Exercise 1.

5 Let Ω ⊂ IR2 be a bounded domain with Lipschitz continuous boundary Γ and let
Γ be decomposed into two non-empty curves Γ1 and Γ2 : Γ = Γ̄1 ∪ Γ̄2 and Γ1 ∩ Γ2 = ∅. For
sufficiently smooth functions f and g give a variational formulation of the mixed boundary
value problem
−∆u + u = f in Ω,
u=0 on Γ1 ,
∂n u = g on Γ2 .

2 Unique solvability of variational formulations

In this section we deal with existence and uniqueness of a solution to the problem

u ∈ V : a(u, v) = L(v) ∀v ∈ V. (2.1)

Here, V is a Hilbert space with inner product h·, ·i and norm k · k, a(·, ·) is a bilinear form and
L : V → IR is a linear form. We will need some properties of a and L.

Definition 2.1 1. The linear form L : V → IR is called continuous or bounded if

∃C > 0 : |L(v)| ≤ Ckvk ∀v ∈ V.

2. Any continuous linear form V → IR is called a linear functional on V . The space consisting
of all linear functionals V → IR is called dual space of V and is denoted by V ′ or L(V, IR).
Its norm is defined by
|L(v)|
kLkV ′ := sup for L ∈ V ′ .
v∈V \{0} kvk

3. The bilinear form a : V × V → IR is called continuous or bounded if

∃Ca > 0 : |a(v, w)| ≤ Ca kvk kwk ∀v, w ∈ V.

4. The bilinear form a : V × V → IR is called V -elliptic (or just elliptic if the space is clear)
if
∃α > 0 : a(v, v) ≥ αkvk2 ∀v ∈ V.

We now have all the properties of bilinear and linear forms to formulate the main result of
this section (Theorem 2.1 below). However, for its proof we need two more classical results from
functional analysis.

Proposition 2.1 (Banach fixed point theorem)

Let V be a Banach space (a complete vector space not necessarily having an inner product) and
let φ : V → V be a contraction, i.e.

∃c, 0 ≤ c < 1 : kφ(v) − φ(w)k ≤ ckv − wk ∀v, w ∈ V.

13
Then there exists a unique u ∈ V such that

φ(u) = u.

Proposition 2.2 (Riesz representation theorem)

Let V be a Hilbert space with inner product h·, ·i and norm k · k. Any element w ∈ V defines
a continuous linear form Lw ∈ V ′ by Lw (v) := hw, vi. On the other hand, for any continuous
linear form L ∈ V ′ there exists a unique element RL ∈ V such that

L(v) = hRL, vi ∀v ∈ V.

Moreover, there holds kRLk = kLkV ′ , i.e.

kRGk
kRkV ′ →V := sup = 1.
G∈V ′ \{0} kGkV ′

Theorem 2.1 (Lax-Milgram) Let V be a Hilbert space, a(·, ·) a continuous, V -elliptic bilinear
form and L a continuous linear form on V . Then the variational problem (2.1) has a unique
solution u ∈ V .

Proof. By the continuity of a we obtain that for any fixed u ∈ V the mapping

Au : v 7→ Au(v) := a(u, v)

is linear and bounded:

Au(v) = a(u, v) ≤ Ca kuk kvk ≤ Ckvk ∀v ∈ V with C := Ca kuk.

This means that

|Au(v)|
kAukV ′ = sup ≤ Ca kuk ∀u ∈ V,
v∈V \{0} kvk
i.e. A : V → V ′ is linear and continuous:
kAukV ′
kAkV →V ′ := sup ≤ Ca .
u∈V \{0} kuk

By the Riesz representation theorem there exists for any element G ∈ V ′ (i.e. any continuous
linear form G : V → IR) a unique element RG ∈ V such that

G(v) = hRG, vi ∀v ∈ V.

Therefore, our problem (2.1) is equivalent to

u ∈ V : RAu = RL.

14
We show that for sufficiently small ρ > 0 the mapping

V → V
Cρ :
v 7→ v − ρ(RAv − RL)

is a contraction. The Banach fixed point theorem then yields the unique existence of u ∈ V
such that
u − ρ(RAu − RL) = u, i.e. RAu = RL,
which proves the theorem.
To establish that, for small ρ, Cρ is a contraction we use the ellipticity of a (with constant
α > 0), the norm property kRkV ′ →V = 1 and the boundedness kAkV →V ′ ≤ Ca . This yields

kv − ρRAvk2 = hv − ρRAv, v − ρRAvi = kvk2 − 2ρhRAv, vi + ρ2 kRAvk2

= kvk2 − 2ρ a(v, v) + ρ2 kRAvk2
≤ kvk2 − 2ραkvk2 + ρ2 kRk2V ′ →V kAk2V →V ′ kvk2 ≤ (1 − 2ρα + ρ2 Ca2 )kvk2 ,

i.e. Cρ is a contraction for ρ ∈ (0, 2α/Ca2 ). 2

Remark 2.1 The proof of the Lax-Milgram lemma implies that the mapping A : V → V ′ is an
isomorphism (linear and bijective). Since

αkvk2 ≤ a(v, v) = Av(v) ≤ kAvkV ′ kvk ∀v ∈ V

one finds that the inverse A−1 of A is continuous with norm

kA−1 Gk
kA−1 kV ′ →V = sup ≤ α−1 .
G∈V \{0}
′ kGkV ′

It follows that the variational formulation (2.1) is well-posed in the sense that there exists a
unique solution which depends continuously on the data (on L):

kuk = kA−1 Lk ≤ α−1 kLkV ′ .

Exercise 2.1 Show, by using the Lax-Milgram lemma, that (1.15) has a unique solution provided
that f ∈ L2 (Ω).

Exercise 2.2 Under appropriate conditions on f and g, prove existence and uniqueness of the
solution to the variational formulation found in Exercise 1.5.
Hint: Use without proof that

∃C > 0 : kvkL2 (Γ) ≤ CkvkH 1 (Ω) ∀v ∈ H 1 (Ω).

15
3 Abstract error estimate for the finite element method
Let us recall the setting. The aim is to find an approximative solution to the continuous problem
(2.1),
u ∈ V : a(u, v) = L(v) ∀v ∈ V,
where we use the notation from before: V is a Hilbert space with inner product h·, ·i and norm
k · k, a(·, ·) is a bilinear form and L : V → IR is a linear form.
The discrete version is as follows. For a given finite-dimensional subspace Vh ⊂ V find
uh ∈ Vh such that
a(uh , v) = L(v) ∀v ∈ Vh . (3.1)

Theorem 3.1 Let a(·, ·) be a continuous and V -elliptic bilinear form and let L ∈ V ′ (i.e. L
is a continuous linear form on V ). Then the discrete variational problem (3.1) has a unique
solution uh ∈ Vh and there holds the stability estimate

kuh k ≤ α−1 kLkV ′ . (3.2)

Proof. Since Vh is a subspace of V the continuity of a and L and the ellipticity of a remain
true on Vh as forms a : Vh × Vh → IR and L : Vh → IR. Therefore, the unique existence of
uh follows from the Lax-Milgram lemma (Theorem 2.1) and the stability estimate is a discrete
version of Remark 2.1. 2

The next theorem is the basis for error estimates of the finite element method.

Theorem 3.2 (Céa’s lemma, quasi-optimal error estimate) Let a be continuous and V -elliptic
bilinear form and let L ∈ V ′ . Then, the solutions u ∈ V and uh ∈ Vh of (2.1) and (3.1),
respectively, satisfy
Ca
ku − uh k ≤ ku − vk ∀v ∈ Vh .
α
Here, α and Ca are the ellipticity and continuity constants of a, respectively (see §2).

Proof. If ku − uh k = 0 then there is nothing to prove. Subtracting the equations of the con-
tinuous and discrete variational formulations, (2.1) and (3.1), yields the Galerkin orthogonality

a(u − uh , w) = 0 ∀w ∈ Vh .

We select an arbitrary v ∈ Vh and define w := uh − v ∈ Vh so that v = uh − w. Then, using the

ellipticity of a, the Galerkin orthogonality and the continuity of a, we find that there holds

αku − uh k2 ≤ a(u − uh , u − uh ) = a(u − uh , u − uh ) + a(u − uh , w)

= a(u − uh , u − uh + w) = a(u − uh , u − v) ≤ Ca ku − uh k ku − vk.

16
Dividing by ku − uh k > 0 and α > 0 gives the result. 2

Céa’s lemma provides an abstract error estimate by a term that includes the unknown
solution u. However, it states two important facts. First, selecting any function v ∈ Vh , the
norm ku − vk is, up to a constant factor (independent of Vh ), an upper bound for the error
ku − uh k. Therefore, based on Céa’s lemma more specific error estimates can be derived if
certain properties of u (regularity) are known. Second, Céa’s lemma states that the finite
element solution uh is almost the best approximation of u among elements of Vh . (“Almost”
refers to the factor Ca /α.) Céa’s lemma is therefore also called a quasi-optimal error estimate.
Note that the estimate can be formulated equivalently by
Ca
ku − uh k ≤ min ku − vk
α v∈Vh
and that minv∈Vh ku − vk is the distance of u to Vh (in the norm of V ). Thus, the finite
element method has the remarkable property of delivering the (almost) best approximation
of an unknown function. Of course, this fact originates from the particular type of (elliptic)
problems we are studying.

3.1 The energy norm

Let us assume that the bilinear form a : V × V → IR is symmetric and positive definite. This
means in fact that a is an inner product on V inducing a norm
p
kvka := a(v, v), v ∈ V.

This norm is called energy norm. Its name has a physical motivation where F (v) := 12 a(v, v) −
L(v) relates to the energy of a physical system. By the ellipticity and continuity of a there holds

αkvk2 ≤ a(v, v) = kvk2a = a(v, v) ≤ Ca kvk2 ∀v ∈ V,

that is
α1/2 kvk ≤ kvka ≤ Ca1/2 kvk ∀v ∈ V. (3.3)
Therefore, k · k and k · ka are equivalent norms in V . The Galerkin orthogonality

a(u − uh , v) = 0 ∀v ∈ Vh

then is in fact an orthogonality: the finite element error u − uh is orthogonal to Vh with respect
to the inner product a(·, ·). As we have seen in §1.3, the Galerkin orthogonality is equivalent to
the best approximation property with respect to the norm induced by the inner product, in this
case
ku − uh ka ≤ ku − vka ∀v ∈ Vh . (3.4)
Therefore, in the case of a symmetric, elliptic bilinear form, Céa’s lemma (Theorem 3.2) can be
improved to a best approximation property by switching from the norm k · k to the energy norm
k · ka .

17