0001.

pdf 0241590001

Exercises
1 2 3 4 5 6 7 8 9 10
1 1 1 1 1 1 1
11
2 2 2 2 2 2 2

3 3 3 3 3 3 3
Surname, First name
4 4 4 4 4 4 4

5 5 5 5 5 5 5

6 6 6 6 6 6 6

7 7 7 7 7 7 7
4MC10 Computational mechanics
4MC10 Final Exam Q1 on campus 8 8 8 8 8 8 8

9 9 9 9 9 9 9

0 0 0 0 0 0 0

a b c d e f →b
a b c d e f →c
b c d e f →a

Fill in your answer(s) to the multiple-choice questions as shown above (circles = one correct answer).

Particular Ans on paper exam instructions

• Write in a black or blue pen.
• You answer open-ended questions by using the text box. Provide your answers on the papers inside
the answer box underneath a question. If you need more space for your answers, use the extra
space at the end of the exam, and clearly indicate there which question you continue answering.
In the text box of the particular question, clearly state that you proceed with your answer on a
different page.
• Hand in all pages. Do not remove the staple. If you remove it anyhow, check that you hand in all
pages.

Dear student,

You are about to take an exam. Write down your name and your student ID at the appropriate places
above. Make sure that you enter your student ID by marking the appropriate boxes. On the examination
attendance card, you fill in the PDF number. You can find the correct number at the top of the first page of
your exam (e.g. 1234.pdf).

Please read the following information carefully:

Date exam: 8 November 2023

Start time: 13.30
End time: 16.30 (+30 minutes for time extension students)

Number of questions: 10
Maximum number of points/distribution of points over questions: 100/see questions
Method of determining the final grade: point score/10

1 / 20
 0001.pdf 0241590002

Answering style:
• Multiple choice questions (Problems 1 to 8): Each of these problems has 5 possible answers (a to
e). The correct answer gives 3 points. For every wrong answer, 1 point is deducted. Answer e (No
answer) gives no points. Mark your answer as explained above, by marking the respective circle
by a cross. Fully colour a circle only to correct an earlier incorrect answer.
• Open questions (Problems 9 and 10): The number of points that may be earned per question is
indicated next to the question. Write your answer in the text box following the question. Make
sure that you fully answer (all parts of) the question. Give sufficient motivation, explanation and
intermediate steps in derivations to make clear how you arrived at your answer.

Permitted examination aids

• Scrap paper (fully blank, provided by invigilators)
• Pen, pencil, eraser, ruler, etc.

Important:
• You are only permitted to visit the toilets under supervision
• Examination scripts (fully completed examination paper, stating name, student number, etc.) must
always be handed in
• The house rules must be observed during the examination
• The instructions of subject experts and invigilators must be followed
• Keep your work place as clean as possible: put pencil case and breadbox away, limit snacks and
drinks
• You are not permitted to share examination aids or lend them to each other
• Do not communicate with any other person by any means

During written examinations, the following actions will in any case be deemed to constitute fraud or
attempted fraud:
• using another person’s proof of identity/campus card (student identity card)
• having a mobile telephone or any other type of media-carrying device on your desk or in your clothes
• using, or attempting to use, unauthorized resources and aids, such as the internet, a mobile telephone,
smartwatch, smart glasses etc.
• having any paper at hand other than that provided by TU/e, unless stated otherwise
• copying (in any form)
• visiting the toilet (or going outside) without permission or supervision

The final grade for this exam will be announced no later than fifteen working days after the date of this
examination.

You can start the exam now, good luck!

2 / 20
 0001.pdf 0241590003

Problem 1
Consider the linear algebraic problem Ax = b. The L and U factors of A and the right-hand side vector b
are given by

⎛1 0 0⎞ ⎛1 2 3⎞ ⎛ 6 ⎞
L=⎜ ⎟
⎜2 3 0⎟ , U =⎜ ⎟
⎜0 4 5⎟ and b = ⎜ ⎟
⎜ 39 ⎟
⎝4 5 6⎠ ⎝0 0 6⎠ ⎝105⎠

3p 1 Which of the following answers is correct?

⎛3⎞
a x=⎜ ⎟
⎜2⎟
⎝1⎠
⎛1⎞
b x=⎜ ⎟
⎜2⎟
⎝3⎠
⎛1⎞
c x=⎜ ⎟
⎜1⎟
⎝1⎠
⎛1⎞
d x=⎜ ⎟
⎜12⎟
⎝36⎠
e No answer

Problem 2
We consider an initial value problem

d ⎛x⎞ ⎛ 0 1⎞ ⎛x⎞ ⎛x(0)⎞ ⎛x0 ⎞

= =
dt ⎝y ⎠ ⎝−4 0⎠ ⎝y ⎠ ⎝y(0)⎠ ⎝ y0 ⎠
where (x0 , y0 )T represents a given vector. In addition, we regard the Runge-Kutta 4 time-integration
scheme, with a stability region as indicated in by the shaded area in the figure below.

3 / 20
 0001.pdf 0241590004

x
3p 2 Which of the following statements is correct?

The solution to the solution to the differential equation is stable, and the numerical scheme is
stable for step size h = 1
a

The solution to the solution to the differential equation is unstable, and the numerical scheme is
stable for step size h = 1
b

The solution to the solution to the differential equation is stable, and the numerical scheme is
unstable for step size h = 1
c

The solution to the solution to the differential equation is unstable, and the numerical scheme is
unstable for step size h = 1
d

e No answer

Problem 3
Consider the following part of a MATLAB script

4 / 20
 0001.pdf 0241590005

3p 3 Which of the following statements is correct?

The script is an implementation of an explicit time integration scheme for the initial value problem
u′ (t) = sin(2πt) for t ∈ (0, 1) and u(1) = 2
a

The script is an implementation of an implicit time integration scheme for the initial value problem
u′ (t) + u(t) = sin(2πt) for t ∈ (0, 1) and u(1) = 2
b

The script is an implementation of an explicit time integration scheme for the initial value problem
u′ (t) = sin(2πt) for t ∈ (0, 1) and u(0) = 2
c

The script is an implementation of an explicit time integration scheme for the initial value problem
u′ (t) + u(t) = sin(2πt) for t ∈ (0, 1) and u(0) = 2
d

e No answer

Problem 4
Consider the initial-value problem

d3 y(t) dy(t)
y(t) + + = exp(t)
dt3 dt
with

d2 y
y(0) = 3, (0) = 6, (0) = π
dy
dt dt2
3p 4 How many points should at least be used in a finite-difference approximation of the above differential
equation to obtain a consistent scheme with an accuracy of order 2?

a 4
b 5
c 6
d 7
e No answer

Problem 5
Consider the one-dimensional boundary value problem given by
⎧
⎪
⎪
⎪
⎪
d2 u
+ 2x − 1 = 0 for 0 < x < 1
⎪
⎪ dx2
⎨
⎪
⎪
⎪
⎪
⎪
⎪
du
=0 at x = 0 and x = 1
⎩ dx

5 / 20
 0001.pdf 0241590006

3p 5 Which of the following statements is correct?

a The problem has two essential boundary conditions and it has a unique solution
b The problem has two essential boundary conditions and it has no unique solution
c The problem has two natural boundary conditions and it has a unique solution
d The problem has two natural boundary conditions and it has no unique solution
e No answer

Problem 6
The sketch below shows a one-dimensional domain (0, L) which is subdivided into 8 quadratic finite
elements. Also sketched are 4 shape functions Ni (x).
x

x
3p 6 Which of these shape functions does not belong to this discretisation?

a Shape function a
b Shape function b
c Shape function c
d Shape function d
e No answer

Problem 7
The Matlab program below computes a finite element solution of a simple boundary value problem, using
a very crude discretisation.

6 / 20
 0001.pdf 0241590007

x
Consider the following two statements on this program:

I It applies one essential and one natural boundary condition

II The condition applied on the final node reads q5 = 0.

3p 7 Which of the above statements is/are correct?

a Both statements are correct

b Statement I is correct but Statement II is wrong
c Statement I is wrong but Statement II is correct
d Both statements are wrong
e No answer

Problem 8
Consider the function

⃗ = ξ2 + ξ2
g(ξ) 1 2

⃗ is defined on the two-dimensional domain Q ∶ (−1, 1) × (−1, 1). We

where ξ⃗ = ξ1 e⃗1 + ξ2 e⃗2 and g(ξ)
approximate the integral of g(ξ) ⃗ on Q,

⃗ dQ
I = ∫ g(ξ)
Q

by a 4-point Gauss rule.

7 / 20
 0001.pdf 0241590008

3p 8 The result of this operation equals:

a I4 = 0
b I4 = 4
3
c I4 = 8
3
d I4 = 4
e No answer

8 / 20
 0001.pdf 0241590009

Problem 9
Consider the initial-value problem

(1) = exp(t) in (0, 1)

du(t)
dt
with initial condition

(2) u(0) = 1

We regard a grid with N + 1 (N > 2) grid points ti = ih (i = 0, 1, 2, . . . , N ) with step size h = 1/N , and the
finite-difference schemes

uhi − uhi−1
(3) = exp(ti ) for i = 1
h
and

α uhi + β uhi−1 + uhi−2

(4) = exp(ti ) for i = 2, 3, . . . , N + 1
2h
where α, β are numbers to be determined, independent of h.

20p 9a Determine the values of α, β for which the above finite-difference scheme in (4) is a consistent
approximation to the given differential equation

x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
X
↶

9 / 20
 0001.pdf 0241590010

↷
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
X
↶

10 / 20
 0001.pdf 0241590011

↷
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
X

11 / 20
 0001.pdf 0241590012

8p 9b Provide a condition for uh0 that is consistent with the above initial condition in (2). Show that the
condition that you have provided satisfies the definition of consistency.

x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
X

12 / 20
 0001.pdf 0241590013

10p 9c We compute the finite-difference approximation for step size h = 3−3 , 3−4 , and regard the
approximation for the end point, uhEND . We find that the error in the approximation uhEND decreases by
a factor of approximately 9 whenever the step size is reduced by a factor of 3. We use Richardson
extrapolation based on uhEND for the aforementioned 2 values of h to get an improved approximation,
i.e. we define

ûEND = a uhEND + b uEND

h/3

for certain coefficients a, b, such that the order of accuracy of ûEND is one higher than that
of uhEND . Determine the coefficients a and b.

x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
X

13 / 20
 0001.pdf 0241590014

Problem 10
Finite element simulations play an essential role in the development and engineering of advanced products
and equipment. As an example, we consider in this problem the high-capacity digital printer shown on
the left in the figure below. The right part of the figure shows a finite element model of one of the rollers
which are involved in the paper transport in the machine. The purpose of the finite element analysis is
to assess the stresses which may be expected to occur in the roller and, based on them, to predict its
lifetime.
x

x
The roller is modelled as a linear elastic solid. A plane strain assumption has been made to reduce the
problem to two dimensions. Furthermore, symmetry has been employed to limit the simulation to a single
roller in contact with a rigid (symmetry) plane instead of with the opposite roller.

The partial differential equation which governs the elastic deformation of the roller reads, in weak form,

⃗ φ⃗T ∶ 4C ∶ ∇⃗
∫ ∇ ⃗ x) ⋅ q⃗(⃗
⃗ u dV = ∫ φ(⃗ x) dS ⃗ x)
∀ φ(⃗
V S
⃗(⃗
where u x) is the unknown displacement field, defined on the two-dimensional domain V , and q⃗(⃗ x)
⃗
the boundary traction on the boundary S of V . Furthermore, φ(⃗ x) denotes a weight function and the
constant fourth-order tensor 4C is the two-dimensional (plane strain) isotropic elasticity tensor.
8p 10a Give the product rule which was used to derive the above weak formulation from the underlying
strong formulation. Take care to correctly represent all inner products.

x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
X
↶

14 / 20
 0001.pdf 0241590015

↷
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
X

10p 10b Based on the above weak formulation and the shape functions N(⃗ x) associated with the finite
~
⃗=q
element mesh shown, derive the discrete system of equations K ⋅ u ⃗. Give expressions for the
— ~ ~
stiffness matrix K and right hand side q
⃗.
— ~
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
X
↶

15 / 20
 0001.pdf 0241590016

↷
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
X

16 / 20
 0001.pdf 0241590017

The sketch below shows the isoparametric finite element used in the discretisation, relative to the local
coordinates ξ1 and ξ2 .
x

x
4p 10c Does this element belong to the Lagrange family or the serendipity family?

x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
X

8p ⃗ belonging to node number 3 of the element.

10d Derive the shape function N3e (ξ)

x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
X
↶

17 / 20
 0001.pdf 0241590018

↷
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
X

8p 10e Explain how the gradient of the shape functions with respect to global coordinates, ∇N ⃗ e , may be
~
⃗ ξ Ne , and the nodal positions, x
computed from their gradient with respect to local coordinates, ∇ ⃗e .
~ ~
x
x
x
x
x
x
x
x
x
x
x
x
x
x
X
↶

18 / 20
 0001.pdf 0241590019

↷
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
X

19 / 20
 0001.pdf 0241590020

Additional answering space

11 Please indicate clearly which question you continue answering here – in the answer box of that
question and in the box below.

x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
x
X

20 / 20
 Answers
Exam Computational Mechanics (4MC10)
Date: 8 November 2023
Time: 13:30–16:30
Location: Vertigo 3.06, 3.15, 4.06, 4.15

Problem 1
C. It holds that A = LU . The solution x therefore corresponds to the solution of the two
equations Ly = b and U x = y. By forward substitution it follows that y = (6, 9, 6)T and,
in turn, it follows by backward substitution that x = (1, 1, 1)T .

Problem 2
A. The eigenvalues of the system matrix follow from the characteristic equation
✓ ◆
1
det = 2+4=0
4

with roots 1,2 = ±2ı where ı stands for the imaginary unit. The purely imaginary eigen-
values belong to the left half of the complex plane, {z 2 C : Re(z)  0} and, hence, the
di↵erential equation is stable. For h = 1 the values h 1,2 lie in the stability region of the RK
scheme. Therefore, the time integration scheme is stable at step size h = 1.

Problem 3
D. The expression in line 10 can be reexpressed as

uhn+1 uhn
+ uhn = sin(2⇡tn )
h
One can recognize (or derive by Taylor series expansions) that this expression is a consistent
approximation of the ordinary di↵erential equation u0 (t) + u(t) = sin(2⇡t). The statement
in line 6 initializes uh0 as uh0 = 2. Here it is important to note the index shift in Matlab, by
which the first entry of the vector in fact pertains to uh0 .

Problem 4
B. The presented di↵erential equation is of order n = 3, i.e. the highest-order derivative that
appears in the equation is the third-order derivative. The number of points that is required
in a finite di↵erence scheme to approximate a derivative of order n with accuracy p is n + p.

Problem 5

1
 D. The boundary condition du/dx = 0 at x = 0 and at x = 1 is of the natural type because
it involves the derivative of u(x) and not u(x) itself. If u⇤ (x) is a solution of boundary
value problem, u⇤ (x) + c, with c an arbitrary constant, is also a solution. So the solution is
non-unique.

Problem 6
B. The shape functions of a quadratic element are element-wise quadratic functions. This
holds for functions a, c and d, but not for b, which has a kink (discontinuous derivative) at
the centre of an element.

Problem 7
B. The program applies an essential boundary condition to Node 5 and a natural condition
to Node 1. The condition for Node 5 reads u5 = 1.

Problem 8
C. The 4-point Gauss rule has integration points at ⇠~k = p13 (±~e1 ± e~2 ) (k = 1, 2, 3, 4), with
weight factors wk = 1. The approximation I4 may thus be computed as
4
X
I4 = wk g(⇠~k ) = 4 23 = 8
3
k=1

Problem 9
a. Define
9
h h ↵ uhi + uhi 1 + uhi 2 =
(L u )i =
2h voor i = 2, 3, . . . , N + 1
;
fih = exp(ti )
The finite-di↵erence scheme is consistent with the given ordinary di↵erential equation if for
any solution u of the di↵erential equation it holds that
fh Lh (I h u) = o(1) as h ! 0 (1)
where I h u is the injection of u, i.e. (I h u)i = u(ti ). Substituting I h u in the left member of
the finite-di↵erence scheme, and applying Taylor-series expansions, one obtains:

fih (Lh (I h u))i

✓ ◆
↵ u(ti ) + u(ti 1 ) + u(ti 2 )
= exp(ti )
2h
✓  ✓ ◆
1 1 2 2 1 3 3 4
= exp(ti ) ↵ u(ti ) + u(ti ) Du(ti )h + D u(ti )h D u(ti )h + O(h )
2h 2 6
✓ ◆
1 2 2 1 3 3 4
+ u(ti ) Du(ti )(2h) + D u(ti )(2h) D u(ti )(2h) + O(h )
2 6
✓ 1 1 4
↵+ +1 2 +2 2 2 3 3 3
= exp(ti ) u(ti ) + h Du(ti ) + 2 h D u(ti ) + 6 h D u(ti )
2h 2h 2h 2h
⇤ ⌘
+ O(h4 )

2
 (2)

Note that the Taylor series are truncated at O(h4 ), because the finite-di↵erence scheme
contains 3 points. We have to choose ↵, such that the coefficient of the term corresponding
to Du(ti ) equals 1, while the coefficient of the term associated to u(ti ) vanishes, i.e. ↵+ +1 =
0 and ( 12 1) = 1. With ↵, independent of h, this holds for ↵ = 3 and = 4. For
these values of ↵ and , we have
1 2 3
fih (Lh (I h u))i = exp(ti ) Du(ti ) h D u(ti )+O(h3 ) = O(h2 ) = o(1) as h ! 0
3
(3)
The penultimate identity holds because u satisfies the ODE by assumption.
b. The initial condition reads u(0) = 1. Since uh0 corresponds to the finite-di↵erence approxima-
tion at time t = 0, the initial condition can be consistently incorporated in the finite-di↵erence
scheme by imposing uh0 = 1. To establish that this condition is indeed consistent, we recall
the definition of consistency: If u satisfies (2), then

gh B h (I h u) = o(1) as h ! 0

where in this case,

g h = 1, B h (uh ) = uh0
where the injection operator is defined as (I h u)i = u(ti ). We therefore have

gh B h (I h u) = 1 u(t0 ) = 1 u(0) = 0 = o(1) as h ! 0

The penultimate identity follows from the fact that u satisfies (2), by assumption.
c. From the fact that the error decreases by a factor 1/9 = (1/3)2 whenever the step size h is
reduced by a factor 3, we can infer that the error is O(h2 ). This is in agreement with the
results of question 9a. It therefore holds that:

uhend = uend + Ch2 + O(h3 )

where uend denotes the exact solution, and C indicates some constant. From the given
Richardson-extrapolation expression, one infers:
h/3
ûend = a uhend + b uend = a uend + Ch2 + O(h3 ) + b uend + C(h/3)2 + O(h3 )
= (a + b)uend + (a + 19 b)Ch2 + O(h3 )

By selecting a, b such that a + b = 1 and a + 19 b = 0, we obtain an approximation ûend =

uend + O(h3 ). From the aforementioned 2 equations for a, b, one obtains:
1 9
a= 8, b= 8

Problem 10
a. The product rule used reads
h i h i
~
r· 4 ~ u · ~ (~x) = ~ (~x) · r·
C : r~ ~ 4C : r~
~u +r~ ~ T : 4C : r~
~u

In this expression we recognise the elasticity operator in the first term of the right hand side.
This terms is eliminated from the weighted residuals form by replacing it by the left hand
side minus the second right hand side term. The latter appears on the left hand side in the
weak formulation and the former gives rise to the boundary integral on the right hand side.

3
b. To discretise the problem, we replace the domain V by the discrete domain V h , its boundary
S by the boundary S h of V h and make the following substitutions:

~u(~x) ! ~uh(~x) = NT(~x) ~u ~ (~x) ! ~ h(~x) = ~ T N(~x) (4)

˜ ˜ ˜ T˜
~ u(~x) ! r~
r~ ~ uh = rN~ T ~u ~ ~T ! r
r ~ ~ hT = ~ rN ~ (5)
˜ ˜ ˜ ˜
where the second column follows from Galerkin’s approach and the second line by di↵eren-
tiation of the first. The weak formulation may now be written as
Z Z
~ T · rN~ · C · rN
4 ~ T
dV · ~u = ~ T
· N(~x) ~q (~x) dS 8 ~
˜ ˜ ˜ ˜ ˜ h ˜ ˜
V h S

T
where ~ and ~u, which do not depend on ~x, have been taken out of the integrals. This
˜ is now˜of the form
equation
~ T · K · ~u = ~ T · ~q 8 ~
˜ ¯ ˜ ˜ ˜ ˜
with the definitions
Z Z
K = rN ~ · 4C · rN
~ T dV ~q = N(~x) ~q (~x) dS
¯ ˜ ˜ ˜ ˜
V h Sh

The above equation holds for all ~ only if

˜
K · ~u = ~q
¯ ˜ ˜
c. The element has 3 ⇥ 3 = 9 nodes and hence belongs to the Lagrange family.
~ should vanish at ⇠1 = 1, ⇠1 = 0, ⇠2 = 1 and ⇠2 = 0, so it must be
d. Shape function N3e (⇠)
of the form
~ = A3 ⇠1 (⇠1 + 1) ⇠2 (⇠2 + 1)
N3e (⇠)

To furthermore ensure that the shape function equals one in the node it belongs to, we need
to require that

N3e (⇠~3e ) = N3e (~e1 + ~e2 ) = 4A3 = 1

1
This results in A3 = 4 and so
~ =
N3e (⇠) 1
4 ⇠1 (⇠1 + 1) ⇠2 (⇠2 + 1)

e. A relationship between the local and global gradients may be established using the chain
rule:
~ ⇠ Ne = r
r ~ ⇠ ~x · rN
~ e
˜ ˜
~ =r
In this expression we recognise the Jacobian J (⇠) ~ ⇠ ~x, which may be computed from the
isoparametric mapping as
~ =r
J (⇠) ~ ⇠ ~x(⇠) = r~ ⇠ NeT(⇠) ~xe = r
~ ⇠ NeT ~xe
˜ ˜ ˜ ˜
1 ~
Pre-multiplying by its inverse, J (⇠), the above chain rule may be rewritten as:
~ e = J 1(⇠)
rN ~ ·r ~ ⇠ Ne
˜ ˜
~ ⇠ Ne and ~xe .
which is now fully computable based on r
˜ ˜