1 TRIGONOMETRIC APPROXIMATION

Finite Element Methods

Module 5

Part 1:

In the last class, we discussed the basic concepts of Weighted residuals method. Then we
have defined the residual term. Next we have discussed the steps of weighted residuals
method. Finally the trial functions are defined and how to obtain the approximate solutions
is discussed.

Recalling required equations from previous class

n
X
ũ(x) = ci Ni (x) = c1 N1 (x) + c2 N2 (x) + ... + cn Nn (x). (0.1a)
i=1

ũ(x) = CT NT = NC. (0.1b)

L (ũ) + f = R, (0.2)

1 Trigonometric Approximation

Another often used set of trial functions is trigonometric approximation based on the
Fourier series. An example of Fourier sine series obtained by taking

kπx
Nk (x) = sin .
L

For a one dimensional problem

n
X iπx
ũ(x) = ci sin .
i=1
L

Since this expansion needs to satisfy the zero boundary conditions, and we notice that
iπx
sin = 0 at x = 0 and x = L.
L

1
 2 WEIGHT FUNCTIONS

There is not much of a restriction, as one can always make the change of variables so
that the boundary conditions become homogeneous.

Remark: With the selection of ũ(x) as the series expansion (0.1), it is evident that the
residual R depends on the unknown parameters ci ’s in the expansion:

R = R(x; C).

If the number of trial functions n is sufficiently large, then in principle, the unknown
parameters ci ’s can be chosen so that the residual R is small over the domain.

Part 2:

2 Weight Functions

In general, the weight function w(x) may be written as

w(x) = = a1 w1 + a2 w2 + ... + an wn = aw
Pn
i=1 ai wi

where a and w are row and column vector given respectively by

 
 w1 
 
  w2 
a= and w = 
  .
a1 a2 .... an
 ... 
 
 wn 
 

Here, wi ’s are known functions of x and ai ’s are constant parameters. Substituting

w(x) = aw in the weighted residual equation, we get
Z
a wRdD = 0.
D

Since a is a constant vector, we have

Z
wRdD = 0. (2.1a)
D

2
 2 WEIGHT FUNCTIONS

or
Z
w1 R(x, ci )dD = 0
ZD
w2 R(x, ci )dD = 0
D

··· (2.1b)

···

···
Z
wn R(x, ci )dD = 0
D

Now we have n equations to determine unknown coefficients ci ’s. Inserting ũ = NC

in equation (0.2), we obtain

R = L (NC) + f = L (N)C + f

and hence the condition (2.1a) becomes

"Z # Z
wL (N)dD C = − w f dD.
D D

Introducing matrix K and f as

Z Z
K= wL (N)dD, f=− w f dD
D D

allows us to write the above equation (2.1) in compact form as

KC = f,

which may be expanded as

 D w1 L (N1 ) dD D w1 L (N2 ) dD L
 R R R     R 
··· D
w 1 (Nn ) dD c1  w 1 f dD

 
    R D
 

··· ··· ··· ··· c2   w 2 f dD
..  = − 
     D 
   .. 
 ··· ··· ··· ···   .   . 
  R
 D wn L (N1 ) dD D wn L (N2 ) dD L
 R R R   
··· w
D n
(N n ) dD cn   w f dD
    D n 

3
 3 COLLOCATION METHOD

The system of equations above in the matrix form, which can be solved for n unknown
coefficients ci ’s provided that a suitable weight function w is selected.

There are (at least) five MWR sub-methods, according to the choices for the Wi ’s.
These five methods are:

1. Collocation method

2. Subdomain method

3. Least-squares method

4. Galerkin method

5. Method of moments etc.

Part 3:

3 Collocation method

3.1 Point collocation method

In the point collocation method, the weight function is selected in such a way that the
residual can be set equal to zero at n distinct points in the domain. This can be achieved by
choosing weight function as the displaced Dirac delta functions. So, for a one-dimensional
case,

 ∞, if x = x i




wi = δ(x − x i ) = 


 0, else,


where the fixed points x i ∈ [ a, b] , (i = 1, 2, · · · , n) are called collocation points. The
number of collocation points selected must be equal to the number of unknown coefficients
ci ’s in the definition of the approximating function. The residual R(x) is forced to be zero
at n collocation points. It can be shown that the point collocation method is equivalent to
the classical finite difference method.

4
3.2 Examples of point collocation method 3 COLLOCATION METHOD

3.2 Examples of point collocation method

Example 3.1. Let us illustrate the application of the point collocation method using a
simple physical problem. We consider a simply supported beam subjected to concentrated
moments at both ends. The problem is governed by the following differential equation
2
E ddxy2 − M0 = 0, x ∈ [0, L ] ,

with boundary conditions y(0) = 0 & y(L) = 0.

Here, the coefficient E represents the resistance of the beam to deflection, M0 is the
applied moment, and L is the length of the beam. The analytical solution of the problem
in the interval [0, L ] is

M0
y(x) = − 2E x(L − x).

The negative sign in the expression shows that the displacement is negative for positive
values of bending moment M0 .

Polynomial approximation to deflection curve: Here we select a second degree poly-

nomial of the form

u(x) = a + bx + cx 2 .

The function which satisfies the prescribed boundary conditions can be obtained by the
application of boundary conditions to the chosen approximating function. Thus, we have

u(x) = cx(x − L).

The second derivative

d2u
dx 2
= 2c.

The residual R is then given by R = E × 2c − M0 . Here R(x) is independent of x, so

the residual can be set to zero at every point in the interval automatically. Therefore,
E × 2c − M0 = 0.

Solving for the unknown coefficient c, we get

5
3.2 Examples of point collocation method 3 COLLOCATION METHOD

c= M0
2E .

Thus, the approximate solution is

M0
u(x) = − 2E x(L − x).

It may be noted that the selection of a second degree polynomial yields an exact solution
since the selected polynomial represents the exact behaviour of the deflection curve.