Pde Fem

Lehrstuhl Informatik V
Scientific Computing I
Module 7: An Introduction to Finite Element Methods
Tobias Neckel
Winter 2015/2016
Tobias Neckel: Scientific Computing I

Module 7: An Introduction to Finite Element Methods, Winter 2015/2016
Part I: Introduction to Finite Element Methods

The Model Problem
FEM Main Ingredients
Weak Forms and Weak Solutions
Test and Shape Functions
Assembling the System of Equations
Example Problem: 1D Poisson
Stiffness Matrix for Linear Nodal Basis Functions
A Road to Theory
Time-Dependent Problems

The Model Problem: Poisson Equation

2D Poisson Equation on unit square:
2
2
u(x,
y)
+
u(x, y ) = f (x, y )
x 2
y 2
in = (0, 1)2
Dirichlet boundary conditions:
u(x, y) = g(x, y)
on


2
2
u(x,
y)
+
u(x, y ) = f (x, y )
x 2
y 2
in = (0, 1)2
u(x, y) = g(x, y)
on
for general operator L instead of : similar concept


2
2
u(x,
y)
+
u(x, y ) = f (x, y )
x 2
y 2
in = (0, 1)2
u(x, y) = g(x, y)
on
for general operator L instead of : similar concept

new for FEM: compute a function to approximate u(x, y )
(instead of approximate values at certain grid points)

Finite Elements Main Ingredients

1. solve weak form of PDE to reduce regularity properties
Z
Z
00
0 0
u = f
v u dx = vf dx
allows additional weak solutions


Z
Z
00
0 0
u = f
v u dx = vf dx
2. compute a function as numerical solution
search in a function space Wh :
X
uh =
uj j (x),
span{1 , . . . , J } = Wh
j


Z
Z
00
0 0
u = f
v u dx = vf dx
2. compute a function as numerical solution
search in a function space Wh :
X
uh =
uj j (x),
span{1 , . . . , J } = Wh
j
3. find weak solutions of simple form:

for example piecewise linear functions and choose basis
functions with local support (hat functions)
leads to system of linear equations
Weak Form and Weak Solutions

The weak form is obtained by 3 steps:
integrate over the domain
from infinetly many equations (strong form has to hold on

every point (x, y)), only one new is obtained



multiply with (any) test function v (from a certain function space
V)
again, infinitely many equations



V)
do integration by parts (i.e. use divergence theorem) to play
derivatives from solution u to test functions v

reduce smoothness assumptions on u



V)
do integration by parts (i.e. use divergence theorem) to play
derivatives from solution u to test functions v

reduce smoothness assumptions on u
real solution u also solves the weak form
however, additional, weak solutions are now allowed

Weak Form of the Poisson Equation

Poisson equation with Dirichlet conditions:
u = f
in , u = 0
on
weak form:
Z
u v d =
f v d
v V

Weak Form of the Poisson Equation

u = f
in , u = 0
on
weak form:
Z
u v d =
f v d
v V
apply divergence theorem:
Z
u v d =
Z
u v d
u v ds
choose functions v such that v = 0 on :
Z
u v d =
f v d
v V

Weak Form of the Poisson Equation (2)

u = f
in , u = 0
on
transformed into weak form:
Z
u v d =
f v d
v V
weaker requirements for a solution u:
twice differentiabale first derivative integrable

hu, v i a bilinear form (i.e. depends on 2 functions u and v );
often written as:

a (u, v ) = hf , v i
v V


search for solution functions uh of the form
uh =
uj j (x)
the j (x) are typically called shape or ansatz functions

the basis functions j (x) build a vector space
(i.e., a function space) Wh

span{1 , . . . , J } = Wh


search for solution functions uh of the form
uh =
uj j (x)
the j (x) are typically called shape or ansatz functions

the basis functions j (x) build a vector space
(i.e., a function space) Wh

span{1 , . . . , J } = Wh
insert into weak formulation
X

Z
uj j (x) v dx = f v dx v V

Test and Shape Functions (2)

choose a basis {i } of the test space Vh
then: if all basis functions i satisfy
X

Z
uj j (x) i dx = f i dx i
then all v Vh satisfy the equation

the {i } are therefore often called test functions


X

Z

we obtain a system of equations for unknowns uj :
one equation for each test function i


X

Z

we obtain a system of equations for unknowns uj :
one equation for each test function i

Vh is often chosen to be identical to Wh (Ritz-Galerkin method)
we then have as many equations as unknowns

Example: Nodal Basis

i (x) :=
1
h (x xi1 )
1
h (xi+1 x)
xi1 < x < xi

xi < x < xi+1
otherwise
0,8
0,6
0,4
0,2
0,2
0,4

0,6
0,8
x
10

linear linear system of equations

Z
j (x) i (x) dx uj = f i dx i
{z
}
|
XZ
j
=:Aij
aim: make matrix A sparse most Aij = 0

11

linear linear system of equations

Z
j (x) i (x) dx uj = f i dx i
{z
}
|
XZ
j
=:Aij
aim: make matrix A sparse most Aij = 0

approach: local basis functions on a discretisation grid
j , j zero everywhere except in grid cells adjacent to grid point
xj
Aij = 0, if i and j dont overlap

11
Example Problem: 1D Poisson

in 1D: u 00 (x) = f (x) on = (0, 1),
hom. Dirichlet boundary cond.: u(0) = u(1) = 0

weak form:
u 0 (x) v 0 (x) dx =
f (x) v (x) dx v V
0
computational grid:
xi = ih, (for i = 1, . . . , n 1); mesh size h = 1/n

Vh = Wh : piecewise linear functions
(on intervals [xi , xi+1 ])

weak form reads:
u 0 (x) 0i (x) dx =
f (x) i (x) dx i = 1, . . . , n 1
0

12
Nodal Basis
i (x) :=
1
h (x xi1 )
1
h (xi+1 x)
xi1 < x < xi

xi < x < xi+1
otherwise
0,8
0,6
0,4
0,2
0,2
0,4

0,6
0,8
x
13
Nodal Basis Derivatives and Integrals

Compute derivative of basis functions:
1
xi1 < x < xi
h
0
1
i (x) :=
h xi < x < xi+1
0
otherwise
Compute integrals of weak form:

Z
Z xi
Z xi+1
1 1
1
1
2
0i (x)0i (x) dx =
dx +
dx =
h
h
h
h
h
xi1
xi

Z
Z xi+1
1
1
1
dx =
0i (x)0i+1 (x) dx =
h h
h
xi
Z
1
0i (x)0i1 (x) dx = (same computation)

h
Z
0
0
i (x)j (x) dx = 0 for all j 6= i, i 1

(supports of basis functions do not overlap!)
14
Nodal Basis System of Equations

stiffness matrix:
1
1
h
1
..
..
. 1
1 2
2
..
.
right hand sides (assume f (x) = R):
i (x) dx = h
f (x) i (x) dx =
0
system of equations very similar to finite differences

note: for other operators L: same approach!
exercise: compute right-hand-side vector, if f (x) :=
fj j (x)

15
Outlook: A Road to Theory
weak formulation is equivalent to variational approach:
solution u minimises an energy functional

16
Outlook: A Road to Theory
weak formulation is equivalent to variational approach:
solution u minimises an energy functional

best approximation property:

u uhFE inf ku v ka
a
v Vh
in terms of the norm induced by the bilinear form a (energy norm)

thus: error bounded by interpolation error
(in energy norm)

(details in lecture Numerical Programming II . . . )

16
Time-Dependent Problems
Example: 1D Heat Equation
ut = uxx + f on domain = [0, 1] for t [0, tend ]
spatial discretisation: weak form
Z
Z
ut v dx =
uxx v dx + f v dx
Z

Z
Z
u v dx
=
uxx v dx + f v dx
spatial discretisation finite elements:
(Mh uh ) = Ah uh + fh
Mh : mass matrix, Ah : stiffness matrix, uh = uh (t)

17
Time-Dependent Problems (2)
Solve a system of ordinary differential equations:

after spatial discretisation (Mh constant):
t Mh
(uh ) = Ah uh + fh
uh a vector of time-dependent functions:
uh = (u1 (t), . . . , ui (t), . . . , un (t))
usually: approximate Mh by a simpler matrix (diagonal matrix,
e.g.) mass lumping

18
Part II: Implementation of Finite Element Methods

Element Stiffness Matrices
Element-Oriented Computation of Stiffness Matrices
Example: 1D Poisson
Example: 2D Poisson
Typical Workflow
Reference Elements and Stiffness Matrices
Element-Oriented Computation on Unstructured Meshes
Accumulation of Global Stiffness Matrix
Simple Example: 1D Poisson
Outlook: Extension to 2D and 3D
Outlook: Further Components and Aspects of FEM

19

Finite Element methods are often used for complicated meshes:
Questions:
how to set up the system of equations efficiently?
is it possible to use stencil notation??

20

Finite Element methods are often used for complicated meshes:
Questions:
how to set up the system of equations efficiently?
is it possible to use stencil notation??
Switch to element stiffness matrices!

20
Element Stiffness Matrices (2)

domain subdivided into finite elements (k ) :
= (1) (2) (n)

observation: basis functions are defined element-wisely
use:
Rb
a
f (x) dx =
Rc
a
f (x) dx +
Rb
c
f (x) dx
element-wise evaluation of the integrals:
Z
u v dx =
XZ
k
Z
f v dx =
XZ
k
u v dx
(k )
f v dx
(k )

21

leads to local stiffness matrices for each element:
Z
j i dx
(k )
{z
(k )
=:Aij
and respective element systems:
A(k) x = b(k)
accumulate to obtain global system:
A(k) x =
b(k )
| {z }
=:A

22

Some comments on notation:
assume: 1D problem, n elements (i.e. intervals)
in each element only two basis functions are non-zero!
(k )
hence, almost all Aij
are zero:
(k )
Aij
Z
j i dx
=
(k)
only 2 2 elements of A(k ) are non-zero

therefore convention to omit zero columns/rows
leaves only unknowns that are in (k)

b (k ) (element stiffness matrix)
notation: write this matrix as A

23
Example: 1D Poisson
= [0, 1] splitted into (k) = [xk , xk +1 ]
nodal basis; leads to element stiffness matrix:
b (k ) =
A
1 1
1 1
consider case with only three unknowns (i.e., 3 basis functions):
A(1) + A(2)
1
= 1
0
1
1
0

0
0 0
0 + 0 1
0
0 1
0
1
1
in stencil notation:
[ 1
1 ] + [ 1 1 ] [ 1
2 1]

24
Example: 2D Poisson
u = f on domain = [0, 1]2
splitted into (i,j) = [xi , xi+1 ] [xj , xj+1 ]
bilinear basis functions
ij (x, y) = i (x)j (y )
pagoda functions

25
Example: 2D Poisson (2)

leads to element stiffness matrix:
12
12
12
12
1
b (k) =
A
21
2
12
12
accumulation leads to 9-point stencil
1
1
1
8
1
1
1
b (k) and assemble global matrix

exercise at home: compute A
(and compare with stencil)
26
Typical Workflow
1. choose grid:
quadratic or cubic cells
triangles (structured, unstructured)
tetrahedra, etc.
2. set up basis functions for each element (k) ;
for example, at all nodes xi (k )
i (xi ) = 1
i (xj ) = 0 for all j 6= i
3. for element stiffness matrix, compute all
Z
(k)
b
Aij =
j i dx
(k)
4. accumulate global stiffness matrix A (and rhs)

27
Element-Oriented Computation on Unstructured

Meshes
Recall: FEM often used for complicated meshes:
(Re-)computation of element matrices required for each element?

28

Meshes

left grid: no (up to scaling with mesh size)
right grid: yes?

28

Meshes

right grid: yes?
requires a closer look at the computation of element matrices

28
Element Stiffness Matrices Notation Revisited

(k )
recall notation for local stiffness matrices Aij :

(k)
Aij
Z
j (x) i (x) dx
=
(k )
which basis functions i , j belong to element k?

switch to element-local numbering of basis functions:
b (k) =
A
(k )
(k)
(x) (x) dx
(k)
(k)
(k)
where (x) = i (x) and = j (x) in (k )

(k)
(k )
requires a mapping i (and
j ),
i.e., a function (k) with i = (k) () and j = (k) ()

29

b (k) :
note the different dimensions of A(k) and A
b (k ) omits zero rows and columns of A(k )
A
formulated via a projection matrix:
(k)
= P

(k) T
b (k ) P (k)
A
with
(k )
P,i

:=
1 if i = (k) ()
0 otherwise
recall computation of global stiffness matrix:
A :=
A(k ) =
P (k )
T
b (k ) P (k )
A

30

b (k) :
note the different dimensions of A(k) and A
b (k ) omits zero rows and columns of A(k )
A
formulated via a projection matrix:
(k)
= P

(k) T
b (k ) P (k)
A
with
(k )
P,i

:=
1 if i = (k) ()
0 otherwise
recall computation of global stiffness matrix:
A :=
A(k ) =
P (k )
T
b (k ) P (k )
A
hence, for each element:
b (k)
1. compute element stiffness matrix A
b (k ) in A(k)
2. determine position of matrix elements A
based on mapping (k) ()
T (k ) (k )
b P to global
3. add non-zero elements of A(k) = P (k ) A
matrix A
30
Element Stiffness Matrices and Meshes

Recall our initial question:
b (k ) required for each element?

(Re-)computation of A

31


right grid: yes?

31


right grid: yes?
b (k) via reference elements

efficient computation of A
31

Consider FEM for 1D Poisson on an adaptive grid:
grid points given as xi , not necessarily equidistant;
define elements (k) = [xk , xk+1 ]
piecewise linear basis functions (hat functions) i (x);
i (x) non-zero only on two elements: (i1) and (i)
mapping local to global indices: (k) (1) := k and (k) (2) := k + 1
R
(k)
(k)
b (k)
wanted: A
(x) (x) dx for every element k
, :=
(k )

32

Consider FEM for 1D Poisson on an adaptive grid:
grid points given as xi , not necessarily equidistant;
define elements (k) = [xk , xk+1 ]
piecewise linear basis functions (hat functions) i (x);
i (x) non-zero only on two elements: (i1) and (i)
mapping local to global indices: (k) (1) := k and (k) (2) := k + 1
R
(k)
(k)
b (k)
wanted: A
(x) (x) dx for every element k
, :=
(k )
Introduce reference element ref with linear basis functions :

1 () :=
2 () := 1
ref := [0, 1]
and a mapping x = (k ) () for each (k) defined as

(k ) () := xk + (xk+1 xk )

(k)
(k )
such that (x) = (k) () := () for x (k) .
32
Simple Example: 1D Poisson (2)

We need to compute
b (k) :=
A
,
xk +1
Z
(k)
(x) (k)
(x) dx =
x
x
xk
(k )
Z
(1)
(k ) (0)
Compare integration by substitution:
g(b)
R
g(a)
b (k)
A
,
(k)
(x) (k)
(x) dx
x
x
Z1
=
f (x) dx =
Rb
f (g(t))g 0 (t) dt
(k) (k) (k ) (k)
()
() (xk +1 xk ) d.
x
x
Note that:

k () = (xk +1 xk ) corresponds to the factor g 0 (t) in the
substitution rule
we still have to deal with the partial derivatives in x
33

We require the chain rule,
f
x
f g
g x ,
to obtain:
(k) (k)
() =
() =
()
x
x
x
x
xk
As x = (k) () := xk + (xk+1 xk ), we have =
and thus:
xk+1 xk
1
()
=
()
xk+1 xk
Finally, we get:
b (k) =
A
,
Z1
(k ) (k) (k) (k )
()
() (xk+1 xk ) d
x
x
Z1
=
1
() ()
(xk+1 xk ) d.
(xk +1 xk )2
0
34

Altogether, element stiffness matrices are computed as:
b (k) =
A
,
1
xk+1 xk
Z1
() () d
With 1 () = and 2 () = 1 , we obtain that

1
1 1
(k)
b
A =
1 1
xk+1 xk

35

Altogether, element stiffness matrices are computed as:
b (k) =
A
,
1
xk+1 xk
Z1
() () d
With 1 () = and 2 () = 1 , we obtain that

1
1 1
(k)
b
A =
1 1
xk+1 xk
Observations:
the integral values only depend on the reference element
ref = [0, 1] and the reference basis functions ,

all other element matrices are obtained by simple scaling
we did not use that , are linear functions;
same computation for more complicated basis functions!

35
Outlook: Extension to 2D and 3D

For 2D, 3D, etc., we require the multidimensional substitution rule:
Z
Z
f (x(, , ), y (, , ), z(, , )) |J | d d d
f (x, y, z) d =
(k )
ref
where |J | is the determinant of the Jacobian:

x x x

y y y
|J | =
z z z

and also the multidimensional chain rule.

Result:
similar, element-wise computation of stiffness matrices via
reference elements
Jacobian instead of factor x 1x ; more complicated, in general
k +1
k
36
Outlook: Further Components and Aspects of FEM

Additional components and features of Finite Element Methods:
mesh generation:
structured/unstructured meshes consisting of triangles,

tetrahedra, rectangles, cuboids, quadrilaterals, hexahedrals,
prisms, . . .
higher-order basis functions:
polynomials of higher degree in various dimensions; nodal vs.

spectral elements, . . .
quadrature rules for integration:
Gaussian quadrature, e.g., to compute integrals of weak forms;

resp. choice of quadrature points for certain elements?
solving the system of equations:
e.g., set up a global system of equations explicitly, or work

directly on element representation?
. . . and many more . . .
37

Pde Fem

Uploaded by

Document Information

Original Description:

Original Title

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

Pde Fem

Uploaded by

Copyright:

Available Formats

Lehrstuhl Informatik V

Tobias Neckel: Scientific Computing I

Part I: Introduction to Finite Element Methods

Tobias Neckel: Scientific Computing I

The Model Problem: Poisson Equation

Dirichlet boundary conditions:

Tobias Neckel: Scientific Computing I

The Model Problem: Poisson Equation

Dirichlet boundary conditions:

for general operator L instead of : similar concept

Tobias Neckel: Scientific Computing I

The Model Problem: Poisson Equation

Dirichlet boundary conditions:

for general operator L instead of : similar concept

(instead of approximate values at certain grid points)

Tobias Neckel: Scientific Computing I

Finite Elements Main Ingredients

Tobias Neckel: Scientific Computing I

Finite Elements Main Ingredients

Tobias Neckel: Scientific Computing I

Finite Elements Main Ingredients

3. find weak solutions of simple form:

Weak Form and Weak Solutions

from infinetly many equations (strong form has to hold on

Tobias Neckel: Scientific Computing I

Weak Form and Weak Solutions

from infinetly many equations (strong form has to hold on

Tobias Neckel: Scientific Computing I

Weak Form and Weak Solutions

from infinetly many equations (strong form has to hold on

derivatives from solution u to test functions v

Tobias Neckel: Scientific Computing I

Weak Form and Weak Solutions

from infinetly many equations (strong form has to hold on

derivatives from solution u to test functions v

Tobias Neckel: Scientific Computing I

Weak Form of the Poisson Equation

Tobias Neckel: Scientific Computing I

Weak Form of the Poisson Equation

apply divergence theorem:

choose functions v such that v = 0 on :

Tobias Neckel: Scientific Computing I

Weak Form of the Poisson Equation (2)

transformed into weak form:

weaker requirements for a solution u:

twice differentiabale first derivative integrable

often written as:

Tobias Neckel: Scientific Computing I

Test and Shape Functions

the j (x) are typically called shape or ansatz functions

(i.e., a function space) Wh

Tobias Neckel: Scientific Computing I

Test and Shape Functions

the j (x) are typically called shape or ansatz functions

(i.e., a function space) Wh

insert into weak formulation

Tobias Neckel: Scientific Computing I

Test and Shape Functions (2)

then all v Vh satisfy the equation

Tobias Neckel: Scientific Computing I

Test and Shape Functions (2)

then all v Vh satisfy the equation