Finite Element Methods

Smartworld.asia Specworld.
in
Contents
Acknowledgements v
Preface vii
1 The Abstract Problem 1
2 Examples 9
3 The Finite Element Method in its Simplest Form 29
smartworlD.asia
4 Examples of Finite Elements 35
5 General Properties of Finite Elements 53
6 Interpolation Theory in Sobolev Spaces 59
7 Applications to Second-Order Problems... 67
8 Numerical Integration 77
9 The Obstacle Problem 95
10 Conforming Finite Element Method for the Plate Problem 103
11 Non-Conforming Methods for the Plate Problem 113
ix
Smartzworld.com 1 jntuworldupdates.org
Smartworld.asia Specworld.in
2 1. The Abstract Problem
Definition 1.1. Let V be a normed linear space. A bilinear form a(·, ·)

on V is said to be V-elliptic if there exists a constant α > 0 such that for
all v ∈ V.
(1.4) a(v, v) ≥ α||v||2 .
Theorem 1.1. Let V be a Banach space and K a closed convex subset

of V. Let a(·, ·) be V-elliptic. Then there exists a unique solution for the
problem (P).
Further this solution is characterised by the property:
(1.5) a(u, v − u) ≥ f (v − u) for all v ∈ K.
Remark 1.1. The inequalities (1.5) are known as variational inequali-

ties.
Proof. The V-ellipticity of a(·, ·) clearly implies that if a(v, v) = 0 then

v = 0. This together with the symmetry and bilinearity of a(·, ·) shows
that a(·, ·) defines an inner-product on V. Further the continuity and the
smartworlD.asia
V-ellipticity of a(·, ·) shows that the norm
1
(1.6) v ∈ V → a(v, v) 2
defined by the inner-product is equivalent to the existing norm on V.

Thus V acquires the structure of a Hilbert space and we apply the Riesz
representation theorem to obtain the following: for all f ∈ V ′ , there
exists σ f ∈ V such that
(1.7) f (v) = a(σ f, v) for all v ∈ V.
The map σ : V ′ → V given by f 7→ σ f is linear. Now,
1
J(v) = a(v, v) − f (v)
2
1
= a(v, v) − a(σ f, v)
2
1. The Abstract Problem 3
1 1
= a(v − σ f, v − σ f ) − a(σ f, σ f ).
2 2
The symmetry of a(·, ·) is essential in obtaining the last equality.
3 For a given f , since σ f is fixed, J is minimised if and only if a(v −
σ f, v − σ f ) is minimised. But this being the distance between v and
σ f , our knowledge of Hilbert space theory tells us that since K is a
closed convex subset, there exists a unique element u ∈ K such that this
minimum is obtained. This proves the existence and uniqueness of the
solution, which is merely the projection of σ f over K.
We know that this projection is characterised by the inequalities:
(1.8) a(σ f − u, v − u) ≤ 0 for all v ∈ K.
Geometrically, this means that the angle between the vectors (σ f −u)
and (v − u) is obtuse. See Fig. 1.1.
smartworlD.asia
Figure 1.1:
Thus, a(σ f, v − u) ≤ a(u, v − u) which by virtue of (1.7) is precisely

the relation (1.5). This completes the proof.
We can state the following
Corollary 1.1. (a) If K is a non-empty closed convex cone with ver-

tex at origin 0, then the solution of (P) is characterised by:



a(u, v) ≥ f (v) for all v ∈ K
(1.9) 

a(u, u) = f (u).
(b) If K is a subspace of V then the solution is characterised by 4
(1.10) a(u, v) = f (v) for all v ∈ K.
Remark 1.2. The relations (1.5), (1.9) and (1.10) are all called varia-
tional formulations of the problem (P).
Proof. (a) If K is a cone with vertex at 0, then for u, v ∈ K, u + v ∈ K.

(cf. Fig. 1.2). If u is the solution to (P), then for all v ∈ K applying
(1.5) to (u + v) we get a(u, v) ≥ f (v) for all v ∈ K. In particular
this applies to u itself. Setting v = 0 in (1.5) we get −a(u, u) ≥
− f (u) which gives the reverse inequality necessary to complete
the proof of (1.9). Conversely, if (1.9) holds, we get (1.5) by just
subtracting one inequality from the other.
smartworlD.asia
Figure 1.2:
(b) Applying (a) to K, since any subspace is a cone with vertex at 0,

we get (b) immediately. For if v ∈ K, then −v ∈ K and applying
(1.9) both to v and −v we get (1.10).
This completes the proof.
Remark 1.3. The solution u of (P) corresponding to f ∈ V ′ (for a fixed

a(·, ·)) defines a map V ′ → V. Since this solution is the projection of
σ f on K, it follows that the above map is linear if and only if K is

a subspace. The problems associated with variational inequalities are
therefore nonlinear in general.
Exercise 1.1. Let V be as in Theorem 1.1. For f1 , f2 ∈ V ′ , let u1 , u2 5

be the corresponding solutions of (P). If || · ||∗ denotes the norm in V ′ ,
prove that
1
||u1 − u2 || ≤ || f1 − f2 ||∗ .
α
Remark 1.4. The above exercise shows, in particular, the continuous
dependence of the solution of f , in the sense described above. This
together with the existence and uniqueness establishes that the problem
(P) is “well-posed” in the terminology of partial differential equations.
Exercise 1.2. If V is a normed linear space, K a given convex subset of V

and J : V → R any functional which is once differentiable everywhere,
then (i) if u ∈ K is such that J(u) = Min J(v), u satisfies, J ′ (u)(v − u) ≥ 0
v∈K
for all v ∈ K. (ii) Conversely, if u ∈ K such that J ′ (u)(v − u) ≥ 0 for
all v ∈ K, and J is everywhere twice differentiable with J ′′ satisfying
J ′′ (v)(w, w) ≥ α||w||2 , for all v, w ∈ K and some α ≥ 0, then J(u) =
smartworlD.asia
Min J(v).
v∈K
Exercise 1.3 (1 ). Apply the previous exercise to the functional

1
J(v) =
a(v, v) = f (v)
2
with a(·, ·) and f as in Theorem 1.1. If K is a subspace of V, show that
J ′ (u)(v) = 0 for all v ∈ K. In particular if K = V, J ′ (u) = 0.
It was essentially the symmetry of the bilinear form which provided

the Hilbert space structure in Theorem 1.1. We now drop the symmetry
assumption on a(·, ·) but we assume V to be a Hilbert space. In addition
we assume that K = V.
Theorem 1.2 (LAX-MILGRAM LEMMA). Let V be a Hilbert space. 6
1
Exercises (1.2)(i) and 1.3 together give relations (1.5)
a(·, ·) a continuous, bilinear, V-elliptic form, f ∈ V ′ . If (P) is the prob-

lem: to find u ∈ V such that for all v ∈ V,
(1.11) a(u, v) = f (v),
then (P) has a unique solution in V 1 .

Proof. Since a(·, ·) is continuous and V-elliptic, there are constants M,
α > 0 such that
|a(u, v)| ≤ M||u|| ||v||,
(1.12)
a(v, v) ≥ α||v||2 ,
for all u, v ∈ V. Fix any u ∈ V. Then the map v 7→ a(u, v) is continuous

and linear. Let us denote it by Au ∈ V ′ . Thus we have a map A : V → V ′
defined by u 7→ Au.
| Au(v)| |a(u, v)|
(1.13) || Au ||∗ = sup = sup ≤ M||u||.
v∈V ||v|| v∈V ||v||
v,0 v,0
Thus A is continuous and ||A|| ≤ M.

We are required to solve the equation
smartworlD.asia
(1.14) Au = f.
Let τ be the Riesz isometry, τ : V ′ → V so that
(1.15) f (v) = ((τ f, v)),
where ((·, ·)) denotes the inner product in V. Then, Au = f if and only
if τ Au = τ f or equivalently.
(1.16) u = u − ρ(τ Au −τ f ),
7 where ρ > 0 is a constant to be specified. We choose ρ such that

g : V → V is a contraction map, where g is defined by
(1.17) g(v) = v − ρ(τAv − τ f ) for v ∈ V.

1
cf. Corollary 1.1(b).
Then the solution to (P) will be the unique fixed point of this con-
traction map, which exists by the contraction mapping theorem.
Let v1 , v2 ∈ V. Set v = v1 − v2 . Then
||g(v1 ) − g(v2 )|| = ||(v1 − v2 ) − ρτA(v1 − v2 )||

= ||v − ρτAv||.
But,
||v − ρτAv||2 = ((v − ρτAv, v − ρτAv))

= ||v||2 − 2ρ((τAv, v)) + ρ2 ||τAv||2
2
= ||v||2 − 2ρAv(v) + ρ2 ||Av||∗
≤ ||v||2 − 2ρα||v||2 + ρ2 M 2 ||v||2
= (1 − 2ρα + ρ2 M 2 )||v||2
2α
since Av(v) = a(v, v) ≥ α||v||2 and ||A|| ≤ M. Choosing ρ ∈]0, 2 [, we
M
get that
(1.18) 1 − 2ρα + ρ2 M 2 < 1
smartworlD.asia
and hence g is a contraction, thus completing the proof.
Remark 1.5. The problem (P) of Theorem 1.2 is well-posed. The ex-

istence and uniqueness were proved in the theorem. For the continuous
dependence of u on f , we have
(1.19) α||u||2 ≤ a(u, u) = f (u) ≤ || f ||∗ · ||u||.
REFERENCE. For Variational Inequalities, see Lions and Stampac-

chia [18].
smartworlD.asia
Chapter 2
Examples
WE GIVE IN this section several examples of the abstract problem for- 9

mulated in Sec. 1. We interpret the solutions of these problems as so-
lutions of classical boundary value problems which often occur in the
theory of Elasticity.
Before we proceed with the examples, we summarize briefly the
results (without proofs) on Sobolev spaces which will prove to be very
useful in our discussion.
smartworlD.asia
Henceforth Ω ⊂ Rn will denote an open set (more often Ω will
be a bounded open set with a specific type of boundary which will be
described presently). A multi-index α will denote an n-tuple (α1 , . . . , αn )
of non-negative integers, and we denote
(2.1) |α| = α1 + · · · + αn ,
and call it the length of the multi-index. If v is a real-valued function on

Ω for which all derivatives upto order m exist, for a multi-index α with
|α| ≤ m we define
∂|α|v
(2.2) ∂αv = .
∂αx11 . . . ∂αxnn
The space of test functions on Ω is given by

(2.3) D(Ω) = v ∈ C ∞ (Ω); supp(v) is a compact subset of Ω .
10 2. Examples
where
(2.4) supp(v) = {x ∈ Ω; v(x) , 0}.
10 Definition 2.1. Let m ≥ 0 be an integer. Then the Sobolev space H m (Ω)

is given by
n o
(2.5) H m (Ω) = v ∈ L2 (Ω); ∂α v ∈ L2 (Ω) for all |α| ≤ m ,
where all derivatives are understood in the sense of distributions.
On H m (Ω) one can define a norm by means of the formula

  12
 X Z 
(2.6) ||v||m,Ω =  |∂α v|2 dx , v ∈ H m (Ω).
|α|≤m Ω
It is easy to check that || · ||m,Ω defines a norm on H m (Ω), which

makes it a Hilbert space. One can also define a semi-norm by
  12
 X Z 
(2.7) |v|m,Ω =  |∂α v|2 dx , v ∈ H m (Ω).
|α|=m Ω
smartworlD.asia
Note that since for all m ≥ 0, D(Ω) ⊂ H m (Ω), we may define,
(2.8) H0m (Ω) = D(Ω),
the closure being taken with respect to the topology of H m (Ω). Since
H0m (Ω) is a closed subspace of H m (Ω), it is also a Hilbert space under
the restriction of the norm || · ||m,Ω . We also have a stronger result:
Theorem 2.1. Assume that Ω is a bounded open set. Then over H0m (Ω)
the semi-norm | · |m,Ω is a norm equivalent to the norm || · ||m,Ω .
This result is a consequence of the following:
Theorem 2.2 (POINCARÉ-FRIEDRICHS’ INEQUALITY). If Ω is a
bounded open set, there exists a constant C = C(Ω) such that, for all
v ∈ H01 (Ω),
(2.9) |v|0,Ω ≤ C|v|1,Ω .
2. Examples 11
11 Henceforth, unless specified to the contrary, the following will be

our standing assumptions: Ω is a bounded open subset of Rn . If Γ is the
boundary of Ω, then Γ is Lipschitz continuous in the sense of Nečas [20].
(Essentially, Γ can be covered by a finite number of local coordinate
systems, such that in each, the corresponding portion of Γ is described
by a Lipschitz continuous function).
If L2 (Γ) is defined in the usual fashion using the Lipschitz continuity
of Γ, one has the following result:
Theorem 2.3. There exists a constant C = C(Ω) such that, for all v ∈
C ∞ (Ω),
(2.10) ||v||L2 (Ω) ≤ C||v||1,Ω .
By virtue of Theorem 2.3 we get that if v ∈ C ∞ (Ω), then its restric-

tion to Γ is an element of L2 (Γ). Thus we have a map from the space
C ∞ (Ω) equipped with the norm || · ||1,Ω into the space L2 (Γ) which is
continuous. We also have:
Theorem 2.4. The space C ∞ (Ω) is dense in H 1 (Ω), for domains with
Lipschitz continuous boundaries.
smartworlD.asia
Consequently, the above map may be extended to a continuous map
H 1 (Ω)→ L2 (Γ) which we denote by trΓ . It is called the trace operator.
An important result on the trace is the characterization:
n o
(2.11) H01 (Ω) = v ∈ H 1 (Ω); trΓ v = 0 .
When no confusion is likely to occur we will merely write v instead

of trΓ v. In fact if v is a “smooth” function then trΓ v is the restriction of
v to Γ.
Retaining our assumption on Ω and Γ, the unit outer normal → −ν is 12
→
−
defined a.e. on Γ. Let ν = (ν1 , . . . , νn ). If v is smooth then we may
∂v
define the outer normal derivative by
∂ν
∂v X ∂v
n
(2.12) = νi .
∂ν i=1 ∂xi
12 2. Examples
∂v
We extend this definition to v ∈ H 2 (Ω). If v ∈ H 2 (Ω), then ∈
∂xi
∂v
H 1 (Ω) and hence trΓ ∈ L2 (Γ). We now define
∂xi
∂v X
n
∂v
(2.13) = νi trΓ .
∂ν i=1 ∂xi
However when there is no confusion we write it in the form of

(2.12). Then one has the following characterization:
( )
2 2 ∂v
(2.14) H0 (Ω) = v ∈ H (Ω); v = = 0 on Γ .
∂ν
Theorem 2.5 (GREEN’S FORMULA IN SOBOLEV SPACES). Let u,

v ∈ H 1 (Ω). Then we have
Z Z Z
∂v ∂u
(2.15) u dx = − v dx + u vνi dγ,
Ω ∂xi Ω ∂xi Γ
for all 1 ≤ i ≤ n.
smartworlD.asia
∂u
If we assume u ∈ H 2 (Ω), we may replace u in (2.15) by ; sum-
∂xi
ming over all 1 ≤ i ≤ n, we get for u ∈ H 2 (Ω), v ∈ H 1 (Ω),
Z X n Z Z
∂u ∂v ∂u
(2.16) dx = − ∆u v dx + v dγ,
Ω i=1 ∂xi ∂xi Ω Ω ∂ν
Pn ∂2
where ∆ = i=1 ∂x2 is the Laplacian.
i
If both u and v are in H 2 (Ω), we may interchange the roles of u and
v in (2.16). Subtracting one formula from the other, we get
Z Z !
∂v ∂u
(2.17) (u∆v − ∆uv)dx = u − v dγ,
Ω Γ ∂ν ∂ν
13 for u, v ∈ H 2 (Ω).
2. Examples 13
Finally replacing u by ∆u in (2.17) we get, for u ∈ H 4 (Ω), v ∈

H 2 (Ω),
Z Z Z Z
2 ∂v ∂(∆u)
(2.18) ∆u∆v dx = ∆ u v dx + ∆u dγ − vdγ.
Ω Ω Γ ∂ν Γ ∂ν
The formulae (2.15) through (2.18) are all known as Green’s formu-
lae in Sobolev spaces.
We derive two results from these formulae. These results will be
useful later.
Lemma 2.1. For all v ∈ H02 (Ω),
(2.19) |∆v|0,Ω = |v|2,Ω .
Consequently over H02 (Ω), the mapping v 7→ |∆v|0,Ω is a norm equiv-
alent to the norm || · ||2,Ω .
Proof. Since D(Ω) is dense in H02 (Ω), it suffices to prove (2.19) for
v ∈ (Ω). Let v ∈ D(Ω). Then
Xn  2 2 X ∂2 v ∂2 v
2  ∂ v 
|∆v| =  2  + 2 .
i=1
∂xi 1≤i< j≤n
∂x2i ∂x2j
smartworlD.asia
By Green’s formula (2.15),
Z 2 2 Z Z !2
∂ v∂ v ∂v ∂3 v ∂2 v
(2.20) 2 2
dx = − 2
dx = dx
Ω ∂xi ∂x j Ω ∂xi ∂xi ∂x j Ω ∂xi ∂x j
for, the integrals over Γ vanish for v ∈ D(Ω). (cf. (2.11)). Now (2.19)
follows directly from (2.20). This proves the lemma.
Lemma 2.2. Let Ω ⊂ R2 . Then for u ∈ H 3 (Ω), v ∈ H 2 (Ω), 14

Z
∂2 u ∂2 v ∂2 u ∂2 v ∂2 u ∂2 v
2 − 2 2 − 2 2 dx
Ω ∂x1 ∂x2 ∂x1 ∂x2 ∂x1 ∂x2 ∂x2 ∂x1
(2.21) Z !
2
∂ u ∂v 2
∂ u ∂v
= − 2 + dγ,
Γ ∂τ ∂ν ∂τ∂ν ∂τ
∂
where denotes the tangential derivative.
∂τ
14 2. Examples
Proof. Let →−ν = (ν , ν ), →

−
1 2 τ = (τ1 , τ2 ) be the unit vectors along the outer
normal and the tangent respectively. Without loss in generality we may
assume τ1 = −ν2 , τ2 = ν1 . Also note that ν21 + ν22 = 1. The second
derivatives occurring in the right hand side are defined by
 2


 ∂ u ∂2 u 2 ∂2 u ∂2 u 2


 = τ + 2 τ1 τ2 + τ

 ∂τ2 ∂x21 1 ∂x1 ∂x2 ∂x22 2
(2.22) 



 ∂2 u ∂2 u ∂2 u ∂2 u

 = ν τ +
 ∂τ∂ν ∂x2 1 1 ∂x1 ∂x2 1 2 (ν τ + ν τ
2 1 ) + ν2 τ2 .
1 ∂x22
Using all these relations we get

 
∂2 u ∂v ∂2 u ∂v  ∂2 u ∂v ∂2 u ∂v 
− 2 + = −  ν1
∂τ ∂ν ∂τ∂ν ∂τ ∂x1 ∂x2 ∂x2 ∂x22 ∂x1
 2 
(2.23)  ∂ u ∂v ∂2 u ∂v 
+  −  ν2
∂x1 ∂x2 ∂x1 ∂x21 ∂x2
→
− −
= X ·→ ν,
→
−
where, X = (X1 , X2 ) and
smartworlD.asia



 ∂2 u ∂v ∂2 u ∂v


 X 1 = −

 ∂x1 ∂x2 ∂x2 ∂x22 ∂x1
(2.24) 



 ∂2 u ∂v ∂2 u ∂v


 X 2 = −
∂x1 ∂x2 ∂x1 ∂x21 ∂x2
Also note that,
∂X1 ∂X2 ∂2 u ∂2 v ∂2 u ∂2 v ∂2 u ∂2 v
(2.25) div X = + =2 − 2 2 − 2 2.
∂x1 ∂x2 ∂x1 ∂x2 ∂x1 ∂x2 ∂x1 ∂x2 ∂x2 ∂x1
15
Now by Green’s formula (2.15) applied to functions vi ∈ H 1 (Ω) and
to the constant function 1,
Z Z
∂vi
dx = vi νi dγ.
Ω ∂xi Γ
2. Examples 15
Hence summing over all i, if →

−v = (v , . . . , v ),
1 n
Z Z
(2.26) div→ −v dx = → −v · →−ν dγ.
Γ Γ
(This is known as the Gauss’ Divergence Theorem and also as the Os-
trogradsky’s formula). From (2.23), (2.25) and (2.26) the result fol-
lows.
With this background, we proceed to examples of the abstract prob-

lem of Sec. 1.
Example 2.1. Let K = V = H01 (Ω). Let a ∈ L∞ (Ω) such that a ≥ 0 a.e.
in Ω. Let f ∈ L2 (Ω). Define the bilinear form a(·, ·) and the functional
f (·), by
 !

 R Pn ∂u ∂v


a(u, v) = Ω i=1 + auv dx,
(2.27) 
 R ∂xi ∂xi


 f (v) =
Ω
f v dx.
The continuity of a(·, ·) and f (·) follows from the Cauchy-Schwarz

inequality. For instance
(2.28)
smartworlD.asia
| f (v)| ≤ | f |0,Ω |v|0,Ω ≤ | f |0,Ω ||v||1,Ω .
We now show that a(·, ·) is V-elliptic.

Z X n !2 

 ∂v 2
a(v, v) =  + av  dx
Ω i=1 ∂xi
Z X n !2
∂v
≥ dx (since a ≥ 0 a.e. in Ω)
Ω i=1 ∂xi
= |v|21,Ω .
16
Since |·|1,Ω is equivalent to ||·||1,Ω over V, this proves the V-ellipticity.
Hence by our results in Sec. 1 there exists a unique function u ∈ V such
that a(u, v) = f (v) for all v ∈ V.
16 2. Examples
Interpretation of this problem: Using the above equation satisfied by

u, we get
Z X n 
 Z
 ∂u ∂v
(2.29)  + auv dx = f v dx, for all v ∈ H01 (Ω).
Ω i=1 ∂xi ∂xi Ω
From the inclusion D(Ω) ⊂ H01 (Ω), we get that u satisfies the equa-
tion −∆u + au = f in the sense of distributions.
If we assume that u is “sufficiently smooth”, for example u ∈ H 2 (Ω),
then we may apply Green’s formula (2.16), which gives
Z Z Z Z
∂u
(2.30) auv dx − ∆uv dx + v dγ = f v dx.
Ω Ω Γ ∂ν Ω
Since v ∈ H01 (Ω), trΓ v = 0. Hence the integral over Γ vanishes.

Thus we get
Z
(2.31) (−∆u + au − f )v dx = 0 for all v ∈ H01 (Ω).
Ω
Varying v over H01 (Ω), we get that u satisfies the equation −∆u+au =
smartworlD.asia
f in Ω. Further since u ∈ H01 (Ω), we get the boundary condition u = 0
on Γ. Thus we may interpret u as the solution of the “classical” boundary
value problem:



−∆u + au = f in Ω,
(2.32) 

u = 0 on Γ.
17
This is known as the homogeneous Dirichlet problem for the opera-
tor −∆u + au.
A particular case of this equation arises in the theory of Elasticity,
for which Ω ⊂ R2 and a = 0. Thus −∆u = f in Ω and u = 0 on Γ. This
corresponds to the membrane problem:
Consider an elastic membrane stretched over Ω and kept fixed along
Γ. Let Fdx be the density of force acting on an element dx of Ω. Let
u(x) be the vertical displacement of the point x ∈ Ω ⊂ R2 , measured
2. Examples 17
in the x3 -direction from the (x1 , x2 )-plane. If t is the ‘tension’ of the

membrane, then u is the solution of the problem



−∆u = f in Ω
(2.33) 

u = 0 on Γ
where f = F/t.
smartworlD.asia Figure 2.1:
Remark 2.1. To solve the problem (2.32) by the classical approach, one
needs hard analysis involving Schauder’s estimates. By the above pro-
cedure viz. the variational method, we have got through with it more
easily.
The above problem is a typical example of a second-order problem. 18
Exercise 2.1. The obstacle problem. Let Ω, Γ be as in example (2.1).

Let X be an “obstacle” in this region. Let X ≤ 0 on Γ. Let F dx be the
density of the force acting on a membrane stretched over Ω, fixed along
Γ. The displacement u(x) at x in the vertical direction is the solution of
the following problem:
n o
If V = H01 (Ω), K = v ∈ H01 (Ω); v ≥ X a.e. in Ω ,
18 2. Examples
Z X n Z
∂u ∂v
let a(u, v) = dx, f (v) = f v dx, f ∈ L2 (Ω),
Ω i=1 ∂xi ∂xi Ω
X ∈ H 2 (Ω), X ≤ 0 on Γ.
Show that K is a closed convex set and hence that this problem ad-
mits of a unique solution. Assuming the regularity result u ∈ H 2 (Ω) ∩
H01 (Ω), show that this problem solves the classical problem,



 u ≥ X in Ω,




 −∆u = f when u > X, ( f = F/t)



u = 0 · Γ
(We will discuss the Obstacle Problem in Sec. 9).
smartworlD.asia
Figure 2.2:
19 Exercise 2.2. Let V = H 1 (Ω). Define a(·, ·), f (·) as in example 2.1.
Assume further that there exists a constant a0 such that a ≥ a0 > 0 in Ω.
If u0 is a given function in H 1 (Ω), define
n o
K = v ∈ H 1 (Ω); v − u0 ∈ H01 (Ω)
n o
= v ∈ H 1 (Ω); trΓ v = trΓ u0 .
2. Examples 19
Check that K is a closed convex subset. Interpret the solution to be

that of the Non-homogeneous Dirichlet problem,



−∆u + au = f in Ω.


u = u0 on Γ.
Example 2.2. Let K = V = H 1 (Ω). Let a ∈ L∞ (Ω) such that a ≥ a0 > 0,

f ∈ L2 (Ω). Define a(·, ·) and f (·) as in example (2.1). The continuity
of a(·, ·) and f (·) follow as usual. For the V-ellipticity, we can no longer
prove it with the semi-norm | · |1,Ω as we did earlier. It is here we use the
additional assumption on a, since
Z X n !2 

 ∂v 2
a(v, v) =  + av  dx
Ω i=1 ∂xi
≥ min(1, a0 )||v||21,Ω .
Thus we have a unique solution u to the abstract problem satisfying

a(u, v) = f (v). If we assume again that u is “sufficiently smooth” to
apply the Green’s formula (2.16), we get
Z Z Z
∂u
smartworlD.asia
(2.34) (−∆u + au)v dx + v dγ = f v dx.
Ω Γ ∂ν Γ
If v ∈ D(Ω), then the integral over Γ will vanish. Thus u satisfies

the equation −∆u + au = f as in Example 2.11 . However we now get a 20
different boundary condition. In example (2.1) the boundary condition
was built in with the assumption u ∈ V = H01 (Ω). Now from (2.34), we
may write:
Z Z
∂u
(2.35) (−∆u + au − f )v dx = − v dγ
Ω Γ ∂ν
for all v ∈ H 1 (Ω).

But the left hand side of (2.35) is zero since u Rsatisfies the differen-
tial equation as above so that for all v ∈ H 1 (Ω), Γ ∂u ∂ν v dγ = 0. Thus
1
As in Example 2.1, the equation −∆u + au = f is always satisfied in the sense of
distributions since D(Ω) ⊂ H 1 (Ω).
20 2. Examples
∂u
= 0 on Γ, and we may interpret this problem as the classical prob-
∂ν
lem:



−∆u + au = f in Ω
(2.36) 

 ∂u = 0 on Γ.
∂ν
This is a homogeneous Neumann problem.
Exercise 2.3. With K, V, a(·, ·) as in example 2.2., define

Z Z 


 f ∈ L2 (Ω),
f (v) = f v dx + gv dγ,  
Ω Γ g ∈ L2 (Ω).
Show that the abstract problem leads to a solution of the non-homo-

geneous Neumann problem



−∆u + au = f in Ω



 ∂u
 = g on Γ.
∂ν
21 Remark 2.2. In these examples one may use the more general bilinear
from defined by
smartworlD.asia
Z  X n


 ∂u ∂v
(2.37) a(u, v) =  ai j + auv dx,
Ω i, j=1
∂x j ∂xi
where the functions ai j ∈ L∞ (Ω) satisfy the condition that for some
ν > 0,
X
n X
n
(2.38) ai j ξi ξ j ≥ ν ξi2
i, j=1 i=1
for all ξ ∈ Rn and a.e. in Ω. This is the classical ellipticity condition for
second order partial differential operators. One should check (exercise!)
in this case that the abstract problem leads to a solution of the boundary
value problem
X n !
∂ ∂u
(2.39) − ai j + au = f in Ω
i, j=1
∂xi ∂x j
2. Examples 21
with the boundary condition




u = 0 on Γ if K = V = H01 (Ω),

 n
(2.40) 
 P ∂u


 ai j ∂x j
νi = 0 on Γ if K = V = H 1 (Ω).
i, j=1
The latter boundary operator in (2.40) is called the conormal deriva-

tive associated with the partial differential operator,
X n !
∂ ∂
− ai j .
i, j=1
∂xi ∂x j
Notice that the term (+ au) contributes nothing.
Example 2.3. System of Elasticity. Let Ω ⊂ R3 , with Lipschitz con-

tinuous boundary Γ. Further assume that Γ can be partitioned into two
portions Γ0 and Γ1 such that the dγ-measure of Γ0 is > 0. Let

K=V= → −v = →
−v = (v , v , v ); v ∈ H 1 (Ω), 1 ≤ i ≤ 3 and → −
0 on Γ0 .
1 2 3 i
Define 22
 !
smartworlD.asia


 →
− 1 ∂vi ∂v j

ǫi j ( v ) = 2 ∂x + ∂x ,
(2.41)  j i



σi j (→−v ) = λ P3 ǫ (→ −v ) δ + 2µǫ (→ −
k=1 kk ij i j v ),
for 1 ≤ i, j ≤ 3. The latter relation is usually known as Hooke’s law.

The constants λ(≥ 0) and µ(> 0) are known as Lame’s coefficients. We
define the bilinear form a(·, ·) by,
Z X
3
a(→
−u ,→
−v ) = σi j (→
−u ǫ (→ −
i j v )dx
Ω i, j=1
(2.42) Z X
3
= (λ div →
−v + 2µ ǫi j (→
−u )ǫ (→ −
i j v )dx.
Ω i, j=1
→
−
Let f = ( f1 , f2 , f3 ), fi ∈ L2 (Ω), and →
−g = (g , g , g ), g ∈ L2 (Γ), be
1 2 3 i
given.
22 2. Examples
Define the linear functional f (·), by,

Z Z
→
− →
− → − →
−g · →
−v dγ.
(2.43) f( v ) = f · v dx +
Ω Γ1
The continuity of a(·, ·) and f (·) follow from the Cauchy-Schwarz

inequality. For the V-ellipticity of a(·, ·) one uses the inequality
Z X
3
a(→
−v ,→
−v ) ≥ 2µ (ǫi j (→
−v ))2 dx
Ω i, j=1
and the fact that the square root of the integral appearing in the right
hand side of the above inequality is a norm over the space V, equivalent
−v = (v , v , v ) 7→ (P3 ||v ||2 ) 12 . This is a nontrivial fact
to the norm → 1 2 3 i=1 i 1,Ω
which uses essentially the fact that Γ0 has measure > 0 and an inequality
known as Körn’s inequality. We omit the proof here.
Again the problem a(→ −u ,→
−v ) = f (→
−v ) admits a unique solution. As-
23 suming sufficient smoothness, we may apply Green’s formula:
Z X
3 Z X
3 !
1 ∂vi ∂v j
σi j (→
−u )ǫ (→ −
i j v )dx =
→
−
σi j ( u ) + dx
Ω i, j=1 2 Ω i, j=1 ∂x j ∂xi
smartworlD.asia
Z X
3
∂vi
= σi j (→
−u ) dx
Ω i, j=1 ∂x j
(since ǫi j (v) is symmetric in i and j)

Z X 3 Z X 3
∂ →
−
=− (σi j ( u ))vi dx + σi j vi ν j dγ.
Ω i, j=1 ∂x j Γ1 i, j=1
Thus, the abstract problem leads to a solution of

 3


 P ∂ →
−


− ∂x j (σi j ( u )) = fi (1 ≤ i ≤ 3) in Ω


 j=1

 →−u = → −
(2.44) 
 0 on Γ0 and





 P
3
→−

 σi j ( u )ν j = gi on Γ1 (1 ≤ i ≤ 3).

j=1
2. Examples 23
Note also that

 3 
X3
∂ X3
∂  X 
− (σi j (→
−u )) = − λ

→
− →
−
ǫkk ( u )δi j + 2µǫi j ( u )
j=1
∂x j j=1
∂x j k=1
 3  !!
X
3
∂  X ∂uk  X3
∂ ∂ui ∂u j
=− λ  δi j − 2µ +
j=1
∂x j k=1
∂xk j=1
∂x j ∂x j ∂xi
= −(λ + µ)(grad div→

−v ) − µ∆u .
i i
Thus the first equation of (2.44) is equivalent to
(2.45) −µ∆→ −u = →
−u − (λ + µ) grad div→ −
f in Ω.
The equations (2.44) constitute the system of linear Elasticity.
smartworlD.asia
Figure 2.3:
24
If we have an elastic three-dimensional body fixed along Γ0 , acted
on by an exterior force of density f dx and force of density g dγ along
Γ1 and if σi j is the stress tensor, the displacement u satisfies (2.44); cf.
Fig. 2.3.
The relation a(→ −u ,→
−v ) = f (→
−v ), viz.,
Z X
3 Z Z
→
− →
(2.46) σi j (→
−u )ǫ (→ −
i j v )dx = f · −v dx + →
−g · →
−v dγ
Ω i, j=1 Ω Γ1
24 2. Examples
for all →−v ∈ V is known as the principle of virtual work. The tensor ǫ
ij
is the strain tensor and the tensor σi j the stress tensor. The expression
1 → − → − →
−
2 a( v , v ) is the strain energy, and the functional f ( v ) is the potential
energy of exterior forces. This example is of fundamental importance in
that the finite element method has been essentially developed for solving
this particular problem or some of its special cases (membranes, plates,
shells, etc.,) and generalizations (nonlinear elasticity, etc. . . ).
smartworlD.asia
2. Examples 25
Remark 2.3. The above problems are all examples of linear problems2 : 25
The map from the right-hand side of the equation and of the boundary
conditions to the solution u is linear. The non-linearity may occur in
three ways:
(i) When K is not a subspace of V. (e.g. Exercises 2.1 and 2.4);
(ii) If in Example 2.3 we have, instead of the first equality in (2.41):
 ! X 
1  ∂vi ∂v j ∂vk ∂vk 
3
→
−
ǫi j ( v ) =  + +  .
2 ∂x j ∂xi k=1
∂xi ∂x j
This is the case for instance when one derives the so-called Von
Karmann’s equations of a clamped plate;
(iii) We may replace Hooke’s law (the second relations in (2.41)) by
non-linear equations connecting ǫi j and σi j , which are known as
non-linear constitutive equations. (e.g. Hencky’s law).
Exercise 2.4. Let V = H 1 (Ω), and a(·, ·) and f (·) be as in example 2.2,
and let n o
1
K = v ∈ H (Ω); v ≥ 0 a.e. on Γ .
smartworlD.asia
Show that K is a closed convex cone with vertex 0. Using the results
of Sec. 1 show that the interpretation is
−∆u + au = f in Ω,
∂u ∂u
u ≥ 0, ≥ 0, u = 0 on Γ.
∂ν ∂ν
(This is called the SIGNORINI problem).
We now examine fourth-order problems.
Example 2.4. Let K = V = H02 (Ω). Define

 R


a(u, v) = ∆u∆v dx,
(2.47) 
 R Ω
 f (v) = f v dx, f ∈ L2 (Ω),
Ω
2
Except in Exercise 2.1.
26 2. Examples
26 for all u, v ∈ V. The continuity follows as usual. For the V-ellipticity

of a(·, ·) we have,
Z
(2.48) a(v, v) = (∆v)2 dx = |∆v|20,Ω = |v|22,Ω .
Ω
(by Lemma 2.1. Since | · |2,Ω and || · ||2,Ω are equivalent on H02 (Ω), the
V-ellipticity follows from (2.48).
Hence there exists a unique function u ∈ H02 (Ω) such that
Z Z
∆u∆v dx = f v dx for all v ∈ H02 (Ω).
Ω Ω
Assuming u to be sufficiently smooth (say, u ∈ H 4 (Ω)), then by

Green’s formula (2.18),
Z Z Z
2 ∂(∆u) ∂v
(2.50) (∆ u − f )v dx = v dγ − ∆u dγ,
Ω Γ ∂ν Γ ∂ν
for all v ∈ H02 (Ω). Hence by varying v over H02 (Ω), we get that u satisfies
∆2 u = f in Ω. Since u ∈ H02 (Ω), the boundary conditions are given by
(2.14). Thus we interpret this problem as the classical problem


 2
∆ u = f in Ω,

smartworlD.asia
(2.51) 
 ∂u

u = = 0 on Γ.
∂ν
This is the homogeneous Dirichlet problem for the operator ∆2 .
When n = 2, this is an important problem in Hydrodynamics. Here
u is known as the stream function and −∆u is the vorticity.
27 A slight modification of a(·, ·) leads to an important problem in Elas-
ticity. Again let n = 2. Let f ∈ L2 (Ω) and if K = V = H02 (Ω), define
(2.52)
 R


 f (v) = Ω f v dx,

 " !#
 R ∂2 u ∂2 v ∂2 u ∂2 v ∂2 u ∂2 v


a(u, v) = ∆u ∆v + (1 − σ) 2 − − dx.
 Ω ∂x1 ∂x2 ∂x1 ∂x2 ∂x21 ∂x22 ∂x22 ∂x21
The integrand occurring in the definition of a(u, v) may also be writ-

ten as
 2 2 2 2

 ∂ u ∂ v ∂ u ∂ v 2∂2 u ∂2 v 
(2.53) σ∆u∆v + (1 − σ)  2 2 + 2 2 +  .
∂x1 ∂x1 ∂x2 ∂x2 ∂x1 ∂x2 ∂x1 ∂x2
2. Examples 27
1
Usually, from physical considerations, 0 < σ < .
2
Note that
(2.54) a(v, v) = σ|∆v|20,Ω + (1 − σ)|v|22,Ω
by (2.53) and this leads to the V-ellipticity of a(·, ·). By virtue of (2.21)
in Lemma 2.2, we get that the relations a(u, v) = f (v) read as
Z Z
2
(2.55) ∆ uv dx = f v dx,
Ω Ω
assuming sufficient smoothness of u. Thus again we get the same equa-

tion as in (2.51). Notice that the additional term in the definition of a(·, ·)
has contributed nothing towards the differential equation.
This latter problem is known as the clamped plate problem:
Consider a plate of “small” thickness e lying on the x1 x2 -plane. Let
E be its Young’s modulus and σ its Poisson coefficient. Let there be a
load F acting on the plate. The displacement u is the solution of (2.51),
where f is given by (cf. Fig. 2.4):
Ee3 f
(2.56) F= .
12(1 − σ2 )
smartworlD.asia
28
Figure 2.4:
We will return to this problem in Sections 10 and 11.
28 2. Examples
Exercise 2.5. Let K = V = {v ∈ H 2 (Ω); v = 0 on Γ} = H 2 (Ω)∩ H01 (Ω),

and define a(·, ·) and f (·) as in the case of the clamped plate. Assuming
the V-ellipticity of a(·, ·) show that the solution of the abstract problem
satisfies ∆2 u = f in Ω and u = 0 on Γ. What is the other boundary
condition? This is known as the problem of the simply supported plate.
Exercise 2.6. Let K = V = H 2 (Ω) ∩ H01 (Ω), and

Z
a(u, v) = ∆u∆v dx;
ZΩ Z
∂v
f (v) = f v dx − λ dγ, where f ∈ L2 (Ω), λ ∈ L2 (Γ).
Ω Γ ∂ν
Show that we may apply the result of Sec. 1 and give an interpreta-
tion of this problem.
REFERENCES. For details on Sobolev spaces, see Nečas [20] and

Lions and Magenes [17]. For the theory of Elasticity, one may refer to
Duvaut and Lions [10] and Landau and Lipschitz [14].
smartworlD.asia
Chapter 3
The Finite Element Method

in its Simplest Form
MAINTAINING OUR ASSUMPTIONS as in the Lax-Milgram Lemma 29

(Theorem 1.2 we concentrate our attention on the following problem
(P):
(P): To find u ∈ V such that a(u, v) = f (v) for all v ∈ V.
smartworlD.asia
Let Vh be a finite-dimensional subspace of V. Then we may state
the following problem:
(Ph ): To find uh ∈ Vh such that a(uh , vh ) = f (vh ) for all vh ∈ Vh .
Vh , being a finite-dimensional subspace, is a Hilbert space for the
norm of V. Hence by Theorem 1.2, uh exists and is unique. We try
to approximate the solution u of (P) by means of solutions uh of the
problem (Ph ) for various subspaces Vh . This is known as the internal
approximation method.
As a first step in this direction, we prove a most fundamental result:
Theorem 3.1. There exists a constant C, which is independent of Vh ,
such that
(3.1) ||u − uh || ≤ C inf ||u − vh ||.

vh ∈Vh
29
30 3. The Finite Element Method in its Simplest Form
Proof. If wh ∈ Vh , then
a(u, wh ) = f (wh ) = a(uh , wh ).
Thus for all wh ∈ Vh
(3.2) a(u − uh , wh ) = 0.
Using this and the V-ellipticity of a(·, ·), we get, for all vh ∈ Vh ,
α||u − uh ||2 ≤ a(u − uh , u − uh )

= a(u − uh , u − vh ) + a(u − uh , vh − uh )
= a(u − uh , u − vh ) by (3.2)
≤ M||u − uh || ||u − vh ||.
30
Hence, ||u − uh || ≤ M/α ||u − vh || for all vh ∈ Vh . Setting C = M/α
and taking infimum on the right-hand side the result follows.
The above result estimates the ‘error’ in the solution of (P) when
instead we solve (Ph ). To get an upper bound for the error, we only need
smartworlD.asia
to compute inf ||u − vh || which is the distance of u from the subspace
vh ∈Vh
Vh . This is a problem in approximation theory.
Remark 3.1. If a(·, ·) is also symmetric then we observe the following:
(i) J(uh ) = inf J(vh ) by Corollary 1.1 (b)

vh ∈Vh
(ii) We saw that a(u − uh , wh ) = 0 for all wh ∈ Vh . Since a(·, ·) is now

an inner product, we get that uh is the projection of u to the closed
subspace Vh in the sense of this inner-product. Therefore,
√ p p √
α||u − uh || ≤ a(u − uh , u − uh ) ≤ a(u − vh , u − vh ) ≤ M||u − vh ||
for all vh ∈ √Vh . Hence the constant C in theorem 3.1 can be taken
to be here M/α ≤ M/α, since the continuity and V-ellipticity
imply jointly that M ≥ α.
3. The Finite Element Method in its Simplest Form 31
We may now describe the finite element method (f.e.m.) in its sim-
plest terms. The method consists in making special choices for the sub-
spaces Vh such that the solutions uh of the problems (Ph ) converge to
u.
We will outline the procedure for obtaining the spaces Vh by consid- 31
ering, for example, a second-order problem.
Let V = H01 (Ω) or H 1 (Ω). Let us assume Ω to be a polygonal domain
in Rn . That is, Ω is a polygon in Rn . We then have the following step-
by-step procedure:
(i) We first establish a finite triangulation kh of the domain Ω such
S
that Ω = K. The sets K are called finite elements. If n = 2,
K∈kh
they will be, in general, triangles. They will be tetrahedral in n =
3 and ‘n-simplices’ in any Rn . These have the further property that
any side of a finite element K is either a portion of the boundary
or the side of an adjacent finite element. (See Fig. 3.1).
smartworlD.asia
Figure 3.1:
(ii) The space Vh is such that for each vh ∈ Vh , its restriction vh |K to

each K belongs to some finite-dimensional space Pk of real valued
functions over K which are preassigned. In practice we choose PK
to be a space of polynomials.
(iii) We then need inclusions such as Vh ⊂ H01 (Ω) or H 1 (Ω). We es-

tablish a simple criterion to realise this.
32 Theorem 3.2. If for every K = kh , PK ⊂ H 1 (K), and Vh ⊂ C 0 (Ω), then

Vh ⊂ H 1 (Ω). If in addition v = 0 on Γ for all v ∈ Vh , then Vh ⊂ H01 (Ω).
Proof. Let v ∈ Vh . Since v|K ∈ L2 (K) for every K ∈ kh , it follows that
v ∈ L2 (Ω). Hence to complete the proof it only remains to show that for
1 ≤ i ≤ n, there exist vi ∈ L2 (Ω) such that for each ϕ = D(Ω), we have,
Z Z
∂ϕ
(3.3) ϕvi dx = − v dx. (1 ≤ i ≤ n).
Ω Ω ∂xi
∂v
Then it will follow that = vi and hence v ∈ H 1 (Ω).
∂xi
∂(v|K)
However, v|K ∈ PK ⊂ H 1 (K) implies that ∈ L2 (K) for
∂xi
1 ≤ i ≤ n. Let ϕ ∈ D(Ω). Since the boundary ∂K of any K of the tri-
angulation is Lipschitz continuous, we apply the Green’s formula (2.15)
to get
Z Z Z
∂(v|K) ∂ϕ
(3.4) ϕdx = − (v|K) dx + (v|K)ϕνi,K dγK ,
K ∂xi K ∂xi ∂K
where dγK is the measure on ∂K and → −ν = (ν , . . . , ν ) is the outer

K 1,K n,K
smartworlD.asia
normal on ∂K. Summing over all the finite elements K, we get
Z X Z ∂(v|K)
ϕvi dx = ϕ dx
Ω K∈kh K
∂xi
(3.5) Z XZ
∂ϕ
=− v dx + ϕ(v|K)νi,K dγK ,
Ω ∂xi K∈k ∂K
h
∂(v|K)
where vi is the function whose restriction to each K is .
∂xi
The summation on the right-hand side of the above equation is zero
for the following reasons:
On the boundary Γ, since ϕ ∈ D(Ω), the integral corresponding to
∂K ∩ Γ is zero. So the problem, if any, is only on the other portions
of the boundary of each K. However, these always occur as common
33 boundaries of adjacent finite elements. The value of v|K on the common
boundary of two adjacent finite elements is the same (Vh ⊂ C 0 (Ω)). But
3. The Finite Element Method in its Simplest Form 33
the outer normals are equal and opposite from orientation considera-
tions. (See Fig. 3.2).
Figure 3.2:
Hence the contributions from each K along the common boundaries

cancel one another. Thus the summation yields only zero. Hence vi
satisfies (3.3) for 1 ≤ i ≤ n, and clearly vi ∈ L2 (Ω). The last part of the
theorem follows the characterization (2.11).
Exercise 3.1. If for all K ∈ kh , PK ⊂ H 2 (K) and Vh ⊂ C 1 (Ω), then

∂v
show that Vh ⊂ H 2 (Ω). Also if v = = 0 on Γ, for all v ∈ Vh , then
∂ν
smartworlD.asia
Vh ⊂ H02 (Ω).
We finally describe the system of linear equations associated with

the space Vh . Suppose {w j ; 1 ≤ j ≤ M} is a basis for Vh . Let uh be the
solution of (Ph ). If uh is given by
X
M
(3.6) uh = u jw j.
j=1
then we have, since a(uh , wi ) = f (wi ) for 1 ≤ i ≤ M,
X
M
(3.7) a(w j , wi )u j = f (wi ), 1 ≤ i ≤ M.
j=1
To find uh , the above system of linear equations must be solved. The

matrix for this system has for its (i, j)-coefficient the value a(w j , wi ). 34
Note that the symmetry of a(·, ·) implies the symmetry of the matrix
and the V-ellipticity says that the matrix is positive definite. In practical
computations these observations are important.
Since we have to handle the matrix of the system, it would be ideal
of course to have a diagonal matrix. We could in principle achieve this
through a Gram-Schmidt orthogonalisation procedure applied to the ba-
sis functions. However such a process is not feasible since it is highly
“numerically unstable”. So the best we may hope for is a matrix with “a
lot of” zeros in it - what is known as a sparse matrix.
For example in the problem given by



−∆u + au = f in Ω


u = 0 on Γ
the (i, j)-coefficient of the matrix is

Z X 

 n ∂w j ∂wi
(3.8) a(w j , wi ) =  + aw j wi  dx.
Ω k=1 ∂xk ∂xk
The matrix will be sparse if the supports of the basis functions are
smartworlD.asia
as “small” as possible so that their inner-products will be most often
zero. We will study subsequently methods to achieve this. This trivial
criterion extends, of course, to all types of problems.
Chapter 4
Examples of Finite Elements
WE SUMMARIZE BELOW our requirements regarding the “finite el- 35

ement subspaces” Vh
(i) Let Ω ⊂ Rn be a polygonal domain. Let kh be a triangulation of

Ω as in Sec. 3. Then Vh is a finite-dimensional vector space such
that for all v ∈ Vh , v|K ∈ PK for every finite element K, where
PK is a vector space of finite dimension. Usually, PK is a space
smartworlD.asia
of polynomials. This is of practical importance in computing the
matrix of the system. We shall see later that it is of theoretical
importance as well. Observe for the moment that if PK consists
of polynomials, then we automatically have that PK ⊂ H 1 (K) or
PK ⊂ H 2 (K).
(ii) By Theorem 3.2, Vh ⊂ C 0 (Ω) implies that Vh ⊂ H 1 (Ω) and by

Exercise 3.1 Vh ⊂ C 1 (Ω) implies that Vh ⊂ H 2 (Ω). Thus we must
choose a proper basis for the “local” spaces PK such that these
“global” inclusions hold.
(iii) There must exist at least one basis {w j } of Vh which consists of

functions with “small” support.
We bear these points in mind when constructing examples of finite

elements. Before we proceed we need a few definitions.
35
36 4. Examples of Finite Elements
Definition 4.1. An n-simplex is the convex hull in Rn of (n + 1) points

{a j }n+1 n
j=1 such that if a j = (ak j )k=1 and A is the matrix
 
a11 a12 . . . a1,n+1 
 .. .. .. 
(4.1) A =  . . . 
an1 an2 . . . an,n+1 
 
1 1 ... 1
36 then det A , 0.
The above definition generalises the notion of a triangle to n dimen-

sions. Geometrically the condition det A , 0 simply means that the
points {a j }n+1
j=1 do not lie in the same hyperplane. For det A is equal to,
by elementary column operations, the determinant of the matrix
 
(a11 − a1,n+1 ) . . . (a1,n − a1,n+1 )
 .. .. 
 . . 
 
(an1 − an,n+1 ) . . . (an,n − an,n+1 )
and that this is non-zero means that (a1 − an+1 ), . . . , (an − an+1 ) are
linearly independent vectors in Rn , which is the same as saying that
a1 , . . . , an+1 do not lie in the same hyperplane.
smartworlD.asia
Definition 4.2. Let {a j }n+1 n
j=1 be (n + 1)-points in R satisfying the condi-
tions of definition 4.1. The barycentric coordinates of any x ∈ Rn with
respect to these points are numbers {λ j }n+1j=1 such that


 P
n+1


 x= λ ja j,

 j=1
(4.2) 



 P
n+1

1 = λ j.
j=1
The barycentric coordinates exist because they are merely the com-
→−
ponents of the unique solution vector λ of the system of (n + 1) linear
equations in (n + 1) unknowns given by
 
→− !  x1 
→
− x  
Aλ = where x =  ... 
→−
1  
xn
4. Examples of Finite Elements 37
The functions λ j = λ j (x) are all affine functions of x. Also λ j (ai ) =

δi j , where δ is the Kronecker symbol. 37
Remark 4.1. Given points {a j }n+1

j=1 as in the definition (4.1), the corre-
sponding n-simplex is given by
 


 X
n+1 X
n+1 


 
K=  x = λ a
j j ; 0 ≤ λ j ≤ 1; λ j = 1
 .

 

j=1 j=1
Definition 4.3. Let k ≥ 0 be an integer. Then, Pk is the space of all

polynomials of degree ≤ k in x1 , . . . , xn .
We now proceed with examples of finite elements.
Example 4.1. The n-simplex of type (1).

P
Let K be an n-simplex. Let PK = P1 . We define a set K =
{p(ai ); 1 ≤ i ≤ n + 1} of degrees of freedom for p ∈ PK , where {ai }n+1
P i=1
are the vertices of K: The set K determines every polynomial p ∈
PK uniquely. For, note that dim PK = dim P1 = n + 1. Consider
λ1 , . . . , λn+1 ∈ P1 , the barycentric coordinate functions. These are lin-
smartworlD.asia
P
early independent since αk λk = 0 implies that its value at each vertex
is zero. Since λk (a j ) = δk j we get that α j = 0 for all j. Thus these
functions form a basis for P1 . Let us write
X
n+1
p= αi λi .
i=1
Then
X
n+1 X
n+1
p(a j ) = αi λi (a j ) = αi δi j = α j .
i=1 i=1
Thus,
X
n+1
(4.4) p= p(ai )λi .
i=1

Let K be an n-simplex with vertices {ai }n+1 i=1 . Let ai j (i < j) be the 38
mid-points of the line joining ai and a j , i.e. ai j = 12 (ai + a j ).
Figure 4.1:
P
Let PK = P2 . We define for p ∈ P2 , the set K = {p(ai ), 1 ≤ i ≤
P
n + 1; p(ai j ), 1 ≤ i < j ≤ n + 1} of degrees of freedom. Again n+2 K
determines p ∈ P2 completely. To see this note that dim PK = 2 and
there are as many functions in the set {λi (2λi − 1), 1 ≤ i ≤ n+ 1; λi λ j , 1 ≤
i ≤ j ≤ n + 1}. There are all functions in P2 . Further since
smartworlD.asia


 1
 2 if i = k or j,
λi (a j ) = δi j , λi (ak j ) = 

0 otherwise,
we see again that these are linearly independent in P2 . Let us write

X
n+1 X
p= αi λi (2λi − 1) + βi j λi λ j .
i=1 1≤i< j≤n+1
Then
X
n+1
p(ak ) = αi δik (2δik − 1) = αk .
i=1
Further,
X
n
p(akl ) = αi (2λ2i (akl ) − λi (akl ))
i=1

X
+ βi j λi (akl )λ j (akl ).
1≤i< j≤n+1
P
n
But since λi (akl ) = 0 or 12 , the first sum is zero. Further, 39
i=1


 1
 4 if (i, j) = (k, l) or (l, k),
λi (akl )λ j (akl ) = 

0 otherwise.
Hence βkl = 4p(akl ). Thus we have
X
n+1 X
(4.5) p= λi (2λi − 1)p(ai ) + 4λi λ j p(ai j ).
i=1 1≤i< j≤n+1

2ai + a j
Let K be an n-simplex with vertices {ai }n+1i=1 . Let aii j = ,
3
ai + a j + ak
i , j. Let ai jk = for i < j < k.
3
smartworlD.asia
Figure 4.2:
Set PK = P3 . Define the set of degrees of freedom

X n
= p(ai ), 1 ≤ i ≤ n + 1; p(aii j ), 1 ≤ i , j ≤ n + 1;
K
o
p(ai jk ), 1 ≤ i < j < k ≤ a + 1 .
2 1
Note that λi (aii j ) = ; λ j (aii j ) = ; λ1 (aii j ) = 0 if 1 , i, 1 , j;
3 3
1
λ1 (ai jk ) = if 1 = i, j or k and 0 otherwise, etc. Using these, one
3
checks the linear independence of the functions
n
λi (3λi − 1)(3λi − 2), 1 ≤ i ≤ n + 1; λi λ j (3λi − 1), 1 ≤ i , j ≤ n + 1;
o
λi λ j λk 1 ≤ i < j < k ≤ n + 1 .
These then form a basis for P3 , for there are as many functions in the
above collection as dim PK . Using the values of λi at the special points
described above, we get
X
n+1
λi (3λi − 1)(3λi − 2)
p= p(ai )
i=1
2
X 9
(4.6) + λi λ j (3λi − 1)p(aii j )
1≤i, j≤n+1
2
X
27λi λ j λk p(ai jk ).
1≤i< j<k≤n+1
40
P
smartworlD.asia
Thus K completely determines p ∈ P3 .
P
The points of K at which the polynomials are evaluated to get K
P
are known as the nodes of the finite element. The set K is the set of
degrees of freedom of the finite element.
Exercise 4.1. Generalize these ideas and describe the n-simplex of type
(k) for any integer k ≥ 1.
We now show how these finite elements may be used to define the
space Vh .
First of all we show the inclusion Vh ⊂ C 0 (Ω). Consider for instance
a triangulation by n-simplices of type (1). Number all the nodes of the
P
triangulation by {b j }. Let us define h = {p(b j ); b j is a node.}: This is
the set of degrees of freedom of the space Vh : A function v in the space
Vh is, by definition, determined over each K ∈ kh by the values v(b j ) for
those nodes b j which belong to K.
Let us examine the two-dimensional case, for simplicity. If K1 and

K2 are two adjacent triangles with common side K ′ (cf. Fig. 4.3), we
need to show that v|K1 = v|K2 along K ′ for any v ∈ Vh . Let t be an
abscissa along K ′ .
Figure 4.3:
41
Now v|K1 along K′ is a polynomial of degree 1 in t. So is v|K2 along
K ′ . But these two agree at the nodes b1 and b3 . Therefore, they must be
identical and hence the continuity of v follows.
smartworlD.asia
This argument can be extended to any simplex of type (k). These
simplices, by Theorem 3.2 yield the inclusion Vh ⊂ H 1 (Ω) and hence
we may use them for second order problems.
Exercise 4.2. The triangle of type (3′ ).

P
Let K be a triangle in R2 . Define K to be the values of p at the
points {ai , 1 ≤ i ≤ 3}, and the points {aii j , 1 ≤ i , j ≤ 3}. If we define
P P
P′3 = {p ∈ P3 ; 12p(a123 ) + 2 3i=1 p(ai ) − 3 i, j p(aii j ) = 0}, then show
P
that K uniquely determines p ∈ P′3 = PK . Further show that P2 ⊂ P′3 .
We now relax our terminological rules about “triangulations” and

admit rectangles (and in higher dimensions, hyper rectangles or hyper-
cubes) in triangulations. We describe below some finite elements which
are rectangles.
We need another space of polynomials.
Definition 4.4. Let k ≥ 1 be an integer. Then

 


 




 X 

 i1 in 

Qk =  p; p(x) = a x
i1 ,...in 1 . . . xn  .


 



 0≤i j ≤k 

1≤ j≤n
42
We have the inclusions Pk ⊂ Qk ⊂ Pnk .
Example 4.4. The Rectangle of type (1).

Let K be the unit square in R2 , i.e., K = [0, 1]n . Let PK = Q1 .
P
The set of degrees of freedom is given by K = {p(ai ), 1 ≤ i ≤ 4}; cf.
Fig. 4.4 in the case n = 2.
smartworlD.asia
Figure 4.4:
P
To show that K indeed determines p ∈ Q1 uniquely we adopt a
different method now. (There are essentially two methods to show that
P
K completely determines P K ; the first was used in the previous ex-
amples where we exhibited a basis for PK such that the corresponding
P
coefficients in the expansion of p in terms of this basis came from K ;
the second is illustrated now).
P
Observe first that dim PK = card K = 2n . To determine a polyno-
P
mial completely in terms of the elements of K we must solve 2n linear
equations in as many unknowns. That every polynomial is determined

this way is deduced from the existence of a solution to this system. But
for such a system the existence and uniqueness of the solution are equiv- 43
alent, and one establishes the latter. Thus we show that if p ∈ PK such
that all its degrees of freedom are zero, the p ≡ 0.
Returning to our example, consider a polynomial p ∈ Q1 such that
p(ai ) = 0 for all 1 ≤ i ≤ 4. On each side p is a polynomial of degree
1 either in x1 alone or in x2 alone. Since it vanishes at two points, the
polynomial p vanishes on the sides of the square. Now consider various
lines parallel to one of the axes. Here too p is a polynomial of degree 1
in one variable only. Since it vanishes at the points where the line meets
the side, it also vanishes on this line. Varying the line we get p ≡ 01 .
Example 4.5. The Rectangle of type (2).

Again consider the unit square (or hypercube in Rn ) to be the finite
P
element K. Set PK = Q2 , and K = {p(ai ), 1 ≤ i ≤ 9} where the ai are
as in the figure below.
smartworlD.asia
Figure 4.5:
Here again one can prove the unisolvency as above. Now let p ∈ Q2
be given such that p(ai ) = 0 for all 1 ≤ i ≤ 9; then p = 0 on the four
1
It is not necessary to restrict ourselves to a square. Any rectangle with sides parallel
to the coordinate axes would do.
sides and on the two central (dotted, in Fig. 4.5) lines. Now take lines
44 parallel to one of the axis and p vanishes on each of these. Thus p ≡ 0
P
on K and we get that K uniquely determines p ∈ Q2 .
Exercise 4.3. Describe the rectangle of type (3) and generalize to hyper-
rectangles of type (k).
Exercise 4.4. Prove that in all the preceding examples, we get Vh ⊂

C 0 (Ω).
Exercise 4.5. The Rectangle of type (2′ ).

Let K be as in example 4.5. However omit the node a9 (the centroid
P
of K). Let K = {p(ai ), 1 ≤ i ≤ 8} and show that this determines
uniquely a function in the space
 


 X
4 X
8 


PK =  p ∈ Q 2 ; 4p(a9 ) + p(a i ) − 2 p(ai ) = 0
 .
 
i=1 i=5
and that P2 ⊂ PK .
We now turn to different types of finite elements. They differ from
smartworlD.asia
the preceding ones in the choice of degrees of freedom as will be seen
presently.
Example 4.6. The Hermite Triangle of Type (3).

Let K ⊂ R2 be a triangle with vertices {a1 , a2 , a3 }. Let λi , 1 ≤ i ≤ 3,
be the barycentric coordinate functions. Then one can check that any
polynomial p ∈ P3 = PK can be expanded as
X
3
p= (−2λ3i + 3λ2i − 7λ1 λ2 λ3 )p(ai ) + 27λ1 λ2 λ3 p(a123 )
i=1
X
3 X
+ λi λ j (2λi + λ j − 1) Dp(ai )(a j − ai ).
i=1 j=1
j,i
P
Thus, K = {p(ai ), 1 ≤ i ≤ 3; Dp(ai )(a j − ai ), 1 ≤ i , j ≤ 3; p(a123 )}
is the corresponding set of degrees of freedom.
45 Note that Dp(ai ) is the Frechet derivative of p evaluated at ai : If

(e1 , . . . , en ) is the standard basis for Rn , then for v : Rn → R, we have,
∂v
(Dv)(x)(ei ) = (x), the usual partial derivative.
∂xi P
Notice that we may replace K by the set
X
′ ( )
∂p ∂p
= p(ai ), 1 ≤ i ≤ 3; p(a123 ); (ai ), (ai ), 1 ≤ i ≤ 3 .
K
∂x1 ∂x2
Remark 4.2. The term “Hermite” means that we assume knowledge of

derivatives at some of the nodes. If only the values of p at the nodes ap-
pear in the set of degrees of freedom, as was the case upto Example 4.5,
we refer to the finite elements as of “Lagrange” type. These ideas will
be made precise in Sec. 5. We usually indicate degrees of freedom in-
volving derivatives by circling the nodes - one circle for first derivatives,
two for first and second derivatives and so on. Thus the finite element
of example 4.6 may be pictured as in Fig. 4.6.
smartworlD.asia
Figure 4.6:
Exercise 4.6. The Hermite Triangle of Type (3′ ).

This is also known
( as the Zienkiewicz triangle in) Engineering lit-
P ∂p ∂p P
erature. Set K = p(ai ), (ai ), (ai ), 1 ≤ i ≤ 3 . Show that K
∂x1 ∂x2
uniquely determines a function in the space
 


 X
3 X
3 


PK = 
 p ∈ P 3 ; 6p(a123 ) − 2 p(ai ) + Dp(ai )(ai − a123 ) = 0
 .
 
i=1 i=1
All examples cited upto now yield the inclusion Vh ⊂ C 0 (Ω) and 46
consequently are useful to solve second-order problems. In order to
solve problems of fourth order, we need the inclusion Vh ⊂ C 1 (Ω). Our
subsequent examples will be in this direction.
Remark 4.3. Consider a 1-simplex K ⊂ R1 . A triangulation is merely

a subdivision of Ω into subintervals. In any subinterval K not only v|K
d(v|K)
but also must be continuous at both end points. Thus we get
dx
4 conditions on v|K. Consequently PK must contain all polynomials
of degree 3 in it. The analogous result (which is non-trivial) is due
to A. Ženišek [24] that is case of R2 , and K a triangle of R2 , at least
polynomials of degree 5 must be contained in PK .
Example 4.7. The Argyris triangle.

This is also known as the 21-degree-of-freedom-triangle. We set
PK = P5 and

X  
 ∂p ∂2 p
=
 p(ai ), (a i ), . . . , (ai ), 1 ≤ i ≤ 3;
 ∂x1 ∂x22
K
)
∂p
smartworlD.asia
(ai j ), 1 ≤ i ≤ j ≤ 3 .
∂ν
Figure 4.7:
∂p
47 The knowledge of the normal derivative is indicated by a line
∂ν
perpendicular to the side at the appropriate point; cf. Fig. 4.7.
P
We now show that any p ∈ PK is uniquely determined by K . Let
p ∈ PK = P5 be given such that all its degrees of freedom are zero. If K ′
is any side of K and t is an abscissa along K ′ then p|K ′ is a polynomial

dp d2 p
p1 (t) of degree 5. The vanishing of p, , at the end points, say b,
′ ′
dt dt2
b , of K imply that all the 6 coefficients of p1 are 0 and hence p1 ≡ 0.
dp ∂p
Thus p = 0 = on K ′ . The polynomial r(t) = (t) is of degree 4 on
dt ∂ν !
′ ′ dr dr ′ b + b′
K and we also have r(b) = r(b ) = (b) = (b ) = r =0
dt dt 2
∂p ∂p
which imply that r ≡ 0 on K ′ . Hence p, , all vanish on the
∂x1 ∂x2
sides of the triangle K. The sides of K are defined by the equations
λi (x1 , x2 ) = 0, (i = 1, 2, 3) where λi are the barycentric coordinate func-
tions. We claim that λ2i divides p for i = 1, 2, 3. To see this it is enough
∂p ∂p
to prove that if p is a polynomial such that p, , vanish on any
∂x1 ∂x2
straight line L = {(x1 , x2 ); λ(x1 , x2 ) = 0} then λ2 divides p. In the special
P j
case, when λ(x1 , x2 ) = x1 writing p(x1 , x2 ) = 5j=0 a j (x2 )x1 (with deg.
∂p
a j ≤ 5 − j) it follows that a0 (x2 ) = a1 (x2 ) = 0 since p = = 0 on
∂x1
L. Thus x21 divides p. The general case reduces to this case by an affine
transformation. In fact, by translating the origin to a point P, fixed ar-
smartworlD.asia
bitrarily on L and by rotation of the coordinate axes we can assume that
L = {(X1 , X2 ); X1 = 0} in the new coordinates. If p′ is the image of p
under this transformation then p′ is also a polynomial (of degree 5) and
∂p′ ∂p′
p′ , , vanish on L by chain rule for differentiation. Hence X12 di-
∂X1 ∂X2
vides p′ . This is the same thing as saying λ2 divides p which proves the
claim. Since λi are mutually coprime we may now write p = qλ21 λ22 λ23 .
Then we necessarily have q(x1 , x2 ) ≡ 0 for, otherwise deg. p ≥ 6 which
P
is impossible since p ∈ P5 . Hence p ≡ 0 on K which proves that K
determines p ∈ P5 .
To define the corresponding space Vh , we number all the vertices of
the triangles by {b j } and all midpoints of the sides by {ck }. The the set
of degrees of freedom of the space Vh is 48
 
X  
 ∂v ∂v ∂2 v ∂2 v ∂2 v ∂v 


=  v(b j ), (b j ), (b j ), (b j ), (b j ), (b j ), (c )
k 

 ∂x1 ∂x2 ∂x 2 ∂x1 ∂x2 ∂x 2 ∂νk 
h 1 2
∂
where is one of the two possible normal derivatives at the mid-point
∂νk
ck .
We now show that Vh ⊂ C 1 (Ω). Consider two adjacent Argyris trian-
gles K1 and K2 with common boundary K ′ along which t is an abscissa
(Fig. 4.8). Let v ∈ Vh . Now v|K1 and v|K2 are polynomials of degree 5 in
t along K ′ and they agree together with their first and second derivatives
at the end points. Thus v|K1 = v|K2 on K ′ , proving continuity.
∂(v|K1 ) ∂(v|K2 )
Now and along K ′ are polynomials of degree 4
∂ν ∂ν
agreeing in their values with first derivatives at end points and agree at
∂(v|K1 ) ∂(v|K2 )
the mid-point in their values. Thus = on K ′ . Similarly,
∂ν ∂ν
∂(v|K1 ) ∂(v|K2 )
= on K ′ and hence v ∈ C 1 (Ω). Thus Vh ⊂ C 1 (Ω).
∂t ∂t
Exercise 4.7. The 18-Degree-of-Freedom-Triangle

Let K be a triangle in R2 . Let PK consist of those polynomials of
degree ≤ 5 for which, along each side of K, the normal derivative is
a polynomial of degree ≤ 3, in one variable of course. Show that a
polynomial in PK is uniquely determined by the following set of degrees
of freedom:
 
X  
 ∂p ∂2 p 


=  p(ai ), (ai ), . . . , (ai ), 1 ≤ i ≤ 3.

 ∂x
1 ∂x 2 
K 2
49
Note that P4 ⊂ PK and dim PK = 18.
Exercise 4.8. The HCT-Triangle.

This element is due to Hsieh, Clough and Tocher. Let a be any inte-
rior point of the triangle K with vertices a1 , a2 , a3 . With a as common
vertex subdivide the triangle into triangles K1 , K2 , K3 ; cf. Fig. 4.9. De-
fine n o
PK = p ∈ C 1 (K); P|Ki ∈ P3 , 1 ≤ i ≤ 3 .
Obviously, P3 ⊂ PK . The degrees of freedom are given by
X ( ∂p ∂p ∂p
)
= p(ai ), (ai ), (ai ), 1 ≤ i ≤ 3; (ai j ), 1 ≤ i < j ≤ 3 .
K
∂x1 ∂x2 ∂ν
Show that
smartworlD.asia
P
K
Figure 4.9:
uniquely determines p ∈ PK .
Note: Since we have to determine 3 polynomials pi = p|Ki each of

degree ≤ 3, we need to determine 30 coefficients on the whole. For this
we have the following conditions:
(i) The values at the vertices together with first derivatives and also
the normal derivative at the mid points give 7 conditions for each
pi = p|Ki . Thus we have 21 conditions from these.
(ii) p1 (a) = p2 (a) = p3 (a) gives 2 conditions.

∂p1 ∂p2 ∂p3
(iii) (a) = (a) = (a) for i = 1, 2, gives 4 more conditions. 50
∂xi ∂xi ∂x1
∂p1 ∂p2
(iv) = along a1 a and two more similar conditions give 3
∂ν ∂ν
conditions.
Thus we have 30 conditions to determine the 30 coefficients. But,

of course this is no proof, which is left as an exercise!
Exercise 4.9. The Bogner-Lox-Schmidt Rectangle; cf. Fig. 4.10.
Figure 4.10:
Let PK = Q3 , the degrees of freedom being given by
smartworlD.asia
X ( ∂p ∂p ∂2 p
)
= p(ai ), (ai ), (ai ), (ai ), 1 ≤ i ≤ 4 .
K
∂x1 ∂x2 ∂x1 ∂x2
P
Show that K determines uniquely a polynomial p ∈ Q3 (a double
dotted arrow indicates that the mixed second derivative is a degree of
freedom). Show also that in this case Vh ⊂ C 1 (Ω).
So far, we have verified requirements (i) and (ii) mentioned at the

beginning of this Section. Let us now examine requirement (iii), which
P
will be fulfilled by a “canonical” choice for the basis functions. Let h
be the set of degrees of freedom of the space Vh derived in an obvious
P P
way from the sets K , K ∈ kh ; Examples of such sets h have been
51 given for n-simplices of type (k) and for Argyris triangles. Then if
X n o
= ϕ jh , 1 ≤ j ≤ M ,
h
we let the basis functions w j , 1 ≤ j ≤ M, be those functions in the space

Vh which satisfies
ϕi (w j ) = δi j , 1 ≤ i ≤ M.
Then it is easily seen that this choice will result in functions with
“small” support: in Fig. 4.11, we have represented there types of sup-
ports encountered in this fashion, depending upon the position of the
node associated
support of basis
function associated support of basis
with node function associated
with node
support of basis
function associated
with node
smartworlD.asia
Figure 4.11:
with the degree of freedom.
smartworlD.asia
Chapter 5
General Properties of Finite

Elements
IT WOULD HAVE been observed that upto now we have not defined fi- 52
nite elements in a precise manner. Various polygons like triangles, rect-
angles, etc. were loosely called finite elements. We rectify this omission
and make precise the ideas expressed in the previous sections.
smartworlD.asia
Definition 5.1. A finite element is a triple (K, Σ, P) such that
(i) K ⊂ Rn with a Lipschitz continuous boundary ∂K and Int K , φ.
(ii) Σ is a finite set of linear forms over C ∞ (K). The set Σ is said to be
the set of degrees of freedom of the finite element.
(iii) P is a finite dimensional space of real-valued functions over K

N and α , 1 ≤ i ≤ N
such that Σ is P-unisolvent: i.e. if Σ = {ϕi }i=1 i
are any scalars, then there exists a unique function p ∈ P such that
(5.1) ϕi (p) = αi , 1 ≤ i ≤ N.
Condition (iii) of definition (5.1) is equivalent to the conditions that

dim P = N = card Σ and that there exists a set of functions {p j }Nj=1 with
ϕi (p j ) = δi j (1 ≤ i, j ≤ N), which forms a basis of P over R. Given any
53
54 5. General Properties of Finite Elements
p ∈ P we may write
X
N
(5.2) p= ϕi (p)pi .
i=1
P P
Instead of (K, , P) one writes at times (K, K , PK ) for the finite
element.
In the various examples we cited in Sec. 4 our set of degrees of free-
dom for a finite element K (which was an n-simplex or hyper-rectangle)
53 had elements of the following type:
Type 1: ϕ0i given by p 7→ p(a0i ). The points {a0i } were the vertices, the
mid-points of sides, etc...
Type 2: ϕ1i,k given by p 7→ Dp(a1i )(ξi,k

1 ). For instance, in the Hermite
triangle of type (3) (cf. Example 4.6), we had a1i = ai , ξi,k

1 = a −a ,
i k
where a1 , a2 , a3 were the vertices.
Type 3: ϕ2i,kl given by p 7→ D2 p(a2i )(ξi,k

2 , ξ 2 ). For example, in the 18-
i,l
degree-of-freedom triangle, a2i = ai , ξi,k2 = e = ξ 2 , the unit vec-
1 i,1
smartworlD.asia
∂2 p
tor in the x1 -direction so that we have D2 p(ai )(e1 , e1 ) = (ai )
∂x21
as a degree of freedom. (cf. Exercise 4.7).
In all these cases the points {ais } for s = 0, 1 and 2, are points of K
and are called the nodes of the finite element.
Definition 5.2. A finite element is called a Lagrange finite element if its

degrees of freedom are only of Type 1. Otherwise it is called a Hermite
finite element. (cf. Remark 4.2)
Let (K, Σ, P) be a finite element and v : K → R be a “smooth”

function on K. Then by virtue of the P-unisolvency of Σ, there exists a
unique element, say, πv ∈ P such that ϕi (πv) = ϕi (v) for all 1 ≤ i ≤ N,
N . The function πv is called the P-interpolate function
where Σ = {ϕi }i=1
of v and the operator π : C ∞ (K) → P is called the P-interpolation
5. General Properties of Finite Elements 55
operator. If {p j }Nj=1 is a basis for P satisfying ϕi (p j ) = δi j for 1 ≤ i,

j ≤ N then we have the explicit expression
X
N
(5.3) π(·) = ϕi (·)pi .
i=1
Example 5.1. In the triangle of type (1) (see Example 4.1), P = P1 ,

Σ = {ϕi ; ϕi (p) = p(ai ), 1 ≤ i ≤ 3} and pi = λi , the barycentric coordinate 54
functions. Thus we also have
X
3
(5.4) πv = v(ai )λi .
i=1
Exercise 5.1. Let K be a triangle with vertices a1 , a2 and a3 . Let ai j (i <

j) be the mid-point of the side joining ai and a j . Define ΣK = {p 7→
p(ai j ), 1 ≤ i ≤ j ≤ 3}. Show that Σ is P1 -unisolvent and that in general
Vh 1 C 0 (Ω) for a triangulation made up of such finite elements.
Exercise 5.2. Let K be a rectangle in R2 with vertices a1 , a2 , a3 , a4 . Let

a5 , a6 , a7 , a8 be the midpoints of the sides as in Fig. 4.5. If Σ = {p 7→
p(ai ), 5 ≤ i ≤ 8}, show that Σ is not Q1 -unisolvent.
smartworlD.asia
Let us now consider a family of finite elements of a given type. To
be more specific, we will consider for instance a family of triangles of
type (2) (see Example 4.2), but our subsequent descriptions extend to all
types of finite elements in all dimensions.
Pick, in particular, a triangle K̂ with vertices {â1 , â2 , â3 } from this
family. Let the mid-points of the sides be {â12 , â23 , â13 }. Set P̂ = PK̂ =
P2 and define accordingly the associated set of degrees of freedom for
K̂ as
X̂ X
= = {p 7→ p(âi ), 1 ≤ i ≤ 3; p 7→ p(âi j ), 1 ≤ i < j ≤ 3}.
K̂
In as much as we consider the finite element (K̂, Σ̂, P̂) as fixed in the
sequel, it will be called the reference finite element of the family.
Given any finite element K with vertices a1 , a2 , a3 in this family,
there exists a unique invertible affine transformation of R2 i.e. of the
form F K ( x̂) = BK x̂ + bK , where BK is an invertible 2 × 2 matrix and

55 bK ∈ R2 , such that F K (K̂) = K and F K (âi ) = ai , 1 ≤ i ≤ 3. It is
easily verified that F K (âi j ) = ai j for 1 ≤ i < j ≤ 3. Also, the space
{p : K → R; p = p̂ ◦ F K−1 , p̂ ∈ P̂} is precisely the space PK = P2 . Hence
the family {(K, Σ, P)} is equivalently defined by means of the following
data:
(i) A reference finite element (K̂, Σ̂, P̂),
(ii) A family of affine mappings {F K } such that F K (K̂) = K, ai =

F K (âi ),
1 ≤ i ≤ 3, ai j = F K (âi j ), 1 ≤ i < j ≤ 3, and

ΣK = {p 7→ p(F K (âi )); p 7→ p(F K (âi j ))},
PK = {p : K → R; p = p̂ ◦ F K−1 , p̂ ∈ P̂}.
This special case leads to the following general definition.
Definition 5.3. Two finite elements (K̂, Σ̂, P̂) and (K, Σ, P) are affine
equivalent if there exists an affine transformation F on Rn such that
smartworlD.asia
(i) F( x̂) = B x̂ + b, b ∈ Rn , B an invertible n × n matrix,
(ii) K = F(K̂),
(iii) ais = F(âis ), s = 0, 1, 2,

1 = Bξ̂ 1 , ξ 2 = Bξ̂ 2 , ξ 2 = Bξ̂ 2
(iv) ξi,k i,k i,k i,k i,l i,l′
and
(v) P{p : K 7→ R; p = p̂ ◦ F −1 , p̂ ∈ P̂}.
This leads to the next definition.
Definition 5.4. A family {(K, ΣK , PK )} of finite elements is called an

affine family if all the finite elements (K, ΣK , PK ) are equivalent to a
single reference finite element (K̂, Σ̂, P̂).
5. General Properties of Finite Elements 57
56 Let us see why the relations given by (iv) must be precisely of that
form. We have by (v), p(x) = p̂( x̂). This must be valid when we use the
basis functions as well. We have:
X X
p̂( x̂) = p̂(â0i ) p̂0i ( x̂) + D p̂(â1i )(ξ̂i,k
1
) p̂1i,k ( x̂)
i i,k
X
2
+ D p̂(â2i )(ξ̂i,k
2 2
, ξ̂i,l ) p̂2i,kl ( x̂).
i,k,l
Now D p̂(â1i )(ξ̂i,k

1 ) = Dp(a1 )Bξ̂ 1 by a simple application of the chain
i i,k
rule and therefore
D p̂(â1i )(ξ̂i,k
1
) = Dp(a1i )ξi,k
1
, by (iv).
By a similar treatment of the second derivative term, we get
X X
p̂( x̂) = p(a0i )p0i (x) + Dp(a1i )(ξi1 )(ξi,k
1
)pi,k (x)
i i,k
X
+ D2 p(a2i )(ξik2 , ξil2 )pikl (x) = p(x).
i,k,l
Thus the relations (iv) and (v) are compatible.
smartworlD.asia
Theorem 5.1. Let (K, Σ, P) and (K̂, Σ̂, P̂) be affine equivalent with F K
as the affine transformation. If v : K → R induces v̂ : K̂ → R by
b = π̂v̂.
v̂( x̂) = v(x) for x̂ ∈ K̂, (x = F K ( x̂)), then πv
N , Σ = {ϕ } N . By definition,
Proof. Let Σ̂ = {ϕ̂i }i=1 i i=1
b = ϕi (πv) = ϕi (v), 1 ≤ i ≤ N. ϕ̂i (π̂v̂) = ϕ̂i (v̂) = ϕi (v), 1 ≤ i ≤ N.

ϕ̂i (πv)
b for 1 ≤ i ≤ N. Hence π̂v̂ = πv
Thus, ϕ̂i (π̂v̂) = ϕ̂i (πv) b by uniqueness
of the function π̂v̂. 57
Let us consider a polygonal domain Ω with a triangulation th . Sup-

pose to each K ∈ th is associated a finite element, (K, ΣK , PK ), ΣK being
the set of degrees of freedom, and PK the finite dimensional space such
that ΣK is PK -unisolvent. Then we have defined the interpolation opera-
tor πK . All these make sense locally i.e. at a particular finite element K.
We now define the global counterparts of these terms. The comparison
is given in the following table.
Table 5.1.
Local definition Global definition

S
1. Finite element K. 1. The set Ω = K
K=tn
2. The boundary of K, ∂K. 2. The boundary of Ω, ∂Ω = Γ.
3. The space PK of functions 3. The space Vh of functions
K → R, which is finite- Ω → R, which is also finite-
dimensional. dimensional.
P N of
4. The set K = {ϕi,K }i=1 4. The set of degrees of free-
P N , where
degrees of freedom of K. dom h = {ϕi }i=1
ϕi (p|K) = ϕi,K (p|K).
5. Basis functions of PK are 5. Basis function of Vh are {w j }.
N
{pi,K }i=1 .
6. The nodes of K are 6. The nodes of th are by def-
S
{a0i , a1i , a2i , . . .}. inition, {Nodes of K} =
K∈th
∪{b j } say.
7. πK is the PK -interpolation 7. The Vh interpolation opera-
operator, defined by tor πh is defined by πh v ∈
smartworlD.asia
ϕi,K (πK v) = ϕi,K (v), for all Vh such that, ϕi,K (πh v|K) =
P P
ϕi,K ∈ K . ϕi,K (v|K) for all ϕi,K ∈ K .
58
Notice that, by definition,
(5.5) (πh v)|K = πK (v|K) for all K ∈ th .
It is this property and the conclusion of theorem 5.1 that will be

essential in our future error analysis.
Definition 5.5. We say that a finite element of a given type is of class

C 0 , resp. of class C 1 , if, whenever it is the generic finite element of a
triangulation, the associated space Vh satisfies the inclusion Vh ⊂ C 0 (Ω),
resp. Vh ⊂ C 1 (Ω). By extension, a triangulation is of class C 0 , resp. of
class C 1 if it is made up of finite elements of class C 0 , resp. of class C 1 .
Reference: A forthcoming book of Ciarlet and Raviart [5].
Chapter 6
Interpolation Theory in
Sobolev Spaces
WE OUTLINED THE internal approximation method in Sec. 3. We are 59

naturally interested in the convergence of the solutions uh ∈ Vh to the
global solution u ∈ V. As a key step in this analysis we obtained the
error estimate (cf. Theorem 3.1):
smartworlD.asia
(6.1) ||u − uh || ≤ C inf ||u − vh ||.
vh ∈Vh
To be more specific let us consider an example. Given Ω ⊂ R2 a

polygon, consider the solution of the following problem, which is there-
fore posed in the space V = H01 (Ω):



−∆u + au = f in Ω,
(6.2) 

u = 0 on Γ.
Let th be a triangulation of Ω by triangles of type (1), (2) or (3).

Then uh ∈ Vh ⊂ H01 (Ω) and (6.1) reads as
(6.3) ||u − uh ||1,Ω ≤ C inf ||u − vh ||1,Ω .

vh ∈Vh
We know ‘a priori’ that u ∈ H01 (Ω). Let us assume for the moment
that u ∈ C 0 (Ω). (Such assumptions are made possible by the various
59
60 6. Interpolation Theory in Sobolev Spaces
regularity theorems. For instance, u ∈ H 2 (Ω) ⊂ C 0 (Ω) if f ∈ L2 (Ω)

and Ω is a convex polygon). If u ∈ C 0 (Ω), then we may define the
Vh -interpolate of u, i.e., πh u by πh u(b j ) = u(b j ) for the nodes b j of the
triangulation. Note also that πh u|K = πK u (cf. (5.5)). Now from (6.3)
we get,
||u − uh ||1,Ω ≤ C||u − πh u||1,Ω

  12
 X 
= C  ||u − πh u||1,K 
2
K∈th
  12
 X 
= C  ||u − πK u||21,K 
K∈th
60
Thus the problem of estimating ||u − uh ||1,Ω is reduced to the problem
of estimating ||u − πK u||1,K . This is one central problem in the finite ele-
ment method and motivates the study of interpolation theory in Sobolev
spaces.
We consider more general types of Sobolev spaces for they are no
more complicated for this purpose than those defined in Sec. 2.
smartworlD.asia
Definition 6.1. Let m ≥ 0 be an integer, and 1 ≤ p ≤ +∞. Then the
Sobolev space W m,p (Ω) for Ω ⊂ Rn , open, is defined by
W m,p (Ω) = {v ∈ L p (Ω); ∂α v ∈ L p (Ω) for all |α| ≤ m}.
Remark 6.1. H m (Ω) = W m,2 (Ω).
On the space W m,p (Ω) we define a norm || · ||m,p,Ω by

Z 1/p
 X 
(6.4) ||v||m,p,Ω =   |∂ v| dx
α p
Ω |α|≤m
and the semi-norm | · |m,p,Ω by

Z 1/p
 X 
(6.5) |v|m,p,Ω =  |∂α v| p dx
Ω |α|=m|
6. Interpolation Theory in Sobolev Spaces 61
If k ≥ 1 is an integer, consider the space W k+1,p (Ω)/Pk . If v̇ stands

for the equivalence class of v ∈ W k+1,p (Ω) we may define the analogues
of (6.4) and (6.5) respectively by
(6.6) ||v̇||k+1,p,Ω = inf ||v + p||k+1,p,Ω

p∈Pk
and 61
(6.7) |v̇|k+1,p,Ω = |v|k+1,p,Ω .
These are obviously well-defined and || · ||k+1,p,Ω defines the quotient

norm on the quotient space above. We then have the following key
result, whose proof may be found in Nečas [20] for instance.
Theorem 6.1. In W k+1,p (Ω)/Pk , the semi-norm |v̇|k+1,p,Ω is a norm
equivalent to the quotient norm ||v̇||k+1,p,Ω , i.e., there exists a constant
C = C(Ω) such that for all v̇ ∈ W k+1,p (Ω)/Pk
(6.8) |v̇|k+1,p,Ω ≤ ||v̇||k+1,p,Ω ≤ C|v̇|k+1,p,Ω .
Equivalently, we may state
smartworlD.asia
Theorem 6.2. There exists a constant C = C(Ω) such that for each
v ∈ W k+1,p (Ω)
(6.9) inf. ||v + p||k+1,p,Ω ≤ C|v|k+1,p,Ω .

p∈Pk
(Note: This result holds if Ω has a continuous boundary and if it is

bounded so that Pk ⊂ W k+1,p (Ω).)
We now prove the following
Theorem 6.3. Let W k+1,p (Ω) and W m,q (Ω) be such that W k+1,p (Ω) ֒→
W m,q (Ω) (continuous injection). Let π ∈ L (W k+1 , ρ(Ω), W m,q (Ω)), i.e.
a continuous linear map, such that for each p ∈ Pk , πp = p. Then there
exists C = C(Ω) such that for each v ∈ W k+1,p (Ω)
|v − πv|m,q,Ω ≤ C||I − π||L (W k+1 ,P(Ω),W m,q (Ω)) |v|k+1,p,Ω
62
Proof. For each v ∈ W k+1,p (Ω) and for each p ∈ Pk , we can write
v − πv = (I − π)(v + p).
Thus,
|v − πv|m,q,Ω ≤ ||v − πv||m,q,Ω

= ||I − π||L (W k+1,p (Ω),W m,q (Ω)) ||v + p||k+1,p,Ω ,
for all p ∈ PK . Hence,
|v − πv|m,q,Ω ≤ ||I − π||L (W k+1,p (Ω),W m,q (Ω)) inf ||v + p||k+1,p,Ω
p∈Pk
≤ |C||I − π||L (W k+1,p (Ω),W m,q (Ω)) |v|k+1,p,Ω
By theorem 6.2, this completes the proof.
Definition 6.2. Two open subsets Ω, Ω̂ of Rn are said to be affine equiv-

alent if there exists an invertible affine map F mapping x̂ to B x̂ + b, B an
invertible (n × n) matrix and b ∈ Rn , such that F(Ω̂) = Ω.
smartworlD.asia
If Ω, Ω̂ are affine equivalent, then we have a bijection between their
points given by x̂ ↔ x = F( x̂). Also we have bijections between smooth
functions on Ω and Ω̂ defined by (v : Ω → R) ↔ (v̂ : Ω̂ → R) where
v(x) = v̂( x̂).
The following theorem gives estimates of |v|m,p,Ω and |v̂|m,p,Ω̂ each
in terms of the other.
63 Theorem 6.4. Let Ω, Ω̂ ⊂ Rn be affine equivalent. Then there exist

constants C, Ĉ such that for all v ∈ W m,p (Ω)
(6.11) |v̂|m,p,Ω̂ ≤ C||B||m || det B|−1/p |v|m,p,Ω
and for all v̂ ∈ W m,p (Ω̂)
(6.12) |v|m,p,Ω ≤ C||B−1 ||m | det B|1/p |v̂|m,p,Ω̂ .
Note:
(i) It suffices to prove either (6.11) or (6.12). We get the other by

merely interchanging the roles of Ω and Ω̂. We will prove the
former.
(ii) ||B|| is the usual norm of the linear transformation defined by B,

||Bx||
viz. ||B|| = sup . (Recall that F(Ω̂) = Ω, F( x̂) = B x̂ + b).
n ||x||
x∈R
x,0
Proof. Let {e1 , . . . , en } be the standard basis for Rn . Let α be a multi-

index with |α| = m. By choosing a suitable collection {e1α , . . . , emα }
with appropriate number of repetitions from the basis, we may write,
(∂α v̂)( x̂) = (Dm v̂)( x̂)(e1α , . . . , emα ),
where Dm v̂ is the mth order Fréchet derivative of v̂ and Dm v̂( x̂) is conse-
quently an m-linear form on Rn . Thus,
|∂α v̂( x̂)| ≤ ||Dm v̂( x̂)|| = sup |Dm v̂( x̂)(ξ1 , . . . , ξm )|.
||ξi ||=1
1≤i≤m
Since this is true for all |α| = m, we get
smartworlD.asia
Z !1/p
m p
(6.13) |v̂|m,p,Ω̂ ≤ C1 ||D v̂( x̂)|| d x̂ ≤ C2 |v̂|m,p,Ω̂ .
Ω̂
The first inequality is a consequence of our preceding argument.

The second follows by a straightforward argument. By composition of
functions in differentiation:
(6.14) Dm v̂( x̂)(ξ1 , . . . , ξm ) = Dm v(x)(Bξ1 , . . . , Bξm);
This gives 64
(6.15) ||Dm v̂( x̂)|| ≤ ||Dm v(x)|| ||B||m .
Hence the first inequality in (6.13) may be rewritten as

Z
|v̂| p ≤ C1p ||B||mp ||Dm v(F( x̂))|| p d x̂
m,p,Ω̂ Ω̂

Z
p
= C1 ||B||mp | det B|−1 ||Dm v(x)|| p dx
Ω
≤ C p ||B||mp | det B|−1 |v|m,p,Ω .
by an inequality similar to the second inequality of (6.13). Raising to
power 1/p on either side we get (6.11). This completes the proof.
We now estimate the norms ||B|| and ||B−1 || in terms of the ‘sizes’ of
Ω and Ω̂. More precisely, if h, (resp. ĥ) the supremum of the diameters
of all balls that can be inscribed in Ω, (resp. Ω̂), we have the following:
Theorem 6.5. ||B|| ≤ h/ρ̂, and ||B−1 || ≤ ĥ/ρ.
Proof. Again it suffices to establish one of these. Now,
!
1
||B|| = sup ||Bξ|| .
||ξ||=ρ̂ ρ̂
Let ξ ∈ Rn with ||ξ|| = ρ̂. Choose ŷ, ẑ ∈ Ω̂ such that ξ = ŷ − ẑ. Then
Bξ = Bŷ − Bẑ = y − z, where F(ŷ) = y, F(ẑ) = z. But y, z ∈ Ω and hence
||y − z|| ≤ h. Thus ||Bξ|| ≤ h. Hence ||B|| ≤ h/ρ̂, which completes the
proof.
smartworlD.asia
We conclude this section with an important, often used, result.
Theorem 6.6. Let (K̂, Σ̂, P̂) be a finite element. Let s(= 0, 1 or 2) be
the maximal order of derivatives occurring in Σ̂. Assume that:
65 (i) W k+1,p (K̂) ֒→ C s (K̂)
(ii) W k+1,p (K̂) ֒→ W m,q (K̂)
(iii) Pk ⊂ P̂ ⊂ W m,q (K̂)
Then there exists a constant C = C(K̂, Σ̂, P̂) such that for all affine
equivalent finite elements (K, Σ, P) we have
1 1 hk+1
(6.16) |v − πK v|m,q,K ≤ C(meas K) q − p K
|v|k+1,p,K
ρm
K
for all v ∈ W k+1,p (K), where hK is the diameter of K and ρK is the

supremum of diameters of all balls inscribed in K.
Proof. Since Pk ⊂ P̂, for any polynomial p ∈ Pk we have π̂p = p. We

may write
X X
π̂v̂ = v̂(â0i ) p̂0i + (Dv̂(â1i )(ξ̂i,k
1
)) p̂1i,k
i i,k
(6.17) X .
= + (D2 v̂(â2i )(ξ̂i,k
2 2
, ξ̂i,l )) p̂2i,k,l ,
i,k,l
all these sums being finite (the second and third may or may not be
present). We claim that π̂ ∈ L (W k+1 , p(K̂), W m,q (K̂)). Since P̂ ⊂
W m,q (K̂) all the basis functions in (6.17) are in W m,q (K̂). Thus,
X
||π̂v̂||m,q,K̂ ≤ |v̂(â0i )| || p̂0i ||m,q,K̂
i
X
(6.18) + |Dv̂(â1i )(ξi,k
1
)| || p̂1ik ||m,q,K̂
i,k
X
+ |D2 v̂(â2i )(ξ̂i,k
2 2
, ξ̂i,l )| || p̂2ikl ||m,q,K̂
i,k,l
Since W k+l,p (K̂) ֒→ C s (K̂) and all the numbers v̂(â0i ), etc. . . , are
bounded by their essential supremum over K̂,
smartworlD.asia ||π̂v̂||m,q,K̂ ≤ C||v̂||k+1,p,K̂ .
Hence the claim is valid. Now by virtue of (ii) and also our obser-
66
vation on preservation of polynomials, we may apply theorem 6.3 to π̂.

Hence there exists C = C(K̂, Σ̂, P̂) such that
|v̂ − π̂v̂|m,q,K̂ ≤ C|v̂|k+1,p,K̂ for v̂ ∈ W k+1,p (K̂).
Notice that π̂v̂ = πd d

K v by Theorem 5.1. Thus v̂ − π̂v̂ = v − πK v. Thus
if F K (K̂) = K where F K ( x̂) = BK x̂ + bK , we get
(6.19) |v − πK v|m,q,K ≤ C1 ||B−1 m

K || | det B K |
1/q
|v̂ − π̂v̂|m,q,K̂ ,
by Theorem 6.4. Also by the same theorem
(6.20) |v̂|k+1,p,K̂ ≤ C2 ||BK ||k+1 | det BK |−1/p |v|k+l,p,K .
Further | det BK | being the Jacobian of the transformation, we have

meas K hK
| det BK | = and ||BK || ≤ , ||B−1
K || ≤ ĥ/ρK by theorem 6.5.
meas K̂ ρ̂
Since ĥ, ρ̂, meas K̂ are constants, combining (6.19), (6.20) and the pre-
ceding observation we complete the proof of the theorem.
References: See Bramble and Hilbert [28], Bramble and Zlq́mal [2],
Ciarlet and Raviart [7], Ciarlet and Wagschal [8], Strang [21], Ženišek
[23, 24] and Zlámal [25, 32].
smartworlD.asia
Chapter 7
Applications to
Second-Order Problems
Over Polygonal Domains
WE APPLY THE results of the preceding section in studying the conver- 67

gence of the finite element method, i.e. the convergence of the solutions
smartworlD.asia
uh of (Ph ) to the solution u of a problem (P) which corresponds to the
choice V = H 1 (Ω) or H01 (Ω), which we saw in Sec. 2 led to second-order
problems.
Let Ω be a polygonal domain throughout.
Definition 7.1. A family (th ) of triangulations of Ω is regular is
(i) for all th and for each K ∈ th , the finite elements (K, Σ, P) are
all affine equivalent to a single finite element, (K̂, Σ̂, P̂) called the
reference finite element of the family;
(ii) there exists a constant σ such that for all th and for each K ∈ th
we have
hK
(7.1) ≤σ
ρK
where hK , ρK are as in Theorem 6.6;
67
68 7. Applications to Second-Order Problems...
(iii) for a given triangulation th if
(7.2) h = max hK ,
K∈th
then h → 0.
Remark 7.1. The condition (ii) in definition 7.1 assures us that as h → 0

the triangles do not become “flat”; cf. Exercise 7.1.
Exercise 7.1. If n = 2 and the sets K are triangles, show that condition
(ii) of definition 7.1 is valid if and only if there exists θ0 > 0 such that
68 for all th and K ∈ th , θK ≥ θ0 > 0, θK being the smallest angle in K.
Exercise 7.2. Consider the space Vh associate with th . Since Vh is finite

dimensional all norms are equivalent and hence
|vh |0,∞,Ω ≤ Ch |vh |0,Ω for all vh ∈ Vh ,
for some constant Ch , a priori dependent upon h, which we may evaluate

as follows: If (th ) is a regular family of triangulations, show that there
exists a constant C, independent of h, such that
smartworlD.asia
C
(7.3) |vh |0,∞,Ω ≤ |vh |0,Ω for vh ∈ Vh .
hn/2
Also show that there exists a constant C such that
C
(7.4) |vh |1,Ω ≤ |vh |0,Ω for all vh ∈ Vh .
h
We now obtain an estimate for the error ||u − uh ||1,Ω when the family
of triangulations is regular, which also gives convergence.
Theorem 7.1. Let (th ) be a regular family of triangulations on Ω of
class C 0 (i.e. Vh ⊂ C 0 (Ω)) with reference finite element (K̂, Σ̂, P̂). We
assume that there exists an integer k ≥ 1 such that
(i) PK ⊂ P̂ ⊂ H 1 (K̂)
(ii) H k+1 (K̂) ֒→ C s (K̂) where s(= 0, 1, or 2) is the maximal order of

P̂
derivatives in .
7. Applications to Second-Order Problems... 69
(iii) u ∈ H k+1 (Ω) (Regularity assumption).

Then there exists a constant C (independent of Vh ) such that
(7.5) ||u − uh ||1,Ω ≤ Chk |u|k+1,Ω .
Proof. Since Vh ⊂ C 0 (Ω), PK ⊂ H 1 (K), we have Vh ⊂ V. By (ii) and 69

(iii) of the hypothesis we have that the Vh -interpolate of u, viz. πh u is
well-defined. Since πh u ∈ Vh , by our fundamental result (see Theo-
rem 3.1 or relation (6.1)), it suffices to estimate ||u − πh u||1,Ω . Now,
H k+1 (K̂) ֒→ C s (K̂),
H k+1 (K̂) ֒→ H 1 (K̂) (k ≥ 1),
Pk ⊂ P ⊂ H 1 (K̂),
and we may apply Theorem 6.6 with p = q = 2, m = 1 to get
hk+1
K
|u − πK u|1,K ≤ C|u|k+1,K
(7.6) ρK
hK
≤ C|u|k+1,K hkK (Since ≤ σ).
ρK
Similarly with m = 0 we get
(7.7)
smartworlD.asia
These together give
|u − πK u|0,K ≤ C|u|k+1,K hk+1
K .
(7.8) ||u − πK u||1,K ≤ ChkK |u|k+1,K .

Now since hK ≤ h,
  12
 X 
||u − πh u||1,Ω =  ||u − πK u||1,K 
2
K∈th
  12
 X 
≤ C hk  |u|2k+1,K 
K∈th
K
= C h |u|k+1,Ω .
Example 7.1. Consider triangulations by triangles of type (1). Then 70

k = 1, P̂ = P1 and if n = 2 or 3, H 2 (K̂) ֒→ C 0 (K̂). If u ∈ H 2 (Ω),
Theorem 7.1 says that
||u − uh ||1,Ω ≤ C h|u|2,Ω .
We conclude the analysis of convergence in the norm || · ||1,Ω with

the following result.
Theorem 7.2. Let (th ) be a regular family of triangulations of Ω, of

class C 0 . Let s = 0 or 1 and let P1 ⊂ P̂ ⊂ H 1 (K̂). Then (with the
assumption that u ∈ V = H 1 (Ω) or H01 (Ω)) we have
(7.9) lim ||u − uh ||1,Ω = 0

h→0
Proof. Let V = V ∩ W 2,∞ (Ω). Since s ≤ 1, W 2,∞ (·) ֒→ C s (·) and

W 2,∞ (·) ֒→ H 1 (·). The second inclusion follows ‘a fortiori’ from the
first with s = 1. Also, P1 ⊂ P̂ ⊂ H 1 (K̂). Thus we may apply Theorem
6.6 with k = 1, p = ∞, m = 1, q = 2. Then for all v ∈ V ,
1
||v − πK v||1,K ≤ C(meas K) 2 h|v|2,∞,K
smartworlD.asia
1
≤ C(meas K) 2 h|v|2,∞,Ω .
Summing over K, we get

  12
 X 
||v − πh v||1,Ω ≤ C h|v|2,∞,Ω  meas K 
K∈th
= C h|v|2,∞,Ω ,
P
since K∈th meas K = meas Ω, a constant. Thus, for all v ∈ V ,
(7.10) lim ||v − πh v||1,Ω = 0

h→0
71
Notice that V = V. Hence choose v0 ∈ V such that ||u−v0 ||1,Ω ≤ ǫ/2
where ǫ > 0 is any preassigned quantity. Then once v0 is chosen, by
(7.10) choose h0 such that for all h ≤ h0 , ||v0 − πh v0 ||1,Ω ≤ ǫ/2. Now, by
(6.1)
||u − uh ||1,Ω ≤ C||u − πh v0 ||1,Ω

≤ C ||u − v0 ||1,Ω + ||v0 − πh v0 ||1,Ω
≤ Cǫ, for h ≤ h0 .
This gives (7.9) and completes the proof.
We now have, by Theorem 7.1, |u − uh |0,Ω ≤ ||u − uh ||1,Ω = 0(hk ).

We now show, by another argument that |u − uh |0,Ω = 0(hk+1 ), (at least
in some cases) there by giving a more rapid convergence than expected.
This is done by the Aubin-Nitsche argument (also known as the duality
argument). We describe this in an abstract setting.
Let V be a normed space with norm denoted by || · ||. Let H be a
Hilbert space with norm | · | and inner product (·, ·) such that



(i) V ֒→ H, and
(7.11) 

(ii) V = H.
smartworlD.asia
For second-order problems: V = H 1 (Ω) or H01 (Ω) and H = L2 (Ω).
Since H is a Hilbert space, we may identify it with its dual. Further
since V is dense in H, we have that H may be identified with a subspace
of V ′ , the dual of V. For, if g ∈ H, define g̃ ∈ V ′ by g̃(v) = (g, v) · g̃ ∈ V ′
since |g̃(v)| ≤ C|g| ||v||. If g̃(v) = 0 for all v ∈ V, then (g, v) = 0 for all
v ∈ H as well since V = H. Thus g = 0. This proves the identification.
In the sequel we will set g = g̃. 72
Recall that u and uh are the solutions of the problems:
(P) a(u, v) = f (v) for all v ∈ V,

(Ph ) a(uh , vh ) = f (vh ) for all vh ∈ Vh ⊂ V,
and that the assumptions on (P) are as in the Lax-Milgram lemma. Then
we have the following theorem.
Theorem 7.3. Let the spaces H and V satisfy (7.11). Then with our
above mentioned notations,
 ( )
 1 
(7.12) |u − uh | ≤ M||u − uh || sup inf ||ϕ − ϕh ||  ,
g∈H |g| ϕh ∈Vh
where for each g ∈ H, ϕ ∈ V is the corresponding unique solution of the

problem
(P∗ ) a(v, ϕ) = (g, v) for all v ∈ V,
and M the constant occurring in the inequality giving continuity of

a(·, ·).
Remark 7.2. Note that unlike in (P), we solve for the second argument
of a(·, ·) in (P∗ ). This is called the adjoint problem of (P). The existence
and uniqueness of the solution of (P∗ ) are proved in an identical manner.
Note that if a(·, ·) is symmetric, then (P) is self-adjoint in the sense that
(P) = (P∗ ).
Proof. From the elementary theory of Hilbert spaces, we have
smartworlD.asia
|(g, u − uh )|
(7.13) |u − uh | = sup .
g∈H |g|
g,0
For a given g ∈ H,
(7.14) (g, u − uh ) = a(u − uh , ϕ)
73 Also if ϕh ∈ Vh we have,
(7.15) a(u − uh , ϕh ) = 0.
Thus (7.14) and (7.15) give
(7.16) (g, u − uh ) = a(u − uh , ϕ − ϕh ),
which gives us
|(g, u − uh )| ≤ M||u − uh || ||ϕ − ϕh ||,
and hence !
||ϕ − ϕh ||
|u − uh | ≤ M||u − uh | sup .
g∈H |g|
g,0
by (7.13). Since this is true for any ϕh ∈ Vh we may take infimum over
Vh to get (7.12), which completes the proof.
For dimensions ≤ 3 and Lagrange finite elements we now show that

|u − uh |0,Ω = 0(hk+1 ). For this we need one more definition.
Definition 7.2. Let V = H 1 (Ω) or H01 (Ω), H = L2 (Ω). The adjoint

problem is said to be regular if the following hold:
(i) for all g ∈ L2 (Ω), the solution ϕ of the adjoint problem for g
belongs to H 2 (Ω) ∩ V;
(ii) there exists a constant C such that for all g ∈ L2 (Ω)
(7.17) ||ϕ||2,Ω ≤ C|g|0,Ω ,
where ϕ is the solution of the adjoint problem for g.
smartworlD.asia
Theorem 7.4. Let (th ) be a regular family of triangulations on Ω with
reference finite element (K̂, Σ̂, P̂). Let s = 0 and n ≤ 3. Suppose there
exists an integer k ≥ 1 such that u ∈ H k+1 (Ω), Pk ⊂ P̂ ⊂ H 1 (K̂). Assume 74
further that the adjoint problem is regular in the sense of Definition 7.2.
Then there exists a constant C independent of h such that
(7.18) |u − uh |0,Ω ≤ C hk+1 |u|k+1,Ω .
Proof. Since n ≤ 3, H 2 (·) ֒→ C 0 (·). Also, H 2 (·) ֒→ H 1 (·) and P1 ⊂ P̂ ⊂

H 1 (Ĥ). Thus for ϕ ∈ H 2 (Ω), by Theorem 7.1,
||ϕ − πh ϕ||1,Ω ≤ C h|ϕ|2,Ω .
Hence
(7.19) inf ||ϕ − ϕh ||1,Ω ≤ C h|ϕ|2,Ω .

ϕh ∈Vh
By (7.12) and (7.19).

!
1
|u − uh |0,Ω ≤ M||u − uh ||1,Ω sup Ch|ϕ|2,Ω .
g∈L2 (Ω) |g|0,Ω
By the regularity of (P∗ ),

|ϕ|2,Ω ||ϕ||2,Ω
(7.20) ≤ ≤ constant.
|g|0,Ω |g|0,Ω
Thus, |u − uh |0,Ω ≤ C h||u − uh ||1,Ω
≤ C h(hk |u|k+1,Ω ) (by theorem 7.1).
This gives (7.18) and completes the proof.
We finally give an estimate for the error in the L∞ -norm.

Theorem 7.5. Let (th ) be a regular family of triangulations on Ω ⊂ Rn ,
where n ≤ 3. Assume further that for all th and K ∈ th .
hK
(7.21) 0<τ≤ ≤, f rm[o]−−, τ being a constant.
h
smartworlD.asia
75 Let u ∈ H 2 (Ω) and P1 ⊂ P̂ ⊂ H 1 (K̂) ∩ L∞ (K̂). If (P∗ ) is regular,
then there exists a constant C independent of h such that



|u − uh |0,∞,Ω ≤ C h|u|2,Ω ; if n = 2
(7.22) 
 √
|u − uh |0,∞,Ω ≤ C h|u|2,Ω if n = 3.
Proof. Assume n = 2. Now
(7.23) |u − uh |0,∞,Ω ≤ |u − πh u|0,∞,Ω + |πh u − uh |0,∞,Ω .
Note that since (uh − πh u) ∈ Vh , we may apply Exercise 7.1 to get

C
(7.24) |uh − πh u|0,∞,Ω ≤ |uh − πh u|0,Ω .
h
Thus,
C
|uh − πh u|0,∞,Ω ≤ |uh − u|0,Ω + |u − πh u|0,Ω
h
Ch i
≤ C1 h2 |u|2,Ω + C2 h2 |u|2,Ω
h
≤ C h|u|2,Ω (by Theorem 7.4 and Theorem 6.6).
Also H 2 (·) ֒→ C 0 (·); H 2 (·) ֒→ L∞ (·) and P1 ⊂ P̂ ⊂ L∞ (K̂). Thus,

by Theorem 6.6 with k = 1, p = 2, m = 0, q = ∞,
1
|u − πK u|0,∞,K ≤ C(meas K)− 2 h2 |u|2,K .
Since n = 2,
C 2 Cτ2 2
meas K ≥ Cρ2K ≥ h ≥ h
σ2 K σ2
1
by (7.1), so that (meas K)− 2 ≤ C h−1 and therefore,
|u − πK u|0,∞,K ≤ C h|u|2,K .
Hence we obtain (7.22) for n = 2 since
|u − πh u|0,∞,Ω = max |u − πK u|0,∞,K ≤ C h|u|2,Ω .

K∈th
smartworlD.asia
76
For n = 3, the only variation in the proof occurs in the fact that
C
|uh − πh u|0,∞,Ω ≤ |uh − πh u|0,Ω
h3/2
as in Exercise 7.1 and that now
C Cτ3 3
meas K ≥ Cρ3K ≥ · h3
K ≥ ·h .
σ3 σ3
References: One may refer to Ciarlet and Raviart [6] for 0(h) conver-
gence in the norm | · |0,∞,Ω for any n. See also Bramble and Thomée
[1].
smartworlD.asia
Chapter 8
Numerical Integration
LET US START with a specific problem. Let Ω be a polygonal domain 77

in Rn . Consider the problem
 n

 P ∂ ∂u


− ∂xi ai j ∂x j = f in Ω,
(8.1) 
 i, j=1


u = 0 on Γ = ∂Ω.
smartworlD.asia
where the (ai j ) and f are functions over Ω which are smooth enough.
Let us further assume that there exists α > 0 such that, for all ξ ∈ Rn ,
X
n X
n
(8.2) ai j ξi ξ j ≥ α ξi2 .
i, j=1 i=1
It has been seen earlier (cf. Remark 2.2) that the above problem (8.1)
is obtained from a problem (P) with a(·, ·) and f (·) being defined by
 R Pn

 ∂u ∂v
a(u, v) = Ω i, j=1 ai j ∂x j ∂xi dx
(8.3) 
 R
 f (v) = f v dx
Ω
for u, v ∈ V = H01 (Ω).

Approximating the solution by the finite element method, i.e. by
constructing a regular family of triangulations (th ) with reference finite
77
78 8. Numerical Integration
element (K̂, Σ̂, P̂), we get the problems (Ph ), i.e., to find uh ∈ Vh such
that
(8.4) a(uh , vh ) = f (vh ), for all vh ∈ Vh .

M for V , then we may write
If we choose a basis {wk }k=1 h
X
M
(8.5) uh = uk wk .
k=1
Thus to solve (Ph ) we have to solve the linear system

X
M
(8.6) a(wk , wm )uk = f (wm ), 1 ≤ m ≤ M.
k=1
78
Notice that
XZ X
n
∂wk ∂wm
a(wk , wm ) = ai j dx
K∈th K i, j=1 ∂x j ∂xi
(8.7) XZ
f (wm ) = f wm dx.
K
smartworlD.asia
K∈th
Thus we have ended up with the computations of integrals over K ∈

th . These are, in general, difficult or impossible to evaluate exactly and
one thus has to resort to numerical methods. We now study briefly how
this may be done.
Let us assume F K (K̂) = K, where, F K ( x̂) = BK x̂+bK , with det BK >
0. There is no loss in generality in the last assumption. Then if ϕ is a
function over K, we have
Z Z
(8.8) ϕ(x)dx = (det BK ) ϕ̂( x̂)d x̂
K K̂
the functions ϕ and ϕ̂ being in the usual correspondence. We then re-

place the expression in the right-hand side by the following:
Z X
L
(8.9) b ∼
ϕ̂( x̂)dx ω̂1 ϕ̂(b̂1 ).
K̂ 1=1
8. Numerical Integration 79
In this section ∼ will denote the right-hand side replacing the ex-
pression in the left hand side in similar relations). In (8.9) the quan-
tities ω̂1 are called the weights and the points b̂1 are called the nodes
of the quadrature scheme. While in general we may assume ω̂1 ∈ R
and b̂1 ∈ Rn , we will restrict ourselves to the most common case where
ω̂1 > 0 and b̂1 ∈ K̂, 1 ≤ l ≤ L.
We may now define the error functional Eˆ by
Z X
L
(8.10) Eˆ (ϕ̂) = ϕ̂( x̂)d x̂ − ω̂1 ϕ̂(b̂1 ).
K̂ 1=1
79
We will be interested in finding spaces of polynomials for which
Eˆ (ϕ̂) = 0, i.e., again we need “polynomial invariance”, an idea already
found in interpolation theory. The above quadrature scheme for K̂ in-
duces one on K as well since if we set



ω1,K = (det BK )ω̂1 ,
(8.11) 

b1,k = F K (b̂1 ),
we then deduce the numerical quadrature scheme
smartworlD.asia
Z X
L
(8.12) ϕ(x)dx ∼ ω1,K ϕ(b1,K ).
K 1=1
We shall therefore define the error functional

Z XL
(8.13) EK (ϕ) = ϕ(x)dx − ω1,K ϕ(b1,K ).
K l=1
Notice that the following relation holds:
(8.14) EK (ϕ) = (det BK )Eˆ (ϕ̂).
Example 8.1. Consider K̂ to be an n-simplex in Rn . Let â be its centroid.

Let Z
ϕ̂( x̂)d x̂ ∼ (meas K̂)ϕ̂(â).
K̂
Exercise 8.1. Show that E (ϕ̂) = 0 for ϕ̂ ∈ P1 in Example 8.1.
Example 8.2. Let K̂ be a triangle in R2 . With the usual notations, set

Z X
1
ϕ̂( x̂)d x̂ ∼ (meas K̂) ϕ̂(âi j ).
K̂ 3 1≤i< j≤3
80 Exercise 8.2. Show that Eˆ (ϕ̂) = 0 for ϕ̂ ∈ P2 in Example 8.2.
Example 8.3. Let K̂ be as in Example 8.2. Let (cf. Fig. 8.1):

Z  
1  X3 X 
ϕ̂( x̂)d x̂ ∼ (meas K̂) 3 ϕ̂(âi ) + 8 ϕ̂(âi j ) + 27ϕ̂(â) .
K̂ 60 i=1 1≤i< j≤3
Exercise 8.3. Show that Eˆ (ϕ̂) = 0 for ϕ̂ ∈ P3 , in Example 8.3.
Let us now review the whole situation. We had the “original” ap-
proximation problem (Ph ): To find uh ∈ Vh such that a(uh , vh ) = f (vh )
for all vh ∈ Vh .
This led to the solution of the linear system (8.6). By virtue of the
quadrature scheme we arrive at a solution of a “modified” approximation
problem (P∗h ): To solve the linear system
X
M
(8.15) ah (wk , wm )u∗k = fh (wm ), 1 ≤ m ≤ M,
k=1
where
 !


 P PL Pn
∂wk ∂wm



ah (wk , wm ) = l=1 ω1,K ai j ∂x j ∂xi (b1,K )
 K∈th i, j=1 !
(8.16) 



 P P L

 fh (wm ) = ω1,K ( f wm )(b1,K ) .
K∈th l=1
While uh was given by (8.5) we now obtain

X
M
(8.17) u∗h = u∗k wk .
k=1
Thus the problem (P∗h ) (not to be confused with any adjoint prob-
lem!) consists in finding u∗h ∈ Vh such that, for all wh ∈ Vh , 81
(8.18) ah (u∗h , wh ) = fh (wh )

where
 !


 P P L Pn
∂vh ∂wh


ah (vh , wh ) = ω1,K ai j ∂x j ∂xi (b1,K )
 K∈th l=1 i, j=1
(8.19) 



 P P L

 fh (vh ) = ω1,K ( f vh )(b1,K ),
K∈th l=1
smartworlD.asia
for vh , wh ∈ Vh .
Remark 8.1. The bilinear form ah (·, ·) : Vh × Vh → R and the linear

form fh : Vh → R are not defined over V in general. For instance if
V = H01 (Ω)(n = 2) in one of the examples, then as they require point
values of the nodes, we see that they are not in general defined over V.
Having obtained the approximate solution u∗h by numerical integra-

tion, we are naturally interested in its efficacy. Thus we require to know
the error ||u−u∗h ||. We now carry out the error analysis, first in an abstract
setting.
Let us maintain our assumptions as in the Lax-Milgram lemma and
consider the problem (P). Then we have problems (P∗h ) to find u∗h ∈ Vh ⊂
V such that for all vh ∈ Vh , ah (u∗h , vh ) = fh (vh ) where fh ∈ Vh′ and ah (·, ·)
is a bilinear form on Vh . Then we would like to answer the following
questions:
(i) What are sufficient conditions such that (P∗h ) have unique solu-
tions?
(ii) Can we find an abstract error estimate for ||u − u∗h ||?
(iii) If ||u − uh || = 0(hk ), i.e., without numerical quadrature, under what

conditions is this order of convergence preserved, i.e. when can
we say ||u − u∗h || = 0(hk )?
82 The assumption of Vh -ellipticity of the bilinear forms ah (·, ·) answers

the first question (by the Lax-Milgram lemma) and we will see in The-
orem 8.2 under which assumptions it is valid. The following theorem
answers the second question.
Theorem 8.1. Let the bilinear forms ah (·, ·) be Vh -elliptic uniformly with
respect to h, i.e., there exists a constant e
α > 0, independent of h, such
that for all h and for all vh ∈ Vh ,
(8.20) α||vh ||2 .

ah (vh , vh ) ≥ e
Then the approximate problems (P∗h ) all have unique solutions u∗h ,
and further we have the estimate:
smartworlD.asia
||u − u∗h || ≤
(8.21)
≤ C inf
vh ∈Vh
(
||u − vh || + sup
wh ∈Vh
|a(vh , wh ) − ah (vh , wh )|
||wh ||
)
+ sup
wh ∈Vh
| f (wh ) − fh (wh )|
||wh ||
!
.
Remark 8.2. If a = ah , f = fh then we get our original estimate (3.1).

Thus (8.21) generalizes our previous result.
Remark 8.3. The terms involving a, ah and f , fh merely mean that if u∗h
is to converge to u, then ah and fh must be “close to” a and f respec-
tively. Their convergence to 0 with h may be viewed as “consistency
conditions” which are so often found in Numerical Analysis.
Proof. The existence and uniqueness of the u∗h are obvious by the Lax-
Milgram lemma applied to the Vh . Since, for all vh ∈ Vh , we have
ah (u∗h , u∗h − vh ) = fh (u∗h − vh ),
a(u, u∗h − vh ) = f (u∗h − vh ),
83 we have the identity
ah (u∗h − vh , u∗h − vh ) = a(u − vh , u∗h − vh ) + {a(vh , u∗h − vh )

(8.22) − ah (vh , u∗h − vh )} + { fh (u∗h − vh ) − f (u∗h − vh )}.
Hence by (8.20) we get
α||u∗h − vh ||2 ≤ M||u − vh || ||u∗h − vh ||

e
+ |a(vh , u∗h − vh ) − ah (vh , u∗h − vh )|
+ | fh (u∗h − vh ) − f (u∗h − vh )|.
Thus,
|a(vh , u∗h − vh ) − ah (vh , u∗h − vh )|

α||u∗h
e − vh || ≤ M||u − vh || +
||u∗h − vh ||
| fh (u∗h − vh ) − f (u∗h − vh )|
+
||u∗h − vh ||
|a(vh , wh ) − ah (vh , wh )|
smartworlD.asia
≤ M||u − vh || + sup
wh ∈Vh ||wh ||
| f (wh ) − fh (wh )|
+ sup
wh ∈Vh ||wh ||
since (u∗h − vh ) ∈ Vh . Hence,
||u − u∗h || ≤ ||u − vh || + ||u∗h − vh ||

M 1 |a(vh , wh ) − ah (vh , wh )|
≤ 1+ ||u − vh || + sup
e
α e
α wh ∈Vh ||wh ||
1 | f (wh ) − fh (wh )|
+ sup .
e
α wh ∈Vh ||wh ||
Varying vh over Vh and taking the infimum, and replacing (1+ M/e α), 84
α) by a larger constant C, we get (8.21), which completes the proof.
(1/e

The following theorem tells us when the uniform Vh -ellipticity as-

sumption of Theorem 8.1 is satisfied in the example we started with.
Theorem 8.2. Let ah (·, ·) and fh (·) be as in (8.19). Assume further that
SL
(i) ω̂1 > 0, 1 ≤ l ≤ L, (ii) P̂ ⊂ Pk , and (iii) {b̂l } contains a Pk′ −1
l=1
unisolvent subset. Then the ah (·, ·) are Vh -elliptic uniformly with respect
to h.
Proof. We must produce an α̃ > 0, free of h, such that ah (vh , vh ) ≥
α̃|vh |21,Ω for all vh ∈ Vh . We have
 
XX L  X n
∂v ∂v 
ah (vh , vh ) = ω1,K   ai j
h h  (b )
 l,K
K∈th l=1 i, j=1
∂x j ∂x i
(8.23)  n !2 
XX L X ∂pK 
≥α ωl,K  (bl.K ) 
K∈t l=1 i=1
∂xi
h
where pK = vh |K . The inequality (8.23) is a result of the ellipticity

e1 >
condition (8.2) on the matrix (ai j ) and the fact that ωl,K > 0, since ω
0 and we assumed without loss in generality that det BK > 0. Now let
smartworlD.asia
p̂K (b
x) = pK (x), where x = BK x̂ + bK . Let BK = (bi j ), so that
X
n
xj = b jl x̂l + bK, j .
l=1
Then
∂ p̂K X ∂pK (x) ∂x j X ∂pK (x)
n n
= = b ji .
∂ x̂i j=1
∂x j ∂ x̂i j=1
∂x j
Thus is
! !
∂ p̂K ( x̂) ∂ p̂K ( x̂) ∂pK (x) ∂pK (x)
D̂ = ,..., and D= ,..., ,
∂ x̂1 ∂ x̂n ∂x1 ∂xn
we have D̂ = DBK . Hence ||D̂||2 ≤ ||D||2 ||BK ||2 . Thus,

Xn !2 Xn !2
∂pK −2 ∂ p̂K
(8.24) (bl,K ) ≥ ||BK || (b̂1 ) .
i=1
∂xi i=1
∂ x̂i
85
Now suppose p̂ ∈ P̂ is such that
X
L Xn !2
∂ p̂
(8.25) ω̂1 (b̂1 ) = 0.
l=1 i=1
∂ x̂i
∂ p̂
Then since ω̂1 > 0, we have (b̂l ) = 0 for all 1 ≤ l ≤ L and 1 ≤ i ≤
∂ x̂i
∂ p̂ ∂ p̂
n. Since P̂ ⊂ Pk′ , we have ∈ Pk′ −1 and hence = 0 by the Pk′ −1 -
∂ x̂i ∂ x̂i
unisolvency. Thus p̂ ∈ P0 , and on the finite dimensional space P̂/P0
(in practice we always have P0 ⊂ P̂) we have a norm defined by the
square-root of the left hand side of (8.25). By the finite dimensionality
this is equivalent to the norm defined by the | · |1,K̂ norm on P. Hence we
have a constant β̂ > 0 such that
X
L Xn !2
∂ p̂
(8.26) ω̂l (b̂l ) ≥ β̂| p̂|2l,K̂ .
l=1 i=1
∂ x̂i
We will apply this to p̂K . We also have
(8.27)
smartworlD.asia
| p̂K |2l,K ≥ C||B−1 −2 −1 2
K || (det B K ) |pK |l,K ,
by Theorem 6.4. Combining the inequalities (8.23), (8.24), (8.26) and

(8.27), we get
X
ah (vh , vh ) ≥ α (det BK )||BK ||−2 β̂C||B−1 −2 −1 2
K || (det B K ) |pK |l,K
K∈th
X
= αβ̂C (||BK || ||B−1 −2 2
K ||) |pK |l,K
K∈th
X
≥ αβ̂Cγ |pK |2l,K
K∈th
= αβ̂Cγ|vh |21,Ω = α̃|vh |2l,Ω
since (||BK || ||B−1

K ||)
−2 ≥ γ by Theorem 6.5. This proves the theorem.
Let us now review our Examples 8.1 through 8.3 to see if the condi- 86
tions of Theorem 8.2 are satisfied.
Let n = 2 and consider Example 8.1. Clearly ω̂ = meas K̂ > 0. Also
P
P̂ = P1 for triangles of type (1). Since K = {p(â)} is P0 -unisolvent, we
have that for triangles of type (1) and the quadrature scheme of Example
8.1 the corresponding ah (·, ·) are Vh -elliptic uniformly with respect to h.
For triangles of type (2), P̂ = P2 . The weights ω̂1 are all > 0 in
Example 8.2. Further we saw in Exercise 5.1 that {âi j }1≤i< j≤3 is P1 -
unisolvent. Hence Theorem 8.2 is valid for this quadrature scheme as
well.
For triangles of type (3), consider the quadrature scheme of Exam-
ple 8.3. We have ω̂1 > 0 and P̂ = P3 . It was seen in Example 4.2 that
the set {ai ; 1 ≤ i ≤ 3} ∪ {ai j ; 1 ≤ i < j ≤ 3} is P2 -unisolvent. Hence
the corresponding bilinear forms ah (·, ·) are Vh -elliptic uniformly with
respect to h.
Exercise 8.4. Let (H, | · |) be a Hilbert space and V a subspace with norm
|| · || such that V ֒→ H and V = H cf. Sec. 7. Then with the usual
notations show that
n1
|u − u∗h | ≤ sup inf (M||u − u∗h || ||ϕ − ϕh || + |a(u∗h , ϕh ) − ah (u∗h , ϕh )|
smartworlD.asia
g∈H |g| ϕh ∈Vh
o
+| f (ϕh ) − fh (ϕh )|)
where ϕ is the solution of adjoint problem for g.
We now turn our attention to the evaluation of the bound for ||u − u∗h ||
given by (8.21). For second-order problems, for which the norm is || ·
||1,Ω , we will take as usual for vh ∈ Vh the element πh u ∈ Vh so that we
now get the bound
"
∗ |a(πh u, wh ) − ah (πh u, wh )|
||u − uh ||1,Ω ≤ C ||u − πh u||1,Ω + sup
wh ∈Vh ||wh ||1,Ω
(8.28) #
| f (wh ) − fh (wh )|
+ sup .
wh ∈Vh ||wh ||1,Ω
87 Let us assume that we may apply Theorem 7.1, so that
(8.29) ||u − πh u||1,Ω ≤ C hk |u|k+1,Ω .
In order to keep the same accuracy, we will therefore try to obtain

estimates of the following form:




|a(πh u, wh ) − ah (πh u, wh )|

 sup ≤ C(u)hk ,

wh ∈Vh ||wh ||1,Ω
(8.30) 
 | f (wh ) − fh (wh )|



wsup
 ≤ C( f )hk ,
h ∈V h
||w h ||1,Ω
and these will in turn be obtained from “local” estimates. (cf. Theo-
rem 8.4 and Exercise 8.5).
As a preliminary step, we need two results which we prove now.
The first of these is a historically important result in the interpolation
theory in Sobolev spaces.
Theorem 8.3 (BRAMBLE-HILBERT LEMMA; cf. Bramble and

Hilbert [27]). Let Ω ⊂ Rn be open with Lipschitz continuous bound-
ary Γ. Let f ∈ W k+1,p (Ω)) which vanishes over Pk . Then there exists a
constant C = C(Ω) such that, for all v ∈ W k+1,p (Ω),
smartworlD.asia
(8.31) | f (v)| ≤ C|| f ||∗k+1,p,Ω |v|k+1,p,Ω .
Proof. For v ∈ W k+1,p (Ω) and all p ∈ Pk , we have
f (v) = f (v + p),
so that
| f (v)| = | f (v + p)| ≤ || f ||∗k+1,p,Ω ||v + p||k+1,p,Ω ,
and thus, 88
| f (v)| ≤ || f ||∗k+1,p,Ω inf ||v + p||k+1,p,Ω

p∈Pk
∗
≤ C|| f ||k+1,p,Ω |v|k+1,p,Ω ,
by Theorem 6.2, which completes the proof.
Lemma 8.1. Let ϕ ∈ W m,q (Ω), w ∈ W m,∞ (Ω). Then ϕw ∈ W m,q (Ω), and
there exists a numerical constant C, independent of ϕ and w such that
X
M
(8.32) |ϕw|m,q,Ω ≤ C |ϕ|m− j,q,Ω |w| j,∞,Ω .
j=0
Proof. The result is an immediate consequence of the Leibniz formula:

For any |α| = m,
X
m X
∂α (ϕw) = Cα,β ∂α−β (ϕ)∂β (w)
j=0 |β|= j
which yields (8.32).
We may now apply Lemma 8.1 and Theorem 8.3 to get the estimates
(8.30). We do this in two stages (Theorem 8.4 and Exercise 8.5) in
which, for the sake of simplicity, we present our results for the special
case PK = P2 .
Theorem 8.4. Let PK = P2 and consider a quadrature scheme such that

for all ϕ̂ ∈ P2 , ξ̂(ϕ̂) = 0. Then there exists a constant C, independent of
smartworlD.asia
K, such that for all ai j ∈ W 2,∞ (K) and for all p, p′ ∈ PK we have
" #
∂p ∂p′ ∂p ∂p′
(8.33) |EK (ai j ) | ≤ C h2K ||ai j ||2,∞,K || ||1,K | |0,K .
∂x j ∂xi ∂x j ∂xi
∂p ∂p′
Proof. Since we have , ∈ P1 , it suffices to find a suitable esti-
∂x j ∂xi
mate for EK (avw), for a ∈ W 2,∞ (K), v, w ∈ P1 . Further, since
(8.34) EK (avw) = (det BK )Eˆ (âv̂ŵ),
89 we will first find an estimate for Eˆ (âv̂ŵ), with â ∈ W 2,∞ (K̂) and v̂,
ŵ ∈ P1 . Let π̂0 ŵ be the orthogonal projection of ŵ onto the subspace P0
in the sense of L2 (K̂). Then we may write
(8.35) Eˆ (âv̂ŵ) = Eˆ (âv̂π̂0 ŵ) + Eˆ (âv̂(ŵ − π̂0 ŵ)).
(i) Estimate for Eˆ (âv̂π̂0 ŵ).
Consider the functional Eˆ : W 2,∞ (K̂) → R defined by

Z X
L
ψ̂ 7→ Eˆ (ψ̂) = ψ̂( x̂)d x̂ = ω̂1 ψ̂(b̂1 ).
K̂ l=1
|Eˆ (ψ̂)| ≤ Ĉ|ψ̂|0,∞,K̂ ≤ Ĉ||ψ̂||2,∞,K̂ . Thus Eˆ is a continuous linear func-

tional on W 2,∞ (K̂). Hence by Theorem 8.3, since Eˆ vanishes on P1 (⊂
P2 )1 , we have a constant Ĉ such that
(8.36) |Eˆ (ψ̂)| ≤ Ĉ|ψ̂|2,∞,K̂ .
Thus
|Eˆ (âv̂π̂0 ŵ)| ≤ Ĉ|âv̂π̂0 ŵ|2,∞,K̂

≤ Ĉ|âv̂|2,∞,K̂ |π̂0 ŵ|0,∞,K̂
since π̂0 ŵ ∈ P0 is a constant function. By Lemma 8.1 (recall that v̂ ∈

P1 ),
smartworlD.asia
h i
|Eˆ (âv̂π̂0 ŵ)| ≤ Ĉ|π̂0 ŵ|0,∞,K̂ |â|1,∞,K̂ |v̂|1,∞,K̂ + |â|2,∞,K̂ |v̂|0,∞,K̂ .
By the equivalence of the L2 and L∞ norms on P0 , and since the

projection has norm less than that of the vector itself in any Hilbert
space we have the chain of inequalities
|π̂0 ŵ|0,∞,K̂ ≤ Ĉ|π̂0 ŵ|0,K̂ ≤ Ĉ|ŵ|0,K̂ .
90
Similarly we may replace |v̂|1,∞,K̂ by |v̂|1,K̂ and |v̂|0,∞,K̂ by |v̂|0,K̂ since
the L2 and L∞ norms are equivalent on P1 . Thus we get
(8.37) |Eˆ (âv̂π̂0 ŵ)| ≤ Ĉ(|â|1,∞,K̂ |v̂|1,K̂ + |â|2,∞,K̂ |v̂|0,K̂ )|ŵ|0,K̂ .

1
In this estimate, we do not use the “full” polynomial invariance of the quadrature
scheme.
(ii) Estimate for Eˆ (âv̂(ŵ − π̂0 ŵ)).
Let ŵ ∈ P1 be fixed and let ϕ̂ ∈ W 2,∞ (K̂). Then
|Eˆ (ϕ̂(ŵ − π̂0 ŵ))| ≤ Ĉ|ϕ̂(ŵ − π̂0 ŵ)|0,∞,K̂

≤ Ĉ|ϕ̂|0,∞,K̂ |ŵ − π̂0 ŵ|0,∞,K̂
≤ Ĉ|ŵ − π̂0 ŵ|0,∞,K̂ ||ϕ̂||2,∞,K̂ .
Thus the functional on W 2,∞ (K̂) defined by ϕ̂ 7→ Eˆ (ϕ̂(ŵ − π̂0 ŵ)) is

continuous, linear with norm ≤ Ĉ|ŵ− π̂0 ŵ|0,∞,K̂ . Since for ϕ̂ ∈ P1 , ϕ̂(ŵ −
π̂0 ŵ) ∈ P2 , we have that the functional vanishes on P1 . By Theorem 8.3,
|Eˆ (ϕ̂(ŵ − π̂0 ŵ))| ≤ Ĉ|ŵ − π̂0 ŵ|0,∞,K̂ |ϕ̂|2,∞,K̂ .
Set ϕ̂ = âv̂. Now,
|âv̂|2,∞,K̂ ≤ Ĉ(|â|2,∞,K̂ |v̂|0,∞,K̂ + |â|1,∞,K̂ |v̂|1,∞,K̂ ).
Again we may use the equivalence between the L∞ -norms of v̂ and

ŵ − π̂0 ŵ and the L2 -norms of the same functions as in (i) since they be-
smartworlD.asia
long to the finite dimensional space P1 . Also, by the triangle inequality,
|ŵ − π̂0 ŵ|0,K̂ ≤ Ĉ|ŵ|0,K̂ .
Thus we get
(8.38) |Eˆ (âv̂(ŵ − π̂0 ŵ))| ≤ Ĉ(|â|2,∞,K̂ |v̂|0,K̂ + |â|1,∞,K̂ |v̂|1,K̂ )|ŵ|0,K̂ .
91
(iii) We can now complete the proof. Recall that EK (avw) = (det BK )
Eˆ (âv̂ŵ). Also,
|â|m,∞,K̂ ≤ C hm
K |a|m,∞,K
1
−2
(8.39) |v̂|2−m,K̂ ≤ C h2−m
K (det B K ) |v|2−m,K .
1
|ŵ|0,K̂ ≤ C(det BK )− 2 |w|0,K ,
by Theorems 6.4 and 6.5. Combining (8.37), (8.38) and (8.39),

we get
|EK (avw)| ≤ C h2K (|a|1,∞,K |v|1,K + |a|2,∞,K |v|0,K )|w|0,K

≤ C h2K ||a||2,∞,K ||v||1,K |w|0,K .
∂p ∂p′
Setting a = Ai j , v = ,w = we obtain (8.33), thus completing
∂x j ∂xi
the proof.
We leave the second stage as an exercise:
Exercise 8.5. Let PK = P2 and let the quadrature scheme be such that
Eˆ (ϕ̂) = 0 for all ϕ̂ ∈ P2 . Then show that for q such that W 2,q (K) ֒→
C 0 (K), there exists C independent of K such that for all f ∈ W 2,q (K)
and all p ∈ PK ,
1 1
|EK ( f p)| ≤ C h2K (det BK ) 2 − q || f ||2,q,K ||p||1,K .
[Hint: If π̂1 is the orthogonal projection to P1 in the L2 -sense then write
Eˆ ( fˆp̂) = E ( fˆπ̂1 p̂) + Eˆ ( fˆ( p̂ − π̂1 p̂))].
smartworlD.asia
Remark 8.4. The inclusion W 2,q (K) ֒→ C 0 (K) is true if, for instance,
2 − nq > 0, by the Sobolev imbedding theorem.
We now come to the final stage in the estimation of ||u − u∗h ||.
Theorem 8.5. Let (th ) be a regular family of triangulations on Ω by 92
n-simplices of type (2). Let us assume that the Vh -ellipticity is uniform
with respect to h. Let Eˆ (ϕ̂) = 0 for all ϕ̂ ∈ P2 . Then if u ∈ H 3 (Ω) ֒→
C 0 (Ω)(n ≤ 5), ai j ∈ W 2,∞ (Ω) and f ∈ W 2,q (Ω) for some q ≥ 2, we have
the estimate
(8.40) ||u − u∗h ||1,Ω ≤ C h2 [||u||3,Ω + || f ||2,q,Ω ].
Proof. We estimate the various quantities in (8.28). We have:
|a(πh u, wh ) − ah (πh u, wh )| ≤
XX
n
∂(πh u|K) ∂(wh |K)
≤ |EK (ai j |
K∈th i, j=1
∂x j ∂xi
XX n
∂(πh u|K ) ∂(wh |K)
≤ C h2K ||ai j ||2,∞,K || ||1,K | |0,K
K∈th i, j=1
∂x j ∂xi
  12   12
X
n  X ∂(πh u|K)   X ∂(wh |K) 
≤h 2
||ai j ||2,∞,Ω  || ||1,K 
2  || ||0,K 
2
i, j=1 K∈th
∂x j K∈th
∂xi
(since hK ≤ h, and we may apply the Cauchy-Schwarz inequality)
≤ C h2 ||πh u||2,Ω ||wh ||1,Ω
Now,
||πh u||2,Ω ≤ ||u||2,Ω + ||u − πh u||2,Ω ≤ C||u||2,Ω ,
using Theorem 6.3 with P1 ⊂ PK = P2 . Therefore, for all wh ∈ Vh , we

have
|a(πh u, wh ) − ah (πh u, wh )|
(8.41) ≤ Ch2 ||u||2,Ω .
smartworlD.asia
||wh ||1,Ω
Similarly, we have
X
| f (wh ) − fh (wh )| ≤ |EK ( f wh |K )|
K∈th
X 1 1
≤ C h2K (meas K) 2 − q || f ||2,q,K ||wh ||1,K .
K∈th
93
1 1
Since q ≥ 2, 2 − ≥ 0 and by the general Hölder’s inequality,
q
X 1 1
(meas K) 2 − q || f ||2,q,K ||wh ||1,K
K∈th
  12 − 1q   1q   12
 X   X   X 
2 
≤  meas K     
q
 || f ||2,q,K    ||w h ||1,K  
K∈th K∈th K∈th
= C|| f ||2,q,Ω ||wh ||1,Ω .
Hence we get, fot all wh ∈ Vh ,
| f (wh ) − fh (wh )|
(8.42) ≤ Ch2 || f ||2,q,Ω .
||wh ||1,Ω
Combining (8.28), (8.29), (8.41) and (8.42) we get (8.40), thus com-
pleting the proof.
Remark 8.5. The condition n ≤ 5 (needed for the continuous inclusion

H 3 (Ω) ֒→ C 0 (Ω)) was already necessary for the definition of πh u.
References: For a survey on numerical quadrature in general one may

refer to Haber’s survey article [13]. For application to the finite element
method, the Sec. 4.2 of the book by Strang and Fix [22] or Chapter 2 of
the forthcoming book of Ciarlet and Raviart [5].
smartworlD.asia
smartworlD.asia
Chapter 9
The Obstacle Problem
In Sec. 2 we cited the Obstacle Problem as an example of a non-linear 94

abstract problem of Sec. 1. Let us recall a few facts about this to start
with.
Consider an elastic membrane (cf. Fig. 9.1) stretched over an open
set Ω ⊂ R2 and fixed along the boundary Γ which is assumed to be
Lipschitz continuous. Let a force of density Fdx act on the membrane.
Let us assume the existence of an obstacle given by χ(x), for x ∈ Ω.
smartworlD.asia
Then vertical displacement given by u is the solution of the abstract
problem where
 R P



2 ∂u ∂v

a(u, v) = Ω dx
(9.1) 
 R i=1 ∂xi ∂xi


 f (v) =
Ω
f v dx, f ∈ L2 (Ω)
for u, v ∈ V = H01 (Ω), where f = F/t, t being the tension. The subset K
is given by
K = {v ∈ H01 (Ω); v ≥ χ a.e. inΩ}.
If v1 , v2 are in K and vi < χ in Ai with meas Ai = 0 for i = 1, 2, then
λv1 + (1 − λ)v2 ≥ χ on (A1 ∪ A2 )c i.e. the complement of A1 ∪ A2 and
meas (A1 ∪ A2 ) = 0. Thus K is convex. If v ∈ K, let vn ∈ K such that
vn → v in H01 (Ω). Let vn ≥ χ in Acn , meas An = 0. Then all the vn are
≥ χ on (∪n An )c and meas (∪n An ) = 0. Hence v ≥ χ a.e. as well. Thus
95
96 9. The Obstacle Problem
v ∈ K and K is closed as well. We have the regularity assumption that

χ ∈ H 2 (Ω). Of course it is
Figure 9.1:
95
The solution u satisfies
(9.2) J(u) = min J(v),

v∈K
where J(v) = 21 a(v, v) − f (v) and is also characterized by the variational

inequalities (cf. Theorem 1.1):
smartworlD.asia
(9.3) a(u, v − u) ≥ f (v − u), for all v ∈ K.
We proposed as a problem to show that this problem is interpreted

as the following classical problem (assuming u ∈ H01 (Ω) ∩ H 2 (Ω)).



 u ≥ χ in Ω,



(9.4) 
 −∆u = f where u > χ,



u = 0 on Γ.
We have a few regularity results which are listed below:

(i) If Ω is convex and Γ is a C 2 -boundary then u ∈ H01 (Ω) ∩ H 2 (Ω).
(ii) If f = 0 and Ω a convex polygon then also u ∈ H01 (Ω) ∩ H 2 (Ω).
(iii) The norm ||u||2,Ω is bounded above by a function of || f ||0,Ω and
96 ||χ||2,Ω in cases (i) and (ii).
9. The Obstacle Problem 97
Our aim in this section is to use the finite element method to approx-
imate this problem and obtain error estimates. We list our assumptions
now:
Let Ω be a convex polygon, f ∈ L2 (Ω), χ ∈ H 2 (Ω) and let u ∈
H (Ω) ∩ H01 (Ω).
2
Remark 9.1. We cannot assume any more smoothness on u other than

H 2 (Ω). For instance in the 1-dimensional case if f = 0, and even if
the function is very smooth the points of contact of u with χ will have
discontinuous second derivatives in general; cf. Fig. 9.2.
smartworlD.asia
Figure 9.2:
With the above assumptions we proceed to the approximate prob-

lems, first in the abstract setting, as usual.
We have the problems (Ph ) associated with the subspaces Vh ⊂ V =
1
H0 (Ω). We now choose closed convex subsets Kh ⊂ Vh . One has to bear
in mind that, in general, Kh 1 K (we will see that this is the case in our
approach, sub-sequently).
We find uh ∈ Kh such that for all vh ∈ Kh ,
(9.5) a(uh , vh − uh ) ≥ f (vh − uh ).
The existence and uniqueness of the uh follow from Theorem 1.1.

Let H be a Hilbert space with norm | · | and inner-product (·, ·). Let 97
(V, || · ||) be a subspace such that V ֒→ H, V = H. Then, as usual, if we
identify H ′ and H, then H will be identified with a subspace of V ′ . (We
will take V = H01 (Ω) and H = L2 (Ω)). Also as in Sec. 1 (cf. proof of
Theorem 1.2), for all u, v ∈ V we have
(9.6) a(u, v) = (Au)(v),
where A : V → V ′ is a linear map. We now pass on to an abstract error

bound.
Theorem 9.1 (FALK). Assume that f ∈ H, Au ∈ H. Then there exists a

constant C, independent of Vh and Kh , such that
" # 21
2
(9.7) ||u − uh || ≤ C inf (||u − vh || + |u − vh |) + inf |uh − v| .
vh ∈Kh v∈K
(Note: The condition Au ∈ H = L2 (Ω) is satisfied if u ∈ H 2 (Ω) since

Au = −∆u ∈ L2 (Ω).)
Proof. Let α stand for the Vh -ellipticity constant. Then
α||u − uh ||2 ≤ a(u − uh , u − uh )

(9.8) = a(u, u) + a(uh , uh ) − a(u, uh ) − a(uh , u).
smartworlD.asia
For any v ∈ K and any vh ∈ Kh , by (9.3) and (9.5), we have



a(u, u) ≤ a(u, v) + f (u − v),
(9.9) 

a(uh , uh ) ≤ a(uh , vh ) + f (uh − vh ).
Substituting in (9.8) we get
α||u − uh ||2 ≤ a(u, v) + f (u − v) + a(uh , vh )

+ f (uh − vn ) − a(u, uh ) − a(uh , u)
= a(u, v − uh ) − f (v − uh ) + a(u, vh − u) − f (vh − u)
+ a(uh − u, vh − u)
= ( f − Au, u − vh ) + ( f − Au, uh − v) + a(uh − u, vh − u)
≤ | f − Au| |u − vh | + | f − Au| |uh − v| + M||uh − u|| ||vh − u||.
98
r r
α M
Notice that since ( ||u − uh || − ||u − vh ||)2 ≥ 0, we have
M α
1α M
||u − uh || ||u − vh || ≤ ||u − uh ||2 + ||u − vh ||2 ,
2 M α
and hence
α M2
α||u − uh ||2 ≤ C[|u − vh | + |uh − v|] + ||u − uh ||2 + ||u − vh ||2
2 2α
Or,
||u − uh ||2 ≤ C[(|u − vh | + ||u − vh ||2 ) + |uh − v|].
Varying vh ∈ Kh and v ∈ K and extracting the square root after
taking the infima we get (9.7). This completes the proof.
Remark 9.2. If we have a linear problem then f = Au gives the solution

and we get the original bound (3.1).
Remark 9.3. From (9.8) we see that this estimate holds even if a(·, ·) is
not symmetric.
smartworlD.asia
We now apply this to the specific membrane problem. Maintaining
our assumptions on Ω, let th be a triangulation by triangles of type (1),
and let Vh be the corresponding subspace of V = H01 (Ω).
Remark 9.4. It is of no practical use if we go to more sophisticated finite

elements, unlike the linear problem. Since u ∈ H 2 (Ω) is the maximum
smoothness, we may atmost use our abstract estimate theorems only on
the spaces P1 .
One may be tempted to try for Kh those vh which are ≥ χ a.e. in Ω.

However this is not of value from numerical and computational points
of view for we do not easily know where exactly our piecewise linear 99
solution functions would touch χ. We set instead
(9.10) Kh = {vh ∈ Vh ; At all nodes b of th , vh (b) ≥ χ(b)}.
Remark 9.5.
Figure 9.3:
As seen in Fig. 9.3, though for nodes b, vh (b) ≥ χ(b), it does not
guarantee that vh ≥ χ a.e. Thus we see that Kh 1 K. Now the relation
(9.10) is very easy to implement using the computer.
We now have our main result on the error bound.

Theorem 9.2 (FALK). There exists a constant C depending on || f ||0,Ω
and ||χ||2,Ω such that for a regular family of triangulations (th ) as above
we have
smartworlD.asia
(9.11) ||u − uh ||1,Ω ≤ C h.
Remark 9.6. The order of convergence is therefore the same as that for
the linear problems when we use piecewise linear approximations.
Proof. By Theorem 9.1,

" # 21
2
||u − uh ||1,Ω ≤ C inf ||u − vh ||1,Ω + |u − vh |0,Ω + inf |uh − v|0,Ω .
vh ∈Kh v∈K
100 (i) We first estimate the infimum over Kh . Note that if vh = πh u, then
vh ∈ Vh . Also for all nodes b, vh (b) = πh u(b) = u(b) ≥ χ(b). Thus
vh ∈ Kh as well. Thus,

inf ||u − vh ||21,Ω + |u − vh |0,Ω ≤ ||u − πh u||21,Ω + |u − πh u|0,Ω
vh ∈Kh

≤ C h2 |u|22,Ω + |u|2,Ω .

(ii) For the infimum over K, consider v1 = max(uh , χ). Clearly v1 ≥ χ

and v belongs to H 1 (Ω) because uh and χ also belong to H 1 (Ω)
(this is a non-trivial result which we assume here). Hence v1 ∈ K,
and thus,
inf |uh − v|0,Ω ≤ |uh − v1 |0,Ω .
v∈K
We have
Z
|uh − v1 |20,Ω = |uh − χ|2 dx, where Λh = {x; χ(x) ≥ uh (x)}.
Λh
If πh χ is the Vh -interpolate of χ, then for all nodes b, uh (b) ≥ χ(b) =

πh χ(b). Since both uh and πh χ are piecewise linear, we may now assert
that uh ≥ πh χ everywhere. Thus uh − πh χ ≥ 0 on Ω. Thus for all x ∈ Λh ,
we have
0 < |(χ − uh )(x)| = (χ − uh )(x) ≤ (χ − πh χ)(x)

≤ |(χ − πh χ)(x)|
and for x ∈ Ω − Λh , (χ − uh )(x) = 0, so that
smartworlD.asia
  21
Z 
 
|uh − v1 |0,Ω ≤  |χ − πh χ|2 dx = |χ − πh χ|=0,Ω ≤ C h2 |χ|2,Ω .
 
Ω
Hence
1
||u − uh ||1,Ω ≤ C(h2 ) 2 = Ch,
where C depends on |χ|2,Ω and |u|2,Ω . However the regularity result (iii) 101
helps us to bound |u|2,Ω above by a constant C depending on | f |0,Ω and
||χ||2,Ω which completes the proof of the theorem.
References: Two important references are Falk [11] and [12]. For
regularity results refer Brezis and Stampacchia [3] and Lewy and Stam-
pacchia [16].
Another references is Mosco and Strang [19].

smartworlD.asia

Chapter 10
Conforming Finite Element

Method for the Plate
Problem
In Sec. 2 (cf. Example 2.4), as an example of fourth-order problem, we 102

described the plate problem. In abstract terms it is to find the solution of
smartworlD.asia
a(u, v) = f (v), for all v ∈ V,
where
(10.2)



 K = V = H02 (Ω), Ω ⊂ R2 ,

 ( )!

 R 2
∂2 v ∂2 u ∂2 v ∂2 u ∂2 v
a(u, v) = ∆u · ∆v + (1 − σ) 2 ∂ u
 − − dx,


 ∂x1 ∂x2 ∂x1 ∂x2 ∂x21 ∂x22 ∂x22 ∂x22


 RΩ

 2
 f (v) = f v dx, f ∈ L (Ω).
Ω
The problem was interpreted as the classical boundary value prob-

lem


 2
∆ u = f in Ω,

(10.3) 
 ∂u

u = = 0 on Γ = ∂Ω.
∂ν
103

104 10. Conforming Finite Element Method for the Plate Problem
Remark 10.1. It was commented in Sec. 2 that the second term of the
integrand in the definition of a(·, ·) does not contribute to the differential
equation. Our method here will be equally applicable to both the cases.
viz. with the second term present (the plate problem) or with that term
absent (as it happens in Hydro-dynamics). In our next section, on non-
conforming methods, we will see that the second term is essential in
order that we may apply that method.
We assume that Ω is a polygonal domain in R2 . We saw in Sec. 3

that for fourth-order problems we need the inclusion Vh ⊂ C 1 (Ω). (cf.
Exercise 3.1). Thus we need to use finite elements of class C 1 , such as
the Argyris triangle, the Bogner-Fox-Schmidt rectangle and so on (cf.
Sec. 4).
103 When such finite elements can be imbedded in an affine family, then
we have the approximation theory, for regular families of triangula-
tions, available to us. We show that this is the case for the Bogner-
Fog-Schmidt rectangle. However for the Argyris triangle or for the 18-
degree-of-freedom triangle such an imbedding is not possible and we
have to modify the usual argument to obtain error estimates. The “min-
smartworlD.asia
imal assumptions” for 0(h) convergence in the || · ||2,Ω norm are that
P2 ⊂ P and that u ∈ H 3 (Ω) ∩ H02 (Ω). We have that if Ω is a convex
polygon and if f ∈ L2 (Ω), then u ∈ H 3 (Ω) ∩ H02 (Ω). This result is due
to Knodratév.
We will go through the various examples of triangulations of class

C 1 and study convergence in these cases.
Example 10.1. The Bogner-Fog-Schmidt rectangle (cf. Exercise 4.9).

Let PK = Q3 (dim PK = 16). We then have (cf. Fig. 10.1):
X ( )
∂p ∂p ∂2 p
(10.4) = p(ai ), (ai ), (ai ), (ai ); 1 ≤ i ≤ 4 .
K
∂x1 ∂x2 ∂x1 ∂x2

10. Conforming Finite Element Method for the Plate Problem 105
Figure 10.1:
Equivalently, one may also use the PK -unisolvent set

X
′

= p(ai ), Dp(ai )(ai+1 − ai ), Dp(ai )(ai−1 − ai ),
(10.5) K

D2 p(ai )(ai+1 − ai , ai−1 − ai ); 1 ≤ i ≤ 4
(all indices being read modulo 4). 104

Recall that for an affine family of finite elements, the degrees of
freedom p(a0i ), Dp(a1i )(ξik1 ), (D2 p(a2i )(ξik2 , ξil2 ) are such that (cf. Sec. 5):
smartworlD.asia


a0i = F(â0i ), . . . , a2i = F(â2i ),

(10.6) 

ξ 1 = BK ξ̂ 1 , . . . , ξ 2 = BK ξ̂ 2
i,k i,k i,1 i,1
for then πdK v = π̂v̂ which is essentially what we need for the abstract
error analysis.
P
In ′K note that
(10.7) ai+1 − ai = F(âi+1 ) − F(âi ) = BK (âi+1 − âi ),
and so on. Thus it is clear that this rectangle can be imbedded in an

affine family of finite elements. Now PK ⊂ P̂ ⊂ Q3 for k = 3. By our
abstract error analysis, we therefore have
(10.8) ||u − uh ||2,Ω ≤ Ch2 |u|4,Ω .
assuming sufficient smoothness on u as usual.

A word about the boundary conditions. As in Exercise 3.1, we get

∂v
that Vh ⊂ H02 (Ω) if v = = 0 on Γ, for v ∈ Vh . Thus in choosing our
∂ν
basis functions we must assure ourselves that this condition is satisfied.
This in turn depends on the values at the boundary nodes. Let b and c be
two nodes on Γ such that the line joining them is parallel to (say) the x1 -
axis. Since we need v = 0 on this line, and since v will be a polynomial
∂v
in x1 of degree ≤ 3 on this line we must have v(b) = v(c) = 0, (b) =
∂x1
∂v ∂v ∂v ∂v
(c) = 0. Also since we need = 0 on this line and since =
∂x1 ∂ν ∂ν ∂x2
∂v ∂v
is a polynomial in x1 of degree ≤ 3, we need to set (b) = (c) = 0
∂x2 ∂x2
2
∂ v 2
∂ v
105 and (b) = (c) = 0. Thus the degrees of freedom on all
∂x1 ∂x2 ∂x, ∂x2
boundary nodes must be zero. The only “free” or “unknown” parameters
are the degrees of freedom at the interior nodes. This takes care of the
boundary conditions.
Let us now turn to the Argyris triangle (cf. Example 4.7).
smartworlD.asia
Figure 10.2:
P
We recall that PK = P5 , dim PK = 21, and K is given by (cf.
Fig. 10.2)
(10.9)  
X  
 ∂p ∂3 p ∂p 


= p(ai ), (ai ), . . . , 2 (ai ), 1 ≤ i ≤ 3; (ai j ), 1 ≤ i < j ≤ 3
 .
 ∂x1 ∂x ∂ν 
K 2

We may replace the first and second derivative values at the vertices
by Dp(ai )(ai+1 − ai ), Dp(ai )(ai−1 − ai ), D2 p(ai )(ai+1 − ai , ai+1 − ai ),
D2 p(ai )(ai+1 − ai , ai−1 − ai ), D2 p(ai )(ai−1 − ai , ai−1 − ai ) in order to get
degrees of freedom for which the relations of the type (10.6) may be
∂p
satisfied. However, one cannot replace the normal derivatives (ai j ),
∂ν
1 ≤ i < j ≤ 3, by such quantities since affine transformations do not
preserve orthogonality.
In order to estimate the errors we describe an “intermediary” finite
element:
Example 10.2. The Hermite Triangle of Type (5).

Let PK = P5 (dim PK = 21). Define
X n
= p(ai ), Dp(ai )(ai+1 − ai ), Dp(ai )(ai−1 − ai ),
K
D2 p(ai )(ai+1 − ai , ai+1 − ai ), D2 p(ai )(ai+1 − ai , ai−1 − ai ),
D2 p(ai )(ai−1 − ai , ai−1 − ai ), 1 ≤ i ≤ 3;
o
Dp(ai j )(ak − ai j ), 1 ≤ i < j ≤ 3, k , 1, , j .
106
smartworlD.asia
That is to say, the only change compared to the Argyris triangle is
that we have replaced the normal derivatives at ai j by the derivatives
along the line joining ai j to ak , the opposite vertex.
Symbolically we can represent such a triangle as in Fig. 10.3.
Figure 10.3:
This element can be put in an affine family as is readily seen. If ΛK

is the associated interpolation operator, our error analysis yields

h6K
(10.10) |v − ΛK v|m,K ≤ C m |v|6,K , 0 ≤ m ≤ 6,
ρK
for v ∈ H 6 (K).
Remark 10.2. Though the Hermite triangle of type (5) yields an affine
family, one cannot use it since Vh ⊂ C 0 (Ω) but, in general, Vh 1 C 1 (Ω)
as is necessary for fourth-order problems. This is so because the adja-
cent triangles will not patch up, in general, in their derivatives along the
medians; cf. 10.4.
smartworlD.asia
Figure 10.4:
107
Again we show how to take care of the boundary conditions in the
∂v
Argyris triangle. We need again v = = 0 on Γ. Let us have two nodes
′
∂ν
b, b , the vertices of a triangle lying on Γ with mid-point c. On this line
v will be a polynomial of degree ≤ 5 in τ, an abscissa along this line.
∂v
will be a polynomial in τ of degree ≤ 4 on this line. Hence for v = 0
∂ν
∂v ∂v ′ ∂2 v
on Γ we need to set, v(b) = v(b′ ) = 0, (b) = (b ) = 0, 2 (b) =
∂τ ∂τ ∂τ
∂2 v ′ ∂v ∂v ∂v ′ ∂v
(b ) = 0. For = 0 on Γ we set, (b) = (b ) = (c) = 0,
∂τ2 ∂ν ∂ν ∂ν ∂ν
∂2 v ∂2 v ′
(b) = (b ) = 0. Thus the only free or unknown parameters
∂τ∂ν ∂τ∂ν
∂2 v
are 2 at vertices on Γ and the degrees of freedom at all interior nodes.
∂ν

We now get an error estimate when we have triangulations of Ar-

gyris triangles. We use our usual terminology more loosely here1 . By a
regular family of triangulations made up of Argyris triangles we mean
hK
that all th consist only of Argyris triangles and that for all K, ≤ σ, a
ρK
constant. We also assume that if h = max hK , then h → 0.
K∈th
Theorem 10.1. For a regular family (th ) of triangulations made up of

Argyris triangles
(10.11) |v − πh v|m,Ω ≤ C h6−m |v|6,Ω , 0 ≤ m ≤ 6.
108
Proof. Let us denote the opposite vertex of ai j (i < j) by ak . Let → −ν

K
be the unit outernormal at ai j and → −τ be the unit vector along the line
K
[ai , a j ], at ai j (cf. Fig. 10.5).
smartworlD.asia
Figure 10.5:
Let πK be the interpolation operator for the Argyris triangle K and

let ΛK be that for the corresponding Hermite triangle of type (5).
1
Because we have to drop the assumption that all the finite elements are affine equiv-
alent to a reference finite element.

Set δ = πK v − ΛK v. Then δ ∈ P5 . Now
∂δ ∂ ∂
(ai j ) = (πK v − ΛK v)(ai j ) = (v − ΛK v)(ai j ).
∂νK ∂νK ∂νK
Also since πK v = ΛK v along any side of K (since the values of

these polynomials of degree 5 as well as those of their first and second
∂δ
derivatives agree at the end-points), we have = 0.
∂τK
Since
∂δ −ν i + ∂δ (a )ha − a ,→
Dδ(ai j )(ak − ai j ) = (ai j )hak − ai j ,→ K ij k
−
i j τ K i,
∂νK ∂τK
∂δ ∂δ
where h·, ·i is the Euclidean inner-product, substituting for and
∂νK ∂τK
at ai j , we get
∂
(10.12) Dδ(ai j )(ak − ai j ) = (v − ΛK v)(ai j )hak − ai j ,→
−ν i
K
∂ν
109 Since δ ∈ P5 , using the unisolvency in the Hermite triangle we may
express δ in terms of its basis functions. Since all degrees of freedom
smartworlD.asia
except those of the type Dδ(ai j )(ak − ai j ) are zero for δ, we have
X ∂
(10.13) δ= (v − ΛK v)(ai j )hak − ai j ,→
−ν ip .
K i jk
1≤i< j≤3
∂ν K
k,i,k, j
Now
(10.14) |hak − ai j ,→
−ν i| ≤ ||a − a || ||→
K k ij
−ν || ≤ h ,
K K
and
∂
| (v − ΛK v)(ai j )| ≤ |v − ΛK v|1,∞,K
∂νK
6
1 h
≤ C(meas K)− 2 K |v|6,K .
ρK

(Theorem 6.6 with m = 1, k = 5, p = 2, q = ∞). Also, meas K ≥ Cρ2K ,

and we have
∂ h6
(10.15) | (v − ΛK v)(ai j )| ≤ C 2K |v|6,K .
∂νK ρK
Finally, by Theorem 6.4 and 6.5

hK
(10.16) |pi jk |m,K ≤ C | p̂i jk |m,K̂ .
ρm
K
Combining (10.14), (10.15) and (10.16), we get

X ∂
|δ|m,K ≤ | (v − ΛK v)(ai j )| |hak − ai j ,→
−ν i| |p |
K i jk m,K
1≤i< j≤3
∂νK
k,i,k, j
h8K
(10.17) ≤C |v|6,K ,
ρm+2
K
and hence,
|v − πK v|m,K ≤ |v − ΛK v|m,K + |δ|m,K
smartworlD.asia
 
h6K  h2K 
≤ C m 1 + 2  |v|6,K (Using (10.10) and (10.17))

ρK ρK
hK
≤ C h6−m |v|6,K (since hK ≤ h, ≤ σ).
ρK
110
This on summing over K gives (10.11), thus completing the proof.

Exercise 10.1. Perform the same analysis for the 18-degree-of freedom
triangle. (cf. Exercise 4.7).
For the interpolation theory of the HCT-triangle (cf. Exercise 4.8),

the normal derivatives are handled as in the present case. However the
arbitrariness of the interior point is an obstacle to be overcome. For a
discussion of this, see Ciarlet [4].

Another finite element, similar in its principle to the HCT-triangle

used in the conforming finite element method for the plate problem is
the Fraeijs de Veubeke and Sander Quadrilateral. See Ciavaldini and
Nédélec [9].
These are all essentially the finite elements used in the “conforming”
methods to approximate the plate problem (we will define such methods
at the beginning of Sec. 11).
smartworlD.asia

Chapter 11
Non-Conforming Methods
for the Plate Problem
WE START WITH a brief classification of finite element methods. The 111

first class of methods are called conforming methods, which we have
described upto now, except when we considered numerical integration.
The second class consists of methods other than conforming. In the
latter class we have the Non-conforming methods included:
smartworlD.asia
Given the abstract problem, the conforming methods deal with the
finding of subspaces Vh ⊂ V and solving the problems
(Ph ) ah (uh , vh ) = fh (vh ), for all vh ∈ Vh ,
where ah = a and fh = f for all h and uh ∈ Vh is the required solution.

When we employ methods other than conforming we commit, in
the terminology of G. Strang, “Variational Crimes”. (See Strang and
Fix [22]). These may occur in the following ways:
(i) When performing numerical integration, we may have ah and fh
different from a and f respectively. However, Vh is a subspace of
V;
(ii) The boundary Γ of Ω may be curved. In this case triangles lying
in the interior will be triangles of straight edges while those meet-
ing the boundary will have curved edges like parabolas. These are
113

114 11. Non-Conforming Methods for the Plate Problem
the so-called “isoparametric” finite elements. Hence if Ωh is the

union of the finite elements of the triangulation th , then, in gen-
eral, Ωh , Ω and consequently Vh 1 V (where Vh is a space of
functions defined over Ωh ), ah , a, fh , f ; for a discussion of
these, see Ciarlet and Raviart [30], [31].
(iii) When employing non-conforming methods (which will be dealt

with subsequently) though Ωh = Ω, fh = f , we will have Vh 1 V
and ah , a.
112 (iv) One may employ any combination of the above three.
Let us return to the plate problem. For a conforming method we

need the inclusion Vh ⊂ H02 (Ω) which essentially results from the in-
clusion Vh ⊂ C 1 (Ω). Because of this necessity, when compared with
smartworlD.asia
second-order problems, we either have the dimension of PK “large” (as
in the case of the Argyris triangle) or that the structure of PK is compli-
cated (as in the HCT-triangle). Also one would like to have just PK = P2
since u is only in H 3 (Ω) in most cases, but this is impossible by the
Ženišek result (cf. Remark 4.3) which stresses that at least polynomials
of degree 5 must be present in PK .
Hence the desire to surmount these difficulties led to the devising of
non-conforming methods, essentially developed by the Engineers.
Since the root of all trouble is the inclusion Vh ⊂ H02 (Ω), we drop
this condition. Thus we start with Vh ⊂ C 0 (Ω) and it is much easier
from the computer programme view point. This of course, works only
for a few finite elements, and we describe one of them.
Example 11.1. The Adini’s rectangle; cf. Fig. 11.1.

11. Non-Conforming Methods for the Plate Problem 115
Figure 11.1:
The element K consists of a rectangle with vertices {ai , 1 ≤ i ≤ 4};

the space PK is given by PK = P3 ⊕ {x1 x32 } ⊕ {x31 x2 }, by which we mean
polynomials of degree ≤ 4 whose only fourth-degree terms are those
involving x1 x32 and x31 x2 . Thus P3 ⊂ PK . We have the set of degrees of
freedom:
X ( )
∂p ∂p
= p(ai ), (ai ), (ai ), 1 ≤ i ≤ 4 .
∂x1 ∂x2
smartworlD.asia
K
113
Of course this element can be used only for plates with sides parallel
to the coordinate axes, such as rectangular plates.
P
Exercise 11.1. Show that in Example 11.1, K is PK -unisolvent and
that Adini’s rectangle is a finite element of class C 0 , and, in general, not
of class C 1 .
Thus we get ‘a priori’ that Vh ⊂ H 1 (Ω). For the boundary condition,

we set all degrees of freedom on the boundary nodes as zero. This gives
us that Vh ⊂ H01 (Ω). Thus the only ‘unknown’ or ‘free’ parameters are
∂vh
the degrees of freedom at the interior nodes. Note that is zero only
∂ν
at the boundary nodes, in general.

In the abstract problem, we have a(·, ·) and f (·) given by

(11.1)
   
 R  2
∂2 v ∂2 u ∂2 v ∂2 u ∂2 v 



 a(u, v) = ∆u∆v + (1 − σ) 2 ∂ u − −  dx,

 ∂x1 ∂x2 ∂x1 ∂x2 ∂x21 ∂x22 ∂x22 ∂x21


 R Ω

 2
 f (v) = f vdx, f ∈ L (Ω).

Ω
The second integral is defined over Vh as well. Thus for the discrete
problem (Ph ) we may set fh = f . However while for u, v ∈ Vh the first
integral is defined over each K ∈ th , we cannot define it over Ω, since
we get Dirac measure-like terms along the boundary. To get over this,
we now define
XZ " ∂2 uh ∂2 vh ∂2 uh ∂2 uh ∂2 vh
!#
ah (uh , vh ) = ∆uh ∆vh + (1 − σ) 2 − − dx
K∈th
∂x1 ∂x2 ∂x1 ∂x2 ∂x21 ∂x22 ∂x21
K
(11.2)X Z " !#
∂2 uh ∂2 vh ∂2 uh ∂2 vh ∂2 uh ∂2 vh
= σ∆uh ∆vh + (1 − σ) + + 2 dx,
K∈t
∂x21 ∂x21 ∂x22 ∂x22 ∂x1 ∂x2 ∂x1 ∂x2
h K
and we have the discrete problem (Ph ): To find uh ∈ Vh such that for all
vh ∈ Vh
(11.3) ah (uh , vh ) = f (vh ).
smartworlD.asia
114
We now prove the existence and uniqueness of the solution uh for
(Ph ). We define on Vh the seminorm
  21
 X 
(11.4) ||vh ||h =  |vh |22,K |22,K  .
K∈th
Notice that this may be defined over V = H02 (Ω) as well and for
v ∈ V, ||v||h = |v|2,Ω . In the same way for u, v ∈ V, ah (u, v) = a(u, v).
We now show that the seminorm || · ||h is indeed a norm on Vh . Let
∂vh
vh ∈ Vh with ||vh ||h = 0. This gives that = constant over any K. But
∂x1
∂vh
given adjacent finite elements the value of at the common vertices
∂x1
∂vh
coincide and hence is constant over Ω. But this is zero on the
∂x1

∂vh ∂vh
boundary nodes. Hence = 0 on Ω. Similarly = 0 on Ω. Since
∂x1 ∂x2
Vh ⊂ C 0 (Ω) and vh = 0 on Γ, the above conditions give that vh = 0 over
Ω. Thus (11.4) defines a norm on Vh .
To show the existence and uniqueness of the solution of (Ph ), we
show that ah (·, ·) is Vh -elliptic. In fact we do more than this. We show
that the ah (·, ·) are Vh -elliptic uniformly with respect to h.
Recall that from physical considerations, 0 < σ < 12 (see Sec. 2).
Now
XZ
a(vh , vh ) = σ(∆vh )2 dx + (1 − σ)||vh ||2h
(11.5) K∈th K
≥ (1 − σ)||vh ||2h .
Remark 11.1. It was mentioned in passing in Sec. 10 that in order to

apply non-conforming methods one needed the second term involving
σ in the integral defining a(·, ·). The uniform Vh -ellipticity could not be
got in the Hydrodynamical case where this term is absent.
We now proceed with the abstract error analysis. 115
smartworlD.asia
Theorem 11.1 (STRANG). Let ah (·, ·) be Vh -elliptic uniformly with re-
spect to h with e
α > 0 so that for all vh ∈ Vh
(11.6) ah (vh , vh ) ≥ e
α||vh ||h .
e such that for all uh , vh ∈ Vh
Let in addition, there exist M
(11.7) e h ||h ||vh ||h .

|a(uh , vh )| ≤ M||u
Assume that ah = a and || · ||h = || · || on V. (These are needed to

extend the definition of ah and || · ||h to V). Then there exists a constant
C, independent of h, such that
( )
| f (wh ) − ah (u, wh )|
(11.8) ||u − uh ||h ≤ C inf ||u − vh ||h + sup
vh ∈Vh wh ∈Vh ||wh ||h
uh being the solution of (Ph ).

Proof. For all vh ∈ Vh we have
(11.9) ||u − uh ||h ≤ ||u − vh ||h + ||uh − vh ||h .
Now for any vh ∈ Vh , f (uh − vh ) = ah (uh , uh − vh ), so that we may

write
α||uh − vh ||2h ≤ ah (uh − vh , uh − vh )

e
= ah (u − vh , uh − vh ) + f (uh − vh ) − ah (u, uh − vh )
e − vh ||h ||uh − vh ||h + | f (uh − vh ) − ah (u, uh − vh )|,
≤ M||u
and thus,
e
M 1 | f (uh − vh ) − ah (u, uh − vh )|
||uh − vh ||h ≤ ||u − vh ||h +
e
α e
α ||uh − vh ||h
e
M 1 | f (wh ) − ah (u, wh )|
≤ ||u − vh ||h + sup .
e
α e
α wh ∈Vh ||wh ||h
116
Substituting in (11.9) and varying vh ∈ Vh and taking the infimum
smartworlD.asia
we get (11.8). This completes the proof.
Remark 11.2. In case the method is conforming, then ah (u, wh ) = a

(u, wh ) = f (wh ) and the second term disappears in (11.8), leaving us
with the original bound (3.1).
We note that for the ah (·, ·) defined for the plate problem by (11.2),
the conditions of Theorem 11.1 are satisfied. The condition (11.6) is
embodied in (11.5). The condition (11.7) follows from the similar (con-
tinuity) condition on a(·, ·) and an application of the Cauchy-Schwarz
inequality.
Exercise 11.2. Let (H) be a Hilbert space with innerproduct (·, ·) and
norm | · |. Let (V, || · ||) be a subspace such that V ֒→ H and V = H. Let
Vh ⊂ H. Define
Eh (u, v) = ( f, v) − ah (u, v), for all u, v ∈ Vh ∪ V.

Then show that

 
 1 
e − uh ||h sup
|u − uh | ≤ M||u inf ||ϕ − ϕh ||h 
g∈H |g| ϕh ∈Vh
" #
1
+ sup inf (Eh (u, ϕ − ϕh ) + Eh (ϕ, u − uh ))
g∈H |g| ϕh ∈Vh
where for all v ∈ V, a(v, ϕ) = g(v).
We now go on with the error analysis and study the order of conver-
gence. We assume that u ∈ H 3 (Ω) ∩ H02 (Ω) which is quite realistic from
the regularity results.
Now, since πh u ∈ Vh , we have that
inf ||u − vh ||h ≤ ||u − πh u||h .

vh ∈Vh
117
Again applying error bounds for each K and then summing over all
K we get
||u − πh u||h ≤ Ch|u|3,Ω .
Thus
(11.10)
smartworlD.asia inf ||u − vh ||h ≤ Ch|u|3,Ω .
vh ∈Vh
Our aim is to get a similar estimate for the second term in (11.8). In
fact we show that
| f (wh ) − ah (u, wh )|
(11.11) sup ≤ Ch||u||3,Ω .
wh ∈Vh ||wh ||h
This entails more work. We define
(11.12) Eh (u, wh ) = f (wh ) − ah (u, wh )

for u ∈ H 3 (Ω) ∩ H02 (Ω), wh ∈ Vh . Since wh ∈ H01 (Ω), there exists a
sequence {wnh } in D(Ω) converging to wh in H01 (Ω). Hence,
Z Z " !#
∂2 u ∂2 wnh ∂2 u ∂2 wnh ∂2 u ∂2 wnh
f wnh dx = ∆u∆wnh + (1 − σ) 2 − 2 − 2 dx
∂x1 ∂x2 ∂x1 ∂x2 ∂x1 ∂x22 ∂x2 ∂x21
Ω Ω


Z
=− (grad ∆u)(grad wnh )dx,
Ω
by Green’s formula. The term involving (1 − σ), by Lemma 2.2, can

be converted to an integral over Γ. All integrals over Γ vanish since
wnh ∈ D(Ω). Since both sides of the above relation are continuous linear
functionals on H01 (Ω), we can pass to the limit to obtain
Z Z
(11.13) f (wh ) = f wh dx = − (grad ∆u)(grad wh )dx
Ω Ω
for all wh ∈ Vh . Now,
XZ " #!
∂2 u ∂2 wh ∂2 u ∂2 wh ∂2 u ∂2 wh
ah (u, wh ) = ∆u∆wh + (1 − σ) 2 − 2 − dx
K∈th
∂x1 ∂x2 ∂x1 ∂x2 ∂x1 ∂x22 ∂x22 ∂x21
K

X  Z
= − grad ∆u grad wh dx

K∈th K

smartworlD.asia
I I ! 
∂wK ∂2 u ∂wh ∂2 u ∂wh 
(11.14) + ∆u h dγ + (1 − σ) − 2 + dγ ,
∂νK ∂τK K∂ν ∂ν K ∂τ K ∂τ K 
∂K ∂K
118 by Green’s formula (2.15) and Lemma 2.2 again. Notice however that
by standard orientation arguments and the continuity of wh over Ω,
XI ∂2 u ∂wh
(11.15) dγ = 0.
K∈th ∂K ∂νK ∂τK ∂τK
Using (11.13), (11.14) and (11.15), we substitute in (11.12) to get
X I  
∂2 u  ∂wh
(11.16) Eh (u, wh ) = 
−∆u + (1 − σ)  dγ.
∂τ 2  ∂ν
K∈t K K
h ∂K

Figure 11.2:
Splitting the boundary into four parts as in Fig. 11.2, we get
X ! !!
∂wh ∂wh
(11.17) Eh (u, wh ) = ∆1,K u, + ∆2,K u, ,
K∈th
∂x1 ∂x2
where for j = 1, 2, we define

! Z   !!
smartworlD.asia
∂wh  ∂2 u  ∂wh ∂wh
∆ j,K u, = −∆u + (1 − σ)  − Λ K dγ
∂x j ∂τ2K ∂x j ∂x j
K ′j
(11.18) Z   !!
 ∂2 u  ∂wh ∂wh
− −∆u + (1 − σ) 2  − ΛK dγ,
′′
∂τK ∂x j ∂x j
Kj
ΛK being the Q1 -interpolation operator associated with the values at the 119
four vertices.
Note that (11.17) is the same as (11.15). This is evident provided
we show that the contribution of the terms involving ΛK is zero. This is
so because! wh = 0 on the boundary Γ and on the common boundaries,
∂wh
ΛK is linear and equal in value for both adjacent finite elements
∂x j
since it agrees at the vertices, but occurs with opposite signs as is obvi-
′ ′′
ous from Fig. 11.3 (K11 = K12 ).

Figure 11.3:
∂wh
We also record that ∈ ∂ j PK where
∂x j K
( )
∂p
(11.19) ∂ j PK = : K → R; p ∈ PK .
∂x j
We now prove a result analogous to the Bramble-Hilbert lemma

which will help us to estimate that ∆ j,K ’s and hence Eh (u, wh ).
Theorem 11.2 (BILINEAR LEMMA). Let Ω ⊂ Rn be open with Lip-

schitz continuous boundary Γ. Let W be a subspace of W l+1,q (Ω) such
smartworlD.asia
that P1 ⊂ W. Let b be a continuous bilinear form over W k+1,p (Ω) × W
such that



b(p, w) = 0 for all p ∈ Pk , w ∈ W


b(v, q) = 0 for all v ∈ W k+1,p (Ω), q ∈ P1 .
120 Then there exists a constant C = C(Ω) such that for all v ∈ W k+1,p
(Ω), w ∈ W,
(11.21) |b(v, w)| ≤ C||b|| |v|k+1,p,Ω |w|l+1,q,Ω .
Proof. For a given w ∈ W, b(·, w) : v 7→ b(v, w) is a continuous lin-

ear form on W k+1,p (Ω) vanishing on Pk . Hence by the Bramble-Hilbert
lemma,
(11.22) |b(v, w)| ≤ C||b(·, w)||∗k+1,p,Ω |v|k+1,q,Ω .

However for all q ∈ P1 , b(v, w) = b(v, w + q) so that
|b(v, w)| ≤ ||b|| ||v||k+1,p,Ω ||w + q||l+1,q,Ω
and hence
|b(v, w)| ≤ ||b|| ||v||k+1,p,Ω inf ||w + q||l+1,q,Ω

q∈P1
≤ C||b|| ||v||k+1,p,Ω |w|l+1,q,Ω , by the theorem 6.2,
so that
||b(·, w)||∗k+1,p,Ω ≤ C||b|| |w|l+1,q,Ω .
and substituting in (11.22), we get (11.21), which completes the proof.

We may now prove the theorem on our error estimate and order of
convergence.
Theorem 11.3. For a regular family (th ) of triangulations made up of
Adini’s rectangles
(11.23) ||u − uh ||h ≤ C h||u||3,Ω .








ϕ = −∆u
smartworlD.asia
Proof. Let us first estimate |∆ j,K | for j = 1, 2. Set
+ (1 − σ)
∂2 u
2
∈ H 1 (K), since u ∈ H 3 (Ω),
 ∂τ K


 ∂wh


v = ∈ ∂1 P K .
∂x1
Define 121
Z Z
(11.24) δ1,K (ϕ, v) = ϕ(v − ΛK v)dγ − ϕ(v − ΛK V)dγ,
K1′ K1′′
for v ∈ ∂1 P, ϕ ∈ H 1 (K). If h2 is the length of K1′ (and K1′′ ) and h1 is that

of K2′ (and K2′′ ) we have by a simple change of variable
(11.25) δ1,K (ϕ, v) = h2 δ1,K̂ (ϕ̂, v̂),

where K̂ is the reference finite element. Since P0 ⊂ Q1 which is pre-

served by ΛK we have that for all v̂ ∈ P0 , ϕ̂ ∈ H 1 (K̂), δ1,K̂ (ϕ̂, v̂) = 0.
Now let ϕ̂ ∈ P0 and v̂ ∈ ∂d1 P. We wish to show that δ1, K̂ (ϕ̂, v̂) = 0:
We may take for K̂, the unit square. Since ϕ̂ ∈ P0 , its value on K̂ is
a constant, say, b0 . Now let v̂ ∈ ∂d
1 P. Then v̂ is of the form
(11.26) v̂ = a0 + a1 x1 + a2 x2 + a3 x21 + a4 x1 x2 + a5 x22 + a6 x21 x2 + a7 x32 .
Taking the values at the four vertices we get
(11.27) ΛK v̂ = a0 + (a1 + a3 )x1 + (a2 + a5 + a7 )x2 + (a4 + a6 )x1 x2 .
Now K1′ is the line x1 = 1 and K1′′ is the line x1 = 0. Thus

v̂ − ΛK v̂ = −(a5 + a7 )x2 + a5 x22 + a7 x32 ,
x1 =0

v̂ − ΛK v̂ = −(a5 + a7 )x2 + a5 x22 + a7 x32 .
x =1 1
Hence,
Z Z1
smartworlD.asia
ϕ̂(v̂ − ΛK v̂)dγ = b0 (−(a5 + a7 )x2 + a5 x22 + a7 x32 )dx2
′ 0
(11.28)K1 Z
= ϕ̂(v̂ − ΛK v̂)dγ.
K1′′
122 Thus δ1,K̂ (ϕ̂, v̂) = 0 for ϕ̂ ∈ P0 , v̂ ∈ ∂d

1 P.
Note further that the bilinear form δ1,K̂ is continuous, for
|δ1,K̂ (ϕ̂, v̂)| ≤ C||ϕ̂||L2 (∂K̂) ||v̂||L2 (∂K̂) .

≤ C||ϕ̂||1,K̂ ||v̂||1,K̂ , by the Trace theorem (cf. Th. 2.3).
Thus we may apply the bilinear lemma to the bilinear form δ1,K̂ with
l = k = 0 to get
(11.29) |δ1,K̂ (ϕ̂, v̂)| ≤ C|ϕ̂|1,K̂ |v̂|1,K̂ .

We also have the relations


 1
|ϕ̂|1,K̂ ≤ C||BK || | det BK |− 2 |ϕ|1,K ,

(11.30) 

|v̂| ≤ C||BK || | det BK |− 12 |v|1,K .
1, K̂
Now ||BK || ≤ C hK and | det BK | = meas K/ meas K̂ ≥ Cρ2K , and thus

1 C hK
||BK || | det BK |− 2 ≤ ≤ C. Also h2 ≤ hK , so that
ρK
(11.31) |∆1,K (ϕ, v)| ≤ C hK ||u||3,K ||wh ||2,K
where 


 ∂2 u
ϕ = −∆u + (1 − σ) ∂τ2K ,



v = ∂wh ,
∂x1
and similarly,
(11.32) |∆2,K (ϕ, v)| ≤ C hK ||u||3,K ||wh ||2,K .
These inequalities lead us to the estimate
(11.33)
smartworlD.asia
| f (wh ) − ah (u, wh )| ≤ Ch||u||3,Ω ||wh ||h ,
for a regular family of triangulations made up of Adini’s rectangles. 123

Thus varying wh over Vh and taking the supremum, we get the estimate
(11.11).
Using (11.10) and (11.11) and substituting in (11.8) we get the re-
quired estimate as given in (11.23). This completes the proof.
Remark 11.3. By the Duality Argument, Lesaint and Lascaux [15] have
proved that ||u − uh ||1,Ω ≤ Ch2 ||u||4,Ω assuming u ∈ H 4 (Ω). They have
also got an improved 0(h2 ) convergence order in the || · ||h norm, when
all the rectangles are equal - a “superconvergence” result.
We close this section with a brief description of other types of finite

elements used in non-conforming methods.

Example 11.2. The Zienkiewicz triangle (cf. Exercise 4.6) cf. Fig. 11.4.
Figure 11.4:
We get Vh ⊂ C 0 (Ω) only and hence the method is non-conforming.

It does not always yield convergence. The method works if the sides of
all triangles are parallel to three directions only, as in Fig. 11.5.
smartworlD.asia
Figure 11.5:
124 This is not so if the number of directions is four, as in Fig. 11.6.

(The Union Jack Problem).

Figure 11.6:
Example 11.3. Morley’s Triangle (cf. Fig. 11.7).
smartworlD.asia
Figure 11.7:
Here PK = P2 . We always get convergence for regular families, of

course. In fact if u ∈ H 4 (Ω), then
(11.34) ||u − uh ||h ≤ Ch||u||4,Ω .
What is astonishing is that this finite element is not even of class C 0 .
Example 11.4. Fraeijs de Veubeke triangle. This finite element is again

a triangle. Apart from the values of the polynomials at the vertices and
the mid-points of the sides, we also take the average normal derivative
along the sides. Here the space PK , which we will not describe, satisfies
the inclusion
P2 ⊂ P K ⊂ P3 ,

and
 


 Z 


X 

 ∂p 


=
 p(ai ), 1 ≤ i ≤ 3; p(aij ), 1 ≤ i < j ≤ 3; dγ, 1 ≤ i ≤ 3
 .


 ∂ν 


K  K′ 
i
125 The finite element is shown symbolically in Fig. 11.8.
Figure 11.8:
smartworlD.asia
Here also the finite element is not of class C 0 in general, but the
method always yields convergence.
References: For general reference on non-conforming methods, see

Strang and Fix [22], for the bilinear lemma, see Ciarlet [29]. For a de-
tailed study of the Zienkiewicz triangle, Moreley’s triangle and Fraeijs
de Veubeke triangle, see Lascaux and Lesaint [15]. For a nonconform-
ing method with penalty, see Babuska and Zlámal [26].

Finite Element Methods

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Finite Element Methods

Uploaded by

Copyright:

Available Formats

Smartworld.asia Specworld.

1 The Abstract Problem 1

3 The Finite Element Method in its Simplest Form 29

5 General Properties of Finite Elements 53

6 Interpolation Theory in Sobolev Spaces 59

7 Applications to Second-Order Problems... 67

9 The Obstacle Problem 95

10 Conforming Finite Element Method for the Plate Problem 103

11 Non-Conforming Methods for the Plate Problem 113

2 1. The Abstract Problem

Definition 1.1. Let V be a normed linear space. A bilinear form a(·, ·)

(1.4) a(v, v) ≥ α||v||2 .

Theorem 1.1. Let V be a Banach space and K a closed convex subset

(1.5) a(u, v − u) ≥ f (v − u) for all v ∈ K.

Remark 1.1. The inequalities (1.5) are known as variational inequali-

Proof. The V-ellipticity of a(·, ·) clearly implies that if a(v, v) = 0 then

defined by the inner-product is equivalent to the existing norm on V.

(1.7) f (v) = a(σ f, v) for all v ∈ V.

The map σ : V ′ → V given by f 7→ σ f is linear. Now,

1. The Abstract Problem 3

(1.8) a(σ f − u, v − u) ≤ 0 for all v ∈ K.

Thus, a(σ f, v − u) ≤ a(u, v − u) which by virtue of (1.7) is precisely

We can state the following

Corollary 1.1. (a) If K is a non-empty closed convex cone with ver-

4 1. The Abstract Problem

(b) If K is a subspace of V then the solution is characterised by 4

(1.10) a(u, v) = f (v) for all v ∈ K.

Proof. (a) If K is a cone with vertex at 0, then for u, v ∈ K, u + v ∈ K.

(b) Applying (a) to K, since any subspace is a cone with vertex at 0,

Remark 1.3. The solution u of (P) corresponding to f ∈ V ′ (for a fixed

1. The Abstract Problem 5

σ f on K, it follows that the above map is linear if and only if K is

Exercise 1.1. Let V be as in Theorem 1.1. For f1 , f2 ∈ V ′ , let u1 , u2 5

Exercise 1.2. If V is a normed linear space, K a given convex subset of V

Exercise 1.3 (1 ). Apply the previous exercise to the functional

It was essentially the symmetry of the bilinear form which provided

6 1. The Abstract Problem

a(·, ·) a continuous, bilinear, V-elliptic form, f ∈ V ′ . If (P) is the prob-

(1.11) a(u, v) = f (v),

then (P) has a unique solution in V 1 .

for all u, v ∈ V. Fix any u ∈ V. Then the map v 7→ a(u, v) is continuous

Thus A is continuous and ||A|| ≤ M.

Let τ be the Riesz isometry, τ : V ′ → V so that

(1.15) f (v) = ((τ f, v)),

7 where ρ > 0 is a constant to be specified. We choose ρ such that

(1.17) g(v) = v − ρ(τAv − τ f ) for v ∈ V.

1. The Abstract Problem 7

||g(v1 ) − g(v2 )|| = ||(v1 − v2 ) − ρτA(v1 − v2 )||

||v − ρτAv||2 = ((v − ρτAv, v − ρτAv))

(1.18) 1 − 2ρα + ρ2 M 2 < 1

(1.19) α||u||2 ≤ a(u, u) = f (u) ≤ || f ||∗ · ||u||.

REFERENCE. For Variational Inequalities, see Lions and Stampac-

WE GIVE IN this section several examples of the abstract problem for- 9

and call it the length of the multi-index. If v is a real-valued function on

The space of test functions on Ω is given by

(2.4) supp(v) = {x ∈ Ω; v(x) , 0}.

10 Definition 2.1. Let m ≥ 0 be an integer. Then the Sobolev space H m (Ω)

where all derivatives are understood in the sense of distributions.

On H m (Ω) one can define a norm by means of the formula

It is easy to check that || · ||m,Ω defines a norm on H m (Ω), which

(2.8) H0m (Ω) = D(Ω),

(2.9) |v|0,Ω ≤ C|v|1,Ω .

11 Henceforth, unless specified to the contrary, the following will be

(2.10) ||v||L2 (Ω) ≤ C||v||1,Ω .