C6.

4 Finite Element Methods for PDEs

Sheet 1 — HT23
Lectures 1–5

Section A
1. Let V be a real vector space. A bilinear form a(u, v) is said to be skew-symmetric if
a(u, v) = −a(v, u). It is said to be alternating if a(u, u) = 0 for all u ∈ V .

(a) Show that every bilinear form on V may be written uniquely as the sum of a
symmetric bilinear form and a skew-symmetric bilinear form.

(b) Show that a bilinear form on V is alternating if and only if it is skew-symmetric.

Solution:

(a) Assume existence of a decomposition a = b + c where b is symmetric and c is

skew-symmetric. Then for u, v ∈ V ,

a(u, v) = b(u, v) + c(u, v),

and

a(v, u) = b(u, v) − c(u, v).

Adding and substracting both equations, we get:

a(u, v) + a(v, u) a(u, v) − a(v, u)

b(u, v) = , c(v, u) = .
2 2
The bilinear forms b and c are symmetric and skew-symmetric, respectively, and
are the unique such forms which sum to a. The existence of a decomposition clearly
holds.

(b) Let a be an alternating bilinear form on V . Then for u, v ∈ V

0 = a(u + v, u + v) = a(u, u) + a(u, v) + a(v, u) + a(v, v) = a(u, v) + a(v, u).

Therefore a(u, v) = −a(v, u).

Let a be a skew-symmetric bilinear form on V . Then for u ∈ V , a(u, u) = −a(u, u),

so 2a(u, u) = 0. Therefore a(u, u) = 0.

Mathematical Institute, University of Oxford Page 1 of 6

Endre Süli: endre.suli@maths.ox.ac.uk
 C6.4 Finite Element Methods for PDEs: Sheet 1 — HT23

2. Let V be a real vector space and let (u, v) be an inner product on V . Let k · k be the
induced norm. By considering the expansion of ku + vk2 , express the inner product
purely in terms of norms.

Solution: Several possible answers:

1
ku + vk2 − ku − vk2


(u, v) =
4
1
ku + vk2 − kuk2 − kvk2


=
2
1
kuk2 + kvk2 − ku − vk2


=
2

These are possible because of the parallelogram law:

2kuk2 + 2kvk2 = ku + vk2 + ku − vk2 .

Mathematical Institute, University of Oxford Page 2 of 6

Endre Süli: endre.suli@maths.ox.ac.uk
 C6.4 Finite Element Methods for PDEs: Sheet 1 — HT23

Section B
3. Suppose u : Ω → R is a piecewise smooth function over a finite partition of a bounded
domain Ω; i.e., Ω = i Ki and u|KĎi ∈ C ∞ (K
Ďi ). Suppose that ∂Ki is a piecewise C 1
S

surface. Show that u ∈ C 0 (Ω) =⇒ u ∈ H 1 (Ω).

4. Given a vector field u : Ω → R3 on a domain Ω ⊂ R3 , its weak divergence ∇ · u : Ω → R

and curl ∇ × u : Ω → R3 are defined by the following identities:
Z Z
u · ∇φ dx = − ∇ · uφ dx ∀ φ ∈ C0∞ (Ω),
Z Ω ZΩ
u · ∇ × φ dx = + (∇ × u) · φ dx ∀ φ ∈ C0∞ (Ω; R3 ).
Ω Ω

Let u : Ω → R3 be a piecewise smooth vector field over a finite partition of a bounded

domain Ω, i.e., Ω = i Ki and u|KĎi ∈ C ∞ (K
Ďi ). Suppose that ∂Ki is a piecewise C 1
S

surface.

(a) Prove that u ∈ H(div, Ω) if its normal component is continuous across cells.

(b) Prove that u ∈ H(curl, Ω) if its tangential components are continuous across cells.

5. Let Ω ⊂ Rd (d ∈ {2, 3}) be an open bounded Lipschitz domain with ∂Ω and outward-
pointing boundary normal n. Let f ∈ L2 (Ω). Consider the sign-positive Helmholtz
equation

−∇2 u + k 2 u = f in Ω,
∇u · n = 0 on ∂Ω,

where k ∈ R with |k| < 1.

(a) State the variational formulation of this problem.

(b) For what values of k is the bilinear form coercive with respect to the norm k·kH 1 (Ω) ?
When the problem is coercive, give the coercivity constant of the bilinear form you
have given in (a) in terms of k, and show that your coercivity bound is tight.

Mathematical Institute, University of Oxford Page 3 of 6

Endre Süli: endre.suli@maths.ox.ac.uk
 C6.4 Finite Element Methods for PDEs: Sheet 1 — HT23

6. The proof of Lax–Milgram given in lectures is constructive: given a u0 ∈ V , one can

apply the fixed-point iteration

uk+1 = T uk , k = 0, . . .

with T defined in (4.2.18) of the notes. By the contraction mapping theorem, uk → u,

the exact solution of the linear variational problem.

(a) For given coercivity and continuity constants α, C, what is the optimal choice of ρ
in the construction of T ?

(b) The constants for linear elasticity applied to steel are α = 75, C = 468. How many
iterations would be required to achieve a relative error of 0.1%? The relative error
is kuk − uk/ku0 − uk, for given initial guess u0 and iterate uk .

(c) The constants for linear elasticity applied to rubber are α = 0.018, C = 2.75. How
many iterations would be required in this case for the same tolerance?

7. Suppose V is a Banach space, and that it has a coercive bounded bilinear form a :
V × V → R. Show that V is in fact a Hilbert space.

Mathematical Institute, University of Oxford Page 4 of 6

Endre Süli: endre.suli@maths.ox.ac.uk
 C6.4 Finite Element Methods for PDEs: Sheet 1 — HT23

Section C
8. Let V be a Hilbert space. We say that a bilinear form a : V × V → R is positive-
semidefinite if a(v, v) ≥ 0 for all v ∈ V .

Assume that a is positive-semidefinite, bounded, symmetric, and that there exists a

constant γ > 0 such that
a(u, v)
γ ≤ inf sup .
u∈V v∈V kukV kvkV
u6=0 v6=0

This condition is referred to as an inf-sup condition.

Prove that these conditions imply that a is coercive. Characterise the coercivity con-
stant of a in terms of its continuity constant and the inf-sup constant γ.

[Hint: by considering a(v + λw, v + λw) for λ ∈ R, first prove that for all v, w ∈ V ,
p p
|a(v, w)| ≤ a(v, v) a(w, w),

and then use the inf-sup condition.]

p p
Solution: We first prove that a(v, w) ≤ a(v, v) a(w, w) for all v, w ∈ V . Consider
for λ ∈ R

0 ≤ a(v + λw, v + λw) = a(v, v) + a(v, λw) + a(λw, v) + λ2 a(w, w)

= a(v, v) + 2λa(v, w) + λ2 a(w, w).

The right-hand side is a quadratic polynomial in λ with real coefficients, and it is non-
negative for all λ ∈ R. Therefore its discriminant is non-positive; it can only be zero or
negative. Thus
(2a(v, w))2 − 4a(v, v)a(w, w) ≤ 0,

yielding the desired inequality.

To prove that well-posedness implies coercivity we argue as follows. We know from the

Mathematical Institute, University of Oxford Page 5 of 6

Endre Süli: endre.suli@maths.ox.ac.uk
 C6.4 Finite Element Methods for PDEs: Sheet 1 — HT23

inf-sup condition that

a(u, v)
γkuk ≤ sup
v∈V kvk
|a(u, v)|
≤ sup
v∈V kvk
p p
a(u, u) a(v, v)
≤ sup
v∈V kvk
p
p a(v, v)
= a(u, u) sup
v∈V kvk
p
p Ckvk2
≤ a(u, u) sup
v∈V kvk
p
= Ca(u, u)

and so
γ2
a(u, u) ≥ kuk2 .
C

Mathematical Institute, University of Oxford Page 6 of 6

Endre Süli: endre.suli@maths.ox.ac.uk