1 Dual Space of W01,p(I)

We denote by W 1,p (I) the dual space of W01,p (I), 1 p < and by H 1 (I) the
0

dual of H01 (I).

Remark: By identifying L2 (I) with its dual, we obtain
H01 (I) L2 (I) H 1 (I)
with continuous and dense embedding.
0 0
Theorem: Let F be in W 1,p (I). Then there exist f0 , f1 in Lp (I) such that
Z Z
hF, vi = f0 v + f1 v 0 , v W01,p (I).

Moreover, if I is bounded, f0 can be taken zero.

Proof : Define the Banach space E = Lp Lp equipped with the norm
khkE = kh0 kp + kh1 kp , h = (h0 , h1 ) E.
T : W01,p (I) E given by T (u) = (u, u0 ) is an isometry.
Let G = T (W01,p (I)) equipped with the norm of E. We define the linear form
: G IR by
(h) = hF, T 1 hi.
It is easy to see that is continuous. So, it can be extended to E by Hahn-Banach
theorem. Call the extension; hence kk = kk = kF k. So by Riesz Representation
0
Theorem, there exists f0 , f1 Lp (I), such that
Z
(h0 , h1 ) = f0 h0 + f1 h1 , (h0 , h1 ) E.

In particular, if u W01,p (I), then

F (u) = hF, T 1 (u, u0 )i
Z
= (u, u0 ) = f0 u + f1 u0 .

When I is bounded, W01,p (I) is equipped with the norm kukw1,p = ku0 kp . We then
repeat a similar reasoning with T : W01,p (I) Lp (I) given by T (u) = u0 .
Remark: f0 and f1 are not unique.
Remark: If v C0 (I), then
Z Z Z
hF, vi = f0 v + f1 v 0 = f0 v f10 v = (f0 f10 )v.

Therefore
F = f0 f10 in D0 (I).
Exercise: Verify that given in the above proof satisfies
kkE 0 = max{kf0 kp0 , kf1 kp0 }.

1
1.1 Bilinear forms and Lax-Milgram Lemma

Definition: Let B : H H IR be a bilinear form on a Hilbert space H. We say

that
1) B is continuous if there exists M > 0 such that

|B(u, v)| M kuk kvk.

2) B is coercive (or elliptic) if there exists > 0 such that

B(u, u) kuk2 , u H.

Theorem: (Lax-Milgram Lemma). Given a Hilbert space H, let B : H H IR be

a continuous and coercive bilinear form and g : h IR be a continuous (bdd) linear
form. Then there exists a unique u in H such that

g(v) = B(u, v), v H.

Application: Consider the problem

(
u00 + u = f in I = (0, 1)
(1.1)
u(0) = u(1) = 0.

For f smooth enough (continuous). This problem can be solved by standard calculus
methods. In this case the solution is of class C 2 (I). It is called a classical solution.
Suppose that f is not regular; say f L2 (I) or f H 1 (I) = dual of H01 (I). Is there
any solution for (1.1)?
Let C01 (I). Multiply by equation (1.1) and integrate over I, assuming u to
be regular, Z Z
1 1
u0 0 + u = f . (1.2)
0 0
Question; Is it possible to find u such that (1.2) is satisfied for all C01 (I)?
Answer: Define the bilinear form B : H01 (I) H01 (I) IR by
Z 1
B(u, v) = u0 v 0 + uv.
0

It is easy to verify that B is continuous and coercive. If f H 1 (I), we then define

the linear form
F : H01 (I) by F (v) = hf, vi.
This is continuous such that kF k = kf k1 . Lax-Milgram lemma guarantees the
existence of a unique u H01 (I) such that

B(u, v) = F (v), v H01 (I).

That is,
Z 1 Z 1
u0 v 0 + uv = hf, vi or f v, if f L2 (I) , v H01 (I).
0 0

2
Definition: We call the weak formulation of (1): find u in H01 (I):
Z 1
u0 v 0 + uv = hf, viH 1 H01 (I) , v H01 (I). (1.3)
0

Definition: We call u H01 (I) satisfying (1.3), the weak solution of (1).
Remark: Since u H01 (I), therefore u is in C(I); hence u(0) = u(1) = 0.
Proposition: If f L2 (I), then u00 = u f L2 (I). Thus u H 2 (I) H01 (I). So,
we have more regularity that is u C 1 (I ).
Proof : C01 (I) Z Z
1 1
u0 0 = (u f ),
0 0

so by definition, u0 has a weak derivative u f L2 (I) u0 H 1 (I), with u00 =

u f L2 (I).
u H01 (I) H 2 (I).
The embedding theorem gives u0 C(I); hence u C 1 (I).
Exercise: 1) Show that

u00 = in I = (1, 1),

u(1) = u(1) = 0

has a solution.
2) Solve

u00 + u = f in I = (0, 1)
u(0) = u(1) = .

for f L2 (I); , IR.

1.2 Neumann Problem

Consider the homogeneous Neumann-condition problem
(
u00 + u = f, 0<x<1
0 0 (1.4)
u (0) = u (1) = 0.

If u is a classical solution of (1.4), then for each v H 1 (I), we have

Z 1 Z 1
0 0
u v + uv = f v. (1.5)
0 0

Again by Lax-Millgram lemma, we have a solution u in H 1 (I).

If f L2 (I), then u is in H 2 (I) since, for each C01 (I), we have
Z 1 Z 1
0 0
u = (f u),
0 0

3
then u0 H 1 (I) with u00 = f u. From (1.5), we obtain
Z 1
(u00 + u f )v + u0 (1)v(1) u0 (0)v(0) = 0, v H 1. (1.6)
0

First, for v H01 (I), we have

Z 1
(u00 + u f )v = 0.
0

So u00 + u f = 0 in L2 (I), hence for almost each x I. Thus (1.6) is reduced to

u0 (1)v(1) u0 (0)v(0) = 0, v H 1 (I).

Since v is arbitrary, then u0 (1) = u0 (0) = 0.