Professional Documents
Culture Documents
and
Approximate Solution of Differential
Equations
Jayadeep U. B.
M.E.D., NIT Calicut
Ref.: Finite Elements and Approximation, Zienkiewicz,
O. C., and Morgan, K., John Wiley & Sons.
Introduction
Why Function Approximation?
Exact variation of the primary variable is not known e.g.,
the temperature profile within an engine block is an unknown
function of spatial coordinates (and possibly of time).
We need the information of variation of field variable to
convert the P.D.E. into a system of algebraic equations, which
can be solved by using computers.
It is preferable to have simplified functions (say polynomials)
representing the field variables Integration, differentiation
etc. can be performed easily.
Before using approximations to solve differential equations,
we have to use them on known functions to ensure
convergence Calibration of the Numerical Procedure.
Lecture - 01
)d W R d 0 for l 1, 2,..., M
W
(
x=0
x=0
Quadratic Functions
Similar to the Second
Moments of functions
W=const
x=0
x=l
W=
1 -x W = x
x=l x
W1 = 1-x2
W2 = (1+x)
(1-x) 2
W3 = x
x=l x
; am N m
m 1
at & N m 0 at
N m - Independent Trial (Basis or Shape) Functions
am - The unknown parameters
Lecture - 01
=x
=x3
N1 = sin(x)
N2 = sin(2x)
)d W ( a N )d 0 for l 1, 2,..., M
W
(
m m
l
l
m 1
W N
l
m 1
a d Wl ( )d for l 1, 2,..., M
m m
a1 Wl N1d a2 Wl N 2 d L aM Wl N M d
Wl ( ) d for l 1, 2,..., M
K11 W1 N1d
Lecture - 01
f1 W1 ( )d
)d W ( a N )d 0 for l 1, 2,..., M
W
(
m m
l
l
m 1
[ K ] a f
(a1 , a2 ,..., aM )
K lm Wl N m d for l , m 1, 2,..., M
f l Wl ( )d for l 1, 2,..., M
Lecture - 01
3
x
x
x a1 sin( x) a2 sin(2 x) dx 0
0
3
x
a
sin(
x
)
a
sin(2
x
)
dx
x
x
2
1
x dx
1
3
(1
x
)
a
sin(
x
)
a
sin(2
x
)
dx
(1
x
)
x
x dx
1
2
Lecture - 01
x sin( x)dx
x sin(2 x)dx
0
0
a 1
1
1
a 2
(1
x
)
sin(
x
)
dx
(1
x
)
sin(2
x
)
dx
0
0
In general:
Lecture - 01
K a f
x x x dx
1
3
(1
x
)
x
x
dx
I (a1 , a1 ,K , an ) ( ) d ( am N m ) 2 d
2
m 1
I
0
al
for 1 l M
Substituting:
M
I
2( am N m ) N l d 0
al
m 1
) d 0 for 1 l M
N
(
Lecture - 01
Lecture - 01
W ( )d N ( a
l
m 1
N m )d 0 for l 1, 2,..., M
[ K ] a f
a (a1 , a2 ,..., aM )
T
K lm N l N m d
for l , m 1, 2,..., M
f l Wl ( )d
for l 1, 2,..., M
m x
N m sin
for m 1, 2,..., M
Nl ( $)d
l x
sin
x0 L
L
m x
a
sin
dx 0 for l 1, 2,..., M
m 1
(a1 , a2 ,..., aM )
K lm N l N m d
Lecture - 02
x=0
sin
l x
( )d
L
f l sin
l x m x
sin
dx for l , m 1, 2,..., M
L
L
for l 1, 2,..., M
L
for l m
l x m x
2
K lm sin
sin
dx
L
L
x 0
0 for l m
L
2
m x
am sin
dx 0 for m 1, 2,..., M
L x 0
L
L
Lecture - 02
L p 0,
where L is a linear differential operator and
p is a function independent of .
For example, consider the 2D steady state heat transfer
equation, assuming constant conductivity:
2 2 Q
2 0
2
x
y
k
2 2
2
2
L 2 2 or L 2 2
x
y
x y
Q
& p
k
Lecture - 02
M r 0 on ,
where M is a linear differential operator and
r is a function independent of .
M or M 1
& r
q 0 on q
x
M k
or M k
x
x
& rq
Lecture - 02
$
am N m p d 0 for 1 l M
Wl (L p)d Wl L
m 1
am N m d Wl L p d for 1 l M
Wl L
m 1
a W L N d W L d W pd
m 1
Lecture - 02
for 1 l M
K a f
K is not symmetric,
even for the Galerkin
Method.
K lm Wl L ( N m )d
(a1 , a2 ,K , aM )
fl Wl L( ) d Wl p d
Lecture - 02
; am N m
m 1
$ p)d W l (M$ r )d
W
(
L
am N m p d Wl M
Wl L
m 1
Lecture - 02
am N m
r d 0
m 1
for 1 l M
M
K a f
K lm Wl L ( N m )d Wl M ( N m )d
a (a1 , a2 ,K , aM )
T
f l Wl p d Wl r d
Lecture - 02
Lecture - 02
2 Q
L p 2 + 0
x
k
on 0 x L
Points (x 0) & (x L)
Let 0 on ( x 0)
and
q
0 on q ( x L)
x k
W L$ p dx W M$ r
L
2 $ Q
Wl
dx Wl $
2
x
k
0
Lecture - 03
W M$ r
l
x 0
x 0
x L
q
Wl
x k
0
0
xL
dx
l
0 x 2 k l
x 0
q
Nl
x k
0
xL
; $ am N m , with & N m 0; 1 m M on
m 1
M
M
2
Q
q
am N m dx N l
am N m
0 Nl x 2
k
k
m 1
m 1
x
L
Lecture - 03
0
xL
2 M
am N m dx N l
0 Nl x 2
x
m 1
am N m
m 1
M
xL
q
2 Q
Nl
dx N l
2
k
x
0
x k
L
x L
Nl
x
am N m
m 1
M
xL
x 0
N l
am N m dx N l
m 1
am N m
m 1
xL
q
2 Q
Nl
dx N l
2
k
x
0
x k x L
Lecture - 03
Nl
x
am N m
m 1
M
Nl
N l
0
xL
am N m
m 1
am N m dx
m 1
q
2 Q
Nl
dx N l
2
k
x
0
x k
L
xL
xL
N l
0 x x
q
2 Q
am N m dx N l
dx N l
2
k
m 1
x
0
x k
M
xL
K a f
L
N l N m
K lm
dx
x x
0
(a1 , a2 ,K , aM )
q
2 Q
fl Nl
dx N l
2
k
x
0
x k
L
xL
Concluding Remarks
Starting with various approximation methods for functions,
we have seen how it can be used for approximate solution of
differential equations.
Function approximation methods (most of them) can be
considered as special forms of Weighted Residual Statements.
The most commonly used type of weighted residual
formulation is the Galerkin Method the trial functions
themselves are used as the Weighting Functions.
The weighted residual formulation leads to the Strong Form,
which can be integrated by parts (in 1D) to obtain the Weak
Form Continuity requirements on trial functions is weak.
Weak Form of Galerkin Statement leads to Symmetric
Stiffness Matrix and Natural B.C. are automatically satisfied
Trial functions need to satisfy the Essential B.C. only!
Still we are missing one of the most important aspect of
F.E.M. The Discretization!!!
Lecture - 03
Assignment 2
Examples 2.2, 2.3, 2.4, 2.5 & 2.6 in the book Finite Elements
and Approximation, by Zienkiewicz and Morgan.
These are solved problems!!! However, all the steps are not
explicitly given in the book. You are expected to do it
completely and submit (e.g. Show how each individual
coefficients used in the matrix equation are obtained!).
Solving these problems should give you a good grip of the
procedure Your understanding of future topics will rely very
much on it!
Lecture - 03