GALERKIN WEIGHTED RESIDUAL

TECHNIQUE
 RESIDUAL MINIMIZATION METHOD
EXAMPLE 1: A uniform rod has been subjected to uniform axial load as shown in Figure. Find
out an appropriate solution to this problem by RMM. The deformation of the bar (u) is governed
by the governing differential equation:
2 q0
d u
AE 2
 qo  0
dx
With the boundary conditions: u 0  0,
du
0 x L
dx x L
Step-1: (Assume the Trial Solution satisfying the boundary conditions)
Let the trial or guess solution to the above differential equation be:

u ( x)  u ( x)  C0  C1 x  C2 x 2
Where the constants C0, C1, C2 are yet to be determined in the trial solution so that the trial
solution satisfies the given boundary conditions.
Considering the first boundary condition: u0  0

u(0)  u (0)  C0  C1 x  C2 x 2  0  C0  0
du
Considering the second boundary condition: 0
 x L dx
du( x) du ( x)
  0  C1  2C2 L  0  C1  2C2 L
dx x  L dx x  L


So the final Trial solution satisfying the boundary conditions is: u ( x)  C2 x  2Lx
2

Since the trial solution contains only one free parameter C2, it is often referred as a One
parameter solution
Step-2: (To find the Domain Residual)
Substituting the Trial solution into the governing differential equation will lead domain
residual indicating the amount of error in the assumed Trial solution.

AE
d 2u ( x)
dx 2
 qo  0 u ( x) 

u ( x)  C
2 x 2
 2Lx
 
d 2u ( x) [As (x ) is not the exact solution of the D.E]
u
AE  q o  0
dx 2

d 2u ( x)
Rd  AE 2
 qo  AE(2C2 )  qo
dx
Step-3: (To minimize the Domain Residual)
The Domain Residual indicates the error through out the domain (Ω) of the given differential
equation. So in order to approximate the Trial Solution with the Exact Solution, the Domain
Residual is tried to be minimized.

d 2u ( x)  qo
Rd  AE 2
 qo  AE(2C2 )  qo  0  C2 
dx 2 AE
So the final Trail Solution to the Governing Differential Equation becomes:
 q

u ( x)  u ( x)  0 2 Lx  x 2
2 AE

For this simple example, as we could make the Domain Residual identically zero every
where in the domain (Ω) , our final Trial Solution matches with the Exact Solution.
 POINT COLLOCATION TECHNIQUE
EXAMPLE 2: Consider the problem of a simply supported beam under uniformly distributed
load q0 as shown in the figure. The governing differential equation is given by:
d 4v
EI 4  qo  0 the boundary conditions :
dx 2 2
 0 vL   0,
d v
v0  0,
d v
0
dx 2 x 0
dx 2 x L

Solution through one-term trial solution:

Step-1: (Assume the Trial Solution satisfying the boundary conditions)
As the Governing D.E is of order four, trial or guess solution for the above differential
equation can be an algebraic trial solution of order four.

v( x)  v ( x)  C0  C1 x  C2 x 2  C3 x 3  C4 x 4
However, we may make the process simple by choosing trigonometric trial functions.

v( x)  v ( x)  C1 sin( x / L)
As per our requirement this single parameter trial solution satisfies all the boundary conditions.
Step-2: (To find the Domain Residual)
Substituting the Trial solution into the governing differential equation will lead domain
residual indicating the amount of error in the assumed Trial solution.
d 4 v( x) 
EI 4
 qo  0 v( x)  v ( x)  C1 sin( x / L)
dx

d 4 v ( x) 
EI 4
 qo  0 [As v (x) is not the exact solution of the D.E]
dx
 
d 4 v ( x) d4
Rd  EI 4
 qo  EI 4 {C1 sin( x / L)}  qo
dx dx

 Rd  C1 ( ) 4 EI sin( x / L)  qo
L
Step-3: (To minimize the Domain Residual)
The Domain Residual indicates the error through out the domain (Ω) of the given differential
equation. So in order to approximate the Trial Solution with the Exact Solution, the Domain
Residual is tried to be minimized.

Rd  C1 ( ) 4 EI sin( x / L)  qo  0
L
qo
 C1  (Which is a function of “x”)
 4
( ) EI sin( x / L)
L
As magnitude of the domain residual depends on the value of C1 which is again a function of
“x’ so domain residual will be varying from point to point within the domain (0< x < L).
So in this case we can not ensure the domain residual to become zero through out the domain,
rather we will try to make it zero at some selected points.
The number of points in the domain about which the domain residual is to be made zero,
are to be decided based on the number of unknown constants we have in the Trial Solution.

v( x)  v ( x)  C1 sin( x / L)
In the present Trial solution we have only one unknown constant (C1) ; hence in order to decide
this constant we have to make the Rd zero at any one point in the domain selectively.
This point can be any random point in the domain.
 This technique is called the Point Collocation Technique where in the residual (In general a
function of “x’) is set to zero at chosen points within the domain. The number of points being
equal to the number of coefficients in the trial function that needed to be determined.
Let us make the domain residual to be zero at some specific locations within the domain and
check the accuracy of the Trial Solution with respect to the Exact Solution.
 2q0 L4
(i) Rd = 0 at x = L/4: Rd  {C1 ( ) 4 EI sin( x / L)  qo }  0  C1  4
L xL / 4  EI
So for making Rd = 0 at x = L/4
 2q0 L4
v ( x)  4 sin( x / L) Rd  2q0 sin( x / L)  qo
 EI
 q0 L4
(ii) Rd = 0 at x = L/2: Rd  {C1 ( ) EI sin( x / L)  qo }
4
 0  C1  4
L xL / 2  EI
So for making Rd = 0 at x = L/2
 q0 L4
v ( x)  4 sin( x / L) Rd  q0 sin( x / L)  qo
 EI
 2q L 4
(iii) Rd = 0 at x = L/3 Rd  {C1 ( ) EI sin( x / L)  qo }  0  C1 
4 0
L xL / 3 3  4
EI
So for making Rd = 0 at x = L/3

 2q0 L4 2
v ( x)  sin( x / L) Rd  q0 sin( x / L)  qo
3 EI
4
3
 Let us now compare the deformation results (v (x)) obtained by the different Trial Solutions
obtained by making the domain residual to vanish at specific points within the domain (x = L/4,
L/2, L/3), with respect to the Exact Solution.
q0 x 3
v( x)  ( L  2 Lx 2  x 3 ) Exact Solution
24EI
 q0 L4
v ( x)  4 sin( x / L) Trial Solution by making Rd = 0 at x = L/2
 EI
 2q0 L4
v ( x)  sin( x / L) Trial Solution by making Rd = 0 at x = L/3
3 EI
4

 2q0 L4
v ( x)  4 sin( x / L) Trial Solution by making Rd = 0 at x = L/4
 EI
Let us compare the results for a specific case where:
The uniformly distributed load = q0 = 1000 N/mm
Young’s Modulus of Elasticity = E = 200,000 MPa
Moment of Inertia of the beam CS = I = 4 x 106 mm4
Length of the beam = L = 1000 mm
 0.2

0.18 Deflection (v) EXACT L/2 L/3 L/4

0.16

0.14

0.12

0.1

0.08

0.06

0.04

0.02

Normalized axis location on the beam (x/L)

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

It may be noted that each of the Trial solution deviates appreciably from the exact solution.
In a similar manner the Residual results obtained for different Trail Solutions may also be
compared with respect to each other.
Rd  q0 sin( x / L)  qo For Trial Solution obtained by making Rd = 0 at x = L/2
2
Rd  q0 sin( x / L)  qo For Trial Solution obtained by making Rd = 0 at x = L/3
3
Rd  2q0 sin( x / L)  qo For Trial Solution obtained by making Rd = 0 at x = L/4
600

400
Residual (Rd)

200

0
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1

-200

-400

RESIDUAL(L/2)
-600
RESIDUAL (L/3)
RESIDUAL (L/4)
-800

-1000

Normalized axis location on the beam (x/L)

-1200
Solution through two-term trial solution:
We may improve the Trial solution by adding one more term to the sine series.

v( x)  v ( x)  C1 sin( x / L)  C3 sin( 3x / L)
It may be verified that this double parameter trial solution also satisfies all the boundary
conditions.
Substituting this modified two term Trial solution into the governing differential equation will
lead domain residual indicating the amount of error in the assumed Trial solution.

d 4 v ( x) d4
Rd  EI 4
 qo  EI 4 {C1 sin( x / L)  C3 sin( 3x / L)}  qo
dx dx
 4 3 4
 Rd  C1 ( ) ( EI ) sin( x / L)  C3 ( ) ( EI ) sin( 3x / L)}  qo
L L
Since we have two constants to be determined, we can set the residual to be zero at two
selected points within the domain.
 4 3 4
Let Rd = 0 at x = L/4 C ( ) ( EI )(1 / 2 )  C ( ) ( EI )(1 / 2 )  q  0
1 3 o
L L
 3
Let Rd = 0 at x = L/3 C1 ( ) 4 ( EI )( 3 / 2)  C3 ( ) 4 ( EI )(0)  qo  0
L L
Solving the above two equations:
2q0 L4 q0 L4
C1  C3  0.00003289
3 4 EI EI
 The final Trail Solution and Residual expressions are:
 2q0 L4 q0 L4
v( x)  v ( x)  sin( x / L)  0.00003289 sin( 3x / L)
3 EI
4
EI
2q0 q0 L4 3 4
 Rd  sin( x / L)  (0.00003289 )( ) ( EI ) sin( 3x / L)}  qo
3 EI L
0.18 200

0.16
0
0 0.2 0.4 0.6 0.8 1
0.14
-200
0.12

0.1 -400
EXACT
0.08 TWO TERM
-600
0.06
-800
0.04

0.02 -1000

0
-1200
0 0.2 0.4 0.6 0.8 1

It may be observed that this approximates the solution perhaps better than the earlier solution
(with one term trial solution), but still deviates appreciably from the exact solution.
Rather than trying to set residual exactly zero at a few selected points and having no
control over the residual at all other points in the domain, we will try to minimize the
residual in an overall sense, this technique is called Weighted Residual Technique.
SOLUTION BY GALERKIN WEIGHTED RESIDUAL TECHNIQUE:
Rather than trying to set residual exactly zero at a few selected points and having no control
over the residual at all other points in the domain, we will try to minimize the residual in an
overall sense, this technique is called Weighted Residual Technique.
L
For this purpose we will formulate this technique as:  W ( x) R ( x)dx  0
0
i d

Where, Wi(x) are appropriately chosen weighting functions which help to achieve the task of
minimizing the residual (Rd) over the entire domain (Ω).
We may choose as many weighting functions as necessary to generate the required number of
equations for the solution of the undetermined coefficients in the trial function.
For example, for a two-term trial solution, we would take two different weighting functions to
generate two equations necessary to solve for the two coefficients.
While the choice of Weighting function Wi(x) is entirely arbitrary, Galerkin (1915) introduced
the idea of letting Wi(x) to be same as the trial functions themselves.
EXAMPLE 3: Consider the problem of a simply supported beam under uniformly distributed
load q0 as shown in the figure. The governing differential equation is given by:
d 4v
EI 4  qo  0 the boundary conditions :
dx 2 2
 0 vL   0,
d v
v0  0,
d v
0
dx 2 x 0
dx 2 x L

Solve it using Galerkin weighted residual technique

Step-1: (Assume the Trial Solution satisfying the boundary conditions)
As the Governing D.E is of order four, trial or guess solution for the above differential
equation can be an algebraic trial solution of order four.

v( x)  v ( x)  C0  C1 x  C2 x 2  C3 x 3  C4 x 4
However, we may make the process simple by choosing trigonometric trial functions.

v( x)  v ( x)  C1 sin( x / L)
As per our requirement this single parameter trial solution satisfies all the boundary conditions.
Step-2: (To find the Domain Residual)
Substituting the Trial solution into the governing differential equation will lead domain
residual indicating the amount of error in the assumed Trial solution.
d 4 v( x) 
EI 4
 q o  0 v ( x)  v ( x)  C1 sin( x / L)
dx

d 4 v ( x) 
EI  q  0 [As v (x) is not the exact solution of the D.E]
dx 4 o

d 4 v ( x) d4
Rd  EI 4
 qo  EI 4 {C1 sin( x / L)}  qo
dx dx
 4
 Rd  C1 ( ) EI sin( x / L)  qo
L
Step-3: (To minimize the Domain Residual)
As per the WRT in order to determine the value of the constant (C1) so that the residual would
be minimized through out the domain in an overall sense we have to multiply the Residual (Rd)
with the weight function Wi(x) and integrate it over the entire domain.
 L

0  EI sin( x / L)  q 0 ]dx  0
4
{sin( x / L )}[C1 ( )
L

Weighting function (Wi (x)) Residual (Rd)

Only one weighting function Wi(x) has been considered as only one constant is to be determined
in the Residual function.
 L L
x
 ( ) 4 EIC1  sin 2 (x / L)dx   q 0 sin(x / L)dx   , dx 
L
d
L 0 0 L 
 
 3 q L x  0   0
 ( ) EIC1  sin 2   d  0  sin   d
L 0
 0 x  L  

 3 (1  cos 2 ) q0 L
 ( ) EIC1 

 d  [ cos ]0
L 0
2 
 q0 L
 ( ) EIC1{[ / 2]  [sin 2 ] }  
1 

3
[cos   cos 0]
L 0 4 0 
 3  2q 0 L
4
q0 L 4 q 0 L4
 ( ) EIC1   C1  ( 5 )  0.01307
L 2  EI  EI
So the final trial solution becomes:
 q 0 L4 4 q 0 L4
v ( x)  ( 5 ) sin( x / L)  0.01307 sin( x / L)
EI  EI
X/L V (TWO-TERM) V(WRM) V (EXACT)
0 0 0 0
0.1 0.046192087 0.050461 0.051094
0.2 0.087625771 0.095987 0.096667
0.3 0.120206753 0.132127 0.132344
0.4 0.140940265 0.155347 0.155
0.5 0.148066378 0.163375 0.16276
0.6 0.141011715 0.155427 0.155
0.7 0.120343819 0.13228 0.132344
0.8 0.087816406 0.096198 0.096667
0.9 0.046418091 0.050708 0.051094
1 0.000238438 0.00026 0

X/L RD (TWO-TERM) RD (WRM) RD (EXACT)

0 -1000 -1000 0
0.1 -433.9057459 -607.568727 0
0.2 -75.19995658 -253.512744 0
0.3 14.2303864 27.54532591 0
0.4 -53.86642588 208.1212177 0
0.5 -104.279839 270.5566751 0
0.6 -54.29913946 208.7462247 0
0.7 14.13367076 28.7342214 0
0.8 -74.09672412 -251.87622 0
0.9 -431.4316694 -605.644609 0
1 -996.9235677 -997.976444 0