Basics on

Finite Element Method

(FEM)
Laboratori de Clcul Numric (LaCN)
Departament de Matemtica Aplicada III
Universitat Politcnica de Catalunya (Barcelona)
http://www-lacan.upc.es
 Weighted residuals:
strong form  weak form

Model problem

with

STEP 1. Multiply by a test function v such that v=0 en d

2
STEP 2. Integration by parts

Integration by parts formula in several dimensions:

(
)

STEP 3. Apply boundary conditions (Neumann BC and v=0 in d)

3
 Weak form

Find uH1() such that u=ud in d and

for all vH1() such that v=0 en d, with

Bilinear,
symmetric
& coercive

The equivalence between strong form and weak form can

be easily proved.

4
 Piece-wise (element-by-element)
polynomial approximation

In the FEM the solution is approximated with a piece-wise
polynomial basis

Ni(x)

Advantages:
compact suport (local basis) sparse matrices
easy computation of integrals
physical meaning of coefficients ui
5
 Prescribed values

The coefficients corresponding to values known by boundary
conditions are set to the prescrived value

uh(x) satisfies (with interpolation error) the Dirichlet

boundary condition u=ud on d
Ni(x)=0 on d for iB (condition for test function v)

Other techniques: Lagrange multipliers, penalty method,

Nitsche method...

6
 Discretization of the weak form
Imposing the weak form for v=Ni(x) with iB, and replacing
the approximation uh(x)

Linear system of equations

7
 1D example (with C0 linear approximation)

Approximation:

8

Replacing the approximation and v=Ni for i=1...5

or, equivalently,

Linear system 55:

9

The matrix of the system is tridiagonal (sparse in general,
with few non-null coefficients)

10
 Symmetric and diagonally dominant matrix:
The matrix is symmetric and positive definite

If the bilinear form a(,) is symmetric and coercive, the
matrix is symmetric and positive definite.

The coefficient (i,j) of the matrix is non-null only if nodes i
and j belong to the same elements: sparse matrices

11
 Computation of integrals: numerical
quadrature in each element

We want to compute integrals as

with element-by-element piece-wise polynomial functions.

Gauss quadrature in each element.

 Elemental matrices

Assembly of elemental matrices and vectors

The elemental matrix Ke has all the non-null integrals in the

element e

where () denotes the local numbering and nnode is the

number of nodes in the element. The connectivity matrix
gives the equivalence between local numbering and global
numbering.
 Example

Mesh geometry
definition

K=

(connectivity
matrix)
 (#) local numbering

K1= K=

K2= K=
K3= K=
 Symmetric and semi-
positive matrix
(positiveness after
imposing boundary
conditions)

K4= K=
Computation of the elemental matrix

Shape functions
Ni(x)=?
(x3,y3) (x2,y2)

Numerical quadrature

e
(x4,y4) Reference
(x1,y1) element
Q1: bilinear, {1,x,y,xy}

Transformacin
isoparamtrica

Q2: biquadratic
 Q1 element
N1 N2

N3 N4
 Q2 element
N1 N2

N8 N9

Triangles:
P1

P1: linear, {1, x, y}

P2: quadratic, {1, x, y, xy,

P2
x2, y2}

P3
 Tetrahedra {1, x, y, z}

Hexahedra {1, x, y, z, xy, xz, yz, xyz}

FIN

26