PROPERTY 2.12:

Given n ∈ N , let V be the vector space of all polynomials P : [ 0 ,1 ] → R of degree less

than or equal to n.
 (i) Given P ∈V , let ‖P‖≔ max {|a0|, … ,|an|} , where P ( t ) =a0 + …+an t n ,

t ∈ [ 0 ,1 ] . Then ‖. ‖ is a norm in V .

(ii) Let 1 ≤ p ≤ ∞ and P ∈V be such that ‖P‖L ( [ 0 ,1 ]) ≤1. Then there exists a constant
p

c=c ( n )> 0 such that |ak|≤ c for all k =1 , … ,n .