Outline Finite Differences Difference Equations FDM FEM

Finite Difference and Finite Element Methods

Georgy Gimel’farb

COMPSCI 369 Computational Science

1 / 39
Outline Finite Differences Difference Equations FDM FEM

1 Finite Differences

2 Difference Equations

3 Finite Difference Methods: Euler FDMs

4 Finite Element Methods (FEM) [optional]

Learning outcomes:
• Be familiar with the finite difference models and methods (Euler FDMs)
• Optional: Runge-Kutta FDMs, more accurate FEMs

Recommended reading:
• M. T. Heath, Scientific Computing: An Introductory Survey. McGraw-Hill, 2002: Chapters 5, 8 – 11
• M. Shäfer, Computational Engineering - Introduction to Numerical Methods. Springer, 2006: Chapters 3, 5
• G. Strang, Computational Science and Engineering. Wellesley-Cambridge Press, 2007: Chapters 1.2, 3, 6

2 / 39
Outline Finite Differences Difference Equations FDM FEM

Let’s Recall Differential Equations

Modern science and engineering assume our world is continuous

and described by differential equations and integral equations
du(x)
Example I: 1st -order differential equation dx = 2; u(0) = 0
General solution: u(x) = 2x + A
Boundary condition: u(0) = 0 gives A = 0
Final solution: → u(x) = 2x

Unknown u(x) is specified by a known instant speed of changes:

du(x) u(x + ∆x) − u(x)

= lim
dx ∆x→0 ∆x

3 / 39
Outline Finite Differences Difference Equations FDM FEM

Let’s Recall Differential Equations, continued

d2 u(x)
Example II: 2nd -order DE dx2
= 2; u(0) = u(1) = 0
General solution: u(x) = x2 + Ax + B
Boundary conditions: u(0) = 0; u(1) = 0 give B = 0; A = 1
Final solution: → u(x) = x2 − x

Unknown u(x) is specified by a known instant acceleration of

changes:
 
d2 u(x) 1 du(x+∆x) du(x)
dx2
= lim ∆x dx − dx
∆x→0
 
1 u(x+2∆x)−u(x+∆x) u(x+∆x)−u(x)
= lim −
∆x→0 ∆x ∆x ∆x

u(x+2∆x)−2u(x+∆x)+u(x)
= lim ∆x2
∆x→0
4 / 39
Outline Finite Differences Difference Equations FDM FEM

Physical Oscillating Second-order Systems

2
• Non-damped mass on a spring: −m d dtx(t) 2 = kx
• x(t) – a displacement
• k – a spring constant (the Hooke’s law)
• m – a mass

Mass m
x
Displacement x

• Damped mass on a spring θ

(γ – the friction constant): R
2 m
−m d dtx(t)
2 = γ dx(t)
dt + kx
2
• Pendulum: −mR2 ddt2θ = mgR sin θ
2 g mg
⇔ − ddt2θ = R sin θ
5 / 39
Outline Finite Differences Difference Equations FDM FEM

Differences Vs. Derivatives

Finite differences: natural approximations to derivatives

du(x) u(x + ∆x) − u(x) ∆u(x) u(x + ∆x) − u(x)

= lim ⇔ ≡
dx ∆x→0 ∆x ∆x ∆x

u u

x x
∆x

Discrete step ∆x can be small but it does not tend to zero!

6 / 39
Outline Finite Differences Difference Equations FDM FEM

Differences Vs. Derivatives

• Three possibilities for the finite difference:

Test for u(x) = x2

Forward difference u(x+h)−u(x) (x+h)2 −x2
h h = 2x + h

Backward difference u(x)−u(x−h) x2 −(x−h)2

h h = 2x − h

Centred difference u(x+h)−u(x−h) (x+h)2 −(x−h)2

2h 2h = 2x

• Results of testing:
du(x)
• The centred difference gives the exact derivative dx = 2x

• The forward and the backward differences differ by h

7 / 39
Outline Finite Differences Difference Equations FDM FEM

Taylor Series Approximation

• Let u(x) be an arbitrary infinitely differentiable function
• In plain words, let all its derivatives exist:

dn u(x)
; 1≤n≤∞
dxn
n u(x) 

• Given u(x0 ) and d dxn  ; n = 1, 2, . . ., at x = x0 , the
x=x0
value u(x0 + h) at x = x0 + h is represented by the Taylor’s
series of the values at x = x0 :
∞ n dn (u(x) 

P h
u(x0 + h) = u(x0 ) + n! n 

n=1 dx
| {z x=x}0
≡u[n] (x0 )
∞
P hn [n]
≡ u(x0 ) + n! u (x0 )
n=1

8 / 39
Outline Finite Differences Difference Equations FDM FEM

A Few Famous Taylor Series

dex 

• Exponential function u(x) = ex ⇔ e0 = 1; dx x=0 = ex |x=0 = 1:
h2 h3 h4
⇒ u(0 + h) ≡ eh = 1 + h + + + + ...
2 6 24
• Polynomial ratio  
1 du(x)  −2x 
u(x) = 1+x 2 ⇔ u(0) = 1; dx  = (1+x2 )2  = 0;
  x=0
 x=0
2
d u(x) 2 8x
= − (1+x 2 )2 + = −2:
 
dx 2  (1+x2 )3 
x=0 x=0

1 1
u(0+h) ≡ 2
= 1−h2 +h4 −. . . ⇒ u(1) = = 1 − 1 + 1 − 1 + . . .
1+h 2
• Trivial for any polynomial
du(x) d2 u(x) d3 u(x)
• u(x) = x3 − 2 ⇔ u(1) = 1; dx
= 3x2 ; dx2
= 6x; dx3
= 6:

h2 h3
⇒ u(1 + h) = −1 + 3h + 6 +6 = −1 + 3h + 3h2 + h3
2 6
9 / 39
Outline Finite Differences Difference Equations FDM FEM

Accuracy of Finite Differences

Taylor series approximations:

u(x + h) = u(x) + hu0 (x) + 12 h2 u00 (x) + 61 h3 u[3] (x) + . . .

u(x − h) = u(x) − hu0 (x) + 21 h2 u00 (x) − 16 h3 u[3] (x) + . . .

• The first-order accuracy O(h) of the one-sided differences:

u(x+h)−u(x)
h = u0 (x) + 12 hu00 (x) + . . .
u(x)−u(x−h)
h = u0 (x) + 12 hu00 (x) + . . .

• The second-order accuracy O(h2 ) of the centred differences:

u(x + h) − u(x − h) 1
= u0 (x) + h2 u[3] (x) + . . .
2h 6

10 / 39
Outline Finite Differences Difference Equations FDM FEM

Second Difference
The second difference approximation of the second derivative:

d2 u ∆2 u u(x + ∆x) − 2u(x) + u(x − ∆x)

2
≈ 2
=
dx ∆x ∆x2

• The second order accuracy O(h2 ) due to the centred ∆2 u:

u(x + ∆x) − 2u(x) + u(x − ∆x) 1

2
= u00 (x) + h2 u[4] (x) + . . .
∆x 12
2 2 +(x−h)2
• Exact second derivatives (x+h) −2x
h2
= 2 for u(x) = x2
(x+h)3 −2x3 +(x−h)3
and h2
= 6x for u(x) = x3
4 4 4
• O(h2 ) accuracy (x+h) −2xh2
+(x−h)
= 12x2 + h2 for u(x) = x4 with
00
the true second derivative u (x) = 12x2
11 / 39
Outline Finite Differences Difference Equations FDM FEM

Finite Difference Equations

2
The 2nd -order differential equation − d dx
u(x)
2 = f (x)
• Known source function f (x)
• Known boundary conditions, e.g. u(0) = 0 and u(1) = 0
Its discrete form – the 2nd -order finite difference equation:
1
• Divide the interval [0, 1] into equal pieces of length h = n+1
that meet at the points x = h, x = 2h, . . . , x = nh:

• Approximate the goal values u(h), . . . , u(nh) at n discrete

points [h, 2h, . . . , nh] inside the interval [0, 1] with the values
2u
u1 , . . . , un , respectively, using the finite difference ∆
∆x2
:

u0 − 2u1 + u2 = f1 ≡ f (h); u1 − 2u2 + u3 = f2 ≡ f (2h);

... un−1 − 2un + un+1 = fn ≡ f (nh)

12 / 39
Outline Finite Differences Difference Equations FDM FEM

Finite Difference Equations: Matrix-Vector Form

For the n unknowns u1 , u2 , . . . , un and the boundary conditions

d2
u0 = un+1 = 0, the differential operator dx 2 is replaced with the
n × (n + 2) matrix:
    
1 −2 1 u0 = 0 f1
 1 −2 1   u1   f2
  
1  . . .

.   ..

− 2
 .. .. ..   .. = .

h 
 
   
 1 −2 1   un   fn−1 
1 −2 1 un+1 = 0 fn

where the source values are specified as fi = f (ih)

2
• For the special case ddxu2 = 1 with the above boundary conditions,
the finite differences give an exact match between u(ih) and ui

13 / 39
Outline Finite Differences Difference Equations FDM FEM

Example with the Constant Source f (x) = 1

2
• Differential equation: − ddxu2 = 1 with u(0) = u(1) = 0
• Complete solution ucomplete = upart + unull :
a particular one for u00 = 1 plus the nullspace one for u00 = 0
2 2
• Particular solution: − ddxu2 = 1 is solved by upart = − x2
2
• Nullspace solution: − ddxu2 = 0 is solved by unull = Cx + D
2
• u(x) = x2 + Cx + D ⇒ From the boundary conditions:
• u(0) = D = 0
• u(1) = − 12 + C + D = 0 → C = − 21 ⇒
2
• u(x) = x−x2

• Difference equation: −ui−1 +2u

h2
i −ui+1
= 1 with u0 = un+1 = 0
2 2
• (!) Just the same parabolic solution: ui = ih−i2 h
• However such a perfect agreement between the discrete ui and
the exact continuous u(ih) is very unusual

14 / 39
Outline Finite Differences Difference Equations FDM FEM

Another Example: Free End Boundary Condition (optional)

Free end condition: zero slope u0 (0) = 0 at x = 0 and u(1) = 0

2
• Differential equation: − ddxu2 = 1 with du
dx (0) = u(1) = 0
2 2
• The complete solution: u(x) = − x2 + Cx + D = 1−x 2
• Boundary conditions:
u0 (0) = C = 0; u(1) = −0.5 + C + D = 0 → D = 0.5
• Difference equation: −ui−1 +2u
h2
i −ui+1
= 1 with u1 −u
h
0
=0
and un+1 = 0
    
1 −1 u1 1
 −1 2 −1  u2   1 
1  .. .. ..

..
 
..

− 2 =
    
. . . . .
h 
 
   
 −1 2 1   un−1   1 
−1 2 un 1
| {z }
Tn

15 / 39
Outline Finite Differences Difference Equations FDM FEM

Free End Boundary Condition, continued

    
u1 n n − 1 n − 2 ... 1 1

 u2 


 n − 1 n − 1 n − 2 ... 1 
 1 

.. n − 2 n − 2 n − 2 ... 1 ..
 = h2 
    
 .  . 
   .. .. .. ..  
 un−1   . . . . 1  1 
un 1 1 1 ... 1 1
| {z }
T−1
n

Discrete solution: ui = h2 (n+i)(n+1−i)

2
• Error w.r.t. the exact solution:
   
i2
u(ih) − ui ≡ u n+1 i
− ui = 12 1 − (n+1) 2 − 12 (n+i)(n+1−i)
(n+1)2
 
= 12 (n+1)
n+1−i
2 = 2
1 1 i
n+1 − (n+1)2 ⇒ O(h)

16 / 39
Outline Finite Differences Difference Equations FDM FEM

Free End Boundary Condition, continued

• Therefore, the one-sided boundary conditions change the

matrix and may result in O(h) error
• A more accurate difference equation can be constructed by
centring the boundary conditions
• Centred difference equation will have 2nd -order errors O(h2 )
• Previous and similar differential and difference equations can
be met in many practical problems
• Simple physical examples of oscillators (see Slide 5):
• a non-damped mass on a spring
• a damped mass on a spring
• a pendulum

17 / 39
Outline Finite Differences Difference Equations FDM FEM

Stiff Differential Equations

A
  z
 }| {      
d v −50 49 v v(0) 2
Example: dt = ; =
w 49 −50 w w(0) 0
Solution: v(t) = e−t + e−99t and w(t) = e−t − e−99t
• Time scales differ by a factor of 99 (the condition number of A)
• Solution decays at the slow time scale of e−t , but computing e−99t
may require a very small step ∆t for stability
• Thus, ∆t is controlled by the fast decaying component, and this is
really counterproductive!
Stiffness comes with any problem involving very different time
scales (chemical kinetics, control theory, circuit simulations, etc.)

18 / 39
Outline Finite Differences Difference Equations FDM FEM

Forward and Backward Euler Methods (optional)

• Differential equation du(t)

dt = f (u, t); an initial value u(0)
• The rate of change u0 is determined by the current state u at
any moment t
• Forward Euler:
un+1 − un
= f (un , tn ) ⇒ un+1 = un + ∆t · fn
∆t
• Backward Euler:
un+1 − un
= f (un+1 , tn+1 ) ⇒ un+1 − ∆t · fn+1 = un
∆t

• Implicit method: if f is linear in u, a linear system is solved at

each step

19 / 39
Outline Finite Differences Difference Equations FDM FEM

Forward and Backward Euler Methods, continued

du(t)
Example: Linear differential equation u0 = au (i.e. dt = au);
u(0) = u0 > 0, with the exact solution: u(t) = eat u(0) ≡ u0 eat
√
Generally, a is a complex number: a = (Re a) + (Im a) −1
• Forward Euler:
one step un+1 = (1 + a∆t)un → un = (1 + a∆t)n u0
• Convergence: (1 + a∆t)T /∆t → eaT as ∆t → 0
• Sharp instability border 1 + a∆t = −1, i.e. for a = −2/∆t
• Backward Euler:
one step (1 − a∆t)un+1 = un → un = (1 − a∆t)−n u0
• Convergence: (1 − a∆t)−1 = 1 + a∆t+ higher order terms, so
that (1 − a∆t)−T /∆t → eaT as ∆t → 0
• Stability whenever u0 = au is stable (i.e. for a having the
negative real part: Re a < 0)

20 / 39
Outline Finite Differences Difference Equations FDM FEM

Forward and Backward Euler Methods, continued

Since forward and backward differences are first order accurate, the
errors from both methods are O(∆t)
Second-order methods:
• Formal notation: fn = f (un , tn ) and fn+1 = f (un+1 , tn+1 )
• Crank-Nicolson:

un+1 − un fn+1 + fn 1 + 12 a∆t

= ⇒ un+1 = un
∆t 2 1 − 12 a∆t

• Stable even for stiff equations, when a  0

21 / 39
Outline Finite Differences Difference Equations FDM FEM

Forward and Backward Euler Methods, continued

Second-order methods:
• Explicit forward Euler:

un+1 − un 3fn − fn−1

=
∆t 2

• Stable if ∆t is small enough: −a∆t ≤ 1 if a is real

• But explicit systems always impose a limit on ∆t
• Implicit backward difference:

3un+1 − 4un + un−1

= fn+1
2∆t

• More stable and accurate than the Crank-Nicolson method

(trapezoidal rule)
22 / 39
Outline Finite Differences Difference Equations FDM FEM

Explicit and Implicit Multistep Methods (optional)

With p earlier values of un , the accuracy is increased to order p

∇u = u(t) − u(t − ∆t)

∇2 u = u(t) − 2u(t − ∆t) + u(t − 2∆t)
∇3 u = u(t) − 3u(t − ∆t) + 3u(t − 2∆t) − u(t − 3∆t)
...
∇p u = u(t) − pt(y − ∆t) + . . . + (−1)p u(t − p∆t)

• Backward
 differences: 
∇ + 12 ∇2 + . . . + p1 ∇p un+1 = ∆t · f (un+1 , tn+1 )

• Backward Euler method: p = 1

• Implicit backward difference method: p = 2

23 / 39
Outline Finite Differences Difference Equations FDM FEM

Explicit and Implicit Multistep Methods, continued

An alternative: using older values of f (un , tn ) instead of un

• Explicit forward Euler methods:

un+1 − un = ∆t · (b1 fn + . . . + bp fn−p+1 )

order of b1 b2 b3 b4 limit on −a∆t

accuracy for stability
p=1 1 2
p=2 3/2 -1/2 1
p=3 23/12 -16/12 5/12 6/11
p=4 55/24 -59/24 37/24 -9/24 3/10

• The like implicit methods are even more stable and accurate

24 / 39
Outline Finite Differences Difference Equations FDM FEM

Runge-Kutta Methods (optional)

• Highly competitive and self-starting if evaluations of f (u, t)

are not too expensive
• Compound 1-step methods, using Euler’s un + ∆tfn inside
the function f
• Simplified RKM:
 
un+1 − un 1
= fn + f (un + ∆tf n , tn+1 )

∆t 2 | {z }
compounding of f

25 / 39
Outline Finite Differences Difference Equations FDM FEM

Simplified Runge-Kutta Method (optional)

• 2nd -order accuracy

• An example for u0 = au:

un + 21 ∆t[aun + a(u
un+1 =  n + ∆taun )]
1 2 2
= 1 + a∆t + a ∆t un
2
| {z }
Growth factor G

• This growth factor G reproduces the exact ea∆t through the

third term 21 a2 (∆t)2 of the Taylor’s series

26 / 39
Outline Finite Differences Difference Equations FDM FEM

Higher-order Runge-Kutta Method (optional)

−un
• Fourth-order RKM: un+1
∆t = 13 (k1 + 2k2 + 2k3 + k4 ) where

k1 = 12 f (un , tn ) k3 = 21 f un + ∆tk2 , tn+1/2

k2 = 21 f un + ∆tk1 , tn+1/2 k4 = 21 f (un + 2∆tk3 , tn+1 )

• This 1-step method needs no special starting instructions

• It is simple to change ∆t during computations
• The growth factor reproduces ea∆t through the fifth term
1 4 4
24 a (∆t) of the Taylor’s series
• Among highly accurate methods, RKM is especially easy
to code and run – probably the easiest there is!
• The stability threshold is −a∆t < 2.78

27 / 39
Outline Finite Differences Difference Equations FDM FEM

Finite Element Methods (FEM) [optional]

Finite difference methods (FDM) approximate the differential equation

(+) Very easy implementation
(−) Low accuracy between grid points
Finite element methods (FEM) approximate directly the solution of the
differential equation
(+) Easy handling of complex geometry and boundaries of a problem
domain
(±) Usually, more accurate than FDM but this depends on a problem
• Unknown solution u(x) as a combination of n basis functions φ(x):
u(x) = u1 φ1 (x) + . . . + un φn (x)
• FDM are special cases of the FEM with specific basis functions

28 / 39
Outline Finite Differences Difference Equations FDM FEM

Finite Element Methods: Weak Form

• Weak (or variational) form of the differential equation

−u00 (x) = f (x) in [0, 1] with u(0) = u(1) = 0:
• For any smooth function ν(x) satisfying the boundary
conditions ν(0) = ν(1) = 0:

Z1 Z1 Z1
00
f (x)ν(x)dx = − u (x)ν(x)dx = u0 (x)ν 0 (x)dx
0 0 0

Recall the calculus: Integration by parts

Zb Zb
0
y(s)z (s)ds = y(s)z(s)|ba − y 0 (s)z(s)ds
a a

29 / 39
Outline Finite Differences Difference Equations FDM FEM

Galerkin’s FEM: −u00 = f in the discrete weak form

Discretisation of ν(x) with n “test functions” ν1 (x), . . . , νn (x)
• The weak form gives one equation for each νj (x) involving
numerical coefficients u1 , . . . , un :
R1 P R1
n 
dφi (x) dνj (x)
ui dx dx dx = f (x)νj (x)dx; j = 1, . . . , n
0 i=1 0

⇒ Ku = f ,
n
P
that is, Kij ui = fj ; j = 1, . . . , n, where
i=1

Z1   Z1
dφi (x) dνj (x)
Kij = dx and fj = f (x)νj (x)dx
dx dx
0 0

30 / 39
Outline Finite Differences Difference Equations FDM FEM

Galerkin’s FEM: −u00 = f (cont.)

• Mostly, the test functions νi ’s are the same as the basis

functions φi ’s for u(x)
• Then the stiffness matrix K is symmetric and positive
definite
• Integrals of products of φi (x) and νi (x) giving the matrix
components Kij should be finite (−∞ < Kij < ∞)
• So step functions are not allowed as the basis and test
functions!
• Derivative of a step-function is the delta-function δ, and the
squared delta-function δ 2 has an infinite integral
• Linear hat functions (having step functions as their derivatives)
can be used as the basis and test functions

31 / 39
Outline Finite Differences Difference Equations FDM FEM

Linear Finite Elements: Hat Functions

1
h = ; u(x) = u1 φ1 (x) + u2 φ2 (x) + u3 φ3 (x)
3
Basis and test functions φi (x) = νi (x); i = 1, 2, 3:
φ1 (x) φ2 (x) φ3 (x)
1

0 h 2h 1
φ1 (x) φ2 (x) φ3 (x)
1
0≤x≤ 3 3x 0 0
1 2
3 ≤x≤ 3 2 − 3x 3x − 1 0
2
3 ≤x≤1 0 3 − 3x 3x − 2
32 / 39
Outline Finite Differences Difference Equations FDM FEM

Linear Finite Elements: An Example with Hat Functions

φ1 (x) φ2 (x) φ3 (x)
1 0.5 u(x)

u1 u2 u3

0 h 2h 1 0 h 2h 1
Piecewise-linear function u(x) = u1 φ1 (x) + u2 φ2 (x) + u3 φ3 (x)
• Differential equation: −u00 = 1 (i.e. f (x) = 1)
Free-end border conditions u(0) = 0; u0 (0) = 1
• Complete solution of this differential eqution:
2
u(x) = A + Bx − x2 (nullspace + a particular solution);
x2
A = 0 and B = 1 (from the border conditions) ⇒ u(x) = x − 2
• Test functions:
νi : hats φ1 and φ2 and one half-hat φ3 ; h = 31

33 / 39
Outline Finite Differences Difference Equations FDM FEM

FEM Example with Hat Functions (cont.)

Stiffness matrix:
 3  
Z1 6 −3 0
dφi (x) dφj (x) 
K = Kij = dx =  −3 6 −3 
dx dx
0 i,j=1
0 −3 3
dφ1 (x) dφ2 (x) dφ3 (x)
dx dx dx
1
0≤x≤ 3 3 0 0
1 2
3 ≤x≤ 3 −3 3 0
2
3 ≤x≤1 0 −3 3
E.g.
R1  dφ1 (x) 2 1 1 1
K11 = dx dx = 9 · 3 +9· 3 +0· 3 =6
0
R1 dφ1 (x) dφ2 (x) 1 1 1
K12 = dx dx dx =0· 3 −9· 3 +0· 3 = −3
0

34 / 39
Outline Finite Differences Difference Equations FDM FEM

FEM Example with Hat Functions (cont.)

1 1 T
1
Vector f = [f1 , f2 , f3 ]T =

3, 3, 6 :

R1 1/3
R 2/3
R
f1 = φ1 (x)dx = 3xdx + (2 − 3x)dx
0 0 1/3
1/3  2/3
3x2  3x2 1 1 1
= + 2x − = + =

2  2  6 6 3
0 1/3
R1 2/3
R R1
f2 = φ2 (x)dx = (3x − 1)dx + (3 − 3x)dx
0 1/3 2/3
 2/3  1
3x2 3x2 1 1 1
= − x  + 3x − = + =
 
2 2  6 6 3
1/3 2/3
R1 R1
f3 = φ3 (x)dx = (3x − 2)dx
0 2/3
 1
3x2 1
= − 2x  =

2 6
2/3

35 / 39
Outline Finite Differences Difference Equations FDM FEM

FEM Example with Hat Functions (cont.)

Solving the finite element equation Ku = f for the mesh values in u:
    
6 −3 0 u1 1/3
 −3 6 −3   u2  =  1/3 
0 −3 3 u3 1/6
| {z } | {z } | {z }
 K  u f 
u1 5/18  5/18 x = 1/3
x2
⇒  u2  =  4/9  ⇔ u(x) = x − 2 = 4/9 x = 2/3
u3 1/2 1/2 x = 1

| {z }
K−1 f

• All three values u1 , u2 , u3 agree exactly with the true solution of

2
the differential equation u(x) = x − x2 at the mesh points
• Values ui by the finite differences were not exact for this equation
36 / 39
Outline Finite Differences Difference Equations FDM FEM

Finite Element Methods Vs. Finite Differences

FEM: two- and three-dimensional meshes of arbitrary geometry,

compared to the inflexibility of a finite difference grid
• Using numerical integration for K and f , FEM allows for any
functions f (x) and c(x)
 in solving
 the differential equation
d du(x)
dx c(x) dx = f (x)
• Linear finite elements:
• 2nd -order accuracy O(h2 ) in u(x) and
• 1st -order accuracy O(h) in du(x)
dx
• Better accuracy – the higher-degree finite elements
• E.g. piecewise cubic polynomials with continuous slopes
• Adding to the hat functions the bubble functions: quadratic
parabolas staying inside each mesh interval

37 / 39
Outline Finite Differences Difference Equations FDM FEM

More Accurate Finite Elements (optional)

Hat + bubble functions → piecewise linear slope u0 (x):

u(x) = u1 φ1 (x)+u2 φ2 (x)+u3 φ3 (x)+u4 φ4 (x)+u5 φ5 (x)+u6 φ6 (x)

u(0.5) = u5 + (u1 + u2 )/2

u(x)
φ4 φ1 φ5 φ2 φ6 φ3 u0 = u2 −u1
1 h

u1 u2 u3

0 h 2h 1 0 h 0.5 2h 1

• Expected accuracy: 3rd -order O(h3 ) in u(x) and

2nd -order O(h2 ) in u0 (x)
• This higher accuracy – without a very fine mesh required by
only linear elements!

38 / 39
Outline Finite Differences Difference Equations FDM FEM

Numerical Linear Algebra

• Applied Mathematics (AM) Vs. Computational Science (CS):

• (AM) Stating a problem and building an equation to describe it
• (CS) Solving that equation using mostly numerical methods
• Numerical linear algebra represents this “build up, break down”
process in its clearest form, with matrix models like Ku = f
• Crucial properties of K:
The algorithm becomes
• symmetric or not
• sparse or not
clearest when it is seen as
• banded or not a factorisation into:
• well conditioned or not triangular matrices or
• Often the computations
orthogonal matrices or
very sparse matrices
break K into simpler pieces

39 / 39