Lesson 5

Method of Weighted Residuals

Classical Solution Technique
The fundamental problem in calculus of variations is to obtain
a function f(x) such that small variations in the function
of(x) will not change the original function

The variational function can be written in general form for a
second-order governing equation (no first derivatives) as



Where o,|, and are prescribed values
2
2
V
1 df
J(f ) f 2 f dV
2 dx
(
| |
= o +|
(
|
\ .
(

}
Classical solution (continued)
An equation containing first-order derivatives may
not have a corresponding variational function. In
some cases, a pseudovariational function can be
used



where C=C(x)
2
2
r
V
1 dC dC
J(C) D Cu K C 2mC dV
2 dx dx
(
| |
= + +
(
|
\ .
(

}
Classical solution (continued)
Example:
Consider a rod of length L. The equation defining heat transfer in the rod is



with boundary conditions


Integrating twice, one obtains


Applying boundary conditions, the final result is

2
2
d T Q
dx k
=
T(0) T(L) 0 = =
2
1 2
Qx
T C x C
2k
= + +
2
Q(Lx x )
T
2k

=
Rayleigh-Ritz Method
FEM variational approach attributed to Lord Rayleigh
(1842-1919) & Walter Ritz (1878-1909)

Let

Assume a quadratic function

with boundary conditions


n
i 1
i i
i 1
T(x) C x (where C are the unknowns)

=
=

2
1 2 3
T C C x C x = + +
1
2 3
T(0) 0 C 0
T(L) 0 C C L
= =
= =
R-R Method (continued)
Hence,



Now integrate



Thus
2
3
T C (x Lx) =
3
dT
C (2x L)
dx
=
2
2
V
1 df
J f 2 f dV (let f T, k, 0, Q)
2 dx
(
| |
= o +| o |
(
|
\ .
(

}
2 2 2
3 3
V
1
J kC (2x L) 2QC (x Lx) Adx
2
( =

}
R-R Method (continued)
which becomes


To find the value of C
3
that makes J a minimum,



Therefore,


2 3 3
3 3
AkC L AC QL
J
6 6
= +
3 3
3
3
2AkC L J AQL
0
C 6 6
c
= + =
c
2
3
Q Q(Lx x )
C or T
2k 2k

= =
Variation Methods
Given a function u(x), the following constraints must
be met
(1) satisfies the constraints u(x
1
)=u
1
and
u(x
2
)=u
2
(2) is twice differentiable in x
1
<x<x
2
(3) minimizes the integral

2
1
x
x
du
J F x, u, dx
dx
| |
=
|
\ .
}
Variational methods (continued)
Then it can be shown that u(x) is also the solution of
the Euler-Lagrange equation



where

(1)
d F F
0
dx u u
c c
| |
=
|
c c
\ .
i
(i)
i
d u
u
dx
=
Variational methods (continued)
For higher derivatives of u,


Hence,


( )
2
1
x
(1) (2) (n)
x
J F x, u, u , u ,..., u dx =
}
2 n
n
(1) 2 (2) n (n)
F d F d F d F
( 1) 0
u dx u dx u dx u
c c c c
| | | | | |
+ =
| | |
c c c c
\ . \ . \ .
Variational methods (continued)
In 2-D, the constraints are
(1) satisfies the constraint u = u
0
on I
(2) is twice differentiable in domain A(x,y)
(3) minimizes the functional


and

u u
J F x, y, u, , d
x y
O
| | c c
= O
|
c c
\ .
}
F F F
0
u x ( u / x) y ( u / y)
| | | | c c c c c
=
| |
c c c c c c c c c
\ . \ .
Variational methods (continued)
Example:
Find the functional statement for the 2-D heat
diffusion equation


Applying the Euler-Lagrange relation


x y
T T
k k Q 0
x x y y
| | c c c c
| |
+ + =
| |
c c c c
\ .
\ .
x y
F d F d F
0
u dx u dy u
| |
| |
c c c
=
|
|
|
c c c
\ .
\ .
Variational methods (continued)
we find that







which yields the final functional form

x y
x y
2
2
y
x
1 2 3
2
2
y
x
F T F T F
k , k , Q
u x u y u
k
k dT dT
F C , F C , F QT C
2 dx 2 dy
k
k dT dT
J(T) QT d
2 dx 2 dy
O
c c c c c

c c c c c
| |
| |
= + = + = +
| |
\ .
\ .
(
| |
| |
= + O (
| |
\ .
( \ .

}
A Rayleigh-Ritz Example
Begin with the equation


with boundary conditions


First find the variational statement (J)

2
2
d u
u x 0 0 x 1
dx
+ + = s s
u(0) u(1) 0 = =
R-R example (continued)
The variational statement is



Assume a quadratic approximation


with boundary conditions

1
2
2
0
du du
J x, u, u 2xu dx
dx dx
(
| | | |
=
(
| |
\ . \ .
(

}
2
0 1 2
u(x) a a x a x = + +
0
1 2
u(0) a 0
u(1) a a 0
= =
= + =
R-R example (continued)
Thus,



Now,


To be a minimum,


Finally




1 1 1
1
u(x) a [x(1 x)] a (x)
du
a (1 2x)
dx
= |
=
1
2 2 2 2 2 2
1 1 1 1
0
J(a ) a (1 2x) a x (1 x) 2a x (1 x) dx ( =
}
1
1
1
J 3a 1 5
0 a
a 10 6 18
c
= = =
c
5
u(x) x(1 x)
18
=
The Weak Statement
Method of Weighted Residuals - one does not
need a strong mathematical background to use
FEM. However, one must be able to integrate.
To illustrate the MWR, let us begin with a simple
example - Conduction of heat in a rod of length L
with source term Q.


2
2
L
d T
K Q 0 x L
dx
dT
K q for x 0
dx
T T for x L
= < <
= =
= =
weak statement (continued)
Integrating,


or


This analytical solution serves as a useful benchmark
for verifying the numerical approach.
( ) ( ) ( )
L y
L
x 0
q 1
T x T L x Q z dz dy
K K
(
(

= + +
} }
( ) ( )
( )
2 2
L
q Q
T x T L x L x
K 2K
= + +
weak statement (continued)
There are basically two ways to numerically solve
this equation using the FEM: Rayleigh Ritz
Method and the Galerkin Method ( which
produces a weak statement)

Consider

T
2 0
2
0
A f 0 in Bu=g on
A B= , ,
x x x x
= O I
(
c c c c
+ +
(
c c c c

weak statement (continued)
Since u = c
i
|
i
(x,y) is an approximate function,
substitution into the above equation may not
satisfy the equation. We set the equation equal to
an error (c)

We now introduce a set of weighting functions (test
functions) W
i
, and construct an inner product
(W
i
,c) that is set to zero this forces the error of
the approximate differential equation to zero
(average).
Au f = c
weak statement (continued)
Hence,



Example:


The inner product becomes

( )
i i j j
d A c f d 0
O O
(
| c O | | + O =

} }
( )
2 2
2 2
u u
f x, y
x y
c c
c =
c c
( ) ( )
2 2
i i i
2 2
u u
, dxdy f x, y dxdy 0
x y
| | c c
| c = | c = | + + =
|
c c
\ .
} } } }
weak statement (continued)
Integrating by parts (Green-Gauss Theorem)



Problem: Use Galerkins Method to solve


( )
i i
i i
u u u u
dy dx f x, y dxdy 0
x y x x y y
| | | | c c c| c c| c
| + + + | =
| |
c c c c c c
\ . \ .
} } }
2
2
d u
f 0 0<x<L
dx
u=0 x=0
du
=0 x=L
dx
+ =
weak statement (continued)
The inner product is

or


The weak statement becomes

( )
1
i i
0
, dx 0 | c = | c =
}
( )
2
L
2
i 1 1 1
2
0
d u
f dx 0 where u=c c x 2xL
dx
| |
| + = | =
|
\ .
}
L L
L
1 1 1
1 0 1
0 0
du d dc
dx fdx 0
dx dx dx
| |
| + | =
} }
weak statement (continued)
Since

we obtain



This is the same as the Rayleigh-Ritz Method
but no variational principle is required.

du
0 at x=L
dx
=
( )
1
2
f
c
2
f
u x 2xL
2
=
=
Weighting Function Choices
Galerkin


Least Squares


Method of Moments



( )
i i
i
W
, 0
= |
| c =
( )
i
, which is the square of the error
c
c
c c
c
( )
i
i
x , 0 i=0,1,2,
W any set of linearly independent functions
c =
=
Weighting Function Choices (continued)
Collocation


Sub domain

( )
( ) ( )
i
i i
i x
W x x
x x , 0
= o
o c =
i
1 for x in subinterval i
W
0 for x outside i

Least Squares Example

Let


For the previous example problem


( )
2
i i i
, dxdy 2 dxdy 0
c c c
c c cc
(
c c c c =

c c c
} } } }
( )
( )
2
1 1 1
1
2
1
2
u c c x 2xL
u
c 2x 2L
x
u
2c
x

= | =

c
=

c
( )
i i
1
, 2 dx 0
c c
2c f
c cc
c c = c =
c c
c = +
}
Least Squares Example (continued)
Thus,






Some observations:
(1) no integration by parts required not a weak form
(2) the Neumann boundary conditions do not appear naturally
(3) in this case |
i
must satisfy global boundary conditions,
which is difficult in most problems

( )
( )
L
1
0
1
2
2 2c f 2dx 0
f
c
2
f
u x 2xL
2
+ =
=
=
}